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 | |
40 | int __init f2fs_init_bioset(void) |
41 | { |
42 | return bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE, |
43 | 0, flags: BIOSET_NEED_BVECS); |
44 | } |
45 | |
46 | void f2fs_destroy_bioset(void) |
47 | { |
48 | bioset_exit(&f2fs_bioset); |
49 | } |
50 | |
51 | static bool __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 (f2fs_is_compressed_page(page)) |
69 | return false; |
70 | if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) || |
71 | page_private_gcing(page)) |
72 | return true; |
73 | return false; |
74 | } |
75 | |
76 | static enum count_type __read_io_type(struct page *page) |
77 | { |
78 | struct address_space *mapping = page_file_mapping(page); |
79 | |
80 | if (mapping) { |
81 | struct inode *inode = mapping->host; |
82 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
83 | |
84 | if (inode->i_ino == F2FS_META_INO(sbi)) |
85 | return F2FS_RD_META; |
86 | |
87 | if (inode->i_ino == F2FS_NODE_INO(sbi)) |
88 | return F2FS_RD_NODE; |
89 | } |
90 | return F2FS_RD_DATA; |
91 | } |
92 | |
93 | /* postprocessing steps for read bios */ |
94 | enum bio_post_read_step { |
95 | #ifdef CONFIG_FS_ENCRYPTION |
96 | STEP_DECRYPT = BIT(0), |
97 | #else |
98 | STEP_DECRYPT = 0, /* compile out the decryption-related code */ |
99 | #endif |
100 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
101 | STEP_DECOMPRESS = BIT(1), |
102 | #else |
103 | STEP_DECOMPRESS = 0, /* compile out the decompression-related code */ |
104 | #endif |
105 | #ifdef CONFIG_FS_VERITY |
106 | STEP_VERITY = BIT(2), |
107 | #else |
108 | STEP_VERITY = 0, /* compile out the verity-related code */ |
109 | #endif |
110 | }; |
111 | |
112 | struct bio_post_read_ctx { |
113 | struct bio *bio; |
114 | struct f2fs_sb_info *sbi; |
115 | struct work_struct work; |
116 | unsigned int enabled_steps; |
117 | /* |
118 | * decompression_attempted keeps track of whether |
119 | * f2fs_end_read_compressed_page() has been called on the pages in the |
120 | * bio that belong to a compressed cluster yet. |
121 | */ |
122 | bool decompression_attempted; |
123 | block_t fs_blkaddr; |
124 | }; |
125 | |
126 | /* |
127 | * Update and unlock a bio's pages, and free the bio. |
128 | * |
129 | * This marks pages up-to-date only if there was no error in the bio (I/O error, |
130 | * decryption error, or verity error), as indicated by bio->bi_status. |
131 | * |
132 | * "Compressed pages" (pagecache pages backed by a compressed cluster on-disk) |
133 | * aren't marked up-to-date here, as decompression is done on a per-compression- |
134 | * cluster basis rather than a per-bio basis. Instead, we only must do two |
135 | * things for each compressed page here: call f2fs_end_read_compressed_page() |
136 | * with failed=true if an error occurred before it would have normally gotten |
137 | * called (i.e., I/O error or decryption error, but *not* verity error), and |
138 | * release the bio's reference to the decompress_io_ctx of the page's cluster. |
139 | */ |
140 | static void f2fs_finish_read_bio(struct bio *bio, bool in_task) |
141 | { |
142 | struct bio_vec *bv; |
143 | struct bvec_iter_all iter_all; |
144 | struct bio_post_read_ctx *ctx = bio->bi_private; |
145 | |
146 | bio_for_each_segment_all(bv, bio, iter_all) { |
147 | struct page *page = bv->bv_page; |
148 | |
149 | if (f2fs_is_compressed_page(page)) { |
150 | if (ctx && !ctx->decompression_attempted) |
151 | f2fs_end_read_compressed_page(page, failed: true, blkaddr: 0, |
152 | in_task); |
153 | f2fs_put_page_dic(page, in_task); |
154 | continue; |
155 | } |
156 | |
157 | if (bio->bi_status) |
158 | ClearPageUptodate(page); |
159 | else |
160 | SetPageUptodate(page); |
161 | dec_page_count(sbi: F2FS_P_SB(page), count_type: __read_io_type(page)); |
162 | unlock_page(page); |
163 | } |
164 | |
165 | if (ctx) |
166 | mempool_free(element: ctx, pool: bio_post_read_ctx_pool); |
167 | bio_put(bio); |
168 | } |
169 | |
170 | static void f2fs_verify_bio(struct work_struct *work) |
171 | { |
172 | struct bio_post_read_ctx *ctx = |
173 | container_of(work, struct bio_post_read_ctx, work); |
174 | struct bio *bio = ctx->bio; |
175 | bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS); |
176 | |
177 | /* |
178 | * fsverity_verify_bio() may call readahead() again, and while verity |
179 | * will be disabled for this, decryption and/or decompression may still |
180 | * be needed, resulting in another bio_post_read_ctx being allocated. |
181 | * So to prevent deadlocks we need to release the current ctx to the |
182 | * mempool first. This assumes that verity is the last post-read step. |
183 | */ |
184 | mempool_free(element: ctx, pool: bio_post_read_ctx_pool); |
185 | bio->bi_private = NULL; |
186 | |
187 | /* |
188 | * Verify the bio's pages with fs-verity. Exclude compressed pages, |
189 | * as those were handled separately by f2fs_end_read_compressed_page(). |
190 | */ |
191 | if (may_have_compressed_pages) { |
192 | struct bio_vec *bv; |
193 | struct bvec_iter_all iter_all; |
194 | |
195 | bio_for_each_segment_all(bv, bio, iter_all) { |
196 | struct page *page = bv->bv_page; |
197 | |
198 | if (!f2fs_is_compressed_page(page) && |
199 | !fsverity_verify_page(page)) { |
200 | bio->bi_status = BLK_STS_IOERR; |
201 | break; |
202 | } |
203 | } |
204 | } else { |
205 | fsverity_verify_bio(bio); |
206 | } |
207 | |
208 | f2fs_finish_read_bio(bio, in_task: true); |
209 | } |
210 | |
211 | /* |
212 | * If the bio's data needs to be verified with fs-verity, then enqueue the |
213 | * verity work for the bio. Otherwise finish the bio now. |
214 | * |
215 | * Note that to avoid deadlocks, the verity work can't be done on the |
216 | * decryption/decompression workqueue. This is because verifying the data pages |
217 | * can involve reading verity metadata pages from the file, and these verity |
218 | * metadata pages may be encrypted and/or compressed. |
219 | */ |
220 | static void f2fs_verify_and_finish_bio(struct bio *bio, bool in_task) |
221 | { |
222 | struct bio_post_read_ctx *ctx = bio->bi_private; |
223 | |
224 | if (ctx && (ctx->enabled_steps & STEP_VERITY)) { |
225 | INIT_WORK(&ctx->work, f2fs_verify_bio); |
226 | fsverity_enqueue_verify_work(work: &ctx->work); |
227 | } else { |
228 | f2fs_finish_read_bio(bio, in_task); |
229 | } |
230 | } |
231 | |
232 | /* |
233 | * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last |
234 | * remaining page was read by @ctx->bio. |
235 | * |
236 | * Note that a bio may span clusters (even a mix of compressed and uncompressed |
237 | * clusters) or be for just part of a cluster. STEP_DECOMPRESS just indicates |
238 | * that the bio includes at least one compressed page. The actual decompression |
239 | * is done on a per-cluster basis, not a per-bio basis. |
240 | */ |
241 | static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx, |
242 | bool in_task) |
243 | { |
244 | struct bio_vec *bv; |
245 | struct bvec_iter_all iter_all; |
246 | bool all_compressed = true; |
247 | block_t blkaddr = ctx->fs_blkaddr; |
248 | |
249 | bio_for_each_segment_all(bv, ctx->bio, iter_all) { |
250 | struct page *page = bv->bv_page; |
251 | |
252 | if (f2fs_is_compressed_page(page)) |
253 | f2fs_end_read_compressed_page(page, failed: false, blkaddr, |
254 | in_task); |
255 | else |
256 | all_compressed = false; |
257 | |
258 | blkaddr++; |
259 | } |
260 | |
261 | ctx->decompression_attempted = true; |
262 | |
263 | /* |
264 | * Optimization: if all the bio's pages are compressed, then scheduling |
265 | * the per-bio verity work is unnecessary, as verity will be fully |
266 | * handled at the compression cluster level. |
267 | */ |
268 | if (all_compressed) |
269 | ctx->enabled_steps &= ~STEP_VERITY; |
270 | } |
271 | |
272 | static void f2fs_post_read_work(struct work_struct *work) |
273 | { |
274 | struct bio_post_read_ctx *ctx = |
275 | container_of(work, struct bio_post_read_ctx, work); |
276 | struct bio *bio = ctx->bio; |
277 | |
278 | if ((ctx->enabled_steps & STEP_DECRYPT) && !fscrypt_decrypt_bio(bio)) { |
279 | f2fs_finish_read_bio(bio, in_task: true); |
280 | return; |
281 | } |
282 | |
283 | if (ctx->enabled_steps & STEP_DECOMPRESS) |
284 | f2fs_handle_step_decompress(ctx, in_task: true); |
285 | |
286 | f2fs_verify_and_finish_bio(bio, in_task: true); |
287 | } |
288 | |
289 | static void f2fs_read_end_io(struct bio *bio) |
290 | { |
291 | struct f2fs_sb_info *sbi = F2FS_P_SB(page: bio_first_page_all(bio)); |
292 | struct bio_post_read_ctx *ctx; |
293 | bool intask = in_task(); |
294 | |
295 | iostat_update_and_unbind_ctx(bio); |
296 | ctx = bio->bi_private; |
297 | |
298 | if (time_to_inject(sbi, FAULT_READ_IO)) |
299 | bio->bi_status = BLK_STS_IOERR; |
300 | |
301 | if (bio->bi_status) { |
302 | f2fs_finish_read_bio(bio, in_task: intask); |
303 | return; |
304 | } |
305 | |
306 | if (ctx) { |
307 | unsigned int enabled_steps = ctx->enabled_steps & |
308 | (STEP_DECRYPT | STEP_DECOMPRESS); |
309 | |
310 | /* |
311 | * If we have only decompression step between decompression and |
312 | * decrypt, we don't need post processing for this. |
313 | */ |
314 | if (enabled_steps == STEP_DECOMPRESS && |
315 | !f2fs_low_mem_mode(sbi)) { |
316 | f2fs_handle_step_decompress(ctx, in_task: intask); |
317 | } else if (enabled_steps) { |
318 | INIT_WORK(&ctx->work, f2fs_post_read_work); |
319 | queue_work(wq: ctx->sbi->post_read_wq, work: &ctx->work); |
320 | return; |
321 | } |
322 | } |
323 | |
324 | f2fs_verify_and_finish_bio(bio, in_task: intask); |
325 | } |
326 | |
327 | static void f2fs_write_end_io(struct bio *bio) |
328 | { |
329 | struct f2fs_sb_info *sbi; |
330 | struct bio_vec *bvec; |
331 | struct bvec_iter_all iter_all; |
332 | |
333 | iostat_update_and_unbind_ctx(bio); |
334 | sbi = bio->bi_private; |
335 | |
336 | if (time_to_inject(sbi, FAULT_WRITE_IO)) |
337 | bio->bi_status = BLK_STS_IOERR; |
338 | |
339 | bio_for_each_segment_all(bvec, bio, iter_all) { |
340 | struct page *page = bvec->bv_page; |
341 | enum count_type type = WB_DATA_TYPE(page); |
342 | |
343 | if (page_private_dummy(page)) { |
344 | clear_page_private_dummy(page); |
345 | unlock_page(page); |
346 | mempool_free(element: page, pool: sbi->write_io_dummy); |
347 | |
348 | if (unlikely(bio->bi_status)) |
349 | f2fs_stop_checkpoint(sbi, end_io: true, |
350 | reason: STOP_CP_REASON_WRITE_FAIL); |
351 | continue; |
352 | } |
353 | |
354 | fscrypt_finalize_bounce_page(pagep: &page); |
355 | |
356 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
357 | if (f2fs_is_compressed_page(page)) { |
358 | f2fs_compress_write_end_io(bio, page); |
359 | continue; |
360 | } |
361 | #endif |
362 | |
363 | if (unlikely(bio->bi_status)) { |
364 | mapping_set_error(mapping: page->mapping, error: -EIO); |
365 | if (type == F2FS_WB_CP_DATA) |
366 | f2fs_stop_checkpoint(sbi, end_io: true, |
367 | reason: STOP_CP_REASON_WRITE_FAIL); |
368 | } |
369 | |
370 | f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) && |
371 | page->index != nid_of_node(page)); |
372 | |
373 | dec_page_count(sbi, count_type: type); |
374 | if (f2fs_in_warm_node_list(sbi, page)) |
375 | f2fs_del_fsync_node_entry(sbi, page); |
376 | clear_page_private_gcing(page); |
377 | end_page_writeback(page); |
378 | } |
379 | if (!get_pages(sbi, count_type: F2FS_WB_CP_DATA) && |
380 | wq_has_sleeper(wq_head: &sbi->cp_wait)) |
381 | wake_up(&sbi->cp_wait); |
382 | |
383 | bio_put(bio); |
384 | } |
385 | |
386 | #ifdef CONFIG_BLK_DEV_ZONED |
387 | static void f2fs_zone_write_end_io(struct bio *bio) |
388 | { |
389 | struct f2fs_bio_info *io = (struct f2fs_bio_info *)bio->bi_private; |
390 | |
391 | bio->bi_private = io->bi_private; |
392 | complete(&io->zone_wait); |
393 | f2fs_write_end_io(bio); |
394 | } |
395 | #endif |
396 | |
397 | struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi, |
398 | block_t blk_addr, sector_t *sector) |
399 | { |
400 | struct block_device *bdev = sbi->sb->s_bdev; |
401 | int i; |
402 | |
403 | if (f2fs_is_multi_device(sbi)) { |
404 | for (i = 0; i < sbi->s_ndevs; i++) { |
405 | if (FDEV(i).start_blk <= blk_addr && |
406 | FDEV(i).end_blk >= blk_addr) { |
407 | blk_addr -= FDEV(i).start_blk; |
408 | bdev = FDEV(i).bdev; |
409 | break; |
410 | } |
411 | } |
412 | } |
413 | |
414 | if (sector) |
415 | *sector = SECTOR_FROM_BLOCK(blk_addr); |
416 | return bdev; |
417 | } |
418 | |
419 | int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr) |
420 | { |
421 | int i; |
422 | |
423 | if (!f2fs_is_multi_device(sbi)) |
424 | return 0; |
425 | |
426 | for (i = 0; i < sbi->s_ndevs; i++) |
427 | if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr) |
428 | return i; |
429 | return 0; |
430 | } |
431 | |
432 | static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio) |
433 | { |
434 | unsigned int temp_mask = GENMASK(NR_TEMP_TYPE - 1, 0); |
435 | unsigned int fua_flag, meta_flag, io_flag; |
436 | blk_opf_t op_flags = 0; |
437 | |
438 | if (fio->op != REQ_OP_WRITE) |
439 | return 0; |
440 | if (fio->type == DATA) |
441 | io_flag = fio->sbi->data_io_flag; |
442 | else if (fio->type == NODE) |
443 | io_flag = fio->sbi->node_io_flag; |
444 | else |
445 | return 0; |
446 | |
447 | fua_flag = io_flag & temp_mask; |
448 | meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask; |
449 | |
450 | /* |
451 | * data/node io flag bits per temp: |
452 | * REQ_META | REQ_FUA | |
453 | * 5 | 4 | 3 | 2 | 1 | 0 | |
454 | * Cold | Warm | Hot | Cold | Warm | Hot | |
455 | */ |
456 | if (BIT(fio->temp) & meta_flag) |
457 | op_flags |= REQ_META; |
458 | if (BIT(fio->temp) & fua_flag) |
459 | op_flags |= REQ_FUA; |
460 | return op_flags; |
461 | } |
462 | |
463 | static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages) |
464 | { |
465 | struct f2fs_sb_info *sbi = fio->sbi; |
466 | struct block_device *bdev; |
467 | sector_t sector; |
468 | struct bio *bio; |
469 | |
470 | bdev = f2fs_target_device(sbi, blk_addr: fio->new_blkaddr, sector: §or); |
471 | bio = bio_alloc_bioset(bdev, nr_vecs: npages, |
472 | opf: fio->op | fio->op_flags | f2fs_io_flags(fio), |
473 | GFP_NOIO, bs: &f2fs_bioset); |
474 | bio->bi_iter.bi_sector = sector; |
475 | if (is_read_io(fio->op)) { |
476 | bio->bi_end_io = f2fs_read_end_io; |
477 | bio->bi_private = NULL; |
478 | } else { |
479 | bio->bi_end_io = f2fs_write_end_io; |
480 | bio->bi_private = sbi; |
481 | } |
482 | iostat_alloc_and_bind_ctx(sbi, bio, NULL); |
483 | |
484 | if (fio->io_wbc) |
485 | wbc_init_bio(wbc: fio->io_wbc, bio); |
486 | |
487 | return bio; |
488 | } |
489 | |
490 | static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode, |
491 | pgoff_t first_idx, |
492 | const struct f2fs_io_info *fio, |
493 | gfp_t gfp_mask) |
494 | { |
495 | /* |
496 | * The f2fs garbage collector sets ->encrypted_page when it wants to |
497 | * read/write raw data without encryption. |
498 | */ |
499 | if (!fio || !fio->encrypted_page) |
500 | fscrypt_set_bio_crypt_ctx(bio, inode, first_lblk: first_idx, gfp_mask); |
501 | } |
502 | |
503 | static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode, |
504 | pgoff_t next_idx, |
505 | const struct f2fs_io_info *fio) |
506 | { |
507 | /* |
508 | * The f2fs garbage collector sets ->encrypted_page when it wants to |
509 | * read/write raw data without encryption. |
510 | */ |
511 | if (fio && fio->encrypted_page) |
512 | return !bio_has_crypt_ctx(bio); |
513 | |
514 | return fscrypt_mergeable_bio(bio, inode, next_lblk: next_idx); |
515 | } |
516 | |
517 | void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio, |
518 | enum page_type type) |
519 | { |
520 | WARN_ON_ONCE(!is_read_io(bio_op(bio))); |
521 | trace_f2fs_submit_read_bio(sb: sbi->sb, type, bio); |
522 | |
523 | iostat_update_submit_ctx(bio, type); |
524 | submit_bio(bio); |
525 | } |
526 | |
527 | static void f2fs_align_write_bio(struct f2fs_sb_info *sbi, struct bio *bio) |
528 | { |
529 | unsigned int start = |
530 | (bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS) % F2FS_IO_SIZE(sbi); |
531 | |
532 | if (start == 0) |
533 | return; |
534 | |
535 | /* fill dummy pages */ |
536 | for (; start < F2FS_IO_SIZE(sbi); start++) { |
537 | struct page *page = |
538 | mempool_alloc(pool: sbi->write_io_dummy, |
539 | GFP_NOIO | __GFP_NOFAIL); |
540 | f2fs_bug_on(sbi, !page); |
541 | |
542 | lock_page(page); |
543 | |
544 | zero_user_segment(page, start: 0, PAGE_SIZE); |
545 | set_page_private_dummy(page); |
546 | |
547 | if (bio_add_page(bio, page, PAGE_SIZE, off: 0) < PAGE_SIZE) |
548 | f2fs_bug_on(sbi, 1); |
549 | } |
550 | } |
551 | |
552 | static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio, |
553 | enum page_type type) |
554 | { |
555 | WARN_ON_ONCE(is_read_io(bio_op(bio))); |
556 | |
557 | if (type == DATA || type == NODE) { |
558 | if (f2fs_lfs_mode(sbi) && current->plug) |
559 | blk_finish_plug(current->plug); |
560 | |
561 | if (F2FS_IO_ALIGNED(sbi)) { |
562 | f2fs_align_write_bio(sbi, bio); |
563 | /* |
564 | * In the NODE case, we lose next block address chain. |
565 | * So, we need to do checkpoint in f2fs_sync_file. |
566 | */ |
567 | if (type == NODE) |
568 | set_sbi_flag(sbi, type: SBI_NEED_CP); |
569 | } |
570 | } |
571 | |
572 | trace_f2fs_submit_write_bio(sb: sbi->sb, type, bio); |
573 | iostat_update_submit_ctx(bio, type); |
574 | submit_bio(bio); |
575 | } |
576 | |
577 | static void __submit_merged_bio(struct f2fs_bio_info *io) |
578 | { |
579 | struct f2fs_io_info *fio = &io->fio; |
580 | |
581 | if (!io->bio) |
582 | return; |
583 | |
584 | if (is_read_io(fio->op)) { |
585 | trace_f2fs_prepare_read_bio(sb: io->sbi->sb, type: fio->type, bio: io->bio); |
586 | f2fs_submit_read_bio(sbi: io->sbi, bio: io->bio, type: fio->type); |
587 | } else { |
588 | trace_f2fs_prepare_write_bio(sb: io->sbi->sb, type: fio->type, bio: io->bio); |
589 | f2fs_submit_write_bio(sbi: io->sbi, bio: io->bio, type: fio->type); |
590 | } |
591 | io->bio = NULL; |
592 | } |
593 | |
594 | static bool __has_merged_page(struct bio *bio, struct inode *inode, |
595 | struct page *page, nid_t ino) |
596 | { |
597 | struct bio_vec *bvec; |
598 | struct bvec_iter_all iter_all; |
599 | |
600 | if (!bio) |
601 | return false; |
602 | |
603 | if (!inode && !page && !ino) |
604 | return true; |
605 | |
606 | bio_for_each_segment_all(bvec, bio, iter_all) { |
607 | struct page *target = bvec->bv_page; |
608 | |
609 | if (fscrypt_is_bounce_page(page: target)) { |
610 | target = fscrypt_pagecache_page(bounce_page: target); |
611 | if (IS_ERR(ptr: target)) |
612 | continue; |
613 | } |
614 | if (f2fs_is_compressed_page(page: target)) { |
615 | target = f2fs_compress_control_page(page: target); |
616 | if (IS_ERR(ptr: target)) |
617 | continue; |
618 | } |
619 | |
620 | if (inode && inode == target->mapping->host) |
621 | return true; |
622 | if (page && page == target) |
623 | return true; |
624 | if (ino && ino == ino_of_node(node_page: target)) |
625 | return true; |
626 | } |
627 | |
628 | return false; |
629 | } |
630 | |
631 | int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi) |
632 | { |
633 | int i; |
634 | |
635 | for (i = 0; i < NR_PAGE_TYPE; i++) { |
636 | int n = (i == META) ? 1 : NR_TEMP_TYPE; |
637 | int j; |
638 | |
639 | sbi->write_io[i] = f2fs_kmalloc(sbi, |
640 | array_size(n, sizeof(struct f2fs_bio_info)), |
641 | GFP_KERNEL); |
642 | if (!sbi->write_io[i]) |
643 | return -ENOMEM; |
644 | |
645 | for (j = HOT; j < n; j++) { |
646 | init_f2fs_rwsem(&sbi->write_io[i][j].io_rwsem); |
647 | sbi->write_io[i][j].sbi = sbi; |
648 | sbi->write_io[i][j].bio = NULL; |
649 | spin_lock_init(&sbi->write_io[i][j].io_lock); |
650 | INIT_LIST_HEAD(list: &sbi->write_io[i][j].io_list); |
651 | INIT_LIST_HEAD(list: &sbi->write_io[i][j].bio_list); |
652 | init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock); |
653 | #ifdef CONFIG_BLK_DEV_ZONED |
654 | init_completion(x: &sbi->write_io[i][j].zone_wait); |
655 | sbi->write_io[i][j].zone_pending_bio = NULL; |
656 | sbi->write_io[i][j].bi_private = NULL; |
657 | #endif |
658 | } |
659 | } |
660 | |
661 | return 0; |
662 | } |
663 | |
664 | static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi, |
665 | enum page_type type, enum temp_type temp) |
666 | { |
667 | enum page_type btype = PAGE_TYPE_OF_BIO(type); |
668 | struct f2fs_bio_info *io = sbi->write_io[btype] + temp; |
669 | |
670 | f2fs_down_write(sem: &io->io_rwsem); |
671 | |
672 | if (!io->bio) |
673 | goto unlock_out; |
674 | |
675 | /* change META to META_FLUSH in the checkpoint procedure */ |
676 | if (type >= META_FLUSH) { |
677 | io->fio.type = META_FLUSH; |
678 | io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC; |
679 | if (!test_opt(sbi, NOBARRIER)) |
680 | io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA; |
681 | } |
682 | __submit_merged_bio(io); |
683 | unlock_out: |
684 | f2fs_up_write(sem: &io->io_rwsem); |
685 | } |
686 | |
687 | static void __submit_merged_write_cond(struct f2fs_sb_info *sbi, |
688 | struct inode *inode, struct page *page, |
689 | nid_t ino, enum page_type type, bool force) |
690 | { |
691 | enum temp_type temp; |
692 | bool ret = true; |
693 | |
694 | for (temp = HOT; temp < NR_TEMP_TYPE; temp++) { |
695 | if (!force) { |
696 | enum page_type btype = PAGE_TYPE_OF_BIO(type); |
697 | struct f2fs_bio_info *io = sbi->write_io[btype] + temp; |
698 | |
699 | f2fs_down_read(sem: &io->io_rwsem); |
700 | ret = __has_merged_page(bio: io->bio, inode, page, ino); |
701 | f2fs_up_read(sem: &io->io_rwsem); |
702 | } |
703 | if (ret) |
704 | __f2fs_submit_merged_write(sbi, type, temp); |
705 | |
706 | /* TODO: use HOT temp only for meta pages now. */ |
707 | if (type >= META) |
708 | break; |
709 | } |
710 | } |
711 | |
712 | void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type) |
713 | { |
714 | __submit_merged_write_cond(sbi, NULL, NULL, ino: 0, type, force: true); |
715 | } |
716 | |
717 | void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi, |
718 | struct inode *inode, struct page *page, |
719 | nid_t ino, enum page_type type) |
720 | { |
721 | __submit_merged_write_cond(sbi, inode, page, ino, type, force: false); |
722 | } |
723 | |
724 | void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi) |
725 | { |
726 | f2fs_submit_merged_write(sbi, type: DATA); |
727 | f2fs_submit_merged_write(sbi, type: NODE); |
728 | f2fs_submit_merged_write(sbi, type: META); |
729 | } |
730 | |
731 | /* |
732 | * Fill the locked page with data located in the block address. |
733 | * A caller needs to unlock the page on failure. |
734 | */ |
735 | int f2fs_submit_page_bio(struct f2fs_io_info *fio) |
736 | { |
737 | struct bio *bio; |
738 | struct page *page = fio->encrypted_page ? |
739 | fio->encrypted_page : fio->page; |
740 | |
741 | if (!f2fs_is_valid_blkaddr(sbi: fio->sbi, blkaddr: fio->new_blkaddr, |
742 | type: fio->is_por ? META_POR : (__is_meta_io(fio) ? |
743 | META_GENERIC : DATA_GENERIC_ENHANCE))) { |
744 | f2fs_handle_error(sbi: fio->sbi, error: ERROR_INVALID_BLKADDR); |
745 | return -EFSCORRUPTED; |
746 | } |
747 | |
748 | trace_f2fs_submit_page_bio(page, fio); |
749 | |
750 | /* Allocate a new bio */ |
751 | bio = __bio_alloc(fio, npages: 1); |
752 | |
753 | f2fs_set_bio_crypt_ctx(bio, inode: fio->page->mapping->host, |
754 | first_idx: fio->page->index, fio, GFP_NOIO); |
755 | |
756 | if (bio_add_page(bio, page, PAGE_SIZE, off: 0) < PAGE_SIZE) { |
757 | bio_put(bio); |
758 | return -EFAULT; |
759 | } |
760 | |
761 | if (fio->io_wbc && !is_read_io(fio->op)) |
762 | wbc_account_cgroup_owner(wbc: fio->io_wbc, page: fio->page, PAGE_SIZE); |
763 | |
764 | inc_page_count(sbi: fio->sbi, is_read_io(fio->op) ? |
765 | __read_io_type(page) : WB_DATA_TYPE(fio->page)); |
766 | |
767 | if (is_read_io(bio_op(bio))) |
768 | f2fs_submit_read_bio(sbi: fio->sbi, bio, type: fio->type); |
769 | else |
770 | f2fs_submit_write_bio(sbi: fio->sbi, bio, type: fio->type); |
771 | return 0; |
772 | } |
773 | |
774 | static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, |
775 | block_t last_blkaddr, block_t cur_blkaddr) |
776 | { |
777 | if (unlikely(sbi->max_io_bytes && |
778 | bio->bi_iter.bi_size >= sbi->max_io_bytes)) |
779 | return false; |
780 | if (last_blkaddr + 1 != cur_blkaddr) |
781 | return false; |
782 | return bio->bi_bdev == f2fs_target_device(sbi, blk_addr: cur_blkaddr, NULL); |
783 | } |
784 | |
785 | static bool io_type_is_mergeable(struct f2fs_bio_info *io, |
786 | struct f2fs_io_info *fio) |
787 | { |
788 | if (io->fio.op != fio->op) |
789 | return false; |
790 | return io->fio.op_flags == fio->op_flags; |
791 | } |
792 | |
793 | static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio, |
794 | struct f2fs_bio_info *io, |
795 | struct f2fs_io_info *fio, |
796 | block_t last_blkaddr, |
797 | block_t cur_blkaddr) |
798 | { |
799 | if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) { |
800 | unsigned int filled_blocks = |
801 | F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size); |
802 | unsigned int io_size = F2FS_IO_SIZE(sbi); |
803 | unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt; |
804 | |
805 | /* IOs in bio is aligned and left space of vectors is not enough */ |
806 | if (!(filled_blocks % io_size) && left_vecs < io_size) |
807 | return false; |
808 | } |
809 | if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr)) |
810 | return false; |
811 | return io_type_is_mergeable(io, fio); |
812 | } |
813 | |
814 | static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio, |
815 | struct page *page, enum temp_type temp) |
816 | { |
817 | struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; |
818 | struct bio_entry *be; |
819 | |
820 | be = f2fs_kmem_cache_alloc(cachep: bio_entry_slab, GFP_NOFS, nofail: true, NULL); |
821 | be->bio = bio; |
822 | bio_get(bio); |
823 | |
824 | if (bio_add_page(bio, page, PAGE_SIZE, off: 0) != PAGE_SIZE) |
825 | f2fs_bug_on(sbi, 1); |
826 | |
827 | f2fs_down_write(sem: &io->bio_list_lock); |
828 | list_add_tail(new: &be->list, head: &io->bio_list); |
829 | f2fs_up_write(sem: &io->bio_list_lock); |
830 | } |
831 | |
832 | static void del_bio_entry(struct bio_entry *be) |
833 | { |
834 | list_del(entry: &be->list); |
835 | kmem_cache_free(s: bio_entry_slab, objp: be); |
836 | } |
837 | |
838 | static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio, |
839 | struct page *page) |
840 | { |
841 | struct f2fs_sb_info *sbi = fio->sbi; |
842 | enum temp_type temp; |
843 | bool found = false; |
844 | int ret = -EAGAIN; |
845 | |
846 | for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { |
847 | struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; |
848 | struct list_head *head = &io->bio_list; |
849 | struct bio_entry *be; |
850 | |
851 | f2fs_down_write(sem: &io->bio_list_lock); |
852 | list_for_each_entry(be, head, list) { |
853 | if (be->bio != *bio) |
854 | continue; |
855 | |
856 | found = true; |
857 | |
858 | f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio, |
859 | *fio->last_block, |
860 | fio->new_blkaddr)); |
861 | if (f2fs_crypt_mergeable_bio(bio: *bio, |
862 | inode: fio->page->mapping->host, |
863 | next_idx: fio->page->index, fio) && |
864 | bio_add_page(bio: *bio, page, PAGE_SIZE, off: 0) == |
865 | PAGE_SIZE) { |
866 | ret = 0; |
867 | break; |
868 | } |
869 | |
870 | /* page can't be merged into bio; submit the bio */ |
871 | del_bio_entry(be); |
872 | f2fs_submit_write_bio(sbi, bio: *bio, type: DATA); |
873 | break; |
874 | } |
875 | f2fs_up_write(sem: &io->bio_list_lock); |
876 | } |
877 | |
878 | if (ret) { |
879 | bio_put(*bio); |
880 | *bio = NULL; |
881 | } |
882 | |
883 | return ret; |
884 | } |
885 | |
886 | void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi, |
887 | struct bio **bio, struct page *page) |
888 | { |
889 | enum temp_type temp; |
890 | bool found = false; |
891 | struct bio *target = bio ? *bio : NULL; |
892 | |
893 | f2fs_bug_on(sbi, !target && !page); |
894 | |
895 | for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) { |
896 | struct f2fs_bio_info *io = sbi->write_io[DATA] + temp; |
897 | struct list_head *head = &io->bio_list; |
898 | struct bio_entry *be; |
899 | |
900 | if (list_empty(head)) |
901 | continue; |
902 | |
903 | f2fs_down_read(sem: &io->bio_list_lock); |
904 | list_for_each_entry(be, head, list) { |
905 | if (target) |
906 | found = (target == be->bio); |
907 | else |
908 | found = __has_merged_page(bio: be->bio, NULL, |
909 | page, ino: 0); |
910 | if (found) |
911 | break; |
912 | } |
913 | f2fs_up_read(sem: &io->bio_list_lock); |
914 | |
915 | if (!found) |
916 | continue; |
917 | |
918 | found = false; |
919 | |
920 | f2fs_down_write(sem: &io->bio_list_lock); |
921 | list_for_each_entry(be, head, list) { |
922 | if (target) |
923 | found = (target == be->bio); |
924 | else |
925 | found = __has_merged_page(bio: be->bio, NULL, |
926 | page, ino: 0); |
927 | if (found) { |
928 | target = be->bio; |
929 | del_bio_entry(be); |
930 | break; |
931 | } |
932 | } |
933 | f2fs_up_write(sem: &io->bio_list_lock); |
934 | } |
935 | |
936 | if (found) |
937 | f2fs_submit_write_bio(sbi, bio: target, type: DATA); |
938 | if (bio && *bio) { |
939 | bio_put(*bio); |
940 | *bio = NULL; |
941 | } |
942 | } |
943 | |
944 | int f2fs_merge_page_bio(struct f2fs_io_info *fio) |
945 | { |
946 | struct bio *bio = *fio->bio; |
947 | struct page *page = fio->encrypted_page ? |
948 | fio->encrypted_page : fio->page; |
949 | |
950 | if (!f2fs_is_valid_blkaddr(sbi: fio->sbi, blkaddr: fio->new_blkaddr, |
951 | __is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) { |
952 | f2fs_handle_error(sbi: fio->sbi, error: ERROR_INVALID_BLKADDR); |
953 | return -EFSCORRUPTED; |
954 | } |
955 | |
956 | trace_f2fs_submit_page_bio(page, fio); |
957 | |
958 | if (bio && !page_is_mergeable(sbi: fio->sbi, bio, last_blkaddr: *fio->last_block, |
959 | cur_blkaddr: fio->new_blkaddr)) |
960 | f2fs_submit_merged_ipu_write(sbi: fio->sbi, bio: &bio, NULL); |
961 | alloc_new: |
962 | if (!bio) { |
963 | bio = __bio_alloc(fio, BIO_MAX_VECS); |
964 | f2fs_set_bio_crypt_ctx(bio, inode: fio->page->mapping->host, |
965 | first_idx: fio->page->index, fio, GFP_NOIO); |
966 | |
967 | add_bio_entry(sbi: fio->sbi, bio, page, temp: fio->temp); |
968 | } else { |
969 | if (add_ipu_page(fio, bio: &bio, page)) |
970 | goto alloc_new; |
971 | } |
972 | |
973 | if (fio->io_wbc) |
974 | wbc_account_cgroup_owner(wbc: fio->io_wbc, page: fio->page, PAGE_SIZE); |
975 | |
976 | inc_page_count(sbi: fio->sbi, WB_DATA_TYPE(page)); |
977 | |
978 | *fio->last_block = fio->new_blkaddr; |
979 | *fio->bio = bio; |
980 | |
981 | return 0; |
982 | } |
983 | |
984 | #ifdef CONFIG_BLK_DEV_ZONED |
985 | static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr) |
986 | { |
987 | int devi = 0; |
988 | |
989 | if (f2fs_is_multi_device(sbi)) { |
990 | devi = f2fs_target_device_index(sbi, blkaddr); |
991 | if (blkaddr < FDEV(devi).start_blk || |
992 | blkaddr > FDEV(devi).end_blk) { |
993 | f2fs_err(sbi, "Invalid block %x" , blkaddr); |
994 | return false; |
995 | } |
996 | blkaddr -= FDEV(devi).start_blk; |
997 | } |
998 | return bdev_zoned_model(FDEV(devi).bdev) == BLK_ZONED_HM && |
999 | f2fs_blkz_is_seq(sbi, devi, blkaddr) && |
1000 | (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1); |
1001 | } |
1002 | #endif |
1003 | |
1004 | void f2fs_submit_page_write(struct f2fs_io_info *fio) |
1005 | { |
1006 | struct f2fs_sb_info *sbi = fio->sbi; |
1007 | enum page_type btype = PAGE_TYPE_OF_BIO(fio->type); |
1008 | struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp; |
1009 | struct page *bio_page; |
1010 | |
1011 | f2fs_bug_on(sbi, is_read_io(fio->op)); |
1012 | |
1013 | f2fs_down_write(sem: &io->io_rwsem); |
1014 | |
1015 | #ifdef CONFIG_BLK_DEV_ZONED |
1016 | if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) { |
1017 | wait_for_completion_io(&io->zone_wait); |
1018 | bio_put(io->zone_pending_bio); |
1019 | io->zone_pending_bio = NULL; |
1020 | io->bi_private = NULL; |
1021 | } |
1022 | #endif |
1023 | |
1024 | next: |
1025 | if (fio->in_list) { |
1026 | spin_lock(lock: &io->io_lock); |
1027 | if (list_empty(head: &io->io_list)) { |
1028 | spin_unlock(lock: &io->io_lock); |
1029 | goto out; |
1030 | } |
1031 | fio = list_first_entry(&io->io_list, |
1032 | struct f2fs_io_info, list); |
1033 | list_del(entry: &fio->list); |
1034 | spin_unlock(lock: &io->io_lock); |
1035 | } |
1036 | |
1037 | verify_fio_blkaddr(fio); |
1038 | |
1039 | if (fio->encrypted_page) |
1040 | bio_page = fio->encrypted_page; |
1041 | else if (fio->compressed_page) |
1042 | bio_page = fio->compressed_page; |
1043 | else |
1044 | bio_page = fio->page; |
1045 | |
1046 | /* set submitted = true as a return value */ |
1047 | fio->submitted = 1; |
1048 | |
1049 | inc_page_count(sbi, WB_DATA_TYPE(bio_page)); |
1050 | |
1051 | if (io->bio && |
1052 | (!io_is_mergeable(sbi, bio: io->bio, io, fio, last_blkaddr: io->last_block_in_bio, |
1053 | cur_blkaddr: fio->new_blkaddr) || |
1054 | !f2fs_crypt_mergeable_bio(bio: io->bio, inode: fio->page->mapping->host, |
1055 | next_idx: bio_page->index, fio))) |
1056 | __submit_merged_bio(io); |
1057 | alloc_new: |
1058 | if (io->bio == NULL) { |
1059 | if (F2FS_IO_ALIGNED(sbi) && |
1060 | (fio->type == DATA || fio->type == NODE) && |
1061 | fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) { |
1062 | dec_page_count(sbi, WB_DATA_TYPE(bio_page)); |
1063 | fio->retry = 1; |
1064 | goto skip; |
1065 | } |
1066 | io->bio = __bio_alloc(fio, BIO_MAX_VECS); |
1067 | f2fs_set_bio_crypt_ctx(bio: io->bio, inode: fio->page->mapping->host, |
1068 | first_idx: bio_page->index, fio, GFP_NOIO); |
1069 | io->fio = *fio; |
1070 | } |
1071 | |
1072 | if (bio_add_page(bio: io->bio, page: bio_page, PAGE_SIZE, off: 0) < PAGE_SIZE) { |
1073 | __submit_merged_bio(io); |
1074 | goto alloc_new; |
1075 | } |
1076 | |
1077 | if (fio->io_wbc) |
1078 | wbc_account_cgroup_owner(wbc: fio->io_wbc, page: fio->page, PAGE_SIZE); |
1079 | |
1080 | io->last_block_in_bio = fio->new_blkaddr; |
1081 | |
1082 | trace_f2fs_submit_page_write(page: fio->page, fio); |
1083 | skip: |
1084 | if (fio->in_list) |
1085 | goto next; |
1086 | out: |
1087 | #ifdef CONFIG_BLK_DEV_ZONED |
1088 | if (f2fs_sb_has_blkzoned(sbi) && btype < META && |
1089 | is_end_zone_blkaddr(sbi, blkaddr: fio->new_blkaddr)) { |
1090 | bio_get(bio: io->bio); |
1091 | reinit_completion(x: &io->zone_wait); |
1092 | io->bi_private = io->bio->bi_private; |
1093 | io->bio->bi_private = io; |
1094 | io->bio->bi_end_io = f2fs_zone_write_end_io; |
1095 | io->zone_pending_bio = io->bio; |
1096 | __submit_merged_bio(io); |
1097 | } |
1098 | #endif |
1099 | if (is_sbi_flag_set(sbi, type: SBI_IS_SHUTDOWN) || |
1100 | !f2fs_is_checkpoint_ready(sbi)) |
1101 | __submit_merged_bio(io); |
1102 | f2fs_up_write(sem: &io->io_rwsem); |
1103 | } |
1104 | |
1105 | static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr, |
1106 | unsigned nr_pages, blk_opf_t op_flag, |
1107 | pgoff_t first_idx, bool for_write) |
1108 | { |
1109 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
1110 | struct bio *bio; |
1111 | struct bio_post_read_ctx *ctx = NULL; |
1112 | unsigned int post_read_steps = 0; |
1113 | sector_t sector; |
1114 | struct block_device *bdev = f2fs_target_device(sbi, blk_addr: blkaddr, sector: §or); |
1115 | |
1116 | bio = bio_alloc_bioset(bdev, nr_vecs: bio_max_segs(nr_segs: nr_pages), |
1117 | opf: REQ_OP_READ | op_flag, |
1118 | gfp_mask: for_write ? GFP_NOIO : GFP_KERNEL, bs: &f2fs_bioset); |
1119 | if (!bio) |
1120 | return ERR_PTR(error: -ENOMEM); |
1121 | bio->bi_iter.bi_sector = sector; |
1122 | f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS); |
1123 | bio->bi_end_io = f2fs_read_end_io; |
1124 | |
1125 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) |
1126 | post_read_steps |= STEP_DECRYPT; |
1127 | |
1128 | if (f2fs_need_verity(inode, idx: first_idx)) |
1129 | post_read_steps |= STEP_VERITY; |
1130 | |
1131 | /* |
1132 | * STEP_DECOMPRESS is handled specially, since a compressed file might |
1133 | * contain both compressed and uncompressed clusters. We'll allocate a |
1134 | * bio_post_read_ctx if the file is compressed, but the caller is |
1135 | * responsible for enabling STEP_DECOMPRESS if it's actually needed. |
1136 | */ |
1137 | |
1138 | if (post_read_steps || f2fs_compressed_file(inode)) { |
1139 | /* Due to the mempool, this never fails. */ |
1140 | ctx = mempool_alloc(pool: bio_post_read_ctx_pool, GFP_NOFS); |
1141 | ctx->bio = bio; |
1142 | ctx->sbi = sbi; |
1143 | ctx->enabled_steps = post_read_steps; |
1144 | ctx->fs_blkaddr = blkaddr; |
1145 | ctx->decompression_attempted = false; |
1146 | bio->bi_private = ctx; |
1147 | } |
1148 | iostat_alloc_and_bind_ctx(sbi, bio, ctx); |
1149 | |
1150 | return bio; |
1151 | } |
1152 | |
1153 | /* This can handle encryption stuffs */ |
1154 | static int f2fs_submit_page_read(struct inode *inode, struct page *page, |
1155 | block_t blkaddr, blk_opf_t op_flags, |
1156 | bool for_write) |
1157 | { |
1158 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
1159 | struct bio *bio; |
1160 | |
1161 | bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages: 1, op_flag: op_flags, |
1162 | first_idx: page->index, for_write); |
1163 | if (IS_ERR(ptr: bio)) |
1164 | return PTR_ERR(ptr: bio); |
1165 | |
1166 | /* wait for GCed page writeback via META_MAPPING */ |
1167 | f2fs_wait_on_block_writeback(inode, blkaddr); |
1168 | |
1169 | if (bio_add_page(bio, page, PAGE_SIZE, off: 0) < PAGE_SIZE) { |
1170 | iostat_update_and_unbind_ctx(bio); |
1171 | if (bio->bi_private) |
1172 | mempool_free(element: bio->bi_private, pool: bio_post_read_ctx_pool); |
1173 | bio_put(bio); |
1174 | return -EFAULT; |
1175 | } |
1176 | inc_page_count(sbi, count_type: F2FS_RD_DATA); |
1177 | f2fs_update_iostat(sbi, NULL, type: FS_DATA_READ_IO, F2FS_BLKSIZE); |
1178 | f2fs_submit_read_bio(sbi, bio, type: DATA); |
1179 | return 0; |
1180 | } |
1181 | |
1182 | static void __set_data_blkaddr(struct dnode_of_data *dn) |
1183 | { |
1184 | struct f2fs_node *rn = F2FS_NODE(page: dn->node_page); |
1185 | __le32 *addr_array; |
1186 | int base = 0; |
1187 | |
1188 | if (IS_INODE(page: dn->node_page) && f2fs_has_extra_attr(inode: dn->inode)) |
1189 | base = get_extra_isize(inode: dn->inode); |
1190 | |
1191 | /* Get physical address of data block */ |
1192 | addr_array = blkaddr_in_node(node: rn); |
1193 | addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr); |
1194 | } |
1195 | |
1196 | /* |
1197 | * Lock ordering for the change of data block address: |
1198 | * ->data_page |
1199 | * ->node_page |
1200 | * update block addresses in the node page |
1201 | */ |
1202 | void f2fs_set_data_blkaddr(struct dnode_of_data *dn) |
1203 | { |
1204 | f2fs_wait_on_page_writeback(page: dn->node_page, type: NODE, ordered: true, locked: true); |
1205 | __set_data_blkaddr(dn); |
1206 | if (set_page_dirty(dn->node_page)) |
1207 | dn->node_changed = true; |
1208 | } |
1209 | |
1210 | void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr) |
1211 | { |
1212 | dn->data_blkaddr = blkaddr; |
1213 | f2fs_set_data_blkaddr(dn); |
1214 | f2fs_update_read_extent_cache(dn); |
1215 | } |
1216 | |
1217 | /* dn->ofs_in_node will be returned with up-to-date last block pointer */ |
1218 | int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count) |
1219 | { |
1220 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode: dn->inode); |
1221 | int err; |
1222 | |
1223 | if (!count) |
1224 | return 0; |
1225 | |
1226 | if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) |
1227 | return -EPERM; |
1228 | if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count)))) |
1229 | return err; |
1230 | |
1231 | trace_f2fs_reserve_new_blocks(inode: dn->inode, nid: dn->nid, |
1232 | ofs_in_node: dn->ofs_in_node, count); |
1233 | |
1234 | f2fs_wait_on_page_writeback(page: dn->node_page, type: NODE, ordered: true, locked: true); |
1235 | |
1236 | for (; count > 0; dn->ofs_in_node++) { |
1237 | block_t blkaddr = f2fs_data_blkaddr(dn); |
1238 | |
1239 | if (blkaddr == NULL_ADDR) { |
1240 | dn->data_blkaddr = NEW_ADDR; |
1241 | __set_data_blkaddr(dn); |
1242 | count--; |
1243 | } |
1244 | } |
1245 | |
1246 | if (set_page_dirty(dn->node_page)) |
1247 | dn->node_changed = true; |
1248 | return 0; |
1249 | } |
1250 | |
1251 | /* Should keep dn->ofs_in_node unchanged */ |
1252 | int f2fs_reserve_new_block(struct dnode_of_data *dn) |
1253 | { |
1254 | unsigned int ofs_in_node = dn->ofs_in_node; |
1255 | int ret; |
1256 | |
1257 | ret = f2fs_reserve_new_blocks(dn, count: 1); |
1258 | dn->ofs_in_node = ofs_in_node; |
1259 | return ret; |
1260 | } |
1261 | |
1262 | int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index) |
1263 | { |
1264 | bool need_put = dn->inode_page ? false : true; |
1265 | int err; |
1266 | |
1267 | err = f2fs_get_dnode_of_data(dn, index, mode: ALLOC_NODE); |
1268 | if (err) |
1269 | return err; |
1270 | |
1271 | if (dn->data_blkaddr == NULL_ADDR) |
1272 | err = f2fs_reserve_new_block(dn); |
1273 | if (err || need_put) |
1274 | f2fs_put_dnode(dn); |
1275 | return err; |
1276 | } |
1277 | |
1278 | struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index, |
1279 | blk_opf_t op_flags, bool for_write, |
1280 | pgoff_t *next_pgofs) |
1281 | { |
1282 | struct address_space *mapping = inode->i_mapping; |
1283 | struct dnode_of_data dn; |
1284 | struct page *page; |
1285 | int err; |
1286 | |
1287 | page = f2fs_grab_cache_page(mapping, index, for_write); |
1288 | if (!page) |
1289 | return ERR_PTR(error: -ENOMEM); |
1290 | |
1291 | if (f2fs_lookup_read_extent_cache_block(inode, index, |
1292 | blkaddr: &dn.data_blkaddr)) { |
1293 | if (!f2fs_is_valid_blkaddr(sbi: F2FS_I_SB(inode), blkaddr: dn.data_blkaddr, |
1294 | type: DATA_GENERIC_ENHANCE_READ)) { |
1295 | err = -EFSCORRUPTED; |
1296 | f2fs_handle_error(sbi: F2FS_I_SB(inode), |
1297 | error: ERROR_INVALID_BLKADDR); |
1298 | goto put_err; |
1299 | } |
1300 | goto got_it; |
1301 | } |
1302 | |
1303 | set_new_dnode(dn: &dn, inode, NULL, NULL, nid: 0); |
1304 | err = f2fs_get_dnode_of_data(dn: &dn, index, mode: LOOKUP_NODE); |
1305 | if (err) { |
1306 | if (err == -ENOENT && next_pgofs) |
1307 | *next_pgofs = f2fs_get_next_page_offset(dn: &dn, pgofs: index); |
1308 | goto put_err; |
1309 | } |
1310 | f2fs_put_dnode(dn: &dn); |
1311 | |
1312 | if (unlikely(dn.data_blkaddr == NULL_ADDR)) { |
1313 | err = -ENOENT; |
1314 | if (next_pgofs) |
1315 | *next_pgofs = index + 1; |
1316 | goto put_err; |
1317 | } |
1318 | if (dn.data_blkaddr != NEW_ADDR && |
1319 | !f2fs_is_valid_blkaddr(sbi: F2FS_I_SB(inode), |
1320 | blkaddr: dn.data_blkaddr, |
1321 | type: DATA_GENERIC_ENHANCE)) { |
1322 | err = -EFSCORRUPTED; |
1323 | f2fs_handle_error(sbi: F2FS_I_SB(inode), |
1324 | error: ERROR_INVALID_BLKADDR); |
1325 | goto put_err; |
1326 | } |
1327 | got_it: |
1328 | if (PageUptodate(page)) { |
1329 | unlock_page(page); |
1330 | return page; |
1331 | } |
1332 | |
1333 | /* |
1334 | * A new dentry page is allocated but not able to be written, since its |
1335 | * new inode page couldn't be allocated due to -ENOSPC. |
1336 | * In such the case, its blkaddr can be remained as NEW_ADDR. |
1337 | * see, f2fs_add_link -> f2fs_get_new_data_page -> |
1338 | * f2fs_init_inode_metadata. |
1339 | */ |
1340 | if (dn.data_blkaddr == NEW_ADDR) { |
1341 | zero_user_segment(page, start: 0, PAGE_SIZE); |
1342 | if (!PageUptodate(page)) |
1343 | SetPageUptodate(page); |
1344 | unlock_page(page); |
1345 | return page; |
1346 | } |
1347 | |
1348 | err = f2fs_submit_page_read(inode, page, blkaddr: dn.data_blkaddr, |
1349 | op_flags, for_write); |
1350 | if (err) |
1351 | goto put_err; |
1352 | return page; |
1353 | |
1354 | put_err: |
1355 | f2fs_put_page(page, unlock: 1); |
1356 | return ERR_PTR(error: err); |
1357 | } |
1358 | |
1359 | struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index, |
1360 | pgoff_t *next_pgofs) |
1361 | { |
1362 | struct address_space *mapping = inode->i_mapping; |
1363 | struct page *page; |
1364 | |
1365 | page = find_get_page(mapping, offset: index); |
1366 | if (page && PageUptodate(page)) |
1367 | return page; |
1368 | f2fs_put_page(page, unlock: 0); |
1369 | |
1370 | page = f2fs_get_read_data_page(inode, index, op_flags: 0, for_write: false, next_pgofs); |
1371 | if (IS_ERR(ptr: page)) |
1372 | return page; |
1373 | |
1374 | if (PageUptodate(page)) |
1375 | return page; |
1376 | |
1377 | wait_on_page_locked(page); |
1378 | if (unlikely(!PageUptodate(page))) { |
1379 | f2fs_put_page(page, unlock: 0); |
1380 | return ERR_PTR(error: -EIO); |
1381 | } |
1382 | return page; |
1383 | } |
1384 | |
1385 | /* |
1386 | * If it tries to access a hole, return an error. |
1387 | * Because, the callers, functions in dir.c and GC, should be able to know |
1388 | * whether this page exists or not. |
1389 | */ |
1390 | struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index, |
1391 | bool for_write) |
1392 | { |
1393 | struct address_space *mapping = inode->i_mapping; |
1394 | struct page *page; |
1395 | |
1396 | page = f2fs_get_read_data_page(inode, index, op_flags: 0, for_write, NULL); |
1397 | if (IS_ERR(ptr: page)) |
1398 | return page; |
1399 | |
1400 | /* wait for read completion */ |
1401 | lock_page(page); |
1402 | if (unlikely(page->mapping != mapping || !PageUptodate(page))) { |
1403 | f2fs_put_page(page, unlock: 1); |
1404 | return ERR_PTR(error: -EIO); |
1405 | } |
1406 | return page; |
1407 | } |
1408 | |
1409 | /* |
1410 | * Caller ensures that this data page is never allocated. |
1411 | * A new zero-filled data page is allocated in the page cache. |
1412 | * |
1413 | * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and |
1414 | * f2fs_unlock_op(). |
1415 | * Note that, ipage is set only by make_empty_dir, and if any error occur, |
1416 | * ipage should be released by this function. |
1417 | */ |
1418 | struct page *f2fs_get_new_data_page(struct inode *inode, |
1419 | struct page *ipage, pgoff_t index, bool new_i_size) |
1420 | { |
1421 | struct address_space *mapping = inode->i_mapping; |
1422 | struct page *page; |
1423 | struct dnode_of_data dn; |
1424 | int err; |
1425 | |
1426 | page = f2fs_grab_cache_page(mapping, index, for_write: true); |
1427 | if (!page) { |
1428 | /* |
1429 | * before exiting, we should make sure ipage will be released |
1430 | * if any error occur. |
1431 | */ |
1432 | f2fs_put_page(page: ipage, unlock: 1); |
1433 | return ERR_PTR(error: -ENOMEM); |
1434 | } |
1435 | |
1436 | set_new_dnode(dn: &dn, inode, ipage, NULL, nid: 0); |
1437 | err = f2fs_reserve_block(dn: &dn, index); |
1438 | if (err) { |
1439 | f2fs_put_page(page, unlock: 1); |
1440 | return ERR_PTR(error: err); |
1441 | } |
1442 | if (!ipage) |
1443 | f2fs_put_dnode(dn: &dn); |
1444 | |
1445 | if (PageUptodate(page)) |
1446 | goto got_it; |
1447 | |
1448 | if (dn.data_blkaddr == NEW_ADDR) { |
1449 | zero_user_segment(page, start: 0, PAGE_SIZE); |
1450 | if (!PageUptodate(page)) |
1451 | SetPageUptodate(page); |
1452 | } else { |
1453 | f2fs_put_page(page, unlock: 1); |
1454 | |
1455 | /* if ipage exists, blkaddr should be NEW_ADDR */ |
1456 | f2fs_bug_on(F2FS_I_SB(inode), ipage); |
1457 | page = f2fs_get_lock_data_page(inode, index, for_write: true); |
1458 | if (IS_ERR(ptr: page)) |
1459 | return page; |
1460 | } |
1461 | got_it: |
1462 | if (new_i_size && i_size_read(inode) < |
1463 | ((loff_t)(index + 1) << PAGE_SHIFT)) |
1464 | f2fs_i_size_write(inode, i_size: ((loff_t)(index + 1) << PAGE_SHIFT)); |
1465 | return page; |
1466 | } |
1467 | |
1468 | static int __allocate_data_block(struct dnode_of_data *dn, int seg_type) |
1469 | { |
1470 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode: dn->inode); |
1471 | struct f2fs_summary sum; |
1472 | struct node_info ni; |
1473 | block_t old_blkaddr; |
1474 | blkcnt_t count = 1; |
1475 | int err; |
1476 | |
1477 | if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC))) |
1478 | return -EPERM; |
1479 | |
1480 | err = f2fs_get_node_info(sbi, nid: dn->nid, ni: &ni, checkpoint_context: false); |
1481 | if (err) |
1482 | return err; |
1483 | |
1484 | dn->data_blkaddr = f2fs_data_blkaddr(dn); |
1485 | if (dn->data_blkaddr == NULL_ADDR) { |
1486 | err = inc_valid_block_count(sbi, inode: dn->inode, count: &count); |
1487 | if (unlikely(err)) |
1488 | return err; |
1489 | } |
1490 | |
1491 | set_summary(sum: &sum, nid: dn->nid, ofs_in_node: dn->ofs_in_node, version: ni.version); |
1492 | old_blkaddr = dn->data_blkaddr; |
1493 | f2fs_allocate_data_block(sbi, NULL, old_blkaddr, new_blkaddr: &dn->data_blkaddr, |
1494 | sum: &sum, type: seg_type, NULL); |
1495 | if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) { |
1496 | invalidate_mapping_pages(mapping: META_MAPPING(sbi), |
1497 | start: old_blkaddr, end: old_blkaddr); |
1498 | f2fs_invalidate_compress_page(sbi, blkaddr: old_blkaddr); |
1499 | } |
1500 | f2fs_update_data_blkaddr(dn, blkaddr: dn->data_blkaddr); |
1501 | return 0; |
1502 | } |
1503 | |
1504 | static void f2fs_map_lock(struct f2fs_sb_info *sbi, int flag) |
1505 | { |
1506 | if (flag == F2FS_GET_BLOCK_PRE_AIO) |
1507 | f2fs_down_read(sem: &sbi->node_change); |
1508 | else |
1509 | f2fs_lock_op(sbi); |
1510 | } |
1511 | |
1512 | static void f2fs_map_unlock(struct f2fs_sb_info *sbi, int flag) |
1513 | { |
1514 | if (flag == F2FS_GET_BLOCK_PRE_AIO) |
1515 | f2fs_up_read(sem: &sbi->node_change); |
1516 | else |
1517 | f2fs_unlock_op(sbi); |
1518 | } |
1519 | |
1520 | int f2fs_get_block_locked(struct dnode_of_data *dn, pgoff_t index) |
1521 | { |
1522 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode: dn->inode); |
1523 | int err = 0; |
1524 | |
1525 | f2fs_map_lock(sbi, flag: F2FS_GET_BLOCK_PRE_AIO); |
1526 | if (!f2fs_lookup_read_extent_cache_block(inode: dn->inode, index, |
1527 | blkaddr: &dn->data_blkaddr)) |
1528 | err = f2fs_reserve_block(dn, index); |
1529 | f2fs_map_unlock(sbi, flag: F2FS_GET_BLOCK_PRE_AIO); |
1530 | |
1531 | return err; |
1532 | } |
1533 | |
1534 | static int f2fs_map_no_dnode(struct inode *inode, |
1535 | struct f2fs_map_blocks *map, struct dnode_of_data *dn, |
1536 | pgoff_t pgoff) |
1537 | { |
1538 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
1539 | |
1540 | /* |
1541 | * There is one exceptional case that read_node_page() may return |
1542 | * -ENOENT due to filesystem has been shutdown or cp_error, return |
1543 | * -EIO in that case. |
1544 | */ |
1545 | if (map->m_may_create && |
1546 | (is_sbi_flag_set(sbi, type: SBI_IS_SHUTDOWN) || f2fs_cp_error(sbi))) |
1547 | return -EIO; |
1548 | |
1549 | if (map->m_next_pgofs) |
1550 | *map->m_next_pgofs = f2fs_get_next_page_offset(dn, pgofs: pgoff); |
1551 | if (map->m_next_extent) |
1552 | *map->m_next_extent = f2fs_get_next_page_offset(dn, pgofs: pgoff); |
1553 | return 0; |
1554 | } |
1555 | |
1556 | static bool f2fs_map_blocks_cached(struct inode *inode, |
1557 | struct f2fs_map_blocks *map, int flag) |
1558 | { |
1559 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
1560 | unsigned int maxblocks = map->m_len; |
1561 | pgoff_t pgoff = (pgoff_t)map->m_lblk; |
1562 | struct extent_info ei = {}; |
1563 | |
1564 | if (!f2fs_lookup_read_extent_cache(inode, pgofs: pgoff, ei: &ei)) |
1565 | return false; |
1566 | |
1567 | map->m_pblk = ei.blk + pgoff - ei.fofs; |
1568 | map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgoff); |
1569 | map->m_flags = F2FS_MAP_MAPPED; |
1570 | if (map->m_next_extent) |
1571 | *map->m_next_extent = pgoff + map->m_len; |
1572 | |
1573 | /* for hardware encryption, but to avoid potential issue in future */ |
1574 | if (flag == F2FS_GET_BLOCK_DIO) |
1575 | f2fs_wait_on_block_writeback_range(inode, |
1576 | blkaddr: map->m_pblk, len: map->m_len); |
1577 | |
1578 | if (f2fs_allow_multi_device_dio(sbi, flag)) { |
1579 | int bidx = f2fs_target_device_index(sbi, blkaddr: map->m_pblk); |
1580 | struct f2fs_dev_info *dev = &sbi->devs[bidx]; |
1581 | |
1582 | map->m_bdev = dev->bdev; |
1583 | map->m_pblk -= dev->start_blk; |
1584 | map->m_len = min(map->m_len, dev->end_blk + 1 - map->m_pblk); |
1585 | } else { |
1586 | map->m_bdev = inode->i_sb->s_bdev; |
1587 | } |
1588 | return true; |
1589 | } |
1590 | |
1591 | /* |
1592 | * f2fs_map_blocks() tries to find or build mapping relationship which |
1593 | * maps continuous logical blocks to physical blocks, and return such |
1594 | * info via f2fs_map_blocks structure. |
1595 | */ |
1596 | int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, int flag) |
1597 | { |
1598 | unsigned int maxblocks = map->m_len; |
1599 | struct dnode_of_data dn; |
1600 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
1601 | int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE; |
1602 | pgoff_t pgofs, end_offset, end; |
1603 | int err = 0, ofs = 1; |
1604 | unsigned int ofs_in_node, last_ofs_in_node; |
1605 | blkcnt_t prealloc; |
1606 | block_t blkaddr; |
1607 | unsigned int start_pgofs; |
1608 | int bidx = 0; |
1609 | bool is_hole; |
1610 | |
1611 | if (!maxblocks) |
1612 | return 0; |
1613 | |
1614 | if (!map->m_may_create && f2fs_map_blocks_cached(inode, map, flag)) |
1615 | goto out; |
1616 | |
1617 | map->m_bdev = inode->i_sb->s_bdev; |
1618 | map->m_multidev_dio = |
1619 | f2fs_allow_multi_device_dio(sbi: F2FS_I_SB(inode), flag); |
1620 | |
1621 | map->m_len = 0; |
1622 | map->m_flags = 0; |
1623 | |
1624 | /* it only supports block size == page size */ |
1625 | pgofs = (pgoff_t)map->m_lblk; |
1626 | end = pgofs + maxblocks; |
1627 | |
1628 | next_dnode: |
1629 | if (map->m_may_create) |
1630 | f2fs_map_lock(sbi, flag); |
1631 | |
1632 | /* When reading holes, we need its node page */ |
1633 | set_new_dnode(dn: &dn, inode, NULL, NULL, nid: 0); |
1634 | err = f2fs_get_dnode_of_data(dn: &dn, index: pgofs, mode); |
1635 | if (err) { |
1636 | if (flag == F2FS_GET_BLOCK_BMAP) |
1637 | map->m_pblk = 0; |
1638 | if (err == -ENOENT) |
1639 | err = f2fs_map_no_dnode(inode, map, dn: &dn, pgoff: pgofs); |
1640 | goto unlock_out; |
1641 | } |
1642 | |
1643 | start_pgofs = pgofs; |
1644 | prealloc = 0; |
1645 | last_ofs_in_node = ofs_in_node = dn.ofs_in_node; |
1646 | end_offset = ADDRS_PER_PAGE(dn.node_page, inode); |
1647 | |
1648 | next_block: |
1649 | blkaddr = f2fs_data_blkaddr(dn: &dn); |
1650 | is_hole = !__is_valid_data_blkaddr(blkaddr); |
1651 | if (!is_hole && |
1652 | !f2fs_is_valid_blkaddr(sbi, blkaddr, type: DATA_GENERIC_ENHANCE)) { |
1653 | err = -EFSCORRUPTED; |
1654 | f2fs_handle_error(sbi, error: ERROR_INVALID_BLKADDR); |
1655 | goto sync_out; |
1656 | } |
1657 | |
1658 | /* use out-place-update for direct IO under LFS mode */ |
1659 | if (map->m_may_create && |
1660 | (is_hole || (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO))) { |
1661 | if (unlikely(f2fs_cp_error(sbi))) { |
1662 | err = -EIO; |
1663 | goto sync_out; |
1664 | } |
1665 | |
1666 | switch (flag) { |
1667 | case F2FS_GET_BLOCK_PRE_AIO: |
1668 | if (blkaddr == NULL_ADDR) { |
1669 | prealloc++; |
1670 | last_ofs_in_node = dn.ofs_in_node; |
1671 | } |
1672 | break; |
1673 | case F2FS_GET_BLOCK_PRE_DIO: |
1674 | case F2FS_GET_BLOCK_DIO: |
1675 | err = __allocate_data_block(dn: &dn, seg_type: map->m_seg_type); |
1676 | if (err) |
1677 | goto sync_out; |
1678 | if (flag == F2FS_GET_BLOCK_PRE_DIO) |
1679 | file_need_truncate(inode); |
1680 | set_inode_flag(inode, flag: FI_APPEND_WRITE); |
1681 | break; |
1682 | default: |
1683 | WARN_ON_ONCE(1); |
1684 | err = -EIO; |
1685 | goto sync_out; |
1686 | } |
1687 | |
1688 | blkaddr = dn.data_blkaddr; |
1689 | if (is_hole) |
1690 | map->m_flags |= F2FS_MAP_NEW; |
1691 | } else if (is_hole) { |
1692 | if (f2fs_compressed_file(inode) && |
1693 | f2fs_sanity_check_cluster(dn: &dn)) { |
1694 | err = -EFSCORRUPTED; |
1695 | f2fs_handle_error(sbi, |
1696 | error: ERROR_CORRUPTED_CLUSTER); |
1697 | goto sync_out; |
1698 | } |
1699 | |
1700 | switch (flag) { |
1701 | case F2FS_GET_BLOCK_PRECACHE: |
1702 | goto sync_out; |
1703 | case F2FS_GET_BLOCK_BMAP: |
1704 | map->m_pblk = 0; |
1705 | goto sync_out; |
1706 | case F2FS_GET_BLOCK_FIEMAP: |
1707 | if (blkaddr == NULL_ADDR) { |
1708 | if (map->m_next_pgofs) |
1709 | *map->m_next_pgofs = pgofs + 1; |
1710 | goto sync_out; |
1711 | } |
1712 | break; |
1713 | default: |
1714 | /* for defragment case */ |
1715 | if (map->m_next_pgofs) |
1716 | *map->m_next_pgofs = pgofs + 1; |
1717 | goto sync_out; |
1718 | } |
1719 | } |
1720 | |
1721 | if (flag == F2FS_GET_BLOCK_PRE_AIO) |
1722 | goto skip; |
1723 | |
1724 | if (map->m_multidev_dio) |
1725 | bidx = f2fs_target_device_index(sbi, blkaddr); |
1726 | |
1727 | if (map->m_len == 0) { |
1728 | /* reserved delalloc block should be mapped for fiemap. */ |
1729 | if (blkaddr == NEW_ADDR) |
1730 | map->m_flags |= F2FS_MAP_DELALLOC; |
1731 | map->m_flags |= F2FS_MAP_MAPPED; |
1732 | |
1733 | map->m_pblk = blkaddr; |
1734 | map->m_len = 1; |
1735 | |
1736 | if (map->m_multidev_dio) |
1737 | map->m_bdev = FDEV(bidx).bdev; |
1738 | } else if ((map->m_pblk != NEW_ADDR && |
1739 | blkaddr == (map->m_pblk + ofs)) || |
1740 | (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) || |
1741 | flag == F2FS_GET_BLOCK_PRE_DIO) { |
1742 | if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev) |
1743 | goto sync_out; |
1744 | ofs++; |
1745 | map->m_len++; |
1746 | } else { |
1747 | goto sync_out; |
1748 | } |
1749 | |
1750 | skip: |
1751 | dn.ofs_in_node++; |
1752 | pgofs++; |
1753 | |
1754 | /* preallocate blocks in batch for one dnode page */ |
1755 | if (flag == F2FS_GET_BLOCK_PRE_AIO && |
1756 | (pgofs == end || dn.ofs_in_node == end_offset)) { |
1757 | |
1758 | dn.ofs_in_node = ofs_in_node; |
1759 | err = f2fs_reserve_new_blocks(dn: &dn, count: prealloc); |
1760 | if (err) |
1761 | goto sync_out; |
1762 | |
1763 | map->m_len += dn.ofs_in_node - ofs_in_node; |
1764 | if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) { |
1765 | err = -ENOSPC; |
1766 | goto sync_out; |
1767 | } |
1768 | dn.ofs_in_node = end_offset; |
1769 | } |
1770 | |
1771 | if (pgofs >= end) |
1772 | goto sync_out; |
1773 | else if (dn.ofs_in_node < end_offset) |
1774 | goto next_block; |
1775 | |
1776 | if (flag == F2FS_GET_BLOCK_PRECACHE) { |
1777 | if (map->m_flags & F2FS_MAP_MAPPED) { |
1778 | unsigned int ofs = start_pgofs - map->m_lblk; |
1779 | |
1780 | f2fs_update_read_extent_cache_range(dn: &dn, |
1781 | fofs: start_pgofs, blkaddr: map->m_pblk + ofs, |
1782 | len: map->m_len - ofs); |
1783 | } |
1784 | } |
1785 | |
1786 | f2fs_put_dnode(dn: &dn); |
1787 | |
1788 | if (map->m_may_create) { |
1789 | f2fs_map_unlock(sbi, flag); |
1790 | f2fs_balance_fs(sbi, need: dn.node_changed); |
1791 | } |
1792 | goto next_dnode; |
1793 | |
1794 | sync_out: |
1795 | |
1796 | if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) { |
1797 | /* |
1798 | * for hardware encryption, but to avoid potential issue |
1799 | * in future |
1800 | */ |
1801 | f2fs_wait_on_block_writeback_range(inode, |
1802 | blkaddr: map->m_pblk, len: map->m_len); |
1803 | |
1804 | if (map->m_multidev_dio) { |
1805 | block_t blk_addr = map->m_pblk; |
1806 | |
1807 | bidx = f2fs_target_device_index(sbi, blkaddr: map->m_pblk); |
1808 | |
1809 | map->m_bdev = FDEV(bidx).bdev; |
1810 | map->m_pblk -= FDEV(bidx).start_blk; |
1811 | |
1812 | if (map->m_may_create) |
1813 | f2fs_update_device_state(sbi, ino: inode->i_ino, |
1814 | blkaddr: blk_addr, blkcnt: map->m_len); |
1815 | |
1816 | f2fs_bug_on(sbi, blk_addr + map->m_len > |
1817 | FDEV(bidx).end_blk + 1); |
1818 | } |
1819 | } |
1820 | |
1821 | if (flag == F2FS_GET_BLOCK_PRECACHE) { |
1822 | if (map->m_flags & F2FS_MAP_MAPPED) { |
1823 | unsigned int ofs = start_pgofs - map->m_lblk; |
1824 | |
1825 | f2fs_update_read_extent_cache_range(dn: &dn, |
1826 | fofs: start_pgofs, blkaddr: map->m_pblk + ofs, |
1827 | len: map->m_len - ofs); |
1828 | } |
1829 | if (map->m_next_extent) |
1830 | *map->m_next_extent = pgofs + 1; |
1831 | } |
1832 | f2fs_put_dnode(dn: &dn); |
1833 | unlock_out: |
1834 | if (map->m_may_create) { |
1835 | f2fs_map_unlock(sbi, flag); |
1836 | f2fs_balance_fs(sbi, need: dn.node_changed); |
1837 | } |
1838 | out: |
1839 | trace_f2fs_map_blocks(inode, map, flag, ret: err); |
1840 | return err; |
1841 | } |
1842 | |
1843 | bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len) |
1844 | { |
1845 | struct f2fs_map_blocks map; |
1846 | block_t last_lblk; |
1847 | int err; |
1848 | |
1849 | if (pos + len > i_size_read(inode)) |
1850 | return false; |
1851 | |
1852 | map.m_lblk = F2FS_BYTES_TO_BLK(pos); |
1853 | map.m_next_pgofs = NULL; |
1854 | map.m_next_extent = NULL; |
1855 | map.m_seg_type = NO_CHECK_TYPE; |
1856 | map.m_may_create = false; |
1857 | last_lblk = F2FS_BLK_ALIGN(pos + len); |
1858 | |
1859 | while (map.m_lblk < last_lblk) { |
1860 | map.m_len = last_lblk - map.m_lblk; |
1861 | err = f2fs_map_blocks(inode, map: &map, flag: F2FS_GET_BLOCK_DEFAULT); |
1862 | if (err || map.m_len == 0) |
1863 | return false; |
1864 | map.m_lblk += map.m_len; |
1865 | } |
1866 | return true; |
1867 | } |
1868 | |
1869 | static inline u64 bytes_to_blks(struct inode *inode, u64 bytes) |
1870 | { |
1871 | return (bytes >> inode->i_blkbits); |
1872 | } |
1873 | |
1874 | static inline u64 blks_to_bytes(struct inode *inode, u64 blks) |
1875 | { |
1876 | return (blks << inode->i_blkbits); |
1877 | } |
1878 | |
1879 | static int f2fs_xattr_fiemap(struct inode *inode, |
1880 | struct fiemap_extent_info *fieinfo) |
1881 | { |
1882 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
1883 | struct page *page; |
1884 | struct node_info ni; |
1885 | __u64 phys = 0, len; |
1886 | __u32 flags; |
1887 | nid_t xnid = F2FS_I(inode)->i_xattr_nid; |
1888 | int err = 0; |
1889 | |
1890 | if (f2fs_has_inline_xattr(inode)) { |
1891 | int offset; |
1892 | |
1893 | page = f2fs_grab_cache_page(mapping: NODE_MAPPING(sbi), |
1894 | index: inode->i_ino, for_write: false); |
1895 | if (!page) |
1896 | return -ENOMEM; |
1897 | |
1898 | err = f2fs_get_node_info(sbi, nid: inode->i_ino, ni: &ni, checkpoint_context: false); |
1899 | if (err) { |
1900 | f2fs_put_page(page, unlock: 1); |
1901 | return err; |
1902 | } |
1903 | |
1904 | phys = blks_to_bytes(inode, blks: ni.blk_addr); |
1905 | offset = offsetof(struct f2fs_inode, i_addr) + |
1906 | sizeof(__le32) * (DEF_ADDRS_PER_INODE - |
1907 | get_inline_xattr_addrs(inode)); |
1908 | |
1909 | phys += offset; |
1910 | len = inline_xattr_size(inode); |
1911 | |
1912 | f2fs_put_page(page, unlock: 1); |
1913 | |
1914 | flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED; |
1915 | |
1916 | if (!xnid) |
1917 | flags |= FIEMAP_EXTENT_LAST; |
1918 | |
1919 | err = fiemap_fill_next_extent(info: fieinfo, logical: 0, phys, len, flags); |
1920 | trace_f2fs_fiemap(inode, lblock: 0, pblock: phys, len, flags, ret: err); |
1921 | if (err) |
1922 | return err; |
1923 | } |
1924 | |
1925 | if (xnid) { |
1926 | page = f2fs_grab_cache_page(mapping: NODE_MAPPING(sbi), index: xnid, for_write: false); |
1927 | if (!page) |
1928 | return -ENOMEM; |
1929 | |
1930 | err = f2fs_get_node_info(sbi, nid: xnid, ni: &ni, checkpoint_context: false); |
1931 | if (err) { |
1932 | f2fs_put_page(page, unlock: 1); |
1933 | return err; |
1934 | } |
1935 | |
1936 | phys = blks_to_bytes(inode, blks: ni.blk_addr); |
1937 | len = inode->i_sb->s_blocksize; |
1938 | |
1939 | f2fs_put_page(page, unlock: 1); |
1940 | |
1941 | flags = FIEMAP_EXTENT_LAST; |
1942 | } |
1943 | |
1944 | if (phys) { |
1945 | err = fiemap_fill_next_extent(info: fieinfo, logical: 0, phys, len, flags); |
1946 | trace_f2fs_fiemap(inode, lblock: 0, pblock: phys, len, flags, ret: err); |
1947 | } |
1948 | |
1949 | return (err < 0 ? err : 0); |
1950 | } |
1951 | |
1952 | static loff_t max_inode_blocks(struct inode *inode) |
1953 | { |
1954 | loff_t result = ADDRS_PER_INODE(inode); |
1955 | loff_t leaf_count = ADDRS_PER_BLOCK(inode); |
1956 | |
1957 | /* two direct node blocks */ |
1958 | result += (leaf_count * 2); |
1959 | |
1960 | /* two indirect node blocks */ |
1961 | leaf_count *= NIDS_PER_BLOCK; |
1962 | result += (leaf_count * 2); |
1963 | |
1964 | /* one double indirect node block */ |
1965 | leaf_count *= NIDS_PER_BLOCK; |
1966 | result += leaf_count; |
1967 | |
1968 | return result; |
1969 | } |
1970 | |
1971 | int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
1972 | u64 start, u64 len) |
1973 | { |
1974 | struct f2fs_map_blocks map; |
1975 | sector_t start_blk, last_blk; |
1976 | pgoff_t next_pgofs; |
1977 | u64 logical = 0, phys = 0, size = 0; |
1978 | u32 flags = 0; |
1979 | int ret = 0; |
1980 | bool compr_cluster = false, compr_appended; |
1981 | unsigned int cluster_size = F2FS_I(inode)->i_cluster_size; |
1982 | unsigned int count_in_cluster = 0; |
1983 | loff_t maxbytes; |
1984 | |
1985 | if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { |
1986 | ret = f2fs_precache_extents(inode); |
1987 | if (ret) |
1988 | return ret; |
1989 | } |
1990 | |
1991 | ret = fiemap_prep(inode, fieinfo, start, len: &len, FIEMAP_FLAG_XATTR); |
1992 | if (ret) |
1993 | return ret; |
1994 | |
1995 | inode_lock(inode); |
1996 | |
1997 | maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS; |
1998 | if (start > maxbytes) { |
1999 | ret = -EFBIG; |
2000 | goto out; |
2001 | } |
2002 | |
2003 | if (len > maxbytes || (maxbytes - len) < start) |
2004 | len = maxbytes - start; |
2005 | |
2006 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { |
2007 | ret = f2fs_xattr_fiemap(inode, fieinfo); |
2008 | goto out; |
2009 | } |
2010 | |
2011 | if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) { |
2012 | ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len); |
2013 | if (ret != -EAGAIN) |
2014 | goto out; |
2015 | } |
2016 | |
2017 | if (bytes_to_blks(inode, bytes: len) == 0) |
2018 | len = blks_to_bytes(inode, blks: 1); |
2019 | |
2020 | start_blk = bytes_to_blks(inode, bytes: start); |
2021 | last_blk = bytes_to_blks(inode, bytes: start + len - 1); |
2022 | |
2023 | next: |
2024 | memset(&map, 0, sizeof(map)); |
2025 | map.m_lblk = start_blk; |
2026 | map.m_len = bytes_to_blks(inode, bytes: len); |
2027 | map.m_next_pgofs = &next_pgofs; |
2028 | map.m_seg_type = NO_CHECK_TYPE; |
2029 | |
2030 | if (compr_cluster) { |
2031 | map.m_lblk += 1; |
2032 | map.m_len = cluster_size - count_in_cluster; |
2033 | } |
2034 | |
2035 | ret = f2fs_map_blocks(inode, map: &map, flag: F2FS_GET_BLOCK_FIEMAP); |
2036 | if (ret) |
2037 | goto out; |
2038 | |
2039 | /* HOLE */ |
2040 | if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) { |
2041 | start_blk = next_pgofs; |
2042 | |
2043 | if (blks_to_bytes(inode, blks: start_blk) < blks_to_bytes(inode, |
2044 | blks: max_inode_blocks(inode))) |
2045 | goto prep_next; |
2046 | |
2047 | flags |= FIEMAP_EXTENT_LAST; |
2048 | } |
2049 | |
2050 | compr_appended = false; |
2051 | /* In a case of compressed cluster, append this to the last extent */ |
2052 | if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) || |
2053 | !(map.m_flags & F2FS_MAP_FLAGS))) { |
2054 | compr_appended = true; |
2055 | goto skip_fill; |
2056 | } |
2057 | |
2058 | if (size) { |
2059 | flags |= FIEMAP_EXTENT_MERGED; |
2060 | if (IS_ENCRYPTED(inode)) |
2061 | flags |= FIEMAP_EXTENT_DATA_ENCRYPTED; |
2062 | |
2063 | ret = fiemap_fill_next_extent(info: fieinfo, logical, |
2064 | phys, len: size, flags); |
2065 | trace_f2fs_fiemap(inode, lblock: logical, pblock: phys, len: size, flags, ret); |
2066 | if (ret) |
2067 | goto out; |
2068 | size = 0; |
2069 | } |
2070 | |
2071 | if (start_blk > last_blk) |
2072 | goto out; |
2073 | |
2074 | skip_fill: |
2075 | if (map.m_pblk == COMPRESS_ADDR) { |
2076 | compr_cluster = true; |
2077 | count_in_cluster = 1; |
2078 | } else if (compr_appended) { |
2079 | unsigned int appended_blks = cluster_size - |
2080 | count_in_cluster + 1; |
2081 | size += blks_to_bytes(inode, blks: appended_blks); |
2082 | start_blk += appended_blks; |
2083 | compr_cluster = false; |
2084 | } else { |
2085 | logical = blks_to_bytes(inode, blks: start_blk); |
2086 | phys = __is_valid_data_blkaddr(blkaddr: map.m_pblk) ? |
2087 | blks_to_bytes(inode, blks: map.m_pblk) : 0; |
2088 | size = blks_to_bytes(inode, blks: map.m_len); |
2089 | flags = 0; |
2090 | |
2091 | if (compr_cluster) { |
2092 | flags = FIEMAP_EXTENT_ENCODED; |
2093 | count_in_cluster += map.m_len; |
2094 | if (count_in_cluster == cluster_size) { |
2095 | compr_cluster = false; |
2096 | size += blks_to_bytes(inode, blks: 1); |
2097 | } |
2098 | } else if (map.m_flags & F2FS_MAP_DELALLOC) { |
2099 | flags = FIEMAP_EXTENT_UNWRITTEN; |
2100 | } |
2101 | |
2102 | start_blk += bytes_to_blks(inode, bytes: size); |
2103 | } |
2104 | |
2105 | prep_next: |
2106 | cond_resched(); |
2107 | if (fatal_signal_pending(current)) |
2108 | ret = -EINTR; |
2109 | else |
2110 | goto next; |
2111 | out: |
2112 | if (ret == 1) |
2113 | ret = 0; |
2114 | |
2115 | inode_unlock(inode); |
2116 | return ret; |
2117 | } |
2118 | |
2119 | static inline loff_t f2fs_readpage_limit(struct inode *inode) |
2120 | { |
2121 | if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) |
2122 | return inode->i_sb->s_maxbytes; |
2123 | |
2124 | return i_size_read(inode); |
2125 | } |
2126 | |
2127 | static int f2fs_read_single_page(struct inode *inode, struct page *page, |
2128 | unsigned nr_pages, |
2129 | struct f2fs_map_blocks *map, |
2130 | struct bio **bio_ret, |
2131 | sector_t *last_block_in_bio, |
2132 | bool is_readahead) |
2133 | { |
2134 | struct bio *bio = *bio_ret; |
2135 | const unsigned blocksize = blks_to_bytes(inode, blks: 1); |
2136 | sector_t block_in_file; |
2137 | sector_t last_block; |
2138 | sector_t last_block_in_file; |
2139 | sector_t block_nr; |
2140 | int ret = 0; |
2141 | |
2142 | block_in_file = (sector_t)page_index(page); |
2143 | last_block = block_in_file + nr_pages; |
2144 | last_block_in_file = bytes_to_blks(inode, |
2145 | bytes: f2fs_readpage_limit(inode) + blocksize - 1); |
2146 | if (last_block > last_block_in_file) |
2147 | last_block = last_block_in_file; |
2148 | |
2149 | /* just zeroing out page which is beyond EOF */ |
2150 | if (block_in_file >= last_block) |
2151 | goto zero_out; |
2152 | /* |
2153 | * Map blocks using the previous result first. |
2154 | */ |
2155 | if ((map->m_flags & F2FS_MAP_MAPPED) && |
2156 | block_in_file > map->m_lblk && |
2157 | block_in_file < (map->m_lblk + map->m_len)) |
2158 | goto got_it; |
2159 | |
2160 | /* |
2161 | * Then do more f2fs_map_blocks() calls until we are |
2162 | * done with this page. |
2163 | */ |
2164 | map->m_lblk = block_in_file; |
2165 | map->m_len = last_block - block_in_file; |
2166 | |
2167 | ret = f2fs_map_blocks(inode, map, flag: F2FS_GET_BLOCK_DEFAULT); |
2168 | if (ret) |
2169 | goto out; |
2170 | got_it: |
2171 | if ((map->m_flags & F2FS_MAP_MAPPED)) { |
2172 | block_nr = map->m_pblk + block_in_file - map->m_lblk; |
2173 | SetPageMappedToDisk(page); |
2174 | |
2175 | if (!f2fs_is_valid_blkaddr(sbi: F2FS_I_SB(inode), blkaddr: block_nr, |
2176 | type: DATA_GENERIC_ENHANCE_READ)) { |
2177 | ret = -EFSCORRUPTED; |
2178 | f2fs_handle_error(sbi: F2FS_I_SB(inode), |
2179 | error: ERROR_INVALID_BLKADDR); |
2180 | goto out; |
2181 | } |
2182 | } else { |
2183 | zero_out: |
2184 | zero_user_segment(page, start: 0, PAGE_SIZE); |
2185 | if (f2fs_need_verity(inode, idx: page->index) && |
2186 | !fsverity_verify_page(page)) { |
2187 | ret = -EIO; |
2188 | goto out; |
2189 | } |
2190 | if (!PageUptodate(page)) |
2191 | SetPageUptodate(page); |
2192 | unlock_page(page); |
2193 | goto out; |
2194 | } |
2195 | |
2196 | /* |
2197 | * This page will go to BIO. Do we need to send this |
2198 | * BIO off first? |
2199 | */ |
2200 | if (bio && (!page_is_mergeable(sbi: F2FS_I_SB(inode), bio, |
2201 | last_blkaddr: *last_block_in_bio, cur_blkaddr: block_nr) || |
2202 | !f2fs_crypt_mergeable_bio(bio, inode, next_idx: page->index, NULL))) { |
2203 | submit_and_realloc: |
2204 | f2fs_submit_read_bio(sbi: F2FS_I_SB(inode), bio, type: DATA); |
2205 | bio = NULL; |
2206 | } |
2207 | if (bio == NULL) { |
2208 | bio = f2fs_grab_read_bio(inode, blkaddr: block_nr, nr_pages, |
2209 | op_flag: is_readahead ? REQ_RAHEAD : 0, first_idx: page->index, |
2210 | for_write: false); |
2211 | if (IS_ERR(ptr: bio)) { |
2212 | ret = PTR_ERR(ptr: bio); |
2213 | bio = NULL; |
2214 | goto out; |
2215 | } |
2216 | } |
2217 | |
2218 | /* |
2219 | * If the page is under writeback, we need to wait for |
2220 | * its completion to see the correct decrypted data. |
2221 | */ |
2222 | f2fs_wait_on_block_writeback(inode, blkaddr: block_nr); |
2223 | |
2224 | if (bio_add_page(bio, page, len: blocksize, off: 0) < blocksize) |
2225 | goto submit_and_realloc; |
2226 | |
2227 | inc_page_count(sbi: F2FS_I_SB(inode), count_type: F2FS_RD_DATA); |
2228 | f2fs_update_iostat(sbi: F2FS_I_SB(inode), NULL, type: FS_DATA_READ_IO, |
2229 | F2FS_BLKSIZE); |
2230 | *last_block_in_bio = block_nr; |
2231 | out: |
2232 | *bio_ret = bio; |
2233 | return ret; |
2234 | } |
2235 | |
2236 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
2237 | int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret, |
2238 | unsigned nr_pages, sector_t *last_block_in_bio, |
2239 | bool is_readahead, bool for_write) |
2240 | { |
2241 | struct dnode_of_data dn; |
2242 | struct inode *inode = cc->inode; |
2243 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
2244 | struct bio *bio = *bio_ret; |
2245 | unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size; |
2246 | sector_t last_block_in_file; |
2247 | const unsigned blocksize = blks_to_bytes(inode, blks: 1); |
2248 | struct decompress_io_ctx *dic = NULL; |
2249 | struct extent_info ei = {}; |
2250 | bool from_dnode = true; |
2251 | int i; |
2252 | int ret = 0; |
2253 | |
2254 | f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc)); |
2255 | |
2256 | last_block_in_file = bytes_to_blks(inode, |
2257 | bytes: f2fs_readpage_limit(inode) + blocksize - 1); |
2258 | |
2259 | /* get rid of pages beyond EOF */ |
2260 | for (i = 0; i < cc->cluster_size; i++) { |
2261 | struct page *page = cc->rpages[i]; |
2262 | |
2263 | if (!page) |
2264 | continue; |
2265 | if ((sector_t)page->index >= last_block_in_file) { |
2266 | zero_user_segment(page, start: 0, PAGE_SIZE); |
2267 | if (!PageUptodate(page)) |
2268 | SetPageUptodate(page); |
2269 | } else if (!PageUptodate(page)) { |
2270 | continue; |
2271 | } |
2272 | unlock_page(page); |
2273 | if (for_write) |
2274 | put_page(page); |
2275 | cc->rpages[i] = NULL; |
2276 | cc->nr_rpages--; |
2277 | } |
2278 | |
2279 | /* we are done since all pages are beyond EOF */ |
2280 | if (f2fs_cluster_is_empty(cc)) |
2281 | goto out; |
2282 | |
2283 | if (f2fs_lookup_read_extent_cache(inode, pgofs: start_idx, ei: &ei)) |
2284 | from_dnode = false; |
2285 | |
2286 | if (!from_dnode) |
2287 | goto skip_reading_dnode; |
2288 | |
2289 | set_new_dnode(dn: &dn, inode, NULL, NULL, nid: 0); |
2290 | ret = f2fs_get_dnode_of_data(dn: &dn, index: start_idx, mode: LOOKUP_NODE); |
2291 | if (ret) |
2292 | goto out; |
2293 | |
2294 | if (unlikely(f2fs_cp_error(sbi))) { |
2295 | ret = -EIO; |
2296 | goto out_put_dnode; |
2297 | } |
2298 | f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR); |
2299 | |
2300 | skip_reading_dnode: |
2301 | for (i = 1; i < cc->cluster_size; i++) { |
2302 | block_t blkaddr; |
2303 | |
2304 | blkaddr = from_dnode ? data_blkaddr(inode: dn.inode, node_page: dn.node_page, |
2305 | offset: dn.ofs_in_node + i) : |
2306 | ei.blk + i - 1; |
2307 | |
2308 | if (!__is_valid_data_blkaddr(blkaddr)) |
2309 | break; |
2310 | |
2311 | if (!f2fs_is_valid_blkaddr(sbi, blkaddr, type: DATA_GENERIC)) { |
2312 | ret = -EFAULT; |
2313 | goto out_put_dnode; |
2314 | } |
2315 | cc->nr_cpages++; |
2316 | |
2317 | if (!from_dnode && i >= ei.c_len) |
2318 | break; |
2319 | } |
2320 | |
2321 | /* nothing to decompress */ |
2322 | if (cc->nr_cpages == 0) { |
2323 | ret = 0; |
2324 | goto out_put_dnode; |
2325 | } |
2326 | |
2327 | dic = f2fs_alloc_dic(cc); |
2328 | if (IS_ERR(ptr: dic)) { |
2329 | ret = PTR_ERR(ptr: dic); |
2330 | goto out_put_dnode; |
2331 | } |
2332 | |
2333 | for (i = 0; i < cc->nr_cpages; i++) { |
2334 | struct page *page = dic->cpages[i]; |
2335 | block_t blkaddr; |
2336 | struct bio_post_read_ctx *ctx; |
2337 | |
2338 | blkaddr = from_dnode ? data_blkaddr(inode: dn.inode, node_page: dn.node_page, |
2339 | offset: dn.ofs_in_node + i + 1) : |
2340 | ei.blk + i; |
2341 | |
2342 | f2fs_wait_on_block_writeback(inode, blkaddr); |
2343 | |
2344 | if (f2fs_load_compressed_page(sbi, page, blkaddr)) { |
2345 | if (atomic_dec_and_test(v: &dic->remaining_pages)) { |
2346 | f2fs_decompress_cluster(dic, in_task: true); |
2347 | break; |
2348 | } |
2349 | continue; |
2350 | } |
2351 | |
2352 | if (bio && (!page_is_mergeable(sbi, bio, |
2353 | last_blkaddr: *last_block_in_bio, cur_blkaddr: blkaddr) || |
2354 | !f2fs_crypt_mergeable_bio(bio, inode, next_idx: page->index, NULL))) { |
2355 | submit_and_realloc: |
2356 | f2fs_submit_read_bio(sbi, bio, type: DATA); |
2357 | bio = NULL; |
2358 | } |
2359 | |
2360 | if (!bio) { |
2361 | bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages, |
2362 | op_flag: is_readahead ? REQ_RAHEAD : 0, |
2363 | first_idx: page->index, for_write); |
2364 | if (IS_ERR(ptr: bio)) { |
2365 | ret = PTR_ERR(ptr: bio); |
2366 | f2fs_decompress_end_io(dic, failed: ret, in_task: true); |
2367 | f2fs_put_dnode(dn: &dn); |
2368 | *bio_ret = NULL; |
2369 | return ret; |
2370 | } |
2371 | } |
2372 | |
2373 | if (bio_add_page(bio, page, len: blocksize, off: 0) < blocksize) |
2374 | goto submit_and_realloc; |
2375 | |
2376 | ctx = get_post_read_ctx(bio); |
2377 | ctx->enabled_steps |= STEP_DECOMPRESS; |
2378 | refcount_inc(r: &dic->refcnt); |
2379 | |
2380 | inc_page_count(sbi, count_type: F2FS_RD_DATA); |
2381 | f2fs_update_iostat(sbi, inode, type: FS_DATA_READ_IO, F2FS_BLKSIZE); |
2382 | *last_block_in_bio = blkaddr; |
2383 | } |
2384 | |
2385 | if (from_dnode) |
2386 | f2fs_put_dnode(dn: &dn); |
2387 | |
2388 | *bio_ret = bio; |
2389 | return 0; |
2390 | |
2391 | out_put_dnode: |
2392 | if (from_dnode) |
2393 | f2fs_put_dnode(dn: &dn); |
2394 | out: |
2395 | for (i = 0; i < cc->cluster_size; i++) { |
2396 | if (cc->rpages[i]) { |
2397 | ClearPageUptodate(page: cc->rpages[i]); |
2398 | unlock_page(page: cc->rpages[i]); |
2399 | } |
2400 | } |
2401 | *bio_ret = bio; |
2402 | return ret; |
2403 | } |
2404 | #endif |
2405 | |
2406 | /* |
2407 | * This function was originally taken from fs/mpage.c, and customized for f2fs. |
2408 | * Major change was from block_size == page_size in f2fs by default. |
2409 | */ |
2410 | static int f2fs_mpage_readpages(struct inode *inode, |
2411 | struct readahead_control *rac, struct page *page) |
2412 | { |
2413 | struct bio *bio = NULL; |
2414 | sector_t last_block_in_bio = 0; |
2415 | struct f2fs_map_blocks map; |
2416 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
2417 | struct compress_ctx cc = { |
2418 | .inode = inode, |
2419 | .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, |
2420 | .cluster_size = F2FS_I(inode)->i_cluster_size, |
2421 | .cluster_idx = NULL_CLUSTER, |
2422 | .rpages = NULL, |
2423 | .cpages = NULL, |
2424 | .nr_rpages = 0, |
2425 | .nr_cpages = 0, |
2426 | }; |
2427 | pgoff_t nc_cluster_idx = NULL_CLUSTER; |
2428 | #endif |
2429 | unsigned nr_pages = rac ? readahead_count(rac) : 1; |
2430 | unsigned max_nr_pages = nr_pages; |
2431 | int ret = 0; |
2432 | |
2433 | map.m_pblk = 0; |
2434 | map.m_lblk = 0; |
2435 | map.m_len = 0; |
2436 | map.m_flags = 0; |
2437 | map.m_next_pgofs = NULL; |
2438 | map.m_next_extent = NULL; |
2439 | map.m_seg_type = NO_CHECK_TYPE; |
2440 | map.m_may_create = false; |
2441 | |
2442 | for (; nr_pages; nr_pages--) { |
2443 | if (rac) { |
2444 | page = readahead_page(ractl: rac); |
2445 | prefetchw(x: &page->flags); |
2446 | } |
2447 | |
2448 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
2449 | if (f2fs_compressed_file(inode)) { |
2450 | /* there are remained compressed pages, submit them */ |
2451 | if (!f2fs_cluster_can_merge_page(cc: &cc, index: page->index)) { |
2452 | ret = f2fs_read_multi_pages(cc: &cc, bio_ret: &bio, |
2453 | nr_pages: max_nr_pages, |
2454 | last_block_in_bio: &last_block_in_bio, |
2455 | is_readahead: rac != NULL, for_write: false); |
2456 | f2fs_destroy_compress_ctx(cc: &cc, reuse: false); |
2457 | if (ret) |
2458 | goto set_error_page; |
2459 | } |
2460 | if (cc.cluster_idx == NULL_CLUSTER) { |
2461 | if (nc_cluster_idx == |
2462 | page->index >> cc.log_cluster_size) { |
2463 | goto read_single_page; |
2464 | } |
2465 | |
2466 | ret = f2fs_is_compressed_cluster(inode, index: page->index); |
2467 | if (ret < 0) |
2468 | goto set_error_page; |
2469 | else if (!ret) { |
2470 | nc_cluster_idx = |
2471 | page->index >> cc.log_cluster_size; |
2472 | goto read_single_page; |
2473 | } |
2474 | |
2475 | nc_cluster_idx = NULL_CLUSTER; |
2476 | } |
2477 | ret = f2fs_init_compress_ctx(cc: &cc); |
2478 | if (ret) |
2479 | goto set_error_page; |
2480 | |
2481 | f2fs_compress_ctx_add_page(cc: &cc, page); |
2482 | |
2483 | goto next_page; |
2484 | } |
2485 | read_single_page: |
2486 | #endif |
2487 | |
2488 | ret = f2fs_read_single_page(inode, page, nr_pages: max_nr_pages, map: &map, |
2489 | bio_ret: &bio, last_block_in_bio: &last_block_in_bio, is_readahead: rac); |
2490 | if (ret) { |
2491 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
2492 | set_error_page: |
2493 | #endif |
2494 | zero_user_segment(page, start: 0, PAGE_SIZE); |
2495 | unlock_page(page); |
2496 | } |
2497 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
2498 | next_page: |
2499 | #endif |
2500 | if (rac) |
2501 | put_page(page); |
2502 | |
2503 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
2504 | if (f2fs_compressed_file(inode)) { |
2505 | /* last page */ |
2506 | if (nr_pages == 1 && !f2fs_cluster_is_empty(cc: &cc)) { |
2507 | ret = f2fs_read_multi_pages(cc: &cc, bio_ret: &bio, |
2508 | nr_pages: max_nr_pages, |
2509 | last_block_in_bio: &last_block_in_bio, |
2510 | is_readahead: rac != NULL, for_write: false); |
2511 | f2fs_destroy_compress_ctx(cc: &cc, reuse: false); |
2512 | } |
2513 | } |
2514 | #endif |
2515 | } |
2516 | if (bio) |
2517 | f2fs_submit_read_bio(sbi: F2FS_I_SB(inode), bio, type: DATA); |
2518 | return ret; |
2519 | } |
2520 | |
2521 | static int f2fs_read_data_folio(struct file *file, struct folio *folio) |
2522 | { |
2523 | struct page *page = &folio->page; |
2524 | struct inode *inode = page_file_mapping(page)->host; |
2525 | int ret = -EAGAIN; |
2526 | |
2527 | trace_f2fs_readpage(page, type: DATA); |
2528 | |
2529 | if (!f2fs_is_compress_backend_ready(inode)) { |
2530 | unlock_page(page); |
2531 | return -EOPNOTSUPP; |
2532 | } |
2533 | |
2534 | /* If the file has inline data, try to read it directly */ |
2535 | if (f2fs_has_inline_data(inode)) |
2536 | ret = f2fs_read_inline_data(inode, page); |
2537 | if (ret == -EAGAIN) |
2538 | ret = f2fs_mpage_readpages(inode, NULL, page); |
2539 | return ret; |
2540 | } |
2541 | |
2542 | static void f2fs_readahead(struct readahead_control *rac) |
2543 | { |
2544 | struct inode *inode = rac->mapping->host; |
2545 | |
2546 | trace_f2fs_readpages(inode, start: readahead_index(rac), nrpage: readahead_count(rac)); |
2547 | |
2548 | if (!f2fs_is_compress_backend_ready(inode)) |
2549 | return; |
2550 | |
2551 | /* If the file has inline data, skip readahead */ |
2552 | if (f2fs_has_inline_data(inode)) |
2553 | return; |
2554 | |
2555 | f2fs_mpage_readpages(inode, rac, NULL); |
2556 | } |
2557 | |
2558 | int f2fs_encrypt_one_page(struct f2fs_io_info *fio) |
2559 | { |
2560 | struct inode *inode = fio->page->mapping->host; |
2561 | struct page *mpage, *page; |
2562 | gfp_t gfp_flags = GFP_NOFS; |
2563 | |
2564 | if (!f2fs_encrypted_file(inode)) |
2565 | return 0; |
2566 | |
2567 | page = fio->compressed_page ? fio->compressed_page : fio->page; |
2568 | |
2569 | /* wait for GCed page writeback via META_MAPPING */ |
2570 | f2fs_wait_on_block_writeback(inode, blkaddr: fio->old_blkaddr); |
2571 | |
2572 | if (fscrypt_inode_uses_inline_crypto(inode)) |
2573 | return 0; |
2574 | |
2575 | retry_encrypt: |
2576 | fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page, |
2577 | PAGE_SIZE, offs: 0, gfp_flags); |
2578 | if (IS_ERR(ptr: fio->encrypted_page)) { |
2579 | /* flush pending IOs and wait for a while in the ENOMEM case */ |
2580 | if (PTR_ERR(ptr: fio->encrypted_page) == -ENOMEM) { |
2581 | f2fs_flush_merged_writes(sbi: fio->sbi); |
2582 | memalloc_retry_wait(GFP_NOFS); |
2583 | gfp_flags |= __GFP_NOFAIL; |
2584 | goto retry_encrypt; |
2585 | } |
2586 | return PTR_ERR(ptr: fio->encrypted_page); |
2587 | } |
2588 | |
2589 | mpage = find_lock_page(mapping: META_MAPPING(sbi: fio->sbi), index: fio->old_blkaddr); |
2590 | if (mpage) { |
2591 | if (PageUptodate(page: mpage)) |
2592 | memcpy(page_address(mpage), |
2593 | page_address(fio->encrypted_page), PAGE_SIZE); |
2594 | f2fs_put_page(page: mpage, unlock: 1); |
2595 | } |
2596 | return 0; |
2597 | } |
2598 | |
2599 | static inline bool check_inplace_update_policy(struct inode *inode, |
2600 | struct f2fs_io_info *fio) |
2601 | { |
2602 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
2603 | |
2604 | if (IS_F2FS_IPU_HONOR_OPU_WRITE(sbi) && |
2605 | is_inode_flag_set(inode, flag: FI_OPU_WRITE)) |
2606 | return false; |
2607 | if (IS_F2FS_IPU_FORCE(sbi)) |
2608 | return true; |
2609 | if (IS_F2FS_IPU_SSR(sbi) && f2fs_need_SSR(sbi)) |
2610 | return true; |
2611 | if (IS_F2FS_IPU_UTIL(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util) |
2612 | return true; |
2613 | if (IS_F2FS_IPU_SSR_UTIL(sbi) && f2fs_need_SSR(sbi) && |
2614 | utilization(sbi) > SM_I(sbi)->min_ipu_util) |
2615 | return true; |
2616 | |
2617 | /* |
2618 | * IPU for rewrite async pages |
2619 | */ |
2620 | if (IS_F2FS_IPU_ASYNC(sbi) && fio && fio->op == REQ_OP_WRITE && |
2621 | !(fio->op_flags & REQ_SYNC) && !IS_ENCRYPTED(inode)) |
2622 | return true; |
2623 | |
2624 | /* this is only set during fdatasync */ |
2625 | if (IS_F2FS_IPU_FSYNC(sbi) && is_inode_flag_set(inode, flag: FI_NEED_IPU)) |
2626 | return true; |
2627 | |
2628 | if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) && |
2629 | !f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) |
2630 | return true; |
2631 | |
2632 | return false; |
2633 | } |
2634 | |
2635 | bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio) |
2636 | { |
2637 | /* swap file is migrating in aligned write mode */ |
2638 | if (is_inode_flag_set(inode, flag: FI_ALIGNED_WRITE)) |
2639 | return false; |
2640 | |
2641 | if (f2fs_is_pinned_file(inode)) |
2642 | return true; |
2643 | |
2644 | /* if this is cold file, we should overwrite to avoid fragmentation */ |
2645 | if (file_is_cold(inode) && !is_inode_flag_set(inode, flag: FI_OPU_WRITE)) |
2646 | return true; |
2647 | |
2648 | return check_inplace_update_policy(inode, fio); |
2649 | } |
2650 | |
2651 | bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) |
2652 | { |
2653 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
2654 | |
2655 | /* The below cases were checked when setting it. */ |
2656 | if (f2fs_is_pinned_file(inode)) |
2657 | return false; |
2658 | if (fio && is_sbi_flag_set(sbi, type: SBI_NEED_FSCK)) |
2659 | return true; |
2660 | if (f2fs_lfs_mode(sbi)) |
2661 | return true; |
2662 | if (S_ISDIR(inode->i_mode)) |
2663 | return true; |
2664 | if (IS_NOQUOTA(inode)) |
2665 | return true; |
2666 | if (f2fs_is_atomic_file(inode)) |
2667 | return true; |
2668 | /* rewrite low ratio compress data w/ OPU mode to avoid fragmentation */ |
2669 | if (f2fs_compressed_file(inode) && |
2670 | F2FS_OPTION(sbi).compress_mode == COMPR_MODE_USER && |
2671 | is_inode_flag_set(inode, flag: FI_ENABLE_COMPRESS)) |
2672 | return true; |
2673 | |
2674 | /* swap file is migrating in aligned write mode */ |
2675 | if (is_inode_flag_set(inode, flag: FI_ALIGNED_WRITE)) |
2676 | return true; |
2677 | |
2678 | if (is_inode_flag_set(inode, flag: FI_OPU_WRITE)) |
2679 | return true; |
2680 | |
2681 | if (fio) { |
2682 | if (page_private_gcing(page: fio->page)) |
2683 | return true; |
2684 | if (page_private_dummy(page: fio->page)) |
2685 | return true; |
2686 | if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) && |
2687 | f2fs_is_checkpointed_data(sbi, fio->old_blkaddr))) |
2688 | return true; |
2689 | } |
2690 | return false; |
2691 | } |
2692 | |
2693 | static inline bool need_inplace_update(struct f2fs_io_info *fio) |
2694 | { |
2695 | struct inode *inode = fio->page->mapping->host; |
2696 | |
2697 | if (f2fs_should_update_outplace(inode, fio)) |
2698 | return false; |
2699 | |
2700 | return f2fs_should_update_inplace(inode, fio); |
2701 | } |
2702 | |
2703 | int f2fs_do_write_data_page(struct f2fs_io_info *fio) |
2704 | { |
2705 | struct page *page = fio->page; |
2706 | struct inode *inode = page->mapping->host; |
2707 | struct dnode_of_data dn; |
2708 | struct node_info ni; |
2709 | bool ipu_force = false; |
2710 | int err = 0; |
2711 | |
2712 | /* Use COW inode to make dnode_of_data for atomic write */ |
2713 | if (f2fs_is_atomic_file(inode)) |
2714 | set_new_dnode(dn: &dn, inode: F2FS_I(inode)->cow_inode, NULL, NULL, nid: 0); |
2715 | else |
2716 | set_new_dnode(dn: &dn, inode, NULL, NULL, nid: 0); |
2717 | |
2718 | if (need_inplace_update(fio) && |
2719 | f2fs_lookup_read_extent_cache_block(inode, index: page->index, |
2720 | blkaddr: &fio->old_blkaddr)) { |
2721 | if (!f2fs_is_valid_blkaddr(sbi: fio->sbi, blkaddr: fio->old_blkaddr, |
2722 | type: DATA_GENERIC_ENHANCE)) { |
2723 | f2fs_handle_error(sbi: fio->sbi, |
2724 | error: ERROR_INVALID_BLKADDR); |
2725 | return -EFSCORRUPTED; |
2726 | } |
2727 | |
2728 | ipu_force = true; |
2729 | fio->need_lock = LOCK_DONE; |
2730 | goto got_it; |
2731 | } |
2732 | |
2733 | /* Deadlock due to between page->lock and f2fs_lock_op */ |
2734 | if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(sbi: fio->sbi)) |
2735 | return -EAGAIN; |
2736 | |
2737 | err = f2fs_get_dnode_of_data(dn: &dn, index: page->index, mode: LOOKUP_NODE); |
2738 | if (err) |
2739 | goto out; |
2740 | |
2741 | fio->old_blkaddr = dn.data_blkaddr; |
2742 | |
2743 | /* This page is already truncated */ |
2744 | if (fio->old_blkaddr == NULL_ADDR) { |
2745 | ClearPageUptodate(page); |
2746 | clear_page_private_gcing(page); |
2747 | goto out_writepage; |
2748 | } |
2749 | got_it: |
2750 | if (__is_valid_data_blkaddr(blkaddr: fio->old_blkaddr) && |
2751 | !f2fs_is_valid_blkaddr(sbi: fio->sbi, blkaddr: fio->old_blkaddr, |
2752 | type: DATA_GENERIC_ENHANCE)) { |
2753 | err = -EFSCORRUPTED; |
2754 | f2fs_handle_error(sbi: fio->sbi, error: ERROR_INVALID_BLKADDR); |
2755 | goto out_writepage; |
2756 | } |
2757 | |
2758 | /* |
2759 | * If current allocation needs SSR, |
2760 | * it had better in-place writes for updated data. |
2761 | */ |
2762 | if (ipu_force || |
2763 | (__is_valid_data_blkaddr(blkaddr: fio->old_blkaddr) && |
2764 | need_inplace_update(fio))) { |
2765 | err = f2fs_encrypt_one_page(fio); |
2766 | if (err) |
2767 | goto out_writepage; |
2768 | |
2769 | set_page_writeback(page); |
2770 | f2fs_put_dnode(dn: &dn); |
2771 | if (fio->need_lock == LOCK_REQ) |
2772 | f2fs_unlock_op(sbi: fio->sbi); |
2773 | err = f2fs_inplace_write_data(fio); |
2774 | if (err) { |
2775 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) |
2776 | fscrypt_finalize_bounce_page(pagep: &fio->encrypted_page); |
2777 | if (PageWriteback(page)) |
2778 | end_page_writeback(page); |
2779 | } else { |
2780 | set_inode_flag(inode, flag: FI_UPDATE_WRITE); |
2781 | } |
2782 | trace_f2fs_do_write_data_page(page: fio->page, type: IPU); |
2783 | return err; |
2784 | } |
2785 | |
2786 | if (fio->need_lock == LOCK_RETRY) { |
2787 | if (!f2fs_trylock_op(sbi: fio->sbi)) { |
2788 | err = -EAGAIN; |
2789 | goto out_writepage; |
2790 | } |
2791 | fio->need_lock = LOCK_REQ; |
2792 | } |
2793 | |
2794 | err = f2fs_get_node_info(sbi: fio->sbi, nid: dn.nid, ni: &ni, checkpoint_context: false); |
2795 | if (err) |
2796 | goto out_writepage; |
2797 | |
2798 | fio->version = ni.version; |
2799 | |
2800 | err = f2fs_encrypt_one_page(fio); |
2801 | if (err) |
2802 | goto out_writepage; |
2803 | |
2804 | set_page_writeback(page); |
2805 | |
2806 | if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR) |
2807 | f2fs_i_compr_blocks_update(inode, blocks: fio->compr_blocks - 1, add: false); |
2808 | |
2809 | /* LFS mode write path */ |
2810 | f2fs_outplace_write_data(dn: &dn, fio); |
2811 | trace_f2fs_do_write_data_page(page, type: OPU); |
2812 | set_inode_flag(inode, flag: FI_APPEND_WRITE); |
2813 | if (page->index == 0) |
2814 | set_inode_flag(inode, flag: FI_FIRST_BLOCK_WRITTEN); |
2815 | out_writepage: |
2816 | f2fs_put_dnode(dn: &dn); |
2817 | out: |
2818 | if (fio->need_lock == LOCK_REQ) |
2819 | f2fs_unlock_op(sbi: fio->sbi); |
2820 | return err; |
2821 | } |
2822 | |
2823 | int f2fs_write_single_data_page(struct page *page, int *submitted, |
2824 | struct bio **bio, |
2825 | sector_t *last_block, |
2826 | struct writeback_control *wbc, |
2827 | enum iostat_type io_type, |
2828 | int compr_blocks, |
2829 | bool allow_balance) |
2830 | { |
2831 | struct inode *inode = page->mapping->host; |
2832 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
2833 | loff_t i_size = i_size_read(inode); |
2834 | const pgoff_t end_index = ((unsigned long long)i_size) |
2835 | >> PAGE_SHIFT; |
2836 | loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT; |
2837 | unsigned offset = 0; |
2838 | bool need_balance_fs = false; |
2839 | bool quota_inode = IS_NOQUOTA(inode); |
2840 | int err = 0; |
2841 | struct f2fs_io_info fio = { |
2842 | .sbi = sbi, |
2843 | .ino = inode->i_ino, |
2844 | .type = DATA, |
2845 | .op = REQ_OP_WRITE, |
2846 | .op_flags = wbc_to_write_flags(wbc), |
2847 | .old_blkaddr = NULL_ADDR, |
2848 | .page = page, |
2849 | .encrypted_page = NULL, |
2850 | .submitted = 0, |
2851 | .compr_blocks = compr_blocks, |
2852 | .need_lock = LOCK_RETRY, |
2853 | .post_read = f2fs_post_read_required(inode) ? 1 : 0, |
2854 | .io_type = io_type, |
2855 | .io_wbc = wbc, |
2856 | .bio = bio, |
2857 | .last_block = last_block, |
2858 | }; |
2859 | |
2860 | trace_f2fs_writepage(page, type: DATA); |
2861 | |
2862 | /* we should bypass data pages to proceed the kworker jobs */ |
2863 | if (unlikely(f2fs_cp_error(sbi))) { |
2864 | mapping_set_error(mapping: page->mapping, error: -EIO); |
2865 | /* |
2866 | * don't drop any dirty dentry pages for keeping lastest |
2867 | * directory structure. |
2868 | */ |
2869 | if (S_ISDIR(inode->i_mode) && |
2870 | !is_sbi_flag_set(sbi, type: SBI_IS_CLOSE)) |
2871 | goto redirty_out; |
2872 | |
2873 | /* keep data pages in remount-ro mode */ |
2874 | if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY) |
2875 | goto redirty_out; |
2876 | goto out; |
2877 | } |
2878 | |
2879 | if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) |
2880 | goto redirty_out; |
2881 | |
2882 | if (page->index < end_index || |
2883 | f2fs_verity_in_progress(inode) || |
2884 | compr_blocks) |
2885 | goto write; |
2886 | |
2887 | /* |
2888 | * If the offset is out-of-range of file size, |
2889 | * this page does not have to be written to disk. |
2890 | */ |
2891 | offset = i_size & (PAGE_SIZE - 1); |
2892 | if ((page->index >= end_index + 1) || !offset) |
2893 | goto out; |
2894 | |
2895 | zero_user_segment(page, start: offset, PAGE_SIZE); |
2896 | write: |
2897 | if (f2fs_is_drop_cache(inode)) |
2898 | goto out; |
2899 | |
2900 | /* Dentry/quota blocks are controlled by checkpoint */ |
2901 | if (S_ISDIR(inode->i_mode) || quota_inode) { |
2902 | /* |
2903 | * We need to wait for node_write to avoid block allocation during |
2904 | * checkpoint. This can only happen to quota writes which can cause |
2905 | * the below discard race condition. |
2906 | */ |
2907 | if (quota_inode) |
2908 | f2fs_down_read(sem: &sbi->node_write); |
2909 | |
2910 | fio.need_lock = LOCK_DONE; |
2911 | err = f2fs_do_write_data_page(fio: &fio); |
2912 | |
2913 | if (quota_inode) |
2914 | f2fs_up_read(sem: &sbi->node_write); |
2915 | |
2916 | goto done; |
2917 | } |
2918 | |
2919 | if (!wbc->for_reclaim) |
2920 | need_balance_fs = true; |
2921 | else if (has_not_enough_free_secs(sbi, freed: 0, needed: 0)) |
2922 | goto redirty_out; |
2923 | else |
2924 | set_inode_flag(inode, flag: FI_HOT_DATA); |
2925 | |
2926 | err = -EAGAIN; |
2927 | if (f2fs_has_inline_data(inode)) { |
2928 | err = f2fs_write_inline_data(inode, page); |
2929 | if (!err) |
2930 | goto out; |
2931 | } |
2932 | |
2933 | if (err == -EAGAIN) { |
2934 | err = f2fs_do_write_data_page(fio: &fio); |
2935 | if (err == -EAGAIN) { |
2936 | fio.need_lock = LOCK_REQ; |
2937 | err = f2fs_do_write_data_page(fio: &fio); |
2938 | } |
2939 | } |
2940 | |
2941 | if (err) { |
2942 | file_set_keep_isize(inode); |
2943 | } else { |
2944 | spin_lock(lock: &F2FS_I(inode)->i_size_lock); |
2945 | if (F2FS_I(inode)->last_disk_size < psize) |
2946 | F2FS_I(inode)->last_disk_size = psize; |
2947 | spin_unlock(lock: &F2FS_I(inode)->i_size_lock); |
2948 | } |
2949 | |
2950 | done: |
2951 | if (err && err != -ENOENT) |
2952 | goto redirty_out; |
2953 | |
2954 | out: |
2955 | inode_dec_dirty_pages(inode); |
2956 | if (err) { |
2957 | ClearPageUptodate(page); |
2958 | clear_page_private_gcing(page); |
2959 | } |
2960 | |
2961 | if (wbc->for_reclaim) { |
2962 | f2fs_submit_merged_write_cond(sbi, NULL, page, ino: 0, type: DATA); |
2963 | clear_inode_flag(inode, flag: FI_HOT_DATA); |
2964 | f2fs_remove_dirty_inode(inode); |
2965 | submitted = NULL; |
2966 | } |
2967 | unlock_page(page); |
2968 | if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) && |
2969 | !F2FS_I(inode)->wb_task && allow_balance) |
2970 | f2fs_balance_fs(sbi, need: need_balance_fs); |
2971 | |
2972 | if (unlikely(f2fs_cp_error(sbi))) { |
2973 | f2fs_submit_merged_write(sbi, type: DATA); |
2974 | if (bio && *bio) |
2975 | f2fs_submit_merged_ipu_write(sbi, bio, NULL); |
2976 | submitted = NULL; |
2977 | } |
2978 | |
2979 | if (submitted) |
2980 | *submitted = fio.submitted; |
2981 | |
2982 | return 0; |
2983 | |
2984 | redirty_out: |
2985 | redirty_page_for_writepage(wbc, page); |
2986 | /* |
2987 | * pageout() in MM translates EAGAIN, so calls handle_write_error() |
2988 | * -> mapping_set_error() -> set_bit(AS_EIO, ...). |
2989 | * file_write_and_wait_range() will see EIO error, which is critical |
2990 | * to return value of fsync() followed by atomic_write failure to user. |
2991 | */ |
2992 | if (!err || wbc->for_reclaim) |
2993 | return AOP_WRITEPAGE_ACTIVATE; |
2994 | unlock_page(page); |
2995 | return err; |
2996 | } |
2997 | |
2998 | static int f2fs_write_data_page(struct page *page, |
2999 | struct writeback_control *wbc) |
3000 | { |
3001 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3002 | struct inode *inode = page->mapping->host; |
3003 | |
3004 | if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) |
3005 | goto out; |
3006 | |
3007 | if (f2fs_compressed_file(inode)) { |
3008 | if (f2fs_is_compressed_cluster(inode, index: page->index)) { |
3009 | redirty_page_for_writepage(wbc, page); |
3010 | return AOP_WRITEPAGE_ACTIVATE; |
3011 | } |
3012 | } |
3013 | out: |
3014 | #endif |
3015 | |
3016 | return f2fs_write_single_data_page(page, NULL, NULL, NULL, |
3017 | wbc, io_type: FS_DATA_IO, compr_blocks: 0, allow_balance: true); |
3018 | } |
3019 | |
3020 | /* |
3021 | * This function was copied from write_cache_pages from mm/page-writeback.c. |
3022 | * The major change is making write step of cold data page separately from |
3023 | * warm/hot data page. |
3024 | */ |
3025 | static int f2fs_write_cache_pages(struct address_space *mapping, |
3026 | struct writeback_control *wbc, |
3027 | enum iostat_type io_type) |
3028 | { |
3029 | int ret = 0; |
3030 | int done = 0, retry = 0; |
3031 | struct page *pages_local[F2FS_ONSTACK_PAGES]; |
3032 | struct page **pages = pages_local; |
3033 | struct folio_batch fbatch; |
3034 | struct f2fs_sb_info *sbi = F2FS_M_SB(mapping); |
3035 | struct bio *bio = NULL; |
3036 | sector_t last_block; |
3037 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3038 | struct inode *inode = mapping->host; |
3039 | struct compress_ctx cc = { |
3040 | .inode = inode, |
3041 | .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, |
3042 | .cluster_size = F2FS_I(inode)->i_cluster_size, |
3043 | .cluster_idx = NULL_CLUSTER, |
3044 | .rpages = NULL, |
3045 | .nr_rpages = 0, |
3046 | .cpages = NULL, |
3047 | .valid_nr_cpages = 0, |
3048 | .rbuf = NULL, |
3049 | .cbuf = NULL, |
3050 | .rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size, |
3051 | .private = NULL, |
3052 | }; |
3053 | #endif |
3054 | int nr_folios, p, idx; |
3055 | int nr_pages; |
3056 | unsigned int max_pages = F2FS_ONSTACK_PAGES; |
3057 | pgoff_t index; |
3058 | pgoff_t end; /* Inclusive */ |
3059 | pgoff_t done_index; |
3060 | int range_whole = 0; |
3061 | xa_mark_t tag; |
3062 | int nwritten = 0; |
3063 | int submitted = 0; |
3064 | int i; |
3065 | |
3066 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3067 | if (f2fs_compressed_file(inode) && |
3068 | 1 << cc.log_cluster_size > F2FS_ONSTACK_PAGES) { |
3069 | pages = f2fs_kzalloc(sbi, size: sizeof(struct page *) << |
3070 | cc.log_cluster_size, GFP_NOFS | __GFP_NOFAIL); |
3071 | max_pages = 1 << cc.log_cluster_size; |
3072 | } |
3073 | #endif |
3074 | |
3075 | folio_batch_init(fbatch: &fbatch); |
3076 | |
3077 | if (get_dirty_pages(inode: mapping->host) <= |
3078 | SM_I(sbi: F2FS_M_SB(mapping))->min_hot_blocks) |
3079 | set_inode_flag(inode: mapping->host, flag: FI_HOT_DATA); |
3080 | else |
3081 | clear_inode_flag(inode: mapping->host, flag: FI_HOT_DATA); |
3082 | |
3083 | if (wbc->range_cyclic) { |
3084 | index = mapping->writeback_index; /* prev offset */ |
3085 | end = -1; |
3086 | } else { |
3087 | index = wbc->range_start >> PAGE_SHIFT; |
3088 | end = wbc->range_end >> PAGE_SHIFT; |
3089 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
3090 | range_whole = 1; |
3091 | } |
3092 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
3093 | tag = PAGECACHE_TAG_TOWRITE; |
3094 | else |
3095 | tag = PAGECACHE_TAG_DIRTY; |
3096 | retry: |
3097 | retry = 0; |
3098 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
3099 | tag_pages_for_writeback(mapping, start: index, end); |
3100 | done_index = index; |
3101 | while (!done && !retry && (index <= end)) { |
3102 | nr_pages = 0; |
3103 | again: |
3104 | nr_folios = filemap_get_folios_tag(mapping, start: &index, end, |
3105 | tag, fbatch: &fbatch); |
3106 | if (nr_folios == 0) { |
3107 | if (nr_pages) |
3108 | goto write; |
3109 | break; |
3110 | } |
3111 | |
3112 | for (i = 0; i < nr_folios; i++) { |
3113 | struct folio *folio = fbatch.folios[i]; |
3114 | |
3115 | idx = 0; |
3116 | p = folio_nr_pages(folio); |
3117 | add_more: |
3118 | pages[nr_pages] = folio_page(folio, idx); |
3119 | folio_get(folio); |
3120 | if (++nr_pages == max_pages) { |
3121 | index = folio->index + idx + 1; |
3122 | folio_batch_release(fbatch: &fbatch); |
3123 | goto write; |
3124 | } |
3125 | if (++idx < p) |
3126 | goto add_more; |
3127 | } |
3128 | folio_batch_release(fbatch: &fbatch); |
3129 | goto again; |
3130 | write: |
3131 | for (i = 0; i < nr_pages; i++) { |
3132 | struct page *page = pages[i]; |
3133 | struct folio *folio = page_folio(page); |
3134 | bool need_readd; |
3135 | readd: |
3136 | need_readd = false; |
3137 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3138 | if (f2fs_compressed_file(inode)) { |
3139 | void *fsdata = NULL; |
3140 | struct page *pagep; |
3141 | int ret2; |
3142 | |
3143 | ret = f2fs_init_compress_ctx(cc: &cc); |
3144 | if (ret) { |
3145 | done = 1; |
3146 | break; |
3147 | } |
3148 | |
3149 | if (!f2fs_cluster_can_merge_page(cc: &cc, |
3150 | index: folio->index)) { |
3151 | ret = f2fs_write_multi_pages(cc: &cc, |
3152 | submitted: &submitted, wbc, io_type); |
3153 | if (!ret) |
3154 | need_readd = true; |
3155 | goto result; |
3156 | } |
3157 | |
3158 | if (unlikely(f2fs_cp_error(sbi))) |
3159 | goto lock_folio; |
3160 | |
3161 | if (!f2fs_cluster_is_empty(cc: &cc)) |
3162 | goto lock_folio; |
3163 | |
3164 | if (f2fs_all_cluster_page_ready(cc: &cc, |
3165 | pages, index: i, nr_pages, uptodate: true)) |
3166 | goto lock_folio; |
3167 | |
3168 | ret2 = f2fs_prepare_compress_overwrite( |
3169 | inode, pagep: &pagep, |
3170 | index: folio->index, fsdata: &fsdata); |
3171 | if (ret2 < 0) { |
3172 | ret = ret2; |
3173 | done = 1; |
3174 | break; |
3175 | } else if (ret2 && |
3176 | (!f2fs_compress_write_end(inode, |
3177 | fsdata, index: folio->index, copied: 1) || |
3178 | !f2fs_all_cluster_page_ready(cc: &cc, |
3179 | pages, index: i, nr_pages, |
3180 | uptodate: false))) { |
3181 | retry = 1; |
3182 | break; |
3183 | } |
3184 | } |
3185 | #endif |
3186 | /* give a priority to WB_SYNC threads */ |
3187 | if (atomic_read(v: &sbi->wb_sync_req[DATA]) && |
3188 | wbc->sync_mode == WB_SYNC_NONE) { |
3189 | done = 1; |
3190 | break; |
3191 | } |
3192 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3193 | lock_folio: |
3194 | #endif |
3195 | done_index = folio->index; |
3196 | retry_write: |
3197 | folio_lock(folio); |
3198 | |
3199 | if (unlikely(folio->mapping != mapping)) { |
3200 | continue_unlock: |
3201 | folio_unlock(folio); |
3202 | continue; |
3203 | } |
3204 | |
3205 | if (!folio_test_dirty(folio)) { |
3206 | /* someone wrote it for us */ |
3207 | goto continue_unlock; |
3208 | } |
3209 | |
3210 | if (folio_test_writeback(folio)) { |
3211 | if (wbc->sync_mode == WB_SYNC_NONE) |
3212 | goto continue_unlock; |
3213 | f2fs_wait_on_page_writeback(page: &folio->page, type: DATA, ordered: true, locked: true); |
3214 | } |
3215 | |
3216 | if (!folio_clear_dirty_for_io(folio)) |
3217 | goto continue_unlock; |
3218 | |
3219 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3220 | if (f2fs_compressed_file(inode)) { |
3221 | folio_get(folio); |
3222 | f2fs_compress_ctx_add_page(cc: &cc, page: &folio->page); |
3223 | continue; |
3224 | } |
3225 | #endif |
3226 | ret = f2fs_write_single_data_page(page: &folio->page, |
3227 | submitted: &submitted, bio: &bio, last_block: &last_block, |
3228 | wbc, io_type, compr_blocks: 0, allow_balance: true); |
3229 | if (ret == AOP_WRITEPAGE_ACTIVATE) |
3230 | folio_unlock(folio); |
3231 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3232 | result: |
3233 | #endif |
3234 | nwritten += submitted; |
3235 | wbc->nr_to_write -= submitted; |
3236 | |
3237 | if (unlikely(ret)) { |
3238 | /* |
3239 | * keep nr_to_write, since vfs uses this to |
3240 | * get # of written pages. |
3241 | */ |
3242 | if (ret == AOP_WRITEPAGE_ACTIVATE) { |
3243 | ret = 0; |
3244 | goto next; |
3245 | } else if (ret == -EAGAIN) { |
3246 | ret = 0; |
3247 | if (wbc->sync_mode == WB_SYNC_ALL) { |
3248 | f2fs_io_schedule_timeout( |
3249 | DEFAULT_IO_TIMEOUT); |
3250 | goto retry_write; |
3251 | } |
3252 | goto next; |
3253 | } |
3254 | done_index = folio_next_index(folio); |
3255 | done = 1; |
3256 | break; |
3257 | } |
3258 | |
3259 | if (wbc->nr_to_write <= 0 && |
3260 | wbc->sync_mode == WB_SYNC_NONE) { |
3261 | done = 1; |
3262 | break; |
3263 | } |
3264 | next: |
3265 | if (need_readd) |
3266 | goto readd; |
3267 | } |
3268 | release_pages(pages, nr: nr_pages); |
3269 | cond_resched(); |
3270 | } |
3271 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3272 | /* flush remained pages in compress cluster */ |
3273 | if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(cc: &cc)) { |
3274 | ret = f2fs_write_multi_pages(cc: &cc, submitted: &submitted, wbc, io_type); |
3275 | nwritten += submitted; |
3276 | wbc->nr_to_write -= submitted; |
3277 | if (ret) { |
3278 | done = 1; |
3279 | retry = 0; |
3280 | } |
3281 | } |
3282 | if (f2fs_compressed_file(inode)) |
3283 | f2fs_destroy_compress_ctx(cc: &cc, reuse: false); |
3284 | #endif |
3285 | if (retry) { |
3286 | index = 0; |
3287 | end = -1; |
3288 | goto retry; |
3289 | } |
3290 | if (wbc->range_cyclic && !done) |
3291 | done_index = 0; |
3292 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
3293 | mapping->writeback_index = done_index; |
3294 | |
3295 | if (nwritten) |
3296 | f2fs_submit_merged_write_cond(sbi: F2FS_M_SB(mapping), inode: mapping->host, |
3297 | NULL, ino: 0, type: DATA); |
3298 | /* submit cached bio of IPU write */ |
3299 | if (bio) |
3300 | f2fs_submit_merged_ipu_write(sbi, bio: &bio, NULL); |
3301 | |
3302 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3303 | if (pages != pages_local) |
3304 | kfree(objp: pages); |
3305 | #endif |
3306 | |
3307 | return ret; |
3308 | } |
3309 | |
3310 | static inline bool __should_serialize_io(struct inode *inode, |
3311 | struct writeback_control *wbc) |
3312 | { |
3313 | /* to avoid deadlock in path of data flush */ |
3314 | if (F2FS_I(inode)->wb_task) |
3315 | return false; |
3316 | |
3317 | if (!S_ISREG(inode->i_mode)) |
3318 | return false; |
3319 | if (IS_NOQUOTA(inode)) |
3320 | return false; |
3321 | |
3322 | if (f2fs_need_compress_data(inode)) |
3323 | return true; |
3324 | if (wbc->sync_mode != WB_SYNC_ALL) |
3325 | return true; |
3326 | if (get_dirty_pages(inode) >= SM_I(sbi: F2FS_I_SB(inode))->min_seq_blocks) |
3327 | return true; |
3328 | return false; |
3329 | } |
3330 | |
3331 | static int __f2fs_write_data_pages(struct address_space *mapping, |
3332 | struct writeback_control *wbc, |
3333 | enum iostat_type io_type) |
3334 | { |
3335 | struct inode *inode = mapping->host; |
3336 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
3337 | struct blk_plug plug; |
3338 | int ret; |
3339 | bool locked = false; |
3340 | |
3341 | /* deal with chardevs and other special file */ |
3342 | if (!mapping->a_ops->writepage) |
3343 | return 0; |
3344 | |
3345 | /* skip writing if there is no dirty page in this inode */ |
3346 | if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE) |
3347 | return 0; |
3348 | |
3349 | /* during POR, we don't need to trigger writepage at all. */ |
3350 | if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) |
3351 | goto skip_write; |
3352 | |
3353 | if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) && |
3354 | wbc->sync_mode == WB_SYNC_NONE && |
3355 | get_dirty_pages(inode) < nr_pages_to_skip(sbi, type: DATA) && |
3356 | f2fs_available_free_memory(sbi, type: DIRTY_DENTS)) |
3357 | goto skip_write; |
3358 | |
3359 | /* skip writing in file defragment preparing stage */ |
3360 | if (is_inode_flag_set(inode, flag: FI_SKIP_WRITES)) |
3361 | goto skip_write; |
3362 | |
3363 | trace_f2fs_writepages(inode: mapping->host, wbc, type: DATA); |
3364 | |
3365 | /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */ |
3366 | if (wbc->sync_mode == WB_SYNC_ALL) |
3367 | atomic_inc(v: &sbi->wb_sync_req[DATA]); |
3368 | else if (atomic_read(v: &sbi->wb_sync_req[DATA])) { |
3369 | /* to avoid potential deadlock */ |
3370 | if (current->plug) |
3371 | blk_finish_plug(current->plug); |
3372 | goto skip_write; |
3373 | } |
3374 | |
3375 | if (__should_serialize_io(inode, wbc)) { |
3376 | mutex_lock(&sbi->writepages); |
3377 | locked = true; |
3378 | } |
3379 | |
3380 | blk_start_plug(&plug); |
3381 | ret = f2fs_write_cache_pages(mapping, wbc, io_type); |
3382 | blk_finish_plug(&plug); |
3383 | |
3384 | if (locked) |
3385 | mutex_unlock(lock: &sbi->writepages); |
3386 | |
3387 | if (wbc->sync_mode == WB_SYNC_ALL) |
3388 | atomic_dec(v: &sbi->wb_sync_req[DATA]); |
3389 | /* |
3390 | * if some pages were truncated, we cannot guarantee its mapping->host |
3391 | * to detect pending bios. |
3392 | */ |
3393 | |
3394 | f2fs_remove_dirty_inode(inode); |
3395 | return ret; |
3396 | |
3397 | skip_write: |
3398 | wbc->pages_skipped += get_dirty_pages(inode); |
3399 | trace_f2fs_writepages(inode: mapping->host, wbc, type: DATA); |
3400 | return 0; |
3401 | } |
3402 | |
3403 | static int f2fs_write_data_pages(struct address_space *mapping, |
3404 | struct writeback_control *wbc) |
3405 | { |
3406 | struct inode *inode = mapping->host; |
3407 | |
3408 | return __f2fs_write_data_pages(mapping, wbc, |
3409 | io_type: F2FS_I(inode)->cp_task == current ? |
3410 | FS_CP_DATA_IO : FS_DATA_IO); |
3411 | } |
3412 | |
3413 | void f2fs_write_failed(struct inode *inode, loff_t to) |
3414 | { |
3415 | loff_t i_size = i_size_read(inode); |
3416 | |
3417 | if (IS_NOQUOTA(inode)) |
3418 | return; |
3419 | |
3420 | /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */ |
3421 | if (to > i_size && !f2fs_verity_in_progress(inode)) { |
3422 | f2fs_down_write(sem: &F2FS_I(inode)->i_gc_rwsem[WRITE]); |
3423 | filemap_invalidate_lock(mapping: inode->i_mapping); |
3424 | |
3425 | truncate_pagecache(inode, new: i_size); |
3426 | f2fs_truncate_blocks(inode, from: i_size, lock: true); |
3427 | |
3428 | filemap_invalidate_unlock(mapping: inode->i_mapping); |
3429 | f2fs_up_write(sem: &F2FS_I(inode)->i_gc_rwsem[WRITE]); |
3430 | } |
3431 | } |
3432 | |
3433 | static int prepare_write_begin(struct f2fs_sb_info *sbi, |
3434 | struct page *page, loff_t pos, unsigned len, |
3435 | block_t *blk_addr, bool *node_changed) |
3436 | { |
3437 | struct inode *inode = page->mapping->host; |
3438 | pgoff_t index = page->index; |
3439 | struct dnode_of_data dn; |
3440 | struct page *ipage; |
3441 | bool locked = false; |
3442 | int flag = F2FS_GET_BLOCK_PRE_AIO; |
3443 | int err = 0; |
3444 | |
3445 | /* |
3446 | * If a whole page is being written and we already preallocated all the |
3447 | * blocks, then there is no need to get a block address now. |
3448 | */ |
3449 | if (len == PAGE_SIZE && is_inode_flag_set(inode, flag: FI_PREALLOCATED_ALL)) |
3450 | return 0; |
3451 | |
3452 | /* f2fs_lock_op avoids race between write CP and convert_inline_page */ |
3453 | if (f2fs_has_inline_data(inode)) { |
3454 | if (pos + len > MAX_INLINE_DATA(inode)) |
3455 | flag = F2FS_GET_BLOCK_DEFAULT; |
3456 | f2fs_map_lock(sbi, flag); |
3457 | locked = true; |
3458 | } else if ((pos & PAGE_MASK) >= i_size_read(inode)) { |
3459 | f2fs_map_lock(sbi, flag); |
3460 | locked = true; |
3461 | } |
3462 | |
3463 | restart: |
3464 | /* check inline_data */ |
3465 | ipage = f2fs_get_node_page(sbi, nid: inode->i_ino); |
3466 | if (IS_ERR(ptr: ipage)) { |
3467 | err = PTR_ERR(ptr: ipage); |
3468 | goto unlock_out; |
3469 | } |
3470 | |
3471 | set_new_dnode(dn: &dn, inode, ipage, npage: ipage, nid: 0); |
3472 | |
3473 | if (f2fs_has_inline_data(inode)) { |
3474 | if (pos + len <= MAX_INLINE_DATA(inode)) { |
3475 | f2fs_do_read_inline_data(page, ipage); |
3476 | set_inode_flag(inode, flag: FI_DATA_EXIST); |
3477 | if (inode->i_nlink) |
3478 | set_page_private_inline(ipage); |
3479 | goto out; |
3480 | } |
3481 | err = f2fs_convert_inline_page(dn: &dn, page); |
3482 | if (err || dn.data_blkaddr != NULL_ADDR) |
3483 | goto out; |
3484 | } |
3485 | |
3486 | if (!f2fs_lookup_read_extent_cache_block(inode, index, |
3487 | blkaddr: &dn.data_blkaddr)) { |
3488 | if (locked) { |
3489 | err = f2fs_reserve_block(dn: &dn, index); |
3490 | goto out; |
3491 | } |
3492 | |
3493 | /* hole case */ |
3494 | err = f2fs_get_dnode_of_data(dn: &dn, index, mode: LOOKUP_NODE); |
3495 | if (!err && dn.data_blkaddr != NULL_ADDR) |
3496 | goto out; |
3497 | f2fs_put_dnode(dn: &dn); |
3498 | f2fs_map_lock(sbi, flag: F2FS_GET_BLOCK_PRE_AIO); |
3499 | WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO); |
3500 | locked = true; |
3501 | goto restart; |
3502 | } |
3503 | out: |
3504 | if (!err) { |
3505 | /* convert_inline_page can make node_changed */ |
3506 | *blk_addr = dn.data_blkaddr; |
3507 | *node_changed = dn.node_changed; |
3508 | } |
3509 | f2fs_put_dnode(dn: &dn); |
3510 | unlock_out: |
3511 | if (locked) |
3512 | f2fs_map_unlock(sbi, flag); |
3513 | return err; |
3514 | } |
3515 | |
3516 | static int __find_data_block(struct inode *inode, pgoff_t index, |
3517 | block_t *blk_addr) |
3518 | { |
3519 | struct dnode_of_data dn; |
3520 | struct page *ipage; |
3521 | int err = 0; |
3522 | |
3523 | ipage = f2fs_get_node_page(sbi: F2FS_I_SB(inode), nid: inode->i_ino); |
3524 | if (IS_ERR(ptr: ipage)) |
3525 | return PTR_ERR(ptr: ipage); |
3526 | |
3527 | set_new_dnode(dn: &dn, inode, ipage, npage: ipage, nid: 0); |
3528 | |
3529 | if (!f2fs_lookup_read_extent_cache_block(inode, index, |
3530 | blkaddr: &dn.data_blkaddr)) { |
3531 | /* hole case */ |
3532 | err = f2fs_get_dnode_of_data(dn: &dn, index, mode: LOOKUP_NODE); |
3533 | if (err) { |
3534 | dn.data_blkaddr = NULL_ADDR; |
3535 | err = 0; |
3536 | } |
3537 | } |
3538 | *blk_addr = dn.data_blkaddr; |
3539 | f2fs_put_dnode(dn: &dn); |
3540 | return err; |
3541 | } |
3542 | |
3543 | static int __reserve_data_block(struct inode *inode, pgoff_t index, |
3544 | block_t *blk_addr, bool *node_changed) |
3545 | { |
3546 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
3547 | struct dnode_of_data dn; |
3548 | struct page *ipage; |
3549 | int err = 0; |
3550 | |
3551 | f2fs_map_lock(sbi, flag: F2FS_GET_BLOCK_PRE_AIO); |
3552 | |
3553 | ipage = f2fs_get_node_page(sbi, nid: inode->i_ino); |
3554 | if (IS_ERR(ptr: ipage)) { |
3555 | err = PTR_ERR(ptr: ipage); |
3556 | goto unlock_out; |
3557 | } |
3558 | set_new_dnode(dn: &dn, inode, ipage, npage: ipage, nid: 0); |
3559 | |
3560 | if (!f2fs_lookup_read_extent_cache_block(inode: dn.inode, index, |
3561 | blkaddr: &dn.data_blkaddr)) |
3562 | err = f2fs_reserve_block(dn: &dn, index); |
3563 | |
3564 | *blk_addr = dn.data_blkaddr; |
3565 | *node_changed = dn.node_changed; |
3566 | f2fs_put_dnode(dn: &dn); |
3567 | |
3568 | unlock_out: |
3569 | f2fs_map_unlock(sbi, flag: F2FS_GET_BLOCK_PRE_AIO); |
3570 | return err; |
3571 | } |
3572 | |
3573 | static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi, |
3574 | struct page *page, loff_t pos, unsigned int len, |
3575 | block_t *blk_addr, bool *node_changed, bool *use_cow) |
3576 | { |
3577 | struct inode *inode = page->mapping->host; |
3578 | struct inode *cow_inode = F2FS_I(inode)->cow_inode; |
3579 | pgoff_t index = page->index; |
3580 | int err = 0; |
3581 | block_t ori_blk_addr = NULL_ADDR; |
3582 | |
3583 | /* If pos is beyond the end of file, reserve a new block in COW inode */ |
3584 | if ((pos & PAGE_MASK) >= i_size_read(inode)) |
3585 | goto reserve_block; |
3586 | |
3587 | /* Look for the block in COW inode first */ |
3588 | err = __find_data_block(inode: cow_inode, index, blk_addr); |
3589 | if (err) { |
3590 | return err; |
3591 | } else if (*blk_addr != NULL_ADDR) { |
3592 | *use_cow = true; |
3593 | return 0; |
3594 | } |
3595 | |
3596 | if (is_inode_flag_set(inode, flag: FI_ATOMIC_REPLACE)) |
3597 | goto reserve_block; |
3598 | |
3599 | /* Look for the block in the original inode */ |
3600 | err = __find_data_block(inode, index, blk_addr: &ori_blk_addr); |
3601 | if (err) |
3602 | return err; |
3603 | |
3604 | reserve_block: |
3605 | /* Finally, we should reserve a new block in COW inode for the update */ |
3606 | err = __reserve_data_block(inode: cow_inode, index, blk_addr, node_changed); |
3607 | if (err) |
3608 | return err; |
3609 | inc_atomic_write_cnt(inode); |
3610 | |
3611 | if (ori_blk_addr != NULL_ADDR) |
3612 | *blk_addr = ori_blk_addr; |
3613 | return 0; |
3614 | } |
3615 | |
3616 | static int f2fs_write_begin(struct file *file, struct address_space *mapping, |
3617 | loff_t pos, unsigned len, struct page **pagep, void **fsdata) |
3618 | { |
3619 | struct inode *inode = mapping->host; |
3620 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
3621 | struct page *page = NULL; |
3622 | pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT; |
3623 | bool need_balance = false; |
3624 | bool use_cow = false; |
3625 | block_t blkaddr = NULL_ADDR; |
3626 | int err = 0; |
3627 | |
3628 | trace_f2fs_write_begin(inode, pos, len); |
3629 | |
3630 | if (!f2fs_is_checkpoint_ready(sbi)) { |
3631 | err = -ENOSPC; |
3632 | goto fail; |
3633 | } |
3634 | |
3635 | /* |
3636 | * We should check this at this moment to avoid deadlock on inode page |
3637 | * and #0 page. The locking rule for inline_data conversion should be: |
3638 | * lock_page(page #0) -> lock_page(inode_page) |
3639 | */ |
3640 | if (index != 0) { |
3641 | err = f2fs_convert_inline_inode(inode); |
3642 | if (err) |
3643 | goto fail; |
3644 | } |
3645 | |
3646 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3647 | if (f2fs_compressed_file(inode)) { |
3648 | int ret; |
3649 | |
3650 | *fsdata = NULL; |
3651 | |
3652 | if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode))) |
3653 | goto repeat; |
3654 | |
3655 | ret = f2fs_prepare_compress_overwrite(inode, pagep, |
3656 | index, fsdata); |
3657 | if (ret < 0) { |
3658 | err = ret; |
3659 | goto fail; |
3660 | } else if (ret) { |
3661 | return 0; |
3662 | } |
3663 | } |
3664 | #endif |
3665 | |
3666 | repeat: |
3667 | /* |
3668 | * Do not use grab_cache_page_write_begin() to avoid deadlock due to |
3669 | * wait_for_stable_page. Will wait that below with our IO control. |
3670 | */ |
3671 | page = f2fs_pagecache_get_page(mapping, index, |
3672 | FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS); |
3673 | if (!page) { |
3674 | err = -ENOMEM; |
3675 | goto fail; |
3676 | } |
3677 | |
3678 | /* TODO: cluster can be compressed due to race with .writepage */ |
3679 | |
3680 | *pagep = page; |
3681 | |
3682 | if (f2fs_is_atomic_file(inode)) |
3683 | err = prepare_atomic_write_begin(sbi, page, pos, len, |
3684 | blk_addr: &blkaddr, node_changed: &need_balance, use_cow: &use_cow); |
3685 | else |
3686 | err = prepare_write_begin(sbi, page, pos, len, |
3687 | blk_addr: &blkaddr, node_changed: &need_balance); |
3688 | if (err) |
3689 | goto fail; |
3690 | |
3691 | if (need_balance && !IS_NOQUOTA(inode) && |
3692 | has_not_enough_free_secs(sbi, freed: 0, needed: 0)) { |
3693 | unlock_page(page); |
3694 | f2fs_balance_fs(sbi, need: true); |
3695 | lock_page(page); |
3696 | if (page->mapping != mapping) { |
3697 | /* The page got truncated from under us */ |
3698 | f2fs_put_page(page, unlock: 1); |
3699 | goto repeat; |
3700 | } |
3701 | } |
3702 | |
3703 | f2fs_wait_on_page_writeback(page, type: DATA, ordered: false, locked: true); |
3704 | |
3705 | if (len == PAGE_SIZE || PageUptodate(page)) |
3706 | return 0; |
3707 | |
3708 | if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) && |
3709 | !f2fs_verity_in_progress(inode)) { |
3710 | zero_user_segment(page, start: len, PAGE_SIZE); |
3711 | return 0; |
3712 | } |
3713 | |
3714 | if (blkaddr == NEW_ADDR) { |
3715 | zero_user_segment(page, start: 0, PAGE_SIZE); |
3716 | SetPageUptodate(page); |
3717 | } else { |
3718 | if (!f2fs_is_valid_blkaddr(sbi, blkaddr, |
3719 | type: DATA_GENERIC_ENHANCE_READ)) { |
3720 | err = -EFSCORRUPTED; |
3721 | f2fs_handle_error(sbi, error: ERROR_INVALID_BLKADDR); |
3722 | goto fail; |
3723 | } |
3724 | err = f2fs_submit_page_read(inode: use_cow ? |
3725 | F2FS_I(inode)->cow_inode : inode, page, |
3726 | blkaddr, op_flags: 0, for_write: true); |
3727 | if (err) |
3728 | goto fail; |
3729 | |
3730 | lock_page(page); |
3731 | if (unlikely(page->mapping != mapping)) { |
3732 | f2fs_put_page(page, unlock: 1); |
3733 | goto repeat; |
3734 | } |
3735 | if (unlikely(!PageUptodate(page))) { |
3736 | err = -EIO; |
3737 | goto fail; |
3738 | } |
3739 | } |
3740 | return 0; |
3741 | |
3742 | fail: |
3743 | f2fs_put_page(page, unlock: 1); |
3744 | f2fs_write_failed(inode, to: pos + len); |
3745 | return err; |
3746 | } |
3747 | |
3748 | static int f2fs_write_end(struct file *file, |
3749 | struct address_space *mapping, |
3750 | loff_t pos, unsigned len, unsigned copied, |
3751 | struct page *page, void *fsdata) |
3752 | { |
3753 | struct inode *inode = page->mapping->host; |
3754 | |
3755 | trace_f2fs_write_end(inode, pos, len, copied); |
3756 | |
3757 | /* |
3758 | * This should be come from len == PAGE_SIZE, and we expect copied |
3759 | * should be PAGE_SIZE. Otherwise, we treat it with zero copied and |
3760 | * let generic_perform_write() try to copy data again through copied=0. |
3761 | */ |
3762 | if (!PageUptodate(page)) { |
3763 | if (unlikely(copied != len)) |
3764 | copied = 0; |
3765 | else |
3766 | SetPageUptodate(page); |
3767 | } |
3768 | |
3769 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3770 | /* overwrite compressed file */ |
3771 | if (f2fs_compressed_file(inode) && fsdata) { |
3772 | f2fs_compress_write_end(inode, fsdata, index: page->index, copied); |
3773 | f2fs_update_time(sbi: F2FS_I_SB(inode), type: REQ_TIME); |
3774 | |
3775 | if (pos + copied > i_size_read(inode) && |
3776 | !f2fs_verity_in_progress(inode)) |
3777 | f2fs_i_size_write(inode, i_size: pos + copied); |
3778 | return copied; |
3779 | } |
3780 | #endif |
3781 | |
3782 | if (!copied) |
3783 | goto unlock_out; |
3784 | |
3785 | set_page_dirty(page); |
3786 | |
3787 | if (pos + copied > i_size_read(inode) && |
3788 | !f2fs_verity_in_progress(inode)) { |
3789 | f2fs_i_size_write(inode, i_size: pos + copied); |
3790 | if (f2fs_is_atomic_file(inode)) |
3791 | f2fs_i_size_write(inode: F2FS_I(inode)->cow_inode, |
3792 | i_size: pos + copied); |
3793 | } |
3794 | unlock_out: |
3795 | f2fs_put_page(page, unlock: 1); |
3796 | f2fs_update_time(sbi: F2FS_I_SB(inode), type: REQ_TIME); |
3797 | return copied; |
3798 | } |
3799 | |
3800 | void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length) |
3801 | { |
3802 | struct inode *inode = folio->mapping->host; |
3803 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
3804 | |
3805 | if (inode->i_ino >= F2FS_ROOT_INO(sbi) && |
3806 | (offset || length != folio_size(folio))) |
3807 | return; |
3808 | |
3809 | if (folio_test_dirty(folio)) { |
3810 | if (inode->i_ino == F2FS_META_INO(sbi)) { |
3811 | dec_page_count(sbi, count_type: F2FS_DIRTY_META); |
3812 | } else if (inode->i_ino == F2FS_NODE_INO(sbi)) { |
3813 | dec_page_count(sbi, count_type: F2FS_DIRTY_NODES); |
3814 | } else { |
3815 | inode_dec_dirty_pages(inode); |
3816 | f2fs_remove_dirty_inode(inode); |
3817 | } |
3818 | } |
3819 | clear_page_private_all(page: &folio->page); |
3820 | } |
3821 | |
3822 | bool f2fs_release_folio(struct folio *folio, gfp_t wait) |
3823 | { |
3824 | /* If this is dirty folio, keep private data */ |
3825 | if (folio_test_dirty(folio)) |
3826 | return false; |
3827 | |
3828 | clear_page_private_all(page: &folio->page); |
3829 | return true; |
3830 | } |
3831 | |
3832 | static bool f2fs_dirty_data_folio(struct address_space *mapping, |
3833 | struct folio *folio) |
3834 | { |
3835 | struct inode *inode = mapping->host; |
3836 | |
3837 | trace_f2fs_set_page_dirty(page: &folio->page, type: DATA); |
3838 | |
3839 | if (!folio_test_uptodate(folio)) |
3840 | folio_mark_uptodate(folio); |
3841 | BUG_ON(folio_test_swapcache(folio)); |
3842 | |
3843 | if (filemap_dirty_folio(mapping, folio)) { |
3844 | f2fs_update_dirty_folio(inode, folio); |
3845 | return true; |
3846 | } |
3847 | return false; |
3848 | } |
3849 | |
3850 | |
3851 | static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block) |
3852 | { |
3853 | #ifdef CONFIG_F2FS_FS_COMPRESSION |
3854 | struct dnode_of_data dn; |
3855 | sector_t start_idx, blknr = 0; |
3856 | int ret; |
3857 | |
3858 | start_idx = round_down(block, F2FS_I(inode)->i_cluster_size); |
3859 | |
3860 | set_new_dnode(dn: &dn, inode, NULL, NULL, nid: 0); |
3861 | ret = f2fs_get_dnode_of_data(dn: &dn, index: start_idx, mode: LOOKUP_NODE); |
3862 | if (ret) |
3863 | return 0; |
3864 | |
3865 | if (dn.data_blkaddr != COMPRESS_ADDR) { |
3866 | dn.ofs_in_node += block - start_idx; |
3867 | blknr = f2fs_data_blkaddr(dn: &dn); |
3868 | if (!__is_valid_data_blkaddr(blkaddr: blknr)) |
3869 | blknr = 0; |
3870 | } |
3871 | |
3872 | f2fs_put_dnode(dn: &dn); |
3873 | return blknr; |
3874 | #else |
3875 | return 0; |
3876 | #endif |
3877 | } |
3878 | |
3879 | |
3880 | static sector_t f2fs_bmap(struct address_space *mapping, sector_t block) |
3881 | { |
3882 | struct inode *inode = mapping->host; |
3883 | sector_t blknr = 0; |
3884 | |
3885 | if (f2fs_has_inline_data(inode)) |
3886 | goto out; |
3887 | |
3888 | /* make sure allocating whole blocks */ |
3889 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
3890 | filemap_write_and_wait(mapping); |
3891 | |
3892 | /* Block number less than F2FS MAX BLOCKS */ |
3893 | if (unlikely(block >= max_file_blocks(inode))) |
3894 | goto out; |
3895 | |
3896 | if (f2fs_compressed_file(inode)) { |
3897 | blknr = f2fs_bmap_compress(inode, block); |
3898 | } else { |
3899 | struct f2fs_map_blocks map; |
3900 | |
3901 | memset(&map, 0, sizeof(map)); |
3902 | map.m_lblk = block; |
3903 | map.m_len = 1; |
3904 | map.m_next_pgofs = NULL; |
3905 | map.m_seg_type = NO_CHECK_TYPE; |
3906 | |
3907 | if (!f2fs_map_blocks(inode, map: &map, flag: F2FS_GET_BLOCK_BMAP)) |
3908 | blknr = map.m_pblk; |
3909 | } |
3910 | out: |
3911 | trace_f2fs_bmap(inode, lblock: block, pblock: blknr); |
3912 | return blknr; |
3913 | } |
3914 | |
3915 | #ifdef CONFIG_SWAP |
3916 | static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk, |
3917 | unsigned int blkcnt) |
3918 | { |
3919 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
3920 | unsigned int blkofs; |
3921 | unsigned int blk_per_sec = BLKS_PER_SEC(sbi); |
3922 | unsigned int secidx = start_blk / blk_per_sec; |
3923 | unsigned int end_sec = secidx + blkcnt / blk_per_sec; |
3924 | int ret = 0; |
3925 | |
3926 | f2fs_down_write(sem: &F2FS_I(inode)->i_gc_rwsem[WRITE]); |
3927 | filemap_invalidate_lock(mapping: inode->i_mapping); |
3928 | |
3929 | set_inode_flag(inode, flag: FI_ALIGNED_WRITE); |
3930 | set_inode_flag(inode, flag: FI_OPU_WRITE); |
3931 | |
3932 | for (; secidx < end_sec; secidx++) { |
3933 | f2fs_down_write(sem: &sbi->pin_sem); |
3934 | |
3935 | f2fs_lock_op(sbi); |
3936 | f2fs_allocate_new_section(sbi, type: CURSEG_COLD_DATA_PINNED, force: false); |
3937 | f2fs_unlock_op(sbi); |
3938 | |
3939 | set_inode_flag(inode, flag: FI_SKIP_WRITES); |
3940 | |
3941 | for (blkofs = 0; blkofs < blk_per_sec; blkofs++) { |
3942 | struct page *page; |
3943 | unsigned int blkidx = secidx * blk_per_sec + blkofs; |
3944 | |
3945 | page = f2fs_get_lock_data_page(inode, index: blkidx, for_write: true); |
3946 | if (IS_ERR(ptr: page)) { |
3947 | f2fs_up_write(sem: &sbi->pin_sem); |
3948 | ret = PTR_ERR(ptr: page); |
3949 | goto done; |
3950 | } |
3951 | |
3952 | set_page_dirty(page); |
3953 | f2fs_put_page(page, unlock: 1); |
3954 | } |
3955 | |
3956 | clear_inode_flag(inode, flag: FI_SKIP_WRITES); |
3957 | |
3958 | ret = filemap_fdatawrite(inode->i_mapping); |
3959 | |
3960 | f2fs_up_write(sem: &sbi->pin_sem); |
3961 | |
3962 | if (ret) |
3963 | break; |
3964 | } |
3965 | |
3966 | done: |
3967 | clear_inode_flag(inode, flag: FI_SKIP_WRITES); |
3968 | clear_inode_flag(inode, flag: FI_OPU_WRITE); |
3969 | clear_inode_flag(inode, flag: FI_ALIGNED_WRITE); |
3970 | |
3971 | filemap_invalidate_unlock(mapping: inode->i_mapping); |
3972 | f2fs_up_write(sem: &F2FS_I(inode)->i_gc_rwsem[WRITE]); |
3973 | |
3974 | return ret; |
3975 | } |
3976 | |
3977 | static int check_swap_activate(struct swap_info_struct *sis, |
3978 | struct file *swap_file, sector_t *span) |
3979 | { |
3980 | struct address_space *mapping = swap_file->f_mapping; |
3981 | struct inode *inode = mapping->host; |
3982 | struct f2fs_sb_info *sbi = F2FS_I_SB(inode); |
3983 | sector_t cur_lblock; |
3984 | sector_t last_lblock; |
3985 | sector_t pblock; |
3986 | sector_t lowest_pblock = -1; |
3987 | sector_t highest_pblock = 0; |
3988 | int nr_extents = 0; |
3989 | unsigned long nr_pblocks; |
3990 | unsigned int blks_per_sec = BLKS_PER_SEC(sbi); |
3991 | unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1; |
3992 | unsigned int not_aligned = 0; |
3993 | int ret = 0; |
3994 | |
3995 | /* |
3996 | * Map all the blocks into the extent list. This code doesn't try |
3997 | * to be very smart. |
3998 | */ |
3999 | cur_lblock = 0; |
4000 | last_lblock = bytes_to_blks(inode, bytes: i_size_read(inode)); |
4001 | |
4002 | while (cur_lblock < last_lblock && cur_lblock < sis->max) { |
4003 | struct f2fs_map_blocks map; |
4004 | retry: |
4005 | cond_resched(); |
4006 | |
4007 | memset(&map, 0, sizeof(map)); |
4008 | map.m_lblk = cur_lblock; |
4009 | map.m_len = last_lblock - cur_lblock; |
4010 | map.m_next_pgofs = NULL; |
4011 | map.m_next_extent = NULL; |
4012 | map.m_seg_type = NO_CHECK_TYPE; |
4013 | map.m_may_create = false; |
4014 | |
4015 | ret = f2fs_map_blocks(inode, map: &map, flag: F2FS_GET_BLOCK_FIEMAP); |
4016 | if (ret) |
4017 | goto out; |
4018 | |
4019 | /* hole */ |
4020 | if (!(map.m_flags & F2FS_MAP_FLAGS)) { |
4021 | f2fs_err(sbi, "Swapfile has holes" ); |
4022 | ret = -EINVAL; |
4023 | goto out; |
4024 | } |
4025 | |
4026 | pblock = map.m_pblk; |
4027 | nr_pblocks = map.m_len; |
4028 | |
4029 | if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask || |
4030 | nr_pblocks & sec_blks_mask) { |
4031 | not_aligned++; |
4032 | |
4033 | nr_pblocks = roundup(nr_pblocks, blks_per_sec); |
4034 | if (cur_lblock + nr_pblocks > sis->max) |
4035 | nr_pblocks -= blks_per_sec; |
4036 | |
4037 | if (!nr_pblocks) { |
4038 | /* this extent is last one */ |
4039 | nr_pblocks = map.m_len; |
4040 | f2fs_warn(sbi, "Swapfile: last extent is not aligned to section" ); |
4041 | goto next; |
4042 | } |
4043 | |
4044 | ret = f2fs_migrate_blocks(inode, start_blk: cur_lblock, |
4045 | blkcnt: nr_pblocks); |
4046 | if (ret) |
4047 | goto out; |
4048 | goto retry; |
4049 | } |
4050 | next: |
4051 | if (cur_lblock + nr_pblocks >= sis->max) |
4052 | nr_pblocks = sis->max - cur_lblock; |
4053 | |
4054 | if (cur_lblock) { /* exclude the header page */ |
4055 | if (pblock < lowest_pblock) |
4056 | lowest_pblock = pblock; |
4057 | if (pblock + nr_pblocks - 1 > highest_pblock) |
4058 | highest_pblock = pblock + nr_pblocks - 1; |
4059 | } |
4060 | |
4061 | /* |
4062 | * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks |
4063 | */ |
4064 | ret = add_swap_extent(sis, start_page: cur_lblock, nr_pages: nr_pblocks, start_block: pblock); |
4065 | if (ret < 0) |
4066 | goto out; |
4067 | nr_extents += ret; |
4068 | cur_lblock += nr_pblocks; |
4069 | } |
4070 | ret = nr_extents; |
4071 | *span = 1 + highest_pblock - lowest_pblock; |
4072 | if (cur_lblock == 0) |
4073 | cur_lblock = 1; /* force Empty message */ |
4074 | sis->max = cur_lblock; |
4075 | sis->pages = cur_lblock - 1; |
4076 | sis->highest_bit = cur_lblock - 1; |
4077 | out: |
4078 | if (not_aligned) |
4079 | f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%lu * N)" , |
4080 | not_aligned, blks_per_sec * F2FS_BLKSIZE); |
4081 | return ret; |
4082 | } |
4083 | |
4084 | static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, |
4085 | sector_t *span) |
4086 | { |
4087 | struct inode *inode = file_inode(f: file); |
4088 | int ret; |
4089 | |
4090 | if (!S_ISREG(inode->i_mode)) |
4091 | return -EINVAL; |
4092 | |
4093 | if (f2fs_readonly(sb: F2FS_I_SB(inode)->sb)) |
4094 | return -EROFS; |
4095 | |
4096 | if (f2fs_lfs_mode(sbi: F2FS_I_SB(inode))) { |
4097 | f2fs_err(F2FS_I_SB(inode), |
4098 | "Swapfile not supported in LFS mode" ); |
4099 | return -EINVAL; |
4100 | } |
4101 | |
4102 | ret = f2fs_convert_inline_inode(inode); |
4103 | if (ret) |
4104 | return ret; |
4105 | |
4106 | if (!f2fs_disable_compressed_file(inode)) |
4107 | return -EINVAL; |
4108 | |
4109 | f2fs_precache_extents(inode); |
4110 | |
4111 | ret = check_swap_activate(sis, swap_file: file, span); |
4112 | if (ret < 0) |
4113 | return ret; |
4114 | |
4115 | stat_inc_swapfile_inode(inode); |
4116 | set_inode_flag(inode, flag: FI_PIN_FILE); |
4117 | f2fs_update_time(sbi: F2FS_I_SB(inode), type: REQ_TIME); |
4118 | return ret; |
4119 | } |
4120 | |
4121 | static void f2fs_swap_deactivate(struct file *file) |
4122 | { |
4123 | struct inode *inode = file_inode(f: file); |
4124 | |
4125 | stat_dec_swapfile_inode(inode); |
4126 | clear_inode_flag(inode, flag: FI_PIN_FILE); |
4127 | } |
4128 | #else |
4129 | static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file, |
4130 | sector_t *span) |
4131 | { |
4132 | return -EOPNOTSUPP; |
4133 | } |
4134 | |
4135 | static void f2fs_swap_deactivate(struct file *file) |
4136 | { |
4137 | } |
4138 | #endif |
4139 | |
4140 | const struct address_space_operations f2fs_dblock_aops = { |
4141 | .read_folio = f2fs_read_data_folio, |
4142 | .readahead = f2fs_readahead, |
4143 | .writepage = f2fs_write_data_page, |
4144 | .writepages = f2fs_write_data_pages, |
4145 | .write_begin = f2fs_write_begin, |
4146 | .write_end = f2fs_write_end, |
4147 | .dirty_folio = f2fs_dirty_data_folio, |
4148 | .migrate_folio = filemap_migrate_folio, |
4149 | .invalidate_folio = f2fs_invalidate_folio, |
4150 | .release_folio = f2fs_release_folio, |
4151 | .bmap = f2fs_bmap, |
4152 | .swap_activate = f2fs_swap_activate, |
4153 | .swap_deactivate = f2fs_swap_deactivate, |
4154 | }; |
4155 | |
4156 | void f2fs_clear_page_cache_dirty_tag(struct page *page) |
4157 | { |
4158 | struct address_space *mapping = page_mapping(page); |
4159 | unsigned long flags; |
4160 | |
4161 | xa_lock_irqsave(&mapping->i_pages, flags); |
4162 | __xa_clear_mark(&mapping->i_pages, index: page_index(page), |
4163 | PAGECACHE_TAG_DIRTY); |
4164 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
4165 | } |
4166 | |
4167 | int __init f2fs_init_post_read_processing(void) |
4168 | { |
4169 | bio_post_read_ctx_cache = |
4170 | kmem_cache_create(name: "f2fs_bio_post_read_ctx" , |
4171 | size: sizeof(struct bio_post_read_ctx), align: 0, flags: 0, NULL); |
4172 | if (!bio_post_read_ctx_cache) |
4173 | goto fail; |
4174 | bio_post_read_ctx_pool = |
4175 | mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, |
4176 | kc: bio_post_read_ctx_cache); |
4177 | if (!bio_post_read_ctx_pool) |
4178 | goto fail_free_cache; |
4179 | return 0; |
4180 | |
4181 | fail_free_cache: |
4182 | kmem_cache_destroy(s: bio_post_read_ctx_cache); |
4183 | fail: |
4184 | return -ENOMEM; |
4185 | } |
4186 | |
4187 | void f2fs_destroy_post_read_processing(void) |
4188 | { |
4189 | mempool_destroy(pool: bio_post_read_ctx_pool); |
4190 | kmem_cache_destroy(s: bio_post_read_ctx_cache); |
4191 | } |
4192 | |
4193 | int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi) |
4194 | { |
4195 | if (!f2fs_sb_has_encrypt(sbi) && |
4196 | !f2fs_sb_has_verity(sbi) && |
4197 | !f2fs_sb_has_compression(sbi)) |
4198 | return 0; |
4199 | |
4200 | sbi->post_read_wq = alloc_workqueue(fmt: "f2fs_post_read_wq" , |
4201 | flags: WQ_UNBOUND | WQ_HIGHPRI, |
4202 | max_active: num_online_cpus()); |
4203 | return sbi->post_read_wq ? 0 : -ENOMEM; |
4204 | } |
4205 | |
4206 | void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi) |
4207 | { |
4208 | if (sbi->post_read_wq) |
4209 | destroy_workqueue(wq: sbi->post_read_wq); |
4210 | } |
4211 | |
4212 | int __init f2fs_init_bio_entry_cache(void) |
4213 | { |
4214 | bio_entry_slab = f2fs_kmem_cache_create(name: "f2fs_bio_entry_slab" , |
4215 | size: sizeof(struct bio_entry)); |
4216 | return bio_entry_slab ? 0 : -ENOMEM; |
4217 | } |
4218 | |
4219 | void f2fs_destroy_bio_entry_cache(void) |
4220 | { |
4221 | kmem_cache_destroy(s: bio_entry_slab); |
4222 | } |
4223 | |
4224 | static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
4225 | unsigned int flags, struct iomap *iomap, |
4226 | struct iomap *srcmap) |
4227 | { |
4228 | struct f2fs_map_blocks map = {}; |
4229 | pgoff_t next_pgofs = 0; |
4230 | int err; |
4231 | |
4232 | map.m_lblk = bytes_to_blks(inode, bytes: offset); |
4233 | map.m_len = bytes_to_blks(inode, bytes: offset + length - 1) - map.m_lblk + 1; |
4234 | map.m_next_pgofs = &next_pgofs; |
4235 | map.m_seg_type = f2fs_rw_hint_to_seg_type(hint: inode->i_write_hint); |
4236 | if (flags & IOMAP_WRITE) |
4237 | map.m_may_create = true; |
4238 | |
4239 | err = f2fs_map_blocks(inode, map: &map, flag: F2FS_GET_BLOCK_DIO); |
4240 | if (err) |
4241 | return err; |
4242 | |
4243 | iomap->offset = blks_to_bytes(inode, blks: map.m_lblk); |
4244 | |
4245 | /* |
4246 | * When inline encryption is enabled, sometimes I/O to an encrypted file |
4247 | * has to be broken up to guarantee DUN contiguity. Handle this by |
4248 | * limiting the length of the mapping returned. |
4249 | */ |
4250 | map.m_len = fscrypt_limit_io_blocks(inode, lblk: map.m_lblk, nr_blocks: map.m_len); |
4251 | |
4252 | /* |
4253 | * We should never see delalloc or compressed extents here based on |
4254 | * prior flushing and checks. |
4255 | */ |
4256 | if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR)) |
4257 | return -EINVAL; |
4258 | if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR)) |
4259 | return -EINVAL; |
4260 | |
4261 | if (map.m_pblk != NULL_ADDR) { |
4262 | iomap->length = blks_to_bytes(inode, blks: map.m_len); |
4263 | iomap->type = IOMAP_MAPPED; |
4264 | iomap->flags |= IOMAP_F_MERGED; |
4265 | iomap->bdev = map.m_bdev; |
4266 | iomap->addr = blks_to_bytes(inode, blks: map.m_pblk); |
4267 | } else { |
4268 | if (flags & IOMAP_WRITE) |
4269 | return -ENOTBLK; |
4270 | iomap->length = blks_to_bytes(inode, blks: next_pgofs) - |
4271 | iomap->offset; |
4272 | iomap->type = IOMAP_HOLE; |
4273 | iomap->addr = IOMAP_NULL_ADDR; |
4274 | } |
4275 | |
4276 | if (map.m_flags & F2FS_MAP_NEW) |
4277 | iomap->flags |= IOMAP_F_NEW; |
4278 | if ((inode->i_state & I_DIRTY_DATASYNC) || |
4279 | offset + length > i_size_read(inode)) |
4280 | iomap->flags |= IOMAP_F_DIRTY; |
4281 | |
4282 | return 0; |
4283 | } |
4284 | |
4285 | const struct iomap_ops f2fs_iomap_ops = { |
4286 | .iomap_begin = f2fs_iomap_begin, |
4287 | }; |
4288 | |