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/signal.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, list_empty(head) || 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)
348 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
349 struct f2fs_inode_info *fi = F2FS_I(inode);
350 struct inmem_pages *cur, *tmp;
351 struct f2fs_io_info fio = {
356 .op_flags = REQ_SYNC | REQ_PRIO,
357 .io_type = FS_DATA_IO,
359 struct list_head revoke_list;
360 pgoff_t last_idx = ULONG_MAX;
363 INIT_LIST_HEAD(&revoke_list);
365 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
366 struct page *page = cur->page;
369 if (page->mapping == inode->i_mapping) {
370 trace_f2fs_commit_inmem_page(page, INMEM);
372 set_page_dirty(page);
373 f2fs_wait_on_page_writeback(page, DATA, true);
374 if (clear_page_dirty_for_io(page)) {
375 inode_dec_dirty_pages(inode);
376 remove_dirty_inode(inode);
380 fio.old_blkaddr = NULL_ADDR;
381 fio.encrypted_page = NULL;
382 fio.need_lock = LOCK_DONE;
383 err = do_write_data_page(&fio);
385 if (err == -ENOMEM) {
386 congestion_wait(BLK_RW_ASYNC, HZ/50);
393 /* record old blkaddr for revoking */
394 cur->old_addr = fio.old_blkaddr;
395 last_idx = page->index;
398 list_move_tail(&cur->list, &revoke_list);
401 if (last_idx != ULONG_MAX)
402 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
406 * try to revoke all committed pages, but still we could fail
407 * due to no memory or other reason, if that happened, EAGAIN
408 * will be returned, which means in such case, transaction is
409 * already not integrity, caller should use journal to do the
410 * recovery or rewrite & commit last transaction. For other
411 * error number, revoking was done by filesystem itself.
413 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
415 /* drop all uncommitted pages */
416 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
418 __revoke_inmem_pages(inode, &revoke_list, false, false);
424 int commit_inmem_pages(struct inode *inode)
426 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
427 struct f2fs_inode_info *fi = F2FS_I(inode);
430 f2fs_balance_fs(sbi, true);
433 set_inode_flag(inode, FI_ATOMIC_COMMIT);
435 mutex_lock(&fi->inmem_lock);
436 err = __commit_inmem_pages(inode);
438 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
439 if (!list_empty(&fi->inmem_ilist))
440 list_del_init(&fi->inmem_ilist);
441 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
442 mutex_unlock(&fi->inmem_lock);
444 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
451 * This function balances dirty node and dentry pages.
452 * In addition, it controls garbage collection.
454 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
456 #ifdef CONFIG_F2FS_FAULT_INJECTION
457 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
458 f2fs_show_injection_info(FAULT_CHECKPOINT);
459 f2fs_stop_checkpoint(sbi, false);
463 /* balance_fs_bg is able to be pending */
464 if (need && excess_cached_nats(sbi))
465 f2fs_balance_fs_bg(sbi);
468 * We should do GC or end up with checkpoint, if there are so many dirty
469 * dir/node pages without enough free segments.
471 if (has_not_enough_free_secs(sbi, 0, 0)) {
472 mutex_lock(&sbi->gc_mutex);
473 f2fs_gc(sbi, false, false, NULL_SEGNO);
477 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
479 /* try to shrink extent cache when there is no enough memory */
480 if (!available_free_memory(sbi, EXTENT_CACHE))
481 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
483 /* check the # of cached NAT entries */
484 if (!available_free_memory(sbi, NAT_ENTRIES))
485 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
487 if (!available_free_memory(sbi, FREE_NIDS))
488 try_to_free_nids(sbi, MAX_FREE_NIDS);
490 build_free_nids(sbi, false, false);
492 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
495 /* checkpoint is the only way to shrink partial cached entries */
496 if (!available_free_memory(sbi, NAT_ENTRIES) ||
497 !available_free_memory(sbi, INO_ENTRIES) ||
498 excess_prefree_segs(sbi) ||
499 excess_dirty_nats(sbi) ||
500 f2fs_time_over(sbi, CP_TIME)) {
501 if (test_opt(sbi, DATA_FLUSH)) {
502 struct blk_plug plug;
504 blk_start_plug(&plug);
505 sync_dirty_inodes(sbi, FILE_INODE);
506 blk_finish_plug(&plug);
508 f2fs_sync_fs(sbi->sb, true);
509 stat_inc_bg_cp_count(sbi->stat_info);
513 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
514 struct block_device *bdev)
516 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
519 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
520 bio_set_dev(bio, bdev);
521 ret = submit_bio_wait(bio);
524 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
525 test_opt(sbi, FLUSH_MERGE), ret);
529 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
535 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
537 for (i = 0; i < sbi->s_ndevs; i++) {
538 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
540 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
547 static int issue_flush_thread(void *data)
549 struct f2fs_sb_info *sbi = data;
550 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
551 wait_queue_head_t *q = &fcc->flush_wait_queue;
553 if (kthread_should_stop())
556 sb_start_intwrite(sbi->sb);
558 if (!llist_empty(&fcc->issue_list)) {
559 struct flush_cmd *cmd, *next;
562 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
563 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
565 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
567 ret = submit_flush_wait(sbi, cmd->ino);
568 atomic_inc(&fcc->issued_flush);
570 llist_for_each_entry_safe(cmd, next,
571 fcc->dispatch_list, llnode) {
573 complete(&cmd->wait);
575 fcc->dispatch_list = NULL;
578 sb_end_intwrite(sbi->sb);
580 wait_event_interruptible(*q,
581 kthread_should_stop() || !llist_empty(&fcc->issue_list));
585 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
587 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
588 struct flush_cmd cmd;
591 if (test_opt(sbi, NOBARRIER))
594 if (!test_opt(sbi, FLUSH_MERGE)) {
595 ret = submit_flush_wait(sbi, ino);
596 atomic_inc(&fcc->issued_flush);
600 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
601 ret = submit_flush_wait(sbi, ino);
602 atomic_dec(&fcc->issing_flush);
604 atomic_inc(&fcc->issued_flush);
609 init_completion(&cmd.wait);
611 llist_add(&cmd.llnode, &fcc->issue_list);
613 /* update issue_list before we wake up issue_flush thread */
616 if (waitqueue_active(&fcc->flush_wait_queue))
617 wake_up(&fcc->flush_wait_queue);
619 if (fcc->f2fs_issue_flush) {
620 wait_for_completion(&cmd.wait);
621 atomic_dec(&fcc->issing_flush);
623 struct llist_node *list;
625 list = llist_del_all(&fcc->issue_list);
627 wait_for_completion(&cmd.wait);
628 atomic_dec(&fcc->issing_flush);
630 struct flush_cmd *tmp, *next;
632 ret = submit_flush_wait(sbi, ino);
634 llist_for_each_entry_safe(tmp, next, list, llnode) {
637 atomic_dec(&fcc->issing_flush);
641 complete(&tmp->wait);
649 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
651 dev_t dev = sbi->sb->s_bdev->bd_dev;
652 struct flush_cmd_control *fcc;
655 if (SM_I(sbi)->fcc_info) {
656 fcc = SM_I(sbi)->fcc_info;
657 if (fcc->f2fs_issue_flush)
662 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
665 atomic_set(&fcc->issued_flush, 0);
666 atomic_set(&fcc->issing_flush, 0);
667 init_waitqueue_head(&fcc->flush_wait_queue);
668 init_llist_head(&fcc->issue_list);
669 SM_I(sbi)->fcc_info = fcc;
670 if (!test_opt(sbi, FLUSH_MERGE))
674 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
675 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
676 if (IS_ERR(fcc->f2fs_issue_flush)) {
677 err = PTR_ERR(fcc->f2fs_issue_flush);
679 SM_I(sbi)->fcc_info = NULL;
686 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
688 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
690 if (fcc && fcc->f2fs_issue_flush) {
691 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
693 fcc->f2fs_issue_flush = NULL;
694 kthread_stop(flush_thread);
698 SM_I(sbi)->fcc_info = NULL;
702 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
709 for (i = 1; i < sbi->s_ndevs; i++) {
710 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
712 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
716 spin_lock(&sbi->dev_lock);
717 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
718 spin_unlock(&sbi->dev_lock);
724 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
725 enum dirty_type dirty_type)
727 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
729 /* need not be added */
730 if (IS_CURSEG(sbi, segno))
733 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
734 dirty_i->nr_dirty[dirty_type]++;
736 if (dirty_type == DIRTY) {
737 struct seg_entry *sentry = get_seg_entry(sbi, segno);
738 enum dirty_type t = sentry->type;
740 if (unlikely(t >= DIRTY)) {
744 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
745 dirty_i->nr_dirty[t]++;
749 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
750 enum dirty_type dirty_type)
752 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
754 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
755 dirty_i->nr_dirty[dirty_type]--;
757 if (dirty_type == DIRTY) {
758 struct seg_entry *sentry = get_seg_entry(sbi, segno);
759 enum dirty_type t = sentry->type;
761 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
762 dirty_i->nr_dirty[t]--;
764 if (get_valid_blocks(sbi, segno, true) == 0)
765 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
766 dirty_i->victim_secmap);
771 * Should not occur error such as -ENOMEM.
772 * Adding dirty entry into seglist is not critical operation.
773 * If a given segment is one of current working segments, it won't be added.
775 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
777 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
778 unsigned short valid_blocks;
780 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
783 mutex_lock(&dirty_i->seglist_lock);
785 valid_blocks = get_valid_blocks(sbi, segno, false);
787 if (valid_blocks == 0) {
788 __locate_dirty_segment(sbi, segno, PRE);
789 __remove_dirty_segment(sbi, segno, DIRTY);
790 } else if (valid_blocks < sbi->blocks_per_seg) {
791 __locate_dirty_segment(sbi, segno, DIRTY);
793 /* Recovery routine with SSR needs this */
794 __remove_dirty_segment(sbi, segno, DIRTY);
797 mutex_unlock(&dirty_i->seglist_lock);
800 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
801 struct block_device *bdev, block_t lstart,
802 block_t start, block_t len)
804 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
805 struct list_head *pend_list;
806 struct discard_cmd *dc;
808 f2fs_bug_on(sbi, !len);
810 pend_list = &dcc->pend_list[plist_idx(len)];
812 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
813 INIT_LIST_HEAD(&dc->list);
821 init_completion(&dc->wait);
822 list_add_tail(&dc->list, pend_list);
823 atomic_inc(&dcc->discard_cmd_cnt);
824 dcc->undiscard_blks += len;
829 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
830 struct block_device *bdev, block_t lstart,
831 block_t start, block_t len,
832 struct rb_node *parent, struct rb_node **p)
834 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
835 struct discard_cmd *dc;
837 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
839 rb_link_node(&dc->rb_node, parent, p);
840 rb_insert_color(&dc->rb_node, &dcc->root);
845 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
846 struct discard_cmd *dc)
848 if (dc->state == D_DONE)
849 atomic_dec(&dcc->issing_discard);
852 rb_erase(&dc->rb_node, &dcc->root);
853 dcc->undiscard_blks -= dc->len;
855 kmem_cache_free(discard_cmd_slab, dc);
857 atomic_dec(&dcc->discard_cmd_cnt);
860 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
861 struct discard_cmd *dc)
863 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
865 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
867 f2fs_bug_on(sbi, dc->ref);
869 if (dc->error == -EOPNOTSUPP)
873 f2fs_msg(sbi->sb, KERN_INFO,
874 "Issue discard(%u, %u, %u) failed, ret: %d",
875 dc->lstart, dc->start, dc->len, dc->error);
876 __detach_discard_cmd(dcc, dc);
879 static void f2fs_submit_discard_endio(struct bio *bio)
881 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
883 dc->error = blk_status_to_errno(bio->bi_status);
885 complete_all(&dc->wait);
889 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
890 block_t start, block_t end)
892 #ifdef CONFIG_F2FS_CHECK_FS
893 struct seg_entry *sentry;
896 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
900 segno = GET_SEGNO(sbi, blk);
901 sentry = get_seg_entry(sbi, segno);
902 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
904 if (end < START_BLOCK(sbi, segno + 1))
905 size = GET_BLKOFF_FROM_SEG0(sbi, end);
908 map = (unsigned long *)(sentry->cur_valid_map);
909 offset = __find_rev_next_bit(map, size, offset);
910 f2fs_bug_on(sbi, offset != size);
911 blk = START_BLOCK(sbi, segno + 1);
916 static void __init_discard_policy(struct f2fs_sb_info *sbi,
917 struct discard_policy *dpolicy,
918 int discard_type, unsigned int granularity)
921 dpolicy->type = discard_type;
922 dpolicy->sync = true;
923 dpolicy->granularity = granularity;
925 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
926 dpolicy->io_aware_gran = MAX_PLIST_NUM;
928 if (discard_type == DPOLICY_BG) {
929 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
930 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
931 dpolicy->io_aware = true;
932 dpolicy->sync = false;
933 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
934 dpolicy->granularity = 1;
935 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
937 } else if (discard_type == DPOLICY_FORCE) {
938 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
939 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
940 dpolicy->io_aware = false;
941 } else if (discard_type == DPOLICY_FSTRIM) {
942 dpolicy->io_aware = false;
943 } else if (discard_type == DPOLICY_UMOUNT) {
944 dpolicy->max_requests = UINT_MAX;
945 dpolicy->io_aware = false;
950 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
951 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
952 struct discard_policy *dpolicy,
953 struct discard_cmd *dc)
955 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
956 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
957 &(dcc->fstrim_list) : &(dcc->wait_list);
958 struct bio *bio = NULL;
959 int flag = dpolicy->sync ? REQ_SYNC : 0;
961 if (dc->state != D_PREP)
964 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
967 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
969 dc->error = __blkdev_issue_discard(dc->bdev,
970 SECTOR_FROM_BLOCK(dc->start),
971 SECTOR_FROM_BLOCK(dc->len),
974 /* should keep before submission to avoid D_DONE right away */
975 dc->state = D_SUBMIT;
976 atomic_inc(&dcc->issued_discard);
977 atomic_inc(&dcc->issing_discard);
979 bio->bi_private = dc;
980 bio->bi_end_io = f2fs_submit_discard_endio;
983 list_move_tail(&dc->list, wait_list);
984 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
986 f2fs_update_iostat(sbi, FS_DISCARD, 1);
989 __remove_discard_cmd(sbi, dc);
993 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
994 struct block_device *bdev, block_t lstart,
995 block_t start, block_t len,
996 struct rb_node **insert_p,
997 struct rb_node *insert_parent)
999 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1001 struct rb_node *parent = NULL;
1002 struct discard_cmd *dc = NULL;
1004 if (insert_p && insert_parent) {
1005 parent = insert_parent;
1010 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1012 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1019 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1020 struct discard_cmd *dc)
1022 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1025 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1026 struct discard_cmd *dc, block_t blkaddr)
1028 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1029 struct discard_info di = dc->di;
1030 bool modified = false;
1032 if (dc->state == D_DONE || dc->len == 1) {
1033 __remove_discard_cmd(sbi, dc);
1037 dcc->undiscard_blks -= di.len;
1039 if (blkaddr > di.lstart) {
1040 dc->len = blkaddr - dc->lstart;
1041 dcc->undiscard_blks += dc->len;
1042 __relocate_discard_cmd(dcc, dc);
1046 if (blkaddr < di.lstart + di.len - 1) {
1048 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1049 di.start + blkaddr + 1 - di.lstart,
1050 di.lstart + di.len - 1 - blkaddr,
1056 dcc->undiscard_blks += dc->len;
1057 __relocate_discard_cmd(dcc, dc);
1062 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1063 struct block_device *bdev, block_t lstart,
1064 block_t start, block_t len)
1066 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1067 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1068 struct discard_cmd *dc;
1069 struct discard_info di = {0};
1070 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1071 block_t end = lstart + len;
1073 mutex_lock(&dcc->cmd_lock);
1075 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1077 (struct rb_entry **)&prev_dc,
1078 (struct rb_entry **)&next_dc,
1079 &insert_p, &insert_parent, true);
1085 di.len = next_dc ? next_dc->lstart - lstart : len;
1086 di.len = min(di.len, len);
1091 struct rb_node *node;
1092 bool merged = false;
1093 struct discard_cmd *tdc = NULL;
1096 di.lstart = prev_dc->lstart + prev_dc->len;
1097 if (di.lstart < lstart)
1099 if (di.lstart >= end)
1102 if (!next_dc || next_dc->lstart > end)
1103 di.len = end - di.lstart;
1105 di.len = next_dc->lstart - di.lstart;
1106 di.start = start + di.lstart - lstart;
1112 if (prev_dc && prev_dc->state == D_PREP &&
1113 prev_dc->bdev == bdev &&
1114 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1115 prev_dc->di.len += di.len;
1116 dcc->undiscard_blks += di.len;
1117 __relocate_discard_cmd(dcc, prev_dc);
1123 if (next_dc && next_dc->state == D_PREP &&
1124 next_dc->bdev == bdev &&
1125 __is_discard_front_mergeable(&di, &next_dc->di)) {
1126 next_dc->di.lstart = di.lstart;
1127 next_dc->di.len += di.len;
1128 next_dc->di.start = di.start;
1129 dcc->undiscard_blks += di.len;
1130 __relocate_discard_cmd(dcc, next_dc);
1132 __remove_discard_cmd(sbi, tdc);
1137 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1138 di.len, NULL, NULL);
1145 node = rb_next(&prev_dc->rb_node);
1146 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1149 mutex_unlock(&dcc->cmd_lock);
1152 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1153 struct block_device *bdev, block_t blkstart, block_t blklen)
1155 block_t lblkstart = blkstart;
1157 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1160 int devi = f2fs_target_device_index(sbi, blkstart);
1162 blkstart -= FDEV(devi).start_blk;
1164 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1168 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1169 struct discard_policy *dpolicy)
1171 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1172 struct list_head *pend_list;
1173 struct discard_cmd *dc, *tmp;
1174 struct blk_plug plug;
1175 int i, iter = 0, issued = 0;
1176 bool io_interrupted = false;
1178 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1179 if (i + 1 < dpolicy->granularity)
1181 pend_list = &dcc->pend_list[i];
1183 mutex_lock(&dcc->cmd_lock);
1184 if (list_empty(pend_list))
1186 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1187 blk_start_plug(&plug);
1188 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1189 f2fs_bug_on(sbi, dc->state != D_PREP);
1191 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1193 io_interrupted = true;
1197 __submit_discard_cmd(sbi, dpolicy, dc);
1200 if (++iter >= dpolicy->max_requests)
1203 blk_finish_plug(&plug);
1205 mutex_unlock(&dcc->cmd_lock);
1207 if (iter >= dpolicy->max_requests)
1211 if (!issued && io_interrupted)
1217 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1219 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1220 struct list_head *pend_list;
1221 struct discard_cmd *dc, *tmp;
1223 bool dropped = false;
1225 mutex_lock(&dcc->cmd_lock);
1226 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1227 pend_list = &dcc->pend_list[i];
1228 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1229 f2fs_bug_on(sbi, dc->state != D_PREP);
1230 __remove_discard_cmd(sbi, dc);
1234 mutex_unlock(&dcc->cmd_lock);
1239 void drop_discard_cmd(struct f2fs_sb_info *sbi)
1241 __drop_discard_cmd(sbi);
1244 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1245 struct discard_cmd *dc)
1247 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1248 unsigned int len = 0;
1250 wait_for_completion_io(&dc->wait);
1251 mutex_lock(&dcc->cmd_lock);
1252 f2fs_bug_on(sbi, dc->state != D_DONE);
1257 __remove_discard_cmd(sbi, dc);
1259 mutex_unlock(&dcc->cmd_lock);
1264 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1265 struct discard_policy *dpolicy,
1266 block_t start, block_t end)
1268 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1269 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1270 &(dcc->fstrim_list) : &(dcc->wait_list);
1271 struct discard_cmd *dc, *tmp;
1273 unsigned int trimmed = 0;
1278 mutex_lock(&dcc->cmd_lock);
1279 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1280 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1282 if (dc->len < dpolicy->granularity)
1284 if (dc->state == D_DONE && !dc->ref) {
1285 wait_for_completion_io(&dc->wait);
1288 __remove_discard_cmd(sbi, dc);
1295 mutex_unlock(&dcc->cmd_lock);
1298 trimmed += __wait_one_discard_bio(sbi, dc);
1305 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1306 struct discard_policy *dpolicy)
1308 struct discard_policy dp;
1311 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1316 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1317 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1318 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1319 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1322 /* This should be covered by global mutex, &sit_i->sentry_lock */
1323 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1325 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1326 struct discard_cmd *dc;
1327 bool need_wait = false;
1329 mutex_lock(&dcc->cmd_lock);
1330 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1332 if (dc->state == D_PREP) {
1333 __punch_discard_cmd(sbi, dc, blkaddr);
1339 mutex_unlock(&dcc->cmd_lock);
1342 __wait_one_discard_bio(sbi, dc);
1345 void stop_discard_thread(struct f2fs_sb_info *sbi)
1347 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1349 if (dcc && dcc->f2fs_issue_discard) {
1350 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1352 dcc->f2fs_issue_discard = NULL;
1353 kthread_stop(discard_thread);
1357 /* This comes from f2fs_put_super */
1358 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1360 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1361 struct discard_policy dpolicy;
1364 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1365 dcc->discard_granularity);
1366 __issue_discard_cmd(sbi, &dpolicy);
1367 dropped = __drop_discard_cmd(sbi);
1369 /* just to make sure there is no pending discard commands */
1370 __wait_all_discard_cmd(sbi, NULL);
1374 static int issue_discard_thread(void *data)
1376 struct f2fs_sb_info *sbi = data;
1377 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1378 wait_queue_head_t *q = &dcc->discard_wait_queue;
1379 struct discard_policy dpolicy;
1380 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1386 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1387 dcc->discard_granularity);
1389 wait_event_interruptible_timeout(*q,
1390 kthread_should_stop() || freezing(current) ||
1392 msecs_to_jiffies(wait_ms));
1393 if (try_to_freeze())
1395 if (f2fs_readonly(sbi->sb))
1397 if (kthread_should_stop())
1399 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1400 wait_ms = dpolicy.max_interval;
1404 if (dcc->discard_wake)
1405 dcc->discard_wake = 0;
1407 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1408 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1410 sb_start_intwrite(sbi->sb);
1412 issued = __issue_discard_cmd(sbi, &dpolicy);
1414 __wait_all_discard_cmd(sbi, &dpolicy);
1415 wait_ms = dpolicy.min_interval;
1417 wait_ms = dpolicy.max_interval;
1420 sb_end_intwrite(sbi->sb);
1422 } while (!kthread_should_stop());
1426 #ifdef CONFIG_BLK_DEV_ZONED
1427 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1428 struct block_device *bdev, block_t blkstart, block_t blklen)
1430 sector_t sector, nr_sects;
1431 block_t lblkstart = blkstart;
1435 devi = f2fs_target_device_index(sbi, blkstart);
1436 blkstart -= FDEV(devi).start_blk;
1440 * We need to know the type of the zone: for conventional zones,
1441 * use regular discard if the drive supports it. For sequential
1442 * zones, reset the zone write pointer.
1444 switch (get_blkz_type(sbi, bdev, blkstart)) {
1446 case BLK_ZONE_TYPE_CONVENTIONAL:
1447 if (!blk_queue_discard(bdev_get_queue(bdev)))
1449 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1450 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1451 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1452 sector = SECTOR_FROM_BLOCK(blkstart);
1453 nr_sects = SECTOR_FROM_BLOCK(blklen);
1455 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1456 nr_sects != bdev_zone_sectors(bdev)) {
1457 f2fs_msg(sbi->sb, KERN_INFO,
1458 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1459 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1463 trace_f2fs_issue_reset_zone(bdev, blkstart);
1464 return blkdev_reset_zones(bdev, sector,
1465 nr_sects, GFP_NOFS);
1467 /* Unknown zone type: broken device ? */
1473 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1474 struct block_device *bdev, block_t blkstart, block_t blklen)
1476 #ifdef CONFIG_BLK_DEV_ZONED
1477 if (f2fs_sb_has_blkzoned(sbi->sb) &&
1478 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1479 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1481 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1484 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1485 block_t blkstart, block_t blklen)
1487 sector_t start = blkstart, len = 0;
1488 struct block_device *bdev;
1489 struct seg_entry *se;
1490 unsigned int offset;
1494 bdev = f2fs_target_device(sbi, blkstart, NULL);
1496 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1498 struct block_device *bdev2 =
1499 f2fs_target_device(sbi, i, NULL);
1501 if (bdev2 != bdev) {
1502 err = __issue_discard_async(sbi, bdev,
1512 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1513 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1515 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1516 sbi->discard_blks--;
1520 err = __issue_discard_async(sbi, bdev, start, len);
1524 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1527 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1528 int max_blocks = sbi->blocks_per_seg;
1529 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1530 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1531 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1532 unsigned long *discard_map = (unsigned long *)se->discard_map;
1533 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1534 unsigned int start = 0, end = -1;
1535 bool force = (cpc->reason & CP_DISCARD);
1536 struct discard_entry *de = NULL;
1537 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1540 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1544 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1545 SM_I(sbi)->dcc_info->nr_discards >=
1546 SM_I(sbi)->dcc_info->max_discards)
1550 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1551 for (i = 0; i < entries; i++)
1552 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1553 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1555 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1556 SM_I(sbi)->dcc_info->max_discards) {
1557 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1558 if (start >= max_blocks)
1561 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1562 if (force && start && end != max_blocks
1563 && (end - start) < cpc->trim_minlen)
1570 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1572 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1573 list_add_tail(&de->list, head);
1576 for (i = start; i < end; i++)
1577 __set_bit_le(i, (void *)de->discard_map);
1579 SM_I(sbi)->dcc_info->nr_discards += end - start;
1584 void release_discard_addrs(struct f2fs_sb_info *sbi)
1586 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1587 struct discard_entry *entry, *this;
1590 list_for_each_entry_safe(entry, this, head, list) {
1591 list_del(&entry->list);
1592 kmem_cache_free(discard_entry_slab, entry);
1597 * Should call clear_prefree_segments after checkpoint is done.
1599 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1601 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1604 mutex_lock(&dirty_i->seglist_lock);
1605 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1606 __set_test_and_free(sbi, segno);
1607 mutex_unlock(&dirty_i->seglist_lock);
1610 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1612 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1613 struct list_head *head = &dcc->entry_list;
1614 struct discard_entry *entry, *this;
1615 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1616 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1617 unsigned int start = 0, end = -1;
1618 unsigned int secno, start_segno;
1619 bool force = (cpc->reason & CP_DISCARD);
1621 mutex_lock(&dirty_i->seglist_lock);
1625 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1626 if (start >= MAIN_SEGS(sbi))
1628 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1631 for (i = start; i < end; i++)
1632 clear_bit(i, prefree_map);
1634 dirty_i->nr_dirty[PRE] -= end - start;
1636 if (!test_opt(sbi, DISCARD))
1639 if (force && start >= cpc->trim_start &&
1640 (end - 1) <= cpc->trim_end)
1643 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1644 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1645 (end - start) << sbi->log_blocks_per_seg);
1649 secno = GET_SEC_FROM_SEG(sbi, start);
1650 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1651 if (!IS_CURSEC(sbi, secno) &&
1652 !get_valid_blocks(sbi, start, true))
1653 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1654 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1656 start = start_segno + sbi->segs_per_sec;
1662 mutex_unlock(&dirty_i->seglist_lock);
1664 /* send small discards */
1665 list_for_each_entry_safe(entry, this, head, list) {
1666 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1667 bool is_valid = test_bit_le(0, entry->discard_map);
1671 next_pos = find_next_zero_bit_le(entry->discard_map,
1672 sbi->blocks_per_seg, cur_pos);
1673 len = next_pos - cur_pos;
1675 if (f2fs_sb_has_blkzoned(sbi->sb) ||
1676 (force && len < cpc->trim_minlen))
1679 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1683 next_pos = find_next_bit_le(entry->discard_map,
1684 sbi->blocks_per_seg, cur_pos);
1688 is_valid = !is_valid;
1690 if (cur_pos < sbi->blocks_per_seg)
1693 list_del(&entry->list);
1694 dcc->nr_discards -= total_len;
1695 kmem_cache_free(discard_entry_slab, entry);
1698 wake_up_discard_thread(sbi, false);
1701 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1703 dev_t dev = sbi->sb->s_bdev->bd_dev;
1704 struct discard_cmd_control *dcc;
1707 if (SM_I(sbi)->dcc_info) {
1708 dcc = SM_I(sbi)->dcc_info;
1712 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1716 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1717 INIT_LIST_HEAD(&dcc->entry_list);
1718 for (i = 0; i < MAX_PLIST_NUM; i++)
1719 INIT_LIST_HEAD(&dcc->pend_list[i]);
1720 INIT_LIST_HEAD(&dcc->wait_list);
1721 INIT_LIST_HEAD(&dcc->fstrim_list);
1722 mutex_init(&dcc->cmd_lock);
1723 atomic_set(&dcc->issued_discard, 0);
1724 atomic_set(&dcc->issing_discard, 0);
1725 atomic_set(&dcc->discard_cmd_cnt, 0);
1726 dcc->nr_discards = 0;
1727 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1728 dcc->undiscard_blks = 0;
1729 dcc->root = RB_ROOT;
1731 init_waitqueue_head(&dcc->discard_wait_queue);
1732 SM_I(sbi)->dcc_info = dcc;
1734 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1735 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1736 if (IS_ERR(dcc->f2fs_issue_discard)) {
1737 err = PTR_ERR(dcc->f2fs_issue_discard);
1739 SM_I(sbi)->dcc_info = NULL;
1746 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1748 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1753 stop_discard_thread(sbi);
1756 SM_I(sbi)->dcc_info = NULL;
1759 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1761 struct sit_info *sit_i = SIT_I(sbi);
1763 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1764 sit_i->dirty_sentries++;
1771 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1772 unsigned int segno, int modified)
1774 struct seg_entry *se = get_seg_entry(sbi, segno);
1777 __mark_sit_entry_dirty(sbi, segno);
1780 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1782 struct seg_entry *se;
1783 unsigned int segno, offset;
1784 long int new_vblocks;
1786 #ifdef CONFIG_F2FS_CHECK_FS
1790 segno = GET_SEGNO(sbi, blkaddr);
1792 se = get_seg_entry(sbi, segno);
1793 new_vblocks = se->valid_blocks + del;
1794 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1796 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1797 (new_vblocks > sbi->blocks_per_seg)));
1799 se->valid_blocks = new_vblocks;
1800 se->mtime = get_mtime(sbi);
1801 SIT_I(sbi)->max_mtime = se->mtime;
1803 /* Update valid block bitmap */
1805 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1806 #ifdef CONFIG_F2FS_CHECK_FS
1807 mir_exist = f2fs_test_and_set_bit(offset,
1808 se->cur_valid_map_mir);
1809 if (unlikely(exist != mir_exist)) {
1810 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1811 "when setting bitmap, blk:%u, old bit:%d",
1813 f2fs_bug_on(sbi, 1);
1816 if (unlikely(exist)) {
1817 f2fs_msg(sbi->sb, KERN_ERR,
1818 "Bitmap was wrongly set, blk:%u", blkaddr);
1819 f2fs_bug_on(sbi, 1);
1824 if (f2fs_discard_en(sbi) &&
1825 !f2fs_test_and_set_bit(offset, se->discard_map))
1826 sbi->discard_blks--;
1828 /* don't overwrite by SSR to keep node chain */
1829 if (IS_NODESEG(se->type)) {
1830 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1831 se->ckpt_valid_blocks++;
1834 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1835 #ifdef CONFIG_F2FS_CHECK_FS
1836 mir_exist = f2fs_test_and_clear_bit(offset,
1837 se->cur_valid_map_mir);
1838 if (unlikely(exist != mir_exist)) {
1839 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1840 "when clearing bitmap, blk:%u, old bit:%d",
1842 f2fs_bug_on(sbi, 1);
1845 if (unlikely(!exist)) {
1846 f2fs_msg(sbi->sb, KERN_ERR,
1847 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1848 f2fs_bug_on(sbi, 1);
1853 if (f2fs_discard_en(sbi) &&
1854 f2fs_test_and_clear_bit(offset, se->discard_map))
1855 sbi->discard_blks++;
1857 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1858 se->ckpt_valid_blocks += del;
1860 __mark_sit_entry_dirty(sbi, segno);
1862 /* update total number of valid blocks to be written in ckpt area */
1863 SIT_I(sbi)->written_valid_blocks += del;
1865 if (sbi->segs_per_sec > 1)
1866 get_sec_entry(sbi, segno)->valid_blocks += del;
1869 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1871 unsigned int segno = GET_SEGNO(sbi, addr);
1872 struct sit_info *sit_i = SIT_I(sbi);
1874 f2fs_bug_on(sbi, addr == NULL_ADDR);
1875 if (addr == NEW_ADDR)
1878 /* add it into sit main buffer */
1879 down_write(&sit_i->sentry_lock);
1881 update_sit_entry(sbi, addr, -1);
1883 /* add it into dirty seglist */
1884 locate_dirty_segment(sbi, segno);
1886 up_write(&sit_i->sentry_lock);
1889 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1891 struct sit_info *sit_i = SIT_I(sbi);
1892 unsigned int segno, offset;
1893 struct seg_entry *se;
1896 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1899 down_read(&sit_i->sentry_lock);
1901 segno = GET_SEGNO(sbi, blkaddr);
1902 se = get_seg_entry(sbi, segno);
1903 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1905 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1908 up_read(&sit_i->sentry_lock);
1914 * This function should be resided under the curseg_mutex lock
1916 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1917 struct f2fs_summary *sum)
1919 struct curseg_info *curseg = CURSEG_I(sbi, type);
1920 void *addr = curseg->sum_blk;
1921 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1922 memcpy(addr, sum, sizeof(struct f2fs_summary));
1926 * Calculate the number of current summary pages for writing
1928 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1930 int valid_sum_count = 0;
1933 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1934 if (sbi->ckpt->alloc_type[i] == SSR)
1935 valid_sum_count += sbi->blocks_per_seg;
1938 valid_sum_count += le16_to_cpu(
1939 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1941 valid_sum_count += curseg_blkoff(sbi, i);
1945 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1946 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1947 if (valid_sum_count <= sum_in_page)
1949 else if ((valid_sum_count - sum_in_page) <=
1950 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1956 * Caller should put this summary page
1958 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1960 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1963 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1965 struct page *page = grab_meta_page(sbi, blk_addr);
1967 memcpy(page_address(page), src, PAGE_SIZE);
1968 set_page_dirty(page);
1969 f2fs_put_page(page, 1);
1972 static void write_sum_page(struct f2fs_sb_info *sbi,
1973 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1975 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1978 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1979 int type, block_t blk_addr)
1981 struct curseg_info *curseg = CURSEG_I(sbi, type);
1982 struct page *page = grab_meta_page(sbi, blk_addr);
1983 struct f2fs_summary_block *src = curseg->sum_blk;
1984 struct f2fs_summary_block *dst;
1986 dst = (struct f2fs_summary_block *)page_address(page);
1987 memset(dst, 0, PAGE_SIZE);
1989 mutex_lock(&curseg->curseg_mutex);
1991 down_read(&curseg->journal_rwsem);
1992 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1993 up_read(&curseg->journal_rwsem);
1995 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1996 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1998 mutex_unlock(&curseg->curseg_mutex);
2000 set_page_dirty(page);
2001 f2fs_put_page(page, 1);
2004 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2006 struct curseg_info *curseg = CURSEG_I(sbi, type);
2007 unsigned int segno = curseg->segno + 1;
2008 struct free_segmap_info *free_i = FREE_I(sbi);
2010 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2011 return !test_bit(segno, free_i->free_segmap);
2016 * Find a new segment from the free segments bitmap to right order
2017 * This function should be returned with success, otherwise BUG
2019 static void get_new_segment(struct f2fs_sb_info *sbi,
2020 unsigned int *newseg, bool new_sec, int dir)
2022 struct free_segmap_info *free_i = FREE_I(sbi);
2023 unsigned int segno, secno, zoneno;
2024 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2025 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2026 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2027 unsigned int left_start = hint;
2032 spin_lock(&free_i->segmap_lock);
2034 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2035 segno = find_next_zero_bit(free_i->free_segmap,
2036 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2037 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2041 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2042 if (secno >= MAIN_SECS(sbi)) {
2043 if (dir == ALLOC_RIGHT) {
2044 secno = find_next_zero_bit(free_i->free_secmap,
2046 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2049 left_start = hint - 1;
2055 while (test_bit(left_start, free_i->free_secmap)) {
2056 if (left_start > 0) {
2060 left_start = find_next_zero_bit(free_i->free_secmap,
2062 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2067 segno = GET_SEG_FROM_SEC(sbi, secno);
2068 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2070 /* give up on finding another zone */
2073 if (sbi->secs_per_zone == 1)
2075 if (zoneno == old_zoneno)
2077 if (dir == ALLOC_LEFT) {
2078 if (!go_left && zoneno + 1 >= total_zones)
2080 if (go_left && zoneno == 0)
2083 for (i = 0; i < NR_CURSEG_TYPE; i++)
2084 if (CURSEG_I(sbi, i)->zone == zoneno)
2087 if (i < NR_CURSEG_TYPE) {
2088 /* zone is in user, try another */
2090 hint = zoneno * sbi->secs_per_zone - 1;
2091 else if (zoneno + 1 >= total_zones)
2094 hint = (zoneno + 1) * sbi->secs_per_zone;
2096 goto find_other_zone;
2099 /* set it as dirty segment in free segmap */
2100 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2101 __set_inuse(sbi, segno);
2103 spin_unlock(&free_i->segmap_lock);
2106 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2108 struct curseg_info *curseg = CURSEG_I(sbi, type);
2109 struct summary_footer *sum_footer;
2111 curseg->segno = curseg->next_segno;
2112 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2113 curseg->next_blkoff = 0;
2114 curseg->next_segno = NULL_SEGNO;
2116 sum_footer = &(curseg->sum_blk->footer);
2117 memset(sum_footer, 0, sizeof(struct summary_footer));
2118 if (IS_DATASEG(type))
2119 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2120 if (IS_NODESEG(type))
2121 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2122 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2125 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2127 /* if segs_per_sec is large than 1, we need to keep original policy. */
2128 if (sbi->segs_per_sec != 1)
2129 return CURSEG_I(sbi, type)->segno;
2131 if (test_opt(sbi, NOHEAP) &&
2132 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2135 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2136 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2138 /* find segments from 0 to reuse freed segments */
2139 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2142 return CURSEG_I(sbi, type)->segno;
2146 * Allocate a current working segment.
2147 * This function always allocates a free segment in LFS manner.
2149 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2151 struct curseg_info *curseg = CURSEG_I(sbi, type);
2152 unsigned int segno = curseg->segno;
2153 int dir = ALLOC_LEFT;
2155 write_sum_page(sbi, curseg->sum_blk,
2156 GET_SUM_BLOCK(sbi, segno));
2157 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2160 if (test_opt(sbi, NOHEAP))
2163 segno = __get_next_segno(sbi, type);
2164 get_new_segment(sbi, &segno, new_sec, dir);
2165 curseg->next_segno = segno;
2166 reset_curseg(sbi, type, 1);
2167 curseg->alloc_type = LFS;
2170 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2171 struct curseg_info *seg, block_t start)
2173 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2174 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2175 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2176 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2177 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2180 for (i = 0; i < entries; i++)
2181 target_map[i] = ckpt_map[i] | cur_map[i];
2183 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2185 seg->next_blkoff = pos;
2189 * If a segment is written by LFS manner, next block offset is just obtained
2190 * by increasing the current block offset. However, if a segment is written by
2191 * SSR manner, next block offset obtained by calling __next_free_blkoff
2193 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2194 struct curseg_info *seg)
2196 if (seg->alloc_type == SSR)
2197 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2203 * This function always allocates a used segment(from dirty seglist) by SSR
2204 * manner, so it should recover the existing segment information of valid blocks
2206 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2208 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2209 struct curseg_info *curseg = CURSEG_I(sbi, type);
2210 unsigned int new_segno = curseg->next_segno;
2211 struct f2fs_summary_block *sum_node;
2212 struct page *sum_page;
2214 write_sum_page(sbi, curseg->sum_blk,
2215 GET_SUM_BLOCK(sbi, curseg->segno));
2216 __set_test_and_inuse(sbi, new_segno);
2218 mutex_lock(&dirty_i->seglist_lock);
2219 __remove_dirty_segment(sbi, new_segno, PRE);
2220 __remove_dirty_segment(sbi, new_segno, DIRTY);
2221 mutex_unlock(&dirty_i->seglist_lock);
2223 reset_curseg(sbi, type, 1);
2224 curseg->alloc_type = SSR;
2225 __next_free_blkoff(sbi, curseg, 0);
2227 sum_page = get_sum_page(sbi, new_segno);
2228 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2229 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2230 f2fs_put_page(sum_page, 1);
2233 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2235 struct curseg_info *curseg = CURSEG_I(sbi, type);
2236 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2237 unsigned segno = NULL_SEGNO;
2239 bool reversed = false;
2241 /* need_SSR() already forces to do this */
2242 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2243 curseg->next_segno = segno;
2247 /* For node segments, let's do SSR more intensively */
2248 if (IS_NODESEG(type)) {
2249 if (type >= CURSEG_WARM_NODE) {
2251 i = CURSEG_COLD_NODE;
2253 i = CURSEG_HOT_NODE;
2255 cnt = NR_CURSEG_NODE_TYPE;
2257 if (type >= CURSEG_WARM_DATA) {
2259 i = CURSEG_COLD_DATA;
2261 i = CURSEG_HOT_DATA;
2263 cnt = NR_CURSEG_DATA_TYPE;
2266 for (; cnt-- > 0; reversed ? i-- : i++) {
2269 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2270 curseg->next_segno = segno;
2278 * flush out current segment and replace it with new segment
2279 * This function should be returned with success, otherwise BUG
2281 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2282 int type, bool force)
2284 struct curseg_info *curseg = CURSEG_I(sbi, type);
2287 new_curseg(sbi, type, true);
2288 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2289 type == CURSEG_WARM_NODE)
2290 new_curseg(sbi, type, false);
2291 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2292 new_curseg(sbi, type, false);
2293 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2294 change_curseg(sbi, type);
2296 new_curseg(sbi, type, false);
2298 stat_inc_seg_type(sbi, curseg);
2301 void allocate_new_segments(struct f2fs_sb_info *sbi)
2303 struct curseg_info *curseg;
2304 unsigned int old_segno;
2307 down_write(&SIT_I(sbi)->sentry_lock);
2309 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2310 curseg = CURSEG_I(sbi, i);
2311 old_segno = curseg->segno;
2312 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2313 locate_dirty_segment(sbi, old_segno);
2316 up_write(&SIT_I(sbi)->sentry_lock);
2319 static const struct segment_allocation default_salloc_ops = {
2320 .allocate_segment = allocate_segment_by_default,
2323 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2325 __u64 trim_start = cpc->trim_start;
2326 bool has_candidate = false;
2328 down_write(&SIT_I(sbi)->sentry_lock);
2329 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2330 if (add_discard_addrs(sbi, cpc, true)) {
2331 has_candidate = true;
2335 up_write(&SIT_I(sbi)->sentry_lock);
2337 cpc->trim_start = trim_start;
2338 return has_candidate;
2341 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2342 struct discard_policy *dpolicy,
2343 unsigned int start, unsigned int end)
2345 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2346 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2347 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2348 struct discard_cmd *dc;
2349 struct blk_plug plug;
2355 mutex_lock(&dcc->cmd_lock);
2356 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
2358 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
2360 (struct rb_entry **)&prev_dc,
2361 (struct rb_entry **)&next_dc,
2362 &insert_p, &insert_parent, true);
2366 blk_start_plug(&plug);
2368 while (dc && dc->lstart <= end) {
2369 struct rb_node *node;
2371 if (dc->len < dpolicy->granularity)
2374 if (dc->state != D_PREP) {
2375 list_move_tail(&dc->list, &dcc->fstrim_list);
2379 __submit_discard_cmd(sbi, dpolicy, dc);
2381 if (++issued >= dpolicy->max_requests) {
2382 start = dc->lstart + dc->len;
2384 blk_finish_plug(&plug);
2385 mutex_unlock(&dcc->cmd_lock);
2386 __wait_all_discard_cmd(sbi, NULL);
2387 congestion_wait(BLK_RW_ASYNC, HZ/50);
2391 node = rb_next(&dc->rb_node);
2392 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2394 if (fatal_signal_pending(current))
2398 blk_finish_plug(&plug);
2399 mutex_unlock(&dcc->cmd_lock);
2402 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2404 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2405 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2406 unsigned int start_segno, end_segno;
2407 block_t start_block, end_block;
2408 struct cp_control cpc;
2409 struct discard_policy dpolicy;
2410 unsigned long long trimmed = 0;
2413 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2416 if (end <= MAIN_BLKADDR(sbi))
2419 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2420 f2fs_msg(sbi->sb, KERN_WARNING,
2421 "Found FS corruption, run fsck to fix.");
2425 /* start/end segment number in main_area */
2426 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2427 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2428 GET_SEGNO(sbi, end);
2430 cpc.reason = CP_DISCARD;
2431 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2432 cpc.trim_start = start_segno;
2433 cpc.trim_end = end_segno;
2435 if (sbi->discard_blks == 0)
2438 mutex_lock(&sbi->gc_mutex);
2439 err = write_checkpoint(sbi, &cpc);
2440 mutex_unlock(&sbi->gc_mutex);
2444 start_block = START_BLOCK(sbi, start_segno);
2445 end_block = START_BLOCK(sbi, end_segno + 1);
2447 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2448 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2451 * We filed discard candidates, but actually we don't need to wait for
2452 * all of them, since they'll be issued in idle time along with runtime
2453 * discard option. User configuration looks like using runtime discard
2454 * or periodic fstrim instead of it.
2456 if (!test_opt(sbi, DISCARD)) {
2457 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2458 start_block, end_block);
2459 range->len = F2FS_BLK_TO_BYTES(trimmed);
2465 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2467 struct curseg_info *curseg = CURSEG_I(sbi, type);
2468 if (curseg->next_blkoff < sbi->blocks_per_seg)
2473 int rw_hint_to_seg_type(enum rw_hint hint)
2476 case WRITE_LIFE_SHORT:
2477 return CURSEG_HOT_DATA;
2478 case WRITE_LIFE_EXTREME:
2479 return CURSEG_COLD_DATA;
2481 return CURSEG_WARM_DATA;
2485 /* This returns write hints for each segment type. This hints will be
2486 * passed down to block layer. There are mapping tables which depend on
2487 * the mount option 'whint_mode'.
2489 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2491 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2495 * META WRITE_LIFE_NOT_SET
2499 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2500 * extension list " "
2503 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2504 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2505 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2506 * WRITE_LIFE_NONE " "
2507 * WRITE_LIFE_MEDIUM " "
2508 * WRITE_LIFE_LONG " "
2511 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2512 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2513 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2514 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2515 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2516 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2518 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2522 * META WRITE_LIFE_MEDIUM;
2523 * HOT_NODE WRITE_LIFE_NOT_SET
2525 * COLD_NODE WRITE_LIFE_NONE
2526 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2527 * extension list " "
2530 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2531 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2532 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2533 * WRITE_LIFE_NONE " "
2534 * WRITE_LIFE_MEDIUM " "
2535 * WRITE_LIFE_LONG " "
2538 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2539 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2540 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2541 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2542 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2543 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2546 enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2547 enum page_type type, enum temp_type temp)
2549 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2552 return WRITE_LIFE_NOT_SET;
2553 else if (temp == HOT)
2554 return WRITE_LIFE_SHORT;
2555 else if (temp == COLD)
2556 return WRITE_LIFE_EXTREME;
2558 return WRITE_LIFE_NOT_SET;
2560 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2563 return WRITE_LIFE_LONG;
2564 else if (temp == HOT)
2565 return WRITE_LIFE_SHORT;
2566 else if (temp == COLD)
2567 return WRITE_LIFE_EXTREME;
2568 } else if (type == NODE) {
2569 if (temp == WARM || temp == HOT)
2570 return WRITE_LIFE_NOT_SET;
2571 else if (temp == COLD)
2572 return WRITE_LIFE_NONE;
2573 } else if (type == META) {
2574 return WRITE_LIFE_MEDIUM;
2577 return WRITE_LIFE_NOT_SET;
2580 static int __get_segment_type_2(struct f2fs_io_info *fio)
2582 if (fio->type == DATA)
2583 return CURSEG_HOT_DATA;
2585 return CURSEG_HOT_NODE;
2588 static int __get_segment_type_4(struct f2fs_io_info *fio)
2590 if (fio->type == DATA) {
2591 struct inode *inode = fio->page->mapping->host;
2593 if (S_ISDIR(inode->i_mode))
2594 return CURSEG_HOT_DATA;
2596 return CURSEG_COLD_DATA;
2598 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2599 return CURSEG_WARM_NODE;
2601 return CURSEG_COLD_NODE;
2605 static int __get_segment_type_6(struct f2fs_io_info *fio)
2607 if (fio->type == DATA) {
2608 struct inode *inode = fio->page->mapping->host;
2610 if (is_cold_data(fio->page) || file_is_cold(inode))
2611 return CURSEG_COLD_DATA;
2612 if (file_is_hot(inode) ||
2613 is_inode_flag_set(inode, FI_HOT_DATA))
2614 return CURSEG_HOT_DATA;
2615 return rw_hint_to_seg_type(inode->i_write_hint);
2617 if (IS_DNODE(fio->page))
2618 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2620 return CURSEG_COLD_NODE;
2624 static int __get_segment_type(struct f2fs_io_info *fio)
2628 switch (F2FS_OPTION(fio->sbi).active_logs) {
2630 type = __get_segment_type_2(fio);
2633 type = __get_segment_type_4(fio);
2636 type = __get_segment_type_6(fio);
2639 f2fs_bug_on(fio->sbi, true);
2644 else if (IS_WARM(type))
2651 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2652 block_t old_blkaddr, block_t *new_blkaddr,
2653 struct f2fs_summary *sum, int type,
2654 struct f2fs_io_info *fio, bool add_list)
2656 struct sit_info *sit_i = SIT_I(sbi);
2657 struct curseg_info *curseg = CURSEG_I(sbi, type);
2659 down_read(&SM_I(sbi)->curseg_lock);
2661 mutex_lock(&curseg->curseg_mutex);
2662 down_write(&sit_i->sentry_lock);
2664 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2666 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2669 * __add_sum_entry should be resided under the curseg_mutex
2670 * because, this function updates a summary entry in the
2671 * current summary block.
2673 __add_sum_entry(sbi, type, sum);
2675 __refresh_next_blkoff(sbi, curseg);
2677 stat_inc_block_count(sbi, curseg);
2680 * SIT information should be updated before segment allocation,
2681 * since SSR needs latest valid block information.
2683 update_sit_entry(sbi, *new_blkaddr, 1);
2684 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2685 update_sit_entry(sbi, old_blkaddr, -1);
2687 if (!__has_curseg_space(sbi, type))
2688 sit_i->s_ops->allocate_segment(sbi, type, false);
2691 * segment dirty status should be updated after segment allocation,
2692 * so we just need to update status only one time after previous
2693 * segment being closed.
2695 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2696 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2698 up_write(&sit_i->sentry_lock);
2700 if (page && IS_NODESEG(type)) {
2701 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2703 f2fs_inode_chksum_set(sbi, page);
2707 struct f2fs_bio_info *io;
2709 INIT_LIST_HEAD(&fio->list);
2710 fio->in_list = true;
2711 io = sbi->write_io[fio->type] + fio->temp;
2712 spin_lock(&io->io_lock);
2713 list_add_tail(&fio->list, &io->io_list);
2714 spin_unlock(&io->io_lock);
2717 mutex_unlock(&curseg->curseg_mutex);
2719 up_read(&SM_I(sbi)->curseg_lock);
2722 static void update_device_state(struct f2fs_io_info *fio)
2724 struct f2fs_sb_info *sbi = fio->sbi;
2725 unsigned int devidx;
2730 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2732 /* update device state for fsync */
2733 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2735 /* update device state for checkpoint */
2736 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2737 spin_lock(&sbi->dev_lock);
2738 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2739 spin_unlock(&sbi->dev_lock);
2743 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2745 int type = __get_segment_type(fio);
2749 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2750 &fio->new_blkaddr, sum, type, fio, true);
2752 /* writeout dirty page into bdev */
2753 err = f2fs_submit_page_write(fio);
2754 if (err == -EAGAIN) {
2755 fio->old_blkaddr = fio->new_blkaddr;
2758 update_device_state(fio);
2762 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2763 enum iostat_type io_type)
2765 struct f2fs_io_info fio = {
2770 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2771 .old_blkaddr = page->index,
2772 .new_blkaddr = page->index,
2774 .encrypted_page = NULL,
2778 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2779 fio.op_flags &= ~REQ_META;
2781 set_page_writeback(page);
2782 ClearPageError(page);
2783 f2fs_submit_page_write(&fio);
2785 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2788 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2790 struct f2fs_summary sum;
2792 set_summary(&sum, nid, 0, 0);
2793 do_write_page(&sum, fio);
2795 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2798 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2800 struct f2fs_sb_info *sbi = fio->sbi;
2801 struct f2fs_summary sum;
2802 struct node_info ni;
2804 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2805 get_node_info(sbi, dn->nid, &ni);
2806 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2807 do_write_page(&sum, fio);
2808 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2810 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2813 int rewrite_data_page(struct f2fs_io_info *fio)
2816 struct f2fs_sb_info *sbi = fio->sbi;
2818 fio->new_blkaddr = fio->old_blkaddr;
2819 /* i/o temperature is needed for passing down write hints */
2820 __get_segment_type(fio);
2822 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2823 GET_SEGNO(sbi, fio->new_blkaddr))->type));
2825 stat_inc_inplace_blocks(fio->sbi);
2827 err = f2fs_submit_page_bio(fio);
2829 update_device_state(fio);
2831 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2836 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2841 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2842 if (CURSEG_I(sbi, i)->segno == segno)
2848 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2849 block_t old_blkaddr, block_t new_blkaddr,
2850 bool recover_curseg, bool recover_newaddr)
2852 struct sit_info *sit_i = SIT_I(sbi);
2853 struct curseg_info *curseg;
2854 unsigned int segno, old_cursegno;
2855 struct seg_entry *se;
2857 unsigned short old_blkoff;
2859 segno = GET_SEGNO(sbi, new_blkaddr);
2860 se = get_seg_entry(sbi, segno);
2863 down_write(&SM_I(sbi)->curseg_lock);
2865 if (!recover_curseg) {
2866 /* for recovery flow */
2867 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2868 if (old_blkaddr == NULL_ADDR)
2869 type = CURSEG_COLD_DATA;
2871 type = CURSEG_WARM_DATA;
2874 if (IS_CURSEG(sbi, segno)) {
2875 /* se->type is volatile as SSR allocation */
2876 type = __f2fs_get_curseg(sbi, segno);
2877 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2879 type = CURSEG_WARM_DATA;
2883 f2fs_bug_on(sbi, !IS_DATASEG(type));
2884 curseg = CURSEG_I(sbi, type);
2886 mutex_lock(&curseg->curseg_mutex);
2887 down_write(&sit_i->sentry_lock);
2889 old_cursegno = curseg->segno;
2890 old_blkoff = curseg->next_blkoff;
2892 /* change the current segment */
2893 if (segno != curseg->segno) {
2894 curseg->next_segno = segno;
2895 change_curseg(sbi, type);
2898 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2899 __add_sum_entry(sbi, type, sum);
2901 if (!recover_curseg || recover_newaddr)
2902 update_sit_entry(sbi, new_blkaddr, 1);
2903 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2904 update_sit_entry(sbi, old_blkaddr, -1);
2906 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2907 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2909 locate_dirty_segment(sbi, old_cursegno);
2911 if (recover_curseg) {
2912 if (old_cursegno != curseg->segno) {
2913 curseg->next_segno = old_cursegno;
2914 change_curseg(sbi, type);
2916 curseg->next_blkoff = old_blkoff;
2919 up_write(&sit_i->sentry_lock);
2920 mutex_unlock(&curseg->curseg_mutex);
2921 up_write(&SM_I(sbi)->curseg_lock);
2924 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2925 block_t old_addr, block_t new_addr,
2926 unsigned char version, bool recover_curseg,
2927 bool recover_newaddr)
2929 struct f2fs_summary sum;
2931 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2933 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2934 recover_curseg, recover_newaddr);
2936 f2fs_update_data_blkaddr(dn, new_addr);
2939 void f2fs_wait_on_page_writeback(struct page *page,
2940 enum page_type type, bool ordered)
2942 if (PageWriteback(page)) {
2943 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2945 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2946 0, page->index, type);
2948 wait_on_page_writeback(page);
2950 wait_for_stable_page(page);
2954 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2958 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
2961 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2963 f2fs_wait_on_page_writeback(cpage, DATA, true);
2964 f2fs_put_page(cpage, 1);
2968 static void read_compacted_summaries(struct f2fs_sb_info *sbi)
2970 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2971 struct curseg_info *seg_i;
2972 unsigned char *kaddr;
2977 start = start_sum_block(sbi);
2979 page = get_meta_page(sbi, start++);
2980 kaddr = (unsigned char *)page_address(page);
2982 /* Step 1: restore nat cache */
2983 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2984 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2986 /* Step 2: restore sit cache */
2987 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2988 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2989 offset = 2 * SUM_JOURNAL_SIZE;
2991 /* Step 3: restore summary entries */
2992 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2993 unsigned short blk_off;
2996 seg_i = CURSEG_I(sbi, i);
2997 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2998 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2999 seg_i->next_segno = segno;
3000 reset_curseg(sbi, i, 0);
3001 seg_i->alloc_type = ckpt->alloc_type[i];
3002 seg_i->next_blkoff = blk_off;
3004 if (seg_i->alloc_type == SSR)
3005 blk_off = sbi->blocks_per_seg;
3007 for (j = 0; j < blk_off; j++) {
3008 struct f2fs_summary *s;
3009 s = (struct f2fs_summary *)(kaddr + offset);
3010 seg_i->sum_blk->entries[j] = *s;
3011 offset += SUMMARY_SIZE;
3012 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3016 f2fs_put_page(page, 1);
3019 page = get_meta_page(sbi, start++);
3020 kaddr = (unsigned char *)page_address(page);
3024 f2fs_put_page(page, 1);
3027 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3029 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3030 struct f2fs_summary_block *sum;
3031 struct curseg_info *curseg;
3033 unsigned short blk_off;
3034 unsigned int segno = 0;
3035 block_t blk_addr = 0;
3037 /* get segment number and block addr */
3038 if (IS_DATASEG(type)) {
3039 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3040 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3042 if (__exist_node_summaries(sbi))
3043 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3045 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3047 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3049 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3051 if (__exist_node_summaries(sbi))
3052 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3053 type - CURSEG_HOT_NODE);
3055 blk_addr = GET_SUM_BLOCK(sbi, segno);
3058 new = get_meta_page(sbi, blk_addr);
3059 sum = (struct f2fs_summary_block *)page_address(new);
3061 if (IS_NODESEG(type)) {
3062 if (__exist_node_summaries(sbi)) {
3063 struct f2fs_summary *ns = &sum->entries[0];
3065 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3067 ns->ofs_in_node = 0;
3070 restore_node_summary(sbi, segno, sum);
3074 /* set uncompleted segment to curseg */
3075 curseg = CURSEG_I(sbi, type);
3076 mutex_lock(&curseg->curseg_mutex);
3078 /* update journal info */
3079 down_write(&curseg->journal_rwsem);
3080 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3081 up_write(&curseg->journal_rwsem);
3083 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3084 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3085 curseg->next_segno = segno;
3086 reset_curseg(sbi, type, 0);
3087 curseg->alloc_type = ckpt->alloc_type[type];
3088 curseg->next_blkoff = blk_off;
3089 mutex_unlock(&curseg->curseg_mutex);
3090 f2fs_put_page(new, 1);
3094 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3096 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3097 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3098 int type = CURSEG_HOT_DATA;
3101 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3102 int npages = npages_for_summary_flush(sbi, true);
3105 ra_meta_pages(sbi, start_sum_block(sbi), npages,
3108 /* restore for compacted data summary */
3109 read_compacted_summaries(sbi);
3110 type = CURSEG_HOT_NODE;
3113 if (__exist_node_summaries(sbi))
3114 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3115 NR_CURSEG_TYPE - type, META_CP, true);
3117 for (; type <= CURSEG_COLD_NODE; type++) {
3118 err = read_normal_summaries(sbi, type);
3123 /* sanity check for summary blocks */
3124 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3125 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3131 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3134 unsigned char *kaddr;
3135 struct f2fs_summary *summary;
3136 struct curseg_info *seg_i;
3137 int written_size = 0;
3140 page = grab_meta_page(sbi, blkaddr++);
3141 kaddr = (unsigned char *)page_address(page);
3142 memset(kaddr, 0, PAGE_SIZE);
3144 /* Step 1: write nat cache */
3145 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3146 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3147 written_size += SUM_JOURNAL_SIZE;
3149 /* Step 2: write sit cache */
3150 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3151 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3152 written_size += SUM_JOURNAL_SIZE;
3154 /* Step 3: write summary entries */
3155 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3156 unsigned short blkoff;
3157 seg_i = CURSEG_I(sbi, i);
3158 if (sbi->ckpt->alloc_type[i] == SSR)
3159 blkoff = sbi->blocks_per_seg;
3161 blkoff = curseg_blkoff(sbi, i);
3163 for (j = 0; j < blkoff; j++) {
3165 page = grab_meta_page(sbi, blkaddr++);
3166 kaddr = (unsigned char *)page_address(page);
3167 memset(kaddr, 0, PAGE_SIZE);
3170 summary = (struct f2fs_summary *)(kaddr + written_size);
3171 *summary = seg_i->sum_blk->entries[j];
3172 written_size += SUMMARY_SIZE;
3174 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3178 set_page_dirty(page);
3179 f2fs_put_page(page, 1);
3184 set_page_dirty(page);
3185 f2fs_put_page(page, 1);
3189 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3190 block_t blkaddr, int type)
3193 if (IS_DATASEG(type))
3194 end = type + NR_CURSEG_DATA_TYPE;
3196 end = type + NR_CURSEG_NODE_TYPE;
3198 for (i = type; i < end; i++)
3199 write_current_sum_page(sbi, i, blkaddr + (i - type));
3202 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3204 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3205 write_compacted_summaries(sbi, start_blk);
3207 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3210 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3212 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3215 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3216 unsigned int val, int alloc)
3220 if (type == NAT_JOURNAL) {
3221 for (i = 0; i < nats_in_cursum(journal); i++) {
3222 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3225 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3226 return update_nats_in_cursum(journal, 1);
3227 } else if (type == SIT_JOURNAL) {
3228 for (i = 0; i < sits_in_cursum(journal); i++)
3229 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3231 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3232 return update_sits_in_cursum(journal, 1);
3237 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3240 return get_meta_page(sbi, current_sit_addr(sbi, segno));
3243 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3246 struct sit_info *sit_i = SIT_I(sbi);
3248 pgoff_t src_off, dst_off;
3250 src_off = current_sit_addr(sbi, start);
3251 dst_off = next_sit_addr(sbi, src_off);
3253 page = grab_meta_page(sbi, dst_off);
3254 seg_info_to_sit_page(sbi, page, start);
3256 set_page_dirty(page);
3257 set_to_next_sit(sit_i, start);
3262 static struct sit_entry_set *grab_sit_entry_set(void)
3264 struct sit_entry_set *ses =
3265 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3268 INIT_LIST_HEAD(&ses->set_list);
3272 static void release_sit_entry_set(struct sit_entry_set *ses)
3274 list_del(&ses->set_list);
3275 kmem_cache_free(sit_entry_set_slab, ses);
3278 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3279 struct list_head *head)
3281 struct sit_entry_set *next = ses;
3283 if (list_is_last(&ses->set_list, head))
3286 list_for_each_entry_continue(next, head, set_list)
3287 if (ses->entry_cnt <= next->entry_cnt)
3290 list_move_tail(&ses->set_list, &next->set_list);
3293 static void add_sit_entry(unsigned int segno, struct list_head *head)
3295 struct sit_entry_set *ses;
3296 unsigned int start_segno = START_SEGNO(segno);
3298 list_for_each_entry(ses, head, set_list) {
3299 if (ses->start_segno == start_segno) {
3301 adjust_sit_entry_set(ses, head);
3306 ses = grab_sit_entry_set();
3308 ses->start_segno = start_segno;
3310 list_add(&ses->set_list, head);
3313 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3315 struct f2fs_sm_info *sm_info = SM_I(sbi);
3316 struct list_head *set_list = &sm_info->sit_entry_set;
3317 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3320 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3321 add_sit_entry(segno, set_list);
3324 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3326 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3327 struct f2fs_journal *journal = curseg->journal;
3330 down_write(&curseg->journal_rwsem);
3331 for (i = 0; i < sits_in_cursum(journal); i++) {
3335 segno = le32_to_cpu(segno_in_journal(journal, i));
3336 dirtied = __mark_sit_entry_dirty(sbi, segno);
3339 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3341 update_sits_in_cursum(journal, -i);
3342 up_write(&curseg->journal_rwsem);
3346 * CP calls this function, which flushes SIT entries including sit_journal,
3347 * and moves prefree segs to free segs.
3349 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3351 struct sit_info *sit_i = SIT_I(sbi);
3352 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3353 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3354 struct f2fs_journal *journal = curseg->journal;
3355 struct sit_entry_set *ses, *tmp;
3356 struct list_head *head = &SM_I(sbi)->sit_entry_set;
3357 bool to_journal = true;
3358 struct seg_entry *se;
3360 down_write(&sit_i->sentry_lock);
3362 if (!sit_i->dirty_sentries)
3366 * add and account sit entries of dirty bitmap in sit entry
3369 add_sits_in_set(sbi);
3372 * if there are no enough space in journal to store dirty sit
3373 * entries, remove all entries from journal and add and account
3374 * them in sit entry set.
3376 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3377 remove_sits_in_journal(sbi);
3380 * there are two steps to flush sit entries:
3381 * #1, flush sit entries to journal in current cold data summary block.
3382 * #2, flush sit entries to sit page.
3384 list_for_each_entry_safe(ses, tmp, head, set_list) {
3385 struct page *page = NULL;
3386 struct f2fs_sit_block *raw_sit = NULL;
3387 unsigned int start_segno = ses->start_segno;
3388 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3389 (unsigned long)MAIN_SEGS(sbi));
3390 unsigned int segno = start_segno;
3393 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3397 down_write(&curseg->journal_rwsem);
3399 page = get_next_sit_page(sbi, start_segno);
3400 raw_sit = page_address(page);
3403 /* flush dirty sit entries in region of current sit set */
3404 for_each_set_bit_from(segno, bitmap, end) {
3405 int offset, sit_offset;
3407 se = get_seg_entry(sbi, segno);
3408 #ifdef CONFIG_F2FS_CHECK_FS
3409 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3410 SIT_VBLOCK_MAP_SIZE))
3411 f2fs_bug_on(sbi, 1);
3414 /* add discard candidates */
3415 if (!(cpc->reason & CP_DISCARD)) {
3416 cpc->trim_start = segno;
3417 add_discard_addrs(sbi, cpc, false);
3421 offset = lookup_journal_in_cursum(journal,
3422 SIT_JOURNAL, segno, 1);
3423 f2fs_bug_on(sbi, offset < 0);
3424 segno_in_journal(journal, offset) =
3426 seg_info_to_raw_sit(se,
3427 &sit_in_journal(journal, offset));
3428 check_block_count(sbi, segno,
3429 &sit_in_journal(journal, offset));
3431 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3432 seg_info_to_raw_sit(se,
3433 &raw_sit->entries[sit_offset]);
3434 check_block_count(sbi, segno,
3435 &raw_sit->entries[sit_offset]);
3438 __clear_bit(segno, bitmap);
3439 sit_i->dirty_sentries--;
3444 up_write(&curseg->journal_rwsem);
3446 f2fs_put_page(page, 1);
3448 f2fs_bug_on(sbi, ses->entry_cnt);
3449 release_sit_entry_set(ses);
3452 f2fs_bug_on(sbi, !list_empty(head));
3453 f2fs_bug_on(sbi, sit_i->dirty_sentries);
3455 if (cpc->reason & CP_DISCARD) {
3456 __u64 trim_start = cpc->trim_start;
3458 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3459 add_discard_addrs(sbi, cpc, false);
3461 cpc->trim_start = trim_start;
3463 up_write(&sit_i->sentry_lock);
3465 set_prefree_as_free_segments(sbi);
3468 static int build_sit_info(struct f2fs_sb_info *sbi)
3470 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3471 struct sit_info *sit_i;
3472 unsigned int sit_segs, start;
3474 unsigned int bitmap_size;
3476 /* allocate memory for SIT information */
3477 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3481 SM_I(sbi)->sit_info = sit_i;
3483 sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
3484 sizeof(struct seg_entry), GFP_KERNEL);
3485 if (!sit_i->sentries)
3488 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3489 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3491 if (!sit_i->dirty_sentries_bitmap)
3494 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3495 sit_i->sentries[start].cur_valid_map
3496 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3497 sit_i->sentries[start].ckpt_valid_map
3498 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3499 if (!sit_i->sentries[start].cur_valid_map ||
3500 !sit_i->sentries[start].ckpt_valid_map)
3503 #ifdef CONFIG_F2FS_CHECK_FS
3504 sit_i->sentries[start].cur_valid_map_mir
3505 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3506 if (!sit_i->sentries[start].cur_valid_map_mir)
3510 if (f2fs_discard_en(sbi)) {
3511 sit_i->sentries[start].discard_map
3512 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3514 if (!sit_i->sentries[start].discard_map)
3519 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3520 if (!sit_i->tmp_map)
3523 if (sbi->segs_per_sec > 1) {
3524 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
3525 sizeof(struct sec_entry), GFP_KERNEL);
3526 if (!sit_i->sec_entries)
3530 /* get information related with SIT */
3531 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3533 /* setup SIT bitmap from ckeckpoint pack */
3534 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3535 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3537 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3538 if (!sit_i->sit_bitmap)
3541 #ifdef CONFIG_F2FS_CHECK_FS
3542 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3543 if (!sit_i->sit_bitmap_mir)
3547 /* init SIT information */
3548 sit_i->s_ops = &default_salloc_ops;
3550 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3551 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3552 sit_i->written_valid_blocks = 0;
3553 sit_i->bitmap_size = bitmap_size;
3554 sit_i->dirty_sentries = 0;
3555 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3556 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3557 sit_i->mounted_time = ktime_get_real_seconds();
3558 init_rwsem(&sit_i->sentry_lock);
3562 static int build_free_segmap(struct f2fs_sb_info *sbi)
3564 struct free_segmap_info *free_i;
3565 unsigned int bitmap_size, sec_bitmap_size;
3567 /* allocate memory for free segmap information */
3568 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3572 SM_I(sbi)->free_info = free_i;
3574 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3575 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3576 if (!free_i->free_segmap)
3579 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3580 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3581 if (!free_i->free_secmap)
3584 /* set all segments as dirty temporarily */
3585 memset(free_i->free_segmap, 0xff, bitmap_size);
3586 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3588 /* init free segmap information */
3589 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3590 free_i->free_segments = 0;
3591 free_i->free_sections = 0;
3592 spin_lock_init(&free_i->segmap_lock);
3596 static int build_curseg(struct f2fs_sb_info *sbi)
3598 struct curseg_info *array;
3601 array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
3605 SM_I(sbi)->curseg_array = array;
3607 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3608 mutex_init(&array[i].curseg_mutex);
3609 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3610 if (!array[i].sum_blk)
3612 init_rwsem(&array[i].journal_rwsem);
3613 array[i].journal = f2fs_kzalloc(sbi,
3614 sizeof(struct f2fs_journal), GFP_KERNEL);
3615 if (!array[i].journal)
3617 array[i].segno = NULL_SEGNO;
3618 array[i].next_blkoff = 0;
3620 return restore_curseg_summaries(sbi);
3623 static int build_sit_entries(struct f2fs_sb_info *sbi)
3625 struct sit_info *sit_i = SIT_I(sbi);
3626 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3627 struct f2fs_journal *journal = curseg->journal;
3628 struct seg_entry *se;
3629 struct f2fs_sit_entry sit;
3630 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3631 unsigned int i, start, end;
3632 unsigned int readed, start_blk = 0;
3636 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3639 start = start_blk * sit_i->sents_per_block;
3640 end = (start_blk + readed) * sit_i->sents_per_block;
3642 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3643 struct f2fs_sit_block *sit_blk;
3646 se = &sit_i->sentries[start];
3647 page = get_current_sit_page(sbi, start);
3648 sit_blk = (struct f2fs_sit_block *)page_address(page);
3649 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3650 f2fs_put_page(page, 1);
3652 err = check_block_count(sbi, start, &sit);
3655 seg_info_from_raw_sit(se, &sit);
3657 /* build discard map only one time */
3658 if (f2fs_discard_en(sbi)) {
3659 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3660 memset(se->discard_map, 0xff,
3661 SIT_VBLOCK_MAP_SIZE);
3663 memcpy(se->discard_map,
3665 SIT_VBLOCK_MAP_SIZE);
3666 sbi->discard_blks +=
3667 sbi->blocks_per_seg -
3672 if (sbi->segs_per_sec > 1)
3673 get_sec_entry(sbi, start)->valid_blocks +=
3676 start_blk += readed;
3677 } while (start_blk < sit_blk_cnt);
3679 down_read(&curseg->journal_rwsem);
3680 for (i = 0; i < sits_in_cursum(journal); i++) {
3681 unsigned int old_valid_blocks;
3683 start = le32_to_cpu(segno_in_journal(journal, i));
3684 if (start >= MAIN_SEGS(sbi)) {
3685 f2fs_msg(sbi->sb, KERN_ERR,
3686 "Wrong journal entry on segno %u",
3688 set_sbi_flag(sbi, SBI_NEED_FSCK);
3693 se = &sit_i->sentries[start];
3694 sit = sit_in_journal(journal, i);
3696 old_valid_blocks = se->valid_blocks;
3698 err = check_block_count(sbi, start, &sit);
3701 seg_info_from_raw_sit(se, &sit);
3703 if (f2fs_discard_en(sbi)) {
3704 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3705 memset(se->discard_map, 0xff,
3706 SIT_VBLOCK_MAP_SIZE);
3708 memcpy(se->discard_map, se->cur_valid_map,
3709 SIT_VBLOCK_MAP_SIZE);
3710 sbi->discard_blks += old_valid_blocks -
3715 if (sbi->segs_per_sec > 1)
3716 get_sec_entry(sbi, start)->valid_blocks +=
3717 se->valid_blocks - old_valid_blocks;
3719 up_read(&curseg->journal_rwsem);
3723 static void init_free_segmap(struct f2fs_sb_info *sbi)
3728 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3729 struct seg_entry *sentry = get_seg_entry(sbi, start);
3730 if (!sentry->valid_blocks)
3731 __set_free(sbi, start);
3733 SIT_I(sbi)->written_valid_blocks +=
3734 sentry->valid_blocks;
3737 /* set use the current segments */
3738 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3739 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3740 __set_test_and_inuse(sbi, curseg_t->segno);
3744 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3746 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3747 struct free_segmap_info *free_i = FREE_I(sbi);
3748 unsigned int segno = 0, offset = 0;
3749 unsigned short valid_blocks;
3752 /* find dirty segment based on free segmap */
3753 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3754 if (segno >= MAIN_SEGS(sbi))
3757 valid_blocks = get_valid_blocks(sbi, segno, false);
3758 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3760 if (valid_blocks > sbi->blocks_per_seg) {
3761 f2fs_bug_on(sbi, 1);
3764 mutex_lock(&dirty_i->seglist_lock);
3765 __locate_dirty_segment(sbi, segno, DIRTY);
3766 mutex_unlock(&dirty_i->seglist_lock);
3770 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3772 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3773 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3775 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3776 if (!dirty_i->victim_secmap)
3781 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3783 struct dirty_seglist_info *dirty_i;
3784 unsigned int bitmap_size, i;
3786 /* allocate memory for dirty segments list information */
3787 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3792 SM_I(sbi)->dirty_info = dirty_i;
3793 mutex_init(&dirty_i->seglist_lock);
3795 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3797 for (i = 0; i < NR_DIRTY_TYPE; i++) {
3798 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3800 if (!dirty_i->dirty_segmap[i])
3804 init_dirty_segmap(sbi);
3805 return init_victim_secmap(sbi);
3809 * Update min, max modified time for cost-benefit GC algorithm
3811 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3813 struct sit_info *sit_i = SIT_I(sbi);
3816 down_write(&sit_i->sentry_lock);
3818 sit_i->min_mtime = LLONG_MAX;
3820 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3822 unsigned long long mtime = 0;
3824 for (i = 0; i < sbi->segs_per_sec; i++)
3825 mtime += get_seg_entry(sbi, segno + i)->mtime;
3827 mtime = div_u64(mtime, sbi->segs_per_sec);
3829 if (sit_i->min_mtime > mtime)
3830 sit_i->min_mtime = mtime;
3832 sit_i->max_mtime = get_mtime(sbi);
3833 up_write(&sit_i->sentry_lock);
3836 int build_segment_manager(struct f2fs_sb_info *sbi)
3838 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3839 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3840 struct f2fs_sm_info *sm_info;
3843 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3848 sbi->sm_info = sm_info;
3849 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3850 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3851 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3852 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3853 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3854 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3855 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3856 sm_info->rec_prefree_segments = sm_info->main_segments *
3857 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3858 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3859 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3861 if (!test_opt(sbi, LFS))
3862 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3863 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3864 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3865 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3866 sm_info->min_ssr_sections = reserved_sections(sbi);
3868 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3870 init_rwsem(&sm_info->curseg_lock);
3872 if (!f2fs_readonly(sbi->sb)) {
3873 err = create_flush_cmd_control(sbi);
3878 err = create_discard_cmd_control(sbi);
3882 err = build_sit_info(sbi);
3885 err = build_free_segmap(sbi);
3888 err = build_curseg(sbi);
3892 /* reinit free segmap based on SIT */
3893 err = build_sit_entries(sbi);
3897 init_free_segmap(sbi);
3898 err = build_dirty_segmap(sbi);
3902 init_min_max_mtime(sbi);
3906 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3907 enum dirty_type dirty_type)
3909 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3911 mutex_lock(&dirty_i->seglist_lock);
3912 kvfree(dirty_i->dirty_segmap[dirty_type]);
3913 dirty_i->nr_dirty[dirty_type] = 0;
3914 mutex_unlock(&dirty_i->seglist_lock);
3917 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3919 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3920 kvfree(dirty_i->victim_secmap);
3923 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3925 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3931 /* discard pre-free/dirty segments list */
3932 for (i = 0; i < NR_DIRTY_TYPE; i++)
3933 discard_dirty_segmap(sbi, i);
3935 destroy_victim_secmap(sbi);
3936 SM_I(sbi)->dirty_info = NULL;
3940 static void destroy_curseg(struct f2fs_sb_info *sbi)
3942 struct curseg_info *array = SM_I(sbi)->curseg_array;
3947 SM_I(sbi)->curseg_array = NULL;
3948 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3949 kfree(array[i].sum_blk);
3950 kfree(array[i].journal);
3955 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3957 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3960 SM_I(sbi)->free_info = NULL;
3961 kvfree(free_i->free_segmap);
3962 kvfree(free_i->free_secmap);
3966 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3968 struct sit_info *sit_i = SIT_I(sbi);
3974 if (sit_i->sentries) {
3975 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3976 kfree(sit_i->sentries[start].cur_valid_map);
3977 #ifdef CONFIG_F2FS_CHECK_FS
3978 kfree(sit_i->sentries[start].cur_valid_map_mir);
3980 kfree(sit_i->sentries[start].ckpt_valid_map);
3981 kfree(sit_i->sentries[start].discard_map);
3984 kfree(sit_i->tmp_map);
3986 kvfree(sit_i->sentries);
3987 kvfree(sit_i->sec_entries);
3988 kvfree(sit_i->dirty_sentries_bitmap);
3990 SM_I(sbi)->sit_info = NULL;
3991 kfree(sit_i->sit_bitmap);
3992 #ifdef CONFIG_F2FS_CHECK_FS
3993 kfree(sit_i->sit_bitmap_mir);
3998 void destroy_segment_manager(struct f2fs_sb_info *sbi)
4000 struct f2fs_sm_info *sm_info = SM_I(sbi);
4004 destroy_flush_cmd_control(sbi, true);
4005 destroy_discard_cmd_control(sbi);
4006 destroy_dirty_segmap(sbi);
4007 destroy_curseg(sbi);
4008 destroy_free_segmap(sbi);
4009 destroy_sit_info(sbi);
4010 sbi->sm_info = NULL;
4014 int __init create_segment_manager_caches(void)
4016 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4017 sizeof(struct discard_entry));
4018 if (!discard_entry_slab)
4021 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4022 sizeof(struct discard_cmd));
4023 if (!discard_cmd_slab)
4024 goto destroy_discard_entry;
4026 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4027 sizeof(struct sit_entry_set));
4028 if (!sit_entry_set_slab)
4029 goto destroy_discard_cmd;
4031 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4032 sizeof(struct inmem_pages));
4033 if (!inmem_entry_slab)
4034 goto destroy_sit_entry_set;
4037 destroy_sit_entry_set:
4038 kmem_cache_destroy(sit_entry_set_slab);
4039 destroy_discard_cmd:
4040 kmem_cache_destroy(discard_cmd_slab);
4041 destroy_discard_entry:
4042 kmem_cache_destroy(discard_entry_slab);
4047 void destroy_segment_manager_caches(void)
4049 kmem_cache_destroy(sit_entry_set_slab);
4050 kmem_cache_destroy(discard_cmd_slab);
4051 kmem_cache_destroy(discard_entry_slab);
4052 kmem_cache_destroy(inmem_entry_slab);