f2fs: call mark_inode_dirty to flush dirty pages
[GitHub/exynos8895/android_kernel_samsung_universal8895.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/aio.h>
16 #include <linux/writeback.h>
17 #include <linux/backing-dev.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21
22 #include "f2fs.h"
23 #include "node.h"
24 #include "segment.h"
25 #include <trace/events/f2fs.h>
26
27 static void f2fs_read_end_io(struct bio *bio, int err)
28 {
29 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
30 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
31
32 do {
33 struct page *page = bvec->bv_page;
34
35 if (--bvec >= bio->bi_io_vec)
36 prefetchw(&bvec->bv_page->flags);
37
38 if (unlikely(!uptodate)) {
39 ClearPageUptodate(page);
40 SetPageError(page);
41 } else {
42 SetPageUptodate(page);
43 }
44 unlock_page(page);
45 } while (bvec >= bio->bi_io_vec);
46
47 bio_put(bio);
48 }
49
50 static void f2fs_write_end_io(struct bio *bio, int err)
51 {
52 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
53 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
54 struct f2fs_sb_info *sbi = F2FS_SB(bvec->bv_page->mapping->host->i_sb);
55
56 do {
57 struct page *page = bvec->bv_page;
58
59 if (--bvec >= bio->bi_io_vec)
60 prefetchw(&bvec->bv_page->flags);
61
62 if (unlikely(!uptodate)) {
63 SetPageError(page);
64 set_bit(AS_EIO, &page->mapping->flags);
65 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
66 sbi->sb->s_flags |= MS_RDONLY;
67 }
68 end_page_writeback(page);
69 dec_page_count(sbi, F2FS_WRITEBACK);
70 } while (bvec >= bio->bi_io_vec);
71
72 if (bio->bi_private)
73 complete(bio->bi_private);
74
75 if (!get_pages(sbi, F2FS_WRITEBACK) &&
76 !list_empty(&sbi->cp_wait.task_list))
77 wake_up(&sbi->cp_wait);
78
79 bio_put(bio);
80 }
81
82 /*
83 * Low-level block read/write IO operations.
84 */
85 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
86 int npages, bool is_read)
87 {
88 struct bio *bio;
89
90 /* No failure on bio allocation */
91 bio = bio_alloc(GFP_NOIO, npages);
92
93 bio->bi_bdev = sbi->sb->s_bdev;
94 bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
95 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
96
97 return bio;
98 }
99
100 static void __submit_merged_bio(struct f2fs_bio_info *io)
101 {
102 struct f2fs_io_info *fio = &io->fio;
103 int rw;
104
105 if (!io->bio)
106 return;
107
108 rw = fio->rw;
109
110 if (is_read_io(rw)) {
111 trace_f2fs_submit_read_bio(io->sbi->sb, rw,
112 fio->type, io->bio);
113 submit_bio(rw, io->bio);
114 } else {
115 trace_f2fs_submit_write_bio(io->sbi->sb, rw,
116 fio->type, io->bio);
117 /*
118 * META_FLUSH is only from the checkpoint procedure, and we
119 * should wait this metadata bio for FS consistency.
120 */
121 if (fio->type == META_FLUSH) {
122 DECLARE_COMPLETION_ONSTACK(wait);
123 io->bio->bi_private = &wait;
124 submit_bio(rw, io->bio);
125 wait_for_completion(&wait);
126 } else {
127 submit_bio(rw, io->bio);
128 }
129 }
130
131 io->bio = NULL;
132 }
133
134 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
135 enum page_type type, int rw)
136 {
137 enum page_type btype = PAGE_TYPE_OF_BIO(type);
138 struct f2fs_bio_info *io;
139
140 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
141
142 mutex_lock(&io->io_mutex);
143
144 /* change META to META_FLUSH in the checkpoint procedure */
145 if (type >= META_FLUSH) {
146 io->fio.type = META_FLUSH;
147 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
148 }
149 __submit_merged_bio(io);
150 mutex_unlock(&io->io_mutex);
151 }
152
153 /*
154 * Fill the locked page with data located in the block address.
155 * Return unlocked page.
156 */
157 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
158 block_t blk_addr, int rw)
159 {
160 struct bio *bio;
161
162 trace_f2fs_submit_page_bio(page, blk_addr, rw);
163
164 /* Allocate a new bio */
165 bio = __bio_alloc(sbi, blk_addr, 1, is_read_io(rw));
166
167 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
168 bio_put(bio);
169 f2fs_put_page(page, 1);
170 return -EFAULT;
171 }
172
173 submit_bio(rw, bio);
174 return 0;
175 }
176
177 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
178 block_t blk_addr, struct f2fs_io_info *fio)
179 {
180 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
181 struct f2fs_bio_info *io;
182 bool is_read = is_read_io(fio->rw);
183
184 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
185
186 verify_block_addr(sbi, blk_addr);
187
188 mutex_lock(&io->io_mutex);
189
190 if (!is_read)
191 inc_page_count(sbi, F2FS_WRITEBACK);
192
193 if (io->bio && (io->last_block_in_bio != blk_addr - 1 ||
194 io->fio.rw != fio->rw))
195 __submit_merged_bio(io);
196 alloc_new:
197 if (io->bio == NULL) {
198 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
199
200 io->bio = __bio_alloc(sbi, blk_addr, bio_blocks, is_read);
201 io->fio = *fio;
202 }
203
204 if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
205 PAGE_CACHE_SIZE) {
206 __submit_merged_bio(io);
207 goto alloc_new;
208 }
209
210 io->last_block_in_bio = blk_addr;
211
212 mutex_unlock(&io->io_mutex);
213 trace_f2fs_submit_page_mbio(page, fio->rw, fio->type, blk_addr);
214 }
215
216 /*
217 * Lock ordering for the change of data block address:
218 * ->data_page
219 * ->node_page
220 * update block addresses in the node page
221 */
222 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
223 {
224 struct f2fs_node *rn;
225 __le32 *addr_array;
226 struct page *node_page = dn->node_page;
227 unsigned int ofs_in_node = dn->ofs_in_node;
228
229 f2fs_wait_on_page_writeback(node_page, NODE);
230
231 rn = F2FS_NODE(node_page);
232
233 /* Get physical address of data block */
234 addr_array = blkaddr_in_node(rn);
235 addr_array[ofs_in_node] = cpu_to_le32(new_addr);
236 set_page_dirty(node_page);
237 }
238
239 int reserve_new_block(struct dnode_of_data *dn)
240 {
241 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
242
243 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
244 return -EPERM;
245 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
246 return -ENOSPC;
247
248 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
249
250 __set_data_blkaddr(dn, NEW_ADDR);
251 dn->data_blkaddr = NEW_ADDR;
252 mark_inode_dirty(dn->inode);
253 sync_inode_page(dn);
254 return 0;
255 }
256
257 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
258 {
259 bool need_put = dn->inode_page ? false : true;
260 int err;
261
262 /* if inode_page exists, index should be zero */
263 f2fs_bug_on(!need_put && index);
264
265 err = get_dnode_of_data(dn, index, ALLOC_NODE);
266 if (err)
267 return err;
268
269 if (dn->data_blkaddr == NULL_ADDR)
270 err = reserve_new_block(dn);
271 if (err || need_put)
272 f2fs_put_dnode(dn);
273 return err;
274 }
275
276 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
277 struct buffer_head *bh_result)
278 {
279 struct f2fs_inode_info *fi = F2FS_I(inode);
280 pgoff_t start_fofs, end_fofs;
281 block_t start_blkaddr;
282
283 if (is_inode_flag_set(fi, FI_NO_EXTENT))
284 return 0;
285
286 read_lock(&fi->ext.ext_lock);
287 if (fi->ext.len == 0) {
288 read_unlock(&fi->ext.ext_lock);
289 return 0;
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_addr;
297
298 if (pgofs >= start_fofs && pgofs <= end_fofs) {
299 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
300 size_t count;
301
302 clear_buffer_new(bh_result);
303 map_bh(bh_result, inode->i_sb,
304 start_blkaddr + pgofs - start_fofs);
305 count = end_fofs - pgofs + 1;
306 if (count < (UINT_MAX >> blkbits))
307 bh_result->b_size = (count << blkbits);
308 else
309 bh_result->b_size = UINT_MAX;
310
311 stat_inc_read_hit(inode->i_sb);
312 read_unlock(&fi->ext.ext_lock);
313 return 1;
314 }
315 read_unlock(&fi->ext.ext_lock);
316 return 0;
317 }
318
319 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
320 {
321 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
322 pgoff_t fofs, start_fofs, end_fofs;
323 block_t start_blkaddr, end_blkaddr;
324 int need_update = true;
325
326 f2fs_bug_on(blk_addr == NEW_ADDR);
327 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
328 dn->ofs_in_node;
329
330 /* Update the page address in the parent node */
331 __set_data_blkaddr(dn, blk_addr);
332
333 if (is_inode_flag_set(fi, FI_NO_EXTENT))
334 return;
335
336 write_lock(&fi->ext.ext_lock);
337
338 start_fofs = fi->ext.fofs;
339 end_fofs = fi->ext.fofs + fi->ext.len - 1;
340 start_blkaddr = fi->ext.blk_addr;
341 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
342
343 /* Drop and initialize the matched extent */
344 if (fi->ext.len == 1 && fofs == start_fofs)
345 fi->ext.len = 0;
346
347 /* Initial extent */
348 if (fi->ext.len == 0) {
349 if (blk_addr != NULL_ADDR) {
350 fi->ext.fofs = fofs;
351 fi->ext.blk_addr = blk_addr;
352 fi->ext.len = 1;
353 }
354 goto end_update;
355 }
356
357 /* Front merge */
358 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
359 fi->ext.fofs--;
360 fi->ext.blk_addr--;
361 fi->ext.len++;
362 goto end_update;
363 }
364
365 /* Back merge */
366 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
367 fi->ext.len++;
368 goto end_update;
369 }
370
371 /* Split the existing extent */
372 if (fi->ext.len > 1 &&
373 fofs >= start_fofs && fofs <= end_fofs) {
374 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
375 fi->ext.len = fofs - start_fofs;
376 } else {
377 fi->ext.fofs = fofs + 1;
378 fi->ext.blk_addr = start_blkaddr +
379 fofs - start_fofs + 1;
380 fi->ext.len -= fofs - start_fofs + 1;
381 }
382 } else {
383 need_update = false;
384 }
385
386 /* Finally, if the extent is very fragmented, let's drop the cache. */
387 if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
388 fi->ext.len = 0;
389 set_inode_flag(fi, FI_NO_EXTENT);
390 need_update = true;
391 }
392 end_update:
393 write_unlock(&fi->ext.ext_lock);
394 if (need_update)
395 sync_inode_page(dn);
396 return;
397 }
398
399 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
400 {
401 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
402 struct address_space *mapping = inode->i_mapping;
403 struct dnode_of_data dn;
404 struct page *page;
405 int err;
406
407 page = find_get_page(mapping, index);
408 if (page && PageUptodate(page))
409 return page;
410 f2fs_put_page(page, 0);
411
412 set_new_dnode(&dn, inode, NULL, NULL, 0);
413 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
414 if (err)
415 return ERR_PTR(err);
416 f2fs_put_dnode(&dn);
417
418 if (dn.data_blkaddr == NULL_ADDR)
419 return ERR_PTR(-ENOENT);
420
421 /* By fallocate(), there is no cached page, but with NEW_ADDR */
422 if (unlikely(dn.data_blkaddr == NEW_ADDR))
423 return ERR_PTR(-EINVAL);
424
425 page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
426 if (!page)
427 return ERR_PTR(-ENOMEM);
428
429 if (PageUptodate(page)) {
430 unlock_page(page);
431 return page;
432 }
433
434 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
435 sync ? READ_SYNC : READA);
436 if (err)
437 return ERR_PTR(err);
438
439 if (sync) {
440 wait_on_page_locked(page);
441 if (unlikely(!PageUptodate(page))) {
442 f2fs_put_page(page, 0);
443 return ERR_PTR(-EIO);
444 }
445 }
446 return page;
447 }
448
449 /*
450 * If it tries to access a hole, return an error.
451 * Because, the callers, functions in dir.c and GC, should be able to know
452 * whether this page exists or not.
453 */
454 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
455 {
456 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
457 struct address_space *mapping = inode->i_mapping;
458 struct dnode_of_data dn;
459 struct page *page;
460 int err;
461
462 repeat:
463 page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
464 if (!page)
465 return ERR_PTR(-ENOMEM);
466
467 set_new_dnode(&dn, inode, NULL, NULL, 0);
468 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
469 if (err) {
470 f2fs_put_page(page, 1);
471 return ERR_PTR(err);
472 }
473 f2fs_put_dnode(&dn);
474
475 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
476 f2fs_put_page(page, 1);
477 return ERR_PTR(-ENOENT);
478 }
479
480 if (PageUptodate(page))
481 return page;
482
483 /*
484 * A new dentry page is allocated but not able to be written, since its
485 * new inode page couldn't be allocated due to -ENOSPC.
486 * In such the case, its blkaddr can be remained as NEW_ADDR.
487 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
488 */
489 if (dn.data_blkaddr == NEW_ADDR) {
490 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
491 SetPageUptodate(page);
492 return page;
493 }
494
495 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, READ_SYNC);
496 if (err)
497 return ERR_PTR(err);
498
499 lock_page(page);
500 if (unlikely(!PageUptodate(page))) {
501 f2fs_put_page(page, 1);
502 return ERR_PTR(-EIO);
503 }
504 if (unlikely(page->mapping != mapping)) {
505 f2fs_put_page(page, 1);
506 goto repeat;
507 }
508 return page;
509 }
510
511 /*
512 * Caller ensures that this data page is never allocated.
513 * A new zero-filled data page is allocated in the page cache.
514 *
515 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
516 * f2fs_unlock_op().
517 * Note that, ipage is set only by make_empty_dir.
518 */
519 struct page *get_new_data_page(struct inode *inode,
520 struct page *ipage, pgoff_t index, bool new_i_size)
521 {
522 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
523 struct address_space *mapping = inode->i_mapping;
524 struct page *page;
525 struct dnode_of_data dn;
526 int err;
527
528 set_new_dnode(&dn, inode, ipage, NULL, 0);
529 err = f2fs_reserve_block(&dn, index);
530 if (err)
531 return ERR_PTR(err);
532 repeat:
533 page = grab_cache_page(mapping, index);
534 if (!page) {
535 err = -ENOMEM;
536 goto put_err;
537 }
538
539 if (PageUptodate(page))
540 return page;
541
542 if (dn.data_blkaddr == NEW_ADDR) {
543 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
544 SetPageUptodate(page);
545 } else {
546 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
547 READ_SYNC);
548 if (err)
549 goto put_err;
550
551 lock_page(page);
552 if (unlikely(!PageUptodate(page))) {
553 f2fs_put_page(page, 1);
554 err = -EIO;
555 goto put_err;
556 }
557 if (unlikely(page->mapping != mapping)) {
558 f2fs_put_page(page, 1);
559 goto repeat;
560 }
561 }
562
563 if (new_i_size &&
564 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
565 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
566 /* Only the directory inode sets new_i_size */
567 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
568 }
569 return page;
570
571 put_err:
572 f2fs_put_dnode(&dn);
573 return ERR_PTR(err);
574 }
575
576 static int __allocate_data_block(struct dnode_of_data *dn)
577 {
578 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
579 struct f2fs_summary sum;
580 block_t new_blkaddr;
581 struct node_info ni;
582 int type;
583
584 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
585 return -EPERM;
586 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
587 return -ENOSPC;
588
589 __set_data_blkaddr(dn, NEW_ADDR);
590 dn->data_blkaddr = NEW_ADDR;
591
592 get_node_info(sbi, dn->nid, &ni);
593 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
594
595 type = CURSEG_WARM_DATA;
596
597 allocate_data_block(sbi, NULL, NULL_ADDR, &new_blkaddr, &sum, type);
598
599 /* direct IO doesn't use extent cache to maximize the performance */
600 set_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
601 update_extent_cache(new_blkaddr, dn);
602 clear_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
603
604 dn->data_blkaddr = new_blkaddr;
605 return 0;
606 }
607
608 /*
609 * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
610 * If original data blocks are allocated, then give them to blockdev.
611 * Otherwise,
612 * a. preallocate requested block addresses
613 * b. do not use extent cache for better performance
614 * c. give the block addresses to blockdev
615 */
616 static int get_data_block(struct inode *inode, sector_t iblock,
617 struct buffer_head *bh_result, int create)
618 {
619 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
620 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
621 unsigned maxblocks = bh_result->b_size >> blkbits;
622 struct dnode_of_data dn;
623 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
624 pgoff_t pgofs, end_offset;
625 int err = 0, ofs = 1;
626 bool allocated = false;
627
628 /* Get the page offset from the block offset(iblock) */
629 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
630
631 if (check_extent_cache(inode, pgofs, bh_result))
632 goto out;
633
634 if (create)
635 f2fs_lock_op(sbi);
636
637 /* When reading holes, we need its node page */
638 set_new_dnode(&dn, inode, NULL, NULL, 0);
639 err = get_dnode_of_data(&dn, pgofs, mode);
640 if (err) {
641 if (err == -ENOENT)
642 err = 0;
643 goto unlock_out;
644 }
645 if (dn.data_blkaddr == NEW_ADDR)
646 goto put_out;
647
648 if (dn.data_blkaddr != NULL_ADDR) {
649 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
650 } else if (create) {
651 err = __allocate_data_block(&dn);
652 if (err)
653 goto put_out;
654 allocated = true;
655 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
656 } else {
657 goto put_out;
658 }
659
660 end_offset = IS_INODE(dn.node_page) ?
661 ADDRS_PER_INODE(F2FS_I(inode)) : ADDRS_PER_BLOCK;
662 bh_result->b_size = (((size_t)1) << blkbits);
663 dn.ofs_in_node++;
664 pgofs++;
665
666 get_next:
667 if (dn.ofs_in_node >= end_offset) {
668 if (allocated)
669 sync_inode_page(&dn);
670 allocated = false;
671 f2fs_put_dnode(&dn);
672
673 set_new_dnode(&dn, inode, NULL, NULL, 0);
674 err = get_dnode_of_data(&dn, pgofs, mode);
675 if (err) {
676 if (err == -ENOENT)
677 err = 0;
678 goto unlock_out;
679 }
680 if (dn.data_blkaddr == NEW_ADDR)
681 goto put_out;
682
683 end_offset = IS_INODE(dn.node_page) ?
684 ADDRS_PER_INODE(F2FS_I(inode)) : ADDRS_PER_BLOCK;
685 }
686
687 if (maxblocks > (bh_result->b_size >> blkbits)) {
688 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
689 if (blkaddr == NULL_ADDR && create) {
690 err = __allocate_data_block(&dn);
691 if (err)
692 goto sync_out;
693 allocated = true;
694 blkaddr = dn.data_blkaddr;
695 }
696 /* Give more consecutive addresses for the read ahead */
697 if (blkaddr == (bh_result->b_blocknr + ofs)) {
698 ofs++;
699 dn.ofs_in_node++;
700 pgofs++;
701 bh_result->b_size += (((size_t)1) << blkbits);
702 goto get_next;
703 }
704 }
705 sync_out:
706 if (allocated)
707 sync_inode_page(&dn);
708 put_out:
709 f2fs_put_dnode(&dn);
710 unlock_out:
711 if (create)
712 f2fs_unlock_op(sbi);
713 out:
714 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
715 return err;
716 }
717
718 static int f2fs_read_data_page(struct file *file, struct page *page)
719 {
720 struct inode *inode = page->mapping->host;
721 int ret;
722
723 /* If the file has inline data, try to read it directlly */
724 if (f2fs_has_inline_data(inode))
725 ret = f2fs_read_inline_data(inode, page);
726 else
727 ret = mpage_readpage(page, get_data_block);
728
729 return ret;
730 }
731
732 static int f2fs_read_data_pages(struct file *file,
733 struct address_space *mapping,
734 struct list_head *pages, unsigned nr_pages)
735 {
736 struct inode *inode = file->f_mapping->host;
737
738 /* If the file has inline data, skip readpages */
739 if (f2fs_has_inline_data(inode))
740 return 0;
741
742 return mpage_readpages(mapping, pages, nr_pages, get_data_block);
743 }
744
745 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
746 {
747 struct inode *inode = page->mapping->host;
748 block_t old_blkaddr, new_blkaddr;
749 struct dnode_of_data dn;
750 int err = 0;
751
752 set_new_dnode(&dn, inode, NULL, NULL, 0);
753 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
754 if (err)
755 return err;
756
757 old_blkaddr = dn.data_blkaddr;
758
759 /* This page is already truncated */
760 if (old_blkaddr == NULL_ADDR)
761 goto out_writepage;
762
763 set_page_writeback(page);
764
765 /*
766 * If current allocation needs SSR,
767 * it had better in-place writes for updated data.
768 */
769 if (unlikely(old_blkaddr != NEW_ADDR &&
770 !is_cold_data(page) &&
771 need_inplace_update(inode))) {
772 rewrite_data_page(page, old_blkaddr, fio);
773 } else {
774 write_data_page(page, &dn, &new_blkaddr, fio);
775 update_extent_cache(new_blkaddr, &dn);
776 }
777 out_writepage:
778 f2fs_put_dnode(&dn);
779 return err;
780 }
781
782 static int f2fs_write_data_page(struct page *page,
783 struct writeback_control *wbc)
784 {
785 struct inode *inode = page->mapping->host;
786 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
787 loff_t i_size = i_size_read(inode);
788 const pgoff_t end_index = ((unsigned long long) i_size)
789 >> PAGE_CACHE_SHIFT;
790 unsigned offset = 0;
791 bool need_balance_fs = false;
792 int err = 0;
793 struct f2fs_io_info fio = {
794 .type = DATA,
795 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
796 };
797
798 if (page->index < end_index)
799 goto write;
800
801 /*
802 * If the offset is out-of-range of file size,
803 * this page does not have to be written to disk.
804 */
805 offset = i_size & (PAGE_CACHE_SIZE - 1);
806 if ((page->index >= end_index + 1) || !offset) {
807 if (S_ISDIR(inode->i_mode)) {
808 dec_page_count(sbi, F2FS_DIRTY_DENTS);
809 inode_dec_dirty_dents(inode);
810 }
811 goto out;
812 }
813
814 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
815 write:
816 if (unlikely(sbi->por_doing)) {
817 err = AOP_WRITEPAGE_ACTIVATE;
818 goto redirty_out;
819 }
820
821 /* Dentry blocks are controlled by checkpoint */
822 if (S_ISDIR(inode->i_mode)) {
823 dec_page_count(sbi, F2FS_DIRTY_DENTS);
824 inode_dec_dirty_dents(inode);
825 err = do_write_data_page(page, &fio);
826 } else {
827 f2fs_lock_op(sbi);
828
829 if (f2fs_has_inline_data(inode) || f2fs_may_inline(inode)) {
830 err = f2fs_write_inline_data(inode, page, offset);
831 f2fs_unlock_op(sbi);
832 goto out;
833 } else {
834 err = do_write_data_page(page, &fio);
835 }
836
837 f2fs_unlock_op(sbi);
838 need_balance_fs = true;
839 }
840 if (err == -ENOENT)
841 goto out;
842 else if (err)
843 goto redirty_out;
844
845 if (wbc->for_reclaim) {
846 f2fs_submit_merged_bio(sbi, DATA, WRITE);
847 need_balance_fs = false;
848 }
849
850 clear_cold_data(page);
851 out:
852 unlock_page(page);
853 if (need_balance_fs)
854 f2fs_balance_fs(sbi);
855 return 0;
856
857 redirty_out:
858 wbc->pages_skipped++;
859 set_page_dirty(page);
860 return err;
861 }
862
863 #define MAX_DESIRED_PAGES_WP 4096
864
865 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
866 void *data)
867 {
868 struct address_space *mapping = data;
869 int ret = mapping->a_ops->writepage(page, wbc);
870 mapping_set_error(mapping, ret);
871 return ret;
872 }
873
874 static int f2fs_write_data_pages(struct address_space *mapping,
875 struct writeback_control *wbc)
876 {
877 struct inode *inode = mapping->host;
878 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
879 bool locked = false;
880 int ret;
881 long excess_nrtw = 0, desired_nrtw;
882
883 /* deal with chardevs and other special file */
884 if (!mapping->a_ops->writepage)
885 return 0;
886
887 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
888 desired_nrtw = MAX_DESIRED_PAGES_WP;
889 excess_nrtw = desired_nrtw - wbc->nr_to_write;
890 wbc->nr_to_write = desired_nrtw;
891 }
892
893 if (!S_ISDIR(inode->i_mode)) {
894 mutex_lock(&sbi->writepages);
895 locked = true;
896 }
897 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
898 if (locked)
899 mutex_unlock(&sbi->writepages);
900
901 f2fs_submit_merged_bio(sbi, DATA, WRITE);
902
903 remove_dirty_dir_inode(inode);
904
905 wbc->nr_to_write -= excess_nrtw;
906 return ret;
907 }
908
909 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
910 loff_t pos, unsigned len, unsigned flags,
911 struct page **pagep, void **fsdata)
912 {
913 struct inode *inode = mapping->host;
914 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
915 struct page *page;
916 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
917 struct dnode_of_data dn;
918 int err = 0;
919
920 f2fs_balance_fs(sbi);
921 repeat:
922 err = f2fs_convert_inline_data(inode, pos + len);
923 if (err)
924 return err;
925
926 page = grab_cache_page_write_begin(mapping, index, flags);
927 if (!page)
928 return -ENOMEM;
929 *pagep = page;
930
931 if (f2fs_has_inline_data(inode) && (pos + len) <= MAX_INLINE_DATA)
932 goto inline_data;
933
934 f2fs_lock_op(sbi);
935 set_new_dnode(&dn, inode, NULL, NULL, 0);
936 err = f2fs_reserve_block(&dn, index);
937 f2fs_unlock_op(sbi);
938
939 if (err) {
940 f2fs_put_page(page, 1);
941 return err;
942 }
943 inline_data:
944 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
945 return 0;
946
947 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
948 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
949 unsigned end = start + len;
950
951 /* Reading beyond i_size is simple: memset to zero */
952 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
953 goto out;
954 }
955
956 if (dn.data_blkaddr == NEW_ADDR) {
957 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
958 } else {
959 if (f2fs_has_inline_data(inode))
960 err = f2fs_read_inline_data(inode, page);
961 else
962 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
963 READ_SYNC);
964 if (err)
965 return err;
966 lock_page(page);
967 if (unlikely(!PageUptodate(page))) {
968 f2fs_put_page(page, 1);
969 return -EIO;
970 }
971 if (unlikely(page->mapping != mapping)) {
972 f2fs_put_page(page, 1);
973 goto repeat;
974 }
975 }
976 out:
977 SetPageUptodate(page);
978 clear_cold_data(page);
979 return 0;
980 }
981
982 static int f2fs_write_end(struct file *file,
983 struct address_space *mapping,
984 loff_t pos, unsigned len, unsigned copied,
985 struct page *page, void *fsdata)
986 {
987 struct inode *inode = page->mapping->host;
988
989 SetPageUptodate(page);
990 set_page_dirty(page);
991
992 if (pos + copied > i_size_read(inode)) {
993 i_size_write(inode, pos + copied);
994 mark_inode_dirty(inode);
995 update_inode_page(inode);
996 }
997
998 f2fs_put_page(page, 1);
999 return copied;
1000 }
1001
1002 static int check_direct_IO(struct inode *inode, int rw,
1003 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
1004 {
1005 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1006 int i;
1007
1008 if (rw == READ)
1009 return 0;
1010
1011 if (offset & blocksize_mask)
1012 return -EINVAL;
1013
1014 for (i = 0; i < nr_segs; i++)
1015 if (iov[i].iov_len & blocksize_mask)
1016 return -EINVAL;
1017 return 0;
1018 }
1019
1020 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1021 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
1022 {
1023 struct file *file = iocb->ki_filp;
1024 struct inode *inode = file->f_mapping->host;
1025
1026 /* Let buffer I/O handle the inline data case. */
1027 if (f2fs_has_inline_data(inode))
1028 return 0;
1029
1030 if (check_direct_IO(inode, rw, iov, offset, nr_segs))
1031 return 0;
1032
1033 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1034 get_data_block);
1035 }
1036
1037 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
1038 unsigned int length)
1039 {
1040 struct inode *inode = page->mapping->host;
1041 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1042 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
1043 dec_page_count(sbi, F2FS_DIRTY_DENTS);
1044 inode_dec_dirty_dents(inode);
1045 }
1046 ClearPagePrivate(page);
1047 }
1048
1049 static int f2fs_release_data_page(struct page *page, gfp_t wait)
1050 {
1051 ClearPagePrivate(page);
1052 return 1;
1053 }
1054
1055 static int f2fs_set_data_page_dirty(struct page *page)
1056 {
1057 struct address_space *mapping = page->mapping;
1058 struct inode *inode = mapping->host;
1059
1060 trace_f2fs_set_page_dirty(page, DATA);
1061
1062 SetPageUptodate(page);
1063 mark_inode_dirty(inode);
1064
1065 if (!PageDirty(page)) {
1066 __set_page_dirty_nobuffers(page);
1067 set_dirty_dir_page(inode, page);
1068 return 1;
1069 }
1070 return 0;
1071 }
1072
1073 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1074 {
1075 return generic_block_bmap(mapping, block, get_data_block);
1076 }
1077
1078 const struct address_space_operations f2fs_dblock_aops = {
1079 .readpage = f2fs_read_data_page,
1080 .readpages = f2fs_read_data_pages,
1081 .writepage = f2fs_write_data_page,
1082 .writepages = f2fs_write_data_pages,
1083 .write_begin = f2fs_write_begin,
1084 .write_end = f2fs_write_end,
1085 .set_page_dirty = f2fs_set_data_page_dirty,
1086 .invalidatepage = f2fs_invalidate_data_page,
1087 .releasepage = f2fs_release_data_page,
1088 .direct_IO = f2fs_direct_IO,
1089 .bmap = f2fs_bmap,
1090 };