xref: /openbmc/linux/fs/f2fs/data.c (revision 55523519bc7227e651fd4febeb3aafdd22b8af1c)
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 		f2fs_show_injection_info(FAULT_IO);
60 		bio->bi_error = -EIO;
61 	}
62 #endif
63 
64 	if (f2fs_bio_encrypted(bio)) {
65 		if (bio->bi_error) {
66 			fscrypt_release_ctx(bio->bi_private);
67 		} else {
68 			fscrypt_decrypt_bio_pages(bio->bi_private, bio);
69 			return;
70 		}
71 	}
72 
73 	bio_for_each_segment_all(bvec, bio, i) {
74 		struct page *page = bvec->bv_page;
75 
76 		if (!bio->bi_error) {
77 			if (!PageUptodate(page))
78 				SetPageUptodate(page);
79 		} else {
80 			ClearPageUptodate(page);
81 			SetPageError(page);
82 		}
83 		unlock_page(page);
84 	}
85 	bio_put(bio);
86 }
87 
88 static void f2fs_write_end_io(struct bio *bio)
89 {
90 	struct f2fs_sb_info *sbi = bio->bi_private;
91 	struct bio_vec *bvec;
92 	int i;
93 
94 	bio_for_each_segment_all(bvec, bio, i) {
95 		struct page *page = bvec->bv_page;
96 		enum count_type type = WB_DATA_TYPE(page);
97 
98 		if (IS_DUMMY_WRITTEN_PAGE(page)) {
99 			set_page_private(page, (unsigned long)NULL);
100 			ClearPagePrivate(page);
101 			unlock_page(page);
102 			mempool_free(page, sbi->write_io_dummy);
103 
104 			if (unlikely(bio->bi_error))
105 				f2fs_stop_checkpoint(sbi, true);
106 			continue;
107 		}
108 
109 		fscrypt_pullback_bio_page(&page, true);
110 
111 		if (unlikely(bio->bi_error)) {
112 			mapping_set_error(page->mapping, -EIO);
113 			f2fs_stop_checkpoint(sbi, true);
114 		}
115 		dec_page_count(sbi, type);
116 		clear_cold_data(page);
117 		end_page_writeback(page);
118 	}
119 	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
120 				wq_has_sleeper(&sbi->cp_wait))
121 		wake_up(&sbi->cp_wait);
122 
123 	bio_put(bio);
124 }
125 
126 /*
127  * Return true, if pre_bio's bdev is same as its target device.
128  */
129 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
130 				block_t blk_addr, struct bio *bio)
131 {
132 	struct block_device *bdev = sbi->sb->s_bdev;
133 	int i;
134 
135 	for (i = 0; i < sbi->s_ndevs; i++) {
136 		if (FDEV(i).start_blk <= blk_addr &&
137 					FDEV(i).end_blk >= blk_addr) {
138 			blk_addr -= FDEV(i).start_blk;
139 			bdev = FDEV(i).bdev;
140 			break;
141 		}
142 	}
143 	if (bio) {
144 		bio->bi_bdev = bdev;
145 		bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
146 	}
147 	return bdev;
148 }
149 
150 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
151 {
152 	int i;
153 
154 	for (i = 0; i < sbi->s_ndevs; i++)
155 		if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
156 			return i;
157 	return 0;
158 }
159 
160 static bool __same_bdev(struct f2fs_sb_info *sbi,
161 				block_t blk_addr, struct bio *bio)
162 {
163 	return f2fs_target_device(sbi, blk_addr, NULL) == bio->bi_bdev;
164 }
165 
166 /*
167  * Low-level block read/write IO operations.
168  */
169 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
170 				int npages, bool is_read)
171 {
172 	struct bio *bio;
173 
174 	bio = f2fs_bio_alloc(npages);
175 
176 	f2fs_target_device(sbi, blk_addr, bio);
177 	bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
178 	bio->bi_private = is_read ? NULL : sbi;
179 
180 	return bio;
181 }
182 
183 static inline void __submit_bio(struct f2fs_sb_info *sbi,
184 				struct bio *bio, enum page_type type)
185 {
186 	if (!is_read_io(bio_op(bio))) {
187 		unsigned int start;
188 
189 		if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
190 			current->plug && (type == DATA || type == NODE))
191 			blk_finish_plug(current->plug);
192 
193 		if (type != DATA && type != NODE)
194 			goto submit_io;
195 
196 		start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
197 		start %= F2FS_IO_SIZE(sbi);
198 
199 		if (start == 0)
200 			goto submit_io;
201 
202 		/* fill dummy pages */
203 		for (; start < F2FS_IO_SIZE(sbi); start++) {
204 			struct page *page =
205 				mempool_alloc(sbi->write_io_dummy,
206 					GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
207 			f2fs_bug_on(sbi, !page);
208 
209 			SetPagePrivate(page);
210 			set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
211 			lock_page(page);
212 			if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
213 				f2fs_bug_on(sbi, 1);
214 		}
215 		/*
216 		 * In the NODE case, we lose next block address chain. So, we
217 		 * need to do checkpoint in f2fs_sync_file.
218 		 */
219 		if (type == NODE)
220 			set_sbi_flag(sbi, SBI_NEED_CP);
221 	}
222 submit_io:
223 	if (is_read_io(bio_op(bio)))
224 		trace_f2fs_submit_read_bio(sbi->sb, type, bio);
225 	else
226 		trace_f2fs_submit_write_bio(sbi->sb, type, bio);
227 	submit_bio(bio);
228 }
229 
230 static void __submit_merged_bio(struct f2fs_bio_info *io)
231 {
232 	struct f2fs_io_info *fio = &io->fio;
233 
234 	if (!io->bio)
235 		return;
236 
237 	bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
238 
239 	if (is_read_io(fio->op))
240 		trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
241 	else
242 		trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
243 
244 	__submit_bio(io->sbi, io->bio, fio->type);
245 	io->bio = NULL;
246 }
247 
248 static bool __has_merged_page(struct f2fs_bio_info *io,
249 				struct inode *inode, nid_t ino, pgoff_t idx)
250 {
251 	struct bio_vec *bvec;
252 	struct page *target;
253 	int i;
254 
255 	if (!io->bio)
256 		return false;
257 
258 	if (!inode && !ino)
259 		return true;
260 
261 	bio_for_each_segment_all(bvec, io->bio, i) {
262 
263 		if (bvec->bv_page->mapping)
264 			target = bvec->bv_page;
265 		else
266 			target = fscrypt_control_page(bvec->bv_page);
267 
268 		if (idx != target->index)
269 			continue;
270 
271 		if (inode && inode == target->mapping->host)
272 			return true;
273 		if (ino && ino == ino_of_node(target))
274 			return true;
275 	}
276 
277 	return false;
278 }
279 
280 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
281 				nid_t ino, pgoff_t idx, enum page_type type)
282 {
283 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
284 	struct f2fs_bio_info *io = &sbi->write_io[btype];
285 	bool ret;
286 
287 	down_read(&io->io_rwsem);
288 	ret = __has_merged_page(io, inode, ino, idx);
289 	up_read(&io->io_rwsem);
290 	return ret;
291 }
292 
293 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
294 				struct inode *inode, nid_t ino, pgoff_t idx,
295 				enum page_type type, int rw)
296 {
297 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
298 	struct f2fs_bio_info *io;
299 
300 	io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
301 
302 	down_write(&io->io_rwsem);
303 
304 	if (!__has_merged_page(io, inode, ino, idx))
305 		goto out;
306 
307 	/* change META to META_FLUSH in the checkpoint procedure */
308 	if (type >= META_FLUSH) {
309 		io->fio.type = META_FLUSH;
310 		io->fio.op = REQ_OP_WRITE;
311 		io->fio.op_flags = REQ_META | REQ_PRIO;
312 		if (!test_opt(sbi, NOBARRIER))
313 			io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
314 	}
315 	__submit_merged_bio(io);
316 out:
317 	up_write(&io->io_rwsem);
318 }
319 
320 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
321 									int rw)
322 {
323 	__f2fs_submit_merged_bio(sbi, NULL, 0, 0, type, rw);
324 }
325 
326 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
327 				struct inode *inode, nid_t ino, pgoff_t idx,
328 				enum page_type type, int rw)
329 {
330 	if (has_merged_page(sbi, inode, ino, idx, type))
331 		__f2fs_submit_merged_bio(sbi, inode, ino, idx, type, rw);
332 }
333 
334 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
335 {
336 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
337 	f2fs_submit_merged_bio(sbi, NODE, WRITE);
338 	f2fs_submit_merged_bio(sbi, META, WRITE);
339 }
340 
341 /*
342  * Fill the locked page with data located in the block address.
343  * Return unlocked page.
344  */
345 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
346 {
347 	struct bio *bio;
348 	struct page *page = fio->encrypted_page ?
349 			fio->encrypted_page : fio->page;
350 
351 	trace_f2fs_submit_page_bio(page, fio);
352 	f2fs_trace_ios(fio, 0);
353 
354 	/* Allocate a new bio */
355 	bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
356 
357 	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
358 		bio_put(bio);
359 		return -EFAULT;
360 	}
361 	bio_set_op_attrs(bio, fio->op, fio->op_flags);
362 
363 	__submit_bio(fio->sbi, bio, fio->type);
364 	return 0;
365 }
366 
367 int f2fs_submit_page_mbio(struct f2fs_io_info *fio)
368 {
369 	struct f2fs_sb_info *sbi = fio->sbi;
370 	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
371 	struct f2fs_bio_info *io;
372 	bool is_read = is_read_io(fio->op);
373 	struct page *bio_page;
374 	int err = 0;
375 
376 	io = is_read ? &sbi->read_io : &sbi->write_io[btype];
377 
378 	if (fio->old_blkaddr != NEW_ADDR)
379 		verify_block_addr(sbi, fio->old_blkaddr);
380 	verify_block_addr(sbi, fio->new_blkaddr);
381 
382 	bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
383 
384 	/* set submitted = 1 as a return value */
385 	fio->submitted = 1;
386 
387 	if (!is_read)
388 		inc_page_count(sbi, WB_DATA_TYPE(bio_page));
389 
390 	down_write(&io->io_rwsem);
391 
392 	if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
393 	    (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
394 			!__same_bdev(sbi, fio->new_blkaddr, io->bio)))
395 		__submit_merged_bio(io);
396 alloc_new:
397 	if (io->bio == NULL) {
398 		if ((fio->type == DATA || fio->type == NODE) &&
399 				fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
400 			err = -EAGAIN;
401 			dec_page_count(sbi, WB_DATA_TYPE(bio_page));
402 			goto out_fail;
403 		}
404 		io->bio = __bio_alloc(sbi, fio->new_blkaddr,
405 						BIO_MAX_PAGES, is_read);
406 		io->fio = *fio;
407 	}
408 
409 	if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
410 							PAGE_SIZE) {
411 		__submit_merged_bio(io);
412 		goto alloc_new;
413 	}
414 
415 	io->last_block_in_bio = fio->new_blkaddr;
416 	f2fs_trace_ios(fio, 0);
417 out_fail:
418 	up_write(&io->io_rwsem);
419 	trace_f2fs_submit_page_mbio(fio->page, fio);
420 	return err;
421 }
422 
423 static void __set_data_blkaddr(struct dnode_of_data *dn)
424 {
425 	struct f2fs_node *rn = F2FS_NODE(dn->node_page);
426 	__le32 *addr_array;
427 
428 	/* Get physical address of data block */
429 	addr_array = blkaddr_in_node(rn);
430 	addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
431 }
432 
433 /*
434  * Lock ordering for the change of data block address:
435  * ->data_page
436  *  ->node_page
437  *    update block addresses in the node page
438  */
439 void set_data_blkaddr(struct dnode_of_data *dn)
440 {
441 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
442 	__set_data_blkaddr(dn);
443 	if (set_page_dirty(dn->node_page))
444 		dn->node_changed = true;
445 }
446 
447 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
448 {
449 	dn->data_blkaddr = blkaddr;
450 	set_data_blkaddr(dn);
451 	f2fs_update_extent_cache(dn);
452 }
453 
454 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
455 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
456 {
457 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
458 
459 	if (!count)
460 		return 0;
461 
462 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
463 		return -EPERM;
464 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
465 		return -ENOSPC;
466 
467 	trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
468 						dn->ofs_in_node, count);
469 
470 	f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
471 
472 	for (; count > 0; dn->ofs_in_node++) {
473 		block_t blkaddr =
474 			datablock_addr(dn->node_page, dn->ofs_in_node);
475 		if (blkaddr == NULL_ADDR) {
476 			dn->data_blkaddr = NEW_ADDR;
477 			__set_data_blkaddr(dn);
478 			count--;
479 		}
480 	}
481 
482 	if (set_page_dirty(dn->node_page))
483 		dn->node_changed = true;
484 	return 0;
485 }
486 
487 /* Should keep dn->ofs_in_node unchanged */
488 int reserve_new_block(struct dnode_of_data *dn)
489 {
490 	unsigned int ofs_in_node = dn->ofs_in_node;
491 	int ret;
492 
493 	ret = reserve_new_blocks(dn, 1);
494 	dn->ofs_in_node = ofs_in_node;
495 	return ret;
496 }
497 
498 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
499 {
500 	bool need_put = dn->inode_page ? false : true;
501 	int err;
502 
503 	err = get_dnode_of_data(dn, index, ALLOC_NODE);
504 	if (err)
505 		return err;
506 
507 	if (dn->data_blkaddr == NULL_ADDR)
508 		err = reserve_new_block(dn);
509 	if (err || need_put)
510 		f2fs_put_dnode(dn);
511 	return err;
512 }
513 
514 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
515 {
516 	struct extent_info ei  = {0,0,0};
517 	struct inode *inode = dn->inode;
518 
519 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
520 		dn->data_blkaddr = ei.blk + index - ei.fofs;
521 		return 0;
522 	}
523 
524 	return f2fs_reserve_block(dn, index);
525 }
526 
527 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
528 						int op_flags, bool for_write)
529 {
530 	struct address_space *mapping = inode->i_mapping;
531 	struct dnode_of_data dn;
532 	struct page *page;
533 	struct extent_info ei = {0,0,0};
534 	int err;
535 	struct f2fs_io_info fio = {
536 		.sbi = F2FS_I_SB(inode),
537 		.type = DATA,
538 		.op = REQ_OP_READ,
539 		.op_flags = op_flags,
540 		.encrypted_page = NULL,
541 	};
542 
543 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
544 		return read_mapping_page(mapping, index, NULL);
545 
546 	page = f2fs_grab_cache_page(mapping, index, for_write);
547 	if (!page)
548 		return ERR_PTR(-ENOMEM);
549 
550 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
551 		dn.data_blkaddr = ei.blk + index - ei.fofs;
552 		goto got_it;
553 	}
554 
555 	set_new_dnode(&dn, inode, NULL, NULL, 0);
556 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
557 	if (err)
558 		goto put_err;
559 	f2fs_put_dnode(&dn);
560 
561 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
562 		err = -ENOENT;
563 		goto put_err;
564 	}
565 got_it:
566 	if (PageUptodate(page)) {
567 		unlock_page(page);
568 		return page;
569 	}
570 
571 	/*
572 	 * A new dentry page is allocated but not able to be written, since its
573 	 * new inode page couldn't be allocated due to -ENOSPC.
574 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
575 	 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
576 	 */
577 	if (dn.data_blkaddr == NEW_ADDR) {
578 		zero_user_segment(page, 0, PAGE_SIZE);
579 		if (!PageUptodate(page))
580 			SetPageUptodate(page);
581 		unlock_page(page);
582 		return page;
583 	}
584 
585 	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
586 	fio.page = page;
587 	err = f2fs_submit_page_bio(&fio);
588 	if (err)
589 		goto put_err;
590 	return page;
591 
592 put_err:
593 	f2fs_put_page(page, 1);
594 	return ERR_PTR(err);
595 }
596 
597 struct page *find_data_page(struct inode *inode, pgoff_t index)
598 {
599 	struct address_space *mapping = inode->i_mapping;
600 	struct page *page;
601 
602 	page = find_get_page(mapping, index);
603 	if (page && PageUptodate(page))
604 		return page;
605 	f2fs_put_page(page, 0);
606 
607 	page = get_read_data_page(inode, index, 0, false);
608 	if (IS_ERR(page))
609 		return page;
610 
611 	if (PageUptodate(page))
612 		return page;
613 
614 	wait_on_page_locked(page);
615 	if (unlikely(!PageUptodate(page))) {
616 		f2fs_put_page(page, 0);
617 		return ERR_PTR(-EIO);
618 	}
619 	return page;
620 }
621 
622 /*
623  * If it tries to access a hole, return an error.
624  * Because, the callers, functions in dir.c and GC, should be able to know
625  * whether this page exists or not.
626  */
627 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
628 							bool for_write)
629 {
630 	struct address_space *mapping = inode->i_mapping;
631 	struct page *page;
632 repeat:
633 	page = get_read_data_page(inode, index, 0, for_write);
634 	if (IS_ERR(page))
635 		return page;
636 
637 	/* wait for read completion */
638 	lock_page(page);
639 	if (unlikely(page->mapping != mapping)) {
640 		f2fs_put_page(page, 1);
641 		goto repeat;
642 	}
643 	if (unlikely(!PageUptodate(page))) {
644 		f2fs_put_page(page, 1);
645 		return ERR_PTR(-EIO);
646 	}
647 	return page;
648 }
649 
650 /*
651  * Caller ensures that this data page is never allocated.
652  * A new zero-filled data page is allocated in the page cache.
653  *
654  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
655  * f2fs_unlock_op().
656  * Note that, ipage is set only by make_empty_dir, and if any error occur,
657  * ipage should be released by this function.
658  */
659 struct page *get_new_data_page(struct inode *inode,
660 		struct page *ipage, pgoff_t index, bool new_i_size)
661 {
662 	struct address_space *mapping = inode->i_mapping;
663 	struct page *page;
664 	struct dnode_of_data dn;
665 	int err;
666 
667 	page = f2fs_grab_cache_page(mapping, index, true);
668 	if (!page) {
669 		/*
670 		 * before exiting, we should make sure ipage will be released
671 		 * if any error occur.
672 		 */
673 		f2fs_put_page(ipage, 1);
674 		return ERR_PTR(-ENOMEM);
675 	}
676 
677 	set_new_dnode(&dn, inode, ipage, NULL, 0);
678 	err = f2fs_reserve_block(&dn, index);
679 	if (err) {
680 		f2fs_put_page(page, 1);
681 		return ERR_PTR(err);
682 	}
683 	if (!ipage)
684 		f2fs_put_dnode(&dn);
685 
686 	if (PageUptodate(page))
687 		goto got_it;
688 
689 	if (dn.data_blkaddr == NEW_ADDR) {
690 		zero_user_segment(page, 0, PAGE_SIZE);
691 		if (!PageUptodate(page))
692 			SetPageUptodate(page);
693 	} else {
694 		f2fs_put_page(page, 1);
695 
696 		/* if ipage exists, blkaddr should be NEW_ADDR */
697 		f2fs_bug_on(F2FS_I_SB(inode), ipage);
698 		page = get_lock_data_page(inode, index, true);
699 		if (IS_ERR(page))
700 			return page;
701 	}
702 got_it:
703 	if (new_i_size && i_size_read(inode) <
704 				((loff_t)(index + 1) << PAGE_SHIFT))
705 		f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
706 	return page;
707 }
708 
709 static int __allocate_data_block(struct dnode_of_data *dn)
710 {
711 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
712 	struct f2fs_summary sum;
713 	struct node_info ni;
714 	pgoff_t fofs;
715 	blkcnt_t count = 1;
716 
717 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
718 		return -EPERM;
719 
720 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
721 	if (dn->data_blkaddr == NEW_ADDR)
722 		goto alloc;
723 
724 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
725 		return -ENOSPC;
726 
727 alloc:
728 	get_node_info(sbi, dn->nid, &ni);
729 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
730 
731 	allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
732 						&sum, CURSEG_WARM_DATA);
733 	set_data_blkaddr(dn);
734 
735 	/* update i_size */
736 	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
737 							dn->ofs_in_node;
738 	if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
739 		f2fs_i_size_write(dn->inode,
740 				((loff_t)(fofs + 1) << PAGE_SHIFT));
741 	return 0;
742 }
743 
744 static inline bool __force_buffered_io(struct inode *inode, int rw)
745 {
746 	return ((f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) ||
747 			(rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
748 			F2FS_I_SB(inode)->s_ndevs);
749 }
750 
751 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
752 {
753 	struct inode *inode = file_inode(iocb->ki_filp);
754 	struct f2fs_map_blocks map;
755 	int err = 0;
756 
757 	if (is_inode_flag_set(inode, FI_NO_PREALLOC))
758 		return 0;
759 
760 	map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
761 	map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
762 	if (map.m_len > map.m_lblk)
763 		map.m_len -= map.m_lblk;
764 	else
765 		map.m_len = 0;
766 
767 	map.m_next_pgofs = NULL;
768 
769 	if (iocb->ki_flags & IOCB_DIRECT) {
770 		err = f2fs_convert_inline_inode(inode);
771 		if (err)
772 			return err;
773 		return f2fs_map_blocks(inode, &map, 1,
774 			__force_buffered_io(inode, WRITE) ?
775 				F2FS_GET_BLOCK_PRE_AIO :
776 				F2FS_GET_BLOCK_PRE_DIO);
777 	}
778 	if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
779 		err = f2fs_convert_inline_inode(inode);
780 		if (err)
781 			return err;
782 	}
783 	if (!f2fs_has_inline_data(inode))
784 		return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
785 	return err;
786 }
787 
788 /*
789  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
790  * f2fs_map_blocks structure.
791  * If original data blocks are allocated, then give them to blockdev.
792  * Otherwise,
793  *     a. preallocate requested block addresses
794  *     b. do not use extent cache for better performance
795  *     c. give the block addresses to blockdev
796  */
797 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
798 						int create, int flag)
799 {
800 	unsigned int maxblocks = map->m_len;
801 	struct dnode_of_data dn;
802 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
803 	int mode = create ? ALLOC_NODE : LOOKUP_NODE;
804 	pgoff_t pgofs, end_offset, end;
805 	int err = 0, ofs = 1;
806 	unsigned int ofs_in_node, last_ofs_in_node;
807 	blkcnt_t prealloc;
808 	struct extent_info ei = {0,0,0};
809 	block_t blkaddr;
810 
811 	if (!maxblocks)
812 		return 0;
813 
814 	map->m_len = 0;
815 	map->m_flags = 0;
816 
817 	/* it only supports block size == page size */
818 	pgofs =	(pgoff_t)map->m_lblk;
819 	end = pgofs + maxblocks;
820 
821 	if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
822 		map->m_pblk = ei.blk + pgofs - ei.fofs;
823 		map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
824 		map->m_flags = F2FS_MAP_MAPPED;
825 		goto out;
826 	}
827 
828 next_dnode:
829 	if (create)
830 		f2fs_lock_op(sbi);
831 
832 	/* When reading holes, we need its node page */
833 	set_new_dnode(&dn, inode, NULL, NULL, 0);
834 	err = get_dnode_of_data(&dn, pgofs, mode);
835 	if (err) {
836 		if (flag == F2FS_GET_BLOCK_BMAP)
837 			map->m_pblk = 0;
838 		if (err == -ENOENT) {
839 			err = 0;
840 			if (map->m_next_pgofs)
841 				*map->m_next_pgofs =
842 					get_next_page_offset(&dn, pgofs);
843 		}
844 		goto unlock_out;
845 	}
846 
847 	prealloc = 0;
848 	last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
849 	end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
850 
851 next_block:
852 	blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
853 
854 	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
855 		if (create) {
856 			if (unlikely(f2fs_cp_error(sbi))) {
857 				err = -EIO;
858 				goto sync_out;
859 			}
860 			if (flag == F2FS_GET_BLOCK_PRE_AIO) {
861 				if (blkaddr == NULL_ADDR) {
862 					prealloc++;
863 					last_ofs_in_node = dn.ofs_in_node;
864 				}
865 			} else {
866 				err = __allocate_data_block(&dn);
867 				if (!err)
868 					set_inode_flag(inode, FI_APPEND_WRITE);
869 			}
870 			if (err)
871 				goto sync_out;
872 			map->m_flags |= F2FS_MAP_NEW;
873 			blkaddr = dn.data_blkaddr;
874 		} else {
875 			if (flag == F2FS_GET_BLOCK_BMAP) {
876 				map->m_pblk = 0;
877 				goto sync_out;
878 			}
879 			if (flag == F2FS_GET_BLOCK_FIEMAP &&
880 						blkaddr == NULL_ADDR) {
881 				if (map->m_next_pgofs)
882 					*map->m_next_pgofs = pgofs + 1;
883 			}
884 			if (flag != F2FS_GET_BLOCK_FIEMAP ||
885 						blkaddr != NEW_ADDR)
886 				goto sync_out;
887 		}
888 	}
889 
890 	if (flag == F2FS_GET_BLOCK_PRE_AIO)
891 		goto skip;
892 
893 	if (map->m_len == 0) {
894 		/* preallocated unwritten block should be mapped for fiemap. */
895 		if (blkaddr == NEW_ADDR)
896 			map->m_flags |= F2FS_MAP_UNWRITTEN;
897 		map->m_flags |= F2FS_MAP_MAPPED;
898 
899 		map->m_pblk = blkaddr;
900 		map->m_len = 1;
901 	} else if ((map->m_pblk != NEW_ADDR &&
902 			blkaddr == (map->m_pblk + ofs)) ||
903 			(map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
904 			flag == F2FS_GET_BLOCK_PRE_DIO) {
905 		ofs++;
906 		map->m_len++;
907 	} else {
908 		goto sync_out;
909 	}
910 
911 skip:
912 	dn.ofs_in_node++;
913 	pgofs++;
914 
915 	/* preallocate blocks in batch for one dnode page */
916 	if (flag == F2FS_GET_BLOCK_PRE_AIO &&
917 			(pgofs == end || dn.ofs_in_node == end_offset)) {
918 
919 		dn.ofs_in_node = ofs_in_node;
920 		err = reserve_new_blocks(&dn, prealloc);
921 		if (err)
922 			goto sync_out;
923 
924 		map->m_len += dn.ofs_in_node - ofs_in_node;
925 		if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
926 			err = -ENOSPC;
927 			goto sync_out;
928 		}
929 		dn.ofs_in_node = end_offset;
930 	}
931 
932 	if (pgofs >= end)
933 		goto sync_out;
934 	else if (dn.ofs_in_node < end_offset)
935 		goto next_block;
936 
937 	f2fs_put_dnode(&dn);
938 
939 	if (create) {
940 		f2fs_unlock_op(sbi);
941 		f2fs_balance_fs(sbi, dn.node_changed);
942 	}
943 	goto next_dnode;
944 
945 sync_out:
946 	f2fs_put_dnode(&dn);
947 unlock_out:
948 	if (create) {
949 		f2fs_unlock_op(sbi);
950 		f2fs_balance_fs(sbi, dn.node_changed);
951 	}
952 out:
953 	trace_f2fs_map_blocks(inode, map, err);
954 	return err;
955 }
956 
957 static int __get_data_block(struct inode *inode, sector_t iblock,
958 			struct buffer_head *bh, int create, int flag,
959 			pgoff_t *next_pgofs)
960 {
961 	struct f2fs_map_blocks map;
962 	int err;
963 
964 	map.m_lblk = iblock;
965 	map.m_len = bh->b_size >> inode->i_blkbits;
966 	map.m_next_pgofs = next_pgofs;
967 
968 	err = f2fs_map_blocks(inode, &map, create, flag);
969 	if (!err) {
970 		map_bh(bh, inode->i_sb, map.m_pblk);
971 		bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
972 		bh->b_size = (u64)map.m_len << inode->i_blkbits;
973 	}
974 	return err;
975 }
976 
977 static int get_data_block(struct inode *inode, sector_t iblock,
978 			struct buffer_head *bh_result, int create, int flag,
979 			pgoff_t *next_pgofs)
980 {
981 	return __get_data_block(inode, iblock, bh_result, create,
982 							flag, next_pgofs);
983 }
984 
985 static int get_data_block_dio(struct inode *inode, sector_t iblock,
986 			struct buffer_head *bh_result, int create)
987 {
988 	return __get_data_block(inode, iblock, bh_result, create,
989 						F2FS_GET_BLOCK_DIO, NULL);
990 }
991 
992 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
993 			struct buffer_head *bh_result, int create)
994 {
995 	/* Block number less than F2FS MAX BLOCKS */
996 	if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
997 		return -EFBIG;
998 
999 	return __get_data_block(inode, iblock, bh_result, create,
1000 						F2FS_GET_BLOCK_BMAP, NULL);
1001 }
1002 
1003 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1004 {
1005 	return (offset >> inode->i_blkbits);
1006 }
1007 
1008 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1009 {
1010 	return (blk << inode->i_blkbits);
1011 }
1012 
1013 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1014 		u64 start, u64 len)
1015 {
1016 	struct buffer_head map_bh;
1017 	sector_t start_blk, last_blk;
1018 	pgoff_t next_pgofs;
1019 	u64 logical = 0, phys = 0, size = 0;
1020 	u32 flags = 0;
1021 	int ret = 0;
1022 
1023 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1024 	if (ret)
1025 		return ret;
1026 
1027 	if (f2fs_has_inline_data(inode)) {
1028 		ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1029 		if (ret != -EAGAIN)
1030 			return ret;
1031 	}
1032 
1033 	inode_lock(inode);
1034 
1035 	if (logical_to_blk(inode, len) == 0)
1036 		len = blk_to_logical(inode, 1);
1037 
1038 	start_blk = logical_to_blk(inode, start);
1039 	last_blk = logical_to_blk(inode, start + len - 1);
1040 
1041 next:
1042 	memset(&map_bh, 0, sizeof(struct buffer_head));
1043 	map_bh.b_size = len;
1044 
1045 	ret = get_data_block(inode, start_blk, &map_bh, 0,
1046 					F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1047 	if (ret)
1048 		goto out;
1049 
1050 	/* HOLE */
1051 	if (!buffer_mapped(&map_bh)) {
1052 		start_blk = next_pgofs;
1053 
1054 		if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1055 					F2FS_I_SB(inode)->max_file_blocks))
1056 			goto prep_next;
1057 
1058 		flags |= FIEMAP_EXTENT_LAST;
1059 	}
1060 
1061 	if (size) {
1062 		if (f2fs_encrypted_inode(inode))
1063 			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1064 
1065 		ret = fiemap_fill_next_extent(fieinfo, logical,
1066 				phys, size, flags);
1067 	}
1068 
1069 	if (start_blk > last_blk || ret)
1070 		goto out;
1071 
1072 	logical = blk_to_logical(inode, start_blk);
1073 	phys = blk_to_logical(inode, map_bh.b_blocknr);
1074 	size = map_bh.b_size;
1075 	flags = 0;
1076 	if (buffer_unwritten(&map_bh))
1077 		flags = FIEMAP_EXTENT_UNWRITTEN;
1078 
1079 	start_blk += logical_to_blk(inode, size);
1080 
1081 prep_next:
1082 	cond_resched();
1083 	if (fatal_signal_pending(current))
1084 		ret = -EINTR;
1085 	else
1086 		goto next;
1087 out:
1088 	if (ret == 1)
1089 		ret = 0;
1090 
1091 	inode_unlock(inode);
1092 	return ret;
1093 }
1094 
1095 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
1096 				 unsigned nr_pages)
1097 {
1098 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1099 	struct fscrypt_ctx *ctx = NULL;
1100 	struct bio *bio;
1101 
1102 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1103 		ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1104 		if (IS_ERR(ctx))
1105 			return ERR_CAST(ctx);
1106 
1107 		/* wait the page to be moved by cleaning */
1108 		f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1109 	}
1110 
1111 	bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
1112 	if (!bio) {
1113 		if (ctx)
1114 			fscrypt_release_ctx(ctx);
1115 		return ERR_PTR(-ENOMEM);
1116 	}
1117 	f2fs_target_device(sbi, blkaddr, bio);
1118 	bio->bi_end_io = f2fs_read_end_io;
1119 	bio->bi_private = ctx;
1120 
1121 	return bio;
1122 }
1123 
1124 /*
1125  * This function was originally taken from fs/mpage.c, and customized for f2fs.
1126  * Major change was from block_size == page_size in f2fs by default.
1127  */
1128 static int f2fs_mpage_readpages(struct address_space *mapping,
1129 			struct list_head *pages, struct page *page,
1130 			unsigned nr_pages)
1131 {
1132 	struct bio *bio = NULL;
1133 	unsigned page_idx;
1134 	sector_t last_block_in_bio = 0;
1135 	struct inode *inode = mapping->host;
1136 	const unsigned blkbits = inode->i_blkbits;
1137 	const unsigned blocksize = 1 << blkbits;
1138 	sector_t block_in_file;
1139 	sector_t last_block;
1140 	sector_t last_block_in_file;
1141 	sector_t block_nr;
1142 	struct f2fs_map_blocks map;
1143 
1144 	map.m_pblk = 0;
1145 	map.m_lblk = 0;
1146 	map.m_len = 0;
1147 	map.m_flags = 0;
1148 	map.m_next_pgofs = NULL;
1149 
1150 	for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1151 
1152 		prefetchw(&page->flags);
1153 		if (pages) {
1154 			page = list_last_entry(pages, struct page, lru);
1155 			list_del(&page->lru);
1156 			if (add_to_page_cache_lru(page, mapping,
1157 						  page->index,
1158 						  readahead_gfp_mask(mapping)))
1159 				goto next_page;
1160 		}
1161 
1162 		block_in_file = (sector_t)page->index;
1163 		last_block = block_in_file + nr_pages;
1164 		last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1165 								blkbits;
1166 		if (last_block > last_block_in_file)
1167 			last_block = last_block_in_file;
1168 
1169 		/*
1170 		 * Map blocks using the previous result first.
1171 		 */
1172 		if ((map.m_flags & F2FS_MAP_MAPPED) &&
1173 				block_in_file > map.m_lblk &&
1174 				block_in_file < (map.m_lblk + map.m_len))
1175 			goto got_it;
1176 
1177 		/*
1178 		 * Then do more f2fs_map_blocks() calls until we are
1179 		 * done with this page.
1180 		 */
1181 		map.m_flags = 0;
1182 
1183 		if (block_in_file < last_block) {
1184 			map.m_lblk = block_in_file;
1185 			map.m_len = last_block - block_in_file;
1186 
1187 			if (f2fs_map_blocks(inode, &map, 0,
1188 						F2FS_GET_BLOCK_READ))
1189 				goto set_error_page;
1190 		}
1191 got_it:
1192 		if ((map.m_flags & F2FS_MAP_MAPPED)) {
1193 			block_nr = map.m_pblk + block_in_file - map.m_lblk;
1194 			SetPageMappedToDisk(page);
1195 
1196 			if (!PageUptodate(page) && !cleancache_get_page(page)) {
1197 				SetPageUptodate(page);
1198 				goto confused;
1199 			}
1200 		} else {
1201 			zero_user_segment(page, 0, PAGE_SIZE);
1202 			if (!PageUptodate(page))
1203 				SetPageUptodate(page);
1204 			unlock_page(page);
1205 			goto next_page;
1206 		}
1207 
1208 		/*
1209 		 * This page will go to BIO.  Do we need to send this
1210 		 * BIO off first?
1211 		 */
1212 		if (bio && (last_block_in_bio != block_nr - 1 ||
1213 			!__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1214 submit_and_realloc:
1215 			__submit_bio(F2FS_I_SB(inode), bio, DATA);
1216 			bio = NULL;
1217 		}
1218 		if (bio == NULL) {
1219 			bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1220 			if (IS_ERR(bio)) {
1221 				bio = NULL;
1222 				goto set_error_page;
1223 			}
1224 			bio_set_op_attrs(bio, REQ_OP_READ, 0);
1225 		}
1226 
1227 		if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1228 			goto submit_and_realloc;
1229 
1230 		last_block_in_bio = block_nr;
1231 		goto next_page;
1232 set_error_page:
1233 		SetPageError(page);
1234 		zero_user_segment(page, 0, PAGE_SIZE);
1235 		unlock_page(page);
1236 		goto next_page;
1237 confused:
1238 		if (bio) {
1239 			__submit_bio(F2FS_I_SB(inode), bio, DATA);
1240 			bio = NULL;
1241 		}
1242 		unlock_page(page);
1243 next_page:
1244 		if (pages)
1245 			put_page(page);
1246 	}
1247 	BUG_ON(pages && !list_empty(pages));
1248 	if (bio)
1249 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
1250 	return 0;
1251 }
1252 
1253 static int f2fs_read_data_page(struct file *file, struct page *page)
1254 {
1255 	struct inode *inode = page->mapping->host;
1256 	int ret = -EAGAIN;
1257 
1258 	trace_f2fs_readpage(page, DATA);
1259 
1260 	/* If the file has inline data, try to read it directly */
1261 	if (f2fs_has_inline_data(inode))
1262 		ret = f2fs_read_inline_data(inode, page);
1263 	if (ret == -EAGAIN)
1264 		ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1265 	return ret;
1266 }
1267 
1268 static int f2fs_read_data_pages(struct file *file,
1269 			struct address_space *mapping,
1270 			struct list_head *pages, unsigned nr_pages)
1271 {
1272 	struct inode *inode = file->f_mapping->host;
1273 	struct page *page = list_last_entry(pages, struct page, lru);
1274 
1275 	trace_f2fs_readpages(inode, page, nr_pages);
1276 
1277 	/* If the file has inline data, skip readpages */
1278 	if (f2fs_has_inline_data(inode))
1279 		return 0;
1280 
1281 	return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1282 }
1283 
1284 int do_write_data_page(struct f2fs_io_info *fio)
1285 {
1286 	struct page *page = fio->page;
1287 	struct inode *inode = page->mapping->host;
1288 	struct dnode_of_data dn;
1289 	int err = 0;
1290 
1291 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1292 	err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1293 	if (err)
1294 		return err;
1295 
1296 	fio->old_blkaddr = dn.data_blkaddr;
1297 
1298 	/* This page is already truncated */
1299 	if (fio->old_blkaddr == NULL_ADDR) {
1300 		ClearPageUptodate(page);
1301 		goto out_writepage;
1302 	}
1303 
1304 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1305 		gfp_t gfp_flags = GFP_NOFS;
1306 
1307 		/* wait for GCed encrypted page writeback */
1308 		f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1309 							fio->old_blkaddr);
1310 retry_encrypt:
1311 		fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1312 							PAGE_SIZE, 0,
1313 							fio->page->index,
1314 							gfp_flags);
1315 		if (IS_ERR(fio->encrypted_page)) {
1316 			err = PTR_ERR(fio->encrypted_page);
1317 			if (err == -ENOMEM) {
1318 				/* flush pending ios and wait for a while */
1319 				f2fs_flush_merged_bios(F2FS_I_SB(inode));
1320 				congestion_wait(BLK_RW_ASYNC, HZ/50);
1321 				gfp_flags |= __GFP_NOFAIL;
1322 				err = 0;
1323 				goto retry_encrypt;
1324 			}
1325 			goto out_writepage;
1326 		}
1327 	}
1328 
1329 	set_page_writeback(page);
1330 
1331 	/*
1332 	 * If current allocation needs SSR,
1333 	 * it had better in-place writes for updated data.
1334 	 */
1335 	if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1336 			!is_cold_data(page) &&
1337 			!IS_ATOMIC_WRITTEN_PAGE(page) &&
1338 			need_inplace_update(inode))) {
1339 		rewrite_data_page(fio);
1340 		set_inode_flag(inode, FI_UPDATE_WRITE);
1341 		trace_f2fs_do_write_data_page(page, IPU);
1342 	} else {
1343 		write_data_page(&dn, fio);
1344 		trace_f2fs_do_write_data_page(page, OPU);
1345 		set_inode_flag(inode, FI_APPEND_WRITE);
1346 		if (page->index == 0)
1347 			set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1348 	}
1349 out_writepage:
1350 	f2fs_put_dnode(&dn);
1351 	return err;
1352 }
1353 
1354 static int __write_data_page(struct page *page, bool *submitted,
1355 				struct writeback_control *wbc)
1356 {
1357 	struct inode *inode = page->mapping->host;
1358 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1359 	loff_t i_size = i_size_read(inode);
1360 	const pgoff_t end_index = ((unsigned long long) i_size)
1361 							>> PAGE_SHIFT;
1362 	loff_t psize = (page->index + 1) << PAGE_SHIFT;
1363 	unsigned offset = 0;
1364 	bool need_balance_fs = false;
1365 	int err = 0;
1366 	struct f2fs_io_info fio = {
1367 		.sbi = sbi,
1368 		.type = DATA,
1369 		.op = REQ_OP_WRITE,
1370 		.op_flags = wbc_to_write_flags(wbc),
1371 		.page = page,
1372 		.encrypted_page = NULL,
1373 		.submitted = false,
1374 	};
1375 
1376 	trace_f2fs_writepage(page, DATA);
1377 
1378 	if (page->index < end_index)
1379 		goto write;
1380 
1381 	/*
1382 	 * If the offset is out-of-range of file size,
1383 	 * this page does not have to be written to disk.
1384 	 */
1385 	offset = i_size & (PAGE_SIZE - 1);
1386 	if ((page->index >= end_index + 1) || !offset)
1387 		goto out;
1388 
1389 	zero_user_segment(page, offset, PAGE_SIZE);
1390 write:
1391 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1392 		goto redirty_out;
1393 	if (f2fs_is_drop_cache(inode))
1394 		goto out;
1395 	/* we should not write 0'th page having journal header */
1396 	if (f2fs_is_volatile_file(inode) && (!page->index ||
1397 			(!wbc->for_reclaim &&
1398 			available_free_memory(sbi, BASE_CHECK))))
1399 		goto redirty_out;
1400 
1401 	/* we should bypass data pages to proceed the kworkder jobs */
1402 	if (unlikely(f2fs_cp_error(sbi))) {
1403 		mapping_set_error(page->mapping, -EIO);
1404 		goto out;
1405 	}
1406 
1407 	/* Dentry blocks are controlled by checkpoint */
1408 	if (S_ISDIR(inode->i_mode)) {
1409 		err = do_write_data_page(&fio);
1410 		goto done;
1411 	}
1412 
1413 	if (!wbc->for_reclaim)
1414 		need_balance_fs = true;
1415 	else if (has_not_enough_free_secs(sbi, 0, 0))
1416 		goto redirty_out;
1417 
1418 	err = -EAGAIN;
1419 	if (f2fs_has_inline_data(inode)) {
1420 		err = f2fs_write_inline_data(inode, page);
1421 		if (!err)
1422 			goto out;
1423 	}
1424 	f2fs_lock_op(sbi);
1425 	if (err == -EAGAIN)
1426 		err = do_write_data_page(&fio);
1427 	if (F2FS_I(inode)->last_disk_size < psize)
1428 		F2FS_I(inode)->last_disk_size = psize;
1429 	f2fs_unlock_op(sbi);
1430 done:
1431 	if (err && err != -ENOENT)
1432 		goto redirty_out;
1433 
1434 out:
1435 	inode_dec_dirty_pages(inode);
1436 	if (err)
1437 		ClearPageUptodate(page);
1438 
1439 	if (wbc->for_reclaim) {
1440 		f2fs_submit_merged_bio_cond(sbi, inode, 0, page->index,
1441 						DATA, WRITE);
1442 		remove_dirty_inode(inode);
1443 		submitted = NULL;
1444 	}
1445 
1446 	unlock_page(page);
1447 	f2fs_balance_fs(sbi, need_balance_fs);
1448 
1449 	if (unlikely(f2fs_cp_error(sbi))) {
1450 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
1451 		submitted = NULL;
1452 	}
1453 
1454 	if (submitted)
1455 		*submitted = fio.submitted;
1456 
1457 	return 0;
1458 
1459 redirty_out:
1460 	redirty_page_for_writepage(wbc, page);
1461 	if (!err)
1462 		return AOP_WRITEPAGE_ACTIVATE;
1463 	unlock_page(page);
1464 	return err;
1465 }
1466 
1467 static int f2fs_write_data_page(struct page *page,
1468 					struct writeback_control *wbc)
1469 {
1470 	return __write_data_page(page, NULL, wbc);
1471 }
1472 
1473 /*
1474  * This function was copied from write_cche_pages from mm/page-writeback.c.
1475  * The major change is making write step of cold data page separately from
1476  * warm/hot data page.
1477  */
1478 static int f2fs_write_cache_pages(struct address_space *mapping,
1479 					struct writeback_control *wbc)
1480 {
1481 	int ret = 0;
1482 	int done = 0;
1483 	struct pagevec pvec;
1484 	int nr_pages;
1485 	pgoff_t uninitialized_var(writeback_index);
1486 	pgoff_t index;
1487 	pgoff_t end;		/* Inclusive */
1488 	pgoff_t done_index;
1489 	pgoff_t last_idx = ULONG_MAX;
1490 	int cycled;
1491 	int range_whole = 0;
1492 	int tag;
1493 
1494 	pagevec_init(&pvec, 0);
1495 
1496 	if (wbc->range_cyclic) {
1497 		writeback_index = mapping->writeback_index; /* prev offset */
1498 		index = writeback_index;
1499 		if (index == 0)
1500 			cycled = 1;
1501 		else
1502 			cycled = 0;
1503 		end = -1;
1504 	} else {
1505 		index = wbc->range_start >> PAGE_SHIFT;
1506 		end = wbc->range_end >> PAGE_SHIFT;
1507 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1508 			range_whole = 1;
1509 		cycled = 1; /* ignore range_cyclic tests */
1510 	}
1511 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1512 		tag = PAGECACHE_TAG_TOWRITE;
1513 	else
1514 		tag = PAGECACHE_TAG_DIRTY;
1515 retry:
1516 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1517 		tag_pages_for_writeback(mapping, index, end);
1518 	done_index = index;
1519 	while (!done && (index <= end)) {
1520 		int i;
1521 
1522 		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1523 			      min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1524 		if (nr_pages == 0)
1525 			break;
1526 
1527 		for (i = 0; i < nr_pages; i++) {
1528 			struct page *page = pvec.pages[i];
1529 			bool submitted = false;
1530 
1531 			if (page->index > end) {
1532 				done = 1;
1533 				break;
1534 			}
1535 
1536 			done_index = page->index;
1537 
1538 			lock_page(page);
1539 
1540 			if (unlikely(page->mapping != mapping)) {
1541 continue_unlock:
1542 				unlock_page(page);
1543 				continue;
1544 			}
1545 
1546 			if (!PageDirty(page)) {
1547 				/* someone wrote it for us */
1548 				goto continue_unlock;
1549 			}
1550 
1551 			if (PageWriteback(page)) {
1552 				if (wbc->sync_mode != WB_SYNC_NONE)
1553 					f2fs_wait_on_page_writeback(page,
1554 								DATA, true);
1555 				else
1556 					goto continue_unlock;
1557 			}
1558 
1559 			BUG_ON(PageWriteback(page));
1560 			if (!clear_page_dirty_for_io(page))
1561 				goto continue_unlock;
1562 
1563 			ret = __write_data_page(page, &submitted, wbc);
1564 			if (unlikely(ret)) {
1565 				/*
1566 				 * keep nr_to_write, since vfs uses this to
1567 				 * get # of written pages.
1568 				 */
1569 				if (ret == AOP_WRITEPAGE_ACTIVATE) {
1570 					unlock_page(page);
1571 					ret = 0;
1572 					continue;
1573 				}
1574 				done_index = page->index + 1;
1575 				done = 1;
1576 				break;
1577 			} else if (submitted) {
1578 				last_idx = page->index;
1579 			}
1580 
1581 			if (--wbc->nr_to_write <= 0 &&
1582 			    wbc->sync_mode == WB_SYNC_NONE) {
1583 				done = 1;
1584 				break;
1585 			}
1586 		}
1587 		pagevec_release(&pvec);
1588 		cond_resched();
1589 	}
1590 
1591 	if (!cycled && !done) {
1592 		cycled = 1;
1593 		index = 0;
1594 		end = writeback_index - 1;
1595 		goto retry;
1596 	}
1597 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1598 		mapping->writeback_index = done_index;
1599 
1600 	if (last_idx != ULONG_MAX)
1601 		f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1602 						0, last_idx, DATA, WRITE);
1603 
1604 	return ret;
1605 }
1606 
1607 static int f2fs_write_data_pages(struct address_space *mapping,
1608 			    struct writeback_control *wbc)
1609 {
1610 	struct inode *inode = mapping->host;
1611 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1612 	struct blk_plug plug;
1613 	int ret;
1614 
1615 	/* deal with chardevs and other special file */
1616 	if (!mapping->a_ops->writepage)
1617 		return 0;
1618 
1619 	/* skip writing if there is no dirty page in this inode */
1620 	if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1621 		return 0;
1622 
1623 	if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1624 			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1625 			available_free_memory(sbi, DIRTY_DENTS))
1626 		goto skip_write;
1627 
1628 	/* skip writing during file defragment */
1629 	if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1630 		goto skip_write;
1631 
1632 	/* during POR, we don't need to trigger writepage at all. */
1633 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1634 		goto skip_write;
1635 
1636 	trace_f2fs_writepages(mapping->host, wbc, DATA);
1637 
1638 	blk_start_plug(&plug);
1639 	ret = f2fs_write_cache_pages(mapping, wbc);
1640 	blk_finish_plug(&plug);
1641 	/*
1642 	 * if some pages were truncated, we cannot guarantee its mapping->host
1643 	 * to detect pending bios.
1644 	 */
1645 
1646 	remove_dirty_inode(inode);
1647 	return ret;
1648 
1649 skip_write:
1650 	wbc->pages_skipped += get_dirty_pages(inode);
1651 	trace_f2fs_writepages(mapping->host, wbc, DATA);
1652 	return 0;
1653 }
1654 
1655 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1656 {
1657 	struct inode *inode = mapping->host;
1658 	loff_t i_size = i_size_read(inode);
1659 
1660 	if (to > i_size) {
1661 		truncate_pagecache(inode, i_size);
1662 		truncate_blocks(inode, i_size, true);
1663 	}
1664 }
1665 
1666 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1667 			struct page *page, loff_t pos, unsigned len,
1668 			block_t *blk_addr, bool *node_changed)
1669 {
1670 	struct inode *inode = page->mapping->host;
1671 	pgoff_t index = page->index;
1672 	struct dnode_of_data dn;
1673 	struct page *ipage;
1674 	bool locked = false;
1675 	struct extent_info ei = {0,0,0};
1676 	int err = 0;
1677 
1678 	/*
1679 	 * we already allocated all the blocks, so we don't need to get
1680 	 * the block addresses when there is no need to fill the page.
1681 	 */
1682 	if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1683 			!is_inode_flag_set(inode, FI_NO_PREALLOC))
1684 		return 0;
1685 
1686 	if (f2fs_has_inline_data(inode) ||
1687 			(pos & PAGE_MASK) >= i_size_read(inode)) {
1688 		f2fs_lock_op(sbi);
1689 		locked = true;
1690 	}
1691 restart:
1692 	/* check inline_data */
1693 	ipage = get_node_page(sbi, inode->i_ino);
1694 	if (IS_ERR(ipage)) {
1695 		err = PTR_ERR(ipage);
1696 		goto unlock_out;
1697 	}
1698 
1699 	set_new_dnode(&dn, inode, ipage, ipage, 0);
1700 
1701 	if (f2fs_has_inline_data(inode)) {
1702 		if (pos + len <= MAX_INLINE_DATA) {
1703 			read_inline_data(page, ipage);
1704 			set_inode_flag(inode, FI_DATA_EXIST);
1705 			if (inode->i_nlink)
1706 				set_inline_node(ipage);
1707 		} else {
1708 			err = f2fs_convert_inline_page(&dn, page);
1709 			if (err)
1710 				goto out;
1711 			if (dn.data_blkaddr == NULL_ADDR)
1712 				err = f2fs_get_block(&dn, index);
1713 		}
1714 	} else if (locked) {
1715 		err = f2fs_get_block(&dn, index);
1716 	} else {
1717 		if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1718 			dn.data_blkaddr = ei.blk + index - ei.fofs;
1719 		} else {
1720 			/* hole case */
1721 			err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1722 			if (err || dn.data_blkaddr == NULL_ADDR) {
1723 				f2fs_put_dnode(&dn);
1724 				f2fs_lock_op(sbi);
1725 				locked = true;
1726 				goto restart;
1727 			}
1728 		}
1729 	}
1730 
1731 	/* convert_inline_page can make node_changed */
1732 	*blk_addr = dn.data_blkaddr;
1733 	*node_changed = dn.node_changed;
1734 out:
1735 	f2fs_put_dnode(&dn);
1736 unlock_out:
1737 	if (locked)
1738 		f2fs_unlock_op(sbi);
1739 	return err;
1740 }
1741 
1742 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1743 		loff_t pos, unsigned len, unsigned flags,
1744 		struct page **pagep, void **fsdata)
1745 {
1746 	struct inode *inode = mapping->host;
1747 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1748 	struct page *page = NULL;
1749 	pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1750 	bool need_balance = false;
1751 	block_t blkaddr = NULL_ADDR;
1752 	int err = 0;
1753 
1754 	trace_f2fs_write_begin(inode, pos, len, flags);
1755 
1756 	/*
1757 	 * We should check this at this moment to avoid deadlock on inode page
1758 	 * and #0 page. The locking rule for inline_data conversion should be:
1759 	 * lock_page(page #0) -> lock_page(inode_page)
1760 	 */
1761 	if (index != 0) {
1762 		err = f2fs_convert_inline_inode(inode);
1763 		if (err)
1764 			goto fail;
1765 	}
1766 repeat:
1767 	/*
1768 	 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1769 	 * wait_for_stable_page. Will wait that below with our IO control.
1770 	 */
1771 	page = pagecache_get_page(mapping, index,
1772 				FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1773 	if (!page) {
1774 		err = -ENOMEM;
1775 		goto fail;
1776 	}
1777 
1778 	*pagep = page;
1779 
1780 	err = prepare_write_begin(sbi, page, pos, len,
1781 					&blkaddr, &need_balance);
1782 	if (err)
1783 		goto fail;
1784 
1785 	if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1786 		unlock_page(page);
1787 		f2fs_balance_fs(sbi, true);
1788 		lock_page(page);
1789 		if (page->mapping != mapping) {
1790 			/* The page got truncated from under us */
1791 			f2fs_put_page(page, 1);
1792 			goto repeat;
1793 		}
1794 	}
1795 
1796 	f2fs_wait_on_page_writeback(page, DATA, false);
1797 
1798 	/* wait for GCed encrypted page writeback */
1799 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1800 		f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1801 
1802 	if (len == PAGE_SIZE || PageUptodate(page))
1803 		return 0;
1804 
1805 	if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
1806 		zero_user_segment(page, len, PAGE_SIZE);
1807 		return 0;
1808 	}
1809 
1810 	if (blkaddr == NEW_ADDR) {
1811 		zero_user_segment(page, 0, PAGE_SIZE);
1812 		SetPageUptodate(page);
1813 	} else {
1814 		struct bio *bio;
1815 
1816 		bio = f2fs_grab_bio(inode, blkaddr, 1);
1817 		if (IS_ERR(bio)) {
1818 			err = PTR_ERR(bio);
1819 			goto fail;
1820 		}
1821 		bio->bi_opf = REQ_OP_READ;
1822 		if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1823 			bio_put(bio);
1824 			err = -EFAULT;
1825 			goto fail;
1826 		}
1827 
1828 		__submit_bio(sbi, bio, DATA);
1829 
1830 		lock_page(page);
1831 		if (unlikely(page->mapping != mapping)) {
1832 			f2fs_put_page(page, 1);
1833 			goto repeat;
1834 		}
1835 		if (unlikely(!PageUptodate(page))) {
1836 			err = -EIO;
1837 			goto fail;
1838 		}
1839 	}
1840 	return 0;
1841 
1842 fail:
1843 	f2fs_put_page(page, 1);
1844 	f2fs_write_failed(mapping, pos + len);
1845 	return err;
1846 }
1847 
1848 static int f2fs_write_end(struct file *file,
1849 			struct address_space *mapping,
1850 			loff_t pos, unsigned len, unsigned copied,
1851 			struct page *page, void *fsdata)
1852 {
1853 	struct inode *inode = page->mapping->host;
1854 
1855 	trace_f2fs_write_end(inode, pos, len, copied);
1856 
1857 	/*
1858 	 * This should be come from len == PAGE_SIZE, and we expect copied
1859 	 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1860 	 * let generic_perform_write() try to copy data again through copied=0.
1861 	 */
1862 	if (!PageUptodate(page)) {
1863 		if (unlikely(copied != len))
1864 			copied = 0;
1865 		else
1866 			SetPageUptodate(page);
1867 	}
1868 	if (!copied)
1869 		goto unlock_out;
1870 
1871 	set_page_dirty(page);
1872 
1873 	if (pos + copied > i_size_read(inode))
1874 		f2fs_i_size_write(inode, pos + copied);
1875 unlock_out:
1876 	f2fs_put_page(page, 1);
1877 	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1878 	return copied;
1879 }
1880 
1881 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1882 			   loff_t offset)
1883 {
1884 	unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1885 
1886 	if (offset & blocksize_mask)
1887 		return -EINVAL;
1888 
1889 	if (iov_iter_alignment(iter) & blocksize_mask)
1890 		return -EINVAL;
1891 
1892 	return 0;
1893 }
1894 
1895 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1896 {
1897 	struct address_space *mapping = iocb->ki_filp->f_mapping;
1898 	struct inode *inode = mapping->host;
1899 	size_t count = iov_iter_count(iter);
1900 	loff_t offset = iocb->ki_pos;
1901 	int rw = iov_iter_rw(iter);
1902 	int err;
1903 
1904 	err = check_direct_IO(inode, iter, offset);
1905 	if (err)
1906 		return err;
1907 
1908 	if (__force_buffered_io(inode, rw))
1909 		return 0;
1910 
1911 	trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1912 
1913 	down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1914 	err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1915 	up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1916 
1917 	if (rw == WRITE) {
1918 		if (err > 0)
1919 			set_inode_flag(inode, FI_UPDATE_WRITE);
1920 		else if (err < 0)
1921 			f2fs_write_failed(mapping, offset + count);
1922 	}
1923 
1924 	trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1925 
1926 	return err;
1927 }
1928 
1929 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1930 							unsigned int length)
1931 {
1932 	struct inode *inode = page->mapping->host;
1933 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1934 
1935 	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1936 		(offset % PAGE_SIZE || length != PAGE_SIZE))
1937 		return;
1938 
1939 	if (PageDirty(page)) {
1940 		if (inode->i_ino == F2FS_META_INO(sbi)) {
1941 			dec_page_count(sbi, F2FS_DIRTY_META);
1942 		} else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
1943 			dec_page_count(sbi, F2FS_DIRTY_NODES);
1944 		} else {
1945 			inode_dec_dirty_pages(inode);
1946 			remove_dirty_inode(inode);
1947 		}
1948 	}
1949 
1950 	/* This is atomic written page, keep Private */
1951 	if (IS_ATOMIC_WRITTEN_PAGE(page))
1952 		return;
1953 
1954 	set_page_private(page, 0);
1955 	ClearPagePrivate(page);
1956 }
1957 
1958 int f2fs_release_page(struct page *page, gfp_t wait)
1959 {
1960 	/* If this is dirty page, keep PagePrivate */
1961 	if (PageDirty(page))
1962 		return 0;
1963 
1964 	/* This is atomic written page, keep Private */
1965 	if (IS_ATOMIC_WRITTEN_PAGE(page))
1966 		return 0;
1967 
1968 	set_page_private(page, 0);
1969 	ClearPagePrivate(page);
1970 	return 1;
1971 }
1972 
1973 /*
1974  * This was copied from __set_page_dirty_buffers which gives higher performance
1975  * in very high speed storages. (e.g., pmem)
1976  */
1977 void f2fs_set_page_dirty_nobuffers(struct page *page)
1978 {
1979 	struct address_space *mapping = page->mapping;
1980 	unsigned long flags;
1981 
1982 	if (unlikely(!mapping))
1983 		return;
1984 
1985 	spin_lock(&mapping->private_lock);
1986 	lock_page_memcg(page);
1987 	SetPageDirty(page);
1988 	spin_unlock(&mapping->private_lock);
1989 
1990 	spin_lock_irqsave(&mapping->tree_lock, flags);
1991 	WARN_ON_ONCE(!PageUptodate(page));
1992 	account_page_dirtied(page, mapping);
1993 	radix_tree_tag_set(&mapping->page_tree,
1994 			page_index(page), PAGECACHE_TAG_DIRTY);
1995 	spin_unlock_irqrestore(&mapping->tree_lock, flags);
1996 	unlock_page_memcg(page);
1997 
1998 	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1999 	return;
2000 }
2001 
2002 static int f2fs_set_data_page_dirty(struct page *page)
2003 {
2004 	struct address_space *mapping = page->mapping;
2005 	struct inode *inode = mapping->host;
2006 
2007 	trace_f2fs_set_page_dirty(page, DATA);
2008 
2009 	if (!PageUptodate(page))
2010 		SetPageUptodate(page);
2011 
2012 	if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2013 		if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2014 			register_inmem_page(inode, page);
2015 			return 1;
2016 		}
2017 		/*
2018 		 * Previously, this page has been registered, we just
2019 		 * return here.
2020 		 */
2021 		return 0;
2022 	}
2023 
2024 	if (!PageDirty(page)) {
2025 		f2fs_set_page_dirty_nobuffers(page);
2026 		update_dirty_page(inode, page);
2027 		return 1;
2028 	}
2029 	return 0;
2030 }
2031 
2032 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2033 {
2034 	struct inode *inode = mapping->host;
2035 
2036 	if (f2fs_has_inline_data(inode))
2037 		return 0;
2038 
2039 	/* make sure allocating whole blocks */
2040 	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2041 		filemap_write_and_wait(mapping);
2042 
2043 	return generic_block_bmap(mapping, block, get_data_block_bmap);
2044 }
2045 
2046 #ifdef CONFIG_MIGRATION
2047 #include <linux/migrate.h>
2048 
2049 int f2fs_migrate_page(struct address_space *mapping,
2050 		struct page *newpage, struct page *page, enum migrate_mode mode)
2051 {
2052 	int rc, extra_count;
2053 	struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2054 	bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2055 
2056 	BUG_ON(PageWriteback(page));
2057 
2058 	/* migrating an atomic written page is safe with the inmem_lock hold */
2059 	if (atomic_written && !mutex_trylock(&fi->inmem_lock))
2060 		return -EAGAIN;
2061 
2062 	/*
2063 	 * A reference is expected if PagePrivate set when move mapping,
2064 	 * however F2FS breaks this for maintaining dirty page counts when
2065 	 * truncating pages. So here adjusting the 'extra_count' make it work.
2066 	 */
2067 	extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2068 	rc = migrate_page_move_mapping(mapping, newpage,
2069 				page, NULL, mode, extra_count);
2070 	if (rc != MIGRATEPAGE_SUCCESS) {
2071 		if (atomic_written)
2072 			mutex_unlock(&fi->inmem_lock);
2073 		return rc;
2074 	}
2075 
2076 	if (atomic_written) {
2077 		struct inmem_pages *cur;
2078 		list_for_each_entry(cur, &fi->inmem_pages, list)
2079 			if (cur->page == page) {
2080 				cur->page = newpage;
2081 				break;
2082 			}
2083 		mutex_unlock(&fi->inmem_lock);
2084 		put_page(page);
2085 		get_page(newpage);
2086 	}
2087 
2088 	if (PagePrivate(page))
2089 		SetPagePrivate(newpage);
2090 	set_page_private(newpage, page_private(page));
2091 
2092 	migrate_page_copy(newpage, page);
2093 
2094 	return MIGRATEPAGE_SUCCESS;
2095 }
2096 #endif
2097 
2098 const struct address_space_operations f2fs_dblock_aops = {
2099 	.readpage	= f2fs_read_data_page,
2100 	.readpages	= f2fs_read_data_pages,
2101 	.writepage	= f2fs_write_data_page,
2102 	.writepages	= f2fs_write_data_pages,
2103 	.write_begin	= f2fs_write_begin,
2104 	.write_end	= f2fs_write_end,
2105 	.set_page_dirty	= f2fs_set_data_page_dirty,
2106 	.invalidatepage	= f2fs_invalidate_page,
2107 	.releasepage	= f2fs_release_page,
2108 	.direct_IO	= f2fs_direct_IO,
2109 	.bmap		= f2fs_bmap,
2110 #ifdef CONFIG_MIGRATION
2111 	.migratepage    = f2fs_migrate_page,
2112 #endif
2113 };
2114