projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: fix to avoid reading out encrypted data in page cache
[deliverable/linux.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)
31 {
32 struct bio_vec *bvec;
33 int i;
34
35 if (f2fs_bio_encrypted(bio)) {
36 if (bio->bi_error) {
37 fscrypt_release_ctx(bio->bi_private);
38 } else {
39 fscrypt_decrypt_bio_pages(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 (!bio->bi_error) {
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)
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 fscrypt_pullback_bio_page(&page, true);
68
69 if (unlikely(bio->bi_error)) {
70 set_bit(AS_EIO, &page->mapping->flags);
71 f2fs_stop_checkpoint(sbi, true);
72 }
73 end_page_writeback(page);
74 }
75 if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
76 wq_has_sleeper(&sbi->cp_wait))
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 bio = f2fs_bio_alloc(npages);
91
92 bio->bi_bdev = sbi->sb->s_bdev;
93 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
94 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
95 bio->bi_private = is_read ? NULL : sbi;
96
97 return bio;
98 }
99
100 static inline void __submit_bio(struct f2fs_sb_info *sbi, int rw,
101 struct bio *bio, enum page_type type)
102 {
103 if (!is_read_io(rw)) {
104 atomic_inc(&sbi->nr_wb_bios);
105 if (f2fs_sb_mounted_hmsmr(sbi->sb) &&
106 current->plug && (type == DATA || type == NODE))
107 blk_finish_plug(current->plug);
108 }
109 submit_bio(rw, bio);
110 }
111
112 static void __submit_merged_bio(struct f2fs_bio_info *io)
113 {
114 struct f2fs_io_info *fio = &io->fio;
115
116 if (!io->bio)
117 return;
118
119 if (is_read_io(fio->rw))
120 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
121 else
122 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
123
124 __submit_bio(io->sbi, fio->rw, io->bio, fio->type);
125 io->bio = NULL;
126 }
127
128 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
129 struct page *page, nid_t ino)
130 {
131 struct bio_vec *bvec;
132 struct page *target;
133 int i;
134
135 if (!io->bio)
136 return false;
137
138 if (!inode && !page && !ino)
139 return true;
140
141 bio_for_each_segment_all(bvec, io->bio, i) {
142
143 if (bvec->bv_page->mapping)
144 target = bvec->bv_page;
145 else
146 target = fscrypt_control_page(bvec->bv_page);
147
148 if (inode && inode == target->mapping->host)
149 return true;
150 if (page && page == target)
151 return true;
152 if (ino && ino == ino_of_node(target))
153 return true;
154 }
155
156 return false;
157 }
158
159 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
160 struct page *page, nid_t ino,
161 enum page_type type)
162 {
163 enum page_type btype = PAGE_TYPE_OF_BIO(type);
164 struct f2fs_bio_info *io = &sbi->write_io[btype];
165 bool ret;
166
167 down_read(&io->io_rwsem);
168 ret = __has_merged_page(io, inode, page, ino);
169 up_read(&io->io_rwsem);
170 return ret;
171 }
172
173 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
174 struct inode *inode, struct page *page,
175 nid_t ino, enum page_type type, int rw)
176 {
177 enum page_type btype = PAGE_TYPE_OF_BIO(type);
178 struct f2fs_bio_info *io;
179
180 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
181
182 down_write(&io->io_rwsem);
183
184 if (!__has_merged_page(io, inode, page, ino))
185 goto out;
186
187 /* change META to META_FLUSH in the checkpoint procedure */
188 if (type >= META_FLUSH) {
189 io->fio.type = META_FLUSH;
190 if (test_opt(sbi, NOBARRIER))
191 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
192 else
193 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
194 }
195 __submit_merged_bio(io);
196 out:
197 up_write(&io->io_rwsem);
198 }
199
200 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
201 int rw)
202 {
203 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
204 }
205
206 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
207 struct inode *inode, struct page *page,
208 nid_t ino, enum page_type type, int rw)
209 {
210 if (has_merged_page(sbi, inode, page, ino, type))
211 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
212 }
213
214 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
215 {
216 f2fs_submit_merged_bio(sbi, DATA, WRITE);
217 f2fs_submit_merged_bio(sbi, NODE, WRITE);
218 f2fs_submit_merged_bio(sbi, META, WRITE);
219 }
220
221 /*
222 * Fill the locked page with data located in the block address.
223 * Return unlocked page.
224 */
225 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
226 {
227 struct bio *bio;
228 struct page *page = fio->encrypted_page ?
229 fio->encrypted_page : fio->page;
230
231 trace_f2fs_submit_page_bio(page, fio);
232 f2fs_trace_ios(fio, 0);
233
234 /* Allocate a new bio */
235 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
236
237 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
238 bio_put(bio);
239 return -EFAULT;
240 }
241
242 __submit_bio(fio->sbi, fio->rw, bio, fio->type);
243 return 0;
244 }
245
246 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
247 {
248 struct f2fs_sb_info *sbi = fio->sbi;
249 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
250 struct f2fs_bio_info *io;
251 bool is_read = is_read_io(fio->rw);
252 struct page *bio_page;
253
254 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
255
256 if (fio->old_blkaddr != NEW_ADDR)
257 verify_block_addr(sbi, fio->old_blkaddr);
258 verify_block_addr(sbi, fio->new_blkaddr);
259
260 down_write(&io->io_rwsem);
261
262 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
263 io->fio.rw != fio->rw))
264 __submit_merged_bio(io);
265 alloc_new:
266 if (io->bio == NULL) {
267 int bio_blocks = MAX_BIO_BLOCKS(sbi);
268
269 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
270 bio_blocks, is_read);
271 io->fio = *fio;
272 }
273
274 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
275
276 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
277 PAGE_SIZE) {
278 __submit_merged_bio(io);
279 goto alloc_new;
280 }
281
282 io->last_block_in_bio = fio->new_blkaddr;
283 f2fs_trace_ios(fio, 0);
284
285 up_write(&io->io_rwsem);
286 trace_f2fs_submit_page_mbio(fio->page, fio);
287 }
288
289 static void __set_data_blkaddr(struct dnode_of_data *dn)
290 {
291 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
292 __le32 *addr_array;
293
294 /* Get physical address of data block */
295 addr_array = blkaddr_in_node(rn);
296 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
297 }
298
299 /*
300 * Lock ordering for the change of data block address:
301 * ->data_page
302 * ->node_page
303 * update block addresses in the node page
304 */
305 void set_data_blkaddr(struct dnode_of_data *dn)
306 {
307 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
308 __set_data_blkaddr(dn);
309 if (set_page_dirty(dn->node_page))
310 dn->node_changed = true;
311 }
312
313 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
314 {
315 dn->data_blkaddr = blkaddr;
316 set_data_blkaddr(dn);
317 f2fs_update_extent_cache(dn);
318 }
319
320 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
321 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
322 {
323 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
324
325 if (!count)
326 return 0;
327
328 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
329 return -EPERM;
330 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
331 return -ENOSPC;
332
333 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
334 dn->ofs_in_node, count);
335
336 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
337
338 for (; count > 0; dn->ofs_in_node++) {
339 block_t blkaddr =
340 datablock_addr(dn->node_page, dn->ofs_in_node);
341 if (blkaddr == NULL_ADDR) {
342 dn->data_blkaddr = NEW_ADDR;
343 __set_data_blkaddr(dn);
344 count--;
345 }
346 }
347
348 if (set_page_dirty(dn->node_page))
349 dn->node_changed = true;
350 return 0;
351 }
352
353 /* Should keep dn->ofs_in_node unchanged */
354 int reserve_new_block(struct dnode_of_data *dn)
355 {
356 unsigned int ofs_in_node = dn->ofs_in_node;
357 int ret;
358
359 ret = reserve_new_blocks(dn, 1);
360 dn->ofs_in_node = ofs_in_node;
361 return ret;
362 }
363
364 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
365 {
366 bool need_put = dn->inode_page ? false : true;
367 int err;
368
369 err = get_dnode_of_data(dn, index, ALLOC_NODE);
370 if (err)
371 return err;
372
373 if (dn->data_blkaddr == NULL_ADDR)
374 err = reserve_new_block(dn);
375 if (err || need_put)
376 f2fs_put_dnode(dn);
377 return err;
378 }
379
380 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
381 {
382 struct extent_info ei;
383 struct inode *inode = dn->inode;
384
385 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
386 dn->data_blkaddr = ei.blk + index - ei.fofs;
387 return 0;
388 }
389
390 return f2fs_reserve_block(dn, index);
391 }
392
393 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
394 int rw, bool for_write)
395 {
396 struct address_space *mapping = inode->i_mapping;
397 struct dnode_of_data dn;
398 struct page *page;
399 struct extent_info ei;
400 int err;
401 struct f2fs_io_info fio = {
402 .sbi = F2FS_I_SB(inode),
403 .type = DATA,
404 .rw = rw,
405 .encrypted_page = NULL,
406 };
407
408 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
409 return read_mapping_page(mapping, index, NULL);
410
411 page = f2fs_grab_cache_page(mapping, index, for_write);
412 if (!page)
413 return ERR_PTR(-ENOMEM);
414
415 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
416 dn.data_blkaddr = ei.blk + index - ei.fofs;
417 goto got_it;
418 }
419
420 set_new_dnode(&dn, inode, NULL, NULL, 0);
421 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
422 if (err)
423 goto put_err;
424 f2fs_put_dnode(&dn);
425
426 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
427 err = -ENOENT;
428 goto put_err;
429 }
430 got_it:
431 if (PageUptodate(page)) {
432 unlock_page(page);
433 return page;
434 }
435
436 /*
437 * A new dentry page is allocated but not able to be written, since its
438 * new inode page couldn't be allocated due to -ENOSPC.
439 * In such the case, its blkaddr can be remained as NEW_ADDR.
440 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
441 */
442 if (dn.data_blkaddr == NEW_ADDR) {
443 zero_user_segment(page, 0, PAGE_SIZE);
444 SetPageUptodate(page);
445 unlock_page(page);
446 return page;
447 }
448
449 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
450 fio.page = page;
451 err = f2fs_submit_page_bio(&fio);
452 if (err)
453 goto put_err;
454 return page;
455
456 put_err:
457 f2fs_put_page(page, 1);
458 return ERR_PTR(err);
459 }
460
461 struct page *find_data_page(struct inode *inode, pgoff_t index)
462 {
463 struct address_space *mapping = inode->i_mapping;
464 struct page *page;
465
466 page = find_get_page(mapping, index);
467 if (page && PageUptodate(page))
468 return page;
469 f2fs_put_page(page, 0);
470
471 page = get_read_data_page(inode, index, READ_SYNC, false);
472 if (IS_ERR(page))
473 return page;
474
475 if (PageUptodate(page))
476 return page;
477
478 wait_on_page_locked(page);
479 if (unlikely(!PageUptodate(page))) {
480 f2fs_put_page(page, 0);
481 return ERR_PTR(-EIO);
482 }
483 return page;
484 }
485
486 /*
487 * If it tries to access a hole, return an error.
488 * Because, the callers, functions in dir.c and GC, should be able to know
489 * whether this page exists or not.
490 */
491 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
492 bool for_write)
493 {
494 struct address_space *mapping = inode->i_mapping;
495 struct page *page;
496 repeat:
497 page = get_read_data_page(inode, index, READ_SYNC, for_write);
498 if (IS_ERR(page))
499 return page;
500
501 /* wait for read completion */
502 lock_page(page);
503 if (unlikely(!PageUptodate(page))) {
504 f2fs_put_page(page, 1);
505 return ERR_PTR(-EIO);
506 }
507 if (unlikely(page->mapping != mapping)) {
508 f2fs_put_page(page, 1);
509 goto repeat;
510 }
511 return page;
512 }
513
514 /*
515 * Caller ensures that this data page is never allocated.
516 * A new zero-filled data page is allocated in the page cache.
517 *
518 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
519 * f2fs_unlock_op().
520 * Note that, ipage is set only by make_empty_dir, and if any error occur,
521 * ipage should be released by this function.
522 */
523 struct page *get_new_data_page(struct inode *inode,
524 struct page *ipage, pgoff_t index, bool new_i_size)
525 {
526 struct address_space *mapping = inode->i_mapping;
527 struct page *page;
528 struct dnode_of_data dn;
529 int err;
530
531 page = f2fs_grab_cache_page(mapping, index, true);
532 if (!page) {
533 /*
534 * before exiting, we should make sure ipage will be released
535 * if any error occur.
536 */
537 f2fs_put_page(ipage, 1);
538 return ERR_PTR(-ENOMEM);
539 }
540
541 set_new_dnode(&dn, inode, ipage, NULL, 0);
542 err = f2fs_reserve_block(&dn, index);
543 if (err) {
544 f2fs_put_page(page, 1);
545 return ERR_PTR(err);
546 }
547 if (!ipage)
548 f2fs_put_dnode(&dn);
549
550 if (PageUptodate(page))
551 goto got_it;
552
553 if (dn.data_blkaddr == NEW_ADDR) {
554 zero_user_segment(page, 0, PAGE_SIZE);
555 SetPageUptodate(page);
556 } else {
557 f2fs_put_page(page, 1);
558
559 /* if ipage exists, blkaddr should be NEW_ADDR */
560 f2fs_bug_on(F2FS_I_SB(inode), ipage);
561 page = get_lock_data_page(inode, index, true);
562 if (IS_ERR(page))
563 return page;
564 }
565 got_it:
566 if (new_i_size && i_size_read(inode) <
567 ((loff_t)(index + 1) << PAGE_SHIFT))
568 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
569 return page;
570 }
571
572 static int __allocate_data_block(struct dnode_of_data *dn)
573 {
574 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
575 struct f2fs_summary sum;
576 struct node_info ni;
577 int seg = CURSEG_WARM_DATA;
578 pgoff_t fofs;
579 blkcnt_t count = 1;
580
581 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
582 return -EPERM;
583
584 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
585 if (dn->data_blkaddr == NEW_ADDR)
586 goto alloc;
587
588 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
589 return -ENOSPC;
590
591 alloc:
592 get_node_info(sbi, dn->nid, &ni);
593 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
594
595 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
596 seg = CURSEG_DIRECT_IO;
597
598 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
599 &sum, seg);
600 set_data_blkaddr(dn);
601
602 /* update i_size */
603 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
604 dn->ofs_in_node;
605 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
606 f2fs_i_size_write(dn->inode,
607 ((loff_t)(fofs + 1) << PAGE_SHIFT));
608 return 0;
609 }
610
611 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
612 {
613 struct inode *inode = file_inode(iocb->ki_filp);
614 struct f2fs_map_blocks map;
615 ssize_t ret = 0;
616
617 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
618 map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
619 map.m_next_pgofs = NULL;
620
621 if (f2fs_encrypted_inode(inode))
622 return 0;
623
624 if (iocb->ki_flags & IOCB_DIRECT) {
625 ret = f2fs_convert_inline_inode(inode);
626 if (ret)
627 return ret;
628 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
629 }
630 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
631 ret = f2fs_convert_inline_inode(inode);
632 if (ret)
633 return ret;
634 }
635 if (!f2fs_has_inline_data(inode))
636 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
637 return ret;
638 }
639
640 /*
641 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
642 * f2fs_map_blocks structure.
643 * If original data blocks are allocated, then give them to blockdev.
644 * Otherwise,
645 * a. preallocate requested block addresses
646 * b. do not use extent cache for better performance
647 * c. give the block addresses to blockdev
648 */
649 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
650 int create, int flag)
651 {
652 unsigned int maxblocks = map->m_len;
653 struct dnode_of_data dn;
654 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
655 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
656 pgoff_t pgofs, end_offset, end;
657 int err = 0, ofs = 1;
658 unsigned int ofs_in_node, last_ofs_in_node;
659 blkcnt_t prealloc;
660 struct extent_info ei;
661 bool allocated = false;
662 block_t blkaddr;
663
664 map->m_len = 0;
665 map->m_flags = 0;
666
667 /* it only supports block size == page size */
668 pgofs = (pgoff_t)map->m_lblk;
669 end = pgofs + maxblocks;
670
671 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
672 map->m_pblk = ei.blk + pgofs - ei.fofs;
673 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
674 map->m_flags = F2FS_MAP_MAPPED;
675 goto out;
676 }
677
678 next_dnode:
679 if (create)
680 f2fs_lock_op(sbi);
681
682 /* When reading holes, we need its node page */
683 set_new_dnode(&dn, inode, NULL, NULL, 0);
684 err = get_dnode_of_data(&dn, pgofs, mode);
685 if (err) {
686 if (flag == F2FS_GET_BLOCK_BMAP)
687 map->m_pblk = 0;
688 if (err == -ENOENT) {
689 err = 0;
690 if (map->m_next_pgofs)
691 *map->m_next_pgofs =
692 get_next_page_offset(&dn, pgofs);
693 }
694 goto unlock_out;
695 }
696
697 prealloc = 0;
698 ofs_in_node = dn.ofs_in_node;
699 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
700
701 next_block:
702 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
703
704 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
705 if (create) {
706 if (unlikely(f2fs_cp_error(sbi))) {
707 err = -EIO;
708 goto sync_out;
709 }
710 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
711 if (blkaddr == NULL_ADDR) {
712 prealloc++;
713 last_ofs_in_node = dn.ofs_in_node;
714 }
715 } else {
716 err = __allocate_data_block(&dn);
717 if (!err) {
718 set_inode_flag(inode, FI_APPEND_WRITE);
719 allocated = true;
720 }
721 }
722 if (err)
723 goto sync_out;
724 map->m_flags = F2FS_MAP_NEW;
725 blkaddr = dn.data_blkaddr;
726 } else {
727 if (flag == F2FS_GET_BLOCK_BMAP) {
728 map->m_pblk = 0;
729 goto sync_out;
730 }
731 if (flag == F2FS_GET_BLOCK_FIEMAP &&
732 blkaddr == NULL_ADDR) {
733 if (map->m_next_pgofs)
734 *map->m_next_pgofs = pgofs + 1;
735 }
736 if (flag != F2FS_GET_BLOCK_FIEMAP ||
737 blkaddr != NEW_ADDR)
738 goto sync_out;
739 }
740 }
741
742 if (flag == F2FS_GET_BLOCK_PRE_AIO)
743 goto skip;
744
745 if (map->m_len == 0) {
746 /* preallocated unwritten block should be mapped for fiemap. */
747 if (blkaddr == NEW_ADDR)
748 map->m_flags |= F2FS_MAP_UNWRITTEN;
749 map->m_flags |= F2FS_MAP_MAPPED;
750
751 map->m_pblk = blkaddr;
752 map->m_len = 1;
753 } else if ((map->m_pblk != NEW_ADDR &&
754 blkaddr == (map->m_pblk + ofs)) ||
755 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
756 flag == F2FS_GET_BLOCK_PRE_DIO) {
757 ofs++;
758 map->m_len++;
759 } else {
760 goto sync_out;
761 }
762
763 skip:
764 dn.ofs_in_node++;
765 pgofs++;
766
767 /* preallocate blocks in batch for one dnode page */
768 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
769 (pgofs == end || dn.ofs_in_node == end_offset)) {
770
771 dn.ofs_in_node = ofs_in_node;
772 err = reserve_new_blocks(&dn, prealloc);
773 if (err)
774 goto sync_out;
775
776 map->m_len += dn.ofs_in_node - ofs_in_node;
777 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
778 err = -ENOSPC;
779 goto sync_out;
780 }
781 dn.ofs_in_node = end_offset;
782 }
783
784 if (pgofs >= end)
785 goto sync_out;
786 else if (dn.ofs_in_node < end_offset)
787 goto next_block;
788
789 f2fs_put_dnode(&dn);
790
791 if (create) {
792 f2fs_unlock_op(sbi);
793 f2fs_balance_fs(sbi, allocated);
794 }
795 allocated = false;
796 goto next_dnode;
797
798 sync_out:
799 f2fs_put_dnode(&dn);
800 unlock_out:
801 if (create) {
802 f2fs_unlock_op(sbi);
803 f2fs_balance_fs(sbi, allocated);
804 }
805 out:
806 trace_f2fs_map_blocks(inode, map, err);
807 return err;
808 }
809
810 static int __get_data_block(struct inode *inode, sector_t iblock,
811 struct buffer_head *bh, int create, int flag,
812 pgoff_t *next_pgofs)
813 {
814 struct f2fs_map_blocks map;
815 int ret;
816
817 map.m_lblk = iblock;
818 map.m_len = bh->b_size >> inode->i_blkbits;
819 map.m_next_pgofs = next_pgofs;
820
821 ret = f2fs_map_blocks(inode, &map, create, flag);
822 if (!ret) {
823 map_bh(bh, inode->i_sb, map.m_pblk);
824 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
825 bh->b_size = map.m_len << inode->i_blkbits;
826 }
827 return ret;
828 }
829
830 static int get_data_block(struct inode *inode, sector_t iblock,
831 struct buffer_head *bh_result, int create, int flag,
832 pgoff_t *next_pgofs)
833 {
834 return __get_data_block(inode, iblock, bh_result, create,
835 flag, next_pgofs);
836 }
837
838 static int get_data_block_dio(struct inode *inode, sector_t iblock,
839 struct buffer_head *bh_result, int create)
840 {
841 return __get_data_block(inode, iblock, bh_result, create,
842 F2FS_GET_BLOCK_DIO, NULL);
843 }
844
845 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
846 struct buffer_head *bh_result, int create)
847 {
848 /* Block number less than F2FS MAX BLOCKS */
849 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
850 return -EFBIG;
851
852 return __get_data_block(inode, iblock, bh_result, create,
853 F2FS_GET_BLOCK_BMAP, NULL);
854 }
855
856 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
857 {
858 return (offset >> inode->i_blkbits);
859 }
860
861 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
862 {
863 return (blk << inode->i_blkbits);
864 }
865
866 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
867 u64 start, u64 len)
868 {
869 struct buffer_head map_bh;
870 sector_t start_blk, last_blk;
871 pgoff_t next_pgofs;
872 loff_t isize;
873 u64 logical = 0, phys = 0, size = 0;
874 u32 flags = 0;
875 int ret = 0;
876
877 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
878 if (ret)
879 return ret;
880
881 if (f2fs_has_inline_data(inode)) {
882 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
883 if (ret != -EAGAIN)
884 return ret;
885 }
886
887 inode_lock(inode);
888
889 isize = i_size_read(inode);
890 if (start >= isize)
891 goto out;
892
893 if (start + len > isize)
894 len = isize - start;
895
896 if (logical_to_blk(inode, len) == 0)
897 len = blk_to_logical(inode, 1);
898
899 start_blk = logical_to_blk(inode, start);
900 last_blk = logical_to_blk(inode, start + len - 1);
901
902 next:
903 memset(&map_bh, 0, sizeof(struct buffer_head));
904 map_bh.b_size = len;
905
906 ret = get_data_block(inode, start_blk, &map_bh, 0,
907 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
908 if (ret)
909 goto out;
910
911 /* HOLE */
912 if (!buffer_mapped(&map_bh)) {
913 start_blk = next_pgofs;
914 /* Go through holes util pass the EOF */
915 if (blk_to_logical(inode, start_blk) < isize)
916 goto prep_next;
917 /* Found a hole beyond isize means no more extents.
918 * Note that the premise is that filesystems don't
919 * punch holes beyond isize and keep size unchanged.
920 */
921 flags |= FIEMAP_EXTENT_LAST;
922 }
923
924 if (size) {
925 if (f2fs_encrypted_inode(inode))
926 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
927
928 ret = fiemap_fill_next_extent(fieinfo, logical,
929 phys, size, flags);
930 }
931
932 if (start_blk > last_blk || ret)
933 goto out;
934
935 logical = blk_to_logical(inode, start_blk);
936 phys = blk_to_logical(inode, map_bh.b_blocknr);
937 size = map_bh.b_size;
938 flags = 0;
939 if (buffer_unwritten(&map_bh))
940 flags = FIEMAP_EXTENT_UNWRITTEN;
941
942 start_blk += logical_to_blk(inode, size);
943
944 prep_next:
945 cond_resched();
946 if (fatal_signal_pending(current))
947 ret = -EINTR;
948 else
949 goto next;
950 out:
951 if (ret == 1)
952 ret = 0;
953
954 inode_unlock(inode);
955 return ret;
956 }
957
958 struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
959 unsigned nr_pages)
960 {
961 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
962 struct fscrypt_ctx *ctx = NULL;
963 struct block_device *bdev = sbi->sb->s_bdev;
964 struct bio *bio;
965
966 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
967 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
968 if (IS_ERR(ctx))
969 return ERR_CAST(ctx);
970
971 /* wait the page to be moved by cleaning */
972 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
973 }
974
975 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
976 if (!bio) {
977 if (ctx)
978 fscrypt_release_ctx(ctx);
979 return ERR_PTR(-ENOMEM);
980 }
981 bio->bi_bdev = bdev;
982 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
983 bio->bi_end_io = f2fs_read_end_io;
984 bio->bi_private = ctx;
985
986 return bio;
987 }
988
989 /*
990 * This function was originally taken from fs/mpage.c, and customized for f2fs.
991 * Major change was from block_size == page_size in f2fs by default.
992 */
993 static int f2fs_mpage_readpages(struct address_space *mapping,
994 struct list_head *pages, struct page *page,
995 unsigned nr_pages)
996 {
997 struct bio *bio = NULL;
998 unsigned page_idx;
999 sector_t last_block_in_bio = 0;
1000 struct inode *inode = mapping->host;
1001 const unsigned blkbits = inode->i_blkbits;
1002 const unsigned blocksize = 1 << blkbits;
1003 sector_t block_in_file;
1004 sector_t last_block;
1005 sector_t last_block_in_file;
1006 sector_t block_nr;
1007 struct f2fs_map_blocks map;
1008
1009 map.m_pblk = 0;
1010 map.m_lblk = 0;
1011 map.m_len = 0;
1012 map.m_flags = 0;
1013 map.m_next_pgofs = NULL;
1014
1015 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1016
1017 prefetchw(&page->flags);
1018 if (pages) {
1019 page = list_entry(pages->prev, struct page, lru);
1020 list_del(&page->lru);
1021 if (add_to_page_cache_lru(page, mapping,
1022 page->index, GFP_KERNEL))
1023 goto next_page;
1024 }
1025
1026 block_in_file = (sector_t)page->index;
1027 last_block = block_in_file + nr_pages;
1028 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1029 blkbits;
1030 if (last_block > last_block_in_file)
1031 last_block = last_block_in_file;
1032
1033 /*
1034 * Map blocks using the previous result first.
1035 */
1036 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1037 block_in_file > map.m_lblk &&
1038 block_in_file < (map.m_lblk + map.m_len))
1039 goto got_it;
1040
1041 /*
1042 * Then do more f2fs_map_blocks() calls until we are
1043 * done with this page.
1044 */
1045 map.m_flags = 0;
1046
1047 if (block_in_file < last_block) {
1048 map.m_lblk = block_in_file;
1049 map.m_len = last_block - block_in_file;
1050
1051 if (f2fs_map_blocks(inode, &map, 0,
1052 F2FS_GET_BLOCK_READ))
1053 goto set_error_page;
1054 }
1055 got_it:
1056 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1057 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1058 SetPageMappedToDisk(page);
1059
1060 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1061 SetPageUptodate(page);
1062 goto confused;
1063 }
1064 } else {
1065 zero_user_segment(page, 0, PAGE_SIZE);
1066 SetPageUptodate(page);
1067 unlock_page(page);
1068 goto next_page;
1069 }
1070
1071 /*
1072 * This page will go to BIO. Do we need to send this
1073 * BIO off first?
1074 */
1075 if (bio && (last_block_in_bio != block_nr - 1)) {
1076 submit_and_realloc:
1077 __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1078 bio = NULL;
1079 }
1080 if (bio == NULL) {
1081 bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1082 if (IS_ERR(bio))
1083 goto set_error_page;
1084 }
1085
1086 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1087 goto submit_and_realloc;
1088
1089 last_block_in_bio = block_nr;
1090 goto next_page;
1091 set_error_page:
1092 SetPageError(page);
1093 zero_user_segment(page, 0, PAGE_SIZE);
1094 unlock_page(page);
1095 goto next_page;
1096 confused:
1097 if (bio) {
1098 __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1099 bio = NULL;
1100 }
1101 unlock_page(page);
1102 next_page:
1103 if (pages)
1104 put_page(page);
1105 }
1106 BUG_ON(pages && !list_empty(pages));
1107 if (bio)
1108 __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1109 return 0;
1110 }
1111
1112 static int f2fs_read_data_page(struct file *file, struct page *page)
1113 {
1114 struct inode *inode = page->mapping->host;
1115 int ret = -EAGAIN;
1116
1117 trace_f2fs_readpage(page, DATA);
1118
1119 /* If the file has inline data, try to read it directly */
1120 if (f2fs_has_inline_data(inode))
1121 ret = f2fs_read_inline_data(inode, page);
1122 if (ret == -EAGAIN)
1123 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1124 return ret;
1125 }
1126
1127 static int f2fs_read_data_pages(struct file *file,
1128 struct address_space *mapping,
1129 struct list_head *pages, unsigned nr_pages)
1130 {
1131 struct inode *inode = file->f_mapping->host;
1132 struct page *page = list_entry(pages->prev, struct page, lru);
1133
1134 trace_f2fs_readpages(inode, page, nr_pages);
1135
1136 /* If the file has inline data, skip readpages */
1137 if (f2fs_has_inline_data(inode))
1138 return 0;
1139
1140 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1141 }
1142
1143 int do_write_data_page(struct f2fs_io_info *fio)
1144 {
1145 struct page *page = fio->page;
1146 struct inode *inode = page->mapping->host;
1147 struct dnode_of_data dn;
1148 int err = 0;
1149
1150 set_new_dnode(&dn, inode, NULL, NULL, 0);
1151 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1152 if (err)
1153 return err;
1154
1155 fio->old_blkaddr = dn.data_blkaddr;
1156
1157 /* This page is already truncated */
1158 if (fio->old_blkaddr == NULL_ADDR) {
1159 ClearPageUptodate(page);
1160 goto out_writepage;
1161 }
1162
1163 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1164 gfp_t gfp_flags = GFP_NOFS;
1165
1166 /* wait for GCed encrypted page writeback */
1167 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1168 fio->old_blkaddr);
1169 retry_encrypt:
1170 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1171 gfp_flags);
1172 if (IS_ERR(fio->encrypted_page)) {
1173 err = PTR_ERR(fio->encrypted_page);
1174 if (err == -ENOMEM) {
1175 /* flush pending ios and wait for a while */
1176 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1177 congestion_wait(BLK_RW_ASYNC, HZ/50);
1178 gfp_flags |= __GFP_NOFAIL;
1179 err = 0;
1180 goto retry_encrypt;
1181 }
1182 goto out_writepage;
1183 }
1184 }
1185
1186 set_page_writeback(page);
1187
1188 /*
1189 * If current allocation needs SSR,
1190 * it had better in-place writes for updated data.
1191 */
1192 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1193 !is_cold_data(page) &&
1194 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1195 need_inplace_update(inode))) {
1196 rewrite_data_page(fio);
1197 set_inode_flag(inode, FI_UPDATE_WRITE);
1198 trace_f2fs_do_write_data_page(page, IPU);
1199 } else {
1200 write_data_page(&dn, fio);
1201 trace_f2fs_do_write_data_page(page, OPU);
1202 set_inode_flag(inode, FI_APPEND_WRITE);
1203 if (page->index == 0)
1204 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1205 }
1206 out_writepage:
1207 f2fs_put_dnode(&dn);
1208 return err;
1209 }
1210
1211 static int f2fs_write_data_page(struct page *page,
1212 struct writeback_control *wbc)
1213 {
1214 struct inode *inode = page->mapping->host;
1215 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1216 loff_t i_size = i_size_read(inode);
1217 const pgoff_t end_index = ((unsigned long long) i_size)
1218 >> PAGE_SHIFT;
1219 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1220 unsigned offset = 0;
1221 bool need_balance_fs = false;
1222 int err = 0;
1223 struct f2fs_io_info fio = {
1224 .sbi = sbi,
1225 .type = DATA,
1226 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1227 .page = page,
1228 .encrypted_page = NULL,
1229 };
1230
1231 trace_f2fs_writepage(page, DATA);
1232
1233 if (page->index < end_index)
1234 goto write;
1235
1236 /*
1237 * If the offset is out-of-range of file size,
1238 * this page does not have to be written to disk.
1239 */
1240 offset = i_size & (PAGE_SIZE - 1);
1241 if ((page->index >= end_index + 1) || !offset)
1242 goto out;
1243
1244 zero_user_segment(page, offset, PAGE_SIZE);
1245 write:
1246 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1247 goto redirty_out;
1248 if (f2fs_is_drop_cache(inode))
1249 goto out;
1250 /* we should not write 0'th page having journal header */
1251 if (f2fs_is_volatile_file(inode) && (!page->index ||
1252 (!wbc->for_reclaim &&
1253 available_free_memory(sbi, BASE_CHECK))))
1254 goto redirty_out;
1255
1256 /* we should bypass data pages to proceed the kworkder jobs */
1257 if (unlikely(f2fs_cp_error(sbi))) {
1258 mapping_set_error(page->mapping, -EIO);
1259 goto out;
1260 }
1261
1262 /* Dentry blocks are controlled by checkpoint */
1263 if (S_ISDIR(inode->i_mode)) {
1264 err = do_write_data_page(&fio);
1265 goto done;
1266 }
1267
1268 if (!wbc->for_reclaim)
1269 need_balance_fs = true;
1270 else if (has_not_enough_free_secs(sbi, 0))
1271 goto redirty_out;
1272
1273 err = -EAGAIN;
1274 f2fs_lock_op(sbi);
1275 if (f2fs_has_inline_data(inode))
1276 err = f2fs_write_inline_data(inode, page);
1277 if (err == -EAGAIN)
1278 err = do_write_data_page(&fio);
1279 if (F2FS_I(inode)->last_disk_size < psize)
1280 F2FS_I(inode)->last_disk_size = psize;
1281 f2fs_unlock_op(sbi);
1282 done:
1283 if (err && err != -ENOENT)
1284 goto redirty_out;
1285
1286 clear_cold_data(page);
1287 out:
1288 inode_dec_dirty_pages(inode);
1289 if (err)
1290 ClearPageUptodate(page);
1291
1292 if (wbc->for_reclaim) {
1293 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1294 remove_dirty_inode(inode);
1295 }
1296
1297 unlock_page(page);
1298 f2fs_balance_fs(sbi, need_balance_fs);
1299
1300 if (unlikely(f2fs_cp_error(sbi)))
1301 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1302
1303 return 0;
1304
1305 redirty_out:
1306 redirty_page_for_writepage(wbc, page);
1307 unlock_page(page);
1308 return err;
1309 }
1310
1311 /*
1312 * This function was copied from write_cche_pages from mm/page-writeback.c.
1313 * The major change is making write step of cold data page separately from
1314 * warm/hot data page.
1315 */
1316 static int f2fs_write_cache_pages(struct address_space *mapping,
1317 struct writeback_control *wbc)
1318 {
1319 int ret = 0;
1320 int done = 0;
1321 struct pagevec pvec;
1322 int nr_pages;
1323 pgoff_t uninitialized_var(writeback_index);
1324 pgoff_t index;
1325 pgoff_t end; /* Inclusive */
1326 pgoff_t done_index;
1327 int cycled;
1328 int range_whole = 0;
1329 int tag;
1330
1331 pagevec_init(&pvec, 0);
1332
1333 if (wbc->range_cyclic) {
1334 writeback_index = mapping->writeback_index; /* prev offset */
1335 index = writeback_index;
1336 if (index == 0)
1337 cycled = 1;
1338 else
1339 cycled = 0;
1340 end = -1;
1341 } else {
1342 index = wbc->range_start >> PAGE_SHIFT;
1343 end = wbc->range_end >> PAGE_SHIFT;
1344 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1345 range_whole = 1;
1346 cycled = 1; /* ignore range_cyclic tests */
1347 }
1348 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1349 tag = PAGECACHE_TAG_TOWRITE;
1350 else
1351 tag = PAGECACHE_TAG_DIRTY;
1352 retry:
1353 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1354 tag_pages_for_writeback(mapping, index, end);
1355 done_index = index;
1356 while (!done && (index <= end)) {
1357 int i;
1358
1359 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1360 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1361 if (nr_pages == 0)
1362 break;
1363
1364 for (i = 0; i < nr_pages; i++) {
1365 struct page *page = pvec.pages[i];
1366
1367 if (page->index > end) {
1368 done = 1;
1369 break;
1370 }
1371
1372 done_index = page->index;
1373
1374 lock_page(page);
1375
1376 if (unlikely(page->mapping != mapping)) {
1377 continue_unlock:
1378 unlock_page(page);
1379 continue;
1380 }
1381
1382 if (!PageDirty(page)) {
1383 /* someone wrote it for us */
1384 goto continue_unlock;
1385 }
1386
1387 if (PageWriteback(page)) {
1388 if (wbc->sync_mode != WB_SYNC_NONE)
1389 f2fs_wait_on_page_writeback(page,
1390 DATA, true);
1391 else
1392 goto continue_unlock;
1393 }
1394
1395 BUG_ON(PageWriteback(page));
1396 if (!clear_page_dirty_for_io(page))
1397 goto continue_unlock;
1398
1399 ret = mapping->a_ops->writepage(page, wbc);
1400 if (unlikely(ret)) {
1401 done_index = page->index + 1;
1402 done = 1;
1403 break;
1404 }
1405
1406 if (--wbc->nr_to_write <= 0 &&
1407 wbc->sync_mode == WB_SYNC_NONE) {
1408 done = 1;
1409 break;
1410 }
1411 }
1412 pagevec_release(&pvec);
1413 cond_resched();
1414 }
1415
1416 if (!cycled && !done) {
1417 cycled = 1;
1418 index = 0;
1419 end = writeback_index - 1;
1420 goto retry;
1421 }
1422 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1423 mapping->writeback_index = done_index;
1424
1425 return ret;
1426 }
1427
1428 static int f2fs_write_data_pages(struct address_space *mapping,
1429 struct writeback_control *wbc)
1430 {
1431 struct inode *inode = mapping->host;
1432 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1433 int ret;
1434
1435 /* deal with chardevs and other special file */
1436 if (!mapping->a_ops->writepage)
1437 return 0;
1438
1439 /* skip writing if there is no dirty page in this inode */
1440 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1441 return 0;
1442
1443 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1444 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1445 available_free_memory(sbi, DIRTY_DENTS))
1446 goto skip_write;
1447
1448 /* skip writing during file defragment */
1449 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1450 goto skip_write;
1451
1452 /* during POR, we don't need to trigger writepage at all. */
1453 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1454 goto skip_write;
1455
1456 trace_f2fs_writepages(mapping->host, wbc, DATA);
1457
1458 ret = f2fs_write_cache_pages(mapping, wbc);
1459 /*
1460 * if some pages were truncated, we cannot guarantee its mapping->host
1461 * to detect pending bios.
1462 */
1463 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1464
1465 remove_dirty_inode(inode);
1466 return ret;
1467
1468 skip_write:
1469 wbc->pages_skipped += get_dirty_pages(inode);
1470 trace_f2fs_writepages(mapping->host, wbc, DATA);
1471 return 0;
1472 }
1473
1474 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1475 {
1476 struct inode *inode = mapping->host;
1477 loff_t i_size = i_size_read(inode);
1478
1479 if (to > i_size) {
1480 truncate_pagecache(inode, i_size);
1481 truncate_blocks(inode, i_size, true);
1482 }
1483 }
1484
1485 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1486 struct page *page, loff_t pos, unsigned len,
1487 block_t *blk_addr, bool *node_changed)
1488 {
1489 struct inode *inode = page->mapping->host;
1490 pgoff_t index = page->index;
1491 struct dnode_of_data dn;
1492 struct page *ipage;
1493 bool locked = false;
1494 struct extent_info ei;
1495 int err = 0;
1496
1497 /*
1498 * we already allocated all the blocks, so we don't need to get
1499 * the block addresses when there is no need to fill the page.
1500 */
1501 if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1502 len == PAGE_SIZE)
1503 return 0;
1504
1505 if (f2fs_has_inline_data(inode) ||
1506 (pos & PAGE_MASK) >= i_size_read(inode)) {
1507 f2fs_lock_op(sbi);
1508 locked = true;
1509 }
1510 restart:
1511 /* check inline_data */
1512 ipage = get_node_page(sbi, inode->i_ino);
1513 if (IS_ERR(ipage)) {
1514 err = PTR_ERR(ipage);
1515 goto unlock_out;
1516 }
1517
1518 set_new_dnode(&dn, inode, ipage, ipage, 0);
1519
1520 if (f2fs_has_inline_data(inode)) {
1521 if (pos + len <= MAX_INLINE_DATA) {
1522 read_inline_data(page, ipage);
1523 set_inode_flag(inode, FI_DATA_EXIST);
1524 if (inode->i_nlink)
1525 set_inline_node(ipage);
1526 } else {
1527 err = f2fs_convert_inline_page(&dn, page);
1528 if (err)
1529 goto out;
1530 if (dn.data_blkaddr == NULL_ADDR)
1531 err = f2fs_get_block(&dn, index);
1532 }
1533 } else if (locked) {
1534 err = f2fs_get_block(&dn, index);
1535 } else {
1536 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1537 dn.data_blkaddr = ei.blk + index - ei.fofs;
1538 } else {
1539 /* hole case */
1540 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1541 if (err || dn.data_blkaddr == NULL_ADDR) {
1542 f2fs_put_dnode(&dn);
1543 f2fs_lock_op(sbi);
1544 locked = true;
1545 goto restart;
1546 }
1547 }
1548 }
1549
1550 /* convert_inline_page can make node_changed */
1551 *blk_addr = dn.data_blkaddr;
1552 *node_changed = dn.node_changed;
1553 out:
1554 f2fs_put_dnode(&dn);
1555 unlock_out:
1556 if (locked)
1557 f2fs_unlock_op(sbi);
1558 return err;
1559 }
1560
1561 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1562 loff_t pos, unsigned len, unsigned flags,
1563 struct page **pagep, void **fsdata)
1564 {
1565 struct inode *inode = mapping->host;
1566 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1567 struct page *page = NULL;
1568 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1569 bool need_balance = false;
1570 block_t blkaddr = NULL_ADDR;
1571 int err = 0;
1572
1573 trace_f2fs_write_begin(inode, pos, len, flags);
1574
1575 /*
1576 * We should check this at this moment to avoid deadlock on inode page
1577 * and #0 page. The locking rule for inline_data conversion should be:
1578 * lock_page(page #0) -> lock_page(inode_page)
1579 */
1580 if (index != 0) {
1581 err = f2fs_convert_inline_inode(inode);
1582 if (err)
1583 goto fail;
1584 }
1585 repeat:
1586 page = grab_cache_page_write_begin(mapping, index, flags);
1587 if (!page) {
1588 err = -ENOMEM;
1589 goto fail;
1590 }
1591
1592 *pagep = page;
1593
1594 err = prepare_write_begin(sbi, page, pos, len,
1595 &blkaddr, &need_balance);
1596 if (err)
1597 goto fail;
1598
1599 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1600 unlock_page(page);
1601 f2fs_balance_fs(sbi, true);
1602 lock_page(page);
1603 if (page->mapping != mapping) {
1604 /* The page got truncated from under us */
1605 f2fs_put_page(page, 1);
1606 goto repeat;
1607 }
1608 }
1609
1610 f2fs_wait_on_page_writeback(page, DATA, false);
1611
1612 /* wait for GCed encrypted page writeback */
1613 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1614 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1615
1616 if (len == PAGE_SIZE)
1617 goto out_update;
1618 if (PageUptodate(page))
1619 goto out_clear;
1620
1621 if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1622 unsigned start = pos & (PAGE_SIZE - 1);
1623 unsigned end = start + len;
1624
1625 /* Reading beyond i_size is simple: memset to zero */
1626 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1627 goto out_update;
1628 }
1629
1630 if (blkaddr == NEW_ADDR) {
1631 zero_user_segment(page, 0, PAGE_SIZE);
1632 } else {
1633 struct bio *bio;
1634
1635 bio = f2fs_grab_bio(inode, blkaddr, 1);
1636 if (IS_ERR(bio)) {
1637 err = PTR_ERR(bio);
1638 goto fail;
1639 }
1640
1641 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1642 bio_put(bio);
1643 err = -EFAULT;
1644 goto fail;
1645 }
1646
1647 __submit_bio(sbi, READ_SYNC, bio, DATA);
1648
1649 lock_page(page);
1650 if (unlikely(!PageUptodate(page))) {
1651 err = -EIO;
1652 goto fail;
1653 }
1654 if (unlikely(page->mapping != mapping)) {
1655 f2fs_put_page(page, 1);
1656 goto repeat;
1657 }
1658 }
1659 out_update:
1660 SetPageUptodate(page);
1661 out_clear:
1662 clear_cold_data(page);
1663 return 0;
1664
1665 fail:
1666 f2fs_put_page(page, 1);
1667 f2fs_write_failed(mapping, pos + len);
1668 return err;
1669 }
1670
1671 static int f2fs_write_end(struct file *file,
1672 struct address_space *mapping,
1673 loff_t pos, unsigned len, unsigned copied,
1674 struct page *page, void *fsdata)
1675 {
1676 struct inode *inode = page->mapping->host;
1677
1678 trace_f2fs_write_end(inode, pos, len, copied);
1679
1680 set_page_dirty(page);
1681
1682 if (pos + copied > i_size_read(inode))
1683 f2fs_i_size_write(inode, pos + copied);
1684
1685 f2fs_put_page(page, 1);
1686 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1687 return copied;
1688 }
1689
1690 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1691 loff_t offset)
1692 {
1693 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1694
1695 if (offset & blocksize_mask)
1696 return -EINVAL;
1697
1698 if (iov_iter_alignment(iter) & blocksize_mask)
1699 return -EINVAL;
1700
1701 return 0;
1702 }
1703
1704 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1705 {
1706 struct address_space *mapping = iocb->ki_filp->f_mapping;
1707 struct inode *inode = mapping->host;
1708 size_t count = iov_iter_count(iter);
1709 loff_t offset = iocb->ki_pos;
1710 int err;
1711
1712 err = check_direct_IO(inode, iter, offset);
1713 if (err)
1714 return err;
1715
1716 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1717 return 0;
1718 if (test_opt(F2FS_I_SB(inode), LFS))
1719 return 0;
1720
1721 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1722
1723 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1724 if (iov_iter_rw(iter) == WRITE) {
1725 if (err > 0)
1726 set_inode_flag(inode, FI_UPDATE_WRITE);
1727 else if (err < 0)
1728 f2fs_write_failed(mapping, offset + count);
1729 }
1730
1731 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1732
1733 return err;
1734 }
1735
1736 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1737 unsigned int length)
1738 {
1739 struct inode *inode = page->mapping->host;
1740 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1741
1742 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1743 (offset % PAGE_SIZE || length != PAGE_SIZE))
1744 return;
1745
1746 if (PageDirty(page)) {
1747 if (inode->i_ino == F2FS_META_INO(sbi))
1748 dec_page_count(sbi, F2FS_DIRTY_META);
1749 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1750 dec_page_count(sbi, F2FS_DIRTY_NODES);
1751 else
1752 inode_dec_dirty_pages(inode);
1753 }
1754
1755 /* This is atomic written page, keep Private */
1756 if (IS_ATOMIC_WRITTEN_PAGE(page))
1757 return;
1758
1759 set_page_private(page, 0);
1760 ClearPagePrivate(page);
1761 }
1762
1763 int f2fs_release_page(struct page *page, gfp_t wait)
1764 {
1765 /* If this is dirty page, keep PagePrivate */
1766 if (PageDirty(page))
1767 return 0;
1768
1769 /* This is atomic written page, keep Private */
1770 if (IS_ATOMIC_WRITTEN_PAGE(page))
1771 return 0;
1772
1773 set_page_private(page, 0);
1774 ClearPagePrivate(page);
1775 return 1;
1776 }
1777
1778 static int f2fs_set_data_page_dirty(struct page *page)
1779 {
1780 struct address_space *mapping = page->mapping;
1781 struct inode *inode = mapping->host;
1782
1783 trace_f2fs_set_page_dirty(page, DATA);
1784
1785 SetPageUptodate(page);
1786
1787 if (f2fs_is_atomic_file(inode)) {
1788 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1789 register_inmem_page(inode, page);
1790 return 1;
1791 }
1792 /*
1793 * Previously, this page has been registered, we just
1794 * return here.
1795 */
1796 return 0;
1797 }
1798
1799 if (!PageDirty(page)) {
1800 __set_page_dirty_nobuffers(page);
1801 update_dirty_page(inode, page);
1802 return 1;
1803 }
1804 return 0;
1805 }
1806
1807 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1808 {
1809 struct inode *inode = mapping->host;
1810
1811 if (f2fs_has_inline_data(inode))
1812 return 0;
1813
1814 /* make sure allocating whole blocks */
1815 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1816 filemap_write_and_wait(mapping);
1817
1818 return generic_block_bmap(mapping, block, get_data_block_bmap);
1819 }
1820
1821 const struct address_space_operations f2fs_dblock_aops = {
1822 .readpage = f2fs_read_data_page,
1823 .readpages = f2fs_read_data_pages,
1824 .writepage = f2fs_write_data_page,
1825 .writepages = f2fs_write_data_pages,
1826 .write_begin = f2fs_write_begin,
1827 .write_end = f2fs_write_end,
1828 .set_page_dirty = f2fs_set_data_page_dirty,
1829 .invalidatepage = f2fs_invalidate_page,
1830 .releasepage = f2fs_release_page,
1831 .direct_IO = f2fs_direct_IO,
1832 .bmap = f2fs_bmap,
1833 };
Linux kernel - Deliverables
This page took 0.097331 seconds and 6 git commands to generate.