projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: expose f2fs_mpage_readpages
[deliverable/linux.git] / fs / f2fs / data.c
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
29 static struct kmem_cache *extent_tree_slab;
30 static struct kmem_cache *extent_node_slab;
31
32 static 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 */
55 static 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
75 static 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 */
103 static 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
119 static 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
135 void 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 */
161 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
162 struct f2fs_io_info *fio)
163 {
164 struct bio *bio;
165
166 trace_f2fs_submit_page_bio(page, fio);
167 f2fs_trace_ios(page, fio, 0);
168
169 /* Allocate a new bio */
170 bio = __bio_alloc(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
182 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
183 struct f2fs_io_info *fio)
184 {
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);
201 alloc_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, 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(page, fio, 0);
217
218 up_write(&io->io_rwsem);
219 trace_f2fs_submit_page_mbio(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 */
228 void 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
245 int 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
263 int 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
279 static 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
308 static 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 }
371 end_update:
372 write_unlock(&fi->ext_lock);
373 return need_update;
374 }
375
376 static 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
396 static 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
407 static 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
424 static 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
450 static 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
478 static 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
499 static 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
518 static 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
556 static 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
585 static 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 }
613 out:
614 write_unlock(&et->lock);
615 atomic_dec(&et->refcount);
616 }
617
618 static 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
650 static 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
692 update_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
732 void 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 }
773 out:
774 read_unlock(&et->lock);
775 atomic_dec(&et->refcount);
776
777 if (sync)
778 update_inode_page(inode);
779 }
780
781 void 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
840 void 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);
872 out:
873 trace_f2fs_destroy_extent_tree(inode, node_cnt);
874 return;
875 }
876
877 void 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
887 static 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
899 void 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
920 struct 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 .type = DATA,
929 .rw = sync ? READ_SYNC : READA,
930 };
931
932 /*
933 * If sync is false, it needs to check its block allocation.
934 * This is need and triggered by two flows:
935 * gc and truncate_partial_data_page.
936 */
937 if (!sync)
938 goto search;
939
940 page = find_get_page(mapping, index);
941 if (page && PageUptodate(page))
942 return page;
943 f2fs_put_page(page, 0);
944 search:
945 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
946 dn.data_blkaddr = ei.blk + index - ei.fofs;
947 goto got_it;
948 }
949
950 set_new_dnode(&dn, inode, NULL, NULL, 0);
951 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
952 if (err)
953 return ERR_PTR(err);
954 f2fs_put_dnode(&dn);
955
956 if (dn.data_blkaddr == NULL_ADDR)
957 return ERR_PTR(-ENOENT);
958
959 /* By fallocate(), there is no cached page, but with NEW_ADDR */
960 if (unlikely(dn.data_blkaddr == NEW_ADDR))
961 return ERR_PTR(-EINVAL);
962
963 got_it:
964 page = grab_cache_page(mapping, index);
965 if (!page)
966 return ERR_PTR(-ENOMEM);
967
968 if (PageUptodate(page)) {
969 unlock_page(page);
970 return page;
971 }
972
973 fio.blk_addr = dn.data_blkaddr;
974 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
975 if (err)
976 return ERR_PTR(err);
977
978 if (sync) {
979 wait_on_page_locked(page);
980 if (unlikely(!PageUptodate(page))) {
981 f2fs_put_page(page, 0);
982 return ERR_PTR(-EIO);
983 }
984 }
985 return page;
986 }
987
988 /*
989 * If it tries to access a hole, return an error.
990 * Because, the callers, functions in dir.c and GC, should be able to know
991 * whether this page exists or not.
992 */
993 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
994 {
995 struct address_space *mapping = inode->i_mapping;
996 struct dnode_of_data dn;
997 struct page *page;
998 struct extent_info ei;
999 int err;
1000 struct f2fs_io_info fio = {
1001 .type = DATA,
1002 .rw = READ_SYNC,
1003 };
1004 repeat:
1005 page = grab_cache_page(mapping, index);
1006 if (!page)
1007 return ERR_PTR(-ENOMEM);
1008
1009 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1010 dn.data_blkaddr = ei.blk + index - ei.fofs;
1011 goto got_it;
1012 }
1013
1014 set_new_dnode(&dn, inode, NULL, NULL, 0);
1015 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1016 if (err) {
1017 f2fs_put_page(page, 1);
1018 return ERR_PTR(err);
1019 }
1020 f2fs_put_dnode(&dn);
1021
1022 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1023 f2fs_put_page(page, 1);
1024 return ERR_PTR(-ENOENT);
1025 }
1026
1027 got_it:
1028 if (PageUptodate(page))
1029 return page;
1030
1031 /*
1032 * A new dentry page is allocated but not able to be written, since its
1033 * new inode page couldn't be allocated due to -ENOSPC.
1034 * In such the case, its blkaddr can be remained as NEW_ADDR.
1035 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
1036 */
1037 if (dn.data_blkaddr == NEW_ADDR) {
1038 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1039 SetPageUptodate(page);
1040 return page;
1041 }
1042
1043 fio.blk_addr = dn.data_blkaddr;
1044 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
1045 if (err)
1046 return ERR_PTR(err);
1047
1048 lock_page(page);
1049 if (unlikely(!PageUptodate(page))) {
1050 f2fs_put_page(page, 1);
1051 return ERR_PTR(-EIO);
1052 }
1053 if (unlikely(page->mapping != mapping)) {
1054 f2fs_put_page(page, 1);
1055 goto repeat;
1056 }
1057 return page;
1058 }
1059
1060 /*
1061 * Caller ensures that this data page is never allocated.
1062 * A new zero-filled data page is allocated in the page cache.
1063 *
1064 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1065 * f2fs_unlock_op().
1066 * Note that, ipage is set only by make_empty_dir.
1067 */
1068 struct page *get_new_data_page(struct inode *inode,
1069 struct page *ipage, pgoff_t index, bool new_i_size)
1070 {
1071 struct address_space *mapping = inode->i_mapping;
1072 struct page *page;
1073 struct dnode_of_data dn;
1074 int err;
1075
1076 set_new_dnode(&dn, inode, ipage, NULL, 0);
1077 err = f2fs_reserve_block(&dn, index);
1078 if (err)
1079 return ERR_PTR(err);
1080 repeat:
1081 page = grab_cache_page(mapping, index);
1082 if (!page) {
1083 err = -ENOMEM;
1084 goto put_err;
1085 }
1086
1087 if (PageUptodate(page))
1088 return page;
1089
1090 if (dn.data_blkaddr == NEW_ADDR) {
1091 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1092 SetPageUptodate(page);
1093 } else {
1094 struct f2fs_io_info fio = {
1095 .type = DATA,
1096 .rw = READ_SYNC,
1097 .blk_addr = dn.data_blkaddr,
1098 };
1099 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
1100 if (err)
1101 goto put_err;
1102
1103 lock_page(page);
1104 if (unlikely(!PageUptodate(page))) {
1105 f2fs_put_page(page, 1);
1106 err = -EIO;
1107 goto put_err;
1108 }
1109 if (unlikely(page->mapping != mapping)) {
1110 f2fs_put_page(page, 1);
1111 goto repeat;
1112 }
1113 }
1114
1115 if (new_i_size &&
1116 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
1117 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
1118 /* Only the directory inode sets new_i_size */
1119 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
1120 }
1121 return page;
1122
1123 put_err:
1124 f2fs_put_dnode(&dn);
1125 return ERR_PTR(err);
1126 }
1127
1128 static int __allocate_data_block(struct dnode_of_data *dn)
1129 {
1130 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1131 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
1132 struct f2fs_summary sum;
1133 struct node_info ni;
1134 int seg = CURSEG_WARM_DATA;
1135 pgoff_t fofs;
1136
1137 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
1138 return -EPERM;
1139
1140 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
1141 if (dn->data_blkaddr == NEW_ADDR)
1142 goto alloc;
1143
1144 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
1145 return -ENOSPC;
1146
1147 alloc:
1148 get_node_info(sbi, dn->nid, &ni);
1149 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1150
1151 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
1152 seg = CURSEG_DIRECT_IO;
1153
1154 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
1155 &sum, seg);
1156
1157 /* direct IO doesn't use extent cache to maximize the performance */
1158 set_data_blkaddr(dn);
1159
1160 /* update i_size */
1161 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
1162 dn->ofs_in_node;
1163 if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
1164 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
1165
1166 return 0;
1167 }
1168
1169 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
1170 size_t count)
1171 {
1172 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1173 struct dnode_of_data dn;
1174 u64 start = F2FS_BYTES_TO_BLK(offset);
1175 u64 len = F2FS_BYTES_TO_BLK(count);
1176 bool allocated;
1177 u64 end_offset;
1178
1179 while (len) {
1180 f2fs_balance_fs(sbi);
1181 f2fs_lock_op(sbi);
1182
1183 /* When reading holes, we need its node page */
1184 set_new_dnode(&dn, inode, NULL, NULL, 0);
1185 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
1186 goto out;
1187
1188 allocated = false;
1189 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1190
1191 while (dn.ofs_in_node < end_offset && len) {
1192 block_t blkaddr;
1193
1194 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
1195 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
1196 if (__allocate_data_block(&dn))
1197 goto sync_out;
1198 allocated = true;
1199 }
1200 len--;
1201 start++;
1202 dn.ofs_in_node++;
1203 }
1204
1205 if (allocated)
1206 sync_inode_page(&dn);
1207
1208 f2fs_put_dnode(&dn);
1209 f2fs_unlock_op(sbi);
1210 }
1211 return;
1212
1213 sync_out:
1214 if (allocated)
1215 sync_inode_page(&dn);
1216 f2fs_put_dnode(&dn);
1217 out:
1218 f2fs_unlock_op(sbi);
1219 return;
1220 }
1221
1222 /*
1223 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1224 * f2fs_map_blocks structure.
1225 * If original data blocks are allocated, then give them to blockdev.
1226 * Otherwise,
1227 * a. preallocate requested block addresses
1228 * b. do not use extent cache for better performance
1229 * c. give the block addresses to blockdev
1230 */
1231 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1232 int create, bool fiemap)
1233 {
1234 unsigned int maxblocks = map->m_len;
1235 struct dnode_of_data dn;
1236 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
1237 pgoff_t pgofs, end_offset;
1238 int err = 0, ofs = 1;
1239 struct extent_info ei;
1240 bool allocated = false;
1241
1242 map->m_len = 0;
1243 map->m_flags = 0;
1244
1245 /* it only supports block size == page size */
1246 pgofs = (pgoff_t)map->m_lblk;
1247
1248 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1249 map->m_pblk = ei.blk + pgofs - ei.fofs;
1250 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1251 map->m_flags = F2FS_MAP_MAPPED;
1252 goto out;
1253 }
1254
1255 if (create)
1256 f2fs_lock_op(F2FS_I_SB(inode));
1257
1258 /* When reading holes, we need its node page */
1259 set_new_dnode(&dn, inode, NULL, NULL, 0);
1260 err = get_dnode_of_data(&dn, pgofs, mode);
1261 if (err) {
1262 if (err == -ENOENT)
1263 err = 0;
1264 goto unlock_out;
1265 }
1266 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
1267 goto put_out;
1268
1269 if (dn.data_blkaddr != NULL_ADDR) {
1270 map->m_flags = F2FS_MAP_MAPPED;
1271 map->m_pblk = dn.data_blkaddr;
1272 } else if (create) {
1273 err = __allocate_data_block(&dn);
1274 if (err)
1275 goto put_out;
1276 allocated = true;
1277 map->m_flags = F2FS_MAP_NEW | F2FS_MAP_MAPPED;
1278 map->m_pblk = dn.data_blkaddr;
1279 } else {
1280 goto put_out;
1281 }
1282
1283 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1284 map->m_len = 1;
1285 dn.ofs_in_node++;
1286 pgofs++;
1287
1288 get_next:
1289 if (dn.ofs_in_node >= end_offset) {
1290 if (allocated)
1291 sync_inode_page(&dn);
1292 allocated = false;
1293 f2fs_put_dnode(&dn);
1294
1295 set_new_dnode(&dn, inode, NULL, NULL, 0);
1296 err = get_dnode_of_data(&dn, pgofs, mode);
1297 if (err) {
1298 if (err == -ENOENT)
1299 err = 0;
1300 goto unlock_out;
1301 }
1302 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
1303 goto put_out;
1304
1305 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
1306 }
1307
1308 if (maxblocks > map->m_len) {
1309 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
1310 if (blkaddr == NULL_ADDR && create) {
1311 err = __allocate_data_block(&dn);
1312 if (err)
1313 goto sync_out;
1314 allocated = true;
1315 map->m_flags |= F2FS_MAP_NEW;
1316 blkaddr = dn.data_blkaddr;
1317 }
1318 /* Give more consecutive addresses for the readahead */
1319 if (map->m_pblk != NEW_ADDR && blkaddr == (map->m_pblk + ofs)) {
1320 ofs++;
1321 dn.ofs_in_node++;
1322 pgofs++;
1323 map->m_len++;
1324 goto get_next;
1325 }
1326 }
1327 sync_out:
1328 if (allocated)
1329 sync_inode_page(&dn);
1330 put_out:
1331 f2fs_put_dnode(&dn);
1332 unlock_out:
1333 if (create)
1334 f2fs_unlock_op(F2FS_I_SB(inode));
1335 out:
1336 trace_f2fs_map_blocks(inode, map, err);
1337 return err;
1338 }
1339
1340 static int __get_data_block(struct inode *inode, sector_t iblock,
1341 struct buffer_head *bh, int create, bool fiemap)
1342 {
1343 struct f2fs_map_blocks map;
1344 int ret;
1345
1346 map.m_lblk = iblock;
1347 map.m_len = bh->b_size >> inode->i_blkbits;
1348
1349 ret = f2fs_map_blocks(inode, &map, create, fiemap);
1350 if (!ret) {
1351 map_bh(bh, inode->i_sb, map.m_pblk);
1352 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1353 bh->b_size = map.m_len << inode->i_blkbits;
1354 }
1355 return ret;
1356 }
1357
1358 static int get_data_block(struct inode *inode, sector_t iblock,
1359 struct buffer_head *bh_result, int create)
1360 {
1361 return __get_data_block(inode, iblock, bh_result, create, false);
1362 }
1363
1364 static int get_data_block_fiemap(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, true);
1368 }
1369
1370 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1371 u64 start, u64 len)
1372 {
1373 return generic_block_fiemap(inode, fieinfo,
1374 start, len, get_data_block_fiemap);
1375 }
1376
1377 /*
1378 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1379 * Major change was from block_size == page_size in f2fs by default.
1380 */
1381 static int f2fs_mpage_readpages(struct address_space *mapping,
1382 struct list_head *pages, struct page *page,
1383 unsigned nr_pages)
1384 {
1385 struct bio *bio = NULL;
1386 unsigned page_idx;
1387 sector_t last_block_in_bio = 0;
1388 struct inode *inode = mapping->host;
1389 const unsigned blkbits = inode->i_blkbits;
1390 const unsigned blocksize = 1 << blkbits;
1391 sector_t block_in_file;
1392 sector_t last_block;
1393 sector_t last_block_in_file;
1394 sector_t block_nr;
1395 struct block_device *bdev = inode->i_sb->s_bdev;
1396 struct f2fs_map_blocks map;
1397
1398 map.m_pblk = 0;
1399 map.m_lblk = 0;
1400 map.m_len = 0;
1401 map.m_flags = 0;
1402
1403 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1404
1405 prefetchw(&page->flags);
1406 if (pages) {
1407 page = list_entry(pages->prev, struct page, lru);
1408 list_del(&page->lru);
1409 if (add_to_page_cache_lru(page, mapping,
1410 page->index, GFP_KERNEL))
1411 goto next_page;
1412 }
1413
1414 block_in_file = (sector_t)page->index;
1415 last_block = block_in_file + nr_pages;
1416 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1417 blkbits;
1418 if (last_block > last_block_in_file)
1419 last_block = last_block_in_file;
1420
1421 /*
1422 * Map blocks using the previous result first.
1423 */
1424 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1425 block_in_file > map.m_lblk &&
1426 block_in_file < (map.m_lblk + map.m_len))
1427 goto got_it;
1428
1429 /*
1430 * Then do more f2fs_map_blocks() calls until we are
1431 * done with this page.
1432 */
1433 map.m_flags = 0;
1434
1435 if (block_in_file < last_block) {
1436 map.m_lblk = block_in_file;
1437 map.m_len = last_block - block_in_file;
1438
1439 if (f2fs_map_blocks(inode, &map, 0, false))
1440 goto set_error_page;
1441 }
1442 got_it:
1443 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1444 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1445 SetPageMappedToDisk(page);
1446
1447 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1448 SetPageUptodate(page);
1449 goto confused;
1450 }
1451 } else {
1452 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1453 SetPageUptodate(page);
1454 unlock_page(page);
1455 goto next_page;
1456 }
1457
1458 /*
1459 * This page will go to BIO. Do we need to send this
1460 * BIO off first?
1461 */
1462 if (bio && (last_block_in_bio != block_nr - 1)) {
1463 submit_and_realloc:
1464 submit_bio(READ, bio);
1465 bio = NULL;
1466 }
1467 if (bio == NULL) {
1468 bio = bio_alloc(GFP_KERNEL,
1469 min_t(int, nr_pages, bio_get_nr_vecs(bdev)));
1470 if (!bio)
1471 goto set_error_page;
1472 bio->bi_bdev = bdev;
1473 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1474 bio->bi_end_io = mpage_end_io;
1475 bio->bi_private = NULL;
1476 }
1477
1478 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1479 goto submit_and_realloc;
1480
1481 last_block_in_bio = block_nr;
1482 goto next_page;
1483 set_error_page:
1484 SetPageError(page);
1485 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1486 unlock_page(page);
1487 goto next_page;
1488 confused:
1489 if (bio) {
1490 submit_bio(READ, bio);
1491 bio = NULL;
1492 }
1493 unlock_page(page);
1494 next_page:
1495 if (pages)
1496 page_cache_release(page);
1497 }
1498 BUG_ON(pages && !list_empty(pages));
1499 if (bio)
1500 submit_bio(READ, bio);
1501 return 0;
1502 }
1503
1504 static int f2fs_read_data_page(struct file *file, struct page *page)
1505 {
1506 struct inode *inode = page->mapping->host;
1507 int ret = -EAGAIN;
1508
1509 trace_f2fs_readpage(page, DATA);
1510
1511 /* If the file has inline data, try to read it directly */
1512 if (f2fs_has_inline_data(inode))
1513 ret = f2fs_read_inline_data(inode, page);
1514 if (ret == -EAGAIN)
1515 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1516 return ret;
1517 }
1518
1519 static int f2fs_read_data_pages(struct file *file,
1520 struct address_space *mapping,
1521 struct list_head *pages, unsigned nr_pages)
1522 {
1523 struct inode *inode = file->f_mapping->host;
1524
1525 /* If the file has inline data, skip readpages */
1526 if (f2fs_has_inline_data(inode))
1527 return 0;
1528
1529 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1530 }
1531
1532 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
1533 {
1534 struct inode *inode = page->mapping->host;
1535 struct dnode_of_data dn;
1536 int err = 0;
1537
1538 set_new_dnode(&dn, inode, NULL, NULL, 0);
1539 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1540 if (err)
1541 return err;
1542
1543 fio->blk_addr = dn.data_blkaddr;
1544
1545 /* This page is already truncated */
1546 if (fio->blk_addr == NULL_ADDR) {
1547 ClearPageUptodate(page);
1548 goto out_writepage;
1549 }
1550
1551 set_page_writeback(page);
1552
1553 /*
1554 * If current allocation needs SSR,
1555 * it had better in-place writes for updated data.
1556 */
1557 if (unlikely(fio->blk_addr != NEW_ADDR &&
1558 !is_cold_data(page) &&
1559 need_inplace_update(inode))) {
1560 rewrite_data_page(page, fio);
1561 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1562 trace_f2fs_do_write_data_page(page, IPU);
1563 } else {
1564 write_data_page(page, &dn, fio);
1565 set_data_blkaddr(&dn);
1566 f2fs_update_extent_cache(&dn);
1567 trace_f2fs_do_write_data_page(page, OPU);
1568 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1569 if (page->index == 0)
1570 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1571 }
1572 out_writepage:
1573 f2fs_put_dnode(&dn);
1574 return err;
1575 }
1576
1577 static int f2fs_write_data_page(struct page *page,
1578 struct writeback_control *wbc)
1579 {
1580 struct inode *inode = page->mapping->host;
1581 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1582 loff_t i_size = i_size_read(inode);
1583 const pgoff_t end_index = ((unsigned long long) i_size)
1584 >> PAGE_CACHE_SHIFT;
1585 unsigned offset = 0;
1586 bool need_balance_fs = false;
1587 int err = 0;
1588 struct f2fs_io_info fio = {
1589 .type = DATA,
1590 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1591 };
1592
1593 trace_f2fs_writepage(page, DATA);
1594
1595 if (page->index < end_index)
1596 goto write;
1597
1598 /*
1599 * If the offset is out-of-range of file size,
1600 * this page does not have to be written to disk.
1601 */
1602 offset = i_size & (PAGE_CACHE_SIZE - 1);
1603 if ((page->index >= end_index + 1) || !offset)
1604 goto out;
1605
1606 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1607 write:
1608 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1609 goto redirty_out;
1610 if (f2fs_is_drop_cache(inode))
1611 goto out;
1612 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1613 available_free_memory(sbi, BASE_CHECK))
1614 goto redirty_out;
1615
1616 /* Dentry blocks are controlled by checkpoint */
1617 if (S_ISDIR(inode->i_mode)) {
1618 if (unlikely(f2fs_cp_error(sbi)))
1619 goto redirty_out;
1620 err = do_write_data_page(page, &fio);
1621 goto done;
1622 }
1623
1624 /* we should bypass data pages to proceed the kworkder jobs */
1625 if (unlikely(f2fs_cp_error(sbi))) {
1626 SetPageError(page);
1627 goto out;
1628 }
1629
1630 if (!wbc->for_reclaim)
1631 need_balance_fs = true;
1632 else if (has_not_enough_free_secs(sbi, 0))
1633 goto redirty_out;
1634
1635 err = -EAGAIN;
1636 f2fs_lock_op(sbi);
1637 if (f2fs_has_inline_data(inode))
1638 err = f2fs_write_inline_data(inode, page);
1639 if (err == -EAGAIN)
1640 err = do_write_data_page(page, &fio);
1641 f2fs_unlock_op(sbi);
1642 done:
1643 if (err && err != -ENOENT)
1644 goto redirty_out;
1645
1646 clear_cold_data(page);
1647 out:
1648 inode_dec_dirty_pages(inode);
1649 if (err)
1650 ClearPageUptodate(page);
1651 unlock_page(page);
1652 if (need_balance_fs)
1653 f2fs_balance_fs(sbi);
1654 if (wbc->for_reclaim)
1655 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1656 return 0;
1657
1658 redirty_out:
1659 redirty_page_for_writepage(wbc, page);
1660 return AOP_WRITEPAGE_ACTIVATE;
1661 }
1662
1663 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1664 void *data)
1665 {
1666 struct address_space *mapping = data;
1667 int ret = mapping->a_ops->writepage(page, wbc);
1668 mapping_set_error(mapping, ret);
1669 return ret;
1670 }
1671
1672 static int f2fs_write_data_pages(struct address_space *mapping,
1673 struct writeback_control *wbc)
1674 {
1675 struct inode *inode = mapping->host;
1676 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1677 bool locked = false;
1678 int ret;
1679 long diff;
1680
1681 trace_f2fs_writepages(mapping->host, wbc, DATA);
1682
1683 /* deal with chardevs and other special file */
1684 if (!mapping->a_ops->writepage)
1685 return 0;
1686
1687 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1688 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1689 available_free_memory(sbi, DIRTY_DENTS))
1690 goto skip_write;
1691
1692 /* during POR, we don't need to trigger writepage at all. */
1693 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1694 goto skip_write;
1695
1696 diff = nr_pages_to_write(sbi, DATA, wbc);
1697
1698 if (!S_ISDIR(inode->i_mode)) {
1699 mutex_lock(&sbi->writepages);
1700 locked = true;
1701 }
1702 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1703 if (locked)
1704 mutex_unlock(&sbi->writepages);
1705
1706 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1707
1708 remove_dirty_dir_inode(inode);
1709
1710 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1711 return ret;
1712
1713 skip_write:
1714 wbc->pages_skipped += get_dirty_pages(inode);
1715 return 0;
1716 }
1717
1718 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1719 {
1720 struct inode *inode = mapping->host;
1721
1722 if (to > inode->i_size) {
1723 truncate_pagecache(inode, inode->i_size);
1724 truncate_blocks(inode, inode->i_size, true);
1725 }
1726 }
1727
1728 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1729 loff_t pos, unsigned len, unsigned flags,
1730 struct page **pagep, void **fsdata)
1731 {
1732 struct inode *inode = mapping->host;
1733 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1734 struct page *page, *ipage;
1735 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1736 struct dnode_of_data dn;
1737 int err = 0;
1738
1739 trace_f2fs_write_begin(inode, pos, len, flags);
1740
1741 f2fs_balance_fs(sbi);
1742
1743 /*
1744 * We should check this at this moment to avoid deadlock on inode page
1745 * and #0 page. The locking rule for inline_data conversion should be:
1746 * lock_page(page #0) -> lock_page(inode_page)
1747 */
1748 if (index != 0) {
1749 err = f2fs_convert_inline_inode(inode);
1750 if (err)
1751 goto fail;
1752 }
1753 repeat:
1754 page = grab_cache_page_write_begin(mapping, index, flags);
1755 if (!page) {
1756 err = -ENOMEM;
1757 goto fail;
1758 }
1759
1760 *pagep = page;
1761
1762 f2fs_lock_op(sbi);
1763
1764 /* check inline_data */
1765 ipage = get_node_page(sbi, inode->i_ino);
1766 if (IS_ERR(ipage)) {
1767 err = PTR_ERR(ipage);
1768 goto unlock_fail;
1769 }
1770
1771 set_new_dnode(&dn, inode, ipage, ipage, 0);
1772
1773 if (f2fs_has_inline_data(inode)) {
1774 if (pos + len <= MAX_INLINE_DATA) {
1775 read_inline_data(page, ipage);
1776 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1777 sync_inode_page(&dn);
1778 goto put_next;
1779 }
1780 err = f2fs_convert_inline_page(&dn, page);
1781 if (err)
1782 goto put_fail;
1783 }
1784 err = f2fs_reserve_block(&dn, index);
1785 if (err)
1786 goto put_fail;
1787 put_next:
1788 f2fs_put_dnode(&dn);
1789 f2fs_unlock_op(sbi);
1790
1791 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
1792 return 0;
1793
1794 f2fs_wait_on_page_writeback(page, DATA);
1795
1796 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1797 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1798 unsigned end = start + len;
1799
1800 /* Reading beyond i_size is simple: memset to zero */
1801 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1802 goto out;
1803 }
1804
1805 if (dn.data_blkaddr == NEW_ADDR) {
1806 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1807 } else {
1808 struct f2fs_io_info fio = {
1809 .type = DATA,
1810 .rw = READ_SYNC,
1811 .blk_addr = dn.data_blkaddr,
1812 };
1813 err = f2fs_submit_page_bio(sbi, page, &fio);
1814 if (err)
1815 goto fail;
1816
1817 lock_page(page);
1818 if (unlikely(!PageUptodate(page))) {
1819 f2fs_put_page(page, 1);
1820 err = -EIO;
1821 goto fail;
1822 }
1823 if (unlikely(page->mapping != mapping)) {
1824 f2fs_put_page(page, 1);
1825 goto repeat;
1826 }
1827 }
1828 out:
1829 SetPageUptodate(page);
1830 clear_cold_data(page);
1831 return 0;
1832
1833 put_fail:
1834 f2fs_put_dnode(&dn);
1835 unlock_fail:
1836 f2fs_unlock_op(sbi);
1837 f2fs_put_page(page, 1);
1838 fail:
1839 f2fs_write_failed(mapping, pos + len);
1840 return err;
1841 }
1842
1843 static int f2fs_write_end(struct file *file,
1844 struct address_space *mapping,
1845 loff_t pos, unsigned len, unsigned copied,
1846 struct page *page, void *fsdata)
1847 {
1848 struct inode *inode = page->mapping->host;
1849
1850 trace_f2fs_write_end(inode, pos, len, copied);
1851
1852 set_page_dirty(page);
1853
1854 if (pos + copied > i_size_read(inode)) {
1855 i_size_write(inode, pos + copied);
1856 mark_inode_dirty(inode);
1857 update_inode_page(inode);
1858 }
1859
1860 f2fs_put_page(page, 1);
1861 return copied;
1862 }
1863
1864 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1865 loff_t offset)
1866 {
1867 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1868
1869 if (iov_iter_rw(iter) == READ)
1870 return 0;
1871
1872 if (offset & blocksize_mask)
1873 return -EINVAL;
1874
1875 if (iov_iter_alignment(iter) & blocksize_mask)
1876 return -EINVAL;
1877
1878 return 0;
1879 }
1880
1881 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1882 loff_t offset)
1883 {
1884 struct file *file = iocb->ki_filp;
1885 struct address_space *mapping = file->f_mapping;
1886 struct inode *inode = mapping->host;
1887 size_t count = iov_iter_count(iter);
1888 int err;
1889
1890 /* we don't need to use inline_data strictly */
1891 if (f2fs_has_inline_data(inode)) {
1892 err = f2fs_convert_inline_inode(inode);
1893 if (err)
1894 return err;
1895 }
1896
1897 if (check_direct_IO(inode, iter, offset))
1898 return 0;
1899
1900 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1901
1902 if (iov_iter_rw(iter) == WRITE)
1903 __allocate_data_blocks(inode, offset, count);
1904
1905 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block);
1906 if (err < 0 && iov_iter_rw(iter) == WRITE)
1907 f2fs_write_failed(mapping, offset + count);
1908
1909 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1910
1911 return err;
1912 }
1913
1914 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1915 unsigned int length)
1916 {
1917 struct inode *inode = page->mapping->host;
1918 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1919
1920 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1921 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1922 return;
1923
1924 if (PageDirty(page)) {
1925 if (inode->i_ino == F2FS_META_INO(sbi))
1926 dec_page_count(sbi, F2FS_DIRTY_META);
1927 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1928 dec_page_count(sbi, F2FS_DIRTY_NODES);
1929 else
1930 inode_dec_dirty_pages(inode);
1931 }
1932 ClearPagePrivate(page);
1933 }
1934
1935 int f2fs_release_page(struct page *page, gfp_t wait)
1936 {
1937 /* If this is dirty page, keep PagePrivate */
1938 if (PageDirty(page))
1939 return 0;
1940
1941 ClearPagePrivate(page);
1942 return 1;
1943 }
1944
1945 static int f2fs_set_data_page_dirty(struct page *page)
1946 {
1947 struct address_space *mapping = page->mapping;
1948 struct inode *inode = mapping->host;
1949
1950 trace_f2fs_set_page_dirty(page, DATA);
1951
1952 SetPageUptodate(page);
1953
1954 if (f2fs_is_atomic_file(inode)) {
1955 register_inmem_page(inode, page);
1956 return 1;
1957 }
1958
1959 mark_inode_dirty(inode);
1960
1961 if (!PageDirty(page)) {
1962 __set_page_dirty_nobuffers(page);
1963 update_dirty_page(inode, page);
1964 return 1;
1965 }
1966 return 0;
1967 }
1968
1969 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1970 {
1971 struct inode *inode = mapping->host;
1972
1973 /* we don't need to use inline_data strictly */
1974 if (f2fs_has_inline_data(inode)) {
1975 int err = f2fs_convert_inline_inode(inode);
1976 if (err)
1977 return err;
1978 }
1979 return generic_block_bmap(mapping, block, get_data_block);
1980 }
1981
1982 void init_extent_cache_info(struct f2fs_sb_info *sbi)
1983 {
1984 INIT_RADIX_TREE(&sbi->extent_tree_root, GFP_NOIO);
1985 init_rwsem(&sbi->extent_tree_lock);
1986 INIT_LIST_HEAD(&sbi->extent_list);
1987 spin_lock_init(&sbi->extent_lock);
1988 sbi->total_ext_tree = 0;
1989 atomic_set(&sbi->total_ext_node, 0);
1990 }
1991
1992 int __init create_extent_cache(void)
1993 {
1994 extent_tree_slab = f2fs_kmem_cache_create("f2fs_extent_tree",
1995 sizeof(struct extent_tree));
1996 if (!extent_tree_slab)
1997 return -ENOMEM;
1998 extent_node_slab = f2fs_kmem_cache_create("f2fs_extent_node",
1999 sizeof(struct extent_node));
2000 if (!extent_node_slab) {
2001 kmem_cache_destroy(extent_tree_slab);
2002 return -ENOMEM;
2003 }
2004 return 0;
2005 }
2006
2007 void destroy_extent_cache(void)
2008 {
2009 kmem_cache_destroy(extent_node_slab);
2010 kmem_cache_destroy(extent_tree_slab);
2011 }
2012
2013 const struct address_space_operations f2fs_dblock_aops = {
2014 .readpage = f2fs_read_data_page,
2015 .readpages = f2fs_read_data_pages,
2016 .writepage = f2fs_write_data_page,
2017 .writepages = f2fs_write_data_pages,
2018 .write_begin = f2fs_write_begin,
2019 .write_end = f2fs_write_end,
2020 .set_page_dirty = f2fs_set_data_page_dirty,
2021 .invalidatepage = f2fs_invalidate_page,
2022 .releasepage = f2fs_release_page,
2023 .direct_IO = f2fs_direct_IO,
2024 .bmap = f2fs_bmap,
2025 };
Linux kernel - Deliverables
This page took 0.094677 seconds and 5 git commands to generate.