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 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->blk_addr,
151 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
154 trace_f2fs_submit_page_bio(page, fio);
155 f2fs_trace_ios(fio, 0);
157 /* Allocate a new bio */
158 bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
160 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
165 submit_bio(fio->rw, bio);
169 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
171 struct f2fs_sb_info *sbi = fio->sbi;
172 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
173 struct f2fs_bio_info *io;
174 bool is_read = is_read_io(fio->rw);
175 struct page *bio_page;
177 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
179 verify_block_addr(fio, fio->blk_addr);
181 down_write(&io->io_rwsem);
184 inc_page_count(sbi, F2FS_WRITEBACK);
186 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
187 io->fio.rw != fio->rw))
188 __submit_merged_bio(io);
190 if (io->bio == NULL) {
191 int bio_blocks = MAX_BIO_BLOCKS(sbi);
193 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
197 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
199 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
201 __submit_merged_bio(io);
205 io->last_block_in_bio = fio->blk_addr;
206 f2fs_trace_ios(fio, 0);
208 up_write(&io->io_rwsem);
209 trace_f2fs_submit_page_mbio(fio->page, fio);
213 * Lock ordering for the change of data block address:
216 * update block addresses in the node page
218 void set_data_blkaddr(struct dnode_of_data *dn)
220 struct f2fs_node *rn;
222 struct page *node_page = dn->node_page;
223 unsigned int ofs_in_node = dn->ofs_in_node;
225 f2fs_wait_on_page_writeback(node_page, NODE);
227 rn = F2FS_NODE(node_page);
229 /* Get physical address of data block */
230 addr_array = blkaddr_in_node(rn);
231 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
232 set_page_dirty(node_page);
235 int reserve_new_block(struct dnode_of_data *dn)
237 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
239 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
241 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
244 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
246 dn->data_blkaddr = NEW_ADDR;
247 set_data_blkaddr(dn);
248 mark_inode_dirty(dn->inode);
253 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
255 bool need_put = dn->inode_page ? false : true;
258 err = get_dnode_of_data(dn, index, ALLOC_NODE);
262 if (dn->data_blkaddr == NULL_ADDR)
263 err = reserve_new_block(dn);
269 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
271 struct extent_info ei;
272 struct inode *inode = dn->inode;
274 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
275 dn->data_blkaddr = ei.blk + index - ei.fofs;
279 return f2fs_reserve_block(dn, index);
282 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
283 int rw, bool for_write)
285 struct address_space *mapping = inode->i_mapping;
286 struct dnode_of_data dn;
288 struct extent_info ei;
290 struct f2fs_io_info fio = {
291 .sbi = F2FS_I_SB(inode),
294 .encrypted_page = NULL,
297 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
298 return read_mapping_page(mapping, index, NULL);
300 page = f2fs_grab_cache_page(mapping, index, for_write);
302 return ERR_PTR(-ENOMEM);
304 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
305 dn.data_blkaddr = ei.blk + index - ei.fofs;
309 set_new_dnode(&dn, inode, NULL, NULL, 0);
310 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
315 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
320 if (PageUptodate(page)) {
326 * A new dentry page is allocated but not able to be written, since its
327 * new inode page couldn't be allocated due to -ENOSPC.
328 * In such the case, its blkaddr can be remained as NEW_ADDR.
329 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
331 if (dn.data_blkaddr == NEW_ADDR) {
332 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
333 SetPageUptodate(page);
338 fio.blk_addr = dn.data_blkaddr;
340 err = f2fs_submit_page_bio(&fio);
346 f2fs_put_page(page, 1);
350 struct page *find_data_page(struct inode *inode, pgoff_t index)
352 struct address_space *mapping = inode->i_mapping;
355 page = find_get_page(mapping, index);
356 if (page && PageUptodate(page))
358 f2fs_put_page(page, 0);
360 page = get_read_data_page(inode, index, READ_SYNC, false);
364 if (PageUptodate(page))
367 wait_on_page_locked(page);
368 if (unlikely(!PageUptodate(page))) {
369 f2fs_put_page(page, 0);
370 return ERR_PTR(-EIO);
376 * If it tries to access a hole, return an error.
377 * Because, the callers, functions in dir.c and GC, should be able to know
378 * whether this page exists or not.
380 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
383 struct address_space *mapping = inode->i_mapping;
386 page = get_read_data_page(inode, index, READ_SYNC, for_write);
390 /* wait for read completion */
392 if (unlikely(!PageUptodate(page))) {
393 f2fs_put_page(page, 1);
394 return ERR_PTR(-EIO);
396 if (unlikely(page->mapping != mapping)) {
397 f2fs_put_page(page, 1);
404 * Caller ensures that this data page is never allocated.
405 * A new zero-filled data page is allocated in the page cache.
407 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
409 * Note that, ipage is set only by make_empty_dir, and if any error occur,
410 * ipage should be released by this function.
412 struct page *get_new_data_page(struct inode *inode,
413 struct page *ipage, pgoff_t index, bool new_i_size)
415 struct address_space *mapping = inode->i_mapping;
417 struct dnode_of_data dn;
420 page = f2fs_grab_cache_page(mapping, index, true);
423 * before exiting, we should make sure ipage will be released
424 * if any error occur.
426 f2fs_put_page(ipage, 1);
427 return ERR_PTR(-ENOMEM);
430 set_new_dnode(&dn, inode, ipage, NULL, 0);
431 err = f2fs_reserve_block(&dn, index);
433 f2fs_put_page(page, 1);
439 if (PageUptodate(page))
442 if (dn.data_blkaddr == NEW_ADDR) {
443 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
444 SetPageUptodate(page);
446 f2fs_put_page(page, 1);
448 page = get_read_data_page(inode, index, READ_SYNC, true);
452 /* wait for read completion */
456 if (new_i_size && i_size_read(inode) <
457 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
458 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
459 /* Only the directory inode sets new_i_size */
460 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
465 static int __allocate_data_block(struct dnode_of_data *dn)
467 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
468 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
469 struct f2fs_summary sum;
471 int seg = CURSEG_WARM_DATA;
474 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
477 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
478 if (dn->data_blkaddr == NEW_ADDR)
481 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
485 get_node_info(sbi, dn->nid, &ni);
486 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
488 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
489 seg = CURSEG_DIRECT_IO;
491 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
493 set_data_blkaddr(dn);
496 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
498 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
499 i_size_write(dn->inode,
500 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
502 /* direct IO doesn't use extent cache to maximize the performance */
503 f2fs_drop_largest_extent(dn->inode, fofs);
508 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
511 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
512 struct dnode_of_data dn;
513 u64 start = F2FS_BYTES_TO_BLK(offset);
514 u64 len = F2FS_BYTES_TO_BLK(count);
519 f2fs_balance_fs(sbi);
522 /* When reading holes, we need its node page */
523 set_new_dnode(&dn, inode, NULL, NULL, 0);
524 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
528 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
530 while (dn.ofs_in_node < end_offset && len) {
533 if (unlikely(f2fs_cp_error(sbi)))
536 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
537 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
538 if (__allocate_data_block(&dn))
548 sync_inode_page(&dn);
557 sync_inode_page(&dn);
565 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
566 * f2fs_map_blocks structure.
567 * If original data blocks are allocated, then give them to blockdev.
569 * a. preallocate requested block addresses
570 * b. do not use extent cache for better performance
571 * c. give the block addresses to blockdev
573 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
574 int create, int flag)
576 unsigned int maxblocks = map->m_len;
577 struct dnode_of_data dn;
578 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
579 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
580 pgoff_t pgofs, end_offset;
581 int err = 0, ofs = 1;
582 struct extent_info ei;
583 bool allocated = false;
588 /* it only supports block size == page size */
589 pgofs = (pgoff_t)map->m_lblk;
591 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
592 map->m_pblk = ei.blk + pgofs - ei.fofs;
593 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
594 map->m_flags = F2FS_MAP_MAPPED;
599 f2fs_lock_op(F2FS_I_SB(inode));
601 /* When reading holes, we need its node page */
602 set_new_dnode(&dn, inode, NULL, NULL, 0);
603 err = get_dnode_of_data(&dn, pgofs, mode);
610 if (__is_valid_data_blkaddr(dn.data_blkaddr) &&
611 !f2fs_is_valid_blkaddr(sbi, dn.data_blkaddr, DATA_GENERIC)) {
616 if (!is_valid_data_blkaddr(sbi, dn.data_blkaddr)) {
618 if (unlikely(f2fs_cp_error(sbi))) {
622 err = __allocate_data_block(&dn);
626 map->m_flags = F2FS_MAP_NEW;
628 if (flag != F2FS_GET_BLOCK_FIEMAP ||
629 dn.data_blkaddr != NEW_ADDR) {
630 if (flag == F2FS_GET_BLOCK_BMAP)
636 * preallocated unwritten block should be mapped
639 if (dn.data_blkaddr == NEW_ADDR)
640 map->m_flags = F2FS_MAP_UNWRITTEN;
644 map->m_flags |= F2FS_MAP_MAPPED;
645 map->m_pblk = dn.data_blkaddr;
648 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
653 if (dn.ofs_in_node >= end_offset) {
655 sync_inode_page(&dn);
659 set_new_dnode(&dn, inode, NULL, NULL, 0);
660 err = get_dnode_of_data(&dn, pgofs, mode);
667 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
670 if (maxblocks > map->m_len) {
671 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
673 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
675 if (unlikely(f2fs_cp_error(sbi))) {
679 err = __allocate_data_block(&dn);
683 map->m_flags |= F2FS_MAP_NEW;
684 blkaddr = dn.data_blkaddr;
687 * we only merge preallocated unwritten blocks
690 if (flag != F2FS_GET_BLOCK_FIEMAP ||
696 /* Give more consecutive addresses for the readahead */
697 if ((map->m_pblk != NEW_ADDR &&
698 blkaddr == (map->m_pblk + ofs)) ||
699 (map->m_pblk == NEW_ADDR &&
700 blkaddr == NEW_ADDR)) {
710 sync_inode_page(&dn);
715 f2fs_unlock_op(F2FS_I_SB(inode));
717 trace_f2fs_map_blocks(inode, map, err);
721 static int __get_data_block(struct inode *inode, sector_t iblock,
722 struct buffer_head *bh, int create, int flag)
724 struct f2fs_map_blocks map;
728 map.m_len = bh->b_size >> inode->i_blkbits;
730 ret = f2fs_map_blocks(inode, &map, create, flag);
732 map_bh(bh, inode->i_sb, map.m_pblk);
733 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
734 bh->b_size = (u64)map.m_len << inode->i_blkbits;
739 static int get_data_block(struct inode *inode, sector_t iblock,
740 struct buffer_head *bh_result, int create, int flag)
742 return __get_data_block(inode, iblock, bh_result, create, flag);
745 static int get_data_block_dio(struct inode *inode, sector_t iblock,
746 struct buffer_head *bh_result, int create)
748 return __get_data_block(inode, iblock, bh_result, create,
752 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
753 struct buffer_head *bh_result, int create)
755 return __get_data_block(inode, iblock, bh_result, create,
756 F2FS_GET_BLOCK_BMAP);
759 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
761 return (offset >> inode->i_blkbits);
764 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
766 return (blk << inode->i_blkbits);
769 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
772 struct buffer_head map_bh;
773 sector_t start_blk, last_blk;
774 loff_t isize = i_size_read(inode);
775 u64 logical = 0, phys = 0, size = 0;
777 bool past_eof = false, whole_file = false;
780 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
784 if (f2fs_has_inline_data(inode)) {
785 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
790 mutex_lock(&inode->i_mutex);
797 if (logical_to_blk(inode, len) == 0)
798 len = blk_to_logical(inode, 1);
800 start_blk = logical_to_blk(inode, start);
801 last_blk = logical_to_blk(inode, start + len - 1);
803 memset(&map_bh, 0, sizeof(struct buffer_head));
806 ret = get_data_block(inode, start_blk, &map_bh, 0,
807 F2FS_GET_BLOCK_FIEMAP);
812 if (!buffer_mapped(&map_bh)) {
815 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
818 if (past_eof && size) {
819 flags |= FIEMAP_EXTENT_LAST;
820 ret = fiemap_fill_next_extent(fieinfo, logical,
823 ret = fiemap_fill_next_extent(fieinfo, logical,
828 /* if we have holes up to/past EOF then we're done */
829 if (start_blk > last_blk || past_eof || ret)
832 if (start_blk > last_blk && !whole_file) {
833 ret = fiemap_fill_next_extent(fieinfo, logical,
839 * if size != 0 then we know we already have an extent
843 ret = fiemap_fill_next_extent(fieinfo, logical,
849 logical = blk_to_logical(inode, start_blk);
850 phys = blk_to_logical(inode, map_bh.b_blocknr);
851 size = map_bh.b_size;
853 if (buffer_unwritten(&map_bh))
854 flags = FIEMAP_EXTENT_UNWRITTEN;
856 start_blk += logical_to_blk(inode, size);
859 * If we are past the EOF, then we need to make sure as
860 * soon as we find a hole that the last extent we found
861 * is marked with FIEMAP_EXTENT_LAST
863 if (!past_eof && logical + size >= isize)
867 if (fatal_signal_pending(current))
875 mutex_unlock(&inode->i_mutex);
879 struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
882 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
883 struct f2fs_crypto_ctx *ctx = NULL;
884 struct block_device *bdev = sbi->sb->s_bdev;
887 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
888 return ERR_PTR(-EFAULT);
890 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
891 ctx = f2fs_get_crypto_ctx(inode);
893 return ERR_CAST(ctx);
895 /* wait the page to be moved by cleaning */
896 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
899 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
902 f2fs_release_crypto_ctx(ctx);
903 return ERR_PTR(-ENOMEM);
906 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
907 bio->bi_end_io = f2fs_read_end_io;
908 bio->bi_private = ctx;
914 * This function was originally taken from fs/mpage.c, and customized for f2fs.
915 * Major change was from block_size == page_size in f2fs by default.
917 static int f2fs_mpage_readpages(struct address_space *mapping,
918 struct list_head *pages, struct page *page,
921 struct bio *bio = NULL;
923 sector_t last_block_in_bio = 0;
924 struct inode *inode = mapping->host;
925 const unsigned blkbits = inode->i_blkbits;
926 const unsigned blocksize = 1 << blkbits;
927 sector_t block_in_file;
929 sector_t last_block_in_file;
931 struct f2fs_map_blocks map;
938 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
940 prefetchw(&page->flags);
942 page = list_entry(pages->prev, struct page, lru);
943 list_del(&page->lru);
944 if (add_to_page_cache_lru(page, mapping,
945 page->index, GFP_KERNEL))
949 block_in_file = (sector_t)page->index;
950 last_block = block_in_file + nr_pages;
951 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
953 if (last_block > last_block_in_file)
954 last_block = last_block_in_file;
957 * Map blocks using the previous result first.
959 if ((map.m_flags & F2FS_MAP_MAPPED) &&
960 block_in_file > map.m_lblk &&
961 block_in_file < (map.m_lblk + map.m_len))
965 * Then do more f2fs_map_blocks() calls until we are
966 * done with this page.
970 if (block_in_file < last_block) {
971 map.m_lblk = block_in_file;
972 map.m_len = last_block - block_in_file;
974 if (f2fs_map_blocks(inode, &map, 0,
975 F2FS_GET_BLOCK_READ))
979 if ((map.m_flags & F2FS_MAP_MAPPED)) {
980 block_nr = map.m_pblk + block_in_file - map.m_lblk;
981 SetPageMappedToDisk(page);
983 if (!PageUptodate(page) && !cleancache_get_page(page)) {
984 SetPageUptodate(page);
988 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
992 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
993 SetPageUptodate(page);
999 * This page will go to BIO. Do we need to send this
1002 if (bio && (last_block_in_bio != block_nr - 1)) {
1004 submit_bio(READ, bio);
1008 bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1010 goto set_error_page;
1013 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1014 goto submit_and_realloc;
1016 last_block_in_bio = block_nr;
1020 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1025 submit_bio(READ, bio);
1031 page_cache_release(page);
1033 BUG_ON(pages && !list_empty(pages));
1035 submit_bio(READ, bio);
1039 static int f2fs_read_data_page(struct file *file, struct page *page)
1041 struct inode *inode = page->mapping->host;
1044 trace_f2fs_readpage(page, DATA);
1046 /* If the file has inline data, try to read it directly */
1047 if (f2fs_has_inline_data(inode))
1048 ret = f2fs_read_inline_data(inode, page);
1050 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1054 static int f2fs_read_data_pages(struct file *file,
1055 struct address_space *mapping,
1056 struct list_head *pages, unsigned nr_pages)
1058 struct inode *inode = file->f_mapping->host;
1059 struct page *page = list_entry(pages->prev, struct page, lru);
1061 trace_f2fs_readpages(inode, page, nr_pages);
1063 /* If the file has inline data, skip readpages */
1064 if (f2fs_has_inline_data(inode))
1067 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1070 int do_write_data_page(struct f2fs_io_info *fio)
1072 struct page *page = fio->page;
1073 struct inode *inode = page->mapping->host;
1074 struct dnode_of_data dn;
1077 set_new_dnode(&dn, inode, NULL, NULL, 0);
1078 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1082 fio->blk_addr = dn.data_blkaddr;
1084 /* This page is already truncated */
1085 if (fio->blk_addr == NULL_ADDR) {
1086 ClearPageUptodate(page);
1090 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1092 /* wait for GCed encrypted page writeback */
1093 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1096 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1097 if (IS_ERR(fio->encrypted_page)) {
1098 err = PTR_ERR(fio->encrypted_page);
1103 set_page_writeback(page);
1105 if (__is_valid_data_blkaddr(fio->blk_addr) &&
1106 !f2fs_is_valid_blkaddr(fio->sbi, fio->blk_addr,
1112 * If current allocation needs SSR,
1113 * it had better in-place writes for updated data.
1115 if (unlikely(is_valid_data_blkaddr(fio->sbi, fio->blk_addr) &&
1116 !is_cold_data(page) &&
1117 need_inplace_update(inode))) {
1118 rewrite_data_page(fio);
1119 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1120 trace_f2fs_do_write_data_page(page, IPU);
1122 write_data_page(&dn, fio);
1123 set_data_blkaddr(&dn);
1124 f2fs_update_extent_cache(&dn);
1125 trace_f2fs_do_write_data_page(page, OPU);
1126 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1127 if (page->index == 0)
1128 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1131 f2fs_put_dnode(&dn);
1135 static int f2fs_write_data_page(struct page *page,
1136 struct writeback_control *wbc)
1138 struct inode *inode = page->mapping->host;
1139 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1140 loff_t i_size = i_size_read(inode);
1141 const pgoff_t end_index = ((unsigned long long) i_size)
1142 >> PAGE_CACHE_SHIFT;
1143 unsigned offset = 0;
1144 bool need_balance_fs = false;
1146 struct f2fs_io_info fio = {
1149 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1151 .encrypted_page = NULL,
1154 trace_f2fs_writepage(page, DATA);
1156 if (page->index < end_index)
1160 * If the offset is out-of-range of file size,
1161 * this page does not have to be written to disk.
1163 offset = i_size & (PAGE_CACHE_SIZE - 1);
1164 if ((page->index >= end_index + 1) || !offset)
1167 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1169 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1171 if (f2fs_is_drop_cache(inode))
1173 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1174 available_free_memory(sbi, BASE_CHECK))
1177 /* Dentry blocks are controlled by checkpoint */
1178 if (S_ISDIR(inode->i_mode)) {
1179 if (unlikely(f2fs_cp_error(sbi)))
1181 err = do_write_data_page(&fio);
1185 /* we should bypass data pages to proceed the kworkder jobs */
1186 if (unlikely(f2fs_cp_error(sbi))) {
1191 if (!wbc->for_reclaim)
1192 need_balance_fs = true;
1193 else if (has_not_enough_free_secs(sbi, 0))
1198 if (f2fs_has_inline_data(inode))
1199 err = f2fs_write_inline_data(inode, page);
1201 err = do_write_data_page(&fio);
1202 f2fs_unlock_op(sbi);
1204 if (err && err != -ENOENT)
1207 clear_cold_data(page);
1209 inode_dec_dirty_pages(inode);
1211 ClearPageUptodate(page);
1213 if (need_balance_fs)
1214 f2fs_balance_fs(sbi);
1215 if (wbc->for_reclaim)
1216 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1220 redirty_page_for_writepage(wbc, page);
1221 return AOP_WRITEPAGE_ACTIVATE;
1224 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1227 struct address_space *mapping = data;
1228 int ret = mapping->a_ops->writepage(page, wbc);
1229 mapping_set_error(mapping, ret);
1234 * This function was copied from write_cche_pages from mm/page-writeback.c.
1235 * The major change is making write step of cold data page separately from
1236 * warm/hot data page.
1238 static int f2fs_write_cache_pages(struct address_space *mapping,
1239 struct writeback_control *wbc, writepage_t writepage,
1244 struct pagevec pvec;
1246 pgoff_t uninitialized_var(writeback_index);
1248 pgoff_t end; /* Inclusive */
1251 int range_whole = 0;
1255 pagevec_init(&pvec, 0);
1257 if (wbc->range_cyclic) {
1258 writeback_index = mapping->writeback_index; /* prev offset */
1259 index = writeback_index;
1266 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1267 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1268 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1270 cycled = 1; /* ignore range_cyclic tests */
1272 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1273 tag = PAGECACHE_TAG_TOWRITE;
1275 tag = PAGECACHE_TAG_DIRTY;
1277 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1278 tag_pages_for_writeback(mapping, index, end);
1280 while (!done && (index <= end)) {
1283 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1284 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1288 for (i = 0; i < nr_pages; i++) {
1289 struct page *page = pvec.pages[i];
1291 if (page->index > end) {
1296 done_index = page->index;
1300 if (unlikely(page->mapping != mapping)) {
1306 if (!PageDirty(page)) {
1307 /* someone wrote it for us */
1308 goto continue_unlock;
1311 if (step == is_cold_data(page))
1312 goto continue_unlock;
1314 if (PageWriteback(page)) {
1315 if (wbc->sync_mode != WB_SYNC_NONE)
1316 f2fs_wait_on_page_writeback(page, DATA);
1318 goto continue_unlock;
1321 BUG_ON(PageWriteback(page));
1322 if (!clear_page_dirty_for_io(page))
1323 goto continue_unlock;
1325 ret = (*writepage)(page, wbc, data);
1326 if (unlikely(ret)) {
1327 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1331 done_index = page->index + 1;
1337 if (--wbc->nr_to_write <= 0 &&
1338 wbc->sync_mode == WB_SYNC_NONE) {
1343 pagevec_release(&pvec);
1352 if (!cycled && !done) {
1355 end = writeback_index - 1;
1358 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1359 mapping->writeback_index = done_index;
1364 static int f2fs_write_data_pages(struct address_space *mapping,
1365 struct writeback_control *wbc)
1367 struct inode *inode = mapping->host;
1368 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1369 bool locked = false;
1373 trace_f2fs_writepages(mapping->host, wbc, DATA);
1375 /* deal with chardevs and other special file */
1376 if (!mapping->a_ops->writepage)
1379 /* skip writing if there is no dirty page in this inode */
1380 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1383 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1384 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1385 available_free_memory(sbi, DIRTY_DENTS))
1388 /* during POR, we don't need to trigger writepage at all. */
1389 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1392 diff = nr_pages_to_write(sbi, DATA, wbc);
1394 if (!S_ISDIR(inode->i_mode)) {
1395 mutex_lock(&sbi->writepages);
1398 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1399 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1401 mutex_unlock(&sbi->writepages);
1403 remove_dirty_dir_inode(inode);
1405 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1409 wbc->pages_skipped += get_dirty_pages(inode);
1413 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1415 struct inode *inode = mapping->host;
1417 if (to > inode->i_size) {
1418 truncate_pagecache(inode, inode->i_size);
1419 truncate_blocks(inode, inode->i_size, true);
1423 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1424 loff_t pos, unsigned len, unsigned flags,
1425 struct page **pagep, void **fsdata)
1427 struct inode *inode = mapping->host;
1428 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1429 struct page *page = NULL;
1431 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1432 struct dnode_of_data dn;
1435 trace_f2fs_write_begin(inode, pos, len, flags);
1437 f2fs_balance_fs(sbi);
1440 * We should check this at this moment to avoid deadlock on inode page
1441 * and #0 page. The locking rule for inline_data conversion should be:
1442 * lock_page(page #0) -> lock_page(inode_page)
1445 err = f2fs_convert_inline_inode(inode);
1451 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1452 * wait_for_stable_page. Will wait that below with our IO control.
1454 page = pagecache_get_page(mapping, index,
1455 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1465 /* check inline_data */
1466 ipage = get_node_page(sbi, inode->i_ino);
1467 if (IS_ERR(ipage)) {
1468 err = PTR_ERR(ipage);
1472 set_new_dnode(&dn, inode, ipage, ipage, 0);
1474 if (f2fs_has_inline_data(inode)) {
1475 if (pos + len <= MAX_INLINE_DATA) {
1476 read_inline_data(page, ipage);
1477 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1478 sync_inode_page(&dn);
1481 err = f2fs_convert_inline_page(&dn, page);
1486 err = f2fs_get_block(&dn, index);
1490 f2fs_put_dnode(&dn);
1491 f2fs_unlock_op(sbi);
1493 f2fs_wait_on_page_writeback(page, DATA);
1495 /* wait for GCed encrypted page writeback */
1496 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1497 f2fs_wait_on_encrypted_page_writeback(sbi, dn.data_blkaddr);
1499 if (len == PAGE_CACHE_SIZE)
1501 if (PageUptodate(page))
1504 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1505 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1506 unsigned end = start + len;
1508 /* Reading beyond i_size is simple: memset to zero */
1509 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1513 if (dn.data_blkaddr == NEW_ADDR) {
1514 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1518 bio = f2fs_grab_bio(inode, dn.data_blkaddr, 1);
1524 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
1530 submit_bio(READ_SYNC, bio);
1533 if (unlikely(!PageUptodate(page))) {
1537 if (unlikely(page->mapping != mapping)) {
1538 f2fs_put_page(page, 1);
1543 SetPageUptodate(page);
1545 clear_cold_data(page);
1549 f2fs_put_dnode(&dn);
1551 f2fs_unlock_op(sbi);
1553 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, struct iov_iter *iter,
1582 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1584 if (offset & blocksize_mask)
1587 if (iov_iter_alignment(iter) & blocksize_mask)
1593 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1596 struct file *file = iocb->ki_filp;
1597 struct address_space *mapping = file->f_mapping;
1598 struct inode *inode = mapping->host;
1599 size_t count = iov_iter_count(iter);
1602 /* we don't need to use inline_data strictly */
1603 if (f2fs_has_inline_data(inode)) {
1604 err = f2fs_convert_inline_inode(inode);
1609 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1612 err = check_direct_IO(inode, iter, offset);
1616 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1618 if (iov_iter_rw(iter) == WRITE) {
1619 __allocate_data_blocks(inode, offset, count);
1620 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1626 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1628 if (err < 0 && iov_iter_rw(iter) == WRITE)
1629 f2fs_write_failed(mapping, offset + count);
1631 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1636 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1637 unsigned int length)
1639 struct inode *inode = page->mapping->host;
1640 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1642 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1643 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1646 if (PageDirty(page)) {
1647 if (inode->i_ino == F2FS_META_INO(sbi))
1648 dec_page_count(sbi, F2FS_DIRTY_META);
1649 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1650 dec_page_count(sbi, F2FS_DIRTY_NODES);
1652 inode_dec_dirty_pages(inode);
1655 /* This is atomic written page, keep Private */
1656 if (IS_ATOMIC_WRITTEN_PAGE(page))
1659 ClearPagePrivate(page);
1662 int f2fs_release_page(struct page *page, gfp_t wait)
1664 /* If this is dirty page, keep PagePrivate */
1665 if (PageDirty(page))
1668 /* This is atomic written page, keep Private */
1669 if (IS_ATOMIC_WRITTEN_PAGE(page))
1672 ClearPagePrivate(page);
1676 static int f2fs_set_data_page_dirty(struct page *page)
1678 struct address_space *mapping = page->mapping;
1679 struct inode *inode = mapping->host;
1681 trace_f2fs_set_page_dirty(page, DATA);
1683 SetPageUptodate(page);
1685 if (f2fs_is_atomic_file(inode)) {
1686 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1687 register_inmem_page(inode, page);
1691 * Previously, this page has been registered, we just
1697 if (!PageDirty(page)) {
1698 __set_page_dirty_nobuffers(page);
1699 update_dirty_page(inode, page);
1705 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1707 struct inode *inode = mapping->host;
1709 if (f2fs_has_inline_data(inode))
1712 /* make sure allocating whole blocks */
1713 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1714 filemap_write_and_wait(mapping);
1716 return generic_block_bmap(mapping, block, get_data_block_bmap);
1719 const struct address_space_operations f2fs_dblock_aops = {
1720 .readpage = f2fs_read_data_page,
1721 .readpages = f2fs_read_data_pages,
1722 .writepage = f2fs_write_data_page,
1723 .writepages = f2fs_write_data_pages,
1724 .write_begin = f2fs_write_begin,
1725 .write_end = f2fs_write_end,
1726 .set_page_dirty = f2fs_set_data_page_dirty,
1727 .invalidatepage = f2fs_invalidate_page,
1728 .releasepage = f2fs_release_page,
1729 .direct_IO = f2fs_direct_IO,