xref: /linux/fs/f2fs/data.c (revision c0c914eca7f251c70facc37dfebeaf176601918d)
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/cleancache.h>
23 
24 #include "f2fs.h"
25 #include "node.h"
26 #include "segment.h"
27 #include "trace.h"
28 #include <trace/events/f2fs.h>
29 
30 static void f2fs_read_end_io(struct bio *bio)
31 {
32 	struct bio_vec *bvec;
33 	int i;
34 
35 	if (f2fs_bio_encrypted(bio)) {
36 		if (bio->bi_error) {
37 			f2fs_release_crypto_ctx(bio->bi_private);
38 		} else {
39 			f2fs_end_io_crypto_work(bio->bi_private, bio);
40 			return;
41 		}
42 	}
43 
44 	bio_for_each_segment_all(bvec, bio, i) {
45 		struct page *page = bvec->bv_page;
46 
47 		if (!bio->bi_error) {
48 			SetPageUptodate(page);
49 		} else {
50 			ClearPageUptodate(page);
51 			SetPageError(page);
52 		}
53 		unlock_page(page);
54 	}
55 	bio_put(bio);
56 }
57 
58 static void f2fs_write_end_io(struct bio *bio)
59 {
60 	struct f2fs_sb_info *sbi = bio->bi_private;
61 	struct bio_vec *bvec;
62 	int i;
63 
64 	bio_for_each_segment_all(bvec, bio, i) {
65 		struct page *page = bvec->bv_page;
66 
67 		f2fs_restore_and_release_control_page(&page);
68 
69 		if (unlikely(bio->bi_error)) {
70 			set_page_dirty(page);
71 			set_bit(AS_EIO, &page->mapping->flags);
72 			f2fs_stop_checkpoint(sbi);
73 		}
74 		end_page_writeback(page);
75 		dec_page_count(sbi, F2FS_WRITEBACK);
76 	}
77 
78 	if (!get_pages(sbi, F2FS_WRITEBACK) &&
79 			!list_empty(&sbi->cp_wait.task_list))
80 		wake_up(&sbi->cp_wait);
81 
82 	bio_put(bio);
83 }
84 
85 /*
86  * Low-level block read/write IO operations.
87  */
88 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
89 				int npages, bool is_read)
90 {
91 	struct bio *bio;
92 
93 	bio = f2fs_bio_alloc(npages);
94 
95 	bio->bi_bdev = sbi->sb->s_bdev;
96 	bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
97 	bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
98 	bio->bi_private = is_read ? NULL : sbi;
99 
100 	return bio;
101 }
102 
103 static void __submit_merged_bio(struct f2fs_bio_info *io)
104 {
105 	struct f2fs_io_info *fio = &io->fio;
106 
107 	if (!io->bio)
108 		return;
109 
110 	if (is_read_io(fio->rw))
111 		trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
112 	else
113 		trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
114 
115 	submit_bio(fio->rw, io->bio);
116 	io->bio = NULL;
117 }
118 
119 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
120 				enum page_type type, int rw)
121 {
122 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
123 	struct f2fs_bio_info *io;
124 
125 	io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
126 
127 	down_write(&io->io_rwsem);
128 
129 	/* change META to META_FLUSH in the checkpoint procedure */
130 	if (type >= META_FLUSH) {
131 		io->fio.type = META_FLUSH;
132 		if (test_opt(sbi, NOBARRIER))
133 			io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
134 		else
135 			io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
136 	}
137 	__submit_merged_bio(io);
138 	up_write(&io->io_rwsem);
139 }
140 
141 /*
142  * Fill the locked page with data located in the block address.
143  * Return unlocked page.
144  */
145 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
146 {
147 	struct bio *bio;
148 	struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
149 
150 	trace_f2fs_submit_page_bio(page, fio);
151 	f2fs_trace_ios(fio, 0);
152 
153 	/* Allocate a new bio */
154 	bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
155 
156 	if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
157 		bio_put(bio);
158 		return -EFAULT;
159 	}
160 
161 	submit_bio(fio->rw, bio);
162 	return 0;
163 }
164 
165 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
166 {
167 	struct f2fs_sb_info *sbi = fio->sbi;
168 	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
169 	struct f2fs_bio_info *io;
170 	bool is_read = is_read_io(fio->rw);
171 	struct page *bio_page;
172 
173 	io = is_read ? &sbi->read_io : &sbi->write_io[btype];
174 
175 	verify_block_addr(sbi, fio->blk_addr);
176 
177 	down_write(&io->io_rwsem);
178 
179 	if (!is_read)
180 		inc_page_count(sbi, F2FS_WRITEBACK);
181 
182 	if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
183 						io->fio.rw != fio->rw))
184 		__submit_merged_bio(io);
185 alloc_new:
186 	if (io->bio == NULL) {
187 		int bio_blocks = MAX_BIO_BLOCKS(sbi);
188 
189 		io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
190 		io->fio = *fio;
191 	}
192 
193 	bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
194 
195 	if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
196 							PAGE_CACHE_SIZE) {
197 		__submit_merged_bio(io);
198 		goto alloc_new;
199 	}
200 
201 	io->last_block_in_bio = fio->blk_addr;
202 	f2fs_trace_ios(fio, 0);
203 
204 	up_write(&io->io_rwsem);
205 	trace_f2fs_submit_page_mbio(fio->page, fio);
206 }
207 
208 /*
209  * Lock ordering for the change of data block address:
210  * ->data_page
211  *  ->node_page
212  *    update block addresses in the node page
213  */
214 void set_data_blkaddr(struct dnode_of_data *dn)
215 {
216 	struct f2fs_node *rn;
217 	__le32 *addr_array;
218 	struct page *node_page = dn->node_page;
219 	unsigned int ofs_in_node = dn->ofs_in_node;
220 
221 	f2fs_wait_on_page_writeback(node_page, NODE);
222 
223 	rn = F2FS_NODE(node_page);
224 
225 	/* Get physical address of data block */
226 	addr_array = blkaddr_in_node(rn);
227 	addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
228 	if (set_page_dirty(node_page))
229 		dn->node_changed = true;
230 }
231 
232 int reserve_new_block(struct dnode_of_data *dn)
233 {
234 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
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 	dn->data_blkaddr = NEW_ADDR;
244 	set_data_blkaddr(dn);
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 	err = get_dnode_of_data(dn, index, ALLOC_NODE);
256 	if (err)
257 		return err;
258 
259 	if (dn->data_blkaddr == NULL_ADDR)
260 		err = reserve_new_block(dn);
261 	if (err || need_put)
262 		f2fs_put_dnode(dn);
263 	return err;
264 }
265 
266 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
267 {
268 	struct extent_info ei;
269 	struct inode *inode = dn->inode;
270 
271 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
272 		dn->data_blkaddr = ei.blk + index - ei.fofs;
273 		return 0;
274 	}
275 
276 	return f2fs_reserve_block(dn, index);
277 }
278 
279 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
280 						int rw, bool for_write)
281 {
282 	struct address_space *mapping = inode->i_mapping;
283 	struct dnode_of_data dn;
284 	struct page *page;
285 	struct extent_info ei;
286 	int err;
287 	struct f2fs_io_info fio = {
288 		.sbi = F2FS_I_SB(inode),
289 		.type = DATA,
290 		.rw = rw,
291 		.encrypted_page = NULL,
292 	};
293 
294 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
295 		return read_mapping_page(mapping, index, NULL);
296 
297 	page = f2fs_grab_cache_page(mapping, index, for_write);
298 	if (!page)
299 		return ERR_PTR(-ENOMEM);
300 
301 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
302 		dn.data_blkaddr = ei.blk + index - ei.fofs;
303 		goto got_it;
304 	}
305 
306 	set_new_dnode(&dn, inode, NULL, NULL, 0);
307 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
308 	if (err)
309 		goto put_err;
310 	f2fs_put_dnode(&dn);
311 
312 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
313 		err = -ENOENT;
314 		goto put_err;
315 	}
316 got_it:
317 	if (PageUptodate(page)) {
318 		unlock_page(page);
319 		return page;
320 	}
321 
322 	/*
323 	 * A new dentry page is allocated but not able to be written, since its
324 	 * new inode page couldn't be allocated due to -ENOSPC.
325 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
326 	 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
327 	 */
328 	if (dn.data_blkaddr == NEW_ADDR) {
329 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
330 		SetPageUptodate(page);
331 		unlock_page(page);
332 		return page;
333 	}
334 
335 	fio.blk_addr = dn.data_blkaddr;
336 	fio.page = page;
337 	err = f2fs_submit_page_bio(&fio);
338 	if (err)
339 		goto put_err;
340 	return page;
341 
342 put_err:
343 	f2fs_put_page(page, 1);
344 	return ERR_PTR(err);
345 }
346 
347 struct page *find_data_page(struct inode *inode, pgoff_t index)
348 {
349 	struct address_space *mapping = inode->i_mapping;
350 	struct page *page;
351 
352 	page = find_get_page(mapping, index);
353 	if (page && PageUptodate(page))
354 		return page;
355 	f2fs_put_page(page, 0);
356 
357 	page = get_read_data_page(inode, index, READ_SYNC, false);
358 	if (IS_ERR(page))
359 		return page;
360 
361 	if (PageUptodate(page))
362 		return page;
363 
364 	wait_on_page_locked(page);
365 	if (unlikely(!PageUptodate(page))) {
366 		f2fs_put_page(page, 0);
367 		return ERR_PTR(-EIO);
368 	}
369 	return page;
370 }
371 
372 /*
373  * If it tries to access a hole, return an error.
374  * Because, the callers, functions in dir.c and GC, should be able to know
375  * whether this page exists or not.
376  */
377 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
378 							bool for_write)
379 {
380 	struct address_space *mapping = inode->i_mapping;
381 	struct page *page;
382 repeat:
383 	page = get_read_data_page(inode, index, READ_SYNC, for_write);
384 	if (IS_ERR(page))
385 		return page;
386 
387 	/* wait for read completion */
388 	lock_page(page);
389 	if (unlikely(!PageUptodate(page))) {
390 		f2fs_put_page(page, 1);
391 		return ERR_PTR(-EIO);
392 	}
393 	if (unlikely(page->mapping != mapping)) {
394 		f2fs_put_page(page, 1);
395 		goto repeat;
396 	}
397 	return page;
398 }
399 
400 /*
401  * Caller ensures that this data page is never allocated.
402  * A new zero-filled data page is allocated in the page cache.
403  *
404  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
405  * f2fs_unlock_op().
406  * Note that, ipage is set only by make_empty_dir, and if any error occur,
407  * ipage should be released by this function.
408  */
409 struct page *get_new_data_page(struct inode *inode,
410 		struct page *ipage, pgoff_t index, bool new_i_size)
411 {
412 	struct address_space *mapping = inode->i_mapping;
413 	struct page *page;
414 	struct dnode_of_data dn;
415 	int err;
416 
417 	page = f2fs_grab_cache_page(mapping, index, true);
418 	if (!page) {
419 		/*
420 		 * before exiting, we should make sure ipage will be released
421 		 * if any error occur.
422 		 */
423 		f2fs_put_page(ipage, 1);
424 		return ERR_PTR(-ENOMEM);
425 	}
426 
427 	set_new_dnode(&dn, inode, ipage, NULL, 0);
428 	err = f2fs_reserve_block(&dn, index);
429 	if (err) {
430 		f2fs_put_page(page, 1);
431 		return ERR_PTR(err);
432 	}
433 	if (!ipage)
434 		f2fs_put_dnode(&dn);
435 
436 	if (PageUptodate(page))
437 		goto got_it;
438 
439 	if (dn.data_blkaddr == NEW_ADDR) {
440 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
441 		SetPageUptodate(page);
442 	} else {
443 		f2fs_put_page(page, 1);
444 
445 		/* if ipage exists, blkaddr should be NEW_ADDR */
446 		f2fs_bug_on(F2FS_I_SB(inode), ipage);
447 		page = get_lock_data_page(inode, index, true);
448 		if (IS_ERR(page))
449 			return page;
450 	}
451 got_it:
452 	if (new_i_size && i_size_read(inode) <
453 				((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
454 		i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
455 		/* Only the directory inode sets new_i_size */
456 		set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
457 	}
458 	return page;
459 }
460 
461 static int __allocate_data_block(struct dnode_of_data *dn)
462 {
463 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
464 	struct f2fs_inode_info *fi = F2FS_I(dn->inode);
465 	struct f2fs_summary sum;
466 	struct node_info ni;
467 	int seg = CURSEG_WARM_DATA;
468 	pgoff_t fofs;
469 
470 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
471 		return -EPERM;
472 
473 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
474 	if (dn->data_blkaddr == NEW_ADDR)
475 		goto alloc;
476 
477 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
478 		return -ENOSPC;
479 
480 alloc:
481 	get_node_info(sbi, dn->nid, &ni);
482 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
483 
484 	if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
485 		seg = CURSEG_DIRECT_IO;
486 
487 	allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
488 								&sum, seg);
489 	set_data_blkaddr(dn);
490 
491 	/* update i_size */
492 	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
493 							dn->ofs_in_node;
494 	if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
495 		i_size_write(dn->inode,
496 				((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
497 	return 0;
498 }
499 
500 static int __allocate_data_blocks(struct inode *inode, loff_t offset,
501 							size_t count)
502 {
503 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
504 	struct dnode_of_data dn;
505 	u64 start = F2FS_BYTES_TO_BLK(offset);
506 	u64 len = F2FS_BYTES_TO_BLK(count);
507 	bool allocated;
508 	u64 end_offset;
509 	int err = 0;
510 
511 	while (len) {
512 		f2fs_lock_op(sbi);
513 
514 		/* When reading holes, we need its node page */
515 		set_new_dnode(&dn, inode, NULL, NULL, 0);
516 		err = get_dnode_of_data(&dn, start, ALLOC_NODE);
517 		if (err)
518 			goto out;
519 
520 		allocated = false;
521 		end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
522 
523 		while (dn.ofs_in_node < end_offset && len) {
524 			block_t blkaddr;
525 
526 			if (unlikely(f2fs_cp_error(sbi))) {
527 				err = -EIO;
528 				goto sync_out;
529 			}
530 
531 			blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
532 			if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
533 				err = __allocate_data_block(&dn);
534 				if (err)
535 					goto sync_out;
536 				allocated = true;
537 			}
538 			len--;
539 			start++;
540 			dn.ofs_in_node++;
541 		}
542 
543 		if (allocated)
544 			sync_inode_page(&dn);
545 
546 		f2fs_put_dnode(&dn);
547 		f2fs_unlock_op(sbi);
548 
549 		f2fs_balance_fs(sbi, dn.node_changed);
550 	}
551 	return err;
552 
553 sync_out:
554 	if (allocated)
555 		sync_inode_page(&dn);
556 	f2fs_put_dnode(&dn);
557 out:
558 	f2fs_unlock_op(sbi);
559 	f2fs_balance_fs(sbi, dn.node_changed);
560 	return err;
561 }
562 
563 /*
564  * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
565  * f2fs_map_blocks structure.
566  * If original data blocks are allocated, then give them to blockdev.
567  * Otherwise,
568  *     a. preallocate requested block addresses
569  *     b. do not use extent cache for better performance
570  *     c. give the block addresses to blockdev
571  */
572 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
573 						int create, int flag)
574 {
575 	unsigned int maxblocks = map->m_len;
576 	struct dnode_of_data dn;
577 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
578 	int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
579 	pgoff_t pgofs, end_offset;
580 	int err = 0, ofs = 1;
581 	struct extent_info ei;
582 	bool allocated = false;
583 	block_t blkaddr;
584 
585 	map->m_len = 0;
586 	map->m_flags = 0;
587 
588 	/* it only supports block size == page size */
589 	pgofs =	(pgoff_t)map->m_lblk;
590 
591 	if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
592 		map->m_pblk = ei.blk + pgofs - ei.fofs;
593 		map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
594 		map->m_flags = F2FS_MAP_MAPPED;
595 		goto out;
596 	}
597 
598 	if (create)
599 		f2fs_lock_op(sbi);
600 
601 	/* When reading holes, we need its node page */
602 	set_new_dnode(&dn, inode, NULL, NULL, 0);
603 	err = get_dnode_of_data(&dn, pgofs, mode);
604 	if (err) {
605 		if (err == -ENOENT)
606 			err = 0;
607 		goto unlock_out;
608 	}
609 
610 	if (dn.data_blkaddr == NEW_ADDR || dn.data_blkaddr == NULL_ADDR) {
611 		if (create) {
612 			if (unlikely(f2fs_cp_error(sbi))) {
613 				err = -EIO;
614 				goto put_out;
615 			}
616 			err = __allocate_data_block(&dn);
617 			if (err)
618 				goto put_out;
619 			allocated = true;
620 			map->m_flags = F2FS_MAP_NEW;
621 		} else {
622 			if (flag != F2FS_GET_BLOCK_FIEMAP ||
623 						dn.data_blkaddr != NEW_ADDR) {
624 				if (flag == F2FS_GET_BLOCK_BMAP)
625 					err = -ENOENT;
626 				goto put_out;
627 			}
628 
629 			/*
630 			 * preallocated unwritten block should be mapped
631 			 * for fiemap.
632 			 */
633 			if (dn.data_blkaddr == NEW_ADDR)
634 				map->m_flags = F2FS_MAP_UNWRITTEN;
635 		}
636 	}
637 
638 	map->m_flags |= F2FS_MAP_MAPPED;
639 	map->m_pblk = dn.data_blkaddr;
640 	map->m_len = 1;
641 
642 	end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
643 	dn.ofs_in_node++;
644 	pgofs++;
645 
646 get_next:
647 	if (map->m_len >= maxblocks)
648 		goto sync_out;
649 
650 	if (dn.ofs_in_node >= end_offset) {
651 		if (allocated)
652 			sync_inode_page(&dn);
653 		allocated = false;
654 		f2fs_put_dnode(&dn);
655 
656 		if (create) {
657 			f2fs_unlock_op(sbi);
658 			f2fs_balance_fs(sbi, dn.node_changed);
659 			f2fs_lock_op(sbi);
660 		}
661 
662 		set_new_dnode(&dn, inode, NULL, NULL, 0);
663 		err = get_dnode_of_data(&dn, pgofs, mode);
664 		if (err) {
665 			if (err == -ENOENT)
666 				err = 0;
667 			goto unlock_out;
668 		}
669 
670 		end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
671 	}
672 
673 	blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
674 
675 	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
676 		if (create) {
677 			if (unlikely(f2fs_cp_error(sbi))) {
678 				err = -EIO;
679 				goto sync_out;
680 			}
681 			err = __allocate_data_block(&dn);
682 			if (err)
683 				goto sync_out;
684 			allocated = true;
685 			map->m_flags |= F2FS_MAP_NEW;
686 			blkaddr = dn.data_blkaddr;
687 		} else {
688 			/*
689 			 * we only merge preallocated unwritten blocks
690 			 * for fiemap.
691 			 */
692 			if (flag != F2FS_GET_BLOCK_FIEMAP ||
693 					blkaddr != NEW_ADDR)
694 				goto sync_out;
695 		}
696 	}
697 
698 	/* Give more consecutive addresses for the readahead */
699 	if ((map->m_pblk != NEW_ADDR &&
700 			blkaddr == (map->m_pblk + ofs)) ||
701 			(map->m_pblk == NEW_ADDR &&
702 			blkaddr == NEW_ADDR)) {
703 		ofs++;
704 		dn.ofs_in_node++;
705 		pgofs++;
706 		map->m_len++;
707 		goto get_next;
708 	}
709 
710 sync_out:
711 	if (allocated)
712 		sync_inode_page(&dn);
713 put_out:
714 	f2fs_put_dnode(&dn);
715 unlock_out:
716 	if (create) {
717 		f2fs_unlock_op(sbi);
718 		f2fs_balance_fs(sbi, dn.node_changed);
719 	}
720 out:
721 	trace_f2fs_map_blocks(inode, map, err);
722 	return err;
723 }
724 
725 static int __get_data_block(struct inode *inode, sector_t iblock,
726 			struct buffer_head *bh, int create, int flag)
727 {
728 	struct f2fs_map_blocks map;
729 	int ret;
730 
731 	map.m_lblk = iblock;
732 	map.m_len = bh->b_size >> inode->i_blkbits;
733 
734 	ret = f2fs_map_blocks(inode, &map, create, flag);
735 	if (!ret) {
736 		map_bh(bh, inode->i_sb, map.m_pblk);
737 		bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
738 		bh->b_size = map.m_len << inode->i_blkbits;
739 	}
740 	return ret;
741 }
742 
743 static int get_data_block(struct inode *inode, sector_t iblock,
744 			struct buffer_head *bh_result, int create, int flag)
745 {
746 	return __get_data_block(inode, iblock, bh_result, create, flag);
747 }
748 
749 static int get_data_block_dio(struct inode *inode, sector_t iblock,
750 			struct buffer_head *bh_result, int create)
751 {
752 	return __get_data_block(inode, iblock, bh_result, create,
753 						F2FS_GET_BLOCK_DIO);
754 }
755 
756 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
757 			struct buffer_head *bh_result, int create)
758 {
759 	/* Block number less than F2FS MAX BLOCKS */
760 	if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
761 		return -EFBIG;
762 
763 	return __get_data_block(inode, iblock, bh_result, create,
764 						F2FS_GET_BLOCK_BMAP);
765 }
766 
767 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
768 {
769 	return (offset >> inode->i_blkbits);
770 }
771 
772 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
773 {
774 	return (blk << inode->i_blkbits);
775 }
776 
777 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
778 		u64 start, u64 len)
779 {
780 	struct buffer_head map_bh;
781 	sector_t start_blk, last_blk;
782 	loff_t isize;
783 	u64 logical = 0, phys = 0, size = 0;
784 	u32 flags = 0;
785 	int ret = 0;
786 
787 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
788 	if (ret)
789 		return ret;
790 
791 	if (f2fs_has_inline_data(inode)) {
792 		ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
793 		if (ret != -EAGAIN)
794 			return ret;
795 	}
796 
797 	inode_lock(inode);
798 
799 	isize = i_size_read(inode);
800 	if (start >= isize)
801 		goto out;
802 
803 	if (start + len > isize)
804 		len = isize - start;
805 
806 	if (logical_to_blk(inode, len) == 0)
807 		len = blk_to_logical(inode, 1);
808 
809 	start_blk = logical_to_blk(inode, start);
810 	last_blk = logical_to_blk(inode, start + len - 1);
811 
812 next:
813 	memset(&map_bh, 0, sizeof(struct buffer_head));
814 	map_bh.b_size = len;
815 
816 	ret = get_data_block(inode, start_blk, &map_bh, 0,
817 					F2FS_GET_BLOCK_FIEMAP);
818 	if (ret)
819 		goto out;
820 
821 	/* HOLE */
822 	if (!buffer_mapped(&map_bh)) {
823 		/* Go through holes util pass the EOF */
824 		if (blk_to_logical(inode, start_blk++) < isize)
825 			goto prep_next;
826 		/* Found a hole beyond isize means no more extents.
827 		 * Note that the premise is that filesystems don't
828 		 * punch holes beyond isize and keep size unchanged.
829 		 */
830 		flags |= FIEMAP_EXTENT_LAST;
831 	}
832 
833 	if (size) {
834 		if (f2fs_encrypted_inode(inode))
835 			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
836 
837 		ret = fiemap_fill_next_extent(fieinfo, logical,
838 				phys, size, flags);
839 	}
840 
841 	if (start_blk > last_blk || ret)
842 		goto out;
843 
844 	logical = blk_to_logical(inode, start_blk);
845 	phys = blk_to_logical(inode, map_bh.b_blocknr);
846 	size = map_bh.b_size;
847 	flags = 0;
848 	if (buffer_unwritten(&map_bh))
849 		flags = FIEMAP_EXTENT_UNWRITTEN;
850 
851 	start_blk += logical_to_blk(inode, size);
852 
853 prep_next:
854 	cond_resched();
855 	if (fatal_signal_pending(current))
856 		ret = -EINTR;
857 	else
858 		goto next;
859 out:
860 	if (ret == 1)
861 		ret = 0;
862 
863 	inode_unlock(inode);
864 	return ret;
865 }
866 
867 /*
868  * This function was originally taken from fs/mpage.c, and customized for f2fs.
869  * Major change was from block_size == page_size in f2fs by default.
870  */
871 static int f2fs_mpage_readpages(struct address_space *mapping,
872 			struct list_head *pages, struct page *page,
873 			unsigned nr_pages)
874 {
875 	struct bio *bio = NULL;
876 	unsigned page_idx;
877 	sector_t last_block_in_bio = 0;
878 	struct inode *inode = mapping->host;
879 	const unsigned blkbits = inode->i_blkbits;
880 	const unsigned blocksize = 1 << blkbits;
881 	sector_t block_in_file;
882 	sector_t last_block;
883 	sector_t last_block_in_file;
884 	sector_t block_nr;
885 	struct block_device *bdev = inode->i_sb->s_bdev;
886 	struct f2fs_map_blocks map;
887 
888 	map.m_pblk = 0;
889 	map.m_lblk = 0;
890 	map.m_len = 0;
891 	map.m_flags = 0;
892 
893 	for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
894 
895 		prefetchw(&page->flags);
896 		if (pages) {
897 			page = list_entry(pages->prev, struct page, lru);
898 			list_del(&page->lru);
899 			if (add_to_page_cache_lru(page, mapping,
900 						  page->index, GFP_KERNEL))
901 				goto next_page;
902 		}
903 
904 		block_in_file = (sector_t)page->index;
905 		last_block = block_in_file + nr_pages;
906 		last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
907 								blkbits;
908 		if (last_block > last_block_in_file)
909 			last_block = last_block_in_file;
910 
911 		/*
912 		 * Map blocks using the previous result first.
913 		 */
914 		if ((map.m_flags & F2FS_MAP_MAPPED) &&
915 				block_in_file > map.m_lblk &&
916 				block_in_file < (map.m_lblk + map.m_len))
917 			goto got_it;
918 
919 		/*
920 		 * Then do more f2fs_map_blocks() calls until we are
921 		 * done with this page.
922 		 */
923 		map.m_flags = 0;
924 
925 		if (block_in_file < last_block) {
926 			map.m_lblk = block_in_file;
927 			map.m_len = last_block - block_in_file;
928 
929 			if (f2fs_map_blocks(inode, &map, 0,
930 							F2FS_GET_BLOCK_READ))
931 				goto set_error_page;
932 		}
933 got_it:
934 		if ((map.m_flags & F2FS_MAP_MAPPED)) {
935 			block_nr = map.m_pblk + block_in_file - map.m_lblk;
936 			SetPageMappedToDisk(page);
937 
938 			if (!PageUptodate(page) && !cleancache_get_page(page)) {
939 				SetPageUptodate(page);
940 				goto confused;
941 			}
942 		} else {
943 			zero_user_segment(page, 0, PAGE_CACHE_SIZE);
944 			SetPageUptodate(page);
945 			unlock_page(page);
946 			goto next_page;
947 		}
948 
949 		/*
950 		 * This page will go to BIO.  Do we need to send this
951 		 * BIO off first?
952 		 */
953 		if (bio && (last_block_in_bio != block_nr - 1)) {
954 submit_and_realloc:
955 			submit_bio(READ, bio);
956 			bio = NULL;
957 		}
958 		if (bio == NULL) {
959 			struct f2fs_crypto_ctx *ctx = NULL;
960 
961 			if (f2fs_encrypted_inode(inode) &&
962 					S_ISREG(inode->i_mode)) {
963 
964 				ctx = f2fs_get_crypto_ctx(inode);
965 				if (IS_ERR(ctx))
966 					goto set_error_page;
967 
968 				/* wait the page to be moved by cleaning */
969 				f2fs_wait_on_encrypted_page_writeback(
970 						F2FS_I_SB(inode), block_nr);
971 			}
972 
973 			bio = bio_alloc(GFP_KERNEL,
974 				min_t(int, nr_pages, BIO_MAX_PAGES));
975 			if (!bio) {
976 				if (ctx)
977 					f2fs_release_crypto_ctx(ctx);
978 				goto set_error_page;
979 			}
980 			bio->bi_bdev = bdev;
981 			bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
982 			bio->bi_end_io = f2fs_read_end_io;
983 			bio->bi_private = ctx;
984 		}
985 
986 		if (bio_add_page(bio, page, blocksize, 0) < blocksize)
987 			goto submit_and_realloc;
988 
989 		last_block_in_bio = block_nr;
990 		goto next_page;
991 set_error_page:
992 		SetPageError(page);
993 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
994 		unlock_page(page);
995 		goto next_page;
996 confused:
997 		if (bio) {
998 			submit_bio(READ, bio);
999 			bio = NULL;
1000 		}
1001 		unlock_page(page);
1002 next_page:
1003 		if (pages)
1004 			page_cache_release(page);
1005 	}
1006 	BUG_ON(pages && !list_empty(pages));
1007 	if (bio)
1008 		submit_bio(READ, bio);
1009 	return 0;
1010 }
1011 
1012 static int f2fs_read_data_page(struct file *file, struct page *page)
1013 {
1014 	struct inode *inode = page->mapping->host;
1015 	int ret = -EAGAIN;
1016 
1017 	trace_f2fs_readpage(page, DATA);
1018 
1019 	/* If the file has inline data, try to read it directly */
1020 	if (f2fs_has_inline_data(inode))
1021 		ret = f2fs_read_inline_data(inode, page);
1022 	if (ret == -EAGAIN)
1023 		ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1024 	return ret;
1025 }
1026 
1027 static int f2fs_read_data_pages(struct file *file,
1028 			struct address_space *mapping,
1029 			struct list_head *pages, unsigned nr_pages)
1030 {
1031 	struct inode *inode = file->f_mapping->host;
1032 	struct page *page = list_entry(pages->prev, struct page, lru);
1033 
1034 	trace_f2fs_readpages(inode, page, nr_pages);
1035 
1036 	/* If the file has inline data, skip readpages */
1037 	if (f2fs_has_inline_data(inode))
1038 		return 0;
1039 
1040 	return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1041 }
1042 
1043 int do_write_data_page(struct f2fs_io_info *fio)
1044 {
1045 	struct page *page = fio->page;
1046 	struct inode *inode = page->mapping->host;
1047 	struct dnode_of_data dn;
1048 	int err = 0;
1049 
1050 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1051 	err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1052 	if (err)
1053 		return err;
1054 
1055 	fio->blk_addr = dn.data_blkaddr;
1056 
1057 	/* This page is already truncated */
1058 	if (fio->blk_addr == NULL_ADDR) {
1059 		ClearPageUptodate(page);
1060 		goto out_writepage;
1061 	}
1062 
1063 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1064 
1065 		/* wait for GCed encrypted page writeback */
1066 		f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1067 							fio->blk_addr);
1068 
1069 		fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1070 		if (IS_ERR(fio->encrypted_page)) {
1071 			err = PTR_ERR(fio->encrypted_page);
1072 			goto out_writepage;
1073 		}
1074 	}
1075 
1076 	set_page_writeback(page);
1077 
1078 	/*
1079 	 * If current allocation needs SSR,
1080 	 * it had better in-place writes for updated data.
1081 	 */
1082 	if (unlikely(fio->blk_addr != NEW_ADDR &&
1083 			!is_cold_data(page) &&
1084 			!IS_ATOMIC_WRITTEN_PAGE(page) &&
1085 			need_inplace_update(inode))) {
1086 		rewrite_data_page(fio);
1087 		set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1088 		trace_f2fs_do_write_data_page(page, IPU);
1089 	} else {
1090 		write_data_page(&dn, fio);
1091 		set_data_blkaddr(&dn);
1092 		f2fs_update_extent_cache(&dn);
1093 		trace_f2fs_do_write_data_page(page, OPU);
1094 		set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1095 		if (page->index == 0)
1096 			set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1097 	}
1098 out_writepage:
1099 	f2fs_put_dnode(&dn);
1100 	return err;
1101 }
1102 
1103 static int f2fs_write_data_page(struct page *page,
1104 					struct writeback_control *wbc)
1105 {
1106 	struct inode *inode = page->mapping->host;
1107 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1108 	loff_t i_size = i_size_read(inode);
1109 	const pgoff_t end_index = ((unsigned long long) i_size)
1110 							>> PAGE_CACHE_SHIFT;
1111 	unsigned offset = 0;
1112 	bool need_balance_fs = false;
1113 	int err = 0;
1114 	struct f2fs_io_info fio = {
1115 		.sbi = sbi,
1116 		.type = DATA,
1117 		.rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1118 		.page = page,
1119 		.encrypted_page = NULL,
1120 	};
1121 
1122 	trace_f2fs_writepage(page, DATA);
1123 
1124 	if (page->index < end_index)
1125 		goto write;
1126 
1127 	/*
1128 	 * If the offset is out-of-range of file size,
1129 	 * this page does not have to be written to disk.
1130 	 */
1131 	offset = i_size & (PAGE_CACHE_SIZE - 1);
1132 	if ((page->index >= end_index + 1) || !offset)
1133 		goto out;
1134 
1135 	zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1136 write:
1137 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1138 		goto redirty_out;
1139 	if (f2fs_is_drop_cache(inode))
1140 		goto out;
1141 	if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1142 			available_free_memory(sbi, BASE_CHECK))
1143 		goto redirty_out;
1144 
1145 	/* Dentry blocks are controlled by checkpoint */
1146 	if (S_ISDIR(inode->i_mode)) {
1147 		if (unlikely(f2fs_cp_error(sbi)))
1148 			goto redirty_out;
1149 		err = do_write_data_page(&fio);
1150 		goto done;
1151 	}
1152 
1153 	/* we should bypass data pages to proceed the kworkder jobs */
1154 	if (unlikely(f2fs_cp_error(sbi))) {
1155 		SetPageError(page);
1156 		goto out;
1157 	}
1158 
1159 	if (!wbc->for_reclaim)
1160 		need_balance_fs = true;
1161 	else if (has_not_enough_free_secs(sbi, 0))
1162 		goto redirty_out;
1163 
1164 	err = -EAGAIN;
1165 	f2fs_lock_op(sbi);
1166 	if (f2fs_has_inline_data(inode))
1167 		err = f2fs_write_inline_data(inode, page);
1168 	if (err == -EAGAIN)
1169 		err = do_write_data_page(&fio);
1170 	f2fs_unlock_op(sbi);
1171 done:
1172 	if (err && err != -ENOENT)
1173 		goto redirty_out;
1174 
1175 	clear_cold_data(page);
1176 out:
1177 	inode_dec_dirty_pages(inode);
1178 	if (err)
1179 		ClearPageUptodate(page);
1180 	unlock_page(page);
1181 	f2fs_balance_fs(sbi, need_balance_fs);
1182 	if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi))) {
1183 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
1184 		remove_dirty_inode(inode);
1185 	}
1186 	return 0;
1187 
1188 redirty_out:
1189 	redirty_page_for_writepage(wbc, page);
1190 	return AOP_WRITEPAGE_ACTIVATE;
1191 }
1192 
1193 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1194 			void *data)
1195 {
1196 	struct address_space *mapping = data;
1197 	int ret = mapping->a_ops->writepage(page, wbc);
1198 	mapping_set_error(mapping, ret);
1199 	return ret;
1200 }
1201 
1202 /*
1203  * This function was copied from write_cche_pages from mm/page-writeback.c.
1204  * The major change is making write step of cold data page separately from
1205  * warm/hot data page.
1206  */
1207 static int f2fs_write_cache_pages(struct address_space *mapping,
1208 			struct writeback_control *wbc, writepage_t writepage,
1209 			void *data)
1210 {
1211 	int ret = 0;
1212 	int done = 0;
1213 	struct pagevec pvec;
1214 	int nr_pages;
1215 	pgoff_t uninitialized_var(writeback_index);
1216 	pgoff_t index;
1217 	pgoff_t end;		/* Inclusive */
1218 	pgoff_t done_index;
1219 	int cycled;
1220 	int range_whole = 0;
1221 	int tag;
1222 	int step = 0;
1223 
1224 	pagevec_init(&pvec, 0);
1225 next:
1226 	if (wbc->range_cyclic) {
1227 		writeback_index = mapping->writeback_index; /* prev offset */
1228 		index = writeback_index;
1229 		if (index == 0)
1230 			cycled = 1;
1231 		else
1232 			cycled = 0;
1233 		end = -1;
1234 	} else {
1235 		index = wbc->range_start >> PAGE_CACHE_SHIFT;
1236 		end = wbc->range_end >> PAGE_CACHE_SHIFT;
1237 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1238 			range_whole = 1;
1239 		cycled = 1; /* ignore range_cyclic tests */
1240 	}
1241 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1242 		tag = PAGECACHE_TAG_TOWRITE;
1243 	else
1244 		tag = PAGECACHE_TAG_DIRTY;
1245 retry:
1246 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1247 		tag_pages_for_writeback(mapping, index, end);
1248 	done_index = index;
1249 	while (!done && (index <= end)) {
1250 		int i;
1251 
1252 		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1253 			      min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1254 		if (nr_pages == 0)
1255 			break;
1256 
1257 		for (i = 0; i < nr_pages; i++) {
1258 			struct page *page = pvec.pages[i];
1259 
1260 			if (page->index > end) {
1261 				done = 1;
1262 				break;
1263 			}
1264 
1265 			done_index = page->index;
1266 
1267 			lock_page(page);
1268 
1269 			if (unlikely(page->mapping != mapping)) {
1270 continue_unlock:
1271 				unlock_page(page);
1272 				continue;
1273 			}
1274 
1275 			if (!PageDirty(page)) {
1276 				/* someone wrote it for us */
1277 				goto continue_unlock;
1278 			}
1279 
1280 			if (step == is_cold_data(page))
1281 				goto continue_unlock;
1282 
1283 			if (PageWriteback(page)) {
1284 				if (wbc->sync_mode != WB_SYNC_NONE)
1285 					f2fs_wait_on_page_writeback(page, DATA);
1286 				else
1287 					goto continue_unlock;
1288 			}
1289 
1290 			BUG_ON(PageWriteback(page));
1291 			if (!clear_page_dirty_for_io(page))
1292 				goto continue_unlock;
1293 
1294 			ret = (*writepage)(page, wbc, data);
1295 			if (unlikely(ret)) {
1296 				if (ret == AOP_WRITEPAGE_ACTIVATE) {
1297 					unlock_page(page);
1298 					ret = 0;
1299 				} else {
1300 					done_index = page->index + 1;
1301 					done = 1;
1302 					break;
1303 				}
1304 			}
1305 
1306 			if (--wbc->nr_to_write <= 0 &&
1307 			    wbc->sync_mode == WB_SYNC_NONE) {
1308 				done = 1;
1309 				break;
1310 			}
1311 		}
1312 		pagevec_release(&pvec);
1313 		cond_resched();
1314 	}
1315 
1316 	if (step < 1) {
1317 		step++;
1318 		goto next;
1319 	}
1320 
1321 	if (!cycled && !done) {
1322 		cycled = 1;
1323 		index = 0;
1324 		end = writeback_index - 1;
1325 		goto retry;
1326 	}
1327 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1328 		mapping->writeback_index = done_index;
1329 
1330 	return ret;
1331 }
1332 
1333 static int f2fs_write_data_pages(struct address_space *mapping,
1334 			    struct writeback_control *wbc)
1335 {
1336 	struct inode *inode = mapping->host;
1337 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1338 	bool locked = false;
1339 	int ret;
1340 	long diff;
1341 
1342 	trace_f2fs_writepages(mapping->host, wbc, DATA);
1343 
1344 	/* deal with chardevs and other special file */
1345 	if (!mapping->a_ops->writepage)
1346 		return 0;
1347 
1348 	/* skip writing if there is no dirty page in this inode */
1349 	if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1350 		return 0;
1351 
1352 	if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1353 			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1354 			available_free_memory(sbi, DIRTY_DENTS))
1355 		goto skip_write;
1356 
1357 	/* skip writing during file defragment */
1358 	if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1359 		goto skip_write;
1360 
1361 	/* during POR, we don't need to trigger writepage at all. */
1362 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1363 		goto skip_write;
1364 
1365 	diff = nr_pages_to_write(sbi, DATA, wbc);
1366 
1367 	if (!S_ISDIR(inode->i_mode)) {
1368 		mutex_lock(&sbi->writepages);
1369 		locked = true;
1370 	}
1371 	ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1372 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
1373 	if (locked)
1374 		mutex_unlock(&sbi->writepages);
1375 
1376 	remove_dirty_inode(inode);
1377 
1378 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1379 	return ret;
1380 
1381 skip_write:
1382 	wbc->pages_skipped += get_dirty_pages(inode);
1383 	return 0;
1384 }
1385 
1386 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1387 {
1388 	struct inode *inode = mapping->host;
1389 	loff_t i_size = i_size_read(inode);
1390 
1391 	if (to > i_size) {
1392 		truncate_pagecache(inode, i_size);
1393 		truncate_blocks(inode, i_size, true);
1394 	}
1395 }
1396 
1397 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1398 			struct page *page, loff_t pos, unsigned len,
1399 			block_t *blk_addr, bool *node_changed)
1400 {
1401 	struct inode *inode = page->mapping->host;
1402 	pgoff_t index = page->index;
1403 	struct dnode_of_data dn;
1404 	struct page *ipage;
1405 	bool locked = false;
1406 	struct extent_info ei;
1407 	int err = 0;
1408 
1409 	if (f2fs_has_inline_data(inode) ||
1410 			(pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1411 		f2fs_lock_op(sbi);
1412 		locked = true;
1413 	}
1414 restart:
1415 	/* check inline_data */
1416 	ipage = get_node_page(sbi, inode->i_ino);
1417 	if (IS_ERR(ipage)) {
1418 		err = PTR_ERR(ipage);
1419 		goto unlock_out;
1420 	}
1421 
1422 	set_new_dnode(&dn, inode, ipage, ipage, 0);
1423 
1424 	if (f2fs_has_inline_data(inode)) {
1425 		if (pos + len <= MAX_INLINE_DATA) {
1426 			read_inline_data(page, ipage);
1427 			set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1428 			sync_inode_page(&dn);
1429 		} else {
1430 			err = f2fs_convert_inline_page(&dn, page);
1431 			if (err)
1432 				goto out;
1433 			if (dn.data_blkaddr == NULL_ADDR)
1434 				err = f2fs_get_block(&dn, index);
1435 		}
1436 	} else if (locked) {
1437 		err = f2fs_get_block(&dn, index);
1438 	} else {
1439 		if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1440 			dn.data_blkaddr = ei.blk + index - ei.fofs;
1441 		} else {
1442 			bool restart = false;
1443 
1444 			/* hole case */
1445 			err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1446 			if (err || (!err && dn.data_blkaddr == NULL_ADDR))
1447 				restart = true;
1448 			if (restart) {
1449 				f2fs_put_dnode(&dn);
1450 				f2fs_lock_op(sbi);
1451 				locked = true;
1452 				goto restart;
1453 			}
1454 		}
1455 	}
1456 
1457 	/* convert_inline_page can make node_changed */
1458 	*blk_addr = dn.data_blkaddr;
1459 	*node_changed = dn.node_changed;
1460 out:
1461 	f2fs_put_dnode(&dn);
1462 unlock_out:
1463 	if (locked)
1464 		f2fs_unlock_op(sbi);
1465 	return err;
1466 }
1467 
1468 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1469 		loff_t pos, unsigned len, unsigned flags,
1470 		struct page **pagep, void **fsdata)
1471 {
1472 	struct inode *inode = mapping->host;
1473 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1474 	struct page *page = NULL;
1475 	pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1476 	bool need_balance = false;
1477 	block_t blkaddr = NULL_ADDR;
1478 	int err = 0;
1479 
1480 	trace_f2fs_write_begin(inode, pos, len, flags);
1481 
1482 	/*
1483 	 * We should check this at this moment to avoid deadlock on inode page
1484 	 * and #0 page. The locking rule for inline_data conversion should be:
1485 	 * lock_page(page #0) -> lock_page(inode_page)
1486 	 */
1487 	if (index != 0) {
1488 		err = f2fs_convert_inline_inode(inode);
1489 		if (err)
1490 			goto fail;
1491 	}
1492 repeat:
1493 	page = grab_cache_page_write_begin(mapping, index, flags);
1494 	if (!page) {
1495 		err = -ENOMEM;
1496 		goto fail;
1497 	}
1498 
1499 	*pagep = page;
1500 
1501 	err = prepare_write_begin(sbi, page, pos, len,
1502 					&blkaddr, &need_balance);
1503 	if (err)
1504 		goto fail;
1505 
1506 	if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1507 		unlock_page(page);
1508 		f2fs_balance_fs(sbi, true);
1509 		lock_page(page);
1510 		if (page->mapping != mapping) {
1511 			/* The page got truncated from under us */
1512 			f2fs_put_page(page, 1);
1513 			goto repeat;
1514 		}
1515 	}
1516 
1517 	f2fs_wait_on_page_writeback(page, DATA);
1518 
1519 	/* wait for GCed encrypted page writeback */
1520 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1521 		f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1522 
1523 	if (len == PAGE_CACHE_SIZE)
1524 		goto out_update;
1525 	if (PageUptodate(page))
1526 		goto out_clear;
1527 
1528 	if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1529 		unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1530 		unsigned end = start + len;
1531 
1532 		/* Reading beyond i_size is simple: memset to zero */
1533 		zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1534 		goto out_update;
1535 	}
1536 
1537 	if (blkaddr == NEW_ADDR) {
1538 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1539 	} else {
1540 		struct f2fs_io_info fio = {
1541 			.sbi = sbi,
1542 			.type = DATA,
1543 			.rw = READ_SYNC,
1544 			.blk_addr = blkaddr,
1545 			.page = page,
1546 			.encrypted_page = NULL,
1547 		};
1548 		err = f2fs_submit_page_bio(&fio);
1549 		if (err)
1550 			goto fail;
1551 
1552 		lock_page(page);
1553 		if (unlikely(!PageUptodate(page))) {
1554 			err = -EIO;
1555 			goto fail;
1556 		}
1557 		if (unlikely(page->mapping != mapping)) {
1558 			f2fs_put_page(page, 1);
1559 			goto repeat;
1560 		}
1561 
1562 		/* avoid symlink page */
1563 		if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1564 			err = f2fs_decrypt_one(inode, page);
1565 			if (err)
1566 				goto fail;
1567 		}
1568 	}
1569 out_update:
1570 	SetPageUptodate(page);
1571 out_clear:
1572 	clear_cold_data(page);
1573 	return 0;
1574 
1575 fail:
1576 	f2fs_put_page(page, 1);
1577 	f2fs_write_failed(mapping, pos + len);
1578 	return err;
1579 }
1580 
1581 static int f2fs_write_end(struct file *file,
1582 			struct address_space *mapping,
1583 			loff_t pos, unsigned len, unsigned copied,
1584 			struct page *page, void *fsdata)
1585 {
1586 	struct inode *inode = page->mapping->host;
1587 
1588 	trace_f2fs_write_end(inode, pos, len, copied);
1589 
1590 	set_page_dirty(page);
1591 
1592 	if (pos + copied > i_size_read(inode)) {
1593 		i_size_write(inode, pos + copied);
1594 		mark_inode_dirty(inode);
1595 		update_inode_page(inode);
1596 	}
1597 
1598 	f2fs_put_page(page, 1);
1599 	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1600 	return copied;
1601 }
1602 
1603 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1604 			   loff_t offset)
1605 {
1606 	unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1607 
1608 	if (offset & blocksize_mask)
1609 		return -EINVAL;
1610 
1611 	if (iov_iter_alignment(iter) & blocksize_mask)
1612 		return -EINVAL;
1613 
1614 	return 0;
1615 }
1616 
1617 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1618 			      loff_t offset)
1619 {
1620 	struct file *file = iocb->ki_filp;
1621 	struct address_space *mapping = file->f_mapping;
1622 	struct inode *inode = mapping->host;
1623 	size_t count = iov_iter_count(iter);
1624 	int err;
1625 
1626 	/* we don't need to use inline_data strictly */
1627 	err = f2fs_convert_inline_inode(inode);
1628 	if (err)
1629 		return err;
1630 
1631 	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1632 		return 0;
1633 
1634 	err = check_direct_IO(inode, iter, offset);
1635 	if (err)
1636 		return err;
1637 
1638 	trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1639 
1640 	if (iov_iter_rw(iter) == WRITE) {
1641 		err = __allocate_data_blocks(inode, offset, count);
1642 		if (err)
1643 			goto out;
1644 	}
1645 
1646 	err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1647 out:
1648 	if (err < 0 && iov_iter_rw(iter) == WRITE)
1649 		f2fs_write_failed(mapping, offset + count);
1650 
1651 	trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1652 
1653 	return err;
1654 }
1655 
1656 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1657 							unsigned int length)
1658 {
1659 	struct inode *inode = page->mapping->host;
1660 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1661 
1662 	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1663 		(offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1664 		return;
1665 
1666 	if (PageDirty(page)) {
1667 		if (inode->i_ino == F2FS_META_INO(sbi))
1668 			dec_page_count(sbi, F2FS_DIRTY_META);
1669 		else if (inode->i_ino == F2FS_NODE_INO(sbi))
1670 			dec_page_count(sbi, F2FS_DIRTY_NODES);
1671 		else
1672 			inode_dec_dirty_pages(inode);
1673 	}
1674 
1675 	/* This is atomic written page, keep Private */
1676 	if (IS_ATOMIC_WRITTEN_PAGE(page))
1677 		return;
1678 
1679 	ClearPagePrivate(page);
1680 }
1681 
1682 int f2fs_release_page(struct page *page, gfp_t wait)
1683 {
1684 	/* If this is dirty page, keep PagePrivate */
1685 	if (PageDirty(page))
1686 		return 0;
1687 
1688 	/* This is atomic written page, keep Private */
1689 	if (IS_ATOMIC_WRITTEN_PAGE(page))
1690 		return 0;
1691 
1692 	ClearPagePrivate(page);
1693 	return 1;
1694 }
1695 
1696 static int f2fs_set_data_page_dirty(struct page *page)
1697 {
1698 	struct address_space *mapping = page->mapping;
1699 	struct inode *inode = mapping->host;
1700 
1701 	trace_f2fs_set_page_dirty(page, DATA);
1702 
1703 	SetPageUptodate(page);
1704 
1705 	if (f2fs_is_atomic_file(inode)) {
1706 		if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1707 			register_inmem_page(inode, page);
1708 			return 1;
1709 		}
1710 		/*
1711 		 * Previously, this page has been registered, we just
1712 		 * return here.
1713 		 */
1714 		return 0;
1715 	}
1716 
1717 	if (!PageDirty(page)) {
1718 		__set_page_dirty_nobuffers(page);
1719 		update_dirty_page(inode, page);
1720 		return 1;
1721 	}
1722 	return 0;
1723 }
1724 
1725 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1726 {
1727 	struct inode *inode = mapping->host;
1728 
1729 	if (f2fs_has_inline_data(inode))
1730 		return 0;
1731 
1732 	/* make sure allocating whole blocks */
1733 	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1734 		filemap_write_and_wait(mapping);
1735 
1736 	return generic_block_bmap(mapping, block, get_data_block_bmap);
1737 }
1738 
1739 const struct address_space_operations f2fs_dblock_aops = {
1740 	.readpage	= f2fs_read_data_page,
1741 	.readpages	= f2fs_read_data_pages,
1742 	.writepage	= f2fs_write_data_page,
1743 	.writepages	= f2fs_write_data_pages,
1744 	.write_begin	= f2fs_write_begin,
1745 	.write_end	= f2fs_write_end,
1746 	.set_page_dirty	= f2fs_set_data_page_dirty,
1747 	.invalidatepage	= f2fs_invalidate_page,
1748 	.releasepage	= f2fs_release_page,
1749 	.direct_IO	= f2fs_direct_IO,
1750 	.bmap		= f2fs_bmap,
1751 };
1752