1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/buffer_head.h>
11 #include <linux/mpage.h>
12 #include <linux/writeback.h>
13 #include <linux/backing-dev.h>
14 #include <linux/pagevec.h>
15 #include <linux/blkdev.h>
16 #include <linux/bio.h>
17 #include <linux/prefetch.h>
18 #include <linux/uio.h>
19 #include <linux/cleancache.h>
20 #include <linux/sched/signal.h>
26 #include <trace/events/f2fs.h>
28 #define NUM_PREALLOC_POST_READ_CTXS 128
30 static struct kmem_cache *bio_post_read_ctx_cache;
31 static mempool_t *bio_post_read_ctx_pool;
33 static bool __is_cp_guaranteed(struct page *page)
35 struct address_space *mapping = page->mapping;
37 struct f2fs_sb_info *sbi;
42 inode = mapping->host;
43 sbi = F2FS_I_SB(inode);
45 if (inode->i_ino == F2FS_META_INO(sbi) ||
46 inode->i_ino == F2FS_NODE_INO(sbi) ||
47 S_ISDIR(inode->i_mode) ||
48 (S_ISREG(inode->i_mode) &&
49 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
55 static enum count_type __read_io_type(struct page *page)
57 struct address_space *mapping = page->mapping;
60 struct inode *inode = mapping->host;
61 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
63 if (inode->i_ino == F2FS_META_INO(sbi))
66 if (inode->i_ino == F2FS_NODE_INO(sbi))
72 /* postprocessing steps for read bios */
73 enum bio_post_read_step {
78 struct bio_post_read_ctx {
80 struct work_struct work;
81 unsigned int cur_step;
82 unsigned int enabled_steps;
85 static void __read_end_io(struct bio *bio)
90 struct bvec_iter_all iter_all;
92 bio_for_each_segment_all(bv, bio, i, 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;
168 struct bvec_iter_all iter_all;
170 if (time_to_inject(sbi, FAULT_WRITE_IO)) {
171 f2fs_show_injection_info(FAULT_WRITE_IO);
172 bio->bi_status = BLK_STS_IOERR;
175 bio_for_each_segment_all(bvec, bio, i, iter_all) {
176 struct page *page = bvec->bv_page;
177 enum count_type type = WB_DATA_TYPE(page);
179 if (IS_DUMMY_WRITTEN_PAGE(page)) {
180 set_page_private(page, (unsigned long)NULL);
181 ClearPagePrivate(page);
183 mempool_free(page, sbi->write_io_dummy);
185 if (unlikely(bio->bi_status))
186 f2fs_stop_checkpoint(sbi, true);
190 fscrypt_pullback_bio_page(&page, true);
192 if (unlikely(bio->bi_status)) {
193 mapping_set_error(page->mapping, -EIO);
194 if (type == F2FS_WB_CP_DATA)
195 f2fs_stop_checkpoint(sbi, true);
198 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
199 page->index != nid_of_node(page));
201 dec_page_count(sbi, type);
202 if (f2fs_in_warm_node_list(sbi, page))
203 f2fs_del_fsync_node_entry(sbi, page);
204 clear_cold_data(page);
205 end_page_writeback(page);
207 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
208 wq_has_sleeper(&sbi->cp_wait))
209 wake_up(&sbi->cp_wait);
215 * Return true, if pre_bio's bdev is same as its target device.
217 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
218 block_t blk_addr, struct bio *bio)
220 struct block_device *bdev = sbi->sb->s_bdev;
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;
232 bio_set_dev(bio, bdev);
233 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
238 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
242 for (i = 0; i < sbi->s_ndevs; i++)
243 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
248 static bool __same_bdev(struct f2fs_sb_info *sbi,
249 block_t blk_addr, struct bio *bio)
251 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
252 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
256 * Low-level block read/write IO operations.
258 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
259 struct writeback_control *wbc,
260 int npages, bool is_read,
261 enum page_type type, enum temp_type temp)
265 bio = f2fs_bio_alloc(sbi, npages, true);
267 f2fs_target_device(sbi, blk_addr, bio);
269 bio->bi_end_io = f2fs_read_end_io;
270 bio->bi_private = NULL;
272 bio->bi_end_io = f2fs_write_end_io;
273 bio->bi_private = sbi;
274 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
277 wbc_init_bio(wbc, bio);
282 static inline void __submit_bio(struct f2fs_sb_info *sbi,
283 struct bio *bio, enum page_type type)
285 if (!is_read_io(bio_op(bio))) {
288 if (type != DATA && type != NODE)
291 if (test_opt(sbi, LFS) && current->plug)
292 blk_finish_plug(current->plug);
294 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
295 start %= F2FS_IO_SIZE(sbi);
300 /* fill dummy pages */
301 for (; start < F2FS_IO_SIZE(sbi); start++) {
303 mempool_alloc(sbi->write_io_dummy,
304 GFP_NOIO | __GFP_NOFAIL);
305 f2fs_bug_on(sbi, !page);
307 zero_user_segment(page, 0, PAGE_SIZE);
308 SetPagePrivate(page);
309 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
311 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
315 * In the NODE case, we lose next block address chain. So, we
316 * need to do checkpoint in f2fs_sync_file.
319 set_sbi_flag(sbi, SBI_NEED_CP);
322 if (is_read_io(bio_op(bio)))
323 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
325 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
329 static void __submit_merged_bio(struct f2fs_bio_info *io)
331 struct f2fs_io_info *fio = &io->fio;
336 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
338 if (is_read_io(fio->op))
339 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
341 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
343 __submit_bio(io->sbi, io->bio, fio->type);
347 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
348 struct page *page, nid_t ino)
350 struct bio_vec *bvec;
353 struct bvec_iter_all iter_all;
358 if (!inode && !page && !ino)
361 bio_for_each_segment_all(bvec, io->bio, i, iter_all) {
363 if (bvec->bv_page->mapping)
364 target = bvec->bv_page;
366 target = fscrypt_control_page(bvec->bv_page);
368 if (inode && inode == target->mapping->host)
370 if (page && page == target)
372 if (ino && ino == ino_of_node(target))
379 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
380 enum page_type type, enum temp_type temp)
382 enum page_type btype = PAGE_TYPE_OF_BIO(type);
383 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
385 down_write(&io->io_rwsem);
387 /* change META to META_FLUSH in the checkpoint procedure */
388 if (type >= META_FLUSH) {
389 io->fio.type = META_FLUSH;
390 io->fio.op = REQ_OP_WRITE;
391 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
392 if (!test_opt(sbi, NOBARRIER))
393 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
395 __submit_merged_bio(io);
396 up_write(&io->io_rwsem);
399 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
400 struct inode *inode, struct page *page,
401 nid_t ino, enum page_type type, bool force)
406 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
408 enum page_type btype = PAGE_TYPE_OF_BIO(type);
409 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
411 down_read(&io->io_rwsem);
412 ret = __has_merged_page(io, inode, page, ino);
413 up_read(&io->io_rwsem);
416 __f2fs_submit_merged_write(sbi, type, temp);
418 /* TODO: use HOT temp only for meta pages now. */
424 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
426 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
429 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
430 struct inode *inode, struct page *page,
431 nid_t ino, enum page_type type)
433 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
436 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
438 f2fs_submit_merged_write(sbi, DATA);
439 f2fs_submit_merged_write(sbi, NODE);
440 f2fs_submit_merged_write(sbi, META);
444 * Fill the locked page with data located in the block address.
445 * A caller needs to unlock the page on failure.
447 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
450 struct page *page = fio->encrypted_page ?
451 fio->encrypted_page : fio->page;
453 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
454 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
457 trace_f2fs_submit_page_bio(page, fio);
458 f2fs_trace_ios(fio, 0);
460 /* Allocate a new bio */
461 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
462 1, is_read_io(fio->op), fio->type, fio->temp);
464 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
469 if (fio->io_wbc && !is_read_io(fio->op))
470 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
472 bio_set_op_attrs(bio, fio->op, fio->op_flags);
474 inc_page_count(fio->sbi, is_read_io(fio->op) ?
475 __read_io_type(page): WB_DATA_TYPE(fio->page));
477 __submit_bio(fio->sbi, bio, fio->type);
481 void f2fs_submit_page_write(struct f2fs_io_info *fio)
483 struct f2fs_sb_info *sbi = fio->sbi;
484 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
485 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
486 struct page *bio_page;
488 f2fs_bug_on(sbi, is_read_io(fio->op));
490 down_write(&io->io_rwsem);
493 spin_lock(&io->io_lock);
494 if (list_empty(&io->io_list)) {
495 spin_unlock(&io->io_lock);
498 fio = list_first_entry(&io->io_list,
499 struct f2fs_io_info, list);
500 list_del(&fio->list);
501 spin_unlock(&io->io_lock);
504 if (__is_valid_data_blkaddr(fio->old_blkaddr))
505 verify_block_addr(fio, fio->old_blkaddr);
506 verify_block_addr(fio, fio->new_blkaddr);
508 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
510 /* set submitted = true as a return value */
511 fio->submitted = true;
513 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
515 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
516 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
517 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
518 __submit_merged_bio(io);
520 if (io->bio == NULL) {
521 if ((fio->type == DATA || fio->type == NODE) &&
522 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
523 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
527 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
528 BIO_MAX_PAGES, false,
529 fio->type, fio->temp);
533 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
534 __submit_merged_bio(io);
539 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
541 io->last_block_in_bio = fio->new_blkaddr;
542 f2fs_trace_ios(fio, 0);
544 trace_f2fs_submit_page_write(fio->page, fio);
549 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
550 f2fs_is_checkpoint_ready(sbi))
551 __submit_merged_bio(io);
552 up_write(&io->io_rwsem);
555 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
556 unsigned nr_pages, unsigned op_flag)
558 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
560 struct bio_post_read_ctx *ctx;
561 unsigned int post_read_steps = 0;
563 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
564 return ERR_PTR(-EFAULT);
566 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
568 return ERR_PTR(-ENOMEM);
569 f2fs_target_device(sbi, blkaddr, bio);
570 bio->bi_end_io = f2fs_read_end_io;
571 bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
573 if (f2fs_encrypted_file(inode))
574 post_read_steps |= 1 << STEP_DECRYPT;
575 if (post_read_steps) {
576 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
579 return ERR_PTR(-ENOMEM);
582 ctx->enabled_steps = post_read_steps;
583 bio->bi_private = ctx;
589 /* This can handle encryption stuffs */
590 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
593 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
598 /* wait for GCed page writeback via META_MAPPING */
599 f2fs_wait_on_block_writeback(inode, blkaddr);
601 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
605 ClearPageError(page);
606 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
607 __submit_bio(F2FS_I_SB(inode), bio, DATA);
611 static void __set_data_blkaddr(struct dnode_of_data *dn)
613 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
617 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
618 base = get_extra_isize(dn->inode);
620 /* Get physical address of data block */
621 addr_array = blkaddr_in_node(rn);
622 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
626 * Lock ordering for the change of data block address:
629 * update block addresses in the node page
631 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
633 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
634 __set_data_blkaddr(dn);
635 if (set_page_dirty(dn->node_page))
636 dn->node_changed = true;
639 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
641 dn->data_blkaddr = blkaddr;
642 f2fs_set_data_blkaddr(dn);
643 f2fs_update_extent_cache(dn);
646 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
647 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
649 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
655 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
657 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
660 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
661 dn->ofs_in_node, count);
663 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
665 for (; count > 0; dn->ofs_in_node++) {
666 block_t blkaddr = datablock_addr(dn->inode,
667 dn->node_page, dn->ofs_in_node);
668 if (blkaddr == NULL_ADDR) {
669 dn->data_blkaddr = NEW_ADDR;
670 __set_data_blkaddr(dn);
675 if (set_page_dirty(dn->node_page))
676 dn->node_changed = true;
680 /* Should keep dn->ofs_in_node unchanged */
681 int f2fs_reserve_new_block(struct dnode_of_data *dn)
683 unsigned int ofs_in_node = dn->ofs_in_node;
686 ret = f2fs_reserve_new_blocks(dn, 1);
687 dn->ofs_in_node = ofs_in_node;
691 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
693 bool need_put = dn->inode_page ? false : true;
696 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
700 if (dn->data_blkaddr == NULL_ADDR)
701 err = f2fs_reserve_new_block(dn);
707 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
709 struct extent_info ei = {0,0,0};
710 struct inode *inode = dn->inode;
712 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
713 dn->data_blkaddr = ei.blk + index - ei.fofs;
717 return f2fs_reserve_block(dn, index);
720 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
721 int op_flags, bool for_write)
723 struct address_space *mapping = inode->i_mapping;
724 struct dnode_of_data dn;
726 struct extent_info ei = {0,0,0};
729 page = f2fs_grab_cache_page(mapping, index, for_write);
731 return ERR_PTR(-ENOMEM);
733 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
734 dn.data_blkaddr = ei.blk + index - ei.fofs;
738 set_new_dnode(&dn, inode, NULL, NULL, 0);
739 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
744 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
749 if (PageUptodate(page)) {
755 * A new dentry page is allocated but not able to be written, since its
756 * new inode page couldn't be allocated due to -ENOSPC.
757 * In such the case, its blkaddr can be remained as NEW_ADDR.
758 * see, f2fs_add_link -> f2fs_get_new_data_page ->
759 * f2fs_init_inode_metadata.
761 if (dn.data_blkaddr == NEW_ADDR) {
762 zero_user_segment(page, 0, PAGE_SIZE);
763 if (!PageUptodate(page))
764 SetPageUptodate(page);
769 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
775 f2fs_put_page(page, 1);
779 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
781 struct address_space *mapping = inode->i_mapping;
784 page = find_get_page(mapping, index);
785 if (page && PageUptodate(page))
787 f2fs_put_page(page, 0);
789 page = f2fs_get_read_data_page(inode, index, 0, false);
793 if (PageUptodate(page))
796 wait_on_page_locked(page);
797 if (unlikely(!PageUptodate(page))) {
798 f2fs_put_page(page, 0);
799 return ERR_PTR(-EIO);
805 * If it tries to access a hole, return an error.
806 * Because, the callers, functions in dir.c and GC, should be able to know
807 * whether this page exists or not.
809 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
812 struct address_space *mapping = inode->i_mapping;
815 page = f2fs_get_read_data_page(inode, index, 0, for_write);
819 /* wait for read completion */
821 if (unlikely(page->mapping != mapping)) {
822 f2fs_put_page(page, 1);
825 if (unlikely(!PageUptodate(page))) {
826 f2fs_put_page(page, 1);
827 return ERR_PTR(-EIO);
833 * Caller ensures that this data page is never allocated.
834 * A new zero-filled data page is allocated in the page cache.
836 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
838 * Note that, ipage is set only by make_empty_dir, and if any error occur,
839 * ipage should be released by this function.
841 struct page *f2fs_get_new_data_page(struct inode *inode,
842 struct page *ipage, pgoff_t index, bool new_i_size)
844 struct address_space *mapping = inode->i_mapping;
846 struct dnode_of_data dn;
849 page = f2fs_grab_cache_page(mapping, index, true);
852 * before exiting, we should make sure ipage will be released
853 * if any error occur.
855 f2fs_put_page(ipage, 1);
856 return ERR_PTR(-ENOMEM);
859 set_new_dnode(&dn, inode, ipage, NULL, 0);
860 err = f2fs_reserve_block(&dn, index);
862 f2fs_put_page(page, 1);
868 if (PageUptodate(page))
871 if (dn.data_blkaddr == NEW_ADDR) {
872 zero_user_segment(page, 0, PAGE_SIZE);
873 if (!PageUptodate(page))
874 SetPageUptodate(page);
876 f2fs_put_page(page, 1);
878 /* if ipage exists, blkaddr should be NEW_ADDR */
879 f2fs_bug_on(F2FS_I_SB(inode), ipage);
880 page = f2fs_get_lock_data_page(inode, index, true);
885 if (new_i_size && i_size_read(inode) <
886 ((loff_t)(index + 1) << PAGE_SHIFT))
887 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
891 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
893 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
894 struct f2fs_summary sum;
900 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
903 err = f2fs_get_node_info(sbi, dn->nid, &ni);
907 dn->data_blkaddr = datablock_addr(dn->inode,
908 dn->node_page, dn->ofs_in_node);
909 if (dn->data_blkaddr != NULL_ADDR)
912 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
916 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
917 old_blkaddr = dn->data_blkaddr;
918 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
919 &sum, seg_type, NULL, false);
920 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
921 invalidate_mapping_pages(META_MAPPING(sbi),
922 old_blkaddr, old_blkaddr);
923 f2fs_set_data_blkaddr(dn);
926 * i_size will be updated by direct_IO. Otherwise, we'll get stale
927 * data from unwritten block via dio_read.
932 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
934 struct inode *inode = file_inode(iocb->ki_filp);
935 struct f2fs_map_blocks map;
938 bool direct_io = iocb->ki_flags & IOCB_DIRECT;
940 /* convert inline data for Direct I/O*/
942 err = f2fs_convert_inline_inode(inode);
947 if (direct_io && allow_outplace_dio(inode, iocb, from))
950 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
953 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
954 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
955 if (map.m_len > map.m_lblk)
956 map.m_len -= map.m_lblk;
960 map.m_next_pgofs = NULL;
961 map.m_next_extent = NULL;
962 map.m_seg_type = NO_CHECK_TYPE;
963 map.m_may_create = true;
966 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
967 flag = f2fs_force_buffered_io(inode, iocb, from) ?
968 F2FS_GET_BLOCK_PRE_AIO :
969 F2FS_GET_BLOCK_PRE_DIO;
972 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
973 err = f2fs_convert_inline_inode(inode);
977 if (f2fs_has_inline_data(inode))
980 flag = F2FS_GET_BLOCK_PRE_AIO;
983 err = f2fs_map_blocks(inode, &map, 1, flag);
984 if (map.m_len > 0 && err == -ENOSPC) {
986 set_inode_flag(inode, FI_NO_PREALLOC);
992 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
994 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
996 down_read(&sbi->node_change);
998 up_read(&sbi->node_change);
1003 f2fs_unlock_op(sbi);
1008 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1009 * f2fs_map_blocks structure.
1010 * If original data blocks are allocated, then give them to blockdev.
1012 * a. preallocate requested block addresses
1013 * b. do not use extent cache for better performance
1014 * c. give the block addresses to blockdev
1016 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1017 int create, int flag)
1019 unsigned int maxblocks = map->m_len;
1020 struct dnode_of_data dn;
1021 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1022 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1023 pgoff_t pgofs, end_offset, end;
1024 int err = 0, ofs = 1;
1025 unsigned int ofs_in_node, last_ofs_in_node;
1027 struct extent_info ei = {0,0,0};
1029 unsigned int start_pgofs;
1037 /* it only supports block size == page size */
1038 pgofs = (pgoff_t)map->m_lblk;
1039 end = pgofs + maxblocks;
1041 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1042 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1046 map->m_pblk = ei.blk + pgofs - ei.fofs;
1047 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1048 map->m_flags = F2FS_MAP_MAPPED;
1049 if (map->m_next_extent)
1050 *map->m_next_extent = pgofs + map->m_len;
1052 /* for hardware encryption, but to avoid potential issue in future */
1053 if (flag == F2FS_GET_BLOCK_DIO)
1054 f2fs_wait_on_block_writeback_range(inode,
1055 map->m_pblk, map->m_len);
1060 if (map->m_may_create)
1061 __do_map_lock(sbi, flag, true);
1063 /* When reading holes, we need its node page */
1064 set_new_dnode(&dn, inode, NULL, NULL, 0);
1065 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1067 if (flag == F2FS_GET_BLOCK_BMAP)
1069 if (err == -ENOENT) {
1071 if (map->m_next_pgofs)
1072 *map->m_next_pgofs =
1073 f2fs_get_next_page_offset(&dn, pgofs);
1074 if (map->m_next_extent)
1075 *map->m_next_extent =
1076 f2fs_get_next_page_offset(&dn, pgofs);
1081 start_pgofs = pgofs;
1083 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1084 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1087 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1089 if (__is_valid_data_blkaddr(blkaddr) &&
1090 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1095 if (is_valid_data_blkaddr(sbi, blkaddr)) {
1096 /* use out-place-update for driect IO under LFS mode */
1097 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1098 map->m_may_create) {
1099 err = __allocate_data_block(&dn, map->m_seg_type);
1101 blkaddr = dn.data_blkaddr;
1102 set_inode_flag(inode, FI_APPEND_WRITE);
1107 if (unlikely(f2fs_cp_error(sbi))) {
1111 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1112 if (blkaddr == NULL_ADDR) {
1114 last_ofs_in_node = dn.ofs_in_node;
1117 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1118 flag != F2FS_GET_BLOCK_DIO);
1119 err = __allocate_data_block(&dn,
1122 set_inode_flag(inode, FI_APPEND_WRITE);
1126 map->m_flags |= F2FS_MAP_NEW;
1127 blkaddr = dn.data_blkaddr;
1129 if (flag == F2FS_GET_BLOCK_BMAP) {
1133 if (flag == F2FS_GET_BLOCK_PRECACHE)
1135 if (flag == F2FS_GET_BLOCK_FIEMAP &&
1136 blkaddr == NULL_ADDR) {
1137 if (map->m_next_pgofs)
1138 *map->m_next_pgofs = pgofs + 1;
1141 if (flag != F2FS_GET_BLOCK_FIEMAP) {
1142 /* for defragment case */
1143 if (map->m_next_pgofs)
1144 *map->m_next_pgofs = pgofs + 1;
1150 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1153 if (map->m_len == 0) {
1154 /* preallocated unwritten block should be mapped for fiemap. */
1155 if (blkaddr == NEW_ADDR)
1156 map->m_flags |= F2FS_MAP_UNWRITTEN;
1157 map->m_flags |= F2FS_MAP_MAPPED;
1159 map->m_pblk = blkaddr;
1161 } else if ((map->m_pblk != NEW_ADDR &&
1162 blkaddr == (map->m_pblk + ofs)) ||
1163 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1164 flag == F2FS_GET_BLOCK_PRE_DIO) {
1175 /* preallocate blocks in batch for one dnode page */
1176 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1177 (pgofs == end || dn.ofs_in_node == end_offset)) {
1179 dn.ofs_in_node = ofs_in_node;
1180 err = f2fs_reserve_new_blocks(&dn, prealloc);
1184 map->m_len += dn.ofs_in_node - ofs_in_node;
1185 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1189 dn.ofs_in_node = end_offset;
1194 else if (dn.ofs_in_node < end_offset)
1197 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1198 if (map->m_flags & F2FS_MAP_MAPPED) {
1199 unsigned int ofs = start_pgofs - map->m_lblk;
1201 f2fs_update_extent_cache_range(&dn,
1202 start_pgofs, map->m_pblk + ofs,
1207 f2fs_put_dnode(&dn);
1209 if (map->m_may_create) {
1210 __do_map_lock(sbi, flag, false);
1211 f2fs_balance_fs(sbi, dn.node_changed);
1217 /* for hardware encryption, but to avoid potential issue in future */
1218 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1219 f2fs_wait_on_block_writeback_range(inode,
1220 map->m_pblk, map->m_len);
1222 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1223 if (map->m_flags & F2FS_MAP_MAPPED) {
1224 unsigned int ofs = start_pgofs - map->m_lblk;
1226 f2fs_update_extent_cache_range(&dn,
1227 start_pgofs, map->m_pblk + ofs,
1230 if (map->m_next_extent)
1231 *map->m_next_extent = pgofs + 1;
1233 f2fs_put_dnode(&dn);
1235 if (map->m_may_create) {
1236 __do_map_lock(sbi, flag, false);
1237 f2fs_balance_fs(sbi, dn.node_changed);
1240 trace_f2fs_map_blocks(inode, map, err);
1244 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1246 struct f2fs_map_blocks map;
1250 if (pos + len > i_size_read(inode))
1253 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1254 map.m_next_pgofs = NULL;
1255 map.m_next_extent = NULL;
1256 map.m_seg_type = NO_CHECK_TYPE;
1257 map.m_may_create = false;
1258 last_lblk = F2FS_BLK_ALIGN(pos + len);
1260 while (map.m_lblk < last_lblk) {
1261 map.m_len = last_lblk - map.m_lblk;
1262 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1263 if (err || map.m_len == 0)
1265 map.m_lblk += map.m_len;
1270 static int __get_data_block(struct inode *inode, sector_t iblock,
1271 struct buffer_head *bh, int create, int flag,
1272 pgoff_t *next_pgofs, int seg_type, bool may_write)
1274 struct f2fs_map_blocks map;
1277 map.m_lblk = iblock;
1278 map.m_len = bh->b_size >> inode->i_blkbits;
1279 map.m_next_pgofs = next_pgofs;
1280 map.m_next_extent = NULL;
1281 map.m_seg_type = seg_type;
1282 map.m_may_create = may_write;
1284 err = f2fs_map_blocks(inode, &map, create, flag);
1286 map_bh(bh, inode->i_sb, map.m_pblk);
1287 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1288 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1293 static int get_data_block(struct inode *inode, sector_t iblock,
1294 struct buffer_head *bh_result, int create, int flag,
1295 pgoff_t *next_pgofs)
1297 return __get_data_block(inode, iblock, bh_result, create,
1299 NO_CHECK_TYPE, create);
1302 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1303 struct buffer_head *bh_result, int create)
1305 return __get_data_block(inode, iblock, bh_result, create,
1306 F2FS_GET_BLOCK_DIO, NULL,
1307 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1311 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1312 struct buffer_head *bh_result, int create)
1314 return __get_data_block(inode, iblock, bh_result, create,
1315 F2FS_GET_BLOCK_DIO, NULL,
1316 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1320 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1321 struct buffer_head *bh_result, int create)
1323 /* Block number less than F2FS MAX BLOCKS */
1324 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1327 return __get_data_block(inode, iblock, bh_result, create,
1328 F2FS_GET_BLOCK_BMAP, NULL,
1329 NO_CHECK_TYPE, create);
1332 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1334 return (offset >> inode->i_blkbits);
1337 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1339 return (blk << inode->i_blkbits);
1342 static int f2fs_xattr_fiemap(struct inode *inode,
1343 struct fiemap_extent_info *fieinfo)
1345 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1347 struct node_info ni;
1348 __u64 phys = 0, len;
1350 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1353 if (f2fs_has_inline_xattr(inode)) {
1356 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1357 inode->i_ino, false);
1361 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1363 f2fs_put_page(page, 1);
1367 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1368 offset = offsetof(struct f2fs_inode, i_addr) +
1369 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1370 get_inline_xattr_addrs(inode));
1373 len = inline_xattr_size(inode);
1375 f2fs_put_page(page, 1);
1377 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1380 flags |= FIEMAP_EXTENT_LAST;
1382 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1383 if (err || err == 1)
1388 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1392 err = f2fs_get_node_info(sbi, xnid, &ni);
1394 f2fs_put_page(page, 1);
1398 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1399 len = inode->i_sb->s_blocksize;
1401 f2fs_put_page(page, 1);
1403 flags = FIEMAP_EXTENT_LAST;
1407 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1409 return (err < 0 ? err : 0);
1412 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1415 struct buffer_head map_bh;
1416 sector_t start_blk, last_blk;
1418 u64 logical = 0, phys = 0, size = 0;
1422 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1423 ret = f2fs_precache_extents(inode);
1428 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1434 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1435 ret = f2fs_xattr_fiemap(inode, fieinfo);
1439 if (f2fs_has_inline_data(inode)) {
1440 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1445 if (logical_to_blk(inode, len) == 0)
1446 len = blk_to_logical(inode, 1);
1448 start_blk = logical_to_blk(inode, start);
1449 last_blk = logical_to_blk(inode, start + len - 1);
1452 memset(&map_bh, 0, sizeof(struct buffer_head));
1453 map_bh.b_size = len;
1455 ret = get_data_block(inode, start_blk, &map_bh, 0,
1456 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1461 if (!buffer_mapped(&map_bh)) {
1462 start_blk = next_pgofs;
1464 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1465 F2FS_I_SB(inode)->max_file_blocks))
1468 flags |= FIEMAP_EXTENT_LAST;
1472 if (IS_ENCRYPTED(inode))
1473 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1475 ret = fiemap_fill_next_extent(fieinfo, logical,
1479 if (start_blk > last_blk || ret)
1482 logical = blk_to_logical(inode, start_blk);
1483 phys = blk_to_logical(inode, map_bh.b_blocknr);
1484 size = map_bh.b_size;
1486 if (buffer_unwritten(&map_bh))
1487 flags = FIEMAP_EXTENT_UNWRITTEN;
1489 start_blk += logical_to_blk(inode, size);
1493 if (fatal_signal_pending(current))
1501 inode_unlock(inode);
1506 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1507 * Major change was from block_size == page_size in f2fs by default.
1509 * Note that the aops->readpages() function is ONLY used for read-ahead. If
1510 * this function ever deviates from doing just read-ahead, it should either
1511 * use ->readpage() or do the necessary surgery to decouple ->readpages()
1514 static int f2fs_mpage_readpages(struct address_space *mapping,
1515 struct list_head *pages, struct page *page,
1516 unsigned nr_pages, bool is_readahead)
1518 struct bio *bio = NULL;
1519 sector_t last_block_in_bio = 0;
1520 struct inode *inode = mapping->host;
1521 const unsigned blkbits = inode->i_blkbits;
1522 const unsigned blocksize = 1 << blkbits;
1523 sector_t block_in_file;
1524 sector_t last_block;
1525 sector_t last_block_in_file;
1527 struct f2fs_map_blocks map;
1533 map.m_next_pgofs = NULL;
1534 map.m_next_extent = NULL;
1535 map.m_seg_type = NO_CHECK_TYPE;
1536 map.m_may_create = false;
1538 for (; nr_pages; nr_pages--) {
1540 page = list_last_entry(pages, struct page, lru);
1542 prefetchw(&page->flags);
1543 list_del(&page->lru);
1544 if (add_to_page_cache_lru(page, mapping,
1546 readahead_gfp_mask(mapping)))
1550 block_in_file = (sector_t)page->index;
1551 last_block = block_in_file + nr_pages;
1552 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1554 if (last_block > last_block_in_file)
1555 last_block = last_block_in_file;
1557 /* just zeroing out page which is beyond EOF */
1558 if (block_in_file >= last_block)
1561 * Map blocks using the previous result first.
1563 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1564 block_in_file > map.m_lblk &&
1565 block_in_file < (map.m_lblk + map.m_len))
1569 * Then do more f2fs_map_blocks() calls until we are
1570 * done with this page.
1572 map.m_lblk = block_in_file;
1573 map.m_len = last_block - block_in_file;
1575 if (f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT))
1576 goto set_error_page;
1578 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1579 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1580 SetPageMappedToDisk(page);
1582 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1583 SetPageUptodate(page);
1587 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1589 goto set_error_page;
1592 zero_user_segment(page, 0, PAGE_SIZE);
1593 if (!PageUptodate(page))
1594 SetPageUptodate(page);
1600 * This page will go to BIO. Do we need to send this
1603 if (bio && (last_block_in_bio != block_nr - 1 ||
1604 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1606 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1610 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1611 is_readahead ? REQ_RAHEAD : 0);
1614 goto set_error_page;
1619 * If the page is under writeback, we need to wait for
1620 * its completion to see the correct decrypted data.
1622 f2fs_wait_on_block_writeback(inode, block_nr);
1624 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1625 goto submit_and_realloc;
1627 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1628 ClearPageError(page);
1629 last_block_in_bio = block_nr;
1633 zero_user_segment(page, 0, PAGE_SIZE);
1638 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1646 BUG_ON(pages && !list_empty(pages));
1648 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1652 static int f2fs_read_data_page(struct file *file, struct page *page)
1654 struct inode *inode = page->mapping->host;
1657 trace_f2fs_readpage(page, DATA);
1659 /* If the file has inline data, try to read it directly */
1660 if (f2fs_has_inline_data(inode))
1661 ret = f2fs_read_inline_data(inode, page);
1663 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1667 static int f2fs_read_data_pages(struct file *file,
1668 struct address_space *mapping,
1669 struct list_head *pages, unsigned nr_pages)
1671 struct inode *inode = mapping->host;
1672 struct page *page = list_last_entry(pages, struct page, lru);
1674 trace_f2fs_readpages(inode, page, nr_pages);
1676 /* If the file has inline data, skip readpages */
1677 if (f2fs_has_inline_data(inode))
1680 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1683 static int encrypt_one_page(struct f2fs_io_info *fio)
1685 struct inode *inode = fio->page->mapping->host;
1687 gfp_t gfp_flags = GFP_NOFS;
1689 if (!f2fs_encrypted_file(inode))
1692 /* wait for GCed page writeback via META_MAPPING */
1693 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1696 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1697 PAGE_SIZE, 0, fio->page->index, gfp_flags);
1698 if (IS_ERR(fio->encrypted_page)) {
1699 /* flush pending IOs and wait for a while in the ENOMEM case */
1700 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1701 f2fs_flush_merged_writes(fio->sbi);
1702 congestion_wait(BLK_RW_ASYNC, HZ/50);
1703 gfp_flags |= __GFP_NOFAIL;
1706 return PTR_ERR(fio->encrypted_page);
1709 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1711 if (PageUptodate(mpage))
1712 memcpy(page_address(mpage),
1713 page_address(fio->encrypted_page), PAGE_SIZE);
1714 f2fs_put_page(mpage, 1);
1719 static inline bool check_inplace_update_policy(struct inode *inode,
1720 struct f2fs_io_info *fio)
1722 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1723 unsigned int policy = SM_I(sbi)->ipu_policy;
1725 if (policy & (0x1 << F2FS_IPU_FORCE))
1727 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1729 if (policy & (0x1 << F2FS_IPU_UTIL) &&
1730 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1732 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1733 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1737 * IPU for rewrite async pages
1739 if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1740 fio && fio->op == REQ_OP_WRITE &&
1741 !(fio->op_flags & REQ_SYNC) &&
1742 !IS_ENCRYPTED(inode))
1745 /* this is only set during fdatasync */
1746 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1747 is_inode_flag_set(inode, FI_NEED_IPU))
1750 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1751 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1757 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1759 if (f2fs_is_pinned_file(inode))
1762 /* if this is cold file, we should overwrite to avoid fragmentation */
1763 if (file_is_cold(inode))
1766 return check_inplace_update_policy(inode, fio);
1769 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1771 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1773 if (test_opt(sbi, LFS))
1775 if (S_ISDIR(inode->i_mode))
1777 if (IS_NOQUOTA(inode))
1779 if (f2fs_is_atomic_file(inode))
1782 if (is_cold_data(fio->page))
1784 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1786 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1787 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1793 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1795 struct inode *inode = fio->page->mapping->host;
1797 if (f2fs_should_update_outplace(inode, fio))
1800 return f2fs_should_update_inplace(inode, fio);
1803 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1805 struct page *page = fio->page;
1806 struct inode *inode = page->mapping->host;
1807 struct dnode_of_data dn;
1808 struct extent_info ei = {0,0,0};
1809 struct node_info ni;
1810 bool ipu_force = false;
1813 set_new_dnode(&dn, inode, NULL, NULL, 0);
1814 if (need_inplace_update(fio) &&
1815 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1816 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1818 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1823 fio->need_lock = LOCK_DONE;
1827 /* Deadlock due to between page->lock and f2fs_lock_op */
1828 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1831 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1835 fio->old_blkaddr = dn.data_blkaddr;
1837 /* This page is already truncated */
1838 if (fio->old_blkaddr == NULL_ADDR) {
1839 ClearPageUptodate(page);
1840 clear_cold_data(page);
1844 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1845 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1851 * If current allocation needs SSR,
1852 * it had better in-place writes for updated data.
1854 if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1855 need_inplace_update(fio))) {
1856 err = encrypt_one_page(fio);
1860 set_page_writeback(page);
1861 ClearPageError(page);
1862 f2fs_put_dnode(&dn);
1863 if (fio->need_lock == LOCK_REQ)
1864 f2fs_unlock_op(fio->sbi);
1865 err = f2fs_inplace_write_data(fio);
1867 if (f2fs_encrypted_file(inode))
1868 fscrypt_pullback_bio_page(&fio->encrypted_page,
1870 if (PageWriteback(page))
1871 end_page_writeback(page);
1873 trace_f2fs_do_write_data_page(fio->page, IPU);
1874 set_inode_flag(inode, FI_UPDATE_WRITE);
1878 if (fio->need_lock == LOCK_RETRY) {
1879 if (!f2fs_trylock_op(fio->sbi)) {
1883 fio->need_lock = LOCK_REQ;
1886 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1890 fio->version = ni.version;
1892 err = encrypt_one_page(fio);
1896 set_page_writeback(page);
1897 ClearPageError(page);
1899 /* LFS mode write path */
1900 f2fs_outplace_write_data(&dn, fio);
1901 trace_f2fs_do_write_data_page(page, OPU);
1902 set_inode_flag(inode, FI_APPEND_WRITE);
1903 if (page->index == 0)
1904 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1906 f2fs_put_dnode(&dn);
1908 if (fio->need_lock == LOCK_REQ)
1909 f2fs_unlock_op(fio->sbi);
1913 static int __write_data_page(struct page *page, bool *submitted,
1914 struct writeback_control *wbc,
1915 enum iostat_type io_type)
1917 struct inode *inode = page->mapping->host;
1918 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1919 loff_t i_size = i_size_read(inode);
1920 const pgoff_t end_index = ((unsigned long long) i_size)
1922 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1923 unsigned offset = 0;
1924 bool need_balance_fs = false;
1926 struct f2fs_io_info fio = {
1928 .ino = inode->i_ino,
1931 .op_flags = wbc_to_write_flags(wbc),
1932 .old_blkaddr = NULL_ADDR,
1934 .encrypted_page = NULL,
1936 .need_lock = LOCK_RETRY,
1941 trace_f2fs_writepage(page, DATA);
1943 /* we should bypass data pages to proceed the kworkder jobs */
1944 if (unlikely(f2fs_cp_error(sbi))) {
1945 mapping_set_error(page->mapping, -EIO);
1947 * don't drop any dirty dentry pages for keeping lastest
1948 * directory structure.
1950 if (S_ISDIR(inode->i_mode))
1955 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1958 if (page->index < end_index)
1962 * If the offset is out-of-range of file size,
1963 * this page does not have to be written to disk.
1965 offset = i_size & (PAGE_SIZE - 1);
1966 if ((page->index >= end_index + 1) || !offset)
1969 zero_user_segment(page, offset, PAGE_SIZE);
1971 if (f2fs_is_drop_cache(inode))
1973 /* we should not write 0'th page having journal header */
1974 if (f2fs_is_volatile_file(inode) && (!page->index ||
1975 (!wbc->for_reclaim &&
1976 f2fs_available_free_memory(sbi, BASE_CHECK))))
1979 /* Dentry blocks are controlled by checkpoint */
1980 if (S_ISDIR(inode->i_mode)) {
1981 fio.need_lock = LOCK_DONE;
1982 err = f2fs_do_write_data_page(&fio);
1986 if (!wbc->for_reclaim)
1987 need_balance_fs = true;
1988 else if (has_not_enough_free_secs(sbi, 0, 0))
1991 set_inode_flag(inode, FI_HOT_DATA);
1994 if (f2fs_has_inline_data(inode)) {
1995 err = f2fs_write_inline_data(inode, page);
2000 if (err == -EAGAIN) {
2001 err = f2fs_do_write_data_page(&fio);
2002 if (err == -EAGAIN) {
2003 fio.need_lock = LOCK_REQ;
2004 err = f2fs_do_write_data_page(&fio);
2009 file_set_keep_isize(inode);
2011 down_write(&F2FS_I(inode)->i_sem);
2012 if (F2FS_I(inode)->last_disk_size < psize)
2013 F2FS_I(inode)->last_disk_size = psize;
2014 up_write(&F2FS_I(inode)->i_sem);
2018 if (err && err != -ENOENT)
2022 inode_dec_dirty_pages(inode);
2024 ClearPageUptodate(page);
2025 clear_cold_data(page);
2028 if (wbc->for_reclaim) {
2029 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2030 clear_inode_flag(inode, FI_HOT_DATA);
2031 f2fs_remove_dirty_inode(inode);
2036 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode))
2037 f2fs_balance_fs(sbi, need_balance_fs);
2039 if (unlikely(f2fs_cp_error(sbi))) {
2040 f2fs_submit_merged_write(sbi, DATA);
2045 *submitted = fio.submitted;
2050 redirty_page_for_writepage(wbc, page);
2052 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2053 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2054 * file_write_and_wait_range() will see EIO error, which is critical
2055 * to return value of fsync() followed by atomic_write failure to user.
2057 if (!err || wbc->for_reclaim)
2058 return AOP_WRITEPAGE_ACTIVATE;
2063 static int f2fs_write_data_page(struct page *page,
2064 struct writeback_control *wbc)
2066 return __write_data_page(page, NULL, wbc, FS_DATA_IO);
2070 * This function was copied from write_cche_pages from mm/page-writeback.c.
2071 * The major change is making write step of cold data page separately from
2072 * warm/hot data page.
2074 static int f2fs_write_cache_pages(struct address_space *mapping,
2075 struct writeback_control *wbc,
2076 enum iostat_type io_type)
2080 struct pagevec pvec;
2081 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2083 pgoff_t uninitialized_var(writeback_index);
2085 pgoff_t end; /* Inclusive */
2088 int range_whole = 0;
2092 pagevec_init(&pvec);
2094 if (get_dirty_pages(mapping->host) <=
2095 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2096 set_inode_flag(mapping->host, FI_HOT_DATA);
2098 clear_inode_flag(mapping->host, FI_HOT_DATA);
2100 if (wbc->range_cyclic) {
2101 writeback_index = mapping->writeback_index; /* prev offset */
2102 index = writeback_index;
2109 index = wbc->range_start >> PAGE_SHIFT;
2110 end = wbc->range_end >> PAGE_SHIFT;
2111 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2113 cycled = 1; /* ignore range_cyclic tests */
2115 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2116 tag = PAGECACHE_TAG_TOWRITE;
2118 tag = PAGECACHE_TAG_DIRTY;
2120 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2121 tag_pages_for_writeback(mapping, index, end);
2123 while (!done && (index <= end)) {
2126 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2131 for (i = 0; i < nr_pages; i++) {
2132 struct page *page = pvec.pages[i];
2133 bool submitted = false;
2135 /* give a priority to WB_SYNC threads */
2136 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2137 wbc->sync_mode == WB_SYNC_NONE) {
2142 done_index = page->index;
2146 if (unlikely(page->mapping != mapping)) {
2152 if (!PageDirty(page)) {
2153 /* someone wrote it for us */
2154 goto continue_unlock;
2157 if (PageWriteback(page)) {
2158 if (wbc->sync_mode != WB_SYNC_NONE)
2159 f2fs_wait_on_page_writeback(page,
2162 goto continue_unlock;
2165 if (!clear_page_dirty_for_io(page))
2166 goto continue_unlock;
2168 ret = __write_data_page(page, &submitted, wbc, io_type);
2169 if (unlikely(ret)) {
2171 * keep nr_to_write, since vfs uses this to
2172 * get # of written pages.
2174 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2178 } else if (ret == -EAGAIN) {
2180 if (wbc->sync_mode == WB_SYNC_ALL) {
2182 congestion_wait(BLK_RW_ASYNC,
2188 done_index = page->index + 1;
2191 } else if (submitted) {
2195 if (--wbc->nr_to_write <= 0 &&
2196 wbc->sync_mode == WB_SYNC_NONE) {
2201 pagevec_release(&pvec);
2205 if (!cycled && !done) {
2208 end = writeback_index - 1;
2211 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2212 mapping->writeback_index = done_index;
2215 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2221 static inline bool __should_serialize_io(struct inode *inode,
2222 struct writeback_control *wbc)
2224 if (!S_ISREG(inode->i_mode))
2226 if (IS_NOQUOTA(inode))
2228 if (wbc->sync_mode != WB_SYNC_ALL)
2230 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2235 static int __f2fs_write_data_pages(struct address_space *mapping,
2236 struct writeback_control *wbc,
2237 enum iostat_type io_type)
2239 struct inode *inode = mapping->host;
2240 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2241 struct blk_plug plug;
2243 bool locked = false;
2245 /* deal with chardevs and other special file */
2246 if (!mapping->a_ops->writepage)
2249 /* skip writing if there is no dirty page in this inode */
2250 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2253 /* during POR, we don't need to trigger writepage at all. */
2254 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2257 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2258 wbc->sync_mode == WB_SYNC_NONE &&
2259 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2260 f2fs_available_free_memory(sbi, DIRTY_DENTS))
2263 /* skip writing during file defragment */
2264 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2267 trace_f2fs_writepages(mapping->host, wbc, DATA);
2269 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2270 if (wbc->sync_mode == WB_SYNC_ALL)
2271 atomic_inc(&sbi->wb_sync_req[DATA]);
2272 else if (atomic_read(&sbi->wb_sync_req[DATA]))
2275 if (__should_serialize_io(inode, wbc)) {
2276 mutex_lock(&sbi->writepages);
2280 blk_start_plug(&plug);
2281 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2282 blk_finish_plug(&plug);
2285 mutex_unlock(&sbi->writepages);
2287 if (wbc->sync_mode == WB_SYNC_ALL)
2288 atomic_dec(&sbi->wb_sync_req[DATA]);
2290 * if some pages were truncated, we cannot guarantee its mapping->host
2291 * to detect pending bios.
2294 f2fs_remove_dirty_inode(inode);
2298 wbc->pages_skipped += get_dirty_pages(inode);
2299 trace_f2fs_writepages(mapping->host, wbc, DATA);
2303 static int f2fs_write_data_pages(struct address_space *mapping,
2304 struct writeback_control *wbc)
2306 struct inode *inode = mapping->host;
2308 return __f2fs_write_data_pages(mapping, wbc,
2309 F2FS_I(inode)->cp_task == current ?
2310 FS_CP_DATA_IO : FS_DATA_IO);
2313 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2315 struct inode *inode = mapping->host;
2316 loff_t i_size = i_size_read(inode);
2319 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2320 down_write(&F2FS_I(inode)->i_mmap_sem);
2322 truncate_pagecache(inode, i_size);
2323 if (!IS_NOQUOTA(inode))
2324 f2fs_truncate_blocks(inode, i_size, true);
2326 up_write(&F2FS_I(inode)->i_mmap_sem);
2327 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2331 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2332 struct page *page, loff_t pos, unsigned len,
2333 block_t *blk_addr, bool *node_changed)
2335 struct inode *inode = page->mapping->host;
2336 pgoff_t index = page->index;
2337 struct dnode_of_data dn;
2339 bool locked = false;
2340 struct extent_info ei = {0,0,0};
2345 * we already allocated all the blocks, so we don't need to get
2346 * the block addresses when there is no need to fill the page.
2348 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2349 !is_inode_flag_set(inode, FI_NO_PREALLOC))
2352 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2353 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2354 flag = F2FS_GET_BLOCK_DEFAULT;
2356 flag = F2FS_GET_BLOCK_PRE_AIO;
2358 if (f2fs_has_inline_data(inode) ||
2359 (pos & PAGE_MASK) >= i_size_read(inode)) {
2360 __do_map_lock(sbi, flag, true);
2364 /* check inline_data */
2365 ipage = f2fs_get_node_page(sbi, inode->i_ino);
2366 if (IS_ERR(ipage)) {
2367 err = PTR_ERR(ipage);
2371 set_new_dnode(&dn, inode, ipage, ipage, 0);
2373 if (f2fs_has_inline_data(inode)) {
2374 if (pos + len <= MAX_INLINE_DATA(inode)) {
2375 f2fs_do_read_inline_data(page, ipage);
2376 set_inode_flag(inode, FI_DATA_EXIST);
2378 set_inline_node(ipage);
2380 err = f2fs_convert_inline_page(&dn, page);
2383 if (dn.data_blkaddr == NULL_ADDR)
2384 err = f2fs_get_block(&dn, index);
2386 } else if (locked) {
2387 err = f2fs_get_block(&dn, index);
2389 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2390 dn.data_blkaddr = ei.blk + index - ei.fofs;
2393 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2394 if (err || dn.data_blkaddr == NULL_ADDR) {
2395 f2fs_put_dnode(&dn);
2396 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2398 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2405 /* convert_inline_page can make node_changed */
2406 *blk_addr = dn.data_blkaddr;
2407 *node_changed = dn.node_changed;
2409 f2fs_put_dnode(&dn);
2412 __do_map_lock(sbi, flag, false);
2416 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2417 loff_t pos, unsigned len, unsigned flags,
2418 struct page **pagep, void **fsdata)
2420 struct inode *inode = mapping->host;
2421 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2422 struct page *page = NULL;
2423 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2424 bool need_balance = false, drop_atomic = false;
2425 block_t blkaddr = NULL_ADDR;
2428 trace_f2fs_write_begin(inode, pos, len, flags);
2430 err = f2fs_is_checkpoint_ready(sbi);
2434 if ((f2fs_is_atomic_file(inode) &&
2435 !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2436 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2443 * We should check this at this moment to avoid deadlock on inode page
2444 * and #0 page. The locking rule for inline_data conversion should be:
2445 * lock_page(page #0) -> lock_page(inode_page)
2448 err = f2fs_convert_inline_inode(inode);
2454 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2455 * wait_for_stable_page. Will wait that below with our IO control.
2457 page = f2fs_pagecache_get_page(mapping, index,
2458 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2466 err = prepare_write_begin(sbi, page, pos, len,
2467 &blkaddr, &need_balance);
2471 if (need_balance && !IS_NOQUOTA(inode) &&
2472 has_not_enough_free_secs(sbi, 0, 0)) {
2474 f2fs_balance_fs(sbi, true);
2476 if (page->mapping != mapping) {
2477 /* The page got truncated from under us */
2478 f2fs_put_page(page, 1);
2483 f2fs_wait_on_page_writeback(page, DATA, false, true);
2485 if (len == PAGE_SIZE || PageUptodate(page))
2488 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2489 zero_user_segment(page, len, PAGE_SIZE);
2493 if (blkaddr == NEW_ADDR) {
2494 zero_user_segment(page, 0, PAGE_SIZE);
2495 SetPageUptodate(page);
2497 err = f2fs_submit_page_read(inode, page, blkaddr);
2502 if (unlikely(page->mapping != mapping)) {
2503 f2fs_put_page(page, 1);
2506 if (unlikely(!PageUptodate(page))) {
2514 f2fs_put_page(page, 1);
2515 f2fs_write_failed(mapping, pos + len);
2517 f2fs_drop_inmem_pages_all(sbi, false);
2521 static int f2fs_write_end(struct file *file,
2522 struct address_space *mapping,
2523 loff_t pos, unsigned len, unsigned copied,
2524 struct page *page, void *fsdata)
2526 struct inode *inode = page->mapping->host;
2528 trace_f2fs_write_end(inode, pos, len, copied);
2531 * This should be come from len == PAGE_SIZE, and we expect copied
2532 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2533 * let generic_perform_write() try to copy data again through copied=0.
2535 if (!PageUptodate(page)) {
2536 if (unlikely(copied != len))
2539 SetPageUptodate(page);
2544 set_page_dirty(page);
2546 if (pos + copied > i_size_read(inode))
2547 f2fs_i_size_write(inode, pos + copied);
2549 f2fs_put_page(page, 1);
2550 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2554 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2557 unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2558 unsigned blkbits = i_blkbits;
2559 unsigned blocksize_mask = (1 << blkbits) - 1;
2560 unsigned long align = offset | iov_iter_alignment(iter);
2561 struct block_device *bdev = inode->i_sb->s_bdev;
2563 if (align & blocksize_mask) {
2565 blkbits = blksize_bits(bdev_logical_block_size(bdev));
2566 blocksize_mask = (1 << blkbits) - 1;
2567 if (align & blocksize_mask)
2574 static void f2fs_dio_end_io(struct bio *bio)
2576 struct f2fs_private_dio *dio = bio->bi_private;
2578 dec_page_count(F2FS_I_SB(dio->inode),
2579 dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2581 bio->bi_private = dio->orig_private;
2582 bio->bi_end_io = dio->orig_end_io;
2589 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2592 struct f2fs_private_dio *dio;
2593 bool write = (bio_op(bio) == REQ_OP_WRITE);
2595 dio = f2fs_kzalloc(F2FS_I_SB(inode),
2596 sizeof(struct f2fs_private_dio), GFP_NOFS);
2601 dio->orig_end_io = bio->bi_end_io;
2602 dio->orig_private = bio->bi_private;
2605 bio->bi_end_io = f2fs_dio_end_io;
2606 bio->bi_private = dio;
2608 inc_page_count(F2FS_I_SB(inode),
2609 write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2614 bio->bi_status = BLK_STS_IOERR;
2618 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2620 struct address_space *mapping = iocb->ki_filp->f_mapping;
2621 struct inode *inode = mapping->host;
2622 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2623 struct f2fs_inode_info *fi = F2FS_I(inode);
2624 size_t count = iov_iter_count(iter);
2625 loff_t offset = iocb->ki_pos;
2626 int rw = iov_iter_rw(iter);
2628 enum rw_hint hint = iocb->ki_hint;
2629 int whint_mode = F2FS_OPTION(sbi).whint_mode;
2632 err = check_direct_IO(inode, iter, offset);
2634 return err < 0 ? err : 0;
2636 if (f2fs_force_buffered_io(inode, iocb, iter))
2639 do_opu = allow_outplace_dio(inode, iocb, iter);
2641 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2643 if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2644 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2646 if (iocb->ki_flags & IOCB_NOWAIT) {
2647 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2648 iocb->ki_hint = hint;
2652 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2653 up_read(&fi->i_gc_rwsem[rw]);
2654 iocb->ki_hint = hint;
2659 down_read(&fi->i_gc_rwsem[rw]);
2661 down_read(&fi->i_gc_rwsem[READ]);
2664 err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2665 iter, rw == WRITE ? get_data_block_dio_write :
2666 get_data_block_dio, NULL, f2fs_dio_submit_bio,
2667 DIO_LOCKING | DIO_SKIP_HOLES);
2670 up_read(&fi->i_gc_rwsem[READ]);
2672 up_read(&fi->i_gc_rwsem[rw]);
2675 if (whint_mode == WHINT_MODE_OFF)
2676 iocb->ki_hint = hint;
2678 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2681 set_inode_flag(inode, FI_UPDATE_WRITE);
2682 } else if (err < 0) {
2683 f2fs_write_failed(mapping, offset + count);
2688 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2693 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2694 unsigned int length)
2696 struct inode *inode = page->mapping->host;
2697 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2699 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2700 (offset % PAGE_SIZE || length != PAGE_SIZE))
2703 if (PageDirty(page)) {
2704 if (inode->i_ino == F2FS_META_INO(sbi)) {
2705 dec_page_count(sbi, F2FS_DIRTY_META);
2706 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2707 dec_page_count(sbi, F2FS_DIRTY_NODES);
2709 inode_dec_dirty_pages(inode);
2710 f2fs_remove_dirty_inode(inode);
2714 clear_cold_data(page);
2716 if (IS_ATOMIC_WRITTEN_PAGE(page))
2717 return f2fs_drop_inmem_page(inode, page);
2719 f2fs_clear_page_private(page);
2722 int f2fs_release_page(struct page *page, gfp_t wait)
2724 /* If this is dirty page, keep PagePrivate */
2725 if (PageDirty(page))
2728 /* This is atomic written page, keep Private */
2729 if (IS_ATOMIC_WRITTEN_PAGE(page))
2732 clear_cold_data(page);
2733 f2fs_clear_page_private(page);
2737 static int f2fs_set_data_page_dirty(struct page *page)
2739 struct address_space *mapping = page->mapping;
2740 struct inode *inode = mapping->host;
2742 trace_f2fs_set_page_dirty(page, DATA);
2744 if (!PageUptodate(page))
2745 SetPageUptodate(page);
2747 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2748 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2749 f2fs_register_inmem_page(inode, page);
2753 * Previously, this page has been registered, we just
2759 if (!PageDirty(page)) {
2760 __set_page_dirty_nobuffers(page);
2761 f2fs_update_dirty_page(inode, page);
2767 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2769 struct inode *inode = mapping->host;
2771 if (f2fs_has_inline_data(inode))
2774 /* make sure allocating whole blocks */
2775 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2776 filemap_write_and_wait(mapping);
2778 return generic_block_bmap(mapping, block, get_data_block_bmap);
2781 #ifdef CONFIG_MIGRATION
2782 #include <linux/migrate.h>
2784 int f2fs_migrate_page(struct address_space *mapping,
2785 struct page *newpage, struct page *page, enum migrate_mode mode)
2787 int rc, extra_count;
2788 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2789 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2791 BUG_ON(PageWriteback(page));
2793 /* migrating an atomic written page is safe with the inmem_lock hold */
2794 if (atomic_written) {
2795 if (mode != MIGRATE_SYNC)
2797 if (!mutex_trylock(&fi->inmem_lock))
2801 /* one extra reference was held for atomic_write page */
2802 extra_count = atomic_written ? 1 : 0;
2803 rc = migrate_page_move_mapping(mapping, newpage,
2804 page, mode, extra_count);
2805 if (rc != MIGRATEPAGE_SUCCESS) {
2807 mutex_unlock(&fi->inmem_lock);
2811 if (atomic_written) {
2812 struct inmem_pages *cur;
2813 list_for_each_entry(cur, &fi->inmem_pages, list)
2814 if (cur->page == page) {
2815 cur->page = newpage;
2818 mutex_unlock(&fi->inmem_lock);
2823 if (PagePrivate(page)) {
2824 f2fs_set_page_private(newpage, page_private(page));
2825 f2fs_clear_page_private(page);
2828 if (mode != MIGRATE_SYNC_NO_COPY)
2829 migrate_page_copy(newpage, page);
2831 migrate_page_states(newpage, page);
2833 return MIGRATEPAGE_SUCCESS;
2837 const struct address_space_operations f2fs_dblock_aops = {
2838 .readpage = f2fs_read_data_page,
2839 .readpages = f2fs_read_data_pages,
2840 .writepage = f2fs_write_data_page,
2841 .writepages = f2fs_write_data_pages,
2842 .write_begin = f2fs_write_begin,
2843 .write_end = f2fs_write_end,
2844 .set_page_dirty = f2fs_set_data_page_dirty,
2845 .invalidatepage = f2fs_invalidate_page,
2846 .releasepage = f2fs_release_page,
2847 .direct_IO = f2fs_direct_IO,
2849 #ifdef CONFIG_MIGRATION
2850 .migratepage = f2fs_migrate_page,
2854 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2856 struct address_space *mapping = page_mapping(page);
2857 unsigned long flags;
2859 xa_lock_irqsave(&mapping->i_pages, flags);
2860 __xa_clear_mark(&mapping->i_pages, page_index(page),
2861 PAGECACHE_TAG_DIRTY);
2862 xa_unlock_irqrestore(&mapping->i_pages, flags);
2865 int __init f2fs_init_post_read_processing(void)
2867 bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2868 if (!bio_post_read_ctx_cache)
2870 bio_post_read_ctx_pool =
2871 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2872 bio_post_read_ctx_cache);
2873 if (!bio_post_read_ctx_pool)
2874 goto fail_free_cache;
2878 kmem_cache_destroy(bio_post_read_ctx_cache);
2883 void __exit f2fs_destroy_post_read_processing(void)
2885 mempool_destroy(bio_post_read_ctx_pool);
2886 kmem_cache_destroy(bio_post_read_ctx_cache);