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 }
469 iostat_alloc_and_bind_ctx(sbi, bio, NULL);
470
471 if (fio->io_wbc)
472 wbc_init_bio(fio->io_wbc, bio);
473
474 return bio;
475 }
476
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)477 static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
478 pgoff_t first_idx,
479 const struct f2fs_io_info *fio,
480 gfp_t gfp_mask)
481 {
482 /*
483 * The f2fs garbage collector sets ->encrypted_page when it wants to
484 * read/write raw data without encryption.
485 */
486 if (!fio || !fio->encrypted_page)
487 fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
488 }
489
f2fs_crypt_mergeable_bio(struct bio * bio,const struct inode * inode,pgoff_t next_idx,const struct f2fs_io_info * fio)490 static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
491 pgoff_t next_idx,
492 const struct f2fs_io_info *fio)
493 {
494 /*
495 * The f2fs garbage collector sets ->encrypted_page when it wants to
496 * read/write raw data without encryption.
497 */
498 if (fio && fio->encrypted_page)
499 return !bio_has_crypt_ctx(bio);
500
501 return fscrypt_mergeable_bio(bio, inode, next_idx);
502 }
503
f2fs_submit_read_bio(struct f2fs_sb_info * sbi,struct bio * bio,enum page_type type)504 void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
505 enum page_type type)
506 {
507 WARN_ON_ONCE(!is_read_io(bio_op(bio)));
508 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
509
510 iostat_update_submit_ctx(bio, type);
511 submit_bio(bio);
512 }
513
f2fs_submit_write_bio(struct f2fs_sb_info * sbi,struct bio * bio,enum page_type type)514 static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio,
515 enum page_type type)
516 {
517 WARN_ON_ONCE(is_read_io(bio_op(bio)));
518
519 if (f2fs_lfs_mode(sbi) && current->plug && PAGE_TYPE_ON_MAIN(type))
520 blk_finish_plug(current->plug);
521
522 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
523 iostat_update_submit_ctx(bio, type);
524 submit_bio(bio);
525 }
526
__submit_merged_bio(struct f2fs_bio_info * io)527 static void __submit_merged_bio(struct f2fs_bio_info *io)
528 {
529 struct f2fs_io_info *fio = &io->fio;
530
531 if (!io->bio)
532 return;
533
534 if (is_read_io(fio->op)) {
535 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
536 f2fs_submit_read_bio(io->sbi, io->bio, fio->type);
537 } else {
538 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
539 f2fs_submit_write_bio(io->sbi, io->bio, fio->type);
540 }
541 io->bio = NULL;
542 }
543
__has_merged_page(struct bio * bio,struct inode * inode,struct page * page,nid_t ino)544 static bool __has_merged_page(struct bio *bio, struct inode *inode,
545 struct page *page, nid_t ino)
546 {
547 struct bio_vec *bvec;
548 struct bvec_iter_all iter_all;
549
550 if (!bio)
551 return false;
552
553 if (!inode && !page && !ino)
554 return true;
555
556 bio_for_each_segment_all(bvec, bio, iter_all) {
557 struct page *target = bvec->bv_page;
558
559 if (fscrypt_is_bounce_page(target)) {
560 target = fscrypt_pagecache_page(target);
561 if (IS_ERR(target))
562 continue;
563 }
564 if (f2fs_is_compressed_page(target)) {
565 target = f2fs_compress_control_page(target);
566 if (IS_ERR(target))
567 continue;
568 }
569
570 if (inode && inode == target->mapping->host)
571 return true;
572 if (page && page == target)
573 return true;
574 if (ino && ino == ino_of_node(target))
575 return true;
576 }
577
578 return false;
579 }
580
f2fs_init_write_merge_io(struct f2fs_sb_info * sbi)581 int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
582 {
583 int i;
584
585 for (i = 0; i < NR_PAGE_TYPE; i++) {
586 int n = (i == META) ? 1 : NR_TEMP_TYPE;
587 int j;
588
589 sbi->write_io[i] = f2fs_kmalloc(sbi,
590 array_size(n, sizeof(struct f2fs_bio_info)),
591 GFP_KERNEL);
592 if (!sbi->write_io[i])
593 return -ENOMEM;
594
595 for (j = HOT; j < n; j++) {
596 init_f2fs_rwsem(&sbi->write_io[i][j].io_rwsem);
597 sbi->write_io[i][j].sbi = sbi;
598 sbi->write_io[i][j].bio = NULL;
599 spin_lock_init(&sbi->write_io[i][j].io_lock);
600 INIT_LIST_HEAD(&sbi->write_io[i][j].io_list);
601 INIT_LIST_HEAD(&sbi->write_io[i][j].bio_list);
602 init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock);
603 #ifdef CONFIG_BLK_DEV_ZONED
604 init_completion(&sbi->write_io[i][j].zone_wait);
605 sbi->write_io[i][j].zone_pending_bio = NULL;
606 sbi->write_io[i][j].bi_private = NULL;
607 #endif
608 }
609 }
610
611 return 0;
612 }
613
__f2fs_submit_merged_write(struct f2fs_sb_info * sbi,enum page_type type,enum temp_type temp)614 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
615 enum page_type type, enum temp_type temp)
616 {
617 enum page_type btype = PAGE_TYPE_OF_BIO(type);
618 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
619
620 f2fs_down_write(&io->io_rwsem);
621
622 if (!io->bio)
623 goto unlock_out;
624
625 /* change META to META_FLUSH in the checkpoint procedure */
626 if (type >= META_FLUSH) {
627 io->fio.type = META_FLUSH;
628 io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
629 if (!test_opt(sbi, NOBARRIER))
630 io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
631 }
632 __submit_merged_bio(io);
633 unlock_out:
634 f2fs_up_write(&io->io_rwsem);
635 }
636
__submit_merged_write_cond(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page,nid_t ino,enum page_type type,bool force)637 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
638 struct inode *inode, struct page *page,
639 nid_t ino, enum page_type type, bool force)
640 {
641 enum temp_type temp;
642 bool ret = true;
643
644 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
645 if (!force) {
646 enum page_type btype = PAGE_TYPE_OF_BIO(type);
647 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
648
649 f2fs_down_read(&io->io_rwsem);
650 ret = __has_merged_page(io->bio, inode, page, ino);
651 f2fs_up_read(&io->io_rwsem);
652 }
653 if (ret)
654 __f2fs_submit_merged_write(sbi, type, temp);
655
656 /* TODO: use HOT temp only for meta pages now. */
657 if (type >= META)
658 break;
659 }
660 }
661
f2fs_submit_merged_write(struct f2fs_sb_info * sbi,enum page_type type)662 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
663 {
664 __submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
665 }
666
f2fs_submit_merged_write_cond(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page,nid_t ino,enum page_type type)667 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
668 struct inode *inode, struct page *page,
669 nid_t ino, enum page_type type)
670 {
671 __submit_merged_write_cond(sbi, inode, page, ino, type, false);
672 }
673
f2fs_flush_merged_writes(struct f2fs_sb_info * sbi)674 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
675 {
676 f2fs_submit_merged_write(sbi, DATA);
677 f2fs_submit_merged_write(sbi, NODE);
678 f2fs_submit_merged_write(sbi, META);
679 }
680
681 /*
682 * Fill the locked page with data located in the block address.
683 * A caller needs to unlock the page on failure.
684 */
f2fs_submit_page_bio(struct f2fs_io_info * fio)685 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
686 {
687 struct bio *bio;
688 struct page *page = fio->encrypted_page ?
689 fio->encrypted_page : fio->page;
690
691 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
692 fio->is_por ? META_POR : (__is_meta_io(fio) ?
693 META_GENERIC : DATA_GENERIC_ENHANCE))) {
694 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
695 return -EFSCORRUPTED;
696 }
697
698 trace_f2fs_submit_page_bio(page, fio);
699
700 /* Allocate a new bio */
701 bio = __bio_alloc(fio, 1);
702
703 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
704 fio->page->index, fio, GFP_NOIO);
705
706 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
707 bio_put(bio);
708 return -EFAULT;
709 }
710
711 if (fio->io_wbc && !is_read_io(fio->op))
712 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
713
714 inc_page_count(fio->sbi, is_read_io(fio->op) ?
715 __read_io_type(page) : WB_DATA_TYPE(fio->page, false));
716
717 if (is_read_io(bio_op(bio)))
718 f2fs_submit_read_bio(fio->sbi, bio, fio->type);
719 else
720 f2fs_submit_write_bio(fio->sbi, bio, fio->type);
721 return 0;
722 }
723
page_is_mergeable(struct f2fs_sb_info * sbi,struct bio * bio,block_t last_blkaddr,block_t cur_blkaddr)724 static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
725 block_t last_blkaddr, block_t cur_blkaddr)
726 {
727 if (unlikely(sbi->max_io_bytes &&
728 bio->bi_iter.bi_size >= sbi->max_io_bytes))
729 return false;
730 if (last_blkaddr + 1 != cur_blkaddr)
731 return false;
732 return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
733 }
734
io_type_is_mergeable(struct f2fs_bio_info * io,struct f2fs_io_info * fio)735 static bool io_type_is_mergeable(struct f2fs_bio_info *io,
736 struct f2fs_io_info *fio)
737 {
738 if (io->fio.op != fio->op)
739 return false;
740 return io->fio.op_flags == fio->op_flags;
741 }
742
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)743 static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
744 struct f2fs_bio_info *io,
745 struct f2fs_io_info *fio,
746 block_t last_blkaddr,
747 block_t cur_blkaddr)
748 {
749 if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
750 return false;
751 return io_type_is_mergeable(io, fio);
752 }
753
add_bio_entry(struct f2fs_sb_info * sbi,struct bio * bio,struct page * page,enum temp_type temp)754 static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
755 struct page *page, enum temp_type temp)
756 {
757 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
758 struct bio_entry *be;
759
760 be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
761 be->bio = bio;
762 bio_get(bio);
763
764 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
765 f2fs_bug_on(sbi, 1);
766
767 f2fs_down_write(&io->bio_list_lock);
768 list_add_tail(&be->list, &io->bio_list);
769 f2fs_up_write(&io->bio_list_lock);
770 }
771
del_bio_entry(struct bio_entry * be)772 static void del_bio_entry(struct bio_entry *be)
773 {
774 list_del(&be->list);
775 kmem_cache_free(bio_entry_slab, be);
776 }
777
add_ipu_page(struct f2fs_io_info * fio,struct bio ** bio,struct page * page)778 static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
779 struct page *page)
780 {
781 struct f2fs_sb_info *sbi = fio->sbi;
782 enum temp_type temp;
783 bool found = false;
784 int ret = -EAGAIN;
785
786 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
787 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
788 struct list_head *head = &io->bio_list;
789 struct bio_entry *be;
790
791 f2fs_down_write(&io->bio_list_lock);
792 list_for_each_entry(be, head, list) {
793 if (be->bio != *bio)
794 continue;
795
796 found = true;
797
798 f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
799 *fio->last_block,
800 fio->new_blkaddr));
801 if (f2fs_crypt_mergeable_bio(*bio,
802 fio->page->mapping->host,
803 fio->page->index, fio) &&
804 bio_add_page(*bio, page, PAGE_SIZE, 0) ==
805 PAGE_SIZE) {
806 ret = 0;
807 break;
808 }
809
810 /* page can't be merged into bio; submit the bio */
811 del_bio_entry(be);
812 f2fs_submit_write_bio(sbi, *bio, DATA);
813 break;
814 }
815 f2fs_up_write(&io->bio_list_lock);
816 }
817
818 if (ret) {
819 bio_put(*bio);
820 *bio = NULL;
821 }
822
823 return ret;
824 }
825
f2fs_submit_merged_ipu_write(struct f2fs_sb_info * sbi,struct bio ** bio,struct page * page)826 void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
827 struct bio **bio, struct page *page)
828 {
829 enum temp_type temp;
830 bool found = false;
831 struct bio *target = bio ? *bio : NULL;
832
833 f2fs_bug_on(sbi, !target && !page);
834
835 for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
836 struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
837 struct list_head *head = &io->bio_list;
838 struct bio_entry *be;
839
840 if (list_empty(head))
841 continue;
842
843 f2fs_down_read(&io->bio_list_lock);
844 list_for_each_entry(be, head, list) {
845 if (target)
846 found = (target == be->bio);
847 else
848 found = __has_merged_page(be->bio, NULL,
849 page, 0);
850 if (found)
851 break;
852 }
853 f2fs_up_read(&io->bio_list_lock);
854
855 if (!found)
856 continue;
857
858 found = false;
859
860 f2fs_down_write(&io->bio_list_lock);
861 list_for_each_entry(be, head, list) {
862 if (target)
863 found = (target == be->bio);
864 else
865 found = __has_merged_page(be->bio, NULL,
866 page, 0);
867 if (found) {
868 target = be->bio;
869 del_bio_entry(be);
870 break;
871 }
872 }
873 f2fs_up_write(&io->bio_list_lock);
874 }
875
876 if (found)
877 f2fs_submit_write_bio(sbi, target, DATA);
878 if (bio && *bio) {
879 bio_put(*bio);
880 *bio = NULL;
881 }
882 }
883
f2fs_merge_page_bio(struct f2fs_io_info * fio)884 int f2fs_merge_page_bio(struct f2fs_io_info *fio)
885 {
886 struct bio *bio = *fio->bio;
887 struct page *page = fio->encrypted_page ?
888 fio->encrypted_page : fio->page;
889
890 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
891 __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) {
892 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
893 return -EFSCORRUPTED;
894 }
895
896 trace_f2fs_submit_page_bio(page, fio);
897
898 if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
899 fio->new_blkaddr))
900 f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
901 alloc_new:
902 if (!bio) {
903 bio = __bio_alloc(fio, BIO_MAX_VECS);
904 f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
905 fio->page->index, fio, GFP_NOIO);
906
907 add_bio_entry(fio->sbi, bio, page, fio->temp);
908 } else {
909 if (add_ipu_page(fio, &bio, page))
910 goto alloc_new;
911 }
912
913 if (fio->io_wbc)
914 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
915
916 inc_page_count(fio->sbi, WB_DATA_TYPE(page, false));
917
918 *fio->last_block = fio->new_blkaddr;
919 *fio->bio = bio;
920
921 return 0;
922 }
923
924 #ifdef CONFIG_BLK_DEV_ZONED
is_end_zone_blkaddr(struct f2fs_sb_info * sbi,block_t blkaddr)925 static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr)
926 {
927 int devi = 0;
928
929 if (f2fs_is_multi_device(sbi)) {
930 devi = f2fs_target_device_index(sbi, blkaddr);
931 if (blkaddr < FDEV(devi).start_blk ||
932 blkaddr > FDEV(devi).end_blk) {
933 f2fs_err(sbi, "Invalid block %x", blkaddr);
934 return false;
935 }
936 blkaddr -= FDEV(devi).start_blk;
937 }
938 return bdev_zoned_model(FDEV(devi).bdev) == BLK_ZONED_HM &&
939 f2fs_blkz_is_seq(sbi, devi, blkaddr) &&
940 (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1);
941 }
942 #endif
943
f2fs_submit_page_write(struct f2fs_io_info * fio)944 void f2fs_submit_page_write(struct f2fs_io_info *fio)
945 {
946 struct f2fs_sb_info *sbi = fio->sbi;
947 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
948 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
949 struct page *bio_page;
950 enum count_type type;
951
952 f2fs_bug_on(sbi, is_read_io(fio->op));
953
954 f2fs_down_write(&io->io_rwsem);
955 next:
956 #ifdef CONFIG_BLK_DEV_ZONED
957 if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) {
958 wait_for_completion_io(&io->zone_wait);
959 bio_put(io->zone_pending_bio);
960 io->zone_pending_bio = NULL;
961 io->bi_private = NULL;
962 }
963 #endif
964
965 if (fio->in_list) {
966 spin_lock(&io->io_lock);
967 if (list_empty(&io->io_list)) {
968 spin_unlock(&io->io_lock);
969 goto out;
970 }
971 fio = list_first_entry(&io->io_list,
972 struct f2fs_io_info, list);
973 list_del(&fio->list);
974 spin_unlock(&io->io_lock);
975 }
976
977 verify_fio_blkaddr(fio);
978
979 if (fio->encrypted_page)
980 bio_page = fio->encrypted_page;
981 else if (fio->compressed_page)
982 bio_page = fio->compressed_page;
983 else
984 bio_page = fio->page;
985
986 /* set submitted = true as a return value */
987 fio->submitted = 1;
988
989 type = WB_DATA_TYPE(bio_page, fio->compressed_page);
990 inc_page_count(sbi, type);
991
992 if (io->bio &&
993 (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
994 fio->new_blkaddr) ||
995 !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
996 bio_page->index, fio)))
997 __submit_merged_bio(io);
998 alloc_new:
999 if (io->bio == NULL) {
1000 io->bio = __bio_alloc(fio, BIO_MAX_VECS);
1001 f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
1002 bio_page->index, fio, GFP_NOIO);
1003 io->fio = *fio;
1004 }
1005
1006 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1007 __submit_merged_bio(io);
1008 goto alloc_new;
1009 }
1010
1011 if (fio->io_wbc)
1012 wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
1013
1014 io->last_block_in_bio = fio->new_blkaddr;
1015
1016 trace_f2fs_submit_page_write(fio->page, fio);
1017 #ifdef CONFIG_BLK_DEV_ZONED
1018 if (f2fs_sb_has_blkzoned(sbi) && btype < META &&
1019 is_end_zone_blkaddr(sbi, fio->new_blkaddr)) {
1020 bio_get(io->bio);
1021 reinit_completion(&io->zone_wait);
1022 io->bi_private = io->bio->bi_private;
1023 io->bio->bi_private = io;
1024 io->bio->bi_end_io = f2fs_zone_write_end_io;
1025 io->zone_pending_bio = io->bio;
1026 __submit_merged_bio(io);
1027 }
1028 #endif
1029 if (fio->in_list)
1030 goto next;
1031 out:
1032 if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1033 !f2fs_is_checkpoint_ready(sbi))
1034 __submit_merged_bio(io);
1035 f2fs_up_write(&io->io_rwsem);
1036 }
1037
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)1038 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1039 unsigned nr_pages, blk_opf_t op_flag,
1040 pgoff_t first_idx, bool for_write)
1041 {
1042 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1043 struct bio *bio;
1044 struct bio_post_read_ctx *ctx = NULL;
1045 unsigned int post_read_steps = 0;
1046 sector_t sector;
1047 struct block_device *bdev = f2fs_target_device(sbi, blkaddr, §or);
1048
1049 bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
1050 REQ_OP_READ | op_flag,
1051 for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
1052 if (!bio)
1053 return ERR_PTR(-ENOMEM);
1054 bio->bi_iter.bi_sector = sector;
1055 f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1056 bio->bi_end_io = f2fs_read_end_io;
1057
1058 if (fscrypt_inode_uses_fs_layer_crypto(inode))
1059 post_read_steps |= STEP_DECRYPT;
1060
1061 if (f2fs_need_verity(inode, first_idx))
1062 post_read_steps |= STEP_VERITY;
1063
1064 /*
1065 * STEP_DECOMPRESS is handled specially, since a compressed file might
1066 * contain both compressed and uncompressed clusters. We'll allocate a
1067 * bio_post_read_ctx if the file is compressed, but the caller is
1068 * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1069 */
1070
1071 if (post_read_steps || f2fs_compressed_file(inode)) {
1072 /* Due to the mempool, this never fails. */
1073 ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1074 ctx->bio = bio;
1075 ctx->sbi = sbi;
1076 ctx->enabled_steps = post_read_steps;
1077 ctx->fs_blkaddr = blkaddr;
1078 ctx->decompression_attempted = false;
1079 bio->bi_private = ctx;
1080 }
1081 iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1082
1083 return bio;
1084 }
1085
1086 /* 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)1087 static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1088 block_t blkaddr, blk_opf_t op_flags,
1089 bool for_write)
1090 {
1091 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1092 struct bio *bio;
1093
1094 bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1095 page->index, for_write);
1096 if (IS_ERR(bio))
1097 return PTR_ERR(bio);
1098
1099 /* wait for GCed page writeback via META_MAPPING */
1100 f2fs_wait_on_block_writeback(inode, blkaddr);
1101
1102 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1103 iostat_update_and_unbind_ctx(bio);
1104 if (bio->bi_private)
1105 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1106 bio_put(bio);
1107 return -EFAULT;
1108 }
1109 inc_page_count(sbi, F2FS_RD_DATA);
1110 f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE);
1111 f2fs_submit_read_bio(sbi, bio, DATA);
1112 return 0;
1113 }
1114
__set_data_blkaddr(struct dnode_of_data * dn,block_t blkaddr)1115 static void __set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1116 {
1117 __le32 *addr = get_dnode_addr(dn->inode, dn->node_page);
1118
1119 dn->data_blkaddr = blkaddr;
1120 addr[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1121 }
1122
1123 /*
1124 * Lock ordering for the change of data block address:
1125 * ->data_page
1126 * ->node_page
1127 * update block addresses in the node page
1128 */
f2fs_set_data_blkaddr(struct dnode_of_data * dn,block_t blkaddr)1129 void f2fs_set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1130 {
1131 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1132 __set_data_blkaddr(dn, blkaddr);
1133 if (set_page_dirty(dn->node_page))
1134 dn->node_changed = true;
1135 }
1136
f2fs_update_data_blkaddr(struct dnode_of_data * dn,block_t blkaddr)1137 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1138 {
1139 f2fs_set_data_blkaddr(dn, blkaddr);
1140 f2fs_update_read_extent_cache(dn);
1141 }
1142
1143 /* 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)1144 int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1145 {
1146 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1147 int err;
1148
1149 if (!count)
1150 return 0;
1151
1152 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1153 return -EPERM;
1154 err = inc_valid_block_count(sbi, dn->inode, &count, true);
1155 if (unlikely(err))
1156 return err;
1157
1158 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1159 dn->ofs_in_node, count);
1160
1161 f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1162
1163 for (; count > 0; dn->ofs_in_node++) {
1164 block_t blkaddr = f2fs_data_blkaddr(dn);
1165
1166 if (blkaddr == NULL_ADDR) {
1167 __set_data_blkaddr(dn, NEW_ADDR);
1168 count--;
1169 }
1170 }
1171
1172 if (set_page_dirty(dn->node_page))
1173 dn->node_changed = true;
1174 return 0;
1175 }
1176
1177 /* Should keep dn->ofs_in_node unchanged */
f2fs_reserve_new_block(struct dnode_of_data * dn)1178 int f2fs_reserve_new_block(struct dnode_of_data *dn)
1179 {
1180 unsigned int ofs_in_node = dn->ofs_in_node;
1181 int ret;
1182
1183 ret = f2fs_reserve_new_blocks(dn, 1);
1184 dn->ofs_in_node = ofs_in_node;
1185 return ret;
1186 }
1187
f2fs_reserve_block(struct dnode_of_data * dn,pgoff_t index)1188 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1189 {
1190 bool need_put = dn->inode_page ? false : true;
1191 int err;
1192
1193 err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1194 if (err)
1195 return err;
1196
1197 if (dn->data_blkaddr == NULL_ADDR)
1198 err = f2fs_reserve_new_block(dn);
1199 if (err || need_put)
1200 f2fs_put_dnode(dn);
1201 return err;
1202 }
1203
f2fs_get_read_data_page(struct inode * inode,pgoff_t index,blk_opf_t op_flags,bool for_write,pgoff_t * next_pgofs)1204 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1205 blk_opf_t op_flags, bool for_write,
1206 pgoff_t *next_pgofs)
1207 {
1208 struct address_space *mapping = inode->i_mapping;
1209 struct dnode_of_data dn;
1210 struct page *page;
1211 int err;
1212
1213 page = f2fs_grab_cache_page(mapping, index, for_write);
1214 if (!page)
1215 return ERR_PTR(-ENOMEM);
1216
1217 if (f2fs_lookup_read_extent_cache_block(inode, index,
1218 &dn.data_blkaddr)) {
1219 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1220 DATA_GENERIC_ENHANCE_READ)) {
1221 err = -EFSCORRUPTED;
1222 f2fs_handle_error(F2FS_I_SB(inode),
1223 ERROR_INVALID_BLKADDR);
1224 goto put_err;
1225 }
1226 goto got_it;
1227 }
1228
1229 set_new_dnode(&dn, inode, NULL, NULL, 0);
1230 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1231 if (err) {
1232 if (err == -ENOENT && next_pgofs)
1233 *next_pgofs = f2fs_get_next_page_offset(&dn, index);
1234 goto put_err;
1235 }
1236 f2fs_put_dnode(&dn);
1237
1238 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1239 err = -ENOENT;
1240 if (next_pgofs)
1241 *next_pgofs = index + 1;
1242 goto put_err;
1243 }
1244 if (dn.data_blkaddr != NEW_ADDR &&
1245 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1246 dn.data_blkaddr,
1247 DATA_GENERIC_ENHANCE)) {
1248 err = -EFSCORRUPTED;
1249 f2fs_handle_error(F2FS_I_SB(inode),
1250 ERROR_INVALID_BLKADDR);
1251 goto put_err;
1252 }
1253 got_it:
1254 if (PageUptodate(page)) {
1255 unlock_page(page);
1256 return page;
1257 }
1258
1259 /*
1260 * A new dentry page is allocated but not able to be written, since its
1261 * new inode page couldn't be allocated due to -ENOSPC.
1262 * In such the case, its blkaddr can be remained as NEW_ADDR.
1263 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1264 * f2fs_init_inode_metadata.
1265 */
1266 if (dn.data_blkaddr == NEW_ADDR) {
1267 zero_user_segment(page, 0, PAGE_SIZE);
1268 if (!PageUptodate(page))
1269 SetPageUptodate(page);
1270 unlock_page(page);
1271 return page;
1272 }
1273
1274 err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1275 op_flags, for_write);
1276 if (err)
1277 goto put_err;
1278 return page;
1279
1280 put_err:
1281 f2fs_put_page(page, 1);
1282 return ERR_PTR(err);
1283 }
1284
f2fs_find_data_page(struct inode * inode,pgoff_t index,pgoff_t * next_pgofs)1285 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index,
1286 pgoff_t *next_pgofs)
1287 {
1288 struct address_space *mapping = inode->i_mapping;
1289 struct page *page;
1290
1291 page = find_get_page(mapping, index);
1292 if (page && PageUptodate(page))
1293 return page;
1294 f2fs_put_page(page, 0);
1295
1296 page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs);
1297 if (IS_ERR(page))
1298 return page;
1299
1300 if (PageUptodate(page))
1301 return page;
1302
1303 wait_on_page_locked(page);
1304 if (unlikely(!PageUptodate(page))) {
1305 f2fs_put_page(page, 0);
1306 return ERR_PTR(-EIO);
1307 }
1308 return page;
1309 }
1310
1311 /*
1312 * If it tries to access a hole, return an error.
1313 * Because, the callers, functions in dir.c and GC, should be able to know
1314 * whether this page exists or not.
1315 */
f2fs_get_lock_data_page(struct inode * inode,pgoff_t index,bool for_write)1316 struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1317 bool for_write)
1318 {
1319 struct address_space *mapping = inode->i_mapping;
1320 struct page *page;
1321
1322 page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL);
1323 if (IS_ERR(page))
1324 return page;
1325
1326 /* wait for read completion */
1327 lock_page(page);
1328 if (unlikely(page->mapping != mapping || !PageUptodate(page))) {
1329 f2fs_put_page(page, 1);
1330 return ERR_PTR(-EIO);
1331 }
1332 return page;
1333 }
1334
1335 /*
1336 * Caller ensures that this data page is never allocated.
1337 * A new zero-filled data page is allocated in the page cache.
1338 *
1339 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1340 * f2fs_unlock_op().
1341 * Note that, ipage is set only by make_empty_dir, and if any error occur,
1342 * ipage should be released by this function.
1343 */
f2fs_get_new_data_page(struct inode * inode,struct page * ipage,pgoff_t index,bool new_i_size)1344 struct page *f2fs_get_new_data_page(struct inode *inode,
1345 struct page *ipage, pgoff_t index, bool new_i_size)
1346 {
1347 struct address_space *mapping = inode->i_mapping;
1348 struct page *page;
1349 struct dnode_of_data dn;
1350 int err;
1351
1352 page = f2fs_grab_cache_page(mapping, index, true);
1353 if (!page) {
1354 /*
1355 * before exiting, we should make sure ipage will be released
1356 * if any error occur.
1357 */
1358 f2fs_put_page(ipage, 1);
1359 return ERR_PTR(-ENOMEM);
1360 }
1361
1362 set_new_dnode(&dn, inode, ipage, NULL, 0);
1363 err = f2fs_reserve_block(&dn, index);
1364 if (err) {
1365 f2fs_put_page(page, 1);
1366 return ERR_PTR(err);
1367 }
1368 if (!ipage)
1369 f2fs_put_dnode(&dn);
1370
1371 if (PageUptodate(page))
1372 goto got_it;
1373
1374 if (dn.data_blkaddr == NEW_ADDR) {
1375 zero_user_segment(page, 0, PAGE_SIZE);
1376 if (!PageUptodate(page))
1377 SetPageUptodate(page);
1378 } else {
1379 f2fs_put_page(page, 1);
1380
1381 /* if ipage exists, blkaddr should be NEW_ADDR */
1382 f2fs_bug_on(F2FS_I_SB(inode), ipage);
1383 page = f2fs_get_lock_data_page(inode, index, true);
1384 if (IS_ERR(page))
1385 return page;
1386 }
1387 got_it:
1388 if (new_i_size && i_size_read(inode) <
1389 ((loff_t)(index + 1) << PAGE_SHIFT))
1390 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1391 return page;
1392 }
1393
__allocate_data_block(struct dnode_of_data * dn,int seg_type)1394 static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1395 {
1396 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1397 struct f2fs_summary sum;
1398 struct node_info ni;
1399 block_t old_blkaddr;
1400 blkcnt_t count = 1;
1401 int err;
1402
1403 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1404 return -EPERM;
1405
1406 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1407 if (err)
1408 return err;
1409
1410 dn->data_blkaddr = f2fs_data_blkaddr(dn);
1411 if (dn->data_blkaddr == NULL_ADDR) {
1412 err = inc_valid_block_count(sbi, dn->inode, &count, true);
1413 if (unlikely(err))
1414 return err;
1415 }
1416
1417 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1418 old_blkaddr = dn->data_blkaddr;
1419 f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1420 &sum, seg_type, NULL);
1421 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1422 f2fs_invalidate_internal_cache(sbi, old_blkaddr);
1423
1424 f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1425 return 0;
1426 }
1427
f2fs_map_lock(struct f2fs_sb_info * sbi,int flag)1428 static void f2fs_map_lock(struct f2fs_sb_info *sbi, int flag)
1429 {
1430 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1431 f2fs_down_read(&sbi->node_change);
1432 else
1433 f2fs_lock_op(sbi);
1434 }
1435
f2fs_map_unlock(struct f2fs_sb_info * sbi,int flag)1436 static void f2fs_map_unlock(struct f2fs_sb_info *sbi, int flag)
1437 {
1438 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1439 f2fs_up_read(&sbi->node_change);
1440 else
1441 f2fs_unlock_op(sbi);
1442 }
1443
f2fs_get_block_locked(struct dnode_of_data * dn,pgoff_t index)1444 int f2fs_get_block_locked(struct dnode_of_data *dn, pgoff_t index)
1445 {
1446 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1447 int err = 0;
1448
1449 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1450 if (!f2fs_lookup_read_extent_cache_block(dn->inode, index,
1451 &dn->data_blkaddr))
1452 err = f2fs_reserve_block(dn, index);
1453 f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1454
1455 return err;
1456 }
1457
f2fs_map_no_dnode(struct inode * inode,struct f2fs_map_blocks * map,struct dnode_of_data * dn,pgoff_t pgoff)1458 static int f2fs_map_no_dnode(struct inode *inode,
1459 struct f2fs_map_blocks *map, struct dnode_of_data *dn,
1460 pgoff_t pgoff)
1461 {
1462 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1463
1464 /*
1465 * There is one exceptional case that read_node_page() may return
1466 * -ENOENT due to filesystem has been shutdown or cp_error, return
1467 * -EIO in that case.
1468 */
1469 if (map->m_may_create &&
1470 (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || f2fs_cp_error(sbi)))
1471 return -EIO;
1472
1473 if (map->m_next_pgofs)
1474 *map->m_next_pgofs = f2fs_get_next_page_offset(dn, pgoff);
1475 if (map->m_next_extent)
1476 *map->m_next_extent = f2fs_get_next_page_offset(dn, pgoff);
1477 return 0;
1478 }
1479
f2fs_map_blocks_cached(struct inode * inode,struct f2fs_map_blocks * map,int flag)1480 static bool f2fs_map_blocks_cached(struct inode *inode,
1481 struct f2fs_map_blocks *map, int flag)
1482 {
1483 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1484 unsigned int maxblocks = map->m_len;
1485 pgoff_t pgoff = (pgoff_t)map->m_lblk;
1486 struct extent_info ei = {};
1487
1488 if (!f2fs_lookup_read_extent_cache(inode, pgoff, &ei))
1489 return false;
1490
1491 map->m_pblk = ei.blk + pgoff - ei.fofs;
1492 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgoff);
1493 map->m_flags = F2FS_MAP_MAPPED;
1494 if (map->m_next_extent)
1495 *map->m_next_extent = pgoff + map->m_len;
1496
1497 /* for hardware encryption, but to avoid potential issue in future */
1498 if (flag == F2FS_GET_BLOCK_DIO)
1499 f2fs_wait_on_block_writeback_range(inode,
1500 map->m_pblk, map->m_len);
1501
1502 if (f2fs_allow_multi_device_dio(sbi, flag)) {
1503 int bidx = f2fs_target_device_index(sbi, map->m_pblk);
1504 struct f2fs_dev_info *dev = &sbi->devs[bidx];
1505
1506 map->m_bdev = dev->bdev;
1507 map->m_pblk -= dev->start_blk;
1508 map->m_len = min(map->m_len, dev->end_blk + 1 - map->m_pblk);
1509 } else {
1510 map->m_bdev = inode->i_sb->s_bdev;
1511 }
1512 return true;
1513 }
1514
1515 /*
1516 * f2fs_map_blocks() tries to find or build mapping relationship which
1517 * maps continuous logical blocks to physical blocks, and return such
1518 * info via f2fs_map_blocks structure.
1519 */
f2fs_map_blocks(struct inode * inode,struct f2fs_map_blocks * map,int flag)1520 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, int flag)
1521 {
1522 unsigned int maxblocks = map->m_len;
1523 struct dnode_of_data dn;
1524 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1525 int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1526 pgoff_t pgofs, end_offset, end;
1527 int err = 0, ofs = 1;
1528 unsigned int ofs_in_node, last_ofs_in_node;
1529 blkcnt_t prealloc;
1530 block_t blkaddr;
1531 unsigned int start_pgofs;
1532 int bidx = 0;
1533 bool is_hole;
1534
1535 if (!maxblocks)
1536 return 0;
1537
1538 if (!map->m_may_create && f2fs_map_blocks_cached(inode, map, flag))
1539 goto out;
1540
1541 map->m_bdev = inode->i_sb->s_bdev;
1542 map->m_multidev_dio =
1543 f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1544
1545 map->m_len = 0;
1546 map->m_flags = 0;
1547
1548 /* it only supports block size == page size */
1549 pgofs = (pgoff_t)map->m_lblk;
1550 end = pgofs + maxblocks;
1551
1552 next_dnode:
1553 if (map->m_may_create)
1554 f2fs_map_lock(sbi, flag);
1555
1556 /* When reading holes, we need its node page */
1557 set_new_dnode(&dn, inode, NULL, NULL, 0);
1558 err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1559 if (err) {
1560 if (flag == F2FS_GET_BLOCK_BMAP)
1561 map->m_pblk = 0;
1562 if (err == -ENOENT)
1563 err = f2fs_map_no_dnode(inode, map, &dn, pgofs);
1564 goto unlock_out;
1565 }
1566
1567 start_pgofs = pgofs;
1568 prealloc = 0;
1569 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1570 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1571
1572 next_block:
1573 blkaddr = f2fs_data_blkaddr(&dn);
1574 is_hole = !__is_valid_data_blkaddr(blkaddr);
1575 if (!is_hole &&
1576 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1577 err = -EFSCORRUPTED;
1578 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1579 goto sync_out;
1580 }
1581
1582 /* use out-place-update for direct IO under LFS mode */
1583 if (map->m_may_create &&
1584 (is_hole || (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO))) {
1585 if (unlikely(f2fs_cp_error(sbi))) {
1586 err = -EIO;
1587 goto sync_out;
1588 }
1589
1590 switch (flag) {
1591 case F2FS_GET_BLOCK_PRE_AIO:
1592 if (blkaddr == NULL_ADDR) {
1593 prealloc++;
1594 last_ofs_in_node = dn.ofs_in_node;
1595 }
1596 break;
1597 case F2FS_GET_BLOCK_PRE_DIO:
1598 case F2FS_GET_BLOCK_DIO:
1599 err = __allocate_data_block(&dn, map->m_seg_type);
1600 if (err)
1601 goto sync_out;
1602 if (flag == F2FS_GET_BLOCK_PRE_DIO)
1603 file_need_truncate(inode);
1604 set_inode_flag(inode, FI_APPEND_WRITE);
1605 break;
1606 default:
1607 WARN_ON_ONCE(1);
1608 err = -EIO;
1609 goto sync_out;
1610 }
1611
1612 blkaddr = dn.data_blkaddr;
1613 if (is_hole)
1614 map->m_flags |= F2FS_MAP_NEW;
1615 } else if (is_hole) {
1616 if (f2fs_compressed_file(inode) &&
1617 f2fs_sanity_check_cluster(&dn)) {
1618 err = -EFSCORRUPTED;
1619 f2fs_handle_error(sbi,
1620 ERROR_CORRUPTED_CLUSTER);
1621 goto sync_out;
1622 }
1623
1624 switch (flag) {
1625 case F2FS_GET_BLOCK_PRECACHE:
1626 goto sync_out;
1627 case F2FS_GET_BLOCK_BMAP:
1628 map->m_pblk = 0;
1629 goto sync_out;
1630 case F2FS_GET_BLOCK_FIEMAP:
1631 if (blkaddr == NULL_ADDR) {
1632 if (map->m_next_pgofs)
1633 *map->m_next_pgofs = pgofs + 1;
1634 goto sync_out;
1635 }
1636 break;
1637 default:
1638 /* for defragment case */
1639 if (map->m_next_pgofs)
1640 *map->m_next_pgofs = pgofs + 1;
1641 goto sync_out;
1642 }
1643 }
1644
1645 if (flag == F2FS_GET_BLOCK_PRE_AIO)
1646 goto skip;
1647
1648 if (map->m_multidev_dio)
1649 bidx = f2fs_target_device_index(sbi, blkaddr);
1650
1651 if (map->m_len == 0) {
1652 /* reserved delalloc block should be mapped for fiemap. */
1653 if (blkaddr == NEW_ADDR)
1654 map->m_flags |= F2FS_MAP_DELALLOC;
1655 map->m_flags |= F2FS_MAP_MAPPED;
1656
1657 map->m_pblk = blkaddr;
1658 map->m_len = 1;
1659
1660 if (map->m_multidev_dio)
1661 map->m_bdev = FDEV(bidx).bdev;
1662 } else if ((map->m_pblk != NEW_ADDR &&
1663 blkaddr == (map->m_pblk + ofs)) ||
1664 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1665 flag == F2FS_GET_BLOCK_PRE_DIO) {
1666 if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1667 goto sync_out;
1668 ofs++;
1669 map->m_len++;
1670 } else {
1671 goto sync_out;
1672 }
1673
1674 skip:
1675 dn.ofs_in_node++;
1676 pgofs++;
1677
1678 /* preallocate blocks in batch for one dnode page */
1679 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1680 (pgofs == end || dn.ofs_in_node == end_offset)) {
1681
1682 dn.ofs_in_node = ofs_in_node;
1683 err = f2fs_reserve_new_blocks(&dn, prealloc);
1684 if (err)
1685 goto sync_out;
1686
1687 map->m_len += dn.ofs_in_node - ofs_in_node;
1688 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1689 err = -ENOSPC;
1690 goto sync_out;
1691 }
1692 dn.ofs_in_node = end_offset;
1693 }
1694
1695 if (pgofs >= end)
1696 goto sync_out;
1697 else if (dn.ofs_in_node < end_offset)
1698 goto next_block;
1699
1700 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1701 if (map->m_flags & F2FS_MAP_MAPPED) {
1702 unsigned int ofs = start_pgofs - map->m_lblk;
1703
1704 f2fs_update_read_extent_cache_range(&dn,
1705 start_pgofs, map->m_pblk + ofs,
1706 map->m_len - ofs);
1707 }
1708 }
1709
1710 f2fs_put_dnode(&dn);
1711
1712 if (map->m_may_create) {
1713 f2fs_map_unlock(sbi, flag);
1714 f2fs_balance_fs(sbi, dn.node_changed);
1715 }
1716 goto next_dnode;
1717
1718 sync_out:
1719
1720 if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1721 /*
1722 * for hardware encryption, but to avoid potential issue
1723 * in future
1724 */
1725 f2fs_wait_on_block_writeback_range(inode,
1726 map->m_pblk, map->m_len);
1727
1728 if (map->m_multidev_dio) {
1729 block_t blk_addr = map->m_pblk;
1730
1731 bidx = f2fs_target_device_index(sbi, map->m_pblk);
1732
1733 map->m_bdev = FDEV(bidx).bdev;
1734 map->m_pblk -= FDEV(bidx).start_blk;
1735
1736 if (map->m_may_create)
1737 f2fs_update_device_state(sbi, inode->i_ino,
1738 blk_addr, map->m_len);
1739
1740 f2fs_bug_on(sbi, blk_addr + map->m_len >
1741 FDEV(bidx).end_blk + 1);
1742 }
1743 }
1744
1745 if (flag == F2FS_GET_BLOCK_PRECACHE) {
1746 if (map->m_flags & F2FS_MAP_MAPPED) {
1747 unsigned int ofs = start_pgofs - map->m_lblk;
1748
1749 f2fs_update_read_extent_cache_range(&dn,
1750 start_pgofs, map->m_pblk + ofs,
1751 map->m_len - ofs);
1752 }
1753 if (map->m_next_extent)
1754 *map->m_next_extent = pgofs + 1;
1755 }
1756 f2fs_put_dnode(&dn);
1757 unlock_out:
1758 if (map->m_may_create) {
1759 f2fs_map_unlock(sbi, flag);
1760 f2fs_balance_fs(sbi, dn.node_changed);
1761 }
1762 out:
1763 trace_f2fs_map_blocks(inode, map, flag, err);
1764 return err;
1765 }
1766
f2fs_overwrite_io(struct inode * inode,loff_t pos,size_t len)1767 bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1768 {
1769 struct f2fs_map_blocks map;
1770 block_t last_lblk;
1771 int err;
1772
1773 if (pos + len > i_size_read(inode))
1774 return false;
1775
1776 map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1777 map.m_next_pgofs = NULL;
1778 map.m_next_extent = NULL;
1779 map.m_seg_type = NO_CHECK_TYPE;
1780 map.m_may_create = false;
1781 last_lblk = F2FS_BLK_ALIGN(pos + len);
1782
1783 while (map.m_lblk < last_lblk) {
1784 map.m_len = last_lblk - map.m_lblk;
1785 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT);
1786 if (err || map.m_len == 0)
1787 return false;
1788 map.m_lblk += map.m_len;
1789 }
1790 return true;
1791 }
1792
bytes_to_blks(struct inode * inode,u64 bytes)1793 static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1794 {
1795 return (bytes >> inode->i_blkbits);
1796 }
1797
blks_to_bytes(struct inode * inode,u64 blks)1798 static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1799 {
1800 return (blks << inode->i_blkbits);
1801 }
1802
f2fs_xattr_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo)1803 static int f2fs_xattr_fiemap(struct inode *inode,
1804 struct fiemap_extent_info *fieinfo)
1805 {
1806 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1807 struct page *page;
1808 struct node_info ni;
1809 __u64 phys = 0, len;
1810 __u32 flags;
1811 nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1812 int err = 0;
1813
1814 if (f2fs_has_inline_xattr(inode)) {
1815 int offset;
1816
1817 page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1818 inode->i_ino, false);
1819 if (!page)
1820 return -ENOMEM;
1821
1822 err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1823 if (err) {
1824 f2fs_put_page(page, 1);
1825 return err;
1826 }
1827
1828 phys = blks_to_bytes(inode, ni.blk_addr);
1829 offset = offsetof(struct f2fs_inode, i_addr) +
1830 sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1831 get_inline_xattr_addrs(inode));
1832
1833 phys += offset;
1834 len = inline_xattr_size(inode);
1835
1836 f2fs_put_page(page, 1);
1837
1838 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1839
1840 if (!xnid)
1841 flags |= FIEMAP_EXTENT_LAST;
1842
1843 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1844 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1845 if (err)
1846 return err;
1847 }
1848
1849 if (xnid) {
1850 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1851 if (!page)
1852 return -ENOMEM;
1853
1854 err = f2fs_get_node_info(sbi, xnid, &ni, false);
1855 if (err) {
1856 f2fs_put_page(page, 1);
1857 return err;
1858 }
1859
1860 phys = blks_to_bytes(inode, ni.blk_addr);
1861 len = inode->i_sb->s_blocksize;
1862
1863 f2fs_put_page(page, 1);
1864
1865 flags = FIEMAP_EXTENT_LAST;
1866 }
1867
1868 if (phys) {
1869 err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1870 trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1871 }
1872
1873 return (err < 0 ? err : 0);
1874 }
1875
max_inode_blocks(struct inode * inode)1876 static loff_t max_inode_blocks(struct inode *inode)
1877 {
1878 loff_t result = ADDRS_PER_INODE(inode);
1879 loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1880
1881 /* two direct node blocks */
1882 result += (leaf_count * 2);
1883
1884 /* two indirect node blocks */
1885 leaf_count *= NIDS_PER_BLOCK;
1886 result += (leaf_count * 2);
1887
1888 /* one double indirect node block */
1889 leaf_count *= NIDS_PER_BLOCK;
1890 result += leaf_count;
1891
1892 return result;
1893 }
1894
f2fs_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 start,u64 len)1895 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1896 u64 start, u64 len)
1897 {
1898 struct f2fs_map_blocks map;
1899 sector_t start_blk, last_blk;
1900 pgoff_t next_pgofs;
1901 u64 logical = 0, phys = 0, size = 0;
1902 u32 flags = 0;
1903 int ret = 0;
1904 bool compr_cluster = false, compr_appended;
1905 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1906 unsigned int count_in_cluster = 0;
1907 loff_t maxbytes;
1908
1909 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1910 ret = f2fs_precache_extents(inode);
1911 if (ret)
1912 return ret;
1913 }
1914
1915 ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1916 if (ret)
1917 return ret;
1918
1919 inode_lock(inode);
1920
1921 maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1922 if (start > maxbytes) {
1923 ret = -EFBIG;
1924 goto out;
1925 }
1926
1927 if (len > maxbytes || (maxbytes - len) < start)
1928 len = maxbytes - start;
1929
1930 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1931 ret = f2fs_xattr_fiemap(inode, fieinfo);
1932 goto out;
1933 }
1934
1935 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
1936 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1937 if (ret != -EAGAIN)
1938 goto out;
1939 }
1940
1941 if (bytes_to_blks(inode, len) == 0)
1942 len = blks_to_bytes(inode, 1);
1943
1944 start_blk = bytes_to_blks(inode, start);
1945 last_blk = bytes_to_blks(inode, start + len - 1);
1946
1947 next:
1948 memset(&map, 0, sizeof(map));
1949 map.m_lblk = start_blk;
1950 map.m_len = bytes_to_blks(inode, len);
1951 map.m_next_pgofs = &next_pgofs;
1952 map.m_seg_type = NO_CHECK_TYPE;
1953
1954 if (compr_cluster) {
1955 map.m_lblk += 1;
1956 map.m_len = cluster_size - count_in_cluster;
1957 }
1958
1959 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
1960 if (ret)
1961 goto out;
1962
1963 /* HOLE */
1964 if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
1965 start_blk = next_pgofs;
1966
1967 if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
1968 max_inode_blocks(inode)))
1969 goto prep_next;
1970
1971 flags |= FIEMAP_EXTENT_LAST;
1972 }
1973
1974 compr_appended = false;
1975 /* In a case of compressed cluster, append this to the last extent */
1976 if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) ||
1977 !(map.m_flags & F2FS_MAP_FLAGS))) {
1978 compr_appended = true;
1979 goto skip_fill;
1980 }
1981
1982 if (size) {
1983 flags |= FIEMAP_EXTENT_MERGED;
1984 if (IS_ENCRYPTED(inode))
1985 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1986
1987 ret = fiemap_fill_next_extent(fieinfo, logical,
1988 phys, size, flags);
1989 trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
1990 if (ret)
1991 goto out;
1992 size = 0;
1993 }
1994
1995 if (start_blk > last_blk)
1996 goto out;
1997
1998 skip_fill:
1999 if (map.m_pblk == COMPRESS_ADDR) {
2000 compr_cluster = true;
2001 count_in_cluster = 1;
2002 } else if (compr_appended) {
2003 unsigned int appended_blks = cluster_size -
2004 count_in_cluster + 1;
2005 size += blks_to_bytes(inode, appended_blks);
2006 start_blk += appended_blks;
2007 compr_cluster = false;
2008 } else {
2009 logical = blks_to_bytes(inode, start_blk);
2010 phys = __is_valid_data_blkaddr(map.m_pblk) ?
2011 blks_to_bytes(inode, map.m_pblk) : 0;
2012 size = blks_to_bytes(inode, map.m_len);
2013 flags = 0;
2014
2015 if (compr_cluster) {
2016 flags = FIEMAP_EXTENT_ENCODED;
2017 count_in_cluster += map.m_len;
2018 if (count_in_cluster == cluster_size) {
2019 compr_cluster = false;
2020 size += blks_to_bytes(inode, 1);
2021 }
2022 } else if (map.m_flags & F2FS_MAP_DELALLOC) {
2023 flags = FIEMAP_EXTENT_UNWRITTEN;
2024 }
2025
2026 start_blk += bytes_to_blks(inode, size);
2027 }
2028
2029 prep_next:
2030 cond_resched();
2031 if (fatal_signal_pending(current))
2032 ret = -EINTR;
2033 else
2034 goto next;
2035 out:
2036 if (ret == 1)
2037 ret = 0;
2038
2039 inode_unlock(inode);
2040 return ret;
2041 }
2042
f2fs_readpage_limit(struct inode * inode)2043 static inline loff_t f2fs_readpage_limit(struct inode *inode)
2044 {
2045 if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
2046 return inode->i_sb->s_maxbytes;
2047
2048 return i_size_read(inode);
2049 }
2050
f2fs_read_single_page(struct inode * inode,struct page * page,unsigned nr_pages,struct f2fs_map_blocks * map,struct bio ** bio_ret,sector_t * last_block_in_bio,bool is_readahead)2051 static int f2fs_read_single_page(struct inode *inode, struct page *page,
2052 unsigned nr_pages,
2053 struct f2fs_map_blocks *map,
2054 struct bio **bio_ret,
2055 sector_t *last_block_in_bio,
2056 bool is_readahead)
2057 {
2058 struct bio *bio = *bio_ret;
2059 const unsigned blocksize = blks_to_bytes(inode, 1);
2060 sector_t block_in_file;
2061 sector_t last_block;
2062 sector_t last_block_in_file;
2063 sector_t block_nr;
2064 int ret = 0;
2065
2066 block_in_file = (sector_t)page_index(page);
2067 last_block = block_in_file + nr_pages;
2068 last_block_in_file = bytes_to_blks(inode,
2069 f2fs_readpage_limit(inode) + blocksize - 1);
2070 if (last_block > last_block_in_file)
2071 last_block = last_block_in_file;
2072
2073 /* just zeroing out page which is beyond EOF */
2074 if (block_in_file >= last_block)
2075 goto zero_out;
2076 /*
2077 * Map blocks using the previous result first.
2078 */
2079 if ((map->m_flags & F2FS_MAP_MAPPED) &&
2080 block_in_file > map->m_lblk &&
2081 block_in_file < (map->m_lblk + map->m_len))
2082 goto got_it;
2083
2084 /*
2085 * Then do more f2fs_map_blocks() calls until we are
2086 * done with this page.
2087 */
2088 map->m_lblk = block_in_file;
2089 map->m_len = last_block - block_in_file;
2090
2091 ret = f2fs_map_blocks(inode, map, F2FS_GET_BLOCK_DEFAULT);
2092 if (ret)
2093 goto out;
2094 got_it:
2095 if ((map->m_flags & F2FS_MAP_MAPPED)) {
2096 block_nr = map->m_pblk + block_in_file - map->m_lblk;
2097 SetPageMappedToDisk(page);
2098
2099 if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2100 DATA_GENERIC_ENHANCE_READ)) {
2101 ret = -EFSCORRUPTED;
2102 f2fs_handle_error(F2FS_I_SB(inode),
2103 ERROR_INVALID_BLKADDR);
2104 goto out;
2105 }
2106 } else {
2107 zero_out:
2108 zero_user_segment(page, 0, PAGE_SIZE);
2109 if (f2fs_need_verity(inode, page->index) &&
2110 !fsverity_verify_page(page)) {
2111 ret = -EIO;
2112 goto out;
2113 }
2114 if (!PageUptodate(page))
2115 SetPageUptodate(page);
2116 unlock_page(page);
2117 goto out;
2118 }
2119
2120 /*
2121 * This page will go to BIO. Do we need to send this
2122 * BIO off first?
2123 */
2124 if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2125 *last_block_in_bio, block_nr) ||
2126 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2127 submit_and_realloc:
2128 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2129 bio = NULL;
2130 }
2131 if (bio == NULL) {
2132 bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2133 is_readahead ? REQ_RAHEAD : 0, page->index,
2134 false);
2135 if (IS_ERR(bio)) {
2136 ret = PTR_ERR(bio);
2137 bio = NULL;
2138 goto out;
2139 }
2140 }
2141
2142 /*
2143 * If the page is under writeback, we need to wait for
2144 * its completion to see the correct decrypted data.
2145 */
2146 f2fs_wait_on_block_writeback(inode, block_nr);
2147
2148 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2149 goto submit_and_realloc;
2150
2151 inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2152 f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO,
2153 F2FS_BLKSIZE);
2154 *last_block_in_bio = block_nr;
2155 out:
2156 *bio_ret = bio;
2157 return ret;
2158 }
2159
2160 #ifdef CONFIG_F2FS_FS_COMPRESSION
f2fs_read_multi_pages(struct compress_ctx * cc,struct bio ** bio_ret,unsigned nr_pages,sector_t * last_block_in_bio,bool is_readahead,bool for_write)2161 int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2162 unsigned nr_pages, sector_t *last_block_in_bio,
2163 bool is_readahead, bool for_write)
2164 {
2165 struct dnode_of_data dn;
2166 struct inode *inode = cc->inode;
2167 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2168 struct bio *bio = *bio_ret;
2169 unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2170 sector_t last_block_in_file;
2171 const unsigned blocksize = blks_to_bytes(inode, 1);
2172 struct decompress_io_ctx *dic = NULL;
2173 struct extent_info ei = {};
2174 bool from_dnode = true;
2175 int i;
2176 int ret = 0;
2177
2178 f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2179
2180 last_block_in_file = bytes_to_blks(inode,
2181 f2fs_readpage_limit(inode) + blocksize - 1);
2182
2183 /* get rid of pages beyond EOF */
2184 for (i = 0; i < cc->cluster_size; i++) {
2185 struct page *page = cc->rpages[i];
2186
2187 if (!page)
2188 continue;
2189 if ((sector_t)page->index >= last_block_in_file) {
2190 zero_user_segment(page, 0, PAGE_SIZE);
2191 if (!PageUptodate(page))
2192 SetPageUptodate(page);
2193 } else if (!PageUptodate(page)) {
2194 continue;
2195 }
2196 unlock_page(page);
2197 if (for_write)
2198 put_page(page);
2199 cc->rpages[i] = NULL;
2200 cc->nr_rpages--;
2201 }
2202
2203 /* we are done since all pages are beyond EOF */
2204 if (f2fs_cluster_is_empty(cc))
2205 goto out;
2206
2207 if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei))
2208 from_dnode = false;
2209
2210 if (!from_dnode)
2211 goto skip_reading_dnode;
2212
2213 set_new_dnode(&dn, inode, NULL, NULL, 0);
2214 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2215 if (ret)
2216 goto out;
2217
2218 if (unlikely(f2fs_cp_error(sbi))) {
2219 ret = -EIO;
2220 goto out_put_dnode;
2221 }
2222 f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2223
2224 skip_reading_dnode:
2225 for (i = 1; i < cc->cluster_size; i++) {
2226 block_t blkaddr;
2227
2228 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2229 dn.ofs_in_node + i) :
2230 ei.blk + i - 1;
2231
2232 if (!__is_valid_data_blkaddr(blkaddr))
2233 break;
2234
2235 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2236 ret = -EFAULT;
2237 goto out_put_dnode;
2238 }
2239 cc->nr_cpages++;
2240
2241 if (!from_dnode && i >= ei.c_len)
2242 break;
2243 }
2244
2245 /* nothing to decompress */
2246 if (cc->nr_cpages == 0) {
2247 ret = 0;
2248 goto out_put_dnode;
2249 }
2250
2251 dic = f2fs_alloc_dic(cc);
2252 if (IS_ERR(dic)) {
2253 ret = PTR_ERR(dic);
2254 goto out_put_dnode;
2255 }
2256
2257 for (i = 0; i < cc->nr_cpages; i++) {
2258 struct page *page = dic->cpages[i];
2259 block_t blkaddr;
2260 struct bio_post_read_ctx *ctx;
2261
2262 blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2263 dn.ofs_in_node + i + 1) :
2264 ei.blk + i;
2265
2266 f2fs_wait_on_block_writeback(inode, blkaddr);
2267
2268 if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2269 if (atomic_dec_and_test(&dic->remaining_pages)) {
2270 f2fs_decompress_cluster(dic, true);
2271 break;
2272 }
2273 continue;
2274 }
2275
2276 if (bio && (!page_is_mergeable(sbi, bio,
2277 *last_block_in_bio, blkaddr) ||
2278 !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2279 submit_and_realloc:
2280 f2fs_submit_read_bio(sbi, bio, DATA);
2281 bio = NULL;
2282 }
2283
2284 if (!bio) {
2285 bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2286 is_readahead ? REQ_RAHEAD : 0,
2287 page->index, for_write);
2288 if (IS_ERR(bio)) {
2289 ret = PTR_ERR(bio);
2290 f2fs_decompress_end_io(dic, ret, true);
2291 f2fs_put_dnode(&dn);
2292 *bio_ret = NULL;
2293 return ret;
2294 }
2295 }
2296
2297 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2298 goto submit_and_realloc;
2299
2300 ctx = get_post_read_ctx(bio);
2301 ctx->enabled_steps |= STEP_DECOMPRESS;
2302 refcount_inc(&dic->refcnt);
2303
2304 inc_page_count(sbi, F2FS_RD_DATA);
2305 f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
2306 *last_block_in_bio = blkaddr;
2307 }
2308
2309 if (from_dnode)
2310 f2fs_put_dnode(&dn);
2311
2312 *bio_ret = bio;
2313 return 0;
2314
2315 out_put_dnode:
2316 if (from_dnode)
2317 f2fs_put_dnode(&dn);
2318 out:
2319 for (i = 0; i < cc->cluster_size; i++) {
2320 if (cc->rpages[i]) {
2321 ClearPageUptodate(cc->rpages[i]);
2322 unlock_page(cc->rpages[i]);
2323 }
2324 }
2325 *bio_ret = bio;
2326 return ret;
2327 }
2328 #endif
2329
2330 /*
2331 * This function was originally taken from fs/mpage.c, and customized for f2fs.
2332 * Major change was from block_size == page_size in f2fs by default.
2333 */
f2fs_mpage_readpages(struct inode * inode,struct readahead_control * rac,struct page * page)2334 static int f2fs_mpage_readpages(struct inode *inode,
2335 struct readahead_control *rac, struct page *page)
2336 {
2337 struct bio *bio = NULL;
2338 sector_t last_block_in_bio = 0;
2339 struct f2fs_map_blocks map;
2340 #ifdef CONFIG_F2FS_FS_COMPRESSION
2341 struct compress_ctx cc = {
2342 .inode = inode,
2343 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2344 .cluster_size = F2FS_I(inode)->i_cluster_size,
2345 .cluster_idx = NULL_CLUSTER,
2346 .rpages = NULL,
2347 .cpages = NULL,
2348 .nr_rpages = 0,
2349 .nr_cpages = 0,
2350 };
2351 pgoff_t nc_cluster_idx = NULL_CLUSTER;
2352 #endif
2353 unsigned nr_pages = rac ? readahead_count(rac) : 1;
2354 unsigned max_nr_pages = nr_pages;
2355 int ret = 0;
2356
2357 map.m_pblk = 0;
2358 map.m_lblk = 0;
2359 map.m_len = 0;
2360 map.m_flags = 0;
2361 map.m_next_pgofs = NULL;
2362 map.m_next_extent = NULL;
2363 map.m_seg_type = NO_CHECK_TYPE;
2364 map.m_may_create = false;
2365
2366 for (; nr_pages; nr_pages--) {
2367 if (rac) {
2368 page = readahead_page(rac);
2369 prefetchw(&page->flags);
2370 }
2371
2372 #ifdef CONFIG_F2FS_FS_COMPRESSION
2373 if (f2fs_compressed_file(inode)) {
2374 /* there are remained compressed pages, submit them */
2375 if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2376 ret = f2fs_read_multi_pages(&cc, &bio,
2377 max_nr_pages,
2378 &last_block_in_bio,
2379 rac != NULL, false);
2380 f2fs_destroy_compress_ctx(&cc, false);
2381 if (ret)
2382 goto set_error_page;
2383 }
2384 if (cc.cluster_idx == NULL_CLUSTER) {
2385 if (nc_cluster_idx ==
2386 page->index >> cc.log_cluster_size) {
2387 goto read_single_page;
2388 }
2389
2390 ret = f2fs_is_compressed_cluster(inode, page->index);
2391 if (ret < 0)
2392 goto set_error_page;
2393 else if (!ret) {
2394 nc_cluster_idx =
2395 page->index >> cc.log_cluster_size;
2396 goto read_single_page;
2397 }
2398
2399 nc_cluster_idx = NULL_CLUSTER;
2400 }
2401 ret = f2fs_init_compress_ctx(&cc);
2402 if (ret)
2403 goto set_error_page;
2404
2405 f2fs_compress_ctx_add_page(&cc, page);
2406
2407 goto next_page;
2408 }
2409 read_single_page:
2410 #endif
2411
2412 ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2413 &bio, &last_block_in_bio, rac);
2414 if (ret) {
2415 #ifdef CONFIG_F2FS_FS_COMPRESSION
2416 set_error_page:
2417 #endif
2418 zero_user_segment(page, 0, PAGE_SIZE);
2419 unlock_page(page);
2420 }
2421 #ifdef CONFIG_F2FS_FS_COMPRESSION
2422 next_page:
2423 #endif
2424 if (rac)
2425 put_page(page);
2426
2427 #ifdef CONFIG_F2FS_FS_COMPRESSION
2428 if (f2fs_compressed_file(inode)) {
2429 /* last page */
2430 if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2431 ret = f2fs_read_multi_pages(&cc, &bio,
2432 max_nr_pages,
2433 &last_block_in_bio,
2434 rac != NULL, false);
2435 f2fs_destroy_compress_ctx(&cc, false);
2436 }
2437 }
2438 #endif
2439 }
2440 if (bio)
2441 f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2442 return ret;
2443 }
2444
f2fs_read_data_folio(struct file * file,struct folio * folio)2445 static int f2fs_read_data_folio(struct file *file, struct folio *folio)
2446 {
2447 struct page *page = &folio->page;
2448 struct inode *inode = page_file_mapping(page)->host;
2449 int ret = -EAGAIN;
2450
2451 trace_f2fs_readpage(page, DATA);
2452
2453 if (!f2fs_is_compress_backend_ready(inode)) {
2454 unlock_page(page);
2455 return -EOPNOTSUPP;
2456 }
2457
2458 /* If the file has inline data, try to read it directly */
2459 if (f2fs_has_inline_data(inode))
2460 ret = f2fs_read_inline_data(inode, page);
2461 if (ret == -EAGAIN)
2462 ret = f2fs_mpage_readpages(inode, NULL, page);
2463 return ret;
2464 }
2465
f2fs_readahead(struct readahead_control * rac)2466 static void f2fs_readahead(struct readahead_control *rac)
2467 {
2468 struct inode *inode = rac->mapping->host;
2469
2470 trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2471
2472 if (!f2fs_is_compress_backend_ready(inode))
2473 return;
2474
2475 /* If the file has inline data, skip readahead */
2476 if (f2fs_has_inline_data(inode))
2477 return;
2478
2479 f2fs_mpage_readpages(inode, rac, NULL);
2480 }
2481
f2fs_encrypt_one_page(struct f2fs_io_info * fio)2482 int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2483 {
2484 struct inode *inode = fio->page->mapping->host;
2485 struct page *mpage, *page;
2486 gfp_t gfp_flags = GFP_NOFS;
2487
2488 if (!f2fs_encrypted_file(inode))
2489 return 0;
2490
2491 page = fio->compressed_page ? fio->compressed_page : fio->page;
2492
2493 if (fscrypt_inode_uses_inline_crypto(inode))
2494 return 0;
2495
2496 retry_encrypt:
2497 fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2498 PAGE_SIZE, 0, gfp_flags);
2499 if (IS_ERR(fio->encrypted_page)) {
2500 /* flush pending IOs and wait for a while in the ENOMEM case */
2501 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2502 f2fs_flush_merged_writes(fio->sbi);
2503 memalloc_retry_wait(GFP_NOFS);
2504 gfp_flags |= __GFP_NOFAIL;
2505 goto retry_encrypt;
2506 }
2507 return PTR_ERR(fio->encrypted_page);
2508 }
2509
2510 mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2511 if (mpage) {
2512 if (PageUptodate(mpage))
2513 memcpy(page_address(mpage),
2514 page_address(fio->encrypted_page), PAGE_SIZE);
2515 f2fs_put_page(mpage, 1);
2516 }
2517 return 0;
2518 }
2519
check_inplace_update_policy(struct inode * inode,struct f2fs_io_info * fio)2520 static inline bool check_inplace_update_policy(struct inode *inode,
2521 struct f2fs_io_info *fio)
2522 {
2523 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2524
2525 if (IS_F2FS_IPU_HONOR_OPU_WRITE(sbi) &&
2526 is_inode_flag_set(inode, FI_OPU_WRITE))
2527 return false;
2528 if (IS_F2FS_IPU_FORCE(sbi))
2529 return true;
2530 if (IS_F2FS_IPU_SSR(sbi) && f2fs_need_SSR(sbi))
2531 return true;
2532 if (IS_F2FS_IPU_UTIL(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util)
2533 return true;
2534 if (IS_F2FS_IPU_SSR_UTIL(sbi) && f2fs_need_SSR(sbi) &&
2535 utilization(sbi) > SM_I(sbi)->min_ipu_util)
2536 return true;
2537
2538 /*
2539 * IPU for rewrite async pages
2540 */
2541 if (IS_F2FS_IPU_ASYNC(sbi) && fio && fio->op == REQ_OP_WRITE &&
2542 !(fio->op_flags & REQ_SYNC) && !IS_ENCRYPTED(inode))
2543 return true;
2544
2545 /* this is only set during fdatasync */
2546 if (IS_F2FS_IPU_FSYNC(sbi) && is_inode_flag_set(inode, FI_NEED_IPU))
2547 return true;
2548
2549 if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2550 !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2551 return true;
2552
2553 return false;
2554 }
2555
f2fs_should_update_inplace(struct inode * inode,struct f2fs_io_info * fio)2556 bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2557 {
2558 /* swap file is migrating in aligned write mode */
2559 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2560 return false;
2561
2562 if (f2fs_is_pinned_file(inode))
2563 return true;
2564
2565 /* if this is cold file, we should overwrite to avoid fragmentation */
2566 if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE))
2567 return true;
2568
2569 return check_inplace_update_policy(inode, fio);
2570 }
2571
f2fs_should_update_outplace(struct inode * inode,struct f2fs_io_info * fio)2572 bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2573 {
2574 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2575
2576 /* The below cases were checked when setting it. */
2577 if (f2fs_is_pinned_file(inode))
2578 return false;
2579 if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2580 return true;
2581 if (f2fs_lfs_mode(sbi))
2582 return true;
2583 if (S_ISDIR(inode->i_mode))
2584 return true;
2585 if (IS_NOQUOTA(inode))
2586 return true;
2587 if (f2fs_used_in_atomic_write(inode))
2588 return true;
2589
2590 /* swap file is migrating in aligned write mode */
2591 if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2592 return true;
2593
2594 if (is_inode_flag_set(inode, FI_OPU_WRITE))
2595 return true;
2596
2597 if (fio) {
2598 if (page_private_gcing(fio->page))
2599 return true;
2600 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2601 f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2602 return true;
2603 }
2604 return false;
2605 }
2606
need_inplace_update(struct f2fs_io_info * fio)2607 static inline bool need_inplace_update(struct f2fs_io_info *fio)
2608 {
2609 struct inode *inode = fio->page->mapping->host;
2610
2611 if (f2fs_should_update_outplace(inode, fio))
2612 return false;
2613
2614 return f2fs_should_update_inplace(inode, fio);
2615 }
2616
f2fs_do_write_data_page(struct f2fs_io_info * fio)2617 int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2618 {
2619 struct page *page = fio->page;
2620 struct inode *inode = page->mapping->host;
2621 struct dnode_of_data dn;
2622 struct node_info ni;
2623 bool ipu_force = false;
2624 bool atomic_commit;
2625 int err = 0;
2626
2627 /* Use COW inode to make dnode_of_data for atomic write */
2628 atomic_commit = f2fs_is_atomic_file(inode) &&
2629 page_private_atomic(fio->page);
2630 if (atomic_commit)
2631 set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
2632 else
2633 set_new_dnode(&dn, inode, NULL, NULL, 0);
2634
2635 if (need_inplace_update(fio) &&
2636 f2fs_lookup_read_extent_cache_block(inode, page->index,
2637 &fio->old_blkaddr)) {
2638 if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2639 DATA_GENERIC_ENHANCE)) {
2640 f2fs_handle_error(fio->sbi,
2641 ERROR_INVALID_BLKADDR);
2642 return -EFSCORRUPTED;
2643 }
2644
2645 ipu_force = true;
2646 fio->need_lock = LOCK_DONE;
2647 goto got_it;
2648 }
2649
2650 /* Deadlock due to between page->lock and f2fs_lock_op */
2651 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2652 return -EAGAIN;
2653
2654 err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2655 if (err)
2656 goto out;
2657
2658 fio->old_blkaddr = dn.data_blkaddr;
2659
2660 /* This page is already truncated */
2661 if (fio->old_blkaddr == NULL_ADDR) {
2662 ClearPageUptodate(page);
2663 clear_page_private_gcing(page);
2664 goto out_writepage;
2665 }
2666 got_it:
2667 if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2668 !f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2669 DATA_GENERIC_ENHANCE)) {
2670 err = -EFSCORRUPTED;
2671 f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
2672 goto out_writepage;
2673 }
2674
2675 /* wait for GCed page writeback via META_MAPPING */
2676 if (fio->meta_gc)
2677 f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2678
2679 /*
2680 * If current allocation needs SSR,
2681 * it had better in-place writes for updated data.
2682 */
2683 if (ipu_force ||
2684 (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2685 need_inplace_update(fio))) {
2686 err = f2fs_encrypt_one_page(fio);
2687 if (err)
2688 goto out_writepage;
2689
2690 set_page_writeback(page);
2691 f2fs_put_dnode(&dn);
2692 if (fio->need_lock == LOCK_REQ)
2693 f2fs_unlock_op(fio->sbi);
2694 err = f2fs_inplace_write_data(fio);
2695 if (err) {
2696 if (fscrypt_inode_uses_fs_layer_crypto(inode))
2697 fscrypt_finalize_bounce_page(&fio->encrypted_page);
2698 if (PageWriteback(page))
2699 end_page_writeback(page);
2700 } else {
2701 set_inode_flag(inode, FI_UPDATE_WRITE);
2702 }
2703 trace_f2fs_do_write_data_page(fio->page, IPU);
2704 return err;
2705 }
2706
2707 if (fio->need_lock == LOCK_RETRY) {
2708 if (!f2fs_trylock_op(fio->sbi)) {
2709 err = -EAGAIN;
2710 goto out_writepage;
2711 }
2712 fio->need_lock = LOCK_REQ;
2713 }
2714
2715 err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2716 if (err)
2717 goto out_writepage;
2718
2719 fio->version = ni.version;
2720
2721 err = f2fs_encrypt_one_page(fio);
2722 if (err)
2723 goto out_writepage;
2724
2725 set_page_writeback(page);
2726
2727 if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2728 f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2729
2730 /* LFS mode write path */
2731 f2fs_outplace_write_data(&dn, fio);
2732 trace_f2fs_do_write_data_page(page, OPU);
2733 set_inode_flag(inode, FI_APPEND_WRITE);
2734 if (atomic_commit)
2735 clear_page_private_atomic(page);
2736 out_writepage:
2737 f2fs_put_dnode(&dn);
2738 out:
2739 if (fio->need_lock == LOCK_REQ)
2740 f2fs_unlock_op(fio->sbi);
2741 return err;
2742 }
2743
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)2744 int f2fs_write_single_data_page(struct page *page, int *submitted,
2745 struct bio **bio,
2746 sector_t *last_block,
2747 struct writeback_control *wbc,
2748 enum iostat_type io_type,
2749 int compr_blocks,
2750 bool allow_balance)
2751 {
2752 struct inode *inode = page->mapping->host;
2753 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2754 loff_t i_size = i_size_read(inode);
2755 const pgoff_t end_index = ((unsigned long long)i_size)
2756 >> PAGE_SHIFT;
2757 loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2758 unsigned offset = 0;
2759 bool need_balance_fs = false;
2760 bool quota_inode = IS_NOQUOTA(inode);
2761 int err = 0;
2762 struct f2fs_io_info fio = {
2763 .sbi = sbi,
2764 .ino = inode->i_ino,
2765 .type = DATA,
2766 .op = REQ_OP_WRITE,
2767 .op_flags = wbc_to_write_flags(wbc),
2768 .old_blkaddr = NULL_ADDR,
2769 .page = page,
2770 .encrypted_page = NULL,
2771 .submitted = 0,
2772 .compr_blocks = compr_blocks,
2773 .need_lock = compr_blocks ? LOCK_DONE : LOCK_RETRY,
2774 .meta_gc = f2fs_meta_inode_gc_required(inode) ? 1 : 0,
2775 .io_type = io_type,
2776 .io_wbc = wbc,
2777 .bio = bio,
2778 .last_block = last_block,
2779 };
2780
2781 trace_f2fs_writepage(page, DATA);
2782
2783 /* we should bypass data pages to proceed the kworker jobs */
2784 if (unlikely(f2fs_cp_error(sbi))) {
2785 mapping_set_error(page->mapping, -EIO);
2786 /*
2787 * don't drop any dirty dentry pages for keeping lastest
2788 * directory structure.
2789 */
2790 if (S_ISDIR(inode->i_mode) &&
2791 !is_sbi_flag_set(sbi, SBI_IS_CLOSE))
2792 goto redirty_out;
2793
2794 /* keep data pages in remount-ro mode */
2795 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
2796 goto redirty_out;
2797 goto out;
2798 }
2799
2800 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2801 goto redirty_out;
2802
2803 if (page->index < end_index ||
2804 f2fs_verity_in_progress(inode) ||
2805 compr_blocks)
2806 goto write;
2807
2808 /*
2809 * If the offset is out-of-range of file size,
2810 * this page does not have to be written to disk.
2811 */
2812 offset = i_size & (PAGE_SIZE - 1);
2813 if ((page->index >= end_index + 1) || !offset)
2814 goto out;
2815
2816 zero_user_segment(page, offset, PAGE_SIZE);
2817 write:
2818 /* Dentry/quota blocks are controlled by checkpoint */
2819 if (S_ISDIR(inode->i_mode) || quota_inode) {
2820 /*
2821 * We need to wait for node_write to avoid block allocation during
2822 * checkpoint. This can only happen to quota writes which can cause
2823 * the below discard race condition.
2824 */
2825 if (quota_inode)
2826 f2fs_down_read(&sbi->node_write);
2827
2828 fio.need_lock = LOCK_DONE;
2829 err = f2fs_do_write_data_page(&fio);
2830
2831 if (quota_inode)
2832 f2fs_up_read(&sbi->node_write);
2833
2834 goto done;
2835 }
2836
2837 if (!wbc->for_reclaim)
2838 need_balance_fs = true;
2839 else if (has_not_enough_free_secs(sbi, 0, 0))
2840 goto redirty_out;
2841 else
2842 set_inode_flag(inode, FI_HOT_DATA);
2843
2844 err = -EAGAIN;
2845 if (f2fs_has_inline_data(inode)) {
2846 err = f2fs_write_inline_data(inode, page);
2847 if (!err)
2848 goto out;
2849 }
2850
2851 if (err == -EAGAIN) {
2852 err = f2fs_do_write_data_page(&fio);
2853 if (err == -EAGAIN) {
2854 f2fs_bug_on(sbi, compr_blocks);
2855 fio.need_lock = LOCK_REQ;
2856 err = f2fs_do_write_data_page(&fio);
2857 }
2858 }
2859
2860 if (err) {
2861 file_set_keep_isize(inode);
2862 } else {
2863 spin_lock(&F2FS_I(inode)->i_size_lock);
2864 if (F2FS_I(inode)->last_disk_size < psize)
2865 F2FS_I(inode)->last_disk_size = psize;
2866 spin_unlock(&F2FS_I(inode)->i_size_lock);
2867 }
2868
2869 done:
2870 if (err && err != -ENOENT)
2871 goto redirty_out;
2872
2873 out:
2874 inode_dec_dirty_pages(inode);
2875 if (err) {
2876 ClearPageUptodate(page);
2877 clear_page_private_gcing(page);
2878 }
2879
2880 if (wbc->for_reclaim) {
2881 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2882 clear_inode_flag(inode, FI_HOT_DATA);
2883 f2fs_remove_dirty_inode(inode);
2884 submitted = NULL;
2885 }
2886 unlock_page(page);
2887 if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2888 !F2FS_I(inode)->wb_task && allow_balance)
2889 f2fs_balance_fs(sbi, need_balance_fs);
2890
2891 if (unlikely(f2fs_cp_error(sbi))) {
2892 f2fs_submit_merged_write(sbi, DATA);
2893 if (bio && *bio)
2894 f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2895 submitted = NULL;
2896 }
2897
2898 if (submitted)
2899 *submitted = fio.submitted;
2900
2901 return 0;
2902
2903 redirty_out:
2904 redirty_page_for_writepage(wbc, page);
2905 /*
2906 * pageout() in MM translates EAGAIN, so calls handle_write_error()
2907 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2908 * file_write_and_wait_range() will see EIO error, which is critical
2909 * to return value of fsync() followed by atomic_write failure to user.
2910 */
2911 if (!err || wbc->for_reclaim)
2912 return AOP_WRITEPAGE_ACTIVATE;
2913 unlock_page(page);
2914 return err;
2915 }
2916
f2fs_write_data_page(struct page * page,struct writeback_control * wbc)2917 static int f2fs_write_data_page(struct page *page,
2918 struct writeback_control *wbc)
2919 {
2920 #ifdef CONFIG_F2FS_FS_COMPRESSION
2921 struct inode *inode = page->mapping->host;
2922
2923 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2924 goto out;
2925
2926 if (f2fs_compressed_file(inode)) {
2927 if (f2fs_is_compressed_cluster(inode, page->index)) {
2928 redirty_page_for_writepage(wbc, page);
2929 return AOP_WRITEPAGE_ACTIVATE;
2930 }
2931 }
2932 out:
2933 #endif
2934
2935 return f2fs_write_single_data_page(page, NULL, NULL, NULL,
2936 wbc, FS_DATA_IO, 0, true);
2937 }
2938
2939 /*
2940 * This function was copied from write_cache_pages from mm/page-writeback.c.
2941 * The major change is making write step of cold data page separately from
2942 * warm/hot data page.
2943 */
f2fs_write_cache_pages(struct address_space * mapping,struct writeback_control * wbc,enum iostat_type io_type)2944 static int f2fs_write_cache_pages(struct address_space *mapping,
2945 struct writeback_control *wbc,
2946 enum iostat_type io_type)
2947 {
2948 int ret = 0;
2949 int done = 0, retry = 0;
2950 struct page *pages_local[F2FS_ONSTACK_PAGES];
2951 struct page **pages = pages_local;
2952 struct folio_batch fbatch;
2953 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2954 struct bio *bio = NULL;
2955 sector_t last_block;
2956 #ifdef CONFIG_F2FS_FS_COMPRESSION
2957 struct inode *inode = mapping->host;
2958 struct compress_ctx cc = {
2959 .inode = inode,
2960 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2961 .cluster_size = F2FS_I(inode)->i_cluster_size,
2962 .cluster_idx = NULL_CLUSTER,
2963 .rpages = NULL,
2964 .nr_rpages = 0,
2965 .cpages = NULL,
2966 .valid_nr_cpages = 0,
2967 .rbuf = NULL,
2968 .cbuf = NULL,
2969 .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
2970 .private = NULL,
2971 };
2972 #endif
2973 int nr_folios, p, idx;
2974 int nr_pages;
2975 unsigned int max_pages = F2FS_ONSTACK_PAGES;
2976 pgoff_t index;
2977 pgoff_t end; /* Inclusive */
2978 pgoff_t done_index;
2979 int range_whole = 0;
2980 xa_mark_t tag;
2981 int nwritten = 0;
2982 int submitted = 0;
2983 int i;
2984
2985 #ifdef CONFIG_F2FS_FS_COMPRESSION
2986 if (f2fs_compressed_file(inode) &&
2987 1 << cc.log_cluster_size > F2FS_ONSTACK_PAGES) {
2988 pages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
2989 cc.log_cluster_size, GFP_NOFS | __GFP_NOFAIL);
2990 max_pages = 1 << cc.log_cluster_size;
2991 }
2992 #endif
2993
2994 folio_batch_init(&fbatch);
2995
2996 if (get_dirty_pages(mapping->host) <=
2997 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
2998 set_inode_flag(mapping->host, FI_HOT_DATA);
2999 else
3000 clear_inode_flag(mapping->host, FI_HOT_DATA);
3001
3002 if (wbc->range_cyclic) {
3003 index = mapping->writeback_index; /* prev offset */
3004 end = -1;
3005 } else {
3006 index = wbc->range_start >> PAGE_SHIFT;
3007 end = wbc->range_end >> PAGE_SHIFT;
3008 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3009 range_whole = 1;
3010 }
3011 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3012 tag = PAGECACHE_TAG_TOWRITE;
3013 else
3014 tag = PAGECACHE_TAG_DIRTY;
3015 retry:
3016 retry = 0;
3017 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3018 tag_pages_for_writeback(mapping, index, end);
3019 done_index = index;
3020 while (!done && !retry && (index <= end)) {
3021 nr_pages = 0;
3022 again:
3023 nr_folios = filemap_get_folios_tag(mapping, &index, end,
3024 tag, &fbatch);
3025 if (nr_folios == 0) {
3026 if (nr_pages)
3027 goto write;
3028 break;
3029 }
3030
3031 for (i = 0; i < nr_folios; i++) {
3032 struct folio *folio = fbatch.folios[i];
3033
3034 idx = 0;
3035 p = folio_nr_pages(folio);
3036 add_more:
3037 pages[nr_pages] = folio_page(folio, idx);
3038 folio_get(folio);
3039 if (++nr_pages == max_pages) {
3040 index = folio->index + idx + 1;
3041 folio_batch_release(&fbatch);
3042 goto write;
3043 }
3044 if (++idx < p)
3045 goto add_more;
3046 }
3047 folio_batch_release(&fbatch);
3048 goto again;
3049 write:
3050 for (i = 0; i < nr_pages; i++) {
3051 struct page *page = pages[i];
3052 struct folio *folio = page_folio(page);
3053 bool need_readd;
3054 readd:
3055 need_readd = false;
3056 #ifdef CONFIG_F2FS_FS_COMPRESSION
3057 if (f2fs_compressed_file(inode)) {
3058 void *fsdata = NULL;
3059 struct page *pagep;
3060 int ret2;
3061
3062 ret = f2fs_init_compress_ctx(&cc);
3063 if (ret) {
3064 done = 1;
3065 break;
3066 }
3067
3068 if (!f2fs_cluster_can_merge_page(&cc,
3069 folio->index)) {
3070 ret = f2fs_write_multi_pages(&cc,
3071 &submitted, wbc, io_type);
3072 if (!ret)
3073 need_readd = true;
3074 goto result;
3075 }
3076
3077 if (unlikely(f2fs_cp_error(sbi)))
3078 goto lock_folio;
3079
3080 if (!f2fs_cluster_is_empty(&cc))
3081 goto lock_folio;
3082
3083 if (f2fs_all_cluster_page_ready(&cc,
3084 pages, i, nr_pages, true))
3085 goto lock_folio;
3086
3087 ret2 = f2fs_prepare_compress_overwrite(
3088 inode, &pagep,
3089 folio->index, &fsdata);
3090 if (ret2 < 0) {
3091 ret = ret2;
3092 done = 1;
3093 break;
3094 } else if (ret2 &&
3095 (!f2fs_compress_write_end(inode,
3096 fsdata, folio->index, 1) ||
3097 !f2fs_all_cluster_page_ready(&cc,
3098 pages, i, nr_pages,
3099 false))) {
3100 retry = 1;
3101 break;
3102 }
3103 }
3104 #endif
3105 /* give a priority to WB_SYNC threads */
3106 if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3107 wbc->sync_mode == WB_SYNC_NONE) {
3108 done = 1;
3109 break;
3110 }
3111 #ifdef CONFIG_F2FS_FS_COMPRESSION
3112 lock_folio:
3113 #endif
3114 done_index = folio->index;
3115 retry_write:
3116 folio_lock(folio);
3117
3118 if (unlikely(folio->mapping != mapping)) {
3119 continue_unlock:
3120 folio_unlock(folio);
3121 continue;
3122 }
3123
3124 if (!folio_test_dirty(folio)) {
3125 /* someone wrote it for us */
3126 goto continue_unlock;
3127 }
3128
3129 if (folio_test_writeback(folio)) {
3130 if (wbc->sync_mode == WB_SYNC_NONE)
3131 goto continue_unlock;
3132 f2fs_wait_on_page_writeback(&folio->page, DATA, true, true);
3133 }
3134
3135 if (!folio_clear_dirty_for_io(folio))
3136 goto continue_unlock;
3137
3138 #ifdef CONFIG_F2FS_FS_COMPRESSION
3139 if (f2fs_compressed_file(inode)) {
3140 folio_get(folio);
3141 f2fs_compress_ctx_add_page(&cc, &folio->page);
3142 continue;
3143 }
3144 #endif
3145 ret = f2fs_write_single_data_page(&folio->page,
3146 &submitted, &bio, &last_block,
3147 wbc, io_type, 0, true);
3148 if (ret == AOP_WRITEPAGE_ACTIVATE)
3149 folio_unlock(folio);
3150 #ifdef CONFIG_F2FS_FS_COMPRESSION
3151 result:
3152 #endif
3153 nwritten += submitted;
3154 wbc->nr_to_write -= submitted;
3155
3156 if (unlikely(ret)) {
3157 /*
3158 * keep nr_to_write, since vfs uses this to
3159 * get # of written pages.
3160 */
3161 if (ret == AOP_WRITEPAGE_ACTIVATE) {
3162 ret = 0;
3163 goto next;
3164 } else if (ret == -EAGAIN) {
3165 ret = 0;
3166 if (wbc->sync_mode == WB_SYNC_ALL) {
3167 f2fs_io_schedule_timeout(
3168 DEFAULT_IO_TIMEOUT);
3169 goto retry_write;
3170 }
3171 goto next;
3172 }
3173 done_index = folio_next_index(folio);
3174 done = 1;
3175 break;
3176 }
3177
3178 if (wbc->nr_to_write <= 0 &&
3179 wbc->sync_mode == WB_SYNC_NONE) {
3180 done = 1;
3181 break;
3182 }
3183 next:
3184 if (need_readd)
3185 goto readd;
3186 }
3187 release_pages(pages, nr_pages);
3188 cond_resched();
3189 }
3190 #ifdef CONFIG_F2FS_FS_COMPRESSION
3191 /* flush remained pages in compress cluster */
3192 if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3193 ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3194 nwritten += submitted;
3195 wbc->nr_to_write -= submitted;
3196 if (ret) {
3197 done = 1;
3198 retry = 0;
3199 }
3200 }
3201 if (f2fs_compressed_file(inode))
3202 f2fs_destroy_compress_ctx(&cc, false);
3203 #endif
3204 if (retry) {
3205 index = 0;
3206 end = -1;
3207 goto retry;
3208 }
3209 if (wbc->range_cyclic && !done)
3210 done_index = 0;
3211 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3212 mapping->writeback_index = done_index;
3213
3214 if (nwritten)
3215 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3216 NULL, 0, DATA);
3217 /* submit cached bio of IPU write */
3218 if (bio)
3219 f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3220
3221 #ifdef CONFIG_F2FS_FS_COMPRESSION
3222 if (pages != pages_local)
3223 kfree(pages);
3224 #endif
3225
3226 return ret;
3227 }
3228
__should_serialize_io(struct inode * inode,struct writeback_control * wbc)3229 static inline bool __should_serialize_io(struct inode *inode,
3230 struct writeback_control *wbc)
3231 {
3232 /* to avoid deadlock in path of data flush */
3233 if (F2FS_I(inode)->wb_task)
3234 return false;
3235
3236 if (!S_ISREG(inode->i_mode))
3237 return false;
3238 if (IS_NOQUOTA(inode))
3239 return false;
3240
3241 if (f2fs_need_compress_data(inode))
3242 return true;
3243 if (wbc->sync_mode != WB_SYNC_ALL)
3244 return true;
3245 if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3246 return true;
3247 return false;
3248 }
3249
__f2fs_write_data_pages(struct address_space * mapping,struct writeback_control * wbc,enum iostat_type io_type)3250 static int __f2fs_write_data_pages(struct address_space *mapping,
3251 struct writeback_control *wbc,
3252 enum iostat_type io_type)
3253 {
3254 struct inode *inode = mapping->host;
3255 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3256 struct blk_plug plug;
3257 int ret;
3258 bool locked = false;
3259
3260 /* deal with chardevs and other special file */
3261 if (!mapping->a_ops->writepage)
3262 return 0;
3263
3264 /* skip writing if there is no dirty page in this inode */
3265 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3266 return 0;
3267
3268 /* during POR, we don't need to trigger writepage at all. */
3269 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3270 goto skip_write;
3271
3272 if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3273 wbc->sync_mode == WB_SYNC_NONE &&
3274 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3275 f2fs_available_free_memory(sbi, DIRTY_DENTS))
3276 goto skip_write;
3277
3278 /* skip writing in file defragment preparing stage */
3279 if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3280 goto skip_write;
3281
3282 trace_f2fs_writepages(mapping->host, wbc, DATA);
3283
3284 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3285 if (wbc->sync_mode == WB_SYNC_ALL)
3286 atomic_inc(&sbi->wb_sync_req[DATA]);
3287 else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3288 /* to avoid potential deadlock */
3289 if (current->plug)
3290 blk_finish_plug(current->plug);
3291 goto skip_write;
3292 }
3293
3294 if (__should_serialize_io(inode, wbc)) {
3295 mutex_lock(&sbi->writepages);
3296 locked = true;
3297 }
3298
3299 blk_start_plug(&plug);
3300 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3301 blk_finish_plug(&plug);
3302
3303 if (locked)
3304 mutex_unlock(&sbi->writepages);
3305
3306 if (wbc->sync_mode == WB_SYNC_ALL)
3307 atomic_dec(&sbi->wb_sync_req[DATA]);
3308 /*
3309 * if some pages were truncated, we cannot guarantee its mapping->host
3310 * to detect pending bios.
3311 */
3312
3313 f2fs_remove_dirty_inode(inode);
3314 return ret;
3315
3316 skip_write:
3317 wbc->pages_skipped += get_dirty_pages(inode);
3318 trace_f2fs_writepages(mapping->host, wbc, DATA);
3319 return 0;
3320 }
3321
f2fs_write_data_pages(struct address_space * mapping,struct writeback_control * wbc)3322 static int f2fs_write_data_pages(struct address_space *mapping,
3323 struct writeback_control *wbc)
3324 {
3325 struct inode *inode = mapping->host;
3326
3327 return __f2fs_write_data_pages(mapping, wbc,
3328 F2FS_I(inode)->cp_task == current ?
3329 FS_CP_DATA_IO : FS_DATA_IO);
3330 }
3331
f2fs_write_failed(struct inode * inode,loff_t to)3332 void f2fs_write_failed(struct inode *inode, loff_t to)
3333 {
3334 loff_t i_size = i_size_read(inode);
3335
3336 if (IS_NOQUOTA(inode))
3337 return;
3338
3339 /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3340 if (to > i_size && !f2fs_verity_in_progress(inode)) {
3341 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3342 filemap_invalidate_lock(inode->i_mapping);
3343
3344 truncate_pagecache(inode, i_size);
3345 f2fs_truncate_blocks(inode, i_size, true);
3346
3347 filemap_invalidate_unlock(inode->i_mapping);
3348 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3349 }
3350 }
3351
prepare_write_begin(struct f2fs_sb_info * sbi,struct page * page,loff_t pos,unsigned len,block_t * blk_addr,bool * node_changed)3352 static int prepare_write_begin(struct f2fs_sb_info *sbi,
3353 struct page *page, loff_t pos, unsigned len,
3354 block_t *blk_addr, bool *node_changed)
3355 {
3356 struct inode *inode = page->mapping->host;
3357 pgoff_t index = page->index;
3358 struct dnode_of_data dn;
3359 struct page *ipage;
3360 bool locked = false;
3361 int flag = F2FS_GET_BLOCK_PRE_AIO;
3362 int err = 0;
3363
3364 /*
3365 * If a whole page is being written and we already preallocated all the
3366 * blocks, then there is no need to get a block address now.
3367 */
3368 if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3369 return 0;
3370
3371 /* f2fs_lock_op avoids race between write CP and convert_inline_page */
3372 if (f2fs_has_inline_data(inode)) {
3373 if (pos + len > MAX_INLINE_DATA(inode))
3374 flag = F2FS_GET_BLOCK_DEFAULT;
3375 f2fs_map_lock(sbi, flag);
3376 locked = true;
3377 } else if ((pos & PAGE_MASK) >= i_size_read(inode)) {
3378 f2fs_map_lock(sbi, flag);
3379 locked = true;
3380 }
3381
3382 restart:
3383 /* check inline_data */
3384 ipage = f2fs_get_node_page(sbi, inode->i_ino);
3385 if (IS_ERR(ipage)) {
3386 err = PTR_ERR(ipage);
3387 goto unlock_out;
3388 }
3389
3390 set_new_dnode(&dn, inode, ipage, ipage, 0);
3391
3392 if (f2fs_has_inline_data(inode)) {
3393 if (pos + len <= MAX_INLINE_DATA(inode)) {
3394 f2fs_do_read_inline_data(page, ipage);
3395 set_inode_flag(inode, FI_DATA_EXIST);
3396 if (inode->i_nlink)
3397 set_page_private_inline(ipage);
3398 goto out;
3399 }
3400 err = f2fs_convert_inline_page(&dn, page);
3401 if (err || dn.data_blkaddr != NULL_ADDR)
3402 goto out;
3403 }
3404
3405 if (!f2fs_lookup_read_extent_cache_block(inode, index,
3406 &dn.data_blkaddr)) {
3407 if (locked) {
3408 err = f2fs_reserve_block(&dn, index);
3409 goto out;
3410 }
3411
3412 /* hole case */
3413 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3414 if (!err && dn.data_blkaddr != NULL_ADDR)
3415 goto out;
3416 f2fs_put_dnode(&dn);
3417 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3418 WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3419 locked = true;
3420 goto restart;
3421 }
3422 out:
3423 if (!err) {
3424 /* convert_inline_page can make node_changed */
3425 *blk_addr = dn.data_blkaddr;
3426 *node_changed = dn.node_changed;
3427 }
3428 f2fs_put_dnode(&dn);
3429 unlock_out:
3430 if (locked)
3431 f2fs_map_unlock(sbi, flag);
3432 return err;
3433 }
3434
__find_data_block(struct inode * inode,pgoff_t index,block_t * blk_addr)3435 static int __find_data_block(struct inode *inode, pgoff_t index,
3436 block_t *blk_addr)
3437 {
3438 struct dnode_of_data dn;
3439 struct page *ipage;
3440 int err = 0;
3441
3442 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
3443 if (IS_ERR(ipage))
3444 return PTR_ERR(ipage);
3445
3446 set_new_dnode(&dn, inode, ipage, ipage, 0);
3447
3448 if (!f2fs_lookup_read_extent_cache_block(inode, index,
3449 &dn.data_blkaddr)) {
3450 /* hole case */
3451 err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3452 if (err) {
3453 dn.data_blkaddr = NULL_ADDR;
3454 err = 0;
3455 }
3456 }
3457 *blk_addr = dn.data_blkaddr;
3458 f2fs_put_dnode(&dn);
3459 return err;
3460 }
3461
__reserve_data_block(struct inode * inode,pgoff_t index,block_t * blk_addr,bool * node_changed)3462 static int __reserve_data_block(struct inode *inode, pgoff_t index,
3463 block_t *blk_addr, bool *node_changed)
3464 {
3465 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3466 struct dnode_of_data dn;
3467 struct page *ipage;
3468 int err = 0;
3469
3470 f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3471
3472 ipage = f2fs_get_node_page(sbi, inode->i_ino);
3473 if (IS_ERR(ipage)) {
3474 err = PTR_ERR(ipage);
3475 goto unlock_out;
3476 }
3477 set_new_dnode(&dn, inode, ipage, ipage, 0);
3478
3479 if (!f2fs_lookup_read_extent_cache_block(dn.inode, index,
3480 &dn.data_blkaddr))
3481 err = f2fs_reserve_block(&dn, index);
3482
3483 *blk_addr = dn.data_blkaddr;
3484 *node_changed = dn.node_changed;
3485 f2fs_put_dnode(&dn);
3486
3487 unlock_out:
3488 f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3489 return err;
3490 }
3491
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)3492 static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
3493 struct page *page, loff_t pos, unsigned int len,
3494 block_t *blk_addr, bool *node_changed, bool *use_cow)
3495 {
3496 struct inode *inode = page->mapping->host;
3497 struct inode *cow_inode = F2FS_I(inode)->cow_inode;
3498 pgoff_t index = page->index;
3499 int err = 0;
3500 block_t ori_blk_addr = NULL_ADDR;
3501
3502 /* If pos is beyond the end of file, reserve a new block in COW inode */
3503 if ((pos & PAGE_MASK) >= i_size_read(inode))
3504 goto reserve_block;
3505
3506 /* Look for the block in COW inode first */
3507 err = __find_data_block(cow_inode, index, blk_addr);
3508 if (err) {
3509 return err;
3510 } else if (*blk_addr != NULL_ADDR) {
3511 *use_cow = true;
3512 return 0;
3513 }
3514
3515 if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE))
3516 goto reserve_block;
3517
3518 /* Look for the block in the original inode */
3519 err = __find_data_block(inode, index, &ori_blk_addr);
3520 if (err)
3521 return err;
3522
3523 reserve_block:
3524 /* Finally, we should reserve a new block in COW inode for the update */
3525 err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
3526 if (err)
3527 return err;
3528 inc_atomic_write_cnt(inode);
3529
3530 if (ori_blk_addr != NULL_ADDR)
3531 *blk_addr = ori_blk_addr;
3532 return 0;
3533 }
3534
f2fs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct page ** pagep,void ** fsdata)3535 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3536 loff_t pos, unsigned len, struct page **pagep, void **fsdata)
3537 {
3538 struct inode *inode = mapping->host;
3539 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3540 struct page *page = NULL;
3541 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3542 bool need_balance = false;
3543 bool use_cow = false;
3544 block_t blkaddr = NULL_ADDR;
3545 int err = 0;
3546
3547 trace_f2fs_write_begin(inode, pos, len);
3548
3549 if (!f2fs_is_checkpoint_ready(sbi)) {
3550 err = -ENOSPC;
3551 goto fail;
3552 }
3553
3554 /*
3555 * We should check this at this moment to avoid deadlock on inode page
3556 * and #0 page. The locking rule for inline_data conversion should be:
3557 * lock_page(page #0) -> lock_page(inode_page)
3558 */
3559 if (index != 0) {
3560 err = f2fs_convert_inline_inode(inode);
3561 if (err)
3562 goto fail;
3563 }
3564
3565 #ifdef CONFIG_F2FS_FS_COMPRESSION
3566 if (f2fs_compressed_file(inode)) {
3567 int ret;
3568
3569 *fsdata = NULL;
3570
3571 if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3572 goto repeat;
3573
3574 ret = f2fs_prepare_compress_overwrite(inode, pagep,
3575 index, fsdata);
3576 if (ret < 0) {
3577 err = ret;
3578 goto fail;
3579 } else if (ret) {
3580 return 0;
3581 }
3582 }
3583 #endif
3584
3585 repeat:
3586 /*
3587 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3588 * wait_for_stable_page. Will wait that below with our IO control.
3589 */
3590 page = f2fs_pagecache_get_page(mapping, index,
3591 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3592 if (!page) {
3593 err = -ENOMEM;
3594 goto fail;
3595 }
3596
3597 /* TODO: cluster can be compressed due to race with .writepage */
3598
3599 *pagep = page;
3600
3601 if (f2fs_is_atomic_file(inode))
3602 err = prepare_atomic_write_begin(sbi, page, pos, len,
3603 &blkaddr, &need_balance, &use_cow);
3604 else
3605 err = prepare_write_begin(sbi, page, pos, len,
3606 &blkaddr, &need_balance);
3607 if (err)
3608 goto fail;
3609
3610 if (need_balance && !IS_NOQUOTA(inode) &&
3611 has_not_enough_free_secs(sbi, 0, 0)) {
3612 unlock_page(page);
3613 f2fs_balance_fs(sbi, true);
3614 lock_page(page);
3615 if (page->mapping != mapping) {
3616 /* The page got truncated from under us */
3617 f2fs_put_page(page, 1);
3618 goto repeat;
3619 }
3620 }
3621
3622 f2fs_wait_on_page_writeback(page, DATA, false, true);
3623
3624 if (len == PAGE_SIZE || PageUptodate(page))
3625 return 0;
3626
3627 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3628 !f2fs_verity_in_progress(inode)) {
3629 zero_user_segment(page, len, PAGE_SIZE);
3630 return 0;
3631 }
3632
3633 if (blkaddr == NEW_ADDR) {
3634 zero_user_segment(page, 0, PAGE_SIZE);
3635 SetPageUptodate(page);
3636 } else {
3637 if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3638 DATA_GENERIC_ENHANCE_READ)) {
3639 err = -EFSCORRUPTED;
3640 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
3641 goto fail;
3642 }
3643 err = f2fs_submit_page_read(use_cow ?
3644 F2FS_I(inode)->cow_inode : inode, page,
3645 blkaddr, 0, true);
3646 if (err)
3647 goto fail;
3648
3649 lock_page(page);
3650 if (unlikely(page->mapping != mapping)) {
3651 f2fs_put_page(page, 1);
3652 goto repeat;
3653 }
3654 if (unlikely(!PageUptodate(page))) {
3655 err = -EIO;
3656 goto fail;
3657 }
3658 }
3659 return 0;
3660
3661 fail:
3662 f2fs_put_page(page, 1);
3663 f2fs_write_failed(inode, pos + len);
3664 return err;
3665 }
3666
f2fs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)3667 static int f2fs_write_end(struct file *file,
3668 struct address_space *mapping,
3669 loff_t pos, unsigned len, unsigned copied,
3670 struct page *page, void *fsdata)
3671 {
3672 struct inode *inode = page->mapping->host;
3673
3674 trace_f2fs_write_end(inode, pos, len, copied);
3675
3676 /*
3677 * This should be come from len == PAGE_SIZE, and we expect copied
3678 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3679 * let generic_perform_write() try to copy data again through copied=0.
3680 */
3681 if (!PageUptodate(page)) {
3682 if (unlikely(copied != len))
3683 copied = 0;
3684 else
3685 SetPageUptodate(page);
3686 }
3687
3688 #ifdef CONFIG_F2FS_FS_COMPRESSION
3689 /* overwrite compressed file */
3690 if (f2fs_compressed_file(inode) && fsdata) {
3691 f2fs_compress_write_end(inode, fsdata, page->index, copied);
3692 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3693
3694 if (pos + copied > i_size_read(inode) &&
3695 !f2fs_verity_in_progress(inode))
3696 f2fs_i_size_write(inode, pos + copied);
3697 return copied;
3698 }
3699 #endif
3700
3701 if (!copied)
3702 goto unlock_out;
3703
3704 set_page_dirty(page);
3705
3706 if (f2fs_is_atomic_file(inode))
3707 set_page_private_atomic(page);
3708
3709 if (pos + copied > i_size_read(inode) &&
3710 !f2fs_verity_in_progress(inode)) {
3711 f2fs_i_size_write(inode, pos + copied);
3712 if (f2fs_is_atomic_file(inode))
3713 f2fs_i_size_write(F2FS_I(inode)->cow_inode,
3714 pos + copied);
3715 }
3716 unlock_out:
3717 f2fs_put_page(page, 1);
3718 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3719 return copied;
3720 }
3721
f2fs_invalidate_folio(struct folio * folio,size_t offset,size_t length)3722 void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3723 {
3724 struct inode *inode = folio->mapping->host;
3725 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3726
3727 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3728 (offset || length != folio_size(folio)))
3729 return;
3730
3731 if (folio_test_dirty(folio)) {
3732 if (inode->i_ino == F2FS_META_INO(sbi)) {
3733 dec_page_count(sbi, F2FS_DIRTY_META);
3734 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3735 dec_page_count(sbi, F2FS_DIRTY_NODES);
3736 } else {
3737 inode_dec_dirty_pages(inode);
3738 f2fs_remove_dirty_inode(inode);
3739 }
3740 }
3741 clear_page_private_all(&folio->page);
3742 }
3743
f2fs_release_folio(struct folio * folio,gfp_t wait)3744 bool f2fs_release_folio(struct folio *folio, gfp_t wait)
3745 {
3746 /* If this is dirty folio, keep private data */
3747 if (folio_test_dirty(folio))
3748 return false;
3749
3750 clear_page_private_all(&folio->page);
3751 return true;
3752 }
3753
f2fs_dirty_data_folio(struct address_space * mapping,struct folio * folio)3754 static bool f2fs_dirty_data_folio(struct address_space *mapping,
3755 struct folio *folio)
3756 {
3757 struct inode *inode = mapping->host;
3758
3759 trace_f2fs_set_page_dirty(&folio->page, DATA);
3760
3761 if (!folio_test_uptodate(folio))
3762 folio_mark_uptodate(folio);
3763 BUG_ON(folio_test_swapcache(folio));
3764
3765 if (filemap_dirty_folio(mapping, folio)) {
3766 f2fs_update_dirty_folio(inode, folio);
3767 return true;
3768 }
3769 return false;
3770 }
3771
3772
f2fs_bmap_compress(struct inode * inode,sector_t block)3773 static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3774 {
3775 #ifdef CONFIG_F2FS_FS_COMPRESSION
3776 struct dnode_of_data dn;
3777 sector_t start_idx, blknr = 0;
3778 int ret;
3779
3780 start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3781
3782 set_new_dnode(&dn, inode, NULL, NULL, 0);
3783 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3784 if (ret)
3785 return 0;
3786
3787 if (dn.data_blkaddr != COMPRESS_ADDR) {
3788 dn.ofs_in_node += block - start_idx;
3789 blknr = f2fs_data_blkaddr(&dn);
3790 if (!__is_valid_data_blkaddr(blknr))
3791 blknr = 0;
3792 }
3793
3794 f2fs_put_dnode(&dn);
3795 return blknr;
3796 #else
3797 return 0;
3798 #endif
3799 }
3800
3801
f2fs_bmap(struct address_space * mapping,sector_t block)3802 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3803 {
3804 struct inode *inode = mapping->host;
3805 sector_t blknr = 0;
3806
3807 if (f2fs_has_inline_data(inode))
3808 goto out;
3809
3810 /* make sure allocating whole blocks */
3811 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3812 filemap_write_and_wait(mapping);
3813
3814 /* Block number less than F2FS MAX BLOCKS */
3815 if (unlikely(block >= max_file_blocks(inode)))
3816 goto out;
3817
3818 if (f2fs_compressed_file(inode)) {
3819 blknr = f2fs_bmap_compress(inode, block);
3820 } else {
3821 struct f2fs_map_blocks map;
3822
3823 memset(&map, 0, sizeof(map));
3824 map.m_lblk = block;
3825 map.m_len = 1;
3826 map.m_next_pgofs = NULL;
3827 map.m_seg_type = NO_CHECK_TYPE;
3828
3829 if (!f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_BMAP))
3830 blknr = map.m_pblk;
3831 }
3832 out:
3833 trace_f2fs_bmap(inode, block, blknr);
3834 return blknr;
3835 }
3836
3837 #ifdef CONFIG_SWAP
f2fs_migrate_blocks(struct inode * inode,block_t start_blk,unsigned int blkcnt)3838 static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3839 unsigned int blkcnt)
3840 {
3841 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3842 unsigned int blkofs;
3843 unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3844 unsigned int secidx = start_blk / blk_per_sec;
3845 unsigned int end_sec;
3846 int ret = 0;
3847
3848 if (!blkcnt)
3849 return 0;
3850 end_sec = secidx + (blkcnt - 1) / blk_per_sec;
3851
3852 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3853 filemap_invalidate_lock(inode->i_mapping);
3854
3855 set_inode_flag(inode, FI_ALIGNED_WRITE);
3856 set_inode_flag(inode, FI_OPU_WRITE);
3857
3858 for (; secidx <= end_sec; secidx++) {
3859 unsigned int blkofs_end = secidx == end_sec ?
3860 (blkcnt - 1) % blk_per_sec : blk_per_sec - 1;
3861
3862 f2fs_down_write(&sbi->pin_sem);
3863
3864 ret = f2fs_allocate_pinning_section(sbi);
3865 if (ret) {
3866 f2fs_up_write(&sbi->pin_sem);
3867 break;
3868 }
3869
3870 set_inode_flag(inode, FI_SKIP_WRITES);
3871
3872 for (blkofs = 0; blkofs <= blkofs_end; blkofs++) {
3873 struct page *page;
3874 unsigned int blkidx = secidx * blk_per_sec + blkofs;
3875
3876 page = f2fs_get_lock_data_page(inode, blkidx, true);
3877 if (IS_ERR(page)) {
3878 f2fs_up_write(&sbi->pin_sem);
3879 ret = PTR_ERR(page);
3880 goto done;
3881 }
3882
3883 set_page_dirty(page);
3884 f2fs_put_page(page, 1);
3885 }
3886
3887 clear_inode_flag(inode, FI_SKIP_WRITES);
3888
3889 ret = filemap_fdatawrite(inode->i_mapping);
3890
3891 f2fs_up_write(&sbi->pin_sem);
3892
3893 if (ret)
3894 break;
3895 }
3896
3897 done:
3898 clear_inode_flag(inode, FI_SKIP_WRITES);
3899 clear_inode_flag(inode, FI_OPU_WRITE);
3900 clear_inode_flag(inode, FI_ALIGNED_WRITE);
3901
3902 filemap_invalidate_unlock(inode->i_mapping);
3903 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3904
3905 return ret;
3906 }
3907
check_swap_activate(struct swap_info_struct * sis,struct file * swap_file,sector_t * span)3908 static int check_swap_activate(struct swap_info_struct *sis,
3909 struct file *swap_file, sector_t *span)
3910 {
3911 struct address_space *mapping = swap_file->f_mapping;
3912 struct inode *inode = mapping->host;
3913 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3914 block_t cur_lblock;
3915 block_t last_lblock;
3916 block_t pblock;
3917 block_t lowest_pblock = -1;
3918 block_t highest_pblock = 0;
3919 int nr_extents = 0;
3920 unsigned int nr_pblocks;
3921 unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3922 unsigned int not_aligned = 0;
3923 int ret = 0;
3924
3925 /*
3926 * Map all the blocks into the extent list. This code doesn't try
3927 * to be very smart.
3928 */
3929 cur_lblock = 0;
3930 last_lblock = bytes_to_blks(inode, i_size_read(inode));
3931
3932 while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3933 struct f2fs_map_blocks map;
3934 retry:
3935 cond_resched();
3936
3937 memset(&map, 0, sizeof(map));
3938 map.m_lblk = cur_lblock;
3939 map.m_len = last_lblock - cur_lblock;
3940 map.m_next_pgofs = NULL;
3941 map.m_next_extent = NULL;
3942 map.m_seg_type = NO_CHECK_TYPE;
3943 map.m_may_create = false;
3944
3945 ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
3946 if (ret)
3947 goto out;
3948
3949 /* hole */
3950 if (!(map.m_flags & F2FS_MAP_FLAGS)) {
3951 f2fs_err(sbi, "Swapfile has holes");
3952 ret = -EINVAL;
3953 goto out;
3954 }
3955
3956 pblock = map.m_pblk;
3957 nr_pblocks = map.m_len;
3958
3959 if ((pblock - SM_I(sbi)->main_blkaddr) % blks_per_sec ||
3960 nr_pblocks % blks_per_sec ||
3961 !f2fs_valid_pinned_area(sbi, pblock)) {
3962 bool last_extent = false;
3963
3964 not_aligned++;
3965
3966 nr_pblocks = roundup(nr_pblocks, blks_per_sec);
3967 if (cur_lblock + nr_pblocks > sis->max)
3968 nr_pblocks -= blks_per_sec;
3969
3970 /* this extent is last one */
3971 if (!nr_pblocks) {
3972 nr_pblocks = last_lblock - cur_lblock;
3973 last_extent = true;
3974 }
3975
3976 ret = f2fs_migrate_blocks(inode, cur_lblock,
3977 nr_pblocks);
3978 if (ret) {
3979 if (ret == -ENOENT)
3980 ret = -EINVAL;
3981 goto out;
3982 }
3983
3984 if (!last_extent)
3985 goto retry;
3986 }
3987
3988 if (cur_lblock + nr_pblocks >= sis->max)
3989 nr_pblocks = sis->max - cur_lblock;
3990
3991 if (cur_lblock) { /* exclude the header page */
3992 if (pblock < lowest_pblock)
3993 lowest_pblock = pblock;
3994 if (pblock + nr_pblocks - 1 > highest_pblock)
3995 highest_pblock = pblock + nr_pblocks - 1;
3996 }
3997
3998 /*
3999 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4000 */
4001 ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
4002 if (ret < 0)
4003 goto out;
4004 nr_extents += ret;
4005 cur_lblock += nr_pblocks;
4006 }
4007 ret = nr_extents;
4008 *span = 1 + highest_pblock - lowest_pblock;
4009 if (cur_lblock == 0)
4010 cur_lblock = 1; /* force Empty message */
4011 sis->max = cur_lblock;
4012 sis->pages = cur_lblock - 1;
4013 sis->highest_bit = cur_lblock - 1;
4014 out:
4015 if (not_aligned)
4016 f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%u * N)",
4017 not_aligned, blks_per_sec * F2FS_BLKSIZE);
4018 return ret;
4019 }
4020
f2fs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)4021 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4022 sector_t *span)
4023 {
4024 struct inode *inode = file_inode(file);
4025 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4026 int ret;
4027
4028 if (!S_ISREG(inode->i_mode))
4029 return -EINVAL;
4030
4031 if (f2fs_readonly(sbi->sb))
4032 return -EROFS;
4033
4034 if (f2fs_lfs_mode(sbi) && !f2fs_sb_has_blkzoned(sbi)) {
4035 f2fs_err(sbi, "Swapfile not supported in LFS mode");
4036 return -EINVAL;
4037 }
4038
4039 ret = f2fs_convert_inline_inode(inode);
4040 if (ret)
4041 return ret;
4042
4043 if (!f2fs_disable_compressed_file(inode))
4044 return -EINVAL;
4045
4046 f2fs_precache_extents(inode);
4047
4048 ret = filemap_fdatawrite(inode->i_mapping);
4049 if (ret < 0)
4050 return ret;
4051
4052 ret = check_swap_activate(sis, file, span);
4053 if (ret < 0)
4054 return ret;
4055
4056 stat_inc_swapfile_inode(inode);
4057 set_inode_flag(inode, FI_PIN_FILE);
4058 f2fs_update_time(sbi, REQ_TIME);
4059 return ret;
4060 }
4061
f2fs_swap_deactivate(struct file * file)4062 static void f2fs_swap_deactivate(struct file *file)
4063 {
4064 struct inode *inode = file_inode(file);
4065
4066 stat_dec_swapfile_inode(inode);
4067 clear_inode_flag(inode, FI_PIN_FILE);
4068 }
4069 #else
f2fs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)4070 static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4071 sector_t *span)
4072 {
4073 return -EOPNOTSUPP;
4074 }
4075
f2fs_swap_deactivate(struct file * file)4076 static void f2fs_swap_deactivate(struct file *file)
4077 {
4078 }
4079 #endif
4080
4081 const struct address_space_operations f2fs_dblock_aops = {
4082 .read_folio = f2fs_read_data_folio,
4083 .readahead = f2fs_readahead,
4084 .writepage = f2fs_write_data_page,
4085 .writepages = f2fs_write_data_pages,
4086 .write_begin = f2fs_write_begin,
4087 .write_end = f2fs_write_end,
4088 .dirty_folio = f2fs_dirty_data_folio,
4089 .migrate_folio = filemap_migrate_folio,
4090 .invalidate_folio = f2fs_invalidate_folio,
4091 .release_folio = f2fs_release_folio,
4092 .bmap = f2fs_bmap,
4093 .swap_activate = f2fs_swap_activate,
4094 .swap_deactivate = f2fs_swap_deactivate,
4095 };
4096
f2fs_clear_page_cache_dirty_tag(struct page * page)4097 void f2fs_clear_page_cache_dirty_tag(struct page *page)
4098 {
4099 struct address_space *mapping = page_mapping(page);
4100 unsigned long flags;
4101
4102 xa_lock_irqsave(&mapping->i_pages, flags);
4103 __xa_clear_mark(&mapping->i_pages, page_index(page),
4104 PAGECACHE_TAG_DIRTY);
4105 xa_unlock_irqrestore(&mapping->i_pages, flags);
4106 }
4107
f2fs_init_post_read_processing(void)4108 int __init f2fs_init_post_read_processing(void)
4109 {
4110 bio_post_read_ctx_cache =
4111 kmem_cache_create("f2fs_bio_post_read_ctx",
4112 sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4113 if (!bio_post_read_ctx_cache)
4114 goto fail;
4115 bio_post_read_ctx_pool =
4116 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4117 bio_post_read_ctx_cache);
4118 if (!bio_post_read_ctx_pool)
4119 goto fail_free_cache;
4120 return 0;
4121
4122 fail_free_cache:
4123 kmem_cache_destroy(bio_post_read_ctx_cache);
4124 fail:
4125 return -ENOMEM;
4126 }
4127
f2fs_destroy_post_read_processing(void)4128 void f2fs_destroy_post_read_processing(void)
4129 {
4130 mempool_destroy(bio_post_read_ctx_pool);
4131 kmem_cache_destroy(bio_post_read_ctx_cache);
4132 }
4133
f2fs_init_post_read_wq(struct f2fs_sb_info * sbi)4134 int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4135 {
4136 if (!f2fs_sb_has_encrypt(sbi) &&
4137 !f2fs_sb_has_verity(sbi) &&
4138 !f2fs_sb_has_compression(sbi))
4139 return 0;
4140
4141 sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4142 WQ_UNBOUND | WQ_HIGHPRI,
4143 num_online_cpus());
4144 return sbi->post_read_wq ? 0 : -ENOMEM;
4145 }
4146
f2fs_destroy_post_read_wq(struct f2fs_sb_info * sbi)4147 void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4148 {
4149 if (sbi->post_read_wq)
4150 destroy_workqueue(sbi->post_read_wq);
4151 }
4152
f2fs_init_bio_entry_cache(void)4153 int __init f2fs_init_bio_entry_cache(void)
4154 {
4155 bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4156 sizeof(struct bio_entry));
4157 return bio_entry_slab ? 0 : -ENOMEM;
4158 }
4159
f2fs_destroy_bio_entry_cache(void)4160 void f2fs_destroy_bio_entry_cache(void)
4161 {
4162 kmem_cache_destroy(bio_entry_slab);
4163 }
4164
f2fs_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)4165 static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4166 unsigned int flags, struct iomap *iomap,
4167 struct iomap *srcmap)
4168 {
4169 struct f2fs_map_blocks map = {};
4170 pgoff_t next_pgofs = 0;
4171 int err;
4172
4173 map.m_lblk = bytes_to_blks(inode, offset);
4174 map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4175 map.m_next_pgofs = &next_pgofs;
4176 map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4177 if (flags & IOMAP_WRITE)
4178 map.m_may_create = true;
4179
4180 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DIO);
4181 if (err)
4182 return err;
4183
4184 iomap->offset = blks_to_bytes(inode, map.m_lblk);
4185
4186 /*
4187 * When inline encryption is enabled, sometimes I/O to an encrypted file
4188 * has to be broken up to guarantee DUN contiguity. Handle this by
4189 * limiting the length of the mapping returned.
4190 */
4191 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4192
4193 /*
4194 * We should never see delalloc or compressed extents here based on
4195 * prior flushing and checks.
4196 */
4197 if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR))
4198 return -EINVAL;
4199 if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR))
4200 return -EINVAL;
4201
4202 if (map.m_flags & F2FS_MAP_MAPPED) {
4203 iomap->length = blks_to_bytes(inode, map.m_len);
4204 iomap->type = IOMAP_MAPPED;
4205 iomap->flags |= IOMAP_F_MERGED;
4206 iomap->bdev = map.m_bdev;
4207 iomap->addr = blks_to_bytes(inode, map.m_pblk);
4208 } else {
4209 if (flags & IOMAP_WRITE)
4210 return -ENOTBLK;
4211 iomap->length = blks_to_bytes(inode, next_pgofs) -
4212 iomap->offset;
4213 iomap->type = IOMAP_HOLE;
4214 iomap->addr = IOMAP_NULL_ADDR;
4215 }
4216
4217 if (map.m_flags & F2FS_MAP_NEW)
4218 iomap->flags |= IOMAP_F_NEW;
4219 if ((inode->i_state & I_DIRTY_DATASYNC) ||
4220 offset + length > i_size_read(inode))
4221 iomap->flags |= IOMAP_F_DIRTY;
4222
4223 return 0;
4224 }
4225
4226 const struct iomap_ops f2fs_iomap_ops = {
4227 .iomap_begin = f2fs_iomap_begin,
4228 };
4229