]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/f2fs/data.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: call f2fs_balance_fs only when node was changed
[mirror_ubuntu-artful-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 void __allocate_data_blocks(struct inode *inode, loff_t offset,
502 size_t count)
503 {
504 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
505 struct dnode_of_data dn;
506 u64 start = F2FS_BYTES_TO_BLK(offset);
507 u64 len = F2FS_BYTES_TO_BLK(count);
508 bool allocated;
509 u64 end_offset;
510
511 while (len) {
512 f2fs_lock_op(sbi);
513
514 /* When reading holes, we need its node page */
515 set_new_dnode(&dn, inode, NULL, NULL, 0);
516 if (get_dnode_of_data(&dn, start, ALLOC_NODE))
517 goto out;
518
519 allocated = false;
520 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
521
522 while (dn.ofs_in_node < end_offset && len) {
523 block_t blkaddr;
524
525 if (unlikely(f2fs_cp_error(sbi)))
526 goto sync_out;
527
528 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
529 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
530 if (__allocate_data_block(&dn))
531 goto sync_out;
532 allocated = true;
533 }
534 len--;
535 start++;
536 dn.ofs_in_node++;
537 }
538
539 if (allocated)
540 sync_inode_page(&dn);
541
542 f2fs_put_dnode(&dn);
543 f2fs_unlock_op(sbi);
544
545 if (dn.node_changed)
546 f2fs_balance_fs(sbi);
547 }
548 return;
549
550 sync_out:
551 if (allocated)
552 sync_inode_page(&dn);
553 f2fs_put_dnode(&dn);
554 out:
555 f2fs_unlock_op(sbi);
556 if (dn.node_changed)
557 f2fs_balance_fs(sbi);
558 return;
559 }
560
561 /*
562 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
563 * f2fs_map_blocks structure.
564 * If original data blocks are allocated, then give them to blockdev.
565 * Otherwise,
566 * a. preallocate requested block addresses
567 * b. do not use extent cache for better performance
568 * c. give the block addresses to blockdev
569 */
570 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
571 int create, int flag)
572 {
573 unsigned int maxblocks = map->m_len;
574 struct dnode_of_data dn;
575 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
576 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
577 pgoff_t pgofs, end_offset;
578 int err = 0, ofs = 1;
579 struct extent_info ei;
580 bool allocated = false;
581 block_t blkaddr;
582
583 map->m_len = 0;
584 map->m_flags = 0;
585
586 /* it only supports block size == page size */
587 pgofs = (pgoff_t)map->m_lblk;
588
589 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
590 map->m_pblk = ei.blk + pgofs - ei.fofs;
591 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
592 map->m_flags = F2FS_MAP_MAPPED;
593 goto out;
594 }
595
596 if (create)
597 f2fs_lock_op(sbi);
598
599 /* When reading holes, we need its node page */
600 set_new_dnode(&dn, inode, NULL, NULL, 0);
601 err = get_dnode_of_data(&dn, pgofs, mode);
602 if (err) {
603 if (err == -ENOENT)
604 err = 0;
605 goto unlock_out;
606 }
607
608 if (dn.data_blkaddr == NEW_ADDR || dn.data_blkaddr == NULL_ADDR) {
609 if (create) {
610 if (unlikely(f2fs_cp_error(sbi))) {
611 err = -EIO;
612 goto put_out;
613 }
614 err = __allocate_data_block(&dn);
615 if (err)
616 goto put_out;
617 allocated = true;
618 map->m_flags = F2FS_MAP_NEW;
619 } else {
620 if (flag != F2FS_GET_BLOCK_FIEMAP ||
621 dn.data_blkaddr != NEW_ADDR) {
622 if (flag == F2FS_GET_BLOCK_BMAP)
623 err = -ENOENT;
624 goto put_out;
625 }
626
627 /*
628 * preallocated unwritten block should be mapped
629 * for fiemap.
630 */
631 if (dn.data_blkaddr == NEW_ADDR)
632 map->m_flags = F2FS_MAP_UNWRITTEN;
633 }
634 }
635
636 map->m_flags |= F2FS_MAP_MAPPED;
637 map->m_pblk = dn.data_blkaddr;
638 map->m_len = 1;
639
640 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
641 dn.ofs_in_node++;
642 pgofs++;
643
644 get_next:
645 if (map->m_len >= maxblocks)
646 goto sync_out;
647
648 if (dn.ofs_in_node >= end_offset) {
649 if (allocated)
650 sync_inode_page(&dn);
651 allocated = false;
652 f2fs_put_dnode(&dn);
653
654 if (create) {
655 f2fs_unlock_op(sbi);
656 if (dn.node_changed)
657 f2fs_balance_fs(sbi);
658 f2fs_lock_op(sbi);
659 }
660
661 set_new_dnode(&dn, inode, NULL, NULL, 0);
662 err = get_dnode_of_data(&dn, pgofs, mode);
663 if (err) {
664 if (err == -ENOENT)
665 err = 0;
666 goto unlock_out;
667 }
668
669 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
670 }
671
672 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
673
674 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
675 if (create) {
676 if (unlikely(f2fs_cp_error(sbi))) {
677 err = -EIO;
678 goto sync_out;
679 }
680 err = __allocate_data_block(&dn);
681 if (err)
682 goto sync_out;
683 allocated = true;
684 map->m_flags |= F2FS_MAP_NEW;
685 blkaddr = dn.data_blkaddr;
686 } else {
687 /*
688 * we only merge preallocated unwritten blocks
689 * for fiemap.
690 */
691 if (flag != F2FS_GET_BLOCK_FIEMAP ||
692 blkaddr != NEW_ADDR)
693 goto sync_out;
694 }
695 }
696
697 /* Give more consecutive addresses for the readahead */
698 if ((map->m_pblk != NEW_ADDR &&
699 blkaddr == (map->m_pblk + ofs)) ||
700 (map->m_pblk == NEW_ADDR &&
701 blkaddr == NEW_ADDR)) {
702 ofs++;
703 dn.ofs_in_node++;
704 pgofs++;
705 map->m_len++;
706 goto get_next;
707 }
708
709 sync_out:
710 if (allocated)
711 sync_inode_page(&dn);
712 put_out:
713 f2fs_put_dnode(&dn);
714 unlock_out:
715 if (create) {
716 f2fs_unlock_op(sbi);
717 if (dn.node_changed)
718 f2fs_balance_fs(sbi);
719 }
720 out:
721 trace_f2fs_map_blocks(inode, map, err);
722 return err;
723 }
724
725 static int __get_data_block(struct inode *inode, sector_t iblock,
726 struct buffer_head *bh, int create, int flag)
727 {
728 struct f2fs_map_blocks map;
729 int ret;
730
731 map.m_lblk = iblock;
732 map.m_len = bh->b_size >> inode->i_blkbits;
733
734 ret = f2fs_map_blocks(inode, &map, create, flag);
735 if (!ret) {
736 map_bh(bh, inode->i_sb, map.m_pblk);
737 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
738 bh->b_size = map.m_len << inode->i_blkbits;
739 }
740 return ret;
741 }
742
743 static int get_data_block(struct inode *inode, sector_t iblock,
744 struct buffer_head *bh_result, int create, int flag)
745 {
746 return __get_data_block(inode, iblock, bh_result, create, flag);
747 }
748
749 static int get_data_block_dio(struct inode *inode, sector_t iblock,
750 struct buffer_head *bh_result, int create)
751 {
752 return __get_data_block(inode, iblock, bh_result, create,
753 F2FS_GET_BLOCK_DIO);
754 }
755
756 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
757 struct buffer_head *bh_result, int create)
758 {
759 return __get_data_block(inode, iblock, bh_result, create,
760 F2FS_GET_BLOCK_BMAP);
761 }
762
763 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
764 {
765 return (offset >> inode->i_blkbits);
766 }
767
768 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
769 {
770 return (blk << inode->i_blkbits);
771 }
772
773 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
774 u64 start, u64 len)
775 {
776 struct buffer_head map_bh;
777 sector_t start_blk, last_blk;
778 loff_t isize = i_size_read(inode);
779 u64 logical = 0, phys = 0, size = 0;
780 u32 flags = 0;
781 bool past_eof = false, whole_file = false;
782 int ret = 0;
783
784 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
785 if (ret)
786 return ret;
787
788 if (f2fs_has_inline_data(inode)) {
789 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
790 if (ret != -EAGAIN)
791 return ret;
792 }
793
794 mutex_lock(&inode->i_mutex);
795
796 if (len >= isize) {
797 whole_file = true;
798 len = isize;
799 }
800
801 if (logical_to_blk(inode, len) == 0)
802 len = blk_to_logical(inode, 1);
803
804 start_blk = logical_to_blk(inode, start);
805 last_blk = logical_to_blk(inode, start + len - 1);
806 next:
807 memset(&map_bh, 0, sizeof(struct buffer_head));
808 map_bh.b_size = len;
809
810 ret = get_data_block(inode, start_blk, &map_bh, 0,
811 F2FS_GET_BLOCK_FIEMAP);
812 if (ret)
813 goto out;
814
815 /* HOLE */
816 if (!buffer_mapped(&map_bh)) {
817 start_blk++;
818
819 if (!past_eof && blk_to_logical(inode, start_blk) >= isize)
820 past_eof = 1;
821
822 if (past_eof && size) {
823 flags |= FIEMAP_EXTENT_LAST;
824 ret = fiemap_fill_next_extent(fieinfo, logical,
825 phys, size, flags);
826 } else if (size) {
827 ret = fiemap_fill_next_extent(fieinfo, logical,
828 phys, size, flags);
829 size = 0;
830 }
831
832 /* if we have holes up to/past EOF then we're done */
833 if (start_blk > last_blk || past_eof || ret)
834 goto out;
835 } else {
836 if (start_blk > last_blk && !whole_file) {
837 ret = fiemap_fill_next_extent(fieinfo, logical,
838 phys, size, flags);
839 goto out;
840 }
841
842 /*
843 * if size != 0 then we know we already have an extent
844 * to add, so add it.
845 */
846 if (size) {
847 ret = fiemap_fill_next_extent(fieinfo, logical,
848 phys, size, flags);
849 if (ret)
850 goto out;
851 }
852
853 logical = blk_to_logical(inode, start_blk);
854 phys = blk_to_logical(inode, map_bh.b_blocknr);
855 size = map_bh.b_size;
856 flags = 0;
857 if (buffer_unwritten(&map_bh))
858 flags = FIEMAP_EXTENT_UNWRITTEN;
859
860 start_blk += logical_to_blk(inode, size);
861
862 /*
863 * If we are past the EOF, then we need to make sure as
864 * soon as we find a hole that the last extent we found
865 * is marked with FIEMAP_EXTENT_LAST
866 */
867 if (!past_eof && logical + size >= isize)
868 past_eof = true;
869 }
870 cond_resched();
871 if (fatal_signal_pending(current))
872 ret = -EINTR;
873 else
874 goto next;
875 out:
876 if (ret == 1)
877 ret = 0;
878
879 mutex_unlock(&inode->i_mutex);
880 return ret;
881 }
882
883 /*
884 * This function was originally taken from fs/mpage.c, and customized for f2fs.
885 * Major change was from block_size == page_size in f2fs by default.
886 */
887 static int f2fs_mpage_readpages(struct address_space *mapping,
888 struct list_head *pages, struct page *page,
889 unsigned nr_pages)
890 {
891 struct bio *bio = NULL;
892 unsigned page_idx;
893 sector_t last_block_in_bio = 0;
894 struct inode *inode = mapping->host;
895 const unsigned blkbits = inode->i_blkbits;
896 const unsigned blocksize = 1 << blkbits;
897 sector_t block_in_file;
898 sector_t last_block;
899 sector_t last_block_in_file;
900 sector_t block_nr;
901 struct block_device *bdev = inode->i_sb->s_bdev;
902 struct f2fs_map_blocks map;
903
904 map.m_pblk = 0;
905 map.m_lblk = 0;
906 map.m_len = 0;
907 map.m_flags = 0;
908
909 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
910
911 prefetchw(&page->flags);
912 if (pages) {
913 page = list_entry(pages->prev, struct page, lru);
914 list_del(&page->lru);
915 if (add_to_page_cache_lru(page, mapping,
916 page->index, GFP_KERNEL))
917 goto next_page;
918 }
919
920 block_in_file = (sector_t)page->index;
921 last_block = block_in_file + nr_pages;
922 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
923 blkbits;
924 if (last_block > last_block_in_file)
925 last_block = last_block_in_file;
926
927 /*
928 * Map blocks using the previous result first.
929 */
930 if ((map.m_flags & F2FS_MAP_MAPPED) &&
931 block_in_file > map.m_lblk &&
932 block_in_file < (map.m_lblk + map.m_len))
933 goto got_it;
934
935 /*
936 * Then do more f2fs_map_blocks() calls until we are
937 * done with this page.
938 */
939 map.m_flags = 0;
940
941 if (block_in_file < last_block) {
942 map.m_lblk = block_in_file;
943 map.m_len = last_block - block_in_file;
944
945 if (f2fs_map_blocks(inode, &map, 0,
946 F2FS_GET_BLOCK_READ))
947 goto set_error_page;
948 }
949 got_it:
950 if ((map.m_flags & F2FS_MAP_MAPPED)) {
951 block_nr = map.m_pblk + block_in_file - map.m_lblk;
952 SetPageMappedToDisk(page);
953
954 if (!PageUptodate(page) && !cleancache_get_page(page)) {
955 SetPageUptodate(page);
956 goto confused;
957 }
958 } else {
959 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
960 SetPageUptodate(page);
961 unlock_page(page);
962 goto next_page;
963 }
964
965 /*
966 * This page will go to BIO. Do we need to send this
967 * BIO off first?
968 */
969 if (bio && (last_block_in_bio != block_nr - 1)) {
970 submit_and_realloc:
971 submit_bio(READ, bio);
972 bio = NULL;
973 }
974 if (bio == NULL) {
975 struct f2fs_crypto_ctx *ctx = NULL;
976
977 if (f2fs_encrypted_inode(inode) &&
978 S_ISREG(inode->i_mode)) {
979
980 ctx = f2fs_get_crypto_ctx(inode);
981 if (IS_ERR(ctx))
982 goto set_error_page;
983
984 /* wait the page to be moved by cleaning */
985 f2fs_wait_on_encrypted_page_writeback(
986 F2FS_I_SB(inode), block_nr);
987 }
988
989 bio = bio_alloc(GFP_KERNEL,
990 min_t(int, nr_pages, BIO_MAX_PAGES));
991 if (!bio) {
992 if (ctx)
993 f2fs_release_crypto_ctx(ctx);
994 goto set_error_page;
995 }
996 bio->bi_bdev = bdev;
997 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
998 bio->bi_end_io = f2fs_read_end_io;
999 bio->bi_private = ctx;
1000 }
1001
1002 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1003 goto submit_and_realloc;
1004
1005 last_block_in_bio = block_nr;
1006 goto next_page;
1007 set_error_page:
1008 SetPageError(page);
1009 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1010 unlock_page(page);
1011 goto next_page;
1012 confused:
1013 if (bio) {
1014 submit_bio(READ, bio);
1015 bio = NULL;
1016 }
1017 unlock_page(page);
1018 next_page:
1019 if (pages)
1020 page_cache_release(page);
1021 }
1022 BUG_ON(pages && !list_empty(pages));
1023 if (bio)
1024 submit_bio(READ, bio);
1025 return 0;
1026 }
1027
1028 static int f2fs_read_data_page(struct file *file, struct page *page)
1029 {
1030 struct inode *inode = page->mapping->host;
1031 int ret = -EAGAIN;
1032
1033 trace_f2fs_readpage(page, DATA);
1034
1035 /* If the file has inline data, try to read it directly */
1036 if (f2fs_has_inline_data(inode))
1037 ret = f2fs_read_inline_data(inode, page);
1038 if (ret == -EAGAIN)
1039 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1040 return ret;
1041 }
1042
1043 static int f2fs_read_data_pages(struct file *file,
1044 struct address_space *mapping,
1045 struct list_head *pages, unsigned nr_pages)
1046 {
1047 struct inode *inode = file->f_mapping->host;
1048 struct page *page = list_entry(pages->prev, struct page, lru);
1049
1050 trace_f2fs_readpages(inode, page, nr_pages);
1051
1052 /* If the file has inline data, skip readpages */
1053 if (f2fs_has_inline_data(inode))
1054 return 0;
1055
1056 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1057 }
1058
1059 int do_write_data_page(struct f2fs_io_info *fio)
1060 {
1061 struct page *page = fio->page;
1062 struct inode *inode = page->mapping->host;
1063 struct dnode_of_data dn;
1064 int err = 0;
1065
1066 set_new_dnode(&dn, inode, NULL, NULL, 0);
1067 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1068 if (err)
1069 return err;
1070
1071 fio->blk_addr = dn.data_blkaddr;
1072
1073 /* This page is already truncated */
1074 if (fio->blk_addr == NULL_ADDR) {
1075 ClearPageUptodate(page);
1076 goto out_writepage;
1077 }
1078
1079 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1080
1081 /* wait for GCed encrypted page writeback */
1082 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1083 fio->blk_addr);
1084
1085 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1086 if (IS_ERR(fio->encrypted_page)) {
1087 err = PTR_ERR(fio->encrypted_page);
1088 goto out_writepage;
1089 }
1090 }
1091
1092 set_page_writeback(page);
1093
1094 /*
1095 * If current allocation needs SSR,
1096 * it had better in-place writes for updated data.
1097 */
1098 if (unlikely(fio->blk_addr != NEW_ADDR &&
1099 !is_cold_data(page) &&
1100 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1101 need_inplace_update(inode))) {
1102 rewrite_data_page(fio);
1103 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1104 trace_f2fs_do_write_data_page(page, IPU);
1105 } else {
1106 write_data_page(&dn, fio);
1107 set_data_blkaddr(&dn);
1108 f2fs_update_extent_cache(&dn);
1109 trace_f2fs_do_write_data_page(page, OPU);
1110 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1111 if (page->index == 0)
1112 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1113 }
1114 out_writepage:
1115 f2fs_put_dnode(&dn);
1116 return err;
1117 }
1118
1119 static int f2fs_write_data_page(struct page *page,
1120 struct writeback_control *wbc)
1121 {
1122 struct inode *inode = page->mapping->host;
1123 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1124 loff_t i_size = i_size_read(inode);
1125 const pgoff_t end_index = ((unsigned long long) i_size)
1126 >> PAGE_CACHE_SHIFT;
1127 unsigned offset = 0;
1128 bool need_balance_fs = false;
1129 int err = 0;
1130 struct f2fs_io_info fio = {
1131 .sbi = sbi,
1132 .type = DATA,
1133 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1134 .page = page,
1135 .encrypted_page = NULL,
1136 };
1137
1138 trace_f2fs_writepage(page, DATA);
1139
1140 if (page->index < end_index)
1141 goto write;
1142
1143 /*
1144 * If the offset is out-of-range of file size,
1145 * this page does not have to be written to disk.
1146 */
1147 offset = i_size & (PAGE_CACHE_SIZE - 1);
1148 if ((page->index >= end_index + 1) || !offset)
1149 goto out;
1150
1151 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1152 write:
1153 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1154 goto redirty_out;
1155 if (f2fs_is_drop_cache(inode))
1156 goto out;
1157 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1158 available_free_memory(sbi, BASE_CHECK))
1159 goto redirty_out;
1160
1161 /* Dentry blocks are controlled by checkpoint */
1162 if (S_ISDIR(inode->i_mode)) {
1163 if (unlikely(f2fs_cp_error(sbi)))
1164 goto redirty_out;
1165 err = do_write_data_page(&fio);
1166 goto done;
1167 }
1168
1169 /* we should bypass data pages to proceed the kworkder jobs */
1170 if (unlikely(f2fs_cp_error(sbi))) {
1171 SetPageError(page);
1172 goto out;
1173 }
1174
1175 if (!wbc->for_reclaim)
1176 need_balance_fs = true;
1177 else if (has_not_enough_free_secs(sbi, 0))
1178 goto redirty_out;
1179
1180 err = -EAGAIN;
1181 f2fs_lock_op(sbi);
1182 if (f2fs_has_inline_data(inode))
1183 err = f2fs_write_inline_data(inode, page);
1184 if (err == -EAGAIN)
1185 err = do_write_data_page(&fio);
1186 f2fs_unlock_op(sbi);
1187 done:
1188 if (err && err != -ENOENT)
1189 goto redirty_out;
1190
1191 clear_cold_data(page);
1192 out:
1193 inode_dec_dirty_pages(inode);
1194 if (err)
1195 ClearPageUptodate(page);
1196 unlock_page(page);
1197 if (need_balance_fs)
1198 f2fs_balance_fs(sbi);
1199 if (wbc->for_reclaim) {
1200 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1201 remove_dirty_inode(inode);
1202 }
1203 return 0;
1204
1205 redirty_out:
1206 redirty_page_for_writepage(wbc, page);
1207 return AOP_WRITEPAGE_ACTIVATE;
1208 }
1209
1210 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1211 void *data)
1212 {
1213 struct address_space *mapping = data;
1214 int ret = mapping->a_ops->writepage(page, wbc);
1215 mapping_set_error(mapping, ret);
1216 return ret;
1217 }
1218
1219 /*
1220 * This function was copied from write_cche_pages from mm/page-writeback.c.
1221 * The major change is making write step of cold data page separately from
1222 * warm/hot data page.
1223 */
1224 static int f2fs_write_cache_pages(struct address_space *mapping,
1225 struct writeback_control *wbc, writepage_t writepage,
1226 void *data)
1227 {
1228 int ret = 0;
1229 int done = 0;
1230 struct pagevec pvec;
1231 int nr_pages;
1232 pgoff_t uninitialized_var(writeback_index);
1233 pgoff_t index;
1234 pgoff_t end; /* Inclusive */
1235 pgoff_t done_index;
1236 int cycled;
1237 int range_whole = 0;
1238 int tag;
1239 int step = 0;
1240
1241 pagevec_init(&pvec, 0);
1242 next:
1243 if (wbc->range_cyclic) {
1244 writeback_index = mapping->writeback_index; /* prev offset */
1245 index = writeback_index;
1246 if (index == 0)
1247 cycled = 1;
1248 else
1249 cycled = 0;
1250 end = -1;
1251 } else {
1252 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1253 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1254 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1255 range_whole = 1;
1256 cycled = 1; /* ignore range_cyclic tests */
1257 }
1258 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1259 tag = PAGECACHE_TAG_TOWRITE;
1260 else
1261 tag = PAGECACHE_TAG_DIRTY;
1262 retry:
1263 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1264 tag_pages_for_writeback(mapping, index, end);
1265 done_index = index;
1266 while (!done && (index <= end)) {
1267 int i;
1268
1269 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1270 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1271 if (nr_pages == 0)
1272 break;
1273
1274 for (i = 0; i < nr_pages; i++) {
1275 struct page *page = pvec.pages[i];
1276
1277 if (page->index > end) {
1278 done = 1;
1279 break;
1280 }
1281
1282 done_index = page->index;
1283
1284 lock_page(page);
1285
1286 if (unlikely(page->mapping != mapping)) {
1287 continue_unlock:
1288 unlock_page(page);
1289 continue;
1290 }
1291
1292 if (!PageDirty(page)) {
1293 /* someone wrote it for us */
1294 goto continue_unlock;
1295 }
1296
1297 if (step == is_cold_data(page))
1298 goto continue_unlock;
1299
1300 if (PageWriteback(page)) {
1301 if (wbc->sync_mode != WB_SYNC_NONE)
1302 f2fs_wait_on_page_writeback(page, DATA);
1303 else
1304 goto continue_unlock;
1305 }
1306
1307 BUG_ON(PageWriteback(page));
1308 if (!clear_page_dirty_for_io(page))
1309 goto continue_unlock;
1310
1311 ret = (*writepage)(page, wbc, data);
1312 if (unlikely(ret)) {
1313 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1314 unlock_page(page);
1315 ret = 0;
1316 } else {
1317 done_index = page->index + 1;
1318 done = 1;
1319 break;
1320 }
1321 }
1322
1323 if (--wbc->nr_to_write <= 0 &&
1324 wbc->sync_mode == WB_SYNC_NONE) {
1325 done = 1;
1326 break;
1327 }
1328 }
1329 pagevec_release(&pvec);
1330 cond_resched();
1331 }
1332
1333 if (step < 1) {
1334 step++;
1335 goto next;
1336 }
1337
1338 if (!cycled && !done) {
1339 cycled = 1;
1340 index = 0;
1341 end = writeback_index - 1;
1342 goto retry;
1343 }
1344 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1345 mapping->writeback_index = done_index;
1346
1347 return ret;
1348 }
1349
1350 static int f2fs_write_data_pages(struct address_space *mapping,
1351 struct writeback_control *wbc)
1352 {
1353 struct inode *inode = mapping->host;
1354 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1355 bool locked = false;
1356 int ret;
1357 long diff;
1358
1359 trace_f2fs_writepages(mapping->host, wbc, DATA);
1360
1361 /* deal with chardevs and other special file */
1362 if (!mapping->a_ops->writepage)
1363 return 0;
1364
1365 /* skip writing if there is no dirty page in this inode */
1366 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1367 return 0;
1368
1369 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1370 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1371 available_free_memory(sbi, DIRTY_DENTS))
1372 goto skip_write;
1373
1374 /* skip writing during file defragment */
1375 if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1376 goto skip_write;
1377
1378 /* during POR, we don't need to trigger writepage at all. */
1379 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1380 goto skip_write;
1381
1382 diff = nr_pages_to_write(sbi, DATA, wbc);
1383
1384 if (!S_ISDIR(inode->i_mode)) {
1385 mutex_lock(&sbi->writepages);
1386 locked = true;
1387 }
1388 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1389 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1390 if (locked)
1391 mutex_unlock(&sbi->writepages);
1392
1393 remove_dirty_inode(inode);
1394
1395 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1396 return ret;
1397
1398 skip_write:
1399 wbc->pages_skipped += get_dirty_pages(inode);
1400 return 0;
1401 }
1402
1403 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1404 {
1405 struct inode *inode = mapping->host;
1406
1407 if (to > inode->i_size) {
1408 truncate_pagecache(inode, inode->i_size);
1409 truncate_blocks(inode, inode->i_size, true);
1410 }
1411 }
1412
1413 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1414 loff_t pos, unsigned len, unsigned flags,
1415 struct page **pagep, void **fsdata)
1416 {
1417 struct inode *inode = mapping->host;
1418 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1419 struct page *page = NULL;
1420 struct page *ipage;
1421 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1422 struct dnode_of_data dn;
1423 int err = 0;
1424
1425 trace_f2fs_write_begin(inode, pos, len, flags);
1426
1427 /*
1428 * We should check this at this moment to avoid deadlock on inode page
1429 * and #0 page. The locking rule for inline_data conversion should be:
1430 * lock_page(page #0) -> lock_page(inode_page)
1431 */
1432 if (index != 0) {
1433 err = f2fs_convert_inline_inode(inode);
1434 if (err)
1435 goto fail;
1436 }
1437 repeat:
1438 page = grab_cache_page_write_begin(mapping, index, flags);
1439 if (!page) {
1440 err = -ENOMEM;
1441 goto fail;
1442 }
1443
1444 *pagep = page;
1445
1446 f2fs_lock_op(sbi);
1447
1448 /* check inline_data */
1449 ipage = get_node_page(sbi, inode->i_ino);
1450 if (IS_ERR(ipage)) {
1451 err = PTR_ERR(ipage);
1452 goto unlock_fail;
1453 }
1454
1455 set_new_dnode(&dn, inode, ipage, ipage, 0);
1456
1457 if (f2fs_has_inline_data(inode)) {
1458 if (pos + len <= MAX_INLINE_DATA) {
1459 read_inline_data(page, ipage);
1460 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1461 sync_inode_page(&dn);
1462 goto put_next;
1463 }
1464 err = f2fs_convert_inline_page(&dn, page);
1465 if (err)
1466 goto put_fail;
1467 }
1468
1469 err = f2fs_get_block(&dn, index);
1470 if (err)
1471 goto put_fail;
1472 put_next:
1473 f2fs_put_dnode(&dn);
1474 f2fs_unlock_op(sbi);
1475
1476 if (dn.node_changed && has_not_enough_free_secs(sbi, 0)) {
1477 unlock_page(page);
1478 f2fs_balance_fs(sbi);
1479 lock_page(page);
1480 if (page->mapping != mapping) {
1481 /* The page got truncated from under us */
1482 f2fs_put_page(page, 1);
1483 goto repeat;
1484 }
1485 }
1486
1487 f2fs_wait_on_page_writeback(page, DATA);
1488
1489 /* wait for GCed encrypted page writeback */
1490 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1491 f2fs_wait_on_encrypted_page_writeback(sbi, dn.data_blkaddr);
1492
1493 if (len == PAGE_CACHE_SIZE)
1494 goto out_update;
1495 if (PageUptodate(page))
1496 goto out_clear;
1497
1498 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1499 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1500 unsigned end = start + len;
1501
1502 /* Reading beyond i_size is simple: memset to zero */
1503 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1504 goto out_update;
1505 }
1506
1507 if (dn.data_blkaddr == NEW_ADDR) {
1508 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1509 } else {
1510 struct f2fs_io_info fio = {
1511 .sbi = sbi,
1512 .type = DATA,
1513 .rw = READ_SYNC,
1514 .blk_addr = dn.data_blkaddr,
1515 .page = page,
1516 .encrypted_page = NULL,
1517 };
1518 err = f2fs_submit_page_bio(&fio);
1519 if (err)
1520 goto fail;
1521
1522 lock_page(page);
1523 if (unlikely(!PageUptodate(page))) {
1524 err = -EIO;
1525 goto fail;
1526 }
1527 if (unlikely(page->mapping != mapping)) {
1528 f2fs_put_page(page, 1);
1529 goto repeat;
1530 }
1531
1532 /* avoid symlink page */
1533 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1534 err = f2fs_decrypt_one(inode, page);
1535 if (err)
1536 goto fail;
1537 }
1538 }
1539 out_update:
1540 SetPageUptodate(page);
1541 out_clear:
1542 clear_cold_data(page);
1543 return 0;
1544
1545 put_fail:
1546 f2fs_put_dnode(&dn);
1547 unlock_fail:
1548 f2fs_unlock_op(sbi);
1549 fail:
1550 f2fs_put_page(page, 1);
1551 f2fs_write_failed(mapping, pos + len);
1552 return err;
1553 }
1554
1555 static int f2fs_write_end(struct file *file,
1556 struct address_space *mapping,
1557 loff_t pos, unsigned len, unsigned copied,
1558 struct page *page, void *fsdata)
1559 {
1560 struct inode *inode = page->mapping->host;
1561
1562 trace_f2fs_write_end(inode, pos, len, copied);
1563
1564 set_page_dirty(page);
1565
1566 if (pos + copied > i_size_read(inode)) {
1567 i_size_write(inode, pos + copied);
1568 mark_inode_dirty(inode);
1569 update_inode_page(inode);
1570 }
1571
1572 f2fs_put_page(page, 1);
1573 return copied;
1574 }
1575
1576 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1577 loff_t offset)
1578 {
1579 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1580
1581 if (offset & blocksize_mask)
1582 return -EINVAL;
1583
1584 if (iov_iter_alignment(iter) & blocksize_mask)
1585 return -EINVAL;
1586
1587 return 0;
1588 }
1589
1590 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1591 loff_t offset)
1592 {
1593 struct file *file = iocb->ki_filp;
1594 struct address_space *mapping = file->f_mapping;
1595 struct inode *inode = mapping->host;
1596 size_t count = iov_iter_count(iter);
1597 int err;
1598
1599 /* we don't need to use inline_data strictly */
1600 err = f2fs_convert_inline_inode(inode);
1601 if (err)
1602 return err;
1603
1604 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1605 return 0;
1606
1607 err = check_direct_IO(inode, iter, offset);
1608 if (err)
1609 return err;
1610
1611 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1612
1613 if (iov_iter_rw(iter) == WRITE) {
1614 __allocate_data_blocks(inode, offset, count);
1615 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1616 err = -EIO;
1617 goto out;
1618 }
1619 }
1620
1621 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1622 out:
1623 if (err < 0 && iov_iter_rw(iter) == WRITE)
1624 f2fs_write_failed(mapping, offset + count);
1625
1626 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1627
1628 return err;
1629 }
1630
1631 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1632 unsigned int length)
1633 {
1634 struct inode *inode = page->mapping->host;
1635 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1636
1637 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1638 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1639 return;
1640
1641 if (PageDirty(page)) {
1642 if (inode->i_ino == F2FS_META_INO(sbi))
1643 dec_page_count(sbi, F2FS_DIRTY_META);
1644 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1645 dec_page_count(sbi, F2FS_DIRTY_NODES);
1646 else
1647 inode_dec_dirty_pages(inode);
1648 }
1649
1650 /* This is atomic written page, keep Private */
1651 if (IS_ATOMIC_WRITTEN_PAGE(page))
1652 return;
1653
1654 ClearPagePrivate(page);
1655 }
1656
1657 int f2fs_release_page(struct page *page, gfp_t wait)
1658 {
1659 /* If this is dirty page, keep PagePrivate */
1660 if (PageDirty(page))
1661 return 0;
1662
1663 /* This is atomic written page, keep Private */
1664 if (IS_ATOMIC_WRITTEN_PAGE(page))
1665 return 0;
1666
1667 ClearPagePrivate(page);
1668 return 1;
1669 }
1670
1671 static int f2fs_set_data_page_dirty(struct page *page)
1672 {
1673 struct address_space *mapping = page->mapping;
1674 struct inode *inode = mapping->host;
1675
1676 trace_f2fs_set_page_dirty(page, DATA);
1677
1678 SetPageUptodate(page);
1679
1680 if (f2fs_is_atomic_file(inode)) {
1681 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1682 register_inmem_page(inode, page);
1683 return 1;
1684 }
1685 /*
1686 * Previously, this page has been registered, we just
1687 * return here.
1688 */
1689 return 0;
1690 }
1691
1692 if (!PageDirty(page)) {
1693 __set_page_dirty_nobuffers(page);
1694 update_dirty_page(inode, page);
1695 return 1;
1696 }
1697 return 0;
1698 }
1699
1700 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1701 {
1702 struct inode *inode = mapping->host;
1703
1704 if (f2fs_has_inline_data(inode))
1705 return 0;
1706
1707 /* make sure allocating whole blocks */
1708 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1709 filemap_write_and_wait(mapping);
1710
1711 return generic_block_bmap(mapping, block, get_data_block_bmap);
1712 }
1713
1714 const struct address_space_operations f2fs_dblock_aops = {
1715 .readpage = f2fs_read_data_page,
1716 .readpages = f2fs_read_data_pages,
1717 .writepage = f2fs_write_data_page,
1718 .writepages = f2fs_write_data_pages,
1719 .write_begin = f2fs_write_begin,
1720 .write_end = f2fs_write_end,
1721 .set_page_dirty = f2fs_set_data_page_dirty,
1722 .invalidatepage = f2fs_invalidate_page,
1723 .releasepage = f2fs_release_page,
1724 .direct_IO = f2fs_direct_IO,
1725 .bmap = f2fs_bmap,
1726 };
Ubuntu Artful kernel mirror