]> git.ipfire.org Git - thirdparty/kernel/stable.git/blob - fs/f2fs/data.c
projects / thirdparty / kernel / stable.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: fix encrypted page memory leak
[thirdparty/kernel/stable.git] / fs / f2fs / data.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/f2fs/data.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/buffer_head.h>
11 #include <linux/mpage.h>
12 #include <linux/writeback.h>
13 #include <linux/backing-dev.h>
14 #include <linux/pagevec.h>
15 #include <linux/blkdev.h>
16 #include <linux/bio.h>
17 #include <linux/prefetch.h>
18 #include <linux/uio.h>
19 #include <linux/cleancache.h>
20 #include <linux/sched/signal.h>
21
22 #include "f2fs.h"
23 #include "node.h"
24 #include "segment.h"
25 #include "trace.h"
26 #include <trace/events/f2fs.h>
27
28 #define NUM_PREALLOC_POST_READ_CTXS 128
29
30 static struct kmem_cache *bio_post_read_ctx_cache;
31 static mempool_t *bio_post_read_ctx_pool;
32
33 static bool __is_cp_guaranteed(struct page *page)
34 {
35 struct address_space *mapping = page->mapping;
36 struct inode *inode;
37 struct f2fs_sb_info *sbi;
38
39 if (!mapping)
40 return false;
41
42 inode = mapping->host;
43 sbi = F2FS_I_SB(inode);
44
45 if (inode->i_ino == F2FS_META_INO(sbi) ||
46 inode->i_ino == F2FS_NODE_INO(sbi) ||
47 S_ISDIR(inode->i_mode) ||
48 (S_ISREG(inode->i_mode) &&
49 (f2fs_is_atomic_file(inode) || IS_NOQUOTA(inode))) ||
50 is_cold_data(page))
51 return true;
52 return false;
53 }
54
55 static enum count_type __read_io_type(struct page *page)
56 {
57 struct address_space *mapping = page->mapping;
58
59 if (mapping) {
60 struct inode *inode = mapping->host;
61 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
62
63 if (inode->i_ino == F2FS_META_INO(sbi))
64 return F2FS_RD_META;
65
66 if (inode->i_ino == F2FS_NODE_INO(sbi))
67 return F2FS_RD_NODE;
68 }
69 return F2FS_RD_DATA;
70 }
71
72 /* postprocessing steps for read bios */
73 enum bio_post_read_step {
74 STEP_INITIAL = 0,
75 STEP_DECRYPT,
76 };
77
78 struct bio_post_read_ctx {
79 struct bio *bio;
80 struct work_struct work;
81 unsigned int cur_step;
82 unsigned int enabled_steps;
83 };
84
85 static void __read_end_io(struct bio *bio)
86 {
87 struct page *page;
88 struct bio_vec *bv;
89 int i;
90
91 bio_for_each_segment_all(bv, bio, i) {
92 page = bv->bv_page;
93
94 /* PG_error was set if any post_read step failed */
95 if (bio->bi_status || PageError(page)) {
96 ClearPageUptodate(page);
97 /* will re-read again later */
98 ClearPageError(page);
99 } else {
100 SetPageUptodate(page);
101 }
102 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
103 unlock_page(page);
104 }
105 if (bio->bi_private)
106 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
107 bio_put(bio);
108 }
109
110 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
111
112 static void decrypt_work(struct work_struct *work)
113 {
114 struct bio_post_read_ctx *ctx =
115 container_of(work, struct bio_post_read_ctx, work);
116
117 fscrypt_decrypt_bio(ctx->bio);
118
119 bio_post_read_processing(ctx);
120 }
121
122 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
123 {
124 switch (++ctx->cur_step) {
125 case STEP_DECRYPT:
126 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
127 INIT_WORK(&ctx->work, decrypt_work);
128 fscrypt_enqueue_decrypt_work(&ctx->work);
129 return;
130 }
131 ctx->cur_step++;
132 /* fall-through */
133 default:
134 __read_end_io(ctx->bio);
135 }
136 }
137
138 static bool f2fs_bio_post_read_required(struct bio *bio)
139 {
140 return bio->bi_private && !bio->bi_status;
141 }
142
143 static void f2fs_read_end_io(struct bio *bio)
144 {
145 if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)),
146 FAULT_READ_IO)) {
147 f2fs_show_injection_info(FAULT_READ_IO);
148 bio->bi_status = BLK_STS_IOERR;
149 }
150
151 if (f2fs_bio_post_read_required(bio)) {
152 struct bio_post_read_ctx *ctx = bio->bi_private;
153
154 ctx->cur_step = STEP_INITIAL;
155 bio_post_read_processing(ctx);
156 return;
157 }
158
159 __read_end_io(bio);
160 }
161
162 static void f2fs_write_end_io(struct bio *bio)
163 {
164 struct f2fs_sb_info *sbi = bio->bi_private;
165 struct bio_vec *bvec;
166 int i;
167
168 if (time_to_inject(sbi, FAULT_WRITE_IO)) {
169 f2fs_show_injection_info(FAULT_WRITE_IO);
170 bio->bi_status = BLK_STS_IOERR;
171 }
172
173 bio_for_each_segment_all(bvec, bio, i) {
174 struct page *page = bvec->bv_page;
175 enum count_type type = WB_DATA_TYPE(page);
176
177 if (IS_DUMMY_WRITTEN_PAGE(page)) {
178 set_page_private(page, (unsigned long)NULL);
179 ClearPagePrivate(page);
180 unlock_page(page);
181 mempool_free(page, sbi->write_io_dummy);
182
183 if (unlikely(bio->bi_status))
184 f2fs_stop_checkpoint(sbi, true);
185 continue;
186 }
187
188 fscrypt_pullback_bio_page(&page, true);
189
190 if (unlikely(bio->bi_status)) {
191 mapping_set_error(page->mapping, -EIO);
192 if (type == F2FS_WB_CP_DATA)
193 f2fs_stop_checkpoint(sbi, true);
194 }
195
196 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
197 page->index != nid_of_node(page));
198
199 dec_page_count(sbi, type);
200 if (f2fs_in_warm_node_list(sbi, page))
201 f2fs_del_fsync_node_entry(sbi, page);
202 clear_cold_data(page);
203 end_page_writeback(page);
204 }
205 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
206 wq_has_sleeper(&sbi->cp_wait))
207 wake_up(&sbi->cp_wait);
208
209 bio_put(bio);
210 }
211
212 /*
213 * Return true, if pre_bio's bdev is same as its target device.
214 */
215 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
216 block_t blk_addr, struct bio *bio)
217 {
218 struct block_device *bdev = sbi->sb->s_bdev;
219 int i;
220
221 for (i = 0; i < sbi->s_ndevs; i++) {
222 if (FDEV(i).start_blk <= blk_addr &&
223 FDEV(i).end_blk >= blk_addr) {
224 blk_addr -= FDEV(i).start_blk;
225 bdev = FDEV(i).bdev;
226 break;
227 }
228 }
229 if (bio) {
230 bio_set_dev(bio, bdev);
231 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
232 }
233 return bdev;
234 }
235
236 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
237 {
238 int i;
239
240 for (i = 0; i < sbi->s_ndevs; i++)
241 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
242 return i;
243 return 0;
244 }
245
246 static bool __same_bdev(struct f2fs_sb_info *sbi,
247 block_t blk_addr, struct bio *bio)
248 {
249 struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
250 return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
251 }
252
253 /*
254 * Low-level block read/write IO operations.
255 */
256 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
257 struct writeback_control *wbc,
258 int npages, bool is_read,
259 enum page_type type, enum temp_type temp)
260 {
261 struct bio *bio;
262
263 bio = f2fs_bio_alloc(sbi, npages, true);
264
265 f2fs_target_device(sbi, blk_addr, bio);
266 if (is_read) {
267 bio->bi_end_io = f2fs_read_end_io;
268 bio->bi_private = NULL;
269 } else {
270 bio->bi_end_io = f2fs_write_end_io;
271 bio->bi_private = sbi;
272 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, type, temp);
273 }
274 if (wbc)
275 wbc_init_bio(wbc, bio);
276
277 return bio;
278 }
279
280 static inline void __submit_bio(struct f2fs_sb_info *sbi,
281 struct bio *bio, enum page_type type)
282 {
283 if (!is_read_io(bio_op(bio))) {
284 unsigned int start;
285
286 if (type != DATA && type != NODE)
287 goto submit_io;
288
289 if (test_opt(sbi, LFS) && current->plug)
290 blk_finish_plug(current->plug);
291
292 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
293 start %= F2FS_IO_SIZE(sbi);
294
295 if (start == 0)
296 goto submit_io;
297
298 /* fill dummy pages */
299 for (; start < F2FS_IO_SIZE(sbi); start++) {
300 struct page *page =
301 mempool_alloc(sbi->write_io_dummy,
302 GFP_NOIO | __GFP_NOFAIL);
303 f2fs_bug_on(sbi, !page);
304
305 zero_user_segment(page, 0, PAGE_SIZE);
306 SetPagePrivate(page);
307 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
308 lock_page(page);
309 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
310 f2fs_bug_on(sbi, 1);
311 }
312 /*
313 * In the NODE case, we lose next block address chain. So, we
314 * need to do checkpoint in f2fs_sync_file.
315 */
316 if (type == NODE)
317 set_sbi_flag(sbi, SBI_NEED_CP);
318 }
319 submit_io:
320 if (is_read_io(bio_op(bio)))
321 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
322 else
323 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
324 submit_bio(bio);
325 }
326
327 static void __submit_merged_bio(struct f2fs_bio_info *io)
328 {
329 struct f2fs_io_info *fio = &io->fio;
330
331 if (!io->bio)
332 return;
333
334 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
335
336 if (is_read_io(fio->op))
337 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
338 else
339 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
340
341 __submit_bio(io->sbi, io->bio, fio->type);
342 io->bio = NULL;
343 }
344
345 static bool __has_merged_page(struct f2fs_bio_info *io, struct inode *inode,
346 struct page *page, nid_t ino)
347 {
348 struct bio_vec *bvec;
349 struct page *target;
350 int i;
351
352 if (!io->bio)
353 return false;
354
355 if (!inode && !page && !ino)
356 return true;
357
358 bio_for_each_segment_all(bvec, io->bio, i) {
359
360 if (bvec->bv_page->mapping)
361 target = bvec->bv_page;
362 else
363 target = fscrypt_control_page(bvec->bv_page);
364
365 if (inode && inode == target->mapping->host)
366 return true;
367 if (page && page == target)
368 return true;
369 if (ino && ino == ino_of_node(target))
370 return true;
371 }
372
373 return false;
374 }
375
376 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
377 enum page_type type, enum temp_type temp)
378 {
379 enum page_type btype = PAGE_TYPE_OF_BIO(type);
380 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
381
382 down_write(&io->io_rwsem);
383
384 /* change META to META_FLUSH in the checkpoint procedure */
385 if (type >= META_FLUSH) {
386 io->fio.type = META_FLUSH;
387 io->fio.op = REQ_OP_WRITE;
388 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
389 if (!test_opt(sbi, NOBARRIER))
390 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
391 }
392 __submit_merged_bio(io);
393 up_write(&io->io_rwsem);
394 }
395
396 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
397 struct inode *inode, struct page *page,
398 nid_t ino, enum page_type type, bool force)
399 {
400 enum temp_type temp;
401 bool ret = true;
402
403 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
404 if (!force) {
405 enum page_type btype = PAGE_TYPE_OF_BIO(type);
406 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
407
408 down_read(&io->io_rwsem);
409 ret = __has_merged_page(io, inode, page, ino);
410 up_read(&io->io_rwsem);
411 }
412 if (ret)
413 __f2fs_submit_merged_write(sbi, type, temp);
414
415 /* TODO: use HOT temp only for meta pages now. */
416 if (type >= META)
417 break;
418 }
419 }
420
421 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
422 {
423 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
424 }
425
426 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
427 struct inode *inode, struct page *page,
428 nid_t ino, enum page_type type)
429 {
430 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
431 }
432
433 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
434 {
435 f2fs_submit_merged_write(sbi, DATA);
436 f2fs_submit_merged_write(sbi, NODE);
437 f2fs_submit_merged_write(sbi, META);
438 }
439
440 /*
441 * Fill the locked page with data located in the block address.
442 * A caller needs to unlock the page on failure.
443 */
444 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
445 {
446 struct bio *bio;
447 struct page *page = fio->encrypted_page ?
448 fio->encrypted_page : fio->page;
449
450 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
451 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
452 return -EFAULT;
453
454 trace_f2fs_submit_page_bio(page, fio);
455 f2fs_trace_ios(fio, 0);
456
457 /* Allocate a new bio */
458 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
459 1, is_read_io(fio->op), fio->type, fio->temp);
460
461 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
462 bio_put(bio);
463 return -EFAULT;
464 }
465
466 if (fio->io_wbc && !is_read_io(fio->op))
467 wbc_account_io(fio->io_wbc, page, PAGE_SIZE);
468
469 bio_set_op_attrs(bio, fio->op, fio->op_flags);
470
471 inc_page_count(fio->sbi, is_read_io(fio->op) ?
472 __read_io_type(page): WB_DATA_TYPE(fio->page));
473
474 __submit_bio(fio->sbi, bio, fio->type);
475 return 0;
476 }
477
478 void f2fs_submit_page_write(struct f2fs_io_info *fio)
479 {
480 struct f2fs_sb_info *sbi = fio->sbi;
481 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
482 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
483 struct page *bio_page;
484
485 f2fs_bug_on(sbi, is_read_io(fio->op));
486
487 down_write(&io->io_rwsem);
488 next:
489 if (fio->in_list) {
490 spin_lock(&io->io_lock);
491 if (list_empty(&io->io_list)) {
492 spin_unlock(&io->io_lock);
493 goto out;
494 }
495 fio = list_first_entry(&io->io_list,
496 struct f2fs_io_info, list);
497 list_del(&fio->list);
498 spin_unlock(&io->io_lock);
499 }
500
501 if (__is_valid_data_blkaddr(fio->old_blkaddr))
502 verify_block_addr(fio, fio->old_blkaddr);
503 verify_block_addr(fio, fio->new_blkaddr);
504
505 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
506
507 /* set submitted = true as a return value */
508 fio->submitted = true;
509
510 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
511
512 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
513 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
514 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
515 __submit_merged_bio(io);
516 alloc_new:
517 if (io->bio == NULL) {
518 if ((fio->type == DATA || fio->type == NODE) &&
519 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
520 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
521 fio->retry = true;
522 goto skip;
523 }
524 io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
525 BIO_MAX_PAGES, false,
526 fio->type, fio->temp);
527 io->fio = *fio;
528 }
529
530 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
531 __submit_merged_bio(io);
532 goto alloc_new;
533 }
534
535 if (fio->io_wbc)
536 wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
537
538 io->last_block_in_bio = fio->new_blkaddr;
539 f2fs_trace_ios(fio, 0);
540
541 trace_f2fs_submit_page_write(fio->page, fio);
542 skip:
543 if (fio->in_list)
544 goto next;
545 out:
546 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
547 f2fs_is_checkpoint_ready(sbi))
548 __submit_merged_bio(io);
549 up_write(&io->io_rwsem);
550 }
551
552 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
553 unsigned nr_pages, unsigned op_flag)
554 {
555 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
556 struct bio *bio;
557 struct bio_post_read_ctx *ctx;
558 unsigned int post_read_steps = 0;
559
560 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC))
561 return ERR_PTR(-EFAULT);
562
563 bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
564 if (!bio)
565 return ERR_PTR(-ENOMEM);
566 f2fs_target_device(sbi, blkaddr, bio);
567 bio->bi_end_io = f2fs_read_end_io;
568 bio_set_op_attrs(bio, REQ_OP_READ, op_flag);
569
570 if (f2fs_encrypted_file(inode))
571 post_read_steps |= 1 << STEP_DECRYPT;
572 if (post_read_steps) {
573 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
574 if (!ctx) {
575 bio_put(bio);
576 return ERR_PTR(-ENOMEM);
577 }
578 ctx->bio = bio;
579 ctx->enabled_steps = post_read_steps;
580 bio->bi_private = ctx;
581 }
582
583 return bio;
584 }
585
586 /* This can handle encryption stuffs */
587 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
588 block_t blkaddr)
589 {
590 struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1, 0);
591
592 if (IS_ERR(bio))
593 return PTR_ERR(bio);
594
595 /* wait for GCed page writeback via META_MAPPING */
596 f2fs_wait_on_block_writeback(inode, blkaddr);
597
598 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
599 bio_put(bio);
600 return -EFAULT;
601 }
602 ClearPageError(page);
603 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
604 __submit_bio(F2FS_I_SB(inode), bio, DATA);
605 return 0;
606 }
607
608 static void __set_data_blkaddr(struct dnode_of_data *dn)
609 {
610 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
611 __le32 *addr_array;
612 int base = 0;
613
614 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
615 base = get_extra_isize(dn->inode);
616
617 /* Get physical address of data block */
618 addr_array = blkaddr_in_node(rn);
619 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
620 }
621
622 /*
623 * Lock ordering for the change of data block address:
624 * ->data_page
625 * ->node_page
626 * update block addresses in the node page
627 */
628 void f2fs_set_data_blkaddr(struct dnode_of_data *dn)
629 {
630 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
631 __set_data_blkaddr(dn);
632 if (set_page_dirty(dn->node_page))
633 dn->node_changed = true;
634 }
635
636 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
637 {
638 dn->data_blkaddr = blkaddr;
639 f2fs_set_data_blkaddr(dn);
640 f2fs_update_extent_cache(dn);
641 }
642
643 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
644 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
645 {
646 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
647 int err;
648
649 if (!count)
650 return 0;
651
652 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
653 return -EPERM;
654 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
655 return err;
656
657 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
658 dn->ofs_in_node, count);
659
660 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
661
662 for (; count > 0; dn->ofs_in_node++) {
663 block_t blkaddr = datablock_addr(dn->inode,
664 dn->node_page, dn->ofs_in_node);
665 if (blkaddr == NULL_ADDR) {
666 dn->data_blkaddr = NEW_ADDR;
667 __set_data_blkaddr(dn);
668 count--;
669 }
670 }
671
672 if (set_page_dirty(dn->node_page))
673 dn->node_changed = true;
674 return 0;
675 }
676
677 /* Should keep dn->ofs_in_node unchanged */
678 int f2fs_reserve_new_block(struct dnode_of_data *dn)
679 {
680 unsigned int ofs_in_node = dn->ofs_in_node;
681 int ret;
682
683 ret = f2fs_reserve_new_blocks(dn, 1);
684 dn->ofs_in_node = ofs_in_node;
685 return ret;
686 }
687
688 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
689 {
690 bool need_put = dn->inode_page ? false : true;
691 int err;
692
693 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
694 if (err)
695 return err;
696
697 if (dn->data_blkaddr == NULL_ADDR)
698 err = f2fs_reserve_new_block(dn);
699 if (err || need_put)
700 f2fs_put_dnode(dn);
701 return err;
702 }
703
704 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
705 {
706 struct extent_info ei = {0,0,0};
707 struct inode *inode = dn->inode;
708
709 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
710 dn->data_blkaddr = ei.blk + index - ei.fofs;
711 return 0;
712 }
713
714 return f2fs_reserve_block(dn, index);
715 }
716
717 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
718 int op_flags, bool for_write)
719 {
720 struct address_space *mapping = inode->i_mapping;
721 struct dnode_of_data dn;
722 struct page *page;
723 struct extent_info ei = {0,0,0};
724 int err;
725
726 page = f2fs_grab_cache_page(mapping, index, for_write);
727 if (!page)
728 return ERR_PTR(-ENOMEM);
729
730 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
731 dn.data_blkaddr = ei.blk + index - ei.fofs;
732 goto got_it;
733 }
734
735 set_new_dnode(&dn, inode, NULL, NULL, 0);
736 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
737 if (err)
738 goto put_err;
739 f2fs_put_dnode(&dn);
740
741 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
742 err = -ENOENT;
743 goto put_err;
744 }
745 got_it:
746 if (PageUptodate(page)) {
747 unlock_page(page);
748 return page;
749 }
750
751 /*
752 * A new dentry page is allocated but not able to be written, since its
753 * new inode page couldn't be allocated due to -ENOSPC.
754 * In such the case, its blkaddr can be remained as NEW_ADDR.
755 * see, f2fs_add_link -> f2fs_get_new_data_page ->
756 * f2fs_init_inode_metadata.
757 */
758 if (dn.data_blkaddr == NEW_ADDR) {
759 zero_user_segment(page, 0, PAGE_SIZE);
760 if (!PageUptodate(page))
761 SetPageUptodate(page);
762 unlock_page(page);
763 return page;
764 }
765
766 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
767 if (err)
768 goto put_err;
769 return page;
770
771 put_err:
772 f2fs_put_page(page, 1);
773 return ERR_PTR(err);
774 }
775
776 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index)
777 {
778 struct address_space *mapping = inode->i_mapping;
779 struct page *page;
780
781 page = find_get_page(mapping, index);
782 if (page && PageUptodate(page))
783 return page;
784 f2fs_put_page(page, 0);
785
786 page = f2fs_get_read_data_page(inode, index, 0, false);
787 if (IS_ERR(page))
788 return page;
789
790 if (PageUptodate(page))
791 return page;
792
793 wait_on_page_locked(page);
794 if (unlikely(!PageUptodate(page))) {
795 f2fs_put_page(page, 0);
796 return ERR_PTR(-EIO);
797 }
798 return page;
799 }
800
801 /*
802 * If it tries to access a hole, return an error.
803 * Because, the callers, functions in dir.c and GC, should be able to know
804 * whether this page exists or not.
805 */
806 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
807 bool for_write)
808 {
809 struct address_space *mapping = inode->i_mapping;
810 struct page *page;
811 repeat:
812 page = f2fs_get_read_data_page(inode, index, 0, for_write);
813 if (IS_ERR(page))
814 return page;
815
816 /* wait for read completion */
817 lock_page(page);
818 if (unlikely(page->mapping != mapping)) {
819 f2fs_put_page(page, 1);
820 goto repeat;
821 }
822 if (unlikely(!PageUptodate(page))) {
823 f2fs_put_page(page, 1);
824 return ERR_PTR(-EIO);
825 }
826 return page;
827 }
828
829 /*
830 * Caller ensures that this data page is never allocated.
831 * A new zero-filled data page is allocated in the page cache.
832 *
833 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
834 * f2fs_unlock_op().
835 * Note that, ipage is set only by make_empty_dir, and if any error occur,
836 * ipage should be released by this function.
837 */
838 struct page *f2fs_get_new_data_page(struct inode *inode,
839 struct page *ipage, pgoff_t index, bool new_i_size)
840 {
841 struct address_space *mapping = inode->i_mapping;
842 struct page *page;
843 struct dnode_of_data dn;
844 int err;
845
846 page = f2fs_grab_cache_page(mapping, index, true);
847 if (!page) {
848 /*
849 * before exiting, we should make sure ipage will be released
850 * if any error occur.
851 */
852 f2fs_put_page(ipage, 1);
853 return ERR_PTR(-ENOMEM);
854 }
855
856 set_new_dnode(&dn, inode, ipage, NULL, 0);
857 err = f2fs_reserve_block(&dn, index);
858 if (err) {
859 f2fs_put_page(page, 1);
860 return ERR_PTR(err);
861 }
862 if (!ipage)
863 f2fs_put_dnode(&dn);
864
865 if (PageUptodate(page))
866 goto got_it;
867
868 if (dn.data_blkaddr == NEW_ADDR) {
869 zero_user_segment(page, 0, PAGE_SIZE);
870 if (!PageUptodate(page))
871 SetPageUptodate(page);
872 } else {
873 f2fs_put_page(page, 1);
874
875 /* if ipage exists, blkaddr should be NEW_ADDR */
876 f2fs_bug_on(F2FS_I_SB(inode), ipage);
877 page = f2fs_get_lock_data_page(inode, index, true);
878 if (IS_ERR(page))
879 return page;
880 }
881 got_it:
882 if (new_i_size && i_size_read(inode) <
883 ((loff_t)(index + 1) << PAGE_SHIFT))
884 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
885 return page;
886 }
887
888 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
889 {
890 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
891 struct f2fs_summary sum;
892 struct node_info ni;
893 block_t old_blkaddr;
894 blkcnt_t count = 1;
895 int err;
896
897 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
898 return -EPERM;
899
900 err = f2fs_get_node_info(sbi, dn->nid, &ni);
901 if (err)
902 return err;
903
904 dn->data_blkaddr = datablock_addr(dn->inode,
905 dn->node_page, dn->ofs_in_node);
906 if (dn->data_blkaddr != NULL_ADDR)
907 goto alloc;
908
909 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
910 return err;
911
912 alloc:
913 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
914 old_blkaddr = dn->data_blkaddr;
915 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
916 &sum, seg_type, NULL, false);
917 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
918 invalidate_mapping_pages(META_MAPPING(sbi),
919 old_blkaddr, old_blkaddr);
920 f2fs_set_data_blkaddr(dn);
921
922 /*
923 * i_size will be updated by direct_IO. Otherwise, we'll get stale
924 * data from unwritten block via dio_read.
925 */
926 return 0;
927 }
928
929 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
930 {
931 struct inode *inode = file_inode(iocb->ki_filp);
932 struct f2fs_map_blocks map;
933 int flag;
934 int err = 0;
935 bool direct_io = iocb->ki_flags & IOCB_DIRECT;
936
937 /* convert inline data for Direct I/O*/
938 if (direct_io) {
939 err = f2fs_convert_inline_inode(inode);
940 if (err)
941 return err;
942 }
943
944 if (direct_io && allow_outplace_dio(inode, iocb, from))
945 return 0;
946
947 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
948 return 0;
949
950 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
951 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
952 if (map.m_len > map.m_lblk)
953 map.m_len -= map.m_lblk;
954 else
955 map.m_len = 0;
956
957 map.m_next_pgofs = NULL;
958 map.m_next_extent = NULL;
959 map.m_seg_type = NO_CHECK_TYPE;
960 map.m_may_create = true;
961
962 if (direct_io) {
963 map.m_seg_type = f2fs_rw_hint_to_seg_type(iocb->ki_hint);
964 flag = f2fs_force_buffered_io(inode, iocb, from) ?
965 F2FS_GET_BLOCK_PRE_AIO :
966 F2FS_GET_BLOCK_PRE_DIO;
967 goto map_blocks;
968 }
969 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
970 err = f2fs_convert_inline_inode(inode);
971 if (err)
972 return err;
973 }
974 if (f2fs_has_inline_data(inode))
975 return err;
976
977 flag = F2FS_GET_BLOCK_PRE_AIO;
978
979 map_blocks:
980 err = f2fs_map_blocks(inode, &map, 1, flag);
981 if (map.m_len > 0 && err == -ENOSPC) {
982 if (!direct_io)
983 set_inode_flag(inode, FI_NO_PREALLOC);
984 err = 0;
985 }
986 return err;
987 }
988
989 void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
990 {
991 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
992 if (lock)
993 down_read(&sbi->node_change);
994 else
995 up_read(&sbi->node_change);
996 } else {
997 if (lock)
998 f2fs_lock_op(sbi);
999 else
1000 f2fs_unlock_op(sbi);
1001 }
1002 }
1003
1004 /*
1005 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
1006 * f2fs_map_blocks structure.
1007 * If original data blocks are allocated, then give them to blockdev.
1008 * Otherwise,
1009 * a. preallocate requested block addresses
1010 * b. do not use extent cache for better performance
1011 * c. give the block addresses to blockdev
1012 */
1013 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
1014 int create, int flag)
1015 {
1016 unsigned int maxblocks = map->m_len;
1017 struct dnode_of_data dn;
1018 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1019 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1020 pgoff_t pgofs, end_offset, end;
1021 int err = 0, ofs = 1;
1022 unsigned int ofs_in_node, last_ofs_in_node;
1023 blkcnt_t prealloc;
1024 struct extent_info ei = {0,0,0};
1025 block_t blkaddr;
1026 unsigned int start_pgofs;
1027
1028 if (!maxblocks)
1029 return 0;
1030
1031 map->m_len = 0;
1032 map->m_flags = 0;
1033
1034 /* it only supports block size == page size */
1035 pgofs = (pgoff_t)map->m_lblk;
1036 end = pgofs + maxblocks;
1037
1038 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
1039 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1040 map->m_may_create)
1041 goto next_dnode;
1042
1043 map->m_pblk = ei.blk + pgofs - ei.fofs;
1044 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
1045 map->m_flags = F2FS_MAP_MAPPED;
1046 if (map->m_next_extent)
1047 *map->m_next_extent = pgofs + map->m_len;
1048
1049 /* for hardware encryption, but to avoid potential issue in future */
1050 if (flag == F2FS_GET_BLOCK_DIO)
1051 f2fs_wait_on_block_writeback_range(inode,
1052 map->m_pblk, map->m_len);
1053 goto out;
1054 }
1055
1056 next_dnode:
1057 if (map->m_may_create)
1058 __do_map_lock(sbi, flag, true);
1059
1060 /* When reading holes, we need its node page */
1061 set_new_dnode(&dn, inode, NULL, NULL, 0);
1062 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1063 if (err) {
1064 if (flag == F2FS_GET_BLOCK_BMAP)
1065 map->m_pblk = 0;
1066 if (err == -ENOENT) {
1067 err = 0;
1068 if (map->m_next_pgofs)
1069 *map->m_next_pgofs =
1070 f2fs_get_next_page_offset(&dn, pgofs);
1071 if (map->m_next_extent)
1072 *map->m_next_extent =
1073 f2fs_get_next_page_offset(&dn, pgofs);
1074 }
1075 goto unlock_out;
1076 }
1077
1078 start_pgofs = pgofs;
1079 prealloc = 0;
1080 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1081 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1082
1083 next_block:
1084 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
1085
1086 if (__is_valid_data_blkaddr(blkaddr) &&
1087 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
1088 err = -EFAULT;
1089 goto sync_out;
1090 }
1091
1092 if (is_valid_data_blkaddr(sbi, blkaddr)) {
1093 /* use out-place-update for driect IO under LFS mode */
1094 if (test_opt(sbi, LFS) && flag == F2FS_GET_BLOCK_DIO &&
1095 map->m_may_create) {
1096 err = __allocate_data_block(&dn, map->m_seg_type);
1097 if (!err) {
1098 blkaddr = dn.data_blkaddr;
1099 set_inode_flag(inode, FI_APPEND_WRITE);
1100 }
1101 }
1102 } else {
1103 if (create) {
1104 if (unlikely(f2fs_cp_error(sbi))) {
1105 err = -EIO;
1106 goto sync_out;
1107 }
1108 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
1109 if (blkaddr == NULL_ADDR) {
1110 prealloc++;
1111 last_ofs_in_node = dn.ofs_in_node;
1112 }
1113 } else {
1114 WARN_ON(flag != F2FS_GET_BLOCK_PRE_DIO &&
1115 flag != F2FS_GET_BLOCK_DIO);
1116 err = __allocate_data_block(&dn,
1117 map->m_seg_type);
1118 if (!err)
1119 set_inode_flag(inode, FI_APPEND_WRITE);
1120 }
1121 if (err)
1122 goto sync_out;
1123 map->m_flags |= F2FS_MAP_NEW;
1124 blkaddr = dn.data_blkaddr;
1125 } else {
1126 if (flag == F2FS_GET_BLOCK_BMAP) {
1127 map->m_pblk = 0;
1128 goto sync_out;
1129 }
1130 if (flag == F2FS_GET_BLOCK_PRECACHE)
1131 goto sync_out;
1132 if (flag == F2FS_GET_BLOCK_FIEMAP &&
1133 blkaddr == NULL_ADDR) {
1134 if (map->m_next_pgofs)
1135 *map->m_next_pgofs = pgofs + 1;
1136 goto sync_out;
1137 }
1138 if (flag != F2FS_GET_BLOCK_FIEMAP) {
1139 /* for defragment case */
1140 if (map->m_next_pgofs)
1141 *map->m_next_pgofs = pgofs + 1;
1142 goto sync_out;
1143 }
1144 }
1145 }
1146
1147 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1148 goto skip;
1149
1150 if (map->m_len == 0) {
1151 /* preallocated unwritten block should be mapped for fiemap. */
1152 if (blkaddr == NEW_ADDR)
1153 map->m_flags |= F2FS_MAP_UNWRITTEN;
1154 map->m_flags |= F2FS_MAP_MAPPED;
1155
1156 map->m_pblk = blkaddr;
1157 map->m_len = 1;
1158 } else if ((map->m_pblk != NEW_ADDR &&
1159 blkaddr == (map->m_pblk + ofs)) ||
1160 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1161 flag == F2FS_GET_BLOCK_PRE_DIO) {
1162 ofs++;
1163 map->m_len++;
1164 } else {
1165 goto sync_out;
1166 }
1167
1168 skip:
1169 dn.ofs_in_node++;
1170 pgofs++;
1171
1172 /* preallocate blocks in batch for one dnode page */
1173 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1174 (pgofs == end || dn.ofs_in_node == end_offset)) {
1175
1176 dn.ofs_in_node = ofs_in_node;
1177 err = f2fs_reserve_new_blocks(&dn, prealloc);
1178 if (err)
1179 goto sync_out;
1180
1181 map->m_len += dn.ofs_in_node - ofs_in_node;
1182 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1183 err = -ENOSPC;
1184 goto sync_out;
1185 }
1186 dn.ofs_in_node = end_offset;
1187 }
1188
1189 if (pgofs >= end)
1190 goto sync_out;
1191 else if (dn.ofs_in_node < end_offset)
1192 goto next_block;
1193
1194 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1195 if (map->m_flags & F2FS_MAP_MAPPED) {
1196 unsigned int ofs = start_pgofs - map->m_lblk;
1197
1198 f2fs_update_extent_cache_range(&dn,
1199 start_pgofs, map->m_pblk + ofs,
1200 map->m_len - ofs);
1201 }
1202 }
1203
1204 f2fs_put_dnode(&dn);
1205
1206 if (map->m_may_create) {
1207 __do_map_lock(sbi, flag, false);
1208 f2fs_balance_fs(sbi, dn.node_changed);
1209 }
1210 goto next_dnode;
1211
1212 sync_out:
1213
1214 /* for hardware encryption, but to avoid potential issue in future */
1215 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED)
1216 f2fs_wait_on_block_writeback_range(inode,
1217 map->m_pblk, map->m_len);
1218
1219 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1220 if (map->m_flags & F2FS_MAP_MAPPED) {
1221 unsigned int ofs = start_pgofs - map->m_lblk;
1222
1223 f2fs_update_extent_cache_range(&dn,
1224 start_pgofs, map->m_pblk + ofs,
1225 map->m_len - ofs);
1226 }
1227 if (map->m_next_extent)
1228 *map->m_next_extent = pgofs + 1;
1229 }
1230 f2fs_put_dnode(&dn);
1231 unlock_out:
1232 if (map->m_may_create) {
1233 __do_map_lock(sbi, flag, false);
1234 f2fs_balance_fs(sbi, dn.node_changed);
1235 }
1236 out:
1237 trace_f2fs_map_blocks(inode, map, err);
1238 return err;
1239 }
1240
1241 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1242 {
1243 struct f2fs_map_blocks map;
1244 block_t last_lblk;
1245 int err;
1246
1247 if (pos + len > i_size_read(inode))
1248 return false;
1249
1250 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1251 map.m_next_pgofs = NULL;
1252 map.m_next_extent = NULL;
1253 map.m_seg_type = NO_CHECK_TYPE;
1254 map.m_may_create = false;
1255 last_lblk = F2FS_BLK_ALIGN(pos + len);
1256
1257 while (map.m_lblk < last_lblk) {
1258 map.m_len = last_lblk - map.m_lblk;
1259 err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1260 if (err || map.m_len == 0)
1261 return false;
1262 map.m_lblk += map.m_len;
1263 }
1264 return true;
1265 }
1266
1267 static int __get_data_block(struct inode *inode, sector_t iblock,
1268 struct buffer_head *bh, int create, int flag,
1269 pgoff_t *next_pgofs, int seg_type, bool may_write)
1270 {
1271 struct f2fs_map_blocks map;
1272 int err;
1273
1274 map.m_lblk = iblock;
1275 map.m_len = bh->b_size >> inode->i_blkbits;
1276 map.m_next_pgofs = next_pgofs;
1277 map.m_next_extent = NULL;
1278 map.m_seg_type = seg_type;
1279 map.m_may_create = may_write;
1280
1281 err = f2fs_map_blocks(inode, &map, create, flag);
1282 if (!err) {
1283 map_bh(bh, inode->i_sb, map.m_pblk);
1284 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1285 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1286 }
1287 return err;
1288 }
1289
1290 static int get_data_block(struct inode *inode, sector_t iblock,
1291 struct buffer_head *bh_result, int create, int flag,
1292 pgoff_t *next_pgofs)
1293 {
1294 return __get_data_block(inode, iblock, bh_result, create,
1295 flag, next_pgofs,
1296 NO_CHECK_TYPE, create);
1297 }
1298
1299 static int get_data_block_dio_write(struct inode *inode, sector_t iblock,
1300 struct buffer_head *bh_result, int create)
1301 {
1302 return __get_data_block(inode, iblock, bh_result, create,
1303 F2FS_GET_BLOCK_DIO, NULL,
1304 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1305 true);
1306 }
1307
1308 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1309 struct buffer_head *bh_result, int create)
1310 {
1311 return __get_data_block(inode, iblock, bh_result, create,
1312 F2FS_GET_BLOCK_DIO, NULL,
1313 f2fs_rw_hint_to_seg_type(inode->i_write_hint),
1314 false);
1315 }
1316
1317 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1318 struct buffer_head *bh_result, int create)
1319 {
1320 /* Block number less than F2FS MAX BLOCKS */
1321 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1322 return -EFBIG;
1323
1324 return __get_data_block(inode, iblock, bh_result, create,
1325 F2FS_GET_BLOCK_BMAP, NULL,
1326 NO_CHECK_TYPE, create);
1327 }
1328
1329 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1330 {
1331 return (offset >> inode->i_blkbits);
1332 }
1333
1334 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1335 {
1336 return (blk << inode->i_blkbits);
1337 }
1338
1339 static int f2fs_xattr_fiemap(struct inode *inode,
1340 struct fiemap_extent_info *fieinfo)
1341 {
1342 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1343 struct page *page;
1344 struct node_info ni;
1345 __u64 phys = 0, len;
1346 __u32 flags;
1347 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1348 int err = 0;
1349
1350 if (f2fs_has_inline_xattr(inode)) {
1351 int offset;
1352
1353 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1354 inode->i_ino, false);
1355 if (!page)
1356 return -ENOMEM;
1357
1358 err = f2fs_get_node_info(sbi, inode->i_ino, &ni);
1359 if (err) {
1360 f2fs_put_page(page, 1);
1361 return err;
1362 }
1363
1364 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1365 offset = offsetof(struct f2fs_inode, i_addr) +
1366 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1367 get_inline_xattr_addrs(inode));
1368
1369 phys += offset;
1370 len = inline_xattr_size(inode);
1371
1372 f2fs_put_page(page, 1);
1373
1374 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1375
1376 if (!xnid)
1377 flags |= FIEMAP_EXTENT_LAST;
1378
1379 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1380 if (err || err == 1)
1381 return err;
1382 }
1383
1384 if (xnid) {
1385 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1386 if (!page)
1387 return -ENOMEM;
1388
1389 err = f2fs_get_node_info(sbi, xnid, &ni);
1390 if (err) {
1391 f2fs_put_page(page, 1);
1392 return err;
1393 }
1394
1395 phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1396 len = inode->i_sb->s_blocksize;
1397
1398 f2fs_put_page(page, 1);
1399
1400 flags = FIEMAP_EXTENT_LAST;
1401 }
1402
1403 if (phys)
1404 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1405
1406 return (err < 0 ? err : 0);
1407 }
1408
1409 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1410 u64 start, u64 len)
1411 {
1412 struct buffer_head map_bh;
1413 sector_t start_blk, last_blk;
1414 pgoff_t next_pgofs;
1415 u64 logical = 0, phys = 0, size = 0;
1416 u32 flags = 0;
1417 int ret = 0;
1418
1419 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1420 ret = f2fs_precache_extents(inode);
1421 if (ret)
1422 return ret;
1423 }
1424
1425 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1426 if (ret)
1427 return ret;
1428
1429 inode_lock(inode);
1430
1431 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1432 ret = f2fs_xattr_fiemap(inode, fieinfo);
1433 goto out;
1434 }
1435
1436 if (f2fs_has_inline_data(inode)) {
1437 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1438 if (ret != -EAGAIN)
1439 goto out;
1440 }
1441
1442 if (logical_to_blk(inode, len) == 0)
1443 len = blk_to_logical(inode, 1);
1444
1445 start_blk = logical_to_blk(inode, start);
1446 last_blk = logical_to_blk(inode, start + len - 1);
1447
1448 next:
1449 memset(&map_bh, 0, sizeof(struct buffer_head));
1450 map_bh.b_size = len;
1451
1452 ret = get_data_block(inode, start_blk, &map_bh, 0,
1453 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1454 if (ret)
1455 goto out;
1456
1457 /* HOLE */
1458 if (!buffer_mapped(&map_bh)) {
1459 start_blk = next_pgofs;
1460
1461 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1462 F2FS_I_SB(inode)->max_file_blocks))
1463 goto prep_next;
1464
1465 flags |= FIEMAP_EXTENT_LAST;
1466 }
1467
1468 if (size) {
1469 if (f2fs_encrypted_inode(inode))
1470 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1471
1472 ret = fiemap_fill_next_extent(fieinfo, logical,
1473 phys, size, flags);
1474 }
1475
1476 if (start_blk > last_blk || ret)
1477 goto out;
1478
1479 logical = blk_to_logical(inode, start_blk);
1480 phys = blk_to_logical(inode, map_bh.b_blocknr);
1481 size = map_bh.b_size;
1482 flags = 0;
1483 if (buffer_unwritten(&map_bh))
1484 flags = FIEMAP_EXTENT_UNWRITTEN;
1485
1486 start_blk += logical_to_blk(inode, size);
1487
1488 prep_next:
1489 cond_resched();
1490 if (fatal_signal_pending(current))
1491 ret = -EINTR;
1492 else
1493 goto next;
1494 out:
1495 if (ret == 1)
1496 ret = 0;
1497
1498 inode_unlock(inode);
1499 return ret;
1500 }
1501
1502 /*
1503 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1504 * Major change was from block_size == page_size in f2fs by default.
1505 *
1506 * Note that the aops->readpages() function is ONLY used for read-ahead. If
1507 * this function ever deviates from doing just read-ahead, it should either
1508 * use ->readpage() or do the necessary surgery to decouple ->readpages()
1509 * from read-ahead.
1510 */
1511 static int f2fs_mpage_readpages(struct address_space *mapping,
1512 struct list_head *pages, struct page *page,
1513 unsigned nr_pages, bool is_readahead)
1514 {
1515 struct bio *bio = NULL;
1516 sector_t last_block_in_bio = 0;
1517 struct inode *inode = mapping->host;
1518 const unsigned blkbits = inode->i_blkbits;
1519 const unsigned blocksize = 1 << blkbits;
1520 sector_t block_in_file;
1521 sector_t last_block;
1522 sector_t last_block_in_file;
1523 sector_t block_nr;
1524 struct f2fs_map_blocks map;
1525
1526 map.m_pblk = 0;
1527 map.m_lblk = 0;
1528 map.m_len = 0;
1529 map.m_flags = 0;
1530 map.m_next_pgofs = NULL;
1531 map.m_next_extent = NULL;
1532 map.m_seg_type = NO_CHECK_TYPE;
1533 map.m_may_create = false;
1534
1535 for (; nr_pages; nr_pages--) {
1536 if (pages) {
1537 page = list_last_entry(pages, struct page, lru);
1538
1539 prefetchw(&page->flags);
1540 list_del(&page->lru);
1541 if (add_to_page_cache_lru(page, mapping,
1542 page->index,
1543 readahead_gfp_mask(mapping)))
1544 goto next_page;
1545 }
1546
1547 block_in_file = (sector_t)page->index;
1548 last_block = block_in_file + nr_pages;
1549 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1550 blkbits;
1551 if (last_block > last_block_in_file)
1552 last_block = last_block_in_file;
1553
1554 /*
1555 * Map blocks using the previous result first.
1556 */
1557 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1558 block_in_file > map.m_lblk &&
1559 block_in_file < (map.m_lblk + map.m_len))
1560 goto got_it;
1561
1562 /*
1563 * Then do more f2fs_map_blocks() calls until we are
1564 * done with this page.
1565 */
1566 map.m_flags = 0;
1567
1568 if (block_in_file < last_block) {
1569 map.m_lblk = block_in_file;
1570 map.m_len = last_block - block_in_file;
1571
1572 if (f2fs_map_blocks(inode, &map, 0,
1573 F2FS_GET_BLOCK_DEFAULT))
1574 goto set_error_page;
1575 }
1576 got_it:
1577 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1578 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1579 SetPageMappedToDisk(page);
1580
1581 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1582 SetPageUptodate(page);
1583 goto confused;
1584 }
1585
1586 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
1587 DATA_GENERIC))
1588 goto set_error_page;
1589 } else {
1590 zero_user_segment(page, 0, PAGE_SIZE);
1591 if (!PageUptodate(page))
1592 SetPageUptodate(page);
1593 unlock_page(page);
1594 goto next_page;
1595 }
1596
1597 /*
1598 * This page will go to BIO. Do we need to send this
1599 * BIO off first?
1600 */
1601 if (bio && (last_block_in_bio != block_nr - 1 ||
1602 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1603 submit_and_realloc:
1604 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1605 bio = NULL;
1606 }
1607 if (bio == NULL) {
1608 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
1609 is_readahead ? REQ_RAHEAD : 0);
1610 if (IS_ERR(bio)) {
1611 bio = NULL;
1612 goto set_error_page;
1613 }
1614 }
1615
1616 /*
1617 * If the page is under writeback, we need to wait for
1618 * its completion to see the correct decrypted data.
1619 */
1620 f2fs_wait_on_block_writeback(inode, block_nr);
1621
1622 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1623 goto submit_and_realloc;
1624
1625 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
1626 ClearPageError(page);
1627 last_block_in_bio = block_nr;
1628 goto next_page;
1629 set_error_page:
1630 SetPageError(page);
1631 zero_user_segment(page, 0, PAGE_SIZE);
1632 unlock_page(page);
1633 goto next_page;
1634 confused:
1635 if (bio) {
1636 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1637 bio = NULL;
1638 }
1639 unlock_page(page);
1640 next_page:
1641 if (pages)
1642 put_page(page);
1643 }
1644 BUG_ON(pages && !list_empty(pages));
1645 if (bio)
1646 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1647 return 0;
1648 }
1649
1650 static int f2fs_read_data_page(struct file *file, struct page *page)
1651 {
1652 struct inode *inode = page->mapping->host;
1653 int ret = -EAGAIN;
1654
1655 trace_f2fs_readpage(page, DATA);
1656
1657 /* If the file has inline data, try to read it directly */
1658 if (f2fs_has_inline_data(inode))
1659 ret = f2fs_read_inline_data(inode, page);
1660 if (ret == -EAGAIN)
1661 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1, false);
1662 return ret;
1663 }
1664
1665 static int f2fs_read_data_pages(struct file *file,
1666 struct address_space *mapping,
1667 struct list_head *pages, unsigned nr_pages)
1668 {
1669 struct inode *inode = mapping->host;
1670 struct page *page = list_last_entry(pages, struct page, lru);
1671
1672 trace_f2fs_readpages(inode, page, nr_pages);
1673
1674 /* If the file has inline data, skip readpages */
1675 if (f2fs_has_inline_data(inode))
1676 return 0;
1677
1678 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages, true);
1679 }
1680
1681 static int encrypt_one_page(struct f2fs_io_info *fio)
1682 {
1683 struct inode *inode = fio->page->mapping->host;
1684 struct page *mpage;
1685 gfp_t gfp_flags = GFP_NOFS;
1686
1687 if (!f2fs_encrypted_file(inode))
1688 return 0;
1689
1690 /* wait for GCed page writeback via META_MAPPING */
1691 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
1692
1693 retry_encrypt:
1694 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1695 PAGE_SIZE, 0, fio->page->index, gfp_flags);
1696 if (IS_ERR(fio->encrypted_page)) {
1697 /* flush pending IOs and wait for a while in the ENOMEM case */
1698 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1699 f2fs_flush_merged_writes(fio->sbi);
1700 congestion_wait(BLK_RW_ASYNC, HZ/50);
1701 gfp_flags |= __GFP_NOFAIL;
1702 goto retry_encrypt;
1703 }
1704 return PTR_ERR(fio->encrypted_page);
1705 }
1706
1707 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
1708 if (mpage) {
1709 if (PageUptodate(mpage))
1710 memcpy(page_address(mpage),
1711 page_address(fio->encrypted_page), PAGE_SIZE);
1712 f2fs_put_page(mpage, 1);
1713 }
1714 return 0;
1715 }
1716
1717 static inline bool check_inplace_update_policy(struct inode *inode,
1718 struct f2fs_io_info *fio)
1719 {
1720 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1721 unsigned int policy = SM_I(sbi)->ipu_policy;
1722
1723 if (policy & (0x1 << F2FS_IPU_FORCE))
1724 return true;
1725 if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
1726 return true;
1727 if (policy & (0x1 << F2FS_IPU_UTIL) &&
1728 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1729 return true;
1730 if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && f2fs_need_SSR(sbi) &&
1731 utilization(sbi) > SM_I(sbi)->min_ipu_util)
1732 return true;
1733
1734 /*
1735 * IPU for rewrite async pages
1736 */
1737 if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1738 fio && fio->op == REQ_OP_WRITE &&
1739 !(fio->op_flags & REQ_SYNC) &&
1740 !f2fs_encrypted_inode(inode))
1741 return true;
1742
1743 /* this is only set during fdatasync */
1744 if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1745 is_inode_flag_set(inode, FI_NEED_IPU))
1746 return true;
1747
1748 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1749 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1750 return true;
1751
1752 return false;
1753 }
1754
1755 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1756 {
1757 if (f2fs_is_pinned_file(inode))
1758 return true;
1759
1760 /* if this is cold file, we should overwrite to avoid fragmentation */
1761 if (file_is_cold(inode))
1762 return true;
1763
1764 return check_inplace_update_policy(inode, fio);
1765 }
1766
1767 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1768 {
1769 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1770
1771 if (test_opt(sbi, LFS))
1772 return true;
1773 if (S_ISDIR(inode->i_mode))
1774 return true;
1775 if (IS_NOQUOTA(inode))
1776 return true;
1777 if (f2fs_is_atomic_file(inode))
1778 return true;
1779 if (fio) {
1780 if (is_cold_data(fio->page))
1781 return true;
1782 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1783 return true;
1784 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1785 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
1786 return true;
1787 }
1788 return false;
1789 }
1790
1791 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1792 {
1793 struct inode *inode = fio->page->mapping->host;
1794
1795 if (f2fs_should_update_outplace(inode, fio))
1796 return false;
1797
1798 return f2fs_should_update_inplace(inode, fio);
1799 }
1800
1801 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
1802 {
1803 struct page *page = fio->page;
1804 struct inode *inode = page->mapping->host;
1805 struct dnode_of_data dn;
1806 struct extent_info ei = {0,0,0};
1807 struct node_info ni;
1808 bool ipu_force = false;
1809 int err = 0;
1810
1811 set_new_dnode(&dn, inode, NULL, NULL, 0);
1812 if (need_inplace_update(fio) &&
1813 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1814 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1815
1816 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1817 DATA_GENERIC))
1818 return -EFAULT;
1819
1820 ipu_force = true;
1821 fio->need_lock = LOCK_DONE;
1822 goto got_it;
1823 }
1824
1825 /* Deadlock due to between page->lock and f2fs_lock_op */
1826 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1827 return -EAGAIN;
1828
1829 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1830 if (err)
1831 goto out;
1832
1833 fio->old_blkaddr = dn.data_blkaddr;
1834
1835 /* This page is already truncated */
1836 if (fio->old_blkaddr == NULL_ADDR) {
1837 ClearPageUptodate(page);
1838 clear_cold_data(page);
1839 goto out_writepage;
1840 }
1841 got_it:
1842 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
1843 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
1844 DATA_GENERIC)) {
1845 err = -EFAULT;
1846 goto out_writepage;
1847 }
1848 /*
1849 * If current allocation needs SSR,
1850 * it had better in-place writes for updated data.
1851 */
1852 if (ipu_force || (is_valid_data_blkaddr(fio->sbi, fio->old_blkaddr) &&
1853 need_inplace_update(fio))) {
1854 err = encrypt_one_page(fio);
1855 if (err)
1856 goto out_writepage;
1857
1858 set_page_writeback(page);
1859 ClearPageError(page);
1860 f2fs_put_dnode(&dn);
1861 if (fio->need_lock == LOCK_REQ)
1862 f2fs_unlock_op(fio->sbi);
1863 err = f2fs_inplace_write_data(fio);
1864 if (err) {
1865 if (f2fs_encrypted_file(inode))
1866 fscrypt_pullback_bio_page(&fio->encrypted_page,
1867 true);
1868 if (PageWriteback(page))
1869 end_page_writeback(page);
1870 }
1871 trace_f2fs_do_write_data_page(fio->page, IPU);
1872 set_inode_flag(inode, FI_UPDATE_WRITE);
1873 return err;
1874 }
1875
1876 if (fio->need_lock == LOCK_RETRY) {
1877 if (!f2fs_trylock_op(fio->sbi)) {
1878 err = -EAGAIN;
1879 goto out_writepage;
1880 }
1881 fio->need_lock = LOCK_REQ;
1882 }
1883
1884 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni);
1885 if (err)
1886 goto out_writepage;
1887
1888 fio->version = ni.version;
1889
1890 err = encrypt_one_page(fio);
1891 if (err)
1892 goto out_writepage;
1893
1894 set_page_writeback(page);
1895 ClearPageError(page);
1896
1897 /* LFS mode write path */
1898 f2fs_outplace_write_data(&dn, fio);
1899 trace_f2fs_do_write_data_page(page, OPU);
1900 set_inode_flag(inode, FI_APPEND_WRITE);
1901 if (page->index == 0)
1902 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1903 out_writepage:
1904 f2fs_put_dnode(&dn);
1905 out:
1906 if (fio->need_lock == LOCK_REQ)
1907 f2fs_unlock_op(fio->sbi);
1908 return err;
1909 }
1910
1911 static int __write_data_page(struct page *page, bool *submitted,
1912 struct writeback_control *wbc,
1913 enum iostat_type io_type)
1914 {
1915 struct inode *inode = page->mapping->host;
1916 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1917 loff_t i_size = i_size_read(inode);
1918 const pgoff_t end_index = ((unsigned long long) i_size)
1919 >> PAGE_SHIFT;
1920 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1921 unsigned offset = 0;
1922 bool need_balance_fs = false;
1923 int err = 0;
1924 struct f2fs_io_info fio = {
1925 .sbi = sbi,
1926 .ino = inode->i_ino,
1927 .type = DATA,
1928 .op = REQ_OP_WRITE,
1929 .op_flags = wbc_to_write_flags(wbc),
1930 .old_blkaddr = NULL_ADDR,
1931 .page = page,
1932 .encrypted_page = NULL,
1933 .submitted = false,
1934 .need_lock = LOCK_RETRY,
1935 .io_type = io_type,
1936 .io_wbc = wbc,
1937 };
1938
1939 trace_f2fs_writepage(page, DATA);
1940
1941 /* we should bypass data pages to proceed the kworkder jobs */
1942 if (unlikely(f2fs_cp_error(sbi))) {
1943 mapping_set_error(page->mapping, -EIO);
1944 /*
1945 * don't drop any dirty dentry pages for keeping lastest
1946 * directory structure.
1947 */
1948 if (S_ISDIR(inode->i_mode))
1949 goto redirty_out;
1950 goto out;
1951 }
1952
1953 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1954 goto redirty_out;
1955
1956 if (page->index < end_index)
1957 goto write;
1958
1959 /*
1960 * If the offset is out-of-range of file size,
1961 * this page does not have to be written to disk.
1962 */
1963 offset = i_size & (PAGE_SIZE - 1);
1964 if ((page->index >= end_index + 1) || !offset)
1965 goto out;
1966
1967 zero_user_segment(page, offset, PAGE_SIZE);
1968 write:
1969 if (f2fs_is_drop_cache(inode))
1970 goto out;
1971 /* we should not write 0'th page having journal header */
1972 if (f2fs_is_volatile_file(inode) && (!page->index ||
1973 (!wbc->for_reclaim &&
1974 f2fs_available_free_memory(sbi, BASE_CHECK))))
1975 goto redirty_out;
1976
1977 /* Dentry blocks are controlled by checkpoint */
1978 if (S_ISDIR(inode->i_mode)) {
1979 fio.need_lock = LOCK_DONE;
1980 err = f2fs_do_write_data_page(&fio);
1981 goto done;
1982 }
1983
1984 if (!wbc->for_reclaim)
1985 need_balance_fs = true;
1986 else if (has_not_enough_free_secs(sbi, 0, 0))
1987 goto redirty_out;
1988 else
1989 set_inode_flag(inode, FI_HOT_DATA);
1990
1991 err = -EAGAIN;
1992 if (f2fs_has_inline_data(inode)) {
1993 err = f2fs_write_inline_data(inode, page);
1994 if (!err)
1995 goto out;
1996 }
1997
1998 if (err == -EAGAIN) {
1999 err = f2fs_do_write_data_page(&fio);
2000 if (err == -EAGAIN) {
2001 fio.need_lock = LOCK_REQ;
2002 err = f2fs_do_write_data_page(&fio);
2003 }
2004 }
2005
2006 if (err) {
2007 file_set_keep_isize(inode);
2008 } else {
2009 down_write(&F2FS_I(inode)->i_sem);
2010 if (F2FS_I(inode)->last_disk_size < psize)
2011 F2FS_I(inode)->last_disk_size = psize;
2012 up_write(&F2FS_I(inode)->i_sem);
2013 }
2014
2015 done:
2016 if (err && err != -ENOENT)
2017 goto redirty_out;
2018
2019 out:
2020 inode_dec_dirty_pages(inode);
2021 if (err) {
2022 ClearPageUptodate(page);
2023 clear_cold_data(page);
2024 }
2025
2026 if (wbc->for_reclaim) {
2027 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2028 clear_inode_flag(inode, FI_HOT_DATA);
2029 f2fs_remove_dirty_inode(inode);
2030 submitted = NULL;
2031 }
2032
2033 unlock_page(page);
2034 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode))
2035 f2fs_balance_fs(sbi, need_balance_fs);
2036
2037 if (unlikely(f2fs_cp_error(sbi))) {
2038 f2fs_submit_merged_write(sbi, DATA);
2039 submitted = NULL;
2040 }
2041
2042 if (submitted)
2043 *submitted = fio.submitted;
2044
2045 return 0;
2046
2047 redirty_out:
2048 redirty_page_for_writepage(wbc, page);
2049 /*
2050 * pageout() in MM traslates EAGAIN, so calls handle_write_error()
2051 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2052 * file_write_and_wait_range() will see EIO error, which is critical
2053 * to return value of fsync() followed by atomic_write failure to user.
2054 */
2055 if (!err || wbc->for_reclaim)
2056 return AOP_WRITEPAGE_ACTIVATE;
2057 unlock_page(page);
2058 return err;
2059 }
2060
2061 static int f2fs_write_data_page(struct page *page,
2062 struct writeback_control *wbc)
2063 {
2064 return __write_data_page(page, NULL, wbc, FS_DATA_IO);
2065 }
2066
2067 /*
2068 * This function was copied from write_cche_pages from mm/page-writeback.c.
2069 * The major change is making write step of cold data page separately from
2070 * warm/hot data page.
2071 */
2072 static int f2fs_write_cache_pages(struct address_space *mapping,
2073 struct writeback_control *wbc,
2074 enum iostat_type io_type)
2075 {
2076 int ret = 0;
2077 int done = 0;
2078 struct pagevec pvec;
2079 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2080 int nr_pages;
2081 pgoff_t uninitialized_var(writeback_index);
2082 pgoff_t index;
2083 pgoff_t end; /* Inclusive */
2084 pgoff_t done_index;
2085 int cycled;
2086 int range_whole = 0;
2087 xa_mark_t tag;
2088 int nwritten = 0;
2089
2090 pagevec_init(&pvec);
2091
2092 if (get_dirty_pages(mapping->host) <=
2093 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2094 set_inode_flag(mapping->host, FI_HOT_DATA);
2095 else
2096 clear_inode_flag(mapping->host, FI_HOT_DATA);
2097
2098 if (wbc->range_cyclic) {
2099 writeback_index = mapping->writeback_index; /* prev offset */
2100 index = writeback_index;
2101 if (index == 0)
2102 cycled = 1;
2103 else
2104 cycled = 0;
2105 end = -1;
2106 } else {
2107 index = wbc->range_start >> PAGE_SHIFT;
2108 end = wbc->range_end >> PAGE_SHIFT;
2109 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2110 range_whole = 1;
2111 cycled = 1; /* ignore range_cyclic tests */
2112 }
2113 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2114 tag = PAGECACHE_TAG_TOWRITE;
2115 else
2116 tag = PAGECACHE_TAG_DIRTY;
2117 retry:
2118 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2119 tag_pages_for_writeback(mapping, index, end);
2120 done_index = index;
2121 while (!done && (index <= end)) {
2122 int i;
2123
2124 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2125 tag);
2126 if (nr_pages == 0)
2127 break;
2128
2129 for (i = 0; i < nr_pages; i++) {
2130 struct page *page = pvec.pages[i];
2131 bool submitted = false;
2132
2133 /* give a priority to WB_SYNC threads */
2134 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
2135 wbc->sync_mode == WB_SYNC_NONE) {
2136 done = 1;
2137 break;
2138 }
2139
2140 done_index = page->index;
2141 retry_write:
2142 lock_page(page);
2143
2144 if (unlikely(page->mapping != mapping)) {
2145 continue_unlock:
2146 unlock_page(page);
2147 continue;
2148 }
2149
2150 if (!PageDirty(page)) {
2151 /* someone wrote it for us */
2152 goto continue_unlock;
2153 }
2154
2155 if (PageWriteback(page)) {
2156 if (wbc->sync_mode != WB_SYNC_NONE)
2157 f2fs_wait_on_page_writeback(page,
2158 DATA, true, true);
2159 else
2160 goto continue_unlock;
2161 }
2162
2163 if (!clear_page_dirty_for_io(page))
2164 goto continue_unlock;
2165
2166 ret = __write_data_page(page, &submitted, wbc, io_type);
2167 if (unlikely(ret)) {
2168 /*
2169 * keep nr_to_write, since vfs uses this to
2170 * get # of written pages.
2171 */
2172 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2173 unlock_page(page);
2174 ret = 0;
2175 continue;
2176 } else if (ret == -EAGAIN) {
2177 ret = 0;
2178 if (wbc->sync_mode == WB_SYNC_ALL) {
2179 cond_resched();
2180 congestion_wait(BLK_RW_ASYNC,
2181 HZ/50);
2182 goto retry_write;
2183 }
2184 continue;
2185 }
2186 done_index = page->index + 1;
2187 done = 1;
2188 break;
2189 } else if (submitted) {
2190 nwritten++;
2191 }
2192
2193 if (--wbc->nr_to_write <= 0 &&
2194 wbc->sync_mode == WB_SYNC_NONE) {
2195 done = 1;
2196 break;
2197 }
2198 }
2199 pagevec_release(&pvec);
2200 cond_resched();
2201 }
2202
2203 if (!cycled && !done) {
2204 cycled = 1;
2205 index = 0;
2206 end = writeback_index - 1;
2207 goto retry;
2208 }
2209 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2210 mapping->writeback_index = done_index;
2211
2212 if (nwritten)
2213 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
2214 NULL, 0, DATA);
2215
2216 return ret;
2217 }
2218
2219 static inline bool __should_serialize_io(struct inode *inode,
2220 struct writeback_control *wbc)
2221 {
2222 if (!S_ISREG(inode->i_mode))
2223 return false;
2224 if (IS_NOQUOTA(inode))
2225 return false;
2226 if (wbc->sync_mode != WB_SYNC_ALL)
2227 return true;
2228 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
2229 return true;
2230 return false;
2231 }
2232
2233 static int __f2fs_write_data_pages(struct address_space *mapping,
2234 struct writeback_control *wbc,
2235 enum iostat_type io_type)
2236 {
2237 struct inode *inode = mapping->host;
2238 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2239 struct blk_plug plug;
2240 int ret;
2241 bool locked = false;
2242
2243 /* deal with chardevs and other special file */
2244 if (!mapping->a_ops->writepage)
2245 return 0;
2246
2247 /* skip writing if there is no dirty page in this inode */
2248 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2249 return 0;
2250
2251 /* during POR, we don't need to trigger writepage at all. */
2252 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2253 goto skip_write;
2254
2255 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
2256 wbc->sync_mode == WB_SYNC_NONE &&
2257 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2258 f2fs_available_free_memory(sbi, DIRTY_DENTS))
2259 goto skip_write;
2260
2261 /* skip writing during file defragment */
2262 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2263 goto skip_write;
2264
2265 trace_f2fs_writepages(mapping->host, wbc, DATA);
2266
2267 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2268 if (wbc->sync_mode == WB_SYNC_ALL)
2269 atomic_inc(&sbi->wb_sync_req[DATA]);
2270 else if (atomic_read(&sbi->wb_sync_req[DATA]))
2271 goto skip_write;
2272
2273 if (__should_serialize_io(inode, wbc)) {
2274 mutex_lock(&sbi->writepages);
2275 locked = true;
2276 }
2277
2278 blk_start_plug(&plug);
2279 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2280 blk_finish_plug(&plug);
2281
2282 if (locked)
2283 mutex_unlock(&sbi->writepages);
2284
2285 if (wbc->sync_mode == WB_SYNC_ALL)
2286 atomic_dec(&sbi->wb_sync_req[DATA]);
2287 /*
2288 * if some pages were truncated, we cannot guarantee its mapping->host
2289 * to detect pending bios.
2290 */
2291
2292 f2fs_remove_dirty_inode(inode);
2293 return ret;
2294
2295 skip_write:
2296 wbc->pages_skipped += get_dirty_pages(inode);
2297 trace_f2fs_writepages(mapping->host, wbc, DATA);
2298 return 0;
2299 }
2300
2301 static int f2fs_write_data_pages(struct address_space *mapping,
2302 struct writeback_control *wbc)
2303 {
2304 struct inode *inode = mapping->host;
2305
2306 return __f2fs_write_data_pages(mapping, wbc,
2307 F2FS_I(inode)->cp_task == current ?
2308 FS_CP_DATA_IO : FS_DATA_IO);
2309 }
2310
2311 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2312 {
2313 struct inode *inode = mapping->host;
2314 loff_t i_size = i_size_read(inode);
2315
2316 if (to > i_size) {
2317 down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2318 down_write(&F2FS_I(inode)->i_mmap_sem);
2319
2320 truncate_pagecache(inode, i_size);
2321 if (!IS_NOQUOTA(inode))
2322 f2fs_truncate_blocks(inode, i_size, true);
2323
2324 up_write(&F2FS_I(inode)->i_mmap_sem);
2325 up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
2326 }
2327 }
2328
2329 static int prepare_write_begin(struct f2fs_sb_info *sbi,
2330 struct page *page, loff_t pos, unsigned len,
2331 block_t *blk_addr, bool *node_changed)
2332 {
2333 struct inode *inode = page->mapping->host;
2334 pgoff_t index = page->index;
2335 struct dnode_of_data dn;
2336 struct page *ipage;
2337 bool locked = false;
2338 struct extent_info ei = {0,0,0};
2339 int err = 0;
2340 int flag;
2341
2342 /*
2343 * we already allocated all the blocks, so we don't need to get
2344 * the block addresses when there is no need to fill the page.
2345 */
2346 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2347 !is_inode_flag_set(inode, FI_NO_PREALLOC))
2348 return 0;
2349
2350 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
2351 if (f2fs_has_inline_data(inode) && pos + len > MAX_INLINE_DATA(inode))
2352 flag = F2FS_GET_BLOCK_DEFAULT;
2353 else
2354 flag = F2FS_GET_BLOCK_PRE_AIO;
2355
2356 if (f2fs_has_inline_data(inode) ||
2357 (pos & PAGE_MASK) >= i_size_read(inode)) {
2358 __do_map_lock(sbi, flag, true);
2359 locked = true;
2360 }
2361 restart:
2362 /* check inline_data */
2363 ipage = f2fs_get_node_page(sbi, inode->i_ino);
2364 if (IS_ERR(ipage)) {
2365 err = PTR_ERR(ipage);
2366 goto unlock_out;
2367 }
2368
2369 set_new_dnode(&dn, inode, ipage, ipage, 0);
2370
2371 if (f2fs_has_inline_data(inode)) {
2372 if (pos + len <= MAX_INLINE_DATA(inode)) {
2373 f2fs_do_read_inline_data(page, ipage);
2374 set_inode_flag(inode, FI_DATA_EXIST);
2375 if (inode->i_nlink)
2376 set_inline_node(ipage);
2377 } else {
2378 err = f2fs_convert_inline_page(&dn, page);
2379 if (err)
2380 goto out;
2381 if (dn.data_blkaddr == NULL_ADDR)
2382 err = f2fs_get_block(&dn, index);
2383 }
2384 } else if (locked) {
2385 err = f2fs_get_block(&dn, index);
2386 } else {
2387 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2388 dn.data_blkaddr = ei.blk + index - ei.fofs;
2389 } else {
2390 /* hole case */
2391 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
2392 if (err || dn.data_blkaddr == NULL_ADDR) {
2393 f2fs_put_dnode(&dn);
2394 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2395 true);
2396 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
2397 locked = true;
2398 goto restart;
2399 }
2400 }
2401 }
2402
2403 /* convert_inline_page can make node_changed */
2404 *blk_addr = dn.data_blkaddr;
2405 *node_changed = dn.node_changed;
2406 out:
2407 f2fs_put_dnode(&dn);
2408 unlock_out:
2409 if (locked)
2410 __do_map_lock(sbi, flag, false);
2411 return err;
2412 }
2413
2414 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2415 loff_t pos, unsigned len, unsigned flags,
2416 struct page **pagep, void **fsdata)
2417 {
2418 struct inode *inode = mapping->host;
2419 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2420 struct page *page = NULL;
2421 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2422 bool need_balance = false, drop_atomic = false;
2423 block_t blkaddr = NULL_ADDR;
2424 int err = 0;
2425
2426 trace_f2fs_write_begin(inode, pos, len, flags);
2427
2428 err = f2fs_is_checkpoint_ready(sbi);
2429 if (err)
2430 goto fail;
2431
2432 if ((f2fs_is_atomic_file(inode) &&
2433 !f2fs_available_free_memory(sbi, INMEM_PAGES)) ||
2434 is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) {
2435 err = -ENOMEM;
2436 drop_atomic = true;
2437 goto fail;
2438 }
2439
2440 /*
2441 * We should check this at this moment to avoid deadlock on inode page
2442 * and #0 page. The locking rule for inline_data conversion should be:
2443 * lock_page(page #0) -> lock_page(inode_page)
2444 */
2445 if (index != 0) {
2446 err = f2fs_convert_inline_inode(inode);
2447 if (err)
2448 goto fail;
2449 }
2450 repeat:
2451 /*
2452 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2453 * wait_for_stable_page. Will wait that below with our IO control.
2454 */
2455 page = f2fs_pagecache_get_page(mapping, index,
2456 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2457 if (!page) {
2458 err = -ENOMEM;
2459 goto fail;
2460 }
2461
2462 *pagep = page;
2463
2464 err = prepare_write_begin(sbi, page, pos, len,
2465 &blkaddr, &need_balance);
2466 if (err)
2467 goto fail;
2468
2469 if (need_balance && !IS_NOQUOTA(inode) &&
2470 has_not_enough_free_secs(sbi, 0, 0)) {
2471 unlock_page(page);
2472 f2fs_balance_fs(sbi, true);
2473 lock_page(page);
2474 if (page->mapping != mapping) {
2475 /* The page got truncated from under us */
2476 f2fs_put_page(page, 1);
2477 goto repeat;
2478 }
2479 }
2480
2481 f2fs_wait_on_page_writeback(page, DATA, false, true);
2482
2483 if (len == PAGE_SIZE || PageUptodate(page))
2484 return 0;
2485
2486 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2487 zero_user_segment(page, len, PAGE_SIZE);
2488 return 0;
2489 }
2490
2491 if (blkaddr == NEW_ADDR) {
2492 zero_user_segment(page, 0, PAGE_SIZE);
2493 SetPageUptodate(page);
2494 } else {
2495 err = f2fs_submit_page_read(inode, page, blkaddr);
2496 if (err)
2497 goto fail;
2498
2499 lock_page(page);
2500 if (unlikely(page->mapping != mapping)) {
2501 f2fs_put_page(page, 1);
2502 goto repeat;
2503 }
2504 if (unlikely(!PageUptodate(page))) {
2505 err = -EIO;
2506 goto fail;
2507 }
2508 }
2509 return 0;
2510
2511 fail:
2512 f2fs_put_page(page, 1);
2513 f2fs_write_failed(mapping, pos + len);
2514 if (drop_atomic)
2515 f2fs_drop_inmem_pages_all(sbi, false);
2516 return err;
2517 }
2518
2519 static int f2fs_write_end(struct file *file,
2520 struct address_space *mapping,
2521 loff_t pos, unsigned len, unsigned copied,
2522 struct page *page, void *fsdata)
2523 {
2524 struct inode *inode = page->mapping->host;
2525
2526 trace_f2fs_write_end(inode, pos, len, copied);
2527
2528 /*
2529 * This should be come from len == PAGE_SIZE, and we expect copied
2530 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2531 * let generic_perform_write() try to copy data again through copied=0.
2532 */
2533 if (!PageUptodate(page)) {
2534 if (unlikely(copied != len))
2535 copied = 0;
2536 else
2537 SetPageUptodate(page);
2538 }
2539 if (!copied)
2540 goto unlock_out;
2541
2542 set_page_dirty(page);
2543
2544 if (pos + copied > i_size_read(inode))
2545 f2fs_i_size_write(inode, pos + copied);
2546 unlock_out:
2547 f2fs_put_page(page, 1);
2548 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2549 return copied;
2550 }
2551
2552 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2553 loff_t offset)
2554 {
2555 unsigned i_blkbits = READ_ONCE(inode->i_blkbits);
2556 unsigned blkbits = i_blkbits;
2557 unsigned blocksize_mask = (1 << blkbits) - 1;
2558 unsigned long align = offset | iov_iter_alignment(iter);
2559 struct block_device *bdev = inode->i_sb->s_bdev;
2560
2561 if (align & blocksize_mask) {
2562 if (bdev)
2563 blkbits = blksize_bits(bdev_logical_block_size(bdev));
2564 blocksize_mask = (1 << blkbits) - 1;
2565 if (align & blocksize_mask)
2566 return -EINVAL;
2567 return 1;
2568 }
2569 return 0;
2570 }
2571
2572 static void f2fs_dio_end_io(struct bio *bio)
2573 {
2574 struct f2fs_private_dio *dio = bio->bi_private;
2575
2576 dec_page_count(F2FS_I_SB(dio->inode),
2577 dio->write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2578
2579 bio->bi_private = dio->orig_private;
2580 bio->bi_end_io = dio->orig_end_io;
2581
2582 kvfree(dio);
2583
2584 bio_endio(bio);
2585 }
2586
2587 static void f2fs_dio_submit_bio(struct bio *bio, struct inode *inode,
2588 loff_t file_offset)
2589 {
2590 struct f2fs_private_dio *dio;
2591 bool write = (bio_op(bio) == REQ_OP_WRITE);
2592
2593 dio = f2fs_kzalloc(F2FS_I_SB(inode),
2594 sizeof(struct f2fs_private_dio), GFP_NOFS);
2595 if (!dio)
2596 goto out;
2597
2598 dio->inode = inode;
2599 dio->orig_end_io = bio->bi_end_io;
2600 dio->orig_private = bio->bi_private;
2601 dio->write = write;
2602
2603 bio->bi_end_io = f2fs_dio_end_io;
2604 bio->bi_private = dio;
2605
2606 inc_page_count(F2FS_I_SB(inode),
2607 write ? F2FS_DIO_WRITE : F2FS_DIO_READ);
2608
2609 submit_bio(bio);
2610 return;
2611 out:
2612 bio->bi_status = BLK_STS_IOERR;
2613 bio_endio(bio);
2614 }
2615
2616 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2617 {
2618 struct address_space *mapping = iocb->ki_filp->f_mapping;
2619 struct inode *inode = mapping->host;
2620 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2621 struct f2fs_inode_info *fi = F2FS_I(inode);
2622 size_t count = iov_iter_count(iter);
2623 loff_t offset = iocb->ki_pos;
2624 int rw = iov_iter_rw(iter);
2625 int err;
2626 enum rw_hint hint = iocb->ki_hint;
2627 int whint_mode = F2FS_OPTION(sbi).whint_mode;
2628 bool do_opu;
2629
2630 err = check_direct_IO(inode, iter, offset);
2631 if (err)
2632 return err < 0 ? err : 0;
2633
2634 if (f2fs_force_buffered_io(inode, iocb, iter))
2635 return 0;
2636
2637 do_opu = allow_outplace_dio(inode, iocb, iter);
2638
2639 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2640
2641 if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2642 iocb->ki_hint = WRITE_LIFE_NOT_SET;
2643
2644 if (iocb->ki_flags & IOCB_NOWAIT) {
2645 if (!down_read_trylock(&fi->i_gc_rwsem[rw])) {
2646 iocb->ki_hint = hint;
2647 err = -EAGAIN;
2648 goto out;
2649 }
2650 if (do_opu && !down_read_trylock(&fi->i_gc_rwsem[READ])) {
2651 up_read(&fi->i_gc_rwsem[rw]);
2652 iocb->ki_hint = hint;
2653 err = -EAGAIN;
2654 goto out;
2655 }
2656 } else {
2657 down_read(&fi->i_gc_rwsem[rw]);
2658 if (do_opu)
2659 down_read(&fi->i_gc_rwsem[READ]);
2660 }
2661
2662 err = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
2663 iter, rw == WRITE ? get_data_block_dio_write :
2664 get_data_block_dio, NULL, f2fs_dio_submit_bio,
2665 DIO_LOCKING | DIO_SKIP_HOLES);
2666
2667 if (do_opu)
2668 up_read(&fi->i_gc_rwsem[READ]);
2669
2670 up_read(&fi->i_gc_rwsem[rw]);
2671
2672 if (rw == WRITE) {
2673 if (whint_mode == WHINT_MODE_OFF)
2674 iocb->ki_hint = hint;
2675 if (err > 0) {
2676 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2677 err);
2678 if (!do_opu)
2679 set_inode_flag(inode, FI_UPDATE_WRITE);
2680 } else if (err < 0) {
2681 f2fs_write_failed(mapping, offset + count);
2682 }
2683 }
2684
2685 out:
2686 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2687
2688 return err;
2689 }
2690
2691 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2692 unsigned int length)
2693 {
2694 struct inode *inode = page->mapping->host;
2695 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2696
2697 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2698 (offset % PAGE_SIZE || length != PAGE_SIZE))
2699 return;
2700
2701 if (PageDirty(page)) {
2702 if (inode->i_ino == F2FS_META_INO(sbi)) {
2703 dec_page_count(sbi, F2FS_DIRTY_META);
2704 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2705 dec_page_count(sbi, F2FS_DIRTY_NODES);
2706 } else {
2707 inode_dec_dirty_pages(inode);
2708 f2fs_remove_dirty_inode(inode);
2709 }
2710 }
2711
2712 clear_cold_data(page);
2713
2714 /* This is atomic written page, keep Private */
2715 if (IS_ATOMIC_WRITTEN_PAGE(page))
2716 return f2fs_drop_inmem_page(inode, page);
2717
2718 set_page_private(page, 0);
2719 ClearPagePrivate(page);
2720 }
2721
2722 int f2fs_release_page(struct page *page, gfp_t wait)
2723 {
2724 /* If this is dirty page, keep PagePrivate */
2725 if (PageDirty(page))
2726 return 0;
2727
2728 /* This is atomic written page, keep Private */
2729 if (IS_ATOMIC_WRITTEN_PAGE(page))
2730 return 0;
2731
2732 clear_cold_data(page);
2733 set_page_private(page, 0);
2734 ClearPagePrivate(page);
2735 return 1;
2736 }
2737
2738 static int f2fs_set_data_page_dirty(struct page *page)
2739 {
2740 struct address_space *mapping = page->mapping;
2741 struct inode *inode = mapping->host;
2742
2743 trace_f2fs_set_page_dirty(page, DATA);
2744
2745 if (!PageUptodate(page))
2746 SetPageUptodate(page);
2747
2748 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2749 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2750 f2fs_register_inmem_page(inode, page);
2751 return 1;
2752 }
2753 /*
2754 * Previously, this page has been registered, we just
2755 * return here.
2756 */
2757 return 0;
2758 }
2759
2760 if (!PageDirty(page)) {
2761 __set_page_dirty_nobuffers(page);
2762 f2fs_update_dirty_page(inode, page);
2763 return 1;
2764 }
2765 return 0;
2766 }
2767
2768 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2769 {
2770 struct inode *inode = mapping->host;
2771
2772 if (f2fs_has_inline_data(inode))
2773 return 0;
2774
2775 /* make sure allocating whole blocks */
2776 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2777 filemap_write_and_wait(mapping);
2778
2779 return generic_block_bmap(mapping, block, get_data_block_bmap);
2780 }
2781
2782 #ifdef CONFIG_MIGRATION
2783 #include <linux/migrate.h>
2784
2785 int f2fs_migrate_page(struct address_space *mapping,
2786 struct page *newpage, struct page *page, enum migrate_mode mode)
2787 {
2788 int rc, extra_count;
2789 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2790 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2791
2792 BUG_ON(PageWriteback(page));
2793
2794 /* migrating an atomic written page is safe with the inmem_lock hold */
2795 if (atomic_written) {
2796 if (mode != MIGRATE_SYNC)
2797 return -EBUSY;
2798 if (!mutex_trylock(&fi->inmem_lock))
2799 return -EAGAIN;
2800 }
2801
2802 /*
2803 * A reference is expected if PagePrivate set when move mapping,
2804 * however F2FS breaks this for maintaining dirty page counts when
2805 * truncating pages. So here adjusting the 'extra_count' make it work.
2806 */
2807 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2808 rc = migrate_page_move_mapping(mapping, newpage,
2809 page, mode, extra_count);
2810 if (rc != MIGRATEPAGE_SUCCESS) {
2811 if (atomic_written)
2812 mutex_unlock(&fi->inmem_lock);
2813 return rc;
2814 }
2815
2816 if (atomic_written) {
2817 struct inmem_pages *cur;
2818 list_for_each_entry(cur, &fi->inmem_pages, list)
2819 if (cur->page == page) {
2820 cur->page = newpage;
2821 break;
2822 }
2823 mutex_unlock(&fi->inmem_lock);
2824 put_page(page);
2825 get_page(newpage);
2826 }
2827
2828 if (PagePrivate(page))
2829 SetPagePrivate(newpage);
2830 set_page_private(newpage, page_private(page));
2831
2832 if (mode != MIGRATE_SYNC_NO_COPY)
2833 migrate_page_copy(newpage, page);
2834 else
2835 migrate_page_states(newpage, page);
2836
2837 return MIGRATEPAGE_SUCCESS;
2838 }
2839 #endif
2840
2841 const struct address_space_operations f2fs_dblock_aops = {
2842 .readpage = f2fs_read_data_page,
2843 .readpages = f2fs_read_data_pages,
2844 .writepage = f2fs_write_data_page,
2845 .writepages = f2fs_write_data_pages,
2846 .write_begin = f2fs_write_begin,
2847 .write_end = f2fs_write_end,
2848 .set_page_dirty = f2fs_set_data_page_dirty,
2849 .invalidatepage = f2fs_invalidate_page,
2850 .releasepage = f2fs_release_page,
2851 .direct_IO = f2fs_direct_IO,
2852 .bmap = f2fs_bmap,
2853 #ifdef CONFIG_MIGRATION
2854 .migratepage = f2fs_migrate_page,
2855 #endif
2856 };
2857
2858 void f2fs_clear_page_cache_dirty_tag(struct page *page)
2859 {
2860 struct address_space *mapping = page_mapping(page);
2861 unsigned long flags;
2862
2863 xa_lock_irqsave(&mapping->i_pages, flags);
2864 __xa_clear_mark(&mapping->i_pages, page_index(page),
2865 PAGECACHE_TAG_DIRTY);
2866 xa_unlock_irqrestore(&mapping->i_pages, flags);
2867 }
2868
2869 int __init f2fs_init_post_read_processing(void)
2870 {
2871 bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, 0);
2872 if (!bio_post_read_ctx_cache)
2873 goto fail;
2874 bio_post_read_ctx_pool =
2875 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
2876 bio_post_read_ctx_cache);
2877 if (!bio_post_read_ctx_pool)
2878 goto fail_free_cache;
2879 return 0;
2880
2881 fail_free_cache:
2882 kmem_cache_destroy(bio_post_read_ctx_cache);
2883 fail:
2884 return -ENOMEM;
2885 }
2886
2887 void __exit f2fs_destroy_post_read_processing(void)
2888 {
2889 mempool_destroy(bio_post_read_ctx_pool);
2890 kmem_cache_destroy(bio_post_read_ctx_cache);
2891 }
Mirror https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git