projects / GitHub / LineageOS / android_kernel_motorola_exynos9610.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge 4.14.4 into android-4.14
[GitHub/LineageOS/android_kernel_motorola_exynos9610.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 #include <linux/sched/signal.h>
26
27 #include "f2fs.h"
28 #include "node.h"
29 #include "segment.h"
30 #include "trace.h"
31 #include <trace/events/f2fs.h>
32 #include <trace/events/android_fs.h>
33
34 static bool __is_cp_guaranteed(struct page *page)
35 {
36 struct address_space *mapping = page->mapping;
37 struct inode *inode;
38 struct f2fs_sb_info *sbi;
39
40 if (!mapping)
41 return false;
42
43 inode = mapping->host;
44 sbi = F2FS_I_SB(inode);
45
46 if (inode->i_ino == F2FS_META_INO(sbi) ||
47 inode->i_ino == F2FS_NODE_INO(sbi) ||
48 S_ISDIR(inode->i_mode) ||
49 is_cold_data(page))
50 return true;
51 return false;
52 }
53
54 static void f2fs_read_end_io(struct bio *bio)
55 {
56 struct bio_vec *bvec;
57 int i;
58
59 #ifdef CONFIG_F2FS_FAULT_INJECTION
60 if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO)) {
61 f2fs_show_injection_info(FAULT_IO);
62 bio->bi_status = BLK_STS_IOERR;
63 }
64 #endif
65
66 if (f2fs_bio_encrypted(bio)) {
67 if (bio->bi_status) {
68 fscrypt_release_ctx(bio->bi_private);
69 } else {
70 fscrypt_decrypt_bio_pages(bio->bi_private, bio);
71 return;
72 }
73 }
74
75 bio_for_each_segment_all(bvec, bio, i) {
76 struct page *page = bvec->bv_page;
77
78 if (!bio->bi_status) {
79 if (!PageUptodate(page))
80 SetPageUptodate(page);
81 } else {
82 ClearPageUptodate(page);
83 SetPageError(page);
84 }
85 unlock_page(page);
86 }
87 bio_put(bio);
88 }
89
90 static void f2fs_write_end_io(struct bio *bio)
91 {
92 struct f2fs_sb_info *sbi = bio->bi_private;
93 struct bio_vec *bvec;
94 int i;
95
96 bio_for_each_segment_all(bvec, bio, i) {
97 struct page *page = bvec->bv_page;
98 enum count_type type = WB_DATA_TYPE(page);
99
100 if (IS_DUMMY_WRITTEN_PAGE(page)) {
101 set_page_private(page, (unsigned long)NULL);
102 ClearPagePrivate(page);
103 unlock_page(page);
104 mempool_free(page, sbi->write_io_dummy);
105
106 if (unlikely(bio->bi_status))
107 f2fs_stop_checkpoint(sbi, true);
108 continue;
109 }
110
111 fscrypt_pullback_bio_page(&page, true);
112
113 if (unlikely(bio->bi_status)) {
114 mapping_set_error(page->mapping, -EIO);
115 f2fs_stop_checkpoint(sbi, true);
116 }
117 dec_page_count(sbi, type);
118 clear_cold_data(page);
119 end_page_writeback(page);
120 }
121 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
122 wq_has_sleeper(&sbi->cp_wait))
123 wake_up(&sbi->cp_wait);
124
125 bio_put(bio);
126 }
127
128 /*
129 * Return true, if pre_bio's bdev is same as its target device.
130 */
131 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
132 block_t blk_addr, struct bio *bio)
133 {
134 struct block_device *bdev = sbi->sb->s_bdev;
135 int i;
136
137 for (i = 0; i < sbi->s_ndevs; i++) {
138 if (FDEV(i).start_blk <= blk_addr &&
139 FDEV(i).end_blk >= blk_addr) {
140 blk_addr -= FDEV(i).start_blk;
141 bdev = FDEV(i).bdev;
142 break;
143 }
144 }
145 if (bio) {
146 bio_set_dev(bio, bdev);
147 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
148 }
149 return bdev;
150 }
151
152 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
153 {
154 int i;
155
156 for (i = 0; i < sbi->s_ndevs; i++)
157 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
158 return i;
159 return 0;
160 }
161
162 static bool __same_bdev(struct f2fs_sb_info *sbi,
163 block_t blk_addr, struct bio *bio)
164 {
165 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
166 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
167 }
168
169 /*
170 * Low-level block read/write IO operations.
171 */
172 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
173 int npages, bool is_read)
174 {
175 struct bio *bio;
176
177 bio = f2fs_bio_alloc(sbi, npages, true);
178
179 f2fs_target_device(sbi, blk_addr, bio);
180 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
181 bio->bi_private = is_read ? NULL : sbi;
182
183 return bio;
184 }
185
186 static inline void __submit_bio(struct f2fs_sb_info *sbi,
187 struct bio *bio, enum page_type type)
188 {
189 if (!is_read_io(bio_op(bio))) {
190 unsigned int start;
191
192 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
193 current->plug && (type == DATA || type == NODE))
194 blk_finish_plug(current->plug);
195
196 if (type != DATA && type != NODE)
197 goto submit_io;
198
199 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
200 start %= F2FS_IO_SIZE(sbi);
201
202 if (start == 0)
203 goto submit_io;
204
205 /* fill dummy pages */
206 for (; start < F2FS_IO_SIZE(sbi); start++) {
207 struct page *page =
208 mempool_alloc(sbi->write_io_dummy,
209 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
210 f2fs_bug_on(sbi, !page);
211
212 SetPagePrivate(page);
213 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
214 lock_page(page);
215 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
216 f2fs_bug_on(sbi, 1);
217 }
218 /*
219 * In the NODE case, we lose next block address chain. So, we
220 * need to do checkpoint in f2fs_sync_file.
221 */
222 if (type == NODE)
223 set_sbi_flag(sbi, SBI_NEED_CP);
224 }
225 submit_io:
226 if (is_read_io(bio_op(bio)))
227 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
228 else
229 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
230 submit_bio(bio);
231 }
232
233 static void __submit_merged_bio(struct f2fs_bio_info *io)
234 {
235 struct f2fs_io_info *fio = &io->fio;
236
237 if (!io->bio)
238 return;
239
240 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
241
242 if (is_read_io(fio->op))
243 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
244 else
245 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
246
247 __submit_bio(io->sbi, io->bio, fio->type);
248 io->bio = NULL;
249 }
250
251 static bool __has_merged_page(struct f2fs_bio_info *io,
252 struct inode *inode, nid_t ino, pgoff_t idx)
253 {
254 struct bio_vec *bvec;
255 struct page *target;
256 int i;
257
258 if (!io->bio)
259 return false;
260
261 if (!inode && !ino)
262 return true;
263
264 bio_for_each_segment_all(bvec, io->bio, i) {
265
266 if (bvec->bv_page->mapping)
267 target = bvec->bv_page;
268 else
269 target = fscrypt_control_page(bvec->bv_page);
270
271 if (idx != target->index)
272 continue;
273
274 if (inode && inode == target->mapping->host)
275 return true;
276 if (ino && ino == ino_of_node(target))
277 return true;
278 }
279
280 return false;
281 }
282
283 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
284 nid_t ino, pgoff_t idx, enum page_type type)
285 {
286 enum page_type btype = PAGE_TYPE_OF_BIO(type);
287 enum temp_type temp;
288 struct f2fs_bio_info *io;
289 bool ret = false;
290
291 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
292 io = sbi->write_io[btype] + temp;
293
294 down_read(&io->io_rwsem);
295 ret = __has_merged_page(io, inode, ino, idx);
296 up_read(&io->io_rwsem);
297
298 /* TODO: use HOT temp only for meta pages now. */
299 if (ret || btype == META)
300 break;
301 }
302 return ret;
303 }
304
305 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
306 enum page_type type, enum temp_type temp)
307 {
308 enum page_type btype = PAGE_TYPE_OF_BIO(type);
309 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
310
311 down_write(&io->io_rwsem);
312
313 /* change META to META_FLUSH in the checkpoint procedure */
314 if (type >= META_FLUSH) {
315 io->fio.type = META_FLUSH;
316 io->fio.op = REQ_OP_WRITE;
317 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
318 if (!test_opt(sbi, NOBARRIER))
319 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
320 }
321 __submit_merged_bio(io);
322 up_write(&io->io_rwsem);
323 }
324
325 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
326 struct inode *inode, nid_t ino, pgoff_t idx,
327 enum page_type type, bool force)
328 {
329 enum temp_type temp;
330
331 if (!force && !has_merged_page(sbi, inode, ino, idx, type))
332 return;
333
334 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
335
336 __f2fs_submit_merged_write(sbi, type, temp);
337
338 /* TODO: use HOT temp only for meta pages now. */
339 if (type >= META)
340 break;
341 }
342 }
343
344 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
345 {
346 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
347 }
348
349 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
350 struct inode *inode, nid_t ino, pgoff_t idx,
351 enum page_type type)
352 {
353 __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
354 }
355
356 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
357 {
358 f2fs_submit_merged_write(sbi, DATA);
359 f2fs_submit_merged_write(sbi, NODE);
360 f2fs_submit_merged_write(sbi, META);
361 }
362
363 /*
364 * Fill the locked page with data located in the block address.
365 * A caller needs to unlock the page on failure.
366 */
367 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
368 {
369 struct bio *bio;
370 struct page *page = fio->encrypted_page ?
371 fio->encrypted_page : fio->page;
372
373 trace_f2fs_submit_page_bio(page, fio);
374 f2fs_trace_ios(fio, 0);
375
376 /* Allocate a new bio */
377 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
378
379 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
380 bio_put(bio);
381 return -EFAULT;
382 }
383 bio_set_op_attrs(bio, fio->op, fio->op_flags);
384
385 __submit_bio(fio->sbi, bio, fio->type);
386
387 if (!is_read_io(fio->op))
388 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
389 return 0;
390 }
391
392 int f2fs_submit_page_write(struct f2fs_io_info *fio)
393 {
394 struct f2fs_sb_info *sbi = fio->sbi;
395 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
396 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
397 struct page *bio_page;
398 int err = 0;
399
400 f2fs_bug_on(sbi, is_read_io(fio->op));
401
402 down_write(&io->io_rwsem);
403 next:
404 if (fio->in_list) {
405 spin_lock(&io->io_lock);
406 if (list_empty(&io->io_list)) {
407 spin_unlock(&io->io_lock);
408 goto out_fail;
409 }
410 fio = list_first_entry(&io->io_list,
411 struct f2fs_io_info, list);
412 list_del(&fio->list);
413 spin_unlock(&io->io_lock);
414 }
415
416 if (fio->old_blkaddr != NEW_ADDR)
417 verify_block_addr(sbi, fio->old_blkaddr);
418 verify_block_addr(sbi, fio->new_blkaddr);
419
420 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
421
422 /* set submitted = true as a return value */
423 fio->submitted = true;
424
425 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
426
427 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
428 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
429 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
430 __submit_merged_bio(io);
431 alloc_new:
432 if (io->bio == NULL) {
433 if ((fio->type == DATA || fio->type == NODE) &&
434 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
435 err = -EAGAIN;
436 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
437 goto out_fail;
438 }
439 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
440 BIO_MAX_PAGES, false);
441 io->fio = *fio;
442 }
443
444 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
445 __submit_merged_bio(io);
446 goto alloc_new;
447 }
448
449 io->last_block_in_bio = fio->new_blkaddr;
450 f2fs_trace_ios(fio, 0);
451
452 trace_f2fs_submit_page_write(fio->page, fio);
453
454 if (fio->in_list)
455 goto next;
456 out_fail:
457 up_write(&io->io_rwsem);
458 return err;
459 }
460
461 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
462 unsigned nr_pages)
463 {
464 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
465 struct fscrypt_ctx *ctx = NULL;
466 struct bio *bio;
467
468 if (f2fs_encrypted_file(inode)) {
469 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
470 if (IS_ERR(ctx))
471 return ERR_CAST(ctx);
472
473 /* wait the page to be moved by cleaning */
474 f2fs_wait_on_block_writeback(sbi, blkaddr);
475 }
476
477 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
478 if (!bio) {
479 if (ctx)
480 fscrypt_release_ctx(ctx);
481 return ERR_PTR(-ENOMEM);
482 }
483 f2fs_target_device(sbi, blkaddr, bio);
484 bio->bi_end_io = f2fs_read_end_io;
485 bio->bi_private = ctx;
486 bio_set_op_attrs(bio, REQ_OP_READ, 0);
487
488 return bio;
489 }
490
491 /* This can handle encryption stuffs */
492 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
493 block_t blkaddr)
494 {
495 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1);
496
497 if (IS_ERR(bio))
498 return PTR_ERR(bio);
499
500 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
501 bio_put(bio);
502 return -EFAULT;
503 }
504 __submit_bio(F2FS_I_SB(inode), bio, DATA);
505 return 0;
506 }
507
508 static void __set_data_blkaddr(struct dnode_of_data *dn)
509 {
510 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
511 __le32 *addr_array;
512 int base = 0;
513
514 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
515 base = get_extra_isize(dn->inode);
516
517 /* Get physical address of data block */
518 addr_array = blkaddr_in_node(rn);
519 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
520 }
521
522 /*
523 * Lock ordering for the change of data block address:
524 * ->data_page
525 * ->node_page
526 * update block addresses in the node page
527 */
528 void set_data_blkaddr(struct dnode_of_data *dn)
529 {
530 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
531 __set_data_blkaddr(dn);
532 if (set_page_dirty(dn->node_page))
533 dn->node_changed = true;
534 }
535
536 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
537 {
538 dn->data_blkaddr = blkaddr;
539 set_data_blkaddr(dn);
540 f2fs_update_extent_cache(dn);
541 }
542
543 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
544 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
545 {
546 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
547 int err;
548
549 if (!count)
550 return 0;
551
552 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
553 return -EPERM;
554 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
555 return err;
556
557 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
558 dn->ofs_in_node, count);
559
560 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
561
562 for (; count > 0; dn->ofs_in_node++) {
563 block_t blkaddr = datablock_addr(dn->inode,
564 dn->node_page, dn->ofs_in_node);
565 if (blkaddr == NULL_ADDR) {
566 dn->data_blkaddr = NEW_ADDR;
567 __set_data_blkaddr(dn);
568 count--;
569 }
570 }
571
572 if (set_page_dirty(dn->node_page))
573 dn->node_changed = true;
574 return 0;
575 }
576
577 /* Should keep dn->ofs_in_node unchanged */
578 int reserve_new_block(struct dnode_of_data *dn)
579 {
580 unsigned int ofs_in_node = dn->ofs_in_node;
581 int ret;
582
583 ret = reserve_new_blocks(dn, 1);
584 dn->ofs_in_node = ofs_in_node;
585 return ret;
586 }
587
588 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
589 {
590 bool need_put = dn->inode_page ? false : true;
591 int err;
592
593 err = get_dnode_of_data(dn, index, ALLOC_NODE);
594 if (err)
595 return err;
596
597 if (dn->data_blkaddr == NULL_ADDR)
598 err = reserve_new_block(dn);
599 if (err || need_put)
600 f2fs_put_dnode(dn);
601 return err;
602 }
603
604 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
605 {
606 struct extent_info ei = {0,0,0};
607 struct inode *inode = dn->inode;
608
609 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
610 dn->data_blkaddr = ei.blk + index - ei.fofs;
611 return 0;
612 }
613
614 return f2fs_reserve_block(dn, index);
615 }
616
617 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
618 int op_flags, bool for_write)
619 {
620 struct address_space *mapping = inode->i_mapping;
621 struct dnode_of_data dn;
622 struct page *page;
623 struct extent_info ei = {0,0,0};
624 int err;
625
626 page = f2fs_grab_cache_page(mapping, index, for_write);
627 if (!page)
628 return ERR_PTR(-ENOMEM);
629
630 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
631 dn.data_blkaddr = ei.blk + index - ei.fofs;
632 goto got_it;
633 }
634
635 set_new_dnode(&dn, inode, NULL, NULL, 0);
636 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
637 if (err)
638 goto put_err;
639 f2fs_put_dnode(&dn);
640
641 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
642 err = -ENOENT;
643 goto put_err;
644 }
645 got_it:
646 if (PageUptodate(page)) {
647 unlock_page(page);
648 return page;
649 }
650
651 /*
652 * A new dentry page is allocated but not able to be written, since its
653 * new inode page couldn't be allocated due to -ENOSPC.
654 * In such the case, its blkaddr can be remained as NEW_ADDR.
655 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
656 */
657 if (dn.data_blkaddr == NEW_ADDR) {
658 zero_user_segment(page, 0, PAGE_SIZE);
659 if (!PageUptodate(page))
660 SetPageUptodate(page);
661 unlock_page(page);
662 return page;
663 }
664
665 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
666 if (err)
667 goto put_err;
668 return page;
669
670 put_err:
671 f2fs_put_page(page, 1);
672 return ERR_PTR(err);
673 }
674
675 struct page *find_data_page(struct inode *inode, pgoff_t index)
676 {
677 struct address_space *mapping = inode->i_mapping;
678 struct page *page;
679
680 page = find_get_page(mapping, index);
681 if (page && PageUptodate(page))
682 return page;
683 f2fs_put_page(page, 0);
684
685 page = get_read_data_page(inode, index, 0, false);
686 if (IS_ERR(page))
687 return page;
688
689 if (PageUptodate(page))
690 return page;
691
692 wait_on_page_locked(page);
693 if (unlikely(!PageUptodate(page))) {
694 f2fs_put_page(page, 0);
695 return ERR_PTR(-EIO);
696 }
697 return page;
698 }
699
700 /*
701 * If it tries to access a hole, return an error.
702 * Because, the callers, functions in dir.c and GC, should be able to know
703 * whether this page exists or not.
704 */
705 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
706 bool for_write)
707 {
708 struct address_space *mapping = inode->i_mapping;
709 struct page *page;
710 repeat:
711 page = get_read_data_page(inode, index, 0, for_write);
712 if (IS_ERR(page))
713 return page;
714
715 /* wait for read completion */
716 lock_page(page);
717 if (unlikely(page->mapping != mapping)) {
718 f2fs_put_page(page, 1);
719 goto repeat;
720 }
721 if (unlikely(!PageUptodate(page))) {
722 f2fs_put_page(page, 1);
723 return ERR_PTR(-EIO);
724 }
725 return page;
726 }
727
728 /*
729 * Caller ensures that this data page is never allocated.
730 * A new zero-filled data page is allocated in the page cache.
731 *
732 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
733 * f2fs_unlock_op().
734 * Note that, ipage is set only by make_empty_dir, and if any error occur,
735 * ipage should be released by this function.
736 */
737 struct page *get_new_data_page(struct inode *inode,
738 struct page *ipage, pgoff_t index, bool new_i_size)
739 {
740 struct address_space *mapping = inode->i_mapping;
741 struct page *page;
742 struct dnode_of_data dn;
743 int err;
744
745 page = f2fs_grab_cache_page(mapping, index, true);
746 if (!page) {
747 /*
748 * before exiting, we should make sure ipage will be released
749 * if any error occur.
750 */
751 f2fs_put_page(ipage, 1);
752 return ERR_PTR(-ENOMEM);
753 }
754
755 set_new_dnode(&dn, inode, ipage, NULL, 0);
756 err = f2fs_reserve_block(&dn, index);
757 if (err) {
758 f2fs_put_page(page, 1);
759 return ERR_PTR(err);
760 }
761 if (!ipage)
762 f2fs_put_dnode(&dn);
763
764 if (PageUptodate(page))
765 goto got_it;
766
767 if (dn.data_blkaddr == NEW_ADDR) {
768 zero_user_segment(page, 0, PAGE_SIZE);
769 if (!PageUptodate(page))
770 SetPageUptodate(page);
771 } else {
772 f2fs_put_page(page, 1);
773
774 /* if ipage exists, blkaddr should be NEW_ADDR */
775 f2fs_bug_on(F2FS_I_SB(inode), ipage);
776 page = get_lock_data_page(inode, index, true);
777 if (IS_ERR(page))
778 return page;
779 }
780 got_it:
781 if (new_i_size && i_size_read(inode) <
782 ((loff_t)(index + 1) << PAGE_SHIFT))
783 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
784 return page;
785 }
786
787 static int __allocate_data_block(struct dnode_of_data *dn)
788 {
789 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
790 struct f2fs_summary sum;
791 struct node_info ni;
792 pgoff_t fofs;
793 blkcnt_t count = 1;
794 int err;
795
796 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
797 return -EPERM;
798
799 dn->data_blkaddr = datablock_addr(dn->inode,
800 dn->node_page, dn->ofs_in_node);
801 if (dn->data_blkaddr == NEW_ADDR)
802 goto alloc;
803
804 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
805 return err;
806
807 alloc:
808 get_node_info(sbi, dn->nid, &ni);
809 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
810
811 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
812 &sum, CURSEG_WARM_DATA, NULL, false);
813 set_data_blkaddr(dn);
814
815 /* update i_size */
816 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
817 dn->ofs_in_node;
818 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
819 f2fs_i_size_write(dn->inode,
820 ((loff_t)(fofs + 1) << PAGE_SHIFT));
821 return 0;
822 }
823
824 static inline bool __force_buffered_io(struct inode *inode, int rw)
825 {
826 return (f2fs_encrypted_file(inode) ||
827 (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
828 F2FS_I_SB(inode)->s_ndevs);
829 }
830
831 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
832 {
833 struct inode *inode = file_inode(iocb->ki_filp);
834 struct f2fs_map_blocks map;
835 int err = 0;
836
837 /* convert inline data for Direct I/O*/
838 if (iocb->ki_flags & IOCB_DIRECT) {
839 err = f2fs_convert_inline_inode(inode);
840 if (err)
841 return err;
842 }
843
844 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
845 return 0;
846
847 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
848 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
849 if (map.m_len > map.m_lblk)
850 map.m_len -= map.m_lblk;
851 else
852 map.m_len = 0;
853
854 map.m_next_pgofs = NULL;
855
856 if (iocb->ki_flags & IOCB_DIRECT)
857 return f2fs_map_blocks(inode, &map, 1,
858 __force_buffered_io(inode, WRITE) ?
859 F2FS_GET_BLOCK_PRE_AIO :
860 F2FS_GET_BLOCK_PRE_DIO);
861 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
862 err = f2fs_convert_inline_inode(inode);
863 if (err)
864 return err;
865 }
866 if (!f2fs_has_inline_data(inode))
867 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
868 return err;
869 }
870
871 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
872 {
873 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
874 if (lock)
875 down_read(&sbi->node_change);
876 else
877 up_read(&sbi->node_change);
878 } else {
879 if (lock)
880 f2fs_lock_op(sbi);
881 else
882 f2fs_unlock_op(sbi);
883 }
884 }
885
886 /*
887 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
888 * f2fs_map_blocks structure.
889 * If original data blocks are allocated, then give them to blockdev.
890 * Otherwise,
891 * a. preallocate requested block addresses
892 * b. do not use extent cache for better performance
893 * c. give the block addresses to blockdev
894 */
895 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
896 int create, int flag)
897 {
898 unsigned int maxblocks = map->m_len;
899 struct dnode_of_data dn;
900 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
901 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
902 pgoff_t pgofs, end_offset, end;
903 int err = 0, ofs = 1;
904 unsigned int ofs_in_node, last_ofs_in_node;
905 blkcnt_t prealloc;
906 struct extent_info ei = {0,0,0};
907 block_t blkaddr;
908
909 if (!maxblocks)
910 return 0;
911
912 map->m_len = 0;
913 map->m_flags = 0;
914
915 /* it only supports block size == page size */
916 pgofs = (pgoff_t)map->m_lblk;
917 end = pgofs + maxblocks;
918
919 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
920 map->m_pblk = ei.blk + pgofs - ei.fofs;
921 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
922 map->m_flags = F2FS_MAP_MAPPED;
923 goto out;
924 }
925
926 next_dnode:
927 if (create)
928 __do_map_lock(sbi, flag, true);
929
930 /* When reading holes, we need its node page */
931 set_new_dnode(&dn, inode, NULL, NULL, 0);
932 err = get_dnode_of_data(&dn, pgofs, mode);
933 if (err) {
934 if (flag == F2FS_GET_BLOCK_BMAP)
935 map->m_pblk = 0;
936 if (err == -ENOENT) {
937 err = 0;
938 if (map->m_next_pgofs)
939 *map->m_next_pgofs =
940 get_next_page_offset(&dn, pgofs);
941 }
942 goto unlock_out;
943 }
944
945 prealloc = 0;
946 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
947 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
948
949 next_block:
950 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
951
952 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
953 if (create) {
954 if (unlikely(f2fs_cp_error(sbi))) {
955 err = -EIO;
956 goto sync_out;
957 }
958 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
959 if (blkaddr == NULL_ADDR) {
960 prealloc++;
961 last_ofs_in_node = dn.ofs_in_node;
962 }
963 } else {
964 err = __allocate_data_block(&dn);
965 if (!err)
966 set_inode_flag(inode, FI_APPEND_WRITE);
967 }
968 if (err)
969 goto sync_out;
970 map->m_flags |= F2FS_MAP_NEW;
971 blkaddr = dn.data_blkaddr;
972 } else {
973 if (flag == F2FS_GET_BLOCK_BMAP) {
974 map->m_pblk = 0;
975 goto sync_out;
976 }
977 if (flag == F2FS_GET_BLOCK_FIEMAP &&
978 blkaddr == NULL_ADDR) {
979 if (map->m_next_pgofs)
980 *map->m_next_pgofs = pgofs + 1;
981 }
982 if (flag != F2FS_GET_BLOCK_FIEMAP ||
983 blkaddr != NEW_ADDR)
984 goto sync_out;
985 }
986 }
987
988 if (flag == F2FS_GET_BLOCK_PRE_AIO)
989 goto skip;
990
991 if (map->m_len == 0) {
992 /* preallocated unwritten block should be mapped for fiemap. */
993 if (blkaddr == NEW_ADDR)
994 map->m_flags |= F2FS_MAP_UNWRITTEN;
995 map->m_flags |= F2FS_MAP_MAPPED;
996
997 map->m_pblk = blkaddr;
998 map->m_len = 1;
999 } else if ((map->m_pblk != NEW_ADDR &&
1000 blkaddr == (map->m_pblk + ofs)) ||
1001 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1002 flag == F2FS_GET_BLOCK_PRE_DIO) {
1003 ofs++;
1004 map->m_len++;
1005 } else {
1006 goto sync_out;
1007 }
1008
1009 skip:
1010 dn.ofs_in_node++;
1011 pgofs++;
1012
1013 /* preallocate blocks in batch for one dnode page */
1014 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1015 (pgofs == end || dn.ofs_in_node == end_offset)) {
1016
1017 dn.ofs_in_node = ofs_in_node;
1018 err = reserve_new_blocks(&dn, prealloc);
1019 if (err)
1020 goto sync_out;
1021
1022 map->m_len += dn.ofs_in_node - ofs_in_node;
1023 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1024 err = -ENOSPC;
1025 goto sync_out;
1026 }
1027 dn.ofs_in_node = end_offset;
1028 }
1029
1030 if (pgofs >= end)
1031 goto sync_out;
1032 else if (dn.ofs_in_node < end_offset)
1033 goto next_block;
1034
1035 f2fs_put_dnode(&dn);
1036
1037 if (create) {
1038 __do_map_lock(sbi, flag, false);
1039 f2fs_balance_fs(sbi, dn.node_changed);
1040 }
1041 goto next_dnode;
1042
1043 sync_out:
1044 f2fs_put_dnode(&dn);
1045 unlock_out:
1046 if (create) {
1047 __do_map_lock(sbi, flag, false);
1048 f2fs_balance_fs(sbi, dn.node_changed);
1049 }
1050 out:
1051 trace_f2fs_map_blocks(inode, map, err);
1052 return err;
1053 }
1054
1055 static int __get_data_block(struct inode *inode, sector_t iblock,
1056 struct buffer_head *bh, int create, int flag,
1057 pgoff_t *next_pgofs)
1058 {
1059 struct f2fs_map_blocks map;
1060 int err;
1061
1062 map.m_lblk = iblock;
1063 map.m_len = bh->b_size >> inode->i_blkbits;
1064 map.m_next_pgofs = next_pgofs;
1065
1066 err = f2fs_map_blocks(inode, &map, create, flag);
1067 if (!err) {
1068 map_bh(bh, inode->i_sb, map.m_pblk);
1069 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1070 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1071 }
1072 return err;
1073 }
1074
1075 static int get_data_block(struct inode *inode, sector_t iblock,
1076 struct buffer_head *bh_result, int create, int flag,
1077 pgoff_t *next_pgofs)
1078 {
1079 return __get_data_block(inode, iblock, bh_result, create,
1080 flag, next_pgofs);
1081 }
1082
1083 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1084 struct buffer_head *bh_result, int create)
1085 {
1086 return __get_data_block(inode, iblock, bh_result, create,
1087 F2FS_GET_BLOCK_DEFAULT, NULL);
1088 }
1089
1090 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1091 struct buffer_head *bh_result, int create)
1092 {
1093 /* Block number less than F2FS MAX BLOCKS */
1094 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1095 return -EFBIG;
1096
1097 return __get_data_block(inode, iblock, bh_result, create,
1098 F2FS_GET_BLOCK_BMAP, NULL);
1099 }
1100
1101 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1102 {
1103 return (offset >> inode->i_blkbits);
1104 }
1105
1106 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1107 {
1108 return (blk << inode->i_blkbits);
1109 }
1110
1111 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1112 u64 start, u64 len)
1113 {
1114 struct buffer_head map_bh;
1115 sector_t start_blk, last_blk;
1116 pgoff_t next_pgofs;
1117 u64 logical = 0, phys = 0, size = 0;
1118 u32 flags = 0;
1119 int ret = 0;
1120
1121 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1122 if (ret)
1123 return ret;
1124
1125 if (f2fs_has_inline_data(inode)) {
1126 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1127 if (ret != -EAGAIN)
1128 return ret;
1129 }
1130
1131 inode_lock(inode);
1132
1133 if (logical_to_blk(inode, len) == 0)
1134 len = blk_to_logical(inode, 1);
1135
1136 start_blk = logical_to_blk(inode, start);
1137 last_blk = logical_to_blk(inode, start + len - 1);
1138
1139 next:
1140 memset(&map_bh, 0, sizeof(struct buffer_head));
1141 map_bh.b_size = len;
1142
1143 ret = get_data_block(inode, start_blk, &map_bh, 0,
1144 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1145 if (ret)
1146 goto out;
1147
1148 /* HOLE */
1149 if (!buffer_mapped(&map_bh)) {
1150 start_blk = next_pgofs;
1151
1152 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1153 F2FS_I_SB(inode)->max_file_blocks))
1154 goto prep_next;
1155
1156 flags |= FIEMAP_EXTENT_LAST;
1157 }
1158
1159 if (size) {
1160 if (f2fs_encrypted_inode(inode))
1161 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1162
1163 ret = fiemap_fill_next_extent(fieinfo, logical,
1164 phys, size, flags);
1165 }
1166
1167 if (start_blk > last_blk || ret)
1168 goto out;
1169
1170 logical = blk_to_logical(inode, start_blk);
1171 phys = blk_to_logical(inode, map_bh.b_blocknr);
1172 size = map_bh.b_size;
1173 flags = 0;
1174 if (buffer_unwritten(&map_bh))
1175 flags = FIEMAP_EXTENT_UNWRITTEN;
1176
1177 start_blk += logical_to_blk(inode, size);
1178
1179 prep_next:
1180 cond_resched();
1181 if (fatal_signal_pending(current))
1182 ret = -EINTR;
1183 else
1184 goto next;
1185 out:
1186 if (ret == 1)
1187 ret = 0;
1188
1189 inode_unlock(inode);
1190 return ret;
1191 }
1192
1193 /*
1194 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1195 * Major change was from block_size == page_size in f2fs by default.
1196 */
1197 static int f2fs_mpage_readpages(struct address_space *mapping,
1198 struct list_head *pages, struct page *page,
1199 unsigned nr_pages)
1200 {
1201 struct bio *bio = NULL;
1202 unsigned page_idx;
1203 sector_t last_block_in_bio = 0;
1204 struct inode *inode = mapping->host;
1205 const unsigned blkbits = inode->i_blkbits;
1206 const unsigned blocksize = 1 << blkbits;
1207 sector_t block_in_file;
1208 sector_t last_block;
1209 sector_t last_block_in_file;
1210 sector_t block_nr;
1211 struct f2fs_map_blocks map;
1212
1213 map.m_pblk = 0;
1214 map.m_lblk = 0;
1215 map.m_len = 0;
1216 map.m_flags = 0;
1217 map.m_next_pgofs = NULL;
1218
1219 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1220
1221 if (pages) {
1222 page = list_last_entry(pages, struct page, lru);
1223
1224 prefetchw(&page->flags);
1225 list_del(&page->lru);
1226 if (add_to_page_cache_lru(page, mapping,
1227 page->index,
1228 readahead_gfp_mask(mapping)))
1229 goto next_page;
1230 }
1231
1232 block_in_file = (sector_t)page->index;
1233 last_block = block_in_file + nr_pages;
1234 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1235 blkbits;
1236 if (last_block > last_block_in_file)
1237 last_block = last_block_in_file;
1238
1239 /*
1240 * Map blocks using the previous result first.
1241 */
1242 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1243 block_in_file > map.m_lblk &&
1244 block_in_file < (map.m_lblk + map.m_len))
1245 goto got_it;
1246
1247 /*
1248 * Then do more f2fs_map_blocks() calls until we are
1249 * done with this page.
1250 */
1251 map.m_flags = 0;
1252
1253 if (block_in_file < last_block) {
1254 map.m_lblk = block_in_file;
1255 map.m_len = last_block - block_in_file;
1256
1257 if (f2fs_map_blocks(inode, &map, 0,
1258 F2FS_GET_BLOCK_DEFAULT))
1259 goto set_error_page;
1260 }
1261 got_it:
1262 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1263 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1264 SetPageMappedToDisk(page);
1265
1266 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1267 SetPageUptodate(page);
1268 goto confused;
1269 }
1270 } else {
1271 zero_user_segment(page, 0, PAGE_SIZE);
1272 if (!PageUptodate(page))
1273 SetPageUptodate(page);
1274 unlock_page(page);
1275 goto next_page;
1276 }
1277
1278 /*
1279 * This page will go to BIO. Do we need to send this
1280 * BIO off first?
1281 */
1282 if (bio && (last_block_in_bio != block_nr - 1 ||
1283 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1284 submit_and_realloc:
1285 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1286 bio = NULL;
1287 }
1288 if (bio == NULL) {
1289 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages);
1290 if (IS_ERR(bio)) {
1291 bio = NULL;
1292 goto set_error_page;
1293 }
1294 }
1295
1296 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1297 goto submit_and_realloc;
1298
1299 last_block_in_bio = block_nr;
1300 goto next_page;
1301 set_error_page:
1302 SetPageError(page);
1303 zero_user_segment(page, 0, PAGE_SIZE);
1304 unlock_page(page);
1305 goto next_page;
1306 confused:
1307 if (bio) {
1308 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1309 bio = NULL;
1310 }
1311 unlock_page(page);
1312 next_page:
1313 if (pages)
1314 put_page(page);
1315 }
1316 BUG_ON(pages && !list_empty(pages));
1317 if (bio)
1318 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1319 return 0;
1320 }
1321
1322 static int f2fs_read_data_page(struct file *file, struct page *page)
1323 {
1324 struct inode *inode = page->mapping->host;
1325 int ret = -EAGAIN;
1326
1327 trace_f2fs_readpage(page, DATA);
1328
1329 /* If the file has inline data, try to read it directly */
1330 if (f2fs_has_inline_data(inode))
1331 ret = f2fs_read_inline_data(inode, page);
1332 if (ret == -EAGAIN)
1333 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1334 return ret;
1335 }
1336
1337 static int f2fs_read_data_pages(struct file *file,
1338 struct address_space *mapping,
1339 struct list_head *pages, unsigned nr_pages)
1340 {
1341 struct inode *inode = mapping->host;
1342 struct page *page = list_last_entry(pages, struct page, lru);
1343
1344 trace_f2fs_readpages(inode, page, nr_pages);
1345
1346 /* If the file has inline data, skip readpages */
1347 if (f2fs_has_inline_data(inode))
1348 return 0;
1349
1350 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1351 }
1352
1353 static int encrypt_one_page(struct f2fs_io_info *fio)
1354 {
1355 struct inode *inode = fio->page->mapping->host;
1356 gfp_t gfp_flags = GFP_NOFS;
1357
1358 if (!f2fs_encrypted_file(inode))
1359 return 0;
1360
1361 /* wait for GCed encrypted page writeback */
1362 f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1363
1364 retry_encrypt:
1365 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1366 PAGE_SIZE, 0, fio->page->index, gfp_flags);
1367 if (!IS_ERR(fio->encrypted_page))
1368 return 0;
1369
1370 /* flush pending IOs and wait for a while in the ENOMEM case */
1371 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1372 f2fs_flush_merged_writes(fio->sbi);
1373 congestion_wait(BLK_RW_ASYNC, HZ/50);
1374 gfp_flags |= __GFP_NOFAIL;
1375 goto retry_encrypt;
1376 }
1377 return PTR_ERR(fio->encrypted_page);
1378 }
1379
1380 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1381 {
1382 struct inode *inode = fio->page->mapping->host;
1383
1384 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
1385 return false;
1386 if (is_cold_data(fio->page))
1387 return false;
1388 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1389 return false;
1390
1391 return need_inplace_update_policy(inode, fio);
1392 }
1393
1394 static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio)
1395 {
1396 if (fio->old_blkaddr == NEW_ADDR)
1397 return false;
1398 if (fio->old_blkaddr == NULL_ADDR)
1399 return false;
1400 return true;
1401 }
1402
1403 int do_write_data_page(struct f2fs_io_info *fio)
1404 {
1405 struct page *page = fio->page;
1406 struct inode *inode = page->mapping->host;
1407 struct dnode_of_data dn;
1408 struct extent_info ei = {0,0,0};
1409 bool ipu_force = false;
1410 int err = 0;
1411
1412 set_new_dnode(&dn, inode, NULL, NULL, 0);
1413 if (need_inplace_update(fio) &&
1414 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1415 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1416
1417 if (valid_ipu_blkaddr(fio)) {
1418 ipu_force = true;
1419 fio->need_lock = LOCK_DONE;
1420 goto got_it;
1421 }
1422 }
1423
1424 /* Deadlock due to between page->lock and f2fs_lock_op */
1425 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1426 return -EAGAIN;
1427
1428 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1429 if (err)
1430 goto out;
1431
1432 fio->old_blkaddr = dn.data_blkaddr;
1433
1434 /* This page is already truncated */
1435 if (fio->old_blkaddr == NULL_ADDR) {
1436 ClearPageUptodate(page);
1437 goto out_writepage;
1438 }
1439 got_it:
1440 /*
1441 * If current allocation needs SSR,
1442 * it had better in-place writes for updated data.
1443 */
1444 if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) {
1445 err = encrypt_one_page(fio);
1446 if (err)
1447 goto out_writepage;
1448
1449 set_page_writeback(page);
1450 f2fs_put_dnode(&dn);
1451 if (fio->need_lock == LOCK_REQ)
1452 f2fs_unlock_op(fio->sbi);
1453 err = rewrite_data_page(fio);
1454 trace_f2fs_do_write_data_page(fio->page, IPU);
1455 set_inode_flag(inode, FI_UPDATE_WRITE);
1456 return err;
1457 }
1458
1459 if (fio->need_lock == LOCK_RETRY) {
1460 if (!f2fs_trylock_op(fio->sbi)) {
1461 err = -EAGAIN;
1462 goto out_writepage;
1463 }
1464 fio->need_lock = LOCK_REQ;
1465 }
1466
1467 err = encrypt_one_page(fio);
1468 if (err)
1469 goto out_writepage;
1470
1471 set_page_writeback(page);
1472
1473 /* LFS mode write path */
1474 write_data_page(&dn, fio);
1475 trace_f2fs_do_write_data_page(page, OPU);
1476 set_inode_flag(inode, FI_APPEND_WRITE);
1477 if (page->index == 0)
1478 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1479 out_writepage:
1480 f2fs_put_dnode(&dn);
1481 out:
1482 if (fio->need_lock == LOCK_REQ)
1483 f2fs_unlock_op(fio->sbi);
1484 return err;
1485 }
1486
1487 static int __write_data_page(struct page *page, bool *submitted,
1488 struct writeback_control *wbc,
1489 enum iostat_type io_type)
1490 {
1491 struct inode *inode = page->mapping->host;
1492 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1493 loff_t i_size = i_size_read(inode);
1494 const pgoff_t end_index = ((unsigned long long) i_size)
1495 >> PAGE_SHIFT;
1496 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1497 unsigned offset = 0;
1498 bool need_balance_fs = false;
1499 int err = 0;
1500 struct f2fs_io_info fio = {
1501 .sbi = sbi,
1502 .ino = inode->i_ino,
1503 .type = DATA,
1504 .op = REQ_OP_WRITE,
1505 .op_flags = wbc_to_write_flags(wbc),
1506 .old_blkaddr = NULL_ADDR,
1507 .page = page,
1508 .encrypted_page = NULL,
1509 .submitted = false,
1510 .need_lock = LOCK_RETRY,
1511 .io_type = io_type,
1512 };
1513
1514 trace_f2fs_writepage(page, DATA);
1515
1516 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1517 goto redirty_out;
1518
1519 if (page->index < end_index)
1520 goto write;
1521
1522 /*
1523 * If the offset is out-of-range of file size,
1524 * this page does not have to be written to disk.
1525 */
1526 offset = i_size & (PAGE_SIZE - 1);
1527 if ((page->index >= end_index + 1) || !offset)
1528 goto out;
1529
1530 zero_user_segment(page, offset, PAGE_SIZE);
1531 write:
1532 if (f2fs_is_drop_cache(inode))
1533 goto out;
1534 /* we should not write 0'th page having journal header */
1535 if (f2fs_is_volatile_file(inode) && (!page->index ||
1536 (!wbc->for_reclaim &&
1537 available_free_memory(sbi, BASE_CHECK))))
1538 goto redirty_out;
1539
1540 /* we should bypass data pages to proceed the kworkder jobs */
1541 if (unlikely(f2fs_cp_error(sbi))) {
1542 mapping_set_error(page->mapping, -EIO);
1543 goto out;
1544 }
1545
1546 /* Dentry blocks are controlled by checkpoint */
1547 if (S_ISDIR(inode->i_mode)) {
1548 fio.need_lock = LOCK_DONE;
1549 err = do_write_data_page(&fio);
1550 goto done;
1551 }
1552
1553 if (!wbc->for_reclaim)
1554 need_balance_fs = true;
1555 else if (has_not_enough_free_secs(sbi, 0, 0))
1556 goto redirty_out;
1557 else
1558 set_inode_flag(inode, FI_HOT_DATA);
1559
1560 err = -EAGAIN;
1561 if (f2fs_has_inline_data(inode)) {
1562 err = f2fs_write_inline_data(inode, page);
1563 if (!err)
1564 goto out;
1565 }
1566
1567 if (err == -EAGAIN) {
1568 err = do_write_data_page(&fio);
1569 if (err == -EAGAIN) {
1570 fio.need_lock = LOCK_REQ;
1571 err = do_write_data_page(&fio);
1572 }
1573 }
1574
1575 down_write(&F2FS_I(inode)->i_sem);
1576 if (F2FS_I(inode)->last_disk_size < psize)
1577 F2FS_I(inode)->last_disk_size = psize;
1578 up_write(&F2FS_I(inode)->i_sem);
1579
1580 done:
1581 if (err && err != -ENOENT)
1582 goto redirty_out;
1583
1584 out:
1585 inode_dec_dirty_pages(inode);
1586 if (err)
1587 ClearPageUptodate(page);
1588
1589 if (wbc->for_reclaim) {
1590 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1591 clear_inode_flag(inode, FI_HOT_DATA);
1592 remove_dirty_inode(inode);
1593 submitted = NULL;
1594 }
1595
1596 unlock_page(page);
1597 if (!S_ISDIR(inode->i_mode))
1598 f2fs_balance_fs(sbi, need_balance_fs);
1599
1600 if (unlikely(f2fs_cp_error(sbi))) {
1601 f2fs_submit_merged_write(sbi, DATA);
1602 submitted = NULL;
1603 }
1604
1605 if (submitted)
1606 *submitted = fio.submitted;
1607
1608 return 0;
1609
1610 redirty_out:
1611 redirty_page_for_writepage(wbc, page);
1612 if (!err)
1613 return AOP_WRITEPAGE_ACTIVATE;
1614 unlock_page(page);
1615 return err;
1616 }
1617
1618 static int f2fs_write_data_page(struct page *page,
1619 struct writeback_control *wbc)
1620 {
1621 return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1622 }
1623
1624 /*
1625 * This function was copied from write_cche_pages from mm/page-writeback.c.
1626 * The major change is making write step of cold data page separately from
1627 * warm/hot data page.
1628 */
1629 static int f2fs_write_cache_pages(struct address_space *mapping,
1630 struct writeback_control *wbc,
1631 enum iostat_type io_type)
1632 {
1633 int ret = 0;
1634 int done = 0;
1635 struct pagevec pvec;
1636 int nr_pages;
1637 pgoff_t uninitialized_var(writeback_index);
1638 pgoff_t index;
1639 pgoff_t end; /* Inclusive */
1640 pgoff_t done_index;
1641 pgoff_t last_idx = ULONG_MAX;
1642 int cycled;
1643 int range_whole = 0;
1644 int tag;
1645
1646 pagevec_init(&pvec, 0);
1647
1648 if (get_dirty_pages(mapping->host) <=
1649 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1650 set_inode_flag(mapping->host, FI_HOT_DATA);
1651 else
1652 clear_inode_flag(mapping->host, FI_HOT_DATA);
1653
1654 if (wbc->range_cyclic) {
1655 writeback_index = mapping->writeback_index; /* prev offset */
1656 index = writeback_index;
1657 if (index == 0)
1658 cycled = 1;
1659 else
1660 cycled = 0;
1661 end = -1;
1662 } else {
1663 index = wbc->range_start >> PAGE_SHIFT;
1664 end = wbc->range_end >> PAGE_SHIFT;
1665 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1666 range_whole = 1;
1667 cycled = 1; /* ignore range_cyclic tests */
1668 }
1669 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1670 tag = PAGECACHE_TAG_TOWRITE;
1671 else
1672 tag = PAGECACHE_TAG_DIRTY;
1673 retry:
1674 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1675 tag_pages_for_writeback(mapping, index, end);
1676 done_index = index;
1677 while (!done && (index <= end)) {
1678 int i;
1679
1680 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1681 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1682 if (nr_pages == 0)
1683 break;
1684
1685 for (i = 0; i < nr_pages; i++) {
1686 struct page *page = pvec.pages[i];
1687 bool submitted = false;
1688
1689 if (page->index > end) {
1690 done = 1;
1691 break;
1692 }
1693
1694 done_index = page->index;
1695 retry_write:
1696 lock_page(page);
1697
1698 if (unlikely(page->mapping != mapping)) {
1699 continue_unlock:
1700 unlock_page(page);
1701 continue;
1702 }
1703
1704 if (!PageDirty(page)) {
1705 /* someone wrote it for us */
1706 goto continue_unlock;
1707 }
1708
1709 if (PageWriteback(page)) {
1710 if (wbc->sync_mode != WB_SYNC_NONE)
1711 f2fs_wait_on_page_writeback(page,
1712 DATA, true);
1713 else
1714 goto continue_unlock;
1715 }
1716
1717 BUG_ON(PageWriteback(page));
1718 if (!clear_page_dirty_for_io(page))
1719 goto continue_unlock;
1720
1721 ret = __write_data_page(page, &submitted, wbc, io_type);
1722 if (unlikely(ret)) {
1723 /*
1724 * keep nr_to_write, since vfs uses this to
1725 * get # of written pages.
1726 */
1727 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1728 unlock_page(page);
1729 ret = 0;
1730 continue;
1731 } else if (ret == -EAGAIN) {
1732 ret = 0;
1733 if (wbc->sync_mode == WB_SYNC_ALL) {
1734 cond_resched();
1735 congestion_wait(BLK_RW_ASYNC,
1736 HZ/50);
1737 goto retry_write;
1738 }
1739 continue;
1740 }
1741 done_index = page->index + 1;
1742 done = 1;
1743 break;
1744 } else if (submitted) {
1745 last_idx = page->index;
1746 }
1747
1748 /* give a priority to WB_SYNC threads */
1749 if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) ||
1750 --wbc->nr_to_write <= 0) &&
1751 wbc->sync_mode == WB_SYNC_NONE) {
1752 done = 1;
1753 break;
1754 }
1755 }
1756 pagevec_release(&pvec);
1757 cond_resched();
1758 }
1759
1760 if (!cycled && !done) {
1761 cycled = 1;
1762 index = 0;
1763 end = writeback_index - 1;
1764 goto retry;
1765 }
1766 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1767 mapping->writeback_index = done_index;
1768
1769 if (last_idx != ULONG_MAX)
1770 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
1771 0, last_idx, DATA);
1772
1773 return ret;
1774 }
1775
1776 int __f2fs_write_data_pages(struct address_space *mapping,
1777 struct writeback_control *wbc,
1778 enum iostat_type io_type)
1779 {
1780 struct inode *inode = mapping->host;
1781 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1782 struct blk_plug plug;
1783 int ret;
1784
1785 /* deal with chardevs and other special file */
1786 if (!mapping->a_ops->writepage)
1787 return 0;
1788
1789 /* skip writing if there is no dirty page in this inode */
1790 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1791 return 0;
1792
1793 /* during POR, we don't need to trigger writepage at all. */
1794 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1795 goto skip_write;
1796
1797 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1798 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1799 available_free_memory(sbi, DIRTY_DENTS))
1800 goto skip_write;
1801
1802 /* skip writing during file defragment */
1803 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1804 goto skip_write;
1805
1806 trace_f2fs_writepages(mapping->host, wbc, DATA);
1807
1808 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
1809 if (wbc->sync_mode == WB_SYNC_ALL)
1810 atomic_inc(&sbi->wb_sync_req);
1811 else if (atomic_read(&sbi->wb_sync_req))
1812 goto skip_write;
1813
1814 blk_start_plug(&plug);
1815 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
1816 blk_finish_plug(&plug);
1817
1818 if (wbc->sync_mode == WB_SYNC_ALL)
1819 atomic_dec(&sbi->wb_sync_req);
1820 /*
1821 * if some pages were truncated, we cannot guarantee its mapping->host
1822 * to detect pending bios.
1823 */
1824
1825 remove_dirty_inode(inode);
1826 return ret;
1827
1828 skip_write:
1829 wbc->pages_skipped += get_dirty_pages(inode);
1830 trace_f2fs_writepages(mapping->host, wbc, DATA);
1831 return 0;
1832 }
1833
1834 static int f2fs_write_data_pages(struct address_space *mapping,
1835 struct writeback_control *wbc)
1836 {
1837 struct inode *inode = mapping->host;
1838
1839 return __f2fs_write_data_pages(mapping, wbc,
1840 F2FS_I(inode)->cp_task == current ?
1841 FS_CP_DATA_IO : FS_DATA_IO);
1842 }
1843
1844 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1845 {
1846 struct inode *inode = mapping->host;
1847 loff_t i_size = i_size_read(inode);
1848
1849 if (to > i_size) {
1850 down_write(&F2FS_I(inode)->i_mmap_sem);
1851 truncate_pagecache(inode, i_size);
1852 truncate_blocks(inode, i_size, true);
1853 up_write(&F2FS_I(inode)->i_mmap_sem);
1854 }
1855 }
1856
1857 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1858 struct page *page, loff_t pos, unsigned len,
1859 block_t *blk_addr, bool *node_changed)
1860 {
1861 struct inode *inode = page->mapping->host;
1862 pgoff_t index = page->index;
1863 struct dnode_of_data dn;
1864 struct page *ipage;
1865 bool locked = false;
1866 struct extent_info ei = {0,0,0};
1867 int err = 0;
1868
1869 /*
1870 * we already allocated all the blocks, so we don't need to get
1871 * the block addresses when there is no need to fill the page.
1872 */
1873 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1874 !is_inode_flag_set(inode, FI_NO_PREALLOC))
1875 return 0;
1876
1877 if (f2fs_has_inline_data(inode) ||
1878 (pos & PAGE_MASK) >= i_size_read(inode)) {
1879 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1880 locked = true;
1881 }
1882 restart:
1883 /* check inline_data */
1884 ipage = get_node_page(sbi, inode->i_ino);
1885 if (IS_ERR(ipage)) {
1886 err = PTR_ERR(ipage);
1887 goto unlock_out;
1888 }
1889
1890 set_new_dnode(&dn, inode, ipage, ipage, 0);
1891
1892 if (f2fs_has_inline_data(inode)) {
1893 if (pos + len <= MAX_INLINE_DATA(inode)) {
1894 read_inline_data(page, ipage);
1895 set_inode_flag(inode, FI_DATA_EXIST);
1896 if (inode->i_nlink)
1897 set_inline_node(ipage);
1898 } else {
1899 err = f2fs_convert_inline_page(&dn, page);
1900 if (err)
1901 goto out;
1902 if (dn.data_blkaddr == NULL_ADDR)
1903 err = f2fs_get_block(&dn, index);
1904 }
1905 } else if (locked) {
1906 err = f2fs_get_block(&dn, index);
1907 } else {
1908 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1909 dn.data_blkaddr = ei.blk + index - ei.fofs;
1910 } else {
1911 /* hole case */
1912 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1913 if (err || dn.data_blkaddr == NULL_ADDR) {
1914 f2fs_put_dnode(&dn);
1915 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
1916 true);
1917 locked = true;
1918 goto restart;
1919 }
1920 }
1921 }
1922
1923 /* convert_inline_page can make node_changed */
1924 *blk_addr = dn.data_blkaddr;
1925 *node_changed = dn.node_changed;
1926 out:
1927 f2fs_put_dnode(&dn);
1928 unlock_out:
1929 if (locked)
1930 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1931 return err;
1932 }
1933
1934 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1935 loff_t pos, unsigned len, unsigned flags,
1936 struct page **pagep, void **fsdata)
1937 {
1938 struct inode *inode = mapping->host;
1939 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1940 struct page *page = NULL;
1941 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1942 bool need_balance = false;
1943 block_t blkaddr = NULL_ADDR;
1944 int err = 0;
1945
1946 if (trace_android_fs_datawrite_start_enabled()) {
1947 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
1948
1949 path = android_fstrace_get_pathname(pathbuf,
1950 MAX_TRACE_PATHBUF_LEN,
1951 inode);
1952 trace_android_fs_datawrite_start(inode, pos, len,
1953 current->pid, path,
1954 current->comm);
1955 }
1956 trace_f2fs_write_begin(inode, pos, len, flags);
1957
1958 if (f2fs_is_atomic_file(inode) &&
1959 !available_free_memory(sbi, INMEM_PAGES)) {
1960 err = -ENOMEM;
1961 goto fail;
1962 }
1963
1964 /*
1965 * We should check this at this moment to avoid deadlock on inode page
1966 * and #0 page. The locking rule for inline_data conversion should be:
1967 * lock_page(page #0) -> lock_page(inode_page)
1968 */
1969 if (index != 0) {
1970 err = f2fs_convert_inline_inode(inode);
1971 if (err)
1972 goto fail;
1973 }
1974 repeat:
1975 /*
1976 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1977 * wait_for_stable_page. Will wait that below with our IO control.
1978 */
1979 page = f2fs_pagecache_get_page(mapping, index,
1980 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1981 if (!page) {
1982 err = -ENOMEM;
1983 goto fail;
1984 }
1985
1986 *pagep = page;
1987
1988 err = prepare_write_begin(sbi, page, pos, len,
1989 &blkaddr, &need_balance);
1990 if (err)
1991 goto fail;
1992
1993 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1994 unlock_page(page);
1995 f2fs_balance_fs(sbi, true);
1996 lock_page(page);
1997 if (page->mapping != mapping) {
1998 /* The page got truncated from under us */
1999 f2fs_put_page(page, 1);
2000 goto repeat;
2001 }
2002 }
2003
2004 f2fs_wait_on_page_writeback(page, DATA, false);
2005
2006 /* wait for GCed encrypted page writeback */
2007 if (f2fs_encrypted_file(inode))
2008 f2fs_wait_on_block_writeback(sbi, blkaddr);
2009
2010 if (len == PAGE_SIZE || PageUptodate(page))
2011 return 0;
2012
2013 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2014 zero_user_segment(page, len, PAGE_SIZE);
2015 return 0;
2016 }
2017
2018 if (blkaddr == NEW_ADDR) {
2019 zero_user_segment(page, 0, PAGE_SIZE);
2020 SetPageUptodate(page);
2021 } else {
2022 err = f2fs_submit_page_read(inode, page, blkaddr);
2023 if (err)
2024 goto fail;
2025
2026 lock_page(page);
2027 if (unlikely(page->mapping != mapping)) {
2028 f2fs_put_page(page, 1);
2029 goto repeat;
2030 }
2031 if (unlikely(!PageUptodate(page))) {
2032 err = -EIO;
2033 goto fail;
2034 }
2035 }
2036 return 0;
2037
2038 fail:
2039 f2fs_put_page(page, 1);
2040 f2fs_write_failed(mapping, pos + len);
2041 if (f2fs_is_atomic_file(inode))
2042 drop_inmem_pages_all(sbi);
2043 return err;
2044 }
2045
2046 static int f2fs_write_end(struct file *file,
2047 struct address_space *mapping,
2048 loff_t pos, unsigned len, unsigned copied,
2049 struct page *page, void *fsdata)
2050 {
2051 struct inode *inode = page->mapping->host;
2052
2053 trace_android_fs_datawrite_end(inode, pos, len);
2054 trace_f2fs_write_end(inode, pos, len, copied);
2055
2056 /*
2057 * This should be come from len == PAGE_SIZE, and we expect copied
2058 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2059 * let generic_perform_write() try to copy data again through copied=0.
2060 */
2061 if (!PageUptodate(page)) {
2062 if (unlikely(copied != len))
2063 copied = 0;
2064 else
2065 SetPageUptodate(page);
2066 }
2067 if (!copied)
2068 goto unlock_out;
2069
2070 set_page_dirty(page);
2071
2072 if (pos + copied > i_size_read(inode))
2073 f2fs_i_size_write(inode, pos + copied);
2074 unlock_out:
2075 f2fs_put_page(page, 1);
2076 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2077 return copied;
2078 }
2079
2080 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2081 loff_t offset)
2082 {
2083 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
2084
2085 if (offset & blocksize_mask)
2086 return -EINVAL;
2087
2088 if (iov_iter_alignment(iter) & blocksize_mask)
2089 return -EINVAL;
2090
2091 return 0;
2092 }
2093
2094 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2095 {
2096 struct address_space *mapping = iocb->ki_filp->f_mapping;
2097 struct inode *inode = mapping->host;
2098 size_t count = iov_iter_count(iter);
2099 loff_t offset = iocb->ki_pos;
2100 int rw = iov_iter_rw(iter);
2101 int err;
2102
2103 err = check_direct_IO(inode, iter, offset);
2104 if (err)
2105 return err;
2106
2107 if (__force_buffered_io(inode, rw))
2108 return 0;
2109
2110 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2111
2112 if (trace_android_fs_dataread_start_enabled() &&
2113 (rw == READ)) {
2114 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
2115
2116 path = android_fstrace_get_pathname(pathbuf,
2117 MAX_TRACE_PATHBUF_LEN,
2118 inode);
2119 trace_android_fs_dataread_start(inode, offset,
2120 count, current->pid, path,
2121 current->comm);
2122 }
2123 if (trace_android_fs_datawrite_start_enabled() &&
2124 (rw == WRITE)) {
2125 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
2126
2127 path = android_fstrace_get_pathname(pathbuf,
2128 MAX_TRACE_PATHBUF_LEN,
2129 inode);
2130 trace_android_fs_datawrite_start(inode, offset, count,
2131 current->pid, path,
2132 current->comm);
2133 }
2134
2135 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
2136 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2137 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
2138
2139 if (rw == WRITE) {
2140 if (err > 0) {
2141 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2142 err);
2143 set_inode_flag(inode, FI_UPDATE_WRITE);
2144 } else if (err < 0) {
2145 f2fs_write_failed(mapping, offset + count);
2146 }
2147 }
2148
2149 if (trace_android_fs_dataread_start_enabled() &&
2150 (rw == READ))
2151 trace_android_fs_dataread_end(inode, offset, count);
2152 if (trace_android_fs_datawrite_start_enabled() &&
2153 (rw == WRITE))
2154 trace_android_fs_datawrite_end(inode, offset, count);
2155
2156 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2157
2158 return err;
2159 }
2160
2161 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2162 unsigned int length)
2163 {
2164 struct inode *inode = page->mapping->host;
2165 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2166
2167 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2168 (offset % PAGE_SIZE || length != PAGE_SIZE))
2169 return;
2170
2171 if (PageDirty(page)) {
2172 if (inode->i_ino == F2FS_META_INO(sbi)) {
2173 dec_page_count(sbi, F2FS_DIRTY_META);
2174 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2175 dec_page_count(sbi, F2FS_DIRTY_NODES);
2176 } else {
2177 inode_dec_dirty_pages(inode);
2178 remove_dirty_inode(inode);
2179 }
2180 }
2181
2182 /* This is atomic written page, keep Private */
2183 if (IS_ATOMIC_WRITTEN_PAGE(page))
2184 return drop_inmem_page(inode, page);
2185
2186 set_page_private(page, 0);
2187 ClearPagePrivate(page);
2188 }
2189
2190 int f2fs_release_page(struct page *page, gfp_t wait)
2191 {
2192 /* If this is dirty page, keep PagePrivate */
2193 if (PageDirty(page))
2194 return 0;
2195
2196 /* This is atomic written page, keep Private */
2197 if (IS_ATOMIC_WRITTEN_PAGE(page))
2198 return 0;
2199
2200 set_page_private(page, 0);
2201 ClearPagePrivate(page);
2202 return 1;
2203 }
2204
2205 /*
2206 * This was copied from __set_page_dirty_buffers which gives higher performance
2207 * in very high speed storages. (e.g., pmem)
2208 */
2209 void f2fs_set_page_dirty_nobuffers(struct page *page)
2210 {
2211 struct address_space *mapping = page->mapping;
2212 unsigned long flags;
2213
2214 if (unlikely(!mapping))
2215 return;
2216
2217 spin_lock(&mapping->private_lock);
2218 lock_page_memcg(page);
2219 SetPageDirty(page);
2220 spin_unlock(&mapping->private_lock);
2221
2222 spin_lock_irqsave(&mapping->tree_lock, flags);
2223 WARN_ON_ONCE(!PageUptodate(page));
2224 account_page_dirtied(page, mapping);
2225 radix_tree_tag_set(&mapping->page_tree,
2226 page_index(page), PAGECACHE_TAG_DIRTY);
2227 spin_unlock_irqrestore(&mapping->tree_lock, flags);
2228 unlock_page_memcg(page);
2229
2230 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2231 return;
2232 }
2233
2234 static int f2fs_set_data_page_dirty(struct page *page)
2235 {
2236 struct address_space *mapping = page->mapping;
2237 struct inode *inode = mapping->host;
2238
2239 trace_f2fs_set_page_dirty(page, DATA);
2240
2241 if (!PageUptodate(page))
2242 SetPageUptodate(page);
2243
2244 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2245 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2246 register_inmem_page(inode, page);
2247 return 1;
2248 }
2249 /*
2250 * Previously, this page has been registered, we just
2251 * return here.
2252 */
2253 return 0;
2254 }
2255
2256 if (!PageDirty(page)) {
2257 f2fs_set_page_dirty_nobuffers(page);
2258 update_dirty_page(inode, page);
2259 return 1;
2260 }
2261 return 0;
2262 }
2263
2264 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2265 {
2266 struct inode *inode = mapping->host;
2267
2268 if (f2fs_has_inline_data(inode))
2269 return 0;
2270
2271 /* make sure allocating whole blocks */
2272 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2273 filemap_write_and_wait(mapping);
2274
2275 return generic_block_bmap(mapping, block, get_data_block_bmap);
2276 }
2277
2278 #ifdef CONFIG_MIGRATION
2279 #include <linux/migrate.h>
2280
2281 int f2fs_migrate_page(struct address_space *mapping,
2282 struct page *newpage, struct page *page, enum migrate_mode mode)
2283 {
2284 int rc, extra_count;
2285 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2286 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2287
2288 BUG_ON(PageWriteback(page));
2289
2290 /* migrating an atomic written page is safe with the inmem_lock hold */
2291 if (atomic_written) {
2292 if (mode != MIGRATE_SYNC)
2293 return -EBUSY;
2294 if (!mutex_trylock(&fi->inmem_lock))
2295 return -EAGAIN;
2296 }
2297
2298 /*
2299 * A reference is expected if PagePrivate set when move mapping,
2300 * however F2FS breaks this for maintaining dirty page counts when
2301 * truncating pages. So here adjusting the 'extra_count' make it work.
2302 */
2303 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2304 rc = migrate_page_move_mapping(mapping, newpage,
2305 page, NULL, mode, extra_count);
2306 if (rc != MIGRATEPAGE_SUCCESS) {
2307 if (atomic_written)
2308 mutex_unlock(&fi->inmem_lock);
2309 return rc;
2310 }
2311
2312 if (atomic_written) {
2313 struct inmem_pages *cur;
2314 list_for_each_entry(cur, &fi->inmem_pages, list)
2315 if (cur->page == page) {
2316 cur->page = newpage;
2317 break;
2318 }
2319 mutex_unlock(&fi->inmem_lock);
2320 put_page(page);
2321 get_page(newpage);
2322 }
2323
2324 if (PagePrivate(page))
2325 SetPagePrivate(newpage);
2326 set_page_private(newpage, page_private(page));
2327
2328 if (mode != MIGRATE_SYNC_NO_COPY)
2329 migrate_page_copy(newpage, page);
2330 else
2331 migrate_page_states(newpage, page);
2332
2333 return MIGRATEPAGE_SUCCESS;
2334 }
2335 #endif
2336
2337 const struct address_space_operations f2fs_dblock_aops = {
2338 .readpage = f2fs_read_data_page,
2339 .readpages = f2fs_read_data_pages,
2340 .writepage = f2fs_write_data_page,
2341 .writepages = f2fs_write_data_pages,
2342 .write_begin = f2fs_write_begin,
2343 .write_end = f2fs_write_end,
2344 .set_page_dirty = f2fs_set_data_page_dirty,
2345 .invalidatepage = f2fs_invalidate_page,
2346 .releasepage = f2fs_release_page,
2347 .direct_IO = f2fs_direct_IO,
2348 .bmap = f2fs_bmap,
2349 #ifdef CONFIG_MIGRATION
2350 .migratepage = f2fs_migrate_page,
2351 #endif
2352 };