xref: /linux/fs/f2fs/data.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
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/blkdev.h>
18 #include <linux/bio.h>
19 #include <linux/prefetch.h>
20 #include <linux/uio.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 struct kmem_cache *extent_tree_slab;
29 static struct kmem_cache *extent_node_slab;
30 
31 static void f2fs_read_end_io(struct bio *bio, int err)
32 {
33 	struct bio_vec *bvec;
34 	int i;
35 
36 	bio_for_each_segment_all(bvec, bio, i) {
37 		struct page *page = bvec->bv_page;
38 
39 		if (!err) {
40 			SetPageUptodate(page);
41 		} else {
42 			ClearPageUptodate(page);
43 			SetPageError(page);
44 		}
45 		unlock_page(page);
46 	}
47 	bio_put(bio);
48 }
49 
50 static void f2fs_write_end_io(struct bio *bio, int err)
51 {
52 	struct f2fs_sb_info *sbi = bio->bi_private;
53 	struct bio_vec *bvec;
54 	int i;
55 
56 	bio_for_each_segment_all(bvec, bio, i) {
57 		struct page *page = bvec->bv_page;
58 
59 		if (unlikely(err)) {
60 			set_page_dirty(page);
61 			set_bit(AS_EIO, &page->mapping->flags);
62 			f2fs_stop_checkpoint(sbi);
63 		}
64 		end_page_writeback(page);
65 		dec_page_count(sbi, F2FS_WRITEBACK);
66 	}
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(blk_addr);
88 	bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
89 	bio->bi_private = sbi;
90 
91 	return bio;
92 }
93 
94 static void __submit_merged_bio(struct f2fs_bio_info *io)
95 {
96 	struct f2fs_io_info *fio = &io->fio;
97 
98 	if (!io->bio)
99 		return;
100 
101 	if (is_read_io(fio->rw))
102 		trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
103 	else
104 		trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
105 
106 	submit_bio(fio->rw, io->bio);
107 	io->bio = NULL;
108 }
109 
110 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
111 				enum page_type type, int rw)
112 {
113 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
114 	struct f2fs_bio_info *io;
115 
116 	io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
117 
118 	down_write(&io->io_rwsem);
119 
120 	/* change META to META_FLUSH in the checkpoint procedure */
121 	if (type >= META_FLUSH) {
122 		io->fio.type = META_FLUSH;
123 		if (test_opt(sbi, NOBARRIER))
124 			io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
125 		else
126 			io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
127 	}
128 	__submit_merged_bio(io);
129 	up_write(&io->io_rwsem);
130 }
131 
132 /*
133  * Fill the locked page with data located in the block address.
134  * Return unlocked page.
135  */
136 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
137 					struct f2fs_io_info *fio)
138 {
139 	struct bio *bio;
140 
141 	trace_f2fs_submit_page_bio(page, fio);
142 	f2fs_trace_ios(page, fio, 0);
143 
144 	/* Allocate a new bio */
145 	bio = __bio_alloc(sbi, fio->blk_addr, 1, is_read_io(fio->rw));
146 
147 	if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
148 		bio_put(bio);
149 		f2fs_put_page(page, 1);
150 		return -EFAULT;
151 	}
152 
153 	submit_bio(fio->rw, bio);
154 	return 0;
155 }
156 
157 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
158 					struct f2fs_io_info *fio)
159 {
160 	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
161 	struct f2fs_bio_info *io;
162 	bool is_read = is_read_io(fio->rw);
163 
164 	io = is_read ? &sbi->read_io : &sbi->write_io[btype];
165 
166 	verify_block_addr(sbi, fio->blk_addr);
167 
168 	down_write(&io->io_rwsem);
169 
170 	if (!is_read)
171 		inc_page_count(sbi, F2FS_WRITEBACK);
172 
173 	if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
174 						io->fio.rw != fio->rw))
175 		__submit_merged_bio(io);
176 alloc_new:
177 	if (io->bio == NULL) {
178 		int bio_blocks = MAX_BIO_BLOCKS(sbi);
179 
180 		io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
181 		io->fio = *fio;
182 	}
183 
184 	if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
185 							PAGE_CACHE_SIZE) {
186 		__submit_merged_bio(io);
187 		goto alloc_new;
188 	}
189 
190 	io->last_block_in_bio = fio->blk_addr;
191 	f2fs_trace_ios(page, fio, 0);
192 
193 	up_write(&io->io_rwsem);
194 	trace_f2fs_submit_page_mbio(page, fio);
195 }
196 
197 /*
198  * Lock ordering for the change of data block address:
199  * ->data_page
200  *  ->node_page
201  *    update block addresses in the node page
202  */
203 void set_data_blkaddr(struct dnode_of_data *dn)
204 {
205 	struct f2fs_node *rn;
206 	__le32 *addr_array;
207 	struct page *node_page = dn->node_page;
208 	unsigned int ofs_in_node = dn->ofs_in_node;
209 
210 	f2fs_wait_on_page_writeback(node_page, NODE);
211 
212 	rn = F2FS_NODE(node_page);
213 
214 	/* Get physical address of data block */
215 	addr_array = blkaddr_in_node(rn);
216 	addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
217 	set_page_dirty(node_page);
218 }
219 
220 int reserve_new_block(struct dnode_of_data *dn)
221 {
222 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
223 
224 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
225 		return -EPERM;
226 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
227 		return -ENOSPC;
228 
229 	trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
230 
231 	dn->data_blkaddr = NEW_ADDR;
232 	set_data_blkaddr(dn);
233 	mark_inode_dirty(dn->inode);
234 	sync_inode_page(dn);
235 	return 0;
236 }
237 
238 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
239 {
240 	bool need_put = dn->inode_page ? false : true;
241 	int err;
242 
243 	err = get_dnode_of_data(dn, index, ALLOC_NODE);
244 	if (err)
245 		return err;
246 
247 	if (dn->data_blkaddr == NULL_ADDR)
248 		err = reserve_new_block(dn);
249 	if (err || need_put)
250 		f2fs_put_dnode(dn);
251 	return err;
252 }
253 
254 static void f2fs_map_bh(struct super_block *sb, pgoff_t pgofs,
255 			struct extent_info *ei, struct buffer_head *bh_result)
256 {
257 	unsigned int blkbits = sb->s_blocksize_bits;
258 	size_t max_size = bh_result->b_size;
259 	size_t mapped_size;
260 
261 	clear_buffer_new(bh_result);
262 	map_bh(bh_result, sb, ei->blk + pgofs - ei->fofs);
263 	mapped_size = (ei->fofs + ei->len - pgofs) << blkbits;
264 	bh_result->b_size = min(max_size, mapped_size);
265 }
266 
267 static bool lookup_extent_info(struct inode *inode, pgoff_t pgofs,
268 							struct extent_info *ei)
269 {
270 	struct f2fs_inode_info *fi = F2FS_I(inode);
271 	pgoff_t start_fofs, end_fofs;
272 	block_t start_blkaddr;
273 
274 	read_lock(&fi->ext_lock);
275 	if (fi->ext.len == 0) {
276 		read_unlock(&fi->ext_lock);
277 		return false;
278 	}
279 
280 	stat_inc_total_hit(inode->i_sb);
281 
282 	start_fofs = fi->ext.fofs;
283 	end_fofs = fi->ext.fofs + fi->ext.len - 1;
284 	start_blkaddr = fi->ext.blk;
285 
286 	if (pgofs >= start_fofs && pgofs <= end_fofs) {
287 		*ei = fi->ext;
288 		stat_inc_read_hit(inode->i_sb);
289 		read_unlock(&fi->ext_lock);
290 		return true;
291 	}
292 	read_unlock(&fi->ext_lock);
293 	return false;
294 }
295 
296 static bool update_extent_info(struct inode *inode, pgoff_t fofs,
297 								block_t blkaddr)
298 {
299 	struct f2fs_inode_info *fi = F2FS_I(inode);
300 	pgoff_t start_fofs, end_fofs;
301 	block_t start_blkaddr, end_blkaddr;
302 	int need_update = true;
303 
304 	write_lock(&fi->ext_lock);
305 
306 	start_fofs = fi->ext.fofs;
307 	end_fofs = fi->ext.fofs + fi->ext.len - 1;
308 	start_blkaddr = fi->ext.blk;
309 	end_blkaddr = fi->ext.blk + fi->ext.len - 1;
310 
311 	/* Drop and initialize the matched extent */
312 	if (fi->ext.len == 1 && fofs == start_fofs)
313 		fi->ext.len = 0;
314 
315 	/* Initial extent */
316 	if (fi->ext.len == 0) {
317 		if (blkaddr != NULL_ADDR) {
318 			fi->ext.fofs = fofs;
319 			fi->ext.blk = blkaddr;
320 			fi->ext.len = 1;
321 		}
322 		goto end_update;
323 	}
324 
325 	/* Front merge */
326 	if (fofs == start_fofs - 1 && blkaddr == start_blkaddr - 1) {
327 		fi->ext.fofs--;
328 		fi->ext.blk--;
329 		fi->ext.len++;
330 		goto end_update;
331 	}
332 
333 	/* Back merge */
334 	if (fofs == end_fofs + 1 && blkaddr == end_blkaddr + 1) {
335 		fi->ext.len++;
336 		goto end_update;
337 	}
338 
339 	/* Split the existing extent */
340 	if (fi->ext.len > 1 &&
341 		fofs >= start_fofs && fofs <= end_fofs) {
342 		if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
343 			fi->ext.len = fofs - start_fofs;
344 		} else {
345 			fi->ext.fofs = fofs + 1;
346 			fi->ext.blk = start_blkaddr + fofs - start_fofs + 1;
347 			fi->ext.len -= fofs - start_fofs + 1;
348 		}
349 	} else {
350 		need_update = false;
351 	}
352 
353 	/* Finally, if the extent is very fragmented, let's drop the cache. */
354 	if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
355 		fi->ext.len = 0;
356 		set_inode_flag(fi, FI_NO_EXTENT);
357 		need_update = true;
358 	}
359 end_update:
360 	write_unlock(&fi->ext_lock);
361 	return need_update;
362 }
363 
364 static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
365 				struct extent_tree *et, struct extent_info *ei,
366 				struct rb_node *parent, struct rb_node **p)
367 {
368 	struct extent_node *en;
369 
370 	en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
371 	if (!en)
372 		return NULL;
373 
374 	en->ei = *ei;
375 	INIT_LIST_HEAD(&en->list);
376 
377 	rb_link_node(&en->rb_node, parent, p);
378 	rb_insert_color(&en->rb_node, &et->root);
379 	et->count++;
380 	atomic_inc(&sbi->total_ext_node);
381 	return en;
382 }
383 
384 static void __detach_extent_node(struct f2fs_sb_info *sbi,
385 				struct extent_tree *et, struct extent_node *en)
386 {
387 	rb_erase(&en->rb_node, &et->root);
388 	et->count--;
389 	atomic_dec(&sbi->total_ext_node);
390 
391 	if (et->cached_en == en)
392 		et->cached_en = NULL;
393 }
394 
395 static struct extent_tree *__find_extent_tree(struct f2fs_sb_info *sbi,
396 							nid_t ino)
397 {
398 	struct extent_tree *et;
399 
400 	down_read(&sbi->extent_tree_lock);
401 	et = radix_tree_lookup(&sbi->extent_tree_root, ino);
402 	if (!et) {
403 		up_read(&sbi->extent_tree_lock);
404 		return NULL;
405 	}
406 	atomic_inc(&et->refcount);
407 	up_read(&sbi->extent_tree_lock);
408 
409 	return et;
410 }
411 
412 static struct extent_tree *__grab_extent_tree(struct inode *inode)
413 {
414 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
415 	struct extent_tree *et;
416 	nid_t ino = inode->i_ino;
417 
418 	down_write(&sbi->extent_tree_lock);
419 	et = radix_tree_lookup(&sbi->extent_tree_root, ino);
420 	if (!et) {
421 		et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
422 		f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
423 		memset(et, 0, sizeof(struct extent_tree));
424 		et->ino = ino;
425 		et->root = RB_ROOT;
426 		et->cached_en = NULL;
427 		rwlock_init(&et->lock);
428 		atomic_set(&et->refcount, 0);
429 		et->count = 0;
430 		sbi->total_ext_tree++;
431 	}
432 	atomic_inc(&et->refcount);
433 	up_write(&sbi->extent_tree_lock);
434 
435 	return et;
436 }
437 
438 static struct extent_node *__lookup_extent_tree(struct extent_tree *et,
439 							unsigned int fofs)
440 {
441 	struct rb_node *node = et->root.rb_node;
442 	struct extent_node *en;
443 
444 	if (et->cached_en) {
445 		struct extent_info *cei = &et->cached_en->ei;
446 
447 		if (cei->fofs <= fofs && cei->fofs + cei->len > fofs)
448 			return et->cached_en;
449 	}
450 
451 	while (node) {
452 		en = rb_entry(node, struct extent_node, rb_node);
453 
454 		if (fofs < en->ei.fofs) {
455 			node = node->rb_left;
456 		} else if (fofs >= en->ei.fofs + en->ei.len) {
457 			node = node->rb_right;
458 		} else {
459 			et->cached_en = en;
460 			return en;
461 		}
462 	}
463 	return NULL;
464 }
465 
466 static struct extent_node *__try_back_merge(struct f2fs_sb_info *sbi,
467 				struct extent_tree *et, struct extent_node *en)
468 {
469 	struct extent_node *prev;
470 	struct rb_node *node;
471 
472 	node = rb_prev(&en->rb_node);
473 	if (!node)
474 		return NULL;
475 
476 	prev = rb_entry(node, struct extent_node, rb_node);
477 	if (__is_back_mergeable(&en->ei, &prev->ei)) {
478 		en->ei.fofs = prev->ei.fofs;
479 		en->ei.blk = prev->ei.blk;
480 		en->ei.len += prev->ei.len;
481 		__detach_extent_node(sbi, et, prev);
482 		return prev;
483 	}
484 	return NULL;
485 }
486 
487 static struct extent_node *__try_front_merge(struct f2fs_sb_info *sbi,
488 				struct extent_tree *et, struct extent_node *en)
489 {
490 	struct extent_node *next;
491 	struct rb_node *node;
492 
493 	node = rb_next(&en->rb_node);
494 	if (!node)
495 		return NULL;
496 
497 	next = rb_entry(node, struct extent_node, rb_node);
498 	if (__is_front_mergeable(&en->ei, &next->ei)) {
499 		en->ei.len += next->ei.len;
500 		__detach_extent_node(sbi, et, next);
501 		return next;
502 	}
503 	return NULL;
504 }
505 
506 static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
507 				struct extent_tree *et, struct extent_info *ei,
508 				struct extent_node **den)
509 {
510 	struct rb_node **p = &et->root.rb_node;
511 	struct rb_node *parent = NULL;
512 	struct extent_node *en;
513 
514 	while (*p) {
515 		parent = *p;
516 		en = rb_entry(parent, struct extent_node, rb_node);
517 
518 		if (ei->fofs < en->ei.fofs) {
519 			if (__is_front_mergeable(ei, &en->ei)) {
520 				f2fs_bug_on(sbi, !den);
521 				en->ei.fofs = ei->fofs;
522 				en->ei.blk = ei->blk;
523 				en->ei.len += ei->len;
524 				*den = __try_back_merge(sbi, et, en);
525 				return en;
526 			}
527 			p = &(*p)->rb_left;
528 		} else if (ei->fofs >= en->ei.fofs + en->ei.len) {
529 			if (__is_back_mergeable(ei, &en->ei)) {
530 				f2fs_bug_on(sbi, !den);
531 				en->ei.len += ei->len;
532 				*den = __try_front_merge(sbi, et, en);
533 				return en;
534 			}
535 			p = &(*p)->rb_right;
536 		} else {
537 			f2fs_bug_on(sbi, 1);
538 		}
539 	}
540 
541 	return __attach_extent_node(sbi, et, ei, parent, p);
542 }
543 
544 static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
545 					struct extent_tree *et, bool free_all)
546 {
547 	struct rb_node *node, *next;
548 	struct extent_node *en;
549 	unsigned int count = et->count;
550 
551 	node = rb_first(&et->root);
552 	while (node) {
553 		next = rb_next(node);
554 		en = rb_entry(node, struct extent_node, rb_node);
555 
556 		if (free_all) {
557 			spin_lock(&sbi->extent_lock);
558 			if (!list_empty(&en->list))
559 				list_del_init(&en->list);
560 			spin_unlock(&sbi->extent_lock);
561 		}
562 
563 		if (free_all || list_empty(&en->list)) {
564 			__detach_extent_node(sbi, et, en);
565 			kmem_cache_free(extent_node_slab, en);
566 		}
567 		node = next;
568 	}
569 
570 	return count - et->count;
571 }
572 
573 static void f2fs_init_extent_tree(struct inode *inode,
574 						struct f2fs_extent *i_ext)
575 {
576 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
577 	struct extent_tree *et;
578 	struct extent_node *en;
579 	struct extent_info ei;
580 
581 	if (le32_to_cpu(i_ext->len) < F2FS_MIN_EXTENT_LEN)
582 		return;
583 
584 	et = __grab_extent_tree(inode);
585 
586 	write_lock(&et->lock);
587 	if (et->count)
588 		goto out;
589 
590 	set_extent_info(&ei, le32_to_cpu(i_ext->fofs),
591 		le32_to_cpu(i_ext->blk), le32_to_cpu(i_ext->len));
592 
593 	en = __insert_extent_tree(sbi, et, &ei, NULL);
594 	if (en) {
595 		et->cached_en = en;
596 
597 		spin_lock(&sbi->extent_lock);
598 		list_add_tail(&en->list, &sbi->extent_list);
599 		spin_unlock(&sbi->extent_lock);
600 	}
601 out:
602 	write_unlock(&et->lock);
603 	atomic_dec(&et->refcount);
604 }
605 
606 static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
607 							struct extent_info *ei)
608 {
609 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
610 	struct extent_tree *et;
611 	struct extent_node *en;
612 
613 	trace_f2fs_lookup_extent_tree_start(inode, pgofs);
614 
615 	et = __find_extent_tree(sbi, inode->i_ino);
616 	if (!et)
617 		return false;
618 
619 	read_lock(&et->lock);
620 	en = __lookup_extent_tree(et, pgofs);
621 	if (en) {
622 		*ei = en->ei;
623 		spin_lock(&sbi->extent_lock);
624 		if (!list_empty(&en->list))
625 			list_move_tail(&en->list, &sbi->extent_list);
626 		spin_unlock(&sbi->extent_lock);
627 		stat_inc_read_hit(sbi->sb);
628 	}
629 	stat_inc_total_hit(sbi->sb);
630 	read_unlock(&et->lock);
631 
632 	trace_f2fs_lookup_extent_tree_end(inode, pgofs, en);
633 
634 	atomic_dec(&et->refcount);
635 	return en ? true : false;
636 }
637 
638 static void f2fs_update_extent_tree(struct inode *inode, pgoff_t fofs,
639 							block_t blkaddr)
640 {
641 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
642 	struct extent_tree *et;
643 	struct extent_node *en = NULL, *en1 = NULL, *en2 = NULL, *en3 = NULL;
644 	struct extent_node *den = NULL;
645 	struct extent_info ei, dei;
646 	unsigned int endofs;
647 
648 	trace_f2fs_update_extent_tree(inode, fofs, blkaddr);
649 
650 	et = __grab_extent_tree(inode);
651 
652 	write_lock(&et->lock);
653 
654 	/* 1. lookup and remove existing extent info in cache */
655 	en = __lookup_extent_tree(et, fofs);
656 	if (!en)
657 		goto update_extent;
658 
659 	dei = en->ei;
660 	__detach_extent_node(sbi, et, en);
661 
662 	/* 2. if extent can be split more, split and insert the left part */
663 	if (dei.len > 1) {
664 		/*  insert left part of split extent into cache */
665 		if (fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
666 			set_extent_info(&ei, dei.fofs, dei.blk,
667 							fofs - dei.fofs);
668 			en1 = __insert_extent_tree(sbi, et, &ei, NULL);
669 		}
670 
671 		/* insert right part of split extent into cache */
672 		endofs = dei.fofs + dei.len - 1;
673 		if (endofs - fofs >= F2FS_MIN_EXTENT_LEN) {
674 			set_extent_info(&ei, fofs + 1,
675 				fofs - dei.fofs + dei.blk, endofs - fofs);
676 			en2 = __insert_extent_tree(sbi, et, &ei, NULL);
677 		}
678 	}
679 
680 update_extent:
681 	/* 3. update extent in extent cache */
682 	if (blkaddr) {
683 		set_extent_info(&ei, fofs, blkaddr, 1);
684 		en3 = __insert_extent_tree(sbi, et, &ei, &den);
685 	}
686 
687 	/* 4. update in global extent list */
688 	spin_lock(&sbi->extent_lock);
689 	if (en && !list_empty(&en->list))
690 		list_del(&en->list);
691 	/*
692 	 * en1 and en2 split from en, they will become more and more smaller
693 	 * fragments after splitting several times. So if the length is smaller
694 	 * than F2FS_MIN_EXTENT_LEN, we will not add them into extent tree.
695 	 */
696 	if (en1)
697 		list_add_tail(&en1->list, &sbi->extent_list);
698 	if (en2)
699 		list_add_tail(&en2->list, &sbi->extent_list);
700 	if (en3) {
701 		if (list_empty(&en3->list))
702 			list_add_tail(&en3->list, &sbi->extent_list);
703 		else
704 			list_move_tail(&en3->list, &sbi->extent_list);
705 	}
706 	if (den && !list_empty(&den->list))
707 		list_del(&den->list);
708 	spin_unlock(&sbi->extent_lock);
709 
710 	/* 5. release extent node */
711 	if (en)
712 		kmem_cache_free(extent_node_slab, en);
713 	if (den)
714 		kmem_cache_free(extent_node_slab, den);
715 
716 	write_unlock(&et->lock);
717 	atomic_dec(&et->refcount);
718 }
719 
720 void f2fs_preserve_extent_tree(struct inode *inode)
721 {
722 	struct extent_tree *et;
723 	struct extent_info *ext = &F2FS_I(inode)->ext;
724 	bool sync = false;
725 
726 	if (!test_opt(F2FS_I_SB(inode), EXTENT_CACHE))
727 		return;
728 
729 	et = __find_extent_tree(F2FS_I_SB(inode), inode->i_ino);
730 	if (!et) {
731 		if (ext->len) {
732 			ext->len = 0;
733 			update_inode_page(inode);
734 		}
735 		return;
736 	}
737 
738 	read_lock(&et->lock);
739 	if (et->count) {
740 		struct extent_node *en;
741 
742 		if (et->cached_en) {
743 			en = et->cached_en;
744 		} else {
745 			struct rb_node *node = rb_first(&et->root);
746 
747 			if (!node)
748 				node = rb_last(&et->root);
749 			en = rb_entry(node, struct extent_node, rb_node);
750 		}
751 
752 		if (__is_extent_same(ext, &en->ei))
753 			goto out;
754 
755 		*ext = en->ei;
756 		sync = true;
757 	} else if (ext->len) {
758 		ext->len = 0;
759 		sync = true;
760 	}
761 out:
762 	read_unlock(&et->lock);
763 	atomic_dec(&et->refcount);
764 
765 	if (sync)
766 		update_inode_page(inode);
767 }
768 
769 void f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
770 {
771 	struct extent_tree *treevec[EXT_TREE_VEC_SIZE];
772 	struct extent_node *en, *tmp;
773 	unsigned long ino = F2FS_ROOT_INO(sbi);
774 	struct radix_tree_iter iter;
775 	void **slot;
776 	unsigned int found;
777 	unsigned int node_cnt = 0, tree_cnt = 0;
778 
779 	if (!test_opt(sbi, EXTENT_CACHE))
780 		return;
781 
782 	if (available_free_memory(sbi, EXTENT_CACHE))
783 		return;
784 
785 	spin_lock(&sbi->extent_lock);
786 	list_for_each_entry_safe(en, tmp, &sbi->extent_list, list) {
787 		if (!nr_shrink--)
788 			break;
789 		list_del_init(&en->list);
790 	}
791 	spin_unlock(&sbi->extent_lock);
792 
793 	down_read(&sbi->extent_tree_lock);
794 	while ((found = radix_tree_gang_lookup(&sbi->extent_tree_root,
795 				(void **)treevec, ino, EXT_TREE_VEC_SIZE))) {
796 		unsigned i;
797 
798 		ino = treevec[found - 1]->ino + 1;
799 		for (i = 0; i < found; i++) {
800 			struct extent_tree *et = treevec[i];
801 
802 			atomic_inc(&et->refcount);
803 			write_lock(&et->lock);
804 			node_cnt += __free_extent_tree(sbi, et, false);
805 			write_unlock(&et->lock);
806 			atomic_dec(&et->refcount);
807 		}
808 	}
809 	up_read(&sbi->extent_tree_lock);
810 
811 	down_write(&sbi->extent_tree_lock);
812 	radix_tree_for_each_slot(slot, &sbi->extent_tree_root, &iter,
813 							F2FS_ROOT_INO(sbi)) {
814 		struct extent_tree *et = (struct extent_tree *)*slot;
815 
816 		if (!atomic_read(&et->refcount) && !et->count) {
817 			radix_tree_delete(&sbi->extent_tree_root, et->ino);
818 			kmem_cache_free(extent_tree_slab, et);
819 			sbi->total_ext_tree--;
820 			tree_cnt++;
821 		}
822 	}
823 	up_write(&sbi->extent_tree_lock);
824 
825 	trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
826 }
827 
828 void f2fs_destroy_extent_tree(struct inode *inode)
829 {
830 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
831 	struct extent_tree *et;
832 	unsigned int node_cnt = 0;
833 
834 	if (!test_opt(sbi, EXTENT_CACHE))
835 		return;
836 
837 	et = __find_extent_tree(sbi, inode->i_ino);
838 	if (!et)
839 		goto out;
840 
841 	/* free all extent info belong to this extent tree */
842 	write_lock(&et->lock);
843 	node_cnt = __free_extent_tree(sbi, et, true);
844 	write_unlock(&et->lock);
845 
846 	atomic_dec(&et->refcount);
847 
848 	/* try to find and delete extent tree entry in radix tree */
849 	down_write(&sbi->extent_tree_lock);
850 	et = radix_tree_lookup(&sbi->extent_tree_root, inode->i_ino);
851 	if (!et) {
852 		up_write(&sbi->extent_tree_lock);
853 		goto out;
854 	}
855 	f2fs_bug_on(sbi, atomic_read(&et->refcount) || et->count);
856 	radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
857 	kmem_cache_free(extent_tree_slab, et);
858 	sbi->total_ext_tree--;
859 	up_write(&sbi->extent_tree_lock);
860 out:
861 	trace_f2fs_destroy_extent_tree(inode, node_cnt);
862 	return;
863 }
864 
865 void f2fs_init_extent_cache(struct inode *inode, struct f2fs_extent *i_ext)
866 {
867 	if (test_opt(F2FS_I_SB(inode), EXTENT_CACHE))
868 		f2fs_init_extent_tree(inode, i_ext);
869 
870 	write_lock(&F2FS_I(inode)->ext_lock);
871 	get_extent_info(&F2FS_I(inode)->ext, *i_ext);
872 	write_unlock(&F2FS_I(inode)->ext_lock);
873 }
874 
875 static bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
876 							struct extent_info *ei)
877 {
878 	if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT))
879 		return false;
880 
881 	if (test_opt(F2FS_I_SB(inode), EXTENT_CACHE))
882 		return f2fs_lookup_extent_tree(inode, pgofs, ei);
883 
884 	return lookup_extent_info(inode, pgofs, ei);
885 }
886 
887 void f2fs_update_extent_cache(struct dnode_of_data *dn)
888 {
889 	struct f2fs_inode_info *fi = F2FS_I(dn->inode);
890 	pgoff_t fofs;
891 
892 	f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR);
893 
894 	if (is_inode_flag_set(fi, FI_NO_EXTENT))
895 		return;
896 
897 	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
898 							dn->ofs_in_node;
899 
900 	if (test_opt(F2FS_I_SB(dn->inode), EXTENT_CACHE))
901 		return f2fs_update_extent_tree(dn->inode, fofs,
902 							dn->data_blkaddr);
903 
904 	if (update_extent_info(dn->inode, fofs, dn->data_blkaddr))
905 		sync_inode_page(dn);
906 }
907 
908 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
909 {
910 	struct address_space *mapping = inode->i_mapping;
911 	struct dnode_of_data dn;
912 	struct page *page;
913 	struct extent_info ei;
914 	int err;
915 	struct f2fs_io_info fio = {
916 		.type = DATA,
917 		.rw = sync ? READ_SYNC : READA,
918 	};
919 
920 	/*
921 	 * If sync is false, it needs to check its block allocation.
922 	 * This is need and triggered by two flows:
923 	 *   gc and truncate_partial_data_page.
924 	 */
925 	if (!sync)
926 		goto search;
927 
928 	page = find_get_page(mapping, index);
929 	if (page && PageUptodate(page))
930 		return page;
931 	f2fs_put_page(page, 0);
932 search:
933 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
934 		dn.data_blkaddr = ei.blk + index - ei.fofs;
935 		goto got_it;
936 	}
937 
938 	set_new_dnode(&dn, inode, NULL, NULL, 0);
939 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
940 	if (err)
941 		return ERR_PTR(err);
942 	f2fs_put_dnode(&dn);
943 
944 	if (dn.data_blkaddr == NULL_ADDR)
945 		return ERR_PTR(-ENOENT);
946 
947 	/* By fallocate(), there is no cached page, but with NEW_ADDR */
948 	if (unlikely(dn.data_blkaddr == NEW_ADDR))
949 		return ERR_PTR(-EINVAL);
950 
951 got_it:
952 	page = grab_cache_page(mapping, index);
953 	if (!page)
954 		return ERR_PTR(-ENOMEM);
955 
956 	if (PageUptodate(page)) {
957 		unlock_page(page);
958 		return page;
959 	}
960 
961 	fio.blk_addr = dn.data_blkaddr;
962 	err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
963 	if (err)
964 		return ERR_PTR(err);
965 
966 	if (sync) {
967 		wait_on_page_locked(page);
968 		if (unlikely(!PageUptodate(page))) {
969 			f2fs_put_page(page, 0);
970 			return ERR_PTR(-EIO);
971 		}
972 	}
973 	return page;
974 }
975 
976 /*
977  * If it tries to access a hole, return an error.
978  * Because, the callers, functions in dir.c and GC, should be able to know
979  * whether this page exists or not.
980  */
981 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
982 {
983 	struct address_space *mapping = inode->i_mapping;
984 	struct dnode_of_data dn;
985 	struct page *page;
986 	struct extent_info ei;
987 	int err;
988 	struct f2fs_io_info fio = {
989 		.type = DATA,
990 		.rw = READ_SYNC,
991 	};
992 repeat:
993 	page = grab_cache_page(mapping, index);
994 	if (!page)
995 		return ERR_PTR(-ENOMEM);
996 
997 	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
998 		dn.data_blkaddr = ei.blk + index - ei.fofs;
999 		goto got_it;
1000 	}
1001 
1002 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1003 	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1004 	if (err) {
1005 		f2fs_put_page(page, 1);
1006 		return ERR_PTR(err);
1007 	}
1008 	f2fs_put_dnode(&dn);
1009 
1010 	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1011 		f2fs_put_page(page, 1);
1012 		return ERR_PTR(-ENOENT);
1013 	}
1014 
1015 got_it:
1016 	if (PageUptodate(page))
1017 		return page;
1018 
1019 	/*
1020 	 * A new dentry page is allocated but not able to be written, since its
1021 	 * new inode page couldn't be allocated due to -ENOSPC.
1022 	 * In such the case, its blkaddr can be remained as NEW_ADDR.
1023 	 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
1024 	 */
1025 	if (dn.data_blkaddr == NEW_ADDR) {
1026 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1027 		SetPageUptodate(page);
1028 		return page;
1029 	}
1030 
1031 	fio.blk_addr = dn.data_blkaddr;
1032 	err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
1033 	if (err)
1034 		return ERR_PTR(err);
1035 
1036 	lock_page(page);
1037 	if (unlikely(!PageUptodate(page))) {
1038 		f2fs_put_page(page, 1);
1039 		return ERR_PTR(-EIO);
1040 	}
1041 	if (unlikely(page->mapping != mapping)) {
1042 		f2fs_put_page(page, 1);
1043 		goto repeat;
1044 	}
1045 	return page;
1046 }
1047 
1048 /*
1049  * Caller ensures that this data page is never allocated.
1050  * A new zero-filled data page is allocated in the page cache.
1051  *
1052  * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1053  * f2fs_unlock_op().
1054  * Note that, ipage is set only by make_empty_dir.
1055  */
1056 struct page *get_new_data_page(struct inode *inode,
1057 		struct page *ipage, pgoff_t index, bool new_i_size)
1058 {
1059 	struct address_space *mapping = inode->i_mapping;
1060 	struct page *page;
1061 	struct dnode_of_data dn;
1062 	int err;
1063 
1064 	set_new_dnode(&dn, inode, ipage, NULL, 0);
1065 	err = f2fs_reserve_block(&dn, index);
1066 	if (err)
1067 		return ERR_PTR(err);
1068 repeat:
1069 	page = grab_cache_page(mapping, index);
1070 	if (!page) {
1071 		err = -ENOMEM;
1072 		goto put_err;
1073 	}
1074 
1075 	if (PageUptodate(page))
1076 		return page;
1077 
1078 	if (dn.data_blkaddr == NEW_ADDR) {
1079 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1080 		SetPageUptodate(page);
1081 	} else {
1082 		struct f2fs_io_info fio = {
1083 			.type = DATA,
1084 			.rw = READ_SYNC,
1085 			.blk_addr = dn.data_blkaddr,
1086 		};
1087 		err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
1088 		if (err)
1089 			goto put_err;
1090 
1091 		lock_page(page);
1092 		if (unlikely(!PageUptodate(page))) {
1093 			f2fs_put_page(page, 1);
1094 			err = -EIO;
1095 			goto put_err;
1096 		}
1097 		if (unlikely(page->mapping != mapping)) {
1098 			f2fs_put_page(page, 1);
1099 			goto repeat;
1100 		}
1101 	}
1102 
1103 	if (new_i_size &&
1104 		i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
1105 		i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
1106 		/* Only the directory inode sets new_i_size */
1107 		set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
1108 	}
1109 	return page;
1110 
1111 put_err:
1112 	f2fs_put_dnode(&dn);
1113 	return ERR_PTR(err);
1114 }
1115 
1116 static int __allocate_data_block(struct dnode_of_data *dn)
1117 {
1118 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1119 	struct f2fs_inode_info *fi = F2FS_I(dn->inode);
1120 	struct f2fs_summary sum;
1121 	struct node_info ni;
1122 	int seg = CURSEG_WARM_DATA;
1123 	pgoff_t fofs;
1124 
1125 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
1126 		return -EPERM;
1127 
1128 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
1129 	if (dn->data_blkaddr == NEW_ADDR)
1130 		goto alloc;
1131 
1132 	if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
1133 		return -ENOSPC;
1134 
1135 alloc:
1136 	get_node_info(sbi, dn->nid, &ni);
1137 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1138 
1139 	if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
1140 		seg = CURSEG_DIRECT_IO;
1141 
1142 	allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
1143 								&sum, seg);
1144 
1145 	/* direct IO doesn't use extent cache to maximize the performance */
1146 	set_data_blkaddr(dn);
1147 
1148 	/* update i_size */
1149 	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
1150 							dn->ofs_in_node;
1151 	if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
1152 		i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
1153 
1154 	return 0;
1155 }
1156 
1157 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
1158 							size_t count)
1159 {
1160 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1161 	struct dnode_of_data dn;
1162 	u64 start = F2FS_BYTES_TO_BLK(offset);
1163 	u64 len = F2FS_BYTES_TO_BLK(count);
1164 	bool allocated;
1165 	u64 end_offset;
1166 
1167 	while (len) {
1168 		f2fs_balance_fs(sbi);
1169 		f2fs_lock_op(sbi);
1170 
1171 		/* When reading holes, we need its node page */
1172 		set_new_dnode(&dn, inode, NULL, NULL, 0);
1173 		if (get_dnode_of_data(&dn, start, ALLOC_NODE))
1174 			goto out;
1175 
1176 		allocated = false;
1177 		end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1178 
1179 		while (dn.ofs_in_node < end_offset && len) {
1180 			block_t blkaddr;
1181 
1182 			blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
1183 			if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
1184 				if (__allocate_data_block(&dn))
1185 					goto sync_out;
1186 				allocated = true;
1187 			}
1188 			len--;
1189 			start++;
1190 			dn.ofs_in_node++;
1191 		}
1192 
1193 		if (allocated)
1194 			sync_inode_page(&dn);
1195 
1196 		f2fs_put_dnode(&dn);
1197 		f2fs_unlock_op(sbi);
1198 	}
1199 	return;
1200 
1201 sync_out:
1202 	if (allocated)
1203 		sync_inode_page(&dn);
1204 	f2fs_put_dnode(&dn);
1205 out:
1206 	f2fs_unlock_op(sbi);
1207 	return;
1208 }
1209 
1210 /*
1211  * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
1212  * If original data blocks are allocated, then give them to blockdev.
1213  * Otherwise,
1214  *     a. preallocate requested block addresses
1215  *     b. do not use extent cache for better performance
1216  *     c. give the block addresses to blockdev
1217  */
1218 static int __get_data_block(struct inode *inode, sector_t iblock,
1219 			struct buffer_head *bh_result, int create, bool fiemap)
1220 {
1221 	unsigned int blkbits = inode->i_sb->s_blocksize_bits;
1222 	unsigned maxblocks = bh_result->b_size >> blkbits;
1223 	struct dnode_of_data dn;
1224 	int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
1225 	pgoff_t pgofs, end_offset;
1226 	int err = 0, ofs = 1;
1227 	struct extent_info ei;
1228 	bool allocated = false;
1229 
1230 	/* Get the page offset from the block offset(iblock) */
1231 	pgofs =	(pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
1232 
1233 	if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1234 		f2fs_map_bh(inode->i_sb, pgofs, &ei, bh_result);
1235 		goto out;
1236 	}
1237 
1238 	if (create)
1239 		f2fs_lock_op(F2FS_I_SB(inode));
1240 
1241 	/* When reading holes, we need its node page */
1242 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1243 	err = get_dnode_of_data(&dn, pgofs, mode);
1244 	if (err) {
1245 		if (err == -ENOENT)
1246 			err = 0;
1247 		goto unlock_out;
1248 	}
1249 	if (dn.data_blkaddr == NEW_ADDR && !fiemap)
1250 		goto put_out;
1251 
1252 	if (dn.data_blkaddr != NULL_ADDR) {
1253 		clear_buffer_new(bh_result);
1254 		map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
1255 	} else if (create) {
1256 		err = __allocate_data_block(&dn);
1257 		if (err)
1258 			goto put_out;
1259 		allocated = true;
1260 		set_buffer_new(bh_result);
1261 		map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
1262 	} else {
1263 		goto put_out;
1264 	}
1265 
1266 	end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1267 	bh_result->b_size = (((size_t)1) << blkbits);
1268 	dn.ofs_in_node++;
1269 	pgofs++;
1270 
1271 get_next:
1272 	if (dn.ofs_in_node >= end_offset) {
1273 		if (allocated)
1274 			sync_inode_page(&dn);
1275 		allocated = false;
1276 		f2fs_put_dnode(&dn);
1277 
1278 		set_new_dnode(&dn, inode, NULL, NULL, 0);
1279 		err = get_dnode_of_data(&dn, pgofs, mode);
1280 		if (err) {
1281 			if (err == -ENOENT)
1282 				err = 0;
1283 			goto unlock_out;
1284 		}
1285 		if (dn.data_blkaddr == NEW_ADDR && !fiemap)
1286 			goto put_out;
1287 
1288 		end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1289 	}
1290 
1291 	if (maxblocks > (bh_result->b_size >> blkbits)) {
1292 		block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
1293 		if (blkaddr == NULL_ADDR && create) {
1294 			err = __allocate_data_block(&dn);
1295 			if (err)
1296 				goto sync_out;
1297 			allocated = true;
1298 			set_buffer_new(bh_result);
1299 			blkaddr = dn.data_blkaddr;
1300 		}
1301 		/* Give more consecutive addresses for the readahead */
1302 		if (blkaddr == (bh_result->b_blocknr + ofs)) {
1303 			ofs++;
1304 			dn.ofs_in_node++;
1305 			pgofs++;
1306 			bh_result->b_size += (((size_t)1) << blkbits);
1307 			goto get_next;
1308 		}
1309 	}
1310 sync_out:
1311 	if (allocated)
1312 		sync_inode_page(&dn);
1313 put_out:
1314 	f2fs_put_dnode(&dn);
1315 unlock_out:
1316 	if (create)
1317 		f2fs_unlock_op(F2FS_I_SB(inode));
1318 out:
1319 	trace_f2fs_get_data_block(inode, iblock, bh_result, err);
1320 	return err;
1321 }
1322 
1323 static int get_data_block(struct inode *inode, sector_t iblock,
1324 			struct buffer_head *bh_result, int create)
1325 {
1326 	return __get_data_block(inode, iblock, bh_result, create, false);
1327 }
1328 
1329 static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
1330 			struct buffer_head *bh_result, int create)
1331 {
1332 	return __get_data_block(inode, iblock, bh_result, create, true);
1333 }
1334 
1335 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1336 		u64 start, u64 len)
1337 {
1338 	return generic_block_fiemap(inode, fieinfo,
1339 				start, len, get_data_block_fiemap);
1340 }
1341 
1342 static int f2fs_read_data_page(struct file *file, struct page *page)
1343 {
1344 	struct inode *inode = page->mapping->host;
1345 	int ret = -EAGAIN;
1346 
1347 	trace_f2fs_readpage(page, DATA);
1348 
1349 	/* If the file has inline data, try to read it directly */
1350 	if (f2fs_has_inline_data(inode))
1351 		ret = f2fs_read_inline_data(inode, page);
1352 	if (ret == -EAGAIN)
1353 		ret = mpage_readpage(page, get_data_block);
1354 
1355 	return ret;
1356 }
1357 
1358 static int f2fs_read_data_pages(struct file *file,
1359 			struct address_space *mapping,
1360 			struct list_head *pages, unsigned nr_pages)
1361 {
1362 	struct inode *inode = file->f_mapping->host;
1363 
1364 	/* If the file has inline data, skip readpages */
1365 	if (f2fs_has_inline_data(inode))
1366 		return 0;
1367 
1368 	return mpage_readpages(mapping, pages, nr_pages, get_data_block);
1369 }
1370 
1371 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
1372 {
1373 	struct inode *inode = page->mapping->host;
1374 	struct dnode_of_data dn;
1375 	int err = 0;
1376 
1377 	set_new_dnode(&dn, inode, NULL, NULL, 0);
1378 	err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1379 	if (err)
1380 		return err;
1381 
1382 	fio->blk_addr = dn.data_blkaddr;
1383 
1384 	/* This page is already truncated */
1385 	if (fio->blk_addr == NULL_ADDR) {
1386 		ClearPageUptodate(page);
1387 		goto out_writepage;
1388 	}
1389 
1390 	set_page_writeback(page);
1391 
1392 	/*
1393 	 * If current allocation needs SSR,
1394 	 * it had better in-place writes for updated data.
1395 	 */
1396 	if (unlikely(fio->blk_addr != NEW_ADDR &&
1397 			!is_cold_data(page) &&
1398 			need_inplace_update(inode))) {
1399 		rewrite_data_page(page, fio);
1400 		set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1401 		trace_f2fs_do_write_data_page(page, IPU);
1402 	} else {
1403 		write_data_page(page, &dn, fio);
1404 		set_data_blkaddr(&dn);
1405 		f2fs_update_extent_cache(&dn);
1406 		trace_f2fs_do_write_data_page(page, OPU);
1407 		set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1408 		if (page->index == 0)
1409 			set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1410 	}
1411 out_writepage:
1412 	f2fs_put_dnode(&dn);
1413 	return err;
1414 }
1415 
1416 static int f2fs_write_data_page(struct page *page,
1417 					struct writeback_control *wbc)
1418 {
1419 	struct inode *inode = page->mapping->host;
1420 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1421 	loff_t i_size = i_size_read(inode);
1422 	const pgoff_t end_index = ((unsigned long long) i_size)
1423 							>> PAGE_CACHE_SHIFT;
1424 	unsigned offset = 0;
1425 	bool need_balance_fs = false;
1426 	int err = 0;
1427 	struct f2fs_io_info fio = {
1428 		.type = DATA,
1429 		.rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1430 	};
1431 
1432 	trace_f2fs_writepage(page, DATA);
1433 
1434 	if (page->index < end_index)
1435 		goto write;
1436 
1437 	/*
1438 	 * If the offset is out-of-range of file size,
1439 	 * this page does not have to be written to disk.
1440 	 */
1441 	offset = i_size & (PAGE_CACHE_SIZE - 1);
1442 	if ((page->index >= end_index + 1) || !offset)
1443 		goto out;
1444 
1445 	zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1446 write:
1447 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1448 		goto redirty_out;
1449 	if (f2fs_is_drop_cache(inode))
1450 		goto out;
1451 	if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1452 			available_free_memory(sbi, BASE_CHECK))
1453 		goto redirty_out;
1454 
1455 	/* Dentry blocks are controlled by checkpoint */
1456 	if (S_ISDIR(inode->i_mode)) {
1457 		if (unlikely(f2fs_cp_error(sbi)))
1458 			goto redirty_out;
1459 		err = do_write_data_page(page, &fio);
1460 		goto done;
1461 	}
1462 
1463 	/* we should bypass data pages to proceed the kworkder jobs */
1464 	if (unlikely(f2fs_cp_error(sbi))) {
1465 		SetPageError(page);
1466 		goto out;
1467 	}
1468 
1469 	if (!wbc->for_reclaim)
1470 		need_balance_fs = true;
1471 	else if (has_not_enough_free_secs(sbi, 0))
1472 		goto redirty_out;
1473 
1474 	err = -EAGAIN;
1475 	f2fs_lock_op(sbi);
1476 	if (f2fs_has_inline_data(inode))
1477 		err = f2fs_write_inline_data(inode, page);
1478 	if (err == -EAGAIN)
1479 		err = do_write_data_page(page, &fio);
1480 	f2fs_unlock_op(sbi);
1481 done:
1482 	if (err && err != -ENOENT)
1483 		goto redirty_out;
1484 
1485 	clear_cold_data(page);
1486 out:
1487 	inode_dec_dirty_pages(inode);
1488 	if (err)
1489 		ClearPageUptodate(page);
1490 	unlock_page(page);
1491 	if (need_balance_fs)
1492 		f2fs_balance_fs(sbi);
1493 	if (wbc->for_reclaim)
1494 		f2fs_submit_merged_bio(sbi, DATA, WRITE);
1495 	return 0;
1496 
1497 redirty_out:
1498 	redirty_page_for_writepage(wbc, page);
1499 	return AOP_WRITEPAGE_ACTIVATE;
1500 }
1501 
1502 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1503 			void *data)
1504 {
1505 	struct address_space *mapping = data;
1506 	int ret = mapping->a_ops->writepage(page, wbc);
1507 	mapping_set_error(mapping, ret);
1508 	return ret;
1509 }
1510 
1511 static int f2fs_write_data_pages(struct address_space *mapping,
1512 			    struct writeback_control *wbc)
1513 {
1514 	struct inode *inode = mapping->host;
1515 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1516 	int ret;
1517 	long diff;
1518 
1519 	trace_f2fs_writepages(mapping->host, wbc, DATA);
1520 
1521 	/* deal with chardevs and other special file */
1522 	if (!mapping->a_ops->writepage)
1523 		return 0;
1524 
1525 	if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1526 			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1527 			available_free_memory(sbi, DIRTY_DENTS))
1528 		goto skip_write;
1529 
1530 	/* during POR, we don't need to trigger writepage at all. */
1531 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1532 		goto skip_write;
1533 
1534 	diff = nr_pages_to_write(sbi, DATA, wbc);
1535 
1536 	ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1537 
1538 	f2fs_submit_merged_bio(sbi, DATA, WRITE);
1539 
1540 	remove_dirty_dir_inode(inode);
1541 
1542 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1543 	return ret;
1544 
1545 skip_write:
1546 	wbc->pages_skipped += get_dirty_pages(inode);
1547 	return 0;
1548 }
1549 
1550 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1551 {
1552 	struct inode *inode = mapping->host;
1553 
1554 	if (to > inode->i_size) {
1555 		truncate_pagecache(inode, inode->i_size);
1556 		truncate_blocks(inode, inode->i_size, true);
1557 	}
1558 }
1559 
1560 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1561 		loff_t pos, unsigned len, unsigned flags,
1562 		struct page **pagep, void **fsdata)
1563 {
1564 	struct inode *inode = mapping->host;
1565 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1566 	struct page *page, *ipage;
1567 	pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1568 	struct dnode_of_data dn;
1569 	int err = 0;
1570 
1571 	trace_f2fs_write_begin(inode, pos, len, flags);
1572 
1573 	f2fs_balance_fs(sbi);
1574 
1575 	/*
1576 	 * We should check this at this moment to avoid deadlock on inode page
1577 	 * and #0 page. The locking rule for inline_data conversion should be:
1578 	 * lock_page(page #0) -> lock_page(inode_page)
1579 	 */
1580 	if (index != 0) {
1581 		err = f2fs_convert_inline_inode(inode);
1582 		if (err)
1583 			goto fail;
1584 	}
1585 repeat:
1586 	page = grab_cache_page_write_begin(mapping, index, flags);
1587 	if (!page) {
1588 		err = -ENOMEM;
1589 		goto fail;
1590 	}
1591 
1592 	*pagep = page;
1593 
1594 	f2fs_lock_op(sbi);
1595 
1596 	/* check inline_data */
1597 	ipage = get_node_page(sbi, inode->i_ino);
1598 	if (IS_ERR(ipage)) {
1599 		err = PTR_ERR(ipage);
1600 		goto unlock_fail;
1601 	}
1602 
1603 	set_new_dnode(&dn, inode, ipage, ipage, 0);
1604 
1605 	if (f2fs_has_inline_data(inode)) {
1606 		if (pos + len <= MAX_INLINE_DATA) {
1607 			read_inline_data(page, ipage);
1608 			set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1609 			sync_inode_page(&dn);
1610 			goto put_next;
1611 		}
1612 		err = f2fs_convert_inline_page(&dn, page);
1613 		if (err)
1614 			goto put_fail;
1615 	}
1616 	err = f2fs_reserve_block(&dn, index);
1617 	if (err)
1618 		goto put_fail;
1619 put_next:
1620 	f2fs_put_dnode(&dn);
1621 	f2fs_unlock_op(sbi);
1622 
1623 	if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
1624 		return 0;
1625 
1626 	f2fs_wait_on_page_writeback(page, DATA);
1627 
1628 	if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1629 		unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1630 		unsigned end = start + len;
1631 
1632 		/* Reading beyond i_size is simple: memset to zero */
1633 		zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1634 		goto out;
1635 	}
1636 
1637 	if (dn.data_blkaddr == NEW_ADDR) {
1638 		zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1639 	} else {
1640 		struct f2fs_io_info fio = {
1641 			.type = DATA,
1642 			.rw = READ_SYNC,
1643 			.blk_addr = dn.data_blkaddr,
1644 		};
1645 		err = f2fs_submit_page_bio(sbi, page, &fio);
1646 		if (err)
1647 			goto fail;
1648 
1649 		lock_page(page);
1650 		if (unlikely(!PageUptodate(page))) {
1651 			f2fs_put_page(page, 1);
1652 			err = -EIO;
1653 			goto fail;
1654 		}
1655 		if (unlikely(page->mapping != mapping)) {
1656 			f2fs_put_page(page, 1);
1657 			goto repeat;
1658 		}
1659 	}
1660 out:
1661 	SetPageUptodate(page);
1662 	clear_cold_data(page);
1663 	return 0;
1664 
1665 put_fail:
1666 	f2fs_put_dnode(&dn);
1667 unlock_fail:
1668 	f2fs_unlock_op(sbi);
1669 	f2fs_put_page(page, 1);
1670 fail:
1671 	f2fs_write_failed(mapping, pos + len);
1672 	return err;
1673 }
1674 
1675 static int f2fs_write_end(struct file *file,
1676 			struct address_space *mapping,
1677 			loff_t pos, unsigned len, unsigned copied,
1678 			struct page *page, void *fsdata)
1679 {
1680 	struct inode *inode = page->mapping->host;
1681 
1682 	trace_f2fs_write_end(inode, pos, len, copied);
1683 
1684 	set_page_dirty(page);
1685 
1686 	if (pos + copied > i_size_read(inode)) {
1687 		i_size_write(inode, pos + copied);
1688 		mark_inode_dirty(inode);
1689 		update_inode_page(inode);
1690 	}
1691 
1692 	f2fs_put_page(page, 1);
1693 	return copied;
1694 }
1695 
1696 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1697 			   loff_t offset)
1698 {
1699 	unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1700 
1701 	if (iov_iter_rw(iter) == READ)
1702 		return 0;
1703 
1704 	if (offset & blocksize_mask)
1705 		return -EINVAL;
1706 
1707 	if (iov_iter_alignment(iter) & blocksize_mask)
1708 		return -EINVAL;
1709 
1710 	return 0;
1711 }
1712 
1713 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1714 			      loff_t offset)
1715 {
1716 	struct file *file = iocb->ki_filp;
1717 	struct address_space *mapping = file->f_mapping;
1718 	struct inode *inode = mapping->host;
1719 	size_t count = iov_iter_count(iter);
1720 	int err;
1721 
1722 	/* we don't need to use inline_data strictly */
1723 	if (f2fs_has_inline_data(inode)) {
1724 		err = f2fs_convert_inline_inode(inode);
1725 		if (err)
1726 			return err;
1727 	}
1728 
1729 	if (check_direct_IO(inode, iter, offset))
1730 		return 0;
1731 
1732 	trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1733 
1734 	if (iov_iter_rw(iter) == WRITE)
1735 		__allocate_data_blocks(inode, offset, count);
1736 
1737 	err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block);
1738 	if (err < 0 && iov_iter_rw(iter) == WRITE)
1739 		f2fs_write_failed(mapping, offset + count);
1740 
1741 	trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1742 
1743 	return err;
1744 }
1745 
1746 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1747 							unsigned int length)
1748 {
1749 	struct inode *inode = page->mapping->host;
1750 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1751 
1752 	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1753 		(offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1754 		return;
1755 
1756 	if (PageDirty(page)) {
1757 		if (inode->i_ino == F2FS_META_INO(sbi))
1758 			dec_page_count(sbi, F2FS_DIRTY_META);
1759 		else if (inode->i_ino == F2FS_NODE_INO(sbi))
1760 			dec_page_count(sbi, F2FS_DIRTY_NODES);
1761 		else
1762 			inode_dec_dirty_pages(inode);
1763 	}
1764 	ClearPagePrivate(page);
1765 }
1766 
1767 int f2fs_release_page(struct page *page, gfp_t wait)
1768 {
1769 	/* If this is dirty page, keep PagePrivate */
1770 	if (PageDirty(page))
1771 		return 0;
1772 
1773 	ClearPagePrivate(page);
1774 	return 1;
1775 }
1776 
1777 static int f2fs_set_data_page_dirty(struct page *page)
1778 {
1779 	struct address_space *mapping = page->mapping;
1780 	struct inode *inode = mapping->host;
1781 
1782 	trace_f2fs_set_page_dirty(page, DATA);
1783 
1784 	SetPageUptodate(page);
1785 
1786 	if (f2fs_is_atomic_file(inode)) {
1787 		register_inmem_page(inode, page);
1788 		return 1;
1789 	}
1790 
1791 	mark_inode_dirty(inode);
1792 
1793 	if (!PageDirty(page)) {
1794 		__set_page_dirty_nobuffers(page);
1795 		update_dirty_page(inode, page);
1796 		return 1;
1797 	}
1798 	return 0;
1799 }
1800 
1801 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1802 {
1803 	struct inode *inode = mapping->host;
1804 
1805 	/* we don't need to use inline_data strictly */
1806 	if (f2fs_has_inline_data(inode)) {
1807 		int err = f2fs_convert_inline_inode(inode);
1808 		if (err)
1809 			return err;
1810 	}
1811 	return generic_block_bmap(mapping, block, get_data_block);
1812 }
1813 
1814 void init_extent_cache_info(struct f2fs_sb_info *sbi)
1815 {
1816 	INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
1817 	init_rwsem(&sbi->extent_tree_lock);
1818 	INIT_LIST_HEAD(&sbi->extent_list);
1819 	spin_lock_init(&sbi->extent_lock);
1820 	sbi->total_ext_tree = 0;
1821 	atomic_set(&sbi->total_ext_node, 0);
1822 }
1823 
1824 int __init create_extent_cache(void)
1825 {
1826 	extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
1827 			sizeof(struct extent_tree));
1828 	if (!extent_tree_slab)
1829 		return -ENOMEM;
1830 	extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
1831 			sizeof(struct extent_node));
1832 	if (!extent_node_slab) {
1833 		kmem_cache_destroy(extent_tree_slab);
1834 		return -ENOMEM;
1835 	}
1836 	return 0;
1837 }
1838 
1839 void destroy_extent_cache(void)
1840 {
1841 	kmem_cache_destroy(extent_node_slab);
1842 	kmem_cache_destroy(extent_tree_slab);
1843 }
1844 
1845 const struct address_space_operations f2fs_dblock_aops = {
1846 	.readpage	= f2fs_read_data_page,
1847 	.readpages	= f2fs_read_data_pages,
1848 	.writepage	= f2fs_write_data_page,
1849 	.writepages	= f2fs_write_data_pages,
1850 	.write_begin	= f2fs_write_begin,
1851 	.write_end	= f2fs_write_end,
1852 	.set_page_dirty	= f2fs_set_data_page_dirty,
1853 	.invalidatepage	= f2fs_invalidate_page,
1854 	.releasepage	= f2fs_release_page,
1855 	.direct_IO	= f2fs_direct_IO,
1856 	.bmap		= f2fs_bmap,
1857 };
1858