projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: add annotation for space utilization of regular/inline dentry
[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/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
22 #include <linux/cleancache.h>
23
24 #include "f2fs.h"
25 #include "node.h"
26 #include "segment.h"
27 #include "trace.h"
28 #include <trace/events/f2fs.h>
29
30 static void f2fs_read_end_io(struct bio *bio, int err)
31 {
32 struct bio_vec *bvec;
33 int i;
34
35 if (f2fs_bio_encrypted(bio)) {
36 if (err) {
37 f2fs_release_crypto_ctx(bio->bi_private);
38 } else {
39 f2fs_end_io_crypto_work(bio->bi_private, bio);
40 return;
41 }
42 }
43
44 bio_for_each_segment_all(bvec, bio, i) {
45 struct page *page = bvec->bv_page;
46
47 if (!err) {
48 SetPageUptodate(page);
49 } else {
50 ClearPageUptodate(page);
51 SetPageError(page);
52 }
53 unlock_page(page);
54 }
55 bio_put(bio);
56 }
57
58 static void f2fs_write_end_io(struct bio *bio, int err)
59 {
60 struct f2fs_sb_info *sbi = bio->bi_private;
61 struct bio_vec *bvec;
62 int i;
63
64 bio_for_each_segment_all(bvec, bio, i) {
65 struct page *page = bvec->bv_page;
66
67 f2fs_restore_and_release_control_page(&page);
68
69 if (unlikely(err)) {
70 set_page_dirty(page);
71 set_bit(AS_EIO, &page->mapping->flags);
72 f2fs_stop_checkpoint(sbi);
73 }
74 end_page_writeback(page);
75 dec_page_count(sbi, F2FS_WRITEBACK);
76 }
77
78 if (!get_pages(sbi, F2FS_WRITEBACK) &&
79 !list_empty(&sbi->cp_wait.task_list))
80 wake_up(&sbi->cp_wait);
81
82 bio_put(bio);
83 }
84
85 /*
86 * Low-level block read/write IO operations.
87 */
88 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
89 int npages, bool is_read)
90 {
91 struct bio *bio;
92
93 bio = f2fs_bio_alloc(npages);
94
95 bio->bi_bdev = sbi->sb->s_bdev;
96 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
97 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
98 bio->bi_private = is_read ? NULL : sbi;
99
100 return bio;
101 }
102
103 static void __submit_merged_bio(struct f2fs_bio_info *io)
104 {
105 struct f2fs_io_info *fio = &io->fio;
106
107 if (!io->bio)
108 return;
109
110 if (is_read_io(fio->rw))
111 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
112 else
113 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
114
115 submit_bio(fio->rw, io->bio);
116 io->bio = NULL;
117 }
118
119 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
120 enum page_type type, int rw)
121 {
122 enum page_type btype = PAGE_TYPE_OF_BIO(type);
123 struct f2fs_bio_info *io;
124
125 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
126
127 down_write(&io->io_rwsem);
128
129 /* change META to META_FLUSH in the checkpoint procedure */
130 if (type >= META_FLUSH) {
131 io->fio.type = META_FLUSH;
132 if (test_opt(sbi, NOBARRIER))
133 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
134 else
135 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
136 }
137 __submit_merged_bio(io);
138 up_write(&io->io_rwsem);
139 }
140
141 /*
142 * Fill the locked page with data located in the block address.
143 * Return unlocked page.
144 */
145 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
146 {
147 struct bio *bio;
148 struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
149
150 trace_f2fs_submit_page_bio(page, fio);
151 f2fs_trace_ios(fio, 0);
152
153 /* Allocate a new bio */
154 bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
155
156 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
157 bio_put(bio);
158 return -EFAULT;
159 }
160
161 submit_bio(fio->rw, bio);
162 return 0;
163 }
164
165 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
166 {
167 struct f2fs_sb_info *sbi = fio->sbi;
168 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
169 struct f2fs_bio_info *io;
170 bool is_read = is_read_io(fio->rw);
171 struct page *bio_page;
172
173 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
174
175 verify_block_addr(sbi, fio->blk_addr);
176
177 down_write(&io->io_rwsem);
178
179 if (!is_read)
180 inc_page_count(sbi, F2FS_WRITEBACK);
181
182 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
183 io->fio.rw != fio->rw))
184 __submit_merged_bio(io);
185 alloc_new:
186 if (io->bio == NULL) {
187 int bio_blocks = MAX_BIO_BLOCKS(sbi);
188
189 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
190 io->fio = *fio;
191 }
192
193 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
194
195 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
196 PAGE_CACHE_SIZE) {
197 __submit_merged_bio(io);
198 goto alloc_new;
199 }
200
201 io->last_block_in_bio = fio->blk_addr;
202 f2fs_trace_ios(fio, 0);
203
204 up_write(&io->io_rwsem);
205 trace_f2fs_submit_page_mbio(fio->page, fio);
206 }
207
208 /*
209 * Lock ordering for the change of data block address:
210 * ->data_page
211 * ->node_page
212 * update block addresses in the node page
213 */
214 void set_data_blkaddr(struct dnode_of_data *dn)
215 {
216 struct f2fs_node *rn;
217 __le32 *addr_array;
218 struct page *node_page = dn->node_page;
219 unsigned int ofs_in_node = dn->ofs_in_node;
220
221 f2fs_wait_on_page_writeback(node_page, NODE);
222
223 rn = F2FS_NODE(node_page);
224
225 /* Get physical address of data block */
226 addr_array = blkaddr_in_node(rn);
227 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
228 set_page_dirty(node_page);
229 }
230
231 int reserve_new_block(struct dnode_of_data *dn)
232 {
233 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
234
235 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
236 return -EPERM;
237 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
238 return -ENOSPC;
239
240 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
241
242 dn->data_blkaddr = NEW_ADDR;
243 set_data_blkaddr(dn);
244 mark_inode_dirty(dn->inode);
245 sync_inode_page(dn);
246 return 0;
247 }
248
249 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
250 {
251 bool need_put = dn->inode_page ? false : true;
252 int err;
253
254 err = get_dnode_of_data(dn, index, ALLOC_NODE);
255 if (err)
256 return err;
257
258 if (dn->data_blkaddr == NULL_ADDR)
259 err = reserve_new_block(dn);
260 if (err || need_put)
261 f2fs_put_dnode(dn);
262 return err;
263 }
264
265 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
266 {
267 struct extent_info ei;
268 struct inode *inode = dn->inode;
269
270 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
271 dn->data_blkaddr = ei.blk + index - ei.fofs;
272 return 0;
273 }
274
275 return f2fs_reserve_block(dn, index);
276 }
277
278 struct page *get_read_data_page(struct inode *inode, pgoff_t index, int rw)
279 {
280 struct address_space *mapping = inode->i_mapping;
281 struct dnode_of_data dn;
282 struct page *page;
283 struct extent_info ei;
284 int err;
285 struct f2fs_io_info fio = {
286 .sbi = F2FS_I_SB(inode),
287 .type = DATA,
288 .rw = rw,
289 .encrypted_page = NULL,
290 };
291
292 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
293 return read_mapping_page(mapping, index, NULL);
294
295 page = grab_cache_page(mapping, index);
296 if (!page)
297 return ERR_PTR(-ENOMEM);
298
299 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
300 dn.data_blkaddr = ei.blk + index - ei.fofs;
301 goto got_it;
302 }
303
304 set_new_dnode(&dn, inode, NULL, NULL, 0);
305 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
306 if (err)
307 goto put_err;
308 f2fs_put_dnode(&dn);
309
310 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
311 err = -ENOENT;
312 goto put_err;
313 }
314 got_it:
315 if (PageUptodate(page)) {
316 unlock_page(page);
317 return page;
318 }
319
320 /*
321 * A new dentry page is allocated but not able to be written, since its
322 * new inode page couldn't be allocated due to -ENOSPC.
323 * In such the case, its blkaddr can be remained as NEW_ADDR.
324 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
325 */
326 if (dn.data_blkaddr == NEW_ADDR) {
327 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
328 SetPageUptodate(page);
329 unlock_page(page);
330 return page;
331 }
332
333 fio.blk_addr = dn.data_blkaddr;
334 fio.page = page;
335 err = f2fs_submit_page_bio(&fio);
336 if (err)
337 goto put_err;
338 return page;
339
340 put_err:
341 f2fs_put_page(page, 1);
342 return ERR_PTR(err);
343 }
344
345 struct page *find_data_page(struct inode *inode, pgoff_t index)
346 {
347 struct address_space *mapping = inode->i_mapping;
348 struct page *page;
349
350 page = find_get_page(mapping, index);
351 if (page && PageUptodate(page))
352 return page;
353 f2fs_put_page(page, 0);
354
355 page = get_read_data_page(inode, index, READ_SYNC);
356 if (IS_ERR(page))
357 return page;
358
359 if (PageUptodate(page))
360 return page;
361
362 wait_on_page_locked(page);
363 if (unlikely(!PageUptodate(page))) {
364 f2fs_put_page(page, 0);
365 return ERR_PTR(-EIO);
366 }
367 return page;
368 }
369
370 /*
371 * If it tries to access a hole, return an error.
372 * Because, the callers, functions in dir.c and GC, should be able to know
373 * whether this page exists or not.
374 */
375 struct page *get_lock_data_page(struct inode *inode, pgoff_t index)
376 {
377 struct address_space *mapping = inode->i_mapping;
378 struct page *page;
379 repeat:
380 page = get_read_data_page(inode, index, READ_SYNC);
381 if (IS_ERR(page))
382 return page;
383
384 /* wait for read completion */
385 lock_page(page);
386 if (unlikely(!PageUptodate(page))) {
387 f2fs_put_page(page, 1);
388 return ERR_PTR(-EIO);
389 }
390 if (unlikely(page->mapping != mapping)) {
391 f2fs_put_page(page, 1);
392 goto repeat;
393 }
394 return page;
395 }
396
397 /*
398 * Caller ensures that this data page is never allocated.
399 * A new zero-filled data page is allocated in the page cache.
400 *
401 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
402 * f2fs_unlock_op().
403 * Note that, ipage is set only by make_empty_dir, and if any error occur,
404 * ipage should be released by this function.
405 */
406 struct page *get_new_data_page(struct inode *inode,
407 struct page *ipage, pgoff_t index, bool new_i_size)
408 {
409 struct address_space *mapping = inode->i_mapping;
410 struct page *page;
411 struct dnode_of_data dn;
412 int err;
413 repeat:
414 page = grab_cache_page(mapping, index);
415 if (!page) {
416 /*
417 * before exiting, we should make sure ipage will be released
418 * if any error occur.
419 */
420 f2fs_put_page(ipage, 1);
421 return ERR_PTR(-ENOMEM);
422 }
423
424 set_new_dnode(&dn, inode, ipage, NULL, 0);
425 err = f2fs_reserve_block(&dn, index);
426 if (err) {
427 f2fs_put_page(page, 1);
428 return ERR_PTR(err);
429 }
430 if (!ipage)
431 f2fs_put_dnode(&dn);
432
433 if (PageUptodate(page))
434 goto got_it;
435
436 if (dn.data_blkaddr == NEW_ADDR) {
437 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
438 SetPageUptodate(page);
439 } else {
440 f2fs_put_page(page, 1);
441
442 page = get_read_data_page(inode, index, READ_SYNC);
443 if (IS_ERR(page))
444 goto repeat;
445
446 /* wait for read completion */
447 lock_page(page);
448 }
449 got_it:
450 if (new_i_size &&
451 i_size_read(inode) < ((index + 1) << PAGE_CACHE_SHIFT)) {
452 i_size_write(inode, ((index + 1) << PAGE_CACHE_SHIFT));
453 /* Only the directory inode sets new_i_size */
454 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
455 }
456 return page;
457 }
458
459 static int __allocate_data_block(struct dnode_of_data *dn)
460 {
461 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
462 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
463 struct f2fs_summary sum;
464 struct node_info ni;
465 int seg = CURSEG_WARM_DATA;
466 pgoff_t fofs;
467
468 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
469 return -EPERM;
470
471 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
472 if (dn->data_blkaddr == NEW_ADDR)
473 goto alloc;
474
475 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
476 return -ENOSPC;
477
478 alloc:
479 get_node_info(sbi, dn->nid, &ni);
480 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
481
482 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
483 seg = CURSEG_DIRECT_IO;
484
485 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
486 &sum, seg);
487 set_data_blkaddr(dn);
488
489 /* update i_size */
490 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
491 dn->ofs_in_node;
492 if (i_size_read(dn->inode) < ((fofs + 1) << PAGE_CACHE_SHIFT))
493 i_size_write(dn->inode, ((fofs + 1) << PAGE_CACHE_SHIFT));
494
495 /* direct IO doesn't use extent cache to maximize the performance */
496 f2fs_drop_largest_extent(dn->inode, fofs);
497
498 return 0;
499 }
500
501 static void __allocate_data_blocks(struct inode *inode, loff_t offset,
502 size_t count)
503 {
504 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
505 struct dnode_of_data dn;
506 u64 start = F2FS_BYTES_TO_BLK(offset);
507 u64 len = F2FS_BYTES_TO_BLK(count);
508 bool allocated;
509 u64 end_offset;
510
511 while (len) {
512 f2fs_balance_fs(sbi);
513 f2fs_lock_op(sbi);
514
515 /* When reading holes, we need its node page */
516 set_new_dnode(&dn, inode, NULL, NULL, 0);
517 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
518 goto out;
519
520 allocated = false;
521 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
522
523 while (dn.ofs_in_node < end_offset && len) {
524 block_t blkaddr;
525
526 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
527 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
528 if (__allocate_data_block(&dn))
529 goto sync_out;
530 allocated = true;
531 }
532 len--;
533 start++;
534 dn.ofs_in_node++;
535 }
536
537 if (allocated)
538 sync_inode_page(&dn);
539
540 f2fs_put_dnode(&dn);
541 f2fs_unlock_op(sbi);
542 }
543 return;
544
545 sync_out:
546 if (allocated)
547 sync_inode_page(&dn);
548 f2fs_put_dnode(&dn);
549 out:
550 f2fs_unlock_op(sbi);
551 return;
552 }
553
554 /*
555 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
556 * f2fs_map_blocks structure.
557 * If original data blocks are allocated, then give them to blockdev.
558 * Otherwise,
559 * a. preallocate requested block addresses
560 * b. do not use extent cache for better performance
561 * c. give the block addresses to blockdev
562 */
563 static int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
564 int create, bool fiemap)
565 {
566 unsigned int maxblocks = map->m_len;
567 struct dnode_of_data dn;
568 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
569 pgoff_t pgofs, end_offset;
570 int err = 0, ofs = 1;
571 struct extent_info ei;
572 bool allocated = false;
573
574 map->m_len = 0;
575 map->m_flags = 0;
576
577 /* it only supports block size == page size */
578 pgofs = (pgoff_t)map->m_lblk;
579
580 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
581 map->m_pblk = ei.blk + pgofs - ei.fofs;
582 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
583 map->m_flags = F2FS_MAP_MAPPED;
584 goto out;
585 }
586
587 if (create)
588 f2fs_lock_op(F2FS_I_SB(inode));
589
590 /* When reading holes, we need its node page */
591 set_new_dnode(&dn, inode, NULL, NULL, 0);
592 err = get_dnode_of_data(&dn, pgofs, mode);
593 if (err) {
594 if (err == -ENOENT)
595 err = 0;
596 goto unlock_out;
597 }
598 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
599 goto put_out;
600
601 if (dn.data_blkaddr != NULL_ADDR) {
602 map->m_flags = F2FS_MAP_MAPPED;
603 map->m_pblk = dn.data_blkaddr;
604 if (dn.data_blkaddr == NEW_ADDR)
605 map->m_flags |= F2FS_MAP_UNWRITTEN;
606 } else if (create) {
607 err = __allocate_data_block(&dn);
608 if (err)
609 goto put_out;
610 allocated = true;
611 map->m_flags = F2FS_MAP_NEW | F2FS_MAP_MAPPED;
612 map->m_pblk = dn.data_blkaddr;
613 } else {
614 goto put_out;
615 }
616
617 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
618 map->m_len = 1;
619 dn.ofs_in_node++;
620 pgofs++;
621
622 get_next:
623 if (dn.ofs_in_node >= end_offset) {
624 if (allocated)
625 sync_inode_page(&dn);
626 allocated = false;
627 f2fs_put_dnode(&dn);
628
629 set_new_dnode(&dn, inode, NULL, NULL, 0);
630 err = get_dnode_of_data(&dn, pgofs, mode);
631 if (err) {
632 if (err == -ENOENT)
633 err = 0;
634 goto unlock_out;
635 }
636 if (dn.data_blkaddr == NEW_ADDR && !fiemap)
637 goto put_out;
638
639 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
640 }
641
642 if (maxblocks > map->m_len) {
643 block_t blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
644 if (blkaddr == NULL_ADDR && create) {
645 err = __allocate_data_block(&dn);
646 if (err)
647 goto sync_out;
648 allocated = true;
649 map->m_flags |= F2FS_MAP_NEW;
650 blkaddr = dn.data_blkaddr;
651 }
652 /* Give more consecutive addresses for the readahead */
653 if ((map->m_pblk != NEW_ADDR &&
654 blkaddr == (map->m_pblk + ofs)) ||
655 (map->m_pblk == NEW_ADDR &&
656 blkaddr == NEW_ADDR)) {
657 ofs++;
658 dn.ofs_in_node++;
659 pgofs++;
660 map->m_len++;
661 goto get_next;
662 }
663 }
664 sync_out:
665 if (allocated)
666 sync_inode_page(&dn);
667 put_out:
668 f2fs_put_dnode(&dn);
669 unlock_out:
670 if (create)
671 f2fs_unlock_op(F2FS_I_SB(inode));
672 out:
673 trace_f2fs_map_blocks(inode, map, err);
674 return err;
675 }
676
677 static int __get_data_block(struct inode *inode, sector_t iblock,
678 struct buffer_head *bh, int create, bool fiemap)
679 {
680 struct f2fs_map_blocks map;
681 int ret;
682
683 map.m_lblk = iblock;
684 map.m_len = bh->b_size >> inode->i_blkbits;
685
686 ret = f2fs_map_blocks(inode, &map, create, fiemap);
687 if (!ret) {
688 map_bh(bh, inode->i_sb, map.m_pblk);
689 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
690 bh->b_size = map.m_len << inode->i_blkbits;
691 }
692 return ret;
693 }
694
695 static int get_data_block(struct inode *inode, sector_t iblock,
696 struct buffer_head *bh_result, int create)
697 {
698 return __get_data_block(inode, iblock, bh_result, create, false);
699 }
700
701 static int get_data_block_fiemap(struct inode *inode, sector_t iblock,
702 struct buffer_head *bh_result, int create)
703 {
704 return __get_data_block(inode, iblock, bh_result, create, true);
705 }
706
707 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
708 {
709 return (offset >> inode->i_blkbits);
710 }
711
712 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
713 {
714 return (blk << inode->i_blkbits);
715 }
716
717 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
718 u64 start, u64 len)
719 {
720 struct buffer_head map_bh;
721 sector_t start_blk, last_blk;
722 loff_t isize = i_size_read(inode);
723 u64 logical = 0, phys = 0, size = 0;
724 u32 flags = 0;
725 bool past_eof = false, whole_file = false;
726 int ret = 0;
727
728 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
729 if (ret)
730 return ret;
731
732 mutex_lock(&inode->i_mutex);
733
734 if (len >= isize) {
735 whole_file = true;
736 len = isize;
737 }
738
739 if (logical_to_blk(inode, len) == 0)
740 len = blk_to_logical(inode, 1);
741
742 start_blk = logical_to_blk(inode, start);
743 last_blk = logical_to_blk(inode, start + len - 1);
744 next:
745 memset(&map_bh, 0, sizeof(struct buffer_head));
746 map_bh.b_size = len;
747
748 ret = get_data_block_fiemap(inode, start_blk, &map_bh, 0);
749 if (ret)
750 goto out;
751
752 /* HOLE */
753 if (!buffer_mapped(&map_bh)) {
754 start_blk++;
755
756 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
757 past_eof = 1;
758
759 if (past_eof && size) {
760 flags |= FIEMAP_EXTENT_LAST;
761 ret = fiemap_fill_next_extent(fieinfo, logical,
762 phys, size, flags);
763 } else if (size) {
764 ret = fiemap_fill_next_extent(fieinfo, logical,
765 phys, size, flags);
766 size = 0;
767 }
768
769 /* if we have holes up to/past EOF then we're done */
770 if (start_blk > last_blk || past_eof || ret)
771 goto out;
772 } else {
773 if (start_blk > last_blk && !whole_file) {
774 ret = fiemap_fill_next_extent(fieinfo, logical,
775 phys, size, flags);
776 goto out;
777 }
778
779 /*
780 * if size != 0 then we know we already have an extent
781 * to add, so add it.
782 */
783 if (size) {
784 ret = fiemap_fill_next_extent(fieinfo, logical,
785 phys, size, flags);
786 if (ret)
787 goto out;
788 }
789
790 logical = blk_to_logical(inode, start_blk);
791 phys = blk_to_logical(inode, map_bh.b_blocknr);
792 size = map_bh.b_size;
793 flags = 0;
794 if (buffer_unwritten(&map_bh))
795 flags = FIEMAP_EXTENT_UNWRITTEN;
796
797 start_blk += logical_to_blk(inode, size);
798
799 /*
800 * If we are past the EOF, then we need to make sure as
801 * soon as we find a hole that the last extent we found
802 * is marked with FIEMAP_EXTENT_LAST
803 */
804 if (!past_eof && logical + size >= isize)
805 past_eof = true;
806 }
807 cond_resched();
808 if (fatal_signal_pending(current))
809 ret = -EINTR;
810 else
811 goto next;
812 out:
813 if (ret == 1)
814 ret = 0;
815
816 mutex_unlock(&inode->i_mutex);
817 return ret;
818 }
819
820 /*
821 * This function was originally taken from fs/mpage.c, and customized for f2fs.
822 * Major change was from block_size == page_size in f2fs by default.
823 */
824 static int f2fs_mpage_readpages(struct address_space *mapping,
825 struct list_head *pages, struct page *page,
826 unsigned nr_pages)
827 {
828 struct bio *bio = NULL;
829 unsigned page_idx;
830 sector_t last_block_in_bio = 0;
831 struct inode *inode = mapping->host;
832 const unsigned blkbits = inode->i_blkbits;
833 const unsigned blocksize = 1 << blkbits;
834 sector_t block_in_file;
835 sector_t last_block;
836 sector_t last_block_in_file;
837 sector_t block_nr;
838 struct block_device *bdev = inode->i_sb->s_bdev;
839 struct f2fs_map_blocks map;
840
841 map.m_pblk = 0;
842 map.m_lblk = 0;
843 map.m_len = 0;
844 map.m_flags = 0;
845
846 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
847
848 prefetchw(&page->flags);
849 if (pages) {
850 page = list_entry(pages->prev, struct page, lru);
851 list_del(&page->lru);
852 if (add_to_page_cache_lru(page, mapping,
853 page->index, GFP_KERNEL))
854 goto next_page;
855 }
856
857 block_in_file = (sector_t)page->index;
858 last_block = block_in_file + nr_pages;
859 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
860 blkbits;
861 if (last_block > last_block_in_file)
862 last_block = last_block_in_file;
863
864 /*
865 * Map blocks using the previous result first.
866 */
867 if ((map.m_flags & F2FS_MAP_MAPPED) &&
868 block_in_file > map.m_lblk &&
869 block_in_file < (map.m_lblk + map.m_len))
870 goto got_it;
871
872 /*
873 * Then do more f2fs_map_blocks() calls until we are
874 * done with this page.
875 */
876 map.m_flags = 0;
877
878 if (block_in_file < last_block) {
879 map.m_lblk = block_in_file;
880 map.m_len = last_block - block_in_file;
881
882 if (f2fs_map_blocks(inode, &map, 0, false))
883 goto set_error_page;
884 }
885 got_it:
886 if ((map.m_flags & F2FS_MAP_MAPPED)) {
887 block_nr = map.m_pblk + block_in_file - map.m_lblk;
888 SetPageMappedToDisk(page);
889
890 if (!PageUptodate(page) && !cleancache_get_page(page)) {
891 SetPageUptodate(page);
892 goto confused;
893 }
894 } else {
895 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
896 SetPageUptodate(page);
897 unlock_page(page);
898 goto next_page;
899 }
900
901 /*
902 * This page will go to BIO. Do we need to send this
903 * BIO off first?
904 */
905 if (bio && (last_block_in_bio != block_nr - 1)) {
906 submit_and_realloc:
907 submit_bio(READ, bio);
908 bio = NULL;
909 }
910 if (bio == NULL) {
911 struct f2fs_crypto_ctx *ctx = NULL;
912
913 if (f2fs_encrypted_inode(inode) &&
914 S_ISREG(inode->i_mode)) {
915 struct page *cpage;
916
917 ctx = f2fs_get_crypto_ctx(inode);
918 if (IS_ERR(ctx))
919 goto set_error_page;
920
921 /* wait the page to be moved by cleaning */
922 cpage = find_lock_page(
923 META_MAPPING(F2FS_I_SB(inode)),
924 block_nr);
925 if (cpage) {
926 f2fs_wait_on_page_writeback(cpage,
927 DATA);
928 f2fs_put_page(cpage, 1);
929 }
930 }
931
932 bio = bio_alloc(GFP_KERNEL,
933 min_t(int, nr_pages, bio_get_nr_vecs(bdev)));
934 if (!bio) {
935 if (ctx)
936 f2fs_release_crypto_ctx(ctx);
937 goto set_error_page;
938 }
939 bio->bi_bdev = bdev;
940 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
941 bio->bi_end_io = f2fs_read_end_io;
942 bio->bi_private = ctx;
943 }
944
945 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
946 goto submit_and_realloc;
947
948 last_block_in_bio = block_nr;
949 goto next_page;
950 set_error_page:
951 SetPageError(page);
952 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
953 unlock_page(page);
954 goto next_page;
955 confused:
956 if (bio) {
957 submit_bio(READ, bio);
958 bio = NULL;
959 }
960 unlock_page(page);
961 next_page:
962 if (pages)
963 page_cache_release(page);
964 }
965 BUG_ON(pages && !list_empty(pages));
966 if (bio)
967 submit_bio(READ, bio);
968 return 0;
969 }
970
971 static int f2fs_read_data_page(struct file *file, struct page *page)
972 {
973 struct inode *inode = page->mapping->host;
974 int ret = -EAGAIN;
975
976 trace_f2fs_readpage(page, DATA);
977
978 /* If the file has inline data, try to read it directly */
979 if (f2fs_has_inline_data(inode))
980 ret = f2fs_read_inline_data(inode, page);
981 if (ret == -EAGAIN)
982 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
983 return ret;
984 }
985
986 static int f2fs_read_data_pages(struct file *file,
987 struct address_space *mapping,
988 struct list_head *pages, unsigned nr_pages)
989 {
990 struct inode *inode = file->f_mapping->host;
991
992 /* If the file has inline data, skip readpages */
993 if (f2fs_has_inline_data(inode))
994 return 0;
995
996 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
997 }
998
999 int do_write_data_page(struct f2fs_io_info *fio)
1000 {
1001 struct page *page = fio->page;
1002 struct inode *inode = page->mapping->host;
1003 struct dnode_of_data dn;
1004 int err = 0;
1005
1006 set_new_dnode(&dn, inode, NULL, NULL, 0);
1007 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1008 if (err)
1009 return err;
1010
1011 fio->blk_addr = dn.data_blkaddr;
1012
1013 /* This page is already truncated */
1014 if (fio->blk_addr == NULL_ADDR) {
1015 ClearPageUptodate(page);
1016 goto out_writepage;
1017 }
1018
1019 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1020 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1021 if (IS_ERR(fio->encrypted_page)) {
1022 err = PTR_ERR(fio->encrypted_page);
1023 goto out_writepage;
1024 }
1025 }
1026
1027 set_page_writeback(page);
1028
1029 /*
1030 * If current allocation needs SSR,
1031 * it had better in-place writes for updated data.
1032 */
1033 if (unlikely(fio->blk_addr != NEW_ADDR &&
1034 !is_cold_data(page) &&
1035 need_inplace_update(inode))) {
1036 rewrite_data_page(fio);
1037 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1038 trace_f2fs_do_write_data_page(page, IPU);
1039 } else {
1040 write_data_page(&dn, fio);
1041 set_data_blkaddr(&dn);
1042 f2fs_update_extent_cache(&dn);
1043 trace_f2fs_do_write_data_page(page, OPU);
1044 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1045 if (page->index == 0)
1046 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1047 }
1048 out_writepage:
1049 f2fs_put_dnode(&dn);
1050 return err;
1051 }
1052
1053 static int f2fs_write_data_page(struct page *page,
1054 struct writeback_control *wbc)
1055 {
1056 struct inode *inode = page->mapping->host;
1057 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1058 loff_t i_size = i_size_read(inode);
1059 const pgoff_t end_index = ((unsigned long long) i_size)
1060 >> PAGE_CACHE_SHIFT;
1061 unsigned offset = 0;
1062 bool need_balance_fs = false;
1063 int err = 0;
1064 struct f2fs_io_info fio = {
1065 .sbi = sbi,
1066 .type = DATA,
1067 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1068 .page = page,
1069 .encrypted_page = NULL,
1070 };
1071
1072 trace_f2fs_writepage(page, DATA);
1073
1074 if (page->index < end_index)
1075 goto write;
1076
1077 /*
1078 * If the offset is out-of-range of file size,
1079 * this page does not have to be written to disk.
1080 */
1081 offset = i_size & (PAGE_CACHE_SIZE - 1);
1082 if ((page->index >= end_index + 1) || !offset)
1083 goto out;
1084
1085 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1086 write:
1087 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1088 goto redirty_out;
1089 if (f2fs_is_drop_cache(inode))
1090 goto out;
1091 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1092 available_free_memory(sbi, BASE_CHECK))
1093 goto redirty_out;
1094
1095 /* Dentry blocks are controlled by checkpoint */
1096 if (S_ISDIR(inode->i_mode)) {
1097 if (unlikely(f2fs_cp_error(sbi)))
1098 goto redirty_out;
1099 err = do_write_data_page(&fio);
1100 goto done;
1101 }
1102
1103 /* we should bypass data pages to proceed the kworkder jobs */
1104 if (unlikely(f2fs_cp_error(sbi))) {
1105 SetPageError(page);
1106 goto out;
1107 }
1108
1109 if (!wbc->for_reclaim)
1110 need_balance_fs = true;
1111 else if (has_not_enough_free_secs(sbi, 0))
1112 goto redirty_out;
1113
1114 err = -EAGAIN;
1115 f2fs_lock_op(sbi);
1116 if (f2fs_has_inline_data(inode))
1117 err = f2fs_write_inline_data(inode, page);
1118 if (err == -EAGAIN)
1119 err = do_write_data_page(&fio);
1120 f2fs_unlock_op(sbi);
1121 done:
1122 if (err && err != -ENOENT)
1123 goto redirty_out;
1124
1125 clear_cold_data(page);
1126 out:
1127 inode_dec_dirty_pages(inode);
1128 if (err)
1129 ClearPageUptodate(page);
1130 unlock_page(page);
1131 if (need_balance_fs)
1132 f2fs_balance_fs(sbi);
1133 if (wbc->for_reclaim)
1134 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1135 return 0;
1136
1137 redirty_out:
1138 redirty_page_for_writepage(wbc, page);
1139 return AOP_WRITEPAGE_ACTIVATE;
1140 }
1141
1142 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1143 void *data)
1144 {
1145 struct address_space *mapping = data;
1146 int ret = mapping->a_ops->writepage(page, wbc);
1147 mapping_set_error(mapping, ret);
1148 return ret;
1149 }
1150
1151 /*
1152 * This function was copied from write_cche_pages from mm/page-writeback.c.
1153 * The major change is making write step of cold data page separately from
1154 * warm/hot data page.
1155 */
1156 static int f2fs_write_cache_pages(struct address_space *mapping,
1157 struct writeback_control *wbc, writepage_t writepage,
1158 void *data)
1159 {
1160 int ret = 0;
1161 int done = 0;
1162 struct pagevec pvec;
1163 int nr_pages;
1164 pgoff_t uninitialized_var(writeback_index);
1165 pgoff_t index;
1166 pgoff_t end; /* Inclusive */
1167 pgoff_t done_index;
1168 int cycled;
1169 int range_whole = 0;
1170 int tag;
1171 int step = 0;
1172
1173 pagevec_init(&pvec, 0);
1174 next:
1175 if (wbc->range_cyclic) {
1176 writeback_index = mapping->writeback_index; /* prev offset */
1177 index = writeback_index;
1178 if (index == 0)
1179 cycled = 1;
1180 else
1181 cycled = 0;
1182 end = -1;
1183 } else {
1184 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1185 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1186 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1187 range_whole = 1;
1188 cycled = 1; /* ignore range_cyclic tests */
1189 }
1190 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1191 tag = PAGECACHE_TAG_TOWRITE;
1192 else
1193 tag = PAGECACHE_TAG_DIRTY;
1194 retry:
1195 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1196 tag_pages_for_writeback(mapping, index, end);
1197 done_index = index;
1198 while (!done && (index <= end)) {
1199 int i;
1200
1201 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1202 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1203 if (nr_pages == 0)
1204 break;
1205
1206 for (i = 0; i < nr_pages; i++) {
1207 struct page *page = pvec.pages[i];
1208
1209 if (page->index > end) {
1210 done = 1;
1211 break;
1212 }
1213
1214 done_index = page->index;
1215
1216 lock_page(page);
1217
1218 if (unlikely(page->mapping != mapping)) {
1219 continue_unlock:
1220 unlock_page(page);
1221 continue;
1222 }
1223
1224 if (!PageDirty(page)) {
1225 /* someone wrote it for us */
1226 goto continue_unlock;
1227 }
1228
1229 if (step == is_cold_data(page))
1230 goto continue_unlock;
1231
1232 if (PageWriteback(page)) {
1233 if (wbc->sync_mode != WB_SYNC_NONE)
1234 f2fs_wait_on_page_writeback(page, DATA);
1235 else
1236 goto continue_unlock;
1237 }
1238
1239 BUG_ON(PageWriteback(page));
1240 if (!clear_page_dirty_for_io(page))
1241 goto continue_unlock;
1242
1243 ret = (*writepage)(page, wbc, data);
1244 if (unlikely(ret)) {
1245 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1246 unlock_page(page);
1247 ret = 0;
1248 } else {
1249 done_index = page->index + 1;
1250 done = 1;
1251 break;
1252 }
1253 }
1254
1255 if (--wbc->nr_to_write <= 0 &&
1256 wbc->sync_mode == WB_SYNC_NONE) {
1257 done = 1;
1258 break;
1259 }
1260 }
1261 pagevec_release(&pvec);
1262 cond_resched();
1263 }
1264
1265 if (step < 1) {
1266 step++;
1267 goto next;
1268 }
1269
1270 if (!cycled && !done) {
1271 cycled = 1;
1272 index = 0;
1273 end = writeback_index - 1;
1274 goto retry;
1275 }
1276 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1277 mapping->writeback_index = done_index;
1278
1279 return ret;
1280 }
1281
1282 static int f2fs_write_data_pages(struct address_space *mapping,
1283 struct writeback_control *wbc)
1284 {
1285 struct inode *inode = mapping->host;
1286 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1287 bool locked = false;
1288 int ret;
1289 long diff;
1290
1291 trace_f2fs_writepages(mapping->host, wbc, DATA);
1292
1293 /* deal with chardevs and other special file */
1294 if (!mapping->a_ops->writepage)
1295 return 0;
1296
1297 /* skip writing if there is no dirty page in this inode */
1298 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1299 return 0;
1300
1301 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1302 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1303 available_free_memory(sbi, DIRTY_DENTS))
1304 goto skip_write;
1305
1306 /* during POR, we don't need to trigger writepage at all. */
1307 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1308 goto skip_write;
1309
1310 diff = nr_pages_to_write(sbi, DATA, wbc);
1311
1312 if (!S_ISDIR(inode->i_mode)) {
1313 mutex_lock(&sbi->writepages);
1314 locked = true;
1315 }
1316 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1317 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1318 if (locked)
1319 mutex_unlock(&sbi->writepages);
1320
1321 remove_dirty_dir_inode(inode);
1322
1323 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1324 return ret;
1325
1326 skip_write:
1327 wbc->pages_skipped += get_dirty_pages(inode);
1328 return 0;
1329 }
1330
1331 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1332 {
1333 struct inode *inode = mapping->host;
1334
1335 if (to > inode->i_size) {
1336 truncate_pagecache(inode, inode->i_size);
1337 truncate_blocks(inode, inode->i_size, true);
1338 }
1339 }
1340
1341 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1342 loff_t pos, unsigned len, unsigned flags,
1343 struct page **pagep, void **fsdata)
1344 {
1345 struct inode *inode = mapping->host;
1346 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1347 struct page *page = NULL;
1348 struct page *ipage;
1349 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1350 struct dnode_of_data dn;
1351 int err = 0;
1352
1353 trace_f2fs_write_begin(inode, pos, len, flags);
1354
1355 f2fs_balance_fs(sbi);
1356
1357 /*
1358 * We should check this at this moment to avoid deadlock on inode page
1359 * and #0 page. The locking rule for inline_data conversion should be:
1360 * lock_page(page #0) -> lock_page(inode_page)
1361 */
1362 if (index != 0) {
1363 err = f2fs_convert_inline_inode(inode);
1364 if (err)
1365 goto fail;
1366 }
1367 repeat:
1368 page = grab_cache_page_write_begin(mapping, index, flags);
1369 if (!page) {
1370 err = -ENOMEM;
1371 goto fail;
1372 }
1373
1374 *pagep = page;
1375
1376 f2fs_lock_op(sbi);
1377
1378 /* check inline_data */
1379 ipage = get_node_page(sbi, inode->i_ino);
1380 if (IS_ERR(ipage)) {
1381 err = PTR_ERR(ipage);
1382 goto unlock_fail;
1383 }
1384
1385 set_new_dnode(&dn, inode, ipage, ipage, 0);
1386
1387 if (f2fs_has_inline_data(inode)) {
1388 if (pos + len <= MAX_INLINE_DATA) {
1389 read_inline_data(page, ipage);
1390 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1391 sync_inode_page(&dn);
1392 goto put_next;
1393 }
1394 err = f2fs_convert_inline_page(&dn, page);
1395 if (err)
1396 goto put_fail;
1397 }
1398
1399 err = f2fs_get_block(&dn, index);
1400 if (err)
1401 goto put_fail;
1402 put_next:
1403 f2fs_put_dnode(&dn);
1404 f2fs_unlock_op(sbi);
1405
1406 f2fs_wait_on_page_writeback(page, DATA);
1407
1408 if (len == PAGE_CACHE_SIZE)
1409 goto out_update;
1410 if (PageUptodate(page))
1411 goto out_clear;
1412
1413 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1414 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1415 unsigned end = start + len;
1416
1417 /* Reading beyond i_size is simple: memset to zero */
1418 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1419 goto out_update;
1420 }
1421
1422 if (dn.data_blkaddr == NEW_ADDR) {
1423 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1424 } else {
1425 struct f2fs_io_info fio = {
1426 .sbi = sbi,
1427 .type = DATA,
1428 .rw = READ_SYNC,
1429 .blk_addr = dn.data_blkaddr,
1430 .page = page,
1431 .encrypted_page = NULL,
1432 };
1433 err = f2fs_submit_page_bio(&fio);
1434 if (err)
1435 goto fail;
1436
1437 lock_page(page);
1438 if (unlikely(!PageUptodate(page))) {
1439 err = -EIO;
1440 goto fail;
1441 }
1442 if (unlikely(page->mapping != mapping)) {
1443 f2fs_put_page(page, 1);
1444 goto repeat;
1445 }
1446
1447 /* avoid symlink page */
1448 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1449 err = f2fs_decrypt_one(inode, page);
1450 if (err)
1451 goto fail;
1452 }
1453 }
1454 out_update:
1455 SetPageUptodate(page);
1456 out_clear:
1457 clear_cold_data(page);
1458 return 0;
1459
1460 put_fail:
1461 f2fs_put_dnode(&dn);
1462 unlock_fail:
1463 f2fs_unlock_op(sbi);
1464 fail:
1465 f2fs_put_page(page, 1);
1466 f2fs_write_failed(mapping, pos + len);
1467 return err;
1468 }
1469
1470 static int f2fs_write_end(struct file *file,
1471 struct address_space *mapping,
1472 loff_t pos, unsigned len, unsigned copied,
1473 struct page *page, void *fsdata)
1474 {
1475 struct inode *inode = page->mapping->host;
1476
1477 trace_f2fs_write_end(inode, pos, len, copied);
1478
1479 set_page_dirty(page);
1480
1481 if (pos + copied > i_size_read(inode)) {
1482 i_size_write(inode, pos + copied);
1483 mark_inode_dirty(inode);
1484 update_inode_page(inode);
1485 }
1486
1487 f2fs_put_page(page, 1);
1488 return copied;
1489 }
1490
1491 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1492 loff_t offset)
1493 {
1494 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1495
1496 if (offset & blocksize_mask)
1497 return -EINVAL;
1498
1499 if (iov_iter_alignment(iter) & blocksize_mask)
1500 return -EINVAL;
1501
1502 return 0;
1503 }
1504
1505 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1506 loff_t offset)
1507 {
1508 struct file *file = iocb->ki_filp;
1509 struct address_space *mapping = file->f_mapping;
1510 struct inode *inode = mapping->host;
1511 size_t count = iov_iter_count(iter);
1512 int err;
1513
1514 /* we don't need to use inline_data strictly */
1515 if (f2fs_has_inline_data(inode)) {
1516 err = f2fs_convert_inline_inode(inode);
1517 if (err)
1518 return err;
1519 }
1520
1521 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1522 return 0;
1523
1524 err = check_direct_IO(inode, iter, offset);
1525 if (err)
1526 return err;
1527
1528 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1529
1530 if (iov_iter_rw(iter) == WRITE)
1531 __allocate_data_blocks(inode, offset, count);
1532
1533 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block);
1534 if (err < 0 && iov_iter_rw(iter) == WRITE)
1535 f2fs_write_failed(mapping, offset + count);
1536
1537 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1538
1539 return err;
1540 }
1541
1542 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1543 unsigned int length)
1544 {
1545 struct inode *inode = page->mapping->host;
1546 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1547
1548 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1549 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1550 return;
1551
1552 if (PageDirty(page)) {
1553 if (inode->i_ino == F2FS_META_INO(sbi))
1554 dec_page_count(sbi, F2FS_DIRTY_META);
1555 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1556 dec_page_count(sbi, F2FS_DIRTY_NODES);
1557 else
1558 inode_dec_dirty_pages(inode);
1559 }
1560
1561 /* This is atomic written page, keep Private */
1562 if (IS_ATOMIC_WRITTEN_PAGE(page))
1563 return;
1564
1565 ClearPagePrivate(page);
1566 }
1567
1568 int f2fs_release_page(struct page *page, gfp_t wait)
1569 {
1570 /* If this is dirty page, keep PagePrivate */
1571 if (PageDirty(page))
1572 return 0;
1573
1574 /* This is atomic written page, keep Private */
1575 if (IS_ATOMIC_WRITTEN_PAGE(page))
1576 return 0;
1577
1578 ClearPagePrivate(page);
1579 return 1;
1580 }
1581
1582 static int f2fs_set_data_page_dirty(struct page *page)
1583 {
1584 struct address_space *mapping = page->mapping;
1585 struct inode *inode = mapping->host;
1586
1587 trace_f2fs_set_page_dirty(page, DATA);
1588
1589 SetPageUptodate(page);
1590
1591 if (f2fs_is_atomic_file(inode)) {
1592 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1593 register_inmem_page(inode, page);
1594 return 1;
1595 }
1596 /*
1597 * Previously, this page has been registered, we just
1598 * return here.
1599 */
1600 return 0;
1601 }
1602
1603 if (!PageDirty(page)) {
1604 __set_page_dirty_nobuffers(page);
1605 update_dirty_page(inode, page);
1606 return 1;
1607 }
1608 return 0;
1609 }
1610
1611 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1612 {
1613 struct inode *inode = mapping->host;
1614
1615 /* we don't need to use inline_data strictly */
1616 if (f2fs_has_inline_data(inode)) {
1617 int err = f2fs_convert_inline_inode(inode);
1618 if (err)
1619 return err;
1620 }
1621 return generic_block_bmap(mapping, block, get_data_block);
1622 }
1623
1624 const struct address_space_operations f2fs_dblock_aops = {
1625 .readpage = f2fs_read_data_page,
1626 .readpages = f2fs_read_data_pages,
1627 .writepage = f2fs_write_data_page,
1628 .writepages = f2fs_write_data_pages,
1629 .write_begin = f2fs_write_begin,
1630 .write_end = f2fs_write_end,
1631 .set_page_dirty = f2fs_set_data_page_dirty,
1632 .invalidatepage = f2fs_invalidate_page,
1633 .releasepage = f2fs_release_page,
1634 .direct_IO = f2fs_direct_IO,
1635 .bmap = f2fs_bmap,
1636 };