projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
f2fs: introduce dot and dotdot name check
[GitHub/exynos8895/android_kernel_samsung_universal8895.git] / fs / f2fs / data.c
... / ...
CommitLineData
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#include <linux/cleancache.h>
22
23#include "f2fs.h"
24#include "node.h"
25#include "segment.h"
26#include "trace.h"
27#include <trace/events/f2fs.h>
28
29static struct kmem_cache *extent_tree_slab;
30static struct kmem_cache *extent_node_slab;
31
32static void f2fs_read_end_io(struct bio *bio, int err)
33{
34 struct bio_vec *bvec;
35 int i;
36
37 bio_for_each_segment_all(bvec, bio, i) {
38 struct page *page = bvec->bv_page;
39
40 if (!err) {
41 SetPageUptodate(page);
42 } else {
43 ClearPageUptodate(page);
44 SetPageError(page);
45 }
46 unlock_page(page);
47 }
48 bio_put(bio);
49}
50
51/*
52 * I/O completion handler for multipage BIOs.
53 * copied from fs/mpage.c
54 */
55static void mpage_end_io(struct bio *bio, int err)
56{
57 struct bio_vec *bv;
58 int i;
59
60 bio_for_each_segment_all(bv, bio, i) {
61 struct page *page = bv->bv_page;
62
63 if (!err) {
64 SetPageUptodate(page);
65 } else {
66 ClearPageUptodate(page);
67 SetPageError(page);
68 }
69 unlock_page(page);
70 }
71
72 bio_put(bio);
73}
74
75static void f2fs_write_end_io(struct bio *bio, int err)
76{
77 struct f2fs_sb_info *sbi = bio->bi_private;
78 struct bio_vec *bvec;
79 int i;
80
81 bio_for_each_segment_all(bvec, bio, i) {
82 struct page *page = bvec->bv_page;
83
84 if (unlikely(err)) {
85 set_page_dirty(page);
86 set_bit(AS_EIO, &page->mapping->flags);
87 f2fs_stop_checkpoint(sbi);
88 }
89 end_page_writeback(page);
90 dec_page_count(sbi, F2FS_WRITEBACK);
91 }
92
93 if (!get_pages(sbi, F2FS_WRITEBACK) &&
94 !list_empty(&sbi->cp_wait.task_list))
95 wake_up(&sbi->cp_wait);
96
97 bio_put(bio);
98}
99
100/*
101 * Low-level block read/write IO operations.
102 */
103static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
104 int npages, bool is_read)
105{
106 struct bio *bio;
107
108 /* No failure on bio allocation */
109 bio = bio_alloc(GFP_NOIO, npages);
110
111 bio->bi_bdev = sbi->sb->s_bdev;
112 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
113 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
114 bio->bi_private = sbi;
115
116 return bio;
117}
118
119static void __submit_merged_bio(struct f2fs_bio_info *io)
120{
121 struct f2fs_io_info *fio = &io->fio;
122
123 if (!io->bio)
124 return;
125
126 if (is_read_io(fio->rw))
127 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
128 else
129 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
130
131 submit_bio(fio->rw, io->bio);
132 io->bio = NULL;
133}
134
135void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
136 enum page_type type, int rw)
137{
138 enum page_type btype = PAGE_TYPE_OF_BIO(type);
139 struct f2fs_bio_info *io;
140
141 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
142
143 down_write(&io->io_rwsem);
144
145 /* change META to META_FLUSH in the checkpoint procedure */
146 if (type >= META_FLUSH) {
147 io->fio.type = META_FLUSH;
148 if (test_opt(sbi, NOBARRIER))
149 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
150 else
151 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
152 }
153 __submit_merged_bio(io);
154 up_write(&io->io_rwsem);
155}
156
157/*
158 * Fill the locked page with data located in the block address.
159 * Return unlocked page.
160 */
161int f2fs_submit_page_bio(struct f2fs_io_info *fio)
162{
163 struct bio *bio;
164 struct page *page = fio->page;
165
166 trace_f2fs_submit_page_bio(page, fio);
167 f2fs_trace_ios(fio, 0);
168
169 /* Allocate a new bio */
170 bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
171
172 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
173 bio_put(bio);
174 f2fs_put_page(page, 1);
175 return -EFAULT;
176 }
177
178 submit_bio(fio->rw, bio);
179 return 0;
180}
181
182void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
183{
184 struct f2fs_sb_info *sbi = fio->sbi;
185 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
186 struct f2fs_bio_info *io;
187 bool is_read = is_read_io(fio->rw);
188
189 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
190
191 verify_block_addr(sbi, fio->blk_addr);
192
193 down_write(&io->io_rwsem);
194
195 if (!is_read)
196 inc_page_count(sbi, F2FS_WRITEBACK);
197
198 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
199 io->fio.rw != fio->rw))
200 __submit_merged_bio(io);
201alloc_new:
202 if (io->bio == NULL) {
203 int bio_blocks = MAX_BIO_BLOCKS(sbi);
204
205 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
206 io->fio = *fio;
207 }
208
209 if (bio_add_page(io->bio, fio->page, PAGE_CACHE_SIZE, 0) <
210 PAGE_CACHE_SIZE) {
211 __submit_merged_bio(io);
212 goto alloc_new;
213 }
214
215 io->last_block_in_bio = fio->blk_addr;
216 f2fs_trace_ios(fio, 0);
217
218 up_write(&io->io_rwsem);
219 trace_f2fs_submit_page_mbio(fio->page, fio);
220}
221
222/*
223 * Lock ordering for the change of data block address:
224 * ->data_page
225 * ->node_page
226 * update block addresses in the node page
227 */
228void set_data_blkaddr(struct dnode_of_data *dn)
229{
230 struct f2fs_node *rn;
231 __le32 *addr_array;
232 struct page *node_page = dn->node_page;
233 unsigned int ofs_in_node = dn->ofs_in_node;
234
235 f2fs_wait_on_page_writeback(node_page, NODE);
236
237 rn = F2FS_NODE(node_page);
238
239 /* Get physical address of data block */
240 addr_array = blkaddr_in_node(rn);
241 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
242 set_page_dirty(node_page);
243}
244
245int reserve_new_block(struct dnode_of_data *dn)
246{
247 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
248
249 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
250 return -EPERM;
251 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
252 return -ENOSPC;
253
254 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
255
256 dn->data_blkaddr = NEW_ADDR;
257 set_data_blkaddr(dn);
258 mark_inode_dirty(dn->inode);
259 sync_inode_page(dn);
260 return 0;
261}
262
263int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
264{
265 bool need_put = dn->inode_page ? false : true;
266 int err;
267
268 err = get_dnode_of_data(dn, index, ALLOC_NODE);
269 if (err)
270 return err;
271
272 if (dn->data_blkaddr == NULL_ADDR)
273 err = reserve_new_block(dn);
274 if (err || need_put)
275 f2fs_put_dnode(dn);
276 return err;
277}
278
279static bool lookup_extent_info(struct inode *inode, pgoff_t pgofs,
280 struct extent_info *ei)
281{
282 struct f2fs_inode_info *fi = F2FS_I(inode);
283 pgoff_t start_fofs, end_fofs;
284 block_t start_blkaddr;
285
286 read_lock(&fi->ext_lock);
287 if (fi->ext.len == 0) {
288 read_unlock(&fi->ext_lock);
289 return false;
290 }
291
292 stat_inc_total_hit(inode->i_sb);
293
294 start_fofs = fi->ext.fofs;
295 end_fofs = fi->ext.fofs + fi->ext.len - 1;
296 start_blkaddr = fi->ext.blk;
297
298 if (pgofs >= start_fofs && pgofs <= end_fofs) {
299 *ei = fi->ext;
300 stat_inc_read_hit(inode->i_sb);
301 read_unlock(&fi->ext_lock);
302 return true;
303 }
304 read_unlock(&fi->ext_lock);
305 return false;
306}
307
308static bool update_extent_info(struct inode *inode, pgoff_t fofs,
309 block_t blkaddr)
310{
311 struct f2fs_inode_info *fi = F2FS_I(inode);
312 pgoff_t start_fofs, end_fofs;
313 block_t start_blkaddr, end_blkaddr;
314 int need_update = true;
315
316 write_lock(&fi->ext_lock);
317
318 start_fofs = fi->ext.fofs;
319 end_fofs = fi->ext.fofs + fi->ext.len - 1;
320 start_blkaddr = fi->ext.blk;
321 end_blkaddr = fi->ext.blk + fi->ext.len - 1;
322
323 /* Drop and initialize the matched extent */
324 if (fi->ext.len == 1 && fofs == start_fofs)
325 fi->ext.len = 0;
326
327 /* Initial extent */
328 if (fi->ext.len == 0) {
329 if (blkaddr != NULL_ADDR) {
330 fi->ext.fofs = fofs;
331 fi->ext.blk = blkaddr;
332 fi->ext.len = 1;
333 }
334 goto end_update;
335 }
336
337 /* Front merge */
338 if (fofs == start_fofs - 1 && blkaddr == start_blkaddr - 1) {
339 fi->ext.fofs--;
340 fi->ext.blk--;
341 fi->ext.len++;
342 goto end_update;
343 }
344
345 /* Back merge */
346 if (fofs == end_fofs + 1 && blkaddr == end_blkaddr + 1) {
347 fi->ext.len++;
348 goto end_update;
349 }
350
351 /* Split the existing extent */
352 if (fi->ext.len > 1 &&
353 fofs >= start_fofs && fofs <= end_fofs) {
354 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
355 fi->ext.len = fofs - start_fofs;
356 } else {
357 fi->ext.fofs = fofs + 1;
358 fi->ext.blk = start_blkaddr + fofs - start_fofs + 1;
359 fi->ext.len -= fofs - start_fofs + 1;
360 }
361 } else {
362 need_update = false;
363 }
364
365 /* Finally, if the extent is very fragmented, let's drop the cache. */
366 if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
367 fi->ext.len = 0;
368 set_inode_flag(fi, FI_NO_EXTENT);
369 need_update = true;
370 }
371end_update:
372 write_unlock(&fi->ext_lock);
373 return need_update;
374}
375
376static struct extent_node *__attach_extent_node(struct f2fs_sb_info *sbi,
377 struct extent_tree *et, struct extent_info *ei,
378 struct rb_node *parent, struct rb_node **p)
379{
380 struct extent_node *en;
381
382 en = kmem_cache_alloc(extent_node_slab, GFP_ATOMIC);
383 if (!en)
384 return NULL;
385
386 en->ei = *ei;
387 INIT_LIST_HEAD(&en->list);
388
389 rb_link_node(&en->rb_node, parent, p);
390 rb_insert_color(&en->rb_node, &et->root);
391 et->count++;
392 atomic_inc(&sbi->total_ext_node);
393 return en;
394}
395
396static void __detach_extent_node(struct f2fs_sb_info *sbi,
397 struct extent_tree *et, struct extent_node *en)
398{
399 rb_erase(&en->rb_node, &et->root);
400 et->count--;
401 atomic_dec(&sbi->total_ext_node);
402
403 if (et->cached_en == en)
404 et->cached_en = NULL;
405}
406
407static struct extent_tree *__find_extent_tree(struct f2fs_sb_info *sbi,
408 nid_t ino)
409{
410 struct extent_tree *et;
411
412 down_read(&sbi->extent_tree_lock);
413 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
414 if (!et) {
415 up_read(&sbi->extent_tree_lock);
416 return NULL;
417 }
418 atomic_inc(&et->refcount);
419 up_read(&sbi->extent_tree_lock);
420
421 return et;
422}
423
424static struct extent_tree *__grab_extent_tree(struct inode *inode)
425{
426 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
427 struct extent_tree *et;
428 nid_t ino = inode->i_ino;
429
430 down_write(&sbi->extent_tree_lock);
431 et = radix_tree_lookup(&sbi->extent_tree_root, ino);
432 if (!et) {
433 et = f2fs_kmem_cache_alloc(extent_tree_slab, GFP_NOFS);
434 f2fs_radix_tree_insert(&sbi->extent_tree_root, ino, et);
435 memset(et, 0, sizeof(struct extent_tree));
436 et->ino = ino;
437 et->root = RB_ROOT;
438 et->cached_en = NULL;
439 rwlock_init(&et->lock);
440 atomic_set(&et->refcount, 0);
441 et->count = 0;
442 sbi->total_ext_tree++;
443 }
444 atomic_inc(&et->refcount);
445 up_write(&sbi->extent_tree_lock);
446
447 return et;
448}
449
450static struct extent_node *__lookup_extent_tree(struct extent_tree *et,
451 unsigned int fofs)
452{
453 struct rb_node *node = et->root.rb_node;
454 struct extent_node *en;
455
456 if (et->cached_en) {
457 struct extent_info *cei = &et->cached_en->ei;
458
459 if (cei->fofs <= fofs && cei->fofs + cei->len > fofs)
460 return et->cached_en;
461 }
462
463 while (node) {
464 en = rb_entry(node, struct extent_node, rb_node);
465
466 if (fofs < en->ei.fofs) {
467 node = node->rb_left;
468 } else if (fofs >= en->ei.fofs + en->ei.len) {
469 node = node->rb_right;
470 } else {
471 et->cached_en = en;
472 return en;
473 }
474 }
475 return NULL;
476}
477
478static struct extent_node *__try_back_merge(struct f2fs_sb_info *sbi,
479 struct extent_tree *et, struct extent_node *en)
480{
481 struct extent_node *prev;
482 struct rb_node *node;
483
484 node = rb_prev(&en->rb_node);
485 if (!node)
486 return NULL;
487
488 prev = rb_entry(node, struct extent_node, rb_node);
489 if (__is_back_mergeable(&en->ei, &prev->ei)) {
490 en->ei.fofs = prev->ei.fofs;
491 en->ei.blk = prev->ei.blk;
492 en->ei.len += prev->ei.len;
493 __detach_extent_node(sbi, et, prev);
494 return prev;
495 }
496 return NULL;
497}
498
499static struct extent_node *__try_front_merge(struct f2fs_sb_info *sbi,
500 struct extent_tree *et, struct extent_node *en)
501{
502 struct extent_node *next;
503 struct rb_node *node;
504
505 node = rb_next(&en->rb_node);
506 if (!node)
507 return NULL;
508
509 next = rb_entry(node, struct extent_node, rb_node);
510 if (__is_front_mergeable(&en->ei, &next->ei)) {
511 en->ei.len += next->ei.len;
512 __detach_extent_node(sbi, et, next);
513 return next;
514 }
515 return NULL;
516}
517
518static struct extent_node *__insert_extent_tree(struct f2fs_sb_info *sbi,
519 struct extent_tree *et, struct extent_info *ei,
520 struct extent_node **den)
521{
522 struct rb_node **p = &et->root.rb_node;
523 struct rb_node *parent = NULL;
524 struct extent_node *en;
525
526 while (*p) {
527 parent = *p;
528 en = rb_entry(parent, struct extent_node, rb_node);
529
530 if (ei->fofs < en->ei.fofs) {
531 if (__is_front_mergeable(ei, &en->ei)) {
532 f2fs_bug_on(sbi, !den);
533 en->ei.fofs = ei->fofs;
534 en->ei.blk = ei->blk;
535 en->ei.len += ei->len;
536 *den = __try_back_merge(sbi, et, en);
537 return en;
538 }
539 p = &(*p)->rb_left;
540 } else if (ei->fofs >= en->ei.fofs + en->ei.len) {
541 if (__is_back_mergeable(ei, &en->ei)) {
542 f2fs_bug_on(sbi, !den);
543 en->ei.len += ei->len;
544 *den = __try_front_merge(sbi, et, en);
545 return en;
546 }
547 p = &(*p)->rb_right;
548 } else {
549 f2fs_bug_on(sbi, 1);
550 }
551 }
552
553 return __attach_extent_node(sbi, et, ei, parent, p);
554}
555
556static unsigned int __free_extent_tree(struct f2fs_sb_info *sbi,
557 struct extent_tree *et, bool free_all)
558{
559 struct rb_node *node, *next;
560 struct extent_node *en;
561 unsigned int count = et->count;
562
563 node = rb_first(&et->root);
564 while (node) {
565 next = rb_next(node);
566 en = rb_entry(node, struct extent_node, rb_node);
567
568 if (free_all) {
569 spin_lock(&sbi->extent_lock);
570 if (!list_empty(&en->list))
571 list_del_init(&en->list);
572 spin_unlock(&sbi->extent_lock);
573 }
574
575 if (free_all || list_empty(&en->list)) {
576 __detach_extent_node(sbi, et, en);
577 kmem_cache_free(extent_node_slab, en);
578 }
579 node = next;
580 }
581
582 return count - et->count;
583}
584
585static void f2fs_init_extent_tree(struct inode *inode,
586 struct f2fs_extent *i_ext)
587{
588 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
589 struct extent_tree *et;
590 struct extent_node *en;
591 struct extent_info ei;
592
593 if (le32_to_cpu(i_ext->len) < F2FS_MIN_EXTENT_LEN)
594 return;
595
596 et = __grab_extent_tree(inode);
597
598 write_lock(&et->lock);
599 if (et->count)
600 goto out;
601
602 set_extent_info(&ei, le32_to_cpu(i_ext->fofs),
603 le32_to_cpu(i_ext->blk), le32_to_cpu(i_ext->len));
604
605 en = __insert_extent_tree(sbi, et, &ei, NULL);
606 if (en) {
607 et->cached_en = en;
608
609 spin_lock(&sbi->extent_lock);
610 list_add_tail(&en->list, &sbi->extent_list);
611 spin_unlock(&sbi->extent_lock);
612 }
613out:
614 write_unlock(&et->lock);
615 atomic_dec(&et->refcount);
616}
617
618static bool f2fs_lookup_extent_tree(struct inode *inode, pgoff_t pgofs,
619 struct extent_info *ei)
620{
621 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
622 struct extent_tree *et;
623 struct extent_node *en;
624
625 trace_f2fs_lookup_extent_tree_start(inode, pgofs);
626
627 et = __find_extent_tree(sbi, inode->i_ino);
628 if (!et)
629 return false;
630
631 read_lock(&et->lock);
632 en = __lookup_extent_tree(et, pgofs);
633 if (en) {
634 *ei = en->ei;
635 spin_lock(&sbi->extent_lock);
636 if (!list_empty(&en->list))
637 list_move_tail(&en->list, &sbi->extent_list);
638 spin_unlock(&sbi->extent_lock);
639 stat_inc_read_hit(sbi->sb);
640 }
641 stat_inc_total_hit(sbi->sb);
642 read_unlock(&et->lock);
643
644 trace_f2fs_lookup_extent_tree_end(inode, pgofs, en);
645
646 atomic_dec(&et->refcount);
647 return en ? true : false;
648}
649
650static void f2fs_update_extent_tree(struct inode *inode, pgoff_t fofs,
651 block_t blkaddr)
652{
653 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
654 struct extent_tree *et;
655 struct extent_node *en = NULL, *en1 = NULL, *en2 = NULL, *en3 = NULL;
656 struct extent_node *den = NULL;
657 struct extent_info ei, dei;
658 unsigned int endofs;
659
660 trace_f2fs_update_extent_tree(inode, fofs, blkaddr);
661
662 et = __grab_extent_tree(inode);
663
664 write_lock(&et->lock);
665
666 /* 1. lookup and remove existing extent info in cache */
667 en = __lookup_extent_tree(et, fofs);
668 if (!en)
669 goto update_extent;
670
671 dei = en->ei;
672 __detach_extent_node(sbi, et, en);
673
674 /* 2. if extent can be split more, split and insert the left part */
675 if (dei.len > 1) {
676 /* insert left part of split extent into cache */
677 if (fofs - dei.fofs >= F2FS_MIN_EXTENT_LEN) {
678 set_extent_info(&ei, dei.fofs, dei.blk,
679 fofs - dei.fofs);
680 en1 = __insert_extent_tree(sbi, et, &ei, NULL);
681 }
682
683 /* insert right part of split extent into cache */
684 endofs = dei.fofs + dei.len - 1;
685 if (endofs - fofs >= F2FS_MIN_EXTENT_LEN) {
686 set_extent_info(&ei, fofs + 1,
687 fofs - dei.fofs + dei.blk, endofs - fofs);
688 en2 = __insert_extent_tree(sbi, et, &ei, NULL);
689 }
690 }
691
692update_extent:
693 /* 3. update extent in extent cache */
694 if (blkaddr) {
695 set_extent_info(&ei, fofs, blkaddr, 1);
696 en3 = __insert_extent_tree(sbi, et, &ei, &den);
697 }
698
699 /* 4. update in global extent list */
700 spin_lock(&sbi->extent_lock);
701 if (en && !list_empty(&en->list))
702 list_del(&en->list);
703 /*
704 * en1 and en2 split from en, they will become more and more smaller
705 * fragments after splitting several times. So if the length is smaller
706 * than F2FS_MIN_EXTENT_LEN, we will not add them into extent tree.
707 */
708 if (en1)
709 list_add_tail(&en1->list, &sbi->extent_list);
710 if (en2)
711 list_add_tail(&en2->list, &sbi->extent_list);
712 if (en3) {
713 if (list_empty(&en3->list))
714 list_add_tail(&en3->list, &sbi->extent_list);
715 else
716 list_move_tail(&en3->list, &sbi->extent_list);
717 }
718 if (den && !list_empty(&den->list))
719 list_del(&den->list);
720 spin_unlock(&sbi->extent_lock);
721
722 /* 5. release extent node */
723 if (en)
724 kmem_cache_free(extent_node_slab, en);
725 if (den)
726 kmem_cache_free(extent_node_slab, den);
727
728 write_unlock(&et->lock);
729 atomic_dec(&et->refcount);
730}
731
732void f2fs_preserve_extent_tree(struct inode *inode)
733{
734 struct extent_tree *et;
735 struct extent_info *ext = &F2FS_I(inode)->ext;
736 bool sync = false;
737
738 if (!test_opt(F2FS_I_SB(inode), EXTENT_CACHE))
739 return;
740
741 et = __find_extent_tree(F2FS_I_SB(inode), inode->i_ino);
742 if (!et) {
743 if (ext->len) {
744 ext->len = 0;
745 update_inode_page(inode);
746 }
747 return;
748 }
749
750 read_lock(&et->lock);
751 if (et->count) {
752 struct extent_node *en;
753
754 if (et->cached_en) {
755 en = et->cached_en;
756 } else {
757 struct rb_node *node = rb_first(&et->root);
758
759 if (!node)
760 node = rb_last(&et->root);
761 en = rb_entry(node, struct extent_node, rb_node);
762 }
763
764 if (__is_extent_same(ext, &en->ei))
765 goto out;
766
767 *ext = en->ei;
768 sync = true;
769 } else if (ext->len) {
770 ext->len = 0;
771 sync = true;
772 }
773out:
774 read_unlock(&et->lock);
775 atomic_dec(&et->refcount);
776
777 if (sync)
778 update_inode_page(inode);
779}
780
781void f2fs_shrink_extent_tree(struct f2fs_sb_info *sbi, int nr_shrink)
782{
783 struct extent_tree *treevec[EXT_TREE_VEC_SIZE];
784 struct extent_node *en, *tmp;
785 unsigned long ino = F2FS_ROOT_INO(sbi);
786 struct radix_tree_iter iter;
787 void **slot;
788 unsigned int found;
789 unsigned int node_cnt = 0, tree_cnt = 0;
790
791 if (!test_opt(sbi, EXTENT_CACHE))
792 return;
793
794 if (available_free_memory(sbi, EXTENT_CACHE))
795 return;
796
797 spin_lock(&sbi->extent_lock);
798 list_for_each_entry_safe(en, tmp, &sbi->extent_list, list) {
799 if (!nr_shrink--)
800 break;
801 list_del_init(&en->list);
802 }
803 spin_unlock(&sbi->extent_lock);
804
805 down_read(&sbi->extent_tree_lock);
806 while ((found = radix_tree_gang_lookup(&sbi->extent_tree_root,
807 (void **)treevec, ino, EXT_TREE_VEC_SIZE))) {
808 unsigned i;
809
810 ino = treevec[found - 1]->ino + 1;
811 for (i = 0; i < found; i++) {
812 struct extent_tree *et = treevec[i];
813
814 atomic_inc(&et->refcount);
815 write_lock(&et->lock);
816 node_cnt += __free_extent_tree(sbi, et, false);
817 write_unlock(&et->lock);
818 atomic_dec(&et->refcount);
819 }
820 }
821 up_read(&sbi->extent_tree_lock);
822
823 down_write(&sbi->extent_tree_lock);
824 radix_tree_for_each_slot(slot, &sbi->extent_tree_root, &iter,
825 F2FS_ROOT_INO(sbi)) {
826 struct extent_tree *et = (struct extent_tree *)*slot;
827
828 if (!atomic_read(&et->refcount) && !et->count) {
829 radix_tree_delete(&sbi->extent_tree_root, et->ino);
830 kmem_cache_free(extent_tree_slab, et);
831 sbi->total_ext_tree--;
832 tree_cnt++;
833 }
834 }
835 up_write(&sbi->extent_tree_lock);
836
837 trace_f2fs_shrink_extent_tree(sbi, node_cnt, tree_cnt);
838}
839
840void f2fs_destroy_extent_tree(struct inode *inode)
841{
842 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
843 struct extent_tree *et;
844 unsigned int node_cnt = 0;
845
846 if (!test_opt(sbi, EXTENT_CACHE))
847 return;
848
849 et = __find_extent_tree(sbi, inode->i_ino);
850 if (!et)
851 goto out;
852
853 /* free all extent info belong to this extent tree */
854 write_lock(&et->lock);
855 node_cnt = __free_extent_tree(sbi, et, true);
856 write_unlock(&et->lock);
857
858 atomic_dec(&et->refcount);
859
860 /* try to find and delete extent tree entry in radix tree */
861 down_write(&sbi->extent_tree_lock);
862 et = radix_tree_lookup(&sbi->extent_tree_root, inode->i_ino);
863 if (!et) {
864 up_write(&sbi->extent_tree_lock);
865 goto out;
866 }
867 f2fs_bug_on(sbi, atomic_read(&et->refcount) || et->count);
868 radix_tree_delete(&sbi->extent_tree_root, inode->i_ino);
869 kmem_cache_free(extent_tree_slab, et);
870 sbi->total_ext_tree--;
871 up_write(&sbi->extent_tree_lock);
872out:
873 trace_f2fs_destroy_extent_tree(inode, node_cnt);
874 return;
875}
876
877void f2fs_init_extent_cache(struct inode *inode, struct f2fs_extent *i_ext)
878{
879 if (test_opt(F2FS_I_SB(inode), EXTENT_CACHE))
880 f2fs_init_extent_tree(inode, i_ext);
881
882 write_lock(&F2FS_I(inode)->ext_lock);
883 get_extent_info(&F2FS_I(inode)->ext, *i_ext);
884 write_unlock(&F2FS_I(inode)->ext_lock);
885}
886
887static bool f2fs_lookup_extent_cache(struct inode *inode, pgoff_t pgofs,
888 struct extent_info *ei)
889{
890 if (is_inode_flag_set(F2FS_I(inode), FI_NO_EXTENT))
891 return false;
892
893 if (test_opt(F2FS_I_SB(inode), EXTENT_CACHE))
894 return f2fs_lookup_extent_tree(inode, pgofs, ei);
895
896 return lookup_extent_info(inode, pgofs, ei);
897}
898
899void f2fs_update_extent_cache(struct dnode_of_data *dn)
900{
901 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
902 pgoff_t fofs;
903
904 f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR);
905
906 if (is_inode_flag_set(fi, FI_NO_EXTENT))
907 return;
908
909 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
910 dn->ofs_in_node;
911
912 if (test_opt(F2FS_I_SB(dn->inode), EXTENT_CACHE))
913 return f2fs_update_extent_tree(dn->inode, fofs,
914 dn->data_blkaddr);
915
916 if (update_extent_info(dn->inode, fofs, dn->data_blkaddr))
917 sync_inode_page(dn);
918}
919
920struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
921{
922 struct address_space *mapping = inode->i_mapping;
923 struct dnode_of_data dn;
924 struct page *page;
925 struct extent_info ei;
926 int err;
927 struct f2fs_io_info fio = {
928 .sbi = F2FS_I_SB(inode),
929 .type = DATA,
930 .rw = sync ? READ_SYNC : READA,
931 };
932
933 /*
934 * If sync is false, it needs to check its block allocation.
935 * This is need and triggered by two flows:
936 * gc and truncate_partial_data_page.
937 */
938 if (!sync)
939 goto search;
940
941 page = find_get_page(mapping, index);
942 if (page && PageUptodate(page))
943 return page;
944 f2fs_put_page(page, 0);
945search:
946 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
947 dn.data_blkaddr = ei.blk + index - ei.fofs;
948 goto got_it;
949 }
950
951 set_new_dnode(&dn, inode, NULL, NULL, 0);
952 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
953 if (err)
954 return ERR_PTR(err);
955 f2fs_put_dnode(&dn);
956
957 if (dn.data_blkaddr == NULL_ADDR)
958 return ERR_PTR(-ENOENT);
959
960 /* By fallocate(), there is no cached page, but with NEW_ADDR */
961 if (unlikely(dn.data_blkaddr == NEW_ADDR))
962 return ERR_PTR(-EINVAL);
963
964got_it:
965 page = grab_cache_page(mapping, index);
966 if (!page)
967 return ERR_PTR(-ENOMEM);
968
969 if (PageUptodate(page)) {
970 unlock_page(page);
971 return page;
972 }
973
974 fio.blk_addr = dn.data_blkaddr;
975 fio.page = page;
976 err = f2fs_submit_page_bio(&fio);
977 if (err)
978 return ERR_PTR(err);
979
980 if (sync) {
981 wait_on_page_locked(page);
982 if (unlikely(!PageUptodate(page))) {
983 f2fs_put_page(page, 0);
984 return ERR_PTR(-EIO);
985 }
986 }
987 return page;
988}
989
990/*
991 * If it tries to access a hole, return an error.
992 * Because, the callers, functions in dir.c and GC, should be able to know
993 * whether this page exists or not.
994 */
995struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
996{
997 struct address_space *mapping = inode->i_mapping;
998 struct dnode_of_data dn;
999 struct page *page;
1000 struct extent_info ei;
1001 int err;
1002 struct f2fs_io_info fio = {
1003 .sbi = F2FS_I_SB(inode),
1004 .type = DATA,
1005 .rw = READ_SYNC,
1006 };
1007repeat:
1008 page = grab_cache_page(mapping, index);
1009 if (!page)
1010 return ERR_PTR(-ENOMEM);
1011
1012 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1013 dn.data_blkaddr = ei.blk + index - ei.fofs;
1014 goto got_it;
1015 }
1016
1017 set_new_dnode(&dn, inode, NULL, NULL, 0);
1018 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1019 if (err) {
1020 f2fs_put_page(page, 1);
1021 return ERR_PTR(err);
1022 }
1023 f2fs_put_dnode(&dn);
1024
1025 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1026 f2fs_put_page(page, 1);
1027 return ERR_PTR(-ENOENT);
1028 }
1029
1030got_it:
1031 if (PageUptodate(page))
1032 return page;
1033
1034 /*
1035 * A new dentry page is allocated but not able to be written, since its
1036 * new inode page couldn't be allocated due to -ENOSPC.
1037 * In such the case, its blkaddr can be remained as NEW_ADDR.
1038 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
1039 */
1040 if (dn.data_blkaddr == NEW_ADDR) {
1041 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1042 SetPageUptodate(page);
1043 return page;
1044 }
1045
1046 fio.blk_addr = dn.data_blkaddr;
1047 fio.page = page;
1048 err = f2fs_submit_page_bio(&fio);
1049 if (err)
1050 return ERR_PTR(err);
1051
1052 lock_page(page);
1053 if (unlikely(!PageUptodate(page))) {
1054 f2fs_put_page(page, 1);
1055 return ERR_PTR(-EIO);
1056 }
1057 if (unlikely(page->mapping != mapping)) {
1058 f2fs_put_page(page, 1);
1059 goto repeat;
1060 }
1061 return page;
1062}
1063
1064/*
1065 * Caller ensures that this data page is never allocated.
1066 * A new zero-filled data page is allocated in the page cache.
1067 *
1068 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1069 * f2fs_unlock_op().
1070 * Note that, ipage is set only by make_empty_dir.
1071 */
1072struct page *get_new_data_page(struct inode *inode,
1073 struct page *ipage, pgoff_t index, bool new_i_size)
1074{
1075 struct address_space *mapping = inode->i_mapping;
1076 struct page *page;
1077 struct dnode_of_data dn;
1078 int err;
1079
1080 set_new_dnode(&dn, inode, ipage, NULL, 0);
1081 err = f2fs_reserve_block(&dn, index);
1082 if (err)
1083 return ERR_PTR(err);
1084repeat:
1085 page = grab_cache_page(mapping, index);
1086 if (!page) {
1087 err = -ENOMEM;
1088 goto put_err;
1089 }
1090
1091 if (PageUptodate(page))
1092 return page;
1093
1094 if (dn.data_blkaddr == NEW_ADDR) {
1095 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1096 SetPageUptodate(page);
1097 } else {
1098 struct f2fs_io_info fio = {
1099 .sbi = F2FS_I_SB(inode),
1100 .type = DATA,
1101 .rw = READ_SYNC,
1102 .blk_addr = dn.data_blkaddr,
1103 .page = page,
1104 };
1105 err = f2fs_submit_page_bio(&fio);
1106 if (err)
1107 goto put_err;
1108
1109 lock_page(page);
1110 if (unlikely(!PageUptodate(page))) {
1111 f2fs_put_page(page, 1);
1112 err = -EIO;
1113 goto put_err;
1114 }
1115 if (unlikely(page->mapping != mapping)) {
1116 f2fs_put_page(page, 1);
1117 goto repeat;
1118 }
1119 }
1120
1121 if (new_i_size &&
1122 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
1123 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
1124 /* Only the directory inode sets new_i_size */
1125 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
1126 }
1127 return page;
1128
1129put_err:
1130 f2fs_put_dnode(&dn);
1131 return ERR_PTR(err);
1132}
1133
1134static int __allocate_data_block(struct dnode_of_data *dn)
1135{
1136 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1137 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
1138 struct f2fs_summary sum;
1139 struct node_info ni;
1140 int seg = CURSEG_WARM_DATA;
1141 pgoff_t fofs;
1142
1143 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
1144 return -EPERM;
1145
1146 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
1147 if (dn->data_blkaddr == NEW_ADDR)
1148 goto alloc;
1149
1150 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
1151 return -ENOSPC;
1152
1153alloc:
1154 get_node_info(sbi, dn->nid, &ni);
1155 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1156
1157 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
1158 seg = CURSEG_DIRECT_IO;
1159
1160 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
1161 &sum, seg);
1162
1163 /* direct IO doesn't use extent cache to maximize the performance */
1164 set_data_blkaddr(dn);
1165
1166 /* update i_size */
1167 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
1168 dn->ofs_in_node;
1169 if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
1170 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
1171
1172 return 0;
1173}
1174
1175static void __allocate_data_blocks(struct inode *inode, loff_t offset,
1176 size_t count)
1177{
1178 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1179 struct dnode_of_data dn;
1180 u64 start = F2FS_BYTES_TO_BLK(offset);
1181 u64 len = F2FS_BYTES_TO_BLK(count);
1182 bool allocated;
1183 u64 end_offset;
1184
1185 while (len) {
1186 f2fs_balance_fs(sbi);
1187 f2fs_lock_op(sbi);
1188
1189 /* When reading holes, we need its node page */
1190 set_new_dnode(&dn, inode, NULL, NULL, 0);
1191 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
1192 goto out;
1193
1194 allocated = false;
1195 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1196
1197 while (dn.ofs_in_node < end_offset && len) {
1198 block_t blkaddr;
1199
1200 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
1201 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
1202 if (__allocate_data_block(&dn))
1203 goto sync_out;
1204 allocated = true;
1205 }
1206 len--;
1207 start++;
1208 dn.ofs_in_node++;
1209 }
1210
1211 if (allocated)
1212 sync_inode_page(&dn);
1213
1214 f2fs_put_dnode(&dn);
1215 f2fs_unlock_op(sbi);
1216 }
1217 return;
1218
1219sync_out:
1220 if (allocated)
1221 sync_inode_page(&dn);
1222 f2fs_put_dnode(&dn);
1223out:
1224 f2fs_unlock_op(sbi);
1225 return;
1226}
1227
1228/*
1229 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1230 * f2fs_map_blocks structure.
1231 * If original data blocks are allocated, then give them to blockdev.
1232 * Otherwise,
1233 * a. preallocate requested block addresses
1234 * b. do not use extent cache for better performance
1235 * c. give the block addresses to blockdev
1236 */
1237static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1238 int create, bool fiemap)
1239{
1240 unsigned int maxblocks = map->m_len;
1241 struct dnode_of_data dn;
1242 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
1243 pgoff_t pgofs, end_offset;
1244 int err = 0, ofs = 1;
1245 struct extent_info ei;
1246 bool allocated = false;
1247
1248 map->m_len = 0;
1249 map->m_flags = 0;
1250
1251 /* it only supports block size == page size */
1252 pgofs = (pgoff_t)map->m_lblk;
1253
1254 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1255 map->m_pblk = ei.blk + pgofs - ei.fofs;
1256 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1257 map->m_flags = F2FS_MAP_MAPPED;
1258 goto out;
1259 }
1260
1261 if (create)
1262 f2fs_lock_op(F2FS_I_SB(inode));
1263
1264 /* When reading holes, we need its node page */
1265 set_new_dnode(&dn, inode, NULL, NULL, 0);
1266 err = get_dnode_of_data(&dn, pgofs, mode);
1267 if (err) {
1268 if (err == -ENOENT)
1269 err = 0;
1270 goto unlock_out;
1271 }
1272 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
1273 goto put_out;
1274
1275 if (dn.data_blkaddr != NULL_ADDR) {
1276 map->m_flags = F2FS_MAP_MAPPED;
1277 map->m_pblk = dn.data_blkaddr;
1278 } else if (create) {
1279 err = __allocate_data_block(&dn);
1280 if (err)
1281 goto put_out;
1282 allocated = true;
1283 map->m_flags = F2FS_MAP_NEW | F2FS_MAP_MAPPED;
1284 map->m_pblk = dn.data_blkaddr;
1285 } else {
1286 goto put_out;
1287 }
1288
1289 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1290 map->m_len = 1;
1291 dn.ofs_in_node++;
1292 pgofs++;
1293
1294get_next:
1295 if (dn.ofs_in_node >= end_offset) {
1296 if (allocated)
1297 sync_inode_page(&dn);
1298 allocated = false;
1299 f2fs_put_dnode(&dn);
1300
1301 set_new_dnode(&dn, inode, NULL, NULL, 0);
1302 err = get_dnode_of_data(&dn, pgofs, mode);
1303 if (err) {
1304 if (err == -ENOENT)
1305 err = 0;
1306 goto unlock_out;
1307 }
1308 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
1309 goto put_out;
1310
1311 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1312 }
1313
1314 if (maxblocks > map->m_len) {
1315 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
1316 if (blkaddr == NULL_ADDR && create) {
1317 err = __allocate_data_block(&dn);
1318 if (err)
1319 goto sync_out;
1320 allocated = true;
1321 map->m_flags |= F2FS_MAP_NEW;
1322 blkaddr = dn.data_blkaddr;
1323 }
1324 /* Give more consecutive addresses for the readahead */
1325 if (map->m_pblk != NEW_ADDR && blkaddr == (map->m_pblk + ofs)) {
1326 ofs++;
1327 dn.ofs_in_node++;
1328 pgofs++;
1329 map->m_len++;
1330 goto get_next;
1331 }
1332 }
1333sync_out:
1334 if (allocated)
1335 sync_inode_page(&dn);
1336put_out:
1337 f2fs_put_dnode(&dn);
1338unlock_out:
1339 if (create)
1340 f2fs_unlock_op(F2FS_I_SB(inode));
1341out:
1342 trace_f2fs_map_blocks(inode, map, err);
1343 return err;
1344}
1345
1346static int __get_data_block(struct inode *inode, sector_t iblock,
1347 struct buffer_head *bh, int create, bool fiemap)
1348{
1349 struct f2fs_map_blocks map;
1350 int ret;
1351
1352 map.m_lblk = iblock;
1353 map.m_len = bh->b_size >> inode->i_blkbits;
1354
1355 ret = f2fs_map_blocks(inode, &map, create, fiemap);
1356 if (!ret) {
1357 map_bh(bh, inode->i_sb, map.m_pblk);
1358 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1359 bh->b_size = map.m_len << inode->i_blkbits;
1360 }
1361 return ret;
1362}
1363
1364static int get_data_block(struct inode *inode, sector_t iblock,
1365 struct buffer_head *bh_result, int create)
1366{
1367 return __get_data_block(inode, iblock, bh_result, create, false);
1368}
1369
1370static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
1371 struct buffer_head *bh_result, int create)
1372{
1373 return __get_data_block(inode, iblock, bh_result, create, true);
1374}
1375
1376int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1377 u64 start, u64 len)
1378{
1379 return generic_block_fiemap(inode, fieinfo,
1380 start, len, get_data_block_fiemap);
1381}
1382
1383/*
1384 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1385 * Major change was from block_size == page_size in f2fs by default.
1386 */
1387static int f2fs_mpage_readpages(struct address_space *mapping,
1388 struct list_head *pages, struct page *page,
1389 unsigned nr_pages)
1390{
1391 struct bio *bio = NULL;
1392 unsigned page_idx;
1393 sector_t last_block_in_bio = 0;
1394 struct inode *inode = mapping->host;
1395 const unsigned blkbits = inode->i_blkbits;
1396 const unsigned blocksize = 1 << blkbits;
1397 sector_t block_in_file;
1398 sector_t last_block;
1399 sector_t last_block_in_file;
1400 sector_t block_nr;
1401 struct block_device *bdev = inode->i_sb->s_bdev;
1402 struct f2fs_map_blocks map;
1403
1404 map.m_pblk = 0;
1405 map.m_lblk = 0;
1406 map.m_len = 0;
1407 map.m_flags = 0;
1408
1409 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1410
1411 prefetchw(&page->flags);
1412 if (pages) {
1413 page = list_entry(pages->prev, struct page, lru);
1414 list_del(&page->lru);
1415 if (add_to_page_cache_lru(page, mapping,
1416 page->index, GFP_KERNEL))
1417 goto next_page;
1418 }
1419
1420 block_in_file = (sector_t)page->index;
1421 last_block = block_in_file + nr_pages;
1422 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1423 blkbits;
1424 if (last_block > last_block_in_file)
1425 last_block = last_block_in_file;
1426
1427 /*
1428 * Map blocks using the previous result first.
1429 */
1430 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1431 block_in_file > map.m_lblk &&
1432 block_in_file < (map.m_lblk + map.m_len))
1433 goto got_it;
1434
1435 /*
1436 * Then do more f2fs_map_blocks() calls until we are
1437 * done with this page.
1438 */
1439 map.m_flags = 0;
1440
1441 if (block_in_file < last_block) {
1442 map.m_lblk = block_in_file;
1443 map.m_len = last_block - block_in_file;
1444
1445 if (f2fs_map_blocks(inode, &map, 0, false))
1446 goto set_error_page;
1447 }
1448got_it:
1449 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1450 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1451 SetPageMappedToDisk(page);
1452
1453 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1454 SetPageUptodate(page);
1455 goto confused;
1456 }
1457 } else {
1458 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1459 SetPageUptodate(page);
1460 unlock_page(page);
1461 goto next_page;
1462 }
1463
1464 /*
1465 * This page will go to BIO. Do we need to send this
1466 * BIO off first?
1467 */
1468 if (bio && (last_block_in_bio != block_nr - 1)) {
1469submit_and_realloc:
1470 submit_bio(READ, bio);
1471 bio = NULL;
1472 }
1473 if (bio == NULL) {
1474 bio = bio_alloc(GFP_KERNEL,
1475 min_t(int, nr_pages, bio_get_nr_vecs(bdev)));
1476 if (!bio)
1477 goto set_error_page;
1478 bio->bi_bdev = bdev;
1479 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1480 bio->bi_end_io = mpage_end_io;
1481 bio->bi_private = NULL;
1482 }
1483
1484 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1485 goto submit_and_realloc;
1486
1487 last_block_in_bio = block_nr;
1488 goto next_page;
1489set_error_page:
1490 SetPageError(page);
1491 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1492 unlock_page(page);
1493 goto next_page;
1494confused:
1495 if (bio) {
1496 submit_bio(READ, bio);
1497 bio = NULL;
1498 }
1499 unlock_page(page);
1500next_page:
1501 if (pages)
1502 page_cache_release(page);
1503 }
1504 BUG_ON(pages && !list_empty(pages));
1505 if (bio)
1506 submit_bio(READ, bio);
1507 return 0;
1508}
1509
1510static int f2fs_read_data_page(struct file *file, struct page *page)
1511{
1512 struct inode *inode = page->mapping->host;
1513 int ret = -EAGAIN;
1514
1515 trace_f2fs_readpage(page, DATA);
1516
1517 /* If the file has inline data, try to read it directly */
1518 if (f2fs_has_inline_data(inode))
1519 ret = f2fs_read_inline_data(inode, page);
1520 if (ret == -EAGAIN)
1521 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1522 return ret;
1523}
1524
1525static int f2fs_read_data_pages(struct file *file,
1526 struct address_space *mapping,
1527 struct list_head *pages, unsigned nr_pages)
1528{
1529 struct inode *inode = file->f_mapping->host;
1530
1531 /* If the file has inline data, skip readpages */
1532 if (f2fs_has_inline_data(inode))
1533 return 0;
1534
1535 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1536}
1537
1538int do_write_data_page(struct f2fs_io_info *fio)
1539{
1540 struct page *page = fio->page;
1541 struct inode *inode = page->mapping->host;
1542 struct dnode_of_data dn;
1543 int err = 0;
1544
1545 set_new_dnode(&dn, inode, NULL, NULL, 0);
1546 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1547 if (err)
1548 return err;
1549
1550 fio->blk_addr = dn.data_blkaddr;
1551
1552 /* This page is already truncated */
1553 if (fio->blk_addr == NULL_ADDR) {
1554 ClearPageUptodate(page);
1555 goto out_writepage;
1556 }
1557
1558 set_page_writeback(page);
1559
1560 /*
1561 * If current allocation needs SSR,
1562 * it had better in-place writes for updated data.
1563 */
1564 if (unlikely(fio->blk_addr != NEW_ADDR &&
1565 !is_cold_data(page) &&
1566 need_inplace_update(inode))) {
1567 rewrite_data_page(fio);
1568 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1569 trace_f2fs_do_write_data_page(page, IPU);
1570 } else {
1571 write_data_page(&dn, fio);
1572 set_data_blkaddr(&dn);
1573 f2fs_update_extent_cache(&dn);
1574 trace_f2fs_do_write_data_page(page, OPU);
1575 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1576 if (page->index == 0)
1577 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1578 }
1579out_writepage:
1580 f2fs_put_dnode(&dn);
1581 return err;
1582}
1583
1584static int f2fs_write_data_page(struct page *page,
1585 struct writeback_control *wbc)
1586{
1587 struct inode *inode = page->mapping->host;
1588 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1589 loff_t i_size = i_size_read(inode);
1590 const pgoff_t end_index = ((unsigned long long) i_size)
1591 >> PAGE_CACHE_SHIFT;
1592 unsigned offset = 0;
1593 bool need_balance_fs = false;
1594 int err = 0;
1595 struct f2fs_io_info fio = {
1596 .sbi = sbi,
1597 .type = DATA,
1598 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1599 .page = page,
1600 };
1601
1602 trace_f2fs_writepage(page, DATA);
1603
1604 if (page->index < end_index)
1605 goto write;
1606
1607 /*
1608 * If the offset is out-of-range of file size,
1609 * this page does not have to be written to disk.
1610 */
1611 offset = i_size & (PAGE_CACHE_SIZE - 1);
1612 if ((page->index >= end_index + 1) || !offset)
1613 goto out;
1614
1615 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1616write:
1617 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1618 goto redirty_out;
1619 if (f2fs_is_drop_cache(inode))
1620 goto out;
1621 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1622 available_free_memory(sbi, BASE_CHECK))
1623 goto redirty_out;
1624
1625 /* Dentry blocks are controlled by checkpoint */
1626 if (S_ISDIR(inode->i_mode)) {
1627 if (unlikely(f2fs_cp_error(sbi)))
1628 goto redirty_out;
1629 err = do_write_data_page(&fio);
1630 goto done;
1631 }
1632
1633 /* we should bypass data pages to proceed the kworkder jobs */
1634 if (unlikely(f2fs_cp_error(sbi))) {
1635 SetPageError(page);
1636 goto out;
1637 }
1638
1639 if (!wbc->for_reclaim)
1640 need_balance_fs = true;
1641 else if (has_not_enough_free_secs(sbi, 0))
1642 goto redirty_out;
1643
1644 err = -EAGAIN;
1645 f2fs_lock_op(sbi);
1646 if (f2fs_has_inline_data(inode))
1647 err = f2fs_write_inline_data(inode, page);
1648 if (err == -EAGAIN)
1649 err = do_write_data_page(&fio);
1650 f2fs_unlock_op(sbi);
1651done:
1652 if (err && err != -ENOENT)
1653 goto redirty_out;
1654
1655 clear_cold_data(page);
1656out:
1657 inode_dec_dirty_pages(inode);
1658 if (err)
1659 ClearPageUptodate(page);
1660 unlock_page(page);
1661 if (need_balance_fs)
1662 f2fs_balance_fs(sbi);
1663 if (wbc->for_reclaim)
1664 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1665 return 0;
1666
1667redirty_out:
1668 redirty_page_for_writepage(wbc, page);
1669 return AOP_WRITEPAGE_ACTIVATE;
1670}
1671
1672static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1673 void *data)
1674{
1675 struct address_space *mapping = data;
1676 int ret = mapping->a_ops->writepage(page, wbc);
1677 mapping_set_error(mapping, ret);
1678 return ret;
1679}
1680
1681static int f2fs_write_data_pages(struct address_space *mapping,
1682 struct writeback_control *wbc)
1683{
1684 struct inode *inode = mapping->host;
1685 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1686 bool locked = false;
1687 int ret;
1688 long diff;
1689
1690 trace_f2fs_writepages(mapping->host, wbc, DATA);
1691
1692 /* deal with chardevs and other special file */
1693 if (!mapping->a_ops->writepage)
1694 return 0;
1695
1696 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1697 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1698 available_free_memory(sbi, DIRTY_DENTS))
1699 goto skip_write;
1700
1701 /* during POR, we don't need to trigger writepage at all. */
1702 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1703 goto skip_write;
1704
1705 diff = nr_pages_to_write(sbi, DATA, wbc);
1706
1707 if (!S_ISDIR(inode->i_mode)) {
1708 mutex_lock(&sbi->writepages);
1709 locked = true;
1710 }
1711 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1712 if (locked)
1713 mutex_unlock(&sbi->writepages);
1714
1715 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1716
1717 remove_dirty_dir_inode(inode);
1718
1719 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1720 return ret;
1721
1722skip_write:
1723 wbc->pages_skipped += get_dirty_pages(inode);
1724 return 0;
1725}
1726
1727static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1728{
1729 struct inode *inode = mapping->host;
1730
1731 if (to > inode->i_size) {
1732 truncate_pagecache(inode, inode->i_size);
1733 truncate_blocks(inode, inode->i_size, true);
1734 }
1735}
1736
1737static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1738 loff_t pos, unsigned len, unsigned flags,
1739 struct page **pagep, void **fsdata)
1740{
1741 struct inode *inode = mapping->host;
1742 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1743 struct page *page, *ipage;
1744 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1745 struct dnode_of_data dn;
1746 int err = 0;
1747
1748 trace_f2fs_write_begin(inode, pos, len, flags);
1749
1750 f2fs_balance_fs(sbi);
1751
1752 /*
1753 * We should check this at this moment to avoid deadlock on inode page
1754 * and #0 page. The locking rule for inline_data conversion should be:
1755 * lock_page(page #0) -> lock_page(inode_page)
1756 */
1757 if (index != 0) {
1758 err = f2fs_convert_inline_inode(inode);
1759 if (err)
1760 goto fail;
1761 }
1762repeat:
1763 page = grab_cache_page_write_begin(mapping, index, flags);
1764 if (!page) {
1765 err = -ENOMEM;
1766 goto fail;
1767 }
1768
1769 *pagep = page;
1770
1771 f2fs_lock_op(sbi);
1772
1773 /* check inline_data */
1774 ipage = get_node_page(sbi, inode->i_ino);
1775 if (IS_ERR(ipage)) {
1776 err = PTR_ERR(ipage);
1777 goto unlock_fail;
1778 }
1779
1780 set_new_dnode(&dn, inode, ipage, ipage, 0);
1781
1782 if (f2fs_has_inline_data(inode)) {
1783 if (pos + len <= MAX_INLINE_DATA) {
1784 read_inline_data(page, ipage);
1785 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1786 sync_inode_page(&dn);
1787 goto put_next;
1788 }
1789 err = f2fs_convert_inline_page(&dn, page);
1790 if (err)
1791 goto put_fail;
1792 }
1793 err = f2fs_reserve_block(&dn, index);
1794 if (err)
1795 goto put_fail;
1796put_next:
1797 f2fs_put_dnode(&dn);
1798 f2fs_unlock_op(sbi);
1799
1800 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
1801 return 0;
1802
1803 f2fs_wait_on_page_writeback(page, DATA);
1804
1805 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1806 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1807 unsigned end = start + len;
1808
1809 /* Reading beyond i_size is simple: memset to zero */
1810 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1811 goto out;
1812 }
1813
1814 if (dn.data_blkaddr == NEW_ADDR) {
1815 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1816 } else {
1817 struct f2fs_io_info fio = {
1818 .sbi = sbi,
1819 .type = DATA,
1820 .rw = READ_SYNC,
1821 .blk_addr = dn.data_blkaddr,
1822 .page = page,
1823 };
1824 err = f2fs_submit_page_bio(&fio);
1825 if (err)
1826 goto fail;
1827
1828 lock_page(page);
1829 if (unlikely(!PageUptodate(page))) {
1830 f2fs_put_page(page, 1);
1831 err = -EIO;
1832 goto fail;
1833 }
1834 if (unlikely(page->mapping != mapping)) {
1835 f2fs_put_page(page, 1);
1836 goto repeat;
1837 }
1838 }
1839out:
1840 SetPageUptodate(page);
1841 clear_cold_data(page);
1842 return 0;
1843
1844put_fail:
1845 f2fs_put_dnode(&dn);
1846unlock_fail:
1847 f2fs_unlock_op(sbi);
1848 f2fs_put_page(page, 1);
1849fail:
1850 f2fs_write_failed(mapping, pos + len);
1851 return err;
1852}
1853
1854static int f2fs_write_end(struct file *file,
1855 struct address_space *mapping,
1856 loff_t pos, unsigned len, unsigned copied,
1857 struct page *page, void *fsdata)
1858{
1859 struct inode *inode = page->mapping->host;
1860
1861 trace_f2fs_write_end(inode, pos, len, copied);
1862
1863 set_page_dirty(page);
1864
1865 if (pos + copied > i_size_read(inode)) {
1866 i_size_write(inode, pos + copied);
1867 mark_inode_dirty(inode);
1868 update_inode_page(inode);
1869 }
1870
1871 f2fs_put_page(page, 1);
1872 return copied;
1873}
1874
1875static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1876 loff_t offset)
1877{
1878 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1879
1880 if (iov_iter_rw(iter) == READ)
1881 return 0;
1882
1883 if (offset & blocksize_mask)
1884 return -EINVAL;
1885
1886 if (iov_iter_alignment(iter) & blocksize_mask)
1887 return -EINVAL;
1888
1889 return 0;
1890}
1891
1892static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1893 loff_t offset)
1894{
1895 struct file *file = iocb->ki_filp;
1896 struct address_space *mapping = file->f_mapping;
1897 struct inode *inode = mapping->host;
1898 size_t count = iov_iter_count(iter);
1899 int err;
1900
1901 /* we don't need to use inline_data strictly */
1902 if (f2fs_has_inline_data(inode)) {
1903 err = f2fs_convert_inline_inode(inode);
1904 if (err)
1905 return err;
1906 }
1907
1908 if (check_direct_IO(inode, iter, offset))
1909 return 0;
1910
1911 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1912
1913 if (iov_iter_rw(iter) == WRITE)
1914 __allocate_data_blocks(inode, offset, count);
1915
1916 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block);
1917 if (err < 0 && iov_iter_rw(iter) == WRITE)
1918 f2fs_write_failed(mapping, offset + count);
1919
1920 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1921
1922 return err;
1923}
1924
1925void f2fs_invalidate_page(struct page *page, unsigned int offset,
1926 unsigned int length)
1927{
1928 struct inode *inode = page->mapping->host;
1929 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1930
1931 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1932 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1933 return;
1934
1935 if (PageDirty(page)) {
1936 if (inode->i_ino == F2FS_META_INO(sbi))
1937 dec_page_count(sbi, F2FS_DIRTY_META);
1938 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1939 dec_page_count(sbi, F2FS_DIRTY_NODES);
1940 else
1941 inode_dec_dirty_pages(inode);
1942 }
1943 ClearPagePrivate(page);
1944}
1945
1946int f2fs_release_page(struct page *page, gfp_t wait)
1947{
1948 /* If this is dirty page, keep PagePrivate */
1949 if (PageDirty(page))
1950 return 0;
1951
1952 ClearPagePrivate(page);
1953 return 1;
1954}
1955
1956static int f2fs_set_data_page_dirty(struct page *page)
1957{
1958 struct address_space *mapping = page->mapping;
1959 struct inode *inode = mapping->host;
1960
1961 trace_f2fs_set_page_dirty(page, DATA);
1962
1963 SetPageUptodate(page);
1964
1965 if (f2fs_is_atomic_file(inode)) {
1966 register_inmem_page(inode, page);
1967 return 1;
1968 }
1969
1970 mark_inode_dirty(inode);
1971
1972 if (!PageDirty(page)) {
1973 __set_page_dirty_nobuffers(page);
1974 update_dirty_page(inode, page);
1975 return 1;
1976 }
1977 return 0;
1978}
1979
1980static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1981{
1982 struct inode *inode = mapping->host;
1983
1984 /* we don't need to use inline_data strictly */
1985 if (f2fs_has_inline_data(inode)) {
1986 int err = f2fs_convert_inline_inode(inode);
1987 if (err)
1988 return err;
1989 }
1990 return generic_block_bmap(mapping, block, get_data_block);
1991}
1992
1993void init_extent_cache_info(struct f2fs_sb_info *sbi)
1994{
1995 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
1996 init_rwsem(&sbi->extent_tree_lock);
1997 INIT_LIST_HEAD(&sbi->extent_list);
1998 spin_lock_init(&sbi->extent_lock);
1999 sbi->total_ext_tree = 0;
2000 atomic_set(&sbi->total_ext_node, 0);
2001}
2002
2003int __init create_extent_cache(void)
2004{
2005 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
2006 sizeof(struct extent_tree));
2007 if (!extent_tree_slab)
2008 return -ENOMEM;
2009 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
2010 sizeof(struct extent_node));
2011 if (!extent_node_slab) {
2012 kmem_cache_destroy(extent_tree_slab);
2013 return -ENOMEM;
2014 }
2015 return 0;
2016}
2017
2018void destroy_extent_cache(void)
2019{
2020 kmem_cache_destroy(extent_node_slab);
2021 kmem_cache_destroy(extent_tree_slab);
2022}
2023
2024const struct address_space_operations f2fs_dblock_aops = {
2025 .readpage = f2fs_read_data_page,
2026 .readpages = f2fs_read_data_pages,
2027 .writepage = f2fs_write_data_page,
2028 .writepages = f2fs_write_data_pages,
2029 .write_begin = f2fs_write_begin,
2030 .write_end = f2fs_write_end,
2031 .set_page_dirty = f2fs_set_data_page_dirty,
2032 .invalidatepage = f2fs_invalidate_page,
2033 .releasepage = f2fs_release_page,
2034 .direct_IO = f2fs_direct_IO,
2035 .bmap = f2fs_bmap,
2036};