xref: /openbmc/linux/fs/f2fs/data.c (revision 54618888)
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/sched/mm.h>
12 #include <linux/mpage.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include <linux/blkdev.h>
16 #include <linux/bio.h>
17 #include <linux/blk-crypto.h>
18 #include <linux/swap.h>
19 #include <linux/prefetch.h>
20 #include <linux/uio.h>
21 #include <linux/sched/signal.h>
22 #include <linux/fiemap.h>
23 #include <linux/iomap.h>
24 
25 #include "f2fs.h"
26 #include "node.h"
27 #include "segment.h"
28 #include "iostat.h"
29 #include <trace/events/f2fs.h>
30 
31 #define NUM_PREALLOC_POST_READ_CTXS	128
32 
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 	return bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
43 					0, BIOSET_NEED_BVECS);
44 }
45 
46 void f2fs_destroy_bioset(void)
47 {
48 	bioset_exit(&f2fs_bioset);
49 }
50 
51 static bool __is_cp_guaranteed(struct page *page)
52 {
53 	struct address_space *mapping = page->mapping;
54 	struct inode *inode;
55 	struct f2fs_sb_info *sbi;
56 
57 	if (!mapping)
58 		return false;
59 
60 	inode = mapping->host;
61 	sbi = F2FS_I_SB(inode);
62 
63 	if (inode->i_ino == F2FS_META_INO(sbi) ||
64 			inode->i_ino == F2FS_NODE_INO(sbi) ||
65 			S_ISDIR(inode->i_mode))
66 		return true;
67 
68 	if (f2fs_is_compressed_page(page))
69 		return false;
70 	if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) ||
71 			page_private_gcing(page))
72 		return true;
73 	return false;
74 }
75 
76 static enum count_type __read_io_type(struct page *page)
77 {
78 	struct address_space *mapping = page_file_mapping(page);
79 
80 	if (mapping) {
81 		struct inode *inode = mapping->host;
82 		struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
83 
84 		if (inode->i_ino == F2FS_META_INO(sbi))
85 			return F2FS_RD_META;
86 
87 		if (inode->i_ino == F2FS_NODE_INO(sbi))
88 			return F2FS_RD_NODE;
89 	}
90 	return F2FS_RD_DATA;
91 }
92 
93 /* postprocessing steps for read bios */
94 enum bio_post_read_step {
95 #ifdef CONFIG_FS_ENCRYPTION
96 	STEP_DECRYPT	= 1 << 0,
97 #else
98 	STEP_DECRYPT	= 0,	/* compile out the decryption-related code */
99 #endif
100 #ifdef CONFIG_F2FS_FS_COMPRESSION
101 	STEP_DECOMPRESS	= 1 << 1,
102 #else
103 	STEP_DECOMPRESS	= 0,	/* compile out the decompression-related code */
104 #endif
105 #ifdef CONFIG_FS_VERITY
106 	STEP_VERITY	= 1 << 2,
107 #else
108 	STEP_VERITY	= 0,	/* compile out the verity-related code */
109 #endif
110 };
111 
112 struct bio_post_read_ctx {
113 	struct bio *bio;
114 	struct f2fs_sb_info *sbi;
115 	struct work_struct work;
116 	unsigned int enabled_steps;
117 	/*
118 	 * decompression_attempted keeps track of whether
119 	 * f2fs_end_read_compressed_page() has been called on the pages in the
120 	 * bio that belong to a compressed cluster yet.
121 	 */
122 	bool decompression_attempted;
123 	block_t fs_blkaddr;
124 };
125 
126 /*
127  * Update and unlock a bio's pages, and free the bio.
128  *
129  * This marks pages up-to-date only if there was no error in the bio (I/O error,
130  * decryption error, or verity error), as indicated by bio->bi_status.
131  *
132  * "Compressed pages" (pagecache pages backed by a compressed cluster on-disk)
133  * aren't marked up-to-date here, as decompression is done on a per-compression-
134  * cluster basis rather than a per-bio basis.  Instead, we only must do two
135  * things for each compressed page here: call f2fs_end_read_compressed_page()
136  * with failed=true if an error occurred before it would have normally gotten
137  * called (i.e., I/O error or decryption error, but *not* verity error), and
138  * release the bio's reference to the decompress_io_ctx of the page's cluster.
139  */
140 static void f2fs_finish_read_bio(struct bio *bio, bool in_task)
141 {
142 	struct bio_vec *bv;
143 	struct bvec_iter_all iter_all;
144 	struct bio_post_read_ctx *ctx = bio->bi_private;
145 
146 	bio_for_each_segment_all(bv, bio, iter_all) {
147 		struct page *page = bv->bv_page;
148 
149 		if (f2fs_is_compressed_page(page)) {
150 			if (ctx && !ctx->decompression_attempted)
151 				f2fs_end_read_compressed_page(page, true, 0,
152 							in_task);
153 			f2fs_put_page_dic(page, in_task);
154 			continue;
155 		}
156 
157 		if (bio->bi_status)
158 			ClearPageUptodate(page);
159 		else
160 			SetPageUptodate(page);
161 		dec_page_count(F2FS_P_SB(page), __read_io_type(page));
162 		unlock_page(page);
163 	}
164 
165 	if (ctx)
166 		mempool_free(ctx, bio_post_read_ctx_pool);
167 	bio_put(bio);
168 }
169 
170 static void f2fs_verify_bio(struct work_struct *work)
171 {
172 	struct bio_post_read_ctx *ctx =
173 		container_of(work, struct bio_post_read_ctx, work);
174 	struct bio *bio = ctx->bio;
175 	bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
176 
177 	/*
178 	 * fsverity_verify_bio() may call readahead() again, and while verity
179 	 * will be disabled for this, decryption and/or decompression may still
180 	 * be needed, resulting in another bio_post_read_ctx being allocated.
181 	 * So to prevent deadlocks we need to release the current ctx to the
182 	 * mempool first.  This assumes that verity is the last post-read step.
183 	 */
184 	mempool_free(ctx, bio_post_read_ctx_pool);
185 	bio->bi_private = NULL;
186 
187 	/*
188 	 * Verify the bio's pages with fs-verity.  Exclude compressed pages,
189 	 * as those were handled separately by f2fs_end_read_compressed_page().
190 	 */
191 	if (may_have_compressed_pages) {
192 		struct bio_vec *bv;
193 		struct bvec_iter_all iter_all;
194 
195 		bio_for_each_segment_all(bv, bio, iter_all) {
196 			struct page *page = bv->bv_page;
197 
198 			if (!f2fs_is_compressed_page(page) &&
199 			    !fsverity_verify_page(page)) {
200 				bio->bi_status = BLK_STS_IOERR;
201 				break;
202 			}
203 		}
204 	} else {
205 		fsverity_verify_bio(bio);
206 	}
207 
208 	f2fs_finish_read_bio(bio, true);
209 }
210 
211 /*
212  * If the bio's data needs to be verified with fs-verity, then enqueue the
213  * verity work for the bio.  Otherwise finish the bio now.
214  *
215  * Note that to avoid deadlocks, the verity work can't be done on the
216  * decryption/decompression workqueue.  This is because verifying the data pages
217  * can involve reading verity metadata pages from the file, and these verity
218  * metadata pages may be encrypted and/or compressed.
219  */
220 static void f2fs_verify_and_finish_bio(struct bio *bio, bool in_task)
221 {
222 	struct bio_post_read_ctx *ctx = bio->bi_private;
223 
224 	if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
225 		INIT_WORK(&ctx->work, f2fs_verify_bio);
226 		fsverity_enqueue_verify_work(&ctx->work);
227 	} else {
228 		f2fs_finish_read_bio(bio, in_task);
229 	}
230 }
231 
232 /*
233  * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
234  * remaining page was read by @ctx->bio.
235  *
236  * Note that a bio may span clusters (even a mix of compressed and uncompressed
237  * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
238  * that the bio includes at least one compressed page.  The actual decompression
239  * is done on a per-cluster basis, not a per-bio basis.
240  */
241 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx,
242 		bool in_task)
243 {
244 	struct bio_vec *bv;
245 	struct bvec_iter_all iter_all;
246 	bool all_compressed = true;
247 	block_t blkaddr = ctx->fs_blkaddr;
248 
249 	bio_for_each_segment_all(bv, ctx->bio, iter_all) {
250 		struct page *page = bv->bv_page;
251 
252 		if (f2fs_is_compressed_page(page))
253 			f2fs_end_read_compressed_page(page, false, blkaddr,
254 						      in_task);
255 		else
256 			all_compressed = false;
257 
258 		blkaddr++;
259 	}
260 
261 	ctx->decompression_attempted = true;
262 
263 	/*
264 	 * Optimization: if all the bio's pages are compressed, then scheduling
265 	 * the per-bio verity work is unnecessary, as verity will be fully
266 	 * handled at the compression cluster level.
267 	 */
268 	if (all_compressed)
269 		ctx->enabled_steps &= ~STEP_VERITY;
270 }
271 
272 static void f2fs_post_read_work(struct work_struct *work)
273 {
274 	struct bio_post_read_ctx *ctx =
275 		container_of(work, struct bio_post_read_ctx, work);
276 	struct bio *bio = ctx->bio;
277 
278 	if ((ctx->enabled_steps & STEP_DECRYPT) && !fscrypt_decrypt_bio(bio)) {
279 		f2fs_finish_read_bio(bio, true);
280 		return;
281 	}
282 
283 	if (ctx->enabled_steps & STEP_DECOMPRESS)
284 		f2fs_handle_step_decompress(ctx, true);
285 
286 	f2fs_verify_and_finish_bio(bio, true);
287 }
288 
289 static void f2fs_read_end_io(struct bio *bio)
290 {
291 	struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
292 	struct bio_post_read_ctx *ctx;
293 	bool intask = in_task();
294 
295 	iostat_update_and_unbind_ctx(bio, 0);
296 	ctx = bio->bi_private;
297 
298 	if (time_to_inject(sbi, FAULT_READ_IO)) {
299 		f2fs_show_injection_info(sbi, FAULT_READ_IO);
300 		bio->bi_status = BLK_STS_IOERR;
301 	}
302 
303 	if (bio->bi_status) {
304 		f2fs_finish_read_bio(bio, intask);
305 		return;
306 	}
307 
308 	if (ctx) {
309 		unsigned int enabled_steps = ctx->enabled_steps &
310 					(STEP_DECRYPT | STEP_DECOMPRESS);
311 
312 		/*
313 		 * If we have only decompression step between decompression and
314 		 * decrypt, we don't need post processing for this.
315 		 */
316 		if (enabled_steps == STEP_DECOMPRESS &&
317 				!f2fs_low_mem_mode(sbi)) {
318 			f2fs_handle_step_decompress(ctx, intask);
319 		} else if (enabled_steps) {
320 			INIT_WORK(&ctx->work, f2fs_post_read_work);
321 			queue_work(ctx->sbi->post_read_wq, &ctx->work);
322 			return;
323 		}
324 	}
325 
326 	f2fs_verify_and_finish_bio(bio, intask);
327 }
328 
329 static void f2fs_write_end_io(struct bio *bio)
330 {
331 	struct f2fs_sb_info *sbi;
332 	struct bio_vec *bvec;
333 	struct bvec_iter_all iter_all;
334 
335 	iostat_update_and_unbind_ctx(bio, 1);
336 	sbi = bio->bi_private;
337 
338 	if (time_to_inject(sbi, FAULT_WRITE_IO)) {
339 		f2fs_show_injection_info(sbi, FAULT_WRITE_IO);
340 		bio->bi_status = BLK_STS_IOERR;
341 	}
342 
343 	bio_for_each_segment_all(bvec, bio, iter_all) {
344 		struct page *page = bvec->bv_page;
345 		enum count_type type = WB_DATA_TYPE(page);
346 
347 		if (page_private_dummy(page)) {
348 			clear_page_private_dummy(page);
349 			unlock_page(page);
350 			mempool_free(page, sbi->write_io_dummy);
351 
352 			if (unlikely(bio->bi_status))
353 				f2fs_stop_checkpoint(sbi, true,
354 						STOP_CP_REASON_WRITE_FAIL);
355 			continue;
356 		}
357 
358 		fscrypt_finalize_bounce_page(&page);
359 
360 #ifdef CONFIG_F2FS_FS_COMPRESSION
361 		if (f2fs_is_compressed_page(page)) {
362 			f2fs_compress_write_end_io(bio, page);
363 			continue;
364 		}
365 #endif
366 
367 		if (unlikely(bio->bi_status)) {
368 			mapping_set_error(page->mapping, -EIO);
369 			if (type == F2FS_WB_CP_DATA)
370 				f2fs_stop_checkpoint(sbi, true,
371 						STOP_CP_REASON_WRITE_FAIL);
372 		}
373 
374 		f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
375 					page->index != nid_of_node(page));
376 
377 		dec_page_count(sbi, type);
378 		if (f2fs_in_warm_node_list(sbi, page))
379 			f2fs_del_fsync_node_entry(sbi, page);
380 		clear_page_private_gcing(page);
381 		end_page_writeback(page);
382 	}
383 	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
384 				wq_has_sleeper(&sbi->cp_wait))
385 		wake_up(&sbi->cp_wait);
386 
387 	bio_put(bio);
388 }
389 
390 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
391 		block_t blk_addr, sector_t *sector)
392 {
393 	struct block_device *bdev = sbi->sb->s_bdev;
394 	int i;
395 
396 	if (f2fs_is_multi_device(sbi)) {
397 		for (i = 0; i < sbi->s_ndevs; i++) {
398 			if (FDEV(i).start_blk <= blk_addr &&
399 			    FDEV(i).end_blk >= blk_addr) {
400 				blk_addr -= FDEV(i).start_blk;
401 				bdev = FDEV(i).bdev;
402 				break;
403 			}
404 		}
405 	}
406 
407 	if (sector)
408 		*sector = SECTOR_FROM_BLOCK(blk_addr);
409 	return bdev;
410 }
411 
412 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
413 {
414 	int i;
415 
416 	if (!f2fs_is_multi_device(sbi))
417 		return 0;
418 
419 	for (i = 0; i < sbi->s_ndevs; i++)
420 		if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
421 			return i;
422 	return 0;
423 }
424 
425 static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio)
426 {
427 	unsigned int temp_mask = (1 << NR_TEMP_TYPE) - 1;
428 	unsigned int fua_flag, meta_flag, io_flag;
429 	blk_opf_t op_flags = 0;
430 
431 	if (fio->op != REQ_OP_WRITE)
432 		return 0;
433 	if (fio->type == DATA)
434 		io_flag = fio->sbi->data_io_flag;
435 	else if (fio->type == NODE)
436 		io_flag = fio->sbi->node_io_flag;
437 	else
438 		return 0;
439 
440 	fua_flag = io_flag & temp_mask;
441 	meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
442 
443 	/*
444 	 * data/node io flag bits per temp:
445 	 *      REQ_META     |      REQ_FUA      |
446 	 *    5 |    4 |   3 |    2 |    1 |   0 |
447 	 * Cold | Warm | Hot | Cold | Warm | Hot |
448 	 */
449 	if ((1 << fio->temp) & meta_flag)
450 		op_flags |= REQ_META;
451 	if ((1 << fio->temp) & fua_flag)
452 		op_flags |= REQ_FUA;
453 	return op_flags;
454 }
455 
456 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
457 {
458 	struct f2fs_sb_info *sbi = fio->sbi;
459 	struct block_device *bdev;
460 	sector_t sector;
461 	struct bio *bio;
462 
463 	bdev = f2fs_target_device(sbi, fio->new_blkaddr, &sector);
464 	bio = bio_alloc_bioset(bdev, npages,
465 				fio->op | fio->op_flags | f2fs_io_flags(fio),
466 				GFP_NOIO, &f2fs_bioset);
467 	bio->bi_iter.bi_sector = sector;
468 	if (is_read_io(fio->op)) {
469 		bio->bi_end_io = f2fs_read_end_io;
470 		bio->bi_private = NULL;
471 	} else {
472 		bio->bi_end_io = f2fs_write_end_io;
473 		bio->bi_private = sbi;
474 	}
475 	iostat_alloc_and_bind_ctx(sbi, bio, NULL);
476 
477 	if (fio->io_wbc)
478 		wbc_init_bio(fio->io_wbc, bio);
479 
480 	return bio;
481 }
482 
483 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
484 				  pgoff_t first_idx,
485 				  const struct f2fs_io_info *fio,
486 				  gfp_t gfp_mask)
487 {
488 	/*
489 	 * The f2fs garbage collector sets ->encrypted_page when it wants to
490 	 * read/write raw data without encryption.
491 	 */
492 	if (!fio || !fio->encrypted_page)
493 		fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
494 }
495 
496 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
497 				     pgoff_t next_idx,
498 				     const struct f2fs_io_info *fio)
499 {
500 	/*
501 	 * The f2fs garbage collector sets ->encrypted_page when it wants to
502 	 * read/write raw data without encryption.
503 	 */
504 	if (fio && fio->encrypted_page)
505 		return !bio_has_crypt_ctx(bio);
506 
507 	return fscrypt_mergeable_bio(bio, inode, next_idx);
508 }
509 
510 static inline void __submit_bio(struct f2fs_sb_info *sbi,
511 				struct bio *bio, enum page_type type)
512 {
513 	if (!is_read_io(bio_op(bio))) {
514 		unsigned int start;
515 
516 		if (type != DATA && type != NODE)
517 			goto submit_io;
518 
519 		if (f2fs_lfs_mode(sbi) && current->plug)
520 			blk_finish_plug(current->plug);
521 
522 		if (!F2FS_IO_ALIGNED(sbi))
523 			goto submit_io;
524 
525 		start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
526 		start %= F2FS_IO_SIZE(sbi);
527 
528 		if (start == 0)
529 			goto submit_io;
530 
531 		/* fill dummy pages */
532 		for (; start < F2FS_IO_SIZE(sbi); start++) {
533 			struct page *page =
534 				mempool_alloc(sbi->write_io_dummy,
535 					      GFP_NOIO | __GFP_NOFAIL);
536 			f2fs_bug_on(sbi, !page);
537 
538 			lock_page(page);
539 
540 			zero_user_segment(page, 0, PAGE_SIZE);
541 			set_page_private_dummy(page);
542 
543 			if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
544 				f2fs_bug_on(sbi, 1);
545 		}
546 		/*
547 		 * In the NODE case, we lose next block address chain. So, we
548 		 * need to do checkpoint in f2fs_sync_file.
549 		 */
550 		if (type == NODE)
551 			set_sbi_flag(sbi, SBI_NEED_CP);
552 	}
553 submit_io:
554 	if (is_read_io(bio_op(bio)))
555 		trace_f2fs_submit_read_bio(sbi->sb, type, bio);
556 	else
557 		trace_f2fs_submit_write_bio(sbi->sb, type, bio);
558 
559 	iostat_update_submit_ctx(bio, type);
560 	submit_bio(bio);
561 }
562 
563 void f2fs_submit_bio(struct f2fs_sb_info *sbi,
564 				struct bio *bio, enum page_type type)
565 {
566 	__submit_bio(sbi, bio, type);
567 }
568 
569 static void __submit_merged_bio(struct f2fs_bio_info *io)
570 {
571 	struct f2fs_io_info *fio = &io->fio;
572 
573 	if (!io->bio)
574 		return;
575 
576 	if (is_read_io(fio->op))
577 		trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
578 	else
579 		trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
580 
581 	__submit_bio(io->sbi, io->bio, fio->type);
582 	io->bio = NULL;
583 }
584 
585 static bool __has_merged_page(struct bio *bio, struct inode *inode,
586 						struct page *page, nid_t ino)
587 {
588 	struct bio_vec *bvec;
589 	struct bvec_iter_all iter_all;
590 
591 	if (!bio)
592 		return false;
593 
594 	if (!inode && !page && !ino)
595 		return true;
596 
597 	bio_for_each_segment_all(bvec, bio, iter_all) {
598 		struct page *target = bvec->bv_page;
599 
600 		if (fscrypt_is_bounce_page(target)) {
601 			target = fscrypt_pagecache_page(target);
602 			if (IS_ERR(target))
603 				continue;
604 		}
605 		if (f2fs_is_compressed_page(target)) {
606 			target = f2fs_compress_control_page(target);
607 			if (IS_ERR(target))
608 				continue;
609 		}
610 
611 		if (inode && inode == target->mapping->host)
612 			return true;
613 		if (page && page == target)
614 			return true;
615 		if (ino && ino == ino_of_node(target))
616 			return true;
617 	}
618 
619 	return false;
620 }
621 
622 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
623 {
624 	int i;
625 
626 	for (i = 0; i < NR_PAGE_TYPE; i++) {
627 		int n = (i == META) ? 1 : NR_TEMP_TYPE;
628 		int j;
629 
630 		sbi->write_io[i] = f2fs_kmalloc(sbi,
631 				array_size(n, sizeof(struct f2fs_bio_info)),
632 				GFP_KERNEL);
633 		if (!sbi->write_io[i])
634 			return -ENOMEM;
635 
636 		for (j = HOT; j < n; j++) {
637 			init_f2fs_rwsem(&sbi->write_io[i][j].io_rwsem);
638 			sbi->write_io[i][j].sbi = sbi;
639 			sbi->write_io[i][j].bio = NULL;
640 			spin_lock_init(&sbi->write_io[i][j].io_lock);
641 			INIT_LIST_HEAD(&sbi->write_io[i][j].io_list);
642 			INIT_LIST_HEAD(&sbi->write_io[i][j].bio_list);
643 			init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock);
644 		}
645 	}
646 
647 	return 0;
648 }
649 
650 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
651 				enum page_type type, enum temp_type temp)
652 {
653 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
654 	struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
655 
656 	f2fs_down_write(&io->io_rwsem);
657 
658 	/* change META to META_FLUSH in the checkpoint procedure */
659 	if (type >= META_FLUSH) {
660 		io->fio.type = META_FLUSH;
661 		io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
662 		if (!test_opt(sbi, NOBARRIER))
663 			io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
664 	}
665 	__submit_merged_bio(io);
666 	f2fs_up_write(&io->io_rwsem);
667 }
668 
669 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
670 				struct inode *inode, struct page *page,
671 				nid_t ino, enum page_type type, bool force)
672 {
673 	enum temp_type temp;
674 	bool ret = true;
675 
676 	for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
677 		if (!force)	{
678 			enum page_type btype = PAGE_TYPE_OF_BIO(type);
679 			struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
680 
681 			f2fs_down_read(&io->io_rwsem);
682 			ret = __has_merged_page(io->bio, inode, page, ino);
683 			f2fs_up_read(&io->io_rwsem);
684 		}
685 		if (ret)
686 			__f2fs_submit_merged_write(sbi, type, temp);
687 
688 		/* TODO: use HOT temp only for meta pages now. */
689 		if (type >= META)
690 			break;
691 	}
692 }
693 
694 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
695 {
696 	__submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
697 }
698 
699 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
700 				struct inode *inode, struct page *page,
701 				nid_t ino, enum page_type type)
702 {
703 	__submit_merged_write_cond(sbi, inode, page, ino, type, false);
704 }
705 
706 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
707 {
708 	f2fs_submit_merged_write(sbi, DATA);
709 	f2fs_submit_merged_write(sbi, NODE);
710 	f2fs_submit_merged_write(sbi, META);
711 }
712 
713 /*
714  * Fill the locked page with data located in the block address.
715  * A caller needs to unlock the page on failure.
716  */
717 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
718 {
719 	struct bio *bio;
720 	struct page *page = fio->encrypted_page ?
721 			fio->encrypted_page : fio->page;
722 
723 	if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
724 			fio->is_por ? META_POR : (__is_meta_io(fio) ?
725 			META_GENERIC : DATA_GENERIC_ENHANCE))) {
726 		f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
727 		return -EFSCORRUPTED;
728 	}
729 
730 	trace_f2fs_submit_page_bio(page, fio);
731 
732 	/* Allocate a new bio */
733 	bio = __bio_alloc(fio, 1);
734 
735 	f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
736 			       fio->page->index, fio, GFP_NOIO);
737 
738 	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
739 		bio_put(bio);
740 		return -EFAULT;
741 	}
742 
743 	if (fio->io_wbc && !is_read_io(fio->op))
744 		wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
745 
746 	inc_page_count(fio->sbi, is_read_io(fio->op) ?
747 			__read_io_type(page) : WB_DATA_TYPE(fio->page));
748 
749 	__submit_bio(fio->sbi, bio, fio->type);
750 	return 0;
751 }
752 
753 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
754 				block_t last_blkaddr, block_t cur_blkaddr)
755 {
756 	if (unlikely(sbi->max_io_bytes &&
757 			bio->bi_iter.bi_size >= sbi->max_io_bytes))
758 		return false;
759 	if (last_blkaddr + 1 != cur_blkaddr)
760 		return false;
761 	return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
762 }
763 
764 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
765 						struct f2fs_io_info *fio)
766 {
767 	if (io->fio.op != fio->op)
768 		return false;
769 	return io->fio.op_flags == fio->op_flags;
770 }
771 
772 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
773 					struct f2fs_bio_info *io,
774 					struct f2fs_io_info *fio,
775 					block_t last_blkaddr,
776 					block_t cur_blkaddr)
777 {
778 	if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
779 		unsigned int filled_blocks =
780 				F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
781 		unsigned int io_size = F2FS_IO_SIZE(sbi);
782 		unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
783 
784 		/* IOs in bio is aligned and left space of vectors is not enough */
785 		if (!(filled_blocks % io_size) && left_vecs < io_size)
786 			return false;
787 	}
788 	if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
789 		return false;
790 	return io_type_is_mergeable(io, fio);
791 }
792 
793 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
794 				struct page *page, enum temp_type temp)
795 {
796 	struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
797 	struct bio_entry *be;
798 
799 	be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
800 	be->bio = bio;
801 	bio_get(bio);
802 
803 	if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
804 		f2fs_bug_on(sbi, 1);
805 
806 	f2fs_down_write(&io->bio_list_lock);
807 	list_add_tail(&be->list, &io->bio_list);
808 	f2fs_up_write(&io->bio_list_lock);
809 }
810 
811 static void del_bio_entry(struct bio_entry *be)
812 {
813 	list_del(&be->list);
814 	kmem_cache_free(bio_entry_slab, be);
815 }
816 
817 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
818 							struct page *page)
819 {
820 	struct f2fs_sb_info *sbi = fio->sbi;
821 	enum temp_type temp;
822 	bool found = false;
823 	int ret = -EAGAIN;
824 
825 	for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
826 		struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
827 		struct list_head *head = &io->bio_list;
828 		struct bio_entry *be;
829 
830 		f2fs_down_write(&io->bio_list_lock);
831 		list_for_each_entry(be, head, list) {
832 			if (be->bio != *bio)
833 				continue;
834 
835 			found = true;
836 
837 			f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
838 							    *fio->last_block,
839 							    fio->new_blkaddr));
840 			if (f2fs_crypt_mergeable_bio(*bio,
841 					fio->page->mapping->host,
842 					fio->page->index, fio) &&
843 			    bio_add_page(*bio, page, PAGE_SIZE, 0) ==
844 					PAGE_SIZE) {
845 				ret = 0;
846 				break;
847 			}
848 
849 			/* page can't be merged into bio; submit the bio */
850 			del_bio_entry(be);
851 			__submit_bio(sbi, *bio, DATA);
852 			break;
853 		}
854 		f2fs_up_write(&io->bio_list_lock);
855 	}
856 
857 	if (ret) {
858 		bio_put(*bio);
859 		*bio = NULL;
860 	}
861 
862 	return ret;
863 }
864 
865 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
866 					struct bio **bio, struct page *page)
867 {
868 	enum temp_type temp;
869 	bool found = false;
870 	struct bio *target = bio ? *bio : NULL;
871 
872 	for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
873 		struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
874 		struct list_head *head = &io->bio_list;
875 		struct bio_entry *be;
876 
877 		if (list_empty(head))
878 			continue;
879 
880 		f2fs_down_read(&io->bio_list_lock);
881 		list_for_each_entry(be, head, list) {
882 			if (target)
883 				found = (target == be->bio);
884 			else
885 				found = __has_merged_page(be->bio, NULL,
886 								page, 0);
887 			if (found)
888 				break;
889 		}
890 		f2fs_up_read(&io->bio_list_lock);
891 
892 		if (!found)
893 			continue;
894 
895 		found = false;
896 
897 		f2fs_down_write(&io->bio_list_lock);
898 		list_for_each_entry(be, head, list) {
899 			if (target)
900 				found = (target == be->bio);
901 			else
902 				found = __has_merged_page(be->bio, NULL,
903 								page, 0);
904 			if (found) {
905 				target = be->bio;
906 				del_bio_entry(be);
907 				break;
908 			}
909 		}
910 		f2fs_up_write(&io->bio_list_lock);
911 	}
912 
913 	if (found)
914 		__submit_bio(sbi, target, DATA);
915 	if (bio && *bio) {
916 		bio_put(*bio);
917 		*bio = NULL;
918 	}
919 }
920 
921 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
922 {
923 	struct bio *bio = *fio->bio;
924 	struct page *page = fio->encrypted_page ?
925 			fio->encrypted_page : fio->page;
926 
927 	if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
928 			__is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) {
929 		f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
930 		return -EFSCORRUPTED;
931 	}
932 
933 	trace_f2fs_submit_page_bio(page, fio);
934 
935 	if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
936 						fio->new_blkaddr))
937 		f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
938 alloc_new:
939 	if (!bio) {
940 		bio = __bio_alloc(fio, BIO_MAX_VECS);
941 		f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
942 				       fio->page->index, fio, GFP_NOIO);
943 
944 		add_bio_entry(fio->sbi, bio, page, fio->temp);
945 	} else {
946 		if (add_ipu_page(fio, &bio, page))
947 			goto alloc_new;
948 	}
949 
950 	if (fio->io_wbc)
951 		wbc_account_cgroup_owner(fio->io_wbc, page, PAGE_SIZE);
952 
953 	inc_page_count(fio->sbi, WB_DATA_TYPE(page));
954 
955 	*fio->last_block = fio->new_blkaddr;
956 	*fio->bio = bio;
957 
958 	return 0;
959 }
960 
961 void f2fs_submit_page_write(struct f2fs_io_info *fio)
962 {
963 	struct f2fs_sb_info *sbi = fio->sbi;
964 	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
965 	struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
966 	struct page *bio_page;
967 
968 	f2fs_bug_on(sbi, is_read_io(fio->op));
969 
970 	f2fs_down_write(&io->io_rwsem);
971 next:
972 	if (fio->in_list) {
973 		spin_lock(&io->io_lock);
974 		if (list_empty(&io->io_list)) {
975 			spin_unlock(&io->io_lock);
976 			goto out;
977 		}
978 		fio = list_first_entry(&io->io_list,
979 						struct f2fs_io_info, list);
980 		list_del(&fio->list);
981 		spin_unlock(&io->io_lock);
982 	}
983 
984 	verify_fio_blkaddr(fio);
985 
986 	if (fio->encrypted_page)
987 		bio_page = fio->encrypted_page;
988 	else if (fio->compressed_page)
989 		bio_page = fio->compressed_page;
990 	else
991 		bio_page = fio->page;
992 
993 	/* set submitted = true as a return value */
994 	fio->submitted = true;
995 
996 	inc_page_count(sbi, WB_DATA_TYPE(bio_page));
997 
998 	if (io->bio &&
999 	    (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
1000 			      fio->new_blkaddr) ||
1001 	     !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
1002 				       bio_page->index, fio)))
1003 		__submit_merged_bio(io);
1004 alloc_new:
1005 	if (io->bio == NULL) {
1006 		if (F2FS_IO_ALIGNED(sbi) &&
1007 				(fio->type == DATA || fio->type == NODE) &&
1008 				fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
1009 			dec_page_count(sbi, WB_DATA_TYPE(bio_page));
1010 			fio->retry = true;
1011 			goto skip;
1012 		}
1013 		io->bio = __bio_alloc(fio, BIO_MAX_VECS);
1014 		f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
1015 				       bio_page->index, fio, GFP_NOIO);
1016 		io->fio = *fio;
1017 	}
1018 
1019 	if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1020 		__submit_merged_bio(io);
1021 		goto alloc_new;
1022 	}
1023 
1024 	if (fio->io_wbc)
1025 		wbc_account_cgroup_owner(fio->io_wbc, bio_page, PAGE_SIZE);
1026 
1027 	io->last_block_in_bio = fio->new_blkaddr;
1028 
1029 	trace_f2fs_submit_page_write(fio->page, fio);
1030 skip:
1031 	if (fio->in_list)
1032 		goto next;
1033 out:
1034 	if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1035 				!f2fs_is_checkpoint_ready(sbi))
1036 		__submit_merged_bio(io);
1037 	f2fs_up_write(&io->io_rwsem);
1038 }
1039 
1040 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1041 				      unsigned nr_pages, blk_opf_t op_flag,
1042 				      pgoff_t first_idx, bool for_write)
1043 {
1044 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1045 	struct bio *bio;
1046 	struct bio_post_read_ctx *ctx = NULL;
1047 	unsigned int post_read_steps = 0;
1048 	sector_t sector;
1049 	struct block_device *bdev = f2fs_target_device(sbi, blkaddr, &sector);
1050 
1051 	bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
1052 			       REQ_OP_READ | op_flag,
1053 			       for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
1054 	if (!bio)
1055 		return ERR_PTR(-ENOMEM);
1056 	bio->bi_iter.bi_sector = sector;
1057 	f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1058 	bio->bi_end_io = f2fs_read_end_io;
1059 
1060 	if (fscrypt_inode_uses_fs_layer_crypto(inode))
1061 		post_read_steps |= STEP_DECRYPT;
1062 
1063 	if (f2fs_need_verity(inode, first_idx))
1064 		post_read_steps |= STEP_VERITY;
1065 
1066 	/*
1067 	 * STEP_DECOMPRESS is handled specially, since a compressed file might
1068 	 * contain both compressed and uncompressed clusters.  We'll allocate a
1069 	 * bio_post_read_ctx if the file is compressed, but the caller is
1070 	 * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1071 	 */
1072 
1073 	if (post_read_steps || f2fs_compressed_file(inode)) {
1074 		/* Due to the mempool, this never fails. */
1075 		ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1076 		ctx->bio = bio;
1077 		ctx->sbi = sbi;
1078 		ctx->enabled_steps = post_read_steps;
1079 		ctx->fs_blkaddr = blkaddr;
1080 		ctx->decompression_attempted = false;
1081 		bio->bi_private = ctx;
1082 	}
1083 	iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1084 
1085 	return bio;
1086 }
1087 
1088 /* This can handle encryption stuffs */
1089 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1090 				 block_t blkaddr, blk_opf_t op_flags,
1091 				 bool for_write)
1092 {
1093 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1094 	struct bio *bio;
1095 
1096 	bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1097 					page->index, for_write);
1098 	if (IS_ERR(bio))
1099 		return PTR_ERR(bio);
1100 
1101 	/* wait for GCed page writeback via META_MAPPING */
1102 	f2fs_wait_on_block_writeback(inode, blkaddr);
1103 
1104 	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1105 		bio_put(bio);
1106 		return -EFAULT;
1107 	}
1108 	inc_page_count(sbi, F2FS_RD_DATA);
1109 	f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE);
1110 	__submit_bio(sbi, bio, DATA);
1111 	return 0;
1112 }
1113 
1114 static void __set_data_blkaddr(struct dnode_of_data *dn)
1115 {
1116 	struct f2fs_node *rn = F2FS_NODE(dn->node_page);
1117 	__le32 *addr_array;
1118 	int base = 0;
1119 
1120 	if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
1121 		base = get_extra_isize(dn->inode);
1122 
1123 	/* Get physical address of data block */
1124 	addr_array = blkaddr_in_node(rn);
1125 	addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1126 }
1127 
1128 /*
1129  * Lock ordering for the change of data block address:
1130  * ->data_page
1131  *  ->node_page
1132  *    update block addresses in the node page
1133  */
1134 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
1135 {
1136 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1137 	__set_data_blkaddr(dn);
1138 	if (set_page_dirty(dn->node_page))
1139 		dn->node_changed = true;
1140 }
1141 
1142 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1143 {
1144 	dn->data_blkaddr = blkaddr;
1145 	f2fs_set_data_blkaddr(dn);
1146 	f2fs_update_read_extent_cache(dn);
1147 }
1148 
1149 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
1150 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1151 {
1152 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1153 	int err;
1154 
1155 	if (!count)
1156 		return 0;
1157 
1158 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1159 		return -EPERM;
1160 	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1161 		return err;
1162 
1163 	trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1164 						dn->ofs_in_node, count);
1165 
1166 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1167 
1168 	for (; count > 0; dn->ofs_in_node++) {
1169 		block_t blkaddr = f2fs_data_blkaddr(dn);
1170 
1171 		if (blkaddr == NULL_ADDR) {
1172 			dn->data_blkaddr = NEW_ADDR;
1173 			__set_data_blkaddr(dn);
1174 			count--;
1175 		}
1176 	}
1177 
1178 	if (set_page_dirty(dn->node_page))
1179 		dn->node_changed = true;
1180 	return 0;
1181 }
1182 
1183 /* Should keep dn->ofs_in_node unchanged */
1184 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1185 {
1186 	unsigned int ofs_in_node = dn->ofs_in_node;
1187 	int ret;
1188 
1189 	ret = f2fs_reserve_new_blocks(dn, 1);
1190 	dn->ofs_in_node = ofs_in_node;
1191 	return ret;
1192 }
1193 
1194 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1195 {
1196 	bool need_put = dn->inode_page ? false : true;
1197 	int err;
1198 
1199 	err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1200 	if (err)
1201 		return err;
1202 
1203 	if (dn->data_blkaddr == NULL_ADDR)
1204 		err = f2fs_reserve_new_block(dn);
1205 	if (err || need_put)
1206 		f2fs_put_dnode(dn);
1207 	return err;
1208 }
1209 
1210 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
1211 {
1212 	struct extent_info ei = {0, };
1213 	struct inode *inode = dn->inode;
1214 
1215 	if (f2fs_lookup_read_extent_cache(inode, index, &ei)) {
1216 		dn->data_blkaddr = ei.blk + index - ei.fofs;
1217 		return 0;
1218 	}
1219 
1220 	return f2fs_reserve_block(dn, index);
1221 }
1222 
1223 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1224 				     blk_opf_t op_flags, bool for_write,
1225 				     pgoff_t *next_pgofs)
1226 {
1227 	struct address_space *mapping = inode->i_mapping;
1228 	struct dnode_of_data dn;
1229 	struct page *page;
1230 	struct extent_info ei = {0, };
1231 	int err;
1232 
1233 	page = f2fs_grab_cache_page(mapping, index, for_write);
1234 	if (!page)
1235 		return ERR_PTR(-ENOMEM);
1236 
1237 	if (f2fs_lookup_read_extent_cache(inode, index, &ei)) {
1238 		dn.data_blkaddr = ei.blk + index - ei.fofs;
1239 		if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1240 						DATA_GENERIC_ENHANCE_READ)) {
1241 			err = -EFSCORRUPTED;
1242 			f2fs_handle_error(F2FS_I_SB(inode),
1243 						ERROR_INVALID_BLKADDR);
1244 			goto put_err;
1245 		}
1246 		goto got_it;
1247 	}
1248 
1249 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1250 	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1251 	if (err) {
1252 		if (err == -ENOENT && next_pgofs)
1253 			*next_pgofs = f2fs_get_next_page_offset(&dn, index);
1254 		goto put_err;
1255 	}
1256 	f2fs_put_dnode(&dn);
1257 
1258 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1259 		err = -ENOENT;
1260 		if (next_pgofs)
1261 			*next_pgofs = index + 1;
1262 		goto put_err;
1263 	}
1264 	if (dn.data_blkaddr != NEW_ADDR &&
1265 			!f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1266 						dn.data_blkaddr,
1267 						DATA_GENERIC_ENHANCE)) {
1268 		err = -EFSCORRUPTED;
1269 		f2fs_handle_error(F2FS_I_SB(inode),
1270 					ERROR_INVALID_BLKADDR);
1271 		goto put_err;
1272 	}
1273 got_it:
1274 	if (PageUptodate(page)) {
1275 		unlock_page(page);
1276 		return page;
1277 	}
1278 
1279 	/*
1280 	 * A new dentry page is allocated but not able to be written, since its
1281 	 * new inode page couldn't be allocated due to -ENOSPC.
1282 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
1283 	 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1284 	 * f2fs_init_inode_metadata.
1285 	 */
1286 	if (dn.data_blkaddr == NEW_ADDR) {
1287 		zero_user_segment(page, 0, PAGE_SIZE);
1288 		if (!PageUptodate(page))
1289 			SetPageUptodate(page);
1290 		unlock_page(page);
1291 		return page;
1292 	}
1293 
1294 	err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1295 						op_flags, for_write);
1296 	if (err)
1297 		goto put_err;
1298 	return page;
1299 
1300 put_err:
1301 	f2fs_put_page(page, 1);
1302 	return ERR_PTR(err);
1303 }
1304 
1305 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index,
1306 					pgoff_t *next_pgofs)
1307 {
1308 	struct address_space *mapping = inode->i_mapping;
1309 	struct page *page;
1310 
1311 	page = find_get_page(mapping, index);
1312 	if (page && PageUptodate(page))
1313 		return page;
1314 	f2fs_put_page(page, 0);
1315 
1316 	page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs);
1317 	if (IS_ERR(page))
1318 		return page;
1319 
1320 	if (PageUptodate(page))
1321 		return page;
1322 
1323 	wait_on_page_locked(page);
1324 	if (unlikely(!PageUptodate(page))) {
1325 		f2fs_put_page(page, 0);
1326 		return ERR_PTR(-EIO);
1327 	}
1328 	return page;
1329 }
1330 
1331 /*
1332  * If it tries to access a hole, return an error.
1333  * Because, the callers, functions in dir.c and GC, should be able to know
1334  * whether this page exists or not.
1335  */
1336 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1337 							bool for_write)
1338 {
1339 	struct address_space *mapping = inode->i_mapping;
1340 	struct page *page;
1341 repeat:
1342 	page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL);
1343 	if (IS_ERR(page))
1344 		return page;
1345 
1346 	/* wait for read completion */
1347 	lock_page(page);
1348 	if (unlikely(page->mapping != mapping)) {
1349 		f2fs_put_page(page, 1);
1350 		goto repeat;
1351 	}
1352 	if (unlikely(!PageUptodate(page))) {
1353 		f2fs_put_page(page, 1);
1354 		return ERR_PTR(-EIO);
1355 	}
1356 	return page;
1357 }
1358 
1359 /*
1360  * Caller ensures that this data page is never allocated.
1361  * A new zero-filled data page is allocated in the page cache.
1362  *
1363  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1364  * f2fs_unlock_op().
1365  * Note that, ipage is set only by make_empty_dir, and if any error occur,
1366  * ipage should be released by this function.
1367  */
1368 struct page *f2fs_get_new_data_page(struct inode *inode,
1369 		struct page *ipage, pgoff_t index, bool new_i_size)
1370 {
1371 	struct address_space *mapping = inode->i_mapping;
1372 	struct page *page;
1373 	struct dnode_of_data dn;
1374 	int err;
1375 
1376 	page = f2fs_grab_cache_page(mapping, index, true);
1377 	if (!page) {
1378 		/*
1379 		 * before exiting, we should make sure ipage will be released
1380 		 * if any error occur.
1381 		 */
1382 		f2fs_put_page(ipage, 1);
1383 		return ERR_PTR(-ENOMEM);
1384 	}
1385 
1386 	set_new_dnode(&dn, inode, ipage, NULL, 0);
1387 	err = f2fs_reserve_block(&dn, index);
1388 	if (err) {
1389 		f2fs_put_page(page, 1);
1390 		return ERR_PTR(err);
1391 	}
1392 	if (!ipage)
1393 		f2fs_put_dnode(&dn);
1394 
1395 	if (PageUptodate(page))
1396 		goto got_it;
1397 
1398 	if (dn.data_blkaddr == NEW_ADDR) {
1399 		zero_user_segment(page, 0, PAGE_SIZE);
1400 		if (!PageUptodate(page))
1401 			SetPageUptodate(page);
1402 	} else {
1403 		f2fs_put_page(page, 1);
1404 
1405 		/* if ipage exists, blkaddr should be NEW_ADDR */
1406 		f2fs_bug_on(F2FS_I_SB(inode), ipage);
1407 		page = f2fs_get_lock_data_page(inode, index, true);
1408 		if (IS_ERR(page))
1409 			return page;
1410 	}
1411 got_it:
1412 	if (new_i_size && i_size_read(inode) <
1413 				((loff_t)(index + 1) << PAGE_SHIFT))
1414 		f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1415 	return page;
1416 }
1417 
1418 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1419 {
1420 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1421 	struct f2fs_summary sum;
1422 	struct node_info ni;
1423 	block_t old_blkaddr;
1424 	blkcnt_t count = 1;
1425 	int err;
1426 
1427 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1428 		return -EPERM;
1429 
1430 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1431 	if (err)
1432 		return err;
1433 
1434 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
1435 	if (dn->data_blkaddr != NULL_ADDR)
1436 		goto alloc;
1437 
1438 	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1439 		return err;
1440 
1441 alloc:
1442 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1443 	old_blkaddr = dn->data_blkaddr;
1444 	f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1445 				&sum, seg_type, NULL);
1446 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
1447 		invalidate_mapping_pages(META_MAPPING(sbi),
1448 					old_blkaddr, old_blkaddr);
1449 		f2fs_invalidate_compress_page(sbi, old_blkaddr);
1450 	}
1451 	f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1452 	return 0;
1453 }
1454 
1455 void f2fs_do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
1456 {
1457 	if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1458 		if (lock)
1459 			f2fs_down_read(&sbi->node_change);
1460 		else
1461 			f2fs_up_read(&sbi->node_change);
1462 	} else {
1463 		if (lock)
1464 			f2fs_lock_op(sbi);
1465 		else
1466 			f2fs_unlock_op(sbi);
1467 	}
1468 }
1469 
1470 /*
1471  * f2fs_map_blocks() tries to find or build mapping relationship which
1472  * maps continuous logical blocks to physical blocks, and return such
1473  * info via f2fs_map_blocks structure.
1474  */
1475 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1476 						int create, int flag)
1477 {
1478 	unsigned int maxblocks = map->m_len;
1479 	struct dnode_of_data dn;
1480 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1481 	int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1482 	pgoff_t pgofs, end_offset, end;
1483 	int err = 0, ofs = 1;
1484 	unsigned int ofs_in_node, last_ofs_in_node;
1485 	blkcnt_t prealloc;
1486 	struct extent_info ei = {0, };
1487 	block_t blkaddr;
1488 	unsigned int start_pgofs;
1489 	int bidx = 0;
1490 
1491 	if (!maxblocks)
1492 		return 0;
1493 
1494 	map->m_bdev = inode->i_sb->s_bdev;
1495 	map->m_multidev_dio =
1496 		f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1497 
1498 	map->m_len = 0;
1499 	map->m_flags = 0;
1500 
1501 	/* it only supports block size == page size */
1502 	pgofs =	(pgoff_t)map->m_lblk;
1503 	end = pgofs + maxblocks;
1504 
1505 	if (!create && f2fs_lookup_read_extent_cache(inode, pgofs, &ei)) {
1506 		if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1507 							map->m_may_create)
1508 			goto next_dnode;
1509 
1510 		map->m_pblk = ei.blk + pgofs - ei.fofs;
1511 		map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1512 		map->m_flags = F2FS_MAP_MAPPED;
1513 		if (map->m_next_extent)
1514 			*map->m_next_extent = pgofs + map->m_len;
1515 
1516 		/* for hardware encryption, but to avoid potential issue in future */
1517 		if (flag == F2FS_GET_BLOCK_DIO)
1518 			f2fs_wait_on_block_writeback_range(inode,
1519 						map->m_pblk, map->m_len);
1520 
1521 		if (map->m_multidev_dio) {
1522 			block_t blk_addr = map->m_pblk;
1523 
1524 			bidx = f2fs_target_device_index(sbi, map->m_pblk);
1525 
1526 			map->m_bdev = FDEV(bidx).bdev;
1527 			map->m_pblk -= FDEV(bidx).start_blk;
1528 			map->m_len = min(map->m_len,
1529 				FDEV(bidx).end_blk + 1 - map->m_pblk);
1530 
1531 			if (map->m_may_create)
1532 				f2fs_update_device_state(sbi, inode->i_ino,
1533 							blk_addr, map->m_len);
1534 		}
1535 		goto out;
1536 	}
1537 
1538 next_dnode:
1539 	if (map->m_may_create)
1540 		f2fs_do_map_lock(sbi, flag, true);
1541 
1542 	/* When reading holes, we need its node page */
1543 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1544 	err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1545 	if (err) {
1546 		if (flag == F2FS_GET_BLOCK_BMAP)
1547 			map->m_pblk = 0;
1548 
1549 		if (err == -ENOENT) {
1550 			/*
1551 			 * There is one exceptional case that read_node_page()
1552 			 * may return -ENOENT due to filesystem has been
1553 			 * shutdown or cp_error, so force to convert error
1554 			 * number to EIO for such case.
1555 			 */
1556 			if (map->m_may_create &&
1557 				(is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1558 				f2fs_cp_error(sbi))) {
1559 				err = -EIO;
1560 				goto unlock_out;
1561 			}
1562 
1563 			err = 0;
1564 			if (map->m_next_pgofs)
1565 				*map->m_next_pgofs =
1566 					f2fs_get_next_page_offset(&dn, pgofs);
1567 			if (map->m_next_extent)
1568 				*map->m_next_extent =
1569 					f2fs_get_next_page_offset(&dn, pgofs);
1570 		}
1571 		goto unlock_out;
1572 	}
1573 
1574 	start_pgofs = pgofs;
1575 	prealloc = 0;
1576 	last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1577 	end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1578 
1579 next_block:
1580 	blkaddr = f2fs_data_blkaddr(&dn);
1581 
1582 	if (__is_valid_data_blkaddr(blkaddr) &&
1583 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1584 		err = -EFSCORRUPTED;
1585 		f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1586 		goto sync_out;
1587 	}
1588 
1589 	if (__is_valid_data_blkaddr(blkaddr)) {
1590 		/* use out-place-update for driect IO under LFS mode */
1591 		if (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO &&
1592 							map->m_may_create) {
1593 			err = __allocate_data_block(&dn, map->m_seg_type);
1594 			if (err)
1595 				goto sync_out;
1596 			blkaddr = dn.data_blkaddr;
1597 			set_inode_flag(inode, FI_APPEND_WRITE);
1598 		}
1599 	} else {
1600 		if (create) {
1601 			if (unlikely(f2fs_cp_error(sbi))) {
1602 				err = -EIO;
1603 				goto sync_out;
1604 			}
1605 			if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1606 				if (blkaddr == NULL_ADDR) {
1607 					prealloc++;
1608 					last_ofs_in_node = dn.ofs_in_node;
1609 				}
1610 			} else {
1611 				WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1612 					flag != F2FS_GET_BLOCK_DIO);
1613 				err = __allocate_data_block(&dn,
1614 							map->m_seg_type);
1615 				if (!err) {
1616 					if (flag == F2FS_GET_BLOCK_PRE_DIO)
1617 						file_need_truncate(inode);
1618 					set_inode_flag(inode, FI_APPEND_WRITE);
1619 				}
1620 			}
1621 			if (err)
1622 				goto sync_out;
1623 			map->m_flags |= F2FS_MAP_NEW;
1624 			blkaddr = dn.data_blkaddr;
1625 		} else {
1626 			if (f2fs_compressed_file(inode) &&
1627 					f2fs_sanity_check_cluster(&dn) &&
1628 					(flag != F2FS_GET_BLOCK_FIEMAP ||
1629 					IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1630 				err = -EFSCORRUPTED;
1631 				f2fs_handle_error(sbi,
1632 						ERROR_CORRUPTED_CLUSTER);
1633 				goto sync_out;
1634 			}
1635 			if (flag == F2FS_GET_BLOCK_BMAP) {
1636 				map->m_pblk = 0;
1637 				goto sync_out;
1638 			}
1639 			if (flag == F2FS_GET_BLOCK_PRECACHE)
1640 				goto sync_out;
1641 			if (flag == F2FS_GET_BLOCK_FIEMAP &&
1642 						blkaddr == NULL_ADDR) {
1643 				if (map->m_next_pgofs)
1644 					*map->m_next_pgofs = pgofs + 1;
1645 				goto sync_out;
1646 			}
1647 			if (flag != F2FS_GET_BLOCK_FIEMAP) {
1648 				/* for defragment case */
1649 				if (map->m_next_pgofs)
1650 					*map->m_next_pgofs = pgofs + 1;
1651 				goto sync_out;
1652 			}
1653 		}
1654 	}
1655 
1656 	if (flag == F2FS_GET_BLOCK_PRE_AIO)
1657 		goto skip;
1658 
1659 	if (map->m_multidev_dio)
1660 		bidx = f2fs_target_device_index(sbi, blkaddr);
1661 
1662 	if (map->m_len == 0) {
1663 		/* preallocated unwritten block should be mapped for fiemap. */
1664 		if (blkaddr == NEW_ADDR)
1665 			map->m_flags |= F2FS_MAP_UNWRITTEN;
1666 		map->m_flags |= F2FS_MAP_MAPPED;
1667 
1668 		map->m_pblk = blkaddr;
1669 		map->m_len = 1;
1670 
1671 		if (map->m_multidev_dio)
1672 			map->m_bdev = FDEV(bidx).bdev;
1673 	} else if ((map->m_pblk != NEW_ADDR &&
1674 			blkaddr == (map->m_pblk + ofs)) ||
1675 			(map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1676 			flag == F2FS_GET_BLOCK_PRE_DIO) {
1677 		if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1678 			goto sync_out;
1679 		ofs++;
1680 		map->m_len++;
1681 	} else {
1682 		goto sync_out;
1683 	}
1684 
1685 skip:
1686 	dn.ofs_in_node++;
1687 	pgofs++;
1688 
1689 	/* preallocate blocks in batch for one dnode page */
1690 	if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1691 			(pgofs == end || dn.ofs_in_node == end_offset)) {
1692 
1693 		dn.ofs_in_node = ofs_in_node;
1694 		err = f2fs_reserve_new_blocks(&dn, prealloc);
1695 		if (err)
1696 			goto sync_out;
1697 
1698 		map->m_len += dn.ofs_in_node - ofs_in_node;
1699 		if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1700 			err = -ENOSPC;
1701 			goto sync_out;
1702 		}
1703 		dn.ofs_in_node = end_offset;
1704 	}
1705 
1706 	if (pgofs >= end)
1707 		goto sync_out;
1708 	else if (dn.ofs_in_node < end_offset)
1709 		goto next_block;
1710 
1711 	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1712 		if (map->m_flags & F2FS_MAP_MAPPED) {
1713 			unsigned int ofs = start_pgofs - map->m_lblk;
1714 
1715 			f2fs_update_read_extent_cache_range(&dn,
1716 				start_pgofs, map->m_pblk + ofs,
1717 				map->m_len - ofs);
1718 		}
1719 	}
1720 
1721 	f2fs_put_dnode(&dn);
1722 
1723 	if (map->m_may_create) {
1724 		f2fs_do_map_lock(sbi, flag, false);
1725 		f2fs_balance_fs(sbi, dn.node_changed);
1726 	}
1727 	goto next_dnode;
1728 
1729 sync_out:
1730 
1731 	if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1732 		/*
1733 		 * for hardware encryption, but to avoid potential issue
1734 		 * in future
1735 		 */
1736 		f2fs_wait_on_block_writeback_range(inode,
1737 						map->m_pblk, map->m_len);
1738 
1739 		if (map->m_multidev_dio) {
1740 			block_t blk_addr = map->m_pblk;
1741 
1742 			bidx = f2fs_target_device_index(sbi, map->m_pblk);
1743 
1744 			map->m_bdev = FDEV(bidx).bdev;
1745 			map->m_pblk -= FDEV(bidx).start_blk;
1746 
1747 			if (map->m_may_create)
1748 				f2fs_update_device_state(sbi, inode->i_ino,
1749 							blk_addr, map->m_len);
1750 
1751 			f2fs_bug_on(sbi, blk_addr + map->m_len >
1752 						FDEV(bidx).end_blk + 1);
1753 		}
1754 	}
1755 
1756 	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1757 		if (map->m_flags & F2FS_MAP_MAPPED) {
1758 			unsigned int ofs = start_pgofs - map->m_lblk;
1759 
1760 			f2fs_update_read_extent_cache_range(&dn,
1761 				start_pgofs, map->m_pblk + ofs,
1762 				map->m_len - ofs);
1763 		}
1764 		if (map->m_next_extent)
1765 			*map->m_next_extent = pgofs + 1;
1766 	}
1767 	f2fs_put_dnode(&dn);
1768 unlock_out:
1769 	if (map->m_may_create) {
1770 		f2fs_do_map_lock(sbi, flag, false);
1771 		f2fs_balance_fs(sbi, dn.node_changed);
1772 	}
1773 out:
1774 	trace_f2fs_map_blocks(inode, map, create, flag, err);
1775 	return err;
1776 }
1777 
1778 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1779 {
1780 	struct f2fs_map_blocks map;
1781 	block_t last_lblk;
1782 	int err;
1783 
1784 	if (pos + len > i_size_read(inode))
1785 		return false;
1786 
1787 	map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1788 	map.m_next_pgofs = NULL;
1789 	map.m_next_extent = NULL;
1790 	map.m_seg_type = NO_CHECK_TYPE;
1791 	map.m_may_create = false;
1792 	last_lblk = F2FS_BLK_ALIGN(pos + len);
1793 
1794 	while (map.m_lblk < last_lblk) {
1795 		map.m_len = last_lblk - map.m_lblk;
1796 		err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1797 		if (err || map.m_len == 0)
1798 			return false;
1799 		map.m_lblk += map.m_len;
1800 	}
1801 	return true;
1802 }
1803 
1804 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1805 {
1806 	return (bytes >> inode->i_blkbits);
1807 }
1808 
1809 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1810 {
1811 	return (blks << inode->i_blkbits);
1812 }
1813 
1814 static int f2fs_xattr_fiemap(struct inode *inode,
1815 				struct fiemap_extent_info *fieinfo)
1816 {
1817 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1818 	struct page *page;
1819 	struct node_info ni;
1820 	__u64 phys = 0, len;
1821 	__u32 flags;
1822 	nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1823 	int err = 0;
1824 
1825 	if (f2fs_has_inline_xattr(inode)) {
1826 		int offset;
1827 
1828 		page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1829 						inode->i_ino, false);
1830 		if (!page)
1831 			return -ENOMEM;
1832 
1833 		err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1834 		if (err) {
1835 			f2fs_put_page(page, 1);
1836 			return err;
1837 		}
1838 
1839 		phys = blks_to_bytes(inode, ni.blk_addr);
1840 		offset = offsetof(struct f2fs_inode, i_addr) +
1841 					sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1842 					get_inline_xattr_addrs(inode));
1843 
1844 		phys += offset;
1845 		len = inline_xattr_size(inode);
1846 
1847 		f2fs_put_page(page, 1);
1848 
1849 		flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1850 
1851 		if (!xnid)
1852 			flags |= FIEMAP_EXTENT_LAST;
1853 
1854 		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1855 		trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1856 		if (err)
1857 			return err;
1858 	}
1859 
1860 	if (xnid) {
1861 		page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1862 		if (!page)
1863 			return -ENOMEM;
1864 
1865 		err = f2fs_get_node_info(sbi, xnid, &ni, false);
1866 		if (err) {
1867 			f2fs_put_page(page, 1);
1868 			return err;
1869 		}
1870 
1871 		phys = blks_to_bytes(inode, ni.blk_addr);
1872 		len = inode->i_sb->s_blocksize;
1873 
1874 		f2fs_put_page(page, 1);
1875 
1876 		flags = FIEMAP_EXTENT_LAST;
1877 	}
1878 
1879 	if (phys) {
1880 		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1881 		trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1882 	}
1883 
1884 	return (err < 0 ? err : 0);
1885 }
1886 
1887 static loff_t max_inode_blocks(struct inode *inode)
1888 {
1889 	loff_t result = ADDRS_PER_INODE(inode);
1890 	loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1891 
1892 	/* two direct node blocks */
1893 	result += (leaf_count * 2);
1894 
1895 	/* two indirect node blocks */
1896 	leaf_count *= NIDS_PER_BLOCK;
1897 	result += (leaf_count * 2);
1898 
1899 	/* one double indirect node block */
1900 	leaf_count *= NIDS_PER_BLOCK;
1901 	result += leaf_count;
1902 
1903 	return result;
1904 }
1905 
1906 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1907 		u64 start, u64 len)
1908 {
1909 	struct f2fs_map_blocks map;
1910 	sector_t start_blk, last_blk;
1911 	pgoff_t next_pgofs;
1912 	u64 logical = 0, phys = 0, size = 0;
1913 	u32 flags = 0;
1914 	int ret = 0;
1915 	bool compr_cluster = false, compr_appended;
1916 	unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1917 	unsigned int count_in_cluster = 0;
1918 	loff_t maxbytes;
1919 
1920 	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1921 		ret = f2fs_precache_extents(inode);
1922 		if (ret)
1923 			return ret;
1924 	}
1925 
1926 	ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1927 	if (ret)
1928 		return ret;
1929 
1930 	inode_lock(inode);
1931 
1932 	maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1933 	if (start > maxbytes) {
1934 		ret = -EFBIG;
1935 		goto out;
1936 	}
1937 
1938 	if (len > maxbytes || (maxbytes - len) < start)
1939 		len = maxbytes - start;
1940 
1941 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1942 		ret = f2fs_xattr_fiemap(inode, fieinfo);
1943 		goto out;
1944 	}
1945 
1946 	if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1947 		ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1948 		if (ret != -EAGAIN)
1949 			goto out;
1950 	}
1951 
1952 	if (bytes_to_blks(inode, len) == 0)
1953 		len = blks_to_bytes(inode, 1);
1954 
1955 	start_blk = bytes_to_blks(inode, start);
1956 	last_blk = bytes_to_blks(inode, start + len - 1);
1957 
1958 next:
1959 	memset(&map, 0, sizeof(map));
1960 	map.m_lblk = start_blk;
1961 	map.m_len = bytes_to_blks(inode, len);
1962 	map.m_next_pgofs = &next_pgofs;
1963 	map.m_seg_type = NO_CHECK_TYPE;
1964 
1965 	if (compr_cluster) {
1966 		map.m_lblk += 1;
1967 		map.m_len = cluster_size - count_in_cluster;
1968 	}
1969 
1970 	ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
1971 	if (ret)
1972 		goto out;
1973 
1974 	/* HOLE */
1975 	if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1976 		start_blk = next_pgofs;
1977 
1978 		if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1979 						max_inode_blocks(inode)))
1980 			goto prep_next;
1981 
1982 		flags |= FIEMAP_EXTENT_LAST;
1983 	}
1984 
1985 	compr_appended = false;
1986 	/* In a case of compressed cluster, append this to the last extent */
1987 	if (compr_cluster && ((map.m_flags & F2FS_MAP_UNWRITTEN) ||
1988 			!(map.m_flags & F2FS_MAP_FLAGS))) {
1989 		compr_appended = true;
1990 		goto skip_fill;
1991 	}
1992 
1993 	if (size) {
1994 		flags |= FIEMAP_EXTENT_MERGED;
1995 		if (IS_ENCRYPTED(inode))
1996 			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1997 
1998 		ret = fiemap_fill_next_extent(fieinfo, logical,
1999 				phys, size, flags);
2000 		trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
2001 		if (ret)
2002 			goto out;
2003 		size = 0;
2004 	}
2005 
2006 	if (start_blk > last_blk)
2007 		goto out;
2008 
2009 skip_fill:
2010 	if (map.m_pblk == COMPRESS_ADDR) {
2011 		compr_cluster = true;
2012 		count_in_cluster = 1;
2013 	} else if (compr_appended) {
2014 		unsigned int appended_blks = cluster_size -
2015 						count_in_cluster + 1;
2016 		size += blks_to_bytes(inode, appended_blks);
2017 		start_blk += appended_blks;
2018 		compr_cluster = false;
2019 	} else {
2020 		logical = blks_to_bytes(inode, start_blk);
2021 		phys = __is_valid_data_blkaddr(map.m_pblk) ?
2022 			blks_to_bytes(inode, map.m_pblk) : 0;
2023 		size = blks_to_bytes(inode, map.m_len);
2024 		flags = 0;
2025 
2026 		if (compr_cluster) {
2027 			flags = FIEMAP_EXTENT_ENCODED;
2028 			count_in_cluster += map.m_len;
2029 			if (count_in_cluster == cluster_size) {
2030 				compr_cluster = false;
2031 				size += blks_to_bytes(inode, 1);
2032 			}
2033 		} else if (map.m_flags & F2FS_MAP_UNWRITTEN) {
2034 			flags = FIEMAP_EXTENT_UNWRITTEN;
2035 		}
2036 
2037 		start_blk += bytes_to_blks(inode, size);
2038 	}
2039 
2040 prep_next:
2041 	cond_resched();
2042 	if (fatal_signal_pending(current))
2043 		ret = -EINTR;
2044 	else
2045 		goto next;
2046 out:
2047 	if (ret == 1)
2048 		ret = 0;
2049 
2050 	inode_unlock(inode);
2051 	return ret;
2052 }
2053 
2054 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2055 {
2056 	if (IS_ENABLED(CONFIG_FS_VERITY) &&
2057 	    (IS_VERITY(inode) || f2fs_verity_in_progress(inode)))
2058 		return inode->i_sb->s_maxbytes;
2059 
2060 	return i_size_read(inode);
2061 }
2062 
2063 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2064 					unsigned nr_pages,
2065 					struct f2fs_map_blocks *map,
2066 					struct bio **bio_ret,
2067 					sector_t *last_block_in_bio,
2068 					bool is_readahead)
2069 {
2070 	struct bio *bio = *bio_ret;
2071 	const unsigned blocksize = blks_to_bytes(inode, 1);
2072 	sector_t block_in_file;
2073 	sector_t last_block;
2074 	sector_t last_block_in_file;
2075 	sector_t block_nr;
2076 	int ret = 0;
2077 
2078 	block_in_file = (sector_t)page_index(page);
2079 	last_block = block_in_file + nr_pages;
2080 	last_block_in_file = bytes_to_blks(inode,
2081 			f2fs_readpage_limit(inode) + blocksize - 1);
2082 	if (last_block > last_block_in_file)
2083 		last_block = last_block_in_file;
2084 
2085 	/* just zeroing out page which is beyond EOF */
2086 	if (block_in_file >= last_block)
2087 		goto zero_out;
2088 	/*
2089 	 * Map blocks using the previous result first.
2090 	 */
2091 	if ((map->m_flags & F2FS_MAP_MAPPED) &&
2092 			block_in_file > map->m_lblk &&
2093 			block_in_file < (map->m_lblk + map->m_len))
2094 		goto got_it;
2095 
2096 	/*
2097 	 * Then do more f2fs_map_blocks() calls until we are
2098 	 * done with this page.
2099 	 */
2100 	map->m_lblk = block_in_file;
2101 	map->m_len = last_block - block_in_file;
2102 
2103 	ret = f2fs_map_blocks(inode, map, 0, F2FS_GET_BLOCK_DEFAULT);
2104 	if (ret)
2105 		goto out;
2106 got_it:
2107 	if ((map->m_flags & F2FS_MAP_MAPPED)) {
2108 		block_nr = map->m_pblk + block_in_file - map->m_lblk;
2109 		SetPageMappedToDisk(page);
2110 
2111 		if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2112 						DATA_GENERIC_ENHANCE_READ)) {
2113 			ret = -EFSCORRUPTED;
2114 			f2fs_handle_error(F2FS_I_SB(inode),
2115 						ERROR_INVALID_BLKADDR);
2116 			goto out;
2117 		}
2118 	} else {
2119 zero_out:
2120 		zero_user_segment(page, 0, PAGE_SIZE);
2121 		if (f2fs_need_verity(inode, page->index) &&
2122 		    !fsverity_verify_page(page)) {
2123 			ret = -EIO;
2124 			goto out;
2125 		}
2126 		if (!PageUptodate(page))
2127 			SetPageUptodate(page);
2128 		unlock_page(page);
2129 		goto out;
2130 	}
2131 
2132 	/*
2133 	 * This page will go to BIO.  Do we need to send this
2134 	 * BIO off first?
2135 	 */
2136 	if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2137 				       *last_block_in_bio, block_nr) ||
2138 		    !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2139 submit_and_realloc:
2140 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
2141 		bio = NULL;
2142 	}
2143 	if (bio == NULL) {
2144 		bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2145 				is_readahead ? REQ_RAHEAD : 0, page->index,
2146 				false);
2147 		if (IS_ERR(bio)) {
2148 			ret = PTR_ERR(bio);
2149 			bio = NULL;
2150 			goto out;
2151 		}
2152 	}
2153 
2154 	/*
2155 	 * If the page is under writeback, we need to wait for
2156 	 * its completion to see the correct decrypted data.
2157 	 */
2158 	f2fs_wait_on_block_writeback(inode, block_nr);
2159 
2160 	if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2161 		goto submit_and_realloc;
2162 
2163 	inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2164 	f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO,
2165 							F2FS_BLKSIZE);
2166 	*last_block_in_bio = block_nr;
2167 	goto out;
2168 out:
2169 	*bio_ret = bio;
2170 	return ret;
2171 }
2172 
2173 #ifdef CONFIG_F2FS_FS_COMPRESSION
2174 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2175 				unsigned nr_pages, sector_t *last_block_in_bio,
2176 				bool is_readahead, bool for_write)
2177 {
2178 	struct dnode_of_data dn;
2179 	struct inode *inode = cc->inode;
2180 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2181 	struct bio *bio = *bio_ret;
2182 	unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2183 	sector_t last_block_in_file;
2184 	const unsigned blocksize = blks_to_bytes(inode, 1);
2185 	struct decompress_io_ctx *dic = NULL;
2186 	struct extent_info ei = {};
2187 	bool from_dnode = true;
2188 	int i;
2189 	int ret = 0;
2190 
2191 	f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2192 
2193 	last_block_in_file = bytes_to_blks(inode,
2194 			f2fs_readpage_limit(inode) + blocksize - 1);
2195 
2196 	/* get rid of pages beyond EOF */
2197 	for (i = 0; i < cc->cluster_size; i++) {
2198 		struct page *page = cc->rpages[i];
2199 
2200 		if (!page)
2201 			continue;
2202 		if ((sector_t)page->index >= last_block_in_file) {
2203 			zero_user_segment(page, 0, PAGE_SIZE);
2204 			if (!PageUptodate(page))
2205 				SetPageUptodate(page);
2206 		} else if (!PageUptodate(page)) {
2207 			continue;
2208 		}
2209 		unlock_page(page);
2210 		if (for_write)
2211 			put_page(page);
2212 		cc->rpages[i] = NULL;
2213 		cc->nr_rpages--;
2214 	}
2215 
2216 	/* we are done since all pages are beyond EOF */
2217 	if (f2fs_cluster_is_empty(cc))
2218 		goto out;
2219 
2220 	if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei))
2221 		from_dnode = false;
2222 
2223 	if (!from_dnode)
2224 		goto skip_reading_dnode;
2225 
2226 	set_new_dnode(&dn, inode, NULL, NULL, 0);
2227 	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2228 	if (ret)
2229 		goto out;
2230 
2231 	f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2232 
2233 skip_reading_dnode:
2234 	for (i = 1; i < cc->cluster_size; i++) {
2235 		block_t blkaddr;
2236 
2237 		blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2238 					dn.ofs_in_node + i) :
2239 					ei.blk + i - 1;
2240 
2241 		if (!__is_valid_data_blkaddr(blkaddr))
2242 			break;
2243 
2244 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2245 			ret = -EFAULT;
2246 			goto out_put_dnode;
2247 		}
2248 		cc->nr_cpages++;
2249 
2250 		if (!from_dnode && i >= ei.c_len)
2251 			break;
2252 	}
2253 
2254 	/* nothing to decompress */
2255 	if (cc->nr_cpages == 0) {
2256 		ret = 0;
2257 		goto out_put_dnode;
2258 	}
2259 
2260 	dic = f2fs_alloc_dic(cc);
2261 	if (IS_ERR(dic)) {
2262 		ret = PTR_ERR(dic);
2263 		goto out_put_dnode;
2264 	}
2265 
2266 	for (i = 0; i < cc->nr_cpages; i++) {
2267 		struct page *page = dic->cpages[i];
2268 		block_t blkaddr;
2269 		struct bio_post_read_ctx *ctx;
2270 
2271 		blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2272 					dn.ofs_in_node + i + 1) :
2273 					ei.blk + i;
2274 
2275 		f2fs_wait_on_block_writeback(inode, blkaddr);
2276 
2277 		if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2278 			if (atomic_dec_and_test(&dic->remaining_pages))
2279 				f2fs_decompress_cluster(dic, true);
2280 			continue;
2281 		}
2282 
2283 		if (bio && (!page_is_mergeable(sbi, bio,
2284 					*last_block_in_bio, blkaddr) ||
2285 		    !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2286 submit_and_realloc:
2287 			__submit_bio(sbi, bio, DATA);
2288 			bio = NULL;
2289 		}
2290 
2291 		if (!bio) {
2292 			bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2293 					is_readahead ? REQ_RAHEAD : 0,
2294 					page->index, for_write);
2295 			if (IS_ERR(bio)) {
2296 				ret = PTR_ERR(bio);
2297 				f2fs_decompress_end_io(dic, ret, true);
2298 				f2fs_put_dnode(&dn);
2299 				*bio_ret = NULL;
2300 				return ret;
2301 			}
2302 		}
2303 
2304 		if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2305 			goto submit_and_realloc;
2306 
2307 		ctx = get_post_read_ctx(bio);
2308 		ctx->enabled_steps |= STEP_DECOMPRESS;
2309 		refcount_inc(&dic->refcnt);
2310 
2311 		inc_page_count(sbi, F2FS_RD_DATA);
2312 		f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
2313 		*last_block_in_bio = blkaddr;
2314 	}
2315 
2316 	if (from_dnode)
2317 		f2fs_put_dnode(&dn);
2318 
2319 	*bio_ret = bio;
2320 	return 0;
2321 
2322 out_put_dnode:
2323 	if (from_dnode)
2324 		f2fs_put_dnode(&dn);
2325 out:
2326 	for (i = 0; i < cc->cluster_size; i++) {
2327 		if (cc->rpages[i]) {
2328 			ClearPageUptodate(cc->rpages[i]);
2329 			unlock_page(cc->rpages[i]);
2330 		}
2331 	}
2332 	*bio_ret = bio;
2333 	return ret;
2334 }
2335 #endif
2336 
2337 /*
2338  * This function was originally taken from fs/mpage.c, and customized for f2fs.
2339  * Major change was from block_size == page_size in f2fs by default.
2340  */
2341 static int f2fs_mpage_readpages(struct inode *inode,
2342 		struct readahead_control *rac, struct page *page)
2343 {
2344 	struct bio *bio = NULL;
2345 	sector_t last_block_in_bio = 0;
2346 	struct f2fs_map_blocks map;
2347 #ifdef CONFIG_F2FS_FS_COMPRESSION
2348 	struct compress_ctx cc = {
2349 		.inode = inode,
2350 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2351 		.cluster_size = F2FS_I(inode)->i_cluster_size,
2352 		.cluster_idx = NULL_CLUSTER,
2353 		.rpages = NULL,
2354 		.cpages = NULL,
2355 		.nr_rpages = 0,
2356 		.nr_cpages = 0,
2357 	};
2358 	pgoff_t nc_cluster_idx = NULL_CLUSTER;
2359 #endif
2360 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
2361 	unsigned max_nr_pages = nr_pages;
2362 	int ret = 0;
2363 
2364 	map.m_pblk = 0;
2365 	map.m_lblk = 0;
2366 	map.m_len = 0;
2367 	map.m_flags = 0;
2368 	map.m_next_pgofs = NULL;
2369 	map.m_next_extent = NULL;
2370 	map.m_seg_type = NO_CHECK_TYPE;
2371 	map.m_may_create = false;
2372 
2373 	for (; nr_pages; nr_pages--) {
2374 		if (rac) {
2375 			page = readahead_page(rac);
2376 			prefetchw(&page->flags);
2377 		}
2378 
2379 #ifdef CONFIG_F2FS_FS_COMPRESSION
2380 		if (f2fs_compressed_file(inode)) {
2381 			/* there are remained comressed pages, submit them */
2382 			if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2383 				ret = f2fs_read_multi_pages(&cc, &bio,
2384 							max_nr_pages,
2385 							&last_block_in_bio,
2386 							rac != NULL, false);
2387 				f2fs_destroy_compress_ctx(&cc, false);
2388 				if (ret)
2389 					goto set_error_page;
2390 			}
2391 			if (cc.cluster_idx == NULL_CLUSTER) {
2392 				if (nc_cluster_idx ==
2393 					page->index >> cc.log_cluster_size) {
2394 					goto read_single_page;
2395 				}
2396 
2397 				ret = f2fs_is_compressed_cluster(inode, page->index);
2398 				if (ret < 0)
2399 					goto set_error_page;
2400 				else if (!ret) {
2401 					nc_cluster_idx =
2402 						page->index >> cc.log_cluster_size;
2403 					goto read_single_page;
2404 				}
2405 
2406 				nc_cluster_idx = NULL_CLUSTER;
2407 			}
2408 			ret = f2fs_init_compress_ctx(&cc);
2409 			if (ret)
2410 				goto set_error_page;
2411 
2412 			f2fs_compress_ctx_add_page(&cc, page);
2413 
2414 			goto next_page;
2415 		}
2416 read_single_page:
2417 #endif
2418 
2419 		ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2420 					&bio, &last_block_in_bio, rac);
2421 		if (ret) {
2422 #ifdef CONFIG_F2FS_FS_COMPRESSION
2423 set_error_page:
2424 #endif
2425 			zero_user_segment(page, 0, PAGE_SIZE);
2426 			unlock_page(page);
2427 		}
2428 #ifdef CONFIG_F2FS_FS_COMPRESSION
2429 next_page:
2430 #endif
2431 		if (rac)
2432 			put_page(page);
2433 
2434 #ifdef CONFIG_F2FS_FS_COMPRESSION
2435 		if (f2fs_compressed_file(inode)) {
2436 			/* last page */
2437 			if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2438 				ret = f2fs_read_multi_pages(&cc, &bio,
2439 							max_nr_pages,
2440 							&last_block_in_bio,
2441 							rac != NULL, false);
2442 				f2fs_destroy_compress_ctx(&cc, false);
2443 			}
2444 		}
2445 #endif
2446 	}
2447 	if (bio)
2448 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
2449 	return ret;
2450 }
2451 
2452 static int f2fs_read_data_folio(struct file *file, struct folio *folio)
2453 {
2454 	struct page *page = &folio->page;
2455 	struct inode *inode = page_file_mapping(page)->host;
2456 	int ret = -EAGAIN;
2457 
2458 	trace_f2fs_readpage(page, DATA);
2459 
2460 	if (!f2fs_is_compress_backend_ready(inode)) {
2461 		unlock_page(page);
2462 		return -EOPNOTSUPP;
2463 	}
2464 
2465 	/* If the file has inline data, try to read it directly */
2466 	if (f2fs_has_inline_data(inode))
2467 		ret = f2fs_read_inline_data(inode, page);
2468 	if (ret == -EAGAIN)
2469 		ret = f2fs_mpage_readpages(inode, NULL, page);
2470 	return ret;
2471 }
2472 
2473 static void f2fs_readahead(struct readahead_control *rac)
2474 {
2475 	struct inode *inode = rac->mapping->host;
2476 
2477 	trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2478 
2479 	if (!f2fs_is_compress_backend_ready(inode))
2480 		return;
2481 
2482 	/* If the file has inline data, skip readahead */
2483 	if (f2fs_has_inline_data(inode))
2484 		return;
2485 
2486 	f2fs_mpage_readpages(inode, rac, NULL);
2487 }
2488 
2489 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2490 {
2491 	struct inode *inode = fio->page->mapping->host;
2492 	struct page *mpage, *page;
2493 	gfp_t gfp_flags = GFP_NOFS;
2494 
2495 	if (!f2fs_encrypted_file(inode))
2496 		return 0;
2497 
2498 	page = fio->compressed_page ? fio->compressed_page : fio->page;
2499 
2500 	/* wait for GCed page writeback via META_MAPPING */
2501 	f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2502 
2503 	if (fscrypt_inode_uses_inline_crypto(inode))
2504 		return 0;
2505 
2506 retry_encrypt:
2507 	fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2508 					PAGE_SIZE, 0, gfp_flags);
2509 	if (IS_ERR(fio->encrypted_page)) {
2510 		/* flush pending IOs and wait for a while in the ENOMEM case */
2511 		if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2512 			f2fs_flush_merged_writes(fio->sbi);
2513 			memalloc_retry_wait(GFP_NOFS);
2514 			gfp_flags |= __GFP_NOFAIL;
2515 			goto retry_encrypt;
2516 		}
2517 		return PTR_ERR(fio->encrypted_page);
2518 	}
2519 
2520 	mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2521 	if (mpage) {
2522 		if (PageUptodate(mpage))
2523 			memcpy(page_address(mpage),
2524 				page_address(fio->encrypted_page), PAGE_SIZE);
2525 		f2fs_put_page(mpage, 1);
2526 	}
2527 	return 0;
2528 }
2529 
2530 static inline bool check_inplace_update_policy(struct inode *inode,
2531 				struct f2fs_io_info *fio)
2532 {
2533 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2534 	unsigned int policy = SM_I(sbi)->ipu_policy;
2535 
2536 	if (policy & (0x1 << F2FS_IPU_HONOR_OPU_WRITE) &&
2537 			is_inode_flag_set(inode, FI_OPU_WRITE))
2538 		return false;
2539 	if (policy & (0x1 << F2FS_IPU_FORCE))
2540 		return true;
2541 	if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
2542 		return true;
2543 	if (policy & (0x1 << F2FS_IPU_UTIL) &&
2544 			utilization(sbi) > SM_I(sbi)->min_ipu_util)
2545 		return true;
2546 	if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
2547 			utilization(sbi) > SM_I(sbi)->min_ipu_util)
2548 		return true;
2549 
2550 	/*
2551 	 * IPU for rewrite async pages
2552 	 */
2553 	if (policy & (0x1 << F2FS_IPU_ASYNC) &&
2554 			fio && fio->op == REQ_OP_WRITE &&
2555 			!(fio->op_flags & REQ_SYNC) &&
2556 			!IS_ENCRYPTED(inode))
2557 		return true;
2558 
2559 	/* this is only set during fdatasync */
2560 	if (policy & (0x1 << F2FS_IPU_FSYNC) &&
2561 			is_inode_flag_set(inode, FI_NEED_IPU))
2562 		return true;
2563 
2564 	if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2565 			!f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2566 		return true;
2567 
2568 	return false;
2569 }
2570 
2571 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2572 {
2573 	/* swap file is migrating in aligned write mode */
2574 	if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2575 		return false;
2576 
2577 	if (f2fs_is_pinned_file(inode))
2578 		return true;
2579 
2580 	/* if this is cold file, we should overwrite to avoid fragmentation */
2581 	if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE))
2582 		return true;
2583 
2584 	return check_inplace_update_policy(inode, fio);
2585 }
2586 
2587 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2588 {
2589 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2590 
2591 	/* The below cases were checked when setting it. */
2592 	if (f2fs_is_pinned_file(inode))
2593 		return false;
2594 	if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2595 		return true;
2596 	if (f2fs_lfs_mode(sbi))
2597 		return true;
2598 	if (S_ISDIR(inode->i_mode))
2599 		return true;
2600 	if (IS_NOQUOTA(inode))
2601 		return true;
2602 	if (f2fs_is_atomic_file(inode))
2603 		return true;
2604 
2605 	/* swap file is migrating in aligned write mode */
2606 	if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2607 		return true;
2608 
2609 	if (is_inode_flag_set(inode, FI_OPU_WRITE))
2610 		return true;
2611 
2612 	if (fio) {
2613 		if (page_private_gcing(fio->page))
2614 			return true;
2615 		if (page_private_dummy(fio->page))
2616 			return true;
2617 		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2618 			f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2619 			return true;
2620 	}
2621 	return false;
2622 }
2623 
2624 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2625 {
2626 	struct inode *inode = fio->page->mapping->host;
2627 
2628 	if (f2fs_should_update_outplace(inode, fio))
2629 		return false;
2630 
2631 	return f2fs_should_update_inplace(inode, fio);
2632 }
2633 
2634 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2635 {
2636 	struct page *page = fio->page;
2637 	struct inode *inode = page->mapping->host;
2638 	struct dnode_of_data dn;
2639 	struct extent_info ei = {0, };
2640 	struct node_info ni;
2641 	bool ipu_force = false;
2642 	int err = 0;
2643 
2644 	/* Use COW inode to make dnode_of_data for atomic write */
2645 	if (f2fs_is_atomic_file(inode))
2646 		set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
2647 	else
2648 		set_new_dnode(&dn, inode, NULL, NULL, 0);
2649 
2650 	if (need_inplace_update(fio) &&
2651 	    f2fs_lookup_read_extent_cache(inode, page->index, &ei)) {
2652 		fio->old_blkaddr = ei.blk + page->index - ei.fofs;
2653 
2654 		if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2655 						DATA_GENERIC_ENHANCE)) {
2656 			f2fs_handle_error(fio->sbi,
2657 						ERROR_INVALID_BLKADDR);
2658 			return -EFSCORRUPTED;
2659 		}
2660 
2661 		ipu_force = true;
2662 		fio->need_lock = LOCK_DONE;
2663 		goto got_it;
2664 	}
2665 
2666 	/* Deadlock due to between page->lock and f2fs_lock_op */
2667 	if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2668 		return -EAGAIN;
2669 
2670 	err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2671 	if (err)
2672 		goto out;
2673 
2674 	fio->old_blkaddr = dn.data_blkaddr;
2675 
2676 	/* This page is already truncated */
2677 	if (fio->old_blkaddr == NULL_ADDR) {
2678 		ClearPageUptodate(page);
2679 		clear_page_private_gcing(page);
2680 		goto out_writepage;
2681 	}
2682 got_it:
2683 	if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2684 		!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2685 						DATA_GENERIC_ENHANCE)) {
2686 		err = -EFSCORRUPTED;
2687 		f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
2688 		goto out_writepage;
2689 	}
2690 
2691 	/*
2692 	 * If current allocation needs SSR,
2693 	 * it had better in-place writes for updated data.
2694 	 */
2695 	if (ipu_force ||
2696 		(__is_valid_data_blkaddr(fio->old_blkaddr) &&
2697 					need_inplace_update(fio))) {
2698 		err = f2fs_encrypt_one_page(fio);
2699 		if (err)
2700 			goto out_writepage;
2701 
2702 		set_page_writeback(page);
2703 		ClearPageError(page);
2704 		f2fs_put_dnode(&dn);
2705 		if (fio->need_lock == LOCK_REQ)
2706 			f2fs_unlock_op(fio->sbi);
2707 		err = f2fs_inplace_write_data(fio);
2708 		if (err) {
2709 			if (fscrypt_inode_uses_fs_layer_crypto(inode))
2710 				fscrypt_finalize_bounce_page(&fio->encrypted_page);
2711 			if (PageWriteback(page))
2712 				end_page_writeback(page);
2713 		} else {
2714 			set_inode_flag(inode, FI_UPDATE_WRITE);
2715 		}
2716 		trace_f2fs_do_write_data_page(fio->page, IPU);
2717 		return err;
2718 	}
2719 
2720 	if (fio->need_lock == LOCK_RETRY) {
2721 		if (!f2fs_trylock_op(fio->sbi)) {
2722 			err = -EAGAIN;
2723 			goto out_writepage;
2724 		}
2725 		fio->need_lock = LOCK_REQ;
2726 	}
2727 
2728 	err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2729 	if (err)
2730 		goto out_writepage;
2731 
2732 	fio->version = ni.version;
2733 
2734 	err = f2fs_encrypt_one_page(fio);
2735 	if (err)
2736 		goto out_writepage;
2737 
2738 	set_page_writeback(page);
2739 	ClearPageError(page);
2740 
2741 	if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2742 		f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2743 
2744 	/* LFS mode write path */
2745 	f2fs_outplace_write_data(&dn, fio);
2746 	trace_f2fs_do_write_data_page(page, OPU);
2747 	set_inode_flag(inode, FI_APPEND_WRITE);
2748 	if (page->index == 0)
2749 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
2750 out_writepage:
2751 	f2fs_put_dnode(&dn);
2752 out:
2753 	if (fio->need_lock == LOCK_REQ)
2754 		f2fs_unlock_op(fio->sbi);
2755 	return err;
2756 }
2757 
2758 int f2fs_write_single_data_page(struct page *page, int *submitted,
2759 				struct bio **bio,
2760 				sector_t *last_block,
2761 				struct writeback_control *wbc,
2762 				enum iostat_type io_type,
2763 				int compr_blocks,
2764 				bool allow_balance)
2765 {
2766 	struct inode *inode = page->mapping->host;
2767 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2768 	loff_t i_size = i_size_read(inode);
2769 	const pgoff_t end_index = ((unsigned long long)i_size)
2770 							>> PAGE_SHIFT;
2771 	loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2772 	unsigned offset = 0;
2773 	bool need_balance_fs = false;
2774 	int err = 0;
2775 	struct f2fs_io_info fio = {
2776 		.sbi = sbi,
2777 		.ino = inode->i_ino,
2778 		.type = DATA,
2779 		.op = REQ_OP_WRITE,
2780 		.op_flags = wbc_to_write_flags(wbc),
2781 		.old_blkaddr = NULL_ADDR,
2782 		.page = page,
2783 		.encrypted_page = NULL,
2784 		.submitted = false,
2785 		.compr_blocks = compr_blocks,
2786 		.need_lock = LOCK_RETRY,
2787 		.post_read = f2fs_post_read_required(inode),
2788 		.io_type = io_type,
2789 		.io_wbc = wbc,
2790 		.bio = bio,
2791 		.last_block = last_block,
2792 	};
2793 
2794 	trace_f2fs_writepage(page, DATA);
2795 
2796 	/* we should bypass data pages to proceed the kworkder jobs */
2797 	if (unlikely(f2fs_cp_error(sbi))) {
2798 		mapping_set_error(page->mapping, -EIO);
2799 		/*
2800 		 * don't drop any dirty dentry pages for keeping lastest
2801 		 * directory structure.
2802 		 */
2803 		if (S_ISDIR(inode->i_mode))
2804 			goto redirty_out;
2805 		goto out;
2806 	}
2807 
2808 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2809 		goto redirty_out;
2810 
2811 	if (page->index < end_index ||
2812 			f2fs_verity_in_progress(inode) ||
2813 			compr_blocks)
2814 		goto write;
2815 
2816 	/*
2817 	 * If the offset is out-of-range of file size,
2818 	 * this page does not have to be written to disk.
2819 	 */
2820 	offset = i_size & (PAGE_SIZE - 1);
2821 	if ((page->index >= end_index + 1) || !offset)
2822 		goto out;
2823 
2824 	zero_user_segment(page, offset, PAGE_SIZE);
2825 write:
2826 	if (f2fs_is_drop_cache(inode))
2827 		goto out;
2828 
2829 	/* Dentry/quota blocks are controlled by checkpoint */
2830 	if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
2831 		/*
2832 		 * We need to wait for node_write to avoid block allocation during
2833 		 * checkpoint. This can only happen to quota writes which can cause
2834 		 * the below discard race condition.
2835 		 */
2836 		if (IS_NOQUOTA(inode))
2837 			f2fs_down_read(&sbi->node_write);
2838 
2839 		fio.need_lock = LOCK_DONE;
2840 		err = f2fs_do_write_data_page(&fio);
2841 
2842 		if (IS_NOQUOTA(inode))
2843 			f2fs_up_read(&sbi->node_write);
2844 
2845 		goto done;
2846 	}
2847 
2848 	if (!wbc->for_reclaim)
2849 		need_balance_fs = true;
2850 	else if (has_not_enough_free_secs(sbi, 0, 0))
2851 		goto redirty_out;
2852 	else
2853 		set_inode_flag(inode, FI_HOT_DATA);
2854 
2855 	err = -EAGAIN;
2856 	if (f2fs_has_inline_data(inode)) {
2857 		err = f2fs_write_inline_data(inode, page);
2858 		if (!err)
2859 			goto out;
2860 	}
2861 
2862 	if (err == -EAGAIN) {
2863 		err = f2fs_do_write_data_page(&fio);
2864 		if (err == -EAGAIN) {
2865 			fio.need_lock = LOCK_REQ;
2866 			err = f2fs_do_write_data_page(&fio);
2867 		}
2868 	}
2869 
2870 	if (err) {
2871 		file_set_keep_isize(inode);
2872 	} else {
2873 		spin_lock(&F2FS_I(inode)->i_size_lock);
2874 		if (F2FS_I(inode)->last_disk_size < psize)
2875 			F2FS_I(inode)->last_disk_size = psize;
2876 		spin_unlock(&F2FS_I(inode)->i_size_lock);
2877 	}
2878 
2879 done:
2880 	if (err && err != -ENOENT)
2881 		goto redirty_out;
2882 
2883 out:
2884 	inode_dec_dirty_pages(inode);
2885 	if (err) {
2886 		ClearPageUptodate(page);
2887 		clear_page_private_gcing(page);
2888 	}
2889 
2890 	if (wbc->for_reclaim) {
2891 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2892 		clear_inode_flag(inode, FI_HOT_DATA);
2893 		f2fs_remove_dirty_inode(inode);
2894 		submitted = NULL;
2895 	}
2896 	unlock_page(page);
2897 	if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2898 			!F2FS_I(inode)->wb_task && allow_balance)
2899 		f2fs_balance_fs(sbi, need_balance_fs);
2900 
2901 	if (unlikely(f2fs_cp_error(sbi))) {
2902 		f2fs_submit_merged_write(sbi, DATA);
2903 		f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2904 		submitted = NULL;
2905 	}
2906 
2907 	if (submitted)
2908 		*submitted = fio.submitted ? 1 : 0;
2909 
2910 	return 0;
2911 
2912 redirty_out:
2913 	redirty_page_for_writepage(wbc, page);
2914 	/*
2915 	 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2916 	 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2917 	 * file_write_and_wait_range() will see EIO error, which is critical
2918 	 * to return value of fsync() followed by atomic_write failure to user.
2919 	 */
2920 	if (!err || wbc->for_reclaim)
2921 		return AOP_WRITEPAGE_ACTIVATE;
2922 	unlock_page(page);
2923 	return err;
2924 }
2925 
2926 static int f2fs_write_data_page(struct page *page,
2927 					struct writeback_control *wbc)
2928 {
2929 #ifdef CONFIG_F2FS_FS_COMPRESSION
2930 	struct inode *inode = page->mapping->host;
2931 
2932 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2933 		goto out;
2934 
2935 	if (f2fs_compressed_file(inode)) {
2936 		if (f2fs_is_compressed_cluster(inode, page->index)) {
2937 			redirty_page_for_writepage(wbc, page);
2938 			return AOP_WRITEPAGE_ACTIVATE;
2939 		}
2940 	}
2941 out:
2942 #endif
2943 
2944 	return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2945 						wbc, FS_DATA_IO, 0, true);
2946 }
2947 
2948 /*
2949  * This function was copied from write_cche_pages from mm/page-writeback.c.
2950  * The major change is making write step of cold data page separately from
2951  * warm/hot data page.
2952  */
2953 static int f2fs_write_cache_pages(struct address_space *mapping,
2954 					struct writeback_control *wbc,
2955 					enum iostat_type io_type)
2956 {
2957 	int ret = 0;
2958 	int done = 0, retry = 0;
2959 	struct page *pages[F2FS_ONSTACK_PAGES];
2960 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2961 	struct bio *bio = NULL;
2962 	sector_t last_block;
2963 #ifdef CONFIG_F2FS_FS_COMPRESSION
2964 	struct inode *inode = mapping->host;
2965 	struct compress_ctx cc = {
2966 		.inode = inode,
2967 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2968 		.cluster_size = F2FS_I(inode)->i_cluster_size,
2969 		.cluster_idx = NULL_CLUSTER,
2970 		.rpages = NULL,
2971 		.nr_rpages = 0,
2972 		.cpages = NULL,
2973 		.valid_nr_cpages = 0,
2974 		.rbuf = NULL,
2975 		.cbuf = NULL,
2976 		.rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2977 		.private = NULL,
2978 	};
2979 #endif
2980 	int nr_pages;
2981 	pgoff_t index;
2982 	pgoff_t end;		/* Inclusive */
2983 	pgoff_t done_index;
2984 	int range_whole = 0;
2985 	xa_mark_t tag;
2986 	int nwritten = 0;
2987 	int submitted = 0;
2988 	int i;
2989 
2990 	if (get_dirty_pages(mapping->host) <=
2991 				SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2992 		set_inode_flag(mapping->host, FI_HOT_DATA);
2993 	else
2994 		clear_inode_flag(mapping->host, FI_HOT_DATA);
2995 
2996 	if (wbc->range_cyclic) {
2997 		index = mapping->writeback_index; /* prev offset */
2998 		end = -1;
2999 	} else {
3000 		index = wbc->range_start >> PAGE_SHIFT;
3001 		end = wbc->range_end >> PAGE_SHIFT;
3002 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3003 			range_whole = 1;
3004 	}
3005 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3006 		tag = PAGECACHE_TAG_TOWRITE;
3007 	else
3008 		tag = PAGECACHE_TAG_DIRTY;
3009 retry:
3010 	retry = 0;
3011 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3012 		tag_pages_for_writeback(mapping, index, end);
3013 	done_index = index;
3014 	while (!done && !retry && (index <= end)) {
3015 		nr_pages = find_get_pages_range_tag(mapping, &index, end,
3016 				tag, F2FS_ONSTACK_PAGES, pages);
3017 		if (nr_pages == 0)
3018 			break;
3019 
3020 		for (i = 0; i < nr_pages; i++) {
3021 			struct page *page = pages[i];
3022 			bool need_readd;
3023 readd:
3024 			need_readd = false;
3025 #ifdef CONFIG_F2FS_FS_COMPRESSION
3026 			if (f2fs_compressed_file(inode)) {
3027 				void *fsdata = NULL;
3028 				struct page *pagep;
3029 				int ret2;
3030 
3031 				ret = f2fs_init_compress_ctx(&cc);
3032 				if (ret) {
3033 					done = 1;
3034 					break;
3035 				}
3036 
3037 				if (!f2fs_cluster_can_merge_page(&cc,
3038 								page->index)) {
3039 					ret = f2fs_write_multi_pages(&cc,
3040 						&submitted, wbc, io_type);
3041 					if (!ret)
3042 						need_readd = true;
3043 					goto result;
3044 				}
3045 
3046 				if (unlikely(f2fs_cp_error(sbi)))
3047 					goto lock_page;
3048 
3049 				if (!f2fs_cluster_is_empty(&cc))
3050 					goto lock_page;
3051 
3052 				if (f2fs_all_cluster_page_ready(&cc,
3053 					pages, i, nr_pages, true))
3054 					goto lock_page;
3055 
3056 				ret2 = f2fs_prepare_compress_overwrite(
3057 							inode, &pagep,
3058 							page->index, &fsdata);
3059 				if (ret2 < 0) {
3060 					ret = ret2;
3061 					done = 1;
3062 					break;
3063 				} else if (ret2 &&
3064 					(!f2fs_compress_write_end(inode,
3065 						fsdata, page->index, 1) ||
3066 					 !f2fs_all_cluster_page_ready(&cc,
3067 						pages, i, nr_pages, false))) {
3068 					retry = 1;
3069 					break;
3070 				}
3071 			}
3072 #endif
3073 			/* give a priority to WB_SYNC threads */
3074 			if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3075 					wbc->sync_mode == WB_SYNC_NONE) {
3076 				done = 1;
3077 				break;
3078 			}
3079 #ifdef CONFIG_F2FS_FS_COMPRESSION
3080 lock_page:
3081 #endif
3082 			done_index = page->index;
3083 retry_write:
3084 			lock_page(page);
3085 
3086 			if (unlikely(page->mapping != mapping)) {
3087 continue_unlock:
3088 				unlock_page(page);
3089 				continue;
3090 			}
3091 
3092 			if (!PageDirty(page)) {
3093 				/* someone wrote it for us */
3094 				goto continue_unlock;
3095 			}
3096 
3097 			if (PageWriteback(page)) {
3098 				if (wbc->sync_mode != WB_SYNC_NONE)
3099 					f2fs_wait_on_page_writeback(page,
3100 							DATA, true, true);
3101 				else
3102 					goto continue_unlock;
3103 			}
3104 
3105 			if (!clear_page_dirty_for_io(page))
3106 				goto continue_unlock;
3107 
3108 #ifdef CONFIG_F2FS_FS_COMPRESSION
3109 			if (f2fs_compressed_file(inode)) {
3110 				get_page(page);
3111 				f2fs_compress_ctx_add_page(&cc, page);
3112 				continue;
3113 			}
3114 #endif
3115 			ret = f2fs_write_single_data_page(page, &submitted,
3116 					&bio, &last_block, wbc, io_type,
3117 					0, true);
3118 			if (ret == AOP_WRITEPAGE_ACTIVATE)
3119 				unlock_page(page);
3120 #ifdef CONFIG_F2FS_FS_COMPRESSION
3121 result:
3122 #endif
3123 			nwritten += submitted;
3124 			wbc->nr_to_write -= submitted;
3125 
3126 			if (unlikely(ret)) {
3127 				/*
3128 				 * keep nr_to_write, since vfs uses this to
3129 				 * get # of written pages.
3130 				 */
3131 				if (ret == AOP_WRITEPAGE_ACTIVATE) {
3132 					ret = 0;
3133 					goto next;
3134 				} else if (ret == -EAGAIN) {
3135 					ret = 0;
3136 					if (wbc->sync_mode == WB_SYNC_ALL) {
3137 						f2fs_io_schedule_timeout(
3138 							DEFAULT_IO_TIMEOUT);
3139 						goto retry_write;
3140 					}
3141 					goto next;
3142 				}
3143 				done_index = page->index + 1;
3144 				done = 1;
3145 				break;
3146 			}
3147 
3148 			if (wbc->nr_to_write <= 0 &&
3149 					wbc->sync_mode == WB_SYNC_NONE) {
3150 				done = 1;
3151 				break;
3152 			}
3153 next:
3154 			if (need_readd)
3155 				goto readd;
3156 		}
3157 		release_pages(pages, nr_pages);
3158 		cond_resched();
3159 	}
3160 #ifdef CONFIG_F2FS_FS_COMPRESSION
3161 	/* flush remained pages in compress cluster */
3162 	if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3163 		ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3164 		nwritten += submitted;
3165 		wbc->nr_to_write -= submitted;
3166 		if (ret) {
3167 			done = 1;
3168 			retry = 0;
3169 		}
3170 	}
3171 	if (f2fs_compressed_file(inode))
3172 		f2fs_destroy_compress_ctx(&cc, false);
3173 #endif
3174 	if (retry) {
3175 		index = 0;
3176 		end = -1;
3177 		goto retry;
3178 	}
3179 	if (wbc->range_cyclic && !done)
3180 		done_index = 0;
3181 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3182 		mapping->writeback_index = done_index;
3183 
3184 	if (nwritten)
3185 		f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3186 								NULL, 0, DATA);
3187 	/* submit cached bio of IPU write */
3188 	if (bio)
3189 		f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3190 
3191 	return ret;
3192 }
3193 
3194 static inline bool __should_serialize_io(struct inode *inode,
3195 					struct writeback_control *wbc)
3196 {
3197 	/* to avoid deadlock in path of data flush */
3198 	if (F2FS_I(inode)->wb_task)
3199 		return false;
3200 
3201 	if (!S_ISREG(inode->i_mode))
3202 		return false;
3203 	if (IS_NOQUOTA(inode))
3204 		return false;
3205 
3206 	if (f2fs_need_compress_data(inode))
3207 		return true;
3208 	if (wbc->sync_mode != WB_SYNC_ALL)
3209 		return true;
3210 	if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3211 		return true;
3212 	return false;
3213 }
3214 
3215 static int __f2fs_write_data_pages(struct address_space *mapping,
3216 						struct writeback_control *wbc,
3217 						enum iostat_type io_type)
3218 {
3219 	struct inode *inode = mapping->host;
3220 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3221 	struct blk_plug plug;
3222 	int ret;
3223 	bool locked = false;
3224 
3225 	/* deal with chardevs and other special file */
3226 	if (!mapping->a_ops->writepage)
3227 		return 0;
3228 
3229 	/* skip writing if there is no dirty page in this inode */
3230 	if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3231 		return 0;
3232 
3233 	/* during POR, we don't need to trigger writepage at all. */
3234 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3235 		goto skip_write;
3236 
3237 	if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3238 			wbc->sync_mode == WB_SYNC_NONE &&
3239 			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3240 			f2fs_available_free_memory(sbi, DIRTY_DENTS))
3241 		goto skip_write;
3242 
3243 	/* skip writing in file defragment preparing stage */
3244 	if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3245 		goto skip_write;
3246 
3247 	trace_f2fs_writepages(mapping->host, wbc, DATA);
3248 
3249 	/* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3250 	if (wbc->sync_mode == WB_SYNC_ALL)
3251 		atomic_inc(&sbi->wb_sync_req[DATA]);
3252 	else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3253 		/* to avoid potential deadlock */
3254 		if (current->plug)
3255 			blk_finish_plug(current->plug);
3256 		goto skip_write;
3257 	}
3258 
3259 	if (__should_serialize_io(inode, wbc)) {
3260 		mutex_lock(&sbi->writepages);
3261 		locked = true;
3262 	}
3263 
3264 	blk_start_plug(&plug);
3265 	ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3266 	blk_finish_plug(&plug);
3267 
3268 	if (locked)
3269 		mutex_unlock(&sbi->writepages);
3270 
3271 	if (wbc->sync_mode == WB_SYNC_ALL)
3272 		atomic_dec(&sbi->wb_sync_req[DATA]);
3273 	/*
3274 	 * if some pages were truncated, we cannot guarantee its mapping->host
3275 	 * to detect pending bios.
3276 	 */
3277 
3278 	f2fs_remove_dirty_inode(inode);
3279 	return ret;
3280 
3281 skip_write:
3282 	wbc->pages_skipped += get_dirty_pages(inode);
3283 	trace_f2fs_writepages(mapping->host, wbc, DATA);
3284 	return 0;
3285 }
3286 
3287 static int f2fs_write_data_pages(struct address_space *mapping,
3288 			    struct writeback_control *wbc)
3289 {
3290 	struct inode *inode = mapping->host;
3291 
3292 	return __f2fs_write_data_pages(mapping, wbc,
3293 			F2FS_I(inode)->cp_task == current ?
3294 			FS_CP_DATA_IO : FS_DATA_IO);
3295 }
3296 
3297 void f2fs_write_failed(struct inode *inode, loff_t to)
3298 {
3299 	loff_t i_size = i_size_read(inode);
3300 
3301 	if (IS_NOQUOTA(inode))
3302 		return;
3303 
3304 	/* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3305 	if (to > i_size && !f2fs_verity_in_progress(inode)) {
3306 		f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3307 		filemap_invalidate_lock(inode->i_mapping);
3308 
3309 		truncate_pagecache(inode, i_size);
3310 		f2fs_truncate_blocks(inode, i_size, true);
3311 
3312 		filemap_invalidate_unlock(inode->i_mapping);
3313 		f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3314 	}
3315 }
3316 
3317 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3318 			struct page *page, loff_t pos, unsigned len,
3319 			block_t *blk_addr, bool *node_changed)
3320 {
3321 	struct inode *inode = page->mapping->host;
3322 	pgoff_t index = page->index;
3323 	struct dnode_of_data dn;
3324 	struct page *ipage;
3325 	bool locked = false;
3326 	struct extent_info ei = {0, };
3327 	int err = 0;
3328 	int flag;
3329 
3330 	/*
3331 	 * If a whole page is being written and we already preallocated all the
3332 	 * blocks, then there is no need to get a block address now.
3333 	 */
3334 	if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3335 		return 0;
3336 
3337 	/* f2fs_lock_op avoids race between write CP and convert_inline_page */
3338 	if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
3339 		flag = F2FS_GET_BLOCK_DEFAULT;
3340 	else
3341 		flag = F2FS_GET_BLOCK_PRE_AIO;
3342 
3343 	if (f2fs_has_inline_data(inode) ||
3344 			(pos & PAGE_MASK) >= i_size_read(inode)) {
3345 		f2fs_do_map_lock(sbi, flag, true);
3346 		locked = true;
3347 	}
3348 
3349 restart:
3350 	/* check inline_data */
3351 	ipage = f2fs_get_node_page(sbi, inode->i_ino);
3352 	if (IS_ERR(ipage)) {
3353 		err = PTR_ERR(ipage);
3354 		goto unlock_out;
3355 	}
3356 
3357 	set_new_dnode(&dn, inode, ipage, ipage, 0);
3358 
3359 	if (f2fs_has_inline_data(inode)) {
3360 		if (pos + len <= MAX_INLINE_DATA(inode)) {
3361 			f2fs_do_read_inline_data(page, ipage);
3362 			set_inode_flag(inode, FI_DATA_EXIST);
3363 			if (inode->i_nlink)
3364 				set_page_private_inline(ipage);
3365 		} else {
3366 			err = f2fs_convert_inline_page(&dn, page);
3367 			if (err)
3368 				goto out;
3369 			if (dn.data_blkaddr == NULL_ADDR)
3370 				err = f2fs_get_block(&dn, index);
3371 		}
3372 	} else if (locked) {
3373 		err = f2fs_get_block(&dn, index);
3374 	} else {
3375 		if (f2fs_lookup_read_extent_cache(inode, index, &ei)) {
3376 			dn.data_blkaddr = ei.blk + index - ei.fofs;
3377 		} else {
3378 			/* hole case */
3379 			err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3380 			if (err || dn.data_blkaddr == NULL_ADDR) {
3381 				f2fs_put_dnode(&dn);
3382 				f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
3383 								true);
3384 				WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3385 				locked = true;
3386 				goto restart;
3387 			}
3388 		}
3389 	}
3390 
3391 	/* convert_inline_page can make node_changed */
3392 	*blk_addr = dn.data_blkaddr;
3393 	*node_changed = dn.node_changed;
3394 out:
3395 	f2fs_put_dnode(&dn);
3396 unlock_out:
3397 	if (locked)
3398 		f2fs_do_map_lock(sbi, flag, false);
3399 	return err;
3400 }
3401 
3402 static int __find_data_block(struct inode *inode, pgoff_t index,
3403 				block_t *blk_addr)
3404 {
3405 	struct dnode_of_data dn;
3406 	struct page *ipage;
3407 	struct extent_info ei = {0, };
3408 	int err = 0;
3409 
3410 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
3411 	if (IS_ERR(ipage))
3412 		return PTR_ERR(ipage);
3413 
3414 	set_new_dnode(&dn, inode, ipage, ipage, 0);
3415 
3416 	if (f2fs_lookup_read_extent_cache(inode, index, &ei)) {
3417 		dn.data_blkaddr = ei.blk + index - ei.fofs;
3418 	} else {
3419 		/* hole case */
3420 		err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3421 		if (err) {
3422 			dn.data_blkaddr = NULL_ADDR;
3423 			err = 0;
3424 		}
3425 	}
3426 	*blk_addr = dn.data_blkaddr;
3427 	f2fs_put_dnode(&dn);
3428 	return err;
3429 }
3430 
3431 static int __reserve_data_block(struct inode *inode, pgoff_t index,
3432 				block_t *blk_addr, bool *node_changed)
3433 {
3434 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3435 	struct dnode_of_data dn;
3436 	struct page *ipage;
3437 	int err = 0;
3438 
3439 	f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
3440 
3441 	ipage = f2fs_get_node_page(sbi, inode->i_ino);
3442 	if (IS_ERR(ipage)) {
3443 		err = PTR_ERR(ipage);
3444 		goto unlock_out;
3445 	}
3446 	set_new_dnode(&dn, inode, ipage, ipage, 0);
3447 
3448 	err = f2fs_get_block(&dn, index);
3449 
3450 	*blk_addr = dn.data_blkaddr;
3451 	*node_changed = dn.node_changed;
3452 	f2fs_put_dnode(&dn);
3453 
3454 unlock_out:
3455 	f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
3456 	return err;
3457 }
3458 
3459 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
3460 			struct page *page, loff_t pos, unsigned int len,
3461 			block_t *blk_addr, bool *node_changed)
3462 {
3463 	struct inode *inode = page->mapping->host;
3464 	struct inode *cow_inode = F2FS_I(inode)->cow_inode;
3465 	pgoff_t index = page->index;
3466 	int err = 0;
3467 	block_t ori_blk_addr = NULL_ADDR;
3468 
3469 	/* If pos is beyond the end of file, reserve a new block in COW inode */
3470 	if ((pos & PAGE_MASK) >= i_size_read(inode))
3471 		goto reserve_block;
3472 
3473 	/* Look for the block in COW inode first */
3474 	err = __find_data_block(cow_inode, index, blk_addr);
3475 	if (err)
3476 		return err;
3477 	else if (*blk_addr != NULL_ADDR)
3478 		return 0;
3479 
3480 	if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE))
3481 		goto reserve_block;
3482 
3483 	/* Look for the block in the original inode */
3484 	err = __find_data_block(inode, index, &ori_blk_addr);
3485 	if (err)
3486 		return err;
3487 
3488 reserve_block:
3489 	/* Finally, we should reserve a new block in COW inode for the update */
3490 	err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
3491 	if (err)
3492 		return err;
3493 	inc_atomic_write_cnt(inode);
3494 
3495 	if (ori_blk_addr != NULL_ADDR)
3496 		*blk_addr = ori_blk_addr;
3497 	return 0;
3498 }
3499 
3500 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3501 		loff_t pos, unsigned len, struct page **pagep, void **fsdata)
3502 {
3503 	struct inode *inode = mapping->host;
3504 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3505 	struct page *page = NULL;
3506 	pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3507 	bool need_balance = false;
3508 	block_t blkaddr = NULL_ADDR;
3509 	int err = 0;
3510 
3511 	trace_f2fs_write_begin(inode, pos, len);
3512 
3513 	if (!f2fs_is_checkpoint_ready(sbi)) {
3514 		err = -ENOSPC;
3515 		goto fail;
3516 	}
3517 
3518 	/*
3519 	 * We should check this at this moment to avoid deadlock on inode page
3520 	 * and #0 page. The locking rule for inline_data conversion should be:
3521 	 * lock_page(page #0) -> lock_page(inode_page)
3522 	 */
3523 	if (index != 0) {
3524 		err = f2fs_convert_inline_inode(inode);
3525 		if (err)
3526 			goto fail;
3527 	}
3528 
3529 #ifdef CONFIG_F2FS_FS_COMPRESSION
3530 	if (f2fs_compressed_file(inode)) {
3531 		int ret;
3532 
3533 		*fsdata = NULL;
3534 
3535 		if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3536 			goto repeat;
3537 
3538 		ret = f2fs_prepare_compress_overwrite(inode, pagep,
3539 							index, fsdata);
3540 		if (ret < 0) {
3541 			err = ret;
3542 			goto fail;
3543 		} else if (ret) {
3544 			return 0;
3545 		}
3546 	}
3547 #endif
3548 
3549 repeat:
3550 	/*
3551 	 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3552 	 * wait_for_stable_page. Will wait that below with our IO control.
3553 	 */
3554 	page = f2fs_pagecache_get_page(mapping, index,
3555 				FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3556 	if (!page) {
3557 		err = -ENOMEM;
3558 		goto fail;
3559 	}
3560 
3561 	/* TODO: cluster can be compressed due to race with .writepage */
3562 
3563 	*pagep = page;
3564 
3565 	if (f2fs_is_atomic_file(inode))
3566 		err = prepare_atomic_write_begin(sbi, page, pos, len,
3567 					&blkaddr, &need_balance);
3568 	else
3569 		err = prepare_write_begin(sbi, page, pos, len,
3570 					&blkaddr, &need_balance);
3571 	if (err)
3572 		goto fail;
3573 
3574 	if (need_balance && !IS_NOQUOTA(inode) &&
3575 			has_not_enough_free_secs(sbi, 0, 0)) {
3576 		unlock_page(page);
3577 		f2fs_balance_fs(sbi, true);
3578 		lock_page(page);
3579 		if (page->mapping != mapping) {
3580 			/* The page got truncated from under us */
3581 			f2fs_put_page(page, 1);
3582 			goto repeat;
3583 		}
3584 	}
3585 
3586 	f2fs_wait_on_page_writeback(page, DATA, false, true);
3587 
3588 	if (len == PAGE_SIZE || PageUptodate(page))
3589 		return 0;
3590 
3591 	if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3592 	    !f2fs_verity_in_progress(inode)) {
3593 		zero_user_segment(page, len, PAGE_SIZE);
3594 		return 0;
3595 	}
3596 
3597 	if (blkaddr == NEW_ADDR) {
3598 		zero_user_segment(page, 0, PAGE_SIZE);
3599 		SetPageUptodate(page);
3600 	} else {
3601 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3602 				DATA_GENERIC_ENHANCE_READ)) {
3603 			err = -EFSCORRUPTED;
3604 			f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
3605 			goto fail;
3606 		}
3607 		err = f2fs_submit_page_read(inode, page, blkaddr, 0, true);
3608 		if (err)
3609 			goto fail;
3610 
3611 		lock_page(page);
3612 		if (unlikely(page->mapping != mapping)) {
3613 			f2fs_put_page(page, 1);
3614 			goto repeat;
3615 		}
3616 		if (unlikely(!PageUptodate(page))) {
3617 			err = -EIO;
3618 			goto fail;
3619 		}
3620 	}
3621 	return 0;
3622 
3623 fail:
3624 	f2fs_put_page(page, 1);
3625 	f2fs_write_failed(inode, pos + len);
3626 	return err;
3627 }
3628 
3629 static int f2fs_write_end(struct file *file,
3630 			struct address_space *mapping,
3631 			loff_t pos, unsigned len, unsigned copied,
3632 			struct page *page, void *fsdata)
3633 {
3634 	struct inode *inode = page->mapping->host;
3635 
3636 	trace_f2fs_write_end(inode, pos, len, copied);
3637 
3638 	/*
3639 	 * This should be come from len == PAGE_SIZE, and we expect copied
3640 	 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3641 	 * let generic_perform_write() try to copy data again through copied=0.
3642 	 */
3643 	if (!PageUptodate(page)) {
3644 		if (unlikely(copied != len))
3645 			copied = 0;
3646 		else
3647 			SetPageUptodate(page);
3648 	}
3649 
3650 #ifdef CONFIG_F2FS_FS_COMPRESSION
3651 	/* overwrite compressed file */
3652 	if (f2fs_compressed_file(inode) && fsdata) {
3653 		f2fs_compress_write_end(inode, fsdata, page->index, copied);
3654 		f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3655 
3656 		if (pos + copied > i_size_read(inode) &&
3657 				!f2fs_verity_in_progress(inode))
3658 			f2fs_i_size_write(inode, pos + copied);
3659 		return copied;
3660 	}
3661 #endif
3662 
3663 	if (!copied)
3664 		goto unlock_out;
3665 
3666 	set_page_dirty(page);
3667 
3668 	if (pos + copied > i_size_read(inode) &&
3669 	    !f2fs_verity_in_progress(inode)) {
3670 		f2fs_i_size_write(inode, pos + copied);
3671 		if (f2fs_is_atomic_file(inode))
3672 			f2fs_i_size_write(F2FS_I(inode)->cow_inode,
3673 					pos + copied);
3674 	}
3675 unlock_out:
3676 	f2fs_put_page(page, 1);
3677 	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3678 	return copied;
3679 }
3680 
3681 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3682 {
3683 	struct inode *inode = folio->mapping->host;
3684 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3685 
3686 	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3687 				(offset || length != folio_size(folio)))
3688 		return;
3689 
3690 	if (folio_test_dirty(folio)) {
3691 		if (inode->i_ino == F2FS_META_INO(sbi)) {
3692 			dec_page_count(sbi, F2FS_DIRTY_META);
3693 		} else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3694 			dec_page_count(sbi, F2FS_DIRTY_NODES);
3695 		} else {
3696 			inode_dec_dirty_pages(inode);
3697 			f2fs_remove_dirty_inode(inode);
3698 		}
3699 	}
3700 
3701 	clear_page_private_gcing(&folio->page);
3702 
3703 	if (test_opt(sbi, COMPRESS_CACHE) &&
3704 			inode->i_ino == F2FS_COMPRESS_INO(sbi))
3705 		clear_page_private_data(&folio->page);
3706 
3707 	folio_detach_private(folio);
3708 }
3709 
3710 bool f2fs_release_folio(struct folio *folio, gfp_t wait)
3711 {
3712 	struct f2fs_sb_info *sbi;
3713 
3714 	/* If this is dirty folio, keep private data */
3715 	if (folio_test_dirty(folio))
3716 		return false;
3717 
3718 	sbi = F2FS_M_SB(folio->mapping);
3719 	if (test_opt(sbi, COMPRESS_CACHE)) {
3720 		struct inode *inode = folio->mapping->host;
3721 
3722 		if (inode->i_ino == F2FS_COMPRESS_INO(sbi))
3723 			clear_page_private_data(&folio->page);
3724 	}
3725 
3726 	clear_page_private_gcing(&folio->page);
3727 
3728 	folio_detach_private(folio);
3729 	return true;
3730 }
3731 
3732 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3733 		struct folio *folio)
3734 {
3735 	struct inode *inode = mapping->host;
3736 
3737 	trace_f2fs_set_page_dirty(&folio->page, DATA);
3738 
3739 	if (!folio_test_uptodate(folio))
3740 		folio_mark_uptodate(folio);
3741 	BUG_ON(folio_test_swapcache(folio));
3742 
3743 	if (filemap_dirty_folio(mapping, folio)) {
3744 		f2fs_update_dirty_folio(inode, folio);
3745 		return true;
3746 	}
3747 	return false;
3748 }
3749 
3750 
3751 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3752 {
3753 #ifdef CONFIG_F2FS_FS_COMPRESSION
3754 	struct dnode_of_data dn;
3755 	sector_t start_idx, blknr = 0;
3756 	int ret;
3757 
3758 	start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3759 
3760 	set_new_dnode(&dn, inode, NULL, NULL, 0);
3761 	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3762 	if (ret)
3763 		return 0;
3764 
3765 	if (dn.data_blkaddr != COMPRESS_ADDR) {
3766 		dn.ofs_in_node += block - start_idx;
3767 		blknr = f2fs_data_blkaddr(&dn);
3768 		if (!__is_valid_data_blkaddr(blknr))
3769 			blknr = 0;
3770 	}
3771 
3772 	f2fs_put_dnode(&dn);
3773 	return blknr;
3774 #else
3775 	return 0;
3776 #endif
3777 }
3778 
3779 
3780 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3781 {
3782 	struct inode *inode = mapping->host;
3783 	sector_t blknr = 0;
3784 
3785 	if (f2fs_has_inline_data(inode))
3786 		goto out;
3787 
3788 	/* make sure allocating whole blocks */
3789 	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3790 		filemap_write_and_wait(mapping);
3791 
3792 	/* Block number less than F2FS MAX BLOCKS */
3793 	if (unlikely(block >= max_file_blocks(inode)))
3794 		goto out;
3795 
3796 	if (f2fs_compressed_file(inode)) {
3797 		blknr = f2fs_bmap_compress(inode, block);
3798 	} else {
3799 		struct f2fs_map_blocks map;
3800 
3801 		memset(&map, 0, sizeof(map));
3802 		map.m_lblk = block;
3803 		map.m_len = 1;
3804 		map.m_next_pgofs = NULL;
3805 		map.m_seg_type = NO_CHECK_TYPE;
3806 
3807 		if (!f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_BMAP))
3808 			blknr = map.m_pblk;
3809 	}
3810 out:
3811 	trace_f2fs_bmap(inode, block, blknr);
3812 	return blknr;
3813 }
3814 
3815 #ifdef CONFIG_SWAP
3816 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3817 							unsigned int blkcnt)
3818 {
3819 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3820 	unsigned int blkofs;
3821 	unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3822 	unsigned int secidx = start_blk / blk_per_sec;
3823 	unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3824 	int ret = 0;
3825 
3826 	f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3827 	filemap_invalidate_lock(inode->i_mapping);
3828 
3829 	set_inode_flag(inode, FI_ALIGNED_WRITE);
3830 	set_inode_flag(inode, FI_OPU_WRITE);
3831 
3832 	for (; secidx < end_sec; secidx++) {
3833 		f2fs_down_write(&sbi->pin_sem);
3834 
3835 		f2fs_lock_op(sbi);
3836 		f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3837 		f2fs_unlock_op(sbi);
3838 
3839 		set_inode_flag(inode, FI_SKIP_WRITES);
3840 
3841 		for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3842 			struct page *page;
3843 			unsigned int blkidx = secidx * blk_per_sec + blkofs;
3844 
3845 			page = f2fs_get_lock_data_page(inode, blkidx, true);
3846 			if (IS_ERR(page)) {
3847 				f2fs_up_write(&sbi->pin_sem);
3848 				ret = PTR_ERR(page);
3849 				goto done;
3850 			}
3851 
3852 			set_page_dirty(page);
3853 			f2fs_put_page(page, 1);
3854 		}
3855 
3856 		clear_inode_flag(inode, FI_SKIP_WRITES);
3857 
3858 		ret = filemap_fdatawrite(inode->i_mapping);
3859 
3860 		f2fs_up_write(&sbi->pin_sem);
3861 
3862 		if (ret)
3863 			break;
3864 	}
3865 
3866 done:
3867 	clear_inode_flag(inode, FI_SKIP_WRITES);
3868 	clear_inode_flag(inode, FI_OPU_WRITE);
3869 	clear_inode_flag(inode, FI_ALIGNED_WRITE);
3870 
3871 	filemap_invalidate_unlock(inode->i_mapping);
3872 	f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3873 
3874 	return ret;
3875 }
3876 
3877 static int check_swap_activate(struct swap_info_struct *sis,
3878 				struct file *swap_file, sector_t *span)
3879 {
3880 	struct address_space *mapping = swap_file->f_mapping;
3881 	struct inode *inode = mapping->host;
3882 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3883 	sector_t cur_lblock;
3884 	sector_t last_lblock;
3885 	sector_t pblock;
3886 	sector_t lowest_pblock = -1;
3887 	sector_t highest_pblock = 0;
3888 	int nr_extents = 0;
3889 	unsigned long nr_pblocks;
3890 	unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3891 	unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3892 	unsigned int not_aligned = 0;
3893 	int ret = 0;
3894 
3895 	/*
3896 	 * Map all the blocks into the extent list.  This code doesn't try
3897 	 * to be very smart.
3898 	 */
3899 	cur_lblock = 0;
3900 	last_lblock = bytes_to_blks(inode, i_size_read(inode));
3901 
3902 	while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3903 		struct f2fs_map_blocks map;
3904 retry:
3905 		cond_resched();
3906 
3907 		memset(&map, 0, sizeof(map));
3908 		map.m_lblk = cur_lblock;
3909 		map.m_len = last_lblock - cur_lblock;
3910 		map.m_next_pgofs = NULL;
3911 		map.m_next_extent = NULL;
3912 		map.m_seg_type = NO_CHECK_TYPE;
3913 		map.m_may_create = false;
3914 
3915 		ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
3916 		if (ret)
3917 			goto out;
3918 
3919 		/* hole */
3920 		if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3921 			f2fs_err(sbi, "Swapfile has holes");
3922 			ret = -EINVAL;
3923 			goto out;
3924 		}
3925 
3926 		pblock = map.m_pblk;
3927 		nr_pblocks = map.m_len;
3928 
3929 		if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
3930 				nr_pblocks & sec_blks_mask) {
3931 			not_aligned++;
3932 
3933 			nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3934 			if (cur_lblock + nr_pblocks > sis->max)
3935 				nr_pblocks -= blks_per_sec;
3936 
3937 			if (!nr_pblocks) {
3938 				/* this extent is last one */
3939 				nr_pblocks = map.m_len;
3940 				f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
3941 				goto next;
3942 			}
3943 
3944 			ret = f2fs_migrate_blocks(inode, cur_lblock,
3945 							nr_pblocks);
3946 			if (ret)
3947 				goto out;
3948 			goto retry;
3949 		}
3950 next:
3951 		if (cur_lblock + nr_pblocks >= sis->max)
3952 			nr_pblocks = sis->max - cur_lblock;
3953 
3954 		if (cur_lblock) {	/* exclude the header page */
3955 			if (pblock < lowest_pblock)
3956 				lowest_pblock = pblock;
3957 			if (pblock + nr_pblocks - 1 > highest_pblock)
3958 				highest_pblock = pblock + nr_pblocks - 1;
3959 		}
3960 
3961 		/*
3962 		 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
3963 		 */
3964 		ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
3965 		if (ret < 0)
3966 			goto out;
3967 		nr_extents += ret;
3968 		cur_lblock += nr_pblocks;
3969 	}
3970 	ret = nr_extents;
3971 	*span = 1 + highest_pblock - lowest_pblock;
3972 	if (cur_lblock == 0)
3973 		cur_lblock = 1;	/* force Empty message */
3974 	sis->max = cur_lblock;
3975 	sis->pages = cur_lblock - 1;
3976 	sis->highest_bit = cur_lblock - 1;
3977 out:
3978 	if (not_aligned)
3979 		f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
3980 			  not_aligned, blks_per_sec * F2FS_BLKSIZE);
3981 	return ret;
3982 }
3983 
3984 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
3985 				sector_t *span)
3986 {
3987 	struct inode *inode = file_inode(file);
3988 	int ret;
3989 
3990 	if (!S_ISREG(inode->i_mode))
3991 		return -EINVAL;
3992 
3993 	if (f2fs_readonly(F2FS_I_SB(inode)->sb))
3994 		return -EROFS;
3995 
3996 	if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
3997 		f2fs_err(F2FS_I_SB(inode),
3998 			"Swapfile not supported in LFS mode");
3999 		return -EINVAL;
4000 	}
4001 
4002 	ret = f2fs_convert_inline_inode(inode);
4003 	if (ret)
4004 		return ret;
4005 
4006 	if (!f2fs_disable_compressed_file(inode))
4007 		return -EINVAL;
4008 
4009 	f2fs_precache_extents(inode);
4010 
4011 	ret = check_swap_activate(sis, file, span);
4012 	if (ret < 0)
4013 		return ret;
4014 
4015 	stat_inc_swapfile_inode(inode);
4016 	set_inode_flag(inode, FI_PIN_FILE);
4017 	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4018 	return ret;
4019 }
4020 
4021 static void f2fs_swap_deactivate(struct file *file)
4022 {
4023 	struct inode *inode = file_inode(file);
4024 
4025 	stat_dec_swapfile_inode(inode);
4026 	clear_inode_flag(inode, FI_PIN_FILE);
4027 }
4028 #else
4029 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4030 				sector_t *span)
4031 {
4032 	return -EOPNOTSUPP;
4033 }
4034 
4035 static void f2fs_swap_deactivate(struct file *file)
4036 {
4037 }
4038 #endif
4039 
4040 const struct address_space_operations f2fs_dblock_aops = {
4041 	.read_folio	= f2fs_read_data_folio,
4042 	.readahead	= f2fs_readahead,
4043 	.writepage	= f2fs_write_data_page,
4044 	.writepages	= f2fs_write_data_pages,
4045 	.write_begin	= f2fs_write_begin,
4046 	.write_end	= f2fs_write_end,
4047 	.dirty_folio	= f2fs_dirty_data_folio,
4048 	.migrate_folio	= filemap_migrate_folio,
4049 	.invalidate_folio = f2fs_invalidate_folio,
4050 	.release_folio	= f2fs_release_folio,
4051 	.direct_IO	= noop_direct_IO,
4052 	.bmap		= f2fs_bmap,
4053 	.swap_activate  = f2fs_swap_activate,
4054 	.swap_deactivate = f2fs_swap_deactivate,
4055 };
4056 
4057 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4058 {
4059 	struct address_space *mapping = page_mapping(page);
4060 	unsigned long flags;
4061 
4062 	xa_lock_irqsave(&mapping->i_pages, flags);
4063 	__xa_clear_mark(&mapping->i_pages, page_index(page),
4064 						PAGECACHE_TAG_DIRTY);
4065 	xa_unlock_irqrestore(&mapping->i_pages, flags);
4066 }
4067 
4068 int __init f2fs_init_post_read_processing(void)
4069 {
4070 	bio_post_read_ctx_cache =
4071 		kmem_cache_create("f2fs_bio_post_read_ctx",
4072 				  sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4073 	if (!bio_post_read_ctx_cache)
4074 		goto fail;
4075 	bio_post_read_ctx_pool =
4076 		mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4077 					 bio_post_read_ctx_cache);
4078 	if (!bio_post_read_ctx_pool)
4079 		goto fail_free_cache;
4080 	return 0;
4081 
4082 fail_free_cache:
4083 	kmem_cache_destroy(bio_post_read_ctx_cache);
4084 fail:
4085 	return -ENOMEM;
4086 }
4087 
4088 void f2fs_destroy_post_read_processing(void)
4089 {
4090 	mempool_destroy(bio_post_read_ctx_pool);
4091 	kmem_cache_destroy(bio_post_read_ctx_cache);
4092 }
4093 
4094 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4095 {
4096 	if (!f2fs_sb_has_encrypt(sbi) &&
4097 		!f2fs_sb_has_verity(sbi) &&
4098 		!f2fs_sb_has_compression(sbi))
4099 		return 0;
4100 
4101 	sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4102 						 WQ_UNBOUND | WQ_HIGHPRI,
4103 						 num_online_cpus());
4104 	return sbi->post_read_wq ? 0 : -ENOMEM;
4105 }
4106 
4107 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4108 {
4109 	if (sbi->post_read_wq)
4110 		destroy_workqueue(sbi->post_read_wq);
4111 }
4112 
4113 int __init f2fs_init_bio_entry_cache(void)
4114 {
4115 	bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4116 			sizeof(struct bio_entry));
4117 	return bio_entry_slab ? 0 : -ENOMEM;
4118 }
4119 
4120 void f2fs_destroy_bio_entry_cache(void)
4121 {
4122 	kmem_cache_destroy(bio_entry_slab);
4123 }
4124 
4125 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4126 			    unsigned int flags, struct iomap *iomap,
4127 			    struct iomap *srcmap)
4128 {
4129 	struct f2fs_map_blocks map = {};
4130 	pgoff_t next_pgofs = 0;
4131 	int err;
4132 
4133 	map.m_lblk = bytes_to_blks(inode, offset);
4134 	map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4135 	map.m_next_pgofs = &next_pgofs;
4136 	map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4137 	if (flags & IOMAP_WRITE)
4138 		map.m_may_create = true;
4139 
4140 	err = f2fs_map_blocks(inode, &map, flags & IOMAP_WRITE,
4141 			      F2FS_GET_BLOCK_DIO);
4142 	if (err)
4143 		return err;
4144 
4145 	iomap->offset = blks_to_bytes(inode, map.m_lblk);
4146 
4147 	/*
4148 	 * When inline encryption is enabled, sometimes I/O to an encrypted file
4149 	 * has to be broken up to guarantee DUN contiguity.  Handle this by
4150 	 * limiting the length of the mapping returned.
4151 	 */
4152 	map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4153 
4154 	if (map.m_flags & (F2FS_MAP_MAPPED | F2FS_MAP_UNWRITTEN)) {
4155 		iomap->length = blks_to_bytes(inode, map.m_len);
4156 		if (map.m_flags & F2FS_MAP_MAPPED) {
4157 			iomap->type = IOMAP_MAPPED;
4158 			iomap->flags |= IOMAP_F_MERGED;
4159 		} else {
4160 			iomap->type = IOMAP_UNWRITTEN;
4161 		}
4162 		if (WARN_ON_ONCE(!__is_valid_data_blkaddr(map.m_pblk)))
4163 			return -EINVAL;
4164 
4165 		iomap->bdev = map.m_bdev;
4166 		iomap->addr = blks_to_bytes(inode, map.m_pblk);
4167 	} else {
4168 		iomap->length = blks_to_bytes(inode, next_pgofs) -
4169 				iomap->offset;
4170 		iomap->type = IOMAP_HOLE;
4171 		iomap->addr = IOMAP_NULL_ADDR;
4172 	}
4173 
4174 	if (map.m_flags & F2FS_MAP_NEW)
4175 		iomap->flags |= IOMAP_F_NEW;
4176 	if ((inode->i_state & I_DIRTY_DATASYNC) ||
4177 	    offset + length > i_size_read(inode))
4178 		iomap->flags |= IOMAP_F_DIRTY;
4179 
4180 	return 0;
4181 }
4182 
4183 const struct iomap_ops f2fs_iomap_ops = {
4184 	.iomap_begin	= f2fs_iomap_begin,
4185 };
4186