xref: /openbmc/linux/fs/f2fs/data.c (revision 8e9356c6)
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 static void f2fs_read_end_io(struct bio *bio, int err)
28 {
29 	struct bio_vec *bvec;
30 	int i;
31 
32 	bio_for_each_segment_all(bvec, bio, i) {
33 		struct page *page = bvec->bv_page;
34 
35 		if (!err) {
36 			SetPageUptodate(page);
37 		} else {
38 			ClearPageUptodate(page);
39 			SetPageError(page);
40 		}
41 		unlock_page(page);
42 	}
43 	bio_put(bio);
44 }
45 
46 static void f2fs_write_end_io(struct bio *bio, int err)
47 {
48 	struct f2fs_sb_info *sbi = F2FS_SB(bio->bi_io_vec->bv_page->mapping->host->i_sb);
49 	struct bio_vec *bvec;
50 	int i;
51 
52 	bio_for_each_segment_all(bvec, bio, i) {
53 		struct page *page = bvec->bv_page;
54 
55 		if (unlikely(err)) {
56 			SetPageError(page);
57 			set_bit(AS_EIO, &page->mapping->flags);
58 			set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
59 			sbi->sb->s_flags |= MS_RDONLY;
60 		}
61 		end_page_writeback(page);
62 		dec_page_count(sbi, F2FS_WRITEBACK);
63 	}
64 
65 	if (bio->bi_private)
66 		complete(bio->bi_private);
67 
68 	if (!get_pages(sbi, F2FS_WRITEBACK) &&
69 			!list_empty(&sbi->cp_wait.task_list))
70 		wake_up(&sbi->cp_wait);
71 
72 	bio_put(bio);
73 }
74 
75 /*
76  * Low-level block read/write IO operations.
77  */
78 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
79 				int npages, bool is_read)
80 {
81 	struct bio *bio;
82 
83 	/* No failure on bio allocation */
84 	bio = bio_alloc(GFP_NOIO, npages);
85 
86 	bio->bi_bdev = sbi->sb->s_bdev;
87 	bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
88 	bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
89 
90 	return bio;
91 }
92 
93 static void __submit_merged_bio(struct f2fs_bio_info *io)
94 {
95 	struct f2fs_io_info *fio = &io->fio;
96 	int rw;
97 
98 	if (!io->bio)
99 		return;
100 
101 	rw = fio->rw;
102 
103 	if (is_read_io(rw)) {
104 		trace_f2fs_submit_read_bio(io->sbi->sb, rw,
105 						fio->type, io->bio);
106 		submit_bio(rw, io->bio);
107 	} else {
108 		trace_f2fs_submit_write_bio(io->sbi->sb, rw,
109 						fio->type, io->bio);
110 		/*
111 		 * META_FLUSH is only from the checkpoint procedure, and we
112 		 * should wait this metadata bio for FS consistency.
113 		 */
114 		if (fio->type == META_FLUSH) {
115 			DECLARE_COMPLETION_ONSTACK(wait);
116 			io->bio->bi_private = &wait;
117 			submit_bio(rw, io->bio);
118 			wait_for_completion(&wait);
119 		} else {
120 			submit_bio(rw, io->bio);
121 		}
122 	}
123 
124 	io->bio = NULL;
125 }
126 
127 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
128 				enum page_type type, int rw)
129 {
130 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
131 	struct f2fs_bio_info *io;
132 
133 	io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
134 
135 	mutex_lock(&io->io_mutex);
136 
137 	/* change META to META_FLUSH in the checkpoint procedure */
138 	if (type >= META_FLUSH) {
139 		io->fio.type = META_FLUSH;
140 		io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
141 	}
142 	__submit_merged_bio(io);
143 	mutex_unlock(&io->io_mutex);
144 }
145 
146 /*
147  * Fill the locked page with data located in the block address.
148  * Return unlocked page.
149  */
150 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
151 					block_t blk_addr, int rw)
152 {
153 	struct bio *bio;
154 
155 	trace_f2fs_submit_page_bio(page, blk_addr, rw);
156 
157 	/* Allocate a new bio */
158 	bio = __bio_alloc(sbi, blk_addr, 1, is_read_io(rw));
159 
160 	if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
161 		bio_put(bio);
162 		f2fs_put_page(page, 1);
163 		return -EFAULT;
164 	}
165 
166 	submit_bio(rw, bio);
167 	return 0;
168 }
169 
170 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
171 			block_t blk_addr, struct f2fs_io_info *fio)
172 {
173 	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
174 	struct f2fs_bio_info *io;
175 	bool is_read = is_read_io(fio->rw);
176 
177 	io = is_read ? &sbi->read_io : &sbi->write_io[btype];
178 
179 	verify_block_addr(sbi, blk_addr);
180 
181 	mutex_lock(&io->io_mutex);
182 
183 	if (!is_read)
184 		inc_page_count(sbi, F2FS_WRITEBACK);
185 
186 	if (io->bio && (io->last_block_in_bio != blk_addr - 1 ||
187 						io->fio.rw != fio->rw))
188 		__submit_merged_bio(io);
189 alloc_new:
190 	if (io->bio == NULL) {
191 		int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
192 
193 		io->bio = __bio_alloc(sbi, blk_addr, bio_blocks, is_read);
194 		io->fio = *fio;
195 	}
196 
197 	if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
198 							PAGE_CACHE_SIZE) {
199 		__submit_merged_bio(io);
200 		goto alloc_new;
201 	}
202 
203 	io->last_block_in_bio = blk_addr;
204 
205 	mutex_unlock(&io->io_mutex);
206 	trace_f2fs_submit_page_mbio(page, fio->rw, fio->type, blk_addr);
207 }
208 
209 /*
210  * Lock ordering for the change of data block address:
211  * ->data_page
212  *  ->node_page
213  *    update block addresses in the node page
214  */
215 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
216 {
217 	struct f2fs_node *rn;
218 	__le32 *addr_array;
219 	struct page *node_page = dn->node_page;
220 	unsigned int ofs_in_node = dn->ofs_in_node;
221 
222 	f2fs_wait_on_page_writeback(node_page, NODE);
223 
224 	rn = F2FS_NODE(node_page);
225 
226 	/* Get physical address of data block */
227 	addr_array = blkaddr_in_node(rn);
228 	addr_array[ofs_in_node] = cpu_to_le32(new_addr);
229 	set_page_dirty(node_page);
230 }
231 
232 int reserve_new_block(struct dnode_of_data *dn)
233 {
234 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
235 
236 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
237 		return -EPERM;
238 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
239 		return -ENOSPC;
240 
241 	trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
242 
243 	__set_data_blkaddr(dn, NEW_ADDR);
244 	dn->data_blkaddr = NEW_ADDR;
245 	mark_inode_dirty(dn->inode);
246 	sync_inode_page(dn);
247 	return 0;
248 }
249 
250 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
251 {
252 	bool need_put = dn->inode_page ? false : true;
253 	int err;
254 
255 	/* if inode_page exists, index should be zero */
256 	f2fs_bug_on(!need_put && index);
257 
258 	err = get_dnode_of_data(dn, index, ALLOC_NODE);
259 	if (err)
260 		return err;
261 
262 	if (dn->data_blkaddr == NULL_ADDR)
263 		err = reserve_new_block(dn);
264 	if (err || need_put)
265 		f2fs_put_dnode(dn);
266 	return err;
267 }
268 
269 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
270 					struct buffer_head *bh_result)
271 {
272 	struct f2fs_inode_info *fi = F2FS_I(inode);
273 	pgoff_t start_fofs, end_fofs;
274 	block_t start_blkaddr;
275 
276 	if (is_inode_flag_set(fi, FI_NO_EXTENT))
277 		return 0;
278 
279 	read_lock(&fi->ext.ext_lock);
280 	if (fi->ext.len == 0) {
281 		read_unlock(&fi->ext.ext_lock);
282 		return 0;
283 	}
284 
285 	stat_inc_total_hit(inode->i_sb);
286 
287 	start_fofs = fi->ext.fofs;
288 	end_fofs = fi->ext.fofs + fi->ext.len - 1;
289 	start_blkaddr = fi->ext.blk_addr;
290 
291 	if (pgofs >= start_fofs && pgofs <= end_fofs) {
292 		unsigned int blkbits = inode->i_sb->s_blocksize_bits;
293 		size_t count;
294 
295 		clear_buffer_new(bh_result);
296 		map_bh(bh_result, inode->i_sb,
297 				start_blkaddr + pgofs - start_fofs);
298 		count = end_fofs - pgofs + 1;
299 		if (count < (UINT_MAX >> blkbits))
300 			bh_result->b_size = (count << blkbits);
301 		else
302 			bh_result->b_size = UINT_MAX;
303 
304 		stat_inc_read_hit(inode->i_sb);
305 		read_unlock(&fi->ext.ext_lock);
306 		return 1;
307 	}
308 	read_unlock(&fi->ext.ext_lock);
309 	return 0;
310 }
311 
312 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
313 {
314 	struct f2fs_inode_info *fi = F2FS_I(dn->inode);
315 	pgoff_t fofs, start_fofs, end_fofs;
316 	block_t start_blkaddr, end_blkaddr;
317 	int need_update = true;
318 
319 	f2fs_bug_on(blk_addr == NEW_ADDR);
320 	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
321 							dn->ofs_in_node;
322 
323 	/* Update the page address in the parent node */
324 	__set_data_blkaddr(dn, blk_addr);
325 
326 	if (is_inode_flag_set(fi, FI_NO_EXTENT))
327 		return;
328 
329 	write_lock(&fi->ext.ext_lock);
330 
331 	start_fofs = fi->ext.fofs;
332 	end_fofs = fi->ext.fofs + fi->ext.len - 1;
333 	start_blkaddr = fi->ext.blk_addr;
334 	end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
335 
336 	/* Drop and initialize the matched extent */
337 	if (fi->ext.len == 1 && fofs == start_fofs)
338 		fi->ext.len = 0;
339 
340 	/* Initial extent */
341 	if (fi->ext.len == 0) {
342 		if (blk_addr != NULL_ADDR) {
343 			fi->ext.fofs = fofs;
344 			fi->ext.blk_addr = blk_addr;
345 			fi->ext.len = 1;
346 		}
347 		goto end_update;
348 	}
349 
350 	/* Front merge */
351 	if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
352 		fi->ext.fofs--;
353 		fi->ext.blk_addr--;
354 		fi->ext.len++;
355 		goto end_update;
356 	}
357 
358 	/* Back merge */
359 	if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
360 		fi->ext.len++;
361 		goto end_update;
362 	}
363 
364 	/* Split the existing extent */
365 	if (fi->ext.len > 1 &&
366 		fofs >= start_fofs && fofs <= end_fofs) {
367 		if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
368 			fi->ext.len = fofs - start_fofs;
369 		} else {
370 			fi->ext.fofs = fofs + 1;
371 			fi->ext.blk_addr = start_blkaddr +
372 					fofs - start_fofs + 1;
373 			fi->ext.len -= fofs - start_fofs + 1;
374 		}
375 	} else {
376 		need_update = false;
377 	}
378 
379 	/* Finally, if the extent is very fragmented, let's drop the cache. */
380 	if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
381 		fi->ext.len = 0;
382 		set_inode_flag(fi, FI_NO_EXTENT);
383 		need_update = true;
384 	}
385 end_update:
386 	write_unlock(&fi->ext.ext_lock);
387 	if (need_update)
388 		sync_inode_page(dn);
389 	return;
390 }
391 
392 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
393 {
394 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
395 	struct address_space *mapping = inode->i_mapping;
396 	struct dnode_of_data dn;
397 	struct page *page;
398 	int err;
399 
400 	page = find_get_page(mapping, index);
401 	if (page && PageUptodate(page))
402 		return page;
403 	f2fs_put_page(page, 0);
404 
405 	set_new_dnode(&dn, inode, NULL, NULL, 0);
406 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
407 	if (err)
408 		return ERR_PTR(err);
409 	f2fs_put_dnode(&dn);
410 
411 	if (dn.data_blkaddr == NULL_ADDR)
412 		return ERR_PTR(-ENOENT);
413 
414 	/* By fallocate(), there is no cached page, but with NEW_ADDR */
415 	if (unlikely(dn.data_blkaddr == NEW_ADDR))
416 		return ERR_PTR(-EINVAL);
417 
418 	page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
419 	if (!page)
420 		return ERR_PTR(-ENOMEM);
421 
422 	if (PageUptodate(page)) {
423 		unlock_page(page);
424 		return page;
425 	}
426 
427 	err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
428 					sync ? READ_SYNC : READA);
429 	if (err)
430 		return ERR_PTR(err);
431 
432 	if (sync) {
433 		wait_on_page_locked(page);
434 		if (unlikely(!PageUptodate(page))) {
435 			f2fs_put_page(page, 0);
436 			return ERR_PTR(-EIO);
437 		}
438 	}
439 	return page;
440 }
441 
442 /*
443  * If it tries to access a hole, return an error.
444  * Because, the callers, functions in dir.c and GC, should be able to know
445  * whether this page exists or not.
446  */
447 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
448 {
449 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
450 	struct address_space *mapping = inode->i_mapping;
451 	struct dnode_of_data dn;
452 	struct page *page;
453 	int err;
454 
455 repeat:
456 	page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
457 	if (!page)
458 		return ERR_PTR(-ENOMEM);
459 
460 	set_new_dnode(&dn, inode, NULL, NULL, 0);
461 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
462 	if (err) {
463 		f2fs_put_page(page, 1);
464 		return ERR_PTR(err);
465 	}
466 	f2fs_put_dnode(&dn);
467 
468 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
469 		f2fs_put_page(page, 1);
470 		return ERR_PTR(-ENOENT);
471 	}
472 
473 	if (PageUptodate(page))
474 		return page;
475 
476 	/*
477 	 * A new dentry page is allocated but not able to be written, since its
478 	 * new inode page couldn't be allocated due to -ENOSPC.
479 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
480 	 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
481 	 */
482 	if (dn.data_blkaddr == NEW_ADDR) {
483 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
484 		SetPageUptodate(page);
485 		return page;
486 	}
487 
488 	err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, READ_SYNC);
489 	if (err)
490 		return ERR_PTR(err);
491 
492 	lock_page(page);
493 	if (unlikely(!PageUptodate(page))) {
494 		f2fs_put_page(page, 1);
495 		return ERR_PTR(-EIO);
496 	}
497 	if (unlikely(page->mapping != mapping)) {
498 		f2fs_put_page(page, 1);
499 		goto repeat;
500 	}
501 	return page;
502 }
503 
504 /*
505  * Caller ensures that this data page is never allocated.
506  * A new zero-filled data page is allocated in the page cache.
507  *
508  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
509  * f2fs_unlock_op().
510  * Note that, ipage is set only by make_empty_dir.
511  */
512 struct page *get_new_data_page(struct inode *inode,
513 		struct page *ipage, pgoff_t index, bool new_i_size)
514 {
515 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
516 	struct address_space *mapping = inode->i_mapping;
517 	struct page *page;
518 	struct dnode_of_data dn;
519 	int err;
520 
521 	set_new_dnode(&dn, inode, ipage, NULL, 0);
522 	err = f2fs_reserve_block(&dn, index);
523 	if (err)
524 		return ERR_PTR(err);
525 repeat:
526 	page = grab_cache_page(mapping, index);
527 	if (!page) {
528 		err = -ENOMEM;
529 		goto put_err;
530 	}
531 
532 	if (PageUptodate(page))
533 		return page;
534 
535 	if (dn.data_blkaddr == NEW_ADDR) {
536 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
537 		SetPageUptodate(page);
538 	} else {
539 		err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
540 								READ_SYNC);
541 		if (err)
542 			goto put_err;
543 
544 		lock_page(page);
545 		if (unlikely(!PageUptodate(page))) {
546 			f2fs_put_page(page, 1);
547 			err = -EIO;
548 			goto put_err;
549 		}
550 		if (unlikely(page->mapping != mapping)) {
551 			f2fs_put_page(page, 1);
552 			goto repeat;
553 		}
554 	}
555 
556 	if (new_i_size &&
557 		i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
558 		i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
559 		/* Only the directory inode sets new_i_size */
560 		set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
561 	}
562 	return page;
563 
564 put_err:
565 	f2fs_put_dnode(&dn);
566 	return ERR_PTR(err);
567 }
568 
569 static int __allocate_data_block(struct dnode_of_data *dn)
570 {
571 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
572 	struct f2fs_summary sum;
573 	block_t new_blkaddr;
574 	struct node_info ni;
575 	int type;
576 
577 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
578 		return -EPERM;
579 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
580 		return -ENOSPC;
581 
582 	__set_data_blkaddr(dn, NEW_ADDR);
583 	dn->data_blkaddr = NEW_ADDR;
584 
585 	get_node_info(sbi, dn->nid, &ni);
586 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
587 
588 	type = CURSEG_WARM_DATA;
589 
590 	allocate_data_block(sbi, NULL, NULL_ADDR, &new_blkaddr, &sum, type);
591 
592 	/* direct IO doesn't use extent cache to maximize the performance */
593 	set_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
594 	update_extent_cache(new_blkaddr, dn);
595 	clear_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
596 
597 	dn->data_blkaddr = new_blkaddr;
598 	return 0;
599 }
600 
601 /*
602  * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
603  * If original data blocks are allocated, then give them to blockdev.
604  * Otherwise,
605  *     a. preallocate requested block addresses
606  *     b. do not use extent cache for better performance
607  *     c. give the block addresses to blockdev
608  */
609 static int get_data_block(struct inode *inode, sector_t iblock,
610 			struct buffer_head *bh_result, int create)
611 {
612 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
613 	unsigned int blkbits = inode->i_sb->s_blocksize_bits;
614 	unsigned maxblocks = bh_result->b_size >> blkbits;
615 	struct dnode_of_data dn;
616 	int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
617 	pgoff_t pgofs, end_offset;
618 	int err = 0, ofs = 1;
619 	bool allocated = false;
620 
621 	/* Get the page offset from the block offset(iblock) */
622 	pgofs =	(pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
623 
624 	if (check_extent_cache(inode, pgofs, bh_result))
625 		goto out;
626 
627 	if (create)
628 		f2fs_lock_op(sbi);
629 
630 	/* When reading holes, we need its node page */
631 	set_new_dnode(&dn, inode, NULL, NULL, 0);
632 	err = get_dnode_of_data(&dn, pgofs, mode);
633 	if (err) {
634 		if (err == -ENOENT)
635 			err = 0;
636 		goto unlock_out;
637 	}
638 	if (dn.data_blkaddr == NEW_ADDR)
639 		goto put_out;
640 
641 	if (dn.data_blkaddr != NULL_ADDR) {
642 		map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
643 	} else if (create) {
644 		err = __allocate_data_block(&dn);
645 		if (err)
646 			goto put_out;
647 		allocated = true;
648 		map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
649 	} else {
650 		goto put_out;
651 	}
652 
653 	end_offset = IS_INODE(dn.node_page) ?
654 			ADDRS_PER_INODE(F2FS_I(inode)) : ADDRS_PER_BLOCK;
655 	bh_result->b_size = (((size_t)1) << blkbits);
656 	dn.ofs_in_node++;
657 	pgofs++;
658 
659 get_next:
660 	if (dn.ofs_in_node >= end_offset) {
661 		if (allocated)
662 			sync_inode_page(&dn);
663 		allocated = false;
664 		f2fs_put_dnode(&dn);
665 
666 		set_new_dnode(&dn, inode, NULL, NULL, 0);
667 		err = get_dnode_of_data(&dn, pgofs, mode);
668 		if (err) {
669 			if (err == -ENOENT)
670 				err = 0;
671 			goto unlock_out;
672 		}
673 		if (dn.data_blkaddr == NEW_ADDR)
674 			goto put_out;
675 
676 		end_offset = IS_INODE(dn.node_page) ?
677 			ADDRS_PER_INODE(F2FS_I(inode)) : ADDRS_PER_BLOCK;
678 	}
679 
680 	if (maxblocks > (bh_result->b_size >> blkbits)) {
681 		block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
682 		if (blkaddr == NULL_ADDR && create) {
683 			err = __allocate_data_block(&dn);
684 			if (err)
685 				goto sync_out;
686 			allocated = true;
687 			blkaddr = dn.data_blkaddr;
688 		}
689 		/* Give more consecutive addresses for the read ahead */
690 		if (blkaddr == (bh_result->b_blocknr + ofs)) {
691 			ofs++;
692 			dn.ofs_in_node++;
693 			pgofs++;
694 			bh_result->b_size += (((size_t)1) << blkbits);
695 			goto get_next;
696 		}
697 	}
698 sync_out:
699 	if (allocated)
700 		sync_inode_page(&dn);
701 put_out:
702 	f2fs_put_dnode(&dn);
703 unlock_out:
704 	if (create)
705 		f2fs_unlock_op(sbi);
706 out:
707 	trace_f2fs_get_data_block(inode, iblock, bh_result, err);
708 	return err;
709 }
710 
711 static int f2fs_read_data_page(struct file *file, struct page *page)
712 {
713 	struct inode *inode = page->mapping->host;
714 	int ret;
715 
716 	/* If the file has inline data, try to read it directlly */
717 	if (f2fs_has_inline_data(inode))
718 		ret = f2fs_read_inline_data(inode, page);
719 	else
720 		ret = mpage_readpage(page, get_data_block);
721 
722 	return ret;
723 }
724 
725 static int f2fs_read_data_pages(struct file *file,
726 			struct address_space *mapping,
727 			struct list_head *pages, unsigned nr_pages)
728 {
729 	struct inode *inode = file->f_mapping->host;
730 
731 	/* If the file has inline data, skip readpages */
732 	if (f2fs_has_inline_data(inode))
733 		return 0;
734 
735 	return mpage_readpages(mapping, pages, nr_pages, get_data_block);
736 }
737 
738 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
739 {
740 	struct inode *inode = page->mapping->host;
741 	block_t old_blkaddr, new_blkaddr;
742 	struct dnode_of_data dn;
743 	int err = 0;
744 
745 	set_new_dnode(&dn, inode, NULL, NULL, 0);
746 	err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
747 	if (err)
748 		return err;
749 
750 	old_blkaddr = dn.data_blkaddr;
751 
752 	/* This page is already truncated */
753 	if (old_blkaddr == NULL_ADDR)
754 		goto out_writepage;
755 
756 	set_page_writeback(page);
757 
758 	/*
759 	 * If current allocation needs SSR,
760 	 * it had better in-place writes for updated data.
761 	 */
762 	if (unlikely(old_blkaddr != NEW_ADDR &&
763 			!is_cold_data(page) &&
764 			need_inplace_update(inode))) {
765 		rewrite_data_page(page, old_blkaddr, fio);
766 	} else {
767 		write_data_page(page, &dn, &new_blkaddr, fio);
768 		update_extent_cache(new_blkaddr, &dn);
769 	}
770 out_writepage:
771 	f2fs_put_dnode(&dn);
772 	return err;
773 }
774 
775 static int f2fs_write_data_page(struct page *page,
776 					struct writeback_control *wbc)
777 {
778 	struct inode *inode = page->mapping->host;
779 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
780 	loff_t i_size = i_size_read(inode);
781 	const pgoff_t end_index = ((unsigned long long) i_size)
782 							>> PAGE_CACHE_SHIFT;
783 	unsigned offset = 0;
784 	bool need_balance_fs = false;
785 	int err = 0;
786 	struct f2fs_io_info fio = {
787 		.type = DATA,
788 		.rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
789 	};
790 
791 	if (page->index < end_index)
792 		goto write;
793 
794 	/*
795 	 * If the offset is out-of-range of file size,
796 	 * this page does not have to be written to disk.
797 	 */
798 	offset = i_size & (PAGE_CACHE_SIZE - 1);
799 	if ((page->index >= end_index + 1) || !offset) {
800 		if (S_ISDIR(inode->i_mode)) {
801 			dec_page_count(sbi, F2FS_DIRTY_DENTS);
802 			inode_dec_dirty_dents(inode);
803 		}
804 		goto out;
805 	}
806 
807 	zero_user_segment(page, offset, PAGE_CACHE_SIZE);
808 write:
809 	if (unlikely(sbi->por_doing)) {
810 		err = AOP_WRITEPAGE_ACTIVATE;
811 		goto redirty_out;
812 	}
813 
814 	/* Dentry blocks are controlled by checkpoint */
815 	if (S_ISDIR(inode->i_mode)) {
816 		dec_page_count(sbi, F2FS_DIRTY_DENTS);
817 		inode_dec_dirty_dents(inode);
818 		err = do_write_data_page(page, &fio);
819 	} else {
820 		f2fs_lock_op(sbi);
821 
822 		if (f2fs_has_inline_data(inode) || f2fs_may_inline(inode)) {
823 			err = f2fs_write_inline_data(inode, page, offset);
824 			f2fs_unlock_op(sbi);
825 			goto out;
826 		} else {
827 			err = do_write_data_page(page, &fio);
828 		}
829 
830 		f2fs_unlock_op(sbi);
831 		need_balance_fs = true;
832 	}
833 	if (err == -ENOENT)
834 		goto out;
835 	else if (err)
836 		goto redirty_out;
837 
838 	if (wbc->for_reclaim) {
839 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
840 		need_balance_fs = false;
841 	}
842 
843 	clear_cold_data(page);
844 out:
845 	unlock_page(page);
846 	if (need_balance_fs)
847 		f2fs_balance_fs(sbi);
848 	return 0;
849 
850 redirty_out:
851 	wbc->pages_skipped++;
852 	set_page_dirty(page);
853 	return err;
854 }
855 
856 #define MAX_DESIRED_PAGES_WP	4096
857 
858 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
859 			void *data)
860 {
861 	struct address_space *mapping = data;
862 	int ret = mapping->a_ops->writepage(page, wbc);
863 	mapping_set_error(mapping, ret);
864 	return ret;
865 }
866 
867 static int f2fs_write_data_pages(struct address_space *mapping,
868 			    struct writeback_control *wbc)
869 {
870 	struct inode *inode = mapping->host;
871 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
872 	bool locked = false;
873 	int ret;
874 	long excess_nrtw = 0, desired_nrtw;
875 
876 	/* deal with chardevs and other special file */
877 	if (!mapping->a_ops->writepage)
878 		return 0;
879 
880 	if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
881 		desired_nrtw = MAX_DESIRED_PAGES_WP;
882 		excess_nrtw = desired_nrtw - wbc->nr_to_write;
883 		wbc->nr_to_write = desired_nrtw;
884 	}
885 
886 	if (!S_ISDIR(inode->i_mode)) {
887 		mutex_lock(&sbi->writepages);
888 		locked = true;
889 	}
890 	ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
891 	if (locked)
892 		mutex_unlock(&sbi->writepages);
893 
894 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
895 
896 	remove_dirty_dir_inode(inode);
897 
898 	wbc->nr_to_write -= excess_nrtw;
899 	return ret;
900 }
901 
902 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
903 		loff_t pos, unsigned len, unsigned flags,
904 		struct page **pagep, void **fsdata)
905 {
906 	struct inode *inode = mapping->host;
907 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
908 	struct page *page;
909 	pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
910 	struct dnode_of_data dn;
911 	int err = 0;
912 
913 	f2fs_balance_fs(sbi);
914 repeat:
915 	err = f2fs_convert_inline_data(inode, pos + len);
916 	if (err)
917 		return err;
918 
919 	page = grab_cache_page_write_begin(mapping, index, flags);
920 	if (!page)
921 		return -ENOMEM;
922 	*pagep = page;
923 
924 	if (f2fs_has_inline_data(inode) && (pos + len) <= MAX_INLINE_DATA)
925 		goto inline_data;
926 
927 	f2fs_lock_op(sbi);
928 	set_new_dnode(&dn, inode, NULL, NULL, 0);
929 	err = f2fs_reserve_block(&dn, index);
930 	f2fs_unlock_op(sbi);
931 
932 	if (err) {
933 		f2fs_put_page(page, 1);
934 		return err;
935 	}
936 inline_data:
937 	if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
938 		return 0;
939 
940 	if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
941 		unsigned start = pos & (PAGE_CACHE_SIZE - 1);
942 		unsigned end = start + len;
943 
944 		/* Reading beyond i_size is simple: memset to zero */
945 		zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
946 		goto out;
947 	}
948 
949 	if (dn.data_blkaddr == NEW_ADDR) {
950 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
951 	} else {
952 		if (f2fs_has_inline_data(inode))
953 			err = f2fs_read_inline_data(inode, page);
954 		else
955 			err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
956 							READ_SYNC);
957 		if (err)
958 			return err;
959 		lock_page(page);
960 		if (unlikely(!PageUptodate(page))) {
961 			f2fs_put_page(page, 1);
962 			return -EIO;
963 		}
964 		if (unlikely(page->mapping != mapping)) {
965 			f2fs_put_page(page, 1);
966 			goto repeat;
967 		}
968 	}
969 out:
970 	SetPageUptodate(page);
971 	clear_cold_data(page);
972 	return 0;
973 }
974 
975 static int f2fs_write_end(struct file *file,
976 			struct address_space *mapping,
977 			loff_t pos, unsigned len, unsigned copied,
978 			struct page *page, void *fsdata)
979 {
980 	struct inode *inode = page->mapping->host;
981 
982 	SetPageUptodate(page);
983 	set_page_dirty(page);
984 
985 	if (pos + copied > i_size_read(inode)) {
986 		i_size_write(inode, pos + copied);
987 		mark_inode_dirty(inode);
988 		update_inode_page(inode);
989 	}
990 
991 	f2fs_put_page(page, 1);
992 	return copied;
993 }
994 
995 static int check_direct_IO(struct inode *inode, int rw,
996 		const struct iovec *iov, loff_t offset, unsigned long nr_segs)
997 {
998 	unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
999 	int i;
1000 
1001 	if (rw == READ)
1002 		return 0;
1003 
1004 	if (offset & blocksize_mask)
1005 		return -EINVAL;
1006 
1007 	for (i = 0; i < nr_segs; i++)
1008 		if (iov[i].iov_len & blocksize_mask)
1009 			return -EINVAL;
1010 	return 0;
1011 }
1012 
1013 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1014 		const struct iovec *iov, loff_t offset, unsigned long nr_segs)
1015 {
1016 	struct file *file = iocb->ki_filp;
1017 	struct inode *inode = file->f_mapping->host;
1018 
1019 	/* Let buffer I/O handle the inline data case. */
1020 	if (f2fs_has_inline_data(inode))
1021 		return 0;
1022 
1023 	if (check_direct_IO(inode, rw, iov, offset, nr_segs))
1024 		return 0;
1025 
1026 	return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1027 							get_data_block);
1028 }
1029 
1030 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
1031 				      unsigned int length)
1032 {
1033 	struct inode *inode = page->mapping->host;
1034 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1035 	if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
1036 		dec_page_count(sbi, F2FS_DIRTY_DENTS);
1037 		inode_dec_dirty_dents(inode);
1038 	}
1039 	ClearPagePrivate(page);
1040 }
1041 
1042 static int f2fs_release_data_page(struct page *page, gfp_t wait)
1043 {
1044 	ClearPagePrivate(page);
1045 	return 1;
1046 }
1047 
1048 static int f2fs_set_data_page_dirty(struct page *page)
1049 {
1050 	struct address_space *mapping = page->mapping;
1051 	struct inode *inode = mapping->host;
1052 
1053 	trace_f2fs_set_page_dirty(page, DATA);
1054 
1055 	SetPageUptodate(page);
1056 	mark_inode_dirty(inode);
1057 
1058 	if (!PageDirty(page)) {
1059 		__set_page_dirty_nobuffers(page);
1060 		set_dirty_dir_page(inode, page);
1061 		return 1;
1062 	}
1063 	return 0;
1064 }
1065 
1066 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1067 {
1068 	return generic_block_bmap(mapping, block, get_data_block);
1069 }
1070 
1071 const struct address_space_operations f2fs_dblock_aops = {
1072 	.readpage	= f2fs_read_data_page,
1073 	.readpages	= f2fs_read_data_pages,
1074 	.writepage	= f2fs_write_data_page,
1075 	.writepages	= f2fs_write_data_pages,
1076 	.write_begin	= f2fs_write_begin,
1077 	.write_end	= f2fs_write_end,
1078 	.set_page_dirty	= f2fs_set_data_page_dirty,
1079 	.invalidatepage	= f2fs_invalidate_data_page,
1080 	.releasepage	= f2fs_release_data_page,
1081 	.direct_IO	= f2fs_direct_IO,
1082 	.bmap		= f2fs_bmap,
1083 };
1084