4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
19 #include <linux/freezer.h>
20 #include <linux/sched.h>
27 #include <trace/events/f2fs.h>
29 #define __reverse_ffz(x) __reverse_ffs(~(x))
31 static struct kmem_cache *discard_entry_slab;
32 static struct kmem_cache *discard_cmd_slab;
33 static struct kmem_cache *sit_entry_set_slab;
34 static struct kmem_cache *inmem_entry_slab;
36 static unsigned long __reverse_ulong(unsigned char *str)
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
41 #if BITS_PER_LONG == 64
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
55 static inline unsigned long __reverse_ffs(unsigned long word)
59 #if BITS_PER_LONG == 64
60 if ((word & 0xffffffff00000000UL) == 0)
65 if ((word & 0xffff0000) == 0)
70 if ((word & 0xff00) == 0)
75 if ((word & 0xf0) == 0)
80 if ((word & 0xc) == 0)
85 if ((word & 0x2) == 0)
91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
93 * @size must be integral times of unsigned long.
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
102 const unsigned long *p = addr + BIT_WORD(offset);
103 unsigned long result = size;
109 size -= (offset & ~(BITS_PER_LONG - 1));
110 offset %= BITS_PER_LONG;
116 tmp = __reverse_ulong((unsigned char *)p);
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
124 if (size <= BITS_PER_LONG)
126 size -= BITS_PER_LONG;
132 return result - size + __reverse_ffs(tmp);
135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
138 const unsigned long *p = addr + BIT_WORD(offset);
139 unsigned long result = size;
145 size -= (offset & ~(BITS_PER_LONG - 1));
146 offset %= BITS_PER_LONG;
152 tmp = __reverse_ulong((unsigned char *)p);
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
161 if (size <= BITS_PER_LONG)
163 size -= BITS_PER_LONG;
169 return result - size + __reverse_ffz(tmp);
172 bool need_SSR(struct f2fs_sb_info *sbi)
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
178 if (test_opt(sbi, LFS))
180 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
184 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
187 void register_inmem_page(struct inode *inode, struct page *page)
189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
190 struct f2fs_inode_info *fi = F2FS_I(inode);
191 struct inmem_pages *new;
193 f2fs_trace_pid(page);
195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 SetPagePrivate(page);
198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
200 /* add atomic page indices to the list */
202 INIT_LIST_HEAD(&new->list);
204 /* increase reference count with clean state */
205 mutex_lock(&fi->inmem_lock);
207 list_add_tail(&new->list, &fi->inmem_pages);
208 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
209 if (list_empty(&fi->inmem_ilist))
210 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
211 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
212 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
213 mutex_unlock(&fi->inmem_lock);
215 trace_f2fs_register_inmem_page(page, INMEM);
218 static int __revoke_inmem_pages(struct inode *inode,
219 struct list_head *head, bool drop, bool recover)
221 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
222 struct inmem_pages *cur, *tmp;
225 list_for_each_entry_safe(cur, tmp, head, list) {
226 struct page *page = cur->page;
229 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
234 struct dnode_of_data dn;
237 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
239 set_new_dnode(&dn, inode, NULL, NULL, 0);
240 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
242 if (err == -ENOMEM) {
243 congestion_wait(BLK_RW_ASYNC, HZ/50);
250 get_node_info(sbi, dn.nid, &ni);
251 if (cur->old_addr == NEW_ADDR) {
252 invalidate_blocks(sbi, dn.data_blkaddr);
253 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
255 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
256 cur->old_addr, ni.version, true, true);
260 /* we don't need to invalidate this in the sccessful status */
262 ClearPageUptodate(page);
263 set_page_private(page, 0);
264 ClearPagePrivate(page);
265 f2fs_put_page(page, 1);
267 list_del(&cur->list);
268 kmem_cache_free(inmem_entry_slab, cur);
269 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
274 void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
276 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
278 struct f2fs_inode_info *fi;
280 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
281 if (list_empty(head)) {
282 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
285 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
286 inode = igrab(&fi->vfs_inode);
287 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
290 drop_inmem_pages(inode);
293 congestion_wait(BLK_RW_ASYNC, HZ/50);
298 void drop_inmem_pages(struct inode *inode)
300 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
301 struct f2fs_inode_info *fi = F2FS_I(inode);
303 mutex_lock(&fi->inmem_lock);
304 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
305 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
306 if (!list_empty(&fi->inmem_ilist))
307 list_del_init(&fi->inmem_ilist);
308 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
309 mutex_unlock(&fi->inmem_lock);
311 clear_inode_flag(inode, FI_ATOMIC_FILE);
312 clear_inode_flag(inode, FI_HOT_DATA);
313 stat_dec_atomic_write(inode);
316 void drop_inmem_page(struct inode *inode, struct page *page)
318 struct f2fs_inode_info *fi = F2FS_I(inode);
319 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
320 struct list_head *head = &fi->inmem_pages;
321 struct inmem_pages *cur = NULL;
323 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
325 mutex_lock(&fi->inmem_lock);
326 list_for_each_entry(cur, head, list) {
327 if (cur->page == page)
331 f2fs_bug_on(sbi, !cur || cur->page != page);
332 list_del(&cur->list);
333 mutex_unlock(&fi->inmem_lock);
335 dec_page_count(sbi, F2FS_INMEM_PAGES);
336 kmem_cache_free(inmem_entry_slab, cur);
338 ClearPageUptodate(page);
339 set_page_private(page, 0);
340 ClearPagePrivate(page);
341 f2fs_put_page(page, 0);
343 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
346 static int __commit_inmem_pages(struct inode *inode,
347 struct list_head *revoke_list)
349 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
350 struct f2fs_inode_info *fi = F2FS_I(inode);
351 struct inmem_pages *cur, *tmp;
352 struct f2fs_io_info fio = {
357 .op_flags = REQ_SYNC | REQ_PRIO,
358 .io_type = FS_DATA_IO,
360 pgoff_t last_idx = ULONG_MAX;
363 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
364 struct page *page = cur->page;
367 if (page->mapping == inode->i_mapping) {
368 trace_f2fs_commit_inmem_page(page, INMEM);
370 set_page_dirty(page);
371 f2fs_wait_on_page_writeback(page, DATA, true);
372 if (clear_page_dirty_for_io(page)) {
373 inode_dec_dirty_pages(inode);
374 remove_dirty_inode(inode);
378 fio.old_blkaddr = NULL_ADDR;
379 fio.encrypted_page = NULL;
380 fio.need_lock = LOCK_DONE;
381 err = do_write_data_page(&fio);
383 if (err == -ENOMEM) {
384 congestion_wait(BLK_RW_ASYNC, HZ/50);
391 /* record old blkaddr for revoking */
392 cur->old_addr = fio.old_blkaddr;
393 last_idx = page->index;
396 list_move_tail(&cur->list, revoke_list);
399 if (last_idx != ULONG_MAX)
400 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
403 __revoke_inmem_pages(inode, revoke_list, false, false);
408 int commit_inmem_pages(struct inode *inode)
410 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
411 struct f2fs_inode_info *fi = F2FS_I(inode);
412 struct list_head revoke_list;
415 INIT_LIST_HEAD(&revoke_list);
416 f2fs_balance_fs(sbi, true);
419 set_inode_flag(inode, FI_ATOMIC_COMMIT);
421 mutex_lock(&fi->inmem_lock);
422 err = __commit_inmem_pages(inode, &revoke_list);
426 * try to revoke all committed pages, but still we could fail
427 * due to no memory or other reason, if that happened, EAGAIN
428 * will be returned, which means in such case, transaction is
429 * already not integrity, caller should use journal to do the
430 * recovery or rewrite & commit last transaction. For other
431 * error number, revoking was done by filesystem itself.
433 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
437 /* drop all uncommitted pages */
438 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
440 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
441 if (!list_empty(&fi->inmem_ilist))
442 list_del_init(&fi->inmem_ilist);
443 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
444 mutex_unlock(&fi->inmem_lock);
446 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
453 * This function balances dirty node and dentry pages.
454 * In addition, it controls garbage collection.
456 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
458 #ifdef CONFIG_F2FS_FAULT_INJECTION
459 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
460 f2fs_show_injection_info(FAULT_CHECKPOINT);
461 f2fs_stop_checkpoint(sbi, false);
465 /* balance_fs_bg is able to be pending */
466 if (need && excess_cached_nats(sbi))
467 f2fs_balance_fs_bg(sbi);
470 * We should do GC or end up with checkpoint, if there are so many dirty
471 * dir/node pages without enough free segments.
473 if (has_not_enough_free_secs(sbi, 0, 0)) {
474 mutex_lock(&sbi->gc_mutex);
475 f2fs_gc(sbi, false, false, NULL_SEGNO);
479 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
481 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
484 /* try to shrink extent cache when there is no enough memory */
485 if (!available_free_memory(sbi, EXTENT_CACHE))
486 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
488 /* check the # of cached NAT entries */
489 if (!available_free_memory(sbi, NAT_ENTRIES))
490 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
492 if (!available_free_memory(sbi, FREE_NIDS))
493 try_to_free_nids(sbi, MAX_FREE_NIDS);
495 build_free_nids(sbi, false, false);
497 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
500 /* checkpoint is the only way to shrink partial cached entries */
501 if (!available_free_memory(sbi, NAT_ENTRIES) ||
502 !available_free_memory(sbi, INO_ENTRIES) ||
503 excess_prefree_segs(sbi) ||
504 excess_dirty_nats(sbi) ||
505 f2fs_time_over(sbi, CP_TIME)) {
506 if (test_opt(sbi, DATA_FLUSH)) {
507 struct blk_plug plug;
509 blk_start_plug(&plug);
510 sync_dirty_inodes(sbi, FILE_INODE);
511 blk_finish_plug(&plug);
513 f2fs_sync_fs(sbi->sb, true);
514 stat_inc_bg_cp_count(sbi->stat_info);
518 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
519 struct block_device *bdev)
521 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
524 bio->bi_rw = REQ_OP_WRITE;
526 ret = submit_bio_wait(WRITE_FLUSH, bio);
529 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
530 test_opt(sbi, FLUSH_MERGE), ret);
534 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
540 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
542 for (i = 0; i < sbi->s_ndevs; i++) {
543 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
545 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
552 static int issue_flush_thread(void *data)
554 struct f2fs_sb_info *sbi = data;
555 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
556 wait_queue_head_t *q = &fcc->flush_wait_queue;
558 if (kthread_should_stop())
561 sb_start_intwrite(sbi->sb);
563 if (!llist_empty(&fcc->issue_list)) {
564 struct flush_cmd *cmd, *next;
567 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
568 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
570 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
572 ret = submit_flush_wait(sbi, cmd->ino);
573 atomic_inc(&fcc->issued_flush);
575 llist_for_each_entry_safe(cmd, next,
576 fcc->dispatch_list, llnode) {
578 complete(&cmd->wait);
580 fcc->dispatch_list = NULL;
583 sb_end_intwrite(sbi->sb);
585 wait_event_interruptible(*q,
586 kthread_should_stop() || !llist_empty(&fcc->issue_list));
590 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
592 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
593 struct flush_cmd cmd;
596 if (test_opt(sbi, NOBARRIER))
599 if (!test_opt(sbi, FLUSH_MERGE)) {
600 ret = submit_flush_wait(sbi, ino);
601 atomic_inc(&fcc->issued_flush);
605 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
606 ret = submit_flush_wait(sbi, ino);
607 atomic_dec(&fcc->issing_flush);
609 atomic_inc(&fcc->issued_flush);
614 init_completion(&cmd.wait);
616 llist_add(&cmd.llnode, &fcc->issue_list);
618 /* update issue_list before we wake up issue_flush thread */
621 if (waitqueue_active(&fcc->flush_wait_queue))
622 wake_up(&fcc->flush_wait_queue);
624 if (fcc->f2fs_issue_flush) {
625 wait_for_completion(&cmd.wait);
626 atomic_dec(&fcc->issing_flush);
628 struct llist_node *list;
630 list = llist_del_all(&fcc->issue_list);
632 wait_for_completion(&cmd.wait);
633 atomic_dec(&fcc->issing_flush);
635 struct flush_cmd *tmp, *next;
637 ret = submit_flush_wait(sbi, ino);
639 llist_for_each_entry_safe(tmp, next, list, llnode) {
642 atomic_dec(&fcc->issing_flush);
646 complete(&tmp->wait);
654 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
656 dev_t dev = sbi->sb->s_bdev->bd_dev;
657 struct flush_cmd_control *fcc;
660 if (SM_I(sbi)->fcc_info) {
661 fcc = SM_I(sbi)->fcc_info;
662 if (fcc->f2fs_issue_flush)
667 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
670 atomic_set(&fcc->issued_flush, 0);
671 atomic_set(&fcc->issing_flush, 0);
672 init_waitqueue_head(&fcc->flush_wait_queue);
673 init_llist_head(&fcc->issue_list);
674 SM_I(sbi)->fcc_info = fcc;
675 if (!test_opt(sbi, FLUSH_MERGE))
679 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
680 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
681 if (IS_ERR(fcc->f2fs_issue_flush)) {
682 err = PTR_ERR(fcc->f2fs_issue_flush);
684 SM_I(sbi)->fcc_info = NULL;
691 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
693 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
695 if (fcc && fcc->f2fs_issue_flush) {
696 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
698 fcc->f2fs_issue_flush = NULL;
699 kthread_stop(flush_thread);
703 SM_I(sbi)->fcc_info = NULL;
707 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
714 for (i = 1; i < sbi->s_ndevs; i++) {
715 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
717 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
721 spin_lock(&sbi->dev_lock);
722 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
723 spin_unlock(&sbi->dev_lock);
729 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
730 enum dirty_type dirty_type)
732 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
734 /* need not be added */
735 if (IS_CURSEG(sbi, segno))
738 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
739 dirty_i->nr_dirty[dirty_type]++;
741 if (dirty_type == DIRTY) {
742 struct seg_entry *sentry = get_seg_entry(sbi, segno);
743 enum dirty_type t = sentry->type;
745 if (unlikely(t >= DIRTY)) {
749 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
750 dirty_i->nr_dirty[t]++;
754 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
755 enum dirty_type dirty_type)
757 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
759 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
760 dirty_i->nr_dirty[dirty_type]--;
762 if (dirty_type == DIRTY) {
763 struct seg_entry *sentry = get_seg_entry(sbi, segno);
764 enum dirty_type t = sentry->type;
766 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
767 dirty_i->nr_dirty[t]--;
769 if (get_valid_blocks(sbi, segno, true) == 0)
770 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
771 dirty_i->victim_secmap);
776 * Should not occur error such as -ENOMEM.
777 * Adding dirty entry into seglist is not critical operation.
778 * If a given segment is one of current working segments, it won't be added.
780 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
782 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
783 unsigned short valid_blocks;
785 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
788 mutex_lock(&dirty_i->seglist_lock);
790 valid_blocks = get_valid_blocks(sbi, segno, false);
792 if (valid_blocks == 0) {
793 __locate_dirty_segment(sbi, segno, PRE);
794 __remove_dirty_segment(sbi, segno, DIRTY);
795 } else if (valid_blocks < sbi->blocks_per_seg) {
796 __locate_dirty_segment(sbi, segno, DIRTY);
798 /* Recovery routine with SSR needs this */
799 __remove_dirty_segment(sbi, segno, DIRTY);
802 mutex_unlock(&dirty_i->seglist_lock);
805 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
806 struct block_device *bdev, block_t lstart,
807 block_t start, block_t len)
809 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
810 struct list_head *pend_list;
811 struct discard_cmd *dc;
813 f2fs_bug_on(sbi, !len);
815 pend_list = &dcc->pend_list[plist_idx(len)];
817 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
818 INIT_LIST_HEAD(&dc->list);
826 init_completion(&dc->wait);
827 list_add_tail(&dc->list, pend_list);
828 atomic_inc(&dcc->discard_cmd_cnt);
829 dcc->undiscard_blks += len;
834 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
835 struct block_device *bdev, block_t lstart,
836 block_t start, block_t len,
837 struct rb_node *parent, struct rb_node **p)
839 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
840 struct discard_cmd *dc;
842 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
844 rb_link_node(&dc->rb_node, parent, p);
845 rb_insert_color(&dc->rb_node, &dcc->root);
850 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
851 struct discard_cmd *dc)
853 if (dc->state == D_DONE)
854 atomic_dec(&dcc->issing_discard);
857 rb_erase(&dc->rb_node, &dcc->root);
858 dcc->undiscard_blks -= dc->len;
860 kmem_cache_free(discard_cmd_slab, dc);
862 atomic_dec(&dcc->discard_cmd_cnt);
865 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
866 struct discard_cmd *dc)
868 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
870 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
872 f2fs_bug_on(sbi, dc->ref);
874 if (dc->error == -EOPNOTSUPP)
878 f2fs_msg(sbi->sb, KERN_INFO,
879 "Issue discard(%u, %u, %u) failed, ret: %d",
880 dc->lstart, dc->start, dc->len, dc->error);
881 __detach_discard_cmd(dcc, dc);
884 static void f2fs_submit_discard_endio(struct bio *bio)
886 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
888 dc->error = bio->bi_error;
890 complete_all(&dc->wait);
894 /* copied from block/blk-lib.c in 4.10-rc1 */
895 static int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
896 sector_t nr_sects, gfp_t gfp_mask, int flags,
899 struct request_queue *q = bdev_get_queue(bdev);
900 struct bio *bio = *biop;
901 unsigned int granularity;
902 int op = REQ_WRITE | REQ_DISCARD;
909 if (!blk_queue_discard(q))
912 if (flags & BLKDEV_DISCARD_SECURE) {
913 if (!blk_queue_secdiscard(q))
918 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
919 if ((sector | nr_sects) & bs_mask)
922 /* Zero-sector (unknown) and one-sector granularities are the same. */
923 granularity = max(q->limits.discard_granularity >> 9, 1U);
924 alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
927 unsigned int req_sects;
928 sector_t end_sect, tmp;
930 /* Make sure bi_size doesn't overflow */
931 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
934 * If splitting a request, and the next starting sector would be
935 * misaligned, stop the discard at the previous aligned sector.
937 end_sect = sector + req_sects;
939 if (req_sects < nr_sects &&
940 sector_div(tmp, granularity) != alignment) {
941 end_sect = end_sect - alignment;
942 sector_div(end_sect, granularity);
943 end_sect = end_sect * granularity + alignment;
944 req_sects = end_sect - sector;
948 int ret = submit_bio_wait(op, bio);
953 bio = bio_alloc(GFP_NOIO | __GFP_NOFAIL, 1);
954 bio->bi_iter.bi_sector = sector;
956 bio_set_op_attrs(bio, op, 0);
958 bio->bi_iter.bi_size = req_sects << 9;
959 nr_sects -= req_sects;
963 * We can loop for a long time in here, if someone does
964 * full device discards (like mkfs). Be nice and allow
965 * us to schedule out to avoid softlocking if preempt
975 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
976 block_t start, block_t end)
978 #ifdef CONFIG_F2FS_CHECK_FS
979 struct seg_entry *sentry;
982 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
986 segno = GET_SEGNO(sbi, blk);
987 sentry = get_seg_entry(sbi, segno);
988 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
990 if (end < START_BLOCK(sbi, segno + 1))
991 size = GET_BLKOFF_FROM_SEG0(sbi, end);
994 map = (unsigned long *)(sentry->cur_valid_map);
995 offset = __find_rev_next_bit(map, size, offset);
996 f2fs_bug_on(sbi, offset != size);
997 blk = START_BLOCK(sbi, segno + 1);
1002 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1003 struct discard_policy *dpolicy,
1004 int discard_type, unsigned int granularity)
1007 dpolicy->type = discard_type;
1008 dpolicy->sync = true;
1009 dpolicy->granularity = granularity;
1011 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1012 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1014 if (discard_type == DPOLICY_BG) {
1015 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1016 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1017 dpolicy->io_aware = true;
1018 dpolicy->sync = false;
1019 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1020 dpolicy->granularity = 1;
1021 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1023 } else if (discard_type == DPOLICY_FORCE) {
1024 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1025 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1026 dpolicy->io_aware = false;
1027 } else if (discard_type == DPOLICY_FSTRIM) {
1028 dpolicy->io_aware = false;
1029 } else if (discard_type == DPOLICY_UMOUNT) {
1030 dpolicy->io_aware = false;
1035 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1036 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
1037 struct discard_policy *dpolicy,
1038 struct discard_cmd *dc)
1040 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1041 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1042 &(dcc->fstrim_list) : &(dcc->wait_list);
1043 struct bio *bio = NULL;
1044 int flag = dpolicy->sync ? REQ_SYNC : 0;
1046 if (dc->state != D_PREP)
1049 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
1051 dc->error = __blkdev_issue_discard(dc->bdev,
1052 SECTOR_FROM_BLOCK(dc->start),
1053 SECTOR_FROM_BLOCK(dc->len),
1056 /* should keep before submission to avoid D_DONE right away */
1057 dc->state = D_SUBMIT;
1058 atomic_inc(&dcc->issued_discard);
1059 atomic_inc(&dcc->issing_discard);
1061 bio->bi_private = dc;
1062 bio->bi_end_io = f2fs_submit_discard_endio;
1063 submit_bio(flag, bio);
1064 list_move_tail(&dc->list, wait_list);
1065 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
1067 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1070 __remove_discard_cmd(sbi, dc);
1074 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1075 struct block_device *bdev, block_t lstart,
1076 block_t start, block_t len,
1077 struct rb_node **insert_p,
1078 struct rb_node *insert_parent)
1080 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1082 struct rb_node *parent = NULL;
1083 struct discard_cmd *dc = NULL;
1085 if (insert_p && insert_parent) {
1086 parent = insert_parent;
1091 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1093 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1100 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1101 struct discard_cmd *dc)
1103 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1106 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1107 struct discard_cmd *dc, block_t blkaddr)
1109 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1110 struct discard_info di = dc->di;
1111 bool modified = false;
1113 if (dc->state == D_DONE || dc->len == 1) {
1114 __remove_discard_cmd(sbi, dc);
1118 dcc->undiscard_blks -= di.len;
1120 if (blkaddr > di.lstart) {
1121 dc->len = blkaddr - dc->lstart;
1122 dcc->undiscard_blks += dc->len;
1123 __relocate_discard_cmd(dcc, dc);
1127 if (blkaddr < di.lstart + di.len - 1) {
1129 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1130 di.start + blkaddr + 1 - di.lstart,
1131 di.lstart + di.len - 1 - blkaddr,
1137 dcc->undiscard_blks += dc->len;
1138 __relocate_discard_cmd(dcc, dc);
1143 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1144 struct block_device *bdev, block_t lstart,
1145 block_t start, block_t len)
1147 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1148 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1149 struct discard_cmd *dc;
1150 struct discard_info di = {0};
1151 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1152 block_t end = lstart + len;
1154 mutex_lock(&dcc->cmd_lock);
1156 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1158 (struct rb_entry **)&prev_dc,
1159 (struct rb_entry **)&next_dc,
1160 &insert_p, &insert_parent, true);
1166 di.len = next_dc ? next_dc->lstart - lstart : len;
1167 di.len = min(di.len, len);
1172 struct rb_node *node;
1173 bool merged = false;
1174 struct discard_cmd *tdc = NULL;
1177 di.lstart = prev_dc->lstart + prev_dc->len;
1178 if (di.lstart < lstart)
1180 if (di.lstart >= end)
1183 if (!next_dc || next_dc->lstart > end)
1184 di.len = end - di.lstart;
1186 di.len = next_dc->lstart - di.lstart;
1187 di.start = start + di.lstart - lstart;
1193 if (prev_dc && prev_dc->state == D_PREP &&
1194 prev_dc->bdev == bdev &&
1195 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1196 prev_dc->di.len += di.len;
1197 dcc->undiscard_blks += di.len;
1198 __relocate_discard_cmd(dcc, prev_dc);
1204 if (next_dc && next_dc->state == D_PREP &&
1205 next_dc->bdev == bdev &&
1206 __is_discard_front_mergeable(&di, &next_dc->di)) {
1207 next_dc->di.lstart = di.lstart;
1208 next_dc->di.len += di.len;
1209 next_dc->di.start = di.start;
1210 dcc->undiscard_blks += di.len;
1211 __relocate_discard_cmd(dcc, next_dc);
1213 __remove_discard_cmd(sbi, tdc);
1218 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1219 di.len, NULL, NULL);
1226 node = rb_next(&prev_dc->rb_node);
1227 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1230 mutex_unlock(&dcc->cmd_lock);
1233 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1234 struct block_device *bdev, block_t blkstart, block_t blklen)
1236 block_t lblkstart = blkstart;
1238 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1241 int devi = f2fs_target_device_index(sbi, blkstart);
1243 blkstart -= FDEV(devi).start_blk;
1245 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1249 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1250 struct discard_policy *dpolicy)
1252 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1253 struct list_head *pend_list;
1254 struct discard_cmd *dc, *tmp;
1255 struct blk_plug plug;
1256 int i, iter = 0, issued = 0;
1257 bool io_interrupted = false;
1259 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1260 if (i + 1 < dpolicy->granularity)
1262 pend_list = &dcc->pend_list[i];
1264 mutex_lock(&dcc->cmd_lock);
1265 if (list_empty(pend_list))
1267 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1268 blk_start_plug(&plug);
1269 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1270 f2fs_bug_on(sbi, dc->state != D_PREP);
1272 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1274 io_interrupted = true;
1278 __submit_discard_cmd(sbi, dpolicy, dc);
1281 if (++iter >= dpolicy->max_requests)
1284 blk_finish_plug(&plug);
1286 mutex_unlock(&dcc->cmd_lock);
1288 if (iter >= dpolicy->max_requests)
1292 if (!issued && io_interrupted)
1298 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1300 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1301 struct list_head *pend_list;
1302 struct discard_cmd *dc, *tmp;
1304 bool dropped = false;
1306 mutex_lock(&dcc->cmd_lock);
1307 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1308 pend_list = &dcc->pend_list[i];
1309 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1310 f2fs_bug_on(sbi, dc->state != D_PREP);
1311 __remove_discard_cmd(sbi, dc);
1315 mutex_unlock(&dcc->cmd_lock);
1320 void drop_discard_cmd(struct f2fs_sb_info *sbi)
1322 __drop_discard_cmd(sbi);
1325 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1326 struct discard_cmd *dc)
1328 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1329 unsigned int len = 0;
1331 wait_for_completion_io(&dc->wait);
1332 mutex_lock(&dcc->cmd_lock);
1333 f2fs_bug_on(sbi, dc->state != D_DONE);
1338 __remove_discard_cmd(sbi, dc);
1340 mutex_unlock(&dcc->cmd_lock);
1345 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1346 struct discard_policy *dpolicy,
1347 block_t start, block_t end)
1349 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1350 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1351 &(dcc->fstrim_list) : &(dcc->wait_list);
1352 struct discard_cmd *dc, *tmp;
1354 unsigned int trimmed = 0;
1359 mutex_lock(&dcc->cmd_lock);
1360 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1361 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1363 if (dc->len < dpolicy->granularity)
1365 if (dc->state == D_DONE && !dc->ref) {
1366 wait_for_completion_io(&dc->wait);
1369 __remove_discard_cmd(sbi, dc);
1376 mutex_unlock(&dcc->cmd_lock);
1379 trimmed += __wait_one_discard_bio(sbi, dc);
1386 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1387 struct discard_policy *dpolicy)
1389 struct discard_policy dp;
1392 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1397 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1398 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1399 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1400 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1403 /* This should be covered by global mutex, &sit_i->sentry_lock */
1404 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1406 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1407 struct discard_cmd *dc;
1408 bool need_wait = false;
1410 mutex_lock(&dcc->cmd_lock);
1411 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1413 if (dc->state == D_PREP) {
1414 __punch_discard_cmd(sbi, dc, blkaddr);
1420 mutex_unlock(&dcc->cmd_lock);
1423 __wait_one_discard_bio(sbi, dc);
1426 void stop_discard_thread(struct f2fs_sb_info *sbi)
1428 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1430 if (dcc && dcc->f2fs_issue_discard) {
1431 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1433 dcc->f2fs_issue_discard = NULL;
1434 kthread_stop(discard_thread);
1438 /* This comes from f2fs_put_super */
1439 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1441 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1442 struct discard_policy dpolicy;
1445 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1446 dcc->discard_granularity);
1447 __issue_discard_cmd(sbi, &dpolicy);
1448 dropped = __drop_discard_cmd(sbi);
1450 /* just to make sure there is no pending discard commands */
1451 __wait_all_discard_cmd(sbi, NULL);
1455 static int issue_discard_thread(void *data)
1457 struct f2fs_sb_info *sbi = data;
1458 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1459 wait_queue_head_t *q = &dcc->discard_wait_queue;
1460 struct discard_policy dpolicy;
1461 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1467 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1468 dcc->discard_granularity);
1470 wait_event_interruptible_timeout(*q,
1471 kthread_should_stop() || freezing(current) ||
1473 msecs_to_jiffies(wait_ms));
1474 if (try_to_freeze())
1476 if (f2fs_readonly(sbi->sb))
1478 if (kthread_should_stop())
1481 if (dcc->discard_wake)
1482 dcc->discard_wake = 0;
1484 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1485 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1487 sb_start_intwrite(sbi->sb);
1489 issued = __issue_discard_cmd(sbi, &dpolicy);
1491 __wait_all_discard_cmd(sbi, &dpolicy);
1492 wait_ms = dpolicy.min_interval;
1494 wait_ms = dpolicy.max_interval;
1497 sb_end_intwrite(sbi->sb);
1499 } while (!kthread_should_stop());
1503 #ifdef CONFIG_BLK_DEV_ZONED
1504 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1505 struct block_device *bdev, block_t blkstart, block_t blklen)
1507 sector_t sector, nr_sects;
1508 block_t lblkstart = blkstart;
1512 devi = f2fs_target_device_index(sbi, blkstart);
1513 blkstart -= FDEV(devi).start_blk;
1517 * We need to know the type of the zone: for conventional zones,
1518 * use regular discard if the drive supports it. For sequential
1519 * zones, reset the zone write pointer.
1521 switch (get_blkz_type(sbi, bdev, blkstart)) {
1523 case BLK_ZONE_TYPE_CONVENTIONAL:
1524 if (!blk_queue_discard(bdev_get_queue(bdev)))
1526 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1527 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1528 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1529 sector = SECTOR_FROM_BLOCK(blkstart);
1530 nr_sects = SECTOR_FROM_BLOCK(blklen);
1532 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1533 nr_sects != bdev_zone_sectors(bdev)) {
1534 f2fs_msg(sbi->sb, KERN_INFO,
1535 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1536 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1540 trace_f2fs_issue_reset_zone(bdev, blkstart);
1541 return blkdev_reset_zones(bdev, sector,
1542 nr_sects, GFP_NOFS);
1544 /* Unknown zone type: broken device ? */
1550 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1551 struct block_device *bdev, block_t blkstart, block_t blklen)
1553 #ifdef CONFIG_BLK_DEV_ZONED
1554 if (f2fs_sb_has_blkzoned(sbi->sb) &&
1555 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1556 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1558 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1561 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1562 block_t blkstart, block_t blklen)
1564 sector_t start = blkstart, len = 0;
1565 struct block_device *bdev;
1566 struct seg_entry *se;
1567 unsigned int offset;
1571 bdev = f2fs_target_device(sbi, blkstart, NULL);
1573 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1575 struct block_device *bdev2 =
1576 f2fs_target_device(sbi, i, NULL);
1578 if (bdev2 != bdev) {
1579 err = __issue_discard_async(sbi, bdev,
1589 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1590 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1592 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1593 sbi->discard_blks--;
1597 err = __issue_discard_async(sbi, bdev, start, len);
1601 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1604 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1605 int max_blocks = sbi->blocks_per_seg;
1606 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1607 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1608 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1609 unsigned long *discard_map = (unsigned long *)se->discard_map;
1610 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1611 unsigned int start = 0, end = -1;
1612 bool force = (cpc->reason & CP_DISCARD);
1613 struct discard_entry *de = NULL;
1614 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1617 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1621 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1622 SM_I(sbi)->dcc_info->nr_discards >=
1623 SM_I(sbi)->dcc_info->max_discards)
1627 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1628 for (i = 0; i < entries; i++)
1629 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1630 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1632 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1633 SM_I(sbi)->dcc_info->max_discards) {
1634 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1635 if (start >= max_blocks)
1638 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1639 if (force && start && end != max_blocks
1640 && (end - start) < cpc->trim_minlen)
1647 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1649 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1650 list_add_tail(&de->list, head);
1653 for (i = start; i < end; i++)
1654 __set_bit_le(i, (void *)de->discard_map);
1656 SM_I(sbi)->dcc_info->nr_discards += end - start;
1661 void release_discard_addrs(struct f2fs_sb_info *sbi)
1663 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1664 struct discard_entry *entry, *this;
1667 list_for_each_entry_safe(entry, this, head, list) {
1668 list_del(&entry->list);
1669 kmem_cache_free(discard_entry_slab, entry);
1674 * Should call clear_prefree_segments after checkpoint is done.
1676 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1678 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1681 mutex_lock(&dirty_i->seglist_lock);
1682 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1683 __set_test_and_free(sbi, segno);
1684 mutex_unlock(&dirty_i->seglist_lock);
1687 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1689 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1690 struct list_head *head = &dcc->entry_list;
1691 struct discard_entry *entry, *this;
1692 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1693 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1694 unsigned int start = 0, end = -1;
1695 unsigned int secno, start_segno;
1696 bool force = (cpc->reason & CP_DISCARD);
1698 mutex_lock(&dirty_i->seglist_lock);
1702 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1703 if (start >= MAIN_SEGS(sbi))
1705 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1708 for (i = start; i < end; i++)
1709 clear_bit(i, prefree_map);
1711 dirty_i->nr_dirty[PRE] -= end - start;
1713 if (!test_opt(sbi, DISCARD))
1716 if (force && start >= cpc->trim_start &&
1717 (end - 1) <= cpc->trim_end)
1720 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1721 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1722 (end - start) << sbi->log_blocks_per_seg);
1726 secno = GET_SEC_FROM_SEG(sbi, start);
1727 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1728 if (!IS_CURSEC(sbi, secno) &&
1729 !get_valid_blocks(sbi, start, true))
1730 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1731 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1733 start = start_segno + sbi->segs_per_sec;
1739 mutex_unlock(&dirty_i->seglist_lock);
1741 /* send small discards */
1742 list_for_each_entry_safe(entry, this, head, list) {
1743 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1744 bool is_valid = test_bit_le(0, entry->discard_map);
1748 next_pos = find_next_zero_bit_le(entry->discard_map,
1749 sbi->blocks_per_seg, cur_pos);
1750 len = next_pos - cur_pos;
1752 if (f2fs_sb_has_blkzoned(sbi->sb) ||
1753 (force && len < cpc->trim_minlen))
1756 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1760 next_pos = find_next_bit_le(entry->discard_map,
1761 sbi->blocks_per_seg, cur_pos);
1765 is_valid = !is_valid;
1767 if (cur_pos < sbi->blocks_per_seg)
1770 list_del(&entry->list);
1771 dcc->nr_discards -= total_len;
1772 kmem_cache_free(discard_entry_slab, entry);
1775 wake_up_discard_thread(sbi, false);
1778 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1780 dev_t dev = sbi->sb->s_bdev->bd_dev;
1781 struct discard_cmd_control *dcc;
1784 if (SM_I(sbi)->dcc_info) {
1785 dcc = SM_I(sbi)->dcc_info;
1789 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1793 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1794 INIT_LIST_HEAD(&dcc->entry_list);
1795 for (i = 0; i < MAX_PLIST_NUM; i++)
1796 INIT_LIST_HEAD(&dcc->pend_list[i]);
1797 INIT_LIST_HEAD(&dcc->wait_list);
1798 INIT_LIST_HEAD(&dcc->fstrim_list);
1799 mutex_init(&dcc->cmd_lock);
1800 atomic_set(&dcc->issued_discard, 0);
1801 atomic_set(&dcc->issing_discard, 0);
1802 atomic_set(&dcc->discard_cmd_cnt, 0);
1803 dcc->nr_discards = 0;
1804 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1805 dcc->undiscard_blks = 0;
1806 dcc->root = RB_ROOT;
1808 init_waitqueue_head(&dcc->discard_wait_queue);
1809 SM_I(sbi)->dcc_info = dcc;
1811 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1812 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1813 if (IS_ERR(dcc->f2fs_issue_discard)) {
1814 err = PTR_ERR(dcc->f2fs_issue_discard);
1816 SM_I(sbi)->dcc_info = NULL;
1823 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1825 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1830 stop_discard_thread(sbi);
1833 SM_I(sbi)->dcc_info = NULL;
1836 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1838 struct sit_info *sit_i = SIT_I(sbi);
1840 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1841 sit_i->dirty_sentries++;
1848 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1849 unsigned int segno, int modified)
1851 struct seg_entry *se = get_seg_entry(sbi, segno);
1854 __mark_sit_entry_dirty(sbi, segno);
1857 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1859 struct seg_entry *se;
1860 unsigned int segno, offset;
1861 long int new_vblocks;
1863 #ifdef CONFIG_F2FS_CHECK_FS
1867 segno = GET_SEGNO(sbi, blkaddr);
1869 se = get_seg_entry(sbi, segno);
1870 new_vblocks = se->valid_blocks + del;
1871 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1873 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1874 (new_vblocks > sbi->blocks_per_seg)));
1876 se->valid_blocks = new_vblocks;
1877 se->mtime = get_mtime(sbi);
1878 SIT_I(sbi)->max_mtime = se->mtime;
1880 /* Update valid block bitmap */
1882 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1883 #ifdef CONFIG_F2FS_CHECK_FS
1884 mir_exist = f2fs_test_and_set_bit(offset,
1885 se->cur_valid_map_mir);
1886 if (unlikely(exist != mir_exist)) {
1887 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1888 "when setting bitmap, blk:%u, old bit:%d",
1890 f2fs_bug_on(sbi, 1);
1893 if (unlikely(exist)) {
1894 f2fs_msg(sbi->sb, KERN_ERR,
1895 "Bitmap was wrongly set, blk:%u", blkaddr);
1896 f2fs_bug_on(sbi, 1);
1901 if (f2fs_discard_en(sbi) &&
1902 !f2fs_test_and_set_bit(offset, se->discard_map))
1903 sbi->discard_blks--;
1905 /* don't overwrite by SSR to keep node chain */
1906 if (IS_NODESEG(se->type)) {
1907 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1908 se->ckpt_valid_blocks++;
1911 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1912 #ifdef CONFIG_F2FS_CHECK_FS
1913 mir_exist = f2fs_test_and_clear_bit(offset,
1914 se->cur_valid_map_mir);
1915 if (unlikely(exist != mir_exist)) {
1916 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1917 "when clearing bitmap, blk:%u, old bit:%d",
1919 f2fs_bug_on(sbi, 1);
1922 if (unlikely(!exist)) {
1923 f2fs_msg(sbi->sb, KERN_ERR,
1924 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1925 f2fs_bug_on(sbi, 1);
1930 if (f2fs_discard_en(sbi) &&
1931 f2fs_test_and_clear_bit(offset, se->discard_map))
1932 sbi->discard_blks++;
1934 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1935 se->ckpt_valid_blocks += del;
1937 __mark_sit_entry_dirty(sbi, segno);
1939 /* update total number of valid blocks to be written in ckpt area */
1940 SIT_I(sbi)->written_valid_blocks += del;
1942 if (sbi->segs_per_sec > 1)
1943 get_sec_entry(sbi, segno)->valid_blocks += del;
1946 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1948 unsigned int segno = GET_SEGNO(sbi, addr);
1949 struct sit_info *sit_i = SIT_I(sbi);
1951 f2fs_bug_on(sbi, addr == NULL_ADDR);
1952 if (addr == NEW_ADDR)
1955 /* add it into sit main buffer */
1956 down_write(&sit_i->sentry_lock);
1958 update_sit_entry(sbi, addr, -1);
1960 /* add it into dirty seglist */
1961 locate_dirty_segment(sbi, segno);
1963 up_write(&sit_i->sentry_lock);
1966 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1968 struct sit_info *sit_i = SIT_I(sbi);
1969 unsigned int segno, offset;
1970 struct seg_entry *se;
1973 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1976 down_read(&sit_i->sentry_lock);
1978 segno = GET_SEGNO(sbi, blkaddr);
1979 se = get_seg_entry(sbi, segno);
1980 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1982 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1985 up_read(&sit_i->sentry_lock);
1991 * This function should be resided under the curseg_mutex lock
1993 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1994 struct f2fs_summary *sum)
1996 struct curseg_info *curseg = CURSEG_I(sbi, type);
1997 void *addr = curseg->sum_blk;
1998 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1999 memcpy(addr, sum, sizeof(struct f2fs_summary));
2003 * Calculate the number of current summary pages for writing
2005 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2007 int valid_sum_count = 0;
2010 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2011 if (sbi->ckpt->alloc_type[i] == SSR)
2012 valid_sum_count += sbi->blocks_per_seg;
2015 valid_sum_count += le16_to_cpu(
2016 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2018 valid_sum_count += curseg_blkoff(sbi, i);
2022 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2023 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2024 if (valid_sum_count <= sum_in_page)
2026 else if ((valid_sum_count - sum_in_page) <=
2027 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2033 * Caller should put this summary page
2035 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2037 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
2040 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
2042 struct page *page = grab_meta_page(sbi, blk_addr);
2044 memcpy(page_address(page), src, PAGE_SIZE);
2045 set_page_dirty(page);
2046 f2fs_put_page(page, 1);
2049 static void write_sum_page(struct f2fs_sb_info *sbi,
2050 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2052 update_meta_page(sbi, (void *)sum_blk, blk_addr);
2055 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2056 int type, block_t blk_addr)
2058 struct curseg_info *curseg = CURSEG_I(sbi, type);
2059 struct page *page = grab_meta_page(sbi, blk_addr);
2060 struct f2fs_summary_block *src = curseg->sum_blk;
2061 struct f2fs_summary_block *dst;
2063 dst = (struct f2fs_summary_block *)page_address(page);
2065 mutex_lock(&curseg->curseg_mutex);
2067 down_read(&curseg->journal_rwsem);
2068 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2069 up_read(&curseg->journal_rwsem);
2071 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2072 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2074 mutex_unlock(&curseg->curseg_mutex);
2076 set_page_dirty(page);
2077 f2fs_put_page(page, 1);
2080 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2082 struct curseg_info *curseg = CURSEG_I(sbi, type);
2083 unsigned int segno = curseg->segno + 1;
2084 struct free_segmap_info *free_i = FREE_I(sbi);
2086 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2087 return !test_bit(segno, free_i->free_segmap);
2092 * Find a new segment from the free segments bitmap to right order
2093 * This function should be returned with success, otherwise BUG
2095 static void get_new_segment(struct f2fs_sb_info *sbi,
2096 unsigned int *newseg, bool new_sec, int dir)
2098 struct free_segmap_info *free_i = FREE_I(sbi);
2099 unsigned int segno, secno, zoneno;
2100 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2101 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2102 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2103 unsigned int left_start = hint;
2108 spin_lock(&free_i->segmap_lock);
2110 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2111 segno = find_next_zero_bit(free_i->free_segmap,
2112 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2113 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2117 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2118 if (secno >= MAIN_SECS(sbi)) {
2119 if (dir == ALLOC_RIGHT) {
2120 secno = find_next_zero_bit(free_i->free_secmap,
2122 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2125 left_start = hint - 1;
2131 while (test_bit(left_start, free_i->free_secmap)) {
2132 if (left_start > 0) {
2136 left_start = find_next_zero_bit(free_i->free_secmap,
2138 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2143 segno = GET_SEG_FROM_SEC(sbi, secno);
2144 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2146 /* give up on finding another zone */
2149 if (sbi->secs_per_zone == 1)
2151 if (zoneno == old_zoneno)
2153 if (dir == ALLOC_LEFT) {
2154 if (!go_left && zoneno + 1 >= total_zones)
2156 if (go_left && zoneno == 0)
2159 for (i = 0; i < NR_CURSEG_TYPE; i++)
2160 if (CURSEG_I(sbi, i)->zone == zoneno)
2163 if (i < NR_CURSEG_TYPE) {
2164 /* zone is in user, try another */
2166 hint = zoneno * sbi->secs_per_zone - 1;
2167 else if (zoneno + 1 >= total_zones)
2170 hint = (zoneno + 1) * sbi->secs_per_zone;
2172 goto find_other_zone;
2175 /* set it as dirty segment in free segmap */
2176 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2177 __set_inuse(sbi, segno);
2179 spin_unlock(&free_i->segmap_lock);
2182 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2184 struct curseg_info *curseg = CURSEG_I(sbi, type);
2185 struct summary_footer *sum_footer;
2187 curseg->segno = curseg->next_segno;
2188 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2189 curseg->next_blkoff = 0;
2190 curseg->next_segno = NULL_SEGNO;
2192 sum_footer = &(curseg->sum_blk->footer);
2193 memset(sum_footer, 0, sizeof(struct summary_footer));
2194 if (IS_DATASEG(type))
2195 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2196 if (IS_NODESEG(type))
2197 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2198 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2201 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2203 /* if segs_per_sec is large than 1, we need to keep original policy. */
2204 if (sbi->segs_per_sec != 1)
2205 return CURSEG_I(sbi, type)->segno;
2207 if (test_opt(sbi, NOHEAP) &&
2208 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2211 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2212 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2214 /* find segments from 0 to reuse freed segments */
2215 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2218 return CURSEG_I(sbi, type)->segno;
2222 * Allocate a current working segment.
2223 * This function always allocates a free segment in LFS manner.
2225 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2227 struct curseg_info *curseg = CURSEG_I(sbi, type);
2228 unsigned int segno = curseg->segno;
2229 int dir = ALLOC_LEFT;
2231 write_sum_page(sbi, curseg->sum_blk,
2232 GET_SUM_BLOCK(sbi, segno));
2233 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2236 if (test_opt(sbi, NOHEAP))
2239 segno = __get_next_segno(sbi, type);
2240 get_new_segment(sbi, &segno, new_sec, dir);
2241 curseg->next_segno = segno;
2242 reset_curseg(sbi, type, 1);
2243 curseg->alloc_type = LFS;
2246 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2247 struct curseg_info *seg, block_t start)
2249 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2250 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2251 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2252 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2253 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2256 for (i = 0; i < entries; i++)
2257 target_map[i] = ckpt_map[i] | cur_map[i];
2259 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2261 seg->next_blkoff = pos;
2265 * If a segment is written by LFS manner, next block offset is just obtained
2266 * by increasing the current block offset. However, if a segment is written by
2267 * SSR manner, next block offset obtained by calling __next_free_blkoff
2269 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2270 struct curseg_info *seg)
2272 if (seg->alloc_type == SSR)
2273 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2279 * This function always allocates a used segment(from dirty seglist) by SSR
2280 * manner, so it should recover the existing segment information of valid blocks
2282 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2284 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2285 struct curseg_info *curseg = CURSEG_I(sbi, type);
2286 unsigned int new_segno = curseg->next_segno;
2287 struct f2fs_summary_block *sum_node;
2288 struct page *sum_page;
2290 write_sum_page(sbi, curseg->sum_blk,
2291 GET_SUM_BLOCK(sbi, curseg->segno));
2292 __set_test_and_inuse(sbi, new_segno);
2294 mutex_lock(&dirty_i->seglist_lock);
2295 __remove_dirty_segment(sbi, new_segno, PRE);
2296 __remove_dirty_segment(sbi, new_segno, DIRTY);
2297 mutex_unlock(&dirty_i->seglist_lock);
2299 reset_curseg(sbi, type, 1);
2300 curseg->alloc_type = SSR;
2301 __next_free_blkoff(sbi, curseg, 0);
2303 sum_page = get_sum_page(sbi, new_segno);
2304 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2305 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2306 f2fs_put_page(sum_page, 1);
2309 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2311 struct curseg_info *curseg = CURSEG_I(sbi, type);
2312 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2313 unsigned segno = NULL_SEGNO;
2315 bool reversed = false;
2317 /* need_SSR() already forces to do this */
2318 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2319 curseg->next_segno = segno;
2323 /* For node segments, let's do SSR more intensively */
2324 if (IS_NODESEG(type)) {
2325 if (type >= CURSEG_WARM_NODE) {
2327 i = CURSEG_COLD_NODE;
2329 i = CURSEG_HOT_NODE;
2331 cnt = NR_CURSEG_NODE_TYPE;
2333 if (type >= CURSEG_WARM_DATA) {
2335 i = CURSEG_COLD_DATA;
2337 i = CURSEG_HOT_DATA;
2339 cnt = NR_CURSEG_DATA_TYPE;
2342 for (; cnt-- > 0; reversed ? i-- : i++) {
2345 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2346 curseg->next_segno = segno;
2354 * flush out current segment and replace it with new segment
2355 * This function should be returned with success, otherwise BUG
2357 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2358 int type, bool force)
2360 struct curseg_info *curseg = CURSEG_I(sbi, type);
2363 new_curseg(sbi, type, true);
2364 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2365 type == CURSEG_WARM_NODE)
2366 new_curseg(sbi, type, false);
2367 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2368 new_curseg(sbi, type, false);
2369 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2370 change_curseg(sbi, type);
2372 new_curseg(sbi, type, false);
2374 stat_inc_seg_type(sbi, curseg);
2377 void allocate_new_segments(struct f2fs_sb_info *sbi)
2379 struct curseg_info *curseg;
2380 unsigned int old_segno;
2383 down_write(&SIT_I(sbi)->sentry_lock);
2385 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2386 curseg = CURSEG_I(sbi, i);
2387 old_segno = curseg->segno;
2388 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2389 locate_dirty_segment(sbi, old_segno);
2392 up_write(&SIT_I(sbi)->sentry_lock);
2395 static const struct segment_allocation default_salloc_ops = {
2396 .allocate_segment = allocate_segment_by_default,
2399 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2401 __u64 trim_start = cpc->trim_start;
2402 bool has_candidate = false;
2404 down_write(&SIT_I(sbi)->sentry_lock);
2405 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2406 if (add_discard_addrs(sbi, cpc, true)) {
2407 has_candidate = true;
2411 up_write(&SIT_I(sbi)->sentry_lock);
2413 cpc->trim_start = trim_start;
2414 return has_candidate;
2417 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2418 struct discard_policy *dpolicy,
2419 unsigned int start, unsigned int end)
2421 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2422 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2423 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2424 struct discard_cmd *dc;
2425 struct blk_plug plug;
2431 mutex_lock(&dcc->cmd_lock);
2432 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
2434 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
2436 (struct rb_entry **)&prev_dc,
2437 (struct rb_entry **)&next_dc,
2438 &insert_p, &insert_parent, true);
2442 blk_start_plug(&plug);
2444 while (dc && dc->lstart <= end) {
2445 struct rb_node *node;
2447 if (dc->len < dpolicy->granularity)
2450 if (dc->state != D_PREP) {
2451 list_move_tail(&dc->list, &dcc->fstrim_list);
2455 __submit_discard_cmd(sbi, dpolicy, dc);
2457 if (++issued >= dpolicy->max_requests) {
2458 start = dc->lstart + dc->len;
2460 blk_finish_plug(&plug);
2461 mutex_unlock(&dcc->cmd_lock);
2462 __wait_all_discard_cmd(sbi, NULL);
2463 congestion_wait(BLK_RW_ASYNC, HZ/50);
2467 node = rb_next(&dc->rb_node);
2468 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2470 if (fatal_signal_pending(current))
2474 blk_finish_plug(&plug);
2475 mutex_unlock(&dcc->cmd_lock);
2478 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2480 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2481 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2482 unsigned int start_segno, end_segno;
2483 block_t start_block, end_block;
2484 struct cp_control cpc;
2485 struct discard_policy dpolicy;
2486 unsigned long long trimmed = 0;
2489 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2492 if (end <= MAIN_BLKADDR(sbi))
2495 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2496 f2fs_msg(sbi->sb, KERN_WARNING,
2497 "Found FS corruption, run fsck to fix.");
2501 /* start/end segment number in main_area */
2502 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2503 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2504 GET_SEGNO(sbi, end);
2506 cpc.reason = CP_DISCARD;
2507 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2508 cpc.trim_start = start_segno;
2509 cpc.trim_end = end_segno;
2511 if (sbi->discard_blks == 0)
2514 mutex_lock(&sbi->gc_mutex);
2515 err = write_checkpoint(sbi, &cpc);
2516 mutex_unlock(&sbi->gc_mutex);
2520 start_block = START_BLOCK(sbi, start_segno);
2521 end_block = START_BLOCK(sbi, end_segno + 1);
2523 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2524 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2527 * We filed discard candidates, but actually we don't need to wait for
2528 * all of them, since they'll be issued in idle time along with runtime
2529 * discard option. User configuration looks like using runtime discard
2530 * or periodic fstrim instead of it.
2532 if (!test_opt(sbi, DISCARD)) {
2533 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2534 start_block, end_block);
2535 range->len = F2FS_BLK_TO_BYTES(trimmed);
2541 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2543 struct curseg_info *curseg = CURSEG_I(sbi, type);
2544 if (curseg->next_blkoff < sbi->blocks_per_seg)
2549 int rw_hint_to_seg_type(enum rw_hint hint)
2552 case WRITE_LIFE_SHORT:
2553 return CURSEG_HOT_DATA;
2554 case WRITE_LIFE_EXTREME:
2555 return CURSEG_COLD_DATA;
2557 return CURSEG_WARM_DATA;
2561 /* This returns write hints for each segment type. This hints will be
2562 * passed down to block layer. There are mapping tables which depend on
2563 * the mount option 'whint_mode'.
2565 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2567 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2571 * META WRITE_LIFE_NOT_SET
2575 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2576 * extension list " "
2579 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2580 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2581 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2582 * WRITE_LIFE_NONE " "
2583 * WRITE_LIFE_MEDIUM " "
2584 * WRITE_LIFE_LONG " "
2587 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2588 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2589 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2590 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2591 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2592 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2594 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2598 * META WRITE_LIFE_MEDIUM;
2599 * HOT_NODE WRITE_LIFE_NOT_SET
2601 * COLD_NODE WRITE_LIFE_NONE
2602 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2603 * extension list " "
2606 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2607 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2608 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2609 * WRITE_LIFE_NONE " "
2610 * WRITE_LIFE_MEDIUM " "
2611 * WRITE_LIFE_LONG " "
2614 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2615 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2616 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2617 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2618 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2619 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2622 enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2623 enum page_type type, enum temp_type temp)
2625 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2628 return WRITE_LIFE_NOT_SET;
2629 else if (temp == HOT)
2630 return WRITE_LIFE_SHORT;
2631 else if (temp == COLD)
2632 return WRITE_LIFE_EXTREME;
2634 return WRITE_LIFE_NOT_SET;
2636 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2639 return WRITE_LIFE_LONG;
2640 else if (temp == HOT)
2641 return WRITE_LIFE_SHORT;
2642 else if (temp == COLD)
2643 return WRITE_LIFE_EXTREME;
2644 } else if (type == NODE) {
2645 if (temp == WARM || temp == HOT)
2646 return WRITE_LIFE_NOT_SET;
2647 else if (temp == COLD)
2648 return WRITE_LIFE_NONE;
2649 } else if (type == META) {
2650 return WRITE_LIFE_MEDIUM;
2653 return WRITE_LIFE_NOT_SET;
2656 static int __get_segment_type_2(struct f2fs_io_info *fio)
2658 if (fio->type == DATA)
2659 return CURSEG_HOT_DATA;
2661 return CURSEG_HOT_NODE;
2664 static int __get_segment_type_4(struct f2fs_io_info *fio)
2666 if (fio->type == DATA) {
2667 struct inode *inode = fio->page->mapping->host;
2669 if (S_ISDIR(inode->i_mode))
2670 return CURSEG_HOT_DATA;
2672 return CURSEG_COLD_DATA;
2674 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2675 return CURSEG_WARM_NODE;
2677 return CURSEG_COLD_NODE;
2681 static int __get_segment_type_6(struct f2fs_io_info *fio)
2683 if (fio->type == DATA) {
2684 struct inode *inode = fio->page->mapping->host;
2686 if (is_cold_data(fio->page) || file_is_cold(inode))
2687 return CURSEG_COLD_DATA;
2688 if (file_is_hot(inode) ||
2689 is_inode_flag_set(inode, FI_HOT_DATA))
2690 return CURSEG_HOT_DATA;
2691 /* rw_hint_to_seg_type(inode->i_write_hint); */
2692 return CURSEG_WARM_DATA;
2694 if (IS_DNODE(fio->page))
2695 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2697 return CURSEG_COLD_NODE;
2701 static int __get_segment_type(struct f2fs_io_info *fio)
2705 switch (F2FS_OPTION(fio->sbi).active_logs) {
2707 type = __get_segment_type_2(fio);
2710 type = __get_segment_type_4(fio);
2713 type = __get_segment_type_6(fio);
2716 f2fs_bug_on(fio->sbi, true);
2721 else if (IS_WARM(type))
2728 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2729 block_t old_blkaddr, block_t *new_blkaddr,
2730 struct f2fs_summary *sum, int type,
2731 struct f2fs_io_info *fio, bool add_list)
2733 struct sit_info *sit_i = SIT_I(sbi);
2734 struct curseg_info *curseg = CURSEG_I(sbi, type);
2736 down_read(&SM_I(sbi)->curseg_lock);
2738 mutex_lock(&curseg->curseg_mutex);
2739 down_write(&sit_i->sentry_lock);
2741 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2743 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2746 * __add_sum_entry should be resided under the curseg_mutex
2747 * because, this function updates a summary entry in the
2748 * current summary block.
2750 __add_sum_entry(sbi, type, sum);
2752 __refresh_next_blkoff(sbi, curseg);
2754 stat_inc_block_count(sbi, curseg);
2757 * SIT information should be updated before segment allocation,
2758 * since SSR needs latest valid block information.
2760 update_sit_entry(sbi, *new_blkaddr, 1);
2761 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2762 update_sit_entry(sbi, old_blkaddr, -1);
2764 if (!__has_curseg_space(sbi, type))
2765 sit_i->s_ops->allocate_segment(sbi, type, false);
2768 * segment dirty status should be updated after segment allocation,
2769 * so we just need to update status only one time after previous
2770 * segment being closed.
2772 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2773 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2775 up_write(&sit_i->sentry_lock);
2777 if (page && IS_NODESEG(type)) {
2778 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2780 f2fs_inode_chksum_set(sbi, page);
2784 struct f2fs_bio_info *io;
2786 INIT_LIST_HEAD(&fio->list);
2787 fio->in_list = true;
2788 io = sbi->write_io[fio->type] + fio->temp;
2789 spin_lock(&io->io_lock);
2790 list_add_tail(&fio->list, &io->io_list);
2791 spin_unlock(&io->io_lock);
2794 mutex_unlock(&curseg->curseg_mutex);
2796 up_read(&SM_I(sbi)->curseg_lock);
2799 static void update_device_state(struct f2fs_io_info *fio)
2801 struct f2fs_sb_info *sbi = fio->sbi;
2802 unsigned int devidx;
2807 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2809 /* update device state for fsync */
2810 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2812 /* update device state for checkpoint */
2813 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2814 spin_lock(&sbi->dev_lock);
2815 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2816 spin_unlock(&sbi->dev_lock);
2820 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2822 int type = __get_segment_type(fio);
2826 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2827 &fio->new_blkaddr, sum, type, fio, true);
2829 /* writeout dirty page into bdev */
2830 err = f2fs_submit_page_write(fio);
2831 if (err == -EAGAIN) {
2832 fio->old_blkaddr = fio->new_blkaddr;
2835 update_device_state(fio);
2839 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2840 enum iostat_type io_type)
2842 struct f2fs_io_info fio = {
2847 .op_flags = REQ_SYNC | REQ_NOIDLE | REQ_META | REQ_PRIO,
2848 .old_blkaddr = page->index,
2849 .new_blkaddr = page->index,
2851 .encrypted_page = NULL,
2855 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2856 fio.op_flags &= ~REQ_META;
2858 set_page_writeback(page);
2859 ClearPageError(page);
2860 f2fs_submit_page_write(&fio);
2862 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2865 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2867 struct f2fs_summary sum;
2869 set_summary(&sum, nid, 0, 0);
2870 do_write_page(&sum, fio);
2872 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2875 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2877 struct f2fs_sb_info *sbi = fio->sbi;
2878 struct f2fs_summary sum;
2879 struct node_info ni;
2881 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2882 get_node_info(sbi, dn->nid, &ni);
2883 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2884 do_write_page(&sum, fio);
2885 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2887 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2890 int rewrite_data_page(struct f2fs_io_info *fio)
2893 struct f2fs_sb_info *sbi = fio->sbi;
2895 fio->new_blkaddr = fio->old_blkaddr;
2896 /* i/o temperature is needed for passing down write hints */
2897 __get_segment_type(fio);
2899 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2900 GET_SEGNO(sbi, fio->new_blkaddr))->type));
2902 stat_inc_inplace_blocks(fio->sbi);
2904 err = f2fs_submit_page_bio(fio);
2906 update_device_state(fio);
2908 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2913 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2918 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2919 if (CURSEG_I(sbi, i)->segno == segno)
2925 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2926 block_t old_blkaddr, block_t new_blkaddr,
2927 bool recover_curseg, bool recover_newaddr)
2929 struct sit_info *sit_i = SIT_I(sbi);
2930 struct curseg_info *curseg;
2931 unsigned int segno, old_cursegno;
2932 struct seg_entry *se;
2934 unsigned short old_blkoff;
2936 segno = GET_SEGNO(sbi, new_blkaddr);
2937 se = get_seg_entry(sbi, segno);
2940 down_write(&SM_I(sbi)->curseg_lock);
2942 if (!recover_curseg) {
2943 /* for recovery flow */
2944 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2945 if (old_blkaddr == NULL_ADDR)
2946 type = CURSEG_COLD_DATA;
2948 type = CURSEG_WARM_DATA;
2951 if (IS_CURSEG(sbi, segno)) {
2952 /* se->type is volatile as SSR allocation */
2953 type = __f2fs_get_curseg(sbi, segno);
2954 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2956 type = CURSEG_WARM_DATA;
2960 f2fs_bug_on(sbi, !IS_DATASEG(type));
2961 curseg = CURSEG_I(sbi, type);
2963 mutex_lock(&curseg->curseg_mutex);
2964 down_write(&sit_i->sentry_lock);
2966 old_cursegno = curseg->segno;
2967 old_blkoff = curseg->next_blkoff;
2969 /* change the current segment */
2970 if (segno != curseg->segno) {
2971 curseg->next_segno = segno;
2972 change_curseg(sbi, type);
2975 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2976 __add_sum_entry(sbi, type, sum);
2978 if (!recover_curseg || recover_newaddr)
2979 update_sit_entry(sbi, new_blkaddr, 1);
2980 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2981 update_sit_entry(sbi, old_blkaddr, -1);
2983 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2984 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2986 locate_dirty_segment(sbi, old_cursegno);
2988 if (recover_curseg) {
2989 if (old_cursegno != curseg->segno) {
2990 curseg->next_segno = old_cursegno;
2991 change_curseg(sbi, type);
2993 curseg->next_blkoff = old_blkoff;
2996 up_write(&sit_i->sentry_lock);
2997 mutex_unlock(&curseg->curseg_mutex);
2998 up_write(&SM_I(sbi)->curseg_lock);
3001 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3002 block_t old_addr, block_t new_addr,
3003 unsigned char version, bool recover_curseg,
3004 bool recover_newaddr)
3006 struct f2fs_summary sum;
3008 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3010 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
3011 recover_curseg, recover_newaddr);
3013 f2fs_update_data_blkaddr(dn, new_addr);
3016 void f2fs_wait_on_page_writeback(struct page *page,
3017 enum page_type type, bool ordered)
3019 if (PageWriteback(page)) {
3020 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3022 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
3023 0, page->index, type);
3025 wait_on_page_writeback(page);
3027 wait_for_stable_page(page);
3031 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
3035 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
3038 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3040 f2fs_wait_on_page_writeback(cpage, DATA, true);
3041 f2fs_put_page(cpage, 1);
3045 static void read_compacted_summaries(struct f2fs_sb_info *sbi)
3047 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3048 struct curseg_info *seg_i;
3049 unsigned char *kaddr;
3054 start = start_sum_block(sbi);
3056 page = get_meta_page(sbi, start++);
3057 kaddr = (unsigned char *)page_address(page);
3059 /* Step 1: restore nat cache */
3060 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3061 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3063 /* Step 2: restore sit cache */
3064 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3065 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3066 offset = 2 * SUM_JOURNAL_SIZE;
3068 /* Step 3: restore summary entries */
3069 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3070 unsigned short blk_off;
3073 seg_i = CURSEG_I(sbi, i);
3074 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3075 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3076 seg_i->next_segno = segno;
3077 reset_curseg(sbi, i, 0);
3078 seg_i->alloc_type = ckpt->alloc_type[i];
3079 seg_i->next_blkoff = blk_off;
3081 if (seg_i->alloc_type == SSR)
3082 blk_off = sbi->blocks_per_seg;
3084 for (j = 0; j < blk_off; j++) {
3085 struct f2fs_summary *s;
3086 s = (struct f2fs_summary *)(kaddr + offset);
3087 seg_i->sum_blk->entries[j] = *s;
3088 offset += SUMMARY_SIZE;
3089 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3093 f2fs_put_page(page, 1);
3096 page = get_meta_page(sbi, start++);
3097 kaddr = (unsigned char *)page_address(page);
3101 f2fs_put_page(page, 1);
3104 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3106 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3107 struct f2fs_summary_block *sum;
3108 struct curseg_info *curseg;
3110 unsigned short blk_off;
3111 unsigned int segno = 0;
3112 block_t blk_addr = 0;
3114 /* get segment number and block addr */
3115 if (IS_DATASEG(type)) {
3116 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3117 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3119 if (__exist_node_summaries(sbi))
3120 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3122 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3124 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3126 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3128 if (__exist_node_summaries(sbi))
3129 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3130 type - CURSEG_HOT_NODE);
3132 blk_addr = GET_SUM_BLOCK(sbi, segno);
3135 new = get_meta_page(sbi, blk_addr);
3136 sum = (struct f2fs_summary_block *)page_address(new);
3138 if (IS_NODESEG(type)) {
3139 if (__exist_node_summaries(sbi)) {
3140 struct f2fs_summary *ns = &sum->entries[0];
3142 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3144 ns->ofs_in_node = 0;
3147 restore_node_summary(sbi, segno, sum);
3151 /* set uncompleted segment to curseg */
3152 curseg = CURSEG_I(sbi, type);
3153 mutex_lock(&curseg->curseg_mutex);
3155 /* update journal info */
3156 down_write(&curseg->journal_rwsem);
3157 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3158 up_write(&curseg->journal_rwsem);
3160 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3161 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3162 curseg->next_segno = segno;
3163 reset_curseg(sbi, type, 0);
3164 curseg->alloc_type = ckpt->alloc_type[type];
3165 curseg->next_blkoff = blk_off;
3166 mutex_unlock(&curseg->curseg_mutex);
3167 f2fs_put_page(new, 1);
3171 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3173 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3174 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3175 int type = CURSEG_HOT_DATA;
3178 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3179 int npages = npages_for_summary_flush(sbi, true);
3182 ra_meta_pages(sbi, start_sum_block(sbi), npages,
3185 /* restore for compacted data summary */
3186 read_compacted_summaries(sbi);
3187 type = CURSEG_HOT_NODE;
3190 if (__exist_node_summaries(sbi))
3191 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3192 NR_CURSEG_TYPE - type, META_CP, true);
3194 for (; type <= CURSEG_COLD_NODE; type++) {
3195 err = read_normal_summaries(sbi, type);
3200 /* sanity check for summary blocks */
3201 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3202 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3208 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3211 unsigned char *kaddr;
3212 struct f2fs_summary *summary;
3213 struct curseg_info *seg_i;
3214 int written_size = 0;
3217 page = grab_meta_page(sbi, blkaddr++);
3218 kaddr = (unsigned char *)page_address(page);
3220 /* Step 1: write nat cache */
3221 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3222 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3223 written_size += SUM_JOURNAL_SIZE;
3225 /* Step 2: write sit cache */
3226 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3227 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3228 written_size += SUM_JOURNAL_SIZE;
3230 /* Step 3: write summary entries */
3231 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3232 unsigned short blkoff;
3233 seg_i = CURSEG_I(sbi, i);
3234 if (sbi->ckpt->alloc_type[i] == SSR)
3235 blkoff = sbi->blocks_per_seg;
3237 blkoff = curseg_blkoff(sbi, i);
3239 for (j = 0; j < blkoff; j++) {
3241 page = grab_meta_page(sbi, blkaddr++);
3242 kaddr = (unsigned char *)page_address(page);
3245 summary = (struct f2fs_summary *)(kaddr + written_size);
3246 *summary = seg_i->sum_blk->entries[j];
3247 written_size += SUMMARY_SIZE;
3249 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3253 set_page_dirty(page);
3254 f2fs_put_page(page, 1);
3259 set_page_dirty(page);
3260 f2fs_put_page(page, 1);
3264 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3265 block_t blkaddr, int type)
3268 if (IS_DATASEG(type))
3269 end = type + NR_CURSEG_DATA_TYPE;
3271 end = type + NR_CURSEG_NODE_TYPE;
3273 for (i = type; i < end; i++)
3274 write_current_sum_page(sbi, i, blkaddr + (i - type));
3277 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3279 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3280 write_compacted_summaries(sbi, start_blk);
3282 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3285 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3287 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3290 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3291 unsigned int val, int alloc)
3295 if (type == NAT_JOURNAL) {
3296 for (i = 0; i < nats_in_cursum(journal); i++) {
3297 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3300 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3301 return update_nats_in_cursum(journal, 1);
3302 } else if (type == SIT_JOURNAL) {
3303 for (i = 0; i < sits_in_cursum(journal); i++)
3304 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3306 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3307 return update_sits_in_cursum(journal, 1);
3312 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3315 return get_meta_page(sbi, current_sit_addr(sbi, segno));
3318 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3321 struct sit_info *sit_i = SIT_I(sbi);
3323 pgoff_t src_off, dst_off;
3325 src_off = current_sit_addr(sbi, start);
3326 dst_off = next_sit_addr(sbi, src_off);
3328 page = grab_meta_page(sbi, dst_off);
3329 seg_info_to_sit_page(sbi, page, start);
3331 set_page_dirty(page);
3332 set_to_next_sit(sit_i, start);
3337 static struct sit_entry_set *grab_sit_entry_set(void)
3339 struct sit_entry_set *ses =
3340 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3343 INIT_LIST_HEAD(&ses->set_list);
3347 static void release_sit_entry_set(struct sit_entry_set *ses)
3349 list_del(&ses->set_list);
3350 kmem_cache_free(sit_entry_set_slab, ses);
3353 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3354 struct list_head *head)
3356 struct sit_entry_set *next = ses;
3358 if (list_is_last(&ses->set_list, head))
3361 list_for_each_entry_continue(next, head, set_list)
3362 if (ses->entry_cnt <= next->entry_cnt)
3365 list_move_tail(&ses->set_list, &next->set_list);
3368 static void add_sit_entry(unsigned int segno, struct list_head *head)
3370 struct sit_entry_set *ses;
3371 unsigned int start_segno = START_SEGNO(segno);
3373 list_for_each_entry(ses, head, set_list) {
3374 if (ses->start_segno == start_segno) {
3376 adjust_sit_entry_set(ses, head);
3381 ses = grab_sit_entry_set();
3383 ses->start_segno = start_segno;
3385 list_add(&ses->set_list, head);
3388 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3390 struct f2fs_sm_info *sm_info = SM_I(sbi);
3391 struct list_head *set_list = &sm_info->sit_entry_set;
3392 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3395 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3396 add_sit_entry(segno, set_list);
3399 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3401 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3402 struct f2fs_journal *journal = curseg->journal;
3405 down_write(&curseg->journal_rwsem);
3406 for (i = 0; i < sits_in_cursum(journal); i++) {
3410 segno = le32_to_cpu(segno_in_journal(journal, i));
3411 dirtied = __mark_sit_entry_dirty(sbi, segno);
3414 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3416 update_sits_in_cursum(journal, -i);
3417 up_write(&curseg->journal_rwsem);
3421 * CP calls this function, which flushes SIT entries including sit_journal,
3422 * and moves prefree segs to free segs.
3424 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3426 struct sit_info *sit_i = SIT_I(sbi);
3427 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3428 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3429 struct f2fs_journal *journal = curseg->journal;
3430 struct sit_entry_set *ses, *tmp;
3431 struct list_head *head = &SM_I(sbi)->sit_entry_set;
3432 bool to_journal = true;
3433 struct seg_entry *se;
3435 down_write(&sit_i->sentry_lock);
3437 if (!sit_i->dirty_sentries)
3441 * add and account sit entries of dirty bitmap in sit entry
3444 add_sits_in_set(sbi);
3447 * if there are no enough space in journal to store dirty sit
3448 * entries, remove all entries from journal and add and account
3449 * them in sit entry set.
3451 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3452 remove_sits_in_journal(sbi);
3455 * there are two steps to flush sit entries:
3456 * #1, flush sit entries to journal in current cold data summary block.
3457 * #2, flush sit entries to sit page.
3459 list_for_each_entry_safe(ses, tmp, head, set_list) {
3460 struct page *page = NULL;
3461 struct f2fs_sit_block *raw_sit = NULL;
3462 unsigned int start_segno = ses->start_segno;
3463 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3464 (unsigned long)MAIN_SEGS(sbi));
3465 unsigned int segno = start_segno;
3468 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3472 down_write(&curseg->journal_rwsem);
3474 page = get_next_sit_page(sbi, start_segno);
3475 raw_sit = page_address(page);
3478 /* flush dirty sit entries in region of current sit set */
3479 for_each_set_bit_from(segno, bitmap, end) {
3480 int offset, sit_offset;
3482 se = get_seg_entry(sbi, segno);
3484 /* add discard candidates */
3485 if (!(cpc->reason & CP_DISCARD)) {
3486 cpc->trim_start = segno;
3487 add_discard_addrs(sbi, cpc, false);
3491 offset = lookup_journal_in_cursum(journal,
3492 SIT_JOURNAL, segno, 1);
3493 f2fs_bug_on(sbi, offset < 0);
3494 segno_in_journal(journal, offset) =
3496 seg_info_to_raw_sit(se,
3497 &sit_in_journal(journal, offset));
3499 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3500 seg_info_to_raw_sit(se,
3501 &raw_sit->entries[sit_offset]);
3504 __clear_bit(segno, bitmap);
3505 sit_i->dirty_sentries--;
3510 up_write(&curseg->journal_rwsem);
3512 f2fs_put_page(page, 1);
3514 f2fs_bug_on(sbi, ses->entry_cnt);
3515 release_sit_entry_set(ses);
3518 f2fs_bug_on(sbi, !list_empty(head));
3519 f2fs_bug_on(sbi, sit_i->dirty_sentries);
3521 if (cpc->reason & CP_DISCARD) {
3522 __u64 trim_start = cpc->trim_start;
3524 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3525 add_discard_addrs(sbi, cpc, false);
3527 cpc->trim_start = trim_start;
3529 up_write(&sit_i->sentry_lock);
3531 set_prefree_as_free_segments(sbi);
3534 static int build_sit_info(struct f2fs_sb_info *sbi)
3536 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3537 struct sit_info *sit_i;
3538 unsigned int sit_segs, start;
3540 unsigned int bitmap_size;
3542 /* allocate memory for SIT information */
3543 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3547 SM_I(sbi)->sit_info = sit_i;
3549 sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
3550 sizeof(struct seg_entry), GFP_KERNEL);
3551 if (!sit_i->sentries)
3554 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3555 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3557 if (!sit_i->dirty_sentries_bitmap)
3560 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3561 sit_i->sentries[start].cur_valid_map
3562 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3563 sit_i->sentries[start].ckpt_valid_map
3564 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3565 if (!sit_i->sentries[start].cur_valid_map ||
3566 !sit_i->sentries[start].ckpt_valid_map)
3569 #ifdef CONFIG_F2FS_CHECK_FS
3570 sit_i->sentries[start].cur_valid_map_mir
3571 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3572 if (!sit_i->sentries[start].cur_valid_map_mir)
3576 if (f2fs_discard_en(sbi)) {
3577 sit_i->sentries[start].discard_map
3578 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3580 if (!sit_i->sentries[start].discard_map)
3585 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3586 if (!sit_i->tmp_map)
3589 if (sbi->segs_per_sec > 1) {
3590 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
3591 sizeof(struct sec_entry), GFP_KERNEL);
3592 if (!sit_i->sec_entries)
3596 /* get information related with SIT */
3597 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3599 /* setup SIT bitmap from ckeckpoint pack */
3600 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3601 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3603 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3604 if (!sit_i->sit_bitmap)
3607 #ifdef CONFIG_F2FS_CHECK_FS
3608 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3609 if (!sit_i->sit_bitmap_mir)
3613 /* init SIT information */
3614 sit_i->s_ops = &default_salloc_ops;
3616 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3617 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3618 sit_i->written_valid_blocks = 0;
3619 sit_i->bitmap_size = bitmap_size;
3620 sit_i->dirty_sentries = 0;
3621 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3622 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3623 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
3624 init_rwsem(&sit_i->sentry_lock);
3628 static int build_free_segmap(struct f2fs_sb_info *sbi)
3630 struct free_segmap_info *free_i;
3631 unsigned int bitmap_size, sec_bitmap_size;
3633 /* allocate memory for free segmap information */
3634 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3638 SM_I(sbi)->free_info = free_i;
3640 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3641 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3642 if (!free_i->free_segmap)
3645 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3646 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3647 if (!free_i->free_secmap)
3650 /* set all segments as dirty temporarily */
3651 memset(free_i->free_segmap, 0xff, bitmap_size);
3652 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3654 /* init free segmap information */
3655 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3656 free_i->free_segments = 0;
3657 free_i->free_sections = 0;
3658 spin_lock_init(&free_i->segmap_lock);
3662 static int build_curseg(struct f2fs_sb_info *sbi)
3664 struct curseg_info *array;
3667 array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
3671 SM_I(sbi)->curseg_array = array;
3673 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3674 mutex_init(&array[i].curseg_mutex);
3675 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3676 if (!array[i].sum_blk)
3678 init_rwsem(&array[i].journal_rwsem);
3679 array[i].journal = f2fs_kzalloc(sbi,
3680 sizeof(struct f2fs_journal), GFP_KERNEL);
3681 if (!array[i].journal)
3683 array[i].segno = NULL_SEGNO;
3684 array[i].next_blkoff = 0;
3686 return restore_curseg_summaries(sbi);
3689 static int build_sit_entries(struct f2fs_sb_info *sbi)
3691 struct sit_info *sit_i = SIT_I(sbi);
3692 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3693 struct f2fs_journal *journal = curseg->journal;
3694 struct seg_entry *se;
3695 struct f2fs_sit_entry sit;
3696 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3697 unsigned int i, start, end;
3698 unsigned int readed, start_blk = 0;
3702 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3705 start = start_blk * sit_i->sents_per_block;
3706 end = (start_blk + readed) * sit_i->sents_per_block;
3708 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3709 struct f2fs_sit_block *sit_blk;
3712 se = &sit_i->sentries[start];
3713 page = get_current_sit_page(sbi, start);
3714 sit_blk = (struct f2fs_sit_block *)page_address(page);
3715 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3716 f2fs_put_page(page, 1);
3718 err = check_block_count(sbi, start, &sit);
3721 seg_info_from_raw_sit(se, &sit);
3723 /* build discard map only one time */
3724 if (f2fs_discard_en(sbi)) {
3725 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3726 memset(se->discard_map, 0xff,
3727 SIT_VBLOCK_MAP_SIZE);
3729 memcpy(se->discard_map,
3731 SIT_VBLOCK_MAP_SIZE);
3732 sbi->discard_blks +=
3733 sbi->blocks_per_seg -
3738 if (sbi->segs_per_sec > 1)
3739 get_sec_entry(sbi, start)->valid_blocks +=
3742 start_blk += readed;
3743 } while (start_blk < sit_blk_cnt);
3745 down_read(&curseg->journal_rwsem);
3746 for (i = 0; i < sits_in_cursum(journal); i++) {
3747 unsigned int old_valid_blocks;
3749 start = le32_to_cpu(segno_in_journal(journal, i));
3750 se = &sit_i->sentries[start];
3751 sit = sit_in_journal(journal, i);
3753 old_valid_blocks = se->valid_blocks;
3755 err = check_block_count(sbi, start, &sit);
3758 seg_info_from_raw_sit(se, &sit);
3760 if (f2fs_discard_en(sbi)) {
3761 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3762 memset(se->discard_map, 0xff,
3763 SIT_VBLOCK_MAP_SIZE);
3765 memcpy(se->discard_map, se->cur_valid_map,
3766 SIT_VBLOCK_MAP_SIZE);
3767 sbi->discard_blks += old_valid_blocks -
3772 if (sbi->segs_per_sec > 1)
3773 get_sec_entry(sbi, start)->valid_blocks +=
3774 se->valid_blocks - old_valid_blocks;
3776 up_read(&curseg->journal_rwsem);
3780 static void init_free_segmap(struct f2fs_sb_info *sbi)
3785 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3786 struct seg_entry *sentry = get_seg_entry(sbi, start);
3787 if (!sentry->valid_blocks)
3788 __set_free(sbi, start);
3790 SIT_I(sbi)->written_valid_blocks +=
3791 sentry->valid_blocks;
3794 /* set use the current segments */
3795 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3796 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3797 __set_test_and_inuse(sbi, curseg_t->segno);
3801 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3803 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3804 struct free_segmap_info *free_i = FREE_I(sbi);
3805 unsigned int segno = 0, offset = 0;
3806 unsigned short valid_blocks;
3809 /* find dirty segment based on free segmap */
3810 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3811 if (segno >= MAIN_SEGS(sbi))
3814 valid_blocks = get_valid_blocks(sbi, segno, false);
3815 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3817 if (valid_blocks > sbi->blocks_per_seg) {
3818 f2fs_bug_on(sbi, 1);
3821 mutex_lock(&dirty_i->seglist_lock);
3822 __locate_dirty_segment(sbi, segno, DIRTY);
3823 mutex_unlock(&dirty_i->seglist_lock);
3827 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3829 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3830 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3832 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3833 if (!dirty_i->victim_secmap)
3838 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3840 struct dirty_seglist_info *dirty_i;
3841 unsigned int bitmap_size, i;
3843 /* allocate memory for dirty segments list information */
3844 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3849 SM_I(sbi)->dirty_info = dirty_i;
3850 mutex_init(&dirty_i->seglist_lock);
3852 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3854 for (i = 0; i < NR_DIRTY_TYPE; i++) {
3855 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3857 if (!dirty_i->dirty_segmap[i])
3861 init_dirty_segmap(sbi);
3862 return init_victim_secmap(sbi);
3866 * Update min, max modified time for cost-benefit GC algorithm
3868 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3870 struct sit_info *sit_i = SIT_I(sbi);
3873 down_write(&sit_i->sentry_lock);
3875 sit_i->min_mtime = LLONG_MAX;
3877 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3879 unsigned long long mtime = 0;
3881 for (i = 0; i < sbi->segs_per_sec; i++)
3882 mtime += get_seg_entry(sbi, segno + i)->mtime;
3884 mtime = div_u64(mtime, sbi->segs_per_sec);
3886 if (sit_i->min_mtime > mtime)
3887 sit_i->min_mtime = mtime;
3889 sit_i->max_mtime = get_mtime(sbi);
3890 up_write(&sit_i->sentry_lock);
3893 int build_segment_manager(struct f2fs_sb_info *sbi)
3895 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3896 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3897 struct f2fs_sm_info *sm_info;
3900 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3905 sbi->sm_info = sm_info;
3906 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3907 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3908 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3909 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3910 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3911 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3912 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3913 sm_info->rec_prefree_segments = sm_info->main_segments *
3914 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3915 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3916 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3918 if (!test_opt(sbi, LFS))
3919 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3920 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3921 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3922 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3923 sm_info->min_ssr_sections = reserved_sections(sbi);
3925 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3927 init_rwsem(&sm_info->curseg_lock);
3929 if (!f2fs_readonly(sbi->sb)) {
3930 err = create_flush_cmd_control(sbi);
3935 err = create_discard_cmd_control(sbi);
3939 err = build_sit_info(sbi);
3942 err = build_free_segmap(sbi);
3945 err = build_curseg(sbi);
3949 /* reinit free segmap based on SIT */
3950 err = build_sit_entries(sbi);
3954 init_free_segmap(sbi);
3955 err = build_dirty_segmap(sbi);
3959 init_min_max_mtime(sbi);
3963 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3964 enum dirty_type dirty_type)
3966 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3968 mutex_lock(&dirty_i->seglist_lock);
3969 kvfree(dirty_i->dirty_segmap[dirty_type]);
3970 dirty_i->nr_dirty[dirty_type] = 0;
3971 mutex_unlock(&dirty_i->seglist_lock);
3974 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3976 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3977 kvfree(dirty_i->victim_secmap);
3980 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3982 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3988 /* discard pre-free/dirty segments list */
3989 for (i = 0; i < NR_DIRTY_TYPE; i++)
3990 discard_dirty_segmap(sbi, i);
3992 destroy_victim_secmap(sbi);
3993 SM_I(sbi)->dirty_info = NULL;
3997 static void destroy_curseg(struct f2fs_sb_info *sbi)
3999 struct curseg_info *array = SM_I(sbi)->curseg_array;
4004 SM_I(sbi)->curseg_array = NULL;
4005 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4006 kfree(array[i].sum_blk);
4007 kfree(array[i].journal);
4012 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4014 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4017 SM_I(sbi)->free_info = NULL;
4018 kvfree(free_i->free_segmap);
4019 kvfree(free_i->free_secmap);
4023 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4025 struct sit_info *sit_i = SIT_I(sbi);
4031 if (sit_i->sentries) {
4032 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4033 kfree(sit_i->sentries[start].cur_valid_map);
4034 #ifdef CONFIG_F2FS_CHECK_FS
4035 kfree(sit_i->sentries[start].cur_valid_map_mir);
4037 kfree(sit_i->sentries[start].ckpt_valid_map);
4038 kfree(sit_i->sentries[start].discard_map);
4041 kfree(sit_i->tmp_map);
4043 kvfree(sit_i->sentries);
4044 kvfree(sit_i->sec_entries);
4045 kvfree(sit_i->dirty_sentries_bitmap);
4047 SM_I(sbi)->sit_info = NULL;
4048 kfree(sit_i->sit_bitmap);
4049 #ifdef CONFIG_F2FS_CHECK_FS
4050 kfree(sit_i->sit_bitmap_mir);
4055 void destroy_segment_manager(struct f2fs_sb_info *sbi)
4057 struct f2fs_sm_info *sm_info = SM_I(sbi);
4061 destroy_flush_cmd_control(sbi, true);
4062 destroy_discard_cmd_control(sbi);
4063 destroy_dirty_segmap(sbi);
4064 destroy_curseg(sbi);
4065 destroy_free_segmap(sbi);
4066 destroy_sit_info(sbi);
4067 sbi->sm_info = NULL;
4071 int __init create_segment_manager_caches(void)
4073 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4074 sizeof(struct discard_entry));
4075 if (!discard_entry_slab)
4078 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4079 sizeof(struct discard_cmd));
4080 if (!discard_cmd_slab)
4081 goto destroy_discard_entry;
4083 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4084 sizeof(struct sit_entry_set));
4085 if (!sit_entry_set_slab)
4086 goto destroy_discard_cmd;
4088 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4089 sizeof(struct inmem_pages));
4090 if (!inmem_entry_slab)
4091 goto destroy_sit_entry_set;
4094 destroy_sit_entry_set:
4095 kmem_cache_destroy(sit_entry_set_slab);
4096 destroy_discard_cmd:
4097 kmem_cache_destroy(discard_cmd_slab);
4098 destroy_discard_entry:
4099 kmem_cache_destroy(discard_entry_slab);
4104 void destroy_segment_manager_caches(void)
4106 kmem_cache_destroy(sit_entry_set_slab);
4107 kmem_cache_destroy(discard_cmd_slab);
4108 kmem_cache_destroy(discard_entry_slab);
4109 kmem_cache_destroy(inmem_entry_slab);