1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/buffer_head.h>
11 #include <linux/mpage.h>
12 #include <linux/writeback.h>
13 #include <linux/backing-dev.h>
14 #include <linux/pagevec.h>
15 #include <linux/blkdev.h>
16 #include <linux/bio.h>
17 #include <linux/prefetch.h>
18 #include <linux/uio.h>
19 #include <linux/cleancache.h>
20 #include <linux/sched/signal.h>
26 #include <trace/events/f2fs.h>
28 #define NUM_PREALLOC_POST_READ_CTXS 128
30 static struct kmem_cache *bio_post_read_ctx_cache;
31 static mempool_t *bio_post_read_ctx_pool;
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) ||
48 (S_ISREG(inode->i_mode) &&
49 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
55 static enum count_type __read_io_type(struct page *page)
57 struct address_space *mapping = page->mapping;
60 struct inode *inode = mapping->host;
61 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
63 if (inode->i_ino == F2FS_META_INO(sbi))
66 if (inode->i_ino == F2FS_NODE_INO(sbi))
72 /* postprocessing steps for read bios */
73 enum bio_post_read_step {
78 struct bio_post_read_ctx {
80 struct work_struct work;
81 unsigned int cur_step;
82 unsigned int enabled_steps;
85 static void __read_end_io(struct bio *bio)
91 bio_for_each_segment_all(bv, bio, i) {
94 /* PG_error was set if any post_read step failed */
95 if (bio->bi_status || PageError(page)) {
96 ClearPageUptodate(page);
97 /* will re-read again later */
100 SetPageUptodate(page);
102 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
106 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
110 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
112 static void decrypt_work(struct work_struct *work)
114 struct bio_post_read_ctx *ctx =
115 container_of(work, struct bio_post_read_ctx, work);
117 fscrypt_decrypt_bio(ctx->bio);
119 bio_post_read_processing(ctx);
122 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
124 switch (++ctx->cur_step) {
126 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
127 INIT_WORK(&ctx->work, decrypt_work);
128 fscrypt_enqueue_decrypt_work(&ctx->work);
134 __read_end_io(ctx->bio);
138 static bool f2fs_bio_post_read_required(struct bio *bio)
140 return bio->bi_private && !bio->bi_status;
143 static void f2fs_read_end_io(struct bio *bio)
145 if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)),
147 f2fs_show_injection_info(FAULT_READ_IO);
148 bio->bi_status = BLK_STS_IOERR;
151 if (f2fs_bio_post_read_required(bio)) {
152 struct bio_post_read_ctx *ctx = bio->bi_private;
154 ctx->cur_step = STEP_INITIAL;
155 bio_post_read_processing(ctx);
162 static void f2fs_write_end_io(struct bio *bio)
164 struct f2fs_sb_info *sbi = bio->bi_private;
165 struct bio_vec *bvec;
168 if (time_to_inject(sbi, FAULT_WRITE_IO)) {
169 f2fs_show_injection_info(FAULT_WRITE_IO);
170 bio->bi_status = BLK_STS_IOERR;
173 bio_for_each_segment_all(bvec, bio, i) {
174 struct page *page = bvec->bv_page;
175 enum count_type type = WB_DATA_TYPE(page);
177 if (IS_DUMMY_WRITTEN_PAGE(page)) {
178 set_page_private(page, (unsigned long)NULL);
179 ClearPagePrivate(page);
181 mempool_free(page, sbi->write_io_dummy);
183 if (unlikely(bio->bi_status))
184 f2fs_stop_checkpoint(sbi, true);
188 fscrypt_pullback_bio_page(&page, true);
190 if (unlikely(bio->bi_status)) {
191 mapping_set_error(page->mapping, -EIO);
192 if (type == F2FS_WB_CP_DATA)
193 f2fs_stop_checkpoint(sbi, true);
196 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
197 page->index != nid_of_node(page));
199 dec_page_count(sbi, type);
200 if (f2fs_in_warm_node_list(sbi, page))
201 f2fs_del_fsync_node_entry(sbi, page);
202 clear_cold_data(page);
203 end_page_writeback(page);
205 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
206 wq_has_sleeper(&sbi->cp_wait))
207 wake_up(&sbi->cp_wait);
213 * Return true, if pre_bio's bdev is same as its target device.
215 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
216 block_t blk_addr, struct bio *bio)
218 struct block_device *bdev = sbi->sb->s_bdev;
221 for (i = 0; i < sbi->s_ndevs; i++) {
222 if (FDEV(i).start_blk <= blk_addr &&
223 FDEV(i).end_blk >= blk_addr) {
224 blk_addr -= FDEV(i).start_blk;
230 bio_set_dev(bio, bdev);
231 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
236 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
240 for (i = 0; i < sbi->s_ndevs; i++)
241 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
246 static bool __same_bdev(struct f2fs_sb_info *sbi,
247 block_t blk_addr, struct bio *bio)
249 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
250 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
254 * Low-level block read/write IO operations.
256 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
257 struct writeback_control *wbc,
258 int npages, bool is_read,
259 enum page_type type, enum temp_type temp)
263 bio = f2fs_bio_alloc(sbi, npages, true);
265 f2fs_target_device(sbi, blk_addr, bio);
267 bio->bi_end_io = f2fs_read_end_io;
268 bio->bi_private = NULL;
270 bio->bi_end_io = f2fs_write_end_io;
271 bio->bi_private = sbi;
272 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
275 wbc_init_bio(wbc, bio);
280 static inline void __submit_bio(struct f2fs_sb_info *sbi,
281 struct bio *bio, enum page_type type)
283 if (!is_read_io(bio_op(bio))) {
286 if (type != DATA && type != NODE)
289 if (test_opt(sbi, LFS) && current->plug)
290 blk_finish_plug(current->plug);
292 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
293 start %= F2FS_IO_SIZE(sbi);
298 /* fill dummy pages */
299 for (; start < F2FS_IO_SIZE(sbi); start++) {
301 mempool_alloc(sbi->write_io_dummy,
302 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
303 f2fs_bug_on(sbi, !page);
305 SetPagePrivate(page);
306 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
308 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
312 * In the NODE case, we lose next block address chain. So, we
313 * need to do checkpoint in f2fs_sync_file.
316 set_sbi_flag(sbi, SBI_NEED_CP);
319 if (is_read_io(bio_op(bio)))
320 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
322 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
326 static void __submit_merged_bio(struct f2fs_bio_info *io)
328 struct f2fs_io_info *fio = &io->fio;
333 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
335 if (is_read_io(fio->op))
336 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
338 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
340 __submit_bio(io->sbi, io->bio, fio->type);
344 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
345 struct page *page, nid_t ino)
347 struct bio_vec *bvec;
354 if (!inode && !page && !ino)
357 bio_for_each_segment_all(bvec, io->bio, i) {
359 if (bvec->bv_page->mapping)
360 target = bvec->bv_page;
362 target = fscrypt_control_page(bvec->bv_page);
364 if (inode && inode == target->mapping->host)
366 if (page && page == target)
368 if (ino && ino == ino_of_node(target))
375 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
376 enum page_type type, enum temp_type temp)
378 enum page_type btype = PAGE_TYPE_OF_BIO(type);
379 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
381 down_write(&io->io_rwsem);
383 /* change META to META_FLUSH in the checkpoint procedure */
384 if (type >= META_FLUSH) {
385 io->fio.type = META_FLUSH;
386 io->fio.op = REQ_OP_WRITE;
387 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
388 if (!test_opt(sbi, NOBARRIER))
389 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
391 __submit_merged_bio(io);
392 up_write(&io->io_rwsem);
395 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
396 struct inode *inode, struct page *page,
397 nid_t ino, enum page_type type, bool force)
402 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
404 enum page_type btype = PAGE_TYPE_OF_BIO(type);
405 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
407 down_read(&io->io_rwsem);
408 ret = __has_merged_page(io, inode, page, ino);
409 up_read(&io->io_rwsem);
412 __f2fs_submit_merged_write(sbi, type, temp);
414 /* TODO: use HOT temp only for meta pages now. */
420 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
422 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
425 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
426 struct inode *inode, struct page *page,
427 nid_t ino, enum page_type type)
429 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
432 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
434 f2fs_submit_merged_write(sbi, DATA);
435 f2fs_submit_merged_write(sbi, NODE);
436 f2fs_submit_merged_write(sbi, META);
440 * Fill the locked page with data located in the block address.
441 * A caller needs to unlock the page on failure.
443 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
446 struct page *page = fio->encrypted_page ?
447 fio->encrypted_page : fio->page;
449 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
450 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
453 trace_f2fs_submit_page_bio(page, fio);
454 f2fs_trace_ios(fio, 0);
456 /* Allocate a new bio */
457 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
458 1, is_read_io(fio->op), fio->type, fio->temp);
460 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
465 if (fio->io_wbc && !is_read_io(fio->op))
466 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
468 bio_set_op_attrs(bio, fio->op, fio->op_flags);
470 inc_page_count(fio->sbi, is_read_io(fio->op) ?
471 __read_io_type(page): WB_DATA_TYPE(fio->page));
473 __submit_bio(fio->sbi, bio, fio->type);
477 void f2fs_submit_page_write(struct f2fs_io_info *fio)
479 struct f2fs_sb_info *sbi = fio->sbi;
480 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
481 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
482 struct page *bio_page;
484 f2fs_bug_on(sbi, is_read_io(fio->op));
486 down_write(&io->io_rwsem);
489 spin_lock(&io->io_lock);
490 if (list_empty(&io->io_list)) {
491 spin_unlock(&io->io_lock);
494 fio = list_first_entry(&io->io_list,
495 struct f2fs_io_info, list);
496 list_del(&fio->list);
497 spin_unlock(&io->io_lock);
500 if (__is_valid_data_blkaddr(fio->old_blkaddr))
501 verify_block_addr(fio, fio->old_blkaddr);
502 verify_block_addr(fio, fio->new_blkaddr);
504 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
506 /* set submitted = true as a return value */
507 fio->submitted = true;
509 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
511 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
512 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
513 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
514 __submit_merged_bio(io);
516 if (io->bio == NULL) {
517 if ((fio->type == DATA || fio->type == NODE) &&
518 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
519 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
523 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
524 BIO_MAX_PAGES, false,
525 fio->type, fio->temp);
529 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
530 __submit_merged_bio(io);
535 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
537 io->last_block_in_bio = fio->new_blkaddr;
538 f2fs_trace_ios(fio, 0);
540 trace_f2fs_submit_page_write(fio->page, fio);
545 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
546 f2fs_is_checkpoint_ready(sbi))
547 __submit_merged_bio(io);
548 up_write(&io->io_rwsem);
551 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
552 unsigned nr_pages, unsigned op_flag)
554 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
556 struct bio_post_read_ctx *ctx;
557 unsigned int post_read_steps = 0;
559 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
560 return ERR_PTR(-EFAULT);
562 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
564 return ERR_PTR(-ENOMEM);
565 f2fs_target_device(sbi, blkaddr, bio);
566 bio->bi_end_io = f2fs_read_end_io;
567 bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
569 if (f2fs_encrypted_file(inode))
570 post_read_steps |= 1 << STEP_DECRYPT;
571 if (post_read_steps) {
572 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
575 return ERR_PTR(-ENOMEM);
578 ctx->enabled_steps = post_read_steps;
579 bio->bi_private = ctx;
585 /* This can handle encryption stuffs */
586 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
589 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
594 /* wait for GCed page writeback via META_MAPPING */
595 f2fs_wait_on_block_writeback(inode, blkaddr);
597 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
601 ClearPageError(page);
602 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
603 __submit_bio(F2FS_I_SB(inode), bio, DATA);
607 static void __set_data_blkaddr(struct dnode_of_data *dn)
609 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
613 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
614 base = get_extra_isize(dn->inode);
616 /* Get physical address of data block */
617 addr_array = blkaddr_in_node(rn);
618 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
622 * Lock ordering for the change of data block address:
625 * update block addresses in the node page
627 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
629 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
630 __set_data_blkaddr(dn);
631 if (set_page_dirty(dn->node_page))
632 dn->node_changed = true;
635 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
637 dn->data_blkaddr = blkaddr;
638 f2fs_set_data_blkaddr(dn);
639 f2fs_update_extent_cache(dn);
642 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
643 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
645 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
651 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
653 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
656 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
657 dn->ofs_in_node, count);
659 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
661 for (; count > 0; dn->ofs_in_node++) {
662 block_t blkaddr = datablock_addr(dn->inode,
663 dn->node_page, dn->ofs_in_node);
664 if (blkaddr == NULL_ADDR) {
665 dn->data_blkaddr = NEW_ADDR;
666 __set_data_blkaddr(dn);
671 if (set_page_dirty(dn->node_page))
672 dn->node_changed = true;
676 /* Should keep dn->ofs_in_node unchanged */
677 int f2fs_reserve_new_block(struct dnode_of_data *dn)
679 unsigned int ofs_in_node = dn->ofs_in_node;
682 ret = f2fs_reserve_new_blocks(dn, 1);
683 dn->ofs_in_node = ofs_in_node;
687 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
689 bool need_put = dn->inode_page ? false : true;
692 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
696 if (dn->data_blkaddr == NULL_ADDR)
697 err = f2fs_reserve_new_block(dn);
703 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
705 struct extent_info ei = {0,0,0};
706 struct inode *inode = dn->inode;
708 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
709 dn->data_blkaddr = ei.blk + index - ei.fofs;
713 return f2fs_reserve_block(dn, index);
716 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
717 int op_flags, bool for_write)
719 struct address_space *mapping = inode->i_mapping;
720 struct dnode_of_data dn;
722 struct extent_info ei = {0,0,0};
725 page = f2fs_grab_cache_page(mapping, index, for_write);
727 return ERR_PTR(-ENOMEM);
729 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
730 dn.data_blkaddr = ei.blk + index - ei.fofs;
734 set_new_dnode(&dn, inode, NULL, NULL, 0);
735 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
740 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
745 if (PageUptodate(page)) {
751 * A new dentry page is allocated but not able to be written, since its
752 * new inode page couldn't be allocated due to -ENOSPC.
753 * In such the case, its blkaddr can be remained as NEW_ADDR.
754 * see, f2fs_add_link -> f2fs_get_new_data_page ->
755 * f2fs_init_inode_metadata.
757 if (dn.data_blkaddr == NEW_ADDR) {
758 zero_user_segment(page, 0, PAGE_SIZE);
759 if (!PageUptodate(page))
760 SetPageUptodate(page);
765 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
771 f2fs_put_page(page, 1);
775 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
777 struct address_space *mapping = inode->i_mapping;
780 page = find_get_page(mapping, index);
781 if (page && PageUptodate(page))
783 f2fs_put_page(page, 0);
785 page = f2fs_get_read_data_page(inode, index, 0, false);
789 if (PageUptodate(page))
792 wait_on_page_locked(page);
793 if (unlikely(!PageUptodate(page))) {
794 f2fs_put_page(page, 0);
795 return ERR_PTR(-EIO);
801 * If it tries to access a hole, return an error.
802 * Because, the callers, functions in dir.c and GC, should be able to know
803 * whether this page exists or not.
805 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
808 struct address_space *mapping = inode->i_mapping;
811 page = f2fs_get_read_data_page(inode, index, 0, for_write);
815 /* wait for read completion */
817 if (unlikely(page->mapping != mapping)) {
818 f2fs_put_page(page, 1);
821 if (unlikely(!PageUptodate(page))) {
822 f2fs_put_page(page, 1);
823 return ERR_PTR(-EIO);
829 * Caller ensures that this data page is never allocated.
830 * A new zero-filled data page is allocated in the page cache.
832 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
834 * Note that, ipage is set only by make_empty_dir, and if any error occur,
835 * ipage should be released by this function.
837 struct page *f2fs_get_new_data_page(struct inode *inode,
838 struct page *ipage, pgoff_t index, bool new_i_size)
840 struct address_space *mapping = inode->i_mapping;
842 struct dnode_of_data dn;
845 page = f2fs_grab_cache_page(mapping, index, true);
848 * before exiting, we should make sure ipage will be released
849 * if any error occur.
851 f2fs_put_page(ipage, 1);
852 return ERR_PTR(-ENOMEM);
855 set_new_dnode(&dn, inode, ipage, NULL, 0);
856 err = f2fs_reserve_block(&dn, index);
858 f2fs_put_page(page, 1);
864 if (PageUptodate(page))
867 if (dn.data_blkaddr == NEW_ADDR) {
868 zero_user_segment(page, 0, PAGE_SIZE);
869 if (!PageUptodate(page))
870 SetPageUptodate(page);
872 f2fs_put_page(page, 1);
874 /* if ipage exists, blkaddr should be NEW_ADDR */
875 f2fs_bug_on(F2FS_I_SB(inode), ipage);
876 page = f2fs_get_lock_data_page(inode, index, true);
881 if (new_i_size && i_size_read(inode) <
882 ((loff_t)(index + 1) << PAGE_SHIFT))
883 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
887 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
889 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
890 struct f2fs_summary sum;
896 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
899 err = f2fs_get_node_info(sbi, dn->nid, &ni);
903 dn->data_blkaddr = datablock_addr(dn->inode,
904 dn->node_page, dn->ofs_in_node);
905 if (dn->data_blkaddr != NULL_ADDR)
908 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
912 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
913 old_blkaddr = dn->data_blkaddr;
914 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
915 &sum, seg_type, NULL, false);
916 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
917 invalidate_mapping_pages(META_MAPPING(sbi),
918 old_blkaddr, old_blkaddr);
919 f2fs_set_data_blkaddr(dn);
922 * i_size will be updated by direct_IO. Otherwise, we'll get stale
923 * data from unwritten block via dio_read.
928 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
930 struct inode *inode = file_inode(iocb->ki_filp);
931 struct f2fs_map_blocks map;
934 bool direct_io = iocb->ki_flags & IOCB_DIRECT;
936 /* convert inline data for Direct I/O*/
938 err = f2fs_convert_inline_inode(inode);
943 if (direct_io && allow_outplace_dio(inode, iocb, from))
946 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
949 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
950 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
951 if (map.m_len > map.m_lblk)
952 map.m_len -= map.m_lblk;
956 map.m_next_pgofs = NULL;
957 map.m_next_extent = NULL;
958 map.m_seg_type = NO_CHECK_TYPE;
959 map.m_may_create = true;
962 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
963 flag = f2fs_force_buffered_io(inode, iocb, from) ?
964 F2FS_GET_BLOCK_PRE_AIO :
965 F2FS_GET_BLOCK_PRE_DIO;
968 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
969 err = f2fs_convert_inline_inode(inode);
973 if (f2fs_has_inline_data(inode))
976 flag = F2FS_GET_BLOCK_PRE_AIO;
979 err = f2fs_map_blocks(inode, &map, 1, flag);
980 if (map.m_len > 0 && err == -ENOSPC) {
982 set_inode_flag(inode, FI_NO_PREALLOC);
988 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
990 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
992 down_read(&sbi->node_change);
994 up_read(&sbi->node_change);
1004 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1005 * f2fs_map_blocks structure.
1006 * If original data blocks are allocated, then give them to blockdev.
1008 * a. preallocate requested block addresses
1009 * b. do not use extent cache for better performance
1010 * c. give the block addresses to blockdev
1012 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1013 int create, int flag)
1015 unsigned int maxblocks = map->m_len;
1016 struct dnode_of_data dn;
1017 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1018 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1019 pgoff_t pgofs, end_offset, end;
1020 int err = 0, ofs = 1;
1021 unsigned int ofs_in_node, last_ofs_in_node;
1023 struct extent_info ei = {0,0,0};
1025 unsigned int start_pgofs;
1033 /* it only supports block size == page size */
1034 pgofs = (pgoff_t)map->m_lblk;
1035 end = pgofs + maxblocks;
1037 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1038 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1042 map->m_pblk = ei.blk + pgofs - ei.fofs;
1043 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1044 map->m_flags = F2FS_MAP_MAPPED;
1045 if (map->m_next_extent)
1046 *map->m_next_extent = pgofs + map->m_len;
1048 /* for hardware encryption, but to avoid potential issue in future */
1049 if (flag == F2FS_GET_BLOCK_DIO)
1050 f2fs_wait_on_block_writeback_range(inode,
1051 map->m_pblk, map->m_len);
1056 if (map->m_may_create)
1057 __do_map_lock(sbi, flag, true);
1059 /* When reading holes, we need its node page */
1060 set_new_dnode(&dn, inode, NULL, NULL, 0);
1061 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1063 if (flag == F2FS_GET_BLOCK_BMAP)
1065 if (err == -ENOENT) {
1067 if (map->m_next_pgofs)
1068 *map->m_next_pgofs =
1069 f2fs_get_next_page_offset(&dn, pgofs);
1070 if (map->m_next_extent)
1071 *map->m_next_extent =
1072 f2fs_get_next_page_offset(&dn, pgofs);
1077 start_pgofs = pgofs;
1079 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1080 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1083 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1085 if (__is_valid_data_blkaddr(blkaddr) &&
1086 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1091 if (is_valid_data_blkaddr(sbi, blkaddr)) {
1092 /* use out-place-update for driect IO under LFS mode */
1093 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1094 map->m_may_create) {
1095 err = __allocate_data_block(&dn, map->m_seg_type);
1097 blkaddr = dn.data_blkaddr;
1098 set_inode_flag(inode, FI_APPEND_WRITE);
1103 if (unlikely(f2fs_cp_error(sbi))) {
1107 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1108 if (blkaddr == NULL_ADDR) {
1110 last_ofs_in_node = dn.ofs_in_node;
1113 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1114 flag != F2FS_GET_BLOCK_DIO);
1115 err = __allocate_data_block(&dn,
1118 set_inode_flag(inode, FI_APPEND_WRITE);
1122 map->m_flags |= F2FS_MAP_NEW;
1123 blkaddr = dn.data_blkaddr;
1125 if (flag == F2FS_GET_BLOCK_BMAP) {
1129 if (flag == F2FS_GET_BLOCK_PRECACHE)
1131 if (flag == F2FS_GET_BLOCK_FIEMAP &&
1132 blkaddr == NULL_ADDR) {
1133 if (map->m_next_pgofs)
1134 *map->m_next_pgofs = pgofs + 1;
1137 if (flag != F2FS_GET_BLOCK_FIEMAP) {
1138 /* for defragment case */
1139 if (map->m_next_pgofs)
1140 *map->m_next_pgofs = pgofs + 1;
1146 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1149 if (map->m_len == 0) {
1150 /* preallocated unwritten block should be mapped for fiemap. */
1151 if (blkaddr == NEW_ADDR)
1152 map->m_flags |= F2FS_MAP_UNWRITTEN;
1153 map->m_flags |= F2FS_MAP_MAPPED;
1155 map->m_pblk = blkaddr;
1157 } else if ((map->m_pblk != NEW_ADDR &&
1158 blkaddr == (map->m_pblk + ofs)) ||
1159 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1160 flag == F2FS_GET_BLOCK_PRE_DIO) {
1171 /* preallocate blocks in batch for one dnode page */
1172 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1173 (pgofs == end || dn.ofs_in_node == end_offset)) {
1175 dn.ofs_in_node = ofs_in_node;
1176 err = f2fs_reserve_new_blocks(&dn, prealloc);
1180 map->m_len += dn.ofs_in_node - ofs_in_node;
1181 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1185 dn.ofs_in_node = end_offset;
1190 else if (dn.ofs_in_node < end_offset)
1193 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1194 if (map->m_flags & F2FS_MAP_MAPPED) {
1195 unsigned int ofs = start_pgofs - map->m_lblk;
1197 f2fs_update_extent_cache_range(&dn,
1198 start_pgofs, map->m_pblk + ofs,
1203 f2fs_put_dnode(&dn);
1205 if (map->m_may_create) {
1206 __do_map_lock(sbi, flag, false);
1207 f2fs_balance_fs(sbi, dn.node_changed);
1213 /* for hardware encryption, but to avoid potential issue in future */
1214 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1215 f2fs_wait_on_block_writeback_range(inode,
1216 map->m_pblk, map->m_len);
1218 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1219 if (map->m_flags & F2FS_MAP_MAPPED) {
1220 unsigned int ofs = start_pgofs - map->m_lblk;
1222 f2fs_update_extent_cache_range(&dn,
1223 start_pgofs, map->m_pblk + ofs,
1226 if (map->m_next_extent)
1227 *map->m_next_extent = pgofs + 1;
1229 f2fs_put_dnode(&dn);
1231 if (map->m_may_create) {
1232 __do_map_lock(sbi, flag, false);
1233 f2fs_balance_fs(sbi, dn.node_changed);
1236 trace_f2fs_map_blocks(inode, map, err);
1240 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1242 struct f2fs_map_blocks map;
1246 if (pos + len > i_size_read(inode))
1249 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1250 map.m_next_pgofs = NULL;
1251 map.m_next_extent = NULL;
1252 map.m_seg_type = NO_CHECK_TYPE;
1253 map.m_may_create = false;
1254 last_lblk = F2FS_BLK_ALIGN(pos + len);
1256 while (map.m_lblk < last_lblk) {
1257 map.m_len = last_lblk - map.m_lblk;
1258 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1259 if (err || map.m_len == 0)
1261 map.m_lblk += map.m_len;
1266 static int __get_data_block(struct inode *inode, sector_t iblock,
1267 struct buffer_head *bh, int create, int flag,
1268 pgoff_t *next_pgofs, int seg_type, bool may_write)
1270 struct f2fs_map_blocks map;
1273 map.m_lblk = iblock;
1274 map.m_len = bh->b_size >> inode->i_blkbits;
1275 map.m_next_pgofs = next_pgofs;
1276 map.m_next_extent = NULL;
1277 map.m_seg_type = seg_type;
1278 map.m_may_create = may_write;
1280 err = f2fs_map_blocks(inode, &map, create, flag);
1282 map_bh(bh, inode->i_sb, map.m_pblk);
1283 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1284 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1289 static int get_data_block(struct inode *inode, sector_t iblock,
1290 struct buffer_head *bh_result, int create, int flag,
1291 pgoff_t *next_pgofs)
1293 return __get_data_block(inode, iblock, bh_result, create,
1295 NO_CHECK_TYPE, create);
1298 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1299 struct buffer_head *bh_result, int create)
1301 return __get_data_block(inode, iblock, bh_result, create,
1302 F2FS_GET_BLOCK_DIO, NULL,
1303 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1307 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1308 struct buffer_head *bh_result, int create)
1310 return __get_data_block(inode, iblock, bh_result, create,
1311 F2FS_GET_BLOCK_DIO, NULL,
1312 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1316 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1317 struct buffer_head *bh_result, int create)
1319 /* Block number less than F2FS MAX BLOCKS */
1320 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1323 return __get_data_block(inode, iblock, bh_result, create,
1324 F2FS_GET_BLOCK_BMAP, NULL,
1325 NO_CHECK_TYPE, create);
1328 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1330 return (offset >> inode->i_blkbits);
1333 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1335 return (blk << inode->i_blkbits);
1338 static int f2fs_xattr_fiemap(struct inode *inode,
1339 struct fiemap_extent_info *fieinfo)
1341 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1343 struct node_info ni;
1344 __u64 phys = 0, len;
1346 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1349 if (f2fs_has_inline_xattr(inode)) {
1352 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1353 inode->i_ino, false);
1357 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1359 f2fs_put_page(page, 1);
1363 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1364 offset = offsetof(struct f2fs_inode, i_addr) +
1365 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1366 get_inline_xattr_addrs(inode));
1369 len = inline_xattr_size(inode);
1371 f2fs_put_page(page, 1);
1373 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1376 flags |= FIEMAP_EXTENT_LAST;
1378 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1379 if (err || err == 1)
1384 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1388 err = f2fs_get_node_info(sbi, xnid, &ni);
1390 f2fs_put_page(page, 1);
1394 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1395 len = inode->i_sb->s_blocksize;
1397 f2fs_put_page(page, 1);
1399 flags = FIEMAP_EXTENT_LAST;
1403 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1405 return (err < 0 ? err : 0);
1408 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1411 struct buffer_head map_bh;
1412 sector_t start_blk, last_blk;
1414 u64 logical = 0, phys = 0, size = 0;
1418 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1419 ret = f2fs_precache_extents(inode);
1424 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1430 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1431 ret = f2fs_xattr_fiemap(inode, fieinfo);
1435 if (f2fs_has_inline_data(inode)) {
1436 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1441 if (logical_to_blk(inode, len) == 0)
1442 len = blk_to_logical(inode, 1);
1444 start_blk = logical_to_blk(inode, start);
1445 last_blk = logical_to_blk(inode, start + len - 1);
1448 memset(&map_bh, 0, sizeof(struct buffer_head));
1449 map_bh.b_size = len;
1451 ret = get_data_block(inode, start_blk, &map_bh, 0,
1452 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1457 if (!buffer_mapped(&map_bh)) {
1458 start_blk = next_pgofs;
1460 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1461 F2FS_I_SB(inode)->max_file_blocks))
1464 flags |= FIEMAP_EXTENT_LAST;
1468 if (f2fs_encrypted_inode(inode))
1469 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1471 ret = fiemap_fill_next_extent(fieinfo, logical,
1475 if (start_blk > last_blk || ret)
1478 logical = blk_to_logical(inode, start_blk);
1479 phys = blk_to_logical(inode, map_bh.b_blocknr);
1480 size = map_bh.b_size;
1482 if (buffer_unwritten(&map_bh))
1483 flags = FIEMAP_EXTENT_UNWRITTEN;
1485 start_blk += logical_to_blk(inode, size);
1489 if (fatal_signal_pending(current))
1497 inode_unlock(inode);
1502 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1503 * Major change was from block_size == page_size in f2fs by default.
1505 * Note that the aops->readpages() function is ONLY used for read-ahead. If
1506 * this function ever deviates from doing just read-ahead, it should either
1507 * use ->readpage() or do the necessary surgery to decouple ->readpages()
1510 static int f2fs_mpage_readpages(struct address_space *mapping,
1511 struct list_head *pages, struct page *page,
1512 unsigned nr_pages, bool is_readahead)
1514 struct bio *bio = NULL;
1515 sector_t last_block_in_bio = 0;
1516 struct inode *inode = mapping->host;
1517 const unsigned blkbits = inode->i_blkbits;
1518 const unsigned blocksize = 1 << blkbits;
1519 sector_t block_in_file;
1520 sector_t last_block;
1521 sector_t last_block_in_file;
1523 struct f2fs_map_blocks map;
1529 map.m_next_pgofs = NULL;
1530 map.m_next_extent = NULL;
1531 map.m_seg_type = NO_CHECK_TYPE;
1532 map.m_may_create = false;
1534 for (; nr_pages; nr_pages--) {
1536 page = list_last_entry(pages, struct page, lru);
1538 prefetchw(&page->flags);
1539 list_del(&page->lru);
1540 if (add_to_page_cache_lru(page, mapping,
1542 readahead_gfp_mask(mapping)))
1546 block_in_file = (sector_t)page->index;
1547 last_block = block_in_file + nr_pages;
1548 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1550 if (last_block > last_block_in_file)
1551 last_block = last_block_in_file;
1554 * Map blocks using the previous result first.
1556 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1557 block_in_file > map.m_lblk &&
1558 block_in_file < (map.m_lblk + map.m_len))
1562 * Then do more f2fs_map_blocks() calls until we are
1563 * done with this page.
1567 if (block_in_file < last_block) {
1568 map.m_lblk = block_in_file;
1569 map.m_len = last_block - block_in_file;
1571 if (f2fs_map_blocks(inode, &map, 0,
1572 F2FS_GET_BLOCK_DEFAULT))
1573 goto set_error_page;
1576 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1577 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1578 SetPageMappedToDisk(page);
1580 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1581 SetPageUptodate(page);
1585 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1587 goto set_error_page;
1589 zero_user_segment(page, 0, PAGE_SIZE);
1590 if (!PageUptodate(page))
1591 SetPageUptodate(page);
1597 * This page will go to BIO. Do we need to send this
1600 if (bio && (last_block_in_bio != block_nr - 1 ||
1601 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1603 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1607 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1608 is_readahead ? REQ_RAHEAD : 0);
1611 goto set_error_page;
1616 * If the page is under writeback, we need to wait for
1617 * its completion to see the correct decrypted data.
1619 f2fs_wait_on_block_writeback(inode, block_nr);
1621 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1622 goto submit_and_realloc;
1624 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1625 ClearPageError(page);
1626 last_block_in_bio = block_nr;
1630 zero_user_segment(page, 0, PAGE_SIZE);
1635 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1643 BUG_ON(pages && !list_empty(pages));
1645 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1649 static int f2fs_read_data_page(struct file *file, struct page *page)
1651 struct inode *inode = page->mapping->host;
1654 trace_f2fs_readpage(page, DATA);
1656 /* If the file has inline data, try to read it directly */
1657 if (f2fs_has_inline_data(inode))
1658 ret = f2fs_read_inline_data(inode, page);
1660 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1664 static int f2fs_read_data_pages(struct file *file,
1665 struct address_space *mapping,
1666 struct list_head *pages, unsigned nr_pages)
1668 struct inode *inode = mapping->host;
1669 struct page *page = list_last_entry(pages, struct page, lru);
1671 trace_f2fs_readpages(inode, page, nr_pages);
1673 /* If the file has inline data, skip readpages */
1674 if (f2fs_has_inline_data(inode))
1677 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1680 static int encrypt_one_page(struct f2fs_io_info *fio)
1682 struct inode *inode = fio->page->mapping->host;
1684 gfp_t gfp_flags = GFP_NOFS;
1686 if (!f2fs_encrypted_file(inode))
1689 /* wait for GCed page writeback via META_MAPPING */
1690 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1693 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1694 PAGE_SIZE, 0, fio->page->index, gfp_flags);
1695 if (IS_ERR(fio->encrypted_page)) {
1696 /* flush pending IOs and wait for a while in the ENOMEM case */
1697 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1698 f2fs_flush_merged_writes(fio->sbi);
1699 congestion_wait(BLK_RW_ASYNC, HZ/50);
1700 gfp_flags |= __GFP_NOFAIL;
1703 return PTR_ERR(fio->encrypted_page);
1706 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1708 if (PageUptodate(mpage))
1709 memcpy(page_address(mpage),
1710 page_address(fio->encrypted_page), PAGE_SIZE);
1711 f2fs_put_page(mpage, 1);
1716 static inline bool check_inplace_update_policy(struct inode *inode,
1717 struct f2fs_io_info *fio)
1719 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1720 unsigned int policy = SM_I(sbi)->ipu_policy;
1722 if (policy & (0x1 << F2FS_IPU_FORCE))
1724 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1726 if (policy & (0x1 << F2FS_IPU_UTIL) &&
1727 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1729 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1730 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1734 * IPU for rewrite async pages
1736 if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1737 fio && fio->op == REQ_OP_WRITE &&
1738 !(fio->op_flags & REQ_SYNC) &&
1739 !f2fs_encrypted_inode(inode))
1742 /* this is only set during fdatasync */
1743 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1744 is_inode_flag_set(inode, FI_NEED_IPU))
1747 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1748 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1754 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1756 if (f2fs_is_pinned_file(inode))
1759 /* if this is cold file, we should overwrite to avoid fragmentation */
1760 if (file_is_cold(inode))
1763 return check_inplace_update_policy(inode, fio);
1766 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1768 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1770 if (test_opt(sbi, LFS))
1772 if (S_ISDIR(inode->i_mode))
1774 if (IS_NOQUOTA(inode))
1776 if (f2fs_is_atomic_file(inode))
1779 if (is_cold_data(fio->page))
1781 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1783 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1784 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1790 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1792 struct inode *inode = fio->page->mapping->host;
1794 if (f2fs_should_update_outplace(inode, fio))
1797 return f2fs_should_update_inplace(inode, fio);
1800 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1802 struct page *page = fio->page;
1803 struct inode *inode = page->mapping->host;
1804 struct dnode_of_data dn;
1805 struct extent_info ei = {0,0,0};
1806 struct node_info ni;
1807 bool ipu_force = false;
1810 set_new_dnode(&dn, inode, NULL, NULL, 0);
1811 if (need_inplace_update(fio) &&
1812 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1813 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1815 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1820 fio->need_lock = LOCK_DONE;
1824 /* Deadlock due to between page->lock and f2fs_lock_op */
1825 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1828 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1832 fio->old_blkaddr = dn.data_blkaddr;
1834 /* This page is already truncated */
1835 if (fio->old_blkaddr == NULL_ADDR) {
1836 ClearPageUptodate(page);
1837 clear_cold_data(page);
1841 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1842 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1848 * If current allocation needs SSR,
1849 * it had better in-place writes for updated data.
1851 if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1852 need_inplace_update(fio))) {
1853 err = encrypt_one_page(fio);
1857 set_page_writeback(page);
1858 ClearPageError(page);
1859 f2fs_put_dnode(&dn);
1860 if (fio->need_lock == LOCK_REQ)
1861 f2fs_unlock_op(fio->sbi);
1862 err = f2fs_inplace_write_data(fio);
1863 if (err && PageWriteback(page))
1864 end_page_writeback(page);
1865 trace_f2fs_do_write_data_page(fio->page, IPU);
1866 set_inode_flag(inode, FI_UPDATE_WRITE);
1870 if (fio->need_lock == LOCK_RETRY) {
1871 if (!f2fs_trylock_op(fio->sbi)) {
1875 fio->need_lock = LOCK_REQ;
1878 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1882 fio->version = ni.version;
1884 err = encrypt_one_page(fio);
1888 set_page_writeback(page);
1889 ClearPageError(page);
1891 /* LFS mode write path */
1892 f2fs_outplace_write_data(&dn, fio);
1893 trace_f2fs_do_write_data_page(page, OPU);
1894 set_inode_flag(inode, FI_APPEND_WRITE);
1895 if (page->index == 0)
1896 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1898 f2fs_put_dnode(&dn);
1900 if (fio->need_lock == LOCK_REQ)
1901 f2fs_unlock_op(fio->sbi);
1905 static int __write_data_page(struct page *page, bool *submitted,
1906 struct writeback_control *wbc,
1907 enum iostat_type io_type)
1909 struct inode *inode = page->mapping->host;
1910 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1911 loff_t i_size = i_size_read(inode);
1912 const pgoff_t end_index = ((unsigned long long) i_size)
1914 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1915 unsigned offset = 0;
1916 bool need_balance_fs = false;
1918 struct f2fs_io_info fio = {
1920 .ino = inode->i_ino,
1923 .op_flags = wbc_to_write_flags(wbc),
1924 .old_blkaddr = NULL_ADDR,
1926 .encrypted_page = NULL,
1928 .need_lock = LOCK_RETRY,
1933 trace_f2fs_writepage(page, DATA);
1935 /* we should bypass data pages to proceed the kworkder jobs */
1936 if (unlikely(f2fs_cp_error(sbi))) {
1937 mapping_set_error(page->mapping, -EIO);
1939 * don't drop any dirty dentry pages for keeping lastest
1940 * directory structure.
1942 if (S_ISDIR(inode->i_mode))
1947 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1950 if (page->index < end_index)
1954 * If the offset is out-of-range of file size,
1955 * this page does not have to be written to disk.
1957 offset = i_size & (PAGE_SIZE - 1);
1958 if ((page->index >= end_index + 1) || !offset)
1961 zero_user_segment(page, offset, PAGE_SIZE);
1963 if (f2fs_is_drop_cache(inode))
1965 /* we should not write 0'th page having journal header */
1966 if (f2fs_is_volatile_file(inode) && (!page->index ||
1967 (!wbc->for_reclaim &&
1968 f2fs_available_free_memory(sbi, BASE_CHECK))))
1971 /* Dentry blocks are controlled by checkpoint */
1972 if (S_ISDIR(inode->i_mode)) {
1973 fio.need_lock = LOCK_DONE;
1974 err = f2fs_do_write_data_page(&fio);
1978 if (!wbc->for_reclaim)
1979 need_balance_fs = true;
1980 else if (has_not_enough_free_secs(sbi, 0, 0))
1983 set_inode_flag(inode, FI_HOT_DATA);
1986 if (f2fs_has_inline_data(inode)) {
1987 err = f2fs_write_inline_data(inode, page);
1992 if (err == -EAGAIN) {
1993 err = f2fs_do_write_data_page(&fio);
1994 if (err == -EAGAIN) {
1995 fio.need_lock = LOCK_REQ;
1996 err = f2fs_do_write_data_page(&fio);
2001 file_set_keep_isize(inode);
2003 down_write(&F2FS_I(inode)->i_sem);
2004 if (F2FS_I(inode)->last_disk_size < psize)
2005 F2FS_I(inode)->last_disk_size = psize;
2006 up_write(&F2FS_I(inode)->i_sem);
2010 if (err && err != -ENOENT)
2014 inode_dec_dirty_pages(inode);
2016 ClearPageUptodate(page);
2017 clear_cold_data(page);
2020 if (wbc->for_reclaim) {
2021 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2022 clear_inode_flag(inode, FI_HOT_DATA);
2023 f2fs_remove_dirty_inode(inode);
2028 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode))
2029 f2fs_balance_fs(sbi, need_balance_fs);
2031 if (unlikely(f2fs_cp_error(sbi))) {
2032 f2fs_submit_merged_write(sbi, DATA);
2037 *submitted = fio.submitted;
2042 redirty_page_for_writepage(wbc, page);
2044 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2045 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2046 * file_write_and_wait_range() will see EIO error, which is critical
2047 * to return value of fsync() followed by atomic_write failure to user.
2049 if (!err || wbc->for_reclaim)
2050 return AOP_WRITEPAGE_ACTIVATE;
2055 static int f2fs_write_data_page(struct page *page,
2056 struct writeback_control *wbc)
2058 return __write_data_page(page, NULL, wbc, FS_DATA_IO);
2062 * This function was copied from write_cche_pages from mm/page-writeback.c.
2063 * The major change is making write step of cold data page separately from
2064 * warm/hot data page.
2066 static int f2fs_write_cache_pages(struct address_space *mapping,
2067 struct writeback_control *wbc,
2068 enum iostat_type io_type)
2072 struct pagevec pvec;
2073 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2075 pgoff_t uninitialized_var(writeback_index);
2077 pgoff_t end; /* Inclusive */
2080 int range_whole = 0;
2084 pagevec_init(&pvec);
2086 if (get_dirty_pages(mapping->host) <=
2087 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2088 set_inode_flag(mapping->host, FI_HOT_DATA);
2090 clear_inode_flag(mapping->host, FI_HOT_DATA);
2092 if (wbc->range_cyclic) {
2093 writeback_index = mapping->writeback_index; /* prev offset */
2094 index = writeback_index;
2101 index = wbc->range_start >> PAGE_SHIFT;
2102 end = wbc->range_end >> PAGE_SHIFT;
2103 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2105 cycled = 1; /* ignore range_cyclic tests */
2107 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2108 tag = PAGECACHE_TAG_TOWRITE;
2110 tag = PAGECACHE_TAG_DIRTY;
2112 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2113 tag_pages_for_writeback(mapping, index, end);
2115 while (!done && (index <= end)) {
2118 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2123 for (i = 0; i < nr_pages; i++) {
2124 struct page *page = pvec.pages[i];
2125 bool submitted = false;
2127 /* give a priority to WB_SYNC threads */
2128 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2129 wbc->sync_mode == WB_SYNC_NONE) {
2134 done_index = page->index;
2138 if (unlikely(page->mapping != mapping)) {
2144 if (!PageDirty(page)) {
2145 /* someone wrote it for us */
2146 goto continue_unlock;
2149 if (PageWriteback(page)) {
2150 if (wbc->sync_mode != WB_SYNC_NONE)
2151 f2fs_wait_on_page_writeback(page,
2154 goto continue_unlock;
2157 if (!clear_page_dirty_for_io(page))
2158 goto continue_unlock;
2160 ret = __write_data_page(page, &submitted, wbc, io_type);
2161 if (unlikely(ret)) {
2163 * keep nr_to_write, since vfs uses this to
2164 * get # of written pages.
2166 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2170 } else if (ret == -EAGAIN) {
2172 if (wbc->sync_mode == WB_SYNC_ALL) {
2174 congestion_wait(BLK_RW_ASYNC,
2180 done_index = page->index + 1;
2183 } else if (submitted) {
2187 if (--wbc->nr_to_write <= 0 &&
2188 wbc->sync_mode == WB_SYNC_NONE) {
2193 pagevec_release(&pvec);
2197 if (!cycled && !done) {
2200 end = writeback_index - 1;
2203 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2204 mapping->writeback_index = done_index;
2207 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2213 static inline bool __should_serialize_io(struct inode *inode,
2214 struct writeback_control *wbc)
2216 if (!S_ISREG(inode->i_mode))
2218 if (IS_NOQUOTA(inode))
2220 if (wbc->sync_mode != WB_SYNC_ALL)
2222 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2227 static int __f2fs_write_data_pages(struct address_space *mapping,
2228 struct writeback_control *wbc,
2229 enum iostat_type io_type)
2231 struct inode *inode = mapping->host;
2232 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2233 struct blk_plug plug;
2235 bool locked = false;
2237 /* deal with chardevs and other special file */
2238 if (!mapping->a_ops->writepage)
2241 /* skip writing if there is no dirty page in this inode */
2242 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2245 /* during POR, we don't need to trigger writepage at all. */
2246 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2249 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2250 wbc->sync_mode == WB_SYNC_NONE &&
2251 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2252 f2fs_available_free_memory(sbi, DIRTY_DENTS))
2255 /* skip writing during file defragment */
2256 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2259 trace_f2fs_writepages(mapping->host, wbc, DATA);
2261 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2262 if (wbc->sync_mode == WB_SYNC_ALL)
2263 atomic_inc(&sbi->wb_sync_req[DATA]);
2264 else if (atomic_read(&sbi->wb_sync_req[DATA]))
2267 if (__should_serialize_io(inode, wbc)) {
2268 mutex_lock(&sbi->writepages);
2272 blk_start_plug(&plug);
2273 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2274 blk_finish_plug(&plug);
2277 mutex_unlock(&sbi->writepages);
2279 if (wbc->sync_mode == WB_SYNC_ALL)
2280 atomic_dec(&sbi->wb_sync_req[DATA]);
2282 * if some pages were truncated, we cannot guarantee its mapping->host
2283 * to detect pending bios.
2286 f2fs_remove_dirty_inode(inode);
2290 wbc->pages_skipped += get_dirty_pages(inode);
2291 trace_f2fs_writepages(mapping->host, wbc, DATA);
2295 static int f2fs_write_data_pages(struct address_space *mapping,
2296 struct writeback_control *wbc)
2298 struct inode *inode = mapping->host;
2300 return __f2fs_write_data_pages(mapping, wbc,
2301 F2FS_I(inode)->cp_task == current ?
2302 FS_CP_DATA_IO : FS_DATA_IO);
2305 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2307 struct inode *inode = mapping->host;
2308 loff_t i_size = i_size_read(inode);
2311 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2312 down_write(&F2FS_I(inode)->i_mmap_sem);
2314 truncate_pagecache(inode, i_size);
2315 f2fs_truncate_blocks(inode, i_size, true, true);
2317 up_write(&F2FS_I(inode)->i_mmap_sem);
2318 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2322 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2323 struct page *page, loff_t pos, unsigned len,
2324 block_t *blk_addr, bool *node_changed)
2326 struct inode *inode = page->mapping->host;
2327 pgoff_t index = page->index;
2328 struct dnode_of_data dn;
2330 bool locked = false;
2331 struct extent_info ei = {0,0,0};
2336 * we already allocated all the blocks, so we don't need to get
2337 * the block addresses when there is no need to fill the page.
2339 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2340 !is_inode_flag_set(inode, FI_NO_PREALLOC))
2343 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2344 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2345 flag = F2FS_GET_BLOCK_DEFAULT;
2347 flag = F2FS_GET_BLOCK_PRE_AIO;
2349 if (f2fs_has_inline_data(inode) ||
2350 (pos & PAGE_MASK) >= i_size_read(inode)) {
2351 __do_map_lock(sbi, flag, true);
2355 /* check inline_data */
2356 ipage = f2fs_get_node_page(sbi, inode->i_ino);
2357 if (IS_ERR(ipage)) {
2358 err = PTR_ERR(ipage);
2362 set_new_dnode(&dn, inode, ipage, ipage, 0);
2364 if (f2fs_has_inline_data(inode)) {
2365 if (pos + len <= MAX_INLINE_DATA(inode)) {
2366 f2fs_do_read_inline_data(page, ipage);
2367 set_inode_flag(inode, FI_DATA_EXIST);
2369 set_inline_node(ipage);
2371 err = f2fs_convert_inline_page(&dn, page);
2374 if (dn.data_blkaddr == NULL_ADDR)
2375 err = f2fs_get_block(&dn, index);
2377 } else if (locked) {
2378 err = f2fs_get_block(&dn, index);
2380 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2381 dn.data_blkaddr = ei.blk + index - ei.fofs;
2384 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2385 if (err || dn.data_blkaddr == NULL_ADDR) {
2386 f2fs_put_dnode(&dn);
2387 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2389 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2396 /* convert_inline_page can make node_changed */
2397 *blk_addr = dn.data_blkaddr;
2398 *node_changed = dn.node_changed;
2400 f2fs_put_dnode(&dn);
2403 __do_map_lock(sbi, flag, false);
2407 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2408 loff_t pos, unsigned len, unsigned flags,
2409 struct page **pagep, void **fsdata)
2411 struct inode *inode = mapping->host;
2412 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2413 struct page *page = NULL;
2414 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2415 bool need_balance = false, drop_atomic = false;
2416 block_t blkaddr = NULL_ADDR;
2419 trace_f2fs_write_begin(inode, pos, len, flags);
2421 err = f2fs_is_checkpoint_ready(sbi);
2425 if ((f2fs_is_atomic_file(inode) &&
2426 !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2427 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2434 * We should check this at this moment to avoid deadlock on inode page
2435 * and #0 page. The locking rule for inline_data conversion should be:
2436 * lock_page(page #0) -> lock_page(inode_page)
2439 err = f2fs_convert_inline_inode(inode);
2445 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2446 * wait_for_stable_page. Will wait that below with our IO control.
2448 page = f2fs_pagecache_get_page(mapping, index,
2449 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2457 err = prepare_write_begin(sbi, page, pos, len,
2458 &blkaddr, &need_balance);
2462 if (need_balance && !IS_NOQUOTA(inode) &&
2463 has_not_enough_free_secs(sbi, 0, 0)) {
2465 f2fs_balance_fs(sbi, true);
2467 if (page->mapping != mapping) {
2468 /* The page got truncated from under us */
2469 f2fs_put_page(page, 1);
2474 f2fs_wait_on_page_writeback(page, DATA, false, true);
2476 if (len == PAGE_SIZE || PageUptodate(page))
2479 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2480 zero_user_segment(page, len, PAGE_SIZE);
2484 if (blkaddr == NEW_ADDR) {
2485 zero_user_segment(page, 0, PAGE_SIZE);
2486 SetPageUptodate(page);
2488 err = f2fs_submit_page_read(inode, page, blkaddr);
2493 if (unlikely(page->mapping != mapping)) {
2494 f2fs_put_page(page, 1);
2497 if (unlikely(!PageUptodate(page))) {
2505 f2fs_put_page(page, 1);
2506 f2fs_write_failed(mapping, pos + len);
2508 f2fs_drop_inmem_pages_all(sbi, false);
2512 static int f2fs_write_end(struct file *file,
2513 struct address_space *mapping,
2514 loff_t pos, unsigned len, unsigned copied,
2515 struct page *page, void *fsdata)
2517 struct inode *inode = page->mapping->host;
2519 trace_f2fs_write_end(inode, pos, len, copied);
2522 * This should be come from len == PAGE_SIZE, and we expect copied
2523 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2524 * let generic_perform_write() try to copy data again through copied=0.
2526 if (!PageUptodate(page)) {
2527 if (unlikely(copied != len))
2530 SetPageUptodate(page);
2535 set_page_dirty(page);
2537 if (pos + copied > i_size_read(inode))
2538 f2fs_i_size_write(inode, pos + copied);
2540 f2fs_put_page(page, 1);
2541 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2545 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2548 unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2549 unsigned blkbits = i_blkbits;
2550 unsigned blocksize_mask = (1 << blkbits) - 1;
2551 unsigned long align = offset | iov_iter_alignment(iter);
2552 struct block_device *bdev = inode->i_sb->s_bdev;
2554 if (align & blocksize_mask) {
2556 blkbits = blksize_bits(bdev_logical_block_size(bdev));
2557 blocksize_mask = (1 << blkbits) - 1;
2558 if (align & blocksize_mask)
2565 static void f2fs_dio_end_io(struct bio *bio)
2567 struct f2fs_private_dio *dio = bio->bi_private;
2569 dec_page_count(F2FS_I_SB(dio->inode),
2570 dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2572 bio->bi_private = dio->orig_private;
2573 bio->bi_end_io = dio->orig_end_io;
2580 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2583 struct f2fs_private_dio *dio;
2584 bool write = (bio_op(bio) == REQ_OP_WRITE);
2587 dio = f2fs_kzalloc(F2FS_I_SB(inode),
2588 sizeof(struct f2fs_private_dio), GFP_NOFS);
2595 dio->orig_end_io = bio->bi_end_io;
2596 dio->orig_private = bio->bi_private;
2599 bio->bi_end_io = f2fs_dio_end_io;
2600 bio->bi_private = dio;
2602 inc_page_count(F2FS_I_SB(inode),
2603 write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2608 bio->bi_status = BLK_STS_IOERR;
2612 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2614 struct address_space *mapping = iocb->ki_filp->f_mapping;
2615 struct inode *inode = mapping->host;
2616 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2617 struct f2fs_inode_info *fi = F2FS_I(inode);
2618 size_t count = iov_iter_count(iter);
2619 loff_t offset = iocb->ki_pos;
2620 int rw = iov_iter_rw(iter);
2622 enum rw_hint hint = iocb->ki_hint;
2623 int whint_mode = F2FS_OPTION(sbi).whint_mode;
2626 err = check_direct_IO(inode, iter, offset);
2628 return err < 0 ? err : 0;
2630 if (f2fs_force_buffered_io(inode, iocb, iter))
2633 do_opu = allow_outplace_dio(inode, iocb, iter);
2635 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2637 if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2638 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2640 if (iocb->ki_flags & IOCB_NOWAIT) {
2641 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2642 iocb->ki_hint = hint;
2646 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2647 up_read(&fi->i_gc_rwsem[rw]);
2648 iocb->ki_hint = hint;
2653 down_read(&fi->i_gc_rwsem[rw]);
2655 down_read(&fi->i_gc_rwsem[READ]);
2658 err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2659 iter, rw == WRITE ? get_data_block_dio_write :
2660 get_data_block_dio, NULL, f2fs_dio_submit_bio,
2661 DIO_LOCKING | DIO_SKIP_HOLES);
2664 up_read(&fi->i_gc_rwsem[READ]);
2666 up_read(&fi->i_gc_rwsem[rw]);
2669 if (whint_mode == WHINT_MODE_OFF)
2670 iocb->ki_hint = hint;
2672 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2675 set_inode_flag(inode, FI_UPDATE_WRITE);
2676 } else if (err < 0) {
2677 f2fs_write_failed(mapping, offset + count);
2682 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2687 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2688 unsigned int length)
2690 struct inode *inode = page->mapping->host;
2691 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2693 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2694 (offset % PAGE_SIZE || length != PAGE_SIZE))
2697 if (PageDirty(page)) {
2698 if (inode->i_ino == F2FS_META_INO(sbi)) {
2699 dec_page_count(sbi, F2FS_DIRTY_META);
2700 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2701 dec_page_count(sbi, F2FS_DIRTY_NODES);
2703 inode_dec_dirty_pages(inode);
2704 f2fs_remove_dirty_inode(inode);
2708 clear_cold_data(page);
2710 /* This is atomic written page, keep Private */
2711 if (IS_ATOMIC_WRITTEN_PAGE(page))
2712 return f2fs_drop_inmem_page(inode, page);
2714 set_page_private(page, 0);
2715 ClearPagePrivate(page);
2718 int f2fs_release_page(struct page *page, gfp_t wait)
2720 /* If this is dirty page, keep PagePrivate */
2721 if (PageDirty(page))
2724 /* This is atomic written page, keep Private */
2725 if (IS_ATOMIC_WRITTEN_PAGE(page))
2728 clear_cold_data(page);
2729 set_page_private(page, 0);
2730 ClearPagePrivate(page);
2734 static int f2fs_set_data_page_dirty(struct page *page)
2736 struct address_space *mapping = page->mapping;
2737 struct inode *inode = mapping->host;
2739 trace_f2fs_set_page_dirty(page, DATA);
2741 if (!PageUptodate(page))
2742 SetPageUptodate(page);
2744 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2745 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2746 f2fs_register_inmem_page(inode, page);
2750 * Previously, this page has been registered, we just
2756 if (!PageDirty(page)) {
2757 __set_page_dirty_nobuffers(page);
2758 f2fs_update_dirty_page(inode, page);
2764 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2766 struct inode *inode = mapping->host;
2768 if (f2fs_has_inline_data(inode))
2771 /* make sure allocating whole blocks */
2772 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2773 filemap_write_and_wait(mapping);
2775 return generic_block_bmap(mapping, block, get_data_block_bmap);
2778 #ifdef CONFIG_MIGRATION
2779 #include <linux/migrate.h>
2781 int f2fs_migrate_page(struct address_space *mapping,
2782 struct page *newpage, struct page *page, enum migrate_mode mode)
2784 int rc, extra_count;
2785 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2786 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2788 BUG_ON(PageWriteback(page));
2790 /* migrating an atomic written page is safe with the inmem_lock hold */
2791 if (atomic_written) {
2792 if (mode != MIGRATE_SYNC)
2794 if (!mutex_trylock(&fi->inmem_lock))
2799 * A reference is expected if PagePrivate set when move mapping,
2800 * however F2FS breaks this for maintaining dirty page counts when
2801 * truncating pages. So here adjusting the 'extra_count' make it work.
2803 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2804 rc = migrate_page_move_mapping(mapping, newpage,
2805 page, mode, extra_count);
2806 if (rc != MIGRATEPAGE_SUCCESS) {
2808 mutex_unlock(&fi->inmem_lock);
2812 if (atomic_written) {
2813 struct inmem_pages *cur;
2814 list_for_each_entry(cur, &fi->inmem_pages, list)
2815 if (cur->page == page) {
2816 cur->page = newpage;
2819 mutex_unlock(&fi->inmem_lock);
2824 if (PagePrivate(page))
2825 SetPagePrivate(newpage);
2826 set_page_private(newpage, page_private(page));
2828 if (mode != MIGRATE_SYNC_NO_COPY)
2829 migrate_page_copy(newpage, page);
2831 migrate_page_states(newpage, page);
2833 return MIGRATEPAGE_SUCCESS;
2837 const struct address_space_operations f2fs_dblock_aops = {
2838 .readpage = f2fs_read_data_page,
2839 .readpages = f2fs_read_data_pages,
2840 .writepage = f2fs_write_data_page,
2841 .writepages = f2fs_write_data_pages,
2842 .write_begin = f2fs_write_begin,
2843 .write_end = f2fs_write_end,
2844 .set_page_dirty = f2fs_set_data_page_dirty,
2845 .invalidatepage = f2fs_invalidate_page,
2846 .releasepage = f2fs_release_page,
2847 .direct_IO = f2fs_direct_IO,
2849 #ifdef CONFIG_MIGRATION
2850 .migratepage = f2fs_migrate_page,
2854 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2856 struct address_space *mapping = page_mapping(page);
2857 unsigned long flags;
2859 xa_lock_irqsave(&mapping->i_pages, flags);
2860 __xa_clear_mark(&mapping->i_pages, page_index(page),
2861 PAGECACHE_TAG_DIRTY);
2862 xa_unlock_irqrestore(&mapping->i_pages, flags);
2865 int __init f2fs_init_post_read_processing(void)
2867 bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2868 if (!bio_post_read_ctx_cache)
2870 bio_post_read_ctx_pool =
2871 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2872 bio_post_read_ctx_cache);
2873 if (!bio_post_read_ctx_pool)
2874 goto fail_free_cache;
2878 kmem_cache_destroy(bio_post_read_ctx_cache);
2883 void __exit f2fs_destroy_post_read_processing(void)
2885 mempool_destroy(bio_post_read_ctx_pool);
2886 kmem_cache_destroy(bio_post_read_ctx_cache);