]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/f2fs/data.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: fix bugs and simplify codes of f2fs_fiemap
[mirror_ubuntu-bionic-kernel.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 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 (!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 f2fs_restore_and_release_control_page(&page);
68
69 if (unlikely(bio->bi_error)) {
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 dn->node_changed = true;
230 }
231
232 int reserve_new_block(struct dnode_of_data *dn)
233 {
234 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
235
236 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
237 return -EPERM;
238 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
239 return -ENOSPC;
240
241 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
242
243 dn->data_blkaddr = NEW_ADDR;
244 set_data_blkaddr(dn);
245 mark_inode_dirty(dn->inode);
246 sync_inode_page(dn);
247 return 0;
248 }
249
250 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
251 {
252 bool need_put = dn->inode_page ? false : true;
253 int err;
254
255 err = get_dnode_of_data(dn, index, ALLOC_NODE);
256 if (err)
257 return err;
258
259 if (dn->data_blkaddr == NULL_ADDR)
260 err = reserve_new_block(dn);
261 if (err || need_put)
262 f2fs_put_dnode(dn);
263 return err;
264 }
265
266 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
267 {
268 struct extent_info ei;
269 struct inode *inode = dn->inode;
270
271 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
272 dn->data_blkaddr = ei.blk + index - ei.fofs;
273 return 0;
274 }
275
276 return f2fs_reserve_block(dn, index);
277 }
278
279 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
280 int rw, bool for_write)
281 {
282 struct address_space *mapping = inode->i_mapping;
283 struct dnode_of_data dn;
284 struct page *page;
285 struct extent_info ei;
286 int err;
287 struct f2fs_io_info fio = {
288 .sbi = F2FS_I_SB(inode),
289 .type = DATA,
290 .rw = rw,
291 .encrypted_page = NULL,
292 };
293
294 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
295 return read_mapping_page(mapping, index, NULL);
296
297 page = f2fs_grab_cache_page(mapping, index, for_write);
298 if (!page)
299 return ERR_PTR(-ENOMEM);
300
301 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
302 dn.data_blkaddr = ei.blk + index - ei.fofs;
303 goto got_it;
304 }
305
306 set_new_dnode(&dn, inode, NULL, NULL, 0);
307 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
308 if (err)
309 goto put_err;
310 f2fs_put_dnode(&dn);
311
312 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
313 err = -ENOENT;
314 goto put_err;
315 }
316 got_it:
317 if (PageUptodate(page)) {
318 unlock_page(page);
319 return page;
320 }
321
322 /*
323 * A new dentry page is allocated but not able to be written, since its
324 * new inode page couldn't be allocated due to -ENOSPC.
325 * In such the case, its blkaddr can be remained as NEW_ADDR.
326 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
327 */
328 if (dn.data_blkaddr == NEW_ADDR) {
329 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
330 SetPageUptodate(page);
331 unlock_page(page);
332 return page;
333 }
334
335 fio.blk_addr = dn.data_blkaddr;
336 fio.page = page;
337 err = f2fs_submit_page_bio(&fio);
338 if (err)
339 goto put_err;
340 return page;
341
342 put_err:
343 f2fs_put_page(page, 1);
344 return ERR_PTR(err);
345 }
346
347 struct page *find_data_page(struct inode *inode, pgoff_t index)
348 {
349 struct address_space *mapping = inode->i_mapping;
350 struct page *page;
351
352 page = find_get_page(mapping, index);
353 if (page && PageUptodate(page))
354 return page;
355 f2fs_put_page(page, 0);
356
357 page = get_read_data_page(inode, index, READ_SYNC, false);
358 if (IS_ERR(page))
359 return page;
360
361 if (PageUptodate(page))
362 return page;
363
364 wait_on_page_locked(page);
365 if (unlikely(!PageUptodate(page))) {
366 f2fs_put_page(page, 0);
367 return ERR_PTR(-EIO);
368 }
369 return page;
370 }
371
372 /*
373 * If it tries to access a hole, return an error.
374 * Because, the callers, functions in dir.c and GC, should be able to know
375 * whether this page exists or not.
376 */
377 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
378 bool for_write)
379 {
380 struct address_space *mapping = inode->i_mapping;
381 struct page *page;
382 repeat:
383 page = get_read_data_page(inode, index, READ_SYNC, for_write);
384 if (IS_ERR(page))
385 return page;
386
387 /* wait for read completion */
388 lock_page(page);
389 if (unlikely(!PageUptodate(page))) {
390 f2fs_put_page(page, 1);
391 return ERR_PTR(-EIO);
392 }
393 if (unlikely(page->mapping != mapping)) {
394 f2fs_put_page(page, 1);
395 goto repeat;
396 }
397 return page;
398 }
399
400 /*
401 * Caller ensures that this data page is never allocated.
402 * A new zero-filled data page is allocated in the page cache.
403 *
404 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
405 * f2fs_unlock_op().
406 * Note that, ipage is set only by make_empty_dir, and if any error occur,
407 * ipage should be released by this function.
408 */
409 struct page *get_new_data_page(struct inode *inode,
410 struct page *ipage, pgoff_t index, bool new_i_size)
411 {
412 struct address_space *mapping = inode->i_mapping;
413 struct page *page;
414 struct dnode_of_data dn;
415 int err;
416 repeat:
417 page = f2fs_grab_cache_page(mapping, index, true);
418 if (!page) {
419 /*
420 * before exiting, we should make sure ipage will be released
421 * if any error occur.
422 */
423 f2fs_put_page(ipage, 1);
424 return ERR_PTR(-ENOMEM);
425 }
426
427 set_new_dnode(&dn, inode, ipage, NULL, 0);
428 err = f2fs_reserve_block(&dn, index);
429 if (err) {
430 f2fs_put_page(page, 1);
431 return ERR_PTR(err);
432 }
433 if (!ipage)
434 f2fs_put_dnode(&dn);
435
436 if (PageUptodate(page))
437 goto got_it;
438
439 if (dn.data_blkaddr == NEW_ADDR) {
440 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
441 SetPageUptodate(page);
442 } else {
443 f2fs_put_page(page, 1);
444
445 page = get_read_data_page(inode, index, READ_SYNC, true);
446 if (IS_ERR(page))
447 goto repeat;
448
449 /* wait for read completion */
450 lock_page(page);
451 }
452 got_it:
453 if (new_i_size && i_size_read(inode) <
454 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
455 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
456 /* Only the directory inode sets new_i_size */
457 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
458 }
459 return page;
460 }
461
462 static int __allocate_data_block(struct dnode_of_data *dn)
463 {
464 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
465 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
466 struct f2fs_summary sum;
467 struct node_info ni;
468 int seg = CURSEG_WARM_DATA;
469 pgoff_t fofs;
470
471 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
472 return -EPERM;
473
474 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
475 if (dn->data_blkaddr == NEW_ADDR)
476 goto alloc;
477
478 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
479 return -ENOSPC;
480
481 alloc:
482 get_node_info(sbi, dn->nid, &ni);
483 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
484
485 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
486 seg = CURSEG_DIRECT_IO;
487
488 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
489 &sum, seg);
490 set_data_blkaddr(dn);
491
492 /* update i_size */
493 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
494 dn->ofs_in_node;
495 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
496 i_size_write(dn->inode,
497 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
498 return 0;
499 }
500
501 static int __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 int err = 0;
511
512 while (len) {
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 err = get_dnode_of_data(&dn, start, ALLOC_NODE);
518 if (err)
519 goto out;
520
521 allocated = false;
522 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
523
524 while (dn.ofs_in_node < end_offset && len) {
525 block_t blkaddr;
526
527 if (unlikely(f2fs_cp_error(sbi))) {
528 err = -EIO;
529 goto sync_out;
530 }
531
532 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
533 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
534 err = __allocate_data_block(&dn);
535 if (err)
536 goto sync_out;
537 allocated = true;
538 }
539 len--;
540 start++;
541 dn.ofs_in_node++;
542 }
543
544 if (allocated)
545 sync_inode_page(&dn);
546
547 f2fs_put_dnode(&dn);
548 f2fs_unlock_op(sbi);
549
550 if (dn.node_changed)
551 f2fs_balance_fs(sbi);
552 }
553 return err;
554
555 sync_out:
556 if (allocated)
557 sync_inode_page(&dn);
558 f2fs_put_dnode(&dn);
559 out:
560 f2fs_unlock_op(sbi);
561 if (dn.node_changed)
562 f2fs_balance_fs(sbi);
563 return err;
564 }
565
566 /*
567 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
568 * f2fs_map_blocks structure.
569 * If original data blocks are allocated, then give them to blockdev.
570 * Otherwise,
571 * a. preallocate requested block addresses
572 * b. do not use extent cache for better performance
573 * c. give the block addresses to blockdev
574 */
575 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
576 int create, int flag)
577 {
578 unsigned int maxblocks = map->m_len;
579 struct dnode_of_data dn;
580 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
581 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
582 pgoff_t pgofs, end_offset;
583 int err = 0, ofs = 1;
584 struct extent_info ei;
585 bool allocated = false;
586 block_t blkaddr;
587
588 map->m_len = 0;
589 map->m_flags = 0;
590
591 /* it only supports block size == page size */
592 pgofs = (pgoff_t)map->m_lblk;
593
594 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
595 map->m_pblk = ei.blk + pgofs - ei.fofs;
596 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
597 map->m_flags = F2FS_MAP_MAPPED;
598 goto out;
599 }
600
601 if (create)
602 f2fs_lock_op(sbi);
603
604 /* When reading holes, we need its node page */
605 set_new_dnode(&dn, inode, NULL, NULL, 0);
606 err = get_dnode_of_data(&dn, pgofs, mode);
607 if (err) {
608 if (err == -ENOENT)
609 err = 0;
610 goto unlock_out;
611 }
612
613 if (dn.data_blkaddr == NEW_ADDR || dn.data_blkaddr == NULL_ADDR) {
614 if (create) {
615 if (unlikely(f2fs_cp_error(sbi))) {
616 err = -EIO;
617 goto put_out;
618 }
619 err = __allocate_data_block(&dn);
620 if (err)
621 goto put_out;
622 allocated = true;
623 map->m_flags = F2FS_MAP_NEW;
624 } else {
625 if (flag != F2FS_GET_BLOCK_FIEMAP ||
626 dn.data_blkaddr != NEW_ADDR) {
627 if (flag == F2FS_GET_BLOCK_BMAP)
628 err = -ENOENT;
629 goto put_out;
630 }
631
632 /*
633 * preallocated unwritten block should be mapped
634 * for fiemap.
635 */
636 if (dn.data_blkaddr == NEW_ADDR)
637 map->m_flags = F2FS_MAP_UNWRITTEN;
638 }
639 }
640
641 map->m_flags |= F2FS_MAP_MAPPED;
642 map->m_pblk = dn.data_blkaddr;
643 map->m_len = 1;
644
645 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
646 dn.ofs_in_node++;
647 pgofs++;
648
649 get_next:
650 if (map->m_len >= maxblocks)
651 goto sync_out;
652
653 if (dn.ofs_in_node >= end_offset) {
654 if (allocated)
655 sync_inode_page(&dn);
656 allocated = false;
657 f2fs_put_dnode(&dn);
658
659 if (create) {
660 f2fs_unlock_op(sbi);
661 if (dn.node_changed)
662 f2fs_balance_fs(sbi);
663 f2fs_lock_op(sbi);
664 }
665
666 set_new_dnode(&dn, inode, NULL, NULL, 0);
667 err = get_dnode_of_data(&dn, pgofs, mode);
668 if (err) {
669 if (err == -ENOENT)
670 err = 0;
671 goto unlock_out;
672 }
673
674 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
675 }
676
677 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
678
679 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
680 if (create) {
681 if (unlikely(f2fs_cp_error(sbi))) {
682 err = -EIO;
683 goto sync_out;
684 }
685 err = __allocate_data_block(&dn);
686 if (err)
687 goto sync_out;
688 allocated = true;
689 map->m_flags |= F2FS_MAP_NEW;
690 blkaddr = dn.data_blkaddr;
691 } else {
692 /*
693 * we only merge preallocated unwritten blocks
694 * for fiemap.
695 */
696 if (flag != F2FS_GET_BLOCK_FIEMAP ||
697 blkaddr != NEW_ADDR)
698 goto sync_out;
699 }
700 }
701
702 /* Give more consecutive addresses for the readahead */
703 if ((map->m_pblk != NEW_ADDR &&
704 blkaddr == (map->m_pblk + ofs)) ||
705 (map->m_pblk == NEW_ADDR &&
706 blkaddr == NEW_ADDR)) {
707 ofs++;
708 dn.ofs_in_node++;
709 pgofs++;
710 map->m_len++;
711 goto get_next;
712 }
713
714 sync_out:
715 if (allocated)
716 sync_inode_page(&dn);
717 put_out:
718 f2fs_put_dnode(&dn);
719 unlock_out:
720 if (create) {
721 f2fs_unlock_op(sbi);
722 if (dn.node_changed)
723 f2fs_balance_fs(sbi);
724 }
725 out:
726 trace_f2fs_map_blocks(inode, map, err);
727 return err;
728 }
729
730 static int __get_data_block(struct inode *inode, sector_t iblock,
731 struct buffer_head *bh, int create, int flag)
732 {
733 struct f2fs_map_blocks map;
734 int ret;
735
736 map.m_lblk = iblock;
737 map.m_len = bh->b_size >> inode->i_blkbits;
738
739 ret = f2fs_map_blocks(inode, &map, create, flag);
740 if (!ret) {
741 map_bh(bh, inode->i_sb, map.m_pblk);
742 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
743 bh->b_size = map.m_len << inode->i_blkbits;
744 }
745 return ret;
746 }
747
748 static int get_data_block(struct inode *inode, sector_t iblock,
749 struct buffer_head *bh_result, int create, int flag)
750 {
751 return __get_data_block(inode, iblock, bh_result, create, flag);
752 }
753
754 static int get_data_block_dio(struct inode *inode, sector_t iblock,
755 struct buffer_head *bh_result, int create)
756 {
757 return __get_data_block(inode, iblock, bh_result, create,
758 F2FS_GET_BLOCK_DIO);
759 }
760
761 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
762 struct buffer_head *bh_result, int create)
763 {
764 return __get_data_block(inode, iblock, bh_result, create,
765 F2FS_GET_BLOCK_BMAP);
766 }
767
768 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
769 {
770 return (offset >> inode->i_blkbits);
771 }
772
773 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
774 {
775 return (blk << inode->i_blkbits);
776 }
777
778 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
779 u64 start, u64 len)
780 {
781 struct buffer_head map_bh;
782 sector_t start_blk, last_blk;
783 loff_t isize = i_size_read(inode);
784 u64 logical = 0, phys = 0, size = 0;
785 u32 flags = 0;
786 int ret = 0;
787
788 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
789 if (ret)
790 return ret;
791
792 if (f2fs_has_inline_data(inode)) {
793 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
794 if (ret != -EAGAIN)
795 return ret;
796 }
797
798 mutex_lock(&inode->i_mutex);
799 if (start >= isize)
800 goto out;
801
802 if (start + len > isize)
803 len = isize - start;
804
805 if (logical_to_blk(inode, len) == 0)
806 len = blk_to_logical(inode, 1);
807
808 start_blk = logical_to_blk(inode, start);
809 last_blk = logical_to_blk(inode, start + len - 1);
810
811 next:
812 memset(&map_bh, 0, sizeof(struct buffer_head));
813 map_bh.b_size = len;
814
815 ret = get_data_block(inode, start_blk, &map_bh, 0,
816 F2FS_GET_BLOCK_FIEMAP);
817 if (ret)
818 goto out;
819
820 /* HOLE */
821 if (!buffer_mapped(&map_bh)) {
822 /* Go through holes util pass the EOF */
823 if (blk_to_logical(inode, start_blk++) < isize)
824 goto prep_next;
825 /* Found a hole beyond isize means no more extents.
826 * Note that the premise is that filesystems don't
827 * punch holes beyond isize and keep size unchanged.
828 */
829 flags |= FIEMAP_EXTENT_LAST;
830 }
831
832 if (size)
833 ret = fiemap_fill_next_extent(fieinfo, logical,
834 phys, size, flags);
835
836 if (start_blk > last_blk || ret)
837 goto out;
838
839 logical = blk_to_logical(inode, start_blk);
840 phys = blk_to_logical(inode, map_bh.b_blocknr);
841 size = map_bh.b_size;
842 flags = 0;
843 if (buffer_unwritten(&map_bh))
844 flags = FIEMAP_EXTENT_UNWRITTEN;
845
846 start_blk += logical_to_blk(inode, size);
847
848 prep_next:
849 cond_resched();
850 if (fatal_signal_pending(current))
851 ret = -EINTR;
852 else
853 goto next;
854 out:
855 if (ret == 1)
856 ret = 0;
857
858 mutex_unlock(&inode->i_mutex);
859 return ret;
860 }
861
862 /*
863 * This function was originally taken from fs/mpage.c, and customized for f2fs.
864 * Major change was from block_size == page_size in f2fs by default.
865 */
866 static int f2fs_mpage_readpages(struct address_space *mapping,
867 struct list_head *pages, struct page *page,
868 unsigned nr_pages)
869 {
870 struct bio *bio = NULL;
871 unsigned page_idx;
872 sector_t last_block_in_bio = 0;
873 struct inode *inode = mapping->host;
874 const unsigned blkbits = inode->i_blkbits;
875 const unsigned blocksize = 1 << blkbits;
876 sector_t block_in_file;
877 sector_t last_block;
878 sector_t last_block_in_file;
879 sector_t block_nr;
880 struct block_device *bdev = inode->i_sb->s_bdev;
881 struct f2fs_map_blocks map;
882
883 map.m_pblk = 0;
884 map.m_lblk = 0;
885 map.m_len = 0;
886 map.m_flags = 0;
887
888 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
889
890 prefetchw(&page->flags);
891 if (pages) {
892 page = list_entry(pages->prev, struct page, lru);
893 list_del(&page->lru);
894 if (add_to_page_cache_lru(page, mapping,
895 page->index, GFP_KERNEL))
896 goto next_page;
897 }
898
899 block_in_file = (sector_t)page->index;
900 last_block = block_in_file + nr_pages;
901 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
902 blkbits;
903 if (last_block > last_block_in_file)
904 last_block = last_block_in_file;
905
906 /*
907 * Map blocks using the previous result first.
908 */
909 if ((map.m_flags & F2FS_MAP_MAPPED) &&
910 block_in_file > map.m_lblk &&
911 block_in_file < (map.m_lblk + map.m_len))
912 goto got_it;
913
914 /*
915 * Then do more f2fs_map_blocks() calls until we are
916 * done with this page.
917 */
918 map.m_flags = 0;
919
920 if (block_in_file < last_block) {
921 map.m_lblk = block_in_file;
922 map.m_len = last_block - block_in_file;
923
924 if (f2fs_map_blocks(inode, &map, 0,
925 F2FS_GET_BLOCK_READ))
926 goto set_error_page;
927 }
928 got_it:
929 if ((map.m_flags & F2FS_MAP_MAPPED)) {
930 block_nr = map.m_pblk + block_in_file - map.m_lblk;
931 SetPageMappedToDisk(page);
932
933 if (!PageUptodate(page) && !cleancache_get_page(page)) {
934 SetPageUptodate(page);
935 goto confused;
936 }
937 } else {
938 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
939 SetPageUptodate(page);
940 unlock_page(page);
941 goto next_page;
942 }
943
944 /*
945 * This page will go to BIO. Do we need to send this
946 * BIO off first?
947 */
948 if (bio && (last_block_in_bio != block_nr - 1)) {
949 submit_and_realloc:
950 submit_bio(READ, bio);
951 bio = NULL;
952 }
953 if (bio == NULL) {
954 struct f2fs_crypto_ctx *ctx = NULL;
955
956 if (f2fs_encrypted_inode(inode) &&
957 S_ISREG(inode->i_mode)) {
958
959 ctx = f2fs_get_crypto_ctx(inode);
960 if (IS_ERR(ctx))
961 goto set_error_page;
962
963 /* wait the page to be moved by cleaning */
964 f2fs_wait_on_encrypted_page_writeback(
965 F2FS_I_SB(inode), block_nr);
966 }
967
968 bio = bio_alloc(GFP_KERNEL,
969 min_t(int, nr_pages, BIO_MAX_PAGES));
970 if (!bio) {
971 if (ctx)
972 f2fs_release_crypto_ctx(ctx);
973 goto set_error_page;
974 }
975 bio->bi_bdev = bdev;
976 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
977 bio->bi_end_io = f2fs_read_end_io;
978 bio->bi_private = ctx;
979 }
980
981 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
982 goto submit_and_realloc;
983
984 last_block_in_bio = block_nr;
985 goto next_page;
986 set_error_page:
987 SetPageError(page);
988 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
989 unlock_page(page);
990 goto next_page;
991 confused:
992 if (bio) {
993 submit_bio(READ, bio);
994 bio = NULL;
995 }
996 unlock_page(page);
997 next_page:
998 if (pages)
999 page_cache_release(page);
1000 }
1001 BUG_ON(pages && !list_empty(pages));
1002 if (bio)
1003 submit_bio(READ, bio);
1004 return 0;
1005 }
1006
1007 static int f2fs_read_data_page(struct file *file, struct page *page)
1008 {
1009 struct inode *inode = page->mapping->host;
1010 int ret = -EAGAIN;
1011
1012 trace_f2fs_readpage(page, DATA);
1013
1014 /* If the file has inline data, try to read it directly */
1015 if (f2fs_has_inline_data(inode))
1016 ret = f2fs_read_inline_data(inode, page);
1017 if (ret == -EAGAIN)
1018 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1019 return ret;
1020 }
1021
1022 static int f2fs_read_data_pages(struct file *file,
1023 struct address_space *mapping,
1024 struct list_head *pages, unsigned nr_pages)
1025 {
1026 struct inode *inode = file->f_mapping->host;
1027 struct page *page = list_entry(pages->prev, struct page, lru);
1028
1029 trace_f2fs_readpages(inode, page, nr_pages);
1030
1031 /* If the file has inline data, skip readpages */
1032 if (f2fs_has_inline_data(inode))
1033 return 0;
1034
1035 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1036 }
1037
1038 int do_write_data_page(struct f2fs_io_info *fio)
1039 {
1040 struct page *page = fio->page;
1041 struct inode *inode = page->mapping->host;
1042 struct dnode_of_data dn;
1043 int err = 0;
1044
1045 set_new_dnode(&dn, inode, NULL, NULL, 0);
1046 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1047 if (err)
1048 return err;
1049
1050 fio->blk_addr = dn.data_blkaddr;
1051
1052 /* This page is already truncated */
1053 if (fio->blk_addr == NULL_ADDR) {
1054 ClearPageUptodate(page);
1055 goto out_writepage;
1056 }
1057
1058 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1059
1060 /* wait for GCed encrypted page writeback */
1061 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1062 fio->blk_addr);
1063
1064 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1065 if (IS_ERR(fio->encrypted_page)) {
1066 err = PTR_ERR(fio->encrypted_page);
1067 goto out_writepage;
1068 }
1069 }
1070
1071 set_page_writeback(page);
1072
1073 /*
1074 * If current allocation needs SSR,
1075 * it had better in-place writes for updated data.
1076 */
1077 if (unlikely(fio->blk_addr != NEW_ADDR &&
1078 !is_cold_data(page) &&
1079 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1080 need_inplace_update(inode))) {
1081 rewrite_data_page(fio);
1082 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1083 trace_f2fs_do_write_data_page(page, IPU);
1084 } else {
1085 write_data_page(&dn, fio);
1086 set_data_blkaddr(&dn);
1087 f2fs_update_extent_cache(&dn);
1088 trace_f2fs_do_write_data_page(page, OPU);
1089 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1090 if (page->index == 0)
1091 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1092 }
1093 out_writepage:
1094 f2fs_put_dnode(&dn);
1095 return err;
1096 }
1097
1098 static int f2fs_write_data_page(struct page *page,
1099 struct writeback_control *wbc)
1100 {
1101 struct inode *inode = page->mapping->host;
1102 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1103 loff_t i_size = i_size_read(inode);
1104 const pgoff_t end_index = ((unsigned long long) i_size)
1105 >> PAGE_CACHE_SHIFT;
1106 unsigned offset = 0;
1107 bool need_balance_fs = false;
1108 int err = 0;
1109 struct f2fs_io_info fio = {
1110 .sbi = sbi,
1111 .type = DATA,
1112 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1113 .page = page,
1114 .encrypted_page = NULL,
1115 };
1116
1117 trace_f2fs_writepage(page, DATA);
1118
1119 if (page->index < end_index)
1120 goto write;
1121
1122 /*
1123 * If the offset is out-of-range of file size,
1124 * this page does not have to be written to disk.
1125 */
1126 offset = i_size & (PAGE_CACHE_SIZE - 1);
1127 if ((page->index >= end_index + 1) || !offset)
1128 goto out;
1129
1130 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1131 write:
1132 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1133 goto redirty_out;
1134 if (f2fs_is_drop_cache(inode))
1135 goto out;
1136 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1137 available_free_memory(sbi, BASE_CHECK))
1138 goto redirty_out;
1139
1140 /* Dentry blocks are controlled by checkpoint */
1141 if (S_ISDIR(inode->i_mode)) {
1142 if (unlikely(f2fs_cp_error(sbi)))
1143 goto redirty_out;
1144 err = do_write_data_page(&fio);
1145 goto done;
1146 }
1147
1148 /* we should bypass data pages to proceed the kworkder jobs */
1149 if (unlikely(f2fs_cp_error(sbi))) {
1150 SetPageError(page);
1151 goto out;
1152 }
1153
1154 if (!wbc->for_reclaim)
1155 need_balance_fs = true;
1156 else if (has_not_enough_free_secs(sbi, 0))
1157 goto redirty_out;
1158
1159 err = -EAGAIN;
1160 f2fs_lock_op(sbi);
1161 if (f2fs_has_inline_data(inode))
1162 err = f2fs_write_inline_data(inode, page);
1163 if (err == -EAGAIN)
1164 err = do_write_data_page(&fio);
1165 f2fs_unlock_op(sbi);
1166 done:
1167 if (err && err != -ENOENT)
1168 goto redirty_out;
1169
1170 clear_cold_data(page);
1171 out:
1172 inode_dec_dirty_pages(inode);
1173 if (err)
1174 ClearPageUptodate(page);
1175 unlock_page(page);
1176 if (need_balance_fs)
1177 f2fs_balance_fs(sbi);
1178 if (wbc->for_reclaim) {
1179 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1180 remove_dirty_inode(inode);
1181 }
1182 return 0;
1183
1184 redirty_out:
1185 redirty_page_for_writepage(wbc, page);
1186 return AOP_WRITEPAGE_ACTIVATE;
1187 }
1188
1189 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1190 void *data)
1191 {
1192 struct address_space *mapping = data;
1193 int ret = mapping->a_ops->writepage(page, wbc);
1194 mapping_set_error(mapping, ret);
1195 return ret;
1196 }
1197
1198 /*
1199 * This function was copied from write_cche_pages from mm/page-writeback.c.
1200 * The major change is making write step of cold data page separately from
1201 * warm/hot data page.
1202 */
1203 static int f2fs_write_cache_pages(struct address_space *mapping,
1204 struct writeback_control *wbc, writepage_t writepage,
1205 void *data)
1206 {
1207 int ret = 0;
1208 int done = 0;
1209 struct pagevec pvec;
1210 int nr_pages;
1211 pgoff_t uninitialized_var(writeback_index);
1212 pgoff_t index;
1213 pgoff_t end; /* Inclusive */
1214 pgoff_t done_index;
1215 int cycled;
1216 int range_whole = 0;
1217 int tag;
1218 int step = 0;
1219
1220 pagevec_init(&pvec, 0);
1221 next:
1222 if (wbc->range_cyclic) {
1223 writeback_index = mapping->writeback_index; /* prev offset */
1224 index = writeback_index;
1225 if (index == 0)
1226 cycled = 1;
1227 else
1228 cycled = 0;
1229 end = -1;
1230 } else {
1231 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1232 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1233 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1234 range_whole = 1;
1235 cycled = 1; /* ignore range_cyclic tests */
1236 }
1237 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1238 tag = PAGECACHE_TAG_TOWRITE;
1239 else
1240 tag = PAGECACHE_TAG_DIRTY;
1241 retry:
1242 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1243 tag_pages_for_writeback(mapping, index, end);
1244 done_index = index;
1245 while (!done && (index <= end)) {
1246 int i;
1247
1248 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1249 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1250 if (nr_pages == 0)
1251 break;
1252
1253 for (i = 0; i < nr_pages; i++) {
1254 struct page *page = pvec.pages[i];
1255
1256 if (page->index > end) {
1257 done = 1;
1258 break;
1259 }
1260
1261 done_index = page->index;
1262
1263 lock_page(page);
1264
1265 if (unlikely(page->mapping != mapping)) {
1266 continue_unlock:
1267 unlock_page(page);
1268 continue;
1269 }
1270
1271 if (!PageDirty(page)) {
1272 /* someone wrote it for us */
1273 goto continue_unlock;
1274 }
1275
1276 if (step == is_cold_data(page))
1277 goto continue_unlock;
1278
1279 if (PageWriteback(page)) {
1280 if (wbc->sync_mode != WB_SYNC_NONE)
1281 f2fs_wait_on_page_writeback(page, DATA);
1282 else
1283 goto continue_unlock;
1284 }
1285
1286 BUG_ON(PageWriteback(page));
1287 if (!clear_page_dirty_for_io(page))
1288 goto continue_unlock;
1289
1290 ret = (*writepage)(page, wbc, data);
1291 if (unlikely(ret)) {
1292 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1293 unlock_page(page);
1294 ret = 0;
1295 } else {
1296 done_index = page->index + 1;
1297 done = 1;
1298 break;
1299 }
1300 }
1301
1302 if (--wbc->nr_to_write <= 0 &&
1303 wbc->sync_mode == WB_SYNC_NONE) {
1304 done = 1;
1305 break;
1306 }
1307 }
1308 pagevec_release(&pvec);
1309 cond_resched();
1310 }
1311
1312 if (step < 1) {
1313 step++;
1314 goto next;
1315 }
1316
1317 if (!cycled && !done) {
1318 cycled = 1;
1319 index = 0;
1320 end = writeback_index - 1;
1321 goto retry;
1322 }
1323 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1324 mapping->writeback_index = done_index;
1325
1326 return ret;
1327 }
1328
1329 static int f2fs_write_data_pages(struct address_space *mapping,
1330 struct writeback_control *wbc)
1331 {
1332 struct inode *inode = mapping->host;
1333 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1334 bool locked = false;
1335 int ret;
1336 long diff;
1337
1338 trace_f2fs_writepages(mapping->host, wbc, DATA);
1339
1340 /* deal with chardevs and other special file */
1341 if (!mapping->a_ops->writepage)
1342 return 0;
1343
1344 /* skip writing if there is no dirty page in this inode */
1345 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1346 return 0;
1347
1348 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1349 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1350 available_free_memory(sbi, DIRTY_DENTS))
1351 goto skip_write;
1352
1353 /* skip writing during file defragment */
1354 if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1355 goto skip_write;
1356
1357 /* during POR, we don't need to trigger writepage at all. */
1358 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1359 goto skip_write;
1360
1361 diff = nr_pages_to_write(sbi, DATA, wbc);
1362
1363 if (!S_ISDIR(inode->i_mode)) {
1364 mutex_lock(&sbi->writepages);
1365 locked = true;
1366 }
1367 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1368 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1369 if (locked)
1370 mutex_unlock(&sbi->writepages);
1371
1372 remove_dirty_inode(inode);
1373
1374 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1375 return ret;
1376
1377 skip_write:
1378 wbc->pages_skipped += get_dirty_pages(inode);
1379 return 0;
1380 }
1381
1382 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1383 {
1384 struct inode *inode = mapping->host;
1385
1386 if (to > inode->i_size) {
1387 truncate_pagecache(inode, inode->i_size);
1388 truncate_blocks(inode, inode->i_size, true);
1389 }
1390 }
1391
1392 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1393 struct page *page, loff_t pos, unsigned len,
1394 block_t *blk_addr, bool *node_changed)
1395 {
1396 struct inode *inode = page->mapping->host;
1397 pgoff_t index = page->index;
1398 struct dnode_of_data dn;
1399 struct page *ipage;
1400 bool locked = false;
1401 struct extent_info ei;
1402 int err = 0;
1403
1404 if (f2fs_has_inline_data(inode) ||
1405 (pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1406 f2fs_lock_op(sbi);
1407 locked = true;
1408 }
1409 restart:
1410 /* check inline_data */
1411 ipage = get_node_page(sbi, inode->i_ino);
1412 if (IS_ERR(ipage)) {
1413 err = PTR_ERR(ipage);
1414 goto unlock_out;
1415 }
1416
1417 set_new_dnode(&dn, inode, ipage, ipage, 0);
1418
1419 if (f2fs_has_inline_data(inode)) {
1420 if (pos + len <= MAX_INLINE_DATA) {
1421 read_inline_data(page, ipage);
1422 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1423 sync_inode_page(&dn);
1424 } else {
1425 err = f2fs_convert_inline_page(&dn, page);
1426 if (err)
1427 goto out;
1428 if (dn.data_blkaddr == NULL_ADDR)
1429 err = f2fs_get_block(&dn, index);
1430 }
1431 } else if (locked) {
1432 err = f2fs_get_block(&dn, index);
1433 } else {
1434 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1435 dn.data_blkaddr = ei.blk + index - ei.fofs;
1436 } else {
1437 bool restart = false;
1438
1439 /* hole case */
1440 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1441 if (err || (!err && dn.data_blkaddr == NULL_ADDR))
1442 restart = true;
1443 if (restart) {
1444 f2fs_put_dnode(&dn);
1445 f2fs_lock_op(sbi);
1446 locked = true;
1447 goto restart;
1448 }
1449 }
1450 }
1451
1452 /* convert_inline_page can make node_changed */
1453 *blk_addr = dn.data_blkaddr;
1454 *node_changed = dn.node_changed;
1455 out:
1456 f2fs_put_dnode(&dn);
1457 unlock_out:
1458 if (locked)
1459 f2fs_unlock_op(sbi);
1460 return err;
1461 }
1462
1463 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1464 loff_t pos, unsigned len, unsigned flags,
1465 struct page **pagep, void **fsdata)
1466 {
1467 struct inode *inode = mapping->host;
1468 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1469 struct page *page = NULL;
1470 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1471 bool need_balance = false;
1472 block_t blkaddr = NULL_ADDR;
1473 int err = 0;
1474
1475 trace_f2fs_write_begin(inode, pos, len, flags);
1476
1477 /*
1478 * We should check this at this moment to avoid deadlock on inode page
1479 * and #0 page. The locking rule for inline_data conversion should be:
1480 * lock_page(page #0) -> lock_page(inode_page)
1481 */
1482 if (index != 0) {
1483 err = f2fs_convert_inline_inode(inode);
1484 if (err)
1485 goto fail;
1486 }
1487 repeat:
1488 page = grab_cache_page_write_begin(mapping, index, flags);
1489 if (!page) {
1490 err = -ENOMEM;
1491 goto fail;
1492 }
1493
1494 *pagep = page;
1495
1496 err = prepare_write_begin(sbi, page, pos, len,
1497 &blkaddr, &need_balance);
1498 if (err)
1499 goto fail;
1500
1501 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1502 unlock_page(page);
1503 f2fs_balance_fs(sbi);
1504 lock_page(page);
1505 if (page->mapping != mapping) {
1506 /* The page got truncated from under us */
1507 f2fs_put_page(page, 1);
1508 goto repeat;
1509 }
1510 }
1511
1512 f2fs_wait_on_page_writeback(page, DATA);
1513
1514 /* wait for GCed encrypted page writeback */
1515 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1516 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1517
1518 if (len == PAGE_CACHE_SIZE)
1519 goto out_update;
1520 if (PageUptodate(page))
1521 goto out_clear;
1522
1523 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1524 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1525 unsigned end = start + len;
1526
1527 /* Reading beyond i_size is simple: memset to zero */
1528 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1529 goto out_update;
1530 }
1531
1532 if (blkaddr == NEW_ADDR) {
1533 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1534 } else {
1535 struct f2fs_io_info fio = {
1536 .sbi = sbi,
1537 .type = DATA,
1538 .rw = READ_SYNC,
1539 .blk_addr = blkaddr,
1540 .page = page,
1541 .encrypted_page = NULL,
1542 };
1543 err = f2fs_submit_page_bio(&fio);
1544 if (err)
1545 goto fail;
1546
1547 lock_page(page);
1548 if (unlikely(!PageUptodate(page))) {
1549 err = -EIO;
1550 goto fail;
1551 }
1552 if (unlikely(page->mapping != mapping)) {
1553 f2fs_put_page(page, 1);
1554 goto repeat;
1555 }
1556
1557 /* avoid symlink page */
1558 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1559 err = f2fs_decrypt_one(inode, page);
1560 if (err)
1561 goto fail;
1562 }
1563 }
1564 out_update:
1565 SetPageUptodate(page);
1566 out_clear:
1567 clear_cold_data(page);
1568 return 0;
1569
1570 fail:
1571 f2fs_put_page(page, 1);
1572 f2fs_write_failed(mapping, pos + len);
1573 return err;
1574 }
1575
1576 static int f2fs_write_end(struct file *file,
1577 struct address_space *mapping,
1578 loff_t pos, unsigned len, unsigned copied,
1579 struct page *page, void *fsdata)
1580 {
1581 struct inode *inode = page->mapping->host;
1582
1583 trace_f2fs_write_end(inode, pos, len, copied);
1584
1585 set_page_dirty(page);
1586
1587 if (pos + copied > i_size_read(inode)) {
1588 i_size_write(inode, pos + copied);
1589 mark_inode_dirty(inode);
1590 update_inode_page(inode);
1591 }
1592
1593 f2fs_put_page(page, 1);
1594 return copied;
1595 }
1596
1597 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1598 loff_t offset)
1599 {
1600 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1601
1602 if (offset & blocksize_mask)
1603 return -EINVAL;
1604
1605 if (iov_iter_alignment(iter) & blocksize_mask)
1606 return -EINVAL;
1607
1608 return 0;
1609 }
1610
1611 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1612 loff_t offset)
1613 {
1614 struct file *file = iocb->ki_filp;
1615 struct address_space *mapping = file->f_mapping;
1616 struct inode *inode = mapping->host;
1617 size_t count = iov_iter_count(iter);
1618 int err;
1619
1620 /* we don't need to use inline_data strictly */
1621 err = f2fs_convert_inline_inode(inode);
1622 if (err)
1623 return err;
1624
1625 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1626 return 0;
1627
1628 err = check_direct_IO(inode, iter, offset);
1629 if (err)
1630 return err;
1631
1632 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1633
1634 if (iov_iter_rw(iter) == WRITE) {
1635 err = __allocate_data_blocks(inode, offset, count);
1636 if (err)
1637 goto out;
1638 }
1639
1640 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1641 out:
1642 if (err < 0 && iov_iter_rw(iter) == WRITE)
1643 f2fs_write_failed(mapping, offset + count);
1644
1645 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1646
1647 return err;
1648 }
1649
1650 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1651 unsigned int length)
1652 {
1653 struct inode *inode = page->mapping->host;
1654 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1655
1656 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1657 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1658 return;
1659
1660 if (PageDirty(page)) {
1661 if (inode->i_ino == F2FS_META_INO(sbi))
1662 dec_page_count(sbi, F2FS_DIRTY_META);
1663 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1664 dec_page_count(sbi, F2FS_DIRTY_NODES);
1665 else
1666 inode_dec_dirty_pages(inode);
1667 }
1668
1669 /* This is atomic written page, keep Private */
1670 if (IS_ATOMIC_WRITTEN_PAGE(page))
1671 return;
1672
1673 ClearPagePrivate(page);
1674 }
1675
1676 int f2fs_release_page(struct page *page, gfp_t wait)
1677 {
1678 /* If this is dirty page, keep PagePrivate */
1679 if (PageDirty(page))
1680 return 0;
1681
1682 /* This is atomic written page, keep Private */
1683 if (IS_ATOMIC_WRITTEN_PAGE(page))
1684 return 0;
1685
1686 ClearPagePrivate(page);
1687 return 1;
1688 }
1689
1690 static int f2fs_set_data_page_dirty(struct page *page)
1691 {
1692 struct address_space *mapping = page->mapping;
1693 struct inode *inode = mapping->host;
1694
1695 trace_f2fs_set_page_dirty(page, DATA);
1696
1697 SetPageUptodate(page);
1698
1699 if (f2fs_is_atomic_file(inode)) {
1700 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1701 register_inmem_page(inode, page);
1702 return 1;
1703 }
1704 /*
1705 * Previously, this page has been registered, we just
1706 * return here.
1707 */
1708 return 0;
1709 }
1710
1711 if (!PageDirty(page)) {
1712 __set_page_dirty_nobuffers(page);
1713 update_dirty_page(inode, page);
1714 return 1;
1715 }
1716 return 0;
1717 }
1718
1719 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1720 {
1721 struct inode *inode = mapping->host;
1722
1723 if (f2fs_has_inline_data(inode))
1724 return 0;
1725
1726 /* make sure allocating whole blocks */
1727 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1728 filemap_write_and_wait(mapping);
1729
1730 return generic_block_bmap(mapping, block, get_data_block_bmap);
1731 }
1732
1733 const struct address_space_operations f2fs_dblock_aops = {
1734 .readpage = f2fs_read_data_page,
1735 .readpages = f2fs_read_data_pages,
1736 .writepage = f2fs_write_data_page,
1737 .writepages = f2fs_write_data_pages,
1738 .write_begin = f2fs_write_begin,
1739 .write_end = f2fs_write_end,
1740 .set_page_dirty = f2fs_set_data_page_dirty,
1741 .invalidatepage = f2fs_invalidate_page,
1742 .releasepage = f2fs_release_page,
1743 .direct_IO = f2fs_direct_IO,
1744 .bmap = f2fs_bmap,
1745 };
ubuntu-bionic-kernel mirror