]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/f2fs/data.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'akpm' (patches from Andrew)
[mirror_ubuntu-artful-kernel.git] / fs / f2fs / data.c
1 /*
2 * fs/f2fs/data.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
22 #include <linux/cleancache.h>
23
24 #include "f2fs.h"
25 #include "node.h"
26 #include "segment.h"
27 #include "trace.h"
28 #include <trace/events/f2fs.h>
29
30 static void f2fs_read_end_io(struct bio *bio)
31 {
32 struct bio_vec *bvec;
33 int i;
34
35 if (f2fs_bio_encrypted(bio)) {
36 if (bio->bi_error) {
37 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, struct bio *bio)
101 {
102 if (!is_read_io(bio_op(bio)))
103 atomic_inc(&sbi->nr_wb_bios);
104 submit_bio(bio);
105 }
106
107 static void __submit_merged_bio(struct f2fs_bio_info *io)
108 {
109 struct f2fs_io_info *fio = &io->fio;
110
111 if (!io->bio)
112 return;
113
114 if (is_read_io(fio->op))
115 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
116 else
117 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
118
119 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
120
121 __submit_bio(io->sbi, io->bio);
122 io->bio = NULL;
123 }
124
125 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
126 struct page *page, nid_t ino)
127 {
128 struct bio_vec *bvec;
129 struct page *target;
130 int i;
131
132 if (!io->bio)
133 return false;
134
135 if (!inode && !page && !ino)
136 return true;
137
138 bio_for_each_segment_all(bvec, io->bio, i) {
139
140 if (bvec->bv_page->mapping)
141 target = bvec->bv_page;
142 else
143 target = fscrypt_control_page(bvec->bv_page);
144
145 if (inode && inode == target->mapping->host)
146 return true;
147 if (page && page == target)
148 return true;
149 if (ino && ino == ino_of_node(target))
150 return true;
151 }
152
153 return false;
154 }
155
156 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
157 struct page *page, nid_t ino,
158 enum page_type type)
159 {
160 enum page_type btype = PAGE_TYPE_OF_BIO(type);
161 struct f2fs_bio_info *io = &sbi->write_io[btype];
162 bool ret;
163
164 down_read(&io->io_rwsem);
165 ret = __has_merged_page(io, inode, page, ino);
166 up_read(&io->io_rwsem);
167 return ret;
168 }
169
170 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
171 struct inode *inode, struct page *page,
172 nid_t ino, enum page_type type, int rw)
173 {
174 enum page_type btype = PAGE_TYPE_OF_BIO(type);
175 struct f2fs_bio_info *io;
176
177 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
178
179 down_write(&io->io_rwsem);
180
181 if (!__has_merged_page(io, inode, page, ino))
182 goto out;
183
184 /* change META to META_FLUSH in the checkpoint procedure */
185 if (type >= META_FLUSH) {
186 io->fio.type = META_FLUSH;
187 io->fio.op = REQ_OP_WRITE;
188 if (test_opt(sbi, NOBARRIER))
189 io->fio.op_flags = WRITE_FLUSH | REQ_META | REQ_PRIO;
190 else
191 io->fio.op_flags = WRITE_FLUSH_FUA | REQ_META |
192 REQ_PRIO;
193 }
194 __submit_merged_bio(io);
195 out:
196 up_write(&io->io_rwsem);
197 }
198
199 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
200 int rw)
201 {
202 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
203 }
204
205 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
206 struct inode *inode, struct page *page,
207 nid_t ino, enum page_type type, int rw)
208 {
209 if (has_merged_page(sbi, inode, page, ino, type))
210 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
211 }
212
213 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
214 {
215 f2fs_submit_merged_bio(sbi, DATA, WRITE);
216 f2fs_submit_merged_bio(sbi, NODE, WRITE);
217 f2fs_submit_merged_bio(sbi, META, WRITE);
218 }
219
220 /*
221 * Fill the locked page with data located in the block address.
222 * Return unlocked page.
223 */
224 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
225 {
226 struct bio *bio;
227 struct page *page = fio->encrypted_page ?
228 fio->encrypted_page : fio->page;
229
230 trace_f2fs_submit_page_bio(page, fio);
231 f2fs_trace_ios(fio, 0);
232
233 /* Allocate a new bio */
234 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
235
236 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
237 bio_put(bio);
238 return -EFAULT;
239 }
240 bio->bi_rw = fio->op_flags;
241 bio_set_op_attrs(bio, fio->op, fio->op_flags);
242
243 __submit_bio(fio->sbi, bio);
244 return 0;
245 }
246
247 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
248 {
249 struct f2fs_sb_info *sbi = fio->sbi;
250 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
251 struct f2fs_bio_info *io;
252 bool is_read = is_read_io(fio->op);
253 struct page *bio_page;
254
255 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
256
257 if (fio->old_blkaddr != NEW_ADDR)
258 verify_block_addr(sbi, fio->old_blkaddr);
259 verify_block_addr(sbi, fio->new_blkaddr);
260
261 down_write(&io->io_rwsem);
262
263 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
264 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags)))
265 __submit_merged_bio(io);
266 alloc_new:
267 if (io->bio == NULL) {
268 int bio_blocks = MAX_BIO_BLOCKS(sbi);
269
270 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
271 bio_blocks, is_read);
272 io->fio = *fio;
273 }
274
275 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
276
277 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
278 PAGE_SIZE) {
279 __submit_merged_bio(io);
280 goto alloc_new;
281 }
282
283 io->last_block_in_bio = fio->new_blkaddr;
284 f2fs_trace_ios(fio, 0);
285
286 up_write(&io->io_rwsem);
287 trace_f2fs_submit_page_mbio(fio->page, fio);
288 }
289
290 static void __set_data_blkaddr(struct dnode_of_data *dn)
291 {
292 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
293 __le32 *addr_array;
294
295 /* Get physical address of data block */
296 addr_array = blkaddr_in_node(rn);
297 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
298 }
299
300 /*
301 * Lock ordering for the change of data block address:
302 * ->data_page
303 * ->node_page
304 * update block addresses in the node page
305 */
306 void set_data_blkaddr(struct dnode_of_data *dn)
307 {
308 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
309 __set_data_blkaddr(dn);
310 if (set_page_dirty(dn->node_page))
311 dn->node_changed = true;
312 }
313
314 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
315 {
316 dn->data_blkaddr = blkaddr;
317 set_data_blkaddr(dn);
318 f2fs_update_extent_cache(dn);
319 }
320
321 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
322 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
323 {
324 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
325
326 if (!count)
327 return 0;
328
329 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
330 return -EPERM;
331 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
332 return -ENOSPC;
333
334 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
335 dn->ofs_in_node, count);
336
337 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
338
339 for (; count > 0; dn->ofs_in_node++) {
340 block_t blkaddr =
341 datablock_addr(dn->node_page, dn->ofs_in_node);
342 if (blkaddr == NULL_ADDR) {
343 dn->data_blkaddr = NEW_ADDR;
344 __set_data_blkaddr(dn);
345 count--;
346 }
347 }
348
349 if (set_page_dirty(dn->node_page))
350 dn->node_changed = true;
351
352 mark_inode_dirty(dn->inode);
353 sync_inode_page(dn);
354 return 0;
355 }
356
357 /* Should keep dn->ofs_in_node unchanged */
358 int reserve_new_block(struct dnode_of_data *dn)
359 {
360 unsigned int ofs_in_node = dn->ofs_in_node;
361 int ret;
362
363 ret = reserve_new_blocks(dn, 1);
364 dn->ofs_in_node = ofs_in_node;
365 return ret;
366 }
367
368 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
369 {
370 bool need_put = dn->inode_page ? false : true;
371 int err;
372
373 err = get_dnode_of_data(dn, index, ALLOC_NODE);
374 if (err)
375 return err;
376
377 if (dn->data_blkaddr == NULL_ADDR)
378 err = reserve_new_block(dn);
379 if (err || need_put)
380 f2fs_put_dnode(dn);
381 return err;
382 }
383
384 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
385 {
386 struct extent_info ei;
387 struct inode *inode = dn->inode;
388
389 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
390 dn->data_blkaddr = ei.blk + index - ei.fofs;
391 return 0;
392 }
393
394 return f2fs_reserve_block(dn, index);
395 }
396
397 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
398 int op_flags, bool for_write)
399 {
400 struct address_space *mapping = inode->i_mapping;
401 struct dnode_of_data dn;
402 struct page *page;
403 struct extent_info ei;
404 int err;
405 struct f2fs_io_info fio = {
406 .sbi = F2FS_I_SB(inode),
407 .type = DATA,
408 .op = REQ_OP_READ,
409 .op_flags = op_flags,
410 .encrypted_page = NULL,
411 };
412
413 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
414 return read_mapping_page(mapping, index, NULL);
415
416 page = f2fs_grab_cache_page(mapping, index, for_write);
417 if (!page)
418 return ERR_PTR(-ENOMEM);
419
420 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
421 dn.data_blkaddr = ei.blk + index - ei.fofs;
422 goto got_it;
423 }
424
425 set_new_dnode(&dn, inode, NULL, NULL, 0);
426 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
427 if (err)
428 goto put_err;
429 f2fs_put_dnode(&dn);
430
431 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
432 err = -ENOENT;
433 goto put_err;
434 }
435 got_it:
436 if (PageUptodate(page)) {
437 unlock_page(page);
438 return page;
439 }
440
441 /*
442 * A new dentry page is allocated but not able to be written, since its
443 * new inode page couldn't be allocated due to -ENOSPC.
444 * In such the case, its blkaddr can be remained as NEW_ADDR.
445 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
446 */
447 if (dn.data_blkaddr == NEW_ADDR) {
448 zero_user_segment(page, 0, PAGE_SIZE);
449 SetPageUptodate(page);
450 unlock_page(page);
451 return page;
452 }
453
454 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
455 fio.page = page;
456 err = f2fs_submit_page_bio(&fio);
457 if (err)
458 goto put_err;
459 return page;
460
461 put_err:
462 f2fs_put_page(page, 1);
463 return ERR_PTR(err);
464 }
465
466 struct page *find_data_page(struct inode *inode, pgoff_t index)
467 {
468 struct address_space *mapping = inode->i_mapping;
469 struct page *page;
470
471 page = find_get_page(mapping, index);
472 if (page && PageUptodate(page))
473 return page;
474 f2fs_put_page(page, 0);
475
476 page = get_read_data_page(inode, index, READ_SYNC, false);
477 if (IS_ERR(page))
478 return page;
479
480 if (PageUptodate(page))
481 return page;
482
483 wait_on_page_locked(page);
484 if (unlikely(!PageUptodate(page))) {
485 f2fs_put_page(page, 0);
486 return ERR_PTR(-EIO);
487 }
488 return page;
489 }
490
491 /*
492 * If it tries to access a hole, return an error.
493 * Because, the callers, functions in dir.c and GC, should be able to know
494 * whether this page exists or not.
495 */
496 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
497 bool for_write)
498 {
499 struct address_space *mapping = inode->i_mapping;
500 struct page *page;
501 repeat:
502 page = get_read_data_page(inode, index, READ_SYNC, for_write);
503 if (IS_ERR(page))
504 return page;
505
506 /* wait for read completion */
507 lock_page(page);
508 if (unlikely(!PageUptodate(page))) {
509 f2fs_put_page(page, 1);
510 return ERR_PTR(-EIO);
511 }
512 if (unlikely(page->mapping != mapping)) {
513 f2fs_put_page(page, 1);
514 goto repeat;
515 }
516 return page;
517 }
518
519 /*
520 * Caller ensures that this data page is never allocated.
521 * A new zero-filled data page is allocated in the page cache.
522 *
523 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
524 * f2fs_unlock_op().
525 * Note that, ipage is set only by make_empty_dir, and if any error occur,
526 * ipage should be released by this function.
527 */
528 struct page *get_new_data_page(struct inode *inode,
529 struct page *ipage, pgoff_t index, bool new_i_size)
530 {
531 struct address_space *mapping = inode->i_mapping;
532 struct page *page;
533 struct dnode_of_data dn;
534 int err;
535
536 page = f2fs_grab_cache_page(mapping, index, true);
537 if (!page) {
538 /*
539 * before exiting, we should make sure ipage will be released
540 * if any error occur.
541 */
542 f2fs_put_page(ipage, 1);
543 return ERR_PTR(-ENOMEM);
544 }
545
546 set_new_dnode(&dn, inode, ipage, NULL, 0);
547 err = f2fs_reserve_block(&dn, index);
548 if (err) {
549 f2fs_put_page(page, 1);
550 return ERR_PTR(err);
551 }
552 if (!ipage)
553 f2fs_put_dnode(&dn);
554
555 if (PageUptodate(page))
556 goto got_it;
557
558 if (dn.data_blkaddr == NEW_ADDR) {
559 zero_user_segment(page, 0, PAGE_SIZE);
560 SetPageUptodate(page);
561 } else {
562 f2fs_put_page(page, 1);
563
564 /* if ipage exists, blkaddr should be NEW_ADDR */
565 f2fs_bug_on(F2FS_I_SB(inode), ipage);
566 page = get_lock_data_page(inode, index, true);
567 if (IS_ERR(page))
568 return page;
569 }
570 got_it:
571 if (new_i_size && i_size_read(inode) <
572 ((loff_t)(index + 1) << PAGE_SHIFT)) {
573 i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
574 /* Only the directory inode sets new_i_size */
575 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
576 }
577 return page;
578 }
579
580 static int __allocate_data_block(struct dnode_of_data *dn)
581 {
582 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
583 struct f2fs_summary sum;
584 struct node_info ni;
585 int seg = CURSEG_WARM_DATA;
586 pgoff_t fofs;
587 blkcnt_t count = 1;
588
589 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
590 return -EPERM;
591
592 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
593 if (dn->data_blkaddr == NEW_ADDR)
594 goto alloc;
595
596 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
597 return -ENOSPC;
598
599 alloc:
600 get_node_info(sbi, dn->nid, &ni);
601 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
602
603 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
604 seg = CURSEG_DIRECT_IO;
605
606 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
607 &sum, seg);
608 set_data_blkaddr(dn);
609
610 /* update i_size */
611 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
612 dn->ofs_in_node;
613 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
614 i_size_write(dn->inode,
615 ((loff_t)(fofs + 1) << PAGE_SHIFT));
616 return 0;
617 }
618
619 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
620 {
621 struct inode *inode = file_inode(iocb->ki_filp);
622 struct f2fs_map_blocks map;
623 ssize_t ret = 0;
624
625 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
626 map.m_len = F2FS_BYTES_TO_BLK(iov_iter_count(from));
627 map.m_next_pgofs = NULL;
628
629 if (f2fs_encrypted_inode(inode))
630 return 0;
631
632 if (iocb->ki_flags & IOCB_DIRECT) {
633 ret = f2fs_convert_inline_inode(inode);
634 if (ret)
635 return ret;
636 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
637 }
638 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
639 ret = f2fs_convert_inline_inode(inode);
640 if (ret)
641 return ret;
642 }
643 if (!f2fs_has_inline_data(inode))
644 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
645 return ret;
646 }
647
648 /*
649 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
650 * f2fs_map_blocks structure.
651 * If original data blocks are allocated, then give them to blockdev.
652 * Otherwise,
653 * a. preallocate requested block addresses
654 * b. do not use extent cache for better performance
655 * c. give the block addresses to blockdev
656 */
657 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
658 int create, int flag)
659 {
660 unsigned int maxblocks = map->m_len;
661 struct dnode_of_data dn;
662 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
663 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
664 pgoff_t pgofs, end_offset, end;
665 int err = 0, ofs = 1;
666 unsigned int ofs_in_node, last_ofs_in_node;
667 blkcnt_t prealloc;
668 struct extent_info ei;
669 bool allocated = false;
670 block_t blkaddr;
671
672 map->m_len = 0;
673 map->m_flags = 0;
674
675 /* it only supports block size == page size */
676 pgofs = (pgoff_t)map->m_lblk;
677 end = pgofs + maxblocks;
678
679 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
680 map->m_pblk = ei.blk + pgofs - ei.fofs;
681 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
682 map->m_flags = F2FS_MAP_MAPPED;
683 goto out;
684 }
685
686 next_dnode:
687 if (create)
688 f2fs_lock_op(sbi);
689
690 /* When reading holes, we need its node page */
691 set_new_dnode(&dn, inode, NULL, NULL, 0);
692 err = get_dnode_of_data(&dn, pgofs, mode);
693 if (err) {
694 if (flag == F2FS_GET_BLOCK_BMAP)
695 map->m_pblk = 0;
696 if (err == -ENOENT) {
697 err = 0;
698 if (map->m_next_pgofs)
699 *map->m_next_pgofs =
700 get_next_page_offset(&dn, pgofs);
701 }
702 goto unlock_out;
703 }
704
705 prealloc = 0;
706 ofs_in_node = dn.ofs_in_node;
707 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
708
709 next_block:
710 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
711
712 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
713 if (create) {
714 if (unlikely(f2fs_cp_error(sbi))) {
715 err = -EIO;
716 goto sync_out;
717 }
718 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
719 if (blkaddr == NULL_ADDR) {
720 prealloc++;
721 last_ofs_in_node = dn.ofs_in_node;
722 }
723 } else {
724 err = __allocate_data_block(&dn);
725 if (!err) {
726 set_inode_flag(F2FS_I(inode),
727 FI_APPEND_WRITE);
728 allocated = true;
729 }
730 }
731 if (err)
732 goto sync_out;
733 map->m_flags = F2FS_MAP_NEW;
734 blkaddr = dn.data_blkaddr;
735 } else {
736 if (flag == F2FS_GET_BLOCK_BMAP) {
737 map->m_pblk = 0;
738 goto sync_out;
739 }
740 if (flag == F2FS_GET_BLOCK_FIEMAP &&
741 blkaddr == NULL_ADDR) {
742 if (map->m_next_pgofs)
743 *map->m_next_pgofs = pgofs + 1;
744 }
745 if (flag != F2FS_GET_BLOCK_FIEMAP ||
746 blkaddr != NEW_ADDR)
747 goto sync_out;
748 }
749 }
750
751 if (flag == F2FS_GET_BLOCK_PRE_AIO)
752 goto skip;
753
754 if (map->m_len == 0) {
755 /* preallocated unwritten block should be mapped for fiemap. */
756 if (blkaddr == NEW_ADDR)
757 map->m_flags |= F2FS_MAP_UNWRITTEN;
758 map->m_flags |= F2FS_MAP_MAPPED;
759
760 map->m_pblk = blkaddr;
761 map->m_len = 1;
762 } else if ((map->m_pblk != NEW_ADDR &&
763 blkaddr == (map->m_pblk + ofs)) ||
764 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
765 flag == F2FS_GET_BLOCK_PRE_DIO) {
766 ofs++;
767 map->m_len++;
768 } else {
769 goto sync_out;
770 }
771
772 skip:
773 dn.ofs_in_node++;
774 pgofs++;
775
776 /* preallocate blocks in batch for one dnode page */
777 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
778 (pgofs == end || dn.ofs_in_node == end_offset)) {
779
780 dn.ofs_in_node = ofs_in_node;
781 err = reserve_new_blocks(&dn, prealloc);
782 if (err)
783 goto sync_out;
784
785 map->m_len += dn.ofs_in_node - ofs_in_node;
786 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
787 err = -ENOSPC;
788 goto sync_out;
789 }
790 dn.ofs_in_node = end_offset;
791 }
792
793 if (pgofs >= end)
794 goto sync_out;
795 else if (dn.ofs_in_node < end_offset)
796 goto next_block;
797
798 if (allocated)
799 sync_inode_page(&dn);
800 f2fs_put_dnode(&dn);
801
802 if (create) {
803 f2fs_unlock_op(sbi);
804 f2fs_balance_fs(sbi, allocated);
805 }
806 allocated = false;
807 goto next_dnode;
808
809 sync_out:
810 if (allocated)
811 sync_inode_page(&dn);
812 f2fs_put_dnode(&dn);
813 unlock_out:
814 if (create) {
815 f2fs_unlock_op(sbi);
816 f2fs_balance_fs(sbi, allocated);
817 }
818 out:
819 trace_f2fs_map_blocks(inode, map, err);
820 return err;
821 }
822
823 static int __get_data_block(struct inode *inode, sector_t iblock,
824 struct buffer_head *bh, int create, int flag,
825 pgoff_t *next_pgofs)
826 {
827 struct f2fs_map_blocks map;
828 int ret;
829
830 map.m_lblk = iblock;
831 map.m_len = bh->b_size >> inode->i_blkbits;
832 map.m_next_pgofs = next_pgofs;
833
834 ret = f2fs_map_blocks(inode, &map, create, flag);
835 if (!ret) {
836 map_bh(bh, inode->i_sb, map.m_pblk);
837 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
838 bh->b_size = map.m_len << inode->i_blkbits;
839 }
840 return ret;
841 }
842
843 static int get_data_block(struct inode *inode, sector_t iblock,
844 struct buffer_head *bh_result, int create, int flag,
845 pgoff_t *next_pgofs)
846 {
847 return __get_data_block(inode, iblock, bh_result, create,
848 flag, next_pgofs);
849 }
850
851 static int get_data_block_dio(struct inode *inode, sector_t iblock,
852 struct buffer_head *bh_result, int create)
853 {
854 return __get_data_block(inode, iblock, bh_result, create,
855 F2FS_GET_BLOCK_DIO, NULL);
856 }
857
858 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
859 struct buffer_head *bh_result, int create)
860 {
861 /* Block number less than F2FS MAX BLOCKS */
862 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
863 return -EFBIG;
864
865 return __get_data_block(inode, iblock, bh_result, create,
866 F2FS_GET_BLOCK_BMAP, NULL);
867 }
868
869 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
870 {
871 return (offset >> inode->i_blkbits);
872 }
873
874 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
875 {
876 return (blk << inode->i_blkbits);
877 }
878
879 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
880 u64 start, u64 len)
881 {
882 struct buffer_head map_bh;
883 sector_t start_blk, last_blk;
884 pgoff_t next_pgofs;
885 loff_t isize;
886 u64 logical = 0, phys = 0, size = 0;
887 u32 flags = 0;
888 int ret = 0;
889
890 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
891 if (ret)
892 return ret;
893
894 if (f2fs_has_inline_data(inode)) {
895 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
896 if (ret != -EAGAIN)
897 return ret;
898 }
899
900 inode_lock(inode);
901
902 isize = i_size_read(inode);
903 if (start >= isize)
904 goto out;
905
906 if (start + len > isize)
907 len = isize - start;
908
909 if (logical_to_blk(inode, len) == 0)
910 len = blk_to_logical(inode, 1);
911
912 start_blk = logical_to_blk(inode, start);
913 last_blk = logical_to_blk(inode, start + len - 1);
914
915 next:
916 memset(&map_bh, 0, sizeof(struct buffer_head));
917 map_bh.b_size = len;
918
919 ret = get_data_block(inode, start_blk, &map_bh, 0,
920 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
921 if (ret)
922 goto out;
923
924 /* HOLE */
925 if (!buffer_mapped(&map_bh)) {
926 start_blk = next_pgofs;
927 /* Go through holes util pass the EOF */
928 if (blk_to_logical(inode, start_blk) < isize)
929 goto prep_next;
930 /* Found a hole beyond isize means no more extents.
931 * Note that the premise is that filesystems don't
932 * punch holes beyond isize and keep size unchanged.
933 */
934 flags |= FIEMAP_EXTENT_LAST;
935 }
936
937 if (size) {
938 if (f2fs_encrypted_inode(inode))
939 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
940
941 ret = fiemap_fill_next_extent(fieinfo, logical,
942 phys, size, flags);
943 }
944
945 if (start_blk > last_blk || ret)
946 goto out;
947
948 logical = blk_to_logical(inode, start_blk);
949 phys = blk_to_logical(inode, map_bh.b_blocknr);
950 size = map_bh.b_size;
951 flags = 0;
952 if (buffer_unwritten(&map_bh))
953 flags = FIEMAP_EXTENT_UNWRITTEN;
954
955 start_blk += logical_to_blk(inode, size);
956
957 prep_next:
958 cond_resched();
959 if (fatal_signal_pending(current))
960 ret = -EINTR;
961 else
962 goto next;
963 out:
964 if (ret == 1)
965 ret = 0;
966
967 inode_unlock(inode);
968 return ret;
969 }
970
971 /*
972 * This function was originally taken from fs/mpage.c, and customized for f2fs.
973 * Major change was from block_size == page_size in f2fs by default.
974 */
975 static int f2fs_mpage_readpages(struct address_space *mapping,
976 struct list_head *pages, struct page *page,
977 unsigned nr_pages)
978 {
979 struct bio *bio = NULL;
980 unsigned page_idx;
981 sector_t last_block_in_bio = 0;
982 struct inode *inode = mapping->host;
983 const unsigned blkbits = inode->i_blkbits;
984 const unsigned blocksize = 1 << blkbits;
985 sector_t block_in_file;
986 sector_t last_block;
987 sector_t last_block_in_file;
988 sector_t block_nr;
989 struct block_device *bdev = inode->i_sb->s_bdev;
990 struct f2fs_map_blocks map;
991
992 map.m_pblk = 0;
993 map.m_lblk = 0;
994 map.m_len = 0;
995 map.m_flags = 0;
996 map.m_next_pgofs = NULL;
997
998 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
999
1000 prefetchw(&page->flags);
1001 if (pages) {
1002 page = list_entry(pages->prev, struct page, lru);
1003 list_del(&page->lru);
1004 if (add_to_page_cache_lru(page, mapping,
1005 page->index,
1006 readahead_gfp_mask(mapping)))
1007 goto next_page;
1008 }
1009
1010 block_in_file = (sector_t)page->index;
1011 last_block = block_in_file + nr_pages;
1012 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1013 blkbits;
1014 if (last_block > last_block_in_file)
1015 last_block = last_block_in_file;
1016
1017 /*
1018 * Map blocks using the previous result first.
1019 */
1020 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1021 block_in_file > map.m_lblk &&
1022 block_in_file < (map.m_lblk + map.m_len))
1023 goto got_it;
1024
1025 /*
1026 * Then do more f2fs_map_blocks() calls until we are
1027 * done with this page.
1028 */
1029 map.m_flags = 0;
1030
1031 if (block_in_file < last_block) {
1032 map.m_lblk = block_in_file;
1033 map.m_len = last_block - block_in_file;
1034
1035 if (f2fs_map_blocks(inode, &map, 0,
1036 F2FS_GET_BLOCK_READ))
1037 goto set_error_page;
1038 }
1039 got_it:
1040 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1041 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1042 SetPageMappedToDisk(page);
1043
1044 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1045 SetPageUptodate(page);
1046 goto confused;
1047 }
1048 } else {
1049 zero_user_segment(page, 0, PAGE_SIZE);
1050 SetPageUptodate(page);
1051 unlock_page(page);
1052 goto next_page;
1053 }
1054
1055 /*
1056 * This page will go to BIO. Do we need to send this
1057 * BIO off first?
1058 */
1059 if (bio && (last_block_in_bio != block_nr - 1)) {
1060 submit_and_realloc:
1061 __submit_bio(F2FS_I_SB(inode), bio);
1062 bio = NULL;
1063 }
1064 if (bio == NULL) {
1065 struct fscrypt_ctx *ctx = NULL;
1066
1067 if (f2fs_encrypted_inode(inode) &&
1068 S_ISREG(inode->i_mode)) {
1069
1070 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1071 if (IS_ERR(ctx))
1072 goto set_error_page;
1073
1074 /* wait the page to be moved by cleaning */
1075 f2fs_wait_on_encrypted_page_writeback(
1076 F2FS_I_SB(inode), block_nr);
1077 }
1078
1079 bio = bio_alloc(GFP_KERNEL,
1080 min_t(int, nr_pages, BIO_MAX_PAGES));
1081 if (!bio) {
1082 if (ctx)
1083 fscrypt_release_ctx(ctx);
1084 goto set_error_page;
1085 }
1086 bio->bi_bdev = bdev;
1087 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1088 bio->bi_end_io = f2fs_read_end_io;
1089 bio->bi_private = ctx;
1090 bio_set_op_attrs(bio, REQ_OP_READ, 0);
1091 }
1092
1093 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1094 goto submit_and_realloc;
1095
1096 last_block_in_bio = block_nr;
1097 goto next_page;
1098 set_error_page:
1099 SetPageError(page);
1100 zero_user_segment(page, 0, PAGE_SIZE);
1101 unlock_page(page);
1102 goto next_page;
1103 confused:
1104 if (bio) {
1105 __submit_bio(F2FS_I_SB(inode), bio);
1106 bio = NULL;
1107 }
1108 unlock_page(page);
1109 next_page:
1110 if (pages)
1111 put_page(page);
1112 }
1113 BUG_ON(pages && !list_empty(pages));
1114 if (bio)
1115 __submit_bio(F2FS_I_SB(inode), bio);
1116 return 0;
1117 }
1118
1119 static int f2fs_read_data_page(struct file *file, struct page *page)
1120 {
1121 struct inode *inode = page->mapping->host;
1122 int ret = -EAGAIN;
1123
1124 trace_f2fs_readpage(page, DATA);
1125
1126 /* If the file has inline data, try to read it directly */
1127 if (f2fs_has_inline_data(inode))
1128 ret = f2fs_read_inline_data(inode, page);
1129 if (ret == -EAGAIN)
1130 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1131 return ret;
1132 }
1133
1134 static int f2fs_read_data_pages(struct file *file,
1135 struct address_space *mapping,
1136 struct list_head *pages, unsigned nr_pages)
1137 {
1138 struct inode *inode = file->f_mapping->host;
1139 struct page *page = list_entry(pages->prev, struct page, lru);
1140
1141 trace_f2fs_readpages(inode, page, nr_pages);
1142
1143 /* If the file has inline data, skip readpages */
1144 if (f2fs_has_inline_data(inode))
1145 return 0;
1146
1147 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1148 }
1149
1150 int do_write_data_page(struct f2fs_io_info *fio)
1151 {
1152 struct page *page = fio->page;
1153 struct inode *inode = page->mapping->host;
1154 struct dnode_of_data dn;
1155 int err = 0;
1156
1157 set_new_dnode(&dn, inode, NULL, NULL, 0);
1158 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1159 if (err)
1160 return err;
1161
1162 fio->old_blkaddr = dn.data_blkaddr;
1163
1164 /* This page is already truncated */
1165 if (fio->old_blkaddr == NULL_ADDR) {
1166 ClearPageUptodate(page);
1167 goto out_writepage;
1168 }
1169
1170 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1171 gfp_t gfp_flags = GFP_NOFS;
1172
1173 /* wait for GCed encrypted page writeback */
1174 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1175 fio->old_blkaddr);
1176 retry_encrypt:
1177 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1178 gfp_flags);
1179 if (IS_ERR(fio->encrypted_page)) {
1180 err = PTR_ERR(fio->encrypted_page);
1181 if (err == -ENOMEM) {
1182 /* flush pending ios and wait for a while */
1183 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1184 congestion_wait(BLK_RW_ASYNC, HZ/50);
1185 gfp_flags |= __GFP_NOFAIL;
1186 err = 0;
1187 goto retry_encrypt;
1188 }
1189 goto out_writepage;
1190 }
1191 }
1192
1193 set_page_writeback(page);
1194
1195 /*
1196 * If current allocation needs SSR,
1197 * it had better in-place writes for updated data.
1198 */
1199 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1200 !is_cold_data(page) &&
1201 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1202 need_inplace_update(inode))) {
1203 rewrite_data_page(fio);
1204 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1205 trace_f2fs_do_write_data_page(page, IPU);
1206 } else {
1207 write_data_page(&dn, fio);
1208 trace_f2fs_do_write_data_page(page, OPU);
1209 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1210 if (page->index == 0)
1211 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1212 }
1213 out_writepage:
1214 f2fs_put_dnode(&dn);
1215 return err;
1216 }
1217
1218 static int f2fs_write_data_page(struct page *page,
1219 struct writeback_control *wbc)
1220 {
1221 struct inode *inode = page->mapping->host;
1222 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1223 loff_t i_size = i_size_read(inode);
1224 const pgoff_t end_index = ((unsigned long long) i_size)
1225 >> PAGE_SHIFT;
1226 unsigned offset = 0;
1227 bool need_balance_fs = false;
1228 int err = 0;
1229 struct f2fs_io_info fio = {
1230 .sbi = sbi,
1231 .type = DATA,
1232 .op = REQ_OP_WRITE,
1233 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1234 .page = page,
1235 .encrypted_page = NULL,
1236 };
1237
1238 trace_f2fs_writepage(page, DATA);
1239
1240 if (page->index < end_index)
1241 goto write;
1242
1243 /*
1244 * If the offset is out-of-range of file size,
1245 * this page does not have to be written to disk.
1246 */
1247 offset = i_size & (PAGE_SIZE - 1);
1248 if ((page->index >= end_index + 1) || !offset)
1249 goto out;
1250
1251 zero_user_segment(page, offset, PAGE_SIZE);
1252 write:
1253 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1254 goto redirty_out;
1255 if (f2fs_is_drop_cache(inode))
1256 goto out;
1257 /* we should not write 0'th page having journal header */
1258 if (f2fs_is_volatile_file(inode) && (!page->index ||
1259 (!wbc->for_reclaim &&
1260 available_free_memory(sbi, BASE_CHECK))))
1261 goto redirty_out;
1262
1263 /* Dentry blocks are controlled by checkpoint */
1264 if (S_ISDIR(inode->i_mode)) {
1265 if (unlikely(f2fs_cp_error(sbi)))
1266 goto redirty_out;
1267 err = do_write_data_page(&fio);
1268 goto done;
1269 }
1270
1271 /* we should bypass data pages to proceed the kworkder jobs */
1272 if (unlikely(f2fs_cp_error(sbi))) {
1273 SetPageError(page);
1274 goto out;
1275 }
1276
1277 if (!wbc->for_reclaim)
1278 need_balance_fs = true;
1279 else if (has_not_enough_free_secs(sbi, 0))
1280 goto redirty_out;
1281
1282 err = -EAGAIN;
1283 f2fs_lock_op(sbi);
1284 if (f2fs_has_inline_data(inode))
1285 err = f2fs_write_inline_data(inode, page);
1286 if (err == -EAGAIN)
1287 err = do_write_data_page(&fio);
1288 f2fs_unlock_op(sbi);
1289 done:
1290 if (err && err != -ENOENT)
1291 goto redirty_out;
1292
1293 clear_cold_data(page);
1294 out:
1295 inode_dec_dirty_pages(inode);
1296 if (err)
1297 ClearPageUptodate(page);
1298
1299 if (wbc->for_reclaim) {
1300 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1301 remove_dirty_inode(inode);
1302 }
1303
1304 unlock_page(page);
1305 f2fs_balance_fs(sbi, need_balance_fs);
1306
1307 if (unlikely(f2fs_cp_error(sbi)))
1308 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1309
1310 return 0;
1311
1312 redirty_out:
1313 redirty_page_for_writepage(wbc, page);
1314 return AOP_WRITEPAGE_ACTIVATE;
1315 }
1316
1317 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1318 void *data)
1319 {
1320 struct address_space *mapping = data;
1321 int ret = mapping->a_ops->writepage(page, wbc);
1322 mapping_set_error(mapping, ret);
1323 return ret;
1324 }
1325
1326 /*
1327 * This function was copied from write_cche_pages from mm/page-writeback.c.
1328 * The major change is making write step of cold data page separately from
1329 * warm/hot data page.
1330 */
1331 static int f2fs_write_cache_pages(struct address_space *mapping,
1332 struct writeback_control *wbc, writepage_t writepage,
1333 void *data)
1334 {
1335 int ret = 0;
1336 int done = 0;
1337 struct pagevec pvec;
1338 int nr_pages;
1339 pgoff_t uninitialized_var(writeback_index);
1340 pgoff_t index;
1341 pgoff_t end; /* Inclusive */
1342 pgoff_t done_index;
1343 int cycled;
1344 int range_whole = 0;
1345 int tag;
1346 int step = 0;
1347
1348 pagevec_init(&pvec, 0);
1349 next:
1350 if (wbc->range_cyclic) {
1351 writeback_index = mapping->writeback_index; /* prev offset */
1352 index = writeback_index;
1353 if (index == 0)
1354 cycled = 1;
1355 else
1356 cycled = 0;
1357 end = -1;
1358 } else {
1359 index = wbc->range_start >> PAGE_SHIFT;
1360 end = wbc->range_end >> PAGE_SHIFT;
1361 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1362 range_whole = 1;
1363 cycled = 1; /* ignore range_cyclic tests */
1364 }
1365 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1366 tag = PAGECACHE_TAG_TOWRITE;
1367 else
1368 tag = PAGECACHE_TAG_DIRTY;
1369 retry:
1370 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1371 tag_pages_for_writeback(mapping, index, end);
1372 done_index = index;
1373 while (!done && (index <= end)) {
1374 int i;
1375
1376 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1377 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1378 if (nr_pages == 0)
1379 break;
1380
1381 for (i = 0; i < nr_pages; i++) {
1382 struct page *page = pvec.pages[i];
1383
1384 if (page->index > end) {
1385 done = 1;
1386 break;
1387 }
1388
1389 done_index = page->index;
1390
1391 lock_page(page);
1392
1393 if (unlikely(page->mapping != mapping)) {
1394 continue_unlock:
1395 unlock_page(page);
1396 continue;
1397 }
1398
1399 if (!PageDirty(page)) {
1400 /* someone wrote it for us */
1401 goto continue_unlock;
1402 }
1403
1404 if (step == is_cold_data(page))
1405 goto continue_unlock;
1406
1407 if (PageWriteback(page)) {
1408 if (wbc->sync_mode != WB_SYNC_NONE)
1409 f2fs_wait_on_page_writeback(page,
1410 DATA, true);
1411 else
1412 goto continue_unlock;
1413 }
1414
1415 BUG_ON(PageWriteback(page));
1416 if (!clear_page_dirty_for_io(page))
1417 goto continue_unlock;
1418
1419 ret = (*writepage)(page, wbc, data);
1420 if (unlikely(ret)) {
1421 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1422 unlock_page(page);
1423 ret = 0;
1424 } else {
1425 done_index = page->index + 1;
1426 done = 1;
1427 break;
1428 }
1429 }
1430
1431 if (--wbc->nr_to_write <= 0 &&
1432 wbc->sync_mode == WB_SYNC_NONE) {
1433 done = 1;
1434 break;
1435 }
1436 }
1437 pagevec_release(&pvec);
1438 cond_resched();
1439 }
1440
1441 if (step < 1) {
1442 step++;
1443 goto next;
1444 }
1445
1446 if (!cycled && !done) {
1447 cycled = 1;
1448 index = 0;
1449 end = writeback_index - 1;
1450 goto retry;
1451 }
1452 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1453 mapping->writeback_index = done_index;
1454
1455 return ret;
1456 }
1457
1458 static int f2fs_write_data_pages(struct address_space *mapping,
1459 struct writeback_control *wbc)
1460 {
1461 struct inode *inode = mapping->host;
1462 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1463 bool locked = false;
1464 int ret;
1465 long diff;
1466
1467 /* deal with chardevs and other special file */
1468 if (!mapping->a_ops->writepage)
1469 return 0;
1470
1471 /* skip writing if there is no dirty page in this inode */
1472 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1473 return 0;
1474
1475 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1476 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1477 available_free_memory(sbi, DIRTY_DENTS))
1478 goto skip_write;
1479
1480 /* skip writing during file defragment */
1481 if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1482 goto skip_write;
1483
1484 /* during POR, we don't need to trigger writepage at all. */
1485 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1486 goto skip_write;
1487
1488 trace_f2fs_writepages(mapping->host, wbc, DATA);
1489
1490 diff = nr_pages_to_write(sbi, DATA, wbc);
1491
1492 if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1493 mutex_lock(&sbi->writepages);
1494 locked = true;
1495 }
1496 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1497 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1498 if (locked)
1499 mutex_unlock(&sbi->writepages);
1500
1501 remove_dirty_inode(inode);
1502
1503 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1504 return ret;
1505
1506 skip_write:
1507 wbc->pages_skipped += get_dirty_pages(inode);
1508 trace_f2fs_writepages(mapping->host, wbc, DATA);
1509 return 0;
1510 }
1511
1512 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1513 {
1514 struct inode *inode = mapping->host;
1515 loff_t i_size = i_size_read(inode);
1516
1517 if (to > i_size) {
1518 truncate_pagecache(inode, i_size);
1519 truncate_blocks(inode, i_size, true);
1520 }
1521 }
1522
1523 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1524 struct page *page, loff_t pos, unsigned len,
1525 block_t *blk_addr, bool *node_changed)
1526 {
1527 struct inode *inode = page->mapping->host;
1528 pgoff_t index = page->index;
1529 struct dnode_of_data dn;
1530 struct page *ipage;
1531 bool locked = false;
1532 struct extent_info ei;
1533 int err = 0;
1534
1535 /*
1536 * we already allocated all the blocks, so we don't need to get
1537 * the block addresses when there is no need to fill the page.
1538 */
1539 if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1540 len == PAGE_SIZE)
1541 return 0;
1542
1543 if (f2fs_has_inline_data(inode) ||
1544 (pos & PAGE_MASK) >= i_size_read(inode)) {
1545 f2fs_lock_op(sbi);
1546 locked = true;
1547 }
1548 restart:
1549 /* check inline_data */
1550 ipage = get_node_page(sbi, inode->i_ino);
1551 if (IS_ERR(ipage)) {
1552 err = PTR_ERR(ipage);
1553 goto unlock_out;
1554 }
1555
1556 set_new_dnode(&dn, inode, ipage, ipage, 0);
1557
1558 if (f2fs_has_inline_data(inode)) {
1559 if (pos + len <= MAX_INLINE_DATA) {
1560 read_inline_data(page, ipage);
1561 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1562 if (inode->i_nlink)
1563 set_inline_node(ipage);
1564 } else {
1565 err = f2fs_convert_inline_page(&dn, page);
1566 if (err)
1567 goto out;
1568 if (dn.data_blkaddr == NULL_ADDR)
1569 err = f2fs_get_block(&dn, index);
1570 }
1571 } else if (locked) {
1572 err = f2fs_get_block(&dn, index);
1573 } else {
1574 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1575 dn.data_blkaddr = ei.blk + index - ei.fofs;
1576 } else {
1577 /* hole case */
1578 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1579 if (err || dn.data_blkaddr == NULL_ADDR) {
1580 f2fs_put_dnode(&dn);
1581 f2fs_lock_op(sbi);
1582 locked = true;
1583 goto restart;
1584 }
1585 }
1586 }
1587
1588 /* convert_inline_page can make node_changed */
1589 *blk_addr = dn.data_blkaddr;
1590 *node_changed = dn.node_changed;
1591 out:
1592 f2fs_put_dnode(&dn);
1593 unlock_out:
1594 if (locked)
1595 f2fs_unlock_op(sbi);
1596 return err;
1597 }
1598
1599 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1600 loff_t pos, unsigned len, unsigned flags,
1601 struct page **pagep, void **fsdata)
1602 {
1603 struct inode *inode = mapping->host;
1604 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1605 struct page *page = NULL;
1606 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1607 bool need_balance = false;
1608 block_t blkaddr = NULL_ADDR;
1609 int err = 0;
1610
1611 trace_f2fs_write_begin(inode, pos, len, flags);
1612
1613 /*
1614 * We should check this at this moment to avoid deadlock on inode page
1615 * and #0 page. The locking rule for inline_data conversion should be:
1616 * lock_page(page #0) -> lock_page(inode_page)
1617 */
1618 if (index != 0) {
1619 err = f2fs_convert_inline_inode(inode);
1620 if (err)
1621 goto fail;
1622 }
1623 repeat:
1624 page = grab_cache_page_write_begin(mapping, index, flags);
1625 if (!page) {
1626 err = -ENOMEM;
1627 goto fail;
1628 }
1629
1630 *pagep = page;
1631
1632 err = prepare_write_begin(sbi, page, pos, len,
1633 &blkaddr, &need_balance);
1634 if (err)
1635 goto fail;
1636
1637 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1638 unlock_page(page);
1639 f2fs_balance_fs(sbi, true);
1640 lock_page(page);
1641 if (page->mapping != mapping) {
1642 /* The page got truncated from under us */
1643 f2fs_put_page(page, 1);
1644 goto repeat;
1645 }
1646 }
1647
1648 f2fs_wait_on_page_writeback(page, DATA, false);
1649
1650 /* wait for GCed encrypted page writeback */
1651 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1652 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1653
1654 if (len == PAGE_SIZE)
1655 goto out_update;
1656 if (PageUptodate(page))
1657 goto out_clear;
1658
1659 if ((pos & PAGE_MASK) >= i_size_read(inode)) {
1660 unsigned start = pos & (PAGE_SIZE - 1);
1661 unsigned end = start + len;
1662
1663 /* Reading beyond i_size is simple: memset to zero */
1664 zero_user_segments(page, 0, start, end, PAGE_SIZE);
1665 goto out_update;
1666 }
1667
1668 if (blkaddr == NEW_ADDR) {
1669 zero_user_segment(page, 0, PAGE_SIZE);
1670 } else {
1671 struct f2fs_io_info fio = {
1672 .sbi = sbi,
1673 .type = DATA,
1674 .op = REQ_OP_READ,
1675 .op_flags = READ_SYNC,
1676 .old_blkaddr = blkaddr,
1677 .new_blkaddr = blkaddr,
1678 .page = page,
1679 .encrypted_page = NULL,
1680 };
1681 err = f2fs_submit_page_bio(&fio);
1682 if (err)
1683 goto fail;
1684
1685 lock_page(page);
1686 if (unlikely(!PageUptodate(page))) {
1687 err = -EIO;
1688 goto fail;
1689 }
1690 if (unlikely(page->mapping != mapping)) {
1691 f2fs_put_page(page, 1);
1692 goto repeat;
1693 }
1694
1695 /* avoid symlink page */
1696 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1697 err = fscrypt_decrypt_page(page);
1698 if (err)
1699 goto fail;
1700 }
1701 }
1702 out_update:
1703 SetPageUptodate(page);
1704 out_clear:
1705 clear_cold_data(page);
1706 return 0;
1707
1708 fail:
1709 f2fs_put_page(page, 1);
1710 f2fs_write_failed(mapping, pos + len);
1711 return err;
1712 }
1713
1714 static int f2fs_write_end(struct file *file,
1715 struct address_space *mapping,
1716 loff_t pos, unsigned len, unsigned copied,
1717 struct page *page, void *fsdata)
1718 {
1719 struct inode *inode = page->mapping->host;
1720
1721 trace_f2fs_write_end(inode, pos, len, copied);
1722
1723 set_page_dirty(page);
1724
1725 if (pos + copied > i_size_read(inode)) {
1726 i_size_write(inode, pos + copied);
1727 mark_inode_dirty(inode);
1728 }
1729
1730 f2fs_put_page(page, 1);
1731 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1732 return copied;
1733 }
1734
1735 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1736 loff_t offset)
1737 {
1738 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1739
1740 if (offset & blocksize_mask)
1741 return -EINVAL;
1742
1743 if (iov_iter_alignment(iter) & blocksize_mask)
1744 return -EINVAL;
1745
1746 return 0;
1747 }
1748
1749 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1750 {
1751 struct address_space *mapping = iocb->ki_filp->f_mapping;
1752 struct inode *inode = mapping->host;
1753 size_t count = iov_iter_count(iter);
1754 loff_t offset = iocb->ki_pos;
1755 int err;
1756
1757 err = check_direct_IO(inode, iter, offset);
1758 if (err)
1759 return err;
1760
1761 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1762 return 0;
1763
1764 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1765
1766 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1767 if (iov_iter_rw(iter) == WRITE) {
1768 if (err > 0)
1769 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1770 else if (err < 0)
1771 f2fs_write_failed(mapping, offset + count);
1772 }
1773
1774 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1775
1776 return err;
1777 }
1778
1779 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1780 unsigned int length)
1781 {
1782 struct inode *inode = page->mapping->host;
1783 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1784
1785 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1786 (offset % PAGE_SIZE || length != PAGE_SIZE))
1787 return;
1788
1789 if (PageDirty(page)) {
1790 if (inode->i_ino == F2FS_META_INO(sbi))
1791 dec_page_count(sbi, F2FS_DIRTY_META);
1792 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1793 dec_page_count(sbi, F2FS_DIRTY_NODES);
1794 else
1795 inode_dec_dirty_pages(inode);
1796 }
1797
1798 /* This is atomic written page, keep Private */
1799 if (IS_ATOMIC_WRITTEN_PAGE(page))
1800 return;
1801
1802 set_page_private(page, 0);
1803 ClearPagePrivate(page);
1804 }
1805
1806 int f2fs_release_page(struct page *page, gfp_t wait)
1807 {
1808 /* If this is dirty page, keep PagePrivate */
1809 if (PageDirty(page))
1810 return 0;
1811
1812 /* This is atomic written page, keep Private */
1813 if (IS_ATOMIC_WRITTEN_PAGE(page))
1814 return 0;
1815
1816 set_page_private(page, 0);
1817 ClearPagePrivate(page);
1818 return 1;
1819 }
1820
1821 static int f2fs_set_data_page_dirty(struct page *page)
1822 {
1823 struct address_space *mapping = page->mapping;
1824 struct inode *inode = mapping->host;
1825
1826 trace_f2fs_set_page_dirty(page, DATA);
1827
1828 SetPageUptodate(page);
1829
1830 if (f2fs_is_atomic_file(inode)) {
1831 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1832 register_inmem_page(inode, page);
1833 return 1;
1834 }
1835 /*
1836 * Previously, this page has been registered, we just
1837 * return here.
1838 */
1839 return 0;
1840 }
1841
1842 if (!PageDirty(page)) {
1843 __set_page_dirty_nobuffers(page);
1844 update_dirty_page(inode, page);
1845 return 1;
1846 }
1847 return 0;
1848 }
1849
1850 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1851 {
1852 struct inode *inode = mapping->host;
1853
1854 if (f2fs_has_inline_data(inode))
1855 return 0;
1856
1857 /* make sure allocating whole blocks */
1858 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1859 filemap_write_and_wait(mapping);
1860
1861 return generic_block_bmap(mapping, block, get_data_block_bmap);
1862 }
1863
1864 const struct address_space_operations f2fs_dblock_aops = {
1865 .readpage = f2fs_read_data_page,
1866 .readpages = f2fs_read_data_pages,
1867 .writepage = f2fs_write_data_page,
1868 .writepages = f2fs_write_data_pages,
1869 .write_begin = f2fs_write_begin,
1870 .write_end = f2fs_write_end,
1871 .set_page_dirty = f2fs_set_data_page_dirty,
1872 .invalidatepage = f2fs_invalidate_page,
1873 .releasepage = f2fs_release_page,
1874 .direct_IO = f2fs_direct_IO,
1875 .bmap = f2fs_bmap,
1876 };
Ubuntu Artful kernel mirror