projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: move all the bio initialization into __bio_alloc
[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 | fio->rw_flag;
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;
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, false);
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 sync_inode_page(dn);
253 return 0;
254 }
255
256 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
257 {
258 bool need_put = dn->inode_page ? false : true;
259 int err;
260
261 err = get_dnode_of_data(dn, index, ALLOC_NODE);
262 if (err)
263 return err;
264 if (dn->data_blkaddr == NULL_ADDR)
265 err = reserve_new_block(dn);
266
267 if (need_put)
268 f2fs_put_dnode(dn);
269 return err;
270 }
271
272 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
273 struct buffer_head *bh_result)
274 {
275 struct f2fs_inode_info *fi = F2FS_I(inode);
276 pgoff_t start_fofs, end_fofs;
277 block_t start_blkaddr;
278
279 if (is_inode_flag_set(fi, FI_NO_EXTENT))
280 return 0;
281
282 read_lock(&fi->ext.ext_lock);
283 if (fi->ext.len == 0) {
284 read_unlock(&fi->ext.ext_lock);
285 return 0;
286 }
287
288 stat_inc_total_hit(inode->i_sb);
289
290 start_fofs = fi->ext.fofs;
291 end_fofs = fi->ext.fofs + fi->ext.len - 1;
292 start_blkaddr = fi->ext.blk_addr;
293
294 if (pgofs >= start_fofs && pgofs <= end_fofs) {
295 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
296 size_t count;
297
298 clear_buffer_new(bh_result);
299 map_bh(bh_result, inode->i_sb,
300 start_blkaddr + pgofs - start_fofs);
301 count = end_fofs - pgofs + 1;
302 if (count < (UINT_MAX >> blkbits))
303 bh_result->b_size = (count << blkbits);
304 else
305 bh_result->b_size = UINT_MAX;
306
307 stat_inc_read_hit(inode->i_sb);
308 read_unlock(&fi->ext.ext_lock);
309 return 1;
310 }
311 read_unlock(&fi->ext.ext_lock);
312 return 0;
313 }
314
315 void update_extent_cache(block_t blk_addr, struct dnode_of_data *dn)
316 {
317 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
318 pgoff_t fofs, start_fofs, end_fofs;
319 block_t start_blkaddr, end_blkaddr;
320 int need_update = true;
321
322 f2fs_bug_on(blk_addr == NEW_ADDR);
323 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
324 dn->ofs_in_node;
325
326 /* Update the page address in the parent node */
327 __set_data_blkaddr(dn, blk_addr);
328
329 if (is_inode_flag_set(fi, FI_NO_EXTENT))
330 return;
331
332 write_lock(&fi->ext.ext_lock);
333
334 start_fofs = fi->ext.fofs;
335 end_fofs = fi->ext.fofs + fi->ext.len - 1;
336 start_blkaddr = fi->ext.blk_addr;
337 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
338
339 /* Drop and initialize the matched extent */
340 if (fi->ext.len == 1 && fofs == start_fofs)
341 fi->ext.len = 0;
342
343 /* Initial extent */
344 if (fi->ext.len == 0) {
345 if (blk_addr != NULL_ADDR) {
346 fi->ext.fofs = fofs;
347 fi->ext.blk_addr = blk_addr;
348 fi->ext.len = 1;
349 }
350 goto end_update;
351 }
352
353 /* Front merge */
354 if (fofs == start_fofs - 1 && blk_addr == start_blkaddr - 1) {
355 fi->ext.fofs--;
356 fi->ext.blk_addr--;
357 fi->ext.len++;
358 goto end_update;
359 }
360
361 /* Back merge */
362 if (fofs == end_fofs + 1 && blk_addr == end_blkaddr + 1) {
363 fi->ext.len++;
364 goto end_update;
365 }
366
367 /* Split the existing extent */
368 if (fi->ext.len > 1 &&
369 fofs >= start_fofs && fofs <= end_fofs) {
370 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
371 fi->ext.len = fofs - start_fofs;
372 } else {
373 fi->ext.fofs = fofs + 1;
374 fi->ext.blk_addr = start_blkaddr +
375 fofs - start_fofs + 1;
376 fi->ext.len -= fofs - start_fofs + 1;
377 }
378 } else {
379 need_update = false;
380 }
381
382 /* Finally, if the extent is very fragmented, let's drop the cache. */
383 if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
384 fi->ext.len = 0;
385 set_inode_flag(fi, FI_NO_EXTENT);
386 need_update = true;
387 }
388 end_update:
389 write_unlock(&fi->ext.ext_lock);
390 if (need_update)
391 sync_inode_page(dn);
392 return;
393 }
394
395 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
396 {
397 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
398 struct address_space *mapping = inode->i_mapping;
399 struct dnode_of_data dn;
400 struct page *page;
401 int err;
402
403 page = find_get_page(mapping, index);
404 if (page && PageUptodate(page))
405 return page;
406 f2fs_put_page(page, 0);
407
408 set_new_dnode(&dn, inode, NULL, NULL, 0);
409 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
410 if (err)
411 return ERR_PTR(err);
412 f2fs_put_dnode(&dn);
413
414 if (dn.data_blkaddr == NULL_ADDR)
415 return ERR_PTR(-ENOENT);
416
417 /* By fallocate(), there is no cached page, but with NEW_ADDR */
418 if (unlikely(dn.data_blkaddr == NEW_ADDR))
419 return ERR_PTR(-EINVAL);
420
421 page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
422 if (!page)
423 return ERR_PTR(-ENOMEM);
424
425 if (PageUptodate(page)) {
426 unlock_page(page);
427 return page;
428 }
429
430 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
431 sync ? READ_SYNC : READA);
432 if (err)
433 return ERR_PTR(err);
434
435 if (sync) {
436 wait_on_page_locked(page);
437 if (unlikely(!PageUptodate(page))) {
438 f2fs_put_page(page, 0);
439 return ERR_PTR(-EIO);
440 }
441 }
442 return page;
443 }
444
445 /*
446 * If it tries to access a hole, return an error.
447 * Because, the callers, functions in dir.c and GC, should be able to know
448 * whether this page exists or not.
449 */
450 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
451 {
452 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
453 struct address_space *mapping = inode->i_mapping;
454 struct dnode_of_data dn;
455 struct page *page;
456 int err;
457
458 repeat:
459 page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
460 if (!page)
461 return ERR_PTR(-ENOMEM);
462
463 set_new_dnode(&dn, inode, NULL, NULL, 0);
464 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
465 if (err) {
466 f2fs_put_page(page, 1);
467 return ERR_PTR(err);
468 }
469 f2fs_put_dnode(&dn);
470
471 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
472 f2fs_put_page(page, 1);
473 return ERR_PTR(-ENOENT);
474 }
475
476 if (PageUptodate(page))
477 return page;
478
479 /*
480 * A new dentry page is allocated but not able to be written, since its
481 * new inode page couldn't be allocated due to -ENOSPC.
482 * In such the case, its blkaddr can be remained as NEW_ADDR.
483 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
484 */
485 if (dn.data_blkaddr == NEW_ADDR) {
486 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
487 SetPageUptodate(page);
488 return page;
489 }
490
491 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr, READ_SYNC);
492 if (err)
493 return ERR_PTR(err);
494
495 lock_page(page);
496 if (unlikely(!PageUptodate(page))) {
497 f2fs_put_page(page, 1);
498 return ERR_PTR(-EIO);
499 }
500 if (unlikely(page->mapping != mapping)) {
501 f2fs_put_page(page, 1);
502 goto repeat;
503 }
504 return page;
505 }
506
507 /*
508 * Caller ensures that this data page is never allocated.
509 * A new zero-filled data page is allocated in the page cache.
510 *
511 * Also, caller should grab and release a mutex by calling mutex_lock_op() and
512 * mutex_unlock_op().
513 * Note that, npage is set only by make_empty_dir.
514 */
515 struct page *get_new_data_page(struct inode *inode,
516 struct page *npage, pgoff_t index, bool new_i_size)
517 {
518 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
519 struct address_space *mapping = inode->i_mapping;
520 struct page *page;
521 struct dnode_of_data dn;
522 int err;
523
524 set_new_dnode(&dn, inode, npage, npage, 0);
525 err = f2fs_reserve_block(&dn, index);
526 if (err)
527 return ERR_PTR(err);
528 repeat:
529 page = grab_cache_page(mapping, index);
530 if (!page)
531 return ERR_PTR(-ENOMEM);
532
533 if (PageUptodate(page))
534 return page;
535
536 if (dn.data_blkaddr == NEW_ADDR) {
537 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
538 SetPageUptodate(page);
539 } else {
540 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
541 READ_SYNC);
542 if (err)
543 return ERR_PTR(err);
544 lock_page(page);
545 if (unlikely(!PageUptodate(page))) {
546 f2fs_put_page(page, 1);
547 return ERR_PTR(-EIO);
548 }
549 if (unlikely(page->mapping != mapping)) {
550 f2fs_put_page(page, 1);
551 goto repeat;
552 }
553 }
554
555 if (new_i_size &&
556 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
557 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
558 /* Only the directory inode sets new_i_size */
559 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
560 mark_inode_dirty_sync(inode);
561 }
562 return page;
563 }
564
565 static int __allocate_data_block(struct dnode_of_data *dn)
566 {
567 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
568 struct f2fs_summary sum;
569 block_t new_blkaddr;
570 struct node_info ni;
571 int type;
572
573 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
574 return -EPERM;
575 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
576 return -ENOSPC;
577
578 __set_data_blkaddr(dn, NEW_ADDR);
579 dn->data_blkaddr = NEW_ADDR;
580
581 get_node_info(sbi, dn->nid, &ni);
582 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
583
584 type = CURSEG_WARM_DATA;
585
586 allocate_data_block(sbi, NULL, NULL_ADDR, &new_blkaddr, &sum, type);
587
588 /* direct IO doesn't use extent cache to maximize the performance */
589 set_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
590 update_extent_cache(new_blkaddr, dn);
591 clear_inode_flag(F2FS_I(dn->inode), FI_NO_EXTENT);
592
593 dn->data_blkaddr = new_blkaddr;
594 return 0;
595 }
596
597 /*
598 * This function should be used by the data read flow only where it
599 * does not check the "create" flag that indicates block allocation.
600 * The reason for this special functionality is to exploit VFS readahead
601 * mechanism.
602 */
603 static int get_data_block(struct inode *inode, sector_t iblock,
604 struct buffer_head *bh_result, int create)
605 {
606 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
607 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
608 unsigned maxblocks = bh_result->b_size >> blkbits;
609 struct dnode_of_data dn;
610 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
611 pgoff_t pgofs, end_offset;
612 int err = 0, ofs = 1;
613 bool allocated = false;
614
615 /* Get the page offset from the block offset(iblock) */
616 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
617
618 if (check_extent_cache(inode, pgofs, bh_result))
619 goto out;
620
621 if (create)
622 f2fs_lock_op(sbi);
623
624 /* When reading holes, we need its node page */
625 set_new_dnode(&dn, inode, NULL, NULL, 0);
626 err = get_dnode_of_data(&dn, pgofs, mode);
627 if (err || dn.data_blkaddr == NEW_ADDR) {
628 if (err == -ENOENT)
629 err = 0;
630 goto unlock_out;
631 }
632
633 if (dn.data_blkaddr != NULL_ADDR) {
634 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
635 } else if (create) {
636 err = __allocate_data_block(&dn);
637 if (err)
638 goto put_out;
639 allocated = true;
640 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
641 } else {
642 goto put_out;
643 }
644
645 end_offset = IS_INODE(dn.node_page) ?
646 ADDRS_PER_INODE(F2FS_I(inode)) : ADDRS_PER_BLOCK;
647 bh_result->b_size = (((size_t)1) << blkbits);
648 dn.ofs_in_node++;
649 pgofs++;
650
651 get_next:
652 if (dn.ofs_in_node >= end_offset) {
653 if (allocated)
654 sync_inode_page(&dn);
655 allocated = false;
656 f2fs_put_dnode(&dn);
657
658 set_new_dnode(&dn, inode, NULL, NULL, 0);
659 err = get_dnode_of_data(&dn, pgofs, mode);
660 if (err || dn.data_blkaddr == NEW_ADDR) {
661 if (err == -ENOENT)
662 err = 0;
663 goto unlock_out;
664 }
665 end_offset = IS_INODE(dn.node_page) ?
666 ADDRS_PER_INODE(F2FS_I(inode)) : ADDRS_PER_BLOCK;
667 }
668
669 if (maxblocks > (bh_result->b_size >> blkbits)) {
670 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
671 if (blkaddr == NULL_ADDR && create) {
672 err = __allocate_data_block(&dn);
673 if (err)
674 goto sync_out;
675 allocated = true;
676 blkaddr = dn.data_blkaddr;
677 }
678 /* Give more consecutive addresses for the read ahead */
679 if (blkaddr == (bh_result->b_blocknr + ofs)) {
680 ofs++;
681 dn.ofs_in_node++;
682 pgofs++;
683 bh_result->b_size += (((size_t)1) << blkbits);
684 goto get_next;
685 }
686 }
687 sync_out:
688 if (allocated)
689 sync_inode_page(&dn);
690 put_out:
691 f2fs_put_dnode(&dn);
692 unlock_out:
693 if (create)
694 f2fs_unlock_op(sbi);
695 out:
696 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
697 return err;
698 }
699
700 static int f2fs_read_data_page(struct file *file, struct page *page)
701 {
702 return mpage_readpage(page, get_data_block);
703 }
704
705 static int f2fs_read_data_pages(struct file *file,
706 struct address_space *mapping,
707 struct list_head *pages, unsigned nr_pages)
708 {
709 return mpage_readpages(mapping, pages, nr_pages, get_data_block);
710 }
711
712 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
713 {
714 struct inode *inode = page->mapping->host;
715 block_t old_blkaddr, new_blkaddr;
716 struct dnode_of_data dn;
717 int err = 0;
718
719 set_new_dnode(&dn, inode, NULL, NULL, 0);
720 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
721 if (err)
722 return err;
723
724 old_blkaddr = dn.data_blkaddr;
725
726 /* This page is already truncated */
727 if (old_blkaddr == NULL_ADDR)
728 goto out_writepage;
729
730 set_page_writeback(page);
731
732 /*
733 * If current allocation needs SSR,
734 * it had better in-place writes for updated data.
735 */
736 if (unlikely(old_blkaddr != NEW_ADDR &&
737 !is_cold_data(page) &&
738 need_inplace_update(inode))) {
739 rewrite_data_page(page, old_blkaddr, fio);
740 } else {
741 write_data_page(page, &dn, &new_blkaddr, fio);
742 update_extent_cache(new_blkaddr, &dn);
743 }
744 out_writepage:
745 f2fs_put_dnode(&dn);
746 return err;
747 }
748
749 static int f2fs_write_data_page(struct page *page,
750 struct writeback_control *wbc)
751 {
752 struct inode *inode = page->mapping->host;
753 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
754 loff_t i_size = i_size_read(inode);
755 const pgoff_t end_index = ((unsigned long long) i_size)
756 >> PAGE_CACHE_SHIFT;
757 unsigned offset;
758 bool need_balance_fs = false;
759 int err = 0;
760 struct f2fs_io_info fio = {
761 .type = DATA,
762 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC: WRITE,
763 .rw_flag = 0,
764 };
765
766 if (page->index < end_index)
767 goto write;
768
769 /*
770 * If the offset is out-of-range of file size,
771 * this page does not have to be written to disk.
772 */
773 offset = i_size & (PAGE_CACHE_SIZE - 1);
774 if ((page->index >= end_index + 1) || !offset) {
775 if (S_ISDIR(inode->i_mode)) {
776 dec_page_count(sbi, F2FS_DIRTY_DENTS);
777 inode_dec_dirty_dents(inode);
778 }
779 goto out;
780 }
781
782 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
783 write:
784 if (unlikely(sbi->por_doing)) {
785 err = AOP_WRITEPAGE_ACTIVATE;
786 goto redirty_out;
787 }
788
789 /* Dentry blocks are controlled by checkpoint */
790 if (S_ISDIR(inode->i_mode)) {
791 dec_page_count(sbi, F2FS_DIRTY_DENTS);
792 inode_dec_dirty_dents(inode);
793 err = do_write_data_page(page, &fio);
794 } else {
795 f2fs_lock_op(sbi);
796 err = do_write_data_page(page, &fio);
797 f2fs_unlock_op(sbi);
798 need_balance_fs = true;
799 }
800 if (err == -ENOENT)
801 goto out;
802 else if (err)
803 goto redirty_out;
804
805 if (wbc->for_reclaim)
806 f2fs_submit_merged_bio(sbi, DATA, WRITE);
807
808 clear_cold_data(page);
809 out:
810 unlock_page(page);
811 if (need_balance_fs)
812 f2fs_balance_fs(sbi);
813 return 0;
814
815 redirty_out:
816 wbc->pages_skipped++;
817 set_page_dirty(page);
818 return err;
819 }
820
821 #define MAX_DESIRED_PAGES_WP 4096
822
823 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
824 void *data)
825 {
826 struct address_space *mapping = data;
827 int ret = mapping->a_ops->writepage(page, wbc);
828 mapping_set_error(mapping, ret);
829 return ret;
830 }
831
832 static int f2fs_write_data_pages(struct address_space *mapping,
833 struct writeback_control *wbc)
834 {
835 struct inode *inode = mapping->host;
836 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
837 bool locked = false;
838 int ret;
839 long excess_nrtw = 0, desired_nrtw;
840
841 /* deal with chardevs and other special file */
842 if (!mapping->a_ops->writepage)
843 return 0;
844
845 if (wbc->nr_to_write < MAX_DESIRED_PAGES_WP) {
846 desired_nrtw = MAX_DESIRED_PAGES_WP;
847 excess_nrtw = desired_nrtw - wbc->nr_to_write;
848 wbc->nr_to_write = desired_nrtw;
849 }
850
851 if (!S_ISDIR(inode->i_mode)) {
852 mutex_lock(&sbi->writepages);
853 locked = true;
854 }
855 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
856 if (locked)
857 mutex_unlock(&sbi->writepages);
858
859 f2fs_submit_merged_bio(sbi, DATA, WRITE);
860
861 remove_dirty_dir_inode(inode);
862
863 wbc->nr_to_write -= excess_nrtw;
864 return ret;
865 }
866
867 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
868 loff_t pos, unsigned len, unsigned flags,
869 struct page **pagep, void **fsdata)
870 {
871 struct inode *inode = mapping->host;
872 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
873 struct page *page;
874 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
875 struct dnode_of_data dn;
876 int err = 0;
877
878 f2fs_balance_fs(sbi);
879 repeat:
880 page = grab_cache_page_write_begin(mapping, index, flags);
881 if (!page)
882 return -ENOMEM;
883 *pagep = page;
884
885 f2fs_lock_op(sbi);
886 set_new_dnode(&dn, inode, NULL, NULL, 0);
887 err = f2fs_reserve_block(&dn, index);
888 f2fs_unlock_op(sbi);
889
890 if (err) {
891 f2fs_put_page(page, 1);
892 return err;
893 }
894
895 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
896 return 0;
897
898 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
899 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
900 unsigned end = start + len;
901
902 /* Reading beyond i_size is simple: memset to zero */
903 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
904 goto out;
905 }
906
907 if (dn.data_blkaddr == NEW_ADDR) {
908 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
909 } else {
910 err = f2fs_submit_page_bio(sbi, page, dn.data_blkaddr,
911 READ_SYNC);
912 if (err)
913 return err;
914 lock_page(page);
915 if (unlikely(!PageUptodate(page))) {
916 f2fs_put_page(page, 1);
917 return -EIO;
918 }
919 if (unlikely(page->mapping != mapping)) {
920 f2fs_put_page(page, 1);
921 goto repeat;
922 }
923 }
924 out:
925 SetPageUptodate(page);
926 clear_cold_data(page);
927 return 0;
928 }
929
930 static int f2fs_write_end(struct file *file,
931 struct address_space *mapping,
932 loff_t pos, unsigned len, unsigned copied,
933 struct page *page, void *fsdata)
934 {
935 struct inode *inode = page->mapping->host;
936
937 SetPageUptodate(page);
938 set_page_dirty(page);
939
940 if (pos + copied > i_size_read(inode)) {
941 i_size_write(inode, pos + copied);
942 mark_inode_dirty(inode);
943 update_inode_page(inode);
944 }
945
946 f2fs_put_page(page, 1);
947 return copied;
948 }
949
950 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
951 const struct iovec *iov, loff_t offset, unsigned long nr_segs)
952 {
953 struct file *file = iocb->ki_filp;
954 struct inode *inode = file->f_mapping->host;
955 return blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
956 get_data_block);
957 }
958
959 static void f2fs_invalidate_data_page(struct page *page, unsigned int offset,
960 unsigned int length)
961 {
962 struct inode *inode = page->mapping->host;
963 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
964 if (S_ISDIR(inode->i_mode) && PageDirty(page)) {
965 dec_page_count(sbi, F2FS_DIRTY_DENTS);
966 inode_dec_dirty_dents(inode);
967 }
968 ClearPagePrivate(page);
969 }
970
971 static int f2fs_release_data_page(struct page *page, gfp_t wait)
972 {
973 ClearPagePrivate(page);
974 return 1;
975 }
976
977 static int f2fs_set_data_page_dirty(struct page *page)
978 {
979 struct address_space *mapping = page->mapping;
980 struct inode *inode = mapping->host;
981
982 trace_f2fs_set_page_dirty(page, DATA);
983
984 SetPageUptodate(page);
985 if (!PageDirty(page)) {
986 __set_page_dirty_nobuffers(page);
987 set_dirty_dir_page(inode, page);
988 return 1;
989 }
990 return 0;
991 }
992
993 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
994 {
995 return generic_block_bmap(mapping, block, get_data_block);
996 }
997
998 const struct address_space_operations f2fs_dblock_aops = {
999 .readpage = f2fs_read_data_page,
1000 .readpages = f2fs_read_data_pages,
1001 .writepage = f2fs_write_data_page,
1002 .writepages = f2fs_write_data_pages,
1003 .write_begin = f2fs_write_begin,
1004 .write_end = f2fs_write_end,
1005 .set_page_dirty = f2fs_set_data_page_dirty,
1006 .invalidatepage = f2fs_invalidate_data_page,
1007 .releasepage = f2fs_release_data_page,
1008 .direct_IO = f2fs_direct_IO,
1009 .bmap = f2fs_bmap,
1010 };