projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge remote-tracking branch 'iommu/next'
[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/mm.h>
23 #include <linux/memcontrol.h>
24 #include <linux/cleancache.h>
25
26 #include "f2fs.h"
27 #include "node.h"
28 #include "segment.h"
29 #include "trace.h"
30 #include <trace/events/f2fs.h>
31
32 static void f2fs_read_end_io(struct bio *bio)
33 {
34 struct bio_vec *bvec;
35 int i;
36
37 if (f2fs_bio_encrypted(bio)) {
38 if (bio->bi_error) {
39 fscrypt_release_ctx(bio->bi_private);
40 } else {
41 fscrypt_decrypt_bio_pages(bio->bi_private, bio);
42 return;
43 }
44 }
45
46 bio_for_each_segment_all(bvec, bio, i) {
47 struct page *page = bvec->bv_page;
48
49 if (!bio->bi_error) {
50 if (!PageUptodate(page))
51 SetPageUptodate(page);
52 } else {
53 ClearPageUptodate(page);
54 SetPageError(page);
55 }
56 unlock_page(page);
57 }
58 bio_put(bio);
59 }
60
61 static void f2fs_write_end_io(struct bio *bio)
62 {
63 struct f2fs_sb_info *sbi = bio->bi_private;
64 struct bio_vec *bvec;
65 int i;
66
67 bio_for_each_segment_all(bvec, bio, i) {
68 struct page *page = bvec->bv_page;
69
70 fscrypt_pullback_bio_page(&page, true);
71
72 if (unlikely(bio->bi_error)) {
73 set_bit(AS_EIO, &page->mapping->flags);
74 f2fs_stop_checkpoint(sbi, true);
75 }
76 end_page_writeback(page);
77 }
78 if (atomic_dec_and_test(&sbi->nr_wb_bios) &&
79 wq_has_sleeper(&sbi->cp_wait))
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 inline void __submit_bio(struct f2fs_sb_info *sbi,
104 struct bio *bio, enum page_type type)
105 {
106 if (!is_read_io(bio_op(bio))) {
107 atomic_inc(&sbi->nr_wb_bios);
108 if (f2fs_sb_mounted_hmsmr(sbi->sb) &&
109 current->plug && (type == DATA || type == NODE))
110 blk_finish_plug(current->plug);
111 }
112 submit_bio(bio);
113 }
114
115 static void __submit_merged_bio(struct f2fs_bio_info *io)
116 {
117 struct f2fs_io_info *fio = &io->fio;
118
119 if (!io->bio)
120 return;
121
122 if (is_read_io(fio->op))
123 trace_f2fs_submit_read_bio(io->sbi->sb, fio, io->bio);
124 else
125 trace_f2fs_submit_write_bio(io->sbi->sb, fio, io->bio);
126
127 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
128
129 __submit_bio(io->sbi, io->bio, fio->type);
130 io->bio = NULL;
131 }
132
133 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
134 struct page *page, nid_t ino)
135 {
136 struct bio_vec *bvec;
137 struct page *target;
138 int i;
139
140 if (!io->bio)
141 return false;
142
143 if (!inode && !page && !ino)
144 return true;
145
146 bio_for_each_segment_all(bvec, io->bio, i) {
147
148 if (bvec->bv_page->mapping)
149 target = bvec->bv_page;
150 else
151 target = fscrypt_control_page(bvec->bv_page);
152
153 if (inode && inode == target->mapping->host)
154 return true;
155 if (page && page == target)
156 return true;
157 if (ino && ino == ino_of_node(target))
158 return true;
159 }
160
161 return false;
162 }
163
164 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
165 struct page *page, nid_t ino,
166 enum page_type type)
167 {
168 enum page_type btype = PAGE_TYPE_OF_BIO(type);
169 struct f2fs_bio_info *io = &sbi->write_io[btype];
170 bool ret;
171
172 down_read(&io->io_rwsem);
173 ret = __has_merged_page(io, inode, page, ino);
174 up_read(&io->io_rwsem);
175 return ret;
176 }
177
178 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
179 struct inode *inode, struct page *page,
180 nid_t ino, enum page_type type, int rw)
181 {
182 enum page_type btype = PAGE_TYPE_OF_BIO(type);
183 struct f2fs_bio_info *io;
184
185 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
186
187 down_write(&io->io_rwsem);
188
189 if (!__has_merged_page(io, inode, page, ino))
190 goto out;
191
192 /* change META to META_FLUSH in the checkpoint procedure */
193 if (type >= META_FLUSH) {
194 io->fio.type = META_FLUSH;
195 io->fio.op = REQ_OP_WRITE;
196 if (test_opt(sbi, NOBARRIER))
197 io->fio.op_flags = WRITE_FLUSH | REQ_META | REQ_PRIO;
198 else
199 io->fio.op_flags = WRITE_FLUSH_FUA | REQ_META |
200 REQ_PRIO;
201 }
202 __submit_merged_bio(io);
203 out:
204 up_write(&io->io_rwsem);
205 }
206
207 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
208 int rw)
209 {
210 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
211 }
212
213 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
214 struct inode *inode, struct page *page,
215 nid_t ino, enum page_type type, int rw)
216 {
217 if (has_merged_page(sbi, inode, page, ino, type))
218 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
219 }
220
221 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
222 {
223 f2fs_submit_merged_bio(sbi, DATA, WRITE);
224 f2fs_submit_merged_bio(sbi, NODE, WRITE);
225 f2fs_submit_merged_bio(sbi, META, WRITE);
226 }
227
228 /*
229 * Fill the locked page with data located in the block address.
230 * Return unlocked page.
231 */
232 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
233 {
234 struct bio *bio;
235 struct page *page = fio->encrypted_page ?
236 fio->encrypted_page : fio->page;
237
238 trace_f2fs_submit_page_bio(page, fio);
239 f2fs_trace_ios(fio, 0);
240
241 /* Allocate a new bio */
242 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
243
244 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
245 bio_put(bio);
246 return -EFAULT;
247 }
248 bio_set_op_attrs(bio, fio->op, fio->op_flags);
249
250 __submit_bio(fio->sbi, bio, fio->type);
251 return 0;
252 }
253
254 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
255 {
256 struct f2fs_sb_info *sbi = fio->sbi;
257 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
258 struct f2fs_bio_info *io;
259 bool is_read = is_read_io(fio->op);
260 struct page *bio_page;
261
262 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
263
264 if (fio->old_blkaddr != NEW_ADDR)
265 verify_block_addr(sbi, fio->old_blkaddr);
266 verify_block_addr(sbi, fio->new_blkaddr);
267
268 down_write(&io->io_rwsem);
269
270 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
271 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags)))
272 __submit_merged_bio(io);
273 alloc_new:
274 if (io->bio == NULL) {
275 int bio_blocks = MAX_BIO_BLOCKS(sbi);
276
277 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
278 bio_blocks, is_read);
279 io->fio = *fio;
280 }
281
282 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
283
284 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) <
285 PAGE_SIZE) {
286 __submit_merged_bio(io);
287 goto alloc_new;
288 }
289
290 io->last_block_in_bio = fio->new_blkaddr;
291 f2fs_trace_ios(fio, 0);
292
293 up_write(&io->io_rwsem);
294 trace_f2fs_submit_page_mbio(fio->page, fio);
295 }
296
297 static void __set_data_blkaddr(struct dnode_of_data *dn)
298 {
299 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
300 __le32 *addr_array;
301
302 /* Get physical address of data block */
303 addr_array = blkaddr_in_node(rn);
304 addr_array[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
305 }
306
307 /*
308 * Lock ordering for the change of data block address:
309 * ->data_page
310 * ->node_page
311 * update block addresses in the node page
312 */
313 void set_data_blkaddr(struct dnode_of_data *dn)
314 {
315 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
316 __set_data_blkaddr(dn);
317 if (set_page_dirty(dn->node_page))
318 dn->node_changed = true;
319 }
320
321 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
322 {
323 dn->data_blkaddr = blkaddr;
324 set_data_blkaddr(dn);
325 f2fs_update_extent_cache(dn);
326 }
327
328 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
329 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
330 {
331 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
332
333 if (!count)
334 return 0;
335
336 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
337 return -EPERM;
338 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
339 return -ENOSPC;
340
341 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
342 dn->ofs_in_node, count);
343
344 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
345
346 for (; count > 0; dn->ofs_in_node++) {
347 block_t blkaddr =
348 datablock_addr(dn->node_page, dn->ofs_in_node);
349 if (blkaddr == NULL_ADDR) {
350 dn->data_blkaddr = NEW_ADDR;
351 __set_data_blkaddr(dn);
352 count--;
353 }
354 }
355
356 if (set_page_dirty(dn->node_page))
357 dn->node_changed = true;
358 return 0;
359 }
360
361 /* Should keep dn->ofs_in_node unchanged */
362 int reserve_new_block(struct dnode_of_data *dn)
363 {
364 unsigned int ofs_in_node = dn->ofs_in_node;
365 int ret;
366
367 ret = reserve_new_blocks(dn, 1);
368 dn->ofs_in_node = ofs_in_node;
369 return ret;
370 }
371
372 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
373 {
374 bool need_put = dn->inode_page ? false : true;
375 int err;
376
377 err = get_dnode_of_data(dn, index, ALLOC_NODE);
378 if (err)
379 return err;
380
381 if (dn->data_blkaddr == NULL_ADDR)
382 err = reserve_new_block(dn);
383 if (err || need_put)
384 f2fs_put_dnode(dn);
385 return err;
386 }
387
388 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
389 {
390 struct extent_info ei;
391 struct inode *inode = dn->inode;
392
393 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
394 dn->data_blkaddr = ei.blk + index - ei.fofs;
395 return 0;
396 }
397
398 return f2fs_reserve_block(dn, index);
399 }
400
401 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
402 int op_flags, bool for_write)
403 {
404 struct address_space *mapping = inode->i_mapping;
405 struct dnode_of_data dn;
406 struct page *page;
407 struct extent_info ei;
408 int err;
409 struct f2fs_io_info fio = {
410 .sbi = F2FS_I_SB(inode),
411 .type = DATA,
412 .op = REQ_OP_READ,
413 .op_flags = op_flags,
414 .encrypted_page = NULL,
415 };
416
417 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
418 return read_mapping_page(mapping, index, NULL);
419
420 page = f2fs_grab_cache_page(mapping, index, for_write);
421 if (!page)
422 return ERR_PTR(-ENOMEM);
423
424 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
425 dn.data_blkaddr = ei.blk + index - ei.fofs;
426 goto got_it;
427 }
428
429 set_new_dnode(&dn, inode, NULL, NULL, 0);
430 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
431 if (err)
432 goto put_err;
433 f2fs_put_dnode(&dn);
434
435 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
436 err = -ENOENT;
437 goto put_err;
438 }
439 got_it:
440 if (PageUptodate(page)) {
441 unlock_page(page);
442 return page;
443 }
444
445 /*
446 * A new dentry page is allocated but not able to be written, since its
447 * new inode page couldn't be allocated due to -ENOSPC.
448 * In such the case, its blkaddr can be remained as NEW_ADDR.
449 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
450 */
451 if (dn.data_blkaddr == NEW_ADDR) {
452 zero_user_segment(page, 0, PAGE_SIZE);
453 if (!PageUptodate(page))
454 SetPageUptodate(page);
455 unlock_page(page);
456 return page;
457 }
458
459 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
460 fio.page = page;
461 err = f2fs_submit_page_bio(&fio);
462 if (err)
463 goto put_err;
464 return page;
465
466 put_err:
467 f2fs_put_page(page, 1);
468 return ERR_PTR(err);
469 }
470
471 struct page *find_data_page(struct inode *inode, pgoff_t index)
472 {
473 struct address_space *mapping = inode->i_mapping;
474 struct page *page;
475
476 page = find_get_page(mapping, index);
477 if (page && PageUptodate(page))
478 return page;
479 f2fs_put_page(page, 0);
480
481 page = get_read_data_page(inode, index, READ_SYNC, false);
482 if (IS_ERR(page))
483 return page;
484
485 if (PageUptodate(page))
486 return page;
487
488 wait_on_page_locked(page);
489 if (unlikely(!PageUptodate(page))) {
490 f2fs_put_page(page, 0);
491 return ERR_PTR(-EIO);
492 }
493 return page;
494 }
495
496 /*
497 * If it tries to access a hole, return an error.
498 * Because, the callers, functions in dir.c and GC, should be able to know
499 * whether this page exists or not.
500 */
501 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
502 bool for_write)
503 {
504 struct address_space *mapping = inode->i_mapping;
505 struct page *page;
506 repeat:
507 page = get_read_data_page(inode, index, READ_SYNC, for_write);
508 if (IS_ERR(page))
509 return page;
510
511 /* wait for read completion */
512 lock_page(page);
513 if (unlikely(page->mapping != mapping)) {
514 f2fs_put_page(page, 1);
515 goto repeat;
516 }
517 if (unlikely(!PageUptodate(page))) {
518 f2fs_put_page(page, 1);
519 return ERR_PTR(-EIO);
520 }
521 return page;
522 }
523
524 /*
525 * Caller ensures that this data page is never allocated.
526 * A new zero-filled data page is allocated in the page cache.
527 *
528 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
529 * f2fs_unlock_op().
530 * Note that, ipage is set only by make_empty_dir, and if any error occur,
531 * ipage should be released by this function.
532 */
533 struct page *get_new_data_page(struct inode *inode,
534 struct page *ipage, pgoff_t index, bool new_i_size)
535 {
536 struct address_space *mapping = inode->i_mapping;
537 struct page *page;
538 struct dnode_of_data dn;
539 int err;
540
541 page = f2fs_grab_cache_page(mapping, index, true);
542 if (!page) {
543 /*
544 * before exiting, we should make sure ipage will be released
545 * if any error occur.
546 */
547 f2fs_put_page(ipage, 1);
548 return ERR_PTR(-ENOMEM);
549 }
550
551 set_new_dnode(&dn, inode, ipage, NULL, 0);
552 err = f2fs_reserve_block(&dn, index);
553 if (err) {
554 f2fs_put_page(page, 1);
555 return ERR_PTR(err);
556 }
557 if (!ipage)
558 f2fs_put_dnode(&dn);
559
560 if (PageUptodate(page))
561 goto got_it;
562
563 if (dn.data_blkaddr == NEW_ADDR) {
564 zero_user_segment(page, 0, PAGE_SIZE);
565 if (!PageUptodate(page))
566 SetPageUptodate(page);
567 } else {
568 f2fs_put_page(page, 1);
569
570 /* if ipage exists, blkaddr should be NEW_ADDR */
571 f2fs_bug_on(F2FS_I_SB(inode), ipage);
572 page = get_lock_data_page(inode, index, true);
573 if (IS_ERR(page))
574 return page;
575 }
576 got_it:
577 if (new_i_size && i_size_read(inode) <
578 ((loff_t)(index + 1) << PAGE_SHIFT))
579 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
580 return page;
581 }
582
583 static int __allocate_data_block(struct dnode_of_data *dn)
584 {
585 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
586 struct f2fs_summary sum;
587 struct node_info ni;
588 int seg = CURSEG_WARM_DATA;
589 pgoff_t fofs;
590 blkcnt_t count = 1;
591
592 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
593 return -EPERM;
594
595 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
596 if (dn->data_blkaddr == NEW_ADDR)
597 goto alloc;
598
599 if (unlikely(!inc_valid_block_count(sbi, dn->inode, &count)))
600 return -ENOSPC;
601
602 alloc:
603 get_node_info(sbi, dn->nid, &ni);
604 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
605
606 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
607 seg = CURSEG_DIRECT_IO;
608
609 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
610 &sum, seg);
611 set_data_blkaddr(dn);
612
613 /* update i_size */
614 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
615 dn->ofs_in_node;
616 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
617 f2fs_i_size_write(dn->inode,
618 ((loff_t)(fofs + 1) << PAGE_SHIFT));
619 return 0;
620 }
621
622 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
623 {
624 struct inode *inode = file_inode(iocb->ki_filp);
625 struct f2fs_map_blocks map;
626 ssize_t ret = 0;
627
628 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
629 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
630 if (map.m_len > map.m_lblk)
631 map.m_len -= map.m_lblk;
632 else
633 map.m_len = 0;
634
635 map.m_next_pgofs = NULL;
636
637 if (f2fs_encrypted_inode(inode))
638 return 0;
639
640 if (iocb->ki_flags & IOCB_DIRECT) {
641 ret = f2fs_convert_inline_inode(inode);
642 if (ret)
643 return ret;
644 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
645 }
646 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
647 ret = f2fs_convert_inline_inode(inode);
648 if (ret)
649 return ret;
650 }
651 if (!f2fs_has_inline_data(inode))
652 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
653 return ret;
654 }
655
656 /*
657 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
658 * f2fs_map_blocks structure.
659 * If original data blocks are allocated, then give them to blockdev.
660 * Otherwise,
661 * a. preallocate requested block addresses
662 * b. do not use extent cache for better performance
663 * c. give the block addresses to blockdev
664 */
665 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
666 int create, int flag)
667 {
668 unsigned int maxblocks = map->m_len;
669 struct dnode_of_data dn;
670 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
671 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
672 pgoff_t pgofs, end_offset, end;
673 int err = 0, ofs = 1;
674 unsigned int ofs_in_node, last_ofs_in_node;
675 blkcnt_t prealloc;
676 struct extent_info ei;
677 bool allocated = false;
678 block_t blkaddr;
679
680 if (!maxblocks)
681 return 0;
682
683 map->m_len = 0;
684 map->m_flags = 0;
685
686 /* it only supports block size == page size */
687 pgofs = (pgoff_t)map->m_lblk;
688 end = pgofs + maxblocks;
689
690 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
691 map->m_pblk = ei.blk + pgofs - ei.fofs;
692 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
693 map->m_flags = F2FS_MAP_MAPPED;
694 goto out;
695 }
696
697 next_dnode:
698 if (create)
699 f2fs_lock_op(sbi);
700
701 /* When reading holes, we need its node page */
702 set_new_dnode(&dn, inode, NULL, NULL, 0);
703 err = get_dnode_of_data(&dn, pgofs, mode);
704 if (err) {
705 if (flag == F2FS_GET_BLOCK_BMAP)
706 map->m_pblk = 0;
707 if (err == -ENOENT) {
708 err = 0;
709 if (map->m_next_pgofs)
710 *map->m_next_pgofs =
711 get_next_page_offset(&dn, pgofs);
712 }
713 goto unlock_out;
714 }
715
716 prealloc = 0;
717 ofs_in_node = dn.ofs_in_node;
718 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
719
720 next_block:
721 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
722
723 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
724 if (create) {
725 if (unlikely(f2fs_cp_error(sbi))) {
726 err = -EIO;
727 goto sync_out;
728 }
729 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
730 if (blkaddr == NULL_ADDR) {
731 prealloc++;
732 last_ofs_in_node = dn.ofs_in_node;
733 }
734 } else {
735 err = __allocate_data_block(&dn);
736 if (!err) {
737 set_inode_flag(inode, FI_APPEND_WRITE);
738 allocated = true;
739 }
740 }
741 if (err)
742 goto sync_out;
743 map->m_flags = F2FS_MAP_NEW;
744 blkaddr = dn.data_blkaddr;
745 } else {
746 if (flag == F2FS_GET_BLOCK_BMAP) {
747 map->m_pblk = 0;
748 goto sync_out;
749 }
750 if (flag == F2FS_GET_BLOCK_FIEMAP &&
751 blkaddr == NULL_ADDR) {
752 if (map->m_next_pgofs)
753 *map->m_next_pgofs = pgofs + 1;
754 }
755 if (flag != F2FS_GET_BLOCK_FIEMAP ||
756 blkaddr != NEW_ADDR)
757 goto sync_out;
758 }
759 }
760
761 if (flag == F2FS_GET_BLOCK_PRE_AIO)
762 goto skip;
763
764 if (map->m_len == 0) {
765 /* preallocated unwritten block should be mapped for fiemap. */
766 if (blkaddr == NEW_ADDR)
767 map->m_flags |= F2FS_MAP_UNWRITTEN;
768 map->m_flags |= F2FS_MAP_MAPPED;
769
770 map->m_pblk = blkaddr;
771 map->m_len = 1;
772 } else if ((map->m_pblk != NEW_ADDR &&
773 blkaddr == (map->m_pblk + ofs)) ||
774 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
775 flag == F2FS_GET_BLOCK_PRE_DIO) {
776 ofs++;
777 map->m_len++;
778 } else {
779 goto sync_out;
780 }
781
782 skip:
783 dn.ofs_in_node++;
784 pgofs++;
785
786 /* preallocate blocks in batch for one dnode page */
787 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
788 (pgofs == end || dn.ofs_in_node == end_offset)) {
789
790 dn.ofs_in_node = ofs_in_node;
791 err = reserve_new_blocks(&dn, prealloc);
792 if (err)
793 goto sync_out;
794 allocated = dn.node_changed;
795
796 map->m_len += dn.ofs_in_node - ofs_in_node;
797 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
798 err = -ENOSPC;
799 goto sync_out;
800 }
801 dn.ofs_in_node = end_offset;
802 }
803
804 if (pgofs >= end)
805 goto sync_out;
806 else if (dn.ofs_in_node < end_offset)
807 goto next_block;
808
809 f2fs_put_dnode(&dn);
810
811 if (create) {
812 f2fs_unlock_op(sbi);
813 f2fs_balance_fs(sbi, allocated);
814 }
815 allocated = false;
816 goto next_dnode;
817
818 sync_out:
819 f2fs_put_dnode(&dn);
820 unlock_out:
821 if (create) {
822 f2fs_unlock_op(sbi);
823 f2fs_balance_fs(sbi, allocated);
824 }
825 out:
826 trace_f2fs_map_blocks(inode, map, err);
827 return err;
828 }
829
830 static int __get_data_block(struct inode *inode, sector_t iblock,
831 struct buffer_head *bh, int create, int flag,
832 pgoff_t *next_pgofs)
833 {
834 struct f2fs_map_blocks map;
835 int ret;
836
837 map.m_lblk = iblock;
838 map.m_len = bh->b_size >> inode->i_blkbits;
839 map.m_next_pgofs = next_pgofs;
840
841 ret = f2fs_map_blocks(inode, &map, create, flag);
842 if (!ret) {
843 map_bh(bh, inode->i_sb, map.m_pblk);
844 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
845 bh->b_size = map.m_len << inode->i_blkbits;
846 }
847 return ret;
848 }
849
850 static int get_data_block(struct inode *inode, sector_t iblock,
851 struct buffer_head *bh_result, int create, int flag,
852 pgoff_t *next_pgofs)
853 {
854 return __get_data_block(inode, iblock, bh_result, create,
855 flag, next_pgofs);
856 }
857
858 static int get_data_block_dio(struct inode *inode, sector_t iblock,
859 struct buffer_head *bh_result, int create)
860 {
861 return __get_data_block(inode, iblock, bh_result, create,
862 F2FS_GET_BLOCK_DIO, NULL);
863 }
864
865 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
866 struct buffer_head *bh_result, int create)
867 {
868 /* Block number less than F2FS MAX BLOCKS */
869 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
870 return -EFBIG;
871
872 return __get_data_block(inode, iblock, bh_result, create,
873 F2FS_GET_BLOCK_BMAP, NULL);
874 }
875
876 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
877 {
878 return (offset >> inode->i_blkbits);
879 }
880
881 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
882 {
883 return (blk << inode->i_blkbits);
884 }
885
886 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
887 u64 start, u64 len)
888 {
889 struct buffer_head map_bh;
890 sector_t start_blk, last_blk;
891 pgoff_t next_pgofs;
892 loff_t isize;
893 u64 logical = 0, phys = 0, size = 0;
894 u32 flags = 0;
895 int ret = 0;
896
897 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
898 if (ret)
899 return ret;
900
901 if (f2fs_has_inline_data(inode)) {
902 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
903 if (ret != -EAGAIN)
904 return ret;
905 }
906
907 inode_lock(inode);
908
909 isize = i_size_read(inode);
910 if (start >= isize)
911 goto out;
912
913 if (start + len > isize)
914 len = isize - start;
915
916 if (logical_to_blk(inode, len) == 0)
917 len = blk_to_logical(inode, 1);
918
919 start_blk = logical_to_blk(inode, start);
920 last_blk = logical_to_blk(inode, start + len - 1);
921
922 next:
923 memset(&map_bh, 0, sizeof(struct buffer_head));
924 map_bh.b_size = len;
925
926 ret = get_data_block(inode, start_blk, &map_bh, 0,
927 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
928 if (ret)
929 goto out;
930
931 /* HOLE */
932 if (!buffer_mapped(&map_bh)) {
933 start_blk = next_pgofs;
934 /* Go through holes util pass the EOF */
935 if (blk_to_logical(inode, start_blk) < isize)
936 goto prep_next;
937 /* Found a hole beyond isize means no more extents.
938 * Note that the premise is that filesystems don't
939 * punch holes beyond isize and keep size unchanged.
940 */
941 flags |= FIEMAP_EXTENT_LAST;
942 }
943
944 if (size) {
945 if (f2fs_encrypted_inode(inode))
946 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
947
948 ret = fiemap_fill_next_extent(fieinfo, logical,
949 phys, size, flags);
950 }
951
952 if (start_blk > last_blk || ret)
953 goto out;
954
955 logical = blk_to_logical(inode, start_blk);
956 phys = blk_to_logical(inode, map_bh.b_blocknr);
957 size = map_bh.b_size;
958 flags = 0;
959 if (buffer_unwritten(&map_bh))
960 flags = FIEMAP_EXTENT_UNWRITTEN;
961
962 start_blk += logical_to_blk(inode, size);
963
964 prep_next:
965 cond_resched();
966 if (fatal_signal_pending(current))
967 ret = -EINTR;
968 else
969 goto next;
970 out:
971 if (ret == 1)
972 ret = 0;
973
974 inode_unlock(inode);
975 return ret;
976 }
977
978 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
979 unsigned nr_pages)
980 {
981 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
982 struct fscrypt_ctx *ctx = NULL;
983 struct block_device *bdev = sbi->sb->s_bdev;
984 struct bio *bio;
985
986 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
987 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
988 if (IS_ERR(ctx))
989 return ERR_CAST(ctx);
990
991 /* wait the page to be moved by cleaning */
992 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
993 }
994
995 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
996 if (!bio) {
997 if (ctx)
998 fscrypt_release_ctx(ctx);
999 return ERR_PTR(-ENOMEM);
1000 }
1001 bio->bi_bdev = bdev;
1002 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blkaddr);
1003 bio->bi_end_io = f2fs_read_end_io;
1004 bio->bi_private = ctx;
1005
1006 return bio;
1007 }
1008
1009 /*
1010 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1011 * Major change was from block_size == page_size in f2fs by default.
1012 */
1013 static int f2fs_mpage_readpages(struct address_space *mapping,
1014 struct list_head *pages, struct page *page,
1015 unsigned nr_pages)
1016 {
1017 struct bio *bio = NULL;
1018 unsigned page_idx;
1019 sector_t last_block_in_bio = 0;
1020 struct inode *inode = mapping->host;
1021 const unsigned blkbits = inode->i_blkbits;
1022 const unsigned blocksize = 1 << blkbits;
1023 sector_t block_in_file;
1024 sector_t last_block;
1025 sector_t last_block_in_file;
1026 sector_t block_nr;
1027 struct f2fs_map_blocks map;
1028
1029 map.m_pblk = 0;
1030 map.m_lblk = 0;
1031 map.m_len = 0;
1032 map.m_flags = 0;
1033 map.m_next_pgofs = NULL;
1034
1035 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1036
1037 prefetchw(&page->flags);
1038 if (pages) {
1039 page = list_entry(pages->prev, struct page, lru);
1040 list_del(&page->lru);
1041 if (add_to_page_cache_lru(page, mapping,
1042 page->index,
1043 readahead_gfp_mask(mapping)))
1044 goto next_page;
1045 }
1046
1047 block_in_file = (sector_t)page->index;
1048 last_block = block_in_file + nr_pages;
1049 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1050 blkbits;
1051 if (last_block > last_block_in_file)
1052 last_block = last_block_in_file;
1053
1054 /*
1055 * Map blocks using the previous result first.
1056 */
1057 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1058 block_in_file > map.m_lblk &&
1059 block_in_file < (map.m_lblk + map.m_len))
1060 goto got_it;
1061
1062 /*
1063 * Then do more f2fs_map_blocks() calls until we are
1064 * done with this page.
1065 */
1066 map.m_flags = 0;
1067
1068 if (block_in_file < last_block) {
1069 map.m_lblk = block_in_file;
1070 map.m_len = last_block - block_in_file;
1071
1072 if (f2fs_map_blocks(inode, &map, 0,
1073 F2FS_GET_BLOCK_READ))
1074 goto set_error_page;
1075 }
1076 got_it:
1077 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1078 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1079 SetPageMappedToDisk(page);
1080
1081 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1082 SetPageUptodate(page);
1083 goto confused;
1084 }
1085 } else {
1086 zero_user_segment(page, 0, PAGE_SIZE);
1087 if (!PageUptodate(page))
1088 SetPageUptodate(page);
1089 unlock_page(page);
1090 goto next_page;
1091 }
1092
1093 /*
1094 * This page will go to BIO. Do we need to send this
1095 * BIO off first?
1096 */
1097 if (bio && (last_block_in_bio != block_nr - 1)) {
1098 submit_and_realloc:
1099 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1100 bio = NULL;
1101 }
1102 if (bio == NULL) {
1103 bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1104 if (IS_ERR(bio)) {
1105 bio = NULL;
1106 goto set_error_page;
1107 }
1108 bio_set_op_attrs(bio, REQ_OP_READ, 0);
1109 }
1110
1111 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1112 goto submit_and_realloc;
1113
1114 last_block_in_bio = block_nr;
1115 goto next_page;
1116 set_error_page:
1117 SetPageError(page);
1118 zero_user_segment(page, 0, PAGE_SIZE);
1119 unlock_page(page);
1120 goto next_page;
1121 confused:
1122 if (bio) {
1123 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1124 bio = NULL;
1125 }
1126 unlock_page(page);
1127 next_page:
1128 if (pages)
1129 put_page(page);
1130 }
1131 BUG_ON(pages && !list_empty(pages));
1132 if (bio)
1133 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1134 return 0;
1135 }
1136
1137 static int f2fs_read_data_page(struct file *file, struct page *page)
1138 {
1139 struct inode *inode = page->mapping->host;
1140 int ret = -EAGAIN;
1141
1142 trace_f2fs_readpage(page, DATA);
1143
1144 /* If the file has inline data, try to read it directly */
1145 if (f2fs_has_inline_data(inode))
1146 ret = f2fs_read_inline_data(inode, page);
1147 if (ret == -EAGAIN)
1148 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1149 return ret;
1150 }
1151
1152 static int f2fs_read_data_pages(struct file *file,
1153 struct address_space *mapping,
1154 struct list_head *pages, unsigned nr_pages)
1155 {
1156 struct inode *inode = file->f_mapping->host;
1157 struct page *page = list_entry(pages->prev, struct page, lru);
1158
1159 trace_f2fs_readpages(inode, page, nr_pages);
1160
1161 /* If the file has inline data, skip readpages */
1162 if (f2fs_has_inline_data(inode))
1163 return 0;
1164
1165 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1166 }
1167
1168 int do_write_data_page(struct f2fs_io_info *fio)
1169 {
1170 struct page *page = fio->page;
1171 struct inode *inode = page->mapping->host;
1172 struct dnode_of_data dn;
1173 int err = 0;
1174
1175 set_new_dnode(&dn, inode, NULL, NULL, 0);
1176 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1177 if (err)
1178 return err;
1179
1180 fio->old_blkaddr = dn.data_blkaddr;
1181
1182 /* This page is already truncated */
1183 if (fio->old_blkaddr == NULL_ADDR) {
1184 ClearPageUptodate(page);
1185 goto out_writepage;
1186 }
1187
1188 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1189 gfp_t gfp_flags = GFP_NOFS;
1190
1191 /* wait for GCed encrypted page writeback */
1192 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1193 fio->old_blkaddr);
1194 retry_encrypt:
1195 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1196 gfp_flags);
1197 if (IS_ERR(fio->encrypted_page)) {
1198 err = PTR_ERR(fio->encrypted_page);
1199 if (err == -ENOMEM) {
1200 /* flush pending ios and wait for a while */
1201 f2fs_flush_merged_bios(F2FS_I_SB(inode));
1202 congestion_wait(BLK_RW_ASYNC, HZ/50);
1203 gfp_flags |= __GFP_NOFAIL;
1204 err = 0;
1205 goto retry_encrypt;
1206 }
1207 goto out_writepage;
1208 }
1209 }
1210
1211 set_page_writeback(page);
1212
1213 /*
1214 * If current allocation needs SSR,
1215 * it had better in-place writes for updated data.
1216 */
1217 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1218 !is_cold_data(page) &&
1219 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1220 need_inplace_update(inode))) {
1221 rewrite_data_page(fio);
1222 set_inode_flag(inode, FI_UPDATE_WRITE);
1223 trace_f2fs_do_write_data_page(page, IPU);
1224 } else {
1225 write_data_page(&dn, fio);
1226 trace_f2fs_do_write_data_page(page, OPU);
1227 set_inode_flag(inode, FI_APPEND_WRITE);
1228 if (page->index == 0)
1229 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1230 }
1231 out_writepage:
1232 f2fs_put_dnode(&dn);
1233 return err;
1234 }
1235
1236 static int f2fs_write_data_page(struct page *page,
1237 struct writeback_control *wbc)
1238 {
1239 struct inode *inode = page->mapping->host;
1240 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1241 loff_t i_size = i_size_read(inode);
1242 const pgoff_t end_index = ((unsigned long long) i_size)
1243 >> PAGE_SHIFT;
1244 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1245 unsigned offset = 0;
1246 bool need_balance_fs = false;
1247 int err = 0;
1248 struct f2fs_io_info fio = {
1249 .sbi = sbi,
1250 .type = DATA,
1251 .op = REQ_OP_WRITE,
1252 .op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1253 .page = page,
1254 .encrypted_page = NULL,
1255 };
1256
1257 trace_f2fs_writepage(page, DATA);
1258
1259 if (page->index < end_index)
1260 goto write;
1261
1262 /*
1263 * If the offset is out-of-range of file size,
1264 * this page does not have to be written to disk.
1265 */
1266 offset = i_size & (PAGE_SIZE - 1);
1267 if ((page->index >= end_index + 1) || !offset)
1268 goto out;
1269
1270 zero_user_segment(page, offset, PAGE_SIZE);
1271 write:
1272 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1273 goto redirty_out;
1274 if (f2fs_is_drop_cache(inode))
1275 goto out;
1276 /* we should not write 0'th page having journal header */
1277 if (f2fs_is_volatile_file(inode) && (!page->index ||
1278 (!wbc->for_reclaim &&
1279 available_free_memory(sbi, BASE_CHECK))))
1280 goto redirty_out;
1281
1282 /* we should bypass data pages to proceed the kworkder jobs */
1283 if (unlikely(f2fs_cp_error(sbi))) {
1284 mapping_set_error(page->mapping, -EIO);
1285 goto out;
1286 }
1287
1288 /* Dentry blocks are controlled by checkpoint */
1289 if (S_ISDIR(inode->i_mode)) {
1290 err = do_write_data_page(&fio);
1291 goto done;
1292 }
1293
1294 if (!wbc->for_reclaim)
1295 need_balance_fs = true;
1296 else if (has_not_enough_free_secs(sbi, 0, 0))
1297 goto redirty_out;
1298
1299 err = -EAGAIN;
1300 f2fs_lock_op(sbi);
1301 if (f2fs_has_inline_data(inode))
1302 err = f2fs_write_inline_data(inode, page);
1303 if (err == -EAGAIN)
1304 err = do_write_data_page(&fio);
1305 if (F2FS_I(inode)->last_disk_size < psize)
1306 F2FS_I(inode)->last_disk_size = psize;
1307 f2fs_unlock_op(sbi);
1308 done:
1309 if (err && err != -ENOENT)
1310 goto redirty_out;
1311
1312 clear_cold_data(page);
1313 out:
1314 inode_dec_dirty_pages(inode);
1315 if (err)
1316 ClearPageUptodate(page);
1317
1318 if (wbc->for_reclaim) {
1319 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1320 remove_dirty_inode(inode);
1321 }
1322
1323 unlock_page(page);
1324 f2fs_balance_fs(sbi, need_balance_fs);
1325
1326 if (unlikely(f2fs_cp_error(sbi)))
1327 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1328
1329 return 0;
1330
1331 redirty_out:
1332 redirty_page_for_writepage(wbc, page);
1333 unlock_page(page);
1334 return err;
1335 }
1336
1337 /*
1338 * This function was copied from write_cche_pages from mm/page-writeback.c.
1339 * The major change is making write step of cold data page separately from
1340 * warm/hot data page.
1341 */
1342 static int f2fs_write_cache_pages(struct address_space *mapping,
1343 struct writeback_control *wbc)
1344 {
1345 int ret = 0;
1346 int done = 0;
1347 struct pagevec pvec;
1348 int nr_pages;
1349 pgoff_t uninitialized_var(writeback_index);
1350 pgoff_t index;
1351 pgoff_t end; /* Inclusive */
1352 pgoff_t done_index;
1353 int cycled;
1354 int range_whole = 0;
1355 int tag;
1356
1357 pagevec_init(&pvec, 0);
1358
1359 if (wbc->range_cyclic) {
1360 writeback_index = mapping->writeback_index; /* prev offset */
1361 index = writeback_index;
1362 if (index == 0)
1363 cycled = 1;
1364 else
1365 cycled = 0;
1366 end = -1;
1367 } else {
1368 index = wbc->range_start >> PAGE_SHIFT;
1369 end = wbc->range_end >> PAGE_SHIFT;
1370 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1371 range_whole = 1;
1372 cycled = 1; /* ignore range_cyclic tests */
1373 }
1374 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1375 tag = PAGECACHE_TAG_TOWRITE;
1376 else
1377 tag = PAGECACHE_TAG_DIRTY;
1378 retry:
1379 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1380 tag_pages_for_writeback(mapping, index, end);
1381 done_index = index;
1382 while (!done && (index <= end)) {
1383 int i;
1384
1385 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1386 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1387 if (nr_pages == 0)
1388 break;
1389
1390 for (i = 0; i < nr_pages; i++) {
1391 struct page *page = pvec.pages[i];
1392
1393 if (page->index > end) {
1394 done = 1;
1395 break;
1396 }
1397
1398 done_index = page->index;
1399
1400 lock_page(page);
1401
1402 if (unlikely(page->mapping != mapping)) {
1403 continue_unlock:
1404 unlock_page(page);
1405 continue;
1406 }
1407
1408 if (!PageDirty(page)) {
1409 /* someone wrote it for us */
1410 goto continue_unlock;
1411 }
1412
1413 if (PageWriteback(page)) {
1414 if (wbc->sync_mode != WB_SYNC_NONE)
1415 f2fs_wait_on_page_writeback(page,
1416 DATA, true);
1417 else
1418 goto continue_unlock;
1419 }
1420
1421 BUG_ON(PageWriteback(page));
1422 if (!clear_page_dirty_for_io(page))
1423 goto continue_unlock;
1424
1425 ret = mapping->a_ops->writepage(page, wbc);
1426 if (unlikely(ret)) {
1427 done_index = page->index + 1;
1428 done = 1;
1429 break;
1430 }
1431
1432 if (--wbc->nr_to_write <= 0 &&
1433 wbc->sync_mode == WB_SYNC_NONE) {
1434 done = 1;
1435 break;
1436 }
1437 }
1438 pagevec_release(&pvec);
1439 cond_resched();
1440 }
1441
1442 if (!cycled && !done) {
1443 cycled = 1;
1444 index = 0;
1445 end = writeback_index - 1;
1446 goto retry;
1447 }
1448 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1449 mapping->writeback_index = done_index;
1450
1451 return ret;
1452 }
1453
1454 static int f2fs_write_data_pages(struct address_space *mapping,
1455 struct writeback_control *wbc)
1456 {
1457 struct inode *inode = mapping->host;
1458 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1459 struct blk_plug plug;
1460 int ret;
1461
1462 /* deal with chardevs and other special file */
1463 if (!mapping->a_ops->writepage)
1464 return 0;
1465
1466 /* skip writing if there is no dirty page in this inode */
1467 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1468 return 0;
1469
1470 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1471 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1472 available_free_memory(sbi, DIRTY_DENTS))
1473 goto skip_write;
1474
1475 /* skip writing during file defragment */
1476 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1477 goto skip_write;
1478
1479 /* during POR, we don't need to trigger writepage at all. */
1480 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1481 goto skip_write;
1482
1483 trace_f2fs_writepages(mapping->host, wbc, DATA);
1484
1485 blk_start_plug(&plug);
1486 ret = f2fs_write_cache_pages(mapping, wbc);
1487 blk_finish_plug(&plug);
1488 /*
1489 * if some pages were truncated, we cannot guarantee its mapping->host
1490 * to detect pending bios.
1491 */
1492 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1493
1494 remove_dirty_inode(inode);
1495 return ret;
1496
1497 skip_write:
1498 wbc->pages_skipped += get_dirty_pages(inode);
1499 trace_f2fs_writepages(mapping->host, wbc, DATA);
1500 return 0;
1501 }
1502
1503 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1504 {
1505 struct inode *inode = mapping->host;
1506 loff_t i_size = i_size_read(inode);
1507
1508 if (to > i_size) {
1509 truncate_pagecache(inode, i_size);
1510 truncate_blocks(inode, i_size, true);
1511 }
1512 }
1513
1514 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1515 struct page *page, loff_t pos, unsigned len,
1516 block_t *blk_addr, bool *node_changed)
1517 {
1518 struct inode *inode = page->mapping->host;
1519 pgoff_t index = page->index;
1520 struct dnode_of_data dn;
1521 struct page *ipage;
1522 bool locked = false;
1523 struct extent_info ei;
1524 int err = 0;
1525
1526 /*
1527 * we already allocated all the blocks, so we don't need to get
1528 * the block addresses when there is no need to fill the page.
1529 */
1530 if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1531 len == PAGE_SIZE)
1532 return 0;
1533
1534 if (f2fs_has_inline_data(inode) ||
1535 (pos & PAGE_MASK) >= i_size_read(inode)) {
1536 f2fs_lock_op(sbi);
1537 locked = true;
1538 }
1539 restart:
1540 /* check inline_data */
1541 ipage = get_node_page(sbi, inode->i_ino);
1542 if (IS_ERR(ipage)) {
1543 err = PTR_ERR(ipage);
1544 goto unlock_out;
1545 }
1546
1547 set_new_dnode(&dn, inode, ipage, ipage, 0);
1548
1549 if (f2fs_has_inline_data(inode)) {
1550 if (pos + len <= MAX_INLINE_DATA) {
1551 read_inline_data(page, ipage);
1552 set_inode_flag(inode, FI_DATA_EXIST);
1553 if (inode->i_nlink)
1554 set_inline_node(ipage);
1555 } else {
1556 err = f2fs_convert_inline_page(&dn, page);
1557 if (err)
1558 goto out;
1559 if (dn.data_blkaddr == NULL_ADDR)
1560 err = f2fs_get_block(&dn, index);
1561 }
1562 } else if (locked) {
1563 err = f2fs_get_block(&dn, index);
1564 } else {
1565 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1566 dn.data_blkaddr = ei.blk + index - ei.fofs;
1567 } else {
1568 /* hole case */
1569 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1570 if (err || dn.data_blkaddr == NULL_ADDR) {
1571 f2fs_put_dnode(&dn);
1572 f2fs_lock_op(sbi);
1573 locked = true;
1574 goto restart;
1575 }
1576 }
1577 }
1578
1579 /* convert_inline_page can make node_changed */
1580 *blk_addr = dn.data_blkaddr;
1581 *node_changed = dn.node_changed;
1582 out:
1583 f2fs_put_dnode(&dn);
1584 unlock_out:
1585 if (locked)
1586 f2fs_unlock_op(sbi);
1587 return err;
1588 }
1589
1590 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1591 loff_t pos, unsigned len, unsigned flags,
1592 struct page **pagep, void **fsdata)
1593 {
1594 struct inode *inode = mapping->host;
1595 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1596 struct page *page = NULL;
1597 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1598 bool need_balance = false;
1599 block_t blkaddr = NULL_ADDR;
1600 int err = 0;
1601
1602 trace_f2fs_write_begin(inode, pos, len, flags);
1603
1604 /*
1605 * We should check this at this moment to avoid deadlock on inode page
1606 * and #0 page. The locking rule for inline_data conversion should be:
1607 * lock_page(page #0) -> lock_page(inode_page)
1608 */
1609 if (index != 0) {
1610 err = f2fs_convert_inline_inode(inode);
1611 if (err)
1612 goto fail;
1613 }
1614 repeat:
1615 page = grab_cache_page_write_begin(mapping, index, flags);
1616 if (!page) {
1617 err = -ENOMEM;
1618 goto fail;
1619 }
1620
1621 *pagep = page;
1622
1623 err = prepare_write_begin(sbi, page, pos, len,
1624 &blkaddr, &need_balance);
1625 if (err)
1626 goto fail;
1627
1628 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1629 unlock_page(page);
1630 f2fs_balance_fs(sbi, true);
1631 lock_page(page);
1632 if (page->mapping != mapping) {
1633 /* The page got truncated from under us */
1634 f2fs_put_page(page, 1);
1635 goto repeat;
1636 }
1637 }
1638
1639 f2fs_wait_on_page_writeback(page, DATA, false);
1640
1641 /* wait for GCed encrypted page writeback */
1642 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1643 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1644
1645 if (len == PAGE_SIZE || PageUptodate(page))
1646 return 0;
1647
1648 if (blkaddr == NEW_ADDR) {
1649 zero_user_segment(page, 0, PAGE_SIZE);
1650 } else {
1651 struct bio *bio;
1652
1653 bio = f2fs_grab_bio(inode, blkaddr, 1);
1654 if (IS_ERR(bio)) {
1655 err = PTR_ERR(bio);
1656 goto fail;
1657 }
1658 bio_set_op_attrs(bio, REQ_OP_READ, READ_SYNC);
1659 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1660 bio_put(bio);
1661 err = -EFAULT;
1662 goto fail;
1663 }
1664
1665 __submit_bio(sbi, bio, DATA);
1666
1667 lock_page(page);
1668 if (unlikely(page->mapping != mapping)) {
1669 f2fs_put_page(page, 1);
1670 goto repeat;
1671 }
1672 if (unlikely(!PageUptodate(page))) {
1673 err = -EIO;
1674 goto fail;
1675 }
1676 }
1677 if (!PageUptodate(page))
1678 SetPageUptodate(page);
1679 return 0;
1680
1681 fail:
1682 f2fs_put_page(page, 1);
1683 f2fs_write_failed(mapping, pos + len);
1684 return err;
1685 }
1686
1687 static int f2fs_write_end(struct file *file,
1688 struct address_space *mapping,
1689 loff_t pos, unsigned len, unsigned copied,
1690 struct page *page, void *fsdata)
1691 {
1692 struct inode *inode = page->mapping->host;
1693
1694 trace_f2fs_write_end(inode, pos, len, copied);
1695
1696 if (!copied)
1697 goto unlock_out;
1698
1699 /*
1700 * This should be come from len == PAGE_SIZE, so we need to fill out
1701 * zeros in the unwritten remaining space.
1702 * The flow was copied from fuse_write_end().
1703 */
1704 if (!PageUptodate(page)) {
1705 size_t endoff = (pos + copied) & ~PAGE_MASK;
1706
1707 if (endoff)
1708 zero_user_segment(page, endoff, PAGE_SIZE);
1709 SetPageUptodate(page);
1710 }
1711
1712 set_page_dirty(page);
1713 clear_cold_data(page);
1714
1715 if (pos + copied > i_size_read(inode))
1716 f2fs_i_size_write(inode, pos + copied);
1717 unlock_out:
1718 f2fs_put_page(page, 1);
1719 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1720 return copied;
1721 }
1722
1723 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1724 loff_t offset)
1725 {
1726 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1727
1728 if (offset & blocksize_mask)
1729 return -EINVAL;
1730
1731 if (iov_iter_alignment(iter) & blocksize_mask)
1732 return -EINVAL;
1733
1734 return 0;
1735 }
1736
1737 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1738 {
1739 struct address_space *mapping = iocb->ki_filp->f_mapping;
1740 struct inode *inode = mapping->host;
1741 size_t count = iov_iter_count(iter);
1742 loff_t offset = iocb->ki_pos;
1743 int rw = iov_iter_rw(iter);
1744 int err;
1745
1746 err = check_direct_IO(inode, iter, offset);
1747 if (err)
1748 return err;
1749
1750 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1751 return 0;
1752 if (test_opt(F2FS_I_SB(inode), LFS))
1753 return 0;
1754
1755 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
1756
1757 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
1758 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
1759 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
1760
1761 if (rw == WRITE) {
1762 if (err > 0)
1763 set_inode_flag(inode, FI_UPDATE_WRITE);
1764 else if (err < 0)
1765 f2fs_write_failed(mapping, offset + count);
1766 }
1767
1768 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
1769
1770 return err;
1771 }
1772
1773 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1774 unsigned int length)
1775 {
1776 struct inode *inode = page->mapping->host;
1777 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1778
1779 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1780 (offset % PAGE_SIZE || length != PAGE_SIZE))
1781 return;
1782
1783 if (PageDirty(page)) {
1784 if (inode->i_ino == F2FS_META_INO(sbi))
1785 dec_page_count(sbi, F2FS_DIRTY_META);
1786 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1787 dec_page_count(sbi, F2FS_DIRTY_NODES);
1788 else
1789 inode_dec_dirty_pages(inode);
1790 }
1791
1792 /* This is atomic written page, keep Private */
1793 if (IS_ATOMIC_WRITTEN_PAGE(page))
1794 return;
1795
1796 set_page_private(page, 0);
1797 ClearPagePrivate(page);
1798 }
1799
1800 int f2fs_release_page(struct page *page, gfp_t wait)
1801 {
1802 /* If this is dirty page, keep PagePrivate */
1803 if (PageDirty(page))
1804 return 0;
1805
1806 /* This is atomic written page, keep Private */
1807 if (IS_ATOMIC_WRITTEN_PAGE(page))
1808 return 0;
1809
1810 set_page_private(page, 0);
1811 ClearPagePrivate(page);
1812 return 1;
1813 }
1814
1815 /*
1816 * This was copied from __set_page_dirty_buffers which gives higher performance
1817 * in very high speed storages. (e.g., pmem)
1818 */
1819 void f2fs_set_page_dirty_nobuffers(struct page *page)
1820 {
1821 struct address_space *mapping = page->mapping;
1822 unsigned long flags;
1823
1824 if (unlikely(!mapping))
1825 return;
1826
1827 spin_lock(&mapping->private_lock);
1828 lock_page_memcg(page);
1829 SetPageDirty(page);
1830 spin_unlock(&mapping->private_lock);
1831
1832 spin_lock_irqsave(&mapping->tree_lock, flags);
1833 WARN_ON_ONCE(!PageUptodate(page));
1834 account_page_dirtied(page, mapping);
1835 radix_tree_tag_set(&mapping->page_tree,
1836 page_index(page), PAGECACHE_TAG_DIRTY);
1837 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1838 unlock_page_memcg(page);
1839
1840 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1841 return;
1842 }
1843
1844 static int f2fs_set_data_page_dirty(struct page *page)
1845 {
1846 struct address_space *mapping = page->mapping;
1847 struct inode *inode = mapping->host;
1848
1849 trace_f2fs_set_page_dirty(page, DATA);
1850
1851 if (!PageUptodate(page))
1852 SetPageUptodate(page);
1853
1854 if (f2fs_is_atomic_file(inode)) {
1855 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1856 register_inmem_page(inode, page);
1857 return 1;
1858 }
1859 /*
1860 * Previously, this page has been registered, we just
1861 * return here.
1862 */
1863 return 0;
1864 }
1865
1866 if (!PageDirty(page)) {
1867 f2fs_set_page_dirty_nobuffers(page);
1868 update_dirty_page(inode, page);
1869 return 1;
1870 }
1871 return 0;
1872 }
1873
1874 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1875 {
1876 struct inode *inode = mapping->host;
1877
1878 if (f2fs_has_inline_data(inode))
1879 return 0;
1880
1881 /* make sure allocating whole blocks */
1882 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1883 filemap_write_and_wait(mapping);
1884
1885 return generic_block_bmap(mapping, block, get_data_block_bmap);
1886 }
1887
1888 const struct address_space_operations f2fs_dblock_aops = {
1889 .readpage = f2fs_read_data_page,
1890 .readpages = f2fs_read_data_pages,
1891 .writepage = f2fs_write_data_page,
1892 .writepages = f2fs_write_data_pages,
1893 .write_begin = f2fs_write_begin,
1894 .write_end = f2fs_write_end,
1895 .set_page_dirty = f2fs_set_data_page_dirty,
1896 .invalidatepage = f2fs_invalidate_page,
1897 .releasepage = f2fs_release_page,
1898 .direct_IO = f2fs_direct_IO,
1899 .bmap = f2fs_bmap,
1900 };
Linux kernel - Deliverables
This page took 0.068118 seconds and 5 git commands to generate.