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/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
22 #include <linux/cleancache.h>
28 #include <trace/events/f2fs.h>
29 #include <trace/events/android_fs.h>
31 static void f2fs_read_end_io(struct bio *bio)
36 if (f2fs_bio_encrypted(bio)) {
38 f2fs_release_crypto_ctx(bio->bi_private);
40 f2fs_end_io_crypto_work(bio->bi_private, bio);
45 bio_for_each_segment_all(bvec, bio, i) {
46 struct page *page = bvec->bv_page;
49 SetPageUptodate(page);
51 ClearPageUptodate(page);
59 static void f2fs_write_end_io(struct bio *bio)
61 struct f2fs_sb_info *sbi = bio->bi_private;
65 bio_for_each_segment_all(bvec, bio, i) {
66 struct page *page = bvec->bv_page;
68 f2fs_restore_and_release_control_page(&page);
70 if (unlikely(bio->bi_error)) {
72 set_bit(AS_EIO, &page->mapping->flags);
73 f2fs_stop_checkpoint(sbi);
75 end_page_writeback(page);
76 dec_page_count(sbi, F2FS_WRITEBACK);
79 if (!get_pages(sbi, F2FS_WRITEBACK) &&
80 !list_empty(&sbi->cp_wait.task_list))
81 wake_up(&sbi->cp_wait);
87 * Low-level block read/write IO operations.
89 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
90 int npages, bool is_read)
94 bio = f2fs_bio_alloc(npages);
96 bio->bi_bdev = sbi->sb->s_bdev;
97 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
98 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
99 bio->bi_private = is_read ? NULL : sbi;
104 static void __submit_merged_bio(struct f2fs_bio_info *io)
106 struct f2fs_io_info *fio = &io->fio;
111 if (is_read_io(fio->rw))
112 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
114 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
116 submit_bio(fio->rw, io->bio);
120 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
121 enum page_type type, int rw)
123 enum page_type btype = PAGE_TYPE_OF_BIO(type);
124 struct f2fs_bio_info *io;
126 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
128 down_write(&io->io_rwsem);
130 /* change META to META_FLUSH in the checkpoint procedure */
131 if (type >= META_FLUSH) {
132 io->fio.type = META_FLUSH;
133 if (test_opt(sbi, NOBARRIER))
134 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
136 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
138 __submit_merged_bio(io);
139 up_write(&io->io_rwsem);
143 * Fill the locked page with data located in the block address.
144 * Return unlocked page.
146 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
149 struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
151 trace_f2fs_submit_page_bio(page, fio);
152 f2fs_trace_ios(fio, 0);
154 /* Allocate a new bio */
155 bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
157 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
162 submit_bio(fio->rw, bio);
166 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
168 struct f2fs_sb_info *sbi = fio->sbi;
169 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
170 struct f2fs_bio_info *io;
171 bool is_read = is_read_io(fio->rw);
172 struct page *bio_page;
174 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
176 verify_block_addr(sbi, fio->blk_addr);
178 down_write(&io->io_rwsem);
181 inc_page_count(sbi, F2FS_WRITEBACK);
183 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
184 io->fio.rw != fio->rw))
185 __submit_merged_bio(io);
187 if (io->bio == NULL) {
188 int bio_blocks = MAX_BIO_BLOCKS(sbi);
190 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
194 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
196 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
198 __submit_merged_bio(io);
202 io->last_block_in_bio = fio->blk_addr;
203 f2fs_trace_ios(fio, 0);
205 up_write(&io->io_rwsem);
206 trace_f2fs_submit_page_mbio(fio->page, fio);
210 * Lock ordering for the change of data block address:
213 * update block addresses in the node page
215 void set_data_blkaddr(struct dnode_of_data *dn)
217 struct f2fs_node *rn;
219 struct page *node_page = dn->node_page;
220 unsigned int ofs_in_node = dn->ofs_in_node;
222 f2fs_wait_on_page_writeback(node_page, NODE);
224 rn = F2FS_NODE(node_page);
226 /* Get physical address of data block */
227 addr_array = blkaddr_in_node(rn);
228 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
229 set_page_dirty(node_page);
232 int reserve_new_block(struct dnode_of_data *dn)
234 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
236 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
238 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
241 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
243 dn->data_blkaddr = NEW_ADDR;
244 set_data_blkaddr(dn);
245 mark_inode_dirty(dn->inode);
250 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
252 bool need_put = dn->inode_page ? false : true;
255 err = get_dnode_of_data(dn, index, ALLOC_NODE);
259 if (dn->data_blkaddr == NULL_ADDR)
260 err = reserve_new_block(dn);
266 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
268 struct extent_info ei;
269 struct inode *inode = dn->inode;
271 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
272 dn->data_blkaddr = ei.blk + index - ei.fofs;
276 return f2fs_reserve_block(dn, index);
279 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
280 int rw, bool for_write)
282 struct address_space *mapping = inode->i_mapping;
283 struct dnode_of_data dn;
285 struct extent_info ei;
287 struct f2fs_io_info fio = {
288 .sbi = F2FS_I_SB(inode),
291 .encrypted_page = NULL,
294 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
295 return read_mapping_page(mapping, index, NULL);
297 page = f2fs_grab_cache_page(mapping, index, for_write);
299 return ERR_PTR(-ENOMEM);
301 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
302 dn.data_blkaddr = ei.blk + index - ei.fofs;
306 set_new_dnode(&dn, inode, NULL, NULL, 0);
307 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
312 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
317 if (PageUptodate(page)) {
323 * A new dentry page is allocated but not able to be written, since its
324 * new inode page couldn't be allocated due to -ENOSPC.
325 * In such the case, its blkaddr can be remained as NEW_ADDR.
326 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
328 if (dn.data_blkaddr == NEW_ADDR) {
329 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
330 SetPageUptodate(page);
335 fio.blk_addr = dn.data_blkaddr;
337 err = f2fs_submit_page_bio(&fio);
343 f2fs_put_page(page, 1);
347 struct page *find_data_page(struct inode *inode, pgoff_t index)
349 struct address_space *mapping = inode->i_mapping;
352 page = find_get_page(mapping, index);
353 if (page && PageUptodate(page))
355 f2fs_put_page(page, 0);
357 page = get_read_data_page(inode, index, READ_SYNC, false);
361 if (PageUptodate(page))
364 wait_on_page_locked(page);
365 if (unlikely(!PageUptodate(page))) {
366 f2fs_put_page(page, 0);
367 return ERR_PTR(-EIO);
373 * If it tries to access a hole, return an error.
374 * Because, the callers, functions in dir.c and GC, should be able to know
375 * whether this page exists or not.
377 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
380 struct address_space *mapping = inode->i_mapping;
383 page = get_read_data_page(inode, index, READ_SYNC, for_write);
387 /* wait for read completion */
389 if (unlikely(!PageUptodate(page))) {
390 f2fs_put_page(page, 1);
391 return ERR_PTR(-EIO);
393 if (unlikely(page->mapping != mapping)) {
394 f2fs_put_page(page, 1);
401 * Caller ensures that this data page is never allocated.
402 * A new zero-filled data page is allocated in the page cache.
404 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
406 * Note that, ipage is set only by make_empty_dir, and if any error occur,
407 * ipage should be released by this function.
409 struct page *get_new_data_page(struct inode *inode,
410 struct page *ipage, pgoff_t index, bool new_i_size)
412 struct address_space *mapping = inode->i_mapping;
414 struct dnode_of_data dn;
417 page = f2fs_grab_cache_page(mapping, index, true);
420 * before exiting, we should make sure ipage will be released
421 * if any error occur.
423 f2fs_put_page(ipage, 1);
424 return ERR_PTR(-ENOMEM);
427 set_new_dnode(&dn, inode, ipage, NULL, 0);
428 err = f2fs_reserve_block(&dn, index);
430 f2fs_put_page(page, 1);
436 if (PageUptodate(page))
439 if (dn.data_blkaddr == NEW_ADDR) {
440 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
441 SetPageUptodate(page);
443 f2fs_put_page(page, 1);
445 page = get_read_data_page(inode, index, READ_SYNC, true);
449 /* wait for read completion */
453 if (new_i_size && i_size_read(inode) <
454 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
455 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
456 /* Only the directory inode sets new_i_size */
457 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
462 static int __allocate_data_block(struct dnode_of_data *dn)
464 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
465 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
466 struct f2fs_summary sum;
468 int seg = CURSEG_WARM_DATA;
471 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
474 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
475 if (dn->data_blkaddr == NEW_ADDR)
478 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
482 get_node_info(sbi, dn->nid, &ni);
483 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
485 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
486 seg = CURSEG_DIRECT_IO;
488 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
490 set_data_blkaddr(dn);
493 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
495 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
496 i_size_write(dn->inode,
497 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
499 /* direct IO doesn't use extent cache to maximize the performance */
500 f2fs_drop_largest_extent(dn->inode, fofs);
505 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
508 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
509 struct dnode_of_data dn;
510 u64 start = F2FS_BYTES_TO_BLK(offset);
511 u64 len = F2FS_BYTES_TO_BLK(count);
516 f2fs_balance_fs(sbi);
519 /* When reading holes, we need its node page */
520 set_new_dnode(&dn, inode, NULL, NULL, 0);
521 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
525 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
527 while (dn.ofs_in_node < end_offset && len) {
530 if (unlikely(f2fs_cp_error(sbi)))
533 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
534 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
535 if (__allocate_data_block(&dn))
545 sync_inode_page(&dn);
554 sync_inode_page(&dn);
562 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
563 * f2fs_map_blocks structure.
564 * If original data blocks are allocated, then give them to blockdev.
566 * a. preallocate requested block addresses
567 * b. do not use extent cache for better performance
568 * c. give the block addresses to blockdev
570 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
571 int create, int flag)
573 unsigned int maxblocks = map->m_len;
574 struct dnode_of_data dn;
575 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
576 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
577 pgoff_t pgofs, end_offset;
578 int err = 0, ofs = 1;
579 struct extent_info ei;
580 bool allocated = false;
585 /* it only supports block size == page size */
586 pgofs = (pgoff_t)map->m_lblk;
588 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
589 map->m_pblk = ei.blk + pgofs - ei.fofs;
590 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
591 map->m_flags = F2FS_MAP_MAPPED;
596 f2fs_lock_op(F2FS_I_SB(inode));
598 /* When reading holes, we need its node page */
599 set_new_dnode(&dn, inode, NULL, NULL, 0);
600 err = get_dnode_of_data(&dn, pgofs, mode);
607 if (dn.data_blkaddr == NEW_ADDR || dn.data_blkaddr == NULL_ADDR) {
609 if (unlikely(f2fs_cp_error(sbi))) {
613 err = __allocate_data_block(&dn);
617 map->m_flags = F2FS_MAP_NEW;
619 if (flag != F2FS_GET_BLOCK_FIEMAP ||
620 dn.data_blkaddr != NEW_ADDR) {
621 if (flag == F2FS_GET_BLOCK_BMAP)
627 * preallocated unwritten block should be mapped
630 if (dn.data_blkaddr == NEW_ADDR)
631 map->m_flags = F2FS_MAP_UNWRITTEN;
635 map->m_flags |= F2FS_MAP_MAPPED;
636 map->m_pblk = dn.data_blkaddr;
639 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
644 if (dn.ofs_in_node >= end_offset) {
646 sync_inode_page(&dn);
650 set_new_dnode(&dn, inode, NULL, NULL, 0);
651 err = get_dnode_of_data(&dn, pgofs, mode);
658 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
661 if (maxblocks > map->m_len) {
662 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
664 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
666 if (unlikely(f2fs_cp_error(sbi))) {
670 err = __allocate_data_block(&dn);
674 map->m_flags |= F2FS_MAP_NEW;
675 blkaddr = dn.data_blkaddr;
678 * we only merge preallocated unwritten blocks
681 if (flag != F2FS_GET_BLOCK_FIEMAP ||
687 /* Give more consecutive addresses for the readahead */
688 if ((map->m_pblk != NEW_ADDR &&
689 blkaddr == (map->m_pblk + ofs)) ||
690 (map->m_pblk == NEW_ADDR &&
691 blkaddr == NEW_ADDR)) {
701 sync_inode_page(&dn);
706 f2fs_unlock_op(F2FS_I_SB(inode));
708 trace_f2fs_map_blocks(inode, map, err);
712 static int __get_data_block(struct inode *inode, sector_t iblock,
713 struct buffer_head *bh, int create, int flag)
715 struct f2fs_map_blocks map;
719 map.m_len = bh->b_size >> inode->i_blkbits;
721 ret = f2fs_map_blocks(inode, &map, create, flag);
723 map_bh(bh, inode->i_sb, map.m_pblk);
724 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
725 bh->b_size = map.m_len << inode->i_blkbits;
730 static int get_data_block(struct inode *inode, sector_t iblock,
731 struct buffer_head *bh_result, int create, int flag)
733 return __get_data_block(inode, iblock, bh_result, create, flag);
736 static int get_data_block_dio(struct inode *inode, sector_t iblock,
737 struct buffer_head *bh_result, int create)
739 return __get_data_block(inode, iblock, bh_result, create,
743 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
744 struct buffer_head *bh_result, int create)
746 return __get_data_block(inode, iblock, bh_result, create,
747 F2FS_GET_BLOCK_BMAP);
750 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
752 return (offset >> inode->i_blkbits);
755 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
757 return (blk << inode->i_blkbits);
760 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
763 struct buffer_head map_bh;
764 sector_t start_blk, last_blk;
765 loff_t isize = i_size_read(inode);
766 u64 logical = 0, phys = 0, size = 0;
768 bool past_eof = false, whole_file = false;
771 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
775 if (f2fs_has_inline_data(inode)) {
776 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
781 mutex_lock(&inode->i_mutex);
788 if (logical_to_blk(inode, len) == 0)
789 len = blk_to_logical(inode, 1);
791 start_blk = logical_to_blk(inode, start);
792 last_blk = logical_to_blk(inode, start + len - 1);
794 memset(&map_bh, 0, sizeof(struct buffer_head));
797 ret = get_data_block(inode, start_blk, &map_bh, 0,
798 F2FS_GET_BLOCK_FIEMAP);
803 if (!buffer_mapped(&map_bh)) {
806 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
809 if (past_eof && size) {
810 flags |= FIEMAP_EXTENT_LAST;
811 ret = fiemap_fill_next_extent(fieinfo, logical,
814 ret = fiemap_fill_next_extent(fieinfo, logical,
819 /* if we have holes up to/past EOF then we're done */
820 if (start_blk > last_blk || past_eof || ret)
823 if (start_blk > last_blk && !whole_file) {
824 ret = fiemap_fill_next_extent(fieinfo, logical,
830 * if size != 0 then we know we already have an extent
834 ret = fiemap_fill_next_extent(fieinfo, logical,
840 logical = blk_to_logical(inode, start_blk);
841 phys = blk_to_logical(inode, map_bh.b_blocknr);
842 size = map_bh.b_size;
844 if (buffer_unwritten(&map_bh))
845 flags = FIEMAP_EXTENT_UNWRITTEN;
847 start_blk += logical_to_blk(inode, size);
850 * If we are past the EOF, then we need to make sure as
851 * soon as we find a hole that the last extent we found
852 * is marked with FIEMAP_EXTENT_LAST
854 if (!past_eof && logical + size >= isize)
858 if (fatal_signal_pending(current))
866 mutex_unlock(&inode->i_mutex);
871 * This function was originally taken from fs/mpage.c, and customized for f2fs.
872 * Major change was from block_size == page_size in f2fs by default.
874 static int f2fs_mpage_readpages(struct address_space *mapping,
875 struct list_head *pages, struct page *page,
878 struct bio *bio = NULL;
880 sector_t last_block_in_bio = 0;
881 struct inode *inode = mapping->host;
882 const unsigned blkbits = inode->i_blkbits;
883 const unsigned blocksize = 1 << blkbits;
884 sector_t block_in_file;
886 sector_t last_block_in_file;
888 struct block_device *bdev = inode->i_sb->s_bdev;
889 struct f2fs_map_blocks map;
896 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
898 prefetchw(&page->flags);
900 page = list_entry(pages->prev, struct page, lru);
901 list_del(&page->lru);
902 if (add_to_page_cache_lru(page, mapping,
903 page->index, GFP_KERNEL))
907 block_in_file = (sector_t)page->index;
908 last_block = block_in_file + nr_pages;
909 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
911 if (last_block > last_block_in_file)
912 last_block = last_block_in_file;
915 * Map blocks using the previous result first.
917 if ((map.m_flags & F2FS_MAP_MAPPED) &&
918 block_in_file > map.m_lblk &&
919 block_in_file < (map.m_lblk + map.m_len))
923 * Then do more f2fs_map_blocks() calls until we are
924 * done with this page.
928 if (block_in_file < last_block) {
929 map.m_lblk = block_in_file;
930 map.m_len = last_block - block_in_file;
932 if (f2fs_map_blocks(inode, &map, 0,
933 F2FS_GET_BLOCK_READ))
937 if ((map.m_flags & F2FS_MAP_MAPPED)) {
938 block_nr = map.m_pblk + block_in_file - map.m_lblk;
939 SetPageMappedToDisk(page);
941 if (!PageUptodate(page) && !cleancache_get_page(page)) {
942 SetPageUptodate(page);
946 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
947 SetPageUptodate(page);
953 * This page will go to BIO. Do we need to send this
956 if (bio && (last_block_in_bio != block_nr - 1)) {
958 submit_bio(READ, bio);
962 struct f2fs_crypto_ctx *ctx = NULL;
964 if (f2fs_encrypted_inode(inode) &&
965 S_ISREG(inode->i_mode)) {
967 ctx = f2fs_get_crypto_ctx(inode);
971 /* wait the page to be moved by cleaning */
972 f2fs_wait_on_encrypted_page_writeback(
973 F2FS_I_SB(inode), block_nr);
976 bio = bio_alloc(GFP_KERNEL,
977 min_t(int, nr_pages, BIO_MAX_PAGES));
980 f2fs_release_crypto_ctx(ctx);
984 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
985 bio->bi_end_io = f2fs_read_end_io;
986 bio->bi_private = ctx;
989 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
990 goto submit_and_realloc;
992 last_block_in_bio = block_nr;
996 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1001 submit_bio(READ, bio);
1007 page_cache_release(page);
1009 BUG_ON(pages && !list_empty(pages));
1011 submit_bio(READ, bio);
1015 static int f2fs_read_data_page(struct file *file, struct page *page)
1017 struct inode *inode = page->mapping->host;
1020 trace_f2fs_readpage(page, DATA);
1022 /* If the file has inline data, try to read it directly */
1023 if (f2fs_has_inline_data(inode))
1024 ret = f2fs_read_inline_data(inode, page);
1026 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1030 static int f2fs_read_data_pages(struct file *file,
1031 struct address_space *mapping,
1032 struct list_head *pages, unsigned nr_pages)
1034 struct inode *inode = file->f_mapping->host;
1035 struct page *page = list_entry(pages->prev, struct page, lru);
1037 trace_f2fs_readpages(inode, page, nr_pages);
1039 /* If the file has inline data, skip readpages */
1040 if (f2fs_has_inline_data(inode))
1043 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1046 int do_write_data_page(struct f2fs_io_info *fio)
1048 struct page *page = fio->page;
1049 struct inode *inode = page->mapping->host;
1050 struct dnode_of_data dn;
1053 set_new_dnode(&dn, inode, NULL, NULL, 0);
1054 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1058 fio->blk_addr = dn.data_blkaddr;
1060 /* This page is already truncated */
1061 if (fio->blk_addr == NULL_ADDR) {
1062 ClearPageUptodate(page);
1066 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1068 /* wait for GCed encrypted page writeback */
1069 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1072 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1073 if (IS_ERR(fio->encrypted_page)) {
1074 err = PTR_ERR(fio->encrypted_page);
1079 set_page_writeback(page);
1082 * If current allocation needs SSR,
1083 * it had better in-place writes for updated data.
1085 if (unlikely(fio->blk_addr != NEW_ADDR &&
1086 !is_cold_data(page) &&
1087 need_inplace_update(inode))) {
1088 rewrite_data_page(fio);
1089 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1090 trace_f2fs_do_write_data_page(page, IPU);
1092 write_data_page(&dn, fio);
1093 set_data_blkaddr(&dn);
1094 f2fs_update_extent_cache(&dn);
1095 trace_f2fs_do_write_data_page(page, OPU);
1096 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1097 if (page->index == 0)
1098 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1101 f2fs_put_dnode(&dn);
1105 static int f2fs_write_data_page(struct page *page,
1106 struct writeback_control *wbc)
1108 struct inode *inode = page->mapping->host;
1109 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1110 loff_t i_size = i_size_read(inode);
1111 const pgoff_t end_index = ((unsigned long long) i_size)
1112 >> PAGE_CACHE_SHIFT;
1113 unsigned offset = 0;
1114 bool need_balance_fs = false;
1116 struct f2fs_io_info fio = {
1119 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1121 .encrypted_page = NULL,
1124 trace_f2fs_writepage(page, DATA);
1126 if (page->index < end_index)
1130 * If the offset is out-of-range of file size,
1131 * this page does not have to be written to disk.
1133 offset = i_size & (PAGE_CACHE_SIZE - 1);
1134 if ((page->index >= end_index + 1) || !offset)
1137 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1139 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1141 if (f2fs_is_drop_cache(inode))
1143 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1144 available_free_memory(sbi, BASE_CHECK))
1147 /* Dentry blocks are controlled by checkpoint */
1148 if (S_ISDIR(inode->i_mode)) {
1149 if (unlikely(f2fs_cp_error(sbi)))
1151 err = do_write_data_page(&fio);
1155 /* we should bypass data pages to proceed the kworkder jobs */
1156 if (unlikely(f2fs_cp_error(sbi))) {
1161 if (!wbc->for_reclaim)
1162 need_balance_fs = true;
1163 else if (has_not_enough_free_secs(sbi, 0))
1168 if (f2fs_has_inline_data(inode))
1169 err = f2fs_write_inline_data(inode, page);
1171 err = do_write_data_page(&fio);
1172 f2fs_unlock_op(sbi);
1174 if (err && err != -ENOENT)
1177 clear_cold_data(page);
1179 inode_dec_dirty_pages(inode);
1181 ClearPageUptodate(page);
1183 if (need_balance_fs)
1184 f2fs_balance_fs(sbi);
1185 if (wbc->for_reclaim)
1186 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1190 redirty_page_for_writepage(wbc, page);
1191 return AOP_WRITEPAGE_ACTIVATE;
1194 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1197 struct address_space *mapping = data;
1198 int ret = mapping->a_ops->writepage(page, wbc);
1199 mapping_set_error(mapping, ret);
1204 * This function was copied from write_cche_pages from mm/page-writeback.c.
1205 * The major change is making write step of cold data page separately from
1206 * warm/hot data page.
1208 static int f2fs_write_cache_pages(struct address_space *mapping,
1209 struct writeback_control *wbc, writepage_t writepage,
1214 struct pagevec pvec;
1216 pgoff_t uninitialized_var(writeback_index);
1218 pgoff_t end; /* Inclusive */
1221 int range_whole = 0;
1225 pagevec_init(&pvec, 0);
1227 if (wbc->range_cyclic) {
1228 writeback_index = mapping->writeback_index; /* prev offset */
1229 index = writeback_index;
1236 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1237 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1238 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1240 cycled = 1; /* ignore range_cyclic tests */
1242 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1243 tag = PAGECACHE_TAG_TOWRITE;
1245 tag = PAGECACHE_TAG_DIRTY;
1247 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1248 tag_pages_for_writeback(mapping, index, end);
1250 while (!done && (index <= end)) {
1253 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1254 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1258 for (i = 0; i < nr_pages; i++) {
1259 struct page *page = pvec.pages[i];
1261 if (page->index > end) {
1266 done_index = page->index;
1270 if (unlikely(page->mapping != mapping)) {
1276 if (!PageDirty(page)) {
1277 /* someone wrote it for us */
1278 goto continue_unlock;
1281 if (step == is_cold_data(page))
1282 goto continue_unlock;
1284 if (PageWriteback(page)) {
1285 if (wbc->sync_mode != WB_SYNC_NONE)
1286 f2fs_wait_on_page_writeback(page, DATA);
1288 goto continue_unlock;
1291 BUG_ON(PageWriteback(page));
1292 if (!clear_page_dirty_for_io(page))
1293 goto continue_unlock;
1295 ret = (*writepage)(page, wbc, data);
1296 if (unlikely(ret)) {
1297 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1301 done_index = page->index + 1;
1307 if (--wbc->nr_to_write <= 0 &&
1308 wbc->sync_mode == WB_SYNC_NONE) {
1313 pagevec_release(&pvec);
1322 if (!cycled && !done) {
1325 end = writeback_index - 1;
1328 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1329 mapping->writeback_index = done_index;
1334 static int f2fs_write_data_pages(struct address_space *mapping,
1335 struct writeback_control *wbc)
1337 struct inode *inode = mapping->host;
1338 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1339 bool locked = false;
1343 trace_f2fs_writepages(mapping->host, wbc, DATA);
1345 /* deal with chardevs and other special file */
1346 if (!mapping->a_ops->writepage)
1349 /* skip writing if there is no dirty page in this inode */
1350 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1353 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1354 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1355 available_free_memory(sbi, DIRTY_DENTS))
1358 /* during POR, we don't need to trigger writepage at all. */
1359 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1362 diff = nr_pages_to_write(sbi, DATA, wbc);
1364 if (!S_ISDIR(inode->i_mode)) {
1365 mutex_lock(&sbi->writepages);
1368 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1369 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1371 mutex_unlock(&sbi->writepages);
1373 remove_dirty_dir_inode(inode);
1375 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1379 wbc->pages_skipped += get_dirty_pages(inode);
1383 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1385 struct inode *inode = mapping->host;
1387 if (to > inode->i_size) {
1388 truncate_pagecache(inode, inode->i_size);
1389 truncate_blocks(inode, inode->i_size, true);
1393 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1394 loff_t pos, unsigned len, unsigned flags,
1395 struct page **pagep, void **fsdata)
1397 struct inode *inode = mapping->host;
1398 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1399 struct page *page = NULL;
1401 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1402 struct dnode_of_data dn;
1405 trace_android_fs_datawrite_start(inode, pos, len,
1406 current->pid, current->comm);
1407 trace_f2fs_write_begin(inode, pos, len, flags);
1409 f2fs_balance_fs(sbi);
1412 * We should check this at this moment to avoid deadlock on inode page
1413 * and #0 page. The locking rule for inline_data conversion should be:
1414 * lock_page(page #0) -> lock_page(inode_page)
1417 err = f2fs_convert_inline_inode(inode);
1422 page = grab_cache_page_write_begin(mapping, index, flags);
1432 /* check inline_data */
1433 ipage = get_node_page(sbi, inode->i_ino);
1434 if (IS_ERR(ipage)) {
1435 err = PTR_ERR(ipage);
1439 set_new_dnode(&dn, inode, ipage, ipage, 0);
1441 if (f2fs_has_inline_data(inode)) {
1442 if (pos + len <= MAX_INLINE_DATA) {
1443 read_inline_data(page, ipage);
1444 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1445 sync_inode_page(&dn);
1448 err = f2fs_convert_inline_page(&dn, page);
1453 err = f2fs_get_block(&dn, index);
1457 f2fs_put_dnode(&dn);
1458 f2fs_unlock_op(sbi);
1460 f2fs_wait_on_page_writeback(page, DATA);
1462 /* wait for GCed encrypted page writeback */
1463 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1464 f2fs_wait_on_encrypted_page_writeback(sbi, dn.data_blkaddr);
1466 if (len == PAGE_CACHE_SIZE)
1468 if (PageUptodate(page))
1471 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1472 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1473 unsigned end = start + len;
1475 /* Reading beyond i_size is simple: memset to zero */
1476 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1480 if (dn.data_blkaddr == NEW_ADDR) {
1481 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1483 struct f2fs_io_info fio = {
1487 .blk_addr = dn.data_blkaddr,
1489 .encrypted_page = NULL,
1491 err = f2fs_submit_page_bio(&fio);
1496 if (unlikely(!PageUptodate(page))) {
1500 if (unlikely(page->mapping != mapping)) {
1501 f2fs_put_page(page, 1);
1505 /* avoid symlink page */
1506 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1507 err = f2fs_decrypt_one(inode, page);
1513 SetPageUptodate(page);
1515 clear_cold_data(page);
1519 f2fs_put_dnode(&dn);
1521 f2fs_unlock_op(sbi);
1523 f2fs_put_page(page, 1);
1524 f2fs_write_failed(mapping, pos + len);
1528 static int f2fs_write_end(struct file *file,
1529 struct address_space *mapping,
1530 loff_t pos, unsigned len, unsigned copied,
1531 struct page *page, void *fsdata)
1533 struct inode *inode = page->mapping->host;
1535 trace_android_fs_datawrite_end(inode, pos, len);
1536 trace_f2fs_write_end(inode, pos, len, copied);
1538 set_page_dirty(page);
1540 if (pos + copied > i_size_read(inode)) {
1541 i_size_write(inode, pos + copied);
1542 mark_inode_dirty(inode);
1543 update_inode_page(inode);
1546 f2fs_put_page(page, 1);
1550 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1553 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1555 if (offset & blocksize_mask)
1558 if (iov_iter_alignment(iter) & blocksize_mask)
1564 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1567 struct file *file = iocb->ki_filp;
1568 struct address_space *mapping = file->f_mapping;
1569 struct inode *inode = mapping->host;
1570 size_t count = iov_iter_count(iter);
1573 /* we don't need to use inline_data strictly */
1574 if (f2fs_has_inline_data(inode)) {
1575 err = f2fs_convert_inline_inode(inode);
1580 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1583 err = check_direct_IO(inode, iter, offset);
1587 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1589 if (trace_android_fs_dataread_start_enabled() &&
1590 (iov_iter_rw(iter) == READ))
1591 trace_android_fs_dataread_start(inode, offset,
1592 count, current->pid,
1594 if (trace_android_fs_datawrite_start_enabled() &&
1595 (iov_iter_rw(iter) == WRITE))
1596 trace_android_fs_datawrite_start(inode, offset, count,
1597 current->pid, current->comm);
1599 if (iov_iter_rw(iter) == WRITE) {
1600 __allocate_data_blocks(inode, offset, count);
1601 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1607 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1609 if (err < 0 && iov_iter_rw(iter) == WRITE)
1610 f2fs_write_failed(mapping, offset + count);
1612 if (trace_android_fs_dataread_start_enabled() &&
1613 (iov_iter_rw(iter) == READ))
1614 trace_android_fs_dataread_end(inode, offset, count);
1615 if (trace_android_fs_datawrite_start_enabled() &&
1616 (iov_iter_rw(iter) == WRITE))
1617 trace_android_fs_datawrite_end(inode, offset, count);
1619 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1624 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1625 unsigned int length)
1627 struct inode *inode = page->mapping->host;
1628 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1630 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1631 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1634 if (PageDirty(page)) {
1635 if (inode->i_ino == F2FS_META_INO(sbi))
1636 dec_page_count(sbi, F2FS_DIRTY_META);
1637 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1638 dec_page_count(sbi, F2FS_DIRTY_NODES);
1640 inode_dec_dirty_pages(inode);
1643 /* This is atomic written page, keep Private */
1644 if (IS_ATOMIC_WRITTEN_PAGE(page))
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 /* This is atomic written page, keep Private */
1657 if (IS_ATOMIC_WRITTEN_PAGE(page))
1660 ClearPagePrivate(page);
1664 static int f2fs_set_data_page_dirty(struct page *page)
1666 struct address_space *mapping = page->mapping;
1667 struct inode *inode = mapping->host;
1669 trace_f2fs_set_page_dirty(page, DATA);
1671 SetPageUptodate(page);
1673 if (f2fs_is_atomic_file(inode)) {
1674 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1675 register_inmem_page(inode, page);
1679 * Previously, this page has been registered, we just
1685 if (!PageDirty(page)) {
1686 __set_page_dirty_nobuffers(page);
1687 update_dirty_page(inode, page);
1693 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1695 struct inode *inode = mapping->host;
1697 if (f2fs_has_inline_data(inode))
1700 /* make sure allocating whole blocks */
1701 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1702 filemap_write_and_wait(mapping);
1704 return generic_block_bmap(mapping, block, get_data_block_bmap);
1707 const struct address_space_operations f2fs_dblock_aops = {
1708 .readpage = f2fs_read_data_page,
1709 .readpages = f2fs_read_data_pages,
1710 .writepage = f2fs_write_data_page,
1711 .writepages = f2fs_write_data_pages,
1712 .write_begin = f2fs_write_begin,
1713 .write_end = f2fs_write_end,
1714 .set_page_dirty = f2fs_set_data_page_dirty,
1715 .invalidatepage = f2fs_invalidate_page,
1716 .releasepage = f2fs_release_page,
1717 .direct_IO = f2fs_direct_IO,