1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/buffer_head.h>
11 #include <linux/mpage.h>
12 #include <linux/writeback.h>
13 #include <linux/backing-dev.h>
14 #include <linux/pagevec.h>
15 #include <linux/blkdev.h>
16 #include <linux/bio.h>
17 #include <linux/swap.h>
18 #include <linux/prefetch.h>
19 #include <linux/uio.h>
20 #include <linux/cleancache.h>
21 #include <linux/sched/signal.h>
27 #include <trace/events/f2fs.h>
29 #define NUM_PREALLOC_POST_READ_CTXS 128
31 static struct kmem_cache *bio_post_read_ctx_cache;
32 static mempool_t *bio_post_read_ctx_pool;
34 static bool __is_cp_guaranteed(struct page *page)
36 struct address_space *mapping = page->mapping;
38 struct f2fs_sb_info *sbi;
43 inode = mapping->host;
44 sbi = F2FS_I_SB(inode);
46 if (inode->i_ino == F2FS_META_INO(sbi) ||
47 inode->i_ino == F2FS_NODE_INO(sbi) ||
48 S_ISDIR(inode->i_mode) ||
49 (S_ISREG(inode->i_mode) &&
50 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
56 static enum count_type __read_io_type(struct page *page)
58 struct address_space *mapping = page_file_mapping(page);
61 struct inode *inode = mapping->host;
62 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
64 if (inode->i_ino == F2FS_META_INO(sbi))
67 if (inode->i_ino == F2FS_NODE_INO(sbi))
73 /* postprocessing steps for read bios */
74 enum bio_post_read_step {
79 struct bio_post_read_ctx {
81 struct work_struct work;
82 unsigned int cur_step;
83 unsigned int enabled_steps;
86 static void __read_end_io(struct bio *bio)
90 struct bvec_iter_all iter_all;
92 bio_for_each_segment_all(bv, bio, iter_all) {
95 /* PG_error was set if any post_read step failed */
96 if (bio->bi_status || PageError(page)) {
97 ClearPageUptodate(page);
98 /* will re-read again later */
101 SetPageUptodate(page);
103 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
107 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
111 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
113 static void decrypt_work(struct work_struct *work)
115 struct bio_post_read_ctx *ctx =
116 container_of(work, struct bio_post_read_ctx, work);
118 fscrypt_decrypt_bio(ctx->bio);
120 bio_post_read_processing(ctx);
123 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
125 switch (++ctx->cur_step) {
127 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
128 INIT_WORK(&ctx->work, decrypt_work);
129 fscrypt_enqueue_decrypt_work(&ctx->work);
135 __read_end_io(ctx->bio);
139 static bool f2fs_bio_post_read_required(struct bio *bio)
141 return bio->bi_private && !bio->bi_status;
144 static void f2fs_read_end_io(struct bio *bio)
146 if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)),
148 f2fs_show_injection_info(FAULT_READ_IO);
149 bio->bi_status = BLK_STS_IOERR;
152 if (f2fs_bio_post_read_required(bio)) {
153 struct bio_post_read_ctx *ctx = bio->bi_private;
155 ctx->cur_step = STEP_INITIAL;
156 bio_post_read_processing(ctx);
163 static void f2fs_write_end_io(struct bio *bio)
165 struct f2fs_sb_info *sbi = bio->bi_private;
166 struct bio_vec *bvec;
167 struct bvec_iter_all iter_all;
169 if (time_to_inject(sbi, FAULT_WRITE_IO)) {
170 f2fs_show_injection_info(FAULT_WRITE_IO);
171 bio->bi_status = BLK_STS_IOERR;
174 bio_for_each_segment_all(bvec, bio, iter_all) {
175 struct page *page = bvec->bv_page;
176 enum count_type type = WB_DATA_TYPE(page);
178 if (IS_DUMMY_WRITTEN_PAGE(page)) {
179 set_page_private(page, (unsigned long)NULL);
180 ClearPagePrivate(page);
182 mempool_free(page, sbi->write_io_dummy);
184 if (unlikely(bio->bi_status))
185 f2fs_stop_checkpoint(sbi, true);
189 fscrypt_finalize_bounce_page(&page);
191 if (unlikely(bio->bi_status)) {
192 mapping_set_error(page->mapping, -EIO);
193 if (type == F2FS_WB_CP_DATA)
194 f2fs_stop_checkpoint(sbi, true);
197 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
198 page->index != nid_of_node(page));
200 dec_page_count(sbi, type);
201 if (f2fs_in_warm_node_list(sbi, page))
202 f2fs_del_fsync_node_entry(sbi, page);
203 clear_cold_data(page);
204 end_page_writeback(page);
206 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
207 wq_has_sleeper(&sbi->cp_wait))
208 wake_up(&sbi->cp_wait);
214 * Return true, if pre_bio's bdev is same as its target device.
216 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
217 block_t blk_addr, struct bio *bio)
219 struct block_device *bdev = sbi->sb->s_bdev;
222 if (f2fs_is_multi_device(sbi)) {
223 for (i = 0; i < sbi->s_ndevs; i++) {
224 if (FDEV(i).start_blk <= blk_addr &&
225 FDEV(i).end_blk >= blk_addr) {
226 blk_addr -= FDEV(i).start_blk;
233 bio_set_dev(bio, bdev);
234 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
239 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
243 if (!f2fs_is_multi_device(sbi))
246 for (i = 0; i < sbi->s_ndevs; i++)
247 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
252 static bool __same_bdev(struct f2fs_sb_info *sbi,
253 block_t blk_addr, struct bio *bio)
255 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
256 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
260 * Low-level block read/write IO operations.
262 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
263 struct writeback_control *wbc,
264 int npages, bool is_read,
265 enum page_type type, enum temp_type temp)
269 bio = f2fs_bio_alloc(sbi, npages, true);
271 f2fs_target_device(sbi, blk_addr, bio);
273 bio->bi_end_io = f2fs_read_end_io;
274 bio->bi_private = NULL;
276 bio->bi_end_io = f2fs_write_end_io;
277 bio->bi_private = sbi;
278 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
281 wbc_init_bio(wbc, bio);
286 static inline void __submit_bio(struct f2fs_sb_info *sbi,
287 struct bio *bio, enum page_type type)
289 if (!is_read_io(bio_op(bio))) {
292 if (type != DATA && type != NODE)
295 if (test_opt(sbi, LFS) && current->plug)
296 blk_finish_plug(current->plug);
298 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
299 start %= F2FS_IO_SIZE(sbi);
304 /* fill dummy pages */
305 for (; start < F2FS_IO_SIZE(sbi); start++) {
307 mempool_alloc(sbi->write_io_dummy,
308 GFP_NOIO | __GFP_NOFAIL);
309 f2fs_bug_on(sbi, !page);
311 zero_user_segment(page, 0, PAGE_SIZE);
312 SetPagePrivate(page);
313 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
315 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
319 * In the NODE case, we lose next block address chain. So, we
320 * need to do checkpoint in f2fs_sync_file.
323 set_sbi_flag(sbi, SBI_NEED_CP);
326 if (is_read_io(bio_op(bio)))
327 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
329 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
333 static void __submit_merged_bio(struct f2fs_bio_info *io)
335 struct f2fs_io_info *fio = &io->fio;
340 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
342 if (is_read_io(fio->op))
343 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
345 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
347 __submit_bio(io->sbi, io->bio, fio->type);
351 static bool __has_merged_page(struct bio *bio, struct inode *inode,
352 struct page *page, nid_t ino)
354 struct bio_vec *bvec;
356 struct bvec_iter_all iter_all;
361 if (!inode && !page && !ino)
364 bio_for_each_segment_all(bvec, bio, iter_all) {
366 target = bvec->bv_page;
367 if (fscrypt_is_bounce_page(target))
368 target = fscrypt_pagecache_page(target);
370 if (inode && inode == target->mapping->host)
372 if (page && page == target)
374 if (ino && ino == ino_of_node(target))
381 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
382 enum page_type type, enum temp_type temp)
384 enum page_type btype = PAGE_TYPE_OF_BIO(type);
385 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
387 down_write(&io->io_rwsem);
389 /* change META to META_FLUSH in the checkpoint procedure */
390 if (type >= META_FLUSH) {
391 io->fio.type = META_FLUSH;
392 io->fio.op = REQ_OP_WRITE;
393 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
394 if (!test_opt(sbi, NOBARRIER))
395 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
397 __submit_merged_bio(io);
398 up_write(&io->io_rwsem);
401 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
402 struct inode *inode, struct page *page,
403 nid_t ino, enum page_type type, bool force)
408 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
410 enum page_type btype = PAGE_TYPE_OF_BIO(type);
411 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
413 down_read(&io->io_rwsem);
414 ret = __has_merged_page(io->bio, inode, page, ino);
415 up_read(&io->io_rwsem);
418 __f2fs_submit_merged_write(sbi, type, temp);
420 /* TODO: use HOT temp only for meta pages now. */
426 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
428 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
431 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
432 struct inode *inode, struct page *page,
433 nid_t ino, enum page_type type)
435 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
438 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
440 f2fs_submit_merged_write(sbi, DATA);
441 f2fs_submit_merged_write(sbi, NODE);
442 f2fs_submit_merged_write(sbi, META);
446 * Fill the locked page with data located in the block address.
447 * A caller needs to unlock the page on failure.
449 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
452 struct page *page = fio->encrypted_page ?
453 fio->encrypted_page : fio->page;
455 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
456 fio->is_por ? META_POR : (__is_meta_io(fio) ?
457 META_GENERIC : DATA_GENERIC_ENHANCE)))
458 return -EFSCORRUPTED;
460 trace_f2fs_submit_page_bio(page, fio);
461 f2fs_trace_ios(fio, 0);
463 /* Allocate a new bio */
464 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
465 1, is_read_io(fio->op), fio->type, fio->temp);
467 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
472 if (fio->io_wbc && !is_read_io(fio->op))
473 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
475 bio_set_op_attrs(bio, fio->op, fio->op_flags);
477 inc_page_count(fio->sbi, is_read_io(fio->op) ?
478 __read_io_type(page): WB_DATA_TYPE(fio->page));
480 __submit_bio(fio->sbi, bio, fio->type);
484 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
486 struct bio *bio = *fio->bio;
487 struct page *page = fio->encrypted_page ?
488 fio->encrypted_page : fio->page;
490 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
491 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
492 return -EFSCORRUPTED;
494 trace_f2fs_submit_page_bio(page, fio);
495 f2fs_trace_ios(fio, 0);
497 if (bio && (*fio->last_block + 1 != fio->new_blkaddr ||
498 !__same_bdev(fio->sbi, fio->new_blkaddr, bio))) {
499 __submit_bio(fio->sbi, bio, fio->type);
504 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
505 BIO_MAX_PAGES, false, fio->type, fio->temp);
506 bio_set_op_attrs(bio, fio->op, fio->op_flags);
509 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
510 __submit_bio(fio->sbi, bio, fio->type);
516 wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
518 inc_page_count(fio->sbi, WB_DATA_TYPE(page));
520 *fio->last_block = fio->new_blkaddr;
526 static void f2fs_submit_ipu_bio(struct f2fs_sb_info *sbi, struct bio **bio,
532 if (!__has_merged_page(*bio, NULL, page, 0))
535 __submit_bio(sbi, *bio, DATA);
539 void f2fs_submit_page_write(struct f2fs_io_info *fio)
541 struct f2fs_sb_info *sbi = fio->sbi;
542 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
543 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
544 struct page *bio_page;
546 f2fs_bug_on(sbi, is_read_io(fio->op));
548 down_write(&io->io_rwsem);
551 spin_lock(&io->io_lock);
552 if (list_empty(&io->io_list)) {
553 spin_unlock(&io->io_lock);
556 fio = list_first_entry(&io->io_list,
557 struct f2fs_io_info, list);
558 list_del(&fio->list);
559 spin_unlock(&io->io_lock);
562 verify_fio_blkaddr(fio);
564 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
566 /* set submitted = true as a return value */
567 fio->submitted = true;
569 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
571 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
572 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
573 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
574 __submit_merged_bio(io);
576 if (io->bio == NULL) {
577 if ((fio->type == DATA || fio->type == NODE) &&
578 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
579 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
583 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
584 BIO_MAX_PAGES, false,
585 fio->type, fio->temp);
589 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
590 __submit_merged_bio(io);
595 wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
597 io->last_block_in_bio = fio->new_blkaddr;
598 f2fs_trace_ios(fio, 0);
600 trace_f2fs_submit_page_write(fio->page, fio);
605 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
606 f2fs_is_checkpoint_ready(sbi))
607 __submit_merged_bio(io);
608 up_write(&io->io_rwsem);
611 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
612 unsigned nr_pages, unsigned op_flag)
614 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
616 struct bio_post_read_ctx *ctx;
617 unsigned int post_read_steps = 0;
619 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
621 return ERR_PTR(-ENOMEM);
622 f2fs_target_device(sbi, blkaddr, bio);
623 bio->bi_end_io = f2fs_read_end_io;
624 bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
626 if (f2fs_encrypted_file(inode))
627 post_read_steps |= 1 << STEP_DECRYPT;
628 if (post_read_steps) {
629 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
632 return ERR_PTR(-ENOMEM);
635 ctx->enabled_steps = post_read_steps;
636 bio->bi_private = ctx;
642 /* This can handle encryption stuffs */
643 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
646 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
649 bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
653 /* wait for GCed page writeback via META_MAPPING */
654 f2fs_wait_on_block_writeback(inode, blkaddr);
656 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
660 ClearPageError(page);
661 inc_page_count(sbi, F2FS_RD_DATA);
662 __submit_bio(sbi, bio, DATA);
666 static void __set_data_blkaddr(struct dnode_of_data *dn)
668 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
672 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
673 base = get_extra_isize(dn->inode);
675 /* Get physical address of data block */
676 addr_array = blkaddr_in_node(rn);
677 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
681 * Lock ordering for the change of data block address:
684 * update block addresses in the node page
686 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
688 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
689 __set_data_blkaddr(dn);
690 if (set_page_dirty(dn->node_page))
691 dn->node_changed = true;
694 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
696 dn->data_blkaddr = blkaddr;
697 f2fs_set_data_blkaddr(dn);
698 f2fs_update_extent_cache(dn);
701 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
702 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
704 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
710 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
712 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
715 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
716 dn->ofs_in_node, count);
718 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
720 for (; count > 0; dn->ofs_in_node++) {
721 block_t blkaddr = datablock_addr(dn->inode,
722 dn->node_page, dn->ofs_in_node);
723 if (blkaddr == NULL_ADDR) {
724 dn->data_blkaddr = NEW_ADDR;
725 __set_data_blkaddr(dn);
730 if (set_page_dirty(dn->node_page))
731 dn->node_changed = true;
735 /* Should keep dn->ofs_in_node unchanged */
736 int f2fs_reserve_new_block(struct dnode_of_data *dn)
738 unsigned int ofs_in_node = dn->ofs_in_node;
741 ret = f2fs_reserve_new_blocks(dn, 1);
742 dn->ofs_in_node = ofs_in_node;
746 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
748 bool need_put = dn->inode_page ? false : true;
751 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
755 if (dn->data_blkaddr == NULL_ADDR)
756 err = f2fs_reserve_new_block(dn);
762 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
764 struct extent_info ei = {0,0,0};
765 struct inode *inode = dn->inode;
767 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
768 dn->data_blkaddr = ei.blk + index - ei.fofs;
772 return f2fs_reserve_block(dn, index);
775 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
776 int op_flags, bool for_write)
778 struct address_space *mapping = inode->i_mapping;
779 struct dnode_of_data dn;
781 struct extent_info ei = {0,0,0};
784 page = f2fs_grab_cache_page(mapping, index, for_write);
786 return ERR_PTR(-ENOMEM);
788 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
789 dn.data_blkaddr = ei.blk + index - ei.fofs;
790 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
791 DATA_GENERIC_ENHANCE_READ)) {
798 set_new_dnode(&dn, inode, NULL, NULL, 0);
799 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
804 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
808 if (dn.data_blkaddr != NEW_ADDR &&
809 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
811 DATA_GENERIC_ENHANCE)) {
816 if (PageUptodate(page)) {
822 * A new dentry page is allocated but not able to be written, since its
823 * new inode page couldn't be allocated due to -ENOSPC.
824 * In such the case, its blkaddr can be remained as NEW_ADDR.
825 * see, f2fs_add_link -> f2fs_get_new_data_page ->
826 * f2fs_init_inode_metadata.
828 if (dn.data_blkaddr == NEW_ADDR) {
829 zero_user_segment(page, 0, PAGE_SIZE);
830 if (!PageUptodate(page))
831 SetPageUptodate(page);
836 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
842 f2fs_put_page(page, 1);
846 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
848 struct address_space *mapping = inode->i_mapping;
851 page = find_get_page(mapping, index);
852 if (page && PageUptodate(page))
854 f2fs_put_page(page, 0);
856 page = f2fs_get_read_data_page(inode, index, 0, false);
860 if (PageUptodate(page))
863 wait_on_page_locked(page);
864 if (unlikely(!PageUptodate(page))) {
865 f2fs_put_page(page, 0);
866 return ERR_PTR(-EIO);
872 * If it tries to access a hole, return an error.
873 * Because, the callers, functions in dir.c and GC, should be able to know
874 * whether this page exists or not.
876 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
879 struct address_space *mapping = inode->i_mapping;
882 page = f2fs_get_read_data_page(inode, index, 0, for_write);
886 /* wait for read completion */
888 if (unlikely(page->mapping != mapping)) {
889 f2fs_put_page(page, 1);
892 if (unlikely(!PageUptodate(page))) {
893 f2fs_put_page(page, 1);
894 return ERR_PTR(-EIO);
900 * Caller ensures that this data page is never allocated.
901 * A new zero-filled data page is allocated in the page cache.
903 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
905 * Note that, ipage is set only by make_empty_dir, and if any error occur,
906 * ipage should be released by this function.
908 struct page *f2fs_get_new_data_page(struct inode *inode,
909 struct page *ipage, pgoff_t index, bool new_i_size)
911 struct address_space *mapping = inode->i_mapping;
913 struct dnode_of_data dn;
916 page = f2fs_grab_cache_page(mapping, index, true);
919 * before exiting, we should make sure ipage will be released
920 * if any error occur.
922 f2fs_put_page(ipage, 1);
923 return ERR_PTR(-ENOMEM);
926 set_new_dnode(&dn, inode, ipage, NULL, 0);
927 err = f2fs_reserve_block(&dn, index);
929 f2fs_put_page(page, 1);
935 if (PageUptodate(page))
938 if (dn.data_blkaddr == NEW_ADDR) {
939 zero_user_segment(page, 0, PAGE_SIZE);
940 if (!PageUptodate(page))
941 SetPageUptodate(page);
943 f2fs_put_page(page, 1);
945 /* if ipage exists, blkaddr should be NEW_ADDR */
946 f2fs_bug_on(F2FS_I_SB(inode), ipage);
947 page = f2fs_get_lock_data_page(inode, index, true);
952 if (new_i_size && i_size_read(inode) <
953 ((loff_t)(index + 1) << PAGE_SHIFT))
954 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
958 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
960 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
961 struct f2fs_summary sum;
967 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
970 err = f2fs_get_node_info(sbi, dn->nid, &ni);
974 dn->data_blkaddr = datablock_addr(dn->inode,
975 dn->node_page, dn->ofs_in_node);
976 if (dn->data_blkaddr != NULL_ADDR)
979 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
983 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
984 old_blkaddr = dn->data_blkaddr;
985 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
986 &sum, seg_type, NULL, false);
987 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
988 invalidate_mapping_pages(META_MAPPING(sbi),
989 old_blkaddr, old_blkaddr);
990 f2fs_set_data_blkaddr(dn);
993 * i_size will be updated by direct_IO. Otherwise, we'll get stale
994 * data from unwritten block via dio_read.
999 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1001 struct inode *inode = file_inode(iocb->ki_filp);
1002 struct f2fs_map_blocks map;
1005 bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1007 /* convert inline data for Direct I/O*/
1009 err = f2fs_convert_inline_inode(inode);
1014 if (direct_io && allow_outplace_dio(inode, iocb, from))
1017 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
1020 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1021 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1022 if (map.m_len > map.m_lblk)
1023 map.m_len -= map.m_lblk;
1027 map.m_next_pgofs = NULL;
1028 map.m_next_extent = NULL;
1029 map.m_seg_type = NO_CHECK_TYPE;
1030 map.m_may_create = true;
1033 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1034 flag = f2fs_force_buffered_io(inode, iocb, from) ?
1035 F2FS_GET_BLOCK_PRE_AIO :
1036 F2FS_GET_BLOCK_PRE_DIO;
1039 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1040 err = f2fs_convert_inline_inode(inode);
1044 if (f2fs_has_inline_data(inode))
1047 flag = F2FS_GET_BLOCK_PRE_AIO;
1050 err = f2fs_map_blocks(inode, &map, 1, flag);
1051 if (map.m_len > 0 && err == -ENOSPC) {
1053 set_inode_flag(inode, FI_NO_PREALLOC);
1059 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1061 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1063 down_read(&sbi->node_change);
1065 up_read(&sbi->node_change);
1070 f2fs_unlock_op(sbi);
1075 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1076 * f2fs_map_blocks structure.
1077 * If original data blocks are allocated, then give them to blockdev.
1079 * a. preallocate requested block addresses
1080 * b. do not use extent cache for better performance
1081 * c. give the block addresses to blockdev
1083 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1084 int create, int flag)
1086 unsigned int maxblocks = map->m_len;
1087 struct dnode_of_data dn;
1088 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1089 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1090 pgoff_t pgofs, end_offset, end;
1091 int err = 0, ofs = 1;
1092 unsigned int ofs_in_node, last_ofs_in_node;
1094 struct extent_info ei = {0,0,0};
1096 unsigned int start_pgofs;
1104 /* it only supports block size == page size */
1105 pgofs = (pgoff_t)map->m_lblk;
1106 end = pgofs + maxblocks;
1108 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1109 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1113 map->m_pblk = ei.blk + pgofs - ei.fofs;
1114 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1115 map->m_flags = F2FS_MAP_MAPPED;
1116 if (map->m_next_extent)
1117 *map->m_next_extent = pgofs + map->m_len;
1119 /* for hardware encryption, but to avoid potential issue in future */
1120 if (flag == F2FS_GET_BLOCK_DIO)
1121 f2fs_wait_on_block_writeback_range(inode,
1122 map->m_pblk, map->m_len);
1127 if (map->m_may_create)
1128 __do_map_lock(sbi, flag, true);
1130 /* When reading holes, we need its node page */
1131 set_new_dnode(&dn, inode, NULL, NULL, 0);
1132 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1134 if (flag == F2FS_GET_BLOCK_BMAP)
1136 if (err == -ENOENT) {
1138 if (map->m_next_pgofs)
1139 *map->m_next_pgofs =
1140 f2fs_get_next_page_offset(&dn, pgofs);
1141 if (map->m_next_extent)
1142 *map->m_next_extent =
1143 f2fs_get_next_page_offset(&dn, pgofs);
1148 start_pgofs = pgofs;
1150 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1151 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1154 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1156 if (__is_valid_data_blkaddr(blkaddr) &&
1157 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1158 err = -EFSCORRUPTED;
1162 if (__is_valid_data_blkaddr(blkaddr)) {
1163 /* use out-place-update for driect IO under LFS mode */
1164 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1165 map->m_may_create) {
1166 err = __allocate_data_block(&dn, map->m_seg_type);
1168 blkaddr = dn.data_blkaddr;
1169 set_inode_flag(inode, FI_APPEND_WRITE);
1174 if (unlikely(f2fs_cp_error(sbi))) {
1178 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1179 if (blkaddr == NULL_ADDR) {
1181 last_ofs_in_node = dn.ofs_in_node;
1184 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1185 flag != F2FS_GET_BLOCK_DIO);
1186 err = __allocate_data_block(&dn,
1189 set_inode_flag(inode, FI_APPEND_WRITE);
1193 map->m_flags |= F2FS_MAP_NEW;
1194 blkaddr = dn.data_blkaddr;
1196 if (flag == F2FS_GET_BLOCK_BMAP) {
1200 if (flag == F2FS_GET_BLOCK_PRECACHE)
1202 if (flag == F2FS_GET_BLOCK_FIEMAP &&
1203 blkaddr == NULL_ADDR) {
1204 if (map->m_next_pgofs)
1205 *map->m_next_pgofs = pgofs + 1;
1208 if (flag != F2FS_GET_BLOCK_FIEMAP) {
1209 /* for defragment case */
1210 if (map->m_next_pgofs)
1211 *map->m_next_pgofs = pgofs + 1;
1217 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1220 if (map->m_len == 0) {
1221 /* preallocated unwritten block should be mapped for fiemap. */
1222 if (blkaddr == NEW_ADDR)
1223 map->m_flags |= F2FS_MAP_UNWRITTEN;
1224 map->m_flags |= F2FS_MAP_MAPPED;
1226 map->m_pblk = blkaddr;
1228 } else if ((map->m_pblk != NEW_ADDR &&
1229 blkaddr == (map->m_pblk + ofs)) ||
1230 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1231 flag == F2FS_GET_BLOCK_PRE_DIO) {
1242 /* preallocate blocks in batch for one dnode page */
1243 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1244 (pgofs == end || dn.ofs_in_node == end_offset)) {
1246 dn.ofs_in_node = ofs_in_node;
1247 err = f2fs_reserve_new_blocks(&dn, prealloc);
1251 map->m_len += dn.ofs_in_node - ofs_in_node;
1252 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1256 dn.ofs_in_node = end_offset;
1261 else if (dn.ofs_in_node < end_offset)
1264 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1265 if (map->m_flags & F2FS_MAP_MAPPED) {
1266 unsigned int ofs = start_pgofs - map->m_lblk;
1268 f2fs_update_extent_cache_range(&dn,
1269 start_pgofs, map->m_pblk + ofs,
1274 f2fs_put_dnode(&dn);
1276 if (map->m_may_create) {
1277 __do_map_lock(sbi, flag, false);
1278 f2fs_balance_fs(sbi, dn.node_changed);
1284 /* for hardware encryption, but to avoid potential issue in future */
1285 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1286 f2fs_wait_on_block_writeback_range(inode,
1287 map->m_pblk, map->m_len);
1289 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1290 if (map->m_flags & F2FS_MAP_MAPPED) {
1291 unsigned int ofs = start_pgofs - map->m_lblk;
1293 f2fs_update_extent_cache_range(&dn,
1294 start_pgofs, map->m_pblk + ofs,
1297 if (map->m_next_extent)
1298 *map->m_next_extent = pgofs + 1;
1300 f2fs_put_dnode(&dn);
1302 if (map->m_may_create) {
1303 __do_map_lock(sbi, flag, false);
1304 f2fs_balance_fs(sbi, dn.node_changed);
1307 trace_f2fs_map_blocks(inode, map, err);
1311 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1313 struct f2fs_map_blocks map;
1317 if (pos + len > i_size_read(inode))
1320 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1321 map.m_next_pgofs = NULL;
1322 map.m_next_extent = NULL;
1323 map.m_seg_type = NO_CHECK_TYPE;
1324 map.m_may_create = false;
1325 last_lblk = F2FS_BLK_ALIGN(pos + len);
1327 while (map.m_lblk < last_lblk) {
1328 map.m_len = last_lblk - map.m_lblk;
1329 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1330 if (err || map.m_len == 0)
1332 map.m_lblk += map.m_len;
1337 static int __get_data_block(struct inode *inode, sector_t iblock,
1338 struct buffer_head *bh, int create, int flag,
1339 pgoff_t *next_pgofs, int seg_type, bool may_write)
1341 struct f2fs_map_blocks map;
1344 map.m_lblk = iblock;
1345 map.m_len = bh->b_size >> inode->i_blkbits;
1346 map.m_next_pgofs = next_pgofs;
1347 map.m_next_extent = NULL;
1348 map.m_seg_type = seg_type;
1349 map.m_may_create = may_write;
1351 err = f2fs_map_blocks(inode, &map, create, flag);
1353 map_bh(bh, inode->i_sb, map.m_pblk);
1354 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1355 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1360 static int get_data_block(struct inode *inode, sector_t iblock,
1361 struct buffer_head *bh_result, int create, int flag,
1362 pgoff_t *next_pgofs)
1364 return __get_data_block(inode, iblock, bh_result, create,
1366 NO_CHECK_TYPE, create);
1369 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1370 struct buffer_head *bh_result, int create)
1372 return __get_data_block(inode, iblock, bh_result, create,
1373 F2FS_GET_BLOCK_DIO, NULL,
1374 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1378 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1379 struct buffer_head *bh_result, int create)
1381 return __get_data_block(inode, iblock, bh_result, create,
1382 F2FS_GET_BLOCK_DIO, NULL,
1383 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1387 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1388 struct buffer_head *bh_result, int create)
1390 /* Block number less than F2FS MAX BLOCKS */
1391 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1394 return __get_data_block(inode, iblock, bh_result, create,
1395 F2FS_GET_BLOCK_BMAP, NULL,
1396 NO_CHECK_TYPE, create);
1399 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1401 return (offset >> inode->i_blkbits);
1404 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1406 return (blk << inode->i_blkbits);
1409 static int f2fs_xattr_fiemap(struct inode *inode,
1410 struct fiemap_extent_info *fieinfo)
1412 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1414 struct node_info ni;
1415 __u64 phys = 0, len;
1417 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1420 if (f2fs_has_inline_xattr(inode)) {
1423 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1424 inode->i_ino, false);
1428 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1430 f2fs_put_page(page, 1);
1434 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1435 offset = offsetof(struct f2fs_inode, i_addr) +
1436 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1437 get_inline_xattr_addrs(inode));
1440 len = inline_xattr_size(inode);
1442 f2fs_put_page(page, 1);
1444 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1447 flags |= FIEMAP_EXTENT_LAST;
1449 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1450 if (err || err == 1)
1455 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1459 err = f2fs_get_node_info(sbi, xnid, &ni);
1461 f2fs_put_page(page, 1);
1465 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1466 len = inode->i_sb->s_blocksize;
1468 f2fs_put_page(page, 1);
1470 flags = FIEMAP_EXTENT_LAST;
1474 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1476 return (err < 0 ? err : 0);
1479 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1482 struct buffer_head map_bh;
1483 sector_t start_blk, last_blk;
1485 u64 logical = 0, phys = 0, size = 0;
1489 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1490 ret = f2fs_precache_extents(inode);
1495 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1501 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1502 ret = f2fs_xattr_fiemap(inode, fieinfo);
1506 if (f2fs_has_inline_data(inode)) {
1507 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1512 if (logical_to_blk(inode, len) == 0)
1513 len = blk_to_logical(inode, 1);
1515 start_blk = logical_to_blk(inode, start);
1516 last_blk = logical_to_blk(inode, start + len - 1);
1519 memset(&map_bh, 0, sizeof(struct buffer_head));
1520 map_bh.b_size = len;
1522 ret = get_data_block(inode, start_blk, &map_bh, 0,
1523 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1528 if (!buffer_mapped(&map_bh)) {
1529 start_blk = next_pgofs;
1531 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1532 F2FS_I_SB(inode)->max_file_blocks))
1535 flags |= FIEMAP_EXTENT_LAST;
1539 if (IS_ENCRYPTED(inode))
1540 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1542 ret = fiemap_fill_next_extent(fieinfo, logical,
1546 if (start_blk > last_blk || ret)
1549 logical = blk_to_logical(inode, start_blk);
1550 phys = blk_to_logical(inode, map_bh.b_blocknr);
1551 size = map_bh.b_size;
1553 if (buffer_unwritten(&map_bh))
1554 flags = FIEMAP_EXTENT_UNWRITTEN;
1556 start_blk += logical_to_blk(inode, size);
1560 if (fatal_signal_pending(current))
1568 inode_unlock(inode);
1572 static int f2fs_read_single_page(struct inode *inode, struct page *page,
1574 struct f2fs_map_blocks *map,
1575 struct bio **bio_ret,
1576 sector_t *last_block_in_bio,
1579 struct bio *bio = *bio_ret;
1580 const unsigned blkbits = inode->i_blkbits;
1581 const unsigned blocksize = 1 << blkbits;
1582 sector_t block_in_file;
1583 sector_t last_block;
1584 sector_t last_block_in_file;
1588 block_in_file = (sector_t)page_index(page);
1589 last_block = block_in_file + nr_pages;
1590 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1592 if (last_block > last_block_in_file)
1593 last_block = last_block_in_file;
1595 /* just zeroing out page which is beyond EOF */
1596 if (block_in_file >= last_block)
1599 * Map blocks using the previous result first.
1601 if ((map->m_flags & F2FS_MAP_MAPPED) &&
1602 block_in_file > map->m_lblk &&
1603 block_in_file < (map->m_lblk + map->m_len))
1607 * Then do more f2fs_map_blocks() calls until we are
1608 * done with this page.
1610 map->m_lblk = block_in_file;
1611 map->m_len = last_block - block_in_file;
1613 ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
1617 if ((map->m_flags & F2FS_MAP_MAPPED)) {
1618 block_nr = map->m_pblk + block_in_file - map->m_lblk;
1619 SetPageMappedToDisk(page);
1621 if (!PageUptodate(page) && (!PageSwapCache(page) &&
1622 !cleancache_get_page(page))) {
1623 SetPageUptodate(page);
1627 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1628 DATA_GENERIC_ENHANCE_READ)) {
1629 ret = -EFSCORRUPTED;
1634 zero_user_segment(page, 0, PAGE_SIZE);
1635 if (!PageUptodate(page))
1636 SetPageUptodate(page);
1642 * This page will go to BIO. Do we need to send this
1645 if (bio && (*last_block_in_bio != block_nr - 1 ||
1646 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1648 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1652 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1653 is_readahead ? REQ_RAHEAD : 0);
1662 * If the page is under writeback, we need to wait for
1663 * its completion to see the correct decrypted data.
1665 f2fs_wait_on_block_writeback(inode, block_nr);
1667 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1668 goto submit_and_realloc;
1670 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1671 ClearPageError(page);
1672 *last_block_in_bio = block_nr;
1676 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1686 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1687 * Major change was from block_size == page_size in f2fs by default.
1689 * Note that the aops->readpages() function is ONLY used for read-ahead. If
1690 * this function ever deviates from doing just read-ahead, it should either
1691 * use ->readpage() or do the necessary surgery to decouple ->readpages()
1694 static int f2fs_mpage_readpages(struct address_space *mapping,
1695 struct list_head *pages, struct page *page,
1696 unsigned nr_pages, bool is_readahead)
1698 struct bio *bio = NULL;
1699 sector_t last_block_in_bio = 0;
1700 struct inode *inode = mapping->host;
1701 struct f2fs_map_blocks map;
1708 map.m_next_pgofs = NULL;
1709 map.m_next_extent = NULL;
1710 map.m_seg_type = NO_CHECK_TYPE;
1711 map.m_may_create = false;
1713 for (; nr_pages; nr_pages--) {
1715 page = list_last_entry(pages, struct page, lru);
1717 prefetchw(&page->flags);
1718 list_del(&page->lru);
1719 if (add_to_page_cache_lru(page, mapping,
1721 readahead_gfp_mask(mapping)))
1725 ret = f2fs_read_single_page(inode, page, nr_pages, &map, &bio,
1726 &last_block_in_bio, is_readahead);
1729 zero_user_segment(page, 0, PAGE_SIZE);
1736 BUG_ON(pages && !list_empty(pages));
1738 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1739 return pages ? 0 : ret;
1742 static int f2fs_read_data_page(struct file *file, struct page *page)
1744 struct inode *inode = page_file_mapping(page)->host;
1747 trace_f2fs_readpage(page, DATA);
1749 /* If the file has inline data, try to read it directly */
1750 if (f2fs_has_inline_data(inode))
1751 ret = f2fs_read_inline_data(inode, page);
1753 ret = f2fs_mpage_readpages(page_file_mapping(page),
1754 NULL, page, 1, false);
1758 static int f2fs_read_data_pages(struct file *file,
1759 struct address_space *mapping,
1760 struct list_head *pages, unsigned nr_pages)
1762 struct inode *inode = mapping->host;
1763 struct page *page = list_last_entry(pages, struct page, lru);
1765 trace_f2fs_readpages(inode, page, nr_pages);
1767 /* If the file has inline data, skip readpages */
1768 if (f2fs_has_inline_data(inode))
1771 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1774 static int encrypt_one_page(struct f2fs_io_info *fio)
1776 struct inode *inode = fio->page->mapping->host;
1778 gfp_t gfp_flags = GFP_NOFS;
1780 if (!f2fs_encrypted_file(inode))
1783 /* wait for GCed page writeback via META_MAPPING */
1784 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1787 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(fio->page,
1790 if (IS_ERR(fio->encrypted_page)) {
1791 /* flush pending IOs and wait for a while in the ENOMEM case */
1792 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1793 f2fs_flush_merged_writes(fio->sbi);
1794 congestion_wait(BLK_RW_ASYNC, HZ/50);
1795 gfp_flags |= __GFP_NOFAIL;
1798 return PTR_ERR(fio->encrypted_page);
1801 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1803 if (PageUptodate(mpage))
1804 memcpy(page_address(mpage),
1805 page_address(fio->encrypted_page), PAGE_SIZE);
1806 f2fs_put_page(mpage, 1);
1811 static inline bool check_inplace_update_policy(struct inode *inode,
1812 struct f2fs_io_info *fio)
1814 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1815 unsigned int policy = SM_I(sbi)->ipu_policy;
1817 if (policy & (0x1 << F2FS_IPU_FORCE))
1819 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1821 if (policy & (0x1 << F2FS_IPU_UTIL) &&
1822 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1824 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1825 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1829 * IPU for rewrite async pages
1831 if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1832 fio && fio->op == REQ_OP_WRITE &&
1833 !(fio->op_flags & REQ_SYNC) &&
1834 !IS_ENCRYPTED(inode))
1837 /* this is only set during fdatasync */
1838 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1839 is_inode_flag_set(inode, FI_NEED_IPU))
1842 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1843 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1849 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1851 if (f2fs_is_pinned_file(inode))
1854 /* if this is cold file, we should overwrite to avoid fragmentation */
1855 if (file_is_cold(inode))
1858 return check_inplace_update_policy(inode, fio);
1861 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1863 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1865 if (test_opt(sbi, LFS))
1867 if (S_ISDIR(inode->i_mode))
1869 if (IS_NOQUOTA(inode))
1871 if (f2fs_is_atomic_file(inode))
1874 if (is_cold_data(fio->page))
1876 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1878 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1879 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1885 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1887 struct inode *inode = fio->page->mapping->host;
1889 if (f2fs_should_update_outplace(inode, fio))
1892 return f2fs_should_update_inplace(inode, fio);
1895 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1897 struct page *page = fio->page;
1898 struct inode *inode = page->mapping->host;
1899 struct dnode_of_data dn;
1900 struct extent_info ei = {0,0,0};
1901 struct node_info ni;
1902 bool ipu_force = false;
1905 set_new_dnode(&dn, inode, NULL, NULL, 0);
1906 if (need_inplace_update(fio) &&
1907 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1908 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1910 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1911 DATA_GENERIC_ENHANCE))
1912 return -EFSCORRUPTED;
1915 fio->need_lock = LOCK_DONE;
1919 /* Deadlock due to between page->lock and f2fs_lock_op */
1920 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1923 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1927 fio->old_blkaddr = dn.data_blkaddr;
1929 /* This page is already truncated */
1930 if (fio->old_blkaddr == NULL_ADDR) {
1931 ClearPageUptodate(page);
1932 clear_cold_data(page);
1936 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1937 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1938 DATA_GENERIC_ENHANCE)) {
1939 err = -EFSCORRUPTED;
1943 * If current allocation needs SSR,
1944 * it had better in-place writes for updated data.
1947 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1948 need_inplace_update(fio))) {
1949 err = encrypt_one_page(fio);
1953 set_page_writeback(page);
1954 ClearPageError(page);
1955 f2fs_put_dnode(&dn);
1956 if (fio->need_lock == LOCK_REQ)
1957 f2fs_unlock_op(fio->sbi);
1958 err = f2fs_inplace_write_data(fio);
1960 if (f2fs_encrypted_file(inode))
1961 fscrypt_finalize_bounce_page(&fio->encrypted_page);
1962 if (PageWriteback(page))
1963 end_page_writeback(page);
1965 set_inode_flag(inode, FI_UPDATE_WRITE);
1967 trace_f2fs_do_write_data_page(fio->page, IPU);
1971 if (fio->need_lock == LOCK_RETRY) {
1972 if (!f2fs_trylock_op(fio->sbi)) {
1976 fio->need_lock = LOCK_REQ;
1979 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1983 fio->version = ni.version;
1985 err = encrypt_one_page(fio);
1989 set_page_writeback(page);
1990 ClearPageError(page);
1992 /* LFS mode write path */
1993 f2fs_outplace_write_data(&dn, fio);
1994 trace_f2fs_do_write_data_page(page, OPU);
1995 set_inode_flag(inode, FI_APPEND_WRITE);
1996 if (page->index == 0)
1997 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1999 f2fs_put_dnode(&dn);
2001 if (fio->need_lock == LOCK_REQ)
2002 f2fs_unlock_op(fio->sbi);
2006 static int __write_data_page(struct page *page, bool *submitted,
2008 sector_t *last_block,
2009 struct writeback_control *wbc,
2010 enum iostat_type io_type)
2012 struct inode *inode = page->mapping->host;
2013 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2014 loff_t i_size = i_size_read(inode);
2015 const pgoff_t end_index = ((unsigned long long) i_size)
2017 loff_t psize = (page->index + 1) << PAGE_SHIFT;
2018 unsigned offset = 0;
2019 bool need_balance_fs = false;
2021 struct f2fs_io_info fio = {
2023 .ino = inode->i_ino,
2026 .op_flags = wbc_to_write_flags(wbc),
2027 .old_blkaddr = NULL_ADDR,
2029 .encrypted_page = NULL,
2031 .need_lock = LOCK_RETRY,
2035 .last_block = last_block,
2038 trace_f2fs_writepage(page, DATA);
2040 /* we should bypass data pages to proceed the kworkder jobs */
2041 if (unlikely(f2fs_cp_error(sbi))) {
2042 mapping_set_error(page->mapping, -EIO);
2044 * don't drop any dirty dentry pages for keeping lastest
2045 * directory structure.
2047 if (S_ISDIR(inode->i_mode))
2052 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2055 if (page->index < end_index)
2059 * If the offset is out-of-range of file size,
2060 * this page does not have to be written to disk.
2062 offset = i_size & (PAGE_SIZE - 1);
2063 if ((page->index >= end_index + 1) || !offset)
2066 zero_user_segment(page, offset, PAGE_SIZE);
2068 if (f2fs_is_drop_cache(inode))
2070 /* we should not write 0'th page having journal header */
2071 if (f2fs_is_volatile_file(inode) && (!page->index ||
2072 (!wbc->for_reclaim &&
2073 f2fs_available_free_memory(sbi, BASE_CHECK))))
2076 /* Dentry blocks are controlled by checkpoint */
2077 if (S_ISDIR(inode->i_mode)) {
2078 fio.need_lock = LOCK_DONE;
2079 err = f2fs_do_write_data_page(&fio);
2083 if (!wbc->for_reclaim)
2084 need_balance_fs = true;
2085 else if (has_not_enough_free_secs(sbi, 0, 0))
2088 set_inode_flag(inode, FI_HOT_DATA);
2091 if (f2fs_has_inline_data(inode)) {
2092 err = f2fs_write_inline_data(inode, page);
2097 if (err == -EAGAIN) {
2098 err = f2fs_do_write_data_page(&fio);
2099 if (err == -EAGAIN) {
2100 fio.need_lock = LOCK_REQ;
2101 err = f2fs_do_write_data_page(&fio);
2106 file_set_keep_isize(inode);
2108 down_write(&F2FS_I(inode)->i_sem);
2109 if (F2FS_I(inode)->last_disk_size < psize)
2110 F2FS_I(inode)->last_disk_size = psize;
2111 up_write(&F2FS_I(inode)->i_sem);
2115 if (err && err != -ENOENT)
2119 inode_dec_dirty_pages(inode);
2121 ClearPageUptodate(page);
2122 clear_cold_data(page);
2125 if (wbc->for_reclaim) {
2126 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2127 clear_inode_flag(inode, FI_HOT_DATA);
2128 f2fs_remove_dirty_inode(inode);
2133 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2134 !F2FS_I(inode)->cp_task) {
2135 f2fs_submit_ipu_bio(sbi, bio, page);
2136 f2fs_balance_fs(sbi, need_balance_fs);
2139 if (unlikely(f2fs_cp_error(sbi))) {
2140 f2fs_submit_ipu_bio(sbi, bio, page);
2141 f2fs_submit_merged_write(sbi, DATA);
2146 *submitted = fio.submitted;
2151 redirty_page_for_writepage(wbc, page);
2153 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2154 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2155 * file_write_and_wait_range() will see EIO error, which is critical
2156 * to return value of fsync() followed by atomic_write failure to user.
2158 if (!err || wbc->for_reclaim)
2159 return AOP_WRITEPAGE_ACTIVATE;
2164 static int f2fs_write_data_page(struct page *page,
2165 struct writeback_control *wbc)
2167 return __write_data_page(page, NULL, NULL, NULL, wbc, FS_DATA_IO);
2171 * This function was copied from write_cche_pages from mm/page-writeback.c.
2172 * The major change is making write step of cold data page separately from
2173 * warm/hot data page.
2175 static int f2fs_write_cache_pages(struct address_space *mapping,
2176 struct writeback_control *wbc,
2177 enum iostat_type io_type)
2181 struct pagevec pvec;
2182 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2183 struct bio *bio = NULL;
2184 sector_t last_block;
2186 pgoff_t uninitialized_var(writeback_index);
2188 pgoff_t end; /* Inclusive */
2191 int range_whole = 0;
2195 pagevec_init(&pvec);
2197 if (get_dirty_pages(mapping->host) <=
2198 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2199 set_inode_flag(mapping->host, FI_HOT_DATA);
2201 clear_inode_flag(mapping->host, FI_HOT_DATA);
2203 if (wbc->range_cyclic) {
2204 writeback_index = mapping->writeback_index; /* prev offset */
2205 index = writeback_index;
2212 index = wbc->range_start >> PAGE_SHIFT;
2213 end = wbc->range_end >> PAGE_SHIFT;
2214 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2216 cycled = 1; /* ignore range_cyclic tests */
2218 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2219 tag = PAGECACHE_TAG_TOWRITE;
2221 tag = PAGECACHE_TAG_DIRTY;
2223 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2224 tag_pages_for_writeback(mapping, index, end);
2226 while (!done && (index <= end)) {
2229 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2234 for (i = 0; i < nr_pages; i++) {
2235 struct page *page = pvec.pages[i];
2236 bool submitted = false;
2238 /* give a priority to WB_SYNC threads */
2239 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2240 wbc->sync_mode == WB_SYNC_NONE) {
2245 done_index = page->index;
2249 if (unlikely(page->mapping != mapping)) {
2255 if (!PageDirty(page)) {
2256 /* someone wrote it for us */
2257 goto continue_unlock;
2260 if (PageWriteback(page)) {
2261 if (wbc->sync_mode != WB_SYNC_NONE) {
2262 f2fs_wait_on_page_writeback(page,
2264 f2fs_submit_ipu_bio(sbi, &bio, page);
2266 goto continue_unlock;
2270 if (!clear_page_dirty_for_io(page))
2271 goto continue_unlock;
2273 ret = __write_data_page(page, &submitted, &bio,
2274 &last_block, wbc, io_type);
2275 if (unlikely(ret)) {
2277 * keep nr_to_write, since vfs uses this to
2278 * get # of written pages.
2280 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2284 } else if (ret == -EAGAIN) {
2286 if (wbc->sync_mode == WB_SYNC_ALL) {
2288 congestion_wait(BLK_RW_ASYNC,
2294 done_index = page->index + 1;
2297 } else if (submitted) {
2301 if (--wbc->nr_to_write <= 0 &&
2302 wbc->sync_mode == WB_SYNC_NONE) {
2307 pagevec_release(&pvec);
2311 if (!cycled && !done) {
2314 end = writeback_index - 1;
2317 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2318 mapping->writeback_index = done_index;
2321 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2323 /* submit cached bio of IPU write */
2325 __submit_bio(sbi, bio, DATA);
2330 static inline bool __should_serialize_io(struct inode *inode,
2331 struct writeback_control *wbc)
2333 if (!S_ISREG(inode->i_mode))
2335 if (IS_NOQUOTA(inode))
2337 /* to avoid deadlock in path of data flush */
2338 if (F2FS_I(inode)->cp_task)
2340 if (wbc->sync_mode != WB_SYNC_ALL)
2342 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2347 static int __f2fs_write_data_pages(struct address_space *mapping,
2348 struct writeback_control *wbc,
2349 enum iostat_type io_type)
2351 struct inode *inode = mapping->host;
2352 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2353 struct blk_plug plug;
2355 bool locked = false;
2357 /* deal with chardevs and other special file */
2358 if (!mapping->a_ops->writepage)
2361 /* skip writing if there is no dirty page in this inode */
2362 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2365 /* during POR, we don't need to trigger writepage at all. */
2366 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2369 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2370 wbc->sync_mode == WB_SYNC_NONE &&
2371 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2372 f2fs_available_free_memory(sbi, DIRTY_DENTS))
2375 /* skip writing during file defragment */
2376 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2379 trace_f2fs_writepages(mapping->host, wbc, DATA);
2381 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2382 if (wbc->sync_mode == WB_SYNC_ALL)
2383 atomic_inc(&sbi->wb_sync_req[DATA]);
2384 else if (atomic_read(&sbi->wb_sync_req[DATA]))
2387 if (__should_serialize_io(inode, wbc)) {
2388 mutex_lock(&sbi->writepages);
2392 blk_start_plug(&plug);
2393 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2394 blk_finish_plug(&plug);
2397 mutex_unlock(&sbi->writepages);
2399 if (wbc->sync_mode == WB_SYNC_ALL)
2400 atomic_dec(&sbi->wb_sync_req[DATA]);
2402 * if some pages were truncated, we cannot guarantee its mapping->host
2403 * to detect pending bios.
2406 f2fs_remove_dirty_inode(inode);
2410 wbc->pages_skipped += get_dirty_pages(inode);
2411 trace_f2fs_writepages(mapping->host, wbc, DATA);
2415 static int f2fs_write_data_pages(struct address_space *mapping,
2416 struct writeback_control *wbc)
2418 struct inode *inode = mapping->host;
2420 return __f2fs_write_data_pages(mapping, wbc,
2421 F2FS_I(inode)->cp_task == current ?
2422 FS_CP_DATA_IO : FS_DATA_IO);
2425 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2427 struct inode *inode = mapping->host;
2428 loff_t i_size = i_size_read(inode);
2431 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2432 down_write(&F2FS_I(inode)->i_mmap_sem);
2434 truncate_pagecache(inode, i_size);
2435 if (!IS_NOQUOTA(inode))
2436 f2fs_truncate_blocks(inode, i_size, true);
2438 up_write(&F2FS_I(inode)->i_mmap_sem);
2439 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2443 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2444 struct page *page, loff_t pos, unsigned len,
2445 block_t *blk_addr, bool *node_changed)
2447 struct inode *inode = page->mapping->host;
2448 pgoff_t index = page->index;
2449 struct dnode_of_data dn;
2451 bool locked = false;
2452 struct extent_info ei = {0,0,0};
2457 * we already allocated all the blocks, so we don't need to get
2458 * the block addresses when there is no need to fill the page.
2460 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2461 !is_inode_flag_set(inode, FI_NO_PREALLOC))
2464 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2465 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2466 flag = F2FS_GET_BLOCK_DEFAULT;
2468 flag = F2FS_GET_BLOCK_PRE_AIO;
2470 if (f2fs_has_inline_data(inode) ||
2471 (pos & PAGE_MASK) >= i_size_read(inode)) {
2472 __do_map_lock(sbi, flag, true);
2476 /* check inline_data */
2477 ipage = f2fs_get_node_page(sbi, inode->i_ino);
2478 if (IS_ERR(ipage)) {
2479 err = PTR_ERR(ipage);
2483 set_new_dnode(&dn, inode, ipage, ipage, 0);
2485 if (f2fs_has_inline_data(inode)) {
2486 if (pos + len <= MAX_INLINE_DATA(inode)) {
2487 f2fs_do_read_inline_data(page, ipage);
2488 set_inode_flag(inode, FI_DATA_EXIST);
2490 set_inline_node(ipage);
2492 err = f2fs_convert_inline_page(&dn, page);
2495 if (dn.data_blkaddr == NULL_ADDR)
2496 err = f2fs_get_block(&dn, index);
2498 } else if (locked) {
2499 err = f2fs_get_block(&dn, index);
2501 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2502 dn.data_blkaddr = ei.blk + index - ei.fofs;
2505 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2506 if (err || dn.data_blkaddr == NULL_ADDR) {
2507 f2fs_put_dnode(&dn);
2508 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2510 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2517 /* convert_inline_page can make node_changed */
2518 *blk_addr = dn.data_blkaddr;
2519 *node_changed = dn.node_changed;
2521 f2fs_put_dnode(&dn);
2524 __do_map_lock(sbi, flag, false);
2528 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2529 loff_t pos, unsigned len, unsigned flags,
2530 struct page **pagep, void **fsdata)
2532 struct inode *inode = mapping->host;
2533 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2534 struct page *page = NULL;
2535 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2536 bool need_balance = false, drop_atomic = false;
2537 block_t blkaddr = NULL_ADDR;
2540 trace_f2fs_write_begin(inode, pos, len, flags);
2542 err = f2fs_is_checkpoint_ready(sbi);
2546 if ((f2fs_is_atomic_file(inode) &&
2547 !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2548 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2555 * We should check this at this moment to avoid deadlock on inode page
2556 * and #0 page. The locking rule for inline_data conversion should be:
2557 * lock_page(page #0) -> lock_page(inode_page)
2560 err = f2fs_convert_inline_inode(inode);
2566 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2567 * wait_for_stable_page. Will wait that below with our IO control.
2569 page = f2fs_pagecache_get_page(mapping, index,
2570 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2578 err = prepare_write_begin(sbi, page, pos, len,
2579 &blkaddr, &need_balance);
2583 if (need_balance && !IS_NOQUOTA(inode) &&
2584 has_not_enough_free_secs(sbi, 0, 0)) {
2586 f2fs_balance_fs(sbi, true);
2588 if (page->mapping != mapping) {
2589 /* The page got truncated from under us */
2590 f2fs_put_page(page, 1);
2595 f2fs_wait_on_page_writeback(page, DATA, false, true);
2597 if (len == PAGE_SIZE || PageUptodate(page))
2600 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2601 zero_user_segment(page, len, PAGE_SIZE);
2605 if (blkaddr == NEW_ADDR) {
2606 zero_user_segment(page, 0, PAGE_SIZE);
2607 SetPageUptodate(page);
2609 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
2610 DATA_GENERIC_ENHANCE_READ)) {
2611 err = -EFSCORRUPTED;
2614 err = f2fs_submit_page_read(inode, page, blkaddr);
2619 if (unlikely(page->mapping != mapping)) {
2620 f2fs_put_page(page, 1);
2623 if (unlikely(!PageUptodate(page))) {
2631 f2fs_put_page(page, 1);
2632 f2fs_write_failed(mapping, pos + len);
2634 f2fs_drop_inmem_pages_all(sbi, false);
2638 static int f2fs_write_end(struct file *file,
2639 struct address_space *mapping,
2640 loff_t pos, unsigned len, unsigned copied,
2641 struct page *page, void *fsdata)
2643 struct inode *inode = page->mapping->host;
2645 trace_f2fs_write_end(inode, pos, len, copied);
2648 * This should be come from len == PAGE_SIZE, and we expect copied
2649 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2650 * let generic_perform_write() try to copy data again through copied=0.
2652 if (!PageUptodate(page)) {
2653 if (unlikely(copied != len))
2656 SetPageUptodate(page);
2661 set_page_dirty(page);
2663 if (pos + copied > i_size_read(inode))
2664 f2fs_i_size_write(inode, pos + copied);
2666 f2fs_put_page(page, 1);
2667 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2671 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2674 unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2675 unsigned blkbits = i_blkbits;
2676 unsigned blocksize_mask = (1 << blkbits) - 1;
2677 unsigned long align = offset | iov_iter_alignment(iter);
2678 struct block_device *bdev = inode->i_sb->s_bdev;
2680 if (align & blocksize_mask) {
2682 blkbits = blksize_bits(bdev_logical_block_size(bdev));
2683 blocksize_mask = (1 << blkbits) - 1;
2684 if (align & blocksize_mask)
2691 static void f2fs_dio_end_io(struct bio *bio)
2693 struct f2fs_private_dio *dio = bio->bi_private;
2695 dec_page_count(F2FS_I_SB(dio->inode),
2696 dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2698 bio->bi_private = dio->orig_private;
2699 bio->bi_end_io = dio->orig_end_io;
2706 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2709 struct f2fs_private_dio *dio;
2710 bool write = (bio_op(bio) == REQ_OP_WRITE);
2712 dio = f2fs_kzalloc(F2FS_I_SB(inode),
2713 sizeof(struct f2fs_private_dio), GFP_NOFS);
2718 dio->orig_end_io = bio->bi_end_io;
2719 dio->orig_private = bio->bi_private;
2722 bio->bi_end_io = f2fs_dio_end_io;
2723 bio->bi_private = dio;
2725 inc_page_count(F2FS_I_SB(inode),
2726 write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2731 bio->bi_status = BLK_STS_IOERR;
2735 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2737 struct address_space *mapping = iocb->ki_filp->f_mapping;
2738 struct inode *inode = mapping->host;
2739 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2740 struct f2fs_inode_info *fi = F2FS_I(inode);
2741 size_t count = iov_iter_count(iter);
2742 loff_t offset = iocb->ki_pos;
2743 int rw = iov_iter_rw(iter);
2745 enum rw_hint hint = iocb->ki_hint;
2746 int whint_mode = F2FS_OPTION(sbi).whint_mode;
2749 err = check_direct_IO(inode, iter, offset);
2751 return err < 0 ? err : 0;
2753 if (f2fs_force_buffered_io(inode, iocb, iter))
2756 do_opu = allow_outplace_dio(inode, iocb, iter);
2758 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2760 if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2761 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2763 if (iocb->ki_flags & IOCB_NOWAIT) {
2764 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2765 iocb->ki_hint = hint;
2769 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2770 up_read(&fi->i_gc_rwsem[rw]);
2771 iocb->ki_hint = hint;
2776 down_read(&fi->i_gc_rwsem[rw]);
2778 down_read(&fi->i_gc_rwsem[READ]);
2781 err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2782 iter, rw == WRITE ? get_data_block_dio_write :
2783 get_data_block_dio, NULL, f2fs_dio_submit_bio,
2784 DIO_LOCKING | DIO_SKIP_HOLES);
2787 up_read(&fi->i_gc_rwsem[READ]);
2789 up_read(&fi->i_gc_rwsem[rw]);
2792 if (whint_mode == WHINT_MODE_OFF)
2793 iocb->ki_hint = hint;
2795 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2798 set_inode_flag(inode, FI_UPDATE_WRITE);
2799 } else if (err < 0) {
2800 f2fs_write_failed(mapping, offset + count);
2805 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2810 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2811 unsigned int length)
2813 struct inode *inode = page->mapping->host;
2814 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2816 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2817 (offset % PAGE_SIZE || length != PAGE_SIZE))
2820 if (PageDirty(page)) {
2821 if (inode->i_ino == F2FS_META_INO(sbi)) {
2822 dec_page_count(sbi, F2FS_DIRTY_META);
2823 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2824 dec_page_count(sbi, F2FS_DIRTY_NODES);
2826 inode_dec_dirty_pages(inode);
2827 f2fs_remove_dirty_inode(inode);
2831 clear_cold_data(page);
2833 if (IS_ATOMIC_WRITTEN_PAGE(page))
2834 return f2fs_drop_inmem_page(inode, page);
2836 f2fs_clear_page_private(page);
2839 int f2fs_release_page(struct page *page, gfp_t wait)
2841 /* If this is dirty page, keep PagePrivate */
2842 if (PageDirty(page))
2845 /* This is atomic written page, keep Private */
2846 if (IS_ATOMIC_WRITTEN_PAGE(page))
2849 clear_cold_data(page);
2850 f2fs_clear_page_private(page);
2854 static int f2fs_set_data_page_dirty(struct page *page)
2856 struct inode *inode = page_file_mapping(page)->host;
2858 trace_f2fs_set_page_dirty(page, DATA);
2860 if (!PageUptodate(page))
2861 SetPageUptodate(page);
2862 if (PageSwapCache(page))
2863 return __set_page_dirty_nobuffers(page);
2865 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2866 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2867 f2fs_register_inmem_page(inode, page);
2871 * Previously, this page has been registered, we just
2877 if (!PageDirty(page)) {
2878 __set_page_dirty_nobuffers(page);
2879 f2fs_update_dirty_page(inode, page);
2885 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2887 struct inode *inode = mapping->host;
2889 if (f2fs_has_inline_data(inode))
2892 /* make sure allocating whole blocks */
2893 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2894 filemap_write_and_wait(mapping);
2896 return generic_block_bmap(mapping, block, get_data_block_bmap);
2899 #ifdef CONFIG_MIGRATION
2900 #include <linux/migrate.h>
2902 int f2fs_migrate_page(struct address_space *mapping,
2903 struct page *newpage, struct page *page, enum migrate_mode mode)
2905 int rc, extra_count;
2906 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2907 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2909 BUG_ON(PageWriteback(page));
2911 /* migrating an atomic written page is safe with the inmem_lock hold */
2912 if (atomic_written) {
2913 if (mode != MIGRATE_SYNC)
2915 if (!mutex_trylock(&fi->inmem_lock))
2919 /* one extra reference was held for atomic_write page */
2920 extra_count = atomic_written ? 1 : 0;
2921 rc = migrate_page_move_mapping(mapping, newpage,
2922 page, mode, extra_count);
2923 if (rc != MIGRATEPAGE_SUCCESS) {
2925 mutex_unlock(&fi->inmem_lock);
2929 if (atomic_written) {
2930 struct inmem_pages *cur;
2931 list_for_each_entry(cur, &fi->inmem_pages, list)
2932 if (cur->page == page) {
2933 cur->page = newpage;
2936 mutex_unlock(&fi->inmem_lock);
2941 if (PagePrivate(page)) {
2942 f2fs_set_page_private(newpage, page_private(page));
2943 f2fs_clear_page_private(page);
2946 if (mode != MIGRATE_SYNC_NO_COPY)
2947 migrate_page_copy(newpage, page);
2949 migrate_page_states(newpage, page);
2951 return MIGRATEPAGE_SUCCESS;
2956 /* Copied from generic_swapfile_activate() to check any holes */
2957 static int check_swap_activate(struct file *swap_file, unsigned int max)
2959 struct address_space *mapping = swap_file->f_mapping;
2960 struct inode *inode = mapping->host;
2961 unsigned blocks_per_page;
2962 unsigned long page_no;
2964 sector_t probe_block;
2965 sector_t last_block;
2966 sector_t lowest_block = -1;
2967 sector_t highest_block = 0;
2969 blkbits = inode->i_blkbits;
2970 blocks_per_page = PAGE_SIZE >> blkbits;
2973 * Map all the blocks into the extent list. This code doesn't try
2978 last_block = i_size_read(inode) >> blkbits;
2979 while ((probe_block + blocks_per_page) <= last_block && page_no < max) {
2980 unsigned block_in_page;
2981 sector_t first_block;
2985 first_block = bmap(inode, probe_block);
2986 if (first_block == 0)
2990 * It must be PAGE_SIZE aligned on-disk
2992 if (first_block & (blocks_per_page - 1)) {
2997 for (block_in_page = 1; block_in_page < blocks_per_page;
3001 block = bmap(inode, probe_block + block_in_page);
3004 if (block != first_block + block_in_page) {
3011 first_block >>= (PAGE_SHIFT - blkbits);
3012 if (page_no) { /* exclude the header page */
3013 if (first_block < lowest_block)
3014 lowest_block = first_block;
3015 if (first_block > highest_block)
3016 highest_block = first_block;
3020 probe_block += blocks_per_page;
3027 pr_err("swapon: swapfile has holes\n");
3031 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3034 struct inode *inode = file_inode(file);
3037 if (!S_ISREG(inode->i_mode))
3040 if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3043 ret = f2fs_convert_inline_inode(inode);
3047 ret = check_swap_activate(file, sis->max);
3051 set_inode_flag(inode, FI_PIN_FILE);
3052 f2fs_precache_extents(inode);
3053 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3057 static void f2fs_swap_deactivate(struct file *file)
3059 struct inode *inode = file_inode(file);
3061 clear_inode_flag(inode, FI_PIN_FILE);
3064 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3070 static void f2fs_swap_deactivate(struct file *file)
3075 const struct address_space_operations f2fs_dblock_aops = {
3076 .readpage = f2fs_read_data_page,
3077 .readpages = f2fs_read_data_pages,
3078 .writepage = f2fs_write_data_page,
3079 .writepages = f2fs_write_data_pages,
3080 .write_begin = f2fs_write_begin,
3081 .write_end = f2fs_write_end,
3082 .set_page_dirty = f2fs_set_data_page_dirty,
3083 .invalidatepage = f2fs_invalidate_page,
3084 .releasepage = f2fs_release_page,
3085 .direct_IO = f2fs_direct_IO,
3087 .swap_activate = f2fs_swap_activate,
3088 .swap_deactivate = f2fs_swap_deactivate,
3089 #ifdef CONFIG_MIGRATION
3090 .migratepage = f2fs_migrate_page,
3094 void f2fs_clear_page_cache_dirty_tag(struct page *page)
3096 struct address_space *mapping = page_mapping(page);
3097 unsigned long flags;
3099 xa_lock_irqsave(&mapping->i_pages, flags);
3100 __xa_clear_mark(&mapping->i_pages, page_index(page),
3101 PAGECACHE_TAG_DIRTY);
3102 xa_unlock_irqrestore(&mapping->i_pages, flags);
3105 int __init f2fs_init_post_read_processing(void)
3107 bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
3108 if (!bio_post_read_ctx_cache)
3110 bio_post_read_ctx_pool =
3111 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
3112 bio_post_read_ctx_cache);
3113 if (!bio_post_read_ctx_pool)
3114 goto fail_free_cache;
3118 kmem_cache_destroy(bio_post_read_ctx_cache);
3123 void __exit f2fs_destroy_post_read_processing(void)
3125 mempool_destroy(bio_post_read_ctx_pool);
3126 kmem_cache_destroy(bio_post_read_ctx_cache);