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/swap.h>
18 #include <linux/prefetch.h>
19 #include <linux/uio.h>
20 #include <linux/cleancache.h>
21 #include <linux/sched/signal.h>
27 #include <trace/events/f2fs.h>
29 #define NUM_PREALLOC_POST_READ_CTXS 128
31 static struct kmem_cache *bio_post_read_ctx_cache;
32 static struct kmem_cache *bio_entry_slab;
33 static mempool_t *bio_post_read_ctx_pool;
35 static bool __is_cp_guaranteed(struct page *page)
37 struct address_space *mapping = page->mapping;
39 struct f2fs_sb_info *sbi;
44 inode = mapping->host;
45 sbi = F2FS_I_SB(inode);
47 if (inode->i_ino == F2FS_META_INO(sbi) ||
48 inode->i_ino == F2FS_NODE_INO(sbi) ||
49 S_ISDIR(inode->i_mode) ||
50 (S_ISREG(inode->i_mode) &&
51 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
57 static enum count_type __read_io_type(struct page *page)
59 struct address_space *mapping = page_file_mapping(page);
62 struct inode *inode = mapping->host;
63 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
65 if (inode->i_ino == F2FS_META_INO(sbi))
68 if (inode->i_ino == F2FS_NODE_INO(sbi))
74 /* postprocessing steps for read bios */
75 enum bio_post_read_step {
81 struct bio_post_read_ctx {
83 struct work_struct work;
84 unsigned int cur_step;
85 unsigned int enabled_steps;
88 static void __read_end_io(struct bio *bio)
92 struct bvec_iter_all iter_all;
94 bio_for_each_segment_all(bv, bio, iter_all) {
97 /* PG_error was set if any post_read step failed */
98 if (bio->bi_status || PageError(page)) {
99 ClearPageUptodate(page);
100 /* will re-read again later */
101 ClearPageError(page);
103 SetPageUptodate(page);
105 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
109 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
113 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
115 static void decrypt_work(struct work_struct *work)
117 struct bio_post_read_ctx *ctx =
118 container_of(work, struct bio_post_read_ctx, work);
120 fscrypt_decrypt_bio(ctx->bio);
122 bio_post_read_processing(ctx);
125 static void verity_work(struct work_struct *work)
127 struct bio_post_read_ctx *ctx =
128 container_of(work, struct bio_post_read_ctx, work);
130 fsverity_verify_bio(ctx->bio);
132 bio_post_read_processing(ctx);
135 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
138 * We use different work queues for decryption and for verity because
139 * verity may require reading metadata pages that need decryption, and
140 * we shouldn't recurse to the same workqueue.
142 switch (++ctx->cur_step) {
144 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
145 INIT_WORK(&ctx->work, decrypt_work);
146 fscrypt_enqueue_decrypt_work(&ctx->work);
152 if (ctx->enabled_steps & (1 << STEP_VERITY)) {
153 INIT_WORK(&ctx->work, verity_work);
154 fsverity_enqueue_verify_work(&ctx->work);
160 __read_end_io(ctx->bio);
164 static bool f2fs_bio_post_read_required(struct bio *bio)
166 return bio->bi_private && !bio->bi_status;
169 static void f2fs_read_end_io(struct bio *bio)
171 struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
173 if (time_to_inject(sbi, FAULT_READ_IO)) {
174 f2fs_show_injection_info(sbi, FAULT_READ_IO);
175 bio->bi_status = BLK_STS_IOERR;
178 if (f2fs_bio_post_read_required(bio)) {
179 struct bio_post_read_ctx *ctx = bio->bi_private;
181 ctx->cur_step = STEP_INITIAL;
182 bio_post_read_processing(ctx);
189 static void f2fs_write_end_io(struct bio *bio)
191 struct f2fs_sb_info *sbi = bio->bi_private;
192 struct bio_vec *bvec;
193 struct bvec_iter_all iter_all;
195 if (time_to_inject(sbi, FAULT_WRITE_IO)) {
196 f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
197 bio->bi_status = BLK_STS_IOERR;
200 bio_for_each_segment_all(bvec, bio, iter_all) {
201 struct page *page = bvec->bv_page;
202 enum count_type type = WB_DATA_TYPE(page);
204 if (IS_DUMMY_WRITTEN_PAGE(page)) {
205 set_page_private(page, (unsigned long)NULL);
206 ClearPagePrivate(page);
208 mempool_free(page, sbi->write_io_dummy);
210 if (unlikely(bio->bi_status))
211 f2fs_stop_checkpoint(sbi, true);
215 fscrypt_finalize_bounce_page(&page);
217 if (unlikely(bio->bi_status)) {
218 mapping_set_error(page->mapping, -EIO);
219 if (type == F2FS_WB_CP_DATA)
220 f2fs_stop_checkpoint(sbi, true);
223 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
224 page->index != nid_of_node(page));
226 dec_page_count(sbi, type);
227 if (f2fs_in_warm_node_list(sbi, page))
228 f2fs_del_fsync_node_entry(sbi, page);
229 clear_cold_data(page);
230 end_page_writeback(page);
232 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
233 wq_has_sleeper(&sbi->cp_wait))
234 wake_up(&sbi->cp_wait);
240 * Return true, if pre_bio's bdev is same as its target device.
242 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
243 block_t blk_addr, struct bio *bio)
245 struct block_device *bdev = sbi->sb->s_bdev;
248 if (f2fs_is_multi_device(sbi)) {
249 for (i = 0; i < sbi->s_ndevs; i++) {
250 if (FDEV(i).start_blk <= blk_addr &&
251 FDEV(i).end_blk >= blk_addr) {
252 blk_addr -= FDEV(i).start_blk;
259 bio_set_dev(bio, bdev);
260 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
265 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
269 if (!f2fs_is_multi_device(sbi))
272 for (i = 0; i < sbi->s_ndevs; i++)
273 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
278 static bool __same_bdev(struct f2fs_sb_info *sbi,
279 block_t blk_addr, struct bio *bio)
281 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
282 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
286 * Low-level block read/write IO operations.
288 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
290 struct f2fs_sb_info *sbi = fio->sbi;
293 bio = f2fs_bio_alloc(sbi, npages, true);
295 f2fs_target_device(sbi, fio->new_blkaddr, bio);
296 if (is_read_io(fio->op)) {
297 bio->bi_end_io = f2fs_read_end_io;
298 bio->bi_private = NULL;
300 bio->bi_end_io = f2fs_write_end_io;
301 bio->bi_private = sbi;
302 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
303 fio->type, fio->temp);
306 wbc_init_bio(fio->io_wbc, bio);
311 static inline void __submit_bio(struct f2fs_sb_info *sbi,
312 struct bio *bio, enum page_type type)
314 if (!is_read_io(bio_op(bio))) {
317 if (type != DATA && type != NODE)
320 if (test_opt(sbi, LFS) && current->plug)
321 blk_finish_plug(current->plug);
323 if (F2FS_IO_ALIGNED(sbi))
326 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
327 start %= F2FS_IO_SIZE(sbi);
332 /* fill dummy pages */
333 for (; start < F2FS_IO_SIZE(sbi); start++) {
335 mempool_alloc(sbi->write_io_dummy,
336 GFP_NOIO | __GFP_NOFAIL);
337 f2fs_bug_on(sbi, !page);
339 zero_user_segment(page, 0, PAGE_SIZE);
340 SetPagePrivate(page);
341 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
343 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
347 * In the NODE case, we lose next block address chain. So, we
348 * need to do checkpoint in f2fs_sync_file.
351 set_sbi_flag(sbi, SBI_NEED_CP);
354 if (is_read_io(bio_op(bio)))
355 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
357 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
361 static void __submit_merged_bio(struct f2fs_bio_info *io)
363 struct f2fs_io_info *fio = &io->fio;
368 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
370 if (is_read_io(fio->op))
371 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
373 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
375 __submit_bio(io->sbi, io->bio, fio->type);
379 static bool __has_merged_page(struct bio *bio, struct inode *inode,
380 struct page *page, nid_t ino)
382 struct bio_vec *bvec;
384 struct bvec_iter_all iter_all;
389 if (!inode && !page && !ino)
392 bio_for_each_segment_all(bvec, bio, iter_all) {
394 target = bvec->bv_page;
395 if (fscrypt_is_bounce_page(target))
396 target = fscrypt_pagecache_page(target);
398 if (inode && inode == target->mapping->host)
400 if (page && page == target)
402 if (ino && ino == ino_of_node(target))
409 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
410 enum page_type type, enum temp_type temp)
412 enum page_type btype = PAGE_TYPE_OF_BIO(type);
413 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
415 down_write(&io->io_rwsem);
417 /* change META to META_FLUSH in the checkpoint procedure */
418 if (type >= META_FLUSH) {
419 io->fio.type = META_FLUSH;
420 io->fio.op = REQ_OP_WRITE;
421 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
422 if (!test_opt(sbi, NOBARRIER))
423 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
425 __submit_merged_bio(io);
426 up_write(&io->io_rwsem);
429 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
430 struct inode *inode, struct page *page,
431 nid_t ino, enum page_type type, bool force)
436 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
438 enum page_type btype = PAGE_TYPE_OF_BIO(type);
439 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
441 down_read(&io->io_rwsem);
442 ret = __has_merged_page(io->bio, inode, page, ino);
443 up_read(&io->io_rwsem);
446 __f2fs_submit_merged_write(sbi, type, temp);
448 /* TODO: use HOT temp only for meta pages now. */
454 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
456 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
459 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
460 struct inode *inode, struct page *page,
461 nid_t ino, enum page_type type)
463 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
466 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
468 f2fs_submit_merged_write(sbi, DATA);
469 f2fs_submit_merged_write(sbi, NODE);
470 f2fs_submit_merged_write(sbi, META);
474 * Fill the locked page with data located in the block address.
475 * A caller needs to unlock the page on failure.
477 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
480 struct page *page = fio->encrypted_page ?
481 fio->encrypted_page : fio->page;
483 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
484 fio->is_por ? META_POR : (__is_meta_io(fio) ?
485 META_GENERIC : DATA_GENERIC_ENHANCE)))
486 return -EFSCORRUPTED;
488 trace_f2fs_submit_page_bio(page, fio);
489 f2fs_trace_ios(fio, 0);
491 /* Allocate a new bio */
492 bio = __bio_alloc(fio, 1);
494 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
499 if (fio->io_wbc && !is_read_io(fio->op))
500 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
502 bio_set_op_attrs(bio, fio->op, fio->op_flags);
504 inc_page_count(fio->sbi, is_read_io(fio->op) ?
505 __read_io_type(page): WB_DATA_TYPE(fio->page));
507 __submit_bio(fio->sbi, bio, fio->type);
511 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
512 block_t last_blkaddr, block_t cur_blkaddr)
514 if (last_blkaddr + 1 != cur_blkaddr)
516 return __same_bdev(sbi, cur_blkaddr, bio);
519 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
520 struct f2fs_io_info *fio)
522 if (io->fio.op != fio->op)
524 return io->fio.op_flags == fio->op_flags;
527 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
528 struct f2fs_bio_info *io,
529 struct f2fs_io_info *fio,
530 block_t last_blkaddr,
533 if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
534 unsigned int filled_blocks =
535 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
536 unsigned int io_size = F2FS_IO_SIZE(sbi);
537 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
539 /* IOs in bio is aligned and left space of vectors is not enough */
540 if (!(filled_blocks % io_size) && left_vecs < io_size)
543 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
545 return io_type_is_mergeable(io, fio);
548 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
549 struct page *page, enum temp_type temp)
551 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
552 struct bio_entry *be;
554 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
558 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
561 down_write(&io->bio_list_lock);
562 list_add_tail(&be->list, &io->bio_list);
563 up_write(&io->bio_list_lock);
566 static void del_bio_entry(struct bio_entry *be)
569 kmem_cache_free(bio_entry_slab, be);
572 static int add_ipu_page(struct f2fs_sb_info *sbi, struct bio **bio,
579 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
580 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
581 struct list_head *head = &io->bio_list;
582 struct bio_entry *be;
584 down_write(&io->bio_list_lock);
585 list_for_each_entry(be, head, list) {
591 if (bio_add_page(*bio, page, PAGE_SIZE, 0) == PAGE_SIZE) {
598 __submit_bio(sbi, *bio, DATA);
601 up_write(&io->bio_list_lock);
612 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
613 struct bio **bio, struct page *page)
617 struct bio *target = bio ? *bio : NULL;
619 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
620 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
621 struct list_head *head = &io->bio_list;
622 struct bio_entry *be;
624 if (list_empty(head))
627 down_read(&io->bio_list_lock);
628 list_for_each_entry(be, head, list) {
630 found = (target == be->bio);
632 found = __has_merged_page(be->bio, NULL,
637 up_read(&io->bio_list_lock);
644 down_write(&io->bio_list_lock);
645 list_for_each_entry(be, head, list) {
647 found = (target == be->bio);
649 found = __has_merged_page(be->bio, NULL,
657 up_write(&io->bio_list_lock);
661 __submit_bio(sbi, target, DATA);
668 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
670 struct bio *bio = *fio->bio;
671 struct page *page = fio->encrypted_page ?
672 fio->encrypted_page : fio->page;
674 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
675 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
676 return -EFSCORRUPTED;
678 trace_f2fs_submit_page_bio(page, fio);
679 f2fs_trace_ios(fio, 0);
681 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
683 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
686 bio = __bio_alloc(fio, BIO_MAX_PAGES);
687 bio_set_op_attrs(bio, fio->op, fio->op_flags);
689 add_bio_entry(fio->sbi, bio, page, fio->temp);
691 if (add_ipu_page(fio->sbi, &bio, page))
696 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
698 inc_page_count(fio->sbi, WB_DATA_TYPE(page));
700 *fio->last_block = fio->new_blkaddr;
706 void f2fs_submit_page_write(struct f2fs_io_info *fio)
708 struct f2fs_sb_info *sbi = fio->sbi;
709 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
710 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
711 struct page *bio_page;
713 f2fs_bug_on(sbi, is_read_io(fio->op));
715 down_write(&io->io_rwsem);
718 spin_lock(&io->io_lock);
719 if (list_empty(&io->io_list)) {
720 spin_unlock(&io->io_lock);
723 fio = list_first_entry(&io->io_list,
724 struct f2fs_io_info, list);
725 list_del(&fio->list);
726 spin_unlock(&io->io_lock);
729 verify_fio_blkaddr(fio);
731 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
733 /* set submitted = true as a return value */
734 fio->submitted = true;
736 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
738 if (io->bio && !io_is_mergeable(sbi, io->bio, io, fio,
739 io->last_block_in_bio, fio->new_blkaddr))
740 __submit_merged_bio(io);
742 if (io->bio == NULL) {
743 if (F2FS_IO_ALIGNED(sbi) &&
744 (fio->type == DATA || fio->type == NODE) &&
745 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
746 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
750 io->bio = __bio_alloc(fio, BIO_MAX_PAGES);
754 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
755 __submit_merged_bio(io);
760 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
762 io->last_block_in_bio = fio->new_blkaddr;
763 f2fs_trace_ios(fio, 0);
765 trace_f2fs_submit_page_write(fio->page, fio);
770 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
771 !f2fs_is_checkpoint_ready(sbi))
772 __submit_merged_bio(io);
773 up_write(&io->io_rwsem);
776 static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx)
778 return fsverity_active(inode) &&
779 idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
782 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
783 unsigned nr_pages, unsigned op_flag,
786 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
788 struct bio_post_read_ctx *ctx;
789 unsigned int post_read_steps = 0;
791 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
793 return ERR_PTR(-ENOMEM);
794 f2fs_target_device(sbi, blkaddr, bio);
795 bio->bi_end_io = f2fs_read_end_io;
796 bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
798 if (f2fs_encrypted_file(inode))
799 post_read_steps |= 1 << STEP_DECRYPT;
801 if (f2fs_need_verity(inode, first_idx))
802 post_read_steps |= 1 << STEP_VERITY;
804 if (post_read_steps) {
805 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
808 return ERR_PTR(-ENOMEM);
811 ctx->enabled_steps = post_read_steps;
812 bio->bi_private = ctx;
818 /* This can handle encryption stuffs */
819 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
822 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
825 bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0, page->index);
829 /* wait for GCed page writeback via META_MAPPING */
830 f2fs_wait_on_block_writeback(inode, blkaddr);
832 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
836 ClearPageError(page);
837 inc_page_count(sbi, F2FS_RD_DATA);
838 __submit_bio(sbi, bio, DATA);
842 static void __set_data_blkaddr(struct dnode_of_data *dn)
844 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
848 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
849 base = get_extra_isize(dn->inode);
851 /* Get physical address of data block */
852 addr_array = blkaddr_in_node(rn);
853 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
857 * Lock ordering for the change of data block address:
860 * update block addresses in the node page
862 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
864 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
865 __set_data_blkaddr(dn);
866 if (set_page_dirty(dn->node_page))
867 dn->node_changed = true;
870 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
872 dn->data_blkaddr = blkaddr;
873 f2fs_set_data_blkaddr(dn);
874 f2fs_update_extent_cache(dn);
877 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
878 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
880 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
886 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
888 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
891 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
892 dn->ofs_in_node, count);
894 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
896 for (; count > 0; dn->ofs_in_node++) {
897 block_t blkaddr = datablock_addr(dn->inode,
898 dn->node_page, dn->ofs_in_node);
899 if (blkaddr == NULL_ADDR) {
900 dn->data_blkaddr = NEW_ADDR;
901 __set_data_blkaddr(dn);
906 if (set_page_dirty(dn->node_page))
907 dn->node_changed = true;
911 /* Should keep dn->ofs_in_node unchanged */
912 int f2fs_reserve_new_block(struct dnode_of_data *dn)
914 unsigned int ofs_in_node = dn->ofs_in_node;
917 ret = f2fs_reserve_new_blocks(dn, 1);
918 dn->ofs_in_node = ofs_in_node;
922 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
924 bool need_put = dn->inode_page ? false : true;
927 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
931 if (dn->data_blkaddr == NULL_ADDR)
932 err = f2fs_reserve_new_block(dn);
938 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
940 struct extent_info ei = {0,0,0};
941 struct inode *inode = dn->inode;
943 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
944 dn->data_blkaddr = ei.blk + index - ei.fofs;
948 return f2fs_reserve_block(dn, index);
951 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
952 int op_flags, bool for_write)
954 struct address_space *mapping = inode->i_mapping;
955 struct dnode_of_data dn;
957 struct extent_info ei = {0,0,0};
960 page = f2fs_grab_cache_page(mapping, index, for_write);
962 return ERR_PTR(-ENOMEM);
964 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
965 dn.data_blkaddr = ei.blk + index - ei.fofs;
966 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
967 DATA_GENERIC_ENHANCE_READ)) {
974 set_new_dnode(&dn, inode, NULL, NULL, 0);
975 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
980 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
984 if (dn.data_blkaddr != NEW_ADDR &&
985 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
987 DATA_GENERIC_ENHANCE)) {
992 if (PageUptodate(page)) {
998 * A new dentry page is allocated but not able to be written, since its
999 * new inode page couldn't be allocated due to -ENOSPC.
1000 * In such the case, its blkaddr can be remained as NEW_ADDR.
1001 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1002 * f2fs_init_inode_metadata.
1004 if (dn.data_blkaddr == NEW_ADDR) {
1005 zero_user_segment(page, 0, PAGE_SIZE);
1006 if (!PageUptodate(page))
1007 SetPageUptodate(page);
1012 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
1018 f2fs_put_page(page, 1);
1019 return ERR_PTR(err);
1022 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1024 struct address_space *mapping = inode->i_mapping;
1027 page = find_get_page(mapping, index);
1028 if (page && PageUptodate(page))
1030 f2fs_put_page(page, 0);
1032 page = f2fs_get_read_data_page(inode, index, 0, false);
1036 if (PageUptodate(page))
1039 wait_on_page_locked(page);
1040 if (unlikely(!PageUptodate(page))) {
1041 f2fs_put_page(page, 0);
1042 return ERR_PTR(-EIO);
1048 * If it tries to access a hole, return an error.
1049 * Because, the callers, functions in dir.c and GC, should be able to know
1050 * whether this page exists or not.
1052 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1055 struct address_space *mapping = inode->i_mapping;
1058 page = f2fs_get_read_data_page(inode, index, 0, for_write);
1062 /* wait for read completion */
1064 if (unlikely(page->mapping != mapping)) {
1065 f2fs_put_page(page, 1);
1068 if (unlikely(!PageUptodate(page))) {
1069 f2fs_put_page(page, 1);
1070 return ERR_PTR(-EIO);
1076 * Caller ensures that this data page is never allocated.
1077 * A new zero-filled data page is allocated in the page cache.
1079 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1081 * Note that, ipage is set only by make_empty_dir, and if any error occur,
1082 * ipage should be released by this function.
1084 struct page *f2fs_get_new_data_page(struct inode *inode,
1085 struct page *ipage, pgoff_t index, bool new_i_size)
1087 struct address_space *mapping = inode->i_mapping;
1089 struct dnode_of_data dn;
1092 page = f2fs_grab_cache_page(mapping, index, true);
1095 * before exiting, we should make sure ipage will be released
1096 * if any error occur.
1098 f2fs_put_page(ipage, 1);
1099 return ERR_PTR(-ENOMEM);
1102 set_new_dnode(&dn, inode, ipage, NULL, 0);
1103 err = f2fs_reserve_block(&dn, index);
1105 f2fs_put_page(page, 1);
1106 return ERR_PTR(err);
1109 f2fs_put_dnode(&dn);
1111 if (PageUptodate(page))
1114 if (dn.data_blkaddr == NEW_ADDR) {
1115 zero_user_segment(page, 0, PAGE_SIZE);
1116 if (!PageUptodate(page))
1117 SetPageUptodate(page);
1119 f2fs_put_page(page, 1);
1121 /* if ipage exists, blkaddr should be NEW_ADDR */
1122 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1123 page = f2fs_get_lock_data_page(inode, index, true);
1128 if (new_i_size && i_size_read(inode) <
1129 ((loff_t)(index + 1) << PAGE_SHIFT))
1130 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1134 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1136 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1137 struct f2fs_summary sum;
1138 struct node_info ni;
1139 block_t old_blkaddr;
1143 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1146 err = f2fs_get_node_info(sbi, dn->nid, &ni);
1150 dn->data_blkaddr = datablock_addr(dn->inode,
1151 dn->node_page, dn->ofs_in_node);
1152 if (dn->data_blkaddr != NULL_ADDR)
1155 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1159 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1160 old_blkaddr = dn->data_blkaddr;
1161 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1162 &sum, seg_type, NULL, false);
1163 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1164 invalidate_mapping_pages(META_MAPPING(sbi),
1165 old_blkaddr, old_blkaddr);
1166 f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1169 * i_size will be updated by direct_IO. Otherwise, we'll get stale
1170 * data from unwritten block via dio_read.
1175 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1177 struct inode *inode = file_inode(iocb->ki_filp);
1178 struct f2fs_map_blocks map;
1181 bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1183 /* convert inline data for Direct I/O*/
1185 err = f2fs_convert_inline_inode(inode);
1190 if (direct_io && allow_outplace_dio(inode, iocb, from))
1193 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
1196 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1197 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1198 if (map.m_len > map.m_lblk)
1199 map.m_len -= map.m_lblk;
1203 map.m_next_pgofs = NULL;
1204 map.m_next_extent = NULL;
1205 map.m_seg_type = NO_CHECK_TYPE;
1206 map.m_may_create = true;
1209 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1210 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1211 F2FS_GET_BLOCK_PRE_AIO :
1212 F2FS_GET_BLOCK_PRE_DIO;
1215 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1216 err = f2fs_convert_inline_inode(inode);
1220 if (f2fs_has_inline_data(inode))
1223 flag = F2FS_GET_BLOCK_PRE_AIO;
1226 err = f2fs_map_blocks(inode, &map, 1, flag);
1227 if (map.m_len > 0 && err == -ENOSPC) {
1229 set_inode_flag(inode, FI_NO_PREALLOC);
1235 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1237 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1239 down_read(&sbi->node_change);
1241 up_read(&sbi->node_change);
1246 f2fs_unlock_op(sbi);
1251 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1252 * f2fs_map_blocks structure.
1253 * If original data blocks are allocated, then give them to blockdev.
1255 * a. preallocate requested block addresses
1256 * b. do not use extent cache for better performance
1257 * c. give the block addresses to blockdev
1259 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1260 int create, int flag)
1262 unsigned int maxblocks = map->m_len;
1263 struct dnode_of_data dn;
1264 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1265 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1266 pgoff_t pgofs, end_offset, end;
1267 int err = 0, ofs = 1;
1268 unsigned int ofs_in_node, last_ofs_in_node;
1270 struct extent_info ei = {0,0,0};
1272 unsigned int start_pgofs;
1280 /* it only supports block size == page size */
1281 pgofs = (pgoff_t)map->m_lblk;
1282 end = pgofs + maxblocks;
1284 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1285 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1289 map->m_pblk = ei.blk + pgofs - ei.fofs;
1290 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1291 map->m_flags = F2FS_MAP_MAPPED;
1292 if (map->m_next_extent)
1293 *map->m_next_extent = pgofs + map->m_len;
1295 /* for hardware encryption, but to avoid potential issue in future */
1296 if (flag == F2FS_GET_BLOCK_DIO)
1297 f2fs_wait_on_block_writeback_range(inode,
1298 map->m_pblk, map->m_len);
1303 if (map->m_may_create)
1304 __do_map_lock(sbi, flag, true);
1306 /* When reading holes, we need its node page */
1307 set_new_dnode(&dn, inode, NULL, NULL, 0);
1308 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1310 if (flag == F2FS_GET_BLOCK_BMAP)
1312 if (err == -ENOENT) {
1314 if (map->m_next_pgofs)
1315 *map->m_next_pgofs =
1316 f2fs_get_next_page_offset(&dn, pgofs);
1317 if (map->m_next_extent)
1318 *map->m_next_extent =
1319 f2fs_get_next_page_offset(&dn, pgofs);
1324 start_pgofs = pgofs;
1326 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1327 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1330 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1332 if (__is_valid_data_blkaddr(blkaddr) &&
1333 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1334 err = -EFSCORRUPTED;
1338 if (__is_valid_data_blkaddr(blkaddr)) {
1339 /* use out-place-update for driect IO under LFS mode */
1340 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1341 map->m_may_create) {
1342 err = __allocate_data_block(&dn, map->m_seg_type);
1345 blkaddr = dn.data_blkaddr;
1346 set_inode_flag(inode, FI_APPEND_WRITE);
1350 if (unlikely(f2fs_cp_error(sbi))) {
1354 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1355 if (blkaddr == NULL_ADDR) {
1357 last_ofs_in_node = dn.ofs_in_node;
1360 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1361 flag != F2FS_GET_BLOCK_DIO);
1362 err = __allocate_data_block(&dn,
1365 set_inode_flag(inode, FI_APPEND_WRITE);
1369 map->m_flags |= F2FS_MAP_NEW;
1370 blkaddr = dn.data_blkaddr;
1372 if (flag == F2FS_GET_BLOCK_BMAP) {
1376 if (flag == F2FS_GET_BLOCK_PRECACHE)
1378 if (flag == F2FS_GET_BLOCK_FIEMAP &&
1379 blkaddr == NULL_ADDR) {
1380 if (map->m_next_pgofs)
1381 *map->m_next_pgofs = pgofs + 1;
1384 if (flag != F2FS_GET_BLOCK_FIEMAP) {
1385 /* for defragment case */
1386 if (map->m_next_pgofs)
1387 *map->m_next_pgofs = pgofs + 1;
1393 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1396 if (map->m_len == 0) {
1397 /* preallocated unwritten block should be mapped for fiemap. */
1398 if (blkaddr == NEW_ADDR)
1399 map->m_flags |= F2FS_MAP_UNWRITTEN;
1400 map->m_flags |= F2FS_MAP_MAPPED;
1402 map->m_pblk = blkaddr;
1404 } else if ((map->m_pblk != NEW_ADDR &&
1405 blkaddr == (map->m_pblk + ofs)) ||
1406 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1407 flag == F2FS_GET_BLOCK_PRE_DIO) {
1418 /* preallocate blocks in batch for one dnode page */
1419 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1420 (pgofs == end || dn.ofs_in_node == end_offset)) {
1422 dn.ofs_in_node = ofs_in_node;
1423 err = f2fs_reserve_new_blocks(&dn, prealloc);
1427 map->m_len += dn.ofs_in_node - ofs_in_node;
1428 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1432 dn.ofs_in_node = end_offset;
1437 else if (dn.ofs_in_node < end_offset)
1440 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1441 if (map->m_flags & F2FS_MAP_MAPPED) {
1442 unsigned int ofs = start_pgofs - map->m_lblk;
1444 f2fs_update_extent_cache_range(&dn,
1445 start_pgofs, map->m_pblk + ofs,
1450 f2fs_put_dnode(&dn);
1452 if (map->m_may_create) {
1453 __do_map_lock(sbi, flag, false);
1454 f2fs_balance_fs(sbi, dn.node_changed);
1460 /* for hardware encryption, but to avoid potential issue in future */
1461 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1462 f2fs_wait_on_block_writeback_range(inode,
1463 map->m_pblk, map->m_len);
1465 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1466 if (map->m_flags & F2FS_MAP_MAPPED) {
1467 unsigned int ofs = start_pgofs - map->m_lblk;
1469 f2fs_update_extent_cache_range(&dn,
1470 start_pgofs, map->m_pblk + ofs,
1473 if (map->m_next_extent)
1474 *map->m_next_extent = pgofs + 1;
1476 f2fs_put_dnode(&dn);
1478 if (map->m_may_create) {
1479 __do_map_lock(sbi, flag, false);
1480 f2fs_balance_fs(sbi, dn.node_changed);
1483 trace_f2fs_map_blocks(inode, map, err);
1487 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1489 struct f2fs_map_blocks map;
1493 if (pos + len > i_size_read(inode))
1496 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1497 map.m_next_pgofs = NULL;
1498 map.m_next_extent = NULL;
1499 map.m_seg_type = NO_CHECK_TYPE;
1500 map.m_may_create = false;
1501 last_lblk = F2FS_BLK_ALIGN(pos + len);
1503 while (map.m_lblk < last_lblk) {
1504 map.m_len = last_lblk - map.m_lblk;
1505 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1506 if (err || map.m_len == 0)
1508 map.m_lblk += map.m_len;
1513 static int __get_data_block(struct inode *inode, sector_t iblock,
1514 struct buffer_head *bh, int create, int flag,
1515 pgoff_t *next_pgofs, int seg_type, bool may_write)
1517 struct f2fs_map_blocks map;
1520 map.m_lblk = iblock;
1521 map.m_len = bh->b_size >> inode->i_blkbits;
1522 map.m_next_pgofs = next_pgofs;
1523 map.m_next_extent = NULL;
1524 map.m_seg_type = seg_type;
1525 map.m_may_create = may_write;
1527 err = f2fs_map_blocks(inode, &map, create, flag);
1529 map_bh(bh, inode->i_sb, map.m_pblk);
1530 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1531 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1536 static int get_data_block(struct inode *inode, sector_t iblock,
1537 struct buffer_head *bh_result, int create, int flag,
1538 pgoff_t *next_pgofs)
1540 return __get_data_block(inode, iblock, bh_result, create,
1542 NO_CHECK_TYPE, create);
1545 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1546 struct buffer_head *bh_result, int create)
1548 return __get_data_block(inode, iblock, bh_result, create,
1549 F2FS_GET_BLOCK_DIO, NULL,
1550 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1551 IS_SWAPFILE(inode) ? false : true);
1554 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1555 struct buffer_head *bh_result, int create)
1557 return __get_data_block(inode, iblock, bh_result, create,
1558 F2FS_GET_BLOCK_DIO, NULL,
1559 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1563 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1564 struct buffer_head *bh_result, int create)
1566 /* Block number less than F2FS MAX BLOCKS */
1567 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1570 return __get_data_block(inode, iblock, bh_result, create,
1571 F2FS_GET_BLOCK_BMAP, NULL,
1572 NO_CHECK_TYPE, create);
1575 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1577 return (offset >> inode->i_blkbits);
1580 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1582 return (blk << inode->i_blkbits);
1585 static int f2fs_xattr_fiemap(struct inode *inode,
1586 struct fiemap_extent_info *fieinfo)
1588 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1590 struct node_info ni;
1591 __u64 phys = 0, len;
1593 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1596 if (f2fs_has_inline_xattr(inode)) {
1599 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1600 inode->i_ino, false);
1604 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1606 f2fs_put_page(page, 1);
1610 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1611 offset = offsetof(struct f2fs_inode, i_addr) +
1612 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1613 get_inline_xattr_addrs(inode));
1616 len = inline_xattr_size(inode);
1618 f2fs_put_page(page, 1);
1620 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1623 flags |= FIEMAP_EXTENT_LAST;
1625 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1626 if (err || err == 1)
1631 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1635 err = f2fs_get_node_info(sbi, xnid, &ni);
1637 f2fs_put_page(page, 1);
1641 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1642 len = inode->i_sb->s_blocksize;
1644 f2fs_put_page(page, 1);
1646 flags = FIEMAP_EXTENT_LAST;
1650 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1652 return (err < 0 ? err : 0);
1655 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1658 struct buffer_head map_bh;
1659 sector_t start_blk, last_blk;
1661 u64 logical = 0, phys = 0, size = 0;
1665 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1666 ret = f2fs_precache_extents(inode);
1671 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1677 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1678 ret = f2fs_xattr_fiemap(inode, fieinfo);
1682 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1683 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1688 if (logical_to_blk(inode, len) == 0)
1689 len = blk_to_logical(inode, 1);
1691 start_blk = logical_to_blk(inode, start);
1692 last_blk = logical_to_blk(inode, start + len - 1);
1695 memset(&map_bh, 0, sizeof(struct buffer_head));
1696 map_bh.b_size = len;
1698 ret = get_data_block(inode, start_blk, &map_bh, 0,
1699 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1704 if (!buffer_mapped(&map_bh)) {
1705 start_blk = next_pgofs;
1707 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1708 F2FS_I_SB(inode)->max_file_blocks))
1711 flags |= FIEMAP_EXTENT_LAST;
1715 if (IS_ENCRYPTED(inode))
1716 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1718 ret = fiemap_fill_next_extent(fieinfo, logical,
1722 if (start_blk > last_blk || ret)
1725 logical = blk_to_logical(inode, start_blk);
1726 phys = blk_to_logical(inode, map_bh.b_blocknr);
1727 size = map_bh.b_size;
1729 if (buffer_unwritten(&map_bh))
1730 flags = FIEMAP_EXTENT_UNWRITTEN;
1732 start_blk += logical_to_blk(inode, size);
1736 if (fatal_signal_pending(current))
1744 inode_unlock(inode);
1748 static inline loff_t f2fs_readpage_limit(struct inode *inode)
1750 if (IS_ENABLED(CONFIG_FS_VERITY) &&
1751 (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
1752 return inode->i_sb->s_maxbytes;
1754 return i_size_read(inode);
1757 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1759 struct f2fs_map_blocks *map,
1760 struct bio **bio_ret,
1761 sector_t *last_block_in_bio,
1764 struct bio *bio = *bio_ret;
1765 const unsigned blkbits = inode->i_blkbits;
1766 const unsigned blocksize = 1 << blkbits;
1767 sector_t block_in_file;
1768 sector_t last_block;
1769 sector_t last_block_in_file;
1773 block_in_file = (sector_t)page_index(page);
1774 last_block = block_in_file + nr_pages;
1775 last_block_in_file = (f2fs_readpage_limit(inode) + blocksize - 1) >>
1777 if (last_block > last_block_in_file)
1778 last_block = last_block_in_file;
1780 /* just zeroing out page which is beyond EOF */
1781 if (block_in_file >= last_block)
1784 * Map blocks using the previous result first.
1786 if ((map->m_flags & F2FS_MAP_MAPPED) &&
1787 block_in_file > map->m_lblk &&
1788 block_in_file < (map->m_lblk + map->m_len))
1792 * Then do more f2fs_map_blocks() calls until we are
1793 * done with this page.
1795 map->m_lblk = block_in_file;
1796 map->m_len = last_block - block_in_file;
1798 ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
1802 if ((map->m_flags & F2FS_MAP_MAPPED)) {
1803 block_nr = map->m_pblk + block_in_file - map->m_lblk;
1804 SetPageMappedToDisk(page);
1806 if (!PageUptodate(page) && (!PageSwapCache(page) &&
1807 !cleancache_get_page(page))) {
1808 SetPageUptodate(page);
1812 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1813 DATA_GENERIC_ENHANCE_READ)) {
1814 ret = -EFSCORRUPTED;
1819 zero_user_segment(page, 0, PAGE_SIZE);
1820 if (f2fs_need_verity(inode, page->index) &&
1821 !fsverity_verify_page(page)) {
1825 if (!PageUptodate(page))
1826 SetPageUptodate(page);
1832 * This page will go to BIO. Do we need to send this
1835 if (bio && !page_is_mergeable(F2FS_I_SB(inode), bio,
1836 *last_block_in_bio, block_nr)) {
1838 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1842 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1843 is_readahead ? REQ_RAHEAD : 0, page->index);
1852 * If the page is under writeback, we need to wait for
1853 * its completion to see the correct decrypted data.
1855 f2fs_wait_on_block_writeback(inode, block_nr);
1857 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1858 goto submit_and_realloc;
1860 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1861 ClearPageError(page);
1862 *last_block_in_bio = block_nr;
1866 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1876 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1877 * Major change was from block_size == page_size in f2fs by default.
1879 * Note that the aops->readpages() function is ONLY used for read-ahead. If
1880 * this function ever deviates from doing just read-ahead, it should either
1881 * use ->readpage() or do the necessary surgery to decouple ->readpages()
1884 int f2fs_mpage_readpages(struct address_space *mapping,
1885 struct list_head *pages, struct page *page,
1886 unsigned nr_pages, bool is_readahead)
1888 struct bio *bio = NULL;
1889 sector_t last_block_in_bio = 0;
1890 struct inode *inode = mapping->host;
1891 struct f2fs_map_blocks map;
1898 map.m_next_pgofs = NULL;
1899 map.m_next_extent = NULL;
1900 map.m_seg_type = NO_CHECK_TYPE;
1901 map.m_may_create = false;
1903 for (; nr_pages; nr_pages--) {
1905 page = list_last_entry(pages, struct page, lru);
1907 prefetchw(&page->flags);
1908 list_del(&page->lru);
1909 if (add_to_page_cache_lru(page, mapping,
1911 readahead_gfp_mask(mapping)))
1915 ret = f2fs_read_single_page(inode, page, nr_pages, &map, &bio,
1916 &last_block_in_bio, is_readahead);
1919 zero_user_segment(page, 0, PAGE_SIZE);
1926 BUG_ON(pages && !list_empty(pages));
1928 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1929 return pages ? 0 : ret;
1932 static int f2fs_read_data_page(struct file *file, struct page *page)
1934 struct inode *inode = page_file_mapping(page)->host;
1937 trace_f2fs_readpage(page, DATA);
1939 /* If the file has inline data, try to read it directly */
1940 if (f2fs_has_inline_data(inode))
1941 ret = f2fs_read_inline_data(inode, page);
1943 ret = f2fs_mpage_readpages(page_file_mapping(page),
1944 NULL, page, 1, false);
1948 static int f2fs_read_data_pages(struct file *file,
1949 struct address_space *mapping,
1950 struct list_head *pages, unsigned nr_pages)
1952 struct inode *inode = mapping->host;
1953 struct page *page = list_last_entry(pages, struct page, lru);
1955 trace_f2fs_readpages(inode, page, nr_pages);
1957 /* If the file has inline data, skip readpages */
1958 if (f2fs_has_inline_data(inode))
1961 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1964 static int encrypt_one_page(struct f2fs_io_info *fio)
1966 struct inode *inode = fio->page->mapping->host;
1968 gfp_t gfp_flags = GFP_NOFS;
1970 if (!f2fs_encrypted_file(inode))
1973 /* wait for GCed page writeback via META_MAPPING */
1974 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1977 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(fio->page,
1980 if (IS_ERR(fio->encrypted_page)) {
1981 /* flush pending IOs and wait for a while in the ENOMEM case */
1982 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1983 f2fs_flush_merged_writes(fio->sbi);
1984 congestion_wait(BLK_RW_ASYNC, HZ/50);
1985 gfp_flags |= __GFP_NOFAIL;
1988 return PTR_ERR(fio->encrypted_page);
1991 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1993 if (PageUptodate(mpage))
1994 memcpy(page_address(mpage),
1995 page_address(fio->encrypted_page), PAGE_SIZE);
1996 f2fs_put_page(mpage, 1);
2001 static inline bool check_inplace_update_policy(struct inode *inode,
2002 struct f2fs_io_info *fio)
2004 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2005 unsigned int policy = SM_I(sbi)->ipu_policy;
2007 if (policy & (0x1 << F2FS_IPU_FORCE))
2009 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2011 if (policy & (0x1 << F2FS_IPU_UTIL) &&
2012 utilization(sbi) > SM_I(sbi)->min_ipu_util)
2014 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2015 utilization(sbi) > SM_I(sbi)->min_ipu_util)
2019 * IPU for rewrite async pages
2021 if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2022 fio && fio->op == REQ_OP_WRITE &&
2023 !(fio->op_flags & REQ_SYNC) &&
2024 !IS_ENCRYPTED(inode))
2027 /* this is only set during fdatasync */
2028 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2029 is_inode_flag_set(inode, FI_NEED_IPU))
2032 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2033 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2039 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2041 if (f2fs_is_pinned_file(inode))
2044 /* if this is cold file, we should overwrite to avoid fragmentation */
2045 if (file_is_cold(inode))
2048 return check_inplace_update_policy(inode, fio);
2051 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2053 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2055 if (test_opt(sbi, LFS))
2057 if (S_ISDIR(inode->i_mode))
2059 if (IS_NOQUOTA(inode))
2061 if (f2fs_is_atomic_file(inode))
2064 if (is_cold_data(fio->page))
2066 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
2068 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2069 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2075 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2077 struct inode *inode = fio->page->mapping->host;
2079 if (f2fs_should_update_outplace(inode, fio))
2082 return f2fs_should_update_inplace(inode, fio);
2085 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2087 struct page *page = fio->page;
2088 struct inode *inode = page->mapping->host;
2089 struct dnode_of_data dn;
2090 struct extent_info ei = {0,0,0};
2091 struct node_info ni;
2092 bool ipu_force = false;
2095 set_new_dnode(&dn, inode, NULL, NULL, 0);
2096 if (need_inplace_update(fio) &&
2097 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2098 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2100 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2101 DATA_GENERIC_ENHANCE))
2102 return -EFSCORRUPTED;
2105 fio->need_lock = LOCK_DONE;
2109 /* Deadlock due to between page->lock and f2fs_lock_op */
2110 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2113 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2117 fio->old_blkaddr = dn.data_blkaddr;
2119 /* This page is already truncated */
2120 if (fio->old_blkaddr == NULL_ADDR) {
2121 ClearPageUptodate(page);
2122 clear_cold_data(page);
2126 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2127 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2128 DATA_GENERIC_ENHANCE)) {
2129 err = -EFSCORRUPTED;
2133 * If current allocation needs SSR,
2134 * it had better in-place writes for updated data.
2137 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2138 need_inplace_update(fio))) {
2139 err = encrypt_one_page(fio);
2143 set_page_writeback(page);
2144 ClearPageError(page);
2145 f2fs_put_dnode(&dn);
2146 if (fio->need_lock == LOCK_REQ)
2147 f2fs_unlock_op(fio->sbi);
2148 err = f2fs_inplace_write_data(fio);
2150 if (f2fs_encrypted_file(inode))
2151 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2152 if (PageWriteback(page))
2153 end_page_writeback(page);
2155 set_inode_flag(inode, FI_UPDATE_WRITE);
2157 trace_f2fs_do_write_data_page(fio->page, IPU);
2161 if (fio->need_lock == LOCK_RETRY) {
2162 if (!f2fs_trylock_op(fio->sbi)) {
2166 fio->need_lock = LOCK_REQ;
2169 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
2173 fio->version = ni.version;
2175 err = encrypt_one_page(fio);
2179 set_page_writeback(page);
2180 ClearPageError(page);
2182 /* LFS mode write path */
2183 f2fs_outplace_write_data(&dn, fio);
2184 trace_f2fs_do_write_data_page(page, OPU);
2185 set_inode_flag(inode, FI_APPEND_WRITE);
2186 if (page->index == 0)
2187 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2189 f2fs_put_dnode(&dn);
2191 if (fio->need_lock == LOCK_REQ)
2192 f2fs_unlock_op(fio->sbi);
2196 static int __write_data_page(struct page *page, bool *submitted,
2198 sector_t *last_block,
2199 struct writeback_control *wbc,
2200 enum iostat_type io_type)
2202 struct inode *inode = page->mapping->host;
2203 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2204 loff_t i_size = i_size_read(inode);
2205 const pgoff_t end_index = ((unsigned long long) i_size)
2207 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2208 unsigned offset = 0;
2209 bool need_balance_fs = false;
2211 struct f2fs_io_info fio = {
2213 .ino = inode->i_ino,
2216 .op_flags = wbc_to_write_flags(wbc),
2217 .old_blkaddr = NULL_ADDR,
2219 .encrypted_page = NULL,
2221 .need_lock = LOCK_RETRY,
2225 .last_block = last_block,
2228 trace_f2fs_writepage(page, DATA);
2230 /* we should bypass data pages to proceed the kworkder jobs */
2231 if (unlikely(f2fs_cp_error(sbi))) {
2232 mapping_set_error(page->mapping, -EIO);
2234 * don't drop any dirty dentry pages for keeping lastest
2235 * directory structure.
2237 if (S_ISDIR(inode->i_mode))
2242 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2245 if (page->index < end_index || f2fs_verity_in_progress(inode))
2249 * If the offset is out-of-range of file size,
2250 * this page does not have to be written to disk.
2252 offset = i_size & (PAGE_SIZE - 1);
2253 if ((page->index >= end_index + 1) || !offset)
2256 zero_user_segment(page, offset, PAGE_SIZE);
2258 if (f2fs_is_drop_cache(inode))
2260 /* we should not write 0'th page having journal header */
2261 if (f2fs_is_volatile_file(inode) && (!page->index ||
2262 (!wbc->for_reclaim &&
2263 f2fs_available_free_memory(sbi, BASE_CHECK))))
2266 /* Dentry blocks are controlled by checkpoint */
2267 if (S_ISDIR(inode->i_mode)) {
2268 fio.need_lock = LOCK_DONE;
2269 err = f2fs_do_write_data_page(&fio);
2273 if (!wbc->for_reclaim)
2274 need_balance_fs = true;
2275 else if (has_not_enough_free_secs(sbi, 0, 0))
2278 set_inode_flag(inode, FI_HOT_DATA);
2281 if (f2fs_has_inline_data(inode)) {
2282 err = f2fs_write_inline_data(inode, page);
2287 if (err == -EAGAIN) {
2288 err = f2fs_do_write_data_page(&fio);
2289 if (err == -EAGAIN) {
2290 fio.need_lock = LOCK_REQ;
2291 err = f2fs_do_write_data_page(&fio);
2296 file_set_keep_isize(inode);
2298 down_write(&F2FS_I(inode)->i_sem);
2299 if (F2FS_I(inode)->last_disk_size < psize)
2300 F2FS_I(inode)->last_disk_size = psize;
2301 up_write(&F2FS_I(inode)->i_sem);
2305 if (err && err != -ENOENT)
2309 inode_dec_dirty_pages(inode);
2311 ClearPageUptodate(page);
2312 clear_cold_data(page);
2315 if (wbc->for_reclaim) {
2316 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2317 clear_inode_flag(inode, FI_HOT_DATA);
2318 f2fs_remove_dirty_inode(inode);
2323 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2324 !F2FS_I(inode)->cp_task)
2325 f2fs_balance_fs(sbi, need_balance_fs);
2327 if (unlikely(f2fs_cp_error(sbi))) {
2328 f2fs_submit_merged_write(sbi, DATA);
2329 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2334 *submitted = fio.submitted;
2339 redirty_page_for_writepage(wbc, page);
2341 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2342 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2343 * file_write_and_wait_range() will see EIO error, which is critical
2344 * to return value of fsync() followed by atomic_write failure to user.
2346 if (!err || wbc->for_reclaim)
2347 return AOP_WRITEPAGE_ACTIVATE;
2352 static int f2fs_write_data_page(struct page *page,
2353 struct writeback_control *wbc)
2355 return __write_data_page(page, NULL, NULL, NULL, wbc, FS_DATA_IO);
2359 * This function was copied from write_cche_pages from mm/page-writeback.c.
2360 * The major change is making write step of cold data page separately from
2361 * warm/hot data page.
2363 static int f2fs_write_cache_pages(struct address_space *mapping,
2364 struct writeback_control *wbc,
2365 enum iostat_type io_type)
2369 struct pagevec pvec;
2370 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2371 struct bio *bio = NULL;
2372 sector_t last_block;
2374 pgoff_t uninitialized_var(writeback_index);
2376 pgoff_t end; /* Inclusive */
2379 int range_whole = 0;
2383 pagevec_init(&pvec);
2385 if (get_dirty_pages(mapping->host) <=
2386 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2387 set_inode_flag(mapping->host, FI_HOT_DATA);
2389 clear_inode_flag(mapping->host, FI_HOT_DATA);
2391 if (wbc->range_cyclic) {
2392 writeback_index = mapping->writeback_index; /* prev offset */
2393 index = writeback_index;
2400 index = wbc->range_start >> PAGE_SHIFT;
2401 end = wbc->range_end >> PAGE_SHIFT;
2402 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2404 cycled = 1; /* ignore range_cyclic tests */
2406 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2407 tag = PAGECACHE_TAG_TOWRITE;
2409 tag = PAGECACHE_TAG_DIRTY;
2411 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2412 tag_pages_for_writeback(mapping, index, end);
2414 while (!done && (index <= end)) {
2417 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2422 for (i = 0; i < nr_pages; i++) {
2423 struct page *page = pvec.pages[i];
2424 bool submitted = false;
2426 /* give a priority to WB_SYNC threads */
2427 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2428 wbc->sync_mode == WB_SYNC_NONE) {
2433 done_index = page->index;
2437 if (unlikely(page->mapping != mapping)) {
2443 if (!PageDirty(page)) {
2444 /* someone wrote it for us */
2445 goto continue_unlock;
2448 if (PageWriteback(page)) {
2449 if (wbc->sync_mode != WB_SYNC_NONE)
2450 f2fs_wait_on_page_writeback(page,
2453 goto continue_unlock;
2456 if (!clear_page_dirty_for_io(page))
2457 goto continue_unlock;
2459 ret = __write_data_page(page, &submitted, &bio,
2460 &last_block, wbc, io_type);
2461 if (unlikely(ret)) {
2463 * keep nr_to_write, since vfs uses this to
2464 * get # of written pages.
2466 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2470 } else if (ret == -EAGAIN) {
2472 if (wbc->sync_mode == WB_SYNC_ALL) {
2474 congestion_wait(BLK_RW_ASYNC,
2480 done_index = page->index + 1;
2483 } else if (submitted) {
2487 if (--wbc->nr_to_write <= 0 &&
2488 wbc->sync_mode == WB_SYNC_NONE) {
2493 pagevec_release(&pvec);
2497 if (!cycled && !done) {
2500 end = writeback_index - 1;
2503 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2504 mapping->writeback_index = done_index;
2507 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2509 /* submit cached bio of IPU write */
2511 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
2516 static inline bool __should_serialize_io(struct inode *inode,
2517 struct writeback_control *wbc)
2519 if (!S_ISREG(inode->i_mode))
2521 if (IS_NOQUOTA(inode))
2523 /* to avoid deadlock in path of data flush */
2524 if (F2FS_I(inode)->cp_task)
2526 if (wbc->sync_mode != WB_SYNC_ALL)
2528 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2533 static int __f2fs_write_data_pages(struct address_space *mapping,
2534 struct writeback_control *wbc,
2535 enum iostat_type io_type)
2537 struct inode *inode = mapping->host;
2538 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2539 struct blk_plug plug;
2541 bool locked = false;
2543 /* deal with chardevs and other special file */
2544 if (!mapping->a_ops->writepage)
2547 /* skip writing if there is no dirty page in this inode */
2548 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2551 /* during POR, we don't need to trigger writepage at all. */
2552 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2555 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2556 wbc->sync_mode == WB_SYNC_NONE &&
2557 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2558 f2fs_available_free_memory(sbi, DIRTY_DENTS))
2561 /* skip writing during file defragment */
2562 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2565 trace_f2fs_writepages(mapping->host, wbc, DATA);
2567 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2568 if (wbc->sync_mode == WB_SYNC_ALL)
2569 atomic_inc(&sbi->wb_sync_req[DATA]);
2570 else if (atomic_read(&sbi->wb_sync_req[DATA]))
2573 if (__should_serialize_io(inode, wbc)) {
2574 mutex_lock(&sbi->writepages);
2578 blk_start_plug(&plug);
2579 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2580 blk_finish_plug(&plug);
2583 mutex_unlock(&sbi->writepages);
2585 if (wbc->sync_mode == WB_SYNC_ALL)
2586 atomic_dec(&sbi->wb_sync_req[DATA]);
2588 * if some pages were truncated, we cannot guarantee its mapping->host
2589 * to detect pending bios.
2592 f2fs_remove_dirty_inode(inode);
2596 wbc->pages_skipped += get_dirty_pages(inode);
2597 trace_f2fs_writepages(mapping->host, wbc, DATA);
2601 static int f2fs_write_data_pages(struct address_space *mapping,
2602 struct writeback_control *wbc)
2604 struct inode *inode = mapping->host;
2606 return __f2fs_write_data_pages(mapping, wbc,
2607 F2FS_I(inode)->cp_task == current ?
2608 FS_CP_DATA_IO : FS_DATA_IO);
2611 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2613 struct inode *inode = mapping->host;
2614 loff_t i_size = i_size_read(inode);
2616 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
2617 if (to > i_size && !f2fs_verity_in_progress(inode)) {
2618 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2619 down_write(&F2FS_I(inode)->i_mmap_sem);
2621 truncate_pagecache(inode, i_size);
2622 if (!IS_NOQUOTA(inode))
2623 f2fs_truncate_blocks(inode, i_size, true);
2625 up_write(&F2FS_I(inode)->i_mmap_sem);
2626 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2630 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2631 struct page *page, loff_t pos, unsigned len,
2632 block_t *blk_addr, bool *node_changed)
2634 struct inode *inode = page->mapping->host;
2635 pgoff_t index = page->index;
2636 struct dnode_of_data dn;
2638 bool locked = false;
2639 struct extent_info ei = {0,0,0};
2644 * we already allocated all the blocks, so we don't need to get
2645 * the block addresses when there is no need to fill the page.
2647 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2648 !is_inode_flag_set(inode, FI_NO_PREALLOC) &&
2649 !f2fs_verity_in_progress(inode))
2652 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2653 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2654 flag = F2FS_GET_BLOCK_DEFAULT;
2656 flag = F2FS_GET_BLOCK_PRE_AIO;
2658 if (f2fs_has_inline_data(inode) ||
2659 (pos & PAGE_MASK) >= i_size_read(inode)) {
2660 __do_map_lock(sbi, flag, true);
2664 /* check inline_data */
2665 ipage = f2fs_get_node_page(sbi, inode->i_ino);
2666 if (IS_ERR(ipage)) {
2667 err = PTR_ERR(ipage);
2671 set_new_dnode(&dn, inode, ipage, ipage, 0);
2673 if (f2fs_has_inline_data(inode)) {
2674 if (pos + len <= MAX_INLINE_DATA(inode)) {
2675 f2fs_do_read_inline_data(page, ipage);
2676 set_inode_flag(inode, FI_DATA_EXIST);
2678 set_inline_node(ipage);
2680 err = f2fs_convert_inline_page(&dn, page);
2683 if (dn.data_blkaddr == NULL_ADDR)
2684 err = f2fs_get_block(&dn, index);
2686 } else if (locked) {
2687 err = f2fs_get_block(&dn, index);
2689 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2690 dn.data_blkaddr = ei.blk + index - ei.fofs;
2693 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2694 if (err || dn.data_blkaddr == NULL_ADDR) {
2695 f2fs_put_dnode(&dn);
2696 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2698 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2705 /* convert_inline_page can make node_changed */
2706 *blk_addr = dn.data_blkaddr;
2707 *node_changed = dn.node_changed;
2709 f2fs_put_dnode(&dn);
2712 __do_map_lock(sbi, flag, false);
2716 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2717 loff_t pos, unsigned len, unsigned flags,
2718 struct page **pagep, void **fsdata)
2720 struct inode *inode = mapping->host;
2721 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2722 struct page *page = NULL;
2723 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2724 bool need_balance = false, drop_atomic = false;
2725 block_t blkaddr = NULL_ADDR;
2728 trace_f2fs_write_begin(inode, pos, len, flags);
2730 if (!f2fs_is_checkpoint_ready(sbi)) {
2735 if ((f2fs_is_atomic_file(inode) &&
2736 !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2737 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2744 * We should check this at this moment to avoid deadlock on inode page
2745 * and #0 page. The locking rule for inline_data conversion should be:
2746 * lock_page(page #0) -> lock_page(inode_page)
2749 err = f2fs_convert_inline_inode(inode);
2755 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2756 * wait_for_stable_page. Will wait that below with our IO control.
2758 page = f2fs_pagecache_get_page(mapping, index,
2759 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2767 err = prepare_write_begin(sbi, page, pos, len,
2768 &blkaddr, &need_balance);
2772 if (need_balance && !IS_NOQUOTA(inode) &&
2773 has_not_enough_free_secs(sbi, 0, 0)) {
2775 f2fs_balance_fs(sbi, true);
2777 if (page->mapping != mapping) {
2778 /* The page got truncated from under us */
2779 f2fs_put_page(page, 1);
2784 f2fs_wait_on_page_writeback(page, DATA, false, true);
2786 if (len == PAGE_SIZE || PageUptodate(page))
2789 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
2790 !f2fs_verity_in_progress(inode)) {
2791 zero_user_segment(page, len, PAGE_SIZE);
2795 if (blkaddr == NEW_ADDR) {
2796 zero_user_segment(page, 0, PAGE_SIZE);
2797 SetPageUptodate(page);
2799 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
2800 DATA_GENERIC_ENHANCE_READ)) {
2801 err = -EFSCORRUPTED;
2804 err = f2fs_submit_page_read(inode, page, blkaddr);
2809 if (unlikely(page->mapping != mapping)) {
2810 f2fs_put_page(page, 1);
2813 if (unlikely(!PageUptodate(page))) {
2821 f2fs_put_page(page, 1);
2822 f2fs_write_failed(mapping, pos + len);
2824 f2fs_drop_inmem_pages_all(sbi, false);
2828 static int f2fs_write_end(struct file *file,
2829 struct address_space *mapping,
2830 loff_t pos, unsigned len, unsigned copied,
2831 struct page *page, void *fsdata)
2833 struct inode *inode = page->mapping->host;
2835 trace_f2fs_write_end(inode, pos, len, copied);
2838 * This should be come from len == PAGE_SIZE, and we expect copied
2839 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2840 * let generic_perform_write() try to copy data again through copied=0.
2842 if (!PageUptodate(page)) {
2843 if (unlikely(copied != len))
2846 SetPageUptodate(page);
2851 set_page_dirty(page);
2853 if (pos + copied > i_size_read(inode) &&
2854 !f2fs_verity_in_progress(inode))
2855 f2fs_i_size_write(inode, pos + copied);
2857 f2fs_put_page(page, 1);
2858 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2862 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2865 unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2866 unsigned blkbits = i_blkbits;
2867 unsigned blocksize_mask = (1 << blkbits) - 1;
2868 unsigned long align = offset | iov_iter_alignment(iter);
2869 struct block_device *bdev = inode->i_sb->s_bdev;
2871 if (align & blocksize_mask) {
2873 blkbits = blksize_bits(bdev_logical_block_size(bdev));
2874 blocksize_mask = (1 << blkbits) - 1;
2875 if (align & blocksize_mask)
2882 static void f2fs_dio_end_io(struct bio *bio)
2884 struct f2fs_private_dio *dio = bio->bi_private;
2886 dec_page_count(F2FS_I_SB(dio->inode),
2887 dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2889 bio->bi_private = dio->orig_private;
2890 bio->bi_end_io = dio->orig_end_io;
2897 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2900 struct f2fs_private_dio *dio;
2901 bool write = (bio_op(bio) == REQ_OP_WRITE);
2903 dio = f2fs_kzalloc(F2FS_I_SB(inode),
2904 sizeof(struct f2fs_private_dio), GFP_NOFS);
2909 dio->orig_end_io = bio->bi_end_io;
2910 dio->orig_private = bio->bi_private;
2913 bio->bi_end_io = f2fs_dio_end_io;
2914 bio->bi_private = dio;
2916 inc_page_count(F2FS_I_SB(inode),
2917 write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2922 bio->bi_status = BLK_STS_IOERR;
2926 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2928 struct address_space *mapping = iocb->ki_filp->f_mapping;
2929 struct inode *inode = mapping->host;
2930 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2931 struct f2fs_inode_info *fi = F2FS_I(inode);
2932 size_t count = iov_iter_count(iter);
2933 loff_t offset = iocb->ki_pos;
2934 int rw = iov_iter_rw(iter);
2936 enum rw_hint hint = iocb->ki_hint;
2937 int whint_mode = F2FS_OPTION(sbi).whint_mode;
2940 err = check_direct_IO(inode, iter, offset);
2942 return err < 0 ? err : 0;
2944 if (f2fs_force_buffered_io(inode, iocb, iter))
2947 do_opu = allow_outplace_dio(inode, iocb, iter);
2949 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2951 if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2952 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2954 if (iocb->ki_flags & IOCB_NOWAIT) {
2955 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2956 iocb->ki_hint = hint;
2960 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2961 up_read(&fi->i_gc_rwsem[rw]);
2962 iocb->ki_hint = hint;
2967 down_read(&fi->i_gc_rwsem[rw]);
2969 down_read(&fi->i_gc_rwsem[READ]);
2972 err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2973 iter, rw == WRITE ? get_data_block_dio_write :
2974 get_data_block_dio, NULL, f2fs_dio_submit_bio,
2975 DIO_LOCKING | DIO_SKIP_HOLES);
2978 up_read(&fi->i_gc_rwsem[READ]);
2980 up_read(&fi->i_gc_rwsem[rw]);
2983 if (whint_mode == WHINT_MODE_OFF)
2984 iocb->ki_hint = hint;
2986 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2989 set_inode_flag(inode, FI_UPDATE_WRITE);
2990 } else if (err < 0) {
2991 f2fs_write_failed(mapping, offset + count);
2996 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
3001 void f2fs_invalidate_page(struct page *page, unsigned int offset,
3002 unsigned int length)
3004 struct inode *inode = page->mapping->host;
3005 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3007 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3008 (offset % PAGE_SIZE || length != PAGE_SIZE))
3011 if (PageDirty(page)) {
3012 if (inode->i_ino == F2FS_META_INO(sbi)) {
3013 dec_page_count(sbi, F2FS_DIRTY_META);
3014 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3015 dec_page_count(sbi, F2FS_DIRTY_NODES);
3017 inode_dec_dirty_pages(inode);
3018 f2fs_remove_dirty_inode(inode);
3022 clear_cold_data(page);
3024 if (IS_ATOMIC_WRITTEN_PAGE(page))
3025 return f2fs_drop_inmem_page(inode, page);
3027 f2fs_clear_page_private(page);
3030 int f2fs_release_page(struct page *page, gfp_t wait)
3032 /* If this is dirty page, keep PagePrivate */
3033 if (PageDirty(page))
3036 /* This is atomic written page, keep Private */
3037 if (IS_ATOMIC_WRITTEN_PAGE(page))
3040 clear_cold_data(page);
3041 f2fs_clear_page_private(page);
3045 static int f2fs_set_data_page_dirty(struct page *page)
3047 struct inode *inode = page_file_mapping(page)->host;
3049 trace_f2fs_set_page_dirty(page, DATA);
3051 if (!PageUptodate(page))
3052 SetPageUptodate(page);
3053 if (PageSwapCache(page))
3054 return __set_page_dirty_nobuffers(page);
3056 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3057 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
3058 f2fs_register_inmem_page(inode, page);
3062 * Previously, this page has been registered, we just
3068 if (!PageDirty(page)) {
3069 __set_page_dirty_nobuffers(page);
3070 f2fs_update_dirty_page(inode, page);
3076 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3078 struct inode *inode = mapping->host;
3080 if (f2fs_has_inline_data(inode))
3083 /* make sure allocating whole blocks */
3084 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3085 filemap_write_and_wait(mapping);
3087 return generic_block_bmap(mapping, block, get_data_block_bmap);
3090 #ifdef CONFIG_MIGRATION
3091 #include <linux/migrate.h>
3093 int f2fs_migrate_page(struct address_space *mapping,
3094 struct page *newpage, struct page *page, enum migrate_mode mode)
3096 int rc, extra_count;
3097 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3098 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
3100 BUG_ON(PageWriteback(page));
3102 /* migrating an atomic written page is safe with the inmem_lock hold */
3103 if (atomic_written) {
3104 if (mode != MIGRATE_SYNC)
3106 if (!mutex_trylock(&fi->inmem_lock))
3110 /* one extra reference was held for atomic_write page */
3111 extra_count = atomic_written ? 1 : 0;
3112 rc = migrate_page_move_mapping(mapping, newpage,
3114 if (rc != MIGRATEPAGE_SUCCESS) {
3116 mutex_unlock(&fi->inmem_lock);
3120 if (atomic_written) {
3121 struct inmem_pages *cur;
3122 list_for_each_entry(cur, &fi->inmem_pages, list)
3123 if (cur->page == page) {
3124 cur->page = newpage;
3127 mutex_unlock(&fi->inmem_lock);
3132 if (PagePrivate(page)) {
3133 f2fs_set_page_private(newpage, page_private(page));
3134 f2fs_clear_page_private(page);
3137 if (mode != MIGRATE_SYNC_NO_COPY)
3138 migrate_page_copy(newpage, page);
3140 migrate_page_states(newpage, page);
3142 return MIGRATEPAGE_SUCCESS;
3147 /* Copied from generic_swapfile_activate() to check any holes */
3148 static int check_swap_activate(struct file *swap_file, unsigned int max)
3150 struct address_space *mapping = swap_file->f_mapping;
3151 struct inode *inode = mapping->host;
3152 unsigned blocks_per_page;
3153 unsigned long page_no;
3155 sector_t probe_block;
3156 sector_t last_block;
3157 sector_t lowest_block = -1;
3158 sector_t highest_block = 0;
3160 blkbits = inode->i_blkbits;
3161 blocks_per_page = PAGE_SIZE >> blkbits;
3164 * Map all the blocks into the extent list. This code doesn't try
3169 last_block = i_size_read(inode) >> blkbits;
3170 while ((probe_block + blocks_per_page) <= last_block && page_no < max) {
3171 unsigned block_in_page;
3172 sector_t first_block;
3176 first_block = bmap(inode, probe_block);
3177 if (first_block == 0)
3181 * It must be PAGE_SIZE aligned on-disk
3183 if (first_block & (blocks_per_page - 1)) {
3188 for (block_in_page = 1; block_in_page < blocks_per_page;
3192 block = bmap(inode, probe_block + block_in_page);
3195 if (block != first_block + block_in_page) {
3202 first_block >>= (PAGE_SHIFT - blkbits);
3203 if (page_no) { /* exclude the header page */
3204 if (first_block < lowest_block)
3205 lowest_block = first_block;
3206 if (first_block > highest_block)
3207 highest_block = first_block;
3211 probe_block += blocks_per_page;
3218 pr_err("swapon: swapfile has holes\n");
3222 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3225 struct inode *inode = file_inode(file);
3228 if (!S_ISREG(inode->i_mode))
3231 if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3234 ret = f2fs_convert_inline_inode(inode);
3238 ret = check_swap_activate(file, sis->max);
3242 set_inode_flag(inode, FI_PIN_FILE);
3243 f2fs_precache_extents(inode);
3244 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3248 static void f2fs_swap_deactivate(struct file *file)
3250 struct inode *inode = file_inode(file);
3252 clear_inode_flag(inode, FI_PIN_FILE);
3255 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3261 static void f2fs_swap_deactivate(struct file *file)
3266 const struct address_space_operations f2fs_dblock_aops = {
3267 .readpage = f2fs_read_data_page,
3268 .readpages = f2fs_read_data_pages,
3269 .writepage = f2fs_write_data_page,
3270 .writepages = f2fs_write_data_pages,
3271 .write_begin = f2fs_write_begin,
3272 .write_end = f2fs_write_end,
3273 .set_page_dirty = f2fs_set_data_page_dirty,
3274 .invalidatepage = f2fs_invalidate_page,
3275 .releasepage = f2fs_release_page,
3276 .direct_IO = f2fs_direct_IO,
3278 .swap_activate = f2fs_swap_activate,
3279 .swap_deactivate = f2fs_swap_deactivate,
3280 #ifdef CONFIG_MIGRATION
3281 .migratepage = f2fs_migrate_page,
3285 void f2fs_clear_page_cache_dirty_tag(struct page *page)
3287 struct address_space *mapping = page_mapping(page);
3288 unsigned long flags;
3290 xa_lock_irqsave(&mapping->i_pages, flags);
3291 __xa_clear_mark(&mapping->i_pages, page_index(page),
3292 PAGECACHE_TAG_DIRTY);
3293 xa_unlock_irqrestore(&mapping->i_pages, flags);
3296 int __init f2fs_init_post_read_processing(void)
3298 bio_post_read_ctx_cache =
3299 kmem_cache_create("f2fs_bio_post_read_ctx",
3300 sizeof(struct bio_post_read_ctx), 0, 0, NULL);
3301 if (!bio_post_read_ctx_cache)
3303 bio_post_read_ctx_pool =
3304 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
3305 bio_post_read_ctx_cache);
3306 if (!bio_post_read_ctx_pool)
3307 goto fail_free_cache;
3311 kmem_cache_destroy(bio_post_read_ctx_cache);
3316 void f2fs_destroy_post_read_processing(void)
3318 mempool_destroy(bio_post_read_ctx_pool);
3319 kmem_cache_destroy(bio_post_read_ctx_cache);
3322 int __init f2fs_init_bio_entry_cache(void)
3324 bio_entry_slab = f2fs_kmem_cache_create("bio_entry_slab",
3325 sizeof(struct bio_entry));
3326 if (!bio_entry_slab)
3331 void __exit f2fs_destroy_bio_entry_cache(void)
3333 kmem_cache_destroy(bio_entry_slab);