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)) {
70 set_bit(AS_EIO, &page->mapping->flags);
71 f2fs_stop_checkpoint(sbi);
73 end_page_writeback(page);
74 dec_page_count(sbi, F2FS_WRITEBACK);
77 if (!get_pages(sbi, F2FS_WRITEBACK) && wq_has_sleeper(&sbi->cp_wait))
78 wake_up(&sbi->cp_wait);
84 * Low-level block read/write IO operations.
86 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
87 int npages, bool is_read)
91 bio = f2fs_bio_alloc(npages);
93 bio->bi_bdev = sbi->sb->s_bdev;
94 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
95 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
96 bio->bi_private = is_read ? NULL : sbi;
101 static void __submit_merged_bio(struct f2fs_bio_info *io)
103 struct f2fs_io_info *fio = &io->fio;
108 if (is_read_io(fio->rw))
109 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
111 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
113 submit_bio(fio->rw, io->bio);
117 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
118 struct page *page, nid_t ino)
120 struct bio_vec *bvec;
127 if (!inode && !page && !ino)
130 bio_for_each_segment_all(bvec, io->bio, i) {
132 if (bvec->bv_page->mapping) {
133 target = bvec->bv_page;
135 struct f2fs_crypto_ctx *ctx;
138 ctx = (struct f2fs_crypto_ctx *)page_private(
140 target = ctx->w.control_page;
143 if (inode && inode == target->mapping->host)
145 if (page && page == target)
147 if (ino && ino == ino_of_node(target))
154 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
155 struct page *page, nid_t ino,
158 enum page_type btype = PAGE_TYPE_OF_BIO(type);
159 struct f2fs_bio_info *io = &sbi->write_io[btype];
162 down_read(&io->io_rwsem);
163 ret = __has_merged_page(io, inode, page, ino);
164 up_read(&io->io_rwsem);
168 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
169 struct inode *inode, struct page *page,
170 nid_t ino, enum page_type type, int rw)
172 enum page_type btype = PAGE_TYPE_OF_BIO(type);
173 struct f2fs_bio_info *io;
175 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
177 down_write(&io->io_rwsem);
179 if (!__has_merged_page(io, inode, page, ino))
182 /* change META to META_FLUSH in the checkpoint procedure */
183 if (type >= META_FLUSH) {
184 io->fio.type = META_FLUSH;
185 if (test_opt(sbi, NOBARRIER))
186 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
188 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
190 __submit_merged_bio(io);
192 up_write(&io->io_rwsem);
195 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
198 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
201 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
202 struct inode *inode, struct page *page,
203 nid_t ino, enum page_type type, int rw)
205 if (has_merged_page(sbi, inode, page, ino, type))
206 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
210 * Fill the locked page with data located in the block address.
211 * Return unlocked page.
213 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
216 struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
218 trace_f2fs_submit_page_bio(page, fio);
219 f2fs_trace_ios(fio, 0);
221 /* Allocate a new bio */
222 bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
224 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
229 submit_bio(fio->rw, bio);
233 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
235 struct f2fs_sb_info *sbi = fio->sbi;
236 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
237 struct f2fs_bio_info *io;
238 bool is_read = is_read_io(fio->rw);
239 struct page *bio_page;
241 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
243 verify_block_addr(sbi, fio->blk_addr);
245 down_write(&io->io_rwsem);
248 inc_page_count(sbi, F2FS_WRITEBACK);
250 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
251 io->fio.rw != fio->rw))
252 __submit_merged_bio(io);
254 if (io->bio == NULL) {
255 int bio_blocks = MAX_BIO_BLOCKS(sbi);
257 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
261 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
263 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
265 __submit_merged_bio(io);
269 io->last_block_in_bio = fio->blk_addr;
270 f2fs_trace_ios(fio, 0);
272 up_write(&io->io_rwsem);
273 trace_f2fs_submit_page_mbio(fio->page, fio);
277 * Lock ordering for the change of data block address:
280 * update block addresses in the node page
282 void set_data_blkaddr(struct dnode_of_data *dn)
284 struct f2fs_node *rn;
286 struct page *node_page = dn->node_page;
287 unsigned int ofs_in_node = dn->ofs_in_node;
289 f2fs_wait_on_page_writeback(node_page, NODE, true);
291 rn = F2FS_NODE(node_page);
293 /* Get physical address of data block */
294 addr_array = blkaddr_in_node(rn);
295 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
296 if (set_page_dirty(node_page))
297 dn->node_changed = true;
300 int reserve_new_block(struct dnode_of_data *dn)
302 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
304 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
306 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
309 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
311 dn->data_blkaddr = NEW_ADDR;
312 set_data_blkaddr(dn);
313 mark_inode_dirty(dn->inode);
318 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
320 bool need_put = dn->inode_page ? false : true;
323 err = get_dnode_of_data(dn, index, ALLOC_NODE);
327 if (dn->data_blkaddr == NULL_ADDR)
328 err = reserve_new_block(dn);
334 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
336 struct extent_info ei;
337 struct inode *inode = dn->inode;
339 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
340 dn->data_blkaddr = ei.blk + index - ei.fofs;
344 return f2fs_reserve_block(dn, index);
347 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
348 int rw, bool for_write)
350 struct address_space *mapping = inode->i_mapping;
351 struct dnode_of_data dn;
353 struct extent_info ei;
355 struct f2fs_io_info fio = {
356 .sbi = F2FS_I_SB(inode),
359 .encrypted_page = NULL,
362 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
363 return read_mapping_page(mapping, index, NULL);
365 page = f2fs_grab_cache_page(mapping, index, for_write);
367 return ERR_PTR(-ENOMEM);
369 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
370 dn.data_blkaddr = ei.blk + index - ei.fofs;
374 set_new_dnode(&dn, inode, NULL, NULL, 0);
375 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
380 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
385 if (PageUptodate(page)) {
391 * A new dentry page is allocated but not able to be written, since its
392 * new inode page couldn't be allocated due to -ENOSPC.
393 * In such the case, its blkaddr can be remained as NEW_ADDR.
394 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
396 if (dn.data_blkaddr == NEW_ADDR) {
397 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
398 SetPageUptodate(page);
403 fio.blk_addr = dn.data_blkaddr;
405 err = f2fs_submit_page_bio(&fio);
411 f2fs_put_page(page, 1);
415 struct page *find_data_page(struct inode *inode, pgoff_t index)
417 struct address_space *mapping = inode->i_mapping;
420 page = find_get_page(mapping, index);
421 if (page && PageUptodate(page))
423 f2fs_put_page(page, 0);
425 page = get_read_data_page(inode, index, READ_SYNC, false);
429 if (PageUptodate(page))
432 wait_on_page_locked(page);
433 if (unlikely(!PageUptodate(page))) {
434 f2fs_put_page(page, 0);
435 return ERR_PTR(-EIO);
441 * If it tries to access a hole, return an error.
442 * Because, the callers, functions in dir.c and GC, should be able to know
443 * whether this page exists or not.
445 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
448 struct address_space *mapping = inode->i_mapping;
451 page = get_read_data_page(inode, index, READ_SYNC, for_write);
455 /* wait for read completion */
457 if (unlikely(!PageUptodate(page))) {
458 f2fs_put_page(page, 1);
459 return ERR_PTR(-EIO);
461 if (unlikely(page->mapping != mapping)) {
462 f2fs_put_page(page, 1);
469 * Caller ensures that this data page is never allocated.
470 * A new zero-filled data page is allocated in the page cache.
472 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
474 * Note that, ipage is set only by make_empty_dir, and if any error occur,
475 * ipage should be released by this function.
477 struct page *get_new_data_page(struct inode *inode,
478 struct page *ipage, pgoff_t index, bool new_i_size)
480 struct address_space *mapping = inode->i_mapping;
482 struct dnode_of_data dn;
485 page = f2fs_grab_cache_page(mapping, index, true);
488 * before exiting, we should make sure ipage will be released
489 * if any error occur.
491 f2fs_put_page(ipage, 1);
492 return ERR_PTR(-ENOMEM);
495 set_new_dnode(&dn, inode, ipage, NULL, 0);
496 err = f2fs_reserve_block(&dn, index);
498 f2fs_put_page(page, 1);
504 if (PageUptodate(page))
507 if (dn.data_blkaddr == NEW_ADDR) {
508 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
509 SetPageUptodate(page);
511 f2fs_put_page(page, 1);
513 /* if ipage exists, blkaddr should be NEW_ADDR */
514 f2fs_bug_on(F2FS_I_SB(inode), ipage);
515 page = get_lock_data_page(inode, index, true);
520 if (new_i_size && i_size_read(inode) <
521 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
522 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
523 /* Only the directory inode sets new_i_size */
524 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
529 static int __allocate_data_block(struct dnode_of_data *dn)
531 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
532 struct f2fs_summary sum;
534 int seg = CURSEG_WARM_DATA;
537 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
540 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
541 if (dn->data_blkaddr == NEW_ADDR)
544 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
548 get_node_info(sbi, dn->nid, &ni);
549 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
551 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
552 seg = CURSEG_DIRECT_IO;
554 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
556 set_data_blkaddr(dn);
559 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
561 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
562 i_size_write(dn->inode,
563 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
567 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
569 struct inode *inode = file_inode(iocb->ki_filp);
570 struct f2fs_map_blocks map;
573 map.m_lblk = F2FS_BYTES_TO_BLK(iocb->ki_pos);
574 map.m_len = F2FS_BLK_ALIGN(iov_iter_count(from));
575 map.m_next_pgofs = NULL;
577 if (f2fs_encrypted_inode(inode))
580 if (iocb->ki_flags & IOCB_DIRECT) {
581 ret = f2fs_convert_inline_inode(inode);
584 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
586 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
587 ret = f2fs_convert_inline_inode(inode);
591 if (!f2fs_has_inline_data(inode))
592 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
597 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
598 * f2fs_map_blocks structure.
599 * If original data blocks are allocated, then give them to blockdev.
601 * a. preallocate requested block addresses
602 * b. do not use extent cache for better performance
603 * c. give the block addresses to blockdev
605 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
606 int create, int flag)
608 unsigned int maxblocks = map->m_len;
609 struct dnode_of_data dn;
610 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
611 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
612 pgoff_t pgofs, end_offset;
613 int err = 0, ofs = 1;
614 struct extent_info ei;
615 bool allocated = false;
621 /* it only supports block size == page size */
622 pgofs = (pgoff_t)map->m_lblk;
624 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
625 map->m_pblk = ei.blk + pgofs - ei.fofs;
626 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
627 map->m_flags = F2FS_MAP_MAPPED;
635 /* When reading holes, we need its node page */
636 set_new_dnode(&dn, inode, NULL, NULL, 0);
637 err = get_dnode_of_data(&dn, pgofs, mode);
639 if (err == -ENOENT) {
641 if (map->m_next_pgofs)
643 get_next_page_offset(&dn, pgofs);
648 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
651 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
653 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
655 if (unlikely(f2fs_cp_error(sbi))) {
659 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
660 if (blkaddr == NULL_ADDR)
661 err = reserve_new_block(&dn);
663 err = __allocate_data_block(&dn);
668 map->m_flags = F2FS_MAP_NEW;
669 blkaddr = dn.data_blkaddr;
671 if (flag == F2FS_GET_BLOCK_FIEMAP &&
672 blkaddr == NULL_ADDR) {
673 if (map->m_next_pgofs)
674 *map->m_next_pgofs = pgofs + 1;
676 if (flag != F2FS_GET_BLOCK_FIEMAP ||
677 blkaddr != NEW_ADDR) {
678 if (flag == F2FS_GET_BLOCK_BMAP)
685 if (map->m_len == 0) {
686 /* preallocated unwritten block should be mapped for fiemap. */
687 if (blkaddr == NEW_ADDR)
688 map->m_flags |= F2FS_MAP_UNWRITTEN;
689 map->m_flags |= F2FS_MAP_MAPPED;
691 map->m_pblk = blkaddr;
693 } else if ((map->m_pblk != NEW_ADDR &&
694 blkaddr == (map->m_pblk + ofs)) ||
695 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
696 flag == F2FS_GET_BLOCK_PRE_DIO ||
697 flag == F2FS_GET_BLOCK_PRE_AIO) {
707 if (map->m_len < maxblocks) {
708 if (dn.ofs_in_node < end_offset)
712 sync_inode_page(&dn);
717 f2fs_balance_fs(sbi, allocated);
725 sync_inode_page(&dn);
730 f2fs_balance_fs(sbi, allocated);
733 trace_f2fs_map_blocks(inode, map, err);
737 static int __get_data_block(struct inode *inode, sector_t iblock,
738 struct buffer_head *bh, int create, int flag,
741 struct f2fs_map_blocks map;
745 map.m_len = bh->b_size >> inode->i_blkbits;
746 map.m_next_pgofs = next_pgofs;
748 ret = f2fs_map_blocks(inode, &map, create, flag);
750 map_bh(bh, inode->i_sb, map.m_pblk);
751 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
752 bh->b_size = map.m_len << inode->i_blkbits;
757 static int get_data_block(struct inode *inode, sector_t iblock,
758 struct buffer_head *bh_result, int create, int flag,
761 return __get_data_block(inode, iblock, bh_result, create,
765 static int get_data_block_dio(struct inode *inode, sector_t iblock,
766 struct buffer_head *bh_result, int create)
768 return __get_data_block(inode, iblock, bh_result, create,
769 F2FS_GET_BLOCK_DIO, NULL);
772 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
773 struct buffer_head *bh_result, int create)
775 /* Block number less than F2FS MAX BLOCKS */
776 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
779 return __get_data_block(inode, iblock, bh_result, create,
780 F2FS_GET_BLOCK_BMAP, NULL);
783 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
785 return (offset >> inode->i_blkbits);
788 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
790 return (blk << inode->i_blkbits);
793 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
796 struct buffer_head map_bh;
797 sector_t start_blk, last_blk;
800 u64 logical = 0, phys = 0, size = 0;
804 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
808 if (f2fs_has_inline_data(inode)) {
809 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
816 isize = i_size_read(inode);
820 if (start + len > isize)
823 if (logical_to_blk(inode, len) == 0)
824 len = blk_to_logical(inode, 1);
826 start_blk = logical_to_blk(inode, start);
827 last_blk = logical_to_blk(inode, start + len - 1);
830 memset(&map_bh, 0, sizeof(struct buffer_head));
833 ret = get_data_block(inode, start_blk, &map_bh, 0,
834 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
839 if (!buffer_mapped(&map_bh)) {
840 start_blk = next_pgofs;
841 /* Go through holes util pass the EOF */
842 if (blk_to_logical(inode, start_blk) < isize)
844 /* Found a hole beyond isize means no more extents.
845 * Note that the premise is that filesystems don't
846 * punch holes beyond isize and keep size unchanged.
848 flags |= FIEMAP_EXTENT_LAST;
852 if (f2fs_encrypted_inode(inode))
853 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
855 ret = fiemap_fill_next_extent(fieinfo, logical,
859 if (start_blk > last_blk || ret)
862 logical = blk_to_logical(inode, start_blk);
863 phys = blk_to_logical(inode, map_bh.b_blocknr);
864 size = map_bh.b_size;
866 if (buffer_unwritten(&map_bh))
867 flags = FIEMAP_EXTENT_UNWRITTEN;
869 start_blk += logical_to_blk(inode, size);
873 if (fatal_signal_pending(current))
886 * This function was originally taken from fs/mpage.c, and customized for f2fs.
887 * Major change was from block_size == page_size in f2fs by default.
889 static int f2fs_mpage_readpages(struct address_space *mapping,
890 struct list_head *pages, struct page *page,
893 struct bio *bio = NULL;
895 sector_t last_block_in_bio = 0;
896 struct inode *inode = mapping->host;
897 const unsigned blkbits = inode->i_blkbits;
898 const unsigned blocksize = 1 << blkbits;
899 sector_t block_in_file;
901 sector_t last_block_in_file;
903 struct block_device *bdev = inode->i_sb->s_bdev;
904 struct f2fs_map_blocks map;
910 map.m_next_pgofs = NULL;
912 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
914 prefetchw(&page->flags);
916 page = list_entry(pages->prev, struct page, lru);
917 list_del(&page->lru);
918 if (add_to_page_cache_lru(page, mapping,
919 page->index, GFP_KERNEL))
923 block_in_file = (sector_t)page->index;
924 last_block = block_in_file + nr_pages;
925 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
927 if (last_block > last_block_in_file)
928 last_block = last_block_in_file;
931 * Map blocks using the previous result first.
933 if ((map.m_flags & F2FS_MAP_MAPPED) &&
934 block_in_file > map.m_lblk &&
935 block_in_file < (map.m_lblk + map.m_len))
939 * Then do more f2fs_map_blocks() calls until we are
940 * done with this page.
944 if (block_in_file < last_block) {
945 map.m_lblk = block_in_file;
946 map.m_len = last_block - block_in_file;
948 if (f2fs_map_blocks(inode, &map, 0,
949 F2FS_GET_BLOCK_READ))
953 if ((map.m_flags & F2FS_MAP_MAPPED)) {
954 block_nr = map.m_pblk + block_in_file - map.m_lblk;
955 SetPageMappedToDisk(page);
957 if (!PageUptodate(page) && !cleancache_get_page(page)) {
958 SetPageUptodate(page);
962 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
963 SetPageUptodate(page);
969 * This page will go to BIO. Do we need to send this
972 if (bio && (last_block_in_bio != block_nr - 1)) {
974 submit_bio(READ, bio);
978 struct f2fs_crypto_ctx *ctx = NULL;
980 if (f2fs_encrypted_inode(inode) &&
981 S_ISREG(inode->i_mode)) {
983 ctx = f2fs_get_crypto_ctx(inode);
987 /* wait the page to be moved by cleaning */
988 f2fs_wait_on_encrypted_page_writeback(
989 F2FS_I_SB(inode), block_nr);
992 bio = bio_alloc(GFP_KERNEL,
993 min_t(int, nr_pages, BIO_MAX_PAGES));
996 f2fs_release_crypto_ctx(ctx);
1000 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1001 bio->bi_end_io = f2fs_read_end_io;
1002 bio->bi_private = ctx;
1005 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1006 goto submit_and_realloc;
1008 last_block_in_bio = block_nr;
1012 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1017 submit_bio(READ, bio);
1023 page_cache_release(page);
1025 BUG_ON(pages && !list_empty(pages));
1027 submit_bio(READ, bio);
1031 static int f2fs_read_data_page(struct file *file, struct page *page)
1033 struct inode *inode = page->mapping->host;
1036 trace_f2fs_readpage(page, DATA);
1038 /* If the file has inline data, try to read it directly */
1039 if (f2fs_has_inline_data(inode))
1040 ret = f2fs_read_inline_data(inode, page);
1042 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1046 static int f2fs_read_data_pages(struct file *file,
1047 struct address_space *mapping,
1048 struct list_head *pages, unsigned nr_pages)
1050 struct inode *inode = file->f_mapping->host;
1051 struct page *page = list_entry(pages->prev, struct page, lru);
1053 trace_f2fs_readpages(inode, page, nr_pages);
1055 /* If the file has inline data, skip readpages */
1056 if (f2fs_has_inline_data(inode))
1059 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1062 int do_write_data_page(struct f2fs_io_info *fio)
1064 struct page *page = fio->page;
1065 struct inode *inode = page->mapping->host;
1066 struct dnode_of_data dn;
1069 set_new_dnode(&dn, inode, NULL, NULL, 0);
1070 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1074 fio->blk_addr = dn.data_blkaddr;
1076 /* This page is already truncated */
1077 if (fio->blk_addr == NULL_ADDR) {
1078 ClearPageUptodate(page);
1082 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1084 /* wait for GCed encrypted page writeback */
1085 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1088 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1089 if (IS_ERR(fio->encrypted_page)) {
1090 err = PTR_ERR(fio->encrypted_page);
1095 set_page_writeback(page);
1098 * If current allocation needs SSR,
1099 * it had better in-place writes for updated data.
1101 if (unlikely(fio->blk_addr != NEW_ADDR &&
1102 !is_cold_data(page) &&
1103 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1104 need_inplace_update(inode))) {
1105 rewrite_data_page(fio);
1106 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1107 trace_f2fs_do_write_data_page(page, IPU);
1109 write_data_page(&dn, fio);
1110 set_data_blkaddr(&dn);
1111 f2fs_update_extent_cache(&dn);
1112 trace_f2fs_do_write_data_page(page, OPU);
1113 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1114 if (page->index == 0)
1115 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1118 f2fs_put_dnode(&dn);
1122 static int f2fs_write_data_page(struct page *page,
1123 struct writeback_control *wbc)
1125 struct inode *inode = page->mapping->host;
1126 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1127 loff_t i_size = i_size_read(inode);
1128 const pgoff_t end_index = ((unsigned long long) i_size)
1129 >> PAGE_CACHE_SHIFT;
1130 unsigned offset = 0;
1131 bool need_balance_fs = false;
1133 struct f2fs_io_info fio = {
1136 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1138 .encrypted_page = NULL,
1141 trace_f2fs_writepage(page, DATA);
1143 if (page->index < end_index)
1147 * If the offset is out-of-range of file size,
1148 * this page does not have to be written to disk.
1150 offset = i_size & (PAGE_CACHE_SIZE - 1);
1151 if ((page->index >= end_index + 1) || !offset)
1154 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1156 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1158 if (f2fs_is_drop_cache(inode))
1160 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1161 available_free_memory(sbi, BASE_CHECK))
1164 /* Dentry blocks are controlled by checkpoint */
1165 if (S_ISDIR(inode->i_mode)) {
1166 if (unlikely(f2fs_cp_error(sbi)))
1168 err = do_write_data_page(&fio);
1172 /* we should bypass data pages to proceed the kworkder jobs */
1173 if (unlikely(f2fs_cp_error(sbi))) {
1178 if (!wbc->for_reclaim)
1179 need_balance_fs = true;
1180 else if (has_not_enough_free_secs(sbi, 0))
1185 if (f2fs_has_inline_data(inode))
1186 err = f2fs_write_inline_data(inode, page);
1188 err = do_write_data_page(&fio);
1189 f2fs_unlock_op(sbi);
1191 if (err && err != -ENOENT)
1194 clear_cold_data(page);
1196 inode_dec_dirty_pages(inode);
1198 ClearPageUptodate(page);
1200 if (wbc->for_reclaim) {
1201 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1202 remove_dirty_inode(inode);
1206 f2fs_balance_fs(sbi, need_balance_fs);
1208 if (unlikely(f2fs_cp_error(sbi)))
1209 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1214 redirty_page_for_writepage(wbc, page);
1215 return AOP_WRITEPAGE_ACTIVATE;
1218 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1221 struct address_space *mapping = data;
1222 int ret = mapping->a_ops->writepage(page, wbc);
1223 mapping_set_error(mapping, ret);
1228 * This function was copied from write_cche_pages from mm/page-writeback.c.
1229 * The major change is making write step of cold data page separately from
1230 * warm/hot data page.
1232 static int f2fs_write_cache_pages(struct address_space *mapping,
1233 struct writeback_control *wbc, writepage_t writepage,
1238 struct pagevec pvec;
1240 pgoff_t uninitialized_var(writeback_index);
1242 pgoff_t end; /* Inclusive */
1245 int range_whole = 0;
1249 pagevec_init(&pvec, 0);
1251 if (wbc->range_cyclic) {
1252 writeback_index = mapping->writeback_index; /* prev offset */
1253 index = writeback_index;
1260 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1261 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1262 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1264 cycled = 1; /* ignore range_cyclic tests */
1266 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1267 tag = PAGECACHE_TAG_TOWRITE;
1269 tag = PAGECACHE_TAG_DIRTY;
1271 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1272 tag_pages_for_writeback(mapping, index, end);
1274 while (!done && (index <= end)) {
1277 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1278 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1282 for (i = 0; i < nr_pages; i++) {
1283 struct page *page = pvec.pages[i];
1285 if (page->index > end) {
1290 done_index = page->index;
1294 if (unlikely(page->mapping != mapping)) {
1300 if (!PageDirty(page)) {
1301 /* someone wrote it for us */
1302 goto continue_unlock;
1305 if (step == is_cold_data(page))
1306 goto continue_unlock;
1308 if (PageWriteback(page)) {
1309 if (wbc->sync_mode != WB_SYNC_NONE)
1310 f2fs_wait_on_page_writeback(page,
1313 goto continue_unlock;
1316 BUG_ON(PageWriteback(page));
1317 if (!clear_page_dirty_for_io(page))
1318 goto continue_unlock;
1320 ret = (*writepage)(page, wbc, data);
1321 if (unlikely(ret)) {
1322 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1326 done_index = page->index + 1;
1332 if (--wbc->nr_to_write <= 0 &&
1333 wbc->sync_mode == WB_SYNC_NONE) {
1338 pagevec_release(&pvec);
1347 if (!cycled && !done) {
1350 end = writeback_index - 1;
1353 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1354 mapping->writeback_index = done_index;
1359 static int f2fs_write_data_pages(struct address_space *mapping,
1360 struct writeback_control *wbc)
1362 struct inode *inode = mapping->host;
1363 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1364 bool locked = false;
1368 /* deal with chardevs and other special file */
1369 if (!mapping->a_ops->writepage)
1372 /* skip writing if there is no dirty page in this inode */
1373 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1376 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1377 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1378 available_free_memory(sbi, DIRTY_DENTS))
1381 /* skip writing during file defragment */
1382 if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1385 /* during POR, we don't need to trigger writepage at all. */
1386 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1389 trace_f2fs_writepages(mapping->host, wbc, DATA);
1391 diff = nr_pages_to_write(sbi, DATA, wbc);
1393 if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1394 mutex_lock(&sbi->writepages);
1397 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1398 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1400 mutex_unlock(&sbi->writepages);
1402 remove_dirty_inode(inode);
1404 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1408 wbc->pages_skipped += get_dirty_pages(inode);
1409 trace_f2fs_writepages(mapping->host, wbc, DATA);
1413 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1415 struct inode *inode = mapping->host;
1416 loff_t i_size = i_size_read(inode);
1419 truncate_pagecache(inode, i_size);
1420 truncate_blocks(inode, i_size, true);
1424 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1425 struct page *page, loff_t pos, unsigned len,
1426 block_t *blk_addr, bool *node_changed)
1428 struct inode *inode = page->mapping->host;
1429 pgoff_t index = page->index;
1430 struct dnode_of_data dn;
1432 bool locked = false;
1433 struct extent_info ei;
1437 * we already allocated all the blocks, so we don't need to get
1438 * the block addresses when there is no need to fill the page.
1440 if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1441 len == PAGE_CACHE_SIZE)
1444 if (f2fs_has_inline_data(inode) ||
1445 (pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1450 /* check inline_data */
1451 ipage = get_node_page(sbi, inode->i_ino);
1452 if (IS_ERR(ipage)) {
1453 err = PTR_ERR(ipage);
1457 set_new_dnode(&dn, inode, ipage, ipage, 0);
1459 if (f2fs_has_inline_data(inode)) {
1460 if (pos + len <= MAX_INLINE_DATA) {
1461 read_inline_data(page, ipage);
1462 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1463 set_inline_node(ipage);
1465 err = f2fs_convert_inline_page(&dn, page);
1468 if (dn.data_blkaddr == NULL_ADDR)
1469 err = f2fs_get_block(&dn, index);
1471 } else if (locked) {
1472 err = f2fs_get_block(&dn, index);
1474 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1475 dn.data_blkaddr = ei.blk + index - ei.fofs;
1478 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1479 if (err || (!err && dn.data_blkaddr == NULL_ADDR)) {
1480 f2fs_put_dnode(&dn);
1488 /* convert_inline_page can make node_changed */
1489 *blk_addr = dn.data_blkaddr;
1490 *node_changed = dn.node_changed;
1492 f2fs_put_dnode(&dn);
1495 f2fs_unlock_op(sbi);
1499 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1500 loff_t pos, unsigned len, unsigned flags,
1501 struct page **pagep, void **fsdata)
1503 struct inode *inode = mapping->host;
1504 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1505 struct page *page = NULL;
1506 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1507 bool need_balance = false;
1508 block_t blkaddr = NULL_ADDR;
1511 trace_f2fs_write_begin(inode, pos, len, flags);
1514 * We should check this at this moment to avoid deadlock on inode page
1515 * and #0 page. The locking rule for inline_data conversion should be:
1516 * lock_page(page #0) -> lock_page(inode_page)
1519 err = f2fs_convert_inline_inode(inode);
1524 page = grab_cache_page_write_begin(mapping, index, flags);
1532 err = prepare_write_begin(sbi, page, pos, len,
1533 &blkaddr, &need_balance);
1537 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1539 f2fs_balance_fs(sbi, true);
1541 if (page->mapping != mapping) {
1542 /* The page got truncated from under us */
1543 f2fs_put_page(page, 1);
1548 f2fs_wait_on_page_writeback(page, DATA, false);
1550 /* wait for GCed encrypted page writeback */
1551 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1552 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1554 if (len == PAGE_CACHE_SIZE)
1556 if (PageUptodate(page))
1559 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1560 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1561 unsigned end = start + len;
1563 /* Reading beyond i_size is simple: memset to zero */
1564 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1568 if (blkaddr == NEW_ADDR) {
1569 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1571 struct f2fs_io_info fio = {
1575 .blk_addr = blkaddr,
1577 .encrypted_page = NULL,
1579 err = f2fs_submit_page_bio(&fio);
1584 if (unlikely(!PageUptodate(page))) {
1588 if (unlikely(page->mapping != mapping)) {
1589 f2fs_put_page(page, 1);
1593 /* avoid symlink page */
1594 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1595 err = f2fs_decrypt_one(inode, page);
1601 SetPageUptodate(page);
1603 clear_cold_data(page);
1607 f2fs_put_page(page, 1);
1608 f2fs_write_failed(mapping, pos + len);
1612 static int f2fs_write_end(struct file *file,
1613 struct address_space *mapping,
1614 loff_t pos, unsigned len, unsigned copied,
1615 struct page *page, void *fsdata)
1617 struct inode *inode = page->mapping->host;
1619 trace_f2fs_write_end(inode, pos, len, copied);
1621 set_page_dirty(page);
1623 if (pos + copied > i_size_read(inode)) {
1624 i_size_write(inode, pos + copied);
1625 mark_inode_dirty(inode);
1628 f2fs_put_page(page, 1);
1629 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1633 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1636 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1638 if (offset & blocksize_mask)
1641 if (iov_iter_alignment(iter) & blocksize_mask)
1647 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1650 struct address_space *mapping = iocb->ki_filp->f_mapping;
1651 struct inode *inode = mapping->host;
1652 size_t count = iov_iter_count(iter);
1655 err = check_direct_IO(inode, iter, offset);
1659 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1662 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1664 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1665 if (err < 0 && iov_iter_rw(iter) == WRITE)
1666 f2fs_write_failed(mapping, offset + count);
1668 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1673 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1674 unsigned int length)
1676 struct inode *inode = page->mapping->host;
1677 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1679 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1680 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1683 if (PageDirty(page)) {
1684 if (inode->i_ino == F2FS_META_INO(sbi))
1685 dec_page_count(sbi, F2FS_DIRTY_META);
1686 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1687 dec_page_count(sbi, F2FS_DIRTY_NODES);
1689 inode_dec_dirty_pages(inode);
1692 /* This is atomic written page, keep Private */
1693 if (IS_ATOMIC_WRITTEN_PAGE(page))
1696 ClearPagePrivate(page);
1699 int f2fs_release_page(struct page *page, gfp_t wait)
1701 /* If this is dirty page, keep PagePrivate */
1702 if (PageDirty(page))
1705 /* This is atomic written page, keep Private */
1706 if (IS_ATOMIC_WRITTEN_PAGE(page))
1709 ClearPagePrivate(page);
1713 static int f2fs_set_data_page_dirty(struct page *page)
1715 struct address_space *mapping = page->mapping;
1716 struct inode *inode = mapping->host;
1718 trace_f2fs_set_page_dirty(page, DATA);
1720 SetPageUptodate(page);
1722 if (f2fs_is_atomic_file(inode)) {
1723 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1724 register_inmem_page(inode, page);
1728 * Previously, this page has been registered, we just
1734 if (!PageDirty(page)) {
1735 __set_page_dirty_nobuffers(page);
1736 update_dirty_page(inode, page);
1742 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1744 struct inode *inode = mapping->host;
1746 if (f2fs_has_inline_data(inode))
1749 /* make sure allocating whole blocks */
1750 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1751 filemap_write_and_wait(mapping);
1753 return generic_block_bmap(mapping, block, get_data_block_bmap);
1756 const struct address_space_operations f2fs_dblock_aops = {
1757 .readpage = f2fs_read_data_page,
1758 .readpages = f2fs_read_data_pages,
1759 .writepage = f2fs_write_data_page,
1760 .writepages = f2fs_write_data_pages,
1761 .write_begin = f2fs_write_begin,
1762 .write_end = f2fs_write_end,
1763 .set_page_dirty = f2fs_set_data_page_dirty,
1764 .invalidatepage = f2fs_invalidate_page,
1765 .releasepage = f2fs_release_page,
1766 .direct_IO = f2fs_direct_IO,