projects / GitHub / LineageOS / android_kernel_motorola_exynos9610.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs/fscrypt: updates to v4.17-rc1
[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 if (type == F2FS_WB_CP_DATA)
116 f2fs_stop_checkpoint(sbi, true);
117 }
118
119 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
120 page->index != nid_of_node(page));
121
122 dec_page_count(sbi, type);
123 clear_cold_data(page);
124 end_page_writeback(page);
125 }
126 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
127 wq_has_sleeper(&sbi->cp_wait))
128 wake_up(&sbi->cp_wait);
129
130 bio_put(bio);
131 }
132
133 /*
134 * Return true, if pre_bio's bdev is same as its target device.
135 */
136 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
137 block_t blk_addr, struct bio *bio)
138 {
139 struct block_device *bdev = sbi->sb->s_bdev;
140 int i;
141
142 for (i = 0; i < sbi->s_ndevs; i++) {
143 if (FDEV(i).start_blk <= blk_addr &&
144 FDEV(i).end_blk >= blk_addr) {
145 blk_addr -= FDEV(i).start_blk;
146 bdev = FDEV(i).bdev;
147 break;
148 }
149 }
150 if (bio) {
151 bio_set_dev(bio, bdev);
152 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
153 }
154 return bdev;
155 }
156
157 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
158 {
159 int i;
160
161 for (i = 0; i < sbi->s_ndevs; i++)
162 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
163 return i;
164 return 0;
165 }
166
167 static bool __same_bdev(struct f2fs_sb_info *sbi,
168 block_t blk_addr, struct bio *bio)
169 {
170 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
171 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
172 }
173
174 /*
175 * Low-level block read/write IO operations.
176 */
177 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
178 struct writeback_control *wbc,
179 int npages, bool is_read,
180 enum page_type type, enum temp_type temp)
181 {
182 struct bio *bio;
183
184 bio = f2fs_bio_alloc(sbi, npages, true);
185
186 f2fs_target_device(sbi, blk_addr, bio);
187 if (is_read) {
188 bio->bi_end_io = f2fs_read_end_io;
189 bio->bi_private = NULL;
190 } else {
191 bio->bi_end_io = f2fs_write_end_io;
192 bio->bi_private = sbi;
193 bio->bi_write_hint = io_type_to_rw_hint(sbi, type, temp);
194 }
195 if (wbc)
196 wbc_init_bio(wbc, bio);
197
198 return bio;
199 }
200
201 static inline void __submit_bio(struct f2fs_sb_info *sbi,
202 struct bio *bio, enum page_type type)
203 {
204 if (!is_read_io(bio_op(bio))) {
205 unsigned int start;
206
207 if (type != DATA && type != NODE)
208 goto submit_io;
209
210 if (f2fs_sb_has_blkzoned(sbi->sb) && current->plug)
211 blk_finish_plug(current->plug);
212
213 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
214 start %= F2FS_IO_SIZE(sbi);
215
216 if (start == 0)
217 goto submit_io;
218
219 /* fill dummy pages */
220 for (; start < F2FS_IO_SIZE(sbi); start++) {
221 struct page *page =
222 mempool_alloc(sbi->write_io_dummy,
223 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
224 f2fs_bug_on(sbi, !page);
225
226 SetPagePrivate(page);
227 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
228 lock_page(page);
229 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
230 f2fs_bug_on(sbi, 1);
231 }
232 /*
233 * In the NODE case, we lose next block address chain. So, we
234 * need to do checkpoint in f2fs_sync_file.
235 */
236 if (type == NODE)
237 set_sbi_flag(sbi, SBI_NEED_CP);
238 }
239 submit_io:
240 if (is_read_io(bio_op(bio)))
241 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
242 else
243 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
244 submit_bio(bio);
245 }
246
247 static void __submit_merged_bio(struct f2fs_bio_info *io)
248 {
249 struct f2fs_io_info *fio = &io->fio;
250
251 if (!io->bio)
252 return;
253
254 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
255
256 if (is_read_io(fio->op))
257 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
258 else
259 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
260
261 __submit_bio(io->sbi, io->bio, fio->type);
262 io->bio = NULL;
263 }
264
265 static bool __has_merged_page(struct f2fs_bio_info *io,
266 struct inode *inode, nid_t ino, pgoff_t idx)
267 {
268 struct bio_vec *bvec;
269 struct page *target;
270 int i;
271
272 if (!io->bio)
273 return false;
274
275 if (!inode && !ino)
276 return true;
277
278 bio_for_each_segment_all(bvec, io->bio, i) {
279
280 if (bvec->bv_page->mapping)
281 target = bvec->bv_page;
282 else
283 target = fscrypt_control_page(bvec->bv_page);
284
285 if (idx != target->index)
286 continue;
287
288 if (inode && inode == target->mapping->host)
289 return true;
290 if (ino && ino == ino_of_node(target))
291 return true;
292 }
293
294 return false;
295 }
296
297 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
298 nid_t ino, pgoff_t idx, enum page_type type)
299 {
300 enum page_type btype = PAGE_TYPE_OF_BIO(type);
301 enum temp_type temp;
302 struct f2fs_bio_info *io;
303 bool ret = false;
304
305 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
306 io = sbi->write_io[btype] + temp;
307
308 down_read(&io->io_rwsem);
309 ret = __has_merged_page(io, inode, ino, idx);
310 up_read(&io->io_rwsem);
311
312 /* TODO: use HOT temp only for meta pages now. */
313 if (ret || btype == META)
314 break;
315 }
316 return ret;
317 }
318
319 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
320 enum page_type type, enum temp_type temp)
321 {
322 enum page_type btype = PAGE_TYPE_OF_BIO(type);
323 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
324
325 down_write(&io->io_rwsem);
326
327 /* change META to META_FLUSH in the checkpoint procedure */
328 if (type >= META_FLUSH) {
329 io->fio.type = META_FLUSH;
330 io->fio.op = REQ_OP_WRITE;
331 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
332 if (!test_opt(sbi, NOBARRIER))
333 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
334 }
335 __submit_merged_bio(io);
336 up_write(&io->io_rwsem);
337 }
338
339 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
340 struct inode *inode, nid_t ino, pgoff_t idx,
341 enum page_type type, bool force)
342 {
343 enum temp_type temp;
344
345 if (!force && !has_merged_page(sbi, inode, ino, idx, type))
346 return;
347
348 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
349
350 __f2fs_submit_merged_write(sbi, type, temp);
351
352 /* TODO: use HOT temp only for meta pages now. */
353 if (type >= META)
354 break;
355 }
356 }
357
358 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
359 {
360 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
361 }
362
363 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
364 struct inode *inode, nid_t ino, pgoff_t idx,
365 enum page_type type)
366 {
367 __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
368 }
369
370 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
371 {
372 f2fs_submit_merged_write(sbi, DATA);
373 f2fs_submit_merged_write(sbi, NODE);
374 f2fs_submit_merged_write(sbi, META);
375 }
376
377 /*
378 * Fill the locked page with data located in the block address.
379 * A caller needs to unlock the page on failure.
380 */
381 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
382 {
383 struct bio *bio;
384 struct page *page = fio->encrypted_page ?
385 fio->encrypted_page : fio->page;
386
387 verify_block_addr(fio, fio->new_blkaddr);
388 trace_f2fs_submit_page_bio(page, fio);
389 f2fs_trace_ios(fio, 0);
390
391 /* Allocate a new bio */
392 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
393 1, is_read_io(fio->op), fio->type, fio->temp);
394
395 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
396 bio_put(bio);
397 return -EFAULT;
398 }
399 bio_set_op_attrs(bio, fio->op, fio->op_flags);
400
401 __submit_bio(fio->sbi, bio, fio->type);
402
403 if (!is_read_io(fio->op))
404 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
405 return 0;
406 }
407
408 int f2fs_submit_page_write(struct f2fs_io_info *fio)
409 {
410 struct f2fs_sb_info *sbi = fio->sbi;
411 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
412 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
413 struct page *bio_page;
414 int err = 0;
415
416 f2fs_bug_on(sbi, is_read_io(fio->op));
417
418 down_write(&io->io_rwsem);
419 next:
420 if (fio->in_list) {
421 spin_lock(&io->io_lock);
422 if (list_empty(&io->io_list)) {
423 spin_unlock(&io->io_lock);
424 goto out_fail;
425 }
426 fio = list_first_entry(&io->io_list,
427 struct f2fs_io_info, list);
428 list_del(&fio->list);
429 spin_unlock(&io->io_lock);
430 }
431
432 if (fio->old_blkaddr != NEW_ADDR)
433 verify_block_addr(fio, fio->old_blkaddr);
434 verify_block_addr(fio, fio->new_blkaddr);
435
436 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
437
438 /* set submitted = true as a return value */
439 fio->submitted = true;
440
441 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
442
443 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
444 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
445 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
446 __submit_merged_bio(io);
447 alloc_new:
448 if (io->bio == NULL) {
449 if ((fio->type == DATA || fio->type == NODE) &&
450 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
451 err = -EAGAIN;
452 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
453 goto out_fail;
454 }
455 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
456 BIO_MAX_PAGES, false,
457 fio->type, fio->temp);
458 io->fio = *fio;
459 }
460
461 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
462 __submit_merged_bio(io);
463 goto alloc_new;
464 }
465
466 if (fio->io_wbc)
467 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
468
469 io->last_block_in_bio = fio->new_blkaddr;
470 f2fs_trace_ios(fio, 0);
471
472 trace_f2fs_submit_page_write(fio->page, fio);
473
474 if (fio->in_list)
475 goto next;
476 out_fail:
477 up_write(&io->io_rwsem);
478 return err;
479 }
480
481 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
482 unsigned nr_pages)
483 {
484 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
485 struct fscrypt_ctx *ctx = NULL;
486 struct bio *bio;
487
488 if (f2fs_encrypted_file(inode)) {
489 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
490 if (IS_ERR(ctx))
491 return ERR_CAST(ctx);
492
493 /* wait the page to be moved by cleaning */
494 f2fs_wait_on_block_writeback(sbi, blkaddr);
495 }
496
497 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
498 if (!bio) {
499 if (ctx)
500 fscrypt_release_ctx(ctx);
501 return ERR_PTR(-ENOMEM);
502 }
503 f2fs_target_device(sbi, blkaddr, bio);
504 bio->bi_end_io = f2fs_read_end_io;
505 bio->bi_private = ctx;
506 bio_set_op_attrs(bio, REQ_OP_READ, 0);
507
508 return bio;
509 }
510
511 /* This can handle encryption stuffs */
512 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
513 block_t blkaddr)
514 {
515 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1);
516
517 if (IS_ERR(bio))
518 return PTR_ERR(bio);
519
520 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
521 bio_put(bio);
522 return -EFAULT;
523 }
524 __submit_bio(F2FS_I_SB(inode), bio, DATA);
525 return 0;
526 }
527
528 static void __set_data_blkaddr(struct dnode_of_data *dn)
529 {
530 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
531 __le32 *addr_array;
532 int base = 0;
533
534 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
535 base = get_extra_isize(dn->inode);
536
537 /* Get physical address of data block */
538 addr_array = blkaddr_in_node(rn);
539 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
540 }
541
542 /*
543 * Lock ordering for the change of data block address:
544 * ->data_page
545 * ->node_page
546 * update block addresses in the node page
547 */
548 void set_data_blkaddr(struct dnode_of_data *dn)
549 {
550 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
551 __set_data_blkaddr(dn);
552 if (set_page_dirty(dn->node_page))
553 dn->node_changed = true;
554 }
555
556 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
557 {
558 dn->data_blkaddr = blkaddr;
559 set_data_blkaddr(dn);
560 f2fs_update_extent_cache(dn);
561 }
562
563 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
564 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
565 {
566 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
567 int err;
568
569 if (!count)
570 return 0;
571
572 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
573 return -EPERM;
574 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
575 return err;
576
577 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
578 dn->ofs_in_node, count);
579
580 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
581
582 for (; count > 0; dn->ofs_in_node++) {
583 block_t blkaddr = datablock_addr(dn->inode,
584 dn->node_page, dn->ofs_in_node);
585 if (blkaddr == NULL_ADDR) {
586 dn->data_blkaddr = NEW_ADDR;
587 __set_data_blkaddr(dn);
588 count--;
589 }
590 }
591
592 if (set_page_dirty(dn->node_page))
593 dn->node_changed = true;
594 return 0;
595 }
596
597 /* Should keep dn->ofs_in_node unchanged */
598 int reserve_new_block(struct dnode_of_data *dn)
599 {
600 unsigned int ofs_in_node = dn->ofs_in_node;
601 int ret;
602
603 ret = reserve_new_blocks(dn, 1);
604 dn->ofs_in_node = ofs_in_node;
605 return ret;
606 }
607
608 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
609 {
610 bool need_put = dn->inode_page ? false : true;
611 int err;
612
613 err = get_dnode_of_data(dn, index, ALLOC_NODE);
614 if (err)
615 return err;
616
617 if (dn->data_blkaddr == NULL_ADDR)
618 err = reserve_new_block(dn);
619 if (err || need_put)
620 f2fs_put_dnode(dn);
621 return err;
622 }
623
624 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
625 {
626 struct extent_info ei = {0,0,0};
627 struct inode *inode = dn->inode;
628
629 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
630 dn->data_blkaddr = ei.blk + index - ei.fofs;
631 return 0;
632 }
633
634 return f2fs_reserve_block(dn, index);
635 }
636
637 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
638 int op_flags, bool for_write)
639 {
640 struct address_space *mapping = inode->i_mapping;
641 struct dnode_of_data dn;
642 struct page *page;
643 struct extent_info ei = {0,0,0};
644 int err;
645
646 page = f2fs_grab_cache_page(mapping, index, for_write);
647 if (!page)
648 return ERR_PTR(-ENOMEM);
649
650 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
651 dn.data_blkaddr = ei.blk + index - ei.fofs;
652 goto got_it;
653 }
654
655 set_new_dnode(&dn, inode, NULL, NULL, 0);
656 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
657 if (err)
658 goto put_err;
659 f2fs_put_dnode(&dn);
660
661 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
662 err = -ENOENT;
663 goto put_err;
664 }
665 got_it:
666 if (PageUptodate(page)) {
667 unlock_page(page);
668 return page;
669 }
670
671 /*
672 * A new dentry page is allocated but not able to be written, since its
673 * new inode page couldn't be allocated due to -ENOSPC.
674 * In such the case, its blkaddr can be remained as NEW_ADDR.
675 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
676 */
677 if (dn.data_blkaddr == NEW_ADDR) {
678 zero_user_segment(page, 0, PAGE_SIZE);
679 if (!PageUptodate(page))
680 SetPageUptodate(page);
681 unlock_page(page);
682 return page;
683 }
684
685 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
686 if (err)
687 goto put_err;
688 return page;
689
690 put_err:
691 f2fs_put_page(page, 1);
692 return ERR_PTR(err);
693 }
694
695 struct page *find_data_page(struct inode *inode, pgoff_t index)
696 {
697 struct address_space *mapping = inode->i_mapping;
698 struct page *page;
699
700 page = find_get_page(mapping, index);
701 if (page && PageUptodate(page))
702 return page;
703 f2fs_put_page(page, 0);
704
705 page = get_read_data_page(inode, index, 0, false);
706 if (IS_ERR(page))
707 return page;
708
709 if (PageUptodate(page))
710 return page;
711
712 wait_on_page_locked(page);
713 if (unlikely(!PageUptodate(page))) {
714 f2fs_put_page(page, 0);
715 return ERR_PTR(-EIO);
716 }
717 return page;
718 }
719
720 /*
721 * If it tries to access a hole, return an error.
722 * Because, the callers, functions in dir.c and GC, should be able to know
723 * whether this page exists or not.
724 */
725 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
726 bool for_write)
727 {
728 struct address_space *mapping = inode->i_mapping;
729 struct page *page;
730 repeat:
731 page = get_read_data_page(inode, index, 0, for_write);
732 if (IS_ERR(page))
733 return page;
734
735 /* wait for read completion */
736 lock_page(page);
737 if (unlikely(page->mapping != mapping)) {
738 f2fs_put_page(page, 1);
739 goto repeat;
740 }
741 if (unlikely(!PageUptodate(page))) {
742 f2fs_put_page(page, 1);
743 return ERR_PTR(-EIO);
744 }
745 return page;
746 }
747
748 /*
749 * Caller ensures that this data page is never allocated.
750 * A new zero-filled data page is allocated in the page cache.
751 *
752 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
753 * f2fs_unlock_op().
754 * Note that, ipage is set only by make_empty_dir, and if any error occur,
755 * ipage should be released by this function.
756 */
757 struct page *get_new_data_page(struct inode *inode,
758 struct page *ipage, pgoff_t index, bool new_i_size)
759 {
760 struct address_space *mapping = inode->i_mapping;
761 struct page *page;
762 struct dnode_of_data dn;
763 int err;
764
765 page = f2fs_grab_cache_page(mapping, index, true);
766 if (!page) {
767 /*
768 * before exiting, we should make sure ipage will be released
769 * if any error occur.
770 */
771 f2fs_put_page(ipage, 1);
772 return ERR_PTR(-ENOMEM);
773 }
774
775 set_new_dnode(&dn, inode, ipage, NULL, 0);
776 err = f2fs_reserve_block(&dn, index);
777 if (err) {
778 f2fs_put_page(page, 1);
779 return ERR_PTR(err);
780 }
781 if (!ipage)
782 f2fs_put_dnode(&dn);
783
784 if (PageUptodate(page))
785 goto got_it;
786
787 if (dn.data_blkaddr == NEW_ADDR) {
788 zero_user_segment(page, 0, PAGE_SIZE);
789 if (!PageUptodate(page))
790 SetPageUptodate(page);
791 } else {
792 f2fs_put_page(page, 1);
793
794 /* if ipage exists, blkaddr should be NEW_ADDR */
795 f2fs_bug_on(F2FS_I_SB(inode), ipage);
796 page = get_lock_data_page(inode, index, true);
797 if (IS_ERR(page))
798 return page;
799 }
800 got_it:
801 if (new_i_size && i_size_read(inode) <
802 ((loff_t)(index + 1) << PAGE_SHIFT))
803 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
804 return page;
805 }
806
807 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
808 {
809 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
810 struct f2fs_summary sum;
811 struct node_info ni;
812 pgoff_t fofs;
813 blkcnt_t count = 1;
814 int err;
815
816 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
817 return -EPERM;
818
819 dn->data_blkaddr = datablock_addr(dn->inode,
820 dn->node_page, dn->ofs_in_node);
821 if (dn->data_blkaddr == NEW_ADDR)
822 goto alloc;
823
824 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
825 return err;
826
827 alloc:
828 get_node_info(sbi, dn->nid, &ni);
829 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
830
831 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
832 &sum, seg_type, NULL, false);
833 set_data_blkaddr(dn);
834
835 /* update i_size */
836 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
837 dn->ofs_in_node;
838 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
839 f2fs_i_size_write(dn->inode,
840 ((loff_t)(fofs + 1) << PAGE_SHIFT));
841 return 0;
842 }
843
844 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
845 {
846 struct inode *inode = file_inode(iocb->ki_filp);
847 struct f2fs_map_blocks map;
848 int flag;
849 int err = 0;
850 bool direct_io = iocb->ki_flags & IOCB_DIRECT;
851
852 /* convert inline data for Direct I/O*/
853 if (direct_io) {
854 err = f2fs_convert_inline_inode(inode);
855 if (err)
856 return err;
857 }
858
859 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
860 return 0;
861
862 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
863 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
864 if (map.m_len > map.m_lblk)
865 map.m_len -= map.m_lblk;
866 else
867 map.m_len = 0;
868
869 map.m_next_pgofs = NULL;
870 map.m_next_extent = NULL;
871 map.m_seg_type = NO_CHECK_TYPE;
872
873 if (direct_io) {
874 map.m_seg_type = rw_hint_to_seg_type(iocb->ki_hint);
875 flag = f2fs_force_buffered_io(inode, WRITE) ?
876 F2FS_GET_BLOCK_PRE_AIO :
877 F2FS_GET_BLOCK_PRE_DIO;
878 goto map_blocks;
879 }
880 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
881 err = f2fs_convert_inline_inode(inode);
882 if (err)
883 return err;
884 }
885 if (f2fs_has_inline_data(inode))
886 return err;
887
888 flag = F2FS_GET_BLOCK_PRE_AIO;
889
890 map_blocks:
891 err = f2fs_map_blocks(inode, &map, 1, flag);
892 if (map.m_len > 0 && err == -ENOSPC) {
893 if (!direct_io)
894 set_inode_flag(inode, FI_NO_PREALLOC);
895 err = 0;
896 }
897 return err;
898 }
899
900 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
901 {
902 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
903 if (lock)
904 down_read(&sbi->node_change);
905 else
906 up_read(&sbi->node_change);
907 } else {
908 if (lock)
909 f2fs_lock_op(sbi);
910 else
911 f2fs_unlock_op(sbi);
912 }
913 }
914
915 /*
916 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
917 * f2fs_map_blocks structure.
918 * If original data blocks are allocated, then give them to blockdev.
919 * Otherwise,
920 * a. preallocate requested block addresses
921 * b. do not use extent cache for better performance
922 * c. give the block addresses to blockdev
923 */
924 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
925 int create, int flag)
926 {
927 unsigned int maxblocks = map->m_len;
928 struct dnode_of_data dn;
929 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
930 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
931 pgoff_t pgofs, end_offset, end;
932 int err = 0, ofs = 1;
933 unsigned int ofs_in_node, last_ofs_in_node;
934 blkcnt_t prealloc;
935 struct extent_info ei = {0,0,0};
936 block_t blkaddr;
937 unsigned int start_pgofs;
938
939 if (!maxblocks)
940 return 0;
941
942 map->m_len = 0;
943 map->m_flags = 0;
944
945 /* it only supports block size == page size */
946 pgofs = (pgoff_t)map->m_lblk;
947 end = pgofs + maxblocks;
948
949 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
950 map->m_pblk = ei.blk + pgofs - ei.fofs;
951 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
952 map->m_flags = F2FS_MAP_MAPPED;
953 if (map->m_next_extent)
954 *map->m_next_extent = pgofs + map->m_len;
955 goto out;
956 }
957
958 next_dnode:
959 if (create)
960 __do_map_lock(sbi, flag, true);
961
962 /* When reading holes, we need its node page */
963 set_new_dnode(&dn, inode, NULL, NULL, 0);
964 err = get_dnode_of_data(&dn, pgofs, mode);
965 if (err) {
966 if (flag == F2FS_GET_BLOCK_BMAP)
967 map->m_pblk = 0;
968 if (err == -ENOENT) {
969 err = 0;
970 if (map->m_next_pgofs)
971 *map->m_next_pgofs =
972 get_next_page_offset(&dn, pgofs);
973 if (map->m_next_extent)
974 *map->m_next_extent =
975 get_next_page_offset(&dn, pgofs);
976 }
977 goto unlock_out;
978 }
979
980 start_pgofs = pgofs;
981 prealloc = 0;
982 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
983 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
984
985 next_block:
986 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
987
988 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
989 if (create) {
990 if (unlikely(f2fs_cp_error(sbi))) {
991 err = -EIO;
992 goto sync_out;
993 }
994 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
995 if (blkaddr == NULL_ADDR) {
996 prealloc++;
997 last_ofs_in_node = dn.ofs_in_node;
998 }
999 } else {
1000 err = __allocate_data_block(&dn,
1001 map->m_seg_type);
1002 if (!err)
1003 set_inode_flag(inode, FI_APPEND_WRITE);
1004 }
1005 if (err)
1006 goto sync_out;
1007 map->m_flags |= F2FS_MAP_NEW;
1008 blkaddr = dn.data_blkaddr;
1009 } else {
1010 if (flag == F2FS_GET_BLOCK_BMAP) {
1011 map->m_pblk = 0;
1012 goto sync_out;
1013 }
1014 if (flag == F2FS_GET_BLOCK_PRECACHE)
1015 goto sync_out;
1016 if (flag == F2FS_GET_BLOCK_FIEMAP &&
1017 blkaddr == NULL_ADDR) {
1018 if (map->m_next_pgofs)
1019 *map->m_next_pgofs = pgofs + 1;
1020 goto sync_out;
1021 }
1022 if (flag != F2FS_GET_BLOCK_FIEMAP) {
1023 /* for defragment case */
1024 if (map->m_next_pgofs)
1025 *map->m_next_pgofs = pgofs + 1;
1026 goto sync_out;
1027 }
1028 }
1029 }
1030
1031 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1032 goto skip;
1033
1034 if (map->m_len == 0) {
1035 /* preallocated unwritten block should be mapped for fiemap. */
1036 if (blkaddr == NEW_ADDR)
1037 map->m_flags |= F2FS_MAP_UNWRITTEN;
1038 map->m_flags |= F2FS_MAP_MAPPED;
1039
1040 map->m_pblk = blkaddr;
1041 map->m_len = 1;
1042 } else if ((map->m_pblk != NEW_ADDR &&
1043 blkaddr == (map->m_pblk + ofs)) ||
1044 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1045 flag == F2FS_GET_BLOCK_PRE_DIO) {
1046 ofs++;
1047 map->m_len++;
1048 } else {
1049 goto sync_out;
1050 }
1051
1052 skip:
1053 dn.ofs_in_node++;
1054 pgofs++;
1055
1056 /* preallocate blocks in batch for one dnode page */
1057 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1058 (pgofs == end || dn.ofs_in_node == end_offset)) {
1059
1060 dn.ofs_in_node = ofs_in_node;
1061 err = reserve_new_blocks(&dn, prealloc);
1062 if (err)
1063 goto sync_out;
1064
1065 map->m_len += dn.ofs_in_node - ofs_in_node;
1066 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1067 err = -ENOSPC;
1068 goto sync_out;
1069 }
1070 dn.ofs_in_node = end_offset;
1071 }
1072
1073 if (pgofs >= end)
1074 goto sync_out;
1075 else if (dn.ofs_in_node < end_offset)
1076 goto next_block;
1077
1078 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1079 if (map->m_flags & F2FS_MAP_MAPPED) {
1080 unsigned int ofs = start_pgofs - map->m_lblk;
1081
1082 f2fs_update_extent_cache_range(&dn,
1083 start_pgofs, map->m_pblk + ofs,
1084 map->m_len - ofs);
1085 }
1086 }
1087
1088 f2fs_put_dnode(&dn);
1089
1090 if (create) {
1091 __do_map_lock(sbi, flag, false);
1092 f2fs_balance_fs(sbi, dn.node_changed);
1093 }
1094 goto next_dnode;
1095
1096 sync_out:
1097 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1098 if (map->m_flags & F2FS_MAP_MAPPED) {
1099 unsigned int ofs = start_pgofs - map->m_lblk;
1100
1101 f2fs_update_extent_cache_range(&dn,
1102 start_pgofs, map->m_pblk + ofs,
1103 map->m_len - ofs);
1104 }
1105 if (map->m_next_extent)
1106 *map->m_next_extent = pgofs + 1;
1107 }
1108 f2fs_put_dnode(&dn);
1109 unlock_out:
1110 if (create) {
1111 __do_map_lock(sbi, flag, false);
1112 f2fs_balance_fs(sbi, dn.node_changed);
1113 }
1114 out:
1115 trace_f2fs_map_blocks(inode, map, err);
1116 return err;
1117 }
1118
1119 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1120 {
1121 struct f2fs_map_blocks map;
1122 block_t last_lblk;
1123 int err;
1124
1125 if (pos + len > i_size_read(inode))
1126 return false;
1127
1128 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1129 map.m_next_pgofs = NULL;
1130 map.m_next_extent = NULL;
1131 map.m_seg_type = NO_CHECK_TYPE;
1132 last_lblk = F2FS_BLK_ALIGN(pos + len);
1133
1134 while (map.m_lblk < last_lblk) {
1135 map.m_len = last_lblk - map.m_lblk;
1136 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1137 if (err || map.m_len == 0)
1138 return false;
1139 map.m_lblk += map.m_len;
1140 }
1141 return true;
1142 }
1143
1144 static int __get_data_block(struct inode *inode, sector_t iblock,
1145 struct buffer_head *bh, int create, int flag,
1146 pgoff_t *next_pgofs, int seg_type)
1147 {
1148 struct f2fs_map_blocks map;
1149 int err;
1150
1151 map.m_lblk = iblock;
1152 map.m_len = bh->b_size >> inode->i_blkbits;
1153 map.m_next_pgofs = next_pgofs;
1154 map.m_next_extent = NULL;
1155 map.m_seg_type = seg_type;
1156
1157 err = f2fs_map_blocks(inode, &map, create, flag);
1158 if (!err) {
1159 map_bh(bh, inode->i_sb, map.m_pblk);
1160 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1161 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1162 }
1163 return err;
1164 }
1165
1166 static int get_data_block(struct inode *inode, sector_t iblock,
1167 struct buffer_head *bh_result, int create, int flag,
1168 pgoff_t *next_pgofs)
1169 {
1170 return __get_data_block(inode, iblock, bh_result, create,
1171 flag, next_pgofs,
1172 NO_CHECK_TYPE);
1173 }
1174
1175 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1176 struct buffer_head *bh_result, int create)
1177 {
1178 return __get_data_block(inode, iblock, bh_result, create,
1179 F2FS_GET_BLOCK_DEFAULT, NULL,
1180 rw_hint_to_seg_type(
1181 inode->i_write_hint));
1182 }
1183
1184 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1185 struct buffer_head *bh_result, int create)
1186 {
1187 /* Block number less than F2FS MAX BLOCKS */
1188 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1189 return -EFBIG;
1190
1191 return __get_data_block(inode, iblock, bh_result, create,
1192 F2FS_GET_BLOCK_BMAP, NULL,
1193 NO_CHECK_TYPE);
1194 }
1195
1196 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1197 {
1198 return (offset >> inode->i_blkbits);
1199 }
1200
1201 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1202 {
1203 return (blk << inode->i_blkbits);
1204 }
1205
1206 static int f2fs_xattr_fiemap(struct inode *inode,
1207 struct fiemap_extent_info *fieinfo)
1208 {
1209 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1210 struct page *page;
1211 struct node_info ni;
1212 __u64 phys = 0, len;
1213 __u32 flags;
1214 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1215 int err = 0;
1216
1217 if (f2fs_has_inline_xattr(inode)) {
1218 int offset;
1219
1220 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1221 inode->i_ino, false);
1222 if (!page)
1223 return -ENOMEM;
1224
1225 get_node_info(sbi, inode->i_ino, &ni);
1226
1227 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1228 offset = offsetof(struct f2fs_inode, i_addr) +
1229 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1230 get_inline_xattr_addrs(inode));
1231
1232 phys += offset;
1233 len = inline_xattr_size(inode);
1234
1235 f2fs_put_page(page, 1);
1236
1237 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1238
1239 if (!xnid)
1240 flags |= FIEMAP_EXTENT_LAST;
1241
1242 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1243 if (err || err == 1)
1244 return err;
1245 }
1246
1247 if (xnid) {
1248 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1249 if (!page)
1250 return -ENOMEM;
1251
1252 get_node_info(sbi, xnid, &ni);
1253
1254 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1255 len = inode->i_sb->s_blocksize;
1256
1257 f2fs_put_page(page, 1);
1258
1259 flags = FIEMAP_EXTENT_LAST;
1260 }
1261
1262 if (phys)
1263 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1264
1265 return (err < 0 ? err : 0);
1266 }
1267
1268 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1269 u64 start, u64 len)
1270 {
1271 struct buffer_head map_bh;
1272 sector_t start_blk, last_blk;
1273 pgoff_t next_pgofs;
1274 u64 logical = 0, phys = 0, size = 0;
1275 u32 flags = 0;
1276 int ret = 0;
1277
1278 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1279 ret = f2fs_precache_extents(inode);
1280 if (ret)
1281 return ret;
1282 }
1283
1284 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1285 if (ret)
1286 return ret;
1287
1288 inode_lock(inode);
1289
1290 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1291 ret = f2fs_xattr_fiemap(inode, fieinfo);
1292 goto out;
1293 }
1294
1295 if (f2fs_has_inline_data(inode)) {
1296 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1297 if (ret != -EAGAIN)
1298 goto out;
1299 }
1300
1301 if (logical_to_blk(inode, len) == 0)
1302 len = blk_to_logical(inode, 1);
1303
1304 start_blk = logical_to_blk(inode, start);
1305 last_blk = logical_to_blk(inode, start + len - 1);
1306
1307 next:
1308 memset(&map_bh, 0, sizeof(struct buffer_head));
1309 map_bh.b_size = len;
1310
1311 ret = get_data_block(inode, start_blk, &map_bh, 0,
1312 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1313 if (ret)
1314 goto out;
1315
1316 /* HOLE */
1317 if (!buffer_mapped(&map_bh)) {
1318 start_blk = next_pgofs;
1319
1320 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1321 F2FS_I_SB(inode)->max_file_blocks))
1322 goto prep_next;
1323
1324 flags |= FIEMAP_EXTENT_LAST;
1325 }
1326
1327 if (size) {
1328 if (f2fs_encrypted_inode(inode))
1329 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1330
1331 ret = fiemap_fill_next_extent(fieinfo, logical,
1332 phys, size, flags);
1333 }
1334
1335 if (start_blk > last_blk || ret)
1336 goto out;
1337
1338 logical = blk_to_logical(inode, start_blk);
1339 phys = blk_to_logical(inode, map_bh.b_blocknr);
1340 size = map_bh.b_size;
1341 flags = 0;
1342 if (buffer_unwritten(&map_bh))
1343 flags = FIEMAP_EXTENT_UNWRITTEN;
1344
1345 start_blk += logical_to_blk(inode, size);
1346
1347 prep_next:
1348 cond_resched();
1349 if (fatal_signal_pending(current))
1350 ret = -EINTR;
1351 else
1352 goto next;
1353 out:
1354 if (ret == 1)
1355 ret = 0;
1356
1357 inode_unlock(inode);
1358 return ret;
1359 }
1360
1361 /*
1362 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1363 * Major change was from block_size == page_size in f2fs by default.
1364 */
1365 static int f2fs_mpage_readpages(struct address_space *mapping,
1366 struct list_head *pages, struct page *page,
1367 unsigned nr_pages)
1368 {
1369 struct bio *bio = NULL;
1370 sector_t last_block_in_bio = 0;
1371 struct inode *inode = mapping->host;
1372 const unsigned blkbits = inode->i_blkbits;
1373 const unsigned blocksize = 1 << blkbits;
1374 sector_t block_in_file;
1375 sector_t last_block;
1376 sector_t last_block_in_file;
1377 sector_t block_nr;
1378 struct f2fs_map_blocks map;
1379
1380 map.m_pblk = 0;
1381 map.m_lblk = 0;
1382 map.m_len = 0;
1383 map.m_flags = 0;
1384 map.m_next_pgofs = NULL;
1385 map.m_next_extent = NULL;
1386 map.m_seg_type = NO_CHECK_TYPE;
1387
1388 for (; nr_pages; nr_pages--) {
1389 if (pages) {
1390 page = list_last_entry(pages, struct page, lru);
1391
1392 prefetchw(&page->flags);
1393 list_del(&page->lru);
1394 if (add_to_page_cache_lru(page, mapping,
1395 page->index,
1396 readahead_gfp_mask(mapping)))
1397 goto next_page;
1398 }
1399
1400 block_in_file = (sector_t)page->index;
1401 last_block = block_in_file + nr_pages;
1402 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1403 blkbits;
1404 if (last_block > last_block_in_file)
1405 last_block = last_block_in_file;
1406
1407 /*
1408 * Map blocks using the previous result first.
1409 */
1410 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1411 block_in_file > map.m_lblk &&
1412 block_in_file < (map.m_lblk + map.m_len))
1413 goto got_it;
1414
1415 /*
1416 * Then do more f2fs_map_blocks() calls until we are
1417 * done with this page.
1418 */
1419 map.m_flags = 0;
1420
1421 if (block_in_file < last_block) {
1422 map.m_lblk = block_in_file;
1423 map.m_len = last_block - block_in_file;
1424
1425 if (f2fs_map_blocks(inode, &map, 0,
1426 F2FS_GET_BLOCK_DEFAULT))
1427 goto set_error_page;
1428 }
1429 got_it:
1430 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1431 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1432 SetPageMappedToDisk(page);
1433
1434 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1435 SetPageUptodate(page);
1436 goto confused;
1437 }
1438 } else {
1439 zero_user_segment(page, 0, PAGE_SIZE);
1440 if (!PageUptodate(page))
1441 SetPageUptodate(page);
1442 unlock_page(page);
1443 goto next_page;
1444 }
1445
1446 /*
1447 * This page will go to BIO. Do we need to send this
1448 * BIO off first?
1449 */
1450 if (bio && (last_block_in_bio != block_nr - 1 ||
1451 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1452 submit_and_realloc:
1453 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1454 bio = NULL;
1455 }
1456 if (bio == NULL) {
1457 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages);
1458 if (IS_ERR(bio)) {
1459 bio = NULL;
1460 goto set_error_page;
1461 }
1462 }
1463
1464 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1465 goto submit_and_realloc;
1466
1467 last_block_in_bio = block_nr;
1468 goto next_page;
1469 set_error_page:
1470 SetPageError(page);
1471 zero_user_segment(page, 0, PAGE_SIZE);
1472 unlock_page(page);
1473 goto next_page;
1474 confused:
1475 if (bio) {
1476 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1477 bio = NULL;
1478 }
1479 unlock_page(page);
1480 next_page:
1481 if (pages)
1482 put_page(page);
1483 }
1484 BUG_ON(pages && !list_empty(pages));
1485 if (bio)
1486 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1487 return 0;
1488 }
1489
1490 static int f2fs_read_data_page(struct file *file, struct page *page)
1491 {
1492 struct inode *inode = page->mapping->host;
1493 int ret = -EAGAIN;
1494
1495 trace_f2fs_readpage(page, DATA);
1496
1497 /* If the file has inline data, try to read it directly */
1498 if (f2fs_has_inline_data(inode))
1499 ret = f2fs_read_inline_data(inode, page);
1500 if (ret == -EAGAIN)
1501 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1502 return ret;
1503 }
1504
1505 static int f2fs_read_data_pages(struct file *file,
1506 struct address_space *mapping,
1507 struct list_head *pages, unsigned nr_pages)
1508 {
1509 struct inode *inode = mapping->host;
1510 struct page *page = list_last_entry(pages, struct page, lru);
1511
1512 trace_f2fs_readpages(inode, page, nr_pages);
1513
1514 /* If the file has inline data, skip readpages */
1515 if (f2fs_has_inline_data(inode))
1516 return 0;
1517
1518 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1519 }
1520
1521 static int encrypt_one_page(struct f2fs_io_info *fio)
1522 {
1523 struct inode *inode = fio->page->mapping->host;
1524 gfp_t gfp_flags = GFP_NOFS;
1525
1526 if (!f2fs_encrypted_file(inode))
1527 return 0;
1528
1529 /* wait for GCed encrypted page writeback */
1530 f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1531
1532 retry_encrypt:
1533 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1534 PAGE_SIZE, 0, fio->page->index, gfp_flags);
1535 if (!IS_ERR(fio->encrypted_page))
1536 return 0;
1537
1538 /* flush pending IOs and wait for a while in the ENOMEM case */
1539 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1540 f2fs_flush_merged_writes(fio->sbi);
1541 congestion_wait(BLK_RW_ASYNC, HZ/50);
1542 gfp_flags |= __GFP_NOFAIL;
1543 goto retry_encrypt;
1544 }
1545 return PTR_ERR(fio->encrypted_page);
1546 }
1547
1548 static inline bool check_inplace_update_policy(struct inode *inode,
1549 struct f2fs_io_info *fio)
1550 {
1551 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1552 unsigned int policy = SM_I(sbi)->ipu_policy;
1553
1554 if (policy & (0x1 << F2FS_IPU_FORCE))
1555 return true;
1556 if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
1557 return true;
1558 if (policy & (0x1 << F2FS_IPU_UTIL) &&
1559 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1560 return true;
1561 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
1562 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1563 return true;
1564
1565 /*
1566 * IPU for rewrite async pages
1567 */
1568 if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1569 fio && fio->op == REQ_OP_WRITE &&
1570 !(fio->op_flags & REQ_SYNC) &&
1571 !f2fs_encrypted_inode(inode))
1572 return true;
1573
1574 /* this is only set during fdatasync */
1575 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1576 is_inode_flag_set(inode, FI_NEED_IPU))
1577 return true;
1578
1579 return false;
1580 }
1581
1582 bool should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1583 {
1584 if (f2fs_is_pinned_file(inode))
1585 return true;
1586
1587 /* if this is cold file, we should overwrite to avoid fragmentation */
1588 if (file_is_cold(inode))
1589 return true;
1590
1591 return check_inplace_update_policy(inode, fio);
1592 }
1593
1594 bool should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1595 {
1596 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1597
1598 if (test_opt(sbi, LFS))
1599 return true;
1600 if (S_ISDIR(inode->i_mode))
1601 return true;
1602 if (f2fs_is_atomic_file(inode))
1603 return true;
1604 if (fio) {
1605 if (is_cold_data(fio->page))
1606 return true;
1607 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1608 return true;
1609 }
1610 return false;
1611 }
1612
1613 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1614 {
1615 struct inode *inode = fio->page->mapping->host;
1616
1617 if (should_update_outplace(inode, fio))
1618 return false;
1619
1620 return should_update_inplace(inode, fio);
1621 }
1622
1623 static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio)
1624 {
1625 if (fio->old_blkaddr == NEW_ADDR)
1626 return false;
1627 if (fio->old_blkaddr == NULL_ADDR)
1628 return false;
1629 return true;
1630 }
1631
1632 int do_write_data_page(struct f2fs_io_info *fio)
1633 {
1634 struct page *page = fio->page;
1635 struct inode *inode = page->mapping->host;
1636 struct dnode_of_data dn;
1637 struct extent_info ei = {0,0,0};
1638 bool ipu_force = false;
1639 int err = 0;
1640
1641 set_new_dnode(&dn, inode, NULL, NULL, 0);
1642 if (need_inplace_update(fio) &&
1643 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1644 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1645
1646 if (valid_ipu_blkaddr(fio)) {
1647 ipu_force = true;
1648 fio->need_lock = LOCK_DONE;
1649 goto got_it;
1650 }
1651 }
1652
1653 /* Deadlock due to between page->lock and f2fs_lock_op */
1654 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1655 return -EAGAIN;
1656
1657 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1658 if (err)
1659 goto out;
1660
1661 fio->old_blkaddr = dn.data_blkaddr;
1662
1663 /* This page is already truncated */
1664 if (fio->old_blkaddr == NULL_ADDR) {
1665 ClearPageUptodate(page);
1666 goto out_writepage;
1667 }
1668 got_it:
1669 /*
1670 * If current allocation needs SSR,
1671 * it had better in-place writes for updated data.
1672 */
1673 if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) {
1674 err = encrypt_one_page(fio);
1675 if (err)
1676 goto out_writepage;
1677
1678 set_page_writeback(page);
1679 f2fs_put_dnode(&dn);
1680 if (fio->need_lock == LOCK_REQ)
1681 f2fs_unlock_op(fio->sbi);
1682 err = rewrite_data_page(fio);
1683 trace_f2fs_do_write_data_page(fio->page, IPU);
1684 set_inode_flag(inode, FI_UPDATE_WRITE);
1685 return err;
1686 }
1687
1688 if (fio->need_lock == LOCK_RETRY) {
1689 if (!f2fs_trylock_op(fio->sbi)) {
1690 err = -EAGAIN;
1691 goto out_writepage;
1692 }
1693 fio->need_lock = LOCK_REQ;
1694 }
1695
1696 err = encrypt_one_page(fio);
1697 if (err)
1698 goto out_writepage;
1699
1700 set_page_writeback(page);
1701
1702 /* LFS mode write path */
1703 write_data_page(&dn, fio);
1704 trace_f2fs_do_write_data_page(page, OPU);
1705 set_inode_flag(inode, FI_APPEND_WRITE);
1706 if (page->index == 0)
1707 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1708 out_writepage:
1709 f2fs_put_dnode(&dn);
1710 out:
1711 if (fio->need_lock == LOCK_REQ)
1712 f2fs_unlock_op(fio->sbi);
1713 return err;
1714 }
1715
1716 static int __write_data_page(struct page *page, bool *submitted,
1717 struct writeback_control *wbc,
1718 enum iostat_type io_type)
1719 {
1720 struct inode *inode = page->mapping->host;
1721 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1722 loff_t i_size = i_size_read(inode);
1723 const pgoff_t end_index = ((unsigned long long) i_size)
1724 >> PAGE_SHIFT;
1725 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1726 unsigned offset = 0;
1727 bool need_balance_fs = false;
1728 int err = 0;
1729 struct f2fs_io_info fio = {
1730 .sbi = sbi,
1731 .ino = inode->i_ino,
1732 .type = DATA,
1733 .op = REQ_OP_WRITE,
1734 .op_flags = wbc_to_write_flags(wbc),
1735 .old_blkaddr = NULL_ADDR,
1736 .page = page,
1737 .encrypted_page = NULL,
1738 .submitted = false,
1739 .need_lock = LOCK_RETRY,
1740 .io_type = io_type,
1741 .io_wbc = wbc,
1742 };
1743
1744 trace_f2fs_writepage(page, DATA);
1745
1746 /* we should bypass data pages to proceed the kworkder jobs */
1747 if (unlikely(f2fs_cp_error(sbi))) {
1748 mapping_set_error(page->mapping, -EIO);
1749 goto out;
1750 }
1751
1752 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1753 goto redirty_out;
1754
1755 if (page->index < end_index)
1756 goto write;
1757
1758 /*
1759 * If the offset is out-of-range of file size,
1760 * this page does not have to be written to disk.
1761 */
1762 offset = i_size & (PAGE_SIZE - 1);
1763 if ((page->index >= end_index + 1) || !offset)
1764 goto out;
1765
1766 zero_user_segment(page, offset, PAGE_SIZE);
1767 write:
1768 if (f2fs_is_drop_cache(inode))
1769 goto out;
1770 /* we should not write 0'th page having journal header */
1771 if (f2fs_is_volatile_file(inode) && (!page->index ||
1772 (!wbc->for_reclaim &&
1773 available_free_memory(sbi, BASE_CHECK))))
1774 goto redirty_out;
1775
1776 /* Dentry blocks are controlled by checkpoint */
1777 if (S_ISDIR(inode->i_mode)) {
1778 fio.need_lock = LOCK_DONE;
1779 err = do_write_data_page(&fio);
1780 goto done;
1781 }
1782
1783 if (!wbc->for_reclaim)
1784 need_balance_fs = true;
1785 else if (has_not_enough_free_secs(sbi, 0, 0))
1786 goto redirty_out;
1787 else
1788 set_inode_flag(inode, FI_HOT_DATA);
1789
1790 err = -EAGAIN;
1791 if (f2fs_has_inline_data(inode)) {
1792 err = f2fs_write_inline_data(inode, page);
1793 if (!err)
1794 goto out;
1795 }
1796
1797 if (err == -EAGAIN) {
1798 err = do_write_data_page(&fio);
1799 if (err == -EAGAIN) {
1800 fio.need_lock = LOCK_REQ;
1801 err = do_write_data_page(&fio);
1802 }
1803 }
1804
1805 if (err) {
1806 file_set_keep_isize(inode);
1807 } else {
1808 down_write(&F2FS_I(inode)->i_sem);
1809 if (F2FS_I(inode)->last_disk_size < psize)
1810 F2FS_I(inode)->last_disk_size = psize;
1811 up_write(&F2FS_I(inode)->i_sem);
1812 }
1813
1814 done:
1815 if (err && err != -ENOENT)
1816 goto redirty_out;
1817
1818 out:
1819 inode_dec_dirty_pages(inode);
1820 if (err)
1821 ClearPageUptodate(page);
1822
1823 if (wbc->for_reclaim) {
1824 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1825 clear_inode_flag(inode, FI_HOT_DATA);
1826 remove_dirty_inode(inode);
1827 submitted = NULL;
1828 }
1829
1830 unlock_page(page);
1831 if (!S_ISDIR(inode->i_mode))
1832 f2fs_balance_fs(sbi, need_balance_fs);
1833
1834 if (unlikely(f2fs_cp_error(sbi))) {
1835 f2fs_submit_merged_write(sbi, DATA);
1836 submitted = NULL;
1837 }
1838
1839 if (submitted)
1840 *submitted = fio.submitted;
1841
1842 return 0;
1843
1844 redirty_out:
1845 redirty_page_for_writepage(wbc, page);
1846 if (!err)
1847 return AOP_WRITEPAGE_ACTIVATE;
1848 unlock_page(page);
1849 return err;
1850 }
1851
1852 static int f2fs_write_data_page(struct page *page,
1853 struct writeback_control *wbc)
1854 {
1855 return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1856 }
1857
1858 /*
1859 * This function was copied from write_cche_pages from mm/page-writeback.c.
1860 * The major change is making write step of cold data page separately from
1861 * warm/hot data page.
1862 */
1863 static int f2fs_write_cache_pages(struct address_space *mapping,
1864 struct writeback_control *wbc,
1865 enum iostat_type io_type)
1866 {
1867 int ret = 0;
1868 int done = 0;
1869 struct pagevec pvec;
1870 int nr_pages;
1871 pgoff_t uninitialized_var(writeback_index);
1872 pgoff_t index;
1873 pgoff_t end; /* Inclusive */
1874 pgoff_t done_index;
1875 pgoff_t last_idx = ULONG_MAX;
1876 int cycled;
1877 int range_whole = 0;
1878 int tag;
1879
1880 pagevec_init(&pvec, 0);
1881
1882 if (get_dirty_pages(mapping->host) <=
1883 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1884 set_inode_flag(mapping->host, FI_HOT_DATA);
1885 else
1886 clear_inode_flag(mapping->host, FI_HOT_DATA);
1887
1888 if (wbc->range_cyclic) {
1889 writeback_index = mapping->writeback_index; /* prev offset */
1890 index = writeback_index;
1891 if (index == 0)
1892 cycled = 1;
1893 else
1894 cycled = 0;
1895 end = -1;
1896 } else {
1897 index = wbc->range_start >> PAGE_SHIFT;
1898 end = wbc->range_end >> PAGE_SHIFT;
1899 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1900 range_whole = 1;
1901 cycled = 1; /* ignore range_cyclic tests */
1902 }
1903 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1904 tag = PAGECACHE_TAG_TOWRITE;
1905 else
1906 tag = PAGECACHE_TAG_DIRTY;
1907 retry:
1908 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1909 tag_pages_for_writeback(mapping, index, end);
1910 done_index = index;
1911 while (!done && (index <= end)) {
1912 int i;
1913
1914 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1915 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1916 if (nr_pages == 0)
1917 break;
1918
1919 for (i = 0; i < nr_pages; i++) {
1920 struct page *page = pvec.pages[i];
1921 bool submitted = false;
1922
1923 if (page->index > end) {
1924 done = 1;
1925 break;
1926 }
1927
1928 done_index = page->index;
1929 retry_write:
1930 lock_page(page);
1931
1932 if (unlikely(page->mapping != mapping)) {
1933 continue_unlock:
1934 unlock_page(page);
1935 continue;
1936 }
1937
1938 if (!PageDirty(page)) {
1939 /* someone wrote it for us */
1940 goto continue_unlock;
1941 }
1942
1943 if (PageWriteback(page)) {
1944 if (wbc->sync_mode != WB_SYNC_NONE)
1945 f2fs_wait_on_page_writeback(page,
1946 DATA, true);
1947 else
1948 goto continue_unlock;
1949 }
1950
1951 BUG_ON(PageWriteback(page));
1952 if (!clear_page_dirty_for_io(page))
1953 goto continue_unlock;
1954
1955 ret = __write_data_page(page, &submitted, wbc, io_type);
1956 if (unlikely(ret)) {
1957 /*
1958 * keep nr_to_write, since vfs uses this to
1959 * get # of written pages.
1960 */
1961 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1962 unlock_page(page);
1963 ret = 0;
1964 continue;
1965 } else if (ret == -EAGAIN) {
1966 ret = 0;
1967 if (wbc->sync_mode == WB_SYNC_ALL) {
1968 cond_resched();
1969 congestion_wait(BLK_RW_ASYNC,
1970 HZ/50);
1971 goto retry_write;
1972 }
1973 continue;
1974 }
1975 done_index = page->index + 1;
1976 done = 1;
1977 break;
1978 } else if (submitted) {
1979 last_idx = page->index;
1980 }
1981
1982 /* give a priority to WB_SYNC threads */
1983 if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) ||
1984 --wbc->nr_to_write <= 0) &&
1985 wbc->sync_mode == WB_SYNC_NONE) {
1986 done = 1;
1987 break;
1988 }
1989 }
1990 pagevec_release(&pvec);
1991 cond_resched();
1992 }
1993
1994 if (!cycled && !done) {
1995 cycled = 1;
1996 index = 0;
1997 end = writeback_index - 1;
1998 goto retry;
1999 }
2000 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2001 mapping->writeback_index = done_index;
2002
2003 if (last_idx != ULONG_MAX)
2004 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2005 0, last_idx, DATA);
2006
2007 return ret;
2008 }
2009
2010 int __f2fs_write_data_pages(struct address_space *mapping,
2011 struct writeback_control *wbc,
2012 enum iostat_type io_type)
2013 {
2014 struct inode *inode = mapping->host;
2015 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2016 struct blk_plug plug;
2017 int ret;
2018
2019 /* deal with chardevs and other special file */
2020 if (!mapping->a_ops->writepage)
2021 return 0;
2022
2023 /* skip writing if there is no dirty page in this inode */
2024 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2025 return 0;
2026
2027 /* during POR, we don't need to trigger writepage at all. */
2028 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2029 goto skip_write;
2030
2031 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
2032 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2033 available_free_memory(sbi, DIRTY_DENTS))
2034 goto skip_write;
2035
2036 /* skip writing during file defragment */
2037 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2038 goto skip_write;
2039
2040 trace_f2fs_writepages(mapping->host, wbc, DATA);
2041
2042 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2043 if (wbc->sync_mode == WB_SYNC_ALL)
2044 atomic_inc(&sbi->wb_sync_req);
2045 else if (atomic_read(&sbi->wb_sync_req))
2046 goto skip_write;
2047
2048 blk_start_plug(&plug);
2049 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2050 blk_finish_plug(&plug);
2051
2052 if (wbc->sync_mode == WB_SYNC_ALL)
2053 atomic_dec(&sbi->wb_sync_req);
2054 /*
2055 * if some pages were truncated, we cannot guarantee its mapping->host
2056 * to detect pending bios.
2057 */
2058
2059 remove_dirty_inode(inode);
2060 return ret;
2061
2062 skip_write:
2063 wbc->pages_skipped += get_dirty_pages(inode);
2064 trace_f2fs_writepages(mapping->host, wbc, DATA);
2065 return 0;
2066 }
2067
2068 static int f2fs_write_data_pages(struct address_space *mapping,
2069 struct writeback_control *wbc)
2070 {
2071 struct inode *inode = mapping->host;
2072
2073 return __f2fs_write_data_pages(mapping, wbc,
2074 F2FS_I(inode)->cp_task == current ?
2075 FS_CP_DATA_IO : FS_DATA_IO);
2076 }
2077
2078 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2079 {
2080 struct inode *inode = mapping->host;
2081 loff_t i_size = i_size_read(inode);
2082
2083 if (to > i_size) {
2084 down_write(&F2FS_I(inode)->i_mmap_sem);
2085 truncate_pagecache(inode, i_size);
2086 truncate_blocks(inode, i_size, true);
2087 up_write(&F2FS_I(inode)->i_mmap_sem);
2088 }
2089 }
2090
2091 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2092 struct page *page, loff_t pos, unsigned len,
2093 block_t *blk_addr, bool *node_changed)
2094 {
2095 struct inode *inode = page->mapping->host;
2096 pgoff_t index = page->index;
2097 struct dnode_of_data dn;
2098 struct page *ipage;
2099 bool locked = false;
2100 struct extent_info ei = {0,0,0};
2101 int err = 0;
2102
2103 /*
2104 * we already allocated all the blocks, so we don't need to get
2105 * the block addresses when there is no need to fill the page.
2106 */
2107 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2108 !is_inode_flag_set(inode, FI_NO_PREALLOC))
2109 return 0;
2110
2111 if (f2fs_has_inline_data(inode) ||
2112 (pos & PAGE_MASK) >= i_size_read(inode)) {
2113 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
2114 locked = true;
2115 }
2116 restart:
2117 /* check inline_data */
2118 ipage = get_node_page(sbi, inode->i_ino);
2119 if (IS_ERR(ipage)) {
2120 err = PTR_ERR(ipage);
2121 goto unlock_out;
2122 }
2123
2124 set_new_dnode(&dn, inode, ipage, ipage, 0);
2125
2126 if (f2fs_has_inline_data(inode)) {
2127 if (pos + len <= MAX_INLINE_DATA(inode)) {
2128 read_inline_data(page, ipage);
2129 set_inode_flag(inode, FI_DATA_EXIST);
2130 if (inode->i_nlink)
2131 set_inline_node(ipage);
2132 } else {
2133 err = f2fs_convert_inline_page(&dn, page);
2134 if (err)
2135 goto out;
2136 if (dn.data_blkaddr == NULL_ADDR)
2137 err = f2fs_get_block(&dn, index);
2138 }
2139 } else if (locked) {
2140 err = f2fs_get_block(&dn, index);
2141 } else {
2142 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2143 dn.data_blkaddr = ei.blk + index - ei.fofs;
2144 } else {
2145 /* hole case */
2146 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
2147 if (err || dn.data_blkaddr == NULL_ADDR) {
2148 f2fs_put_dnode(&dn);
2149 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2150 true);
2151 locked = true;
2152 goto restart;
2153 }
2154 }
2155 }
2156
2157 /* convert_inline_page can make node_changed */
2158 *blk_addr = dn.data_blkaddr;
2159 *node_changed = dn.node_changed;
2160 out:
2161 f2fs_put_dnode(&dn);
2162 unlock_out:
2163 if (locked)
2164 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
2165 return err;
2166 }
2167
2168 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2169 loff_t pos, unsigned len, unsigned flags,
2170 struct page **pagep, void **fsdata)
2171 {
2172 struct inode *inode = mapping->host;
2173 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2174 struct page *page = NULL;
2175 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2176 bool need_balance = false, drop_atomic = false;
2177 block_t blkaddr = NULL_ADDR;
2178 int err = 0;
2179
2180 if (trace_android_fs_datawrite_start_enabled()) {
2181 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
2182
2183 path = android_fstrace_get_pathname(pathbuf,
2184 MAX_TRACE_PATHBUF_LEN,
2185 inode);
2186 trace_android_fs_datawrite_start(inode, pos, len,
2187 current->pid, path,
2188 current->comm);
2189 }
2190 trace_f2fs_write_begin(inode, pos, len, flags);
2191
2192 if (f2fs_is_atomic_file(inode) &&
2193 !available_free_memory(sbi, INMEM_PAGES)) {
2194 err = -ENOMEM;
2195 drop_atomic = true;
2196 goto fail;
2197 }
2198
2199 /*
2200 * We should check this at this moment to avoid deadlock on inode page
2201 * and #0 page. The locking rule for inline_data conversion should be:
2202 * lock_page(page #0) -> lock_page(inode_page)
2203 */
2204 if (index != 0) {
2205 err = f2fs_convert_inline_inode(inode);
2206 if (err)
2207 goto fail;
2208 }
2209 repeat:
2210 /*
2211 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2212 * wait_for_stable_page. Will wait that below with our IO control.
2213 */
2214 page = f2fs_pagecache_get_page(mapping, index,
2215 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2216 if (!page) {
2217 err = -ENOMEM;
2218 goto fail;
2219 }
2220
2221 *pagep = page;
2222
2223 err = prepare_write_begin(sbi, page, pos, len,
2224 &blkaddr, &need_balance);
2225 if (err)
2226 goto fail;
2227
2228 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
2229 unlock_page(page);
2230 f2fs_balance_fs(sbi, true);
2231 lock_page(page);
2232 if (page->mapping != mapping) {
2233 /* The page got truncated from under us */
2234 f2fs_put_page(page, 1);
2235 goto repeat;
2236 }
2237 }
2238
2239 f2fs_wait_on_page_writeback(page, DATA, false);
2240
2241 /* wait for GCed encrypted page writeback */
2242 if (f2fs_encrypted_file(inode))
2243 f2fs_wait_on_block_writeback(sbi, blkaddr);
2244
2245 if (len == PAGE_SIZE || PageUptodate(page))
2246 return 0;
2247
2248 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2249 zero_user_segment(page, len, PAGE_SIZE);
2250 return 0;
2251 }
2252
2253 if (blkaddr == NEW_ADDR) {
2254 zero_user_segment(page, 0, PAGE_SIZE);
2255 SetPageUptodate(page);
2256 } else {
2257 err = f2fs_submit_page_read(inode, page, blkaddr);
2258 if (err)
2259 goto fail;
2260
2261 lock_page(page);
2262 if (unlikely(page->mapping != mapping)) {
2263 f2fs_put_page(page, 1);
2264 goto repeat;
2265 }
2266 if (unlikely(!PageUptodate(page))) {
2267 err = -EIO;
2268 goto fail;
2269 }
2270 }
2271 return 0;
2272
2273 fail:
2274 f2fs_put_page(page, 1);
2275 f2fs_write_failed(mapping, pos + len);
2276 if (drop_atomic)
2277 drop_inmem_pages_all(sbi);
2278 return err;
2279 }
2280
2281 static int f2fs_write_end(struct file *file,
2282 struct address_space *mapping,
2283 loff_t pos, unsigned len, unsigned copied,
2284 struct page *page, void *fsdata)
2285 {
2286 struct inode *inode = page->mapping->host;
2287
2288 trace_android_fs_datawrite_end(inode, pos, len);
2289 trace_f2fs_write_end(inode, pos, len, copied);
2290
2291 /*
2292 * This should be come from len == PAGE_SIZE, and we expect copied
2293 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2294 * let generic_perform_write() try to copy data again through copied=0.
2295 */
2296 if (!PageUptodate(page)) {
2297 if (unlikely(copied != len))
2298 copied = 0;
2299 else
2300 SetPageUptodate(page);
2301 }
2302 if (!copied)
2303 goto unlock_out;
2304
2305 set_page_dirty(page);
2306
2307 if (pos + copied > i_size_read(inode))
2308 f2fs_i_size_write(inode, pos + copied);
2309 unlock_out:
2310 f2fs_put_page(page, 1);
2311 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2312 return copied;
2313 }
2314
2315 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2316 loff_t offset)
2317 {
2318 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
2319
2320 if (offset & blocksize_mask)
2321 return -EINVAL;
2322
2323 if (iov_iter_alignment(iter) & blocksize_mask)
2324 return -EINVAL;
2325
2326 return 0;
2327 }
2328
2329 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2330 {
2331 struct address_space *mapping = iocb->ki_filp->f_mapping;
2332 struct inode *inode = mapping->host;
2333 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2334 size_t count = iov_iter_count(iter);
2335 loff_t offset = iocb->ki_pos;
2336 int rw = iov_iter_rw(iter);
2337 int err;
2338 enum rw_hint hint = iocb->ki_hint;
2339 int whint_mode = F2FS_OPTION(sbi).whint_mode;
2340
2341 err = check_direct_IO(inode, iter, offset);
2342 if (err)
2343 return err;
2344
2345 if (f2fs_force_buffered_io(inode, rw))
2346 return 0;
2347
2348 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2349
2350 if (trace_android_fs_dataread_start_enabled() &&
2351 (rw == READ)) {
2352 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
2353
2354 path = android_fstrace_get_pathname(pathbuf,
2355 MAX_TRACE_PATHBUF_LEN,
2356 inode);
2357 trace_android_fs_dataread_start(inode, offset,
2358 count, current->pid, path,
2359 current->comm);
2360 }
2361 if (trace_android_fs_datawrite_start_enabled() &&
2362 (rw == WRITE)) {
2363 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
2364
2365 path = android_fstrace_get_pathname(pathbuf,
2366 MAX_TRACE_PATHBUF_LEN,
2367 inode);
2368 trace_android_fs_datawrite_start(inode, offset, count,
2369 current->pid, path,
2370 current->comm);
2371 }
2372 if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2373 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2374
2375 if (!down_read_trylock(&F2FS_I(inode)->dio_rwsem[rw])) {
2376 if (iocb->ki_flags & IOCB_NOWAIT) {
2377 iocb->ki_hint = hint;
2378 err = -EAGAIN;
2379 goto out;
2380 }
2381 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
2382 }
2383
2384 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2385 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
2386
2387 if (rw == WRITE) {
2388 if (whint_mode == WHINT_MODE_OFF)
2389 iocb->ki_hint = hint;
2390 if (err > 0) {
2391 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2392 err);
2393 set_inode_flag(inode, FI_UPDATE_WRITE);
2394 } else if (err < 0) {
2395 f2fs_write_failed(mapping, offset + count);
2396 }
2397 }
2398 out:
2399 if (trace_android_fs_dataread_start_enabled() &&
2400 (rw == READ))
2401 trace_android_fs_dataread_end(inode, offset, count);
2402 if (trace_android_fs_datawrite_start_enabled() &&
2403 (rw == WRITE))
2404 trace_android_fs_datawrite_end(inode, offset, count);
2405
2406 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2407
2408 return err;
2409 }
2410
2411 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2412 unsigned int length)
2413 {
2414 struct inode *inode = page->mapping->host;
2415 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2416
2417 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2418 (offset % PAGE_SIZE || length != PAGE_SIZE))
2419 return;
2420
2421 if (PageDirty(page)) {
2422 if (inode->i_ino == F2FS_META_INO(sbi)) {
2423 dec_page_count(sbi, F2FS_DIRTY_META);
2424 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2425 dec_page_count(sbi, F2FS_DIRTY_NODES);
2426 } else {
2427 inode_dec_dirty_pages(inode);
2428 remove_dirty_inode(inode);
2429 }
2430 }
2431
2432 /* This is atomic written page, keep Private */
2433 if (IS_ATOMIC_WRITTEN_PAGE(page))
2434 return drop_inmem_page(inode, page);
2435
2436 set_page_private(page, 0);
2437 ClearPagePrivate(page);
2438 }
2439
2440 int f2fs_release_page(struct page *page, gfp_t wait)
2441 {
2442 /* If this is dirty page, keep PagePrivate */
2443 if (PageDirty(page))
2444 return 0;
2445
2446 /* This is atomic written page, keep Private */
2447 if (IS_ATOMIC_WRITTEN_PAGE(page))
2448 return 0;
2449
2450 set_page_private(page, 0);
2451 ClearPagePrivate(page);
2452 return 1;
2453 }
2454
2455 /*
2456 * This was copied from __set_page_dirty_buffers which gives higher performance
2457 * in very high speed storages. (e.g., pmem)
2458 */
2459 void f2fs_set_page_dirty_nobuffers(struct page *page)
2460 {
2461 struct address_space *mapping = page->mapping;
2462 unsigned long flags;
2463
2464 if (unlikely(!mapping))
2465 return;
2466
2467 spin_lock(&mapping->private_lock);
2468 lock_page_memcg(page);
2469 SetPageDirty(page);
2470 spin_unlock(&mapping->private_lock);
2471
2472 spin_lock_irqsave(&mapping->tree_lock, flags);
2473 WARN_ON_ONCE(!PageUptodate(page));
2474 account_page_dirtied(page, mapping);
2475 radix_tree_tag_set(&mapping->page_tree,
2476 page_index(page), PAGECACHE_TAG_DIRTY);
2477 spin_unlock_irqrestore(&mapping->tree_lock, flags);
2478 unlock_page_memcg(page);
2479
2480 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2481 return;
2482 }
2483
2484 static int f2fs_set_data_page_dirty(struct page *page)
2485 {
2486 struct address_space *mapping = page->mapping;
2487 struct inode *inode = mapping->host;
2488
2489 trace_f2fs_set_page_dirty(page, DATA);
2490
2491 if (!PageUptodate(page))
2492 SetPageUptodate(page);
2493
2494 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2495 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2496 register_inmem_page(inode, page);
2497 return 1;
2498 }
2499 /*
2500 * Previously, this page has been registered, we just
2501 * return here.
2502 */
2503 return 0;
2504 }
2505
2506 if (!PageDirty(page)) {
2507 f2fs_set_page_dirty_nobuffers(page);
2508 update_dirty_page(inode, page);
2509 return 1;
2510 }
2511 return 0;
2512 }
2513
2514 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2515 {
2516 struct inode *inode = mapping->host;
2517
2518 if (f2fs_has_inline_data(inode))
2519 return 0;
2520
2521 /* make sure allocating whole blocks */
2522 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2523 filemap_write_and_wait(mapping);
2524
2525 return generic_block_bmap(mapping, block, get_data_block_bmap);
2526 }
2527
2528 #ifdef CONFIG_MIGRATION
2529 #include <linux/migrate.h>
2530
2531 int f2fs_migrate_page(struct address_space *mapping,
2532 struct page *newpage, struct page *page, enum migrate_mode mode)
2533 {
2534 int rc, extra_count;
2535 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2536 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2537
2538 BUG_ON(PageWriteback(page));
2539
2540 /* migrating an atomic written page is safe with the inmem_lock hold */
2541 if (atomic_written) {
2542 if (mode != MIGRATE_SYNC)
2543 return -EBUSY;
2544 if (!mutex_trylock(&fi->inmem_lock))
2545 return -EAGAIN;
2546 }
2547
2548 /*
2549 * A reference is expected if PagePrivate set when move mapping,
2550 * however F2FS breaks this for maintaining dirty page counts when
2551 * truncating pages. So here adjusting the 'extra_count' make it work.
2552 */
2553 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2554 rc = migrate_page_move_mapping(mapping, newpage,
2555 page, NULL, mode, extra_count);
2556 if (rc != MIGRATEPAGE_SUCCESS) {
2557 if (atomic_written)
2558 mutex_unlock(&fi->inmem_lock);
2559 return rc;
2560 }
2561
2562 if (atomic_written) {
2563 struct inmem_pages *cur;
2564 list_for_each_entry(cur, &fi->inmem_pages, list)
2565 if (cur->page == page) {
2566 cur->page = newpage;
2567 break;
2568 }
2569 mutex_unlock(&fi->inmem_lock);
2570 put_page(page);
2571 get_page(newpage);
2572 }
2573
2574 if (PagePrivate(page))
2575 SetPagePrivate(newpage);
2576 set_page_private(newpage, page_private(page));
2577
2578 if (mode != MIGRATE_SYNC_NO_COPY)
2579 migrate_page_copy(newpage, page);
2580 else
2581 migrate_page_states(newpage, page);
2582
2583 return MIGRATEPAGE_SUCCESS;
2584 }
2585 #endif
2586
2587 const struct address_space_operations f2fs_dblock_aops = {
2588 .readpage = f2fs_read_data_page,
2589 .readpages = f2fs_read_data_pages,
2590 .writepage = f2fs_write_data_page,
2591 .writepages = f2fs_write_data_pages,
2592 .write_begin = f2fs_write_begin,
2593 .write_end = f2fs_write_end,
2594 .set_page_dirty = f2fs_set_data_page_dirty,
2595 .invalidatepage = f2fs_invalidate_page,
2596 .releasepage = f2fs_release_page,
2597 .direct_IO = f2fs_direct_IO,
2598 .bmap = f2fs_bmap,
2599 #ifdef CONFIG_MIGRATION
2600 .migratepage = f2fs_migrate_page,
2601 #endif
2602 };