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 readahead() 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 unsigned int f2fs_io_flags(struct f2fs_io_info *fio)
393 unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
394 unsigned int fua_flag, meta_flag, io_flag;
395 unsigned int op_flags = 0;
397 if (fio->op != REQ_OP_WRITE)
399 if (fio->type == DATA)
400 io_flag = fio->sbi->data_io_flag;
401 else if (fio->type == NODE)
402 io_flag = fio->sbi->node_io_flag;
406 fua_flag = io_flag & temp_mask;
407 meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
410 * data/node io flag bits per temp:
411 * REQ_META | REQ_FUA |
412 * 5 | 4 | 3 | 2 | 1 | 0 |
413 * Cold | Warm | Hot | Cold | Warm | Hot |
415 if ((1 << fio->temp) & meta_flag)
416 op_flags |= REQ_META;
417 if ((1 << fio->temp) & fua_flag)
422 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
424 struct f2fs_sb_info *sbi = fio->sbi;
425 struct block_device *bdev;
429 bdev = f2fs_target_device(sbi, fio->new_blkaddr, §or);
430 bio = bio_alloc_bioset(bdev, npages,
431 fio->op | fio->op_flags | f2fs_io_flags(fio),
432 GFP_NOIO, &f2fs_bioset);
433 bio->bi_iter.bi_sector = sector;
434 if (is_read_io(fio->op)) {
435 bio->bi_end_io = f2fs_read_end_io;
436 bio->bi_private = NULL;
438 bio->bi_end_io = f2fs_write_end_io;
439 bio->bi_private = sbi;
441 iostat_alloc_and_bind_ctx(sbi, bio, NULL);
444 wbc_init_bio(fio->io_wbc, bio);
449 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
451 const struct f2fs_io_info *fio,
455 * The f2fs garbage collector sets ->encrypted_page when it wants to
456 * read/write raw data without encryption.
458 if (!fio || !fio->encrypted_page)
459 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
462 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
464 const struct f2fs_io_info *fio)
467 * The f2fs garbage collector sets ->encrypted_page when it wants to
468 * read/write raw data without encryption.
470 if (fio && fio->encrypted_page)
471 return !bio_has_crypt_ctx(bio);
473 return fscrypt_mergeable_bio(bio, inode, next_idx);
476 static inline void __submit_bio(struct f2fs_sb_info *sbi,
477 struct bio *bio, enum page_type type)
479 if (!is_read_io(bio_op(bio))) {
482 if (type != DATA && type != NODE)
485 if (f2fs_lfs_mode(sbi) && current->plug)
486 blk_finish_plug(current->plug);
488 if (!F2FS_IO_ALIGNED(sbi))
491 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
492 start %= F2FS_IO_SIZE(sbi);
497 /* fill dummy pages */
498 for (; start < F2FS_IO_SIZE(sbi); start++) {
500 mempool_alloc(sbi->write_io_dummy,
501 GFP_NOIO | __GFP_NOFAIL);
502 f2fs_bug_on(sbi, !page);
506 zero_user_segment(page, 0, PAGE_SIZE);
507 set_page_private_dummy(page);
509 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
513 * In the NODE case, we lose next block address chain. So, we
514 * need to do checkpoint in f2fs_sync_file.
517 set_sbi_flag(sbi, SBI_NEED_CP);
520 if (is_read_io(bio_op(bio)))
521 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
523 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
525 iostat_update_submit_ctx(bio, type);
529 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
530 struct bio *bio, enum page_type type)
532 __submit_bio(sbi, bio, type);
535 static void __submit_merged_bio(struct f2fs_bio_info *io)
537 struct f2fs_io_info *fio = &io->fio;
542 if (is_read_io(fio->op))
543 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
545 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
547 __submit_bio(io->sbi, io->bio, fio->type);
551 static bool __has_merged_page(struct bio *bio, struct inode *inode,
552 struct page *page, nid_t ino)
554 struct bio_vec *bvec;
555 struct bvec_iter_all iter_all;
560 if (!inode && !page && !ino)
563 bio_for_each_segment_all(bvec, bio, iter_all) {
564 struct page *target = bvec->bv_page;
566 if (fscrypt_is_bounce_page(target)) {
567 target = fscrypt_pagecache_page(target);
571 if (f2fs_is_compressed_page(target)) {
572 target = f2fs_compress_control_page(target);
577 if (inode && inode == target->mapping->host)
579 if (page && page == target)
581 if (ino && ino == ino_of_node(target))
588 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
589 enum page_type type, enum temp_type temp)
591 enum page_type btype = PAGE_TYPE_OF_BIO(type);
592 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
594 f2fs_down_write(&io->io_rwsem);
596 /* change META to META_FLUSH in the checkpoint procedure */
597 if (type >= META_FLUSH) {
598 io->fio.type = META_FLUSH;
599 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
600 if (!test_opt(sbi, NOBARRIER))
601 io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
603 __submit_merged_bio(io);
604 f2fs_up_write(&io->io_rwsem);
607 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
608 struct inode *inode, struct page *page,
609 nid_t ino, enum page_type type, bool force)
614 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
616 enum page_type btype = PAGE_TYPE_OF_BIO(type);
617 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
619 f2fs_down_read(&io->io_rwsem);
620 ret = __has_merged_page(io->bio, inode, page, ino);
621 f2fs_up_read(&io->io_rwsem);
624 __f2fs_submit_merged_write(sbi, type, temp);
626 /* TODO: use HOT temp only for meta pages now. */
632 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
634 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
637 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
638 struct inode *inode, struct page *page,
639 nid_t ino, enum page_type type)
641 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
644 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
646 f2fs_submit_merged_write(sbi, DATA);
647 f2fs_submit_merged_write(sbi, NODE);
648 f2fs_submit_merged_write(sbi, META);
652 * Fill the locked page with data located in the block address.
653 * A caller needs to unlock the page on failure.
655 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
658 struct page *page = fio->encrypted_page ?
659 fio->encrypted_page : fio->page;
661 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
662 fio->is_por ? META_POR : (__is_meta_io(fio) ?
663 META_GENERIC : DATA_GENERIC_ENHANCE)))
664 return -EFSCORRUPTED;
666 trace_f2fs_submit_page_bio(page, fio);
668 /* Allocate a new bio */
669 bio = __bio_alloc(fio, 1);
671 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
672 fio->page->index, fio, GFP_NOIO);
674 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
679 if (fio->io_wbc && !is_read_io(fio->op))
680 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
682 inc_page_count(fio->sbi, is_read_io(fio->op) ?
683 __read_io_type(page): WB_DATA_TYPE(fio->page));
685 __submit_bio(fio->sbi, bio, fio->type);
689 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
690 block_t last_blkaddr, block_t cur_blkaddr)
692 if (unlikely(sbi->max_io_bytes &&
693 bio->bi_iter.bi_size >= sbi->max_io_bytes))
695 if (last_blkaddr + 1 != cur_blkaddr)
697 return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
700 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
701 struct f2fs_io_info *fio)
703 if (io->fio.op != fio->op)
705 return io->fio.op_flags == fio->op_flags;
708 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
709 struct f2fs_bio_info *io,
710 struct f2fs_io_info *fio,
711 block_t last_blkaddr,
714 if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
715 unsigned int filled_blocks =
716 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
717 unsigned int io_size = F2FS_IO_SIZE(sbi);
718 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
720 /* IOs in bio is aligned and left space of vectors is not enough */
721 if (!(filled_blocks % io_size) && left_vecs < io_size)
724 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
726 return io_type_is_mergeable(io, fio);
729 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
730 struct page *page, enum temp_type temp)
732 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
733 struct bio_entry *be;
735 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
739 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
742 f2fs_down_write(&io->bio_list_lock);
743 list_add_tail(&be->list, &io->bio_list);
744 f2fs_up_write(&io->bio_list_lock);
747 static void del_bio_entry(struct bio_entry *be)
750 kmem_cache_free(bio_entry_slab, be);
753 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
756 struct f2fs_sb_info *sbi = fio->sbi;
761 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
762 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
763 struct list_head *head = &io->bio_list;
764 struct bio_entry *be;
766 f2fs_down_write(&io->bio_list_lock);
767 list_for_each_entry(be, head, list) {
773 f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
776 if (f2fs_crypt_mergeable_bio(*bio,
777 fio->page->mapping->host,
778 fio->page->index, fio) &&
779 bio_add_page(*bio, page, PAGE_SIZE, 0) ==
785 /* page can't be merged into bio; submit the bio */
787 __submit_bio(sbi, *bio, DATA);
790 f2fs_up_write(&io->bio_list_lock);
801 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
802 struct bio **bio, struct page *page)
806 struct bio *target = bio ? *bio : NULL;
808 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
809 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
810 struct list_head *head = &io->bio_list;
811 struct bio_entry *be;
813 if (list_empty(head))
816 f2fs_down_read(&io->bio_list_lock);
817 list_for_each_entry(be, head, list) {
819 found = (target == be->bio);
821 found = __has_merged_page(be->bio, NULL,
826 f2fs_up_read(&io->bio_list_lock);
833 f2fs_down_write(&io->bio_list_lock);
834 list_for_each_entry(be, head, list) {
836 found = (target == be->bio);
838 found = __has_merged_page(be->bio, NULL,
846 f2fs_up_write(&io->bio_list_lock);
850 __submit_bio(sbi, target, DATA);
857 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
859 struct bio *bio = *fio->bio;
860 struct page *page = fio->encrypted_page ?
861 fio->encrypted_page : fio->page;
863 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
864 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
865 return -EFSCORRUPTED;
867 trace_f2fs_submit_page_bio(page, fio);
869 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
871 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
874 bio = __bio_alloc(fio, BIO_MAX_VECS);
875 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
876 fio->page->index, fio, GFP_NOIO);
878 add_bio_entry(fio->sbi, bio, page, fio->temp);
880 if (add_ipu_page(fio, &bio, page))
885 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
887 inc_page_count(fio->sbi, WB_DATA_TYPE(page));
889 *fio->last_block = fio->new_blkaddr;
895 void f2fs_submit_page_write(struct f2fs_io_info *fio)
897 struct f2fs_sb_info *sbi = fio->sbi;
898 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
899 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
900 struct page *bio_page;
902 f2fs_bug_on(sbi, is_read_io(fio->op));
904 f2fs_down_write(&io->io_rwsem);
907 spin_lock(&io->io_lock);
908 if (list_empty(&io->io_list)) {
909 spin_unlock(&io->io_lock);
912 fio = list_first_entry(&io->io_list,
913 struct f2fs_io_info, list);
914 list_del(&fio->list);
915 spin_unlock(&io->io_lock);
918 verify_fio_blkaddr(fio);
920 if (fio->encrypted_page)
921 bio_page = fio->encrypted_page;
922 else if (fio->compressed_page)
923 bio_page = fio->compressed_page;
925 bio_page = fio->page;
927 /* set submitted = true as a return value */
928 fio->submitted = true;
930 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
933 (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
935 !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
936 bio_page->index, fio)))
937 __submit_merged_bio(io);
939 if (io->bio == NULL) {
940 if (F2FS_IO_ALIGNED(sbi) &&
941 (fio->type == DATA || fio->type == NODE) &&
942 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
943 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
947 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
948 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
949 bio_page->index, fio, GFP_NOIO);
953 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
954 __submit_merged_bio(io);
959 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
961 io->last_block_in_bio = fio->new_blkaddr;
963 trace_f2fs_submit_page_write(fio->page, fio);
968 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
969 !f2fs_is_checkpoint_ready(sbi))
970 __submit_merged_bio(io);
971 f2fs_up_write(&io->io_rwsem);
974 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
975 unsigned nr_pages, unsigned op_flag,
976 pgoff_t first_idx, bool for_write)
978 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
980 struct bio_post_read_ctx *ctx = NULL;
981 unsigned int post_read_steps = 0;
983 struct block_device *bdev = f2fs_target_device(sbi, blkaddr, §or);
985 bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
986 REQ_OP_READ | op_flag,
987 for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
989 return ERR_PTR(-ENOMEM);
990 bio->bi_iter.bi_sector = sector;
991 f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
992 bio->bi_end_io = f2fs_read_end_io;
994 if (fscrypt_inode_uses_fs_layer_crypto(inode))
995 post_read_steps |= STEP_DECRYPT;
997 if (f2fs_need_verity(inode, first_idx))
998 post_read_steps |= STEP_VERITY;
1001 * STEP_DECOMPRESS is handled specially, since a compressed file might
1002 * contain both compressed and uncompressed clusters. We'll allocate a
1003 * bio_post_read_ctx if the file is compressed, but the caller is
1004 * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1007 if (post_read_steps || f2fs_compressed_file(inode)) {
1008 /* Due to the mempool, this never fails. */
1009 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1012 ctx->enabled_steps = post_read_steps;
1013 ctx->fs_blkaddr = blkaddr;
1014 bio->bi_private = ctx;
1016 iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1021 /* This can handle encryption stuffs */
1022 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1023 block_t blkaddr, int op_flags, bool for_write)
1025 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1028 bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1029 page->index, for_write);
1031 return PTR_ERR(bio);
1033 /* wait for GCed page writeback via META_MAPPING */
1034 f2fs_wait_on_block_writeback(inode, blkaddr);
1036 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1040 ClearPageError(page);
1041 inc_page_count(sbi, F2FS_RD_DATA);
1042 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1043 __submit_bio(sbi, bio, DATA);
1047 static void __set_data_blkaddr(struct dnode_of_data *dn)
1049 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1053 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1054 base = get_extra_isize(dn->inode);
1056 /* Get physical address of data block */
1057 addr_array = blkaddr_in_node(rn);
1058 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1062 * Lock ordering for the change of data block address:
1065 * update block addresses in the node page
1067 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1069 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1070 __set_data_blkaddr(dn);
1071 if (set_page_dirty(dn->node_page))
1072 dn->node_changed = true;
1075 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1077 dn->data_blkaddr = blkaddr;
1078 f2fs_set_data_blkaddr(dn);
1079 f2fs_update_extent_cache(dn);
1082 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1083 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1085 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1091 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1093 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1096 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1097 dn->ofs_in_node, count);
1099 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1101 for (; count > 0; dn->ofs_in_node++) {
1102 block_t blkaddr = f2fs_data_blkaddr(dn);
1104 if (blkaddr == NULL_ADDR) {
1105 dn->data_blkaddr = NEW_ADDR;
1106 __set_data_blkaddr(dn);
1111 if (set_page_dirty(dn->node_page))
1112 dn->node_changed = true;
1116 /* Should keep dn->ofs_in_node unchanged */
1117 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1119 unsigned int ofs_in_node = dn->ofs_in_node;
1122 ret = f2fs_reserve_new_blocks(dn, 1);
1123 dn->ofs_in_node = ofs_in_node;
1127 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1129 bool need_put = dn->inode_page ? false : true;
1132 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1136 if (dn->data_blkaddr == NULL_ADDR)
1137 err = f2fs_reserve_new_block(dn);
1138 if (err || need_put)
1143 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1145 struct extent_info ei = {0, };
1146 struct inode *inode = dn->inode;
1148 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1149 dn->data_blkaddr = ei.blk + index - ei.fofs;
1153 return f2fs_reserve_block(dn, index);
1156 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1157 int op_flags, bool for_write)
1159 struct address_space *mapping = inode->i_mapping;
1160 struct dnode_of_data dn;
1162 struct extent_info ei = {0, };
1165 page = f2fs_grab_cache_page(mapping, index, for_write);
1167 return ERR_PTR(-ENOMEM);
1169 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1170 dn.data_blkaddr = ei.blk + index - ei.fofs;
1171 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1172 DATA_GENERIC_ENHANCE_READ)) {
1173 err = -EFSCORRUPTED;
1179 set_new_dnode(&dn, inode, NULL, NULL, 0);
1180 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1183 f2fs_put_dnode(&dn);
1185 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1189 if (dn.data_blkaddr != NEW_ADDR &&
1190 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1192 DATA_GENERIC_ENHANCE)) {
1193 err = -EFSCORRUPTED;
1197 if (PageUptodate(page)) {
1203 * A new dentry page is allocated but not able to be written, since its
1204 * new inode page couldn't be allocated due to -ENOSPC.
1205 * In such the case, its blkaddr can be remained as NEW_ADDR.
1206 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1207 * f2fs_init_inode_metadata.
1209 if (dn.data_blkaddr == NEW_ADDR) {
1210 zero_user_segment(page, 0, PAGE_SIZE);
1211 if (!PageUptodate(page))
1212 SetPageUptodate(page);
1217 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1218 op_flags, for_write);
1224 f2fs_put_page(page, 1);
1225 return ERR_PTR(err);
1228 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1230 struct address_space *mapping = inode->i_mapping;
1233 page = find_get_page(mapping, index);
1234 if (page && PageUptodate(page))
1236 f2fs_put_page(page, 0);
1238 page = f2fs_get_read_data_page(inode, index, 0, false);
1242 if (PageUptodate(page))
1245 wait_on_page_locked(page);
1246 if (unlikely(!PageUptodate(page))) {
1247 f2fs_put_page(page, 0);
1248 return ERR_PTR(-EIO);
1254 * If it tries to access a hole, return an error.
1255 * Because, the callers, functions in dir.c and GC, should be able to know
1256 * whether this page exists or not.
1258 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1261 struct address_space *mapping = inode->i_mapping;
1264 page = f2fs_get_read_data_page(inode, index, 0, for_write);
1268 /* wait for read completion */
1270 if (unlikely(page->mapping != mapping)) {
1271 f2fs_put_page(page, 1);
1274 if (unlikely(!PageUptodate(page))) {
1275 f2fs_put_page(page, 1);
1276 return ERR_PTR(-EIO);
1282 * Caller ensures that this data page is never allocated.
1283 * A new zero-filled data page is allocated in the page cache.
1285 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1287 * Note that, ipage is set only by make_empty_dir, and if any error occur,
1288 * ipage should be released by this function.
1290 struct page *f2fs_get_new_data_page(struct inode *inode,
1291 struct page *ipage, pgoff_t index, bool new_i_size)
1293 struct address_space *mapping = inode->i_mapping;
1295 struct dnode_of_data dn;
1298 page = f2fs_grab_cache_page(mapping, index, true);
1301 * before exiting, we should make sure ipage will be released
1302 * if any error occur.
1304 f2fs_put_page(ipage, 1);
1305 return ERR_PTR(-ENOMEM);
1308 set_new_dnode(&dn, inode, ipage, NULL, 0);
1309 err = f2fs_reserve_block(&dn, index);
1311 f2fs_put_page(page, 1);
1312 return ERR_PTR(err);
1315 f2fs_put_dnode(&dn);
1317 if (PageUptodate(page))
1320 if (dn.data_blkaddr == NEW_ADDR) {
1321 zero_user_segment(page, 0, PAGE_SIZE);
1322 if (!PageUptodate(page))
1323 SetPageUptodate(page);
1325 f2fs_put_page(page, 1);
1327 /* if ipage exists, blkaddr should be NEW_ADDR */
1328 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1329 page = f2fs_get_lock_data_page(inode, index, true);
1334 if (new_i_size && i_size_read(inode) <
1335 ((loff_t)(index + 1) << PAGE_SHIFT))
1336 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1340 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1342 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1343 struct f2fs_summary sum;
1344 struct node_info ni;
1345 block_t old_blkaddr;
1349 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1352 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1356 dn->data_blkaddr = f2fs_data_blkaddr(dn);
1357 if (dn->data_blkaddr != NULL_ADDR)
1360 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1364 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1365 old_blkaddr = dn->data_blkaddr;
1366 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1367 &sum, seg_type, NULL);
1368 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1369 invalidate_mapping_pages(META_MAPPING(sbi),
1370 old_blkaddr, old_blkaddr);
1371 f2fs_invalidate_compress_page(sbi, old_blkaddr);
1373 f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1377 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1379 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1381 f2fs_down_read(&sbi->node_change);
1383 f2fs_up_read(&sbi->node_change);
1388 f2fs_unlock_op(sbi);
1393 * f2fs_map_blocks() tries to find or build mapping relationship which
1394 * maps continuous logical blocks to physical blocks, and return such
1395 * info via f2fs_map_blocks structure.
1397 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1398 int create, int flag)
1400 unsigned int maxblocks = map->m_len;
1401 struct dnode_of_data dn;
1402 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1403 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1404 pgoff_t pgofs, end_offset, end;
1405 int err = 0, ofs = 1;
1406 unsigned int ofs_in_node, last_ofs_in_node;
1408 struct extent_info ei = {0, };
1410 unsigned int start_pgofs;
1416 map->m_bdev = inode->i_sb->s_bdev;
1417 map->m_multidev_dio =
1418 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1423 /* it only supports block size == page size */
1424 pgofs = (pgoff_t)map->m_lblk;
1425 end = pgofs + maxblocks;
1427 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1428 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1432 map->m_pblk = ei.blk + pgofs - ei.fofs;
1433 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1434 map->m_flags = F2FS_MAP_MAPPED;
1435 if (map->m_next_extent)
1436 *map->m_next_extent = pgofs + map->m_len;
1438 /* for hardware encryption, but to avoid potential issue in future */
1439 if (flag == F2FS_GET_BLOCK_DIO)
1440 f2fs_wait_on_block_writeback_range(inode,
1441 map->m_pblk, map->m_len);
1443 if (map->m_multidev_dio) {
1444 block_t blk_addr = map->m_pblk;
1446 bidx = f2fs_target_device_index(sbi, map->m_pblk);
1448 map->m_bdev = FDEV(bidx).bdev;
1449 map->m_pblk -= FDEV(bidx).start_blk;
1450 map->m_len = min(map->m_len,
1451 FDEV(bidx).end_blk + 1 - map->m_pblk);
1453 if (map->m_may_create)
1454 f2fs_update_device_state(sbi, inode->i_ino,
1455 blk_addr, map->m_len);
1461 if (map->m_may_create)
1462 f2fs_do_map_lock(sbi, flag, true);
1464 /* When reading holes, we need its node page */
1465 set_new_dnode(&dn, inode, NULL, NULL, 0);
1466 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1468 if (flag == F2FS_GET_BLOCK_BMAP)
1471 if (err == -ENOENT) {
1473 * There is one exceptional case that read_node_page()
1474 * may return -ENOENT due to filesystem has been
1475 * shutdown or cp_error, so force to convert error
1476 * number to EIO for such case.
1478 if (map->m_may_create &&
1479 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1480 f2fs_cp_error(sbi))) {
1486 if (map->m_next_pgofs)
1487 *map->m_next_pgofs =
1488 f2fs_get_next_page_offset(&dn, pgofs);
1489 if (map->m_next_extent)
1490 *map->m_next_extent =
1491 f2fs_get_next_page_offset(&dn, pgofs);
1496 start_pgofs = pgofs;
1498 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1499 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1502 blkaddr = f2fs_data_blkaddr(&dn);
1504 if (__is_valid_data_blkaddr(blkaddr) &&
1505 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1506 err = -EFSCORRUPTED;
1510 if (__is_valid_data_blkaddr(blkaddr)) {
1511 /* use out-place-update for driect IO under LFS mode */
1512 if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1513 map->m_may_create) {
1514 err = __allocate_data_block(&dn, map->m_seg_type);
1517 blkaddr = dn.data_blkaddr;
1518 set_inode_flag(inode, FI_APPEND_WRITE);
1522 if (unlikely(f2fs_cp_error(sbi))) {
1526 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1527 if (blkaddr == NULL_ADDR) {
1529 last_ofs_in_node = dn.ofs_in_node;
1532 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1533 flag != F2FS_GET_BLOCK_DIO);
1534 err = __allocate_data_block(&dn,
1537 if (flag == F2FS_GET_BLOCK_PRE_DIO)
1538 file_need_truncate(inode);
1539 set_inode_flag(inode, FI_APPEND_WRITE);
1544 map->m_flags |= F2FS_MAP_NEW;
1545 blkaddr = dn.data_blkaddr;
1547 if (f2fs_compressed_file(inode) &&
1548 f2fs_sanity_check_cluster(&dn) &&
1549 (flag != F2FS_GET_BLOCK_FIEMAP ||
1550 IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1551 err = -EFSCORRUPTED;
1554 if (flag == F2FS_GET_BLOCK_BMAP) {
1558 if (flag == F2FS_GET_BLOCK_PRECACHE)
1560 if (flag == F2FS_GET_BLOCK_FIEMAP &&
1561 blkaddr == NULL_ADDR) {
1562 if (map->m_next_pgofs)
1563 *map->m_next_pgofs = pgofs + 1;
1566 if (flag != F2FS_GET_BLOCK_FIEMAP) {
1567 /* for defragment case */
1568 if (map->m_next_pgofs)
1569 *map->m_next_pgofs = pgofs + 1;
1575 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1578 if (map->m_multidev_dio)
1579 bidx = f2fs_target_device_index(sbi, blkaddr);
1581 if (map->m_len == 0) {
1582 /* preallocated unwritten block should be mapped for fiemap. */
1583 if (blkaddr == NEW_ADDR)
1584 map->m_flags |= F2FS_MAP_UNWRITTEN;
1585 map->m_flags |= F2FS_MAP_MAPPED;
1587 map->m_pblk = blkaddr;
1590 if (map->m_multidev_dio)
1591 map->m_bdev = FDEV(bidx).bdev;
1592 } else if ((map->m_pblk != NEW_ADDR &&
1593 blkaddr == (map->m_pblk + ofs)) ||
1594 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1595 flag == F2FS_GET_BLOCK_PRE_DIO) {
1596 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1608 /* preallocate blocks in batch for one dnode page */
1609 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1610 (pgofs == end || dn.ofs_in_node == end_offset)) {
1612 dn.ofs_in_node = ofs_in_node;
1613 err = f2fs_reserve_new_blocks(&dn, prealloc);
1617 map->m_len += dn.ofs_in_node - ofs_in_node;
1618 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1622 dn.ofs_in_node = end_offset;
1627 else if (dn.ofs_in_node < end_offset)
1630 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1631 if (map->m_flags & F2FS_MAP_MAPPED) {
1632 unsigned int ofs = start_pgofs - map->m_lblk;
1634 f2fs_update_extent_cache_range(&dn,
1635 start_pgofs, map->m_pblk + ofs,
1640 f2fs_put_dnode(&dn);
1642 if (map->m_may_create) {
1643 f2fs_do_map_lock(sbi, flag, false);
1644 f2fs_balance_fs(sbi, dn.node_changed);
1650 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1652 * for hardware encryption, but to avoid potential issue
1655 f2fs_wait_on_block_writeback_range(inode,
1656 map->m_pblk, map->m_len);
1657 invalidate_mapping_pages(META_MAPPING(sbi),
1658 map->m_pblk, map->m_pblk);
1660 if (map->m_multidev_dio) {
1661 block_t blk_addr = map->m_pblk;
1663 bidx = f2fs_target_device_index(sbi, map->m_pblk);
1665 map->m_bdev = FDEV(bidx).bdev;
1666 map->m_pblk -= FDEV(bidx).start_blk;
1668 if (map->m_may_create)
1669 f2fs_update_device_state(sbi, inode->i_ino,
1670 blk_addr, map->m_len);
1672 f2fs_bug_on(sbi, blk_addr + map->m_len >
1673 FDEV(bidx).end_blk + 1);
1677 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1678 if (map->m_flags & F2FS_MAP_MAPPED) {
1679 unsigned int ofs = start_pgofs - map->m_lblk;
1681 f2fs_update_extent_cache_range(&dn,
1682 start_pgofs, map->m_pblk + ofs,
1685 if (map->m_next_extent)
1686 *map->m_next_extent = pgofs + 1;
1688 f2fs_put_dnode(&dn);
1690 if (map->m_may_create) {
1691 f2fs_do_map_lock(sbi, flag, false);
1692 f2fs_balance_fs(sbi, dn.node_changed);
1695 trace_f2fs_map_blocks(inode, map, create, flag, err);
1699 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1701 struct f2fs_map_blocks map;
1705 if (pos + len > i_size_read(inode))
1708 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1709 map.m_next_pgofs = NULL;
1710 map.m_next_extent = NULL;
1711 map.m_seg_type = NO_CHECK_TYPE;
1712 map.m_may_create = false;
1713 last_lblk = F2FS_BLK_ALIGN(pos + len);
1715 while (map.m_lblk < last_lblk) {
1716 map.m_len = last_lblk - map.m_lblk;
1717 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1718 if (err || map.m_len == 0)
1720 map.m_lblk += map.m_len;
1725 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1727 return (bytes >> inode->i_blkbits);
1730 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1732 return (blks << inode->i_blkbits);
1735 static int f2fs_xattr_fiemap(struct inode *inode,
1736 struct fiemap_extent_info *fieinfo)
1738 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1740 struct node_info ni;
1741 __u64 phys = 0, len;
1743 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1746 if (f2fs_has_inline_xattr(inode)) {
1749 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1750 inode->i_ino, false);
1754 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1756 f2fs_put_page(page, 1);
1760 phys = blks_to_bytes(inode, ni.blk_addr);
1761 offset = offsetof(struct f2fs_inode, i_addr) +
1762 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1763 get_inline_xattr_addrs(inode));
1766 len = inline_xattr_size(inode);
1768 f2fs_put_page(page, 1);
1770 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1773 flags |= FIEMAP_EXTENT_LAST;
1775 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1776 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1777 if (err || err == 1)
1782 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1786 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1788 f2fs_put_page(page, 1);
1792 phys = blks_to_bytes(inode, ni.blk_addr);
1793 len = inode->i_sb->s_blocksize;
1795 f2fs_put_page(page, 1);
1797 flags = FIEMAP_EXTENT_LAST;
1801 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1802 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1805 return (err < 0 ? err : 0);
1808 static loff_t max_inode_blocks(struct inode *inode)
1810 loff_t result = ADDRS_PER_INODE(inode);
1811 loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1813 /* two direct node blocks */
1814 result += (leaf_count * 2);
1816 /* two indirect node blocks */
1817 leaf_count *= NIDS_PER_BLOCK;
1818 result += (leaf_count * 2);
1820 /* one double indirect node block */
1821 leaf_count *= NIDS_PER_BLOCK;
1822 result += leaf_count;
1827 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1830 struct f2fs_map_blocks map;
1831 sector_t start_blk, last_blk;
1833 u64 logical = 0, phys = 0, size = 0;
1836 bool compr_cluster = false, compr_appended;
1837 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1838 unsigned int count_in_cluster = 0;
1841 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1842 ret = f2fs_precache_extents(inode);
1847 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1853 maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1854 if (start > maxbytes) {
1859 if (len > maxbytes || (maxbytes - len) < start)
1860 len = maxbytes - start;
1862 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1863 ret = f2fs_xattr_fiemap(inode, fieinfo);
1867 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1868 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1873 if (bytes_to_blks(inode, len) == 0)
1874 len = blks_to_bytes(inode, 1);
1876 start_blk = bytes_to_blks(inode, start);
1877 last_blk = bytes_to_blks(inode, start + len - 1);
1880 memset(&map, 0, sizeof(map));
1881 map.m_lblk = start_blk;
1882 map.m_len = bytes_to_blks(inode, len);
1883 map.m_next_pgofs = &next_pgofs;
1884 map.m_seg_type = NO_CHECK_TYPE;
1886 if (compr_cluster) {
1888 map.m_len = cluster_size - count_in_cluster;
1891 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1896 if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1897 start_blk = next_pgofs;
1899 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1900 max_inode_blocks(inode)))
1903 flags |= FIEMAP_EXTENT_LAST;
1906 compr_appended = false;
1907 /* In a case of compressed cluster, append this to the last extent */
1908 if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) ||
1909 !(map.m_flags & F2FS_MAP_FLAGS))) {
1910 compr_appended = true;
1915 flags |= FIEMAP_EXTENT_MERGED;
1916 if (IS_ENCRYPTED(inode))
1917 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1919 ret = fiemap_fill_next_extent(fieinfo, logical,
1921 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1927 if (start_blk > last_blk)
1931 if (map.m_pblk == COMPRESS_ADDR) {
1932 compr_cluster = true;
1933 count_in_cluster = 1;
1934 } else if (compr_appended) {
1935 unsigned int appended_blks = cluster_size -
1936 count_in_cluster + 1;
1937 size += blks_to_bytes(inode, appended_blks);
1938 start_blk += appended_blks;
1939 compr_cluster = false;
1941 logical = blks_to_bytes(inode, start_blk);
1942 phys = __is_valid_data_blkaddr(map.m_pblk) ?
1943 blks_to_bytes(inode, map.m_pblk) : 0;
1944 size = blks_to_bytes(inode, map.m_len);
1947 if (compr_cluster) {
1948 flags = FIEMAP_EXTENT_ENCODED;
1949 count_in_cluster += map.m_len;
1950 if (count_in_cluster == cluster_size) {
1951 compr_cluster = false;
1952 size += blks_to_bytes(inode, 1);
1954 } else if (map.m_flags & F2FS_MAP_UNWRITTEN) {
1955 flags = FIEMAP_EXTENT_UNWRITTEN;
1958 start_blk += bytes_to_blks(inode, size);
1963 if (fatal_signal_pending(current))
1971 inode_unlock(inode);
1975 static inline loff_t f2fs_readpage_limit(struct inode *inode)
1977 if (IS_ENABLED(CONFIG_FS_VERITY) &&
1978 (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
1979 return inode->i_sb->s_maxbytes;
1981 return i_size_read(inode);
1984 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1986 struct f2fs_map_blocks *map,
1987 struct bio **bio_ret,
1988 sector_t *last_block_in_bio,
1991 struct bio *bio = *bio_ret;
1992 const unsigned blocksize = blks_to_bytes(inode, 1);
1993 sector_t block_in_file;
1994 sector_t last_block;
1995 sector_t last_block_in_file;
1999 block_in_file = (sector_t)page_index(page);
2000 last_block = block_in_file + nr_pages;
2001 last_block_in_file = bytes_to_blks(inode,
2002 f2fs_readpage_limit(inode) + blocksize - 1);
2003 if (last_block > last_block_in_file)
2004 last_block = last_block_in_file;
2006 /* just zeroing out page which is beyond EOF */
2007 if (block_in_file >= last_block)
2010 * Map blocks using the previous result first.
2012 if ((map->m_flags & F2FS_MAP_MAPPED) &&
2013 block_in_file > map->m_lblk &&
2014 block_in_file < (map->m_lblk + map->m_len))
2018 * Then do more f2fs_map_blocks() calls until we are
2019 * done with this page.
2021 map->m_lblk = block_in_file;
2022 map->m_len = last_block - block_in_file;
2024 ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2028 if ((map->m_flags & F2FS_MAP_MAPPED)) {
2029 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2030 SetPageMappedToDisk(page);
2032 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2033 DATA_GENERIC_ENHANCE_READ)) {
2034 ret = -EFSCORRUPTED;
2039 zero_user_segment(page, 0, PAGE_SIZE);
2040 if (f2fs_need_verity(inode, page->index) &&
2041 !fsverity_verify_page(page)) {
2045 if (!PageUptodate(page))
2046 SetPageUptodate(page);
2052 * This page will go to BIO. Do we need to send this
2055 if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2056 *last_block_in_bio, block_nr) ||
2057 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2059 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2063 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2064 is_readahead ? REQ_RAHEAD : 0, page->index,
2074 * If the page is under writeback, we need to wait for
2075 * its completion to see the correct decrypted data.
2077 f2fs_wait_on_block_writeback(inode, block_nr);
2079 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2080 goto submit_and_realloc;
2082 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2083 f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2084 ClearPageError(page);
2085 *last_block_in_bio = block_nr;
2092 #ifdef CONFIG_F2FS_FS_COMPRESSION
2093 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2094 unsigned nr_pages, sector_t *last_block_in_bio,
2095 bool is_readahead, bool for_write)
2097 struct dnode_of_data dn;
2098 struct inode *inode = cc->inode;
2099 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2100 struct bio *bio = *bio_ret;
2101 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2102 sector_t last_block_in_file;
2103 const unsigned blocksize = blks_to_bytes(inode, 1);
2104 struct decompress_io_ctx *dic = NULL;
2105 struct extent_info ei = {0, };
2106 bool from_dnode = true;
2110 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2112 last_block_in_file = bytes_to_blks(inode,
2113 f2fs_readpage_limit(inode) + blocksize - 1);
2115 /* get rid of pages beyond EOF */
2116 for (i = 0; i < cc->cluster_size; i++) {
2117 struct page *page = cc->rpages[i];
2121 if ((sector_t)page->index >= last_block_in_file) {
2122 zero_user_segment(page, 0, PAGE_SIZE);
2123 if (!PageUptodate(page))
2124 SetPageUptodate(page);
2125 } else if (!PageUptodate(page)) {
2131 cc->rpages[i] = NULL;
2135 /* we are done since all pages are beyond EOF */
2136 if (f2fs_cluster_is_empty(cc))
2139 if (f2fs_lookup_extent_cache(inode, start_idx, &ei))
2143 goto skip_reading_dnode;
2145 set_new_dnode(&dn, inode, NULL, NULL, 0);
2146 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2150 f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2153 for (i = 1; i < cc->cluster_size; i++) {
2156 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2157 dn.ofs_in_node + i) :
2160 if (!__is_valid_data_blkaddr(blkaddr))
2163 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2169 if (!from_dnode && i >= ei.c_len)
2173 /* nothing to decompress */
2174 if (cc->nr_cpages == 0) {
2179 dic = f2fs_alloc_dic(cc);
2185 for (i = 0; i < cc->nr_cpages; i++) {
2186 struct page *page = dic->cpages[i];
2188 struct bio_post_read_ctx *ctx;
2190 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2191 dn.ofs_in_node + i + 1) :
2194 f2fs_wait_on_block_writeback(inode, blkaddr);
2196 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2197 if (atomic_dec_and_test(&dic->remaining_pages))
2198 f2fs_decompress_cluster(dic);
2202 if (bio && (!page_is_mergeable(sbi, bio,
2203 *last_block_in_bio, blkaddr) ||
2204 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2206 __submit_bio(sbi, bio, DATA);
2211 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2212 is_readahead ? REQ_RAHEAD : 0,
2213 page->index, for_write);
2216 f2fs_decompress_end_io(dic, ret);
2217 f2fs_put_dnode(&dn);
2223 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2224 goto submit_and_realloc;
2226 ctx = get_post_read_ctx(bio);
2227 ctx->enabled_steps |= STEP_DECOMPRESS;
2228 refcount_inc(&dic->refcnt);
2230 inc_page_count(sbi, F2FS_RD_DATA);
2231 f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2232 f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2233 ClearPageError(page);
2234 *last_block_in_bio = blkaddr;
2238 f2fs_put_dnode(&dn);
2245 f2fs_put_dnode(&dn);
2247 for (i = 0; i < cc->cluster_size; i++) {
2248 if (cc->rpages[i]) {
2249 ClearPageUptodate(cc->rpages[i]);
2250 ClearPageError(cc->rpages[i]);
2251 unlock_page(cc->rpages[i]);
2260 * This function was originally taken from fs/mpage.c, and customized for f2fs.
2261 * Major change was from block_size == page_size in f2fs by default.
2263 static int f2fs_mpage_readpages(struct inode *inode,
2264 struct readahead_control *rac, struct page *page)
2266 struct bio *bio = NULL;
2267 sector_t last_block_in_bio = 0;
2268 struct f2fs_map_blocks map;
2269 #ifdef CONFIG_F2FS_FS_COMPRESSION
2270 struct compress_ctx cc = {
2272 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2273 .cluster_size = F2FS_I(inode)->i_cluster_size,
2274 .cluster_idx = NULL_CLUSTER,
2280 pgoff_t nc_cluster_idx = NULL_CLUSTER;
2282 unsigned nr_pages = rac ? readahead_count(rac) : 1;
2283 unsigned max_nr_pages = nr_pages;
2290 map.m_next_pgofs = NULL;
2291 map.m_next_extent = NULL;
2292 map.m_seg_type = NO_CHECK_TYPE;
2293 map.m_may_create = false;
2295 for (; nr_pages; nr_pages--) {
2297 page = readahead_page(rac);
2298 prefetchw(&page->flags);
2301 #ifdef CONFIG_F2FS_FS_COMPRESSION
2302 if (f2fs_compressed_file(inode)) {
2303 /* there are remained comressed pages, submit them */
2304 if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2305 ret = f2fs_read_multi_pages(&cc, &bio,
2308 rac != NULL, false);
2309 f2fs_destroy_compress_ctx(&cc, false);
2311 goto set_error_page;
2313 if (cc.cluster_idx == NULL_CLUSTER) {
2314 if (nc_cluster_idx ==
2315 page->index >> cc.log_cluster_size) {
2316 goto read_single_page;
2319 ret = f2fs_is_compressed_cluster(inode, page->index);
2321 goto set_error_page;
2324 page->index >> cc.log_cluster_size;
2325 goto read_single_page;
2328 nc_cluster_idx = NULL_CLUSTER;
2330 ret = f2fs_init_compress_ctx(&cc);
2332 goto set_error_page;
2334 f2fs_compress_ctx_add_page(&cc, page);
2341 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2342 &bio, &last_block_in_bio, rac);
2344 #ifdef CONFIG_F2FS_FS_COMPRESSION
2348 zero_user_segment(page, 0, PAGE_SIZE);
2351 #ifdef CONFIG_F2FS_FS_COMPRESSION
2357 #ifdef CONFIG_F2FS_FS_COMPRESSION
2358 if (f2fs_compressed_file(inode)) {
2360 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2361 ret = f2fs_read_multi_pages(&cc, &bio,
2364 rac != NULL, false);
2365 f2fs_destroy_compress_ctx(&cc, false);
2371 __submit_bio(F2FS_I_SB(inode), bio, DATA);
2375 static int f2fs_read_data_page(struct file *file, struct page *page)
2377 struct inode *inode = page_file_mapping(page)->host;
2380 trace_f2fs_readpage(page, DATA);
2382 if (!f2fs_is_compress_backend_ready(inode)) {
2387 /* If the file has inline data, try to read it directly */
2388 if (f2fs_has_inline_data(inode))
2389 ret = f2fs_read_inline_data(inode, page);
2391 ret = f2fs_mpage_readpages(inode, NULL, page);
2395 static void f2fs_readahead(struct readahead_control *rac)
2397 struct inode *inode = rac->mapping->host;
2399 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2401 if (!f2fs_is_compress_backend_ready(inode))
2404 /* If the file has inline data, skip readahead */
2405 if (f2fs_has_inline_data(inode))
2408 f2fs_mpage_readpages(inode, rac, NULL);
2411 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2413 struct inode *inode = fio->page->mapping->host;
2414 struct page *mpage, *page;
2415 gfp_t gfp_flags = GFP_NOFS;
2417 if (!f2fs_encrypted_file(inode))
2420 page = fio->compressed_page ? fio->compressed_page : fio->page;
2422 /* wait for GCed page writeback via META_MAPPING */
2423 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2425 if (fscrypt_inode_uses_inline_crypto(inode))
2429 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2430 PAGE_SIZE, 0, gfp_flags);
2431 if (IS_ERR(fio->encrypted_page)) {
2432 /* flush pending IOs and wait for a while in the ENOMEM case */
2433 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2434 f2fs_flush_merged_writes(fio->sbi);
2435 memalloc_retry_wait(GFP_NOFS);
2436 gfp_flags |= __GFP_NOFAIL;
2439 return PTR_ERR(fio->encrypted_page);
2442 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2444 if (PageUptodate(mpage))
2445 memcpy(page_address(mpage),
2446 page_address(fio->encrypted_page), PAGE_SIZE);
2447 f2fs_put_page(mpage, 1);
2452 static inline bool check_inplace_update_policy(struct inode *inode,
2453 struct f2fs_io_info *fio)
2455 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2456 unsigned int policy = SM_I(sbi)->ipu_policy;
2458 if (policy & (0x1 << F2FS_IPU_HONOR_OPU_WRITE) &&
2459 is_inode_flag_set(inode, FI_OPU_WRITE))
2461 if (policy & (0x1 << F2FS_IPU_FORCE))
2463 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2465 if (policy & (0x1 << F2FS_IPU_UTIL) &&
2466 utilization(sbi) > SM_I(sbi)->min_ipu_util)
2468 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2469 utilization(sbi) > SM_I(sbi)->min_ipu_util)
2473 * IPU for rewrite async pages
2475 if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2476 fio && fio->op == REQ_OP_WRITE &&
2477 !(fio->op_flags & REQ_SYNC) &&
2478 !IS_ENCRYPTED(inode))
2481 /* this is only set during fdatasync */
2482 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2483 is_inode_flag_set(inode, FI_NEED_IPU))
2486 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2487 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2493 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2495 /* swap file is migrating in aligned write mode */
2496 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2499 if (f2fs_is_pinned_file(inode))
2502 /* if this is cold file, we should overwrite to avoid fragmentation */
2503 if (file_is_cold(inode))
2506 return check_inplace_update_policy(inode, fio);
2509 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2511 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2513 /* The below cases were checked when setting it. */
2514 if (f2fs_is_pinned_file(inode))
2516 if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2518 if (f2fs_lfs_mode(sbi))
2520 if (S_ISDIR(inode->i_mode))
2522 if (IS_NOQUOTA(inode))
2524 if (f2fs_is_atomic_file(inode))
2527 /* swap file is migrating in aligned write mode */
2528 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2531 if (is_inode_flag_set(inode, FI_OPU_WRITE))
2535 if (page_private_gcing(fio->page))
2537 if (page_private_dummy(fio->page))
2539 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2540 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2546 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2548 struct inode *inode = fio->page->mapping->host;
2550 if (f2fs_should_update_outplace(inode, fio))
2553 return f2fs_should_update_inplace(inode, fio);
2556 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2558 struct page *page = fio->page;
2559 struct inode *inode = page->mapping->host;
2560 struct dnode_of_data dn;
2561 struct extent_info ei = {0, };
2562 struct node_info ni;
2563 bool ipu_force = false;
2566 set_new_dnode(&dn, inode, NULL, NULL, 0);
2567 if (need_inplace_update(fio) &&
2568 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2569 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2571 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2572 DATA_GENERIC_ENHANCE))
2573 return -EFSCORRUPTED;
2576 fio->need_lock = LOCK_DONE;
2580 /* Deadlock due to between page->lock and f2fs_lock_op */
2581 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2584 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2588 fio->old_blkaddr = dn.data_blkaddr;
2590 /* This page is already truncated */
2591 if (fio->old_blkaddr == NULL_ADDR) {
2592 ClearPageUptodate(page);
2593 clear_page_private_gcing(page);
2597 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2598 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2599 DATA_GENERIC_ENHANCE)) {
2600 err = -EFSCORRUPTED;
2604 * If current allocation needs SSR,
2605 * it had better in-place writes for updated data.
2608 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2609 need_inplace_update(fio))) {
2610 err = f2fs_encrypt_one_page(fio);
2614 set_page_writeback(page);
2615 ClearPageError(page);
2616 f2fs_put_dnode(&dn);
2617 if (fio->need_lock == LOCK_REQ)
2618 f2fs_unlock_op(fio->sbi);
2619 err = f2fs_inplace_write_data(fio);
2621 if (fscrypt_inode_uses_fs_layer_crypto(inode))
2622 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2623 if (PageWriteback(page))
2624 end_page_writeback(page);
2626 set_inode_flag(inode, FI_UPDATE_WRITE);
2628 trace_f2fs_do_write_data_page(fio->page, IPU);
2632 if (fio->need_lock == LOCK_RETRY) {
2633 if (!f2fs_trylock_op(fio->sbi)) {
2637 fio->need_lock = LOCK_REQ;
2640 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2644 fio->version = ni.version;
2646 err = f2fs_encrypt_one_page(fio);
2650 set_page_writeback(page);
2651 ClearPageError(page);
2653 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2654 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2656 /* LFS mode write path */
2657 f2fs_outplace_write_data(&dn, fio);
2658 trace_f2fs_do_write_data_page(page, OPU);
2659 set_inode_flag(inode, FI_APPEND_WRITE);
2660 if (page->index == 0)
2661 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2663 f2fs_put_dnode(&dn);
2665 if (fio->need_lock == LOCK_REQ)
2666 f2fs_unlock_op(fio->sbi);
2670 int f2fs_write_single_data_page(struct page *page, int *submitted,
2672 sector_t *last_block,
2673 struct writeback_control *wbc,
2674 enum iostat_type io_type,
2678 struct inode *inode = page->mapping->host;
2679 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2680 loff_t i_size = i_size_read(inode);
2681 const pgoff_t end_index = ((unsigned long long)i_size)
2683 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2684 unsigned offset = 0;
2685 bool need_balance_fs = false;
2687 struct f2fs_io_info fio = {
2689 .ino = inode->i_ino,
2692 .op_flags = wbc_to_write_flags(wbc),
2693 .old_blkaddr = NULL_ADDR,
2695 .encrypted_page = NULL,
2697 .compr_blocks = compr_blocks,
2698 .need_lock = LOCK_RETRY,
2702 .last_block = last_block,
2705 trace_f2fs_writepage(page, DATA);
2707 /* we should bypass data pages to proceed the kworkder jobs */
2708 if (unlikely(f2fs_cp_error(sbi))) {
2709 mapping_set_error(page->mapping, -EIO);
2711 * don't drop any dirty dentry pages for keeping lastest
2712 * directory structure.
2714 if (S_ISDIR(inode->i_mode))
2719 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2722 if (page->index < end_index ||
2723 f2fs_verity_in_progress(inode) ||
2728 * If the offset is out-of-range of file size,
2729 * this page does not have to be written to disk.
2731 offset = i_size & (PAGE_SIZE - 1);
2732 if ((page->index >= end_index + 1) || !offset)
2735 zero_user_segment(page, offset, PAGE_SIZE);
2737 if (f2fs_is_drop_cache(inode))
2739 /* we should not write 0'th page having journal header */
2740 if (f2fs_is_volatile_file(inode) && (!page->index ||
2741 (!wbc->for_reclaim &&
2742 f2fs_available_free_memory(sbi, BASE_CHECK))))
2745 /* Dentry/quota blocks are controlled by checkpoint */
2746 if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2748 * We need to wait for node_write to avoid block allocation during
2749 * checkpoint. This can only happen to quota writes which can cause
2750 * the below discard race condition.
2752 if (IS_NOQUOTA(inode))
2753 f2fs_down_read(&sbi->node_write);
2755 fio.need_lock = LOCK_DONE;
2756 err = f2fs_do_write_data_page(&fio);
2758 if (IS_NOQUOTA(inode))
2759 f2fs_up_read(&sbi->node_write);
2764 if (!wbc->for_reclaim)
2765 need_balance_fs = true;
2766 else if (has_not_enough_free_secs(sbi, 0, 0))
2769 set_inode_flag(inode, FI_HOT_DATA);
2772 if (f2fs_has_inline_data(inode)) {
2773 err = f2fs_write_inline_data(inode, page);
2778 if (err == -EAGAIN) {
2779 err = f2fs_do_write_data_page(&fio);
2780 if (err == -EAGAIN) {
2781 fio.need_lock = LOCK_REQ;
2782 err = f2fs_do_write_data_page(&fio);
2787 file_set_keep_isize(inode);
2789 spin_lock(&F2FS_I(inode)->i_size_lock);
2790 if (F2FS_I(inode)->last_disk_size < psize)
2791 F2FS_I(inode)->last_disk_size = psize;
2792 spin_unlock(&F2FS_I(inode)->i_size_lock);
2796 if (err && err != -ENOENT)
2800 inode_dec_dirty_pages(inode);
2802 ClearPageUptodate(page);
2803 clear_page_private_gcing(page);
2806 if (wbc->for_reclaim) {
2807 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2808 clear_inode_flag(inode, FI_HOT_DATA);
2809 f2fs_remove_dirty_inode(inode);
2813 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2814 !F2FS_I(inode)->cp_task && allow_balance)
2815 f2fs_balance_fs(sbi, need_balance_fs);
2817 if (unlikely(f2fs_cp_error(sbi))) {
2818 f2fs_submit_merged_write(sbi, DATA);
2819 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2824 *submitted = fio.submitted ? 1 : 0;
2829 redirty_page_for_writepage(wbc, page);
2831 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2832 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2833 * file_write_and_wait_range() will see EIO error, which is critical
2834 * to return value of fsync() followed by atomic_write failure to user.
2836 if (!err || wbc->for_reclaim)
2837 return AOP_WRITEPAGE_ACTIVATE;
2842 static int f2fs_write_data_page(struct page *page,
2843 struct writeback_control *wbc)
2845 #ifdef CONFIG_F2FS_FS_COMPRESSION
2846 struct inode *inode = page->mapping->host;
2848 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2851 if (f2fs_compressed_file(inode)) {
2852 if (f2fs_is_compressed_cluster(inode, page->index)) {
2853 redirty_page_for_writepage(wbc, page);
2854 return AOP_WRITEPAGE_ACTIVATE;
2860 return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2861 wbc, FS_DATA_IO, 0, true);
2865 * This function was copied from write_cche_pages from mm/page-writeback.c.
2866 * The major change is making write step of cold data page separately from
2867 * warm/hot data page.
2869 static int f2fs_write_cache_pages(struct address_space *mapping,
2870 struct writeback_control *wbc,
2871 enum iostat_type io_type)
2874 int done = 0, retry = 0;
2875 struct pagevec pvec;
2876 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2877 struct bio *bio = NULL;
2878 sector_t last_block;
2879 #ifdef CONFIG_F2FS_FS_COMPRESSION
2880 struct inode *inode = mapping->host;
2881 struct compress_ctx cc = {
2883 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2884 .cluster_size = F2FS_I(inode)->i_cluster_size,
2885 .cluster_idx = NULL_CLUSTER,
2889 .valid_nr_cpages = 0,
2892 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2898 pgoff_t end; /* Inclusive */
2900 int range_whole = 0;
2906 pagevec_init(&pvec);
2908 if (get_dirty_pages(mapping->host) <=
2909 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2910 set_inode_flag(mapping->host, FI_HOT_DATA);
2912 clear_inode_flag(mapping->host, FI_HOT_DATA);
2914 if (wbc->range_cyclic) {
2915 index = mapping->writeback_index; /* prev offset */
2918 index = wbc->range_start >> PAGE_SHIFT;
2919 end = wbc->range_end >> PAGE_SHIFT;
2920 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2923 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2924 tag = PAGECACHE_TAG_TOWRITE;
2926 tag = PAGECACHE_TAG_DIRTY;
2929 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2930 tag_pages_for_writeback(mapping, index, end);
2932 while (!done && !retry && (index <= end)) {
2933 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2938 for (i = 0; i < nr_pages; i++) {
2939 struct page *page = pvec.pages[i];
2943 #ifdef CONFIG_F2FS_FS_COMPRESSION
2944 if (f2fs_compressed_file(inode)) {
2945 void *fsdata = NULL;
2949 ret = f2fs_init_compress_ctx(&cc);
2955 if (!f2fs_cluster_can_merge_page(&cc,
2957 ret = f2fs_write_multi_pages(&cc,
2958 &submitted, wbc, io_type);
2964 if (unlikely(f2fs_cp_error(sbi)))
2967 if (!f2fs_cluster_is_empty(&cc))
2970 ret2 = f2fs_prepare_compress_overwrite(
2972 page->index, &fsdata);
2978 (!f2fs_compress_write_end(inode,
2979 fsdata, page->index, 1) ||
2980 !f2fs_all_cluster_page_loaded(&cc,
2981 &pvec, i, nr_pages))) {
2987 /* give a priority to WB_SYNC threads */
2988 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2989 wbc->sync_mode == WB_SYNC_NONE) {
2993 #ifdef CONFIG_F2FS_FS_COMPRESSION
2996 done_index = page->index;
3000 if (unlikely(page->mapping != mapping)) {
3006 if (!PageDirty(page)) {
3007 /* someone wrote it for us */
3008 goto continue_unlock;
3011 if (PageWriteback(page)) {
3012 if (wbc->sync_mode != WB_SYNC_NONE)
3013 f2fs_wait_on_page_writeback(page,
3016 goto continue_unlock;
3019 if (!clear_page_dirty_for_io(page))
3020 goto continue_unlock;
3022 #ifdef CONFIG_F2FS_FS_COMPRESSION
3023 if (f2fs_compressed_file(inode)) {
3025 f2fs_compress_ctx_add_page(&cc, page);
3029 ret = f2fs_write_single_data_page(page, &submitted,
3030 &bio, &last_block, wbc, io_type,
3032 if (ret == AOP_WRITEPAGE_ACTIVATE)
3034 #ifdef CONFIG_F2FS_FS_COMPRESSION
3037 nwritten += submitted;
3038 wbc->nr_to_write -= submitted;
3040 if (unlikely(ret)) {
3042 * keep nr_to_write, since vfs uses this to
3043 * get # of written pages.
3045 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3048 } else if (ret == -EAGAIN) {
3050 if (wbc->sync_mode == WB_SYNC_ALL) {
3051 f2fs_io_schedule_timeout(
3052 DEFAULT_IO_TIMEOUT);
3057 done_index = page->index + 1;
3062 if (wbc->nr_to_write <= 0 &&
3063 wbc->sync_mode == WB_SYNC_NONE) {
3071 pagevec_release(&pvec);
3074 #ifdef CONFIG_F2FS_FS_COMPRESSION
3075 /* flush remained pages in compress cluster */
3076 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3077 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3078 nwritten += submitted;
3079 wbc->nr_to_write -= submitted;
3085 if (f2fs_compressed_file(inode))
3086 f2fs_destroy_compress_ctx(&cc, false);
3093 if (wbc->range_cyclic && !done)
3095 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3096 mapping->writeback_index = done_index;
3099 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3101 /* submit cached bio of IPU write */
3103 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3108 static inline bool __should_serialize_io(struct inode *inode,
3109 struct writeback_control *wbc)
3111 /* to avoid deadlock in path of data flush */
3112 if (F2FS_I(inode)->cp_task)
3115 if (!S_ISREG(inode->i_mode))
3117 if (IS_NOQUOTA(inode))
3120 if (f2fs_need_compress_data(inode))
3122 if (wbc->sync_mode != WB_SYNC_ALL)
3124 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3129 static int __f2fs_write_data_pages(struct address_space *mapping,
3130 struct writeback_control *wbc,
3131 enum iostat_type io_type)
3133 struct inode *inode = mapping->host;
3134 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3135 struct blk_plug plug;
3137 bool locked = false;
3139 /* deal with chardevs and other special file */
3140 if (!mapping->a_ops->writepage)
3143 /* skip writing if there is no dirty page in this inode */
3144 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3147 /* during POR, we don't need to trigger writepage at all. */
3148 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3151 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3152 wbc->sync_mode == WB_SYNC_NONE &&
3153 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3154 f2fs_available_free_memory(sbi, DIRTY_DENTS))
3157 /* skip writing in file defragment preparing stage */
3158 if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3161 trace_f2fs_writepages(mapping->host, wbc, DATA);
3163 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3164 if (wbc->sync_mode == WB_SYNC_ALL)
3165 atomic_inc(&sbi->wb_sync_req[DATA]);
3166 else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3167 /* to avoid potential deadlock */
3169 blk_finish_plug(current->plug);
3173 if (__should_serialize_io(inode, wbc)) {
3174 mutex_lock(&sbi->writepages);
3178 blk_start_plug(&plug);
3179 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3180 blk_finish_plug(&plug);
3183 mutex_unlock(&sbi->writepages);
3185 if (wbc->sync_mode == WB_SYNC_ALL)
3186 atomic_dec(&sbi->wb_sync_req[DATA]);
3188 * if some pages were truncated, we cannot guarantee its mapping->host
3189 * to detect pending bios.
3192 f2fs_remove_dirty_inode(inode);
3196 wbc->pages_skipped += get_dirty_pages(inode);
3197 trace_f2fs_writepages(mapping->host, wbc, DATA);
3201 static int f2fs_write_data_pages(struct address_space *mapping,
3202 struct writeback_control *wbc)
3204 struct inode *inode = mapping->host;
3206 return __f2fs_write_data_pages(mapping, wbc,
3207 F2FS_I(inode)->cp_task == current ?
3208 FS_CP_DATA_IO : FS_DATA_IO);
3211 void f2fs_write_failed(struct inode *inode, loff_t to)
3213 loff_t i_size = i_size_read(inode);
3215 if (IS_NOQUOTA(inode))
3218 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3219 if (to > i_size && !f2fs_verity_in_progress(inode)) {
3220 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3221 filemap_invalidate_lock(inode->i_mapping);
3223 truncate_pagecache(inode, i_size);
3224 f2fs_truncate_blocks(inode, i_size, true);
3226 filemap_invalidate_unlock(inode->i_mapping);
3227 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3231 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3232 struct page *page, loff_t pos, unsigned len,
3233 block_t *blk_addr, bool *node_changed)
3235 struct inode *inode = page->mapping->host;
3236 pgoff_t index = page->index;
3237 struct dnode_of_data dn;
3239 bool locked = false;
3240 struct extent_info ei = {0, };
3245 * If a whole page is being written and we already preallocated all the
3246 * blocks, then there is no need to get a block address now.
3248 if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3251 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3252 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3253 flag = F2FS_GET_BLOCK_DEFAULT;
3255 flag = F2FS_GET_BLOCK_PRE_AIO;
3257 if (f2fs_has_inline_data(inode) ||
3258 (pos & PAGE_MASK) >= i_size_read(inode)) {
3259 f2fs_do_map_lock(sbi, flag, true);
3264 /* check inline_data */
3265 ipage = f2fs_get_node_page(sbi, inode->i_ino);
3266 if (IS_ERR(ipage)) {
3267 err = PTR_ERR(ipage);
3271 set_new_dnode(&dn, inode, ipage, ipage, 0);
3273 if (f2fs_has_inline_data(inode)) {
3274 if (pos + len <= MAX_INLINE_DATA(inode)) {
3275 f2fs_do_read_inline_data(page, ipage);
3276 set_inode_flag(inode, FI_DATA_EXIST);
3278 set_page_private_inline(ipage);
3280 err = f2fs_convert_inline_page(&dn, page);
3283 if (dn.data_blkaddr == NULL_ADDR)
3284 err = f2fs_get_block(&dn, index);
3286 } else if (locked) {
3287 err = f2fs_get_block(&dn, index);
3289 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3290 dn.data_blkaddr = ei.blk + index - ei.fofs;
3293 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3294 if (err || dn.data_blkaddr == NULL_ADDR) {
3295 f2fs_put_dnode(&dn);
3296 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3298 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3305 /* convert_inline_page can make node_changed */
3306 *blk_addr = dn.data_blkaddr;
3307 *node_changed = dn.node_changed;
3309 f2fs_put_dnode(&dn);
3312 f2fs_do_map_lock(sbi, flag, false);
3316 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3317 loff_t pos, unsigned len, unsigned flags,
3318 struct page **pagep, void **fsdata)
3320 struct inode *inode = mapping->host;
3321 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3322 struct page *page = NULL;
3323 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3324 bool need_balance = false, drop_atomic = false;
3325 block_t blkaddr = NULL_ADDR;
3328 trace_f2fs_write_begin(inode, pos, len, flags);
3330 if (!f2fs_is_checkpoint_ready(sbi)) {
3335 if ((f2fs_is_atomic_file(inode) &&
3336 !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3337 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3344 * We should check this at this moment to avoid deadlock on inode page
3345 * and #0 page. The locking rule for inline_data conversion should be:
3346 * lock_page(page #0) -> lock_page(inode_page)
3349 err = f2fs_convert_inline_inode(inode);
3354 #ifdef CONFIG_F2FS_FS_COMPRESSION
3355 if (f2fs_compressed_file(inode)) {
3360 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3363 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3376 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3377 * wait_for_stable_page. Will wait that below with our IO control.
3379 page = f2fs_pagecache_get_page(mapping, index,
3380 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3386 /* TODO: cluster can be compressed due to race with .writepage */
3390 err = prepare_write_begin(sbi, page, pos, len,
3391 &blkaddr, &need_balance);
3395 if (need_balance && !IS_NOQUOTA(inode) &&
3396 has_not_enough_free_secs(sbi, 0, 0)) {
3398 f2fs_balance_fs(sbi, true);
3400 if (page->mapping != mapping) {
3401 /* The page got truncated from under us */
3402 f2fs_put_page(page, 1);
3407 f2fs_wait_on_page_writeback(page, DATA, false, true);
3409 if (len == PAGE_SIZE || PageUptodate(page))
3412 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3413 !f2fs_verity_in_progress(inode)) {
3414 zero_user_segment(page, len, PAGE_SIZE);
3418 if (blkaddr == NEW_ADDR) {
3419 zero_user_segment(page, 0, PAGE_SIZE);
3420 SetPageUptodate(page);
3422 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3423 DATA_GENERIC_ENHANCE_READ)) {
3424 err = -EFSCORRUPTED;
3427 err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3432 if (unlikely(page->mapping != mapping)) {
3433 f2fs_put_page(page, 1);
3436 if (unlikely(!PageUptodate(page))) {
3444 f2fs_put_page(page, 1);
3445 f2fs_write_failed(inode, pos + len);
3447 f2fs_drop_inmem_pages_all(sbi, false);
3451 static int f2fs_write_end(struct file *file,
3452 struct address_space *mapping,
3453 loff_t pos, unsigned len, unsigned copied,
3454 struct page *page, void *fsdata)
3456 struct inode *inode = page->mapping->host;
3458 trace_f2fs_write_end(inode, pos, len, copied);
3461 * This should be come from len == PAGE_SIZE, and we expect copied
3462 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3463 * let generic_perform_write() try to copy data again through copied=0.
3465 if (!PageUptodate(page)) {
3466 if (unlikely(copied != len))
3469 SetPageUptodate(page);
3472 #ifdef CONFIG_F2FS_FS_COMPRESSION
3473 /* overwrite compressed file */
3474 if (f2fs_compressed_file(inode) && fsdata) {
3475 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3476 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3478 if (pos + copied > i_size_read(inode) &&
3479 !f2fs_verity_in_progress(inode))
3480 f2fs_i_size_write(inode, pos + copied);
3488 set_page_dirty(page);
3490 if (pos + copied > i_size_read(inode) &&
3491 !f2fs_verity_in_progress(inode))
3492 f2fs_i_size_write(inode, pos + copied);
3494 f2fs_put_page(page, 1);
3495 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3499 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3501 struct inode *inode = folio->mapping->host;
3502 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3504 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3505 (offset || length != folio_size(folio)))
3508 if (folio_test_dirty(folio)) {
3509 if (inode->i_ino == F2FS_META_INO(sbi)) {
3510 dec_page_count(sbi, F2FS_DIRTY_META);
3511 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3512 dec_page_count(sbi, F2FS_DIRTY_NODES);
3514 inode_dec_dirty_pages(inode);
3515 f2fs_remove_dirty_inode(inode);
3519 clear_page_private_gcing(&folio->page);
3521 if (test_opt(sbi, COMPRESS_CACHE) &&
3522 inode->i_ino == F2FS_COMPRESS_INO(sbi))
3523 clear_page_private_data(&folio->page);
3525 if (page_private_atomic(&folio->page))
3526 return f2fs_drop_inmem_page(inode, &folio->page);
3528 folio_detach_private(folio);
3531 int f2fs_release_page(struct page *page, gfp_t wait)
3533 /* If this is dirty page, keep PagePrivate */
3534 if (PageDirty(page))
3537 /* This is atomic written page, keep Private */
3538 if (page_private_atomic(page))
3541 if (test_opt(F2FS_P_SB(page), COMPRESS_CACHE)) {
3542 struct inode *inode = page->mapping->host;
3544 if (inode->i_ino == F2FS_COMPRESS_INO(F2FS_I_SB(inode)))
3545 clear_page_private_data(page);
3548 clear_page_private_gcing(page);
3550 detach_page_private(page);
3551 set_page_private(page, 0);
3555 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3556 struct folio *folio)
3558 struct inode *inode = mapping->host;
3560 trace_f2fs_set_page_dirty(&folio->page, DATA);
3562 if (!folio_test_uptodate(folio))
3563 folio_mark_uptodate(folio);
3564 BUG_ON(folio_test_swapcache(folio));
3566 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3567 if (!page_private_atomic(&folio->page)) {
3568 f2fs_register_inmem_page(inode, &folio->page);
3572 * Previously, this page has been registered, we just
3578 if (!folio_test_dirty(folio)) {
3579 filemap_dirty_folio(mapping, folio);
3580 f2fs_update_dirty_folio(inode, folio);
3587 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3589 #ifdef CONFIG_F2FS_FS_COMPRESSION
3590 struct dnode_of_data dn;
3591 sector_t start_idx, blknr = 0;
3594 start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3596 set_new_dnode(&dn, inode, NULL, NULL, 0);
3597 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3601 if (dn.data_blkaddr != COMPRESS_ADDR) {
3602 dn.ofs_in_node += block - start_idx;
3603 blknr = f2fs_data_blkaddr(&dn);
3604 if (!__is_valid_data_blkaddr(blknr))
3608 f2fs_put_dnode(&dn);
3616 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3618 struct inode *inode = mapping->host;
3621 if (f2fs_has_inline_data(inode))
3624 /* make sure allocating whole blocks */
3625 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3626 filemap_write_and_wait(mapping);
3628 /* Block number less than F2FS MAX BLOCKS */
3629 if (unlikely(block >= max_file_blocks(inode)))
3632 if (f2fs_compressed_file(inode)) {
3633 blknr = f2fs_bmap_compress(inode, block);
3635 struct f2fs_map_blocks map;
3637 memset(&map, 0, sizeof(map));
3640 map.m_next_pgofs = NULL;
3641 map.m_seg_type = NO_CHECK_TYPE;
3643 if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3647 trace_f2fs_bmap(inode, block, blknr);
3651 #ifdef CONFIG_MIGRATION
3652 #include <linux/migrate.h>
3654 int f2fs_migrate_page(struct address_space *mapping,
3655 struct page *newpage, struct page *page, enum migrate_mode mode)
3657 int rc, extra_count;
3658 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3659 bool atomic_written = page_private_atomic(page);
3661 BUG_ON(PageWriteback(page));
3663 /* migrating an atomic written page is safe with the inmem_lock hold */
3664 if (atomic_written) {
3665 if (mode != MIGRATE_SYNC)
3667 if (!mutex_trylock(&fi->inmem_lock))
3671 /* one extra reference was held for atomic_write page */
3672 extra_count = atomic_written ? 1 : 0;
3673 rc = migrate_page_move_mapping(mapping, newpage,
3675 if (rc != MIGRATEPAGE_SUCCESS) {
3677 mutex_unlock(&fi->inmem_lock);
3681 if (atomic_written) {
3682 struct inmem_pages *cur;
3684 list_for_each_entry(cur, &fi->inmem_pages, list)
3685 if (cur->page == page) {
3686 cur->page = newpage;
3689 mutex_unlock(&fi->inmem_lock);
3694 /* guarantee to start from no stale private field */
3695 set_page_private(newpage, 0);
3696 if (PagePrivate(page)) {
3697 set_page_private(newpage, page_private(page));
3698 SetPagePrivate(newpage);
3701 set_page_private(page, 0);
3702 ClearPagePrivate(page);
3706 if (mode != MIGRATE_SYNC_NO_COPY)
3707 migrate_page_copy(newpage, page);
3709 migrate_page_states(newpage, page);
3711 return MIGRATEPAGE_SUCCESS;
3716 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3717 unsigned int blkcnt)
3719 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3720 unsigned int blkofs;
3721 unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3722 unsigned int secidx = start_blk / blk_per_sec;
3723 unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3726 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3727 filemap_invalidate_lock(inode->i_mapping);
3729 set_inode_flag(inode, FI_ALIGNED_WRITE);
3730 set_inode_flag(inode, FI_OPU_WRITE);
3732 for (; secidx < end_sec; secidx++) {
3733 f2fs_down_write(&sbi->pin_sem);
3736 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3737 f2fs_unlock_op(sbi);
3739 set_inode_flag(inode, FI_SKIP_WRITES);
3741 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3743 unsigned int blkidx = secidx * blk_per_sec + blkofs;
3745 page = f2fs_get_lock_data_page(inode, blkidx, true);
3747 f2fs_up_write(&sbi->pin_sem);
3748 ret = PTR_ERR(page);
3752 set_page_dirty(page);
3753 f2fs_put_page(page, 1);
3756 clear_inode_flag(inode, FI_SKIP_WRITES);
3758 ret = filemap_fdatawrite(inode->i_mapping);
3760 f2fs_up_write(&sbi->pin_sem);
3767 clear_inode_flag(inode, FI_SKIP_WRITES);
3768 clear_inode_flag(inode, FI_OPU_WRITE);
3769 clear_inode_flag(inode, FI_ALIGNED_WRITE);
3771 filemap_invalidate_unlock(inode->i_mapping);
3772 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3777 static int check_swap_activate(struct swap_info_struct *sis,
3778 struct file *swap_file, sector_t *span)
3780 struct address_space *mapping = swap_file->f_mapping;
3781 struct inode *inode = mapping->host;
3782 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3783 sector_t cur_lblock;
3784 sector_t last_lblock;
3786 sector_t lowest_pblock = -1;
3787 sector_t highest_pblock = 0;
3789 unsigned long nr_pblocks;
3790 unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3791 unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3792 unsigned int not_aligned = 0;
3796 * Map all the blocks into the extent list. This code doesn't try
3800 last_lblock = bytes_to_blks(inode, i_size_read(inode));
3802 while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3803 struct f2fs_map_blocks map;
3807 memset(&map, 0, sizeof(map));
3808 map.m_lblk = cur_lblock;
3809 map.m_len = last_lblock - cur_lblock;
3810 map.m_next_pgofs = NULL;
3811 map.m_next_extent = NULL;
3812 map.m_seg_type = NO_CHECK_TYPE;
3813 map.m_may_create = false;
3815 ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3820 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3821 f2fs_err(sbi, "Swapfile has holes");
3826 pblock = map.m_pblk;
3827 nr_pblocks = map.m_len;
3829 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
3830 nr_pblocks & sec_blks_mask) {
3833 nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3834 if (cur_lblock + nr_pblocks > sis->max)
3835 nr_pblocks -= blks_per_sec;
3838 /* this extent is last one */
3839 nr_pblocks = map.m_len;
3840 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
3844 ret = f2fs_migrate_blocks(inode, cur_lblock,
3851 if (cur_lblock + nr_pblocks >= sis->max)
3852 nr_pblocks = sis->max - cur_lblock;
3854 if (cur_lblock) { /* exclude the header page */
3855 if (pblock < lowest_pblock)
3856 lowest_pblock = pblock;
3857 if (pblock + nr_pblocks - 1 > highest_pblock)
3858 highest_pblock = pblock + nr_pblocks - 1;
3862 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3864 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3868 cur_lblock += nr_pblocks;
3871 *span = 1 + highest_pblock - lowest_pblock;
3872 if (cur_lblock == 0)
3873 cur_lblock = 1; /* force Empty message */
3874 sis->max = cur_lblock;
3875 sis->pages = cur_lblock - 1;
3876 sis->highest_bit = cur_lblock - 1;
3879 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
3880 not_aligned, blks_per_sec * F2FS_BLKSIZE);
3884 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3887 struct inode *inode = file_inode(file);
3890 if (!S_ISREG(inode->i_mode))
3893 if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3896 if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
3897 f2fs_err(F2FS_I_SB(inode),
3898 "Swapfile not supported in LFS mode");
3902 ret = f2fs_convert_inline_inode(inode);
3906 if (!f2fs_disable_compressed_file(inode))
3909 f2fs_precache_extents(inode);
3911 ret = check_swap_activate(sis, file, span);
3915 set_inode_flag(inode, FI_PIN_FILE);
3916 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3920 static void f2fs_swap_deactivate(struct file *file)
3922 struct inode *inode = file_inode(file);
3924 clear_inode_flag(inode, FI_PIN_FILE);
3927 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3933 static void f2fs_swap_deactivate(struct file *file)
3938 const struct address_space_operations f2fs_dblock_aops = {
3939 .readpage = f2fs_read_data_page,
3940 .readahead = f2fs_readahead,
3941 .writepage = f2fs_write_data_page,
3942 .writepages = f2fs_write_data_pages,
3943 .write_begin = f2fs_write_begin,
3944 .write_end = f2fs_write_end,
3945 .dirty_folio = f2fs_dirty_data_folio,
3946 .invalidate_folio = f2fs_invalidate_folio,
3947 .releasepage = f2fs_release_page,
3948 .direct_IO = noop_direct_IO,
3950 .swap_activate = f2fs_swap_activate,
3951 .swap_deactivate = f2fs_swap_deactivate,
3952 #ifdef CONFIG_MIGRATION
3953 .migratepage = f2fs_migrate_page,
3957 void f2fs_clear_page_cache_dirty_tag(struct page *page)
3959 struct address_space *mapping = page_mapping(page);
3960 unsigned long flags;
3962 xa_lock_irqsave(&mapping->i_pages, flags);
3963 __xa_clear_mark(&mapping->i_pages, page_index(page),
3964 PAGECACHE_TAG_DIRTY);
3965 xa_unlock_irqrestore(&mapping->i_pages, flags);
3968 int __init f2fs_init_post_read_processing(void)
3970 bio_post_read_ctx_cache =
3971 kmem_cache_create("f2fs_bio_post_read_ctx",
3972 sizeof(struct bio_post_read_ctx), 0, 0, NULL);
3973 if (!bio_post_read_ctx_cache)
3975 bio_post_read_ctx_pool =
3976 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
3977 bio_post_read_ctx_cache);
3978 if (!bio_post_read_ctx_pool)
3979 goto fail_free_cache;
3983 kmem_cache_destroy(bio_post_read_ctx_cache);
3988 void f2fs_destroy_post_read_processing(void)
3990 mempool_destroy(bio_post_read_ctx_pool);
3991 kmem_cache_destroy(bio_post_read_ctx_cache);
3994 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
3996 if (!f2fs_sb_has_encrypt(sbi) &&
3997 !f2fs_sb_has_verity(sbi) &&
3998 !f2fs_sb_has_compression(sbi))
4001 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4002 WQ_UNBOUND | WQ_HIGHPRI,
4004 if (!sbi->post_read_wq)
4009 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4011 if (sbi->post_read_wq)
4012 destroy_workqueue(sbi->post_read_wq);
4015 int __init f2fs_init_bio_entry_cache(void)
4017 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4018 sizeof(struct bio_entry));
4019 if (!bio_entry_slab)
4024 void f2fs_destroy_bio_entry_cache(void)
4026 kmem_cache_destroy(bio_entry_slab);
4029 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4030 unsigned int flags, struct iomap *iomap,
4031 struct iomap *srcmap)
4033 struct f2fs_map_blocks map = {};
4034 pgoff_t next_pgofs = 0;
4037 map.m_lblk = bytes_to_blks(inode, offset);
4038 map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4039 map.m_next_pgofs = &next_pgofs;
4040 map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4041 if (flags & IOMAP_WRITE)
4042 map.m_may_create = true;
4044 err = f2fs_map_blocks(inode, &map, flags & IOMAP_WRITE,
4045 F2FS_GET_BLOCK_DIO);
4049 iomap->offset = blks_to_bytes(inode, map.m_lblk);
4052 * When inline encryption is enabled, sometimes I/O to an encrypted file
4053 * has to be broken up to guarantee DUN contiguity. Handle this by
4054 * limiting the length of the mapping returned.
4056 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4058 if (map.m_flags & (F2FS_MAP_MAPPED | F2FS_MAP_UNWRITTEN)) {
4059 iomap->length = blks_to_bytes(inode, map.m_len);
4060 if (map.m_flags & F2FS_MAP_MAPPED) {
4061 iomap->type = IOMAP_MAPPED;
4062 iomap->flags |= IOMAP_F_MERGED;
4064 iomap->type = IOMAP_UNWRITTEN;
4066 if (WARN_ON_ONCE(!__is_valid_data_blkaddr(map.m_pblk)))
4069 iomap->bdev = map.m_bdev;
4070 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4072 iomap->length = blks_to_bytes(inode, next_pgofs) -
4074 iomap->type = IOMAP_HOLE;
4075 iomap->addr = IOMAP_NULL_ADDR;
4078 if (map.m_flags & F2FS_MAP_NEW)
4079 iomap->flags |= IOMAP_F_NEW;
4080 if ((inode->i_state & I_DIRTY_DATASYNC) ||
4081 offset + length > i_size_read(inode))
4082 iomap->flags |= IOMAP_F_DIRTY;
4087 const struct iomap_ops f2fs_iomap_ops = {
4088 .iomap_begin = f2fs_iomap_begin,