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/aio.h>
16 #include <linux/writeback.h>
17 #include <linux/backing-dev.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
25 #include <trace/events/f2fs.h>
27 static void f2fs_read_end_io(struct bio *bio, int err)
32 bio_for_each_segment_all(bvec, bio, i) {
33 struct page *page = bvec->bv_page;
36 SetPageUptodate(page);
38 ClearPageUptodate(page);
46 static void f2fs_write_end_io(struct bio *bio, int err)
48 struct f2fs_sb_info *sbi = bio->bi_private;
52 bio_for_each_segment_all(bvec, bio, i) {
53 struct page *page = bvec->bv_page;
57 set_bit(AS_EIO, &page->mapping->flags);
58 f2fs_stop_checkpoint(sbi);
60 end_page_writeback(page);
61 dec_page_count(sbi, F2FS_WRITEBACK);
65 complete(sbi->wait_io);
69 if (!get_pages(sbi, F2FS_WRITEBACK) &&
70 !list_empty(&sbi->cp_wait.task_list))
71 wake_up(&sbi->cp_wait);
77 * Low-level block read/write IO operations.
79 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
80 int npages, bool is_read)
84 /* No failure on bio allocation */
85 bio = bio_alloc(GFP_NOIO, npages);
87 bio->bi_bdev = sbi->sb->s_bdev;
88 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
89 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
90 bio->bi_private = sbi;
95 static void __submit_merged_bio(struct f2fs_bio_info *io)
97 struct f2fs_io_info *fio = &io->fio;
105 if (is_read_io(rw)) {
106 trace_f2fs_submit_read_bio(io->sbi->sb, rw,
108 submit_bio(rw, io->bio);
110 trace_f2fs_submit_write_bio(io->sbi->sb, rw,
113 * META_FLUSH is only from the checkpoint procedure, and we
114 * should wait this metadata bio for FS consistency.
116 if (fio->type == META_FLUSH) {
117 DECLARE_COMPLETION_ONSTACK(wait);
118 io->sbi->wait_io = &wait;
119 submit_bio(rw, io->bio);
120 wait_for_completion(&wait);
122 submit_bio(rw, io->bio);
129 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
130 enum page_type type, int rw)
132 enum page_type btype = PAGE_TYPE_OF_BIO(type);
133 struct f2fs_bio_info *io;
135 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
137 down_write(&io->io_rwsem);
139 /* change META to META_FLUSH in the checkpoint procedure */
140 if (type >= META_FLUSH) {
141 io->fio.type = META_FLUSH;
142 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
144 __submit_merged_bio(io);
145 up_write(&io->io_rwsem);
149 * Fill the locked page with data located in the block address.
150 * Return unlocked page.
152 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
153 block_t blk_addr, int rw)
157 trace_f2fs_submit_page_bio(page, blk_addr, rw);
159 /* Allocate a new bio */
160 bio = __bio_alloc(sbi, blk_addr, 1, is_read_io(rw));
162 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
164 f2fs_put_page(page, 1);
172 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
173 block_t blk_addr, struct f2fs_io_info *fio)
175 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
176 struct f2fs_bio_info *io;
177 bool is_read = is_read_io(fio->rw);
179 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
181 verify_block_addr(sbi, blk_addr);
183 down_write(&io->io_rwsem);
186 inc_page_count(sbi, F2FS_WRITEBACK);
188 if (io->bio && (io->last_block_in_bio != blk_addr - 1 ||
189 io->fio.rw != fio->rw))
190 __submit_merged_bio(io);
192 if (io->bio == NULL) {
193 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
195 io->bio = __bio_alloc(sbi, blk_addr, bio_blocks, is_read);
199 if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
201 __submit_merged_bio(io);
205 io->last_block_in_bio = blk_addr;
207 up_write(&io->io_rwsem);
208 trace_f2fs_submit_page_mbio(page, fio->rw, fio->type, blk_addr);
212 * Lock ordering for the change of data block address:
215 * update block addresses in the node page
217 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
219 struct f2fs_node *rn;
221 struct page *node_page = dn->node_page;
222 unsigned int ofs_in_node = dn->ofs_in_node;
224 f2fs_wait_on_page_writeback(node_page, NODE);
226 rn = F2FS_NODE(node_page);
228 /* Get physical address of data block */
229 addr_array = blkaddr_in_node(rn);
230 addr_array[ofs_in_node] = cpu_to_le32(new_addr);
231 set_page_dirty(node_page);
234 int reserve_new_block(struct dnode_of_data *dn)
236 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
238 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
240 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
243 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
245 __set_data_blkaddr(dn, NEW_ADDR);
246 dn->data_blkaddr = NEW_ADDR;
247 mark_inode_dirty(dn->inode);
252 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
254 bool need_put = dn->inode_page ? false : true;
257 /* if inode_page exists, index should be zero */
258 f2fs_bug_on(!need_put && index);
260 err = get_dnode_of_data(dn, index, ALLOC_NODE);
264 if (dn->data_blkaddr == NULL_ADDR)
265 err = reserve_new_block(dn);
271 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
272 struct buffer_head *bh_result)
274 struct f2fs_inode_info *fi = F2FS_I(inode);
275 pgoff_t start_fofs, end_fofs;
276 block_t start_blkaddr;
278 if (is_inode_flag_set(fi, FI_NO_EXTENT))
281 read_lock(&fi->ext.ext_lock);
282 if (fi->ext.len == 0) {
283 read_unlock(&fi->ext.ext_lock);
287 stat_inc_total_hit(inode->i_sb);
289 start_fofs = fi->ext.fofs;
290 end_fofs = fi->ext.fofs + fi->ext.len - 1;
291 start_blkaddr = fi->ext.blk_addr;
293 if (pgofs >= start_fofs && pgofs <= end_fofs) {
294 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
297 clear_buffer_new(bh_result);
298 map_bh(bh_result, inode->i_sb,
299 start_blkaddr + pgofs - start_fofs);
300 count = end_fofs - pgofs + 1;
301 if (count < (UINT_MAX >> blkbits))
302 bh_result->b_size = (count << blkbits);
304 bh_result->b_size = UINT_MAX;
306 stat_inc_read_hit(inode->i_sb);
307 read_unlock(&fi->ext.ext_lock);
310 read_unlock(&fi->ext.ext_lock);
314 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
316 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
317 pgoff_t fofs, start_fofs, end_fofs;
318 block_t start_blkaddr, end_blkaddr;
319 int need_update = true;
321 f2fs_bug_on(blk_addr == NEW_ADDR);
322 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
325 /* Update the page address in the parent node */
326 __set_data_blkaddr(dn, blk_addr);
328 if (is_inode_flag_set(fi, FI_NO_EXTENT))
331 write_lock(&fi->ext.ext_lock);
333 start_fofs = fi->ext.fofs;
334 end_fofs = fi->ext.fofs + fi->ext.len - 1;
335 start_blkaddr = fi->ext.blk_addr;
336 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
338 /* Drop and initialize the matched extent */
339 if (fi->ext.len == 1 && fofs == start_fofs)
343 if (fi->ext.len == 0) {
344 if (blk_addr != NULL_ADDR) {
346 fi->ext.blk_addr = blk_addr;
353 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
361 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
366 /* Split the existing extent */
367 if (fi->ext.len > 1 &&
368 fofs >= start_fofs && fofs <= end_fofs) {
369 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
370 fi->ext.len = fofs - start_fofs;
372 fi->ext.fofs = fofs + 1;
373 fi->ext.blk_addr = start_blkaddr +
374 fofs - start_fofs + 1;
375 fi->ext.len -= fofs - start_fofs + 1;
381 /* Finally, if the extent is very fragmented, let's drop the cache. */
382 if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
384 set_inode_flag(fi, FI_NO_EXTENT);
388 write_unlock(&fi->ext.ext_lock);
394 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
396 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
397 struct address_space *mapping = inode->i_mapping;
398 struct dnode_of_data dn;
402 page = find_get_page(mapping, index);
403 if (page && PageUptodate(page))
405 f2fs_put_page(page, 0);
407 set_new_dnode(&dn, inode, NULL, NULL, 0);
408 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
413 if (dn.data_blkaddr == NULL_ADDR)
414 return ERR_PTR(-ENOENT);
416 /* By fallocate(), there is no cached page, but with NEW_ADDR */
417 if (unlikely(dn.data_blkaddr == NEW_ADDR))
418 return ERR_PTR(-EINVAL);
420 page = grab_cache_page(mapping, index);
422 return ERR_PTR(-ENOMEM);
424 if (PageUptodate(page)) {
429 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
430 sync ? READ_SYNC : READA);
435 wait_on_page_locked(page);
436 if (unlikely(!PageUptodate(page))) {
437 f2fs_put_page(page, 0);
438 return ERR_PTR(-EIO);
445 * If it tries to access a hole, return an error.
446 * Because, the callers, functions in dir.c and GC, should be able to know
447 * whether this page exists or not.
449 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
451 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
452 struct address_space *mapping = inode->i_mapping;
453 struct dnode_of_data dn;
458 page = grab_cache_page(mapping, index);
460 return ERR_PTR(-ENOMEM);
462 set_new_dnode(&dn, inode, NULL, NULL, 0);
463 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
465 f2fs_put_page(page, 1);
470 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
471 f2fs_put_page(page, 1);
472 return ERR_PTR(-ENOENT);
475 if (PageUptodate(page))
479 * A new dentry page is allocated but not able to be written, since its
480 * new inode page couldn't be allocated due to -ENOSPC.
481 * In such the case, its blkaddr can be remained as NEW_ADDR.
482 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
484 if (dn.data_blkaddr == NEW_ADDR) {
485 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
486 SetPageUptodate(page);
490 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, READ_SYNC);
495 if (unlikely(!PageUptodate(page))) {
496 f2fs_put_page(page, 1);
497 return ERR_PTR(-EIO);
499 if (unlikely(page->mapping != mapping)) {
500 f2fs_put_page(page, 1);
507 * Caller ensures that this data page is never allocated.
508 * A new zero-filled data page is allocated in the page cache.
510 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
512 * Note that, ipage is set only by make_empty_dir.
514 struct page *get_new_data_page(struct inode *inode,
515 struct page *ipage, pgoff_t index, bool new_i_size)
517 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
518 struct address_space *mapping = inode->i_mapping;
520 struct dnode_of_data dn;
523 set_new_dnode(&dn, inode, ipage, NULL, 0);
524 err = f2fs_reserve_block(&dn, index);
528 page = grab_cache_page(mapping, index);
534 if (PageUptodate(page))
537 if (dn.data_blkaddr == NEW_ADDR) {
538 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
539 SetPageUptodate(page);
541 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
547 if (unlikely(!PageUptodate(page))) {
548 f2fs_put_page(page, 1);
552 if (unlikely(page->mapping != mapping)) {
553 f2fs_put_page(page, 1);
559 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
560 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
561 /* Only the directory inode sets new_i_size */
562 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
571 static int __allocate_data_block(struct dnode_of_data *dn)
573 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
574 struct f2fs_summary sum;
579 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
581 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
584 __set_data_blkaddr(dn, NEW_ADDR);
585 dn->data_blkaddr = NEW_ADDR;
587 get_node_info(sbi, dn->nid, &ni);
588 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
590 type = CURSEG_WARM_DATA;
592 allocate_data_block(sbi, NULL, NULL_ADDR, &new_blkaddr, &sum, type);
594 /* direct IO doesn't use extent cache to maximize the performance */
595 set_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
596 update_extent_cache(new_blkaddr, dn);
597 clear_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
599 dn->data_blkaddr = new_blkaddr;
604 * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
605 * If original data blocks are allocated, then give them to blockdev.
607 * a. preallocate requested block addresses
608 * b. do not use extent cache for better performance
609 * c. give the block addresses to blockdev
611 static int get_data_block(struct inode *inode, sector_t iblock,
612 struct buffer_head *bh_result, int create)
614 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
615 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
616 unsigned maxblocks = bh_result->b_size >> blkbits;
617 struct dnode_of_data dn;
618 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
619 pgoff_t pgofs, end_offset;
620 int err = 0, ofs = 1;
621 bool allocated = false;
623 /* Get the page offset from the block offset(iblock) */
624 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
626 if (check_extent_cache(inode, pgofs, bh_result))
632 /* When reading holes, we need its node page */
633 set_new_dnode(&dn, inode, NULL, NULL, 0);
634 err = get_dnode_of_data(&dn, pgofs, mode);
640 if (dn.data_blkaddr == NEW_ADDR)
643 if (dn.data_blkaddr != NULL_ADDR) {
644 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
646 err = __allocate_data_block(&dn);
650 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
655 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
656 bh_result->b_size = (((size_t)1) << blkbits);
661 if (dn.ofs_in_node >= end_offset) {
663 sync_inode_page(&dn);
667 set_new_dnode(&dn, inode, NULL, NULL, 0);
668 err = get_dnode_of_data(&dn, pgofs, mode);
674 if (dn.data_blkaddr == NEW_ADDR)
677 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
680 if (maxblocks > (bh_result->b_size >> blkbits)) {
681 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
682 if (blkaddr == NULL_ADDR && create) {
683 err = __allocate_data_block(&dn);
687 blkaddr = dn.data_blkaddr;
689 /* Give more consecutive addresses for the read ahead */
690 if (blkaddr == (bh_result->b_blocknr + ofs)) {
694 bh_result->b_size += (((size_t)1) << blkbits);
700 sync_inode_page(&dn);
707 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
711 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
714 return generic_block_fiemap(inode, fieinfo, start, len, get_data_block);
717 static int f2fs_read_data_page(struct file *file, struct page *page)
719 struct inode *inode = page->mapping->host;
722 trace_f2fs_readpage(page, DATA);
724 /* If the file has inline data, try to read it directlly */
725 if (f2fs_has_inline_data(inode))
726 ret = f2fs_read_inline_data(inode, page);
728 ret = mpage_readpage(page, get_data_block);
733 static int f2fs_read_data_pages(struct file *file,
734 struct address_space *mapping,
735 struct list_head *pages, unsigned nr_pages)
737 struct inode *inode = file->f_mapping->host;
739 /* If the file has inline data, skip readpages */
740 if (f2fs_has_inline_data(inode))
743 return mpage_readpages(mapping, pages, nr_pages, get_data_block);
746 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
748 struct inode *inode = page->mapping->host;
749 block_t old_blkaddr, new_blkaddr;
750 struct dnode_of_data dn;
753 set_new_dnode(&dn, inode, NULL, NULL, 0);
754 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
758 old_blkaddr = dn.data_blkaddr;
760 /* This page is already truncated */
761 if (old_blkaddr == NULL_ADDR)
764 set_page_writeback(page);
767 * If current allocation needs SSR,
768 * it had better in-place writes for updated data.
770 if (unlikely(old_blkaddr != NEW_ADDR &&
771 !is_cold_data(page) &&
772 need_inplace_update(inode))) {
773 rewrite_data_page(page, old_blkaddr, fio);
775 write_data_page(page, &dn, &new_blkaddr, fio);
776 update_extent_cache(new_blkaddr, &dn);
783 static int f2fs_write_data_page(struct page *page,
784 struct writeback_control *wbc)
786 struct inode *inode = page->mapping->host;
787 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
788 loff_t i_size = i_size_read(inode);
789 const pgoff_t end_index = ((unsigned long long) i_size)
792 bool need_balance_fs = false;
794 struct f2fs_io_info fio = {
796 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
799 trace_f2fs_writepage(page, DATA);
801 if (page->index < end_index)
805 * If the offset is out-of-range of file size,
806 * this page does not have to be written to disk.
808 offset = i_size & (PAGE_CACHE_SIZE - 1);
809 if ((page->index >= end_index + 1) || !offset)
812 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
814 if (unlikely(sbi->por_doing))
817 /* Dentry blocks are controlled by checkpoint */
818 if (S_ISDIR(inode->i_mode)) {
819 err = do_write_data_page(page, &fio);
823 if (!wbc->for_reclaim)
824 need_balance_fs = true;
825 else if (has_not_enough_free_secs(sbi, 0))
829 if (f2fs_has_inline_data(inode) || f2fs_may_inline(inode))
830 err = f2fs_write_inline_data(inode, page, offset);
832 err = do_write_data_page(page, &fio);
835 if (err && err != -ENOENT)
838 clear_cold_data(page);
840 inode_dec_dirty_dents(inode);
843 f2fs_balance_fs(sbi);
844 if (wbc->for_reclaim)
845 f2fs_submit_merged_bio(sbi, DATA, WRITE);
849 redirty_page_for_writepage(wbc, page);
850 return AOP_WRITEPAGE_ACTIVATE;
853 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
856 struct address_space *mapping = data;
857 int ret = mapping->a_ops->writepage(page, wbc);
858 mapping_set_error(mapping, ret);
862 static int f2fs_write_data_pages(struct address_space *mapping,
863 struct writeback_control *wbc)
865 struct inode *inode = mapping->host;
866 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
871 trace_f2fs_writepages(mapping->host, wbc, DATA);
873 /* deal with chardevs and other special file */
874 if (!mapping->a_ops->writepage)
877 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
878 get_dirty_dents(inode) < nr_pages_to_skip(sbi, DATA) &&
879 available_free_memory(sbi, DIRTY_DENTS))
882 diff = nr_pages_to_write(sbi, DATA, wbc);
884 if (!S_ISDIR(inode->i_mode)) {
885 mutex_lock(&sbi->writepages);
888 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
890 mutex_unlock(&sbi->writepages);
892 f2fs_submit_merged_bio(sbi, DATA, WRITE);
894 remove_dirty_dir_inode(inode);
896 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
900 wbc->pages_skipped += get_dirty_dents(inode);
904 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
905 loff_t pos, unsigned len, unsigned flags,
906 struct page **pagep, void **fsdata)
908 struct inode *inode = mapping->host;
909 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
911 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
912 struct dnode_of_data dn;
915 trace_f2fs_write_begin(inode, pos, len, flags);
917 f2fs_balance_fs(sbi);
919 err = f2fs_convert_inline_data(inode, pos + len);
923 page = grab_cache_page_write_begin(mapping, index, flags);
927 /* to avoid latency during memory pressure */
932 if (f2fs_has_inline_data(inode) && (pos + len) <= MAX_INLINE_DATA)
936 set_new_dnode(&dn, inode, NULL, NULL, 0);
937 err = f2fs_reserve_block(&dn, index);
941 f2fs_put_page(page, 0);
946 if (unlikely(page->mapping != mapping)) {
947 f2fs_put_page(page, 1);
951 f2fs_wait_on_page_writeback(page, DATA);
953 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
956 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
957 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
958 unsigned end = start + len;
960 /* Reading beyond i_size is simple: memset to zero */
961 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
965 if (dn.data_blkaddr == NEW_ADDR) {
966 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
968 if (f2fs_has_inline_data(inode)) {
969 err = f2fs_read_inline_data(inode, page);
971 page_cache_release(page);
975 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
982 if (unlikely(!PageUptodate(page))) {
983 f2fs_put_page(page, 1);
986 if (unlikely(page->mapping != mapping)) {
987 f2fs_put_page(page, 1);
992 SetPageUptodate(page);
993 clear_cold_data(page);
997 static int f2fs_write_end(struct file *file,
998 struct address_space *mapping,
999 loff_t pos, unsigned len, unsigned copied,
1000 struct page *page, void *fsdata)
1002 struct inode *inode = page->mapping->host;
1004 trace_f2fs_write_end(inode, pos, len, copied);
1006 SetPageUptodate(page);
1007 set_page_dirty(page);
1009 if (pos + copied > i_size_read(inode)) {
1010 i_size_write(inode, pos + copied);
1011 mark_inode_dirty(inode);
1012 update_inode_page(inode);
1015 f2fs_put_page(page, 1);
1019 static int check_direct_IO(struct inode *inode, int rw,
1020 struct iov_iter *iter, loff_t offset)
1022 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1027 if (offset & blocksize_mask)
1030 if (iov_iter_alignment(iter) & blocksize_mask)
1036 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1037 struct iov_iter *iter, loff_t offset)
1039 struct file *file = iocb->ki_filp;
1040 struct inode *inode = file->f_mapping->host;
1042 /* Let buffer I/O handle the inline data case. */
1043 if (f2fs_has_inline_data(inode))
1046 if (check_direct_IO(inode, rw, iter, offset))
1049 /* clear fsync mark to recover these blocks */
1050 fsync_mark_clear(F2FS_SB(inode->i_sb), inode->i_ino);
1052 return blockdev_direct_IO(rw, iocb, inode, iter, offset,
1056 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
1057 unsigned int length)
1059 struct inode *inode = page->mapping->host;
1060 if (PageDirty(page))
1061 inode_dec_dirty_dents(inode);
1062 ClearPagePrivate(page);
1065 static int f2fs_release_data_page(struct page *page, gfp_t wait)
1067 ClearPagePrivate(page);
1071 static int f2fs_set_data_page_dirty(struct page *page)
1073 struct address_space *mapping = page->mapping;
1074 struct inode *inode = mapping->host;
1076 trace_f2fs_set_page_dirty(page, DATA);
1078 SetPageUptodate(page);
1079 mark_inode_dirty(inode);
1081 if (!PageDirty(page)) {
1082 __set_page_dirty_nobuffers(page);
1083 set_dirty_dir_page(inode, page);
1089 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1091 struct inode *inode = mapping->host;
1093 if (f2fs_has_inline_data(inode))
1096 return generic_block_bmap(mapping, block, get_data_block);
1099 const struct address_space_operations f2fs_dblock_aops = {
1100 .readpage = f2fs_read_data_page,
1101 .readpages = f2fs_read_data_pages,
1102 .writepage = f2fs_write_data_page,
1103 .writepages = f2fs_write_data_pages,
1104 .write_begin = f2fs_write_begin,
1105 .write_end = f2fs_write_end,
1106 .set_page_dirty = f2fs_set_data_page_dirty,
1107 .invalidatepage = f2fs_invalidate_data_page,
1108 .releasepage = f2fs_release_data_page,
1109 .direct_IO = f2fs_direct_IO,