data.c revision adf9ea89
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * fs/f2fs/data.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 *             http://www.samsung.com/
7 */
8#include <linux/fs.h>
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/blk-crypto.h>
18#include <linux/swap.h>
19#include <linux/prefetch.h>
20#include <linux/uio.h>
21#include <linux/cleancache.h>
22#include <linux/sched/signal.h>
23#include <linux/fiemap.h>
24
25#include "f2fs.h"
26#include "node.h"
27#include "segment.h"
28#include "iostat.h"
29#include <trace/events/f2fs.h>
30
31#define NUM_PREALLOC_POST_READ_CTXS	128
32
33static struct kmem_cache *bio_post_read_ctx_cache;
34static struct kmem_cache *bio_entry_slab;
35static mempool_t *bio_post_read_ctx_pool;
36static struct bio_set f2fs_bioset;
37
38#define	F2FS_BIO_POOL_SIZE	NR_CURSEG_TYPE
39
40int __init f2fs_init_bioset(void)
41{
42	if (bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
43					0, BIOSET_NEED_BVECS))
44		return -ENOMEM;
45	return 0;
46}
47
48void f2fs_destroy_bioset(void)
49{
50	bioset_exit(&f2fs_bioset);
51}
52
53static bool __is_cp_guaranteed(struct page *page)
54{
55	struct address_space *mapping = page->mapping;
56	struct inode *inode;
57	struct f2fs_sb_info *sbi;
58
59	if (!mapping)
60		return false;
61
62	inode = mapping->host;
63	sbi = F2FS_I_SB(inode);
64
65	if (inode->i_ino == F2FS_META_INO(sbi) ||
66			inode->i_ino == F2FS_NODE_INO(sbi) ||
67			S_ISDIR(inode->i_mode))
68		return true;
69
70	if (f2fs_is_compressed_page(page))
71		return false;
72	if ((S_ISREG(inode->i_mode) &&
73			(f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
74			page_private_gcing(page))
75		return true;
76	return false;
77}
78
79static enum count_type __read_io_type(struct page *page)
80{
81	struct address_space *mapping = page_file_mapping(page);
82
83	if (mapping) {
84		struct inode *inode = mapping->host;
85		struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
86
87		if (inode->i_ino == F2FS_META_INO(sbi))
88			return F2FS_RD_META;
89
90		if (inode->i_ino == F2FS_NODE_INO(sbi))
91			return F2FS_RD_NODE;
92	}
93	return F2FS_RD_DATA;
94}
95
96/* postprocessing steps for read bios */
97enum bio_post_read_step {
98#ifdef CONFIG_FS_ENCRYPTION
99	STEP_DECRYPT	= 1 << 0,
100#else
101	STEP_DECRYPT	= 0,	/* compile out the decryption-related code */
102#endif
103#ifdef CONFIG_F2FS_FS_COMPRESSION
104	STEP_DECOMPRESS	= 1 << 1,
105#else
106	STEP_DECOMPRESS	= 0,	/* compile out the decompression-related code */
107#endif
108#ifdef CONFIG_FS_VERITY
109	STEP_VERITY	= 1 << 2,
110#else
111	STEP_VERITY	= 0,	/* compile out the verity-related code */
112#endif
113};
114
115struct bio_post_read_ctx {
116	struct bio *bio;
117	struct f2fs_sb_info *sbi;
118	struct work_struct work;
119	unsigned int enabled_steps;
120	block_t fs_blkaddr;
121};
122
123static void f2fs_finish_read_bio(struct bio *bio)
124{
125	struct bio_vec *bv;
126	struct bvec_iter_all iter_all;
127
128	/*
129	 * Update and unlock the bio's pagecache pages, and put the
130	 * decompression context for any compressed pages.
131	 */
132	bio_for_each_segment_all(bv, bio, iter_all) {
133		struct page *page = bv->bv_page;
134
135		if (f2fs_is_compressed_page(page)) {
136			if (bio->bi_status)
137				f2fs_end_read_compressed_page(page, true, 0);
138			f2fs_put_page_dic(page);
139			continue;
140		}
141
142		/* PG_error was set if decryption or verity failed. */
143		if (bio->bi_status || PageError(page)) {
144			ClearPageUptodate(page);
145			/* will re-read again later */
146			ClearPageError(page);
147		} else {
148			SetPageUptodate(page);
149		}
150		dec_page_count(F2FS_P_SB(page), __read_io_type(page));
151		unlock_page(page);
152	}
153
154	if (bio->bi_private)
155		mempool_free(bio->bi_private, bio_post_read_ctx_pool);
156	bio_put(bio);
157}
158
159static void f2fs_verify_bio(struct work_struct *work)
160{
161	struct bio_post_read_ctx *ctx =
162		container_of(work, struct bio_post_read_ctx, work);
163	struct bio *bio = ctx->bio;
164	bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
165
166	/*
167	 * fsverity_verify_bio() may call readpages() again, and while verity
168	 * will be disabled for this, decryption and/or decompression may still
169	 * be needed, resulting in another bio_post_read_ctx being allocated.
170	 * So to prevent deadlocks we need to release the current ctx to the
171	 * mempool first.  This assumes that verity is the last post-read step.
172	 */
173	mempool_free(ctx, bio_post_read_ctx_pool);
174	bio->bi_private = NULL;
175
176	/*
177	 * Verify the bio's pages with fs-verity.  Exclude compressed pages,
178	 * as those were handled separately by f2fs_end_read_compressed_page().
179	 */
180	if (may_have_compressed_pages) {
181		struct bio_vec *bv;
182		struct bvec_iter_all iter_all;
183
184		bio_for_each_segment_all(bv, bio, iter_all) {
185			struct page *page = bv->bv_page;
186
187			if (!f2fs_is_compressed_page(page) &&
188			    !PageError(page) && !fsverity_verify_page(page))
189				SetPageError(page);
190		}
191	} else {
192		fsverity_verify_bio(bio);
193	}
194
195	f2fs_finish_read_bio(bio);
196}
197
198/*
199 * If the bio's data needs to be verified with fs-verity, then enqueue the
200 * verity work for the bio.  Otherwise finish the bio now.
201 *
202 * Note that to avoid deadlocks, the verity work can't be done on the
203 * decryption/decompression workqueue.  This is because verifying the data pages
204 * can involve reading verity metadata pages from the file, and these verity
205 * metadata pages may be encrypted and/or compressed.
206 */
207static void f2fs_verify_and_finish_bio(struct bio *bio)
208{
209	struct bio_post_read_ctx *ctx = bio->bi_private;
210
211	if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
212		INIT_WORK(&ctx->work, f2fs_verify_bio);
213		fsverity_enqueue_verify_work(&ctx->work);
214	} else {
215		f2fs_finish_read_bio(bio);
216	}
217}
218
219/*
220 * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
221 * remaining page was read by @ctx->bio.
222 *
223 * Note that a bio may span clusters (even a mix of compressed and uncompressed
224 * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
225 * that the bio includes at least one compressed page.  The actual decompression
226 * is done on a per-cluster basis, not a per-bio basis.
227 */
228static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx)
229{
230	struct bio_vec *bv;
231	struct bvec_iter_all iter_all;
232	bool all_compressed = true;
233	block_t blkaddr = ctx->fs_blkaddr;
234
235	bio_for_each_segment_all(bv, ctx->bio, iter_all) {
236		struct page *page = bv->bv_page;
237
238		/* PG_error was set if decryption failed. */
239		if (f2fs_is_compressed_page(page))
240			f2fs_end_read_compressed_page(page, PageError(page),
241						blkaddr);
242		else
243			all_compressed = false;
244
245		blkaddr++;
246	}
247
248	/*
249	 * Optimization: if all the bio's pages are compressed, then scheduling
250	 * the per-bio verity work is unnecessary, as verity will be fully
251	 * handled at the compression cluster level.
252	 */
253	if (all_compressed)
254		ctx->enabled_steps &= ~STEP_VERITY;
255}
256
257static void f2fs_post_read_work(struct work_struct *work)
258{
259	struct bio_post_read_ctx *ctx =
260		container_of(work, struct bio_post_read_ctx, work);
261
262	if (ctx->enabled_steps & STEP_DECRYPT)
263		fscrypt_decrypt_bio(ctx->bio);
264
265	if (ctx->enabled_steps & STEP_DECOMPRESS)
266		f2fs_handle_step_decompress(ctx);
267
268	f2fs_verify_and_finish_bio(ctx->bio);
269}
270
271static void f2fs_read_end_io(struct bio *bio)
272{
273	struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
274	struct bio_post_read_ctx *ctx;
275
276	iostat_update_and_unbind_ctx(bio, 0);
277	ctx = bio->bi_private;
278
279	if (time_to_inject(sbi, FAULT_READ_IO)) {
280		f2fs_show_injection_info(sbi, FAULT_READ_IO);
281		bio->bi_status = BLK_STS_IOERR;
282	}
283
284	if (bio->bi_status) {
285		f2fs_finish_read_bio(bio);
286		return;
287	}
288
289	if (ctx && (ctx->enabled_steps & (STEP_DECRYPT | STEP_DECOMPRESS))) {
290		INIT_WORK(&ctx->work, f2fs_post_read_work);
291		queue_work(ctx->sbi->post_read_wq, &ctx->work);
292	} else {
293		f2fs_verify_and_finish_bio(bio);
294	}
295}
296
297static void f2fs_write_end_io(struct bio *bio)
298{
299	struct f2fs_sb_info *sbi;
300	struct bio_vec *bvec;
301	struct bvec_iter_all iter_all;
302
303	iostat_update_and_unbind_ctx(bio, 1);
304	sbi = bio->bi_private;
305
306	if (time_to_inject(sbi, FAULT_WRITE_IO)) {
307		f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
308		bio->bi_status = BLK_STS_IOERR;
309	}
310
311	bio_for_each_segment_all(bvec, bio, iter_all) {
312		struct page *page = bvec->bv_page;
313		enum count_type type = WB_DATA_TYPE(page);
314
315		if (page_private_dummy(page)) {
316			clear_page_private_dummy(page);
317			unlock_page(page);
318			mempool_free(page, sbi->write_io_dummy);
319
320			if (unlikely(bio->bi_status))
321				f2fs_stop_checkpoint(sbi, true);
322			continue;
323		}
324
325		fscrypt_finalize_bounce_page(&page);
326
327#ifdef CONFIG_F2FS_FS_COMPRESSION
328		if (f2fs_is_compressed_page(page)) {
329			f2fs_compress_write_end_io(bio, page);
330			continue;
331		}
332#endif
333
334		if (unlikely(bio->bi_status)) {
335			mapping_set_error(page->mapping, -EIO);
336			if (type == F2FS_WB_CP_DATA)
337				f2fs_stop_checkpoint(sbi, true);
338		}
339
340		f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
341					page->index != nid_of_node(page));
342
343		dec_page_count(sbi, type);
344		if (f2fs_in_warm_node_list(sbi, page))
345			f2fs_del_fsync_node_entry(sbi, page);
346		clear_page_private_gcing(page);
347		end_page_writeback(page);
348	}
349	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
350				wq_has_sleeper(&sbi->cp_wait))
351		wake_up(&sbi->cp_wait);
352
353	bio_put(bio);
354}
355
356struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
357				block_t blk_addr, struct bio *bio)
358{
359	struct block_device *bdev = sbi->sb->s_bdev;
360	int i;
361
362	if (f2fs_is_multi_device(sbi)) {
363		for (i = 0; i < sbi->s_ndevs; i++) {
364			if (FDEV(i).start_blk <= blk_addr &&
365			    FDEV(i).end_blk >= blk_addr) {
366				blk_addr -= FDEV(i).start_blk;
367				bdev = FDEV(i).bdev;
368				break;
369			}
370		}
371	}
372	if (bio) {
373		bio_set_dev(bio, bdev);
374		bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
375	}
376	return bdev;
377}
378
379int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
380{
381	int i;
382
383	if (!f2fs_is_multi_device(sbi))
384		return 0;
385
386	for (i = 0; i < sbi->s_ndevs; i++)
387		if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
388			return i;
389	return 0;
390}
391
392static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
393{
394	struct f2fs_sb_info *sbi = fio->sbi;
395	struct bio *bio;
396
397	bio = bio_alloc_bioset(GFP_NOIO, npages, &f2fs_bioset);
398
399	f2fs_target_device(sbi, fio->new_blkaddr, bio);
400	if (is_read_io(fio->op)) {
401		bio->bi_end_io = f2fs_read_end_io;
402		bio->bi_private = NULL;
403	} else {
404		bio->bi_end_io = f2fs_write_end_io;
405		bio->bi_private = sbi;
406		bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
407						fio->type, fio->temp);
408	}
409	iostat_alloc_and_bind_ctx(sbi, bio, NULL);
410
411	if (fio->io_wbc)
412		wbc_init_bio(fio->io_wbc, bio);
413
414	return bio;
415}
416
417static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
418				  pgoff_t first_idx,
419				  const struct f2fs_io_info *fio,
420				  gfp_t gfp_mask)
421{
422	/*
423	 * The f2fs garbage collector sets ->encrypted_page when it wants to
424	 * read/write raw data without encryption.
425	 */
426	if (!fio || !fio->encrypted_page)
427		fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
428}
429
430static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
431				     pgoff_t next_idx,
432				     const struct f2fs_io_info *fio)
433{
434	/*
435	 * The f2fs garbage collector sets ->encrypted_page when it wants to
436	 * read/write raw data without encryption.
437	 */
438	if (fio && fio->encrypted_page)
439		return !bio_has_crypt_ctx(bio);
440
441	return fscrypt_mergeable_bio(bio, inode, next_idx);
442}
443
444static inline void __submit_bio(struct f2fs_sb_info *sbi,
445				struct bio *bio, enum page_type type)
446{
447	if (!is_read_io(bio_op(bio))) {
448		unsigned int start;
449
450		if (type != DATA && type != NODE)
451			goto submit_io;
452
453		if (f2fs_lfs_mode(sbi) && current->plug)
454			blk_finish_plug(current->plug);
455
456		if (!F2FS_IO_ALIGNED(sbi))
457			goto submit_io;
458
459		start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
460		start %= F2FS_IO_SIZE(sbi);
461
462		if (start == 0)
463			goto submit_io;
464
465		/* fill dummy pages */
466		for (; start < F2FS_IO_SIZE(sbi); start++) {
467			struct page *page =
468				mempool_alloc(sbi->write_io_dummy,
469					      GFP_NOIO | __GFP_NOFAIL);
470			f2fs_bug_on(sbi, !page);
471
472			lock_page(page);
473
474			zero_user_segment(page, 0, PAGE_SIZE);
475			set_page_private_dummy(page);
476
477			if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
478				f2fs_bug_on(sbi, 1);
479		}
480		/*
481		 * In the NODE case, we lose next block address chain. So, we
482		 * need to do checkpoint in f2fs_sync_file.
483		 */
484		if (type == NODE)
485			set_sbi_flag(sbi, SBI_NEED_CP);
486	}
487submit_io:
488	if (is_read_io(bio_op(bio)))
489		trace_f2fs_submit_read_bio(sbi->sb, type, bio);
490	else
491		trace_f2fs_submit_write_bio(sbi->sb, type, bio);
492
493	iostat_update_submit_ctx(bio, type);
494	submit_bio(bio);
495}
496
497void f2fs_submit_bio(struct f2fs_sb_info *sbi,
498				struct bio *bio, enum page_type type)
499{
500	__submit_bio(sbi, bio, type);
501}
502
503static void __attach_io_flag(struct f2fs_io_info *fio)
504{
505	struct f2fs_sb_info *sbi = fio->sbi;
506	unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
507	unsigned int io_flag, fua_flag, meta_flag;
508
509	if (fio->type == DATA)
510		io_flag = sbi->data_io_flag;
511	else if (fio->type == NODE)
512		io_flag = sbi->node_io_flag;
513	else
514		return;
515
516	fua_flag = io_flag & temp_mask;
517	meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
518
519	/*
520	 * data/node io flag bits per temp:
521	 *      REQ_META     |      REQ_FUA      |
522	 *    5 |    4 |   3 |    2 |    1 |   0 |
523	 * Cold | Warm | Hot | Cold | Warm | Hot |
524	 */
525	if ((1 << fio->temp) & meta_flag)
526		fio->op_flags |= REQ_META;
527	if ((1 << fio->temp) & fua_flag)
528		fio->op_flags |= REQ_FUA;
529}
530
531static void __submit_merged_bio(struct f2fs_bio_info *io)
532{
533	struct f2fs_io_info *fio = &io->fio;
534
535	if (!io->bio)
536		return;
537
538	__attach_io_flag(fio);
539	bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
540
541	if (is_read_io(fio->op))
542		trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
543	else
544		trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
545
546	__submit_bio(io->sbi, io->bio, fio->type);
547	io->bio = NULL;
548}
549
550static bool __has_merged_page(struct bio *bio, struct inode *inode,
551						struct page *page, nid_t ino)
552{
553	struct bio_vec *bvec;
554	struct bvec_iter_all iter_all;
555
556	if (!bio)
557		return false;
558
559	if (!inode && !page && !ino)
560		return true;
561
562	bio_for_each_segment_all(bvec, bio, iter_all) {
563		struct page *target = bvec->bv_page;
564
565		if (fscrypt_is_bounce_page(target)) {
566			target = fscrypt_pagecache_page(target);
567			if (IS_ERR(target))
568				continue;
569		}
570		if (f2fs_is_compressed_page(target)) {
571			target = f2fs_compress_control_page(target);
572			if (IS_ERR(target))
573				continue;
574		}
575
576		if (inode && inode == target->mapping->host)
577			return true;
578		if (page && page == target)
579			return true;
580		if (ino && ino == ino_of_node(target))
581			return true;
582	}
583
584	return false;
585}
586
587static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
588				enum page_type type, enum temp_type temp)
589{
590	enum page_type btype = PAGE_TYPE_OF_BIO(type);
591	struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
592
593	down_write(&io->io_rwsem);
594
595	/* change META to META_FLUSH in the checkpoint procedure */
596	if (type >= META_FLUSH) {
597		io->fio.type = META_FLUSH;
598		io->fio.op = REQ_OP_WRITE;
599		io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
600		if (!test_opt(sbi, NOBARRIER))
601			io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
602	}
603	__submit_merged_bio(io);
604	up_write(&io->io_rwsem);
605}
606
607static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
608				struct inode *inode, struct page *page,
609				nid_t ino, enum page_type type, bool force)
610{
611	enum temp_type temp;
612	bool ret = true;
613
614	for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
615		if (!force)	{
616			enum page_type btype = PAGE_TYPE_OF_BIO(type);
617			struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
618
619			down_read(&io->io_rwsem);
620			ret = __has_merged_page(io->bio, inode, page, ino);
621			up_read(&io->io_rwsem);
622		}
623		if (ret)
624			__f2fs_submit_merged_write(sbi, type, temp);
625
626		/* TODO: use HOT temp only for meta pages now. */
627		if (type >= META)
628			break;
629	}
630}
631
632void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
633{
634	__submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
635}
636
637void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
638				struct inode *inode, struct page *page,
639				nid_t ino, enum page_type type)
640{
641	__submit_merged_write_cond(sbi, inode, page, ino, type, false);
642}
643
644void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
645{
646	f2fs_submit_merged_write(sbi, DATA);
647	f2fs_submit_merged_write(sbi, NODE);
648	f2fs_submit_merged_write(sbi, META);
649}
650
651/*
652 * Fill the locked page with data located in the block address.
653 * A caller needs to unlock the page on failure.
654 */
655int f2fs_submit_page_bio(struct f2fs_io_info *fio)
656{
657	struct bio *bio;
658	struct page *page = fio->encrypted_page ?
659			fio->encrypted_page : fio->page;
660
661	if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
662			fio->is_por ? META_POR : (__is_meta_io(fio) ?
663			META_GENERIC : DATA_GENERIC_ENHANCE)))
664		return -EFSCORRUPTED;
665
666	trace_f2fs_submit_page_bio(page, fio);
667
668	/* Allocate a new bio */
669	bio = __bio_alloc(fio, 1);
670
671	f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
672			       fio->page->index, fio, GFP_NOIO);
673
674	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
675		bio_put(bio);
676		return -EFAULT;
677	}
678
679	if (fio->io_wbc && !is_read_io(fio->op))
680		wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
681
682	__attach_io_flag(fio);
683	bio_set_op_attrs(bio, fio->op, fio->op_flags);
684
685	inc_page_count(fio->sbi, is_read_io(fio->op) ?
686			__read_io_type(page): WB_DATA_TYPE(fio->page));
687
688	__submit_bio(fio->sbi, bio, fio->type);
689	return 0;
690}
691
692static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
693				block_t last_blkaddr, block_t cur_blkaddr)
694{
695	if (unlikely(sbi->max_io_bytes &&
696			bio->bi_iter.bi_size >= sbi->max_io_bytes))
697		return false;
698	if (last_blkaddr + 1 != cur_blkaddr)
699		return false;
700	return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
701}
702
703static bool io_type_is_mergeable(struct f2fs_bio_info *io,
704						struct f2fs_io_info *fio)
705{
706	if (io->fio.op != fio->op)
707		return false;
708	return io->fio.op_flags == fio->op_flags;
709}
710
711static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
712					struct f2fs_bio_info *io,
713					struct f2fs_io_info *fio,
714					block_t last_blkaddr,
715					block_t cur_blkaddr)
716{
717	if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
718		unsigned int filled_blocks =
719				F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
720		unsigned int io_size = F2FS_IO_SIZE(sbi);
721		unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
722
723		/* IOs in bio is aligned and left space of vectors is not enough */
724		if (!(filled_blocks % io_size) && left_vecs < io_size)
725			return false;
726	}
727	if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
728		return false;
729	return io_type_is_mergeable(io, fio);
730}
731
732static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
733				struct page *page, enum temp_type temp)
734{
735	struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
736	struct bio_entry *be;
737
738	be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
739	be->bio = bio;
740	bio_get(bio);
741
742	if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
743		f2fs_bug_on(sbi, 1);
744
745	down_write(&io->bio_list_lock);
746	list_add_tail(&be->list, &io->bio_list);
747	up_write(&io->bio_list_lock);
748}
749
750static void del_bio_entry(struct bio_entry *be)
751{
752	list_del(&be->list);
753	kmem_cache_free(bio_entry_slab, be);
754}
755
756static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
757							struct page *page)
758{
759	struct f2fs_sb_info *sbi = fio->sbi;
760	enum temp_type temp;
761	bool found = false;
762	int ret = -EAGAIN;
763
764	for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
765		struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
766		struct list_head *head = &io->bio_list;
767		struct bio_entry *be;
768
769		down_write(&io->bio_list_lock);
770		list_for_each_entry(be, head, list) {
771			if (be->bio != *bio)
772				continue;
773
774			found = true;
775
776			f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
777							    *fio->last_block,
778							    fio->new_blkaddr));
779			if (f2fs_crypt_mergeable_bio(*bio,
780					fio->page->mapping->host,
781					fio->page->index, fio) &&
782			    bio_add_page(*bio, page, PAGE_SIZE, 0) ==
783					PAGE_SIZE) {
784				ret = 0;
785				break;
786			}
787
788			/* page can't be merged into bio; submit the bio */
789			del_bio_entry(be);
790			__submit_bio(sbi, *bio, DATA);
791			break;
792		}
793		up_write(&io->bio_list_lock);
794	}
795
796	if (ret) {
797		bio_put(*bio);
798		*bio = NULL;
799	}
800
801	return ret;
802}
803
804void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
805					struct bio **bio, struct page *page)
806{
807	enum temp_type temp;
808	bool found = false;
809	struct bio *target = bio ? *bio : NULL;
810
811	for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
812		struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
813		struct list_head *head = &io->bio_list;
814		struct bio_entry *be;
815
816		if (list_empty(head))
817			continue;
818
819		down_read(&io->bio_list_lock);
820		list_for_each_entry(be, head, list) {
821			if (target)
822				found = (target == be->bio);
823			else
824				found = __has_merged_page(be->bio, NULL,
825								page, 0);
826			if (found)
827				break;
828		}
829		up_read(&io->bio_list_lock);
830
831		if (!found)
832			continue;
833
834		found = false;
835
836		down_write(&io->bio_list_lock);
837		list_for_each_entry(be, head, list) {
838			if (target)
839				found = (target == be->bio);
840			else
841				found = __has_merged_page(be->bio, NULL,
842								page, 0);
843			if (found) {
844				target = be->bio;
845				del_bio_entry(be);
846				break;
847			}
848		}
849		up_write(&io->bio_list_lock);
850	}
851
852	if (found)
853		__submit_bio(sbi, target, DATA);
854	if (bio && *bio) {
855		bio_put(*bio);
856		*bio = NULL;
857	}
858}
859
860int f2fs_merge_page_bio(struct f2fs_io_info *fio)
861{
862	struct bio *bio = *fio->bio;
863	struct page *page = fio->encrypted_page ?
864			fio->encrypted_page : fio->page;
865
866	if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
867			__is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
868		return -EFSCORRUPTED;
869
870	trace_f2fs_submit_page_bio(page, fio);
871
872	if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
873						fio->new_blkaddr))
874		f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
875alloc_new:
876	if (!bio) {
877		bio = __bio_alloc(fio, BIO_MAX_VECS);
878		__attach_io_flag(fio);
879		f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
880				       fio->page->index, fio, GFP_NOIO);
881		bio_set_op_attrs(bio, fio->op, fio->op_flags);
882
883		add_bio_entry(fio->sbi, bio, page, fio->temp);
884	} else {
885		if (add_ipu_page(fio, &bio, page))
886			goto alloc_new;
887	}
888
889	if (fio->io_wbc)
890		wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
891
892	inc_page_count(fio->sbi, WB_DATA_TYPE(page));
893
894	*fio->last_block = fio->new_blkaddr;
895	*fio->bio = bio;
896
897	return 0;
898}
899
900void f2fs_submit_page_write(struct f2fs_io_info *fio)
901{
902	struct f2fs_sb_info *sbi = fio->sbi;
903	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
904	struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
905	struct page *bio_page;
906
907	f2fs_bug_on(sbi, is_read_io(fio->op));
908
909	down_write(&io->io_rwsem);
910next:
911	if (fio->in_list) {
912		spin_lock(&io->io_lock);
913		if (list_empty(&io->io_list)) {
914			spin_unlock(&io->io_lock);
915			goto out;
916		}
917		fio = list_first_entry(&io->io_list,
918						struct f2fs_io_info, list);
919		list_del(&fio->list);
920		spin_unlock(&io->io_lock);
921	}
922
923	verify_fio_blkaddr(fio);
924
925	if (fio->encrypted_page)
926		bio_page = fio->encrypted_page;
927	else if (fio->compressed_page)
928		bio_page = fio->compressed_page;
929	else
930		bio_page = fio->page;
931
932	/* set submitted = true as a return value */
933	fio->submitted = true;
934
935	inc_page_count(sbi, WB_DATA_TYPE(bio_page));
936
937	if (io->bio &&
938	    (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
939			      fio->new_blkaddr) ||
940	     !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
941				       bio_page->index, fio)))
942		__submit_merged_bio(io);
943alloc_new:
944	if (io->bio == NULL) {
945		if (F2FS_IO_ALIGNED(sbi) &&
946				(fio->type == DATA || fio->type == NODE) &&
947				fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
948			dec_page_count(sbi, WB_DATA_TYPE(bio_page));
949			fio->retry = true;
950			goto skip;
951		}
952		io->bio = __bio_alloc(fio, BIO_MAX_VECS);
953		f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
954				       bio_page->index, fio, GFP_NOIO);
955		io->fio = *fio;
956	}
957
958	if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
959		__submit_merged_bio(io);
960		goto alloc_new;
961	}
962
963	if (fio->io_wbc)
964		wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
965
966	io->last_block_in_bio = fio->new_blkaddr;
967
968	trace_f2fs_submit_page_write(fio->page, fio);
969skip:
970	if (fio->in_list)
971		goto next;
972out:
973	if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
974				!f2fs_is_checkpoint_ready(sbi))
975		__submit_merged_bio(io);
976	up_write(&io->io_rwsem);
977}
978
979static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
980				      unsigned nr_pages, unsigned op_flag,
981				      pgoff_t first_idx, bool for_write)
982{
983	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
984	struct bio *bio;
985	struct bio_post_read_ctx *ctx = NULL;
986	unsigned int post_read_steps = 0;
987
988	bio = bio_alloc_bioset(for_write ? GFP_NOIO : GFP_KERNEL,
989			       bio_max_segs(nr_pages), &f2fs_bioset);
990	if (!bio)
991		return ERR_PTR(-ENOMEM);
992
993	f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
994
995	f2fs_target_device(sbi, blkaddr, bio);
996	bio->bi_end_io = f2fs_read_end_io;
997	bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
998
999	if (fscrypt_inode_uses_fs_layer_crypto(inode))
1000		post_read_steps |= STEP_DECRYPT;
1001
1002	if (f2fs_need_verity(inode, first_idx))
1003		post_read_steps |= STEP_VERITY;
1004
1005	/*
1006	 * STEP_DECOMPRESS is handled specially, since a compressed file might
1007	 * contain both compressed and uncompressed clusters.  We'll allocate a
1008	 * bio_post_read_ctx if the file is compressed, but the caller is
1009	 * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1010	 */
1011
1012	if (post_read_steps || f2fs_compressed_file(inode)) {
1013		/* Due to the mempool, this never fails. */
1014		ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1015		ctx->bio = bio;
1016		ctx->sbi = sbi;
1017		ctx->enabled_steps = post_read_steps;
1018		ctx->fs_blkaddr = blkaddr;
1019		bio->bi_private = ctx;
1020	}
1021	iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1022
1023	return bio;
1024}
1025
1026/* This can handle encryption stuffs */
1027static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1028				 block_t blkaddr, int op_flags, bool for_write)
1029{
1030	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1031	struct bio *bio;
1032
1033	bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1034					page->index, for_write);
1035	if (IS_ERR(bio))
1036		return PTR_ERR(bio);
1037
1038	/* wait for GCed page writeback via META_MAPPING */
1039	f2fs_wait_on_block_writeback(inode, blkaddr);
1040
1041	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1042		bio_put(bio);
1043		return -EFAULT;
1044	}
1045	ClearPageError(page);
1046	inc_page_count(sbi, F2FS_RD_DATA);
1047	f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1048	__submit_bio(sbi, bio, DATA);
1049	return 0;
1050}
1051
1052static void __set_data_blkaddr(struct dnode_of_data *dn)
1053{
1054	struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1055	__le32 *addr_array;
1056	int base = 0;
1057
1058	if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1059		base = get_extra_isize(dn->inode);
1060
1061	/* Get physical address of data block */
1062	addr_array = blkaddr_in_node(rn);
1063	addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1064}
1065
1066/*
1067 * Lock ordering for the change of data block address:
1068 * ->data_page
1069 *  ->node_page
1070 *    update block addresses in the node page
1071 */
1072void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1073{
1074	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1075	__set_data_blkaddr(dn);
1076	if (set_page_dirty(dn->node_page))
1077		dn->node_changed = true;
1078}
1079
1080void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1081{
1082	dn->data_blkaddr = blkaddr;
1083	f2fs_set_data_blkaddr(dn);
1084	f2fs_update_extent_cache(dn);
1085}
1086
1087/* dn->ofs_in_node will be returned with up-to-date last block pointer */
1088int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1089{
1090	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1091	int err;
1092
1093	if (!count)
1094		return 0;
1095
1096	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1097		return -EPERM;
1098	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1099		return err;
1100
1101	trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1102						dn->ofs_in_node, count);
1103
1104	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1105
1106	for (; count > 0; dn->ofs_in_node++) {
1107		block_t blkaddr = f2fs_data_blkaddr(dn);
1108
1109		if (blkaddr == NULL_ADDR) {
1110			dn->data_blkaddr = NEW_ADDR;
1111			__set_data_blkaddr(dn);
1112			count--;
1113		}
1114	}
1115
1116	if (set_page_dirty(dn->node_page))
1117		dn->node_changed = true;
1118	return 0;
1119}
1120
1121/* Should keep dn->ofs_in_node unchanged */
1122int f2fs_reserve_new_block(struct dnode_of_data *dn)
1123{
1124	unsigned int ofs_in_node = dn->ofs_in_node;
1125	int ret;
1126
1127	ret = f2fs_reserve_new_blocks(dn, 1);
1128	dn->ofs_in_node = ofs_in_node;
1129	return ret;
1130}
1131
1132int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1133{
1134	bool need_put = dn->inode_page ? false : true;
1135	int err;
1136
1137	err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1138	if (err)
1139		return err;
1140
1141	if (dn->data_blkaddr == NULL_ADDR)
1142		err = f2fs_reserve_new_block(dn);
1143	if (err || need_put)
1144		f2fs_put_dnode(dn);
1145	return err;
1146}
1147
1148int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1149{
1150	struct extent_info ei = {0, };
1151	struct inode *inode = dn->inode;
1152
1153	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1154		dn->data_blkaddr = ei.blk + index - ei.fofs;
1155		return 0;
1156	}
1157
1158	return f2fs_reserve_block(dn, index);
1159}
1160
1161struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1162						int op_flags, bool for_write)
1163{
1164	struct address_space *mapping = inode->i_mapping;
1165	struct dnode_of_data dn;
1166	struct page *page;
1167	struct extent_info ei = {0, };
1168	int err;
1169
1170	page = f2fs_grab_cache_page(mapping, index, for_write);
1171	if (!page)
1172		return ERR_PTR(-ENOMEM);
1173
1174	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1175		dn.data_blkaddr = ei.blk + index - ei.fofs;
1176		if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1177						DATA_GENERIC_ENHANCE_READ)) {
1178			err = -EFSCORRUPTED;
1179			goto put_err;
1180		}
1181		goto got_it;
1182	}
1183
1184	set_new_dnode(&dn, inode, NULL, NULL, 0);
1185	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1186	if (err)
1187		goto put_err;
1188	f2fs_put_dnode(&dn);
1189
1190	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1191		err = -ENOENT;
1192		goto put_err;
1193	}
1194	if (dn.data_blkaddr != NEW_ADDR &&
1195			!f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1196						dn.data_blkaddr,
1197						DATA_GENERIC_ENHANCE)) {
1198		err = -EFSCORRUPTED;
1199		goto put_err;
1200	}
1201got_it:
1202	if (PageUptodate(page)) {
1203		unlock_page(page);
1204		return page;
1205	}
1206
1207	/*
1208	 * A new dentry page is allocated but not able to be written, since its
1209	 * new inode page couldn't be allocated due to -ENOSPC.
1210	 * In such the case, its blkaddr can be remained as NEW_ADDR.
1211	 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1212	 * f2fs_init_inode_metadata.
1213	 */
1214	if (dn.data_blkaddr == NEW_ADDR) {
1215		zero_user_segment(page, 0, PAGE_SIZE);
1216		if (!PageUptodate(page))
1217			SetPageUptodate(page);
1218		unlock_page(page);
1219		return page;
1220	}
1221
1222	err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1223						op_flags, for_write);
1224	if (err)
1225		goto put_err;
1226	return page;
1227
1228put_err:
1229	f2fs_put_page(page, 1);
1230	return ERR_PTR(err);
1231}
1232
1233struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1234{
1235	struct address_space *mapping = inode->i_mapping;
1236	struct page *page;
1237
1238	page = find_get_page(mapping, index);
1239	if (page && PageUptodate(page))
1240		return page;
1241	f2fs_put_page(page, 0);
1242
1243	page = f2fs_get_read_data_page(inode, index, 0, false);
1244	if (IS_ERR(page))
1245		return page;
1246
1247	if (PageUptodate(page))
1248		return page;
1249
1250	wait_on_page_locked(page);
1251	if (unlikely(!PageUptodate(page))) {
1252		f2fs_put_page(page, 0);
1253		return ERR_PTR(-EIO);
1254	}
1255	return page;
1256}
1257
1258/*
1259 * If it tries to access a hole, return an error.
1260 * Because, the callers, functions in dir.c and GC, should be able to know
1261 * whether this page exists or not.
1262 */
1263struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1264							bool for_write)
1265{
1266	struct address_space *mapping = inode->i_mapping;
1267	struct page *page;
1268repeat:
1269	page = f2fs_get_read_data_page(inode, index, 0, for_write);
1270	if (IS_ERR(page))
1271		return page;
1272
1273	/* wait for read completion */
1274	lock_page(page);
1275	if (unlikely(page->mapping != mapping)) {
1276		f2fs_put_page(page, 1);
1277		goto repeat;
1278	}
1279	if (unlikely(!PageUptodate(page))) {
1280		f2fs_put_page(page, 1);
1281		return ERR_PTR(-EIO);
1282	}
1283	return page;
1284}
1285
1286/*
1287 * Caller ensures that this data page is never allocated.
1288 * A new zero-filled data page is allocated in the page cache.
1289 *
1290 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1291 * f2fs_unlock_op().
1292 * Note that, ipage is set only by make_empty_dir, and if any error occur,
1293 * ipage should be released by this function.
1294 */
1295struct page *f2fs_get_new_data_page(struct inode *inode,
1296		struct page *ipage, pgoff_t index, bool new_i_size)
1297{
1298	struct address_space *mapping = inode->i_mapping;
1299	struct page *page;
1300	struct dnode_of_data dn;
1301	int err;
1302
1303	page = f2fs_grab_cache_page(mapping, index, true);
1304	if (!page) {
1305		/*
1306		 * before exiting, we should make sure ipage will be released
1307		 * if any error occur.
1308		 */
1309		f2fs_put_page(ipage, 1);
1310		return ERR_PTR(-ENOMEM);
1311	}
1312
1313	set_new_dnode(&dn, inode, ipage, NULL, 0);
1314	err = f2fs_reserve_block(&dn, index);
1315	if (err) {
1316		f2fs_put_page(page, 1);
1317		return ERR_PTR(err);
1318	}
1319	if (!ipage)
1320		f2fs_put_dnode(&dn);
1321
1322	if (PageUptodate(page))
1323		goto got_it;
1324
1325	if (dn.data_blkaddr == NEW_ADDR) {
1326		zero_user_segment(page, 0, PAGE_SIZE);
1327		if (!PageUptodate(page))
1328			SetPageUptodate(page);
1329	} else {
1330		f2fs_put_page(page, 1);
1331
1332		/* if ipage exists, blkaddr should be NEW_ADDR */
1333		f2fs_bug_on(F2FS_I_SB(inode), ipage);
1334		page = f2fs_get_lock_data_page(inode, index, true);
1335		if (IS_ERR(page))
1336			return page;
1337	}
1338got_it:
1339	if (new_i_size && i_size_read(inode) <
1340				((loff_t)(index + 1) << PAGE_SHIFT))
1341		f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1342	return page;
1343}
1344
1345static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1346{
1347	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1348	struct f2fs_summary sum;
1349	struct node_info ni;
1350	block_t old_blkaddr;
1351	blkcnt_t count = 1;
1352	int err;
1353
1354	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1355		return -EPERM;
1356
1357	err = f2fs_get_node_info(sbi, dn->nid, &ni);
1358	if (err)
1359		return err;
1360
1361	dn->data_blkaddr = f2fs_data_blkaddr(dn);
1362	if (dn->data_blkaddr != NULL_ADDR)
1363		goto alloc;
1364
1365	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1366		return err;
1367
1368alloc:
1369	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1370	old_blkaddr = dn->data_blkaddr;
1371	f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1372				&sum, seg_type, NULL);
1373	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1374		invalidate_mapping_pages(META_MAPPING(sbi),
1375					old_blkaddr, old_blkaddr);
1376		f2fs_invalidate_compress_page(sbi, old_blkaddr);
1377	}
1378	f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1379
1380	/*
1381	 * i_size will be updated by direct_IO. Otherwise, we'll get stale
1382	 * data from unwritten block via dio_read.
1383	 */
1384	return 0;
1385}
1386
1387int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1388{
1389	struct inode *inode = file_inode(iocb->ki_filp);
1390	struct f2fs_map_blocks map;
1391	int flag;
1392	int err = 0;
1393	bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1394
1395	map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1396	map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1397	if (map.m_len > map.m_lblk)
1398		map.m_len -= map.m_lblk;
1399	else
1400		map.m_len = 0;
1401
1402	map.m_next_pgofs = NULL;
1403	map.m_next_extent = NULL;
1404	map.m_seg_type = NO_CHECK_TYPE;
1405	map.m_may_create = true;
1406
1407	if (direct_io) {
1408		map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1409		flag = f2fs_force_buffered_io(inode, iocb, from) ?
1410					F2FS_GET_BLOCK_PRE_AIO :
1411					F2FS_GET_BLOCK_PRE_DIO;
1412		goto map_blocks;
1413	}
1414	if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1415		err = f2fs_convert_inline_inode(inode);
1416		if (err)
1417			return err;
1418	}
1419	if (f2fs_has_inline_data(inode))
1420		return err;
1421
1422	flag = F2FS_GET_BLOCK_PRE_AIO;
1423
1424map_blocks:
1425	err = f2fs_map_blocks(inode, &map, 1, flag);
1426	if (map.m_len > 0 && err == -ENOSPC) {
1427		if (!direct_io)
1428			set_inode_flag(inode, FI_NO_PREALLOC);
1429		err = 0;
1430	}
1431	return err;
1432}
1433
1434void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1435{
1436	if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1437		if (lock)
1438			down_read(&sbi->node_change);
1439		else
1440			up_read(&sbi->node_change);
1441	} else {
1442		if (lock)
1443			f2fs_lock_op(sbi);
1444		else
1445			f2fs_unlock_op(sbi);
1446	}
1447}
1448
1449/*
1450 * f2fs_map_blocks() tries to find or build mapping relationship which
1451 * maps continuous logical blocks to physical blocks, and return such
1452 * info via f2fs_map_blocks structure.
1453 */
1454int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1455						int create, int flag)
1456{
1457	unsigned int maxblocks = map->m_len;
1458	struct dnode_of_data dn;
1459	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1460	int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1461	pgoff_t pgofs, end_offset, end;
1462	int err = 0, ofs = 1;
1463	unsigned int ofs_in_node, last_ofs_in_node;
1464	blkcnt_t prealloc;
1465	struct extent_info ei = {0, };
1466	block_t blkaddr;
1467	unsigned int start_pgofs;
1468
1469	if (!maxblocks)
1470		return 0;
1471
1472	map->m_len = 0;
1473	map->m_flags = 0;
1474
1475	/* it only supports block size == page size */
1476	pgofs =	(pgoff_t)map->m_lblk;
1477	end = pgofs + maxblocks;
1478
1479	if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1480		if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1481							map->m_may_create)
1482			goto next_dnode;
1483
1484		map->m_pblk = ei.blk + pgofs - ei.fofs;
1485		map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1486		map->m_flags = F2FS_MAP_MAPPED;
1487		if (map->m_next_extent)
1488			*map->m_next_extent = pgofs + map->m_len;
1489
1490		/* for hardware encryption, but to avoid potential issue in future */
1491		if (flag == F2FS_GET_BLOCK_DIO)
1492			f2fs_wait_on_block_writeback_range(inode,
1493						map->m_pblk, map->m_len);
1494		goto out;
1495	}
1496
1497next_dnode:
1498	if (map->m_may_create)
1499		f2fs_do_map_lock(sbi, flag, true);
1500
1501	/* When reading holes, we need its node page */
1502	set_new_dnode(&dn, inode, NULL, NULL, 0);
1503	err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1504	if (err) {
1505		if (flag == F2FS_GET_BLOCK_BMAP)
1506			map->m_pblk = 0;
1507
1508		if (err == -ENOENT) {
1509			/*
1510			 * There is one exceptional case that read_node_page()
1511			 * may return -ENOENT due to filesystem has been
1512			 * shutdown or cp_error, so force to convert error
1513			 * number to EIO for such case.
1514			 */
1515			if (map->m_may_create &&
1516				(is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1517				f2fs_cp_error(sbi))) {
1518				err = -EIO;
1519				goto unlock_out;
1520			}
1521
1522			err = 0;
1523			if (map->m_next_pgofs)
1524				*map->m_next_pgofs =
1525					f2fs_get_next_page_offset(&dn, pgofs);
1526			if (map->m_next_extent)
1527				*map->m_next_extent =
1528					f2fs_get_next_page_offset(&dn, pgofs);
1529		}
1530		goto unlock_out;
1531	}
1532
1533	start_pgofs = pgofs;
1534	prealloc = 0;
1535	last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1536	end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1537
1538next_block:
1539	blkaddr = f2fs_data_blkaddr(&dn);
1540
1541	if (__is_valid_data_blkaddr(blkaddr) &&
1542		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1543		err = -EFSCORRUPTED;
1544		goto sync_out;
1545	}
1546
1547	if (__is_valid_data_blkaddr(blkaddr)) {
1548		/* use out-place-update for driect IO under LFS mode */
1549		if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1550							map->m_may_create) {
1551			err = __allocate_data_block(&dn, map->m_seg_type);
1552			if (err)
1553				goto sync_out;
1554			blkaddr = dn.data_blkaddr;
1555			set_inode_flag(inode, FI_APPEND_WRITE);
1556		}
1557	} else {
1558		if (create) {
1559			if (unlikely(f2fs_cp_error(sbi))) {
1560				err = -EIO;
1561				goto sync_out;
1562			}
1563			if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1564				if (blkaddr == NULL_ADDR) {
1565					prealloc++;
1566					last_ofs_in_node = dn.ofs_in_node;
1567				}
1568			} else {
1569				WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1570					flag != F2FS_GET_BLOCK_DIO);
1571				err = __allocate_data_block(&dn,
1572							map->m_seg_type);
1573				if (!err)
1574					set_inode_flag(inode, FI_APPEND_WRITE);
1575			}
1576			if (err)
1577				goto sync_out;
1578			map->m_flags |= F2FS_MAP_NEW;
1579			blkaddr = dn.data_blkaddr;
1580		} else {
1581			if (f2fs_compressed_file(inode) &&
1582					f2fs_sanity_check_cluster(&dn) &&
1583					(flag != F2FS_GET_BLOCK_FIEMAP ||
1584					IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1585				err = -EFSCORRUPTED;
1586				goto sync_out;
1587			}
1588			if (flag == F2FS_GET_BLOCK_BMAP) {
1589				map->m_pblk = 0;
1590				goto sync_out;
1591			}
1592			if (flag == F2FS_GET_BLOCK_PRECACHE)
1593				goto sync_out;
1594			if (flag == F2FS_GET_BLOCK_FIEMAP &&
1595						blkaddr == NULL_ADDR) {
1596				if (map->m_next_pgofs)
1597					*map->m_next_pgofs = pgofs + 1;
1598				goto sync_out;
1599			}
1600			if (flag != F2FS_GET_BLOCK_FIEMAP) {
1601				/* for defragment case */
1602				if (map->m_next_pgofs)
1603					*map->m_next_pgofs = pgofs + 1;
1604				goto sync_out;
1605			}
1606		}
1607	}
1608
1609	if (flag == F2FS_GET_BLOCK_PRE_AIO)
1610		goto skip;
1611
1612	if (map->m_len == 0) {
1613		/* preallocated unwritten block should be mapped for fiemap. */
1614		if (blkaddr == NEW_ADDR)
1615			map->m_flags |= F2FS_MAP_UNWRITTEN;
1616		map->m_flags |= F2FS_MAP_MAPPED;
1617
1618		map->m_pblk = blkaddr;
1619		map->m_len = 1;
1620	} else if ((map->m_pblk != NEW_ADDR &&
1621			blkaddr == (map->m_pblk + ofs)) ||
1622			(map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1623			flag == F2FS_GET_BLOCK_PRE_DIO) {
1624		ofs++;
1625		map->m_len++;
1626	} else {
1627		goto sync_out;
1628	}
1629
1630skip:
1631	dn.ofs_in_node++;
1632	pgofs++;
1633
1634	/* preallocate blocks in batch for one dnode page */
1635	if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1636			(pgofs == end || dn.ofs_in_node == end_offset)) {
1637
1638		dn.ofs_in_node = ofs_in_node;
1639		err = f2fs_reserve_new_blocks(&dn, prealloc);
1640		if (err)
1641			goto sync_out;
1642
1643		map->m_len += dn.ofs_in_node - ofs_in_node;
1644		if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1645			err = -ENOSPC;
1646			goto sync_out;
1647		}
1648		dn.ofs_in_node = end_offset;
1649	}
1650
1651	if (pgofs >= end)
1652		goto sync_out;
1653	else if (dn.ofs_in_node < end_offset)
1654		goto next_block;
1655
1656	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1657		if (map->m_flags & F2FS_MAP_MAPPED) {
1658			unsigned int ofs = start_pgofs - map->m_lblk;
1659
1660			f2fs_update_extent_cache_range(&dn,
1661				start_pgofs, map->m_pblk + ofs,
1662				map->m_len - ofs);
1663		}
1664	}
1665
1666	f2fs_put_dnode(&dn);
1667
1668	if (map->m_may_create) {
1669		f2fs_do_map_lock(sbi, flag, false);
1670		f2fs_balance_fs(sbi, dn.node_changed);
1671	}
1672	goto next_dnode;
1673
1674sync_out:
1675
1676	/* for hardware encryption, but to avoid potential issue in future */
1677	if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1678		f2fs_wait_on_block_writeback_range(inode,
1679						map->m_pblk, map->m_len);
1680
1681	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1682		if (map->m_flags & F2FS_MAP_MAPPED) {
1683			unsigned int ofs = start_pgofs - map->m_lblk;
1684
1685			f2fs_update_extent_cache_range(&dn,
1686				start_pgofs, map->m_pblk + ofs,
1687				map->m_len - ofs);
1688		}
1689		if (map->m_next_extent)
1690			*map->m_next_extent = pgofs + 1;
1691	}
1692	f2fs_put_dnode(&dn);
1693unlock_out:
1694	if (map->m_may_create) {
1695		f2fs_do_map_lock(sbi, flag, false);
1696		f2fs_balance_fs(sbi, dn.node_changed);
1697	}
1698out:
1699	trace_f2fs_map_blocks(inode, map, err);
1700	return err;
1701}
1702
1703bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1704{
1705	struct f2fs_map_blocks map;
1706	block_t last_lblk;
1707	int err;
1708
1709	if (pos + len > i_size_read(inode))
1710		return false;
1711
1712	map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1713	map.m_next_pgofs = NULL;
1714	map.m_next_extent = NULL;
1715	map.m_seg_type = NO_CHECK_TYPE;
1716	map.m_may_create = false;
1717	last_lblk = F2FS_BLK_ALIGN(pos + len);
1718
1719	while (map.m_lblk < last_lblk) {
1720		map.m_len = last_lblk - map.m_lblk;
1721		err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1722		if (err || map.m_len == 0)
1723			return false;
1724		map.m_lblk += map.m_len;
1725	}
1726	return true;
1727}
1728
1729static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1730{
1731	return (bytes >> inode->i_blkbits);
1732}
1733
1734static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1735{
1736	return (blks << inode->i_blkbits);
1737}
1738
1739static int __get_data_block(struct inode *inode, sector_t iblock,
1740			struct buffer_head *bh, int create, int flag,
1741			pgoff_t *next_pgofs, int seg_type, bool may_write)
1742{
1743	struct f2fs_map_blocks map;
1744	int err;
1745
1746	map.m_lblk = iblock;
1747	map.m_len = bytes_to_blks(inode, bh->b_size);
1748	map.m_next_pgofs = next_pgofs;
1749	map.m_next_extent = NULL;
1750	map.m_seg_type = seg_type;
1751	map.m_may_create = may_write;
1752
1753	err = f2fs_map_blocks(inode, &map, create, flag);
1754	if (!err) {
1755		map_bh(bh, inode->i_sb, map.m_pblk);
1756		bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1757		bh->b_size = blks_to_bytes(inode, map.m_len);
1758	}
1759	return err;
1760}
1761
1762static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1763			struct buffer_head *bh_result, int create)
1764{
1765	return __get_data_block(inode, iblock, bh_result, create,
1766				F2FS_GET_BLOCK_DIO, NULL,
1767				f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1768				true);
1769}
1770
1771static int get_data_block_dio(struct inode *inode, sector_t iblock,
1772			struct buffer_head *bh_result, int create)
1773{
1774	return __get_data_block(inode, iblock, bh_result, create,
1775				F2FS_GET_BLOCK_DIO, NULL,
1776				f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1777				false);
1778}
1779
1780static int f2fs_xattr_fiemap(struct inode *inode,
1781				struct fiemap_extent_info *fieinfo)
1782{
1783	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1784	struct page *page;
1785	struct node_info ni;
1786	__u64 phys = 0, len;
1787	__u32 flags;
1788	nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1789	int err = 0;
1790
1791	if (f2fs_has_inline_xattr(inode)) {
1792		int offset;
1793
1794		page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1795						inode->i_ino, false);
1796		if (!page)
1797			return -ENOMEM;
1798
1799		err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1800		if (err) {
1801			f2fs_put_page(page, 1);
1802			return err;
1803		}
1804
1805		phys = blks_to_bytes(inode, ni.blk_addr);
1806		offset = offsetof(struct f2fs_inode, i_addr) +
1807					sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1808					get_inline_xattr_addrs(inode));
1809
1810		phys += offset;
1811		len = inline_xattr_size(inode);
1812
1813		f2fs_put_page(page, 1);
1814
1815		flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1816
1817		if (!xnid)
1818			flags |= FIEMAP_EXTENT_LAST;
1819
1820		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1821		trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1822		if (err || err == 1)
1823			return err;
1824	}
1825
1826	if (xnid) {
1827		page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1828		if (!page)
1829			return -ENOMEM;
1830
1831		err = f2fs_get_node_info(sbi, xnid, &ni);
1832		if (err) {
1833			f2fs_put_page(page, 1);
1834			return err;
1835		}
1836
1837		phys = blks_to_bytes(inode, ni.blk_addr);
1838		len = inode->i_sb->s_blocksize;
1839
1840		f2fs_put_page(page, 1);
1841
1842		flags = FIEMAP_EXTENT_LAST;
1843	}
1844
1845	if (phys) {
1846		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1847		trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1848	}
1849
1850	return (err < 0 ? err : 0);
1851}
1852
1853static loff_t max_inode_blocks(struct inode *inode)
1854{
1855	loff_t result = ADDRS_PER_INODE(inode);
1856	loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1857
1858	/* two direct node blocks */
1859	result += (leaf_count * 2);
1860
1861	/* two indirect node blocks */
1862	leaf_count *= NIDS_PER_BLOCK;
1863	result += (leaf_count * 2);
1864
1865	/* one double indirect node block */
1866	leaf_count *= NIDS_PER_BLOCK;
1867	result += leaf_count;
1868
1869	return result;
1870}
1871
1872int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1873		u64 start, u64 len)
1874{
1875	struct f2fs_map_blocks map;
1876	sector_t start_blk, last_blk;
1877	pgoff_t next_pgofs;
1878	u64 logical = 0, phys = 0, size = 0;
1879	u32 flags = 0;
1880	int ret = 0;
1881	bool compr_cluster = false, compr_appended;
1882	unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1883	unsigned int count_in_cluster = 0;
1884	loff_t maxbytes;
1885
1886	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1887		ret = f2fs_precache_extents(inode);
1888		if (ret)
1889			return ret;
1890	}
1891
1892	ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1893	if (ret)
1894		return ret;
1895
1896	inode_lock(inode);
1897
1898	maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1899	if (start > maxbytes) {
1900		ret = -EFBIG;
1901		goto out;
1902	}
1903
1904	if (len > maxbytes || (maxbytes - len) < start)
1905		len = maxbytes - start;
1906
1907	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1908		ret = f2fs_xattr_fiemap(inode, fieinfo);
1909		goto out;
1910	}
1911
1912	if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1913		ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1914		if (ret != -EAGAIN)
1915			goto out;
1916	}
1917
1918	if (bytes_to_blks(inode, len) == 0)
1919		len = blks_to_bytes(inode, 1);
1920
1921	start_blk = bytes_to_blks(inode, start);
1922	last_blk = bytes_to_blks(inode, start + len - 1);
1923
1924next:
1925	memset(&map, 0, sizeof(map));
1926	map.m_lblk = start_blk;
1927	map.m_len = bytes_to_blks(inode, len);
1928	map.m_next_pgofs = &next_pgofs;
1929	map.m_seg_type = NO_CHECK_TYPE;
1930
1931	if (compr_cluster) {
1932		map.m_lblk += 1;
1933		map.m_len = cluster_size - count_in_cluster;
1934	}
1935
1936	ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1937	if (ret)
1938		goto out;
1939
1940	/* HOLE */
1941	if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1942		start_blk = next_pgofs;
1943
1944		if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1945						max_inode_blocks(inode)))
1946			goto prep_next;
1947
1948		flags |= FIEMAP_EXTENT_LAST;
1949	}
1950
1951	compr_appended = false;
1952	/* In a case of compressed cluster, append this to the last extent */
1953	if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) ||
1954			!(map.m_flags & F2FS_MAP_FLAGS))) {
1955		compr_appended = true;
1956		goto skip_fill;
1957	}
1958
1959	if (size) {
1960		flags |= FIEMAP_EXTENT_MERGED;
1961		if (IS_ENCRYPTED(inode))
1962			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1963
1964		ret = fiemap_fill_next_extent(fieinfo, logical,
1965				phys, size, flags);
1966		trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1967		if (ret)
1968			goto out;
1969		size = 0;
1970	}
1971
1972	if (start_blk > last_blk)
1973		goto out;
1974
1975skip_fill:
1976	if (map.m_pblk == COMPRESS_ADDR) {
1977		compr_cluster = true;
1978		count_in_cluster = 1;
1979	} else if (compr_appended) {
1980		unsigned int appended_blks = cluster_size -
1981						count_in_cluster + 1;
1982		size += blks_to_bytes(inode, appended_blks);
1983		start_blk += appended_blks;
1984		compr_cluster = false;
1985	} else {
1986		logical = blks_to_bytes(inode, start_blk);
1987		phys = __is_valid_data_blkaddr(map.m_pblk) ?
1988			blks_to_bytes(inode, map.m_pblk) : 0;
1989		size = blks_to_bytes(inode, map.m_len);
1990		flags = 0;
1991
1992		if (compr_cluster) {
1993			flags = FIEMAP_EXTENT_ENCODED;
1994			count_in_cluster += map.m_len;
1995			if (count_in_cluster == cluster_size) {
1996				compr_cluster = false;
1997				size += blks_to_bytes(inode, 1);
1998			}
1999		} else if (map.m_flags & F2FS_MAP_UNWRITTEN) {
2000			flags = FIEMAP_EXTENT_UNWRITTEN;
2001		}
2002
2003		start_blk += bytes_to_blks(inode, size);
2004	}
2005
2006prep_next:
2007	cond_resched();
2008	if (fatal_signal_pending(current))
2009		ret = -EINTR;
2010	else
2011		goto next;
2012out:
2013	if (ret == 1)
2014		ret = 0;
2015
2016	inode_unlock(inode);
2017	return ret;
2018}
2019
2020static inline loff_t f2fs_readpage_limit(struct inode *inode)
2021{
2022	if (IS_ENABLED(CONFIG_FS_VERITY) &&
2023	    (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
2024		return inode->i_sb->s_maxbytes;
2025
2026	return i_size_read(inode);
2027}
2028
2029static int f2fs_read_single_page(struct inode *inode, struct page *page,
2030					unsigned nr_pages,
2031					struct f2fs_map_blocks *map,
2032					struct bio **bio_ret,
2033					sector_t *last_block_in_bio,
2034					bool is_readahead)
2035{
2036	struct bio *bio = *bio_ret;
2037	const unsigned blocksize = blks_to_bytes(inode, 1);
2038	sector_t block_in_file;
2039	sector_t last_block;
2040	sector_t last_block_in_file;
2041	sector_t block_nr;
2042	int ret = 0;
2043
2044	block_in_file = (sector_t)page_index(page);
2045	last_block = block_in_file + nr_pages;
2046	last_block_in_file = bytes_to_blks(inode,
2047			f2fs_readpage_limit(inode) + blocksize - 1);
2048	if (last_block > last_block_in_file)
2049		last_block = last_block_in_file;
2050
2051	/* just zeroing out page which is beyond EOF */
2052	if (block_in_file >= last_block)
2053		goto zero_out;
2054	/*
2055	 * Map blocks using the previous result first.
2056	 */
2057	if ((map->m_flags & F2FS_MAP_MAPPED) &&
2058			block_in_file > map->m_lblk &&
2059			block_in_file < (map->m_lblk + map->m_len))
2060		goto got_it;
2061
2062	/*
2063	 * Then do more f2fs_map_blocks() calls until we are
2064	 * done with this page.
2065	 */
2066	map->m_lblk = block_in_file;
2067	map->m_len = last_block - block_in_file;
2068
2069	ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2070	if (ret)
2071		goto out;
2072got_it:
2073	if ((map->m_flags & F2FS_MAP_MAPPED)) {
2074		block_nr = map->m_pblk + block_in_file - map->m_lblk;
2075		SetPageMappedToDisk(page);
2076
2077		if (!PageUptodate(page) && (!PageSwapCache(page) &&
2078					!cleancache_get_page(page))) {
2079			SetPageUptodate(page);
2080			goto confused;
2081		}
2082
2083		if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2084						DATA_GENERIC_ENHANCE_READ)) {
2085			ret = -EFSCORRUPTED;
2086			goto out;
2087		}
2088	} else {
2089zero_out:
2090		zero_user_segment(page, 0, PAGE_SIZE);
2091		if (f2fs_need_verity(inode, page->index) &&
2092		    !fsverity_verify_page(page)) {
2093			ret = -EIO;
2094			goto out;
2095		}
2096		if (!PageUptodate(page))
2097			SetPageUptodate(page);
2098		unlock_page(page);
2099		goto out;
2100	}
2101
2102	/*
2103	 * This page will go to BIO.  Do we need to send this
2104	 * BIO off first?
2105	 */
2106	if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2107				       *last_block_in_bio, block_nr) ||
2108		    !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2109submit_and_realloc:
2110		__submit_bio(F2FS_I_SB(inode), bio, DATA);
2111		bio = NULL;
2112	}
2113	if (bio == NULL) {
2114		bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2115				is_readahead ? REQ_RAHEAD : 0, page->index,
2116				false);
2117		if (IS_ERR(bio)) {
2118			ret = PTR_ERR(bio);
2119			bio = NULL;
2120			goto out;
2121		}
2122	}
2123
2124	/*
2125	 * If the page is under writeback, we need to wait for
2126	 * its completion to see the correct decrypted data.
2127	 */
2128	f2fs_wait_on_block_writeback(inode, block_nr);
2129
2130	if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2131		goto submit_and_realloc;
2132
2133	inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2134	f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2135	ClearPageError(page);
2136	*last_block_in_bio = block_nr;
2137	goto out;
2138confused:
2139	if (bio) {
2140		__submit_bio(F2FS_I_SB(inode), bio, DATA);
2141		bio = NULL;
2142	}
2143	unlock_page(page);
2144out:
2145	*bio_ret = bio;
2146	return ret;
2147}
2148
2149#ifdef CONFIG_F2FS_FS_COMPRESSION
2150int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2151				unsigned nr_pages, sector_t *last_block_in_bio,
2152				bool is_readahead, bool for_write)
2153{
2154	struct dnode_of_data dn;
2155	struct inode *inode = cc->inode;
2156	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2157	struct bio *bio = *bio_ret;
2158	unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2159	sector_t last_block_in_file;
2160	const unsigned blocksize = blks_to_bytes(inode, 1);
2161	struct decompress_io_ctx *dic = NULL;
2162	struct extent_info ei = {0, };
2163	bool from_dnode = true;
2164	int i;
2165	int ret = 0;
2166
2167	f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2168
2169	last_block_in_file = bytes_to_blks(inode,
2170			f2fs_readpage_limit(inode) + blocksize - 1);
2171
2172	/* get rid of pages beyond EOF */
2173	for (i = 0; i < cc->cluster_size; i++) {
2174		struct page *page = cc->rpages[i];
2175
2176		if (!page)
2177			continue;
2178		if ((sector_t)page->index >= last_block_in_file) {
2179			zero_user_segment(page, 0, PAGE_SIZE);
2180			if (!PageUptodate(page))
2181				SetPageUptodate(page);
2182		} else if (!PageUptodate(page)) {
2183			continue;
2184		}
2185		unlock_page(page);
2186		cc->rpages[i] = NULL;
2187		cc->nr_rpages--;
2188	}
2189
2190	/* we are done since all pages are beyond EOF */
2191	if (f2fs_cluster_is_empty(cc))
2192		goto out;
2193
2194	if (f2fs_lookup_extent_cache(inode, start_idx, &ei))
2195		from_dnode = false;
2196
2197	if (!from_dnode)
2198		goto skip_reading_dnode;
2199
2200	set_new_dnode(&dn, inode, NULL, NULL, 0);
2201	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2202	if (ret)
2203		goto out;
2204
2205	f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2206
2207skip_reading_dnode:
2208	for (i = 1; i < cc->cluster_size; i++) {
2209		block_t blkaddr;
2210
2211		blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2212					dn.ofs_in_node + i) :
2213					ei.blk + i - 1;
2214
2215		if (!__is_valid_data_blkaddr(blkaddr))
2216			break;
2217
2218		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2219			ret = -EFAULT;
2220			goto out_put_dnode;
2221		}
2222		cc->nr_cpages++;
2223
2224		if (!from_dnode && i >= ei.c_len)
2225			break;
2226	}
2227
2228	/* nothing to decompress */
2229	if (cc->nr_cpages == 0) {
2230		ret = 0;
2231		goto out_put_dnode;
2232	}
2233
2234	dic = f2fs_alloc_dic(cc);
2235	if (IS_ERR(dic)) {
2236		ret = PTR_ERR(dic);
2237		goto out_put_dnode;
2238	}
2239
2240	for (i = 0; i < cc->nr_cpages; i++) {
2241		struct page *page = dic->cpages[i];
2242		block_t blkaddr;
2243		struct bio_post_read_ctx *ctx;
2244
2245		blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2246					dn.ofs_in_node + i + 1) :
2247					ei.blk + i;
2248
2249		f2fs_wait_on_block_writeback(inode, blkaddr);
2250
2251		if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2252			if (atomic_dec_and_test(&dic->remaining_pages))
2253				f2fs_decompress_cluster(dic);
2254			continue;
2255		}
2256
2257		if (bio && (!page_is_mergeable(sbi, bio,
2258					*last_block_in_bio, blkaddr) ||
2259		    !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2260submit_and_realloc:
2261			__submit_bio(sbi, bio, DATA);
2262			bio = NULL;
2263		}
2264
2265		if (!bio) {
2266			bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2267					is_readahead ? REQ_RAHEAD : 0,
2268					page->index, for_write);
2269			if (IS_ERR(bio)) {
2270				ret = PTR_ERR(bio);
2271				f2fs_decompress_end_io(dic, ret);
2272				f2fs_put_dnode(&dn);
2273				*bio_ret = NULL;
2274				return ret;
2275			}
2276		}
2277
2278		if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2279			goto submit_and_realloc;
2280
2281		ctx = get_post_read_ctx(bio);
2282		ctx->enabled_steps |= STEP_DECOMPRESS;
2283		refcount_inc(&dic->refcnt);
2284
2285		inc_page_count(sbi, F2FS_RD_DATA);
2286		f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2287		f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2288		ClearPageError(page);
2289		*last_block_in_bio = blkaddr;
2290	}
2291
2292	if (from_dnode)
2293		f2fs_put_dnode(&dn);
2294
2295	*bio_ret = bio;
2296	return 0;
2297
2298out_put_dnode:
2299	if (from_dnode)
2300		f2fs_put_dnode(&dn);
2301out:
2302	for (i = 0; i < cc->cluster_size; i++) {
2303		if (cc->rpages[i]) {
2304			ClearPageUptodate(cc->rpages[i]);
2305			ClearPageError(cc->rpages[i]);
2306			unlock_page(cc->rpages[i]);
2307		}
2308	}
2309	*bio_ret = bio;
2310	return ret;
2311}
2312#endif
2313
2314/*
2315 * This function was originally taken from fs/mpage.c, and customized for f2fs.
2316 * Major change was from block_size == page_size in f2fs by default.
2317 */
2318static int f2fs_mpage_readpages(struct inode *inode,
2319		struct readahead_control *rac, struct page *page)
2320{
2321	struct bio *bio = NULL;
2322	sector_t last_block_in_bio = 0;
2323	struct f2fs_map_blocks map;
2324#ifdef CONFIG_F2FS_FS_COMPRESSION
2325	struct compress_ctx cc = {
2326		.inode = inode,
2327		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2328		.cluster_size = F2FS_I(inode)->i_cluster_size,
2329		.cluster_idx = NULL_CLUSTER,
2330		.rpages = NULL,
2331		.cpages = NULL,
2332		.nr_rpages = 0,
2333		.nr_cpages = 0,
2334	};
2335	pgoff_t nc_cluster_idx = NULL_CLUSTER;
2336#endif
2337	unsigned nr_pages = rac ? readahead_count(rac) : 1;
2338	unsigned max_nr_pages = nr_pages;
2339	int ret = 0;
2340
2341	map.m_pblk = 0;
2342	map.m_lblk = 0;
2343	map.m_len = 0;
2344	map.m_flags = 0;
2345	map.m_next_pgofs = NULL;
2346	map.m_next_extent = NULL;
2347	map.m_seg_type = NO_CHECK_TYPE;
2348	map.m_may_create = false;
2349
2350	for (; nr_pages; nr_pages--) {
2351		if (rac) {
2352			page = readahead_page(rac);
2353			prefetchw(&page->flags);
2354		}
2355
2356#ifdef CONFIG_F2FS_FS_COMPRESSION
2357		if (f2fs_compressed_file(inode)) {
2358			/* there are remained comressed pages, submit them */
2359			if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2360				ret = f2fs_read_multi_pages(&cc, &bio,
2361							max_nr_pages,
2362							&last_block_in_bio,
2363							rac != NULL, false);
2364				f2fs_destroy_compress_ctx(&cc, false);
2365				if (ret)
2366					goto set_error_page;
2367			}
2368			if (cc.cluster_idx == NULL_CLUSTER) {
2369				if (nc_cluster_idx ==
2370					page->index >> cc.log_cluster_size) {
2371					goto read_single_page;
2372				}
2373
2374				ret = f2fs_is_compressed_cluster(inode, page->index);
2375				if (ret < 0)
2376					goto set_error_page;
2377				else if (!ret) {
2378					nc_cluster_idx =
2379						page->index >> cc.log_cluster_size;
2380					goto read_single_page;
2381				}
2382
2383				nc_cluster_idx = NULL_CLUSTER;
2384			}
2385			ret = f2fs_init_compress_ctx(&cc);
2386			if (ret)
2387				goto set_error_page;
2388
2389			f2fs_compress_ctx_add_page(&cc, page);
2390
2391			goto next_page;
2392		}
2393read_single_page:
2394#endif
2395
2396		ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2397					&bio, &last_block_in_bio, rac);
2398		if (ret) {
2399#ifdef CONFIG_F2FS_FS_COMPRESSION
2400set_error_page:
2401#endif
2402			SetPageError(page);
2403			zero_user_segment(page, 0, PAGE_SIZE);
2404			unlock_page(page);
2405		}
2406#ifdef CONFIG_F2FS_FS_COMPRESSION
2407next_page:
2408#endif
2409		if (rac)
2410			put_page(page);
2411
2412#ifdef CONFIG_F2FS_FS_COMPRESSION
2413		if (f2fs_compressed_file(inode)) {
2414			/* last page */
2415			if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2416				ret = f2fs_read_multi_pages(&cc, &bio,
2417							max_nr_pages,
2418							&last_block_in_bio,
2419							rac != NULL, false);
2420				f2fs_destroy_compress_ctx(&cc, false);
2421			}
2422		}
2423#endif
2424	}
2425	if (bio)
2426		__submit_bio(F2FS_I_SB(inode), bio, DATA);
2427	return ret;
2428}
2429
2430static int f2fs_read_data_page(struct file *file, struct page *page)
2431{
2432	struct inode *inode = page_file_mapping(page)->host;
2433	int ret = -EAGAIN;
2434
2435	trace_f2fs_readpage(page, DATA);
2436
2437	if (!f2fs_is_compress_backend_ready(inode)) {
2438		unlock_page(page);
2439		return -EOPNOTSUPP;
2440	}
2441
2442	/* If the file has inline data, try to read it directly */
2443	if (f2fs_has_inline_data(inode))
2444		ret = f2fs_read_inline_data(inode, page);
2445	if (ret == -EAGAIN)
2446		ret = f2fs_mpage_readpages(inode, NULL, page);
2447	return ret;
2448}
2449
2450static void f2fs_readahead(struct readahead_control *rac)
2451{
2452	struct inode *inode = rac->mapping->host;
2453
2454	trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2455
2456	if (!f2fs_is_compress_backend_ready(inode))
2457		return;
2458
2459	/* If the file has inline data, skip readpages */
2460	if (f2fs_has_inline_data(inode))
2461		return;
2462
2463	f2fs_mpage_readpages(inode, rac, NULL);
2464}
2465
2466int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2467{
2468	struct inode *inode = fio->page->mapping->host;
2469	struct page *mpage, *page;
2470	gfp_t gfp_flags = GFP_NOFS;
2471
2472	if (!f2fs_encrypted_file(inode))
2473		return 0;
2474
2475	page = fio->compressed_page ? fio->compressed_page : fio->page;
2476
2477	/* wait for GCed page writeback via META_MAPPING */
2478	f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2479
2480	if (fscrypt_inode_uses_inline_crypto(inode))
2481		return 0;
2482
2483retry_encrypt:
2484	fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2485					PAGE_SIZE, 0, gfp_flags);
2486	if (IS_ERR(fio->encrypted_page)) {
2487		/* flush pending IOs and wait for a while in the ENOMEM case */
2488		if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2489			f2fs_flush_merged_writes(fio->sbi);
2490			congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2491			gfp_flags |= __GFP_NOFAIL;
2492			goto retry_encrypt;
2493		}
2494		return PTR_ERR(fio->encrypted_page);
2495	}
2496
2497	mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2498	if (mpage) {
2499		if (PageUptodate(mpage))
2500			memcpy(page_address(mpage),
2501				page_address(fio->encrypted_page), PAGE_SIZE);
2502		f2fs_put_page(mpage, 1);
2503	}
2504	return 0;
2505}
2506
2507static inline bool check_inplace_update_policy(struct inode *inode,
2508				struct f2fs_io_info *fio)
2509{
2510	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2511	unsigned int policy = SM_I(sbi)->ipu_policy;
2512
2513	if (policy & (0x1 << F2FS_IPU_FORCE))
2514		return true;
2515	if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2516		return true;
2517	if (policy & (0x1 << F2FS_IPU_UTIL) &&
2518			utilization(sbi) > SM_I(sbi)->min_ipu_util)
2519		return true;
2520	if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2521			utilization(sbi) > SM_I(sbi)->min_ipu_util)
2522		return true;
2523
2524	/*
2525	 * IPU for rewrite async pages
2526	 */
2527	if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2528			fio && fio->op == REQ_OP_WRITE &&
2529			!(fio->op_flags & REQ_SYNC) &&
2530			!IS_ENCRYPTED(inode))
2531		return true;
2532
2533	/* this is only set during fdatasync */
2534	if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2535			is_inode_flag_set(inode, FI_NEED_IPU))
2536		return true;
2537
2538	if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2539			!f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2540		return true;
2541
2542	return false;
2543}
2544
2545bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2546{
2547	/* swap file is migrating in aligned write mode */
2548	if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2549		return false;
2550
2551	if (f2fs_is_pinned_file(inode))
2552		return true;
2553
2554	/* if this is cold file, we should overwrite to avoid fragmentation */
2555	if (file_is_cold(inode))
2556		return true;
2557
2558	return check_inplace_update_policy(inode, fio);
2559}
2560
2561bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2562{
2563	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2564
2565	if (f2fs_lfs_mode(sbi))
2566		return true;
2567	if (S_ISDIR(inode->i_mode))
2568		return true;
2569	if (IS_NOQUOTA(inode))
2570		return true;
2571	if (f2fs_is_atomic_file(inode))
2572		return true;
2573	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2574		return true;
2575
2576	/* swap file is migrating in aligned write mode */
2577	if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2578		return true;
2579
2580	if (fio) {
2581		if (page_private_gcing(fio->page))
2582			return true;
2583		if (page_private_dummy(fio->page))
2584			return true;
2585		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2586			f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2587			return true;
2588	}
2589	return false;
2590}
2591
2592static inline bool need_inplace_update(struct f2fs_io_info *fio)
2593{
2594	struct inode *inode = fio->page->mapping->host;
2595
2596	if (f2fs_should_update_outplace(inode, fio))
2597		return false;
2598
2599	return f2fs_should_update_inplace(inode, fio);
2600}
2601
2602int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2603{
2604	struct page *page = fio->page;
2605	struct inode *inode = page->mapping->host;
2606	struct dnode_of_data dn;
2607	struct extent_info ei = {0, };
2608	struct node_info ni;
2609	bool ipu_force = false;
2610	int err = 0;
2611
2612	set_new_dnode(&dn, inode, NULL, NULL, 0);
2613	if (need_inplace_update(fio) &&
2614			f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2615		fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2616
2617		if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2618						DATA_GENERIC_ENHANCE))
2619			return -EFSCORRUPTED;
2620
2621		ipu_force = true;
2622		fio->need_lock = LOCK_DONE;
2623		goto got_it;
2624	}
2625
2626	/* Deadlock due to between page->lock and f2fs_lock_op */
2627	if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2628		return -EAGAIN;
2629
2630	err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2631	if (err)
2632		goto out;
2633
2634	fio->old_blkaddr = dn.data_blkaddr;
2635
2636	/* This page is already truncated */
2637	if (fio->old_blkaddr == NULL_ADDR) {
2638		ClearPageUptodate(page);
2639		clear_page_private_gcing(page);
2640		goto out_writepage;
2641	}
2642got_it:
2643	if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2644		!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2645						DATA_GENERIC_ENHANCE)) {
2646		err = -EFSCORRUPTED;
2647		goto out_writepage;
2648	}
2649	/*
2650	 * If current allocation needs SSR,
2651	 * it had better in-place writes for updated data.
2652	 */
2653	if (ipu_force ||
2654		(__is_valid_data_blkaddr(fio->old_blkaddr) &&
2655					need_inplace_update(fio))) {
2656		err = f2fs_encrypt_one_page(fio);
2657		if (err)
2658			goto out_writepage;
2659
2660		set_page_writeback(page);
2661		ClearPageError(page);
2662		f2fs_put_dnode(&dn);
2663		if (fio->need_lock == LOCK_REQ)
2664			f2fs_unlock_op(fio->sbi);
2665		err = f2fs_inplace_write_data(fio);
2666		if (err) {
2667			if (fscrypt_inode_uses_fs_layer_crypto(inode))
2668				fscrypt_finalize_bounce_page(&fio->encrypted_page);
2669			if (PageWriteback(page))
2670				end_page_writeback(page);
2671		} else {
2672			set_inode_flag(inode, FI_UPDATE_WRITE);
2673		}
2674		trace_f2fs_do_write_data_page(fio->page, IPU);
2675		return err;
2676	}
2677
2678	if (fio->need_lock == LOCK_RETRY) {
2679		if (!f2fs_trylock_op(fio->sbi)) {
2680			err = -EAGAIN;
2681			goto out_writepage;
2682		}
2683		fio->need_lock = LOCK_REQ;
2684	}
2685
2686	err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
2687	if (err)
2688		goto out_writepage;
2689
2690	fio->version = ni.version;
2691
2692	err = f2fs_encrypt_one_page(fio);
2693	if (err)
2694		goto out_writepage;
2695
2696	set_page_writeback(page);
2697	ClearPageError(page);
2698
2699	if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2700		f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2701
2702	/* LFS mode write path */
2703	f2fs_outplace_write_data(&dn, fio);
2704	trace_f2fs_do_write_data_page(page, OPU);
2705	set_inode_flag(inode, FI_APPEND_WRITE);
2706	if (page->index == 0)
2707		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2708out_writepage:
2709	f2fs_put_dnode(&dn);
2710out:
2711	if (fio->need_lock == LOCK_REQ)
2712		f2fs_unlock_op(fio->sbi);
2713	return err;
2714}
2715
2716int f2fs_write_single_data_page(struct page *page, int *submitted,
2717				struct bio **bio,
2718				sector_t *last_block,
2719				struct writeback_control *wbc,
2720				enum iostat_type io_type,
2721				int compr_blocks,
2722				bool allow_balance)
2723{
2724	struct inode *inode = page->mapping->host;
2725	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2726	loff_t i_size = i_size_read(inode);
2727	const pgoff_t end_index = ((unsigned long long)i_size)
2728							>> PAGE_SHIFT;
2729	loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2730	unsigned offset = 0;
2731	bool need_balance_fs = false;
2732	int err = 0;
2733	struct f2fs_io_info fio = {
2734		.sbi = sbi,
2735		.ino = inode->i_ino,
2736		.type = DATA,
2737		.op = REQ_OP_WRITE,
2738		.op_flags = wbc_to_write_flags(wbc),
2739		.old_blkaddr = NULL_ADDR,
2740		.page = page,
2741		.encrypted_page = NULL,
2742		.submitted = false,
2743		.compr_blocks = compr_blocks,
2744		.need_lock = LOCK_RETRY,
2745		.io_type = io_type,
2746		.io_wbc = wbc,
2747		.bio = bio,
2748		.last_block = last_block,
2749	};
2750
2751	trace_f2fs_writepage(page, DATA);
2752
2753	/* we should bypass data pages to proceed the kworkder jobs */
2754	if (unlikely(f2fs_cp_error(sbi))) {
2755		mapping_set_error(page->mapping, -EIO);
2756		/*
2757		 * don't drop any dirty dentry pages for keeping lastest
2758		 * directory structure.
2759		 */
2760		if (S_ISDIR(inode->i_mode))
2761			goto redirty_out;
2762		goto out;
2763	}
2764
2765	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2766		goto redirty_out;
2767
2768	if (page->index < end_index ||
2769			f2fs_verity_in_progress(inode) ||
2770			compr_blocks)
2771		goto write;
2772
2773	/*
2774	 * If the offset is out-of-range of file size,
2775	 * this page does not have to be written to disk.
2776	 */
2777	offset = i_size & (PAGE_SIZE - 1);
2778	if ((page->index >= end_index + 1) || !offset)
2779		goto out;
2780
2781	zero_user_segment(page, offset, PAGE_SIZE);
2782write:
2783	if (f2fs_is_drop_cache(inode))
2784		goto out;
2785	/* we should not write 0'th page having journal header */
2786	if (f2fs_is_volatile_file(inode) && (!page->index ||
2787			(!wbc->for_reclaim &&
2788			f2fs_available_free_memory(sbi, BASE_CHECK))))
2789		goto redirty_out;
2790
2791	/* Dentry/quota blocks are controlled by checkpoint */
2792	if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2793		/*
2794		 * We need to wait for node_write to avoid block allocation during
2795		 * checkpoint. This can only happen to quota writes which can cause
2796		 * the below discard race condition.
2797		 */
2798		if (IS_NOQUOTA(inode))
2799			down_read(&sbi->node_write);
2800
2801		fio.need_lock = LOCK_DONE;
2802		err = f2fs_do_write_data_page(&fio);
2803
2804		if (IS_NOQUOTA(inode))
2805			up_read(&sbi->node_write);
2806
2807		goto done;
2808	}
2809
2810	if (!wbc->for_reclaim)
2811		need_balance_fs = true;
2812	else if (has_not_enough_free_secs(sbi, 0, 0))
2813		goto redirty_out;
2814	else
2815		set_inode_flag(inode, FI_HOT_DATA);
2816
2817	err = -EAGAIN;
2818	if (f2fs_has_inline_data(inode)) {
2819		err = f2fs_write_inline_data(inode, page);
2820		if (!err)
2821			goto out;
2822	}
2823
2824	if (err == -EAGAIN) {
2825		err = f2fs_do_write_data_page(&fio);
2826		if (err == -EAGAIN) {
2827			fio.need_lock = LOCK_REQ;
2828			err = f2fs_do_write_data_page(&fio);
2829		}
2830	}
2831
2832	if (err) {
2833		file_set_keep_isize(inode);
2834	} else {
2835		spin_lock(&F2FS_I(inode)->i_size_lock);
2836		if (F2FS_I(inode)->last_disk_size < psize)
2837			F2FS_I(inode)->last_disk_size = psize;
2838		spin_unlock(&F2FS_I(inode)->i_size_lock);
2839	}
2840
2841done:
2842	if (err && err != -ENOENT)
2843		goto redirty_out;
2844
2845out:
2846	inode_dec_dirty_pages(inode);
2847	if (err) {
2848		ClearPageUptodate(page);
2849		clear_page_private_gcing(page);
2850	}
2851
2852	if (wbc->for_reclaim) {
2853		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2854		clear_inode_flag(inode, FI_HOT_DATA);
2855		f2fs_remove_dirty_inode(inode);
2856		submitted = NULL;
2857	}
2858	unlock_page(page);
2859	if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2860			!F2FS_I(inode)->cp_task && allow_balance)
2861		f2fs_balance_fs(sbi, need_balance_fs);
2862
2863	if (unlikely(f2fs_cp_error(sbi))) {
2864		f2fs_submit_merged_write(sbi, DATA);
2865		f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2866		submitted = NULL;
2867	}
2868
2869	if (submitted)
2870		*submitted = fio.submitted ? 1 : 0;
2871
2872	return 0;
2873
2874redirty_out:
2875	redirty_page_for_writepage(wbc, page);
2876	/*
2877	 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2878	 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2879	 * file_write_and_wait_range() will see EIO error, which is critical
2880	 * to return value of fsync() followed by atomic_write failure to user.
2881	 */
2882	if (!err || wbc->for_reclaim)
2883		return AOP_WRITEPAGE_ACTIVATE;
2884	unlock_page(page);
2885	return err;
2886}
2887
2888static int f2fs_write_data_page(struct page *page,
2889					struct writeback_control *wbc)
2890{
2891#ifdef CONFIG_F2FS_FS_COMPRESSION
2892	struct inode *inode = page->mapping->host;
2893
2894	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2895		goto out;
2896
2897	if (f2fs_compressed_file(inode)) {
2898		if (f2fs_is_compressed_cluster(inode, page->index)) {
2899			redirty_page_for_writepage(wbc, page);
2900			return AOP_WRITEPAGE_ACTIVATE;
2901		}
2902	}
2903out:
2904#endif
2905
2906	return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2907						wbc, FS_DATA_IO, 0, true);
2908}
2909
2910/*
2911 * This function was copied from write_cche_pages from mm/page-writeback.c.
2912 * The major change is making write step of cold data page separately from
2913 * warm/hot data page.
2914 */
2915static int f2fs_write_cache_pages(struct address_space *mapping,
2916					struct writeback_control *wbc,
2917					enum iostat_type io_type)
2918{
2919	int ret = 0;
2920	int done = 0, retry = 0;
2921	struct pagevec pvec;
2922	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2923	struct bio *bio = NULL;
2924	sector_t last_block;
2925#ifdef CONFIG_F2FS_FS_COMPRESSION
2926	struct inode *inode = mapping->host;
2927	struct compress_ctx cc = {
2928		.inode = inode,
2929		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2930		.cluster_size = F2FS_I(inode)->i_cluster_size,
2931		.cluster_idx = NULL_CLUSTER,
2932		.rpages = NULL,
2933		.nr_rpages = 0,
2934		.cpages = NULL,
2935		.rbuf = NULL,
2936		.cbuf = NULL,
2937		.rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2938		.private = NULL,
2939	};
2940#endif
2941	int nr_pages;
2942	pgoff_t index;
2943	pgoff_t end;		/* Inclusive */
2944	pgoff_t done_index;
2945	int range_whole = 0;
2946	xa_mark_t tag;
2947	int nwritten = 0;
2948	int submitted = 0;
2949	int i;
2950
2951	pagevec_init(&pvec);
2952
2953	if (get_dirty_pages(mapping->host) <=
2954				SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2955		set_inode_flag(mapping->host, FI_HOT_DATA);
2956	else
2957		clear_inode_flag(mapping->host, FI_HOT_DATA);
2958
2959	if (wbc->range_cyclic) {
2960		index = mapping->writeback_index; /* prev offset */
2961		end = -1;
2962	} else {
2963		index = wbc->range_start >> PAGE_SHIFT;
2964		end = wbc->range_end >> PAGE_SHIFT;
2965		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2966			range_whole = 1;
2967	}
2968	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2969		tag = PAGECACHE_TAG_TOWRITE;
2970	else
2971		tag = PAGECACHE_TAG_DIRTY;
2972retry:
2973	retry = 0;
2974	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2975		tag_pages_for_writeback(mapping, index, end);
2976	done_index = index;
2977	while (!done && !retry && (index <= end)) {
2978		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2979				tag);
2980		if (nr_pages == 0)
2981			break;
2982
2983		for (i = 0; i < nr_pages; i++) {
2984			struct page *page = pvec.pages[i];
2985			bool need_readd;
2986readd:
2987			need_readd = false;
2988#ifdef CONFIG_F2FS_FS_COMPRESSION
2989			if (f2fs_compressed_file(inode)) {
2990				ret = f2fs_init_compress_ctx(&cc);
2991				if (ret) {
2992					done = 1;
2993					break;
2994				}
2995
2996				if (!f2fs_cluster_can_merge_page(&cc,
2997								page->index)) {
2998					ret = f2fs_write_multi_pages(&cc,
2999						&submitted, wbc, io_type);
3000					if (!ret)
3001						need_readd = true;
3002					goto result;
3003				}
3004
3005				if (unlikely(f2fs_cp_error(sbi)))
3006					goto lock_page;
3007
3008				if (f2fs_cluster_is_empty(&cc)) {
3009					void *fsdata = NULL;
3010					struct page *pagep;
3011					int ret2;
3012
3013					ret2 = f2fs_prepare_compress_overwrite(
3014							inode, &pagep,
3015							page->index, &fsdata);
3016					if (ret2 < 0) {
3017						ret = ret2;
3018						done = 1;
3019						break;
3020					} else if (ret2 &&
3021						!f2fs_compress_write_end(inode,
3022								fsdata, page->index,
3023								1)) {
3024						retry = 1;
3025						break;
3026					}
3027				} else {
3028					goto lock_page;
3029				}
3030			}
3031#endif
3032			/* give a priority to WB_SYNC threads */
3033			if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3034					wbc->sync_mode == WB_SYNC_NONE) {
3035				done = 1;
3036				break;
3037			}
3038#ifdef CONFIG_F2FS_FS_COMPRESSION
3039lock_page:
3040#endif
3041			done_index = page->index;
3042retry_write:
3043			lock_page(page);
3044
3045			if (unlikely(page->mapping != mapping)) {
3046continue_unlock:
3047				unlock_page(page);
3048				continue;
3049			}
3050
3051			if (!PageDirty(page)) {
3052				/* someone wrote it for us */
3053				goto continue_unlock;
3054			}
3055
3056			if (PageWriteback(page)) {
3057				if (wbc->sync_mode != WB_SYNC_NONE)
3058					f2fs_wait_on_page_writeback(page,
3059							DATA, true, true);
3060				else
3061					goto continue_unlock;
3062			}
3063
3064			if (!clear_page_dirty_for_io(page))
3065				goto continue_unlock;
3066
3067#ifdef CONFIG_F2FS_FS_COMPRESSION
3068			if (f2fs_compressed_file(inode)) {
3069				get_page(page);
3070				f2fs_compress_ctx_add_page(&cc, page);
3071				continue;
3072			}
3073#endif
3074			ret = f2fs_write_single_data_page(page, &submitted,
3075					&bio, &last_block, wbc, io_type,
3076					0, true);
3077			if (ret == AOP_WRITEPAGE_ACTIVATE)
3078				unlock_page(page);
3079#ifdef CONFIG_F2FS_FS_COMPRESSION
3080result:
3081#endif
3082			nwritten += submitted;
3083			wbc->nr_to_write -= submitted;
3084
3085			if (unlikely(ret)) {
3086				/*
3087				 * keep nr_to_write, since vfs uses this to
3088				 * get # of written pages.
3089				 */
3090				if (ret == AOP_WRITEPAGE_ACTIVATE) {
3091					ret = 0;
3092					goto next;
3093				} else if (ret == -EAGAIN) {
3094					ret = 0;
3095					if (wbc->sync_mode == WB_SYNC_ALL) {
3096						cond_resched();
3097						congestion_wait(BLK_RW_ASYNC,
3098							DEFAULT_IO_TIMEOUT);
3099						goto retry_write;
3100					}
3101					goto next;
3102				}
3103				done_index = page->index + 1;
3104				done = 1;
3105				break;
3106			}
3107
3108			if (wbc->nr_to_write <= 0 &&
3109					wbc->sync_mode == WB_SYNC_NONE) {
3110				done = 1;
3111				break;
3112			}
3113next:
3114			if (need_readd)
3115				goto readd;
3116		}
3117		pagevec_release(&pvec);
3118		cond_resched();
3119	}
3120#ifdef CONFIG_F2FS_FS_COMPRESSION
3121	/* flush remained pages in compress cluster */
3122	if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3123		ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3124		nwritten += submitted;
3125		wbc->nr_to_write -= submitted;
3126		if (ret) {
3127			done = 1;
3128			retry = 0;
3129		}
3130	}
3131	if (f2fs_compressed_file(inode))
3132		f2fs_destroy_compress_ctx(&cc, false);
3133#endif
3134	if (retry) {
3135		index = 0;
3136		end = -1;
3137		goto retry;
3138	}
3139	if (wbc->range_cyclic && !done)
3140		done_index = 0;
3141	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3142		mapping->writeback_index = done_index;
3143
3144	if (nwritten)
3145		f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3146								NULL, 0, DATA);
3147	/* submit cached bio of IPU write */
3148	if (bio)
3149		f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3150
3151	return ret;
3152}
3153
3154static inline bool __should_serialize_io(struct inode *inode,
3155					struct writeback_control *wbc)
3156{
3157	/* to avoid deadlock in path of data flush */
3158	if (F2FS_I(inode)->cp_task)
3159		return false;
3160
3161	if (!S_ISREG(inode->i_mode))
3162		return false;
3163	if (IS_NOQUOTA(inode))
3164		return false;
3165
3166	if (f2fs_need_compress_data(inode))
3167		return true;
3168	if (wbc->sync_mode != WB_SYNC_ALL)
3169		return true;
3170	if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3171		return true;
3172	return false;
3173}
3174
3175static int __f2fs_write_data_pages(struct address_space *mapping,
3176						struct writeback_control *wbc,
3177						enum iostat_type io_type)
3178{
3179	struct inode *inode = mapping->host;
3180	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3181	struct blk_plug plug;
3182	int ret;
3183	bool locked = false;
3184
3185	/* deal with chardevs and other special file */
3186	if (!mapping->a_ops->writepage)
3187		return 0;
3188
3189	/* skip writing if there is no dirty page in this inode */
3190	if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3191		return 0;
3192
3193	/* during POR, we don't need to trigger writepage at all. */
3194	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3195		goto skip_write;
3196
3197	if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3198			wbc->sync_mode == WB_SYNC_NONE &&
3199			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3200			f2fs_available_free_memory(sbi, DIRTY_DENTS))
3201		goto skip_write;
3202
3203	/* skip writing during file defragment */
3204	if (is_inode_flag_set(inode, FI_DO_DEFRAG))
3205		goto skip_write;
3206
3207	trace_f2fs_writepages(mapping->host, wbc, DATA);
3208
3209	/* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3210	if (wbc->sync_mode == WB_SYNC_ALL)
3211		atomic_inc(&sbi->wb_sync_req[DATA]);
3212	else if (atomic_read(&sbi->wb_sync_req[DATA]))
3213		goto skip_write;
3214
3215	if (__should_serialize_io(inode, wbc)) {
3216		mutex_lock(&sbi->writepages);
3217		locked = true;
3218	}
3219
3220	blk_start_plug(&plug);
3221	ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3222	blk_finish_plug(&plug);
3223
3224	if (locked)
3225		mutex_unlock(&sbi->writepages);
3226
3227	if (wbc->sync_mode == WB_SYNC_ALL)
3228		atomic_dec(&sbi->wb_sync_req[DATA]);
3229	/*
3230	 * if some pages were truncated, we cannot guarantee its mapping->host
3231	 * to detect pending bios.
3232	 */
3233
3234	f2fs_remove_dirty_inode(inode);
3235	return ret;
3236
3237skip_write:
3238	wbc->pages_skipped += get_dirty_pages(inode);
3239	trace_f2fs_writepages(mapping->host, wbc, DATA);
3240	return 0;
3241}
3242
3243static int f2fs_write_data_pages(struct address_space *mapping,
3244			    struct writeback_control *wbc)
3245{
3246	struct inode *inode = mapping->host;
3247
3248	return __f2fs_write_data_pages(mapping, wbc,
3249			F2FS_I(inode)->cp_task == current ?
3250			FS_CP_DATA_IO : FS_DATA_IO);
3251}
3252
3253static void f2fs_write_failed(struct inode *inode, loff_t to)
3254{
3255	loff_t i_size = i_size_read(inode);
3256
3257	if (IS_NOQUOTA(inode))
3258		return;
3259
3260	/* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3261	if (to > i_size && !f2fs_verity_in_progress(inode)) {
3262		down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3263		down_write(&F2FS_I(inode)->i_mmap_sem);
3264
3265		truncate_pagecache(inode, i_size);
3266		f2fs_truncate_blocks(inode, i_size, true);
3267
3268		up_write(&F2FS_I(inode)->i_mmap_sem);
3269		up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3270	}
3271}
3272
3273static int prepare_write_begin(struct f2fs_sb_info *sbi,
3274			struct page *page, loff_t pos, unsigned len,
3275			block_t *blk_addr, bool *node_changed)
3276{
3277	struct inode *inode = page->mapping->host;
3278	pgoff_t index = page->index;
3279	struct dnode_of_data dn;
3280	struct page *ipage;
3281	bool locked = false;
3282	struct extent_info ei = {0, };
3283	int err = 0;
3284	int flag;
3285
3286	/*
3287	 * we already allocated all the blocks, so we don't need to get
3288	 * the block addresses when there is no need to fill the page.
3289	 */
3290	if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
3291	    !is_inode_flag_set(inode, FI_NO_PREALLOC) &&
3292	    !f2fs_verity_in_progress(inode))
3293		return 0;
3294
3295	/* f2fs_lock_op avoids race between write CP and convert_inline_page */
3296	if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3297		flag = F2FS_GET_BLOCK_DEFAULT;
3298	else
3299		flag = F2FS_GET_BLOCK_PRE_AIO;
3300
3301	if (f2fs_has_inline_data(inode) ||
3302			(pos & PAGE_MASK) >= i_size_read(inode)) {
3303		f2fs_do_map_lock(sbi, flag, true);
3304		locked = true;
3305	}
3306
3307restart:
3308	/* check inline_data */
3309	ipage = f2fs_get_node_page(sbi, inode->i_ino);
3310	if (IS_ERR(ipage)) {
3311		err = PTR_ERR(ipage);
3312		goto unlock_out;
3313	}
3314
3315	set_new_dnode(&dn, inode, ipage, ipage, 0);
3316
3317	if (f2fs_has_inline_data(inode)) {
3318		if (pos + len <= MAX_INLINE_DATA(inode)) {
3319			f2fs_do_read_inline_data(page, ipage);
3320			set_inode_flag(inode, FI_DATA_EXIST);
3321			if (inode->i_nlink)
3322				set_page_private_inline(ipage);
3323		} else {
3324			err = f2fs_convert_inline_page(&dn, page);
3325			if (err)
3326				goto out;
3327			if (dn.data_blkaddr == NULL_ADDR)
3328				err = f2fs_get_block(&dn, index);
3329		}
3330	} else if (locked) {
3331		err = f2fs_get_block(&dn, index);
3332	} else {
3333		if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3334			dn.data_blkaddr = ei.blk + index - ei.fofs;
3335		} else {
3336			/* hole case */
3337			err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3338			if (err || dn.data_blkaddr == NULL_ADDR) {
3339				f2fs_put_dnode(&dn);
3340				f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3341								true);
3342				WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3343				locked = true;
3344				goto restart;
3345			}
3346		}
3347	}
3348
3349	/* convert_inline_page can make node_changed */
3350	*blk_addr = dn.data_blkaddr;
3351	*node_changed = dn.node_changed;
3352out:
3353	f2fs_put_dnode(&dn);
3354unlock_out:
3355	if (locked)
3356		f2fs_do_map_lock(sbi, flag, false);
3357	return err;
3358}
3359
3360static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3361		loff_t pos, unsigned len, unsigned flags,
3362		struct page **pagep, void **fsdata)
3363{
3364	struct inode *inode = mapping->host;
3365	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3366	struct page *page = NULL;
3367	pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3368	bool need_balance = false, drop_atomic = false;
3369	block_t blkaddr = NULL_ADDR;
3370	int err = 0;
3371
3372	trace_f2fs_write_begin(inode, pos, len, flags);
3373
3374	if (!f2fs_is_checkpoint_ready(sbi)) {
3375		err = -ENOSPC;
3376		goto fail;
3377	}
3378
3379	if ((f2fs_is_atomic_file(inode) &&
3380			!f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3381			is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3382		err = -ENOMEM;
3383		drop_atomic = true;
3384		goto fail;
3385	}
3386
3387	/*
3388	 * We should check this at this moment to avoid deadlock on inode page
3389	 * and #0 page. The locking rule for inline_data conversion should be:
3390	 * lock_page(page #0) -> lock_page(inode_page)
3391	 */
3392	if (index != 0) {
3393		err = f2fs_convert_inline_inode(inode);
3394		if (err)
3395			goto fail;
3396	}
3397
3398#ifdef CONFIG_F2FS_FS_COMPRESSION
3399	if (f2fs_compressed_file(inode)) {
3400		int ret;
3401
3402		*fsdata = NULL;
3403
3404		if (len == PAGE_SIZE)
3405			goto repeat;
3406
3407		ret = f2fs_prepare_compress_overwrite(inode, pagep,
3408							index, fsdata);
3409		if (ret < 0) {
3410			err = ret;
3411			goto fail;
3412		} else if (ret) {
3413			return 0;
3414		}
3415	}
3416#endif
3417
3418repeat:
3419	/*
3420	 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3421	 * wait_for_stable_page. Will wait that below with our IO control.
3422	 */
3423	page = f2fs_pagecache_get_page(mapping, index,
3424				FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3425	if (!page) {
3426		err = -ENOMEM;
3427		goto fail;
3428	}
3429
3430	/* TODO: cluster can be compressed due to race with .writepage */
3431
3432	*pagep = page;
3433
3434	err = prepare_write_begin(sbi, page, pos, len,
3435					&blkaddr, &need_balance);
3436	if (err)
3437		goto fail;
3438
3439	if (need_balance && !IS_NOQUOTA(inode) &&
3440			has_not_enough_free_secs(sbi, 0, 0)) {
3441		unlock_page(page);
3442		f2fs_balance_fs(sbi, true);
3443		lock_page(page);
3444		if (page->mapping != mapping) {
3445			/* The page got truncated from under us */
3446			f2fs_put_page(page, 1);
3447			goto repeat;
3448		}
3449	}
3450
3451	f2fs_wait_on_page_writeback(page, DATA, false, true);
3452
3453	if (len == PAGE_SIZE || PageUptodate(page))
3454		return 0;
3455
3456	if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3457	    !f2fs_verity_in_progress(inode)) {
3458		zero_user_segment(page, len, PAGE_SIZE);
3459		return 0;
3460	}
3461
3462	if (blkaddr == NEW_ADDR) {
3463		zero_user_segment(page, 0, PAGE_SIZE);
3464		SetPageUptodate(page);
3465	} else {
3466		if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3467				DATA_GENERIC_ENHANCE_READ)) {
3468			err = -EFSCORRUPTED;
3469			goto fail;
3470		}
3471		err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3472		if (err)
3473			goto fail;
3474
3475		lock_page(page);
3476		if (unlikely(page->mapping != mapping)) {
3477			f2fs_put_page(page, 1);
3478			goto repeat;
3479		}
3480		if (unlikely(!PageUptodate(page))) {
3481			err = -EIO;
3482			goto fail;
3483		}
3484	}
3485	return 0;
3486
3487fail:
3488	f2fs_put_page(page, 1);
3489	f2fs_write_failed(inode, pos + len);
3490	if (drop_atomic)
3491		f2fs_drop_inmem_pages_all(sbi, false);
3492	return err;
3493}
3494
3495static int f2fs_write_end(struct file *file,
3496			struct address_space *mapping,
3497			loff_t pos, unsigned len, unsigned copied,
3498			struct page *page, void *fsdata)
3499{
3500	struct inode *inode = page->mapping->host;
3501
3502	trace_f2fs_write_end(inode, pos, len, copied);
3503
3504	/*
3505	 * This should be come from len == PAGE_SIZE, and we expect copied
3506	 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3507	 * let generic_perform_write() try to copy data again through copied=0.
3508	 */
3509	if (!PageUptodate(page)) {
3510		if (unlikely(copied != len))
3511			copied = 0;
3512		else
3513			SetPageUptodate(page);
3514	}
3515
3516#ifdef CONFIG_F2FS_FS_COMPRESSION
3517	/* overwrite compressed file */
3518	if (f2fs_compressed_file(inode) && fsdata) {
3519		f2fs_compress_write_end(inode, fsdata, page->index, copied);
3520		f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3521
3522		if (pos + copied > i_size_read(inode) &&
3523				!f2fs_verity_in_progress(inode))
3524			f2fs_i_size_write(inode, pos + copied);
3525		return copied;
3526	}
3527#endif
3528
3529	if (!copied)
3530		goto unlock_out;
3531
3532	set_page_dirty(page);
3533
3534	if (pos + copied > i_size_read(inode) &&
3535	    !f2fs_verity_in_progress(inode))
3536		f2fs_i_size_write(inode, pos + copied);
3537unlock_out:
3538	f2fs_put_page(page, 1);
3539	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3540	return copied;
3541}
3542
3543static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
3544			   loff_t offset)
3545{
3546	unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
3547	unsigned blkbits = i_blkbits;
3548	unsigned blocksize_mask = (1 << blkbits) - 1;
3549	unsigned long align = offset | iov_iter_alignment(iter);
3550	struct block_device *bdev = inode->i_sb->s_bdev;
3551
3552	if (iov_iter_rw(iter) == READ && offset >= i_size_read(inode))
3553		return 1;
3554
3555	if (align & blocksize_mask) {
3556		if (bdev)
3557			blkbits = blksize_bits(bdev_logical_block_size(bdev));
3558		blocksize_mask = (1 << blkbits) - 1;
3559		if (align & blocksize_mask)
3560			return -EINVAL;
3561		return 1;
3562	}
3563	return 0;
3564}
3565
3566static void f2fs_dio_end_io(struct bio *bio)
3567{
3568	struct f2fs_private_dio *dio = bio->bi_private;
3569
3570	dec_page_count(F2FS_I_SB(dio->inode),
3571			dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3572
3573	bio->bi_private = dio->orig_private;
3574	bio->bi_end_io = dio->orig_end_io;
3575
3576	kfree(dio);
3577
3578	bio_endio(bio);
3579}
3580
3581static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
3582							loff_t file_offset)
3583{
3584	struct f2fs_private_dio *dio;
3585	bool write = (bio_op(bio) == REQ_OP_WRITE);
3586
3587	dio = f2fs_kzalloc(F2FS_I_SB(inode),
3588			sizeof(struct f2fs_private_dio), GFP_NOFS);
3589	if (!dio)
3590		goto out;
3591
3592	dio->inode = inode;
3593	dio->orig_end_io = bio->bi_end_io;
3594	dio->orig_private = bio->bi_private;
3595	dio->write = write;
3596
3597	bio->bi_end_io = f2fs_dio_end_io;
3598	bio->bi_private = dio;
3599
3600	inc_page_count(F2FS_I_SB(inode),
3601			write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3602
3603	submit_bio(bio);
3604	return;
3605out:
3606	bio->bi_status = BLK_STS_IOERR;
3607	bio_endio(bio);
3608}
3609
3610static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3611{
3612	struct address_space *mapping = iocb->ki_filp->f_mapping;
3613	struct inode *inode = mapping->host;
3614	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3615	struct f2fs_inode_info *fi = F2FS_I(inode);
3616	size_t count = iov_iter_count(iter);
3617	loff_t offset = iocb->ki_pos;
3618	int rw = iov_iter_rw(iter);
3619	int err;
3620	enum rw_hint hint = iocb->ki_hint;
3621	int whint_mode = F2FS_OPTION(sbi).whint_mode;
3622	bool do_opu;
3623
3624	err = check_direct_IO(inode, iter, offset);
3625	if (err)
3626		return err < 0 ? err : 0;
3627
3628	if (f2fs_force_buffered_io(inode, iocb, iter))
3629		return 0;
3630
3631	do_opu = rw == WRITE && f2fs_lfs_mode(sbi);
3632
3633	trace_f2fs_direct_IO_enter(inode, offset, count, rw);
3634
3635	if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
3636		iocb->ki_hint = WRITE_LIFE_NOT_SET;
3637
3638	if (iocb->ki_flags & IOCB_NOWAIT) {
3639		if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
3640			iocb->ki_hint = hint;
3641			err = -EAGAIN;
3642			goto out;
3643		}
3644		if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
3645			up_read(&fi->i_gc_rwsem[rw]);
3646			iocb->ki_hint = hint;
3647			err = -EAGAIN;
3648			goto out;
3649		}
3650	} else {
3651		down_read(&fi->i_gc_rwsem[rw]);
3652		if (do_opu)
3653			down_read(&fi->i_gc_rwsem[READ]);
3654	}
3655
3656	err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3657			iter, rw == WRITE ? get_data_block_dio_write :
3658			get_data_block_dio, NULL, f2fs_dio_submit_bio,
3659			rw == WRITE ? DIO_LOCKING | DIO_SKIP_HOLES :
3660			DIO_SKIP_HOLES);
3661
3662	if (do_opu)
3663		up_read(&fi->i_gc_rwsem[READ]);
3664
3665	up_read(&fi->i_gc_rwsem[rw]);
3666
3667	if (rw == WRITE) {
3668		if (whint_mode == WHINT_MODE_OFF)
3669			iocb->ki_hint = hint;
3670		if (err > 0) {
3671			f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3672									err);
3673			if (!do_opu)
3674				set_inode_flag(inode, FI_UPDATE_WRITE);
3675		} else if (err == -EIOCBQUEUED) {
3676			f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3677						count - iov_iter_count(iter));
3678		} else if (err < 0) {
3679			f2fs_write_failed(inode, offset + count);
3680		}
3681	} else {
3682		if (err > 0)
3683			f2fs_update_iostat(sbi, APP_DIRECT_READ_IO, err);
3684		else if (err == -EIOCBQUEUED)
3685			f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_READ_IO,
3686						count - iov_iter_count(iter));
3687	}
3688
3689out:
3690	trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
3691
3692	return err;
3693}
3694
3695void f2fs_invalidate_page(struct page *page, unsigned int offset,
3696							unsigned int length)
3697{
3698	struct inode *inode = page->mapping->host;
3699	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3700
3701	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3702		(offset % PAGE_SIZE || length != PAGE_SIZE))
3703		return;
3704
3705	if (PageDirty(page)) {
3706		if (inode->i_ino == F2FS_META_INO(sbi)) {
3707			dec_page_count(sbi, F2FS_DIRTY_META);
3708		} else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3709			dec_page_count(sbi, F2FS_DIRTY_NODES);
3710		} else {
3711			inode_dec_dirty_pages(inode);
3712			f2fs_remove_dirty_inode(inode);
3713		}
3714	}
3715
3716	clear_page_private_gcing(page);
3717
3718	if (test_opt(sbi, COMPRESS_CACHE)) {
3719		if (f2fs_compressed_file(inode))
3720			f2fs_invalidate_compress_pages(sbi, inode->i_ino);
3721		if (inode->i_ino == F2FS_COMPRESS_INO(sbi))
3722			clear_page_private_data(page);
3723	}
3724
3725	if (page_private_atomic(page))
3726		return f2fs_drop_inmem_page(inode, page);
3727
3728	detach_page_private(page);
3729	set_page_private(page, 0);
3730}
3731
3732int f2fs_release_page(struct page *page, gfp_t wait)
3733{
3734	/* If this is dirty page, keep PagePrivate */
3735	if (PageDirty(page))
3736		return 0;
3737
3738	/* This is atomic written page, keep Private */
3739	if (page_private_atomic(page))
3740		return 0;
3741
3742	if (test_opt(F2FS_P_SB(page), COMPRESS_CACHE)) {
3743		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3744		struct inode *inode = page->mapping->host;
3745
3746		if (f2fs_compressed_file(inode))
3747			f2fs_invalidate_compress_pages(sbi, inode->i_ino);
3748		if (inode->i_ino == F2FS_COMPRESS_INO(sbi))
3749			clear_page_private_data(page);
3750	}
3751
3752	clear_page_private_gcing(page);
3753
3754	detach_page_private(page);
3755	set_page_private(page, 0);
3756	return 1;
3757}
3758
3759static int f2fs_set_data_page_dirty(struct page *page)
3760{
3761	struct inode *inode = page_file_mapping(page)->host;
3762
3763	trace_f2fs_set_page_dirty(page, DATA);
3764
3765	if (!PageUptodate(page))
3766		SetPageUptodate(page);
3767	if (PageSwapCache(page))
3768		return __set_page_dirty_nobuffers(page);
3769
3770	if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3771		if (!page_private_atomic(page)) {
3772			f2fs_register_inmem_page(inode, page);
3773			return 1;
3774		}
3775		/*
3776		 * Previously, this page has been registered, we just
3777		 * return here.
3778		 */
3779		return 0;
3780	}
3781
3782	if (!PageDirty(page)) {
3783		__set_page_dirty_nobuffers(page);
3784		f2fs_update_dirty_page(inode, page);
3785		return 1;
3786	}
3787	return 0;
3788}
3789
3790
3791static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3792{
3793#ifdef CONFIG_F2FS_FS_COMPRESSION
3794	struct dnode_of_data dn;
3795	sector_t start_idx, blknr = 0;
3796	int ret;
3797
3798	start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3799
3800	set_new_dnode(&dn, inode, NULL, NULL, 0);
3801	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3802	if (ret)
3803		return 0;
3804
3805	if (dn.data_blkaddr != COMPRESS_ADDR) {
3806		dn.ofs_in_node += block - start_idx;
3807		blknr = f2fs_data_blkaddr(&dn);
3808		if (!__is_valid_data_blkaddr(blknr))
3809			blknr = 0;
3810	}
3811
3812	f2fs_put_dnode(&dn);
3813	return blknr;
3814#else
3815	return 0;
3816#endif
3817}
3818
3819
3820static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3821{
3822	struct inode *inode = mapping->host;
3823	sector_t blknr = 0;
3824
3825	if (f2fs_has_inline_data(inode))
3826		goto out;
3827
3828	/* make sure allocating whole blocks */
3829	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3830		filemap_write_and_wait(mapping);
3831
3832	/* Block number less than F2FS MAX BLOCKS */
3833	if (unlikely(block >= max_file_blocks(inode)))
3834		goto out;
3835
3836	if (f2fs_compressed_file(inode)) {
3837		blknr = f2fs_bmap_compress(inode, block);
3838	} else {
3839		struct f2fs_map_blocks map;
3840
3841		memset(&map, 0, sizeof(map));
3842		map.m_lblk = block;
3843		map.m_len = 1;
3844		map.m_next_pgofs = NULL;
3845		map.m_seg_type = NO_CHECK_TYPE;
3846
3847		if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3848			blknr = map.m_pblk;
3849	}
3850out:
3851	trace_f2fs_bmap(inode, block, blknr);
3852	return blknr;
3853}
3854
3855#ifdef CONFIG_MIGRATION
3856#include <linux/migrate.h>
3857
3858int f2fs_migrate_page(struct address_space *mapping,
3859		struct page *newpage, struct page *page, enum migrate_mode mode)
3860{
3861	int rc, extra_count;
3862	struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3863	bool atomic_written = page_private_atomic(page);
3864
3865	BUG_ON(PageWriteback(page));
3866
3867	/* migrating an atomic written page is safe with the inmem_lock hold */
3868	if (atomic_written) {
3869		if (mode != MIGRATE_SYNC)
3870			return -EBUSY;
3871		if (!mutex_trylock(&fi->inmem_lock))
3872			return -EAGAIN;
3873	}
3874
3875	/* one extra reference was held for atomic_write page */
3876	extra_count = atomic_written ? 1 : 0;
3877	rc = migrate_page_move_mapping(mapping, newpage,
3878				page, extra_count);
3879	if (rc != MIGRATEPAGE_SUCCESS) {
3880		if (atomic_written)
3881			mutex_unlock(&fi->inmem_lock);
3882		return rc;
3883	}
3884
3885	if (atomic_written) {
3886		struct inmem_pages *cur;
3887
3888		list_for_each_entry(cur, &fi->inmem_pages, list)
3889			if (cur->page == page) {
3890				cur->page = newpage;
3891				break;
3892			}
3893		mutex_unlock(&fi->inmem_lock);
3894		put_page(page);
3895		get_page(newpage);
3896	}
3897
3898	/* guarantee to start from no stale private field */
3899	set_page_private(newpage, 0);
3900	if (PagePrivate(page)) {
3901		set_page_private(newpage, page_private(page));
3902		SetPagePrivate(newpage);
3903		get_page(newpage);
3904
3905		set_page_private(page, 0);
3906		ClearPagePrivate(page);
3907		put_page(page);
3908	}
3909
3910	if (mode != MIGRATE_SYNC_NO_COPY)
3911		migrate_page_copy(newpage, page);
3912	else
3913		migrate_page_states(newpage, page);
3914
3915	return MIGRATEPAGE_SUCCESS;
3916}
3917#endif
3918
3919#ifdef CONFIG_SWAP
3920static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3921							unsigned int blkcnt)
3922{
3923	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3924	unsigned int blkofs;
3925	unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3926	unsigned int secidx = start_blk / blk_per_sec;
3927	unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3928	int ret = 0;
3929
3930	down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3931	down_write(&F2FS_I(inode)->i_mmap_sem);
3932
3933	set_inode_flag(inode, FI_ALIGNED_WRITE);
3934
3935	for (; secidx < end_sec; secidx++) {
3936		down_write(&sbi->pin_sem);
3937
3938		f2fs_lock_op(sbi);
3939		f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3940		f2fs_unlock_op(sbi);
3941
3942		set_inode_flag(inode, FI_DO_DEFRAG);
3943
3944		for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3945			struct page *page;
3946			unsigned int blkidx = secidx * blk_per_sec + blkofs;
3947
3948			page = f2fs_get_lock_data_page(inode, blkidx, true);
3949			if (IS_ERR(page)) {
3950				up_write(&sbi->pin_sem);
3951				ret = PTR_ERR(page);
3952				goto done;
3953			}
3954
3955			set_page_dirty(page);
3956			f2fs_put_page(page, 1);
3957		}
3958
3959		clear_inode_flag(inode, FI_DO_DEFRAG);
3960
3961		ret = filemap_fdatawrite(inode->i_mapping);
3962
3963		up_write(&sbi->pin_sem);
3964
3965		if (ret)
3966			break;
3967	}
3968
3969done:
3970	clear_inode_flag(inode, FI_DO_DEFRAG);
3971	clear_inode_flag(inode, FI_ALIGNED_WRITE);
3972
3973	up_write(&F2FS_I(inode)->i_mmap_sem);
3974	up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3975
3976	return ret;
3977}
3978
3979static int check_swap_activate(struct swap_info_struct *sis,
3980				struct file *swap_file, sector_t *span)
3981{
3982	struct address_space *mapping = swap_file->f_mapping;
3983	struct inode *inode = mapping->host;
3984	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3985	sector_t cur_lblock;
3986	sector_t last_lblock;
3987	sector_t pblock;
3988	sector_t lowest_pblock = -1;
3989	sector_t highest_pblock = 0;
3990	int nr_extents = 0;
3991	unsigned long nr_pblocks;
3992	unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3993	unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3994	unsigned int not_aligned = 0;
3995	int ret = 0;
3996
3997	/*
3998	 * Map all the blocks into the extent list.  This code doesn't try
3999	 * to be very smart.
4000	 */
4001	cur_lblock = 0;
4002	last_lblock = bytes_to_blks(inode, i_size_read(inode));
4003
4004	while (cur_lblock < last_lblock && cur_lblock < sis->max) {
4005		struct f2fs_map_blocks map;
4006retry:
4007		cond_resched();
4008
4009		memset(&map, 0, sizeof(map));
4010		map.m_lblk = cur_lblock;
4011		map.m_len = last_lblock - cur_lblock;
4012		map.m_next_pgofs = NULL;
4013		map.m_next_extent = NULL;
4014		map.m_seg_type = NO_CHECK_TYPE;
4015		map.m_may_create = false;
4016
4017		ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
4018		if (ret)
4019			goto out;
4020
4021		/* hole */
4022		if (!(map.m_flags & F2FS_MAP_FLAGS)) {
4023			f2fs_err(sbi, "Swapfile has holes");
4024			ret = -EINVAL;
4025			goto out;
4026		}
4027
4028		pblock = map.m_pblk;
4029		nr_pblocks = map.m_len;
4030
4031		if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
4032				nr_pblocks & sec_blks_mask) {
4033			not_aligned++;
4034
4035			nr_pblocks = roundup(nr_pblocks, blks_per_sec);
4036			if (cur_lblock + nr_pblocks > sis->max)
4037				nr_pblocks -= blks_per_sec;
4038
4039			if (!nr_pblocks) {
4040				/* this extent is last one */
4041				nr_pblocks = map.m_len;
4042				f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
4043				goto next;
4044			}
4045
4046			ret = f2fs_migrate_blocks(inode, cur_lblock,
4047							nr_pblocks);
4048			if (ret)
4049				goto out;
4050			goto retry;
4051		}
4052next:
4053		if (cur_lblock + nr_pblocks >= sis->max)
4054			nr_pblocks = sis->max - cur_lblock;
4055
4056		if (cur_lblock) {	/* exclude the header page */
4057			if (pblock < lowest_pblock)
4058				lowest_pblock = pblock;
4059			if (pblock + nr_pblocks - 1 > highest_pblock)
4060				highest_pblock = pblock + nr_pblocks - 1;
4061		}
4062
4063		/*
4064		 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4065		 */
4066		ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
4067		if (ret < 0)
4068			goto out;
4069		nr_extents += ret;
4070		cur_lblock += nr_pblocks;
4071	}
4072	ret = nr_extents;
4073	*span = 1 + highest_pblock - lowest_pblock;
4074	if (cur_lblock == 0)
4075		cur_lblock = 1;	/* force Empty message */
4076	sis->max = cur_lblock;
4077	sis->pages = cur_lblock - 1;
4078	sis->highest_bit = cur_lblock - 1;
4079out:
4080	if (not_aligned)
4081		f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
4082			  not_aligned, blks_per_sec * F2FS_BLKSIZE);
4083	return ret;
4084}
4085
4086static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4087				sector_t *span)
4088{
4089	struct inode *inode = file_inode(file);
4090	int ret;
4091
4092	if (!S_ISREG(inode->i_mode))
4093		return -EINVAL;
4094
4095	if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4096		return -EROFS;
4097
4098	if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
4099		f2fs_err(F2FS_I_SB(inode),
4100			"Swapfile not supported in LFS mode");
4101		return -EINVAL;
4102	}
4103
4104	ret = f2fs_convert_inline_inode(inode);
4105	if (ret)
4106		return ret;
4107
4108	if (!f2fs_disable_compressed_file(inode))
4109		return -EINVAL;
4110
4111	f2fs_precache_extents(inode);
4112
4113	ret = check_swap_activate(sis, file, span);
4114	if (ret < 0)
4115		return ret;
4116
4117	set_inode_flag(inode, FI_PIN_FILE);
4118	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4119	return ret;
4120}
4121
4122static void f2fs_swap_deactivate(struct file *file)
4123{
4124	struct inode *inode = file_inode(file);
4125
4126	clear_inode_flag(inode, FI_PIN_FILE);
4127}
4128#else
4129static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4130				sector_t *span)
4131{
4132	return -EOPNOTSUPP;
4133}
4134
4135static void f2fs_swap_deactivate(struct file *file)
4136{
4137}
4138#endif
4139
4140const struct address_space_operations f2fs_dblock_aops = {
4141	.readpage	= f2fs_read_data_page,
4142	.readahead	= f2fs_readahead,
4143	.writepage	= f2fs_write_data_page,
4144	.writepages	= f2fs_write_data_pages,
4145	.write_begin	= f2fs_write_begin,
4146	.write_end	= f2fs_write_end,
4147	.set_page_dirty	= f2fs_set_data_page_dirty,
4148	.invalidatepage	= f2fs_invalidate_page,
4149	.releasepage	= f2fs_release_page,
4150	.direct_IO	= f2fs_direct_IO,
4151	.bmap		= f2fs_bmap,
4152	.swap_activate  = f2fs_swap_activate,
4153	.swap_deactivate = f2fs_swap_deactivate,
4154#ifdef CONFIG_MIGRATION
4155	.migratepage    = f2fs_migrate_page,
4156#endif
4157};
4158
4159void f2fs_clear_page_cache_dirty_tag(struct page *page)
4160{
4161	struct address_space *mapping = page_mapping(page);
4162	unsigned long flags;
4163
4164	xa_lock_irqsave(&mapping->i_pages, flags);
4165	__xa_clear_mark(&mapping->i_pages, page_index(page),
4166						PAGECACHE_TAG_DIRTY);
4167	xa_unlock_irqrestore(&mapping->i_pages, flags);
4168}
4169
4170int __init f2fs_init_post_read_processing(void)
4171{
4172	bio_post_read_ctx_cache =
4173		kmem_cache_create("f2fs_bio_post_read_ctx",
4174				  sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4175	if (!bio_post_read_ctx_cache)
4176		goto fail;
4177	bio_post_read_ctx_pool =
4178		mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4179					 bio_post_read_ctx_cache);
4180	if (!bio_post_read_ctx_pool)
4181		goto fail_free_cache;
4182	return 0;
4183
4184fail_free_cache:
4185	kmem_cache_destroy(bio_post_read_ctx_cache);
4186fail:
4187	return -ENOMEM;
4188}
4189
4190void f2fs_destroy_post_read_processing(void)
4191{
4192	mempool_destroy(bio_post_read_ctx_pool);
4193	kmem_cache_destroy(bio_post_read_ctx_cache);
4194}
4195
4196int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4197{
4198	if (!f2fs_sb_has_encrypt(sbi) &&
4199		!f2fs_sb_has_verity(sbi) &&
4200		!f2fs_sb_has_compression(sbi))
4201		return 0;
4202
4203	sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4204						 WQ_UNBOUND | WQ_HIGHPRI,
4205						 num_online_cpus());
4206	if (!sbi->post_read_wq)
4207		return -ENOMEM;
4208	return 0;
4209}
4210
4211void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4212{
4213	if (sbi->post_read_wq)
4214		destroy_workqueue(sbi->post_read_wq);
4215}
4216
4217int __init f2fs_init_bio_entry_cache(void)
4218{
4219	bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4220			sizeof(struct bio_entry));
4221	if (!bio_entry_slab)
4222		return -ENOMEM;
4223	return 0;
4224}
4225
4226void f2fs_destroy_bio_entry_cache(void)
4227{
4228	kmem_cache_destroy(bio_entry_slab);
4229}
4230