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/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/aio.h>
16 #include <linux/writeback.h>
17 #include <linux/backing-dev.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
26 #include <trace/events/f2fs.h>
28 static struct kmem_cache *extent_tree_slab;
29 static struct kmem_cache *extent_node_slab;
31 static void f2fs_read_end_io(struct bio *bio, int err)
36 bio_for_each_segment_all(bvec, bio, i) {
37 struct page *page = bvec->bv_page;
40 SetPageUptodate(page);
42 ClearPageUptodate(page);
50 static void f2fs_write_end_io(struct bio *bio, int err)
52 struct f2fs_sb_info *sbi = bio->bi_private;
56 bio_for_each_segment_all(bvec, bio, i) {
57 struct page *page = bvec->bv_page;
61 set_bit(AS_EIO, &page->mapping->flags);
62 f2fs_stop_checkpoint(sbi);
64 end_page_writeback(page);
65 dec_page_count(sbi, F2FS_WRITEBACK);
68 if (!get_pages(sbi, F2FS_WRITEBACK) &&
69 !list_empty(&sbi->cp_wait.task_list))
70 wake_up(&sbi->cp_wait);
76 * Low-level block read/write IO operations.
78 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
79 int npages, bool is_read)
83 /* No failure on bio allocation */
84 bio = bio_alloc(GFP_NOIO, npages);
86 bio->bi_bdev = sbi->sb->s_bdev;
87 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
88 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
89 bio->bi_private = sbi;
94 static void __submit_merged_bio(struct f2fs_bio_info *io)
96 struct f2fs_io_info *fio = &io->fio;
101 if (is_read_io(fio->rw))
102 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
104 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
106 submit_bio(fio->rw, io->bio);
110 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
111 enum page_type type, int rw)
113 enum page_type btype = PAGE_TYPE_OF_BIO(type);
114 struct f2fs_bio_info *io;
116 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
118 down_write(&io->io_rwsem);
120 /* change META to META_FLUSH in the checkpoint procedure */
121 if (type >= META_FLUSH) {
122 io->fio.type = META_FLUSH;
123 if (test_opt(sbi, NOBARRIER))
124 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
126 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
128 __submit_merged_bio(io);
129 up_write(&io->io_rwsem);
133 * Fill the locked page with data located in the block address.
134 * Return unlocked page.
136 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
137 struct f2fs_io_info *fio)
141 trace_f2fs_submit_page_bio(page, fio);
142 f2fs_trace_ios(page, fio, 0);
144 /* Allocate a new bio */
145 bio = __bio_alloc(sbi, fio->blk_addr, 1, is_read_io(fio->rw));
147 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
149 f2fs_put_page(page, 1);
153 submit_bio(fio->rw, bio);
157 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
158 struct f2fs_io_info *fio)
160 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
161 struct f2fs_bio_info *io;
162 bool is_read = is_read_io(fio->rw);
164 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
166 verify_block_addr(sbi, fio->blk_addr);
168 down_write(&io->io_rwsem);
171 inc_page_count(sbi, F2FS_WRITEBACK);
173 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
174 io->fio.rw != fio->rw))
175 __submit_merged_bio(io);
177 if (io->bio == NULL) {
178 int bio_blocks = MAX_BIO_BLOCKS(sbi);
180 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
184 if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
186 __submit_merged_bio(io);
190 io->last_block_in_bio = fio->blk_addr;
191 f2fs_trace_ios(page, fio, 0);
193 up_write(&io->io_rwsem);
194 trace_f2fs_submit_page_mbio(page, fio);
198 * Lock ordering for the change of data block address:
201 * update block addresses in the node page
203 static void __set_data_blkaddr(struct dnode_of_data *dn)
205 struct f2fs_node *rn;
207 struct page *node_page = dn->node_page;
208 unsigned int ofs_in_node = dn->ofs_in_node;
210 f2fs_wait_on_page_writeback(node_page, NODE);
212 rn = F2FS_NODE(node_page);
214 /* Get physical address of data block */
215 addr_array = blkaddr_in_node(rn);
216 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
217 set_page_dirty(node_page);
220 int reserve_new_block(struct dnode_of_data *dn)
222 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
224 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
226 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
229 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
231 dn->data_blkaddr = NEW_ADDR;
232 __set_data_blkaddr(dn);
233 mark_inode_dirty(dn->inode);
238 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
240 bool need_put = dn->inode_page ? false : true;
243 err = get_dnode_of_data(dn, index, ALLOC_NODE);
247 if (dn->data_blkaddr == NULL_ADDR)
248 err = reserve_new_block(dn);
254 static void f2fs_map_bh(struct super_block *sb, pgoff_t pgofs,
255 struct extent_info *ei, struct buffer_head *bh_result)
257 unsigned int blkbits = sb->s_blocksize_bits;
260 set_buffer_new(bh_result);
261 map_bh(bh_result, sb, ei->blk + pgofs - ei->fofs);
262 count = ei->fofs + ei->len - pgofs;
263 if (count < (UINT_MAX >> blkbits))
264 bh_result->b_size = (count << blkbits);
266 bh_result->b_size = UINT_MAX;
269 static bool lookup_extent_info(struct inode *inode, pgoff_t pgofs,
270 struct extent_info *ei)
272 struct f2fs_inode_info *fi = F2FS_I(inode);
273 pgoff_t start_fofs, end_fofs;
274 block_t start_blkaddr;
276 read_lock(&fi->ext_lock);
277 if (fi->ext.len == 0) {
278 read_unlock(&fi->ext_lock);
282 stat_inc_total_hit(inode->i_sb);
284 start_fofs = fi->ext.fofs;
285 end_fofs = fi->ext.fofs + fi->ext.len - 1;
286 start_blkaddr = fi->ext.blk;
288 if (pgofs >= start_fofs && pgofs <= end_fofs) {
290 stat_inc_read_hit(inode->i_sb);
291 read_unlock(&fi->ext_lock);
294 read_unlock(&fi->ext_lock);
298 static bool update_extent_info(struct inode *inode, pgoff_t fofs,
301 struct f2fs_inode_info *fi = F2FS_I(inode);
302 pgoff_t start_fofs, end_fofs;
303 block_t start_blkaddr, end_blkaddr;
304 int need_update = true;
306 write_lock(&fi->ext_lock);
308 start_fofs = fi->ext.fofs;
309 end_fofs = fi->ext.fofs + fi->ext.len - 1;
310 start_blkaddr = fi->ext.blk;
311 end_blkaddr = fi->ext.blk + fi->ext.len - 1;
313 /* Drop and initialize the matched extent */
314 if (fi->ext.len == 1 && fofs == start_fofs)
318 if (fi->ext.len == 0) {
319 if (blkaddr != NULL_ADDR) {
321 fi->ext.blk = blkaddr;
328 if (fofs == start_fofs - 1 && blkaddr == start_blkaddr - 1) {
336 if (fofs == end_fofs + 1 && blkaddr == end_blkaddr + 1) {
341 /* Split the existing extent */
342 if (fi->ext.len > 1 &&
343 fofs >= start_fofs && fofs <= end_fofs) {
344 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
345 fi->ext.len = fofs - start_fofs;
347 fi->ext.fofs = fofs + 1;
348 fi->ext.blk = start_blkaddr + fofs - start_fofs + 1;
349 fi->ext.len -= fofs - start_fofs + 1;
355 /* Finally, if the extent is very fragmented, let's drop the cache. */
356 if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
358 set_inode_flag(fi, FI_NO_EXTENT);
362 write_unlock(&fi->ext_lock);
366 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
367 struct extent_tree *et, struct extent_info *ei,
368 struct rb_node *parent, struct rb_node **p)
370 struct extent_node *en;
372 en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
377 INIT_LIST_HEAD(&en->list);
379 rb_link_node(&en->rb_node, parent, p);
380 rb_insert_color(&en->rb_node, &et->root);
382 atomic_inc(&sbi->total_ext_node);
386 static void __detach_extent_node(struct f2fs_sb_info *sbi,
387 struct extent_tree *et, struct extent_node *en)
389 rb_erase(&en->rb_node, &et->root);
391 atomic_dec(&sbi->total_ext_node);
393 if (et->cached_en == en)
394 et->cached_en = NULL;
397 static struct extent_node *__lookup_extent_tree(struct extent_tree *et,
400 struct rb_node *node = et->root.rb_node;
401 struct extent_node *en;
404 struct extent_info *cei = &et->cached_en->ei;
406 if (cei->fofs <= fofs && cei->fofs + cei->len > fofs)
407 return et->cached_en;
411 en = rb_entry(node, struct extent_node, rb_node);
413 if (fofs < en->ei.fofs) {
414 node = node->rb_left;
415 } else if (fofs >= en->ei.fofs + en->ei.len) {
416 node = node->rb_right;
425 static struct extent_node *__try_back_merge(struct f2fs_sb_info *sbi,
426 struct extent_tree *et, struct extent_node *en)
428 struct extent_node *prev;
429 struct rb_node *node;
431 node = rb_prev(&en->rb_node);
435 prev = rb_entry(node, struct extent_node, rb_node);
436 if (__is_back_mergeable(&en->ei, &prev->ei)) {
437 en->ei.fofs = prev->ei.fofs;
438 en->ei.blk = prev->ei.blk;
439 en->ei.len += prev->ei.len;
440 __detach_extent_node(sbi, et, prev);
446 static struct extent_node *__try_front_merge(struct f2fs_sb_info *sbi,
447 struct extent_tree *et, struct extent_node *en)
449 struct extent_node *next;
450 struct rb_node *node;
452 node = rb_next(&en->rb_node);
456 next = rb_entry(node, struct extent_node, rb_node);
457 if (__is_front_mergeable(&en->ei, &next->ei)) {
458 en->ei.len += next->ei.len;
459 __detach_extent_node(sbi, et, next);
465 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
466 struct extent_tree *et, struct extent_info *ei,
467 struct extent_node **den)
469 struct rb_node **p = &et->root.rb_node;
470 struct rb_node *parent = NULL;
471 struct extent_node *en;
475 en = rb_entry(parent, struct extent_node, rb_node);
477 if (ei->fofs < en->ei.fofs) {
478 if (__is_front_mergeable(ei, &en->ei)) {
479 f2fs_bug_on(sbi, !den);
480 en->ei.fofs = ei->fofs;
481 en->ei.blk = ei->blk;
482 en->ei.len += ei->len;
483 *den = __try_back_merge(sbi, et, en);
487 } else if (ei->fofs >= en->ei.fofs + en->ei.len) {
488 if (__is_back_mergeable(ei, &en->ei)) {
489 f2fs_bug_on(sbi, !den);
490 en->ei.len += ei->len;
491 *den = __try_front_merge(sbi, et, en);
500 return __attach_extent_node(sbi, et, ei, parent, p);
503 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
504 struct extent_tree *et, bool free_all)
506 struct rb_node *node, *next;
507 struct extent_node *en;
508 unsigned int count = et->count;
510 node = rb_first(&et->root);
512 next = rb_next(node);
513 en = rb_entry(node, struct extent_node, rb_node);
516 spin_lock(&sbi->extent_lock);
517 if (!list_empty(&en->list))
518 list_del_init(&en->list);
519 spin_unlock(&sbi->extent_lock);
522 if (free_all || list_empty(&en->list)) {
523 __detach_extent_node(sbi, et, en);
524 kmem_cache_free(extent_node_slab, en);
529 return count - et->count;
532 static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
533 struct extent_info *ei)
535 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
536 struct extent_tree *et;
537 struct extent_node *en;
539 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
541 down_read(&sbi->extent_tree_lock);
542 et = radix_tree_lookup(&sbi->extent_tree_root, inode->i_ino);
544 up_read(&sbi->extent_tree_lock);
547 atomic_inc(&et->refcount);
548 up_read(&sbi->extent_tree_lock);
550 read_lock(&et->lock);
551 en = __lookup_extent_tree(et, pgofs);
554 spin_lock(&sbi->extent_lock);
555 if (!list_empty(&en->list))
556 list_move_tail(&en->list, &sbi->extent_list);
557 spin_unlock(&sbi->extent_lock);
558 stat_inc_read_hit(sbi->sb);
560 stat_inc_total_hit(sbi->sb);
561 read_unlock(&et->lock);
563 trace_f2fs_lookup_extent_tree_end(inode, pgofs, en);
565 atomic_dec(&et->refcount);
566 return en ? true : false;
569 static void f2fs_update_extent_tree(struct inode *inode, pgoff_t fofs,
572 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
573 nid_t ino = inode->i_ino;
574 struct extent_tree *et;
575 struct extent_node *en = NULL, *en1 = NULL, *en2 = NULL, *en3 = NULL;
576 struct extent_node *den = NULL;
577 struct extent_info ei, dei;
580 trace_f2fs_update_extent_tree(inode, fofs, blkaddr);
582 down_write(&sbi->extent_tree_lock);
583 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
585 et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
586 f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
587 memset(et, 0, sizeof(struct extent_tree));
590 et->cached_en = NULL;
591 rwlock_init(&et->lock);
592 atomic_set(&et->refcount, 0);
594 sbi->total_ext_tree++;
596 atomic_inc(&et->refcount);
597 up_write(&sbi->extent_tree_lock);
599 write_lock(&et->lock);
601 /* 1. lookup and remove existing extent info in cache */
602 en = __lookup_extent_tree(et, fofs);
607 __detach_extent_node(sbi, et, en);
609 /* 2. if extent can be split more, split and insert the left part */
611 /* insert left part of split extent into cache */
612 if (fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
613 set_extent_info(&ei, dei.fofs, dei.blk,
615 en1 = __insert_extent_tree(sbi, et, &ei, NULL);
618 /* insert right part of split extent into cache */
619 endofs = dei.fofs + dei.len - 1;
620 if (endofs - fofs >= F2FS_MIN_EXTENT_LEN) {
621 set_extent_info(&ei, fofs + 1,
622 fofs - dei.fofs + dei.blk, endofs - fofs);
623 en2 = __insert_extent_tree(sbi, et, &ei, NULL);
628 /* 3. update extent in extent cache */
630 set_extent_info(&ei, fofs, blkaddr, 1);
631 en3 = __insert_extent_tree(sbi, et, &ei, &den);
634 /* 4. update in global extent list */
635 spin_lock(&sbi->extent_lock);
636 if (en && !list_empty(&en->list))
639 * en1 and en2 split from en, they will become more and more smaller
640 * fragments after splitting several times. So if the length is smaller
641 * than F2FS_MIN_EXTENT_LEN, we will not add them into extent tree.
644 list_add_tail(&en1->list, &sbi->extent_list);
646 list_add_tail(&en2->list, &sbi->extent_list);
648 if (list_empty(&en3->list))
649 list_add_tail(&en3->list, &sbi->extent_list);
651 list_move_tail(&en3->list, &sbi->extent_list);
653 if (den && !list_empty(&den->list))
654 list_del(&den->list);
655 spin_unlock(&sbi->extent_lock);
657 /* 5. release extent node */
659 kmem_cache_free(extent_node_slab, en);
661 kmem_cache_free(extent_node_slab, den);
663 write_unlock(&et->lock);
664 atomic_dec(&et->refcount);
667 void f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
669 struct extent_tree *treevec[EXT_TREE_VEC_SIZE];
670 struct extent_node *en, *tmp;
671 unsigned long ino = F2FS_ROOT_INO(sbi);
672 struct radix_tree_iter iter;
675 unsigned int node_cnt = 0, tree_cnt = 0;
677 if (!test_opt(sbi, EXTENT_CACHE))
680 if (available_free_memory(sbi, EXTENT_CACHE))
683 spin_lock(&sbi->extent_lock);
684 list_for_each_entry_safe(en, tmp, &sbi->extent_list, list) {
687 list_del_init(&en->list);
689 spin_unlock(&sbi->extent_lock);
691 down_read(&sbi->extent_tree_lock);
692 while ((found = radix_tree_gang_lookup(&sbi->extent_tree_root,
693 (void **)treevec, ino, EXT_TREE_VEC_SIZE))) {
696 ino = treevec[found - 1]->ino + 1;
697 for (i = 0; i < found; i++) {
698 struct extent_tree *et = treevec[i];
700 atomic_inc(&et->refcount);
701 write_lock(&et->lock);
702 node_cnt += __free_extent_tree(sbi, et, false);
703 write_unlock(&et->lock);
704 atomic_dec(&et->refcount);
707 up_read(&sbi->extent_tree_lock);
709 down_write(&sbi->extent_tree_lock);
710 radix_tree_for_each_slot(slot, &sbi->extent_tree_root, &iter,
711 F2FS_ROOT_INO(sbi)) {
712 struct extent_tree *et = (struct extent_tree *)*slot;
714 if (!atomic_read(&et->refcount) && !et->count) {
715 radix_tree_delete(&sbi->extent_tree_root, et->ino);
716 kmem_cache_free(extent_tree_slab, et);
717 sbi->total_ext_tree--;
721 up_write(&sbi->extent_tree_lock);
723 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
726 void f2fs_destroy_extent_tree(struct inode *inode)
728 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
729 struct extent_tree *et;
730 unsigned int node_cnt = 0;
732 if (!test_opt(sbi, EXTENT_CACHE))
735 down_read(&sbi->extent_tree_lock);
736 et = radix_tree_lookup(&sbi->extent_tree_root, inode->i_ino);
738 up_read(&sbi->extent_tree_lock);
741 atomic_inc(&et->refcount);
742 up_read(&sbi->extent_tree_lock);
744 /* free all extent info belong to this extent tree */
745 write_lock(&et->lock);
746 node_cnt = __free_extent_tree(sbi, et, true);
747 write_unlock(&et->lock);
749 atomic_dec(&et->refcount);
751 /* try to find and delete extent tree entry in radix tree */
752 down_write(&sbi->extent_tree_lock);
753 et = radix_tree_lookup(&sbi->extent_tree_root, inode->i_ino);
755 up_write(&sbi->extent_tree_lock);
758 f2fs_bug_on(sbi, atomic_read(&et->refcount) || et->count);
759 radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
760 kmem_cache_free(extent_tree_slab, et);
761 sbi->total_ext_tree--;
762 up_write(&sbi->extent_tree_lock);
764 trace_f2fs_destroy_extent_tree(inode, node_cnt);
768 static bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
769 struct extent_info *ei)
771 if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT))
774 if (test_opt(F2FS_I_SB(inode), EXTENT_CACHE))
775 return f2fs_lookup_extent_tree(inode, pgofs, ei);
777 return lookup_extent_info(inode, pgofs, ei);
780 void f2fs_update_extent_cache(struct dnode_of_data *dn)
782 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
785 f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR);
787 /* Update the page address in the parent node */
788 __set_data_blkaddr(dn);
790 if (is_inode_flag_set(fi, FI_NO_EXTENT))
793 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
796 if (test_opt(F2FS_I_SB(dn->inode), EXTENT_CACHE))
797 return f2fs_update_extent_tree(dn->inode, fofs,
800 if (update_extent_info(dn->inode, fofs, dn->data_blkaddr))
804 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
806 struct address_space *mapping = inode->i_mapping;
807 struct dnode_of_data dn;
809 struct extent_info ei;
811 struct f2fs_io_info fio = {
813 .rw = sync ? READ_SYNC : READA,
817 * If sync is false, it needs to check its block allocation.
818 * This is need and triggered by two flows:
819 * gc and truncate_partial_data_page.
824 page = find_get_page(mapping, index);
825 if (page && PageUptodate(page))
827 f2fs_put_page(page, 0);
829 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
830 dn.data_blkaddr = ei.blk + index - ei.fofs;
834 set_new_dnode(&dn, inode, NULL, NULL, 0);
835 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
840 if (dn.data_blkaddr == NULL_ADDR)
841 return ERR_PTR(-ENOENT);
843 /* By fallocate(), there is no cached page, but with NEW_ADDR */
844 if (unlikely(dn.data_blkaddr == NEW_ADDR))
845 return ERR_PTR(-EINVAL);
848 page = grab_cache_page(mapping, index);
850 return ERR_PTR(-ENOMEM);
852 if (PageUptodate(page)) {
857 fio.blk_addr = dn.data_blkaddr;
858 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
863 wait_on_page_locked(page);
864 if (unlikely(!PageUptodate(page))) {
865 f2fs_put_page(page, 0);
866 return ERR_PTR(-EIO);
873 * If it tries to access a hole, return an error.
874 * Because, the callers, functions in dir.c and GC, should be able to know
875 * whether this page exists or not.
877 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
879 struct address_space *mapping = inode->i_mapping;
880 struct dnode_of_data dn;
882 struct extent_info ei;
884 struct f2fs_io_info fio = {
889 page = grab_cache_page(mapping, index);
891 return ERR_PTR(-ENOMEM);
893 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
894 dn.data_blkaddr = ei.blk + index - ei.fofs;
898 set_new_dnode(&dn, inode, NULL, NULL, 0);
899 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
901 f2fs_put_page(page, 1);
906 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
907 f2fs_put_page(page, 1);
908 return ERR_PTR(-ENOENT);
912 if (PageUptodate(page))
916 * A new dentry page is allocated but not able to be written, since its
917 * new inode page couldn't be allocated due to -ENOSPC.
918 * In such the case, its blkaddr can be remained as NEW_ADDR.
919 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
921 if (dn.data_blkaddr == NEW_ADDR) {
922 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
923 SetPageUptodate(page);
927 fio.blk_addr = dn.data_blkaddr;
928 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
933 if (unlikely(!PageUptodate(page))) {
934 f2fs_put_page(page, 1);
935 return ERR_PTR(-EIO);
937 if (unlikely(page->mapping != mapping)) {
938 f2fs_put_page(page, 1);
945 * Caller ensures that this data page is never allocated.
946 * A new zero-filled data page is allocated in the page cache.
948 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
950 * Note that, ipage is set only by make_empty_dir.
952 struct page *get_new_data_page(struct inode *inode,
953 struct page *ipage, pgoff_t index, bool new_i_size)
955 struct address_space *mapping = inode->i_mapping;
957 struct dnode_of_data dn;
960 set_new_dnode(&dn, inode, ipage, NULL, 0);
961 err = f2fs_reserve_block(&dn, index);
965 page = grab_cache_page(mapping, index);
971 if (PageUptodate(page))
974 if (dn.data_blkaddr == NEW_ADDR) {
975 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
976 SetPageUptodate(page);
978 struct f2fs_io_info fio = {
981 .blk_addr = dn.data_blkaddr,
983 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
988 if (unlikely(!PageUptodate(page))) {
989 f2fs_put_page(page, 1);
993 if (unlikely(page->mapping != mapping)) {
994 f2fs_put_page(page, 1);
1000 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
1001 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
1002 /* Only the directory inode sets new_i_size */
1003 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
1008 f2fs_put_dnode(&dn);
1009 return ERR_PTR(err);
1012 static int __allocate_data_block(struct dnode_of_data *dn)
1014 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1015 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
1016 struct f2fs_summary sum;
1017 struct node_info ni;
1018 int seg = CURSEG_WARM_DATA;
1021 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
1023 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
1026 get_node_info(sbi, dn->nid, &ni);
1027 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1029 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
1030 seg = CURSEG_DIRECT_IO;
1032 allocate_data_block(sbi, NULL, NULL_ADDR, &dn->data_blkaddr, &sum, seg);
1034 /* direct IO doesn't use extent cache to maximize the performance */
1035 __set_data_blkaddr(dn);
1038 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
1040 if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
1041 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
1046 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
1049 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1050 struct dnode_of_data dn;
1051 u64 start = F2FS_BYTES_TO_BLK(offset);
1052 u64 len = F2FS_BYTES_TO_BLK(count);
1057 f2fs_balance_fs(sbi);
1060 /* When reading holes, we need its node page */
1061 set_new_dnode(&dn, inode, NULL, NULL, 0);
1062 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
1066 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1068 while (dn.ofs_in_node < end_offset && len) {
1069 if (dn.data_blkaddr == NULL_ADDR) {
1070 if (__allocate_data_block(&dn))
1080 sync_inode_page(&dn);
1082 f2fs_put_dnode(&dn);
1083 f2fs_unlock_op(sbi);
1089 sync_inode_page(&dn);
1090 f2fs_put_dnode(&dn);
1092 f2fs_unlock_op(sbi);
1097 * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
1098 * If original data blocks are allocated, then give them to blockdev.
1100 * a. preallocate requested block addresses
1101 * b. do not use extent cache for better performance
1102 * c. give the block addresses to blockdev
1104 static int __get_data_block(struct inode *inode, sector_t iblock,
1105 struct buffer_head *bh_result, int create, bool fiemap)
1107 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
1108 unsigned maxblocks = bh_result->b_size >> blkbits;
1109 struct dnode_of_data dn;
1110 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
1111 pgoff_t pgofs, end_offset;
1112 int err = 0, ofs = 1;
1113 struct extent_info ei;
1114 bool allocated = false;
1116 /* Get the page offset from the block offset(iblock) */
1117 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
1119 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1120 f2fs_map_bh(inode->i_sb, pgofs, &ei, bh_result);
1125 f2fs_lock_op(F2FS_I_SB(inode));
1127 /* When reading holes, we need its node page */
1128 set_new_dnode(&dn, inode, NULL, NULL, 0);
1129 err = get_dnode_of_data(&dn, pgofs, mode);
1135 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
1138 if (dn.data_blkaddr != NULL_ADDR) {
1139 set_buffer_new(bh_result);
1140 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
1141 } else if (create) {
1142 err = __allocate_data_block(&dn);
1146 set_buffer_new(bh_result);
1147 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
1152 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1153 bh_result->b_size = (((size_t)1) << blkbits);
1158 if (dn.ofs_in_node >= end_offset) {
1160 sync_inode_page(&dn);
1162 f2fs_put_dnode(&dn);
1164 set_new_dnode(&dn, inode, NULL, NULL, 0);
1165 err = get_dnode_of_data(&dn, pgofs, mode);
1171 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
1174 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1177 if (maxblocks > (bh_result->b_size >> blkbits)) {
1178 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
1179 if (blkaddr == NULL_ADDR && create) {
1180 err = __allocate_data_block(&dn);
1184 blkaddr = dn.data_blkaddr;
1186 /* Give more consecutive addresses for the readahead */
1187 if (blkaddr == (bh_result->b_blocknr + ofs)) {
1191 bh_result->b_size += (((size_t)1) << blkbits);
1197 sync_inode_page(&dn);
1199 f2fs_put_dnode(&dn);
1202 f2fs_unlock_op(F2FS_I_SB(inode));
1204 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
1208 static int get_data_block(struct inode *inode, sector_t iblock,
1209 struct buffer_head *bh_result, int create)
1211 return __get_data_block(inode, iblock, bh_result, create, false);
1214 static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
1215 struct buffer_head *bh_result, int create)
1217 return __get_data_block(inode, iblock, bh_result, create, true);
1220 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1223 return generic_block_fiemap(inode, fieinfo,
1224 start, len, get_data_block_fiemap);
1227 static int f2fs_read_data_page(struct file *file, struct page *page)
1229 struct inode *inode = page->mapping->host;
1232 trace_f2fs_readpage(page, DATA);
1234 /* If the file has inline data, try to read it directly */
1235 if (f2fs_has_inline_data(inode))
1236 ret = f2fs_read_inline_data(inode, page);
1238 ret = mpage_readpage(page, get_data_block);
1243 static int f2fs_read_data_pages(struct file *file,
1244 struct address_space *mapping,
1245 struct list_head *pages, unsigned nr_pages)
1247 struct inode *inode = file->f_mapping->host;
1249 /* If the file has inline data, skip readpages */
1250 if (f2fs_has_inline_data(inode))
1253 return mpage_readpages(mapping, pages, nr_pages, get_data_block);
1256 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
1258 struct inode *inode = page->mapping->host;
1259 struct dnode_of_data dn;
1262 set_new_dnode(&dn, inode, NULL, NULL, 0);
1263 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1267 fio->blk_addr = dn.data_blkaddr;
1269 /* This page is already truncated */
1270 if (fio->blk_addr == NULL_ADDR) {
1271 ClearPageUptodate(page);
1275 set_page_writeback(page);
1278 * If current allocation needs SSR,
1279 * it had better in-place writes for updated data.
1281 if (unlikely(fio->blk_addr != NEW_ADDR &&
1282 !is_cold_data(page) &&
1283 need_inplace_update(inode))) {
1284 rewrite_data_page(page, fio);
1285 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1287 write_data_page(page, &dn, fio);
1288 f2fs_update_extent_cache(&dn);
1289 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1292 f2fs_put_dnode(&dn);
1296 static int f2fs_write_data_page(struct page *page,
1297 struct writeback_control *wbc)
1299 struct inode *inode = page->mapping->host;
1300 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1301 loff_t i_size = i_size_read(inode);
1302 const pgoff_t end_index = ((unsigned long long) i_size)
1303 >> PAGE_CACHE_SHIFT;
1304 unsigned offset = 0;
1305 bool need_balance_fs = false;
1307 struct f2fs_io_info fio = {
1309 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1312 trace_f2fs_writepage(page, DATA);
1314 if (page->index < end_index)
1318 * If the offset is out-of-range of file size,
1319 * this page does not have to be written to disk.
1321 offset = i_size & (PAGE_CACHE_SIZE - 1);
1322 if ((page->index >= end_index + 1) || !offset)
1325 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1327 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1329 if (f2fs_is_drop_cache(inode))
1331 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1332 available_free_memory(sbi, BASE_CHECK))
1335 /* Dentry blocks are controlled by checkpoint */
1336 if (S_ISDIR(inode->i_mode)) {
1337 if (unlikely(f2fs_cp_error(sbi)))
1339 err = do_write_data_page(page, &fio);
1343 /* we should bypass data pages to proceed the kworkder jobs */
1344 if (unlikely(f2fs_cp_error(sbi))) {
1349 if (!wbc->for_reclaim)
1350 need_balance_fs = true;
1351 else if (has_not_enough_free_secs(sbi, 0))
1356 if (f2fs_has_inline_data(inode))
1357 err = f2fs_write_inline_data(inode, page);
1359 err = do_write_data_page(page, &fio);
1360 f2fs_unlock_op(sbi);
1362 if (err && err != -ENOENT)
1365 clear_cold_data(page);
1367 inode_dec_dirty_pages(inode);
1369 ClearPageUptodate(page);
1371 if (need_balance_fs)
1372 f2fs_balance_fs(sbi);
1373 if (wbc->for_reclaim)
1374 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1378 redirty_page_for_writepage(wbc, page);
1379 return AOP_WRITEPAGE_ACTIVATE;
1382 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1385 struct address_space *mapping = data;
1386 int ret = mapping->a_ops->writepage(page, wbc);
1387 mapping_set_error(mapping, ret);
1391 static int f2fs_write_data_pages(struct address_space *mapping,
1392 struct writeback_control *wbc)
1394 struct inode *inode = mapping->host;
1395 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1396 bool locked = false;
1400 trace_f2fs_writepages(mapping->host, wbc, DATA);
1402 /* deal with chardevs and other special file */
1403 if (!mapping->a_ops->writepage)
1406 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1407 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1408 available_free_memory(sbi, DIRTY_DENTS))
1411 diff = nr_pages_to_write(sbi, DATA, wbc);
1413 if (!S_ISDIR(inode->i_mode)) {
1414 mutex_lock(&sbi->writepages);
1417 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1419 mutex_unlock(&sbi->writepages);
1421 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1423 remove_dirty_dir_inode(inode);
1425 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1429 wbc->pages_skipped += get_dirty_pages(inode);
1433 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1435 struct inode *inode = mapping->host;
1437 if (to > inode->i_size) {
1438 truncate_pagecache(inode, inode->i_size);
1439 truncate_blocks(inode, inode->i_size, true);
1443 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1444 loff_t pos, unsigned len, unsigned flags,
1445 struct page **pagep, void **fsdata)
1447 struct inode *inode = mapping->host;
1448 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1449 struct page *page, *ipage;
1450 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1451 struct dnode_of_data dn;
1454 trace_f2fs_write_begin(inode, pos, len, flags);
1456 f2fs_balance_fs(sbi);
1459 * We should check this at this moment to avoid deadlock on inode page
1460 * and #0 page. The locking rule for inline_data conversion should be:
1461 * lock_page(page #0) -> lock_page(inode_page)
1464 err = f2fs_convert_inline_inode(inode);
1469 page = grab_cache_page_write_begin(mapping, index, flags);
1479 /* check inline_data */
1480 ipage = get_node_page(sbi, inode->i_ino);
1481 if (IS_ERR(ipage)) {
1482 err = PTR_ERR(ipage);
1486 set_new_dnode(&dn, inode, ipage, ipage, 0);
1488 if (f2fs_has_inline_data(inode)) {
1489 if (pos + len <= MAX_INLINE_DATA) {
1490 read_inline_data(page, ipage);
1491 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1492 sync_inode_page(&dn);
1495 err = f2fs_convert_inline_page(&dn, page);
1499 err = f2fs_reserve_block(&dn, index);
1503 f2fs_put_dnode(&dn);
1504 f2fs_unlock_op(sbi);
1506 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
1509 f2fs_wait_on_page_writeback(page, DATA);
1511 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1512 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1513 unsigned end = start + len;
1515 /* Reading beyond i_size is simple: memset to zero */
1516 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1520 if (dn.data_blkaddr == NEW_ADDR) {
1521 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1523 struct f2fs_io_info fio = {
1526 .blk_addr = dn.data_blkaddr,
1528 err = f2fs_submit_page_bio(sbi, page, &fio);
1533 if (unlikely(!PageUptodate(page))) {
1534 f2fs_put_page(page, 1);
1538 if (unlikely(page->mapping != mapping)) {
1539 f2fs_put_page(page, 1);
1544 SetPageUptodate(page);
1545 clear_cold_data(page);
1549 f2fs_put_dnode(&dn);
1551 f2fs_unlock_op(sbi);
1552 f2fs_put_page(page, 1);
1554 f2fs_write_failed(mapping, pos + len);
1558 static int f2fs_write_end(struct file *file,
1559 struct address_space *mapping,
1560 loff_t pos, unsigned len, unsigned copied,
1561 struct page *page, void *fsdata)
1563 struct inode *inode = page->mapping->host;
1565 trace_f2fs_write_end(inode, pos, len, copied);
1567 set_page_dirty(page);
1569 if (pos + copied > i_size_read(inode)) {
1570 i_size_write(inode, pos + copied);
1571 mark_inode_dirty(inode);
1572 update_inode_page(inode);
1575 f2fs_put_page(page, 1);
1579 static int check_direct_IO(struct inode *inode, int rw,
1580 struct iov_iter *iter, loff_t offset)
1582 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1587 if (offset & blocksize_mask)
1590 if (iov_iter_alignment(iter) & blocksize_mask)
1596 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1597 struct iov_iter *iter, loff_t offset)
1599 struct file *file = iocb->ki_filp;
1600 struct address_space *mapping = file->f_mapping;
1601 struct inode *inode = mapping->host;
1602 size_t count = iov_iter_count(iter);
1605 /* we don't need to use inline_data strictly */
1606 if (f2fs_has_inline_data(inode)) {
1607 err = f2fs_convert_inline_inode(inode);
1612 if (check_direct_IO(inode, rw, iter, offset))
1615 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1618 __allocate_data_blocks(inode, offset, count);
1620 err = blockdev_direct_IO(rw, iocb, inode, iter, offset, get_data_block);
1621 if (err < 0 && (rw & WRITE))
1622 f2fs_write_failed(mapping, offset + count);
1624 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1629 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1630 unsigned int length)
1632 struct inode *inode = page->mapping->host;
1633 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1635 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1636 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1639 if (PageDirty(page)) {
1640 if (inode->i_ino == F2FS_META_INO(sbi))
1641 dec_page_count(sbi, F2FS_DIRTY_META);
1642 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1643 dec_page_count(sbi, F2FS_DIRTY_NODES);
1645 inode_dec_dirty_pages(inode);
1647 ClearPagePrivate(page);
1650 int f2fs_release_page(struct page *page, gfp_t wait)
1652 /* If this is dirty page, keep PagePrivate */
1653 if (PageDirty(page))
1656 ClearPagePrivate(page);
1660 static int f2fs_set_data_page_dirty(struct page *page)
1662 struct address_space *mapping = page->mapping;
1663 struct inode *inode = mapping->host;
1665 trace_f2fs_set_page_dirty(page, DATA);
1667 SetPageUptodate(page);
1669 if (f2fs_is_atomic_file(inode)) {
1670 register_inmem_page(inode, page);
1674 mark_inode_dirty(inode);
1676 if (!PageDirty(page)) {
1677 __set_page_dirty_nobuffers(page);
1678 update_dirty_page(inode, page);
1684 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1686 struct inode *inode = mapping->host;
1688 /* we don't need to use inline_data strictly */
1689 if (f2fs_has_inline_data(inode)) {
1690 int err = f2fs_convert_inline_inode(inode);
1694 return generic_block_bmap(mapping, block, get_data_block);
1697 void init_extent_cache_info(struct f2fs_sb_info *sbi)
1699 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
1700 init_rwsem(&sbi->extent_tree_lock);
1701 INIT_LIST_HEAD(&sbi->extent_list);
1702 spin_lock_init(&sbi->extent_lock);
1703 sbi->total_ext_tree = 0;
1704 atomic_set(&sbi->total_ext_node, 0);
1707 int __init create_extent_cache(void)
1709 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
1710 sizeof(struct extent_tree));
1711 if (!extent_tree_slab)
1713 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
1714 sizeof(struct extent_node));
1715 if (!extent_node_slab) {
1716 kmem_cache_destroy(extent_tree_slab);
1722 void destroy_extent_cache(void)
1724 kmem_cache_destroy(extent_node_slab);
1725 kmem_cache_destroy(extent_tree_slab);
1728 const struct address_space_operations f2fs_dblock_aops = {
1729 .readpage = f2fs_read_data_page,
1730 .readpages = f2fs_read_data_pages,
1731 .writepage = f2fs_write_data_page,
1732 .writepages = f2fs_write_data_pages,
1733 .write_begin = f2fs_write_begin,
1734 .write_end = f2fs_write_end,
1735 .set_page_dirty = f2fs_set_data_page_dirty,
1736 .invalidatepage = f2fs_invalidate_page,
1737 .releasepage = f2fs_release_page,
1738 .direct_IO = f2fs_direct_IO,