xref: /openbmc/linux/fs/f2fs/data.c (revision 4c5a116a)
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 "trace.h"
29 #include <trace/events/f2fs.h>
30 
31 #define NUM_PREALLOC_POST_READ_CTXS	128
32 
33 static struct kmem_cache *bio_post_read_ctx_cache;
34 static struct kmem_cache *bio_entry_slab;
35 static mempool_t *bio_post_read_ctx_pool;
36 static struct bio_set f2fs_bioset;
37 
38 #define	F2FS_BIO_POOL_SIZE	NR_CURSEG_TYPE
39 
40 int __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 
48 void f2fs_destroy_bioset(void)
49 {
50 	bioset_exit(&f2fs_bioset);
51 }
52 
53 static inline struct bio *__f2fs_bio_alloc(gfp_t gfp_mask,
54 						unsigned int nr_iovecs)
55 {
56 	return bio_alloc_bioset(gfp_mask, nr_iovecs, &f2fs_bioset);
57 }
58 
59 struct bio *f2fs_bio_alloc(struct f2fs_sb_info *sbi, int npages, bool noio)
60 {
61 	if (noio) {
62 		/* No failure on bio allocation */
63 		return __f2fs_bio_alloc(GFP_NOIO, npages);
64 	}
65 
66 	if (time_to_inject(sbi, FAULT_ALLOC_BIO)) {
67 		f2fs_show_injection_info(sbi, FAULT_ALLOC_BIO);
68 		return NULL;
69 	}
70 
71 	return __f2fs_bio_alloc(GFP_KERNEL, npages);
72 }
73 
74 static bool __is_cp_guaranteed(struct page *page)
75 {
76 	struct address_space *mapping = page->mapping;
77 	struct inode *inode;
78 	struct f2fs_sb_info *sbi;
79 
80 	if (!mapping)
81 		return false;
82 
83 	if (f2fs_is_compressed_page(page))
84 		return false;
85 
86 	inode = mapping->host;
87 	sbi = F2FS_I_SB(inode);
88 
89 	if (inode->i_ino == F2FS_META_INO(sbi) ||
90 			inode->i_ino ==  F2FS_NODE_INO(sbi) ||
91 			S_ISDIR(inode->i_mode) ||
92 			(S_ISREG(inode->i_mode) &&
93 			(f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
94 			is_cold_data(page))
95 		return true;
96 	return false;
97 }
98 
99 static enum count_type __read_io_type(struct page *page)
100 {
101 	struct address_space *mapping = page_file_mapping(page);
102 
103 	if (mapping) {
104 		struct inode *inode = mapping->host;
105 		struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
106 
107 		if (inode->i_ino == F2FS_META_INO(sbi))
108 			return F2FS_RD_META;
109 
110 		if (inode->i_ino == F2FS_NODE_INO(sbi))
111 			return F2FS_RD_NODE;
112 	}
113 	return F2FS_RD_DATA;
114 }
115 
116 /* postprocessing steps for read bios */
117 enum bio_post_read_step {
118 	STEP_DECRYPT,
119 	STEP_DECOMPRESS_NOWQ,		/* handle normal cluster data inplace */
120 	STEP_DECOMPRESS,		/* handle compressed cluster data in workqueue */
121 	STEP_VERITY,
122 };
123 
124 struct bio_post_read_ctx {
125 	struct bio *bio;
126 	struct f2fs_sb_info *sbi;
127 	struct work_struct work;
128 	unsigned int enabled_steps;
129 };
130 
131 static void __read_end_io(struct bio *bio, bool compr, bool verity)
132 {
133 	struct page *page;
134 	struct bio_vec *bv;
135 	struct bvec_iter_all iter_all;
136 
137 	bio_for_each_segment_all(bv, bio, iter_all) {
138 		page = bv->bv_page;
139 
140 #ifdef CONFIG_F2FS_FS_COMPRESSION
141 		if (compr && f2fs_is_compressed_page(page)) {
142 			f2fs_decompress_pages(bio, page, verity);
143 			continue;
144 		}
145 		if (verity)
146 			continue;
147 #endif
148 
149 		/* PG_error was set if any post_read step failed */
150 		if (bio->bi_status || PageError(page)) {
151 			ClearPageUptodate(page);
152 			/* will re-read again later */
153 			ClearPageError(page);
154 		} else {
155 			SetPageUptodate(page);
156 		}
157 		dec_page_count(F2FS_P_SB(page), __read_io_type(page));
158 		unlock_page(page);
159 	}
160 }
161 
162 static void f2fs_release_read_bio(struct bio *bio);
163 static void __f2fs_read_end_io(struct bio *bio, bool compr, bool verity)
164 {
165 	if (!compr)
166 		__read_end_io(bio, false, verity);
167 	f2fs_release_read_bio(bio);
168 }
169 
170 static void f2fs_decompress_bio(struct bio *bio, bool verity)
171 {
172 	__read_end_io(bio, true, verity);
173 }
174 
175 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
176 
177 static void f2fs_decrypt_work(struct bio_post_read_ctx *ctx)
178 {
179 	fscrypt_decrypt_bio(ctx->bio);
180 }
181 
182 static void f2fs_decompress_work(struct bio_post_read_ctx *ctx)
183 {
184 	f2fs_decompress_bio(ctx->bio, ctx->enabled_steps & (1 << STEP_VERITY));
185 }
186 
187 #ifdef CONFIG_F2FS_FS_COMPRESSION
188 static void f2fs_verify_pages(struct page **rpages, unsigned int cluster_size)
189 {
190 	f2fs_decompress_end_io(rpages, cluster_size, false, true);
191 }
192 
193 static void f2fs_verify_bio(struct bio *bio)
194 {
195 	struct bio_vec *bv;
196 	struct bvec_iter_all iter_all;
197 
198 	bio_for_each_segment_all(bv, bio, iter_all) {
199 		struct page *page = bv->bv_page;
200 		struct decompress_io_ctx *dic;
201 
202 		dic = (struct decompress_io_ctx *)page_private(page);
203 
204 		if (dic) {
205 			if (refcount_dec_not_one(&dic->ref))
206 				continue;
207 			f2fs_verify_pages(dic->rpages,
208 						dic->cluster_size);
209 			f2fs_free_dic(dic);
210 			continue;
211 		}
212 
213 		if (bio->bi_status || PageError(page))
214 			goto clear_uptodate;
215 
216 		if (fsverity_verify_page(page)) {
217 			SetPageUptodate(page);
218 			goto unlock;
219 		}
220 clear_uptodate:
221 		ClearPageUptodate(page);
222 		ClearPageError(page);
223 unlock:
224 		dec_page_count(F2FS_P_SB(page), __read_io_type(page));
225 		unlock_page(page);
226 	}
227 }
228 #endif
229 
230 static void f2fs_verity_work(struct work_struct *work)
231 {
232 	struct bio_post_read_ctx *ctx =
233 		container_of(work, struct bio_post_read_ctx, work);
234 	struct bio *bio = ctx->bio;
235 #ifdef CONFIG_F2FS_FS_COMPRESSION
236 	unsigned int enabled_steps = ctx->enabled_steps;
237 #endif
238 
239 	/*
240 	 * fsverity_verify_bio() may call readpages() again, and while verity
241 	 * will be disabled for this, decryption may still be needed, resulting
242 	 * in another bio_post_read_ctx being allocated.  So to prevent
243 	 * deadlocks we need to release the current ctx to the mempool first.
244 	 * This assumes that verity is the last post-read step.
245 	 */
246 	mempool_free(ctx, bio_post_read_ctx_pool);
247 	bio->bi_private = NULL;
248 
249 #ifdef CONFIG_F2FS_FS_COMPRESSION
250 	/* previous step is decompression */
251 	if (enabled_steps & (1 << STEP_DECOMPRESS)) {
252 		f2fs_verify_bio(bio);
253 		f2fs_release_read_bio(bio);
254 		return;
255 	}
256 #endif
257 
258 	fsverity_verify_bio(bio);
259 	__f2fs_read_end_io(bio, false, false);
260 }
261 
262 static void f2fs_post_read_work(struct work_struct *work)
263 {
264 	struct bio_post_read_ctx *ctx =
265 		container_of(work, struct bio_post_read_ctx, work);
266 
267 	if (ctx->enabled_steps & (1 << STEP_DECRYPT))
268 		f2fs_decrypt_work(ctx);
269 
270 	if (ctx->enabled_steps & (1 << STEP_DECOMPRESS))
271 		f2fs_decompress_work(ctx);
272 
273 	if (ctx->enabled_steps & (1 << STEP_VERITY)) {
274 		INIT_WORK(&ctx->work, f2fs_verity_work);
275 		fsverity_enqueue_verify_work(&ctx->work);
276 		return;
277 	}
278 
279 	__f2fs_read_end_io(ctx->bio,
280 		ctx->enabled_steps & (1 << STEP_DECOMPRESS), false);
281 }
282 
283 static void f2fs_enqueue_post_read_work(struct f2fs_sb_info *sbi,
284 						struct work_struct *work)
285 {
286 	queue_work(sbi->post_read_wq, work);
287 }
288 
289 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
290 {
291 	/*
292 	 * We use different work queues for decryption and for verity because
293 	 * verity may require reading metadata pages that need decryption, and
294 	 * we shouldn't recurse to the same workqueue.
295 	 */
296 
297 	if (ctx->enabled_steps & (1 << STEP_DECRYPT) ||
298 		ctx->enabled_steps & (1 << STEP_DECOMPRESS)) {
299 		INIT_WORK(&ctx->work, f2fs_post_read_work);
300 		f2fs_enqueue_post_read_work(ctx->sbi, &ctx->work);
301 		return;
302 	}
303 
304 	if (ctx->enabled_steps & (1 << STEP_VERITY)) {
305 		INIT_WORK(&ctx->work, f2fs_verity_work);
306 		fsverity_enqueue_verify_work(&ctx->work);
307 		return;
308 	}
309 
310 	__f2fs_read_end_io(ctx->bio, false, false);
311 }
312 
313 static bool f2fs_bio_post_read_required(struct bio *bio)
314 {
315 	return bio->bi_private;
316 }
317 
318 static void f2fs_read_end_io(struct bio *bio)
319 {
320 	struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
321 
322 	if (time_to_inject(sbi, FAULT_READ_IO)) {
323 		f2fs_show_injection_info(sbi, FAULT_READ_IO);
324 		bio->bi_status = BLK_STS_IOERR;
325 	}
326 
327 	if (f2fs_bio_post_read_required(bio)) {
328 		struct bio_post_read_ctx *ctx = bio->bi_private;
329 
330 		bio_post_read_processing(ctx);
331 		return;
332 	}
333 
334 	__f2fs_read_end_io(bio, false, false);
335 }
336 
337 static void f2fs_write_end_io(struct bio *bio)
338 {
339 	struct f2fs_sb_info *sbi = bio->bi_private;
340 	struct bio_vec *bvec;
341 	struct bvec_iter_all iter_all;
342 
343 	if (time_to_inject(sbi, FAULT_WRITE_IO)) {
344 		f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
345 		bio->bi_status = BLK_STS_IOERR;
346 	}
347 
348 	bio_for_each_segment_all(bvec, bio, iter_all) {
349 		struct page *page = bvec->bv_page;
350 		enum count_type type = WB_DATA_TYPE(page);
351 
352 		if (IS_DUMMY_WRITTEN_PAGE(page)) {
353 			set_page_private(page, (unsigned long)NULL);
354 			ClearPagePrivate(page);
355 			unlock_page(page);
356 			mempool_free(page, sbi->write_io_dummy);
357 
358 			if (unlikely(bio->bi_status))
359 				f2fs_stop_checkpoint(sbi, true);
360 			continue;
361 		}
362 
363 		fscrypt_finalize_bounce_page(&page);
364 
365 #ifdef CONFIG_F2FS_FS_COMPRESSION
366 		if (f2fs_is_compressed_page(page)) {
367 			f2fs_compress_write_end_io(bio, page);
368 			continue;
369 		}
370 #endif
371 
372 		if (unlikely(bio->bi_status)) {
373 			mapping_set_error(page->mapping, -EIO);
374 			if (type == F2FS_WB_CP_DATA)
375 				f2fs_stop_checkpoint(sbi, true);
376 		}
377 
378 		f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
379 					page->index != nid_of_node(page));
380 
381 		dec_page_count(sbi, type);
382 		if (f2fs_in_warm_node_list(sbi, page))
383 			f2fs_del_fsync_node_entry(sbi, page);
384 		clear_cold_data(page);
385 		end_page_writeback(page);
386 	}
387 	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
388 				wq_has_sleeper(&sbi->cp_wait))
389 		wake_up(&sbi->cp_wait);
390 
391 	bio_put(bio);
392 }
393 
394 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
395 				block_t blk_addr, struct bio *bio)
396 {
397 	struct block_device *bdev = sbi->sb->s_bdev;
398 	int i;
399 
400 	if (f2fs_is_multi_device(sbi)) {
401 		for (i = 0; i < sbi->s_ndevs; i++) {
402 			if (FDEV(i).start_blk <= blk_addr &&
403 			    FDEV(i).end_blk >= blk_addr) {
404 				blk_addr -= FDEV(i).start_blk;
405 				bdev = FDEV(i).bdev;
406 				break;
407 			}
408 		}
409 	}
410 	if (bio) {
411 		bio_set_dev(bio, bdev);
412 		bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
413 	}
414 	return bdev;
415 }
416 
417 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
418 {
419 	int i;
420 
421 	if (!f2fs_is_multi_device(sbi))
422 		return 0;
423 
424 	for (i = 0; i < sbi->s_ndevs; i++)
425 		if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
426 			return i;
427 	return 0;
428 }
429 
430 /*
431  * Return true, if pre_bio's bdev is same as its target device.
432  */
433 static bool __same_bdev(struct f2fs_sb_info *sbi,
434 				block_t blk_addr, struct bio *bio)
435 {
436 	struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
437 	return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
438 }
439 
440 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
441 {
442 	struct f2fs_sb_info *sbi = fio->sbi;
443 	struct bio *bio;
444 
445 	bio = f2fs_bio_alloc(sbi, npages, true);
446 
447 	f2fs_target_device(sbi, fio->new_blkaddr, bio);
448 	if (is_read_io(fio->op)) {
449 		bio->bi_end_io = f2fs_read_end_io;
450 		bio->bi_private = NULL;
451 	} else {
452 		bio->bi_end_io = f2fs_write_end_io;
453 		bio->bi_private = sbi;
454 		bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
455 						fio->type, fio->temp);
456 	}
457 	if (fio->io_wbc)
458 		wbc_init_bio(fio->io_wbc, bio);
459 
460 	return bio;
461 }
462 
463 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
464 				  pgoff_t first_idx,
465 				  const struct f2fs_io_info *fio,
466 				  gfp_t gfp_mask)
467 {
468 	/*
469 	 * The f2fs garbage collector sets ->encrypted_page when it wants to
470 	 * read/write raw data without encryption.
471 	 */
472 	if (!fio || !fio->encrypted_page)
473 		fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
474 }
475 
476 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
477 				     pgoff_t next_idx,
478 				     const struct f2fs_io_info *fio)
479 {
480 	/*
481 	 * The f2fs garbage collector sets ->encrypted_page when it wants to
482 	 * read/write raw data without encryption.
483 	 */
484 	if (fio && fio->encrypted_page)
485 		return !bio_has_crypt_ctx(bio);
486 
487 	return fscrypt_mergeable_bio(bio, inode, next_idx);
488 }
489 
490 static inline void __submit_bio(struct f2fs_sb_info *sbi,
491 				struct bio *bio, enum page_type type)
492 {
493 	if (!is_read_io(bio_op(bio))) {
494 		unsigned int start;
495 
496 		if (type != DATA && type != NODE)
497 			goto submit_io;
498 
499 		if (f2fs_lfs_mode(sbi) && current->plug)
500 			blk_finish_plug(current->plug);
501 
502 		if (F2FS_IO_ALIGNED(sbi))
503 			goto submit_io;
504 
505 		start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
506 		start %= F2FS_IO_SIZE(sbi);
507 
508 		if (start == 0)
509 			goto submit_io;
510 
511 		/* fill dummy pages */
512 		for (; start < F2FS_IO_SIZE(sbi); start++) {
513 			struct page *page =
514 				mempool_alloc(sbi->write_io_dummy,
515 					      GFP_NOIO | __GFP_NOFAIL);
516 			f2fs_bug_on(sbi, !page);
517 
518 			zero_user_segment(page, 0, PAGE_SIZE);
519 			SetPagePrivate(page);
520 			set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
521 			lock_page(page);
522 			if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
523 				f2fs_bug_on(sbi, 1);
524 		}
525 		/*
526 		 * In the NODE case, we lose next block address chain. So, we
527 		 * need to do checkpoint in f2fs_sync_file.
528 		 */
529 		if (type == NODE)
530 			set_sbi_flag(sbi, SBI_NEED_CP);
531 	}
532 submit_io:
533 	if (is_read_io(bio_op(bio)))
534 		trace_f2fs_submit_read_bio(sbi->sb, type, bio);
535 	else
536 		trace_f2fs_submit_write_bio(sbi->sb, type, bio);
537 	submit_bio(bio);
538 }
539 
540 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
541 				struct bio *bio, enum page_type type)
542 {
543 	__submit_bio(sbi, bio, type);
544 }
545 
546 static void __attach_io_flag(struct f2fs_io_info *fio)
547 {
548 	struct f2fs_sb_info *sbi = fio->sbi;
549 	unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
550 	unsigned int io_flag, fua_flag, meta_flag;
551 
552 	if (fio->type == DATA)
553 		io_flag = sbi->data_io_flag;
554 	else if (fio->type == NODE)
555 		io_flag = sbi->node_io_flag;
556 	else
557 		return;
558 
559 	fua_flag = io_flag & temp_mask;
560 	meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
561 
562 	/*
563 	 * data/node io flag bits per temp:
564 	 *      REQ_META     |      REQ_FUA      |
565 	 *    5 |    4 |   3 |    2 |    1 |   0 |
566 	 * Cold | Warm | Hot | Cold | Warm | Hot |
567 	 */
568 	if ((1 << fio->temp) & meta_flag)
569 		fio->op_flags |= REQ_META;
570 	if ((1 << fio->temp) & fua_flag)
571 		fio->op_flags |= REQ_FUA;
572 }
573 
574 static void __submit_merged_bio(struct f2fs_bio_info *io)
575 {
576 	struct f2fs_io_info *fio = &io->fio;
577 
578 	if (!io->bio)
579 		return;
580 
581 	__attach_io_flag(fio);
582 	bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
583 
584 	if (is_read_io(fio->op))
585 		trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
586 	else
587 		trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
588 
589 	__submit_bio(io->sbi, io->bio, fio->type);
590 	io->bio = NULL;
591 }
592 
593 static bool __has_merged_page(struct bio *bio, struct inode *inode,
594 						struct page *page, nid_t ino)
595 {
596 	struct bio_vec *bvec;
597 	struct bvec_iter_all iter_all;
598 
599 	if (!bio)
600 		return false;
601 
602 	if (!inode && !page && !ino)
603 		return true;
604 
605 	bio_for_each_segment_all(bvec, bio, iter_all) {
606 		struct page *target = bvec->bv_page;
607 
608 		if (fscrypt_is_bounce_page(target)) {
609 			target = fscrypt_pagecache_page(target);
610 			if (IS_ERR(target))
611 				continue;
612 		}
613 		if (f2fs_is_compressed_page(target)) {
614 			target = f2fs_compress_control_page(target);
615 			if (IS_ERR(target))
616 				continue;
617 		}
618 
619 		if (inode && inode == target->mapping->host)
620 			return true;
621 		if (page && page == target)
622 			return true;
623 		if (ino && ino == ino_of_node(target))
624 			return true;
625 	}
626 
627 	return false;
628 }
629 
630 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
631 				enum page_type type, enum temp_type temp)
632 {
633 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
634 	struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
635 
636 	down_write(&io->io_rwsem);
637 
638 	/* change META to META_FLUSH in the checkpoint procedure */
639 	if (type >= META_FLUSH) {
640 		io->fio.type = META_FLUSH;
641 		io->fio.op = REQ_OP_WRITE;
642 		io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
643 		if (!test_opt(sbi, NOBARRIER))
644 			io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
645 	}
646 	__submit_merged_bio(io);
647 	up_write(&io->io_rwsem);
648 }
649 
650 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
651 				struct inode *inode, struct page *page,
652 				nid_t ino, enum page_type type, bool force)
653 {
654 	enum temp_type temp;
655 	bool ret = true;
656 
657 	for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
658 		if (!force)	{
659 			enum page_type btype = PAGE_TYPE_OF_BIO(type);
660 			struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
661 
662 			down_read(&io->io_rwsem);
663 			ret = __has_merged_page(io->bio, inode, page, ino);
664 			up_read(&io->io_rwsem);
665 		}
666 		if (ret)
667 			__f2fs_submit_merged_write(sbi, type, temp);
668 
669 		/* TODO: use HOT temp only for meta pages now. */
670 		if (type >= META)
671 			break;
672 	}
673 }
674 
675 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
676 {
677 	__submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
678 }
679 
680 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
681 				struct inode *inode, struct page *page,
682 				nid_t ino, enum page_type type)
683 {
684 	__submit_merged_write_cond(sbi, inode, page, ino, type, false);
685 }
686 
687 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
688 {
689 	f2fs_submit_merged_write(sbi, DATA);
690 	f2fs_submit_merged_write(sbi, NODE);
691 	f2fs_submit_merged_write(sbi, META);
692 }
693 
694 /*
695  * Fill the locked page with data located in the block address.
696  * A caller needs to unlock the page on failure.
697  */
698 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
699 {
700 	struct bio *bio;
701 	struct page *page = fio->encrypted_page ?
702 			fio->encrypted_page : fio->page;
703 
704 	if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
705 			fio->is_por ? META_POR : (__is_meta_io(fio) ?
706 			META_GENERIC : DATA_GENERIC_ENHANCE)))
707 		return -EFSCORRUPTED;
708 
709 	trace_f2fs_submit_page_bio(page, fio);
710 	f2fs_trace_ios(fio, 0);
711 
712 	/* Allocate a new bio */
713 	bio = __bio_alloc(fio, 1);
714 
715 	f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
716 			       fio->page->index, fio, GFP_NOIO);
717 
718 	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
719 		bio_put(bio);
720 		return -EFAULT;
721 	}
722 
723 	if (fio->io_wbc && !is_read_io(fio->op))
724 		wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
725 
726 	__attach_io_flag(fio);
727 	bio_set_op_attrs(bio, fio->op, fio->op_flags);
728 
729 	inc_page_count(fio->sbi, is_read_io(fio->op) ?
730 			__read_io_type(page): WB_DATA_TYPE(fio->page));
731 
732 	__submit_bio(fio->sbi, bio, fio->type);
733 	return 0;
734 }
735 
736 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
737 				block_t last_blkaddr, block_t cur_blkaddr)
738 {
739 	if (last_blkaddr + 1 != cur_blkaddr)
740 		return false;
741 	return __same_bdev(sbi, cur_blkaddr, bio);
742 }
743 
744 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
745 						struct f2fs_io_info *fio)
746 {
747 	if (io->fio.op != fio->op)
748 		return false;
749 	return io->fio.op_flags == fio->op_flags;
750 }
751 
752 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
753 					struct f2fs_bio_info *io,
754 					struct f2fs_io_info *fio,
755 					block_t last_blkaddr,
756 					block_t cur_blkaddr)
757 {
758 	if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
759 		unsigned int filled_blocks =
760 				F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
761 		unsigned int io_size = F2FS_IO_SIZE(sbi);
762 		unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
763 
764 		/* IOs in bio is aligned and left space of vectors is not enough */
765 		if (!(filled_blocks % io_size) && left_vecs < io_size)
766 			return false;
767 	}
768 	if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
769 		return false;
770 	return io_type_is_mergeable(io, fio);
771 }
772 
773 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
774 				struct page *page, enum temp_type temp)
775 {
776 	struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
777 	struct bio_entry *be;
778 
779 	be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
780 	be->bio = bio;
781 	bio_get(bio);
782 
783 	if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
784 		f2fs_bug_on(sbi, 1);
785 
786 	down_write(&io->bio_list_lock);
787 	list_add_tail(&be->list, &io->bio_list);
788 	up_write(&io->bio_list_lock);
789 }
790 
791 static void del_bio_entry(struct bio_entry *be)
792 {
793 	list_del(&be->list);
794 	kmem_cache_free(bio_entry_slab, be);
795 }
796 
797 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
798 							struct page *page)
799 {
800 	struct f2fs_sb_info *sbi = fio->sbi;
801 	enum temp_type temp;
802 	bool found = false;
803 	int ret = -EAGAIN;
804 
805 	for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
806 		struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
807 		struct list_head *head = &io->bio_list;
808 		struct bio_entry *be;
809 
810 		down_write(&io->bio_list_lock);
811 		list_for_each_entry(be, head, list) {
812 			if (be->bio != *bio)
813 				continue;
814 
815 			found = true;
816 
817 			f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
818 							    *fio->last_block,
819 							    fio->new_blkaddr));
820 			if (f2fs_crypt_mergeable_bio(*bio,
821 					fio->page->mapping->host,
822 					fio->page->index, fio) &&
823 			    bio_add_page(*bio, page, PAGE_SIZE, 0) ==
824 					PAGE_SIZE) {
825 				ret = 0;
826 				break;
827 			}
828 
829 			/* page can't be merged into bio; submit the bio */
830 			del_bio_entry(be);
831 			__submit_bio(sbi, *bio, DATA);
832 			break;
833 		}
834 		up_write(&io->bio_list_lock);
835 	}
836 
837 	if (ret) {
838 		bio_put(*bio);
839 		*bio = NULL;
840 	}
841 
842 	return ret;
843 }
844 
845 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
846 					struct bio **bio, struct page *page)
847 {
848 	enum temp_type temp;
849 	bool found = false;
850 	struct bio *target = bio ? *bio : NULL;
851 
852 	for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
853 		struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
854 		struct list_head *head = &io->bio_list;
855 		struct bio_entry *be;
856 
857 		if (list_empty(head))
858 			continue;
859 
860 		down_read(&io->bio_list_lock);
861 		list_for_each_entry(be, head, list) {
862 			if (target)
863 				found = (target == be->bio);
864 			else
865 				found = __has_merged_page(be->bio, NULL,
866 								page, 0);
867 			if (found)
868 				break;
869 		}
870 		up_read(&io->bio_list_lock);
871 
872 		if (!found)
873 			continue;
874 
875 		found = false;
876 
877 		down_write(&io->bio_list_lock);
878 		list_for_each_entry(be, head, list) {
879 			if (target)
880 				found = (target == be->bio);
881 			else
882 				found = __has_merged_page(be->bio, NULL,
883 								page, 0);
884 			if (found) {
885 				target = be->bio;
886 				del_bio_entry(be);
887 				break;
888 			}
889 		}
890 		up_write(&io->bio_list_lock);
891 	}
892 
893 	if (found)
894 		__submit_bio(sbi, target, DATA);
895 	if (bio && *bio) {
896 		bio_put(*bio);
897 		*bio = NULL;
898 	}
899 }
900 
901 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
902 {
903 	struct bio *bio = *fio->bio;
904 	struct page *page = fio->encrypted_page ?
905 			fio->encrypted_page : fio->page;
906 
907 	if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
908 			__is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
909 		return -EFSCORRUPTED;
910 
911 	trace_f2fs_submit_page_bio(page, fio);
912 	f2fs_trace_ios(fio, 0);
913 
914 	if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
915 						fio->new_blkaddr))
916 		f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
917 alloc_new:
918 	if (!bio) {
919 		bio = __bio_alloc(fio, BIO_MAX_PAGES);
920 		__attach_io_flag(fio);
921 		f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
922 				       fio->page->index, fio, GFP_NOIO);
923 		bio_set_op_attrs(bio, fio->op, fio->op_flags);
924 
925 		add_bio_entry(fio->sbi, bio, page, fio->temp);
926 	} else {
927 		if (add_ipu_page(fio, &bio, page))
928 			goto alloc_new;
929 	}
930 
931 	if (fio->io_wbc)
932 		wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
933 
934 	inc_page_count(fio->sbi, WB_DATA_TYPE(page));
935 
936 	*fio->last_block = fio->new_blkaddr;
937 	*fio->bio = bio;
938 
939 	return 0;
940 }
941 
942 void f2fs_submit_page_write(struct f2fs_io_info *fio)
943 {
944 	struct f2fs_sb_info *sbi = fio->sbi;
945 	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
946 	struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
947 	struct page *bio_page;
948 
949 	f2fs_bug_on(sbi, is_read_io(fio->op));
950 
951 	down_write(&io->io_rwsem);
952 next:
953 	if (fio->in_list) {
954 		spin_lock(&io->io_lock);
955 		if (list_empty(&io->io_list)) {
956 			spin_unlock(&io->io_lock);
957 			goto out;
958 		}
959 		fio = list_first_entry(&io->io_list,
960 						struct f2fs_io_info, list);
961 		list_del(&fio->list);
962 		spin_unlock(&io->io_lock);
963 	}
964 
965 	verify_fio_blkaddr(fio);
966 
967 	if (fio->encrypted_page)
968 		bio_page = fio->encrypted_page;
969 	else if (fio->compressed_page)
970 		bio_page = fio->compressed_page;
971 	else
972 		bio_page = fio->page;
973 
974 	/* set submitted = true as a return value */
975 	fio->submitted = true;
976 
977 	inc_page_count(sbi, WB_DATA_TYPE(bio_page));
978 
979 	if (io->bio &&
980 	    (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
981 			      fio->new_blkaddr) ||
982 	     !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
983 				       bio_page->index, fio)))
984 		__submit_merged_bio(io);
985 alloc_new:
986 	if (io->bio == NULL) {
987 		if (F2FS_IO_ALIGNED(sbi) &&
988 				(fio->type == DATA || fio->type == NODE) &&
989 				fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
990 			dec_page_count(sbi, WB_DATA_TYPE(bio_page));
991 			fio->retry = true;
992 			goto skip;
993 		}
994 		io->bio = __bio_alloc(fio, BIO_MAX_PAGES);
995 		f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
996 				       bio_page->index, fio, GFP_NOIO);
997 		io->fio = *fio;
998 	}
999 
1000 	if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1001 		__submit_merged_bio(io);
1002 		goto alloc_new;
1003 	}
1004 
1005 	if (fio->io_wbc)
1006 		wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
1007 
1008 	io->last_block_in_bio = fio->new_blkaddr;
1009 	f2fs_trace_ios(fio, 0);
1010 
1011 	trace_f2fs_submit_page_write(fio->page, fio);
1012 skip:
1013 	if (fio->in_list)
1014 		goto next;
1015 out:
1016 	if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1017 				!f2fs_is_checkpoint_ready(sbi))
1018 		__submit_merged_bio(io);
1019 	up_write(&io->io_rwsem);
1020 }
1021 
1022 static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx)
1023 {
1024 	return fsverity_active(inode) &&
1025 	       idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
1026 }
1027 
1028 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1029 				      unsigned nr_pages, unsigned op_flag,
1030 				      pgoff_t first_idx, bool for_write)
1031 {
1032 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1033 	struct bio *bio;
1034 	struct bio_post_read_ctx *ctx;
1035 	unsigned int post_read_steps = 0;
1036 
1037 	bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES),
1038 								for_write);
1039 	if (!bio)
1040 		return ERR_PTR(-ENOMEM);
1041 
1042 	f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1043 
1044 	f2fs_target_device(sbi, blkaddr, bio);
1045 	bio->bi_end_io = f2fs_read_end_io;
1046 	bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
1047 
1048 	if (fscrypt_inode_uses_fs_layer_crypto(inode))
1049 		post_read_steps |= 1 << STEP_DECRYPT;
1050 	if (f2fs_compressed_file(inode))
1051 		post_read_steps |= 1 << STEP_DECOMPRESS_NOWQ;
1052 	if (f2fs_need_verity(inode, first_idx))
1053 		post_read_steps |= 1 << STEP_VERITY;
1054 
1055 	if (post_read_steps) {
1056 		/* Due to the mempool, this never fails. */
1057 		ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1058 		ctx->bio = bio;
1059 		ctx->sbi = sbi;
1060 		ctx->enabled_steps = post_read_steps;
1061 		bio->bi_private = ctx;
1062 	}
1063 
1064 	return bio;
1065 }
1066 
1067 static void f2fs_release_read_bio(struct bio *bio)
1068 {
1069 	if (bio->bi_private)
1070 		mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1071 	bio_put(bio);
1072 }
1073 
1074 /* This can handle encryption stuffs */
1075 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1076 						block_t blkaddr, bool for_write)
1077 {
1078 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1079 	struct bio *bio;
1080 
1081 	bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0, page->index, for_write);
1082 	if (IS_ERR(bio))
1083 		return PTR_ERR(bio);
1084 
1085 	/* wait for GCed page writeback via META_MAPPING */
1086 	f2fs_wait_on_block_writeback(inode, blkaddr);
1087 
1088 	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1089 		bio_put(bio);
1090 		return -EFAULT;
1091 	}
1092 	ClearPageError(page);
1093 	inc_page_count(sbi, F2FS_RD_DATA);
1094 	f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
1095 	__submit_bio(sbi, bio, DATA);
1096 	return 0;
1097 }
1098 
1099 static void __set_data_blkaddr(struct dnode_of_data *dn)
1100 {
1101 	struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1102 	__le32 *addr_array;
1103 	int base = 0;
1104 
1105 	if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1106 		base = get_extra_isize(dn->inode);
1107 
1108 	/* Get physical address of data block */
1109 	addr_array = blkaddr_in_node(rn);
1110 	addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1111 }
1112 
1113 /*
1114  * Lock ordering for the change of data block address:
1115  * ->data_page
1116  *  ->node_page
1117  *    update block addresses in the node page
1118  */
1119 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1120 {
1121 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1122 	__set_data_blkaddr(dn);
1123 	if (set_page_dirty(dn->node_page))
1124 		dn->node_changed = true;
1125 }
1126 
1127 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1128 {
1129 	dn->data_blkaddr = blkaddr;
1130 	f2fs_set_data_blkaddr(dn);
1131 	f2fs_update_extent_cache(dn);
1132 }
1133 
1134 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1135 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1136 {
1137 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1138 	int err;
1139 
1140 	if (!count)
1141 		return 0;
1142 
1143 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1144 		return -EPERM;
1145 	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1146 		return err;
1147 
1148 	trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1149 						dn->ofs_in_node, count);
1150 
1151 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1152 
1153 	for (; count > 0; dn->ofs_in_node++) {
1154 		block_t blkaddr = f2fs_data_blkaddr(dn);
1155 		if (blkaddr == NULL_ADDR) {
1156 			dn->data_blkaddr = NEW_ADDR;
1157 			__set_data_blkaddr(dn);
1158 			count--;
1159 		}
1160 	}
1161 
1162 	if (set_page_dirty(dn->node_page))
1163 		dn->node_changed = true;
1164 	return 0;
1165 }
1166 
1167 /* Should keep dn->ofs_in_node unchanged */
1168 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1169 {
1170 	unsigned int ofs_in_node = dn->ofs_in_node;
1171 	int ret;
1172 
1173 	ret = f2fs_reserve_new_blocks(dn, 1);
1174 	dn->ofs_in_node = ofs_in_node;
1175 	return ret;
1176 }
1177 
1178 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1179 {
1180 	bool need_put = dn->inode_page ? false : true;
1181 	int err;
1182 
1183 	err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1184 	if (err)
1185 		return err;
1186 
1187 	if (dn->data_blkaddr == NULL_ADDR)
1188 		err = f2fs_reserve_new_block(dn);
1189 	if (err || need_put)
1190 		f2fs_put_dnode(dn);
1191 	return err;
1192 }
1193 
1194 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1195 {
1196 	struct extent_info ei  = {0,0,0};
1197 	struct inode *inode = dn->inode;
1198 
1199 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1200 		dn->data_blkaddr = ei.blk + index - ei.fofs;
1201 		return 0;
1202 	}
1203 
1204 	return f2fs_reserve_block(dn, index);
1205 }
1206 
1207 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1208 						int op_flags, bool for_write)
1209 {
1210 	struct address_space *mapping = inode->i_mapping;
1211 	struct dnode_of_data dn;
1212 	struct page *page;
1213 	struct extent_info ei = {0,0,0};
1214 	int err;
1215 
1216 	page = f2fs_grab_cache_page(mapping, index, for_write);
1217 	if (!page)
1218 		return ERR_PTR(-ENOMEM);
1219 
1220 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1221 		dn.data_blkaddr = ei.blk + index - ei.fofs;
1222 		if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1223 						DATA_GENERIC_ENHANCE_READ)) {
1224 			err = -EFSCORRUPTED;
1225 			goto put_err;
1226 		}
1227 		goto got_it;
1228 	}
1229 
1230 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1231 	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1232 	if (err)
1233 		goto put_err;
1234 	f2fs_put_dnode(&dn);
1235 
1236 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1237 		err = -ENOENT;
1238 		goto put_err;
1239 	}
1240 	if (dn.data_blkaddr != NEW_ADDR &&
1241 			!f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1242 						dn.data_blkaddr,
1243 						DATA_GENERIC_ENHANCE)) {
1244 		err = -EFSCORRUPTED;
1245 		goto put_err;
1246 	}
1247 got_it:
1248 	if (PageUptodate(page)) {
1249 		unlock_page(page);
1250 		return page;
1251 	}
1252 
1253 	/*
1254 	 * A new dentry page is allocated but not able to be written, since its
1255 	 * new inode page couldn't be allocated due to -ENOSPC.
1256 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
1257 	 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1258 	 * f2fs_init_inode_metadata.
1259 	 */
1260 	if (dn.data_blkaddr == NEW_ADDR) {
1261 		zero_user_segment(page, 0, PAGE_SIZE);
1262 		if (!PageUptodate(page))
1263 			SetPageUptodate(page);
1264 		unlock_page(page);
1265 		return page;
1266 	}
1267 
1268 	err = f2fs_submit_page_read(inode, page, dn.data_blkaddr, for_write);
1269 	if (err)
1270 		goto put_err;
1271 	return page;
1272 
1273 put_err:
1274 	f2fs_put_page(page, 1);
1275 	return ERR_PTR(err);
1276 }
1277 
1278 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
1279 {
1280 	struct address_space *mapping = inode->i_mapping;
1281 	struct page *page;
1282 
1283 	page = find_get_page(mapping, index);
1284 	if (page && PageUptodate(page))
1285 		return page;
1286 	f2fs_put_page(page, 0);
1287 
1288 	page = f2fs_get_read_data_page(inode, index, 0, false);
1289 	if (IS_ERR(page))
1290 		return page;
1291 
1292 	if (PageUptodate(page))
1293 		return page;
1294 
1295 	wait_on_page_locked(page);
1296 	if (unlikely(!PageUptodate(page))) {
1297 		f2fs_put_page(page, 0);
1298 		return ERR_PTR(-EIO);
1299 	}
1300 	return page;
1301 }
1302 
1303 /*
1304  * If it tries to access a hole, return an error.
1305  * Because, the callers, functions in dir.c and GC, should be able to know
1306  * whether this page exists or not.
1307  */
1308 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1309 							bool for_write)
1310 {
1311 	struct address_space *mapping = inode->i_mapping;
1312 	struct page *page;
1313 repeat:
1314 	page = f2fs_get_read_data_page(inode, index, 0, for_write);
1315 	if (IS_ERR(page))
1316 		return page;
1317 
1318 	/* wait for read completion */
1319 	lock_page(page);
1320 	if (unlikely(page->mapping != mapping)) {
1321 		f2fs_put_page(page, 1);
1322 		goto repeat;
1323 	}
1324 	if (unlikely(!PageUptodate(page))) {
1325 		f2fs_put_page(page, 1);
1326 		return ERR_PTR(-EIO);
1327 	}
1328 	return page;
1329 }
1330 
1331 /*
1332  * Caller ensures that this data page is never allocated.
1333  * A new zero-filled data page is allocated in the page cache.
1334  *
1335  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1336  * f2fs_unlock_op().
1337  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1338  * ipage should be released by this function.
1339  */
1340 struct page *f2fs_get_new_data_page(struct inode *inode,
1341 		struct page *ipage, pgoff_t index, bool new_i_size)
1342 {
1343 	struct address_space *mapping = inode->i_mapping;
1344 	struct page *page;
1345 	struct dnode_of_data dn;
1346 	int err;
1347 
1348 	page = f2fs_grab_cache_page(mapping, index, true);
1349 	if (!page) {
1350 		/*
1351 		 * before exiting, we should make sure ipage will be released
1352 		 * if any error occur.
1353 		 */
1354 		f2fs_put_page(ipage, 1);
1355 		return ERR_PTR(-ENOMEM);
1356 	}
1357 
1358 	set_new_dnode(&dn, inode, ipage, NULL, 0);
1359 	err = f2fs_reserve_block(&dn, index);
1360 	if (err) {
1361 		f2fs_put_page(page, 1);
1362 		return ERR_PTR(err);
1363 	}
1364 	if (!ipage)
1365 		f2fs_put_dnode(&dn);
1366 
1367 	if (PageUptodate(page))
1368 		goto got_it;
1369 
1370 	if (dn.data_blkaddr == NEW_ADDR) {
1371 		zero_user_segment(page, 0, PAGE_SIZE);
1372 		if (!PageUptodate(page))
1373 			SetPageUptodate(page);
1374 	} else {
1375 		f2fs_put_page(page, 1);
1376 
1377 		/* if ipage exists, blkaddr should be NEW_ADDR */
1378 		f2fs_bug_on(F2FS_I_SB(inode), ipage);
1379 		page = f2fs_get_lock_data_page(inode, index, true);
1380 		if (IS_ERR(page))
1381 			return page;
1382 	}
1383 got_it:
1384 	if (new_i_size && i_size_read(inode) <
1385 				((loff_t)(index + 1) << PAGE_SHIFT))
1386 		f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1387 	return page;
1388 }
1389 
1390 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1391 {
1392 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1393 	struct f2fs_summary sum;
1394 	struct node_info ni;
1395 	block_t old_blkaddr;
1396 	blkcnt_t count = 1;
1397 	int err;
1398 
1399 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1400 		return -EPERM;
1401 
1402 	err = f2fs_get_node_info(sbi, dn->nid, &ni);
1403 	if (err)
1404 		return err;
1405 
1406 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
1407 	if (dn->data_blkaddr != NULL_ADDR)
1408 		goto alloc;
1409 
1410 	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1411 		return err;
1412 
1413 alloc:
1414 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1415 	old_blkaddr = dn->data_blkaddr;
1416 	f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1417 					&sum, seg_type, NULL, false);
1418 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1419 		invalidate_mapping_pages(META_MAPPING(sbi),
1420 					old_blkaddr, old_blkaddr);
1421 	f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1422 
1423 	/*
1424 	 * i_size will be updated by direct_IO. Otherwise, we'll get stale
1425 	 * data from unwritten block via dio_read.
1426 	 */
1427 	return 0;
1428 }
1429 
1430 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
1431 {
1432 	struct inode *inode = file_inode(iocb->ki_filp);
1433 	struct f2fs_map_blocks map;
1434 	int flag;
1435 	int err = 0;
1436 	bool direct_io = iocb->ki_flags & IOCB_DIRECT;
1437 
1438 	map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
1439 	map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
1440 	if (map.m_len > map.m_lblk)
1441 		map.m_len -= map.m_lblk;
1442 	else
1443 		map.m_len = 0;
1444 
1445 	map.m_next_pgofs = NULL;
1446 	map.m_next_extent = NULL;
1447 	map.m_seg_type = NO_CHECK_TYPE;
1448 	map.m_may_create = true;
1449 
1450 	if (direct_io) {
1451 		map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
1452 		flag = f2fs_force_buffered_io(inode, iocb, from) ?
1453 					F2FS_GET_BLOCK_PRE_AIO :
1454 					F2FS_GET_BLOCK_PRE_DIO;
1455 		goto map_blocks;
1456 	}
1457 	if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
1458 		err = f2fs_convert_inline_inode(inode);
1459 		if (err)
1460 			return err;
1461 	}
1462 	if (f2fs_has_inline_data(inode))
1463 		return err;
1464 
1465 	flag = F2FS_GET_BLOCK_PRE_AIO;
1466 
1467 map_blocks:
1468 	err = f2fs_map_blocks(inode, &map, 1, flag);
1469 	if (map.m_len > 0 && err == -ENOSPC) {
1470 		if (!direct_io)
1471 			set_inode_flag(inode, FI_NO_PREALLOC);
1472 		err = 0;
1473 	}
1474 	return err;
1475 }
1476 
1477 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1478 {
1479 	if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1480 		if (lock)
1481 			down_read(&sbi->node_change);
1482 		else
1483 			up_read(&sbi->node_change);
1484 	} else {
1485 		if (lock)
1486 			f2fs_lock_op(sbi);
1487 		else
1488 			f2fs_unlock_op(sbi);
1489 	}
1490 }
1491 
1492 /*
1493  * f2fs_map_blocks() tries to find or build mapping relationship which
1494  * maps continuous logical blocks to physical blocks, and return such
1495  * info via f2fs_map_blocks structure.
1496  */
1497 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1498 						int create, int flag)
1499 {
1500 	unsigned int maxblocks = map->m_len;
1501 	struct dnode_of_data dn;
1502 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1503 	int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1504 	pgoff_t pgofs, end_offset, end;
1505 	int err = 0, ofs = 1;
1506 	unsigned int ofs_in_node, last_ofs_in_node;
1507 	blkcnt_t prealloc;
1508 	struct extent_info ei = {0,0,0};
1509 	block_t blkaddr;
1510 	unsigned int start_pgofs;
1511 
1512 	if (!maxblocks)
1513 		return 0;
1514 
1515 	map->m_len = 0;
1516 	map->m_flags = 0;
1517 
1518 	/* it only supports block size == page size */
1519 	pgofs =	(pgoff_t)map->m_lblk;
1520 	end = pgofs + maxblocks;
1521 
1522 	if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1523 		if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1524 							map->m_may_create)
1525 			goto next_dnode;
1526 
1527 		map->m_pblk = ei.blk + pgofs - ei.fofs;
1528 		map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1529 		map->m_flags = F2FS_MAP_MAPPED;
1530 		if (map->m_next_extent)
1531 			*map->m_next_extent = pgofs + map->m_len;
1532 
1533 		/* for hardware encryption, but to avoid potential issue in future */
1534 		if (flag == F2FS_GET_BLOCK_DIO)
1535 			f2fs_wait_on_block_writeback_range(inode,
1536 						map->m_pblk, map->m_len);
1537 		goto out;
1538 	}
1539 
1540 next_dnode:
1541 	if (map->m_may_create)
1542 		__do_map_lock(sbi, flag, true);
1543 
1544 	/* When reading holes, we need its node page */
1545 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1546 	err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1547 	if (err) {
1548 		if (flag == F2FS_GET_BLOCK_BMAP)
1549 			map->m_pblk = 0;
1550 		if (err == -ENOENT) {
1551 			err = 0;
1552 			if (map->m_next_pgofs)
1553 				*map->m_next_pgofs =
1554 					f2fs_get_next_page_offset(&dn, pgofs);
1555 			if (map->m_next_extent)
1556 				*map->m_next_extent =
1557 					f2fs_get_next_page_offset(&dn, pgofs);
1558 		}
1559 		goto unlock_out;
1560 	}
1561 
1562 	start_pgofs = pgofs;
1563 	prealloc = 0;
1564 	last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1565 	end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1566 
1567 next_block:
1568 	blkaddr = f2fs_data_blkaddr(&dn);
1569 
1570 	if (__is_valid_data_blkaddr(blkaddr) &&
1571 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1572 		err = -EFSCORRUPTED;
1573 		goto sync_out;
1574 	}
1575 
1576 	if (__is_valid_data_blkaddr(blkaddr)) {
1577 		/* use out-place-update for driect IO under LFS mode */
1578 		if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1579 							map->m_may_create) {
1580 			err = __allocate_data_block(&dn, map->m_seg_type);
1581 			if (err)
1582 				goto sync_out;
1583 			blkaddr = dn.data_blkaddr;
1584 			set_inode_flag(inode, FI_APPEND_WRITE);
1585 		}
1586 	} else {
1587 		if (create) {
1588 			if (unlikely(f2fs_cp_error(sbi))) {
1589 				err = -EIO;
1590 				goto sync_out;
1591 			}
1592 			if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1593 				if (blkaddr == NULL_ADDR) {
1594 					prealloc++;
1595 					last_ofs_in_node = dn.ofs_in_node;
1596 				}
1597 			} else {
1598 				WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1599 					flag != F2FS_GET_BLOCK_DIO);
1600 				err = __allocate_data_block(&dn,
1601 							map->m_seg_type);
1602 				if (!err)
1603 					set_inode_flag(inode, FI_APPEND_WRITE);
1604 			}
1605 			if (err)
1606 				goto sync_out;
1607 			map->m_flags |= F2FS_MAP_NEW;
1608 			blkaddr = dn.data_blkaddr;
1609 		} else {
1610 			if (flag == F2FS_GET_BLOCK_BMAP) {
1611 				map->m_pblk = 0;
1612 				goto sync_out;
1613 			}
1614 			if (flag == F2FS_GET_BLOCK_PRECACHE)
1615 				goto sync_out;
1616 			if (flag == F2FS_GET_BLOCK_FIEMAP &&
1617 						blkaddr == NULL_ADDR) {
1618 				if (map->m_next_pgofs)
1619 					*map->m_next_pgofs = pgofs + 1;
1620 				goto sync_out;
1621 			}
1622 			if (flag != F2FS_GET_BLOCK_FIEMAP) {
1623 				/* for defragment case */
1624 				if (map->m_next_pgofs)
1625 					*map->m_next_pgofs = pgofs + 1;
1626 				goto sync_out;
1627 			}
1628 		}
1629 	}
1630 
1631 	if (flag == F2FS_GET_BLOCK_PRE_AIO)
1632 		goto skip;
1633 
1634 	if (map->m_len == 0) {
1635 		/* preallocated unwritten block should be mapped for fiemap. */
1636 		if (blkaddr == NEW_ADDR)
1637 			map->m_flags |= F2FS_MAP_UNWRITTEN;
1638 		map->m_flags |= F2FS_MAP_MAPPED;
1639 
1640 		map->m_pblk = blkaddr;
1641 		map->m_len = 1;
1642 	} else if ((map->m_pblk != NEW_ADDR &&
1643 			blkaddr == (map->m_pblk + ofs)) ||
1644 			(map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1645 			flag == F2FS_GET_BLOCK_PRE_DIO) {
1646 		ofs++;
1647 		map->m_len++;
1648 	} else {
1649 		goto sync_out;
1650 	}
1651 
1652 skip:
1653 	dn.ofs_in_node++;
1654 	pgofs++;
1655 
1656 	/* preallocate blocks in batch for one dnode page */
1657 	if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1658 			(pgofs == end || dn.ofs_in_node == end_offset)) {
1659 
1660 		dn.ofs_in_node = ofs_in_node;
1661 		err = f2fs_reserve_new_blocks(&dn, prealloc);
1662 		if (err)
1663 			goto sync_out;
1664 
1665 		map->m_len += dn.ofs_in_node - ofs_in_node;
1666 		if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1667 			err = -ENOSPC;
1668 			goto sync_out;
1669 		}
1670 		dn.ofs_in_node = end_offset;
1671 	}
1672 
1673 	if (pgofs >= end)
1674 		goto sync_out;
1675 	else if (dn.ofs_in_node < end_offset)
1676 		goto next_block;
1677 
1678 	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1679 		if (map->m_flags & F2FS_MAP_MAPPED) {
1680 			unsigned int ofs = start_pgofs - map->m_lblk;
1681 
1682 			f2fs_update_extent_cache_range(&dn,
1683 				start_pgofs, map->m_pblk + ofs,
1684 				map->m_len - ofs);
1685 		}
1686 	}
1687 
1688 	f2fs_put_dnode(&dn);
1689 
1690 	if (map->m_may_create) {
1691 		__do_map_lock(sbi, flag, false);
1692 		f2fs_balance_fs(sbi, dn.node_changed);
1693 	}
1694 	goto next_dnode;
1695 
1696 sync_out:
1697 
1698 	/* for hardware encryption, but to avoid potential issue in future */
1699 	if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1700 		f2fs_wait_on_block_writeback_range(inode,
1701 						map->m_pblk, map->m_len);
1702 
1703 	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1704 		if (map->m_flags & F2FS_MAP_MAPPED) {
1705 			unsigned int ofs = start_pgofs - map->m_lblk;
1706 
1707 			f2fs_update_extent_cache_range(&dn,
1708 				start_pgofs, map->m_pblk + ofs,
1709 				map->m_len - ofs);
1710 		}
1711 		if (map->m_next_extent)
1712 			*map->m_next_extent = pgofs + 1;
1713 	}
1714 	f2fs_put_dnode(&dn);
1715 unlock_out:
1716 	if (map->m_may_create) {
1717 		__do_map_lock(sbi, flag, false);
1718 		f2fs_balance_fs(sbi, dn.node_changed);
1719 	}
1720 out:
1721 	trace_f2fs_map_blocks(inode, map, err);
1722 	return err;
1723 }
1724 
1725 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1726 {
1727 	struct f2fs_map_blocks map;
1728 	block_t last_lblk;
1729 	int err;
1730 
1731 	if (pos + len > i_size_read(inode))
1732 		return false;
1733 
1734 	map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1735 	map.m_next_pgofs = NULL;
1736 	map.m_next_extent = NULL;
1737 	map.m_seg_type = NO_CHECK_TYPE;
1738 	map.m_may_create = false;
1739 	last_lblk = F2FS_BLK_ALIGN(pos + len);
1740 
1741 	while (map.m_lblk < last_lblk) {
1742 		map.m_len = last_lblk - map.m_lblk;
1743 		err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1744 		if (err || map.m_len == 0)
1745 			return false;
1746 		map.m_lblk += map.m_len;
1747 	}
1748 	return true;
1749 }
1750 
1751 static int __get_data_block(struct inode *inode, sector_t iblock,
1752 			struct buffer_head *bh, int create, int flag,
1753 			pgoff_t *next_pgofs, int seg_type, bool may_write)
1754 {
1755 	struct f2fs_map_blocks map;
1756 	int err;
1757 
1758 	map.m_lblk = iblock;
1759 	map.m_len = bh->b_size >> inode->i_blkbits;
1760 	map.m_next_pgofs = next_pgofs;
1761 	map.m_next_extent = NULL;
1762 	map.m_seg_type = seg_type;
1763 	map.m_may_create = may_write;
1764 
1765 	err = f2fs_map_blocks(inode, &map, create, flag);
1766 	if (!err) {
1767 		map_bh(bh, inode->i_sb, map.m_pblk);
1768 		bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1769 		bh->b_size = (u64)map.m_len << inode->i_blkbits;
1770 	}
1771 	return err;
1772 }
1773 
1774 static int get_data_block(struct inode *inode, sector_t iblock,
1775 			struct buffer_head *bh_result, int create, int flag,
1776 			pgoff_t *next_pgofs)
1777 {
1778 	return __get_data_block(inode, iblock, bh_result, create,
1779 							flag, next_pgofs,
1780 							NO_CHECK_TYPE, create);
1781 }
1782 
1783 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1784 			struct buffer_head *bh_result, int create)
1785 {
1786 	return __get_data_block(inode, iblock, bh_result, create,
1787 				F2FS_GET_BLOCK_DIO, NULL,
1788 				f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1789 				IS_SWAPFILE(inode) ? false : true);
1790 }
1791 
1792 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1793 			struct buffer_head *bh_result, int create)
1794 {
1795 	return __get_data_block(inode, iblock, bh_result, create,
1796 				F2FS_GET_BLOCK_DIO, NULL,
1797 				f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1798 				false);
1799 }
1800 
1801 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1802 			struct buffer_head *bh_result, int create)
1803 {
1804 	/* Block number less than F2FS MAX BLOCKS */
1805 	if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1806 		return -EFBIG;
1807 
1808 	return __get_data_block(inode, iblock, bh_result, create,
1809 						F2FS_GET_BLOCK_BMAP, NULL,
1810 						NO_CHECK_TYPE, create);
1811 }
1812 
1813 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1814 {
1815 	return (offset >> inode->i_blkbits);
1816 }
1817 
1818 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1819 {
1820 	return (blk << inode->i_blkbits);
1821 }
1822 
1823 static int f2fs_xattr_fiemap(struct inode *inode,
1824 				struct fiemap_extent_info *fieinfo)
1825 {
1826 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1827 	struct page *page;
1828 	struct node_info ni;
1829 	__u64 phys = 0, len;
1830 	__u32 flags;
1831 	nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1832 	int err = 0;
1833 
1834 	if (f2fs_has_inline_xattr(inode)) {
1835 		int offset;
1836 
1837 		page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1838 						inode->i_ino, false);
1839 		if (!page)
1840 			return -ENOMEM;
1841 
1842 		err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1843 		if (err) {
1844 			f2fs_put_page(page, 1);
1845 			return err;
1846 		}
1847 
1848 		phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1849 		offset = offsetof(struct f2fs_inode, i_addr) +
1850 					sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1851 					get_inline_xattr_addrs(inode));
1852 
1853 		phys += offset;
1854 		len = inline_xattr_size(inode);
1855 
1856 		f2fs_put_page(page, 1);
1857 
1858 		flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1859 
1860 		if (!xnid)
1861 			flags |= FIEMAP_EXTENT_LAST;
1862 
1863 		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1864 		if (err || err == 1)
1865 			return err;
1866 	}
1867 
1868 	if (xnid) {
1869 		page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1870 		if (!page)
1871 			return -ENOMEM;
1872 
1873 		err = f2fs_get_node_info(sbi, xnid, &ni);
1874 		if (err) {
1875 			f2fs_put_page(page, 1);
1876 			return err;
1877 		}
1878 
1879 		phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1880 		len = inode->i_sb->s_blocksize;
1881 
1882 		f2fs_put_page(page, 1);
1883 
1884 		flags = FIEMAP_EXTENT_LAST;
1885 	}
1886 
1887 	if (phys)
1888 		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1889 
1890 	return (err < 0 ? err : 0);
1891 }
1892 
1893 static loff_t max_inode_blocks(struct inode *inode)
1894 {
1895 	loff_t result = ADDRS_PER_INODE(inode);
1896 	loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1897 
1898 	/* two direct node blocks */
1899 	result += (leaf_count * 2);
1900 
1901 	/* two indirect node blocks */
1902 	leaf_count *= NIDS_PER_BLOCK;
1903 	result += (leaf_count * 2);
1904 
1905 	/* one double indirect node block */
1906 	leaf_count *= NIDS_PER_BLOCK;
1907 	result += leaf_count;
1908 
1909 	return result;
1910 }
1911 
1912 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1913 		u64 start, u64 len)
1914 {
1915 	struct buffer_head map_bh;
1916 	sector_t start_blk, last_blk;
1917 	pgoff_t next_pgofs;
1918 	u64 logical = 0, phys = 0, size = 0;
1919 	u32 flags = 0;
1920 	int ret = 0;
1921 	bool compr_cluster = false;
1922 	unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1923 
1924 	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1925 		ret = f2fs_precache_extents(inode);
1926 		if (ret)
1927 			return ret;
1928 	}
1929 
1930 	ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1931 	if (ret)
1932 		return ret;
1933 
1934 	inode_lock(inode);
1935 
1936 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1937 		ret = f2fs_xattr_fiemap(inode, fieinfo);
1938 		goto out;
1939 	}
1940 
1941 	if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1942 		ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1943 		if (ret != -EAGAIN)
1944 			goto out;
1945 	}
1946 
1947 	if (logical_to_blk(inode, len) == 0)
1948 		len = blk_to_logical(inode, 1);
1949 
1950 	start_blk = logical_to_blk(inode, start);
1951 	last_blk = logical_to_blk(inode, start + len - 1);
1952 
1953 next:
1954 	memset(&map_bh, 0, sizeof(struct buffer_head));
1955 	map_bh.b_size = len;
1956 
1957 	if (compr_cluster)
1958 		map_bh.b_size = blk_to_logical(inode, cluster_size - 1);
1959 
1960 	ret = get_data_block(inode, start_blk, &map_bh, 0,
1961 					F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1962 	if (ret)
1963 		goto out;
1964 
1965 	/* HOLE */
1966 	if (!buffer_mapped(&map_bh)) {
1967 		start_blk = next_pgofs;
1968 
1969 		if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1970 						max_inode_blocks(inode)))
1971 			goto prep_next;
1972 
1973 		flags |= FIEMAP_EXTENT_LAST;
1974 	}
1975 
1976 	if (size) {
1977 		if (IS_ENCRYPTED(inode))
1978 			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1979 
1980 		ret = fiemap_fill_next_extent(fieinfo, logical,
1981 				phys, size, flags);
1982 		if (ret)
1983 			goto out;
1984 		size = 0;
1985 	}
1986 
1987 	if (start_blk > last_blk)
1988 		goto out;
1989 
1990 	if (compr_cluster) {
1991 		compr_cluster = false;
1992 
1993 
1994 		logical = blk_to_logical(inode, start_blk - 1);
1995 		phys = blk_to_logical(inode, map_bh.b_blocknr);
1996 		size = blk_to_logical(inode, cluster_size);
1997 
1998 		flags |= FIEMAP_EXTENT_ENCODED;
1999 
2000 		start_blk += cluster_size - 1;
2001 
2002 		if (start_blk > last_blk)
2003 			goto out;
2004 
2005 		goto prep_next;
2006 	}
2007 
2008 	if (map_bh.b_blocknr == COMPRESS_ADDR) {
2009 		compr_cluster = true;
2010 		start_blk++;
2011 		goto prep_next;
2012 	}
2013 
2014 	logical = blk_to_logical(inode, start_blk);
2015 	phys = blk_to_logical(inode, map_bh.b_blocknr);
2016 	size = map_bh.b_size;
2017 	flags = 0;
2018 	if (buffer_unwritten(&map_bh))
2019 		flags = FIEMAP_EXTENT_UNWRITTEN;
2020 
2021 	start_blk += logical_to_blk(inode, size);
2022 
2023 prep_next:
2024 	cond_resched();
2025 	if (fatal_signal_pending(current))
2026 		ret = -EINTR;
2027 	else
2028 		goto next;
2029 out:
2030 	if (ret == 1)
2031 		ret = 0;
2032 
2033 	inode_unlock(inode);
2034 	return ret;
2035 }
2036 
2037 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2038 {
2039 	if (IS_ENABLED(CONFIG_FS_VERITY) &&
2040 	    (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
2041 		return inode->i_sb->s_maxbytes;
2042 
2043 	return i_size_read(inode);
2044 }
2045 
2046 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2047 					unsigned nr_pages,
2048 					struct f2fs_map_blocks *map,
2049 					struct bio **bio_ret,
2050 					sector_t *last_block_in_bio,
2051 					bool is_readahead)
2052 {
2053 	struct bio *bio = *bio_ret;
2054 	const unsigned blkbits = inode->i_blkbits;
2055 	const unsigned blocksize = 1 << blkbits;
2056 	sector_t block_in_file;
2057 	sector_t last_block;
2058 	sector_t last_block_in_file;
2059 	sector_t block_nr;
2060 	int ret = 0;
2061 
2062 	block_in_file = (sector_t)page_index(page);
2063 	last_block = block_in_file + nr_pages;
2064 	last_block_in_file = (f2fs_readpage_limit(inode) + blocksize - 1) >>
2065 							blkbits;
2066 	if (last_block > last_block_in_file)
2067 		last_block = last_block_in_file;
2068 
2069 	/* just zeroing out page which is beyond EOF */
2070 	if (block_in_file >= last_block)
2071 		goto zero_out;
2072 	/*
2073 	 * Map blocks using the previous result first.
2074 	 */
2075 	if ((map->m_flags & F2FS_MAP_MAPPED) &&
2076 			block_in_file > map->m_lblk &&
2077 			block_in_file < (map->m_lblk + map->m_len))
2078 		goto got_it;
2079 
2080 	/*
2081 	 * Then do more f2fs_map_blocks() calls until we are
2082 	 * done with this page.
2083 	 */
2084 	map->m_lblk = block_in_file;
2085 	map->m_len = last_block - block_in_file;
2086 
2087 	ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2088 	if (ret)
2089 		goto out;
2090 got_it:
2091 	if ((map->m_flags & F2FS_MAP_MAPPED)) {
2092 		block_nr = map->m_pblk + block_in_file - map->m_lblk;
2093 		SetPageMappedToDisk(page);
2094 
2095 		if (!PageUptodate(page) && (!PageSwapCache(page) &&
2096 					!cleancache_get_page(page))) {
2097 			SetPageUptodate(page);
2098 			goto confused;
2099 		}
2100 
2101 		if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2102 						DATA_GENERIC_ENHANCE_READ)) {
2103 			ret = -EFSCORRUPTED;
2104 			goto out;
2105 		}
2106 	} else {
2107 zero_out:
2108 		zero_user_segment(page, 0, PAGE_SIZE);
2109 		if (f2fs_need_verity(inode, page->index) &&
2110 		    !fsverity_verify_page(page)) {
2111 			ret = -EIO;
2112 			goto out;
2113 		}
2114 		if (!PageUptodate(page))
2115 			SetPageUptodate(page);
2116 		unlock_page(page);
2117 		goto out;
2118 	}
2119 
2120 	/*
2121 	 * This page will go to BIO.  Do we need to send this
2122 	 * BIO off first?
2123 	 */
2124 	if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2125 				       *last_block_in_bio, block_nr) ||
2126 		    !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2127 submit_and_realloc:
2128 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
2129 		bio = NULL;
2130 	}
2131 	if (bio == NULL) {
2132 		bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2133 				is_readahead ? REQ_RAHEAD : 0, page->index,
2134 				false);
2135 		if (IS_ERR(bio)) {
2136 			ret = PTR_ERR(bio);
2137 			bio = NULL;
2138 			goto out;
2139 		}
2140 	}
2141 
2142 	/*
2143 	 * If the page is under writeback, we need to wait for
2144 	 * its completion to see the correct decrypted data.
2145 	 */
2146 	f2fs_wait_on_block_writeback(inode, block_nr);
2147 
2148 	if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2149 		goto submit_and_realloc;
2150 
2151 	inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2152 	f2fs_update_iostat(F2FS_I_SB(inode), FS_DATA_READ_IO, F2FS_BLKSIZE);
2153 	ClearPageError(page);
2154 	*last_block_in_bio = block_nr;
2155 	goto out;
2156 confused:
2157 	if (bio) {
2158 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
2159 		bio = NULL;
2160 	}
2161 	unlock_page(page);
2162 out:
2163 	*bio_ret = bio;
2164 	return ret;
2165 }
2166 
2167 #ifdef CONFIG_F2FS_FS_COMPRESSION
2168 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2169 				unsigned nr_pages, sector_t *last_block_in_bio,
2170 				bool is_readahead, bool for_write)
2171 {
2172 	struct dnode_of_data dn;
2173 	struct inode *inode = cc->inode;
2174 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2175 	struct bio *bio = *bio_ret;
2176 	unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2177 	sector_t last_block_in_file;
2178 	const unsigned blkbits = inode->i_blkbits;
2179 	const unsigned blocksize = 1 << blkbits;
2180 	struct decompress_io_ctx *dic = NULL;
2181 	int i;
2182 	int ret = 0;
2183 
2184 	f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2185 
2186 	last_block_in_file = (f2fs_readpage_limit(inode) +
2187 					blocksize - 1) >> blkbits;
2188 
2189 	/* get rid of pages beyond EOF */
2190 	for (i = 0; i < cc->cluster_size; i++) {
2191 		struct page *page = cc->rpages[i];
2192 
2193 		if (!page)
2194 			continue;
2195 		if ((sector_t)page->index >= last_block_in_file) {
2196 			zero_user_segment(page, 0, PAGE_SIZE);
2197 			if (!PageUptodate(page))
2198 				SetPageUptodate(page);
2199 		} else if (!PageUptodate(page)) {
2200 			continue;
2201 		}
2202 		unlock_page(page);
2203 		cc->rpages[i] = NULL;
2204 		cc->nr_rpages--;
2205 	}
2206 
2207 	/* we are done since all pages are beyond EOF */
2208 	if (f2fs_cluster_is_empty(cc))
2209 		goto out;
2210 
2211 	set_new_dnode(&dn, inode, NULL, NULL, 0);
2212 	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2213 	if (ret)
2214 		goto out;
2215 
2216 	/* cluster was overwritten as normal cluster */
2217 	if (dn.data_blkaddr != COMPRESS_ADDR)
2218 		goto out;
2219 
2220 	for (i = 1; i < cc->cluster_size; i++) {
2221 		block_t blkaddr;
2222 
2223 		blkaddr = data_blkaddr(dn.inode, dn.node_page,
2224 						dn.ofs_in_node + i);
2225 
2226 		if (!__is_valid_data_blkaddr(blkaddr))
2227 			break;
2228 
2229 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2230 			ret = -EFAULT;
2231 			goto out_put_dnode;
2232 		}
2233 		cc->nr_cpages++;
2234 	}
2235 
2236 	/* nothing to decompress */
2237 	if (cc->nr_cpages == 0) {
2238 		ret = 0;
2239 		goto out_put_dnode;
2240 	}
2241 
2242 	dic = f2fs_alloc_dic(cc);
2243 	if (IS_ERR(dic)) {
2244 		ret = PTR_ERR(dic);
2245 		goto out_put_dnode;
2246 	}
2247 
2248 	for (i = 0; i < dic->nr_cpages; i++) {
2249 		struct page *page = dic->cpages[i];
2250 		block_t blkaddr;
2251 		struct bio_post_read_ctx *ctx;
2252 
2253 		blkaddr = data_blkaddr(dn.inode, dn.node_page,
2254 						dn.ofs_in_node + i + 1);
2255 
2256 		if (bio && (!page_is_mergeable(sbi, bio,
2257 					*last_block_in_bio, blkaddr) ||
2258 		    !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2259 submit_and_realloc:
2260 			__submit_bio(sbi, bio, DATA);
2261 			bio = NULL;
2262 		}
2263 
2264 		if (!bio) {
2265 			bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2266 					is_readahead ? REQ_RAHEAD : 0,
2267 					page->index, for_write);
2268 			if (IS_ERR(bio)) {
2269 				ret = PTR_ERR(bio);
2270 				dic->failed = true;
2271 				if (refcount_sub_and_test(dic->nr_cpages - i,
2272 							&dic->ref)) {
2273 					f2fs_decompress_end_io(dic->rpages,
2274 							cc->cluster_size, true,
2275 							false);
2276 					f2fs_free_dic(dic);
2277 				}
2278 				f2fs_put_dnode(&dn);
2279 				*bio_ret = NULL;
2280 				return ret;
2281 			}
2282 		}
2283 
2284 		f2fs_wait_on_block_writeback(inode, blkaddr);
2285 
2286 		if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2287 			goto submit_and_realloc;
2288 
2289 		/* tag STEP_DECOMPRESS to handle IO in wq */
2290 		ctx = bio->bi_private;
2291 		if (!(ctx->enabled_steps & (1 << STEP_DECOMPRESS)))
2292 			ctx->enabled_steps |= 1 << STEP_DECOMPRESS;
2293 
2294 		inc_page_count(sbi, F2FS_RD_DATA);
2295 		f2fs_update_iostat(sbi, FS_DATA_READ_IO, F2FS_BLKSIZE);
2296 		f2fs_update_iostat(sbi, FS_CDATA_READ_IO, F2FS_BLKSIZE);
2297 		ClearPageError(page);
2298 		*last_block_in_bio = blkaddr;
2299 	}
2300 
2301 	f2fs_put_dnode(&dn);
2302 
2303 	*bio_ret = bio;
2304 	return 0;
2305 
2306 out_put_dnode:
2307 	f2fs_put_dnode(&dn);
2308 out:
2309 	f2fs_decompress_end_io(cc->rpages, cc->cluster_size, true, false);
2310 	*bio_ret = bio;
2311 	return ret;
2312 }
2313 #endif
2314 
2315 /*
2316  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2317  * Major change was from block_size == page_size in f2fs by default.
2318  *
2319  * Note that the aops->readpages() function is ONLY used for read-ahead. If
2320  * this function ever deviates from doing just read-ahead, it should either
2321  * use ->readpage() or do the necessary surgery to decouple ->readpages()
2322  * from read-ahead.
2323  */
2324 static int f2fs_mpage_readpages(struct inode *inode,
2325 		struct readahead_control *rac, struct page *page)
2326 {
2327 	struct bio *bio = NULL;
2328 	sector_t last_block_in_bio = 0;
2329 	struct f2fs_map_blocks map;
2330 #ifdef CONFIG_F2FS_FS_COMPRESSION
2331 	struct compress_ctx cc = {
2332 		.inode = inode,
2333 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2334 		.cluster_size = F2FS_I(inode)->i_cluster_size,
2335 		.cluster_idx = NULL_CLUSTER,
2336 		.rpages = NULL,
2337 		.cpages = NULL,
2338 		.nr_rpages = 0,
2339 		.nr_cpages = 0,
2340 	};
2341 #endif
2342 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
2343 	unsigned max_nr_pages = nr_pages;
2344 	int ret = 0;
2345 
2346 	map.m_pblk = 0;
2347 	map.m_lblk = 0;
2348 	map.m_len = 0;
2349 	map.m_flags = 0;
2350 	map.m_next_pgofs = NULL;
2351 	map.m_next_extent = NULL;
2352 	map.m_seg_type = NO_CHECK_TYPE;
2353 	map.m_may_create = false;
2354 
2355 	for (; nr_pages; nr_pages--) {
2356 		if (rac) {
2357 			page = readahead_page(rac);
2358 			prefetchw(&page->flags);
2359 		}
2360 
2361 #ifdef CONFIG_F2FS_FS_COMPRESSION
2362 		if (f2fs_compressed_file(inode)) {
2363 			/* there are remained comressed pages, submit them */
2364 			if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2365 				ret = f2fs_read_multi_pages(&cc, &bio,
2366 							max_nr_pages,
2367 							&last_block_in_bio,
2368 							rac != NULL, false);
2369 				f2fs_destroy_compress_ctx(&cc);
2370 				if (ret)
2371 					goto set_error_page;
2372 			}
2373 			ret = f2fs_is_compressed_cluster(inode, page->index);
2374 			if (ret < 0)
2375 				goto set_error_page;
2376 			else if (!ret)
2377 				goto read_single_page;
2378 
2379 			ret = f2fs_init_compress_ctx(&cc);
2380 			if (ret)
2381 				goto set_error_page;
2382 
2383 			f2fs_compress_ctx_add_page(&cc, page);
2384 
2385 			goto next_page;
2386 		}
2387 read_single_page:
2388 #endif
2389 
2390 		ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2391 					&bio, &last_block_in_bio, rac);
2392 		if (ret) {
2393 #ifdef CONFIG_F2FS_FS_COMPRESSION
2394 set_error_page:
2395 #endif
2396 			SetPageError(page);
2397 			zero_user_segment(page, 0, PAGE_SIZE);
2398 			unlock_page(page);
2399 		}
2400 #ifdef CONFIG_F2FS_FS_COMPRESSION
2401 next_page:
2402 #endif
2403 		if (rac)
2404 			put_page(page);
2405 
2406 #ifdef CONFIG_F2FS_FS_COMPRESSION
2407 		if (f2fs_compressed_file(inode)) {
2408 			/* last page */
2409 			if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2410 				ret = f2fs_read_multi_pages(&cc, &bio,
2411 							max_nr_pages,
2412 							&last_block_in_bio,
2413 							rac != NULL, false);
2414 				f2fs_destroy_compress_ctx(&cc);
2415 			}
2416 		}
2417 #endif
2418 	}
2419 	if (bio)
2420 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
2421 	return ret;
2422 }
2423 
2424 static int f2fs_read_data_page(struct file *file, struct page *page)
2425 {
2426 	struct inode *inode = page_file_mapping(page)->host;
2427 	int ret = -EAGAIN;
2428 
2429 	trace_f2fs_readpage(page, DATA);
2430 
2431 	if (!f2fs_is_compress_backend_ready(inode)) {
2432 		unlock_page(page);
2433 		return -EOPNOTSUPP;
2434 	}
2435 
2436 	/* If the file has inline data, try to read it directly */
2437 	if (f2fs_has_inline_data(inode))
2438 		ret = f2fs_read_inline_data(inode, page);
2439 	if (ret == -EAGAIN)
2440 		ret = f2fs_mpage_readpages(inode, NULL, page);
2441 	return ret;
2442 }
2443 
2444 static void f2fs_readahead(struct readahead_control *rac)
2445 {
2446 	struct inode *inode = rac->mapping->host;
2447 
2448 	trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2449 
2450 	if (!f2fs_is_compress_backend_ready(inode))
2451 		return;
2452 
2453 	/* If the file has inline data, skip readpages */
2454 	if (f2fs_has_inline_data(inode))
2455 		return;
2456 
2457 	f2fs_mpage_readpages(inode, rac, NULL);
2458 }
2459 
2460 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2461 {
2462 	struct inode *inode = fio->page->mapping->host;
2463 	struct page *mpage, *page;
2464 	gfp_t gfp_flags = GFP_NOFS;
2465 
2466 	if (!f2fs_encrypted_file(inode))
2467 		return 0;
2468 
2469 	page = fio->compressed_page ? fio->compressed_page : fio->page;
2470 
2471 	/* wait for GCed page writeback via META_MAPPING */
2472 	f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2473 
2474 	if (fscrypt_inode_uses_inline_crypto(inode))
2475 		return 0;
2476 
2477 retry_encrypt:
2478 	fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2479 					PAGE_SIZE, 0, gfp_flags);
2480 	if (IS_ERR(fio->encrypted_page)) {
2481 		/* flush pending IOs and wait for a while in the ENOMEM case */
2482 		if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2483 			f2fs_flush_merged_writes(fio->sbi);
2484 			congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2485 			gfp_flags |= __GFP_NOFAIL;
2486 			goto retry_encrypt;
2487 		}
2488 		return PTR_ERR(fio->encrypted_page);
2489 	}
2490 
2491 	mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2492 	if (mpage) {
2493 		if (PageUptodate(mpage))
2494 			memcpy(page_address(mpage),
2495 				page_address(fio->encrypted_page), PAGE_SIZE);
2496 		f2fs_put_page(mpage, 1);
2497 	}
2498 	return 0;
2499 }
2500 
2501 static inline bool check_inplace_update_policy(struct inode *inode,
2502 				struct f2fs_io_info *fio)
2503 {
2504 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2505 	unsigned int policy = SM_I(sbi)->ipu_policy;
2506 
2507 	if (policy & (0x1 << F2FS_IPU_FORCE))
2508 		return true;
2509 	if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2510 		return true;
2511 	if (policy & (0x1 << F2FS_IPU_UTIL) &&
2512 			utilization(sbi) > SM_I(sbi)->min_ipu_util)
2513 		return true;
2514 	if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2515 			utilization(sbi) > SM_I(sbi)->min_ipu_util)
2516 		return true;
2517 
2518 	/*
2519 	 * IPU for rewrite async pages
2520 	 */
2521 	if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2522 			fio && fio->op == REQ_OP_WRITE &&
2523 			!(fio->op_flags & REQ_SYNC) &&
2524 			!IS_ENCRYPTED(inode))
2525 		return true;
2526 
2527 	/* this is only set during fdatasync */
2528 	if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2529 			is_inode_flag_set(inode, FI_NEED_IPU))
2530 		return true;
2531 
2532 	if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2533 			!f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2534 		return true;
2535 
2536 	return false;
2537 }
2538 
2539 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2540 {
2541 	if (f2fs_is_pinned_file(inode))
2542 		return true;
2543 
2544 	/* if this is cold file, we should overwrite to avoid fragmentation */
2545 	if (file_is_cold(inode))
2546 		return true;
2547 
2548 	return check_inplace_update_policy(inode, fio);
2549 }
2550 
2551 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2552 {
2553 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2554 
2555 	if (f2fs_lfs_mode(sbi))
2556 		return true;
2557 	if (S_ISDIR(inode->i_mode))
2558 		return true;
2559 	if (IS_NOQUOTA(inode))
2560 		return true;
2561 	if (f2fs_is_atomic_file(inode))
2562 		return true;
2563 	if (fio) {
2564 		if (is_cold_data(fio->page))
2565 			return true;
2566 		if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
2567 			return true;
2568 		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2569 			f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2570 			return true;
2571 	}
2572 	return false;
2573 }
2574 
2575 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2576 {
2577 	struct inode *inode = fio->page->mapping->host;
2578 
2579 	if (f2fs_should_update_outplace(inode, fio))
2580 		return false;
2581 
2582 	return f2fs_should_update_inplace(inode, fio);
2583 }
2584 
2585 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2586 {
2587 	struct page *page = fio->page;
2588 	struct inode *inode = page->mapping->host;
2589 	struct dnode_of_data dn;
2590 	struct extent_info ei = {0,0,0};
2591 	struct node_info ni;
2592 	bool ipu_force = false;
2593 	int err = 0;
2594 
2595 	set_new_dnode(&dn, inode, NULL, NULL, 0);
2596 	if (need_inplace_update(fio) &&
2597 			f2fs_lookup_extent_cache(inode, page->index, &ei)) {
2598 		fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2599 
2600 		if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2601 						DATA_GENERIC_ENHANCE))
2602 			return -EFSCORRUPTED;
2603 
2604 		ipu_force = true;
2605 		fio->need_lock = LOCK_DONE;
2606 		goto got_it;
2607 	}
2608 
2609 	/* Deadlock due to between page->lock and f2fs_lock_op */
2610 	if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2611 		return -EAGAIN;
2612 
2613 	err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2614 	if (err)
2615 		goto out;
2616 
2617 	fio->old_blkaddr = dn.data_blkaddr;
2618 
2619 	/* This page is already truncated */
2620 	if (fio->old_blkaddr == NULL_ADDR) {
2621 		ClearPageUptodate(page);
2622 		clear_cold_data(page);
2623 		goto out_writepage;
2624 	}
2625 got_it:
2626 	if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2627 		!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2628 						DATA_GENERIC_ENHANCE)) {
2629 		err = -EFSCORRUPTED;
2630 		goto out_writepage;
2631 	}
2632 	/*
2633 	 * If current allocation needs SSR,
2634 	 * it had better in-place writes for updated data.
2635 	 */
2636 	if (ipu_force ||
2637 		(__is_valid_data_blkaddr(fio->old_blkaddr) &&
2638 					need_inplace_update(fio))) {
2639 		err = f2fs_encrypt_one_page(fio);
2640 		if (err)
2641 			goto out_writepage;
2642 
2643 		set_page_writeback(page);
2644 		ClearPageError(page);
2645 		f2fs_put_dnode(&dn);
2646 		if (fio->need_lock == LOCK_REQ)
2647 			f2fs_unlock_op(fio->sbi);
2648 		err = f2fs_inplace_write_data(fio);
2649 		if (err) {
2650 			if (fscrypt_inode_uses_fs_layer_crypto(inode))
2651 				fscrypt_finalize_bounce_page(&fio->encrypted_page);
2652 			if (PageWriteback(page))
2653 				end_page_writeback(page);
2654 		} else {
2655 			set_inode_flag(inode, FI_UPDATE_WRITE);
2656 		}
2657 		trace_f2fs_do_write_data_page(fio->page, IPU);
2658 		return err;
2659 	}
2660 
2661 	if (fio->need_lock == LOCK_RETRY) {
2662 		if (!f2fs_trylock_op(fio->sbi)) {
2663 			err = -EAGAIN;
2664 			goto out_writepage;
2665 		}
2666 		fio->need_lock = LOCK_REQ;
2667 	}
2668 
2669 	err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
2670 	if (err)
2671 		goto out_writepage;
2672 
2673 	fio->version = ni.version;
2674 
2675 	err = f2fs_encrypt_one_page(fio);
2676 	if (err)
2677 		goto out_writepage;
2678 
2679 	set_page_writeback(page);
2680 	ClearPageError(page);
2681 
2682 	if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2683 		f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2684 
2685 	/* LFS mode write path */
2686 	f2fs_outplace_write_data(&dn, fio);
2687 	trace_f2fs_do_write_data_page(page, OPU);
2688 	set_inode_flag(inode, FI_APPEND_WRITE);
2689 	if (page->index == 0)
2690 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2691 out_writepage:
2692 	f2fs_put_dnode(&dn);
2693 out:
2694 	if (fio->need_lock == LOCK_REQ)
2695 		f2fs_unlock_op(fio->sbi);
2696 	return err;
2697 }
2698 
2699 int f2fs_write_single_data_page(struct page *page, int *submitted,
2700 				struct bio **bio,
2701 				sector_t *last_block,
2702 				struct writeback_control *wbc,
2703 				enum iostat_type io_type,
2704 				int compr_blocks)
2705 {
2706 	struct inode *inode = page->mapping->host;
2707 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2708 	loff_t i_size = i_size_read(inode);
2709 	const pgoff_t end_index = ((unsigned long long)i_size)
2710 							>> PAGE_SHIFT;
2711 	loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2712 	unsigned offset = 0;
2713 	bool need_balance_fs = false;
2714 	int err = 0;
2715 	struct f2fs_io_info fio = {
2716 		.sbi = sbi,
2717 		.ino = inode->i_ino,
2718 		.type = DATA,
2719 		.op = REQ_OP_WRITE,
2720 		.op_flags = wbc_to_write_flags(wbc),
2721 		.old_blkaddr = NULL_ADDR,
2722 		.page = page,
2723 		.encrypted_page = NULL,
2724 		.submitted = false,
2725 		.compr_blocks = compr_blocks,
2726 		.need_lock = LOCK_RETRY,
2727 		.io_type = io_type,
2728 		.io_wbc = wbc,
2729 		.bio = bio,
2730 		.last_block = last_block,
2731 	};
2732 
2733 	trace_f2fs_writepage(page, DATA);
2734 
2735 	/* we should bypass data pages to proceed the kworkder jobs */
2736 	if (unlikely(f2fs_cp_error(sbi))) {
2737 		mapping_set_error(page->mapping, -EIO);
2738 		/*
2739 		 * don't drop any dirty dentry pages for keeping lastest
2740 		 * directory structure.
2741 		 */
2742 		if (S_ISDIR(inode->i_mode))
2743 			goto redirty_out;
2744 		goto out;
2745 	}
2746 
2747 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2748 		goto redirty_out;
2749 
2750 	if (page->index < end_index ||
2751 			f2fs_verity_in_progress(inode) ||
2752 			compr_blocks)
2753 		goto write;
2754 
2755 	/*
2756 	 * If the offset is out-of-range of file size,
2757 	 * this page does not have to be written to disk.
2758 	 */
2759 	offset = i_size & (PAGE_SIZE - 1);
2760 	if ((page->index >= end_index + 1) || !offset)
2761 		goto out;
2762 
2763 	zero_user_segment(page, offset, PAGE_SIZE);
2764 write:
2765 	if (f2fs_is_drop_cache(inode))
2766 		goto out;
2767 	/* we should not write 0'th page having journal header */
2768 	if (f2fs_is_volatile_file(inode) && (!page->index ||
2769 			(!wbc->for_reclaim &&
2770 			f2fs_available_free_memory(sbi, BASE_CHECK))))
2771 		goto redirty_out;
2772 
2773 	/* Dentry/quota blocks are controlled by checkpoint */
2774 	if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2775 		fio.need_lock = LOCK_DONE;
2776 		err = f2fs_do_write_data_page(&fio);
2777 		goto done;
2778 	}
2779 
2780 	if (!wbc->for_reclaim)
2781 		need_balance_fs = true;
2782 	else if (has_not_enough_free_secs(sbi, 0, 0))
2783 		goto redirty_out;
2784 	else
2785 		set_inode_flag(inode, FI_HOT_DATA);
2786 
2787 	err = -EAGAIN;
2788 	if (f2fs_has_inline_data(inode)) {
2789 		err = f2fs_write_inline_data(inode, page);
2790 		if (!err)
2791 			goto out;
2792 	}
2793 
2794 	if (err == -EAGAIN) {
2795 		err = f2fs_do_write_data_page(&fio);
2796 		if (err == -EAGAIN) {
2797 			fio.need_lock = LOCK_REQ;
2798 			err = f2fs_do_write_data_page(&fio);
2799 		}
2800 	}
2801 
2802 	if (err) {
2803 		file_set_keep_isize(inode);
2804 	} else {
2805 		spin_lock(&F2FS_I(inode)->i_size_lock);
2806 		if (F2FS_I(inode)->last_disk_size < psize)
2807 			F2FS_I(inode)->last_disk_size = psize;
2808 		spin_unlock(&F2FS_I(inode)->i_size_lock);
2809 	}
2810 
2811 done:
2812 	if (err && err != -ENOENT)
2813 		goto redirty_out;
2814 
2815 out:
2816 	inode_dec_dirty_pages(inode);
2817 	if (err) {
2818 		ClearPageUptodate(page);
2819 		clear_cold_data(page);
2820 	}
2821 
2822 	if (wbc->for_reclaim) {
2823 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2824 		clear_inode_flag(inode, FI_HOT_DATA);
2825 		f2fs_remove_dirty_inode(inode);
2826 		submitted = NULL;
2827 	}
2828 	unlock_page(page);
2829 	if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2830 					!F2FS_I(inode)->cp_task)
2831 		f2fs_balance_fs(sbi, need_balance_fs);
2832 
2833 	if (unlikely(f2fs_cp_error(sbi))) {
2834 		f2fs_submit_merged_write(sbi, DATA);
2835 		f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2836 		submitted = NULL;
2837 	}
2838 
2839 	if (submitted)
2840 		*submitted = fio.submitted ? 1 : 0;
2841 
2842 	return 0;
2843 
2844 redirty_out:
2845 	redirty_page_for_writepage(wbc, page);
2846 	/*
2847 	 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2848 	 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2849 	 * file_write_and_wait_range() will see EIO error, which is critical
2850 	 * to return value of fsync() followed by atomic_write failure to user.
2851 	 */
2852 	if (!err || wbc->for_reclaim)
2853 		return AOP_WRITEPAGE_ACTIVATE;
2854 	unlock_page(page);
2855 	return err;
2856 }
2857 
2858 static int f2fs_write_data_page(struct page *page,
2859 					struct writeback_control *wbc)
2860 {
2861 #ifdef CONFIG_F2FS_FS_COMPRESSION
2862 	struct inode *inode = page->mapping->host;
2863 
2864 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2865 		goto out;
2866 
2867 	if (f2fs_compressed_file(inode)) {
2868 		if (f2fs_is_compressed_cluster(inode, page->index)) {
2869 			redirty_page_for_writepage(wbc, page);
2870 			return AOP_WRITEPAGE_ACTIVATE;
2871 		}
2872 	}
2873 out:
2874 #endif
2875 
2876 	return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2877 						wbc, FS_DATA_IO, 0);
2878 }
2879 
2880 /*
2881  * This function was copied from write_cche_pages from mm/page-writeback.c.
2882  * The major change is making write step of cold data page separately from
2883  * warm/hot data page.
2884  */
2885 static int f2fs_write_cache_pages(struct address_space *mapping,
2886 					struct writeback_control *wbc,
2887 					enum iostat_type io_type)
2888 {
2889 	int ret = 0;
2890 	int done = 0, retry = 0;
2891 	struct pagevec pvec;
2892 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2893 	struct bio *bio = NULL;
2894 	sector_t last_block;
2895 #ifdef CONFIG_F2FS_FS_COMPRESSION
2896 	struct inode *inode = mapping->host;
2897 	struct compress_ctx cc = {
2898 		.inode = inode,
2899 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2900 		.cluster_size = F2FS_I(inode)->i_cluster_size,
2901 		.cluster_idx = NULL_CLUSTER,
2902 		.rpages = NULL,
2903 		.nr_rpages = 0,
2904 		.cpages = NULL,
2905 		.rbuf = NULL,
2906 		.cbuf = NULL,
2907 		.rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2908 		.private = NULL,
2909 	};
2910 #endif
2911 	int nr_pages;
2912 	pgoff_t index;
2913 	pgoff_t end;		/* Inclusive */
2914 	pgoff_t done_index;
2915 	int range_whole = 0;
2916 	xa_mark_t tag;
2917 	int nwritten = 0;
2918 	int submitted = 0;
2919 	int i;
2920 
2921 	pagevec_init(&pvec);
2922 
2923 	if (get_dirty_pages(mapping->host) <=
2924 				SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2925 		set_inode_flag(mapping->host, FI_HOT_DATA);
2926 	else
2927 		clear_inode_flag(mapping->host, FI_HOT_DATA);
2928 
2929 	if (wbc->range_cyclic) {
2930 		index = mapping->writeback_index; /* prev offset */
2931 		end = -1;
2932 	} else {
2933 		index = wbc->range_start >> PAGE_SHIFT;
2934 		end = wbc->range_end >> PAGE_SHIFT;
2935 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2936 			range_whole = 1;
2937 	}
2938 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2939 		tag = PAGECACHE_TAG_TOWRITE;
2940 	else
2941 		tag = PAGECACHE_TAG_DIRTY;
2942 retry:
2943 	retry = 0;
2944 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2945 		tag_pages_for_writeback(mapping, index, end);
2946 	done_index = index;
2947 	while (!done && !retry && (index <= end)) {
2948 		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2949 				tag);
2950 		if (nr_pages == 0)
2951 			break;
2952 
2953 		for (i = 0; i < nr_pages; i++) {
2954 			struct page *page = pvec.pages[i];
2955 			bool need_readd;
2956 readd:
2957 			need_readd = false;
2958 #ifdef CONFIG_F2FS_FS_COMPRESSION
2959 			if (f2fs_compressed_file(inode)) {
2960 				ret = f2fs_init_compress_ctx(&cc);
2961 				if (ret) {
2962 					done = 1;
2963 					break;
2964 				}
2965 
2966 				if (!f2fs_cluster_can_merge_page(&cc,
2967 								page->index)) {
2968 					ret = f2fs_write_multi_pages(&cc,
2969 						&submitted, wbc, io_type);
2970 					if (!ret)
2971 						need_readd = true;
2972 					goto result;
2973 				}
2974 
2975 				if (unlikely(f2fs_cp_error(sbi)))
2976 					goto lock_page;
2977 
2978 				if (f2fs_cluster_is_empty(&cc)) {
2979 					void *fsdata = NULL;
2980 					struct page *pagep;
2981 					int ret2;
2982 
2983 					ret2 = f2fs_prepare_compress_overwrite(
2984 							inode, &pagep,
2985 							page->index, &fsdata);
2986 					if (ret2 < 0) {
2987 						ret = ret2;
2988 						done = 1;
2989 						break;
2990 					} else if (ret2 &&
2991 						!f2fs_compress_write_end(inode,
2992 								fsdata, page->index,
2993 								1)) {
2994 						retry = 1;
2995 						break;
2996 					}
2997 				} else {
2998 					goto lock_page;
2999 				}
3000 			}
3001 #endif
3002 			/* give a priority to WB_SYNC threads */
3003 			if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3004 					wbc->sync_mode == WB_SYNC_NONE) {
3005 				done = 1;
3006 				break;
3007 			}
3008 #ifdef CONFIG_F2FS_FS_COMPRESSION
3009 lock_page:
3010 #endif
3011 			done_index = page->index;
3012 retry_write:
3013 			lock_page(page);
3014 
3015 			if (unlikely(page->mapping != mapping)) {
3016 continue_unlock:
3017 				unlock_page(page);
3018 				continue;
3019 			}
3020 
3021 			if (!PageDirty(page)) {
3022 				/* someone wrote it for us */
3023 				goto continue_unlock;
3024 			}
3025 
3026 			if (PageWriteback(page)) {
3027 				if (wbc->sync_mode != WB_SYNC_NONE)
3028 					f2fs_wait_on_page_writeback(page,
3029 							DATA, true, true);
3030 				else
3031 					goto continue_unlock;
3032 			}
3033 
3034 			if (!clear_page_dirty_for_io(page))
3035 				goto continue_unlock;
3036 
3037 #ifdef CONFIG_F2FS_FS_COMPRESSION
3038 			if (f2fs_compressed_file(inode)) {
3039 				get_page(page);
3040 				f2fs_compress_ctx_add_page(&cc, page);
3041 				continue;
3042 			}
3043 #endif
3044 			ret = f2fs_write_single_data_page(page, &submitted,
3045 					&bio, &last_block, wbc, io_type, 0);
3046 			if (ret == AOP_WRITEPAGE_ACTIVATE)
3047 				unlock_page(page);
3048 #ifdef CONFIG_F2FS_FS_COMPRESSION
3049 result:
3050 #endif
3051 			nwritten += submitted;
3052 			wbc->nr_to_write -= submitted;
3053 
3054 			if (unlikely(ret)) {
3055 				/*
3056 				 * keep nr_to_write, since vfs uses this to
3057 				 * get # of written pages.
3058 				 */
3059 				if (ret == AOP_WRITEPAGE_ACTIVATE) {
3060 					ret = 0;
3061 					goto next;
3062 				} else if (ret == -EAGAIN) {
3063 					ret = 0;
3064 					if (wbc->sync_mode == WB_SYNC_ALL) {
3065 						cond_resched();
3066 						congestion_wait(BLK_RW_ASYNC,
3067 							DEFAULT_IO_TIMEOUT);
3068 						goto retry_write;
3069 					}
3070 					goto next;
3071 				}
3072 				done_index = page->index + 1;
3073 				done = 1;
3074 				break;
3075 			}
3076 
3077 			if (wbc->nr_to_write <= 0 &&
3078 					wbc->sync_mode == WB_SYNC_NONE) {
3079 				done = 1;
3080 				break;
3081 			}
3082 next:
3083 			if (need_readd)
3084 				goto readd;
3085 		}
3086 		pagevec_release(&pvec);
3087 		cond_resched();
3088 	}
3089 #ifdef CONFIG_F2FS_FS_COMPRESSION
3090 	/* flush remained pages in compress cluster */
3091 	if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3092 		ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3093 		nwritten += submitted;
3094 		wbc->nr_to_write -= submitted;
3095 		if (ret) {
3096 			done = 1;
3097 			retry = 0;
3098 		}
3099 	}
3100 #endif
3101 	if (retry) {
3102 		index = 0;
3103 		end = -1;
3104 		goto retry;
3105 	}
3106 	if (wbc->range_cyclic && !done)
3107 		done_index = 0;
3108 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3109 		mapping->writeback_index = done_index;
3110 
3111 	if (nwritten)
3112 		f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3113 								NULL, 0, DATA);
3114 	/* submit cached bio of IPU write */
3115 	if (bio)
3116 		f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3117 
3118 	return ret;
3119 }
3120 
3121 static inline bool __should_serialize_io(struct inode *inode,
3122 					struct writeback_control *wbc)
3123 {
3124 	/* to avoid deadlock in path of data flush */
3125 	if (F2FS_I(inode)->cp_task)
3126 		return false;
3127 
3128 	if (!S_ISREG(inode->i_mode))
3129 		return false;
3130 	if (IS_NOQUOTA(inode))
3131 		return false;
3132 
3133 	if (f2fs_compressed_file(inode))
3134 		return true;
3135 	if (wbc->sync_mode != WB_SYNC_ALL)
3136 		return true;
3137 	if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3138 		return true;
3139 	return false;
3140 }
3141 
3142 static int __f2fs_write_data_pages(struct address_space *mapping,
3143 						struct writeback_control *wbc,
3144 						enum iostat_type io_type)
3145 {
3146 	struct inode *inode = mapping->host;
3147 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3148 	struct blk_plug plug;
3149 	int ret;
3150 	bool locked = false;
3151 
3152 	/* deal with chardevs and other special file */
3153 	if (!mapping->a_ops->writepage)
3154 		return 0;
3155 
3156 	/* skip writing if there is no dirty page in this inode */
3157 	if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3158 		return 0;
3159 
3160 	/* during POR, we don't need to trigger writepage at all. */
3161 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3162 		goto skip_write;
3163 
3164 	if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3165 			wbc->sync_mode == WB_SYNC_NONE &&
3166 			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3167 			f2fs_available_free_memory(sbi, DIRTY_DENTS))
3168 		goto skip_write;
3169 
3170 	/* skip writing during file defragment */
3171 	if (is_inode_flag_set(inode, FI_DO_DEFRAG))
3172 		goto skip_write;
3173 
3174 	trace_f2fs_writepages(mapping->host, wbc, DATA);
3175 
3176 	/* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3177 	if (wbc->sync_mode == WB_SYNC_ALL)
3178 		atomic_inc(&sbi->wb_sync_req[DATA]);
3179 	else if (atomic_read(&sbi->wb_sync_req[DATA]))
3180 		goto skip_write;
3181 
3182 	if (__should_serialize_io(inode, wbc)) {
3183 		mutex_lock(&sbi->writepages);
3184 		locked = true;
3185 	}
3186 
3187 	blk_start_plug(&plug);
3188 	ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3189 	blk_finish_plug(&plug);
3190 
3191 	if (locked)
3192 		mutex_unlock(&sbi->writepages);
3193 
3194 	if (wbc->sync_mode == WB_SYNC_ALL)
3195 		atomic_dec(&sbi->wb_sync_req[DATA]);
3196 	/*
3197 	 * if some pages were truncated, we cannot guarantee its mapping->host
3198 	 * to detect pending bios.
3199 	 */
3200 
3201 	f2fs_remove_dirty_inode(inode);
3202 	return ret;
3203 
3204 skip_write:
3205 	wbc->pages_skipped += get_dirty_pages(inode);
3206 	trace_f2fs_writepages(mapping->host, wbc, DATA);
3207 	return 0;
3208 }
3209 
3210 static int f2fs_write_data_pages(struct address_space *mapping,
3211 			    struct writeback_control *wbc)
3212 {
3213 	struct inode *inode = mapping->host;
3214 
3215 	return __f2fs_write_data_pages(mapping, wbc,
3216 			F2FS_I(inode)->cp_task == current ?
3217 			FS_CP_DATA_IO : FS_DATA_IO);
3218 }
3219 
3220 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
3221 {
3222 	struct inode *inode = mapping->host;
3223 	loff_t i_size = i_size_read(inode);
3224 
3225 	if (IS_NOQUOTA(inode))
3226 		return;
3227 
3228 	/* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3229 	if (to > i_size && !f2fs_verity_in_progress(inode)) {
3230 		down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3231 		down_write(&F2FS_I(inode)->i_mmap_sem);
3232 
3233 		truncate_pagecache(inode, i_size);
3234 		f2fs_truncate_blocks(inode, i_size, true);
3235 
3236 		up_write(&F2FS_I(inode)->i_mmap_sem);
3237 		up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3238 	}
3239 }
3240 
3241 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3242 			struct page *page, loff_t pos, unsigned len,
3243 			block_t *blk_addr, bool *node_changed)
3244 {
3245 	struct inode *inode = page->mapping->host;
3246 	pgoff_t index = page->index;
3247 	struct dnode_of_data dn;
3248 	struct page *ipage;
3249 	bool locked = false;
3250 	struct extent_info ei = {0,0,0};
3251 	int err = 0;
3252 	int flag;
3253 
3254 	/*
3255 	 * we already allocated all the blocks, so we don't need to get
3256 	 * the block addresses when there is no need to fill the page.
3257 	 */
3258 	if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
3259 	    !is_inode_flag_set(inode, FI_NO_PREALLOC) &&
3260 	    !f2fs_verity_in_progress(inode))
3261 		return 0;
3262 
3263 	/* f2fs_lock_op avoids race between write CP and convert_inline_page */
3264 	if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3265 		flag = F2FS_GET_BLOCK_DEFAULT;
3266 	else
3267 		flag = F2FS_GET_BLOCK_PRE_AIO;
3268 
3269 	if (f2fs_has_inline_data(inode) ||
3270 			(pos & PAGE_MASK) >= i_size_read(inode)) {
3271 		__do_map_lock(sbi, flag, true);
3272 		locked = true;
3273 	}
3274 
3275 restart:
3276 	/* check inline_data */
3277 	ipage = f2fs_get_node_page(sbi, inode->i_ino);
3278 	if (IS_ERR(ipage)) {
3279 		err = PTR_ERR(ipage);
3280 		goto unlock_out;
3281 	}
3282 
3283 	set_new_dnode(&dn, inode, ipage, ipage, 0);
3284 
3285 	if (f2fs_has_inline_data(inode)) {
3286 		if (pos + len <= MAX_INLINE_DATA(inode)) {
3287 			f2fs_do_read_inline_data(page, ipage);
3288 			set_inode_flag(inode, FI_DATA_EXIST);
3289 			if (inode->i_nlink)
3290 				set_inline_node(ipage);
3291 		} else {
3292 			err = f2fs_convert_inline_page(&dn, page);
3293 			if (err)
3294 				goto out;
3295 			if (dn.data_blkaddr == NULL_ADDR)
3296 				err = f2fs_get_block(&dn, index);
3297 		}
3298 	} else if (locked) {
3299 		err = f2fs_get_block(&dn, index);
3300 	} else {
3301 		if (f2fs_lookup_extent_cache(inode, index, &ei)) {
3302 			dn.data_blkaddr = ei.blk + index - ei.fofs;
3303 		} else {
3304 			/* hole case */
3305 			err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3306 			if (err || dn.data_blkaddr == NULL_ADDR) {
3307 				f2fs_put_dnode(&dn);
3308 				__do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3309 								true);
3310 				WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3311 				locked = true;
3312 				goto restart;
3313 			}
3314 		}
3315 	}
3316 
3317 	/* convert_inline_page can make node_changed */
3318 	*blk_addr = dn.data_blkaddr;
3319 	*node_changed = dn.node_changed;
3320 out:
3321 	f2fs_put_dnode(&dn);
3322 unlock_out:
3323 	if (locked)
3324 		__do_map_lock(sbi, flag, false);
3325 	return err;
3326 }
3327 
3328 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3329 		loff_t pos, unsigned len, unsigned flags,
3330 		struct page **pagep, void **fsdata)
3331 {
3332 	struct inode *inode = mapping->host;
3333 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3334 	struct page *page = NULL;
3335 	pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3336 	bool need_balance = false, drop_atomic = false;
3337 	block_t blkaddr = NULL_ADDR;
3338 	int err = 0;
3339 
3340 	trace_f2fs_write_begin(inode, pos, len, flags);
3341 
3342 	if (!f2fs_is_checkpoint_ready(sbi)) {
3343 		err = -ENOSPC;
3344 		goto fail;
3345 	}
3346 
3347 	if ((f2fs_is_atomic_file(inode) &&
3348 			!f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
3349 			is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
3350 		err = -ENOMEM;
3351 		drop_atomic = true;
3352 		goto fail;
3353 	}
3354 
3355 	/*
3356 	 * We should check this at this moment to avoid deadlock on inode page
3357 	 * and #0 page. The locking rule for inline_data conversion should be:
3358 	 * lock_page(page #0) -> lock_page(inode_page)
3359 	 */
3360 	if (index != 0) {
3361 		err = f2fs_convert_inline_inode(inode);
3362 		if (err)
3363 			goto fail;
3364 	}
3365 
3366 #ifdef CONFIG_F2FS_FS_COMPRESSION
3367 	if (f2fs_compressed_file(inode)) {
3368 		int ret;
3369 
3370 		*fsdata = NULL;
3371 
3372 		ret = f2fs_prepare_compress_overwrite(inode, pagep,
3373 							index, fsdata);
3374 		if (ret < 0) {
3375 			err = ret;
3376 			goto fail;
3377 		} else if (ret) {
3378 			return 0;
3379 		}
3380 	}
3381 #endif
3382 
3383 repeat:
3384 	/*
3385 	 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3386 	 * wait_for_stable_page. Will wait that below with our IO control.
3387 	 */
3388 	page = f2fs_pagecache_get_page(mapping, index,
3389 				FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3390 	if (!page) {
3391 		err = -ENOMEM;
3392 		goto fail;
3393 	}
3394 
3395 	/* TODO: cluster can be compressed due to race with .writepage */
3396 
3397 	*pagep = page;
3398 
3399 	err = prepare_write_begin(sbi, page, pos, len,
3400 					&blkaddr, &need_balance);
3401 	if (err)
3402 		goto fail;
3403 
3404 	if (need_balance && !IS_NOQUOTA(inode) &&
3405 			has_not_enough_free_secs(sbi, 0, 0)) {
3406 		unlock_page(page);
3407 		f2fs_balance_fs(sbi, true);
3408 		lock_page(page);
3409 		if (page->mapping != mapping) {
3410 			/* The page got truncated from under us */
3411 			f2fs_put_page(page, 1);
3412 			goto repeat;
3413 		}
3414 	}
3415 
3416 	f2fs_wait_on_page_writeback(page, DATA, false, true);
3417 
3418 	if (len == PAGE_SIZE || PageUptodate(page))
3419 		return 0;
3420 
3421 	if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3422 	    !f2fs_verity_in_progress(inode)) {
3423 		zero_user_segment(page, len, PAGE_SIZE);
3424 		return 0;
3425 	}
3426 
3427 	if (blkaddr == NEW_ADDR) {
3428 		zero_user_segment(page, 0, PAGE_SIZE);
3429 		SetPageUptodate(page);
3430 	} else {
3431 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3432 				DATA_GENERIC_ENHANCE_READ)) {
3433 			err = -EFSCORRUPTED;
3434 			goto fail;
3435 		}
3436 		err = f2fs_submit_page_read(inode, page, blkaddr, true);
3437 		if (err)
3438 			goto fail;
3439 
3440 		lock_page(page);
3441 		if (unlikely(page->mapping != mapping)) {
3442 			f2fs_put_page(page, 1);
3443 			goto repeat;
3444 		}
3445 		if (unlikely(!PageUptodate(page))) {
3446 			err = -EIO;
3447 			goto fail;
3448 		}
3449 	}
3450 	return 0;
3451 
3452 fail:
3453 	f2fs_put_page(page, 1);
3454 	f2fs_write_failed(mapping, pos + len);
3455 	if (drop_atomic)
3456 		f2fs_drop_inmem_pages_all(sbi, false);
3457 	return err;
3458 }
3459 
3460 static int f2fs_write_end(struct file *file,
3461 			struct address_space *mapping,
3462 			loff_t pos, unsigned len, unsigned copied,
3463 			struct page *page, void *fsdata)
3464 {
3465 	struct inode *inode = page->mapping->host;
3466 
3467 	trace_f2fs_write_end(inode, pos, len, copied);
3468 
3469 	/*
3470 	 * This should be come from len == PAGE_SIZE, and we expect copied
3471 	 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3472 	 * let generic_perform_write() try to copy data again through copied=0.
3473 	 */
3474 	if (!PageUptodate(page)) {
3475 		if (unlikely(copied != len))
3476 			copied = 0;
3477 		else
3478 			SetPageUptodate(page);
3479 	}
3480 
3481 #ifdef CONFIG_F2FS_FS_COMPRESSION
3482 	/* overwrite compressed file */
3483 	if (f2fs_compressed_file(inode) && fsdata) {
3484 		f2fs_compress_write_end(inode, fsdata, page->index, copied);
3485 		f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3486 		return copied;
3487 	}
3488 #endif
3489 
3490 	if (!copied)
3491 		goto unlock_out;
3492 
3493 	set_page_dirty(page);
3494 
3495 	if (pos + copied > i_size_read(inode) &&
3496 	    !f2fs_verity_in_progress(inode))
3497 		f2fs_i_size_write(inode, pos + copied);
3498 unlock_out:
3499 	f2fs_put_page(page, 1);
3500 	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3501 	return copied;
3502 }
3503 
3504 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
3505 			   loff_t offset)
3506 {
3507 	unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
3508 	unsigned blkbits = i_blkbits;
3509 	unsigned blocksize_mask = (1 << blkbits) - 1;
3510 	unsigned long align = offset | iov_iter_alignment(iter);
3511 	struct block_device *bdev = inode->i_sb->s_bdev;
3512 
3513 	if (align & blocksize_mask) {
3514 		if (bdev)
3515 			blkbits = blksize_bits(bdev_logical_block_size(bdev));
3516 		blocksize_mask = (1 << blkbits) - 1;
3517 		if (align & blocksize_mask)
3518 			return -EINVAL;
3519 		return 1;
3520 	}
3521 	return 0;
3522 }
3523 
3524 static void f2fs_dio_end_io(struct bio *bio)
3525 {
3526 	struct f2fs_private_dio *dio = bio->bi_private;
3527 
3528 	dec_page_count(F2FS_I_SB(dio->inode),
3529 			dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3530 
3531 	bio->bi_private = dio->orig_private;
3532 	bio->bi_end_io = dio->orig_end_io;
3533 
3534 	kvfree(dio);
3535 
3536 	bio_endio(bio);
3537 }
3538 
3539 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
3540 							loff_t file_offset)
3541 {
3542 	struct f2fs_private_dio *dio;
3543 	bool write = (bio_op(bio) == REQ_OP_WRITE);
3544 
3545 	dio = f2fs_kzalloc(F2FS_I_SB(inode),
3546 			sizeof(struct f2fs_private_dio), GFP_NOFS);
3547 	if (!dio)
3548 		goto out;
3549 
3550 	dio->inode = inode;
3551 	dio->orig_end_io = bio->bi_end_io;
3552 	dio->orig_private = bio->bi_private;
3553 	dio->write = write;
3554 
3555 	bio->bi_end_io = f2fs_dio_end_io;
3556 	bio->bi_private = dio;
3557 
3558 	inc_page_count(F2FS_I_SB(inode),
3559 			write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
3560 
3561 	submit_bio(bio);
3562 	return;
3563 out:
3564 	bio->bi_status = BLK_STS_IOERR;
3565 	bio_endio(bio);
3566 }
3567 
3568 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3569 {
3570 	struct address_space *mapping = iocb->ki_filp->f_mapping;
3571 	struct inode *inode = mapping->host;
3572 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3573 	struct f2fs_inode_info *fi = F2FS_I(inode);
3574 	size_t count = iov_iter_count(iter);
3575 	loff_t offset = iocb->ki_pos;
3576 	int rw = iov_iter_rw(iter);
3577 	int err;
3578 	enum rw_hint hint = iocb->ki_hint;
3579 	int whint_mode = F2FS_OPTION(sbi).whint_mode;
3580 	bool do_opu;
3581 
3582 	err = check_direct_IO(inode, iter, offset);
3583 	if (err)
3584 		return err < 0 ? err : 0;
3585 
3586 	if (f2fs_force_buffered_io(inode, iocb, iter))
3587 		return 0;
3588 
3589 	do_opu = allow_outplace_dio(inode, iocb, iter);
3590 
3591 	trace_f2fs_direct_IO_enter(inode, offset, count, rw);
3592 
3593 	if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
3594 		iocb->ki_hint = WRITE_LIFE_NOT_SET;
3595 
3596 	if (iocb->ki_flags & IOCB_NOWAIT) {
3597 		if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
3598 			iocb->ki_hint = hint;
3599 			err = -EAGAIN;
3600 			goto out;
3601 		}
3602 		if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
3603 			up_read(&fi->i_gc_rwsem[rw]);
3604 			iocb->ki_hint = hint;
3605 			err = -EAGAIN;
3606 			goto out;
3607 		}
3608 	} else {
3609 		down_read(&fi->i_gc_rwsem[rw]);
3610 		if (do_opu)
3611 			down_read(&fi->i_gc_rwsem[READ]);
3612 	}
3613 
3614 	err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3615 			iter, rw == WRITE ? get_data_block_dio_write :
3616 			get_data_block_dio, NULL, f2fs_dio_submit_bio,
3617 			rw == WRITE ? DIO_LOCKING | DIO_SKIP_HOLES :
3618 			DIO_SKIP_HOLES);
3619 
3620 	if (do_opu)
3621 		up_read(&fi->i_gc_rwsem[READ]);
3622 
3623 	up_read(&fi->i_gc_rwsem[rw]);
3624 
3625 	if (rw == WRITE) {
3626 		if (whint_mode == WHINT_MODE_OFF)
3627 			iocb->ki_hint = hint;
3628 		if (err > 0) {
3629 			f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
3630 									err);
3631 			if (!do_opu)
3632 				set_inode_flag(inode, FI_UPDATE_WRITE);
3633 		} else if (err < 0) {
3634 			f2fs_write_failed(mapping, offset + count);
3635 		}
3636 	} else {
3637 		if (err > 0)
3638 			f2fs_update_iostat(sbi, APP_DIRECT_READ_IO, err);
3639 	}
3640 
3641 out:
3642 	trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
3643 
3644 	return err;
3645 }
3646 
3647 void f2fs_invalidate_page(struct page *page, unsigned int offset,
3648 							unsigned int length)
3649 {
3650 	struct inode *inode = page->mapping->host;
3651 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3652 
3653 	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3654 		(offset % PAGE_SIZE || length != PAGE_SIZE))
3655 		return;
3656 
3657 	if (PageDirty(page)) {
3658 		if (inode->i_ino == F2FS_META_INO(sbi)) {
3659 			dec_page_count(sbi, F2FS_DIRTY_META);
3660 		} else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3661 			dec_page_count(sbi, F2FS_DIRTY_NODES);
3662 		} else {
3663 			inode_dec_dirty_pages(inode);
3664 			f2fs_remove_dirty_inode(inode);
3665 		}
3666 	}
3667 
3668 	clear_cold_data(page);
3669 
3670 	if (IS_ATOMIC_WRITTEN_PAGE(page))
3671 		return f2fs_drop_inmem_page(inode, page);
3672 
3673 	f2fs_clear_page_private(page);
3674 }
3675 
3676 int f2fs_release_page(struct page *page, gfp_t wait)
3677 {
3678 	/* If this is dirty page, keep PagePrivate */
3679 	if (PageDirty(page))
3680 		return 0;
3681 
3682 	/* This is atomic written page, keep Private */
3683 	if (IS_ATOMIC_WRITTEN_PAGE(page))
3684 		return 0;
3685 
3686 	clear_cold_data(page);
3687 	f2fs_clear_page_private(page);
3688 	return 1;
3689 }
3690 
3691 static int f2fs_set_data_page_dirty(struct page *page)
3692 {
3693 	struct inode *inode = page_file_mapping(page)->host;
3694 
3695 	trace_f2fs_set_page_dirty(page, DATA);
3696 
3697 	if (!PageUptodate(page))
3698 		SetPageUptodate(page);
3699 	if (PageSwapCache(page))
3700 		return __set_page_dirty_nobuffers(page);
3701 
3702 	if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
3703 		if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
3704 			f2fs_register_inmem_page(inode, page);
3705 			return 1;
3706 		}
3707 		/*
3708 		 * Previously, this page has been registered, we just
3709 		 * return here.
3710 		 */
3711 		return 0;
3712 	}
3713 
3714 	if (!PageDirty(page)) {
3715 		__set_page_dirty_nobuffers(page);
3716 		f2fs_update_dirty_page(inode, page);
3717 		return 1;
3718 	}
3719 	return 0;
3720 }
3721 
3722 
3723 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3724 {
3725 #ifdef CONFIG_F2FS_FS_COMPRESSION
3726 	struct dnode_of_data dn;
3727 	sector_t start_idx, blknr = 0;
3728 	int ret;
3729 
3730 	start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3731 
3732 	set_new_dnode(&dn, inode, NULL, NULL, 0);
3733 	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3734 	if (ret)
3735 		return 0;
3736 
3737 	if (dn.data_blkaddr != COMPRESS_ADDR) {
3738 		dn.ofs_in_node += block - start_idx;
3739 		blknr = f2fs_data_blkaddr(&dn);
3740 		if (!__is_valid_data_blkaddr(blknr))
3741 			blknr = 0;
3742 	}
3743 
3744 	f2fs_put_dnode(&dn);
3745 
3746 	return blknr;
3747 #else
3748 	return -EOPNOTSUPP;
3749 #endif
3750 }
3751 
3752 
3753 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3754 {
3755 	struct inode *inode = mapping->host;
3756 
3757 	if (f2fs_has_inline_data(inode))
3758 		return 0;
3759 
3760 	/* make sure allocating whole blocks */
3761 	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3762 		filemap_write_and_wait(mapping);
3763 
3764 	if (f2fs_compressed_file(inode))
3765 		return f2fs_bmap_compress(inode, block);
3766 
3767 	return generic_block_bmap(mapping, block, get_data_block_bmap);
3768 }
3769 
3770 #ifdef CONFIG_MIGRATION
3771 #include <linux/migrate.h>
3772 
3773 int f2fs_migrate_page(struct address_space *mapping,
3774 		struct page *newpage, struct page *page, enum migrate_mode mode)
3775 {
3776 	int rc, extra_count;
3777 	struct f2fs_inode_info *fi = F2FS_I(mapping->host);
3778 	bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
3779 
3780 	BUG_ON(PageWriteback(page));
3781 
3782 	/* migrating an atomic written page is safe with the inmem_lock hold */
3783 	if (atomic_written) {
3784 		if (mode != MIGRATE_SYNC)
3785 			return -EBUSY;
3786 		if (!mutex_trylock(&fi->inmem_lock))
3787 			return -EAGAIN;
3788 	}
3789 
3790 	/* one extra reference was held for atomic_write page */
3791 	extra_count = atomic_written ? 1 : 0;
3792 	rc = migrate_page_move_mapping(mapping, newpage,
3793 				page, extra_count);
3794 	if (rc != MIGRATEPAGE_SUCCESS) {
3795 		if (atomic_written)
3796 			mutex_unlock(&fi->inmem_lock);
3797 		return rc;
3798 	}
3799 
3800 	if (atomic_written) {
3801 		struct inmem_pages *cur;
3802 		list_for_each_entry(cur, &fi->inmem_pages, list)
3803 			if (cur->page == page) {
3804 				cur->page = newpage;
3805 				break;
3806 			}
3807 		mutex_unlock(&fi->inmem_lock);
3808 		put_page(page);
3809 		get_page(newpage);
3810 	}
3811 
3812 	if (PagePrivate(page)) {
3813 		f2fs_set_page_private(newpage, page_private(page));
3814 		f2fs_clear_page_private(page);
3815 	}
3816 
3817 	if (mode != MIGRATE_SYNC_NO_COPY)
3818 		migrate_page_copy(newpage, page);
3819 	else
3820 		migrate_page_states(newpage, page);
3821 
3822 	return MIGRATEPAGE_SUCCESS;
3823 }
3824 #endif
3825 
3826 #ifdef CONFIG_SWAP
3827 /* Copied from generic_swapfile_activate() to check any holes */
3828 static int check_swap_activate(struct swap_info_struct *sis,
3829 				struct file *swap_file, sector_t *span)
3830 {
3831 	struct address_space *mapping = swap_file->f_mapping;
3832 	struct inode *inode = mapping->host;
3833 	unsigned blocks_per_page;
3834 	unsigned long page_no;
3835 	unsigned blkbits;
3836 	sector_t probe_block;
3837 	sector_t last_block;
3838 	sector_t lowest_block = -1;
3839 	sector_t highest_block = 0;
3840 	int nr_extents = 0;
3841 	int ret;
3842 
3843 	blkbits = inode->i_blkbits;
3844 	blocks_per_page = PAGE_SIZE >> blkbits;
3845 
3846 	/*
3847 	 * Map all the blocks into the extent list.  This code doesn't try
3848 	 * to be very smart.
3849 	 */
3850 	probe_block = 0;
3851 	page_no = 0;
3852 	last_block = i_size_read(inode) >> blkbits;
3853 	while ((probe_block + blocks_per_page) <= last_block &&
3854 			page_no < sis->max) {
3855 		unsigned block_in_page;
3856 		sector_t first_block;
3857 		sector_t block = 0;
3858 		int	 err = 0;
3859 
3860 		cond_resched();
3861 
3862 		block = probe_block;
3863 		err = bmap(inode, &block);
3864 		if (err || !block)
3865 			goto bad_bmap;
3866 		first_block = block;
3867 
3868 		/*
3869 		 * It must be PAGE_SIZE aligned on-disk
3870 		 */
3871 		if (first_block & (blocks_per_page - 1)) {
3872 			probe_block++;
3873 			goto reprobe;
3874 		}
3875 
3876 		for (block_in_page = 1; block_in_page < blocks_per_page;
3877 					block_in_page++) {
3878 
3879 			block = probe_block + block_in_page;
3880 			err = bmap(inode, &block);
3881 
3882 			if (err || !block)
3883 				goto bad_bmap;
3884 
3885 			if (block != first_block + block_in_page) {
3886 				/* Discontiguity */
3887 				probe_block++;
3888 				goto reprobe;
3889 			}
3890 		}
3891 
3892 		first_block >>= (PAGE_SHIFT - blkbits);
3893 		if (page_no) {	/* exclude the header page */
3894 			if (first_block < lowest_block)
3895 				lowest_block = first_block;
3896 			if (first_block > highest_block)
3897 				highest_block = first_block;
3898 		}
3899 
3900 		/*
3901 		 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3902 		 */
3903 		ret = add_swap_extent(sis, page_no, 1, first_block);
3904 		if (ret < 0)
3905 			goto out;
3906 		nr_extents += ret;
3907 		page_no++;
3908 		probe_block += blocks_per_page;
3909 reprobe:
3910 		continue;
3911 	}
3912 	ret = nr_extents;
3913 	*span = 1 + highest_block - lowest_block;
3914 	if (page_no == 0)
3915 		page_no = 1;	/* force Empty message */
3916 	sis->max = page_no;
3917 	sis->pages = page_no - 1;
3918 	sis->highest_bit = page_no - 1;
3919 out:
3920 	return ret;
3921 bad_bmap:
3922 	pr_err("swapon: swapfile has holes\n");
3923 	return -EINVAL;
3924 }
3925 
3926 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3927 				sector_t *span)
3928 {
3929 	struct inode *inode = file_inode(file);
3930 	int ret;
3931 
3932 	if (!S_ISREG(inode->i_mode))
3933 		return -EINVAL;
3934 
3935 	if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3936 		return -EROFS;
3937 
3938 	ret = f2fs_convert_inline_inode(inode);
3939 	if (ret)
3940 		return ret;
3941 
3942 	if (f2fs_disable_compressed_file(inode))
3943 		return -EINVAL;
3944 
3945 	ret = check_swap_activate(sis, file, span);
3946 	if (ret < 0)
3947 		return ret;
3948 
3949 	set_inode_flag(inode, FI_PIN_FILE);
3950 	f2fs_precache_extents(inode);
3951 	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3952 	return ret;
3953 }
3954 
3955 static void f2fs_swap_deactivate(struct file *file)
3956 {
3957 	struct inode *inode = file_inode(file);
3958 
3959 	clear_inode_flag(inode, FI_PIN_FILE);
3960 }
3961 #else
3962 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3963 				sector_t *span)
3964 {
3965 	return -EOPNOTSUPP;
3966 }
3967 
3968 static void f2fs_swap_deactivate(struct file *file)
3969 {
3970 }
3971 #endif
3972 
3973 const struct address_space_operations f2fs_dblock_aops = {
3974 	.readpage	= f2fs_read_data_page,
3975 	.readahead	= f2fs_readahead,
3976 	.writepage	= f2fs_write_data_page,
3977 	.writepages	= f2fs_write_data_pages,
3978 	.write_begin	= f2fs_write_begin,
3979 	.write_end	= f2fs_write_end,
3980 	.set_page_dirty	= f2fs_set_data_page_dirty,
3981 	.invalidatepage	= f2fs_invalidate_page,
3982 	.releasepage	= f2fs_release_page,
3983 	.direct_IO	= f2fs_direct_IO,
3984 	.bmap		= f2fs_bmap,
3985 	.swap_activate  = f2fs_swap_activate,
3986 	.swap_deactivate = f2fs_swap_deactivate,
3987 #ifdef CONFIG_MIGRATION
3988 	.migratepage    = f2fs_migrate_page,
3989 #endif
3990 };
3991 
3992 void f2fs_clear_page_cache_dirty_tag(struct page *page)
3993 {
3994 	struct address_space *mapping = page_mapping(page);
3995 	unsigned long flags;
3996 
3997 	xa_lock_irqsave(&mapping->i_pages, flags);
3998 	__xa_clear_mark(&mapping->i_pages, page_index(page),
3999 						PAGECACHE_TAG_DIRTY);
4000 	xa_unlock_irqrestore(&mapping->i_pages, flags);
4001 }
4002 
4003 int __init f2fs_init_post_read_processing(void)
4004 {
4005 	bio_post_read_ctx_cache =
4006 		kmem_cache_create("f2fs_bio_post_read_ctx",
4007 				  sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4008 	if (!bio_post_read_ctx_cache)
4009 		goto fail;
4010 	bio_post_read_ctx_pool =
4011 		mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4012 					 bio_post_read_ctx_cache);
4013 	if (!bio_post_read_ctx_pool)
4014 		goto fail_free_cache;
4015 	return 0;
4016 
4017 fail_free_cache:
4018 	kmem_cache_destroy(bio_post_read_ctx_cache);
4019 fail:
4020 	return -ENOMEM;
4021 }
4022 
4023 void f2fs_destroy_post_read_processing(void)
4024 {
4025 	mempool_destroy(bio_post_read_ctx_pool);
4026 	kmem_cache_destroy(bio_post_read_ctx_cache);
4027 }
4028 
4029 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4030 {
4031 	if (!f2fs_sb_has_encrypt(sbi) &&
4032 		!f2fs_sb_has_verity(sbi) &&
4033 		!f2fs_sb_has_compression(sbi))
4034 		return 0;
4035 
4036 	sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4037 						 WQ_UNBOUND | WQ_HIGHPRI,
4038 						 num_online_cpus());
4039 	if (!sbi->post_read_wq)
4040 		return -ENOMEM;
4041 	return 0;
4042 }
4043 
4044 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4045 {
4046 	if (sbi->post_read_wq)
4047 		destroy_workqueue(sbi->post_read_wq);
4048 }
4049 
4050 int __init f2fs_init_bio_entry_cache(void)
4051 {
4052 	bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4053 			sizeof(struct bio_entry));
4054 	if (!bio_entry_slab)
4055 		return -ENOMEM;
4056 	return 0;
4057 }
4058 
4059 void f2fs_destroy_bio_entry_cache(void)
4060 {
4061 	kmem_cache_destroy(bio_entry_slab);
4062 }
4063