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 if (f2fs_is_multi_device(sbi)) {
137 for (i = 0; i < sbi->s_ndevs; i++) {
138 if (FDEV(i).start_blk <= blk_addr &&
139 FDEV(i).end_blk >= blk_addr) {
140 blk_addr -= FDEV(i).start_blk;
147 bio_set_dev(bio, bdev);
148 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
153 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
157 if (!f2fs_is_multi_device(sbi))
160 for (i = 0; i < sbi->s_ndevs; i++)
161 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
166 static bool __same_bdev(struct f2fs_sb_info *sbi,
167 block_t blk_addr, struct bio *bio)
169 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
170 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
174 * Low-level block read/write IO operations.
176 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
177 int npages, bool is_read)
181 bio = f2fs_bio_alloc(npages);
183 f2fs_target_device(sbi, blk_addr, bio);
184 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
185 bio->bi_private = is_read ? NULL : sbi;
190 static inline void __submit_bio(struct f2fs_sb_info *sbi,
191 struct bio *bio, enum page_type type)
193 if (!is_read_io(bio_op(bio))) {
196 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
197 current->plug && (type == DATA || type == NODE))
198 blk_finish_plug(current->plug);
200 if (type != DATA && type != NODE)
203 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
204 start %= F2FS_IO_SIZE(sbi);
209 /* fill dummy pages */
210 for (; start < F2FS_IO_SIZE(sbi); start++) {
212 mempool_alloc(sbi->write_io_dummy,
213 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
214 f2fs_bug_on(sbi, !page);
216 SetPagePrivate(page);
217 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
219 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
223 * In the NODE case, we lose next block address chain. So, we
224 * need to do checkpoint in f2fs_sync_file.
227 set_sbi_flag(sbi, SBI_NEED_CP);
230 if (is_read_io(bio_op(bio)))
231 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
233 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
237 static void __submit_merged_bio(struct f2fs_bio_info *io)
239 struct f2fs_io_info *fio = &io->fio;
244 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
246 if (is_read_io(fio->op))
247 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
249 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
251 __submit_bio(io->sbi, io->bio, fio->type);
255 static bool __has_merged_page(struct f2fs_bio_info *io,
256 struct inode *inode, nid_t ino, pgoff_t idx)
258 struct bio_vec *bvec;
268 bio_for_each_segment_all(bvec, io->bio, i) {
270 if (bvec->bv_page->mapping)
271 target = bvec->bv_page;
273 target = fscrypt_control_page(bvec->bv_page);
275 if (idx != target->index)
278 if (inode && inode == target->mapping->host)
280 if (ino && ino == ino_of_node(target))
287 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
288 nid_t ino, pgoff_t idx, enum page_type type)
290 enum page_type btype = PAGE_TYPE_OF_BIO(type);
292 struct f2fs_bio_info *io;
295 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
296 io = sbi->write_io[btype] + temp;
298 down_read(&io->io_rwsem);
299 ret = __has_merged_page(io, inode, ino, idx);
300 up_read(&io->io_rwsem);
302 /* TODO: use HOT temp only for meta pages now. */
303 if (ret || btype == META)
309 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
310 enum page_type type, enum temp_type temp)
312 enum page_type btype = PAGE_TYPE_OF_BIO(type);
313 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
315 down_write(&io->io_rwsem);
317 /* change META to META_FLUSH in the checkpoint procedure */
318 if (type >= META_FLUSH) {
319 io->fio.type = META_FLUSH;
320 io->fio.op = REQ_OP_WRITE;
321 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
322 if (!test_opt(sbi, NOBARRIER))
323 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
325 __submit_merged_bio(io);
326 up_write(&io->io_rwsem);
329 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
330 struct inode *inode, nid_t ino, pgoff_t idx,
331 enum page_type type, bool force)
335 if (!force && !has_merged_page(sbi, inode, ino, idx, type))
338 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
340 __f2fs_submit_merged_write(sbi, type, temp);
342 /* TODO: use HOT temp only for meta pages now. */
348 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
350 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
353 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
354 struct inode *inode, nid_t ino, pgoff_t idx,
357 __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
360 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
362 f2fs_submit_merged_write(sbi, DATA);
363 f2fs_submit_merged_write(sbi, NODE);
364 f2fs_submit_merged_write(sbi, META);
368 * Fill the locked page with data located in the block address.
369 * A caller needs to unlock the page on failure.
371 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
374 struct page *page = fio->encrypted_page ?
375 fio->encrypted_page : fio->page;
377 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
378 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
379 return -EFSCORRUPTED;
381 trace_f2fs_submit_page_bio(page, fio);
382 f2fs_trace_ios(fio, 0);
384 /* Allocate a new bio */
385 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
387 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
391 bio_set_op_attrs(bio, fio->op, fio->op_flags);
393 if (!is_read_io(fio->op))
394 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
396 __submit_bio(fio->sbi, bio, fio->type);
400 int f2fs_submit_page_write(struct f2fs_io_info *fio)
402 struct f2fs_sb_info *sbi = fio->sbi;
403 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
404 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
405 struct page *bio_page;
408 f2fs_bug_on(sbi, is_read_io(fio->op));
410 down_write(&io->io_rwsem);
413 spin_lock(&io->io_lock);
414 if (list_empty(&io->io_list)) {
415 spin_unlock(&io->io_lock);
418 fio = list_first_entry(&io->io_list,
419 struct f2fs_io_info, list);
420 list_del(&fio->list);
421 spin_unlock(&io->io_lock);
424 if (__is_valid_data_blkaddr(fio->old_blkaddr))
425 verify_block_addr(fio, fio->old_blkaddr);
426 verify_block_addr(fio, fio->new_blkaddr);
428 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
430 /* set submitted = 1 as a return value */
433 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
435 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
436 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
437 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
438 __submit_merged_bio(io);
440 if (io->bio == NULL) {
441 if ((fio->type == DATA || fio->type == NODE) &&
442 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
444 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
447 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
448 BIO_MAX_PAGES, false);
452 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
453 __submit_merged_bio(io);
457 io->last_block_in_bio = fio->new_blkaddr;
458 f2fs_trace_ios(fio, 0);
460 trace_f2fs_submit_page_write(fio->page, fio);
465 up_write(&io->io_rwsem);
469 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
472 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
473 struct fscrypt_ctx *ctx = NULL;
476 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
477 return ERR_PTR(-EFAULT);
479 if (f2fs_encrypted_file(inode)) {
480 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
482 return ERR_CAST(ctx);
484 /* wait the page to be moved by cleaning */
485 f2fs_wait_on_block_writeback(sbi, blkaddr);
488 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
491 fscrypt_release_ctx(ctx);
492 return ERR_PTR(-ENOMEM);
494 f2fs_target_device(sbi, blkaddr, bio);
495 bio->bi_end_io = f2fs_read_end_io;
496 bio->bi_private = ctx;
497 bio_set_op_attrs(bio, REQ_OP_READ, 0);
502 /* This can handle encryption stuffs */
503 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
506 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1);
511 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
515 __submit_bio(F2FS_I_SB(inode), bio, DATA);
519 static void __set_data_blkaddr(struct dnode_of_data *dn)
521 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
525 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
526 base = get_extra_isize(dn->inode);
528 /* Get physical address of data block */
529 addr_array = blkaddr_in_node(rn);
530 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
534 * Lock ordering for the change of data block address:
537 * update block addresses in the node page
539 void set_data_blkaddr(struct dnode_of_data *dn)
541 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
542 __set_data_blkaddr(dn);
543 if (set_page_dirty(dn->node_page))
544 dn->node_changed = true;
547 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
549 dn->data_blkaddr = blkaddr;
550 set_data_blkaddr(dn);
551 f2fs_update_extent_cache(dn);
554 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
555 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
557 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
563 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
565 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
568 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
569 dn->ofs_in_node, count);
571 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
573 for (; count > 0; dn->ofs_in_node++) {
574 block_t blkaddr = datablock_addr(dn->inode,
575 dn->node_page, dn->ofs_in_node);
576 if (blkaddr == NULL_ADDR) {
577 dn->data_blkaddr = NEW_ADDR;
578 __set_data_blkaddr(dn);
583 if (set_page_dirty(dn->node_page))
584 dn->node_changed = true;
588 /* Should keep dn->ofs_in_node unchanged */
589 int reserve_new_block(struct dnode_of_data *dn)
591 unsigned int ofs_in_node = dn->ofs_in_node;
594 ret = reserve_new_blocks(dn, 1);
595 dn->ofs_in_node = ofs_in_node;
599 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
601 bool need_put = dn->inode_page ? false : true;
604 err = get_dnode_of_data(dn, index, ALLOC_NODE);
608 if (dn->data_blkaddr == NULL_ADDR)
609 err = reserve_new_block(dn);
615 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
617 struct extent_info ei = {0,0,0};
618 struct inode *inode = dn->inode;
620 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
621 dn->data_blkaddr = ei.blk + index - ei.fofs;
625 return f2fs_reserve_block(dn, index);
628 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
629 int op_flags, bool for_write)
631 struct address_space *mapping = inode->i_mapping;
632 struct dnode_of_data dn;
634 struct extent_info ei = {0,0,0};
637 page = f2fs_grab_cache_page(mapping, index, for_write);
639 return ERR_PTR(-ENOMEM);
641 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
642 dn.data_blkaddr = ei.blk + index - ei.fofs;
646 set_new_dnode(&dn, inode, NULL, NULL, 0);
647 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
652 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
657 if (PageUptodate(page)) {
663 * A new dentry page is allocated but not able to be written, since its
664 * new inode page couldn't be allocated due to -ENOSPC.
665 * In such the case, its blkaddr can be remained as NEW_ADDR.
666 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
668 if (dn.data_blkaddr == NEW_ADDR) {
669 zero_user_segment(page, 0, PAGE_SIZE);
670 if (!PageUptodate(page))
671 SetPageUptodate(page);
676 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
682 f2fs_put_page(page, 1);
686 struct page *find_data_page(struct inode *inode, pgoff_t index)
688 struct address_space *mapping = inode->i_mapping;
691 page = find_get_page(mapping, index);
692 if (page && PageUptodate(page))
694 f2fs_put_page(page, 0);
696 page = get_read_data_page(inode, index, 0, false);
700 if (PageUptodate(page))
703 wait_on_page_locked(page);
704 if (unlikely(!PageUptodate(page))) {
705 f2fs_put_page(page, 0);
706 return ERR_PTR(-EIO);
712 * If it tries to access a hole, return an error.
713 * Because, the callers, functions in dir.c and GC, should be able to know
714 * whether this page exists or not.
716 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
719 struct address_space *mapping = inode->i_mapping;
722 page = get_read_data_page(inode, index, 0, for_write);
726 /* wait for read completion */
728 if (unlikely(page->mapping != mapping)) {
729 f2fs_put_page(page, 1);
732 if (unlikely(!PageUptodate(page))) {
733 f2fs_put_page(page, 1);
734 return ERR_PTR(-EIO);
740 * Caller ensures that this data page is never allocated.
741 * A new zero-filled data page is allocated in the page cache.
743 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
745 * Note that, ipage is set only by make_empty_dir, and if any error occur,
746 * ipage should be released by this function.
748 struct page *get_new_data_page(struct inode *inode,
749 struct page *ipage, pgoff_t index, bool new_i_size)
751 struct address_space *mapping = inode->i_mapping;
753 struct dnode_of_data dn;
756 page = f2fs_grab_cache_page(mapping, index, true);
759 * before exiting, we should make sure ipage will be released
760 * if any error occur.
762 f2fs_put_page(ipage, 1);
763 return ERR_PTR(-ENOMEM);
766 set_new_dnode(&dn, inode, ipage, NULL, 0);
767 err = f2fs_reserve_block(&dn, index);
769 f2fs_put_page(page, 1);
775 if (PageUptodate(page))
778 if (dn.data_blkaddr == NEW_ADDR) {
779 zero_user_segment(page, 0, PAGE_SIZE);
780 if (!PageUptodate(page))
781 SetPageUptodate(page);
783 f2fs_put_page(page, 1);
785 /* if ipage exists, blkaddr should be NEW_ADDR */
786 f2fs_bug_on(F2FS_I_SB(inode), ipage);
787 page = get_lock_data_page(inode, index, true);
792 if (new_i_size && i_size_read(inode) <
793 ((loff_t)(index + 1) << PAGE_SHIFT))
794 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
798 static int __allocate_data_block(struct dnode_of_data *dn)
800 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
801 struct f2fs_summary sum;
807 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
810 dn->data_blkaddr = datablock_addr(dn->inode,
811 dn->node_page, dn->ofs_in_node);
812 if (dn->data_blkaddr == NEW_ADDR)
815 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
819 get_node_info(sbi, dn->nid, &ni);
820 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
822 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
823 &sum, CURSEG_WARM_DATA, NULL, false);
824 set_data_blkaddr(dn);
827 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
829 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
830 f2fs_i_size_write(dn->inode,
831 ((loff_t)(fofs + 1) << PAGE_SHIFT));
835 static inline bool __force_buffered_io(struct inode *inode, int rw)
837 return (f2fs_encrypted_file(inode) ||
838 (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
839 F2FS_I_SB(inode)->s_ndevs);
842 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
844 struct inode *inode = file_inode(iocb->ki_filp);
845 struct f2fs_map_blocks map;
848 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
851 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
852 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
853 if (map.m_len > map.m_lblk)
854 map.m_len -= map.m_lblk;
858 map.m_next_pgofs = NULL;
860 if (iocb->ki_flags & IOCB_DIRECT) {
861 err = f2fs_convert_inline_inode(inode);
864 return f2fs_map_blocks(inode, &map, 1,
865 __force_buffered_io(inode, WRITE) ?
866 F2FS_GET_BLOCK_PRE_AIO :
867 F2FS_GET_BLOCK_PRE_DIO);
869 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
870 err = f2fs_convert_inline_inode(inode);
874 if (!f2fs_has_inline_data(inode))
875 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
879 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
881 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
883 down_read(&sbi->node_change);
885 up_read(&sbi->node_change);
895 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
896 * f2fs_map_blocks structure.
897 * If original data blocks are allocated, then give them to blockdev.
899 * a. preallocate requested block addresses
900 * b. do not use extent cache for better performance
901 * c. give the block addresses to blockdev
903 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
904 int create, int flag)
906 unsigned int maxblocks = map->m_len;
907 struct dnode_of_data dn;
908 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
909 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
910 pgoff_t pgofs, end_offset, end;
911 int err = 0, ofs = 1;
912 unsigned int ofs_in_node, last_ofs_in_node;
914 struct extent_info ei = {0,0,0};
923 /* it only supports block size == page size */
924 pgofs = (pgoff_t)map->m_lblk;
925 end = pgofs + maxblocks;
927 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
928 map->m_pblk = ei.blk + pgofs - ei.fofs;
929 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
930 map->m_flags = F2FS_MAP_MAPPED;
936 __do_map_lock(sbi, flag, true);
938 /* When reading holes, we need its node page */
939 set_new_dnode(&dn, inode, NULL, NULL, 0);
940 err = get_dnode_of_data(&dn, pgofs, mode);
942 if (flag == F2FS_GET_BLOCK_BMAP)
944 if (err == -ENOENT) {
946 if (map->m_next_pgofs)
948 get_next_page_offset(&dn, pgofs);
954 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
955 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
958 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
960 if (__is_valid_data_blkaddr(blkaddr) &&
961 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
966 if (!is_valid_data_blkaddr(sbi, blkaddr)) {
968 if (unlikely(f2fs_cp_error(sbi))) {
972 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
973 if (blkaddr == NULL_ADDR) {
975 last_ofs_in_node = dn.ofs_in_node;
978 err = __allocate_data_block(&dn);
980 set_inode_flag(inode, FI_APPEND_WRITE);
984 map->m_flags |= F2FS_MAP_NEW;
985 blkaddr = dn.data_blkaddr;
987 if (flag == F2FS_GET_BLOCK_BMAP) {
991 if (flag == F2FS_GET_BLOCK_FIEMAP &&
992 blkaddr == NULL_ADDR) {
993 if (map->m_next_pgofs)
994 *map->m_next_pgofs = pgofs + 1;
996 if (flag != F2FS_GET_BLOCK_FIEMAP ||
1002 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1005 if (map->m_len == 0) {
1006 /* preallocated unwritten block should be mapped for fiemap. */
1007 if (blkaddr == NEW_ADDR)
1008 map->m_flags |= F2FS_MAP_UNWRITTEN;
1009 map->m_flags |= F2FS_MAP_MAPPED;
1011 map->m_pblk = blkaddr;
1013 } else if ((map->m_pblk != NEW_ADDR &&
1014 blkaddr == (map->m_pblk + ofs)) ||
1015 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1016 flag == F2FS_GET_BLOCK_PRE_DIO) {
1027 /* preallocate blocks in batch for one dnode page */
1028 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1029 (pgofs == end || dn.ofs_in_node == end_offset)) {
1031 dn.ofs_in_node = ofs_in_node;
1032 err = reserve_new_blocks(&dn, prealloc);
1036 map->m_len += dn.ofs_in_node - ofs_in_node;
1037 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1041 dn.ofs_in_node = end_offset;
1046 else if (dn.ofs_in_node < end_offset)
1049 f2fs_put_dnode(&dn);
1052 __do_map_lock(sbi, flag, false);
1053 f2fs_balance_fs(sbi, dn.node_changed);
1058 f2fs_put_dnode(&dn);
1061 __do_map_lock(sbi, flag, false);
1062 f2fs_balance_fs(sbi, dn.node_changed);
1065 trace_f2fs_map_blocks(inode, map, err);
1069 static int __get_data_block(struct inode *inode, sector_t iblock,
1070 struct buffer_head *bh, int create, int flag,
1071 pgoff_t *next_pgofs)
1073 struct f2fs_map_blocks map;
1076 map.m_lblk = iblock;
1077 map.m_len = bh->b_size >> inode->i_blkbits;
1078 map.m_next_pgofs = next_pgofs;
1080 err = f2fs_map_blocks(inode, &map, create, flag);
1082 map_bh(bh, inode->i_sb, map.m_pblk);
1083 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1084 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1089 static int get_data_block(struct inode *inode, sector_t iblock,
1090 struct buffer_head *bh_result, int create, int flag,
1091 pgoff_t *next_pgofs)
1093 return __get_data_block(inode, iblock, bh_result, create,
1097 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1098 struct buffer_head *bh_result, int create)
1100 return __get_data_block(inode, iblock, bh_result, create,
1101 F2FS_GET_BLOCK_DEFAULT, NULL);
1104 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1105 struct buffer_head *bh_result, int create)
1107 /* Block number less than F2FS MAX BLOCKS */
1108 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1111 return __get_data_block(inode, iblock, bh_result, create,
1112 F2FS_GET_BLOCK_BMAP, NULL);
1115 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1117 return (offset >> inode->i_blkbits);
1120 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1122 return (blk << inode->i_blkbits);
1125 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1128 struct buffer_head map_bh;
1129 sector_t start_blk, last_blk;
1131 u64 logical = 0, phys = 0, size = 0;
1135 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1139 if (f2fs_has_inline_data(inode)) {
1140 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1147 if (logical_to_blk(inode, len) == 0)
1148 len = blk_to_logical(inode, 1);
1150 start_blk = logical_to_blk(inode, start);
1151 last_blk = logical_to_blk(inode, start + len - 1);
1154 memset(&map_bh, 0, sizeof(struct buffer_head));
1155 map_bh.b_size = len;
1157 ret = get_data_block(inode, start_blk, &map_bh, 0,
1158 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1163 if (!buffer_mapped(&map_bh)) {
1164 start_blk = next_pgofs;
1166 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1167 F2FS_I_SB(inode)->max_file_blocks))
1170 flags |= FIEMAP_EXTENT_LAST;
1174 if (f2fs_encrypted_inode(inode))
1175 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1177 ret = fiemap_fill_next_extent(fieinfo, logical,
1181 if (start_blk > last_blk || ret)
1184 logical = blk_to_logical(inode, start_blk);
1185 phys = blk_to_logical(inode, map_bh.b_blocknr);
1186 size = map_bh.b_size;
1188 if (buffer_unwritten(&map_bh))
1189 flags = FIEMAP_EXTENT_UNWRITTEN;
1191 start_blk += logical_to_blk(inode, size);
1195 if (fatal_signal_pending(current))
1203 inode_unlock(inode);
1208 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1209 * Major change was from block_size == page_size in f2fs by default.
1211 static int f2fs_mpage_readpages(struct address_space *mapping,
1212 struct list_head *pages, struct page *page,
1215 struct bio *bio = NULL;
1217 sector_t last_block_in_bio = 0;
1218 struct inode *inode = mapping->host;
1219 const unsigned blkbits = inode->i_blkbits;
1220 const unsigned blocksize = 1 << blkbits;
1221 sector_t block_in_file;
1222 sector_t last_block;
1223 sector_t last_block_in_file;
1225 struct f2fs_map_blocks map;
1231 map.m_next_pgofs = NULL;
1233 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1236 page = list_last_entry(pages, struct page, lru);
1238 prefetchw(&page->flags);
1239 list_del(&page->lru);
1240 if (add_to_page_cache_lru(page, mapping,
1242 readahead_gfp_mask(mapping)))
1246 block_in_file = (sector_t)page->index;
1247 last_block = block_in_file + nr_pages;
1248 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1250 if (last_block > last_block_in_file)
1251 last_block = last_block_in_file;
1254 * Map blocks using the previous result first.
1256 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1257 block_in_file > map.m_lblk &&
1258 block_in_file < (map.m_lblk + map.m_len))
1262 * Then do more f2fs_map_blocks() calls until we are
1263 * done with this page.
1267 if (block_in_file < last_block) {
1268 map.m_lblk = block_in_file;
1269 map.m_len = last_block - block_in_file;
1271 if (f2fs_map_blocks(inode, &map, 0,
1272 F2FS_GET_BLOCK_DEFAULT))
1273 goto set_error_page;
1276 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1277 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1278 SetPageMappedToDisk(page);
1280 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1281 SetPageUptodate(page);
1285 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1287 goto set_error_page;
1289 zero_user_segment(page, 0, PAGE_SIZE);
1290 if (!PageUptodate(page))
1291 SetPageUptodate(page);
1297 * This page will go to BIO. Do we need to send this
1300 if (bio && (last_block_in_bio != block_nr - 1 ||
1301 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1303 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1307 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages);
1310 goto set_error_page;
1314 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1315 goto submit_and_realloc;
1317 last_block_in_bio = block_nr;
1321 zero_user_segment(page, 0, PAGE_SIZE);
1326 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1334 BUG_ON(pages && !list_empty(pages));
1336 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1340 static int f2fs_read_data_page(struct file *file, struct page *page)
1342 struct inode *inode = page->mapping->host;
1345 trace_f2fs_readpage(page, DATA);
1347 /* If the file has inline data, try to read it directly */
1348 if (f2fs_has_inline_data(inode))
1349 ret = f2fs_read_inline_data(inode, page);
1351 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1355 static int f2fs_read_data_pages(struct file *file,
1356 struct address_space *mapping,
1357 struct list_head *pages, unsigned nr_pages)
1359 struct inode *inode = file->f_mapping->host;
1360 struct page *page = list_last_entry(pages, struct page, lru);
1362 trace_f2fs_readpages(inode, page, nr_pages);
1364 /* If the file has inline data, skip readpages */
1365 if (f2fs_has_inline_data(inode))
1368 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1371 static int encrypt_one_page(struct f2fs_io_info *fio)
1373 struct inode *inode = fio->page->mapping->host;
1374 gfp_t gfp_flags = GFP_NOFS;
1376 if (!f2fs_encrypted_file(inode))
1379 /* wait for GCed encrypted page writeback */
1380 f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1383 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1384 PAGE_SIZE, 0, fio->page->index, gfp_flags);
1385 if (!IS_ERR(fio->encrypted_page))
1388 /* flush pending IOs and wait for a while in the ENOMEM case */
1389 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1390 f2fs_flush_merged_writes(fio->sbi);
1391 congestion_wait(BLK_RW_ASYNC, HZ/50);
1392 gfp_flags |= __GFP_NOFAIL;
1395 return PTR_ERR(fio->encrypted_page);
1398 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1400 struct inode *inode = fio->page->mapping->host;
1402 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
1404 if (is_cold_data(fio->page))
1406 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1409 return need_inplace_update_policy(inode, fio);
1412 int do_write_data_page(struct f2fs_io_info *fio)
1414 struct page *page = fio->page;
1415 struct inode *inode = page->mapping->host;
1416 struct dnode_of_data dn;
1417 struct extent_info ei = {0,0,0};
1418 bool ipu_force = false;
1421 set_new_dnode(&dn, inode, NULL, NULL, 0);
1422 if (need_inplace_update(fio) &&
1423 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1424 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1426 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1428 return -EFSCORRUPTED;
1431 fio->need_lock = LOCK_DONE;
1435 /* Deadlock due to between page->lock and f2fs_lock_op */
1436 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1439 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1443 fio->old_blkaddr = dn.data_blkaddr;
1445 /* This page is already truncated */
1446 if (fio->old_blkaddr == NULL_ADDR) {
1447 ClearPageUptodate(page);
1448 clear_cold_data(page);
1452 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1453 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1455 err = -EFSCORRUPTED;
1459 * If current allocation needs SSR,
1460 * it had better in-place writes for updated data.
1462 if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1463 need_inplace_update(fio))) {
1464 err = encrypt_one_page(fio);
1468 set_page_writeback(page);
1469 f2fs_put_dnode(&dn);
1470 if (fio->need_lock == LOCK_REQ)
1471 f2fs_unlock_op(fio->sbi);
1472 err = rewrite_data_page(fio);
1473 trace_f2fs_do_write_data_page(fio->page, IPU);
1474 set_inode_flag(inode, FI_UPDATE_WRITE);
1478 if (fio->need_lock == LOCK_RETRY) {
1479 if (!f2fs_trylock_op(fio->sbi)) {
1483 fio->need_lock = LOCK_REQ;
1486 err = encrypt_one_page(fio);
1490 set_page_writeback(page);
1492 /* LFS mode write path */
1493 write_data_page(&dn, fio);
1494 trace_f2fs_do_write_data_page(page, OPU);
1495 set_inode_flag(inode, FI_APPEND_WRITE);
1496 if (page->index == 0)
1497 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1499 f2fs_put_dnode(&dn);
1501 if (fio->need_lock == LOCK_REQ)
1502 f2fs_unlock_op(fio->sbi);
1506 static int __write_data_page(struct page *page, bool *submitted,
1507 struct writeback_control *wbc,
1508 enum iostat_type io_type)
1510 struct inode *inode = page->mapping->host;
1511 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1512 loff_t i_size = i_size_read(inode);
1513 const pgoff_t end_index = ((unsigned long long) i_size)
1515 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
1516 unsigned offset = 0;
1517 bool need_balance_fs = false;
1519 struct f2fs_io_info fio = {
1523 .op_flags = wbc_to_write_flags(wbc),
1524 .old_blkaddr = NULL_ADDR,
1526 .encrypted_page = NULL,
1528 .need_lock = LOCK_RETRY,
1532 trace_f2fs_writepage(page, DATA);
1534 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1537 if (page->index < end_index)
1541 * If the offset is out-of-range of file size,
1542 * this page does not have to be written to disk.
1544 offset = i_size & (PAGE_SIZE - 1);
1545 if ((page->index >= end_index + 1) || !offset)
1548 zero_user_segment(page, offset, PAGE_SIZE);
1550 if (f2fs_is_drop_cache(inode))
1552 /* we should not write 0'th page having journal header */
1553 if (f2fs_is_volatile_file(inode) && (!page->index ||
1554 (!wbc->for_reclaim &&
1555 available_free_memory(sbi, BASE_CHECK))))
1558 /* we should bypass data pages to proceed the kworkder jobs */
1559 if (unlikely(f2fs_cp_error(sbi))) {
1560 mapping_set_error(page->mapping, -EIO);
1564 /* Dentry blocks are controlled by checkpoint */
1565 if (S_ISDIR(inode->i_mode)) {
1566 fio.need_lock = LOCK_DONE;
1567 err = do_write_data_page(&fio);
1571 if (!wbc->for_reclaim)
1572 need_balance_fs = true;
1573 else if (has_not_enough_free_secs(sbi, 0, 0))
1576 set_inode_flag(inode, FI_HOT_DATA);
1579 if (f2fs_has_inline_data(inode)) {
1580 err = f2fs_write_inline_data(inode, page);
1585 if (err == -EAGAIN) {
1586 err = do_write_data_page(&fio);
1587 if (err == -EAGAIN) {
1588 fio.need_lock = LOCK_REQ;
1589 err = do_write_data_page(&fio);
1592 if (F2FS_I(inode)->last_disk_size < psize)
1593 F2FS_I(inode)->last_disk_size = psize;
1596 if (err && err != -ENOENT)
1600 inode_dec_dirty_pages(inode);
1602 ClearPageUptodate(page);
1603 clear_cold_data(page);
1606 if (wbc->for_reclaim) {
1607 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1608 clear_inode_flag(inode, FI_HOT_DATA);
1609 remove_dirty_inode(inode);
1614 if (!S_ISDIR(inode->i_mode))
1615 f2fs_balance_fs(sbi, need_balance_fs);
1617 if (unlikely(f2fs_cp_error(sbi))) {
1618 f2fs_submit_merged_write(sbi, DATA);
1623 *submitted = fio.submitted;
1628 redirty_page_for_writepage(wbc, page);
1630 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
1631 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
1632 * file_write_and_wait_range() will see EIO error, which is critical
1633 * to return value of fsync() followed by atomic_write failure to user.
1635 if (!err || wbc->for_reclaim)
1636 return AOP_WRITEPAGE_ACTIVATE;
1641 static int f2fs_write_data_page(struct page *page,
1642 struct writeback_control *wbc)
1644 return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1648 * This function was copied from write_cche_pages from mm/page-writeback.c.
1649 * The major change is making write step of cold data page separately from
1650 * warm/hot data page.
1652 static int f2fs_write_cache_pages(struct address_space *mapping,
1653 struct writeback_control *wbc,
1654 enum iostat_type io_type)
1658 struct pagevec pvec;
1660 pgoff_t uninitialized_var(writeback_index);
1662 pgoff_t end; /* Inclusive */
1664 pgoff_t last_idx = ULONG_MAX;
1666 int range_whole = 0;
1669 pagevec_init(&pvec, 0);
1671 if (get_dirty_pages(mapping->host) <=
1672 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1673 set_inode_flag(mapping->host, FI_HOT_DATA);
1675 clear_inode_flag(mapping->host, FI_HOT_DATA);
1677 if (wbc->range_cyclic) {
1678 writeback_index = mapping->writeback_index; /* prev offset */
1679 index = writeback_index;
1686 index = wbc->range_start >> PAGE_SHIFT;
1687 end = wbc->range_end >> PAGE_SHIFT;
1688 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1690 cycled = 1; /* ignore range_cyclic tests */
1692 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1693 tag = PAGECACHE_TAG_TOWRITE;
1695 tag = PAGECACHE_TAG_DIRTY;
1697 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1698 tag_pages_for_writeback(mapping, index, end);
1700 while (!done && (index <= end)) {
1703 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1704 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1708 for (i = 0; i < nr_pages; i++) {
1709 struct page *page = pvec.pages[i];
1710 bool submitted = false;
1712 if (page->index > end) {
1717 done_index = page->index;
1721 if (unlikely(page->mapping != mapping)) {
1727 if (!PageDirty(page)) {
1728 /* someone wrote it for us */
1729 goto continue_unlock;
1732 if (PageWriteback(page)) {
1733 if (wbc->sync_mode != WB_SYNC_NONE)
1734 f2fs_wait_on_page_writeback(page,
1737 goto continue_unlock;
1740 BUG_ON(PageWriteback(page));
1741 if (!clear_page_dirty_for_io(page))
1742 goto continue_unlock;
1744 ret = __write_data_page(page, &submitted, wbc, io_type);
1745 if (unlikely(ret)) {
1747 * keep nr_to_write, since vfs uses this to
1748 * get # of written pages.
1750 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1754 } else if (ret == -EAGAIN) {
1756 if (wbc->sync_mode == WB_SYNC_ALL) {
1758 congestion_wait(BLK_RW_ASYNC,
1764 done_index = page->index + 1;
1767 } else if (submitted) {
1768 last_idx = page->index;
1771 /* give a priority to WB_SYNC threads */
1772 if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) ||
1773 --wbc->nr_to_write <= 0) &&
1774 wbc->sync_mode == WB_SYNC_NONE) {
1779 pagevec_release(&pvec);
1783 if (!cycled && !done) {
1786 end = writeback_index - 1;
1789 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1790 mapping->writeback_index = done_index;
1792 if (last_idx != ULONG_MAX)
1793 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
1799 int __f2fs_write_data_pages(struct address_space *mapping,
1800 struct writeback_control *wbc,
1801 enum iostat_type io_type)
1803 struct inode *inode = mapping->host;
1804 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1805 struct blk_plug plug;
1808 /* deal with chardevs and other special file */
1809 if (!mapping->a_ops->writepage)
1812 /* skip writing if there is no dirty page in this inode */
1813 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1816 /* during POR, we don't need to trigger writepage at all. */
1817 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1820 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1821 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1822 available_free_memory(sbi, DIRTY_DENTS))
1825 /* skip writing during file defragment */
1826 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1829 trace_f2fs_writepages(mapping->host, wbc, DATA);
1831 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
1832 if (wbc->sync_mode == WB_SYNC_ALL)
1833 atomic_inc(&sbi->wb_sync_req);
1834 else if (atomic_read(&sbi->wb_sync_req))
1837 blk_start_plug(&plug);
1838 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
1839 blk_finish_plug(&plug);
1841 if (wbc->sync_mode == WB_SYNC_ALL)
1842 atomic_dec(&sbi->wb_sync_req);
1844 * if some pages were truncated, we cannot guarantee its mapping->host
1845 * to detect pending bios.
1848 remove_dirty_inode(inode);
1852 wbc->pages_skipped += get_dirty_pages(inode);
1853 trace_f2fs_writepages(mapping->host, wbc, DATA);
1857 static int f2fs_write_data_pages(struct address_space *mapping,
1858 struct writeback_control *wbc)
1860 struct inode *inode = mapping->host;
1862 return __f2fs_write_data_pages(mapping, wbc,
1863 F2FS_I(inode)->cp_task == current ?
1864 FS_CP_DATA_IO : FS_DATA_IO);
1867 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1869 struct inode *inode = mapping->host;
1870 loff_t i_size = i_size_read(inode);
1873 down_write(&F2FS_I(inode)->i_mmap_sem);
1874 truncate_pagecache(inode, i_size);
1875 truncate_blocks(inode, i_size, true);
1876 up_write(&F2FS_I(inode)->i_mmap_sem);
1880 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1881 struct page *page, loff_t pos, unsigned len,
1882 block_t *blk_addr, bool *node_changed)
1884 struct inode *inode = page->mapping->host;
1885 pgoff_t index = page->index;
1886 struct dnode_of_data dn;
1888 bool locked = false;
1889 struct extent_info ei = {0,0,0};
1894 * we already allocated all the blocks, so we don't need to get
1895 * the block addresses when there is no need to fill the page.
1897 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1898 !is_inode_flag_set(inode, FI_NO_PREALLOC))
1901 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
1902 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
1903 flag = F2FS_GET_BLOCK_DEFAULT;
1905 flag = F2FS_GET_BLOCK_PRE_AIO;
1907 if (f2fs_has_inline_data(inode) ||
1908 (pos & PAGE_MASK) >= i_size_read(inode)) {
1909 __do_map_lock(sbi, flag, true);
1913 /* check inline_data */
1914 ipage = get_node_page(sbi, inode->i_ino);
1915 if (IS_ERR(ipage)) {
1916 err = PTR_ERR(ipage);
1920 set_new_dnode(&dn, inode, ipage, ipage, 0);
1922 if (f2fs_has_inline_data(inode)) {
1923 if (pos + len <= MAX_INLINE_DATA(inode)) {
1924 read_inline_data(page, ipage);
1925 set_inode_flag(inode, FI_DATA_EXIST);
1927 set_inline_node(ipage);
1929 err = f2fs_convert_inline_page(&dn, page);
1932 if (dn.data_blkaddr == NULL_ADDR)
1933 err = f2fs_get_block(&dn, index);
1935 } else if (locked) {
1936 err = f2fs_get_block(&dn, index);
1938 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1939 dn.data_blkaddr = ei.blk + index - ei.fofs;
1942 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1943 if (err || dn.data_blkaddr == NULL_ADDR) {
1944 f2fs_put_dnode(&dn);
1945 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
1947 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
1954 /* convert_inline_page can make node_changed */
1955 *blk_addr = dn.data_blkaddr;
1956 *node_changed = dn.node_changed;
1958 f2fs_put_dnode(&dn);
1961 __do_map_lock(sbi, flag, false);
1965 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1966 loff_t pos, unsigned len, unsigned flags,
1967 struct page **pagep, void **fsdata)
1969 struct inode *inode = mapping->host;
1970 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1971 struct page *page = NULL;
1972 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1973 bool need_balance = false;
1974 block_t blkaddr = NULL_ADDR;
1977 trace_f2fs_write_begin(inode, pos, len, flags);
1980 * We should check this at this moment to avoid deadlock on inode page
1981 * and #0 page. The locking rule for inline_data conversion should be:
1982 * lock_page(page #0) -> lock_page(inode_page)
1985 err = f2fs_convert_inline_inode(inode);
1991 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1992 * wait_for_stable_page. Will wait that below with our IO control.
1994 page = pagecache_get_page(mapping, index,
1995 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2003 err = prepare_write_begin(sbi, page, pos, len,
2004 &blkaddr, &need_balance);
2008 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
2010 f2fs_balance_fs(sbi, true);
2012 if (page->mapping != mapping) {
2013 /* The page got truncated from under us */
2014 f2fs_put_page(page, 1);
2019 f2fs_wait_on_page_writeback(page, DATA, false);
2021 /* wait for GCed encrypted page writeback */
2022 if (f2fs_encrypted_file(inode))
2023 f2fs_wait_on_block_writeback(sbi, blkaddr);
2025 if (len == PAGE_SIZE || PageUptodate(page))
2028 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2029 zero_user_segment(page, len, PAGE_SIZE);
2033 if (blkaddr == NEW_ADDR) {
2034 zero_user_segment(page, 0, PAGE_SIZE);
2035 SetPageUptodate(page);
2037 err = f2fs_submit_page_read(inode, page, blkaddr);
2042 if (unlikely(page->mapping != mapping)) {
2043 f2fs_put_page(page, 1);
2046 if (unlikely(!PageUptodate(page))) {
2054 f2fs_put_page(page, 1);
2055 f2fs_write_failed(mapping, pos + len);
2059 static int f2fs_write_end(struct file *file,
2060 struct address_space *mapping,
2061 loff_t pos, unsigned len, unsigned copied,
2062 struct page *page, void *fsdata)
2064 struct inode *inode = page->mapping->host;
2066 trace_f2fs_write_end(inode, pos, len, copied);
2069 * This should be come from len == PAGE_SIZE, and we expect copied
2070 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2071 * let generic_perform_write() try to copy data again through copied=0.
2073 if (!PageUptodate(page)) {
2074 if (unlikely(copied != len))
2077 SetPageUptodate(page);
2082 set_page_dirty(page);
2084 if (pos + copied > i_size_read(inode))
2085 f2fs_i_size_write(inode, pos + copied);
2087 f2fs_put_page(page, 1);
2088 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2092 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2095 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
2097 if (offset & blocksize_mask)
2100 if (iov_iter_alignment(iter) & blocksize_mask)
2106 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2108 struct address_space *mapping = iocb->ki_filp->f_mapping;
2109 struct inode *inode = mapping->host;
2110 size_t count = iov_iter_count(iter);
2111 loff_t offset = iocb->ki_pos;
2112 int rw = iov_iter_rw(iter);
2115 err = check_direct_IO(inode, iter, offset);
2119 if (__force_buffered_io(inode, rw))
2122 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2124 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
2125 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2126 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
2130 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2132 set_inode_flag(inode, FI_UPDATE_WRITE);
2133 } else if (err < 0) {
2134 f2fs_write_failed(mapping, offset + count);
2138 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2143 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2144 unsigned int length)
2146 struct inode *inode = page->mapping->host;
2147 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2149 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2150 (offset % PAGE_SIZE || length != PAGE_SIZE))
2153 if (PageDirty(page)) {
2154 if (inode->i_ino == F2FS_META_INO(sbi)) {
2155 dec_page_count(sbi, F2FS_DIRTY_META);
2156 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2157 dec_page_count(sbi, F2FS_DIRTY_NODES);
2159 inode_dec_dirty_pages(inode);
2160 remove_dirty_inode(inode);
2164 clear_cold_data(page);
2166 /* This is atomic written page, keep Private */
2167 if (IS_ATOMIC_WRITTEN_PAGE(page))
2168 return drop_inmem_page(inode, page);
2170 set_page_private(page, 0);
2171 ClearPagePrivate(page);
2174 int f2fs_release_page(struct page *page, gfp_t wait)
2176 /* If this is dirty page, keep PagePrivate */
2177 if (PageDirty(page))
2180 /* This is atomic written page, keep Private */
2181 if (IS_ATOMIC_WRITTEN_PAGE(page))
2184 clear_cold_data(page);
2185 set_page_private(page, 0);
2186 ClearPagePrivate(page);
2191 * This was copied from __set_page_dirty_buffers which gives higher performance
2192 * in very high speed storages. (e.g., pmem)
2194 void f2fs_set_page_dirty_nobuffers(struct page *page)
2196 struct address_space *mapping = page->mapping;
2197 unsigned long flags;
2199 if (unlikely(!mapping))
2202 spin_lock(&mapping->private_lock);
2203 lock_page_memcg(page);
2205 spin_unlock(&mapping->private_lock);
2207 spin_lock_irqsave(&mapping->tree_lock, flags);
2208 WARN_ON_ONCE(!PageUptodate(page));
2209 account_page_dirtied(page, mapping);
2210 radix_tree_tag_set(&mapping->page_tree,
2211 page_index(page), PAGECACHE_TAG_DIRTY);
2212 spin_unlock_irqrestore(&mapping->tree_lock, flags);
2213 unlock_page_memcg(page);
2215 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2219 static int f2fs_set_data_page_dirty(struct page *page)
2221 struct address_space *mapping = page->mapping;
2222 struct inode *inode = mapping->host;
2224 trace_f2fs_set_page_dirty(page, DATA);
2226 if (!PageUptodate(page))
2227 SetPageUptodate(page);
2229 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2230 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2231 register_inmem_page(inode, page);
2235 * Previously, this page has been registered, we just
2241 if (!PageDirty(page)) {
2242 f2fs_set_page_dirty_nobuffers(page);
2243 update_dirty_page(inode, page);
2249 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2251 struct inode *inode = mapping->host;
2253 if (f2fs_has_inline_data(inode))
2256 /* make sure allocating whole blocks */
2257 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2258 filemap_write_and_wait(mapping);
2260 return generic_block_bmap(mapping, block, get_data_block_bmap);
2263 #ifdef CONFIG_MIGRATION
2264 #include <linux/migrate.h>
2266 int f2fs_migrate_page(struct address_space *mapping,
2267 struct page *newpage, struct page *page, enum migrate_mode mode)
2269 int rc, extra_count;
2270 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2271 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2273 BUG_ON(PageWriteback(page));
2275 /* migrating an atomic written page is safe with the inmem_lock hold */
2276 if (atomic_written) {
2277 if (mode != MIGRATE_SYNC)
2279 if (!mutex_trylock(&fi->inmem_lock))
2284 * A reference is expected if PagePrivate set when move mapping,
2285 * however F2FS breaks this for maintaining dirty page counts when
2286 * truncating pages. So here adjusting the 'extra_count' make it work.
2288 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2289 rc = migrate_page_move_mapping(mapping, newpage,
2290 page, NULL, mode, extra_count);
2291 if (rc != MIGRATEPAGE_SUCCESS) {
2293 mutex_unlock(&fi->inmem_lock);
2297 if (atomic_written) {
2298 struct inmem_pages *cur;
2299 list_for_each_entry(cur, &fi->inmem_pages, list)
2300 if (cur->page == page) {
2301 cur->page = newpage;
2304 mutex_unlock(&fi->inmem_lock);
2309 if (PagePrivate(page))
2310 SetPagePrivate(newpage);
2311 set_page_private(newpage, page_private(page));
2313 if (mode != MIGRATE_SYNC_NO_COPY)
2314 migrate_page_copy(newpage, page);
2316 migrate_page_states(newpage, page);
2318 return MIGRATEPAGE_SUCCESS;
2322 const struct address_space_operations f2fs_dblock_aops = {
2323 .readpage = f2fs_read_data_page,
2324 .readpages = f2fs_read_data_pages,
2325 .writepage = f2fs_write_data_page,
2326 .writepages = f2fs_write_data_pages,
2327 .write_begin = f2fs_write_begin,
2328 .write_end = f2fs_write_end,
2329 .set_page_dirty = f2fs_set_data_page_dirty,
2330 .invalidatepage = f2fs_invalidate_page,
2331 .releasepage = f2fs_release_page,
2332 .direct_IO = f2fs_direct_IO,
2334 #ifdef CONFIG_MIGRATION
2335 .migratepage = f2fs_migrate_page,