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(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 if (test_opt(sbi, NOBARRIER))
143 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
145 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
147 __submit_merged_bio(io);
148 up_write(&io->io_rwsem);
152 * Fill the locked page with data located in the block address.
153 * Return unlocked page.
155 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
156 block_t blk_addr, int rw)
160 trace_f2fs_submit_page_bio(page, blk_addr, rw);
162 /* Allocate a new bio */
163 bio = __bio_alloc(sbi, blk_addr, 1, is_read_io(rw));
165 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
167 f2fs_put_page(page, 1);
175 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
176 block_t blk_addr, struct f2fs_io_info *fio)
178 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
179 struct f2fs_bio_info *io;
180 bool is_read = is_read_io(fio->rw);
182 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
184 verify_block_addr(sbi, blk_addr);
186 down_write(&io->io_rwsem);
189 inc_page_count(sbi, F2FS_WRITEBACK);
191 if (io->bio && (io->last_block_in_bio != blk_addr - 1 ||
192 io->fio.rw != fio->rw))
193 __submit_merged_bio(io);
195 if (io->bio == NULL) {
196 int bio_blocks = MAX_BIO_BLOCKS(sbi);
198 io->bio = __bio_alloc(sbi, blk_addr, bio_blocks, is_read);
202 if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
204 __submit_merged_bio(io);
208 io->last_block_in_bio = blk_addr;
210 up_write(&io->io_rwsem);
211 trace_f2fs_submit_page_mbio(page, fio->rw, fio->type, blk_addr);
215 * Lock ordering for the change of data block address:
218 * update block addresses in the node page
220 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
222 struct f2fs_node *rn;
224 struct page *node_page = dn->node_page;
225 unsigned int ofs_in_node = dn->ofs_in_node;
227 f2fs_wait_on_page_writeback(node_page, NODE);
229 rn = F2FS_NODE(node_page);
231 /* Get physical address of data block */
232 addr_array = blkaddr_in_node(rn);
233 addr_array[ofs_in_node] = cpu_to_le32(new_addr);
234 set_page_dirty(node_page);
237 int reserve_new_block(struct dnode_of_data *dn)
239 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
241 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
243 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
246 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
248 __set_data_blkaddr(dn, NEW_ADDR);
249 dn->data_blkaddr = NEW_ADDR;
250 mark_inode_dirty(dn->inode);
255 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
257 bool need_put = dn->inode_page ? false : true;
260 /* if inode_page exists, index should be zero */
261 f2fs_bug_on(F2FS_I_SB(dn->inode), !need_put && index);
263 err = get_dnode_of_data(dn, index, ALLOC_NODE);
267 if (dn->data_blkaddr == NULL_ADDR)
268 err = reserve_new_block(dn);
274 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
275 struct buffer_head *bh_result)
277 struct f2fs_inode_info *fi = F2FS_I(inode);
278 pgoff_t start_fofs, end_fofs;
279 block_t start_blkaddr;
281 if (is_inode_flag_set(fi, FI_NO_EXTENT))
284 read_lock(&fi->ext.ext_lock);
285 if (fi->ext.len == 0) {
286 read_unlock(&fi->ext.ext_lock);
290 stat_inc_total_hit(inode->i_sb);
292 start_fofs = fi->ext.fofs;
293 end_fofs = fi->ext.fofs + fi->ext.len - 1;
294 start_blkaddr = fi->ext.blk_addr;
296 if (pgofs >= start_fofs && pgofs <= end_fofs) {
297 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
300 clear_buffer_new(bh_result);
301 map_bh(bh_result, inode->i_sb,
302 start_blkaddr + pgofs - start_fofs);
303 count = end_fofs - pgofs + 1;
304 if (count < (UINT_MAX >> blkbits))
305 bh_result->b_size = (count << blkbits);
307 bh_result->b_size = UINT_MAX;
309 stat_inc_read_hit(inode->i_sb);
310 read_unlock(&fi->ext.ext_lock);
313 read_unlock(&fi->ext.ext_lock);
317 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
319 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
320 pgoff_t fofs, start_fofs, end_fofs;
321 block_t start_blkaddr, end_blkaddr;
322 int need_update = true;
324 f2fs_bug_on(F2FS_I_SB(dn->inode), blk_addr == NEW_ADDR);
325 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
328 /* Update the page address in the parent node */
329 __set_data_blkaddr(dn, blk_addr);
331 if (is_inode_flag_set(fi, FI_NO_EXTENT))
334 write_lock(&fi->ext.ext_lock);
336 start_fofs = fi->ext.fofs;
337 end_fofs = fi->ext.fofs + fi->ext.len - 1;
338 start_blkaddr = fi->ext.blk_addr;
339 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
341 /* Drop and initialize the matched extent */
342 if (fi->ext.len == 1 && fofs == start_fofs)
346 if (fi->ext.len == 0) {
347 if (blk_addr != NULL_ADDR) {
349 fi->ext.blk_addr = blk_addr;
356 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
364 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
369 /* Split the existing extent */
370 if (fi->ext.len > 1 &&
371 fofs >= start_fofs && fofs <= end_fofs) {
372 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
373 fi->ext.len = fofs - start_fofs;
375 fi->ext.fofs = fofs + 1;
376 fi->ext.blk_addr = start_blkaddr +
377 fofs - start_fofs + 1;
378 fi->ext.len -= fofs - start_fofs + 1;
384 /* Finally, if the extent is very fragmented, let's drop the cache. */
385 if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
387 set_inode_flag(fi, FI_NO_EXTENT);
391 write_unlock(&fi->ext.ext_lock);
397 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
399 struct address_space *mapping = inode->i_mapping;
400 struct dnode_of_data dn;
404 page = find_get_page(mapping, index);
405 if (page && PageUptodate(page))
407 f2fs_put_page(page, 0);
409 set_new_dnode(&dn, inode, NULL, NULL, 0);
410 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
415 if (dn.data_blkaddr == NULL_ADDR)
416 return ERR_PTR(-ENOENT);
418 /* By fallocate(), there is no cached page, but with NEW_ADDR */
419 if (unlikely(dn.data_blkaddr == NEW_ADDR))
420 return ERR_PTR(-EINVAL);
422 page = grab_cache_page(mapping, index);
424 return ERR_PTR(-ENOMEM);
426 if (PageUptodate(page)) {
431 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, dn.data_blkaddr,
432 sync ? READ_SYNC : READA);
437 wait_on_page_locked(page);
438 if (unlikely(!PageUptodate(page))) {
439 f2fs_put_page(page, 0);
440 return ERR_PTR(-EIO);
447 * If it tries to access a hole, return an error.
448 * Because, the callers, functions in dir.c and GC, should be able to know
449 * whether this page exists or not.
451 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
453 struct address_space *mapping = inode->i_mapping;
454 struct dnode_of_data dn;
459 page = grab_cache_page(mapping, index);
461 return ERR_PTR(-ENOMEM);
463 set_new_dnode(&dn, inode, NULL, NULL, 0);
464 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
466 f2fs_put_page(page, 1);
471 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
472 f2fs_put_page(page, 1);
473 return ERR_PTR(-ENOENT);
476 if (PageUptodate(page))
480 * A new dentry page is allocated but not able to be written, since its
481 * new inode page couldn't be allocated due to -ENOSPC.
482 * In such the case, its blkaddr can be remained as NEW_ADDR.
483 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
485 if (dn.data_blkaddr == NEW_ADDR) {
486 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
487 SetPageUptodate(page);
491 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page,
492 dn.data_blkaddr, READ_SYNC);
497 if (unlikely(!PageUptodate(page))) {
498 f2fs_put_page(page, 1);
499 return ERR_PTR(-EIO);
501 if (unlikely(page->mapping != mapping)) {
502 f2fs_put_page(page, 1);
509 * Caller ensures that this data page is never allocated.
510 * A new zero-filled data page is allocated in the page cache.
512 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
514 * Note that, ipage is set only by make_empty_dir.
516 struct page *get_new_data_page(struct inode *inode,
517 struct page *ipage, pgoff_t index, bool new_i_size)
519 struct address_space *mapping = inode->i_mapping;
521 struct dnode_of_data dn;
524 set_new_dnode(&dn, inode, ipage, NULL, 0);
525 err = f2fs_reserve_block(&dn, index);
529 page = grab_cache_page(mapping, index);
535 if (PageUptodate(page))
538 if (dn.data_blkaddr == NEW_ADDR) {
539 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
540 SetPageUptodate(page);
542 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page,
543 dn.data_blkaddr, READ_SYNC);
548 if (unlikely(!PageUptodate(page))) {
549 f2fs_put_page(page, 1);
553 if (unlikely(page->mapping != mapping)) {
554 f2fs_put_page(page, 1);
560 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
561 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
562 /* Only the directory inode sets new_i_size */
563 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
572 static int __allocate_data_block(struct dnode_of_data *dn)
574 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
575 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
576 struct f2fs_summary sum;
582 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
584 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
587 __set_data_blkaddr(dn, NEW_ADDR);
588 dn->data_blkaddr = NEW_ADDR;
590 get_node_info(sbi, dn->nid, &ni);
591 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
593 type = CURSEG_WARM_DATA;
595 allocate_data_block(sbi, NULL, NULL_ADDR, &new_blkaddr, &sum, type);
597 /* direct IO doesn't use extent cache to maximize the performance */
598 set_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
599 update_extent_cache(new_blkaddr, dn);
600 clear_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
603 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
605 if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
606 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
608 dn->data_blkaddr = new_blkaddr;
613 * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
614 * If original data blocks are allocated, then give them to blockdev.
616 * a. preallocate requested block addresses
617 * b. do not use extent cache for better performance
618 * c. give the block addresses to blockdev
620 static int __get_data_block(struct inode *inode, sector_t iblock,
621 struct buffer_head *bh_result, int create, bool fiemap)
623 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
624 unsigned maxblocks = bh_result->b_size >> blkbits;
625 struct dnode_of_data dn;
626 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
627 pgoff_t pgofs, end_offset;
628 int err = 0, ofs = 1;
629 bool allocated = false;
631 /* Get the page offset from the block offset(iblock) */
632 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
634 if (check_extent_cache(inode, pgofs, bh_result))
638 f2fs_balance_fs(F2FS_I_SB(inode));
639 f2fs_lock_op(F2FS_I_SB(inode));
642 /* When reading holes, we need its node page */
643 set_new_dnode(&dn, inode, NULL, NULL, 0);
644 err = get_dnode_of_data(&dn, pgofs, mode);
650 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
653 if (dn.data_blkaddr != NULL_ADDR) {
654 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
656 err = __allocate_data_block(&dn);
660 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
665 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
666 bh_result->b_size = (((size_t)1) << blkbits);
671 if (dn.ofs_in_node >= end_offset) {
673 sync_inode_page(&dn);
677 set_new_dnode(&dn, inode, NULL, NULL, 0);
678 err = get_dnode_of_data(&dn, pgofs, mode);
684 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
687 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
690 if (maxblocks > (bh_result->b_size >> blkbits)) {
691 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
692 if (blkaddr == NULL_ADDR && create) {
693 err = __allocate_data_block(&dn);
697 blkaddr = dn.data_blkaddr;
699 /* Give more consecutive addresses for the readahead */
700 if (blkaddr == (bh_result->b_blocknr + ofs)) {
704 bh_result->b_size += (((size_t)1) << blkbits);
710 sync_inode_page(&dn);
715 f2fs_unlock_op(F2FS_I_SB(inode));
717 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
721 static int get_data_block(struct inode *inode, sector_t iblock,
722 struct buffer_head *bh_result, int create)
724 return __get_data_block(inode, iblock, bh_result, create, false);
727 static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
728 struct buffer_head *bh_result, int create)
730 return __get_data_block(inode, iblock, bh_result, create, true);
733 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
736 return generic_block_fiemap(inode, fieinfo,
737 start, len, get_data_block_fiemap);
740 static int f2fs_read_data_page(struct file *file, struct page *page)
742 struct inode *inode = page->mapping->host;
745 trace_f2fs_readpage(page, DATA);
747 /* If the file has inline data, try to read it directly */
748 if (f2fs_has_inline_data(inode))
749 ret = f2fs_read_inline_data(inode, page);
751 ret = mpage_readpage(page, get_data_block);
756 static int f2fs_read_data_pages(struct file *file,
757 struct address_space *mapping,
758 struct list_head *pages, unsigned nr_pages)
760 struct inode *inode = file->f_mapping->host;
762 /* If the file has inline data, skip readpages */
763 if (f2fs_has_inline_data(inode))
766 return mpage_readpages(mapping, pages, nr_pages, get_data_block);
769 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
771 struct inode *inode = page->mapping->host;
772 block_t old_blkaddr, new_blkaddr;
773 struct dnode_of_data dn;
776 set_new_dnode(&dn, inode, NULL, NULL, 0);
777 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
781 old_blkaddr = dn.data_blkaddr;
783 /* This page is already truncated */
784 if (old_blkaddr == NULL_ADDR)
787 set_page_writeback(page);
790 * If current allocation needs SSR,
791 * it had better in-place writes for updated data.
793 if (unlikely(old_blkaddr != NEW_ADDR &&
794 !is_cold_data(page) &&
795 need_inplace_update(inode))) {
796 rewrite_data_page(page, old_blkaddr, fio);
797 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
799 write_data_page(page, &dn, &new_blkaddr, fio);
800 update_extent_cache(new_blkaddr, &dn);
801 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
808 static int f2fs_write_data_page(struct page *page,
809 struct writeback_control *wbc)
811 struct inode *inode = page->mapping->host;
812 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
813 loff_t i_size = i_size_read(inode);
814 const pgoff_t end_index = ((unsigned long long) i_size)
817 bool need_balance_fs = false;
819 struct f2fs_io_info fio = {
821 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
824 trace_f2fs_writepage(page, DATA);
826 if (page->index < end_index)
830 * If the offset is out-of-range of file size,
831 * this page does not have to be written to disk.
833 offset = i_size & (PAGE_CACHE_SIZE - 1);
834 if ((page->index >= end_index + 1) || !offset)
837 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
839 if (unlikely(sbi->por_doing))
842 /* Dentry blocks are controlled by checkpoint */
843 if (S_ISDIR(inode->i_mode)) {
844 if (unlikely(f2fs_cp_error(sbi)))
846 err = do_write_data_page(page, &fio);
850 /* we should bypass data pages to proceed the kworkder jobs */
851 if (unlikely(f2fs_cp_error(sbi))) {
857 if (!wbc->for_reclaim)
858 need_balance_fs = true;
859 else if (has_not_enough_free_secs(sbi, 0))
863 if (f2fs_has_inline_data(inode) || f2fs_may_inline(inode))
864 err = f2fs_write_inline_data(inode, page, offset);
866 err = do_write_data_page(page, &fio);
869 if (err && err != -ENOENT)
872 clear_cold_data(page);
874 inode_dec_dirty_pages(inode);
877 f2fs_balance_fs(sbi);
878 if (wbc->for_reclaim)
879 f2fs_submit_merged_bio(sbi, DATA, WRITE);
883 redirty_page_for_writepage(wbc, page);
884 return AOP_WRITEPAGE_ACTIVATE;
887 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
890 struct address_space *mapping = data;
891 int ret = mapping->a_ops->writepage(page, wbc);
892 mapping_set_error(mapping, ret);
896 static int f2fs_write_data_pages(struct address_space *mapping,
897 struct writeback_control *wbc)
899 struct inode *inode = mapping->host;
900 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
905 trace_f2fs_writepages(mapping->host, wbc, DATA);
907 /* deal with chardevs and other special file */
908 if (!mapping->a_ops->writepage)
911 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
912 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
913 available_free_memory(sbi, DIRTY_DENTS))
916 diff = nr_pages_to_write(sbi, DATA, wbc);
918 if (!S_ISDIR(inode->i_mode)) {
919 mutex_lock(&sbi->writepages);
922 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
924 mutex_unlock(&sbi->writepages);
926 f2fs_submit_merged_bio(sbi, DATA, WRITE);
928 remove_dirty_dir_inode(inode);
930 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
934 wbc->pages_skipped += get_dirty_pages(inode);
938 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
940 struct inode *inode = mapping->host;
942 if (to > inode->i_size) {
943 truncate_pagecache(inode, inode->i_size);
944 truncate_blocks(inode, inode->i_size, true);
948 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
949 loff_t pos, unsigned len, unsigned flags,
950 struct page **pagep, void **fsdata)
952 struct inode *inode = mapping->host;
953 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
955 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
956 struct dnode_of_data dn;
959 trace_f2fs_write_begin(inode, pos, len, flags);
961 f2fs_balance_fs(sbi);
963 err = f2fs_convert_inline_data(inode, pos + len, NULL);
967 page = grab_cache_page_write_begin(mapping, index, flags);
973 /* to avoid latency during memory pressure */
978 if (f2fs_has_inline_data(inode) && (pos + len) <= MAX_INLINE_DATA)
982 set_new_dnode(&dn, inode, NULL, NULL, 0);
983 err = f2fs_reserve_block(&dn, index);
986 f2fs_put_page(page, 0);
991 if (unlikely(page->mapping != mapping)) {
992 f2fs_put_page(page, 1);
996 f2fs_wait_on_page_writeback(page, DATA);
998 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
1001 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1002 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1003 unsigned end = start + len;
1005 /* Reading beyond i_size is simple: memset to zero */
1006 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1010 if (dn.data_blkaddr == NEW_ADDR) {
1011 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1013 if (f2fs_has_inline_data(inode)) {
1014 err = f2fs_read_inline_data(inode, page);
1016 page_cache_release(page);
1020 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
1027 if (unlikely(!PageUptodate(page))) {
1028 f2fs_put_page(page, 1);
1032 if (unlikely(page->mapping != mapping)) {
1033 f2fs_put_page(page, 1);
1038 SetPageUptodate(page);
1039 clear_cold_data(page);
1042 f2fs_write_failed(mapping, pos + len);
1046 static int f2fs_write_end(struct file *file,
1047 struct address_space *mapping,
1048 loff_t pos, unsigned len, unsigned copied,
1049 struct page *page, void *fsdata)
1051 struct inode *inode = page->mapping->host;
1053 trace_f2fs_write_end(inode, pos, len, copied);
1055 if (f2fs_is_atomic_file(inode) || f2fs_is_volatile_file(inode))
1056 register_inmem_page(inode, page);
1058 set_page_dirty(page);
1060 if (pos + copied > i_size_read(inode)) {
1061 i_size_write(inode, pos + copied);
1062 mark_inode_dirty(inode);
1063 update_inode_page(inode);
1066 f2fs_put_page(page, 1);
1070 static int check_direct_IO(struct inode *inode, int rw,
1071 struct iov_iter *iter, loff_t offset)
1073 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1078 if (offset & blocksize_mask)
1081 if (iov_iter_alignment(iter) & blocksize_mask)
1087 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1088 struct iov_iter *iter, loff_t offset)
1090 struct file *file = iocb->ki_filp;
1091 struct address_space *mapping = file->f_mapping;
1092 struct inode *inode = mapping->host;
1093 size_t count = iov_iter_count(iter);
1096 /* Let buffer I/O handle the inline data case. */
1097 if (f2fs_has_inline_data(inode))
1100 if (check_direct_IO(inode, rw, iter, offset))
1103 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1105 err = blockdev_direct_IO(rw, iocb, inode, iter, offset, get_data_block);
1106 if (err < 0 && (rw & WRITE))
1107 f2fs_write_failed(mapping, offset + count);
1109 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1114 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
1115 unsigned int length)
1117 struct inode *inode = page->mapping->host;
1119 if (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE)
1122 if (PageDirty(page))
1123 inode_dec_dirty_pages(inode);
1124 ClearPagePrivate(page);
1127 static int f2fs_release_data_page(struct page *page, gfp_t wait)
1129 ClearPagePrivate(page);
1133 static int f2fs_set_data_page_dirty(struct page *page)
1135 struct address_space *mapping = page->mapping;
1136 struct inode *inode = mapping->host;
1138 trace_f2fs_set_page_dirty(page, DATA);
1140 SetPageUptodate(page);
1141 mark_inode_dirty(inode);
1143 if (!PageDirty(page)) {
1144 __set_page_dirty_nobuffers(page);
1145 update_dirty_page(inode, page);
1151 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1153 struct inode *inode = mapping->host;
1155 if (f2fs_has_inline_data(inode))
1158 return generic_block_bmap(mapping, block, get_data_block);
1161 const struct address_space_operations f2fs_dblock_aops = {
1162 .readpage = f2fs_read_data_page,
1163 .readpages = f2fs_read_data_pages,
1164 .writepage = f2fs_write_data_page,
1165 .writepages = f2fs_write_data_pages,
1166 .write_begin = f2fs_write_begin,
1167 .write_end = f2fs_write_end,
1168 .set_page_dirty = f2fs_set_data_page_dirty,
1169 .invalidatepage = f2fs_invalidate_data_page,
1170 .releasepage = f2fs_release_data_page,
1171 .direct_IO = f2fs_direct_IO,