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


bb60e6afbb722ba09ce3942d9ae1c89578157795
[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_bit(AS_EIO, &page->mapping->flags);
71 f2fs_stop_checkpoint(sbi);
72 }
73 end_page_writeback(page);
74 dec_page_count(sbi, F2FS_WRITEBACK);
75 }
76
77 if (!get_pages(sbi, F2FS_WRITEBACK) && wq_has_sleeper(&sbi->cp_wait))
78 wake_up(&sbi->cp_wait);
79
80 bio_put(bio);
81 }
82
83 /*
84 * Low-level block read/write IO operations.
85 */
86 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
87 int npages, bool is_read)
88 {
89 struct bio *bio;
90
91 bio = f2fs_bio_alloc(npages);
92
93 bio->bi_bdev = sbi->sb->s_bdev;
94 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
95 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
96 bio->bi_private = is_read ? NULL : sbi;
97
98 return bio;
99 }
100
101 static void __submit_merged_bio(struct f2fs_bio_info *io)
102 {
103 struct f2fs_io_info *fio = &io->fio;
104
105 if (!io->bio)
106 return;
107
108 if (is_read_io(fio->rw))
109 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
110 else
111 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
112
113 submit_bio(fio->rw, io->bio);
114 io->bio = NULL;
115 }
116
117 bool is_merged_page(struct f2fs_sb_info *sbi, struct page *page,
118 enum page_type type)
119 {
120 enum page_type btype = PAGE_TYPE_OF_BIO(type);
121 struct f2fs_bio_info *io = &sbi->write_io[btype];
122 struct bio_vec *bvec;
123 struct page *target;
124 int i;
125
126 down_read(&io->io_rwsem);
127 if (!io->bio) {
128 up_read(&io->io_rwsem);
129 return false;
130 }
131
132 bio_for_each_segment_all(bvec, io->bio, i) {
133
134 if (bvec->bv_page->mapping) {
135 target = bvec->bv_page;
136 } else {
137 struct f2fs_crypto_ctx *ctx;
138
139 /* encrypted page */
140 ctx = (struct f2fs_crypto_ctx *)page_private(
141 bvec->bv_page);
142 target = ctx->w.control_page;
143 }
144
145 if (page == target) {
146 up_read(&io->io_rwsem);
147 return true;
148 }
149 }
150
151 up_read(&io->io_rwsem);
152 return false;
153 }
154
155 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
156 enum page_type type, int rw)
157 {
158 enum page_type btype = PAGE_TYPE_OF_BIO(type);
159 struct f2fs_bio_info *io;
160
161 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
162
163 down_write(&io->io_rwsem);
164
165 /* change META to META_FLUSH in the checkpoint procedure */
166 if (type >= META_FLUSH) {
167 io->fio.type = META_FLUSH;
168 if (test_opt(sbi, NOBARRIER))
169 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
170 else
171 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
172 }
173 __submit_merged_bio(io);
174 up_write(&io->io_rwsem);
175 }
176
177 /*
178 * Fill the locked page with data located in the block address.
179 * Return unlocked page.
180 */
181 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
182 {
183 struct bio *bio;
184 struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
185
186 trace_f2fs_submit_page_bio(page, fio);
187 f2fs_trace_ios(fio, 0);
188
189 /* Allocate a new bio */
190 bio = __bio_alloc(fio->sbi, fio->blk_addr, 1, is_read_io(fio->rw));
191
192 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
193 bio_put(bio);
194 return -EFAULT;
195 }
196
197 submit_bio(fio->rw, bio);
198 return 0;
199 }
200
201 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
202 {
203 struct f2fs_sb_info *sbi = fio->sbi;
204 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
205 struct f2fs_bio_info *io;
206 bool is_read = is_read_io(fio->rw);
207 struct page *bio_page;
208
209 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
210
211 verify_block_addr(sbi, fio->blk_addr);
212
213 down_write(&io->io_rwsem);
214
215 if (!is_read)
216 inc_page_count(sbi, F2FS_WRITEBACK);
217
218 if (io->bio && (io->last_block_in_bio != fio->blk_addr - 1 ||
219 io->fio.rw != fio->rw))
220 __submit_merged_bio(io);
221 alloc_new:
222 if (io->bio == NULL) {
223 int bio_blocks = MAX_BIO_BLOCKS(sbi);
224
225 io->bio = __bio_alloc(sbi, fio->blk_addr, bio_blocks, is_read);
226 io->fio = *fio;
227 }
228
229 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
230
231 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
232 PAGE_CACHE_SIZE) {
233 __submit_merged_bio(io);
234 goto alloc_new;
235 }
236
237 io->last_block_in_bio = fio->blk_addr;
238 f2fs_trace_ios(fio, 0);
239
240 up_write(&io->io_rwsem);
241 trace_f2fs_submit_page_mbio(fio->page, fio);
242 }
243
244 /*
245 * Lock ordering for the change of data block address:
246 * ->data_page
247 * ->node_page
248 * update block addresses in the node page
249 */
250 void set_data_blkaddr(struct dnode_of_data *dn)
251 {
252 struct f2fs_node *rn;
253 __le32 *addr_array;
254 struct page *node_page = dn->node_page;
255 unsigned int ofs_in_node = dn->ofs_in_node;
256
257 f2fs_wait_on_page_writeback(node_page, NODE, true);
258
259 rn = F2FS_NODE(node_page);
260
261 /* Get physical address of data block */
262 addr_array = blkaddr_in_node(rn);
263 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
264 if (set_page_dirty(node_page))
265 dn->node_changed = true;
266 }
267
268 int reserve_new_block(struct dnode_of_data *dn)
269 {
270 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
271
272 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
273 return -EPERM;
274 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
275 return -ENOSPC;
276
277 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
278
279 dn->data_blkaddr = NEW_ADDR;
280 set_data_blkaddr(dn);
281 mark_inode_dirty(dn->inode);
282 sync_inode_page(dn);
283 return 0;
284 }
285
286 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
287 {
288 bool need_put = dn->inode_page ? false : true;
289 int err;
290
291 err = get_dnode_of_data(dn, index, ALLOC_NODE);
292 if (err)
293 return err;
294
295 if (dn->data_blkaddr == NULL_ADDR)
296 err = reserve_new_block(dn);
297 if (err || need_put)
298 f2fs_put_dnode(dn);
299 return err;
300 }
301
302 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
303 {
304 struct extent_info ei;
305 struct inode *inode = dn->inode;
306
307 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
308 dn->data_blkaddr = ei.blk + index - ei.fofs;
309 return 0;
310 }
311
312 return f2fs_reserve_block(dn, index);
313 }
314
315 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
316 int rw, bool for_write)
317 {
318 struct address_space *mapping = inode->i_mapping;
319 struct dnode_of_data dn;
320 struct page *page;
321 struct extent_info ei;
322 int err;
323 struct f2fs_io_info fio = {
324 .sbi = F2FS_I_SB(inode),
325 .type = DATA,
326 .rw = rw,
327 .encrypted_page = NULL,
328 };
329
330 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
331 return read_mapping_page(mapping, index, NULL);
332
333 page = f2fs_grab_cache_page(mapping, index, for_write);
334 if (!page)
335 return ERR_PTR(-ENOMEM);
336
337 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
338 dn.data_blkaddr = ei.blk + index - ei.fofs;
339 goto got_it;
340 }
341
342 set_new_dnode(&dn, inode, NULL, NULL, 0);
343 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
344 if (err)
345 goto put_err;
346 f2fs_put_dnode(&dn);
347
348 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
349 err = -ENOENT;
350 goto put_err;
351 }
352 got_it:
353 if (PageUptodate(page)) {
354 unlock_page(page);
355 return page;
356 }
357
358 /*
359 * A new dentry page is allocated but not able to be written, since its
360 * new inode page couldn't be allocated due to -ENOSPC.
361 * In such the case, its blkaddr can be remained as NEW_ADDR.
362 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
363 */
364 if (dn.data_blkaddr == NEW_ADDR) {
365 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
366 SetPageUptodate(page);
367 unlock_page(page);
368 return page;
369 }
370
371 fio.blk_addr = dn.data_blkaddr;
372 fio.page = page;
373 err = f2fs_submit_page_bio(&fio);
374 if (err)
375 goto put_err;
376 return page;
377
378 put_err:
379 f2fs_put_page(page, 1);
380 return ERR_PTR(err);
381 }
382
383 struct page *find_data_page(struct inode *inode, pgoff_t index)
384 {
385 struct address_space *mapping = inode->i_mapping;
386 struct page *page;
387
388 page = find_get_page(mapping, index);
389 if (page && PageUptodate(page))
390 return page;
391 f2fs_put_page(page, 0);
392
393 page = get_read_data_page(inode, index, READ_SYNC, false);
394 if (IS_ERR(page))
395 return page;
396
397 if (PageUptodate(page))
398 return page;
399
400 wait_on_page_locked(page);
401 if (unlikely(!PageUptodate(page))) {
402 f2fs_put_page(page, 0);
403 return ERR_PTR(-EIO);
404 }
405 return page;
406 }
407
408 /*
409 * If it tries to access a hole, return an error.
410 * Because, the callers, functions in dir.c and GC, should be able to know
411 * whether this page exists or not.
412 */
413 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
414 bool for_write)
415 {
416 struct address_space *mapping = inode->i_mapping;
417 struct page *page;
418 repeat:
419 page = get_read_data_page(inode, index, READ_SYNC, for_write);
420 if (IS_ERR(page))
421 return page;
422
423 /* wait for read completion */
424 lock_page(page);
425 if (unlikely(!PageUptodate(page))) {
426 f2fs_put_page(page, 1);
427 return ERR_PTR(-EIO);
428 }
429 if (unlikely(page->mapping != mapping)) {
430 f2fs_put_page(page, 1);
431 goto repeat;
432 }
433 return page;
434 }
435
436 /*
437 * Caller ensures that this data page is never allocated.
438 * A new zero-filled data page is allocated in the page cache.
439 *
440 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
441 * f2fs_unlock_op().
442 * Note that, ipage is set only by make_empty_dir, and if any error occur,
443 * ipage should be released by this function.
444 */
445 struct page *get_new_data_page(struct inode *inode,
446 struct page *ipage, pgoff_t index, bool new_i_size)
447 {
448 struct address_space *mapping = inode->i_mapping;
449 struct page *page;
450 struct dnode_of_data dn;
451 int err;
452
453 page = f2fs_grab_cache_page(mapping, index, true);
454 if (!page) {
455 /*
456 * before exiting, we should make sure ipage will be released
457 * if any error occur.
458 */
459 f2fs_put_page(ipage, 1);
460 return ERR_PTR(-ENOMEM);
461 }
462
463 set_new_dnode(&dn, inode, ipage, NULL, 0);
464 err = f2fs_reserve_block(&dn, index);
465 if (err) {
466 f2fs_put_page(page, 1);
467 return ERR_PTR(err);
468 }
469 if (!ipage)
470 f2fs_put_dnode(&dn);
471
472 if (PageUptodate(page))
473 goto got_it;
474
475 if (dn.data_blkaddr == NEW_ADDR) {
476 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
477 SetPageUptodate(page);
478 } else {
479 f2fs_put_page(page, 1);
480
481 /* if ipage exists, blkaddr should be NEW_ADDR */
482 f2fs_bug_on(F2FS_I_SB(inode), ipage);
483 page = get_lock_data_page(inode, index, true);
484 if (IS_ERR(page))
485 return page;
486 }
487 got_it:
488 if (new_i_size && i_size_read(inode) <
489 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
490 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
491 /* Only the directory inode sets new_i_size */
492 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
493 }
494 return page;
495 }
496
497 static int __allocate_data_block(struct dnode_of_data *dn)
498 {
499 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
500 struct f2fs_summary sum;
501 struct node_info ni;
502 int seg = CURSEG_WARM_DATA;
503 pgoff_t fofs;
504
505 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
506 return -EPERM;
507
508 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
509 if (dn->data_blkaddr == NEW_ADDR)
510 goto alloc;
511
512 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
513 return -ENOSPC;
514
515 alloc:
516 get_node_info(sbi, dn->nid, &ni);
517 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
518
519 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
520 seg = CURSEG_DIRECT_IO;
521
522 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
523 &sum, seg);
524 set_data_blkaddr(dn);
525
526 /* update i_size */
527 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
528 dn->ofs_in_node;
529 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
530 i_size_write(dn->inode,
531 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
532 return 0;
533 }
534
535 static int __allocate_data_blocks(struct inode *inode, loff_t offset,
536 size_t count)
537 {
538 struct f2fs_map_blocks map;
539
540 map.m_lblk = F2FS_BYTES_TO_BLK(offset);
541 map.m_len = F2FS_BYTES_TO_BLK(count);
542
543 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_DIO);
544 }
545
546 /*
547 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
548 * f2fs_map_blocks structure.
549 * If original data blocks are allocated, then give them to blockdev.
550 * Otherwise,
551 * a. preallocate requested block addresses
552 * b. do not use extent cache for better performance
553 * c. give the block addresses to blockdev
554 */
555 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
556 int create, int flag)
557 {
558 unsigned int maxblocks = map->m_len;
559 struct dnode_of_data dn;
560 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
561 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
562 pgoff_t pgofs, end_offset;
563 int err = 0, ofs = 1;
564 struct extent_info ei;
565 bool allocated = false;
566 block_t blkaddr;
567
568 map->m_len = 0;
569 map->m_flags = 0;
570
571 /* it only supports block size == page size */
572 pgofs = (pgoff_t)map->m_lblk;
573
574 if (f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
575 map->m_pblk = ei.blk + pgofs - ei.fofs;
576 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
577 map->m_flags = F2FS_MAP_MAPPED;
578 goto out;
579 }
580
581 next_dnode:
582 if (create)
583 f2fs_lock_op(sbi);
584
585 /* When reading holes, we need its node page */
586 set_new_dnode(&dn, inode, NULL, NULL, 0);
587 err = get_dnode_of_data(&dn, pgofs, mode);
588 if (err) {
589 if (err == -ENOENT)
590 err = 0;
591 goto unlock_out;
592 }
593
594 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
595
596 next_block:
597 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
598
599 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
600 if (create) {
601 if (unlikely(f2fs_cp_error(sbi))) {
602 err = -EIO;
603 goto sync_out;
604 }
605 err = __allocate_data_block(&dn);
606 if (err)
607 goto sync_out;
608 allocated = true;
609 map->m_flags = F2FS_MAP_NEW;
610 blkaddr = dn.data_blkaddr;
611 } else {
612 if (flag != F2FS_GET_BLOCK_FIEMAP ||
613 blkaddr != NEW_ADDR) {
614 if (flag == F2FS_GET_BLOCK_BMAP)
615 err = -ENOENT;
616 goto sync_out;
617 }
618 }
619 }
620
621 if (map->m_len == 0) {
622 /* preallocated unwritten block should be mapped for fiemap. */
623 if (blkaddr == NEW_ADDR)
624 map->m_flags |= F2FS_MAP_UNWRITTEN;
625 map->m_flags |= F2FS_MAP_MAPPED;
626
627 map->m_pblk = blkaddr;
628 map->m_len = 1;
629 } else if ((map->m_pblk != NEW_ADDR &&
630 blkaddr == (map->m_pblk + ofs)) ||
631 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR)) {
632 ofs++;
633 map->m_len++;
634 } else {
635 goto sync_out;
636 }
637
638 dn.ofs_in_node++;
639 pgofs++;
640
641 if (map->m_len < maxblocks) {
642 if (dn.ofs_in_node < end_offset)
643 goto next_block;
644
645 if (allocated)
646 sync_inode_page(&dn);
647 f2fs_put_dnode(&dn);
648
649 if (create) {
650 f2fs_unlock_op(sbi);
651 f2fs_balance_fs(sbi, allocated);
652 }
653 allocated = false;
654 goto next_dnode;
655 }
656
657 sync_out:
658 if (allocated)
659 sync_inode_page(&dn);
660 f2fs_put_dnode(&dn);
661 unlock_out:
662 if (create) {
663 f2fs_unlock_op(sbi);
664 f2fs_balance_fs(sbi, allocated);
665 }
666 out:
667 trace_f2fs_map_blocks(inode, map, err);
668 return err;
669 }
670
671 static int __get_data_block(struct inode *inode, sector_t iblock,
672 struct buffer_head *bh, int create, int flag)
673 {
674 struct f2fs_map_blocks map;
675 int ret;
676
677 map.m_lblk = iblock;
678 map.m_len = bh->b_size >> inode->i_blkbits;
679
680 ret = f2fs_map_blocks(inode, &map, create, flag);
681 if (!ret) {
682 map_bh(bh, inode->i_sb, map.m_pblk);
683 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
684 bh->b_size = map.m_len << inode->i_blkbits;
685 }
686 return ret;
687 }
688
689 static int get_data_block(struct inode *inode, sector_t iblock,
690 struct buffer_head *bh_result, int create, int flag)
691 {
692 return __get_data_block(inode, iblock, bh_result, create, flag);
693 }
694
695 static int get_data_block_dio(struct inode *inode, sector_t iblock,
696 struct buffer_head *bh_result, int create)
697 {
698 return __get_data_block(inode, iblock, bh_result, create,
699 F2FS_GET_BLOCK_DIO);
700 }
701
702 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
703 struct buffer_head *bh_result, int create)
704 {
705 /* Block number less than F2FS MAX BLOCKS */
706 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
707 return -EFBIG;
708
709 return __get_data_block(inode, iblock, bh_result, create,
710 F2FS_GET_BLOCK_BMAP);
711 }
712
713 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
714 {
715 return (offset >> inode->i_blkbits);
716 }
717
718 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
719 {
720 return (blk << inode->i_blkbits);
721 }
722
723 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
724 u64 start, u64 len)
725 {
726 struct buffer_head map_bh;
727 sector_t start_blk, last_blk;
728 loff_t isize;
729 u64 logical = 0, phys = 0, size = 0;
730 u32 flags = 0;
731 int ret = 0;
732
733 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
734 if (ret)
735 return ret;
736
737 if (f2fs_has_inline_data(inode)) {
738 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
739 if (ret != -EAGAIN)
740 return ret;
741 }
742
743 inode_lock(inode);
744
745 isize = i_size_read(inode);
746 if (start >= isize)
747 goto out;
748
749 if (start + len > isize)
750 len = isize - start;
751
752 if (logical_to_blk(inode, len) == 0)
753 len = blk_to_logical(inode, 1);
754
755 start_blk = logical_to_blk(inode, start);
756 last_blk = logical_to_blk(inode, start + len - 1);
757
758 next:
759 memset(&map_bh, 0, sizeof(struct buffer_head));
760 map_bh.b_size = len;
761
762 ret = get_data_block(inode, start_blk, &map_bh, 0,
763 F2FS_GET_BLOCK_FIEMAP);
764 if (ret)
765 goto out;
766
767 /* HOLE */
768 if (!buffer_mapped(&map_bh)) {
769 /* Go through holes util pass the EOF */
770 if (blk_to_logical(inode, start_blk++) < isize)
771 goto prep_next;
772 /* Found a hole beyond isize means no more extents.
773 * Note that the premise is that filesystems don't
774 * punch holes beyond isize and keep size unchanged.
775 */
776 flags |= FIEMAP_EXTENT_LAST;
777 }
778
779 if (size) {
780 if (f2fs_encrypted_inode(inode))
781 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
782
783 ret = fiemap_fill_next_extent(fieinfo, logical,
784 phys, size, flags);
785 }
786
787 if (start_blk > last_blk || ret)
788 goto out;
789
790 logical = blk_to_logical(inode, start_blk);
791 phys = blk_to_logical(inode, map_bh.b_blocknr);
792 size = map_bh.b_size;
793 flags = 0;
794 if (buffer_unwritten(&map_bh))
795 flags = FIEMAP_EXTENT_UNWRITTEN;
796
797 start_blk += logical_to_blk(inode, size);
798
799 prep_next:
800 cond_resched();
801 if (fatal_signal_pending(current))
802 ret = -EINTR;
803 else
804 goto next;
805 out:
806 if (ret == 1)
807 ret = 0;
808
809 inode_unlock(inode);
810 return ret;
811 }
812
813 /*
814 * This function was originally taken from fs/mpage.c, and customized for f2fs.
815 * Major change was from block_size == page_size in f2fs by default.
816 */
817 static int f2fs_mpage_readpages(struct address_space *mapping,
818 struct list_head *pages, struct page *page,
819 unsigned nr_pages)
820 {
821 struct bio *bio = NULL;
822 unsigned page_idx;
823 sector_t last_block_in_bio = 0;
824 struct inode *inode = mapping->host;
825 const unsigned blkbits = inode->i_blkbits;
826 const unsigned blocksize = 1 << blkbits;
827 sector_t block_in_file;
828 sector_t last_block;
829 sector_t last_block_in_file;
830 sector_t block_nr;
831 struct block_device *bdev = inode->i_sb->s_bdev;
832 struct f2fs_map_blocks map;
833
834 map.m_pblk = 0;
835 map.m_lblk = 0;
836 map.m_len = 0;
837 map.m_flags = 0;
838
839 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
840
841 prefetchw(&page->flags);
842 if (pages) {
843 page = list_entry(pages->prev, struct page, lru);
844 list_del(&page->lru);
845 if (add_to_page_cache_lru(page, mapping,
846 page->index, GFP_KERNEL))
847 goto next_page;
848 }
849
850 block_in_file = (sector_t)page->index;
851 last_block = block_in_file + nr_pages;
852 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
853 blkbits;
854 if (last_block > last_block_in_file)
855 last_block = last_block_in_file;
856
857 /*
858 * Map blocks using the previous result first.
859 */
860 if ((map.m_flags & F2FS_MAP_MAPPED) &&
861 block_in_file > map.m_lblk &&
862 block_in_file < (map.m_lblk + map.m_len))
863 goto got_it;
864
865 /*
866 * Then do more f2fs_map_blocks() calls until we are
867 * done with this page.
868 */
869 map.m_flags = 0;
870
871 if (block_in_file < last_block) {
872 map.m_lblk = block_in_file;
873 map.m_len = last_block - block_in_file;
874
875 if (f2fs_map_blocks(inode, &map, 0,
876 F2FS_GET_BLOCK_READ))
877 goto set_error_page;
878 }
879 got_it:
880 if ((map.m_flags & F2FS_MAP_MAPPED)) {
881 block_nr = map.m_pblk + block_in_file - map.m_lblk;
882 SetPageMappedToDisk(page);
883
884 if (!PageUptodate(page) && !cleancache_get_page(page)) {
885 SetPageUptodate(page);
886 goto confused;
887 }
888 } else {
889 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
890 SetPageUptodate(page);
891 unlock_page(page);
892 goto next_page;
893 }
894
895 /*
896 * This page will go to BIO. Do we need to send this
897 * BIO off first?
898 */
899 if (bio && (last_block_in_bio != block_nr - 1)) {
900 submit_and_realloc:
901 submit_bio(READ, bio);
902 bio = NULL;
903 }
904 if (bio == NULL) {
905 struct f2fs_crypto_ctx *ctx = NULL;
906
907 if (f2fs_encrypted_inode(inode) &&
908 S_ISREG(inode->i_mode)) {
909
910 ctx = f2fs_get_crypto_ctx(inode);
911 if (IS_ERR(ctx))
912 goto set_error_page;
913
914 /* wait the page to be moved by cleaning */
915 f2fs_wait_on_encrypted_page_writeback(
916 F2FS_I_SB(inode), block_nr);
917 }
918
919 bio = bio_alloc(GFP_KERNEL,
920 min_t(int, nr_pages, BIO_MAX_PAGES));
921 if (!bio) {
922 if (ctx)
923 f2fs_release_crypto_ctx(ctx);
924 goto set_error_page;
925 }
926 bio->bi_bdev = bdev;
927 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
928 bio->bi_end_io = f2fs_read_end_io;
929 bio->bi_private = ctx;
930 }
931
932 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
933 goto submit_and_realloc;
934
935 last_block_in_bio = block_nr;
936 goto next_page;
937 set_error_page:
938 SetPageError(page);
939 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
940 unlock_page(page);
941 goto next_page;
942 confused:
943 if (bio) {
944 submit_bio(READ, bio);
945 bio = NULL;
946 }
947 unlock_page(page);
948 next_page:
949 if (pages)
950 page_cache_release(page);
951 }
952 BUG_ON(pages && !list_empty(pages));
953 if (bio)
954 submit_bio(READ, bio);
955 return 0;
956 }
957
958 static int f2fs_read_data_page(struct file *file, struct page *page)
959 {
960 struct inode *inode = page->mapping->host;
961 int ret = -EAGAIN;
962
963 trace_f2fs_readpage(page, DATA);
964
965 /* If the file has inline data, try to read it directly */
966 if (f2fs_has_inline_data(inode))
967 ret = f2fs_read_inline_data(inode, page);
968 if (ret == -EAGAIN)
969 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
970 return ret;
971 }
972
973 static int f2fs_read_data_pages(struct file *file,
974 struct address_space *mapping,
975 struct list_head *pages, unsigned nr_pages)
976 {
977 struct inode *inode = file->f_mapping->host;
978 struct page *page = list_entry(pages->prev, struct page, lru);
979
980 trace_f2fs_readpages(inode, page, nr_pages);
981
982 /* If the file has inline data, skip readpages */
983 if (f2fs_has_inline_data(inode))
984 return 0;
985
986 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
987 }
988
989 int do_write_data_page(struct f2fs_io_info *fio)
990 {
991 struct page *page = fio->page;
992 struct inode *inode = page->mapping->host;
993 struct dnode_of_data dn;
994 int err = 0;
995
996 set_new_dnode(&dn, inode, NULL, NULL, 0);
997 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
998 if (err)
999 return err;
1000
1001 fio->blk_addr = dn.data_blkaddr;
1002
1003 /* This page is already truncated */
1004 if (fio->blk_addr == NULL_ADDR) {
1005 ClearPageUptodate(page);
1006 goto out_writepage;
1007 }
1008
1009 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1010
1011 /* wait for GCed encrypted page writeback */
1012 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1013 fio->blk_addr);
1014
1015 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1016 if (IS_ERR(fio->encrypted_page)) {
1017 err = PTR_ERR(fio->encrypted_page);
1018 goto out_writepage;
1019 }
1020 }
1021
1022 set_page_writeback(page);
1023
1024 /*
1025 * If current allocation needs SSR,
1026 * it had better in-place writes for updated data.
1027 */
1028 if (unlikely(fio->blk_addr != NEW_ADDR &&
1029 !is_cold_data(page) &&
1030 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1031 need_inplace_update(inode))) {
1032 rewrite_data_page(fio);
1033 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1034 trace_f2fs_do_write_data_page(page, IPU);
1035 } else {
1036 write_data_page(&dn, fio);
1037 set_data_blkaddr(&dn);
1038 f2fs_update_extent_cache(&dn);
1039 trace_f2fs_do_write_data_page(page, OPU);
1040 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1041 if (page->index == 0)
1042 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1043 }
1044 out_writepage:
1045 f2fs_put_dnode(&dn);
1046 return err;
1047 }
1048
1049 static int f2fs_write_data_page(struct page *page,
1050 struct writeback_control *wbc)
1051 {
1052 struct inode *inode = page->mapping->host;
1053 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1054 loff_t i_size = i_size_read(inode);
1055 const pgoff_t end_index = ((unsigned long long) i_size)
1056 >> PAGE_CACHE_SHIFT;
1057 unsigned offset = 0;
1058 bool need_balance_fs = false;
1059 int err = 0;
1060 struct f2fs_io_info fio = {
1061 .sbi = sbi,
1062 .type = DATA,
1063 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1064 .page = page,
1065 .encrypted_page = NULL,
1066 };
1067
1068 trace_f2fs_writepage(page, DATA);
1069
1070 if (page->index < end_index)
1071 goto write;
1072
1073 /*
1074 * If the offset is out-of-range of file size,
1075 * this page does not have to be written to disk.
1076 */
1077 offset = i_size & (PAGE_CACHE_SIZE - 1);
1078 if ((page->index >= end_index + 1) || !offset)
1079 goto out;
1080
1081 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1082 write:
1083 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1084 goto redirty_out;
1085 if (f2fs_is_drop_cache(inode))
1086 goto out;
1087 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1088 available_free_memory(sbi, BASE_CHECK))
1089 goto redirty_out;
1090
1091 /* Dentry blocks are controlled by checkpoint */
1092 if (S_ISDIR(inode->i_mode)) {
1093 if (unlikely(f2fs_cp_error(sbi)))
1094 goto redirty_out;
1095 err = do_write_data_page(&fio);
1096 goto done;
1097 }
1098
1099 /* we should bypass data pages to proceed the kworkder jobs */
1100 if (unlikely(f2fs_cp_error(sbi))) {
1101 SetPageError(page);
1102 goto out;
1103 }
1104
1105 if (!wbc->for_reclaim)
1106 need_balance_fs = true;
1107 else if (has_not_enough_free_secs(sbi, 0))
1108 goto redirty_out;
1109
1110 err = -EAGAIN;
1111 f2fs_lock_op(sbi);
1112 if (f2fs_has_inline_data(inode))
1113 err = f2fs_write_inline_data(inode, page);
1114 if (err == -EAGAIN)
1115 err = do_write_data_page(&fio);
1116 f2fs_unlock_op(sbi);
1117 done:
1118 if (err && err != -ENOENT)
1119 goto redirty_out;
1120
1121 clear_cold_data(page);
1122 out:
1123 inode_dec_dirty_pages(inode);
1124 if (err)
1125 ClearPageUptodate(page);
1126 unlock_page(page);
1127 f2fs_balance_fs(sbi, need_balance_fs);
1128 if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi))) {
1129 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1130 remove_dirty_inode(inode);
1131 }
1132 return 0;
1133
1134 redirty_out:
1135 redirty_page_for_writepage(wbc, page);
1136 return AOP_WRITEPAGE_ACTIVATE;
1137 }
1138
1139 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1140 void *data)
1141 {
1142 struct address_space *mapping = data;
1143 int ret = mapping->a_ops->writepage(page, wbc);
1144 mapping_set_error(mapping, ret);
1145 return ret;
1146 }
1147
1148 /*
1149 * This function was copied from write_cche_pages from mm/page-writeback.c.
1150 * The major change is making write step of cold data page separately from
1151 * warm/hot data page.
1152 */
1153 static int f2fs_write_cache_pages(struct address_space *mapping,
1154 struct writeback_control *wbc, writepage_t writepage,
1155 void *data)
1156 {
1157 int ret = 0;
1158 int done = 0;
1159 struct pagevec pvec;
1160 int nr_pages;
1161 pgoff_t uninitialized_var(writeback_index);
1162 pgoff_t index;
1163 pgoff_t end; /* Inclusive */
1164 pgoff_t done_index;
1165 int cycled;
1166 int range_whole = 0;
1167 int tag;
1168 int step = 0;
1169
1170 pagevec_init(&pvec, 0);
1171 next:
1172 if (wbc->range_cyclic) {
1173 writeback_index = mapping->writeback_index; /* prev offset */
1174 index = writeback_index;
1175 if (index == 0)
1176 cycled = 1;
1177 else
1178 cycled = 0;
1179 end = -1;
1180 } else {
1181 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1182 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1183 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1184 range_whole = 1;
1185 cycled = 1; /* ignore range_cyclic tests */
1186 }
1187 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1188 tag = PAGECACHE_TAG_TOWRITE;
1189 else
1190 tag = PAGECACHE_TAG_DIRTY;
1191 retry:
1192 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1193 tag_pages_for_writeback(mapping, index, end);
1194 done_index = index;
1195 while (!done && (index <= end)) {
1196 int i;
1197
1198 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1199 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1200 if (nr_pages == 0)
1201 break;
1202
1203 for (i = 0; i < nr_pages; i++) {
1204 struct page *page = pvec.pages[i];
1205
1206 if (page->index > end) {
1207 done = 1;
1208 break;
1209 }
1210
1211 done_index = page->index;
1212
1213 lock_page(page);
1214
1215 if (unlikely(page->mapping != mapping)) {
1216 continue_unlock:
1217 unlock_page(page);
1218 continue;
1219 }
1220
1221 if (!PageDirty(page)) {
1222 /* someone wrote it for us */
1223 goto continue_unlock;
1224 }
1225
1226 if (step == is_cold_data(page))
1227 goto continue_unlock;
1228
1229 if (PageWriteback(page)) {
1230 if (wbc->sync_mode != WB_SYNC_NONE)
1231 f2fs_wait_on_page_writeback(page,
1232 DATA, true);
1233 else
1234 goto continue_unlock;
1235 }
1236
1237 BUG_ON(PageWriteback(page));
1238 if (!clear_page_dirty_for_io(page))
1239 goto continue_unlock;
1240
1241 ret = (*writepage)(page, wbc, data);
1242 if (unlikely(ret)) {
1243 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1244 unlock_page(page);
1245 ret = 0;
1246 } else {
1247 done_index = page->index + 1;
1248 done = 1;
1249 break;
1250 }
1251 }
1252
1253 if (--wbc->nr_to_write <= 0 &&
1254 wbc->sync_mode == WB_SYNC_NONE) {
1255 done = 1;
1256 break;
1257 }
1258 }
1259 pagevec_release(&pvec);
1260 cond_resched();
1261 }
1262
1263 if (step < 1) {
1264 step++;
1265 goto next;
1266 }
1267
1268 if (!cycled && !done) {
1269 cycled = 1;
1270 index = 0;
1271 end = writeback_index - 1;
1272 goto retry;
1273 }
1274 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1275 mapping->writeback_index = done_index;
1276
1277 return ret;
1278 }
1279
1280 static int f2fs_write_data_pages(struct address_space *mapping,
1281 struct writeback_control *wbc)
1282 {
1283 struct inode *inode = mapping->host;
1284 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1285 bool locked = false;
1286 int ret;
1287 long diff;
1288
1289 trace_f2fs_writepages(mapping->host, wbc, DATA);
1290
1291 /* deal with chardevs and other special file */
1292 if (!mapping->a_ops->writepage)
1293 return 0;
1294
1295 /* skip writing if there is no dirty page in this inode */
1296 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1297 return 0;
1298
1299 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1300 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1301 available_free_memory(sbi, DIRTY_DENTS))
1302 goto skip_write;
1303
1304 /* skip writing during file defragment */
1305 if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1306 goto skip_write;
1307
1308 /* during POR, we don't need to trigger writepage at all. */
1309 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1310 goto skip_write;
1311
1312 diff = nr_pages_to_write(sbi, DATA, wbc);
1313
1314 if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1315 mutex_lock(&sbi->writepages);
1316 locked = true;
1317 }
1318 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1319 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1320 if (locked)
1321 mutex_unlock(&sbi->writepages);
1322
1323 remove_dirty_inode(inode);
1324
1325 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1326 return ret;
1327
1328 skip_write:
1329 wbc->pages_skipped += get_dirty_pages(inode);
1330 return 0;
1331 }
1332
1333 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1334 {
1335 struct inode *inode = mapping->host;
1336 loff_t i_size = i_size_read(inode);
1337
1338 if (to > i_size) {
1339 truncate_pagecache(inode, i_size);
1340 truncate_blocks(inode, i_size, true);
1341 }
1342 }
1343
1344 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1345 struct page *page, loff_t pos, unsigned len,
1346 block_t *blk_addr, bool *node_changed)
1347 {
1348 struct inode *inode = page->mapping->host;
1349 pgoff_t index = page->index;
1350 struct dnode_of_data dn;
1351 struct page *ipage;
1352 bool locked = false;
1353 struct extent_info ei;
1354 int err = 0;
1355
1356 if (f2fs_has_inline_data(inode) ||
1357 (pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1358 f2fs_lock_op(sbi);
1359 locked = true;
1360 }
1361 restart:
1362 /* check inline_data */
1363 ipage = get_node_page(sbi, inode->i_ino);
1364 if (IS_ERR(ipage)) {
1365 err = PTR_ERR(ipage);
1366 goto unlock_out;
1367 }
1368
1369 set_new_dnode(&dn, inode, ipage, ipage, 0);
1370
1371 if (f2fs_has_inline_data(inode)) {
1372 if (pos + len <= MAX_INLINE_DATA) {
1373 read_inline_data(page, ipage);
1374 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1375 set_inline_node(ipage);
1376 } else {
1377 err = f2fs_convert_inline_page(&dn, page);
1378 if (err)
1379 goto out;
1380 if (dn.data_blkaddr == NULL_ADDR)
1381 err = f2fs_get_block(&dn, index);
1382 }
1383 } else if (locked) {
1384 err = f2fs_get_block(&dn, index);
1385 } else {
1386 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1387 dn.data_blkaddr = ei.blk + index - ei.fofs;
1388 } else {
1389 /* hole case */
1390 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1391 if (err || (!err && dn.data_blkaddr == NULL_ADDR)) {
1392 f2fs_put_dnode(&dn);
1393 f2fs_lock_op(sbi);
1394 locked = true;
1395 goto restart;
1396 }
1397 }
1398 }
1399
1400 /* convert_inline_page can make node_changed */
1401 *blk_addr = dn.data_blkaddr;
1402 *node_changed = dn.node_changed;
1403 out:
1404 f2fs_put_dnode(&dn);
1405 unlock_out:
1406 if (locked)
1407 f2fs_unlock_op(sbi);
1408 return err;
1409 }
1410
1411 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1412 loff_t pos, unsigned len, unsigned flags,
1413 struct page **pagep, void **fsdata)
1414 {
1415 struct inode *inode = mapping->host;
1416 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1417 struct page *page = NULL;
1418 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1419 bool need_balance = false;
1420 block_t blkaddr = NULL_ADDR;
1421 int err = 0;
1422
1423 trace_f2fs_write_begin(inode, pos, len, flags);
1424
1425 /*
1426 * We should check this at this moment to avoid deadlock on inode page
1427 * and #0 page. The locking rule for inline_data conversion should be:
1428 * lock_page(page #0) -> lock_page(inode_page)
1429 */
1430 if (index != 0) {
1431 err = f2fs_convert_inline_inode(inode);
1432 if (err)
1433 goto fail;
1434 }
1435 repeat:
1436 page = grab_cache_page_write_begin(mapping, index, flags);
1437 if (!page) {
1438 err = -ENOMEM;
1439 goto fail;
1440 }
1441
1442 *pagep = page;
1443
1444 err = prepare_write_begin(sbi, page, pos, len,
1445 &blkaddr, &need_balance);
1446 if (err)
1447 goto fail;
1448
1449 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1450 unlock_page(page);
1451 f2fs_balance_fs(sbi, true);
1452 lock_page(page);
1453 if (page->mapping != mapping) {
1454 /* The page got truncated from under us */
1455 f2fs_put_page(page, 1);
1456 goto repeat;
1457 }
1458 }
1459
1460 f2fs_wait_on_page_writeback(page, DATA, false);
1461
1462 /* wait for GCed encrypted page writeback */
1463 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1464 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1465
1466 if (len == PAGE_CACHE_SIZE)
1467 goto out_update;
1468 if (PageUptodate(page))
1469 goto out_clear;
1470
1471 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1472 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1473 unsigned end = start + len;
1474
1475 /* Reading beyond i_size is simple: memset to zero */
1476 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1477 goto out_update;
1478 }
1479
1480 if (blkaddr == NEW_ADDR) {
1481 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1482 } else {
1483 struct f2fs_io_info fio = {
1484 .sbi = sbi,
1485 .type = DATA,
1486 .rw = READ_SYNC,
1487 .blk_addr = blkaddr,
1488 .page = page,
1489 .encrypted_page = NULL,
1490 };
1491 err = f2fs_submit_page_bio(&fio);
1492 if (err)
1493 goto fail;
1494
1495 lock_page(page);
1496 if (unlikely(!PageUptodate(page))) {
1497 err = -EIO;
1498 goto fail;
1499 }
1500 if (unlikely(page->mapping != mapping)) {
1501 f2fs_put_page(page, 1);
1502 goto repeat;
1503 }
1504
1505 /* avoid symlink page */
1506 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1507 err = f2fs_decrypt_one(inode, page);
1508 if (err)
1509 goto fail;
1510 }
1511 }
1512 out_update:
1513 SetPageUptodate(page);
1514 out_clear:
1515 clear_cold_data(page);
1516 return 0;
1517
1518 fail:
1519 f2fs_put_page(page, 1);
1520 f2fs_write_failed(mapping, pos + len);
1521 return err;
1522 }
1523
1524 static int f2fs_write_end(struct file *file,
1525 struct address_space *mapping,
1526 loff_t pos, unsigned len, unsigned copied,
1527 struct page *page, void *fsdata)
1528 {
1529 struct inode *inode = page->mapping->host;
1530
1531 trace_f2fs_write_end(inode, pos, len, copied);
1532
1533 set_page_dirty(page);
1534
1535 if (pos + copied > i_size_read(inode)) {
1536 i_size_write(inode, pos + copied);
1537 mark_inode_dirty(inode);
1538 }
1539
1540 f2fs_put_page(page, 1);
1541 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1542 return copied;
1543 }
1544
1545 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1546 loff_t offset)
1547 {
1548 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1549
1550 if (offset & blocksize_mask)
1551 return -EINVAL;
1552
1553 if (iov_iter_alignment(iter) & blocksize_mask)
1554 return -EINVAL;
1555
1556 return 0;
1557 }
1558
1559 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1560 loff_t offset)
1561 {
1562 struct file *file = iocb->ki_filp;
1563 struct address_space *mapping = file->f_mapping;
1564 struct inode *inode = mapping->host;
1565 size_t count = iov_iter_count(iter);
1566 int err;
1567
1568 /* we don't need to use inline_data strictly */
1569 err = f2fs_convert_inline_inode(inode);
1570 if (err)
1571 return err;
1572
1573 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1574 return 0;
1575
1576 err = check_direct_IO(inode, iter, offset);
1577 if (err)
1578 return err;
1579
1580 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1581
1582 if (iov_iter_rw(iter) == WRITE) {
1583 err = __allocate_data_blocks(inode, offset, count);
1584 if (err)
1585 goto out;
1586 }
1587
1588 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1589 out:
1590 if (err < 0 && iov_iter_rw(iter) == WRITE)
1591 f2fs_write_failed(mapping, offset + count);
1592
1593 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1594
1595 return err;
1596 }
1597
1598 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1599 unsigned int length)
1600 {
1601 struct inode *inode = page->mapping->host;
1602 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1603
1604 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1605 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1606 return;
1607
1608 if (PageDirty(page)) {
1609 if (inode->i_ino == F2FS_META_INO(sbi))
1610 dec_page_count(sbi, F2FS_DIRTY_META);
1611 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1612 dec_page_count(sbi, F2FS_DIRTY_NODES);
1613 else
1614 inode_dec_dirty_pages(inode);
1615 }
1616
1617 /* This is atomic written page, keep Private */
1618 if (IS_ATOMIC_WRITTEN_PAGE(page))
1619 return;
1620
1621 ClearPagePrivate(page);
1622 }
1623
1624 int f2fs_release_page(struct page *page, gfp_t wait)
1625 {
1626 /* If this is dirty page, keep PagePrivate */
1627 if (PageDirty(page))
1628 return 0;
1629
1630 /* This is atomic written page, keep Private */
1631 if (IS_ATOMIC_WRITTEN_PAGE(page))
1632 return 0;
1633
1634 ClearPagePrivate(page);
1635 return 1;
1636 }
1637
1638 static int f2fs_set_data_page_dirty(struct page *page)
1639 {
1640 struct address_space *mapping = page->mapping;
1641 struct inode *inode = mapping->host;
1642
1643 trace_f2fs_set_page_dirty(page, DATA);
1644
1645 SetPageUptodate(page);
1646
1647 if (f2fs_is_atomic_file(inode)) {
1648 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1649 register_inmem_page(inode, page);
1650 return 1;
1651 }
1652 /*
1653 * Previously, this page has been registered, we just
1654 * return here.
1655 */
1656 return 0;
1657 }
1658
1659 if (!PageDirty(page)) {
1660 __set_page_dirty_nobuffers(page);
1661 update_dirty_page(inode, page);
1662 return 1;
1663 }
1664 return 0;
1665 }
1666
1667 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1668 {
1669 struct inode *inode = mapping->host;
1670
1671 if (f2fs_has_inline_data(inode))
1672 return 0;
1673
1674 /* make sure allocating whole blocks */
1675 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1676 filemap_write_and_wait(mapping);
1677
1678 return generic_block_bmap(mapping, block, get_data_block_bmap);
1679 }
1680
1681 const struct address_space_operations f2fs_dblock_aops = {
1682 .readpage = f2fs_read_data_page,
1683 .readpages = f2fs_read_data_pages,
1684 .writepage = f2fs_write_data_page,
1685 .writepages = f2fs_write_data_pages,
1686 .write_begin = f2fs_write_begin,
1687 .write_end = f2fs_write_end,
1688 .set_page_dirty = f2fs_set_data_page_dirty,
1689 .invalidatepage = f2fs_invalidate_page,
1690 .releasepage = f2fs_release_page,
1691 .direct_IO = f2fs_direct_IO,
1692 .bmap = f2fs_bmap,
1693 };
Linux kernel - Deliverables
This page took 0.08751 seconds and 4 git commands to generate.