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 *discard_cmd_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 mutex_lock(&fi->inmem_lock);
246 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
247 mutex_unlock(&fi->inmem_lock);
249 clear_inode_flag(inode, FI_ATOMIC_FILE);
250 stat_dec_atomic_write(inode);
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 = REQ_SYNC | REQ_PRIO,
264 .encrypted_page = NULL,
266 pgoff_t last_idx = ULONG_MAX;
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;
292 last_idx = page->index;
295 list_move_tail(&cur->list, revoke_list);
298 if (last_idx != ULONG_MAX)
299 f2fs_submit_merged_bio_cond(sbi, inode, 0, last_idx,
303 __revoke_inmem_pages(inode, revoke_list, false, false);
308 int commit_inmem_pages(struct inode *inode)
310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
311 struct f2fs_inode_info *fi = F2FS_I(inode);
312 struct list_head revoke_list;
315 INIT_LIST_HEAD(&revoke_list);
316 f2fs_balance_fs(sbi, true);
319 set_inode_flag(inode, FI_ATOMIC_COMMIT);
321 mutex_lock(&fi->inmem_lock);
322 err = __commit_inmem_pages(inode, &revoke_list);
326 * try to revoke all committed pages, but still we could fail
327 * due to no memory or other reason, if that happened, EAGAIN
328 * will be returned, which means in such case, transaction is
329 * already not integrity, caller should use journal to do the
330 * recovery or rewrite & commit last transaction. For other
331 * error number, revoking was done by filesystem itself.
333 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
337 /* drop all uncommitted pages */
338 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
340 mutex_unlock(&fi->inmem_lock);
342 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
349 * This function balances dirty node and dentry pages.
350 * In addition, it controls garbage collection.
352 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
354 #ifdef CONFIG_F2FS_FAULT_INJECTION
355 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
356 f2fs_show_injection_info(FAULT_CHECKPOINT);
357 f2fs_stop_checkpoint(sbi, false);
364 /* balance_fs_bg is able to be pending */
365 if (excess_cached_nats(sbi))
366 f2fs_balance_fs_bg(sbi);
369 * We should do GC or end up with checkpoint, if there are so many dirty
370 * dir/node pages without enough free segments.
372 if (has_not_enough_free_secs(sbi, 0, 0)) {
373 mutex_lock(&sbi->gc_mutex);
374 f2fs_gc(sbi, false, false);
378 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
380 /* try to shrink extent cache when there is no enough memory */
381 if (!available_free_memory(sbi, EXTENT_CACHE))
382 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
384 /* check the # of cached NAT entries */
385 if (!available_free_memory(sbi, NAT_ENTRIES))
386 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
388 if (!available_free_memory(sbi, FREE_NIDS))
389 try_to_free_nids(sbi, MAX_FREE_NIDS);
391 build_free_nids(sbi, false, false);
396 /* checkpoint is the only way to shrink partial cached entries */
397 if (!available_free_memory(sbi, NAT_ENTRIES) ||
398 !available_free_memory(sbi, INO_ENTRIES) ||
399 excess_prefree_segs(sbi) ||
400 excess_dirty_nats(sbi) ||
401 f2fs_time_over(sbi, CP_TIME)) {
402 if (test_opt(sbi, DATA_FLUSH)) {
403 struct blk_plug plug;
405 blk_start_plug(&plug);
406 sync_dirty_inodes(sbi, FILE_INODE);
407 blk_finish_plug(&plug);
409 f2fs_sync_fs(sbi->sb, true);
410 stat_inc_bg_cp_count(sbi->stat_info);
414 static int __submit_flush_wait(struct block_device *bdev)
416 struct bio *bio = f2fs_bio_alloc(0);
419 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
421 ret = submit_bio_wait(bio);
426 static int submit_flush_wait(struct f2fs_sb_info *sbi)
428 int ret = __submit_flush_wait(sbi->sb->s_bdev);
431 if (sbi->s_ndevs && !ret) {
432 for (i = 1; i < sbi->s_ndevs; i++) {
433 trace_f2fs_issue_flush(FDEV(i).bdev,
434 test_opt(sbi, NOBARRIER),
435 test_opt(sbi, FLUSH_MERGE));
436 ret = __submit_flush_wait(FDEV(i).bdev);
444 static int issue_flush_thread(void *data)
446 struct f2fs_sb_info *sbi = data;
447 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
448 wait_queue_head_t *q = &fcc->flush_wait_queue;
450 if (kthread_should_stop())
453 if (!llist_empty(&fcc->issue_list)) {
454 struct flush_cmd *cmd, *next;
457 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
458 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
460 ret = submit_flush_wait(sbi);
461 llist_for_each_entry_safe(cmd, next,
462 fcc->dispatch_list, llnode) {
464 complete(&cmd->wait);
466 fcc->dispatch_list = NULL;
469 wait_event_interruptible(*q,
470 kthread_should_stop() || !llist_empty(&fcc->issue_list));
474 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
476 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
477 struct flush_cmd cmd;
479 if (test_opt(sbi, NOBARRIER))
482 if (!test_opt(sbi, FLUSH_MERGE))
483 return submit_flush_wait(sbi);
485 if (!atomic_read(&fcc->submit_flush)) {
488 atomic_inc(&fcc->submit_flush);
489 ret = submit_flush_wait(sbi);
490 atomic_dec(&fcc->submit_flush);
494 init_completion(&cmd.wait);
496 atomic_inc(&fcc->submit_flush);
497 llist_add(&cmd.llnode, &fcc->issue_list);
499 if (!fcc->dispatch_list)
500 wake_up(&fcc->flush_wait_queue);
502 if (fcc->f2fs_issue_flush) {
503 wait_for_completion(&cmd.wait);
504 atomic_dec(&fcc->submit_flush);
506 llist_del_all(&fcc->issue_list);
507 atomic_set(&fcc->submit_flush, 0);
513 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
515 dev_t dev = sbi->sb->s_bdev->bd_dev;
516 struct flush_cmd_control *fcc;
519 if (SM_I(sbi)->fcc_info) {
520 fcc = SM_I(sbi)->fcc_info;
524 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
527 atomic_set(&fcc->submit_flush, 0);
528 init_waitqueue_head(&fcc->flush_wait_queue);
529 init_llist_head(&fcc->issue_list);
530 SM_I(sbi)->fcc_info = fcc;
532 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
533 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
534 if (IS_ERR(fcc->f2fs_issue_flush)) {
535 err = PTR_ERR(fcc->f2fs_issue_flush);
537 SM_I(sbi)->fcc_info = NULL;
544 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
546 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
548 if (fcc && fcc->f2fs_issue_flush) {
549 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
551 fcc->f2fs_issue_flush = NULL;
552 kthread_stop(flush_thread);
556 SM_I(sbi)->fcc_info = NULL;
560 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
561 enum dirty_type dirty_type)
563 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
565 /* need not be added */
566 if (IS_CURSEG(sbi, segno))
569 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
570 dirty_i->nr_dirty[dirty_type]++;
572 if (dirty_type == DIRTY) {
573 struct seg_entry *sentry = get_seg_entry(sbi, segno);
574 enum dirty_type t = sentry->type;
576 if (unlikely(t >= DIRTY)) {
580 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
581 dirty_i->nr_dirty[t]++;
585 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
586 enum dirty_type dirty_type)
588 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
590 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
591 dirty_i->nr_dirty[dirty_type]--;
593 if (dirty_type == DIRTY) {
594 struct seg_entry *sentry = get_seg_entry(sbi, segno);
595 enum dirty_type t = sentry->type;
597 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
598 dirty_i->nr_dirty[t]--;
600 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
601 clear_bit(GET_SECNO(sbi, segno),
602 dirty_i->victim_secmap);
607 * Should not occur error such as -ENOMEM.
608 * Adding dirty entry into seglist is not critical operation.
609 * If a given segment is one of current working segments, it won't be added.
611 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
613 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
614 unsigned short valid_blocks;
616 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
619 mutex_lock(&dirty_i->seglist_lock);
621 valid_blocks = get_valid_blocks(sbi, segno, 0);
623 if (valid_blocks == 0) {
624 __locate_dirty_segment(sbi, segno, PRE);
625 __remove_dirty_segment(sbi, segno, DIRTY);
626 } else if (valid_blocks < sbi->blocks_per_seg) {
627 __locate_dirty_segment(sbi, segno, DIRTY);
629 /* Recovery routine with SSR needs this */
630 __remove_dirty_segment(sbi, segno, DIRTY);
633 mutex_unlock(&dirty_i->seglist_lock);
636 static void __add_discard_cmd(struct f2fs_sb_info *sbi,
637 struct bio *bio, block_t lstart, block_t len)
639 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
640 struct list_head *cmd_list = &(dcc->discard_cmd_list);
641 struct discard_cmd *dc;
643 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
644 INIT_LIST_HEAD(&dc->list);
646 bio->bi_private = dc;
650 init_completion(&dc->wait);
652 mutex_lock(&dcc->cmd_lock);
653 list_add_tail(&dc->list, cmd_list);
654 mutex_unlock(&dcc->cmd_lock);
657 static void __remove_discard_cmd(struct f2fs_sb_info *sbi, struct discard_cmd *dc)
659 int err = dc->bio->bi_error;
661 if (dc->state == D_DONE)
662 atomic_dec(&(SM_I(sbi)->dcc_info->submit_discard));
664 if (err == -EOPNOTSUPP)
668 f2fs_msg(sbi->sb, KERN_INFO,
669 "Issue discard failed, ret: %d", err);
672 kmem_cache_free(discard_cmd_slab, dc);
675 /* This should be covered by global mutex, &sit_i->sentry_lock */
676 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
678 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
679 struct list_head *wait_list = &(dcc->discard_cmd_list);
680 struct discard_cmd *dc, *tmp;
681 struct blk_plug plug;
683 mutex_lock(&dcc->cmd_lock);
685 blk_start_plug(&plug);
687 list_for_each_entry_safe(dc, tmp, wait_list, list) {
689 if (blkaddr == NULL_ADDR) {
690 if (dc->state == D_PREP) {
691 dc->state = D_SUBMIT;
693 atomic_inc(&dcc->submit_discard);
698 if (dc->lstart <= blkaddr && blkaddr < dc->lstart + dc->len) {
699 if (dc->state == D_SUBMIT)
700 wait_for_completion_io(&dc->wait);
702 __remove_discard_cmd(sbi, dc);
705 blk_finish_plug(&plug);
707 /* this comes from f2fs_put_super */
708 if (blkaddr == NULL_ADDR) {
709 list_for_each_entry_safe(dc, tmp, wait_list, list) {
710 wait_for_completion_io(&dc->wait);
711 __remove_discard_cmd(sbi, dc);
714 mutex_unlock(&dcc->cmd_lock);
717 static void f2fs_submit_discard_endio(struct bio *bio)
719 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
725 static int issue_discard_thread(void *data)
727 struct f2fs_sb_info *sbi = data;
728 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
729 wait_queue_head_t *q = &dcc->discard_wait_queue;
730 struct list_head *cmd_list = &dcc->discard_cmd_list;
731 struct discard_cmd *dc, *tmp;
732 struct blk_plug plug;
735 if (kthread_should_stop())
738 blk_start_plug(&plug);
740 mutex_lock(&dcc->cmd_lock);
741 list_for_each_entry_safe(dc, tmp, cmd_list, list) {
742 if (dc->state == D_PREP) {
743 dc->state = D_SUBMIT;
745 atomic_inc(&dcc->submit_discard);
746 if (iter++ > DISCARD_ISSUE_RATE)
748 } else if (dc->state == D_DONE) {
749 __remove_discard_cmd(sbi, dc);
752 mutex_unlock(&dcc->cmd_lock);
754 blk_finish_plug(&plug);
757 congestion_wait(BLK_RW_SYNC, HZ/50);
759 wait_event_interruptible(*q,
760 kthread_should_stop() || !list_empty(&dcc->discard_cmd_list));
765 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
766 static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi,
767 struct block_device *bdev, block_t blkstart, block_t blklen)
769 struct bio *bio = NULL;
770 block_t lblkstart = blkstart;
773 trace_f2fs_issue_discard(bdev, blkstart, blklen);
776 int devi = f2fs_target_device_index(sbi, blkstart);
778 blkstart -= FDEV(devi).start_blk;
780 err = __blkdev_issue_discard(bdev,
781 SECTOR_FROM_BLOCK(blkstart),
782 SECTOR_FROM_BLOCK(blklen),
785 bio->bi_end_io = f2fs_submit_discard_endio;
786 bio->bi_opf |= REQ_SYNC;
788 __add_discard_cmd(sbi, bio, lblkstart, blklen);
789 wake_up(&SM_I(sbi)->dcc_info->discard_wait_queue);
794 #ifdef CONFIG_BLK_DEV_ZONED
795 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
796 struct block_device *bdev, block_t blkstart, block_t blklen)
798 sector_t sector, nr_sects;
802 devi = f2fs_target_device_index(sbi, blkstart);
803 blkstart -= FDEV(devi).start_blk;
807 * We need to know the type of the zone: for conventional zones,
808 * use regular discard if the drive supports it. For sequential
809 * zones, reset the zone write pointer.
811 switch (get_blkz_type(sbi, bdev, blkstart)) {
813 case BLK_ZONE_TYPE_CONVENTIONAL:
814 if (!blk_queue_discard(bdev_get_queue(bdev)))
816 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
817 case BLK_ZONE_TYPE_SEQWRITE_REQ:
818 case BLK_ZONE_TYPE_SEQWRITE_PREF:
819 sector = SECTOR_FROM_BLOCK(blkstart);
820 nr_sects = SECTOR_FROM_BLOCK(blklen);
822 if (sector & (bdev_zone_sectors(bdev) - 1) ||
823 nr_sects != bdev_zone_sectors(bdev)) {
824 f2fs_msg(sbi->sb, KERN_INFO,
825 "(%d) %s: Unaligned discard attempted (block %x + %x)",
826 devi, sbi->s_ndevs ? FDEV(devi).path: "",
830 trace_f2fs_issue_reset_zone(bdev, blkstart);
831 return blkdev_reset_zones(bdev, sector,
834 /* Unknown zone type: broken device ? */
840 static int __issue_discard_async(struct f2fs_sb_info *sbi,
841 struct block_device *bdev, block_t blkstart, block_t blklen)
843 #ifdef CONFIG_BLK_DEV_ZONED
844 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
845 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
846 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
848 return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
851 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
852 block_t blkstart, block_t blklen)
854 sector_t start = blkstart, len = 0;
855 struct block_device *bdev;
856 struct seg_entry *se;
861 bdev = f2fs_target_device(sbi, blkstart, NULL);
863 for (i = blkstart; i < blkstart + blklen; i++, len++) {
865 struct block_device *bdev2 =
866 f2fs_target_device(sbi, i, NULL);
869 err = __issue_discard_async(sbi, bdev,
879 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
880 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
882 if (!f2fs_test_and_set_bit(offset, se->discard_map))
887 err = __issue_discard_async(sbi, bdev, start, len);
891 static void __add_discard_entry(struct f2fs_sb_info *sbi,
892 struct cp_control *cpc, struct seg_entry *se,
893 unsigned int start, unsigned int end)
895 struct list_head *head = &SM_I(sbi)->dcc_info->discard_entry_list;
896 struct discard_entry *new, *last;
898 if (!list_empty(head)) {
899 last = list_last_entry(head, struct discard_entry, list);
900 if (START_BLOCK(sbi, cpc->trim_start) + start ==
901 last->blkaddr + last->len &&
902 last->len < MAX_DISCARD_BLOCKS(sbi)) {
903 last->len += end - start;
908 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
909 INIT_LIST_HEAD(&new->list);
910 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
911 new->len = end - start;
912 list_add_tail(&new->list, head);
914 SM_I(sbi)->dcc_info->nr_discards += end - start;
917 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
920 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
921 int max_blocks = sbi->blocks_per_seg;
922 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
923 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
924 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
925 unsigned long *discard_map = (unsigned long *)se->discard_map;
926 unsigned long *dmap = SIT_I(sbi)->tmp_map;
927 unsigned int start = 0, end = -1;
928 bool force = (cpc->reason == CP_DISCARD);
931 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
935 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
936 SM_I(sbi)->dcc_info->nr_discards >=
937 SM_I(sbi)->dcc_info->max_discards)
941 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
942 for (i = 0; i < entries; i++)
943 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
944 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
946 while (force || SM_I(sbi)->dcc_info->nr_discards <=
947 SM_I(sbi)->dcc_info->max_discards) {
948 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
949 if (start >= max_blocks)
952 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
953 if (force && start && end != max_blocks
954 && (end - start) < cpc->trim_minlen)
960 __add_discard_entry(sbi, cpc, se, start, end);
965 void release_discard_addrs(struct f2fs_sb_info *sbi)
967 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
968 struct discard_entry *entry, *this;
971 list_for_each_entry_safe(entry, this, head, list) {
972 list_del(&entry->list);
973 kmem_cache_free(discard_entry_slab, entry);
978 * Should call clear_prefree_segments after checkpoint is done.
980 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
982 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
985 mutex_lock(&dirty_i->seglist_lock);
986 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
987 __set_test_and_free(sbi, segno);
988 mutex_unlock(&dirty_i->seglist_lock);
991 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
993 struct list_head *head = &(SM_I(sbi)->dcc_info->discard_entry_list);
994 struct discard_entry *entry, *this;
995 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
996 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
997 unsigned int start = 0, end = -1;
998 unsigned int secno, start_segno;
999 bool force = (cpc->reason == CP_DISCARD);
1001 mutex_lock(&dirty_i->seglist_lock);
1005 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1006 if (start >= MAIN_SEGS(sbi))
1008 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1011 for (i = start; i < end; i++)
1012 clear_bit(i, prefree_map);
1014 dirty_i->nr_dirty[PRE] -= end - start;
1016 if (!test_opt(sbi, DISCARD))
1019 if (force && start >= cpc->trim_start &&
1020 (end - 1) <= cpc->trim_end)
1023 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1024 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1025 (end - start) << sbi->log_blocks_per_seg);
1029 secno = GET_SECNO(sbi, start);
1030 start_segno = secno * sbi->segs_per_sec;
1031 if (!IS_CURSEC(sbi, secno) &&
1032 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
1033 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1034 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1036 start = start_segno + sbi->segs_per_sec;
1042 mutex_unlock(&dirty_i->seglist_lock);
1044 /* send small discards */
1045 list_for_each_entry_safe(entry, this, head, list) {
1046 if (force && entry->len < cpc->trim_minlen)
1048 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
1049 cpc->trimmed += entry->len;
1051 list_del(&entry->list);
1052 SM_I(sbi)->dcc_info->nr_discards -= entry->len;
1053 kmem_cache_free(discard_entry_slab, entry);
1057 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1059 dev_t dev = sbi->sb->s_bdev->bd_dev;
1060 struct discard_cmd_control *dcc;
1063 if (SM_I(sbi)->dcc_info) {
1064 dcc = SM_I(sbi)->dcc_info;
1068 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1072 INIT_LIST_HEAD(&dcc->discard_entry_list);
1073 INIT_LIST_HEAD(&dcc->discard_cmd_list);
1074 mutex_init(&dcc->cmd_lock);
1075 atomic_set(&dcc->submit_discard, 0);
1076 dcc->nr_discards = 0;
1077 dcc->max_discards = 0;
1079 init_waitqueue_head(&dcc->discard_wait_queue);
1080 SM_I(sbi)->dcc_info = dcc;
1082 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1083 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1084 if (IS_ERR(dcc->f2fs_issue_discard)) {
1085 err = PTR_ERR(dcc->f2fs_issue_discard);
1087 SM_I(sbi)->dcc_info = NULL;
1094 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi, bool free)
1096 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1098 if (dcc && dcc->f2fs_issue_discard) {
1099 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1101 dcc->f2fs_issue_discard = NULL;
1102 kthread_stop(discard_thread);
1106 SM_I(sbi)->dcc_info = NULL;
1110 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1112 struct sit_info *sit_i = SIT_I(sbi);
1114 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1115 sit_i->dirty_sentries++;
1122 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1123 unsigned int segno, int modified)
1125 struct seg_entry *se = get_seg_entry(sbi, segno);
1128 __mark_sit_entry_dirty(sbi, segno);
1131 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1133 struct seg_entry *se;
1134 unsigned int segno, offset;
1135 long int new_vblocks;
1137 segno = GET_SEGNO(sbi, blkaddr);
1139 se = get_seg_entry(sbi, segno);
1140 new_vblocks = se->valid_blocks + del;
1141 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1143 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1144 (new_vblocks > sbi->blocks_per_seg)));
1146 se->valid_blocks = new_vblocks;
1147 se->mtime = get_mtime(sbi);
1148 SIT_I(sbi)->max_mtime = se->mtime;
1150 /* Update valid block bitmap */
1152 if (f2fs_test_and_set_bit(offset, se->cur_valid_map)) {
1153 #ifdef CONFIG_F2FS_CHECK_FS
1154 if (f2fs_test_and_set_bit(offset,
1155 se->cur_valid_map_mir))
1156 f2fs_bug_on(sbi, 1);
1160 f2fs_bug_on(sbi, 1);
1163 if (f2fs_discard_en(sbi) &&
1164 !f2fs_test_and_set_bit(offset, se->discard_map))
1165 sbi->discard_blks--;
1167 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1168 #ifdef CONFIG_F2FS_CHECK_FS
1169 if (!f2fs_test_and_clear_bit(offset,
1170 se->cur_valid_map_mir))
1171 f2fs_bug_on(sbi, 1);
1175 f2fs_bug_on(sbi, 1);
1178 if (f2fs_discard_en(sbi) &&
1179 f2fs_test_and_clear_bit(offset, se->discard_map))
1180 sbi->discard_blks++;
1182 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1183 se->ckpt_valid_blocks += del;
1185 __mark_sit_entry_dirty(sbi, segno);
1187 /* update total number of valid blocks to be written in ckpt area */
1188 SIT_I(sbi)->written_valid_blocks += del;
1190 if (sbi->segs_per_sec > 1)
1191 get_sec_entry(sbi, segno)->valid_blocks += del;
1194 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1196 update_sit_entry(sbi, new, 1);
1197 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1198 update_sit_entry(sbi, old, -1);
1200 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1201 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1204 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1206 unsigned int segno = GET_SEGNO(sbi, addr);
1207 struct sit_info *sit_i = SIT_I(sbi);
1209 f2fs_bug_on(sbi, addr == NULL_ADDR);
1210 if (addr == NEW_ADDR)
1213 /* add it into sit main buffer */
1214 mutex_lock(&sit_i->sentry_lock);
1216 update_sit_entry(sbi, addr, -1);
1218 /* add it into dirty seglist */
1219 locate_dirty_segment(sbi, segno);
1221 mutex_unlock(&sit_i->sentry_lock);
1224 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1226 struct sit_info *sit_i = SIT_I(sbi);
1227 unsigned int segno, offset;
1228 struct seg_entry *se;
1231 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1234 mutex_lock(&sit_i->sentry_lock);
1236 segno = GET_SEGNO(sbi, blkaddr);
1237 se = get_seg_entry(sbi, segno);
1238 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1240 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1243 mutex_unlock(&sit_i->sentry_lock);
1249 * This function should be resided under the curseg_mutex lock
1251 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1252 struct f2fs_summary *sum)
1254 struct curseg_info *curseg = CURSEG_I(sbi, type);
1255 void *addr = curseg->sum_blk;
1256 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1257 memcpy(addr, sum, sizeof(struct f2fs_summary));
1261 * Calculate the number of current summary pages for writing
1263 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1265 int valid_sum_count = 0;
1268 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1269 if (sbi->ckpt->alloc_type[i] == SSR)
1270 valid_sum_count += sbi->blocks_per_seg;
1273 valid_sum_count += le16_to_cpu(
1274 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1276 valid_sum_count += curseg_blkoff(sbi, i);
1280 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1281 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1282 if (valid_sum_count <= sum_in_page)
1284 else if ((valid_sum_count - sum_in_page) <=
1285 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1291 * Caller should put this summary page
1293 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1295 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1298 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1300 struct page *page = grab_meta_page(sbi, blk_addr);
1301 void *dst = page_address(page);
1304 memcpy(dst, src, PAGE_SIZE);
1306 memset(dst, 0, PAGE_SIZE);
1307 set_page_dirty(page);
1308 f2fs_put_page(page, 1);
1311 static void write_sum_page(struct f2fs_sb_info *sbi,
1312 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1314 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1317 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1318 int type, block_t blk_addr)
1320 struct curseg_info *curseg = CURSEG_I(sbi, type);
1321 struct page *page = grab_meta_page(sbi, blk_addr);
1322 struct f2fs_summary_block *src = curseg->sum_blk;
1323 struct f2fs_summary_block *dst;
1325 dst = (struct f2fs_summary_block *)page_address(page);
1327 mutex_lock(&curseg->curseg_mutex);
1329 down_read(&curseg->journal_rwsem);
1330 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1331 up_read(&curseg->journal_rwsem);
1333 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1334 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1336 mutex_unlock(&curseg->curseg_mutex);
1338 set_page_dirty(page);
1339 f2fs_put_page(page, 1);
1343 * Find a new segment from the free segments bitmap to right order
1344 * This function should be returned with success, otherwise BUG
1346 static void get_new_segment(struct f2fs_sb_info *sbi,
1347 unsigned int *newseg, bool new_sec, int dir)
1349 struct free_segmap_info *free_i = FREE_I(sbi);
1350 unsigned int segno, secno, zoneno;
1351 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1352 unsigned int hint = *newseg / sbi->segs_per_sec;
1353 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1354 unsigned int left_start = hint;
1359 spin_lock(&free_i->segmap_lock);
1361 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1362 segno = find_next_zero_bit(free_i->free_segmap,
1363 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1364 if (segno < (hint + 1) * sbi->segs_per_sec)
1368 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1369 if (secno >= MAIN_SECS(sbi)) {
1370 if (dir == ALLOC_RIGHT) {
1371 secno = find_next_zero_bit(free_i->free_secmap,
1373 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1376 left_start = hint - 1;
1382 while (test_bit(left_start, free_i->free_secmap)) {
1383 if (left_start > 0) {
1387 left_start = find_next_zero_bit(free_i->free_secmap,
1389 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1395 segno = secno * sbi->segs_per_sec;
1396 zoneno = secno / sbi->secs_per_zone;
1398 /* give up on finding another zone */
1401 if (sbi->secs_per_zone == 1)
1403 if (zoneno == old_zoneno)
1405 if (dir == ALLOC_LEFT) {
1406 if (!go_left && zoneno + 1 >= total_zones)
1408 if (go_left && zoneno == 0)
1411 for (i = 0; i < NR_CURSEG_TYPE; i++)
1412 if (CURSEG_I(sbi, i)->zone == zoneno)
1415 if (i < NR_CURSEG_TYPE) {
1416 /* zone is in user, try another */
1418 hint = zoneno * sbi->secs_per_zone - 1;
1419 else if (zoneno + 1 >= total_zones)
1422 hint = (zoneno + 1) * sbi->secs_per_zone;
1424 goto find_other_zone;
1427 /* set it as dirty segment in free segmap */
1428 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1429 __set_inuse(sbi, segno);
1431 spin_unlock(&free_i->segmap_lock);
1434 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1436 struct curseg_info *curseg = CURSEG_I(sbi, type);
1437 struct summary_footer *sum_footer;
1439 curseg->segno = curseg->next_segno;
1440 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1441 curseg->next_blkoff = 0;
1442 curseg->next_segno = NULL_SEGNO;
1444 sum_footer = &(curseg->sum_blk->footer);
1445 memset(sum_footer, 0, sizeof(struct summary_footer));
1446 if (IS_DATASEG(type))
1447 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1448 if (IS_NODESEG(type))
1449 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1450 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1454 * Allocate a current working segment.
1455 * This function always allocates a free segment in LFS manner.
1457 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1459 struct curseg_info *curseg = CURSEG_I(sbi, type);
1460 unsigned int segno = curseg->segno;
1461 int dir = ALLOC_LEFT;
1463 write_sum_page(sbi, curseg->sum_blk,
1464 GET_SUM_BLOCK(sbi, segno));
1465 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1468 if (test_opt(sbi, NOHEAP))
1471 get_new_segment(sbi, &segno, new_sec, dir);
1472 curseg->next_segno = segno;
1473 reset_curseg(sbi, type, 1);
1474 curseg->alloc_type = LFS;
1477 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1478 struct curseg_info *seg, block_t start)
1480 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1481 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1482 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1483 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1484 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1487 for (i = 0; i < entries; i++)
1488 target_map[i] = ckpt_map[i] | cur_map[i];
1490 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1492 seg->next_blkoff = pos;
1496 * If a segment is written by LFS manner, next block offset is just obtained
1497 * by increasing the current block offset. However, if a segment is written by
1498 * SSR manner, next block offset obtained by calling __next_free_blkoff
1500 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1501 struct curseg_info *seg)
1503 if (seg->alloc_type == SSR)
1504 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1510 * This function always allocates a used segment(from dirty seglist) by SSR
1511 * manner, so it should recover the existing segment information of valid blocks
1513 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1515 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1516 struct curseg_info *curseg = CURSEG_I(sbi, type);
1517 unsigned int new_segno = curseg->next_segno;
1518 struct f2fs_summary_block *sum_node;
1519 struct page *sum_page;
1521 write_sum_page(sbi, curseg->sum_blk,
1522 GET_SUM_BLOCK(sbi, curseg->segno));
1523 __set_test_and_inuse(sbi, new_segno);
1525 mutex_lock(&dirty_i->seglist_lock);
1526 __remove_dirty_segment(sbi, new_segno, PRE);
1527 __remove_dirty_segment(sbi, new_segno, DIRTY);
1528 mutex_unlock(&dirty_i->seglist_lock);
1530 reset_curseg(sbi, type, 1);
1531 curseg->alloc_type = SSR;
1532 __next_free_blkoff(sbi, curseg, 0);
1535 sum_page = get_sum_page(sbi, new_segno);
1536 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1537 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1538 f2fs_put_page(sum_page, 1);
1542 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1544 struct curseg_info *curseg = CURSEG_I(sbi, type);
1545 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1547 bool reversed = false;
1549 /* need_SSR() already forces to do this */
1550 if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR))
1553 /* For node segments, let's do SSR more intensively */
1554 if (IS_NODESEG(type)) {
1555 if (type >= CURSEG_WARM_NODE) {
1557 i = CURSEG_COLD_NODE;
1559 i = CURSEG_HOT_NODE;
1561 cnt = NR_CURSEG_NODE_TYPE;
1563 if (type >= CURSEG_WARM_DATA) {
1565 i = CURSEG_COLD_DATA;
1567 i = CURSEG_HOT_DATA;
1569 cnt = NR_CURSEG_DATA_TYPE;
1572 for (; cnt-- > 0; reversed ? i-- : i++) {
1575 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1583 * flush out current segment and replace it with new segment
1584 * This function should be returned with success, otherwise BUG
1586 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1587 int type, bool force)
1590 new_curseg(sbi, type, true);
1591 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1592 type == CURSEG_WARM_NODE)
1593 new_curseg(sbi, type, false);
1594 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1595 change_curseg(sbi, type, true);
1597 new_curseg(sbi, type, false);
1599 stat_inc_seg_type(sbi, CURSEG_I(sbi, type));
1602 void allocate_new_segments(struct f2fs_sb_info *sbi)
1604 struct curseg_info *curseg;
1605 unsigned int old_segno;
1608 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1609 curseg = CURSEG_I(sbi, i);
1610 old_segno = curseg->segno;
1611 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1612 locate_dirty_segment(sbi, old_segno);
1616 static const struct segment_allocation default_salloc_ops = {
1617 .allocate_segment = allocate_segment_by_default,
1620 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1622 __u64 trim_start = cpc->trim_start;
1623 bool has_candidate = false;
1625 mutex_lock(&SIT_I(sbi)->sentry_lock);
1626 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1627 if (add_discard_addrs(sbi, cpc, true)) {
1628 has_candidate = true;
1632 mutex_unlock(&SIT_I(sbi)->sentry_lock);
1634 cpc->trim_start = trim_start;
1635 return has_candidate;
1638 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1640 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1641 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1642 unsigned int start_segno, end_segno;
1643 struct cp_control cpc;
1646 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1650 if (end <= MAIN_BLKADDR(sbi))
1653 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1654 f2fs_msg(sbi->sb, KERN_WARNING,
1655 "Found FS corruption, run fsck to fix.");
1659 /* start/end segment number in main_area */
1660 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1661 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1662 GET_SEGNO(sbi, end);
1663 cpc.reason = CP_DISCARD;
1664 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1666 /* do checkpoint to issue discard commands safely */
1667 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1668 cpc.trim_start = start_segno;
1670 if (sbi->discard_blks == 0)
1672 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1673 cpc.trim_end = end_segno;
1675 cpc.trim_end = min_t(unsigned int,
1676 rounddown(start_segno +
1677 BATCHED_TRIM_SEGMENTS(sbi),
1678 sbi->segs_per_sec) - 1, end_segno);
1680 mutex_lock(&sbi->gc_mutex);
1681 err = write_checkpoint(sbi, &cpc);
1682 mutex_unlock(&sbi->gc_mutex);
1689 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1693 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1695 struct curseg_info *curseg = CURSEG_I(sbi, type);
1696 if (curseg->next_blkoff < sbi->blocks_per_seg)
1701 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1704 return CURSEG_HOT_DATA;
1706 return CURSEG_HOT_NODE;
1709 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1711 if (p_type == DATA) {
1712 struct inode *inode = page->mapping->host;
1714 if (S_ISDIR(inode->i_mode))
1715 return CURSEG_HOT_DATA;
1717 return CURSEG_COLD_DATA;
1719 if (IS_DNODE(page) && is_cold_node(page))
1720 return CURSEG_WARM_NODE;
1722 return CURSEG_COLD_NODE;
1726 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1728 if (p_type == DATA) {
1729 struct inode *inode = page->mapping->host;
1731 if (S_ISDIR(inode->i_mode))
1732 return CURSEG_HOT_DATA;
1733 else if (is_cold_data(page) || file_is_cold(inode))
1734 return CURSEG_COLD_DATA;
1736 return CURSEG_WARM_DATA;
1739 return is_cold_node(page) ? CURSEG_WARM_NODE :
1742 return CURSEG_COLD_NODE;
1746 static int __get_segment_type(struct page *page, enum page_type p_type)
1748 switch (F2FS_P_SB(page)->active_logs) {
1750 return __get_segment_type_2(page, p_type);
1752 return __get_segment_type_4(page, p_type);
1754 /* NR_CURSEG_TYPE(6) logs by default */
1755 f2fs_bug_on(F2FS_P_SB(page),
1756 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1757 return __get_segment_type_6(page, p_type);
1760 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1761 block_t old_blkaddr, block_t *new_blkaddr,
1762 struct f2fs_summary *sum, int type)
1764 struct sit_info *sit_i = SIT_I(sbi);
1765 struct curseg_info *curseg = CURSEG_I(sbi, type);
1767 mutex_lock(&curseg->curseg_mutex);
1768 mutex_lock(&sit_i->sentry_lock);
1770 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1772 f2fs_wait_discard_bio(sbi, *new_blkaddr);
1775 * __add_sum_entry should be resided under the curseg_mutex
1776 * because, this function updates a summary entry in the
1777 * current summary block.
1779 __add_sum_entry(sbi, type, sum);
1781 __refresh_next_blkoff(sbi, curseg);
1783 stat_inc_block_count(sbi, curseg);
1786 * SIT information should be updated before segment allocation,
1787 * since SSR needs latest valid block information.
1789 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1791 if (!__has_curseg_space(sbi, type))
1792 sit_i->s_ops->allocate_segment(sbi, type, false);
1794 mutex_unlock(&sit_i->sentry_lock);
1796 if (page && IS_NODESEG(type))
1797 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1799 mutex_unlock(&curseg->curseg_mutex);
1802 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1804 int type = __get_segment_type(fio->page, fio->type);
1807 if (fio->type == NODE || fio->type == DATA)
1808 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1810 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1811 &fio->new_blkaddr, sum, type);
1813 /* writeout dirty page into bdev */
1814 err = f2fs_submit_page_mbio(fio);
1815 if (err == -EAGAIN) {
1816 fio->old_blkaddr = fio->new_blkaddr;
1820 if (fio->type == NODE || fio->type == DATA)
1821 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1824 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1826 struct f2fs_io_info fio = {
1830 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
1831 .old_blkaddr = page->index,
1832 .new_blkaddr = page->index,
1834 .encrypted_page = NULL,
1837 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1838 fio.op_flags &= ~REQ_META;
1840 set_page_writeback(page);
1841 f2fs_submit_page_mbio(&fio);
1844 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1846 struct f2fs_summary sum;
1848 set_summary(&sum, nid, 0, 0);
1849 do_write_page(&sum, fio);
1852 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1854 struct f2fs_sb_info *sbi = fio->sbi;
1855 struct f2fs_summary sum;
1856 struct node_info ni;
1858 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1859 get_node_info(sbi, dn->nid, &ni);
1860 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1861 do_write_page(&sum, fio);
1862 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1865 void rewrite_data_page(struct f2fs_io_info *fio)
1867 fio->new_blkaddr = fio->old_blkaddr;
1868 stat_inc_inplace_blocks(fio->sbi);
1869 f2fs_submit_page_mbio(fio);
1872 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1873 block_t old_blkaddr, block_t new_blkaddr,
1874 bool recover_curseg, bool recover_newaddr)
1876 struct sit_info *sit_i = SIT_I(sbi);
1877 struct curseg_info *curseg;
1878 unsigned int segno, old_cursegno;
1879 struct seg_entry *se;
1881 unsigned short old_blkoff;
1883 segno = GET_SEGNO(sbi, new_blkaddr);
1884 se = get_seg_entry(sbi, segno);
1887 if (!recover_curseg) {
1888 /* for recovery flow */
1889 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1890 if (old_blkaddr == NULL_ADDR)
1891 type = CURSEG_COLD_DATA;
1893 type = CURSEG_WARM_DATA;
1896 if (!IS_CURSEG(sbi, segno))
1897 type = CURSEG_WARM_DATA;
1900 curseg = CURSEG_I(sbi, type);
1902 mutex_lock(&curseg->curseg_mutex);
1903 mutex_lock(&sit_i->sentry_lock);
1905 old_cursegno = curseg->segno;
1906 old_blkoff = curseg->next_blkoff;
1908 /* change the current segment */
1909 if (segno != curseg->segno) {
1910 curseg->next_segno = segno;
1911 change_curseg(sbi, type, true);
1914 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1915 __add_sum_entry(sbi, type, sum);
1917 if (!recover_curseg || recover_newaddr)
1918 update_sit_entry(sbi, new_blkaddr, 1);
1919 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1920 update_sit_entry(sbi, old_blkaddr, -1);
1922 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1923 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1925 locate_dirty_segment(sbi, old_cursegno);
1927 if (recover_curseg) {
1928 if (old_cursegno != curseg->segno) {
1929 curseg->next_segno = old_cursegno;
1930 change_curseg(sbi, type, true);
1932 curseg->next_blkoff = old_blkoff;
1935 mutex_unlock(&sit_i->sentry_lock);
1936 mutex_unlock(&curseg->curseg_mutex);
1939 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1940 block_t old_addr, block_t new_addr,
1941 unsigned char version, bool recover_curseg,
1942 bool recover_newaddr)
1944 struct f2fs_summary sum;
1946 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1948 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1949 recover_curseg, recover_newaddr);
1951 f2fs_update_data_blkaddr(dn, new_addr);
1954 void f2fs_wait_on_page_writeback(struct page *page,
1955 enum page_type type, bool ordered)
1957 if (PageWriteback(page)) {
1958 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1960 f2fs_submit_merged_bio_cond(sbi, page->mapping->host,
1961 0, page->index, type, WRITE);
1963 wait_on_page_writeback(page);
1965 wait_for_stable_page(page);
1969 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1974 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1977 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1979 f2fs_wait_on_page_writeback(cpage, DATA, true);
1980 f2fs_put_page(cpage, 1);
1984 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1986 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1987 struct curseg_info *seg_i;
1988 unsigned char *kaddr;
1993 start = start_sum_block(sbi);
1995 page = get_meta_page(sbi, start++);
1996 kaddr = (unsigned char *)page_address(page);
1998 /* Step 1: restore nat cache */
1999 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2000 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2002 /* Step 2: restore sit cache */
2003 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2004 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2005 offset = 2 * SUM_JOURNAL_SIZE;
2007 /* Step 3: restore summary entries */
2008 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2009 unsigned short blk_off;
2012 seg_i = CURSEG_I(sbi, i);
2013 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2014 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2015 seg_i->next_segno = segno;
2016 reset_curseg(sbi, i, 0);
2017 seg_i->alloc_type = ckpt->alloc_type[i];
2018 seg_i->next_blkoff = blk_off;
2020 if (seg_i->alloc_type == SSR)
2021 blk_off = sbi->blocks_per_seg;
2023 for (j = 0; j < blk_off; j++) {
2024 struct f2fs_summary *s;
2025 s = (struct f2fs_summary *)(kaddr + offset);
2026 seg_i->sum_blk->entries[j] = *s;
2027 offset += SUMMARY_SIZE;
2028 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2032 f2fs_put_page(page, 1);
2035 page = get_meta_page(sbi, start++);
2036 kaddr = (unsigned char *)page_address(page);
2040 f2fs_put_page(page, 1);
2044 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2046 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2047 struct f2fs_summary_block *sum;
2048 struct curseg_info *curseg;
2050 unsigned short blk_off;
2051 unsigned int segno = 0;
2052 block_t blk_addr = 0;
2054 /* get segment number and block addr */
2055 if (IS_DATASEG(type)) {
2056 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2057 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2059 if (__exist_node_summaries(sbi))
2060 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2062 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2064 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2066 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2068 if (__exist_node_summaries(sbi))
2069 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2070 type - CURSEG_HOT_NODE);
2072 blk_addr = GET_SUM_BLOCK(sbi, segno);
2075 new = get_meta_page(sbi, blk_addr);
2076 sum = (struct f2fs_summary_block *)page_address(new);
2078 if (IS_NODESEG(type)) {
2079 if (__exist_node_summaries(sbi)) {
2080 struct f2fs_summary *ns = &sum->entries[0];
2082 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2084 ns->ofs_in_node = 0;
2089 err = restore_node_summary(sbi, segno, sum);
2091 f2fs_put_page(new, 1);
2097 /* set uncompleted segment to curseg */
2098 curseg = CURSEG_I(sbi, type);
2099 mutex_lock(&curseg->curseg_mutex);
2101 /* update journal info */
2102 down_write(&curseg->journal_rwsem);
2103 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2104 up_write(&curseg->journal_rwsem);
2106 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2107 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2108 curseg->next_segno = segno;
2109 reset_curseg(sbi, type, 0);
2110 curseg->alloc_type = ckpt->alloc_type[type];
2111 curseg->next_blkoff = blk_off;
2112 mutex_unlock(&curseg->curseg_mutex);
2113 f2fs_put_page(new, 1);
2117 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2119 int type = CURSEG_HOT_DATA;
2122 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2123 int npages = npages_for_summary_flush(sbi, true);
2126 ra_meta_pages(sbi, start_sum_block(sbi), npages,
2129 /* restore for compacted data summary */
2130 if (read_compacted_summaries(sbi))
2132 type = CURSEG_HOT_NODE;
2135 if (__exist_node_summaries(sbi))
2136 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2137 NR_CURSEG_TYPE - type, META_CP, true);
2139 for (; type <= CURSEG_COLD_NODE; type++) {
2140 err = read_normal_summaries(sbi, type);
2148 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2151 unsigned char *kaddr;
2152 struct f2fs_summary *summary;
2153 struct curseg_info *seg_i;
2154 int written_size = 0;
2157 page = grab_meta_page(sbi, blkaddr++);
2158 kaddr = (unsigned char *)page_address(page);
2160 /* Step 1: write nat cache */
2161 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2162 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2163 written_size += SUM_JOURNAL_SIZE;
2165 /* Step 2: write sit cache */
2166 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2167 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2168 written_size += SUM_JOURNAL_SIZE;
2170 /* Step 3: write summary entries */
2171 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2172 unsigned short blkoff;
2173 seg_i = CURSEG_I(sbi, i);
2174 if (sbi->ckpt->alloc_type[i] == SSR)
2175 blkoff = sbi->blocks_per_seg;
2177 blkoff = curseg_blkoff(sbi, i);
2179 for (j = 0; j < blkoff; j++) {
2181 page = grab_meta_page(sbi, blkaddr++);
2182 kaddr = (unsigned char *)page_address(page);
2185 summary = (struct f2fs_summary *)(kaddr + written_size);
2186 *summary = seg_i->sum_blk->entries[j];
2187 written_size += SUMMARY_SIZE;
2189 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2193 set_page_dirty(page);
2194 f2fs_put_page(page, 1);
2199 set_page_dirty(page);
2200 f2fs_put_page(page, 1);
2204 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2205 block_t blkaddr, int type)
2208 if (IS_DATASEG(type))
2209 end = type + NR_CURSEG_DATA_TYPE;
2211 end = type + NR_CURSEG_NODE_TYPE;
2213 for (i = type; i < end; i++)
2214 write_current_sum_page(sbi, i, blkaddr + (i - type));
2217 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2219 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2220 write_compacted_summaries(sbi, start_blk);
2222 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2225 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2227 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2230 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2231 unsigned int val, int alloc)
2235 if (type == NAT_JOURNAL) {
2236 for (i = 0; i < nats_in_cursum(journal); i++) {
2237 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2240 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2241 return update_nats_in_cursum(journal, 1);
2242 } else if (type == SIT_JOURNAL) {
2243 for (i = 0; i < sits_in_cursum(journal); i++)
2244 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2246 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2247 return update_sits_in_cursum(journal, 1);
2252 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2255 return get_meta_page(sbi, current_sit_addr(sbi, segno));
2258 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2261 struct sit_info *sit_i = SIT_I(sbi);
2262 struct page *src_page, *dst_page;
2263 pgoff_t src_off, dst_off;
2264 void *src_addr, *dst_addr;
2266 src_off = current_sit_addr(sbi, start);
2267 dst_off = next_sit_addr(sbi, src_off);
2269 /* get current sit block page without lock */
2270 src_page = get_meta_page(sbi, src_off);
2271 dst_page = grab_meta_page(sbi, dst_off);
2272 f2fs_bug_on(sbi, PageDirty(src_page));
2274 src_addr = page_address(src_page);
2275 dst_addr = page_address(dst_page);
2276 memcpy(dst_addr, src_addr, PAGE_SIZE);
2278 set_page_dirty(dst_page);
2279 f2fs_put_page(src_page, 1);
2281 set_to_next_sit(sit_i, start);
2286 static struct sit_entry_set *grab_sit_entry_set(void)
2288 struct sit_entry_set *ses =
2289 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2292 INIT_LIST_HEAD(&ses->set_list);
2296 static void release_sit_entry_set(struct sit_entry_set *ses)
2298 list_del(&ses->set_list);
2299 kmem_cache_free(sit_entry_set_slab, ses);
2302 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2303 struct list_head *head)
2305 struct sit_entry_set *next = ses;
2307 if (list_is_last(&ses->set_list, head))
2310 list_for_each_entry_continue(next, head, set_list)
2311 if (ses->entry_cnt <= next->entry_cnt)
2314 list_move_tail(&ses->set_list, &next->set_list);
2317 static void add_sit_entry(unsigned int segno, struct list_head *head)
2319 struct sit_entry_set *ses;
2320 unsigned int start_segno = START_SEGNO(segno);
2322 list_for_each_entry(ses, head, set_list) {
2323 if (ses->start_segno == start_segno) {
2325 adjust_sit_entry_set(ses, head);
2330 ses = grab_sit_entry_set();
2332 ses->start_segno = start_segno;
2334 list_add(&ses->set_list, head);
2337 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2339 struct f2fs_sm_info *sm_info = SM_I(sbi);
2340 struct list_head *set_list = &sm_info->sit_entry_set;
2341 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2344 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2345 add_sit_entry(segno, set_list);
2348 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2350 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2351 struct f2fs_journal *journal = curseg->journal;
2354 down_write(&curseg->journal_rwsem);
2355 for (i = 0; i < sits_in_cursum(journal); i++) {
2359 segno = le32_to_cpu(segno_in_journal(journal, i));
2360 dirtied = __mark_sit_entry_dirty(sbi, segno);
2363 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2365 update_sits_in_cursum(journal, -i);
2366 up_write(&curseg->journal_rwsem);
2370 * CP calls this function, which flushes SIT entries including sit_journal,
2371 * and moves prefree segs to free segs.
2373 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2375 struct sit_info *sit_i = SIT_I(sbi);
2376 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2377 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2378 struct f2fs_journal *journal = curseg->journal;
2379 struct sit_entry_set *ses, *tmp;
2380 struct list_head *head = &SM_I(sbi)->sit_entry_set;
2381 bool to_journal = true;
2382 struct seg_entry *se;
2384 mutex_lock(&sit_i->sentry_lock);
2386 if (!sit_i->dirty_sentries)
2390 * add and account sit entries of dirty bitmap in sit entry
2393 add_sits_in_set(sbi);
2396 * if there are no enough space in journal to store dirty sit
2397 * entries, remove all entries from journal and add and account
2398 * them in sit entry set.
2400 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2401 remove_sits_in_journal(sbi);
2404 * there are two steps to flush sit entries:
2405 * #1, flush sit entries to journal in current cold data summary block.
2406 * #2, flush sit entries to sit page.
2408 list_for_each_entry_safe(ses, tmp, head, set_list) {
2409 struct page *page = NULL;
2410 struct f2fs_sit_block *raw_sit = NULL;
2411 unsigned int start_segno = ses->start_segno;
2412 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2413 (unsigned long)MAIN_SEGS(sbi));
2414 unsigned int segno = start_segno;
2417 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2421 down_write(&curseg->journal_rwsem);
2423 page = get_next_sit_page(sbi, start_segno);
2424 raw_sit = page_address(page);
2427 /* flush dirty sit entries in region of current sit set */
2428 for_each_set_bit_from(segno, bitmap, end) {
2429 int offset, sit_offset;
2431 se = get_seg_entry(sbi, segno);
2433 /* add discard candidates */
2434 if (cpc->reason != CP_DISCARD) {
2435 cpc->trim_start = segno;
2436 add_discard_addrs(sbi, cpc, false);
2440 offset = lookup_journal_in_cursum(journal,
2441 SIT_JOURNAL, segno, 1);
2442 f2fs_bug_on(sbi, offset < 0);
2443 segno_in_journal(journal, offset) =
2445 seg_info_to_raw_sit(se,
2446 &sit_in_journal(journal, offset));
2448 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2449 seg_info_to_raw_sit(se,
2450 &raw_sit->entries[sit_offset]);
2453 __clear_bit(segno, bitmap);
2454 sit_i->dirty_sentries--;
2459 up_write(&curseg->journal_rwsem);
2461 f2fs_put_page(page, 1);
2463 f2fs_bug_on(sbi, ses->entry_cnt);
2464 release_sit_entry_set(ses);
2467 f2fs_bug_on(sbi, !list_empty(head));
2468 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2470 if (cpc->reason == CP_DISCARD) {
2471 __u64 trim_start = cpc->trim_start;
2473 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2474 add_discard_addrs(sbi, cpc, false);
2476 cpc->trim_start = trim_start;
2478 mutex_unlock(&sit_i->sentry_lock);
2480 set_prefree_as_free_segments(sbi);
2483 static int build_sit_info(struct f2fs_sb_info *sbi)
2485 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2486 struct sit_info *sit_i;
2487 unsigned int sit_segs, start;
2489 unsigned int bitmap_size;
2491 /* allocate memory for SIT information */
2492 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2496 SM_I(sbi)->sit_info = sit_i;
2498 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2499 sizeof(struct seg_entry), GFP_KERNEL);
2500 if (!sit_i->sentries)
2503 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2504 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2505 if (!sit_i->dirty_sentries_bitmap)
2508 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2509 sit_i->sentries[start].cur_valid_map
2510 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2511 sit_i->sentries[start].ckpt_valid_map
2512 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2513 if (!sit_i->sentries[start].cur_valid_map ||
2514 !sit_i->sentries[start].ckpt_valid_map)
2517 #ifdef CONFIG_F2FS_CHECK_FS
2518 sit_i->sentries[start].cur_valid_map_mir
2519 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2520 if (!sit_i->sentries[start].cur_valid_map_mir)
2524 if (f2fs_discard_en(sbi)) {
2525 sit_i->sentries[start].discard_map
2526 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2527 if (!sit_i->sentries[start].discard_map)
2532 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2533 if (!sit_i->tmp_map)
2536 if (sbi->segs_per_sec > 1) {
2537 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2538 sizeof(struct sec_entry), GFP_KERNEL);
2539 if (!sit_i->sec_entries)
2543 /* get information related with SIT */
2544 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2546 /* setup SIT bitmap from ckeckpoint pack */
2547 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2548 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2550 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2551 if (!sit_i->sit_bitmap)
2554 #ifdef CONFIG_F2FS_CHECK_FS
2555 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2556 if (!sit_i->sit_bitmap_mir)
2560 /* init SIT information */
2561 sit_i->s_ops = &default_salloc_ops;
2563 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2564 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2565 sit_i->written_valid_blocks = 0;
2566 sit_i->bitmap_size = bitmap_size;
2567 sit_i->dirty_sentries = 0;
2568 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2569 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2570 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2571 mutex_init(&sit_i->sentry_lock);
2575 static int build_free_segmap(struct f2fs_sb_info *sbi)
2577 struct free_segmap_info *free_i;
2578 unsigned int bitmap_size, sec_bitmap_size;
2580 /* allocate memory for free segmap information */
2581 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2585 SM_I(sbi)->free_info = free_i;
2587 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2588 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2589 if (!free_i->free_segmap)
2592 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2593 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2594 if (!free_i->free_secmap)
2597 /* set all segments as dirty temporarily */
2598 memset(free_i->free_segmap, 0xff, bitmap_size);
2599 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2601 /* init free segmap information */
2602 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2603 free_i->free_segments = 0;
2604 free_i->free_sections = 0;
2605 spin_lock_init(&free_i->segmap_lock);
2609 static int build_curseg(struct f2fs_sb_info *sbi)
2611 struct curseg_info *array;
2614 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2618 SM_I(sbi)->curseg_array = array;
2620 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2621 mutex_init(&array[i].curseg_mutex);
2622 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2623 if (!array[i].sum_blk)
2625 init_rwsem(&array[i].journal_rwsem);
2626 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2628 if (!array[i].journal)
2630 array[i].segno = NULL_SEGNO;
2631 array[i].next_blkoff = 0;
2633 return restore_curseg_summaries(sbi);
2636 static void build_sit_entries(struct f2fs_sb_info *sbi)
2638 struct sit_info *sit_i = SIT_I(sbi);
2639 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2640 struct f2fs_journal *journal = curseg->journal;
2641 struct seg_entry *se;
2642 struct f2fs_sit_entry sit;
2643 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2644 unsigned int i, start, end;
2645 unsigned int readed, start_blk = 0;
2648 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2651 start = start_blk * sit_i->sents_per_block;
2652 end = (start_blk + readed) * sit_i->sents_per_block;
2654 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2655 struct f2fs_sit_block *sit_blk;
2658 se = &sit_i->sentries[start];
2659 page = get_current_sit_page(sbi, start);
2660 sit_blk = (struct f2fs_sit_block *)page_address(page);
2661 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2662 f2fs_put_page(page, 1);
2664 check_block_count(sbi, start, &sit);
2665 seg_info_from_raw_sit(se, &sit);
2667 /* build discard map only one time */
2668 if (f2fs_discard_en(sbi)) {
2669 memcpy(se->discard_map, se->cur_valid_map,
2670 SIT_VBLOCK_MAP_SIZE);
2671 sbi->discard_blks += sbi->blocks_per_seg -
2675 if (sbi->segs_per_sec > 1)
2676 get_sec_entry(sbi, start)->valid_blocks +=
2679 start_blk += readed;
2680 } while (start_blk < sit_blk_cnt);
2682 down_read(&curseg->journal_rwsem);
2683 for (i = 0; i < sits_in_cursum(journal); i++) {
2684 unsigned int old_valid_blocks;
2686 start = le32_to_cpu(segno_in_journal(journal, i));
2687 se = &sit_i->sentries[start];
2688 sit = sit_in_journal(journal, i);
2690 old_valid_blocks = se->valid_blocks;
2692 check_block_count(sbi, start, &sit);
2693 seg_info_from_raw_sit(se, &sit);
2695 if (f2fs_discard_en(sbi)) {
2696 memcpy(se->discard_map, se->cur_valid_map,
2697 SIT_VBLOCK_MAP_SIZE);
2698 sbi->discard_blks += old_valid_blocks -
2702 if (sbi->segs_per_sec > 1)
2703 get_sec_entry(sbi, start)->valid_blocks +=
2704 se->valid_blocks - old_valid_blocks;
2706 up_read(&curseg->journal_rwsem);
2709 static void init_free_segmap(struct f2fs_sb_info *sbi)
2714 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2715 struct seg_entry *sentry = get_seg_entry(sbi, start);
2716 if (!sentry->valid_blocks)
2717 __set_free(sbi, start);
2719 SIT_I(sbi)->written_valid_blocks +=
2720 sentry->valid_blocks;
2723 /* set use the current segments */
2724 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2725 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2726 __set_test_and_inuse(sbi, curseg_t->segno);
2730 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2732 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2733 struct free_segmap_info *free_i = FREE_I(sbi);
2734 unsigned int segno = 0, offset = 0;
2735 unsigned short valid_blocks;
2738 /* find dirty segment based on free segmap */
2739 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2740 if (segno >= MAIN_SEGS(sbi))
2743 valid_blocks = get_valid_blocks(sbi, segno, 0);
2744 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2746 if (valid_blocks > sbi->blocks_per_seg) {
2747 f2fs_bug_on(sbi, 1);
2750 mutex_lock(&dirty_i->seglist_lock);
2751 __locate_dirty_segment(sbi, segno, DIRTY);
2752 mutex_unlock(&dirty_i->seglist_lock);
2756 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2758 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2759 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2761 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2762 if (!dirty_i->victim_secmap)
2767 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2769 struct dirty_seglist_info *dirty_i;
2770 unsigned int bitmap_size, i;
2772 /* allocate memory for dirty segments list information */
2773 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2777 SM_I(sbi)->dirty_info = dirty_i;
2778 mutex_init(&dirty_i->seglist_lock);
2780 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2782 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2783 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2784 if (!dirty_i->dirty_segmap[i])
2788 init_dirty_segmap(sbi);
2789 return init_victim_secmap(sbi);
2793 * Update min, max modified time for cost-benefit GC algorithm
2795 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2797 struct sit_info *sit_i = SIT_I(sbi);
2800 mutex_lock(&sit_i->sentry_lock);
2802 sit_i->min_mtime = LLONG_MAX;
2804 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2806 unsigned long long mtime = 0;
2808 for (i = 0; i < sbi->segs_per_sec; i++)
2809 mtime += get_seg_entry(sbi, segno + i)->mtime;
2811 mtime = div_u64(mtime, sbi->segs_per_sec);
2813 if (sit_i->min_mtime > mtime)
2814 sit_i->min_mtime = mtime;
2816 sit_i->max_mtime = get_mtime(sbi);
2817 mutex_unlock(&sit_i->sentry_lock);
2820 int build_segment_manager(struct f2fs_sb_info *sbi)
2822 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2823 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2824 struct f2fs_sm_info *sm_info;
2827 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2832 sbi->sm_info = sm_info;
2833 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2834 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2835 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2836 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2837 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2838 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2839 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2840 sm_info->rec_prefree_segments = sm_info->main_segments *
2841 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2842 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2843 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2845 if (!test_opt(sbi, LFS))
2846 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2847 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2848 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2850 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2852 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2854 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2855 err = create_flush_cmd_control(sbi);
2860 err = create_discard_cmd_control(sbi);
2864 err = build_sit_info(sbi);
2867 err = build_free_segmap(sbi);
2870 err = build_curseg(sbi);
2874 /* reinit free segmap based on SIT */
2875 build_sit_entries(sbi);
2877 init_free_segmap(sbi);
2878 err = build_dirty_segmap(sbi);
2882 init_min_max_mtime(sbi);
2886 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2887 enum dirty_type dirty_type)
2889 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2891 mutex_lock(&dirty_i->seglist_lock);
2892 kvfree(dirty_i->dirty_segmap[dirty_type]);
2893 dirty_i->nr_dirty[dirty_type] = 0;
2894 mutex_unlock(&dirty_i->seglist_lock);
2897 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2899 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2900 kvfree(dirty_i->victim_secmap);
2903 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2905 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2911 /* discard pre-free/dirty segments list */
2912 for (i = 0; i < NR_DIRTY_TYPE; i++)
2913 discard_dirty_segmap(sbi, i);
2915 destroy_victim_secmap(sbi);
2916 SM_I(sbi)->dirty_info = NULL;
2920 static void destroy_curseg(struct f2fs_sb_info *sbi)
2922 struct curseg_info *array = SM_I(sbi)->curseg_array;
2927 SM_I(sbi)->curseg_array = NULL;
2928 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2929 kfree(array[i].sum_blk);
2930 kfree(array[i].journal);
2935 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2937 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2940 SM_I(sbi)->free_info = NULL;
2941 kvfree(free_i->free_segmap);
2942 kvfree(free_i->free_secmap);
2946 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2948 struct sit_info *sit_i = SIT_I(sbi);
2954 if (sit_i->sentries) {
2955 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2956 kfree(sit_i->sentries[start].cur_valid_map);
2957 #ifdef CONFIG_F2FS_CHECK_FS
2958 kfree(sit_i->sentries[start].cur_valid_map_mir);
2960 kfree(sit_i->sentries[start].ckpt_valid_map);
2961 kfree(sit_i->sentries[start].discard_map);
2964 kfree(sit_i->tmp_map);
2966 kvfree(sit_i->sentries);
2967 kvfree(sit_i->sec_entries);
2968 kvfree(sit_i->dirty_sentries_bitmap);
2970 SM_I(sbi)->sit_info = NULL;
2971 kfree(sit_i->sit_bitmap);
2972 #ifdef CONFIG_F2FS_CHECK_FS
2973 kfree(sit_i->sit_bitmap_mir);
2978 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2980 struct f2fs_sm_info *sm_info = SM_I(sbi);
2984 destroy_flush_cmd_control(sbi, true);
2985 destroy_discard_cmd_control(sbi, true);
2986 destroy_dirty_segmap(sbi);
2987 destroy_curseg(sbi);
2988 destroy_free_segmap(sbi);
2989 destroy_sit_info(sbi);
2990 sbi->sm_info = NULL;
2994 int __init create_segment_manager_caches(void)
2996 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2997 sizeof(struct discard_entry));
2998 if (!discard_entry_slab)
3001 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3002 sizeof(struct discard_cmd));
3003 if (!discard_cmd_slab)
3004 goto destroy_discard_entry;
3006 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3007 sizeof(struct sit_entry_set));
3008 if (!sit_entry_set_slab)
3009 goto destroy_discard_cmd;
3011 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3012 sizeof(struct inmem_pages));
3013 if (!inmem_entry_slab)
3014 goto destroy_sit_entry_set;
3017 destroy_sit_entry_set:
3018 kmem_cache_destroy(sit_entry_set_slab);
3019 destroy_discard_cmd:
3020 kmem_cache_destroy(discard_cmd_slab);
3021 destroy_discard_entry:
3022 kmem_cache_destroy(discard_entry_slab);
3027 void destroy_segment_manager_caches(void)
3029 kmem_cache_destroy(sit_entry_set_slab);
3030 kmem_cache_destroy(discard_cmd_slab);
3031 kmem_cache_destroy(discard_entry_slab);
3032 kmem_cache_destroy(inmem_entry_slab);