projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: call set_buffer_new for get_block
[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.h"
26 #include <trace/events/f2fs.h>
27
28 static void f2fs_read_end_io(struct bio *bio, int err)
29 {
30 struct bio_vec *bvec;
31 int i;
32
33 bio_for_each_segment_all(bvec, bio, i) {
34 struct page *page = bvec->bv_page;
35
36 if (!err) {
37 SetPageUptodate(page);
38 } else {
39 ClearPageUptodate(page);
40 SetPageError(page);
41 }
42 unlock_page(page);
43 }
44 bio_put(bio);
45 }
46
47 static void f2fs_write_end_io(struct bio *bio, int err)
48 {
49 struct f2fs_sb_info *sbi = bio->bi_private;
50 struct bio_vec *bvec;
51 int i;
52
53 bio_for_each_segment_all(bvec, bio, i) {
54 struct page *page = bvec->bv_page;
55
56 if (unlikely(err)) {
57 set_page_dirty(page);
58 set_bit(AS_EIO, &page->mapping->flags);
59 f2fs_stop_checkpoint(sbi);
60 }
61 end_page_writeback(page);
62 dec_page_count(sbi, F2FS_WRITEBACK);
63 }
64
65 if (!get_pages(sbi, F2FS_WRITEBACK) &&
66 !list_empty(&sbi->cp_wait.task_list))
67 wake_up(&sbi->cp_wait);
68
69 bio_put(bio);
70 }
71
72 /*
73 * Low-level block read/write IO operations.
74 */
75 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
76 int npages, bool is_read)
77 {
78 struct bio *bio;
79
80 /* No failure on bio allocation */
81 bio = bio_alloc(GFP_NOIO, npages);
82
83 bio->bi_bdev = sbi->sb->s_bdev;
84 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
85 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
86 bio->bi_private = sbi;
87
88 return bio;
89 }
90
91 static void __submit_merged_bio(struct f2fs_bio_info *io)
92 {
93 struct f2fs_io_info *fio = &io->fio;
94
95 if (!io->bio)
96 return;
97
98 if (is_read_io(fio->rw))
99 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
100 else
101 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
102
103 submit_bio(fio->rw, io->bio);
104 io->bio = NULL;
105 }
106
107 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
108 enum page_type type, int rw)
109 {
110 enum page_type btype = PAGE_TYPE_OF_BIO(type);
111 struct f2fs_bio_info *io;
112
113 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
114
115 down_write(&io->io_rwsem);
116
117 /* change META to META_FLUSH in the checkpoint procedure */
118 if (type >= META_FLUSH) {
119 io->fio.type = META_FLUSH;
120 if (test_opt(sbi, NOBARRIER))
121 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
122 else
123 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
124 }
125 __submit_merged_bio(io);
126 up_write(&io->io_rwsem);
127 }
128
129 /*
130 * Fill the locked page with data located in the block address.
131 * Return unlocked page.
132 */
133 int f2fs_submit_page_bio(struct f2fs_sb_info *sbi, struct page *page,
134 struct f2fs_io_info *fio)
135 {
136 struct bio *bio;
137
138 trace_f2fs_submit_page_bio(page, fio);
139 f2fs_trace_ios(page, fio, 0);
140
141 /* Allocate a new bio */
142 bio = __bio_alloc(sbi, fio->blk_addr, 1, is_read_io(fio->rw));
143
144 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
145 bio_put(bio);
146 f2fs_put_page(page, 1);
147 return -EFAULT;
148 }
149
150 submit_bio(fio->rw, bio);
151 return 0;
152 }
153
154 void f2fs_submit_page_mbio(struct f2fs_sb_info *sbi, struct page *page,
155 struct f2fs_io_info *fio)
156 {
157 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
158 struct f2fs_bio_info *io;
159 bool is_read = is_read_io(fio->rw);
160
161 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
162
163 verify_block_addr(sbi, fio->blk_addr);
164
165 down_write(&io->io_rwsem);
166
167 if (!is_read)
168 inc_page_count(sbi, F2FS_WRITEBACK);
169
170 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
171 io->fio.rw != fio->rw))
172 __submit_merged_bio(io);
173 alloc_new:
174 if (io->bio == NULL) {
175 int bio_blocks = MAX_BIO_BLOCKS(sbi);
176
177 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
178 io->fio = *fio;
179 }
180
181 if (bio_add_page(io->bio, page, PAGE_CACHE_SIZE, 0) <
182 PAGE_CACHE_SIZE) {
183 __submit_merged_bio(io);
184 goto alloc_new;
185 }
186
187 io->last_block_in_bio = fio->blk_addr;
188 f2fs_trace_ios(page, fio, 0);
189
190 up_write(&io->io_rwsem);
191 trace_f2fs_submit_page_mbio(page, fio);
192 }
193
194 /*
195 * Lock ordering for the change of data block address:
196 * ->data_page
197 * ->node_page
198 * update block addresses in the node page
199 */
200 static void __set_data_blkaddr(struct dnode_of_data *dn)
201 {
202 struct f2fs_node *rn;
203 __le32 *addr_array;
204 struct page *node_page = dn->node_page;
205 unsigned int ofs_in_node = dn->ofs_in_node;
206
207 f2fs_wait_on_page_writeback(node_page, NODE);
208
209 rn = F2FS_NODE(node_page);
210
211 /* Get physical address of data block */
212 addr_array = blkaddr_in_node(rn);
213 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
214 set_page_dirty(node_page);
215 }
216
217 int reserve_new_block(struct dnode_of_data *dn)
218 {
219 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
220
221 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
222 return -EPERM;
223 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
224 return -ENOSPC;
225
226 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
227
228 dn->data_blkaddr = NEW_ADDR;
229 __set_data_blkaddr(dn);
230 mark_inode_dirty(dn->inode);
231 sync_inode_page(dn);
232 return 0;
233 }
234
235 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
236 {
237 bool need_put = dn->inode_page ? false : true;
238 int err;
239
240 err = get_dnode_of_data(dn, index, ALLOC_NODE);
241 if (err)
242 return err;
243
244 if (dn->data_blkaddr == NULL_ADDR)
245 err = reserve_new_block(dn);
246 if (err || need_put)
247 f2fs_put_dnode(dn);
248 return err;
249 }
250
251 static int check_extent_cache(struct inode *inode, pgoff_t pgofs,
252 struct buffer_head *bh_result)
253 {
254 struct f2fs_inode_info *fi = F2FS_I(inode);
255 pgoff_t start_fofs, end_fofs;
256 block_t start_blkaddr;
257
258 if (is_inode_flag_set(fi, FI_NO_EXTENT))
259 return 0;
260
261 read_lock(&fi->ext.ext_lock);
262 if (fi->ext.len == 0) {
263 read_unlock(&fi->ext.ext_lock);
264 return 0;
265 }
266
267 stat_inc_total_hit(inode->i_sb);
268
269 start_fofs = fi->ext.fofs;
270 end_fofs = fi->ext.fofs + fi->ext.len - 1;
271 start_blkaddr = fi->ext.blk_addr;
272
273 if (pgofs >= start_fofs && pgofs <= end_fofs) {
274 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
275 size_t count;
276
277 set_buffer_new(bh_result);
278 map_bh(bh_result, inode->i_sb,
279 start_blkaddr + pgofs - start_fofs);
280 count = end_fofs - pgofs + 1;
281 if (count < (UINT_MAX >> blkbits))
282 bh_result->b_size = (count << blkbits);
283 else
284 bh_result->b_size = UINT_MAX;
285
286 stat_inc_read_hit(inode->i_sb);
287 read_unlock(&fi->ext.ext_lock);
288 return 1;
289 }
290 read_unlock(&fi->ext.ext_lock);
291 return 0;
292 }
293
294 void update_extent_cache(struct dnode_of_data *dn)
295 {
296 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
297 pgoff_t fofs, start_fofs, end_fofs;
298 block_t start_blkaddr, end_blkaddr;
299 int need_update = true;
300
301 f2fs_bug_on(F2FS_I_SB(dn->inode), dn->data_blkaddr == NEW_ADDR);
302
303 /* Update the page address in the parent node */
304 __set_data_blkaddr(dn);
305
306 if (is_inode_flag_set(fi, FI_NO_EXTENT))
307 return;
308
309 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
310 dn->ofs_in_node;
311
312 write_lock(&fi->ext.ext_lock);
313
314 start_fofs = fi->ext.fofs;
315 end_fofs = fi->ext.fofs + fi->ext.len - 1;
316 start_blkaddr = fi->ext.blk_addr;
317 end_blkaddr = fi->ext.blk_addr + fi->ext.len - 1;
318
319 /* Drop and initialize the matched extent */
320 if (fi->ext.len == 1 && fofs == start_fofs)
321 fi->ext.len = 0;
322
323 /* Initial extent */
324 if (fi->ext.len == 0) {
325 if (dn->data_blkaddr != NULL_ADDR) {
326 fi->ext.fofs = fofs;
327 fi->ext.blk_addr = dn->data_blkaddr;
328 fi->ext.len = 1;
329 }
330 goto end_update;
331 }
332
333 /* Front merge */
334 if (fofs == start_fofs - 1 && dn->data_blkaddr == start_blkaddr - 1) {
335 fi->ext.fofs--;
336 fi->ext.blk_addr--;
337 fi->ext.len++;
338 goto end_update;
339 }
340
341 /* Back merge */
342 if (fofs == end_fofs + 1 && dn->data_blkaddr == end_blkaddr + 1) {
343 fi->ext.len++;
344 goto end_update;
345 }
346
347 /* Split the existing extent */
348 if (fi->ext.len > 1 &&
349 fofs >= start_fofs && fofs <= end_fofs) {
350 if ((end_fofs - fofs) < (fi->ext.len >> 1)) {
351 fi->ext.len = fofs - start_fofs;
352 } else {
353 fi->ext.fofs = fofs + 1;
354 fi->ext.blk_addr = start_blkaddr +
355 fofs - start_fofs + 1;
356 fi->ext.len -= fofs - start_fofs + 1;
357 }
358 } else {
359 need_update = false;
360 }
361
362 /* Finally, if the extent is very fragmented, let's drop the cache. */
363 if (fi->ext.len < F2FS_MIN_EXTENT_LEN) {
364 fi->ext.len = 0;
365 set_inode_flag(fi, FI_NO_EXTENT);
366 need_update = true;
367 }
368 end_update:
369 write_unlock(&fi->ext.ext_lock);
370 if (need_update)
371 sync_inode_page(dn);
372 return;
373 }
374
375 struct page *find_data_page(struct inode *inode, pgoff_t index, bool sync)
376 {
377 struct address_space *mapping = inode->i_mapping;
378 struct dnode_of_data dn;
379 struct page *page;
380 int err;
381 struct f2fs_io_info fio = {
382 .type = DATA,
383 .rw = sync ? READ_SYNC : READA,
384 };
385
386 page = find_get_page(mapping, index);
387 if (page && PageUptodate(page))
388 return page;
389 f2fs_put_page(page, 0);
390
391 set_new_dnode(&dn, inode, NULL, NULL, 0);
392 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
393 if (err)
394 return ERR_PTR(err);
395 f2fs_put_dnode(&dn);
396
397 if (dn.data_blkaddr == NULL_ADDR)
398 return ERR_PTR(-ENOENT);
399
400 /* By fallocate(), there is no cached page, but with NEW_ADDR */
401 if (unlikely(dn.data_blkaddr == NEW_ADDR))
402 return ERR_PTR(-EINVAL);
403
404 page = grab_cache_page(mapping, index);
405 if (!page)
406 return ERR_PTR(-ENOMEM);
407
408 if (PageUptodate(page)) {
409 unlock_page(page);
410 return page;
411 }
412
413 fio.blk_addr = dn.data_blkaddr;
414 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
415 if (err)
416 return ERR_PTR(err);
417
418 if (sync) {
419 wait_on_page_locked(page);
420 if (unlikely(!PageUptodate(page))) {
421 f2fs_put_page(page, 0);
422 return ERR_PTR(-EIO);
423 }
424 }
425 return page;
426 }
427
428 /*
429 * If it tries to access a hole, return an error.
430 * Because, the callers, functions in dir.c and GC, should be able to know
431 * whether this page exists or not.
432 */
433 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
434 {
435 struct address_space *mapping = inode->i_mapping;
436 struct dnode_of_data dn;
437 struct page *page;
438 int err;
439 struct f2fs_io_info fio = {
440 .type = DATA,
441 .rw = READ_SYNC,
442 };
443 repeat:
444 page = grab_cache_page(mapping, index);
445 if (!page)
446 return ERR_PTR(-ENOMEM);
447
448 set_new_dnode(&dn, inode, NULL, NULL, 0);
449 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
450 if (err) {
451 f2fs_put_page(page, 1);
452 return ERR_PTR(err);
453 }
454 f2fs_put_dnode(&dn);
455
456 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
457 f2fs_put_page(page, 1);
458 return ERR_PTR(-ENOENT);
459 }
460
461 if (PageUptodate(page))
462 return page;
463
464 /*
465 * A new dentry page is allocated but not able to be written, since its
466 * new inode page couldn't be allocated due to -ENOSPC.
467 * In such the case, its blkaddr can be remained as NEW_ADDR.
468 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
469 */
470 if (dn.data_blkaddr == NEW_ADDR) {
471 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
472 SetPageUptodate(page);
473 return page;
474 }
475
476 fio.blk_addr = dn.data_blkaddr;
477 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
478 if (err)
479 return ERR_PTR(err);
480
481 lock_page(page);
482 if (unlikely(!PageUptodate(page))) {
483 f2fs_put_page(page, 1);
484 return ERR_PTR(-EIO);
485 }
486 if (unlikely(page->mapping != mapping)) {
487 f2fs_put_page(page, 1);
488 goto repeat;
489 }
490 return page;
491 }
492
493 /*
494 * Caller ensures that this data page is never allocated.
495 * A new zero-filled data page is allocated in the page cache.
496 *
497 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
498 * f2fs_unlock_op().
499 * Note that, ipage is set only by make_empty_dir.
500 */
501 struct page *get_new_data_page(struct inode *inode,
502 struct page *ipage, pgoff_t index, bool new_i_size)
503 {
504 struct address_space *mapping = inode->i_mapping;
505 struct page *page;
506 struct dnode_of_data dn;
507 int err;
508
509 set_new_dnode(&dn, inode, ipage, NULL, 0);
510 err = f2fs_reserve_block(&dn, index);
511 if (err)
512 return ERR_PTR(err);
513 repeat:
514 page = grab_cache_page(mapping, index);
515 if (!page) {
516 err = -ENOMEM;
517 goto put_err;
518 }
519
520 if (PageUptodate(page))
521 return page;
522
523 if (dn.data_blkaddr == NEW_ADDR) {
524 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
525 SetPageUptodate(page);
526 } else {
527 struct f2fs_io_info fio = {
528 .type = DATA,
529 .rw = READ_SYNC,
530 .blk_addr = dn.data_blkaddr,
531 };
532 err = f2fs_submit_page_bio(F2FS_I_SB(inode), page, &fio);
533 if (err)
534 goto put_err;
535
536 lock_page(page);
537 if (unlikely(!PageUptodate(page))) {
538 f2fs_put_page(page, 1);
539 err = -EIO;
540 goto put_err;
541 }
542 if (unlikely(page->mapping != mapping)) {
543 f2fs_put_page(page, 1);
544 goto repeat;
545 }
546 }
547
548 if (new_i_size &&
549 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
550 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
551 /* Only the directory inode sets new_i_size */
552 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
553 }
554 return page;
555
556 put_err:
557 f2fs_put_dnode(&dn);
558 return ERR_PTR(err);
559 }
560
561 static int __allocate_data_block(struct dnode_of_data *dn)
562 {
563 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
564 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
565 struct f2fs_summary sum;
566 struct node_info ni;
567 int seg = CURSEG_WARM_DATA;
568 pgoff_t fofs;
569
570 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
571 return -EPERM;
572 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
573 return -ENOSPC;
574
575 get_node_info(sbi, dn->nid, &ni);
576 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
577
578 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
579 seg = CURSEG_DIRECT_IO;
580
581 allocate_data_block(sbi, NULL, NULL_ADDR, &dn->data_blkaddr, &sum, seg);
582
583 /* direct IO doesn't use extent cache to maximize the performance */
584 __set_data_blkaddr(dn);
585
586 /* update i_size */
587 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
588 dn->ofs_in_node;
589 if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
590 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
591
592 return 0;
593 }
594
595 /*
596 * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
597 * If original data blocks are allocated, then give them to blockdev.
598 * Otherwise,
599 * a. preallocate requested block addresses
600 * b. do not use extent cache for better performance
601 * c. give the block addresses to blockdev
602 */
603 static int __get_data_block(struct inode *inode, sector_t iblock,
604 struct buffer_head *bh_result, int create, bool fiemap)
605 {
606 unsigned int blkbits = inode->i_sb->s_blocksize_bits;
607 unsigned maxblocks = bh_result->b_size >> blkbits;
608 struct dnode_of_data dn;
609 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
610 pgoff_t pgofs, end_offset;
611 int err = 0, ofs = 1;
612 bool allocated = false;
613
614 /* Get the page offset from the block offset(iblock) */
615 pgofs = (pgoff_t)(iblock >> (PAGE_CACHE_SHIFT - blkbits));
616
617 if (check_extent_cache(inode, pgofs, bh_result))
618 goto out;
619
620 if (create) {
621 f2fs_balance_fs(F2FS_I_SB(inode));
622 f2fs_lock_op(F2FS_I_SB(inode));
623 }
624
625 /* When reading holes, we need its node page */
626 set_new_dnode(&dn, inode, NULL, NULL, 0);
627 err = get_dnode_of_data(&dn, pgofs, mode);
628 if (err) {
629 if (err == -ENOENT)
630 err = 0;
631 goto unlock_out;
632 }
633 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
634 goto put_out;
635
636 if (dn.data_blkaddr != NULL_ADDR) {
637 set_buffer_new(bh_result);
638 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
639 } else if (create) {
640 err = __allocate_data_block(&dn);
641 if (err)
642 goto put_out;
643 allocated = true;
644 set_buffer_new(bh_result);
645 map_bh(bh_result, inode->i_sb, dn.data_blkaddr);
646 } else {
647 goto put_out;
648 }
649
650 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
651 bh_result->b_size = (((size_t)1) << blkbits);
652 dn.ofs_in_node++;
653 pgofs++;
654
655 get_next:
656 if (dn.ofs_in_node >= end_offset) {
657 if (allocated)
658 sync_inode_page(&dn);
659 allocated = false;
660 f2fs_put_dnode(&dn);
661
662 set_new_dnode(&dn, inode, NULL, NULL, 0);
663 err = get_dnode_of_data(&dn, pgofs, mode);
664 if (err) {
665 if (err == -ENOENT)
666 err = 0;
667 goto unlock_out;
668 }
669 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
670 goto put_out;
671
672 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
673 }
674
675 if (maxblocks > (bh_result->b_size >> blkbits)) {
676 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
677 if (blkaddr == NULL_ADDR && create) {
678 err = __allocate_data_block(&dn);
679 if (err)
680 goto sync_out;
681 allocated = true;
682 blkaddr = dn.data_blkaddr;
683 }
684 /* Give more consecutive addresses for the readahead */
685 if (blkaddr == (bh_result->b_blocknr + ofs)) {
686 ofs++;
687 dn.ofs_in_node++;
688 pgofs++;
689 bh_result->b_size += (((size_t)1) << blkbits);
690 goto get_next;
691 }
692 }
693 sync_out:
694 if (allocated)
695 sync_inode_page(&dn);
696 put_out:
697 f2fs_put_dnode(&dn);
698 unlock_out:
699 if (create)
700 f2fs_unlock_op(F2FS_I_SB(inode));
701 out:
702 trace_f2fs_get_data_block(inode, iblock, bh_result, err);
703 return err;
704 }
705
706 static int get_data_block(struct inode *inode, sector_t iblock,
707 struct buffer_head *bh_result, int create)
708 {
709 return __get_data_block(inode, iblock, bh_result, create, false);
710 }
711
712 static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
713 struct buffer_head *bh_result, int create)
714 {
715 return __get_data_block(inode, iblock, bh_result, create, true);
716 }
717
718 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
719 u64 start, u64 len)
720 {
721 return generic_block_fiemap(inode, fieinfo,
722 start, len, get_data_block_fiemap);
723 }
724
725 static int f2fs_read_data_page(struct file *file, struct page *page)
726 {
727 struct inode *inode = page->mapping->host;
728 int ret = -EAGAIN;
729
730 trace_f2fs_readpage(page, DATA);
731
732 /* If the file has inline data, try to read it directly */
733 if (f2fs_has_inline_data(inode))
734 ret = f2fs_read_inline_data(inode, page);
735 if (ret == -EAGAIN)
736 ret = mpage_readpage(page, get_data_block);
737
738 return ret;
739 }
740
741 static int f2fs_read_data_pages(struct file *file,
742 struct address_space *mapping,
743 struct list_head *pages, unsigned nr_pages)
744 {
745 struct inode *inode = file->f_mapping->host;
746
747 /* If the file has inline data, skip readpages */
748 if (f2fs_has_inline_data(inode))
749 return 0;
750
751 return mpage_readpages(mapping, pages, nr_pages, get_data_block);
752 }
753
754 int do_write_data_page(struct page *page, struct f2fs_io_info *fio)
755 {
756 struct inode *inode = page->mapping->host;
757 struct dnode_of_data dn;
758 int err = 0;
759
760 set_new_dnode(&dn, inode, NULL, NULL, 0);
761 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
762 if (err)
763 return err;
764
765 fio->blk_addr = dn.data_blkaddr;
766
767 /* This page is already truncated */
768 if (fio->blk_addr == NULL_ADDR)
769 goto out_writepage;
770
771 set_page_writeback(page);
772
773 /*
774 * If current allocation needs SSR,
775 * it had better in-place writes for updated data.
776 */
777 if (unlikely(fio->blk_addr != NEW_ADDR &&
778 !is_cold_data(page) &&
779 need_inplace_update(inode))) {
780 rewrite_data_page(page, fio);
781 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
782 } else {
783 write_data_page(page, &dn, fio);
784 update_extent_cache(&dn);
785 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
786 }
787 out_writepage:
788 f2fs_put_dnode(&dn);
789 return err;
790 }
791
792 static int f2fs_write_data_page(struct page *page,
793 struct writeback_control *wbc)
794 {
795 struct inode *inode = page->mapping->host;
796 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
797 loff_t i_size = i_size_read(inode);
798 const pgoff_t end_index = ((unsigned long long) i_size)
799 >> PAGE_CACHE_SHIFT;
800 unsigned offset = 0;
801 bool need_balance_fs = false;
802 int err = 0;
803 struct f2fs_io_info fio = {
804 .type = DATA,
805 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
806 };
807
808 trace_f2fs_writepage(page, DATA);
809
810 if (page->index < end_index)
811 goto write;
812
813 /*
814 * If the offset is out-of-range of file size,
815 * this page does not have to be written to disk.
816 */
817 offset = i_size & (PAGE_CACHE_SIZE - 1);
818 if ((page->index >= end_index + 1) || !offset)
819 goto out;
820
821 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
822 write:
823 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
824 goto redirty_out;
825 if (f2fs_is_drop_cache(inode))
826 goto out;
827 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
828 available_free_memory(sbi, BASE_CHECK))
829 goto redirty_out;
830
831 /* Dentry blocks are controlled by checkpoint */
832 if (S_ISDIR(inode->i_mode)) {
833 if (unlikely(f2fs_cp_error(sbi)))
834 goto redirty_out;
835 err = do_write_data_page(page, &fio);
836 goto done;
837 }
838
839 /* we should bypass data pages to proceed the kworkder jobs */
840 if (unlikely(f2fs_cp_error(sbi))) {
841 SetPageError(page);
842 goto out;
843 }
844
845 if (!wbc->for_reclaim)
846 need_balance_fs = true;
847 else if (has_not_enough_free_secs(sbi, 0))
848 goto redirty_out;
849
850 err = -EAGAIN;
851 f2fs_lock_op(sbi);
852 if (f2fs_has_inline_data(inode))
853 err = f2fs_write_inline_data(inode, page);
854 if (err == -EAGAIN)
855 err = do_write_data_page(page, &fio);
856 f2fs_unlock_op(sbi);
857 done:
858 if (err && err != -ENOENT)
859 goto redirty_out;
860
861 clear_cold_data(page);
862 out:
863 inode_dec_dirty_pages(inode);
864 unlock_page(page);
865 if (need_balance_fs)
866 f2fs_balance_fs(sbi);
867 if (wbc->for_reclaim)
868 f2fs_submit_merged_bio(sbi, DATA, WRITE);
869 return 0;
870
871 redirty_out:
872 redirty_page_for_writepage(wbc, page);
873 return AOP_WRITEPAGE_ACTIVATE;
874 }
875
876 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
877 void *data)
878 {
879 struct address_space *mapping = data;
880 int ret = mapping->a_ops->writepage(page, wbc);
881 mapping_set_error(mapping, ret);
882 return ret;
883 }
884
885 static int f2fs_write_data_pages(struct address_space *mapping,
886 struct writeback_control *wbc)
887 {
888 struct inode *inode = mapping->host;
889 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
890 bool locked = false;
891 int ret;
892 long diff;
893
894 trace_f2fs_writepages(mapping->host, wbc, DATA);
895
896 /* deal with chardevs and other special file */
897 if (!mapping->a_ops->writepage)
898 return 0;
899
900 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
901 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
902 available_free_memory(sbi, DIRTY_DENTS))
903 goto skip_write;
904
905 diff = nr_pages_to_write(sbi, DATA, wbc);
906
907 if (!S_ISDIR(inode->i_mode)) {
908 mutex_lock(&sbi->writepages);
909 locked = true;
910 }
911 ret = write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
912 if (locked)
913 mutex_unlock(&sbi->writepages);
914
915 f2fs_submit_merged_bio(sbi, DATA, WRITE);
916
917 remove_dirty_dir_inode(inode);
918
919 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
920 return ret;
921
922 skip_write:
923 wbc->pages_skipped += get_dirty_pages(inode);
924 return 0;
925 }
926
927 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
928 {
929 struct inode *inode = mapping->host;
930
931 if (to > inode->i_size) {
932 truncate_pagecache(inode, inode->i_size);
933 truncate_blocks(inode, inode->i_size, true);
934 }
935 }
936
937 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
938 loff_t pos, unsigned len, unsigned flags,
939 struct page **pagep, void **fsdata)
940 {
941 struct inode *inode = mapping->host;
942 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
943 struct page *page, *ipage;
944 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
945 struct dnode_of_data dn;
946 int err = 0;
947
948 trace_f2fs_write_begin(inode, pos, len, flags);
949
950 f2fs_balance_fs(sbi);
951
952 /*
953 * We should check this at this moment to avoid deadlock on inode page
954 * and #0 page. The locking rule for inline_data conversion should be:
955 * lock_page(page #0) -> lock_page(inode_page)
956 */
957 if (index != 0) {
958 err = f2fs_convert_inline_inode(inode);
959 if (err)
960 goto fail;
961 }
962 repeat:
963 page = grab_cache_page_write_begin(mapping, index, flags);
964 if (!page) {
965 err = -ENOMEM;
966 goto fail;
967 }
968
969 *pagep = page;
970
971 f2fs_lock_op(sbi);
972
973 /* check inline_data */
974 ipage = get_node_page(sbi, inode->i_ino);
975 if (IS_ERR(ipage)) {
976 err = PTR_ERR(ipage);
977 goto unlock_fail;
978 }
979
980 set_new_dnode(&dn, inode, ipage, ipage, 0);
981
982 if (f2fs_has_inline_data(inode)) {
983 if (pos + len <= MAX_INLINE_DATA) {
984 read_inline_data(page, ipage);
985 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
986 sync_inode_page(&dn);
987 goto put_next;
988 }
989 err = f2fs_convert_inline_page(&dn, page);
990 if (err)
991 goto put_fail;
992 }
993 err = f2fs_reserve_block(&dn, index);
994 if (err)
995 goto put_fail;
996 put_next:
997 f2fs_put_dnode(&dn);
998 f2fs_unlock_op(sbi);
999
1000 if ((len == PAGE_CACHE_SIZE) || PageUptodate(page))
1001 return 0;
1002
1003 f2fs_wait_on_page_writeback(page, DATA);
1004
1005 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1006 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1007 unsigned end = start + len;
1008
1009 /* Reading beyond i_size is simple: memset to zero */
1010 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1011 goto out;
1012 }
1013
1014 if (dn.data_blkaddr == NEW_ADDR) {
1015 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1016 } else {
1017 struct f2fs_io_info fio = {
1018 .type = DATA,
1019 .rw = READ_SYNC,
1020 .blk_addr = dn.data_blkaddr,
1021 };
1022 err = f2fs_submit_page_bio(sbi, page, &fio);
1023 if (err)
1024 goto fail;
1025
1026 lock_page(page);
1027 if (unlikely(!PageUptodate(page))) {
1028 f2fs_put_page(page, 1);
1029 err = -EIO;
1030 goto fail;
1031 }
1032 if (unlikely(page->mapping != mapping)) {
1033 f2fs_put_page(page, 1);
1034 goto repeat;
1035 }
1036 }
1037 out:
1038 SetPageUptodate(page);
1039 clear_cold_data(page);
1040 return 0;
1041
1042 put_fail:
1043 f2fs_put_dnode(&dn);
1044 unlock_fail:
1045 f2fs_unlock_op(sbi);
1046 f2fs_put_page(page, 1);
1047 fail:
1048 f2fs_write_failed(mapping, pos + len);
1049 return err;
1050 }
1051
1052 static int f2fs_write_end(struct file *file,
1053 struct address_space *mapping,
1054 loff_t pos, unsigned len, unsigned copied,
1055 struct page *page, void *fsdata)
1056 {
1057 struct inode *inode = page->mapping->host;
1058
1059 trace_f2fs_write_end(inode, pos, len, copied);
1060
1061 set_page_dirty(page);
1062
1063 if (pos + copied > i_size_read(inode)) {
1064 i_size_write(inode, pos + copied);
1065 mark_inode_dirty(inode);
1066 update_inode_page(inode);
1067 }
1068
1069 f2fs_put_page(page, 1);
1070 return copied;
1071 }
1072
1073 static int check_direct_IO(struct inode *inode, int rw,
1074 struct iov_iter *iter, loff_t offset)
1075 {
1076 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1077
1078 if (rw == READ)
1079 return 0;
1080
1081 if (offset & blocksize_mask)
1082 return -EINVAL;
1083
1084 if (iov_iter_alignment(iter) & blocksize_mask)
1085 return -EINVAL;
1086
1087 return 0;
1088 }
1089
1090 static ssize_t f2fs_direct_IO(int rw, struct kiocb *iocb,
1091 struct iov_iter *iter, loff_t offset)
1092 {
1093 struct file *file = iocb->ki_filp;
1094 struct address_space *mapping = file->f_mapping;
1095 struct inode *inode = mapping->host;
1096 size_t count = iov_iter_count(iter);
1097 int err;
1098
1099 /* we don't need to use inline_data strictly */
1100 if (f2fs_has_inline_data(inode)) {
1101 err = f2fs_convert_inline_inode(inode);
1102 if (err)
1103 return err;
1104 }
1105
1106 if (check_direct_IO(inode, rw, iter, offset))
1107 return 0;
1108
1109 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1110
1111 err = blockdev_direct_IO(rw, iocb, inode, iter, offset, get_data_block);
1112 if (err < 0 && (rw & WRITE))
1113 f2fs_write_failed(mapping, offset + count);
1114
1115 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1116
1117 return err;
1118 }
1119
1120 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1121 unsigned int length)
1122 {
1123 struct inode *inode = page->mapping->host;
1124 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1125
1126 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1127 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1128 return;
1129
1130 if (PageDirty(page)) {
1131 if (inode->i_ino == F2FS_META_INO(sbi))
1132 dec_page_count(sbi, F2FS_DIRTY_META);
1133 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1134 dec_page_count(sbi, F2FS_DIRTY_NODES);
1135 else
1136 inode_dec_dirty_pages(inode);
1137 }
1138 ClearPagePrivate(page);
1139 }
1140
1141 int f2fs_release_page(struct page *page, gfp_t wait)
1142 {
1143 /* If this is dirty page, keep PagePrivate */
1144 if (PageDirty(page))
1145 return 0;
1146
1147 ClearPagePrivate(page);
1148 return 1;
1149 }
1150
1151 static int f2fs_set_data_page_dirty(struct page *page)
1152 {
1153 struct address_space *mapping = page->mapping;
1154 struct inode *inode = mapping->host;
1155
1156 trace_f2fs_set_page_dirty(page, DATA);
1157
1158 SetPageUptodate(page);
1159
1160 if (f2fs_is_atomic_file(inode)) {
1161 register_inmem_page(inode, page);
1162 return 1;
1163 }
1164
1165 mark_inode_dirty(inode);
1166
1167 if (!PageDirty(page)) {
1168 __set_page_dirty_nobuffers(page);
1169 update_dirty_page(inode, page);
1170 return 1;
1171 }
1172 return 0;
1173 }
1174
1175 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1176 {
1177 struct inode *inode = mapping->host;
1178
1179 /* we don't need to use inline_data strictly */
1180 if (f2fs_has_inline_data(inode)) {
1181 int err = f2fs_convert_inline_inode(inode);
1182 if (err)
1183 return err;
1184 }
1185 return generic_block_bmap(mapping, block, get_data_block);
1186 }
1187
1188 const struct address_space_operations f2fs_dblock_aops = {
1189 .readpage = f2fs_read_data_page,
1190 .readpages = f2fs_read_data_pages,
1191 .writepage = f2fs_write_data_page,
1192 .writepages = f2fs_write_data_pages,
1193 .write_begin = f2fs_write_begin,
1194 .write_end = f2fs_write_end,
1195 .set_page_dirty = f2fs_set_data_page_dirty,
1196 .invalidatepage = f2fs_invalidate_page,
1197 .releasepage = f2fs_release_page,
1198 .direct_IO = f2fs_direct_IO,
1199 .bmap = f2fs_bmap,
1200 };