projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: introduce f2fs_i_size_write with mark_inode_dirty_sync
[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)
102 {
103 if (!is_read_io(rw))
104 atomic_inc(&sbi->nr_wb_bios);
105 submit_bio(rw, bio);
106 }
107
108 static void __submit_merged_bio(struct f2fs_bio_info *io)
109 {
110 struct f2fs_io_info *fio = &io->fio;
111
112 if (!io->bio)
113 return;
114
115 if (is_read_io(fio->rw))
116 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
117 else
118 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
119
120 __submit_bio(io->sbi, fio->rw, io->bio);
121 io->bio = NULL;
122 }
123
124 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
125 struct page *page, nid_t ino)
126 {
127 struct bio_vec *bvec;
128 struct page *target;
129 int i;
130
131 if (!io->bio)
132 return false;
133
134 if (!inode && !page && !ino)
135 return true;
136
137 bio_for_each_segment_all(bvec, io->bio, i) {
138
139 if (bvec->bv_page->mapping)
140 target = bvec->bv_page;
141 else
142 target = fscrypt_control_page(bvec->bv_page);
143
144 if (inode && inode == target->mapping->host)
145 return true;
146 if (page && page == target)
147 return true;
148 if (ino && ino == ino_of_node(target))
149 return true;
150 }
151
152 return false;
153 }
154
155 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
156 struct page *page, nid_t ino,
157 enum page_type type)
158 {
159 enum page_type btype = PAGE_TYPE_OF_BIO(type);
160 struct f2fs_bio_info *io = &sbi->write_io[btype];
161 bool ret;
162
163 down_read(&io->io_rwsem);
164 ret = __has_merged_page(io, inode, page, ino);
165 up_read(&io->io_rwsem);
166 return ret;
167 }
168
169 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
170 struct inode *inode, struct page *page,
171 nid_t ino, enum page_type type, int rw)
172 {
173 enum page_type btype = PAGE_TYPE_OF_BIO(type);
174 struct f2fs_bio_info *io;
175
176 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
177
178 down_write(&io->io_rwsem);
179
180 if (!__has_merged_page(io, inode, page, ino))
181 goto out;
182
183 /* change META to META_FLUSH in the checkpoint procedure */
184 if (type >= META_FLUSH) {
185 io->fio.type = META_FLUSH;
186 if (test_opt(sbi, NOBARRIER))
187 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
188 else
189 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
190 }
191 __submit_merged_bio(io);
192 out:
193 up_write(&io->io_rwsem);
194 }
195
196 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
197 int rw)
198 {
199 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
200 }
201
202 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
203 struct inode *inode, struct page *page,
204 nid_t ino, enum page_type type, int rw)
205 {
206 if (has_merged_page(sbi, inode, page, ino, type))
207 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
208 }
209
210 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
211 {
212 f2fs_submit_merged_bio(sbi, DATA, WRITE);
213 f2fs_submit_merged_bio(sbi, NODE, WRITE);
214 f2fs_submit_merged_bio(sbi, META, WRITE);
215 }
216
217 /*
218 * Fill the locked page with data located in the block address.
219 * Return unlocked page.
220 */
221 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
222 {
223 struct bio *bio;
224 struct page *page = fio->encrypted_page ?
225 fio->encrypted_page : fio->page;
226
227 trace_f2fs_submit_page_bio(page, fio);
228 f2fs_trace_ios(fio, 0);
229
230 /* Allocate a new bio */
231 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
232
233 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
234 bio_put(bio);
235 return -EFAULT;
236 }
237
238 __submit_bio(fio->sbi, fio->rw, bio);
239 return 0;
240 }
241
242 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
243 {
244 struct f2fs_sb_info *sbi = fio->sbi;
245 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
246 struct f2fs_bio_info *io;
247 bool is_read = is_read_io(fio->rw);
248 struct page *bio_page;
249
250 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
251
252 if (fio->old_blkaddr != NEW_ADDR)
253 verify_block_addr(sbi, fio->old_blkaddr);
254 verify_block_addr(sbi, fio->new_blkaddr);
255
256 down_write(&io->io_rwsem);
257
258 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
259 io->fio.rw != fio->rw))
260 __submit_merged_bio(io);
261 alloc_new:
262 if (io->bio == NULL) {
263 int bio_blocks = MAX_BIO_BLOCKS(sbi);
264
265 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
266 bio_blocks, is_read);
267 io->fio = *fio;
268 }
269
270 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
271
272 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
273 PAGE_SIZE) {
274 __submit_merged_bio(io);
275 goto alloc_new;
276 }
277
278 io->last_block_in_bio = fio->new_blkaddr;
279 f2fs_trace_ios(fio, 0);
280
281 up_write(&io->io_rwsem);
282 trace_f2fs_submit_page_mbio(fio->page, fio);
283 }
284
285 static void __set_data_blkaddr(struct dnode_of_data *dn)
286 {
287 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
288 __le32 *addr_array;
289
290 /* Get physical address of data block */
291 addr_array = blkaddr_in_node(rn);
292 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
293 }
294
295 /*
296 * Lock ordering for the change of data block address:
297 * ->data_page
298 * ->node_page
299 * update block addresses in the node page
300 */
301 void set_data_blkaddr(struct dnode_of_data *dn)
302 {
303 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
304 __set_data_blkaddr(dn);
305 if (set_page_dirty(dn->node_page))
306 dn->node_changed = true;
307 }
308
309 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
310 {
311 dn->data_blkaddr = blkaddr;
312 set_data_blkaddr(dn);
313 f2fs_update_extent_cache(dn);
314 }
315
316 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
317 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
318 {
319 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
320
321 if (!count)
322 return 0;
323
324 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
325 return -EPERM;
326 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
327 return -ENOSPC;
328
329 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
330 dn->ofs_in_node, count);
331
332 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
333
334 for (; count > 0; dn->ofs_in_node++) {
335 block_t blkaddr =
336 datablock_addr(dn->node_page, dn->ofs_in_node);
337 if (blkaddr == NULL_ADDR) {
338 dn->data_blkaddr = NEW_ADDR;
339 __set_data_blkaddr(dn);
340 count--;
341 }
342 }
343
344 if (set_page_dirty(dn->node_page))
345 dn->node_changed = true;
346
347 mark_inode_dirty(dn->inode);
348 sync_inode_page(dn);
349 return 0;
350 }
351
352 /* Should keep dn->ofs_in_node unchanged */
353 int reserve_new_block(struct dnode_of_data *dn)
354 {
355 unsigned int ofs_in_node = dn->ofs_in_node;
356 int ret;
357
358 ret = reserve_new_blocks(dn, 1);
359 dn->ofs_in_node = ofs_in_node;
360 return ret;
361 }
362
363 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
364 {
365 bool need_put = dn->inode_page ? false : true;
366 int err;
367
368 err = get_dnode_of_data(dn, index, ALLOC_NODE);
369 if (err)
370 return err;
371
372 if (dn->data_blkaddr == NULL_ADDR)
373 err = reserve_new_block(dn);
374 if (err || need_put)
375 f2fs_put_dnode(dn);
376 return err;
377 }
378
379 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
380 {
381 struct extent_info ei;
382 struct inode *inode = dn->inode;
383
384 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
385 dn->data_blkaddr = ei.blk + index - ei.fofs;
386 return 0;
387 }
388
389 return f2fs_reserve_block(dn, index);
390 }
391
392 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
393 int rw, bool for_write)
394 {
395 struct address_space *mapping = inode->i_mapping;
396 struct dnode_of_data dn;
397 struct page *page;
398 struct extent_info ei;
399 int err;
400 struct f2fs_io_info fio = {
401 .sbi = F2FS_I_SB(inode),
402 .type = DATA,
403 .rw = rw,
404 .encrypted_page = NULL,
405 };
406
407 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
408 return read_mapping_page(mapping, index, NULL);
409
410 page = f2fs_grab_cache_page(mapping, index, for_write);
411 if (!page)
412 return ERR_PTR(-ENOMEM);
413
414 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
415 dn.data_blkaddr = ei.blk + index - ei.fofs;
416 goto got_it;
417 }
418
419 set_new_dnode(&dn, inode, NULL, NULL, 0);
420 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
421 if (err)
422 goto put_err;
423 f2fs_put_dnode(&dn);
424
425 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
426 err = -ENOENT;
427 goto put_err;
428 }
429 got_it:
430 if (PageUptodate(page)) {
431 unlock_page(page);
432 return page;
433 }
434
435 /*
436 * A new dentry page is allocated but not able to be written, since its
437 * new inode page couldn't be allocated due to -ENOSPC.
438 * In such the case, its blkaddr can be remained as NEW_ADDR.
439 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
440 */
441 if (dn.data_blkaddr == NEW_ADDR) {
442 zero_user_segment(page, 0, PAGE_SIZE);
443 SetPageUptodate(page);
444 unlock_page(page);
445 return page;
446 }
447
448 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
449 fio.page = page;
450 err = f2fs_submit_page_bio(&fio);
451 if (err)
452 goto put_err;
453 return page;
454
455 put_err:
456 f2fs_put_page(page, 1);
457 return ERR_PTR(err);
458 }
459
460 struct page *find_data_page(struct inode *inode, pgoff_t index)
461 {
462 struct address_space *mapping = inode->i_mapping;
463 struct page *page;
464
465 page = find_get_page(mapping, index);
466 if (page && PageUptodate(page))
467 return page;
468 f2fs_put_page(page, 0);
469
470 page = get_read_data_page(inode, index, READ_SYNC, false);
471 if (IS_ERR(page))
472 return page;
473
474 if (PageUptodate(page))
475 return page;
476
477 wait_on_page_locked(page);
478 if (unlikely(!PageUptodate(page))) {
479 f2fs_put_page(page, 0);
480 return ERR_PTR(-EIO);
481 }
482 return page;
483 }
484
485 /*
486 * If it tries to access a hole, return an error.
487 * Because, the callers, functions in dir.c and GC, should be able to know
488 * whether this page exists or not.
489 */
490 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
491 bool for_write)
492 {
493 struct address_space *mapping = inode->i_mapping;
494 struct page *page;
495 repeat:
496 page = get_read_data_page(inode, index, READ_SYNC, for_write);
497 if (IS_ERR(page))
498 return page;
499
500 /* wait for read completion */
501 lock_page(page);
502 if (unlikely(!PageUptodate(page))) {
503 f2fs_put_page(page, 1);
504 return ERR_PTR(-EIO);
505 }
506 if (unlikely(page->mapping != mapping)) {
507 f2fs_put_page(page, 1);
508 goto repeat;
509 }
510 return page;
511 }
512
513 /*
514 * Caller ensures that this data page is never allocated.
515 * A new zero-filled data page is allocated in the page cache.
516 *
517 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
518 * f2fs_unlock_op().
519 * Note that, ipage is set only by make_empty_dir, and if any error occur,
520 * ipage should be released by this function.
521 */
522 struct page *get_new_data_page(struct inode *inode,
523 struct page *ipage, pgoff_t index, bool new_i_size)
524 {
525 struct address_space *mapping = inode->i_mapping;
526 struct page *page;
527 struct dnode_of_data dn;
528 int err;
529
530 page = f2fs_grab_cache_page(mapping, index, true);
531 if (!page) {
532 /*
533 * before exiting, we should make sure ipage will be released
534 * if any error occur.
535 */
536 f2fs_put_page(ipage, 1);
537 return ERR_PTR(-ENOMEM);
538 }
539
540 set_new_dnode(&dn, inode, ipage, NULL, 0);
541 err = f2fs_reserve_block(&dn, index);
542 if (err) {
543 f2fs_put_page(page, 1);
544 return ERR_PTR(err);
545 }
546 if (!ipage)
547 f2fs_put_dnode(&dn);
548
549 if (PageUptodate(page))
550 goto got_it;
551
552 if (dn.data_blkaddr == NEW_ADDR) {
553 zero_user_segment(page, 0, PAGE_SIZE);
554 SetPageUptodate(page);
555 } else {
556 f2fs_put_page(page, 1);
557
558 /* if ipage exists, blkaddr should be NEW_ADDR */
559 f2fs_bug_on(F2FS_I_SB(inode), ipage);
560 page = get_lock_data_page(inode, index, true);
561 if (IS_ERR(page))
562 return page;
563 }
564 got_it:
565 if (new_i_size && i_size_read(inode) <
566 ((loff_t)(index + 1) << PAGE_SHIFT)) {
567 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
568 /* Only the directory inode sets new_i_size */
569 set_inode_flag(inode, FI_UPDATE_DIR);
570 }
571 return page;
572 }
573
574 static int __allocate_data_block(struct dnode_of_data *dn)
575 {
576 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
577 struct f2fs_summary sum;
578 struct node_info ni;
579 int seg = CURSEG_WARM_DATA;
580 pgoff_t fofs;
581 blkcnt_t count = 1;
582
583 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
584 return -EPERM;
585
586 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
587 if (dn->data_blkaddr == NEW_ADDR)
588 goto alloc;
589
590 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
591 return -ENOSPC;
592
593 alloc:
594 get_node_info(sbi, dn->nid, &ni);
595 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
596
597 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
598 seg = CURSEG_DIRECT_IO;
599
600 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
601 &sum, seg);
602 set_data_blkaddr(dn);
603
604 /* update i_size */
605 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
606 dn->ofs_in_node;
607 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
608 f2fs_i_size_write(dn->inode,
609 ((loff_t)(fofs + 1) << PAGE_SHIFT));
610 return 0;
611 }
612
613 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
614 {
615 struct inode *inode = file_inode(iocb->ki_filp);
616 struct f2fs_map_blocks map;
617 ssize_t ret = 0;
618
619 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
620 map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
621 map.m_next_pgofs = NULL;
622
623 if (f2fs_encrypted_inode(inode))
624 return 0;
625
626 if (iocb->ki_flags & IOCB_DIRECT) {
627 ret = f2fs_convert_inline_inode(inode);
628 if (ret)
629 return ret;
630 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
631 }
632 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
633 ret = f2fs_convert_inline_inode(inode);
634 if (ret)
635 return ret;
636 }
637 if (!f2fs_has_inline_data(inode))
638 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
639 return ret;
640 }
641
642 /*
643 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
644 * f2fs_map_blocks structure.
645 * If original data blocks are allocated, then give them to blockdev.
646 * Otherwise,
647 * a. preallocate requested block addresses
648 * b. do not use extent cache for better performance
649 * c. give the block addresses to blockdev
650 */
651 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
652 int create, int flag)
653 {
654 unsigned int maxblocks = map->m_len;
655 struct dnode_of_data dn;
656 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
657 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
658 pgoff_t pgofs, end_offset, end;
659 int err = 0, ofs = 1;
660 unsigned int ofs_in_node, last_ofs_in_node;
661 blkcnt_t prealloc;
662 struct extent_info ei;
663 bool allocated = false;
664 block_t blkaddr;
665
666 map->m_len = 0;
667 map->m_flags = 0;
668
669 /* it only supports block size == page size */
670 pgofs = (pgoff_t)map->m_lblk;
671 end = pgofs + maxblocks;
672
673 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
674 map->m_pblk = ei.blk + pgofs - ei.fofs;
675 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
676 map->m_flags = F2FS_MAP_MAPPED;
677 goto out;
678 }
679
680 next_dnode:
681 if (create)
682 f2fs_lock_op(sbi);
683
684 /* When reading holes, we need its node page */
685 set_new_dnode(&dn, inode, NULL, NULL, 0);
686 err = get_dnode_of_data(&dn, pgofs, mode);
687 if (err) {
688 if (flag == F2FS_GET_BLOCK_BMAP)
689 map->m_pblk = 0;
690 if (err == -ENOENT) {
691 err = 0;
692 if (map->m_next_pgofs)
693 *map->m_next_pgofs =
694 get_next_page_offset(&dn, pgofs);
695 }
696 goto unlock_out;
697 }
698
699 prealloc = 0;
700 ofs_in_node = dn.ofs_in_node;
701 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
702
703 next_block:
704 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
705
706 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
707 if (create) {
708 if (unlikely(f2fs_cp_error(sbi))) {
709 err = -EIO;
710 goto sync_out;
711 }
712 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
713 if (blkaddr == NULL_ADDR) {
714 prealloc++;
715 last_ofs_in_node = dn.ofs_in_node;
716 }
717 } else {
718 err = __allocate_data_block(&dn);
719 if (!err) {
720 set_inode_flag(inode, FI_APPEND_WRITE);
721 allocated = true;
722 }
723 }
724 if (err)
725 goto sync_out;
726 map->m_flags = F2FS_MAP_NEW;
727 blkaddr = dn.data_blkaddr;
728 } else {
729 if (flag == F2FS_GET_BLOCK_BMAP) {
730 map->m_pblk = 0;
731 goto sync_out;
732 }
733 if (flag == F2FS_GET_BLOCK_FIEMAP &&
734 blkaddr == NULL_ADDR) {
735 if (map->m_next_pgofs)
736 *map->m_next_pgofs = pgofs + 1;
737 }
738 if (flag != F2FS_GET_BLOCK_FIEMAP ||
739 blkaddr != NEW_ADDR)
740 goto sync_out;
741 }
742 }
743
744 if (flag == F2FS_GET_BLOCK_PRE_AIO)
745 goto skip;
746
747 if (map->m_len == 0) {
748 /* preallocated unwritten block should be mapped for fiemap. */
749 if (blkaddr == NEW_ADDR)
750 map->m_flags |= F2FS_MAP_UNWRITTEN;
751 map->m_flags |= F2FS_MAP_MAPPED;
752
753 map->m_pblk = blkaddr;
754 map->m_len = 1;
755 } else if ((map->m_pblk != NEW_ADDR &&
756 blkaddr == (map->m_pblk + ofs)) ||
757 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
758 flag == F2FS_GET_BLOCK_PRE_DIO) {
759 ofs++;
760 map->m_len++;
761 } else {
762 goto sync_out;
763 }
764
765 skip:
766 dn.ofs_in_node++;
767 pgofs++;
768
769 /* preallocate blocks in batch for one dnode page */
770 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
771 (pgofs == end || dn.ofs_in_node == end_offset)) {
772
773 dn.ofs_in_node = ofs_in_node;
774 err = reserve_new_blocks(&dn, prealloc);
775 if (err)
776 goto sync_out;
777
778 map->m_len += dn.ofs_in_node - ofs_in_node;
779 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
780 err = -ENOSPC;
781 goto sync_out;
782 }
783 dn.ofs_in_node = end_offset;
784 }
785
786 if (pgofs >= end)
787 goto sync_out;
788 else if (dn.ofs_in_node < end_offset)
789 goto next_block;
790
791 if (allocated)
792 sync_inode_page(&dn);
793 f2fs_put_dnode(&dn);
794
795 if (create) {
796 f2fs_unlock_op(sbi);
797 f2fs_balance_fs(sbi, allocated);
798 }
799 allocated = false;
800 goto next_dnode;
801
802 sync_out:
803 if (allocated)
804 sync_inode_page(&dn);
805 f2fs_put_dnode(&dn);
806 unlock_out:
807 if (create) {
808 f2fs_unlock_op(sbi);
809 f2fs_balance_fs(sbi, allocated);
810 }
811 out:
812 trace_f2fs_map_blocks(inode, map, err);
813 return err;
814 }
815
816 static int __get_data_block(struct inode *inode, sector_t iblock,
817 struct buffer_head *bh, int create, int flag,
818 pgoff_t *next_pgofs)
819 {
820 struct f2fs_map_blocks map;
821 int ret;
822
823 map.m_lblk = iblock;
824 map.m_len = bh->b_size >> inode->i_blkbits;
825 map.m_next_pgofs = next_pgofs;
826
827 ret = f2fs_map_blocks(inode, &map, create, flag);
828 if (!ret) {
829 map_bh(bh, inode->i_sb, map.m_pblk);
830 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
831 bh->b_size = map.m_len << inode->i_blkbits;
832 }
833 return ret;
834 }
835
836 static int get_data_block(struct inode *inode, sector_t iblock,
837 struct buffer_head *bh_result, int create, int flag,
838 pgoff_t *next_pgofs)
839 {
840 return __get_data_block(inode, iblock, bh_result, create,
841 flag, next_pgofs);
842 }
843
844 static int get_data_block_dio(struct inode *inode, sector_t iblock,
845 struct buffer_head *bh_result, int create)
846 {
847 return __get_data_block(inode, iblock, bh_result, create,
848 F2FS_GET_BLOCK_DIO, NULL);
849 }
850
851 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
852 struct buffer_head *bh_result, int create)
853 {
854 /* Block number less than F2FS MAX BLOCKS */
855 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
856 return -EFBIG;
857
858 return __get_data_block(inode, iblock, bh_result, create,
859 F2FS_GET_BLOCK_BMAP, NULL);
860 }
861
862 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
863 {
864 return (offset >> inode->i_blkbits);
865 }
866
867 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
868 {
869 return (blk << inode->i_blkbits);
870 }
871
872 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
873 u64 start, u64 len)
874 {
875 struct buffer_head map_bh;
876 sector_t start_blk, last_blk;
877 pgoff_t next_pgofs;
878 loff_t isize;
879 u64 logical = 0, phys = 0, size = 0;
880 u32 flags = 0;
881 int ret = 0;
882
883 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
884 if (ret)
885 return ret;
886
887 if (f2fs_has_inline_data(inode)) {
888 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
889 if (ret != -EAGAIN)
890 return ret;
891 }
892
893 inode_lock(inode);
894
895 isize = i_size_read(inode);
896 if (start >= isize)
897 goto out;
898
899 if (start + len > isize)
900 len = isize - start;
901
902 if (logical_to_blk(inode, len) == 0)
903 len = blk_to_logical(inode, 1);
904
905 start_blk = logical_to_blk(inode, start);
906 last_blk = logical_to_blk(inode, start + len - 1);
907
908 next:
909 memset(&map_bh, 0, sizeof(struct buffer_head));
910 map_bh.b_size = len;
911
912 ret = get_data_block(inode, start_blk, &map_bh, 0,
913 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
914 if (ret)
915 goto out;
916
917 /* HOLE */
918 if (!buffer_mapped(&map_bh)) {
919 start_blk = next_pgofs;
920 /* Go through holes util pass the EOF */
921 if (blk_to_logical(inode, start_blk) < isize)
922 goto prep_next;
923 /* Found a hole beyond isize means no more extents.
924 * Note that the premise is that filesystems don't
925 * punch holes beyond isize and keep size unchanged.
926 */
927 flags |= FIEMAP_EXTENT_LAST;
928 }
929
930 if (size) {
931 if (f2fs_encrypted_inode(inode))
932 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
933
934 ret = fiemap_fill_next_extent(fieinfo, logical,
935 phys, size, flags);
936 }
937
938 if (start_blk > last_blk || ret)
939 goto out;
940
941 logical = blk_to_logical(inode, start_blk);
942 phys = blk_to_logical(inode, map_bh.b_blocknr);
943 size = map_bh.b_size;
944 flags = 0;
945 if (buffer_unwritten(&map_bh))
946 flags = FIEMAP_EXTENT_UNWRITTEN;
947
948 start_blk += logical_to_blk(inode, size);
949
950 prep_next:
951 cond_resched();
952 if (fatal_signal_pending(current))
953 ret = -EINTR;
954 else
955 goto next;
956 out:
957 if (ret == 1)
958 ret = 0;
959
960 inode_unlock(inode);
961 return ret;
962 }
963
964 /*
965 * This function was originally taken from fs/mpage.c, and customized for f2fs.
966 * Major change was from block_size == page_size in f2fs by default.
967 */
968 static int f2fs_mpage_readpages(struct address_space *mapping,
969 struct list_head *pages, struct page *page,
970 unsigned nr_pages)
971 {
972 struct bio *bio = NULL;
973 unsigned page_idx;
974 sector_t last_block_in_bio = 0;
975 struct inode *inode = mapping->host;
976 const unsigned blkbits = inode->i_blkbits;
977 const unsigned blocksize = 1 << blkbits;
978 sector_t block_in_file;
979 sector_t last_block;
980 sector_t last_block_in_file;
981 sector_t block_nr;
982 struct block_device *bdev = inode->i_sb->s_bdev;
983 struct f2fs_map_blocks map;
984
985 map.m_pblk = 0;
986 map.m_lblk = 0;
987 map.m_len = 0;
988 map.m_flags = 0;
989 map.m_next_pgofs = NULL;
990
991 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
992
993 prefetchw(&page->flags);
994 if (pages) {
995 page = list_entry(pages->prev, struct page, lru);
996 list_del(&page->lru);
997 if (add_to_page_cache_lru(page, mapping,
998 page->index, GFP_KERNEL))
999 goto next_page;
1000 }
1001
1002 block_in_file = (sector_t)page->index;
1003 last_block = block_in_file + nr_pages;
1004 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1005 blkbits;
1006 if (last_block > last_block_in_file)
1007 last_block = last_block_in_file;
1008
1009 /*
1010 * Map blocks using the previous result first.
1011 */
1012 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1013 block_in_file > map.m_lblk &&
1014 block_in_file < (map.m_lblk + map.m_len))
1015 goto got_it;
1016
1017 /*
1018 * Then do more f2fs_map_blocks() calls until we are
1019 * done with this page.
1020 */
1021 map.m_flags = 0;
1022
1023 if (block_in_file < last_block) {
1024 map.m_lblk = block_in_file;
1025 map.m_len = last_block - block_in_file;
1026
1027 if (f2fs_map_blocks(inode, &map, 0,
1028 F2FS_GET_BLOCK_READ))
1029 goto set_error_page;
1030 }
1031 got_it:
1032 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1033 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1034 SetPageMappedToDisk(page);
1035
1036 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1037 SetPageUptodate(page);
1038 goto confused;
1039 }
1040 } else {
1041 zero_user_segment(page, 0, PAGE_SIZE);
1042 SetPageUptodate(page);
1043 unlock_page(page);
1044 goto next_page;
1045 }
1046
1047 /*
1048 * This page will go to BIO. Do we need to send this
1049 * BIO off first?
1050 */
1051 if (bio && (last_block_in_bio != block_nr - 1)) {
1052 submit_and_realloc:
1053 __submit_bio(F2FS_I_SB(inode), READ, bio);
1054 bio = NULL;
1055 }
1056 if (bio == NULL) {
1057 struct fscrypt_ctx *ctx = NULL;
1058
1059 if (f2fs_encrypted_inode(inode) &&
1060 S_ISREG(inode->i_mode)) {
1061
1062 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1063 if (IS_ERR(ctx))
1064 goto set_error_page;
1065
1066 /* wait the page to be moved by cleaning */
1067 f2fs_wait_on_encrypted_page_writeback(
1068 F2FS_I_SB(inode), block_nr);
1069 }
1070
1071 bio = bio_alloc(GFP_KERNEL,
1072 min_t(int, nr_pages, BIO_MAX_PAGES));
1073 if (!bio) {
1074 if (ctx)
1075 fscrypt_release_ctx(ctx);
1076 goto set_error_page;
1077 }
1078 bio->bi_bdev = bdev;
1079 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1080 bio->bi_end_io = f2fs_read_end_io;
1081 bio->bi_private = ctx;
1082 }
1083
1084 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1085 goto submit_and_realloc;
1086
1087 last_block_in_bio = block_nr;
1088 goto next_page;
1089 set_error_page:
1090 SetPageError(page);
1091 zero_user_segment(page, 0, PAGE_SIZE);
1092 unlock_page(page);
1093 goto next_page;
1094 confused:
1095 if (bio) {
1096 __submit_bio(F2FS_I_SB(inode), READ, bio);
1097 bio = NULL;
1098 }
1099 unlock_page(page);
1100 next_page:
1101 if (pages)
1102 put_page(page);
1103 }
1104 BUG_ON(pages && !list_empty(pages));
1105 if (bio)
1106 __submit_bio(F2FS_I_SB(inode), READ, bio);
1107 return 0;
1108 }
1109
1110 static int f2fs_read_data_page(struct file *file, struct page *page)
1111 {
1112 struct inode *inode = page->mapping->host;
1113 int ret = -EAGAIN;
1114
1115 trace_f2fs_readpage(page, DATA);
1116
1117 /* If the file has inline data, try to read it directly */
1118 if (f2fs_has_inline_data(inode))
1119 ret = f2fs_read_inline_data(inode, page);
1120 if (ret == -EAGAIN)
1121 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1122 return ret;
1123 }
1124
1125 static int f2fs_read_data_pages(struct file *file,
1126 struct address_space *mapping,
1127 struct list_head *pages, unsigned nr_pages)
1128 {
1129 struct inode *inode = file->f_mapping->host;
1130 struct page *page = list_entry(pages->prev, struct page, lru);
1131
1132 trace_f2fs_readpages(inode, page, nr_pages);
1133
1134 /* If the file has inline data, skip readpages */
1135 if (f2fs_has_inline_data(inode))
1136 return 0;
1137
1138 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1139 }
1140
1141 int do_write_data_page(struct f2fs_io_info *fio)
1142 {
1143 struct page *page = fio->page;
1144 struct inode *inode = page->mapping->host;
1145 struct dnode_of_data dn;
1146 int err = 0;
1147
1148 set_new_dnode(&dn, inode, NULL, NULL, 0);
1149 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1150 if (err)
1151 return err;
1152
1153 fio->old_blkaddr = dn.data_blkaddr;
1154
1155 /* This page is already truncated */
1156 if (fio->old_blkaddr == NULL_ADDR) {
1157 ClearPageUptodate(page);
1158 goto out_writepage;
1159 }
1160
1161 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1162 gfp_t gfp_flags = GFP_NOFS;
1163
1164 /* wait for GCed encrypted page writeback */
1165 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1166 fio->old_blkaddr);
1167 retry_encrypt:
1168 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1169 gfp_flags);
1170 if (IS_ERR(fio->encrypted_page)) {
1171 err = PTR_ERR(fio->encrypted_page);
1172 if (err == -ENOMEM) {
1173 /* flush pending ios and wait for a while */
1174 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1175 congestion_wait(BLK_RW_ASYNC, HZ/50);
1176 gfp_flags |= __GFP_NOFAIL;
1177 err = 0;
1178 goto retry_encrypt;
1179 }
1180 goto out_writepage;
1181 }
1182 }
1183
1184 set_page_writeback(page);
1185
1186 /*
1187 * If current allocation needs SSR,
1188 * it had better in-place writes for updated data.
1189 */
1190 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1191 !is_cold_data(page) &&
1192 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1193 need_inplace_update(inode))) {
1194 rewrite_data_page(fio);
1195 set_inode_flag(inode, FI_UPDATE_WRITE);
1196 trace_f2fs_do_write_data_page(page, IPU);
1197 } else {
1198 write_data_page(&dn, fio);
1199 trace_f2fs_do_write_data_page(page, OPU);
1200 set_inode_flag(inode, FI_APPEND_WRITE);
1201 if (page->index == 0)
1202 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1203 }
1204 out_writepage:
1205 f2fs_put_dnode(&dn);
1206 return err;
1207 }
1208
1209 static int f2fs_write_data_page(struct page *page,
1210 struct writeback_control *wbc)
1211 {
1212 struct inode *inode = page->mapping->host;
1213 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1214 loff_t i_size = i_size_read(inode);
1215 const pgoff_t end_index = ((unsigned long long) i_size)
1216 >> PAGE_SHIFT;
1217 unsigned offset = 0;
1218 bool need_balance_fs = false;
1219 int err = 0;
1220 struct f2fs_io_info fio = {
1221 .sbi = sbi,
1222 .type = DATA,
1223 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1224 .page = page,
1225 .encrypted_page = NULL,
1226 };
1227
1228 trace_f2fs_writepage(page, DATA);
1229
1230 if (page->index < end_index)
1231 goto write;
1232
1233 /*
1234 * If the offset is out-of-range of file size,
1235 * this page does not have to be written to disk.
1236 */
1237 offset = i_size & (PAGE_SIZE - 1);
1238 if ((page->index >= end_index + 1) || !offset)
1239 goto out;
1240
1241 zero_user_segment(page, offset, PAGE_SIZE);
1242 write:
1243 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1244 goto redirty_out;
1245 if (f2fs_is_drop_cache(inode))
1246 goto out;
1247 /* we should not write 0'th page having journal header */
1248 if (f2fs_is_volatile_file(inode) && (!page->index ||
1249 (!wbc->for_reclaim &&
1250 available_free_memory(sbi, BASE_CHECK))))
1251 goto redirty_out;
1252
1253 /* Dentry blocks are controlled by checkpoint */
1254 if (S_ISDIR(inode->i_mode)) {
1255 if (unlikely(f2fs_cp_error(sbi)))
1256 goto redirty_out;
1257 err = do_write_data_page(&fio);
1258 goto done;
1259 }
1260
1261 /* we should bypass data pages to proceed the kworkder jobs */
1262 if (unlikely(f2fs_cp_error(sbi))) {
1263 SetPageError(page);
1264 goto out;
1265 }
1266
1267 if (!wbc->for_reclaim)
1268 need_balance_fs = true;
1269 else if (has_not_enough_free_secs(sbi, 0))
1270 goto redirty_out;
1271
1272 err = -EAGAIN;
1273 f2fs_lock_op(sbi);
1274 if (f2fs_has_inline_data(inode))
1275 err = f2fs_write_inline_data(inode, page);
1276 if (err == -EAGAIN)
1277 err = do_write_data_page(&fio);
1278 f2fs_unlock_op(sbi);
1279 done:
1280 if (err && err != -ENOENT)
1281 goto redirty_out;
1282
1283 clear_cold_data(page);
1284 out:
1285 inode_dec_dirty_pages(inode);
1286 if (err)
1287 ClearPageUptodate(page);
1288
1289 if (wbc->for_reclaim) {
1290 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1291 remove_dirty_inode(inode);
1292 }
1293
1294 unlock_page(page);
1295 f2fs_balance_fs(sbi, need_balance_fs);
1296
1297 if (unlikely(f2fs_cp_error(sbi)))
1298 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1299
1300 return 0;
1301
1302 redirty_out:
1303 redirty_page_for_writepage(wbc, page);
1304 return AOP_WRITEPAGE_ACTIVATE;
1305 }
1306
1307 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1308 void *data)
1309 {
1310 struct address_space *mapping = data;
1311 int ret = mapping->a_ops->writepage(page, wbc);
1312 mapping_set_error(mapping, ret);
1313 return ret;
1314 }
1315
1316 /*
1317 * This function was copied from write_cche_pages from mm/page-writeback.c.
1318 * The major change is making write step of cold data page separately from
1319 * warm/hot data page.
1320 */
1321 static int f2fs_write_cache_pages(struct address_space *mapping,
1322 struct writeback_control *wbc, writepage_t writepage,
1323 void *data)
1324 {
1325 int ret = 0;
1326 int done = 0;
1327 struct pagevec pvec;
1328 int nr_pages;
1329 pgoff_t uninitialized_var(writeback_index);
1330 pgoff_t index;
1331 pgoff_t end; /* Inclusive */
1332 pgoff_t done_index;
1333 int cycled;
1334 int range_whole = 0;
1335 int tag;
1336 int step = 0;
1337
1338 pagevec_init(&pvec, 0);
1339 next:
1340 if (wbc->range_cyclic) {
1341 writeback_index = mapping->writeback_index; /* prev offset */
1342 index = writeback_index;
1343 if (index == 0)
1344 cycled = 1;
1345 else
1346 cycled = 0;
1347 end = -1;
1348 } else {
1349 index = wbc->range_start >> PAGE_SHIFT;
1350 end = wbc->range_end >> PAGE_SHIFT;
1351 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1352 range_whole = 1;
1353 cycled = 1; /* ignore range_cyclic tests */
1354 }
1355 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1356 tag = PAGECACHE_TAG_TOWRITE;
1357 else
1358 tag = PAGECACHE_TAG_DIRTY;
1359 retry:
1360 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1361 tag_pages_for_writeback(mapping, index, end);
1362 done_index = index;
1363 while (!done && (index <= end)) {
1364 int i;
1365
1366 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1367 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1368 if (nr_pages == 0)
1369 break;
1370
1371 for (i = 0; i < nr_pages; i++) {
1372 struct page *page = pvec.pages[i];
1373
1374 if (page->index > end) {
1375 done = 1;
1376 break;
1377 }
1378
1379 done_index = page->index;
1380
1381 lock_page(page);
1382
1383 if (unlikely(page->mapping != mapping)) {
1384 continue_unlock:
1385 unlock_page(page);
1386 continue;
1387 }
1388
1389 if (!PageDirty(page)) {
1390 /* someone wrote it for us */
1391 goto continue_unlock;
1392 }
1393
1394 if (step == is_cold_data(page))
1395 goto continue_unlock;
1396
1397 if (PageWriteback(page)) {
1398 if (wbc->sync_mode != WB_SYNC_NONE)
1399 f2fs_wait_on_page_writeback(page,
1400 DATA, true);
1401 else
1402 goto continue_unlock;
1403 }
1404
1405 BUG_ON(PageWriteback(page));
1406 if (!clear_page_dirty_for_io(page))
1407 goto continue_unlock;
1408
1409 ret = (*writepage)(page, wbc, data);
1410 if (unlikely(ret)) {
1411 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1412 unlock_page(page);
1413 ret = 0;
1414 } else {
1415 done_index = page->index + 1;
1416 done = 1;
1417 break;
1418 }
1419 }
1420
1421 if (--wbc->nr_to_write <= 0 &&
1422 wbc->sync_mode == WB_SYNC_NONE) {
1423 done = 1;
1424 break;
1425 }
1426 }
1427 pagevec_release(&pvec);
1428 cond_resched();
1429 }
1430
1431 if (step < 1) {
1432 step++;
1433 goto next;
1434 }
1435
1436 if (!cycled && !done) {
1437 cycled = 1;
1438 index = 0;
1439 end = writeback_index - 1;
1440 goto retry;
1441 }
1442 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1443 mapping->writeback_index = done_index;
1444
1445 return ret;
1446 }
1447
1448 static int f2fs_write_data_pages(struct address_space *mapping,
1449 struct writeback_control *wbc)
1450 {
1451 struct inode *inode = mapping->host;
1452 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1453 bool locked = false;
1454 int ret;
1455 long diff;
1456
1457 /* deal with chardevs and other special file */
1458 if (!mapping->a_ops->writepage)
1459 return 0;
1460
1461 /* skip writing if there is no dirty page in this inode */
1462 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1463 return 0;
1464
1465 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1466 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1467 available_free_memory(sbi, DIRTY_DENTS))
1468 goto skip_write;
1469
1470 /* skip writing during file defragment */
1471 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1472 goto skip_write;
1473
1474 /* during POR, we don't need to trigger writepage at all. */
1475 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1476 goto skip_write;
1477
1478 trace_f2fs_writepages(mapping->host, wbc, DATA);
1479
1480 diff = nr_pages_to_write(sbi, DATA, wbc);
1481
1482 if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1483 mutex_lock(&sbi->writepages);
1484 locked = true;
1485 }
1486 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1487 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1488 if (locked)
1489 mutex_unlock(&sbi->writepages);
1490
1491 remove_dirty_inode(inode);
1492
1493 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1494 return ret;
1495
1496 skip_write:
1497 wbc->pages_skipped += get_dirty_pages(inode);
1498 trace_f2fs_writepages(mapping->host, wbc, DATA);
1499 return 0;
1500 }
1501
1502 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1503 {
1504 struct inode *inode = mapping->host;
1505 loff_t i_size = i_size_read(inode);
1506
1507 if (to > i_size) {
1508 truncate_pagecache(inode, i_size);
1509 truncate_blocks(inode, i_size, true);
1510 }
1511 }
1512
1513 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1514 struct page *page, loff_t pos, unsigned len,
1515 block_t *blk_addr, bool *node_changed)
1516 {
1517 struct inode *inode = page->mapping->host;
1518 pgoff_t index = page->index;
1519 struct dnode_of_data dn;
1520 struct page *ipage;
1521 bool locked = false;
1522 struct extent_info ei;
1523 int err = 0;
1524
1525 /*
1526 * we already allocated all the blocks, so we don't need to get
1527 * the block addresses when there is no need to fill the page.
1528 */
1529 if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1530 len == PAGE_SIZE)
1531 return 0;
1532
1533 if (f2fs_has_inline_data(inode) ||
1534 (pos & PAGE_MASK) >= i_size_read(inode)) {
1535 f2fs_lock_op(sbi);
1536 locked = true;
1537 }
1538 restart:
1539 /* check inline_data */
1540 ipage = get_node_page(sbi, inode->i_ino);
1541 if (IS_ERR(ipage)) {
1542 err = PTR_ERR(ipage);
1543 goto unlock_out;
1544 }
1545
1546 set_new_dnode(&dn, inode, ipage, ipage, 0);
1547
1548 if (f2fs_has_inline_data(inode)) {
1549 if (pos + len <= MAX_INLINE_DATA) {
1550 read_inline_data(page, ipage);
1551 set_inode_flag(inode, FI_DATA_EXIST);
1552 if (inode->i_nlink)
1553 set_inline_node(ipage);
1554 } else {
1555 err = f2fs_convert_inline_page(&dn, page);
1556 if (err)
1557 goto out;
1558 if (dn.data_blkaddr == NULL_ADDR)
1559 err = f2fs_get_block(&dn, index);
1560 }
1561 } else if (locked) {
1562 err = f2fs_get_block(&dn, index);
1563 } else {
1564 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1565 dn.data_blkaddr = ei.blk + index - ei.fofs;
1566 } else {
1567 /* hole case */
1568 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1569 if (err || dn.data_blkaddr == NULL_ADDR) {
1570 f2fs_put_dnode(&dn);
1571 f2fs_lock_op(sbi);
1572 locked = true;
1573 goto restart;
1574 }
1575 }
1576 }
1577
1578 /* convert_inline_page can make node_changed */
1579 *blk_addr = dn.data_blkaddr;
1580 *node_changed = dn.node_changed;
1581 out:
1582 f2fs_put_dnode(&dn);
1583 unlock_out:
1584 if (locked)
1585 f2fs_unlock_op(sbi);
1586 return err;
1587 }
1588
1589 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1590 loff_t pos, unsigned len, unsigned flags,
1591 struct page **pagep, void **fsdata)
1592 {
1593 struct inode *inode = mapping->host;
1594 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1595 struct page *page = NULL;
1596 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1597 bool need_balance = false;
1598 block_t blkaddr = NULL_ADDR;
1599 int err = 0;
1600
1601 trace_f2fs_write_begin(inode, pos, len, flags);
1602
1603 /*
1604 * We should check this at this moment to avoid deadlock on inode page
1605 * and #0 page. The locking rule for inline_data conversion should be:
1606 * lock_page(page #0) -> lock_page(inode_page)
1607 */
1608 if (index != 0) {
1609 err = f2fs_convert_inline_inode(inode);
1610 if (err)
1611 goto fail;
1612 }
1613 repeat:
1614 page = grab_cache_page_write_begin(mapping, index, flags);
1615 if (!page) {
1616 err = -ENOMEM;
1617 goto fail;
1618 }
1619
1620 *pagep = page;
1621
1622 err = prepare_write_begin(sbi, page, pos, len,
1623 &blkaddr, &need_balance);
1624 if (err)
1625 goto fail;
1626
1627 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1628 unlock_page(page);
1629 f2fs_balance_fs(sbi, true);
1630 lock_page(page);
1631 if (page->mapping != mapping) {
1632 /* The page got truncated from under us */
1633 f2fs_put_page(page, 1);
1634 goto repeat;
1635 }
1636 }
1637
1638 f2fs_wait_on_page_writeback(page, DATA, false);
1639
1640 /* wait for GCed encrypted page writeback */
1641 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1642 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1643
1644 if (len == PAGE_SIZE)
1645 goto out_update;
1646 if (PageUptodate(page))
1647 goto out_clear;
1648
1649 if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1650 unsigned start = pos & (PAGE_SIZE - 1);
1651 unsigned end = start + len;
1652
1653 /* Reading beyond i_size is simple: memset to zero */
1654 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1655 goto out_update;
1656 }
1657
1658 if (blkaddr == NEW_ADDR) {
1659 zero_user_segment(page, 0, PAGE_SIZE);
1660 } else {
1661 struct f2fs_io_info fio = {
1662 .sbi = sbi,
1663 .type = DATA,
1664 .rw = READ_SYNC,
1665 .old_blkaddr = blkaddr,
1666 .new_blkaddr = blkaddr,
1667 .page = page,
1668 .encrypted_page = NULL,
1669 };
1670 err = f2fs_submit_page_bio(&fio);
1671 if (err)
1672 goto fail;
1673
1674 lock_page(page);
1675 if (unlikely(!PageUptodate(page))) {
1676 err = -EIO;
1677 goto fail;
1678 }
1679 if (unlikely(page->mapping != mapping)) {
1680 f2fs_put_page(page, 1);
1681 goto repeat;
1682 }
1683
1684 /* avoid symlink page */
1685 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1686 err = fscrypt_decrypt_page(page);
1687 if (err)
1688 goto fail;
1689 }
1690 }
1691 out_update:
1692 SetPageUptodate(page);
1693 out_clear:
1694 clear_cold_data(page);
1695 return 0;
1696
1697 fail:
1698 f2fs_put_page(page, 1);
1699 f2fs_write_failed(mapping, pos + len);
1700 return err;
1701 }
1702
1703 static int f2fs_write_end(struct file *file,
1704 struct address_space *mapping,
1705 loff_t pos, unsigned len, unsigned copied,
1706 struct page *page, void *fsdata)
1707 {
1708 struct inode *inode = page->mapping->host;
1709
1710 trace_f2fs_write_end(inode, pos, len, copied);
1711
1712 set_page_dirty(page);
1713
1714 if (pos + copied > i_size_read(inode))
1715 f2fs_i_size_write(inode, pos + copied);
1716
1717 f2fs_put_page(page, 1);
1718 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1719 return copied;
1720 }
1721
1722 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1723 loff_t offset)
1724 {
1725 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1726
1727 if (offset & blocksize_mask)
1728 return -EINVAL;
1729
1730 if (iov_iter_alignment(iter) & blocksize_mask)
1731 return -EINVAL;
1732
1733 return 0;
1734 }
1735
1736 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1737 {
1738 struct address_space *mapping = iocb->ki_filp->f_mapping;
1739 struct inode *inode = mapping->host;
1740 size_t count = iov_iter_count(iter);
1741 loff_t offset = iocb->ki_pos;
1742 int err;
1743
1744 err = check_direct_IO(inode, iter, offset);
1745 if (err)
1746 return err;
1747
1748 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1749 return 0;
1750
1751 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1752
1753 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1754 if (iov_iter_rw(iter) == WRITE) {
1755 if (err > 0)
1756 set_inode_flag(inode, FI_UPDATE_WRITE);
1757 else if (err < 0)
1758 f2fs_write_failed(mapping, offset + count);
1759 }
1760
1761 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1762
1763 return err;
1764 }
1765
1766 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1767 unsigned int length)
1768 {
1769 struct inode *inode = page->mapping->host;
1770 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1771
1772 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1773 (offset % PAGE_SIZE || length != PAGE_SIZE))
1774 return;
1775
1776 if (PageDirty(page)) {
1777 if (inode->i_ino == F2FS_META_INO(sbi))
1778 dec_page_count(sbi, F2FS_DIRTY_META);
1779 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1780 dec_page_count(sbi, F2FS_DIRTY_NODES);
1781 else
1782 inode_dec_dirty_pages(inode);
1783 }
1784
1785 /* This is atomic written page, keep Private */
1786 if (IS_ATOMIC_WRITTEN_PAGE(page))
1787 return;
1788
1789 set_page_private(page, 0);
1790 ClearPagePrivate(page);
1791 }
1792
1793 int f2fs_release_page(struct page *page, gfp_t wait)
1794 {
1795 /* If this is dirty page, keep PagePrivate */
1796 if (PageDirty(page))
1797 return 0;
1798
1799 /* This is atomic written page, keep Private */
1800 if (IS_ATOMIC_WRITTEN_PAGE(page))
1801 return 0;
1802
1803 set_page_private(page, 0);
1804 ClearPagePrivate(page);
1805 return 1;
1806 }
1807
1808 static int f2fs_set_data_page_dirty(struct page *page)
1809 {
1810 struct address_space *mapping = page->mapping;
1811 struct inode *inode = mapping->host;
1812
1813 trace_f2fs_set_page_dirty(page, DATA);
1814
1815 SetPageUptodate(page);
1816
1817 if (f2fs_is_atomic_file(inode)) {
1818 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1819 register_inmem_page(inode, page);
1820 return 1;
1821 }
1822 /*
1823 * Previously, this page has been registered, we just
1824 * return here.
1825 */
1826 return 0;
1827 }
1828
1829 if (!PageDirty(page)) {
1830 __set_page_dirty_nobuffers(page);
1831 update_dirty_page(inode, page);
1832 return 1;
1833 }
1834 return 0;
1835 }
1836
1837 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1838 {
1839 struct inode *inode = mapping->host;
1840
1841 if (f2fs_has_inline_data(inode))
1842 return 0;
1843
1844 /* make sure allocating whole blocks */
1845 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1846 filemap_write_and_wait(mapping);
1847
1848 return generic_block_bmap(mapping, block, get_data_block_bmap);
1849 }
1850
1851 const struct address_space_operations f2fs_dblock_aops = {
1852 .readpage = f2fs_read_data_page,
1853 .readpages = f2fs_read_data_pages,
1854 .writepage = f2fs_write_data_page,
1855 .writepages = f2fs_write_data_pages,
1856 .write_begin = f2fs_write_begin,
1857 .write_end = f2fs_write_end,
1858 .set_page_dirty = f2fs_set_data_page_dirty,
1859 .invalidatepage = f2fs_invalidate_page,
1860 .releasepage = f2fs_release_page,
1861 .direct_IO = f2fs_direct_IO,
1862 .bmap = f2fs_bmap,
1863 };
Linux kernel - Deliverables
This page took 0.077305 seconds and 5 git commands to generate.