xref: /openbmc/linux/fs/f2fs/data.c (revision cd5d5810)
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/aio.h>
16 #include <linux/writeback.h>
17 #include <linux/backing-dev.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 
22 #include "f2fs.h"
23 #include "node.h"
24 #include "segment.h"
25 #include <trace/events/f2fs.h>
26 
27 /*
28  * Lock ordering for the change of data block address:
29  * ->data_page
30  *  ->node_page
31  *    update block addresses in the node page
32  */
33 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
34 {
35 	struct f2fs_node *rn;
36 	__le32 *addr_array;
37 	struct page *node_page = dn->node_page;
38 	unsigned int ofs_in_node = dn->ofs_in_node;
39 
40 	f2fs_wait_on_page_writeback(node_page, NODE, false);
41 
42 	rn = F2FS_NODE(node_page);
43 
44 	/* Get physical address of data block */
45 	addr_array = blkaddr_in_node(rn);
46 	addr_array[ofs_in_node] = cpu_to_le32(new_addr);
47 	set_page_dirty(node_page);
48 }
49 
50 int reserve_new_block(struct dnode_of_data *dn)
51 {
52 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
53 
54 	if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
55 		return -EPERM;
56 	if (!inc_valid_block_count(sbi, dn->inode, 1))
57 		return -ENOSPC;
58 
59 	trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
60 
61 	__set_data_blkaddr(dn, NEW_ADDR);
62 	dn->data_blkaddr = NEW_ADDR;
63 	sync_inode_page(dn);
64 	return 0;
65 }
66 
67 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
68 					struct buffer_head *bh_result)
69 {
70 	struct f2fs_inode_info *fi = F2FS_I(inode);
71 #ifdef CONFIG_F2FS_STAT_FS
72 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
73 #endif
74 	pgoff_t start_fofs, end_fofs;
75 	block_t start_blkaddr;
76 
77 	read_lock(&fi->ext.ext_lock);
78 	if (fi->ext.len == 0) {
79 		read_unlock(&fi->ext.ext_lock);
80 		return 0;
81 	}
82 
83 #ifdef CONFIG_F2FS_STAT_FS
84 	sbi->total_hit_ext++;
85 #endif
86 	start_fofs = fi->ext.fofs;
87 	end_fofs = fi->ext.fofs + fi->ext.len - 1;
88 	start_blkaddr = fi->ext.blk_addr;
89 
90 	if (pgofs >= start_fofs && pgofs <= end_fofs) {
91 		unsigned int blkbits = inode->i_sb->s_blocksize_bits;
92 		size_t count;
93 
94 		clear_buffer_new(bh_result);
95 		map_bh(bh_result, inode->i_sb,
96 				start_blkaddr + pgofs - start_fofs);
97 		count = end_fofs - pgofs + 1;
98 		if (count < (UINT_MAX >> blkbits))
99 			bh_result->b_size = (count << blkbits);
100 		else
101 			bh_result->b_size = UINT_MAX;
102 
103 #ifdef CONFIG_F2FS_STAT_FS
104 		sbi->read_hit_ext++;
105 #endif
106 		read_unlock(&fi->ext.ext_lock);
107 		return 1;
108 	}
109 	read_unlock(&fi->ext.ext_lock);
110 	return 0;
111 }
112 
113 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
114 {
115 	struct f2fs_inode_info *fi = F2FS_I(dn->inode);
116 	pgoff_t fofs, start_fofs, end_fofs;
117 	block_t start_blkaddr, end_blkaddr;
118 
119 	BUG_ON(blk_addr == NEW_ADDR);
120 	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
121 							dn->ofs_in_node;
122 
123 	/* Update the page address in the parent node */
124 	__set_data_blkaddr(dn, blk_addr);
125 
126 	write_lock(&fi->ext.ext_lock);
127 
128 	start_fofs = fi->ext.fofs;
129 	end_fofs = fi->ext.fofs + fi->ext.len - 1;
130 	start_blkaddr = fi->ext.blk_addr;
131 	end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
132 
133 	/* Drop and initialize the matched extent */
134 	if (fi->ext.len == 1 && fofs == start_fofs)
135 		fi->ext.len = 0;
136 
137 	/* Initial extent */
138 	if (fi->ext.len == 0) {
139 		if (blk_addr != NULL_ADDR) {
140 			fi->ext.fofs = fofs;
141 			fi->ext.blk_addr = blk_addr;
142 			fi->ext.len = 1;
143 		}
144 		goto end_update;
145 	}
146 
147 	/* Front merge */
148 	if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
149 		fi->ext.fofs--;
150 		fi->ext.blk_addr--;
151 		fi->ext.len++;
152 		goto end_update;
153 	}
154 
155 	/* Back merge */
156 	if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
157 		fi->ext.len++;
158 		goto end_update;
159 	}
160 
161 	/* Split the existing extent */
162 	if (fi->ext.len > 1 &&
163 		fofs >= start_fofs && fofs <= end_fofs) {
164 		if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
165 			fi->ext.len = fofs - start_fofs;
166 		} else {
167 			fi->ext.fofs = fofs + 1;
168 			fi->ext.blk_addr = start_blkaddr +
169 					fofs - start_fofs + 1;
170 			fi->ext.len -= fofs - start_fofs + 1;
171 		}
172 		goto end_update;
173 	}
174 	write_unlock(&fi->ext.ext_lock);
175 	return;
176 
177 end_update:
178 	write_unlock(&fi->ext.ext_lock);
179 	sync_inode_page(dn);
180 }
181 
182 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
183 {
184 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
185 	struct address_space *mapping = inode->i_mapping;
186 	struct dnode_of_data dn;
187 	struct page *page;
188 	int err;
189 
190 	page = find_get_page(mapping, index);
191 	if (page && PageUptodate(page))
192 		return page;
193 	f2fs_put_page(page, 0);
194 
195 	set_new_dnode(&dn, inode, NULL, NULL, 0);
196 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
197 	if (err)
198 		return ERR_PTR(err);
199 	f2fs_put_dnode(&dn);
200 
201 	if (dn.data_blkaddr == NULL_ADDR)
202 		return ERR_PTR(-ENOENT);
203 
204 	/* By fallocate(), there is no cached page, but with NEW_ADDR */
205 	if (dn.data_blkaddr == NEW_ADDR)
206 		return ERR_PTR(-EINVAL);
207 
208 	page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
209 	if (!page)
210 		return ERR_PTR(-ENOMEM);
211 
212 	if (PageUptodate(page)) {
213 		unlock_page(page);
214 		return page;
215 	}
216 
217 	err = f2fs_readpage(sbi, page, dn.data_blkaddr,
218 					sync ? READ_SYNC : READA);
219 	if (sync) {
220 		wait_on_page_locked(page);
221 		if (!PageUptodate(page)) {
222 			f2fs_put_page(page, 0);
223 			return ERR_PTR(-EIO);
224 		}
225 	}
226 	return page;
227 }
228 
229 /*
230  * If it tries to access a hole, return an error.
231  * Because, the callers, functions in dir.c and GC, should be able to know
232  * whether this page exists or not.
233  */
234 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
235 {
236 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
237 	struct address_space *mapping = inode->i_mapping;
238 	struct dnode_of_data dn;
239 	struct page *page;
240 	int err;
241 
242 repeat:
243 	page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
244 	if (!page)
245 		return ERR_PTR(-ENOMEM);
246 
247 	set_new_dnode(&dn, inode, NULL, NULL, 0);
248 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
249 	if (err) {
250 		f2fs_put_page(page, 1);
251 		return ERR_PTR(err);
252 	}
253 	f2fs_put_dnode(&dn);
254 
255 	if (dn.data_blkaddr == NULL_ADDR) {
256 		f2fs_put_page(page, 1);
257 		return ERR_PTR(-ENOENT);
258 	}
259 
260 	if (PageUptodate(page))
261 		return page;
262 
263 	/*
264 	 * A new dentry page is allocated but not able to be written, since its
265 	 * new inode page couldn't be allocated due to -ENOSPC.
266 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
267 	 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
268 	 */
269 	if (dn.data_blkaddr == NEW_ADDR) {
270 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
271 		SetPageUptodate(page);
272 		return page;
273 	}
274 
275 	err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
276 	if (err)
277 		return ERR_PTR(err);
278 
279 	lock_page(page);
280 	if (!PageUptodate(page)) {
281 		f2fs_put_page(page, 1);
282 		return ERR_PTR(-EIO);
283 	}
284 	if (page->mapping != mapping) {
285 		f2fs_put_page(page, 1);
286 		goto repeat;
287 	}
288 	return page;
289 }
290 
291 /*
292  * Caller ensures that this data page is never allocated.
293  * A new zero-filled data page is allocated in the page cache.
294  *
295  * Also, caller should grab and release a mutex by calling mutex_lock_op() and
296  * mutex_unlock_op().
297  * Note that, npage is set only by make_empty_dir.
298  */
299 struct page *get_new_data_page(struct inode *inode,
300 		struct page *npage, pgoff_t index, bool new_i_size)
301 {
302 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
303 	struct address_space *mapping = inode->i_mapping;
304 	struct page *page;
305 	struct dnode_of_data dn;
306 	int err;
307 
308 	set_new_dnode(&dn, inode, npage, npage, 0);
309 	err = get_dnode_of_data(&dn, index, ALLOC_NODE);
310 	if (err)
311 		return ERR_PTR(err);
312 
313 	if (dn.data_blkaddr == NULL_ADDR) {
314 		if (reserve_new_block(&dn)) {
315 			if (!npage)
316 				f2fs_put_dnode(&dn);
317 			return ERR_PTR(-ENOSPC);
318 		}
319 	}
320 	if (!npage)
321 		f2fs_put_dnode(&dn);
322 repeat:
323 	page = grab_cache_page(mapping, index);
324 	if (!page)
325 		return ERR_PTR(-ENOMEM);
326 
327 	if (PageUptodate(page))
328 		return page;
329 
330 	if (dn.data_blkaddr == NEW_ADDR) {
331 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
332 		SetPageUptodate(page);
333 	} else {
334 		err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
335 		if (err)
336 			return ERR_PTR(err);
337 		lock_page(page);
338 		if (!PageUptodate(page)) {
339 			f2fs_put_page(page, 1);
340 			return ERR_PTR(-EIO);
341 		}
342 		if (page->mapping != mapping) {
343 			f2fs_put_page(page, 1);
344 			goto repeat;
345 		}
346 	}
347 
348 	if (new_i_size &&
349 		i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
350 		i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
351 		/* Only the directory inode sets new_i_size */
352 		set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
353 		mark_inode_dirty_sync(inode);
354 	}
355 	return page;
356 }
357 
358 static void read_end_io(struct bio *bio, int err)
359 {
360 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
361 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
362 
363 	do {
364 		struct page *page = bvec->bv_page;
365 
366 		if (--bvec >= bio->bi_io_vec)
367 			prefetchw(&bvec->bv_page->flags);
368 
369 		if (uptodate) {
370 			SetPageUptodate(page);
371 		} else {
372 			ClearPageUptodate(page);
373 			SetPageError(page);
374 		}
375 		unlock_page(page);
376 	} while (bvec >= bio->bi_io_vec);
377 	bio_put(bio);
378 }
379 
380 /*
381  * Fill the locked page with data located in the block address.
382  * Return unlocked page.
383  */
384 int f2fs_readpage(struct f2fs_sb_info *sbi, struct page *page,
385 					block_t blk_addr, int type)
386 {
387 	struct block_device *bdev = sbi->sb->s_bdev;
388 	struct bio *bio;
389 
390 	trace_f2fs_readpage(page, blk_addr, type);
391 
392 	down_read(&sbi->bio_sem);
393 
394 	/* Allocate a new bio */
395 	bio = f2fs_bio_alloc(bdev, 1);
396 
397 	/* Initialize the bio */
398 	bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
399 	bio->bi_end_io = read_end_io;
400 
401 	if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
402 		bio_put(bio);
403 		up_read(&sbi->bio_sem);
404 		f2fs_put_page(page, 1);
405 		return -EFAULT;
406 	}
407 
408 	submit_bio(type, bio);
409 	up_read(&sbi->bio_sem);
410 	return 0;
411 }
412 
413 /*
414  * This function should be used by the data read flow only where it
415  * does not check the "create" flag that indicates block allocation.
416  * The reason for this special functionality is to exploit VFS readahead
417  * mechanism.
418  */
419 static int get_data_block_ro(struct inode *inode, sector_t iblock,
420 			struct buffer_head *bh_result, int create)
421 {
422 	unsigned int blkbits = inode->i_sb->s_blocksize_bits;
423 	unsigned maxblocks = bh_result->b_size >> blkbits;
424 	struct dnode_of_data dn;
425 	pgoff_t pgofs;
426 	int err;
427 
428 	/* Get the page offset from the block offset(iblock) */
429 	pgofs =	(pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
430 
431 	if (check_extent_cache(inode, pgofs, bh_result)) {
432 		trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
433 		return 0;
434 	}
435 
436 	/* When reading holes, we need its node page */
437 	set_new_dnode(&dn, inode, NULL, NULL, 0);
438 	err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
439 	if (err) {
440 		trace_f2fs_get_data_block(inode, iblock, bh_result, err);
441 		return (err == -ENOENT) ? 0 : err;
442 	}
443 
444 	/* It does not support data allocation */
445 	BUG_ON(create);
446 
447 	if (dn.data_blkaddr != NEW_ADDR && dn.data_blkaddr != NULL_ADDR) {
448 		int i;
449 		unsigned int end_offset;
450 
451 		end_offset = IS_INODE(dn.node_page) ?
452 				ADDRS_PER_INODE(F2FS_I(inode)) :
453 				ADDRS_PER_BLOCK;
454 
455 		clear_buffer_new(bh_result);
456 
457 		/* Give more consecutive addresses for the read ahead */
458 		for (i = 0; i < end_offset - dn.ofs_in_node; i++)
459 			if (((datablock_addr(dn.node_page,
460 							dn.ofs_in_node + i))
461 				!= (dn.data_blkaddr + i)) || maxblocks == i)
462 				break;
463 		map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
464 		bh_result->b_size = (i << blkbits);
465 	}
466 	f2fs_put_dnode(&dn);
467 	trace_f2fs_get_data_block(inode, iblock, bh_result, 0);
468 	return 0;
469 }
470 
471 static int f2fs_read_data_page(struct file *file, struct page *page)
472 {
473 	return mpage_readpage(page, get_data_block_ro);
474 }
475 
476 static int f2fs_read_data_pages(struct file *file,
477 			struct address_space *mapping,
478 			struct list_head *pages, unsigned nr_pages)
479 {
480 	return mpage_readpages(mapping, pages, nr_pages, get_data_block_ro);
481 }
482 
483 int do_write_data_page(struct page *page)
484 {
485 	struct inode *inode = page->mapping->host;
486 	block_t old_blk_addr, new_blk_addr;
487 	struct dnode_of_data dn;
488 	int err = 0;
489 
490 	set_new_dnode(&dn, inode, NULL, NULL, 0);
491 	err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
492 	if (err)
493 		return err;
494 
495 	old_blk_addr = dn.data_blkaddr;
496 
497 	/* This page is already truncated */
498 	if (old_blk_addr == NULL_ADDR)
499 		goto out_writepage;
500 
501 	set_page_writeback(page);
502 
503 	/*
504 	 * If current allocation needs SSR,
505 	 * it had better in-place writes for updated data.
506 	 */
507 	if (unlikely(old_blk_addr != NEW_ADDR &&
508 			!is_cold_data(page) &&
509 			need_inplace_update(inode))) {
510 		rewrite_data_page(F2FS_SB(inode->i_sb), page,
511 						old_blk_addr);
512 	} else {
513 		write_data_page(inode, page, &dn,
514 				old_blk_addr, &new_blk_addr);
515 		update_extent_cache(new_blk_addr, &dn);
516 	}
517 out_writepage:
518 	f2fs_put_dnode(&dn);
519 	return err;
520 }
521 
522 static int f2fs_write_data_page(struct page *page,
523 					struct writeback_control *wbc)
524 {
525 	struct inode *inode = page->mapping->host;
526 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
527 	loff_t i_size = i_size_read(inode);
528 	const pgoff_t end_index = ((unsigned long long) i_size)
529 							>> PAGE_CACHE_SHIFT;
530 	unsigned offset;
531 	bool need_balance_fs = false;
532 	int err = 0;
533 
534 	if (page->index < end_index)
535 		goto write;
536 
537 	/*
538 	 * If the offset is out-of-range of file size,
539 	 * this page does not have to be written to disk.
540 	 */
541 	offset = i_size & (PAGE_CACHE_SIZE - 1);
542 	if ((page->index >= end_index + 1) || !offset) {
543 		if (S_ISDIR(inode->i_mode)) {
544 			dec_page_count(sbi, F2FS_DIRTY_DENTS);
545 			inode_dec_dirty_dents(inode);
546 		}
547 		goto out;
548 	}
549 
550 	zero_user_segment(page, offset, PAGE_CACHE_SIZE);
551 write:
552 	if (sbi->por_doing) {
553 		err = AOP_WRITEPAGE_ACTIVATE;
554 		goto redirty_out;
555 	}
556 
557 	/* Dentry blocks are controlled by checkpoint */
558 	if (S_ISDIR(inode->i_mode)) {
559 		dec_page_count(sbi, F2FS_DIRTY_DENTS);
560 		inode_dec_dirty_dents(inode);
561 		err = do_write_data_page(page);
562 	} else {
563 		int ilock = mutex_lock_op(sbi);
564 		err = do_write_data_page(page);
565 		mutex_unlock_op(sbi, ilock);
566 		need_balance_fs = true;
567 	}
568 	if (err == -ENOENT)
569 		goto out;
570 	else if (err)
571 		goto redirty_out;
572 
573 	if (wbc->for_reclaim)
574 		f2fs_submit_bio(sbi, DATA, true);
575 
576 	clear_cold_data(page);
577 out:
578 	unlock_page(page);
579 	if (need_balance_fs)
580 		f2fs_balance_fs(sbi);
581 	return 0;
582 
583 redirty_out:
584 	wbc->pages_skipped++;
585 	set_page_dirty(page);
586 	return err;
587 }
588 
589 #define MAX_DESIRED_PAGES_WP	4096
590 
591 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
592 			void *data)
593 {
594 	struct address_space *mapping = data;
595 	int ret = mapping->a_ops->writepage(page, wbc);
596 	mapping_set_error(mapping, ret);
597 	return ret;
598 }
599 
600 static int f2fs_write_data_pages(struct address_space *mapping,
601 			    struct writeback_control *wbc)
602 {
603 	struct inode *inode = mapping->host;
604 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
605 	bool locked = false;
606 	int ret;
607 	long excess_nrtw = 0, desired_nrtw;
608 
609 	/* deal with chardevs and other special file */
610 	if (!mapping->a_ops->writepage)
611 		return 0;
612 
613 	if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
614 		desired_nrtw = MAX_DESIRED_PAGES_WP;
615 		excess_nrtw = desired_nrtw - wbc->nr_to_write;
616 		wbc->nr_to_write = desired_nrtw;
617 	}
618 
619 	if (!S_ISDIR(inode->i_mode)) {
620 		mutex_lock(&sbi->writepages);
621 		locked = true;
622 	}
623 	ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
624 	if (locked)
625 		mutex_unlock(&sbi->writepages);
626 	f2fs_submit_bio(sbi, DATA, (wbc->sync_mode == WB_SYNC_ALL));
627 
628 	remove_dirty_dir_inode(inode);
629 
630 	wbc->nr_to_write -= excess_nrtw;
631 	return ret;
632 }
633 
634 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
635 		loff_t pos, unsigned len, unsigned flags,
636 		struct page **pagep, void **fsdata)
637 {
638 	struct inode *inode = mapping->host;
639 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
640 	struct page *page;
641 	pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
642 	struct dnode_of_data dn;
643 	int err = 0;
644 	int ilock;
645 
646 	f2fs_balance_fs(sbi);
647 repeat:
648 	page = grab_cache_page_write_begin(mapping, index, flags);
649 	if (!page)
650 		return -ENOMEM;
651 	*pagep = page;
652 
653 	ilock = mutex_lock_op(sbi);
654 
655 	set_new_dnode(&dn, inode, NULL, NULL, 0);
656 	err = get_dnode_of_data(&dn, index, ALLOC_NODE);
657 	if (err)
658 		goto err;
659 
660 	if (dn.data_blkaddr == NULL_ADDR)
661 		err = reserve_new_block(&dn);
662 
663 	f2fs_put_dnode(&dn);
664 	if (err)
665 		goto err;
666 
667 	mutex_unlock_op(sbi, ilock);
668 
669 	if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
670 		return 0;
671 
672 	if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
673 		unsigned start = pos & (PAGE_CACHE_SIZE - 1);
674 		unsigned end = start + len;
675 
676 		/* Reading beyond i_size is simple: memset to zero */
677 		zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
678 		goto out;
679 	}
680 
681 	if (dn.data_blkaddr == NEW_ADDR) {
682 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
683 	} else {
684 		err = f2fs_readpage(sbi, page, dn.data_blkaddr, READ_SYNC);
685 		if (err)
686 			return err;
687 		lock_page(page);
688 		if (!PageUptodate(page)) {
689 			f2fs_put_page(page, 1);
690 			return -EIO;
691 		}
692 		if (page->mapping != mapping) {
693 			f2fs_put_page(page, 1);
694 			goto repeat;
695 		}
696 	}
697 out:
698 	SetPageUptodate(page);
699 	clear_cold_data(page);
700 	return 0;
701 
702 err:
703 	mutex_unlock_op(sbi, ilock);
704 	f2fs_put_page(page, 1);
705 	return err;
706 }
707 
708 static int f2fs_write_end(struct file *file,
709 			struct address_space *mapping,
710 			loff_t pos, unsigned len, unsigned copied,
711 			struct page *page, void *fsdata)
712 {
713 	struct inode *inode = page->mapping->host;
714 
715 	SetPageUptodate(page);
716 	set_page_dirty(page);
717 
718 	if (pos + copied > i_size_read(inode)) {
719 		i_size_write(inode, pos + copied);
720 		mark_inode_dirty(inode);
721 		update_inode_page(inode);
722 	}
723 
724 	unlock_page(page);
725 	page_cache_release(page);
726 	return copied;
727 }
728 
729 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
730 		const struct iovec *iov, loff_t offset, unsigned long nr_segs)
731 {
732 	struct file *file = iocb->ki_filp;
733 	struct inode *inode = file->f_mapping->host;
734 
735 	if (rw == WRITE)
736 		return 0;
737 
738 	/* Needs synchronization with the cleaner */
739 	return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
740 						  get_data_block_ro);
741 }
742 
743 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
744 				      unsigned int length)
745 {
746 	struct inode *inode = page->mapping->host;
747 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
748 	if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
749 		dec_page_count(sbi, F2FS_DIRTY_DENTS);
750 		inode_dec_dirty_dents(inode);
751 	}
752 	ClearPagePrivate(page);
753 }
754 
755 static int f2fs_release_data_page(struct page *page, gfp_t wait)
756 {
757 	ClearPagePrivate(page);
758 	return 1;
759 }
760 
761 static int f2fs_set_data_page_dirty(struct page *page)
762 {
763 	struct address_space *mapping = page->mapping;
764 	struct inode *inode = mapping->host;
765 
766 	SetPageUptodate(page);
767 	if (!PageDirty(page)) {
768 		__set_page_dirty_nobuffers(page);
769 		set_dirty_dir_page(inode, page);
770 		return 1;
771 	}
772 	return 0;
773 }
774 
775 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
776 {
777 	return generic_block_bmap(mapping, block, get_data_block_ro);
778 }
779 
780 const struct address_space_operations f2fs_dblock_aops = {
781 	.readpage	= f2fs_read_data_page,
782 	.readpages	= f2fs_read_data_pages,
783 	.writepage	= f2fs_write_data_page,
784 	.writepages	= f2fs_write_data_pages,
785 	.write_begin	= f2fs_write_begin,
786 	.write_end	= f2fs_write_end,
787 	.set_page_dirty	= f2fs_set_data_page_dirty,
788 	.invalidatepage	= f2fs_invalidate_data_page,
789 	.releasepage	= f2fs_release_data_page,
790 	.direct_IO	= f2fs_direct_IO,
791 	.bmap		= f2fs_bmap,
792 };
793