4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
22 #include <linux/cleancache.h>
28 #include <trace/events/f2fs.h>
30 static void f2fs_read_end_io(struct bio *bio)
35 if (f2fs_bio_encrypted(bio)) {
37 f2fs_release_crypto_ctx(bio->bi_private);
39 f2fs_end_io_crypto_work(bio->bi_private, bio);
44 bio_for_each_segment_all(bvec, bio, i) {
45 struct page *page = bvec->bv_page;
48 SetPageUptodate(page);
50 ClearPageUptodate(page);
58 static void f2fs_write_end_io(struct bio *bio)
60 struct f2fs_sb_info *sbi = bio->bi_private;
64 bio_for_each_segment_all(bvec, bio, i) {
65 struct page *page = bvec->bv_page;
67 f2fs_restore_and_release_control_page(&page);
69 if (unlikely(bio->bi_error)) {
71 set_bit(AS_EIO, &page->mapping->flags);
72 f2fs_stop_checkpoint(sbi);
74 end_page_writeback(page);
75 dec_page_count(sbi, F2FS_WRITEBACK);
78 if (!get_pages(sbi, F2FS_WRITEBACK) &&
79 !list_empty(&sbi->cp_wait.task_list))
80 wake_up(&sbi->cp_wait);
86 * Low-level block read/write IO operations.
88 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
89 int npages, bool is_read)
93 bio = f2fs_bio_alloc(npages);
95 bio->bi_bdev = sbi->sb->s_bdev;
96 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
97 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
98 bio->bi_private = is_read ? NULL : sbi;
103 static void __submit_merged_bio(struct f2fs_bio_info *io)
105 struct f2fs_io_info *fio = &io->fio;
110 if (is_read_io(fio->rw))
111 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
113 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
115 submit_bio(fio->rw, io->bio);
119 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
120 enum page_type type, int rw)
122 enum page_type btype = PAGE_TYPE_OF_BIO(type);
123 struct f2fs_bio_info *io;
125 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
127 down_write(&io->io_rwsem);
129 /* change META to META_FLUSH in the checkpoint procedure */
130 if (type >= META_FLUSH) {
131 io->fio.type = META_FLUSH;
132 if (test_opt(sbi, NOBARRIER))
133 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
135 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
137 __submit_merged_bio(io);
138 up_write(&io->io_rwsem);
142 * Fill the locked page with data located in the block address.
143 * Return unlocked page.
145 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
148 struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
150 trace_f2fs_submit_page_bio(page, fio);
151 f2fs_trace_ios(fio, 0);
153 /* Allocate a new bio */
154 bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
156 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
161 submit_bio(fio->rw, bio);
165 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
167 struct f2fs_sb_info *sbi = fio->sbi;
168 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
169 struct f2fs_bio_info *io;
170 bool is_read = is_read_io(fio->rw);
171 struct page *bio_page;
173 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
175 verify_block_addr(sbi, fio->blk_addr);
177 down_write(&io->io_rwsem);
180 inc_page_count(sbi, F2FS_WRITEBACK);
182 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
183 io->fio.rw != fio->rw))
184 __submit_merged_bio(io);
186 if (io->bio == NULL) {
187 int bio_blocks = MAX_BIO_BLOCKS(sbi);
189 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
193 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
195 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
197 __submit_merged_bio(io);
201 io->last_block_in_bio = fio->blk_addr;
202 f2fs_trace_ios(fio, 0);
204 up_write(&io->io_rwsem);
205 trace_f2fs_submit_page_mbio(fio->page, fio);
209 * Lock ordering for the change of data block address:
212 * update block addresses in the node page
214 void set_data_blkaddr(struct dnode_of_data *dn)
216 struct f2fs_node *rn;
218 struct page *node_page = dn->node_page;
219 unsigned int ofs_in_node = dn->ofs_in_node;
221 f2fs_wait_on_page_writeback(node_page, NODE);
223 rn = F2FS_NODE(node_page);
225 /* Get physical address of data block */
226 addr_array = blkaddr_in_node(rn);
227 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
228 if (set_page_dirty(node_page))
229 dn->node_changed = true;
232 int reserve_new_block(struct dnode_of_data *dn)
234 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
236 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
238 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
241 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
243 dn->data_blkaddr = NEW_ADDR;
244 set_data_blkaddr(dn);
245 mark_inode_dirty(dn->inode);
250 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
252 bool need_put = dn->inode_page ? false : true;
255 err = get_dnode_of_data(dn, index, ALLOC_NODE);
259 if (dn->data_blkaddr == NULL_ADDR)
260 err = reserve_new_block(dn);
266 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
268 struct extent_info ei;
269 struct inode *inode = dn->inode;
271 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
272 dn->data_blkaddr = ei.blk + index - ei.fofs;
276 return f2fs_reserve_block(dn, index);
279 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
280 int rw, bool for_write)
282 struct address_space *mapping = inode->i_mapping;
283 struct dnode_of_data dn;
285 struct extent_info ei;
287 struct f2fs_io_info fio = {
288 .sbi = F2FS_I_SB(inode),
291 .encrypted_page = NULL,
294 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
295 return read_mapping_page(mapping, index, NULL);
297 page = f2fs_grab_cache_page(mapping, index, for_write);
299 return ERR_PTR(-ENOMEM);
301 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
302 dn.data_blkaddr = ei.blk + index - ei.fofs;
306 set_new_dnode(&dn, inode, NULL, NULL, 0);
307 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
312 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
317 if (PageUptodate(page)) {
323 * A new dentry page is allocated but not able to be written, since its
324 * new inode page couldn't be allocated due to -ENOSPC.
325 * In such the case, its blkaddr can be remained as NEW_ADDR.
326 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
328 if (dn.data_blkaddr == NEW_ADDR) {
329 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
330 SetPageUptodate(page);
335 fio.blk_addr = dn.data_blkaddr;
337 err = f2fs_submit_page_bio(&fio);
343 f2fs_put_page(page, 1);
347 struct page *find_data_page(struct inode *inode, pgoff_t index)
349 struct address_space *mapping = inode->i_mapping;
352 page = find_get_page(mapping, index);
353 if (page && PageUptodate(page))
355 f2fs_put_page(page, 0);
357 page = get_read_data_page(inode, index, READ_SYNC, false);
361 if (PageUptodate(page))
364 wait_on_page_locked(page);
365 if (unlikely(!PageUptodate(page))) {
366 f2fs_put_page(page, 0);
367 return ERR_PTR(-EIO);
373 * If it tries to access a hole, return an error.
374 * Because, the callers, functions in dir.c and GC, should be able to know
375 * whether this page exists or not.
377 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
380 struct address_space *mapping = inode->i_mapping;
383 page = get_read_data_page(inode, index, READ_SYNC, for_write);
387 /* wait for read completion */
389 if (unlikely(!PageUptodate(page))) {
390 f2fs_put_page(page, 1);
391 return ERR_PTR(-EIO);
393 if (unlikely(page->mapping != mapping)) {
394 f2fs_put_page(page, 1);
401 * Caller ensures that this data page is never allocated.
402 * A new zero-filled data page is allocated in the page cache.
404 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
406 * Note that, ipage is set only by make_empty_dir, and if any error occur,
407 * ipage should be released by this function.
409 struct page *get_new_data_page(struct inode *inode,
410 struct page *ipage, pgoff_t index, bool new_i_size)
412 struct address_space *mapping = inode->i_mapping;
414 struct dnode_of_data dn;
417 page = f2fs_grab_cache_page(mapping, index, true);
420 * before exiting, we should make sure ipage will be released
421 * if any error occur.
423 f2fs_put_page(ipage, 1);
424 return ERR_PTR(-ENOMEM);
427 set_new_dnode(&dn, inode, ipage, NULL, 0);
428 err = f2fs_reserve_block(&dn, index);
430 f2fs_put_page(page, 1);
436 if (PageUptodate(page))
439 if (dn.data_blkaddr == NEW_ADDR) {
440 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
441 SetPageUptodate(page);
443 f2fs_put_page(page, 1);
445 /* if ipage exists, blkaddr should be NEW_ADDR */
446 f2fs_bug_on(F2FS_I_SB(inode), ipage);
447 page = get_lock_data_page(inode, index, true);
452 if (new_i_size && i_size_read(inode) <
453 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
454 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
455 /* Only the directory inode sets new_i_size */
456 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
461 static int __allocate_data_block(struct dnode_of_data *dn)
463 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
464 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
465 struct f2fs_summary sum;
467 int seg = CURSEG_WARM_DATA;
470 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
473 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
474 if (dn->data_blkaddr == NEW_ADDR)
477 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
481 get_node_info(sbi, dn->nid, &ni);
482 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
484 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
485 seg = CURSEG_DIRECT_IO;
487 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
489 set_data_blkaddr(dn);
492 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
494 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
495 i_size_write(dn->inode,
496 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
500 static int __allocate_data_blocks(struct inode *inode, loff_t offset,
503 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
504 struct dnode_of_data dn;
505 u64 start = F2FS_BYTES_TO_BLK(offset);
506 u64 len = F2FS_BYTES_TO_BLK(count);
514 /* When reading holes, we need its node page */
515 set_new_dnode(&dn, inode, NULL, NULL, 0);
516 err = get_dnode_of_data(&dn, start, ALLOC_NODE);
521 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
523 while (dn.ofs_in_node < end_offset && len) {
526 if (unlikely(f2fs_cp_error(sbi))) {
531 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
532 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
533 err = __allocate_data_block(&dn);
544 sync_inode_page(&dn);
549 f2fs_balance_fs(sbi, dn.node_changed);
555 sync_inode_page(&dn);
559 f2fs_balance_fs(sbi, dn.node_changed);
564 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
565 * f2fs_map_blocks structure.
566 * If original data blocks are allocated, then give them to blockdev.
568 * a. preallocate requested block addresses
569 * b. do not use extent cache for better performance
570 * c. give the block addresses to blockdev
572 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
573 int create, int flag)
575 unsigned int maxblocks = map->m_len;
576 struct dnode_of_data dn;
577 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
578 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
579 pgoff_t pgofs, end_offset;
580 int err = 0, ofs = 1;
581 struct extent_info ei;
582 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;
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 (dn.data_blkaddr == NEW_ADDR || dn.data_blkaddr == NULL_ADDR) {
612 if (unlikely(f2fs_cp_error(sbi))) {
616 err = __allocate_data_block(&dn);
620 map->m_flags = F2FS_MAP_NEW;
622 if (flag != F2FS_GET_BLOCK_FIEMAP ||
623 dn.data_blkaddr != NEW_ADDR) {
624 if (flag == F2FS_GET_BLOCK_BMAP)
630 * preallocated unwritten block should be mapped
633 if (dn.data_blkaddr == NEW_ADDR)
634 map->m_flags = F2FS_MAP_UNWRITTEN;
638 map->m_flags |= F2FS_MAP_MAPPED;
639 map->m_pblk = dn.data_blkaddr;
642 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
647 if (map->m_len >= maxblocks)
650 if (dn.ofs_in_node >= end_offset) {
652 sync_inode_page(&dn);
658 f2fs_balance_fs(sbi, dn.node_changed);
662 set_new_dnode(&dn, inode, NULL, NULL, 0);
663 err = get_dnode_of_data(&dn, pgofs, mode);
670 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
673 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
675 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
677 if (unlikely(f2fs_cp_error(sbi))) {
681 err = __allocate_data_block(&dn);
685 map->m_flags |= F2FS_MAP_NEW;
686 blkaddr = dn.data_blkaddr;
689 * we only merge preallocated unwritten blocks
692 if (flag != F2FS_GET_BLOCK_FIEMAP ||
698 /* Give more consecutive addresses for the readahead */
699 if ((map->m_pblk != NEW_ADDR &&
700 blkaddr == (map->m_pblk + ofs)) ||
701 (map->m_pblk == NEW_ADDR &&
702 blkaddr == NEW_ADDR)) {
712 sync_inode_page(&dn);
718 f2fs_balance_fs(sbi, dn.node_changed);
721 trace_f2fs_map_blocks(inode, map, err);
725 static int __get_data_block(struct inode *inode, sector_t iblock,
726 struct buffer_head *bh, int create, int flag)
728 struct f2fs_map_blocks map;
732 map.m_len = bh->b_size >> inode->i_blkbits;
734 ret = f2fs_map_blocks(inode, &map, create, flag);
736 map_bh(bh, inode->i_sb, map.m_pblk);
737 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
738 bh->b_size = map.m_len << inode->i_blkbits;
743 static int get_data_block(struct inode *inode, sector_t iblock,
744 struct buffer_head *bh_result, int create, int flag)
746 return __get_data_block(inode, iblock, bh_result, create, flag);
749 static int get_data_block_dio(struct inode *inode, sector_t iblock,
750 struct buffer_head *bh_result, int create)
752 return __get_data_block(inode, iblock, bh_result, create,
756 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
757 struct buffer_head *bh_result, int create)
759 /* Block number less than F2FS MAX BLOCKS */
760 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
763 return __get_data_block(inode, iblock, bh_result, create,
764 F2FS_GET_BLOCK_BMAP);
767 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
769 return (offset >> inode->i_blkbits);
772 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
774 return (blk << inode->i_blkbits);
777 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
780 struct buffer_head map_bh;
781 sector_t start_blk, last_blk;
783 u64 logical = 0, phys = 0, size = 0;
787 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
791 if (f2fs_has_inline_data(inode)) {
792 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
799 isize = i_size_read(inode);
803 if (start + len > isize)
806 if (logical_to_blk(inode, len) == 0)
807 len = blk_to_logical(inode, 1);
809 start_blk = logical_to_blk(inode, start);
810 last_blk = logical_to_blk(inode, start + len - 1);
813 memset(&map_bh, 0, sizeof(struct buffer_head));
816 ret = get_data_block(inode, start_blk, &map_bh, 0,
817 F2FS_GET_BLOCK_FIEMAP);
822 if (!buffer_mapped(&map_bh)) {
823 /* Go through holes util pass the EOF */
824 if (blk_to_logical(inode, start_blk++) < isize)
826 /* Found a hole beyond isize means no more extents.
827 * Note that the premise is that filesystems don't
828 * punch holes beyond isize and keep size unchanged.
830 flags |= FIEMAP_EXTENT_LAST;
834 if (f2fs_encrypted_inode(inode))
835 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
837 ret = fiemap_fill_next_extent(fieinfo, logical,
841 if (start_blk > last_blk || ret)
844 logical = blk_to_logical(inode, start_blk);
845 phys = blk_to_logical(inode, map_bh.b_blocknr);
846 size = map_bh.b_size;
848 if (buffer_unwritten(&map_bh))
849 flags = FIEMAP_EXTENT_UNWRITTEN;
851 start_blk += logical_to_blk(inode, size);
855 if (fatal_signal_pending(current))
868 * This function was originally taken from fs/mpage.c, and customized for f2fs.
869 * Major change was from block_size == page_size in f2fs by default.
871 static int f2fs_mpage_readpages(struct address_space *mapping,
872 struct list_head *pages, struct page *page,
875 struct bio *bio = NULL;
877 sector_t last_block_in_bio = 0;
878 struct inode *inode = mapping->host;
879 const unsigned blkbits = inode->i_blkbits;
880 const unsigned blocksize = 1 << blkbits;
881 sector_t block_in_file;
883 sector_t last_block_in_file;
885 struct block_device *bdev = inode->i_sb->s_bdev;
886 struct f2fs_map_blocks map;
893 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
895 prefetchw(&page->flags);
897 page = list_entry(pages->prev, struct page, lru);
898 list_del(&page->lru);
899 if (add_to_page_cache_lru(page, mapping,
900 page->index, GFP_KERNEL))
904 block_in_file = (sector_t)page->index;
905 last_block = block_in_file + nr_pages;
906 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
908 if (last_block > last_block_in_file)
909 last_block = last_block_in_file;
912 * Map blocks using the previous result first.
914 if ((map.m_flags & F2FS_MAP_MAPPED) &&
915 block_in_file > map.m_lblk &&
916 block_in_file < (map.m_lblk + map.m_len))
920 * Then do more f2fs_map_blocks() calls until we are
921 * done with this page.
925 if (block_in_file < last_block) {
926 map.m_lblk = block_in_file;
927 map.m_len = last_block - block_in_file;
929 if (f2fs_map_blocks(inode, &map, 0,
930 F2FS_GET_BLOCK_READ))
934 if ((map.m_flags & F2FS_MAP_MAPPED)) {
935 block_nr = map.m_pblk + block_in_file - map.m_lblk;
936 SetPageMappedToDisk(page);
938 if (!PageUptodate(page) && !cleancache_get_page(page)) {
939 SetPageUptodate(page);
943 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
944 SetPageUptodate(page);
950 * This page will go to BIO. Do we need to send this
953 if (bio && (last_block_in_bio != block_nr - 1)) {
955 submit_bio(READ, bio);
959 struct f2fs_crypto_ctx *ctx = NULL;
961 if (f2fs_encrypted_inode(inode) &&
962 S_ISREG(inode->i_mode)) {
964 ctx = f2fs_get_crypto_ctx(inode);
968 /* wait the page to be moved by cleaning */
969 f2fs_wait_on_encrypted_page_writeback(
970 F2FS_I_SB(inode), block_nr);
973 bio = bio_alloc(GFP_KERNEL,
974 min_t(int, nr_pages, BIO_MAX_PAGES));
977 f2fs_release_crypto_ctx(ctx);
981 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
982 bio->bi_end_io = f2fs_read_end_io;
983 bio->bi_private = ctx;
986 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
987 goto submit_and_realloc;
989 last_block_in_bio = block_nr;
993 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
998 submit_bio(READ, bio);
1004 page_cache_release(page);
1006 BUG_ON(pages && !list_empty(pages));
1008 submit_bio(READ, bio);
1012 static int f2fs_read_data_page(struct file *file, struct page *page)
1014 struct inode *inode = page->mapping->host;
1017 trace_f2fs_readpage(page, DATA);
1019 /* If the file has inline data, try to read it directly */
1020 if (f2fs_has_inline_data(inode))
1021 ret = f2fs_read_inline_data(inode, page);
1023 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1027 static int f2fs_read_data_pages(struct file *file,
1028 struct address_space *mapping,
1029 struct list_head *pages, unsigned nr_pages)
1031 struct inode *inode = file->f_mapping->host;
1032 struct page *page = list_entry(pages->prev, struct page, lru);
1034 trace_f2fs_readpages(inode, page, nr_pages);
1036 /* If the file has inline data, skip readpages */
1037 if (f2fs_has_inline_data(inode))
1040 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1043 int do_write_data_page(struct f2fs_io_info *fio)
1045 struct page *page = fio->page;
1046 struct inode *inode = page->mapping->host;
1047 struct dnode_of_data dn;
1050 set_new_dnode(&dn, inode, NULL, NULL, 0);
1051 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1055 fio->blk_addr = dn.data_blkaddr;
1057 /* This page is already truncated */
1058 if (fio->blk_addr == NULL_ADDR) {
1059 ClearPageUptodate(page);
1063 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1065 /* wait for GCed encrypted page writeback */
1066 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1069 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1070 if (IS_ERR(fio->encrypted_page)) {
1071 err = PTR_ERR(fio->encrypted_page);
1076 set_page_writeback(page);
1079 * If current allocation needs SSR,
1080 * it had better in-place writes for updated data.
1082 if (unlikely(fio->blk_addr != NEW_ADDR &&
1083 !is_cold_data(page) &&
1084 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1085 need_inplace_update(inode))) {
1086 rewrite_data_page(fio);
1087 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1088 trace_f2fs_do_write_data_page(page, IPU);
1090 write_data_page(&dn, fio);
1091 set_data_blkaddr(&dn);
1092 f2fs_update_extent_cache(&dn);
1093 trace_f2fs_do_write_data_page(page, OPU);
1094 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1095 if (page->index == 0)
1096 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1099 f2fs_put_dnode(&dn);
1103 static int f2fs_write_data_page(struct page *page,
1104 struct writeback_control *wbc)
1106 struct inode *inode = page->mapping->host;
1107 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1108 loff_t i_size = i_size_read(inode);
1109 const pgoff_t end_index = ((unsigned long long) i_size)
1110 >> PAGE_CACHE_SHIFT;
1111 unsigned offset = 0;
1112 bool need_balance_fs = false;
1114 struct f2fs_io_info fio = {
1117 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1119 .encrypted_page = NULL,
1122 trace_f2fs_writepage(page, DATA);
1124 if (page->index < end_index)
1128 * If the offset is out-of-range of file size,
1129 * this page does not have to be written to disk.
1131 offset = i_size & (PAGE_CACHE_SIZE - 1);
1132 if ((page->index >= end_index + 1) || !offset)
1135 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1137 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1139 if (f2fs_is_drop_cache(inode))
1141 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1142 available_free_memory(sbi, BASE_CHECK))
1145 /* Dentry blocks are controlled by checkpoint */
1146 if (S_ISDIR(inode->i_mode)) {
1147 if (unlikely(f2fs_cp_error(sbi)))
1149 err = do_write_data_page(&fio);
1153 /* we should bypass data pages to proceed the kworkder jobs */
1154 if (unlikely(f2fs_cp_error(sbi))) {
1159 if (!wbc->for_reclaim)
1160 need_balance_fs = true;
1161 else if (has_not_enough_free_secs(sbi, 0))
1166 if (f2fs_has_inline_data(inode))
1167 err = f2fs_write_inline_data(inode, page);
1169 err = do_write_data_page(&fio);
1170 f2fs_unlock_op(sbi);
1172 if (err && err != -ENOENT)
1175 clear_cold_data(page);
1177 inode_dec_dirty_pages(inode);
1179 ClearPageUptodate(page);
1181 f2fs_balance_fs(sbi, need_balance_fs);
1182 if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi))) {
1183 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1184 remove_dirty_inode(inode);
1189 redirty_page_for_writepage(wbc, page);
1190 return AOP_WRITEPAGE_ACTIVATE;
1193 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1196 struct address_space *mapping = data;
1197 int ret = mapping->a_ops->writepage(page, wbc);
1198 mapping_set_error(mapping, ret);
1203 * This function was copied from write_cche_pages from mm/page-writeback.c.
1204 * The major change is making write step of cold data page separately from
1205 * warm/hot data page.
1207 static int f2fs_write_cache_pages(struct address_space *mapping,
1208 struct writeback_control *wbc, writepage_t writepage,
1213 struct pagevec pvec;
1215 pgoff_t uninitialized_var(writeback_index);
1217 pgoff_t end; /* Inclusive */
1220 int range_whole = 0;
1224 pagevec_init(&pvec, 0);
1226 if (wbc->range_cyclic) {
1227 writeback_index = mapping->writeback_index; /* prev offset */
1228 index = writeback_index;
1235 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1236 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1237 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1239 cycled = 1; /* ignore range_cyclic tests */
1241 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1242 tag = PAGECACHE_TAG_TOWRITE;
1244 tag = PAGECACHE_TAG_DIRTY;
1246 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1247 tag_pages_for_writeback(mapping, index, end);
1249 while (!done && (index <= end)) {
1252 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1253 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1257 for (i = 0; i < nr_pages; i++) {
1258 struct page *page = pvec.pages[i];
1260 if (page->index > end) {
1265 done_index = page->index;
1269 if (unlikely(page->mapping != mapping)) {
1275 if (!PageDirty(page)) {
1276 /* someone wrote it for us */
1277 goto continue_unlock;
1280 if (step == is_cold_data(page))
1281 goto continue_unlock;
1283 if (PageWriteback(page)) {
1284 if (wbc->sync_mode != WB_SYNC_NONE)
1285 f2fs_wait_on_page_writeback(page, DATA);
1287 goto continue_unlock;
1290 BUG_ON(PageWriteback(page));
1291 if (!clear_page_dirty_for_io(page))
1292 goto continue_unlock;
1294 ret = (*writepage)(page, wbc, data);
1295 if (unlikely(ret)) {
1296 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1300 done_index = page->index + 1;
1306 if (--wbc->nr_to_write <= 0 &&
1307 wbc->sync_mode == WB_SYNC_NONE) {
1312 pagevec_release(&pvec);
1321 if (!cycled && !done) {
1324 end = writeback_index - 1;
1327 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1328 mapping->writeback_index = done_index;
1333 static int f2fs_write_data_pages(struct address_space *mapping,
1334 struct writeback_control *wbc)
1336 struct inode *inode = mapping->host;
1337 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1338 bool locked = false;
1342 trace_f2fs_writepages(mapping->host, wbc, DATA);
1344 /* deal with chardevs and other special file */
1345 if (!mapping->a_ops->writepage)
1348 /* skip writing if there is no dirty page in this inode */
1349 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1352 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1353 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1354 available_free_memory(sbi, DIRTY_DENTS))
1357 /* skip writing during file defragment */
1358 if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1361 /* during POR, we don't need to trigger writepage at all. */
1362 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1365 diff = nr_pages_to_write(sbi, DATA, wbc);
1367 if (!S_ISDIR(inode->i_mode)) {
1368 mutex_lock(&sbi->writepages);
1371 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1372 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1374 mutex_unlock(&sbi->writepages);
1376 remove_dirty_inode(inode);
1378 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1382 wbc->pages_skipped += get_dirty_pages(inode);
1386 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1388 struct inode *inode = mapping->host;
1389 loff_t i_size = i_size_read(inode);
1392 truncate_pagecache(inode, i_size);
1393 truncate_blocks(inode, i_size, true);
1397 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1398 struct page *page, loff_t pos, unsigned len,
1399 block_t *blk_addr, bool *node_changed)
1401 struct inode *inode = page->mapping->host;
1402 pgoff_t index = page->index;
1403 struct dnode_of_data dn;
1405 bool locked = false;
1406 struct extent_info ei;
1409 if (f2fs_has_inline_data(inode) ||
1410 (pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1415 /* check inline_data */
1416 ipage = get_node_page(sbi, inode->i_ino);
1417 if (IS_ERR(ipage)) {
1418 err = PTR_ERR(ipage);
1422 set_new_dnode(&dn, inode, ipage, ipage, 0);
1424 if (f2fs_has_inline_data(inode)) {
1425 if (pos + len <= MAX_INLINE_DATA) {
1426 read_inline_data(page, ipage);
1427 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1428 sync_inode_page(&dn);
1430 err = f2fs_convert_inline_page(&dn, page);
1433 if (dn.data_blkaddr == NULL_ADDR)
1434 err = f2fs_get_block(&dn, index);
1436 } else if (locked) {
1437 err = f2fs_get_block(&dn, index);
1439 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1440 dn.data_blkaddr = ei.blk + index - ei.fofs;
1442 bool restart = false;
1445 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1446 if (err || (!err && dn.data_blkaddr == NULL_ADDR))
1449 f2fs_put_dnode(&dn);
1457 /* convert_inline_page can make node_changed */
1458 *blk_addr = dn.data_blkaddr;
1459 *node_changed = dn.node_changed;
1461 f2fs_put_dnode(&dn);
1464 f2fs_unlock_op(sbi);
1468 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1469 loff_t pos, unsigned len, unsigned flags,
1470 struct page **pagep, void **fsdata)
1472 struct inode *inode = mapping->host;
1473 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1474 struct page *page = NULL;
1475 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1476 bool need_balance = false;
1477 block_t blkaddr = NULL_ADDR;
1480 trace_f2fs_write_begin(inode, pos, len, flags);
1483 * We should check this at this moment to avoid deadlock on inode page
1484 * and #0 page. The locking rule for inline_data conversion should be:
1485 * lock_page(page #0) -> lock_page(inode_page)
1488 err = f2fs_convert_inline_inode(inode);
1493 page = grab_cache_page_write_begin(mapping, index, flags);
1501 err = prepare_write_begin(sbi, page, pos, len,
1502 &blkaddr, &need_balance);
1506 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1508 f2fs_balance_fs(sbi, true);
1510 if (page->mapping != mapping) {
1511 /* The page got truncated from under us */
1512 f2fs_put_page(page, 1);
1517 f2fs_wait_on_page_writeback(page, DATA);
1519 /* wait for GCed encrypted page writeback */
1520 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1521 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1523 if (len == PAGE_CACHE_SIZE)
1525 if (PageUptodate(page))
1528 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1529 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1530 unsigned end = start + len;
1532 /* Reading beyond i_size is simple: memset to zero */
1533 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1537 if (blkaddr == NEW_ADDR) {
1538 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1540 struct f2fs_io_info fio = {
1544 .blk_addr = blkaddr,
1546 .encrypted_page = NULL,
1548 err = f2fs_submit_page_bio(&fio);
1553 if (unlikely(!PageUptodate(page))) {
1557 if (unlikely(page->mapping != mapping)) {
1558 f2fs_put_page(page, 1);
1562 /* avoid symlink page */
1563 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1564 err = f2fs_decrypt_one(inode, page);
1570 SetPageUptodate(page);
1572 clear_cold_data(page);
1576 f2fs_put_page(page, 1);
1577 f2fs_write_failed(mapping, pos + len);
1581 static int f2fs_write_end(struct file *file,
1582 struct address_space *mapping,
1583 loff_t pos, unsigned len, unsigned copied,
1584 struct page *page, void *fsdata)
1586 struct inode *inode = page->mapping->host;
1588 trace_f2fs_write_end(inode, pos, len, copied);
1590 set_page_dirty(page);
1592 if (pos + copied > i_size_read(inode)) {
1593 i_size_write(inode, pos + copied);
1594 mark_inode_dirty(inode);
1595 update_inode_page(inode);
1598 f2fs_put_page(page, 1);
1599 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1603 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1606 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1608 if (offset & blocksize_mask)
1611 if (iov_iter_alignment(iter) & blocksize_mask)
1617 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1620 struct file *file = iocb->ki_filp;
1621 struct address_space *mapping = file->f_mapping;
1622 struct inode *inode = mapping->host;
1623 size_t count = iov_iter_count(iter);
1626 /* we don't need to use inline_data strictly */
1627 err = f2fs_convert_inline_inode(inode);
1631 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1634 err = check_direct_IO(inode, iter, offset);
1638 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1640 if (iov_iter_rw(iter) == WRITE) {
1641 err = __allocate_data_blocks(inode, offset, count);
1646 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1648 if (err < 0 && iov_iter_rw(iter) == WRITE)
1649 f2fs_write_failed(mapping, offset + count);
1651 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1656 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1657 unsigned int length)
1659 struct inode *inode = page->mapping->host;
1660 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1662 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1663 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1666 if (PageDirty(page)) {
1667 if (inode->i_ino == F2FS_META_INO(sbi))
1668 dec_page_count(sbi, F2FS_DIRTY_META);
1669 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1670 dec_page_count(sbi, F2FS_DIRTY_NODES);
1672 inode_dec_dirty_pages(inode);
1675 /* This is atomic written page, keep Private */
1676 if (IS_ATOMIC_WRITTEN_PAGE(page))
1679 ClearPagePrivate(page);
1682 int f2fs_release_page(struct page *page, gfp_t wait)
1684 /* If this is dirty page, keep PagePrivate */
1685 if (PageDirty(page))
1688 /* This is atomic written page, keep Private */
1689 if (IS_ATOMIC_WRITTEN_PAGE(page))
1692 ClearPagePrivate(page);
1696 static int f2fs_set_data_page_dirty(struct page *page)
1698 struct address_space *mapping = page->mapping;
1699 struct inode *inode = mapping->host;
1701 trace_f2fs_set_page_dirty(page, DATA);
1703 SetPageUptodate(page);
1705 if (f2fs_is_atomic_file(inode)) {
1706 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1707 register_inmem_page(inode, page);
1711 * Previously, this page has been registered, we just
1717 if (!PageDirty(page)) {
1718 __set_page_dirty_nobuffers(page);
1719 update_dirty_page(inode, page);
1725 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1727 struct inode *inode = mapping->host;
1729 if (f2fs_has_inline_data(inode))
1732 /* make sure allocating whole blocks */
1733 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1734 filemap_write_and_wait(mapping);
1736 return generic_block_bmap(mapping, block, get_data_block_bmap);
1739 const struct address_space_operations f2fs_dblock_aops = {
1740 .readpage = f2fs_read_data_page,
1741 .readpages = f2fs_read_data_pages,
1742 .writepage = f2fs_write_data_page,
1743 .writepages = f2fs_write_data_pages,
1744 .write_begin = f2fs_write_begin,
1745 .write_end = f2fs_write_end,
1746 .set_page_dirty = f2fs_set_data_page_dirty,
1747 .invalidatepage = f2fs_invalidate_page,
1748 .releasepage = f2fs_release_page,
1749 .direct_IO = f2fs_direct_IO,