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 = 1 as a return value */
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 void __set_data_blkaddr(struct dnode_of_data *dn)
462 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
465 /* Get physical address of data block */
466 addr_array = blkaddr_in_node(rn);
467 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
471 * Lock ordering for the change of data block address:
474 * update block addresses in the node page
476 void set_data_blkaddr(struct dnode_of_data *dn)
478 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
479 __set_data_blkaddr(dn);
480 if (set_page_dirty(dn->node_page))
481 dn->node_changed = true;
484 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
486 dn->data_blkaddr = blkaddr;
487 set_data_blkaddr(dn);
488 f2fs_update_extent_cache(dn);
491 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
492 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
494 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
500 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
502 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
505 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
506 dn->ofs_in_node, count);
508 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
510 for (; count > 0; dn->ofs_in_node++) {
512 datablock_addr(dn->node_page, dn->ofs_in_node);
513 if (blkaddr == NULL_ADDR) {
514 dn->data_blkaddr = NEW_ADDR;
515 __set_data_blkaddr(dn);
520 if (set_page_dirty(dn->node_page))
521 dn->node_changed = true;
525 /* Should keep dn->ofs_in_node unchanged */
526 int reserve_new_block(struct dnode_of_data *dn)
528 unsigned int ofs_in_node = dn->ofs_in_node;
531 ret = reserve_new_blocks(dn, 1);
532 dn->ofs_in_node = ofs_in_node;
536 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
538 bool need_put = dn->inode_page ? false : true;
541 err = get_dnode_of_data(dn, index, ALLOC_NODE);
545 if (dn->data_blkaddr == NULL_ADDR)
546 err = reserve_new_block(dn);
552 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
554 struct extent_info ei = {0,0,0};
555 struct inode *inode = dn->inode;
557 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
558 dn->data_blkaddr = ei.blk + index - ei.fofs;
562 return f2fs_reserve_block(dn, index);
565 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
566 int op_flags, bool for_write)
568 struct address_space *mapping = inode->i_mapping;
569 struct dnode_of_data dn;
571 struct extent_info ei = {0,0,0};
573 struct f2fs_io_info fio = {
574 .sbi = F2FS_I_SB(inode),
577 .op_flags = op_flags,
578 .encrypted_page = NULL,
581 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
582 return read_mapping_page(mapping, index, NULL);
584 page = f2fs_grab_cache_page(mapping, index, for_write);
586 return ERR_PTR(-ENOMEM);
588 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
589 dn.data_blkaddr = ei.blk + index - ei.fofs;
593 set_new_dnode(&dn, inode, NULL, NULL, 0);
594 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
599 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
604 if (PageUptodate(page)) {
610 * A new dentry page is allocated but not able to be written, since its
611 * new inode page couldn't be allocated due to -ENOSPC.
612 * In such the case, its blkaddr can be remained as NEW_ADDR.
613 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
615 if (dn.data_blkaddr == NEW_ADDR) {
616 zero_user_segment(page, 0, PAGE_SIZE);
617 if (!PageUptodate(page))
618 SetPageUptodate(page);
623 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
625 err = f2fs_submit_page_bio(&fio);
631 f2fs_put_page(page, 1);
635 struct page *find_data_page(struct inode *inode, pgoff_t index)
637 struct address_space *mapping = inode->i_mapping;
640 page = find_get_page(mapping, index);
641 if (page && PageUptodate(page))
643 f2fs_put_page(page, 0);
645 page = get_read_data_page(inode, index, 0, false);
649 if (PageUptodate(page))
652 wait_on_page_locked(page);
653 if (unlikely(!PageUptodate(page))) {
654 f2fs_put_page(page, 0);
655 return ERR_PTR(-EIO);
661 * If it tries to access a hole, return an error.
662 * Because, the callers, functions in dir.c and GC, should be able to know
663 * whether this page exists or not.
665 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
668 struct address_space *mapping = inode->i_mapping;
671 page = get_read_data_page(inode, index, 0, for_write);
675 /* wait for read completion */
677 if (unlikely(page->mapping != mapping)) {
678 f2fs_put_page(page, 1);
681 if (unlikely(!PageUptodate(page))) {
682 f2fs_put_page(page, 1);
683 return ERR_PTR(-EIO);
689 * Caller ensures that this data page is never allocated.
690 * A new zero-filled data page is allocated in the page cache.
692 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
694 * Note that, ipage is set only by make_empty_dir, and if any error occur,
695 * ipage should be released by this function.
697 struct page *get_new_data_page(struct inode *inode,
698 struct page *ipage, pgoff_t index, bool new_i_size)
700 struct address_space *mapping = inode->i_mapping;
702 struct dnode_of_data dn;
705 page = f2fs_grab_cache_page(mapping, index, true);
708 * before exiting, we should make sure ipage will be released
709 * if any error occur.
711 f2fs_put_page(ipage, 1);
712 return ERR_PTR(-ENOMEM);
715 set_new_dnode(&dn, inode, ipage, NULL, 0);
716 err = f2fs_reserve_block(&dn, index);
718 f2fs_put_page(page, 1);
724 if (PageUptodate(page))
727 if (dn.data_blkaddr == NEW_ADDR) {
728 zero_user_segment(page, 0, PAGE_SIZE);
729 if (!PageUptodate(page))
730 SetPageUptodate(page);
732 f2fs_put_page(page, 1);
734 /* if ipage exists, blkaddr should be NEW_ADDR */
735 f2fs_bug_on(F2FS_I_SB(inode), ipage);
736 page = get_lock_data_page(inode, index, true);
741 if (new_i_size && i_size_read(inode) <
742 ((loff_t)(index + 1) << PAGE_SHIFT))
743 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
747 static int __allocate_data_block(struct dnode_of_data *dn)
749 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
750 struct f2fs_summary sum;
756 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
759 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
760 if (dn->data_blkaddr == NEW_ADDR)
763 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
767 get_node_info(sbi, dn->nid, &ni);
768 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
770 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
771 &sum, CURSEG_WARM_DATA, NULL, false);
772 set_data_blkaddr(dn);
775 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
777 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
778 f2fs_i_size_write(dn->inode,
779 ((loff_t)(fofs + 1) << PAGE_SHIFT));
783 static inline bool __force_buffered_io(struct inode *inode, int rw)
785 return ((f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) ||
786 (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
787 F2FS_I_SB(inode)->s_ndevs);
790 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
792 struct inode *inode = file_inode(iocb->ki_filp);
793 struct f2fs_map_blocks map;
796 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
799 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
800 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
801 if (map.m_len > map.m_lblk)
802 map.m_len -= map.m_lblk;
806 map.m_next_pgofs = NULL;
808 if (iocb->ki_flags & IOCB_DIRECT) {
809 err = f2fs_convert_inline_inode(inode);
812 return f2fs_map_blocks(inode, &map, 1,
813 __force_buffered_io(inode, WRITE) ?
814 F2FS_GET_BLOCK_PRE_AIO :
815 F2FS_GET_BLOCK_PRE_DIO);
817 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
818 err = f2fs_convert_inline_inode(inode);
822 if (!f2fs_has_inline_data(inode))
823 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
827 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
829 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
831 down_read(&sbi->node_change);
833 up_read(&sbi->node_change);
843 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
844 * f2fs_map_blocks structure.
845 * If original data blocks are allocated, then give them to blockdev.
847 * a. preallocate requested block addresses
848 * b. do not use extent cache for better performance
849 * c. give the block addresses to blockdev
851 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
852 int create, int flag)
854 unsigned int maxblocks = map->m_len;
855 struct dnode_of_data dn;
856 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
857 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
858 pgoff_t pgofs, end_offset, end;
859 int err = 0, ofs = 1;
860 unsigned int ofs_in_node, last_ofs_in_node;
862 struct extent_info ei = {0,0,0};
871 /* it only supports block size == page size */
872 pgofs = (pgoff_t)map->m_lblk;
873 end = pgofs + maxblocks;
875 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
876 map->m_pblk = ei.blk + pgofs - ei.fofs;
877 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
878 map->m_flags = F2FS_MAP_MAPPED;
884 __do_map_lock(sbi, flag, true);
886 /* When reading holes, we need its node page */
887 set_new_dnode(&dn, inode, NULL, NULL, 0);
888 err = get_dnode_of_data(&dn, pgofs, mode);
890 if (flag == F2FS_GET_BLOCK_BMAP)
892 if (err == -ENOENT) {
894 if (map->m_next_pgofs)
896 get_next_page_offset(&dn, pgofs);
902 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
903 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
906 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
908 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
910 if (unlikely(f2fs_cp_error(sbi))) {
914 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
915 if (blkaddr == NULL_ADDR) {
917 last_ofs_in_node = dn.ofs_in_node;
920 err = __allocate_data_block(&dn);
922 set_inode_flag(inode, FI_APPEND_WRITE);
926 map->m_flags |= F2FS_MAP_NEW;
927 blkaddr = dn.data_blkaddr;
929 if (flag == F2FS_GET_BLOCK_BMAP) {
933 if (flag == F2FS_GET_BLOCK_FIEMAP &&
934 blkaddr == NULL_ADDR) {
935 if (map->m_next_pgofs)
936 *map->m_next_pgofs = pgofs + 1;
938 if (flag != F2FS_GET_BLOCK_FIEMAP ||
944 if (flag == F2FS_GET_BLOCK_PRE_AIO)
947 if (map->m_len == 0) {
948 /* preallocated unwritten block should be mapped for fiemap. */
949 if (blkaddr == NEW_ADDR)
950 map->m_flags |= F2FS_MAP_UNWRITTEN;
951 map->m_flags |= F2FS_MAP_MAPPED;
953 map->m_pblk = blkaddr;
955 } else if ((map->m_pblk != NEW_ADDR &&
956 blkaddr == (map->m_pblk + ofs)) ||
957 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
958 flag == F2FS_GET_BLOCK_PRE_DIO) {
969 /* preallocate blocks in batch for one dnode page */
970 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
971 (pgofs == end || dn.ofs_in_node == end_offset)) {
973 dn.ofs_in_node = ofs_in_node;
974 err = reserve_new_blocks(&dn, prealloc);
978 map->m_len += dn.ofs_in_node - ofs_in_node;
979 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
983 dn.ofs_in_node = end_offset;
988 else if (dn.ofs_in_node < end_offset)
994 __do_map_lock(sbi, flag, false);
995 f2fs_balance_fs(sbi, dn.node_changed);
1000 f2fs_put_dnode(&dn);
1003 __do_map_lock(sbi, flag, false);
1004 f2fs_balance_fs(sbi, dn.node_changed);
1007 trace_f2fs_map_blocks(inode, map, err);
1011 static int __get_data_block(struct inode *inode, sector_t iblock,
1012 struct buffer_head *bh, int create, int flag,
1013 pgoff_t *next_pgofs)
1015 struct f2fs_map_blocks map;
1018 map.m_lblk = iblock;
1019 map.m_len = bh->b_size >> inode->i_blkbits;
1020 map.m_next_pgofs = next_pgofs;
1022 err = f2fs_map_blocks(inode, &map, create, flag);
1024 map_bh(bh, inode->i_sb, map.m_pblk);
1025 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1026 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1031 static int get_data_block(struct inode *inode, sector_t iblock,
1032 struct buffer_head *bh_result, int create, int flag,
1033 pgoff_t *next_pgofs)
1035 return __get_data_block(inode, iblock, bh_result, create,
1039 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1040 struct buffer_head *bh_result, int create)
1042 return __get_data_block(inode, iblock, bh_result, create,
1043 F2FS_GET_BLOCK_DIO, NULL);
1046 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1047 struct buffer_head *bh_result, int create)
1049 /* Block number less than F2FS MAX BLOCKS */
1050 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1053 return __get_data_block(inode, iblock, bh_result, create,
1054 F2FS_GET_BLOCK_BMAP, NULL);
1057 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1059 return (offset >> inode->i_blkbits);
1062 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1064 return (blk << inode->i_blkbits);
1067 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1070 struct buffer_head map_bh;
1071 sector_t start_blk, last_blk;
1073 u64 logical = 0, phys = 0, size = 0;
1077 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1081 if (f2fs_has_inline_data(inode)) {
1082 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1089 if (logical_to_blk(inode, len) == 0)
1090 len = blk_to_logical(inode, 1);
1092 start_blk = logical_to_blk(inode, start);
1093 last_blk = logical_to_blk(inode, start + len - 1);
1096 memset(&map_bh, 0, sizeof(struct buffer_head));
1097 map_bh.b_size = len;
1099 ret = get_data_block(inode, start_blk, &map_bh, 0,
1100 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1105 if (!buffer_mapped(&map_bh)) {
1106 start_blk = next_pgofs;
1108 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1109 F2FS_I_SB(inode)->max_file_blocks))
1112 flags |= FIEMAP_EXTENT_LAST;
1116 if (f2fs_encrypted_inode(inode))
1117 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1119 ret = fiemap_fill_next_extent(fieinfo, logical,
1123 if (start_blk > last_blk || ret)
1126 logical = blk_to_logical(inode, start_blk);
1127 phys = blk_to_logical(inode, map_bh.b_blocknr);
1128 size = map_bh.b_size;
1130 if (buffer_unwritten(&map_bh))
1131 flags = FIEMAP_EXTENT_UNWRITTEN;
1133 start_blk += logical_to_blk(inode, size);
1137 if (fatal_signal_pending(current))
1145 inode_unlock(inode);
1149 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
1152 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1153 struct fscrypt_ctx *ctx = NULL;
1156 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1157 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1159 return ERR_CAST(ctx);
1161 /* wait the page to be moved by cleaning */
1162 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1165 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
1168 fscrypt_release_ctx(ctx);
1169 return ERR_PTR(-ENOMEM);
1171 f2fs_target_device(sbi, blkaddr, bio);
1172 bio->bi_end_io = f2fs_read_end_io;
1173 bio->bi_private = ctx;
1179 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1180 * Major change was from block_size == page_size in f2fs by default.
1182 static int f2fs_mpage_readpages(struct address_space *mapping,
1183 struct list_head *pages, struct page *page,
1186 struct bio *bio = NULL;
1188 sector_t last_block_in_bio = 0;
1189 struct inode *inode = mapping->host;
1190 const unsigned blkbits = inode->i_blkbits;
1191 const unsigned blocksize = 1 << blkbits;
1192 sector_t block_in_file;
1193 sector_t last_block;
1194 sector_t last_block_in_file;
1196 struct f2fs_map_blocks map;
1202 map.m_next_pgofs = NULL;
1204 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1207 page = list_last_entry(pages, struct page, lru);
1209 prefetchw(&page->flags);
1210 list_del(&page->lru);
1211 if (add_to_page_cache_lru(page, mapping,
1213 readahead_gfp_mask(mapping)))
1217 block_in_file = (sector_t)page->index;
1218 last_block = block_in_file + nr_pages;
1219 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1221 if (last_block > last_block_in_file)
1222 last_block = last_block_in_file;
1225 * Map blocks using the previous result first.
1227 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1228 block_in_file > map.m_lblk &&
1229 block_in_file < (map.m_lblk + map.m_len))
1233 * Then do more f2fs_map_blocks() calls until we are
1234 * done with this page.
1238 if (block_in_file < last_block) {
1239 map.m_lblk = block_in_file;
1240 map.m_len = last_block - block_in_file;
1242 if (f2fs_map_blocks(inode, &map, 0,
1243 F2FS_GET_BLOCK_READ))
1244 goto set_error_page;
1247 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1248 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1249 SetPageMappedToDisk(page);
1251 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1252 SetPageUptodate(page);
1256 zero_user_segment(page, 0, PAGE_SIZE);
1257 if (!PageUptodate(page))
1258 SetPageUptodate(page);
1264 * This page will go to BIO. Do we need to send this
1267 if (bio && (last_block_in_bio != block_nr - 1 ||
1268 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1270 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1274 bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1277 goto set_error_page;
1279 bio_set_op_attrs(bio, REQ_OP_READ, 0);
1282 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1283 goto submit_and_realloc;
1285 last_block_in_bio = block_nr;
1289 zero_user_segment(page, 0, PAGE_SIZE);
1294 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1302 BUG_ON(pages && !list_empty(pages));
1304 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1308 static int f2fs_read_data_page(struct file *file, struct page *page)
1310 struct inode *inode = page->mapping->host;
1313 trace_f2fs_readpage(page, DATA);
1315 /* If the file has inline data, try to read it directly */
1316 if (f2fs_has_inline_data(inode))
1317 ret = f2fs_read_inline_data(inode, page);
1319 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1323 static int f2fs_read_data_pages(struct file *file,
1324 struct address_space *mapping,
1325 struct list_head *pages, unsigned nr_pages)
1327 struct inode *inode = file->f_mapping->host;
1328 struct page *page = list_last_entry(pages, struct page, lru);
1330 trace_f2fs_readpages(inode, page, nr_pages);
1332 /* If the file has inline data, skip readpages */
1333 if (f2fs_has_inline_data(inode))
1336 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1339 static int encrypt_one_page(struct f2fs_io_info *fio)
1341 struct inode *inode = fio->page->mapping->host;
1342 gfp_t gfp_flags = GFP_NOFS;
1344 if (!f2fs_encrypted_inode(inode) || !S_ISREG(inode->i_mode))
1347 /* wait for GCed encrypted page writeback */
1348 f2fs_wait_on_encrypted_page_writeback(fio->sbi, fio->old_blkaddr);
1351 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1352 PAGE_SIZE, 0, fio->page->index, gfp_flags);
1353 if (!IS_ERR(fio->encrypted_page))
1356 /* flush pending IOs and wait for a while in the ENOMEM case */
1357 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1358 f2fs_flush_merged_writes(fio->sbi);
1359 congestion_wait(BLK_RW_ASYNC, HZ/50);
1360 gfp_flags |= __GFP_NOFAIL;
1363 return PTR_ERR(fio->encrypted_page);
1366 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1368 struct inode *inode = fio->page->mapping->host;
1370 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
1372 if (is_cold_data(fio->page))
1374 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1377 return need_inplace_update_policy(inode, fio);
1380 static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio)
1382 if (fio->old_blkaddr == NEW_ADDR)
1384 if (fio->old_blkaddr == NULL_ADDR)
1389 int do_write_data_page(struct f2fs_io_info *fio)
1391 struct page *page = fio->page;
1392 struct inode *inode = page->mapping->host;
1393 struct dnode_of_data dn;
1394 struct extent_info ei = {0,0,0};
1395 bool ipu_force = false;
1398 set_new_dnode(&dn, inode, NULL, NULL, 0);
1399 if (need_inplace_update(fio) &&
1400 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1401 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1403 if (valid_ipu_blkaddr(fio)) {
1405 fio->need_lock = LOCK_DONE;
1410 /* Deadlock due to between page->lock and f2fs_lock_op */
1411 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1414 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1418 fio->old_blkaddr = dn.data_blkaddr;
1420 /* This page is already truncated */
1421 if (fio->old_blkaddr == NULL_ADDR) {
1422 ClearPageUptodate(page);
1427 * If current allocation needs SSR,
1428 * it had better in-place writes for updated data.
1430 if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) {
1431 err = encrypt_one_page(fio);
1435 set_page_writeback(page);
1436 f2fs_put_dnode(&dn);
1437 if (fio->need_lock == LOCK_REQ)
1438 f2fs_unlock_op(fio->sbi);
1439 err = rewrite_data_page(fio);
1440 trace_f2fs_do_write_data_page(fio->page, IPU);
1441 set_inode_flag(inode, FI_UPDATE_WRITE);
1445 if (fio->need_lock == LOCK_RETRY) {
1446 if (!f2fs_trylock_op(fio->sbi)) {
1450 fio->need_lock = LOCK_REQ;
1453 err = encrypt_one_page(fio);
1457 set_page_writeback(page);
1459 /* LFS mode write path */
1460 write_data_page(&dn, fio);
1461 trace_f2fs_do_write_data_page(page, OPU);
1462 set_inode_flag(inode, FI_APPEND_WRITE);
1463 if (page->index == 0)
1464 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1466 f2fs_put_dnode(&dn);
1468 if (fio->need_lock == LOCK_REQ)
1469 f2fs_unlock_op(fio->sbi);
1473 static int __write_data_page(struct page *page, bool *submitted,
1474 struct writeback_control *wbc)
1476 struct inode *inode = page->mapping->host;
1477 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1478 loff_t i_size = i_size_read(inode);
1479 const pgoff_t end_index = ((unsigned long long) i_size)
1481 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1482 unsigned offset = 0;
1483 bool need_balance_fs = false;
1485 struct f2fs_io_info fio = {
1489 .op_flags = wbc_to_write_flags(wbc),
1490 .old_blkaddr = NULL_ADDR,
1492 .encrypted_page = NULL,
1494 .need_lock = LOCK_RETRY,
1497 trace_f2fs_writepage(page, DATA);
1499 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1502 if (page->index < end_index)
1506 * If the offset is out-of-range of file size,
1507 * this page does not have to be written to disk.
1509 offset = i_size & (PAGE_SIZE - 1);
1510 if ((page->index >= end_index + 1) || !offset)
1513 zero_user_segment(page, offset, PAGE_SIZE);
1515 if (f2fs_is_drop_cache(inode))
1517 /* we should not write 0'th page having journal header */
1518 if (f2fs_is_volatile_file(inode) && (!page->index ||
1519 (!wbc->for_reclaim &&
1520 available_free_memory(sbi, BASE_CHECK))))
1523 /* we should bypass data pages to proceed the kworkder jobs */
1524 if (unlikely(f2fs_cp_error(sbi))) {
1525 mapping_set_error(page->mapping, -EIO);
1529 /* Dentry blocks are controlled by checkpoint */
1530 if (S_ISDIR(inode->i_mode)) {
1531 fio.need_lock = LOCK_DONE;
1532 err = do_write_data_page(&fio);
1536 if (!wbc->for_reclaim)
1537 need_balance_fs = true;
1538 else if (has_not_enough_free_secs(sbi, 0, 0))
1541 set_inode_flag(inode, FI_HOT_DATA);
1544 if (f2fs_has_inline_data(inode)) {
1545 err = f2fs_write_inline_data(inode, page);
1550 if (err == -EAGAIN) {
1551 err = do_write_data_page(&fio);
1552 if (err == -EAGAIN) {
1553 fio.need_lock = LOCK_REQ;
1554 err = do_write_data_page(&fio);
1557 if (F2FS_I(inode)->last_disk_size < psize)
1558 F2FS_I(inode)->last_disk_size = psize;
1561 if (err && err != -ENOENT)
1565 inode_dec_dirty_pages(inode);
1567 ClearPageUptodate(page);
1569 if (wbc->for_reclaim) {
1570 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1571 clear_inode_flag(inode, FI_HOT_DATA);
1572 remove_dirty_inode(inode);
1577 if (!S_ISDIR(inode->i_mode))
1578 f2fs_balance_fs(sbi, need_balance_fs);
1580 if (unlikely(f2fs_cp_error(sbi))) {
1581 f2fs_submit_merged_write(sbi, DATA);
1586 *submitted = fio.submitted;
1591 redirty_page_for_writepage(wbc, page);
1593 return AOP_WRITEPAGE_ACTIVATE;
1598 static int f2fs_write_data_page(struct page *page,
1599 struct writeback_control *wbc)
1601 return __write_data_page(page, NULL, wbc);
1605 * This function was copied from write_cche_pages from mm/page-writeback.c.
1606 * The major change is making write step of cold data page separately from
1607 * warm/hot data page.
1609 static int f2fs_write_cache_pages(struct address_space *mapping,
1610 struct writeback_control *wbc)
1614 struct pagevec pvec;
1616 pgoff_t uninitialized_var(writeback_index);
1618 pgoff_t end; /* Inclusive */
1620 pgoff_t last_idx = ULONG_MAX;
1622 int range_whole = 0;
1625 pagevec_init(&pvec, 0);
1627 if (get_dirty_pages(mapping->host) <=
1628 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1629 set_inode_flag(mapping->host, FI_HOT_DATA);
1631 clear_inode_flag(mapping->host, FI_HOT_DATA);
1633 if (wbc->range_cyclic) {
1634 writeback_index = mapping->writeback_index; /* prev offset */
1635 index = writeback_index;
1642 index = wbc->range_start >> PAGE_SHIFT;
1643 end = wbc->range_end >> PAGE_SHIFT;
1644 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1646 cycled = 1; /* ignore range_cyclic tests */
1648 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1649 tag = PAGECACHE_TAG_TOWRITE;
1651 tag = PAGECACHE_TAG_DIRTY;
1653 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1654 tag_pages_for_writeback(mapping, index, end);
1656 while (!done && (index <= end)) {
1659 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1660 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1664 for (i = 0; i < nr_pages; i++) {
1665 struct page *page = pvec.pages[i];
1666 bool submitted = false;
1668 if (page->index > end) {
1673 done_index = page->index;
1677 if (unlikely(page->mapping != mapping)) {
1683 if (!PageDirty(page)) {
1684 /* someone wrote it for us */
1685 goto continue_unlock;
1688 if (PageWriteback(page)) {
1689 if (wbc->sync_mode != WB_SYNC_NONE)
1690 f2fs_wait_on_page_writeback(page,
1693 goto continue_unlock;
1696 BUG_ON(PageWriteback(page));
1697 if (!clear_page_dirty_for_io(page))
1698 goto continue_unlock;
1700 ret = __write_data_page(page, &submitted, wbc);
1701 if (unlikely(ret)) {
1703 * keep nr_to_write, since vfs uses this to
1704 * get # of written pages.
1706 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1710 } else if (ret == -EAGAIN) {
1712 if (wbc->sync_mode == WB_SYNC_ALL) {
1714 congestion_wait(BLK_RW_ASYNC,
1720 done_index = page->index + 1;
1723 } else if (submitted) {
1724 last_idx = page->index;
1727 /* give a priority to WB_SYNC threads */
1728 if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) ||
1729 --wbc->nr_to_write <= 0) &&
1730 wbc->sync_mode == WB_SYNC_NONE) {
1735 pagevec_release(&pvec);
1739 if (!cycled && !done) {
1742 end = writeback_index - 1;
1745 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1746 mapping->writeback_index = done_index;
1748 if (last_idx != ULONG_MAX)
1749 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
1755 static int f2fs_write_data_pages(struct address_space *mapping,
1756 struct writeback_control *wbc)
1758 struct inode *inode = mapping->host;
1759 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1760 struct blk_plug plug;
1763 /* deal with chardevs and other special file */
1764 if (!mapping->a_ops->writepage)
1767 /* skip writing if there is no dirty page in this inode */
1768 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1771 /* during POR, we don't need to trigger writepage at all. */
1772 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1775 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1776 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1777 available_free_memory(sbi, DIRTY_DENTS))
1780 /* skip writing during file defragment */
1781 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1784 trace_f2fs_writepages(mapping->host, wbc, DATA);
1786 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
1787 if (wbc->sync_mode == WB_SYNC_ALL)
1788 atomic_inc(&sbi->wb_sync_req);
1789 else if (atomic_read(&sbi->wb_sync_req))
1792 blk_start_plug(&plug);
1793 ret = f2fs_write_cache_pages(mapping, wbc);
1794 blk_finish_plug(&plug);
1796 if (wbc->sync_mode == WB_SYNC_ALL)
1797 atomic_dec(&sbi->wb_sync_req);
1799 * if some pages were truncated, we cannot guarantee its mapping->host
1800 * to detect pending bios.
1803 remove_dirty_inode(inode);
1807 wbc->pages_skipped += get_dirty_pages(inode);
1808 trace_f2fs_writepages(mapping->host, wbc, DATA);
1812 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1814 struct inode *inode = mapping->host;
1815 loff_t i_size = i_size_read(inode);
1818 down_write(&F2FS_I(inode)->i_mmap_sem);
1819 truncate_pagecache(inode, i_size);
1820 truncate_blocks(inode, i_size, true);
1821 up_write(&F2FS_I(inode)->i_mmap_sem);
1825 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1826 struct page *page, loff_t pos, unsigned len,
1827 block_t *blk_addr, bool *node_changed)
1829 struct inode *inode = page->mapping->host;
1830 pgoff_t index = page->index;
1831 struct dnode_of_data dn;
1833 bool locked = false;
1834 struct extent_info ei = {0,0,0};
1838 * we already allocated all the blocks, so we don't need to get
1839 * the block addresses when there is no need to fill the page.
1841 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1842 !is_inode_flag_set(inode, FI_NO_PREALLOC))
1845 if (f2fs_has_inline_data(inode) ||
1846 (pos & PAGE_MASK) >= i_size_read(inode)) {
1847 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1851 /* check inline_data */
1852 ipage = get_node_page(sbi, inode->i_ino);
1853 if (IS_ERR(ipage)) {
1854 err = PTR_ERR(ipage);
1858 set_new_dnode(&dn, inode, ipage, ipage, 0);
1860 if (f2fs_has_inline_data(inode)) {
1861 if (pos + len <= MAX_INLINE_DATA) {
1862 read_inline_data(page, ipage);
1863 set_inode_flag(inode, FI_DATA_EXIST);
1865 set_inline_node(ipage);
1867 err = f2fs_convert_inline_page(&dn, page);
1870 if (dn.data_blkaddr == NULL_ADDR)
1871 err = f2fs_get_block(&dn, index);
1873 } else if (locked) {
1874 err = f2fs_get_block(&dn, index);
1876 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1877 dn.data_blkaddr = ei.blk + index - ei.fofs;
1880 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1881 if (err || dn.data_blkaddr == NULL_ADDR) {
1882 f2fs_put_dnode(&dn);
1883 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
1891 /* convert_inline_page can make node_changed */
1892 *blk_addr = dn.data_blkaddr;
1893 *node_changed = dn.node_changed;
1895 f2fs_put_dnode(&dn);
1898 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1902 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1903 loff_t pos, unsigned len, unsigned flags,
1904 struct page **pagep, void **fsdata)
1906 struct inode *inode = mapping->host;
1907 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1908 struct page *page = NULL;
1909 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1910 bool need_balance = false;
1911 block_t blkaddr = NULL_ADDR;
1914 trace_f2fs_write_begin(inode, pos, len, flags);
1917 * We should check this at this moment to avoid deadlock on inode page
1918 * and #0 page. The locking rule for inline_data conversion should be:
1919 * lock_page(page #0) -> lock_page(inode_page)
1922 err = f2fs_convert_inline_inode(inode);
1928 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1929 * wait_for_stable_page. Will wait that below with our IO control.
1931 page = pagecache_get_page(mapping, index,
1932 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1940 err = prepare_write_begin(sbi, page, pos, len,
1941 &blkaddr, &need_balance);
1945 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1947 f2fs_balance_fs(sbi, true);
1949 if (page->mapping != mapping) {
1950 /* The page got truncated from under us */
1951 f2fs_put_page(page, 1);
1956 f2fs_wait_on_page_writeback(page, DATA, false);
1958 /* wait for GCed encrypted page writeback */
1959 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1960 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1962 if (len == PAGE_SIZE || PageUptodate(page))
1965 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
1966 zero_user_segment(page, len, PAGE_SIZE);
1970 if (blkaddr == NEW_ADDR) {
1971 zero_user_segment(page, 0, PAGE_SIZE);
1972 SetPageUptodate(page);
1976 bio = f2fs_grab_bio(inode, blkaddr, 1);
1981 bio->bi_opf = REQ_OP_READ;
1982 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1988 __submit_bio(sbi, bio, DATA);
1991 if (unlikely(page->mapping != mapping)) {
1992 f2fs_put_page(page, 1);
1995 if (unlikely(!PageUptodate(page))) {
2003 f2fs_put_page(page, 1);
2004 f2fs_write_failed(mapping, pos + len);
2008 static int f2fs_write_end(struct file *file,
2009 struct address_space *mapping,
2010 loff_t pos, unsigned len, unsigned copied,
2011 struct page *page, void *fsdata)
2013 struct inode *inode = page->mapping->host;
2015 trace_f2fs_write_end(inode, pos, len, copied);
2018 * This should be come from len == PAGE_SIZE, and we expect copied
2019 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2020 * let generic_perform_write() try to copy data again through copied=0.
2022 if (!PageUptodate(page)) {
2023 if (unlikely(copied != len))
2026 SetPageUptodate(page);
2031 set_page_dirty(page);
2033 if (pos + copied > i_size_read(inode))
2034 f2fs_i_size_write(inode, pos + copied);
2036 f2fs_put_page(page, 1);
2037 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2041 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2044 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
2046 if (offset & blocksize_mask)
2049 if (iov_iter_alignment(iter) & blocksize_mask)
2055 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2057 struct address_space *mapping = iocb->ki_filp->f_mapping;
2058 struct inode *inode = mapping->host;
2059 size_t count = iov_iter_count(iter);
2060 loff_t offset = iocb->ki_pos;
2061 int rw = iov_iter_rw(iter);
2064 err = check_direct_IO(inode, iter, offset);
2068 if (__force_buffered_io(inode, rw))
2071 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2073 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
2074 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2075 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
2079 set_inode_flag(inode, FI_UPDATE_WRITE);
2081 f2fs_write_failed(mapping, offset + count);
2084 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2089 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2090 unsigned int length)
2092 struct inode *inode = page->mapping->host;
2093 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2095 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2096 (offset % PAGE_SIZE || length != PAGE_SIZE))
2099 if (PageDirty(page)) {
2100 if (inode->i_ino == F2FS_META_INO(sbi)) {
2101 dec_page_count(sbi, F2FS_DIRTY_META);
2102 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2103 dec_page_count(sbi, F2FS_DIRTY_NODES);
2105 inode_dec_dirty_pages(inode);
2106 remove_dirty_inode(inode);
2110 /* This is atomic written page, keep Private */
2111 if (IS_ATOMIC_WRITTEN_PAGE(page))
2112 return drop_inmem_page(inode, page);
2114 set_page_private(page, 0);
2115 ClearPagePrivate(page);
2118 int f2fs_release_page(struct page *page, gfp_t wait)
2120 /* If this is dirty page, keep PagePrivate */
2121 if (PageDirty(page))
2124 /* This is atomic written page, keep Private */
2125 if (IS_ATOMIC_WRITTEN_PAGE(page))
2128 set_page_private(page, 0);
2129 ClearPagePrivate(page);
2134 * This was copied from __set_page_dirty_buffers which gives higher performance
2135 * in very high speed storages. (e.g., pmem)
2137 void f2fs_set_page_dirty_nobuffers(struct page *page)
2139 struct address_space *mapping = page->mapping;
2140 unsigned long flags;
2142 if (unlikely(!mapping))
2145 spin_lock(&mapping->private_lock);
2146 lock_page_memcg(page);
2148 spin_unlock(&mapping->private_lock);
2150 spin_lock_irqsave(&mapping->tree_lock, flags);
2151 WARN_ON_ONCE(!PageUptodate(page));
2152 account_page_dirtied(page, mapping);
2153 radix_tree_tag_set(&mapping->page_tree,
2154 page_index(page), PAGECACHE_TAG_DIRTY);
2155 spin_unlock_irqrestore(&mapping->tree_lock, flags);
2156 unlock_page_memcg(page);
2158 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2162 static int f2fs_set_data_page_dirty(struct page *page)
2164 struct address_space *mapping = page->mapping;
2165 struct inode *inode = mapping->host;
2167 trace_f2fs_set_page_dirty(page, DATA);
2169 if (!PageUptodate(page))
2170 SetPageUptodate(page);
2172 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2173 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2174 register_inmem_page(inode, page);
2178 * Previously, this page has been registered, we just
2184 if (!PageDirty(page)) {
2185 f2fs_set_page_dirty_nobuffers(page);
2186 update_dirty_page(inode, page);
2192 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2194 struct inode *inode = mapping->host;
2196 if (f2fs_has_inline_data(inode))
2199 /* make sure allocating whole blocks */
2200 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2201 filemap_write_and_wait(mapping);
2203 return generic_block_bmap(mapping, block, get_data_block_bmap);
2206 #ifdef CONFIG_MIGRATION
2207 #include <linux/migrate.h>
2209 int f2fs_migrate_page(struct address_space *mapping,
2210 struct page *newpage, struct page *page, enum migrate_mode mode)
2212 int rc, extra_count;
2213 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2214 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2216 BUG_ON(PageWriteback(page));
2218 /* migrating an atomic written page is safe with the inmem_lock hold */
2219 if (atomic_written) {
2220 if (mode != MIGRATE_SYNC)
2222 if (!mutex_trylock(&fi->inmem_lock))
2227 * A reference is expected if PagePrivate set when move mapping,
2228 * however F2FS breaks this for maintaining dirty page counts when
2229 * truncating pages. So here adjusting the 'extra_count' make it work.
2231 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2232 rc = migrate_page_move_mapping(mapping, newpage,
2233 page, NULL, mode, extra_count);
2234 if (rc != MIGRATEPAGE_SUCCESS) {
2236 mutex_unlock(&fi->inmem_lock);
2240 if (atomic_written) {
2241 struct inmem_pages *cur;
2242 list_for_each_entry(cur, &fi->inmem_pages, list)
2243 if (cur->page == page) {
2244 cur->page = newpage;
2247 mutex_unlock(&fi->inmem_lock);
2252 if (PagePrivate(page))
2253 SetPagePrivate(newpage);
2254 set_page_private(newpage, page_private(page));
2256 migrate_page_copy(newpage, page);
2258 return MIGRATEPAGE_SUCCESS;
2262 const struct address_space_operations f2fs_dblock_aops = {
2263 .readpage = f2fs_read_data_page,
2264 .readpages = f2fs_read_data_pages,
2265 .writepage = f2fs_write_data_page,
2266 .writepages = f2fs_write_data_pages,
2267 .write_begin = f2fs_write_begin,
2268 .write_end = f2fs_write_end,
2269 .set_page_dirty = f2fs_set_data_page_dirty,
2270 .invalidatepage = f2fs_invalidate_page,
2271 .releasepage = f2fs_release_page,
2272 .direct_IO = f2fs_direct_IO,
2274 #ifdef CONFIG_MIGRATION
2275 .migratepage = f2fs_migrate_page,