projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: be aware of extent beyond EOF in fiemap
[GitHub/exynos8895/android_kernel_samsung_universal8895.git] / fs / f2fs / data.c
1 /*
2 * fs/f2fs/data.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
22 #include <linux/mm.h>
23 #include <linux/memcontrol.h>
24 #include <linux/cleancache.h>
25
26 #include "f2fs.h"
27 #include "node.h"
28 #include "segment.h"
29 #include "trace.h"
30 #include <trace/events/f2fs.h>
31 #include <trace/events/android_fs.h>
32
33 static void f2fs_read_end_io(struct bio *bio)
34 {
35 struct bio_vec *bvec;
36 int i;
37
38 #ifdef CONFIG_F2FS_FAULT_INJECTION
39 if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO))
40 bio->bi_error = -EIO;
41 #endif
42
43 if (f2fs_bio_encrypted(bio)) {
44 if (bio->bi_error) {
45 fscrypt_release_ctx(bio->bi_private);
46 } else {
47 fscrypt_decrypt_bio_pages(bio->bi_private, bio);
48 return;
49 }
50 }
51
52 bio_for_each_segment_all(bvec, bio, i) {
53 struct page *page = bvec->bv_page;
54
55 if (!bio->bi_error) {
56 if (!PageUptodate(page))
57 SetPageUptodate(page);
58 } else {
59 ClearPageUptodate(page);
60 SetPageError(page);
61 }
62 unlock_page(page);
63 }
64 bio_put(bio);
65 }
66
67 static void f2fs_write_end_io(struct bio *bio)
68 {
69 struct f2fs_sb_info *sbi = bio->bi_private;
70 struct bio_vec *bvec;
71 int i;
72
73 bio_for_each_segment_all(bvec, bio, i) {
74 struct page *page = bvec->bv_page;
75
76 fscrypt_pullback_bio_page(&page, true);
77
78 if (unlikely(bio->bi_error)) {
79 set_bit(AS_EIO, &page->mapping->flags);
80 f2fs_stop_checkpoint(sbi, true);
81 }
82 end_page_writeback(page);
83 }
84 if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
85 wq_has_sleeper(&sbi->cp_wait))
86 wake_up(&sbi->cp_wait);
87
88 bio_put(bio);
89 }
90
91 /*
92 * Low-level block read/write IO operations.
93 */
94 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
95 int npages, bool is_read)
96 {
97 struct bio *bio;
98
99 bio = f2fs_bio_alloc(npages);
100
101 bio->bi_bdev = sbi->sb->s_bdev;
102 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
103 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
104 bio->bi_private = is_read ? NULL : sbi;
105
106 return bio;
107 }
108
109 static inline void __submit_bio(struct f2fs_sb_info *sbi, int rw,
110 struct bio *bio, enum page_type type)
111 {
112 if (!is_read_io(rw)) {
113 atomic_inc(&sbi->nr_wb_bios);
114 if (f2fs_sb_mounted_hmsmr(sbi->sb) &&
115 current->plug && (type == DATA || type == NODE))
116 blk_finish_plug(current->plug);
117 }
118 submit_bio(rw, bio);
119 }
120
121 static void __submit_merged_bio(struct f2fs_bio_info *io)
122 {
123 struct f2fs_io_info *fio = &io->fio;
124
125 if (!io->bio)
126 return;
127
128 if (is_read_io(fio->rw))
129 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
130 else
131 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
132
133 __submit_bio(io->sbi, fio->rw, io->bio, fio->type);
134 io->bio = NULL;
135 }
136
137 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
138 struct page *page, nid_t ino)
139 {
140 struct bio_vec *bvec;
141 struct page *target;
142 int i;
143
144 if (!io->bio)
145 return false;
146
147 if (!inode && !page && !ino)
148 return true;
149
150 bio_for_each_segment_all(bvec, io->bio, i) {
151
152 if (bvec->bv_page->mapping)
153 target = bvec->bv_page;
154 else
155 target = fscrypt_control_page(bvec->bv_page);
156
157 if (inode && inode == target->mapping->host)
158 return true;
159 if (page && page == target)
160 return true;
161 if (ino && ino == ino_of_node(target))
162 return true;
163 }
164
165 return false;
166 }
167
168 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
169 struct page *page, nid_t ino,
170 enum page_type type)
171 {
172 enum page_type btype = PAGE_TYPE_OF_BIO(type);
173 struct f2fs_bio_info *io = &sbi->write_io[btype];
174 bool ret;
175
176 down_read(&io->io_rwsem);
177 ret = __has_merged_page(io, inode, page, ino);
178 up_read(&io->io_rwsem);
179 return ret;
180 }
181
182 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
183 struct inode *inode, struct page *page,
184 nid_t ino, enum page_type type, int rw)
185 {
186 enum page_type btype = PAGE_TYPE_OF_BIO(type);
187 struct f2fs_bio_info *io;
188
189 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
190
191 down_write(&io->io_rwsem);
192
193 if (!__has_merged_page(io, inode, page, ino))
194 goto out;
195
196 /* change META to META_FLUSH in the checkpoint procedure */
197 if (type >= META_FLUSH) {
198 io->fio.type = META_FLUSH;
199 if (test_opt(sbi, NOBARRIER))
200 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
201 else
202 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
203 }
204 __submit_merged_bio(io);
205 out:
206 up_write(&io->io_rwsem);
207 }
208
209 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
210 int rw)
211 {
212 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
213 }
214
215 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
216 struct inode *inode, struct page *page,
217 nid_t ino, enum page_type type, int rw)
218 {
219 if (has_merged_page(sbi, inode, page, ino, type))
220 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
221 }
222
223 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
224 {
225 f2fs_submit_merged_bio(sbi, DATA, WRITE);
226 f2fs_submit_merged_bio(sbi, NODE, WRITE);
227 f2fs_submit_merged_bio(sbi, META, WRITE);
228 }
229
230 /*
231 * Fill the locked page with data located in the block address.
232 * Return unlocked page.
233 */
234 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
235 {
236 struct bio *bio;
237 struct page *page = fio->encrypted_page ?
238 fio->encrypted_page : fio->page;
239
240 trace_f2fs_submit_page_bio(page, fio);
241 f2fs_trace_ios(fio, 0);
242
243 /* Allocate a new bio */
244 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
245
246 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
247 bio_put(bio);
248 return -EFAULT;
249 }
250
251 __submit_bio(fio->sbi, fio->rw, bio, fio->type);
252 return 0;
253 }
254
255 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
256 {
257 struct f2fs_sb_info *sbi = fio->sbi;
258 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
259 struct f2fs_bio_info *io;
260 bool is_read = is_read_io(fio->rw);
261 struct page *bio_page;
262
263 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
264
265 if (fio->old_blkaddr != NEW_ADDR)
266 verify_block_addr(sbi, fio->old_blkaddr);
267 verify_block_addr(sbi, fio->new_blkaddr);
268
269 down_write(&io->io_rwsem);
270
271 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
272 io->fio.rw != fio->rw))
273 __submit_merged_bio(io);
274 alloc_new:
275 if (io->bio == NULL) {
276 int bio_blocks = MAX_BIO_BLOCKS(sbi);
277
278 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
279 bio_blocks, is_read);
280 io->fio = *fio;
281 }
282
283 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
284
285 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
286 PAGE_SIZE) {
287 __submit_merged_bio(io);
288 goto alloc_new;
289 }
290
291 io->last_block_in_bio = fio->new_blkaddr;
292 f2fs_trace_ios(fio, 0);
293
294 up_write(&io->io_rwsem);
295 trace_f2fs_submit_page_mbio(fio->page, fio);
296 }
297
298 static void __set_data_blkaddr(struct dnode_of_data *dn)
299 {
300 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
301 __le32 *addr_array;
302
303 /* Get physical address of data block */
304 addr_array = blkaddr_in_node(rn);
305 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
306 }
307
308 /*
309 * Lock ordering for the change of data block address:
310 * ->data_page
311 * ->node_page
312 * update block addresses in the node page
313 */
314 void set_data_blkaddr(struct dnode_of_data *dn)
315 {
316 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
317 __set_data_blkaddr(dn);
318 if (set_page_dirty(dn->node_page))
319 dn->node_changed = true;
320 }
321
322 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
323 {
324 dn->data_blkaddr = blkaddr;
325 set_data_blkaddr(dn);
326 f2fs_update_extent_cache(dn);
327 }
328
329 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
330 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
331 {
332 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
333
334 if (!count)
335 return 0;
336
337 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
338 return -EPERM;
339 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
340 return -ENOSPC;
341
342 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
343 dn->ofs_in_node, count);
344
345 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
346
347 for (; count > 0; dn->ofs_in_node++) {
348 block_t blkaddr =
349 datablock_addr(dn->node_page, dn->ofs_in_node);
350 if (blkaddr == NULL_ADDR) {
351 dn->data_blkaddr = NEW_ADDR;
352 __set_data_blkaddr(dn);
353 count--;
354 }
355 }
356
357 if (set_page_dirty(dn->node_page))
358 dn->node_changed = true;
359 return 0;
360 }
361
362 /* Should keep dn->ofs_in_node unchanged */
363 int reserve_new_block(struct dnode_of_data *dn)
364 {
365 unsigned int ofs_in_node = dn->ofs_in_node;
366 int ret;
367
368 ret = reserve_new_blocks(dn, 1);
369 dn->ofs_in_node = ofs_in_node;
370 return ret;
371 }
372
373 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
374 {
375 bool need_put = dn->inode_page ? false : true;
376 int err;
377
378 err = get_dnode_of_data(dn, index, ALLOC_NODE);
379 if (err)
380 return err;
381
382 if (dn->data_blkaddr == NULL_ADDR)
383 err = reserve_new_block(dn);
384 if (err || need_put)
385 f2fs_put_dnode(dn);
386 return err;
387 }
388
389 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
390 {
391 struct extent_info ei;
392 struct inode *inode = dn->inode;
393
394 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
395 dn->data_blkaddr = ei.blk + index - ei.fofs;
396 return 0;
397 }
398
399 return f2fs_reserve_block(dn, index);
400 }
401
402 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
403 int rw, bool for_write)
404 {
405 struct address_space *mapping = inode->i_mapping;
406 struct dnode_of_data dn;
407 struct page *page;
408 struct extent_info ei;
409 int err;
410 struct f2fs_io_info fio = {
411 .sbi = F2FS_I_SB(inode),
412 .type = DATA,
413 .rw = rw,
414 .encrypted_page = NULL,
415 };
416
417 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
418 return read_mapping_page(mapping, index, NULL);
419
420 page = f2fs_grab_cache_page(mapping, index, for_write);
421 if (!page)
422 return ERR_PTR(-ENOMEM);
423
424 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
425 dn.data_blkaddr = ei.blk + index - ei.fofs;
426 goto got_it;
427 }
428
429 set_new_dnode(&dn, inode, NULL, NULL, 0);
430 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
431 if (err)
432 goto put_err;
433 f2fs_put_dnode(&dn);
434
435 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
436 err = -ENOENT;
437 goto put_err;
438 }
439 got_it:
440 if (PageUptodate(page)) {
441 unlock_page(page);
442 return page;
443 }
444
445 /*
446 * A new dentry page is allocated but not able to be written, since its
447 * new inode page couldn't be allocated due to -ENOSPC.
448 * In such the case, its blkaddr can be remained as NEW_ADDR.
449 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
450 */
451 if (dn.data_blkaddr == NEW_ADDR) {
452 zero_user_segment(page, 0, PAGE_SIZE);
453 if (!PageUptodate(page))
454 SetPageUptodate(page);
455 unlock_page(page);
456 return page;
457 }
458
459 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
460 fio.page = page;
461 err = f2fs_submit_page_bio(&fio);
462 if (err)
463 goto put_err;
464 return page;
465
466 put_err:
467 f2fs_put_page(page, 1);
468 return ERR_PTR(err);
469 }
470
471 struct page *find_data_page(struct inode *inode, pgoff_t index)
472 {
473 struct address_space *mapping = inode->i_mapping;
474 struct page *page;
475
476 page = find_get_page(mapping, index);
477 if (page && PageUptodate(page))
478 return page;
479 f2fs_put_page(page, 0);
480
481 page = get_read_data_page(inode, index, READ_SYNC, false);
482 if (IS_ERR(page))
483 return page;
484
485 if (PageUptodate(page))
486 return page;
487
488 wait_on_page_locked(page);
489 if (unlikely(!PageUptodate(page))) {
490 f2fs_put_page(page, 0);
491 return ERR_PTR(-EIO);
492 }
493 return page;
494 }
495
496 /*
497 * If it tries to access a hole, return an error.
498 * Because, the callers, functions in dir.c and GC, should be able to know
499 * whether this page exists or not.
500 */
501 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
502 bool for_write)
503 {
504 struct address_space *mapping = inode->i_mapping;
505 struct page *page;
506 repeat:
507 page = get_read_data_page(inode, index, READ_SYNC, for_write);
508 if (IS_ERR(page))
509 return page;
510
511 /* wait for read completion */
512 lock_page(page);
513 if (unlikely(page->mapping != mapping)) {
514 f2fs_put_page(page, 1);
515 goto repeat;
516 }
517 if (unlikely(!PageUptodate(page))) {
518 f2fs_put_page(page, 1);
519 return ERR_PTR(-EIO);
520 }
521 return page;
522 }
523
524 /*
525 * Caller ensures that this data page is never allocated.
526 * A new zero-filled data page is allocated in the page cache.
527 *
528 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
529 * f2fs_unlock_op().
530 * Note that, ipage is set only by make_empty_dir, and if any error occur,
531 * ipage should be released by this function.
532 */
533 struct page *get_new_data_page(struct inode *inode,
534 struct page *ipage, pgoff_t index, bool new_i_size)
535 {
536 struct address_space *mapping = inode->i_mapping;
537 struct page *page;
538 struct dnode_of_data dn;
539 int err;
540
541 page = f2fs_grab_cache_page(mapping, index, true);
542 if (!page) {
543 /*
544 * before exiting, we should make sure ipage will be released
545 * if any error occur.
546 */
547 f2fs_put_page(ipage, 1);
548 return ERR_PTR(-ENOMEM);
549 }
550
551 set_new_dnode(&dn, inode, ipage, NULL, 0);
552 err = f2fs_reserve_block(&dn, index);
553 if (err) {
554 f2fs_put_page(page, 1);
555 return ERR_PTR(err);
556 }
557 if (!ipage)
558 f2fs_put_dnode(&dn);
559
560 if (PageUptodate(page))
561 goto got_it;
562
563 if (dn.data_blkaddr == NEW_ADDR) {
564 zero_user_segment(page, 0, PAGE_SIZE);
565 if (!PageUptodate(page))
566 SetPageUptodate(page);
567 } else {
568 f2fs_put_page(page, 1);
569
570 /* if ipage exists, blkaddr should be NEW_ADDR */
571 f2fs_bug_on(F2FS_I_SB(inode), ipage);
572 page = get_lock_data_page(inode, index, true);
573 if (IS_ERR(page))
574 return page;
575 }
576 got_it:
577 if (new_i_size && i_size_read(inode) <
578 ((loff_t)(index + 1) << PAGE_SHIFT))
579 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
580 return page;
581 }
582
583 static int __allocate_data_block(struct dnode_of_data *dn)
584 {
585 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
586 struct f2fs_summary sum;
587 struct node_info ni;
588 int seg = CURSEG_WARM_DATA;
589 pgoff_t fofs;
590 blkcnt_t count = 1;
591
592 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
593 return -EPERM;
594
595 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
596 if (dn->data_blkaddr == NEW_ADDR)
597 goto alloc;
598
599 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
600 return -ENOSPC;
601
602 alloc:
603 get_node_info(sbi, dn->nid, &ni);
604 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
605
606 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
607 seg = CURSEG_DIRECT_IO;
608
609 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
610 &sum, seg);
611 set_data_blkaddr(dn);
612
613 /* update i_size */
614 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
615 dn->ofs_in_node;
616 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
617 f2fs_i_size_write(dn->inode,
618 ((loff_t)(fofs + 1) << PAGE_SHIFT));
619 return 0;
620 }
621
622 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
623 {
624 struct inode *inode = file_inode(iocb->ki_filp);
625 struct f2fs_map_blocks map;
626 ssize_t ret = 0;
627
628 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
629 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
630 if (map.m_len > map.m_lblk)
631 map.m_len -= map.m_lblk;
632 else
633 map.m_len = 0;
634
635 map.m_next_pgofs = NULL;
636
637 if (iocb->ki_flags & IOCB_DIRECT) {
638 ret = f2fs_convert_inline_inode(inode);
639 if (ret)
640 return ret;
641 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
642 }
643 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
644 ret = f2fs_convert_inline_inode(inode);
645 if (ret)
646 return ret;
647 }
648 if (!f2fs_has_inline_data(inode))
649 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
650 return ret;
651 }
652
653 /*
654 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
655 * f2fs_map_blocks structure.
656 * If original data blocks are allocated, then give them to blockdev.
657 * Otherwise,
658 * a. preallocate requested block addresses
659 * b. do not use extent cache for better performance
660 * c. give the block addresses to blockdev
661 */
662 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
663 int create, int flag)
664 {
665 unsigned int maxblocks = map->m_len;
666 struct dnode_of_data dn;
667 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
668 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
669 pgoff_t pgofs, end_offset, end;
670 int err = 0, ofs = 1;
671 unsigned int ofs_in_node, last_ofs_in_node;
672 blkcnt_t prealloc;
673 struct extent_info ei;
674 block_t blkaddr;
675
676 if (!maxblocks)
677 return 0;
678
679 map->m_len = 0;
680 map->m_flags = 0;
681
682 /* it only supports block size == page size */
683 pgofs = (pgoff_t)map->m_lblk;
684 end = pgofs + maxblocks;
685
686 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
687 map->m_pblk = ei.blk + pgofs - ei.fofs;
688 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
689 map->m_flags = F2FS_MAP_MAPPED;
690 goto out;
691 }
692
693 next_dnode:
694 if (create)
695 f2fs_lock_op(sbi);
696
697 /* When reading holes, we need its node page */
698 set_new_dnode(&dn, inode, NULL, NULL, 0);
699 err = get_dnode_of_data(&dn, pgofs, mode);
700 if (err) {
701 if (flag == F2FS_GET_BLOCK_BMAP)
702 map->m_pblk = 0;
703 if (err == -ENOENT) {
704 err = 0;
705 if (map->m_next_pgofs)
706 *map->m_next_pgofs =
707 get_next_page_offset(&dn, pgofs);
708 }
709 goto unlock_out;
710 }
711
712 prealloc = 0;
713 ofs_in_node = dn.ofs_in_node;
714 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
715
716 next_block:
717 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
718
719 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
720 if (create) {
721 if (unlikely(f2fs_cp_error(sbi))) {
722 err = -EIO;
723 goto sync_out;
724 }
725 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
726 if (blkaddr == NULL_ADDR) {
727 prealloc++;
728 last_ofs_in_node = dn.ofs_in_node;
729 }
730 } else {
731 err = __allocate_data_block(&dn);
732 if (!err)
733 set_inode_flag(inode, FI_APPEND_WRITE);
734 }
735 if (err)
736 goto sync_out;
737 map->m_flags = F2FS_MAP_NEW;
738 blkaddr = dn.data_blkaddr;
739 } else {
740 if (flag == F2FS_GET_BLOCK_BMAP) {
741 map->m_pblk = 0;
742 goto sync_out;
743 }
744 if (flag == F2FS_GET_BLOCK_FIEMAP &&
745 blkaddr == NULL_ADDR) {
746 if (map->m_next_pgofs)
747 *map->m_next_pgofs = pgofs + 1;
748 }
749 if (flag != F2FS_GET_BLOCK_FIEMAP ||
750 blkaddr != NEW_ADDR)
751 goto sync_out;
752 }
753 }
754
755 if (flag == F2FS_GET_BLOCK_PRE_AIO)
756 goto skip;
757
758 if (map->m_len == 0) {
759 /* preallocated unwritten block should be mapped for fiemap. */
760 if (blkaddr == NEW_ADDR)
761 map->m_flags |= F2FS_MAP_UNWRITTEN;
762 map->m_flags |= F2FS_MAP_MAPPED;
763
764 map->m_pblk = blkaddr;
765 map->m_len = 1;
766 } else if ((map->m_pblk != NEW_ADDR &&
767 blkaddr == (map->m_pblk + ofs)) ||
768 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
769 flag == F2FS_GET_BLOCK_PRE_DIO) {
770 ofs++;
771 map->m_len++;
772 } else {
773 goto sync_out;
774 }
775
776 skip:
777 dn.ofs_in_node++;
778 pgofs++;
779
780 /* preallocate blocks in batch for one dnode page */
781 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
782 (pgofs == end || dn.ofs_in_node == end_offset)) {
783
784 dn.ofs_in_node = ofs_in_node;
785 err = reserve_new_blocks(&dn, prealloc);
786 if (err)
787 goto sync_out;
788
789 map->m_len += dn.ofs_in_node - ofs_in_node;
790 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
791 err = -ENOSPC;
792 goto sync_out;
793 }
794 dn.ofs_in_node = end_offset;
795 }
796
797 if (pgofs >= end)
798 goto sync_out;
799 else if (dn.ofs_in_node < end_offset)
800 goto next_block;
801
802 f2fs_put_dnode(&dn);
803
804 if (create) {
805 f2fs_unlock_op(sbi);
806 f2fs_balance_fs(sbi, dn.node_changed);
807 }
808 goto next_dnode;
809
810 sync_out:
811 f2fs_put_dnode(&dn);
812 unlock_out:
813 if (create) {
814 f2fs_unlock_op(sbi);
815 f2fs_balance_fs(sbi, dn.node_changed);
816 }
817 out:
818 trace_f2fs_map_blocks(inode, map, err);
819 return err;
820 }
821
822 static int __get_data_block(struct inode *inode, sector_t iblock,
823 struct buffer_head *bh, int create, int flag,
824 pgoff_t *next_pgofs)
825 {
826 struct f2fs_map_blocks map;
827 int ret;
828
829 map.m_lblk = iblock;
830 map.m_len = bh->b_size >> inode->i_blkbits;
831 map.m_next_pgofs = next_pgofs;
832
833 ret = f2fs_map_blocks(inode, &map, create, flag);
834 if (!ret) {
835 map_bh(bh, inode->i_sb, map.m_pblk);
836 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
837 bh->b_size = map.m_len << inode->i_blkbits;
838 }
839 return ret;
840 }
841
842 static int get_data_block(struct inode *inode, sector_t iblock,
843 struct buffer_head *bh_result, int create, int flag,
844 pgoff_t *next_pgofs)
845 {
846 return __get_data_block(inode, iblock, bh_result, create,
847 flag, next_pgofs);
848 }
849
850 static int get_data_block_dio(struct inode *inode, sector_t iblock,
851 struct buffer_head *bh_result, int create)
852 {
853 return __get_data_block(inode, iblock, bh_result, create,
854 F2FS_GET_BLOCK_DIO, NULL);
855 }
856
857 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
858 struct buffer_head *bh_result, int create)
859 {
860 /* Block number less than F2FS MAX BLOCKS */
861 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
862 return -EFBIG;
863
864 return __get_data_block(inode, iblock, bh_result, create,
865 F2FS_GET_BLOCK_BMAP, NULL);
866 }
867
868 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
869 {
870 return (offset >> inode->i_blkbits);
871 }
872
873 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
874 {
875 return (blk << inode->i_blkbits);
876 }
877
878 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
879 u64 start, u64 len)
880 {
881 struct buffer_head map_bh;
882 sector_t start_blk, last_blk;
883 pgoff_t next_pgofs;
884 u64 logical = 0, phys = 0, size = 0;
885 u32 flags = 0;
886 int ret = 0;
887
888 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
889 if (ret)
890 return ret;
891
892 if (f2fs_has_inline_data(inode)) {
893 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
894 if (ret != -EAGAIN)
895 return ret;
896 }
897
898 inode_lock(inode);
899
900 if (logical_to_blk(inode, len) == 0)
901 len = blk_to_logical(inode, 1);
902
903 start_blk = logical_to_blk(inode, start);
904 last_blk = logical_to_blk(inode, start + len - 1);
905
906 next:
907 memset(&map_bh, 0, sizeof(struct buffer_head));
908 map_bh.b_size = len;
909
910 ret = get_data_block(inode, start_blk, &map_bh, 0,
911 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
912 if (ret)
913 goto out;
914
915 /* HOLE */
916 if (!buffer_mapped(&map_bh)) {
917 start_blk = next_pgofs;
918
919 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
920 F2FS_I_SB(inode)->max_file_blocks))
921 goto prep_next;
922
923 flags |= FIEMAP_EXTENT_LAST;
924 }
925
926 if (size) {
927 if (f2fs_encrypted_inode(inode))
928 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
929
930 ret = fiemap_fill_next_extent(fieinfo, logical,
931 phys, size, flags);
932 }
933
934 if (start_blk > last_blk || ret)
935 goto out;
936
937 logical = blk_to_logical(inode, start_blk);
938 phys = blk_to_logical(inode, map_bh.b_blocknr);
939 size = map_bh.b_size;
940 flags = 0;
941 if (buffer_unwritten(&map_bh))
942 flags = FIEMAP_EXTENT_UNWRITTEN;
943
944 start_blk += logical_to_blk(inode, size);
945
946 prep_next:
947 cond_resched();
948 if (fatal_signal_pending(current))
949 ret = -EINTR;
950 else
951 goto next;
952 out:
953 if (ret == 1)
954 ret = 0;
955
956 inode_unlock(inode);
957 return ret;
958 }
959
960 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
961 unsigned nr_pages)
962 {
963 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
964 struct fscrypt_ctx *ctx = NULL;
965 struct block_device *bdev = sbi->sb->s_bdev;
966 struct bio *bio;
967
968 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
969 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
970 if (IS_ERR(ctx))
971 return ERR_CAST(ctx);
972
973 /* wait the page to be moved by cleaning */
974 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
975 }
976
977 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
978 if (!bio) {
979 if (ctx)
980 fscrypt_release_ctx(ctx);
981 return ERR_PTR(-ENOMEM);
982 }
983 bio->bi_bdev = bdev;
984 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
985 bio->bi_end_io = f2fs_read_end_io;
986 bio->bi_private = ctx;
987
988 return bio;
989 }
990
991 /*
992 * This function was originally taken from fs/mpage.c, and customized for f2fs.
993 * Major change was from block_size == page_size in f2fs by default.
994 */
995 static int f2fs_mpage_readpages(struct address_space *mapping,
996 struct list_head *pages, struct page *page,
997 unsigned nr_pages)
998 {
999 struct bio *bio = NULL;
1000 unsigned page_idx;
1001 sector_t last_block_in_bio = 0;
1002 struct inode *inode = mapping->host;
1003 const unsigned blkbits = inode->i_blkbits;
1004 const unsigned blocksize = 1 << blkbits;
1005 sector_t block_in_file;
1006 sector_t last_block;
1007 sector_t last_block_in_file;
1008 sector_t block_nr;
1009 struct f2fs_map_blocks map;
1010
1011 map.m_pblk = 0;
1012 map.m_lblk = 0;
1013 map.m_len = 0;
1014 map.m_flags = 0;
1015 map.m_next_pgofs = NULL;
1016
1017 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1018
1019 prefetchw(&page->flags);
1020 if (pages) {
1021 page = list_entry(pages->prev, struct page, lru);
1022 list_del(&page->lru);
1023 if (add_to_page_cache_lru(page, mapping,
1024 page->index, GFP_KERNEL))
1025 goto next_page;
1026 }
1027
1028 block_in_file = (sector_t)page->index;
1029 last_block = block_in_file + nr_pages;
1030 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1031 blkbits;
1032 if (last_block > last_block_in_file)
1033 last_block = last_block_in_file;
1034
1035 /*
1036 * Map blocks using the previous result first.
1037 */
1038 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1039 block_in_file > map.m_lblk &&
1040 block_in_file < (map.m_lblk + map.m_len))
1041 goto got_it;
1042
1043 /*
1044 * Then do more f2fs_map_blocks() calls until we are
1045 * done with this page.
1046 */
1047 map.m_flags = 0;
1048
1049 if (block_in_file < last_block) {
1050 map.m_lblk = block_in_file;
1051 map.m_len = last_block - block_in_file;
1052
1053 if (f2fs_map_blocks(inode, &map, 0,
1054 F2FS_GET_BLOCK_READ))
1055 goto set_error_page;
1056 }
1057 got_it:
1058 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1059 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1060 SetPageMappedToDisk(page);
1061
1062 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1063 SetPageUptodate(page);
1064 goto confused;
1065 }
1066 } else {
1067 zero_user_segment(page, 0, PAGE_SIZE);
1068 if (!PageUptodate(page))
1069 SetPageUptodate(page);
1070 unlock_page(page);
1071 goto next_page;
1072 }
1073
1074 /*
1075 * This page will go to BIO. Do we need to send this
1076 * BIO off first?
1077 */
1078 if (bio && (last_block_in_bio != block_nr - 1)) {
1079 submit_and_realloc:
1080 __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1081 bio = NULL;
1082 }
1083 if (bio == NULL) {
1084 bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1085 if (IS_ERR(bio)) {
1086 bio = NULL;
1087 goto set_error_page;
1088 }
1089 }
1090
1091 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1092 goto submit_and_realloc;
1093
1094 last_block_in_bio = block_nr;
1095 goto next_page;
1096 set_error_page:
1097 SetPageError(page);
1098 zero_user_segment(page, 0, PAGE_SIZE);
1099 unlock_page(page);
1100 goto next_page;
1101 confused:
1102 if (bio) {
1103 __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1104 bio = NULL;
1105 }
1106 unlock_page(page);
1107 next_page:
1108 if (pages)
1109 put_page(page);
1110 }
1111 BUG_ON(pages && !list_empty(pages));
1112 if (bio)
1113 __submit_bio(F2FS_I_SB(inode), READ, bio, DATA);
1114 return 0;
1115 }
1116
1117 static int f2fs_read_data_page(struct file *file, struct page *page)
1118 {
1119 struct inode *inode = page->mapping->host;
1120 int ret = -EAGAIN;
1121
1122 trace_f2fs_readpage(page, DATA);
1123
1124 /* If the file has inline data, try to read it directly */
1125 if (f2fs_has_inline_data(inode))
1126 ret = f2fs_read_inline_data(inode, page);
1127 if (ret == -EAGAIN)
1128 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1129 return ret;
1130 }
1131
1132 static int f2fs_read_data_pages(struct file *file,
1133 struct address_space *mapping,
1134 struct list_head *pages, unsigned nr_pages)
1135 {
1136 struct inode *inode = file->f_mapping->host;
1137 struct page *page = list_entry(pages->prev, struct page, lru);
1138
1139 trace_f2fs_readpages(inode, page, nr_pages);
1140
1141 /* If the file has inline data, skip readpages */
1142 if (f2fs_has_inline_data(inode))
1143 return 0;
1144
1145 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1146 }
1147
1148 int do_write_data_page(struct f2fs_io_info *fio)
1149 {
1150 struct page *page = fio->page;
1151 struct inode *inode = page->mapping->host;
1152 struct dnode_of_data dn;
1153 int err = 0;
1154
1155 set_new_dnode(&dn, inode, NULL, NULL, 0);
1156 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1157 if (err)
1158 return err;
1159
1160 fio->old_blkaddr = dn.data_blkaddr;
1161
1162 /* This page is already truncated */
1163 if (fio->old_blkaddr == NULL_ADDR) {
1164 ClearPageUptodate(page);
1165 goto out_writepage;
1166 }
1167
1168 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1169 gfp_t gfp_flags = GFP_NOFS;
1170
1171 /* wait for GCed encrypted page writeback */
1172 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1173 fio->old_blkaddr);
1174 retry_encrypt:
1175 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1176 gfp_flags);
1177 if (IS_ERR(fio->encrypted_page)) {
1178 err = PTR_ERR(fio->encrypted_page);
1179 if (err == -ENOMEM) {
1180 /* flush pending ios and wait for a while */
1181 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1182 congestion_wait(BLK_RW_ASYNC, HZ/50);
1183 gfp_flags |= __GFP_NOFAIL;
1184 err = 0;
1185 goto retry_encrypt;
1186 }
1187 goto out_writepage;
1188 }
1189 }
1190
1191 set_page_writeback(page);
1192
1193 /*
1194 * If current allocation needs SSR,
1195 * it had better in-place writes for updated data.
1196 */
1197 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1198 !is_cold_data(page) &&
1199 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1200 need_inplace_update(inode))) {
1201 rewrite_data_page(fio);
1202 set_inode_flag(inode, FI_UPDATE_WRITE);
1203 trace_f2fs_do_write_data_page(page, IPU);
1204 } else {
1205 write_data_page(&dn, fio);
1206 trace_f2fs_do_write_data_page(page, OPU);
1207 set_inode_flag(inode, FI_APPEND_WRITE);
1208 if (page->index == 0)
1209 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1210 }
1211 out_writepage:
1212 f2fs_put_dnode(&dn);
1213 return err;
1214 }
1215
1216 static int f2fs_write_data_page(struct page *page,
1217 struct writeback_control *wbc)
1218 {
1219 struct inode *inode = page->mapping->host;
1220 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1221 loff_t i_size = i_size_read(inode);
1222 const pgoff_t end_index = ((unsigned long long) i_size)
1223 >> PAGE_SHIFT;
1224 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1225 unsigned offset = 0;
1226 bool need_balance_fs = false;
1227 int err = 0;
1228 struct f2fs_io_info fio = {
1229 .sbi = sbi,
1230 .type = DATA,
1231 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1232 .page = page,
1233 .encrypted_page = NULL,
1234 };
1235
1236 trace_f2fs_writepage(page, DATA);
1237
1238 if (page->index < end_index)
1239 goto write;
1240
1241 /*
1242 * If the offset is out-of-range of file size,
1243 * this page does not have to be written to disk.
1244 */
1245 offset = i_size & (PAGE_SIZE - 1);
1246 if ((page->index >= end_index + 1) || !offset)
1247 goto out;
1248
1249 zero_user_segment(page, offset, PAGE_SIZE);
1250 write:
1251 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1252 goto redirty_out;
1253 if (f2fs_is_drop_cache(inode))
1254 goto out;
1255 /* we should not write 0'th page having journal header */
1256 if (f2fs_is_volatile_file(inode) && (!page->index ||
1257 (!wbc->for_reclaim &&
1258 available_free_memory(sbi, BASE_CHECK))))
1259 goto redirty_out;
1260
1261 /* we should bypass data pages to proceed the kworkder jobs */
1262 if (unlikely(f2fs_cp_error(sbi))) {
1263 mapping_set_error(page->mapping, -EIO);
1264 goto out;
1265 }
1266
1267 /* Dentry blocks are controlled by checkpoint */
1268 if (S_ISDIR(inode->i_mode)) {
1269 err = do_write_data_page(&fio);
1270 goto done;
1271 }
1272
1273 if (!wbc->for_reclaim)
1274 need_balance_fs = true;
1275 else if (has_not_enough_free_secs(sbi, 0, 0))
1276 goto redirty_out;
1277
1278 err = -EAGAIN;
1279 f2fs_lock_op(sbi);
1280 if (f2fs_has_inline_data(inode))
1281 err = f2fs_write_inline_data(inode, page);
1282 if (err == -EAGAIN)
1283 err = do_write_data_page(&fio);
1284 if (F2FS_I(inode)->last_disk_size < psize)
1285 F2FS_I(inode)->last_disk_size = psize;
1286 f2fs_unlock_op(sbi);
1287 done:
1288 if (err && err != -ENOENT)
1289 goto redirty_out;
1290
1291 clear_cold_data(page);
1292 out:
1293 inode_dec_dirty_pages(inode);
1294 if (err)
1295 ClearPageUptodate(page);
1296
1297 if (wbc->for_reclaim) {
1298 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1299 remove_dirty_inode(inode);
1300 }
1301
1302 unlock_page(page);
1303 f2fs_balance_fs(sbi, need_balance_fs);
1304
1305 if (unlikely(f2fs_cp_error(sbi)))
1306 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1307
1308 return 0;
1309
1310 redirty_out:
1311 redirty_page_for_writepage(wbc, page);
1312 unlock_page(page);
1313 return err;
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)
1323 {
1324 int ret = 0;
1325 int done = 0;
1326 struct pagevec pvec;
1327 int nr_pages;
1328 pgoff_t uninitialized_var(writeback_index);
1329 pgoff_t index;
1330 pgoff_t end; /* Inclusive */
1331 pgoff_t done_index;
1332 int cycled;
1333 int range_whole = 0;
1334 int tag;
1335 int nwritten = 0;
1336
1337 pagevec_init(&pvec, 0);
1338
1339 if (wbc->range_cyclic) {
1340 writeback_index = mapping->writeback_index; /* prev offset */
1341 index = writeback_index;
1342 if (index == 0)
1343 cycled = 1;
1344 else
1345 cycled = 0;
1346 end = -1;
1347 } else {
1348 index = wbc->range_start >> PAGE_SHIFT;
1349 end = wbc->range_end >> PAGE_SHIFT;
1350 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1351 range_whole = 1;
1352 cycled = 1; /* ignore range_cyclic tests */
1353 }
1354 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1355 tag = PAGECACHE_TAG_TOWRITE;
1356 else
1357 tag = PAGECACHE_TAG_DIRTY;
1358 retry:
1359 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1360 tag_pages_for_writeback(mapping, index, end);
1361 done_index = index;
1362 while (!done && (index <= end)) {
1363 int i;
1364
1365 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1366 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1367 if (nr_pages == 0)
1368 break;
1369
1370 for (i = 0; i < nr_pages; i++) {
1371 struct page *page = pvec.pages[i];
1372
1373 if (page->index > end) {
1374 done = 1;
1375 break;
1376 }
1377
1378 done_index = page->index;
1379
1380 lock_page(page);
1381
1382 if (unlikely(page->mapping != mapping)) {
1383 continue_unlock:
1384 unlock_page(page);
1385 continue;
1386 }
1387
1388 if (!PageDirty(page)) {
1389 /* someone wrote it for us */
1390 goto continue_unlock;
1391 }
1392
1393 if (PageWriteback(page)) {
1394 if (wbc->sync_mode != WB_SYNC_NONE)
1395 f2fs_wait_on_page_writeback(page,
1396 DATA, true);
1397 else
1398 goto continue_unlock;
1399 }
1400
1401 BUG_ON(PageWriteback(page));
1402 if (!clear_page_dirty_for_io(page))
1403 goto continue_unlock;
1404
1405 ret = mapping->a_ops->writepage(page, wbc);
1406 if (unlikely(ret)) {
1407 done_index = page->index + 1;
1408 done = 1;
1409 break;
1410 } else {
1411 nwritten++;
1412 }
1413
1414 if (--wbc->nr_to_write <= 0 &&
1415 wbc->sync_mode == WB_SYNC_NONE) {
1416 done = 1;
1417 break;
1418 }
1419 }
1420 pagevec_release(&pvec);
1421 cond_resched();
1422 }
1423
1424 if (!cycled && !done) {
1425 cycled = 1;
1426 index = 0;
1427 end = writeback_index - 1;
1428 goto retry;
1429 }
1430 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1431 mapping->writeback_index = done_index;
1432
1433 if (nwritten)
1434 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping), mapping->host,
1435 NULL, 0, DATA, WRITE);
1436
1437 return ret;
1438 }
1439
1440 static int f2fs_write_data_pages(struct address_space *mapping,
1441 struct writeback_control *wbc)
1442 {
1443 struct inode *inode = mapping->host;
1444 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1445 struct blk_plug plug;
1446 int ret;
1447
1448 /* deal with chardevs and other special file */
1449 if (!mapping->a_ops->writepage)
1450 return 0;
1451
1452 /* skip writing if there is no dirty page in this inode */
1453 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1454 return 0;
1455
1456 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1457 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1458 available_free_memory(sbi, DIRTY_DENTS))
1459 goto skip_write;
1460
1461 /* skip writing during file defragment */
1462 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1463 goto skip_write;
1464
1465 /* during POR, we don't need to trigger writepage at all. */
1466 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1467 goto skip_write;
1468
1469 trace_f2fs_writepages(mapping->host, wbc, DATA);
1470
1471 blk_start_plug(&plug);
1472 ret = f2fs_write_cache_pages(mapping, wbc);
1473 blk_finish_plug(&plug);
1474 /*
1475 * if some pages were truncated, we cannot guarantee its mapping->host
1476 * to detect pending bios.
1477 */
1478
1479 remove_dirty_inode(inode);
1480 return ret;
1481
1482 skip_write:
1483 wbc->pages_skipped += get_dirty_pages(inode);
1484 trace_f2fs_writepages(mapping->host, wbc, DATA);
1485 return 0;
1486 }
1487
1488 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1489 {
1490 struct inode *inode = mapping->host;
1491 loff_t i_size = i_size_read(inode);
1492
1493 if (to > i_size) {
1494 truncate_pagecache(inode, i_size);
1495 truncate_blocks(inode, i_size, true);
1496 }
1497 }
1498
1499 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1500 struct page *page, loff_t pos, unsigned len,
1501 block_t *blk_addr, bool *node_changed)
1502 {
1503 struct inode *inode = page->mapping->host;
1504 pgoff_t index = page->index;
1505 struct dnode_of_data dn;
1506 struct page *ipage;
1507 bool locked = false;
1508 struct extent_info ei;
1509 int err = 0;
1510
1511 /*
1512 * we already allocated all the blocks, so we don't need to get
1513 * the block addresses when there is no need to fill the page.
1514 */
1515 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE)
1516 return 0;
1517
1518 if (f2fs_has_inline_data(inode) ||
1519 (pos & PAGE_MASK) >= i_size_read(inode)) {
1520 f2fs_lock_op(sbi);
1521 locked = true;
1522 }
1523 restart:
1524 /* check inline_data */
1525 ipage = get_node_page(sbi, inode->i_ino);
1526 if (IS_ERR(ipage)) {
1527 err = PTR_ERR(ipage);
1528 goto unlock_out;
1529 }
1530
1531 set_new_dnode(&dn, inode, ipage, ipage, 0);
1532
1533 if (f2fs_has_inline_data(inode)) {
1534 if (pos + len <= MAX_INLINE_DATA) {
1535 read_inline_data(page, ipage);
1536 set_inode_flag(inode, FI_DATA_EXIST);
1537 if (inode->i_nlink)
1538 set_inline_node(ipage);
1539 } else {
1540 err = f2fs_convert_inline_page(&dn, page);
1541 if (err)
1542 goto out;
1543 if (dn.data_blkaddr == NULL_ADDR)
1544 err = f2fs_get_block(&dn, index);
1545 }
1546 } else if (locked) {
1547 err = f2fs_get_block(&dn, index);
1548 } else {
1549 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1550 dn.data_blkaddr = ei.blk + index - ei.fofs;
1551 } else {
1552 /* hole case */
1553 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1554 if (err || dn.data_blkaddr == NULL_ADDR) {
1555 f2fs_put_dnode(&dn);
1556 f2fs_lock_op(sbi);
1557 locked = true;
1558 goto restart;
1559 }
1560 }
1561 }
1562
1563 /* convert_inline_page can make node_changed */
1564 *blk_addr = dn.data_blkaddr;
1565 *node_changed = dn.node_changed;
1566 out:
1567 f2fs_put_dnode(&dn);
1568 unlock_out:
1569 if (locked)
1570 f2fs_unlock_op(sbi);
1571 return err;
1572 }
1573
1574 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1575 loff_t pos, unsigned len, unsigned flags,
1576 struct page **pagep, void **fsdata)
1577 {
1578 struct inode *inode = mapping->host;
1579 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1580 struct page *page = NULL;
1581 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1582 bool need_balance = false;
1583 block_t blkaddr = NULL_ADDR;
1584 int err = 0;
1585
1586 if (trace_android_fs_datawrite_start_enabled()) {
1587 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
1588
1589 path = android_fstrace_get_pathname(pathbuf,
1590 MAX_TRACE_PATHBUF_LEN,
1591 inode);
1592 trace_android_fs_datawrite_start(inode, pos, len,
1593 current->pid, path,
1594 current->comm);
1595 }
1596 trace_f2fs_write_begin(inode, pos, len, flags);
1597
1598 /*
1599 * We should check this at this moment to avoid deadlock on inode page
1600 * and #0 page. The locking rule for inline_data conversion should be:
1601 * lock_page(page #0) -> lock_page(inode_page)
1602 */
1603 if (index != 0) {
1604 err = f2fs_convert_inline_inode(inode);
1605 if (err)
1606 goto fail;
1607 }
1608 repeat:
1609 page = grab_cache_page_write_begin(mapping, index, flags);
1610 if (!page) {
1611 err = -ENOMEM;
1612 goto fail;
1613 }
1614
1615 *pagep = page;
1616
1617 err = prepare_write_begin(sbi, page, pos, len,
1618 &blkaddr, &need_balance);
1619 if (err)
1620 goto fail;
1621
1622 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1623 unlock_page(page);
1624 f2fs_balance_fs(sbi, true);
1625 lock_page(page);
1626 if (page->mapping != mapping) {
1627 /* The page got truncated from under us */
1628 f2fs_put_page(page, 1);
1629 goto repeat;
1630 }
1631 }
1632
1633 f2fs_wait_on_page_writeback(page, DATA, false);
1634
1635 /* wait for GCed encrypted page writeback */
1636 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1637 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1638
1639 if (len == PAGE_SIZE || PageUptodate(page))
1640 return 0;
1641
1642 if (blkaddr == NEW_ADDR) {
1643 zero_user_segment(page, 0, PAGE_SIZE);
1644 SetPageUptodate(page);
1645 } else {
1646 struct bio *bio;
1647
1648 bio = f2fs_grab_bio(inode, blkaddr, 1);
1649 if (IS_ERR(bio)) {
1650 err = PTR_ERR(bio);
1651 goto fail;
1652 }
1653
1654 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1655 bio_put(bio);
1656 err = -EFAULT;
1657 goto fail;
1658 }
1659
1660 __submit_bio(sbi, READ_SYNC, bio, DATA);
1661
1662 lock_page(page);
1663 if (unlikely(page->mapping != mapping)) {
1664 f2fs_put_page(page, 1);
1665 goto repeat;
1666 }
1667 if (unlikely(!PageUptodate(page))) {
1668 err = -EIO;
1669 goto fail;
1670 }
1671 }
1672 return 0;
1673
1674 fail:
1675 f2fs_put_page(page, 1);
1676 f2fs_write_failed(mapping, pos + len);
1677 return err;
1678 }
1679
1680 static int f2fs_write_end(struct file *file,
1681 struct address_space *mapping,
1682 loff_t pos, unsigned len, unsigned copied,
1683 struct page *page, void *fsdata)
1684 {
1685 struct inode *inode = page->mapping->host;
1686
1687 trace_android_fs_datawrite_end(inode, pos, len);
1688 trace_f2fs_write_end(inode, pos, len, copied);
1689
1690 /*
1691 * This should be come from len == PAGE_SIZE, and we expect copied
1692 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1693 * let generic_perform_write() try to copy data again through copied=0.
1694 */
1695 if (!PageUptodate(page)) {
1696 if (unlikely(copied != PAGE_SIZE))
1697 copied = 0;
1698 else
1699 SetPageUptodate(page);
1700 }
1701 if (!copied)
1702 goto unlock_out;
1703
1704 set_page_dirty(page);
1705 clear_cold_data(page);
1706
1707 if (pos + copied > i_size_read(inode))
1708 f2fs_i_size_write(inode, pos + copied);
1709 unlock_out:
1710 f2fs_put_page(page, 1);
1711 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1712 return copied;
1713 }
1714
1715 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1716 loff_t offset)
1717 {
1718 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1719
1720 if (offset & blocksize_mask)
1721 return -EINVAL;
1722
1723 if (iov_iter_alignment(iter) & blocksize_mask)
1724 return -EINVAL;
1725
1726 return 0;
1727 }
1728
1729 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1730 loff_t offset)
1731 {
1732 struct address_space *mapping = iocb->ki_filp->f_mapping;
1733 struct inode *inode = mapping->host;
1734 size_t count = iov_iter_count(iter);
1735 int rw = iov_iter_rw(iter);
1736 int err;
1737
1738 err = check_direct_IO(inode, iter, offset);
1739 if (err)
1740 return err;
1741
1742 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1743 return 0;
1744 if (test_opt(F2FS_I_SB(inode), LFS))
1745 return 0;
1746
1747 if (trace_android_fs_dataread_start_enabled() &&
1748 (iov_iter_rw(iter) == READ)) {
1749 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
1750
1751 path = android_fstrace_get_pathname(pathbuf,
1752 MAX_TRACE_PATHBUF_LEN,
1753 inode);
1754 trace_android_fs_dataread_start(inode, offset,
1755 count, current->pid, path,
1756 current->comm);
1757 }
1758 if (trace_android_fs_datawrite_start_enabled() &&
1759 (iov_iter_rw(iter) == WRITE)) {
1760 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
1761
1762 path = android_fstrace_get_pathname(pathbuf,
1763 MAX_TRACE_PATHBUF_LEN,
1764 inode);
1765 trace_android_fs_datawrite_start(inode, offset, count,
1766 current->pid, path,
1767 current->comm);
1768 }
1769 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1770
1771 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1772 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1773 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1774
1775 if (rw == WRITE) {
1776 if (err > 0)
1777 set_inode_flag(inode, FI_UPDATE_WRITE);
1778 else if (err < 0)
1779 f2fs_write_failed(mapping, offset + count);
1780 }
1781
1782 if (trace_android_fs_dataread_start_enabled() &&
1783 (iov_iter_rw(iter) == READ))
1784 trace_android_fs_dataread_end(inode, offset, count);
1785 if (trace_android_fs_datawrite_start_enabled() &&
1786 (iov_iter_rw(iter) == WRITE))
1787 trace_android_fs_datawrite_end(inode, offset, count);
1788
1789 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1790
1791 return err;
1792 }
1793
1794 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1795 unsigned int length)
1796 {
1797 struct inode *inode = page->mapping->host;
1798 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1799
1800 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1801 (offset % PAGE_SIZE || length != PAGE_SIZE))
1802 return;
1803
1804 if (PageDirty(page)) {
1805 if (inode->i_ino == F2FS_META_INO(sbi)) {
1806 dec_page_count(sbi, F2FS_DIRTY_META);
1807 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
1808 dec_page_count(sbi, F2FS_DIRTY_NODES);
1809 } else {
1810 inode_dec_dirty_pages(inode);
1811 remove_dirty_inode(inode);
1812 }
1813 }
1814
1815 /* This is atomic written page, keep Private */
1816 if (IS_ATOMIC_WRITTEN_PAGE(page))
1817 return;
1818
1819 set_page_private(page, 0);
1820 ClearPagePrivate(page);
1821 }
1822
1823 int f2fs_release_page(struct page *page, gfp_t wait)
1824 {
1825 /* If this is dirty page, keep PagePrivate */
1826 if (PageDirty(page))
1827 return 0;
1828
1829 /* This is atomic written page, keep Private */
1830 if (IS_ATOMIC_WRITTEN_PAGE(page))
1831 return 0;
1832
1833 set_page_private(page, 0);
1834 ClearPagePrivate(page);
1835 return 1;
1836 }
1837
1838 /*
1839 * This was copied from __set_page_dirty_buffers which gives higher performance
1840 * in very high speed storages. (e.g., pmem)
1841 */
1842 void f2fs_set_page_dirty_nobuffers(struct page *page)
1843 {
1844 struct address_space *mapping = page->mapping;
1845 struct mem_cgroup *memcg;
1846 unsigned long flags;
1847
1848 if (unlikely(!mapping))
1849 return;
1850
1851 spin_lock(&mapping->private_lock);
1852 memcg = mem_cgroup_begin_page_stat(page);
1853 SetPageDirty(page);
1854 spin_unlock(&mapping->private_lock);
1855
1856 spin_lock_irqsave(&mapping->tree_lock, flags);
1857 WARN_ON_ONCE(!PageUptodate(page));
1858 account_page_dirtied(page, mapping, memcg);
1859 radix_tree_tag_set(&mapping->page_tree,
1860 page_index(page), PAGECACHE_TAG_DIRTY);
1861 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1862
1863 mem_cgroup_end_page_stat(memcg);
1864
1865 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1866 return;
1867 }
1868
1869 static int f2fs_set_data_page_dirty(struct page *page)
1870 {
1871 struct address_space *mapping = page->mapping;
1872 struct inode *inode = mapping->host;
1873
1874 trace_f2fs_set_page_dirty(page, DATA);
1875
1876 if (!PageUptodate(page))
1877 SetPageUptodate(page);
1878
1879 if (f2fs_is_atomic_file(inode)) {
1880 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1881 register_inmem_page(inode, page);
1882 return 1;
1883 }
1884 /*
1885 * Previously, this page has been registered, we just
1886 * return here.
1887 */
1888 return 0;
1889 }
1890
1891 if (!PageDirty(page)) {
1892 f2fs_set_page_dirty_nobuffers(page);
1893 update_dirty_page(inode, page);
1894 return 1;
1895 }
1896 return 0;
1897 }
1898
1899 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1900 {
1901 struct inode *inode = mapping->host;
1902
1903 if (f2fs_has_inline_data(inode))
1904 return 0;
1905
1906 /* make sure allocating whole blocks */
1907 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1908 filemap_write_and_wait(mapping);
1909
1910 return generic_block_bmap(mapping, block, get_data_block_bmap);
1911 }
1912
1913 #ifdef CONFIG_MIGRATION
1914 #include <linux/migrate.h>
1915
1916 int f2fs_migrate_page(struct address_space *mapping,
1917 struct page *newpage, struct page *page, enum migrate_mode mode)
1918 {
1919 int rc, extra_count;
1920 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
1921 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
1922
1923 BUG_ON(PageWriteback(page));
1924
1925 /* migrating an atomic written page is safe with the inmem_lock hold */
1926 if (atomic_written && !mutex_trylock(&fi->inmem_lock))
1927 return -EAGAIN;
1928
1929 /*
1930 * A reference is expected if PagePrivate set when move mapping,
1931 * however F2FS breaks this for maintaining dirty page counts when
1932 * truncating pages. So here adjusting the 'extra_count' make it work.
1933 */
1934 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
1935 rc = migrate_page_move_mapping(mapping, newpage,
1936 page, NULL, mode, extra_count);
1937 if (rc != MIGRATEPAGE_SUCCESS) {
1938 if (atomic_written)
1939 mutex_unlock(&fi->inmem_lock);
1940 return rc;
1941 }
1942
1943 if (atomic_written) {
1944 struct inmem_pages *cur;
1945 list_for_each_entry(cur, &fi->inmem_pages, list)
1946 if (cur->page == page) {
1947 cur->page = newpage;
1948 break;
1949 }
1950 mutex_unlock(&fi->inmem_lock);
1951 put_page(page);
1952 get_page(newpage);
1953 }
1954
1955 if (PagePrivate(page))
1956 SetPagePrivate(newpage);
1957 set_page_private(newpage, page_private(page));
1958
1959 migrate_page_copy(newpage, page);
1960
1961 return MIGRATEPAGE_SUCCESS;
1962 }
1963 #endif
1964
1965 const struct address_space_operations f2fs_dblock_aops = {
1966 .readpage = f2fs_read_data_page,
1967 .readpages = f2fs_read_data_pages,
1968 .writepage = f2fs_write_data_page,
1969 .writepages = f2fs_write_data_pages,
1970 .write_begin = f2fs_write_begin,
1971 .write_end = f2fs_write_end,
1972 .set_page_dirty = f2fs_set_data_page_dirty,
1973 .invalidatepage = f2fs_invalidate_page,
1974 .releasepage = f2fs_release_page,
1975 .direct_IO = f2fs_direct_IO,
1976 .bmap = f2fs_bmap,
1977 #ifdef CONFIG_MIGRATION
1978 .migratepage = f2fs_migrate_page,
1979 #endif
1980 };