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.
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
19 #include <linux/timer.h>
25 #include <trace/events/f2fs.h>
27 #define __reverse_ffz(x) __reverse_ffs(~(x))
29 static struct kmem_cache *discard_entry_slab;
30 static struct kmem_cache *bio_entry_slab;
31 static struct kmem_cache *sit_entry_set_slab;
32 static struct kmem_cache *inmem_entry_slab;
34 static unsigned long __reverse_ulong(unsigned char *str)
36 unsigned long tmp = 0;
37 int shift = 24, idx = 0;
39 #if BITS_PER_LONG == 64
43 tmp |= (unsigned long)str[idx++] << shift;
44 shift -= BITS_PER_BYTE;
50 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
51 * MSB and LSB are reversed in a byte by f2fs_set_bit.
53 static inline unsigned long __reverse_ffs(unsigned long word)
57 #if BITS_PER_LONG == 64
58 if ((word & 0xffffffff00000000UL) == 0)
63 if ((word & 0xffff0000) == 0)
68 if ((word & 0xff00) == 0)
73 if ((word & 0xf0) == 0)
78 if ((word & 0xc) == 0)
83 if ((word & 0x2) == 0)
89 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
90 * f2fs_set_bit makes MSB and LSB reversed in a byte.
91 * @size must be integral times of unsigned long.
94 * f2fs_set_bit(0, bitmap) => 1000 0000
95 * f2fs_set_bit(7, bitmap) => 0000 0001
97 static unsigned long __find_rev_next_bit(const unsigned long *addr,
98 unsigned long size, unsigned long offset)
100 const unsigned long *p = addr + BIT_WORD(offset);
101 unsigned long result = size;
107 size -= (offset & ~(BITS_PER_LONG - 1));
108 offset %= BITS_PER_LONG;
114 tmp = __reverse_ulong((unsigned char *)p);
116 tmp &= ~0UL >> offset;
117 if (size < BITS_PER_LONG)
118 tmp &= (~0UL << (BITS_PER_LONG - size));
122 if (size <= BITS_PER_LONG)
124 size -= BITS_PER_LONG;
130 return result - size + __reverse_ffs(tmp);
133 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
134 unsigned long size, unsigned long offset)
136 const unsigned long *p = addr + BIT_WORD(offset);
137 unsigned long result = size;
143 size -= (offset & ~(BITS_PER_LONG - 1));
144 offset %= BITS_PER_LONG;
150 tmp = __reverse_ulong((unsigned char *)p);
153 tmp |= ~0UL << (BITS_PER_LONG - offset);
154 if (size < BITS_PER_LONG)
159 if (size <= BITS_PER_LONG)
161 size -= BITS_PER_LONG;
167 return result - size + __reverse_ffz(tmp);
170 void register_inmem_page(struct inode *inode, struct page *page)
172 struct f2fs_inode_info *fi = F2FS_I(inode);
173 struct inmem_pages *new;
175 f2fs_trace_pid(page);
177 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
178 SetPagePrivate(page);
180 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
182 /* add atomic page indices to the list */
184 INIT_LIST_HEAD(&new->list);
186 /* increase reference count with clean state */
187 mutex_lock(&fi->inmem_lock);
189 list_add_tail(&new->list, &fi->inmem_pages);
190 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
191 mutex_unlock(&fi->inmem_lock);
193 trace_f2fs_register_inmem_page(page, INMEM);
196 static int __revoke_inmem_pages(struct inode *inode,
197 struct list_head *head, bool drop, bool recover)
199 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
200 struct inmem_pages *cur, *tmp;
203 list_for_each_entry_safe(cur, tmp, head, list) {
204 struct page *page = cur->page;
207 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
212 struct dnode_of_data dn;
215 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
217 set_new_dnode(&dn, inode, NULL, NULL, 0);
218 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
222 get_node_info(sbi, dn.nid, &ni);
223 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
224 cur->old_addr, ni.version, true, true);
228 /* we don't need to invalidate this in the sccessful status */
230 ClearPageUptodate(page);
231 set_page_private(page, 0);
232 ClearPagePrivate(page);
233 f2fs_put_page(page, 1);
235 list_del(&cur->list);
236 kmem_cache_free(inmem_entry_slab, cur);
237 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
242 void drop_inmem_pages(struct inode *inode)
244 struct f2fs_inode_info *fi = F2FS_I(inode);
246 clear_inode_flag(inode, FI_ATOMIC_FILE);
248 mutex_lock(&fi->inmem_lock);
249 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
250 mutex_unlock(&fi->inmem_lock);
253 static int __commit_inmem_pages(struct inode *inode,
254 struct list_head *revoke_list)
256 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
257 struct f2fs_inode_info *fi = F2FS_I(inode);
258 struct inmem_pages *cur, *tmp;
259 struct f2fs_io_info fio = {
263 .op_flags = WRITE_SYNC | REQ_PRIO,
264 .encrypted_page = NULL,
266 bool submit_bio = false;
269 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
270 struct page *page = cur->page;
273 if (page->mapping == inode->i_mapping) {
274 trace_f2fs_commit_inmem_page(page, INMEM);
276 set_page_dirty(page);
277 f2fs_wait_on_page_writeback(page, DATA, true);
278 if (clear_page_dirty_for_io(page)) {
279 inode_dec_dirty_pages(inode);
280 remove_dirty_inode(inode);
284 err = do_write_data_page(&fio);
290 /* record old blkaddr for revoking */
291 cur->old_addr = fio.old_blkaddr;
293 clear_cold_data(page);
297 list_move_tail(&cur->list, revoke_list);
301 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
304 __revoke_inmem_pages(inode, revoke_list, false, false);
309 int commit_inmem_pages(struct inode *inode)
311 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
312 struct f2fs_inode_info *fi = F2FS_I(inode);
313 struct list_head revoke_list;
316 INIT_LIST_HEAD(&revoke_list);
317 f2fs_balance_fs(sbi, true);
320 mutex_lock(&fi->inmem_lock);
321 err = __commit_inmem_pages(inode, &revoke_list);
325 * try to revoke all committed pages, but still we could fail
326 * due to no memory or other reason, if that happened, EAGAIN
327 * will be returned, which means in such case, transaction is
328 * already not integrity, caller should use journal to do the
329 * recovery or rewrite & commit last transaction. For other
330 * error number, revoking was done by filesystem itself.
332 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
336 /* drop all uncommitted pages */
337 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
339 mutex_unlock(&fi->inmem_lock);
346 * This function balances dirty node and dentry pages.
347 * In addition, it controls garbage collection.
349 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
351 #ifdef CONFIG_F2FS_FAULT_INJECTION
352 if (time_to_inject(sbi, FAULT_CHECKPOINT))
353 f2fs_stop_checkpoint(sbi, false);
359 /* balance_fs_bg is able to be pending */
360 if (excess_cached_nats(sbi))
361 f2fs_balance_fs_bg(sbi);
364 * We should do GC or end up with checkpoint, if there are so many dirty
365 * dir/node pages without enough free segments.
367 if (has_not_enough_free_secs(sbi, 0, 0)) {
368 mutex_lock(&sbi->gc_mutex);
369 f2fs_gc(sbi, false, false);
373 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
375 /* try to shrink extent cache when there is no enough memory */
376 if (!available_free_memory(sbi, EXTENT_CACHE))
377 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
379 /* check the # of cached NAT entries */
380 if (!available_free_memory(sbi, NAT_ENTRIES))
381 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
383 if (!available_free_memory(sbi, FREE_NIDS))
384 try_to_free_nids(sbi, MAX_FREE_NIDS);
386 build_free_nids(sbi, false);
388 /* checkpoint is the only way to shrink partial cached entries */
389 if (!available_free_memory(sbi, NAT_ENTRIES) ||
390 !available_free_memory(sbi, INO_ENTRIES) ||
391 excess_prefree_segs(sbi) ||
392 excess_dirty_nats(sbi) ||
393 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
394 if (test_opt(sbi, DATA_FLUSH)) {
395 struct blk_plug plug;
397 blk_start_plug(&plug);
398 sync_dirty_inodes(sbi, FILE_INODE);
399 blk_finish_plug(&plug);
401 f2fs_sync_fs(sbi->sb, true);
402 stat_inc_bg_cp_count(sbi->stat_info);
406 static int __submit_flush_wait(struct block_device *bdev)
408 struct bio *bio = f2fs_bio_alloc(0);
411 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
413 ret = submit_bio_wait(bio);
418 static int submit_flush_wait(struct f2fs_sb_info *sbi)
420 int ret = __submit_flush_wait(sbi->sb->s_bdev);
423 if (sbi->s_ndevs && !ret) {
424 for (i = 1; i < sbi->s_ndevs; i++) {
425 ret = __submit_flush_wait(FDEV(i).bdev);
433 static int issue_flush_thread(void *data)
435 struct f2fs_sb_info *sbi = data;
436 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
437 wait_queue_head_t *q = &fcc->flush_wait_queue;
439 if (kthread_should_stop())
442 if (!llist_empty(&fcc->issue_list)) {
443 struct flush_cmd *cmd, *next;
446 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
447 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
449 ret = submit_flush_wait(sbi);
450 llist_for_each_entry_safe(cmd, next,
451 fcc->dispatch_list, llnode) {
453 complete(&cmd->wait);
455 fcc->dispatch_list = NULL;
458 wait_event_interruptible(*q,
459 kthread_should_stop() || !llist_empty(&fcc->issue_list));
463 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
465 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
466 struct flush_cmd cmd;
468 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
469 test_opt(sbi, FLUSH_MERGE));
471 if (test_opt(sbi, NOBARRIER))
474 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
477 atomic_inc(&fcc->submit_flush);
478 ret = submit_flush_wait(sbi);
479 atomic_dec(&fcc->submit_flush);
483 init_completion(&cmd.wait);
485 atomic_inc(&fcc->submit_flush);
486 llist_add(&cmd.llnode, &fcc->issue_list);
488 if (!fcc->dispatch_list)
489 wake_up(&fcc->flush_wait_queue);
491 wait_for_completion(&cmd.wait);
492 atomic_dec(&fcc->submit_flush);
497 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
499 dev_t dev = sbi->sb->s_bdev->bd_dev;
500 struct flush_cmd_control *fcc;
503 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
506 atomic_set(&fcc->submit_flush, 0);
507 init_waitqueue_head(&fcc->flush_wait_queue);
508 init_llist_head(&fcc->issue_list);
509 SM_I(sbi)->cmd_control_info = fcc;
510 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
511 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
512 if (IS_ERR(fcc->f2fs_issue_flush)) {
513 err = PTR_ERR(fcc->f2fs_issue_flush);
515 SM_I(sbi)->cmd_control_info = NULL;
522 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
524 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
526 if (fcc && fcc->f2fs_issue_flush)
527 kthread_stop(fcc->f2fs_issue_flush);
529 SM_I(sbi)->cmd_control_info = NULL;
532 static void __locate_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 /* need not be added */
538 if (IS_CURSEG(sbi, segno))
541 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
542 dirty_i->nr_dirty[dirty_type]++;
544 if (dirty_type == DIRTY) {
545 struct seg_entry *sentry = get_seg_entry(sbi, segno);
546 enum dirty_type t = sentry->type;
548 if (unlikely(t >= DIRTY)) {
552 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
553 dirty_i->nr_dirty[t]++;
557 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
558 enum dirty_type dirty_type)
560 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
562 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
563 dirty_i->nr_dirty[dirty_type]--;
565 if (dirty_type == DIRTY) {
566 struct seg_entry *sentry = get_seg_entry(sbi, segno);
567 enum dirty_type t = sentry->type;
569 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
570 dirty_i->nr_dirty[t]--;
572 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
573 clear_bit(GET_SECNO(sbi, segno),
574 dirty_i->victim_secmap);
579 * Should not occur error such as -ENOMEM.
580 * Adding dirty entry into seglist is not critical operation.
581 * If a given segment is one of current working segments, it won't be added.
583 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
585 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
586 unsigned short valid_blocks;
588 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
591 mutex_lock(&dirty_i->seglist_lock);
593 valid_blocks = get_valid_blocks(sbi, segno, 0);
595 if (valid_blocks == 0) {
596 __locate_dirty_segment(sbi, segno, PRE);
597 __remove_dirty_segment(sbi, segno, DIRTY);
598 } else if (valid_blocks < sbi->blocks_per_seg) {
599 __locate_dirty_segment(sbi, segno, DIRTY);
601 /* Recovery routine with SSR needs this */
602 __remove_dirty_segment(sbi, segno, DIRTY);
605 mutex_unlock(&dirty_i->seglist_lock);
608 static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
611 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
612 struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
614 INIT_LIST_HEAD(&be->list);
616 init_completion(&be->event);
617 list_add_tail(&be->list, wait_list);
622 void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
624 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
625 struct bio_entry *be, *tmp;
627 list_for_each_entry_safe(be, tmp, wait_list, list) {
628 struct bio *bio = be->bio;
631 wait_for_completion_io(&be->event);
633 if (err == -EOPNOTSUPP)
637 f2fs_msg(sbi->sb, KERN_INFO,
638 "Issue discard failed, ret: %d", err);
642 kmem_cache_free(bio_entry_slab, be);
646 static void f2fs_submit_bio_wait_endio(struct bio *bio)
648 struct bio_entry *be = (struct bio_entry *)bio->bi_private;
650 be->error = bio->bi_error;
651 complete(&be->event);
654 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
655 static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi,
656 struct block_device *bdev, block_t blkstart, block_t blklen)
658 struct bio *bio = NULL;
661 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
664 int devi = f2fs_target_device_index(sbi, blkstart);
666 blkstart -= FDEV(devi).start_blk;
668 err = __blkdev_issue_discard(bdev,
669 SECTOR_FROM_BLOCK(blkstart),
670 SECTOR_FROM_BLOCK(blklen),
673 struct bio_entry *be = __add_bio_entry(sbi, bio);
675 bio->bi_private = be;
676 bio->bi_end_io = f2fs_submit_bio_wait_endio;
677 bio->bi_opf |= REQ_SYNC;
684 #ifdef CONFIG_BLK_DEV_ZONED
685 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
686 struct block_device *bdev, block_t blkstart, block_t blklen)
688 sector_t nr_sects = SECTOR_FROM_BLOCK(blklen);
693 devi = f2fs_target_device_index(sbi, blkstart);
694 blkstart -= FDEV(devi).start_blk;
696 sector = SECTOR_FROM_BLOCK(blkstart);
698 if (sector % bdev_zone_size(bdev) || nr_sects != bdev_zone_size(bdev)) {
699 f2fs_msg(sbi->sb, KERN_INFO,
700 "(%d) %s: Unaligned discard attempted (block %x + %x)",
701 devi, sbi->s_ndevs ? FDEV(devi).path: "",
707 * We need to know the type of the zone: for conventional zones,
708 * use regular discard if the drive supports it. For sequential
709 * zones, reset the zone write pointer.
711 switch (get_blkz_type(sbi, bdev, blkstart)) {
713 case BLK_ZONE_TYPE_CONVENTIONAL:
714 if (!blk_queue_discard(bdev_get_queue(bdev)))
716 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
717 case BLK_ZONE_TYPE_SEQWRITE_REQ:
718 case BLK_ZONE_TYPE_SEQWRITE_PREF:
719 trace_f2fs_issue_reset_zone(sbi->sb, blkstart);
720 return blkdev_reset_zones(bdev, sector,
723 /* Unknown zone type: broken device ? */
729 static int __issue_discard_async(struct f2fs_sb_info *sbi,
730 struct block_device *bdev, block_t blkstart, block_t blklen)
732 #ifdef CONFIG_BLK_DEV_ZONED
733 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
734 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
735 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
737 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
740 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
741 block_t blkstart, block_t blklen)
743 sector_t start = blkstart, len = 0;
744 struct block_device *bdev;
745 struct seg_entry *se;
750 bdev = f2fs_target_device(sbi, blkstart, NULL);
752 for (i = blkstart; i < blkstart + blklen; i++, len++) {
754 struct block_device *bdev2 =
755 f2fs_target_device(sbi, i, NULL);
758 err = __issue_discard_async(sbi, bdev,
768 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
769 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
771 if (!f2fs_test_and_set_bit(offset, se->discard_map))
776 err = __issue_discard_async(sbi, bdev, start, len);
780 static void __add_discard_entry(struct f2fs_sb_info *sbi,
781 struct cp_control *cpc, struct seg_entry *se,
782 unsigned int start, unsigned int end)
784 struct list_head *head = &SM_I(sbi)->discard_list;
785 struct discard_entry *new, *last;
787 if (!list_empty(head)) {
788 last = list_last_entry(head, struct discard_entry, list);
789 if (START_BLOCK(sbi, cpc->trim_start) + start ==
790 last->blkaddr + last->len) {
791 last->len += end - start;
796 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
797 INIT_LIST_HEAD(&new->list);
798 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
799 new->len = end - start;
800 list_add_tail(&new->list, head);
802 SM_I(sbi)->nr_discards += end - start;
805 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
807 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
808 int max_blocks = sbi->blocks_per_seg;
809 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
810 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
811 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
812 unsigned long *discard_map = (unsigned long *)se->discard_map;
813 unsigned long *dmap = SIT_I(sbi)->tmp_map;
814 unsigned int start = 0, end = -1;
815 bool force = (cpc->reason == CP_DISCARD);
818 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
822 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
823 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
827 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
828 for (i = 0; i < entries; i++)
829 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
830 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
832 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
833 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
834 if (start >= max_blocks)
837 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
838 if (force && start && end != max_blocks
839 && (end - start) < cpc->trim_minlen)
842 __add_discard_entry(sbi, cpc, se, start, end);
846 void release_discard_addrs(struct f2fs_sb_info *sbi)
848 struct list_head *head = &(SM_I(sbi)->discard_list);
849 struct discard_entry *entry, *this;
852 list_for_each_entry_safe(entry, this, head, list) {
853 list_del(&entry->list);
854 kmem_cache_free(discard_entry_slab, entry);
859 * Should call clear_prefree_segments after checkpoint is done.
861 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
863 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
866 mutex_lock(&dirty_i->seglist_lock);
867 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
868 __set_test_and_free(sbi, segno);
869 mutex_unlock(&dirty_i->seglist_lock);
872 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
874 struct list_head *head = &(SM_I(sbi)->discard_list);
875 struct discard_entry *entry, *this;
876 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
877 struct blk_plug plug;
878 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
879 unsigned int start = 0, end = -1;
880 unsigned int secno, start_segno;
881 bool force = (cpc->reason == CP_DISCARD);
883 blk_start_plug(&plug);
885 mutex_lock(&dirty_i->seglist_lock);
889 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
890 if (start >= MAIN_SEGS(sbi))
892 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
895 for (i = start; i < end; i++)
896 clear_bit(i, prefree_map);
898 dirty_i->nr_dirty[PRE] -= end - start;
900 if (force || !test_opt(sbi, DISCARD))
903 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
904 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
905 (end - start) << sbi->log_blocks_per_seg);
909 secno = GET_SECNO(sbi, start);
910 start_segno = secno * sbi->segs_per_sec;
911 if (!IS_CURSEC(sbi, secno) &&
912 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
913 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
914 sbi->segs_per_sec << sbi->log_blocks_per_seg);
916 start = start_segno + sbi->segs_per_sec;
920 mutex_unlock(&dirty_i->seglist_lock);
922 /* send small discards */
923 list_for_each_entry_safe(entry, this, head, list) {
924 if (force && entry->len < cpc->trim_minlen)
926 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
927 cpc->trimmed += entry->len;
929 list_del(&entry->list);
930 SM_I(sbi)->nr_discards -= entry->len;
931 kmem_cache_free(discard_entry_slab, entry);
934 blk_finish_plug(&plug);
937 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
939 struct sit_info *sit_i = SIT_I(sbi);
941 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
942 sit_i->dirty_sentries++;
949 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
950 unsigned int segno, int modified)
952 struct seg_entry *se = get_seg_entry(sbi, segno);
955 __mark_sit_entry_dirty(sbi, segno);
958 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
960 struct seg_entry *se;
961 unsigned int segno, offset;
962 long int new_vblocks;
964 segno = GET_SEGNO(sbi, blkaddr);
966 se = get_seg_entry(sbi, segno);
967 new_vblocks = se->valid_blocks + del;
968 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
970 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
971 (new_vblocks > sbi->blocks_per_seg)));
973 se->valid_blocks = new_vblocks;
974 se->mtime = get_mtime(sbi);
975 SIT_I(sbi)->max_mtime = se->mtime;
977 /* Update valid block bitmap */
979 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
981 if (f2fs_discard_en(sbi) &&
982 !f2fs_test_and_set_bit(offset, se->discard_map))
985 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
987 if (f2fs_discard_en(sbi) &&
988 f2fs_test_and_clear_bit(offset, se->discard_map))
991 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
992 se->ckpt_valid_blocks += del;
994 __mark_sit_entry_dirty(sbi, segno);
996 /* update total number of valid blocks to be written in ckpt area */
997 SIT_I(sbi)->written_valid_blocks += del;
999 if (sbi->segs_per_sec > 1)
1000 get_sec_entry(sbi, segno)->valid_blocks += del;
1003 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1005 update_sit_entry(sbi, new, 1);
1006 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1007 update_sit_entry(sbi, old, -1);
1009 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1010 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1013 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1015 unsigned int segno = GET_SEGNO(sbi, addr);
1016 struct sit_info *sit_i = SIT_I(sbi);
1018 f2fs_bug_on(sbi, addr == NULL_ADDR);
1019 if (addr == NEW_ADDR)
1022 /* add it into sit main buffer */
1023 mutex_lock(&sit_i->sentry_lock);
1025 update_sit_entry(sbi, addr, -1);
1027 /* add it into dirty seglist */
1028 locate_dirty_segment(sbi, segno);
1030 mutex_unlock(&sit_i->sentry_lock);
1033 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1035 struct sit_info *sit_i = SIT_I(sbi);
1036 unsigned int segno, offset;
1037 struct seg_entry *se;
1040 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1043 mutex_lock(&sit_i->sentry_lock);
1045 segno = GET_SEGNO(sbi, blkaddr);
1046 se = get_seg_entry(sbi, segno);
1047 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1049 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1052 mutex_unlock(&sit_i->sentry_lock);
1058 * This function should be resided under the curseg_mutex lock
1060 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1061 struct f2fs_summary *sum)
1063 struct curseg_info *curseg = CURSEG_I(sbi, type);
1064 void *addr = curseg->sum_blk;
1065 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1066 memcpy(addr, sum, sizeof(struct f2fs_summary));
1070 * Calculate the number of current summary pages for writing
1072 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1074 int valid_sum_count = 0;
1077 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1078 if (sbi->ckpt->alloc_type[i] == SSR)
1079 valid_sum_count += sbi->blocks_per_seg;
1082 valid_sum_count += le16_to_cpu(
1083 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1085 valid_sum_count += curseg_blkoff(sbi, i);
1089 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1090 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1091 if (valid_sum_count <= sum_in_page)
1093 else if ((valid_sum_count - sum_in_page) <=
1094 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1100 * Caller should put this summary page
1102 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1104 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1107 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1109 struct page *page = grab_meta_page(sbi, blk_addr);
1110 void *dst = page_address(page);
1113 memcpy(dst, src, PAGE_SIZE);
1115 memset(dst, 0, PAGE_SIZE);
1116 set_page_dirty(page);
1117 f2fs_put_page(page, 1);
1120 static void write_sum_page(struct f2fs_sb_info *sbi,
1121 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1123 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1126 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1127 int type, block_t blk_addr)
1129 struct curseg_info *curseg = CURSEG_I(sbi, type);
1130 struct page *page = grab_meta_page(sbi, blk_addr);
1131 struct f2fs_summary_block *src = curseg->sum_blk;
1132 struct f2fs_summary_block *dst;
1134 dst = (struct f2fs_summary_block *)page_address(page);
1136 mutex_lock(&curseg->curseg_mutex);
1138 down_read(&curseg->journal_rwsem);
1139 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1140 up_read(&curseg->journal_rwsem);
1142 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1143 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1145 mutex_unlock(&curseg->curseg_mutex);
1147 set_page_dirty(page);
1148 f2fs_put_page(page, 1);
1151 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1153 struct curseg_info *curseg = CURSEG_I(sbi, type);
1154 unsigned int segno = curseg->segno + 1;
1155 struct free_segmap_info *free_i = FREE_I(sbi);
1157 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1158 return !test_bit(segno, free_i->free_segmap);
1163 * Find a new segment from the free segments bitmap to right order
1164 * This function should be returned with success, otherwise BUG
1166 static void get_new_segment(struct f2fs_sb_info *sbi,
1167 unsigned int *newseg, bool new_sec, int dir)
1169 struct free_segmap_info *free_i = FREE_I(sbi);
1170 unsigned int segno, secno, zoneno;
1171 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1172 unsigned int hint = *newseg / sbi->segs_per_sec;
1173 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1174 unsigned int left_start = hint;
1179 spin_lock(&free_i->segmap_lock);
1181 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1182 segno = find_next_zero_bit(free_i->free_segmap,
1183 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1184 if (segno < (hint + 1) * sbi->segs_per_sec)
1188 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1189 if (secno >= MAIN_SECS(sbi)) {
1190 if (dir == ALLOC_RIGHT) {
1191 secno = find_next_zero_bit(free_i->free_secmap,
1193 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1196 left_start = hint - 1;
1202 while (test_bit(left_start, free_i->free_secmap)) {
1203 if (left_start > 0) {
1207 left_start = find_next_zero_bit(free_i->free_secmap,
1209 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1215 segno = secno * sbi->segs_per_sec;
1216 zoneno = secno / sbi->secs_per_zone;
1218 /* give up on finding another zone */
1221 if (sbi->secs_per_zone == 1)
1223 if (zoneno == old_zoneno)
1225 if (dir == ALLOC_LEFT) {
1226 if (!go_left && zoneno + 1 >= total_zones)
1228 if (go_left && zoneno == 0)
1231 for (i = 0; i < NR_CURSEG_TYPE; i++)
1232 if (CURSEG_I(sbi, i)->zone == zoneno)
1235 if (i < NR_CURSEG_TYPE) {
1236 /* zone is in user, try another */
1238 hint = zoneno * sbi->secs_per_zone - 1;
1239 else if (zoneno + 1 >= total_zones)
1242 hint = (zoneno + 1) * sbi->secs_per_zone;
1244 goto find_other_zone;
1247 /* set it as dirty segment in free segmap */
1248 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1249 __set_inuse(sbi, segno);
1251 spin_unlock(&free_i->segmap_lock);
1254 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1256 struct curseg_info *curseg = CURSEG_I(sbi, type);
1257 struct summary_footer *sum_footer;
1259 curseg->segno = curseg->next_segno;
1260 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1261 curseg->next_blkoff = 0;
1262 curseg->next_segno = NULL_SEGNO;
1264 sum_footer = &(curseg->sum_blk->footer);
1265 memset(sum_footer, 0, sizeof(struct summary_footer));
1266 if (IS_DATASEG(type))
1267 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1268 if (IS_NODESEG(type))
1269 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1270 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1274 * Allocate a current working segment.
1275 * This function always allocates a free segment in LFS manner.
1277 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1279 struct curseg_info *curseg = CURSEG_I(sbi, type);
1280 unsigned int segno = curseg->segno;
1281 int dir = ALLOC_LEFT;
1283 write_sum_page(sbi, curseg->sum_blk,
1284 GET_SUM_BLOCK(sbi, segno));
1285 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1288 if (test_opt(sbi, NOHEAP))
1291 get_new_segment(sbi, &segno, new_sec, dir);
1292 curseg->next_segno = segno;
1293 reset_curseg(sbi, type, 1);
1294 curseg->alloc_type = LFS;
1297 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1298 struct curseg_info *seg, block_t start)
1300 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1301 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1302 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1303 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1304 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1307 for (i = 0; i < entries; i++)
1308 target_map[i] = ckpt_map[i] | cur_map[i];
1310 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1312 seg->next_blkoff = pos;
1316 * If a segment is written by LFS manner, next block offset is just obtained
1317 * by increasing the current block offset. However, if a segment is written by
1318 * SSR manner, next block offset obtained by calling __next_free_blkoff
1320 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1321 struct curseg_info *seg)
1323 if (seg->alloc_type == SSR)
1324 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1330 * This function always allocates a used segment(from dirty seglist) by SSR
1331 * manner, so it should recover the existing segment information of valid blocks
1333 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1335 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1336 struct curseg_info *curseg = CURSEG_I(sbi, type);
1337 unsigned int new_segno = curseg->next_segno;
1338 struct f2fs_summary_block *sum_node;
1339 struct page *sum_page;
1341 write_sum_page(sbi, curseg->sum_blk,
1342 GET_SUM_BLOCK(sbi, curseg->segno));
1343 __set_test_and_inuse(sbi, new_segno);
1345 mutex_lock(&dirty_i->seglist_lock);
1346 __remove_dirty_segment(sbi, new_segno, PRE);
1347 __remove_dirty_segment(sbi, new_segno, DIRTY);
1348 mutex_unlock(&dirty_i->seglist_lock);
1350 reset_curseg(sbi, type, 1);
1351 curseg->alloc_type = SSR;
1352 __next_free_blkoff(sbi, curseg, 0);
1355 sum_page = get_sum_page(sbi, new_segno);
1356 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1357 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1358 f2fs_put_page(sum_page, 1);
1362 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1364 struct curseg_info *curseg = CURSEG_I(sbi, type);
1365 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1367 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0, 0))
1368 return v_ops->get_victim(sbi,
1369 &(curseg)->next_segno, BG_GC, type, SSR);
1371 /* For data segments, let's do SSR more intensively */
1372 for (; type >= CURSEG_HOT_DATA; type--)
1373 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1380 * flush out current segment and replace it with new segment
1381 * This function should be returned with success, otherwise BUG
1383 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1384 int type, bool force)
1386 struct curseg_info *curseg = CURSEG_I(sbi, type);
1389 new_curseg(sbi, type, true);
1390 else if (type == CURSEG_WARM_NODE)
1391 new_curseg(sbi, type, false);
1392 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1393 new_curseg(sbi, type, false);
1394 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1395 change_curseg(sbi, type, true);
1397 new_curseg(sbi, type, false);
1399 stat_inc_seg_type(sbi, curseg);
1402 void allocate_new_segments(struct f2fs_sb_info *sbi)
1404 struct curseg_info *curseg;
1405 unsigned int old_segno;
1408 if (test_opt(sbi, LFS))
1411 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1412 curseg = CURSEG_I(sbi, i);
1413 old_segno = curseg->segno;
1414 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1415 locate_dirty_segment(sbi, old_segno);
1419 static const struct segment_allocation default_salloc_ops = {
1420 .allocate_segment = allocate_segment_by_default,
1423 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1425 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1426 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1427 unsigned int start_segno, end_segno;
1428 struct cp_control cpc;
1431 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1435 if (end <= MAIN_BLKADDR(sbi))
1438 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1439 f2fs_msg(sbi->sb, KERN_WARNING,
1440 "Found FS corruption, run fsck to fix.");
1444 /* start/end segment number in main_area */
1445 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1446 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1447 GET_SEGNO(sbi, end);
1448 cpc.reason = CP_DISCARD;
1449 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1451 /* do checkpoint to issue discard commands safely */
1452 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1453 cpc.trim_start = start_segno;
1455 if (sbi->discard_blks == 0)
1457 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1458 cpc.trim_end = end_segno;
1460 cpc.trim_end = min_t(unsigned int,
1461 rounddown(start_segno +
1462 BATCHED_TRIM_SEGMENTS(sbi),
1463 sbi->segs_per_sec) - 1, end_segno);
1465 mutex_lock(&sbi->gc_mutex);
1466 err = write_checkpoint(sbi, &cpc);
1467 mutex_unlock(&sbi->gc_mutex);
1474 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1478 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1480 struct curseg_info *curseg = CURSEG_I(sbi, type);
1481 if (curseg->next_blkoff < sbi->blocks_per_seg)
1486 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1489 return CURSEG_HOT_DATA;
1491 return CURSEG_HOT_NODE;
1494 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1496 if (p_type == DATA) {
1497 struct inode *inode = page->mapping->host;
1499 if (S_ISDIR(inode->i_mode))
1500 return CURSEG_HOT_DATA;
1502 return CURSEG_COLD_DATA;
1504 if (IS_DNODE(page) && is_cold_node(page))
1505 return CURSEG_WARM_NODE;
1507 return CURSEG_COLD_NODE;
1511 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1513 if (p_type == DATA) {
1514 struct inode *inode = page->mapping->host;
1516 if (S_ISDIR(inode->i_mode))
1517 return CURSEG_HOT_DATA;
1518 else if (is_cold_data(page) || file_is_cold(inode))
1519 return CURSEG_COLD_DATA;
1521 return CURSEG_WARM_DATA;
1524 return is_cold_node(page) ? CURSEG_WARM_NODE :
1527 return CURSEG_COLD_NODE;
1531 static int __get_segment_type(struct page *page, enum page_type p_type)
1533 switch (F2FS_P_SB(page)->active_logs) {
1535 return __get_segment_type_2(page, p_type);
1537 return __get_segment_type_4(page, p_type);
1539 /* NR_CURSEG_TYPE(6) logs by default */
1540 f2fs_bug_on(F2FS_P_SB(page),
1541 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1542 return __get_segment_type_6(page, p_type);
1545 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1546 block_t old_blkaddr, block_t *new_blkaddr,
1547 struct f2fs_summary *sum, int type)
1549 struct sit_info *sit_i = SIT_I(sbi);
1550 struct curseg_info *curseg = CURSEG_I(sbi, type);
1552 mutex_lock(&curseg->curseg_mutex);
1553 mutex_lock(&sit_i->sentry_lock);
1555 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1558 * __add_sum_entry should be resided under the curseg_mutex
1559 * because, this function updates a summary entry in the
1560 * current summary block.
1562 __add_sum_entry(sbi, type, sum);
1564 __refresh_next_blkoff(sbi, curseg);
1566 stat_inc_block_count(sbi, curseg);
1568 if (!__has_curseg_space(sbi, type))
1569 sit_i->s_ops->allocate_segment(sbi, type, false);
1571 * SIT information should be updated before segment allocation,
1572 * since SSR needs latest valid block information.
1574 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1576 mutex_unlock(&sit_i->sentry_lock);
1578 if (page && IS_NODESEG(type))
1579 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1581 mutex_unlock(&curseg->curseg_mutex);
1584 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1586 int type = __get_segment_type(fio->page, fio->type);
1588 if (fio->type == NODE || fio->type == DATA)
1589 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1591 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1592 &fio->new_blkaddr, sum, type);
1594 /* writeout dirty page into bdev */
1595 f2fs_submit_page_mbio(fio);
1597 if (fio->type == NODE || fio->type == DATA)
1598 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1601 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1603 struct f2fs_io_info fio = {
1607 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
1608 .old_blkaddr = page->index,
1609 .new_blkaddr = page->index,
1611 .encrypted_page = NULL,
1614 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1615 fio.op_flags &= ~REQ_META;
1617 set_page_writeback(page);
1618 f2fs_submit_page_mbio(&fio);
1621 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1623 struct f2fs_summary sum;
1625 set_summary(&sum, nid, 0, 0);
1626 do_write_page(&sum, fio);
1629 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1631 struct f2fs_sb_info *sbi = fio->sbi;
1632 struct f2fs_summary sum;
1633 struct node_info ni;
1635 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1636 get_node_info(sbi, dn->nid, &ni);
1637 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1638 do_write_page(&sum, fio);
1639 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1642 void rewrite_data_page(struct f2fs_io_info *fio)
1644 fio->new_blkaddr = fio->old_blkaddr;
1645 stat_inc_inplace_blocks(fio->sbi);
1646 f2fs_submit_page_mbio(fio);
1649 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1650 block_t old_blkaddr, block_t new_blkaddr,
1651 bool recover_curseg, bool recover_newaddr)
1653 struct sit_info *sit_i = SIT_I(sbi);
1654 struct curseg_info *curseg;
1655 unsigned int segno, old_cursegno;
1656 struct seg_entry *se;
1658 unsigned short old_blkoff;
1660 segno = GET_SEGNO(sbi, new_blkaddr);
1661 se = get_seg_entry(sbi, segno);
1664 if (!recover_curseg) {
1665 /* for recovery flow */
1666 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1667 if (old_blkaddr == NULL_ADDR)
1668 type = CURSEG_COLD_DATA;
1670 type = CURSEG_WARM_DATA;
1673 if (!IS_CURSEG(sbi, segno))
1674 type = CURSEG_WARM_DATA;
1677 curseg = CURSEG_I(sbi, type);
1679 mutex_lock(&curseg->curseg_mutex);
1680 mutex_lock(&sit_i->sentry_lock);
1682 old_cursegno = curseg->segno;
1683 old_blkoff = curseg->next_blkoff;
1685 /* change the current segment */
1686 if (segno != curseg->segno) {
1687 curseg->next_segno = segno;
1688 change_curseg(sbi, type, true);
1691 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1692 __add_sum_entry(sbi, type, sum);
1694 if (!recover_curseg || recover_newaddr)
1695 update_sit_entry(sbi, new_blkaddr, 1);
1696 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1697 update_sit_entry(sbi, old_blkaddr, -1);
1699 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1700 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1702 locate_dirty_segment(sbi, old_cursegno);
1704 if (recover_curseg) {
1705 if (old_cursegno != curseg->segno) {
1706 curseg->next_segno = old_cursegno;
1707 change_curseg(sbi, type, true);
1709 curseg->next_blkoff = old_blkoff;
1712 mutex_unlock(&sit_i->sentry_lock);
1713 mutex_unlock(&curseg->curseg_mutex);
1716 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1717 block_t old_addr, block_t new_addr,
1718 unsigned char version, bool recover_curseg,
1719 bool recover_newaddr)
1721 struct f2fs_summary sum;
1723 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1725 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1726 recover_curseg, recover_newaddr);
1728 f2fs_update_data_blkaddr(dn, new_addr);
1731 void f2fs_wait_on_page_writeback(struct page *page,
1732 enum page_type type, bool ordered)
1734 if (PageWriteback(page)) {
1735 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1737 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1739 wait_on_page_writeback(page);
1741 wait_for_stable_page(page);
1745 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1750 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1753 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1755 f2fs_wait_on_page_writeback(cpage, DATA, true);
1756 f2fs_put_page(cpage, 1);
1760 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1762 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1763 struct curseg_info *seg_i;
1764 unsigned char *kaddr;
1769 start = start_sum_block(sbi);
1771 page = get_meta_page(sbi, start++);
1772 kaddr = (unsigned char *)page_address(page);
1774 /* Step 1: restore nat cache */
1775 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1776 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1778 /* Step 2: restore sit cache */
1779 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1780 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1781 offset = 2 * SUM_JOURNAL_SIZE;
1783 /* Step 3: restore summary entries */
1784 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1785 unsigned short blk_off;
1788 seg_i = CURSEG_I(sbi, i);
1789 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1790 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1791 seg_i->next_segno = segno;
1792 reset_curseg(sbi, i, 0);
1793 seg_i->alloc_type = ckpt->alloc_type[i];
1794 seg_i->next_blkoff = blk_off;
1796 if (seg_i->alloc_type == SSR)
1797 blk_off = sbi->blocks_per_seg;
1799 for (j = 0; j < blk_off; j++) {
1800 struct f2fs_summary *s;
1801 s = (struct f2fs_summary *)(kaddr + offset);
1802 seg_i->sum_blk->entries[j] = *s;
1803 offset += SUMMARY_SIZE;
1804 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1808 f2fs_put_page(page, 1);
1811 page = get_meta_page(sbi, start++);
1812 kaddr = (unsigned char *)page_address(page);
1816 f2fs_put_page(page, 1);
1820 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1822 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1823 struct f2fs_summary_block *sum;
1824 struct curseg_info *curseg;
1826 unsigned short blk_off;
1827 unsigned int segno = 0;
1828 block_t blk_addr = 0;
1830 /* get segment number and block addr */
1831 if (IS_DATASEG(type)) {
1832 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1833 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1835 if (__exist_node_summaries(sbi))
1836 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1838 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1840 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1842 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1844 if (__exist_node_summaries(sbi))
1845 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1846 type - CURSEG_HOT_NODE);
1848 blk_addr = GET_SUM_BLOCK(sbi, segno);
1851 new = get_meta_page(sbi, blk_addr);
1852 sum = (struct f2fs_summary_block *)page_address(new);
1854 if (IS_NODESEG(type)) {
1855 if (__exist_node_summaries(sbi)) {
1856 struct f2fs_summary *ns = &sum->entries[0];
1858 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1860 ns->ofs_in_node = 0;
1865 err = restore_node_summary(sbi, segno, sum);
1867 f2fs_put_page(new, 1);
1873 /* set uncompleted segment to curseg */
1874 curseg = CURSEG_I(sbi, type);
1875 mutex_lock(&curseg->curseg_mutex);
1877 /* update journal info */
1878 down_write(&curseg->journal_rwsem);
1879 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1880 up_write(&curseg->journal_rwsem);
1882 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1883 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1884 curseg->next_segno = segno;
1885 reset_curseg(sbi, type, 0);
1886 curseg->alloc_type = ckpt->alloc_type[type];
1887 curseg->next_blkoff = blk_off;
1888 mutex_unlock(&curseg->curseg_mutex);
1889 f2fs_put_page(new, 1);
1893 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1895 int type = CURSEG_HOT_DATA;
1898 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
1899 int npages = npages_for_summary_flush(sbi, true);
1902 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1905 /* restore for compacted data summary */
1906 if (read_compacted_summaries(sbi))
1908 type = CURSEG_HOT_NODE;
1911 if (__exist_node_summaries(sbi))
1912 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1913 NR_CURSEG_TYPE - type, META_CP, true);
1915 for (; type <= CURSEG_COLD_NODE; type++) {
1916 err = read_normal_summaries(sbi, type);
1924 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1927 unsigned char *kaddr;
1928 struct f2fs_summary *summary;
1929 struct curseg_info *seg_i;
1930 int written_size = 0;
1933 page = grab_meta_page(sbi, blkaddr++);
1934 kaddr = (unsigned char *)page_address(page);
1936 /* Step 1: write nat cache */
1937 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1938 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1939 written_size += SUM_JOURNAL_SIZE;
1941 /* Step 2: write sit cache */
1942 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1943 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1944 written_size += SUM_JOURNAL_SIZE;
1946 /* Step 3: write summary entries */
1947 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1948 unsigned short blkoff;
1949 seg_i = CURSEG_I(sbi, i);
1950 if (sbi->ckpt->alloc_type[i] == SSR)
1951 blkoff = sbi->blocks_per_seg;
1953 blkoff = curseg_blkoff(sbi, i);
1955 for (j = 0; j < blkoff; j++) {
1957 page = grab_meta_page(sbi, blkaddr++);
1958 kaddr = (unsigned char *)page_address(page);
1961 summary = (struct f2fs_summary *)(kaddr + written_size);
1962 *summary = seg_i->sum_blk->entries[j];
1963 written_size += SUMMARY_SIZE;
1965 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1969 set_page_dirty(page);
1970 f2fs_put_page(page, 1);
1975 set_page_dirty(page);
1976 f2fs_put_page(page, 1);
1980 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1981 block_t blkaddr, int type)
1984 if (IS_DATASEG(type))
1985 end = type + NR_CURSEG_DATA_TYPE;
1987 end = type + NR_CURSEG_NODE_TYPE;
1989 for (i = type; i < end; i++)
1990 write_current_sum_page(sbi, i, blkaddr + (i - type));
1993 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1995 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
1996 write_compacted_summaries(sbi, start_blk);
1998 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2001 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2003 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2006 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2007 unsigned int val, int alloc)
2011 if (type == NAT_JOURNAL) {
2012 for (i = 0; i < nats_in_cursum(journal); i++) {
2013 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2016 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2017 return update_nats_in_cursum(journal, 1);
2018 } else if (type == SIT_JOURNAL) {
2019 for (i = 0; i < sits_in_cursum(journal); i++)
2020 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2022 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2023 return update_sits_in_cursum(journal, 1);
2028 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2031 return get_meta_page(sbi, current_sit_addr(sbi, segno));
2034 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2037 struct sit_info *sit_i = SIT_I(sbi);
2038 struct page *src_page, *dst_page;
2039 pgoff_t src_off, dst_off;
2040 void *src_addr, *dst_addr;
2042 src_off = current_sit_addr(sbi, start);
2043 dst_off = next_sit_addr(sbi, src_off);
2045 /* get current sit block page without lock */
2046 src_page = get_meta_page(sbi, src_off);
2047 dst_page = grab_meta_page(sbi, dst_off);
2048 f2fs_bug_on(sbi, PageDirty(src_page));
2050 src_addr = page_address(src_page);
2051 dst_addr = page_address(dst_page);
2052 memcpy(dst_addr, src_addr, PAGE_SIZE);
2054 set_page_dirty(dst_page);
2055 f2fs_put_page(src_page, 1);
2057 set_to_next_sit(sit_i, start);
2062 static struct sit_entry_set *grab_sit_entry_set(void)
2064 struct sit_entry_set *ses =
2065 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2068 INIT_LIST_HEAD(&ses->set_list);
2072 static void release_sit_entry_set(struct sit_entry_set *ses)
2074 list_del(&ses->set_list);
2075 kmem_cache_free(sit_entry_set_slab, ses);
2078 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2079 struct list_head *head)
2081 struct sit_entry_set *next = ses;
2083 if (list_is_last(&ses->set_list, head))
2086 list_for_each_entry_continue(next, head, set_list)
2087 if (ses->entry_cnt <= next->entry_cnt)
2090 list_move_tail(&ses->set_list, &next->set_list);
2093 static void add_sit_entry(unsigned int segno, struct list_head *head)
2095 struct sit_entry_set *ses;
2096 unsigned int start_segno = START_SEGNO(segno);
2098 list_for_each_entry(ses, head, set_list) {
2099 if (ses->start_segno == start_segno) {
2101 adjust_sit_entry_set(ses, head);
2106 ses = grab_sit_entry_set();
2108 ses->start_segno = start_segno;
2110 list_add(&ses->set_list, head);
2113 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2115 struct f2fs_sm_info *sm_info = SM_I(sbi);
2116 struct list_head *set_list = &sm_info->sit_entry_set;
2117 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2120 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2121 add_sit_entry(segno, set_list);
2124 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2126 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2127 struct f2fs_journal *journal = curseg->journal;
2130 down_write(&curseg->journal_rwsem);
2131 for (i = 0; i < sits_in_cursum(journal); i++) {
2135 segno = le32_to_cpu(segno_in_journal(journal, i));
2136 dirtied = __mark_sit_entry_dirty(sbi, segno);
2139 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2141 update_sits_in_cursum(journal, -i);
2142 up_write(&curseg->journal_rwsem);
2146 * CP calls this function, which flushes SIT entries including sit_journal,
2147 * and moves prefree segs to free segs.
2149 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2151 struct sit_info *sit_i = SIT_I(sbi);
2152 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2153 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2154 struct f2fs_journal *journal = curseg->journal;
2155 struct sit_entry_set *ses, *tmp;
2156 struct list_head *head = &SM_I(sbi)->sit_entry_set;
2157 bool to_journal = true;
2158 struct seg_entry *se;
2160 mutex_lock(&sit_i->sentry_lock);
2162 if (!sit_i->dirty_sentries)
2166 * add and account sit entries of dirty bitmap in sit entry
2169 add_sits_in_set(sbi);
2172 * if there are no enough space in journal to store dirty sit
2173 * entries, remove all entries from journal and add and account
2174 * them in sit entry set.
2176 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2177 remove_sits_in_journal(sbi);
2180 * there are two steps to flush sit entries:
2181 * #1, flush sit entries to journal in current cold data summary block.
2182 * #2, flush sit entries to sit page.
2184 list_for_each_entry_safe(ses, tmp, head, set_list) {
2185 struct page *page = NULL;
2186 struct f2fs_sit_block *raw_sit = NULL;
2187 unsigned int start_segno = ses->start_segno;
2188 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2189 (unsigned long)MAIN_SEGS(sbi));
2190 unsigned int segno = start_segno;
2193 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2197 down_write(&curseg->journal_rwsem);
2199 page = get_next_sit_page(sbi, start_segno);
2200 raw_sit = page_address(page);
2203 /* flush dirty sit entries in region of current sit set */
2204 for_each_set_bit_from(segno, bitmap, end) {
2205 int offset, sit_offset;
2207 se = get_seg_entry(sbi, segno);
2209 /* add discard candidates */
2210 if (cpc->reason != CP_DISCARD) {
2211 cpc->trim_start = segno;
2212 add_discard_addrs(sbi, cpc);
2216 offset = lookup_journal_in_cursum(journal,
2217 SIT_JOURNAL, segno, 1);
2218 f2fs_bug_on(sbi, offset < 0);
2219 segno_in_journal(journal, offset) =
2221 seg_info_to_raw_sit(se,
2222 &sit_in_journal(journal, offset));
2224 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2225 seg_info_to_raw_sit(se,
2226 &raw_sit->entries[sit_offset]);
2229 __clear_bit(segno, bitmap);
2230 sit_i->dirty_sentries--;
2235 up_write(&curseg->journal_rwsem);
2237 f2fs_put_page(page, 1);
2239 f2fs_bug_on(sbi, ses->entry_cnt);
2240 release_sit_entry_set(ses);
2243 f2fs_bug_on(sbi, !list_empty(head));
2244 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2246 if (cpc->reason == CP_DISCARD) {
2247 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2248 add_discard_addrs(sbi, cpc);
2250 mutex_unlock(&sit_i->sentry_lock);
2252 set_prefree_as_free_segments(sbi);
2255 static int build_sit_info(struct f2fs_sb_info *sbi)
2257 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2258 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2259 struct sit_info *sit_i;
2260 unsigned int sit_segs, start;
2261 char *src_bitmap, *dst_bitmap;
2262 unsigned int bitmap_size;
2264 /* allocate memory for SIT information */
2265 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2269 SM_I(sbi)->sit_info = sit_i;
2271 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2272 sizeof(struct seg_entry), GFP_KERNEL);
2273 if (!sit_i->sentries)
2276 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2277 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2278 if (!sit_i->dirty_sentries_bitmap)
2281 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2282 sit_i->sentries[start].cur_valid_map
2283 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2284 sit_i->sentries[start].ckpt_valid_map
2285 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2286 if (!sit_i->sentries[start].cur_valid_map ||
2287 !sit_i->sentries[start].ckpt_valid_map)
2290 if (f2fs_discard_en(sbi)) {
2291 sit_i->sentries[start].discard_map
2292 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2293 if (!sit_i->sentries[start].discard_map)
2298 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2299 if (!sit_i->tmp_map)
2302 if (sbi->segs_per_sec > 1) {
2303 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2304 sizeof(struct sec_entry), GFP_KERNEL);
2305 if (!sit_i->sec_entries)
2309 /* get information related with SIT */
2310 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2312 /* setup SIT bitmap from ckeckpoint pack */
2313 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2314 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2316 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2320 /* init SIT information */
2321 sit_i->s_ops = &default_salloc_ops;
2323 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2324 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2325 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2326 sit_i->sit_bitmap = dst_bitmap;
2327 sit_i->bitmap_size = bitmap_size;
2328 sit_i->dirty_sentries = 0;
2329 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2330 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2331 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2332 mutex_init(&sit_i->sentry_lock);
2336 static int build_free_segmap(struct f2fs_sb_info *sbi)
2338 struct free_segmap_info *free_i;
2339 unsigned int bitmap_size, sec_bitmap_size;
2341 /* allocate memory for free segmap information */
2342 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2346 SM_I(sbi)->free_info = free_i;
2348 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2349 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2350 if (!free_i->free_segmap)
2353 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2354 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2355 if (!free_i->free_secmap)
2358 /* set all segments as dirty temporarily */
2359 memset(free_i->free_segmap, 0xff, bitmap_size);
2360 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2362 /* init free segmap information */
2363 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2364 free_i->free_segments = 0;
2365 free_i->free_sections = 0;
2366 spin_lock_init(&free_i->segmap_lock);
2370 static int build_curseg(struct f2fs_sb_info *sbi)
2372 struct curseg_info *array;
2375 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2379 SM_I(sbi)->curseg_array = array;
2381 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2382 mutex_init(&array[i].curseg_mutex);
2383 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2384 if (!array[i].sum_blk)
2386 init_rwsem(&array[i].journal_rwsem);
2387 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2389 if (!array[i].journal)
2391 array[i].segno = NULL_SEGNO;
2392 array[i].next_blkoff = 0;
2394 return restore_curseg_summaries(sbi);
2397 static void build_sit_entries(struct f2fs_sb_info *sbi)
2399 struct sit_info *sit_i = SIT_I(sbi);
2400 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2401 struct f2fs_journal *journal = curseg->journal;
2402 struct seg_entry *se;
2403 struct f2fs_sit_entry sit;
2404 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2405 unsigned int i, start, end;
2406 unsigned int readed, start_blk = 0;
2409 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2412 start = start_blk * sit_i->sents_per_block;
2413 end = (start_blk + readed) * sit_i->sents_per_block;
2415 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2416 struct f2fs_sit_block *sit_blk;
2419 se = &sit_i->sentries[start];
2420 page = get_current_sit_page(sbi, start);
2421 sit_blk = (struct f2fs_sit_block *)page_address(page);
2422 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2423 f2fs_put_page(page, 1);
2425 check_block_count(sbi, start, &sit);
2426 seg_info_from_raw_sit(se, &sit);
2428 /* build discard map only one time */
2429 if (f2fs_discard_en(sbi)) {
2430 memcpy(se->discard_map, se->cur_valid_map,
2431 SIT_VBLOCK_MAP_SIZE);
2432 sbi->discard_blks += sbi->blocks_per_seg -
2436 if (sbi->segs_per_sec > 1)
2437 get_sec_entry(sbi, start)->valid_blocks +=
2440 start_blk += readed;
2441 } while (start_blk < sit_blk_cnt);
2443 down_read(&curseg->journal_rwsem);
2444 for (i = 0; i < sits_in_cursum(journal); i++) {
2445 unsigned int old_valid_blocks;
2447 start = le32_to_cpu(segno_in_journal(journal, i));
2448 se = &sit_i->sentries[start];
2449 sit = sit_in_journal(journal, i);
2451 old_valid_blocks = se->valid_blocks;
2453 check_block_count(sbi, start, &sit);
2454 seg_info_from_raw_sit(se, &sit);
2456 if (f2fs_discard_en(sbi)) {
2457 memcpy(se->discard_map, se->cur_valid_map,
2458 SIT_VBLOCK_MAP_SIZE);
2459 sbi->discard_blks += old_valid_blocks -
2463 if (sbi->segs_per_sec > 1)
2464 get_sec_entry(sbi, start)->valid_blocks +=
2465 se->valid_blocks - old_valid_blocks;
2467 up_read(&curseg->journal_rwsem);
2470 static void init_free_segmap(struct f2fs_sb_info *sbi)
2475 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2476 struct seg_entry *sentry = get_seg_entry(sbi, start);
2477 if (!sentry->valid_blocks)
2478 __set_free(sbi, start);
2481 /* set use the current segments */
2482 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2483 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2484 __set_test_and_inuse(sbi, curseg_t->segno);
2488 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2490 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2491 struct free_segmap_info *free_i = FREE_I(sbi);
2492 unsigned int segno = 0, offset = 0;
2493 unsigned short valid_blocks;
2496 /* find dirty segment based on free segmap */
2497 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2498 if (segno >= MAIN_SEGS(sbi))
2501 valid_blocks = get_valid_blocks(sbi, segno, 0);
2502 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2504 if (valid_blocks > sbi->blocks_per_seg) {
2505 f2fs_bug_on(sbi, 1);
2508 mutex_lock(&dirty_i->seglist_lock);
2509 __locate_dirty_segment(sbi, segno, DIRTY);
2510 mutex_unlock(&dirty_i->seglist_lock);
2514 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2516 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2517 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2519 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2520 if (!dirty_i->victim_secmap)
2525 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2527 struct dirty_seglist_info *dirty_i;
2528 unsigned int bitmap_size, i;
2530 /* allocate memory for dirty segments list information */
2531 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2535 SM_I(sbi)->dirty_info = dirty_i;
2536 mutex_init(&dirty_i->seglist_lock);
2538 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2540 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2541 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2542 if (!dirty_i->dirty_segmap[i])
2546 init_dirty_segmap(sbi);
2547 return init_victim_secmap(sbi);
2551 * Update min, max modified time for cost-benefit GC algorithm
2553 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2555 struct sit_info *sit_i = SIT_I(sbi);
2558 mutex_lock(&sit_i->sentry_lock);
2560 sit_i->min_mtime = LLONG_MAX;
2562 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2564 unsigned long long mtime = 0;
2566 for (i = 0; i < sbi->segs_per_sec; i++)
2567 mtime += get_seg_entry(sbi, segno + i)->mtime;
2569 mtime = div_u64(mtime, sbi->segs_per_sec);
2571 if (sit_i->min_mtime > mtime)
2572 sit_i->min_mtime = mtime;
2574 sit_i->max_mtime = get_mtime(sbi);
2575 mutex_unlock(&sit_i->sentry_lock);
2578 int build_segment_manager(struct f2fs_sb_info *sbi)
2580 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2581 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2582 struct f2fs_sm_info *sm_info;
2585 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2590 sbi->sm_info = sm_info;
2591 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2592 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2593 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2594 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2595 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2596 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2597 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2598 sm_info->rec_prefree_segments = sm_info->main_segments *
2599 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2600 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2601 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2603 if (!test_opt(sbi, LFS))
2604 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2605 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2606 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2608 INIT_LIST_HEAD(&sm_info->discard_list);
2609 INIT_LIST_HEAD(&sm_info->wait_list);
2610 sm_info->nr_discards = 0;
2611 sm_info->max_discards = 0;
2613 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2615 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2617 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2618 err = create_flush_cmd_control(sbi);
2623 err = build_sit_info(sbi);
2626 err = build_free_segmap(sbi);
2629 err = build_curseg(sbi);
2633 /* reinit free segmap based on SIT */
2634 build_sit_entries(sbi);
2636 init_free_segmap(sbi);
2637 err = build_dirty_segmap(sbi);
2641 init_min_max_mtime(sbi);
2645 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2646 enum dirty_type dirty_type)
2648 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2650 mutex_lock(&dirty_i->seglist_lock);
2651 kvfree(dirty_i->dirty_segmap[dirty_type]);
2652 dirty_i->nr_dirty[dirty_type] = 0;
2653 mutex_unlock(&dirty_i->seglist_lock);
2656 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2658 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2659 kvfree(dirty_i->victim_secmap);
2662 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2664 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2670 /* discard pre-free/dirty segments list */
2671 for (i = 0; i < NR_DIRTY_TYPE; i++)
2672 discard_dirty_segmap(sbi, i);
2674 destroy_victim_secmap(sbi);
2675 SM_I(sbi)->dirty_info = NULL;
2679 static void destroy_curseg(struct f2fs_sb_info *sbi)
2681 struct curseg_info *array = SM_I(sbi)->curseg_array;
2686 SM_I(sbi)->curseg_array = NULL;
2687 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2688 kfree(array[i].sum_blk);
2689 kfree(array[i].journal);
2694 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2696 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2699 SM_I(sbi)->free_info = NULL;
2700 kvfree(free_i->free_segmap);
2701 kvfree(free_i->free_secmap);
2705 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2707 struct sit_info *sit_i = SIT_I(sbi);
2713 if (sit_i->sentries) {
2714 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2715 kfree(sit_i->sentries[start].cur_valid_map);
2716 kfree(sit_i->sentries[start].ckpt_valid_map);
2717 kfree(sit_i->sentries[start].discard_map);
2720 kfree(sit_i->tmp_map);
2722 kvfree(sit_i->sentries);
2723 kvfree(sit_i->sec_entries);
2724 kvfree(sit_i->dirty_sentries_bitmap);
2726 SM_I(sbi)->sit_info = NULL;
2727 kfree(sit_i->sit_bitmap);
2731 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2733 struct f2fs_sm_info *sm_info = SM_I(sbi);
2737 destroy_flush_cmd_control(sbi);
2738 destroy_dirty_segmap(sbi);
2739 destroy_curseg(sbi);
2740 destroy_free_segmap(sbi);
2741 destroy_sit_info(sbi);
2742 sbi->sm_info = NULL;
2746 int __init create_segment_manager_caches(void)
2748 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2749 sizeof(struct discard_entry));
2750 if (!discard_entry_slab)
2753 bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2754 sizeof(struct bio_entry));
2755 if (!bio_entry_slab)
2756 goto destroy_discard_entry;
2758 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2759 sizeof(struct sit_entry_set));
2760 if (!sit_entry_set_slab)
2761 goto destroy_bio_entry;
2763 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2764 sizeof(struct inmem_pages));
2765 if (!inmem_entry_slab)
2766 goto destroy_sit_entry_set;
2769 destroy_sit_entry_set:
2770 kmem_cache_destroy(sit_entry_set_slab);
2772 kmem_cache_destroy(bio_entry_slab);
2773 destroy_discard_entry:
2774 kmem_cache_destroy(discard_entry_slab);
2779 void destroy_segment_manager_caches(void)
2781 kmem_cache_destroy(sit_entry_set_slab);
2782 kmem_cache_destroy(bio_entry_slab);
2783 kmem_cache_destroy(discard_entry_slab);
2784 kmem_cache_destroy(inmem_entry_slab);