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>
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 *bio_entry_slab;
30 static struct kmem_cache *sit_entry_set_slab;
31 static struct kmem_cache *inmem_entry_slab;
33 static unsigned long __reverse_ulong(unsigned char *str)
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
38 #if BITS_PER_LONG == 64
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
52 static inline unsigned long __reverse_ffs(unsigned long word)
56 #if BITS_PER_LONG == 64
57 if ((word & 0xffffffff00000000UL) == 0)
62 if ((word & 0xffff0000) == 0)
67 if ((word & 0xff00) == 0)
72 if ((word & 0xf0) == 0)
77 if ((word & 0xc) == 0)
82 if ((word & 0x2) == 0)
88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
90 * @size must be integral times of unsigned long.
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
99 const unsigned long *p = addr + BIT_WORD(offset);
100 unsigned long result = size;
106 size -= (offset & ~(BITS_PER_LONG - 1));
107 offset %= BITS_PER_LONG;
113 tmp = __reverse_ulong((unsigned char *)p);
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
121 if (size <= BITS_PER_LONG)
123 size -= BITS_PER_LONG;
129 return result - size + __reverse_ffs(tmp);
132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
135 const unsigned long *p = addr + BIT_WORD(offset);
136 unsigned long result = size;
142 size -= (offset & ~(BITS_PER_LONG - 1));
143 offset %= BITS_PER_LONG;
149 tmp = __reverse_ulong((unsigned char *)p);
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
158 if (size <= BITS_PER_LONG)
160 size -= BITS_PER_LONG;
166 return result - size + __reverse_ffz(tmp);
169 void register_inmem_page(struct inode *inode, struct page *page)
171 struct f2fs_inode_info *fi = F2FS_I(inode);
172 struct inmem_pages *new;
174 f2fs_trace_pid(page);
176 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
177 SetPagePrivate(page);
179 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
181 /* add atomic page indices to the list */
183 INIT_LIST_HEAD(&new->list);
185 /* increase reference count with clean state */
186 mutex_lock(&fi->inmem_lock);
188 list_add_tail(&new->list, &fi->inmem_pages);
189 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
190 mutex_unlock(&fi->inmem_lock);
192 trace_f2fs_register_inmem_page(page, INMEM);
195 static int __revoke_inmem_pages(struct inode *inode,
196 struct list_head *head, bool drop, bool recover)
198 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
199 struct inmem_pages *cur, *tmp;
202 list_for_each_entry_safe(cur, tmp, head, list) {
203 struct page *page = cur->page;
206 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
211 struct dnode_of_data dn;
214 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
216 set_new_dnode(&dn, inode, NULL, NULL, 0);
217 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
221 get_node_info(sbi, dn.nid, &ni);
222 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
223 cur->old_addr, ni.version, true, true);
227 /* we don't need to invalidate this in the sccessful status */
229 ClearPageUptodate(page);
230 set_page_private(page, 0);
231 ClearPagePrivate(page);
232 f2fs_put_page(page, 1);
234 list_del(&cur->list);
235 kmem_cache_free(inmem_entry_slab, cur);
236 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
241 void drop_inmem_pages(struct inode *inode)
243 struct f2fs_inode_info *fi = F2FS_I(inode);
245 clear_inode_flag(inode, FI_ATOMIC_FILE);
247 mutex_lock(&fi->inmem_lock);
248 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
249 mutex_unlock(&fi->inmem_lock);
252 static int __commit_inmem_pages(struct inode *inode,
253 struct list_head *revoke_list)
255 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
256 struct f2fs_inode_info *fi = F2FS_I(inode);
257 struct inmem_pages *cur, *tmp;
258 struct f2fs_io_info fio = {
262 .op_flags = REQ_SYNC | REQ_PRIO,
263 .encrypted_page = NULL,
265 bool submit_bio = false;
268 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
269 struct page *page = cur->page;
272 if (page->mapping == inode->i_mapping) {
273 trace_f2fs_commit_inmem_page(page, INMEM);
275 set_page_dirty(page);
276 f2fs_wait_on_page_writeback(page, DATA, true);
277 if (clear_page_dirty_for_io(page)) {
278 inode_dec_dirty_pages(inode);
279 remove_dirty_inode(inode);
283 err = do_write_data_page(&fio);
289 /* record old blkaddr for revoking */
290 cur->old_addr = fio.old_blkaddr;
295 list_move_tail(&cur->list, revoke_list);
299 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
302 __revoke_inmem_pages(inode, revoke_list, false, false);
307 int commit_inmem_pages(struct inode *inode)
309 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
310 struct f2fs_inode_info *fi = F2FS_I(inode);
311 struct list_head revoke_list;
314 INIT_LIST_HEAD(&revoke_list);
315 f2fs_balance_fs(sbi, true);
318 mutex_lock(&fi->inmem_lock);
319 err = __commit_inmem_pages(inode, &revoke_list);
323 * try to revoke all committed pages, but still we could fail
324 * due to no memory or other reason, if that happened, EAGAIN
325 * will be returned, which means in such case, transaction is
326 * already not integrity, caller should use journal to do the
327 * recovery or rewrite & commit last transaction. For other
328 * error number, revoking was done by filesystem itself.
330 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
334 /* drop all uncommitted pages */
335 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
337 mutex_unlock(&fi->inmem_lock);
344 * This function balances dirty node and dentry pages.
345 * In addition, it controls garbage collection.
347 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
349 #ifdef CONFIG_F2FS_FAULT_INJECTION
350 if (time_to_inject(sbi, FAULT_CHECKPOINT))
351 f2fs_stop_checkpoint(sbi, false);
357 /* balance_fs_bg is able to be pending */
358 if (excess_cached_nats(sbi))
359 f2fs_balance_fs_bg(sbi);
362 * We should do GC or end up with checkpoint, if there are so many dirty
363 * dir/node pages without enough free segments.
365 if (has_not_enough_free_secs(sbi, 0, 0)) {
366 mutex_lock(&sbi->gc_mutex);
367 f2fs_gc(sbi, false, false);
371 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
373 /* try to shrink extent cache when there is no enough memory */
374 if (!available_free_memory(sbi, EXTENT_CACHE))
375 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
377 /* check the # of cached NAT entries */
378 if (!available_free_memory(sbi, NAT_ENTRIES))
379 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
381 if (!available_free_memory(sbi, FREE_NIDS))
382 try_to_free_nids(sbi, MAX_FREE_NIDS);
384 build_free_nids(sbi, false);
389 /* checkpoint is the only way to shrink partial cached entries */
390 if (!available_free_memory(sbi, NAT_ENTRIES) ||
391 !available_free_memory(sbi, INO_ENTRIES) ||
392 excess_prefree_segs(sbi) ||
393 excess_dirty_nats(sbi) ||
394 f2fs_time_over(sbi, CP_TIME)) {
395 if (test_opt(sbi, DATA_FLUSH)) {
396 struct blk_plug plug;
398 blk_start_plug(&plug);
399 sync_dirty_inodes(sbi, FILE_INODE);
400 blk_finish_plug(&plug);
402 f2fs_sync_fs(sbi->sb, true);
403 stat_inc_bg_cp_count(sbi->stat_info);
407 static int __submit_flush_wait(struct block_device *bdev)
409 struct bio *bio = f2fs_bio_alloc(0);
412 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
414 ret = submit_bio_wait(bio);
419 static int submit_flush_wait(struct f2fs_sb_info *sbi)
421 int ret = __submit_flush_wait(sbi->sb->s_bdev);
424 if (sbi->s_ndevs && !ret) {
425 for (i = 1; i < sbi->s_ndevs; i++) {
426 ret = __submit_flush_wait(FDEV(i).bdev);
434 static int issue_flush_thread(void *data)
436 struct f2fs_sb_info *sbi = data;
437 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
438 wait_queue_head_t *q = &fcc->flush_wait_queue;
440 if (kthread_should_stop())
443 if (!llist_empty(&fcc->issue_list)) {
444 struct flush_cmd *cmd, *next;
447 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
448 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
450 ret = submit_flush_wait(sbi);
451 llist_for_each_entry_safe(cmd, next,
452 fcc->dispatch_list, llnode) {
454 complete(&cmd->wait);
456 fcc->dispatch_list = NULL;
459 wait_event_interruptible(*q,
460 kthread_should_stop() || !llist_empty(&fcc->issue_list));
464 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
466 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
467 struct flush_cmd cmd;
469 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
470 test_opt(sbi, FLUSH_MERGE));
472 if (test_opt(sbi, NOBARRIER))
475 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
478 atomic_inc(&fcc->submit_flush);
479 ret = submit_flush_wait(sbi);
480 atomic_dec(&fcc->submit_flush);
484 init_completion(&cmd.wait);
486 atomic_inc(&fcc->submit_flush);
487 llist_add(&cmd.llnode, &fcc->issue_list);
489 if (!fcc->dispatch_list)
490 wake_up(&fcc->flush_wait_queue);
492 if (fcc->f2fs_issue_flush) {
493 wait_for_completion(&cmd.wait);
494 atomic_dec(&fcc->submit_flush);
496 llist_del_all(&fcc->issue_list);
497 atomic_set(&fcc->submit_flush, 0);
503 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
505 dev_t dev = sbi->sb->s_bdev->bd_dev;
506 struct flush_cmd_control *fcc;
509 if (SM_I(sbi)->cmd_control_info) {
510 fcc = SM_I(sbi)->cmd_control_info;
514 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
517 atomic_set(&fcc->submit_flush, 0);
518 init_waitqueue_head(&fcc->flush_wait_queue);
519 init_llist_head(&fcc->issue_list);
520 SM_I(sbi)->cmd_control_info = fcc;
522 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
523 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
524 if (IS_ERR(fcc->f2fs_issue_flush)) {
525 err = PTR_ERR(fcc->f2fs_issue_flush);
527 SM_I(sbi)->cmd_control_info = NULL;
534 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
536 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
538 if (fcc && fcc->f2fs_issue_flush) {
539 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
541 fcc->f2fs_issue_flush = NULL;
542 kthread_stop(flush_thread);
546 SM_I(sbi)->cmd_control_info = NULL;
550 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
551 enum dirty_type dirty_type)
553 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
555 /* need not be added */
556 if (IS_CURSEG(sbi, segno))
559 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
560 dirty_i->nr_dirty[dirty_type]++;
562 if (dirty_type == DIRTY) {
563 struct seg_entry *sentry = get_seg_entry(sbi, segno);
564 enum dirty_type t = sentry->type;
566 if (unlikely(t >= DIRTY)) {
570 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
571 dirty_i->nr_dirty[t]++;
575 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
576 enum dirty_type dirty_type)
578 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
580 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
581 dirty_i->nr_dirty[dirty_type]--;
583 if (dirty_type == DIRTY) {
584 struct seg_entry *sentry = get_seg_entry(sbi, segno);
585 enum dirty_type t = sentry->type;
587 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
588 dirty_i->nr_dirty[t]--;
590 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
591 clear_bit(GET_SECNO(sbi, segno),
592 dirty_i->victim_secmap);
597 * Should not occur error such as -ENOMEM.
598 * Adding dirty entry into seglist is not critical operation.
599 * If a given segment is one of current working segments, it won't be added.
601 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
603 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
604 unsigned short valid_blocks;
606 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
609 mutex_lock(&dirty_i->seglist_lock);
611 valid_blocks = get_valid_blocks(sbi, segno, 0);
613 if (valid_blocks == 0) {
614 __locate_dirty_segment(sbi, segno, PRE);
615 __remove_dirty_segment(sbi, segno, DIRTY);
616 } else if (valid_blocks < sbi->blocks_per_seg) {
617 __locate_dirty_segment(sbi, segno, DIRTY);
619 /* Recovery routine with SSR needs this */
620 __remove_dirty_segment(sbi, segno, DIRTY);
623 mutex_unlock(&dirty_i->seglist_lock);
626 static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
629 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
630 struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
632 INIT_LIST_HEAD(&be->list);
634 init_completion(&be->event);
635 list_add_tail(&be->list, wait_list);
640 void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
642 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
643 struct bio_entry *be, *tmp;
645 list_for_each_entry_safe(be, tmp, wait_list, list) {
646 struct bio *bio = be->bio;
649 wait_for_completion_io(&be->event);
651 if (err == -EOPNOTSUPP)
655 f2fs_msg(sbi->sb, KERN_INFO,
656 "Issue discard failed, ret: %d", err);
660 kmem_cache_free(bio_entry_slab, be);
664 static void f2fs_submit_bio_wait_endio(struct bio *bio)
666 struct bio_entry *be = (struct bio_entry *)bio->bi_private;
668 be->error = bio->bi_error;
669 complete(&be->event);
672 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
673 static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi,
674 struct block_device *bdev, block_t blkstart, block_t blklen)
676 struct bio *bio = NULL;
679 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
682 int devi = f2fs_target_device_index(sbi, blkstart);
684 blkstart -= FDEV(devi).start_blk;
686 err = __blkdev_issue_discard(bdev,
687 SECTOR_FROM_BLOCK(blkstart),
688 SECTOR_FROM_BLOCK(blklen),
691 struct bio_entry *be = __add_bio_entry(sbi, bio);
693 bio->bi_private = be;
694 bio->bi_end_io = f2fs_submit_bio_wait_endio;
695 bio->bi_opf |= REQ_SYNC;
702 #ifdef CONFIG_BLK_DEV_ZONED
703 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
704 struct block_device *bdev, block_t blkstart, block_t blklen)
706 sector_t nr_sects = SECTOR_FROM_BLOCK(blklen);
711 devi = f2fs_target_device_index(sbi, blkstart);
712 blkstart -= FDEV(devi).start_blk;
714 sector = SECTOR_FROM_BLOCK(blkstart);
716 if (sector & (bdev_zone_sectors(bdev) - 1) ||
717 nr_sects != bdev_zone_sectors(bdev)) {
718 f2fs_msg(sbi->sb, KERN_INFO,
719 "(%d) %s: Unaligned discard attempted (block %x + %x)",
720 devi, sbi->s_ndevs ? FDEV(devi).path: "",
726 * We need to know the type of the zone: for conventional zones,
727 * use regular discard if the drive supports it. For sequential
728 * zones, reset the zone write pointer.
730 switch (get_blkz_type(sbi, bdev, blkstart)) {
732 case BLK_ZONE_TYPE_CONVENTIONAL:
733 if (!blk_queue_discard(bdev_get_queue(bdev)))
735 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
736 case BLK_ZONE_TYPE_SEQWRITE_REQ:
737 case BLK_ZONE_TYPE_SEQWRITE_PREF:
738 trace_f2fs_issue_reset_zone(sbi->sb, blkstart);
739 return blkdev_reset_zones(bdev, sector,
742 /* Unknown zone type: broken device ? */
748 static int __issue_discard_async(struct f2fs_sb_info *sbi,
749 struct block_device *bdev, block_t blkstart, block_t blklen)
751 #ifdef CONFIG_BLK_DEV_ZONED
752 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
753 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
754 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
756 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
759 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
760 block_t blkstart, block_t blklen)
762 sector_t start = blkstart, len = 0;
763 struct block_device *bdev;
764 struct seg_entry *se;
769 bdev = f2fs_target_device(sbi, blkstart, NULL);
771 for (i = blkstart; i < blkstart + blklen; i++, len++) {
773 struct block_device *bdev2 =
774 f2fs_target_device(sbi, i, NULL);
777 err = __issue_discard_async(sbi, bdev,
787 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
788 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
790 if (!f2fs_test_and_set_bit(offset, se->discard_map))
795 err = __issue_discard_async(sbi, bdev, start, len);
799 static void __add_discard_entry(struct f2fs_sb_info *sbi,
800 struct cp_control *cpc, struct seg_entry *se,
801 unsigned int start, unsigned int end)
803 struct list_head *head = &SM_I(sbi)->discard_list;
804 struct discard_entry *new, *last;
806 if (!list_empty(head)) {
807 last = list_last_entry(head, struct discard_entry, list);
808 if (START_BLOCK(sbi, cpc->trim_start) + start ==
809 last->blkaddr + last->len) {
810 last->len += end - start;
815 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
816 INIT_LIST_HEAD(&new->list);
817 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
818 new->len = end - start;
819 list_add_tail(&new->list, head);
821 SM_I(sbi)->nr_discards += end - start;
824 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
826 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
827 int max_blocks = sbi->blocks_per_seg;
828 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
829 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
830 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
831 unsigned long *discard_map = (unsigned long *)se->discard_map;
832 unsigned long *dmap = SIT_I(sbi)->tmp_map;
833 unsigned int start = 0, end = -1;
834 bool force = (cpc->reason == CP_DISCARD);
837 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
841 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
842 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
846 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
847 for (i = 0; i < entries; i++)
848 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
849 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
851 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
852 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
853 if (start >= max_blocks)
856 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
857 if (force && start && end != max_blocks
858 && (end - start) < cpc->trim_minlen)
861 __add_discard_entry(sbi, cpc, se, start, end);
865 void release_discard_addrs(struct f2fs_sb_info *sbi)
867 struct list_head *head = &(SM_I(sbi)->discard_list);
868 struct discard_entry *entry, *this;
871 list_for_each_entry_safe(entry, this, head, list) {
872 list_del(&entry->list);
873 kmem_cache_free(discard_entry_slab, entry);
878 * Should call clear_prefree_segments after checkpoint is done.
880 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
882 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
885 mutex_lock(&dirty_i->seglist_lock);
886 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
887 __set_test_and_free(sbi, segno);
888 mutex_unlock(&dirty_i->seglist_lock);
891 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
893 struct list_head *head = &(SM_I(sbi)->discard_list);
894 struct discard_entry *entry, *this;
895 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
896 struct blk_plug plug;
897 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
898 unsigned int start = 0, end = -1;
899 unsigned int secno, start_segno;
900 bool force = (cpc->reason == CP_DISCARD);
902 blk_start_plug(&plug);
904 mutex_lock(&dirty_i->seglist_lock);
908 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
909 if (start >= MAIN_SEGS(sbi))
911 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
914 for (i = start; i < end; i++)
915 clear_bit(i, prefree_map);
917 dirty_i->nr_dirty[PRE] -= end - start;
919 if (!test_opt(sbi, DISCARD))
922 if (force && start >= cpc->trim_start &&
923 (end - 1) <= cpc->trim_end)
926 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
927 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
928 (end - start) << sbi->log_blocks_per_seg);
932 secno = GET_SECNO(sbi, start);
933 start_segno = secno * sbi->segs_per_sec;
934 if (!IS_CURSEC(sbi, secno) &&
935 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
936 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
937 sbi->segs_per_sec << sbi->log_blocks_per_seg);
939 start = start_segno + sbi->segs_per_sec;
943 mutex_unlock(&dirty_i->seglist_lock);
945 /* send small discards */
946 list_for_each_entry_safe(entry, this, head, list) {
947 if (force && entry->len < cpc->trim_minlen)
949 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
950 cpc->trimmed += entry->len;
952 list_del(&entry->list);
953 SM_I(sbi)->nr_discards -= entry->len;
954 kmem_cache_free(discard_entry_slab, entry);
957 blk_finish_plug(&plug);
960 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
962 struct sit_info *sit_i = SIT_I(sbi);
964 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
965 sit_i->dirty_sentries++;
972 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
973 unsigned int segno, int modified)
975 struct seg_entry *se = get_seg_entry(sbi, segno);
978 __mark_sit_entry_dirty(sbi, segno);
981 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
983 struct seg_entry *se;
984 unsigned int segno, offset;
985 long int new_vblocks;
987 segno = GET_SEGNO(sbi, blkaddr);
989 se = get_seg_entry(sbi, segno);
990 new_vblocks = se->valid_blocks + del;
991 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
993 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
994 (new_vblocks > sbi->blocks_per_seg)));
996 se->valid_blocks = new_vblocks;
997 se->mtime = get_mtime(sbi);
998 SIT_I(sbi)->max_mtime = se->mtime;
1000 /* Update valid block bitmap */
1002 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
1003 f2fs_bug_on(sbi, 1);
1004 if (f2fs_discard_en(sbi) &&
1005 !f2fs_test_and_set_bit(offset, se->discard_map))
1006 sbi->discard_blks--;
1008 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
1009 f2fs_bug_on(sbi, 1);
1010 if (f2fs_discard_en(sbi) &&
1011 f2fs_test_and_clear_bit(offset, se->discard_map))
1012 sbi->discard_blks++;
1014 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1015 se->ckpt_valid_blocks += del;
1017 __mark_sit_entry_dirty(sbi, segno);
1019 /* update total number of valid blocks to be written in ckpt area */
1020 SIT_I(sbi)->written_valid_blocks += del;
1022 if (sbi->segs_per_sec > 1)
1023 get_sec_entry(sbi, segno)->valid_blocks += del;
1026 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1028 update_sit_entry(sbi, new, 1);
1029 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1030 update_sit_entry(sbi, old, -1);
1032 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1033 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1036 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1038 unsigned int segno = GET_SEGNO(sbi, addr);
1039 struct sit_info *sit_i = SIT_I(sbi);
1041 f2fs_bug_on(sbi, addr == NULL_ADDR);
1042 if (addr == NEW_ADDR)
1045 /* add it into sit main buffer */
1046 mutex_lock(&sit_i->sentry_lock);
1048 update_sit_entry(sbi, addr, -1);
1050 /* add it into dirty seglist */
1051 locate_dirty_segment(sbi, segno);
1053 mutex_unlock(&sit_i->sentry_lock);
1056 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1058 struct sit_info *sit_i = SIT_I(sbi);
1059 unsigned int segno, offset;
1060 struct seg_entry *se;
1063 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1066 mutex_lock(&sit_i->sentry_lock);
1068 segno = GET_SEGNO(sbi, blkaddr);
1069 se = get_seg_entry(sbi, segno);
1070 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1072 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1075 mutex_unlock(&sit_i->sentry_lock);
1081 * This function should be resided under the curseg_mutex lock
1083 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1084 struct f2fs_summary *sum)
1086 struct curseg_info *curseg = CURSEG_I(sbi, type);
1087 void *addr = curseg->sum_blk;
1088 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1089 memcpy(addr, sum, sizeof(struct f2fs_summary));
1093 * Calculate the number of current summary pages for writing
1095 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1097 int valid_sum_count = 0;
1100 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1101 if (sbi->ckpt->alloc_type[i] == SSR)
1102 valid_sum_count += sbi->blocks_per_seg;
1105 valid_sum_count += le16_to_cpu(
1106 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1108 valid_sum_count += curseg_blkoff(sbi, i);
1112 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1113 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1114 if (valid_sum_count <= sum_in_page)
1116 else if ((valid_sum_count - sum_in_page) <=
1117 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1123 * Caller should put this summary page
1125 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1127 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1130 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1132 struct page *page = grab_meta_page(sbi, blk_addr);
1133 void *dst = page_address(page);
1136 memcpy(dst, src, PAGE_SIZE);
1138 memset(dst, 0, PAGE_SIZE);
1139 set_page_dirty(page);
1140 f2fs_put_page(page, 1);
1143 static void write_sum_page(struct f2fs_sb_info *sbi,
1144 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1146 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1149 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1150 int type, block_t blk_addr)
1152 struct curseg_info *curseg = CURSEG_I(sbi, type);
1153 struct page *page = grab_meta_page(sbi, blk_addr);
1154 struct f2fs_summary_block *src = curseg->sum_blk;
1155 struct f2fs_summary_block *dst;
1157 dst = (struct f2fs_summary_block *)page_address(page);
1159 mutex_lock(&curseg->curseg_mutex);
1161 down_read(&curseg->journal_rwsem);
1162 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1163 up_read(&curseg->journal_rwsem);
1165 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1166 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1168 mutex_unlock(&curseg->curseg_mutex);
1170 set_page_dirty(page);
1171 f2fs_put_page(page, 1);
1174 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1176 struct curseg_info *curseg = CURSEG_I(sbi, type);
1177 unsigned int segno = curseg->segno + 1;
1178 struct free_segmap_info *free_i = FREE_I(sbi);
1180 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1181 return !test_bit(segno, free_i->free_segmap);
1186 * Find a new segment from the free segments bitmap to right order
1187 * This function should be returned with success, otherwise BUG
1189 static void get_new_segment(struct f2fs_sb_info *sbi,
1190 unsigned int *newseg, bool new_sec, int dir)
1192 struct free_segmap_info *free_i = FREE_I(sbi);
1193 unsigned int segno, secno, zoneno;
1194 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1195 unsigned int hint = *newseg / sbi->segs_per_sec;
1196 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1197 unsigned int left_start = hint;
1202 spin_lock(&free_i->segmap_lock);
1204 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1205 segno = find_next_zero_bit(free_i->free_segmap,
1206 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1207 if (segno < (hint + 1) * sbi->segs_per_sec)
1211 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1212 if (secno >= MAIN_SECS(sbi)) {
1213 if (dir == ALLOC_RIGHT) {
1214 secno = find_next_zero_bit(free_i->free_secmap,
1216 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1219 left_start = hint - 1;
1225 while (test_bit(left_start, free_i->free_secmap)) {
1226 if (left_start > 0) {
1230 left_start = find_next_zero_bit(free_i->free_secmap,
1232 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1238 segno = secno * sbi->segs_per_sec;
1239 zoneno = secno / sbi->secs_per_zone;
1241 /* give up on finding another zone */
1244 if (sbi->secs_per_zone == 1)
1246 if (zoneno == old_zoneno)
1248 if (dir == ALLOC_LEFT) {
1249 if (!go_left && zoneno + 1 >= total_zones)
1251 if (go_left && zoneno == 0)
1254 for (i = 0; i < NR_CURSEG_TYPE; i++)
1255 if (CURSEG_I(sbi, i)->zone == zoneno)
1258 if (i < NR_CURSEG_TYPE) {
1259 /* zone is in user, try another */
1261 hint = zoneno * sbi->secs_per_zone - 1;
1262 else if (zoneno + 1 >= total_zones)
1265 hint = (zoneno + 1) * sbi->secs_per_zone;
1267 goto find_other_zone;
1270 /* set it as dirty segment in free segmap */
1271 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1272 __set_inuse(sbi, segno);
1274 spin_unlock(&free_i->segmap_lock);
1277 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1279 struct curseg_info *curseg = CURSEG_I(sbi, type);
1280 struct summary_footer *sum_footer;
1282 curseg->segno = curseg->next_segno;
1283 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1284 curseg->next_blkoff = 0;
1285 curseg->next_segno = NULL_SEGNO;
1287 sum_footer = &(curseg->sum_blk->footer);
1288 memset(sum_footer, 0, sizeof(struct summary_footer));
1289 if (IS_DATASEG(type))
1290 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1291 if (IS_NODESEG(type))
1292 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1293 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1297 * Allocate a current working segment.
1298 * This function always allocates a free segment in LFS manner.
1300 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1302 struct curseg_info *curseg = CURSEG_I(sbi, type);
1303 unsigned int segno = curseg->segno;
1304 int dir = ALLOC_LEFT;
1306 write_sum_page(sbi, curseg->sum_blk,
1307 GET_SUM_BLOCK(sbi, segno));
1308 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1311 if (test_opt(sbi, NOHEAP))
1314 get_new_segment(sbi, &segno, new_sec, dir);
1315 curseg->next_segno = segno;
1316 reset_curseg(sbi, type, 1);
1317 curseg->alloc_type = LFS;
1320 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1321 struct curseg_info *seg, block_t start)
1323 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1324 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1325 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1326 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1327 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1330 for (i = 0; i < entries; i++)
1331 target_map[i] = ckpt_map[i] | cur_map[i];
1333 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1335 seg->next_blkoff = pos;
1339 * If a segment is written by LFS manner, next block offset is just obtained
1340 * by increasing the current block offset. However, if a segment is written by
1341 * SSR manner, next block offset obtained by calling __next_free_blkoff
1343 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1344 struct curseg_info *seg)
1346 if (seg->alloc_type == SSR)
1347 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1353 * This function always allocates a used segment(from dirty seglist) by SSR
1354 * manner, so it should recover the existing segment information of valid blocks
1356 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1358 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1359 struct curseg_info *curseg = CURSEG_I(sbi, type);
1360 unsigned int new_segno = curseg->next_segno;
1361 struct f2fs_summary_block *sum_node;
1362 struct page *sum_page;
1364 write_sum_page(sbi, curseg->sum_blk,
1365 GET_SUM_BLOCK(sbi, curseg->segno));
1366 __set_test_and_inuse(sbi, new_segno);
1368 mutex_lock(&dirty_i->seglist_lock);
1369 __remove_dirty_segment(sbi, new_segno, PRE);
1370 __remove_dirty_segment(sbi, new_segno, DIRTY);
1371 mutex_unlock(&dirty_i->seglist_lock);
1373 reset_curseg(sbi, type, 1);
1374 curseg->alloc_type = SSR;
1375 __next_free_blkoff(sbi, curseg, 0);
1378 sum_page = get_sum_page(sbi, new_segno);
1379 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1380 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1381 f2fs_put_page(sum_page, 1);
1385 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1387 struct curseg_info *curseg = CURSEG_I(sbi, type);
1388 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1390 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0, 0))
1391 return v_ops->get_victim(sbi,
1392 &(curseg)->next_segno, BG_GC, type, SSR);
1394 /* For data segments, let's do SSR more intensively */
1395 for (; type >= CURSEG_HOT_DATA; type--)
1396 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1403 * flush out current segment and replace it with new segment
1404 * This function should be returned with success, otherwise BUG
1406 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1407 int type, bool force)
1409 struct curseg_info *curseg = CURSEG_I(sbi, type);
1412 new_curseg(sbi, type, true);
1413 else if (type == CURSEG_WARM_NODE)
1414 new_curseg(sbi, type, false);
1415 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1416 new_curseg(sbi, type, false);
1417 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1418 change_curseg(sbi, type, true);
1420 new_curseg(sbi, type, false);
1422 stat_inc_seg_type(sbi, curseg);
1425 void allocate_new_segments(struct f2fs_sb_info *sbi)
1427 struct curseg_info *curseg;
1428 unsigned int old_segno;
1431 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1432 curseg = CURSEG_I(sbi, i);
1433 old_segno = curseg->segno;
1434 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1435 locate_dirty_segment(sbi, old_segno);
1439 static const struct segment_allocation default_salloc_ops = {
1440 .allocate_segment = allocate_segment_by_default,
1443 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1445 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1446 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1447 unsigned int start_segno, end_segno;
1448 struct cp_control cpc;
1451 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1455 if (end <= MAIN_BLKADDR(sbi))
1458 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1459 f2fs_msg(sbi->sb, KERN_WARNING,
1460 "Found FS corruption, run fsck to fix.");
1464 /* start/end segment number in main_area */
1465 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1466 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1467 GET_SEGNO(sbi, end);
1468 cpc.reason = CP_DISCARD;
1469 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1471 /* do checkpoint to issue discard commands safely */
1472 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1473 cpc.trim_start = start_segno;
1475 if (sbi->discard_blks == 0)
1477 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1478 cpc.trim_end = end_segno;
1480 cpc.trim_end = min_t(unsigned int,
1481 rounddown(start_segno +
1482 BATCHED_TRIM_SEGMENTS(sbi),
1483 sbi->segs_per_sec) - 1, end_segno);
1485 mutex_lock(&sbi->gc_mutex);
1486 err = write_checkpoint(sbi, &cpc);
1487 mutex_unlock(&sbi->gc_mutex);
1494 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1498 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1500 struct curseg_info *curseg = CURSEG_I(sbi, type);
1501 if (curseg->next_blkoff < sbi->blocks_per_seg)
1506 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1509 return CURSEG_HOT_DATA;
1511 return CURSEG_HOT_NODE;
1514 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1516 if (p_type == DATA) {
1517 struct inode *inode = page->mapping->host;
1519 if (S_ISDIR(inode->i_mode))
1520 return CURSEG_HOT_DATA;
1522 return CURSEG_COLD_DATA;
1524 if (IS_DNODE(page) && is_cold_node(page))
1525 return CURSEG_WARM_NODE;
1527 return CURSEG_COLD_NODE;
1531 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1533 if (p_type == DATA) {
1534 struct inode *inode = page->mapping->host;
1536 if (S_ISDIR(inode->i_mode))
1537 return CURSEG_HOT_DATA;
1538 else if (is_cold_data(page) || file_is_cold(inode))
1539 return CURSEG_COLD_DATA;
1541 return CURSEG_WARM_DATA;
1544 return is_cold_node(page) ? CURSEG_WARM_NODE :
1547 return CURSEG_COLD_NODE;
1551 static int __get_segment_type(struct page *page, enum page_type p_type)
1553 switch (F2FS_P_SB(page)->active_logs) {
1555 return __get_segment_type_2(page, p_type);
1557 return __get_segment_type_4(page, p_type);
1559 /* NR_CURSEG_TYPE(6) logs by default */
1560 f2fs_bug_on(F2FS_P_SB(page),
1561 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1562 return __get_segment_type_6(page, p_type);
1565 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1566 block_t old_blkaddr, block_t *new_blkaddr,
1567 struct f2fs_summary *sum, int type)
1569 struct sit_info *sit_i = SIT_I(sbi);
1570 struct curseg_info *curseg = CURSEG_I(sbi, type);
1572 mutex_lock(&curseg->curseg_mutex);
1573 mutex_lock(&sit_i->sentry_lock);
1575 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1578 * __add_sum_entry should be resided under the curseg_mutex
1579 * because, this function updates a summary entry in the
1580 * current summary block.
1582 __add_sum_entry(sbi, type, sum);
1584 __refresh_next_blkoff(sbi, curseg);
1586 stat_inc_block_count(sbi, curseg);
1588 if (!__has_curseg_space(sbi, type))
1589 sit_i->s_ops->allocate_segment(sbi, type, false);
1591 * SIT information should be updated before segment allocation,
1592 * since SSR needs latest valid block information.
1594 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1596 mutex_unlock(&sit_i->sentry_lock);
1598 if (page && IS_NODESEG(type))
1599 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1601 mutex_unlock(&curseg->curseg_mutex);
1604 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1606 int type = __get_segment_type(fio->page, fio->type);
1608 if (fio->type == NODE || fio->type == DATA)
1609 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1611 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1612 &fio->new_blkaddr, sum, type);
1614 /* writeout dirty page into bdev */
1615 f2fs_submit_page_mbio(fio);
1617 if (fio->type == NODE || fio->type == DATA)
1618 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1621 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1623 struct f2fs_io_info fio = {
1627 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
1628 .old_blkaddr = page->index,
1629 .new_blkaddr = page->index,
1631 .encrypted_page = NULL,
1634 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1635 fio.op_flags &= ~REQ_META;
1637 set_page_writeback(page);
1638 f2fs_submit_page_mbio(&fio);
1641 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1643 struct f2fs_summary sum;
1645 set_summary(&sum, nid, 0, 0);
1646 do_write_page(&sum, fio);
1649 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1651 struct f2fs_sb_info *sbi = fio->sbi;
1652 struct f2fs_summary sum;
1653 struct node_info ni;
1655 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1656 get_node_info(sbi, dn->nid, &ni);
1657 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1658 do_write_page(&sum, fio);
1659 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1662 void rewrite_data_page(struct f2fs_io_info *fio)
1664 fio->new_blkaddr = fio->old_blkaddr;
1665 stat_inc_inplace_blocks(fio->sbi);
1666 f2fs_submit_page_mbio(fio);
1669 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1670 block_t old_blkaddr, block_t new_blkaddr,
1671 bool recover_curseg, bool recover_newaddr)
1673 struct sit_info *sit_i = SIT_I(sbi);
1674 struct curseg_info *curseg;
1675 unsigned int segno, old_cursegno;
1676 struct seg_entry *se;
1678 unsigned short old_blkoff;
1680 segno = GET_SEGNO(sbi, new_blkaddr);
1681 se = get_seg_entry(sbi, segno);
1684 if (!recover_curseg) {
1685 /* for recovery flow */
1686 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1687 if (old_blkaddr == NULL_ADDR)
1688 type = CURSEG_COLD_DATA;
1690 type = CURSEG_WARM_DATA;
1693 if (!IS_CURSEG(sbi, segno))
1694 type = CURSEG_WARM_DATA;
1697 curseg = CURSEG_I(sbi, type);
1699 mutex_lock(&curseg->curseg_mutex);
1700 mutex_lock(&sit_i->sentry_lock);
1702 old_cursegno = curseg->segno;
1703 old_blkoff = curseg->next_blkoff;
1705 /* change the current segment */
1706 if (segno != curseg->segno) {
1707 curseg->next_segno = segno;
1708 change_curseg(sbi, type, true);
1711 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1712 __add_sum_entry(sbi, type, sum);
1714 if (!recover_curseg || recover_newaddr)
1715 update_sit_entry(sbi, new_blkaddr, 1);
1716 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1717 update_sit_entry(sbi, old_blkaddr, -1);
1719 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1720 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1722 locate_dirty_segment(sbi, old_cursegno);
1724 if (recover_curseg) {
1725 if (old_cursegno != curseg->segno) {
1726 curseg->next_segno = old_cursegno;
1727 change_curseg(sbi, type, true);
1729 curseg->next_blkoff = old_blkoff;
1732 mutex_unlock(&sit_i->sentry_lock);
1733 mutex_unlock(&curseg->curseg_mutex);
1736 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1737 block_t old_addr, block_t new_addr,
1738 unsigned char version, bool recover_curseg,
1739 bool recover_newaddr)
1741 struct f2fs_summary sum;
1743 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1745 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1746 recover_curseg, recover_newaddr);
1748 f2fs_update_data_blkaddr(dn, new_addr);
1751 void f2fs_wait_on_page_writeback(struct page *page,
1752 enum page_type type, bool ordered)
1754 if (PageWriteback(page)) {
1755 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1757 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1759 wait_on_page_writeback(page);
1761 wait_for_stable_page(page);
1765 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1770 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1773 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1775 f2fs_wait_on_page_writeback(cpage, DATA, true);
1776 f2fs_put_page(cpage, 1);
1780 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1782 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1783 struct curseg_info *seg_i;
1784 unsigned char *kaddr;
1789 start = start_sum_block(sbi);
1791 page = get_meta_page(sbi, start++);
1792 kaddr = (unsigned char *)page_address(page);
1794 /* Step 1: restore nat cache */
1795 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1796 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1798 /* Step 2: restore sit cache */
1799 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1800 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1801 offset = 2 * SUM_JOURNAL_SIZE;
1803 /* Step 3: restore summary entries */
1804 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1805 unsigned short blk_off;
1808 seg_i = CURSEG_I(sbi, i);
1809 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1810 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1811 seg_i->next_segno = segno;
1812 reset_curseg(sbi, i, 0);
1813 seg_i->alloc_type = ckpt->alloc_type[i];
1814 seg_i->next_blkoff = blk_off;
1816 if (seg_i->alloc_type == SSR)
1817 blk_off = sbi->blocks_per_seg;
1819 for (j = 0; j < blk_off; j++) {
1820 struct f2fs_summary *s;
1821 s = (struct f2fs_summary *)(kaddr + offset);
1822 seg_i->sum_blk->entries[j] = *s;
1823 offset += SUMMARY_SIZE;
1824 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1828 f2fs_put_page(page, 1);
1831 page = get_meta_page(sbi, start++);
1832 kaddr = (unsigned char *)page_address(page);
1836 f2fs_put_page(page, 1);
1840 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1842 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1843 struct f2fs_summary_block *sum;
1844 struct curseg_info *curseg;
1846 unsigned short blk_off;
1847 unsigned int segno = 0;
1848 block_t blk_addr = 0;
1850 /* get segment number and block addr */
1851 if (IS_DATASEG(type)) {
1852 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1853 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1855 if (__exist_node_summaries(sbi))
1856 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1858 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1860 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1862 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1864 if (__exist_node_summaries(sbi))
1865 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1866 type - CURSEG_HOT_NODE);
1868 blk_addr = GET_SUM_BLOCK(sbi, segno);
1871 new = get_meta_page(sbi, blk_addr);
1872 sum = (struct f2fs_summary_block *)page_address(new);
1874 if (IS_NODESEG(type)) {
1875 if (__exist_node_summaries(sbi)) {
1876 struct f2fs_summary *ns = &sum->entries[0];
1878 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1880 ns->ofs_in_node = 0;
1885 err = restore_node_summary(sbi, segno, sum);
1887 f2fs_put_page(new, 1);
1893 /* set uncompleted segment to curseg */
1894 curseg = CURSEG_I(sbi, type);
1895 mutex_lock(&curseg->curseg_mutex);
1897 /* update journal info */
1898 down_write(&curseg->journal_rwsem);
1899 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1900 up_write(&curseg->journal_rwsem);
1902 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1903 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1904 curseg->next_segno = segno;
1905 reset_curseg(sbi, type, 0);
1906 curseg->alloc_type = ckpt->alloc_type[type];
1907 curseg->next_blkoff = blk_off;
1908 mutex_unlock(&curseg->curseg_mutex);
1909 f2fs_put_page(new, 1);
1913 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1915 int type = CURSEG_HOT_DATA;
1918 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
1919 int npages = npages_for_summary_flush(sbi, true);
1922 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1925 /* restore for compacted data summary */
1926 if (read_compacted_summaries(sbi))
1928 type = CURSEG_HOT_NODE;
1931 if (__exist_node_summaries(sbi))
1932 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1933 NR_CURSEG_TYPE - type, META_CP, true);
1935 for (; type <= CURSEG_COLD_NODE; type++) {
1936 err = read_normal_summaries(sbi, type);
1944 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1947 unsigned char *kaddr;
1948 struct f2fs_summary *summary;
1949 struct curseg_info *seg_i;
1950 int written_size = 0;
1953 page = grab_meta_page(sbi, blkaddr++);
1954 kaddr = (unsigned char *)page_address(page);
1956 /* Step 1: write nat cache */
1957 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1958 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1959 written_size += SUM_JOURNAL_SIZE;
1961 /* Step 2: write sit cache */
1962 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1963 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1964 written_size += SUM_JOURNAL_SIZE;
1966 /* Step 3: write summary entries */
1967 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1968 unsigned short blkoff;
1969 seg_i = CURSEG_I(sbi, i);
1970 if (sbi->ckpt->alloc_type[i] == SSR)
1971 blkoff = sbi->blocks_per_seg;
1973 blkoff = curseg_blkoff(sbi, i);
1975 for (j = 0; j < blkoff; j++) {
1977 page = grab_meta_page(sbi, blkaddr++);
1978 kaddr = (unsigned char *)page_address(page);
1981 summary = (struct f2fs_summary *)(kaddr + written_size);
1982 *summary = seg_i->sum_blk->entries[j];
1983 written_size += SUMMARY_SIZE;
1985 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1989 set_page_dirty(page);
1990 f2fs_put_page(page, 1);
1995 set_page_dirty(page);
1996 f2fs_put_page(page, 1);
2000 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2001 block_t blkaddr, int type)
2004 if (IS_DATASEG(type))
2005 end = type + NR_CURSEG_DATA_TYPE;
2007 end = type + NR_CURSEG_NODE_TYPE;
2009 for (i = type; i < end; i++)
2010 write_current_sum_page(sbi, i, blkaddr + (i - type));
2013 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2015 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2016 write_compacted_summaries(sbi, start_blk);
2018 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2021 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2023 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2026 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2027 unsigned int val, int alloc)
2031 if (type == NAT_JOURNAL) {
2032 for (i = 0; i < nats_in_cursum(journal); i++) {
2033 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2036 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2037 return update_nats_in_cursum(journal, 1);
2038 } else if (type == SIT_JOURNAL) {
2039 for (i = 0; i < sits_in_cursum(journal); i++)
2040 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2042 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2043 return update_sits_in_cursum(journal, 1);
2048 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2051 return get_meta_page(sbi, current_sit_addr(sbi, segno));
2054 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2057 struct sit_info *sit_i = SIT_I(sbi);
2058 struct page *src_page, *dst_page;
2059 pgoff_t src_off, dst_off;
2060 void *src_addr, *dst_addr;
2062 src_off = current_sit_addr(sbi, start);
2063 dst_off = next_sit_addr(sbi, src_off);
2065 /* get current sit block page without lock */
2066 src_page = get_meta_page(sbi, src_off);
2067 dst_page = grab_meta_page(sbi, dst_off);
2068 f2fs_bug_on(sbi, PageDirty(src_page));
2070 src_addr = page_address(src_page);
2071 dst_addr = page_address(dst_page);
2072 memcpy(dst_addr, src_addr, PAGE_SIZE);
2074 set_page_dirty(dst_page);
2075 f2fs_put_page(src_page, 1);
2077 set_to_next_sit(sit_i, start);
2082 static struct sit_entry_set *grab_sit_entry_set(void)
2084 struct sit_entry_set *ses =
2085 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2088 INIT_LIST_HEAD(&ses->set_list);
2092 static void release_sit_entry_set(struct sit_entry_set *ses)
2094 list_del(&ses->set_list);
2095 kmem_cache_free(sit_entry_set_slab, ses);
2098 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2099 struct list_head *head)
2101 struct sit_entry_set *next = ses;
2103 if (list_is_last(&ses->set_list, head))
2106 list_for_each_entry_continue(next, head, set_list)
2107 if (ses->entry_cnt <= next->entry_cnt)
2110 list_move_tail(&ses->set_list, &next->set_list);
2113 static void add_sit_entry(unsigned int segno, struct list_head *head)
2115 struct sit_entry_set *ses;
2116 unsigned int start_segno = START_SEGNO(segno);
2118 list_for_each_entry(ses, head, set_list) {
2119 if (ses->start_segno == start_segno) {
2121 adjust_sit_entry_set(ses, head);
2126 ses = grab_sit_entry_set();
2128 ses->start_segno = start_segno;
2130 list_add(&ses->set_list, head);
2133 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2135 struct f2fs_sm_info *sm_info = SM_I(sbi);
2136 struct list_head *set_list = &sm_info->sit_entry_set;
2137 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2140 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2141 add_sit_entry(segno, set_list);
2144 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2146 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2147 struct f2fs_journal *journal = curseg->journal;
2150 down_write(&curseg->journal_rwsem);
2151 for (i = 0; i < sits_in_cursum(journal); i++) {
2155 segno = le32_to_cpu(segno_in_journal(journal, i));
2156 dirtied = __mark_sit_entry_dirty(sbi, segno);
2159 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2161 update_sits_in_cursum(journal, -i);
2162 up_write(&curseg->journal_rwsem);
2166 * CP calls this function, which flushes SIT entries including sit_journal,
2167 * and moves prefree segs to free segs.
2169 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2171 struct sit_info *sit_i = SIT_I(sbi);
2172 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2173 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2174 struct f2fs_journal *journal = curseg->journal;
2175 struct sit_entry_set *ses, *tmp;
2176 struct list_head *head = &SM_I(sbi)->sit_entry_set;
2177 bool to_journal = true;
2178 struct seg_entry *se;
2180 mutex_lock(&sit_i->sentry_lock);
2182 if (!sit_i->dirty_sentries)
2186 * add and account sit entries of dirty bitmap in sit entry
2189 add_sits_in_set(sbi);
2192 * if there are no enough space in journal to store dirty sit
2193 * entries, remove all entries from journal and add and account
2194 * them in sit entry set.
2196 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2197 remove_sits_in_journal(sbi);
2200 * there are two steps to flush sit entries:
2201 * #1, flush sit entries to journal in current cold data summary block.
2202 * #2, flush sit entries to sit page.
2204 list_for_each_entry_safe(ses, tmp, head, set_list) {
2205 struct page *page = NULL;
2206 struct f2fs_sit_block *raw_sit = NULL;
2207 unsigned int start_segno = ses->start_segno;
2208 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2209 (unsigned long)MAIN_SEGS(sbi));
2210 unsigned int segno = start_segno;
2213 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2217 down_write(&curseg->journal_rwsem);
2219 page = get_next_sit_page(sbi, start_segno);
2220 raw_sit = page_address(page);
2223 /* flush dirty sit entries in region of current sit set */
2224 for_each_set_bit_from(segno, bitmap, end) {
2225 int offset, sit_offset;
2227 se = get_seg_entry(sbi, segno);
2229 /* add discard candidates */
2230 if (cpc->reason != CP_DISCARD) {
2231 cpc->trim_start = segno;
2232 add_discard_addrs(sbi, cpc);
2236 offset = lookup_journal_in_cursum(journal,
2237 SIT_JOURNAL, segno, 1);
2238 f2fs_bug_on(sbi, offset < 0);
2239 segno_in_journal(journal, offset) =
2241 seg_info_to_raw_sit(se,
2242 &sit_in_journal(journal, offset));
2244 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2245 seg_info_to_raw_sit(se,
2246 &raw_sit->entries[sit_offset]);
2249 __clear_bit(segno, bitmap);
2250 sit_i->dirty_sentries--;
2255 up_write(&curseg->journal_rwsem);
2257 f2fs_put_page(page, 1);
2259 f2fs_bug_on(sbi, ses->entry_cnt);
2260 release_sit_entry_set(ses);
2263 f2fs_bug_on(sbi, !list_empty(head));
2264 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2266 if (cpc->reason == CP_DISCARD) {
2267 __u64 trim_start = cpc->trim_start;
2269 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2270 add_discard_addrs(sbi, cpc);
2272 cpc->trim_start = trim_start;
2274 mutex_unlock(&sit_i->sentry_lock);
2276 set_prefree_as_free_segments(sbi);
2279 static int build_sit_info(struct f2fs_sb_info *sbi)
2281 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2282 struct sit_info *sit_i;
2283 unsigned int sit_segs, start;
2284 char *src_bitmap, *dst_bitmap;
2285 unsigned int bitmap_size;
2287 /* allocate memory for SIT information */
2288 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2292 SM_I(sbi)->sit_info = sit_i;
2294 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2295 sizeof(struct seg_entry), GFP_KERNEL);
2296 if (!sit_i->sentries)
2299 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2300 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2301 if (!sit_i->dirty_sentries_bitmap)
2304 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2305 sit_i->sentries[start].cur_valid_map
2306 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2307 sit_i->sentries[start].ckpt_valid_map
2308 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2309 if (!sit_i->sentries[start].cur_valid_map ||
2310 !sit_i->sentries[start].ckpt_valid_map)
2313 if (f2fs_discard_en(sbi)) {
2314 sit_i->sentries[start].discard_map
2315 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2316 if (!sit_i->sentries[start].discard_map)
2321 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2322 if (!sit_i->tmp_map)
2325 if (sbi->segs_per_sec > 1) {
2326 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2327 sizeof(struct sec_entry), GFP_KERNEL);
2328 if (!sit_i->sec_entries)
2332 /* get information related with SIT */
2333 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2335 /* setup SIT bitmap from ckeckpoint pack */
2336 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2337 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2339 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2343 /* init SIT information */
2344 sit_i->s_ops = &default_salloc_ops;
2346 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2347 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2348 sit_i->written_valid_blocks = 0;
2349 sit_i->sit_bitmap = dst_bitmap;
2350 sit_i->bitmap_size = bitmap_size;
2351 sit_i->dirty_sentries = 0;
2352 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2353 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2354 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2355 mutex_init(&sit_i->sentry_lock);
2359 static int build_free_segmap(struct f2fs_sb_info *sbi)
2361 struct free_segmap_info *free_i;
2362 unsigned int bitmap_size, sec_bitmap_size;
2364 /* allocate memory for free segmap information */
2365 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2369 SM_I(sbi)->free_info = free_i;
2371 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2372 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2373 if (!free_i->free_segmap)
2376 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2377 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2378 if (!free_i->free_secmap)
2381 /* set all segments as dirty temporarily */
2382 memset(free_i->free_segmap, 0xff, bitmap_size);
2383 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2385 /* init free segmap information */
2386 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2387 free_i->free_segments = 0;
2388 free_i->free_sections = 0;
2389 spin_lock_init(&free_i->segmap_lock);
2393 static int build_curseg(struct f2fs_sb_info *sbi)
2395 struct curseg_info *array;
2398 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2402 SM_I(sbi)->curseg_array = array;
2404 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2405 mutex_init(&array[i].curseg_mutex);
2406 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2407 if (!array[i].sum_blk)
2409 init_rwsem(&array[i].journal_rwsem);
2410 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2412 if (!array[i].journal)
2414 array[i].segno = NULL_SEGNO;
2415 array[i].next_blkoff = 0;
2417 return restore_curseg_summaries(sbi);
2420 static void build_sit_entries(struct f2fs_sb_info *sbi)
2422 struct sit_info *sit_i = SIT_I(sbi);
2423 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2424 struct f2fs_journal *journal = curseg->journal;
2425 struct seg_entry *se;
2426 struct f2fs_sit_entry sit;
2427 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2428 unsigned int i, start, end;
2429 unsigned int readed, start_blk = 0;
2432 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2435 start = start_blk * sit_i->sents_per_block;
2436 end = (start_blk + readed) * sit_i->sents_per_block;
2438 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2439 struct f2fs_sit_block *sit_blk;
2442 se = &sit_i->sentries[start];
2443 page = get_current_sit_page(sbi, start);
2444 sit_blk = (struct f2fs_sit_block *)page_address(page);
2445 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2446 f2fs_put_page(page, 1);
2448 check_block_count(sbi, start, &sit);
2449 seg_info_from_raw_sit(se, &sit);
2451 /* build discard map only one time */
2452 if (f2fs_discard_en(sbi)) {
2453 memcpy(se->discard_map, se->cur_valid_map,
2454 SIT_VBLOCK_MAP_SIZE);
2455 sbi->discard_blks += sbi->blocks_per_seg -
2459 if (sbi->segs_per_sec > 1)
2460 get_sec_entry(sbi, start)->valid_blocks +=
2463 start_blk += readed;
2464 } while (start_blk < sit_blk_cnt);
2466 down_read(&curseg->journal_rwsem);
2467 for (i = 0; i < sits_in_cursum(journal); i++) {
2468 unsigned int old_valid_blocks;
2470 start = le32_to_cpu(segno_in_journal(journal, i));
2471 se = &sit_i->sentries[start];
2472 sit = sit_in_journal(journal, i);
2474 old_valid_blocks = se->valid_blocks;
2476 check_block_count(sbi, start, &sit);
2477 seg_info_from_raw_sit(se, &sit);
2479 if (f2fs_discard_en(sbi)) {
2480 memcpy(se->discard_map, se->cur_valid_map,
2481 SIT_VBLOCK_MAP_SIZE);
2482 sbi->discard_blks += old_valid_blocks -
2486 if (sbi->segs_per_sec > 1)
2487 get_sec_entry(sbi, start)->valid_blocks +=
2488 se->valid_blocks - old_valid_blocks;
2490 up_read(&curseg->journal_rwsem);
2493 static void init_free_segmap(struct f2fs_sb_info *sbi)
2498 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2499 struct seg_entry *sentry = get_seg_entry(sbi, start);
2500 if (!sentry->valid_blocks)
2501 __set_free(sbi, start);
2503 SIT_I(sbi)->written_valid_blocks +=
2504 sentry->valid_blocks;
2507 /* set use the current segments */
2508 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2509 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2510 __set_test_and_inuse(sbi, curseg_t->segno);
2514 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2516 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2517 struct free_segmap_info *free_i = FREE_I(sbi);
2518 unsigned int segno = 0, offset = 0;
2519 unsigned short valid_blocks;
2522 /* find dirty segment based on free segmap */
2523 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2524 if (segno >= MAIN_SEGS(sbi))
2527 valid_blocks = get_valid_blocks(sbi, segno, 0);
2528 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2530 if (valid_blocks > sbi->blocks_per_seg) {
2531 f2fs_bug_on(sbi, 1);
2534 mutex_lock(&dirty_i->seglist_lock);
2535 __locate_dirty_segment(sbi, segno, DIRTY);
2536 mutex_unlock(&dirty_i->seglist_lock);
2540 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2542 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2543 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2545 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2546 if (!dirty_i->victim_secmap)
2551 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2553 struct dirty_seglist_info *dirty_i;
2554 unsigned int bitmap_size, i;
2556 /* allocate memory for dirty segments list information */
2557 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2561 SM_I(sbi)->dirty_info = dirty_i;
2562 mutex_init(&dirty_i->seglist_lock);
2564 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2566 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2567 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2568 if (!dirty_i->dirty_segmap[i])
2572 init_dirty_segmap(sbi);
2573 return init_victim_secmap(sbi);
2577 * Update min, max modified time for cost-benefit GC algorithm
2579 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2581 struct sit_info *sit_i = SIT_I(sbi);
2584 mutex_lock(&sit_i->sentry_lock);
2586 sit_i->min_mtime = LLONG_MAX;
2588 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2590 unsigned long long mtime = 0;
2592 for (i = 0; i < sbi->segs_per_sec; i++)
2593 mtime += get_seg_entry(sbi, segno + i)->mtime;
2595 mtime = div_u64(mtime, sbi->segs_per_sec);
2597 if (sit_i->min_mtime > mtime)
2598 sit_i->min_mtime = mtime;
2600 sit_i->max_mtime = get_mtime(sbi);
2601 mutex_unlock(&sit_i->sentry_lock);
2604 int build_segment_manager(struct f2fs_sb_info *sbi)
2606 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2607 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2608 struct f2fs_sm_info *sm_info;
2611 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2616 sbi->sm_info = sm_info;
2617 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2618 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2619 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2620 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2621 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2622 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2623 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2624 sm_info->rec_prefree_segments = sm_info->main_segments *
2625 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2626 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2627 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2629 if (!test_opt(sbi, LFS))
2630 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2631 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2632 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2634 INIT_LIST_HEAD(&sm_info->discard_list);
2635 INIT_LIST_HEAD(&sm_info->wait_list);
2636 sm_info->nr_discards = 0;
2637 sm_info->max_discards = 0;
2639 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2641 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2643 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2644 err = create_flush_cmd_control(sbi);
2649 err = build_sit_info(sbi);
2652 err = build_free_segmap(sbi);
2655 err = build_curseg(sbi);
2659 /* reinit free segmap based on SIT */
2660 build_sit_entries(sbi);
2662 init_free_segmap(sbi);
2663 err = build_dirty_segmap(sbi);
2667 init_min_max_mtime(sbi);
2671 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2672 enum dirty_type dirty_type)
2674 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2676 mutex_lock(&dirty_i->seglist_lock);
2677 kvfree(dirty_i->dirty_segmap[dirty_type]);
2678 dirty_i->nr_dirty[dirty_type] = 0;
2679 mutex_unlock(&dirty_i->seglist_lock);
2682 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2684 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2685 kvfree(dirty_i->victim_secmap);
2688 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2690 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2696 /* discard pre-free/dirty segments list */
2697 for (i = 0; i < NR_DIRTY_TYPE; i++)
2698 discard_dirty_segmap(sbi, i);
2700 destroy_victim_secmap(sbi);
2701 SM_I(sbi)->dirty_info = NULL;
2705 static void destroy_curseg(struct f2fs_sb_info *sbi)
2707 struct curseg_info *array = SM_I(sbi)->curseg_array;
2712 SM_I(sbi)->curseg_array = NULL;
2713 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2714 kfree(array[i].sum_blk);
2715 kfree(array[i].journal);
2720 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2722 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2725 SM_I(sbi)->free_info = NULL;
2726 kvfree(free_i->free_segmap);
2727 kvfree(free_i->free_secmap);
2731 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2733 struct sit_info *sit_i = SIT_I(sbi);
2739 if (sit_i->sentries) {
2740 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2741 kfree(sit_i->sentries[start].cur_valid_map);
2742 kfree(sit_i->sentries[start].ckpt_valid_map);
2743 kfree(sit_i->sentries[start].discard_map);
2746 kfree(sit_i->tmp_map);
2748 kvfree(sit_i->sentries);
2749 kvfree(sit_i->sec_entries);
2750 kvfree(sit_i->dirty_sentries_bitmap);
2752 SM_I(sbi)->sit_info = NULL;
2753 kfree(sit_i->sit_bitmap);
2757 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2759 struct f2fs_sm_info *sm_info = SM_I(sbi);
2763 destroy_flush_cmd_control(sbi, true);
2764 destroy_dirty_segmap(sbi);
2765 destroy_curseg(sbi);
2766 destroy_free_segmap(sbi);
2767 destroy_sit_info(sbi);
2768 sbi->sm_info = NULL;
2772 int __init create_segment_manager_caches(void)
2774 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2775 sizeof(struct discard_entry));
2776 if (!discard_entry_slab)
2779 bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2780 sizeof(struct bio_entry));
2781 if (!bio_entry_slab)
2782 goto destroy_discard_entry;
2784 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2785 sizeof(struct sit_entry_set));
2786 if (!sit_entry_set_slab)
2787 goto destroy_bio_entry;
2789 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2790 sizeof(struct inmem_pages));
2791 if (!inmem_entry_slab)
2792 goto destroy_sit_entry_set;
2795 destroy_sit_entry_set:
2796 kmem_cache_destroy(sit_entry_set_slab);
2798 kmem_cache_destroy(bio_entry_slab);
2799 destroy_discard_entry:
2800 kmem_cache_destroy(discard_entry_slab);
2805 void destroy_segment_manager_caches(void)
2807 kmem_cache_destroy(sit_entry_set_slab);
2808 kmem_cache_destroy(bio_entry_slab);
2809 kmem_cache_destroy(discard_entry_slab);
2810 kmem_cache_destroy(inmem_entry_slab);