xref: /openbmc/linux/fs/f2fs/data.c (revision dd7b2333e6cd31584682382fcf0a1c1e5140b936)
1 /*
2  * fs/f2fs/data.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
22 #include <linux/mm.h>
23 #include <linux/memcontrol.h>
24 #include <linux/cleancache.h>
25 
26 #include "f2fs.h"
27 #include "node.h"
28 #include "segment.h"
29 #include "trace.h"
30 #include <trace/events/f2fs.h>
31 
32 static bool __is_cp_guaranteed(struct page *page)
33 {
34 	struct address_space *mapping = page->mapping;
35 	struct inode *inode;
36 	struct f2fs_sb_info *sbi;
37 
38 	if (!mapping)
39 		return false;
40 
41 	inode = mapping->host;
42 	sbi = F2FS_I_SB(inode);
43 
44 	if (inode->i_ino == F2FS_META_INO(sbi) ||
45 			inode->i_ino ==  F2FS_NODE_INO(sbi) ||
46 			S_ISDIR(inode->i_mode) ||
47 			is_cold_data(page))
48 		return true;
49 	return false;
50 }
51 
52 static void f2fs_read_end_io(struct bio *bio)
53 {
54 	struct bio_vec *bvec;
55 	int i;
56 
57 #ifdef CONFIG_F2FS_FAULT_INJECTION
58 	if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO))
59 		bio->bi_error = -EIO;
60 #endif
61 
62 	if (f2fs_bio_encrypted(bio)) {
63 		if (bio->bi_error) {
64 			fscrypt_release_ctx(bio->bi_private);
65 		} else {
66 			fscrypt_decrypt_bio_pages(bio->bi_private, bio);
67 			return;
68 		}
69 	}
70 
71 	bio_for_each_segment_all(bvec, bio, i) {
72 		struct page *page = bvec->bv_page;
73 
74 		if (!bio->bi_error) {
75 			if (!PageUptodate(page))
76 				SetPageUptodate(page);
77 		} else {
78 			ClearPageUptodate(page);
79 			SetPageError(page);
80 		}
81 		unlock_page(page);
82 	}
83 	bio_put(bio);
84 }
85 
86 static void f2fs_write_end_io(struct bio *bio)
87 {
88 	struct f2fs_sb_info *sbi = bio->bi_private;
89 	struct bio_vec *bvec;
90 	int i;
91 
92 	bio_for_each_segment_all(bvec, bio, i) {
93 		struct page *page = bvec->bv_page;
94 		enum count_type type = WB_DATA_TYPE(page);
95 
96 		if (IS_DUMMY_WRITTEN_PAGE(page)) {
97 			set_page_private(page, (unsigned long)NULL);
98 			ClearPagePrivate(page);
99 			unlock_page(page);
100 			mempool_free(page, sbi->write_io_dummy);
101 
102 			if (unlikely(bio->bi_error))
103 				f2fs_stop_checkpoint(sbi, true);
104 			continue;
105 		}
106 
107 		fscrypt_pullback_bio_page(&page, true);
108 
109 		if (unlikely(bio->bi_error)) {
110 			mapping_set_error(page->mapping, -EIO);
111 			f2fs_stop_checkpoint(sbi, true);
112 		}
113 		dec_page_count(sbi, type);
114 		clear_cold_data(page);
115 		end_page_writeback(page);
116 	}
117 	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
118 				wq_has_sleeper(&sbi->cp_wait))
119 		wake_up(&sbi->cp_wait);
120 
121 	bio_put(bio);
122 }
123 
124 /*
125  * Return true, if pre_bio's bdev is same as its target device.
126  */
127 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
128 				block_t blk_addr, struct bio *bio)
129 {
130 	struct block_device *bdev = sbi->sb->s_bdev;
131 	int i;
132 
133 	for (i = 0; i < sbi->s_ndevs; i++) {
134 		if (FDEV(i).start_blk <= blk_addr &&
135 					FDEV(i).end_blk >= blk_addr) {
136 			blk_addr -= FDEV(i).start_blk;
137 			bdev = FDEV(i).bdev;
138 			break;
139 		}
140 	}
141 	if (bio) {
142 		bio->bi_bdev = bdev;
143 		bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
144 	}
145 	return bdev;
146 }
147 
148 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
149 {
150 	int i;
151 
152 	for (i = 0; i < sbi->s_ndevs; i++)
153 		if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
154 			return i;
155 	return 0;
156 }
157 
158 static bool __same_bdev(struct f2fs_sb_info *sbi,
159 				block_t blk_addr, struct bio *bio)
160 {
161 	return f2fs_target_device(sbi, blk_addr, NULL) == bio->bi_bdev;
162 }
163 
164 /*
165  * Low-level block read/write IO operations.
166  */
167 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
168 				int npages, bool is_read)
169 {
170 	struct bio *bio;
171 
172 	bio = f2fs_bio_alloc(npages);
173 
174 	f2fs_target_device(sbi, blk_addr, bio);
175 	bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
176 	bio->bi_private = is_read ? NULL : sbi;
177 
178 	return bio;
179 }
180 
181 static inline void __submit_bio(struct f2fs_sb_info *sbi,
182 				struct bio *bio, enum page_type type)
183 {
184 	if (!is_read_io(bio_op(bio))) {
185 		unsigned int start;
186 
187 		if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
188 			current->plug && (type == DATA || type == NODE))
189 			blk_finish_plug(current->plug);
190 
191 		if (type != DATA && type != NODE)
192 			goto submit_io;
193 
194 		start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
195 		start %= F2FS_IO_SIZE(sbi);
196 
197 		if (start == 0)
198 			goto submit_io;
199 
200 		/* fill dummy pages */
201 		for (; start < F2FS_IO_SIZE(sbi); start++) {
202 			struct page *page =
203 				mempool_alloc(sbi->write_io_dummy,
204 					GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
205 			f2fs_bug_on(sbi, !page);
206 
207 			SetPagePrivate(page);
208 			set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
209 			lock_page(page);
210 			if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
211 				f2fs_bug_on(sbi, 1);
212 		}
213 		/*
214 		 * In the NODE case, we lose next block address chain. So, we
215 		 * need to do checkpoint in f2fs_sync_file.
216 		 */
217 		if (type == NODE)
218 			set_sbi_flag(sbi, SBI_NEED_CP);
219 	}
220 submit_io:
221 	if (is_read_io(bio_op(bio)))
222 		trace_f2fs_submit_read_bio(sbi->sb, type, bio);
223 	else
224 		trace_f2fs_submit_write_bio(sbi->sb, type, bio);
225 	submit_bio(bio);
226 }
227 
228 static void __submit_merged_bio(struct f2fs_bio_info *io)
229 {
230 	struct f2fs_io_info *fio = &io->fio;
231 
232 	if (!io->bio)
233 		return;
234 
235 	bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
236 
237 	if (is_read_io(fio->op))
238 		trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
239 	else
240 		trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
241 
242 	__submit_bio(io->sbi, io->bio, fio->type);
243 	io->bio = NULL;
244 }
245 
246 static bool __has_merged_page(struct f2fs_bio_info *io,
247 				struct inode *inode, nid_t ino, pgoff_t idx)
248 {
249 	struct bio_vec *bvec;
250 	struct page *target;
251 	int i;
252 
253 	if (!io->bio)
254 		return false;
255 
256 	if (!inode && !ino)
257 		return true;
258 
259 	bio_for_each_segment_all(bvec, io->bio, i) {
260 
261 		if (bvec->bv_page->mapping)
262 			target = bvec->bv_page;
263 		else
264 			target = fscrypt_control_page(bvec->bv_page);
265 
266 		if (idx != target->index)
267 			continue;
268 
269 		if (inode && inode == target->mapping->host)
270 			return true;
271 		if (ino && ino == ino_of_node(target))
272 			return true;
273 	}
274 
275 	return false;
276 }
277 
278 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
279 				nid_t ino, pgoff_t idx, enum page_type type)
280 {
281 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
282 	struct f2fs_bio_info *io = &sbi->write_io[btype];
283 	bool ret;
284 
285 	down_read(&io->io_rwsem);
286 	ret = __has_merged_page(io, inode, ino, idx);
287 	up_read(&io->io_rwsem);
288 	return ret;
289 }
290 
291 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
292 				struct inode *inode, nid_t ino, pgoff_t idx,
293 				enum page_type type, int rw)
294 {
295 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
296 	struct f2fs_bio_info *io;
297 
298 	io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
299 
300 	down_write(&io->io_rwsem);
301 
302 	if (!__has_merged_page(io, inode, ino, idx))
303 		goto out;
304 
305 	/* change META to META_FLUSH in the checkpoint procedure */
306 	if (type >= META_FLUSH) {
307 		io->fio.type = META_FLUSH;
308 		io->fio.op = REQ_OP_WRITE;
309 		io->fio.op_flags = REQ_META | REQ_PRIO;
310 		if (!test_opt(sbi, NOBARRIER))
311 			io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
312 	}
313 	__submit_merged_bio(io);
314 out:
315 	up_write(&io->io_rwsem);
316 }
317 
318 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
319 									int rw)
320 {
321 	__f2fs_submit_merged_bio(sbi, NULL, 0, 0, type, rw);
322 }
323 
324 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
325 				struct inode *inode, nid_t ino, pgoff_t idx,
326 				enum page_type type, int rw)
327 {
328 	if (has_merged_page(sbi, inode, ino, idx, type))
329 		__f2fs_submit_merged_bio(sbi, inode, ino, idx, type, rw);
330 }
331 
332 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
333 {
334 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
335 	f2fs_submit_merged_bio(sbi, NODE, WRITE);
336 	f2fs_submit_merged_bio(sbi, META, WRITE);
337 }
338 
339 /*
340  * Fill the locked page with data located in the block address.
341  * Return unlocked page.
342  */
343 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
344 {
345 	struct bio *bio;
346 	struct page *page = fio->encrypted_page ?
347 			fio->encrypted_page : fio->page;
348 
349 	trace_f2fs_submit_page_bio(page, fio);
350 	f2fs_trace_ios(fio, 0);
351 
352 	/* Allocate a new bio */
353 	bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
354 
355 	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
356 		bio_put(bio);
357 		return -EFAULT;
358 	}
359 	bio_set_op_attrs(bio, fio->op, fio->op_flags);
360 
361 	__submit_bio(fio->sbi, bio, fio->type);
362 	return 0;
363 }
364 
365 int f2fs_submit_page_mbio(struct f2fs_io_info *fio)
366 {
367 	struct f2fs_sb_info *sbi = fio->sbi;
368 	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
369 	struct f2fs_bio_info *io;
370 	bool is_read = is_read_io(fio->op);
371 	struct page *bio_page;
372 	int err = 0;
373 
374 	io = is_read ? &sbi->read_io : &sbi->write_io[btype];
375 
376 	if (fio->old_blkaddr != NEW_ADDR)
377 		verify_block_addr(sbi, fio->old_blkaddr);
378 	verify_block_addr(sbi, fio->new_blkaddr);
379 
380 	bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
381 
382 	/* set submitted = 1 as a return value */
383 	fio->submitted = 1;
384 
385 	if (!is_read)
386 		inc_page_count(sbi, WB_DATA_TYPE(bio_page));
387 
388 	down_write(&io->io_rwsem);
389 
390 	if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
391 	    (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
392 			!__same_bdev(sbi, fio->new_blkaddr, io->bio)))
393 		__submit_merged_bio(io);
394 alloc_new:
395 	if (io->bio == NULL) {
396 		if ((fio->type == DATA || fio->type == NODE) &&
397 				fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
398 			err = -EAGAIN;
399 			dec_page_count(sbi, WB_DATA_TYPE(bio_page));
400 			goto out_fail;
401 		}
402 		io->bio = __bio_alloc(sbi, fio->new_blkaddr,
403 						BIO_MAX_PAGES, is_read);
404 		io->fio = *fio;
405 	}
406 
407 	if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
408 							PAGE_SIZE) {
409 		__submit_merged_bio(io);
410 		goto alloc_new;
411 	}
412 
413 	io->last_block_in_bio = fio->new_blkaddr;
414 	f2fs_trace_ios(fio, 0);
415 out_fail:
416 	up_write(&io->io_rwsem);
417 	trace_f2fs_submit_page_mbio(fio->page, fio);
418 	return err;
419 }
420 
421 static void __set_data_blkaddr(struct dnode_of_data *dn)
422 {
423 	struct f2fs_node *rn = F2FS_NODE(dn->node_page);
424 	__le32 *addr_array;
425 
426 	/* Get physical address of data block */
427 	addr_array = blkaddr_in_node(rn);
428 	addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
429 }
430 
431 /*
432  * Lock ordering for the change of data block address:
433  * ->data_page
434  *  ->node_page
435  *    update block addresses in the node page
436  */
437 void set_data_blkaddr(struct dnode_of_data *dn)
438 {
439 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
440 	__set_data_blkaddr(dn);
441 	if (set_page_dirty(dn->node_page))
442 		dn->node_changed = true;
443 }
444 
445 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
446 {
447 	dn->data_blkaddr = blkaddr;
448 	set_data_blkaddr(dn);
449 	f2fs_update_extent_cache(dn);
450 }
451 
452 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
453 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
454 {
455 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
456 
457 	if (!count)
458 		return 0;
459 
460 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
461 		return -EPERM;
462 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
463 		return -ENOSPC;
464 
465 	trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
466 						dn->ofs_in_node, count);
467 
468 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
469 
470 	for (; count > 0; dn->ofs_in_node++) {
471 		block_t blkaddr =
472 			datablock_addr(dn->node_page, dn->ofs_in_node);
473 		if (blkaddr == NULL_ADDR) {
474 			dn->data_blkaddr = NEW_ADDR;
475 			__set_data_blkaddr(dn);
476 			count--;
477 		}
478 	}
479 
480 	if (set_page_dirty(dn->node_page))
481 		dn->node_changed = true;
482 	return 0;
483 }
484 
485 /* Should keep dn->ofs_in_node unchanged */
486 int reserve_new_block(struct dnode_of_data *dn)
487 {
488 	unsigned int ofs_in_node = dn->ofs_in_node;
489 	int ret;
490 
491 	ret = reserve_new_blocks(dn, 1);
492 	dn->ofs_in_node = ofs_in_node;
493 	return ret;
494 }
495 
496 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
497 {
498 	bool need_put = dn->inode_page ? false : true;
499 	int err;
500 
501 	err = get_dnode_of_data(dn, index, ALLOC_NODE);
502 	if (err)
503 		return err;
504 
505 	if (dn->data_blkaddr == NULL_ADDR)
506 		err = reserve_new_block(dn);
507 	if (err || need_put)
508 		f2fs_put_dnode(dn);
509 	return err;
510 }
511 
512 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
513 {
514 	struct extent_info ei  = {0,0,0};
515 	struct inode *inode = dn->inode;
516 
517 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
518 		dn->data_blkaddr = ei.blk + index - ei.fofs;
519 		return 0;
520 	}
521 
522 	return f2fs_reserve_block(dn, index);
523 }
524 
525 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
526 						int op_flags, bool for_write)
527 {
528 	struct address_space *mapping = inode->i_mapping;
529 	struct dnode_of_data dn;
530 	struct page *page;
531 	struct extent_info ei = {0,0,0};
532 	int err;
533 	struct f2fs_io_info fio = {
534 		.sbi = F2FS_I_SB(inode),
535 		.type = DATA,
536 		.op = REQ_OP_READ,
537 		.op_flags = op_flags,
538 		.encrypted_page = NULL,
539 	};
540 
541 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
542 		return read_mapping_page(mapping, index, NULL);
543 
544 	page = f2fs_grab_cache_page(mapping, index, for_write);
545 	if (!page)
546 		return ERR_PTR(-ENOMEM);
547 
548 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
549 		dn.data_blkaddr = ei.blk + index - ei.fofs;
550 		goto got_it;
551 	}
552 
553 	set_new_dnode(&dn, inode, NULL, NULL, 0);
554 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
555 	if (err)
556 		goto put_err;
557 	f2fs_put_dnode(&dn);
558 
559 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
560 		err = -ENOENT;
561 		goto put_err;
562 	}
563 got_it:
564 	if (PageUptodate(page)) {
565 		unlock_page(page);
566 		return page;
567 	}
568 
569 	/*
570 	 * A new dentry page is allocated but not able to be written, since its
571 	 * new inode page couldn't be allocated due to -ENOSPC.
572 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
573 	 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
574 	 */
575 	if (dn.data_blkaddr == NEW_ADDR) {
576 		zero_user_segment(page, 0, PAGE_SIZE);
577 		if (!PageUptodate(page))
578 			SetPageUptodate(page);
579 		unlock_page(page);
580 		return page;
581 	}
582 
583 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
584 	fio.page = page;
585 	err = f2fs_submit_page_bio(&fio);
586 	if (err)
587 		goto put_err;
588 	return page;
589 
590 put_err:
591 	f2fs_put_page(page, 1);
592 	return ERR_PTR(err);
593 }
594 
595 struct page *find_data_page(struct inode *inode, pgoff_t index)
596 {
597 	struct address_space *mapping = inode->i_mapping;
598 	struct page *page;
599 
600 	page = find_get_page(mapping, index);
601 	if (page && PageUptodate(page))
602 		return page;
603 	f2fs_put_page(page, 0);
604 
605 	page = get_read_data_page(inode, index, 0, false);
606 	if (IS_ERR(page))
607 		return page;
608 
609 	if (PageUptodate(page))
610 		return page;
611 
612 	wait_on_page_locked(page);
613 	if (unlikely(!PageUptodate(page))) {
614 		f2fs_put_page(page, 0);
615 		return ERR_PTR(-EIO);
616 	}
617 	return page;
618 }
619 
620 /*
621  * If it tries to access a hole, return an error.
622  * Because, the callers, functions in dir.c and GC, should be able to know
623  * whether this page exists or not.
624  */
625 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
626 							bool for_write)
627 {
628 	struct address_space *mapping = inode->i_mapping;
629 	struct page *page;
630 repeat:
631 	page = get_read_data_page(inode, index, 0, for_write);
632 	if (IS_ERR(page))
633 		return page;
634 
635 	/* wait for read completion */
636 	lock_page(page);
637 	if (unlikely(page->mapping != mapping)) {
638 		f2fs_put_page(page, 1);
639 		goto repeat;
640 	}
641 	if (unlikely(!PageUptodate(page))) {
642 		f2fs_put_page(page, 1);
643 		return ERR_PTR(-EIO);
644 	}
645 	return page;
646 }
647 
648 /*
649  * Caller ensures that this data page is never allocated.
650  * A new zero-filled data page is allocated in the page cache.
651  *
652  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
653  * f2fs_unlock_op().
654  * Note that, ipage is set only by make_empty_dir, and if any error occur,
655  * ipage should be released by this function.
656  */
657 struct page *get_new_data_page(struct inode *inode,
658 		struct page *ipage, pgoff_t index, bool new_i_size)
659 {
660 	struct address_space *mapping = inode->i_mapping;
661 	struct page *page;
662 	struct dnode_of_data dn;
663 	int err;
664 
665 	page = f2fs_grab_cache_page(mapping, index, true);
666 	if (!page) {
667 		/*
668 		 * before exiting, we should make sure ipage will be released
669 		 * if any error occur.
670 		 */
671 		f2fs_put_page(ipage, 1);
672 		return ERR_PTR(-ENOMEM);
673 	}
674 
675 	set_new_dnode(&dn, inode, ipage, NULL, 0);
676 	err = f2fs_reserve_block(&dn, index);
677 	if (err) {
678 		f2fs_put_page(page, 1);
679 		return ERR_PTR(err);
680 	}
681 	if (!ipage)
682 		f2fs_put_dnode(&dn);
683 
684 	if (PageUptodate(page))
685 		goto got_it;
686 
687 	if (dn.data_blkaddr == NEW_ADDR) {
688 		zero_user_segment(page, 0, PAGE_SIZE);
689 		if (!PageUptodate(page))
690 			SetPageUptodate(page);
691 	} else {
692 		f2fs_put_page(page, 1);
693 
694 		/* if ipage exists, blkaddr should be NEW_ADDR */
695 		f2fs_bug_on(F2FS_I_SB(inode), ipage);
696 		page = get_lock_data_page(inode, index, true);
697 		if (IS_ERR(page))
698 			return page;
699 	}
700 got_it:
701 	if (new_i_size && i_size_read(inode) <
702 				((loff_t)(index + 1) << PAGE_SHIFT))
703 		f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
704 	return page;
705 }
706 
707 static int __allocate_data_block(struct dnode_of_data *dn)
708 {
709 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
710 	struct f2fs_summary sum;
711 	struct node_info ni;
712 	pgoff_t fofs;
713 	blkcnt_t count = 1;
714 
715 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
716 		return -EPERM;
717 
718 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
719 	if (dn->data_blkaddr == NEW_ADDR)
720 		goto alloc;
721 
722 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
723 		return -ENOSPC;
724 
725 alloc:
726 	get_node_info(sbi, dn->nid, &ni);
727 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
728 
729 	allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
730 						&sum, CURSEG_WARM_DATA);
731 	set_data_blkaddr(dn);
732 
733 	/* update i_size */
734 	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
735 							dn->ofs_in_node;
736 	if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
737 		f2fs_i_size_write(dn->inode,
738 				((loff_t)(fofs + 1) << PAGE_SHIFT));
739 	return 0;
740 }
741 
742 static inline bool __force_buffered_io(struct inode *inode, int rw)
743 {
744 	return ((f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) ||
745 			(rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
746 			F2FS_I_SB(inode)->s_ndevs);
747 }
748 
749 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
750 {
751 	struct inode *inode = file_inode(iocb->ki_filp);
752 	struct f2fs_map_blocks map;
753 	int err = 0;
754 
755 	if (is_inode_flag_set(inode, FI_NO_PREALLOC))
756 		return 0;
757 
758 	map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
759 	map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
760 	if (map.m_len > map.m_lblk)
761 		map.m_len -= map.m_lblk;
762 	else
763 		map.m_len = 0;
764 
765 	map.m_next_pgofs = NULL;
766 
767 	if (iocb->ki_flags & IOCB_DIRECT) {
768 		err = f2fs_convert_inline_inode(inode);
769 		if (err)
770 			return err;
771 		return f2fs_map_blocks(inode, &map, 1,
772 			__force_buffered_io(inode, WRITE) ?
773 				F2FS_GET_BLOCK_PRE_AIO :
774 				F2FS_GET_BLOCK_PRE_DIO);
775 	}
776 	if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
777 		err = f2fs_convert_inline_inode(inode);
778 		if (err)
779 			return err;
780 	}
781 	if (!f2fs_has_inline_data(inode))
782 		return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
783 	return err;
784 }
785 
786 /*
787  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
788  * f2fs_map_blocks structure.
789  * If original data blocks are allocated, then give them to blockdev.
790  * Otherwise,
791  *     a. preallocate requested block addresses
792  *     b. do not use extent cache for better performance
793  *     c. give the block addresses to blockdev
794  */
795 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
796 						int create, int flag)
797 {
798 	unsigned int maxblocks = map->m_len;
799 	struct dnode_of_data dn;
800 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
801 	int mode = create ? ALLOC_NODE : LOOKUP_NODE;
802 	pgoff_t pgofs, end_offset, end;
803 	int err = 0, ofs = 1;
804 	unsigned int ofs_in_node, last_ofs_in_node;
805 	blkcnt_t prealloc;
806 	struct extent_info ei = {0,0,0};
807 	block_t blkaddr;
808 
809 	if (!maxblocks)
810 		return 0;
811 
812 	map->m_len = 0;
813 	map->m_flags = 0;
814 
815 	/* it only supports block size == page size */
816 	pgofs =	(pgoff_t)map->m_lblk;
817 	end = pgofs + maxblocks;
818 
819 	if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
820 		map->m_pblk = ei.blk + pgofs - ei.fofs;
821 		map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
822 		map->m_flags = F2FS_MAP_MAPPED;
823 		goto out;
824 	}
825 
826 next_dnode:
827 	if (create)
828 		f2fs_lock_op(sbi);
829 
830 	/* When reading holes, we need its node page */
831 	set_new_dnode(&dn, inode, NULL, NULL, 0);
832 	err = get_dnode_of_data(&dn, pgofs, mode);
833 	if (err) {
834 		if (flag == F2FS_GET_BLOCK_BMAP)
835 			map->m_pblk = 0;
836 		if (err == -ENOENT) {
837 			err = 0;
838 			if (map->m_next_pgofs)
839 				*map->m_next_pgofs =
840 					get_next_page_offset(&dn, pgofs);
841 		}
842 		goto unlock_out;
843 	}
844 
845 	prealloc = 0;
846 	last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
847 	end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
848 
849 next_block:
850 	blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
851 
852 	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
853 		if (create) {
854 			if (unlikely(f2fs_cp_error(sbi))) {
855 				err = -EIO;
856 				goto sync_out;
857 			}
858 			if (flag == F2FS_GET_BLOCK_PRE_AIO) {
859 				if (blkaddr == NULL_ADDR) {
860 					prealloc++;
861 					last_ofs_in_node = dn.ofs_in_node;
862 				}
863 			} else {
864 				err = __allocate_data_block(&dn);
865 				if (!err)
866 					set_inode_flag(inode, FI_APPEND_WRITE);
867 			}
868 			if (err)
869 				goto sync_out;
870 			map->m_flags |= F2FS_MAP_NEW;
871 			blkaddr = dn.data_blkaddr;
872 		} else {
873 			if (flag == F2FS_GET_BLOCK_BMAP) {
874 				map->m_pblk = 0;
875 				goto sync_out;
876 			}
877 			if (flag == F2FS_GET_BLOCK_FIEMAP &&
878 						blkaddr == NULL_ADDR) {
879 				if (map->m_next_pgofs)
880 					*map->m_next_pgofs = pgofs + 1;
881 			}
882 			if (flag != F2FS_GET_BLOCK_FIEMAP ||
883 						blkaddr != NEW_ADDR)
884 				goto sync_out;
885 		}
886 	}
887 
888 	if (flag == F2FS_GET_BLOCK_PRE_AIO)
889 		goto skip;
890 
891 	if (map->m_len == 0) {
892 		/* preallocated unwritten block should be mapped for fiemap. */
893 		if (blkaddr == NEW_ADDR)
894 			map->m_flags |= F2FS_MAP_UNWRITTEN;
895 		map->m_flags |= F2FS_MAP_MAPPED;
896 
897 		map->m_pblk = blkaddr;
898 		map->m_len = 1;
899 	} else if ((map->m_pblk != NEW_ADDR &&
900 			blkaddr == (map->m_pblk + ofs)) ||
901 			(map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
902 			flag == F2FS_GET_BLOCK_PRE_DIO) {
903 		ofs++;
904 		map->m_len++;
905 	} else {
906 		goto sync_out;
907 	}
908 
909 skip:
910 	dn.ofs_in_node++;
911 	pgofs++;
912 
913 	/* preallocate blocks in batch for one dnode page */
914 	if (flag == F2FS_GET_BLOCK_PRE_AIO &&
915 			(pgofs == end || dn.ofs_in_node == end_offset)) {
916 
917 		dn.ofs_in_node = ofs_in_node;
918 		err = reserve_new_blocks(&dn, prealloc);
919 		if (err)
920 			goto sync_out;
921 
922 		map->m_len += dn.ofs_in_node - ofs_in_node;
923 		if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
924 			err = -ENOSPC;
925 			goto sync_out;
926 		}
927 		dn.ofs_in_node = end_offset;
928 	}
929 
930 	if (pgofs >= end)
931 		goto sync_out;
932 	else if (dn.ofs_in_node < end_offset)
933 		goto next_block;
934 
935 	f2fs_put_dnode(&dn);
936 
937 	if (create) {
938 		f2fs_unlock_op(sbi);
939 		f2fs_balance_fs(sbi, dn.node_changed);
940 	}
941 	goto next_dnode;
942 
943 sync_out:
944 	f2fs_put_dnode(&dn);
945 unlock_out:
946 	if (create) {
947 		f2fs_unlock_op(sbi);
948 		f2fs_balance_fs(sbi, dn.node_changed);
949 	}
950 out:
951 	trace_f2fs_map_blocks(inode, map, err);
952 	return err;
953 }
954 
955 static int __get_data_block(struct inode *inode, sector_t iblock,
956 			struct buffer_head *bh, int create, int flag,
957 			pgoff_t *next_pgofs)
958 {
959 	struct f2fs_map_blocks map;
960 	int err;
961 
962 	map.m_lblk = iblock;
963 	map.m_len = bh->b_size >> inode->i_blkbits;
964 	map.m_next_pgofs = next_pgofs;
965 
966 	err = f2fs_map_blocks(inode, &map, create, flag);
967 	if (!err) {
968 		map_bh(bh, inode->i_sb, map.m_pblk);
969 		bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
970 		bh->b_size = (u64)map.m_len << inode->i_blkbits;
971 	}
972 	return err;
973 }
974 
975 static int get_data_block(struct inode *inode, sector_t iblock,
976 			struct buffer_head *bh_result, int create, int flag,
977 			pgoff_t *next_pgofs)
978 {
979 	return __get_data_block(inode, iblock, bh_result, create,
980 							flag, next_pgofs);
981 }
982 
983 static int get_data_block_dio(struct inode *inode, sector_t iblock,
984 			struct buffer_head *bh_result, int create)
985 {
986 	return __get_data_block(inode, iblock, bh_result, create,
987 						F2FS_GET_BLOCK_DIO, NULL);
988 }
989 
990 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
991 			struct buffer_head *bh_result, int create)
992 {
993 	/* Block number less than F2FS MAX BLOCKS */
994 	if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
995 		return -EFBIG;
996 
997 	return __get_data_block(inode, iblock, bh_result, create,
998 						F2FS_GET_BLOCK_BMAP, NULL);
999 }
1000 
1001 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1002 {
1003 	return (offset >> inode->i_blkbits);
1004 }
1005 
1006 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1007 {
1008 	return (blk << inode->i_blkbits);
1009 }
1010 
1011 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1012 		u64 start, u64 len)
1013 {
1014 	struct buffer_head map_bh;
1015 	sector_t start_blk, last_blk;
1016 	pgoff_t next_pgofs;
1017 	u64 logical = 0, phys = 0, size = 0;
1018 	u32 flags = 0;
1019 	int ret = 0;
1020 
1021 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1022 	if (ret)
1023 		return ret;
1024 
1025 	if (f2fs_has_inline_data(inode)) {
1026 		ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1027 		if (ret != -EAGAIN)
1028 			return ret;
1029 	}
1030 
1031 	inode_lock(inode);
1032 
1033 	if (logical_to_blk(inode, len) == 0)
1034 		len = blk_to_logical(inode, 1);
1035 
1036 	start_blk = logical_to_blk(inode, start);
1037 	last_blk = logical_to_blk(inode, start + len - 1);
1038 
1039 next:
1040 	memset(&map_bh, 0, sizeof(struct buffer_head));
1041 	map_bh.b_size = len;
1042 
1043 	ret = get_data_block(inode, start_blk, &map_bh, 0,
1044 					F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1045 	if (ret)
1046 		goto out;
1047 
1048 	/* HOLE */
1049 	if (!buffer_mapped(&map_bh)) {
1050 		start_blk = next_pgofs;
1051 
1052 		if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1053 					F2FS_I_SB(inode)->max_file_blocks))
1054 			goto prep_next;
1055 
1056 		flags |= FIEMAP_EXTENT_LAST;
1057 	}
1058 
1059 	if (size) {
1060 		if (f2fs_encrypted_inode(inode))
1061 			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1062 
1063 		ret = fiemap_fill_next_extent(fieinfo, logical,
1064 				phys, size, flags);
1065 	}
1066 
1067 	if (start_blk > last_blk || ret)
1068 		goto out;
1069 
1070 	logical = blk_to_logical(inode, start_blk);
1071 	phys = blk_to_logical(inode, map_bh.b_blocknr);
1072 	size = map_bh.b_size;
1073 	flags = 0;
1074 	if (buffer_unwritten(&map_bh))
1075 		flags = FIEMAP_EXTENT_UNWRITTEN;
1076 
1077 	start_blk += logical_to_blk(inode, size);
1078 
1079 prep_next:
1080 	cond_resched();
1081 	if (fatal_signal_pending(current))
1082 		ret = -EINTR;
1083 	else
1084 		goto next;
1085 out:
1086 	if (ret == 1)
1087 		ret = 0;
1088 
1089 	inode_unlock(inode);
1090 	return ret;
1091 }
1092 
1093 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
1094 				 unsigned nr_pages)
1095 {
1096 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1097 	struct fscrypt_ctx *ctx = NULL;
1098 	struct bio *bio;
1099 
1100 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1101 		ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1102 		if (IS_ERR(ctx))
1103 			return ERR_CAST(ctx);
1104 
1105 		/* wait the page to be moved by cleaning */
1106 		f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1107 	}
1108 
1109 	bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
1110 	if (!bio) {
1111 		if (ctx)
1112 			fscrypt_release_ctx(ctx);
1113 		return ERR_PTR(-ENOMEM);
1114 	}
1115 	f2fs_target_device(sbi, blkaddr, bio);
1116 	bio->bi_end_io = f2fs_read_end_io;
1117 	bio->bi_private = ctx;
1118 
1119 	return bio;
1120 }
1121 
1122 /*
1123  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1124  * Major change was from block_size == page_size in f2fs by default.
1125  */
1126 static int f2fs_mpage_readpages(struct address_space *mapping,
1127 			struct list_head *pages, struct page *page,
1128 			unsigned nr_pages)
1129 {
1130 	struct bio *bio = NULL;
1131 	unsigned page_idx;
1132 	sector_t last_block_in_bio = 0;
1133 	struct inode *inode = mapping->host;
1134 	const unsigned blkbits = inode->i_blkbits;
1135 	const unsigned blocksize = 1 << blkbits;
1136 	sector_t block_in_file;
1137 	sector_t last_block;
1138 	sector_t last_block_in_file;
1139 	sector_t block_nr;
1140 	struct f2fs_map_blocks map;
1141 
1142 	map.m_pblk = 0;
1143 	map.m_lblk = 0;
1144 	map.m_len = 0;
1145 	map.m_flags = 0;
1146 	map.m_next_pgofs = NULL;
1147 
1148 	for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1149 
1150 		prefetchw(&page->flags);
1151 		if (pages) {
1152 			page = list_last_entry(pages, struct page, lru);
1153 			list_del(&page->lru);
1154 			if (add_to_page_cache_lru(page, mapping,
1155 						  page->index,
1156 						  readahead_gfp_mask(mapping)))
1157 				goto next_page;
1158 		}
1159 
1160 		block_in_file = (sector_t)page->index;
1161 		last_block = block_in_file + nr_pages;
1162 		last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1163 								blkbits;
1164 		if (last_block > last_block_in_file)
1165 			last_block = last_block_in_file;
1166 
1167 		/*
1168 		 * Map blocks using the previous result first.
1169 		 */
1170 		if ((map.m_flags & F2FS_MAP_MAPPED) &&
1171 				block_in_file > map.m_lblk &&
1172 				block_in_file < (map.m_lblk + map.m_len))
1173 			goto got_it;
1174 
1175 		/*
1176 		 * Then do more f2fs_map_blocks() calls until we are
1177 		 * done with this page.
1178 		 */
1179 		map.m_flags = 0;
1180 
1181 		if (block_in_file < last_block) {
1182 			map.m_lblk = block_in_file;
1183 			map.m_len = last_block - block_in_file;
1184 
1185 			if (f2fs_map_blocks(inode, &map, 0,
1186 						F2FS_GET_BLOCK_READ))
1187 				goto set_error_page;
1188 		}
1189 got_it:
1190 		if ((map.m_flags & F2FS_MAP_MAPPED)) {
1191 			block_nr = map.m_pblk + block_in_file - map.m_lblk;
1192 			SetPageMappedToDisk(page);
1193 
1194 			if (!PageUptodate(page) && !cleancache_get_page(page)) {
1195 				SetPageUptodate(page);
1196 				goto confused;
1197 			}
1198 		} else {
1199 			zero_user_segment(page, 0, PAGE_SIZE);
1200 			if (!PageUptodate(page))
1201 				SetPageUptodate(page);
1202 			unlock_page(page);
1203 			goto next_page;
1204 		}
1205 
1206 		/*
1207 		 * This page will go to BIO.  Do we need to send this
1208 		 * BIO off first?
1209 		 */
1210 		if (bio && (last_block_in_bio != block_nr - 1 ||
1211 			!__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1212 submit_and_realloc:
1213 			__submit_bio(F2FS_I_SB(inode), bio, DATA);
1214 			bio = NULL;
1215 		}
1216 		if (bio == NULL) {
1217 			bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1218 			if (IS_ERR(bio)) {
1219 				bio = NULL;
1220 				goto set_error_page;
1221 			}
1222 			bio_set_op_attrs(bio, REQ_OP_READ, 0);
1223 		}
1224 
1225 		if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1226 			goto submit_and_realloc;
1227 
1228 		last_block_in_bio = block_nr;
1229 		goto next_page;
1230 set_error_page:
1231 		SetPageError(page);
1232 		zero_user_segment(page, 0, PAGE_SIZE);
1233 		unlock_page(page);
1234 		goto next_page;
1235 confused:
1236 		if (bio) {
1237 			__submit_bio(F2FS_I_SB(inode), bio, DATA);
1238 			bio = NULL;
1239 		}
1240 		unlock_page(page);
1241 next_page:
1242 		if (pages)
1243 			put_page(page);
1244 	}
1245 	BUG_ON(pages && !list_empty(pages));
1246 	if (bio)
1247 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
1248 	return 0;
1249 }
1250 
1251 static int f2fs_read_data_page(struct file *file, struct page *page)
1252 {
1253 	struct inode *inode = page->mapping->host;
1254 	int ret = -EAGAIN;
1255 
1256 	trace_f2fs_readpage(page, DATA);
1257 
1258 	/* If the file has inline data, try to read it directly */
1259 	if (f2fs_has_inline_data(inode))
1260 		ret = f2fs_read_inline_data(inode, page);
1261 	if (ret == -EAGAIN)
1262 		ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1263 	return ret;
1264 }
1265 
1266 static int f2fs_read_data_pages(struct file *file,
1267 			struct address_space *mapping,
1268 			struct list_head *pages, unsigned nr_pages)
1269 {
1270 	struct inode *inode = file->f_mapping->host;
1271 	struct page *page = list_last_entry(pages, struct page, lru);
1272 
1273 	trace_f2fs_readpages(inode, page, nr_pages);
1274 
1275 	/* If the file has inline data, skip readpages */
1276 	if (f2fs_has_inline_data(inode))
1277 		return 0;
1278 
1279 	return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1280 }
1281 
1282 int do_write_data_page(struct f2fs_io_info *fio)
1283 {
1284 	struct page *page = fio->page;
1285 	struct inode *inode = page->mapping->host;
1286 	struct dnode_of_data dn;
1287 	int err = 0;
1288 
1289 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1290 	err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1291 	if (err)
1292 		return err;
1293 
1294 	fio->old_blkaddr = dn.data_blkaddr;
1295 
1296 	/* This page is already truncated */
1297 	if (fio->old_blkaddr == NULL_ADDR) {
1298 		ClearPageUptodate(page);
1299 		goto out_writepage;
1300 	}
1301 
1302 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1303 		gfp_t gfp_flags = GFP_NOFS;
1304 
1305 		/* wait for GCed encrypted page writeback */
1306 		f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1307 							fio->old_blkaddr);
1308 retry_encrypt:
1309 		fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1310 							PAGE_SIZE, 0,
1311 							fio->page->index,
1312 							gfp_flags);
1313 		if (IS_ERR(fio->encrypted_page)) {
1314 			err = PTR_ERR(fio->encrypted_page);
1315 			if (err == -ENOMEM) {
1316 				/* flush pending ios and wait for a while */
1317 				f2fs_flush_merged_bios(F2FS_I_SB(inode));
1318 				congestion_wait(BLK_RW_ASYNC, HZ/50);
1319 				gfp_flags |= __GFP_NOFAIL;
1320 				err = 0;
1321 				goto retry_encrypt;
1322 			}
1323 			goto out_writepage;
1324 		}
1325 	}
1326 
1327 	set_page_writeback(page);
1328 
1329 	/*
1330 	 * If current allocation needs SSR,
1331 	 * it had better in-place writes for updated data.
1332 	 */
1333 	if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1334 			!is_cold_data(page) &&
1335 			!IS_ATOMIC_WRITTEN_PAGE(page) &&
1336 			need_inplace_update(inode))) {
1337 		rewrite_data_page(fio);
1338 		set_inode_flag(inode, FI_UPDATE_WRITE);
1339 		trace_f2fs_do_write_data_page(page, IPU);
1340 	} else {
1341 		write_data_page(&dn, fio);
1342 		trace_f2fs_do_write_data_page(page, OPU);
1343 		set_inode_flag(inode, FI_APPEND_WRITE);
1344 		if (page->index == 0)
1345 			set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1346 	}
1347 out_writepage:
1348 	f2fs_put_dnode(&dn);
1349 	return err;
1350 }
1351 
1352 static int __write_data_page(struct page *page, bool *submitted,
1353 				struct writeback_control *wbc)
1354 {
1355 	struct inode *inode = page->mapping->host;
1356 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1357 	loff_t i_size = i_size_read(inode);
1358 	const pgoff_t end_index = ((unsigned long long) i_size)
1359 							>> PAGE_SHIFT;
1360 	loff_t psize = (page->index + 1) << PAGE_SHIFT;
1361 	unsigned offset = 0;
1362 	bool need_balance_fs = false;
1363 	int err = 0;
1364 	struct f2fs_io_info fio = {
1365 		.sbi = sbi,
1366 		.type = DATA,
1367 		.op = REQ_OP_WRITE,
1368 		.op_flags = wbc_to_write_flags(wbc),
1369 		.page = page,
1370 		.encrypted_page = NULL,
1371 		.submitted = false,
1372 	};
1373 
1374 	trace_f2fs_writepage(page, DATA);
1375 
1376 	if (page->index < end_index)
1377 		goto write;
1378 
1379 	/*
1380 	 * If the offset is out-of-range of file size,
1381 	 * this page does not have to be written to disk.
1382 	 */
1383 	offset = i_size & (PAGE_SIZE - 1);
1384 	if ((page->index >= end_index + 1) || !offset)
1385 		goto out;
1386 
1387 	zero_user_segment(page, offset, PAGE_SIZE);
1388 write:
1389 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1390 		goto redirty_out;
1391 	if (f2fs_is_drop_cache(inode))
1392 		goto out;
1393 	/* we should not write 0'th page having journal header */
1394 	if (f2fs_is_volatile_file(inode) && (!page->index ||
1395 			(!wbc->for_reclaim &&
1396 			available_free_memory(sbi, BASE_CHECK))))
1397 		goto redirty_out;
1398 
1399 	/* we should bypass data pages to proceed the kworkder jobs */
1400 	if (unlikely(f2fs_cp_error(sbi))) {
1401 		mapping_set_error(page->mapping, -EIO);
1402 		goto out;
1403 	}
1404 
1405 	/* Dentry blocks are controlled by checkpoint */
1406 	if (S_ISDIR(inode->i_mode)) {
1407 		err = do_write_data_page(&fio);
1408 		goto done;
1409 	}
1410 
1411 	if (!wbc->for_reclaim)
1412 		need_balance_fs = true;
1413 	else if (has_not_enough_free_secs(sbi, 0, 0))
1414 		goto redirty_out;
1415 
1416 	err = -EAGAIN;
1417 	if (f2fs_has_inline_data(inode)) {
1418 		err = f2fs_write_inline_data(inode, page);
1419 		if (!err)
1420 			goto out;
1421 	}
1422 	f2fs_lock_op(sbi);
1423 	if (err == -EAGAIN)
1424 		err = do_write_data_page(&fio);
1425 	if (F2FS_I(inode)->last_disk_size < psize)
1426 		F2FS_I(inode)->last_disk_size = psize;
1427 	f2fs_unlock_op(sbi);
1428 done:
1429 	if (err && err != -ENOENT)
1430 		goto redirty_out;
1431 
1432 out:
1433 	inode_dec_dirty_pages(inode);
1434 	if (err)
1435 		ClearPageUptodate(page);
1436 
1437 	if (wbc->for_reclaim) {
1438 		f2fs_submit_merged_bio_cond(sbi, inode, 0, page->index,
1439 						DATA, WRITE);
1440 		remove_dirty_inode(inode);
1441 		submitted = NULL;
1442 	}
1443 
1444 	unlock_page(page);
1445 	f2fs_balance_fs(sbi, need_balance_fs);
1446 
1447 	if (unlikely(f2fs_cp_error(sbi))) {
1448 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
1449 		submitted = NULL;
1450 	}
1451 
1452 	if (submitted)
1453 		*submitted = fio.submitted;
1454 
1455 	return 0;
1456 
1457 redirty_out:
1458 	redirty_page_for_writepage(wbc, page);
1459 	if (!err)
1460 		return AOP_WRITEPAGE_ACTIVATE;
1461 	unlock_page(page);
1462 	return err;
1463 }
1464 
1465 static int f2fs_write_data_page(struct page *page,
1466 					struct writeback_control *wbc)
1467 {
1468 	return __write_data_page(page, NULL, wbc);
1469 }
1470 
1471 /*
1472  * This function was copied from write_cche_pages from mm/page-writeback.c.
1473  * The major change is making write step of cold data page separately from
1474  * warm/hot data page.
1475  */
1476 static int f2fs_write_cache_pages(struct address_space *mapping,
1477 					struct writeback_control *wbc)
1478 {
1479 	int ret = 0;
1480 	int done = 0;
1481 	struct pagevec pvec;
1482 	int nr_pages;
1483 	pgoff_t uninitialized_var(writeback_index);
1484 	pgoff_t index;
1485 	pgoff_t end;		/* Inclusive */
1486 	pgoff_t done_index;
1487 	pgoff_t last_idx = ULONG_MAX;
1488 	int cycled;
1489 	int range_whole = 0;
1490 	int tag;
1491 
1492 	pagevec_init(&pvec, 0);
1493 
1494 	if (wbc->range_cyclic) {
1495 		writeback_index = mapping->writeback_index; /* prev offset */
1496 		index = writeback_index;
1497 		if (index == 0)
1498 			cycled = 1;
1499 		else
1500 			cycled = 0;
1501 		end = -1;
1502 	} else {
1503 		index = wbc->range_start >> PAGE_SHIFT;
1504 		end = wbc->range_end >> PAGE_SHIFT;
1505 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1506 			range_whole = 1;
1507 		cycled = 1; /* ignore range_cyclic tests */
1508 	}
1509 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1510 		tag = PAGECACHE_TAG_TOWRITE;
1511 	else
1512 		tag = PAGECACHE_TAG_DIRTY;
1513 retry:
1514 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1515 		tag_pages_for_writeback(mapping, index, end);
1516 	done_index = index;
1517 	while (!done && (index <= end)) {
1518 		int i;
1519 
1520 		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1521 			      min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1522 		if (nr_pages == 0)
1523 			break;
1524 
1525 		for (i = 0; i < nr_pages; i++) {
1526 			struct page *page = pvec.pages[i];
1527 			bool submitted = false;
1528 
1529 			if (page->index > end) {
1530 				done = 1;
1531 				break;
1532 			}
1533 
1534 			done_index = page->index;
1535 
1536 			lock_page(page);
1537 
1538 			if (unlikely(page->mapping != mapping)) {
1539 continue_unlock:
1540 				unlock_page(page);
1541 				continue;
1542 			}
1543 
1544 			if (!PageDirty(page)) {
1545 				/* someone wrote it for us */
1546 				goto continue_unlock;
1547 			}
1548 
1549 			if (PageWriteback(page)) {
1550 				if (wbc->sync_mode != WB_SYNC_NONE)
1551 					f2fs_wait_on_page_writeback(page,
1552 								DATA, true);
1553 				else
1554 					goto continue_unlock;
1555 			}
1556 
1557 			BUG_ON(PageWriteback(page));
1558 			if (!clear_page_dirty_for_io(page))
1559 				goto continue_unlock;
1560 
1561 			ret = __write_data_page(page, &submitted, wbc);
1562 			if (unlikely(ret)) {
1563 				/*
1564 				 * keep nr_to_write, since vfs uses this to
1565 				 * get # of written pages.
1566 				 */
1567 				if (ret == AOP_WRITEPAGE_ACTIVATE) {
1568 					unlock_page(page);
1569 					ret = 0;
1570 					continue;
1571 				}
1572 				done_index = page->index + 1;
1573 				done = 1;
1574 				break;
1575 			} else if (submitted) {
1576 				last_idx = page->index;
1577 			}
1578 
1579 			if (--wbc->nr_to_write <= 0 &&
1580 			    wbc->sync_mode == WB_SYNC_NONE) {
1581 				done = 1;
1582 				break;
1583 			}
1584 		}
1585 		pagevec_release(&pvec);
1586 		cond_resched();
1587 	}
1588 
1589 	if (!cycled && !done) {
1590 		cycled = 1;
1591 		index = 0;
1592 		end = writeback_index - 1;
1593 		goto retry;
1594 	}
1595 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1596 		mapping->writeback_index = done_index;
1597 
1598 	if (last_idx != ULONG_MAX)
1599 		f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1600 						0, last_idx, DATA, WRITE);
1601 
1602 	return ret;
1603 }
1604 
1605 static int f2fs_write_data_pages(struct address_space *mapping,
1606 			    struct writeback_control *wbc)
1607 {
1608 	struct inode *inode = mapping->host;
1609 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1610 	struct blk_plug plug;
1611 	int ret;
1612 
1613 	/* deal with chardevs and other special file */
1614 	if (!mapping->a_ops->writepage)
1615 		return 0;
1616 
1617 	/* skip writing if there is no dirty page in this inode */
1618 	if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1619 		return 0;
1620 
1621 	if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1622 			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1623 			available_free_memory(sbi, DIRTY_DENTS))
1624 		goto skip_write;
1625 
1626 	/* skip writing during file defragment */
1627 	if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1628 		goto skip_write;
1629 
1630 	/* during POR, we don't need to trigger writepage at all. */
1631 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1632 		goto skip_write;
1633 
1634 	trace_f2fs_writepages(mapping->host, wbc, DATA);
1635 
1636 	blk_start_plug(&plug);
1637 	ret = f2fs_write_cache_pages(mapping, wbc);
1638 	blk_finish_plug(&plug);
1639 	/*
1640 	 * if some pages were truncated, we cannot guarantee its mapping->host
1641 	 * to detect pending bios.
1642 	 */
1643 
1644 	remove_dirty_inode(inode);
1645 	return ret;
1646 
1647 skip_write:
1648 	wbc->pages_skipped += get_dirty_pages(inode);
1649 	trace_f2fs_writepages(mapping->host, wbc, DATA);
1650 	return 0;
1651 }
1652 
1653 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1654 {
1655 	struct inode *inode = mapping->host;
1656 	loff_t i_size = i_size_read(inode);
1657 
1658 	if (to > i_size) {
1659 		truncate_pagecache(inode, i_size);
1660 		truncate_blocks(inode, i_size, true);
1661 	}
1662 }
1663 
1664 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1665 			struct page *page, loff_t pos, unsigned len,
1666 			block_t *blk_addr, bool *node_changed)
1667 {
1668 	struct inode *inode = page->mapping->host;
1669 	pgoff_t index = page->index;
1670 	struct dnode_of_data dn;
1671 	struct page *ipage;
1672 	bool locked = false;
1673 	struct extent_info ei = {0,0,0};
1674 	int err = 0;
1675 
1676 	/*
1677 	 * we already allocated all the blocks, so we don't need to get
1678 	 * the block addresses when there is no need to fill the page.
1679 	 */
1680 	if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1681 			!is_inode_flag_set(inode, FI_NO_PREALLOC))
1682 		return 0;
1683 
1684 	if (f2fs_has_inline_data(inode) ||
1685 			(pos & PAGE_MASK) >= i_size_read(inode)) {
1686 		f2fs_lock_op(sbi);
1687 		locked = true;
1688 	}
1689 restart:
1690 	/* check inline_data */
1691 	ipage = get_node_page(sbi, inode->i_ino);
1692 	if (IS_ERR(ipage)) {
1693 		err = PTR_ERR(ipage);
1694 		goto unlock_out;
1695 	}
1696 
1697 	set_new_dnode(&dn, inode, ipage, ipage, 0);
1698 
1699 	if (f2fs_has_inline_data(inode)) {
1700 		if (pos + len <= MAX_INLINE_DATA) {
1701 			read_inline_data(page, ipage);
1702 			set_inode_flag(inode, FI_DATA_EXIST);
1703 			if (inode->i_nlink)
1704 				set_inline_node(ipage);
1705 		} else {
1706 			err = f2fs_convert_inline_page(&dn, page);
1707 			if (err)
1708 				goto out;
1709 			if (dn.data_blkaddr == NULL_ADDR)
1710 				err = f2fs_get_block(&dn, index);
1711 		}
1712 	} else if (locked) {
1713 		err = f2fs_get_block(&dn, index);
1714 	} else {
1715 		if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1716 			dn.data_blkaddr = ei.blk + index - ei.fofs;
1717 		} else {
1718 			/* hole case */
1719 			err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1720 			if (err || dn.data_blkaddr == NULL_ADDR) {
1721 				f2fs_put_dnode(&dn);
1722 				f2fs_lock_op(sbi);
1723 				locked = true;
1724 				goto restart;
1725 			}
1726 		}
1727 	}
1728 
1729 	/* convert_inline_page can make node_changed */
1730 	*blk_addr = dn.data_blkaddr;
1731 	*node_changed = dn.node_changed;
1732 out:
1733 	f2fs_put_dnode(&dn);
1734 unlock_out:
1735 	if (locked)
1736 		f2fs_unlock_op(sbi);
1737 	return err;
1738 }
1739 
1740 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1741 		loff_t pos, unsigned len, unsigned flags,
1742 		struct page **pagep, void **fsdata)
1743 {
1744 	struct inode *inode = mapping->host;
1745 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1746 	struct page *page = NULL;
1747 	pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1748 	bool need_balance = false;
1749 	block_t blkaddr = NULL_ADDR;
1750 	int err = 0;
1751 
1752 	trace_f2fs_write_begin(inode, pos, len, flags);
1753 
1754 	/*
1755 	 * We should check this at this moment to avoid deadlock on inode page
1756 	 * and #0 page. The locking rule for inline_data conversion should be:
1757 	 * lock_page(page #0) -> lock_page(inode_page)
1758 	 */
1759 	if (index != 0) {
1760 		err = f2fs_convert_inline_inode(inode);
1761 		if (err)
1762 			goto fail;
1763 	}
1764 repeat:
1765 	/*
1766 	 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1767 	 * wait_for_stable_page. Will wait that below with our IO control.
1768 	 */
1769 	page = pagecache_get_page(mapping, index,
1770 				FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1771 	if (!page) {
1772 		err = -ENOMEM;
1773 		goto fail;
1774 	}
1775 
1776 	*pagep = page;
1777 
1778 	err = prepare_write_begin(sbi, page, pos, len,
1779 					&blkaddr, &need_balance);
1780 	if (err)
1781 		goto fail;
1782 
1783 	if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1784 		unlock_page(page);
1785 		f2fs_balance_fs(sbi, true);
1786 		lock_page(page);
1787 		if (page->mapping != mapping) {
1788 			/* The page got truncated from under us */
1789 			f2fs_put_page(page, 1);
1790 			goto repeat;
1791 		}
1792 	}
1793 
1794 	f2fs_wait_on_page_writeback(page, DATA, false);
1795 
1796 	/* wait for GCed encrypted page writeback */
1797 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1798 		f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1799 
1800 	if (len == PAGE_SIZE || PageUptodate(page))
1801 		return 0;
1802 
1803 	if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
1804 		zero_user_segment(page, len, PAGE_SIZE);
1805 		return 0;
1806 	}
1807 
1808 	if (blkaddr == NEW_ADDR) {
1809 		zero_user_segment(page, 0, PAGE_SIZE);
1810 		SetPageUptodate(page);
1811 	} else {
1812 		struct bio *bio;
1813 
1814 		bio = f2fs_grab_bio(inode, blkaddr, 1);
1815 		if (IS_ERR(bio)) {
1816 			err = PTR_ERR(bio);
1817 			goto fail;
1818 		}
1819 		bio->bi_opf = REQ_OP_READ;
1820 		if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1821 			bio_put(bio);
1822 			err = -EFAULT;
1823 			goto fail;
1824 		}
1825 
1826 		__submit_bio(sbi, bio, DATA);
1827 
1828 		lock_page(page);
1829 		if (unlikely(page->mapping != mapping)) {
1830 			f2fs_put_page(page, 1);
1831 			goto repeat;
1832 		}
1833 		if (unlikely(!PageUptodate(page))) {
1834 			err = -EIO;
1835 			goto fail;
1836 		}
1837 	}
1838 	return 0;
1839 
1840 fail:
1841 	f2fs_put_page(page, 1);
1842 	f2fs_write_failed(mapping, pos + len);
1843 	return err;
1844 }
1845 
1846 static int f2fs_write_end(struct file *file,
1847 			struct address_space *mapping,
1848 			loff_t pos, unsigned len, unsigned copied,
1849 			struct page *page, void *fsdata)
1850 {
1851 	struct inode *inode = page->mapping->host;
1852 
1853 	trace_f2fs_write_end(inode, pos, len, copied);
1854 
1855 	/*
1856 	 * This should be come from len == PAGE_SIZE, and we expect copied
1857 	 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1858 	 * let generic_perform_write() try to copy data again through copied=0.
1859 	 */
1860 	if (!PageUptodate(page)) {
1861 		if (unlikely(copied != len))
1862 			copied = 0;
1863 		else
1864 			SetPageUptodate(page);
1865 	}
1866 	if (!copied)
1867 		goto unlock_out;
1868 
1869 	set_page_dirty(page);
1870 
1871 	if (pos + copied > i_size_read(inode))
1872 		f2fs_i_size_write(inode, pos + copied);
1873 unlock_out:
1874 	f2fs_put_page(page, 1);
1875 	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1876 	return copied;
1877 }
1878 
1879 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1880 			   loff_t offset)
1881 {
1882 	unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1883 
1884 	if (offset & blocksize_mask)
1885 		return -EINVAL;
1886 
1887 	if (iov_iter_alignment(iter) & blocksize_mask)
1888 		return -EINVAL;
1889 
1890 	return 0;
1891 }
1892 
1893 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1894 {
1895 	struct address_space *mapping = iocb->ki_filp->f_mapping;
1896 	struct inode *inode = mapping->host;
1897 	size_t count = iov_iter_count(iter);
1898 	loff_t offset = iocb->ki_pos;
1899 	int rw = iov_iter_rw(iter);
1900 	int err;
1901 
1902 	err = check_direct_IO(inode, iter, offset);
1903 	if (err)
1904 		return err;
1905 
1906 	if (__force_buffered_io(inode, rw))
1907 		return 0;
1908 
1909 	trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1910 
1911 	down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1912 	err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1913 	up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1914 
1915 	if (rw == WRITE) {
1916 		if (err > 0)
1917 			set_inode_flag(inode, FI_UPDATE_WRITE);
1918 		else if (err < 0)
1919 			f2fs_write_failed(mapping, offset + count);
1920 	}
1921 
1922 	trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1923 
1924 	return err;
1925 }
1926 
1927 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1928 							unsigned int length)
1929 {
1930 	struct inode *inode = page->mapping->host;
1931 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1932 
1933 	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1934 		(offset % PAGE_SIZE || length != PAGE_SIZE))
1935 		return;
1936 
1937 	if (PageDirty(page)) {
1938 		if (inode->i_ino == F2FS_META_INO(sbi)) {
1939 			dec_page_count(sbi, F2FS_DIRTY_META);
1940 		} else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
1941 			dec_page_count(sbi, F2FS_DIRTY_NODES);
1942 		} else {
1943 			inode_dec_dirty_pages(inode);
1944 			remove_dirty_inode(inode);
1945 		}
1946 	}
1947 
1948 	/* This is atomic written page, keep Private */
1949 	if (IS_ATOMIC_WRITTEN_PAGE(page))
1950 		return;
1951 
1952 	set_page_private(page, 0);
1953 	ClearPagePrivate(page);
1954 }
1955 
1956 int f2fs_release_page(struct page *page, gfp_t wait)
1957 {
1958 	/* If this is dirty page, keep PagePrivate */
1959 	if (PageDirty(page))
1960 		return 0;
1961 
1962 	/* This is atomic written page, keep Private */
1963 	if (IS_ATOMIC_WRITTEN_PAGE(page))
1964 		return 0;
1965 
1966 	set_page_private(page, 0);
1967 	ClearPagePrivate(page);
1968 	return 1;
1969 }
1970 
1971 /*
1972  * This was copied from __set_page_dirty_buffers which gives higher performance
1973  * in very high speed storages. (e.g., pmem)
1974  */
1975 void f2fs_set_page_dirty_nobuffers(struct page *page)
1976 {
1977 	struct address_space *mapping = page->mapping;
1978 	unsigned long flags;
1979 
1980 	if (unlikely(!mapping))
1981 		return;
1982 
1983 	spin_lock(&mapping->private_lock);
1984 	lock_page_memcg(page);
1985 	SetPageDirty(page);
1986 	spin_unlock(&mapping->private_lock);
1987 
1988 	spin_lock_irqsave(&mapping->tree_lock, flags);
1989 	WARN_ON_ONCE(!PageUptodate(page));
1990 	account_page_dirtied(page, mapping);
1991 	radix_tree_tag_set(&mapping->page_tree,
1992 			page_index(page), PAGECACHE_TAG_DIRTY);
1993 	spin_unlock_irqrestore(&mapping->tree_lock, flags);
1994 	unlock_page_memcg(page);
1995 
1996 	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1997 	return;
1998 }
1999 
2000 static int f2fs_set_data_page_dirty(struct page *page)
2001 {
2002 	struct address_space *mapping = page->mapping;
2003 	struct inode *inode = mapping->host;
2004 
2005 	trace_f2fs_set_page_dirty(page, DATA);
2006 
2007 	if (!PageUptodate(page))
2008 		SetPageUptodate(page);
2009 
2010 	if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2011 		if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2012 			register_inmem_page(inode, page);
2013 			return 1;
2014 		}
2015 		/*
2016 		 * Previously, this page has been registered, we just
2017 		 * return here.
2018 		 */
2019 		return 0;
2020 	}
2021 
2022 	if (!PageDirty(page)) {
2023 		f2fs_set_page_dirty_nobuffers(page);
2024 		update_dirty_page(inode, page);
2025 		return 1;
2026 	}
2027 	return 0;
2028 }
2029 
2030 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2031 {
2032 	struct inode *inode = mapping->host;
2033 
2034 	if (f2fs_has_inline_data(inode))
2035 		return 0;
2036 
2037 	/* make sure allocating whole blocks */
2038 	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2039 		filemap_write_and_wait(mapping);
2040 
2041 	return generic_block_bmap(mapping, block, get_data_block_bmap);
2042 }
2043 
2044 #ifdef CONFIG_MIGRATION
2045 #include <linux/migrate.h>
2046 
2047 int f2fs_migrate_page(struct address_space *mapping,
2048 		struct page *newpage, struct page *page, enum migrate_mode mode)
2049 {
2050 	int rc, extra_count;
2051 	struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2052 	bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2053 
2054 	BUG_ON(PageWriteback(page));
2055 
2056 	/* migrating an atomic written page is safe with the inmem_lock hold */
2057 	if (atomic_written && !mutex_trylock(&fi->inmem_lock))
2058 		return -EAGAIN;
2059 
2060 	/*
2061 	 * A reference is expected if PagePrivate set when move mapping,
2062 	 * however F2FS breaks this for maintaining dirty page counts when
2063 	 * truncating pages. So here adjusting the 'extra_count' make it work.
2064 	 */
2065 	extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2066 	rc = migrate_page_move_mapping(mapping, newpage,
2067 				page, NULL, mode, extra_count);
2068 	if (rc != MIGRATEPAGE_SUCCESS) {
2069 		if (atomic_written)
2070 			mutex_unlock(&fi->inmem_lock);
2071 		return rc;
2072 	}
2073 
2074 	if (atomic_written) {
2075 		struct inmem_pages *cur;
2076 		list_for_each_entry(cur, &fi->inmem_pages, list)
2077 			if (cur->page == page) {
2078 				cur->page = newpage;
2079 				break;
2080 			}
2081 		mutex_unlock(&fi->inmem_lock);
2082 		put_page(page);
2083 		get_page(newpage);
2084 	}
2085 
2086 	if (PagePrivate(page))
2087 		SetPagePrivate(newpage);
2088 	set_page_private(newpage, page_private(page));
2089 
2090 	migrate_page_copy(newpage, page);
2091 
2092 	return MIGRATEPAGE_SUCCESS;
2093 }
2094 #endif
2095 
2096 const struct address_space_operations f2fs_dblock_aops = {
2097 	.readpage	= f2fs_read_data_page,
2098 	.readpages	= f2fs_read_data_pages,
2099 	.writepage	= f2fs_write_data_page,
2100 	.writepages	= f2fs_write_data_pages,
2101 	.write_begin	= f2fs_write_begin,
2102 	.write_end	= f2fs_write_end,
2103 	.set_page_dirty	= f2fs_set_data_page_dirty,
2104 	.invalidatepage	= f2fs_invalidate_page,
2105 	.releasepage	= f2fs_release_page,
2106 	.direct_IO	= f2fs_direct_IO,
2107 	.bmap		= f2fs_bmap,
2108 #ifdef CONFIG_MIGRATION
2109 	.migratepage    = f2fs_migrate_page,
2110 #endif
2111 };
2112