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 fscrypt_release_ctx(bio->bi_private);
39 fscrypt_decrypt_bio_pages(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 fscrypt_pullback_bio_page(&page, true);
69 if (unlikely(bio->bi_error)) {
70 set_bit(AS_EIO, &page->mapping->flags);
71 f2fs_stop_checkpoint(sbi, true);
73 end_page_writeback(page);
75 if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
76 wq_has_sleeper(&sbi->cp_wait))
77 wake_up(&sbi->cp_wait);
83 * Low-level block read/write IO operations.
85 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
86 int npages, bool is_read)
90 bio = f2fs_bio_alloc(npages);
92 bio->bi_bdev = sbi->sb->s_bdev;
93 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
94 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
95 bio->bi_private = is_read ? NULL : sbi;
100 static inline void __submit_bio(struct f2fs_sb_info *sbi, int rw,
104 atomic_inc(&sbi->nr_wb_bios);
108 static void __submit_merged_bio(struct f2fs_bio_info *io)
110 struct f2fs_io_info *fio = &io->fio;
115 if (is_read_io(fio->rw))
116 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
118 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
120 __submit_bio(io->sbi, fio->rw, io->bio);
124 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
125 struct page *page, nid_t ino)
127 struct bio_vec *bvec;
134 if (!inode && !page && !ino)
137 bio_for_each_segment_all(bvec, io->bio, i) {
139 if (bvec->bv_page->mapping)
140 target = bvec->bv_page;
142 target = fscrypt_control_page(bvec->bv_page);
144 if (inode && inode == target->mapping->host)
146 if (page && page == target)
148 if (ino && ino == ino_of_node(target))
155 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
156 struct page *page, nid_t ino,
159 enum page_type btype = PAGE_TYPE_OF_BIO(type);
160 struct f2fs_bio_info *io = &sbi->write_io[btype];
163 down_read(&io->io_rwsem);
164 ret = __has_merged_page(io, inode, page, ino);
165 up_read(&io->io_rwsem);
169 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
170 struct inode *inode, struct page *page,
171 nid_t ino, enum page_type type, int rw)
173 enum page_type btype = PAGE_TYPE_OF_BIO(type);
174 struct f2fs_bio_info *io;
176 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
178 down_write(&io->io_rwsem);
180 if (!__has_merged_page(io, inode, page, ino))
183 /* change META to META_FLUSH in the checkpoint procedure */
184 if (type >= META_FLUSH) {
185 io->fio.type = META_FLUSH;
186 if (test_opt(sbi, NOBARRIER))
187 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
189 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
191 __submit_merged_bio(io);
193 up_write(&io->io_rwsem);
196 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
199 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
202 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
203 struct inode *inode, struct page *page,
204 nid_t ino, enum page_type type, int rw)
206 if (has_merged_page(sbi, inode, page, ino, type))
207 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
210 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
212 f2fs_submit_merged_bio(sbi, DATA, WRITE);
213 f2fs_submit_merged_bio(sbi, NODE, WRITE);
214 f2fs_submit_merged_bio(sbi, META, WRITE);
218 * Fill the locked page with data located in the block address.
219 * Return unlocked page.
221 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
224 struct page *page = fio->encrypted_page ?
225 fio->encrypted_page : fio->page;
227 trace_f2fs_submit_page_bio(page, fio);
228 f2fs_trace_ios(fio, 0);
230 /* Allocate a new bio */
231 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
233 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
238 __submit_bio(fio->sbi, fio->rw, bio);
242 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
244 struct f2fs_sb_info *sbi = fio->sbi;
245 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
246 struct f2fs_bio_info *io;
247 bool is_read = is_read_io(fio->rw);
248 struct page *bio_page;
250 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
252 if (fio->old_blkaddr != NEW_ADDR)
253 verify_block_addr(sbi, fio->old_blkaddr);
254 verify_block_addr(sbi, fio->new_blkaddr);
256 down_write(&io->io_rwsem);
258 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
259 io->fio.rw != fio->rw))
260 __submit_merged_bio(io);
262 if (io->bio == NULL) {
263 int bio_blocks = MAX_BIO_BLOCKS(sbi);
265 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
266 bio_blocks, is_read);
270 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
272 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
274 __submit_merged_bio(io);
278 io->last_block_in_bio = fio->new_blkaddr;
279 f2fs_trace_ios(fio, 0);
281 up_write(&io->io_rwsem);
282 trace_f2fs_submit_page_mbio(fio->page, fio);
285 static void __set_data_blkaddr(struct dnode_of_data *dn)
287 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
290 /* Get physical address of data block */
291 addr_array = blkaddr_in_node(rn);
292 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
296 * Lock ordering for the change of data block address:
299 * update block addresses in the node page
301 void set_data_blkaddr(struct dnode_of_data *dn)
303 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
304 __set_data_blkaddr(dn);
305 if (set_page_dirty(dn->node_page))
306 dn->node_changed = true;
309 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
311 dn->data_blkaddr = blkaddr;
312 set_data_blkaddr(dn);
313 f2fs_update_extent_cache(dn);
316 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
317 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
319 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
324 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
326 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
329 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
330 dn->ofs_in_node, count);
332 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
334 for (; count > 0; dn->ofs_in_node++) {
336 datablock_addr(dn->node_page, dn->ofs_in_node);
337 if (blkaddr == NULL_ADDR) {
338 dn->data_blkaddr = NEW_ADDR;
339 __set_data_blkaddr(dn);
344 if (set_page_dirty(dn->node_page))
345 dn->node_changed = true;
349 /* Should keep dn->ofs_in_node unchanged */
350 int reserve_new_block(struct dnode_of_data *dn)
352 unsigned int ofs_in_node = dn->ofs_in_node;
355 ret = reserve_new_blocks(dn, 1);
356 dn->ofs_in_node = ofs_in_node;
360 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
362 bool need_put = dn->inode_page ? false : true;
365 err = get_dnode_of_data(dn, index, ALLOC_NODE);
369 if (dn->data_blkaddr == NULL_ADDR)
370 err = reserve_new_block(dn);
376 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
378 struct extent_info ei;
379 struct inode *inode = dn->inode;
381 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
382 dn->data_blkaddr = ei.blk + index - ei.fofs;
386 return f2fs_reserve_block(dn, index);
389 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
390 int rw, bool for_write)
392 struct address_space *mapping = inode->i_mapping;
393 struct dnode_of_data dn;
395 struct extent_info ei;
397 struct f2fs_io_info fio = {
398 .sbi = F2FS_I_SB(inode),
401 .encrypted_page = NULL,
404 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
405 return read_mapping_page(mapping, index, NULL);
407 page = f2fs_grab_cache_page(mapping, index, for_write);
409 return ERR_PTR(-ENOMEM);
411 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
412 dn.data_blkaddr = ei.blk + index - ei.fofs;
416 set_new_dnode(&dn, inode, NULL, NULL, 0);
417 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
422 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
427 if (PageUptodate(page)) {
433 * A new dentry page is allocated but not able to be written, since its
434 * new inode page couldn't be allocated due to -ENOSPC.
435 * In such the case, its blkaddr can be remained as NEW_ADDR.
436 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
438 if (dn.data_blkaddr == NEW_ADDR) {
439 zero_user_segment(page, 0, PAGE_SIZE);
440 SetPageUptodate(page);
445 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
447 err = f2fs_submit_page_bio(&fio);
453 f2fs_put_page(page, 1);
457 struct page *find_data_page(struct inode *inode, pgoff_t index)
459 struct address_space *mapping = inode->i_mapping;
462 page = find_get_page(mapping, index);
463 if (page && PageUptodate(page))
465 f2fs_put_page(page, 0);
467 page = get_read_data_page(inode, index, READ_SYNC, false);
471 if (PageUptodate(page))
474 wait_on_page_locked(page);
475 if (unlikely(!PageUptodate(page))) {
476 f2fs_put_page(page, 0);
477 return ERR_PTR(-EIO);
483 * If it tries to access a hole, return an error.
484 * Because, the callers, functions in dir.c and GC, should be able to know
485 * whether this page exists or not.
487 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
490 struct address_space *mapping = inode->i_mapping;
493 page = get_read_data_page(inode, index, READ_SYNC, for_write);
497 /* wait for read completion */
499 if (unlikely(!PageUptodate(page))) {
500 f2fs_put_page(page, 1);
501 return ERR_PTR(-EIO);
503 if (unlikely(page->mapping != mapping)) {
504 f2fs_put_page(page, 1);
511 * Caller ensures that this data page is never allocated.
512 * A new zero-filled data page is allocated in the page cache.
514 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
516 * Note that, ipage is set only by make_empty_dir, and if any error occur,
517 * ipage should be released by this function.
519 struct page *get_new_data_page(struct inode *inode,
520 struct page *ipage, pgoff_t index, bool new_i_size)
522 struct address_space *mapping = inode->i_mapping;
524 struct dnode_of_data dn;
527 page = f2fs_grab_cache_page(mapping, index, true);
530 * before exiting, we should make sure ipage will be released
531 * if any error occur.
533 f2fs_put_page(ipage, 1);
534 return ERR_PTR(-ENOMEM);
537 set_new_dnode(&dn, inode, ipage, NULL, 0);
538 err = f2fs_reserve_block(&dn, index);
540 f2fs_put_page(page, 1);
546 if (PageUptodate(page))
549 if (dn.data_blkaddr == NEW_ADDR) {
550 zero_user_segment(page, 0, PAGE_SIZE);
551 SetPageUptodate(page);
553 f2fs_put_page(page, 1);
555 /* if ipage exists, blkaddr should be NEW_ADDR */
556 f2fs_bug_on(F2FS_I_SB(inode), ipage);
557 page = get_lock_data_page(inode, index, true);
562 if (new_i_size && i_size_read(inode) <
563 ((loff_t)(index + 1) << PAGE_SHIFT))
564 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
568 static int __allocate_data_block(struct dnode_of_data *dn)
570 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
571 struct f2fs_summary sum;
573 int seg = CURSEG_WARM_DATA;
577 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
580 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
581 if (dn->data_blkaddr == NEW_ADDR)
584 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
588 get_node_info(sbi, dn->nid, &ni);
589 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
591 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
592 seg = CURSEG_DIRECT_IO;
594 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
596 set_data_blkaddr(dn);
599 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
601 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
602 f2fs_i_size_write(dn->inode,
603 ((loff_t)(fofs + 1) << PAGE_SHIFT));
607 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
609 struct inode *inode = file_inode(iocb->ki_filp);
610 struct f2fs_map_blocks map;
613 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
614 map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
615 map.m_next_pgofs = NULL;
617 if (f2fs_encrypted_inode(inode))
620 if (iocb->ki_flags & IOCB_DIRECT) {
621 ret = f2fs_convert_inline_inode(inode);
624 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
626 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
627 ret = f2fs_convert_inline_inode(inode);
631 if (!f2fs_has_inline_data(inode))
632 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
637 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
638 * f2fs_map_blocks structure.
639 * If original data blocks are allocated, then give them to blockdev.
641 * a. preallocate requested block addresses
642 * b. do not use extent cache for better performance
643 * c. give the block addresses to blockdev
645 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
646 int create, int flag)
648 unsigned int maxblocks = map->m_len;
649 struct dnode_of_data dn;
650 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
651 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
652 pgoff_t pgofs, end_offset, end;
653 int err = 0, ofs = 1;
654 unsigned int ofs_in_node, last_ofs_in_node;
656 struct extent_info ei;
657 bool allocated = false;
663 /* it only supports block size == page size */
664 pgofs = (pgoff_t)map->m_lblk;
665 end = pgofs + maxblocks;
667 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
668 map->m_pblk = ei.blk + pgofs - ei.fofs;
669 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
670 map->m_flags = F2FS_MAP_MAPPED;
678 /* When reading holes, we need its node page */
679 set_new_dnode(&dn, inode, NULL, NULL, 0);
680 err = get_dnode_of_data(&dn, pgofs, mode);
682 if (flag == F2FS_GET_BLOCK_BMAP)
684 if (err == -ENOENT) {
686 if (map->m_next_pgofs)
688 get_next_page_offset(&dn, pgofs);
694 ofs_in_node = dn.ofs_in_node;
695 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
698 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
700 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
702 if (unlikely(f2fs_cp_error(sbi))) {
706 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
707 if (blkaddr == NULL_ADDR) {
709 last_ofs_in_node = dn.ofs_in_node;
712 err = __allocate_data_block(&dn);
714 set_inode_flag(inode, FI_APPEND_WRITE);
720 map->m_flags = F2FS_MAP_NEW;
721 blkaddr = dn.data_blkaddr;
723 if (flag == F2FS_GET_BLOCK_BMAP) {
727 if (flag == F2FS_GET_BLOCK_FIEMAP &&
728 blkaddr == NULL_ADDR) {
729 if (map->m_next_pgofs)
730 *map->m_next_pgofs = pgofs + 1;
732 if (flag != F2FS_GET_BLOCK_FIEMAP ||
738 if (flag == F2FS_GET_BLOCK_PRE_AIO)
741 if (map->m_len == 0) {
742 /* preallocated unwritten block should be mapped for fiemap. */
743 if (blkaddr == NEW_ADDR)
744 map->m_flags |= F2FS_MAP_UNWRITTEN;
745 map->m_flags |= F2FS_MAP_MAPPED;
747 map->m_pblk = blkaddr;
749 } else if ((map->m_pblk != NEW_ADDR &&
750 blkaddr == (map->m_pblk + ofs)) ||
751 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
752 flag == F2FS_GET_BLOCK_PRE_DIO) {
763 /* preallocate blocks in batch for one dnode page */
764 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
765 (pgofs == end || dn.ofs_in_node == end_offset)) {
767 dn.ofs_in_node = ofs_in_node;
768 err = reserve_new_blocks(&dn, prealloc);
772 map->m_len += dn.ofs_in_node - ofs_in_node;
773 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
777 dn.ofs_in_node = end_offset;
782 else if (dn.ofs_in_node < end_offset)
789 f2fs_balance_fs(sbi, allocated);
799 f2fs_balance_fs(sbi, allocated);
802 trace_f2fs_map_blocks(inode, map, err);
806 static int __get_data_block(struct inode *inode, sector_t iblock,
807 struct buffer_head *bh, int create, int flag,
810 struct f2fs_map_blocks map;
814 map.m_len = bh->b_size >> inode->i_blkbits;
815 map.m_next_pgofs = next_pgofs;
817 ret = f2fs_map_blocks(inode, &map, create, flag);
819 map_bh(bh, inode->i_sb, map.m_pblk);
820 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
821 bh->b_size = map.m_len << inode->i_blkbits;
826 static int get_data_block(struct inode *inode, sector_t iblock,
827 struct buffer_head *bh_result, int create, int flag,
830 return __get_data_block(inode, iblock, bh_result, create,
834 static int get_data_block_dio(struct inode *inode, sector_t iblock,
835 struct buffer_head *bh_result, int create)
837 return __get_data_block(inode, iblock, bh_result, create,
838 F2FS_GET_BLOCK_DIO, NULL);
841 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
842 struct buffer_head *bh_result, int create)
844 /* Block number less than F2FS MAX BLOCKS */
845 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
848 return __get_data_block(inode, iblock, bh_result, create,
849 F2FS_GET_BLOCK_BMAP, NULL);
852 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
854 return (offset >> inode->i_blkbits);
857 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
859 return (blk << inode->i_blkbits);
862 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
865 struct buffer_head map_bh;
866 sector_t start_blk, last_blk;
869 u64 logical = 0, phys = 0, size = 0;
873 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
877 if (f2fs_has_inline_data(inode)) {
878 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
885 isize = i_size_read(inode);
889 if (start + len > isize)
892 if (logical_to_blk(inode, len) == 0)
893 len = blk_to_logical(inode, 1);
895 start_blk = logical_to_blk(inode, start);
896 last_blk = logical_to_blk(inode, start + len - 1);
899 memset(&map_bh, 0, sizeof(struct buffer_head));
902 ret = get_data_block(inode, start_blk, &map_bh, 0,
903 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
908 if (!buffer_mapped(&map_bh)) {
909 start_blk = next_pgofs;
910 /* Go through holes util pass the EOF */
911 if (blk_to_logical(inode, start_blk) < isize)
913 /* Found a hole beyond isize means no more extents.
914 * Note that the premise is that filesystems don't
915 * punch holes beyond isize and keep size unchanged.
917 flags |= FIEMAP_EXTENT_LAST;
921 if (f2fs_encrypted_inode(inode))
922 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
924 ret = fiemap_fill_next_extent(fieinfo, logical,
928 if (start_blk > last_blk || ret)
931 logical = blk_to_logical(inode, start_blk);
932 phys = blk_to_logical(inode, map_bh.b_blocknr);
933 size = map_bh.b_size;
935 if (buffer_unwritten(&map_bh))
936 flags = FIEMAP_EXTENT_UNWRITTEN;
938 start_blk += logical_to_blk(inode, size);
942 if (fatal_signal_pending(current))
955 * This function was originally taken from fs/mpage.c, and customized for f2fs.
956 * Major change was from block_size == page_size in f2fs by default.
958 static int f2fs_mpage_readpages(struct address_space *mapping,
959 struct list_head *pages, struct page *page,
962 struct bio *bio = NULL;
964 sector_t last_block_in_bio = 0;
965 struct inode *inode = mapping->host;
966 const unsigned blkbits = inode->i_blkbits;
967 const unsigned blocksize = 1 << blkbits;
968 sector_t block_in_file;
970 sector_t last_block_in_file;
972 struct block_device *bdev = inode->i_sb->s_bdev;
973 struct f2fs_map_blocks map;
979 map.m_next_pgofs = NULL;
981 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
983 prefetchw(&page->flags);
985 page = list_entry(pages->prev, struct page, lru);
986 list_del(&page->lru);
987 if (add_to_page_cache_lru(page, mapping,
988 page->index, GFP_KERNEL))
992 block_in_file = (sector_t)page->index;
993 last_block = block_in_file + nr_pages;
994 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
996 if (last_block > last_block_in_file)
997 last_block = last_block_in_file;
1000 * Map blocks using the previous result first.
1002 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1003 block_in_file > map.m_lblk &&
1004 block_in_file < (map.m_lblk + map.m_len))
1008 * Then do more f2fs_map_blocks() calls until we are
1009 * done with this page.
1013 if (block_in_file < last_block) {
1014 map.m_lblk = block_in_file;
1015 map.m_len = last_block - block_in_file;
1017 if (f2fs_map_blocks(inode, &map, 0,
1018 F2FS_GET_BLOCK_READ))
1019 goto set_error_page;
1022 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1023 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1024 SetPageMappedToDisk(page);
1026 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1027 SetPageUptodate(page);
1031 zero_user_segment(page, 0, PAGE_SIZE);
1032 SetPageUptodate(page);
1038 * This page will go to BIO. Do we need to send this
1041 if (bio && (last_block_in_bio != block_nr - 1)) {
1043 __submit_bio(F2FS_I_SB(inode), READ, bio);
1047 struct fscrypt_ctx *ctx = NULL;
1049 if (f2fs_encrypted_inode(inode) &&
1050 S_ISREG(inode->i_mode)) {
1052 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1054 goto set_error_page;
1056 /* wait the page to be moved by cleaning */
1057 f2fs_wait_on_encrypted_page_writeback(
1058 F2FS_I_SB(inode), block_nr);
1061 bio = bio_alloc(GFP_KERNEL,
1062 min_t(int, nr_pages, BIO_MAX_PAGES));
1065 fscrypt_release_ctx(ctx);
1066 goto set_error_page;
1068 bio->bi_bdev = bdev;
1069 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1070 bio->bi_end_io = f2fs_read_end_io;
1071 bio->bi_private = ctx;
1074 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1075 goto submit_and_realloc;
1077 last_block_in_bio = block_nr;
1081 zero_user_segment(page, 0, PAGE_SIZE);
1086 __submit_bio(F2FS_I_SB(inode), READ, bio);
1094 BUG_ON(pages && !list_empty(pages));
1096 __submit_bio(F2FS_I_SB(inode), READ, bio);
1100 static int f2fs_read_data_page(struct file *file, struct page *page)
1102 struct inode *inode = page->mapping->host;
1105 trace_f2fs_readpage(page, DATA);
1107 /* If the file has inline data, try to read it directly */
1108 if (f2fs_has_inline_data(inode))
1109 ret = f2fs_read_inline_data(inode, page);
1111 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1115 static int f2fs_read_data_pages(struct file *file,
1116 struct address_space *mapping,
1117 struct list_head *pages, unsigned nr_pages)
1119 struct inode *inode = file->f_mapping->host;
1120 struct page *page = list_entry(pages->prev, struct page, lru);
1122 trace_f2fs_readpages(inode, page, nr_pages);
1124 /* If the file has inline data, skip readpages */
1125 if (f2fs_has_inline_data(inode))
1128 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1131 int do_write_data_page(struct f2fs_io_info *fio)
1133 struct page *page = fio->page;
1134 struct inode *inode = page->mapping->host;
1135 struct dnode_of_data dn;
1138 set_new_dnode(&dn, inode, NULL, NULL, 0);
1139 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1143 fio->old_blkaddr = dn.data_blkaddr;
1145 /* This page is already truncated */
1146 if (fio->old_blkaddr == NULL_ADDR) {
1147 ClearPageUptodate(page);
1151 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1152 gfp_t gfp_flags = GFP_NOFS;
1154 /* wait for GCed encrypted page writeback */
1155 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1158 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1160 if (IS_ERR(fio->encrypted_page)) {
1161 err = PTR_ERR(fio->encrypted_page);
1162 if (err == -ENOMEM) {
1163 /* flush pending ios and wait for a while */
1164 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1165 congestion_wait(BLK_RW_ASYNC, HZ/50);
1166 gfp_flags |= __GFP_NOFAIL;
1174 set_page_writeback(page);
1177 * If current allocation needs SSR,
1178 * it had better in-place writes for updated data.
1180 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1181 !is_cold_data(page) &&
1182 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1183 need_inplace_update(inode))) {
1184 rewrite_data_page(fio);
1185 set_inode_flag(inode, FI_UPDATE_WRITE);
1186 trace_f2fs_do_write_data_page(page, IPU);
1188 write_data_page(&dn, fio);
1189 trace_f2fs_do_write_data_page(page, OPU);
1190 set_inode_flag(inode, FI_APPEND_WRITE);
1191 if (page->index == 0)
1192 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1195 f2fs_put_dnode(&dn);
1199 static int f2fs_write_data_page(struct page *page,
1200 struct writeback_control *wbc)
1202 struct inode *inode = page->mapping->host;
1203 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1204 loff_t i_size = i_size_read(inode);
1205 const pgoff_t end_index = ((unsigned long long) i_size)
1207 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1208 unsigned offset = 0;
1209 bool need_balance_fs = false;
1211 struct f2fs_io_info fio = {
1214 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1216 .encrypted_page = NULL,
1219 trace_f2fs_writepage(page, DATA);
1221 if (page->index < end_index)
1225 * If the offset is out-of-range of file size,
1226 * this page does not have to be written to disk.
1228 offset = i_size & (PAGE_SIZE - 1);
1229 if ((page->index >= end_index + 1) || !offset)
1232 zero_user_segment(page, offset, PAGE_SIZE);
1234 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1236 if (f2fs_is_drop_cache(inode))
1238 /* we should not write 0'th page having journal header */
1239 if (f2fs_is_volatile_file(inode) && (!page->index ||
1240 (!wbc->for_reclaim &&
1241 available_free_memory(sbi, BASE_CHECK))))
1244 /* Dentry blocks are controlled by checkpoint */
1245 if (S_ISDIR(inode->i_mode)) {
1246 if (unlikely(f2fs_cp_error(sbi)))
1248 err = do_write_data_page(&fio);
1252 /* we should bypass data pages to proceed the kworkder jobs */
1253 if (unlikely(f2fs_cp_error(sbi))) {
1258 if (!wbc->for_reclaim)
1259 need_balance_fs = true;
1260 else if (has_not_enough_free_secs(sbi, 0))
1265 if (f2fs_has_inline_data(inode))
1266 err = f2fs_write_inline_data(inode, page);
1268 err = do_write_data_page(&fio);
1269 if (F2FS_I(inode)->last_disk_size < psize)
1270 F2FS_I(inode)->last_disk_size = psize;
1271 f2fs_unlock_op(sbi);
1273 if (err && err != -ENOENT)
1276 clear_cold_data(page);
1278 inode_dec_dirty_pages(inode);
1280 ClearPageUptodate(page);
1282 if (wbc->for_reclaim) {
1283 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1284 remove_dirty_inode(inode);
1288 f2fs_balance_fs(sbi, need_balance_fs);
1290 if (unlikely(f2fs_cp_error(sbi)))
1291 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1296 redirty_page_for_writepage(wbc, page);
1297 return AOP_WRITEPAGE_ACTIVATE;
1300 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1303 struct address_space *mapping = data;
1304 int ret = mapping->a_ops->writepage(page, wbc);
1305 mapping_set_error(mapping, ret);
1310 * This function was copied from write_cche_pages from mm/page-writeback.c.
1311 * The major change is making write step of cold data page separately from
1312 * warm/hot data page.
1314 static int f2fs_write_cache_pages(struct address_space *mapping,
1315 struct writeback_control *wbc, writepage_t writepage,
1320 struct pagevec pvec;
1322 pgoff_t uninitialized_var(writeback_index);
1324 pgoff_t end; /* Inclusive */
1327 int range_whole = 0;
1331 pagevec_init(&pvec, 0);
1333 if (wbc->range_cyclic) {
1334 writeback_index = mapping->writeback_index; /* prev offset */
1335 index = writeback_index;
1342 index = wbc->range_start >> PAGE_SHIFT;
1343 end = wbc->range_end >> PAGE_SHIFT;
1344 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1346 cycled = 1; /* ignore range_cyclic tests */
1348 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1349 tag = PAGECACHE_TAG_TOWRITE;
1351 tag = PAGECACHE_TAG_DIRTY;
1353 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1354 tag_pages_for_writeback(mapping, index, end);
1356 while (!done && (index <= end)) {
1359 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1360 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1364 for (i = 0; i < nr_pages; i++) {
1365 struct page *page = pvec.pages[i];
1367 if (page->index > end) {
1372 done_index = page->index;
1376 if (unlikely(page->mapping != mapping)) {
1382 if (!PageDirty(page)) {
1383 /* someone wrote it for us */
1384 goto continue_unlock;
1387 if (step == is_cold_data(page))
1388 goto continue_unlock;
1390 if (PageWriteback(page)) {
1391 if (wbc->sync_mode != WB_SYNC_NONE)
1392 f2fs_wait_on_page_writeback(page,
1395 goto continue_unlock;
1398 BUG_ON(PageWriteback(page));
1399 if (!clear_page_dirty_for_io(page))
1400 goto continue_unlock;
1402 ret = (*writepage)(page, wbc, data);
1403 if (unlikely(ret)) {
1404 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1408 done_index = page->index + 1;
1414 if (--wbc->nr_to_write <= 0 &&
1415 wbc->sync_mode == WB_SYNC_NONE) {
1420 pagevec_release(&pvec);
1429 if (!cycled && !done) {
1432 end = writeback_index - 1;
1435 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1436 mapping->writeback_index = done_index;
1441 static int f2fs_write_data_pages(struct address_space *mapping,
1442 struct writeback_control *wbc)
1444 struct inode *inode = mapping->host;
1445 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1446 bool locked = false;
1450 /* deal with chardevs and other special file */
1451 if (!mapping->a_ops->writepage)
1454 /* skip writing if there is no dirty page in this inode */
1455 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1458 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1459 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1460 available_free_memory(sbi, DIRTY_DENTS))
1463 /* skip writing during file defragment */
1464 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1467 /* during POR, we don't need to trigger writepage at all. */
1468 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1471 trace_f2fs_writepages(mapping->host, wbc, DATA);
1473 diff = nr_pages_to_write(sbi, DATA, wbc);
1475 if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1476 mutex_lock(&sbi->writepages);
1479 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1480 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1482 mutex_unlock(&sbi->writepages);
1484 remove_dirty_inode(inode);
1486 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1490 wbc->pages_skipped += get_dirty_pages(inode);
1491 trace_f2fs_writepages(mapping->host, wbc, DATA);
1495 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1497 struct inode *inode = mapping->host;
1498 loff_t i_size = i_size_read(inode);
1501 truncate_pagecache(inode, i_size);
1502 truncate_blocks(inode, i_size, true);
1506 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1507 struct page *page, loff_t pos, unsigned len,
1508 block_t *blk_addr, bool *node_changed)
1510 struct inode *inode = page->mapping->host;
1511 pgoff_t index = page->index;
1512 struct dnode_of_data dn;
1514 bool locked = false;
1515 struct extent_info ei;
1519 * we already allocated all the blocks, so we don't need to get
1520 * the block addresses when there is no need to fill the page.
1522 if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1526 if (f2fs_has_inline_data(inode) ||
1527 (pos & PAGE_MASK) >= i_size_read(inode)) {
1532 /* check inline_data */
1533 ipage = get_node_page(sbi, inode->i_ino);
1534 if (IS_ERR(ipage)) {
1535 err = PTR_ERR(ipage);
1539 set_new_dnode(&dn, inode, ipage, ipage, 0);
1541 if (f2fs_has_inline_data(inode)) {
1542 if (pos + len <= MAX_INLINE_DATA) {
1543 read_inline_data(page, ipage);
1544 set_inode_flag(inode, FI_DATA_EXIST);
1546 set_inline_node(ipage);
1548 err = f2fs_convert_inline_page(&dn, page);
1551 if (dn.data_blkaddr == NULL_ADDR)
1552 err = f2fs_get_block(&dn, index);
1554 } else if (locked) {
1555 err = f2fs_get_block(&dn, index);
1557 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1558 dn.data_blkaddr = ei.blk + index - ei.fofs;
1561 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1562 if (err || dn.data_blkaddr == NULL_ADDR) {
1563 f2fs_put_dnode(&dn);
1571 /* convert_inline_page can make node_changed */
1572 *blk_addr = dn.data_blkaddr;
1573 *node_changed = dn.node_changed;
1575 f2fs_put_dnode(&dn);
1578 f2fs_unlock_op(sbi);
1582 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1583 loff_t pos, unsigned len, unsigned flags,
1584 struct page **pagep, void **fsdata)
1586 struct inode *inode = mapping->host;
1587 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1588 struct page *page = NULL;
1589 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1590 bool need_balance = false;
1591 block_t blkaddr = NULL_ADDR;
1594 trace_f2fs_write_begin(inode, pos, len, flags);
1597 * We should check this at this moment to avoid deadlock on inode page
1598 * and #0 page. The locking rule for inline_data conversion should be:
1599 * lock_page(page #0) -> lock_page(inode_page)
1602 err = f2fs_convert_inline_inode(inode);
1607 page = grab_cache_page_write_begin(mapping, index, flags);
1615 err = prepare_write_begin(sbi, page, pos, len,
1616 &blkaddr, &need_balance);
1620 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1622 f2fs_balance_fs(sbi, true);
1624 if (page->mapping != mapping) {
1625 /* The page got truncated from under us */
1626 f2fs_put_page(page, 1);
1631 f2fs_wait_on_page_writeback(page, DATA, false);
1633 /* wait for GCed encrypted page writeback */
1634 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1635 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1637 if (len == PAGE_SIZE)
1639 if (PageUptodate(page))
1642 if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1643 unsigned start = pos & (PAGE_SIZE - 1);
1644 unsigned end = start + len;
1646 /* Reading beyond i_size is simple: memset to zero */
1647 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1651 if (blkaddr == NEW_ADDR) {
1652 zero_user_segment(page, 0, PAGE_SIZE);
1654 struct f2fs_io_info fio = {
1658 .old_blkaddr = blkaddr,
1659 .new_blkaddr = blkaddr,
1661 .encrypted_page = NULL,
1663 err = f2fs_submit_page_bio(&fio);
1668 if (unlikely(!PageUptodate(page))) {
1672 if (unlikely(page->mapping != mapping)) {
1673 f2fs_put_page(page, 1);
1677 /* avoid symlink page */
1678 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1679 err = fscrypt_decrypt_page(page);
1685 SetPageUptodate(page);
1687 clear_cold_data(page);
1691 f2fs_put_page(page, 1);
1692 f2fs_write_failed(mapping, pos + len);
1696 static int f2fs_write_end(struct file *file,
1697 struct address_space *mapping,
1698 loff_t pos, unsigned len, unsigned copied,
1699 struct page *page, void *fsdata)
1701 struct inode *inode = page->mapping->host;
1703 trace_f2fs_write_end(inode, pos, len, copied);
1705 set_page_dirty(page);
1707 if (pos + copied > i_size_read(inode))
1708 f2fs_i_size_write(inode, pos + copied);
1710 f2fs_put_page(page, 1);
1711 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1715 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1718 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1720 if (offset & blocksize_mask)
1723 if (iov_iter_alignment(iter) & blocksize_mask)
1729 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1731 struct address_space *mapping = iocb->ki_filp->f_mapping;
1732 struct inode *inode = mapping->host;
1733 size_t count = iov_iter_count(iter);
1734 loff_t offset = iocb->ki_pos;
1737 err = check_direct_IO(inode, iter, offset);
1741 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1744 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1746 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1747 if (iov_iter_rw(iter) == WRITE) {
1749 set_inode_flag(inode, FI_UPDATE_WRITE);
1751 f2fs_write_failed(mapping, offset + count);
1754 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1759 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1760 unsigned int length)
1762 struct inode *inode = page->mapping->host;
1763 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1765 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1766 (offset % PAGE_SIZE || length != PAGE_SIZE))
1769 if (PageDirty(page)) {
1770 if (inode->i_ino == F2FS_META_INO(sbi))
1771 dec_page_count(sbi, F2FS_DIRTY_META);
1772 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1773 dec_page_count(sbi, F2FS_DIRTY_NODES);
1775 inode_dec_dirty_pages(inode);
1778 /* This is atomic written page, keep Private */
1779 if (IS_ATOMIC_WRITTEN_PAGE(page))
1782 set_page_private(page, 0);
1783 ClearPagePrivate(page);
1786 int f2fs_release_page(struct page *page, gfp_t wait)
1788 /* If this is dirty page, keep PagePrivate */
1789 if (PageDirty(page))
1792 /* This is atomic written page, keep Private */
1793 if (IS_ATOMIC_WRITTEN_PAGE(page))
1796 set_page_private(page, 0);
1797 ClearPagePrivate(page);
1801 static int f2fs_set_data_page_dirty(struct page *page)
1803 struct address_space *mapping = page->mapping;
1804 struct inode *inode = mapping->host;
1806 trace_f2fs_set_page_dirty(page, DATA);
1808 SetPageUptodate(page);
1810 if (f2fs_is_atomic_file(inode)) {
1811 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1812 register_inmem_page(inode, page);
1816 * Previously, this page has been registered, we just
1822 if (!PageDirty(page)) {
1823 __set_page_dirty_nobuffers(page);
1824 update_dirty_page(inode, page);
1830 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1832 struct inode *inode = mapping->host;
1834 if (f2fs_has_inline_data(inode))
1837 /* make sure allocating whole blocks */
1838 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1839 filemap_write_and_wait(mapping);
1841 return generic_block_bmap(mapping, block, get_data_block_bmap);
1844 const struct address_space_operations f2fs_dblock_aops = {
1845 .readpage = f2fs_read_data_page,
1846 .readpages = f2fs_read_data_pages,
1847 .writepage = f2fs_write_data_page,
1848 .writepages = f2fs_write_data_pages,
1849 .write_begin = f2fs_write_begin,
1850 .write_end = f2fs_write_end,
1851 .set_page_dirty = f2fs_set_data_page_dirty,
1852 .invalidatepage = f2fs_invalidate_page,
1853 .releasepage = f2fs_release_page,
1854 .direct_IO = f2fs_direct_IO,