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 /* don't overwrite by SSR to keep node chain */
1168 if (se->type == CURSEG_WARM_NODE) {
1169 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1170 se->ckpt_valid_blocks++;
1173 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map)) {
1174 #ifdef CONFIG_F2FS_CHECK_FS
1175 if (!f2fs_test_and_clear_bit(offset,
1176 se->cur_valid_map_mir))
1177 f2fs_bug_on(sbi, 1);
1181 f2fs_bug_on(sbi, 1);
1184 if (f2fs_discard_en(sbi) &&
1185 f2fs_test_and_clear_bit(offset, se->discard_map))
1186 sbi->discard_blks++;
1188 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1189 se->ckpt_valid_blocks += del;
1191 __mark_sit_entry_dirty(sbi, segno);
1193 /* update total number of valid blocks to be written in ckpt area */
1194 SIT_I(sbi)->written_valid_blocks += del;
1196 if (sbi->segs_per_sec > 1)
1197 get_sec_entry(sbi, segno)->valid_blocks += del;
1200 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1202 update_sit_entry(sbi, new, 1);
1203 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1204 update_sit_entry(sbi, old, -1);
1206 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1207 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1210 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1212 unsigned int segno = GET_SEGNO(sbi, addr);
1213 struct sit_info *sit_i = SIT_I(sbi);
1215 f2fs_bug_on(sbi, addr == NULL_ADDR);
1216 if (addr == NEW_ADDR)
1219 /* add it into sit main buffer */
1220 mutex_lock(&sit_i->sentry_lock);
1222 update_sit_entry(sbi, addr, -1);
1224 /* add it into dirty seglist */
1225 locate_dirty_segment(sbi, segno);
1227 mutex_unlock(&sit_i->sentry_lock);
1230 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1232 struct sit_info *sit_i = SIT_I(sbi);
1233 unsigned int segno, offset;
1234 struct seg_entry *se;
1237 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1240 mutex_lock(&sit_i->sentry_lock);
1242 segno = GET_SEGNO(sbi, blkaddr);
1243 se = get_seg_entry(sbi, segno);
1244 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1246 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1249 mutex_unlock(&sit_i->sentry_lock);
1255 * This function should be resided under the curseg_mutex lock
1257 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1258 struct f2fs_summary *sum)
1260 struct curseg_info *curseg = CURSEG_I(sbi, type);
1261 void *addr = curseg->sum_blk;
1262 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1263 memcpy(addr, sum, sizeof(struct f2fs_summary));
1267 * Calculate the number of current summary pages for writing
1269 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1271 int valid_sum_count = 0;
1274 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1275 if (sbi->ckpt->alloc_type[i] == SSR)
1276 valid_sum_count += sbi->blocks_per_seg;
1279 valid_sum_count += le16_to_cpu(
1280 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1282 valid_sum_count += curseg_blkoff(sbi, i);
1286 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1287 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1288 if (valid_sum_count <= sum_in_page)
1290 else if ((valid_sum_count - sum_in_page) <=
1291 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1297 * Caller should put this summary page
1299 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1301 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1304 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1306 struct page *page = grab_meta_page(sbi, blk_addr);
1307 void *dst = page_address(page);
1310 memcpy(dst, src, PAGE_SIZE);
1312 memset(dst, 0, PAGE_SIZE);
1313 set_page_dirty(page);
1314 f2fs_put_page(page, 1);
1317 static void write_sum_page(struct f2fs_sb_info *sbi,
1318 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1320 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1323 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1324 int type, block_t blk_addr)
1326 struct curseg_info *curseg = CURSEG_I(sbi, type);
1327 struct page *page = grab_meta_page(sbi, blk_addr);
1328 struct f2fs_summary_block *src = curseg->sum_blk;
1329 struct f2fs_summary_block *dst;
1331 dst = (struct f2fs_summary_block *)page_address(page);
1333 mutex_lock(&curseg->curseg_mutex);
1335 down_read(&curseg->journal_rwsem);
1336 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1337 up_read(&curseg->journal_rwsem);
1339 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1340 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1342 mutex_unlock(&curseg->curseg_mutex);
1344 set_page_dirty(page);
1345 f2fs_put_page(page, 1);
1349 * Find a new segment from the free segments bitmap to right order
1350 * This function should be returned with success, otherwise BUG
1352 static void get_new_segment(struct f2fs_sb_info *sbi,
1353 unsigned int *newseg, bool new_sec, int dir)
1355 struct free_segmap_info *free_i = FREE_I(sbi);
1356 unsigned int segno, secno, zoneno;
1357 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1358 unsigned int hint = *newseg / sbi->segs_per_sec;
1359 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1360 unsigned int left_start = hint;
1365 spin_lock(&free_i->segmap_lock);
1367 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1368 segno = find_next_zero_bit(free_i->free_segmap,
1369 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1370 if (segno < (hint + 1) * sbi->segs_per_sec)
1374 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1375 if (secno >= MAIN_SECS(sbi)) {
1376 if (dir == ALLOC_RIGHT) {
1377 secno = find_next_zero_bit(free_i->free_secmap,
1379 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1382 left_start = hint - 1;
1388 while (test_bit(left_start, free_i->free_secmap)) {
1389 if (left_start > 0) {
1393 left_start = find_next_zero_bit(free_i->free_secmap,
1395 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1401 segno = secno * sbi->segs_per_sec;
1402 zoneno = secno / sbi->secs_per_zone;
1404 /* give up on finding another zone */
1407 if (sbi->secs_per_zone == 1)
1409 if (zoneno == old_zoneno)
1411 if (dir == ALLOC_LEFT) {
1412 if (!go_left && zoneno + 1 >= total_zones)
1414 if (go_left && zoneno == 0)
1417 for (i = 0; i < NR_CURSEG_TYPE; i++)
1418 if (CURSEG_I(sbi, i)->zone == zoneno)
1421 if (i < NR_CURSEG_TYPE) {
1422 /* zone is in user, try another */
1424 hint = zoneno * sbi->secs_per_zone - 1;
1425 else if (zoneno + 1 >= total_zones)
1428 hint = (zoneno + 1) * sbi->secs_per_zone;
1430 goto find_other_zone;
1433 /* set it as dirty segment in free segmap */
1434 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1435 __set_inuse(sbi, segno);
1437 spin_unlock(&free_i->segmap_lock);
1440 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1442 struct curseg_info *curseg = CURSEG_I(sbi, type);
1443 struct summary_footer *sum_footer;
1445 curseg->segno = curseg->next_segno;
1446 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1447 curseg->next_blkoff = 0;
1448 curseg->next_segno = NULL_SEGNO;
1450 sum_footer = &(curseg->sum_blk->footer);
1451 memset(sum_footer, 0, sizeof(struct summary_footer));
1452 if (IS_DATASEG(type))
1453 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1454 if (IS_NODESEG(type))
1455 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1456 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1460 * Allocate a current working segment.
1461 * This function always allocates a free segment in LFS manner.
1463 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1465 struct curseg_info *curseg = CURSEG_I(sbi, type);
1466 unsigned int segno = curseg->segno;
1467 int dir = ALLOC_LEFT;
1469 write_sum_page(sbi, curseg->sum_blk,
1470 GET_SUM_BLOCK(sbi, segno));
1471 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1474 if (test_opt(sbi, NOHEAP))
1477 get_new_segment(sbi, &segno, new_sec, dir);
1478 curseg->next_segno = segno;
1479 reset_curseg(sbi, type, 1);
1480 curseg->alloc_type = LFS;
1483 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1484 struct curseg_info *seg, block_t start)
1486 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1487 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1488 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1489 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1490 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1493 for (i = 0; i < entries; i++)
1494 target_map[i] = ckpt_map[i] | cur_map[i];
1496 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1498 seg->next_blkoff = pos;
1502 * If a segment is written by LFS manner, next block offset is just obtained
1503 * by increasing the current block offset. However, if a segment is written by
1504 * SSR manner, next block offset obtained by calling __next_free_blkoff
1506 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1507 struct curseg_info *seg)
1509 if (seg->alloc_type == SSR)
1510 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1516 * This function always allocates a used segment(from dirty seglist) by SSR
1517 * manner, so it should recover the existing segment information of valid blocks
1519 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1521 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1522 struct curseg_info *curseg = CURSEG_I(sbi, type);
1523 unsigned int new_segno = curseg->next_segno;
1524 struct f2fs_summary_block *sum_node;
1525 struct page *sum_page;
1527 write_sum_page(sbi, curseg->sum_blk,
1528 GET_SUM_BLOCK(sbi, curseg->segno));
1529 __set_test_and_inuse(sbi, new_segno);
1531 mutex_lock(&dirty_i->seglist_lock);
1532 __remove_dirty_segment(sbi, new_segno, PRE);
1533 __remove_dirty_segment(sbi, new_segno, DIRTY);
1534 mutex_unlock(&dirty_i->seglist_lock);
1536 reset_curseg(sbi, type, 1);
1537 curseg->alloc_type = SSR;
1538 __next_free_blkoff(sbi, curseg, 0);
1541 sum_page = get_sum_page(sbi, new_segno);
1542 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1543 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1544 f2fs_put_page(sum_page, 1);
1548 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1550 struct curseg_info *curseg = CURSEG_I(sbi, type);
1551 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1553 bool reversed = false;
1555 /* need_SSR() already forces to do this */
1556 if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR))
1559 /* For node segments, let's do SSR more intensively */
1560 if (IS_NODESEG(type)) {
1561 if (type >= CURSEG_WARM_NODE) {
1563 i = CURSEG_COLD_NODE;
1565 i = CURSEG_HOT_NODE;
1567 cnt = NR_CURSEG_NODE_TYPE;
1569 if (type >= CURSEG_WARM_DATA) {
1571 i = CURSEG_COLD_DATA;
1573 i = CURSEG_HOT_DATA;
1575 cnt = NR_CURSEG_DATA_TYPE;
1578 for (; cnt-- > 0; reversed ? i-- : i++) {
1581 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1589 * flush out current segment and replace it with new segment
1590 * This function should be returned with success, otherwise BUG
1592 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1593 int type, bool force)
1596 new_curseg(sbi, type, true);
1597 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
1598 type == CURSEG_WARM_NODE)
1599 new_curseg(sbi, type, false);
1600 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1601 change_curseg(sbi, type, true);
1603 new_curseg(sbi, type, false);
1605 stat_inc_seg_type(sbi, CURSEG_I(sbi, type));
1608 void allocate_new_segments(struct f2fs_sb_info *sbi)
1610 struct curseg_info *curseg;
1611 unsigned int old_segno;
1614 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1615 curseg = CURSEG_I(sbi, i);
1616 old_segno = curseg->segno;
1617 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1618 locate_dirty_segment(sbi, old_segno);
1622 static const struct segment_allocation default_salloc_ops = {
1623 .allocate_segment = allocate_segment_by_default,
1626 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1628 __u64 trim_start = cpc->trim_start;
1629 bool has_candidate = false;
1631 mutex_lock(&SIT_I(sbi)->sentry_lock);
1632 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
1633 if (add_discard_addrs(sbi, cpc, true)) {
1634 has_candidate = true;
1638 mutex_unlock(&SIT_I(sbi)->sentry_lock);
1640 cpc->trim_start = trim_start;
1641 return has_candidate;
1644 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1646 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1647 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1648 unsigned int start_segno, end_segno;
1649 struct cp_control cpc;
1652 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1656 if (end <= MAIN_BLKADDR(sbi))
1659 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1660 f2fs_msg(sbi->sb, KERN_WARNING,
1661 "Found FS corruption, run fsck to fix.");
1665 /* start/end segment number in main_area */
1666 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1667 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1668 GET_SEGNO(sbi, end);
1669 cpc.reason = CP_DISCARD;
1670 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1672 /* do checkpoint to issue discard commands safely */
1673 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1674 cpc.trim_start = start_segno;
1676 if (sbi->discard_blks == 0)
1678 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1679 cpc.trim_end = end_segno;
1681 cpc.trim_end = min_t(unsigned int,
1682 rounddown(start_segno +
1683 BATCHED_TRIM_SEGMENTS(sbi),
1684 sbi->segs_per_sec) - 1, end_segno);
1686 mutex_lock(&sbi->gc_mutex);
1687 err = write_checkpoint(sbi, &cpc);
1688 mutex_unlock(&sbi->gc_mutex);
1695 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1699 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1701 struct curseg_info *curseg = CURSEG_I(sbi, type);
1702 if (curseg->next_blkoff < sbi->blocks_per_seg)
1707 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1710 return CURSEG_HOT_DATA;
1712 return CURSEG_HOT_NODE;
1715 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1717 if (p_type == DATA) {
1718 struct inode *inode = page->mapping->host;
1720 if (S_ISDIR(inode->i_mode))
1721 return CURSEG_HOT_DATA;
1723 return CURSEG_COLD_DATA;
1725 if (IS_DNODE(page) && is_cold_node(page))
1726 return CURSEG_WARM_NODE;
1728 return CURSEG_COLD_NODE;
1732 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1734 if (p_type == DATA) {
1735 struct inode *inode = page->mapping->host;
1737 if (S_ISDIR(inode->i_mode))
1738 return CURSEG_HOT_DATA;
1739 else if (is_cold_data(page) || file_is_cold(inode))
1740 return CURSEG_COLD_DATA;
1742 return CURSEG_WARM_DATA;
1745 return is_cold_node(page) ? CURSEG_WARM_NODE :
1748 return CURSEG_COLD_NODE;
1752 static int __get_segment_type(struct page *page, enum page_type p_type)
1754 switch (F2FS_P_SB(page)->active_logs) {
1756 return __get_segment_type_2(page, p_type);
1758 return __get_segment_type_4(page, p_type);
1760 /* NR_CURSEG_TYPE(6) logs by default */
1761 f2fs_bug_on(F2FS_P_SB(page),
1762 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1763 return __get_segment_type_6(page, p_type);
1766 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1767 block_t old_blkaddr, block_t *new_blkaddr,
1768 struct f2fs_summary *sum, int type)
1770 struct sit_info *sit_i = SIT_I(sbi);
1771 struct curseg_info *curseg = CURSEG_I(sbi, type);
1773 mutex_lock(&curseg->curseg_mutex);
1774 mutex_lock(&sit_i->sentry_lock);
1776 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1778 f2fs_wait_discard_bio(sbi, *new_blkaddr);
1781 * __add_sum_entry should be resided under the curseg_mutex
1782 * because, this function updates a summary entry in the
1783 * current summary block.
1785 __add_sum_entry(sbi, type, sum);
1787 __refresh_next_blkoff(sbi, curseg);
1789 stat_inc_block_count(sbi, curseg);
1792 * SIT information should be updated before segment allocation,
1793 * since SSR needs latest valid block information.
1795 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1797 if (!__has_curseg_space(sbi, type))
1798 sit_i->s_ops->allocate_segment(sbi, type, false);
1800 mutex_unlock(&sit_i->sentry_lock);
1802 if (page && IS_NODESEG(type))
1803 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1805 mutex_unlock(&curseg->curseg_mutex);
1808 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1810 int type = __get_segment_type(fio->page, fio->type);
1813 if (fio->type == NODE || fio->type == DATA)
1814 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1816 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1817 &fio->new_blkaddr, sum, type);
1819 /* writeout dirty page into bdev */
1820 err = f2fs_submit_page_mbio(fio);
1821 if (err == -EAGAIN) {
1822 fio->old_blkaddr = fio->new_blkaddr;
1826 if (fio->type == NODE || fio->type == DATA)
1827 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1830 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1832 struct f2fs_io_info fio = {
1836 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
1837 .old_blkaddr = page->index,
1838 .new_blkaddr = page->index,
1840 .encrypted_page = NULL,
1843 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1844 fio.op_flags &= ~REQ_META;
1846 set_page_writeback(page);
1847 f2fs_submit_page_mbio(&fio);
1850 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1852 struct f2fs_summary sum;
1854 set_summary(&sum, nid, 0, 0);
1855 do_write_page(&sum, fio);
1858 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1860 struct f2fs_sb_info *sbi = fio->sbi;
1861 struct f2fs_summary sum;
1862 struct node_info ni;
1864 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1865 get_node_info(sbi, dn->nid, &ni);
1866 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1867 do_write_page(&sum, fio);
1868 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1871 void rewrite_data_page(struct f2fs_io_info *fio)
1873 fio->new_blkaddr = fio->old_blkaddr;
1874 stat_inc_inplace_blocks(fio->sbi);
1875 f2fs_submit_page_mbio(fio);
1878 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1879 block_t old_blkaddr, block_t new_blkaddr,
1880 bool recover_curseg, bool recover_newaddr)
1882 struct sit_info *sit_i = SIT_I(sbi);
1883 struct curseg_info *curseg;
1884 unsigned int segno, old_cursegno;
1885 struct seg_entry *se;
1887 unsigned short old_blkoff;
1889 segno = GET_SEGNO(sbi, new_blkaddr);
1890 se = get_seg_entry(sbi, segno);
1893 if (!recover_curseg) {
1894 /* for recovery flow */
1895 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1896 if (old_blkaddr == NULL_ADDR)
1897 type = CURSEG_COLD_DATA;
1899 type = CURSEG_WARM_DATA;
1902 if (!IS_CURSEG(sbi, segno))
1903 type = CURSEG_WARM_DATA;
1906 curseg = CURSEG_I(sbi, type);
1908 mutex_lock(&curseg->curseg_mutex);
1909 mutex_lock(&sit_i->sentry_lock);
1911 old_cursegno = curseg->segno;
1912 old_blkoff = curseg->next_blkoff;
1914 /* change the current segment */
1915 if (segno != curseg->segno) {
1916 curseg->next_segno = segno;
1917 change_curseg(sbi, type, true);
1920 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1921 __add_sum_entry(sbi, type, sum);
1923 if (!recover_curseg || recover_newaddr)
1924 update_sit_entry(sbi, new_blkaddr, 1);
1925 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1926 update_sit_entry(sbi, old_blkaddr, -1);
1928 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1929 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1931 locate_dirty_segment(sbi, old_cursegno);
1933 if (recover_curseg) {
1934 if (old_cursegno != curseg->segno) {
1935 curseg->next_segno = old_cursegno;
1936 change_curseg(sbi, type, true);
1938 curseg->next_blkoff = old_blkoff;
1941 mutex_unlock(&sit_i->sentry_lock);
1942 mutex_unlock(&curseg->curseg_mutex);
1945 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1946 block_t old_addr, block_t new_addr,
1947 unsigned char version, bool recover_curseg,
1948 bool recover_newaddr)
1950 struct f2fs_summary sum;
1952 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1954 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1955 recover_curseg, recover_newaddr);
1957 f2fs_update_data_blkaddr(dn, new_addr);
1960 void f2fs_wait_on_page_writeback(struct page *page,
1961 enum page_type type, bool ordered)
1963 if (PageWriteback(page)) {
1964 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1966 f2fs_submit_merged_bio_cond(sbi, page->mapping->host,
1967 0, page->index, type, WRITE);
1969 wait_on_page_writeback(page);
1971 wait_for_stable_page(page);
1975 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1980 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1983 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1985 f2fs_wait_on_page_writeback(cpage, DATA, true);
1986 f2fs_put_page(cpage, 1);
1990 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1992 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1993 struct curseg_info *seg_i;
1994 unsigned char *kaddr;
1999 start = start_sum_block(sbi);
2001 page = get_meta_page(sbi, start++);
2002 kaddr = (unsigned char *)page_address(page);
2004 /* Step 1: restore nat cache */
2005 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2006 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2008 /* Step 2: restore sit cache */
2009 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2010 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2011 offset = 2 * SUM_JOURNAL_SIZE;
2013 /* Step 3: restore summary entries */
2014 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2015 unsigned short blk_off;
2018 seg_i = CURSEG_I(sbi, i);
2019 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2020 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2021 seg_i->next_segno = segno;
2022 reset_curseg(sbi, i, 0);
2023 seg_i->alloc_type = ckpt->alloc_type[i];
2024 seg_i->next_blkoff = blk_off;
2026 if (seg_i->alloc_type == SSR)
2027 blk_off = sbi->blocks_per_seg;
2029 for (j = 0; j < blk_off; j++) {
2030 struct f2fs_summary *s;
2031 s = (struct f2fs_summary *)(kaddr + offset);
2032 seg_i->sum_blk->entries[j] = *s;
2033 offset += SUMMARY_SIZE;
2034 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2038 f2fs_put_page(page, 1);
2041 page = get_meta_page(sbi, start++);
2042 kaddr = (unsigned char *)page_address(page);
2046 f2fs_put_page(page, 1);
2050 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2052 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2053 struct f2fs_summary_block *sum;
2054 struct curseg_info *curseg;
2056 unsigned short blk_off;
2057 unsigned int segno = 0;
2058 block_t blk_addr = 0;
2060 /* get segment number and block addr */
2061 if (IS_DATASEG(type)) {
2062 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2063 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2065 if (__exist_node_summaries(sbi))
2066 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2068 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2070 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2072 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2074 if (__exist_node_summaries(sbi))
2075 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2076 type - CURSEG_HOT_NODE);
2078 blk_addr = GET_SUM_BLOCK(sbi, segno);
2081 new = get_meta_page(sbi, blk_addr);
2082 sum = (struct f2fs_summary_block *)page_address(new);
2084 if (IS_NODESEG(type)) {
2085 if (__exist_node_summaries(sbi)) {
2086 struct f2fs_summary *ns = &sum->entries[0];
2088 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2090 ns->ofs_in_node = 0;
2095 err = restore_node_summary(sbi, segno, sum);
2097 f2fs_put_page(new, 1);
2103 /* set uncompleted segment to curseg */
2104 curseg = CURSEG_I(sbi, type);
2105 mutex_lock(&curseg->curseg_mutex);
2107 /* update journal info */
2108 down_write(&curseg->journal_rwsem);
2109 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2110 up_write(&curseg->journal_rwsem);
2112 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2113 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2114 curseg->next_segno = segno;
2115 reset_curseg(sbi, type, 0);
2116 curseg->alloc_type = ckpt->alloc_type[type];
2117 curseg->next_blkoff = blk_off;
2118 mutex_unlock(&curseg->curseg_mutex);
2119 f2fs_put_page(new, 1);
2123 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
2125 int type = CURSEG_HOT_DATA;
2128 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
2129 int npages = npages_for_summary_flush(sbi, true);
2132 ra_meta_pages(sbi, start_sum_block(sbi), npages,
2135 /* restore for compacted data summary */
2136 if (read_compacted_summaries(sbi))
2138 type = CURSEG_HOT_NODE;
2141 if (__exist_node_summaries(sbi))
2142 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
2143 NR_CURSEG_TYPE - type, META_CP, true);
2145 for (; type <= CURSEG_COLD_NODE; type++) {
2146 err = read_normal_summaries(sbi, type);
2154 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
2157 unsigned char *kaddr;
2158 struct f2fs_summary *summary;
2159 struct curseg_info *seg_i;
2160 int written_size = 0;
2163 page = grab_meta_page(sbi, blkaddr++);
2164 kaddr = (unsigned char *)page_address(page);
2166 /* Step 1: write nat cache */
2167 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2168 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
2169 written_size += SUM_JOURNAL_SIZE;
2171 /* Step 2: write sit cache */
2172 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2173 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
2174 written_size += SUM_JOURNAL_SIZE;
2176 /* Step 3: write summary entries */
2177 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2178 unsigned short blkoff;
2179 seg_i = CURSEG_I(sbi, i);
2180 if (sbi->ckpt->alloc_type[i] == SSR)
2181 blkoff = sbi->blocks_per_seg;
2183 blkoff = curseg_blkoff(sbi, i);
2185 for (j = 0; j < blkoff; j++) {
2187 page = grab_meta_page(sbi, blkaddr++);
2188 kaddr = (unsigned char *)page_address(page);
2191 summary = (struct f2fs_summary *)(kaddr + written_size);
2192 *summary = seg_i->sum_blk->entries[j];
2193 written_size += SUMMARY_SIZE;
2195 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
2199 set_page_dirty(page);
2200 f2fs_put_page(page, 1);
2205 set_page_dirty(page);
2206 f2fs_put_page(page, 1);
2210 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2211 block_t blkaddr, int type)
2214 if (IS_DATASEG(type))
2215 end = type + NR_CURSEG_DATA_TYPE;
2217 end = type + NR_CURSEG_NODE_TYPE;
2219 for (i = type; i < end; i++)
2220 write_current_sum_page(sbi, i, blkaddr + (i - type));
2223 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2225 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2226 write_compacted_summaries(sbi, start_blk);
2228 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2231 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2233 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2236 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2237 unsigned int val, int alloc)
2241 if (type == NAT_JOURNAL) {
2242 for (i = 0; i < nats_in_cursum(journal); i++) {
2243 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2246 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2247 return update_nats_in_cursum(journal, 1);
2248 } else if (type == SIT_JOURNAL) {
2249 for (i = 0; i < sits_in_cursum(journal); i++)
2250 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2252 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2253 return update_sits_in_cursum(journal, 1);
2258 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2261 return get_meta_page(sbi, current_sit_addr(sbi, segno));
2264 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2267 struct sit_info *sit_i = SIT_I(sbi);
2268 struct page *src_page, *dst_page;
2269 pgoff_t src_off, dst_off;
2270 void *src_addr, *dst_addr;
2272 src_off = current_sit_addr(sbi, start);
2273 dst_off = next_sit_addr(sbi, src_off);
2275 /* get current sit block page without lock */
2276 src_page = get_meta_page(sbi, src_off);
2277 dst_page = grab_meta_page(sbi, dst_off);
2278 f2fs_bug_on(sbi, PageDirty(src_page));
2280 src_addr = page_address(src_page);
2281 dst_addr = page_address(dst_page);
2282 memcpy(dst_addr, src_addr, PAGE_SIZE);
2284 set_page_dirty(dst_page);
2285 f2fs_put_page(src_page, 1);
2287 set_to_next_sit(sit_i, start);
2292 static struct sit_entry_set *grab_sit_entry_set(void)
2294 struct sit_entry_set *ses =
2295 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2298 INIT_LIST_HEAD(&ses->set_list);
2302 static void release_sit_entry_set(struct sit_entry_set *ses)
2304 list_del(&ses->set_list);
2305 kmem_cache_free(sit_entry_set_slab, ses);
2308 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2309 struct list_head *head)
2311 struct sit_entry_set *next = ses;
2313 if (list_is_last(&ses->set_list, head))
2316 list_for_each_entry_continue(next, head, set_list)
2317 if (ses->entry_cnt <= next->entry_cnt)
2320 list_move_tail(&ses->set_list, &next->set_list);
2323 static void add_sit_entry(unsigned int segno, struct list_head *head)
2325 struct sit_entry_set *ses;
2326 unsigned int start_segno = START_SEGNO(segno);
2328 list_for_each_entry(ses, head, set_list) {
2329 if (ses->start_segno == start_segno) {
2331 adjust_sit_entry_set(ses, head);
2336 ses = grab_sit_entry_set();
2338 ses->start_segno = start_segno;
2340 list_add(&ses->set_list, head);
2343 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2345 struct f2fs_sm_info *sm_info = SM_I(sbi);
2346 struct list_head *set_list = &sm_info->sit_entry_set;
2347 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2350 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2351 add_sit_entry(segno, set_list);
2354 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2356 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2357 struct f2fs_journal *journal = curseg->journal;
2360 down_write(&curseg->journal_rwsem);
2361 for (i = 0; i < sits_in_cursum(journal); i++) {
2365 segno = le32_to_cpu(segno_in_journal(journal, i));
2366 dirtied = __mark_sit_entry_dirty(sbi, segno);
2369 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2371 update_sits_in_cursum(journal, -i);
2372 up_write(&curseg->journal_rwsem);
2376 * CP calls this function, which flushes SIT entries including sit_journal,
2377 * and moves prefree segs to free segs.
2379 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2381 struct sit_info *sit_i = SIT_I(sbi);
2382 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2383 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2384 struct f2fs_journal *journal = curseg->journal;
2385 struct sit_entry_set *ses, *tmp;
2386 struct list_head *head = &SM_I(sbi)->sit_entry_set;
2387 bool to_journal = true;
2388 struct seg_entry *se;
2390 mutex_lock(&sit_i->sentry_lock);
2392 if (!sit_i->dirty_sentries)
2396 * add and account sit entries of dirty bitmap in sit entry
2399 add_sits_in_set(sbi);
2402 * if there are no enough space in journal to store dirty sit
2403 * entries, remove all entries from journal and add and account
2404 * them in sit entry set.
2406 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2407 remove_sits_in_journal(sbi);
2410 * there are two steps to flush sit entries:
2411 * #1, flush sit entries to journal in current cold data summary block.
2412 * #2, flush sit entries to sit page.
2414 list_for_each_entry_safe(ses, tmp, head, set_list) {
2415 struct page *page = NULL;
2416 struct f2fs_sit_block *raw_sit = NULL;
2417 unsigned int start_segno = ses->start_segno;
2418 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2419 (unsigned long)MAIN_SEGS(sbi));
2420 unsigned int segno = start_segno;
2423 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2427 down_write(&curseg->journal_rwsem);
2429 page = get_next_sit_page(sbi, start_segno);
2430 raw_sit = page_address(page);
2433 /* flush dirty sit entries in region of current sit set */
2434 for_each_set_bit_from(segno, bitmap, end) {
2435 int offset, sit_offset;
2437 se = get_seg_entry(sbi, segno);
2439 /* add discard candidates */
2440 if (cpc->reason != CP_DISCARD) {
2441 cpc->trim_start = segno;
2442 add_discard_addrs(sbi, cpc, false);
2446 offset = lookup_journal_in_cursum(journal,
2447 SIT_JOURNAL, segno, 1);
2448 f2fs_bug_on(sbi, offset < 0);
2449 segno_in_journal(journal, offset) =
2451 seg_info_to_raw_sit(se,
2452 &sit_in_journal(journal, offset));
2454 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2455 seg_info_to_raw_sit(se,
2456 &raw_sit->entries[sit_offset]);
2459 __clear_bit(segno, bitmap);
2460 sit_i->dirty_sentries--;
2465 up_write(&curseg->journal_rwsem);
2467 f2fs_put_page(page, 1);
2469 f2fs_bug_on(sbi, ses->entry_cnt);
2470 release_sit_entry_set(ses);
2473 f2fs_bug_on(sbi, !list_empty(head));
2474 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2476 if (cpc->reason == CP_DISCARD) {
2477 __u64 trim_start = cpc->trim_start;
2479 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2480 add_discard_addrs(sbi, cpc, false);
2482 cpc->trim_start = trim_start;
2484 mutex_unlock(&sit_i->sentry_lock);
2486 set_prefree_as_free_segments(sbi);
2489 static int build_sit_info(struct f2fs_sb_info *sbi)
2491 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2492 struct sit_info *sit_i;
2493 unsigned int sit_segs, start;
2495 unsigned int bitmap_size;
2497 /* allocate memory for SIT information */
2498 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2502 SM_I(sbi)->sit_info = sit_i;
2504 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2505 sizeof(struct seg_entry), GFP_KERNEL);
2506 if (!sit_i->sentries)
2509 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2510 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2511 if (!sit_i->dirty_sentries_bitmap)
2514 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2515 sit_i->sentries[start].cur_valid_map
2516 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2517 sit_i->sentries[start].ckpt_valid_map
2518 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2519 if (!sit_i->sentries[start].cur_valid_map ||
2520 !sit_i->sentries[start].ckpt_valid_map)
2523 #ifdef CONFIG_F2FS_CHECK_FS
2524 sit_i->sentries[start].cur_valid_map_mir
2525 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2526 if (!sit_i->sentries[start].cur_valid_map_mir)
2530 if (f2fs_discard_en(sbi)) {
2531 sit_i->sentries[start].discard_map
2532 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2533 if (!sit_i->sentries[start].discard_map)
2538 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2539 if (!sit_i->tmp_map)
2542 if (sbi->segs_per_sec > 1) {
2543 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2544 sizeof(struct sec_entry), GFP_KERNEL);
2545 if (!sit_i->sec_entries)
2549 /* get information related with SIT */
2550 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2552 /* setup SIT bitmap from ckeckpoint pack */
2553 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2554 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2556 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2557 if (!sit_i->sit_bitmap)
2560 #ifdef CONFIG_F2FS_CHECK_FS
2561 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2562 if (!sit_i->sit_bitmap_mir)
2566 /* init SIT information */
2567 sit_i->s_ops = &default_salloc_ops;
2569 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2570 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2571 sit_i->written_valid_blocks = 0;
2572 sit_i->bitmap_size = bitmap_size;
2573 sit_i->dirty_sentries = 0;
2574 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2575 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2576 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2577 mutex_init(&sit_i->sentry_lock);
2581 static int build_free_segmap(struct f2fs_sb_info *sbi)
2583 struct free_segmap_info *free_i;
2584 unsigned int bitmap_size, sec_bitmap_size;
2586 /* allocate memory for free segmap information */
2587 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2591 SM_I(sbi)->free_info = free_i;
2593 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2594 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2595 if (!free_i->free_segmap)
2598 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2599 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2600 if (!free_i->free_secmap)
2603 /* set all segments as dirty temporarily */
2604 memset(free_i->free_segmap, 0xff, bitmap_size);
2605 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2607 /* init free segmap information */
2608 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2609 free_i->free_segments = 0;
2610 free_i->free_sections = 0;
2611 spin_lock_init(&free_i->segmap_lock);
2615 static int build_curseg(struct f2fs_sb_info *sbi)
2617 struct curseg_info *array;
2620 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2624 SM_I(sbi)->curseg_array = array;
2626 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2627 mutex_init(&array[i].curseg_mutex);
2628 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2629 if (!array[i].sum_blk)
2631 init_rwsem(&array[i].journal_rwsem);
2632 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2634 if (!array[i].journal)
2636 array[i].segno = NULL_SEGNO;
2637 array[i].next_blkoff = 0;
2639 return restore_curseg_summaries(sbi);
2642 static void build_sit_entries(struct f2fs_sb_info *sbi)
2644 struct sit_info *sit_i = SIT_I(sbi);
2645 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2646 struct f2fs_journal *journal = curseg->journal;
2647 struct seg_entry *se;
2648 struct f2fs_sit_entry sit;
2649 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2650 unsigned int i, start, end;
2651 unsigned int readed, start_blk = 0;
2654 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2657 start = start_blk * sit_i->sents_per_block;
2658 end = (start_blk + readed) * sit_i->sents_per_block;
2660 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2661 struct f2fs_sit_block *sit_blk;
2664 se = &sit_i->sentries[start];
2665 page = get_current_sit_page(sbi, start);
2666 sit_blk = (struct f2fs_sit_block *)page_address(page);
2667 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2668 f2fs_put_page(page, 1);
2670 check_block_count(sbi, start, &sit);
2671 seg_info_from_raw_sit(se, &sit);
2673 /* build discard map only one time */
2674 if (f2fs_discard_en(sbi)) {
2675 memcpy(se->discard_map, se->cur_valid_map,
2676 SIT_VBLOCK_MAP_SIZE);
2677 sbi->discard_blks += sbi->blocks_per_seg -
2681 if (sbi->segs_per_sec > 1)
2682 get_sec_entry(sbi, start)->valid_blocks +=
2685 start_blk += readed;
2686 } while (start_blk < sit_blk_cnt);
2688 down_read(&curseg->journal_rwsem);
2689 for (i = 0; i < sits_in_cursum(journal); i++) {
2690 unsigned int old_valid_blocks;
2692 start = le32_to_cpu(segno_in_journal(journal, i));
2693 se = &sit_i->sentries[start];
2694 sit = sit_in_journal(journal, i);
2696 old_valid_blocks = se->valid_blocks;
2698 check_block_count(sbi, start, &sit);
2699 seg_info_from_raw_sit(se, &sit);
2701 if (f2fs_discard_en(sbi)) {
2702 memcpy(se->discard_map, se->cur_valid_map,
2703 SIT_VBLOCK_MAP_SIZE);
2704 sbi->discard_blks += old_valid_blocks -
2708 if (sbi->segs_per_sec > 1)
2709 get_sec_entry(sbi, start)->valid_blocks +=
2710 se->valid_blocks - old_valid_blocks;
2712 up_read(&curseg->journal_rwsem);
2715 static void init_free_segmap(struct f2fs_sb_info *sbi)
2720 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2721 struct seg_entry *sentry = get_seg_entry(sbi, start);
2722 if (!sentry->valid_blocks)
2723 __set_free(sbi, start);
2725 SIT_I(sbi)->written_valid_blocks +=
2726 sentry->valid_blocks;
2729 /* set use the current segments */
2730 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2731 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2732 __set_test_and_inuse(sbi, curseg_t->segno);
2736 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2738 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2739 struct free_segmap_info *free_i = FREE_I(sbi);
2740 unsigned int segno = 0, offset = 0;
2741 unsigned short valid_blocks;
2744 /* find dirty segment based on free segmap */
2745 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2746 if (segno >= MAIN_SEGS(sbi))
2749 valid_blocks = get_valid_blocks(sbi, segno, 0);
2750 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2752 if (valid_blocks > sbi->blocks_per_seg) {
2753 f2fs_bug_on(sbi, 1);
2756 mutex_lock(&dirty_i->seglist_lock);
2757 __locate_dirty_segment(sbi, segno, DIRTY);
2758 mutex_unlock(&dirty_i->seglist_lock);
2762 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2764 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2765 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2767 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2768 if (!dirty_i->victim_secmap)
2773 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2775 struct dirty_seglist_info *dirty_i;
2776 unsigned int bitmap_size, i;
2778 /* allocate memory for dirty segments list information */
2779 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2783 SM_I(sbi)->dirty_info = dirty_i;
2784 mutex_init(&dirty_i->seglist_lock);
2786 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2788 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2789 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2790 if (!dirty_i->dirty_segmap[i])
2794 init_dirty_segmap(sbi);
2795 return init_victim_secmap(sbi);
2799 * Update min, max modified time for cost-benefit GC algorithm
2801 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2803 struct sit_info *sit_i = SIT_I(sbi);
2806 mutex_lock(&sit_i->sentry_lock);
2808 sit_i->min_mtime = LLONG_MAX;
2810 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2812 unsigned long long mtime = 0;
2814 for (i = 0; i < sbi->segs_per_sec; i++)
2815 mtime += get_seg_entry(sbi, segno + i)->mtime;
2817 mtime = div_u64(mtime, sbi->segs_per_sec);
2819 if (sit_i->min_mtime > mtime)
2820 sit_i->min_mtime = mtime;
2822 sit_i->max_mtime = get_mtime(sbi);
2823 mutex_unlock(&sit_i->sentry_lock);
2826 int build_segment_manager(struct f2fs_sb_info *sbi)
2828 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2829 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2830 struct f2fs_sm_info *sm_info;
2833 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2838 sbi->sm_info = sm_info;
2839 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2840 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2841 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2842 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2843 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2844 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2845 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2846 sm_info->rec_prefree_segments = sm_info->main_segments *
2847 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2848 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2849 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2851 if (!test_opt(sbi, LFS))
2852 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2853 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2854 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2856 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2858 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2860 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2861 err = create_flush_cmd_control(sbi);
2866 err = create_discard_cmd_control(sbi);
2870 err = build_sit_info(sbi);
2873 err = build_free_segmap(sbi);
2876 err = build_curseg(sbi);
2880 /* reinit free segmap based on SIT */
2881 build_sit_entries(sbi);
2883 init_free_segmap(sbi);
2884 err = build_dirty_segmap(sbi);
2888 init_min_max_mtime(sbi);
2892 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2893 enum dirty_type dirty_type)
2895 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2897 mutex_lock(&dirty_i->seglist_lock);
2898 kvfree(dirty_i->dirty_segmap[dirty_type]);
2899 dirty_i->nr_dirty[dirty_type] = 0;
2900 mutex_unlock(&dirty_i->seglist_lock);
2903 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2905 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2906 kvfree(dirty_i->victim_secmap);
2909 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2911 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2917 /* discard pre-free/dirty segments list */
2918 for (i = 0; i < NR_DIRTY_TYPE; i++)
2919 discard_dirty_segmap(sbi, i);
2921 destroy_victim_secmap(sbi);
2922 SM_I(sbi)->dirty_info = NULL;
2926 static void destroy_curseg(struct f2fs_sb_info *sbi)
2928 struct curseg_info *array = SM_I(sbi)->curseg_array;
2933 SM_I(sbi)->curseg_array = NULL;
2934 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2935 kfree(array[i].sum_blk);
2936 kfree(array[i].journal);
2941 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2943 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2946 SM_I(sbi)->free_info = NULL;
2947 kvfree(free_i->free_segmap);
2948 kvfree(free_i->free_secmap);
2952 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2954 struct sit_info *sit_i = SIT_I(sbi);
2960 if (sit_i->sentries) {
2961 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2962 kfree(sit_i->sentries[start].cur_valid_map);
2963 #ifdef CONFIG_F2FS_CHECK_FS
2964 kfree(sit_i->sentries[start].cur_valid_map_mir);
2966 kfree(sit_i->sentries[start].ckpt_valid_map);
2967 kfree(sit_i->sentries[start].discard_map);
2970 kfree(sit_i->tmp_map);
2972 kvfree(sit_i->sentries);
2973 kvfree(sit_i->sec_entries);
2974 kvfree(sit_i->dirty_sentries_bitmap);
2976 SM_I(sbi)->sit_info = NULL;
2977 kfree(sit_i->sit_bitmap);
2978 #ifdef CONFIG_F2FS_CHECK_FS
2979 kfree(sit_i->sit_bitmap_mir);
2984 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2986 struct f2fs_sm_info *sm_info = SM_I(sbi);
2990 destroy_flush_cmd_control(sbi, true);
2991 destroy_discard_cmd_control(sbi, true);
2992 destroy_dirty_segmap(sbi);
2993 destroy_curseg(sbi);
2994 destroy_free_segmap(sbi);
2995 destroy_sit_info(sbi);
2996 sbi->sm_info = NULL;
3000 int __init create_segment_manager_caches(void)
3002 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3003 sizeof(struct discard_entry));
3004 if (!discard_entry_slab)
3007 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3008 sizeof(struct discard_cmd));
3009 if (!discard_cmd_slab)
3010 goto destroy_discard_entry;
3012 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3013 sizeof(struct sit_entry_set));
3014 if (!sit_entry_set_slab)
3015 goto destroy_discard_cmd;
3017 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3018 sizeof(struct inmem_pages));
3019 if (!inmem_entry_slab)
3020 goto destroy_sit_entry_set;
3023 destroy_sit_entry_set:
3024 kmem_cache_destroy(sit_entry_set_slab);
3025 destroy_discard_cmd:
3026 kmem_cache_destroy(discard_cmd_slab);
3027 destroy_discard_entry:
3028 kmem_cache_destroy(discard_entry_slab);
3033 void destroy_segment_manager_caches(void)
3035 kmem_cache_destroy(sit_entry_set_slab);
3036 kmem_cache_destroy(discard_cmd_slab);
3037 kmem_cache_destroy(discard_entry_slab);
3038 kmem_cache_destroy(inmem_entry_slab);