Loading...
Note: File does not exist in v3.5.6.
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
27static void f2fs_read_end_io(struct bio *bio, int err)
28{
29 struct bio_vec *bvec;
30 int i;
31
32 bio_for_each_segment_all(bvec, bio, i) {
33 struct page *page = bvec->bv_page;
34
35 if (!err) {
36 SetPageUptodate(page);
37 } else {
38 ClearPageUptodate(page);
39 SetPageError(page);
40 }
41 unlock_page(page);
42 }
43 bio_put(bio);
44}
45
46static void f2fs_write_end_io(struct bio *bio, int err)
47{
48 struct f2fs_sb_info *sbi = bio->bi_private;
49 struct bio_vec *bvec;
50 int i;
51
52 bio_for_each_segment_all(bvec, bio, i) {
53 struct page *page = bvec->bv_page;
54
55 if (unlikely(err)) {
56 SetPageError(page);
57 set_bit(AS_EIO, &page->mapping->flags);
58 f2fs_stop_checkpoint(sbi);
59 }
60 end_page_writeback(page);
61 dec_page_count(sbi, F2FS_WRITEBACK);
62 }
63
64 if (sbi->wait_io) {
65 complete(sbi->wait_io);
66 sbi->wait_io = NULL;
67 }
68
69 if (!get_pages(sbi, F2FS_WRITEBACK) &&
70 !list_empty(&sbi->cp_wait.task_list))
71 wake_up(&sbi->cp_wait);
72
73 bio_put(bio);
74}
75
76/*
77 * Low-level block read/write IO operations.
78 */
79static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
80 int npages, bool is_read)
81{
82 struct bio *bio;
83
84 /* No failure on bio allocation */
85 bio = bio_alloc(GFP_NOIO, npages);
86
87 bio->bi_bdev = sbi->sb->s_bdev;
88 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
89 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
90 bio->bi_private = sbi;
91
92 return bio;
93}
94
95static void __submit_merged_bio(struct f2fs_bio_info *io)
96{
97 struct f2fs_io_info *fio = &io->fio;
98 int rw;
99
100 if (!io->bio)
101 return;
102
103 rw = fio->rw;
104
105 if (is_read_io(rw)) {
106 trace_f2fs_submit_read_bio(io->sbi->sb, rw,
107 fio->type, io->bio);
108 submit_bio(rw, io->bio);
109 } else {
110 trace_f2fs_submit_write_bio(io->sbi->sb, rw,
111 fio->type, io->bio);
112 /*
113 * META_FLUSH is only from the checkpoint procedure, and we
114 * should wait this metadata bio for FS consistency.
115 */
116 if (fio->type == META_FLUSH) {
117 DECLARE_COMPLETION_ONSTACK(wait);
118 io->sbi->wait_io = &wait;
119 submit_bio(rw, io->bio);
120 wait_for_completion(&wait);
121 } else {
122 submit_bio(rw, io->bio);
123 }
124 }
125
126 io->bio = NULL;
127}
128
129void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
130 enum page_type type, int rw)
131{
132 enum page_type btype = PAGE_TYPE_OF_BIO(type);
133 struct f2fs_bio_info *io;
134
135 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
136
137 down_write(&io->io_rwsem);
138
139 /* change META to META_FLUSH in the checkpoint procedure */
140 if (type >= META_FLUSH) {
141 io->fio.type = META_FLUSH;
142 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
143 }
144 __submit_merged_bio(io);
145 up_write(&io->io_rwsem);
146}
147
148/*
149 * Fill the locked page with data located in the block address.
150 * Return unlocked page.
151 */
152int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
153 block_t blk_addr, int rw)
154{
155 struct bio *bio;
156
157 trace_f2fs_submit_page_bio(page, blk_addr, rw);
158
159 /* Allocate a new bio */
160 bio = __bio_alloc(sbi, blk_addr, 1, is_read_io(rw));
161
162 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
163 bio_put(bio);
164 f2fs_put_page(page, 1);
165 return -EFAULT;
166 }
167
168 submit_bio(rw, bio);
169 return 0;
170}
171
172void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
173 block_t blk_addr, struct f2fs_io_info *fio)
174{
175 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
176 struct f2fs_bio_info *io;
177 bool is_read = is_read_io(fio->rw);
178
179 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
180
181 verify_block_addr(sbi, blk_addr);
182
183 down_write(&io->io_rwsem);
184
185 if (!is_read)
186 inc_page_count(sbi, F2FS_WRITEBACK);
187
188 if (io->bio && (io->last_block_in_bio != blk_addr - 1 ||
189 io->fio.rw != fio->rw))
190 __submit_merged_bio(io);
191alloc_new:
192 if (io->bio == NULL) {
193 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
194
195 io->bio = __bio_alloc(sbi, blk_addr, bio_blocks, is_read);
196 io->fio = *fio;
197 }
198
199 if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
200 PAGE_CACHE_SIZE) {
201 __submit_merged_bio(io);
202 goto alloc_new;
203 }
204
205 io->last_block_in_bio = blk_addr;
206
207 up_write(&io->io_rwsem);
208 trace_f2fs_submit_page_mbio(page, fio->rw, fio->type, blk_addr);
209}
210
211/*
212 * Lock ordering for the change of data block address:
213 * ->data_page
214 * ->node_page
215 * update block addresses in the node page
216 */
217static void __set_data_blkaddr(struct dnode_of_data *dn, block_t new_addr)
218{
219 struct f2fs_node *rn;
220 __le32 *addr_array;
221 struct page *node_page = dn->node_page;
222 unsigned int ofs_in_node = dn->ofs_in_node;
223
224 f2fs_wait_on_page_writeback(node_page, NODE);
225
226 rn = F2FS_NODE(node_page);
227
228 /* Get physical address of data block */
229 addr_array = blkaddr_in_node(rn);
230 addr_array[ofs_in_node] = cpu_to_le32(new_addr);
231 set_page_dirty(node_page);
232}
233
234int reserve_new_block(struct dnode_of_data *dn)
235{
236 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
237
238 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
239 return -EPERM;
240 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
241 return -ENOSPC;
242
243 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
244
245 __set_data_blkaddr(dn, NEW_ADDR);
246 dn->data_blkaddr = NEW_ADDR;
247 mark_inode_dirty(dn->inode);
248 sync_inode_page(dn);
249 return 0;
250}
251
252int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
253{
254 bool need_put = dn->inode_page ? false : true;
255 int err;
256
257 /* if inode_page exists, index should be zero */
258 f2fs_bug_on(!need_put && index);
259
260 err = get_dnode_of_data(dn, index, ALLOC_NODE);
261 if (err)
262 return err;
263
264 if (dn->data_blkaddr == NULL_ADDR)
265 err = reserve_new_block(dn);
266 if (err || need_put)
267 f2fs_put_dnode(dn);
268 return err;
269}
270
271static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
272 struct buffer_head *bh_result)
273{
274 struct f2fs_inode_info *fi = F2FS_I(inode);
275 pgoff_t start_fofs, end_fofs;
276 block_t start_blkaddr;
277
278 if (is_inode_flag_set(fi, FI_NO_EXTENT))
279 return 0;
280
281 read_lock(&fi->ext.ext_lock);
282 if (fi->ext.len == 0) {
283 read_unlock(&fi->ext.ext_lock);
284 return 0;
285 }
286
287 stat_inc_total_hit(inode->i_sb);
288
289 start_fofs = fi->ext.fofs;
290 end_fofs = fi->ext.fofs + fi->ext.len - 1;
291 start_blkaddr = fi->ext.blk_addr;
292
293 if (pgofs >= start_fofs && pgofs <= end_fofs) {
294 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
295 size_t count;
296
297 clear_buffer_new(bh_result);
298 map_bh(bh_result, inode->i_sb,
299 start_blkaddr + pgofs - start_fofs);
300 count = end_fofs - pgofs + 1;
301 if (count < (UINT_MAX >> blkbits))
302 bh_result->b_size = (count << blkbits);
303 else
304 bh_result->b_size = UINT_MAX;
305
306 stat_inc_read_hit(inode->i_sb);
307 read_unlock(&fi->ext.ext_lock);
308 return 1;
309 }
310 read_unlock(&fi->ext.ext_lock);
311 return 0;
312}
313
314void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
315{
316 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
317 pgoff_t fofs, start_fofs, end_fofs;
318 block_t start_blkaddr, end_blkaddr;
319 int need_update = true;
320
321 f2fs_bug_on(blk_addr == NEW_ADDR);
322 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
323 dn->ofs_in_node;
324
325 /* Update the page address in the parent node */
326 __set_data_blkaddr(dn, blk_addr);
327
328 if (is_inode_flag_set(fi, FI_NO_EXTENT))
329 return;
330
331 write_lock(&fi->ext.ext_lock);
332
333 start_fofs = fi->ext.fofs;
334 end_fofs = fi->ext.fofs + fi->ext.len - 1;
335 start_blkaddr = fi->ext.blk_addr;
336 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
337
338 /* Drop and initialize the matched extent */
339 if (fi->ext.len == 1 && fofs == start_fofs)
340 fi->ext.len = 0;
341
342 /* Initial extent */
343 if (fi->ext.len == 0) {
344 if (blk_addr != NULL_ADDR) {
345 fi->ext.fofs = fofs;
346 fi->ext.blk_addr = blk_addr;
347 fi->ext.len = 1;
348 }
349 goto end_update;
350 }
351
352 /* Front merge */
353 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
354 fi->ext.fofs--;
355 fi->ext.blk_addr--;
356 fi->ext.len++;
357 goto end_update;
358 }
359
360 /* Back merge */
361 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
362 fi->ext.len++;
363 goto end_update;
364 }
365
366 /* Split the existing extent */
367 if (fi->ext.len > 1 &&
368 fofs >= start_fofs && fofs <= end_fofs) {
369 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
370 fi->ext.len = fofs - start_fofs;
371 } else {
372 fi->ext.fofs = fofs + 1;
373 fi->ext.blk_addr = start_blkaddr +
374 fofs - start_fofs + 1;
375 fi->ext.len -= fofs - start_fofs + 1;
376 }
377 } else {
378 need_update = false;
379 }
380
381 /* Finally, if the extent is very fragmented, let's drop the cache. */
382 if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
383 fi->ext.len = 0;
384 set_inode_flag(fi, FI_NO_EXTENT);
385 need_update = true;
386 }
387end_update:
388 write_unlock(&fi->ext.ext_lock);
389 if (need_update)
390 sync_inode_page(dn);
391 return;
392}
393
394struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
395{
396 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
397 struct address_space *mapping = inode->i_mapping;
398 struct dnode_of_data dn;
399 struct page *page;
400 int err;
401
402 page = find_get_page(mapping, index);
403 if (page && PageUptodate(page))
404 return page;
405 f2fs_put_page(page, 0);
406
407 set_new_dnode(&dn, inode, NULL, NULL, 0);
408 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
409 if (err)
410 return ERR_PTR(err);
411 f2fs_put_dnode(&dn);
412
413 if (dn.data_blkaddr == NULL_ADDR)
414 return ERR_PTR(-ENOENT);
415
416 /* By fallocate(), there is no cached page, but with NEW_ADDR */
417 if (unlikely(dn.data_blkaddr == NEW_ADDR))
418 return ERR_PTR(-EINVAL);
419
420 page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
421 if (!page)
422 return ERR_PTR(-ENOMEM);
423
424 if (PageUptodate(page)) {
425 unlock_page(page);
426 return page;
427 }
428
429 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
430 sync ? READ_SYNC : READA);
431 if (err)
432 return ERR_PTR(err);
433
434 if (sync) {
435 wait_on_page_locked(page);
436 if (unlikely(!PageUptodate(page))) {
437 f2fs_put_page(page, 0);
438 return ERR_PTR(-EIO);
439 }
440 }
441 return page;
442}
443
444/*
445 * If it tries to access a hole, return an error.
446 * Because, the callers, functions in dir.c and GC, should be able to know
447 * whether this page exists or not.
448 */
449struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
450{
451 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
452 struct address_space *mapping = inode->i_mapping;
453 struct dnode_of_data dn;
454 struct page *page;
455 int err;
456
457repeat:
458 page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
459 if (!page)
460 return ERR_PTR(-ENOMEM);
461
462 set_new_dnode(&dn, inode, NULL, NULL, 0);
463 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
464 if (err) {
465 f2fs_put_page(page, 1);
466 return ERR_PTR(err);
467 }
468 f2fs_put_dnode(&dn);
469
470 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
471 f2fs_put_page(page, 1);
472 return ERR_PTR(-ENOENT);
473 }
474
475 if (PageUptodate(page))
476 return page;
477
478 /*
479 * A new dentry page is allocated but not able to be written, since its
480 * new inode page couldn't be allocated due to -ENOSPC.
481 * In such the case, its blkaddr can be remained as NEW_ADDR.
482 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
483 */
484 if (dn.data_blkaddr == NEW_ADDR) {
485 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
486 SetPageUptodate(page);
487 return page;
488 }
489
490 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, READ_SYNC);
491 if (err)
492 return ERR_PTR(err);
493
494 lock_page(page);
495 if (unlikely(!PageUptodate(page))) {
496 f2fs_put_page(page, 1);
497 return ERR_PTR(-EIO);
498 }
499 if (unlikely(page->mapping != mapping)) {
500 f2fs_put_page(page, 1);
501 goto repeat;
502 }
503 return page;
504}
505
506/*
507 * Caller ensures that this data page is never allocated.
508 * A new zero-filled data page is allocated in the page cache.
509 *
510 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
511 * f2fs_unlock_op().
512 * Note that, ipage is set only by make_empty_dir.
513 */
514struct page *get_new_data_page(struct inode *inode,
515 struct page *ipage, pgoff_t index, bool new_i_size)
516{
517 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
518 struct address_space *mapping = inode->i_mapping;
519 struct page *page;
520 struct dnode_of_data dn;
521 int err;
522
523 set_new_dnode(&dn, inode, ipage, NULL, 0);
524 err = f2fs_reserve_block(&dn, index);
525 if (err)
526 return ERR_PTR(err);
527repeat:
528 page = grab_cache_page(mapping, index);
529 if (!page) {
530 err = -ENOMEM;
531 goto put_err;
532 }
533
534 if (PageUptodate(page))
535 return page;
536
537 if (dn.data_blkaddr == NEW_ADDR) {
538 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
539 SetPageUptodate(page);
540 } else {
541 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
542 READ_SYNC);
543 if (err)
544 goto put_err;
545
546 lock_page(page);
547 if (unlikely(!PageUptodate(page))) {
548 f2fs_put_page(page, 1);
549 err = -EIO;
550 goto put_err;
551 }
552 if (unlikely(page->mapping != mapping)) {
553 f2fs_put_page(page, 1);
554 goto repeat;
555 }
556 }
557
558 if (new_i_size &&
559 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
560 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
561 /* Only the directory inode sets new_i_size */
562 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
563 }
564 return page;
565
566put_err:
567 f2fs_put_dnode(&dn);
568 return ERR_PTR(err);
569}
570
571static int __allocate_data_block(struct dnode_of_data *dn)
572{
573 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
574 struct f2fs_summary sum;
575 block_t new_blkaddr;
576 struct node_info ni;
577 int type;
578
579 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
580 return -EPERM;
581 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
582 return -ENOSPC;
583
584 __set_data_blkaddr(dn, NEW_ADDR);
585 dn->data_blkaddr = NEW_ADDR;
586
587 get_node_info(sbi, dn->nid, &ni);
588 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
589
590 type = CURSEG_WARM_DATA;
591
592 allocate_data_block(sbi, NULL, NULL_ADDR, &new_blkaddr, &sum, type);
593
594 /* direct IO doesn't use extent cache to maximize the performance */
595 set_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
596 update_extent_cache(new_blkaddr, dn);
597 clear_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
598
599 dn->data_blkaddr = new_blkaddr;
600 return 0;
601}
602
603/*
604 * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
605 * If original data blocks are allocated, then give them to blockdev.
606 * Otherwise,
607 * a. preallocate requested block addresses
608 * b. do not use extent cache for better performance
609 * c. give the block addresses to blockdev
610 */
611static int get_data_block(struct inode *inode, sector_t iblock,
612 struct buffer_head *bh_result, int create)
613{
614 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
615 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
616 unsigned maxblocks = bh_result->b_size >> blkbits;
617 struct dnode_of_data dn;
618 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
619 pgoff_t pgofs, end_offset;
620 int err = 0, ofs = 1;
621 bool allocated = false;
622
623 /* Get the page offset from the block offset(iblock) */
624 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
625
626 if (check_extent_cache(inode, pgofs, bh_result))
627 goto out;
628
629 if (create)
630 f2fs_lock_op(sbi);
631
632 /* When reading holes, we need its node page */
633 set_new_dnode(&dn, inode, NULL, NULL, 0);
634 err = get_dnode_of_data(&dn, pgofs, mode);
635 if (err) {
636 if (err == -ENOENT)
637 err = 0;
638 goto unlock_out;
639 }
640 if (dn.data_blkaddr == NEW_ADDR)
641 goto put_out;
642
643 if (dn.data_blkaddr != NULL_ADDR) {
644 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
645 } else if (create) {
646 err = __allocate_data_block(&dn);
647 if (err)
648 goto put_out;
649 allocated = true;
650 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
651 } else {
652 goto put_out;
653 }
654
655 end_offset = IS_INODE(dn.node_page) ?
656 ADDRS_PER_INODE(F2FS_I(inode)) : ADDRS_PER_BLOCK;
657 bh_result->b_size = (((size_t)1) << blkbits);
658 dn.ofs_in_node++;
659 pgofs++;
660
661get_next:
662 if (dn.ofs_in_node >= end_offset) {
663 if (allocated)
664 sync_inode_page(&dn);
665 allocated = false;
666 f2fs_put_dnode(&dn);
667
668 set_new_dnode(&dn, inode, NULL, NULL, 0);
669 err = get_dnode_of_data(&dn, pgofs, mode);
670 if (err) {
671 if (err == -ENOENT)
672 err = 0;
673 goto unlock_out;
674 }
675 if (dn.data_blkaddr == NEW_ADDR)
676 goto put_out;
677
678 end_offset = IS_INODE(dn.node_page) ?
679 ADDRS_PER_INODE(F2FS_I(inode)) : ADDRS_PER_BLOCK;
680 }
681
682 if (maxblocks > (bh_result->b_size >> blkbits)) {
683 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
684 if (blkaddr == NULL_ADDR && create) {
685 err = __allocate_data_block(&dn);
686 if (err)
687 goto sync_out;
688 allocated = true;
689 blkaddr = dn.data_blkaddr;
690 }
691 /* Give more consecutive addresses for the read ahead */
692 if (blkaddr == (bh_result->b_blocknr + ofs)) {
693 ofs++;
694 dn.ofs_in_node++;
695 pgofs++;
696 bh_result->b_size += (((size_t)1) << blkbits);
697 goto get_next;
698 }
699 }
700sync_out:
701 if (allocated)
702 sync_inode_page(&dn);
703put_out:
704 f2fs_put_dnode(&dn);
705unlock_out:
706 if (create)
707 f2fs_unlock_op(sbi);
708out:
709 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
710 return err;
711}
712
713static int f2fs_read_data_page(struct file *file, struct page *page)
714{
715 struct inode *inode = page->mapping->host;
716 int ret;
717
718 /* If the file has inline data, try to read it directlly */
719 if (f2fs_has_inline_data(inode))
720 ret = f2fs_read_inline_data(inode, page);
721 else
722 ret = mpage_readpage(page, get_data_block);
723
724 return ret;
725}
726
727static int f2fs_read_data_pages(struct file *file,
728 struct address_space *mapping,
729 struct list_head *pages, unsigned nr_pages)
730{
731 struct inode *inode = file->f_mapping->host;
732
733 /* If the file has inline data, skip readpages */
734 if (f2fs_has_inline_data(inode))
735 return 0;
736
737 return mpage_readpages(mapping, pages, nr_pages, get_data_block);
738}
739
740int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
741{
742 struct inode *inode = page->mapping->host;
743 block_t old_blkaddr, new_blkaddr;
744 struct dnode_of_data dn;
745 int err = 0;
746
747 set_new_dnode(&dn, inode, NULL, NULL, 0);
748 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
749 if (err)
750 return err;
751
752 old_blkaddr = dn.data_blkaddr;
753
754 /* This page is already truncated */
755 if (old_blkaddr == NULL_ADDR)
756 goto out_writepage;
757
758 set_page_writeback(page);
759
760 /*
761 * If current allocation needs SSR,
762 * it had better in-place writes for updated data.
763 */
764 if (unlikely(old_blkaddr != NEW_ADDR &&
765 !is_cold_data(page) &&
766 need_inplace_update(inode))) {
767 rewrite_data_page(page, old_blkaddr, fio);
768 } else {
769 write_data_page(page, &dn, &new_blkaddr, fio);
770 update_extent_cache(new_blkaddr, &dn);
771 }
772out_writepage:
773 f2fs_put_dnode(&dn);
774 return err;
775}
776
777static int f2fs_write_data_page(struct page *page,
778 struct writeback_control *wbc)
779{
780 struct inode *inode = page->mapping->host;
781 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
782 loff_t i_size = i_size_read(inode);
783 const pgoff_t end_index = ((unsigned long long) i_size)
784 >> PAGE_CACHE_SHIFT;
785 unsigned offset = 0;
786 bool need_balance_fs = false;
787 int err = 0;
788 struct f2fs_io_info fio = {
789 .type = DATA,
790 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
791 };
792
793 if (page->index < end_index)
794 goto write;
795
796 /*
797 * If the offset is out-of-range of file size,
798 * this page does not have to be written to disk.
799 */
800 offset = i_size & (PAGE_CACHE_SIZE - 1);
801 if ((page->index >= end_index + 1) || !offset) {
802 inode_dec_dirty_dents(inode);
803 goto out;
804 }
805
806 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
807write:
808 if (unlikely(sbi->por_doing))
809 goto redirty_out;
810
811 /* Dentry blocks are controlled by checkpoint */
812 if (S_ISDIR(inode->i_mode)) {
813 inode_dec_dirty_dents(inode);
814 err = do_write_data_page(page, &fio);
815 goto done;
816 }
817
818 if (!wbc->for_reclaim)
819 need_balance_fs = true;
820 else if (has_not_enough_free_secs(sbi, 0))
821 goto redirty_out;
822
823 f2fs_lock_op(sbi);
824 if (f2fs_has_inline_data(inode) || f2fs_may_inline(inode))
825 err = f2fs_write_inline_data(inode, page, offset);
826 else
827 err = do_write_data_page(page, &fio);
828 f2fs_unlock_op(sbi);
829done:
830 if (err && err != -ENOENT)
831 goto redirty_out;
832
833 clear_cold_data(page);
834out:
835 unlock_page(page);
836 if (need_balance_fs)
837 f2fs_balance_fs(sbi);
838 return 0;
839
840redirty_out:
841 wbc->pages_skipped++;
842 account_page_redirty(page);
843 set_page_dirty(page);
844 return AOP_WRITEPAGE_ACTIVATE;
845}
846
847static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
848 void *data)
849{
850 struct address_space *mapping = data;
851 int ret = mapping->a_ops->writepage(page, wbc);
852 mapping_set_error(mapping, ret);
853 return ret;
854}
855
856static int f2fs_write_data_pages(struct address_space *mapping,
857 struct writeback_control *wbc)
858{
859 struct inode *inode = mapping->host;
860 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
861 bool locked = false;
862 int ret;
863 long diff;
864
865 /* deal with chardevs and other special file */
866 if (!mapping->a_ops->writepage)
867 return 0;
868
869 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
870 get_dirty_dents(inode) < nr_pages_to_skip(sbi, DATA))
871 goto skip_write;
872
873 diff = nr_pages_to_write(sbi, DATA, wbc);
874
875 if (!S_ISDIR(inode->i_mode)) {
876 mutex_lock(&sbi->writepages);
877 locked = true;
878 }
879 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
880 if (locked)
881 mutex_unlock(&sbi->writepages);
882
883 f2fs_submit_merged_bio(sbi, DATA, WRITE);
884
885 remove_dirty_dir_inode(inode);
886
887 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
888 return ret;
889
890skip_write:
891 wbc->pages_skipped += get_dirty_dents(inode);
892 return 0;
893}
894
895static int f2fs_write_begin(struct file *file, struct address_space *mapping,
896 loff_t pos, unsigned len, unsigned flags,
897 struct page **pagep, void **fsdata)
898{
899 struct inode *inode = mapping->host;
900 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
901 struct page *page;
902 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
903 struct dnode_of_data dn;
904 int err = 0;
905
906 f2fs_balance_fs(sbi);
907repeat:
908 err = f2fs_convert_inline_data(inode, pos + len);
909 if (err)
910 return err;
911
912 page = grab_cache_page_write_begin(mapping, index, flags);
913 if (!page)
914 return -ENOMEM;
915 *pagep = page;
916
917 if (f2fs_has_inline_data(inode) && (pos + len) <= MAX_INLINE_DATA)
918 goto inline_data;
919
920 f2fs_lock_op(sbi);
921 set_new_dnode(&dn, inode, NULL, NULL, 0);
922 err = f2fs_reserve_block(&dn, index);
923 f2fs_unlock_op(sbi);
924
925 if (err) {
926 f2fs_put_page(page, 1);
927 return err;
928 }
929inline_data:
930 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
931 return 0;
932
933 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
934 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
935 unsigned end = start + len;
936
937 /* Reading beyond i_size is simple: memset to zero */
938 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
939 goto out;
940 }
941
942 if (dn.data_blkaddr == NEW_ADDR) {
943 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
944 } else {
945 if (f2fs_has_inline_data(inode)) {
946 err = f2fs_read_inline_data(inode, page);
947 if (err) {
948 page_cache_release(page);
949 return err;
950 }
951 } else {
952 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
953 READ_SYNC);
954 if (err)
955 return err;
956 }
957
958 lock_page(page);
959 if (unlikely(!PageUptodate(page))) {
960 f2fs_put_page(page, 1);
961 return -EIO;
962 }
963 if (unlikely(page->mapping != mapping)) {
964 f2fs_put_page(page, 1);
965 goto repeat;
966 }
967 }
968out:
969 SetPageUptodate(page);
970 clear_cold_data(page);
971 return 0;
972}
973
974static int f2fs_write_end(struct file *file,
975 struct address_space *mapping,
976 loff_t pos, unsigned len, unsigned copied,
977 struct page *page, void *fsdata)
978{
979 struct inode *inode = page->mapping->host;
980
981 SetPageUptodate(page);
982 set_page_dirty(page);
983
984 if (pos + copied > i_size_read(inode)) {
985 i_size_write(inode, pos + copied);
986 mark_inode_dirty(inode);
987 update_inode_page(inode);
988 }
989
990 f2fs_put_page(page, 1);
991 return copied;
992}
993
994static int check_direct_IO(struct inode *inode, int rw,
995 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
996{
997 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
998 int i;
999
1000 if (rw == READ)
1001 return 0;
1002
1003 if (offset & blocksize_mask)
1004 return -EINVAL;
1005
1006 for (i = 0; i < nr_segs; i++)
1007 if (iov[i].iov_len & blocksize_mask)
1008 return -EINVAL;
1009 return 0;
1010}
1011
1012static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1013 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
1014{
1015 struct file *file = iocb->ki_filp;
1016 struct inode *inode = file->f_mapping->host;
1017
1018 /* Let buffer I/O handle the inline data case. */
1019 if (f2fs_has_inline_data(inode))
1020 return 0;
1021
1022 if (check_direct_IO(inode, rw, iov, offset, nr_segs))
1023 return 0;
1024
1025 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1026 get_data_block);
1027}
1028
1029static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
1030 unsigned int length)
1031{
1032 struct inode *inode = page->mapping->host;
1033 if (PageDirty(page))
1034 inode_dec_dirty_dents(inode);
1035 ClearPagePrivate(page);
1036}
1037
1038static int f2fs_release_data_page(struct page *page, gfp_t wait)
1039{
1040 ClearPagePrivate(page);
1041 return 1;
1042}
1043
1044static int f2fs_set_data_page_dirty(struct page *page)
1045{
1046 struct address_space *mapping = page->mapping;
1047 struct inode *inode = mapping->host;
1048
1049 trace_f2fs_set_page_dirty(page, DATA);
1050
1051 SetPageUptodate(page);
1052 mark_inode_dirty(inode);
1053
1054 if (!PageDirty(page)) {
1055 __set_page_dirty_nobuffers(page);
1056 set_dirty_dir_page(inode, page);
1057 return 1;
1058 }
1059 return 0;
1060}
1061
1062static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1063{
1064 return generic_block_bmap(mapping, block, get_data_block);
1065}
1066
1067const struct address_space_operations f2fs_dblock_aops = {
1068 .readpage = f2fs_read_data_page,
1069 .readpages = f2fs_read_data_pages,
1070 .writepage = f2fs_write_data_page,
1071 .writepages = f2fs_write_data_pages,
1072 .write_begin = f2fs_write_begin,
1073 .write_end = f2fs_write_end,
1074 .set_page_dirty = f2fs_set_data_page_dirty,
1075 .invalidatepage = f2fs_invalidate_data_page,
1076 .releasepage = f2fs_release_data_page,
1077 .direct_IO = f2fs_direct_IO,
1078 .bmap = f2fs_bmap,
1079};