projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: introduce f2fs_flush_merged_bios for cleanup
[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 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
118 struct page *page, nid_t ino)
119 {
120 struct bio_vec *bvec;
121 struct page *target;
122 int i;
123
124 if (!io->bio)
125 return false;
126
127 if (!inode && !page && !ino)
128 return true;
129
130 bio_for_each_segment_all(bvec, io->bio, i) {
131
132 if (bvec->bv_page->mapping) {
133 target = bvec->bv_page;
134 } else {
135 struct f2fs_crypto_ctx *ctx;
136
137 /* encrypted page */
138 ctx = (struct f2fs_crypto_ctx *)page_private(
139 bvec->bv_page);
140 target = ctx->w.control_page;
141 }
142
143 if (inode && inode == target->mapping->host)
144 return true;
145 if (page && page == target)
146 return true;
147 if (ino && ino == ino_of_node(target))
148 return true;
149 }
150
151 return false;
152 }
153
154 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
155 struct page *page, nid_t ino,
156 enum page_type type)
157 {
158 enum page_type btype = PAGE_TYPE_OF_BIO(type);
159 struct f2fs_bio_info *io = &sbi->write_io[btype];
160 bool ret;
161
162 down_read(&io->io_rwsem);
163 ret = __has_merged_page(io, inode, page, ino);
164 up_read(&io->io_rwsem);
165 return ret;
166 }
167
168 static void __f2fs_submit_merged_bio(struct f2fs_sb_info *sbi,
169 struct inode *inode, struct page *page,
170 nid_t ino, enum page_type type, int rw)
171 {
172 enum page_type btype = PAGE_TYPE_OF_BIO(type);
173 struct f2fs_bio_info *io;
174
175 io = is_read_io(rw) ? &sbi->read_io : &sbi->write_io[btype];
176
177 down_write(&io->io_rwsem);
178
179 if (!__has_merged_page(io, inode, page, ino))
180 goto out;
181
182 /* change META to META_FLUSH in the checkpoint procedure */
183 if (type >= META_FLUSH) {
184 io->fio.type = META_FLUSH;
185 if (test_opt(sbi, NOBARRIER))
186 io->fio.rw = WRITE_FLUSH | REQ_META | REQ_PRIO;
187 else
188 io->fio.rw = WRITE_FLUSH_FUA | REQ_META | REQ_PRIO;
189 }
190 __submit_merged_bio(io);
191 out:
192 up_write(&io->io_rwsem);
193 }
194
195 void f2fs_submit_merged_bio(struct f2fs_sb_info *sbi, enum page_type type,
196 int rw)
197 {
198 __f2fs_submit_merged_bio(sbi, NULL, NULL, 0, type, rw);
199 }
200
201 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info *sbi,
202 struct inode *inode, struct page *page,
203 nid_t ino, enum page_type type, int rw)
204 {
205 if (has_merged_page(sbi, inode, page, ino, type))
206 __f2fs_submit_merged_bio(sbi, inode, page, ino, type, rw);
207 }
208
209 void f2fs_flush_merged_bios(struct f2fs_sb_info *sbi)
210 {
211 f2fs_submit_merged_bio(sbi, DATA, WRITE);
212 f2fs_submit_merged_bio(sbi, NODE, WRITE);
213 f2fs_submit_merged_bio(sbi, META, WRITE);
214 }
215
216 /*
217 * Fill the locked page with data located in the block address.
218 * Return unlocked page.
219 */
220 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
221 {
222 struct bio *bio;
223 struct page *page = fio->encrypted_page ? fio->encrypted_page : fio->page;
224
225 trace_f2fs_submit_page_bio(page, fio);
226 f2fs_trace_ios(fio, 0);
227
228 /* Allocate a new bio */
229 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->rw));
230
231 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
232 bio_put(bio);
233 return -EFAULT;
234 }
235
236 submit_bio(fio->rw, bio);
237 return 0;
238 }
239
240 void f2fs_submit_page_mbio(struct f2fs_io_info *fio)
241 {
242 struct f2fs_sb_info *sbi = fio->sbi;
243 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
244 struct f2fs_bio_info *io;
245 bool is_read = is_read_io(fio->rw);
246 struct page *bio_page;
247
248 io = is_read ? &sbi->read_io : &sbi->write_io[btype];
249
250 if (fio->old_blkaddr != NEW_ADDR)
251 verify_block_addr(sbi, fio->old_blkaddr);
252 verify_block_addr(sbi, fio->new_blkaddr);
253
254 down_write(&io->io_rwsem);
255
256 if (!is_read)
257 inc_page_count(sbi, F2FS_WRITEBACK);
258
259 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
260 io->fio.rw != fio->rw))
261 __submit_merged_bio(io);
262 alloc_new:
263 if (io->bio == NULL) {
264 int bio_blocks = MAX_BIO_BLOCKS(sbi);
265
266 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
267 bio_blocks, is_read);
268 io->fio = *fio;
269 }
270
271 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
272
273 if (bio_add_page(io->bio, bio_page, PAGE_CACHE_SIZE, 0) <
274 PAGE_CACHE_SIZE) {
275 __submit_merged_bio(io);
276 goto alloc_new;
277 }
278
279 io->last_block_in_bio = fio->new_blkaddr;
280 f2fs_trace_ios(fio, 0);
281
282 up_write(&io->io_rwsem);
283 trace_f2fs_submit_page_mbio(fio->page, fio);
284 }
285
286 /*
287 * Lock ordering for the change of data block address:
288 * ->data_page
289 * ->node_page
290 * update block addresses in the node page
291 */
292 void set_data_blkaddr(struct dnode_of_data *dn)
293 {
294 struct f2fs_node *rn;
295 __le32 *addr_array;
296 struct page *node_page = dn->node_page;
297 unsigned int ofs_in_node = dn->ofs_in_node;
298
299 f2fs_wait_on_page_writeback(node_page, NODE, true);
300
301 rn = F2FS_NODE(node_page);
302
303 /* Get physical address of data block */
304 addr_array = blkaddr_in_node(rn);
305 addr_array[ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
306 if (set_page_dirty(node_page))
307 dn->node_changed = true;
308 }
309
310 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
311 {
312 dn->data_blkaddr = blkaddr;
313 set_data_blkaddr(dn);
314 f2fs_update_extent_cache(dn);
315 }
316
317 int reserve_new_block(struct dnode_of_data *dn)
318 {
319 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
320
321 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
322 return -EPERM;
323 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
324 return -ENOSPC;
325
326 trace_f2fs_reserve_new_block(dn->inode, dn->nid, dn->ofs_in_node);
327
328 dn->data_blkaddr = NEW_ADDR;
329 set_data_blkaddr(dn);
330 mark_inode_dirty(dn->inode);
331 sync_inode_page(dn);
332 return 0;
333 }
334
335 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
336 {
337 bool need_put = dn->inode_page ? false : true;
338 int err;
339
340 err = get_dnode_of_data(dn, index, ALLOC_NODE);
341 if (err)
342 return err;
343
344 if (dn->data_blkaddr == NULL_ADDR)
345 err = reserve_new_block(dn);
346 if (err || need_put)
347 f2fs_put_dnode(dn);
348 return err;
349 }
350
351 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
352 {
353 struct extent_info ei;
354 struct inode *inode = dn->inode;
355
356 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
357 dn->data_blkaddr = ei.blk + index - ei.fofs;
358 return 0;
359 }
360
361 return f2fs_reserve_block(dn, index);
362 }
363
364 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
365 int rw, bool for_write)
366 {
367 struct address_space *mapping = inode->i_mapping;
368 struct dnode_of_data dn;
369 struct page *page;
370 struct extent_info ei;
371 int err;
372 struct f2fs_io_info fio = {
373 .sbi = F2FS_I_SB(inode),
374 .type = DATA,
375 .rw = rw,
376 .encrypted_page = NULL,
377 };
378
379 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
380 return read_mapping_page(mapping, index, NULL);
381
382 page = f2fs_grab_cache_page(mapping, index, for_write);
383 if (!page)
384 return ERR_PTR(-ENOMEM);
385
386 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
387 dn.data_blkaddr = ei.blk + index - ei.fofs;
388 goto got_it;
389 }
390
391 set_new_dnode(&dn, inode, NULL, NULL, 0);
392 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
393 if (err)
394 goto put_err;
395 f2fs_put_dnode(&dn);
396
397 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
398 err = -ENOENT;
399 goto put_err;
400 }
401 got_it:
402 if (PageUptodate(page)) {
403 unlock_page(page);
404 return page;
405 }
406
407 /*
408 * A new dentry page is allocated but not able to be written, since its
409 * new inode page couldn't be allocated due to -ENOSPC.
410 * In such the case, its blkaddr can be remained as NEW_ADDR.
411 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
412 */
413 if (dn.data_blkaddr == NEW_ADDR) {
414 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
415 SetPageUptodate(page);
416 unlock_page(page);
417 return page;
418 }
419
420 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
421 fio.page = page;
422 err = f2fs_submit_page_bio(&fio);
423 if (err)
424 goto put_err;
425 return page;
426
427 put_err:
428 f2fs_put_page(page, 1);
429 return ERR_PTR(err);
430 }
431
432 struct page *find_data_page(struct inode *inode, pgoff_t index)
433 {
434 struct address_space *mapping = inode->i_mapping;
435 struct page *page;
436
437 page = find_get_page(mapping, index);
438 if (page && PageUptodate(page))
439 return page;
440 f2fs_put_page(page, 0);
441
442 page = get_read_data_page(inode, index, READ_SYNC, false);
443 if (IS_ERR(page))
444 return page;
445
446 if (PageUptodate(page))
447 return page;
448
449 wait_on_page_locked(page);
450 if (unlikely(!PageUptodate(page))) {
451 f2fs_put_page(page, 0);
452 return ERR_PTR(-EIO);
453 }
454 return page;
455 }
456
457 /*
458 * If it tries to access a hole, return an error.
459 * Because, the callers, functions in dir.c and GC, should be able to know
460 * whether this page exists or not.
461 */
462 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
463 bool for_write)
464 {
465 struct address_space *mapping = inode->i_mapping;
466 struct page *page;
467 repeat:
468 page = get_read_data_page(inode, index, READ_SYNC, for_write);
469 if (IS_ERR(page))
470 return page;
471
472 /* wait for read completion */
473 lock_page(page);
474 if (unlikely(!PageUptodate(page))) {
475 f2fs_put_page(page, 1);
476 return ERR_PTR(-EIO);
477 }
478 if (unlikely(page->mapping != mapping)) {
479 f2fs_put_page(page, 1);
480 goto repeat;
481 }
482 return page;
483 }
484
485 /*
486 * Caller ensures that this data page is never allocated.
487 * A new zero-filled data page is allocated in the page cache.
488 *
489 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
490 * f2fs_unlock_op().
491 * Note that, ipage is set only by make_empty_dir, and if any error occur,
492 * ipage should be released by this function.
493 */
494 struct page *get_new_data_page(struct inode *inode,
495 struct page *ipage, pgoff_t index, bool new_i_size)
496 {
497 struct address_space *mapping = inode->i_mapping;
498 struct page *page;
499 struct dnode_of_data dn;
500 int err;
501
502 page = f2fs_grab_cache_page(mapping, index, true);
503 if (!page) {
504 /*
505 * before exiting, we should make sure ipage will be released
506 * if any error occur.
507 */
508 f2fs_put_page(ipage, 1);
509 return ERR_PTR(-ENOMEM);
510 }
511
512 set_new_dnode(&dn, inode, ipage, NULL, 0);
513 err = f2fs_reserve_block(&dn, index);
514 if (err) {
515 f2fs_put_page(page, 1);
516 return ERR_PTR(err);
517 }
518 if (!ipage)
519 f2fs_put_dnode(&dn);
520
521 if (PageUptodate(page))
522 goto got_it;
523
524 if (dn.data_blkaddr == NEW_ADDR) {
525 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
526 SetPageUptodate(page);
527 } else {
528 f2fs_put_page(page, 1);
529
530 /* if ipage exists, blkaddr should be NEW_ADDR */
531 f2fs_bug_on(F2FS_I_SB(inode), ipage);
532 page = get_lock_data_page(inode, index, true);
533 if (IS_ERR(page))
534 return page;
535 }
536 got_it:
537 if (new_i_size && i_size_read(inode) <
538 ((loff_t)(index + 1) << PAGE_CACHE_SHIFT)) {
539 i_size_write(inode, ((loff_t)(index + 1) << PAGE_CACHE_SHIFT));
540 /* Only the directory inode sets new_i_size */
541 set_inode_flag(F2FS_I(inode), FI_UPDATE_DIR);
542 }
543 return page;
544 }
545
546 static int __allocate_data_block(struct dnode_of_data *dn)
547 {
548 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
549 struct f2fs_summary sum;
550 struct node_info ni;
551 int seg = CURSEG_WARM_DATA;
552 pgoff_t fofs;
553
554 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
555 return -EPERM;
556
557 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
558 if (dn->data_blkaddr == NEW_ADDR)
559 goto alloc;
560
561 if (unlikely(!inc_valid_block_count(sbi, dn->inode, 1)))
562 return -ENOSPC;
563
564 alloc:
565 get_node_info(sbi, dn->nid, &ni);
566 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
567
568 if (dn->ofs_in_node == 0 && dn->inode_page == dn->node_page)
569 seg = CURSEG_DIRECT_IO;
570
571 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
572 &sum, seg);
573 set_data_blkaddr(dn);
574
575 /* update i_size */
576 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
577 dn->ofs_in_node;
578 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT))
579 i_size_write(dn->inode,
580 ((loff_t)(fofs + 1) << PAGE_CACHE_SHIFT));
581 return 0;
582 }
583
584 ssize_t f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
585 {
586 struct inode *inode = file_inode(iocb->ki_filp);
587 struct f2fs_map_blocks map;
588 ssize_t ret = 0;
589
590 map.m_lblk = F2FS_BYTES_TO_BLK(iocb->ki_pos);
591 map.m_len = F2FS_BLK_ALIGN(iov_iter_count(from));
592 map.m_next_pgofs = NULL;
593
594 if (f2fs_encrypted_inode(inode))
595 return 0;
596
597 if (iocb->ki_flags & IOCB_DIRECT) {
598 ret = f2fs_convert_inline_inode(inode);
599 if (ret)
600 return ret;
601 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
602 }
603 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA) {
604 ret = f2fs_convert_inline_inode(inode);
605 if (ret)
606 return ret;
607 }
608 if (!f2fs_has_inline_data(inode))
609 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
610 return ret;
611 }
612
613 /*
614 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
615 * f2fs_map_blocks structure.
616 * If original data blocks are allocated, then give them to blockdev.
617 * Otherwise,
618 * a. preallocate requested block addresses
619 * b. do not use extent cache for better performance
620 * c. give the block addresses to blockdev
621 */
622 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
623 int create, int flag)
624 {
625 unsigned int maxblocks = map->m_len;
626 struct dnode_of_data dn;
627 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
628 int mode = create ? ALLOC_NODE : LOOKUP_NODE_RA;
629 pgoff_t pgofs, end_offset;
630 int err = 0, ofs = 1;
631 struct extent_info ei;
632 bool allocated = false;
633 block_t blkaddr;
634
635 map->m_len = 0;
636 map->m_flags = 0;
637
638 /* it only supports block size == page size */
639 pgofs = (pgoff_t)map->m_lblk;
640
641 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
642 map->m_pblk = ei.blk + pgofs - ei.fofs;
643 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
644 map->m_flags = F2FS_MAP_MAPPED;
645 goto out;
646 }
647
648 next_dnode:
649 if (create)
650 f2fs_lock_op(sbi);
651
652 /* When reading holes, we need its node page */
653 set_new_dnode(&dn, inode, NULL, NULL, 0);
654 err = get_dnode_of_data(&dn, pgofs, mode);
655 if (err) {
656 if (err == -ENOENT) {
657 err = 0;
658 if (map->m_next_pgofs)
659 *map->m_next_pgofs =
660 get_next_page_offset(&dn, pgofs);
661 }
662 goto unlock_out;
663 }
664
665 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
666
667 next_block:
668 blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
669
670 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
671 if (create) {
672 if (unlikely(f2fs_cp_error(sbi))) {
673 err = -EIO;
674 goto sync_out;
675 }
676 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
677 if (blkaddr == NULL_ADDR)
678 err = reserve_new_block(&dn);
679 } else {
680 err = __allocate_data_block(&dn);
681 }
682 if (err)
683 goto sync_out;
684 allocated = true;
685 map->m_flags = F2FS_MAP_NEW;
686 blkaddr = dn.data_blkaddr;
687 } else {
688 if (flag == F2FS_GET_BLOCK_FIEMAP &&
689 blkaddr == NULL_ADDR) {
690 if (map->m_next_pgofs)
691 *map->m_next_pgofs = pgofs + 1;
692 }
693 if (flag != F2FS_GET_BLOCK_FIEMAP ||
694 blkaddr != NEW_ADDR) {
695 if (flag == F2FS_GET_BLOCK_BMAP)
696 err = -ENOENT;
697 goto sync_out;
698 }
699 }
700 }
701
702 if (map->m_len == 0) {
703 /* preallocated unwritten block should be mapped for fiemap. */
704 if (blkaddr == NEW_ADDR)
705 map->m_flags |= F2FS_MAP_UNWRITTEN;
706 map->m_flags |= F2FS_MAP_MAPPED;
707
708 map->m_pblk = blkaddr;
709 map->m_len = 1;
710 } else if ((map->m_pblk != NEW_ADDR &&
711 blkaddr == (map->m_pblk + ofs)) ||
712 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
713 flag == F2FS_GET_BLOCK_PRE_DIO ||
714 flag == F2FS_GET_BLOCK_PRE_AIO) {
715 ofs++;
716 map->m_len++;
717 } else {
718 goto sync_out;
719 }
720
721 dn.ofs_in_node++;
722 pgofs++;
723
724 if (map->m_len < maxblocks) {
725 if (dn.ofs_in_node < end_offset)
726 goto next_block;
727
728 if (allocated)
729 sync_inode_page(&dn);
730 f2fs_put_dnode(&dn);
731
732 if (create) {
733 f2fs_unlock_op(sbi);
734 f2fs_balance_fs(sbi, allocated);
735 }
736 allocated = false;
737 goto next_dnode;
738 }
739
740 sync_out:
741 if (allocated)
742 sync_inode_page(&dn);
743 f2fs_put_dnode(&dn);
744 unlock_out:
745 if (create) {
746 f2fs_unlock_op(sbi);
747 f2fs_balance_fs(sbi, allocated);
748 }
749 out:
750 trace_f2fs_map_blocks(inode, map, err);
751 return err;
752 }
753
754 static int __get_data_block(struct inode *inode, sector_t iblock,
755 struct buffer_head *bh, int create, int flag,
756 pgoff_t *next_pgofs)
757 {
758 struct f2fs_map_blocks map;
759 int ret;
760
761 map.m_lblk = iblock;
762 map.m_len = bh->b_size >> inode->i_blkbits;
763 map.m_next_pgofs = next_pgofs;
764
765 ret = f2fs_map_blocks(inode, &map, create, flag);
766 if (!ret) {
767 map_bh(bh, inode->i_sb, map.m_pblk);
768 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
769 bh->b_size = map.m_len << inode->i_blkbits;
770 }
771 return ret;
772 }
773
774 static int get_data_block(struct inode *inode, sector_t iblock,
775 struct buffer_head *bh_result, int create, int flag,
776 pgoff_t *next_pgofs)
777 {
778 return __get_data_block(inode, iblock, bh_result, create,
779 flag, next_pgofs);
780 }
781
782 static int get_data_block_dio(struct inode *inode, sector_t iblock,
783 struct buffer_head *bh_result, int create)
784 {
785 return __get_data_block(inode, iblock, bh_result, create,
786 F2FS_GET_BLOCK_DIO, NULL);
787 }
788
789 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
790 struct buffer_head *bh_result, int create)
791 {
792 /* Block number less than F2FS MAX BLOCKS */
793 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
794 return -EFBIG;
795
796 return __get_data_block(inode, iblock, bh_result, create,
797 F2FS_GET_BLOCK_BMAP, NULL);
798 }
799
800 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
801 {
802 return (offset >> inode->i_blkbits);
803 }
804
805 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
806 {
807 return (blk << inode->i_blkbits);
808 }
809
810 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
811 u64 start, u64 len)
812 {
813 struct buffer_head map_bh;
814 sector_t start_blk, last_blk;
815 pgoff_t next_pgofs;
816 loff_t isize;
817 u64 logical = 0, phys = 0, size = 0;
818 u32 flags = 0;
819 int ret = 0;
820
821 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
822 if (ret)
823 return ret;
824
825 if (f2fs_has_inline_data(inode)) {
826 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
827 if (ret != -EAGAIN)
828 return ret;
829 }
830
831 inode_lock(inode);
832
833 isize = i_size_read(inode);
834 if (start >= isize)
835 goto out;
836
837 if (start + len > isize)
838 len = isize - start;
839
840 if (logical_to_blk(inode, len) == 0)
841 len = blk_to_logical(inode, 1);
842
843 start_blk = logical_to_blk(inode, start);
844 last_blk = logical_to_blk(inode, start + len - 1);
845
846 next:
847 memset(&map_bh, 0, sizeof(struct buffer_head));
848 map_bh.b_size = len;
849
850 ret = get_data_block(inode, start_blk, &map_bh, 0,
851 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
852 if (ret)
853 goto out;
854
855 /* HOLE */
856 if (!buffer_mapped(&map_bh)) {
857 start_blk = next_pgofs;
858 /* Go through holes util pass the EOF */
859 if (blk_to_logical(inode, start_blk) < isize)
860 goto prep_next;
861 /* Found a hole beyond isize means no more extents.
862 * Note that the premise is that filesystems don't
863 * punch holes beyond isize and keep size unchanged.
864 */
865 flags |= FIEMAP_EXTENT_LAST;
866 }
867
868 if (size) {
869 if (f2fs_encrypted_inode(inode))
870 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
871
872 ret = fiemap_fill_next_extent(fieinfo, logical,
873 phys, size, flags);
874 }
875
876 if (start_blk > last_blk || ret)
877 goto out;
878
879 logical = blk_to_logical(inode, start_blk);
880 phys = blk_to_logical(inode, map_bh.b_blocknr);
881 size = map_bh.b_size;
882 flags = 0;
883 if (buffer_unwritten(&map_bh))
884 flags = FIEMAP_EXTENT_UNWRITTEN;
885
886 start_blk += logical_to_blk(inode, size);
887
888 prep_next:
889 cond_resched();
890 if (fatal_signal_pending(current))
891 ret = -EINTR;
892 else
893 goto next;
894 out:
895 if (ret == 1)
896 ret = 0;
897
898 inode_unlock(inode);
899 return ret;
900 }
901
902 /*
903 * This function was originally taken from fs/mpage.c, and customized for f2fs.
904 * Major change was from block_size == page_size in f2fs by default.
905 */
906 static int f2fs_mpage_readpages(struct address_space *mapping,
907 struct list_head *pages, struct page *page,
908 unsigned nr_pages)
909 {
910 struct bio *bio = NULL;
911 unsigned page_idx;
912 sector_t last_block_in_bio = 0;
913 struct inode *inode = mapping->host;
914 const unsigned blkbits = inode->i_blkbits;
915 const unsigned blocksize = 1 << blkbits;
916 sector_t block_in_file;
917 sector_t last_block;
918 sector_t last_block_in_file;
919 sector_t block_nr;
920 struct block_device *bdev = inode->i_sb->s_bdev;
921 struct f2fs_map_blocks map;
922
923 map.m_pblk = 0;
924 map.m_lblk = 0;
925 map.m_len = 0;
926 map.m_flags = 0;
927 map.m_next_pgofs = NULL;
928
929 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
930
931 prefetchw(&page->flags);
932 if (pages) {
933 page = list_entry(pages->prev, struct page, lru);
934 list_del(&page->lru);
935 if (add_to_page_cache_lru(page, mapping,
936 page->index, GFP_KERNEL))
937 goto next_page;
938 }
939
940 block_in_file = (sector_t)page->index;
941 last_block = block_in_file + nr_pages;
942 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
943 blkbits;
944 if (last_block > last_block_in_file)
945 last_block = last_block_in_file;
946
947 /*
948 * Map blocks using the previous result first.
949 */
950 if ((map.m_flags & F2FS_MAP_MAPPED) &&
951 block_in_file > map.m_lblk &&
952 block_in_file < (map.m_lblk + map.m_len))
953 goto got_it;
954
955 /*
956 * Then do more f2fs_map_blocks() calls until we are
957 * done with this page.
958 */
959 map.m_flags = 0;
960
961 if (block_in_file < last_block) {
962 map.m_lblk = block_in_file;
963 map.m_len = last_block - block_in_file;
964
965 if (f2fs_map_blocks(inode, &map, 0,
966 F2FS_GET_BLOCK_READ))
967 goto set_error_page;
968 }
969 got_it:
970 if ((map.m_flags & F2FS_MAP_MAPPED)) {
971 block_nr = map.m_pblk + block_in_file - map.m_lblk;
972 SetPageMappedToDisk(page);
973
974 if (!PageUptodate(page) && !cleancache_get_page(page)) {
975 SetPageUptodate(page);
976 goto confused;
977 }
978 } else {
979 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
980 SetPageUptodate(page);
981 unlock_page(page);
982 goto next_page;
983 }
984
985 /*
986 * This page will go to BIO. Do we need to send this
987 * BIO off first?
988 */
989 if (bio && (last_block_in_bio != block_nr - 1)) {
990 submit_and_realloc:
991 submit_bio(READ, bio);
992 bio = NULL;
993 }
994 if (bio == NULL) {
995 struct f2fs_crypto_ctx *ctx = NULL;
996
997 if (f2fs_encrypted_inode(inode) &&
998 S_ISREG(inode->i_mode)) {
999
1000 ctx = f2fs_get_crypto_ctx(inode);
1001 if (IS_ERR(ctx))
1002 goto set_error_page;
1003
1004 /* wait the page to be moved by cleaning */
1005 f2fs_wait_on_encrypted_page_writeback(
1006 F2FS_I_SB(inode), block_nr);
1007 }
1008
1009 bio = bio_alloc(GFP_KERNEL,
1010 min_t(int, nr_pages, BIO_MAX_PAGES));
1011 if (!bio) {
1012 if (ctx)
1013 f2fs_release_crypto_ctx(ctx);
1014 goto set_error_page;
1015 }
1016 bio->bi_bdev = bdev;
1017 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(block_nr);
1018 bio->bi_end_io = f2fs_read_end_io;
1019 bio->bi_private = ctx;
1020 }
1021
1022 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1023 goto submit_and_realloc;
1024
1025 last_block_in_bio = block_nr;
1026 goto next_page;
1027 set_error_page:
1028 SetPageError(page);
1029 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1030 unlock_page(page);
1031 goto next_page;
1032 confused:
1033 if (bio) {
1034 submit_bio(READ, bio);
1035 bio = NULL;
1036 }
1037 unlock_page(page);
1038 next_page:
1039 if (pages)
1040 page_cache_release(page);
1041 }
1042 BUG_ON(pages && !list_empty(pages));
1043 if (bio)
1044 submit_bio(READ, bio);
1045 return 0;
1046 }
1047
1048 static int f2fs_read_data_page(struct file *file, struct page *page)
1049 {
1050 struct inode *inode = page->mapping->host;
1051 int ret = -EAGAIN;
1052
1053 trace_f2fs_readpage(page, DATA);
1054
1055 /* If the file has inline data, try to read it directly */
1056 if (f2fs_has_inline_data(inode))
1057 ret = f2fs_read_inline_data(inode, page);
1058 if (ret == -EAGAIN)
1059 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1060 return ret;
1061 }
1062
1063 static int f2fs_read_data_pages(struct file *file,
1064 struct address_space *mapping,
1065 struct list_head *pages, unsigned nr_pages)
1066 {
1067 struct inode *inode = file->f_mapping->host;
1068 struct page *page = list_entry(pages->prev, struct page, lru);
1069
1070 trace_f2fs_readpages(inode, page, nr_pages);
1071
1072 /* If the file has inline data, skip readpages */
1073 if (f2fs_has_inline_data(inode))
1074 return 0;
1075
1076 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1077 }
1078
1079 int do_write_data_page(struct f2fs_io_info *fio)
1080 {
1081 struct page *page = fio->page;
1082 struct inode *inode = page->mapping->host;
1083 struct dnode_of_data dn;
1084 int err = 0;
1085
1086 set_new_dnode(&dn, inode, NULL, NULL, 0);
1087 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1088 if (err)
1089 return err;
1090
1091 fio->old_blkaddr = dn.data_blkaddr;
1092
1093 /* This page is already truncated */
1094 if (fio->old_blkaddr == NULL_ADDR) {
1095 ClearPageUptodate(page);
1096 goto out_writepage;
1097 }
1098
1099 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1100
1101 /* wait for GCed encrypted page writeback */
1102 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode),
1103 fio->old_blkaddr);
1104
1105 fio->encrypted_page = f2fs_encrypt(inode, fio->page);
1106 if (IS_ERR(fio->encrypted_page)) {
1107 err = PTR_ERR(fio->encrypted_page);
1108 goto out_writepage;
1109 }
1110 }
1111
1112 set_page_writeback(page);
1113
1114 /*
1115 * If current allocation needs SSR,
1116 * it had better in-place writes for updated data.
1117 */
1118 if (unlikely(fio->old_blkaddr != NEW_ADDR &&
1119 !is_cold_data(page) &&
1120 !IS_ATOMIC_WRITTEN_PAGE(page) &&
1121 need_inplace_update(inode))) {
1122 rewrite_data_page(fio);
1123 set_inode_flag(F2FS_I(inode), FI_UPDATE_WRITE);
1124 trace_f2fs_do_write_data_page(page, IPU);
1125 } else {
1126 write_data_page(&dn, fio);
1127 trace_f2fs_do_write_data_page(page, OPU);
1128 set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
1129 if (page->index == 0)
1130 set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
1131 }
1132 out_writepage:
1133 f2fs_put_dnode(&dn);
1134 return err;
1135 }
1136
1137 static int f2fs_write_data_page(struct page *page,
1138 struct writeback_control *wbc)
1139 {
1140 struct inode *inode = page->mapping->host;
1141 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1142 loff_t i_size = i_size_read(inode);
1143 const pgoff_t end_index = ((unsigned long long) i_size)
1144 >> PAGE_CACHE_SHIFT;
1145 unsigned offset = 0;
1146 bool need_balance_fs = false;
1147 int err = 0;
1148 struct f2fs_io_info fio = {
1149 .sbi = sbi,
1150 .type = DATA,
1151 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1152 .page = page,
1153 .encrypted_page = NULL,
1154 };
1155
1156 trace_f2fs_writepage(page, DATA);
1157
1158 if (page->index < end_index)
1159 goto write;
1160
1161 /*
1162 * If the offset is out-of-range of file size,
1163 * this page does not have to be written to disk.
1164 */
1165 offset = i_size & (PAGE_CACHE_SIZE - 1);
1166 if ((page->index >= end_index + 1) || !offset)
1167 goto out;
1168
1169 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
1170 write:
1171 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1172 goto redirty_out;
1173 if (f2fs_is_drop_cache(inode))
1174 goto out;
1175 if (f2fs_is_volatile_file(inode) && !wbc->for_reclaim &&
1176 available_free_memory(sbi, BASE_CHECK))
1177 goto redirty_out;
1178
1179 /* Dentry blocks are controlled by checkpoint */
1180 if (S_ISDIR(inode->i_mode)) {
1181 if (unlikely(f2fs_cp_error(sbi)))
1182 goto redirty_out;
1183 err = do_write_data_page(&fio);
1184 goto done;
1185 }
1186
1187 /* we should bypass data pages to proceed the kworkder jobs */
1188 if (unlikely(f2fs_cp_error(sbi))) {
1189 SetPageError(page);
1190 goto out;
1191 }
1192
1193 if (!wbc->for_reclaim)
1194 need_balance_fs = true;
1195 else if (has_not_enough_free_secs(sbi, 0))
1196 goto redirty_out;
1197
1198 err = -EAGAIN;
1199 f2fs_lock_op(sbi);
1200 if (f2fs_has_inline_data(inode))
1201 err = f2fs_write_inline_data(inode, page);
1202 if (err == -EAGAIN)
1203 err = do_write_data_page(&fio);
1204 f2fs_unlock_op(sbi);
1205 done:
1206 if (err && err != -ENOENT)
1207 goto redirty_out;
1208
1209 clear_cold_data(page);
1210 out:
1211 inode_dec_dirty_pages(inode);
1212 if (err)
1213 ClearPageUptodate(page);
1214
1215 if (wbc->for_reclaim) {
1216 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, DATA, WRITE);
1217 remove_dirty_inode(inode);
1218 }
1219
1220 unlock_page(page);
1221 f2fs_balance_fs(sbi, need_balance_fs);
1222
1223 if (unlikely(f2fs_cp_error(sbi)))
1224 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1225
1226 return 0;
1227
1228 redirty_out:
1229 redirty_page_for_writepage(wbc, page);
1230 return AOP_WRITEPAGE_ACTIVATE;
1231 }
1232
1233 static int __f2fs_writepage(struct page *page, struct writeback_control *wbc,
1234 void *data)
1235 {
1236 struct address_space *mapping = data;
1237 int ret = mapping->a_ops->writepage(page, wbc);
1238 mapping_set_error(mapping, ret);
1239 return ret;
1240 }
1241
1242 /*
1243 * This function was copied from write_cche_pages from mm/page-writeback.c.
1244 * The major change is making write step of cold data page separately from
1245 * warm/hot data page.
1246 */
1247 static int f2fs_write_cache_pages(struct address_space *mapping,
1248 struct writeback_control *wbc, writepage_t writepage,
1249 void *data)
1250 {
1251 int ret = 0;
1252 int done = 0;
1253 struct pagevec pvec;
1254 int nr_pages;
1255 pgoff_t uninitialized_var(writeback_index);
1256 pgoff_t index;
1257 pgoff_t end; /* Inclusive */
1258 pgoff_t done_index;
1259 int cycled;
1260 int range_whole = 0;
1261 int tag;
1262 int step = 0;
1263
1264 pagevec_init(&pvec, 0);
1265 next:
1266 if (wbc->range_cyclic) {
1267 writeback_index = mapping->writeback_index; /* prev offset */
1268 index = writeback_index;
1269 if (index == 0)
1270 cycled = 1;
1271 else
1272 cycled = 0;
1273 end = -1;
1274 } else {
1275 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1276 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1277 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1278 range_whole = 1;
1279 cycled = 1; /* ignore range_cyclic tests */
1280 }
1281 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1282 tag = PAGECACHE_TAG_TOWRITE;
1283 else
1284 tag = PAGECACHE_TAG_DIRTY;
1285 retry:
1286 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1287 tag_pages_for_writeback(mapping, index, end);
1288 done_index = index;
1289 while (!done && (index <= end)) {
1290 int i;
1291
1292 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1293 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1294 if (nr_pages == 0)
1295 break;
1296
1297 for (i = 0; i < nr_pages; i++) {
1298 struct page *page = pvec.pages[i];
1299
1300 if (page->index > end) {
1301 done = 1;
1302 break;
1303 }
1304
1305 done_index = page->index;
1306
1307 lock_page(page);
1308
1309 if (unlikely(page->mapping != mapping)) {
1310 continue_unlock:
1311 unlock_page(page);
1312 continue;
1313 }
1314
1315 if (!PageDirty(page)) {
1316 /* someone wrote it for us */
1317 goto continue_unlock;
1318 }
1319
1320 if (step == is_cold_data(page))
1321 goto continue_unlock;
1322
1323 if (PageWriteback(page)) {
1324 if (wbc->sync_mode != WB_SYNC_NONE)
1325 f2fs_wait_on_page_writeback(page,
1326 DATA, true);
1327 else
1328 goto continue_unlock;
1329 }
1330
1331 BUG_ON(PageWriteback(page));
1332 if (!clear_page_dirty_for_io(page))
1333 goto continue_unlock;
1334
1335 ret = (*writepage)(page, wbc, data);
1336 if (unlikely(ret)) {
1337 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1338 unlock_page(page);
1339 ret = 0;
1340 } else {
1341 done_index = page->index + 1;
1342 done = 1;
1343 break;
1344 }
1345 }
1346
1347 if (--wbc->nr_to_write <= 0 &&
1348 wbc->sync_mode == WB_SYNC_NONE) {
1349 done = 1;
1350 break;
1351 }
1352 }
1353 pagevec_release(&pvec);
1354 cond_resched();
1355 }
1356
1357 if (step < 1) {
1358 step++;
1359 goto next;
1360 }
1361
1362 if (!cycled && !done) {
1363 cycled = 1;
1364 index = 0;
1365 end = writeback_index - 1;
1366 goto retry;
1367 }
1368 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1369 mapping->writeback_index = done_index;
1370
1371 return ret;
1372 }
1373
1374 static int f2fs_write_data_pages(struct address_space *mapping,
1375 struct writeback_control *wbc)
1376 {
1377 struct inode *inode = mapping->host;
1378 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1379 bool locked = false;
1380 int ret;
1381 long diff;
1382
1383 /* deal with chardevs and other special file */
1384 if (!mapping->a_ops->writepage)
1385 return 0;
1386
1387 /* skip writing if there is no dirty page in this inode */
1388 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1389 return 0;
1390
1391 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1392 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1393 available_free_memory(sbi, DIRTY_DENTS))
1394 goto skip_write;
1395
1396 /* skip writing during file defragment */
1397 if (is_inode_flag_set(F2FS_I(inode), FI_DO_DEFRAG))
1398 goto skip_write;
1399
1400 /* during POR, we don't need to trigger writepage at all. */
1401 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1402 goto skip_write;
1403
1404 trace_f2fs_writepages(mapping->host, wbc, DATA);
1405
1406 diff = nr_pages_to_write(sbi, DATA, wbc);
1407
1408 if (!S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_ALL) {
1409 mutex_lock(&sbi->writepages);
1410 locked = true;
1411 }
1412 ret = f2fs_write_cache_pages(mapping, wbc, __f2fs_writepage, mapping);
1413 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
1414 if (locked)
1415 mutex_unlock(&sbi->writepages);
1416
1417 remove_dirty_inode(inode);
1418
1419 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1420 return ret;
1421
1422 skip_write:
1423 wbc->pages_skipped += get_dirty_pages(inode);
1424 trace_f2fs_writepages(mapping->host, wbc, DATA);
1425 return 0;
1426 }
1427
1428 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1429 {
1430 struct inode *inode = mapping->host;
1431 loff_t i_size = i_size_read(inode);
1432
1433 if (to > i_size) {
1434 truncate_pagecache(inode, i_size);
1435 truncate_blocks(inode, i_size, true);
1436 }
1437 }
1438
1439 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1440 struct page *page, loff_t pos, unsigned len,
1441 block_t *blk_addr, bool *node_changed)
1442 {
1443 struct inode *inode = page->mapping->host;
1444 pgoff_t index = page->index;
1445 struct dnode_of_data dn;
1446 struct page *ipage;
1447 bool locked = false;
1448 struct extent_info ei;
1449 int err = 0;
1450
1451 /*
1452 * we already allocated all the blocks, so we don't need to get
1453 * the block addresses when there is no need to fill the page.
1454 */
1455 if (!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
1456 len == PAGE_CACHE_SIZE)
1457 return 0;
1458
1459 if (f2fs_has_inline_data(inode) ||
1460 (pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1461 f2fs_lock_op(sbi);
1462 locked = true;
1463 }
1464 restart:
1465 /* check inline_data */
1466 ipage = get_node_page(sbi, inode->i_ino);
1467 if (IS_ERR(ipage)) {
1468 err = PTR_ERR(ipage);
1469 goto unlock_out;
1470 }
1471
1472 set_new_dnode(&dn, inode, ipage, ipage, 0);
1473
1474 if (f2fs_has_inline_data(inode)) {
1475 if (pos + len <= MAX_INLINE_DATA) {
1476 read_inline_data(page, ipage);
1477 set_inode_flag(F2FS_I(inode), FI_DATA_EXIST);
1478 set_inline_node(ipage);
1479 } else {
1480 err = f2fs_convert_inline_page(&dn, page);
1481 if (err)
1482 goto out;
1483 if (dn.data_blkaddr == NULL_ADDR)
1484 err = f2fs_get_block(&dn, index);
1485 }
1486 } else if (locked) {
1487 err = f2fs_get_block(&dn, index);
1488 } else {
1489 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1490 dn.data_blkaddr = ei.blk + index - ei.fofs;
1491 } else {
1492 /* hole case */
1493 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1494 if (err || (!err && dn.data_blkaddr == NULL_ADDR)) {
1495 f2fs_put_dnode(&dn);
1496 f2fs_lock_op(sbi);
1497 locked = true;
1498 goto restart;
1499 }
1500 }
1501 }
1502
1503 /* convert_inline_page can make node_changed */
1504 *blk_addr = dn.data_blkaddr;
1505 *node_changed = dn.node_changed;
1506 out:
1507 f2fs_put_dnode(&dn);
1508 unlock_out:
1509 if (locked)
1510 f2fs_unlock_op(sbi);
1511 return err;
1512 }
1513
1514 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1515 loff_t pos, unsigned len, unsigned flags,
1516 struct page **pagep, void **fsdata)
1517 {
1518 struct inode *inode = mapping->host;
1519 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1520 struct page *page = NULL;
1521 pgoff_t index = ((unsigned long long) pos) >> PAGE_CACHE_SHIFT;
1522 bool need_balance = false;
1523 block_t blkaddr = NULL_ADDR;
1524 int err = 0;
1525
1526 trace_f2fs_write_begin(inode, pos, len, flags);
1527
1528 /*
1529 * We should check this at this moment to avoid deadlock on inode page
1530 * and #0 page. The locking rule for inline_data conversion should be:
1531 * lock_page(page #0) -> lock_page(inode_page)
1532 */
1533 if (index != 0) {
1534 err = f2fs_convert_inline_inode(inode);
1535 if (err)
1536 goto fail;
1537 }
1538 repeat:
1539 page = grab_cache_page_write_begin(mapping, index, flags);
1540 if (!page) {
1541 err = -ENOMEM;
1542 goto fail;
1543 }
1544
1545 *pagep = page;
1546
1547 err = prepare_write_begin(sbi, page, pos, len,
1548 &blkaddr, &need_balance);
1549 if (err)
1550 goto fail;
1551
1552 if (need_balance && has_not_enough_free_secs(sbi, 0)) {
1553 unlock_page(page);
1554 f2fs_balance_fs(sbi, true);
1555 lock_page(page);
1556 if (page->mapping != mapping) {
1557 /* The page got truncated from under us */
1558 f2fs_put_page(page, 1);
1559 goto repeat;
1560 }
1561 }
1562
1563 f2fs_wait_on_page_writeback(page, DATA, false);
1564
1565 /* wait for GCed encrypted page writeback */
1566 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1567 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1568
1569 if (len == PAGE_CACHE_SIZE)
1570 goto out_update;
1571 if (PageUptodate(page))
1572 goto out_clear;
1573
1574 if ((pos & PAGE_CACHE_MASK) >= i_size_read(inode)) {
1575 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
1576 unsigned end = start + len;
1577
1578 /* Reading beyond i_size is simple: memset to zero */
1579 zero_user_segments(page, 0, start, end, PAGE_CACHE_SIZE);
1580 goto out_update;
1581 }
1582
1583 if (blkaddr == NEW_ADDR) {
1584 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1585 } else {
1586 struct f2fs_io_info fio = {
1587 .sbi = sbi,
1588 .type = DATA,
1589 .rw = READ_SYNC,
1590 .old_blkaddr = blkaddr,
1591 .new_blkaddr = blkaddr,
1592 .page = page,
1593 .encrypted_page = NULL,
1594 };
1595 err = f2fs_submit_page_bio(&fio);
1596 if (err)
1597 goto fail;
1598
1599 lock_page(page);
1600 if (unlikely(!PageUptodate(page))) {
1601 err = -EIO;
1602 goto fail;
1603 }
1604 if (unlikely(page->mapping != mapping)) {
1605 f2fs_put_page(page, 1);
1606 goto repeat;
1607 }
1608
1609 /* avoid symlink page */
1610 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
1611 err = f2fs_decrypt(page);
1612 if (err)
1613 goto fail;
1614 }
1615 }
1616 out_update:
1617 SetPageUptodate(page);
1618 out_clear:
1619 clear_cold_data(page);
1620 return 0;
1621
1622 fail:
1623 f2fs_put_page(page, 1);
1624 f2fs_write_failed(mapping, pos + len);
1625 return err;
1626 }
1627
1628 static int f2fs_write_end(struct file *file,
1629 struct address_space *mapping,
1630 loff_t pos, unsigned len, unsigned copied,
1631 struct page *page, void *fsdata)
1632 {
1633 struct inode *inode = page->mapping->host;
1634
1635 trace_f2fs_write_end(inode, pos, len, copied);
1636
1637 set_page_dirty(page);
1638
1639 if (pos + copied > i_size_read(inode)) {
1640 i_size_write(inode, pos + copied);
1641 mark_inode_dirty(inode);
1642 }
1643
1644 f2fs_put_page(page, 1);
1645 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
1646 return copied;
1647 }
1648
1649 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
1650 loff_t offset)
1651 {
1652 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
1653
1654 if (offset & blocksize_mask)
1655 return -EINVAL;
1656
1657 if (iov_iter_alignment(iter) & blocksize_mask)
1658 return -EINVAL;
1659
1660 return 0;
1661 }
1662
1663 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1664 loff_t offset)
1665 {
1666 struct address_space *mapping = iocb->ki_filp->f_mapping;
1667 struct inode *inode = mapping->host;
1668 size_t count = iov_iter_count(iter);
1669 int err;
1670
1671 err = check_direct_IO(inode, iter, offset);
1672 if (err)
1673 return err;
1674
1675 if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
1676 return 0;
1677
1678 trace_f2fs_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
1679
1680 err = blockdev_direct_IO(iocb, inode, iter, offset, get_data_block_dio);
1681 if (err < 0 && iov_iter_rw(iter) == WRITE)
1682 f2fs_write_failed(mapping, offset + count);
1683
1684 trace_f2fs_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), err);
1685
1686 return err;
1687 }
1688
1689 void f2fs_invalidate_page(struct page *page, unsigned int offset,
1690 unsigned int length)
1691 {
1692 struct inode *inode = page->mapping->host;
1693 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1694
1695 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
1696 (offset % PAGE_CACHE_SIZE || length != PAGE_CACHE_SIZE))
1697 return;
1698
1699 if (PageDirty(page)) {
1700 if (inode->i_ino == F2FS_META_INO(sbi))
1701 dec_page_count(sbi, F2FS_DIRTY_META);
1702 else if (inode->i_ino == F2FS_NODE_INO(sbi))
1703 dec_page_count(sbi, F2FS_DIRTY_NODES);
1704 else
1705 inode_dec_dirty_pages(inode);
1706 }
1707
1708 /* This is atomic written page, keep Private */
1709 if (IS_ATOMIC_WRITTEN_PAGE(page))
1710 return;
1711
1712 ClearPagePrivate(page);
1713 }
1714
1715 int f2fs_release_page(struct page *page, gfp_t wait)
1716 {
1717 /* If this is dirty page, keep PagePrivate */
1718 if (PageDirty(page))
1719 return 0;
1720
1721 /* This is atomic written page, keep Private */
1722 if (IS_ATOMIC_WRITTEN_PAGE(page))
1723 return 0;
1724
1725 ClearPagePrivate(page);
1726 return 1;
1727 }
1728
1729 static int f2fs_set_data_page_dirty(struct page *page)
1730 {
1731 struct address_space *mapping = page->mapping;
1732 struct inode *inode = mapping->host;
1733
1734 trace_f2fs_set_page_dirty(page, DATA);
1735
1736 SetPageUptodate(page);
1737
1738 if (f2fs_is_atomic_file(inode)) {
1739 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
1740 register_inmem_page(inode, page);
1741 return 1;
1742 }
1743 /*
1744 * Previously, this page has been registered, we just
1745 * return here.
1746 */
1747 return 0;
1748 }
1749
1750 if (!PageDirty(page)) {
1751 __set_page_dirty_nobuffers(page);
1752 update_dirty_page(inode, page);
1753 return 1;
1754 }
1755 return 0;
1756 }
1757
1758 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
1759 {
1760 struct inode *inode = mapping->host;
1761
1762 if (f2fs_has_inline_data(inode))
1763 return 0;
1764
1765 /* make sure allocating whole blocks */
1766 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1767 filemap_write_and_wait(mapping);
1768
1769 return generic_block_bmap(mapping, block, get_data_block_bmap);
1770 }
1771
1772 const struct address_space_operations f2fs_dblock_aops = {
1773 .readpage = f2fs_read_data_page,
1774 .readpages = f2fs_read_data_pages,
1775 .writepage = f2fs_write_data_page,
1776 .writepages = f2fs_write_data_pages,
1777 .write_begin = f2fs_write_begin,
1778 .write_end = f2fs_write_end,
1779 .set_page_dirty = f2fs_set_data_page_dirty,
1780 .invalidatepage = f2fs_invalidate_page,
1781 .releasepage = f2fs_release_page,
1782 .direct_IO = f2fs_direct_IO,
1783 .bmap = f2fs_bmap,
1784 };
Linux kernel - Deliverables
This page took 0.100618 seconds and 5 git commands to generate.