xref: /linux/fs/f2fs/data.c (revision 037fc3368be46dc1a2a90f6e50c8cbce49d75fd6)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/data.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/buffer_head.h>
11 #include <linux/mpage.h>
12 #include <linux/writeback.h>
13 #include <linux/backing-dev.h>
14 #include <linux/pagevec.h>
15 #include <linux/blkdev.h>
16 #include <linux/bio.h>
17 #include <linux/prefetch.h>
18 #include <linux/uio.h>
19 #include <linux/cleancache.h>
20 #include <linux/sched/signal.h>
21 
22 #include "f2fs.h"
23 #include "node.h"
24 #include "segment.h"
25 #include "trace.h"
26 #include <trace/events/f2fs.h>
27 
28 #define NUM_PREALLOC_POST_READ_CTXS	128
29 
30 static struct kmem_cache *bio_post_read_ctx_cache;
31 static mempool_t *bio_post_read_ctx_pool;
32 
33 static bool __is_cp_guaranteed(struct page *page)
34 {
35 	struct address_space *mapping = page->mapping;
36 	struct inode *inode;
37 	struct f2fs_sb_info *sbi;
38 
39 	if (!mapping)
40 		return false;
41 
42 	inode = mapping->host;
43 	sbi = F2FS_I_SB(inode);
44 
45 	if (inode->i_ino == F2FS_META_INO(sbi) ||
46 			inode->i_ino ==  F2FS_NODE_INO(sbi) ||
47 			S_ISDIR(inode->i_mode) ||
48 			(S_ISREG(inode->i_mode) &&
49 			(f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
50 			is_cold_data(page))
51 		return true;
52 	return false;
53 }
54 
55 static enum count_type __read_io_type(struct page *page)
56 {
57 	struct address_space *mapping = page->mapping;
58 
59 	if (mapping) {
60 		struct inode *inode = mapping->host;
61 		struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
62 
63 		if (inode->i_ino == F2FS_META_INO(sbi))
64 			return F2FS_RD_META;
65 
66 		if (inode->i_ino == F2FS_NODE_INO(sbi))
67 			return F2FS_RD_NODE;
68 	}
69 	return F2FS_RD_DATA;
70 }
71 
72 /* postprocessing steps for read bios */
73 enum bio_post_read_step {
74 	STEP_INITIAL = 0,
75 	STEP_DECRYPT,
76 };
77 
78 struct bio_post_read_ctx {
79 	struct bio *bio;
80 	struct work_struct work;
81 	unsigned int cur_step;
82 	unsigned int enabled_steps;
83 };
84 
85 static void __read_end_io(struct bio *bio)
86 {
87 	struct page *page;
88 	struct bio_vec *bv;
89 	int i;
90 	struct bvec_iter_all iter_all;
91 
92 	bio_for_each_segment_all(bv, bio, i, iter_all) {
93 		page = bv->bv_page;
94 
95 		/* PG_error was set if any post_read step failed */
96 		if (bio->bi_status || PageError(page)) {
97 			ClearPageUptodate(page);
98 			/* will re-read again later */
99 			ClearPageError(page);
100 		} else {
101 			SetPageUptodate(page);
102 		}
103 		dec_page_count(F2FS_P_SB(page), __read_io_type(page));
104 		unlock_page(page);
105 	}
106 	if (bio->bi_private)
107 		mempool_free(bio->bi_private, bio_post_read_ctx_pool);
108 	bio_put(bio);
109 }
110 
111 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
112 
113 static void decrypt_work(struct work_struct *work)
114 {
115 	struct bio_post_read_ctx *ctx =
116 		container_of(work, struct bio_post_read_ctx, work);
117 
118 	fscrypt_decrypt_bio(ctx->bio);
119 
120 	bio_post_read_processing(ctx);
121 }
122 
123 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
124 {
125 	switch (++ctx->cur_step) {
126 	case STEP_DECRYPT:
127 		if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
128 			INIT_WORK(&ctx->work, decrypt_work);
129 			fscrypt_enqueue_decrypt_work(&ctx->work);
130 			return;
131 		}
132 		ctx->cur_step++;
133 		/* fall-through */
134 	default:
135 		__read_end_io(ctx->bio);
136 	}
137 }
138 
139 static bool f2fs_bio_post_read_required(struct bio *bio)
140 {
141 	return bio->bi_private && !bio->bi_status;
142 }
143 
144 static void f2fs_read_end_io(struct bio *bio)
145 {
146 	if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)),
147 						FAULT_READ_IO)) {
148 		f2fs_show_injection_info(FAULT_READ_IO);
149 		bio->bi_status = BLK_STS_IOERR;
150 	}
151 
152 	if (f2fs_bio_post_read_required(bio)) {
153 		struct bio_post_read_ctx *ctx = bio->bi_private;
154 
155 		ctx->cur_step = STEP_INITIAL;
156 		bio_post_read_processing(ctx);
157 		return;
158 	}
159 
160 	__read_end_io(bio);
161 }
162 
163 static void f2fs_write_end_io(struct bio *bio)
164 {
165 	struct f2fs_sb_info *sbi = bio->bi_private;
166 	struct bio_vec *bvec;
167 	int i;
168 	struct bvec_iter_all iter_all;
169 
170 	if (time_to_inject(sbi, FAULT_WRITE_IO)) {
171 		f2fs_show_injection_info(FAULT_WRITE_IO);
172 		bio->bi_status = BLK_STS_IOERR;
173 	}
174 
175 	bio_for_each_segment_all(bvec, bio, i, iter_all) {
176 		struct page *page = bvec->bv_page;
177 		enum count_type type = WB_DATA_TYPE(page);
178 
179 		if (IS_DUMMY_WRITTEN_PAGE(page)) {
180 			set_page_private(page, (unsigned long)NULL);
181 			ClearPagePrivate(page);
182 			unlock_page(page);
183 			mempool_free(page, sbi->write_io_dummy);
184 
185 			if (unlikely(bio->bi_status))
186 				f2fs_stop_checkpoint(sbi, true);
187 			continue;
188 		}
189 
190 		fscrypt_pullback_bio_page(&page, true);
191 
192 		if (unlikely(bio->bi_status)) {
193 			mapping_set_error(page->mapping, -EIO);
194 			if (type == F2FS_WB_CP_DATA)
195 				f2fs_stop_checkpoint(sbi, true);
196 		}
197 
198 		f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
199 					page->index != nid_of_node(page));
200 
201 		dec_page_count(sbi, type);
202 		if (f2fs_in_warm_node_list(sbi, page))
203 			f2fs_del_fsync_node_entry(sbi, page);
204 		clear_cold_data(page);
205 		end_page_writeback(page);
206 	}
207 	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
208 				wq_has_sleeper(&sbi->cp_wait))
209 		wake_up(&sbi->cp_wait);
210 
211 	bio_put(bio);
212 }
213 
214 /*
215  * Return true, if pre_bio's bdev is same as its target device.
216  */
217 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
218 				block_t blk_addr, struct bio *bio)
219 {
220 	struct block_device *bdev = sbi->sb->s_bdev;
221 	int i;
222 
223 	for (i = 0; i < sbi->s_ndevs; i++) {
224 		if (FDEV(i).start_blk <= blk_addr &&
225 					FDEV(i).end_blk >= blk_addr) {
226 			blk_addr -= FDEV(i).start_blk;
227 			bdev = FDEV(i).bdev;
228 			break;
229 		}
230 	}
231 	if (bio) {
232 		bio_set_dev(bio, bdev);
233 		bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
234 	}
235 	return bdev;
236 }
237 
238 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
239 {
240 	int i;
241 
242 	for (i = 0; i < sbi->s_ndevs; i++)
243 		if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
244 			return i;
245 	return 0;
246 }
247 
248 static bool __same_bdev(struct f2fs_sb_info *sbi,
249 				block_t blk_addr, struct bio *bio)
250 {
251 	struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
252 	return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
253 }
254 
255 /*
256  * Low-level block read/write IO operations.
257  */
258 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
259 				struct writeback_control *wbc,
260 				int npages, bool is_read,
261 				enum page_type type, enum temp_type temp)
262 {
263 	struct bio *bio;
264 
265 	bio = f2fs_bio_alloc(sbi, npages, true);
266 
267 	f2fs_target_device(sbi, blk_addr, bio);
268 	if (is_read) {
269 		bio->bi_end_io = f2fs_read_end_io;
270 		bio->bi_private = NULL;
271 	} else {
272 		bio->bi_end_io = f2fs_write_end_io;
273 		bio->bi_private = sbi;
274 		bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
275 	}
276 	if (wbc)
277 		wbc_init_bio(wbc, bio);
278 
279 	return bio;
280 }
281 
282 static inline void __submit_bio(struct f2fs_sb_info *sbi,
283 				struct bio *bio, enum page_type type)
284 {
285 	if (!is_read_io(bio_op(bio))) {
286 		unsigned int start;
287 
288 		if (type != DATA && type != NODE)
289 			goto submit_io;
290 
291 		if (test_opt(sbi, LFS) && current->plug)
292 			blk_finish_plug(current->plug);
293 
294 		start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
295 		start %= F2FS_IO_SIZE(sbi);
296 
297 		if (start == 0)
298 			goto submit_io;
299 
300 		/* fill dummy pages */
301 		for (; start < F2FS_IO_SIZE(sbi); start++) {
302 			struct page *page =
303 				mempool_alloc(sbi->write_io_dummy,
304 					GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
305 			f2fs_bug_on(sbi, !page);
306 
307 			SetPagePrivate(page);
308 			set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
309 			lock_page(page);
310 			if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
311 				f2fs_bug_on(sbi, 1);
312 		}
313 		/*
314 		 * In the NODE case, we lose next block address chain. So, we
315 		 * need to do checkpoint in f2fs_sync_file.
316 		 */
317 		if (type == NODE)
318 			set_sbi_flag(sbi, SBI_NEED_CP);
319 	}
320 submit_io:
321 	if (is_read_io(bio_op(bio)))
322 		trace_f2fs_submit_read_bio(sbi->sb, type, bio);
323 	else
324 		trace_f2fs_submit_write_bio(sbi->sb, type, bio);
325 	submit_bio(bio);
326 }
327 
328 static void __submit_merged_bio(struct f2fs_bio_info *io)
329 {
330 	struct f2fs_io_info *fio = &io->fio;
331 
332 	if (!io->bio)
333 		return;
334 
335 	bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
336 
337 	if (is_read_io(fio->op))
338 		trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
339 	else
340 		trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
341 
342 	__submit_bio(io->sbi, io->bio, fio->type);
343 	io->bio = NULL;
344 }
345 
346 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
347 						struct page *page, nid_t ino)
348 {
349 	struct bio_vec *bvec;
350 	struct page *target;
351 	int i;
352 	struct bvec_iter_all iter_all;
353 
354 	if (!io->bio)
355 		return false;
356 
357 	if (!inode && !page && !ino)
358 		return true;
359 
360 	bio_for_each_segment_all(bvec, io->bio, i, iter_all) {
361 
362 		if (bvec->bv_page->mapping)
363 			target = bvec->bv_page;
364 		else
365 			target = fscrypt_control_page(bvec->bv_page);
366 
367 		if (inode && inode == target->mapping->host)
368 			return true;
369 		if (page && page == target)
370 			return true;
371 		if (ino && ino == ino_of_node(target))
372 			return true;
373 	}
374 
375 	return false;
376 }
377 
378 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
379 				enum page_type type, enum temp_type temp)
380 {
381 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
382 	struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
383 
384 	down_write(&io->io_rwsem);
385 
386 	/* change META to META_FLUSH in the checkpoint procedure */
387 	if (type >= META_FLUSH) {
388 		io->fio.type = META_FLUSH;
389 		io->fio.op = REQ_OP_WRITE;
390 		io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
391 		if (!test_opt(sbi, NOBARRIER))
392 			io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
393 	}
394 	__submit_merged_bio(io);
395 	up_write(&io->io_rwsem);
396 }
397 
398 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
399 				struct inode *inode, struct page *page,
400 				nid_t ino, enum page_type type, bool force)
401 {
402 	enum temp_type temp;
403 	bool ret = true;
404 
405 	for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
406 		if (!force)	{
407 			enum page_type btype = PAGE_TYPE_OF_BIO(type);
408 			struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
409 
410 			down_read(&io->io_rwsem);
411 			ret = __has_merged_page(io, inode, page, ino);
412 			up_read(&io->io_rwsem);
413 		}
414 		if (ret)
415 			__f2fs_submit_merged_write(sbi, type, temp);
416 
417 		/* TODO: use HOT temp only for meta pages now. */
418 		if (type >= META)
419 			break;
420 	}
421 }
422 
423 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
424 {
425 	__submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
426 }
427 
428 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
429 				struct inode *inode, struct page *page,
430 				nid_t ino, enum page_type type)
431 {
432 	__submit_merged_write_cond(sbi, inode, page, ino, type, false);
433 }
434 
435 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
436 {
437 	f2fs_submit_merged_write(sbi, DATA);
438 	f2fs_submit_merged_write(sbi, NODE);
439 	f2fs_submit_merged_write(sbi, META);
440 }
441 
442 /*
443  * Fill the locked page with data located in the block address.
444  * A caller needs to unlock the page on failure.
445  */
446 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
447 {
448 	struct bio *bio;
449 	struct page *page = fio->encrypted_page ?
450 			fio->encrypted_page : fio->page;
451 
452 	if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
453 			__is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
454 		return -EFAULT;
455 
456 	trace_f2fs_submit_page_bio(page, fio);
457 	f2fs_trace_ios(fio, 0);
458 
459 	/* Allocate a new bio */
460 	bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
461 				1, is_read_io(fio->op), fio->type, fio->temp);
462 
463 	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
464 		bio_put(bio);
465 		return -EFAULT;
466 	}
467 
468 	if (fio->io_wbc && !is_read_io(fio->op))
469 		wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
470 
471 	bio_set_op_attrs(bio, fio->op, fio->op_flags);
472 
473 	inc_page_count(fio->sbi, is_read_io(fio->op) ?
474 			__read_io_type(page): WB_DATA_TYPE(fio->page));
475 
476 	__submit_bio(fio->sbi, bio, fio->type);
477 	return 0;
478 }
479 
480 void f2fs_submit_page_write(struct f2fs_io_info *fio)
481 {
482 	struct f2fs_sb_info *sbi = fio->sbi;
483 	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
484 	struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
485 	struct page *bio_page;
486 
487 	f2fs_bug_on(sbi, is_read_io(fio->op));
488 
489 	down_write(&io->io_rwsem);
490 next:
491 	if (fio->in_list) {
492 		spin_lock(&io->io_lock);
493 		if (list_empty(&io->io_list)) {
494 			spin_unlock(&io->io_lock);
495 			goto out;
496 		}
497 		fio = list_first_entry(&io->io_list,
498 						struct f2fs_io_info, list);
499 		list_del(&fio->list);
500 		spin_unlock(&io->io_lock);
501 	}
502 
503 	if (__is_valid_data_blkaddr(fio->old_blkaddr))
504 		verify_block_addr(fio, fio->old_blkaddr);
505 	verify_block_addr(fio, fio->new_blkaddr);
506 
507 	bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
508 
509 	/* set submitted = true as a return value */
510 	fio->submitted = true;
511 
512 	inc_page_count(sbi, WB_DATA_TYPE(bio_page));
513 
514 	if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
515 	    (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
516 			!__same_bdev(sbi, fio->new_blkaddr, io->bio)))
517 		__submit_merged_bio(io);
518 alloc_new:
519 	if (io->bio == NULL) {
520 		if ((fio->type == DATA || fio->type == NODE) &&
521 				fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
522 			dec_page_count(sbi, WB_DATA_TYPE(bio_page));
523 			fio->retry = true;
524 			goto skip;
525 		}
526 		io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
527 						BIO_MAX_PAGES, false,
528 						fio->type, fio->temp);
529 		io->fio = *fio;
530 	}
531 
532 	if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
533 		__submit_merged_bio(io);
534 		goto alloc_new;
535 	}
536 
537 	if (fio->io_wbc)
538 		wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
539 
540 	io->last_block_in_bio = fio->new_blkaddr;
541 	f2fs_trace_ios(fio, 0);
542 
543 	trace_f2fs_submit_page_write(fio->page, fio);
544 skip:
545 	if (fio->in_list)
546 		goto next;
547 out:
548 	if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
549 				f2fs_is_checkpoint_ready(sbi))
550 		__submit_merged_bio(io);
551 	up_write(&io->io_rwsem);
552 }
553 
554 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
555 					unsigned nr_pages, unsigned op_flag)
556 {
557 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
558 	struct bio *bio;
559 	struct bio_post_read_ctx *ctx;
560 	unsigned int post_read_steps = 0;
561 
562 	if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
563 		return ERR_PTR(-EFAULT);
564 
565 	bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
566 	if (!bio)
567 		return ERR_PTR(-ENOMEM);
568 	f2fs_target_device(sbi, blkaddr, bio);
569 	bio->bi_end_io = f2fs_read_end_io;
570 	bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
571 
572 	if (f2fs_encrypted_file(inode))
573 		post_read_steps |= 1 << STEP_DECRYPT;
574 	if (post_read_steps) {
575 		ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
576 		if (!ctx) {
577 			bio_put(bio);
578 			return ERR_PTR(-ENOMEM);
579 		}
580 		ctx->bio = bio;
581 		ctx->enabled_steps = post_read_steps;
582 		bio->bi_private = ctx;
583 	}
584 
585 	return bio;
586 }
587 
588 /* This can handle encryption stuffs */
589 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
590 							block_t blkaddr)
591 {
592 	struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
593 
594 	if (IS_ERR(bio))
595 		return PTR_ERR(bio);
596 
597 	/* wait for GCed page writeback via META_MAPPING */
598 	f2fs_wait_on_block_writeback(inode, blkaddr);
599 
600 	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
601 		bio_put(bio);
602 		return -EFAULT;
603 	}
604 	ClearPageError(page);
605 	inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
606 	__submit_bio(F2FS_I_SB(inode), bio, DATA);
607 	return 0;
608 }
609 
610 static void __set_data_blkaddr(struct dnode_of_data *dn)
611 {
612 	struct f2fs_node *rn = F2FS_NODE(dn->node_page);
613 	__le32 *addr_array;
614 	int base = 0;
615 
616 	if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
617 		base = get_extra_isize(dn->inode);
618 
619 	/* Get physical address of data block */
620 	addr_array = blkaddr_in_node(rn);
621 	addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
622 }
623 
624 /*
625  * Lock ordering for the change of data block address:
626  * ->data_page
627  *  ->node_page
628  *    update block addresses in the node page
629  */
630 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
631 {
632 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
633 	__set_data_blkaddr(dn);
634 	if (set_page_dirty(dn->node_page))
635 		dn->node_changed = true;
636 }
637 
638 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
639 {
640 	dn->data_blkaddr = blkaddr;
641 	f2fs_set_data_blkaddr(dn);
642 	f2fs_update_extent_cache(dn);
643 }
644 
645 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
646 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
647 {
648 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
649 	int err;
650 
651 	if (!count)
652 		return 0;
653 
654 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
655 		return -EPERM;
656 	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
657 		return err;
658 
659 	trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
660 						dn->ofs_in_node, count);
661 
662 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
663 
664 	for (; count > 0; dn->ofs_in_node++) {
665 		block_t blkaddr = datablock_addr(dn->inode,
666 					dn->node_page, dn->ofs_in_node);
667 		if (blkaddr == NULL_ADDR) {
668 			dn->data_blkaddr = NEW_ADDR;
669 			__set_data_blkaddr(dn);
670 			count--;
671 		}
672 	}
673 
674 	if (set_page_dirty(dn->node_page))
675 		dn->node_changed = true;
676 	return 0;
677 }
678 
679 /* Should keep dn->ofs_in_node unchanged */
680 int f2fs_reserve_new_block(struct dnode_of_data *dn)
681 {
682 	unsigned int ofs_in_node = dn->ofs_in_node;
683 	int ret;
684 
685 	ret = f2fs_reserve_new_blocks(dn, 1);
686 	dn->ofs_in_node = ofs_in_node;
687 	return ret;
688 }
689 
690 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
691 {
692 	bool need_put = dn->inode_page ? false : true;
693 	int err;
694 
695 	err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
696 	if (err)
697 		return err;
698 
699 	if (dn->data_blkaddr == NULL_ADDR)
700 		err = f2fs_reserve_new_block(dn);
701 	if (err || need_put)
702 		f2fs_put_dnode(dn);
703 	return err;
704 }
705 
706 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
707 {
708 	struct extent_info ei  = {0,0,0};
709 	struct inode *inode = dn->inode;
710 
711 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
712 		dn->data_blkaddr = ei.blk + index - ei.fofs;
713 		return 0;
714 	}
715 
716 	return f2fs_reserve_block(dn, index);
717 }
718 
719 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
720 						int op_flags, bool for_write)
721 {
722 	struct address_space *mapping = inode->i_mapping;
723 	struct dnode_of_data dn;
724 	struct page *page;
725 	struct extent_info ei = {0,0,0};
726 	int err;
727 
728 	page = f2fs_grab_cache_page(mapping, index, for_write);
729 	if (!page)
730 		return ERR_PTR(-ENOMEM);
731 
732 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
733 		dn.data_blkaddr = ei.blk + index - ei.fofs;
734 		goto got_it;
735 	}
736 
737 	set_new_dnode(&dn, inode, NULL, NULL, 0);
738 	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
739 	if (err)
740 		goto put_err;
741 	f2fs_put_dnode(&dn);
742 
743 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
744 		err = -ENOENT;
745 		goto put_err;
746 	}
747 got_it:
748 	if (PageUptodate(page)) {
749 		unlock_page(page);
750 		return page;
751 	}
752 
753 	/*
754 	 * A new dentry page is allocated but not able to be written, since its
755 	 * new inode page couldn't be allocated due to -ENOSPC.
756 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
757 	 * see, f2fs_add_link -> f2fs_get_new_data_page ->
758 	 * f2fs_init_inode_metadata.
759 	 */
760 	if (dn.data_blkaddr == NEW_ADDR) {
761 		zero_user_segment(page, 0, PAGE_SIZE);
762 		if (!PageUptodate(page))
763 			SetPageUptodate(page);
764 		unlock_page(page);
765 		return page;
766 	}
767 
768 	err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
769 	if (err)
770 		goto put_err;
771 	return page;
772 
773 put_err:
774 	f2fs_put_page(page, 1);
775 	return ERR_PTR(err);
776 }
777 
778 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
779 {
780 	struct address_space *mapping = inode->i_mapping;
781 	struct page *page;
782 
783 	page = find_get_page(mapping, index);
784 	if (page && PageUptodate(page))
785 		return page;
786 	f2fs_put_page(page, 0);
787 
788 	page = f2fs_get_read_data_page(inode, index, 0, false);
789 	if (IS_ERR(page))
790 		return page;
791 
792 	if (PageUptodate(page))
793 		return page;
794 
795 	wait_on_page_locked(page);
796 	if (unlikely(!PageUptodate(page))) {
797 		f2fs_put_page(page, 0);
798 		return ERR_PTR(-EIO);
799 	}
800 	return page;
801 }
802 
803 /*
804  * If it tries to access a hole, return an error.
805  * Because, the callers, functions in dir.c and GC, should be able to know
806  * whether this page exists or not.
807  */
808 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
809 							bool for_write)
810 {
811 	struct address_space *mapping = inode->i_mapping;
812 	struct page *page;
813 repeat:
814 	page = f2fs_get_read_data_page(inode, index, 0, for_write);
815 	if (IS_ERR(page))
816 		return page;
817 
818 	/* wait for read completion */
819 	lock_page(page);
820 	if (unlikely(page->mapping != mapping)) {
821 		f2fs_put_page(page, 1);
822 		goto repeat;
823 	}
824 	if (unlikely(!PageUptodate(page))) {
825 		f2fs_put_page(page, 1);
826 		return ERR_PTR(-EIO);
827 	}
828 	return page;
829 }
830 
831 /*
832  * Caller ensures that this data page is never allocated.
833  * A new zero-filled data page is allocated in the page cache.
834  *
835  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
836  * f2fs_unlock_op().
837  * Note that, ipage is set only by make_empty_dir, and if any error occur,
838  * ipage should be released by this function.
839  */
840 struct page *f2fs_get_new_data_page(struct inode *inode,
841 		struct page *ipage, pgoff_t index, bool new_i_size)
842 {
843 	struct address_space *mapping = inode->i_mapping;
844 	struct page *page;
845 	struct dnode_of_data dn;
846 	int err;
847 
848 	page = f2fs_grab_cache_page(mapping, index, true);
849 	if (!page) {
850 		/*
851 		 * before exiting, we should make sure ipage will be released
852 		 * if any error occur.
853 		 */
854 		f2fs_put_page(ipage, 1);
855 		return ERR_PTR(-ENOMEM);
856 	}
857 
858 	set_new_dnode(&dn, inode, ipage, NULL, 0);
859 	err = f2fs_reserve_block(&dn, index);
860 	if (err) {
861 		f2fs_put_page(page, 1);
862 		return ERR_PTR(err);
863 	}
864 	if (!ipage)
865 		f2fs_put_dnode(&dn);
866 
867 	if (PageUptodate(page))
868 		goto got_it;
869 
870 	if (dn.data_blkaddr == NEW_ADDR) {
871 		zero_user_segment(page, 0, PAGE_SIZE);
872 		if (!PageUptodate(page))
873 			SetPageUptodate(page);
874 	} else {
875 		f2fs_put_page(page, 1);
876 
877 		/* if ipage exists, blkaddr should be NEW_ADDR */
878 		f2fs_bug_on(F2FS_I_SB(inode), ipage);
879 		page = f2fs_get_lock_data_page(inode, index, true);
880 		if (IS_ERR(page))
881 			return page;
882 	}
883 got_it:
884 	if (new_i_size && i_size_read(inode) <
885 				((loff_t)(index + 1) << PAGE_SHIFT))
886 		f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
887 	return page;
888 }
889 
890 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
891 {
892 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
893 	struct f2fs_summary sum;
894 	struct node_info ni;
895 	block_t old_blkaddr;
896 	blkcnt_t count = 1;
897 	int err;
898 
899 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
900 		return -EPERM;
901 
902 	err = f2fs_get_node_info(sbi, dn->nid, &ni);
903 	if (err)
904 		return err;
905 
906 	dn->data_blkaddr = datablock_addr(dn->inode,
907 				dn->node_page, dn->ofs_in_node);
908 	if (dn->data_blkaddr != NULL_ADDR)
909 		goto alloc;
910 
911 	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
912 		return err;
913 
914 alloc:
915 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
916 	old_blkaddr = dn->data_blkaddr;
917 	f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
918 					&sum, seg_type, NULL, false);
919 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
920 		invalidate_mapping_pages(META_MAPPING(sbi),
921 					old_blkaddr, old_blkaddr);
922 	f2fs_set_data_blkaddr(dn);
923 
924 	/*
925 	 * i_size will be updated by direct_IO. Otherwise, we'll get stale
926 	 * data from unwritten block via dio_read.
927 	 */
928 	return 0;
929 }
930 
931 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
932 {
933 	struct inode *inode = file_inode(iocb->ki_filp);
934 	struct f2fs_map_blocks map;
935 	int flag;
936 	int err = 0;
937 	bool direct_io = iocb->ki_flags & IOCB_DIRECT;
938 
939 	/* convert inline data for Direct I/O*/
940 	if (direct_io) {
941 		err = f2fs_convert_inline_inode(inode);
942 		if (err)
943 			return err;
944 	}
945 
946 	if (direct_io && allow_outplace_dio(inode, iocb, from))
947 		return 0;
948 
949 	if (is_inode_flag_set(inode, FI_NO_PREALLOC))
950 		return 0;
951 
952 	map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
953 	map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
954 	if (map.m_len > map.m_lblk)
955 		map.m_len -= map.m_lblk;
956 	else
957 		map.m_len = 0;
958 
959 	map.m_next_pgofs = NULL;
960 	map.m_next_extent = NULL;
961 	map.m_seg_type = NO_CHECK_TYPE;
962 	map.m_may_create = true;
963 
964 	if (direct_io) {
965 		map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
966 		flag = f2fs_force_buffered_io(inode, iocb, from) ?
967 					F2FS_GET_BLOCK_PRE_AIO :
968 					F2FS_GET_BLOCK_PRE_DIO;
969 		goto map_blocks;
970 	}
971 	if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
972 		err = f2fs_convert_inline_inode(inode);
973 		if (err)
974 			return err;
975 	}
976 	if (f2fs_has_inline_data(inode))
977 		return err;
978 
979 	flag = F2FS_GET_BLOCK_PRE_AIO;
980 
981 map_blocks:
982 	err = f2fs_map_blocks(inode, &map, 1, flag);
983 	if (map.m_len > 0 && err == -ENOSPC) {
984 		if (!direct_io)
985 			set_inode_flag(inode, FI_NO_PREALLOC);
986 		err = 0;
987 	}
988 	return err;
989 }
990 
991 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
992 {
993 	if (flag == F2FS_GET_BLOCK_PRE_AIO) {
994 		if (lock)
995 			down_read(&sbi->node_change);
996 		else
997 			up_read(&sbi->node_change);
998 	} else {
999 		if (lock)
1000 			f2fs_lock_op(sbi);
1001 		else
1002 			f2fs_unlock_op(sbi);
1003 	}
1004 }
1005 
1006 /*
1007  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1008  * f2fs_map_blocks structure.
1009  * If original data blocks are allocated, then give them to blockdev.
1010  * Otherwise,
1011  *     a. preallocate requested block addresses
1012  *     b. do not use extent cache for better performance
1013  *     c. give the block addresses to blockdev
1014  */
1015 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1016 						int create, int flag)
1017 {
1018 	unsigned int maxblocks = map->m_len;
1019 	struct dnode_of_data dn;
1020 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1021 	int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1022 	pgoff_t pgofs, end_offset, end;
1023 	int err = 0, ofs = 1;
1024 	unsigned int ofs_in_node, last_ofs_in_node;
1025 	blkcnt_t prealloc;
1026 	struct extent_info ei = {0,0,0};
1027 	block_t blkaddr;
1028 	unsigned int start_pgofs;
1029 
1030 	if (!maxblocks)
1031 		return 0;
1032 
1033 	map->m_len = 0;
1034 	map->m_flags = 0;
1035 
1036 	/* it only supports block size == page size */
1037 	pgofs =	(pgoff_t)map->m_lblk;
1038 	end = pgofs + maxblocks;
1039 
1040 	if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1041 		if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1042 							map->m_may_create)
1043 			goto next_dnode;
1044 
1045 		map->m_pblk = ei.blk + pgofs - ei.fofs;
1046 		map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1047 		map->m_flags = F2FS_MAP_MAPPED;
1048 		if (map->m_next_extent)
1049 			*map->m_next_extent = pgofs + map->m_len;
1050 
1051 		/* for hardware encryption, but to avoid potential issue in future */
1052 		if (flag == F2FS_GET_BLOCK_DIO)
1053 			f2fs_wait_on_block_writeback_range(inode,
1054 						map->m_pblk, map->m_len);
1055 		goto out;
1056 	}
1057 
1058 next_dnode:
1059 	if (map->m_may_create)
1060 		__do_map_lock(sbi, flag, true);
1061 
1062 	/* When reading holes, we need its node page */
1063 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1064 	err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1065 	if (err) {
1066 		if (flag == F2FS_GET_BLOCK_BMAP)
1067 			map->m_pblk = 0;
1068 		if (err == -ENOENT) {
1069 			err = 0;
1070 			if (map->m_next_pgofs)
1071 				*map->m_next_pgofs =
1072 					f2fs_get_next_page_offset(&dn, pgofs);
1073 			if (map->m_next_extent)
1074 				*map->m_next_extent =
1075 					f2fs_get_next_page_offset(&dn, pgofs);
1076 		}
1077 		goto unlock_out;
1078 	}
1079 
1080 	start_pgofs = pgofs;
1081 	prealloc = 0;
1082 	last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1083 	end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1084 
1085 next_block:
1086 	blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1087 
1088 	if (__is_valid_data_blkaddr(blkaddr) &&
1089 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1090 		err = -EFAULT;
1091 		goto sync_out;
1092 	}
1093 
1094 	if (is_valid_data_blkaddr(sbi, blkaddr)) {
1095 		/* use out-place-update for driect IO under LFS mode */
1096 		if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1097 							map->m_may_create) {
1098 			err = __allocate_data_block(&dn, map->m_seg_type);
1099 			if (!err) {
1100 				blkaddr = dn.data_blkaddr;
1101 				set_inode_flag(inode, FI_APPEND_WRITE);
1102 			}
1103 		}
1104 	} else {
1105 		if (create) {
1106 			if (unlikely(f2fs_cp_error(sbi))) {
1107 				err = -EIO;
1108 				goto sync_out;
1109 			}
1110 			if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1111 				if (blkaddr == NULL_ADDR) {
1112 					prealloc++;
1113 					last_ofs_in_node = dn.ofs_in_node;
1114 				}
1115 			} else {
1116 				WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1117 					flag != F2FS_GET_BLOCK_DIO);
1118 				err = __allocate_data_block(&dn,
1119 							map->m_seg_type);
1120 				if (!err)
1121 					set_inode_flag(inode, FI_APPEND_WRITE);
1122 			}
1123 			if (err)
1124 				goto sync_out;
1125 			map->m_flags |= F2FS_MAP_NEW;
1126 			blkaddr = dn.data_blkaddr;
1127 		} else {
1128 			if (flag == F2FS_GET_BLOCK_BMAP) {
1129 				map->m_pblk = 0;
1130 				goto sync_out;
1131 			}
1132 			if (flag == F2FS_GET_BLOCK_PRECACHE)
1133 				goto sync_out;
1134 			if (flag == F2FS_GET_BLOCK_FIEMAP &&
1135 						blkaddr == NULL_ADDR) {
1136 				if (map->m_next_pgofs)
1137 					*map->m_next_pgofs = pgofs + 1;
1138 				goto sync_out;
1139 			}
1140 			if (flag != F2FS_GET_BLOCK_FIEMAP) {
1141 				/* for defragment case */
1142 				if (map->m_next_pgofs)
1143 					*map->m_next_pgofs = pgofs + 1;
1144 				goto sync_out;
1145 			}
1146 		}
1147 	}
1148 
1149 	if (flag == F2FS_GET_BLOCK_PRE_AIO)
1150 		goto skip;
1151 
1152 	if (map->m_len == 0) {
1153 		/* preallocated unwritten block should be mapped for fiemap. */
1154 		if (blkaddr == NEW_ADDR)
1155 			map->m_flags |= F2FS_MAP_UNWRITTEN;
1156 		map->m_flags |= F2FS_MAP_MAPPED;
1157 
1158 		map->m_pblk = blkaddr;
1159 		map->m_len = 1;
1160 	} else if ((map->m_pblk != NEW_ADDR &&
1161 			blkaddr == (map->m_pblk + ofs)) ||
1162 			(map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1163 			flag == F2FS_GET_BLOCK_PRE_DIO) {
1164 		ofs++;
1165 		map->m_len++;
1166 	} else {
1167 		goto sync_out;
1168 	}
1169 
1170 skip:
1171 	dn.ofs_in_node++;
1172 	pgofs++;
1173 
1174 	/* preallocate blocks in batch for one dnode page */
1175 	if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1176 			(pgofs == end || dn.ofs_in_node == end_offset)) {
1177 
1178 		dn.ofs_in_node = ofs_in_node;
1179 		err = f2fs_reserve_new_blocks(&dn, prealloc);
1180 		if (err)
1181 			goto sync_out;
1182 
1183 		map->m_len += dn.ofs_in_node - ofs_in_node;
1184 		if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1185 			err = -ENOSPC;
1186 			goto sync_out;
1187 		}
1188 		dn.ofs_in_node = end_offset;
1189 	}
1190 
1191 	if (pgofs >= end)
1192 		goto sync_out;
1193 	else if (dn.ofs_in_node < end_offset)
1194 		goto next_block;
1195 
1196 	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1197 		if (map->m_flags & F2FS_MAP_MAPPED) {
1198 			unsigned int ofs = start_pgofs - map->m_lblk;
1199 
1200 			f2fs_update_extent_cache_range(&dn,
1201 				start_pgofs, map->m_pblk + ofs,
1202 				map->m_len - ofs);
1203 		}
1204 	}
1205 
1206 	f2fs_put_dnode(&dn);
1207 
1208 	if (map->m_may_create) {
1209 		__do_map_lock(sbi, flag, false);
1210 		f2fs_balance_fs(sbi, dn.node_changed);
1211 	}
1212 	goto next_dnode;
1213 
1214 sync_out:
1215 
1216 	/* for hardware encryption, but to avoid potential issue in future */
1217 	if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1218 		f2fs_wait_on_block_writeback_range(inode,
1219 						map->m_pblk, map->m_len);
1220 
1221 	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1222 		if (map->m_flags & F2FS_MAP_MAPPED) {
1223 			unsigned int ofs = start_pgofs - map->m_lblk;
1224 
1225 			f2fs_update_extent_cache_range(&dn,
1226 				start_pgofs, map->m_pblk + ofs,
1227 				map->m_len - ofs);
1228 		}
1229 		if (map->m_next_extent)
1230 			*map->m_next_extent = pgofs + 1;
1231 	}
1232 	f2fs_put_dnode(&dn);
1233 unlock_out:
1234 	if (map->m_may_create) {
1235 		__do_map_lock(sbi, flag, false);
1236 		f2fs_balance_fs(sbi, dn.node_changed);
1237 	}
1238 out:
1239 	trace_f2fs_map_blocks(inode, map, err);
1240 	return err;
1241 }
1242 
1243 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1244 {
1245 	struct f2fs_map_blocks map;
1246 	block_t last_lblk;
1247 	int err;
1248 
1249 	if (pos + len > i_size_read(inode))
1250 		return false;
1251 
1252 	map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1253 	map.m_next_pgofs = NULL;
1254 	map.m_next_extent = NULL;
1255 	map.m_seg_type = NO_CHECK_TYPE;
1256 	map.m_may_create = false;
1257 	last_lblk = F2FS_BLK_ALIGN(pos + len);
1258 
1259 	while (map.m_lblk < last_lblk) {
1260 		map.m_len = last_lblk - map.m_lblk;
1261 		err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1262 		if (err || map.m_len == 0)
1263 			return false;
1264 		map.m_lblk += map.m_len;
1265 	}
1266 	return true;
1267 }
1268 
1269 static int __get_data_block(struct inode *inode, sector_t iblock,
1270 			struct buffer_head *bh, int create, int flag,
1271 			pgoff_t *next_pgofs, int seg_type, bool may_write)
1272 {
1273 	struct f2fs_map_blocks map;
1274 	int err;
1275 
1276 	map.m_lblk = iblock;
1277 	map.m_len = bh->b_size >> inode->i_blkbits;
1278 	map.m_next_pgofs = next_pgofs;
1279 	map.m_next_extent = NULL;
1280 	map.m_seg_type = seg_type;
1281 	map.m_may_create = may_write;
1282 
1283 	err = f2fs_map_blocks(inode, &map, create, flag);
1284 	if (!err) {
1285 		map_bh(bh, inode->i_sb, map.m_pblk);
1286 		bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1287 		bh->b_size = (u64)map.m_len << inode->i_blkbits;
1288 	}
1289 	return err;
1290 }
1291 
1292 static int get_data_block(struct inode *inode, sector_t iblock,
1293 			struct buffer_head *bh_result, int create, int flag,
1294 			pgoff_t *next_pgofs)
1295 {
1296 	return __get_data_block(inode, iblock, bh_result, create,
1297 							flag, next_pgofs,
1298 							NO_CHECK_TYPE, create);
1299 }
1300 
1301 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1302 			struct buffer_head *bh_result, int create)
1303 {
1304 	return __get_data_block(inode, iblock, bh_result, create,
1305 				F2FS_GET_BLOCK_DIO, NULL,
1306 				f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1307 				true);
1308 }
1309 
1310 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1311 			struct buffer_head *bh_result, int create)
1312 {
1313 	return __get_data_block(inode, iblock, bh_result, create,
1314 				F2FS_GET_BLOCK_DIO, NULL,
1315 				f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1316 				false);
1317 }
1318 
1319 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1320 			struct buffer_head *bh_result, int create)
1321 {
1322 	/* Block number less than F2FS MAX BLOCKS */
1323 	if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1324 		return -EFBIG;
1325 
1326 	return __get_data_block(inode, iblock, bh_result, create,
1327 						F2FS_GET_BLOCK_BMAP, NULL,
1328 						NO_CHECK_TYPE, create);
1329 }
1330 
1331 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1332 {
1333 	return (offset >> inode->i_blkbits);
1334 }
1335 
1336 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1337 {
1338 	return (blk << inode->i_blkbits);
1339 }
1340 
1341 static int f2fs_xattr_fiemap(struct inode *inode,
1342 				struct fiemap_extent_info *fieinfo)
1343 {
1344 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1345 	struct page *page;
1346 	struct node_info ni;
1347 	__u64 phys = 0, len;
1348 	__u32 flags;
1349 	nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1350 	int err = 0;
1351 
1352 	if (f2fs_has_inline_xattr(inode)) {
1353 		int offset;
1354 
1355 		page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1356 						inode->i_ino, false);
1357 		if (!page)
1358 			return -ENOMEM;
1359 
1360 		err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1361 		if (err) {
1362 			f2fs_put_page(page, 1);
1363 			return err;
1364 		}
1365 
1366 		phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1367 		offset = offsetof(struct f2fs_inode, i_addr) +
1368 					sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1369 					get_inline_xattr_addrs(inode));
1370 
1371 		phys += offset;
1372 		len = inline_xattr_size(inode);
1373 
1374 		f2fs_put_page(page, 1);
1375 
1376 		flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1377 
1378 		if (!xnid)
1379 			flags |= FIEMAP_EXTENT_LAST;
1380 
1381 		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1382 		if (err || err == 1)
1383 			return err;
1384 	}
1385 
1386 	if (xnid) {
1387 		page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1388 		if (!page)
1389 			return -ENOMEM;
1390 
1391 		err = f2fs_get_node_info(sbi, xnid, &ni);
1392 		if (err) {
1393 			f2fs_put_page(page, 1);
1394 			return err;
1395 		}
1396 
1397 		phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1398 		len = inode->i_sb->s_blocksize;
1399 
1400 		f2fs_put_page(page, 1);
1401 
1402 		flags = FIEMAP_EXTENT_LAST;
1403 	}
1404 
1405 	if (phys)
1406 		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1407 
1408 	return (err < 0 ? err : 0);
1409 }
1410 
1411 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1412 		u64 start, u64 len)
1413 {
1414 	struct buffer_head map_bh;
1415 	sector_t start_blk, last_blk;
1416 	pgoff_t next_pgofs;
1417 	u64 logical = 0, phys = 0, size = 0;
1418 	u32 flags = 0;
1419 	int ret = 0;
1420 
1421 	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1422 		ret = f2fs_precache_extents(inode);
1423 		if (ret)
1424 			return ret;
1425 	}
1426 
1427 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1428 	if (ret)
1429 		return ret;
1430 
1431 	inode_lock(inode);
1432 
1433 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1434 		ret = f2fs_xattr_fiemap(inode, fieinfo);
1435 		goto out;
1436 	}
1437 
1438 	if (f2fs_has_inline_data(inode)) {
1439 		ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1440 		if (ret != -EAGAIN)
1441 			goto out;
1442 	}
1443 
1444 	if (logical_to_blk(inode, len) == 0)
1445 		len = blk_to_logical(inode, 1);
1446 
1447 	start_blk = logical_to_blk(inode, start);
1448 	last_blk = logical_to_blk(inode, start + len - 1);
1449 
1450 next:
1451 	memset(&map_bh, 0, sizeof(struct buffer_head));
1452 	map_bh.b_size = len;
1453 
1454 	ret = get_data_block(inode, start_blk, &map_bh, 0,
1455 					F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1456 	if (ret)
1457 		goto out;
1458 
1459 	/* HOLE */
1460 	if (!buffer_mapped(&map_bh)) {
1461 		start_blk = next_pgofs;
1462 
1463 		if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1464 					F2FS_I_SB(inode)->max_file_blocks))
1465 			goto prep_next;
1466 
1467 		flags |= FIEMAP_EXTENT_LAST;
1468 	}
1469 
1470 	if (size) {
1471 		if (IS_ENCRYPTED(inode))
1472 			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1473 
1474 		ret = fiemap_fill_next_extent(fieinfo, logical,
1475 				phys, size, flags);
1476 	}
1477 
1478 	if (start_blk > last_blk || ret)
1479 		goto out;
1480 
1481 	logical = blk_to_logical(inode, start_blk);
1482 	phys = blk_to_logical(inode, map_bh.b_blocknr);
1483 	size = map_bh.b_size;
1484 	flags = 0;
1485 	if (buffer_unwritten(&map_bh))
1486 		flags = FIEMAP_EXTENT_UNWRITTEN;
1487 
1488 	start_blk += logical_to_blk(inode, size);
1489 
1490 prep_next:
1491 	cond_resched();
1492 	if (fatal_signal_pending(current))
1493 		ret = -EINTR;
1494 	else
1495 		goto next;
1496 out:
1497 	if (ret == 1)
1498 		ret = 0;
1499 
1500 	inode_unlock(inode);
1501 	return ret;
1502 }
1503 
1504 /*
1505  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1506  * Major change was from block_size == page_size in f2fs by default.
1507  *
1508  * Note that the aops->readpages() function is ONLY used for read-ahead. If
1509  * this function ever deviates from doing just read-ahead, it should either
1510  * use ->readpage() or do the necessary surgery to decouple ->readpages()
1511  * from read-ahead.
1512  */
1513 static int f2fs_mpage_readpages(struct address_space *mapping,
1514 			struct list_head *pages, struct page *page,
1515 			unsigned nr_pages, bool is_readahead)
1516 {
1517 	struct bio *bio = NULL;
1518 	sector_t last_block_in_bio = 0;
1519 	struct inode *inode = mapping->host;
1520 	const unsigned blkbits = inode->i_blkbits;
1521 	const unsigned blocksize = 1 << blkbits;
1522 	sector_t block_in_file;
1523 	sector_t last_block;
1524 	sector_t last_block_in_file;
1525 	sector_t block_nr;
1526 	struct f2fs_map_blocks map;
1527 
1528 	map.m_pblk = 0;
1529 	map.m_lblk = 0;
1530 	map.m_len = 0;
1531 	map.m_flags = 0;
1532 	map.m_next_pgofs = NULL;
1533 	map.m_next_extent = NULL;
1534 	map.m_seg_type = NO_CHECK_TYPE;
1535 	map.m_may_create = false;
1536 
1537 	for (; nr_pages; nr_pages--) {
1538 		if (pages) {
1539 			page = list_last_entry(pages, struct page, lru);
1540 
1541 			prefetchw(&page->flags);
1542 			list_del(&page->lru);
1543 			if (add_to_page_cache_lru(page, mapping,
1544 						  page->index,
1545 						  readahead_gfp_mask(mapping)))
1546 				goto next_page;
1547 		}
1548 
1549 		block_in_file = (sector_t)page->index;
1550 		last_block = block_in_file + nr_pages;
1551 		last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1552 								blkbits;
1553 		if (last_block > last_block_in_file)
1554 			last_block = last_block_in_file;
1555 
1556 		/*
1557 		 * Map blocks using the previous result first.
1558 		 */
1559 		if ((map.m_flags & F2FS_MAP_MAPPED) &&
1560 				block_in_file > map.m_lblk &&
1561 				block_in_file < (map.m_lblk + map.m_len))
1562 			goto got_it;
1563 
1564 		/*
1565 		 * Then do more f2fs_map_blocks() calls until we are
1566 		 * done with this page.
1567 		 */
1568 		map.m_flags = 0;
1569 
1570 		if (block_in_file < last_block) {
1571 			map.m_lblk = block_in_file;
1572 			map.m_len = last_block - block_in_file;
1573 
1574 			if (f2fs_map_blocks(inode, &map, 0,
1575 						F2FS_GET_BLOCK_DEFAULT))
1576 				goto set_error_page;
1577 		}
1578 got_it:
1579 		if ((map.m_flags & F2FS_MAP_MAPPED)) {
1580 			block_nr = map.m_pblk + block_in_file - map.m_lblk;
1581 			SetPageMappedToDisk(page);
1582 
1583 			if (!PageUptodate(page) && !cleancache_get_page(page)) {
1584 				SetPageUptodate(page);
1585 				goto confused;
1586 			}
1587 
1588 			if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1589 								DATA_GENERIC))
1590 				goto set_error_page;
1591 		} else {
1592 			zero_user_segment(page, 0, PAGE_SIZE);
1593 			if (!PageUptodate(page))
1594 				SetPageUptodate(page);
1595 			unlock_page(page);
1596 			goto next_page;
1597 		}
1598 
1599 		/*
1600 		 * This page will go to BIO.  Do we need to send this
1601 		 * BIO off first?
1602 		 */
1603 		if (bio && (last_block_in_bio != block_nr - 1 ||
1604 			!__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1605 submit_and_realloc:
1606 			__submit_bio(F2FS_I_SB(inode), bio, DATA);
1607 			bio = NULL;
1608 		}
1609 		if (bio == NULL) {
1610 			bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1611 					is_readahead ? REQ_RAHEAD : 0);
1612 			if (IS_ERR(bio)) {
1613 				bio = NULL;
1614 				goto set_error_page;
1615 			}
1616 		}
1617 
1618 		/*
1619 		 * If the page is under writeback, we need to wait for
1620 		 * its completion to see the correct decrypted data.
1621 		 */
1622 		f2fs_wait_on_block_writeback(inode, block_nr);
1623 
1624 		if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1625 			goto submit_and_realloc;
1626 
1627 		inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1628 		ClearPageError(page);
1629 		last_block_in_bio = block_nr;
1630 		goto next_page;
1631 set_error_page:
1632 		SetPageError(page);
1633 		zero_user_segment(page, 0, PAGE_SIZE);
1634 		unlock_page(page);
1635 		goto next_page;
1636 confused:
1637 		if (bio) {
1638 			__submit_bio(F2FS_I_SB(inode), bio, DATA);
1639 			bio = NULL;
1640 		}
1641 		unlock_page(page);
1642 next_page:
1643 		if (pages)
1644 			put_page(page);
1645 	}
1646 	BUG_ON(pages && !list_empty(pages));
1647 	if (bio)
1648 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
1649 	return 0;
1650 }
1651 
1652 static int f2fs_read_data_page(struct file *file, struct page *page)
1653 {
1654 	struct inode *inode = page->mapping->host;
1655 	int ret = -EAGAIN;
1656 
1657 	trace_f2fs_readpage(page, DATA);
1658 
1659 	/* If the file has inline data, try to read it directly */
1660 	if (f2fs_has_inline_data(inode))
1661 		ret = f2fs_read_inline_data(inode, page);
1662 	if (ret == -EAGAIN)
1663 		ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1664 	return ret;
1665 }
1666 
1667 static int f2fs_read_data_pages(struct file *file,
1668 			struct address_space *mapping,
1669 			struct list_head *pages, unsigned nr_pages)
1670 {
1671 	struct inode *inode = mapping->host;
1672 	struct page *page = list_last_entry(pages, struct page, lru);
1673 
1674 	trace_f2fs_readpages(inode, page, nr_pages);
1675 
1676 	/* If the file has inline data, skip readpages */
1677 	if (f2fs_has_inline_data(inode))
1678 		return 0;
1679 
1680 	return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1681 }
1682 
1683 static int encrypt_one_page(struct f2fs_io_info *fio)
1684 {
1685 	struct inode *inode = fio->page->mapping->host;
1686 	struct page *mpage;
1687 	gfp_t gfp_flags = GFP_NOFS;
1688 
1689 	if (!f2fs_encrypted_file(inode))
1690 		return 0;
1691 
1692 	/* wait for GCed page writeback via META_MAPPING */
1693 	f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1694 
1695 retry_encrypt:
1696 	fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1697 			PAGE_SIZE, 0, fio->page->index, gfp_flags);
1698 	if (IS_ERR(fio->encrypted_page)) {
1699 		/* flush pending IOs and wait for a while in the ENOMEM case */
1700 		if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1701 			f2fs_flush_merged_writes(fio->sbi);
1702 			congestion_wait(BLK_RW_ASYNC, HZ/50);
1703 			gfp_flags |= __GFP_NOFAIL;
1704 			goto retry_encrypt;
1705 		}
1706 		return PTR_ERR(fio->encrypted_page);
1707 	}
1708 
1709 	mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1710 	if (mpage) {
1711 		if (PageUptodate(mpage))
1712 			memcpy(page_address(mpage),
1713 				page_address(fio->encrypted_page), PAGE_SIZE);
1714 		f2fs_put_page(mpage, 1);
1715 	}
1716 	return 0;
1717 }
1718 
1719 static inline bool check_inplace_update_policy(struct inode *inode,
1720 				struct f2fs_io_info *fio)
1721 {
1722 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1723 	unsigned int policy = SM_I(sbi)->ipu_policy;
1724 
1725 	if (policy & (0x1 << F2FS_IPU_FORCE))
1726 		return true;
1727 	if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1728 		return true;
1729 	if (policy & (0x1 << F2FS_IPU_UTIL) &&
1730 			utilization(sbi) > SM_I(sbi)->min_ipu_util)
1731 		return true;
1732 	if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1733 			utilization(sbi) > SM_I(sbi)->min_ipu_util)
1734 		return true;
1735 
1736 	/*
1737 	 * IPU for rewrite async pages
1738 	 */
1739 	if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1740 			fio && fio->op == REQ_OP_WRITE &&
1741 			!(fio->op_flags & REQ_SYNC) &&
1742 			!IS_ENCRYPTED(inode))
1743 		return true;
1744 
1745 	/* this is only set during fdatasync */
1746 	if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1747 			is_inode_flag_set(inode, FI_NEED_IPU))
1748 		return true;
1749 
1750 	if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1751 			!f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1752 		return true;
1753 
1754 	return false;
1755 }
1756 
1757 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1758 {
1759 	if (f2fs_is_pinned_file(inode))
1760 		return true;
1761 
1762 	/* if this is cold file, we should overwrite to avoid fragmentation */
1763 	if (file_is_cold(inode))
1764 		return true;
1765 
1766 	return check_inplace_update_policy(inode, fio);
1767 }
1768 
1769 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1770 {
1771 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1772 
1773 	if (test_opt(sbi, LFS))
1774 		return true;
1775 	if (S_ISDIR(inode->i_mode))
1776 		return true;
1777 	if (IS_NOQUOTA(inode))
1778 		return true;
1779 	if (f2fs_is_atomic_file(inode))
1780 		return true;
1781 	if (fio) {
1782 		if (is_cold_data(fio->page))
1783 			return true;
1784 		if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1785 			return true;
1786 		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1787 			f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1788 			return true;
1789 	}
1790 	return false;
1791 }
1792 
1793 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1794 {
1795 	struct inode *inode = fio->page->mapping->host;
1796 
1797 	if (f2fs_should_update_outplace(inode, fio))
1798 		return false;
1799 
1800 	return f2fs_should_update_inplace(inode, fio);
1801 }
1802 
1803 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1804 {
1805 	struct page *page = fio->page;
1806 	struct inode *inode = page->mapping->host;
1807 	struct dnode_of_data dn;
1808 	struct extent_info ei = {0,0,0};
1809 	struct node_info ni;
1810 	bool ipu_force = false;
1811 	int err = 0;
1812 
1813 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1814 	if (need_inplace_update(fio) &&
1815 			f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1816 		fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1817 
1818 		if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1819 							DATA_GENERIC))
1820 			return -EFAULT;
1821 
1822 		ipu_force = true;
1823 		fio->need_lock = LOCK_DONE;
1824 		goto got_it;
1825 	}
1826 
1827 	/* Deadlock due to between page->lock and f2fs_lock_op */
1828 	if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1829 		return -EAGAIN;
1830 
1831 	err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1832 	if (err)
1833 		goto out;
1834 
1835 	fio->old_blkaddr = dn.data_blkaddr;
1836 
1837 	/* This page is already truncated */
1838 	if (fio->old_blkaddr == NULL_ADDR) {
1839 		ClearPageUptodate(page);
1840 		clear_cold_data(page);
1841 		goto out_writepage;
1842 	}
1843 got_it:
1844 	if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1845 		!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1846 							DATA_GENERIC)) {
1847 		err = -EFAULT;
1848 		goto out_writepage;
1849 	}
1850 	/*
1851 	 * If current allocation needs SSR,
1852 	 * it had better in-place writes for updated data.
1853 	 */
1854 	if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1855 					need_inplace_update(fio))) {
1856 		err = encrypt_one_page(fio);
1857 		if (err)
1858 			goto out_writepage;
1859 
1860 		set_page_writeback(page);
1861 		ClearPageError(page);
1862 		f2fs_put_dnode(&dn);
1863 		if (fio->need_lock == LOCK_REQ)
1864 			f2fs_unlock_op(fio->sbi);
1865 		err = f2fs_inplace_write_data(fio);
1866 		if (err && PageWriteback(page))
1867 			end_page_writeback(page);
1868 		trace_f2fs_do_write_data_page(fio->page, IPU);
1869 		set_inode_flag(inode, FI_UPDATE_WRITE);
1870 		return err;
1871 	}
1872 
1873 	if (fio->need_lock == LOCK_RETRY) {
1874 		if (!f2fs_trylock_op(fio->sbi)) {
1875 			err = -EAGAIN;
1876 			goto out_writepage;
1877 		}
1878 		fio->need_lock = LOCK_REQ;
1879 	}
1880 
1881 	err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1882 	if (err)
1883 		goto out_writepage;
1884 
1885 	fio->version = ni.version;
1886 
1887 	err = encrypt_one_page(fio);
1888 	if (err)
1889 		goto out_writepage;
1890 
1891 	set_page_writeback(page);
1892 	ClearPageError(page);
1893 
1894 	/* LFS mode write path */
1895 	f2fs_outplace_write_data(&dn, fio);
1896 	trace_f2fs_do_write_data_page(page, OPU);
1897 	set_inode_flag(inode, FI_APPEND_WRITE);
1898 	if (page->index == 0)
1899 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1900 out_writepage:
1901 	f2fs_put_dnode(&dn);
1902 out:
1903 	if (fio->need_lock == LOCK_REQ)
1904 		f2fs_unlock_op(fio->sbi);
1905 	return err;
1906 }
1907 
1908 static int __write_data_page(struct page *page, bool *submitted,
1909 				struct writeback_control *wbc,
1910 				enum iostat_type io_type)
1911 {
1912 	struct inode *inode = page->mapping->host;
1913 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1914 	loff_t i_size = i_size_read(inode);
1915 	const pgoff_t end_index = ((unsigned long long) i_size)
1916 							>> PAGE_SHIFT;
1917 	loff_t psize = (page->index + 1) << PAGE_SHIFT;
1918 	unsigned offset = 0;
1919 	bool need_balance_fs = false;
1920 	int err = 0;
1921 	struct f2fs_io_info fio = {
1922 		.sbi = sbi,
1923 		.ino = inode->i_ino,
1924 		.type = DATA,
1925 		.op = REQ_OP_WRITE,
1926 		.op_flags = wbc_to_write_flags(wbc),
1927 		.old_blkaddr = NULL_ADDR,
1928 		.page = page,
1929 		.encrypted_page = NULL,
1930 		.submitted = false,
1931 		.need_lock = LOCK_RETRY,
1932 		.io_type = io_type,
1933 		.io_wbc = wbc,
1934 	};
1935 
1936 	trace_f2fs_writepage(page, DATA);
1937 
1938 	/* we should bypass data pages to proceed the kworkder jobs */
1939 	if (unlikely(f2fs_cp_error(sbi))) {
1940 		mapping_set_error(page->mapping, -EIO);
1941 		/*
1942 		 * don't drop any dirty dentry pages for keeping lastest
1943 		 * directory structure.
1944 		 */
1945 		if (S_ISDIR(inode->i_mode))
1946 			goto redirty_out;
1947 		goto out;
1948 	}
1949 
1950 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1951 		goto redirty_out;
1952 
1953 	if (page->index < end_index)
1954 		goto write;
1955 
1956 	/*
1957 	 * If the offset is out-of-range of file size,
1958 	 * this page does not have to be written to disk.
1959 	 */
1960 	offset = i_size & (PAGE_SIZE - 1);
1961 	if ((page->index >= end_index + 1) || !offset)
1962 		goto out;
1963 
1964 	zero_user_segment(page, offset, PAGE_SIZE);
1965 write:
1966 	if (f2fs_is_drop_cache(inode))
1967 		goto out;
1968 	/* we should not write 0'th page having journal header */
1969 	if (f2fs_is_volatile_file(inode) && (!page->index ||
1970 			(!wbc->for_reclaim &&
1971 			f2fs_available_free_memory(sbi, BASE_CHECK))))
1972 		goto redirty_out;
1973 
1974 	/* Dentry blocks are controlled by checkpoint */
1975 	if (S_ISDIR(inode->i_mode)) {
1976 		fio.need_lock = LOCK_DONE;
1977 		err = f2fs_do_write_data_page(&fio);
1978 		goto done;
1979 	}
1980 
1981 	if (!wbc->for_reclaim)
1982 		need_balance_fs = true;
1983 	else if (has_not_enough_free_secs(sbi, 0, 0))
1984 		goto redirty_out;
1985 	else
1986 		set_inode_flag(inode, FI_HOT_DATA);
1987 
1988 	err = -EAGAIN;
1989 	if (f2fs_has_inline_data(inode)) {
1990 		err = f2fs_write_inline_data(inode, page);
1991 		if (!err)
1992 			goto out;
1993 	}
1994 
1995 	if (err == -EAGAIN) {
1996 		err = f2fs_do_write_data_page(&fio);
1997 		if (err == -EAGAIN) {
1998 			fio.need_lock = LOCK_REQ;
1999 			err = f2fs_do_write_data_page(&fio);
2000 		}
2001 	}
2002 
2003 	if (err) {
2004 		file_set_keep_isize(inode);
2005 	} else {
2006 		down_write(&F2FS_I(inode)->i_sem);
2007 		if (F2FS_I(inode)->last_disk_size < psize)
2008 			F2FS_I(inode)->last_disk_size = psize;
2009 		up_write(&F2FS_I(inode)->i_sem);
2010 	}
2011 
2012 done:
2013 	if (err && err != -ENOENT)
2014 		goto redirty_out;
2015 
2016 out:
2017 	inode_dec_dirty_pages(inode);
2018 	if (err) {
2019 		ClearPageUptodate(page);
2020 		clear_cold_data(page);
2021 	}
2022 
2023 	if (wbc->for_reclaim) {
2024 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2025 		clear_inode_flag(inode, FI_HOT_DATA);
2026 		f2fs_remove_dirty_inode(inode);
2027 		submitted = NULL;
2028 	}
2029 
2030 	unlock_page(page);
2031 	if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode))
2032 		f2fs_balance_fs(sbi, need_balance_fs);
2033 
2034 	if (unlikely(f2fs_cp_error(sbi))) {
2035 		f2fs_submit_merged_write(sbi, DATA);
2036 		submitted = NULL;
2037 	}
2038 
2039 	if (submitted)
2040 		*submitted = fio.submitted;
2041 
2042 	return 0;
2043 
2044 redirty_out:
2045 	redirty_page_for_writepage(wbc, page);
2046 	/*
2047 	 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2048 	 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2049 	 * file_write_and_wait_range() will see EIO error, which is critical
2050 	 * to return value of fsync() followed by atomic_write failure to user.
2051 	 */
2052 	if (!err || wbc->for_reclaim)
2053 		return AOP_WRITEPAGE_ACTIVATE;
2054 	unlock_page(page);
2055 	return err;
2056 }
2057 
2058 static int f2fs_write_data_page(struct page *page,
2059 					struct writeback_control *wbc)
2060 {
2061 	return __write_data_page(page, NULL, wbc, FS_DATA_IO);
2062 }
2063 
2064 /*
2065  * This function was copied from write_cche_pages from mm/page-writeback.c.
2066  * The major change is making write step of cold data page separately from
2067  * warm/hot data page.
2068  */
2069 static int f2fs_write_cache_pages(struct address_space *mapping,
2070 					struct writeback_control *wbc,
2071 					enum iostat_type io_type)
2072 {
2073 	int ret = 0;
2074 	int done = 0;
2075 	struct pagevec pvec;
2076 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2077 	int nr_pages;
2078 	pgoff_t uninitialized_var(writeback_index);
2079 	pgoff_t index;
2080 	pgoff_t end;		/* Inclusive */
2081 	pgoff_t done_index;
2082 	int cycled;
2083 	int range_whole = 0;
2084 	xa_mark_t tag;
2085 	int nwritten = 0;
2086 
2087 	pagevec_init(&pvec);
2088 
2089 	if (get_dirty_pages(mapping->host) <=
2090 				SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2091 		set_inode_flag(mapping->host, FI_HOT_DATA);
2092 	else
2093 		clear_inode_flag(mapping->host, FI_HOT_DATA);
2094 
2095 	if (wbc->range_cyclic) {
2096 		writeback_index = mapping->writeback_index; /* prev offset */
2097 		index = writeback_index;
2098 		if (index == 0)
2099 			cycled = 1;
2100 		else
2101 			cycled = 0;
2102 		end = -1;
2103 	} else {
2104 		index = wbc->range_start >> PAGE_SHIFT;
2105 		end = wbc->range_end >> PAGE_SHIFT;
2106 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2107 			range_whole = 1;
2108 		cycled = 1; /* ignore range_cyclic tests */
2109 	}
2110 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2111 		tag = PAGECACHE_TAG_TOWRITE;
2112 	else
2113 		tag = PAGECACHE_TAG_DIRTY;
2114 retry:
2115 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2116 		tag_pages_for_writeback(mapping, index, end);
2117 	done_index = index;
2118 	while (!done && (index <= end)) {
2119 		int i;
2120 
2121 		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2122 				tag);
2123 		if (nr_pages == 0)
2124 			break;
2125 
2126 		for (i = 0; i < nr_pages; i++) {
2127 			struct page *page = pvec.pages[i];
2128 			bool submitted = false;
2129 
2130 			/* give a priority to WB_SYNC threads */
2131 			if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2132 					wbc->sync_mode == WB_SYNC_NONE) {
2133 				done = 1;
2134 				break;
2135 			}
2136 
2137 			done_index = page->index;
2138 retry_write:
2139 			lock_page(page);
2140 
2141 			if (unlikely(page->mapping != mapping)) {
2142 continue_unlock:
2143 				unlock_page(page);
2144 				continue;
2145 			}
2146 
2147 			if (!PageDirty(page)) {
2148 				/* someone wrote it for us */
2149 				goto continue_unlock;
2150 			}
2151 
2152 			if (PageWriteback(page)) {
2153 				if (wbc->sync_mode != WB_SYNC_NONE)
2154 					f2fs_wait_on_page_writeback(page,
2155 							DATA, true, true);
2156 				else
2157 					goto continue_unlock;
2158 			}
2159 
2160 			if (!clear_page_dirty_for_io(page))
2161 				goto continue_unlock;
2162 
2163 			ret = __write_data_page(page, &submitted, wbc, io_type);
2164 			if (unlikely(ret)) {
2165 				/*
2166 				 * keep nr_to_write, since vfs uses this to
2167 				 * get # of written pages.
2168 				 */
2169 				if (ret == AOP_WRITEPAGE_ACTIVATE) {
2170 					unlock_page(page);
2171 					ret = 0;
2172 					continue;
2173 				} else if (ret == -EAGAIN) {
2174 					ret = 0;
2175 					if (wbc->sync_mode == WB_SYNC_ALL) {
2176 						cond_resched();
2177 						congestion_wait(BLK_RW_ASYNC,
2178 									HZ/50);
2179 						goto retry_write;
2180 					}
2181 					continue;
2182 				}
2183 				done_index = page->index + 1;
2184 				done = 1;
2185 				break;
2186 			} else if (submitted) {
2187 				nwritten++;
2188 			}
2189 
2190 			if (--wbc->nr_to_write <= 0 &&
2191 					wbc->sync_mode == WB_SYNC_NONE) {
2192 				done = 1;
2193 				break;
2194 			}
2195 		}
2196 		pagevec_release(&pvec);
2197 		cond_resched();
2198 	}
2199 
2200 	if (!cycled && !done) {
2201 		cycled = 1;
2202 		index = 0;
2203 		end = writeback_index - 1;
2204 		goto retry;
2205 	}
2206 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2207 		mapping->writeback_index = done_index;
2208 
2209 	if (nwritten)
2210 		f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2211 								NULL, 0, DATA);
2212 
2213 	return ret;
2214 }
2215 
2216 static inline bool __should_serialize_io(struct inode *inode,
2217 					struct writeback_control *wbc)
2218 {
2219 	if (!S_ISREG(inode->i_mode))
2220 		return false;
2221 	if (IS_NOQUOTA(inode))
2222 		return false;
2223 	if (wbc->sync_mode != WB_SYNC_ALL)
2224 		return true;
2225 	if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2226 		return true;
2227 	return false;
2228 }
2229 
2230 static int __f2fs_write_data_pages(struct address_space *mapping,
2231 						struct writeback_control *wbc,
2232 						enum iostat_type io_type)
2233 {
2234 	struct inode *inode = mapping->host;
2235 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2236 	struct blk_plug plug;
2237 	int ret;
2238 	bool locked = false;
2239 
2240 	/* deal with chardevs and other special file */
2241 	if (!mapping->a_ops->writepage)
2242 		return 0;
2243 
2244 	/* skip writing if there is no dirty page in this inode */
2245 	if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2246 		return 0;
2247 
2248 	/* during POR, we don't need to trigger writepage at all. */
2249 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2250 		goto skip_write;
2251 
2252 	if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2253 			wbc->sync_mode == WB_SYNC_NONE &&
2254 			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2255 			f2fs_available_free_memory(sbi, DIRTY_DENTS))
2256 		goto skip_write;
2257 
2258 	/* skip writing during file defragment */
2259 	if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2260 		goto skip_write;
2261 
2262 	trace_f2fs_writepages(mapping->host, wbc, DATA);
2263 
2264 	/* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2265 	if (wbc->sync_mode == WB_SYNC_ALL)
2266 		atomic_inc(&sbi->wb_sync_req[DATA]);
2267 	else if (atomic_read(&sbi->wb_sync_req[DATA]))
2268 		goto skip_write;
2269 
2270 	if (__should_serialize_io(inode, wbc)) {
2271 		mutex_lock(&sbi->writepages);
2272 		locked = true;
2273 	}
2274 
2275 	blk_start_plug(&plug);
2276 	ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2277 	blk_finish_plug(&plug);
2278 
2279 	if (locked)
2280 		mutex_unlock(&sbi->writepages);
2281 
2282 	if (wbc->sync_mode == WB_SYNC_ALL)
2283 		atomic_dec(&sbi->wb_sync_req[DATA]);
2284 	/*
2285 	 * if some pages were truncated, we cannot guarantee its mapping->host
2286 	 * to detect pending bios.
2287 	 */
2288 
2289 	f2fs_remove_dirty_inode(inode);
2290 	return ret;
2291 
2292 skip_write:
2293 	wbc->pages_skipped += get_dirty_pages(inode);
2294 	trace_f2fs_writepages(mapping->host, wbc, DATA);
2295 	return 0;
2296 }
2297 
2298 static int f2fs_write_data_pages(struct address_space *mapping,
2299 			    struct writeback_control *wbc)
2300 {
2301 	struct inode *inode = mapping->host;
2302 
2303 	return __f2fs_write_data_pages(mapping, wbc,
2304 			F2FS_I(inode)->cp_task == current ?
2305 			FS_CP_DATA_IO : FS_DATA_IO);
2306 }
2307 
2308 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2309 {
2310 	struct inode *inode = mapping->host;
2311 	loff_t i_size = i_size_read(inode);
2312 
2313 	if (to > i_size) {
2314 		down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2315 		down_write(&F2FS_I(inode)->i_mmap_sem);
2316 
2317 		truncate_pagecache(inode, i_size);
2318 		f2fs_truncate_blocks(inode, i_size, true, true);
2319 
2320 		up_write(&F2FS_I(inode)->i_mmap_sem);
2321 		up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2322 	}
2323 }
2324 
2325 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2326 			struct page *page, loff_t pos, unsigned len,
2327 			block_t *blk_addr, bool *node_changed)
2328 {
2329 	struct inode *inode = page->mapping->host;
2330 	pgoff_t index = page->index;
2331 	struct dnode_of_data dn;
2332 	struct page *ipage;
2333 	bool locked = false;
2334 	struct extent_info ei = {0,0,0};
2335 	int err = 0;
2336 	int flag;
2337 
2338 	/*
2339 	 * we already allocated all the blocks, so we don't need to get
2340 	 * the block addresses when there is no need to fill the page.
2341 	 */
2342 	if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2343 			!is_inode_flag_set(inode, FI_NO_PREALLOC))
2344 		return 0;
2345 
2346 	/* f2fs_lock_op avoids race between write CP and convert_inline_page */
2347 	if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2348 		flag = F2FS_GET_BLOCK_DEFAULT;
2349 	else
2350 		flag = F2FS_GET_BLOCK_PRE_AIO;
2351 
2352 	if (f2fs_has_inline_data(inode) ||
2353 			(pos & PAGE_MASK) >= i_size_read(inode)) {
2354 		__do_map_lock(sbi, flag, true);
2355 		locked = true;
2356 	}
2357 restart:
2358 	/* check inline_data */
2359 	ipage = f2fs_get_node_page(sbi, inode->i_ino);
2360 	if (IS_ERR(ipage)) {
2361 		err = PTR_ERR(ipage);
2362 		goto unlock_out;
2363 	}
2364 
2365 	set_new_dnode(&dn, inode, ipage, ipage, 0);
2366 
2367 	if (f2fs_has_inline_data(inode)) {
2368 		if (pos + len <= MAX_INLINE_DATA(inode)) {
2369 			f2fs_do_read_inline_data(page, ipage);
2370 			set_inode_flag(inode, FI_DATA_EXIST);
2371 			if (inode->i_nlink)
2372 				set_inline_node(ipage);
2373 		} else {
2374 			err = f2fs_convert_inline_page(&dn, page);
2375 			if (err)
2376 				goto out;
2377 			if (dn.data_blkaddr == NULL_ADDR)
2378 				err = f2fs_get_block(&dn, index);
2379 		}
2380 	} else if (locked) {
2381 		err = f2fs_get_block(&dn, index);
2382 	} else {
2383 		if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2384 			dn.data_blkaddr = ei.blk + index - ei.fofs;
2385 		} else {
2386 			/* hole case */
2387 			err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2388 			if (err || dn.data_blkaddr == NULL_ADDR) {
2389 				f2fs_put_dnode(&dn);
2390 				__do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2391 								true);
2392 				WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2393 				locked = true;
2394 				goto restart;
2395 			}
2396 		}
2397 	}
2398 
2399 	/* convert_inline_page can make node_changed */
2400 	*blk_addr = dn.data_blkaddr;
2401 	*node_changed = dn.node_changed;
2402 out:
2403 	f2fs_put_dnode(&dn);
2404 unlock_out:
2405 	if (locked)
2406 		__do_map_lock(sbi, flag, false);
2407 	return err;
2408 }
2409 
2410 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2411 		loff_t pos, unsigned len, unsigned flags,
2412 		struct page **pagep, void **fsdata)
2413 {
2414 	struct inode *inode = mapping->host;
2415 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2416 	struct page *page = NULL;
2417 	pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2418 	bool need_balance = false, drop_atomic = false;
2419 	block_t blkaddr = NULL_ADDR;
2420 	int err = 0;
2421 
2422 	trace_f2fs_write_begin(inode, pos, len, flags);
2423 
2424 	err = f2fs_is_checkpoint_ready(sbi);
2425 	if (err)
2426 		goto fail;
2427 
2428 	if ((f2fs_is_atomic_file(inode) &&
2429 			!f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2430 			is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2431 		err = -ENOMEM;
2432 		drop_atomic = true;
2433 		goto fail;
2434 	}
2435 
2436 	/*
2437 	 * We should check this at this moment to avoid deadlock on inode page
2438 	 * and #0 page. The locking rule for inline_data conversion should be:
2439 	 * lock_page(page #0) -> lock_page(inode_page)
2440 	 */
2441 	if (index != 0) {
2442 		err = f2fs_convert_inline_inode(inode);
2443 		if (err)
2444 			goto fail;
2445 	}
2446 repeat:
2447 	/*
2448 	 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2449 	 * wait_for_stable_page. Will wait that below with our IO control.
2450 	 */
2451 	page = f2fs_pagecache_get_page(mapping, index,
2452 				FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2453 	if (!page) {
2454 		err = -ENOMEM;
2455 		goto fail;
2456 	}
2457 
2458 	*pagep = page;
2459 
2460 	err = prepare_write_begin(sbi, page, pos, len,
2461 					&blkaddr, &need_balance);
2462 	if (err)
2463 		goto fail;
2464 
2465 	if (need_balance && !IS_NOQUOTA(inode) &&
2466 			has_not_enough_free_secs(sbi, 0, 0)) {
2467 		unlock_page(page);
2468 		f2fs_balance_fs(sbi, true);
2469 		lock_page(page);
2470 		if (page->mapping != mapping) {
2471 			/* The page got truncated from under us */
2472 			f2fs_put_page(page, 1);
2473 			goto repeat;
2474 		}
2475 	}
2476 
2477 	f2fs_wait_on_page_writeback(page, DATA, false, true);
2478 
2479 	if (len == PAGE_SIZE || PageUptodate(page))
2480 		return 0;
2481 
2482 	if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2483 		zero_user_segment(page, len, PAGE_SIZE);
2484 		return 0;
2485 	}
2486 
2487 	if (blkaddr == NEW_ADDR) {
2488 		zero_user_segment(page, 0, PAGE_SIZE);
2489 		SetPageUptodate(page);
2490 	} else {
2491 		err = f2fs_submit_page_read(inode, page, blkaddr);
2492 		if (err)
2493 			goto fail;
2494 
2495 		lock_page(page);
2496 		if (unlikely(page->mapping != mapping)) {
2497 			f2fs_put_page(page, 1);
2498 			goto repeat;
2499 		}
2500 		if (unlikely(!PageUptodate(page))) {
2501 			err = -EIO;
2502 			goto fail;
2503 		}
2504 	}
2505 	return 0;
2506 
2507 fail:
2508 	f2fs_put_page(page, 1);
2509 	f2fs_write_failed(mapping, pos + len);
2510 	if (drop_atomic)
2511 		f2fs_drop_inmem_pages_all(sbi, false);
2512 	return err;
2513 }
2514 
2515 static int f2fs_write_end(struct file *file,
2516 			struct address_space *mapping,
2517 			loff_t pos, unsigned len, unsigned copied,
2518 			struct page *page, void *fsdata)
2519 {
2520 	struct inode *inode = page->mapping->host;
2521 
2522 	trace_f2fs_write_end(inode, pos, len, copied);
2523 
2524 	/*
2525 	 * This should be come from len == PAGE_SIZE, and we expect copied
2526 	 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2527 	 * let generic_perform_write() try to copy data again through copied=0.
2528 	 */
2529 	if (!PageUptodate(page)) {
2530 		if (unlikely(copied != len))
2531 			copied = 0;
2532 		else
2533 			SetPageUptodate(page);
2534 	}
2535 	if (!copied)
2536 		goto unlock_out;
2537 
2538 	set_page_dirty(page);
2539 
2540 	if (pos + copied > i_size_read(inode))
2541 		f2fs_i_size_write(inode, pos + copied);
2542 unlock_out:
2543 	f2fs_put_page(page, 1);
2544 	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2545 	return copied;
2546 }
2547 
2548 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2549 			   loff_t offset)
2550 {
2551 	unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2552 	unsigned blkbits = i_blkbits;
2553 	unsigned blocksize_mask = (1 << blkbits) - 1;
2554 	unsigned long align = offset | iov_iter_alignment(iter);
2555 	struct block_device *bdev = inode->i_sb->s_bdev;
2556 
2557 	if (align & blocksize_mask) {
2558 		if (bdev)
2559 			blkbits = blksize_bits(bdev_logical_block_size(bdev));
2560 		blocksize_mask = (1 << blkbits) - 1;
2561 		if (align & blocksize_mask)
2562 			return -EINVAL;
2563 		return 1;
2564 	}
2565 	return 0;
2566 }
2567 
2568 static void f2fs_dio_end_io(struct bio *bio)
2569 {
2570 	struct f2fs_private_dio *dio = bio->bi_private;
2571 
2572 	dec_page_count(F2FS_I_SB(dio->inode),
2573 			dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2574 
2575 	bio->bi_private = dio->orig_private;
2576 	bio->bi_end_io = dio->orig_end_io;
2577 
2578 	kvfree(dio);
2579 
2580 	bio_endio(bio);
2581 }
2582 
2583 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2584 							loff_t file_offset)
2585 {
2586 	struct f2fs_private_dio *dio;
2587 	bool write = (bio_op(bio) == REQ_OP_WRITE);
2588 	int err;
2589 
2590 	dio = f2fs_kzalloc(F2FS_I_SB(inode),
2591 			sizeof(struct f2fs_private_dio), GFP_NOFS);
2592 	if (!dio) {
2593 		err = -ENOMEM;
2594 		goto out;
2595 	}
2596 
2597 	dio->inode = inode;
2598 	dio->orig_end_io = bio->bi_end_io;
2599 	dio->orig_private = bio->bi_private;
2600 	dio->write = write;
2601 
2602 	bio->bi_end_io = f2fs_dio_end_io;
2603 	bio->bi_private = dio;
2604 
2605 	inc_page_count(F2FS_I_SB(inode),
2606 			write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2607 
2608 	submit_bio(bio);
2609 	return;
2610 out:
2611 	bio->bi_status = BLK_STS_IOERR;
2612 	bio_endio(bio);
2613 }
2614 
2615 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2616 {
2617 	struct address_space *mapping = iocb->ki_filp->f_mapping;
2618 	struct inode *inode = mapping->host;
2619 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2620 	struct f2fs_inode_info *fi = F2FS_I(inode);
2621 	size_t count = iov_iter_count(iter);
2622 	loff_t offset = iocb->ki_pos;
2623 	int rw = iov_iter_rw(iter);
2624 	int err;
2625 	enum rw_hint hint = iocb->ki_hint;
2626 	int whint_mode = F2FS_OPTION(sbi).whint_mode;
2627 	bool do_opu;
2628 
2629 	err = check_direct_IO(inode, iter, offset);
2630 	if (err)
2631 		return err < 0 ? err : 0;
2632 
2633 	if (f2fs_force_buffered_io(inode, iocb, iter))
2634 		return 0;
2635 
2636 	do_opu = allow_outplace_dio(inode, iocb, iter);
2637 
2638 	trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2639 
2640 	if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2641 		iocb->ki_hint = WRITE_LIFE_NOT_SET;
2642 
2643 	if (iocb->ki_flags & IOCB_NOWAIT) {
2644 		if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2645 			iocb->ki_hint = hint;
2646 			err = -EAGAIN;
2647 			goto out;
2648 		}
2649 		if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2650 			up_read(&fi->i_gc_rwsem[rw]);
2651 			iocb->ki_hint = hint;
2652 			err = -EAGAIN;
2653 			goto out;
2654 		}
2655 	} else {
2656 		down_read(&fi->i_gc_rwsem[rw]);
2657 		if (do_opu)
2658 			down_read(&fi->i_gc_rwsem[READ]);
2659 	}
2660 
2661 	err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2662 			iter, rw == WRITE ? get_data_block_dio_write :
2663 			get_data_block_dio, NULL, f2fs_dio_submit_bio,
2664 			DIO_LOCKING | DIO_SKIP_HOLES);
2665 
2666 	if (do_opu)
2667 		up_read(&fi->i_gc_rwsem[READ]);
2668 
2669 	up_read(&fi->i_gc_rwsem[rw]);
2670 
2671 	if (rw == WRITE) {
2672 		if (whint_mode == WHINT_MODE_OFF)
2673 			iocb->ki_hint = hint;
2674 		if (err > 0) {
2675 			f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2676 									err);
2677 			if (!do_opu)
2678 				set_inode_flag(inode, FI_UPDATE_WRITE);
2679 		} else if (err < 0) {
2680 			f2fs_write_failed(mapping, offset + count);
2681 		}
2682 	}
2683 
2684 out:
2685 	trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2686 
2687 	return err;
2688 }
2689 
2690 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2691 							unsigned int length)
2692 {
2693 	struct inode *inode = page->mapping->host;
2694 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2695 
2696 	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2697 		(offset % PAGE_SIZE || length != PAGE_SIZE))
2698 		return;
2699 
2700 	if (PageDirty(page)) {
2701 		if (inode->i_ino == F2FS_META_INO(sbi)) {
2702 			dec_page_count(sbi, F2FS_DIRTY_META);
2703 		} else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2704 			dec_page_count(sbi, F2FS_DIRTY_NODES);
2705 		} else {
2706 			inode_dec_dirty_pages(inode);
2707 			f2fs_remove_dirty_inode(inode);
2708 		}
2709 	}
2710 
2711 	clear_cold_data(page);
2712 
2713 	/* This is atomic written page, keep Private */
2714 	if (IS_ATOMIC_WRITTEN_PAGE(page))
2715 		return f2fs_drop_inmem_page(inode, page);
2716 
2717 	set_page_private(page, 0);
2718 	ClearPagePrivate(page);
2719 }
2720 
2721 int f2fs_release_page(struct page *page, gfp_t wait)
2722 {
2723 	/* If this is dirty page, keep PagePrivate */
2724 	if (PageDirty(page))
2725 		return 0;
2726 
2727 	/* This is atomic written page, keep Private */
2728 	if (IS_ATOMIC_WRITTEN_PAGE(page))
2729 		return 0;
2730 
2731 	clear_cold_data(page);
2732 	set_page_private(page, 0);
2733 	ClearPagePrivate(page);
2734 	return 1;
2735 }
2736 
2737 static int f2fs_set_data_page_dirty(struct page *page)
2738 {
2739 	struct address_space *mapping = page->mapping;
2740 	struct inode *inode = mapping->host;
2741 
2742 	trace_f2fs_set_page_dirty(page, DATA);
2743 
2744 	if (!PageUptodate(page))
2745 		SetPageUptodate(page);
2746 
2747 	if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2748 		if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2749 			f2fs_register_inmem_page(inode, page);
2750 			return 1;
2751 		}
2752 		/*
2753 		 * Previously, this page has been registered, we just
2754 		 * return here.
2755 		 */
2756 		return 0;
2757 	}
2758 
2759 	if (!PageDirty(page)) {
2760 		__set_page_dirty_nobuffers(page);
2761 		f2fs_update_dirty_page(inode, page);
2762 		return 1;
2763 	}
2764 	return 0;
2765 }
2766 
2767 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2768 {
2769 	struct inode *inode = mapping->host;
2770 
2771 	if (f2fs_has_inline_data(inode))
2772 		return 0;
2773 
2774 	/* make sure allocating whole blocks */
2775 	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2776 		filemap_write_and_wait(mapping);
2777 
2778 	return generic_block_bmap(mapping, block, get_data_block_bmap);
2779 }
2780 
2781 #ifdef CONFIG_MIGRATION
2782 #include <linux/migrate.h>
2783 
2784 int f2fs_migrate_page(struct address_space *mapping,
2785 		struct page *newpage, struct page *page, enum migrate_mode mode)
2786 {
2787 	int rc, extra_count;
2788 	struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2789 	bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2790 
2791 	BUG_ON(PageWriteback(page));
2792 
2793 	/* migrating an atomic written page is safe with the inmem_lock hold */
2794 	if (atomic_written) {
2795 		if (mode != MIGRATE_SYNC)
2796 			return -EBUSY;
2797 		if (!mutex_trylock(&fi->inmem_lock))
2798 			return -EAGAIN;
2799 	}
2800 
2801 	/*
2802 	 * A reference is expected if PagePrivate set when move mapping,
2803 	 * however F2FS breaks this for maintaining dirty page counts when
2804 	 * truncating pages. So here adjusting the 'extra_count' make it work.
2805 	 */
2806 	extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2807 	rc = migrate_page_move_mapping(mapping, newpage,
2808 				page, mode, extra_count);
2809 	if (rc != MIGRATEPAGE_SUCCESS) {
2810 		if (atomic_written)
2811 			mutex_unlock(&fi->inmem_lock);
2812 		return rc;
2813 	}
2814 
2815 	if (atomic_written) {
2816 		struct inmem_pages *cur;
2817 		list_for_each_entry(cur, &fi->inmem_pages, list)
2818 			if (cur->page == page) {
2819 				cur->page = newpage;
2820 				break;
2821 			}
2822 		mutex_unlock(&fi->inmem_lock);
2823 		put_page(page);
2824 		get_page(newpage);
2825 	}
2826 
2827 	if (PagePrivate(page))
2828 		SetPagePrivate(newpage);
2829 	set_page_private(newpage, page_private(page));
2830 
2831 	if (mode != MIGRATE_SYNC_NO_COPY)
2832 		migrate_page_copy(newpage, page);
2833 	else
2834 		migrate_page_states(newpage, page);
2835 
2836 	return MIGRATEPAGE_SUCCESS;
2837 }
2838 #endif
2839 
2840 const struct address_space_operations f2fs_dblock_aops = {
2841 	.readpage	= f2fs_read_data_page,
2842 	.readpages	= f2fs_read_data_pages,
2843 	.writepage	= f2fs_write_data_page,
2844 	.writepages	= f2fs_write_data_pages,
2845 	.write_begin	= f2fs_write_begin,
2846 	.write_end	= f2fs_write_end,
2847 	.set_page_dirty	= f2fs_set_data_page_dirty,
2848 	.invalidatepage	= f2fs_invalidate_page,
2849 	.releasepage	= f2fs_release_page,
2850 	.direct_IO	= f2fs_direct_IO,
2851 	.bmap		= f2fs_bmap,
2852 #ifdef CONFIG_MIGRATION
2853 	.migratepage    = f2fs_migrate_page,
2854 #endif
2855 };
2856 
2857 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2858 {
2859 	struct address_space *mapping = page_mapping(page);
2860 	unsigned long flags;
2861 
2862 	xa_lock_irqsave(&mapping->i_pages, flags);
2863 	__xa_clear_mark(&mapping->i_pages, page_index(page),
2864 						PAGECACHE_TAG_DIRTY);
2865 	xa_unlock_irqrestore(&mapping->i_pages, flags);
2866 }
2867 
2868 int __init f2fs_init_post_read_processing(void)
2869 {
2870 	bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2871 	if (!bio_post_read_ctx_cache)
2872 		goto fail;
2873 	bio_post_read_ctx_pool =
2874 		mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2875 					 bio_post_read_ctx_cache);
2876 	if (!bio_post_read_ctx_pool)
2877 		goto fail_free_cache;
2878 	return 0;
2879 
2880 fail_free_cache:
2881 	kmem_cache_destroy(bio_post_read_ctx_cache);
2882 fail:
2883 	return -ENOMEM;
2884 }
2885 
2886 void __exit f2fs_destroy_post_read_processing(void)
2887 {
2888 	mempool_destroy(bio_post_read_ctx_pool);
2889 	kmem_cache_destroy(bio_post_read_ctx_cache);
2890 }
2891