projects / deliverable / linux.git / blob
? search:


7fff6ac3d0cb90fce727e788bdee866c448eb7e4
[deliverable/linux.git] / fs / f2fs / data.c
1 /*
2 * fs/f2fs/data.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
22 #include <linux/cleancache.h>
23
24 #include "f2fs.h"
25 #include "node.h"
26 #include "segment.h"
27 #include "trace.h"
28 #include <trace/events/f2fs.h>
29
30 static void f2fs_read_end_io(struct bio *bio)
31 {
32 struct bio_vec *bvec;
33 int i;
34
35 if (f2fs_bio_encrypted(bio)) {
36 if (bio->bi_error) {
37 f2fs_release_crypto_ctx(bio->bi_private);
38 } else {
39 f2fs_end_io_crypto_work(bio->bi_private, bio);
40 return;
41 }
42 }
43
44 bio_for_each_segment_all(bvec, bio, i) {
45 struct page *page = bvec->bv_page;
46
47 if (!bio->bi_error) {
48 SetPageUptodate(page);
49 } else {
50 ClearPageUptodate(page);
51 SetPageError(page);
52 }
53 unlock_page(page);
54 }
55 bio_put(bio);
56 }
57
58 static void f2fs_write_end_io(struct bio *bio)
59 {
60 struct f2fs_sb_info *sbi = bio->bi_private;
61 struct bio_vec *bvec;
62 int i;
63
64 bio_for_each_segment_all(bvec, bio, i) {
65 struct page *page = bvec->bv_page;
66
67 f2fs_restore_and_release_control_page(&page);
68
69 if (unlikely(bio->bi_error)) {
70 set_page_dirty(page);
71 set_bit(AS_EIO, &page->mapping->flags);
72 f2fs_stop_checkpoint(sbi);
73 }
74 end_page_writeback(page);
75 dec_page_count(sbi, F2FS_WRITEBACK);
76 }
77
78 if (!get_pages(sbi, F2FS_WRITEBACK) &&
79 !list_empty(&sbi->cp_wait.task_list))
80 wake_up(&sbi->cp_wait);
81
82 bio_put(bio);
83 }
84
85 /*
86 * Low-level block read/write IO operations.
87 */
88 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
89 int npages, bool is_read)
90 {
91 struct bio *bio;
92
93 bio = f2fs_bio_alloc(npages);
94
95 bio->bi_bdev = sbi->sb->s_bdev;
96 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
97 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
98 bio->bi_private = is_read ? NULL : sbi;
99
100 return bio;
101 }
102
103 static void __submit_merged_bio(struct f2fs_bio_info *io)
104 {
105 struct f2fs_io_info *fio = &io->fio;
106
107 if (!io->bio)
108 return;
109
110 if (is_read_io(fio->rw))
111 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
112 else
113 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
114
115 submit_bio(fio->rw, io->bio);
116 io->bio = NULL;
117 }
118
119 bool is_merged_page(struct f2fs_sb_info *sbi, struct page *page,
120 enum page_type type)
121 {
122 enum page_type btype = PAGE_TYPE_OF_BIO(type);
123 struct f2fs_bio_info *io = &sbi->write_io[btype];
124 struct bio_vec *bvec;
125 struct page *target;
126 int i;
127
128 down_read(&io->io_rwsem);
129 if (!io->bio) {
130 up_read(&io->io_rwsem);
131 return false;
132 }
133
134 bio_for_each_segment_all(bvec, io->bio, i) {
135
136 if (bvec->bv_page->mapping) {
137 target = bvec->bv_page;
138 } else {
139 struct f2fs_crypto_ctx *ctx;
140
141 /* encrypted page */
142 ctx = (struct f2fs_crypto_ctx *)page_private(
143 bvec->bv_page);
144 target = ctx->w.control_page;
145 }
146
147 if (page == target) {
148 up_read(&io->io_rwsem);
149 return true;
150 }
151 }
152
153 up_read(&io->io_rwsem);
154 return false;
155 }
156
157 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
158 enum page_type type, int rw)
159 {
160 enum page_type btype = PAGE_TYPE_OF_BIO(type);
161 struct f2fs_bio_info *io;
162
163 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
164
165 down_write(&io->io_rwsem);
166
167 /* change META to META_FLUSH in the checkpoint procedure */
168 if (type >= META_FLUSH) {
169 io->fio.type = META_FLUSH;
170 if (test_opt(sbi, NOBARRIER))
171 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
172 else
173 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
174 }
175 __submit_merged_bio(io);
176 up_write(&io->io_rwsem);
177 }
178
179 /*
180 * Fill the locked page with data located in the block address.
181 * Return unlocked page.
182 */
183 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
184 {
185 struct bio *bio;
186 struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
187
188 trace_f2fs_submit_page_bio(page, fio);
189 f2fs_trace_ios(fio, 0);
190
191 /* Allocate a new bio */
192 bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
193
194 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
195 bio_put(bio);
196 return -EFAULT;
197 }
198
199 submit_bio(fio->rw, bio);
200 return 0;
201 }
202
203 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
204 {
205 struct f2fs_sb_info *sbi = fio->sbi;
206 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
207 struct f2fs_bio_info *io;
208 bool is_read = is_read_io(fio->rw);
209 struct page *bio_page;
210
211 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
212
213 verify_block_addr(sbi, fio->blk_addr);
214
215 down_write(&io->io_rwsem);
216
217 if (!is_read)
218 inc_page_count(sbi, F2FS_WRITEBACK);
219
220 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
221 io->fio.rw != fio->rw))
222 __submit_merged_bio(io);
223 alloc_new:
224 if (io->bio == NULL) {
225 int bio_blocks = MAX_BIO_BLOCKS(sbi);
226
227 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
228 io->fio = *fio;
229 }
230
231 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
232
233 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
234 PAGE_CACHE_SIZE) {
235 __submit_merged_bio(io);
236 goto alloc_new;
237 }
238
239 io->last_block_in_bio = fio->blk_addr;
240 f2fs_trace_ios(fio, 0);
241
242 up_write(&io->io_rwsem);
243 trace_f2fs_submit_page_mbio(fio->page, fio);
244 }
245
246 /*
247 * Lock ordering for the change of data block address:
248 * ->data_page
249 * ->node_page
250 * update block addresses in the node page
251 */
252 void set_data_blkaddr(struct dnode_of_data *dn)
253 {
254 struct f2fs_node *rn;
255 __le32 *addr_array;
256 struct page *node_page = dn->node_page;
257 unsigned int ofs_in_node = dn->ofs_in_node;
258
259 f2fs_wait_on_page_writeback(node_page, NODE);
260
261 rn = F2FS_NODE(node_page);
262
263 /* Get physical address of data block */
264 addr_array = blkaddr_in_node(rn);
265 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
266 if (set_page_dirty(node_page))
267 dn->node_changed = true;
268 }
269
270 int reserve_new_block(struct dnode_of_data *dn)
271 {
272 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
273
274 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
275 return -EPERM;
276 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
277 return -ENOSPC;
278
279 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
280
281 dn->data_blkaddr = NEW_ADDR;
282 set_data_blkaddr(dn);
283 mark_inode_dirty(dn->inode);
284 sync_inode_page(dn);
285 return 0;
286 }
287
288 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
289 {
290 bool need_put = dn->inode_page ? false : true;
291 int err;
292
293 err = get_dnode_of_data(dn, index, ALLOC_NODE);
294 if (err)
295 return err;
296
297 if (dn->data_blkaddr == NULL_ADDR)
298 err = reserve_new_block(dn);
299 if (err || need_put)
300 f2fs_put_dnode(dn);
301 return err;
302 }
303
304 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
305 {
306 struct extent_info ei;
307 struct inode *inode = dn->inode;
308
309 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
310 dn->data_blkaddr = ei.blk + index - ei.fofs;
311 return 0;
312 }
313
314 return f2fs_reserve_block(dn, index);
315 }
316
317 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
318 int rw, bool for_write)
319 {
320 struct address_space *mapping = inode->i_mapping;
321 struct dnode_of_data dn;
322 struct page *page;
323 struct extent_info ei;
324 int err;
325 struct f2fs_io_info fio = {
326 .sbi = F2FS_I_SB(inode),
327 .type = DATA,
328 .rw = rw,
329 .encrypted_page = NULL,
330 };
331
332 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
333 return read_mapping_page(mapping, index, NULL);
334
335 page = f2fs_grab_cache_page(mapping, index, for_write);
336 if (!page)
337 return ERR_PTR(-ENOMEM);
338
339 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
340 dn.data_blkaddr = ei.blk + index - ei.fofs;
341 goto got_it;
342 }
343
344 set_new_dnode(&dn, inode, NULL, NULL, 0);
345 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
346 if (err)
347 goto put_err;
348 f2fs_put_dnode(&dn);
349
350 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
351 err = -ENOENT;
352 goto put_err;
353 }
354 got_it:
355 if (PageUptodate(page)) {
356 unlock_page(page);
357 return page;
358 }
359
360 /*
361 * A new dentry page is allocated but not able to be written, since its
362 * new inode page couldn't be allocated due to -ENOSPC.
363 * In such the case, its blkaddr can be remained as NEW_ADDR.
364 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
365 */
366 if (dn.data_blkaddr == NEW_ADDR) {
367 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
368 SetPageUptodate(page);
369 unlock_page(page);
370 return page;
371 }
372
373 fio.blk_addr = dn.data_blkaddr;
374 fio.page = page;
375 err = f2fs_submit_page_bio(&fio);
376 if (err)
377 goto put_err;
378 return page;
379
380 put_err:
381 f2fs_put_page(page, 1);
382 return ERR_PTR(err);
383 }
384
385 struct page *find_data_page(struct inode *inode, pgoff_t index)
386 {
387 struct address_space *mapping = inode->i_mapping;
388 struct page *page;
389
390 page = find_get_page(mapping, index);
391 if (page && PageUptodate(page))
392 return page;
393 f2fs_put_page(page, 0);
394
395 page = get_read_data_page(inode, index, READ_SYNC, false);
396 if (IS_ERR(page))
397 return page;
398
399 if (PageUptodate(page))
400 return page;
401
402 wait_on_page_locked(page);
403 if (unlikely(!PageUptodate(page))) {
404 f2fs_put_page(page, 0);
405 return ERR_PTR(-EIO);
406 }
407 return page;
408 }
409
410 /*
411 * If it tries to access a hole, return an error.
412 * Because, the callers, functions in dir.c and GC, should be able to know
413 * whether this page exists or not.
414 */
415 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
416 bool for_write)
417 {
418 struct address_space *mapping = inode->i_mapping;
419 struct page *page;
420 repeat:
421 page = get_read_data_page(inode, index, READ_SYNC, for_write);
422 if (IS_ERR(page))
423 return page;
424
425 /* wait for read completion */
426 lock_page(page);
427 if (unlikely(!PageUptodate(page))) {
428 f2fs_put_page(page, 1);
429 return ERR_PTR(-EIO);
430 }
431 if (unlikely(page->mapping != mapping)) {
432 f2fs_put_page(page, 1);
433 goto repeat;
434 }
435 return page;
436 }
437
438 /*
439 * Caller ensures that this data page is never allocated.
440 * A new zero-filled data page is allocated in the page cache.
441 *
442 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
443 * f2fs_unlock_op().
444 * Note that, ipage is set only by make_empty_dir, and if any error occur,
445 * ipage should be released by this function.
446 */
447 struct page *get_new_data_page(struct inode *inode,
448 struct page *ipage, pgoff_t index, bool new_i_size)
449 {
450 struct address_space *mapping = inode->i_mapping;
451 struct page *page;
452 struct dnode_of_data dn;
453 int err;
454
455 page = f2fs_grab_cache_page(mapping, index, true);
456 if (!page) {
457 /*
458 * before exiting, we should make sure ipage will be released
459 * if any error occur.
460 */
461 f2fs_put_page(ipage, 1);
462 return ERR_PTR(-ENOMEM);
463 }
464
465 set_new_dnode(&dn, inode, ipage, NULL, 0);
466 err = f2fs_reserve_block(&dn, index);
467 if (err) {
468 f2fs_put_page(page, 1);
469 return ERR_PTR(err);
470 }
471 if (!ipage)
472 f2fs_put_dnode(&dn);
473
474 if (PageUptodate(page))
475 goto got_it;
476
477 if (dn.data_blkaddr == NEW_ADDR) {
478 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
479 SetPageUptodate(page);
480 } else {
481 f2fs_put_page(page, 1);
482
483 /* if ipage exists, blkaddr should be NEW_ADDR */
484 f2fs_bug_on(F2FS_I_SB(inode), ipage);
485 page = get_lock_data_page(inode, index, true);
486 if (IS_ERR(page))
487 return page;
488 }
489 got_it:
490 if (new_i_size && i_size_read(inode) <
491 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
492 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
493 /* Only the directory inode sets new_i_size */
494 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
495 }
496 return page;
497 }
498
499 static int __allocate_data_block(struct dnode_of_data *dn)
500 {
501 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
502 struct f2fs_inode_info *fi = F2FS_I(dn->inode);
503 struct f2fs_summary sum;
504 struct node_info ni;
505 int seg = CURSEG_WARM_DATA;
506 pgoff_t fofs;
507
508 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
509 return -EPERM;
510
511 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
512 if (dn->data_blkaddr == NEW_ADDR)
513 goto alloc;
514
515 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
516 return -ENOSPC;
517
518 alloc:
519 get_node_info(sbi, dn->nid, &ni);
520 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
521
522 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
523 seg = CURSEG_DIRECT_IO;
524
525 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
526 &sum, seg);
527 set_data_blkaddr(dn);
528
529 /* update i_size */
530 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), fi) +
531 dn->ofs_in_node;
532 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
533 i_size_write(dn->inode,
534 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
535 return 0;
536 }
537
538 static int __allocate_data_blocks(struct inode *inode, loff_t offset,
539 size_t count)
540 {
541 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
542 struct dnode_of_data dn;
543 u64 start = F2FS_BYTES_TO_BLK(offset);
544 u64 len = F2FS_BYTES_TO_BLK(count);
545 bool allocated;
546 u64 end_offset;
547 int err = 0;
548
549 while (len) {
550 f2fs_lock_op(sbi);
551
552 /* When reading holes, we need its node page */
553 set_new_dnode(&dn, inode, NULL, NULL, 0);
554 err = get_dnode_of_data(&dn, start, ALLOC_NODE);
555 if (err)
556 goto out;
557
558 allocated = false;
559 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
560
561 while (dn.ofs_in_node < end_offset && len) {
562 block_t blkaddr;
563
564 if (unlikely(f2fs_cp_error(sbi))) {
565 err = -EIO;
566 goto sync_out;
567 }
568
569 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
570 if (blkaddr == NULL_ADDR || blkaddr == NEW_ADDR) {
571 err = __allocate_data_block(&dn);
572 if (err)
573 goto sync_out;
574 allocated = true;
575 }
576 len--;
577 start++;
578 dn.ofs_in_node++;
579 }
580
581 if (allocated)
582 sync_inode_page(&dn);
583
584 f2fs_put_dnode(&dn);
585 f2fs_unlock_op(sbi);
586
587 f2fs_balance_fs(sbi, dn.node_changed);
588 }
589 return err;
590
591 sync_out:
592 if (allocated)
593 sync_inode_page(&dn);
594 f2fs_put_dnode(&dn);
595 out:
596 f2fs_unlock_op(sbi);
597 f2fs_balance_fs(sbi, dn.node_changed);
598 return err;
599 }
600
601 /*
602 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
603 * f2fs_map_blocks structure.
604 * If original data blocks are allocated, then give them to blockdev.
605 * Otherwise,
606 * a. preallocate requested block addresses
607 * b. do not use extent cache for better performance
608 * c. give the block addresses to blockdev
609 */
610 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
611 int create, int flag)
612 {
613 unsigned int maxblocks = map->m_len;
614 struct dnode_of_data dn;
615 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
616 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
617 pgoff_t pgofs, end_offset;
618 int err = 0, ofs = 1;
619 struct extent_info ei;
620 bool allocated = false;
621 block_t blkaddr;
622
623 map->m_len = 0;
624 map->m_flags = 0;
625
626 /* it only supports block size == page size */
627 pgofs = (pgoff_t)map->m_lblk;
628
629 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
630 map->m_pblk = ei.blk + pgofs - ei.fofs;
631 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
632 map->m_flags = F2FS_MAP_MAPPED;
633 goto out;
634 }
635
636 if (create)
637 f2fs_lock_op(sbi);
638
639 /* When reading holes, we need its node page */
640 set_new_dnode(&dn, inode, NULL, NULL, 0);
641 err = get_dnode_of_data(&dn, pgofs, mode);
642 if (err) {
643 if (err == -ENOENT)
644 err = 0;
645 goto unlock_out;
646 }
647
648 if (dn.data_blkaddr == NEW_ADDR || dn.data_blkaddr == NULL_ADDR) {
649 if (create) {
650 if (unlikely(f2fs_cp_error(sbi))) {
651 err = -EIO;
652 goto put_out;
653 }
654 err = __allocate_data_block(&dn);
655 if (err)
656 goto put_out;
657 allocated = true;
658 map->m_flags = F2FS_MAP_NEW;
659 } else {
660 if (flag != F2FS_GET_BLOCK_FIEMAP ||
661 dn.data_blkaddr != NEW_ADDR) {
662 if (flag == F2FS_GET_BLOCK_BMAP)
663 err = -ENOENT;
664 goto put_out;
665 }
666
667 /*
668 * preallocated unwritten block should be mapped
669 * for fiemap.
670 */
671 if (dn.data_blkaddr == NEW_ADDR)
672 map->m_flags = F2FS_MAP_UNWRITTEN;
673 }
674 }
675
676 map->m_flags |= F2FS_MAP_MAPPED;
677 map->m_pblk = dn.data_blkaddr;
678 map->m_len = 1;
679
680 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
681 dn.ofs_in_node++;
682 pgofs++;
683
684 get_next:
685 if (map->m_len >= maxblocks)
686 goto sync_out;
687
688 if (dn.ofs_in_node >= end_offset) {
689 if (allocated)
690 sync_inode_page(&dn);
691 allocated = false;
692 f2fs_put_dnode(&dn);
693
694 if (create) {
695 f2fs_unlock_op(sbi);
696 f2fs_balance_fs(sbi, dn.node_changed);
697 f2fs_lock_op(sbi);
698 }
699
700 set_new_dnode(&dn, inode, NULL, NULL, 0);
701 err = get_dnode_of_data(&dn, pgofs, mode);
702 if (err) {
703 if (err == -ENOENT)
704 err = 0;
705 goto unlock_out;
706 }
707
708 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
709 }
710
711 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
712
713 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
714 if (create) {
715 if (unlikely(f2fs_cp_error(sbi))) {
716 err = -EIO;
717 goto sync_out;
718 }
719 err = __allocate_data_block(&dn);
720 if (err)
721 goto sync_out;
722 allocated = true;
723 map->m_flags |= F2FS_MAP_NEW;
724 blkaddr = dn.data_blkaddr;
725 } else {
726 /*
727 * we only merge preallocated unwritten blocks
728 * for fiemap.
729 */
730 if (flag != F2FS_GET_BLOCK_FIEMAP ||
731 blkaddr != NEW_ADDR)
732 goto sync_out;
733 }
734 }
735
736 /* Give more consecutive addresses for the readahead */
737 if ((map->m_pblk != NEW_ADDR &&
738 blkaddr == (map->m_pblk + ofs)) ||
739 (map->m_pblk == NEW_ADDR &&
740 blkaddr == NEW_ADDR)) {
741 ofs++;
742 dn.ofs_in_node++;
743 pgofs++;
744 map->m_len++;
745 goto get_next;
746 }
747
748 sync_out:
749 if (allocated)
750 sync_inode_page(&dn);
751 put_out:
752 f2fs_put_dnode(&dn);
753 unlock_out:
754 if (create) {
755 f2fs_unlock_op(sbi);
756 f2fs_balance_fs(sbi, dn.node_changed);
757 }
758 out:
759 trace_f2fs_map_blocks(inode, map, err);
760 return err;
761 }
762
763 static int __get_data_block(struct inode *inode, sector_t iblock,
764 struct buffer_head *bh, int create, int flag)
765 {
766 struct f2fs_map_blocks map;
767 int ret;
768
769 map.m_lblk = iblock;
770 map.m_len = bh->b_size >> inode->i_blkbits;
771
772 ret = f2fs_map_blocks(inode, &map, create, flag);
773 if (!ret) {
774 map_bh(bh, inode->i_sb, map.m_pblk);
775 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
776 bh->b_size = map.m_len << inode->i_blkbits;
777 }
778 return ret;
779 }
780
781 static int get_data_block(struct inode *inode, sector_t iblock,
782 struct buffer_head *bh_result, int create, int flag)
783 {
784 return __get_data_block(inode, iblock, bh_result, create, flag);
785 }
786
787 static int get_data_block_dio(struct inode *inode, sector_t iblock,
788 struct buffer_head *bh_result, int create)
789 {
790 return __get_data_block(inode, iblock, bh_result, create,
791 F2FS_GET_BLOCK_DIO);
792 }
793
794 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
795 struct buffer_head *bh_result, int create)
796 {
797 /* Block number less than F2FS MAX BLOCKS */
798 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
799 return -EFBIG;
800
801 return __get_data_block(inode, iblock, bh_result, create,
802 F2FS_GET_BLOCK_BMAP);
803 }
804
805 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
806 {
807 return (offset >> inode->i_blkbits);
808 }
809
810 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
811 {
812 return (blk << inode->i_blkbits);
813 }
814
815 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
816 u64 start, u64 len)
817 {
818 struct buffer_head map_bh;
819 sector_t start_blk, last_blk;
820 loff_t isize;
821 u64 logical = 0, phys = 0, size = 0;
822 u32 flags = 0;
823 int ret = 0;
824
825 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
826 if (ret)
827 return ret;
828
829 if (f2fs_has_inline_data(inode)) {
830 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
831 if (ret != -EAGAIN)
832 return ret;
833 }
834
835 inode_lock(inode);
836
837 isize = i_size_read(inode);
838 if (start >= isize)
839 goto out;
840
841 if (start + len > isize)
842 len = isize - start;
843
844 if (logical_to_blk(inode, len) == 0)
845 len = blk_to_logical(inode, 1);
846
847 start_blk = logical_to_blk(inode, start);
848 last_blk = logical_to_blk(inode, start + len - 1);
849
850 next:
851 memset(&map_bh, 0, sizeof(struct buffer_head));
852 map_bh.b_size = len;
853
854 ret = get_data_block(inode, start_blk, &map_bh, 0,
855 F2FS_GET_BLOCK_FIEMAP);
856 if (ret)
857 goto out;
858
859 /* HOLE */
860 if (!buffer_mapped(&map_bh)) {
861 /* Go through holes util pass the EOF */
862 if (blk_to_logical(inode, start_blk++) < isize)
863 goto prep_next;
864 /* Found a hole beyond isize means no more extents.
865 * Note that the premise is that filesystems don't
866 * punch holes beyond isize and keep size unchanged.
867 */
868 flags |= FIEMAP_EXTENT_LAST;
869 }
870
871 if (size) {
872 if (f2fs_encrypted_inode(inode))
873 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
874
875 ret = fiemap_fill_next_extent(fieinfo, logical,
876 phys, size, flags);
877 }
878
879 if (start_blk > last_blk || ret)
880 goto out;
881
882 logical = blk_to_logical(inode, start_blk);
883 phys = blk_to_logical(inode, map_bh.b_blocknr);
884 size = map_bh.b_size;
885 flags = 0;
886 if (buffer_unwritten(&map_bh))
887 flags = FIEMAP_EXTENT_UNWRITTEN;
888
889 start_blk += logical_to_blk(inode, size);
890
891 prep_next:
892 cond_resched();
893 if (fatal_signal_pending(current))
894 ret = -EINTR;
895 else
896 goto next;
897 out:
898 if (ret == 1)
899 ret = 0;
900
901 inode_unlock(inode);
902 return ret;
903 }
904
905 /*
906 * This function was originally taken from fs/mpage.c, and customized for f2fs.
907 * Major change was from block_size == page_size in f2fs by default.
908 */
909 static int f2fs_mpage_readpages(struct address_space *mapping,
910 struct list_head *pages, struct page *page,
911 unsigned nr_pages)
912 {
913 struct bio *bio = NULL;
914 unsigned page_idx;
915 sector_t last_block_in_bio = 0;
916 struct inode *inode = mapping->host;
917 const unsigned blkbits = inode->i_blkbits;
918 const unsigned blocksize = 1 << blkbits;
919 sector_t block_in_file;
920 sector_t last_block;
921 sector_t last_block_in_file;
922 sector_t block_nr;
923 struct block_device *bdev = inode->i_sb->s_bdev;
924 struct f2fs_map_blocks map;
925
926 map.m_pblk = 0;
927 map.m_lblk = 0;
928 map.m_len = 0;
929 map.m_flags = 0;
930
931 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
932
933 prefetchw(&page->flags);
934 if (pages) {
935 page = list_entry(pages->prev, struct page, lru);
936 list_del(&page->lru);
937 if (add_to_page_cache_lru(page, mapping,
938 page->index, GFP_KERNEL))
939 goto next_page;
940 }
941
942 block_in_file = (sector_t)page->index;
943 last_block = block_in_file + nr_pages;
944 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
945 blkbits;
946 if (last_block > last_block_in_file)
947 last_block = last_block_in_file;
948
949 /*
950 * Map blocks using the previous result first.
951 */
952 if ((map.m_flags & F2FS_MAP_MAPPED) &&
953 block_in_file > map.m_lblk &&
954 block_in_file < (map.m_lblk + map.m_len))
955 goto got_it;
956
957 /*
958 * Then do more f2fs_map_blocks() calls until we are
959 * done with this page.
960 */
961 map.m_flags = 0;
962
963 if (block_in_file < last_block) {
964 map.m_lblk = block_in_file;
965 map.m_len = last_block - block_in_file;
966
967 if (f2fs_map_blocks(inode, &map, 0,
968 F2FS_GET_BLOCK_READ))
969 goto set_error_page;
970 }
971 got_it:
972 if ((map.m_flags & F2FS_MAP_MAPPED)) {
973 block_nr = map.m_pblk + block_in_file - map.m_lblk;
974 SetPageMappedToDisk(page);
975
976 if (!PageUptodate(page) && !cleancache_get_page(page)) {
977 SetPageUptodate(page);
978 goto confused;
979 }
980 } else {
981 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
982 SetPageUptodate(page);
983 unlock_page(page);
984 goto next_page;
985 }
986
987 /*
988 * This page will go to BIO. Do we need to send this
989 * BIO off first?
990 */
991 if (bio && (last_block_in_bio != block_nr - 1)) {
992 submit_and_realloc:
993 submit_bio(READ, bio);
994 bio = NULL;
995 }
996 if (bio == NULL) {
997 struct f2fs_crypto_ctx *ctx = NULL;
998
999 if (f2fs_encrypted_inode(inode) &&
1000 S_ISREG(inode->i_mode)) {
1001
1002 ctx = f2fs_get_crypto_ctx(inode);
1003 if (IS_ERR(ctx))
1004 goto set_error_page;
1005
1006 /* wait the page to be moved by cleaning */
1007 f2fs_wait_on_encrypted_page_writeback(
1008 F2FS_I_SB(inode), block_nr);
1009 }
1010
1011 bio = bio_alloc(GFP_KERNEL,
1012 min_t(int, nr_pages, BIO_MAX_PAGES));
1013 if (!bio) {
1014 if (ctx)
1015 f2fs_release_crypto_ctx(ctx);
1016 goto set_error_page;
1017 }
1018 bio->bi_bdev = bdev;
1019 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1020 bio->bi_end_io = f2fs_read_end_io;
1021 bio->bi_private = ctx;
1022 }
1023
1024 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1025 goto submit_and_realloc;
1026
1027 last_block_in_bio = block_nr;
1028 goto next_page;
1029 set_error_page:
1030 SetPageError(page);
1031 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1032 unlock_page(page);
1033 goto next_page;
1034 confused:
1035 if (bio) {
1036 submit_bio(READ, bio);
1037 bio = NULL;
1038 }
1039 unlock_page(page);
1040 next_page:
1041 if (pages)
1042 page_cache_release(page);
1043 }
1044 BUG_ON(pages && !list_empty(pages));
1045 if (bio)
1046 submit_bio(READ, bio);
1047 return 0;
1048 }
1049
1050 static int f2fs_read_data_page(struct file *file, struct page *page)
1051 {
1052 struct inode *inode = page->mapping->host;
1053 int ret = -EAGAIN;
1054
1055 trace_f2fs_readpage(page, DATA);
1056
1057 /* If the file has inline data, try to read it directly */
1058 if (f2fs_has_inline_data(inode))
1059 ret = f2fs_read_inline_data(inode, page);
1060 if (ret == -EAGAIN)
1061 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1062 return ret;
1063 }
1064
1065 static int f2fs_read_data_pages(struct file *file,
1066 struct address_space *mapping,
1067 struct list_head *pages, unsigned nr_pages)
1068 {
1069 struct inode *inode = file->f_mapping->host;
1070 struct page *page = list_entry(pages->prev, struct page, lru);
1071
1072 trace_f2fs_readpages(inode, page, nr_pages);
1073
1074 /* If the file has inline data, skip readpages */
1075 if (f2fs_has_inline_data(inode))
1076 return 0;
1077
1078 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1079 }
1080
1081 int do_write_data_page(struct f2fs_io_info *fio)
1082 {
1083 struct page *page = fio->page;
1084 struct inode *inode = page->mapping->host;
1085 struct dnode_of_data dn;
1086 int err = 0;
1087
1088 set_new_dnode(&dn, inode, NULL, NULL, 0);
1089 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1090 if (err)
1091 return err;
1092
1093 fio->blk_addr = dn.data_blkaddr;
1094
1095 /* This page is already truncated */
1096 if (fio->blk_addr == NULL_ADDR) {
1097 ClearPageUptodate(page);
1098 goto out_writepage;
1099 }
1100
1101 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1102
1103 /* wait for GCed encrypted page writeback */
1104 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1105 fio->blk_addr);
1106
1107 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1108 if (IS_ERR(fio->encrypted_page)) {
1109 err = PTR_ERR(fio->encrypted_page);
1110 goto out_writepage;
1111 }
1112 }
1113
1114 set_page_writeback(page);
1115
1116 /*
1117 * If current allocation needs SSR,
1118 * it had better in-place writes for updated data.
1119 */
1120 if (unlikely(fio->blk_addr != NEW_ADDR &&
1121 !is_cold_data(page) &&
1122 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1123 need_inplace_update(inode))) {
1124 rewrite_data_page(fio);
1125 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1126 trace_f2fs_do_write_data_page(page, IPU);
1127 } else {
1128 write_data_page(&dn, fio);
1129 set_data_blkaddr(&dn);
1130 f2fs_update_extent_cache(&dn);
1131 trace_f2fs_do_write_data_page(page, OPU);
1132 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1133 if (page->index == 0)
1134 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1135 }
1136 out_writepage:
1137 f2fs_put_dnode(&dn);
1138 return err;
1139 }
1140
1141 static int f2fs_write_data_page(struct page *page,
1142 struct writeback_control *wbc)
1143 {
1144 struct inode *inode = page->mapping->host;
1145 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1146 loff_t i_size = i_size_read(inode);
1147 const pgoff_t end_index = ((unsigned long long) i_size)
1148 >> PAGE_CACHE_SHIFT;
1149 unsigned offset = 0;
1150 bool need_balance_fs = false;
1151 int err = 0;
1152 struct f2fs_io_info fio = {
1153 .sbi = sbi,
1154 .type = DATA,
1155 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1156 .page = page,
1157 .encrypted_page = NULL,
1158 };
1159
1160 trace_f2fs_writepage(page, DATA);
1161
1162 if (page->index < end_index)
1163 goto write;
1164
1165 /*
1166 * If the offset is out-of-range of file size,
1167 * this page does not have to be written to disk.
1168 */
1169 offset = i_size & (PAGE_CACHE_SIZE - 1);
1170 if ((page->index >= end_index + 1) || !offset)
1171 goto out;
1172
1173 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1174 write:
1175 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1176 goto redirty_out;
1177 if (f2fs_is_drop_cache(inode))
1178 goto out;
1179 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1180 available_free_memory(sbi, BASE_CHECK))
1181 goto redirty_out;
1182
1183 /* Dentry blocks are controlled by checkpoint */
1184 if (S_ISDIR(inode->i_mode)) {
1185 if (unlikely(f2fs_cp_error(sbi)))
1186 goto redirty_out;
1187 err = do_write_data_page(&fio);
1188 goto done;
1189 }
1190
1191 /* we should bypass data pages to proceed the kworkder jobs */
1192 if (unlikely(f2fs_cp_error(sbi))) {
1193 SetPageError(page);
1194 goto out;
1195 }
1196
1197 if (!wbc->for_reclaim)
1198 need_balance_fs = true;
1199 else if (has_not_enough_free_secs(sbi, 0))
1200 goto redirty_out;
1201
1202 err = -EAGAIN;
1203 f2fs_lock_op(sbi);
1204 if (f2fs_has_inline_data(inode))
1205 err = f2fs_write_inline_data(inode, page);
1206 if (err == -EAGAIN)
1207 err = do_write_data_page(&fio);
1208 f2fs_unlock_op(sbi);
1209 done:
1210 if (err && err != -ENOENT)
1211 goto redirty_out;
1212
1213 clear_cold_data(page);
1214 out:
1215 inode_dec_dirty_pages(inode);
1216 if (err)
1217 ClearPageUptodate(page);
1218 unlock_page(page);
1219 f2fs_balance_fs(sbi, need_balance_fs);
1220 if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi))) {
1221 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1222 remove_dirty_inode(inode);
1223 }
1224 return 0;
1225
1226 redirty_out:
1227 redirty_page_for_writepage(wbc, page);
1228 return AOP_WRITEPAGE_ACTIVATE;
1229 }
1230
1231 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1232 void *data)
1233 {
1234 struct address_space *mapping = data;
1235 int ret = mapping->a_ops->writepage(page, wbc);
1236 mapping_set_error(mapping, ret);
1237 return ret;
1238 }
1239
1240 /*
1241 * This function was copied from write_cche_pages from mm/page-writeback.c.
1242 * The major change is making write step of cold data page separately from
1243 * warm/hot data page.
1244 */
1245 static int f2fs_write_cache_pages(struct address_space *mapping,
1246 struct writeback_control *wbc, writepage_t writepage,
1247 void *data)
1248 {
1249 int ret = 0;
1250 int done = 0;
1251 struct pagevec pvec;
1252 int nr_pages;
1253 pgoff_t uninitialized_var(writeback_index);
1254 pgoff_t index;
1255 pgoff_t end; /* Inclusive */
1256 pgoff_t done_index;
1257 int cycled;
1258 int range_whole = 0;
1259 int tag;
1260 int step = 0;
1261
1262 pagevec_init(&pvec, 0);
1263 next:
1264 if (wbc->range_cyclic) {
1265 writeback_index = mapping->writeback_index; /* prev offset */
1266 index = writeback_index;
1267 if (index == 0)
1268 cycled = 1;
1269 else
1270 cycled = 0;
1271 end = -1;
1272 } else {
1273 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1274 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1275 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1276 range_whole = 1;
1277 cycled = 1; /* ignore range_cyclic tests */
1278 }
1279 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1280 tag = PAGECACHE_TAG_TOWRITE;
1281 else
1282 tag = PAGECACHE_TAG_DIRTY;
1283 retry:
1284 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1285 tag_pages_for_writeback(mapping, index, end);
1286 done_index = index;
1287 while (!done && (index <= end)) {
1288 int i;
1289
1290 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1291 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1292 if (nr_pages == 0)
1293 break;
1294
1295 for (i = 0; i < nr_pages; i++) {
1296 struct page *page = pvec.pages[i];
1297
1298 if (page->index > end) {
1299 done = 1;
1300 break;
1301 }
1302
1303 done_index = page->index;
1304
1305 lock_page(page);
1306
1307 if (unlikely(page->mapping != mapping)) {
1308 continue_unlock:
1309 unlock_page(page);
1310 continue;
1311 }
1312
1313 if (!PageDirty(page)) {
1314 /* someone wrote it for us */
1315 goto continue_unlock;
1316 }
1317
1318 if (step == is_cold_data(page))
1319 goto continue_unlock;
1320
1321 if (PageWriteback(page)) {
1322 if (wbc->sync_mode != WB_SYNC_NONE)
1323 f2fs_wait_on_page_writeback(page, DATA);
1324 else
1325 goto continue_unlock;
1326 }
1327
1328 BUG_ON(PageWriteback(page));
1329 if (!clear_page_dirty_for_io(page))
1330 goto continue_unlock;
1331
1332 ret = (*writepage)(page, wbc, data);
1333 if (unlikely(ret)) {
1334 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1335 unlock_page(page);
1336 ret = 0;
1337 } else {
1338 done_index = page->index + 1;
1339 done = 1;
1340 break;
1341 }
1342 }
1343
1344 if (--wbc->nr_to_write <= 0 &&
1345 wbc->sync_mode == WB_SYNC_NONE) {
1346 done = 1;
1347 break;
1348 }
1349 }
1350 pagevec_release(&pvec);
1351 cond_resched();
1352 }
1353
1354 if (step < 1) {
1355 step++;
1356 goto next;
1357 }
1358
1359 if (!cycled && !done) {
1360 cycled = 1;
1361 index = 0;
1362 end = writeback_index - 1;
1363 goto retry;
1364 }
1365 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1366 mapping->writeback_index = done_index;
1367
1368 return ret;
1369 }
1370
1371 static int f2fs_write_data_pages(struct address_space *mapping,
1372 struct writeback_control *wbc)
1373 {
1374 struct inode *inode = mapping->host;
1375 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1376 bool locked = false;
1377 int ret;
1378 long diff;
1379
1380 trace_f2fs_writepages(mapping->host, wbc, DATA);
1381
1382 /* deal with chardevs and other special file */
1383 if (!mapping->a_ops->writepage)
1384 return 0;
1385
1386 /* skip writing if there is no dirty page in this inode */
1387 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1388 return 0;
1389
1390 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1391 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1392 available_free_memory(sbi, DIRTY_DENTS))
1393 goto skip_write;
1394
1395 /* skip writing during file defragment */
1396 if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1397 goto skip_write;
1398
1399 /* during POR, we don't need to trigger writepage at all. */
1400 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1401 goto skip_write;
1402
1403 diff = nr_pages_to_write(sbi, DATA, wbc);
1404
1405 if (!S_ISDIR(inode->i_mode)) {
1406 mutex_lock(&sbi->writepages);
1407 locked = true;
1408 }
1409 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1410 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1411 if (locked)
1412 mutex_unlock(&sbi->writepages);
1413
1414 remove_dirty_inode(inode);
1415
1416 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1417 return ret;
1418
1419 skip_write:
1420 wbc->pages_skipped += get_dirty_pages(inode);
1421 return 0;
1422 }
1423
1424 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1425 {
1426 struct inode *inode = mapping->host;
1427 loff_t i_size = i_size_read(inode);
1428
1429 if (to > i_size) {
1430 truncate_pagecache(inode, i_size);
1431 truncate_blocks(inode, i_size, true);
1432 }
1433 }
1434
1435 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1436 struct page *page, loff_t pos, unsigned len,
1437 block_t *blk_addr, bool *node_changed)
1438 {
1439 struct inode *inode = page->mapping->host;
1440 pgoff_t index = page->index;
1441 struct dnode_of_data dn;
1442 struct page *ipage;
1443 bool locked = false;
1444 struct extent_info ei;
1445 int err = 0;
1446
1447 if (f2fs_has_inline_data(inode) ||
1448 (pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1449 f2fs_lock_op(sbi);
1450 locked = true;
1451 }
1452 restart:
1453 /* check inline_data */
1454 ipage = get_node_page(sbi, inode->i_ino);
1455 if (IS_ERR(ipage)) {
1456 err = PTR_ERR(ipage);
1457 goto unlock_out;
1458 }
1459
1460 set_new_dnode(&dn, inode, ipage, ipage, 0);
1461
1462 if (f2fs_has_inline_data(inode)) {
1463 if (pos + len <= MAX_INLINE_DATA) {
1464 read_inline_data(page, ipage);
1465 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1466 sync_inode_page(&dn);
1467 } else {
1468 err = f2fs_convert_inline_page(&dn, page);
1469 if (err)
1470 goto out;
1471 if (dn.data_blkaddr == NULL_ADDR)
1472 err = f2fs_get_block(&dn, index);
1473 }
1474 } else if (locked) {
1475 err = f2fs_get_block(&dn, index);
1476 } else {
1477 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1478 dn.data_blkaddr = ei.blk + index - ei.fofs;
1479 } else {
1480 bool restart = false;
1481
1482 /* hole case */
1483 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1484 if (err || (!err && dn.data_blkaddr == NULL_ADDR))
1485 restart = true;
1486 if (restart) {
1487 f2fs_put_dnode(&dn);
1488 f2fs_lock_op(sbi);
1489 locked = true;
1490 goto restart;
1491 }
1492 }
1493 }
1494
1495 /* convert_inline_page can make node_changed */
1496 *blk_addr = dn.data_blkaddr;
1497 *node_changed = dn.node_changed;
1498 out:
1499 f2fs_put_dnode(&dn);
1500 unlock_out:
1501 if (locked)
1502 f2fs_unlock_op(sbi);
1503 return err;
1504 }
1505
1506 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1507 loff_t pos, unsigned len, unsigned flags,
1508 struct page **pagep, void **fsdata)
1509 {
1510 struct inode *inode = mapping->host;
1511 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1512 struct page *page = NULL;
1513 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1514 bool need_balance = false;
1515 block_t blkaddr = NULL_ADDR;
1516 int err = 0;
1517
1518 trace_f2fs_write_begin(inode, pos, len, flags);
1519
1520 /*
1521 * We should check this at this moment to avoid deadlock on inode page
1522 * and #0 page. The locking rule for inline_data conversion should be:
1523 * lock_page(page #0) -> lock_page(inode_page)
1524 */
1525 if (index != 0) {
1526 err = f2fs_convert_inline_inode(inode);
1527 if (err)
1528 goto fail;
1529 }
1530 repeat:
1531 page = grab_cache_page_write_begin(mapping, index, flags);
1532 if (!page) {
1533 err = -ENOMEM;
1534 goto fail;
1535 }
1536
1537 *pagep = page;
1538
1539 err = prepare_write_begin(sbi, page, pos, len,
1540 &blkaddr, &need_balance);
1541 if (err)
1542 goto fail;
1543
1544 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1545 unlock_page(page);
1546 f2fs_balance_fs(sbi, true);
1547 lock_page(page);
1548 if (page->mapping != mapping) {
1549 /* The page got truncated from under us */
1550 f2fs_put_page(page, 1);
1551 goto repeat;
1552 }
1553 }
1554
1555 f2fs_wait_on_page_writeback(page, DATA);
1556
1557 /* wait for GCed encrypted page writeback */
1558 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1559 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1560
1561 if (len == PAGE_CACHE_SIZE)
1562 goto out_update;
1563 if (PageUptodate(page))
1564 goto out_clear;
1565
1566 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1567 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1568 unsigned end = start + len;
1569
1570 /* Reading beyond i_size is simple: memset to zero */
1571 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1572 goto out_update;
1573 }
1574
1575 if (blkaddr == NEW_ADDR) {
1576 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1577 } else {
1578 struct f2fs_io_info fio = {
1579 .sbi = sbi,
1580 .type = DATA,
1581 .rw = READ_SYNC,
1582 .blk_addr = blkaddr,
1583 .page = page,
1584 .encrypted_page = NULL,
1585 };
1586 err = f2fs_submit_page_bio(&fio);
1587 if (err)
1588 goto fail;
1589
1590 lock_page(page);
1591 if (unlikely(!PageUptodate(page))) {
1592 err = -EIO;
1593 goto fail;
1594 }
1595 if (unlikely(page->mapping != mapping)) {
1596 f2fs_put_page(page, 1);
1597 goto repeat;
1598 }
1599
1600 /* avoid symlink page */
1601 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1602 err = f2fs_decrypt_one(inode, page);
1603 if (err)
1604 goto fail;
1605 }
1606 }
1607 out_update:
1608 SetPageUptodate(page);
1609 out_clear:
1610 clear_cold_data(page);
1611 return 0;
1612
1613 fail:
1614 f2fs_put_page(page, 1);
1615 f2fs_write_failed(mapping, pos + len);
1616 return err;
1617 }
1618
1619 static int f2fs_write_end(struct file *file,
1620 struct address_space *mapping,
1621 loff_t pos, unsigned len, unsigned copied,
1622 struct page *page, void *fsdata)
1623 {
1624 struct inode *inode = page->mapping->host;
1625
1626 trace_f2fs_write_end(inode, pos, len, copied);
1627
1628 set_page_dirty(page);
1629
1630 if (pos + copied > i_size_read(inode)) {
1631 i_size_write(inode, pos + copied);
1632 mark_inode_dirty(inode);
1633 update_inode_page(inode);
1634 }
1635
1636 f2fs_put_page(page, 1);
1637 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1638 return copied;
1639 }
1640
1641 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1642 loff_t offset)
1643 {
1644 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1645
1646 if (offset & blocksize_mask)
1647 return -EINVAL;
1648
1649 if (iov_iter_alignment(iter) & blocksize_mask)
1650 return -EINVAL;
1651
1652 return 0;
1653 }
1654
1655 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1656 loff_t offset)
1657 {
1658 struct file *file = iocb->ki_filp;
1659 struct address_space *mapping = file->f_mapping;
1660 struct inode *inode = mapping->host;
1661 size_t count = iov_iter_count(iter);
1662 int err;
1663
1664 /* we don't need to use inline_data strictly */
1665 err = f2fs_convert_inline_inode(inode);
1666 if (err)
1667 return err;
1668
1669 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1670 return 0;
1671
1672 err = check_direct_IO(inode, iter, offset);
1673 if (err)
1674 return err;
1675
1676 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1677
1678 if (iov_iter_rw(iter) == WRITE) {
1679 err = __allocate_data_blocks(inode, offset, count);
1680 if (err)
1681 goto out;
1682 }
1683
1684 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1685 out:
1686 if (err < 0 && iov_iter_rw(iter) == WRITE)
1687 f2fs_write_failed(mapping, offset + count);
1688
1689 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1690
1691 return err;
1692 }
1693
1694 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1695 unsigned int length)
1696 {
1697 struct inode *inode = page->mapping->host;
1698 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1699
1700 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1701 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1702 return;
1703
1704 if (PageDirty(page)) {
1705 if (inode->i_ino == F2FS_META_INO(sbi))
1706 dec_page_count(sbi, F2FS_DIRTY_META);
1707 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1708 dec_page_count(sbi, F2FS_DIRTY_NODES);
1709 else
1710 inode_dec_dirty_pages(inode);
1711 }
1712
1713 /* This is atomic written page, keep Private */
1714 if (IS_ATOMIC_WRITTEN_PAGE(page))
1715 return;
1716
1717 ClearPagePrivate(page);
1718 }
1719
1720 int f2fs_release_page(struct page *page, gfp_t wait)
1721 {
1722 /* If this is dirty page, keep PagePrivate */
1723 if (PageDirty(page))
1724 return 0;
1725
1726 /* This is atomic written page, keep Private */
1727 if (IS_ATOMIC_WRITTEN_PAGE(page))
1728 return 0;
1729
1730 ClearPagePrivate(page);
1731 return 1;
1732 }
1733
1734 static int f2fs_set_data_page_dirty(struct page *page)
1735 {
1736 struct address_space *mapping = page->mapping;
1737 struct inode *inode = mapping->host;
1738
1739 trace_f2fs_set_page_dirty(page, DATA);
1740
1741 SetPageUptodate(page);
1742
1743 if (f2fs_is_atomic_file(inode)) {
1744 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1745 register_inmem_page(inode, page);
1746 return 1;
1747 }
1748 /*
1749 * Previously, this page has been registered, we just
1750 * return here.
1751 */
1752 return 0;
1753 }
1754
1755 if (!PageDirty(page)) {
1756 __set_page_dirty_nobuffers(page);
1757 update_dirty_page(inode, page);
1758 return 1;
1759 }
1760 return 0;
1761 }
1762
1763 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1764 {
1765 struct inode *inode = mapping->host;
1766
1767 if (f2fs_has_inline_data(inode))
1768 return 0;
1769
1770 /* make sure allocating whole blocks */
1771 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1772 filemap_write_and_wait(mapping);
1773
1774 return generic_block_bmap(mapping, block, get_data_block_bmap);
1775 }
1776
1777 const struct address_space_operations f2fs_dblock_aops = {
1778 .readpage = f2fs_read_data_page,
1779 .readpages = f2fs_read_data_pages,
1780 .writepage = f2fs_write_data_page,
1781 .writepages = f2fs_write_data_pages,
1782 .write_begin = f2fs_write_begin,
1783 .write_end = f2fs_write_end,
1784 .set_page_dirty = f2fs_set_data_page_dirty,
1785 .invalidatepage = f2fs_invalidate_page,
1786 .releasepage = f2fs_release_page,
1787 .direct_IO = f2fs_direct_IO,
1788 .bmap = f2fs_bmap,
1789 };
Linux kernel - Deliverables
This page took 0.237565 seconds and 4 git commands to generate.