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/sched/mm.h>
12 #include <linux/mpage.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include <linux/blkdev.h>
16 #include <linux/bio.h>
17 #include <linux/blk-crypto.h>
18 #include <linux/swap.h>
19 #include <linux/prefetch.h>
20 #include <linux/uio.h>
21 #include <linux/sched/signal.h>
22 #include <linux/fiemap.h>
23 #include <linux/iomap.h>
24
25 #include "f2fs.h"
26 #include "node.h"
27 #include "segment.h"
28 #include "iostat.h"
29 #include <trace/events/f2fs.h>
30
31 #define NUM_PREALLOC_POST_READ_CTXS 128
32
33 static struct kmem_cache *bio_post_read_ctx_cache;
34 static struct kmem_cache *bio_entry_slab;
35 static mempool_t *bio_post_read_ctx_pool;
36 static struct bio_set f2fs_bioset;
37
38 #define F2FS_BIO_POOL_SIZE NR_CURSEG_TYPE
39
f2fs_init_bioset(void)40 int __init f2fs_init_bioset(void)
41 {
42 return bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
43 0, BIOSET_NEED_BVECS);
44 }
45
f2fs_destroy_bioset(void)46 void f2fs_destroy_bioset(void)
47 {
48 bioset_exit(&f2fs_bioset);
49 }
50
f2fs_is_cp_guaranteed(struct page * page)51 bool f2fs_is_cp_guaranteed(struct page *page)
52 {
53 struct address_space *mapping = page->mapping;
54 struct inode *inode;
55 struct f2fs_sb_info *sbi;
56
57 if (!mapping)
58 return false;
59
60 inode = mapping->host;
61 sbi = F2FS_I_SB(inode);
62
63 if (inode->i_ino == F2FS_META_INO(sbi) ||
64 inode->i_ino == F2FS_NODE_INO(sbi) ||
65 S_ISDIR(inode->i_mode))
66 return true;
67
68 if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) ||
69 page_private_gcing(page))
70 return true;
71 return false;
72 }
73
__read_io_type(struct page * page)74 static enum count_type __read_io_type(struct page *page)
75 {
76 struct address_space *mapping = page_file_mapping(page);
77
78 if (mapping) {
79 struct inode *inode = mapping->host;
80 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
81
82 if (inode->i_ino == F2FS_META_INO(sbi))
83 return F2FS_RD_META;
84
85 if (inode->i_ino == F2FS_NODE_INO(sbi))
86 return F2FS_RD_NODE;
87 }
88 return F2FS_RD_DATA;
89 }
90
91 /* postprocessing steps for read bios */
92 enum bio_post_read_step {
93 #ifdef CONFIG_FS_ENCRYPTION
94 STEP_DECRYPT = BIT(0),
95 #else
96 STEP_DECRYPT = 0, /* compile out the decryption-related code */
97 #endif
98 #ifdef CONFIG_F2FS_FS_COMPRESSION
99 STEP_DECOMPRESS = BIT(1),
100 #else
101 STEP_DECOMPRESS = 0, /* compile out the decompression-related code */
102 #endif
103 #ifdef CONFIG_FS_VERITY
104 STEP_VERITY = BIT(2),
105 #else
106 STEP_VERITY = 0, /* compile out the verity-related code */
107 #endif
108 };
109
110 struct bio_post_read_ctx {
111 struct bio *bio;
112 struct f2fs_sb_info *sbi;
113 struct work_struct work;
114 unsigned int enabled_steps;
115 /*
116 * decompression_attempted keeps track of whether
117 * f2fs_end_read_compressed_page() has been called on the pages in the
118 * bio that belong to a compressed cluster yet.
119 */
120 bool decompression_attempted;
121 block_t fs_blkaddr;
122 };
123
124 /*
125 * Update and unlock a bio's pages, and free the bio.
126 *
127 * This marks pages up-to-date only if there was no error in the bio (I/O error,
128 * decryption error, or verity error), as indicated by bio->bi_status.
129 *
130 * "Compressed pages" (pagecache pages backed by a compressed cluster on-disk)
131 * aren't marked up-to-date here, as decompression is done on a per-compression-
132 * cluster basis rather than a per-bio basis. Instead, we only must do two
133 * things for each compressed page here: call f2fs_end_read_compressed_page()
134 * with failed=true if an error occurred before it would have normally gotten
135 * called (i.e., I/O error or decryption error, but *not* verity error), and
136 * release the bio's reference to the decompress_io_ctx of the page's cluster.
137 */
f2fs_finish_read_bio(struct bio * bio,bool in_task)138 static void f2fs_finish_read_bio(struct bio *bio, bool in_task)
139 {
140 struct bio_vec *bv;
141 struct bvec_iter_all iter_all;
142 struct bio_post_read_ctx *ctx = bio->bi_private;
143
144 bio_for_each_segment_all(bv, bio, iter_all) {
145 struct page *page = bv->bv_page;
146
147 if (f2fs_is_compressed_page(page)) {
148 if (ctx && !ctx->decompression_attempted)
149 f2fs_end_read_compressed_page(page, true, 0,
150 in_task);
151 f2fs_put_page_dic(page, in_task);
152 continue;
153 }
154
155 if (bio->bi_status)
156 ClearPageUptodate(page);
157 else
158 SetPageUptodate(page);
159 dec_page_count(F2FS_P_SB(page), __read_io_type(page));
160 unlock_page(page);
161 }
162
163 if (ctx)
164 mempool_free(ctx, bio_post_read_ctx_pool);
165 bio_put(bio);
166 }
167
f2fs_verify_bio(struct work_struct * work)168 static void f2fs_verify_bio(struct work_struct *work)
169 {
170 struct bio_post_read_ctx *ctx =
171 container_of(work, struct bio_post_read_ctx, work);
172 struct bio *bio = ctx->bio;
173 bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
174
175 /*
176 * fsverity_verify_bio() may call readahead() again, and while verity
177 * will be disabled for this, decryption and/or decompression may still
178 * be needed, resulting in another bio_post_read_ctx being allocated.
179 * So to prevent deadlocks we need to release the current ctx to the
180 * mempool first. This assumes that verity is the last post-read step.
181 */
182 mempool_free(ctx, bio_post_read_ctx_pool);
183 bio->bi_private = NULL;
184
185 /*
186 * Verify the bio's pages with fs-verity. Exclude compressed pages,
187 * as those were handled separately by f2fs_end_read_compressed_page().
188 */
189 if (may_have_compressed_pages) {
190 struct bio_vec *bv;
191 struct bvec_iter_all iter_all;
192
193 bio_for_each_segment_all(bv, bio, iter_all) {
194 struct page *page = bv->bv_page;
195
196 if (!f2fs_is_compressed_page(page) &&
197 !fsverity_verify_page(page)) {
198 bio->bi_status = BLK_STS_IOERR;
199 break;
200 }
201 }
202 } else {
203 fsverity_verify_bio(bio);
204 }
205
206 f2fs_finish_read_bio(bio, true);
207 }
208
209 /*
210 * If the bio's data needs to be verified with fs-verity, then enqueue the
211 * verity work for the bio. Otherwise finish the bio now.
212 *
213 * Note that to avoid deadlocks, the verity work can't be done on the
214 * decryption/decompression workqueue. This is because verifying the data pages
215 * can involve reading verity metadata pages from the file, and these verity
216 * metadata pages may be encrypted and/or compressed.
217 */
f2fs_verify_and_finish_bio(struct bio * bio,bool in_task)218 static void f2fs_verify_and_finish_bio(struct bio *bio, bool in_task)
219 {
220 struct bio_post_read_ctx *ctx = bio->bi_private;
221
222 if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
223 INIT_WORK(&ctx->work, f2fs_verify_bio);
224 fsverity_enqueue_verify_work(&ctx->work);
225 } else {
226 f2fs_finish_read_bio(bio, in_task);
227 }
228 }
229
230 /*
231 * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
232 * remaining page was read by @ctx->bio.
233 *
234 * Note that a bio may span clusters (even a mix of compressed and uncompressed
235 * clusters) or be for just part of a cluster. STEP_DECOMPRESS just indicates
236 * that the bio includes at least one compressed page. The actual decompression
237 * is done on a per-cluster basis, not a per-bio basis.
238 */
f2fs_handle_step_decompress(struct bio_post_read_ctx * ctx,bool in_task)239 static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx,
240 bool in_task)
241 {
242 struct bio_vec *bv;
243 struct bvec_iter_all iter_all;
244 bool all_compressed = true;
245 block_t blkaddr = ctx->fs_blkaddr;
246
247 bio_for_each_segment_all(bv, ctx->bio, iter_all) {
248 struct page *page = bv->bv_page;
249
250 if (f2fs_is_compressed_page(page))
251 f2fs_end_read_compressed_page(page, false, blkaddr,
252 in_task);
253 else
254 all_compressed = false;
255
256 blkaddr++;
257 }
258
259 ctx->decompression_attempted = true;
260
261 /*
262 * Optimization: if all the bio's pages are compressed, then scheduling
263 * the per-bio verity work is unnecessary, as verity will be fully
264 * handled at the compression cluster level.
265 */
266 if (all_compressed)
267 ctx->enabled_steps &= ~STEP_VERITY;
268 }
269
f2fs_post_read_work(struct work_struct * work)270 static void f2fs_post_read_work(struct work_struct *work)
271 {
272 struct bio_post_read_ctx *ctx =
273 container_of(work, struct bio_post_read_ctx, work);
274 struct bio *bio = ctx->bio;
275
276 if ((ctx->enabled_steps & STEP_DECRYPT) && !fscrypt_decrypt_bio(bio)) {
277 f2fs_finish_read_bio(bio, true);
278 return;
279 }
280
281 if (ctx->enabled_steps & STEP_DECOMPRESS)
282 f2fs_handle_step_decompress(ctx, true);
283
284 f2fs_verify_and_finish_bio(bio, true);
285 }
286
f2fs_read_end_io(struct bio * bio)287 static void f2fs_read_end_io(struct bio *bio)
288 {
289 struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
290 struct bio_post_read_ctx *ctx;
291 bool intask = in_task();
292
293 iostat_update_and_unbind_ctx(bio);
294 ctx = bio->bi_private;
295
296 if (time_to_inject(sbi, FAULT_READ_IO))
297 bio->bi_status = BLK_STS_IOERR;
298
299 if (bio->bi_status) {
300 f2fs_finish_read_bio(bio, intask);
301 return;
302 }
303
304 if (ctx) {
305 unsigned int enabled_steps = ctx->enabled_steps &
306 (STEP_DECRYPT | STEP_DECOMPRESS);
307
308 /*
309 * If we have only decompression step between decompression and
310 * decrypt, we don't need post processing for this.
311 */
312 if (enabled_steps == STEP_DECOMPRESS &&
313 !f2fs_low_mem_mode(sbi)) {
314 f2fs_handle_step_decompress(ctx, intask);
315 } else if (enabled_steps) {
316 INIT_WORK(&ctx->work, f2fs_post_read_work);
317 queue_work(ctx->sbi->post_read_wq, &ctx->work);
318 return;
319 }
320 }
321
322 f2fs_verify_and_finish_bio(bio, intask);
323 }
324
f2fs_write_end_io(struct bio * bio)325 static void f2fs_write_end_io(struct bio *bio)
326 {
327 struct f2fs_sb_info *sbi;
328 struct bio_vec *bvec;
329 struct bvec_iter_all iter_all;
330
331 iostat_update_and_unbind_ctx(bio);
332 sbi = bio->bi_private;
333
334 if (time_to_inject(sbi, FAULT_WRITE_IO))
335 bio->bi_status = BLK_STS_IOERR;
336
337 bio_for_each_segment_all(bvec, bio, iter_all) {
338 struct page *page = bvec->bv_page;
339 enum count_type type = WB_DATA_TYPE(page, false);
340
341 if (page_private_dummy(page)) {
342 clear_page_private_dummy(page);
343 unlock_page(page);
344 mempool_free(page, sbi->write_io_dummy);
345
346 if (unlikely(bio->bi_status))
347 f2fs_stop_checkpoint(sbi, true,
348 STOP_CP_REASON_WRITE_FAIL);
349 continue;
350 }
351
352 fscrypt_finalize_bounce_page(&page);
353
354 #ifdef CONFIG_F2FS_FS_COMPRESSION
355 if (f2fs_is_compressed_page(page)) {
356 f2fs_compress_write_end_io(bio, page);
357 continue;
358 }
359 #endif
360
361 if (unlikely(bio->bi_status)) {
362 mapping_set_error(page->mapping, -EIO);
363 if (type == F2FS_WB_CP_DATA)
364 f2fs_stop_checkpoint(sbi, true,
365 STOP_CP_REASON_WRITE_FAIL);
366 }
367
368 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
369 page->index != nid_of_node(page));
370
371 dec_page_count(sbi, type);
372 if (f2fs_in_warm_node_list(sbi, page))
373 f2fs_del_fsync_node_entry(sbi, page);
374 clear_page_private_gcing(page);
375 end_page_writeback(page);
376 }
377 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
378 wq_has_sleeper(&sbi->cp_wait))
379 wake_up(&sbi->cp_wait);
380
381 bio_put(bio);
382 }
383
384 #ifdef CONFIG_BLK_DEV_ZONED
f2fs_zone_write_end_io(struct bio * bio)385 static void f2fs_zone_write_end_io(struct bio *bio)
386 {
387 struct f2fs_bio_info *io = (struct f2fs_bio_info *)bio->bi_private;
388
389 bio->bi_private = io->bi_private;
390 complete(&io->zone_wait);
391 f2fs_write_end_io(bio);
392 }
393 #endif
394
f2fs_target_device(struct f2fs_sb_info * sbi,block_t blk_addr,sector_t * sector)395 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
396 block_t blk_addr, sector_t *sector)
397 {
398 struct block_device *bdev = sbi->sb->s_bdev;
399 int i;
400
401 if (f2fs_is_multi_device(sbi)) {
402 for (i = 0; i < sbi->s_ndevs; i++) {
403 if (FDEV(i).start_blk <= blk_addr &&
404 FDEV(i).end_blk >= blk_addr) {
405 blk_addr -= FDEV(i).start_blk;
406 bdev = FDEV(i).bdev;
407 break;
408 }
409 }
410 }
411
412 if (sector)
413 *sector = SECTOR_FROM_BLOCK(blk_addr);
414 return bdev;
415 }
416
f2fs_target_device_index(struct f2fs_sb_info * sbi,block_t blkaddr)417 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
418 {
419 int i;
420
421 if (!f2fs_is_multi_device(sbi))
422 return 0;
423
424 for (i = 0; i < sbi->s_ndevs; i++)
425 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
426 return i;
427 return 0;
428 }
429
f2fs_io_flags(struct f2fs_io_info * fio)430 static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio)
431 {
432 unsigned int temp_mask = GENMASK(NR_TEMP_TYPE - 1, 0);
433 unsigned int fua_flag, meta_flag, io_flag;
434 blk_opf_t op_flags = 0;
435
436 if (fio->op != REQ_OP_WRITE)
437 return 0;
438 if (fio->type == DATA)
439 io_flag = fio->sbi->data_io_flag;
440 else if (fio->type == NODE)
441 io_flag = fio->sbi->node_io_flag;
442 else
443 return 0;
444
445 fua_flag = io_flag & temp_mask;
446 meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
447
448 /*
449 * data/node io flag bits per temp:
450 * REQ_META | REQ_FUA |
451 * 5 | 4 | 3 | 2 | 1 | 0 |
452 * Cold | Warm | Hot | Cold | Warm | Hot |
453 */
454 if (BIT(fio->temp) & meta_flag)
455 op_flags |= REQ_META;
456 if (BIT(fio->temp) & fua_flag)
457 op_flags |= REQ_FUA;
458 return op_flags;
459 }
460
__bio_alloc(struct f2fs_io_info * fio,int npages)461 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
462 {
463 struct f2fs_sb_info *sbi = fio->sbi;
464 struct block_device *bdev;
465 sector_t sector;
466 struct bio *bio;
467
468 bdev = f2fs_target_device(sbi, fio->new_blkaddr, §or);
469 bio = bio_alloc_bioset(bdev, npages,
470 fio->op | fio->op_flags | f2fs_io_flags(fio),
471 GFP_NOIO, &f2fs_bioset);
472 bio->bi_iter.bi_sector = sector;
473 if (is_read_io(fio->op)) {
474 bio->bi_end_io = f2fs_read_end_io;
475 bio->bi_private = NULL;
476 } else {
477 bio->bi_end_io = f2fs_write_end_io;
478 bio->bi_private = sbi;
479 }
480 iostat_alloc_and_bind_ctx(sbi, bio, NULL);
481
482 if (fio->io_wbc)
483 wbc_init_bio(fio->io_wbc, bio);
484
485 return bio;
486 }
487
f2fs_set_bio_crypt_ctx(struct bio * bio,const struct inode * inode,pgoff_t first_idx,const struct f2fs_io_info * fio,gfp_t gfp_mask)488 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
489 pgoff_t first_idx,
490 const struct f2fs_io_info *fio,
491 gfp_t gfp_mask)
492 {
493 /*
494 * The f2fs garbage collector sets ->encrypted_page when it wants to
495 * read/write raw data without encryption.
496 */
497 if (!fio || !fio->encrypted_page)
498 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
499 }
500
f2fs_crypt_mergeable_bio(struct bio * bio,const struct inode * inode,pgoff_t next_idx,const struct f2fs_io_info * fio)501 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
502 pgoff_t next_idx,
503 const struct f2fs_io_info *fio)
504 {
505 /*
506 * The f2fs garbage collector sets ->encrypted_page when it wants to
507 * read/write raw data without encryption.
508 */
509 if (fio && fio->encrypted_page)
510 return !bio_has_crypt_ctx(bio);
511
512 return fscrypt_mergeable_bio(bio, inode, next_idx);
513 }
514
f2fs_submit_read_bio(struct f2fs_sb_info * sbi,struct bio * bio,enum page_type type)515 void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
516 enum page_type type)
517 {
518 WARN_ON_ONCE(!is_read_io(bio_op(bio)));
519 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
520
521 iostat_update_submit_ctx(bio, type);
522 submit_bio(bio);
523 }
524
f2fs_align_write_bio(struct f2fs_sb_info * sbi,struct bio * bio)525 static void f2fs_align_write_bio(struct f2fs_sb_info *sbi, struct bio *bio)
526 {
527 unsigned int start =
528 (bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS) % F2FS_IO_SIZE(sbi);
529
530 if (start == 0)
531 return;
532
533 /* fill dummy pages */
534 for (; start < F2FS_IO_SIZE(sbi); start++) {
535 struct page *page =
536 mempool_alloc(sbi->write_io_dummy,
537 GFP_NOIO | __GFP_NOFAIL);
538 f2fs_bug_on(sbi, !page);
539
540 lock_page(page);
541
542 zero_user_segment(page, 0, PAGE_SIZE);
543 set_page_private_dummy(page);
544
545 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
546 f2fs_bug_on(sbi, 1);
547 }
548 }
549
f2fs_submit_write_bio(struct f2fs_sb_info * sbi,struct bio * bio,enum page_type type)550 static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio,
551 enum page_type type)
552 {
553 WARN_ON_ONCE(is_read_io(bio_op(bio)));
554
555 if (type == DATA || type == NODE) {
556 if (f2fs_lfs_mode(sbi) && current->plug)
557 blk_finish_plug(current->plug);
558
559 if (F2FS_IO_ALIGNED(sbi)) {
560 f2fs_align_write_bio(sbi, bio);
561 /*
562 * In the NODE case, we lose next block address chain.
563 * So, we need to do checkpoint in f2fs_sync_file.
564 */
565 if (type == NODE)
566 set_sbi_flag(sbi, SBI_NEED_CP);
567 }
568 }
569
570 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
571 iostat_update_submit_ctx(bio, type);
572 submit_bio(bio);
573 }
574
__submit_merged_bio(struct f2fs_bio_info * io)575 static void __submit_merged_bio(struct f2fs_bio_info *io)
576 {
577 struct f2fs_io_info *fio = &io->fio;
578
579 if (!io->bio)
580 return;
581
582 if (is_read_io(fio->op)) {
583 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
584 f2fs_submit_read_bio(io->sbi, io->bio, fio->type);
585 } else {
586 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
587 f2fs_submit_write_bio(io->sbi, io->bio, fio->type);
588 }
589 io->bio = NULL;
590 }
591
__has_merged_page(struct bio * bio,struct inode * inode,struct page * page,nid_t ino)592 static bool __has_merged_page(struct bio *bio, struct inode *inode,
593 struct page *page, nid_t ino)
594 {
595 struct bio_vec *bvec;
596 struct bvec_iter_all iter_all;
597
598 if (!bio)
599 return false;
600
601 if (!inode && !page && !ino)
602 return true;
603
604 bio_for_each_segment_all(bvec, bio, iter_all) {
605 struct page *target = bvec->bv_page;
606
607 if (fscrypt_is_bounce_page(target)) {
608 target = fscrypt_pagecache_page(target);
609 if (IS_ERR(target))
610 continue;
611 }
612 if (f2fs_is_compressed_page(target)) {
613 target = f2fs_compress_control_page(target);
614 if (IS_ERR(target))
615 continue;
616 }
617
618 if (inode && inode == target->mapping->host)
619 return true;
620 if (page && page == target)
621 return true;
622 if (ino && ino == ino_of_node(target))
623 return true;
624 }
625
626 return false;
627 }
628
f2fs_init_write_merge_io(struct f2fs_sb_info * sbi)629 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
630 {
631 int i;
632
633 for (i = 0; i < NR_PAGE_TYPE; i++) {
634 int n = (i == META) ? 1 : NR_TEMP_TYPE;
635 int j;
636
637 sbi->write_io[i] = f2fs_kmalloc(sbi,
638 array_size(n, sizeof(struct f2fs_bio_info)),
639 GFP_KERNEL);
640 if (!sbi->write_io[i])
641 return -ENOMEM;
642
643 for (j = HOT; j < n; j++) {
644 init_f2fs_rwsem(&sbi->write_io[i][j].io_rwsem);
645 sbi->write_io[i][j].sbi = sbi;
646 sbi->write_io[i][j].bio = NULL;
647 spin_lock_init(&sbi->write_io[i][j].io_lock);
648 INIT_LIST_HEAD(&sbi->write_io[i][j].io_list);
649 INIT_LIST_HEAD(&sbi->write_io[i][j].bio_list);
650 init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock);
651 #ifdef CONFIG_BLK_DEV_ZONED
652 init_completion(&sbi->write_io[i][j].zone_wait);
653 sbi->write_io[i][j].zone_pending_bio = NULL;
654 sbi->write_io[i][j].bi_private = NULL;
655 #endif
656 }
657 }
658
659 return 0;
660 }
661
__f2fs_submit_merged_write(struct f2fs_sb_info * sbi,enum page_type type,enum temp_type temp)662 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
663 enum page_type type, enum temp_type temp)
664 {
665 enum page_type btype = PAGE_TYPE_OF_BIO(type);
666 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
667
668 f2fs_down_write(&io->io_rwsem);
669
670 if (!io->bio)
671 goto unlock_out;
672
673 /* change META to META_FLUSH in the checkpoint procedure */
674 if (type >= META_FLUSH) {
675 io->fio.type = META_FLUSH;
676 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
677 if (!test_opt(sbi, NOBARRIER))
678 io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
679 }
680 __submit_merged_bio(io);
681 unlock_out:
682 f2fs_up_write(&io->io_rwsem);
683 }
684
__submit_merged_write_cond(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page,nid_t ino,enum page_type type,bool force)685 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
686 struct inode *inode, struct page *page,
687 nid_t ino, enum page_type type, bool force)
688 {
689 enum temp_type temp;
690 bool ret = true;
691
692 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
693 if (!force) {
694 enum page_type btype = PAGE_TYPE_OF_BIO(type);
695 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
696
697 f2fs_down_read(&io->io_rwsem);
698 ret = __has_merged_page(io->bio, inode, page, ino);
699 f2fs_up_read(&io->io_rwsem);
700 }
701 if (ret)
702 __f2fs_submit_merged_write(sbi, type, temp);
703
704 /* TODO: use HOT temp only for meta pages now. */
705 if (type >= META)
706 break;
707 }
708 }
709
f2fs_submit_merged_write(struct f2fs_sb_info * sbi,enum page_type type)710 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
711 {
712 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
713 }
714
f2fs_submit_merged_write_cond(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page,nid_t ino,enum page_type type)715 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
716 struct inode *inode, struct page *page,
717 nid_t ino, enum page_type type)
718 {
719 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
720 }
721
f2fs_flush_merged_writes(struct f2fs_sb_info * sbi)722 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
723 {
724 f2fs_submit_merged_write(sbi, DATA);
725 f2fs_submit_merged_write(sbi, NODE);
726 f2fs_submit_merged_write(sbi, META);
727 }
728
729 /*
730 * Fill the locked page with data located in the block address.
731 * A caller needs to unlock the page on failure.
732 */
f2fs_submit_page_bio(struct f2fs_io_info * fio)733 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
734 {
735 struct bio *bio;
736 struct page *page = fio->encrypted_page ?
737 fio->encrypted_page : fio->page;
738
739 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
740 fio->is_por ? META_POR : (__is_meta_io(fio) ?
741 META_GENERIC : DATA_GENERIC_ENHANCE))) {
742 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
743 return -EFSCORRUPTED;
744 }
745
746 trace_f2fs_submit_page_bio(page, fio);
747
748 /* Allocate a new bio */
749 bio = __bio_alloc(fio, 1);
750
751 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
752 fio->page->index, fio, GFP_NOIO);
753
754 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
755 bio_put(bio);
756 return -EFAULT;
757 }
758
759 if (fio->io_wbc && !is_read_io(fio->op))
760 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
761
762 inc_page_count(fio->sbi, is_read_io(fio->op) ?
763 __read_io_type(page) : WB_DATA_TYPE(fio->page, false));
764
765 if (is_read_io(bio_op(bio)))
766 f2fs_submit_read_bio(fio->sbi, bio, fio->type);
767 else
768 f2fs_submit_write_bio(fio->sbi, bio, fio->type);
769 return 0;
770 }
771
page_is_mergeable(struct f2fs_sb_info * sbi,struct bio * bio,block_t last_blkaddr,block_t cur_blkaddr)772 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
773 block_t last_blkaddr, block_t cur_blkaddr)
774 {
775 if (unlikely(sbi->max_io_bytes &&
776 bio->bi_iter.bi_size >= sbi->max_io_bytes))
777 return false;
778 if (last_blkaddr + 1 != cur_blkaddr)
779 return false;
780 return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
781 }
782
io_type_is_mergeable(struct f2fs_bio_info * io,struct f2fs_io_info * fio)783 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
784 struct f2fs_io_info *fio)
785 {
786 if (io->fio.op != fio->op)
787 return false;
788 return io->fio.op_flags == fio->op_flags;
789 }
790
io_is_mergeable(struct f2fs_sb_info * sbi,struct bio * bio,struct f2fs_bio_info * io,struct f2fs_io_info * fio,block_t last_blkaddr,block_t cur_blkaddr)791 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
792 struct f2fs_bio_info *io,
793 struct f2fs_io_info *fio,
794 block_t last_blkaddr,
795 block_t cur_blkaddr)
796 {
797 if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
798 unsigned int filled_blocks =
799 F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
800 unsigned int io_size = F2FS_IO_SIZE(sbi);
801 unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
802
803 /* IOs in bio is aligned and left space of vectors is not enough */
804 if (!(filled_blocks % io_size) && left_vecs < io_size)
805 return false;
806 }
807 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
808 return false;
809 return io_type_is_mergeable(io, fio);
810 }
811
add_bio_entry(struct f2fs_sb_info * sbi,struct bio * bio,struct page * page,enum temp_type temp)812 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
813 struct page *page, enum temp_type temp)
814 {
815 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
816 struct bio_entry *be;
817
818 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
819 be->bio = bio;
820 bio_get(bio);
821
822 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
823 f2fs_bug_on(sbi, 1);
824
825 f2fs_down_write(&io->bio_list_lock);
826 list_add_tail(&be->list, &io->bio_list);
827 f2fs_up_write(&io->bio_list_lock);
828 }
829
del_bio_entry(struct bio_entry * be)830 static void del_bio_entry(struct bio_entry *be)
831 {
832 list_del(&be->list);
833 kmem_cache_free(bio_entry_slab, be);
834 }
835
add_ipu_page(struct f2fs_io_info * fio,struct bio ** bio,struct page * page)836 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
837 struct page *page)
838 {
839 struct f2fs_sb_info *sbi = fio->sbi;
840 enum temp_type temp;
841 bool found = false;
842 int ret = -EAGAIN;
843
844 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
845 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
846 struct list_head *head = &io->bio_list;
847 struct bio_entry *be;
848
849 f2fs_down_write(&io->bio_list_lock);
850 list_for_each_entry(be, head, list) {
851 if (be->bio != *bio)
852 continue;
853
854 found = true;
855
856 f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
857 *fio->last_block,
858 fio->new_blkaddr));
859 if (f2fs_crypt_mergeable_bio(*bio,
860 fio->page->mapping->host,
861 fio->page->index, fio) &&
862 bio_add_page(*bio, page, PAGE_SIZE, 0) ==
863 PAGE_SIZE) {
864 ret = 0;
865 break;
866 }
867
868 /* page can't be merged into bio; submit the bio */
869 del_bio_entry(be);
870 f2fs_submit_write_bio(sbi, *bio, DATA);
871 break;
872 }
873 f2fs_up_write(&io->bio_list_lock);
874 }
875
876 if (ret) {
877 bio_put(*bio);
878 *bio = NULL;
879 }
880
881 return ret;
882 }
883
f2fs_submit_merged_ipu_write(struct f2fs_sb_info * sbi,struct bio ** bio,struct page * page)884 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
885 struct bio **bio, struct page *page)
886 {
887 enum temp_type temp;
888 bool found = false;
889 struct bio *target = bio ? *bio : NULL;
890
891 f2fs_bug_on(sbi, !target && !page);
892
893 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
894 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
895 struct list_head *head = &io->bio_list;
896 struct bio_entry *be;
897
898 if (list_empty(head))
899 continue;
900
901 f2fs_down_read(&io->bio_list_lock);
902 list_for_each_entry(be, head, list) {
903 if (target)
904 found = (target == be->bio);
905 else
906 found = __has_merged_page(be->bio, NULL,
907 page, 0);
908 if (found)
909 break;
910 }
911 f2fs_up_read(&io->bio_list_lock);
912
913 if (!found)
914 continue;
915
916 found = false;
917
918 f2fs_down_write(&io->bio_list_lock);
919 list_for_each_entry(be, head, list) {
920 if (target)
921 found = (target == be->bio);
922 else
923 found = __has_merged_page(be->bio, NULL,
924 page, 0);
925 if (found) {
926 target = be->bio;
927 del_bio_entry(be);
928 break;
929 }
930 }
931 f2fs_up_write(&io->bio_list_lock);
932 }
933
934 if (found)
935 f2fs_submit_write_bio(sbi, target, DATA);
936 if (bio && *bio) {
937 bio_put(*bio);
938 *bio = NULL;
939 }
940 }
941
f2fs_merge_page_bio(struct f2fs_io_info * fio)942 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
943 {
944 struct bio *bio = *fio->bio;
945 struct page *page = fio->encrypted_page ?
946 fio->encrypted_page : fio->page;
947
948 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
949 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) {
950 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
951 return -EFSCORRUPTED;
952 }
953
954 trace_f2fs_submit_page_bio(page, fio);
955
956 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
957 fio->new_blkaddr))
958 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
959 alloc_new:
960 if (!bio) {
961 bio = __bio_alloc(fio, BIO_MAX_VECS);
962 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
963 fio->page->index, fio, GFP_NOIO);
964
965 add_bio_entry(fio->sbi, bio, page, fio->temp);
966 } else {
967 if (add_ipu_page(fio, &bio, page))
968 goto alloc_new;
969 }
970
971 if (fio->io_wbc)
972 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
973
974 inc_page_count(fio->sbi, WB_DATA_TYPE(page, false));
975
976 *fio->last_block = fio->new_blkaddr;
977 *fio->bio = bio;
978
979 return 0;
980 }
981
982 #ifdef CONFIG_BLK_DEV_ZONED
is_end_zone_blkaddr(struct f2fs_sb_info * sbi,block_t blkaddr)983 static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr)
984 {
985 int devi = 0;
986
987 if (f2fs_is_multi_device(sbi)) {
988 devi = f2fs_target_device_index(sbi, blkaddr);
989 if (blkaddr < FDEV(devi).start_blk ||
990 blkaddr > FDEV(devi).end_blk) {
991 f2fs_err(sbi, "Invalid block %x", blkaddr);
992 return false;
993 }
994 blkaddr -= FDEV(devi).start_blk;
995 }
996 return bdev_zoned_model(FDEV(devi).bdev) == BLK_ZONED_HM &&
997 f2fs_blkz_is_seq(sbi, devi, blkaddr) &&
998 (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1);
999 }
1000 #endif
1001
f2fs_submit_page_write(struct f2fs_io_info * fio)1002 void f2fs_submit_page_write(struct f2fs_io_info *fio)
1003 {
1004 struct f2fs_sb_info *sbi = fio->sbi;
1005 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
1006 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
1007 struct page *bio_page;
1008 enum count_type type;
1009
1010 f2fs_bug_on(sbi, is_read_io(fio->op));
1011
1012 f2fs_down_write(&io->io_rwsem);
1013 next:
1014 #ifdef CONFIG_BLK_DEV_ZONED
1015 if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) {
1016 wait_for_completion_io(&io->zone_wait);
1017 bio_put(io->zone_pending_bio);
1018 io->zone_pending_bio = NULL;
1019 io->bi_private = NULL;
1020 }
1021 #endif
1022
1023 if (fio->in_list) {
1024 spin_lock(&io->io_lock);
1025 if (list_empty(&io->io_list)) {
1026 spin_unlock(&io->io_lock);
1027 goto out;
1028 }
1029 fio = list_first_entry(&io->io_list,
1030 struct f2fs_io_info, list);
1031 list_del(&fio->list);
1032 spin_unlock(&io->io_lock);
1033 }
1034
1035 verify_fio_blkaddr(fio);
1036
1037 if (fio->encrypted_page)
1038 bio_page = fio->encrypted_page;
1039 else if (fio->compressed_page)
1040 bio_page = fio->compressed_page;
1041 else
1042 bio_page = fio->page;
1043
1044 /* set submitted = true as a return value */
1045 fio->submitted = 1;
1046
1047 type = WB_DATA_TYPE(bio_page, fio->compressed_page);
1048 inc_page_count(sbi, type);
1049
1050 if (io->bio &&
1051 (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
1052 fio->new_blkaddr) ||
1053 !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
1054 bio_page->index, fio)))
1055 __submit_merged_bio(io);
1056 alloc_new:
1057 if (io->bio == NULL) {
1058 if (F2FS_IO_ALIGNED(sbi) &&
1059 (fio->type == DATA || fio->type == NODE) &&
1060 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
1061 dec_page_count(sbi, WB_DATA_TYPE(bio_page,
1062 fio->compressed_page));
1063 fio->retry = 1;
1064 goto skip;
1065 }
1066 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
1067 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
1068 bio_page->index, fio, GFP_NOIO);
1069 io->fio = *fio;
1070 }
1071
1072 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1073 __submit_merged_bio(io);
1074 goto alloc_new;
1075 }
1076
1077 if (fio->io_wbc)
1078 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
1079
1080 io->last_block_in_bio = fio->new_blkaddr;
1081
1082 trace_f2fs_submit_page_write(fio->page, fio);
1083 #ifdef CONFIG_BLK_DEV_ZONED
1084 if (f2fs_sb_has_blkzoned(sbi) && btype < META &&
1085 is_end_zone_blkaddr(sbi, fio->new_blkaddr)) {
1086 bio_get(io->bio);
1087 reinit_completion(&io->zone_wait);
1088 io->bi_private = io->bio->bi_private;
1089 io->bio->bi_private = io;
1090 io->bio->bi_end_io = f2fs_zone_write_end_io;
1091 io->zone_pending_bio = io->bio;
1092 __submit_merged_bio(io);
1093 }
1094 #endif
1095 skip:
1096 if (fio->in_list)
1097 goto next;
1098 out:
1099 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1100 !f2fs_is_checkpoint_ready(sbi))
1101 __submit_merged_bio(io);
1102 f2fs_up_write(&io->io_rwsem);
1103 }
1104
f2fs_grab_read_bio(struct inode * inode,block_t blkaddr,unsigned nr_pages,blk_opf_t op_flag,pgoff_t first_idx,bool for_write)1105 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1106 unsigned nr_pages, blk_opf_t op_flag,
1107 pgoff_t first_idx, bool for_write)
1108 {
1109 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1110 struct bio *bio;
1111 struct bio_post_read_ctx *ctx = NULL;
1112 unsigned int post_read_steps = 0;
1113 sector_t sector;
1114 struct block_device *bdev = f2fs_target_device(sbi, blkaddr, §or);
1115
1116 bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
1117 REQ_OP_READ | op_flag,
1118 for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
1119 if (!bio)
1120 return ERR_PTR(-ENOMEM);
1121 bio->bi_iter.bi_sector = sector;
1122 f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1123 bio->bi_end_io = f2fs_read_end_io;
1124
1125 if (fscrypt_inode_uses_fs_layer_crypto(inode))
1126 post_read_steps |= STEP_DECRYPT;
1127
1128 if (f2fs_need_verity(inode, first_idx))
1129 post_read_steps |= STEP_VERITY;
1130
1131 /*
1132 * STEP_DECOMPRESS is handled specially, since a compressed file might
1133 * contain both compressed and uncompressed clusters. We'll allocate a
1134 * bio_post_read_ctx if the file is compressed, but the caller is
1135 * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1136 */
1137
1138 if (post_read_steps || f2fs_compressed_file(inode)) {
1139 /* Due to the mempool, this never fails. */
1140 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1141 ctx->bio = bio;
1142 ctx->sbi = sbi;
1143 ctx->enabled_steps = post_read_steps;
1144 ctx->fs_blkaddr = blkaddr;
1145 ctx->decompression_attempted = false;
1146 bio->bi_private = ctx;
1147 }
1148 iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1149
1150 return bio;
1151 }
1152
1153 /* This can handle encryption stuffs */
f2fs_submit_page_read(struct inode * inode,struct page * page,block_t blkaddr,blk_opf_t op_flags,bool for_write)1154 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1155 block_t blkaddr, blk_opf_t op_flags,
1156 bool for_write)
1157 {
1158 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1159 struct bio *bio;
1160
1161 bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1162 page->index, for_write);
1163 if (IS_ERR(bio))
1164 return PTR_ERR(bio);
1165
1166 /* wait for GCed page writeback via META_MAPPING */
1167 f2fs_wait_on_block_writeback(inode, blkaddr);
1168
1169 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1170 iostat_update_and_unbind_ctx(bio);
1171 if (bio->bi_private)
1172 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1173 bio_put(bio);
1174 return -EFAULT;
1175 }
1176 inc_page_count(sbi, F2FS_RD_DATA);
1177 f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE);
1178 f2fs_submit_read_bio(sbi, bio, DATA);
1179 return 0;
1180 }
1181
__set_data_blkaddr(struct dnode_of_data * dn,block_t blkaddr)1182 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1183 {
1184 __le32 *addr = get_dnode_addr(dn->inode, dn->node_page);
1185
1186 dn->data_blkaddr = blkaddr;
1187 addr[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1188 }
1189
1190 /*
1191 * Lock ordering for the change of data block address:
1192 * ->data_page
1193 * ->node_page
1194 * update block addresses in the node page
1195 */
f2fs_set_data_blkaddr(struct dnode_of_data * dn,block_t blkaddr)1196 void f2fs_set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1197 {
1198 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1199 __set_data_blkaddr(dn, blkaddr);
1200 if (set_page_dirty(dn->node_page))
1201 dn->node_changed = true;
1202 }
1203
f2fs_update_data_blkaddr(struct dnode_of_data * dn,block_t blkaddr)1204 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1205 {
1206 f2fs_set_data_blkaddr(dn, blkaddr);
1207 f2fs_update_read_extent_cache(dn);
1208 }
1209
1210 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
f2fs_reserve_new_blocks(struct dnode_of_data * dn,blkcnt_t count)1211 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1212 {
1213 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1214 int err;
1215
1216 if (!count)
1217 return 0;
1218
1219 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1220 return -EPERM;
1221 err = inc_valid_block_count(sbi, dn->inode, &count, true);
1222 if (unlikely(err))
1223 return err;
1224
1225 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1226 dn->ofs_in_node, count);
1227
1228 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1229
1230 for (; count > 0; dn->ofs_in_node++) {
1231 block_t blkaddr = f2fs_data_blkaddr(dn);
1232
1233 if (blkaddr == NULL_ADDR) {
1234 __set_data_blkaddr(dn, NEW_ADDR);
1235 count--;
1236 }
1237 }
1238
1239 if (set_page_dirty(dn->node_page))
1240 dn->node_changed = true;
1241 return 0;
1242 }
1243
1244 /* Should keep dn->ofs_in_node unchanged */
f2fs_reserve_new_block(struct dnode_of_data * dn)1245 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1246 {
1247 unsigned int ofs_in_node = dn->ofs_in_node;
1248 int ret;
1249
1250 ret = f2fs_reserve_new_blocks(dn, 1);
1251 dn->ofs_in_node = ofs_in_node;
1252 return ret;
1253 }
1254
f2fs_reserve_block(struct dnode_of_data * dn,pgoff_t index)1255 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1256 {
1257 bool need_put = dn->inode_page ? false : true;
1258 int err;
1259
1260 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1261 if (err)
1262 return err;
1263
1264 if (dn->data_blkaddr == NULL_ADDR)
1265 err = f2fs_reserve_new_block(dn);
1266 if (err || need_put)
1267 f2fs_put_dnode(dn);
1268 return err;
1269 }
1270
f2fs_get_read_data_page(struct inode * inode,pgoff_t index,blk_opf_t op_flags,bool for_write,pgoff_t * next_pgofs)1271 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1272 blk_opf_t op_flags, bool for_write,
1273 pgoff_t *next_pgofs)
1274 {
1275 struct address_space *mapping = inode->i_mapping;
1276 struct dnode_of_data dn;
1277 struct page *page;
1278 int err;
1279
1280 page = f2fs_grab_cache_page(mapping, index, for_write);
1281 if (!page)
1282 return ERR_PTR(-ENOMEM);
1283
1284 if (f2fs_lookup_read_extent_cache_block(inode, index,
1285 &dn.data_blkaddr)) {
1286 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1287 DATA_GENERIC_ENHANCE_READ)) {
1288 err = -EFSCORRUPTED;
1289 f2fs_handle_error(F2FS_I_SB(inode),
1290 ERROR_INVALID_BLKADDR);
1291 goto put_err;
1292 }
1293 goto got_it;
1294 }
1295
1296 set_new_dnode(&dn, inode, NULL, NULL, 0);
1297 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1298 if (err) {
1299 if (err == -ENOENT && next_pgofs)
1300 *next_pgofs = f2fs_get_next_page_offset(&dn, index);
1301 goto put_err;
1302 }
1303 f2fs_put_dnode(&dn);
1304
1305 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1306 err = -ENOENT;
1307 if (next_pgofs)
1308 *next_pgofs = index + 1;
1309 goto put_err;
1310 }
1311 if (dn.data_blkaddr != NEW_ADDR &&
1312 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1313 dn.data_blkaddr,
1314 DATA_GENERIC_ENHANCE)) {
1315 err = -EFSCORRUPTED;
1316 f2fs_handle_error(F2FS_I_SB(inode),
1317 ERROR_INVALID_BLKADDR);
1318 goto put_err;
1319 }
1320 got_it:
1321 if (PageUptodate(page)) {
1322 unlock_page(page);
1323 return page;
1324 }
1325
1326 /*
1327 * A new dentry page is allocated but not able to be written, since its
1328 * new inode page couldn't be allocated due to -ENOSPC.
1329 * In such the case, its blkaddr can be remained as NEW_ADDR.
1330 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1331 * f2fs_init_inode_metadata.
1332 */
1333 if (dn.data_blkaddr == NEW_ADDR) {
1334 zero_user_segment(page, 0, PAGE_SIZE);
1335 if (!PageUptodate(page))
1336 SetPageUptodate(page);
1337 unlock_page(page);
1338 return page;
1339 }
1340
1341 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1342 op_flags, for_write);
1343 if (err)
1344 goto put_err;
1345 return page;
1346
1347 put_err:
1348 f2fs_put_page(page, 1);
1349 return ERR_PTR(err);
1350 }
1351
f2fs_find_data_page(struct inode * inode,pgoff_t index,pgoff_t * next_pgofs)1352 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index,
1353 pgoff_t *next_pgofs)
1354 {
1355 struct address_space *mapping = inode->i_mapping;
1356 struct page *page;
1357
1358 page = find_get_page(mapping, index);
1359 if (page && PageUptodate(page))
1360 return page;
1361 f2fs_put_page(page, 0);
1362
1363 page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs);
1364 if (IS_ERR(page))
1365 return page;
1366
1367 if (PageUptodate(page))
1368 return page;
1369
1370 wait_on_page_locked(page);
1371 if (unlikely(!PageUptodate(page))) {
1372 f2fs_put_page(page, 0);
1373 return ERR_PTR(-EIO);
1374 }
1375 return page;
1376 }
1377
1378 /*
1379 * If it tries to access a hole, return an error.
1380 * Because, the callers, functions in dir.c and GC, should be able to know
1381 * whether this page exists or not.
1382 */
f2fs_get_lock_data_page(struct inode * inode,pgoff_t index,bool for_write)1383 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1384 bool for_write)
1385 {
1386 struct address_space *mapping = inode->i_mapping;
1387 struct page *page;
1388
1389 page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL);
1390 if (IS_ERR(page))
1391 return page;
1392
1393 /* wait for read completion */
1394 lock_page(page);
1395 if (unlikely(page->mapping != mapping || !PageUptodate(page))) {
1396 f2fs_put_page(page, 1);
1397 return ERR_PTR(-EIO);
1398 }
1399 return page;
1400 }
1401
1402 /*
1403 * Caller ensures that this data page is never allocated.
1404 * A new zero-filled data page is allocated in the page cache.
1405 *
1406 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1407 * f2fs_unlock_op().
1408 * Note that, ipage is set only by make_empty_dir, and if any error occur,
1409 * ipage should be released by this function.
1410 */
f2fs_get_new_data_page(struct inode * inode,struct page * ipage,pgoff_t index,bool new_i_size)1411 struct page *f2fs_get_new_data_page(struct inode *inode,
1412 struct page *ipage, pgoff_t index, bool new_i_size)
1413 {
1414 struct address_space *mapping = inode->i_mapping;
1415 struct page *page;
1416 struct dnode_of_data dn;
1417 int err;
1418
1419 page = f2fs_grab_cache_page(mapping, index, true);
1420 if (!page) {
1421 /*
1422 * before exiting, we should make sure ipage will be released
1423 * if any error occur.
1424 */
1425 f2fs_put_page(ipage, 1);
1426 return ERR_PTR(-ENOMEM);
1427 }
1428
1429 set_new_dnode(&dn, inode, ipage, NULL, 0);
1430 err = f2fs_reserve_block(&dn, index);
1431 if (err) {
1432 f2fs_put_page(page, 1);
1433 return ERR_PTR(err);
1434 }
1435 if (!ipage)
1436 f2fs_put_dnode(&dn);
1437
1438 if (PageUptodate(page))
1439 goto got_it;
1440
1441 if (dn.data_blkaddr == NEW_ADDR) {
1442 zero_user_segment(page, 0, PAGE_SIZE);
1443 if (!PageUptodate(page))
1444 SetPageUptodate(page);
1445 } else {
1446 f2fs_put_page(page, 1);
1447
1448 /* if ipage exists, blkaddr should be NEW_ADDR */
1449 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1450 page = f2fs_get_lock_data_page(inode, index, true);
1451 if (IS_ERR(page))
1452 return page;
1453 }
1454 got_it:
1455 if (new_i_size && i_size_read(inode) <
1456 ((loff_t)(index + 1) << PAGE_SHIFT))
1457 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1458 return page;
1459 }
1460
__allocate_data_block(struct dnode_of_data * dn,int seg_type)1461 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1462 {
1463 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1464 struct f2fs_summary sum;
1465 struct node_info ni;
1466 block_t old_blkaddr;
1467 blkcnt_t count = 1;
1468 int err;
1469
1470 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1471 return -EPERM;
1472
1473 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1474 if (err)
1475 return err;
1476
1477 dn->data_blkaddr = f2fs_data_blkaddr(dn);
1478 if (dn->data_blkaddr == NULL_ADDR) {
1479 err = inc_valid_block_count(sbi, dn->inode, &count, true);
1480 if (unlikely(err))
1481 return err;
1482 }
1483
1484 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1485 old_blkaddr = dn->data_blkaddr;
1486 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1487 &sum, seg_type, NULL);
1488 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1489 f2fs_invalidate_internal_cache(sbi, old_blkaddr);
1490
1491 f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1492 return 0;
1493 }
1494
f2fs_map_lock(struct f2fs_sb_info * sbi,int flag)1495 static void f2fs_map_lock(struct f2fs_sb_info *sbi, int flag)
1496 {
1497 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1498 f2fs_down_read(&sbi->node_change);
1499 else
1500 f2fs_lock_op(sbi);
1501 }
1502
f2fs_map_unlock(struct f2fs_sb_info * sbi,int flag)1503 static void f2fs_map_unlock(struct f2fs_sb_info *sbi, int flag)
1504 {
1505 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1506 f2fs_up_read(&sbi->node_change);
1507 else
1508 f2fs_unlock_op(sbi);
1509 }
1510
f2fs_get_block_locked(struct dnode_of_data * dn,pgoff_t index)1511 int f2fs_get_block_locked(struct dnode_of_data *dn, pgoff_t index)
1512 {
1513 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1514 int err = 0;
1515
1516 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1517 if (!f2fs_lookup_read_extent_cache_block(dn->inode, index,
1518 &dn->data_blkaddr))
1519 err = f2fs_reserve_block(dn, index);
1520 f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1521
1522 return err;
1523 }
1524
f2fs_map_no_dnode(struct inode * inode,struct f2fs_map_blocks * map,struct dnode_of_data * dn,pgoff_t pgoff)1525 static int f2fs_map_no_dnode(struct inode *inode,
1526 struct f2fs_map_blocks *map, struct dnode_of_data *dn,
1527 pgoff_t pgoff)
1528 {
1529 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1530
1531 /*
1532 * There is one exceptional case that read_node_page() may return
1533 * -ENOENT due to filesystem has been shutdown or cp_error, return
1534 * -EIO in that case.
1535 */
1536 if (map->m_may_create &&
1537 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || f2fs_cp_error(sbi)))
1538 return -EIO;
1539
1540 if (map->m_next_pgofs)
1541 *map->m_next_pgofs = f2fs_get_next_page_offset(dn, pgoff);
1542 if (map->m_next_extent)
1543 *map->m_next_extent = f2fs_get_next_page_offset(dn, pgoff);
1544 return 0;
1545 }
1546
f2fs_map_blocks_cached(struct inode * inode,struct f2fs_map_blocks * map,int flag)1547 static bool f2fs_map_blocks_cached(struct inode *inode,
1548 struct f2fs_map_blocks *map, int flag)
1549 {
1550 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1551 unsigned int maxblocks = map->m_len;
1552 pgoff_t pgoff = (pgoff_t)map->m_lblk;
1553 struct extent_info ei = {};
1554
1555 if (!f2fs_lookup_read_extent_cache(inode, pgoff, &ei))
1556 return false;
1557
1558 map->m_pblk = ei.blk + pgoff - ei.fofs;
1559 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgoff);
1560 map->m_flags = F2FS_MAP_MAPPED;
1561 if (map->m_next_extent)
1562 *map->m_next_extent = pgoff + map->m_len;
1563
1564 /* for hardware encryption, but to avoid potential issue in future */
1565 if (flag == F2FS_GET_BLOCK_DIO)
1566 f2fs_wait_on_block_writeback_range(inode,
1567 map->m_pblk, map->m_len);
1568
1569 if (f2fs_allow_multi_device_dio(sbi, flag)) {
1570 int bidx = f2fs_target_device_index(sbi, map->m_pblk);
1571 struct f2fs_dev_info *dev = &sbi->devs[bidx];
1572
1573 map->m_bdev = dev->bdev;
1574 map->m_pblk -= dev->start_blk;
1575 map->m_len = min(map->m_len, dev->end_blk + 1 - map->m_pblk);
1576 } else {
1577 map->m_bdev = inode->i_sb->s_bdev;
1578 }
1579 return true;
1580 }
1581
1582 /*
1583 * f2fs_map_blocks() tries to find or build mapping relationship which
1584 * maps continuous logical blocks to physical blocks, and return such
1585 * info via f2fs_map_blocks structure.
1586 */
f2fs_map_blocks(struct inode * inode,struct f2fs_map_blocks * map,int flag)1587 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, int flag)
1588 {
1589 unsigned int maxblocks = map->m_len;
1590 struct dnode_of_data dn;
1591 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1592 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1593 pgoff_t pgofs, end_offset, end;
1594 int err = 0, ofs = 1;
1595 unsigned int ofs_in_node, last_ofs_in_node;
1596 blkcnt_t prealloc;
1597 block_t blkaddr;
1598 unsigned int start_pgofs;
1599 int bidx = 0;
1600 bool is_hole;
1601
1602 if (!maxblocks)
1603 return 0;
1604
1605 if (!map->m_may_create && f2fs_map_blocks_cached(inode, map, flag))
1606 goto out;
1607
1608 map->m_bdev = inode->i_sb->s_bdev;
1609 map->m_multidev_dio =
1610 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1611
1612 map->m_len = 0;
1613 map->m_flags = 0;
1614
1615 /* it only supports block size == page size */
1616 pgofs = (pgoff_t)map->m_lblk;
1617 end = pgofs + maxblocks;
1618
1619 next_dnode:
1620 if (map->m_may_create)
1621 f2fs_map_lock(sbi, flag);
1622
1623 /* When reading holes, we need its node page */
1624 set_new_dnode(&dn, inode, NULL, NULL, 0);
1625 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1626 if (err) {
1627 if (flag == F2FS_GET_BLOCK_BMAP)
1628 map->m_pblk = 0;
1629 if (err == -ENOENT)
1630 err = f2fs_map_no_dnode(inode, map, &dn, pgofs);
1631 goto unlock_out;
1632 }
1633
1634 start_pgofs = pgofs;
1635 prealloc = 0;
1636 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1637 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1638
1639 next_block:
1640 blkaddr = f2fs_data_blkaddr(&dn);
1641 is_hole = !__is_valid_data_blkaddr(blkaddr);
1642 if (!is_hole &&
1643 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1644 err = -EFSCORRUPTED;
1645 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1646 goto sync_out;
1647 }
1648
1649 /* use out-place-update for direct IO under LFS mode */
1650 if (map->m_may_create &&
1651 (is_hole || (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO))) {
1652 if (unlikely(f2fs_cp_error(sbi))) {
1653 err = -EIO;
1654 goto sync_out;
1655 }
1656
1657 switch (flag) {
1658 case F2FS_GET_BLOCK_PRE_AIO:
1659 if (blkaddr == NULL_ADDR) {
1660 prealloc++;
1661 last_ofs_in_node = dn.ofs_in_node;
1662 }
1663 break;
1664 case F2FS_GET_BLOCK_PRE_DIO:
1665 case F2FS_GET_BLOCK_DIO:
1666 err = __allocate_data_block(&dn, map->m_seg_type);
1667 if (err)
1668 goto sync_out;
1669 if (flag == F2FS_GET_BLOCK_PRE_DIO)
1670 file_need_truncate(inode);
1671 set_inode_flag(inode, FI_APPEND_WRITE);
1672 break;
1673 default:
1674 WARN_ON_ONCE(1);
1675 err = -EIO;
1676 goto sync_out;
1677 }
1678
1679 blkaddr = dn.data_blkaddr;
1680 if (is_hole)
1681 map->m_flags |= F2FS_MAP_NEW;
1682 } else if (is_hole) {
1683 if (f2fs_compressed_file(inode) &&
1684 f2fs_sanity_check_cluster(&dn) &&
1685 (flag != F2FS_GET_BLOCK_FIEMAP ||
1686 IS_ENABLED(CONFIG_F2FS_CHECK_FS))) {
1687 err = -EFSCORRUPTED;
1688 f2fs_handle_error(sbi,
1689 ERROR_CORRUPTED_CLUSTER);
1690 goto sync_out;
1691 }
1692
1693 switch (flag) {
1694 case F2FS_GET_BLOCK_PRECACHE:
1695 goto sync_out;
1696 case F2FS_GET_BLOCK_BMAP:
1697 map->m_pblk = 0;
1698 goto sync_out;
1699 case F2FS_GET_BLOCK_FIEMAP:
1700 if (blkaddr == NULL_ADDR) {
1701 if (map->m_next_pgofs)
1702 *map->m_next_pgofs = pgofs + 1;
1703 goto sync_out;
1704 }
1705 break;
1706 default:
1707 /* for defragment case */
1708 if (map->m_next_pgofs)
1709 *map->m_next_pgofs = pgofs + 1;
1710 goto sync_out;
1711 }
1712 }
1713
1714 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1715 goto skip;
1716
1717 if (map->m_multidev_dio)
1718 bidx = f2fs_target_device_index(sbi, blkaddr);
1719
1720 if (map->m_len == 0) {
1721 /* reserved delalloc block should be mapped for fiemap. */
1722 if (blkaddr == NEW_ADDR)
1723 map->m_flags |= F2FS_MAP_DELALLOC;
1724 map->m_flags |= F2FS_MAP_MAPPED;
1725
1726 map->m_pblk = blkaddr;
1727 map->m_len = 1;
1728
1729 if (map->m_multidev_dio)
1730 map->m_bdev = FDEV(bidx).bdev;
1731 } else if ((map->m_pblk != NEW_ADDR &&
1732 blkaddr == (map->m_pblk + ofs)) ||
1733 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1734 flag == F2FS_GET_BLOCK_PRE_DIO) {
1735 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1736 goto sync_out;
1737 ofs++;
1738 map->m_len++;
1739 } else {
1740 goto sync_out;
1741 }
1742
1743 skip:
1744 dn.ofs_in_node++;
1745 pgofs++;
1746
1747 /* preallocate blocks in batch for one dnode page */
1748 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1749 (pgofs == end || dn.ofs_in_node == end_offset)) {
1750
1751 dn.ofs_in_node = ofs_in_node;
1752 err = f2fs_reserve_new_blocks(&dn, prealloc);
1753 if (err)
1754 goto sync_out;
1755
1756 map->m_len += dn.ofs_in_node - ofs_in_node;
1757 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1758 err = -ENOSPC;
1759 goto sync_out;
1760 }
1761 dn.ofs_in_node = end_offset;
1762 }
1763
1764 if (pgofs >= end)
1765 goto sync_out;
1766 else if (dn.ofs_in_node < end_offset)
1767 goto next_block;
1768
1769 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1770 if (map->m_flags & F2FS_MAP_MAPPED) {
1771 unsigned int ofs = start_pgofs - map->m_lblk;
1772
1773 f2fs_update_read_extent_cache_range(&dn,
1774 start_pgofs, map->m_pblk + ofs,
1775 map->m_len - ofs);
1776 }
1777 }
1778
1779 f2fs_put_dnode(&dn);
1780
1781 if (map->m_may_create) {
1782 f2fs_map_unlock(sbi, flag);
1783 f2fs_balance_fs(sbi, dn.node_changed);
1784 }
1785 goto next_dnode;
1786
1787 sync_out:
1788
1789 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1790 /*
1791 * for hardware encryption, but to avoid potential issue
1792 * in future
1793 */
1794 f2fs_wait_on_block_writeback_range(inode,
1795 map->m_pblk, map->m_len);
1796
1797 if (map->m_multidev_dio) {
1798 block_t blk_addr = map->m_pblk;
1799
1800 bidx = f2fs_target_device_index(sbi, map->m_pblk);
1801
1802 map->m_bdev = FDEV(bidx).bdev;
1803 map->m_pblk -= FDEV(bidx).start_blk;
1804
1805 if (map->m_may_create)
1806 f2fs_update_device_state(sbi, inode->i_ino,
1807 blk_addr, map->m_len);
1808
1809 f2fs_bug_on(sbi, blk_addr + map->m_len >
1810 FDEV(bidx).end_blk + 1);
1811 }
1812 }
1813
1814 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1815 if (map->m_flags & F2FS_MAP_MAPPED) {
1816 unsigned int ofs = start_pgofs - map->m_lblk;
1817
1818 f2fs_update_read_extent_cache_range(&dn,
1819 start_pgofs, map->m_pblk + ofs,
1820 map->m_len - ofs);
1821 }
1822 if (map->m_next_extent)
1823 *map->m_next_extent = pgofs + 1;
1824 }
1825 f2fs_put_dnode(&dn);
1826 unlock_out:
1827 if (map->m_may_create) {
1828 f2fs_map_unlock(sbi, flag);
1829 f2fs_balance_fs(sbi, dn.node_changed);
1830 }
1831 out:
1832 trace_f2fs_map_blocks(inode, map, flag, err);
1833 return err;
1834 }
1835
f2fs_overwrite_io(struct inode * inode,loff_t pos,size_t len)1836 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1837 {
1838 struct f2fs_map_blocks map;
1839 block_t last_lblk;
1840 int err;
1841
1842 if (pos + len > i_size_read(inode))
1843 return false;
1844
1845 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1846 map.m_next_pgofs = NULL;
1847 map.m_next_extent = NULL;
1848 map.m_seg_type = NO_CHECK_TYPE;
1849 map.m_may_create = false;
1850 last_lblk = F2FS_BLK_ALIGN(pos + len);
1851
1852 while (map.m_lblk < last_lblk) {
1853 map.m_len = last_lblk - map.m_lblk;
1854 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT);
1855 if (err || map.m_len == 0)
1856 return false;
1857 map.m_lblk += map.m_len;
1858 }
1859 return true;
1860 }
1861
bytes_to_blks(struct inode * inode,u64 bytes)1862 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1863 {
1864 return (bytes >> inode->i_blkbits);
1865 }
1866
blks_to_bytes(struct inode * inode,u64 blks)1867 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1868 {
1869 return (blks << inode->i_blkbits);
1870 }
1871
f2fs_xattr_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo)1872 static int f2fs_xattr_fiemap(struct inode *inode,
1873 struct fiemap_extent_info *fieinfo)
1874 {
1875 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1876 struct page *page;
1877 struct node_info ni;
1878 __u64 phys = 0, len;
1879 __u32 flags;
1880 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1881 int err = 0;
1882
1883 if (f2fs_has_inline_xattr(inode)) {
1884 int offset;
1885
1886 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1887 inode->i_ino, false);
1888 if (!page)
1889 return -ENOMEM;
1890
1891 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1892 if (err) {
1893 f2fs_put_page(page, 1);
1894 return err;
1895 }
1896
1897 phys = blks_to_bytes(inode, ni.blk_addr);
1898 offset = offsetof(struct f2fs_inode, i_addr) +
1899 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1900 get_inline_xattr_addrs(inode));
1901
1902 phys += offset;
1903 len = inline_xattr_size(inode);
1904
1905 f2fs_put_page(page, 1);
1906
1907 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1908
1909 if (!xnid)
1910 flags |= FIEMAP_EXTENT_LAST;
1911
1912 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1913 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1914 if (err)
1915 return err;
1916 }
1917
1918 if (xnid) {
1919 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1920 if (!page)
1921 return -ENOMEM;
1922
1923 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1924 if (err) {
1925 f2fs_put_page(page, 1);
1926 return err;
1927 }
1928
1929 phys = blks_to_bytes(inode, ni.blk_addr);
1930 len = inode->i_sb->s_blocksize;
1931
1932 f2fs_put_page(page, 1);
1933
1934 flags = FIEMAP_EXTENT_LAST;
1935 }
1936
1937 if (phys) {
1938 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1939 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1940 }
1941
1942 return (err < 0 ? err : 0);
1943 }
1944
max_inode_blocks(struct inode * inode)1945 static loff_t max_inode_blocks(struct inode *inode)
1946 {
1947 loff_t result = ADDRS_PER_INODE(inode);
1948 loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1949
1950 /* two direct node blocks */
1951 result += (leaf_count * 2);
1952
1953 /* two indirect node blocks */
1954 leaf_count *= NIDS_PER_BLOCK;
1955 result += (leaf_count * 2);
1956
1957 /* one double indirect node block */
1958 leaf_count *= NIDS_PER_BLOCK;
1959 result += leaf_count;
1960
1961 return result;
1962 }
1963
f2fs_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 start,u64 len)1964 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1965 u64 start, u64 len)
1966 {
1967 struct f2fs_map_blocks map;
1968 sector_t start_blk, last_blk;
1969 pgoff_t next_pgofs;
1970 u64 logical = 0, phys = 0, size = 0;
1971 u32 flags = 0;
1972 int ret = 0;
1973 bool compr_cluster = false, compr_appended;
1974 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1975 unsigned int count_in_cluster = 0;
1976 loff_t maxbytes;
1977
1978 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1979 ret = f2fs_precache_extents(inode);
1980 if (ret)
1981 return ret;
1982 }
1983
1984 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1985 if (ret)
1986 return ret;
1987
1988 inode_lock(inode);
1989
1990 maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1991 if (start > maxbytes) {
1992 ret = -EFBIG;
1993 goto out;
1994 }
1995
1996 if (len > maxbytes || (maxbytes - len) < start)
1997 len = maxbytes - start;
1998
1999 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
2000 ret = f2fs_xattr_fiemap(inode, fieinfo);
2001 goto out;
2002 }
2003
2004 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
2005 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
2006 if (ret != -EAGAIN)
2007 goto out;
2008 }
2009
2010 if (bytes_to_blks(inode, len) == 0)
2011 len = blks_to_bytes(inode, 1);
2012
2013 start_blk = bytes_to_blks(inode, start);
2014 last_blk = bytes_to_blks(inode, start + len - 1);
2015
2016 next:
2017 memset(&map, 0, sizeof(map));
2018 map.m_lblk = start_blk;
2019 map.m_len = bytes_to_blks(inode, len);
2020 map.m_next_pgofs = &next_pgofs;
2021 map.m_seg_type = NO_CHECK_TYPE;
2022
2023 if (compr_cluster) {
2024 map.m_lblk += 1;
2025 map.m_len = cluster_size - count_in_cluster;
2026 }
2027
2028 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
2029 if (ret)
2030 goto out;
2031
2032 /* HOLE */
2033 if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
2034 start_blk = next_pgofs;
2035
2036 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
2037 max_inode_blocks(inode)))
2038 goto prep_next;
2039
2040 flags |= FIEMAP_EXTENT_LAST;
2041 }
2042
2043 compr_appended = false;
2044 /* In a case of compressed cluster, append this to the last extent */
2045 if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) ||
2046 !(map.m_flags & F2FS_MAP_FLAGS))) {
2047 compr_appended = true;
2048 goto skip_fill;
2049 }
2050
2051 if (size) {
2052 flags |= FIEMAP_EXTENT_MERGED;
2053 if (IS_ENCRYPTED(inode))
2054 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
2055
2056 ret = fiemap_fill_next_extent(fieinfo, logical,
2057 phys, size, flags);
2058 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
2059 if (ret)
2060 goto out;
2061 size = 0;
2062 }
2063
2064 if (start_blk > last_blk)
2065 goto out;
2066
2067 skip_fill:
2068 if (map.m_pblk == COMPRESS_ADDR) {
2069 compr_cluster = true;
2070 count_in_cluster = 1;
2071 } else if (compr_appended) {
2072 unsigned int appended_blks = cluster_size -
2073 count_in_cluster + 1;
2074 size += blks_to_bytes(inode, appended_blks);
2075 start_blk += appended_blks;
2076 compr_cluster = false;
2077 } else {
2078 logical = blks_to_bytes(inode, start_blk);
2079 phys = __is_valid_data_blkaddr(map.m_pblk) ?
2080 blks_to_bytes(inode, map.m_pblk) : 0;
2081 size = blks_to_bytes(inode, map.m_len);
2082 flags = 0;
2083
2084 if (compr_cluster) {
2085 flags = FIEMAP_EXTENT_ENCODED;
2086 count_in_cluster += map.m_len;
2087 if (count_in_cluster == cluster_size) {
2088 compr_cluster = false;
2089 size += blks_to_bytes(inode, 1);
2090 }
2091 } else if (map.m_flags & F2FS_MAP_DELALLOC) {
2092 flags = FIEMAP_EXTENT_UNWRITTEN;
2093 }
2094
2095 start_blk += bytes_to_blks(inode, size);
2096 }
2097
2098 prep_next:
2099 cond_resched();
2100 if (fatal_signal_pending(current))
2101 ret = -EINTR;
2102 else
2103 goto next;
2104 out:
2105 if (ret == 1)
2106 ret = 0;
2107
2108 inode_unlock(inode);
2109 return ret;
2110 }
2111
f2fs_readpage_limit(struct inode * inode)2112 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2113 {
2114 if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
2115 return inode->i_sb->s_maxbytes;
2116
2117 return i_size_read(inode);
2118 }
2119
f2fs_read_single_page(struct inode * inode,struct page * page,unsigned nr_pages,struct f2fs_map_blocks * map,struct bio ** bio_ret,sector_t * last_block_in_bio,bool is_readahead)2120 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2121 unsigned nr_pages,
2122 struct f2fs_map_blocks *map,
2123 struct bio **bio_ret,
2124 sector_t *last_block_in_bio,
2125 bool is_readahead)
2126 {
2127 struct bio *bio = *bio_ret;
2128 const unsigned blocksize = blks_to_bytes(inode, 1);
2129 sector_t block_in_file;
2130 sector_t last_block;
2131 sector_t last_block_in_file;
2132 sector_t block_nr;
2133 int ret = 0;
2134
2135 block_in_file = (sector_t)page_index(page);
2136 last_block = block_in_file + nr_pages;
2137 last_block_in_file = bytes_to_blks(inode,
2138 f2fs_readpage_limit(inode) + blocksize - 1);
2139 if (last_block > last_block_in_file)
2140 last_block = last_block_in_file;
2141
2142 /* just zeroing out page which is beyond EOF */
2143 if (block_in_file >= last_block)
2144 goto zero_out;
2145 /*
2146 * Map blocks using the previous result first.
2147 */
2148 if ((map->m_flags & F2FS_MAP_MAPPED) &&
2149 block_in_file > map->m_lblk &&
2150 block_in_file < (map->m_lblk + map->m_len))
2151 goto got_it;
2152
2153 /*
2154 * Then do more f2fs_map_blocks() calls until we are
2155 * done with this page.
2156 */
2157 map->m_lblk = block_in_file;
2158 map->m_len = last_block - block_in_file;
2159
2160 ret = f2fs_map_blocks(inode, map, F2FS_GET_BLOCK_DEFAULT);
2161 if (ret)
2162 goto out;
2163 got_it:
2164 if ((map->m_flags & F2FS_MAP_MAPPED)) {
2165 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2166 SetPageMappedToDisk(page);
2167
2168 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2169 DATA_GENERIC_ENHANCE_READ)) {
2170 ret = -EFSCORRUPTED;
2171 f2fs_handle_error(F2FS_I_SB(inode),
2172 ERROR_INVALID_BLKADDR);
2173 goto out;
2174 }
2175 } else {
2176 zero_out:
2177 zero_user_segment(page, 0, PAGE_SIZE);
2178 if (f2fs_need_verity(inode, page->index) &&
2179 !fsverity_verify_page(page)) {
2180 ret = -EIO;
2181 goto out;
2182 }
2183 if (!PageUptodate(page))
2184 SetPageUptodate(page);
2185 unlock_page(page);
2186 goto out;
2187 }
2188
2189 /*
2190 * This page will go to BIO. Do we need to send this
2191 * BIO off first?
2192 */
2193 if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2194 *last_block_in_bio, block_nr) ||
2195 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2196 submit_and_realloc:
2197 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2198 bio = NULL;
2199 }
2200 if (bio == NULL) {
2201 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2202 is_readahead ? REQ_RAHEAD : 0, page->index,
2203 false);
2204 if (IS_ERR(bio)) {
2205 ret = PTR_ERR(bio);
2206 bio = NULL;
2207 goto out;
2208 }
2209 }
2210
2211 /*
2212 * If the page is under writeback, we need to wait for
2213 * its completion to see the correct decrypted data.
2214 */
2215 f2fs_wait_on_block_writeback(inode, block_nr);
2216
2217 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2218 goto submit_and_realloc;
2219
2220 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2221 f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO,
2222 F2FS_BLKSIZE);
2223 *last_block_in_bio = block_nr;
2224 out:
2225 *bio_ret = bio;
2226 return ret;
2227 }
2228
2229 #ifdef CONFIG_F2FS_FS_COMPRESSION
f2fs_read_multi_pages(struct compress_ctx * cc,struct bio ** bio_ret,unsigned nr_pages,sector_t * last_block_in_bio,bool is_readahead,bool for_write)2230 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2231 unsigned nr_pages, sector_t *last_block_in_bio,
2232 bool is_readahead, bool for_write)
2233 {
2234 struct dnode_of_data dn;
2235 struct inode *inode = cc->inode;
2236 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2237 struct bio *bio = *bio_ret;
2238 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2239 sector_t last_block_in_file;
2240 const unsigned blocksize = blks_to_bytes(inode, 1);
2241 struct decompress_io_ctx *dic = NULL;
2242 struct extent_info ei = {};
2243 bool from_dnode = true;
2244 int i;
2245 int ret = 0;
2246
2247 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2248
2249 last_block_in_file = bytes_to_blks(inode,
2250 f2fs_readpage_limit(inode) + blocksize - 1);
2251
2252 /* get rid of pages beyond EOF */
2253 for (i = 0; i < cc->cluster_size; i++) {
2254 struct page *page = cc->rpages[i];
2255
2256 if (!page)
2257 continue;
2258 if ((sector_t)page->index >= last_block_in_file) {
2259 zero_user_segment(page, 0, PAGE_SIZE);
2260 if (!PageUptodate(page))
2261 SetPageUptodate(page);
2262 } else if (!PageUptodate(page)) {
2263 continue;
2264 }
2265 unlock_page(page);
2266 if (for_write)
2267 put_page(page);
2268 cc->rpages[i] = NULL;
2269 cc->nr_rpages--;
2270 }
2271
2272 /* we are done since all pages are beyond EOF */
2273 if (f2fs_cluster_is_empty(cc))
2274 goto out;
2275
2276 if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei))
2277 from_dnode = false;
2278
2279 if (!from_dnode)
2280 goto skip_reading_dnode;
2281
2282 set_new_dnode(&dn, inode, NULL, NULL, 0);
2283 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2284 if (ret)
2285 goto out;
2286
2287 if (unlikely(f2fs_cp_error(sbi))) {
2288 ret = -EIO;
2289 goto out_put_dnode;
2290 }
2291 f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2292
2293 skip_reading_dnode:
2294 for (i = 1; i < cc->cluster_size; i++) {
2295 block_t blkaddr;
2296
2297 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2298 dn.ofs_in_node + i) :
2299 ei.blk + i - 1;
2300
2301 if (!__is_valid_data_blkaddr(blkaddr))
2302 break;
2303
2304 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2305 ret = -EFAULT;
2306 goto out_put_dnode;
2307 }
2308 cc->nr_cpages++;
2309
2310 if (!from_dnode && i >= ei.c_len)
2311 break;
2312 }
2313
2314 /* nothing to decompress */
2315 if (cc->nr_cpages == 0) {
2316 ret = 0;
2317 goto out_put_dnode;
2318 }
2319
2320 dic = f2fs_alloc_dic(cc);
2321 if (IS_ERR(dic)) {
2322 ret = PTR_ERR(dic);
2323 goto out_put_dnode;
2324 }
2325
2326 for (i = 0; i < cc->nr_cpages; i++) {
2327 struct page *page = dic->cpages[i];
2328 block_t blkaddr;
2329 struct bio_post_read_ctx *ctx;
2330
2331 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2332 dn.ofs_in_node + i + 1) :
2333 ei.blk + i;
2334
2335 f2fs_wait_on_block_writeback(inode, blkaddr);
2336
2337 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2338 if (atomic_dec_and_test(&dic->remaining_pages)) {
2339 f2fs_decompress_cluster(dic, true);
2340 break;
2341 }
2342 continue;
2343 }
2344
2345 if (bio && (!page_is_mergeable(sbi, bio,
2346 *last_block_in_bio, blkaddr) ||
2347 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2348 submit_and_realloc:
2349 f2fs_submit_read_bio(sbi, bio, DATA);
2350 bio = NULL;
2351 }
2352
2353 if (!bio) {
2354 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2355 is_readahead ? REQ_RAHEAD : 0,
2356 page->index, for_write);
2357 if (IS_ERR(bio)) {
2358 ret = PTR_ERR(bio);
2359 f2fs_decompress_end_io(dic, ret, true);
2360 f2fs_put_dnode(&dn);
2361 *bio_ret = NULL;
2362 return ret;
2363 }
2364 }
2365
2366 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2367 goto submit_and_realloc;
2368
2369 ctx = get_post_read_ctx(bio);
2370 ctx->enabled_steps |= STEP_DECOMPRESS;
2371 refcount_inc(&dic->refcnt);
2372
2373 inc_page_count(sbi, F2FS_RD_DATA);
2374 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
2375 *last_block_in_bio = blkaddr;
2376 }
2377
2378 if (from_dnode)
2379 f2fs_put_dnode(&dn);
2380
2381 *bio_ret = bio;
2382 return 0;
2383
2384 out_put_dnode:
2385 if (from_dnode)
2386 f2fs_put_dnode(&dn);
2387 out:
2388 for (i = 0; i < cc->cluster_size; i++) {
2389 if (cc->rpages[i]) {
2390 ClearPageUptodate(cc->rpages[i]);
2391 unlock_page(cc->rpages[i]);
2392 }
2393 }
2394 *bio_ret = bio;
2395 return ret;
2396 }
2397 #endif
2398
2399 /*
2400 * This function was originally taken from fs/mpage.c, and customized for f2fs.
2401 * Major change was from block_size == page_size in f2fs by default.
2402 */
f2fs_mpage_readpages(struct inode * inode,struct readahead_control * rac,struct page * page)2403 static int f2fs_mpage_readpages(struct inode *inode,
2404 struct readahead_control *rac, struct page *page)
2405 {
2406 struct bio *bio = NULL;
2407 sector_t last_block_in_bio = 0;
2408 struct f2fs_map_blocks map;
2409 #ifdef CONFIG_F2FS_FS_COMPRESSION
2410 struct compress_ctx cc = {
2411 .inode = inode,
2412 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2413 .cluster_size = F2FS_I(inode)->i_cluster_size,
2414 .cluster_idx = NULL_CLUSTER,
2415 .rpages = NULL,
2416 .cpages = NULL,
2417 .nr_rpages = 0,
2418 .nr_cpages = 0,
2419 };
2420 pgoff_t nc_cluster_idx = NULL_CLUSTER;
2421 #endif
2422 unsigned nr_pages = rac ? readahead_count(rac) : 1;
2423 unsigned max_nr_pages = nr_pages;
2424 int ret = 0;
2425
2426 map.m_pblk = 0;
2427 map.m_lblk = 0;
2428 map.m_len = 0;
2429 map.m_flags = 0;
2430 map.m_next_pgofs = NULL;
2431 map.m_next_extent = NULL;
2432 map.m_seg_type = NO_CHECK_TYPE;
2433 map.m_may_create = false;
2434
2435 for (; nr_pages; nr_pages--) {
2436 if (rac) {
2437 page = readahead_page(rac);
2438 prefetchw(&page->flags);
2439 }
2440
2441 #ifdef CONFIG_F2FS_FS_COMPRESSION
2442 if (f2fs_compressed_file(inode)) {
2443 /* there are remained compressed pages, submit them */
2444 if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2445 ret = f2fs_read_multi_pages(&cc, &bio,
2446 max_nr_pages,
2447 &last_block_in_bio,
2448 rac != NULL, false);
2449 f2fs_destroy_compress_ctx(&cc, false);
2450 if (ret)
2451 goto set_error_page;
2452 }
2453 if (cc.cluster_idx == NULL_CLUSTER) {
2454 if (nc_cluster_idx ==
2455 page->index >> cc.log_cluster_size) {
2456 goto read_single_page;
2457 }
2458
2459 ret = f2fs_is_compressed_cluster(inode, page->index);
2460 if (ret < 0)
2461 goto set_error_page;
2462 else if (!ret) {
2463 nc_cluster_idx =
2464 page->index >> cc.log_cluster_size;
2465 goto read_single_page;
2466 }
2467
2468 nc_cluster_idx = NULL_CLUSTER;
2469 }
2470 ret = f2fs_init_compress_ctx(&cc);
2471 if (ret)
2472 goto set_error_page;
2473
2474 f2fs_compress_ctx_add_page(&cc, page);
2475
2476 goto next_page;
2477 }
2478 read_single_page:
2479 #endif
2480
2481 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2482 &bio, &last_block_in_bio, rac);
2483 if (ret) {
2484 #ifdef CONFIG_F2FS_FS_COMPRESSION
2485 set_error_page:
2486 #endif
2487 zero_user_segment(page, 0, PAGE_SIZE);
2488 unlock_page(page);
2489 }
2490 #ifdef CONFIG_F2FS_FS_COMPRESSION
2491 next_page:
2492 #endif
2493 if (rac)
2494 put_page(page);
2495
2496 #ifdef CONFIG_F2FS_FS_COMPRESSION
2497 if (f2fs_compressed_file(inode)) {
2498 /* last page */
2499 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2500 ret = f2fs_read_multi_pages(&cc, &bio,
2501 max_nr_pages,
2502 &last_block_in_bio,
2503 rac != NULL, false);
2504 f2fs_destroy_compress_ctx(&cc, false);
2505 }
2506 }
2507 #endif
2508 }
2509 if (bio)
2510 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2511 return ret;
2512 }
2513
f2fs_read_data_folio(struct file * file,struct folio * folio)2514 static int f2fs_read_data_folio(struct file *file, struct folio *folio)
2515 {
2516 struct page *page = &folio->page;
2517 struct inode *inode = page_file_mapping(page)->host;
2518 int ret = -EAGAIN;
2519
2520 trace_f2fs_readpage(page, DATA);
2521
2522 if (!f2fs_is_compress_backend_ready(inode)) {
2523 unlock_page(page);
2524 return -EOPNOTSUPP;
2525 }
2526
2527 /* If the file has inline data, try to read it directly */
2528 if (f2fs_has_inline_data(inode))
2529 ret = f2fs_read_inline_data(inode, page);
2530 if (ret == -EAGAIN)
2531 ret = f2fs_mpage_readpages(inode, NULL, page);
2532 return ret;
2533 }
2534
f2fs_readahead(struct readahead_control * rac)2535 static void f2fs_readahead(struct readahead_control *rac)
2536 {
2537 struct inode *inode = rac->mapping->host;
2538
2539 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2540
2541 if (!f2fs_is_compress_backend_ready(inode))
2542 return;
2543
2544 /* If the file has inline data, skip readahead */
2545 if (f2fs_has_inline_data(inode))
2546 return;
2547
2548 f2fs_mpage_readpages(inode, rac, NULL);
2549 }
2550
f2fs_encrypt_one_page(struct f2fs_io_info * fio)2551 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2552 {
2553 struct inode *inode = fio->page->mapping->host;
2554 struct page *mpage, *page;
2555 gfp_t gfp_flags = GFP_NOFS;
2556
2557 if (!f2fs_encrypted_file(inode))
2558 return 0;
2559
2560 page = fio->compressed_page ? fio->compressed_page : fio->page;
2561
2562 if (fscrypt_inode_uses_inline_crypto(inode))
2563 return 0;
2564
2565 retry_encrypt:
2566 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2567 PAGE_SIZE, 0, gfp_flags);
2568 if (IS_ERR(fio->encrypted_page)) {
2569 /* flush pending IOs and wait for a while in the ENOMEM case */
2570 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2571 f2fs_flush_merged_writes(fio->sbi);
2572 memalloc_retry_wait(GFP_NOFS);
2573 gfp_flags |= __GFP_NOFAIL;
2574 goto retry_encrypt;
2575 }
2576 return PTR_ERR(fio->encrypted_page);
2577 }
2578
2579 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2580 if (mpage) {
2581 if (PageUptodate(mpage))
2582 memcpy(page_address(mpage),
2583 page_address(fio->encrypted_page), PAGE_SIZE);
2584 f2fs_put_page(mpage, 1);
2585 }
2586 return 0;
2587 }
2588
check_inplace_update_policy(struct inode * inode,struct f2fs_io_info * fio)2589 static inline bool check_inplace_update_policy(struct inode *inode,
2590 struct f2fs_io_info *fio)
2591 {
2592 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2593
2594 if (IS_F2FS_IPU_HONOR_OPU_WRITE(sbi) &&
2595 is_inode_flag_set(inode, FI_OPU_WRITE))
2596 return false;
2597 if (IS_F2FS_IPU_FORCE(sbi))
2598 return true;
2599 if (IS_F2FS_IPU_SSR(sbi) && f2fs_need_SSR(sbi))
2600 return true;
2601 if (IS_F2FS_IPU_UTIL(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util)
2602 return true;
2603 if (IS_F2FS_IPU_SSR_UTIL(sbi) && f2fs_need_SSR(sbi) &&
2604 utilization(sbi) > SM_I(sbi)->min_ipu_util)
2605 return true;
2606
2607 /*
2608 * IPU for rewrite async pages
2609 */
2610 if (IS_F2FS_IPU_ASYNC(sbi) && fio && fio->op == REQ_OP_WRITE &&
2611 !(fio->op_flags & REQ_SYNC) && !IS_ENCRYPTED(inode))
2612 return true;
2613
2614 /* this is only set during fdatasync */
2615 if (IS_F2FS_IPU_FSYNC(sbi) && is_inode_flag_set(inode, FI_NEED_IPU))
2616 return true;
2617
2618 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2619 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2620 return true;
2621
2622 return false;
2623 }
2624
f2fs_should_update_inplace(struct inode * inode,struct f2fs_io_info * fio)2625 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2626 {
2627 /* swap file is migrating in aligned write mode */
2628 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2629 return false;
2630
2631 if (f2fs_is_pinned_file(inode))
2632 return true;
2633
2634 /* if this is cold file, we should overwrite to avoid fragmentation */
2635 if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE))
2636 return true;
2637
2638 return check_inplace_update_policy(inode, fio);
2639 }
2640
f2fs_should_update_outplace(struct inode * inode,struct f2fs_io_info * fio)2641 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2642 {
2643 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2644
2645 /* The below cases were checked when setting it. */
2646 if (f2fs_is_pinned_file(inode))
2647 return false;
2648 if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2649 return true;
2650 if (f2fs_lfs_mode(sbi))
2651 return true;
2652 if (S_ISDIR(inode->i_mode))
2653 return true;
2654 if (IS_NOQUOTA(inode))
2655 return true;
2656 if (f2fs_is_atomic_file(inode))
2657 return true;
2658
2659 /* swap file is migrating in aligned write mode */
2660 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2661 return true;
2662
2663 if (is_inode_flag_set(inode, FI_OPU_WRITE))
2664 return true;
2665
2666 if (fio) {
2667 if (page_private_gcing(fio->page))
2668 return true;
2669 if (page_private_dummy(fio->page))
2670 return true;
2671 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2672 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2673 return true;
2674 }
2675 return false;
2676 }
2677
need_inplace_update(struct f2fs_io_info * fio)2678 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2679 {
2680 struct inode *inode = fio->page->mapping->host;
2681
2682 if (f2fs_should_update_outplace(inode, fio))
2683 return false;
2684
2685 return f2fs_should_update_inplace(inode, fio);
2686 }
2687
f2fs_do_write_data_page(struct f2fs_io_info * fio)2688 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2689 {
2690 struct page *page = fio->page;
2691 struct inode *inode = page->mapping->host;
2692 struct dnode_of_data dn;
2693 struct node_info ni;
2694 bool ipu_force = false;
2695 int err = 0;
2696
2697 /* Use COW inode to make dnode_of_data for atomic write */
2698 if (f2fs_is_atomic_file(inode))
2699 set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
2700 else
2701 set_new_dnode(&dn, inode, NULL, NULL, 0);
2702
2703 if (need_inplace_update(fio) &&
2704 f2fs_lookup_read_extent_cache_block(inode, page->index,
2705 &fio->old_blkaddr)) {
2706 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2707 DATA_GENERIC_ENHANCE)) {
2708 f2fs_handle_error(fio->sbi,
2709 ERROR_INVALID_BLKADDR);
2710 return -EFSCORRUPTED;
2711 }
2712
2713 ipu_force = true;
2714 fio->need_lock = LOCK_DONE;
2715 goto got_it;
2716 }
2717
2718 /* Deadlock due to between page->lock and f2fs_lock_op */
2719 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2720 return -EAGAIN;
2721
2722 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2723 if (err)
2724 goto out;
2725
2726 fio->old_blkaddr = dn.data_blkaddr;
2727
2728 /* This page is already truncated */
2729 if (fio->old_blkaddr == NULL_ADDR) {
2730 ClearPageUptodate(page);
2731 clear_page_private_gcing(page);
2732 goto out_writepage;
2733 }
2734 got_it:
2735 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2736 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2737 DATA_GENERIC_ENHANCE)) {
2738 err = -EFSCORRUPTED;
2739 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
2740 goto out_writepage;
2741 }
2742
2743 /* wait for GCed page writeback via META_MAPPING */
2744 if (fio->post_read)
2745 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2746
2747 /*
2748 * If current allocation needs SSR,
2749 * it had better in-place writes for updated data.
2750 */
2751 if (ipu_force ||
2752 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2753 need_inplace_update(fio))) {
2754 err = f2fs_encrypt_one_page(fio);
2755 if (err)
2756 goto out_writepage;
2757
2758 set_page_writeback(page);
2759 f2fs_put_dnode(&dn);
2760 if (fio->need_lock == LOCK_REQ)
2761 f2fs_unlock_op(fio->sbi);
2762 err = f2fs_inplace_write_data(fio);
2763 if (err) {
2764 if (fscrypt_inode_uses_fs_layer_crypto(inode))
2765 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2766 if (PageWriteback(page))
2767 end_page_writeback(page);
2768 } else {
2769 set_inode_flag(inode, FI_UPDATE_WRITE);
2770 }
2771 trace_f2fs_do_write_data_page(fio->page, IPU);
2772 return err;
2773 }
2774
2775 if (fio->need_lock == LOCK_RETRY) {
2776 if (!f2fs_trylock_op(fio->sbi)) {
2777 err = -EAGAIN;
2778 goto out_writepage;
2779 }
2780 fio->need_lock = LOCK_REQ;
2781 }
2782
2783 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2784 if (err)
2785 goto out_writepage;
2786
2787 fio->version = ni.version;
2788
2789 err = f2fs_encrypt_one_page(fio);
2790 if (err)
2791 goto out_writepage;
2792
2793 set_page_writeback(page);
2794
2795 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2796 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2797
2798 /* LFS mode write path */
2799 f2fs_outplace_write_data(&dn, fio);
2800 trace_f2fs_do_write_data_page(page, OPU);
2801 set_inode_flag(inode, FI_APPEND_WRITE);
2802 out_writepage:
2803 f2fs_put_dnode(&dn);
2804 out:
2805 if (fio->need_lock == LOCK_REQ)
2806 f2fs_unlock_op(fio->sbi);
2807 return err;
2808 }
2809
f2fs_write_single_data_page(struct page * page,int * submitted,struct bio ** bio,sector_t * last_block,struct writeback_control * wbc,enum iostat_type io_type,int compr_blocks,bool allow_balance)2810 int f2fs_write_single_data_page(struct page *page, int *submitted,
2811 struct bio **bio,
2812 sector_t *last_block,
2813 struct writeback_control *wbc,
2814 enum iostat_type io_type,
2815 int compr_blocks,
2816 bool allow_balance)
2817 {
2818 struct inode *inode = page->mapping->host;
2819 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2820 loff_t i_size = i_size_read(inode);
2821 const pgoff_t end_index = ((unsigned long long)i_size)
2822 >> PAGE_SHIFT;
2823 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2824 unsigned offset = 0;
2825 bool need_balance_fs = false;
2826 bool quota_inode = IS_NOQUOTA(inode);
2827 int err = 0;
2828 struct f2fs_io_info fio = {
2829 .sbi = sbi,
2830 .ino = inode->i_ino,
2831 .type = DATA,
2832 .op = REQ_OP_WRITE,
2833 .op_flags = wbc_to_write_flags(wbc),
2834 .old_blkaddr = NULL_ADDR,
2835 .page = page,
2836 .encrypted_page = NULL,
2837 .submitted = 0,
2838 .compr_blocks = compr_blocks,
2839 .need_lock = compr_blocks ? LOCK_DONE : LOCK_RETRY,
2840 .post_read = f2fs_post_read_required(inode) ? 1 : 0,
2841 .io_type = io_type,
2842 .io_wbc = wbc,
2843 .bio = bio,
2844 .last_block = last_block,
2845 };
2846
2847 trace_f2fs_writepage(page, DATA);
2848
2849 /* we should bypass data pages to proceed the kworker jobs */
2850 if (unlikely(f2fs_cp_error(sbi))) {
2851 mapping_set_error(page->mapping, -EIO);
2852 /*
2853 * don't drop any dirty dentry pages for keeping lastest
2854 * directory structure.
2855 */
2856 if (S_ISDIR(inode->i_mode) &&
2857 !is_sbi_flag_set(sbi, SBI_IS_CLOSE))
2858 goto redirty_out;
2859
2860 /* keep data pages in remount-ro mode */
2861 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
2862 goto redirty_out;
2863 goto out;
2864 }
2865
2866 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2867 goto redirty_out;
2868
2869 if (page->index < end_index ||
2870 f2fs_verity_in_progress(inode) ||
2871 compr_blocks)
2872 goto write;
2873
2874 /*
2875 * If the offset is out-of-range of file size,
2876 * this page does not have to be written to disk.
2877 */
2878 offset = i_size & (PAGE_SIZE - 1);
2879 if ((page->index >= end_index + 1) || !offset)
2880 goto out;
2881
2882 zero_user_segment(page, offset, PAGE_SIZE);
2883 write:
2884 /* Dentry/quota blocks are controlled by checkpoint */
2885 if (S_ISDIR(inode->i_mode) || quota_inode) {
2886 /*
2887 * We need to wait for node_write to avoid block allocation during
2888 * checkpoint. This can only happen to quota writes which can cause
2889 * the below discard race condition.
2890 */
2891 if (quota_inode)
2892 f2fs_down_read(&sbi->node_write);
2893
2894 fio.need_lock = LOCK_DONE;
2895 err = f2fs_do_write_data_page(&fio);
2896
2897 if (quota_inode)
2898 f2fs_up_read(&sbi->node_write);
2899
2900 goto done;
2901 }
2902
2903 if (!wbc->for_reclaim)
2904 need_balance_fs = true;
2905 else if (has_not_enough_free_secs(sbi, 0, 0))
2906 goto redirty_out;
2907 else
2908 set_inode_flag(inode, FI_HOT_DATA);
2909
2910 err = -EAGAIN;
2911 if (f2fs_has_inline_data(inode)) {
2912 err = f2fs_write_inline_data(inode, page);
2913 if (!err)
2914 goto out;
2915 }
2916
2917 if (err == -EAGAIN) {
2918 err = f2fs_do_write_data_page(&fio);
2919 if (err == -EAGAIN) {
2920 f2fs_bug_on(sbi, compr_blocks);
2921 fio.need_lock = LOCK_REQ;
2922 err = f2fs_do_write_data_page(&fio);
2923 }
2924 }
2925
2926 if (err) {
2927 file_set_keep_isize(inode);
2928 } else {
2929 spin_lock(&F2FS_I(inode)->i_size_lock);
2930 if (F2FS_I(inode)->last_disk_size < psize)
2931 F2FS_I(inode)->last_disk_size = psize;
2932 spin_unlock(&F2FS_I(inode)->i_size_lock);
2933 }
2934
2935 done:
2936 if (err && err != -ENOENT)
2937 goto redirty_out;
2938
2939 out:
2940 inode_dec_dirty_pages(inode);
2941 if (err) {
2942 ClearPageUptodate(page);
2943 clear_page_private_gcing(page);
2944 }
2945
2946 if (wbc->for_reclaim) {
2947 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2948 clear_inode_flag(inode, FI_HOT_DATA);
2949 f2fs_remove_dirty_inode(inode);
2950 submitted = NULL;
2951 }
2952 unlock_page(page);
2953 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2954 !F2FS_I(inode)->wb_task && allow_balance)
2955 f2fs_balance_fs(sbi, need_balance_fs);
2956
2957 if (unlikely(f2fs_cp_error(sbi))) {
2958 f2fs_submit_merged_write(sbi, DATA);
2959 if (bio && *bio)
2960 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2961 submitted = NULL;
2962 }
2963
2964 if (submitted)
2965 *submitted = fio.submitted;
2966
2967 return 0;
2968
2969 redirty_out:
2970 redirty_page_for_writepage(wbc, page);
2971 /*
2972 * pageout() in MM translates EAGAIN, so calls handle_write_error()
2973 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2974 * file_write_and_wait_range() will see EIO error, which is critical
2975 * to return value of fsync() followed by atomic_write failure to user.
2976 */
2977 if (!err || wbc->for_reclaim)
2978 return AOP_WRITEPAGE_ACTIVATE;
2979 unlock_page(page);
2980 return err;
2981 }
2982
f2fs_write_data_page(struct page * page,struct writeback_control * wbc)2983 static int f2fs_write_data_page(struct page *page,
2984 struct writeback_control *wbc)
2985 {
2986 #ifdef CONFIG_F2FS_FS_COMPRESSION
2987 struct inode *inode = page->mapping->host;
2988
2989 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2990 goto out;
2991
2992 if (f2fs_compressed_file(inode)) {
2993 if (f2fs_is_compressed_cluster(inode, page->index)) {
2994 redirty_page_for_writepage(wbc, page);
2995 return AOP_WRITEPAGE_ACTIVATE;
2996 }
2997 }
2998 out:
2999 #endif
3000
3001 return f2fs_write_single_data_page(page, NULL, NULL, NULL,
3002 wbc, FS_DATA_IO, 0, true);
3003 }
3004
3005 /*
3006 * This function was copied from write_cache_pages from mm/page-writeback.c.
3007 * The major change is making write step of cold data page separately from
3008 * warm/hot data page.
3009 */
f2fs_write_cache_pages(struct address_space * mapping,struct writeback_control * wbc,enum iostat_type io_type)3010 static int f2fs_write_cache_pages(struct address_space *mapping,
3011 struct writeback_control *wbc,
3012 enum iostat_type io_type)
3013 {
3014 int ret = 0;
3015 int done = 0, retry = 0;
3016 struct page *pages_local[F2FS_ONSTACK_PAGES];
3017 struct page **pages = pages_local;
3018 struct folio_batch fbatch;
3019 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
3020 struct bio *bio = NULL;
3021 sector_t last_block;
3022 #ifdef CONFIG_F2FS_FS_COMPRESSION
3023 struct inode *inode = mapping->host;
3024 struct compress_ctx cc = {
3025 .inode = inode,
3026 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
3027 .cluster_size = F2FS_I(inode)->i_cluster_size,
3028 .cluster_idx = NULL_CLUSTER,
3029 .rpages = NULL,
3030 .nr_rpages = 0,
3031 .cpages = NULL,
3032 .valid_nr_cpages = 0,
3033 .rbuf = NULL,
3034 .cbuf = NULL,
3035 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
3036 .private = NULL,
3037 };
3038 #endif
3039 int nr_folios, p, idx;
3040 int nr_pages;
3041 unsigned int max_pages = F2FS_ONSTACK_PAGES;
3042 pgoff_t index;
3043 pgoff_t end; /* Inclusive */
3044 pgoff_t done_index;
3045 int range_whole = 0;
3046 xa_mark_t tag;
3047 int nwritten = 0;
3048 int submitted = 0;
3049 int i;
3050
3051 #ifdef CONFIG_F2FS_FS_COMPRESSION
3052 if (f2fs_compressed_file(inode) &&
3053 1 << cc.log_cluster_size > F2FS_ONSTACK_PAGES) {
3054 pages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
3055 cc.log_cluster_size, GFP_NOFS | __GFP_NOFAIL);
3056 max_pages = 1 << cc.log_cluster_size;
3057 }
3058 #endif
3059
3060 folio_batch_init(&fbatch);
3061
3062 if (get_dirty_pages(mapping->host) <=
3063 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
3064 set_inode_flag(mapping->host, FI_HOT_DATA);
3065 else
3066 clear_inode_flag(mapping->host, FI_HOT_DATA);
3067
3068 if (wbc->range_cyclic) {
3069 index = mapping->writeback_index; /* prev offset */
3070 end = -1;
3071 } else {
3072 index = wbc->range_start >> PAGE_SHIFT;
3073 end = wbc->range_end >> PAGE_SHIFT;
3074 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3075 range_whole = 1;
3076 }
3077 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3078 tag = PAGECACHE_TAG_TOWRITE;
3079 else
3080 tag = PAGECACHE_TAG_DIRTY;
3081 retry:
3082 retry = 0;
3083 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3084 tag_pages_for_writeback(mapping, index, end);
3085 done_index = index;
3086 while (!done && !retry && (index <= end)) {
3087 nr_pages = 0;
3088 again:
3089 nr_folios = filemap_get_folios_tag(mapping, &index, end,
3090 tag, &fbatch);
3091 if (nr_folios == 0) {
3092 if (nr_pages)
3093 goto write;
3094 break;
3095 }
3096
3097 for (i = 0; i < nr_folios; i++) {
3098 struct folio *folio = fbatch.folios[i];
3099
3100 idx = 0;
3101 p = folio_nr_pages(folio);
3102 add_more:
3103 pages[nr_pages] = folio_page(folio, idx);
3104 folio_get(folio);
3105 if (++nr_pages == max_pages) {
3106 index = folio->index + idx + 1;
3107 folio_batch_release(&fbatch);
3108 goto write;
3109 }
3110 if (++idx < p)
3111 goto add_more;
3112 }
3113 folio_batch_release(&fbatch);
3114 goto again;
3115 write:
3116 for (i = 0; i < nr_pages; i++) {
3117 struct page *page = pages[i];
3118 struct folio *folio = page_folio(page);
3119 bool need_readd;
3120 readd:
3121 need_readd = false;
3122 #ifdef CONFIG_F2FS_FS_COMPRESSION
3123 if (f2fs_compressed_file(inode)) {
3124 void *fsdata = NULL;
3125 struct page *pagep;
3126 int ret2;
3127
3128 ret = f2fs_init_compress_ctx(&cc);
3129 if (ret) {
3130 done = 1;
3131 break;
3132 }
3133
3134 if (!f2fs_cluster_can_merge_page(&cc,
3135 folio->index)) {
3136 ret = f2fs_write_multi_pages(&cc,
3137 &submitted, wbc, io_type);
3138 if (!ret)
3139 need_readd = true;
3140 goto result;
3141 }
3142
3143 if (unlikely(f2fs_cp_error(sbi)))
3144 goto lock_folio;
3145
3146 if (!f2fs_cluster_is_empty(&cc))
3147 goto lock_folio;
3148
3149 if (f2fs_all_cluster_page_ready(&cc,
3150 pages, i, nr_pages, true))
3151 goto lock_folio;
3152
3153 ret2 = f2fs_prepare_compress_overwrite(
3154 inode, &pagep,
3155 folio->index, &fsdata);
3156 if (ret2 < 0) {
3157 ret = ret2;
3158 done = 1;
3159 break;
3160 } else if (ret2 &&
3161 (!f2fs_compress_write_end(inode,
3162 fsdata, folio->index, 1) ||
3163 !f2fs_all_cluster_page_ready(&cc,
3164 pages, i, nr_pages,
3165 false))) {
3166 retry = 1;
3167 break;
3168 }
3169 }
3170 #endif
3171 /* give a priority to WB_SYNC threads */
3172 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3173 wbc->sync_mode == WB_SYNC_NONE) {
3174 done = 1;
3175 break;
3176 }
3177 #ifdef CONFIG_F2FS_FS_COMPRESSION
3178 lock_folio:
3179 #endif
3180 done_index = folio->index;
3181 retry_write:
3182 folio_lock(folio);
3183
3184 if (unlikely(folio->mapping != mapping)) {
3185 continue_unlock:
3186 folio_unlock(folio);
3187 continue;
3188 }
3189
3190 if (!folio_test_dirty(folio)) {
3191 /* someone wrote it for us */
3192 goto continue_unlock;
3193 }
3194
3195 if (folio_test_writeback(folio)) {
3196 if (wbc->sync_mode == WB_SYNC_NONE)
3197 goto continue_unlock;
3198 f2fs_wait_on_page_writeback(&folio->page, DATA, true, true);
3199 }
3200
3201 if (!folio_clear_dirty_for_io(folio))
3202 goto continue_unlock;
3203
3204 #ifdef CONFIG_F2FS_FS_COMPRESSION
3205 if (f2fs_compressed_file(inode)) {
3206 folio_get(folio);
3207 f2fs_compress_ctx_add_page(&cc, &folio->page);
3208 continue;
3209 }
3210 #endif
3211 ret = f2fs_write_single_data_page(&folio->page,
3212 &submitted, &bio, &last_block,
3213 wbc, io_type, 0, true);
3214 if (ret == AOP_WRITEPAGE_ACTIVATE)
3215 folio_unlock(folio);
3216 #ifdef CONFIG_F2FS_FS_COMPRESSION
3217 result:
3218 #endif
3219 nwritten += submitted;
3220 wbc->nr_to_write -= submitted;
3221
3222 if (unlikely(ret)) {
3223 /*
3224 * keep nr_to_write, since vfs uses this to
3225 * get # of written pages.
3226 */
3227 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3228 ret = 0;
3229 goto next;
3230 } else if (ret == -EAGAIN) {
3231 ret = 0;
3232 if (wbc->sync_mode == WB_SYNC_ALL) {
3233 f2fs_io_schedule_timeout(
3234 DEFAULT_IO_TIMEOUT);
3235 goto retry_write;
3236 }
3237 goto next;
3238 }
3239 done_index = folio_next_index(folio);
3240 done = 1;
3241 break;
3242 }
3243
3244 if (wbc->nr_to_write <= 0 &&
3245 wbc->sync_mode == WB_SYNC_NONE) {
3246 done = 1;
3247 break;
3248 }
3249 next:
3250 if (need_readd)
3251 goto readd;
3252 }
3253 release_pages(pages, nr_pages);
3254 cond_resched();
3255 }
3256 #ifdef CONFIG_F2FS_FS_COMPRESSION
3257 /* flush remained pages in compress cluster */
3258 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3259 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3260 nwritten += submitted;
3261 wbc->nr_to_write -= submitted;
3262 if (ret) {
3263 done = 1;
3264 retry = 0;
3265 }
3266 }
3267 if (f2fs_compressed_file(inode))
3268 f2fs_destroy_compress_ctx(&cc, false);
3269 #endif
3270 if (retry) {
3271 index = 0;
3272 end = -1;
3273 goto retry;
3274 }
3275 if (wbc->range_cyclic && !done)
3276 done_index = 0;
3277 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3278 mapping->writeback_index = done_index;
3279
3280 if (nwritten)
3281 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3282 NULL, 0, DATA);
3283 /* submit cached bio of IPU write */
3284 if (bio)
3285 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3286
3287 #ifdef CONFIG_F2FS_FS_COMPRESSION
3288 if (pages != pages_local)
3289 kfree(pages);
3290 #endif
3291
3292 return ret;
3293 }
3294
__should_serialize_io(struct inode * inode,struct writeback_control * wbc)3295 static inline bool __should_serialize_io(struct inode *inode,
3296 struct writeback_control *wbc)
3297 {
3298 /* to avoid deadlock in path of data flush */
3299 if (F2FS_I(inode)->wb_task)
3300 return false;
3301
3302 if (!S_ISREG(inode->i_mode))
3303 return false;
3304 if (IS_NOQUOTA(inode))
3305 return false;
3306
3307 if (f2fs_need_compress_data(inode))
3308 return true;
3309 if (wbc->sync_mode != WB_SYNC_ALL)
3310 return true;
3311 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3312 return true;
3313 return false;
3314 }
3315
__f2fs_write_data_pages(struct address_space * mapping,struct writeback_control * wbc,enum iostat_type io_type)3316 static int __f2fs_write_data_pages(struct address_space *mapping,
3317 struct writeback_control *wbc,
3318 enum iostat_type io_type)
3319 {
3320 struct inode *inode = mapping->host;
3321 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3322 struct blk_plug plug;
3323 int ret;
3324 bool locked = false;
3325
3326 /* deal with chardevs and other special file */
3327 if (!mapping->a_ops->writepage)
3328 return 0;
3329
3330 /* skip writing if there is no dirty page in this inode */
3331 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3332 return 0;
3333
3334 /* during POR, we don't need to trigger writepage at all. */
3335 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3336 goto skip_write;
3337
3338 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3339 wbc->sync_mode == WB_SYNC_NONE &&
3340 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3341 f2fs_available_free_memory(sbi, DIRTY_DENTS))
3342 goto skip_write;
3343
3344 /* skip writing in file defragment preparing stage */
3345 if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3346 goto skip_write;
3347
3348 trace_f2fs_writepages(mapping->host, wbc, DATA);
3349
3350 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3351 if (wbc->sync_mode == WB_SYNC_ALL)
3352 atomic_inc(&sbi->wb_sync_req[DATA]);
3353 else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3354 /* to avoid potential deadlock */
3355 if (current->plug)
3356 blk_finish_plug(current->plug);
3357 goto skip_write;
3358 }
3359
3360 if (__should_serialize_io(inode, wbc)) {
3361 mutex_lock(&sbi->writepages);
3362 locked = true;
3363 }
3364
3365 blk_start_plug(&plug);
3366 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3367 blk_finish_plug(&plug);
3368
3369 if (locked)
3370 mutex_unlock(&sbi->writepages);
3371
3372 if (wbc->sync_mode == WB_SYNC_ALL)
3373 atomic_dec(&sbi->wb_sync_req[DATA]);
3374 /*
3375 * if some pages were truncated, we cannot guarantee its mapping->host
3376 * to detect pending bios.
3377 */
3378
3379 f2fs_remove_dirty_inode(inode);
3380 return ret;
3381
3382 skip_write:
3383 wbc->pages_skipped += get_dirty_pages(inode);
3384 trace_f2fs_writepages(mapping->host, wbc, DATA);
3385 return 0;
3386 }
3387
f2fs_write_data_pages(struct address_space * mapping,struct writeback_control * wbc)3388 static int f2fs_write_data_pages(struct address_space *mapping,
3389 struct writeback_control *wbc)
3390 {
3391 struct inode *inode = mapping->host;
3392
3393 return __f2fs_write_data_pages(mapping, wbc,
3394 F2FS_I(inode)->cp_task == current ?
3395 FS_CP_DATA_IO : FS_DATA_IO);
3396 }
3397
f2fs_write_failed(struct inode * inode,loff_t to)3398 void f2fs_write_failed(struct inode *inode, loff_t to)
3399 {
3400 loff_t i_size = i_size_read(inode);
3401
3402 if (IS_NOQUOTA(inode))
3403 return;
3404
3405 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3406 if (to > i_size && !f2fs_verity_in_progress(inode)) {
3407 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3408 filemap_invalidate_lock(inode->i_mapping);
3409
3410 truncate_pagecache(inode, i_size);
3411 f2fs_truncate_blocks(inode, i_size, true);
3412
3413 filemap_invalidate_unlock(inode->i_mapping);
3414 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3415 }
3416 }
3417
prepare_write_begin(struct f2fs_sb_info * sbi,struct page * page,loff_t pos,unsigned len,block_t * blk_addr,bool * node_changed)3418 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3419 struct page *page, loff_t pos, unsigned len,
3420 block_t *blk_addr, bool *node_changed)
3421 {
3422 struct inode *inode = page->mapping->host;
3423 pgoff_t index = page->index;
3424 struct dnode_of_data dn;
3425 struct page *ipage;
3426 bool locked = false;
3427 int flag = F2FS_GET_BLOCK_PRE_AIO;
3428 int err = 0;
3429
3430 /*
3431 * If a whole page is being written and we already preallocated all the
3432 * blocks, then there is no need to get a block address now.
3433 */
3434 if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3435 return 0;
3436
3437 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3438 if (f2fs_has_inline_data(inode)) {
3439 if (pos + len > MAX_INLINE_DATA(inode))
3440 flag = F2FS_GET_BLOCK_DEFAULT;
3441 f2fs_map_lock(sbi, flag);
3442 locked = true;
3443 } else if ((pos & PAGE_MASK) >= i_size_read(inode)) {
3444 f2fs_map_lock(sbi, flag);
3445 locked = true;
3446 }
3447
3448 restart:
3449 /* check inline_data */
3450 ipage = f2fs_get_node_page(sbi, inode->i_ino);
3451 if (IS_ERR(ipage)) {
3452 err = PTR_ERR(ipage);
3453 goto unlock_out;
3454 }
3455
3456 set_new_dnode(&dn, inode, ipage, ipage, 0);
3457
3458 if (f2fs_has_inline_data(inode)) {
3459 if (pos + len <= MAX_INLINE_DATA(inode)) {
3460 f2fs_do_read_inline_data(page, ipage);
3461 set_inode_flag(inode, FI_DATA_EXIST);
3462 if (inode->i_nlink)
3463 set_page_private_inline(ipage);
3464 goto out;
3465 }
3466 err = f2fs_convert_inline_page(&dn, page);
3467 if (err || dn.data_blkaddr != NULL_ADDR)
3468 goto out;
3469 }
3470
3471 if (!f2fs_lookup_read_extent_cache_block(inode, index,
3472 &dn.data_blkaddr)) {
3473 if (locked) {
3474 err = f2fs_reserve_block(&dn, index);
3475 goto out;
3476 }
3477
3478 /* hole case */
3479 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3480 if (!err && dn.data_blkaddr != NULL_ADDR)
3481 goto out;
3482 f2fs_put_dnode(&dn);
3483 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3484 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3485 locked = true;
3486 goto restart;
3487 }
3488 out:
3489 if (!err) {
3490 /* convert_inline_page can make node_changed */
3491 *blk_addr = dn.data_blkaddr;
3492 *node_changed = dn.node_changed;
3493 }
3494 f2fs_put_dnode(&dn);
3495 unlock_out:
3496 if (locked)
3497 f2fs_map_unlock(sbi, flag);
3498 return err;
3499 }
3500
__find_data_block(struct inode * inode,pgoff_t index,block_t * blk_addr)3501 static int __find_data_block(struct inode *inode, pgoff_t index,
3502 block_t *blk_addr)
3503 {
3504 struct dnode_of_data dn;
3505 struct page *ipage;
3506 int err = 0;
3507
3508 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
3509 if (IS_ERR(ipage))
3510 return PTR_ERR(ipage);
3511
3512 set_new_dnode(&dn, inode, ipage, ipage, 0);
3513
3514 if (!f2fs_lookup_read_extent_cache_block(inode, index,
3515 &dn.data_blkaddr)) {
3516 /* hole case */
3517 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3518 if (err) {
3519 dn.data_blkaddr = NULL_ADDR;
3520 err = 0;
3521 }
3522 }
3523 *blk_addr = dn.data_blkaddr;
3524 f2fs_put_dnode(&dn);
3525 return err;
3526 }
3527
__reserve_data_block(struct inode * inode,pgoff_t index,block_t * blk_addr,bool * node_changed)3528 static int __reserve_data_block(struct inode *inode, pgoff_t index,
3529 block_t *blk_addr, bool *node_changed)
3530 {
3531 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3532 struct dnode_of_data dn;
3533 struct page *ipage;
3534 int err = 0;
3535
3536 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3537
3538 ipage = f2fs_get_node_page(sbi, inode->i_ino);
3539 if (IS_ERR(ipage)) {
3540 err = PTR_ERR(ipage);
3541 goto unlock_out;
3542 }
3543 set_new_dnode(&dn, inode, ipage, ipage, 0);
3544
3545 if (!f2fs_lookup_read_extent_cache_block(dn.inode, index,
3546 &dn.data_blkaddr))
3547 err = f2fs_reserve_block(&dn, index);
3548
3549 *blk_addr = dn.data_blkaddr;
3550 *node_changed = dn.node_changed;
3551 f2fs_put_dnode(&dn);
3552
3553 unlock_out:
3554 f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3555 return err;
3556 }
3557
prepare_atomic_write_begin(struct f2fs_sb_info * sbi,struct page * page,loff_t pos,unsigned int len,block_t * blk_addr,bool * node_changed,bool * use_cow)3558 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
3559 struct page *page, loff_t pos, unsigned int len,
3560 block_t *blk_addr, bool *node_changed, bool *use_cow)
3561 {
3562 struct inode *inode = page->mapping->host;
3563 struct inode *cow_inode = F2FS_I(inode)->cow_inode;
3564 pgoff_t index = page->index;
3565 int err = 0;
3566 block_t ori_blk_addr = NULL_ADDR;
3567
3568 /* If pos is beyond the end of file, reserve a new block in COW inode */
3569 if ((pos & PAGE_MASK) >= i_size_read(inode))
3570 goto reserve_block;
3571
3572 /* Look for the block in COW inode first */
3573 err = __find_data_block(cow_inode, index, blk_addr);
3574 if (err) {
3575 return err;
3576 } else if (*blk_addr != NULL_ADDR) {
3577 *use_cow = true;
3578 return 0;
3579 }
3580
3581 if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE))
3582 goto reserve_block;
3583
3584 /* Look for the block in the original inode */
3585 err = __find_data_block(inode, index, &ori_blk_addr);
3586 if (err)
3587 return err;
3588
3589 reserve_block:
3590 /* Finally, we should reserve a new block in COW inode for the update */
3591 err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
3592 if (err)
3593 return err;
3594 inc_atomic_write_cnt(inode);
3595
3596 if (ori_blk_addr != NULL_ADDR)
3597 *blk_addr = ori_blk_addr;
3598 return 0;
3599 }
3600
f2fs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct page ** pagep,void ** fsdata)3601 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3602 loff_t pos, unsigned len, struct page **pagep, void **fsdata)
3603 {
3604 struct inode *inode = mapping->host;
3605 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3606 struct page *page = NULL;
3607 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3608 bool need_balance = false;
3609 bool use_cow = false;
3610 block_t blkaddr = NULL_ADDR;
3611 int err = 0;
3612
3613 trace_f2fs_write_begin(inode, pos, len);
3614
3615 if (!f2fs_is_checkpoint_ready(sbi)) {
3616 err = -ENOSPC;
3617 goto fail;
3618 }
3619
3620 /*
3621 * We should check this at this moment to avoid deadlock on inode page
3622 * and #0 page. The locking rule for inline_data conversion should be:
3623 * lock_page(page #0) -> lock_page(inode_page)
3624 */
3625 if (index != 0) {
3626 err = f2fs_convert_inline_inode(inode);
3627 if (err)
3628 goto fail;
3629 }
3630
3631 #ifdef CONFIG_F2FS_FS_COMPRESSION
3632 if (f2fs_compressed_file(inode)) {
3633 int ret;
3634
3635 *fsdata = NULL;
3636
3637 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3638 goto repeat;
3639
3640 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3641 index, fsdata);
3642 if (ret < 0) {
3643 err = ret;
3644 goto fail;
3645 } else if (ret) {
3646 return 0;
3647 }
3648 }
3649 #endif
3650
3651 repeat:
3652 /*
3653 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3654 * wait_for_stable_page. Will wait that below with our IO control.
3655 */
3656 page = f2fs_pagecache_get_page(mapping, index,
3657 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3658 if (!page) {
3659 err = -ENOMEM;
3660 goto fail;
3661 }
3662
3663 /* TODO: cluster can be compressed due to race with .writepage */
3664
3665 *pagep = page;
3666
3667 if (f2fs_is_atomic_file(inode))
3668 err = prepare_atomic_write_begin(sbi, page, pos, len,
3669 &blkaddr, &need_balance, &use_cow);
3670 else
3671 err = prepare_write_begin(sbi, page, pos, len,
3672 &blkaddr, &need_balance);
3673 if (err)
3674 goto fail;
3675
3676 if (need_balance && !IS_NOQUOTA(inode) &&
3677 has_not_enough_free_secs(sbi, 0, 0)) {
3678 unlock_page(page);
3679 f2fs_balance_fs(sbi, true);
3680 lock_page(page);
3681 if (page->mapping != mapping) {
3682 /* The page got truncated from under us */
3683 f2fs_put_page(page, 1);
3684 goto repeat;
3685 }
3686 }
3687
3688 f2fs_wait_on_page_writeback(page, DATA, false, true);
3689
3690 if (len == PAGE_SIZE || PageUptodate(page))
3691 return 0;
3692
3693 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3694 !f2fs_verity_in_progress(inode)) {
3695 zero_user_segment(page, len, PAGE_SIZE);
3696 return 0;
3697 }
3698
3699 if (blkaddr == NEW_ADDR) {
3700 zero_user_segment(page, 0, PAGE_SIZE);
3701 SetPageUptodate(page);
3702 } else {
3703 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3704 DATA_GENERIC_ENHANCE_READ)) {
3705 err = -EFSCORRUPTED;
3706 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
3707 goto fail;
3708 }
3709 err = f2fs_submit_page_read(use_cow ?
3710 F2FS_I(inode)->cow_inode : inode, page,
3711 blkaddr, 0, true);
3712 if (err)
3713 goto fail;
3714
3715 lock_page(page);
3716 if (unlikely(page->mapping != mapping)) {
3717 f2fs_put_page(page, 1);
3718 goto repeat;
3719 }
3720 if (unlikely(!PageUptodate(page))) {
3721 err = -EIO;
3722 goto fail;
3723 }
3724 }
3725 return 0;
3726
3727 fail:
3728 f2fs_put_page(page, 1);
3729 f2fs_write_failed(inode, pos + len);
3730 return err;
3731 }
3732
f2fs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)3733 static int f2fs_write_end(struct file *file,
3734 struct address_space *mapping,
3735 loff_t pos, unsigned len, unsigned copied,
3736 struct page *page, void *fsdata)
3737 {
3738 struct inode *inode = page->mapping->host;
3739
3740 trace_f2fs_write_end(inode, pos, len, copied);
3741
3742 /*
3743 * This should be come from len == PAGE_SIZE, and we expect copied
3744 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3745 * let generic_perform_write() try to copy data again through copied=0.
3746 */
3747 if (!PageUptodate(page)) {
3748 if (unlikely(copied != len))
3749 copied = 0;
3750 else
3751 SetPageUptodate(page);
3752 }
3753
3754 #ifdef CONFIG_F2FS_FS_COMPRESSION
3755 /* overwrite compressed file */
3756 if (f2fs_compressed_file(inode) && fsdata) {
3757 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3758 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3759
3760 if (pos + copied > i_size_read(inode) &&
3761 !f2fs_verity_in_progress(inode))
3762 f2fs_i_size_write(inode, pos + copied);
3763 return copied;
3764 }
3765 #endif
3766
3767 if (!copied)
3768 goto unlock_out;
3769
3770 set_page_dirty(page);
3771
3772 if (pos + copied > i_size_read(inode) &&
3773 !f2fs_verity_in_progress(inode)) {
3774 f2fs_i_size_write(inode, pos + copied);
3775 if (f2fs_is_atomic_file(inode))
3776 f2fs_i_size_write(F2FS_I(inode)->cow_inode,
3777 pos + copied);
3778 }
3779 unlock_out:
3780 f2fs_put_page(page, 1);
3781 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3782 return copied;
3783 }
3784
f2fs_invalidate_folio(struct folio * folio,size_t offset,size_t length)3785 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3786 {
3787 struct inode *inode = folio->mapping->host;
3788 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3789
3790 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3791 (offset || length != folio_size(folio)))
3792 return;
3793
3794 if (folio_test_dirty(folio)) {
3795 if (inode->i_ino == F2FS_META_INO(sbi)) {
3796 dec_page_count(sbi, F2FS_DIRTY_META);
3797 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3798 dec_page_count(sbi, F2FS_DIRTY_NODES);
3799 } else {
3800 inode_dec_dirty_pages(inode);
3801 f2fs_remove_dirty_inode(inode);
3802 }
3803 }
3804 clear_page_private_all(&folio->page);
3805 }
3806
f2fs_release_folio(struct folio * folio,gfp_t wait)3807 bool f2fs_release_folio(struct folio *folio, gfp_t wait)
3808 {
3809 /* If this is dirty folio, keep private data */
3810 if (folio_test_dirty(folio))
3811 return false;
3812
3813 clear_page_private_all(&folio->page);
3814 return true;
3815 }
3816
f2fs_dirty_data_folio(struct address_space * mapping,struct folio * folio)3817 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3818 struct folio *folio)
3819 {
3820 struct inode *inode = mapping->host;
3821
3822 trace_f2fs_set_page_dirty(&folio->page, DATA);
3823
3824 if (!folio_test_uptodate(folio))
3825 folio_mark_uptodate(folio);
3826 BUG_ON(folio_test_swapcache(folio));
3827
3828 if (filemap_dirty_folio(mapping, folio)) {
3829 f2fs_update_dirty_folio(inode, folio);
3830 return true;
3831 }
3832 return false;
3833 }
3834
3835
f2fs_bmap_compress(struct inode * inode,sector_t block)3836 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3837 {
3838 #ifdef CONFIG_F2FS_FS_COMPRESSION
3839 struct dnode_of_data dn;
3840 sector_t start_idx, blknr = 0;
3841 int ret;
3842
3843 start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3844
3845 set_new_dnode(&dn, inode, NULL, NULL, 0);
3846 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3847 if (ret)
3848 return 0;
3849
3850 if (dn.data_blkaddr != COMPRESS_ADDR) {
3851 dn.ofs_in_node += block - start_idx;
3852 blknr = f2fs_data_blkaddr(&dn);
3853 if (!__is_valid_data_blkaddr(blknr))
3854 blknr = 0;
3855 }
3856
3857 f2fs_put_dnode(&dn);
3858 return blknr;
3859 #else
3860 return 0;
3861 #endif
3862 }
3863
3864
f2fs_bmap(struct address_space * mapping,sector_t block)3865 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3866 {
3867 struct inode *inode = mapping->host;
3868 sector_t blknr = 0;
3869
3870 if (f2fs_has_inline_data(inode))
3871 goto out;
3872
3873 /* make sure allocating whole blocks */
3874 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3875 filemap_write_and_wait(mapping);
3876
3877 /* Block number less than F2FS MAX BLOCKS */
3878 if (unlikely(block >= max_file_blocks(inode)))
3879 goto out;
3880
3881 if (f2fs_compressed_file(inode)) {
3882 blknr = f2fs_bmap_compress(inode, block);
3883 } else {
3884 struct f2fs_map_blocks map;
3885
3886 memset(&map, 0, sizeof(map));
3887 map.m_lblk = block;
3888 map.m_len = 1;
3889 map.m_next_pgofs = NULL;
3890 map.m_seg_type = NO_CHECK_TYPE;
3891
3892 if (!f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_BMAP))
3893 blknr = map.m_pblk;
3894 }
3895 out:
3896 trace_f2fs_bmap(inode, block, blknr);
3897 return blknr;
3898 }
3899
3900 #ifdef CONFIG_SWAP
f2fs_migrate_blocks(struct inode * inode,block_t start_blk,unsigned int blkcnt)3901 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3902 unsigned int blkcnt)
3903 {
3904 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3905 unsigned int blkofs;
3906 unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3907 unsigned int secidx = start_blk / blk_per_sec;
3908 unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3909 int ret = 0;
3910
3911 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3912 filemap_invalidate_lock(inode->i_mapping);
3913
3914 set_inode_flag(inode, FI_ALIGNED_WRITE);
3915 set_inode_flag(inode, FI_OPU_WRITE);
3916
3917 for (; secidx < end_sec; secidx++) {
3918 f2fs_down_write(&sbi->pin_sem);
3919
3920 f2fs_lock_op(sbi);
3921 f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3922 f2fs_unlock_op(sbi);
3923
3924 set_inode_flag(inode, FI_SKIP_WRITES);
3925
3926 for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3927 struct page *page;
3928 unsigned int blkidx = secidx * blk_per_sec + blkofs;
3929
3930 page = f2fs_get_lock_data_page(inode, blkidx, true);
3931 if (IS_ERR(page)) {
3932 f2fs_up_write(&sbi->pin_sem);
3933 ret = PTR_ERR(page);
3934 goto done;
3935 }
3936
3937 set_page_dirty(page);
3938 f2fs_put_page(page, 1);
3939 }
3940
3941 clear_inode_flag(inode, FI_SKIP_WRITES);
3942
3943 ret = filemap_fdatawrite(inode->i_mapping);
3944
3945 f2fs_up_write(&sbi->pin_sem);
3946
3947 if (ret)
3948 break;
3949 }
3950
3951 done:
3952 clear_inode_flag(inode, FI_SKIP_WRITES);
3953 clear_inode_flag(inode, FI_OPU_WRITE);
3954 clear_inode_flag(inode, FI_ALIGNED_WRITE);
3955
3956 filemap_invalidate_unlock(inode->i_mapping);
3957 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3958
3959 return ret;
3960 }
3961
check_swap_activate(struct swap_info_struct * sis,struct file * swap_file,sector_t * span)3962 static int check_swap_activate(struct swap_info_struct *sis,
3963 struct file *swap_file, sector_t *span)
3964 {
3965 struct address_space *mapping = swap_file->f_mapping;
3966 struct inode *inode = mapping->host;
3967 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3968 sector_t cur_lblock;
3969 sector_t last_lblock;
3970 sector_t pblock;
3971 sector_t lowest_pblock = -1;
3972 sector_t highest_pblock = 0;
3973 int nr_extents = 0;
3974 unsigned long nr_pblocks;
3975 unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3976 unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3977 unsigned int not_aligned = 0;
3978 int ret = 0;
3979
3980 /*
3981 * Map all the blocks into the extent list. This code doesn't try
3982 * to be very smart.
3983 */
3984 cur_lblock = 0;
3985 last_lblock = bytes_to_blks(inode, i_size_read(inode));
3986
3987 while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3988 struct f2fs_map_blocks map;
3989 retry:
3990 cond_resched();
3991
3992 memset(&map, 0, sizeof(map));
3993 map.m_lblk = cur_lblock;
3994 map.m_len = last_lblock - cur_lblock;
3995 map.m_next_pgofs = NULL;
3996 map.m_next_extent = NULL;
3997 map.m_seg_type = NO_CHECK_TYPE;
3998 map.m_may_create = false;
3999
4000 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
4001 if (ret)
4002 goto out;
4003
4004 /* hole */
4005 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
4006 f2fs_err(sbi, "Swapfile has holes");
4007 ret = -EINVAL;
4008 goto out;
4009 }
4010
4011 pblock = map.m_pblk;
4012 nr_pblocks = map.m_len;
4013
4014 if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
4015 nr_pblocks & sec_blks_mask) {
4016 not_aligned++;
4017
4018 nr_pblocks = roundup(nr_pblocks, blks_per_sec);
4019 if (cur_lblock + nr_pblocks > sis->max)
4020 nr_pblocks -= blks_per_sec;
4021
4022 if (!nr_pblocks) {
4023 /* this extent is last one */
4024 nr_pblocks = map.m_len;
4025 f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
4026 goto next;
4027 }
4028
4029 ret = f2fs_migrate_blocks(inode, cur_lblock,
4030 nr_pblocks);
4031 if (ret)
4032 goto out;
4033 goto retry;
4034 }
4035 next:
4036 if (cur_lblock + nr_pblocks >= sis->max)
4037 nr_pblocks = sis->max - cur_lblock;
4038
4039 if (cur_lblock) { /* exclude the header page */
4040 if (pblock < lowest_pblock)
4041 lowest_pblock = pblock;
4042 if (pblock + nr_pblocks - 1 > highest_pblock)
4043 highest_pblock = pblock + nr_pblocks - 1;
4044 }
4045
4046 /*
4047 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4048 */
4049 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
4050 if (ret < 0)
4051 goto out;
4052 nr_extents += ret;
4053 cur_lblock += nr_pblocks;
4054 }
4055 ret = nr_extents;
4056 *span = 1 + highest_pblock - lowest_pblock;
4057 if (cur_lblock == 0)
4058 cur_lblock = 1; /* force Empty message */
4059 sis->max = cur_lblock;
4060 sis->pages = cur_lblock - 1;
4061 sis->highest_bit = cur_lblock - 1;
4062 out:
4063 if (not_aligned)
4064 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
4065 not_aligned, blks_per_sec * F2FS_BLKSIZE);
4066 return ret;
4067 }
4068
f2fs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)4069 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4070 sector_t *span)
4071 {
4072 struct inode *inode = file_inode(file);
4073 int ret;
4074
4075 if (!S_ISREG(inode->i_mode))
4076 return -EINVAL;
4077
4078 if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4079 return -EROFS;
4080
4081 if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
4082 f2fs_err(F2FS_I_SB(inode),
4083 "Swapfile not supported in LFS mode");
4084 return -EINVAL;
4085 }
4086
4087 ret = f2fs_convert_inline_inode(inode);
4088 if (ret)
4089 return ret;
4090
4091 if (!f2fs_disable_compressed_file(inode))
4092 return -EINVAL;
4093
4094 f2fs_precache_extents(inode);
4095
4096 ret = check_swap_activate(sis, file, span);
4097 if (ret < 0)
4098 return ret;
4099
4100 stat_inc_swapfile_inode(inode);
4101 set_inode_flag(inode, FI_PIN_FILE);
4102 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4103 return ret;
4104 }
4105
f2fs_swap_deactivate(struct file * file)4106 static void f2fs_swap_deactivate(struct file *file)
4107 {
4108 struct inode *inode = file_inode(file);
4109
4110 stat_dec_swapfile_inode(inode);
4111 clear_inode_flag(inode, FI_PIN_FILE);
4112 }
4113 #else
f2fs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)4114 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4115 sector_t *span)
4116 {
4117 return -EOPNOTSUPP;
4118 }
4119
f2fs_swap_deactivate(struct file * file)4120 static void f2fs_swap_deactivate(struct file *file)
4121 {
4122 }
4123 #endif
4124
4125 const struct address_space_operations f2fs_dblock_aops = {
4126 .read_folio = f2fs_read_data_folio,
4127 .readahead = f2fs_readahead,
4128 .writepage = f2fs_write_data_page,
4129 .writepages = f2fs_write_data_pages,
4130 .write_begin = f2fs_write_begin,
4131 .write_end = f2fs_write_end,
4132 .dirty_folio = f2fs_dirty_data_folio,
4133 .migrate_folio = filemap_migrate_folio,
4134 .invalidate_folio = f2fs_invalidate_folio,
4135 .release_folio = f2fs_release_folio,
4136 .bmap = f2fs_bmap,
4137 .swap_activate = f2fs_swap_activate,
4138 .swap_deactivate = f2fs_swap_deactivate,
4139 };
4140
f2fs_clear_page_cache_dirty_tag(struct page * page)4141 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4142 {
4143 struct address_space *mapping = page_mapping(page);
4144 unsigned long flags;
4145
4146 xa_lock_irqsave(&mapping->i_pages, flags);
4147 __xa_clear_mark(&mapping->i_pages, page_index(page),
4148 PAGECACHE_TAG_DIRTY);
4149 xa_unlock_irqrestore(&mapping->i_pages, flags);
4150 }
4151
f2fs_init_post_read_processing(void)4152 int __init f2fs_init_post_read_processing(void)
4153 {
4154 bio_post_read_ctx_cache =
4155 kmem_cache_create("f2fs_bio_post_read_ctx",
4156 sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4157 if (!bio_post_read_ctx_cache)
4158 goto fail;
4159 bio_post_read_ctx_pool =
4160 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4161 bio_post_read_ctx_cache);
4162 if (!bio_post_read_ctx_pool)
4163 goto fail_free_cache;
4164 return 0;
4165
4166 fail_free_cache:
4167 kmem_cache_destroy(bio_post_read_ctx_cache);
4168 fail:
4169 return -ENOMEM;
4170 }
4171
f2fs_destroy_post_read_processing(void)4172 void f2fs_destroy_post_read_processing(void)
4173 {
4174 mempool_destroy(bio_post_read_ctx_pool);
4175 kmem_cache_destroy(bio_post_read_ctx_cache);
4176 }
4177
f2fs_init_post_read_wq(struct f2fs_sb_info * sbi)4178 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4179 {
4180 if (!f2fs_sb_has_encrypt(sbi) &&
4181 !f2fs_sb_has_verity(sbi) &&
4182 !f2fs_sb_has_compression(sbi))
4183 return 0;
4184
4185 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4186 WQ_UNBOUND | WQ_HIGHPRI,
4187 num_online_cpus());
4188 return sbi->post_read_wq ? 0 : -ENOMEM;
4189 }
4190
f2fs_destroy_post_read_wq(struct f2fs_sb_info * sbi)4191 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4192 {
4193 if (sbi->post_read_wq)
4194 destroy_workqueue(sbi->post_read_wq);
4195 }
4196
f2fs_init_bio_entry_cache(void)4197 int __init f2fs_init_bio_entry_cache(void)
4198 {
4199 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4200 sizeof(struct bio_entry));
4201 return bio_entry_slab ? 0 : -ENOMEM;
4202 }
4203
f2fs_destroy_bio_entry_cache(void)4204 void f2fs_destroy_bio_entry_cache(void)
4205 {
4206 kmem_cache_destroy(bio_entry_slab);
4207 }
4208
f2fs_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)4209 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4210 unsigned int flags, struct iomap *iomap,
4211 struct iomap *srcmap)
4212 {
4213 struct f2fs_map_blocks map = {};
4214 pgoff_t next_pgofs = 0;
4215 int err;
4216
4217 map.m_lblk = bytes_to_blks(inode, offset);
4218 map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4219 map.m_next_pgofs = &next_pgofs;
4220 map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4221 if (flags & IOMAP_WRITE)
4222 map.m_may_create = true;
4223
4224 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DIO);
4225 if (err)
4226 return err;
4227
4228 iomap->offset = blks_to_bytes(inode, map.m_lblk);
4229
4230 /*
4231 * When inline encryption is enabled, sometimes I/O to an encrypted file
4232 * has to be broken up to guarantee DUN contiguity. Handle this by
4233 * limiting the length of the mapping returned.
4234 */
4235 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4236
4237 /*
4238 * We should never see delalloc or compressed extents here based on
4239 * prior flushing and checks.
4240 */
4241 if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR))
4242 return -EINVAL;
4243 if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR))
4244 return -EINVAL;
4245
4246 if (map.m_pblk != NULL_ADDR) {
4247 iomap->length = blks_to_bytes(inode, map.m_len);
4248 iomap->type = IOMAP_MAPPED;
4249 iomap->flags |= IOMAP_F_MERGED;
4250 iomap->bdev = map.m_bdev;
4251 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4252 } else {
4253 if (flags & IOMAP_WRITE)
4254 return -ENOTBLK;
4255 iomap->length = blks_to_bytes(inode, next_pgofs) -
4256 iomap->offset;
4257 iomap->type = IOMAP_HOLE;
4258 iomap->addr = IOMAP_NULL_ADDR;
4259 }
4260
4261 if (map.m_flags & F2FS_MAP_NEW)
4262 iomap->flags |= IOMAP_F_NEW;
4263 if ((inode->i_state & I_DIRTY_DATASYNC) ||
4264 offset + length > i_size_read(inode))
4265 iomap->flags |= IOMAP_F_DIRTY;
4266
4267 return 0;
4268 }
4269
4270 const struct iomap_ops f2fs_iomap_ops = {
4271 .iomap_begin = f2fs_iomap_begin,
4272 };
4273