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 fscrypt_finalize_bounce_page(&page);
342
343 #ifdef CONFIG_F2FS_FS_COMPRESSION
344 if (f2fs_is_compressed_page(page)) {
345 f2fs_compress_write_end_io(bio, page);
346 continue;
347 }
348 #endif
349
350 if (unlikely(bio->bi_status)) {
351 mapping_set_error(page->mapping, -EIO);
352 if (type == F2FS_WB_CP_DATA)
353 f2fs_stop_checkpoint(sbi, true,
354 STOP_CP_REASON_WRITE_FAIL);
355 }
356
357 f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
358 page->index != nid_of_node(page));
359
360 dec_page_count(sbi, type);
361 if (f2fs_in_warm_node_list(sbi, page))
362 f2fs_del_fsync_node_entry(sbi, page);
363 clear_page_private_gcing(page);
364 end_page_writeback(page);
365 }
366 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
367 wq_has_sleeper(&sbi->cp_wait))
368 wake_up(&sbi->cp_wait);
369
370 bio_put(bio);
371 }
372
373 #ifdef CONFIG_BLK_DEV_ZONED
f2fs_zone_write_end_io(struct bio * bio)374 static void f2fs_zone_write_end_io(struct bio *bio)
375 {
376 struct f2fs_bio_info *io = (struct f2fs_bio_info *)bio->bi_private;
377
378 bio->bi_private = io->bi_private;
379 complete(&io->zone_wait);
380 f2fs_write_end_io(bio);
381 }
382 #endif
383
f2fs_target_device(struct f2fs_sb_info * sbi,block_t blk_addr,sector_t * sector)384 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
385 block_t blk_addr, sector_t *sector)
386 {
387 struct block_device *bdev = sbi->sb->s_bdev;
388 int i;
389
390 if (f2fs_is_multi_device(sbi)) {
391 for (i = 0; i < sbi->s_ndevs; i++) {
392 if (FDEV(i).start_blk <= blk_addr &&
393 FDEV(i).end_blk >= blk_addr) {
394 blk_addr -= FDEV(i).start_blk;
395 bdev = FDEV(i).bdev;
396 break;
397 }
398 }
399 }
400
401 if (sector)
402 *sector = SECTOR_FROM_BLOCK(blk_addr);
403 return bdev;
404 }
405
f2fs_target_device_index(struct f2fs_sb_info * sbi,block_t blkaddr)406 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
407 {
408 int i;
409
410 if (!f2fs_is_multi_device(sbi))
411 return 0;
412
413 for (i = 0; i < sbi->s_ndevs; i++)
414 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
415 return i;
416 return 0;
417 }
418
f2fs_io_flags(struct f2fs_io_info * fio)419 static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio)
420 {
421 unsigned int temp_mask = GENMASK(NR_TEMP_TYPE - 1, 0);
422 unsigned int fua_flag, meta_flag, io_flag;
423 blk_opf_t op_flags = 0;
424
425 if (fio->op != REQ_OP_WRITE)
426 return 0;
427 if (fio->type == DATA)
428 io_flag = fio->sbi->data_io_flag;
429 else if (fio->type == NODE)
430 io_flag = fio->sbi->node_io_flag;
431 else
432 return 0;
433
434 fua_flag = io_flag & temp_mask;
435 meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
436
437 /*
438 * data/node io flag bits per temp:
439 * REQ_META | REQ_FUA |
440 * 5 | 4 | 3 | 2 | 1 | 0 |
441 * Cold | Warm | Hot | Cold | Warm | Hot |
442 */
443 if (BIT(fio->temp) & meta_flag)
444 op_flags |= REQ_META;
445 if (BIT(fio->temp) & fua_flag)
446 op_flags |= REQ_FUA;
447 return op_flags;
448 }
449
__bio_alloc(struct f2fs_io_info * fio,int npages)450 static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
451 {
452 struct f2fs_sb_info *sbi = fio->sbi;
453 struct block_device *bdev;
454 sector_t sector;
455 struct bio *bio;
456
457 bdev = f2fs_target_device(sbi, fio->new_blkaddr, §or);
458 bio = bio_alloc_bioset(bdev, npages,
459 fio->op | fio->op_flags | f2fs_io_flags(fio),
460 GFP_NOIO, &f2fs_bioset);
461 bio->bi_iter.bi_sector = sector;
462 if (is_read_io(fio->op)) {
463 bio->bi_end_io = f2fs_read_end_io;
464 bio->bi_private = NULL;
465 } else {
466 bio->bi_end_io = f2fs_write_end_io;
467 bio->bi_private = sbi;
468 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi,
469 fio->type, fio->temp);
470 }
471 iostat_alloc_and_bind_ctx(sbi, bio, NULL);
472
473 if (fio->io_wbc)
474 wbc_init_bio(fio->io_wbc, bio);
475
476 return bio;
477 }
478
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)479 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
480 pgoff_t first_idx,
481 const struct f2fs_io_info *fio,
482 gfp_t gfp_mask)
483 {
484 /*
485 * The f2fs garbage collector sets ->encrypted_page when it wants to
486 * read/write raw data without encryption.
487 */
488 if (!fio || !fio->encrypted_page)
489 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
490 }
491
f2fs_crypt_mergeable_bio(struct bio * bio,const struct inode * inode,pgoff_t next_idx,const struct f2fs_io_info * fio)492 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
493 pgoff_t next_idx,
494 const struct f2fs_io_info *fio)
495 {
496 /*
497 * The f2fs garbage collector sets ->encrypted_page when it wants to
498 * read/write raw data without encryption.
499 */
500 if (fio && fio->encrypted_page)
501 return !bio_has_crypt_ctx(bio);
502
503 return fscrypt_mergeable_bio(bio, inode, next_idx);
504 }
505
f2fs_submit_read_bio(struct f2fs_sb_info * sbi,struct bio * bio,enum page_type type)506 void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
507 enum page_type type)
508 {
509 WARN_ON_ONCE(!is_read_io(bio_op(bio)));
510 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
511
512 iostat_update_submit_ctx(bio, type);
513 submit_bio(bio);
514 }
515
f2fs_submit_write_bio(struct f2fs_sb_info * sbi,struct bio * bio,enum page_type type)516 static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio,
517 enum page_type type)
518 {
519 WARN_ON_ONCE(is_read_io(bio_op(bio)));
520
521 if (f2fs_lfs_mode(sbi) && current->plug && PAGE_TYPE_ON_MAIN(type))
522 blk_finish_plug(current->plug);
523
524 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
525 iostat_update_submit_ctx(bio, type);
526 submit_bio(bio);
527 }
528
__submit_merged_bio(struct f2fs_bio_info * io)529 static void __submit_merged_bio(struct f2fs_bio_info *io)
530 {
531 struct f2fs_io_info *fio = &io->fio;
532
533 if (!io->bio)
534 return;
535
536 if (is_read_io(fio->op)) {
537 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
538 f2fs_submit_read_bio(io->sbi, io->bio, fio->type);
539 } else {
540 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
541 f2fs_submit_write_bio(io->sbi, io->bio, fio->type);
542 }
543 io->bio = NULL;
544 }
545
__has_merged_page(struct bio * bio,struct inode * inode,struct page * page,nid_t ino)546 static bool __has_merged_page(struct bio *bio, struct inode *inode,
547 struct page *page, nid_t ino)
548 {
549 struct bio_vec *bvec;
550 struct bvec_iter_all iter_all;
551
552 if (!bio)
553 return false;
554
555 if (!inode && !page && !ino)
556 return true;
557
558 bio_for_each_segment_all(bvec, bio, iter_all) {
559 struct page *target = bvec->bv_page;
560
561 if (fscrypt_is_bounce_page(target)) {
562 target = fscrypt_pagecache_page(target);
563 if (IS_ERR(target))
564 continue;
565 }
566 if (f2fs_is_compressed_page(target)) {
567 target = f2fs_compress_control_page(target);
568 if (IS_ERR(target))
569 continue;
570 }
571
572 if (inode && inode == target->mapping->host)
573 return true;
574 if (page && page == target)
575 return true;
576 if (ino && ino == ino_of_node(target))
577 return true;
578 }
579
580 return false;
581 }
582
f2fs_init_write_merge_io(struct f2fs_sb_info * sbi)583 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
584 {
585 int i;
586
587 for (i = 0; i < NR_PAGE_TYPE; i++) {
588 int n = (i == META) ? 1 : NR_TEMP_TYPE;
589 int j;
590
591 sbi->write_io[i] = f2fs_kmalloc(sbi,
592 array_size(n, sizeof(struct f2fs_bio_info)),
593 GFP_KERNEL);
594 if (!sbi->write_io[i])
595 return -ENOMEM;
596
597 for (j = HOT; j < n; j++) {
598 struct f2fs_bio_info *io = &sbi->write_io[i][j];
599
600 init_f2fs_rwsem(&io->io_rwsem);
601 io->sbi = sbi;
602 io->bio = NULL;
603 io->last_block_in_bio = 0;
604 spin_lock_init(&io->io_lock);
605 INIT_LIST_HEAD(&io->io_list);
606 INIT_LIST_HEAD(&io->bio_list);
607 init_f2fs_rwsem(&io->bio_list_lock);
608 #ifdef CONFIG_BLK_DEV_ZONED
609 init_completion(&io->zone_wait);
610 io->zone_pending_bio = NULL;
611 io->bi_private = NULL;
612 #endif
613 }
614 }
615
616 return 0;
617 }
618
__f2fs_submit_merged_write(struct f2fs_sb_info * sbi,enum page_type type,enum temp_type temp)619 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
620 enum page_type type, enum temp_type temp)
621 {
622 enum page_type btype = PAGE_TYPE_OF_BIO(type);
623 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
624
625 f2fs_down_write(&io->io_rwsem);
626
627 if (!io->bio)
628 goto unlock_out;
629
630 /* change META to META_FLUSH in the checkpoint procedure */
631 if (type >= META_FLUSH) {
632 io->fio.type = META_FLUSH;
633 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
634 if (!test_opt(sbi, NOBARRIER))
635 io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
636 }
637 __submit_merged_bio(io);
638 unlock_out:
639 f2fs_up_write(&io->io_rwsem);
640 }
641
__submit_merged_write_cond(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page,nid_t ino,enum page_type type,bool force)642 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
643 struct inode *inode, struct page *page,
644 nid_t ino, enum page_type type, bool force)
645 {
646 enum temp_type temp;
647 bool ret = true;
648
649 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
650 if (!force) {
651 enum page_type btype = PAGE_TYPE_OF_BIO(type);
652 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
653
654 f2fs_down_read(&io->io_rwsem);
655 ret = __has_merged_page(io->bio, inode, page, ino);
656 f2fs_up_read(&io->io_rwsem);
657 }
658 if (ret)
659 __f2fs_submit_merged_write(sbi, type, temp);
660
661 /* TODO: use HOT temp only for meta pages now. */
662 if (type >= META)
663 break;
664 }
665 }
666
f2fs_submit_merged_write(struct f2fs_sb_info * sbi,enum page_type type)667 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
668 {
669 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
670 }
671
f2fs_submit_merged_write_cond(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page,nid_t ino,enum page_type type)672 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
673 struct inode *inode, struct page *page,
674 nid_t ino, enum page_type type)
675 {
676 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
677 }
678
f2fs_flush_merged_writes(struct f2fs_sb_info * sbi)679 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
680 {
681 f2fs_submit_merged_write(sbi, DATA);
682 f2fs_submit_merged_write(sbi, NODE);
683 f2fs_submit_merged_write(sbi, META);
684 }
685
686 /*
687 * Fill the locked page with data located in the block address.
688 * A caller needs to unlock the page on failure.
689 */
f2fs_submit_page_bio(struct f2fs_io_info * fio)690 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
691 {
692 struct bio *bio;
693 struct page *page = fio->encrypted_page ?
694 fio->encrypted_page : fio->page;
695
696 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
697 fio->is_por ? META_POR : (__is_meta_io(fio) ?
698 META_GENERIC : DATA_GENERIC_ENHANCE)))
699 return -EFSCORRUPTED;
700
701 trace_f2fs_submit_page_bio(page, fio);
702
703 /* Allocate a new bio */
704 bio = __bio_alloc(fio, 1);
705
706 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
707 fio->page->index, fio, GFP_NOIO);
708
709 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
710 bio_put(bio);
711 return -EFAULT;
712 }
713
714 if (fio->io_wbc && !is_read_io(fio->op))
715 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
716
717 inc_page_count(fio->sbi, is_read_io(fio->op) ?
718 __read_io_type(page) : WB_DATA_TYPE(fio->page, false));
719
720 if (is_read_io(bio_op(bio)))
721 f2fs_submit_read_bio(fio->sbi, bio, fio->type);
722 else
723 f2fs_submit_write_bio(fio->sbi, bio, fio->type);
724 return 0;
725 }
726
page_is_mergeable(struct f2fs_sb_info * sbi,struct bio * bio,block_t last_blkaddr,block_t cur_blkaddr)727 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
728 block_t last_blkaddr, block_t cur_blkaddr)
729 {
730 if (unlikely(sbi->max_io_bytes &&
731 bio->bi_iter.bi_size >= sbi->max_io_bytes))
732 return false;
733 if (last_blkaddr + 1 != cur_blkaddr)
734 return false;
735 return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
736 }
737
io_type_is_mergeable(struct f2fs_bio_info * io,struct f2fs_io_info * fio)738 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
739 struct f2fs_io_info *fio)
740 {
741 if (io->fio.op != fio->op)
742 return false;
743 return io->fio.op_flags == fio->op_flags;
744 }
745
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)746 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
747 struct f2fs_bio_info *io,
748 struct f2fs_io_info *fio,
749 block_t last_blkaddr,
750 block_t cur_blkaddr)
751 {
752 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
753 return false;
754 return io_type_is_mergeable(io, fio);
755 }
756
add_bio_entry(struct f2fs_sb_info * sbi,struct bio * bio,struct page * page,enum temp_type temp)757 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
758 struct page *page, enum temp_type temp)
759 {
760 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
761 struct bio_entry *be;
762
763 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
764 be->bio = bio;
765 bio_get(bio);
766
767 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
768 f2fs_bug_on(sbi, 1);
769
770 f2fs_down_write(&io->bio_list_lock);
771 list_add_tail(&be->list, &io->bio_list);
772 f2fs_up_write(&io->bio_list_lock);
773 }
774
del_bio_entry(struct bio_entry * be)775 static void del_bio_entry(struct bio_entry *be)
776 {
777 list_del(&be->list);
778 kmem_cache_free(bio_entry_slab, be);
779 }
780
add_ipu_page(struct f2fs_io_info * fio,struct bio ** bio,struct page * page)781 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
782 struct page *page)
783 {
784 struct f2fs_sb_info *sbi = fio->sbi;
785 enum temp_type temp;
786 bool found = false;
787 int ret = -EAGAIN;
788
789 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
790 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
791 struct list_head *head = &io->bio_list;
792 struct bio_entry *be;
793
794 f2fs_down_write(&io->bio_list_lock);
795 list_for_each_entry(be, head, list) {
796 if (be->bio != *bio)
797 continue;
798
799 found = true;
800
801 f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
802 *fio->last_block,
803 fio->new_blkaddr));
804 if (f2fs_crypt_mergeable_bio(*bio,
805 fio->page->mapping->host,
806 fio->page->index, fio) &&
807 bio_add_page(*bio, page, PAGE_SIZE, 0) ==
808 PAGE_SIZE) {
809 ret = 0;
810 break;
811 }
812
813 /* page can't be merged into bio; submit the bio */
814 del_bio_entry(be);
815 f2fs_submit_write_bio(sbi, *bio, DATA);
816 break;
817 }
818 f2fs_up_write(&io->bio_list_lock);
819 }
820
821 if (ret) {
822 bio_put(*bio);
823 *bio = NULL;
824 }
825
826 return ret;
827 }
828
f2fs_submit_merged_ipu_write(struct f2fs_sb_info * sbi,struct bio ** bio,struct page * page)829 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
830 struct bio **bio, struct page *page)
831 {
832 enum temp_type temp;
833 bool found = false;
834 struct bio *target = bio ? *bio : NULL;
835
836 f2fs_bug_on(sbi, !target && !page);
837
838 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
839 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
840 struct list_head *head = &io->bio_list;
841 struct bio_entry *be;
842
843 if (list_empty(head))
844 continue;
845
846 f2fs_down_read(&io->bio_list_lock);
847 list_for_each_entry(be, head, list) {
848 if (target)
849 found = (target == be->bio);
850 else
851 found = __has_merged_page(be->bio, NULL,
852 page, 0);
853 if (found)
854 break;
855 }
856 f2fs_up_read(&io->bio_list_lock);
857
858 if (!found)
859 continue;
860
861 found = false;
862
863 f2fs_down_write(&io->bio_list_lock);
864 list_for_each_entry(be, head, list) {
865 if (target)
866 found = (target == be->bio);
867 else
868 found = __has_merged_page(be->bio, NULL,
869 page, 0);
870 if (found) {
871 target = be->bio;
872 del_bio_entry(be);
873 break;
874 }
875 }
876 f2fs_up_write(&io->bio_list_lock);
877 }
878
879 if (found)
880 f2fs_submit_write_bio(sbi, target, DATA);
881 if (bio && *bio) {
882 bio_put(*bio);
883 *bio = NULL;
884 }
885 }
886
f2fs_merge_page_bio(struct f2fs_io_info * fio)887 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
888 {
889 struct bio *bio = *fio->bio;
890 struct page *page = fio->encrypted_page ?
891 fio->encrypted_page : fio->page;
892
893 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
894 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC))
895 return -EFSCORRUPTED;
896
897 trace_f2fs_submit_page_bio(page, fio);
898
899 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
900 fio->new_blkaddr))
901 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
902 alloc_new:
903 if (!bio) {
904 bio = __bio_alloc(fio, BIO_MAX_VECS);
905 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
906 fio->page->index, fio, GFP_NOIO);
907
908 add_bio_entry(fio->sbi, bio, page, fio->temp);
909 } else {
910 if (add_ipu_page(fio, &bio, page))
911 goto alloc_new;
912 }
913
914 if (fio->io_wbc)
915 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
916
917 inc_page_count(fio->sbi, WB_DATA_TYPE(page, false));
918
919 *fio->last_block = fio->new_blkaddr;
920 *fio->bio = bio;
921
922 return 0;
923 }
924
925 #ifdef CONFIG_BLK_DEV_ZONED
is_end_zone_blkaddr(struct f2fs_sb_info * sbi,block_t blkaddr)926 static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr)
927 {
928 struct block_device *bdev = sbi->sb->s_bdev;
929 int devi = 0;
930
931 if (f2fs_is_multi_device(sbi)) {
932 devi = f2fs_target_device_index(sbi, blkaddr);
933 if (blkaddr < FDEV(devi).start_blk ||
934 blkaddr > FDEV(devi).end_blk) {
935 f2fs_err(sbi, "Invalid block %x", blkaddr);
936 return false;
937 }
938 blkaddr -= FDEV(devi).start_blk;
939 bdev = FDEV(devi).bdev;
940 }
941 return bdev_is_zoned(bdev) &&
942 f2fs_blkz_is_seq(sbi, devi, blkaddr) &&
943 (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1);
944 }
945 #endif
946
f2fs_submit_page_write(struct f2fs_io_info * fio)947 void f2fs_submit_page_write(struct f2fs_io_info *fio)
948 {
949 struct f2fs_sb_info *sbi = fio->sbi;
950 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
951 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
952 struct page *bio_page;
953 enum count_type type;
954
955 f2fs_bug_on(sbi, is_read_io(fio->op));
956
957 f2fs_down_write(&io->io_rwsem);
958 next:
959 #ifdef CONFIG_BLK_DEV_ZONED
960 if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) {
961 wait_for_completion_io(&io->zone_wait);
962 bio_put(io->zone_pending_bio);
963 io->zone_pending_bio = NULL;
964 io->bi_private = NULL;
965 }
966 #endif
967
968 if (fio->in_list) {
969 spin_lock(&io->io_lock);
970 if (list_empty(&io->io_list)) {
971 spin_unlock(&io->io_lock);
972 goto out;
973 }
974 fio = list_first_entry(&io->io_list,
975 struct f2fs_io_info, list);
976 list_del(&fio->list);
977 spin_unlock(&io->io_lock);
978 }
979
980 verify_fio_blkaddr(fio);
981
982 if (fio->encrypted_page)
983 bio_page = fio->encrypted_page;
984 else if (fio->compressed_page)
985 bio_page = fio->compressed_page;
986 else
987 bio_page = fio->page;
988
989 /* set submitted = true as a return value */
990 fio->submitted = 1;
991
992 type = WB_DATA_TYPE(bio_page, fio->compressed_page);
993 inc_page_count(sbi, type);
994
995 if (io->bio &&
996 (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
997 fio->new_blkaddr) ||
998 !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
999 bio_page->index, fio)))
1000 __submit_merged_bio(io);
1001 alloc_new:
1002 if (io->bio == NULL) {
1003 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
1004 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
1005 bio_page->index, fio, GFP_NOIO);
1006 io->fio = *fio;
1007 }
1008
1009 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1010 __submit_merged_bio(io);
1011 goto alloc_new;
1012 }
1013
1014 if (fio->io_wbc)
1015 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
1016
1017 io->last_block_in_bio = fio->new_blkaddr;
1018
1019 trace_f2fs_submit_page_write(fio->page, fio);
1020 #ifdef CONFIG_BLK_DEV_ZONED
1021 if (f2fs_sb_has_blkzoned(sbi) && btype < META &&
1022 is_end_zone_blkaddr(sbi, fio->new_blkaddr)) {
1023 bio_get(io->bio);
1024 reinit_completion(&io->zone_wait);
1025 io->bi_private = io->bio->bi_private;
1026 io->bio->bi_private = io;
1027 io->bio->bi_end_io = f2fs_zone_write_end_io;
1028 io->zone_pending_bio = io->bio;
1029 __submit_merged_bio(io);
1030 }
1031 #endif
1032 if (fio->in_list)
1033 goto next;
1034 out:
1035 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1036 !f2fs_is_checkpoint_ready(sbi))
1037 __submit_merged_bio(io);
1038 f2fs_up_write(&io->io_rwsem);
1039 }
1040
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)1041 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1042 unsigned nr_pages, blk_opf_t op_flag,
1043 pgoff_t first_idx, bool for_write)
1044 {
1045 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1046 struct bio *bio;
1047 struct bio_post_read_ctx *ctx = NULL;
1048 unsigned int post_read_steps = 0;
1049 sector_t sector;
1050 struct block_device *bdev = f2fs_target_device(sbi, blkaddr, §or);
1051
1052 bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
1053 REQ_OP_READ | op_flag,
1054 for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
1055 if (!bio)
1056 return ERR_PTR(-ENOMEM);
1057 bio->bi_iter.bi_sector = sector;
1058 f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1059 bio->bi_end_io = f2fs_read_end_io;
1060
1061 if (fscrypt_inode_uses_fs_layer_crypto(inode))
1062 post_read_steps |= STEP_DECRYPT;
1063
1064 if (f2fs_need_verity(inode, first_idx))
1065 post_read_steps |= STEP_VERITY;
1066
1067 /*
1068 * STEP_DECOMPRESS is handled specially, since a compressed file might
1069 * contain both compressed and uncompressed clusters. We'll allocate a
1070 * bio_post_read_ctx if the file is compressed, but the caller is
1071 * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1072 */
1073
1074 if (post_read_steps || f2fs_compressed_file(inode)) {
1075 /* Due to the mempool, this never fails. */
1076 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1077 ctx->bio = bio;
1078 ctx->sbi = sbi;
1079 ctx->enabled_steps = post_read_steps;
1080 ctx->fs_blkaddr = blkaddr;
1081 ctx->decompression_attempted = false;
1082 bio->bi_private = ctx;
1083 }
1084 iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1085
1086 return bio;
1087 }
1088
1089 /* 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)1090 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1091 block_t blkaddr, blk_opf_t op_flags,
1092 bool for_write)
1093 {
1094 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1095 struct bio *bio;
1096
1097 bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1098 page->index, for_write);
1099 if (IS_ERR(bio))
1100 return PTR_ERR(bio);
1101
1102 /* wait for GCed page writeback via META_MAPPING */
1103 f2fs_wait_on_block_writeback(inode, blkaddr);
1104
1105 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1106 iostat_update_and_unbind_ctx(bio);
1107 if (bio->bi_private)
1108 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1109 bio_put(bio);
1110 return -EFAULT;
1111 }
1112 inc_page_count(sbi, F2FS_RD_DATA);
1113 f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE);
1114 f2fs_submit_read_bio(sbi, bio, DATA);
1115 return 0;
1116 }
1117
__set_data_blkaddr(struct dnode_of_data * dn,block_t blkaddr)1118 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1119 {
1120 __le32 *addr = get_dnode_addr(dn->inode, dn->node_page);
1121
1122 dn->data_blkaddr = blkaddr;
1123 addr[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1124 }
1125
1126 /*
1127 * Lock ordering for the change of data block address:
1128 * ->data_page
1129 * ->node_page
1130 * update block addresses in the node page
1131 */
f2fs_set_data_blkaddr(struct dnode_of_data * dn,block_t blkaddr)1132 void f2fs_set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1133 {
1134 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1135 __set_data_blkaddr(dn, blkaddr);
1136 if (set_page_dirty(dn->node_page))
1137 dn->node_changed = true;
1138 }
1139
f2fs_update_data_blkaddr(struct dnode_of_data * dn,block_t blkaddr)1140 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1141 {
1142 f2fs_set_data_blkaddr(dn, blkaddr);
1143 f2fs_update_read_extent_cache(dn);
1144 }
1145
1146 /* 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)1147 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1148 {
1149 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1150 int err;
1151
1152 if (!count)
1153 return 0;
1154
1155 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1156 return -EPERM;
1157 err = inc_valid_block_count(sbi, dn->inode, &count, true);
1158 if (unlikely(err))
1159 return err;
1160
1161 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1162 dn->ofs_in_node, count);
1163
1164 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1165
1166 for (; count > 0; dn->ofs_in_node++) {
1167 block_t blkaddr = f2fs_data_blkaddr(dn);
1168
1169 if (blkaddr == NULL_ADDR) {
1170 __set_data_blkaddr(dn, NEW_ADDR);
1171 count--;
1172 }
1173 }
1174
1175 if (set_page_dirty(dn->node_page))
1176 dn->node_changed = true;
1177 return 0;
1178 }
1179
1180 /* Should keep dn->ofs_in_node unchanged */
f2fs_reserve_new_block(struct dnode_of_data * dn)1181 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1182 {
1183 unsigned int ofs_in_node = dn->ofs_in_node;
1184 int ret;
1185
1186 ret = f2fs_reserve_new_blocks(dn, 1);
1187 dn->ofs_in_node = ofs_in_node;
1188 return ret;
1189 }
1190
f2fs_reserve_block(struct dnode_of_data * dn,pgoff_t index)1191 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1192 {
1193 bool need_put = dn->inode_page ? false : true;
1194 int err;
1195
1196 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1197 if (err)
1198 return err;
1199
1200 if (dn->data_blkaddr == NULL_ADDR)
1201 err = f2fs_reserve_new_block(dn);
1202 if (err || need_put)
1203 f2fs_put_dnode(dn);
1204 return err;
1205 }
1206
f2fs_get_read_data_page(struct inode * inode,pgoff_t index,blk_opf_t op_flags,bool for_write,pgoff_t * next_pgofs)1207 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1208 blk_opf_t op_flags, bool for_write,
1209 pgoff_t *next_pgofs)
1210 {
1211 struct address_space *mapping = inode->i_mapping;
1212 struct dnode_of_data dn;
1213 struct page *page;
1214 int err;
1215
1216 page = f2fs_grab_cache_page(mapping, index, for_write);
1217 if (!page)
1218 return ERR_PTR(-ENOMEM);
1219
1220 if (f2fs_lookup_read_extent_cache_block(inode, index,
1221 &dn.data_blkaddr)) {
1222 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1223 DATA_GENERIC_ENHANCE_READ)) {
1224 err = -EFSCORRUPTED;
1225 goto put_err;
1226 }
1227 goto got_it;
1228 }
1229
1230 set_new_dnode(&dn, inode, NULL, NULL, 0);
1231 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1232 if (err) {
1233 if (err == -ENOENT && next_pgofs)
1234 *next_pgofs = f2fs_get_next_page_offset(&dn, index);
1235 goto put_err;
1236 }
1237 f2fs_put_dnode(&dn);
1238
1239 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1240 err = -ENOENT;
1241 if (next_pgofs)
1242 *next_pgofs = index + 1;
1243 goto put_err;
1244 }
1245 if (dn.data_blkaddr != NEW_ADDR &&
1246 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1247 dn.data_blkaddr,
1248 DATA_GENERIC_ENHANCE)) {
1249 err = -EFSCORRUPTED;
1250 goto put_err;
1251 }
1252 got_it:
1253 if (PageUptodate(page)) {
1254 unlock_page(page);
1255 return page;
1256 }
1257
1258 /*
1259 * A new dentry page is allocated but not able to be written, since its
1260 * new inode page couldn't be allocated due to -ENOSPC.
1261 * In such the case, its blkaddr can be remained as NEW_ADDR.
1262 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1263 * f2fs_init_inode_metadata.
1264 */
1265 if (dn.data_blkaddr == NEW_ADDR) {
1266 zero_user_segment(page, 0, PAGE_SIZE);
1267 if (!PageUptodate(page))
1268 SetPageUptodate(page);
1269 unlock_page(page);
1270 return page;
1271 }
1272
1273 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1274 op_flags, for_write);
1275 if (err)
1276 goto put_err;
1277 return page;
1278
1279 put_err:
1280 f2fs_put_page(page, 1);
1281 return ERR_PTR(err);
1282 }
1283
f2fs_find_data_page(struct inode * inode,pgoff_t index,pgoff_t * next_pgofs)1284 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index,
1285 pgoff_t *next_pgofs)
1286 {
1287 struct address_space *mapping = inode->i_mapping;
1288 struct page *page;
1289
1290 page = find_get_page(mapping, index);
1291 if (page && PageUptodate(page))
1292 return page;
1293 f2fs_put_page(page, 0);
1294
1295 page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs);
1296 if (IS_ERR(page))
1297 return page;
1298
1299 if (PageUptodate(page))
1300 return page;
1301
1302 wait_on_page_locked(page);
1303 if (unlikely(!PageUptodate(page))) {
1304 f2fs_put_page(page, 0);
1305 return ERR_PTR(-EIO);
1306 }
1307 return page;
1308 }
1309
1310 /*
1311 * If it tries to access a hole, return an error.
1312 * Because, the callers, functions in dir.c and GC, should be able to know
1313 * whether this page exists or not.
1314 */
f2fs_get_lock_data_page(struct inode * inode,pgoff_t index,bool for_write)1315 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1316 bool for_write)
1317 {
1318 struct address_space *mapping = inode->i_mapping;
1319 struct page *page;
1320
1321 page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL);
1322 if (IS_ERR(page))
1323 return page;
1324
1325 /* wait for read completion */
1326 lock_page(page);
1327 if (unlikely(page->mapping != mapping || !PageUptodate(page))) {
1328 f2fs_put_page(page, 1);
1329 return ERR_PTR(-EIO);
1330 }
1331 return page;
1332 }
1333
1334 /*
1335 * Caller ensures that this data page is never allocated.
1336 * A new zero-filled data page is allocated in the page cache.
1337 *
1338 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1339 * f2fs_unlock_op().
1340 * Note that, ipage is set only by make_empty_dir, and if any error occur,
1341 * ipage should be released by this function.
1342 */
f2fs_get_new_data_page(struct inode * inode,struct page * ipage,pgoff_t index,bool new_i_size)1343 struct page *f2fs_get_new_data_page(struct inode *inode,
1344 struct page *ipage, pgoff_t index, bool new_i_size)
1345 {
1346 struct address_space *mapping = inode->i_mapping;
1347 struct page *page;
1348 struct dnode_of_data dn;
1349 int err;
1350
1351 page = f2fs_grab_cache_page(mapping, index, true);
1352 if (!page) {
1353 /*
1354 * before exiting, we should make sure ipage will be released
1355 * if any error occur.
1356 */
1357 f2fs_put_page(ipage, 1);
1358 return ERR_PTR(-ENOMEM);
1359 }
1360
1361 set_new_dnode(&dn, inode, ipage, NULL, 0);
1362 err = f2fs_reserve_block(&dn, index);
1363 if (err) {
1364 f2fs_put_page(page, 1);
1365 return ERR_PTR(err);
1366 }
1367 if (!ipage)
1368 f2fs_put_dnode(&dn);
1369
1370 if (PageUptodate(page))
1371 goto got_it;
1372
1373 if (dn.data_blkaddr == NEW_ADDR) {
1374 zero_user_segment(page, 0, PAGE_SIZE);
1375 if (!PageUptodate(page))
1376 SetPageUptodate(page);
1377 } else {
1378 f2fs_put_page(page, 1);
1379
1380 /* if ipage exists, blkaddr should be NEW_ADDR */
1381 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1382 page = f2fs_get_lock_data_page(inode, index, true);
1383 if (IS_ERR(page))
1384 return page;
1385 }
1386 got_it:
1387 if (new_i_size && i_size_read(inode) <
1388 ((loff_t)(index + 1) << PAGE_SHIFT))
1389 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1390 return page;
1391 }
1392
__allocate_data_block(struct dnode_of_data * dn,int seg_type)1393 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1394 {
1395 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1396 struct f2fs_summary sum;
1397 struct node_info ni;
1398 block_t old_blkaddr;
1399 blkcnt_t count = 1;
1400 int err;
1401
1402 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1403 return -EPERM;
1404
1405 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1406 if (err)
1407 return err;
1408
1409 dn->data_blkaddr = f2fs_data_blkaddr(dn);
1410 if (dn->data_blkaddr == NULL_ADDR) {
1411 err = inc_valid_block_count(sbi, dn->inode, &count, true);
1412 if (unlikely(err))
1413 return err;
1414 }
1415
1416 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1417 old_blkaddr = dn->data_blkaddr;
1418 err = f2fs_allocate_data_block(sbi, NULL, old_blkaddr,
1419 &dn->data_blkaddr, &sum, seg_type, NULL);
1420 if (err)
1421 return err;
1422
1423 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1424 f2fs_invalidate_internal_cache(sbi, old_blkaddr);
1425
1426 f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1427 return 0;
1428 }
1429
f2fs_map_lock(struct f2fs_sb_info * sbi,int flag)1430 static void f2fs_map_lock(struct f2fs_sb_info *sbi, int flag)
1431 {
1432 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1433 f2fs_down_read(&sbi->node_change);
1434 else
1435 f2fs_lock_op(sbi);
1436 }
1437
f2fs_map_unlock(struct f2fs_sb_info * sbi,int flag)1438 static void f2fs_map_unlock(struct f2fs_sb_info *sbi, int flag)
1439 {
1440 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1441 f2fs_up_read(&sbi->node_change);
1442 else
1443 f2fs_unlock_op(sbi);
1444 }
1445
f2fs_get_block_locked(struct dnode_of_data * dn,pgoff_t index)1446 int f2fs_get_block_locked(struct dnode_of_data *dn, pgoff_t index)
1447 {
1448 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1449 int err = 0;
1450
1451 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1452 if (!f2fs_lookup_read_extent_cache_block(dn->inode, index,
1453 &dn->data_blkaddr))
1454 err = f2fs_reserve_block(dn, index);
1455 f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1456
1457 return err;
1458 }
1459
f2fs_map_no_dnode(struct inode * inode,struct f2fs_map_blocks * map,struct dnode_of_data * dn,pgoff_t pgoff)1460 static int f2fs_map_no_dnode(struct inode *inode,
1461 struct f2fs_map_blocks *map, struct dnode_of_data *dn,
1462 pgoff_t pgoff)
1463 {
1464 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1465
1466 /*
1467 * There is one exceptional case that read_node_page() may return
1468 * -ENOENT due to filesystem has been shutdown or cp_error, return
1469 * -EIO in that case.
1470 */
1471 if (map->m_may_create &&
1472 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || f2fs_cp_error(sbi)))
1473 return -EIO;
1474
1475 if (map->m_next_pgofs)
1476 *map->m_next_pgofs = f2fs_get_next_page_offset(dn, pgoff);
1477 if (map->m_next_extent)
1478 *map->m_next_extent = f2fs_get_next_page_offset(dn, pgoff);
1479 return 0;
1480 }
1481
f2fs_map_blocks_cached(struct inode * inode,struct f2fs_map_blocks * map,int flag)1482 static bool f2fs_map_blocks_cached(struct inode *inode,
1483 struct f2fs_map_blocks *map, int flag)
1484 {
1485 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1486 unsigned int maxblocks = map->m_len;
1487 pgoff_t pgoff = (pgoff_t)map->m_lblk;
1488 struct extent_info ei = {};
1489
1490 if (!f2fs_lookup_read_extent_cache(inode, pgoff, &ei))
1491 return false;
1492
1493 map->m_pblk = ei.blk + pgoff - ei.fofs;
1494 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgoff);
1495 map->m_flags = F2FS_MAP_MAPPED;
1496 if (map->m_next_extent)
1497 *map->m_next_extent = pgoff + map->m_len;
1498
1499 /* for hardware encryption, but to avoid potential issue in future */
1500 if (flag == F2FS_GET_BLOCK_DIO)
1501 f2fs_wait_on_block_writeback_range(inode,
1502 map->m_pblk, map->m_len);
1503
1504 if (f2fs_allow_multi_device_dio(sbi, flag)) {
1505 int bidx = f2fs_target_device_index(sbi, map->m_pblk);
1506 struct f2fs_dev_info *dev = &sbi->devs[bidx];
1507
1508 map->m_bdev = dev->bdev;
1509 map->m_pblk -= dev->start_blk;
1510 map->m_len = min(map->m_len, dev->end_blk + 1 - map->m_pblk);
1511 } else {
1512 map->m_bdev = inode->i_sb->s_bdev;
1513 }
1514 return true;
1515 }
1516
map_is_mergeable(struct f2fs_sb_info * sbi,struct f2fs_map_blocks * map,block_t blkaddr,int flag,int bidx,int ofs)1517 static bool map_is_mergeable(struct f2fs_sb_info *sbi,
1518 struct f2fs_map_blocks *map,
1519 block_t blkaddr, int flag, int bidx,
1520 int ofs)
1521 {
1522 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1523 return false;
1524 if (map->m_pblk != NEW_ADDR && blkaddr == (map->m_pblk + ofs))
1525 return true;
1526 if (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR)
1527 return true;
1528 if (flag == F2FS_GET_BLOCK_PRE_DIO)
1529 return true;
1530 if (flag == F2FS_GET_BLOCK_DIO &&
1531 map->m_pblk == NULL_ADDR && blkaddr == NULL_ADDR)
1532 return true;
1533 return false;
1534 }
1535
1536 /*
1537 * f2fs_map_blocks() tries to find or build mapping relationship which
1538 * maps continuous logical blocks to physical blocks, and return such
1539 * info via f2fs_map_blocks structure.
1540 */
f2fs_map_blocks(struct inode * inode,struct f2fs_map_blocks * map,int flag)1541 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, int flag)
1542 {
1543 unsigned int maxblocks = map->m_len;
1544 struct dnode_of_data dn;
1545 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1546 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1547 pgoff_t pgofs, end_offset, end;
1548 int err = 0, ofs = 1;
1549 unsigned int ofs_in_node, last_ofs_in_node;
1550 blkcnt_t prealloc;
1551 block_t blkaddr;
1552 unsigned int start_pgofs;
1553 int bidx = 0;
1554 bool is_hole;
1555
1556 if (!maxblocks)
1557 return 0;
1558
1559 if (!map->m_may_create && f2fs_map_blocks_cached(inode, map, flag))
1560 goto out;
1561
1562 map->m_bdev = inode->i_sb->s_bdev;
1563 map->m_multidev_dio =
1564 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1565
1566 map->m_len = 0;
1567 map->m_flags = 0;
1568
1569 /* it only supports block size == page size */
1570 pgofs = (pgoff_t)map->m_lblk;
1571 end = pgofs + maxblocks;
1572
1573 next_dnode:
1574 if (map->m_may_create)
1575 f2fs_map_lock(sbi, flag);
1576
1577 /* When reading holes, we need its node page */
1578 set_new_dnode(&dn, inode, NULL, NULL, 0);
1579 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1580 if (err) {
1581 if (flag == F2FS_GET_BLOCK_BMAP)
1582 map->m_pblk = 0;
1583 if (err == -ENOENT)
1584 err = f2fs_map_no_dnode(inode, map, &dn, pgofs);
1585 goto unlock_out;
1586 }
1587
1588 start_pgofs = pgofs;
1589 prealloc = 0;
1590 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1591 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1592
1593 next_block:
1594 blkaddr = f2fs_data_blkaddr(&dn);
1595 is_hole = !__is_valid_data_blkaddr(blkaddr);
1596 if (!is_hole &&
1597 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1598 err = -EFSCORRUPTED;
1599 goto sync_out;
1600 }
1601
1602 /* use out-place-update for direct IO under LFS mode */
1603 if (map->m_may_create && (is_hole ||
1604 (flag == F2FS_GET_BLOCK_DIO && f2fs_lfs_mode(sbi) &&
1605 !f2fs_is_pinned_file(inode)))) {
1606 if (unlikely(f2fs_cp_error(sbi))) {
1607 err = -EIO;
1608 goto sync_out;
1609 }
1610
1611 switch (flag) {
1612 case F2FS_GET_BLOCK_PRE_AIO:
1613 if (blkaddr == NULL_ADDR) {
1614 prealloc++;
1615 last_ofs_in_node = dn.ofs_in_node;
1616 }
1617 break;
1618 case F2FS_GET_BLOCK_PRE_DIO:
1619 case F2FS_GET_BLOCK_DIO:
1620 err = __allocate_data_block(&dn, map->m_seg_type);
1621 if (err)
1622 goto sync_out;
1623 if (flag == F2FS_GET_BLOCK_PRE_DIO)
1624 file_need_truncate(inode);
1625 set_inode_flag(inode, FI_APPEND_WRITE);
1626 break;
1627 default:
1628 WARN_ON_ONCE(1);
1629 err = -EIO;
1630 goto sync_out;
1631 }
1632
1633 blkaddr = dn.data_blkaddr;
1634 if (is_hole)
1635 map->m_flags |= F2FS_MAP_NEW;
1636 } else if (is_hole) {
1637 if (f2fs_compressed_file(inode) &&
1638 f2fs_sanity_check_cluster(&dn)) {
1639 err = -EFSCORRUPTED;
1640 f2fs_handle_error(sbi,
1641 ERROR_CORRUPTED_CLUSTER);
1642 goto sync_out;
1643 }
1644
1645 switch (flag) {
1646 case F2FS_GET_BLOCK_PRECACHE:
1647 goto sync_out;
1648 case F2FS_GET_BLOCK_BMAP:
1649 map->m_pblk = 0;
1650 goto sync_out;
1651 case F2FS_GET_BLOCK_FIEMAP:
1652 if (blkaddr == NULL_ADDR) {
1653 if (map->m_next_pgofs)
1654 *map->m_next_pgofs = pgofs + 1;
1655 goto sync_out;
1656 }
1657 break;
1658 case F2FS_GET_BLOCK_DIO:
1659 if (map->m_next_pgofs)
1660 *map->m_next_pgofs = pgofs + 1;
1661 break;
1662 default:
1663 /* for defragment case */
1664 if (map->m_next_pgofs)
1665 *map->m_next_pgofs = pgofs + 1;
1666 goto sync_out;
1667 }
1668 }
1669
1670 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1671 goto skip;
1672
1673 if (map->m_multidev_dio)
1674 bidx = f2fs_target_device_index(sbi, blkaddr);
1675
1676 if (map->m_len == 0) {
1677 /* reserved delalloc block should be mapped for fiemap. */
1678 if (blkaddr == NEW_ADDR)
1679 map->m_flags |= F2FS_MAP_DELALLOC;
1680 if (flag != F2FS_GET_BLOCK_DIO || !is_hole)
1681 map->m_flags |= F2FS_MAP_MAPPED;
1682
1683 map->m_pblk = blkaddr;
1684 map->m_len = 1;
1685
1686 if (map->m_multidev_dio)
1687 map->m_bdev = FDEV(bidx).bdev;
1688 } else if (map_is_mergeable(sbi, map, blkaddr, flag, bidx, ofs)) {
1689 ofs++;
1690 map->m_len++;
1691 } else {
1692 goto sync_out;
1693 }
1694
1695 skip:
1696 dn.ofs_in_node++;
1697 pgofs++;
1698
1699 /* preallocate blocks in batch for one dnode page */
1700 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1701 (pgofs == end || dn.ofs_in_node == end_offset)) {
1702
1703 dn.ofs_in_node = ofs_in_node;
1704 err = f2fs_reserve_new_blocks(&dn, prealloc);
1705 if (err)
1706 goto sync_out;
1707
1708 map->m_len += dn.ofs_in_node - ofs_in_node;
1709 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1710 err = -ENOSPC;
1711 goto sync_out;
1712 }
1713 dn.ofs_in_node = end_offset;
1714 }
1715
1716 if (pgofs >= end)
1717 goto sync_out;
1718 else if (dn.ofs_in_node < end_offset)
1719 goto next_block;
1720
1721 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1722 if (map->m_flags & F2FS_MAP_MAPPED) {
1723 unsigned int ofs = start_pgofs - map->m_lblk;
1724
1725 f2fs_update_read_extent_cache_range(&dn,
1726 start_pgofs, map->m_pblk + ofs,
1727 map->m_len - ofs);
1728 }
1729 }
1730
1731 f2fs_put_dnode(&dn);
1732
1733 if (map->m_may_create) {
1734 f2fs_map_unlock(sbi, flag);
1735 f2fs_balance_fs(sbi, dn.node_changed);
1736 }
1737 goto next_dnode;
1738
1739 sync_out:
1740
1741 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1742 /*
1743 * for hardware encryption, but to avoid potential issue
1744 * in future
1745 */
1746 f2fs_wait_on_block_writeback_range(inode,
1747 map->m_pblk, map->m_len);
1748
1749 if (map->m_multidev_dio) {
1750 block_t blk_addr = map->m_pblk;
1751
1752 bidx = f2fs_target_device_index(sbi, map->m_pblk);
1753
1754 map->m_bdev = FDEV(bidx).bdev;
1755 map->m_pblk -= FDEV(bidx).start_blk;
1756
1757 if (map->m_may_create)
1758 f2fs_update_device_state(sbi, inode->i_ino,
1759 blk_addr, map->m_len);
1760
1761 f2fs_bug_on(sbi, blk_addr + map->m_len >
1762 FDEV(bidx).end_blk + 1);
1763 }
1764 }
1765
1766 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1767 if (map->m_flags & F2FS_MAP_MAPPED) {
1768 unsigned int ofs = start_pgofs - map->m_lblk;
1769
1770 f2fs_update_read_extent_cache_range(&dn,
1771 start_pgofs, map->m_pblk + ofs,
1772 map->m_len - ofs);
1773 }
1774 if (map->m_next_extent)
1775 *map->m_next_extent = pgofs + 1;
1776 }
1777 f2fs_put_dnode(&dn);
1778 unlock_out:
1779 if (map->m_may_create) {
1780 f2fs_map_unlock(sbi, flag);
1781 f2fs_balance_fs(sbi, dn.node_changed);
1782 }
1783 out:
1784 trace_f2fs_map_blocks(inode, map, flag, err);
1785 return err;
1786 }
1787
f2fs_overwrite_io(struct inode * inode,loff_t pos,size_t len)1788 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1789 {
1790 struct f2fs_map_blocks map;
1791 block_t last_lblk;
1792 int err;
1793
1794 if (pos + len > i_size_read(inode))
1795 return false;
1796
1797 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1798 map.m_next_pgofs = NULL;
1799 map.m_next_extent = NULL;
1800 map.m_seg_type = NO_CHECK_TYPE;
1801 map.m_may_create = false;
1802 last_lblk = F2FS_BLK_ALIGN(pos + len);
1803
1804 while (map.m_lblk < last_lblk) {
1805 map.m_len = last_lblk - map.m_lblk;
1806 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT);
1807 if (err || map.m_len == 0)
1808 return false;
1809 map.m_lblk += map.m_len;
1810 }
1811 return true;
1812 }
1813
bytes_to_blks(struct inode * inode,u64 bytes)1814 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1815 {
1816 return (bytes >> inode->i_blkbits);
1817 }
1818
blks_to_bytes(struct inode * inode,u64 blks)1819 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1820 {
1821 return (blks << inode->i_blkbits);
1822 }
1823
f2fs_xattr_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo)1824 static int f2fs_xattr_fiemap(struct inode *inode,
1825 struct fiemap_extent_info *fieinfo)
1826 {
1827 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1828 struct page *page;
1829 struct node_info ni;
1830 __u64 phys = 0, len;
1831 __u32 flags;
1832 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1833 int err = 0;
1834
1835 if (f2fs_has_inline_xattr(inode)) {
1836 int offset;
1837
1838 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1839 inode->i_ino, false);
1840 if (!page)
1841 return -ENOMEM;
1842
1843 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1844 if (err) {
1845 f2fs_put_page(page, 1);
1846 return err;
1847 }
1848
1849 phys = blks_to_bytes(inode, ni.blk_addr);
1850 offset = offsetof(struct f2fs_inode, i_addr) +
1851 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1852 get_inline_xattr_addrs(inode));
1853
1854 phys += offset;
1855 len = inline_xattr_size(inode);
1856
1857 f2fs_put_page(page, 1);
1858
1859 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1860
1861 if (!xnid)
1862 flags |= FIEMAP_EXTENT_LAST;
1863
1864 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1865 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1866 if (err)
1867 return err;
1868 }
1869
1870 if (xnid) {
1871 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1872 if (!page)
1873 return -ENOMEM;
1874
1875 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1876 if (err) {
1877 f2fs_put_page(page, 1);
1878 return err;
1879 }
1880
1881 phys = blks_to_bytes(inode, ni.blk_addr);
1882 len = inode->i_sb->s_blocksize;
1883
1884 f2fs_put_page(page, 1);
1885
1886 flags = FIEMAP_EXTENT_LAST;
1887 }
1888
1889 if (phys) {
1890 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1891 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1892 }
1893
1894 return (err < 0 ? err : 0);
1895 }
1896
max_inode_blocks(struct inode * inode)1897 static loff_t max_inode_blocks(struct inode *inode)
1898 {
1899 loff_t result = ADDRS_PER_INODE(inode);
1900 loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1901
1902 /* two direct node blocks */
1903 result += (leaf_count * 2);
1904
1905 /* two indirect node blocks */
1906 leaf_count *= NIDS_PER_BLOCK;
1907 result += (leaf_count * 2);
1908
1909 /* one double indirect node block */
1910 leaf_count *= NIDS_PER_BLOCK;
1911 result += leaf_count;
1912
1913 return result;
1914 }
1915
f2fs_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 start,u64 len)1916 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1917 u64 start, u64 len)
1918 {
1919 struct f2fs_map_blocks map;
1920 sector_t start_blk, last_blk;
1921 pgoff_t next_pgofs;
1922 u64 logical = 0, phys = 0, size = 0;
1923 u32 flags = 0;
1924 int ret = 0;
1925 bool compr_cluster = false, compr_appended;
1926 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1927 unsigned int count_in_cluster = 0;
1928 loff_t maxbytes;
1929
1930 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1931 ret = f2fs_precache_extents(inode);
1932 if (ret)
1933 return ret;
1934 }
1935
1936 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1937 if (ret)
1938 return ret;
1939
1940 inode_lock_shared(inode);
1941
1942 maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1943 if (start > maxbytes) {
1944 ret = -EFBIG;
1945 goto out;
1946 }
1947
1948 if (len > maxbytes || (maxbytes - len) < start)
1949 len = maxbytes - start;
1950
1951 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1952 ret = f2fs_xattr_fiemap(inode, fieinfo);
1953 goto out;
1954 }
1955
1956 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1957 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1958 if (ret != -EAGAIN)
1959 goto out;
1960 }
1961
1962 if (bytes_to_blks(inode, len) == 0)
1963 len = blks_to_bytes(inode, 1);
1964
1965 start_blk = bytes_to_blks(inode, start);
1966 last_blk = bytes_to_blks(inode, start + len - 1);
1967
1968 next:
1969 memset(&map, 0, sizeof(map));
1970 map.m_lblk = start_blk;
1971 map.m_len = bytes_to_blks(inode, len);
1972 map.m_next_pgofs = &next_pgofs;
1973 map.m_seg_type = NO_CHECK_TYPE;
1974
1975 if (compr_cluster) {
1976 map.m_lblk += 1;
1977 map.m_len = cluster_size - count_in_cluster;
1978 }
1979
1980 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
1981 if (ret)
1982 goto out;
1983
1984 /* HOLE */
1985 if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1986 start_blk = next_pgofs;
1987
1988 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1989 max_inode_blocks(inode)))
1990 goto prep_next;
1991
1992 flags |= FIEMAP_EXTENT_LAST;
1993 }
1994
1995 compr_appended = false;
1996 /* In a case of compressed cluster, append this to the last extent */
1997 if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) ||
1998 !(map.m_flags & F2FS_MAP_FLAGS))) {
1999 compr_appended = true;
2000 goto skip_fill;
2001 }
2002
2003 if (size) {
2004 flags |= FIEMAP_EXTENT_MERGED;
2005 if (IS_ENCRYPTED(inode))
2006 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
2007
2008 ret = fiemap_fill_next_extent(fieinfo, logical,
2009 phys, size, flags);
2010 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
2011 if (ret)
2012 goto out;
2013 size = 0;
2014 }
2015
2016 if (start_blk > last_blk)
2017 goto out;
2018
2019 skip_fill:
2020 if (map.m_pblk == COMPRESS_ADDR) {
2021 compr_cluster = true;
2022 count_in_cluster = 1;
2023 } else if (compr_appended) {
2024 unsigned int appended_blks = cluster_size -
2025 count_in_cluster + 1;
2026 size += blks_to_bytes(inode, appended_blks);
2027 start_blk += appended_blks;
2028 compr_cluster = false;
2029 } else {
2030 logical = blks_to_bytes(inode, start_blk);
2031 phys = __is_valid_data_blkaddr(map.m_pblk) ?
2032 blks_to_bytes(inode, map.m_pblk) : 0;
2033 size = blks_to_bytes(inode, map.m_len);
2034 flags = 0;
2035
2036 if (compr_cluster) {
2037 flags = FIEMAP_EXTENT_ENCODED;
2038 count_in_cluster += map.m_len;
2039 if (count_in_cluster == cluster_size) {
2040 compr_cluster = false;
2041 size += blks_to_bytes(inode, 1);
2042 }
2043 } else if (map.m_flags & F2FS_MAP_DELALLOC) {
2044 flags = FIEMAP_EXTENT_UNWRITTEN;
2045 }
2046
2047 start_blk += bytes_to_blks(inode, size);
2048 }
2049
2050 prep_next:
2051 cond_resched();
2052 if (fatal_signal_pending(current))
2053 ret = -EINTR;
2054 else
2055 goto next;
2056 out:
2057 if (ret == 1)
2058 ret = 0;
2059
2060 inode_unlock_shared(inode);
2061 return ret;
2062 }
2063
f2fs_readpage_limit(struct inode * inode)2064 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2065 {
2066 if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
2067 return inode->i_sb->s_maxbytes;
2068
2069 return i_size_read(inode);
2070 }
2071
f2fs_ra_op_flags(struct readahead_control * rac)2072 static inline blk_opf_t f2fs_ra_op_flags(struct readahead_control *rac)
2073 {
2074 return rac ? REQ_RAHEAD : 0;
2075 }
2076
f2fs_read_single_page(struct inode * inode,struct folio * folio,unsigned nr_pages,struct f2fs_map_blocks * map,struct bio ** bio_ret,sector_t * last_block_in_bio,struct readahead_control * rac)2077 static int f2fs_read_single_page(struct inode *inode, struct folio *folio,
2078 unsigned nr_pages,
2079 struct f2fs_map_blocks *map,
2080 struct bio **bio_ret,
2081 sector_t *last_block_in_bio,
2082 struct readahead_control *rac)
2083 {
2084 struct bio *bio = *bio_ret;
2085 const unsigned blocksize = blks_to_bytes(inode, 1);
2086 sector_t block_in_file;
2087 sector_t last_block;
2088 sector_t last_block_in_file;
2089 sector_t block_nr;
2090 pgoff_t index = folio_index(folio);
2091 int ret = 0;
2092
2093 block_in_file = (sector_t)index;
2094 last_block = block_in_file + nr_pages;
2095 last_block_in_file = bytes_to_blks(inode,
2096 f2fs_readpage_limit(inode) + blocksize - 1);
2097 if (last_block > last_block_in_file)
2098 last_block = last_block_in_file;
2099
2100 /* just zeroing out page which is beyond EOF */
2101 if (block_in_file >= last_block)
2102 goto zero_out;
2103 /*
2104 * Map blocks using the previous result first.
2105 */
2106 if ((map->m_flags & F2FS_MAP_MAPPED) &&
2107 block_in_file > map->m_lblk &&
2108 block_in_file < (map->m_lblk + map->m_len))
2109 goto got_it;
2110
2111 /*
2112 * Then do more f2fs_map_blocks() calls until we are
2113 * done with this page.
2114 */
2115 map->m_lblk = block_in_file;
2116 map->m_len = last_block - block_in_file;
2117
2118 ret = f2fs_map_blocks(inode, map, F2FS_GET_BLOCK_DEFAULT);
2119 if (ret)
2120 goto out;
2121 got_it:
2122 if ((map->m_flags & F2FS_MAP_MAPPED)) {
2123 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2124 folio_set_mappedtodisk(folio);
2125
2126 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2127 DATA_GENERIC_ENHANCE_READ)) {
2128 ret = -EFSCORRUPTED;
2129 goto out;
2130 }
2131 } else {
2132 zero_out:
2133 folio_zero_segment(folio, 0, folio_size(folio));
2134 if (f2fs_need_verity(inode, index) &&
2135 !fsverity_verify_folio(folio)) {
2136 ret = -EIO;
2137 goto out;
2138 }
2139 if (!folio_test_uptodate(folio))
2140 folio_mark_uptodate(folio);
2141 folio_unlock(folio);
2142 goto out;
2143 }
2144
2145 /*
2146 * This page will go to BIO. Do we need to send this
2147 * BIO off first?
2148 */
2149 if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2150 *last_block_in_bio, block_nr) ||
2151 !f2fs_crypt_mergeable_bio(bio, inode, index, NULL))) {
2152 submit_and_realloc:
2153 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2154 bio = NULL;
2155 }
2156 if (bio == NULL) {
2157 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2158 f2fs_ra_op_flags(rac), index,
2159 false);
2160 if (IS_ERR(bio)) {
2161 ret = PTR_ERR(bio);
2162 bio = NULL;
2163 goto out;
2164 }
2165 }
2166
2167 /*
2168 * If the page is under writeback, we need to wait for
2169 * its completion to see the correct decrypted data.
2170 */
2171 f2fs_wait_on_block_writeback(inode, block_nr);
2172
2173 if (!bio_add_folio(bio, folio, blocksize, 0))
2174 goto submit_and_realloc;
2175
2176 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2177 f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO,
2178 F2FS_BLKSIZE);
2179 *last_block_in_bio = block_nr;
2180 out:
2181 *bio_ret = bio;
2182 return ret;
2183 }
2184
2185 #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,struct readahead_control * rac,bool for_write)2186 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2187 unsigned nr_pages, sector_t *last_block_in_bio,
2188 struct readahead_control *rac, bool for_write)
2189 {
2190 struct dnode_of_data dn;
2191 struct inode *inode = cc->inode;
2192 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2193 struct bio *bio = *bio_ret;
2194 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2195 sector_t last_block_in_file;
2196 const unsigned blocksize = blks_to_bytes(inode, 1);
2197 struct decompress_io_ctx *dic = NULL;
2198 struct extent_info ei = {};
2199 bool from_dnode = true;
2200 int i;
2201 int ret = 0;
2202
2203 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2204
2205 last_block_in_file = bytes_to_blks(inode,
2206 f2fs_readpage_limit(inode) + blocksize - 1);
2207
2208 /* get rid of pages beyond EOF */
2209 for (i = 0; i < cc->cluster_size; i++) {
2210 struct page *page = cc->rpages[i];
2211
2212 if (!page)
2213 continue;
2214 if ((sector_t)page->index >= last_block_in_file) {
2215 zero_user_segment(page, 0, PAGE_SIZE);
2216 if (!PageUptodate(page))
2217 SetPageUptodate(page);
2218 } else if (!PageUptodate(page)) {
2219 continue;
2220 }
2221 unlock_page(page);
2222 if (for_write)
2223 put_page(page);
2224 cc->rpages[i] = NULL;
2225 cc->nr_rpages--;
2226 }
2227
2228 /* we are done since all pages are beyond EOF */
2229 if (f2fs_cluster_is_empty(cc))
2230 goto out;
2231
2232 if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei))
2233 from_dnode = false;
2234
2235 if (!from_dnode)
2236 goto skip_reading_dnode;
2237
2238 set_new_dnode(&dn, inode, NULL, NULL, 0);
2239 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2240 if (ret)
2241 goto out;
2242
2243 if (unlikely(f2fs_cp_error(sbi))) {
2244 ret = -EIO;
2245 goto out_put_dnode;
2246 }
2247 f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2248
2249 skip_reading_dnode:
2250 for (i = 1; i < cc->cluster_size; i++) {
2251 block_t blkaddr;
2252
2253 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2254 dn.ofs_in_node + i) :
2255 ei.blk + i - 1;
2256
2257 if (!__is_valid_data_blkaddr(blkaddr))
2258 break;
2259
2260 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2261 ret = -EFAULT;
2262 goto out_put_dnode;
2263 }
2264 cc->nr_cpages++;
2265
2266 if (!from_dnode && i >= ei.c_len)
2267 break;
2268 }
2269
2270 /* nothing to decompress */
2271 if (cc->nr_cpages == 0) {
2272 ret = 0;
2273 goto out_put_dnode;
2274 }
2275
2276 dic = f2fs_alloc_dic(cc);
2277 if (IS_ERR(dic)) {
2278 ret = PTR_ERR(dic);
2279 goto out_put_dnode;
2280 }
2281
2282 for (i = 0; i < cc->nr_cpages; i++) {
2283 struct page *page = dic->cpages[i];
2284 block_t blkaddr;
2285 struct bio_post_read_ctx *ctx;
2286
2287 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2288 dn.ofs_in_node + i + 1) :
2289 ei.blk + i;
2290
2291 f2fs_wait_on_block_writeback(inode, blkaddr);
2292
2293 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2294 if (atomic_dec_and_test(&dic->remaining_pages)) {
2295 f2fs_decompress_cluster(dic, true);
2296 break;
2297 }
2298 continue;
2299 }
2300
2301 if (bio && (!page_is_mergeable(sbi, bio,
2302 *last_block_in_bio, blkaddr) ||
2303 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2304 submit_and_realloc:
2305 f2fs_submit_read_bio(sbi, bio, DATA);
2306 bio = NULL;
2307 }
2308
2309 if (!bio) {
2310 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2311 f2fs_ra_op_flags(rac),
2312 page->index, for_write);
2313 if (IS_ERR(bio)) {
2314 ret = PTR_ERR(bio);
2315 f2fs_decompress_end_io(dic, ret, true);
2316 f2fs_put_dnode(&dn);
2317 *bio_ret = NULL;
2318 return ret;
2319 }
2320 }
2321
2322 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2323 goto submit_and_realloc;
2324
2325 ctx = get_post_read_ctx(bio);
2326 ctx->enabled_steps |= STEP_DECOMPRESS;
2327 refcount_inc(&dic->refcnt);
2328
2329 inc_page_count(sbi, F2FS_RD_DATA);
2330 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
2331 *last_block_in_bio = blkaddr;
2332 }
2333
2334 if (from_dnode)
2335 f2fs_put_dnode(&dn);
2336
2337 *bio_ret = bio;
2338 return 0;
2339
2340 out_put_dnode:
2341 if (from_dnode)
2342 f2fs_put_dnode(&dn);
2343 out:
2344 for (i = 0; i < cc->cluster_size; i++) {
2345 if (cc->rpages[i]) {
2346 ClearPageUptodate(cc->rpages[i]);
2347 unlock_page(cc->rpages[i]);
2348 }
2349 }
2350 *bio_ret = bio;
2351 return ret;
2352 }
2353 #endif
2354
2355 /*
2356 * This function was originally taken from fs/mpage.c, and customized for f2fs.
2357 * Major change was from block_size == page_size in f2fs by default.
2358 */
f2fs_mpage_readpages(struct inode * inode,struct readahead_control * rac,struct folio * folio)2359 static int f2fs_mpage_readpages(struct inode *inode,
2360 struct readahead_control *rac, struct folio *folio)
2361 {
2362 struct bio *bio = NULL;
2363 sector_t last_block_in_bio = 0;
2364 struct f2fs_map_blocks map;
2365 #ifdef CONFIG_F2FS_FS_COMPRESSION
2366 struct compress_ctx cc = {
2367 .inode = inode,
2368 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2369 .cluster_size = F2FS_I(inode)->i_cluster_size,
2370 .cluster_idx = NULL_CLUSTER,
2371 .rpages = NULL,
2372 .cpages = NULL,
2373 .nr_rpages = 0,
2374 .nr_cpages = 0,
2375 };
2376 pgoff_t nc_cluster_idx = NULL_CLUSTER;
2377 #endif
2378 unsigned nr_pages = rac ? readahead_count(rac) : 1;
2379 unsigned max_nr_pages = nr_pages;
2380 pgoff_t index;
2381 int ret = 0;
2382
2383 map.m_pblk = 0;
2384 map.m_lblk = 0;
2385 map.m_len = 0;
2386 map.m_flags = 0;
2387 map.m_next_pgofs = NULL;
2388 map.m_next_extent = NULL;
2389 map.m_seg_type = NO_CHECK_TYPE;
2390 map.m_may_create = false;
2391
2392 for (; nr_pages; nr_pages--) {
2393 if (rac) {
2394 folio = readahead_folio(rac);
2395 prefetchw(&folio->flags);
2396 }
2397
2398 index = folio_index(folio);
2399
2400 #ifdef CONFIG_F2FS_FS_COMPRESSION
2401 if (!f2fs_compressed_file(inode))
2402 goto read_single_page;
2403
2404 /* there are remained compressed pages, submit them */
2405 if (!f2fs_cluster_can_merge_page(&cc, index)) {
2406 ret = f2fs_read_multi_pages(&cc, &bio,
2407 max_nr_pages,
2408 &last_block_in_bio,
2409 rac, false);
2410 f2fs_destroy_compress_ctx(&cc, false);
2411 if (ret)
2412 goto set_error_page;
2413 }
2414 if (cc.cluster_idx == NULL_CLUSTER) {
2415 if (nc_cluster_idx == index >> cc.log_cluster_size)
2416 goto read_single_page;
2417
2418 ret = f2fs_is_compressed_cluster(inode, index);
2419 if (ret < 0)
2420 goto set_error_page;
2421 else if (!ret) {
2422 nc_cluster_idx =
2423 index >> cc.log_cluster_size;
2424 goto read_single_page;
2425 }
2426
2427 nc_cluster_idx = NULL_CLUSTER;
2428 }
2429 ret = f2fs_init_compress_ctx(&cc);
2430 if (ret)
2431 goto set_error_page;
2432
2433 f2fs_compress_ctx_add_page(&cc, &folio->page);
2434
2435 goto next_page;
2436 read_single_page:
2437 #endif
2438
2439 ret = f2fs_read_single_page(inode, folio, max_nr_pages, &map,
2440 &bio, &last_block_in_bio, rac);
2441 if (ret) {
2442 #ifdef CONFIG_F2FS_FS_COMPRESSION
2443 set_error_page:
2444 #endif
2445 folio_zero_segment(folio, 0, folio_size(folio));
2446 folio_unlock(folio);
2447 }
2448 #ifdef CONFIG_F2FS_FS_COMPRESSION
2449 next_page:
2450 #endif
2451
2452 #ifdef CONFIG_F2FS_FS_COMPRESSION
2453 if (f2fs_compressed_file(inode)) {
2454 /* last page */
2455 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2456 ret = f2fs_read_multi_pages(&cc, &bio,
2457 max_nr_pages,
2458 &last_block_in_bio,
2459 rac, false);
2460 f2fs_destroy_compress_ctx(&cc, false);
2461 }
2462 }
2463 #endif
2464 }
2465 if (bio)
2466 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2467 return ret;
2468 }
2469
f2fs_read_data_folio(struct file * file,struct folio * folio)2470 static int f2fs_read_data_folio(struct file *file, struct folio *folio)
2471 {
2472 struct inode *inode = folio_file_mapping(folio)->host;
2473 int ret = -EAGAIN;
2474
2475 trace_f2fs_readpage(folio, DATA);
2476
2477 if (!f2fs_is_compress_backend_ready(inode)) {
2478 folio_unlock(folio);
2479 return -EOPNOTSUPP;
2480 }
2481
2482 /* If the file has inline data, try to read it directly */
2483 if (f2fs_has_inline_data(inode))
2484 ret = f2fs_read_inline_data(inode, folio);
2485 if (ret == -EAGAIN)
2486 ret = f2fs_mpage_readpages(inode, NULL, folio);
2487 return ret;
2488 }
2489
f2fs_readahead(struct readahead_control * rac)2490 static void f2fs_readahead(struct readahead_control *rac)
2491 {
2492 struct inode *inode = rac->mapping->host;
2493
2494 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2495
2496 if (!f2fs_is_compress_backend_ready(inode))
2497 return;
2498
2499 /* If the file has inline data, skip readahead */
2500 if (f2fs_has_inline_data(inode))
2501 return;
2502
2503 f2fs_mpage_readpages(inode, rac, NULL);
2504 }
2505
f2fs_encrypt_one_page(struct f2fs_io_info * fio)2506 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2507 {
2508 struct inode *inode = fio->page->mapping->host;
2509 struct page *mpage, *page;
2510 gfp_t gfp_flags = GFP_NOFS;
2511
2512 if (!f2fs_encrypted_file(inode))
2513 return 0;
2514
2515 page = fio->compressed_page ? fio->compressed_page : fio->page;
2516
2517 if (fscrypt_inode_uses_inline_crypto(inode))
2518 return 0;
2519
2520 retry_encrypt:
2521 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2522 PAGE_SIZE, 0, gfp_flags);
2523 if (IS_ERR(fio->encrypted_page)) {
2524 /* flush pending IOs and wait for a while in the ENOMEM case */
2525 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2526 f2fs_flush_merged_writes(fio->sbi);
2527 memalloc_retry_wait(GFP_NOFS);
2528 gfp_flags |= __GFP_NOFAIL;
2529 goto retry_encrypt;
2530 }
2531 return PTR_ERR(fio->encrypted_page);
2532 }
2533
2534 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2535 if (mpage) {
2536 if (PageUptodate(mpage))
2537 memcpy(page_address(mpage),
2538 page_address(fio->encrypted_page), PAGE_SIZE);
2539 f2fs_put_page(mpage, 1);
2540 }
2541 return 0;
2542 }
2543
check_inplace_update_policy(struct inode * inode,struct f2fs_io_info * fio)2544 static inline bool check_inplace_update_policy(struct inode *inode,
2545 struct f2fs_io_info *fio)
2546 {
2547 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2548
2549 if (IS_F2FS_IPU_HONOR_OPU_WRITE(sbi) &&
2550 is_inode_flag_set(inode, FI_OPU_WRITE))
2551 return false;
2552 if (IS_F2FS_IPU_FORCE(sbi))
2553 return true;
2554 if (IS_F2FS_IPU_SSR(sbi) && f2fs_need_SSR(sbi))
2555 return true;
2556 if (IS_F2FS_IPU_UTIL(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util)
2557 return true;
2558 if (IS_F2FS_IPU_SSR_UTIL(sbi) && f2fs_need_SSR(sbi) &&
2559 utilization(sbi) > SM_I(sbi)->min_ipu_util)
2560 return true;
2561
2562 /*
2563 * IPU for rewrite async pages
2564 */
2565 if (IS_F2FS_IPU_ASYNC(sbi) && fio && fio->op == REQ_OP_WRITE &&
2566 !(fio->op_flags & REQ_SYNC) && !IS_ENCRYPTED(inode))
2567 return true;
2568
2569 /* this is only set during fdatasync */
2570 if (IS_F2FS_IPU_FSYNC(sbi) && is_inode_flag_set(inode, FI_NEED_IPU))
2571 return true;
2572
2573 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2574 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2575 return true;
2576
2577 return false;
2578 }
2579
f2fs_should_update_inplace(struct inode * inode,struct f2fs_io_info * fio)2580 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2581 {
2582 /* swap file is migrating in aligned write mode */
2583 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2584 return false;
2585
2586 if (f2fs_is_pinned_file(inode))
2587 return true;
2588
2589 /* if this is cold file, we should overwrite to avoid fragmentation */
2590 if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE))
2591 return true;
2592
2593 return check_inplace_update_policy(inode, fio);
2594 }
2595
f2fs_should_update_outplace(struct inode * inode,struct f2fs_io_info * fio)2596 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2597 {
2598 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2599
2600 /* The below cases were checked when setting it. */
2601 if (f2fs_is_pinned_file(inode))
2602 return false;
2603 if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2604 return true;
2605 if (f2fs_lfs_mode(sbi))
2606 return true;
2607 if (S_ISDIR(inode->i_mode))
2608 return true;
2609 if (IS_NOQUOTA(inode))
2610 return true;
2611 if (f2fs_used_in_atomic_write(inode))
2612 return true;
2613 /* rewrite low ratio compress data w/ OPU mode to avoid fragmentation */
2614 if (f2fs_compressed_file(inode) &&
2615 F2FS_OPTION(sbi).compress_mode == COMPR_MODE_USER &&
2616 is_inode_flag_set(inode, FI_ENABLE_COMPRESS))
2617 return true;
2618
2619 /* swap file is migrating in aligned write mode */
2620 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2621 return true;
2622
2623 if (is_inode_flag_set(inode, FI_OPU_WRITE))
2624 return true;
2625
2626 if (fio) {
2627 if (page_private_gcing(fio->page))
2628 return true;
2629 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2630 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2631 return true;
2632 }
2633 return false;
2634 }
2635
need_inplace_update(struct f2fs_io_info * fio)2636 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2637 {
2638 struct inode *inode = fio->page->mapping->host;
2639
2640 if (f2fs_should_update_outplace(inode, fio))
2641 return false;
2642
2643 return f2fs_should_update_inplace(inode, fio);
2644 }
2645
f2fs_do_write_data_page(struct f2fs_io_info * fio)2646 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2647 {
2648 struct page *page = fio->page;
2649 struct inode *inode = page->mapping->host;
2650 struct dnode_of_data dn;
2651 struct node_info ni;
2652 bool ipu_force = false;
2653 int err = 0;
2654
2655 /* Use COW inode to make dnode_of_data for atomic write */
2656 if (f2fs_is_atomic_file(inode))
2657 set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
2658 else
2659 set_new_dnode(&dn, inode, NULL, NULL, 0);
2660
2661 if (need_inplace_update(fio) &&
2662 f2fs_lookup_read_extent_cache_block(inode, page->index,
2663 &fio->old_blkaddr)) {
2664 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2665 DATA_GENERIC_ENHANCE))
2666 return -EFSCORRUPTED;
2667
2668 ipu_force = true;
2669 fio->need_lock = LOCK_DONE;
2670 goto got_it;
2671 }
2672
2673 /* Deadlock due to between page->lock and f2fs_lock_op */
2674 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2675 return -EAGAIN;
2676
2677 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2678 if (err)
2679 goto out;
2680
2681 fio->old_blkaddr = dn.data_blkaddr;
2682
2683 /* This page is already truncated */
2684 if (fio->old_blkaddr == NULL_ADDR) {
2685 ClearPageUptodate(page);
2686 clear_page_private_gcing(page);
2687 goto out_writepage;
2688 }
2689 got_it:
2690 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2691 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2692 DATA_GENERIC_ENHANCE)) {
2693 err = -EFSCORRUPTED;
2694 goto out_writepage;
2695 }
2696
2697 /* wait for GCed page writeback via META_MAPPING */
2698 if (fio->meta_gc)
2699 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2700
2701 /*
2702 * If current allocation needs SSR,
2703 * it had better in-place writes for updated data.
2704 */
2705 if (ipu_force ||
2706 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2707 need_inplace_update(fio))) {
2708 err = f2fs_encrypt_one_page(fio);
2709 if (err)
2710 goto out_writepage;
2711
2712 set_page_writeback(page);
2713 f2fs_put_dnode(&dn);
2714 if (fio->need_lock == LOCK_REQ)
2715 f2fs_unlock_op(fio->sbi);
2716 err = f2fs_inplace_write_data(fio);
2717 if (err) {
2718 if (fscrypt_inode_uses_fs_layer_crypto(inode))
2719 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2720 end_page_writeback(page);
2721 } else {
2722 set_inode_flag(inode, FI_UPDATE_WRITE);
2723 }
2724 trace_f2fs_do_write_data_page(page_folio(page), IPU);
2725 return err;
2726 }
2727
2728 if (fio->need_lock == LOCK_RETRY) {
2729 if (!f2fs_trylock_op(fio->sbi)) {
2730 err = -EAGAIN;
2731 goto out_writepage;
2732 }
2733 fio->need_lock = LOCK_REQ;
2734 }
2735
2736 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2737 if (err)
2738 goto out_writepage;
2739
2740 fio->version = ni.version;
2741
2742 err = f2fs_encrypt_one_page(fio);
2743 if (err)
2744 goto out_writepage;
2745
2746 set_page_writeback(page);
2747
2748 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2749 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2750
2751 /* LFS mode write path */
2752 f2fs_outplace_write_data(&dn, fio);
2753 trace_f2fs_do_write_data_page(page_folio(page), OPU);
2754 set_inode_flag(inode, FI_APPEND_WRITE);
2755 out_writepage:
2756 f2fs_put_dnode(&dn);
2757 out:
2758 if (fio->need_lock == LOCK_REQ)
2759 f2fs_unlock_op(fio->sbi);
2760 return err;
2761 }
2762
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)2763 int f2fs_write_single_data_page(struct page *page, int *submitted,
2764 struct bio **bio,
2765 sector_t *last_block,
2766 struct writeback_control *wbc,
2767 enum iostat_type io_type,
2768 int compr_blocks,
2769 bool allow_balance)
2770 {
2771 struct inode *inode = page->mapping->host;
2772 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2773 loff_t i_size = i_size_read(inode);
2774 const pgoff_t end_index = ((unsigned long long)i_size)
2775 >> PAGE_SHIFT;
2776 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2777 unsigned offset = 0;
2778 bool need_balance_fs = false;
2779 bool quota_inode = IS_NOQUOTA(inode);
2780 int err = 0;
2781 struct f2fs_io_info fio = {
2782 .sbi = sbi,
2783 .ino = inode->i_ino,
2784 .type = DATA,
2785 .op = REQ_OP_WRITE,
2786 .op_flags = wbc_to_write_flags(wbc),
2787 .old_blkaddr = NULL_ADDR,
2788 .page = page,
2789 .encrypted_page = NULL,
2790 .submitted = 0,
2791 .compr_blocks = compr_blocks,
2792 .need_lock = compr_blocks ? LOCK_DONE : LOCK_RETRY,
2793 .meta_gc = f2fs_meta_inode_gc_required(inode) ? 1 : 0,
2794 .io_type = io_type,
2795 .io_wbc = wbc,
2796 .bio = bio,
2797 .last_block = last_block,
2798 };
2799
2800 trace_f2fs_writepage(page_folio(page), DATA);
2801
2802 /* we should bypass data pages to proceed the kworker jobs */
2803 if (unlikely(f2fs_cp_error(sbi))) {
2804 mapping_set_error(page->mapping, -EIO);
2805 /*
2806 * don't drop any dirty dentry pages for keeping lastest
2807 * directory structure.
2808 */
2809 if (S_ISDIR(inode->i_mode) &&
2810 !is_sbi_flag_set(sbi, SBI_IS_CLOSE))
2811 goto redirty_out;
2812
2813 /* keep data pages in remount-ro mode */
2814 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
2815 goto redirty_out;
2816 goto out;
2817 }
2818
2819 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2820 goto redirty_out;
2821
2822 if (page->index < end_index ||
2823 f2fs_verity_in_progress(inode) ||
2824 compr_blocks)
2825 goto write;
2826
2827 /*
2828 * If the offset is out-of-range of file size,
2829 * this page does not have to be written to disk.
2830 */
2831 offset = i_size & (PAGE_SIZE - 1);
2832 if ((page->index >= end_index + 1) || !offset)
2833 goto out;
2834
2835 zero_user_segment(page, offset, PAGE_SIZE);
2836 write:
2837 /* Dentry/quota blocks are controlled by checkpoint */
2838 if (S_ISDIR(inode->i_mode) || quota_inode) {
2839 /*
2840 * We need to wait for node_write to avoid block allocation during
2841 * checkpoint. This can only happen to quota writes which can cause
2842 * the below discard race condition.
2843 */
2844 if (quota_inode)
2845 f2fs_down_read(&sbi->node_write);
2846
2847 fio.need_lock = LOCK_DONE;
2848 err = f2fs_do_write_data_page(&fio);
2849
2850 if (quota_inode)
2851 f2fs_up_read(&sbi->node_write);
2852
2853 goto done;
2854 }
2855
2856 if (!wbc->for_reclaim)
2857 need_balance_fs = true;
2858 else if (has_not_enough_free_secs(sbi, 0, 0))
2859 goto redirty_out;
2860 else
2861 set_inode_flag(inode, FI_HOT_DATA);
2862
2863 err = -EAGAIN;
2864 if (f2fs_has_inline_data(inode)) {
2865 err = f2fs_write_inline_data(inode, page);
2866 if (!err)
2867 goto out;
2868 }
2869
2870 if (err == -EAGAIN) {
2871 err = f2fs_do_write_data_page(&fio);
2872 if (err == -EAGAIN) {
2873 f2fs_bug_on(sbi, compr_blocks);
2874 fio.need_lock = LOCK_REQ;
2875 err = f2fs_do_write_data_page(&fio);
2876 }
2877 }
2878
2879 if (err) {
2880 file_set_keep_isize(inode);
2881 } else {
2882 spin_lock(&F2FS_I(inode)->i_size_lock);
2883 if (F2FS_I(inode)->last_disk_size < psize)
2884 F2FS_I(inode)->last_disk_size = psize;
2885 spin_unlock(&F2FS_I(inode)->i_size_lock);
2886 }
2887
2888 done:
2889 if (err && err != -ENOENT)
2890 goto redirty_out;
2891
2892 out:
2893 inode_dec_dirty_pages(inode);
2894 if (err) {
2895 ClearPageUptodate(page);
2896 clear_page_private_gcing(page);
2897 }
2898
2899 if (wbc->for_reclaim) {
2900 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2901 clear_inode_flag(inode, FI_HOT_DATA);
2902 f2fs_remove_dirty_inode(inode);
2903 submitted = NULL;
2904 }
2905 unlock_page(page);
2906 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2907 !F2FS_I(inode)->wb_task && allow_balance)
2908 f2fs_balance_fs(sbi, need_balance_fs);
2909
2910 if (unlikely(f2fs_cp_error(sbi))) {
2911 f2fs_submit_merged_write(sbi, DATA);
2912 if (bio && *bio)
2913 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2914 submitted = NULL;
2915 }
2916
2917 if (submitted)
2918 *submitted = fio.submitted;
2919
2920 return 0;
2921
2922 redirty_out:
2923 redirty_page_for_writepage(wbc, page);
2924 /*
2925 * pageout() in MM translates EAGAIN, so calls handle_write_error()
2926 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2927 * file_write_and_wait_range() will see EIO error, which is critical
2928 * to return value of fsync() followed by atomic_write failure to user.
2929 */
2930 if (!err || wbc->for_reclaim)
2931 return AOP_WRITEPAGE_ACTIVATE;
2932 unlock_page(page);
2933 return err;
2934 }
2935
f2fs_write_data_page(struct page * page,struct writeback_control * wbc)2936 static int f2fs_write_data_page(struct page *page,
2937 struct writeback_control *wbc)
2938 {
2939 #ifdef CONFIG_F2FS_FS_COMPRESSION
2940 struct inode *inode = page->mapping->host;
2941
2942 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2943 goto out;
2944
2945 if (f2fs_compressed_file(inode)) {
2946 if (f2fs_is_compressed_cluster(inode, page->index)) {
2947 redirty_page_for_writepage(wbc, page);
2948 return AOP_WRITEPAGE_ACTIVATE;
2949 }
2950 }
2951 out:
2952 #endif
2953
2954 return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2955 wbc, FS_DATA_IO, 0, true);
2956 }
2957
2958 /*
2959 * This function was copied from write_cache_pages from mm/page-writeback.c.
2960 * The major change is making write step of cold data page separately from
2961 * warm/hot data page.
2962 */
f2fs_write_cache_pages(struct address_space * mapping,struct writeback_control * wbc,enum iostat_type io_type)2963 static int f2fs_write_cache_pages(struct address_space *mapping,
2964 struct writeback_control *wbc,
2965 enum iostat_type io_type)
2966 {
2967 int ret = 0;
2968 int done = 0, retry = 0;
2969 struct page *pages_local[F2FS_ONSTACK_PAGES];
2970 struct page **pages = pages_local;
2971 struct folio_batch fbatch;
2972 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2973 struct bio *bio = NULL;
2974 sector_t last_block;
2975 #ifdef CONFIG_F2FS_FS_COMPRESSION
2976 struct inode *inode = mapping->host;
2977 struct compress_ctx cc = {
2978 .inode = inode,
2979 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2980 .cluster_size = F2FS_I(inode)->i_cluster_size,
2981 .cluster_idx = NULL_CLUSTER,
2982 .rpages = NULL,
2983 .nr_rpages = 0,
2984 .cpages = NULL,
2985 .valid_nr_cpages = 0,
2986 .rbuf = NULL,
2987 .cbuf = NULL,
2988 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2989 .private = NULL,
2990 };
2991 #endif
2992 int nr_folios, p, idx;
2993 int nr_pages;
2994 unsigned int max_pages = F2FS_ONSTACK_PAGES;
2995 pgoff_t index;
2996 pgoff_t end; /* Inclusive */
2997 pgoff_t done_index;
2998 int range_whole = 0;
2999 xa_mark_t tag;
3000 int nwritten = 0;
3001 int submitted = 0;
3002 int i;
3003
3004 #ifdef CONFIG_F2FS_FS_COMPRESSION
3005 if (f2fs_compressed_file(inode) &&
3006 1 << cc.log_cluster_size > F2FS_ONSTACK_PAGES) {
3007 pages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
3008 cc.log_cluster_size, GFP_NOFS | __GFP_NOFAIL);
3009 max_pages = 1 << cc.log_cluster_size;
3010 }
3011 #endif
3012
3013 folio_batch_init(&fbatch);
3014
3015 if (get_dirty_pages(mapping->host) <=
3016 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
3017 set_inode_flag(mapping->host, FI_HOT_DATA);
3018 else
3019 clear_inode_flag(mapping->host, FI_HOT_DATA);
3020
3021 if (wbc->range_cyclic) {
3022 index = mapping->writeback_index; /* prev offset */
3023 end = -1;
3024 } else {
3025 index = wbc->range_start >> PAGE_SHIFT;
3026 end = wbc->range_end >> PAGE_SHIFT;
3027 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3028 range_whole = 1;
3029 }
3030 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3031 tag = PAGECACHE_TAG_TOWRITE;
3032 else
3033 tag = PAGECACHE_TAG_DIRTY;
3034 retry:
3035 retry = 0;
3036 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3037 tag_pages_for_writeback(mapping, index, end);
3038 done_index = index;
3039 while (!done && !retry && (index <= end)) {
3040 nr_pages = 0;
3041 again:
3042 nr_folios = filemap_get_folios_tag(mapping, &index, end,
3043 tag, &fbatch);
3044 if (nr_folios == 0) {
3045 if (nr_pages)
3046 goto write;
3047 break;
3048 }
3049
3050 for (i = 0; i < nr_folios; i++) {
3051 struct folio *folio = fbatch.folios[i];
3052
3053 idx = 0;
3054 p = folio_nr_pages(folio);
3055 add_more:
3056 pages[nr_pages] = folio_page(folio, idx);
3057 folio_get(folio);
3058 if (++nr_pages == max_pages) {
3059 index = folio->index + idx + 1;
3060 folio_batch_release(&fbatch);
3061 goto write;
3062 }
3063 if (++idx < p)
3064 goto add_more;
3065 }
3066 folio_batch_release(&fbatch);
3067 goto again;
3068 write:
3069 for (i = 0; i < nr_pages; i++) {
3070 struct page *page = pages[i];
3071 struct folio *folio = page_folio(page);
3072 bool need_readd;
3073 readd:
3074 need_readd = false;
3075 #ifdef CONFIG_F2FS_FS_COMPRESSION
3076 if (f2fs_compressed_file(inode)) {
3077 void *fsdata = NULL;
3078 struct page *pagep;
3079 int ret2;
3080
3081 ret = f2fs_init_compress_ctx(&cc);
3082 if (ret) {
3083 done = 1;
3084 break;
3085 }
3086
3087 if (!f2fs_cluster_can_merge_page(&cc,
3088 folio->index)) {
3089 ret = f2fs_write_multi_pages(&cc,
3090 &submitted, wbc, io_type);
3091 if (!ret)
3092 need_readd = true;
3093 goto result;
3094 }
3095
3096 if (unlikely(f2fs_cp_error(sbi)))
3097 goto lock_folio;
3098
3099 if (!f2fs_cluster_is_empty(&cc))
3100 goto lock_folio;
3101
3102 if (f2fs_all_cluster_page_ready(&cc,
3103 pages, i, nr_pages, true))
3104 goto lock_folio;
3105
3106 ret2 = f2fs_prepare_compress_overwrite(
3107 inode, &pagep,
3108 folio->index, &fsdata);
3109 if (ret2 < 0) {
3110 ret = ret2;
3111 done = 1;
3112 break;
3113 } else if (ret2 &&
3114 (!f2fs_compress_write_end(inode,
3115 fsdata, folio->index, 1) ||
3116 !f2fs_all_cluster_page_ready(&cc,
3117 pages, i, nr_pages,
3118 false))) {
3119 retry = 1;
3120 break;
3121 }
3122 }
3123 #endif
3124 /* give a priority to WB_SYNC threads */
3125 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3126 wbc->sync_mode == WB_SYNC_NONE) {
3127 done = 1;
3128 break;
3129 }
3130 #ifdef CONFIG_F2FS_FS_COMPRESSION
3131 lock_folio:
3132 #endif
3133 done_index = folio->index;
3134 retry_write:
3135 folio_lock(folio);
3136
3137 if (unlikely(folio->mapping != mapping)) {
3138 continue_unlock:
3139 folio_unlock(folio);
3140 continue;
3141 }
3142
3143 if (!folio_test_dirty(folio)) {
3144 /* someone wrote it for us */
3145 goto continue_unlock;
3146 }
3147
3148 if (folio_test_writeback(folio)) {
3149 if (wbc->sync_mode == WB_SYNC_NONE)
3150 goto continue_unlock;
3151 f2fs_wait_on_page_writeback(&folio->page, DATA, true, true);
3152 }
3153
3154 if (!folio_clear_dirty_for_io(folio))
3155 goto continue_unlock;
3156
3157 #ifdef CONFIG_F2FS_FS_COMPRESSION
3158 if (f2fs_compressed_file(inode)) {
3159 folio_get(folio);
3160 f2fs_compress_ctx_add_page(&cc, &folio->page);
3161 continue;
3162 }
3163 #endif
3164 ret = f2fs_write_single_data_page(&folio->page,
3165 &submitted, &bio, &last_block,
3166 wbc, io_type, 0, true);
3167 if (ret == AOP_WRITEPAGE_ACTIVATE)
3168 folio_unlock(folio);
3169 #ifdef CONFIG_F2FS_FS_COMPRESSION
3170 result:
3171 #endif
3172 nwritten += submitted;
3173 wbc->nr_to_write -= submitted;
3174
3175 if (unlikely(ret)) {
3176 /*
3177 * keep nr_to_write, since vfs uses this to
3178 * get # of written pages.
3179 */
3180 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3181 ret = 0;
3182 goto next;
3183 } else if (ret == -EAGAIN) {
3184 ret = 0;
3185 if (wbc->sync_mode == WB_SYNC_ALL) {
3186 f2fs_io_schedule_timeout(
3187 DEFAULT_IO_TIMEOUT);
3188 goto retry_write;
3189 }
3190 goto next;
3191 }
3192 done_index = folio_next_index(folio);
3193 done = 1;
3194 break;
3195 }
3196
3197 if (wbc->nr_to_write <= 0 &&
3198 wbc->sync_mode == WB_SYNC_NONE) {
3199 done = 1;
3200 break;
3201 }
3202 next:
3203 if (need_readd)
3204 goto readd;
3205 }
3206 release_pages(pages, nr_pages);
3207 cond_resched();
3208 }
3209 #ifdef CONFIG_F2FS_FS_COMPRESSION
3210 /* flush remained pages in compress cluster */
3211 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3212 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3213 nwritten += submitted;
3214 wbc->nr_to_write -= submitted;
3215 if (ret) {
3216 done = 1;
3217 retry = 0;
3218 }
3219 }
3220 if (f2fs_compressed_file(inode))
3221 f2fs_destroy_compress_ctx(&cc, false);
3222 #endif
3223 if (retry) {
3224 index = 0;
3225 end = -1;
3226 goto retry;
3227 }
3228 if (wbc->range_cyclic && !done)
3229 done_index = 0;
3230 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3231 mapping->writeback_index = done_index;
3232
3233 if (nwritten)
3234 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3235 NULL, 0, DATA);
3236 /* submit cached bio of IPU write */
3237 if (bio)
3238 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3239
3240 #ifdef CONFIG_F2FS_FS_COMPRESSION
3241 if (pages != pages_local)
3242 kfree(pages);
3243 #endif
3244
3245 return ret;
3246 }
3247
__should_serialize_io(struct inode * inode,struct writeback_control * wbc)3248 static inline bool __should_serialize_io(struct inode *inode,
3249 struct writeback_control *wbc)
3250 {
3251 /* to avoid deadlock in path of data flush */
3252 if (F2FS_I(inode)->wb_task)
3253 return false;
3254
3255 if (!S_ISREG(inode->i_mode))
3256 return false;
3257 if (IS_NOQUOTA(inode))
3258 return false;
3259
3260 if (f2fs_need_compress_data(inode))
3261 return true;
3262 if (wbc->sync_mode != WB_SYNC_ALL)
3263 return true;
3264 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3265 return true;
3266 return false;
3267 }
3268
__f2fs_write_data_pages(struct address_space * mapping,struct writeback_control * wbc,enum iostat_type io_type)3269 static int __f2fs_write_data_pages(struct address_space *mapping,
3270 struct writeback_control *wbc,
3271 enum iostat_type io_type)
3272 {
3273 struct inode *inode = mapping->host;
3274 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3275 struct blk_plug plug;
3276 int ret;
3277 bool locked = false;
3278
3279 /* deal with chardevs and other special file */
3280 if (!mapping->a_ops->writepage)
3281 return 0;
3282
3283 /* skip writing if there is no dirty page in this inode */
3284 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3285 return 0;
3286
3287 /* during POR, we don't need to trigger writepage at all. */
3288 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3289 goto skip_write;
3290
3291 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3292 wbc->sync_mode == WB_SYNC_NONE &&
3293 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3294 f2fs_available_free_memory(sbi, DIRTY_DENTS))
3295 goto skip_write;
3296
3297 /* skip writing in file defragment preparing stage */
3298 if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3299 goto skip_write;
3300
3301 trace_f2fs_writepages(mapping->host, wbc, DATA);
3302
3303 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3304 if (wbc->sync_mode == WB_SYNC_ALL)
3305 atomic_inc(&sbi->wb_sync_req[DATA]);
3306 else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3307 /* to avoid potential deadlock */
3308 if (current->plug)
3309 blk_finish_plug(current->plug);
3310 goto skip_write;
3311 }
3312
3313 if (__should_serialize_io(inode, wbc)) {
3314 mutex_lock(&sbi->writepages);
3315 locked = true;
3316 }
3317
3318 blk_start_plug(&plug);
3319 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3320 blk_finish_plug(&plug);
3321
3322 if (locked)
3323 mutex_unlock(&sbi->writepages);
3324
3325 if (wbc->sync_mode == WB_SYNC_ALL)
3326 atomic_dec(&sbi->wb_sync_req[DATA]);
3327 /*
3328 * if some pages were truncated, we cannot guarantee its mapping->host
3329 * to detect pending bios.
3330 */
3331
3332 f2fs_remove_dirty_inode(inode);
3333 return ret;
3334
3335 skip_write:
3336 wbc->pages_skipped += get_dirty_pages(inode);
3337 trace_f2fs_writepages(mapping->host, wbc, DATA);
3338 return 0;
3339 }
3340
f2fs_write_data_pages(struct address_space * mapping,struct writeback_control * wbc)3341 static int f2fs_write_data_pages(struct address_space *mapping,
3342 struct writeback_control *wbc)
3343 {
3344 struct inode *inode = mapping->host;
3345
3346 return __f2fs_write_data_pages(mapping, wbc,
3347 F2FS_I(inode)->cp_task == current ?
3348 FS_CP_DATA_IO : FS_DATA_IO);
3349 }
3350
f2fs_write_failed(struct inode * inode,loff_t to)3351 void f2fs_write_failed(struct inode *inode, loff_t to)
3352 {
3353 loff_t i_size = i_size_read(inode);
3354
3355 if (IS_NOQUOTA(inode))
3356 return;
3357
3358 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3359 if (to > i_size && !f2fs_verity_in_progress(inode)) {
3360 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3361 filemap_invalidate_lock(inode->i_mapping);
3362
3363 truncate_pagecache(inode, i_size);
3364 f2fs_truncate_blocks(inode, i_size, true);
3365
3366 filemap_invalidate_unlock(inode->i_mapping);
3367 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3368 }
3369 }
3370
prepare_write_begin(struct f2fs_sb_info * sbi,struct page * page,loff_t pos,unsigned len,block_t * blk_addr,bool * node_changed)3371 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3372 struct page *page, loff_t pos, unsigned len,
3373 block_t *blk_addr, bool *node_changed)
3374 {
3375 struct inode *inode = page->mapping->host;
3376 pgoff_t index = page->index;
3377 struct dnode_of_data dn;
3378 struct page *ipage;
3379 bool locked = false;
3380 int flag = F2FS_GET_BLOCK_PRE_AIO;
3381 int err = 0;
3382
3383 /*
3384 * If a whole page is being written and we already preallocated all the
3385 * blocks, then there is no need to get a block address now.
3386 */
3387 if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3388 return 0;
3389
3390 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3391 if (f2fs_has_inline_data(inode)) {
3392 if (pos + len > MAX_INLINE_DATA(inode))
3393 flag = F2FS_GET_BLOCK_DEFAULT;
3394 f2fs_map_lock(sbi, flag);
3395 locked = true;
3396 } else if ((pos & PAGE_MASK) >= i_size_read(inode)) {
3397 f2fs_map_lock(sbi, flag);
3398 locked = true;
3399 }
3400
3401 restart:
3402 /* check inline_data */
3403 ipage = f2fs_get_node_page(sbi, inode->i_ino);
3404 if (IS_ERR(ipage)) {
3405 err = PTR_ERR(ipage);
3406 goto unlock_out;
3407 }
3408
3409 set_new_dnode(&dn, inode, ipage, ipage, 0);
3410
3411 if (f2fs_has_inline_data(inode)) {
3412 if (pos + len <= MAX_INLINE_DATA(inode)) {
3413 f2fs_do_read_inline_data(page_folio(page), ipage);
3414 set_inode_flag(inode, FI_DATA_EXIST);
3415 if (inode->i_nlink)
3416 set_page_private_inline(ipage);
3417 goto out;
3418 }
3419 err = f2fs_convert_inline_page(&dn, page);
3420 if (err || dn.data_blkaddr != NULL_ADDR)
3421 goto out;
3422 }
3423
3424 if (!f2fs_lookup_read_extent_cache_block(inode, index,
3425 &dn.data_blkaddr)) {
3426 if (locked) {
3427 err = f2fs_reserve_block(&dn, index);
3428 goto out;
3429 }
3430
3431 /* hole case */
3432 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3433 if (!err && dn.data_blkaddr != NULL_ADDR)
3434 goto out;
3435 f2fs_put_dnode(&dn);
3436 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3437 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3438 locked = true;
3439 goto restart;
3440 }
3441 out:
3442 if (!err) {
3443 /* convert_inline_page can make node_changed */
3444 *blk_addr = dn.data_blkaddr;
3445 *node_changed = dn.node_changed;
3446 }
3447 f2fs_put_dnode(&dn);
3448 unlock_out:
3449 if (locked)
3450 f2fs_map_unlock(sbi, flag);
3451 return err;
3452 }
3453
__find_data_block(struct inode * inode,pgoff_t index,block_t * blk_addr)3454 static int __find_data_block(struct inode *inode, pgoff_t index,
3455 block_t *blk_addr)
3456 {
3457 struct dnode_of_data dn;
3458 struct page *ipage;
3459 int err = 0;
3460
3461 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
3462 if (IS_ERR(ipage))
3463 return PTR_ERR(ipage);
3464
3465 set_new_dnode(&dn, inode, ipage, ipage, 0);
3466
3467 if (!f2fs_lookup_read_extent_cache_block(inode, index,
3468 &dn.data_blkaddr)) {
3469 /* hole case */
3470 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3471 if (err) {
3472 dn.data_blkaddr = NULL_ADDR;
3473 err = 0;
3474 }
3475 }
3476 *blk_addr = dn.data_blkaddr;
3477 f2fs_put_dnode(&dn);
3478 return err;
3479 }
3480
__reserve_data_block(struct inode * inode,pgoff_t index,block_t * blk_addr,bool * node_changed)3481 static int __reserve_data_block(struct inode *inode, pgoff_t index,
3482 block_t *blk_addr, bool *node_changed)
3483 {
3484 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3485 struct dnode_of_data dn;
3486 struct page *ipage;
3487 int err = 0;
3488
3489 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3490
3491 ipage = f2fs_get_node_page(sbi, inode->i_ino);
3492 if (IS_ERR(ipage)) {
3493 err = PTR_ERR(ipage);
3494 goto unlock_out;
3495 }
3496 set_new_dnode(&dn, inode, ipage, ipage, 0);
3497
3498 if (!f2fs_lookup_read_extent_cache_block(dn.inode, index,
3499 &dn.data_blkaddr))
3500 err = f2fs_reserve_block(&dn, index);
3501
3502 *blk_addr = dn.data_blkaddr;
3503 *node_changed = dn.node_changed;
3504 f2fs_put_dnode(&dn);
3505
3506 unlock_out:
3507 f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3508 return err;
3509 }
3510
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)3511 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
3512 struct page *page, loff_t pos, unsigned int len,
3513 block_t *blk_addr, bool *node_changed, bool *use_cow)
3514 {
3515 struct inode *inode = page->mapping->host;
3516 struct inode *cow_inode = F2FS_I(inode)->cow_inode;
3517 pgoff_t index = page->index;
3518 int err = 0;
3519 block_t ori_blk_addr = NULL_ADDR;
3520
3521 /* If pos is beyond the end of file, reserve a new block in COW inode */
3522 if ((pos & PAGE_MASK) >= i_size_read(inode))
3523 goto reserve_block;
3524
3525 /* Look for the block in COW inode first */
3526 err = __find_data_block(cow_inode, index, blk_addr);
3527 if (err) {
3528 return err;
3529 } else if (*blk_addr != NULL_ADDR) {
3530 *use_cow = true;
3531 return 0;
3532 }
3533
3534 if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE))
3535 goto reserve_block;
3536
3537 /* Look for the block in the original inode */
3538 err = __find_data_block(inode, index, &ori_blk_addr);
3539 if (err)
3540 return err;
3541
3542 reserve_block:
3543 /* Finally, we should reserve a new block in COW inode for the update */
3544 err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
3545 if (err)
3546 return err;
3547 inc_atomic_write_cnt(inode);
3548
3549 if (ori_blk_addr != NULL_ADDR)
3550 *blk_addr = ori_blk_addr;
3551 return 0;
3552 }
3553
f2fs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct folio ** foliop,void ** fsdata)3554 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3555 loff_t pos, unsigned len, struct folio **foliop, void **fsdata)
3556 {
3557 struct inode *inode = mapping->host;
3558 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3559 struct folio *folio;
3560 pgoff_t index = pos >> PAGE_SHIFT;
3561 bool need_balance = false;
3562 bool use_cow = false;
3563 block_t blkaddr = NULL_ADDR;
3564 int err = 0;
3565
3566 trace_f2fs_write_begin(inode, pos, len);
3567
3568 if (!f2fs_is_checkpoint_ready(sbi)) {
3569 err = -ENOSPC;
3570 goto fail;
3571 }
3572
3573 /*
3574 * We should check this at this moment to avoid deadlock on inode page
3575 * and #0 page. The locking rule for inline_data conversion should be:
3576 * folio_lock(folio #0) -> folio_lock(inode_page)
3577 */
3578 if (index != 0) {
3579 err = f2fs_convert_inline_inode(inode);
3580 if (err)
3581 goto fail;
3582 }
3583
3584 #ifdef CONFIG_F2FS_FS_COMPRESSION
3585 if (f2fs_compressed_file(inode)) {
3586 int ret;
3587 struct page *page;
3588
3589 *fsdata = NULL;
3590
3591 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3592 goto repeat;
3593
3594 ret = f2fs_prepare_compress_overwrite(inode, &page,
3595 index, fsdata);
3596 if (ret < 0) {
3597 err = ret;
3598 goto fail;
3599 } else if (ret) {
3600 *foliop = page_folio(page);
3601 return 0;
3602 }
3603 }
3604 #endif
3605
3606 repeat:
3607 /*
3608 * Do not use FGP_STABLE to avoid deadlock.
3609 * Will wait that below with our IO control.
3610 */
3611 folio = __filemap_get_folio(mapping, index,
3612 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3613 if (IS_ERR(folio)) {
3614 err = PTR_ERR(folio);
3615 goto fail;
3616 }
3617
3618 /* TODO: cluster can be compressed due to race with .writepage */
3619
3620 *foliop = folio;
3621
3622 if (f2fs_is_atomic_file(inode))
3623 err = prepare_atomic_write_begin(sbi, &folio->page, pos, len,
3624 &blkaddr, &need_balance, &use_cow);
3625 else
3626 err = prepare_write_begin(sbi, &folio->page, pos, len,
3627 &blkaddr, &need_balance);
3628 if (err)
3629 goto put_folio;
3630
3631 if (need_balance && !IS_NOQUOTA(inode) &&
3632 has_not_enough_free_secs(sbi, 0, 0)) {
3633 folio_unlock(folio);
3634 f2fs_balance_fs(sbi, true);
3635 folio_lock(folio);
3636 if (folio->mapping != mapping) {
3637 /* The folio got truncated from under us */
3638 folio_unlock(folio);
3639 folio_put(folio);
3640 goto repeat;
3641 }
3642 }
3643
3644 f2fs_wait_on_page_writeback(&folio->page, DATA, false, true);
3645
3646 if (len == folio_size(folio) || folio_test_uptodate(folio))
3647 return 0;
3648
3649 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3650 !f2fs_verity_in_progress(inode)) {
3651 folio_zero_segment(folio, len, PAGE_SIZE);
3652 return 0;
3653 }
3654
3655 if (blkaddr == NEW_ADDR) {
3656 folio_zero_segment(folio, 0, folio_size(folio));
3657 folio_mark_uptodate(folio);
3658 } else {
3659 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3660 DATA_GENERIC_ENHANCE_READ)) {
3661 err = -EFSCORRUPTED;
3662 goto put_folio;
3663 }
3664 err = f2fs_submit_page_read(use_cow ?
3665 F2FS_I(inode)->cow_inode : inode, &folio->page,
3666 blkaddr, 0, true);
3667 if (err)
3668 goto put_folio;
3669
3670 folio_lock(folio);
3671 if (unlikely(folio->mapping != mapping)) {
3672 folio_unlock(folio);
3673 folio_put(folio);
3674 goto repeat;
3675 }
3676 if (unlikely(!folio_test_uptodate(folio))) {
3677 err = -EIO;
3678 goto put_folio;
3679 }
3680 }
3681 return 0;
3682
3683 put_folio:
3684 folio_unlock(folio);
3685 folio_put(folio);
3686 fail:
3687 f2fs_write_failed(inode, pos + len);
3688 return err;
3689 }
3690
f2fs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct folio * folio,void * fsdata)3691 static int f2fs_write_end(struct file *file,
3692 struct address_space *mapping,
3693 loff_t pos, unsigned len, unsigned copied,
3694 struct folio *folio, void *fsdata)
3695 {
3696 struct inode *inode = folio->mapping->host;
3697
3698 trace_f2fs_write_end(inode, pos, len, copied);
3699
3700 /*
3701 * This should be come from len == PAGE_SIZE, and we expect copied
3702 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3703 * let generic_perform_write() try to copy data again through copied=0.
3704 */
3705 if (!folio_test_uptodate(folio)) {
3706 if (unlikely(copied != len))
3707 copied = 0;
3708 else
3709 folio_mark_uptodate(folio);
3710 }
3711
3712 #ifdef CONFIG_F2FS_FS_COMPRESSION
3713 /* overwrite compressed file */
3714 if (f2fs_compressed_file(inode) && fsdata) {
3715 f2fs_compress_write_end(inode, fsdata, folio->index, copied);
3716 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3717
3718 if (pos + copied > i_size_read(inode) &&
3719 !f2fs_verity_in_progress(inode))
3720 f2fs_i_size_write(inode, pos + copied);
3721 return copied;
3722 }
3723 #endif
3724
3725 if (!copied)
3726 goto unlock_out;
3727
3728 folio_mark_dirty(folio);
3729
3730 if (pos + copied > i_size_read(inode) &&
3731 !f2fs_verity_in_progress(inode)) {
3732 f2fs_i_size_write(inode, pos + copied);
3733 if (f2fs_is_atomic_file(inode))
3734 f2fs_i_size_write(F2FS_I(inode)->cow_inode,
3735 pos + copied);
3736 }
3737 unlock_out:
3738 folio_unlock(folio);
3739 folio_put(folio);
3740 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3741 return copied;
3742 }
3743
f2fs_invalidate_folio(struct folio * folio,size_t offset,size_t length)3744 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3745 {
3746 struct inode *inode = folio->mapping->host;
3747 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3748
3749 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3750 (offset || length != folio_size(folio)))
3751 return;
3752
3753 if (folio_test_dirty(folio)) {
3754 if (inode->i_ino == F2FS_META_INO(sbi)) {
3755 dec_page_count(sbi, F2FS_DIRTY_META);
3756 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3757 dec_page_count(sbi, F2FS_DIRTY_NODES);
3758 } else {
3759 inode_dec_dirty_pages(inode);
3760 f2fs_remove_dirty_inode(inode);
3761 }
3762 }
3763 clear_page_private_all(&folio->page);
3764 }
3765
f2fs_release_folio(struct folio * folio,gfp_t wait)3766 bool f2fs_release_folio(struct folio *folio, gfp_t wait)
3767 {
3768 /* If this is dirty folio, keep private data */
3769 if (folio_test_dirty(folio))
3770 return false;
3771
3772 clear_page_private_all(&folio->page);
3773 return true;
3774 }
3775
f2fs_dirty_data_folio(struct address_space * mapping,struct folio * folio)3776 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3777 struct folio *folio)
3778 {
3779 struct inode *inode = mapping->host;
3780
3781 trace_f2fs_set_page_dirty(folio, DATA);
3782
3783 if (!folio_test_uptodate(folio))
3784 folio_mark_uptodate(folio);
3785 BUG_ON(folio_test_swapcache(folio));
3786
3787 if (filemap_dirty_folio(mapping, folio)) {
3788 f2fs_update_dirty_folio(inode, folio);
3789 return true;
3790 }
3791 return false;
3792 }
3793
3794
f2fs_bmap_compress(struct inode * inode,sector_t block)3795 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3796 {
3797 #ifdef CONFIG_F2FS_FS_COMPRESSION
3798 struct dnode_of_data dn;
3799 sector_t start_idx, blknr = 0;
3800 int ret;
3801
3802 start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3803
3804 set_new_dnode(&dn, inode, NULL, NULL, 0);
3805 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3806 if (ret)
3807 return 0;
3808
3809 if (dn.data_blkaddr != COMPRESS_ADDR) {
3810 dn.ofs_in_node += block - start_idx;
3811 blknr = f2fs_data_blkaddr(&dn);
3812 if (!__is_valid_data_blkaddr(blknr))
3813 blknr = 0;
3814 }
3815
3816 f2fs_put_dnode(&dn);
3817 return blknr;
3818 #else
3819 return 0;
3820 #endif
3821 }
3822
3823
f2fs_bmap(struct address_space * mapping,sector_t block)3824 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3825 {
3826 struct inode *inode = mapping->host;
3827 sector_t blknr = 0;
3828
3829 if (f2fs_has_inline_data(inode))
3830 goto out;
3831
3832 /* make sure allocating whole blocks */
3833 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3834 filemap_write_and_wait(mapping);
3835
3836 /* Block number less than F2FS MAX BLOCKS */
3837 if (unlikely(block >= max_file_blocks(inode)))
3838 goto out;
3839
3840 if (f2fs_compressed_file(inode)) {
3841 blknr = f2fs_bmap_compress(inode, block);
3842 } else {
3843 struct f2fs_map_blocks map;
3844
3845 memset(&map, 0, sizeof(map));
3846 map.m_lblk = block;
3847 map.m_len = 1;
3848 map.m_next_pgofs = NULL;
3849 map.m_seg_type = NO_CHECK_TYPE;
3850
3851 if (!f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_BMAP))
3852 blknr = map.m_pblk;
3853 }
3854 out:
3855 trace_f2fs_bmap(inode, block, blknr);
3856 return blknr;
3857 }
3858
3859 #ifdef CONFIG_SWAP
f2fs_migrate_blocks(struct inode * inode,block_t start_blk,unsigned int blkcnt)3860 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3861 unsigned int blkcnt)
3862 {
3863 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3864 unsigned int blkofs;
3865 unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3866 unsigned int end_blk = start_blk + blkcnt - 1;
3867 unsigned int secidx = start_blk / blk_per_sec;
3868 unsigned int end_sec;
3869 int ret = 0;
3870
3871 if (!blkcnt)
3872 return 0;
3873 end_sec = end_blk / blk_per_sec;
3874
3875 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3876 filemap_invalidate_lock(inode->i_mapping);
3877
3878 set_inode_flag(inode, FI_ALIGNED_WRITE);
3879 set_inode_flag(inode, FI_OPU_WRITE);
3880
3881 for (; secidx <= end_sec; secidx++) {
3882 unsigned int blkofs_end = secidx == end_sec ?
3883 end_blk % blk_per_sec : blk_per_sec - 1;
3884
3885 f2fs_down_write(&sbi->pin_sem);
3886
3887 ret = f2fs_allocate_pinning_section(sbi);
3888 if (ret) {
3889 f2fs_up_write(&sbi->pin_sem);
3890 break;
3891 }
3892
3893 set_inode_flag(inode, FI_SKIP_WRITES);
3894
3895 for (blkofs = 0; blkofs <= blkofs_end; blkofs++) {
3896 struct page *page;
3897 unsigned int blkidx = secidx * blk_per_sec + blkofs;
3898
3899 page = f2fs_get_lock_data_page(inode, blkidx, true);
3900 if (IS_ERR(page)) {
3901 f2fs_up_write(&sbi->pin_sem);
3902 ret = PTR_ERR(page);
3903 goto done;
3904 }
3905
3906 set_page_dirty(page);
3907 f2fs_put_page(page, 1);
3908 }
3909
3910 clear_inode_flag(inode, FI_SKIP_WRITES);
3911
3912 ret = filemap_fdatawrite(inode->i_mapping);
3913
3914 f2fs_up_write(&sbi->pin_sem);
3915
3916 if (ret)
3917 break;
3918 }
3919
3920 done:
3921 clear_inode_flag(inode, FI_SKIP_WRITES);
3922 clear_inode_flag(inode, FI_OPU_WRITE);
3923 clear_inode_flag(inode, FI_ALIGNED_WRITE);
3924
3925 filemap_invalidate_unlock(inode->i_mapping);
3926 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3927
3928 return ret;
3929 }
3930
check_swap_activate(struct swap_info_struct * sis,struct file * swap_file,sector_t * span)3931 static int check_swap_activate(struct swap_info_struct *sis,
3932 struct file *swap_file, sector_t *span)
3933 {
3934 struct address_space *mapping = swap_file->f_mapping;
3935 struct inode *inode = mapping->host;
3936 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3937 block_t cur_lblock;
3938 block_t last_lblock;
3939 block_t pblock;
3940 block_t lowest_pblock = -1;
3941 block_t highest_pblock = 0;
3942 int nr_extents = 0;
3943 unsigned int nr_pblocks;
3944 unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3945 unsigned int not_aligned = 0;
3946 int ret = 0;
3947
3948 /*
3949 * Map all the blocks into the extent list. This code doesn't try
3950 * to be very smart.
3951 */
3952 cur_lblock = 0;
3953 last_lblock = bytes_to_blks(inode, i_size_read(inode));
3954
3955 while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3956 struct f2fs_map_blocks map;
3957 retry:
3958 cond_resched();
3959
3960 memset(&map, 0, sizeof(map));
3961 map.m_lblk = cur_lblock;
3962 map.m_len = last_lblock - cur_lblock;
3963 map.m_next_pgofs = NULL;
3964 map.m_next_extent = NULL;
3965 map.m_seg_type = NO_CHECK_TYPE;
3966 map.m_may_create = false;
3967
3968 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
3969 if (ret)
3970 goto out;
3971
3972 /* hole */
3973 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3974 f2fs_err(sbi, "Swapfile has holes");
3975 ret = -EINVAL;
3976 goto out;
3977 }
3978
3979 pblock = map.m_pblk;
3980 nr_pblocks = map.m_len;
3981
3982 if ((pblock - SM_I(sbi)->main_blkaddr) % blks_per_sec ||
3983 nr_pblocks % blks_per_sec ||
3984 !f2fs_valid_pinned_area(sbi, pblock)) {
3985 bool last_extent = false;
3986
3987 not_aligned++;
3988
3989 nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3990 if (cur_lblock + nr_pblocks > sis->max)
3991 nr_pblocks -= blks_per_sec;
3992
3993 /* this extent is last one */
3994 if (!nr_pblocks) {
3995 nr_pblocks = last_lblock - cur_lblock;
3996 last_extent = true;
3997 }
3998
3999 ret = f2fs_migrate_blocks(inode, cur_lblock,
4000 nr_pblocks);
4001 if (ret) {
4002 if (ret == -ENOENT)
4003 ret = -EINVAL;
4004 goto out;
4005 }
4006
4007 if (!last_extent)
4008 goto retry;
4009 }
4010
4011 if (cur_lblock + nr_pblocks >= sis->max)
4012 nr_pblocks = sis->max - cur_lblock;
4013
4014 if (cur_lblock) { /* exclude the header page */
4015 if (pblock < lowest_pblock)
4016 lowest_pblock = pblock;
4017 if (pblock + nr_pblocks - 1 > highest_pblock)
4018 highest_pblock = pblock + nr_pblocks - 1;
4019 }
4020
4021 /*
4022 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4023 */
4024 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
4025 if (ret < 0)
4026 goto out;
4027 nr_extents += ret;
4028 cur_lblock += nr_pblocks;
4029 }
4030 ret = nr_extents;
4031 *span = 1 + highest_pblock - lowest_pblock;
4032 if (cur_lblock == 0)
4033 cur_lblock = 1; /* force Empty message */
4034 sis->max = cur_lblock;
4035 sis->pages = cur_lblock - 1;
4036 sis->highest_bit = cur_lblock - 1;
4037 out:
4038 if (not_aligned)
4039 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%lu * N)",
4040 not_aligned, blks_per_sec * F2FS_BLKSIZE);
4041 return ret;
4042 }
4043
f2fs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)4044 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4045 sector_t *span)
4046 {
4047 struct inode *inode = file_inode(file);
4048 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4049 int ret;
4050
4051 if (!S_ISREG(inode->i_mode))
4052 return -EINVAL;
4053
4054 if (f2fs_readonly(sbi->sb))
4055 return -EROFS;
4056
4057 if (f2fs_lfs_mode(sbi) && !f2fs_sb_has_blkzoned(sbi)) {
4058 f2fs_err(sbi, "Swapfile not supported in LFS mode");
4059 return -EINVAL;
4060 }
4061
4062 ret = f2fs_convert_inline_inode(inode);
4063 if (ret)
4064 return ret;
4065
4066 if (!f2fs_disable_compressed_file(inode))
4067 return -EINVAL;
4068
4069 ret = filemap_fdatawrite(inode->i_mapping);
4070 if (ret < 0)
4071 return ret;
4072
4073 f2fs_precache_extents(inode);
4074
4075 ret = check_swap_activate(sis, file, span);
4076 if (ret < 0)
4077 return ret;
4078
4079 stat_inc_swapfile_inode(inode);
4080 set_inode_flag(inode, FI_PIN_FILE);
4081 f2fs_update_time(sbi, REQ_TIME);
4082 return ret;
4083 }
4084
f2fs_swap_deactivate(struct file * file)4085 static void f2fs_swap_deactivate(struct file *file)
4086 {
4087 struct inode *inode = file_inode(file);
4088
4089 stat_dec_swapfile_inode(inode);
4090 clear_inode_flag(inode, FI_PIN_FILE);
4091 }
4092 #else
f2fs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)4093 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4094 sector_t *span)
4095 {
4096 return -EOPNOTSUPP;
4097 }
4098
f2fs_swap_deactivate(struct file * file)4099 static void f2fs_swap_deactivate(struct file *file)
4100 {
4101 }
4102 #endif
4103
4104 const struct address_space_operations f2fs_dblock_aops = {
4105 .read_folio = f2fs_read_data_folio,
4106 .readahead = f2fs_readahead,
4107 .writepage = f2fs_write_data_page,
4108 .writepages = f2fs_write_data_pages,
4109 .write_begin = f2fs_write_begin,
4110 .write_end = f2fs_write_end,
4111 .dirty_folio = f2fs_dirty_data_folio,
4112 .migrate_folio = filemap_migrate_folio,
4113 .invalidate_folio = f2fs_invalidate_folio,
4114 .release_folio = f2fs_release_folio,
4115 .bmap = f2fs_bmap,
4116 .swap_activate = f2fs_swap_activate,
4117 .swap_deactivate = f2fs_swap_deactivate,
4118 };
4119
f2fs_clear_page_cache_dirty_tag(struct page * page)4120 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4121 {
4122 struct folio *folio = page_folio(page);
4123 struct address_space *mapping = folio->mapping;
4124 unsigned long flags;
4125
4126 xa_lock_irqsave(&mapping->i_pages, flags);
4127 __xa_clear_mark(&mapping->i_pages, folio->index,
4128 PAGECACHE_TAG_DIRTY);
4129 xa_unlock_irqrestore(&mapping->i_pages, flags);
4130 }
4131
f2fs_init_post_read_processing(void)4132 int __init f2fs_init_post_read_processing(void)
4133 {
4134 bio_post_read_ctx_cache =
4135 kmem_cache_create("f2fs_bio_post_read_ctx",
4136 sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4137 if (!bio_post_read_ctx_cache)
4138 goto fail;
4139 bio_post_read_ctx_pool =
4140 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4141 bio_post_read_ctx_cache);
4142 if (!bio_post_read_ctx_pool)
4143 goto fail_free_cache;
4144 return 0;
4145
4146 fail_free_cache:
4147 kmem_cache_destroy(bio_post_read_ctx_cache);
4148 fail:
4149 return -ENOMEM;
4150 }
4151
f2fs_destroy_post_read_processing(void)4152 void f2fs_destroy_post_read_processing(void)
4153 {
4154 mempool_destroy(bio_post_read_ctx_pool);
4155 kmem_cache_destroy(bio_post_read_ctx_cache);
4156 }
4157
f2fs_init_post_read_wq(struct f2fs_sb_info * sbi)4158 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4159 {
4160 if (!f2fs_sb_has_encrypt(sbi) &&
4161 !f2fs_sb_has_verity(sbi) &&
4162 !f2fs_sb_has_compression(sbi))
4163 return 0;
4164
4165 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4166 WQ_UNBOUND | WQ_HIGHPRI,
4167 num_online_cpus());
4168 return sbi->post_read_wq ? 0 : -ENOMEM;
4169 }
4170
f2fs_destroy_post_read_wq(struct f2fs_sb_info * sbi)4171 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4172 {
4173 if (sbi->post_read_wq)
4174 destroy_workqueue(sbi->post_read_wq);
4175 }
4176
f2fs_init_bio_entry_cache(void)4177 int __init f2fs_init_bio_entry_cache(void)
4178 {
4179 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4180 sizeof(struct bio_entry));
4181 return bio_entry_slab ? 0 : -ENOMEM;
4182 }
4183
f2fs_destroy_bio_entry_cache(void)4184 void f2fs_destroy_bio_entry_cache(void)
4185 {
4186 kmem_cache_destroy(bio_entry_slab);
4187 }
4188
f2fs_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)4189 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4190 unsigned int flags, struct iomap *iomap,
4191 struct iomap *srcmap)
4192 {
4193 struct f2fs_map_blocks map = {};
4194 pgoff_t next_pgofs = 0;
4195 int err;
4196
4197 map.m_lblk = bytes_to_blks(inode, offset);
4198 map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4199 map.m_next_pgofs = &next_pgofs;
4200 map.m_seg_type = f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
4201 inode->i_write_hint);
4202 if (flags & IOMAP_WRITE)
4203 map.m_may_create = true;
4204
4205 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DIO);
4206 if (err)
4207 return err;
4208
4209 iomap->offset = blks_to_bytes(inode, map.m_lblk);
4210
4211 /*
4212 * When inline encryption is enabled, sometimes I/O to an encrypted file
4213 * has to be broken up to guarantee DUN contiguity. Handle this by
4214 * limiting the length of the mapping returned.
4215 */
4216 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4217
4218 /*
4219 * We should never see delalloc or compressed extents here based on
4220 * prior flushing and checks.
4221 */
4222 if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR))
4223 return -EINVAL;
4224
4225 if (map.m_flags & F2FS_MAP_MAPPED) {
4226 if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR))
4227 return -EINVAL;
4228
4229 iomap->length = blks_to_bytes(inode, map.m_len);
4230 iomap->type = IOMAP_MAPPED;
4231 iomap->flags |= IOMAP_F_MERGED;
4232 iomap->bdev = map.m_bdev;
4233 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4234 } else {
4235 if (flags & IOMAP_WRITE)
4236 return -ENOTBLK;
4237
4238 if (map.m_pblk == NULL_ADDR) {
4239 iomap->length = blks_to_bytes(inode, next_pgofs) -
4240 iomap->offset;
4241 iomap->type = IOMAP_HOLE;
4242 } else if (map.m_pblk == NEW_ADDR) {
4243 iomap->length = blks_to_bytes(inode, map.m_len);
4244 iomap->type = IOMAP_UNWRITTEN;
4245 } else {
4246 f2fs_bug_on(F2FS_I_SB(inode), 1);
4247 }
4248 iomap->addr = IOMAP_NULL_ADDR;
4249 }
4250
4251 if (map.m_flags & F2FS_MAP_NEW)
4252 iomap->flags |= IOMAP_F_NEW;
4253 if ((inode->i_state & I_DIRTY_DATASYNC) ||
4254 offset + length > i_size_read(inode))
4255 iomap->flags |= IOMAP_F_DIRTY;
4256
4257 return 0;
4258 }
4259
4260 const struct iomap_ops f2fs_iomap_ops = {
4261 .iomap_begin = f2fs_iomap_begin,
4262 };
4263