Loading...
Note: File does not exist in v3.15.
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * bio-integrity.c - bio data integrity extensions
4 *
5 * Copyright (C) 2007, 2008, 2009 Oracle Corporation
6 * Written by: Martin K. Petersen <martin.petersen@oracle.com>
7 */
8
9#include <linux/blk-integrity.h>
10#include <linux/mempool.h>
11#include <linux/export.h>
12#include <linux/bio.h>
13#include <linux/workqueue.h>
14#include <linux/slab.h>
15#include "blk.h"
16
17static struct kmem_cache *bip_slab;
18static struct workqueue_struct *kintegrityd_wq;
19
20void blk_flush_integrity(void)
21{
22 flush_workqueue(kintegrityd_wq);
23}
24
25static void __bio_integrity_free(struct bio_set *bs,
26 struct bio_integrity_payload *bip)
27{
28 if (bs && mempool_initialized(&bs->bio_integrity_pool)) {
29 if (bip->bip_vec)
30 bvec_free(&bs->bvec_integrity_pool, bip->bip_vec,
31 bip->bip_max_vcnt);
32 mempool_free(bip, &bs->bio_integrity_pool);
33 } else {
34 kfree(bip);
35 }
36}
37
38/**
39 * bio_integrity_alloc - Allocate integrity payload and attach it to bio
40 * @bio: bio to attach integrity metadata to
41 * @gfp_mask: Memory allocation mask
42 * @nr_vecs: Number of integrity metadata scatter-gather elements
43 *
44 * Description: This function prepares a bio for attaching integrity
45 * metadata. nr_vecs specifies the maximum number of pages containing
46 * integrity metadata that can be attached.
47 */
48struct bio_integrity_payload *bio_integrity_alloc(struct bio *bio,
49 gfp_t gfp_mask,
50 unsigned int nr_vecs)
51{
52 struct bio_integrity_payload *bip;
53 struct bio_set *bs = bio->bi_pool;
54 unsigned inline_vecs;
55
56 if (WARN_ON_ONCE(bio_has_crypt_ctx(bio)))
57 return ERR_PTR(-EOPNOTSUPP);
58
59 if (!bs || !mempool_initialized(&bs->bio_integrity_pool)) {
60 bip = kmalloc(struct_size(bip, bip_inline_vecs, nr_vecs), gfp_mask);
61 inline_vecs = nr_vecs;
62 } else {
63 bip = mempool_alloc(&bs->bio_integrity_pool, gfp_mask);
64 inline_vecs = BIO_INLINE_VECS;
65 }
66
67 if (unlikely(!bip))
68 return ERR_PTR(-ENOMEM);
69
70 memset(bip, 0, sizeof(*bip));
71
72 /* always report as many vecs as asked explicitly, not inline vecs */
73 bip->bip_max_vcnt = nr_vecs;
74 if (nr_vecs > inline_vecs) {
75 bip->bip_vec = bvec_alloc(&bs->bvec_integrity_pool,
76 &bip->bip_max_vcnt, gfp_mask);
77 if (!bip->bip_vec)
78 goto err;
79 } else {
80 bip->bip_vec = bip->bip_inline_vecs;
81 }
82
83 bip->bip_bio = bio;
84 bio->bi_integrity = bip;
85 bio->bi_opf |= REQ_INTEGRITY;
86
87 return bip;
88err:
89 __bio_integrity_free(bs, bip);
90 return ERR_PTR(-ENOMEM);
91}
92EXPORT_SYMBOL(bio_integrity_alloc);
93
94static void bio_integrity_unpin_bvec(struct bio_vec *bv, int nr_vecs,
95 bool dirty)
96{
97 int i;
98
99 for (i = 0; i < nr_vecs; i++) {
100 if (dirty && !PageCompound(bv[i].bv_page))
101 set_page_dirty_lock(bv[i].bv_page);
102 unpin_user_page(bv[i].bv_page);
103 }
104}
105
106static void bio_integrity_uncopy_user(struct bio_integrity_payload *bip)
107{
108 unsigned short nr_vecs = bip->bip_max_vcnt - 1;
109 struct bio_vec *copy = &bip->bip_vec[1];
110 size_t bytes = bip->bip_iter.bi_size;
111 struct iov_iter iter;
112 int ret;
113
114 iov_iter_bvec(&iter, ITER_DEST, copy, nr_vecs, bytes);
115 ret = copy_to_iter(bvec_virt(bip->bip_vec), bytes, &iter);
116 WARN_ON_ONCE(ret != bytes);
117
118 bio_integrity_unpin_bvec(copy, nr_vecs, true);
119}
120
121static void bio_integrity_unmap_user(struct bio_integrity_payload *bip)
122{
123 bool dirty = bio_data_dir(bip->bip_bio) == READ;
124
125 if (bip->bip_flags & BIP_COPY_USER) {
126 if (dirty)
127 bio_integrity_uncopy_user(bip);
128 kfree(bvec_virt(bip->bip_vec));
129 return;
130 }
131
132 bio_integrity_unpin_bvec(bip->bip_vec, bip->bip_max_vcnt, dirty);
133}
134
135/**
136 * bio_integrity_free - Free bio integrity payload
137 * @bio: bio containing bip to be freed
138 *
139 * Description: Used to free the integrity portion of a bio. Usually
140 * called from bio_free().
141 */
142void bio_integrity_free(struct bio *bio)
143{
144 struct bio_integrity_payload *bip = bio_integrity(bio);
145 struct bio_set *bs = bio->bi_pool;
146
147 if (bip->bip_flags & BIP_BLOCK_INTEGRITY)
148 kfree(bvec_virt(bip->bip_vec));
149 else if (bip->bip_flags & BIP_INTEGRITY_USER)
150 bio_integrity_unmap_user(bip);
151
152 __bio_integrity_free(bs, bip);
153 bio->bi_integrity = NULL;
154 bio->bi_opf &= ~REQ_INTEGRITY;
155}
156
157/**
158 * bio_integrity_add_page - Attach integrity metadata
159 * @bio: bio to update
160 * @page: page containing integrity metadata
161 * @len: number of bytes of integrity metadata in page
162 * @offset: start offset within page
163 *
164 * Description: Attach a page containing integrity metadata to bio.
165 */
166int bio_integrity_add_page(struct bio *bio, struct page *page,
167 unsigned int len, unsigned int offset)
168{
169 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
170 struct bio_integrity_payload *bip = bio_integrity(bio);
171
172 if (((bip->bip_iter.bi_size + len) >> SECTOR_SHIFT) >
173 queue_max_hw_sectors(q))
174 return 0;
175
176 if (bip->bip_vcnt > 0) {
177 struct bio_vec *bv = &bip->bip_vec[bip->bip_vcnt - 1];
178 bool same_page = false;
179
180 if (bvec_try_merge_hw_page(q, bv, page, len, offset,
181 &same_page)) {
182 bip->bip_iter.bi_size += len;
183 return len;
184 }
185
186 if (bip->bip_vcnt >=
187 min(bip->bip_max_vcnt, queue_max_integrity_segments(q)))
188 return 0;
189
190 /*
191 * If the queue doesn't support SG gaps and adding this segment
192 * would create a gap, disallow it.
193 */
194 if (bvec_gap_to_prev(&q->limits, bv, offset))
195 return 0;
196 }
197
198 bvec_set_page(&bip->bip_vec[bip->bip_vcnt], page, len, offset);
199 bip->bip_vcnt++;
200 bip->bip_iter.bi_size += len;
201
202 return len;
203}
204EXPORT_SYMBOL(bio_integrity_add_page);
205
206static int bio_integrity_copy_user(struct bio *bio, struct bio_vec *bvec,
207 int nr_vecs, unsigned int len,
208 unsigned int direction, u32 seed)
209{
210 bool write = direction == ITER_SOURCE;
211 struct bio_integrity_payload *bip;
212 struct iov_iter iter;
213 void *buf;
214 int ret;
215
216 buf = kmalloc(len, GFP_KERNEL);
217 if (!buf)
218 return -ENOMEM;
219
220 if (write) {
221 iov_iter_bvec(&iter, direction, bvec, nr_vecs, len);
222 if (!copy_from_iter_full(buf, len, &iter)) {
223 ret = -EFAULT;
224 goto free_buf;
225 }
226
227 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
228 } else {
229 memset(buf, 0, len);
230
231 /*
232 * We need to preserve the original bvec and the number of vecs
233 * in it for completion handling
234 */
235 bip = bio_integrity_alloc(bio, GFP_KERNEL, nr_vecs + 1);
236 }
237
238 if (IS_ERR(bip)) {
239 ret = PTR_ERR(bip);
240 goto free_buf;
241 }
242
243 if (write)
244 bio_integrity_unpin_bvec(bvec, nr_vecs, false);
245 else
246 memcpy(&bip->bip_vec[1], bvec, nr_vecs * sizeof(*bvec));
247
248 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
249 offset_in_page(buf));
250 if (ret != len) {
251 ret = -ENOMEM;
252 goto free_bip;
253 }
254
255 bip->bip_flags |= BIP_INTEGRITY_USER | BIP_COPY_USER;
256 bip->bip_iter.bi_sector = seed;
257 return 0;
258free_bip:
259 bio_integrity_free(bio);
260free_buf:
261 kfree(buf);
262 return ret;
263}
264
265static int bio_integrity_init_user(struct bio *bio, struct bio_vec *bvec,
266 int nr_vecs, unsigned int len, u32 seed)
267{
268 struct bio_integrity_payload *bip;
269
270 bip = bio_integrity_alloc(bio, GFP_KERNEL, nr_vecs);
271 if (IS_ERR(bip))
272 return PTR_ERR(bip);
273
274 memcpy(bip->bip_vec, bvec, nr_vecs * sizeof(*bvec));
275 bip->bip_flags |= BIP_INTEGRITY_USER;
276 bip->bip_iter.bi_sector = seed;
277 bip->bip_iter.bi_size = len;
278 return 0;
279}
280
281static unsigned int bvec_from_pages(struct bio_vec *bvec, struct page **pages,
282 int nr_vecs, ssize_t bytes, ssize_t offset)
283{
284 unsigned int nr_bvecs = 0;
285 int i, j;
286
287 for (i = 0; i < nr_vecs; i = j) {
288 size_t size = min_t(size_t, bytes, PAGE_SIZE - offset);
289 struct folio *folio = page_folio(pages[i]);
290
291 bytes -= size;
292 for (j = i + 1; j < nr_vecs; j++) {
293 size_t next = min_t(size_t, PAGE_SIZE, bytes);
294
295 if (page_folio(pages[j]) != folio ||
296 pages[j] != pages[j - 1] + 1)
297 break;
298 unpin_user_page(pages[j]);
299 size += next;
300 bytes -= next;
301 }
302
303 bvec_set_page(&bvec[nr_bvecs], pages[i], size, offset);
304 offset = 0;
305 nr_bvecs++;
306 }
307
308 return nr_bvecs;
309}
310
311int bio_integrity_map_user(struct bio *bio, void __user *ubuf, ssize_t bytes,
312 u32 seed)
313{
314 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
315 unsigned int align = q->dma_pad_mask | queue_dma_alignment(q);
316 struct page *stack_pages[UIO_FASTIOV], **pages = stack_pages;
317 struct bio_vec stack_vec[UIO_FASTIOV], *bvec = stack_vec;
318 unsigned int direction, nr_bvecs;
319 struct iov_iter iter;
320 int ret, nr_vecs;
321 size_t offset;
322 bool copy;
323
324 if (bio_integrity(bio))
325 return -EINVAL;
326 if (bytes >> SECTOR_SHIFT > queue_max_hw_sectors(q))
327 return -E2BIG;
328
329 if (bio_data_dir(bio) == READ)
330 direction = ITER_DEST;
331 else
332 direction = ITER_SOURCE;
333
334 iov_iter_ubuf(&iter, direction, ubuf, bytes);
335 nr_vecs = iov_iter_npages(&iter, BIO_MAX_VECS + 1);
336 if (nr_vecs > BIO_MAX_VECS)
337 return -E2BIG;
338 if (nr_vecs > UIO_FASTIOV) {
339 bvec = kcalloc(nr_vecs, sizeof(*bvec), GFP_KERNEL);
340 if (!bvec)
341 return -ENOMEM;
342 pages = NULL;
343 }
344
345 copy = !iov_iter_is_aligned(&iter, align, align);
346 ret = iov_iter_extract_pages(&iter, &pages, bytes, nr_vecs, 0, &offset);
347 if (unlikely(ret < 0))
348 goto free_bvec;
349
350 nr_bvecs = bvec_from_pages(bvec, pages, nr_vecs, bytes, offset);
351 if (pages != stack_pages)
352 kvfree(pages);
353 if (nr_bvecs > queue_max_integrity_segments(q))
354 copy = true;
355
356 if (copy)
357 ret = bio_integrity_copy_user(bio, bvec, nr_bvecs, bytes,
358 direction, seed);
359 else
360 ret = bio_integrity_init_user(bio, bvec, nr_bvecs, bytes, seed);
361 if (ret)
362 goto release_pages;
363 if (bvec != stack_vec)
364 kfree(bvec);
365
366 return 0;
367
368release_pages:
369 bio_integrity_unpin_bvec(bvec, nr_bvecs, false);
370free_bvec:
371 if (bvec != stack_vec)
372 kfree(bvec);
373 return ret;
374}
375EXPORT_SYMBOL_GPL(bio_integrity_map_user);
376
377/**
378 * bio_integrity_process - Process integrity metadata for a bio
379 * @bio: bio to generate/verify integrity metadata for
380 * @proc_iter: iterator to process
381 * @proc_fn: Pointer to the relevant processing function
382 */
383static blk_status_t bio_integrity_process(struct bio *bio,
384 struct bvec_iter *proc_iter, integrity_processing_fn *proc_fn)
385{
386 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
387 struct blk_integrity_iter iter;
388 struct bvec_iter bviter;
389 struct bio_vec bv;
390 struct bio_integrity_payload *bip = bio_integrity(bio);
391 blk_status_t ret = BLK_STS_OK;
392
393 iter.disk_name = bio->bi_bdev->bd_disk->disk_name;
394 iter.interval = 1 << bi->interval_exp;
395 iter.tuple_size = bi->tuple_size;
396 iter.seed = proc_iter->bi_sector;
397 iter.prot_buf = bvec_virt(bip->bip_vec);
398
399 __bio_for_each_segment(bv, bio, bviter, *proc_iter) {
400 void *kaddr = bvec_kmap_local(&bv);
401
402 iter.data_buf = kaddr;
403 iter.data_size = bv.bv_len;
404 ret = proc_fn(&iter);
405 kunmap_local(kaddr);
406
407 if (ret)
408 break;
409
410 }
411 return ret;
412}
413
414/**
415 * bio_integrity_prep - Prepare bio for integrity I/O
416 * @bio: bio to prepare
417 *
418 * Description: Checks if the bio already has an integrity payload attached.
419 * If it does, the payload has been generated by another kernel subsystem,
420 * and we just pass it through. Otherwise allocates integrity payload.
421 * The bio must have data direction, target device and start sector set priot
422 * to calling. In the WRITE case, integrity metadata will be generated using
423 * the block device's integrity function. In the READ case, the buffer
424 * will be prepared for DMA and a suitable end_io handler set up.
425 */
426bool bio_integrity_prep(struct bio *bio)
427{
428 struct bio_integrity_payload *bip;
429 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
430 void *buf;
431 unsigned long start, end;
432 unsigned int len, nr_pages;
433 unsigned int bytes, offset, i;
434
435 if (!bi)
436 return true;
437
438 if (bio_op(bio) != REQ_OP_READ && bio_op(bio) != REQ_OP_WRITE)
439 return true;
440
441 if (!bio_sectors(bio))
442 return true;
443
444 /* Already protected? */
445 if (bio_integrity(bio))
446 return true;
447
448 if (bio_data_dir(bio) == READ) {
449 if (!bi->profile->verify_fn ||
450 !(bi->flags & BLK_INTEGRITY_VERIFY))
451 return true;
452 } else {
453 if (!bi->profile->generate_fn ||
454 !(bi->flags & BLK_INTEGRITY_GENERATE))
455 return true;
456 }
457
458 /* Allocate kernel buffer for protection data */
459 len = bio_integrity_bytes(bi, bio_sectors(bio));
460 buf = kmalloc(len, GFP_NOIO);
461 if (unlikely(buf == NULL)) {
462 printk(KERN_ERR "could not allocate integrity buffer\n");
463 goto err_end_io;
464 }
465
466 end = (((unsigned long) buf) + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
467 start = ((unsigned long) buf) >> PAGE_SHIFT;
468 nr_pages = end - start;
469
470 /* Allocate bio integrity payload and integrity vectors */
471 bip = bio_integrity_alloc(bio, GFP_NOIO, nr_pages);
472 if (IS_ERR(bip)) {
473 printk(KERN_ERR "could not allocate data integrity bioset\n");
474 kfree(buf);
475 goto err_end_io;
476 }
477
478 bip->bip_flags |= BIP_BLOCK_INTEGRITY;
479 bip_set_seed(bip, bio->bi_iter.bi_sector);
480
481 if (bi->flags & BLK_INTEGRITY_IP_CHECKSUM)
482 bip->bip_flags |= BIP_IP_CHECKSUM;
483
484 /* Map it */
485 offset = offset_in_page(buf);
486 for (i = 0; i < nr_pages && len > 0; i++) {
487 bytes = PAGE_SIZE - offset;
488
489 if (bytes > len)
490 bytes = len;
491
492 if (bio_integrity_add_page(bio, virt_to_page(buf),
493 bytes, offset) < bytes) {
494 printk(KERN_ERR "could not attach integrity payload\n");
495 goto err_end_io;
496 }
497
498 buf += bytes;
499 len -= bytes;
500 offset = 0;
501 }
502
503 /* Auto-generate integrity metadata if this is a write */
504 if (bio_data_dir(bio) == WRITE) {
505 bio_integrity_process(bio, &bio->bi_iter,
506 bi->profile->generate_fn);
507 } else {
508 bip->bio_iter = bio->bi_iter;
509 }
510 return true;
511
512err_end_io:
513 bio->bi_status = BLK_STS_RESOURCE;
514 bio_endio(bio);
515 return false;
516}
517EXPORT_SYMBOL(bio_integrity_prep);
518
519/**
520 * bio_integrity_verify_fn - Integrity I/O completion worker
521 * @work: Work struct stored in bio to be verified
522 *
523 * Description: This workqueue function is called to complete a READ
524 * request. The function verifies the transferred integrity metadata
525 * and then calls the original bio end_io function.
526 */
527static void bio_integrity_verify_fn(struct work_struct *work)
528{
529 struct bio_integrity_payload *bip =
530 container_of(work, struct bio_integrity_payload, bip_work);
531 struct bio *bio = bip->bip_bio;
532 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
533
534 /*
535 * At the moment verify is called bio's iterator was advanced
536 * during split and completion, we need to rewind iterator to
537 * it's original position.
538 */
539 bio->bi_status = bio_integrity_process(bio, &bip->bio_iter,
540 bi->profile->verify_fn);
541 bio_integrity_free(bio);
542 bio_endio(bio);
543}
544
545/**
546 * __bio_integrity_endio - Integrity I/O completion function
547 * @bio: Protected bio
548 *
549 * Description: Completion for integrity I/O
550 *
551 * Normally I/O completion is done in interrupt context. However,
552 * verifying I/O integrity is a time-consuming task which must be run
553 * in process context. This function postpones completion
554 * accordingly.
555 */
556bool __bio_integrity_endio(struct bio *bio)
557{
558 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
559 struct bio_integrity_payload *bip = bio_integrity(bio);
560
561 if (bio_op(bio) == REQ_OP_READ && !bio->bi_status &&
562 (bip->bip_flags & BIP_BLOCK_INTEGRITY) && bi->profile->verify_fn) {
563 INIT_WORK(&bip->bip_work, bio_integrity_verify_fn);
564 queue_work(kintegrityd_wq, &bip->bip_work);
565 return false;
566 }
567
568 bio_integrity_free(bio);
569 return true;
570}
571
572/**
573 * bio_integrity_advance - Advance integrity vector
574 * @bio: bio whose integrity vector to update
575 * @bytes_done: number of data bytes that have been completed
576 *
577 * Description: This function calculates how many integrity bytes the
578 * number of completed data bytes correspond to and advances the
579 * integrity vector accordingly.
580 */
581void bio_integrity_advance(struct bio *bio, unsigned int bytes_done)
582{
583 struct bio_integrity_payload *bip = bio_integrity(bio);
584 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
585 unsigned bytes = bio_integrity_bytes(bi, bytes_done >> 9);
586
587 bip->bip_iter.bi_sector += bio_integrity_intervals(bi, bytes_done >> 9);
588 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, bytes);
589}
590
591/**
592 * bio_integrity_trim - Trim integrity vector
593 * @bio: bio whose integrity vector to update
594 *
595 * Description: Used to trim the integrity vector in a cloned bio.
596 */
597void bio_integrity_trim(struct bio *bio)
598{
599 struct bio_integrity_payload *bip = bio_integrity(bio);
600 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
601
602 bip->bip_iter.bi_size = bio_integrity_bytes(bi, bio_sectors(bio));
603}
604EXPORT_SYMBOL(bio_integrity_trim);
605
606/**
607 * bio_integrity_clone - Callback for cloning bios with integrity metadata
608 * @bio: New bio
609 * @bio_src: Original bio
610 * @gfp_mask: Memory allocation mask
611 *
612 * Description: Called to allocate a bip when cloning a bio
613 */
614int bio_integrity_clone(struct bio *bio, struct bio *bio_src,
615 gfp_t gfp_mask)
616{
617 struct bio_integrity_payload *bip_src = bio_integrity(bio_src);
618 struct bio_integrity_payload *bip;
619
620 BUG_ON(bip_src == NULL);
621
622 bip = bio_integrity_alloc(bio, gfp_mask, bip_src->bip_vcnt);
623 if (IS_ERR(bip))
624 return PTR_ERR(bip);
625
626 memcpy(bip->bip_vec, bip_src->bip_vec,
627 bip_src->bip_vcnt * sizeof(struct bio_vec));
628
629 bip->bip_vcnt = bip_src->bip_vcnt;
630 bip->bip_iter = bip_src->bip_iter;
631 bip->bip_flags = bip_src->bip_flags & ~BIP_BLOCK_INTEGRITY;
632
633 return 0;
634}
635
636int bioset_integrity_create(struct bio_set *bs, int pool_size)
637{
638 if (mempool_initialized(&bs->bio_integrity_pool))
639 return 0;
640
641 if (mempool_init_slab_pool(&bs->bio_integrity_pool,
642 pool_size, bip_slab))
643 return -1;
644
645 if (biovec_init_pool(&bs->bvec_integrity_pool, pool_size)) {
646 mempool_exit(&bs->bio_integrity_pool);
647 return -1;
648 }
649
650 return 0;
651}
652EXPORT_SYMBOL(bioset_integrity_create);
653
654void bioset_integrity_free(struct bio_set *bs)
655{
656 mempool_exit(&bs->bio_integrity_pool);
657 mempool_exit(&bs->bvec_integrity_pool);
658}
659
660void __init bio_integrity_init(void)
661{
662 /*
663 * kintegrityd won't block much but may burn a lot of CPU cycles.
664 * Make it highpri CPU intensive wq with max concurrency of 1.
665 */
666 kintegrityd_wq = alloc_workqueue("kintegrityd", WQ_MEM_RECLAIM |
667 WQ_HIGHPRI | WQ_CPU_INTENSIVE, 1);
668 if (!kintegrityd_wq)
669 panic("Failed to create kintegrityd\n");
670
671 bip_slab = kmem_cache_create("bio_integrity_payload",
672 sizeof(struct bio_integrity_payload) +
673 sizeof(struct bio_vec) * BIO_INLINE_VECS,
674 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
675}