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>
23 #include <linux/memcontrol.h>
24 #include <linux/cleancache.h>
25 #include <linux/sched/signal.h>
31 #include <trace/events/f2fs.h>
33 static bool __is_cp_guaranteed(struct page *page)
35 struct address_space *mapping = page->mapping;
37 struct f2fs_sb_info *sbi;
42 inode = mapping->host;
43 sbi = F2FS_I_SB(inode);
45 if (inode->i_ino == F2FS_META_INO(sbi) ||
46 inode->i_ino == F2FS_NODE_INO(sbi) ||
47 S_ISDIR(inode->i_mode) ||
53 static void f2fs_read_end_io(struct bio *bio)
58 #ifdef CONFIG_F2FS_FAULT_INJECTION
59 if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO)) {
60 f2fs_show_injection_info(FAULT_IO);
61 bio->bi_status = BLK_STS_IOERR;
65 if (f2fs_bio_encrypted(bio)) {
67 fscrypt_release_ctx(bio->bi_private);
69 fscrypt_decrypt_bio_pages(bio->bi_private, bio);
74 bio_for_each_segment_all(bvec, bio, i) {
75 struct page *page = bvec->bv_page;
77 if (!bio->bi_status) {
78 if (!PageUptodate(page))
79 SetPageUptodate(page);
81 ClearPageUptodate(page);
89 static void f2fs_write_end_io(struct bio *bio)
91 struct f2fs_sb_info *sbi = bio->bi_private;
95 bio_for_each_segment_all(bvec, bio, i) {
96 struct page *page = bvec->bv_page;
97 enum count_type type = WB_DATA_TYPE(page);
99 if (IS_DUMMY_WRITTEN_PAGE(page)) {
100 set_page_private(page, (unsigned long)NULL);
101 ClearPagePrivate(page);
103 mempool_free(page, sbi->write_io_dummy);
105 if (unlikely(bio->bi_status))
106 f2fs_stop_checkpoint(sbi, true);
110 fscrypt_pullback_bio_page(&page, true);
112 if (unlikely(bio->bi_status)) {
113 mapping_set_error(page->mapping, -EIO);
114 f2fs_stop_checkpoint(sbi, true);
116 dec_page_count(sbi, type);
117 clear_cold_data(page);
118 end_page_writeback(page);
120 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
121 wq_has_sleeper(&sbi->cp_wait))
122 wake_up(&sbi->cp_wait);
128 * Return true, if pre_bio's bdev is same as its target device.
130 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
131 block_t blk_addr, struct bio *bio)
133 struct block_device *bdev = sbi->sb->s_bdev;
136 for (i = 0; i < sbi->s_ndevs; i++) {
137 if (FDEV(i).start_blk <= blk_addr &&
138 FDEV(i).end_blk >= blk_addr) {
139 blk_addr -= FDEV(i).start_blk;
145 bio_set_dev(bio, bdev);
146 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
151 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
155 for (i = 0; i < sbi->s_ndevs; i++)
156 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
161 static bool __same_bdev(struct f2fs_sb_info *sbi,
162 block_t blk_addr, struct bio *bio)
164 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
165 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
169 * Low-level block read/write IO operations.
171 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
172 int npages, bool is_read)
176 bio = f2fs_bio_alloc(npages);
178 f2fs_target_device(sbi, blk_addr, bio);
179 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
180 bio->bi_private = is_read ? NULL : sbi;
185 static inline void __submit_bio(struct f2fs_sb_info *sbi,
186 struct bio *bio, enum page_type type)
188 if (!is_read_io(bio_op(bio))) {
191 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
192 current->plug && (type == DATA || type == NODE))
193 blk_finish_plug(current->plug);
195 if (type != DATA && type != NODE)
198 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
199 start %= F2FS_IO_SIZE(sbi);
204 /* fill dummy pages */
205 for (; start < F2FS_IO_SIZE(sbi); start++) {
207 mempool_alloc(sbi->write_io_dummy,
208 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
209 f2fs_bug_on(sbi, !page);
211 SetPagePrivate(page);
212 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
214 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
218 * In the NODE case, we lose next block address chain. So, we
219 * need to do checkpoint in f2fs_sync_file.
222 set_sbi_flag(sbi, SBI_NEED_CP);
225 if (is_read_io(bio_op(bio)))
226 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
228 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
232 static void __submit_merged_bio(struct f2fs_bio_info *io)
234 struct f2fs_io_info *fio = &io->fio;
239 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
241 if (is_read_io(fio->op))
242 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
244 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
246 __submit_bio(io->sbi, io->bio, fio->type);
250 static bool __has_merged_page(struct f2fs_bio_info *io,
251 struct inode *inode, nid_t ino, pgoff_t idx)
253 struct bio_vec *bvec;
263 bio_for_each_segment_all(bvec, io->bio, i) {
265 if (bvec->bv_page->mapping)
266 target = bvec->bv_page;
268 target = fscrypt_control_page(bvec->bv_page);
270 if (idx != target->index)
273 if (inode && inode == target->mapping->host)
275 if (ino && ino == ino_of_node(target))
282 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
283 nid_t ino, pgoff_t idx, enum page_type type)
285 enum page_type btype = PAGE_TYPE_OF_BIO(type);
287 struct f2fs_bio_info *io;
290 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
291 io = sbi->write_io[btype] + temp;
293 down_read(&io->io_rwsem);
294 ret = __has_merged_page(io, inode, ino, idx);
295 up_read(&io->io_rwsem);
297 /* TODO: use HOT temp only for meta pages now. */
298 if (ret || btype == META)
304 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
305 enum page_type type, enum temp_type temp)
307 enum page_type btype = PAGE_TYPE_OF_BIO(type);
308 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
310 down_write(&io->io_rwsem);
312 /* change META to META_FLUSH in the checkpoint procedure */
313 if (type >= META_FLUSH) {
314 io->fio.type = META_FLUSH;
315 io->fio.op = REQ_OP_WRITE;
316 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
317 if (!test_opt(sbi, NOBARRIER))
318 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
320 __submit_merged_bio(io);
321 up_write(&io->io_rwsem);
324 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
325 struct inode *inode, nid_t ino, pgoff_t idx,
326 enum page_type type, bool force)
330 if (!force && !has_merged_page(sbi, inode, ino, idx, type))
333 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
335 __f2fs_submit_merged_write(sbi, type, temp);
337 /* TODO: use HOT temp only for meta pages now. */
343 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
345 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
348 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
349 struct inode *inode, nid_t ino, pgoff_t idx,
352 __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
355 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
357 f2fs_submit_merged_write(sbi, DATA);
358 f2fs_submit_merged_write(sbi, NODE);
359 f2fs_submit_merged_write(sbi, META);
363 * Fill the locked page with data located in the block address.
364 * A caller needs to unlock the page on failure.
366 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
369 struct page *page = fio->encrypted_page ?
370 fio->encrypted_page : fio->page;
372 trace_f2fs_submit_page_bio(page, fio);
373 f2fs_trace_ios(fio, 0);
375 /* Allocate a new bio */
376 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
378 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
382 bio_set_op_attrs(bio, fio->op, fio->op_flags);
384 __submit_bio(fio->sbi, bio, fio->type);
386 if (!is_read_io(fio->op))
387 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
391 int f2fs_submit_page_write(struct f2fs_io_info *fio)
393 struct f2fs_sb_info *sbi = fio->sbi;
394 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
395 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
396 struct page *bio_page;
399 f2fs_bug_on(sbi, is_read_io(fio->op));
401 down_write(&io->io_rwsem);
404 spin_lock(&io->io_lock);
405 if (list_empty(&io->io_list)) {
406 spin_unlock(&io->io_lock);
409 fio = list_first_entry(&io->io_list,
410 struct f2fs_io_info, list);
411 list_del(&fio->list);
412 spin_unlock(&io->io_lock);
415 if (fio->old_blkaddr != NEW_ADDR)
416 verify_block_addr(sbi, fio->old_blkaddr);
417 verify_block_addr(sbi, fio->new_blkaddr);
419 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
421 /* set submitted = true as a return value */
422 fio->submitted = true;
424 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
426 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
427 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
428 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
429 __submit_merged_bio(io);
431 if (io->bio == NULL) {
432 if ((fio->type == DATA || fio->type == NODE) &&
433 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
435 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
438 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
439 BIO_MAX_PAGES, false);
443 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
444 __submit_merged_bio(io);
448 io->last_block_in_bio = fio->new_blkaddr;
449 f2fs_trace_ios(fio, 0);
451 trace_f2fs_submit_page_write(fio->page, fio);
456 up_write(&io->io_rwsem);
460 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
463 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
464 struct fscrypt_ctx *ctx = NULL;
467 if (f2fs_encrypted_file(inode)) {
468 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
470 return ERR_CAST(ctx);
472 /* wait the page to be moved by cleaning */
473 f2fs_wait_on_block_writeback(sbi, blkaddr);
476 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
479 fscrypt_release_ctx(ctx);
480 return ERR_PTR(-ENOMEM);
482 f2fs_target_device(sbi, blkaddr, bio);
483 bio->bi_end_io = f2fs_read_end_io;
484 bio->bi_private = ctx;
485 bio_set_op_attrs(bio, REQ_OP_READ, 0);
490 /* This can handle encryption stuffs */
491 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
494 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1);
499 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
503 __submit_bio(F2FS_I_SB(inode), bio, DATA);
507 static void __set_data_blkaddr(struct dnode_of_data *dn)
509 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
513 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
514 base = get_extra_isize(dn->inode);
516 /* Get physical address of data block */
517 addr_array = blkaddr_in_node(rn);
518 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
522 * Lock ordering for the change of data block address:
525 * update block addresses in the node page
527 void set_data_blkaddr(struct dnode_of_data *dn)
529 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
530 __set_data_blkaddr(dn);
531 if (set_page_dirty(dn->node_page))
532 dn->node_changed = true;
535 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
537 dn->data_blkaddr = blkaddr;
538 set_data_blkaddr(dn);
539 f2fs_update_extent_cache(dn);
542 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
543 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
545 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
551 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
553 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
556 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
557 dn->ofs_in_node, count);
559 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
561 for (; count > 0; dn->ofs_in_node++) {
562 block_t blkaddr = datablock_addr(dn->inode,
563 dn->node_page, dn->ofs_in_node);
564 if (blkaddr == NULL_ADDR) {
565 dn->data_blkaddr = NEW_ADDR;
566 __set_data_blkaddr(dn);
571 if (set_page_dirty(dn->node_page))
572 dn->node_changed = true;
576 /* Should keep dn->ofs_in_node unchanged */
577 int reserve_new_block(struct dnode_of_data *dn)
579 unsigned int ofs_in_node = dn->ofs_in_node;
582 ret = reserve_new_blocks(dn, 1);
583 dn->ofs_in_node = ofs_in_node;
587 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
589 bool need_put = dn->inode_page ? false : true;
592 err = get_dnode_of_data(dn, index, ALLOC_NODE);
596 if (dn->data_blkaddr == NULL_ADDR)
597 err = reserve_new_block(dn);
603 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
605 struct extent_info ei = {0,0,0};
606 struct inode *inode = dn->inode;
608 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
609 dn->data_blkaddr = ei.blk + index - ei.fofs;
613 return f2fs_reserve_block(dn, index);
616 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
617 int op_flags, bool for_write)
619 struct address_space *mapping = inode->i_mapping;
620 struct dnode_of_data dn;
622 struct extent_info ei = {0,0,0};
625 page = f2fs_grab_cache_page(mapping, index, for_write);
627 return ERR_PTR(-ENOMEM);
629 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
630 dn.data_blkaddr = ei.blk + index - ei.fofs;
634 set_new_dnode(&dn, inode, NULL, NULL, 0);
635 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
640 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
645 if (PageUptodate(page)) {
651 * A new dentry page is allocated but not able to be written, since its
652 * new inode page couldn't be allocated due to -ENOSPC.
653 * In such the case, its blkaddr can be remained as NEW_ADDR.
654 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
656 if (dn.data_blkaddr == NEW_ADDR) {
657 zero_user_segment(page, 0, PAGE_SIZE);
658 if (!PageUptodate(page))
659 SetPageUptodate(page);
664 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
670 f2fs_put_page(page, 1);
674 struct page *find_data_page(struct inode *inode, pgoff_t index)
676 struct address_space *mapping = inode->i_mapping;
679 page = find_get_page(mapping, index);
680 if (page && PageUptodate(page))
682 f2fs_put_page(page, 0);
684 page = get_read_data_page(inode, index, 0, false);
688 if (PageUptodate(page))
691 wait_on_page_locked(page);
692 if (unlikely(!PageUptodate(page))) {
693 f2fs_put_page(page, 0);
694 return ERR_PTR(-EIO);
700 * If it tries to access a hole, return an error.
701 * Because, the callers, functions in dir.c and GC, should be able to know
702 * whether this page exists or not.
704 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
707 struct address_space *mapping = inode->i_mapping;
710 page = get_read_data_page(inode, index, 0, for_write);
714 /* wait for read completion */
716 if (unlikely(page->mapping != mapping)) {
717 f2fs_put_page(page, 1);
720 if (unlikely(!PageUptodate(page))) {
721 f2fs_put_page(page, 1);
722 return ERR_PTR(-EIO);
728 * Caller ensures that this data page is never allocated.
729 * A new zero-filled data page is allocated in the page cache.
731 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
733 * Note that, ipage is set only by make_empty_dir, and if any error occur,
734 * ipage should be released by this function.
736 struct page *get_new_data_page(struct inode *inode,
737 struct page *ipage, pgoff_t index, bool new_i_size)
739 struct address_space *mapping = inode->i_mapping;
741 struct dnode_of_data dn;
744 page = f2fs_grab_cache_page(mapping, index, true);
747 * before exiting, we should make sure ipage will be released
748 * if any error occur.
750 f2fs_put_page(ipage, 1);
751 return ERR_PTR(-ENOMEM);
754 set_new_dnode(&dn, inode, ipage, NULL, 0);
755 err = f2fs_reserve_block(&dn, index);
757 f2fs_put_page(page, 1);
763 if (PageUptodate(page))
766 if (dn.data_blkaddr == NEW_ADDR) {
767 zero_user_segment(page, 0, PAGE_SIZE);
768 if (!PageUptodate(page))
769 SetPageUptodate(page);
771 f2fs_put_page(page, 1);
773 /* if ipage exists, blkaddr should be NEW_ADDR */
774 f2fs_bug_on(F2FS_I_SB(inode), ipage);
775 page = get_lock_data_page(inode, index, true);
780 if (new_i_size && i_size_read(inode) <
781 ((loff_t)(index + 1) << PAGE_SHIFT))
782 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
786 static int __allocate_data_block(struct dnode_of_data *dn)
788 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
789 struct f2fs_summary sum;
795 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
798 dn->data_blkaddr = datablock_addr(dn->inode,
799 dn->node_page, dn->ofs_in_node);
800 if (dn->data_blkaddr == NEW_ADDR)
803 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
807 get_node_info(sbi, dn->nid, &ni);
808 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
810 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
811 &sum, CURSEG_WARM_DATA, NULL, false);
812 set_data_blkaddr(dn);
815 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
817 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
818 f2fs_i_size_write(dn->inode,
819 ((loff_t)(fofs + 1) << PAGE_SHIFT));
823 static inline bool __force_buffered_io(struct inode *inode, int rw)
825 return (f2fs_encrypted_file(inode) ||
826 (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
827 F2FS_I_SB(inode)->s_ndevs);
830 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
832 struct inode *inode = file_inode(iocb->ki_filp);
833 struct f2fs_map_blocks map;
836 /* convert inline data for Direct I/O*/
837 if (iocb->ki_flags & IOCB_DIRECT) {
838 err = f2fs_convert_inline_inode(inode);
843 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
846 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
847 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
848 if (map.m_len > map.m_lblk)
849 map.m_len -= map.m_lblk;
853 map.m_next_pgofs = NULL;
855 if (iocb->ki_flags & IOCB_DIRECT)
856 return f2fs_map_blocks(inode, &map, 1,
857 __force_buffered_io(inode, WRITE) ?
858 F2FS_GET_BLOCK_PRE_AIO :
859 F2FS_GET_BLOCK_PRE_DIO);
860 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
861 err = f2fs_convert_inline_inode(inode);
865 if (!f2fs_has_inline_data(inode))
866 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
870 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
872 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
874 down_read(&sbi->node_change);
876 up_read(&sbi->node_change);
886 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
887 * f2fs_map_blocks structure.
888 * If original data blocks are allocated, then give them to blockdev.
890 * a. preallocate requested block addresses
891 * b. do not use extent cache for better performance
892 * c. give the block addresses to blockdev
894 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
895 int create, int flag)
897 unsigned int maxblocks = map->m_len;
898 struct dnode_of_data dn;
899 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
900 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
901 pgoff_t pgofs, end_offset, end;
902 int err = 0, ofs = 1;
903 unsigned int ofs_in_node, last_ofs_in_node;
905 struct extent_info ei = {0,0,0};
914 /* it only supports block size == page size */
915 pgofs = (pgoff_t)map->m_lblk;
916 end = pgofs + maxblocks;
918 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
919 map->m_pblk = ei.blk + pgofs - ei.fofs;
920 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
921 map->m_flags = F2FS_MAP_MAPPED;
927 __do_map_lock(sbi, flag, true);
929 /* When reading holes, we need its node page */
930 set_new_dnode(&dn, inode, NULL, NULL, 0);
931 err = get_dnode_of_data(&dn, pgofs, mode);
933 if (flag == F2FS_GET_BLOCK_BMAP)
935 if (err == -ENOENT) {
937 if (map->m_next_pgofs)
939 get_next_page_offset(&dn, pgofs);
945 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
946 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
949 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
951 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
953 if (unlikely(f2fs_cp_error(sbi))) {
957 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
958 if (blkaddr == NULL_ADDR) {
960 last_ofs_in_node = dn.ofs_in_node;
963 err = __allocate_data_block(&dn);
965 set_inode_flag(inode, FI_APPEND_WRITE);
969 map->m_flags |= F2FS_MAP_NEW;
970 blkaddr = dn.data_blkaddr;
972 if (flag == F2FS_GET_BLOCK_BMAP) {
976 if (flag == F2FS_GET_BLOCK_FIEMAP &&
977 blkaddr == NULL_ADDR) {
978 if (map->m_next_pgofs)
979 *map->m_next_pgofs = pgofs + 1;
981 if (flag != F2FS_GET_BLOCK_FIEMAP ||
987 if (flag == F2FS_GET_BLOCK_PRE_AIO)
990 if (map->m_len == 0) {
991 /* preallocated unwritten block should be mapped for fiemap. */
992 if (blkaddr == NEW_ADDR)
993 map->m_flags |= F2FS_MAP_UNWRITTEN;
994 map->m_flags |= F2FS_MAP_MAPPED;
996 map->m_pblk = blkaddr;
998 } else if ((map->m_pblk != NEW_ADDR &&
999 blkaddr == (map->m_pblk + ofs)) ||
1000 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1001 flag == F2FS_GET_BLOCK_PRE_DIO) {
1012 /* preallocate blocks in batch for one dnode page */
1013 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1014 (pgofs == end || dn.ofs_in_node == end_offset)) {
1016 dn.ofs_in_node = ofs_in_node;
1017 err = reserve_new_blocks(&dn, prealloc);
1021 map->m_len += dn.ofs_in_node - ofs_in_node;
1022 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1026 dn.ofs_in_node = end_offset;
1031 else if (dn.ofs_in_node < end_offset)
1034 f2fs_put_dnode(&dn);
1037 __do_map_lock(sbi, flag, false);
1038 f2fs_balance_fs(sbi, dn.node_changed);
1043 f2fs_put_dnode(&dn);
1046 __do_map_lock(sbi, flag, false);
1047 f2fs_balance_fs(sbi, dn.node_changed);
1050 trace_f2fs_map_blocks(inode, map, err);
1054 static int __get_data_block(struct inode *inode, sector_t iblock,
1055 struct buffer_head *bh, int create, int flag,
1056 pgoff_t *next_pgofs)
1058 struct f2fs_map_blocks map;
1061 map.m_lblk = iblock;
1062 map.m_len = bh->b_size >> inode->i_blkbits;
1063 map.m_next_pgofs = next_pgofs;
1065 err = f2fs_map_blocks(inode, &map, create, flag);
1067 map_bh(bh, inode->i_sb, map.m_pblk);
1068 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1069 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1074 static int get_data_block(struct inode *inode, sector_t iblock,
1075 struct buffer_head *bh_result, int create, int flag,
1076 pgoff_t *next_pgofs)
1078 return __get_data_block(inode, iblock, bh_result, create,
1082 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1083 struct buffer_head *bh_result, int create)
1085 return __get_data_block(inode, iblock, bh_result, create,
1086 F2FS_GET_BLOCK_DEFAULT, NULL);
1089 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1090 struct buffer_head *bh_result, int create)
1092 /* Block number less than F2FS MAX BLOCKS */
1093 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1096 return __get_data_block(inode, iblock, bh_result, create,
1097 F2FS_GET_BLOCK_BMAP, NULL);
1100 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1102 return (offset >> inode->i_blkbits);
1105 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1107 return (blk << inode->i_blkbits);
1110 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1113 struct buffer_head map_bh;
1114 sector_t start_blk, last_blk;
1116 u64 logical = 0, phys = 0, size = 0;
1120 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1124 if (f2fs_has_inline_data(inode)) {
1125 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1132 if (logical_to_blk(inode, len) == 0)
1133 len = blk_to_logical(inode, 1);
1135 start_blk = logical_to_blk(inode, start);
1136 last_blk = logical_to_blk(inode, start + len - 1);
1139 memset(&map_bh, 0, sizeof(struct buffer_head));
1140 map_bh.b_size = len;
1142 ret = get_data_block(inode, start_blk, &map_bh, 0,
1143 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1148 if (!buffer_mapped(&map_bh)) {
1149 start_blk = next_pgofs;
1151 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1152 F2FS_I_SB(inode)->max_file_blocks))
1155 flags |= FIEMAP_EXTENT_LAST;
1159 if (f2fs_encrypted_inode(inode))
1160 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1162 ret = fiemap_fill_next_extent(fieinfo, logical,
1166 if (start_blk > last_blk || ret)
1169 logical = blk_to_logical(inode, start_blk);
1170 phys = blk_to_logical(inode, map_bh.b_blocknr);
1171 size = map_bh.b_size;
1173 if (buffer_unwritten(&map_bh))
1174 flags = FIEMAP_EXTENT_UNWRITTEN;
1176 start_blk += logical_to_blk(inode, size);
1180 if (fatal_signal_pending(current))
1188 inode_unlock(inode);
1193 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1194 * Major change was from block_size == page_size in f2fs by default.
1196 static int f2fs_mpage_readpages(struct address_space *mapping,
1197 struct list_head *pages, struct page *page,
1200 struct bio *bio = NULL;
1202 sector_t last_block_in_bio = 0;
1203 struct inode *inode = mapping->host;
1204 const unsigned blkbits = inode->i_blkbits;
1205 const unsigned blocksize = 1 << blkbits;
1206 sector_t block_in_file;
1207 sector_t last_block;
1208 sector_t last_block_in_file;
1210 struct f2fs_map_blocks map;
1216 map.m_next_pgofs = NULL;
1218 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1221 page = list_last_entry(pages, struct page, lru);
1223 prefetchw(&page->flags);
1224 list_del(&page->lru);
1225 if (add_to_page_cache_lru(page, mapping,
1227 readahead_gfp_mask(mapping)))
1231 block_in_file = (sector_t)page->index;
1232 last_block = block_in_file + nr_pages;
1233 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1235 if (last_block > last_block_in_file)
1236 last_block = last_block_in_file;
1239 * Map blocks using the previous result first.
1241 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1242 block_in_file > map.m_lblk &&
1243 block_in_file < (map.m_lblk + map.m_len))
1247 * Then do more f2fs_map_blocks() calls until we are
1248 * done with this page.
1252 if (block_in_file < last_block) {
1253 map.m_lblk = block_in_file;
1254 map.m_len = last_block - block_in_file;
1256 if (f2fs_map_blocks(inode, &map, 0,
1257 F2FS_GET_BLOCK_DEFAULT))
1258 goto set_error_page;
1261 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1262 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1263 SetPageMappedToDisk(page);
1265 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1266 SetPageUptodate(page);
1270 zero_user_segment(page, 0, PAGE_SIZE);
1271 if (!PageUptodate(page))
1272 SetPageUptodate(page);
1278 * This page will go to BIO. Do we need to send this
1281 if (bio && (last_block_in_bio != block_nr - 1 ||
1282 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1284 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1288 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages);
1291 goto set_error_page;
1295 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1296 goto submit_and_realloc;
1298 last_block_in_bio = block_nr;
1302 zero_user_segment(page, 0, PAGE_SIZE);
1307 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1315 BUG_ON(pages && !list_empty(pages));
1317 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1321 static int f2fs_read_data_page(struct file *file, struct page *page)
1323 struct inode *inode = page->mapping->host;
1326 trace_f2fs_readpage(page, DATA);
1328 /* If the file has inline data, try to read it directly */
1329 if (f2fs_has_inline_data(inode))
1330 ret = f2fs_read_inline_data(inode, page);
1332 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1336 static int f2fs_read_data_pages(struct file *file,
1337 struct address_space *mapping,
1338 struct list_head *pages, unsigned nr_pages)
1340 struct inode *inode = mapping->host;
1341 struct page *page = list_last_entry(pages, struct page, lru);
1343 trace_f2fs_readpages(inode, page, nr_pages);
1345 /* If the file has inline data, skip readpages */
1346 if (f2fs_has_inline_data(inode))
1349 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1352 static int encrypt_one_page(struct f2fs_io_info *fio)
1354 struct inode *inode = fio->page->mapping->host;
1355 gfp_t gfp_flags = GFP_NOFS;
1357 if (!f2fs_encrypted_file(inode))
1360 /* wait for GCed encrypted page writeback */
1361 f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1364 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1365 PAGE_SIZE, 0, fio->page->index, gfp_flags);
1366 if (!IS_ERR(fio->encrypted_page))
1369 /* flush pending IOs and wait for a while in the ENOMEM case */
1370 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1371 f2fs_flush_merged_writes(fio->sbi);
1372 congestion_wait(BLK_RW_ASYNC, HZ/50);
1373 gfp_flags |= __GFP_NOFAIL;
1376 return PTR_ERR(fio->encrypted_page);
1379 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1381 struct inode *inode = fio->page->mapping->host;
1383 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
1385 if (is_cold_data(fio->page))
1387 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1390 return need_inplace_update_policy(inode, fio);
1393 static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio)
1395 if (fio->old_blkaddr == NEW_ADDR)
1397 if (fio->old_blkaddr == NULL_ADDR)
1402 int do_write_data_page(struct f2fs_io_info *fio)
1404 struct page *page = fio->page;
1405 struct inode *inode = page->mapping->host;
1406 struct dnode_of_data dn;
1407 struct extent_info ei = {0,0,0};
1408 bool ipu_force = false;
1411 set_new_dnode(&dn, inode, NULL, NULL, 0);
1412 if (need_inplace_update(fio) &&
1413 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1414 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1416 if (valid_ipu_blkaddr(fio)) {
1418 fio->need_lock = LOCK_DONE;
1423 /* Deadlock due to between page->lock and f2fs_lock_op */
1424 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1427 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1431 fio->old_blkaddr = dn.data_blkaddr;
1433 /* This page is already truncated */
1434 if (fio->old_blkaddr == NULL_ADDR) {
1435 ClearPageUptodate(page);
1440 * If current allocation needs SSR,
1441 * it had better in-place writes for updated data.
1443 if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) {
1444 err = encrypt_one_page(fio);
1448 set_page_writeback(page);
1449 f2fs_put_dnode(&dn);
1450 if (fio->need_lock == LOCK_REQ)
1451 f2fs_unlock_op(fio->sbi);
1452 err = rewrite_data_page(fio);
1453 trace_f2fs_do_write_data_page(fio->page, IPU);
1454 set_inode_flag(inode, FI_UPDATE_WRITE);
1458 if (fio->need_lock == LOCK_RETRY) {
1459 if (!f2fs_trylock_op(fio->sbi)) {
1463 fio->need_lock = LOCK_REQ;
1466 err = encrypt_one_page(fio);
1470 set_page_writeback(page);
1472 /* LFS mode write path */
1473 write_data_page(&dn, fio);
1474 trace_f2fs_do_write_data_page(page, OPU);
1475 set_inode_flag(inode, FI_APPEND_WRITE);
1476 if (page->index == 0)
1477 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1479 f2fs_put_dnode(&dn);
1481 if (fio->need_lock == LOCK_REQ)
1482 f2fs_unlock_op(fio->sbi);
1486 static int __write_data_page(struct page *page, bool *submitted,
1487 struct writeback_control *wbc,
1488 enum iostat_type io_type)
1490 struct inode *inode = page->mapping->host;
1491 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1492 loff_t i_size = i_size_read(inode);
1493 const pgoff_t end_index = ((unsigned long long) i_size)
1495 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1496 unsigned offset = 0;
1497 bool need_balance_fs = false;
1499 struct f2fs_io_info fio = {
1501 .ino = inode->i_ino,
1504 .op_flags = wbc_to_write_flags(wbc),
1505 .old_blkaddr = NULL_ADDR,
1507 .encrypted_page = NULL,
1509 .need_lock = LOCK_RETRY,
1513 trace_f2fs_writepage(page, DATA);
1515 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1518 if (page->index < end_index)
1522 * If the offset is out-of-range of file size,
1523 * this page does not have to be written to disk.
1525 offset = i_size & (PAGE_SIZE - 1);
1526 if ((page->index >= end_index + 1) || !offset)
1529 zero_user_segment(page, offset, PAGE_SIZE);
1531 if (f2fs_is_drop_cache(inode))
1533 /* we should not write 0'th page having journal header */
1534 if (f2fs_is_volatile_file(inode) && (!page->index ||
1535 (!wbc->for_reclaim &&
1536 available_free_memory(sbi, BASE_CHECK))))
1539 /* we should bypass data pages to proceed the kworkder jobs */
1540 if (unlikely(f2fs_cp_error(sbi))) {
1541 mapping_set_error(page->mapping, -EIO);
1545 /* Dentry blocks are controlled by checkpoint */
1546 if (S_ISDIR(inode->i_mode)) {
1547 fio.need_lock = LOCK_DONE;
1548 err = do_write_data_page(&fio);
1552 if (!wbc->for_reclaim)
1553 need_balance_fs = true;
1554 else if (has_not_enough_free_secs(sbi, 0, 0))
1557 set_inode_flag(inode, FI_HOT_DATA);
1560 if (f2fs_has_inline_data(inode)) {
1561 err = f2fs_write_inline_data(inode, page);
1566 if (err == -EAGAIN) {
1567 err = do_write_data_page(&fio);
1568 if (err == -EAGAIN) {
1569 fio.need_lock = LOCK_REQ;
1570 err = do_write_data_page(&fio);
1574 down_write(&F2FS_I(inode)->i_sem);
1575 if (F2FS_I(inode)->last_disk_size < psize)
1576 F2FS_I(inode)->last_disk_size = psize;
1577 up_write(&F2FS_I(inode)->i_sem);
1580 if (err && err != -ENOENT)
1584 inode_dec_dirty_pages(inode);
1586 ClearPageUptodate(page);
1588 if (wbc->for_reclaim) {
1589 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1590 clear_inode_flag(inode, FI_HOT_DATA);
1591 remove_dirty_inode(inode);
1596 if (!S_ISDIR(inode->i_mode))
1597 f2fs_balance_fs(sbi, need_balance_fs);
1599 if (unlikely(f2fs_cp_error(sbi))) {
1600 f2fs_submit_merged_write(sbi, DATA);
1605 *submitted = fio.submitted;
1610 redirty_page_for_writepage(wbc, page);
1612 return AOP_WRITEPAGE_ACTIVATE;
1617 static int f2fs_write_data_page(struct page *page,
1618 struct writeback_control *wbc)
1620 return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1624 * This function was copied from write_cche_pages from mm/page-writeback.c.
1625 * The major change is making write step of cold data page separately from
1626 * warm/hot data page.
1628 static int f2fs_write_cache_pages(struct address_space *mapping,
1629 struct writeback_control *wbc,
1630 enum iostat_type io_type)
1634 struct pagevec pvec;
1636 pgoff_t uninitialized_var(writeback_index);
1638 pgoff_t end; /* Inclusive */
1640 pgoff_t last_idx = ULONG_MAX;
1642 int range_whole = 0;
1645 pagevec_init(&pvec, 0);
1647 if (get_dirty_pages(mapping->host) <=
1648 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1649 set_inode_flag(mapping->host, FI_HOT_DATA);
1651 clear_inode_flag(mapping->host, FI_HOT_DATA);
1653 if (wbc->range_cyclic) {
1654 writeback_index = mapping->writeback_index; /* prev offset */
1655 index = writeback_index;
1662 index = wbc->range_start >> PAGE_SHIFT;
1663 end = wbc->range_end >> PAGE_SHIFT;
1664 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1666 cycled = 1; /* ignore range_cyclic tests */
1668 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1669 tag = PAGECACHE_TAG_TOWRITE;
1671 tag = PAGECACHE_TAG_DIRTY;
1673 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1674 tag_pages_for_writeback(mapping, index, end);
1676 while (!done && (index <= end)) {
1679 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1680 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1684 for (i = 0; i < nr_pages; i++) {
1685 struct page *page = pvec.pages[i];
1686 bool submitted = false;
1688 if (page->index > end) {
1693 done_index = page->index;
1697 if (unlikely(page->mapping != mapping)) {
1703 if (!PageDirty(page)) {
1704 /* someone wrote it for us */
1705 goto continue_unlock;
1708 if (PageWriteback(page)) {
1709 if (wbc->sync_mode != WB_SYNC_NONE)
1710 f2fs_wait_on_page_writeback(page,
1713 goto continue_unlock;
1716 BUG_ON(PageWriteback(page));
1717 if (!clear_page_dirty_for_io(page))
1718 goto continue_unlock;
1720 ret = __write_data_page(page, &submitted, wbc, io_type);
1721 if (unlikely(ret)) {
1723 * keep nr_to_write, since vfs uses this to
1724 * get # of written pages.
1726 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1730 } else if (ret == -EAGAIN) {
1732 if (wbc->sync_mode == WB_SYNC_ALL) {
1734 congestion_wait(BLK_RW_ASYNC,
1740 done_index = page->index + 1;
1743 } else if (submitted) {
1744 last_idx = page->index;
1747 /* give a priority to WB_SYNC threads */
1748 if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) ||
1749 --wbc->nr_to_write <= 0) &&
1750 wbc->sync_mode == WB_SYNC_NONE) {
1755 pagevec_release(&pvec);
1759 if (!cycled && !done) {
1762 end = writeback_index - 1;
1765 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1766 mapping->writeback_index = done_index;
1768 if (last_idx != ULONG_MAX)
1769 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
1775 int __f2fs_write_data_pages(struct address_space *mapping,
1776 struct writeback_control *wbc,
1777 enum iostat_type io_type)
1779 struct inode *inode = mapping->host;
1780 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1781 struct blk_plug plug;
1784 /* deal with chardevs and other special file */
1785 if (!mapping->a_ops->writepage)
1788 /* skip writing if there is no dirty page in this inode */
1789 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1792 /* during POR, we don't need to trigger writepage at all. */
1793 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1796 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1797 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1798 available_free_memory(sbi, DIRTY_DENTS))
1801 /* skip writing during file defragment */
1802 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1805 trace_f2fs_writepages(mapping->host, wbc, DATA);
1807 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
1808 if (wbc->sync_mode == WB_SYNC_ALL)
1809 atomic_inc(&sbi->wb_sync_req);
1810 else if (atomic_read(&sbi->wb_sync_req))
1813 blk_start_plug(&plug);
1814 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
1815 blk_finish_plug(&plug);
1817 if (wbc->sync_mode == WB_SYNC_ALL)
1818 atomic_dec(&sbi->wb_sync_req);
1820 * if some pages were truncated, we cannot guarantee its mapping->host
1821 * to detect pending bios.
1824 remove_dirty_inode(inode);
1828 wbc->pages_skipped += get_dirty_pages(inode);
1829 trace_f2fs_writepages(mapping->host, wbc, DATA);
1833 static int f2fs_write_data_pages(struct address_space *mapping,
1834 struct writeback_control *wbc)
1836 struct inode *inode = mapping->host;
1838 return __f2fs_write_data_pages(mapping, wbc,
1839 F2FS_I(inode)->cp_task == current ?
1840 FS_CP_DATA_IO : FS_DATA_IO);
1843 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1845 struct inode *inode = mapping->host;
1846 loff_t i_size = i_size_read(inode);
1849 down_write(&F2FS_I(inode)->i_mmap_sem);
1850 truncate_pagecache(inode, i_size);
1851 truncate_blocks(inode, i_size, true);
1852 up_write(&F2FS_I(inode)->i_mmap_sem);
1856 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1857 struct page *page, loff_t pos, unsigned len,
1858 block_t *blk_addr, bool *node_changed)
1860 struct inode *inode = page->mapping->host;
1861 pgoff_t index = page->index;
1862 struct dnode_of_data dn;
1864 bool locked = false;
1865 struct extent_info ei = {0,0,0};
1869 * we already allocated all the blocks, so we don't need to get
1870 * the block addresses when there is no need to fill the page.
1872 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1873 !is_inode_flag_set(inode, FI_NO_PREALLOC))
1876 if (f2fs_has_inline_data(inode) ||
1877 (pos & PAGE_MASK) >= i_size_read(inode)) {
1878 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1882 /* check inline_data */
1883 ipage = get_node_page(sbi, inode->i_ino);
1884 if (IS_ERR(ipage)) {
1885 err = PTR_ERR(ipage);
1889 set_new_dnode(&dn, inode, ipage, ipage, 0);
1891 if (f2fs_has_inline_data(inode)) {
1892 if (pos + len <= MAX_INLINE_DATA(inode)) {
1893 read_inline_data(page, ipage);
1894 set_inode_flag(inode, FI_DATA_EXIST);
1896 set_inline_node(ipage);
1898 err = f2fs_convert_inline_page(&dn, page);
1901 if (dn.data_blkaddr == NULL_ADDR)
1902 err = f2fs_get_block(&dn, index);
1904 } else if (locked) {
1905 err = f2fs_get_block(&dn, index);
1907 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1908 dn.data_blkaddr = ei.blk + index - ei.fofs;
1911 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1912 if (err || dn.data_blkaddr == NULL_ADDR) {
1913 f2fs_put_dnode(&dn);
1914 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
1922 /* convert_inline_page can make node_changed */
1923 *blk_addr = dn.data_blkaddr;
1924 *node_changed = dn.node_changed;
1926 f2fs_put_dnode(&dn);
1929 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1933 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1934 loff_t pos, unsigned len, unsigned flags,
1935 struct page **pagep, void **fsdata)
1937 struct inode *inode = mapping->host;
1938 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1939 struct page *page = NULL;
1940 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1941 bool need_balance = false;
1942 block_t blkaddr = NULL_ADDR;
1945 trace_f2fs_write_begin(inode, pos, len, flags);
1947 if (f2fs_is_atomic_file(inode) &&
1948 !available_free_memory(sbi, INMEM_PAGES)) {
1954 * We should check this at this moment to avoid deadlock on inode page
1955 * and #0 page. The locking rule for inline_data conversion should be:
1956 * lock_page(page #0) -> lock_page(inode_page)
1959 err = f2fs_convert_inline_inode(inode);
1965 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1966 * wait_for_stable_page. Will wait that below with our IO control.
1968 page = pagecache_get_page(mapping, index,
1969 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1977 err = prepare_write_begin(sbi, page, pos, len,
1978 &blkaddr, &need_balance);
1982 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1984 f2fs_balance_fs(sbi, true);
1986 if (page->mapping != mapping) {
1987 /* The page got truncated from under us */
1988 f2fs_put_page(page, 1);
1993 f2fs_wait_on_page_writeback(page, DATA, false);
1995 /* wait for GCed encrypted page writeback */
1996 if (f2fs_encrypted_file(inode))
1997 f2fs_wait_on_block_writeback(sbi, blkaddr);
1999 if (len == PAGE_SIZE || PageUptodate(page))
2002 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2003 zero_user_segment(page, len, PAGE_SIZE);
2007 if (blkaddr == NEW_ADDR) {
2008 zero_user_segment(page, 0, PAGE_SIZE);
2009 SetPageUptodate(page);
2011 err = f2fs_submit_page_read(inode, page, blkaddr);
2016 if (unlikely(page->mapping != mapping)) {
2017 f2fs_put_page(page, 1);
2020 if (unlikely(!PageUptodate(page))) {
2028 f2fs_put_page(page, 1);
2029 f2fs_write_failed(mapping, pos + len);
2030 if (f2fs_is_atomic_file(inode))
2031 drop_inmem_pages_all(sbi);
2035 static int f2fs_write_end(struct file *file,
2036 struct address_space *mapping,
2037 loff_t pos, unsigned len, unsigned copied,
2038 struct page *page, void *fsdata)
2040 struct inode *inode = page->mapping->host;
2042 trace_f2fs_write_end(inode, pos, len, copied);
2045 * This should be come from len == PAGE_SIZE, and we expect copied
2046 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2047 * let generic_perform_write() try to copy data again through copied=0.
2049 if (!PageUptodate(page)) {
2050 if (unlikely(copied != len))
2053 SetPageUptodate(page);
2058 set_page_dirty(page);
2060 if (pos + copied > i_size_read(inode))
2061 f2fs_i_size_write(inode, pos + copied);
2063 f2fs_put_page(page, 1);
2064 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2068 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2071 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
2073 if (offset & blocksize_mask)
2076 if (iov_iter_alignment(iter) & blocksize_mask)
2082 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2084 struct address_space *mapping = iocb->ki_filp->f_mapping;
2085 struct inode *inode = mapping->host;
2086 size_t count = iov_iter_count(iter);
2087 loff_t offset = iocb->ki_pos;
2088 int rw = iov_iter_rw(iter);
2091 err = check_direct_IO(inode, iter, offset);
2095 if (__force_buffered_io(inode, rw))
2098 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2100 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
2101 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2102 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
2106 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2108 set_inode_flag(inode, FI_UPDATE_WRITE);
2109 } else if (err < 0) {
2110 f2fs_write_failed(mapping, offset + count);
2114 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2119 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2120 unsigned int length)
2122 struct inode *inode = page->mapping->host;
2123 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2125 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2126 (offset % PAGE_SIZE || length != PAGE_SIZE))
2129 if (PageDirty(page)) {
2130 if (inode->i_ino == F2FS_META_INO(sbi)) {
2131 dec_page_count(sbi, F2FS_DIRTY_META);
2132 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2133 dec_page_count(sbi, F2FS_DIRTY_NODES);
2135 inode_dec_dirty_pages(inode);
2136 remove_dirty_inode(inode);
2140 /* This is atomic written page, keep Private */
2141 if (IS_ATOMIC_WRITTEN_PAGE(page))
2142 return drop_inmem_page(inode, page);
2144 set_page_private(page, 0);
2145 ClearPagePrivate(page);
2148 int f2fs_release_page(struct page *page, gfp_t wait)
2150 /* If this is dirty page, keep PagePrivate */
2151 if (PageDirty(page))
2154 /* This is atomic written page, keep Private */
2155 if (IS_ATOMIC_WRITTEN_PAGE(page))
2158 set_page_private(page, 0);
2159 ClearPagePrivate(page);
2164 * This was copied from __set_page_dirty_buffers which gives higher performance
2165 * in very high speed storages. (e.g., pmem)
2167 void f2fs_set_page_dirty_nobuffers(struct page *page)
2169 struct address_space *mapping = page->mapping;
2170 unsigned long flags;
2172 if (unlikely(!mapping))
2175 spin_lock(&mapping->private_lock);
2176 lock_page_memcg(page);
2178 spin_unlock(&mapping->private_lock);
2180 spin_lock_irqsave(&mapping->tree_lock, flags);
2181 WARN_ON_ONCE(!PageUptodate(page));
2182 account_page_dirtied(page, mapping);
2183 radix_tree_tag_set(&mapping->page_tree,
2184 page_index(page), PAGECACHE_TAG_DIRTY);
2185 spin_unlock_irqrestore(&mapping->tree_lock, flags);
2186 unlock_page_memcg(page);
2188 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2192 static int f2fs_set_data_page_dirty(struct page *page)
2194 struct address_space *mapping = page->mapping;
2195 struct inode *inode = mapping->host;
2197 trace_f2fs_set_page_dirty(page, DATA);
2199 if (!PageUptodate(page))
2200 SetPageUptodate(page);
2202 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2203 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2204 register_inmem_page(inode, page);
2208 * Previously, this page has been registered, we just
2214 if (!PageDirty(page)) {
2215 f2fs_set_page_dirty_nobuffers(page);
2216 update_dirty_page(inode, page);
2222 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2224 struct inode *inode = mapping->host;
2226 if (f2fs_has_inline_data(inode))
2229 /* make sure allocating whole blocks */
2230 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2231 filemap_write_and_wait(mapping);
2233 return generic_block_bmap(mapping, block, get_data_block_bmap);
2236 #ifdef CONFIG_MIGRATION
2237 #include <linux/migrate.h>
2239 int f2fs_migrate_page(struct address_space *mapping,
2240 struct page *newpage, struct page *page, enum migrate_mode mode)
2242 int rc, extra_count;
2243 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2244 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2246 BUG_ON(PageWriteback(page));
2248 /* migrating an atomic written page is safe with the inmem_lock hold */
2249 if (atomic_written) {
2250 if (mode != MIGRATE_SYNC)
2252 if (!mutex_trylock(&fi->inmem_lock))
2257 * A reference is expected if PagePrivate set when move mapping,
2258 * however F2FS breaks this for maintaining dirty page counts when
2259 * truncating pages. So here adjusting the 'extra_count' make it work.
2261 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2262 rc = migrate_page_move_mapping(mapping, newpage,
2263 page, NULL, mode, extra_count);
2264 if (rc != MIGRATEPAGE_SUCCESS) {
2266 mutex_unlock(&fi->inmem_lock);
2270 if (atomic_written) {
2271 struct inmem_pages *cur;
2272 list_for_each_entry(cur, &fi->inmem_pages, list)
2273 if (cur->page == page) {
2274 cur->page = newpage;
2277 mutex_unlock(&fi->inmem_lock);
2282 if (PagePrivate(page))
2283 SetPagePrivate(newpage);
2284 set_page_private(newpage, page_private(page));
2286 if (mode != MIGRATE_SYNC_NO_COPY)
2287 migrate_page_copy(newpage, page);
2289 migrate_page_states(newpage, page);
2291 return MIGRATEPAGE_SUCCESS;
2295 const struct address_space_operations f2fs_dblock_aops = {
2296 .readpage = f2fs_read_data_page,
2297 .readpages = f2fs_read_data_pages,
2298 .writepage = f2fs_write_data_page,
2299 .writepages = f2fs_write_data_pages,
2300 .write_begin = f2fs_write_begin,
2301 .write_end = f2fs_write_end,
2302 .set_page_dirty = f2fs_set_data_page_dirty,
2303 .invalidatepage = f2fs_invalidate_page,
2304 .releasepage = f2fs_release_page,
2305 .direct_IO = f2fs_direct_IO,
2307 #ifdef CONFIG_MIGRATION
2308 .migratepage = f2fs_migrate_page,