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>
30 static void f2fs_read_end_io(struct bio *bio)
35 if (f2fs_bio_encrypted(bio)) {
37 f2fs_release_crypto_ctx(bio->bi_private);
39 f2fs_end_io_crypto_work(bio->bi_private, bio);
44 bio_for_each_segment_all(bvec, bio, i) {
45 struct page *page = bvec->bv_page;
48 SetPageUptodate(page);
50 ClearPageUptodate(page);
58 static void f2fs_write_end_io(struct bio *bio)
60 struct f2fs_sb_info *sbi = bio->bi_private;
64 bio_for_each_segment_all(bvec, bio, i) {
65 struct page *page = bvec->bv_page;
67 f2fs_restore_and_release_control_page(&page);
69 if (unlikely(bio->bi_error)) {
71 set_bit(AS_EIO, &page->mapping->flags);
72 f2fs_stop_checkpoint(sbi);
74 end_page_writeback(page);
75 dec_page_count(sbi, F2FS_WRITEBACK);
78 if (!get_pages(sbi, F2FS_WRITEBACK) &&
79 !list_empty(&sbi->cp_wait.task_list))
80 wake_up(&sbi->cp_wait);
86 * Low-level block read/write IO operations.
88 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
89 int npages, bool is_read)
93 bio = f2fs_bio_alloc(npages);
95 bio->bi_bdev = sbi->sb->s_bdev;
96 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
97 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
98 bio->bi_private = is_read ? NULL : sbi;
103 static void __submit_merged_bio(struct f2fs_bio_info *io)
105 struct f2fs_io_info *fio = &io->fio;
110 if (is_read_io(fio->rw))
111 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
113 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
115 submit_bio(fio->rw, io->bio);
119 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
120 enum page_type type, int rw)
122 enum page_type btype = PAGE_TYPE_OF_BIO(type);
123 struct f2fs_bio_info *io;
125 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
127 down_write(&io->io_rwsem);
129 /* change META to META_FLUSH in the checkpoint procedure */
130 if (type >= META_FLUSH) {
131 io->fio.type = META_FLUSH;
132 if (test_opt(sbi, NOBARRIER))
133 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
135 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
137 __submit_merged_bio(io);
138 up_write(&io->io_rwsem);
142 * Fill the locked page with data located in the block address.
143 * Return unlocked page.
145 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
148 struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
150 trace_f2fs_submit_page_bio(page, fio);
151 f2fs_trace_ios(fio, 0);
153 /* Allocate a new bio */
154 bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
156 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
161 submit_bio(fio->rw, bio);
165 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
167 struct f2fs_sb_info *sbi = fio->sbi;
168 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
169 struct f2fs_bio_info *io;
170 bool is_read = is_read_io(fio->rw);
171 struct page *bio_page;
173 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
175 verify_block_addr(sbi, fio->blk_addr);
177 down_write(&io->io_rwsem);
180 inc_page_count(sbi, F2FS_WRITEBACK);
182 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
183 io->fio.rw != fio->rw))
184 __submit_merged_bio(io);
186 if (io->bio == NULL) {
187 int bio_blocks = MAX_BIO_BLOCKS(sbi);
189 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
193 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
195 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
197 __submit_merged_bio(io);
201 io->last_block_in_bio = fio->blk_addr;
202 f2fs_trace_ios(fio, 0);
204 up_write(&io->io_rwsem);
205 trace_f2fs_submit_page_mbio(fio->page, fio);
209 * Lock ordering for the change of data block address:
212 * update block addresses in the node page
214 void set_data_blkaddr(struct dnode_of_data *dn)
216 struct f2fs_node *rn;
218 struct page *node_page = dn->node_page;
219 unsigned int ofs_in_node = dn->ofs_in_node;
221 f2fs_wait_on_page_writeback(node_page, NODE);
223 rn = F2FS_NODE(node_page);
225 /* Get physical address of data block */
226 addr_array = blkaddr_in_node(rn);
227 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
228 set_page_dirty(node_page);
231 int reserve_new_block(struct dnode_of_data *dn)
233 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
235 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
237 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
240 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
242 dn->data_blkaddr = NEW_ADDR;
243 set_data_blkaddr(dn);
244 mark_inode_dirty(dn->inode);
249 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
251 bool need_put = dn->inode_page ? false : true;
254 err = get_dnode_of_data(dn, index, ALLOC_NODE);
258 if (dn->data_blkaddr == NULL_ADDR)
259 err = reserve_new_block(dn);
265 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
267 struct extent_info ei;
268 struct inode *inode = dn->inode;
270 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
271 dn->data_blkaddr = ei.blk + index - ei.fofs;
275 return f2fs_reserve_block(dn, index);
278 struct page *get_read_data_page(struct inode *inode, pgoff_t index, int rw)
280 struct address_space *mapping = inode->i_mapping;
281 struct dnode_of_data dn;
283 struct extent_info ei;
285 struct f2fs_io_info fio = {
286 .sbi = F2FS_I_SB(inode),
289 .encrypted_page = NULL,
292 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
293 return read_mapping_page(mapping, index, NULL);
295 page = grab_cache_page(mapping, index);
297 return ERR_PTR(-ENOMEM);
299 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
300 dn.data_blkaddr = ei.blk + index - ei.fofs;
304 set_new_dnode(&dn, inode, NULL, NULL, 0);
305 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
310 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
315 if (PageUptodate(page)) {
321 * A new dentry page is allocated but not able to be written, since its
322 * new inode page couldn't be allocated due to -ENOSPC.
323 * In such the case, its blkaddr can be remained as NEW_ADDR.
324 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
326 if (dn.data_blkaddr == NEW_ADDR) {
327 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
328 SetPageUptodate(page);
333 fio.blk_addr = dn.data_blkaddr;
335 err = f2fs_submit_page_bio(&fio);
341 f2fs_put_page(page, 1);
345 struct page *find_data_page(struct inode *inode, pgoff_t index)
347 struct address_space *mapping = inode->i_mapping;
350 page = find_get_page(mapping, index);
351 if (page && PageUptodate(page))
353 f2fs_put_page(page, 0);
355 page = get_read_data_page(inode, index, READ_SYNC);
359 if (PageUptodate(page))
362 wait_on_page_locked(page);
363 if (unlikely(!PageUptodate(page))) {
364 f2fs_put_page(page, 0);
365 return ERR_PTR(-EIO);
371 * If it tries to access a hole, return an error.
372 * Because, the callers, functions in dir.c and GC, should be able to know
373 * whether this page exists or not.
375 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
377 struct address_space *mapping = inode->i_mapping;
380 page = get_read_data_page(inode, index, READ_SYNC);
384 /* wait for read completion */
386 if (unlikely(!PageUptodate(page))) {
387 f2fs_put_page(page, 1);
388 return ERR_PTR(-EIO);
390 if (unlikely(page->mapping != mapping)) {
391 f2fs_put_page(page, 1);
398 * Caller ensures that this data page is never allocated.
399 * A new zero-filled data page is allocated in the page cache.
401 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
403 * Note that, ipage is set only by make_empty_dir, and if any error occur,
404 * ipage should be released by this function.
406 struct page *get_new_data_page(struct inode *inode,
407 struct page *ipage, pgoff_t index, bool new_i_size)
409 struct address_space *mapping = inode->i_mapping;
411 struct dnode_of_data dn;
414 page = grab_cache_page(mapping, index);
417 * before exiting, we should make sure ipage will be released
418 * if any error occur.
420 f2fs_put_page(ipage, 1);
421 return ERR_PTR(-ENOMEM);
424 set_new_dnode(&dn, inode, ipage, NULL, 0);
425 err = f2fs_reserve_block(&dn, index);
427 f2fs_put_page(page, 1);
433 if (PageUptodate(page))
436 if (dn.data_blkaddr == NEW_ADDR) {
437 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
438 SetPageUptodate(page);
440 f2fs_put_page(page, 1);
442 page = get_read_data_page(inode, index, READ_SYNC);
446 /* wait for read completion */
451 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
452 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
453 /* Only the directory inode sets new_i_size */
454 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
459 static int __allocate_data_block(struct dnode_of_data *dn)
461 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
462 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
463 struct f2fs_summary sum;
465 int seg = CURSEG_WARM_DATA;
468 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
471 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
472 if (dn->data_blkaddr == NEW_ADDR)
475 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
479 get_node_info(sbi, dn->nid, &ni);
480 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
482 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
483 seg = CURSEG_DIRECT_IO;
485 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
487 set_data_blkaddr(dn);
490 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
492 if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
493 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
495 /* direct IO doesn't use extent cache to maximize the performance */
496 f2fs_drop_largest_extent(dn->inode, fofs);
501 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
504 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
505 struct dnode_of_data dn;
506 u64 start = F2FS_BYTES_TO_BLK(offset);
507 u64 len = F2FS_BYTES_TO_BLK(count);
512 f2fs_balance_fs(sbi);
515 /* When reading holes, we need its node page */
516 set_new_dnode(&dn, inode, NULL, NULL, 0);
517 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
521 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
523 while (dn.ofs_in_node < end_offset && len) {
526 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
527 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
528 if (__allocate_data_block(&dn))
538 sync_inode_page(&dn);
547 sync_inode_page(&dn);
555 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
556 * f2fs_map_blocks structure.
557 * If original data blocks are allocated, then give them to blockdev.
559 * a. preallocate requested block addresses
560 * b. do not use extent cache for better performance
561 * c. give the block addresses to blockdev
563 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
564 int create, int flag)
566 unsigned int maxblocks = map->m_len;
567 struct dnode_of_data dn;
568 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
569 pgoff_t pgofs, end_offset;
570 int err = 0, ofs = 1;
571 struct extent_info ei;
572 bool allocated = false;
577 /* it only supports block size == page size */
578 pgofs = (pgoff_t)map->m_lblk;
580 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
581 map->m_pblk = ei.blk + pgofs - ei.fofs;
582 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
583 map->m_flags = F2FS_MAP_MAPPED;
588 f2fs_lock_op(F2FS_I_SB(inode));
590 /* When reading holes, we need its node page */
591 set_new_dnode(&dn, inode, NULL, NULL, 0);
592 err = get_dnode_of_data(&dn, pgofs, mode);
598 if (dn.data_blkaddr == NEW_ADDR) {
599 if (flag == F2FS_GET_BLOCK_BMAP) {
602 } else if (flag == F2FS_GET_BLOCK_READ ||
603 flag == F2FS_GET_BLOCK_DIO) {
607 * if it is in fiemap call path (flag = F2FS_GET_BLOCK_FIEMAP),
608 * mark it as mapped and unwritten block.
612 if (dn.data_blkaddr != NULL_ADDR) {
613 map->m_flags = F2FS_MAP_MAPPED;
614 map->m_pblk = dn.data_blkaddr;
615 if (dn.data_blkaddr == NEW_ADDR)
616 map->m_flags |= F2FS_MAP_UNWRITTEN;
618 err = __allocate_data_block(&dn);
622 map->m_flags = F2FS_MAP_NEW | F2FS_MAP_MAPPED;
623 map->m_pblk = dn.data_blkaddr;
625 if (flag == F2FS_GET_BLOCK_BMAP)
630 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
636 if (dn.ofs_in_node >= end_offset) {
638 sync_inode_page(&dn);
642 set_new_dnode(&dn, inode, NULL, NULL, 0);
643 err = get_dnode_of_data(&dn, pgofs, mode);
650 if (dn.data_blkaddr == NEW_ADDR &&
651 flag != F2FS_GET_BLOCK_FIEMAP)
654 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
657 if (maxblocks > map->m_len) {
658 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
659 if (blkaddr == NULL_ADDR && create) {
660 err = __allocate_data_block(&dn);
664 map->m_flags |= F2FS_MAP_NEW;
665 blkaddr = dn.data_blkaddr;
667 /* Give more consecutive addresses for the readahead */
668 if ((map->m_pblk != NEW_ADDR &&
669 blkaddr == (map->m_pblk + ofs)) ||
670 (map->m_pblk == NEW_ADDR &&
671 blkaddr == NEW_ADDR)) {
681 sync_inode_page(&dn);
686 f2fs_unlock_op(F2FS_I_SB(inode));
688 trace_f2fs_map_blocks(inode, map, err);
692 static int __get_data_block(struct inode *inode, sector_t iblock,
693 struct buffer_head *bh, int create, int flag)
695 struct f2fs_map_blocks map;
699 map.m_len = bh->b_size >> inode->i_blkbits;
701 ret = f2fs_map_blocks(inode, &map, create, flag);
703 map_bh(bh, inode->i_sb, map.m_pblk);
704 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
705 bh->b_size = map.m_len << inode->i_blkbits;
710 static int get_data_block(struct inode *inode, sector_t iblock,
711 struct buffer_head *bh_result, int create, int flag)
713 return __get_data_block(inode, iblock, bh_result, create, flag);
716 static int get_data_block_dio(struct inode *inode, sector_t iblock,
717 struct buffer_head *bh_result, int create)
719 return __get_data_block(inode, iblock, bh_result, create,
723 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
724 struct buffer_head *bh_result, int create)
726 return __get_data_block(inode, iblock, bh_result, create,
727 F2FS_GET_BLOCK_BMAP);
730 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
732 return (offset >> inode->i_blkbits);
735 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
737 return (blk << inode->i_blkbits);
740 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
743 struct buffer_head map_bh;
744 sector_t start_blk, last_blk;
745 loff_t isize = i_size_read(inode);
746 u64 logical = 0, phys = 0, size = 0;
748 bool past_eof = false, whole_file = false;
751 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
755 mutex_lock(&inode->i_mutex);
762 if (logical_to_blk(inode, len) == 0)
763 len = blk_to_logical(inode, 1);
765 start_blk = logical_to_blk(inode, start);
766 last_blk = logical_to_blk(inode, start + len - 1);
768 memset(&map_bh, 0, sizeof(struct buffer_head));
771 ret = get_data_block(inode, start_blk, &map_bh, 0,
772 F2FS_GET_BLOCK_FIEMAP);
777 if (!buffer_mapped(&map_bh)) {
780 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
783 if (past_eof && size) {
784 flags |= FIEMAP_EXTENT_LAST;
785 ret = fiemap_fill_next_extent(fieinfo, logical,
788 ret = fiemap_fill_next_extent(fieinfo, logical,
793 /* if we have holes up to/past EOF then we're done */
794 if (start_blk > last_blk || past_eof || ret)
797 if (start_blk > last_blk && !whole_file) {
798 ret = fiemap_fill_next_extent(fieinfo, logical,
804 * if size != 0 then we know we already have an extent
808 ret = fiemap_fill_next_extent(fieinfo, logical,
814 logical = blk_to_logical(inode, start_blk);
815 phys = blk_to_logical(inode, map_bh.b_blocknr);
816 size = map_bh.b_size;
818 if (buffer_unwritten(&map_bh))
819 flags = FIEMAP_EXTENT_UNWRITTEN;
821 start_blk += logical_to_blk(inode, size);
824 * If we are past the EOF, then we need to make sure as
825 * soon as we find a hole that the last extent we found
826 * is marked with FIEMAP_EXTENT_LAST
828 if (!past_eof && logical + size >= isize)
832 if (fatal_signal_pending(current))
840 mutex_unlock(&inode->i_mutex);
845 * This function was originally taken from fs/mpage.c, and customized for f2fs.
846 * Major change was from block_size == page_size in f2fs by default.
848 static int f2fs_mpage_readpages(struct address_space *mapping,
849 struct list_head *pages, struct page *page,
852 struct bio *bio = NULL;
854 sector_t last_block_in_bio = 0;
855 struct inode *inode = mapping->host;
856 const unsigned blkbits = inode->i_blkbits;
857 const unsigned blocksize = 1 << blkbits;
858 sector_t block_in_file;
860 sector_t last_block_in_file;
862 struct block_device *bdev = inode->i_sb->s_bdev;
863 struct f2fs_map_blocks map;
870 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
872 prefetchw(&page->flags);
874 page = list_entry(pages->prev, struct page, lru);
875 list_del(&page->lru);
876 if (add_to_page_cache_lru(page, mapping,
877 page->index, GFP_KERNEL))
881 block_in_file = (sector_t)page->index;
882 last_block = block_in_file + nr_pages;
883 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
885 if (last_block > last_block_in_file)
886 last_block = last_block_in_file;
889 * Map blocks using the previous result first.
891 if ((map.m_flags & F2FS_MAP_MAPPED) &&
892 block_in_file > map.m_lblk &&
893 block_in_file < (map.m_lblk + map.m_len))
897 * Then do more f2fs_map_blocks() calls until we are
898 * done with this page.
902 if (block_in_file < last_block) {
903 map.m_lblk = block_in_file;
904 map.m_len = last_block - block_in_file;
906 if (f2fs_map_blocks(inode, &map, 0, false))
910 if ((map.m_flags & F2FS_MAP_MAPPED)) {
911 block_nr = map.m_pblk + block_in_file - map.m_lblk;
912 SetPageMappedToDisk(page);
914 if (!PageUptodate(page) && !cleancache_get_page(page)) {
915 SetPageUptodate(page);
919 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
920 SetPageUptodate(page);
926 * This page will go to BIO. Do we need to send this
929 if (bio && (last_block_in_bio != block_nr - 1)) {
931 submit_bio(READ, bio);
935 struct f2fs_crypto_ctx *ctx = NULL;
937 if (f2fs_encrypted_inode(inode) &&
938 S_ISREG(inode->i_mode)) {
941 ctx = f2fs_get_crypto_ctx(inode);
945 /* wait the page to be moved by cleaning */
946 cpage = find_lock_page(
947 META_MAPPING(F2FS_I_SB(inode)),
950 f2fs_wait_on_page_writeback(cpage,
952 f2fs_put_page(cpage, 1);
956 bio = bio_alloc(GFP_KERNEL,
957 min_t(int, nr_pages, BIO_MAX_PAGES));
960 f2fs_release_crypto_ctx(ctx);
964 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
965 bio->bi_end_io = f2fs_read_end_io;
966 bio->bi_private = ctx;
969 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
970 goto submit_and_realloc;
972 last_block_in_bio = block_nr;
976 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
981 submit_bio(READ, bio);
987 page_cache_release(page);
989 BUG_ON(pages && !list_empty(pages));
991 submit_bio(READ, bio);
995 static int f2fs_read_data_page(struct file *file, struct page *page)
997 struct inode *inode = page->mapping->host;
1000 trace_f2fs_readpage(page, DATA);
1002 /* If the file has inline data, try to read it directly */
1003 if (f2fs_has_inline_data(inode))
1004 ret = f2fs_read_inline_data(inode, page);
1006 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1010 static int f2fs_read_data_pages(struct file *file,
1011 struct address_space *mapping,
1012 struct list_head *pages, unsigned nr_pages)
1014 struct inode *inode = file->f_mapping->host;
1016 /* If the file has inline data, skip readpages */
1017 if (f2fs_has_inline_data(inode))
1020 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1023 int do_write_data_page(struct f2fs_io_info *fio)
1025 struct page *page = fio->page;
1026 struct inode *inode = page->mapping->host;
1027 struct dnode_of_data dn;
1030 set_new_dnode(&dn, inode, NULL, NULL, 0);
1031 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1035 fio->blk_addr = dn.data_blkaddr;
1037 /* This page is already truncated */
1038 if (fio->blk_addr == NULL_ADDR) {
1039 ClearPageUptodate(page);
1043 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1044 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1045 if (IS_ERR(fio->encrypted_page)) {
1046 err = PTR_ERR(fio->encrypted_page);
1051 set_page_writeback(page);
1054 * If current allocation needs SSR,
1055 * it had better in-place writes for updated data.
1057 if (unlikely(fio->blk_addr != NEW_ADDR &&
1058 !is_cold_data(page) &&
1059 need_inplace_update(inode))) {
1060 rewrite_data_page(fio);
1061 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1062 trace_f2fs_do_write_data_page(page, IPU);
1064 write_data_page(&dn, fio);
1065 set_data_blkaddr(&dn);
1066 f2fs_update_extent_cache(&dn);
1067 trace_f2fs_do_write_data_page(page, OPU);
1068 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1069 if (page->index == 0)
1070 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1073 f2fs_put_dnode(&dn);
1077 static int f2fs_write_data_page(struct page *page,
1078 struct writeback_control *wbc)
1080 struct inode *inode = page->mapping->host;
1081 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1082 loff_t i_size = i_size_read(inode);
1083 const pgoff_t end_index = ((unsigned long long) i_size)
1084 >> PAGE_CACHE_SHIFT;
1085 unsigned offset = 0;
1086 bool need_balance_fs = false;
1088 struct f2fs_io_info fio = {
1091 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1093 .encrypted_page = NULL,
1096 trace_f2fs_writepage(page, DATA);
1098 if (page->index < end_index)
1102 * If the offset is out-of-range of file size,
1103 * this page does not have to be written to disk.
1105 offset = i_size & (PAGE_CACHE_SIZE - 1);
1106 if ((page->index >= end_index + 1) || !offset)
1109 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1111 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1113 if (f2fs_is_drop_cache(inode))
1115 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1116 available_free_memory(sbi, BASE_CHECK))
1119 /* Dentry blocks are controlled by checkpoint */
1120 if (S_ISDIR(inode->i_mode)) {
1121 if (unlikely(f2fs_cp_error(sbi)))
1123 err = do_write_data_page(&fio);
1127 /* we should bypass data pages to proceed the kworkder jobs */
1128 if (unlikely(f2fs_cp_error(sbi))) {
1133 if (!wbc->for_reclaim)
1134 need_balance_fs = true;
1135 else if (has_not_enough_free_secs(sbi, 0))
1140 if (f2fs_has_inline_data(inode))
1141 err = f2fs_write_inline_data(inode, page);
1143 err = do_write_data_page(&fio);
1144 f2fs_unlock_op(sbi);
1146 if (err && err != -ENOENT)
1149 clear_cold_data(page);
1151 inode_dec_dirty_pages(inode);
1153 ClearPageUptodate(page);
1155 if (need_balance_fs)
1156 f2fs_balance_fs(sbi);
1157 if (wbc->for_reclaim)
1158 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1162 redirty_page_for_writepage(wbc, page);
1163 return AOP_WRITEPAGE_ACTIVATE;
1166 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1169 struct address_space *mapping = data;
1170 int ret = mapping->a_ops->writepage(page, wbc);
1171 mapping_set_error(mapping, ret);
1176 * This function was copied from write_cche_pages from mm/page-writeback.c.
1177 * The major change is making write step of cold data page separately from
1178 * warm/hot data page.
1180 static int f2fs_write_cache_pages(struct address_space *mapping,
1181 struct writeback_control *wbc, writepage_t writepage,
1186 struct pagevec pvec;
1188 pgoff_t uninitialized_var(writeback_index);
1190 pgoff_t end; /* Inclusive */
1193 int range_whole = 0;
1197 pagevec_init(&pvec, 0);
1199 if (wbc->range_cyclic) {
1200 writeback_index = mapping->writeback_index; /* prev offset */
1201 index = writeback_index;
1208 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1209 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1210 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1212 cycled = 1; /* ignore range_cyclic tests */
1214 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1215 tag = PAGECACHE_TAG_TOWRITE;
1217 tag = PAGECACHE_TAG_DIRTY;
1219 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1220 tag_pages_for_writeback(mapping, index, end);
1222 while (!done && (index <= end)) {
1225 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1226 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1230 for (i = 0; i < nr_pages; i++) {
1231 struct page *page = pvec.pages[i];
1233 if (page->index > end) {
1238 done_index = page->index;
1242 if (unlikely(page->mapping != mapping)) {
1248 if (!PageDirty(page)) {
1249 /* someone wrote it for us */
1250 goto continue_unlock;
1253 if (step == is_cold_data(page))
1254 goto continue_unlock;
1256 if (PageWriteback(page)) {
1257 if (wbc->sync_mode != WB_SYNC_NONE)
1258 f2fs_wait_on_page_writeback(page, DATA);
1260 goto continue_unlock;
1263 BUG_ON(PageWriteback(page));
1264 if (!clear_page_dirty_for_io(page))
1265 goto continue_unlock;
1267 ret = (*writepage)(page, wbc, data);
1268 if (unlikely(ret)) {
1269 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1273 done_index = page->index + 1;
1279 if (--wbc->nr_to_write <= 0 &&
1280 wbc->sync_mode == WB_SYNC_NONE) {
1285 pagevec_release(&pvec);
1294 if (!cycled && !done) {
1297 end = writeback_index - 1;
1300 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1301 mapping->writeback_index = done_index;
1306 static int f2fs_write_data_pages(struct address_space *mapping,
1307 struct writeback_control *wbc)
1309 struct inode *inode = mapping->host;
1310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1311 bool locked = false;
1315 trace_f2fs_writepages(mapping->host, wbc, DATA);
1317 /* deal with chardevs and other special file */
1318 if (!mapping->a_ops->writepage)
1321 /* skip writing if there is no dirty page in this inode */
1322 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1325 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1326 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1327 available_free_memory(sbi, DIRTY_DENTS))
1330 /* during POR, we don't need to trigger writepage at all. */
1331 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1334 diff = nr_pages_to_write(sbi, DATA, wbc);
1336 if (!S_ISDIR(inode->i_mode)) {
1337 mutex_lock(&sbi->writepages);
1340 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1341 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1343 mutex_unlock(&sbi->writepages);
1345 remove_dirty_dir_inode(inode);
1347 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1351 wbc->pages_skipped += get_dirty_pages(inode);
1355 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1357 struct inode *inode = mapping->host;
1359 if (to > inode->i_size) {
1360 truncate_pagecache(inode, inode->i_size);
1361 truncate_blocks(inode, inode->i_size, true);
1365 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1366 loff_t pos, unsigned len, unsigned flags,
1367 struct page **pagep, void **fsdata)
1369 struct inode *inode = mapping->host;
1370 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1371 struct page *page = NULL;
1373 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1374 struct dnode_of_data dn;
1377 trace_f2fs_write_begin(inode, pos, len, flags);
1379 f2fs_balance_fs(sbi);
1382 * We should check this at this moment to avoid deadlock on inode page
1383 * and #0 page. The locking rule for inline_data conversion should be:
1384 * lock_page(page #0) -> lock_page(inode_page)
1387 err = f2fs_convert_inline_inode(inode);
1392 page = grab_cache_page_write_begin(mapping, index, flags);
1402 /* check inline_data */
1403 ipage = get_node_page(sbi, inode->i_ino);
1404 if (IS_ERR(ipage)) {
1405 err = PTR_ERR(ipage);
1409 set_new_dnode(&dn, inode, ipage, ipage, 0);
1411 if (f2fs_has_inline_data(inode)) {
1412 if (pos + len <= MAX_INLINE_DATA) {
1413 read_inline_data(page, ipage);
1414 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1415 sync_inode_page(&dn);
1418 err = f2fs_convert_inline_page(&dn, page);
1423 err = f2fs_get_block(&dn, index);
1427 f2fs_put_dnode(&dn);
1428 f2fs_unlock_op(sbi);
1430 f2fs_wait_on_page_writeback(page, DATA);
1432 if (len == PAGE_CACHE_SIZE)
1434 if (PageUptodate(page))
1437 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1438 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1439 unsigned end = start + len;
1441 /* Reading beyond i_size is simple: memset to zero */
1442 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1446 if (dn.data_blkaddr == NEW_ADDR) {
1447 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1449 struct f2fs_io_info fio = {
1453 .blk_addr = dn.data_blkaddr,
1455 .encrypted_page = NULL,
1457 err = f2fs_submit_page_bio(&fio);
1462 if (unlikely(!PageUptodate(page))) {
1466 if (unlikely(page->mapping != mapping)) {
1467 f2fs_put_page(page, 1);
1471 /* avoid symlink page */
1472 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1473 err = f2fs_decrypt_one(inode, page);
1479 SetPageUptodate(page);
1481 clear_cold_data(page);
1485 f2fs_put_dnode(&dn);
1487 f2fs_unlock_op(sbi);
1489 f2fs_put_page(page, 1);
1490 f2fs_write_failed(mapping, pos + len);
1494 static int f2fs_write_end(struct file *file,
1495 struct address_space *mapping,
1496 loff_t pos, unsigned len, unsigned copied,
1497 struct page *page, void *fsdata)
1499 struct inode *inode = page->mapping->host;
1501 trace_f2fs_write_end(inode, pos, len, copied);
1503 set_page_dirty(page);
1505 if (pos + copied > i_size_read(inode)) {
1506 i_size_write(inode, pos + copied);
1507 mark_inode_dirty(inode);
1508 update_inode_page(inode);
1511 f2fs_put_page(page, 1);
1515 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1518 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1520 if (offset & blocksize_mask)
1523 if (iov_iter_alignment(iter) & blocksize_mask)
1529 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1532 struct file *file = iocb->ki_filp;
1533 struct address_space *mapping = file->f_mapping;
1534 struct inode *inode = mapping->host;
1535 size_t count = iov_iter_count(iter);
1538 /* we don't need to use inline_data strictly */
1539 if (f2fs_has_inline_data(inode)) {
1540 err = f2fs_convert_inline_inode(inode);
1545 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1548 err = check_direct_IO(inode, iter, offset);
1552 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1554 if (iov_iter_rw(iter) == WRITE)
1555 __allocate_data_blocks(inode, offset, count);
1557 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1558 if (err < 0 && iov_iter_rw(iter) == WRITE)
1559 f2fs_write_failed(mapping, offset + count);
1561 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1566 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1567 unsigned int length)
1569 struct inode *inode = page->mapping->host;
1570 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1572 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1573 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1576 if (PageDirty(page)) {
1577 if (inode->i_ino == F2FS_META_INO(sbi))
1578 dec_page_count(sbi, F2FS_DIRTY_META);
1579 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1580 dec_page_count(sbi, F2FS_DIRTY_NODES);
1582 inode_dec_dirty_pages(inode);
1585 /* This is atomic written page, keep Private */
1586 if (IS_ATOMIC_WRITTEN_PAGE(page))
1589 ClearPagePrivate(page);
1592 int f2fs_release_page(struct page *page, gfp_t wait)
1594 /* If this is dirty page, keep PagePrivate */
1595 if (PageDirty(page))
1598 /* This is atomic written page, keep Private */
1599 if (IS_ATOMIC_WRITTEN_PAGE(page))
1602 ClearPagePrivate(page);
1606 static int f2fs_set_data_page_dirty(struct page *page)
1608 struct address_space *mapping = page->mapping;
1609 struct inode *inode = mapping->host;
1611 trace_f2fs_set_page_dirty(page, DATA);
1613 SetPageUptodate(page);
1615 if (f2fs_is_atomic_file(inode)) {
1616 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1617 register_inmem_page(inode, page);
1621 * Previously, this page has been registered, we just
1627 if (!PageDirty(page)) {
1628 __set_page_dirty_nobuffers(page);
1629 update_dirty_page(inode, page);
1635 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1637 struct inode *inode = mapping->host;
1639 /* we don't need to use inline_data strictly */
1640 if (f2fs_has_inline_data(inode)) {
1641 int err = f2fs_convert_inline_inode(inode);
1645 return generic_block_bmap(mapping, block, get_data_block_bmap);
1648 const struct address_space_operations f2fs_dblock_aops = {
1649 .readpage = f2fs_read_data_page,
1650 .readpages = f2fs_read_data_pages,
1651 .writepage = f2fs_write_data_page,
1652 .writepages = f2fs_write_data_pages,
1653 .write_begin = f2fs_write_begin,
1654 .write_end = f2fs_write_end,
1655 .set_page_dirty = f2fs_set_data_page_dirty,
1656 .invalidatepage = f2fs_invalidate_page,
1657 .releasepage = f2fs_release_page,
1658 .direct_IO = f2fs_direct_IO,