xref: /openbmc/linux/fs/f2fs/data.c (revision 034f90b3)
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.h"
26 #include <trace/events/f2fs.h>
27 
28 static void f2fs_read_end_io(struct bio *bio, int err)
29 {
30 	struct bio_vec *bvec;
31 	int i;
32 
33 	bio_for_each_segment_all(bvec, bio, i) {
34 		struct page *page = bvec->bv_page;
35 
36 		if (!err) {
37 			SetPageUptodate(page);
38 		} else {
39 			ClearPageUptodate(page);
40 			SetPageError(page);
41 		}
42 		unlock_page(page);
43 	}
44 	bio_put(bio);
45 }
46 
47 static void f2fs_write_end_io(struct bio *bio, int err)
48 {
49 	struct f2fs_sb_info *sbi = bio->bi_private;
50 	struct bio_vec *bvec;
51 	int i;
52 
53 	bio_for_each_segment_all(bvec, bio, i) {
54 		struct page *page = bvec->bv_page;
55 
56 		if (unlikely(err)) {
57 			set_page_dirty(page);
58 			set_bit(AS_EIO, &page->mapping->flags);
59 			f2fs_stop_checkpoint(sbi);
60 		}
61 		end_page_writeback(page);
62 		dec_page_count(sbi, F2FS_WRITEBACK);
63 	}
64 
65 	if (!get_pages(sbi, F2FS_WRITEBACK) &&
66 			!list_empty(&sbi->cp_wait.task_list))
67 		wake_up(&sbi->cp_wait);
68 
69 	bio_put(bio);
70 }
71 
72 /*
73  * Low-level block read/write IO operations.
74  */
75 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
76 				int npages, bool is_read)
77 {
78 	struct bio *bio;
79 
80 	/* No failure on bio allocation */
81 	bio = bio_alloc(GFP_NOIO, npages);
82 
83 	bio->bi_bdev = sbi->sb->s_bdev;
84 	bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
85 	bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
86 	bio->bi_private = sbi;
87 
88 	return bio;
89 }
90 
91 static void __submit_merged_bio(struct f2fs_bio_info *io)
92 {
93 	struct f2fs_io_info *fio = &io->fio;
94 
95 	if (!io->bio)
96 		return;
97 
98 	if (is_read_io(fio->rw))
99 		trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
100 	else
101 		trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
102 
103 	submit_bio(fio->rw, io->bio);
104 	io->bio = NULL;
105 }
106 
107 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
108 				enum page_type type, int rw)
109 {
110 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
111 	struct f2fs_bio_info *io;
112 
113 	io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
114 
115 	down_write(&io->io_rwsem);
116 
117 	/* change META to META_FLUSH in the checkpoint procedure */
118 	if (type >= META_FLUSH) {
119 		io->fio.type = META_FLUSH;
120 		if (test_opt(sbi, NOBARRIER))
121 			io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
122 		else
123 			io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
124 	}
125 	__submit_merged_bio(io);
126 	up_write(&io->io_rwsem);
127 }
128 
129 /*
130  * Fill the locked page with data located in the block address.
131  * Return unlocked page.
132  */
133 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
134 					struct f2fs_io_info *fio)
135 {
136 	struct bio *bio;
137 
138 	trace_f2fs_submit_page_bio(page, fio);
139 	f2fs_trace_ios(page, fio, 0);
140 
141 	/* Allocate a new bio */
142 	bio = __bio_alloc(sbi, fio->blk_addr, 1, is_read_io(fio->rw));
143 
144 	if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
145 		bio_put(bio);
146 		f2fs_put_page(page, 1);
147 		return -EFAULT;
148 	}
149 
150 	submit_bio(fio->rw, bio);
151 	return 0;
152 }
153 
154 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
155 					struct f2fs_io_info *fio)
156 {
157 	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
158 	struct f2fs_bio_info *io;
159 	bool is_read = is_read_io(fio->rw);
160 
161 	io = is_read ? &sbi->read_io : &sbi->write_io[btype];
162 
163 	verify_block_addr(sbi, fio->blk_addr);
164 
165 	down_write(&io->io_rwsem);
166 
167 	if (!is_read)
168 		inc_page_count(sbi, F2FS_WRITEBACK);
169 
170 	if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
171 						io->fio.rw != fio->rw))
172 		__submit_merged_bio(io);
173 alloc_new:
174 	if (io->bio == NULL) {
175 		int bio_blocks = MAX_BIO_BLOCKS(sbi);
176 
177 		io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
178 		io->fio = *fio;
179 	}
180 
181 	if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
182 							PAGE_CACHE_SIZE) {
183 		__submit_merged_bio(io);
184 		goto alloc_new;
185 	}
186 
187 	io->last_block_in_bio = fio->blk_addr;
188 	f2fs_trace_ios(page, fio, 0);
189 
190 	up_write(&io->io_rwsem);
191 	trace_f2fs_submit_page_mbio(page, fio);
192 }
193 
194 /*
195  * Lock ordering for the change of data block address:
196  * ->data_page
197  *  ->node_page
198  *    update block addresses in the node page
199  */
200 static void __set_data_blkaddr(struct dnode_of_data *dn)
201 {
202 	struct f2fs_node *rn;
203 	__le32 *addr_array;
204 	struct page *node_page = dn->node_page;
205 	unsigned int ofs_in_node = dn->ofs_in_node;
206 
207 	f2fs_wait_on_page_writeback(node_page, NODE);
208 
209 	rn = F2FS_NODE(node_page);
210 
211 	/* Get physical address of data block */
212 	addr_array = blkaddr_in_node(rn);
213 	addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
214 	set_page_dirty(node_page);
215 }
216 
217 int reserve_new_block(struct dnode_of_data *dn)
218 {
219 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
220 
221 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
222 		return -EPERM;
223 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
224 		return -ENOSPC;
225 
226 	trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
227 
228 	dn->data_blkaddr = NEW_ADDR;
229 	__set_data_blkaddr(dn);
230 	mark_inode_dirty(dn->inode);
231 	sync_inode_page(dn);
232 	return 0;
233 }
234 
235 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
236 {
237 	bool need_put = dn->inode_page ? false : true;
238 	int err;
239 
240 	err = get_dnode_of_data(dn, index, ALLOC_NODE);
241 	if (err)
242 		return err;
243 
244 	if (dn->data_blkaddr == NULL_ADDR)
245 		err = reserve_new_block(dn);
246 	if (err || need_put)
247 		f2fs_put_dnode(dn);
248 	return err;
249 }
250 
251 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
252 					struct buffer_head *bh_result)
253 {
254 	struct f2fs_inode_info *fi = F2FS_I(inode);
255 	pgoff_t start_fofs, end_fofs;
256 	block_t start_blkaddr;
257 
258 	if (is_inode_flag_set(fi, FI_NO_EXTENT))
259 		return 0;
260 
261 	read_lock(&fi->ext.ext_lock);
262 	if (fi->ext.len == 0) {
263 		read_unlock(&fi->ext.ext_lock);
264 		return 0;
265 	}
266 
267 	stat_inc_total_hit(inode->i_sb);
268 
269 	start_fofs = fi->ext.fofs;
270 	end_fofs = fi->ext.fofs + fi->ext.len - 1;
271 	start_blkaddr = fi->ext.blk_addr;
272 
273 	if (pgofs >= start_fofs && pgofs <= end_fofs) {
274 		unsigned int blkbits = inode->i_sb->s_blocksize_bits;
275 		size_t count;
276 
277 		set_buffer_new(bh_result);
278 		map_bh(bh_result, inode->i_sb,
279 				start_blkaddr + pgofs - start_fofs);
280 		count = end_fofs - pgofs + 1;
281 		if (count < (UINT_MAX >> blkbits))
282 			bh_result->b_size = (count << blkbits);
283 		else
284 			bh_result->b_size = UINT_MAX;
285 
286 		stat_inc_read_hit(inode->i_sb);
287 		read_unlock(&fi->ext.ext_lock);
288 		return 1;
289 	}
290 	read_unlock(&fi->ext.ext_lock);
291 	return 0;
292 }
293 
294 void update_extent_cache(struct dnode_of_data *dn)
295 {
296 	struct f2fs_inode_info *fi = F2FS_I(dn->inode);
297 	pgoff_t fofs, start_fofs, end_fofs;
298 	block_t start_blkaddr, end_blkaddr;
299 	int need_update = true;
300 
301 	f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR);
302 
303 	/* Update the page address in the parent node */
304 	__set_data_blkaddr(dn);
305 
306 	if (is_inode_flag_set(fi, FI_NO_EXTENT))
307 		return;
308 
309 	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
310 							dn->ofs_in_node;
311 
312 	write_lock(&fi->ext.ext_lock);
313 
314 	start_fofs = fi->ext.fofs;
315 	end_fofs = fi->ext.fofs + fi->ext.len - 1;
316 	start_blkaddr = fi->ext.blk_addr;
317 	end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
318 
319 	/* Drop and initialize the matched extent */
320 	if (fi->ext.len == 1 && fofs == start_fofs)
321 		fi->ext.len = 0;
322 
323 	/* Initial extent */
324 	if (fi->ext.len == 0) {
325 		if (dn->data_blkaddr != NULL_ADDR) {
326 			fi->ext.fofs = fofs;
327 			fi->ext.blk_addr = dn->data_blkaddr;
328 			fi->ext.len = 1;
329 		}
330 		goto end_update;
331 	}
332 
333 	/* Front merge */
334 	if (fofs == start_fofs - 1 && dn->data_blkaddr == start_blkaddr - 1) {
335 		fi->ext.fofs--;
336 		fi->ext.blk_addr--;
337 		fi->ext.len++;
338 		goto end_update;
339 	}
340 
341 	/* Back merge */
342 	if (fofs == end_fofs + 1 && dn->data_blkaddr == end_blkaddr + 1) {
343 		fi->ext.len++;
344 		goto end_update;
345 	}
346 
347 	/* Split the existing extent */
348 	if (fi->ext.len > 1 &&
349 		fofs >= start_fofs && fofs <= end_fofs) {
350 		if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
351 			fi->ext.len = fofs - start_fofs;
352 		} else {
353 			fi->ext.fofs = fofs + 1;
354 			fi->ext.blk_addr = start_blkaddr +
355 					fofs - start_fofs + 1;
356 			fi->ext.len -= fofs - start_fofs + 1;
357 		}
358 	} else {
359 		need_update = false;
360 	}
361 
362 	/* Finally, if the extent is very fragmented, let's drop the cache. */
363 	if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
364 		fi->ext.len = 0;
365 		set_inode_flag(fi, FI_NO_EXTENT);
366 		need_update = true;
367 	}
368 end_update:
369 	write_unlock(&fi->ext.ext_lock);
370 	if (need_update)
371 		sync_inode_page(dn);
372 	return;
373 }
374 
375 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
376 {
377 	struct address_space *mapping = inode->i_mapping;
378 	struct dnode_of_data dn;
379 	struct page *page;
380 	int err;
381 	struct f2fs_io_info fio = {
382 		.type = DATA,
383 		.rw = sync ? READ_SYNC : READA,
384 	};
385 
386 	page = find_get_page(mapping, index);
387 	if (page && PageUptodate(page))
388 		return page;
389 	f2fs_put_page(page, 0);
390 
391 	set_new_dnode(&dn, inode, NULL, NULL, 0);
392 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
393 	if (err)
394 		return ERR_PTR(err);
395 	f2fs_put_dnode(&dn);
396 
397 	if (dn.data_blkaddr == NULL_ADDR)
398 		return ERR_PTR(-ENOENT);
399 
400 	/* By fallocate(), there is no cached page, but with NEW_ADDR */
401 	if (unlikely(dn.data_blkaddr == NEW_ADDR))
402 		return ERR_PTR(-EINVAL);
403 
404 	page = grab_cache_page(mapping, index);
405 	if (!page)
406 		return ERR_PTR(-ENOMEM);
407 
408 	if (PageUptodate(page)) {
409 		unlock_page(page);
410 		return page;
411 	}
412 
413 	fio.blk_addr = dn.data_blkaddr;
414 	err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
415 	if (err)
416 		return ERR_PTR(err);
417 
418 	if (sync) {
419 		wait_on_page_locked(page);
420 		if (unlikely(!PageUptodate(page))) {
421 			f2fs_put_page(page, 0);
422 			return ERR_PTR(-EIO);
423 		}
424 	}
425 	return page;
426 }
427 
428 /*
429  * If it tries to access a hole, return an error.
430  * Because, the callers, functions in dir.c and GC, should be able to know
431  * whether this page exists or not.
432  */
433 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
434 {
435 	struct address_space *mapping = inode->i_mapping;
436 	struct dnode_of_data dn;
437 	struct page *page;
438 	int err;
439 	struct f2fs_io_info fio = {
440 		.type = DATA,
441 		.rw = READ_SYNC,
442 	};
443 repeat:
444 	page = grab_cache_page(mapping, index);
445 	if (!page)
446 		return ERR_PTR(-ENOMEM);
447 
448 	set_new_dnode(&dn, inode, NULL, NULL, 0);
449 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
450 	if (err) {
451 		f2fs_put_page(page, 1);
452 		return ERR_PTR(err);
453 	}
454 	f2fs_put_dnode(&dn);
455 
456 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
457 		f2fs_put_page(page, 1);
458 		return ERR_PTR(-ENOENT);
459 	}
460 
461 	if (PageUptodate(page))
462 		return page;
463 
464 	/*
465 	 * A new dentry page is allocated but not able to be written, since its
466 	 * new inode page couldn't be allocated due to -ENOSPC.
467 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
468 	 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
469 	 */
470 	if (dn.data_blkaddr == NEW_ADDR) {
471 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
472 		SetPageUptodate(page);
473 		return page;
474 	}
475 
476 	fio.blk_addr = dn.data_blkaddr;
477 	err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
478 	if (err)
479 		return ERR_PTR(err);
480 
481 	lock_page(page);
482 	if (unlikely(!PageUptodate(page))) {
483 		f2fs_put_page(page, 1);
484 		return ERR_PTR(-EIO);
485 	}
486 	if (unlikely(page->mapping != mapping)) {
487 		f2fs_put_page(page, 1);
488 		goto repeat;
489 	}
490 	return page;
491 }
492 
493 /*
494  * Caller ensures that this data page is never allocated.
495  * A new zero-filled data page is allocated in the page cache.
496  *
497  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
498  * f2fs_unlock_op().
499  * Note that, ipage is set only by make_empty_dir.
500  */
501 struct page *get_new_data_page(struct inode *inode,
502 		struct page *ipage, pgoff_t index, bool new_i_size)
503 {
504 	struct address_space *mapping = inode->i_mapping;
505 	struct page *page;
506 	struct dnode_of_data dn;
507 	int err;
508 
509 	set_new_dnode(&dn, inode, ipage, NULL, 0);
510 	err = f2fs_reserve_block(&dn, index);
511 	if (err)
512 		return ERR_PTR(err);
513 repeat:
514 	page = grab_cache_page(mapping, index);
515 	if (!page) {
516 		err = -ENOMEM;
517 		goto put_err;
518 	}
519 
520 	if (PageUptodate(page))
521 		return page;
522 
523 	if (dn.data_blkaddr == NEW_ADDR) {
524 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
525 		SetPageUptodate(page);
526 	} else {
527 		struct f2fs_io_info fio = {
528 			.type = DATA,
529 			.rw = READ_SYNC,
530 			.blk_addr = dn.data_blkaddr,
531 		};
532 		err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
533 		if (err)
534 			goto put_err;
535 
536 		lock_page(page);
537 		if (unlikely(!PageUptodate(page))) {
538 			f2fs_put_page(page, 1);
539 			err = -EIO;
540 			goto put_err;
541 		}
542 		if (unlikely(page->mapping != mapping)) {
543 			f2fs_put_page(page, 1);
544 			goto repeat;
545 		}
546 	}
547 
548 	if (new_i_size &&
549 		i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
550 		i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
551 		/* Only the directory inode sets new_i_size */
552 		set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
553 	}
554 	return page;
555 
556 put_err:
557 	f2fs_put_dnode(&dn);
558 	return ERR_PTR(err);
559 }
560 
561 static int __allocate_data_block(struct dnode_of_data *dn)
562 {
563 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
564 	struct f2fs_inode_info *fi = F2FS_I(dn->inode);
565 	struct f2fs_summary sum;
566 	struct node_info ni;
567 	int seg = CURSEG_WARM_DATA;
568 	pgoff_t fofs;
569 
570 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
571 		return -EPERM;
572 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
573 		return -ENOSPC;
574 
575 	get_node_info(sbi, dn->nid, &ni);
576 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
577 
578 	if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
579 		seg = CURSEG_DIRECT_IO;
580 
581 	allocate_data_block(sbi, NULL, NULL_ADDR, &dn->data_blkaddr, &sum, seg);
582 
583 	/* direct IO doesn't use extent cache to maximize the performance */
584 	__set_data_blkaddr(dn);
585 
586 	/* update i_size */
587 	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
588 							dn->ofs_in_node;
589 	if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
590 		i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
591 
592 	return 0;
593 }
594 
595 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
596 							size_t count)
597 {
598 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
599 	struct dnode_of_data dn;
600 	u64 start = F2FS_BYTES_TO_BLK(offset);
601 	u64 len = F2FS_BYTES_TO_BLK(count);
602 	bool allocated;
603 	u64 end_offset;
604 
605 	while (len) {
606 		f2fs_balance_fs(sbi);
607 		f2fs_lock_op(sbi);
608 
609 		/* When reading holes, we need its node page */
610 		set_new_dnode(&dn, inode, NULL, NULL, 0);
611 		if (get_dnode_of_data(&dn, start, ALLOC_NODE))
612 			goto out;
613 
614 		allocated = false;
615 		end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
616 
617 		while (dn.ofs_in_node < end_offset && len) {
618 			if (dn.data_blkaddr == NULL_ADDR) {
619 				if (__allocate_data_block(&dn))
620 					goto sync_out;
621 				allocated = true;
622 			}
623 			len--;
624 			start++;
625 			dn.ofs_in_node++;
626 		}
627 
628 		if (allocated)
629 			sync_inode_page(&dn);
630 
631 		f2fs_put_dnode(&dn);
632 		f2fs_unlock_op(sbi);
633 	}
634 	return;
635 
636 sync_out:
637 	if (allocated)
638 		sync_inode_page(&dn);
639 	f2fs_put_dnode(&dn);
640 out:
641 	f2fs_unlock_op(sbi);
642 	return;
643 }
644 
645 /*
646  * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
647  * If original data blocks are allocated, then give them to blockdev.
648  * Otherwise,
649  *     a. preallocate requested block addresses
650  *     b. do not use extent cache for better performance
651  *     c. give the block addresses to blockdev
652  */
653 static int __get_data_block(struct inode *inode, sector_t iblock,
654 			struct buffer_head *bh_result, int create, bool fiemap)
655 {
656 	unsigned int blkbits = inode->i_sb->s_blocksize_bits;
657 	unsigned maxblocks = bh_result->b_size >> blkbits;
658 	struct dnode_of_data dn;
659 	int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
660 	pgoff_t pgofs, end_offset;
661 	int err = 0, ofs = 1;
662 	bool allocated = false;
663 
664 	/* Get the page offset from the block offset(iblock) */
665 	pgofs =	(pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
666 
667 	if (check_extent_cache(inode, pgofs, bh_result))
668 		goto out;
669 
670 	if (create)
671 		f2fs_lock_op(F2FS_I_SB(inode));
672 
673 	/* When reading holes, we need its node page */
674 	set_new_dnode(&dn, inode, NULL, NULL, 0);
675 	err = get_dnode_of_data(&dn, pgofs, mode);
676 	if (err) {
677 		if (err == -ENOENT)
678 			err = 0;
679 		goto unlock_out;
680 	}
681 	if (dn.data_blkaddr == NEW_ADDR && !fiemap)
682 		goto put_out;
683 
684 	if (dn.data_blkaddr != NULL_ADDR) {
685 		set_buffer_new(bh_result);
686 		map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
687 	} else if (create) {
688 		err = __allocate_data_block(&dn);
689 		if (err)
690 			goto put_out;
691 		allocated = true;
692 		set_buffer_new(bh_result);
693 		map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
694 	} else {
695 		goto put_out;
696 	}
697 
698 	end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
699 	bh_result->b_size = (((size_t)1) << blkbits);
700 	dn.ofs_in_node++;
701 	pgofs++;
702 
703 get_next:
704 	if (dn.ofs_in_node >= end_offset) {
705 		if (allocated)
706 			sync_inode_page(&dn);
707 		allocated = false;
708 		f2fs_put_dnode(&dn);
709 
710 		set_new_dnode(&dn, inode, NULL, NULL, 0);
711 		err = get_dnode_of_data(&dn, pgofs, mode);
712 		if (err) {
713 			if (err == -ENOENT)
714 				err = 0;
715 			goto unlock_out;
716 		}
717 		if (dn.data_blkaddr == NEW_ADDR && !fiemap)
718 			goto put_out;
719 
720 		end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
721 	}
722 
723 	if (maxblocks > (bh_result->b_size >> blkbits)) {
724 		block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
725 		if (blkaddr == NULL_ADDR && create) {
726 			err = __allocate_data_block(&dn);
727 			if (err)
728 				goto sync_out;
729 			allocated = true;
730 			blkaddr = dn.data_blkaddr;
731 		}
732 		/* Give more consecutive addresses for the readahead */
733 		if (blkaddr == (bh_result->b_blocknr + ofs)) {
734 			ofs++;
735 			dn.ofs_in_node++;
736 			pgofs++;
737 			bh_result->b_size += (((size_t)1) << blkbits);
738 			goto get_next;
739 		}
740 	}
741 sync_out:
742 	if (allocated)
743 		sync_inode_page(&dn);
744 put_out:
745 	f2fs_put_dnode(&dn);
746 unlock_out:
747 	if (create)
748 		f2fs_unlock_op(F2FS_I_SB(inode));
749 out:
750 	trace_f2fs_get_data_block(inode, iblock, bh_result, err);
751 	return err;
752 }
753 
754 static int get_data_block(struct inode *inode, sector_t iblock,
755 			struct buffer_head *bh_result, int create)
756 {
757 	return __get_data_block(inode, iblock, bh_result, create, false);
758 }
759 
760 static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
761 			struct buffer_head *bh_result, int create)
762 {
763 	return __get_data_block(inode, iblock, bh_result, create, true);
764 }
765 
766 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
767 		u64 start, u64 len)
768 {
769 	return generic_block_fiemap(inode, fieinfo,
770 				start, len, get_data_block_fiemap);
771 }
772 
773 static int f2fs_read_data_page(struct file *file, struct page *page)
774 {
775 	struct inode *inode = page->mapping->host;
776 	int ret = -EAGAIN;
777 
778 	trace_f2fs_readpage(page, DATA);
779 
780 	/* If the file has inline data, try to read it directly */
781 	if (f2fs_has_inline_data(inode))
782 		ret = f2fs_read_inline_data(inode, page);
783 	if (ret == -EAGAIN)
784 		ret = mpage_readpage(page, get_data_block);
785 
786 	return ret;
787 }
788 
789 static int f2fs_read_data_pages(struct file *file,
790 			struct address_space *mapping,
791 			struct list_head *pages, unsigned nr_pages)
792 {
793 	struct inode *inode = file->f_mapping->host;
794 
795 	/* If the file has inline data, skip readpages */
796 	if (f2fs_has_inline_data(inode))
797 		return 0;
798 
799 	return mpage_readpages(mapping, pages, nr_pages, get_data_block);
800 }
801 
802 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
803 {
804 	struct inode *inode = page->mapping->host;
805 	struct dnode_of_data dn;
806 	int err = 0;
807 
808 	set_new_dnode(&dn, inode, NULL, NULL, 0);
809 	err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
810 	if (err)
811 		return err;
812 
813 	fio->blk_addr = dn.data_blkaddr;
814 
815 	/* This page is already truncated */
816 	if (fio->blk_addr == NULL_ADDR)
817 		goto out_writepage;
818 
819 	set_page_writeback(page);
820 
821 	/*
822 	 * If current allocation needs SSR,
823 	 * it had better in-place writes for updated data.
824 	 */
825 	if (unlikely(fio->blk_addr != NEW_ADDR &&
826 			!is_cold_data(page) &&
827 			need_inplace_update(inode))) {
828 		rewrite_data_page(page, fio);
829 		set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
830 	} else {
831 		write_data_page(page, &dn, fio);
832 		update_extent_cache(&dn);
833 		set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
834 	}
835 out_writepage:
836 	f2fs_put_dnode(&dn);
837 	return err;
838 }
839 
840 static int f2fs_write_data_page(struct page *page,
841 					struct writeback_control *wbc)
842 {
843 	struct inode *inode = page->mapping->host;
844 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
845 	loff_t i_size = i_size_read(inode);
846 	const pgoff_t end_index = ((unsigned long long) i_size)
847 							>> PAGE_CACHE_SHIFT;
848 	unsigned offset = 0;
849 	bool need_balance_fs = false;
850 	int err = 0;
851 	struct f2fs_io_info fio = {
852 		.type = DATA,
853 		.rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
854 	};
855 
856 	trace_f2fs_writepage(page, DATA);
857 
858 	if (page->index < end_index)
859 		goto write;
860 
861 	/*
862 	 * If the offset is out-of-range of file size,
863 	 * this page does not have to be written to disk.
864 	 */
865 	offset = i_size & (PAGE_CACHE_SIZE - 1);
866 	if ((page->index >= end_index + 1) || !offset)
867 		goto out;
868 
869 	zero_user_segment(page, offset, PAGE_CACHE_SIZE);
870 write:
871 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
872 		goto redirty_out;
873 	if (f2fs_is_drop_cache(inode))
874 		goto out;
875 	if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
876 			available_free_memory(sbi, BASE_CHECK))
877 		goto redirty_out;
878 
879 	/* Dentry blocks are controlled by checkpoint */
880 	if (S_ISDIR(inode->i_mode)) {
881 		if (unlikely(f2fs_cp_error(sbi)))
882 			goto redirty_out;
883 		err = do_write_data_page(page, &fio);
884 		goto done;
885 	}
886 
887 	/* we should bypass data pages to proceed the kworkder jobs */
888 	if (unlikely(f2fs_cp_error(sbi))) {
889 		SetPageError(page);
890 		goto out;
891 	}
892 
893 	if (!wbc->for_reclaim)
894 		need_balance_fs = true;
895 	else if (has_not_enough_free_secs(sbi, 0))
896 		goto redirty_out;
897 
898 	err = -EAGAIN;
899 	f2fs_lock_op(sbi);
900 	if (f2fs_has_inline_data(inode))
901 		err = f2fs_write_inline_data(inode, page);
902 	if (err == -EAGAIN)
903 		err = do_write_data_page(page, &fio);
904 	f2fs_unlock_op(sbi);
905 done:
906 	if (err && err != -ENOENT)
907 		goto redirty_out;
908 
909 	clear_cold_data(page);
910 out:
911 	inode_dec_dirty_pages(inode);
912 	unlock_page(page);
913 	if (need_balance_fs)
914 		f2fs_balance_fs(sbi);
915 	if (wbc->for_reclaim)
916 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
917 	return 0;
918 
919 redirty_out:
920 	redirty_page_for_writepage(wbc, page);
921 	return AOP_WRITEPAGE_ACTIVATE;
922 }
923 
924 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
925 			void *data)
926 {
927 	struct address_space *mapping = data;
928 	int ret = mapping->a_ops->writepage(page, wbc);
929 	mapping_set_error(mapping, ret);
930 	return ret;
931 }
932 
933 static int f2fs_write_data_pages(struct address_space *mapping,
934 			    struct writeback_control *wbc)
935 {
936 	struct inode *inode = mapping->host;
937 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
938 	bool locked = false;
939 	int ret;
940 	long diff;
941 
942 	trace_f2fs_writepages(mapping->host, wbc, DATA);
943 
944 	/* deal with chardevs and other special file */
945 	if (!mapping->a_ops->writepage)
946 		return 0;
947 
948 	if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
949 			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
950 			available_free_memory(sbi, DIRTY_DENTS))
951 		goto skip_write;
952 
953 	diff = nr_pages_to_write(sbi, DATA, wbc);
954 
955 	if (!S_ISDIR(inode->i_mode)) {
956 		mutex_lock(&sbi->writepages);
957 		locked = true;
958 	}
959 	ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
960 	if (locked)
961 		mutex_unlock(&sbi->writepages);
962 
963 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
964 
965 	remove_dirty_dir_inode(inode);
966 
967 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
968 	return ret;
969 
970 skip_write:
971 	wbc->pages_skipped += get_dirty_pages(inode);
972 	return 0;
973 }
974 
975 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
976 {
977 	struct inode *inode = mapping->host;
978 
979 	if (to > inode->i_size) {
980 		truncate_pagecache(inode, inode->i_size);
981 		truncate_blocks(inode, inode->i_size, true);
982 	}
983 }
984 
985 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
986 		loff_t pos, unsigned len, unsigned flags,
987 		struct page **pagep, void **fsdata)
988 {
989 	struct inode *inode = mapping->host;
990 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
991 	struct page *page, *ipage;
992 	pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
993 	struct dnode_of_data dn;
994 	int err = 0;
995 
996 	trace_f2fs_write_begin(inode, pos, len, flags);
997 
998 	f2fs_balance_fs(sbi);
999 
1000 	/*
1001 	 * We should check this at this moment to avoid deadlock on inode page
1002 	 * and #0 page. The locking rule for inline_data conversion should be:
1003 	 * lock_page(page #0) -> lock_page(inode_page)
1004 	 */
1005 	if (index != 0) {
1006 		err = f2fs_convert_inline_inode(inode);
1007 		if (err)
1008 			goto fail;
1009 	}
1010 repeat:
1011 	page = grab_cache_page_write_begin(mapping, index, flags);
1012 	if (!page) {
1013 		err = -ENOMEM;
1014 		goto fail;
1015 	}
1016 
1017 	*pagep = page;
1018 
1019 	f2fs_lock_op(sbi);
1020 
1021 	/* check inline_data */
1022 	ipage = get_node_page(sbi, inode->i_ino);
1023 	if (IS_ERR(ipage)) {
1024 		err = PTR_ERR(ipage);
1025 		goto unlock_fail;
1026 	}
1027 
1028 	set_new_dnode(&dn, inode, ipage, ipage, 0);
1029 
1030 	if (f2fs_has_inline_data(inode)) {
1031 		if (pos + len <= MAX_INLINE_DATA) {
1032 			read_inline_data(page, ipage);
1033 			set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1034 			sync_inode_page(&dn);
1035 			goto put_next;
1036 		}
1037 		err = f2fs_convert_inline_page(&dn, page);
1038 		if (err)
1039 			goto put_fail;
1040 	}
1041 	err = f2fs_reserve_block(&dn, index);
1042 	if (err)
1043 		goto put_fail;
1044 put_next:
1045 	f2fs_put_dnode(&dn);
1046 	f2fs_unlock_op(sbi);
1047 
1048 	if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
1049 		return 0;
1050 
1051 	f2fs_wait_on_page_writeback(page, DATA);
1052 
1053 	if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1054 		unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1055 		unsigned end = start + len;
1056 
1057 		/* Reading beyond i_size is simple: memset to zero */
1058 		zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1059 		goto out;
1060 	}
1061 
1062 	if (dn.data_blkaddr == NEW_ADDR) {
1063 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1064 	} else {
1065 		struct f2fs_io_info fio = {
1066 			.type = DATA,
1067 			.rw = READ_SYNC,
1068 			.blk_addr = dn.data_blkaddr,
1069 		};
1070 		err = f2fs_submit_page_bio(sbi, page, &fio);
1071 		if (err)
1072 			goto fail;
1073 
1074 		lock_page(page);
1075 		if (unlikely(!PageUptodate(page))) {
1076 			f2fs_put_page(page, 1);
1077 			err = -EIO;
1078 			goto fail;
1079 		}
1080 		if (unlikely(page->mapping != mapping)) {
1081 			f2fs_put_page(page, 1);
1082 			goto repeat;
1083 		}
1084 	}
1085 out:
1086 	SetPageUptodate(page);
1087 	clear_cold_data(page);
1088 	return 0;
1089 
1090 put_fail:
1091 	f2fs_put_dnode(&dn);
1092 unlock_fail:
1093 	f2fs_unlock_op(sbi);
1094 	f2fs_put_page(page, 1);
1095 fail:
1096 	f2fs_write_failed(mapping, pos + len);
1097 	return err;
1098 }
1099 
1100 static int f2fs_write_end(struct file *file,
1101 			struct address_space *mapping,
1102 			loff_t pos, unsigned len, unsigned copied,
1103 			struct page *page, void *fsdata)
1104 {
1105 	struct inode *inode = page->mapping->host;
1106 
1107 	trace_f2fs_write_end(inode, pos, len, copied);
1108 
1109 	set_page_dirty(page);
1110 
1111 	if (pos + copied > i_size_read(inode)) {
1112 		i_size_write(inode, pos + copied);
1113 		mark_inode_dirty(inode);
1114 		update_inode_page(inode);
1115 	}
1116 
1117 	f2fs_put_page(page, 1);
1118 	return copied;
1119 }
1120 
1121 static int check_direct_IO(struct inode *inode, int rw,
1122 		struct iov_iter *iter, loff_t offset)
1123 {
1124 	unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1125 
1126 	if (rw == READ)
1127 		return 0;
1128 
1129 	if (offset & blocksize_mask)
1130 		return -EINVAL;
1131 
1132 	if (iov_iter_alignment(iter) & blocksize_mask)
1133 		return -EINVAL;
1134 
1135 	return 0;
1136 }
1137 
1138 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1139 		struct iov_iter *iter, loff_t offset)
1140 {
1141 	struct file *file = iocb->ki_filp;
1142 	struct address_space *mapping = file->f_mapping;
1143 	struct inode *inode = mapping->host;
1144 	size_t count = iov_iter_count(iter);
1145 	int err;
1146 
1147 	/* we don't need to use inline_data strictly */
1148 	if (f2fs_has_inline_data(inode)) {
1149 		err = f2fs_convert_inline_inode(inode);
1150 		if (err)
1151 			return err;
1152 	}
1153 
1154 	if (check_direct_IO(inode, rw, iter, offset))
1155 		return 0;
1156 
1157 	trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1158 
1159 	if (rw & WRITE)
1160 		__allocate_data_blocks(inode, offset, count);
1161 
1162 	err = blockdev_direct_IO(rw, iocb, inode, iter, offset, get_data_block);
1163 	if (err < 0 && (rw & WRITE))
1164 		f2fs_write_failed(mapping, offset + count);
1165 
1166 	trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1167 
1168 	return err;
1169 }
1170 
1171 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1172 							unsigned int length)
1173 {
1174 	struct inode *inode = page->mapping->host;
1175 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1176 
1177 	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1178 		(offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1179 		return;
1180 
1181 	if (PageDirty(page)) {
1182 		if (inode->i_ino == F2FS_META_INO(sbi))
1183 			dec_page_count(sbi, F2FS_DIRTY_META);
1184 		else if (inode->i_ino == F2FS_NODE_INO(sbi))
1185 			dec_page_count(sbi, F2FS_DIRTY_NODES);
1186 		else
1187 			inode_dec_dirty_pages(inode);
1188 	}
1189 	ClearPagePrivate(page);
1190 }
1191 
1192 int f2fs_release_page(struct page *page, gfp_t wait)
1193 {
1194 	/* If this is dirty page, keep PagePrivate */
1195 	if (PageDirty(page))
1196 		return 0;
1197 
1198 	ClearPagePrivate(page);
1199 	return 1;
1200 }
1201 
1202 static int f2fs_set_data_page_dirty(struct page *page)
1203 {
1204 	struct address_space *mapping = page->mapping;
1205 	struct inode *inode = mapping->host;
1206 
1207 	trace_f2fs_set_page_dirty(page, DATA);
1208 
1209 	SetPageUptodate(page);
1210 
1211 	if (f2fs_is_atomic_file(inode)) {
1212 		register_inmem_page(inode, page);
1213 		return 1;
1214 	}
1215 
1216 	mark_inode_dirty(inode);
1217 
1218 	if (!PageDirty(page)) {
1219 		__set_page_dirty_nobuffers(page);
1220 		update_dirty_page(inode, page);
1221 		return 1;
1222 	}
1223 	return 0;
1224 }
1225 
1226 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1227 {
1228 	struct inode *inode = mapping->host;
1229 
1230 	/* we don't need to use inline_data strictly */
1231 	if (f2fs_has_inline_data(inode)) {
1232 		int err = f2fs_convert_inline_inode(inode);
1233 		if (err)
1234 			return err;
1235 	}
1236 	return generic_block_bmap(mapping, block, get_data_block);
1237 }
1238 
1239 const struct address_space_operations f2fs_dblock_aops = {
1240 	.readpage	= f2fs_read_data_page,
1241 	.readpages	= f2fs_read_data_pages,
1242 	.writepage	= f2fs_write_data_page,
1243 	.writepages	= f2fs_write_data_pages,
1244 	.write_begin	= f2fs_write_begin,
1245 	.write_end	= f2fs_write_end,
1246 	.set_page_dirty	= f2fs_set_data_page_dirty,
1247 	.invalidatepage	= f2fs_invalidate_page,
1248 	.releasepage	= f2fs_release_page,
1249 	.direct_IO	= f2fs_direct_IO,
1250 	.bmap		= f2fs_bmap,
1251 };
1252