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/sched/mm.h>
12 #include <linux/mpage.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include <linux/blkdev.h>
16 #include <linux/bio.h>
17 #include <linux/blk-crypto.h>
18 #include <linux/swap.h>
19 #include <linux/prefetch.h>
20 #include <linux/uio.h>
21 #include <linux/sched/signal.h>
22 #include <linux/fiemap.h>
23 #include <linux/iomap.h>
29 #include <trace/events/f2fs.h>
31 #define NUM_PREALLOC_POST_READ_CTXS 128
33 static struct kmem_cache *bio_post_read_ctx_cache;
34 static struct kmem_cache *bio_entry_slab;
35 static mempool_t *bio_post_read_ctx_pool;
36 static struct bio_set f2fs_bioset;
38 #define F2FS_BIO_POOL_SIZE NR_CURSEG_TYPE
40 int __init f2fs_init_bioset(void)
42 if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
43 0, BIOSET_NEED_BVECS))
48 void f2fs_destroy_bioset(void)
50 bioset_exit(&f2fs_bioset);
53 static bool __is_cp_guaranteed(struct page *page)
55 struct address_space *mapping = page->mapping;
57 struct f2fs_sb_info *sbi;
62 inode = mapping->host;
63 sbi = F2FS_I_SB(inode);
65 if (inode->i_ino == F2FS_META_INO(sbi) ||
66 inode->i_ino == F2FS_NODE_INO(sbi) ||
67 S_ISDIR(inode->i_mode))
70 if (f2fs_is_compressed_page(page))
72 if ((S_ISREG(inode->i_mode) &&
73 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
74 page_private_gcing(page))
79 static enum count_type __read_io_type(struct page *page)
81 struct address_space *mapping = page_file_mapping(page);
84 struct inode *inode = mapping->host;
85 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
87 if (inode->i_ino == F2FS_META_INO(sbi))
90 if (inode->i_ino == F2FS_NODE_INO(sbi))
96 /* postprocessing steps for read bios */
97 enum bio_post_read_step {
98 #ifdef CONFIG_FS_ENCRYPTION
99 STEP_DECRYPT = 1 << 0,
101 STEP_DECRYPT = 0, /* compile out the decryption-related code */
103 #ifdef CONFIG_F2FS_FS_COMPRESSION
104 STEP_DECOMPRESS = 1 << 1,
106 STEP_DECOMPRESS = 0, /* compile out the decompression-related code */
108 #ifdef CONFIG_FS_VERITY
109 STEP_VERITY = 1 << 2,
111 STEP_VERITY = 0, /* compile out the verity-related code */
115 struct bio_post_read_ctx {
117 struct f2fs_sb_info *sbi;
118 struct work_struct work;
119 unsigned int enabled_steps;
123 static void f2fs_finish_read_bio(struct bio *bio)
126 struct bvec_iter_all iter_all;
129 * Update and unlock the bio's pagecache pages, and put the
130 * decompression context for any compressed pages.
132 bio_for_each_segment_all(bv, bio, iter_all) {
133 struct page *page = bv->bv_page;
135 if (f2fs_is_compressed_page(page)) {
137 f2fs_end_read_compressed_page(page, true, 0);
138 f2fs_put_page_dic(page);
142 /* PG_error was set if decryption or verity failed. */
143 if (bio->bi_status || PageError(page)) {
144 ClearPageUptodate(page);
145 /* will re-read again later */
146 ClearPageError(page);
148 SetPageUptodate(page);
150 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
155 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
159 static void f2fs_verify_bio(struct work_struct *work)
161 struct bio_post_read_ctx *ctx =
162 container_of(work, struct bio_post_read_ctx, work);
163 struct bio *bio = ctx->bio;
164 bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
167 * fsverity_verify_bio() may call readpages() again, and while verity
168 * will be disabled for this, decryption and/or decompression may still
169 * be needed, resulting in another bio_post_read_ctx being allocated.
170 * So to prevent deadlocks we need to release the current ctx to the
171 * mempool first. This assumes that verity is the last post-read step.
173 mempool_free(ctx, bio_post_read_ctx_pool);
174 bio->bi_private = NULL;
177 * Verify the bio's pages with fs-verity. Exclude compressed pages,
178 * as those were handled separately by f2fs_end_read_compressed_page().
180 if (may_have_compressed_pages) {
182 struct bvec_iter_all iter_all;
184 bio_for_each_segment_all(bv, bio, iter_all) {
185 struct page *page = bv->bv_page;
187 if (!f2fs_is_compressed_page(page) &&
188 !PageError(page) && !fsverity_verify_page(page))
192 fsverity_verify_bio(bio);
195 f2fs_finish_read_bio(bio);
199 * If the bio's data needs to be verified with fs-verity, then enqueue the
200 * verity work for the bio. Otherwise finish the bio now.
202 * Note that to avoid deadlocks, the verity work can't be done on the
203 * decryption/decompression workqueue. This is because verifying the data pages
204 * can involve reading verity metadata pages from the file, and these verity
205 * metadata pages may be encrypted and/or compressed.
207 static void f2fs_verify_and_finish_bio(struct bio *bio)
209 struct bio_post_read_ctx *ctx = bio->bi_private;
211 if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
212 INIT_WORK(&ctx->work, f2fs_verify_bio);
213 fsverity_enqueue_verify_work(&ctx->work);
215 f2fs_finish_read_bio(bio);
220 * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
221 * remaining page was read by @ctx->bio.
223 * Note that a bio may span clusters (even a mix of compressed and uncompressed
224 * clusters) or be for just part of a cluster. STEP_DECOMPRESS just indicates
225 * that the bio includes at least one compressed page. The actual decompression
226 * is done on a per-cluster basis, not a per-bio basis.
228 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx)
231 struct bvec_iter_all iter_all;
232 bool all_compressed = true;
233 block_t blkaddr = ctx->fs_blkaddr;
235 bio_for_each_segment_all(bv, ctx->bio, iter_all) {
236 struct page *page = bv->bv_page;
238 /* PG_error was set if decryption failed. */
239 if (f2fs_is_compressed_page(page))
240 f2fs_end_read_compressed_page(page, PageError(page),
243 all_compressed = false;
249 * Optimization: if all the bio's pages are compressed, then scheduling
250 * the per-bio verity work is unnecessary, as verity will be fully
251 * handled at the compression cluster level.
254 ctx->enabled_steps &= ~STEP_VERITY;
257 static void f2fs_post_read_work(struct work_struct *work)
259 struct bio_post_read_ctx *ctx =
260 container_of(work, struct bio_post_read_ctx, work);
262 if (ctx->enabled_steps & STEP_DECRYPT)
263 fscrypt_decrypt_bio(ctx->bio);
265 if (ctx->enabled_steps & STEP_DECOMPRESS)
266 f2fs_handle_step_decompress(ctx);
268 f2fs_verify_and_finish_bio(ctx->bio);
271 static void f2fs_read_end_io(struct bio *bio)
273 struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
274 struct bio_post_read_ctx *ctx;
276 iostat_update_and_unbind_ctx(bio, 0);
277 ctx = bio->bi_private;
279 if (time_to_inject(sbi, FAULT_READ_IO)) {
280 f2fs_show_injection_info(sbi, FAULT_READ_IO);
281 bio->bi_status = BLK_STS_IOERR;
284 if (bio->bi_status) {
285 f2fs_finish_read_bio(bio);
289 if (ctx && (ctx->enabled_steps & (STEP_DECRYPT | STEP_DECOMPRESS))) {
290 INIT_WORK(&ctx->work, f2fs_post_read_work);
291 queue_work(ctx->sbi->post_read_wq, &ctx->work);
293 f2fs_verify_and_finish_bio(bio);
297 static void f2fs_write_end_io(struct bio *bio)
299 struct f2fs_sb_info *sbi;
300 struct bio_vec *bvec;
301 struct bvec_iter_all iter_all;
303 iostat_update_and_unbind_ctx(bio, 1);
304 sbi = bio->bi_private;
306 if (time_to_inject(sbi, FAULT_WRITE_IO)) {
307 f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
308 bio->bi_status = BLK_STS_IOERR;
311 bio_for_each_segment_all(bvec, bio, iter_all) {
312 struct page *page = bvec->bv_page;
313 enum count_type type = WB_DATA_TYPE(page);
315 if (page_private_dummy(page)) {
316 clear_page_private_dummy(page);
318 mempool_free(page, sbi->write_io_dummy);
320 if (unlikely(bio->bi_status))
321 f2fs_stop_checkpoint(sbi, true);
325 fscrypt_finalize_bounce_page(&page);
327 #ifdef CONFIG_F2FS_FS_COMPRESSION
328 if (f2fs_is_compressed_page(page)) {
329 f2fs_compress_write_end_io(bio, page);
334 if (unlikely(bio->bi_status)) {
335 mapping_set_error(page->mapping, -EIO);
336 if (type == F2FS_WB_CP_DATA)
337 f2fs_stop_checkpoint(sbi, true);
340 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
341 page->index != nid_of_node(page));
343 dec_page_count(sbi, type);
344 if (f2fs_in_warm_node_list(sbi, page))
345 f2fs_del_fsync_node_entry(sbi, page);
346 clear_page_private_gcing(page);
347 end_page_writeback(page);
349 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
350 wq_has_sleeper(&sbi->cp_wait))
351 wake_up(&sbi->cp_wait);
356 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
357 block_t blk_addr, sector_t *sector)
359 struct block_device *bdev = sbi->sb->s_bdev;
362 if (f2fs_is_multi_device(sbi)) {
363 for (i = 0; i < sbi->s_ndevs; i++) {
364 if (FDEV(i).start_blk <= blk_addr &&
365 FDEV(i).end_blk >= blk_addr) {
366 blk_addr -= FDEV(i).start_blk;
374 *sector = SECTOR_FROM_BLOCK(blk_addr);
378 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
382 if (!f2fs_is_multi_device(sbi))
385 for (i = 0; i < sbi->s_ndevs; i++)
386 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
391 static void __attach_io_flag(struct f2fs_io_info *fio, unsigned int io_flag)
393 unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
394 unsigned int fua_flag = io_flag & temp_mask;
395 unsigned int meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
398 * data/node io flag bits per temp:
399 * REQ_META | REQ_FUA |
400 * 5 | 4 | 3 | 2 | 1 | 0 |
401 * Cold | Warm | Hot | Cold | Warm | Hot |
403 if ((1 << fio->temp) & meta_flag)
404 fio->op_flags |= REQ_META;
405 if ((1 << fio->temp) & fua_flag)
406 fio->op_flags |= REQ_FUA;
409 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
411 struct f2fs_sb_info *sbi = fio->sbi;
412 struct block_device *bdev;
416 if (fio->type == DATA)
417 __attach_io_flag(fio, sbi->data_io_flag);
418 else if (fio->type == NODE)
419 __attach_io_flag(fio, sbi->node_io_flag);
421 bdev = f2fs_target_device(sbi, fio->new_blkaddr, §or);
422 bio = bio_alloc_bioset(bdev, npages, fio->op | fio->op_flags, GFP_NOIO,
424 bio->bi_iter.bi_sector = sector;
425 if (is_read_io(fio->op)) {
426 bio->bi_end_io = f2fs_read_end_io;
427 bio->bi_private = NULL;
429 bio->bi_end_io = f2fs_write_end_io;
430 bio->bi_private = sbi;
431 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
432 fio->type, fio->temp);
434 iostat_alloc_and_bind_ctx(sbi, bio, NULL);
437 wbc_init_bio(fio->io_wbc, bio);
442 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
444 const struct f2fs_io_info *fio,
448 * The f2fs garbage collector sets ->encrypted_page when it wants to
449 * read/write raw data without encryption.
451 if (!fio || !fio->encrypted_page)
452 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
455 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
457 const struct f2fs_io_info *fio)
460 * The f2fs garbage collector sets ->encrypted_page when it wants to
461 * read/write raw data without encryption.
463 if (fio && fio->encrypted_page)
464 return !bio_has_crypt_ctx(bio);
466 return fscrypt_mergeable_bio(bio, inode, next_idx);
469 static inline void __submit_bio(struct f2fs_sb_info *sbi,
470 struct bio *bio, enum page_type type)
472 if (!is_read_io(bio_op(bio))) {
475 if (type != DATA && type != NODE)
478 if (f2fs_lfs_mode(sbi) && current->plug)
479 blk_finish_plug(current->plug);
481 if (!F2FS_IO_ALIGNED(sbi))
484 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
485 start %= F2FS_IO_SIZE(sbi);
490 /* fill dummy pages */
491 for (; start < F2FS_IO_SIZE(sbi); start++) {
493 mempool_alloc(sbi->write_io_dummy,
494 GFP_NOIO | __GFP_NOFAIL);
495 f2fs_bug_on(sbi, !page);
499 zero_user_segment(page, 0, PAGE_SIZE);
500 set_page_private_dummy(page);
502 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
506 * In the NODE case, we lose next block address chain. So, we
507 * need to do checkpoint in f2fs_sync_file.
510 set_sbi_flag(sbi, SBI_NEED_CP);
513 if (is_read_io(bio_op(bio)))
514 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
516 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
518 iostat_update_submit_ctx(bio, type);
522 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
523 struct bio *bio, enum page_type type)
525 __submit_bio(sbi, bio, type);
528 static void __submit_merged_bio(struct f2fs_bio_info *io)
530 struct f2fs_io_info *fio = &io->fio;
535 if (is_read_io(fio->op))
536 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
538 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
540 __submit_bio(io->sbi, io->bio, fio->type);
544 static bool __has_merged_page(struct bio *bio, struct inode *inode,
545 struct page *page, nid_t ino)
547 struct bio_vec *bvec;
548 struct bvec_iter_all iter_all;
553 if (!inode && !page && !ino)
556 bio_for_each_segment_all(bvec, bio, iter_all) {
557 struct page *target = bvec->bv_page;
559 if (fscrypt_is_bounce_page(target)) {
560 target = fscrypt_pagecache_page(target);
564 if (f2fs_is_compressed_page(target)) {
565 target = f2fs_compress_control_page(target);
570 if (inode && inode == target->mapping->host)
572 if (page && page == target)
574 if (ino && ino == ino_of_node(target))
581 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
582 enum page_type type, enum temp_type temp)
584 enum page_type btype = PAGE_TYPE_OF_BIO(type);
585 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
587 f2fs_down_write(&io->io_rwsem);
589 /* change META to META_FLUSH in the checkpoint procedure */
590 if (type >= META_FLUSH) {
591 io->fio.type = META_FLUSH;
592 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
593 if (!test_opt(sbi, NOBARRIER))
594 io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
596 __submit_merged_bio(io);
597 f2fs_up_write(&io->io_rwsem);
600 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
601 struct inode *inode, struct page *page,
602 nid_t ino, enum page_type type, bool force)
607 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
609 enum page_type btype = PAGE_TYPE_OF_BIO(type);
610 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
612 f2fs_down_read(&io->io_rwsem);
613 ret = __has_merged_page(io->bio, inode, page, ino);
614 f2fs_up_read(&io->io_rwsem);
617 __f2fs_submit_merged_write(sbi, type, temp);
619 /* TODO: use HOT temp only for meta pages now. */
625 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
627 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
630 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
631 struct inode *inode, struct page *page,
632 nid_t ino, enum page_type type)
634 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
637 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
639 f2fs_submit_merged_write(sbi, DATA);
640 f2fs_submit_merged_write(sbi, NODE);
641 f2fs_submit_merged_write(sbi, META);
645 * Fill the locked page with data located in the block address.
646 * A caller needs to unlock the page on failure.
648 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
651 struct page *page = fio->encrypted_page ?
652 fio->encrypted_page : fio->page;
654 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
655 fio->is_por ? META_POR : (__is_meta_io(fio) ?
656 META_GENERIC : DATA_GENERIC_ENHANCE)))
657 return -EFSCORRUPTED;
659 trace_f2fs_submit_page_bio(page, fio);
661 /* Allocate a new bio */
662 bio = __bio_alloc(fio, 1);
664 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
665 fio->page->index, fio, GFP_NOIO);
667 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
672 if (fio->io_wbc && !is_read_io(fio->op))
673 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
675 inc_page_count(fio->sbi, is_read_io(fio->op) ?
676 __read_io_type(page): WB_DATA_TYPE(fio->page));
678 __submit_bio(fio->sbi, bio, fio->type);
682 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
683 block_t last_blkaddr, block_t cur_blkaddr)
685 if (unlikely(sbi->max_io_bytes &&
686 bio->bi_iter.bi_size >= sbi->max_io_bytes))
688 if (last_blkaddr + 1 != cur_blkaddr)
690 return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
693 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
694 struct f2fs_io_info *fio)
696 if (io->fio.op != fio->op)
698 return io->fio.op_flags == fio->op_flags;
701 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
702 struct f2fs_bio_info *io,
703 struct f2fs_io_info *fio,
704 block_t last_blkaddr,
707 if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
708 unsigned int filled_blocks =
709 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
710 unsigned int io_size = F2FS_IO_SIZE(sbi);
711 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
713 /* IOs in bio is aligned and left space of vectors is not enough */
714 if (!(filled_blocks % io_size) && left_vecs < io_size)
717 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
719 return io_type_is_mergeable(io, fio);
722 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
723 struct page *page, enum temp_type temp)
725 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
726 struct bio_entry *be;
728 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
732 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
735 f2fs_down_write(&io->bio_list_lock);
736 list_add_tail(&be->list, &io->bio_list);
737 f2fs_up_write(&io->bio_list_lock);
740 static void del_bio_entry(struct bio_entry *be)
743 kmem_cache_free(bio_entry_slab, be);
746 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
749 struct f2fs_sb_info *sbi = fio->sbi;
754 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
755 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
756 struct list_head *head = &io->bio_list;
757 struct bio_entry *be;
759 f2fs_down_write(&io->bio_list_lock);
760 list_for_each_entry(be, head, list) {
766 f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
769 if (f2fs_crypt_mergeable_bio(*bio,
770 fio->page->mapping->host,
771 fio->page->index, fio) &&
772 bio_add_page(*bio, page, PAGE_SIZE, 0) ==
778 /* page can't be merged into bio; submit the bio */
780 __submit_bio(sbi, *bio, DATA);
783 f2fs_up_write(&io->bio_list_lock);
794 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
795 struct bio **bio, struct page *page)
799 struct bio *target = bio ? *bio : NULL;
801 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
802 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
803 struct list_head *head = &io->bio_list;
804 struct bio_entry *be;
806 if (list_empty(head))
809 f2fs_down_read(&io->bio_list_lock);
810 list_for_each_entry(be, head, list) {
812 found = (target == be->bio);
814 found = __has_merged_page(be->bio, NULL,
819 f2fs_up_read(&io->bio_list_lock);
826 f2fs_down_write(&io->bio_list_lock);
827 list_for_each_entry(be, head, list) {
829 found = (target == be->bio);
831 found = __has_merged_page(be->bio, NULL,
839 f2fs_up_write(&io->bio_list_lock);
843 __submit_bio(sbi, target, DATA);
850 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
852 struct bio *bio = *fio->bio;
853 struct page *page = fio->encrypted_page ?
854 fio->encrypted_page : fio->page;
856 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
857 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
858 return -EFSCORRUPTED;
860 trace_f2fs_submit_page_bio(page, fio);
862 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
864 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
867 bio = __bio_alloc(fio, BIO_MAX_VECS);
868 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
869 fio->page->index, fio, GFP_NOIO);
871 add_bio_entry(fio->sbi, bio, page, fio->temp);
873 if (add_ipu_page(fio, &bio, page))
878 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
880 inc_page_count(fio->sbi, WB_DATA_TYPE(page));
882 *fio->last_block = fio->new_blkaddr;
888 void f2fs_submit_page_write(struct f2fs_io_info *fio)
890 struct f2fs_sb_info *sbi = fio->sbi;
891 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
892 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
893 struct page *bio_page;
895 f2fs_bug_on(sbi, is_read_io(fio->op));
897 f2fs_down_write(&io->io_rwsem);
900 spin_lock(&io->io_lock);
901 if (list_empty(&io->io_list)) {
902 spin_unlock(&io->io_lock);
905 fio = list_first_entry(&io->io_list,
906 struct f2fs_io_info, list);
907 list_del(&fio->list);
908 spin_unlock(&io->io_lock);
911 verify_fio_blkaddr(fio);
913 if (fio->encrypted_page)
914 bio_page = fio->encrypted_page;
915 else if (fio->compressed_page)
916 bio_page = fio->compressed_page;
918 bio_page = fio->page;
920 /* set submitted = true as a return value */
921 fio->submitted = true;
923 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
926 (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
928 !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
929 bio_page->index, fio)))
930 __submit_merged_bio(io);
932 if (io->bio == NULL) {
933 if (F2FS_IO_ALIGNED(sbi) &&
934 (fio->type == DATA || fio->type == NODE) &&
935 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
936 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
940 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
941 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
942 bio_page->index, fio, GFP_NOIO);
946 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
947 __submit_merged_bio(io);
952 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
954 io->last_block_in_bio = fio->new_blkaddr;
956 trace_f2fs_submit_page_write(fio->page, fio);
961 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
962 !f2fs_is_checkpoint_ready(sbi))
963 __submit_merged_bio(io);
964 f2fs_up_write(&io->io_rwsem);
967 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
968 unsigned nr_pages, unsigned op_flag,
969 pgoff_t first_idx, bool for_write)
971 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
973 struct bio_post_read_ctx *ctx = NULL;
974 unsigned int post_read_steps = 0;
976 struct block_device *bdev = f2fs_target_device(sbi, blkaddr, §or);
978 bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
979 REQ_OP_READ | op_flag,
980 for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
982 return ERR_PTR(-ENOMEM);
983 bio->bi_iter.bi_sector = sector;
984 f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
985 bio->bi_end_io = f2fs_read_end_io;
987 if (fscrypt_inode_uses_fs_layer_crypto(inode))
988 post_read_steps |= STEP_DECRYPT;
990 if (f2fs_need_verity(inode, first_idx))
991 post_read_steps |= STEP_VERITY;
994 * STEP_DECOMPRESS is handled specially, since a compressed file might
995 * contain both compressed and uncompressed clusters. We'll allocate a
996 * bio_post_read_ctx if the file is compressed, but the caller is
997 * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1000 if (post_read_steps || f2fs_compressed_file(inode)) {
1001 /* Due to the mempool, this never fails. */
1002 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1005 ctx->enabled_steps = post_read_steps;
1006 ctx->fs_blkaddr = blkaddr;
1007 bio->bi_private = ctx;
1009 iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1014 /* This can handle encryption stuffs */
1015 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1016 block_t blkaddr, int op_flags, bool for_write)
1018 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1021 bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1022 page->index, for_write);
1024 return PTR_ERR(bio);
1026 /* wait for GCed page writeback via META_MAPPING */
1027 f2fs_wait_on_block_writeback(inode, blkaddr);
1029 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1033 ClearPageError(page);
1034 inc_page_count(sbi, F2FS_RD_DATA);
1035 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1036 __submit_bio(sbi, bio, DATA);
1040 static void __set_data_blkaddr(struct dnode_of_data *dn)
1042 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1046 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1047 base = get_extra_isize(dn->inode);
1049 /* Get physical address of data block */
1050 addr_array = blkaddr_in_node(rn);
1051 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1055 * Lock ordering for the change of data block address:
1058 * update block addresses in the node page
1060 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1062 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1063 __set_data_blkaddr(dn);
1064 if (set_page_dirty(dn->node_page))
1065 dn->node_changed = true;
1068 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1070 dn->data_blkaddr = blkaddr;
1071 f2fs_set_data_blkaddr(dn);
1072 f2fs_update_extent_cache(dn);
1075 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1076 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1078 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1084 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1086 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1089 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1090 dn->ofs_in_node, count);
1092 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1094 for (; count > 0; dn->ofs_in_node++) {
1095 block_t blkaddr = f2fs_data_blkaddr(dn);
1097 if (blkaddr == NULL_ADDR) {
1098 dn->data_blkaddr = NEW_ADDR;
1099 __set_data_blkaddr(dn);
1104 if (set_page_dirty(dn->node_page))
1105 dn->node_changed = true;
1109 /* Should keep dn->ofs_in_node unchanged */
1110 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1112 unsigned int ofs_in_node = dn->ofs_in_node;
1115 ret = f2fs_reserve_new_blocks(dn, 1);
1116 dn->ofs_in_node = ofs_in_node;
1120 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1122 bool need_put = dn->inode_page ? false : true;
1125 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1129 if (dn->data_blkaddr == NULL_ADDR)
1130 err = f2fs_reserve_new_block(dn);
1131 if (err || need_put)
1136 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1138 struct extent_info ei = {0, };
1139 struct inode *inode = dn->inode;
1141 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1142 dn->data_blkaddr = ei.blk + index - ei.fofs;
1146 return f2fs_reserve_block(dn, index);
1149 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1150 int op_flags, bool for_write)
1152 struct address_space *mapping = inode->i_mapping;
1153 struct dnode_of_data dn;
1155 struct extent_info ei = {0, };
1158 page = f2fs_grab_cache_page(mapping, index, for_write);
1160 return ERR_PTR(-ENOMEM);
1162 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1163 dn.data_blkaddr = ei.blk + index - ei.fofs;
1164 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1165 DATA_GENERIC_ENHANCE_READ)) {
1166 err = -EFSCORRUPTED;
1172 set_new_dnode(&dn, inode, NULL, NULL, 0);
1173 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1176 f2fs_put_dnode(&dn);
1178 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1182 if (dn.data_blkaddr != NEW_ADDR &&
1183 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1185 DATA_GENERIC_ENHANCE)) {
1186 err = -EFSCORRUPTED;
1190 if (PageUptodate(page)) {
1196 * A new dentry page is allocated but not able to be written, since its
1197 * new inode page couldn't be allocated due to -ENOSPC.
1198 * In such the case, its blkaddr can be remained as NEW_ADDR.
1199 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1200 * f2fs_init_inode_metadata.
1202 if (dn.data_blkaddr == NEW_ADDR) {
1203 zero_user_segment(page, 0, PAGE_SIZE);
1204 if (!PageUptodate(page))
1205 SetPageUptodate(page);
1210 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1211 op_flags, for_write);
1217 f2fs_put_page(page, 1);
1218 return ERR_PTR(err);
1221 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1223 struct address_space *mapping = inode->i_mapping;
1226 page = find_get_page(mapping, index);
1227 if (page && PageUptodate(page))
1229 f2fs_put_page(page, 0);
1231 page = f2fs_get_read_data_page(inode, index, 0, false);
1235 if (PageUptodate(page))
1238 wait_on_page_locked(page);
1239 if (unlikely(!PageUptodate(page))) {
1240 f2fs_put_page(page, 0);
1241 return ERR_PTR(-EIO);
1247 * If it tries to access a hole, return an error.
1248 * Because, the callers, functions in dir.c and GC, should be able to know
1249 * whether this page exists or not.
1251 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1254 struct address_space *mapping = inode->i_mapping;
1257 page = f2fs_get_read_data_page(inode, index, 0, for_write);
1261 /* wait for read completion */
1263 if (unlikely(page->mapping != mapping)) {
1264 f2fs_put_page(page, 1);
1267 if (unlikely(!PageUptodate(page))) {
1268 f2fs_put_page(page, 1);
1269 return ERR_PTR(-EIO);
1275 * Caller ensures that this data page is never allocated.
1276 * A new zero-filled data page is allocated in the page cache.
1278 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1280 * Note that, ipage is set only by make_empty_dir, and if any error occur,
1281 * ipage should be released by this function.
1283 struct page *f2fs_get_new_data_page(struct inode *inode,
1284 struct page *ipage, pgoff_t index, bool new_i_size)
1286 struct address_space *mapping = inode->i_mapping;
1288 struct dnode_of_data dn;
1291 page = f2fs_grab_cache_page(mapping, index, true);
1294 * before exiting, we should make sure ipage will be released
1295 * if any error occur.
1297 f2fs_put_page(ipage, 1);
1298 return ERR_PTR(-ENOMEM);
1301 set_new_dnode(&dn, inode, ipage, NULL, 0);
1302 err = f2fs_reserve_block(&dn, index);
1304 f2fs_put_page(page, 1);
1305 return ERR_PTR(err);
1308 f2fs_put_dnode(&dn);
1310 if (PageUptodate(page))
1313 if (dn.data_blkaddr == NEW_ADDR) {
1314 zero_user_segment(page, 0, PAGE_SIZE);
1315 if (!PageUptodate(page))
1316 SetPageUptodate(page);
1318 f2fs_put_page(page, 1);
1320 /* if ipage exists, blkaddr should be NEW_ADDR */
1321 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1322 page = f2fs_get_lock_data_page(inode, index, true);
1327 if (new_i_size && i_size_read(inode) <
1328 ((loff_t)(index + 1) << PAGE_SHIFT))
1329 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1333 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1335 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1336 struct f2fs_summary sum;
1337 struct node_info ni;
1338 block_t old_blkaddr;
1342 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1345 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1349 dn->data_blkaddr = f2fs_data_blkaddr(dn);
1350 if (dn->data_blkaddr != NULL_ADDR)
1353 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1357 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1358 old_blkaddr = dn->data_blkaddr;
1359 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1360 &sum, seg_type, NULL);
1361 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1362 invalidate_mapping_pages(META_MAPPING(sbi),
1363 old_blkaddr, old_blkaddr);
1364 f2fs_invalidate_compress_page(sbi, old_blkaddr);
1366 f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1370 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1372 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1374 f2fs_down_read(&sbi->node_change);
1376 f2fs_up_read(&sbi->node_change);
1381 f2fs_unlock_op(sbi);
1386 * f2fs_map_blocks() tries to find or build mapping relationship which
1387 * maps continuous logical blocks to physical blocks, and return such
1388 * info via f2fs_map_blocks structure.
1390 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1391 int create, int flag)
1393 unsigned int maxblocks = map->m_len;
1394 struct dnode_of_data dn;
1395 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1396 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1397 pgoff_t pgofs, end_offset, end;
1398 int err = 0, ofs = 1;
1399 unsigned int ofs_in_node, last_ofs_in_node;
1401 struct extent_info ei = {0, };
1403 unsigned int start_pgofs;
1409 map->m_bdev = inode->i_sb->s_bdev;
1410 map->m_multidev_dio =
1411 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1416 /* it only supports block size == page size */
1417 pgofs = (pgoff_t)map->m_lblk;
1418 end = pgofs + maxblocks;
1420 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1421 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1425 map->m_pblk = ei.blk + pgofs - ei.fofs;
1426 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1427 map->m_flags = F2FS_MAP_MAPPED;
1428 if (map->m_next_extent)
1429 *map->m_next_extent = pgofs + map->m_len;
1431 /* for hardware encryption, but to avoid potential issue in future */
1432 if (flag == F2FS_GET_BLOCK_DIO)
1433 f2fs_wait_on_block_writeback_range(inode,
1434 map->m_pblk, map->m_len);
1436 if (map->m_multidev_dio) {
1437 block_t blk_addr = map->m_pblk;
1439 bidx = f2fs_target_device_index(sbi, map->m_pblk);
1441 map->m_bdev = FDEV(bidx).bdev;
1442 map->m_pblk -= FDEV(bidx).start_blk;
1443 map->m_len = min(map->m_len,
1444 FDEV(bidx).end_blk + 1 - map->m_pblk);
1446 if (map->m_may_create)
1447 f2fs_update_device_state(sbi, inode->i_ino,
1448 blk_addr, map->m_len);
1454 if (map->m_may_create)
1455 f2fs_do_map_lock(sbi, flag, true);
1457 /* When reading holes, we need its node page */
1458 set_new_dnode(&dn, inode, NULL, NULL, 0);
1459 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1461 if (flag == F2FS_GET_BLOCK_BMAP)
1464 if (err == -ENOENT) {
1466 * There is one exceptional case that read_node_page()
1467 * may return -ENOENT due to filesystem has been
1468 * shutdown or cp_error, so force to convert error
1469 * number to EIO for such case.
1471 if (map->m_may_create &&
1472 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1473 f2fs_cp_error(sbi))) {
1479 if (map->m_next_pgofs)
1480 *map->m_next_pgofs =
1481 f2fs_get_next_page_offset(&dn, pgofs);
1482 if (map->m_next_extent)
1483 *map->m_next_extent =
1484 f2fs_get_next_page_offset(&dn, pgofs);
1489 start_pgofs = pgofs;
1491 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1492 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1495 blkaddr = f2fs_data_blkaddr(&dn);
1497 if (__is_valid_data_blkaddr(blkaddr) &&
1498 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1499 err = -EFSCORRUPTED;
1503 if (__is_valid_data_blkaddr(blkaddr)) {
1504 /* use out-place-update for driect IO under LFS mode */
1505 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1506 map->m_may_create) {
1507 err = __allocate_data_block(&dn, map->m_seg_type);
1510 blkaddr = dn.data_blkaddr;
1511 set_inode_flag(inode, FI_APPEND_WRITE);
1515 if (unlikely(f2fs_cp_error(sbi))) {
1519 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1520 if (blkaddr == NULL_ADDR) {
1522 last_ofs_in_node = dn.ofs_in_node;
1525 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1526 flag != F2FS_GET_BLOCK_DIO);
1527 err = __allocate_data_block(&dn,
1530 if (flag == F2FS_GET_BLOCK_PRE_DIO)
1531 file_need_truncate(inode);
1532 set_inode_flag(inode, FI_APPEND_WRITE);
1537 map->m_flags |= F2FS_MAP_NEW;
1538 blkaddr = dn.data_blkaddr;
1540 if (f2fs_compressed_file(inode) &&
1541 f2fs_sanity_check_cluster(&dn) &&
1542 (flag != F2FS_GET_BLOCK_FIEMAP ||
1543 IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1544 err = -EFSCORRUPTED;
1547 if (flag == F2FS_GET_BLOCK_BMAP) {
1551 if (flag == F2FS_GET_BLOCK_PRECACHE)
1553 if (flag == F2FS_GET_BLOCK_FIEMAP &&
1554 blkaddr == NULL_ADDR) {
1555 if (map->m_next_pgofs)
1556 *map->m_next_pgofs = pgofs + 1;
1559 if (flag != F2FS_GET_BLOCK_FIEMAP) {
1560 /* for defragment case */
1561 if (map->m_next_pgofs)
1562 *map->m_next_pgofs = pgofs + 1;
1568 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1571 if (map->m_multidev_dio)
1572 bidx = f2fs_target_device_index(sbi, blkaddr);
1574 if (map->m_len == 0) {
1575 /* preallocated unwritten block should be mapped for fiemap. */
1576 if (blkaddr == NEW_ADDR)
1577 map->m_flags |= F2FS_MAP_UNWRITTEN;
1578 map->m_flags |= F2FS_MAP_MAPPED;
1580 map->m_pblk = blkaddr;
1583 if (map->m_multidev_dio)
1584 map->m_bdev = FDEV(bidx).bdev;
1585 } else if ((map->m_pblk != NEW_ADDR &&
1586 blkaddr == (map->m_pblk + ofs)) ||
1587 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1588 flag == F2FS_GET_BLOCK_PRE_DIO) {
1589 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1601 /* preallocate blocks in batch for one dnode page */
1602 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1603 (pgofs == end || dn.ofs_in_node == end_offset)) {
1605 dn.ofs_in_node = ofs_in_node;
1606 err = f2fs_reserve_new_blocks(&dn, prealloc);
1610 map->m_len += dn.ofs_in_node - ofs_in_node;
1611 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1615 dn.ofs_in_node = end_offset;
1620 else if (dn.ofs_in_node < end_offset)
1623 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1624 if (map->m_flags & F2FS_MAP_MAPPED) {
1625 unsigned int ofs = start_pgofs - map->m_lblk;
1627 f2fs_update_extent_cache_range(&dn,
1628 start_pgofs, map->m_pblk + ofs,
1633 f2fs_put_dnode(&dn);
1635 if (map->m_may_create) {
1636 f2fs_do_map_lock(sbi, flag, false);
1637 f2fs_balance_fs(sbi, dn.node_changed);
1643 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1645 * for hardware encryption, but to avoid potential issue
1648 f2fs_wait_on_block_writeback_range(inode,
1649 map->m_pblk, map->m_len);
1650 invalidate_mapping_pages(META_MAPPING(sbi),
1651 map->m_pblk, map->m_pblk);
1653 if (map->m_multidev_dio) {
1654 block_t blk_addr = map->m_pblk;
1656 bidx = f2fs_target_device_index(sbi, map->m_pblk);
1658 map->m_bdev = FDEV(bidx).bdev;
1659 map->m_pblk -= FDEV(bidx).start_blk;
1661 if (map->m_may_create)
1662 f2fs_update_device_state(sbi, inode->i_ino,
1663 blk_addr, map->m_len);
1665 f2fs_bug_on(sbi, blk_addr + map->m_len >
1666 FDEV(bidx).end_blk + 1);
1670 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1671 if (map->m_flags & F2FS_MAP_MAPPED) {
1672 unsigned int ofs = start_pgofs - map->m_lblk;
1674 f2fs_update_extent_cache_range(&dn,
1675 start_pgofs, map->m_pblk + ofs,
1678 if (map->m_next_extent)
1679 *map->m_next_extent = pgofs + 1;
1681 f2fs_put_dnode(&dn);
1683 if (map->m_may_create) {
1684 f2fs_do_map_lock(sbi, flag, false);
1685 f2fs_balance_fs(sbi, dn.node_changed);
1688 trace_f2fs_map_blocks(inode, map, create, flag, err);
1692 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1694 struct f2fs_map_blocks map;
1698 if (pos + len > i_size_read(inode))
1701 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1702 map.m_next_pgofs = NULL;
1703 map.m_next_extent = NULL;
1704 map.m_seg_type = NO_CHECK_TYPE;
1705 map.m_may_create = false;
1706 last_lblk = F2FS_BLK_ALIGN(pos + len);
1708 while (map.m_lblk < last_lblk) {
1709 map.m_len = last_lblk - map.m_lblk;
1710 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1711 if (err || map.m_len == 0)
1713 map.m_lblk += map.m_len;
1718 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1720 return (bytes >> inode->i_blkbits);
1723 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1725 return (blks << inode->i_blkbits);
1728 static int f2fs_xattr_fiemap(struct inode *inode,
1729 struct fiemap_extent_info *fieinfo)
1731 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1733 struct node_info ni;
1734 __u64 phys = 0, len;
1736 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1739 if (f2fs_has_inline_xattr(inode)) {
1742 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1743 inode->i_ino, false);
1747 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1749 f2fs_put_page(page, 1);
1753 phys = blks_to_bytes(inode, ni.blk_addr);
1754 offset = offsetof(struct f2fs_inode, i_addr) +
1755 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1756 get_inline_xattr_addrs(inode));
1759 len = inline_xattr_size(inode);
1761 f2fs_put_page(page, 1);
1763 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1766 flags |= FIEMAP_EXTENT_LAST;
1768 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1769 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1770 if (err || err == 1)
1775 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1779 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1781 f2fs_put_page(page, 1);
1785 phys = blks_to_bytes(inode, ni.blk_addr);
1786 len = inode->i_sb->s_blocksize;
1788 f2fs_put_page(page, 1);
1790 flags = FIEMAP_EXTENT_LAST;
1794 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1795 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1798 return (err < 0 ? err : 0);
1801 static loff_t max_inode_blocks(struct inode *inode)
1803 loff_t result = ADDRS_PER_INODE(inode);
1804 loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1806 /* two direct node blocks */
1807 result += (leaf_count * 2);
1809 /* two indirect node blocks */
1810 leaf_count *= NIDS_PER_BLOCK;
1811 result += (leaf_count * 2);
1813 /* one double indirect node block */
1814 leaf_count *= NIDS_PER_BLOCK;
1815 result += leaf_count;
1820 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1823 struct f2fs_map_blocks map;
1824 sector_t start_blk, last_blk;
1826 u64 logical = 0, phys = 0, size = 0;
1829 bool compr_cluster = false, compr_appended;
1830 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1831 unsigned int count_in_cluster = 0;
1834 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1835 ret = f2fs_precache_extents(inode);
1840 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1846 maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1847 if (start > maxbytes) {
1852 if (len > maxbytes || (maxbytes - len) < start)
1853 len = maxbytes - start;
1855 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1856 ret = f2fs_xattr_fiemap(inode, fieinfo);
1860 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1861 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1866 if (bytes_to_blks(inode, len) == 0)
1867 len = blks_to_bytes(inode, 1);
1869 start_blk = bytes_to_blks(inode, start);
1870 last_blk = bytes_to_blks(inode, start + len - 1);
1873 memset(&map, 0, sizeof(map));
1874 map.m_lblk = start_blk;
1875 map.m_len = bytes_to_blks(inode, len);
1876 map.m_next_pgofs = &next_pgofs;
1877 map.m_seg_type = NO_CHECK_TYPE;
1879 if (compr_cluster) {
1881 map.m_len = cluster_size - count_in_cluster;
1884 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1889 if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1890 start_blk = next_pgofs;
1892 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1893 max_inode_blocks(inode)))
1896 flags |= FIEMAP_EXTENT_LAST;
1899 compr_appended = false;
1900 /* In a case of compressed cluster, append this to the last extent */
1901 if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) ||
1902 !(map.m_flags & F2FS_MAP_FLAGS))) {
1903 compr_appended = true;
1908 flags |= FIEMAP_EXTENT_MERGED;
1909 if (IS_ENCRYPTED(inode))
1910 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1912 ret = fiemap_fill_next_extent(fieinfo, logical,
1914 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1920 if (start_blk > last_blk)
1924 if (map.m_pblk == COMPRESS_ADDR) {
1925 compr_cluster = true;
1926 count_in_cluster = 1;
1927 } else if (compr_appended) {
1928 unsigned int appended_blks = cluster_size -
1929 count_in_cluster + 1;
1930 size += blks_to_bytes(inode, appended_blks);
1931 start_blk += appended_blks;
1932 compr_cluster = false;
1934 logical = blks_to_bytes(inode, start_blk);
1935 phys = __is_valid_data_blkaddr(map.m_pblk) ?
1936 blks_to_bytes(inode, map.m_pblk) : 0;
1937 size = blks_to_bytes(inode, map.m_len);
1940 if (compr_cluster) {
1941 flags = FIEMAP_EXTENT_ENCODED;
1942 count_in_cluster += map.m_len;
1943 if (count_in_cluster == cluster_size) {
1944 compr_cluster = false;
1945 size += blks_to_bytes(inode, 1);
1947 } else if (map.m_flags & F2FS_MAP_UNWRITTEN) {
1948 flags = FIEMAP_EXTENT_UNWRITTEN;
1951 start_blk += bytes_to_blks(inode, size);
1956 if (fatal_signal_pending(current))
1964 inode_unlock(inode);
1968 static inline loff_t f2fs_readpage_limit(struct inode *inode)
1970 if (IS_ENABLED(CONFIG_FS_VERITY) &&
1971 (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
1972 return inode->i_sb->s_maxbytes;
1974 return i_size_read(inode);
1977 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1979 struct f2fs_map_blocks *map,
1980 struct bio **bio_ret,
1981 sector_t *last_block_in_bio,
1984 struct bio *bio = *bio_ret;
1985 const unsigned blocksize = blks_to_bytes(inode, 1);
1986 sector_t block_in_file;
1987 sector_t last_block;
1988 sector_t last_block_in_file;
1992 block_in_file = (sector_t)page_index(page);
1993 last_block = block_in_file + nr_pages;
1994 last_block_in_file = bytes_to_blks(inode,
1995 f2fs_readpage_limit(inode) + blocksize - 1);
1996 if (last_block > last_block_in_file)
1997 last_block = last_block_in_file;
1999 /* just zeroing out page which is beyond EOF */
2000 if (block_in_file >= last_block)
2003 * Map blocks using the previous result first.
2005 if ((map->m_flags & F2FS_MAP_MAPPED) &&
2006 block_in_file > map->m_lblk &&
2007 block_in_file < (map->m_lblk + map->m_len))
2011 * Then do more f2fs_map_blocks() calls until we are
2012 * done with this page.
2014 map->m_lblk = block_in_file;
2015 map->m_len = last_block - block_in_file;
2017 ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2021 if ((map->m_flags & F2FS_MAP_MAPPED)) {
2022 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2023 SetPageMappedToDisk(page);
2025 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2026 DATA_GENERIC_ENHANCE_READ)) {
2027 ret = -EFSCORRUPTED;
2032 zero_user_segment(page, 0, PAGE_SIZE);
2033 if (f2fs_need_verity(inode, page->index) &&
2034 !fsverity_verify_page(page)) {
2038 if (!PageUptodate(page))
2039 SetPageUptodate(page);
2045 * This page will go to BIO. Do we need to send this
2048 if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2049 *last_block_in_bio, block_nr) ||
2050 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2052 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2056 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2057 is_readahead ? REQ_RAHEAD : 0, page->index,
2067 * If the page is under writeback, we need to wait for
2068 * its completion to see the correct decrypted data.
2070 f2fs_wait_on_block_writeback(inode, block_nr);
2072 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2073 goto submit_and_realloc;
2075 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2076 f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2077 ClearPageError(page);
2078 *last_block_in_bio = block_nr;
2085 #ifdef CONFIG_F2FS_FS_COMPRESSION
2086 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2087 unsigned nr_pages, sector_t *last_block_in_bio,
2088 bool is_readahead, bool for_write)
2090 struct dnode_of_data dn;
2091 struct inode *inode = cc->inode;
2092 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2093 struct bio *bio = *bio_ret;
2094 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2095 sector_t last_block_in_file;
2096 const unsigned blocksize = blks_to_bytes(inode, 1);
2097 struct decompress_io_ctx *dic = NULL;
2098 struct extent_info ei = {0, };
2099 bool from_dnode = true;
2103 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2105 last_block_in_file = bytes_to_blks(inode,
2106 f2fs_readpage_limit(inode) + blocksize - 1);
2108 /* get rid of pages beyond EOF */
2109 for (i = 0; i < cc->cluster_size; i++) {
2110 struct page *page = cc->rpages[i];
2114 if ((sector_t)page->index >= last_block_in_file) {
2115 zero_user_segment(page, 0, PAGE_SIZE);
2116 if (!PageUptodate(page))
2117 SetPageUptodate(page);
2118 } else if (!PageUptodate(page)) {
2124 cc->rpages[i] = NULL;
2128 /* we are done since all pages are beyond EOF */
2129 if (f2fs_cluster_is_empty(cc))
2132 if (f2fs_lookup_extent_cache(inode, start_idx, &ei))
2136 goto skip_reading_dnode;
2138 set_new_dnode(&dn, inode, NULL, NULL, 0);
2139 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2143 f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2146 for (i = 1; i < cc->cluster_size; i++) {
2149 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2150 dn.ofs_in_node + i) :
2153 if (!__is_valid_data_blkaddr(blkaddr))
2156 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2162 if (!from_dnode && i >= ei.c_len)
2166 /* nothing to decompress */
2167 if (cc->nr_cpages == 0) {
2172 dic = f2fs_alloc_dic(cc);
2178 for (i = 0; i < cc->nr_cpages; i++) {
2179 struct page *page = dic->cpages[i];
2181 struct bio_post_read_ctx *ctx;
2183 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2184 dn.ofs_in_node + i + 1) :
2187 f2fs_wait_on_block_writeback(inode, blkaddr);
2189 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2190 if (atomic_dec_and_test(&dic->remaining_pages))
2191 f2fs_decompress_cluster(dic);
2195 if (bio && (!page_is_mergeable(sbi, bio,
2196 *last_block_in_bio, blkaddr) ||
2197 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2199 __submit_bio(sbi, bio, DATA);
2204 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2205 is_readahead ? REQ_RAHEAD : 0,
2206 page->index, for_write);
2209 f2fs_decompress_end_io(dic, ret);
2210 f2fs_put_dnode(&dn);
2216 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2217 goto submit_and_realloc;
2219 ctx = get_post_read_ctx(bio);
2220 ctx->enabled_steps |= STEP_DECOMPRESS;
2221 refcount_inc(&dic->refcnt);
2223 inc_page_count(sbi, F2FS_RD_DATA);
2224 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2225 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2226 ClearPageError(page);
2227 *last_block_in_bio = blkaddr;
2231 f2fs_put_dnode(&dn);
2238 f2fs_put_dnode(&dn);
2240 for (i = 0; i < cc->cluster_size; i++) {
2241 if (cc->rpages[i]) {
2242 ClearPageUptodate(cc->rpages[i]);
2243 ClearPageError(cc->rpages[i]);
2244 unlock_page(cc->rpages[i]);
2253 * This function was originally taken from fs/mpage.c, and customized for f2fs.
2254 * Major change was from block_size == page_size in f2fs by default.
2256 static int f2fs_mpage_readpages(struct inode *inode,
2257 struct readahead_control *rac, struct page *page)
2259 struct bio *bio = NULL;
2260 sector_t last_block_in_bio = 0;
2261 struct f2fs_map_blocks map;
2262 #ifdef CONFIG_F2FS_FS_COMPRESSION
2263 struct compress_ctx cc = {
2265 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2266 .cluster_size = F2FS_I(inode)->i_cluster_size,
2267 .cluster_idx = NULL_CLUSTER,
2273 pgoff_t nc_cluster_idx = NULL_CLUSTER;
2275 unsigned nr_pages = rac ? readahead_count(rac) : 1;
2276 unsigned max_nr_pages = nr_pages;
2283 map.m_next_pgofs = NULL;
2284 map.m_next_extent = NULL;
2285 map.m_seg_type = NO_CHECK_TYPE;
2286 map.m_may_create = false;
2288 for (; nr_pages; nr_pages--) {
2290 page = readahead_page(rac);
2291 prefetchw(&page->flags);
2294 #ifdef CONFIG_F2FS_FS_COMPRESSION
2295 if (f2fs_compressed_file(inode)) {
2296 /* there are remained comressed pages, submit them */
2297 if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2298 ret = f2fs_read_multi_pages(&cc, &bio,
2301 rac != NULL, false);
2302 f2fs_destroy_compress_ctx(&cc, false);
2304 goto set_error_page;
2306 if (cc.cluster_idx == NULL_CLUSTER) {
2307 if (nc_cluster_idx ==
2308 page->index >> cc.log_cluster_size) {
2309 goto read_single_page;
2312 ret = f2fs_is_compressed_cluster(inode, page->index);
2314 goto set_error_page;
2317 page->index >> cc.log_cluster_size;
2318 goto read_single_page;
2321 nc_cluster_idx = NULL_CLUSTER;
2323 ret = f2fs_init_compress_ctx(&cc);
2325 goto set_error_page;
2327 f2fs_compress_ctx_add_page(&cc, page);
2334 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2335 &bio, &last_block_in_bio, rac);
2337 #ifdef CONFIG_F2FS_FS_COMPRESSION
2341 zero_user_segment(page, 0, PAGE_SIZE);
2344 #ifdef CONFIG_F2FS_FS_COMPRESSION
2350 #ifdef CONFIG_F2FS_FS_COMPRESSION
2351 if (f2fs_compressed_file(inode)) {
2353 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2354 ret = f2fs_read_multi_pages(&cc, &bio,
2357 rac != NULL, false);
2358 f2fs_destroy_compress_ctx(&cc, false);
2364 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2368 static int f2fs_read_data_page(struct file *file, struct page *page)
2370 struct inode *inode = page_file_mapping(page)->host;
2373 trace_f2fs_readpage(page, DATA);
2375 if (!f2fs_is_compress_backend_ready(inode)) {
2380 /* If the file has inline data, try to read it directly */
2381 if (f2fs_has_inline_data(inode))
2382 ret = f2fs_read_inline_data(inode, page);
2384 ret = f2fs_mpage_readpages(inode, NULL, page);
2388 static void f2fs_readahead(struct readahead_control *rac)
2390 struct inode *inode = rac->mapping->host;
2392 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2394 if (!f2fs_is_compress_backend_ready(inode))
2397 /* If the file has inline data, skip readpages */
2398 if (f2fs_has_inline_data(inode))
2401 f2fs_mpage_readpages(inode, rac, NULL);
2404 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2406 struct inode *inode = fio->page->mapping->host;
2407 struct page *mpage, *page;
2408 gfp_t gfp_flags = GFP_NOFS;
2410 if (!f2fs_encrypted_file(inode))
2413 page = fio->compressed_page ? fio->compressed_page : fio->page;
2415 /* wait for GCed page writeback via META_MAPPING */
2416 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2418 if (fscrypt_inode_uses_inline_crypto(inode))
2422 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2423 PAGE_SIZE, 0, gfp_flags);
2424 if (IS_ERR(fio->encrypted_page)) {
2425 /* flush pending IOs and wait for a while in the ENOMEM case */
2426 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2427 f2fs_flush_merged_writes(fio->sbi);
2428 memalloc_retry_wait(GFP_NOFS);
2429 gfp_flags |= __GFP_NOFAIL;
2432 return PTR_ERR(fio->encrypted_page);
2435 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2437 if (PageUptodate(mpage))
2438 memcpy(page_address(mpage),
2439 page_address(fio->encrypted_page), PAGE_SIZE);
2440 f2fs_put_page(mpage, 1);
2445 static inline bool check_inplace_update_policy(struct inode *inode,
2446 struct f2fs_io_info *fio)
2448 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2449 unsigned int policy = SM_I(sbi)->ipu_policy;
2451 if (policy & (0x1 << F2FS_IPU_HONOR_OPU_WRITE) &&
2452 is_inode_flag_set(inode, FI_OPU_WRITE))
2454 if (policy & (0x1 << F2FS_IPU_FORCE))
2456 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2458 if (policy & (0x1 << F2FS_IPU_UTIL) &&
2459 utilization(sbi) > SM_I(sbi)->min_ipu_util)
2461 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2462 utilization(sbi) > SM_I(sbi)->min_ipu_util)
2466 * IPU for rewrite async pages
2468 if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2469 fio && fio->op == REQ_OP_WRITE &&
2470 !(fio->op_flags & REQ_SYNC) &&
2471 !IS_ENCRYPTED(inode))
2474 /* this is only set during fdatasync */
2475 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2476 is_inode_flag_set(inode, FI_NEED_IPU))
2479 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2480 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2486 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2488 /* swap file is migrating in aligned write mode */
2489 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2492 if (f2fs_is_pinned_file(inode))
2495 /* if this is cold file, we should overwrite to avoid fragmentation */
2496 if (file_is_cold(inode))
2499 return check_inplace_update_policy(inode, fio);
2502 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2504 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2506 /* The below cases were checked when setting it. */
2507 if (f2fs_is_pinned_file(inode))
2509 if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2511 if (f2fs_lfs_mode(sbi))
2513 if (S_ISDIR(inode->i_mode))
2515 if (IS_NOQUOTA(inode))
2517 if (f2fs_is_atomic_file(inode))
2520 /* swap file is migrating in aligned write mode */
2521 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2524 if (is_inode_flag_set(inode, FI_OPU_WRITE))
2528 if (page_private_gcing(fio->page))
2530 if (page_private_dummy(fio->page))
2532 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2533 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2539 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2541 struct inode *inode = fio->page->mapping->host;
2543 if (f2fs_should_update_outplace(inode, fio))
2546 return f2fs_should_update_inplace(inode, fio);
2549 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2551 struct page *page = fio->page;
2552 struct inode *inode = page->mapping->host;
2553 struct dnode_of_data dn;
2554 struct extent_info ei = {0, };
2555 struct node_info ni;
2556 bool ipu_force = false;
2559 set_new_dnode(&dn, inode, NULL, NULL, 0);
2560 if (need_inplace_update(fio) &&
2561 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2562 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2564 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2565 DATA_GENERIC_ENHANCE))
2566 return -EFSCORRUPTED;
2569 fio->need_lock = LOCK_DONE;
2573 /* Deadlock due to between page->lock and f2fs_lock_op */
2574 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2577 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2581 fio->old_blkaddr = dn.data_blkaddr;
2583 /* This page is already truncated */
2584 if (fio->old_blkaddr == NULL_ADDR) {
2585 ClearPageUptodate(page);
2586 clear_page_private_gcing(page);
2590 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2591 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2592 DATA_GENERIC_ENHANCE)) {
2593 err = -EFSCORRUPTED;
2597 * If current allocation needs SSR,
2598 * it had better in-place writes for updated data.
2601 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2602 need_inplace_update(fio))) {
2603 err = f2fs_encrypt_one_page(fio);
2607 set_page_writeback(page);
2608 ClearPageError(page);
2609 f2fs_put_dnode(&dn);
2610 if (fio->need_lock == LOCK_REQ)
2611 f2fs_unlock_op(fio->sbi);
2612 err = f2fs_inplace_write_data(fio);
2614 if (fscrypt_inode_uses_fs_layer_crypto(inode))
2615 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2616 if (PageWriteback(page))
2617 end_page_writeback(page);
2619 set_inode_flag(inode, FI_UPDATE_WRITE);
2621 trace_f2fs_do_write_data_page(fio->page, IPU);
2625 if (fio->need_lock == LOCK_RETRY) {
2626 if (!f2fs_trylock_op(fio->sbi)) {
2630 fio->need_lock = LOCK_REQ;
2633 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2637 fio->version = ni.version;
2639 err = f2fs_encrypt_one_page(fio);
2643 set_page_writeback(page);
2644 ClearPageError(page);
2646 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2647 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2649 /* LFS mode write path */
2650 f2fs_outplace_write_data(&dn, fio);
2651 trace_f2fs_do_write_data_page(page, OPU);
2652 set_inode_flag(inode, FI_APPEND_WRITE);
2653 if (page->index == 0)
2654 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2656 f2fs_put_dnode(&dn);
2658 if (fio->need_lock == LOCK_REQ)
2659 f2fs_unlock_op(fio->sbi);
2663 int f2fs_write_single_data_page(struct page *page, int *submitted,
2665 sector_t *last_block,
2666 struct writeback_control *wbc,
2667 enum iostat_type io_type,
2671 struct inode *inode = page->mapping->host;
2672 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2673 loff_t i_size = i_size_read(inode);
2674 const pgoff_t end_index = ((unsigned long long)i_size)
2676 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2677 unsigned offset = 0;
2678 bool need_balance_fs = false;
2680 struct f2fs_io_info fio = {
2682 .ino = inode->i_ino,
2685 .op_flags = wbc_to_write_flags(wbc),
2686 .old_blkaddr = NULL_ADDR,
2688 .encrypted_page = NULL,
2690 .compr_blocks = compr_blocks,
2691 .need_lock = LOCK_RETRY,
2695 .last_block = last_block,
2698 trace_f2fs_writepage(page, DATA);
2700 /* we should bypass data pages to proceed the kworkder jobs */
2701 if (unlikely(f2fs_cp_error(sbi))) {
2702 mapping_set_error(page->mapping, -EIO);
2704 * don't drop any dirty dentry pages for keeping lastest
2705 * directory structure.
2707 if (S_ISDIR(inode->i_mode))
2712 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2715 if (page->index < end_index ||
2716 f2fs_verity_in_progress(inode) ||
2721 * If the offset is out-of-range of file size,
2722 * this page does not have to be written to disk.
2724 offset = i_size & (PAGE_SIZE - 1);
2725 if ((page->index >= end_index + 1) || !offset)
2728 zero_user_segment(page, offset, PAGE_SIZE);
2730 if (f2fs_is_drop_cache(inode))
2732 /* we should not write 0'th page having journal header */
2733 if (f2fs_is_volatile_file(inode) && (!page->index ||
2734 (!wbc->for_reclaim &&
2735 f2fs_available_free_memory(sbi, BASE_CHECK))))
2738 /* Dentry/quota blocks are controlled by checkpoint */
2739 if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2741 * We need to wait for node_write to avoid block allocation during
2742 * checkpoint. This can only happen to quota writes which can cause
2743 * the below discard race condition.
2745 if (IS_NOQUOTA(inode))
2746 f2fs_down_read(&sbi->node_write);
2748 fio.need_lock = LOCK_DONE;
2749 err = f2fs_do_write_data_page(&fio);
2751 if (IS_NOQUOTA(inode))
2752 f2fs_up_read(&sbi->node_write);
2757 if (!wbc->for_reclaim)
2758 need_balance_fs = true;
2759 else if (has_not_enough_free_secs(sbi, 0, 0))
2762 set_inode_flag(inode, FI_HOT_DATA);
2765 if (f2fs_has_inline_data(inode)) {
2766 err = f2fs_write_inline_data(inode, page);
2771 if (err == -EAGAIN) {
2772 err = f2fs_do_write_data_page(&fio);
2773 if (err == -EAGAIN) {
2774 fio.need_lock = LOCK_REQ;
2775 err = f2fs_do_write_data_page(&fio);
2780 file_set_keep_isize(inode);
2782 spin_lock(&F2FS_I(inode)->i_size_lock);
2783 if (F2FS_I(inode)->last_disk_size < psize)
2784 F2FS_I(inode)->last_disk_size = psize;
2785 spin_unlock(&F2FS_I(inode)->i_size_lock);
2789 if (err && err != -ENOENT)
2793 inode_dec_dirty_pages(inode);
2795 ClearPageUptodate(page);
2796 clear_page_private_gcing(page);
2799 if (wbc->for_reclaim) {
2800 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2801 clear_inode_flag(inode, FI_HOT_DATA);
2802 f2fs_remove_dirty_inode(inode);
2806 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2807 !F2FS_I(inode)->cp_task && allow_balance)
2808 f2fs_balance_fs(sbi, need_balance_fs);
2810 if (unlikely(f2fs_cp_error(sbi))) {
2811 f2fs_submit_merged_write(sbi, DATA);
2812 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2817 *submitted = fio.submitted ? 1 : 0;
2822 redirty_page_for_writepage(wbc, page);
2824 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2825 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2826 * file_write_and_wait_range() will see EIO error, which is critical
2827 * to return value of fsync() followed by atomic_write failure to user.
2829 if (!err || wbc->for_reclaim)
2830 return AOP_WRITEPAGE_ACTIVATE;
2835 static int f2fs_write_data_page(struct page *page,
2836 struct writeback_control *wbc)
2838 #ifdef CONFIG_F2FS_FS_COMPRESSION
2839 struct inode *inode = page->mapping->host;
2841 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2844 if (f2fs_compressed_file(inode)) {
2845 if (f2fs_is_compressed_cluster(inode, page->index)) {
2846 redirty_page_for_writepage(wbc, page);
2847 return AOP_WRITEPAGE_ACTIVATE;
2853 return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2854 wbc, FS_DATA_IO, 0, true);
2858 * This function was copied from write_cche_pages from mm/page-writeback.c.
2859 * The major change is making write step of cold data page separately from
2860 * warm/hot data page.
2862 static int f2fs_write_cache_pages(struct address_space *mapping,
2863 struct writeback_control *wbc,
2864 enum iostat_type io_type)
2867 int done = 0, retry = 0;
2868 struct pagevec pvec;
2869 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2870 struct bio *bio = NULL;
2871 sector_t last_block;
2872 #ifdef CONFIG_F2FS_FS_COMPRESSION
2873 struct inode *inode = mapping->host;
2874 struct compress_ctx cc = {
2876 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2877 .cluster_size = F2FS_I(inode)->i_cluster_size,
2878 .cluster_idx = NULL_CLUSTER,
2882 .valid_nr_cpages = 0,
2885 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2891 pgoff_t end; /* Inclusive */
2893 int range_whole = 0;
2899 pagevec_init(&pvec);
2901 if (get_dirty_pages(mapping->host) <=
2902 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2903 set_inode_flag(mapping->host, FI_HOT_DATA);
2905 clear_inode_flag(mapping->host, FI_HOT_DATA);
2907 if (wbc->range_cyclic) {
2908 index = mapping->writeback_index; /* prev offset */
2911 index = wbc->range_start >> PAGE_SHIFT;
2912 end = wbc->range_end >> PAGE_SHIFT;
2913 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2916 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2917 tag = PAGECACHE_TAG_TOWRITE;
2919 tag = PAGECACHE_TAG_DIRTY;
2922 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2923 tag_pages_for_writeback(mapping, index, end);
2925 while (!done && !retry && (index <= end)) {
2926 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2931 for (i = 0; i < nr_pages; i++) {
2932 struct page *page = pvec.pages[i];
2936 #ifdef CONFIG_F2FS_FS_COMPRESSION
2937 if (f2fs_compressed_file(inode)) {
2938 void *fsdata = NULL;
2942 ret = f2fs_init_compress_ctx(&cc);
2948 if (!f2fs_cluster_can_merge_page(&cc,
2950 ret = f2fs_write_multi_pages(&cc,
2951 &submitted, wbc, io_type);
2957 if (unlikely(f2fs_cp_error(sbi)))
2960 if (!f2fs_cluster_is_empty(&cc))
2963 ret2 = f2fs_prepare_compress_overwrite(
2965 page->index, &fsdata);
2971 (!f2fs_compress_write_end(inode,
2972 fsdata, page->index, 1) ||
2973 !f2fs_all_cluster_page_loaded(&cc,
2974 &pvec, i, nr_pages))) {
2980 /* give a priority to WB_SYNC threads */
2981 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2982 wbc->sync_mode == WB_SYNC_NONE) {
2986 #ifdef CONFIG_F2FS_FS_COMPRESSION
2989 done_index = page->index;
2993 if (unlikely(page->mapping != mapping)) {
2999 if (!PageDirty(page)) {
3000 /* someone wrote it for us */
3001 goto continue_unlock;
3004 if (PageWriteback(page)) {
3005 if (wbc->sync_mode != WB_SYNC_NONE)
3006 f2fs_wait_on_page_writeback(page,
3009 goto continue_unlock;
3012 if (!clear_page_dirty_for_io(page))
3013 goto continue_unlock;
3015 #ifdef CONFIG_F2FS_FS_COMPRESSION
3016 if (f2fs_compressed_file(inode)) {
3018 f2fs_compress_ctx_add_page(&cc, page);
3022 ret = f2fs_write_single_data_page(page, &submitted,
3023 &bio, &last_block, wbc, io_type,
3025 if (ret == AOP_WRITEPAGE_ACTIVATE)
3027 #ifdef CONFIG_F2FS_FS_COMPRESSION
3030 nwritten += submitted;
3031 wbc->nr_to_write -= submitted;
3033 if (unlikely(ret)) {
3035 * keep nr_to_write, since vfs uses this to
3036 * get # of written pages.
3038 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3041 } else if (ret == -EAGAIN) {
3043 if (wbc->sync_mode == WB_SYNC_ALL) {
3044 f2fs_io_schedule_timeout(
3045 DEFAULT_IO_TIMEOUT);
3050 done_index = page->index + 1;
3055 if (wbc->nr_to_write <= 0 &&
3056 wbc->sync_mode == WB_SYNC_NONE) {
3064 pagevec_release(&pvec);
3067 #ifdef CONFIG_F2FS_FS_COMPRESSION
3068 /* flush remained pages in compress cluster */
3069 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3070 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3071 nwritten += submitted;
3072 wbc->nr_to_write -= submitted;
3078 if (f2fs_compressed_file(inode))
3079 f2fs_destroy_compress_ctx(&cc, false);
3086 if (wbc->range_cyclic && !done)
3088 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3089 mapping->writeback_index = done_index;
3092 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3094 /* submit cached bio of IPU write */
3096 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3101 static inline bool __should_serialize_io(struct inode *inode,
3102 struct writeback_control *wbc)
3104 /* to avoid deadlock in path of data flush */
3105 if (F2FS_I(inode)->cp_task)
3108 if (!S_ISREG(inode->i_mode))
3110 if (IS_NOQUOTA(inode))
3113 if (f2fs_need_compress_data(inode))
3115 if (wbc->sync_mode != WB_SYNC_ALL)
3117 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3122 static int __f2fs_write_data_pages(struct address_space *mapping,
3123 struct writeback_control *wbc,
3124 enum iostat_type io_type)
3126 struct inode *inode = mapping->host;
3127 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3128 struct blk_plug plug;
3130 bool locked = false;
3132 /* deal with chardevs and other special file */
3133 if (!mapping->a_ops->writepage)
3136 /* skip writing if there is no dirty page in this inode */
3137 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3140 /* during POR, we don't need to trigger writepage at all. */
3141 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3144 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3145 wbc->sync_mode == WB_SYNC_NONE &&
3146 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3147 f2fs_available_free_memory(sbi, DIRTY_DENTS))
3150 /* skip writing in file defragment preparing stage */
3151 if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3154 trace_f2fs_writepages(mapping->host, wbc, DATA);
3156 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3157 if (wbc->sync_mode == WB_SYNC_ALL)
3158 atomic_inc(&sbi->wb_sync_req[DATA]);
3159 else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3160 /* to avoid potential deadlock */
3162 blk_finish_plug(current->plug);
3166 if (__should_serialize_io(inode, wbc)) {
3167 mutex_lock(&sbi->writepages);
3171 blk_start_plug(&plug);
3172 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3173 blk_finish_plug(&plug);
3176 mutex_unlock(&sbi->writepages);
3178 if (wbc->sync_mode == WB_SYNC_ALL)
3179 atomic_dec(&sbi->wb_sync_req[DATA]);
3181 * if some pages were truncated, we cannot guarantee its mapping->host
3182 * to detect pending bios.
3185 f2fs_remove_dirty_inode(inode);
3189 wbc->pages_skipped += get_dirty_pages(inode);
3190 trace_f2fs_writepages(mapping->host, wbc, DATA);
3194 static int f2fs_write_data_pages(struct address_space *mapping,
3195 struct writeback_control *wbc)
3197 struct inode *inode = mapping->host;
3199 return __f2fs_write_data_pages(mapping, wbc,
3200 F2FS_I(inode)->cp_task == current ?
3201 FS_CP_DATA_IO : FS_DATA_IO);
3204 void f2fs_write_failed(struct inode *inode, loff_t to)
3206 loff_t i_size = i_size_read(inode);
3208 if (IS_NOQUOTA(inode))
3211 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3212 if (to > i_size && !f2fs_verity_in_progress(inode)) {
3213 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3214 filemap_invalidate_lock(inode->i_mapping);
3216 truncate_pagecache(inode, i_size);
3217 f2fs_truncate_blocks(inode, i_size, true);
3219 filemap_invalidate_unlock(inode->i_mapping);
3220 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3224 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3225 struct page *page, loff_t pos, unsigned len,
3226 block_t *blk_addr, bool *node_changed)
3228 struct inode *inode = page->mapping->host;
3229 pgoff_t index = page->index;
3230 struct dnode_of_data dn;
3232 bool locked = false;
3233 struct extent_info ei = {0, };
3238 * If a whole page is being written and we already preallocated all the
3239 * blocks, then there is no need to get a block address now.
3241 if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3244 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3245 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3246 flag = F2FS_GET_BLOCK_DEFAULT;
3248 flag = F2FS_GET_BLOCK_PRE_AIO;
3250 if (f2fs_has_inline_data(inode) ||
3251 (pos & PAGE_MASK) >= i_size_read(inode)) {
3252 f2fs_do_map_lock(sbi, flag, true);
3257 /* check inline_data */
3258 ipage = f2fs_get_node_page(sbi, inode->i_ino);
3259 if (IS_ERR(ipage)) {
3260 err = PTR_ERR(ipage);
3264 set_new_dnode(&dn, inode, ipage, ipage, 0);
3266 if (f2fs_has_inline_data(inode)) {
3267 if (pos + len <= MAX_INLINE_DATA(inode)) {
3268 f2fs_do_read_inline_data(page, ipage);
3269 set_inode_flag(inode, FI_DATA_EXIST);
3271 set_page_private_inline(ipage);
3273 err = f2fs_convert_inline_page(&dn, page);
3276 if (dn.data_blkaddr == NULL_ADDR)
3277 err = f2fs_get_block(&dn, index);
3279 } else if (locked) {
3280 err = f2fs_get_block(&dn, index);
3282 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3283 dn.data_blkaddr = ei.blk + index - ei.fofs;
3286 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3287 if (err || dn.data_blkaddr == NULL_ADDR) {
3288 f2fs_put_dnode(&dn);
3289 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3291 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3298 /* convert_inline_page can make node_changed */
3299 *blk_addr = dn.data_blkaddr;
3300 *node_changed = dn.node_changed;
3302 f2fs_put_dnode(&dn);
3305 f2fs_do_map_lock(sbi, flag, false);
3309 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3310 loff_t pos, unsigned len, unsigned flags,
3311 struct page **pagep, void **fsdata)
3313 struct inode *inode = mapping->host;
3314 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3315 struct page *page = NULL;
3316 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3317 bool need_balance = false, drop_atomic = false;
3318 block_t blkaddr = NULL_ADDR;
3321 trace_f2fs_write_begin(inode, pos, len, flags);
3323 if (!f2fs_is_checkpoint_ready(sbi)) {
3328 if ((f2fs_is_atomic_file(inode) &&
3329 !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3330 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3337 * We should check this at this moment to avoid deadlock on inode page
3338 * and #0 page. The locking rule for inline_data conversion should be:
3339 * lock_page(page #0) -> lock_page(inode_page)
3342 err = f2fs_convert_inline_inode(inode);
3347 #ifdef CONFIG_F2FS_FS_COMPRESSION
3348 if (f2fs_compressed_file(inode)) {
3353 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3356 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3369 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3370 * wait_for_stable_page. Will wait that below with our IO control.
3372 page = f2fs_pagecache_get_page(mapping, index,
3373 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3379 /* TODO: cluster can be compressed due to race with .writepage */
3383 err = prepare_write_begin(sbi, page, pos, len,
3384 &blkaddr, &need_balance);
3388 if (need_balance && !IS_NOQUOTA(inode) &&
3389 has_not_enough_free_secs(sbi, 0, 0)) {
3391 f2fs_balance_fs(sbi, true);
3393 if (page->mapping != mapping) {
3394 /* The page got truncated from under us */
3395 f2fs_put_page(page, 1);
3400 f2fs_wait_on_page_writeback(page, DATA, false, true);
3402 if (len == PAGE_SIZE || PageUptodate(page))
3405 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3406 !f2fs_verity_in_progress(inode)) {
3407 zero_user_segment(page, len, PAGE_SIZE);
3411 if (blkaddr == NEW_ADDR) {
3412 zero_user_segment(page, 0, PAGE_SIZE);
3413 SetPageUptodate(page);
3415 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3416 DATA_GENERIC_ENHANCE_READ)) {
3417 err = -EFSCORRUPTED;
3420 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3425 if (unlikely(page->mapping != mapping)) {
3426 f2fs_put_page(page, 1);
3429 if (unlikely(!PageUptodate(page))) {
3437 f2fs_put_page(page, 1);
3438 f2fs_write_failed(inode, pos + len);
3440 f2fs_drop_inmem_pages_all(sbi, false);
3444 static int f2fs_write_end(struct file *file,
3445 struct address_space *mapping,
3446 loff_t pos, unsigned len, unsigned copied,
3447 struct page *page, void *fsdata)
3449 struct inode *inode = page->mapping->host;
3451 trace_f2fs_write_end(inode, pos, len, copied);
3454 * This should be come from len == PAGE_SIZE, and we expect copied
3455 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3456 * let generic_perform_write() try to copy data again through copied=0.
3458 if (!PageUptodate(page)) {
3459 if (unlikely(copied != len))
3462 SetPageUptodate(page);
3465 #ifdef CONFIG_F2FS_FS_COMPRESSION
3466 /* overwrite compressed file */
3467 if (f2fs_compressed_file(inode) && fsdata) {
3468 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3469 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3471 if (pos + copied > i_size_read(inode) &&
3472 !f2fs_verity_in_progress(inode))
3473 f2fs_i_size_write(inode, pos + copied);
3481 set_page_dirty(page);
3483 if (pos + copied > i_size_read(inode) &&
3484 !f2fs_verity_in_progress(inode))
3485 f2fs_i_size_write(inode, pos + copied);
3487 f2fs_put_page(page, 1);
3488 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3492 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3494 struct inode *inode = folio->mapping->host;
3495 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3497 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3498 (offset || length != folio_size(folio)))
3501 if (folio_test_dirty(folio)) {
3502 if (inode->i_ino == F2FS_META_INO(sbi)) {
3503 dec_page_count(sbi, F2FS_DIRTY_META);
3504 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3505 dec_page_count(sbi, F2FS_DIRTY_NODES);
3507 inode_dec_dirty_pages(inode);
3508 f2fs_remove_dirty_inode(inode);
3512 clear_page_private_gcing(&folio->page);
3514 if (test_opt(sbi, COMPRESS_CACHE) &&
3515 inode->i_ino == F2FS_COMPRESS_INO(sbi))
3516 clear_page_private_data(&folio->page);
3518 if (page_private_atomic(&folio->page))
3519 return f2fs_drop_inmem_page(inode, &folio->page);
3521 folio_detach_private(folio);
3524 int f2fs_release_page(struct page *page, gfp_t wait)
3526 /* If this is dirty page, keep PagePrivate */
3527 if (PageDirty(page))
3530 /* This is atomic written page, keep Private */
3531 if (page_private_atomic(page))
3534 if (test_opt(F2FS_P_SB(page), COMPRESS_CACHE)) {
3535 struct inode *inode = page->mapping->host;
3537 if (inode->i_ino == F2FS_COMPRESS_INO(F2FS_I_SB(inode)))
3538 clear_page_private_data(page);
3541 clear_page_private_gcing(page);
3543 detach_page_private(page);
3544 set_page_private(page, 0);
3548 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3549 struct folio *folio)
3551 struct inode *inode = mapping->host;
3553 trace_f2fs_set_page_dirty(&folio->page, DATA);
3555 if (!folio_test_uptodate(folio))
3556 folio_mark_uptodate(folio);
3557 BUG_ON(folio_test_swapcache(folio));
3559 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3560 if (!page_private_atomic(&folio->page)) {
3561 f2fs_register_inmem_page(inode, &folio->page);
3565 * Previously, this page has been registered, we just
3571 if (!folio_test_dirty(folio)) {
3572 filemap_dirty_folio(mapping, folio);
3573 f2fs_update_dirty_folio(inode, folio);
3580 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3582 #ifdef CONFIG_F2FS_FS_COMPRESSION
3583 struct dnode_of_data dn;
3584 sector_t start_idx, blknr = 0;
3587 start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3589 set_new_dnode(&dn, inode, NULL, NULL, 0);
3590 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3594 if (dn.data_blkaddr != COMPRESS_ADDR) {
3595 dn.ofs_in_node += block - start_idx;
3596 blknr = f2fs_data_blkaddr(&dn);
3597 if (!__is_valid_data_blkaddr(blknr))
3601 f2fs_put_dnode(&dn);
3609 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3611 struct inode *inode = mapping->host;
3614 if (f2fs_has_inline_data(inode))
3617 /* make sure allocating whole blocks */
3618 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3619 filemap_write_and_wait(mapping);
3621 /* Block number less than F2FS MAX BLOCKS */
3622 if (unlikely(block >= max_file_blocks(inode)))
3625 if (f2fs_compressed_file(inode)) {
3626 blknr = f2fs_bmap_compress(inode, block);
3628 struct f2fs_map_blocks map;
3630 memset(&map, 0, sizeof(map));
3633 map.m_next_pgofs = NULL;
3634 map.m_seg_type = NO_CHECK_TYPE;
3636 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3640 trace_f2fs_bmap(inode, block, blknr);
3644 #ifdef CONFIG_MIGRATION
3645 #include <linux/migrate.h>
3647 int f2fs_migrate_page(struct address_space *mapping,
3648 struct page *newpage, struct page *page, enum migrate_mode mode)
3650 int rc, extra_count;
3651 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3652 bool atomic_written = page_private_atomic(page);
3654 BUG_ON(PageWriteback(page));
3656 /* migrating an atomic written page is safe with the inmem_lock hold */
3657 if (atomic_written) {
3658 if (mode != MIGRATE_SYNC)
3660 if (!mutex_trylock(&fi->inmem_lock))
3664 /* one extra reference was held for atomic_write page */
3665 extra_count = atomic_written ? 1 : 0;
3666 rc = migrate_page_move_mapping(mapping, newpage,
3668 if (rc != MIGRATEPAGE_SUCCESS) {
3670 mutex_unlock(&fi->inmem_lock);
3674 if (atomic_written) {
3675 struct inmem_pages *cur;
3677 list_for_each_entry(cur, &fi->inmem_pages, list)
3678 if (cur->page == page) {
3679 cur->page = newpage;
3682 mutex_unlock(&fi->inmem_lock);
3687 /* guarantee to start from no stale private field */
3688 set_page_private(newpage, 0);
3689 if (PagePrivate(page)) {
3690 set_page_private(newpage, page_private(page));
3691 SetPagePrivate(newpage);
3694 set_page_private(page, 0);
3695 ClearPagePrivate(page);
3699 if (mode != MIGRATE_SYNC_NO_COPY)
3700 migrate_page_copy(newpage, page);
3702 migrate_page_states(newpage, page);
3704 return MIGRATEPAGE_SUCCESS;
3709 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3710 unsigned int blkcnt)
3712 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3713 unsigned int blkofs;
3714 unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3715 unsigned int secidx = start_blk / blk_per_sec;
3716 unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3719 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3720 filemap_invalidate_lock(inode->i_mapping);
3722 set_inode_flag(inode, FI_ALIGNED_WRITE);
3723 set_inode_flag(inode, FI_OPU_WRITE);
3725 for (; secidx < end_sec; secidx++) {
3726 f2fs_down_write(&sbi->pin_sem);
3729 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3730 f2fs_unlock_op(sbi);
3732 set_inode_flag(inode, FI_SKIP_WRITES);
3734 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3736 unsigned int blkidx = secidx * blk_per_sec + blkofs;
3738 page = f2fs_get_lock_data_page(inode, blkidx, true);
3740 f2fs_up_write(&sbi->pin_sem);
3741 ret = PTR_ERR(page);
3745 set_page_dirty(page);
3746 f2fs_put_page(page, 1);
3749 clear_inode_flag(inode, FI_SKIP_WRITES);
3751 ret = filemap_fdatawrite(inode->i_mapping);
3753 f2fs_up_write(&sbi->pin_sem);
3760 clear_inode_flag(inode, FI_SKIP_WRITES);
3761 clear_inode_flag(inode, FI_OPU_WRITE);
3762 clear_inode_flag(inode, FI_ALIGNED_WRITE);
3764 filemap_invalidate_unlock(inode->i_mapping);
3765 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3770 static int check_swap_activate(struct swap_info_struct *sis,
3771 struct file *swap_file, sector_t *span)
3773 struct address_space *mapping = swap_file->f_mapping;
3774 struct inode *inode = mapping->host;
3775 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3776 sector_t cur_lblock;
3777 sector_t last_lblock;
3779 sector_t lowest_pblock = -1;
3780 sector_t highest_pblock = 0;
3782 unsigned long nr_pblocks;
3783 unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3784 unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3785 unsigned int not_aligned = 0;
3789 * Map all the blocks into the extent list. This code doesn't try
3793 last_lblock = bytes_to_blks(inode, i_size_read(inode));
3795 while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3796 struct f2fs_map_blocks map;
3800 memset(&map, 0, sizeof(map));
3801 map.m_lblk = cur_lblock;
3802 map.m_len = last_lblock - cur_lblock;
3803 map.m_next_pgofs = NULL;
3804 map.m_next_extent = NULL;
3805 map.m_seg_type = NO_CHECK_TYPE;
3806 map.m_may_create = false;
3808 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3813 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3814 f2fs_err(sbi, "Swapfile has holes");
3819 pblock = map.m_pblk;
3820 nr_pblocks = map.m_len;
3822 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
3823 nr_pblocks & sec_blks_mask) {
3826 nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3827 if (cur_lblock + nr_pblocks > sis->max)
3828 nr_pblocks -= blks_per_sec;
3831 /* this extent is last one */
3832 nr_pblocks = map.m_len;
3833 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
3837 ret = f2fs_migrate_blocks(inode, cur_lblock,
3844 if (cur_lblock + nr_pblocks >= sis->max)
3845 nr_pblocks = sis->max - cur_lblock;
3847 if (cur_lblock) { /* exclude the header page */
3848 if (pblock < lowest_pblock)
3849 lowest_pblock = pblock;
3850 if (pblock + nr_pblocks - 1 > highest_pblock)
3851 highest_pblock = pblock + nr_pblocks - 1;
3855 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3857 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3861 cur_lblock += nr_pblocks;
3864 *span = 1 + highest_pblock - lowest_pblock;
3865 if (cur_lblock == 0)
3866 cur_lblock = 1; /* force Empty message */
3867 sis->max = cur_lblock;
3868 sis->pages = cur_lblock - 1;
3869 sis->highest_bit = cur_lblock - 1;
3872 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
3873 not_aligned, blks_per_sec * F2FS_BLKSIZE);
3877 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3880 struct inode *inode = file_inode(file);
3883 if (!S_ISREG(inode->i_mode))
3886 if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3889 if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
3890 f2fs_err(F2FS_I_SB(inode),
3891 "Swapfile not supported in LFS mode");
3895 ret = f2fs_convert_inline_inode(inode);
3899 if (!f2fs_disable_compressed_file(inode))
3902 f2fs_precache_extents(inode);
3904 ret = check_swap_activate(sis, file, span);
3908 set_inode_flag(inode, FI_PIN_FILE);
3909 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3913 static void f2fs_swap_deactivate(struct file *file)
3915 struct inode *inode = file_inode(file);
3917 clear_inode_flag(inode, FI_PIN_FILE);
3920 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3926 static void f2fs_swap_deactivate(struct file *file)
3931 const struct address_space_operations f2fs_dblock_aops = {
3932 .readpage = f2fs_read_data_page,
3933 .readahead = f2fs_readahead,
3934 .writepage = f2fs_write_data_page,
3935 .writepages = f2fs_write_data_pages,
3936 .write_begin = f2fs_write_begin,
3937 .write_end = f2fs_write_end,
3938 .dirty_folio = f2fs_dirty_data_folio,
3939 .invalidate_folio = f2fs_invalidate_folio,
3940 .releasepage = f2fs_release_page,
3941 .direct_IO = noop_direct_IO,
3943 .swap_activate = f2fs_swap_activate,
3944 .swap_deactivate = f2fs_swap_deactivate,
3945 #ifdef CONFIG_MIGRATION
3946 .migratepage = f2fs_migrate_page,
3950 void f2fs_clear_page_cache_dirty_tag(struct page *page)
3952 struct address_space *mapping = page_mapping(page);
3953 unsigned long flags;
3955 xa_lock_irqsave(&mapping->i_pages, flags);
3956 __xa_clear_mark(&mapping->i_pages, page_index(page),
3957 PAGECACHE_TAG_DIRTY);
3958 xa_unlock_irqrestore(&mapping->i_pages, flags);
3961 int __init f2fs_init_post_read_processing(void)
3963 bio_post_read_ctx_cache =
3964 kmem_cache_create("f2fs_bio_post_read_ctx",
3965 sizeof(struct bio_post_read_ctx), 0, 0, NULL);
3966 if (!bio_post_read_ctx_cache)
3968 bio_post_read_ctx_pool =
3969 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
3970 bio_post_read_ctx_cache);
3971 if (!bio_post_read_ctx_pool)
3972 goto fail_free_cache;
3976 kmem_cache_destroy(bio_post_read_ctx_cache);
3981 void f2fs_destroy_post_read_processing(void)
3983 mempool_destroy(bio_post_read_ctx_pool);
3984 kmem_cache_destroy(bio_post_read_ctx_cache);
3987 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
3989 if (!f2fs_sb_has_encrypt(sbi) &&
3990 !f2fs_sb_has_verity(sbi) &&
3991 !f2fs_sb_has_compression(sbi))
3994 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
3995 WQ_UNBOUND | WQ_HIGHPRI,
3997 if (!sbi->post_read_wq)
4002 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4004 if (sbi->post_read_wq)
4005 destroy_workqueue(sbi->post_read_wq);
4008 int __init f2fs_init_bio_entry_cache(void)
4010 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4011 sizeof(struct bio_entry));
4012 if (!bio_entry_slab)
4017 void f2fs_destroy_bio_entry_cache(void)
4019 kmem_cache_destroy(bio_entry_slab);
4022 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4023 unsigned int flags, struct iomap *iomap,
4024 struct iomap *srcmap)
4026 struct f2fs_map_blocks map = {};
4027 pgoff_t next_pgofs = 0;
4030 map.m_lblk = bytes_to_blks(inode, offset);
4031 map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4032 map.m_next_pgofs = &next_pgofs;
4033 map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4034 if (flags & IOMAP_WRITE)
4035 map.m_may_create = true;
4037 err = f2fs_map_blocks(inode, &map, flags & IOMAP_WRITE,
4038 F2FS_GET_BLOCK_DIO);
4042 iomap->offset = blks_to_bytes(inode, map.m_lblk);
4045 * When inline encryption is enabled, sometimes I/O to an encrypted file
4046 * has to be broken up to guarantee DUN contiguity. Handle this by
4047 * limiting the length of the mapping returned.
4049 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4051 if (map.m_flags & (F2FS_MAP_MAPPED | F2FS_MAP_UNWRITTEN)) {
4052 iomap->length = blks_to_bytes(inode, map.m_len);
4053 if (map.m_flags & F2FS_MAP_MAPPED) {
4054 iomap->type = IOMAP_MAPPED;
4055 iomap->flags |= IOMAP_F_MERGED;
4057 iomap->type = IOMAP_UNWRITTEN;
4059 if (WARN_ON_ONCE(!__is_valid_data_blkaddr(map.m_pblk)))
4062 iomap->bdev = map.m_bdev;
4063 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4065 iomap->length = blks_to_bytes(inode, next_pgofs) -
4067 iomap->type = IOMAP_HOLE;
4068 iomap->addr = IOMAP_NULL_ADDR;
4071 if (map.m_flags & F2FS_MAP_NEW)
4072 iomap->flags |= IOMAP_F_NEW;
4073 if ((inode->i_state & I_DIRTY_DATASYNC) ||
4074 offset + length > i_size_read(inode))
4075 iomap->flags |= IOMAP_F_DIRTY;
4080 const struct iomap_ops f2fs_iomap_ops = {
4081 .iomap_begin = f2fs_iomap_begin,