xref: /linux/block/bio-integrity.c (revision 34e3c4500cdc06094b37a41b622598098308ba8f)
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 
17 static struct kmem_cache *bip_slab;
18 static struct workqueue_struct *kintegrityd_wq;
19 
20 void blk_flush_integrity(void)
21 {
22 	flush_workqueue(kintegrityd_wq);
23 }
24 
25 /**
26  * bio_integrity_free - Free bio integrity payload
27  * @bio:	bio containing bip to be freed
28  *
29  * Description: Free the integrity portion of a bio.
30  */
31 void bio_integrity_free(struct bio *bio)
32 {
33 	struct bio_integrity_payload *bip = bio_integrity(bio);
34 	struct bio_set *bs = bio->bi_pool;
35 
36 	if (bs && mempool_initialized(&bs->bio_integrity_pool)) {
37 		if (bip->bip_vec)
38 			bvec_free(&bs->bvec_integrity_pool, bip->bip_vec,
39 				  bip->bip_max_vcnt);
40 		mempool_free(bip, &bs->bio_integrity_pool);
41 	} else {
42 		kfree(bip);
43 	}
44 	bio->bi_integrity = NULL;
45 	bio->bi_opf &= ~REQ_INTEGRITY;
46 }
47 
48 /**
49  * bio_integrity_alloc - Allocate integrity payload and attach it to bio
50  * @bio:	bio to attach integrity metadata to
51  * @gfp_mask:	Memory allocation mask
52  * @nr_vecs:	Number of integrity metadata scatter-gather elements
53  *
54  * Description: This function prepares a bio for attaching integrity
55  * metadata.  nr_vecs specifies the maximum number of pages containing
56  * integrity metadata that can be attached.
57  */
58 struct bio_integrity_payload *bio_integrity_alloc(struct bio *bio,
59 						  gfp_t gfp_mask,
60 						  unsigned int nr_vecs)
61 {
62 	struct bio_integrity_payload *bip;
63 	struct bio_set *bs = bio->bi_pool;
64 	unsigned inline_vecs;
65 
66 	if (WARN_ON_ONCE(bio_has_crypt_ctx(bio)))
67 		return ERR_PTR(-EOPNOTSUPP);
68 
69 	if (!bs || !mempool_initialized(&bs->bio_integrity_pool)) {
70 		bip = kmalloc(struct_size(bip, bip_inline_vecs, nr_vecs), gfp_mask);
71 		inline_vecs = nr_vecs;
72 	} else {
73 		bip = mempool_alloc(&bs->bio_integrity_pool, gfp_mask);
74 		inline_vecs = BIO_INLINE_VECS;
75 	}
76 
77 	if (unlikely(!bip))
78 		return ERR_PTR(-ENOMEM);
79 
80 	memset(bip, 0, sizeof(*bip));
81 
82 	/* always report as many vecs as asked explicitly, not inline vecs */
83 	bip->bip_max_vcnt = nr_vecs;
84 	if (nr_vecs > inline_vecs) {
85 		bip->bip_vec = bvec_alloc(&bs->bvec_integrity_pool,
86 					  &bip->bip_max_vcnt, gfp_mask);
87 		if (!bip->bip_vec)
88 			goto err;
89 	} else if (nr_vecs) {
90 		bip->bip_vec = bip->bip_inline_vecs;
91 	}
92 
93 	bip->bip_bio = bio;
94 	bio->bi_integrity = bip;
95 	bio->bi_opf |= REQ_INTEGRITY;
96 
97 	return bip;
98 err:
99 	if (bs && mempool_initialized(&bs->bio_integrity_pool))
100 		mempool_free(bip, &bs->bio_integrity_pool);
101 	else
102 		kfree(bip);
103 	return ERR_PTR(-ENOMEM);
104 }
105 EXPORT_SYMBOL(bio_integrity_alloc);
106 
107 static void bio_integrity_unpin_bvec(struct bio_vec *bv, int nr_vecs,
108 				     bool dirty)
109 {
110 	int i;
111 
112 	for (i = 0; i < nr_vecs; i++) {
113 		if (dirty && !PageCompound(bv[i].bv_page))
114 			set_page_dirty_lock(bv[i].bv_page);
115 		unpin_user_page(bv[i].bv_page);
116 	}
117 }
118 
119 static void bio_integrity_uncopy_user(struct bio_integrity_payload *bip)
120 {
121 	unsigned short nr_vecs = bip->bip_max_vcnt - 1;
122 	struct bio_vec *copy = &bip->bip_vec[1];
123 	size_t bytes = bip->bip_iter.bi_size;
124 	struct iov_iter iter;
125 	int ret;
126 
127 	iov_iter_bvec(&iter, ITER_DEST, copy, nr_vecs, bytes);
128 	ret = copy_to_iter(bvec_virt(bip->bip_vec), bytes, &iter);
129 	WARN_ON_ONCE(ret != bytes);
130 
131 	bio_integrity_unpin_bvec(copy, nr_vecs, true);
132 }
133 
134 /**
135  * bio_integrity_unmap_user - Unmap user integrity payload
136  * @bio:	bio containing bip to be unmapped
137  *
138  * Unmap the user mapped integrity portion of a bio.
139  */
140 void bio_integrity_unmap_user(struct bio *bio)
141 {
142 	struct bio_integrity_payload *bip = bio_integrity(bio);
143 
144 	if (bip->bip_flags & BIP_COPY_USER) {
145 		if (bio_data_dir(bio) == READ)
146 			bio_integrity_uncopy_user(bip);
147 		kfree(bvec_virt(bip->bip_vec));
148 		return;
149 	}
150 
151 	bio_integrity_unpin_bvec(bip->bip_vec, bip->bip_max_vcnt,
152 			bio_data_dir(bio) == READ);
153 }
154 
155 /**
156  * bio_integrity_add_page - Attach integrity metadata
157  * @bio:	bio to update
158  * @page:	page containing integrity metadata
159  * @len:	number of bytes of integrity metadata in page
160  * @offset:	start offset within page
161  *
162  * Description: Attach a page containing integrity metadata to bio.
163  */
164 int bio_integrity_add_page(struct bio *bio, struct page *page,
165 			   unsigned int len, unsigned int offset)
166 {
167 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
168 	struct bio_integrity_payload *bip = bio_integrity(bio);
169 
170 	if (bip->bip_vcnt > 0) {
171 		struct bio_vec *bv = &bip->bip_vec[bip->bip_vcnt - 1];
172 		bool same_page = false;
173 
174 		if (bvec_try_merge_hw_page(q, bv, page, len, offset,
175 					   &same_page)) {
176 			bip->bip_iter.bi_size += len;
177 			return len;
178 		}
179 
180 		if (bip->bip_vcnt >=
181 		    min(bip->bip_max_vcnt, queue_max_integrity_segments(q)))
182 			return 0;
183 
184 		/*
185 		 * If the queue doesn't support SG gaps and adding this segment
186 		 * would create a gap, disallow it.
187 		 */
188 		if (bvec_gap_to_prev(&q->limits, bv, offset))
189 			return 0;
190 	}
191 
192 	bvec_set_page(&bip->bip_vec[bip->bip_vcnt], page, len, offset);
193 	bip->bip_vcnt++;
194 	bip->bip_iter.bi_size += len;
195 
196 	return len;
197 }
198 EXPORT_SYMBOL(bio_integrity_add_page);
199 
200 static int bio_integrity_copy_user(struct bio *bio, struct bio_vec *bvec,
201 				   int nr_vecs, unsigned int len,
202 				   unsigned int direction, u32 seed)
203 {
204 	bool write = direction == ITER_SOURCE;
205 	struct bio_integrity_payload *bip;
206 	struct iov_iter iter;
207 	void *buf;
208 	int ret;
209 
210 	buf = kmalloc(len, GFP_KERNEL);
211 	if (!buf)
212 		return -ENOMEM;
213 
214 	if (write) {
215 		iov_iter_bvec(&iter, direction, bvec, nr_vecs, len);
216 		if (!copy_from_iter_full(buf, len, &iter)) {
217 			ret = -EFAULT;
218 			goto free_buf;
219 		}
220 
221 		bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
222 	} else {
223 		memset(buf, 0, len);
224 
225 		/*
226 		 * We need to preserve the original bvec and the number of vecs
227 		 * in it for completion handling
228 		 */
229 		bip = bio_integrity_alloc(bio, GFP_KERNEL, nr_vecs + 1);
230 	}
231 
232 	if (IS_ERR(bip)) {
233 		ret = PTR_ERR(bip);
234 		goto free_buf;
235 	}
236 
237 	if (write)
238 		bio_integrity_unpin_bvec(bvec, nr_vecs, false);
239 	else
240 		memcpy(&bip->bip_vec[1], bvec, nr_vecs * sizeof(*bvec));
241 
242 	ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
243 				     offset_in_page(buf));
244 	if (ret != len) {
245 		ret = -ENOMEM;
246 		goto free_bip;
247 	}
248 
249 	bip->bip_flags |= BIP_COPY_USER;
250 	bip->bip_iter.bi_sector = seed;
251 	bip->bip_vcnt = nr_vecs;
252 	return 0;
253 free_bip:
254 	bio_integrity_free(bio);
255 free_buf:
256 	kfree(buf);
257 	return ret;
258 }
259 
260 static int bio_integrity_init_user(struct bio *bio, struct bio_vec *bvec,
261 				   int nr_vecs, unsigned int len, u32 seed)
262 {
263 	struct bio_integrity_payload *bip;
264 
265 	bip = bio_integrity_alloc(bio, GFP_KERNEL, nr_vecs);
266 	if (IS_ERR(bip))
267 		return PTR_ERR(bip);
268 
269 	memcpy(bip->bip_vec, bvec, nr_vecs * sizeof(*bvec));
270 	bip->bip_iter.bi_sector = seed;
271 	bip->bip_iter.bi_size = len;
272 	bip->bip_vcnt = nr_vecs;
273 	return 0;
274 }
275 
276 static unsigned int bvec_from_pages(struct bio_vec *bvec, struct page **pages,
277 				    int nr_vecs, ssize_t bytes, ssize_t offset)
278 {
279 	unsigned int nr_bvecs = 0;
280 	int i, j;
281 
282 	for (i = 0; i < nr_vecs; i = j) {
283 		size_t size = min_t(size_t, bytes, PAGE_SIZE - offset);
284 		struct folio *folio = page_folio(pages[i]);
285 
286 		bytes -= size;
287 		for (j = i + 1; j < nr_vecs; j++) {
288 			size_t next = min_t(size_t, PAGE_SIZE, bytes);
289 
290 			if (page_folio(pages[j]) != folio ||
291 			    pages[j] != pages[j - 1] + 1)
292 				break;
293 			unpin_user_page(pages[j]);
294 			size += next;
295 			bytes -= next;
296 		}
297 
298 		bvec_set_page(&bvec[nr_bvecs], pages[i], size, offset);
299 		offset = 0;
300 		nr_bvecs++;
301 	}
302 
303 	return nr_bvecs;
304 }
305 
306 int bio_integrity_map_user(struct bio *bio, void __user *ubuf, ssize_t bytes,
307 			   u32 seed)
308 {
309 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
310 	unsigned int align = blk_lim_dma_alignment_and_pad(&q->limits);
311 	struct page *stack_pages[UIO_FASTIOV], **pages = stack_pages;
312 	struct bio_vec stack_vec[UIO_FASTIOV], *bvec = stack_vec;
313 	unsigned int direction, nr_bvecs;
314 	struct iov_iter iter;
315 	int ret, nr_vecs;
316 	size_t offset;
317 	bool copy;
318 
319 	if (bio_integrity(bio))
320 		return -EINVAL;
321 	if (bytes >> SECTOR_SHIFT > queue_max_hw_sectors(q))
322 		return -E2BIG;
323 
324 	if (bio_data_dir(bio) == READ)
325 		direction = ITER_DEST;
326 	else
327 		direction = ITER_SOURCE;
328 
329 	iov_iter_ubuf(&iter, direction, ubuf, bytes);
330 	nr_vecs = iov_iter_npages(&iter, BIO_MAX_VECS + 1);
331 	if (nr_vecs > BIO_MAX_VECS)
332 		return -E2BIG;
333 	if (nr_vecs > UIO_FASTIOV) {
334 		bvec = kcalloc(nr_vecs, sizeof(*bvec), GFP_KERNEL);
335 		if (!bvec)
336 			return -ENOMEM;
337 		pages = NULL;
338 	}
339 
340 	copy = !iov_iter_is_aligned(&iter, align, align);
341 	ret = iov_iter_extract_pages(&iter, &pages, bytes, nr_vecs, 0, &offset);
342 	if (unlikely(ret < 0))
343 		goto free_bvec;
344 
345 	nr_bvecs = bvec_from_pages(bvec, pages, nr_vecs, bytes, offset);
346 	if (pages != stack_pages)
347 		kvfree(pages);
348 	if (nr_bvecs > queue_max_integrity_segments(q))
349 		copy = true;
350 
351 	if (copy)
352 		ret = bio_integrity_copy_user(bio, bvec, nr_bvecs, bytes,
353 					      direction, seed);
354 	else
355 		ret = bio_integrity_init_user(bio, bvec, nr_bvecs, bytes, seed);
356 	if (ret)
357 		goto release_pages;
358 	if (bvec != stack_vec)
359 		kfree(bvec);
360 
361 	return 0;
362 
363 release_pages:
364 	bio_integrity_unpin_bvec(bvec, nr_bvecs, false);
365 free_bvec:
366 	if (bvec != stack_vec)
367 		kfree(bvec);
368 	return ret;
369 }
370 EXPORT_SYMBOL_GPL(bio_integrity_map_user);
371 
372 /**
373  * bio_integrity_prep - Prepare bio for integrity I/O
374  * @bio:	bio to prepare
375  *
376  * Description:  Checks if the bio already has an integrity payload attached.
377  * If it does, the payload has been generated by another kernel subsystem,
378  * and we just pass it through. Otherwise allocates integrity payload.
379  * The bio must have data direction, target device and start sector set priot
380  * to calling.  In the WRITE case, integrity metadata will be generated using
381  * the block device's integrity function.  In the READ case, the buffer
382  * will be prepared for DMA and a suitable end_io handler set up.
383  */
384 bool bio_integrity_prep(struct bio *bio)
385 {
386 	struct bio_integrity_payload *bip;
387 	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
388 	unsigned int len;
389 	void *buf;
390 	gfp_t gfp = GFP_NOIO;
391 
392 	if (!bi)
393 		return true;
394 
395 	if (!bio_sectors(bio))
396 		return true;
397 
398 	/* Already protected? */
399 	if (bio_integrity(bio))
400 		return true;
401 
402 	switch (bio_op(bio)) {
403 	case REQ_OP_READ:
404 		if (bi->flags & BLK_INTEGRITY_NOVERIFY)
405 			return true;
406 		break;
407 	case REQ_OP_WRITE:
408 		if (bi->flags & BLK_INTEGRITY_NOGENERATE)
409 			return true;
410 
411 		/*
412 		 * Zero the memory allocated to not leak uninitialized kernel
413 		 * memory to disk for non-integrity metadata where nothing else
414 		 * initializes the memory.
415 		 */
416 		if (bi->csum_type == BLK_INTEGRITY_CSUM_NONE)
417 			gfp |= __GFP_ZERO;
418 		break;
419 	default:
420 		return true;
421 	}
422 
423 	/* Allocate kernel buffer for protection data */
424 	len = bio_integrity_bytes(bi, bio_sectors(bio));
425 	buf = kmalloc(len, gfp);
426 	if (unlikely(buf == NULL)) {
427 		goto err_end_io;
428 	}
429 
430 	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
431 	if (IS_ERR(bip)) {
432 		kfree(buf);
433 		goto err_end_io;
434 	}
435 
436 	bip->bip_flags |= BIP_BLOCK_INTEGRITY;
437 	bip_set_seed(bip, bio->bi_iter.bi_sector);
438 
439 	if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)
440 		bip->bip_flags |= BIP_IP_CHECKSUM;
441 
442 	if (bio_integrity_add_page(bio, virt_to_page(buf), len,
443 			offset_in_page(buf)) < len) {
444 		printk(KERN_ERR "could not attach integrity payload\n");
445 		goto err_end_io;
446 	}
447 
448 	/* Auto-generate integrity metadata if this is a write */
449 	if (bio_data_dir(bio) == WRITE)
450 		blk_integrity_generate(bio);
451 	else
452 		bip->bio_iter = bio->bi_iter;
453 	return true;
454 
455 err_end_io:
456 	bio->bi_status = BLK_STS_RESOURCE;
457 	bio_endio(bio);
458 	return false;
459 }
460 EXPORT_SYMBOL(bio_integrity_prep);
461 
462 /**
463  * bio_integrity_verify_fn - Integrity I/O completion worker
464  * @work:	Work struct stored in bio to be verified
465  *
466  * Description: This workqueue function is called to complete a READ
467  * request.  The function verifies the transferred integrity metadata
468  * and then calls the original bio end_io function.
469  */
470 static void bio_integrity_verify_fn(struct work_struct *work)
471 {
472 	struct bio_integrity_payload *bip =
473 		container_of(work, struct bio_integrity_payload, bip_work);
474 	struct bio *bio = bip->bip_bio;
475 
476 	blk_integrity_verify(bio);
477 
478 	kfree(bvec_virt(bip->bip_vec));
479 	bio_integrity_free(bio);
480 	bio_endio(bio);
481 }
482 
483 /**
484  * __bio_integrity_endio - Integrity I/O completion function
485  * @bio:	Protected bio
486  *
487  * Description: Completion for integrity I/O
488  *
489  * Normally I/O completion is done in interrupt context.  However,
490  * verifying I/O integrity is a time-consuming task which must be run
491  * in process context.	This function postpones completion
492  * accordingly.
493  */
494 bool __bio_integrity_endio(struct bio *bio)
495 {
496 	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
497 	struct bio_integrity_payload *bip = bio_integrity(bio);
498 
499 	if (bio_op(bio) == REQ_OP_READ && !bio->bi_status && bi->csum_type) {
500 		INIT_WORK(&bip->bip_work, bio_integrity_verify_fn);
501 		queue_work(kintegrityd_wq, &bip->bip_work);
502 		return false;
503 	}
504 
505 	kfree(bvec_virt(bip->bip_vec));
506 	bio_integrity_free(bio);
507 	return true;
508 }
509 
510 /**
511  * bio_integrity_advance - Advance integrity vector
512  * @bio:	bio whose integrity vector to update
513  * @bytes_done:	number of data bytes that have been completed
514  *
515  * Description: This function calculates how many integrity bytes the
516  * number of completed data bytes correspond to and advances the
517  * integrity vector accordingly.
518  */
519 void bio_integrity_advance(struct bio *bio, unsigned int bytes_done)
520 {
521 	struct bio_integrity_payload *bip = bio_integrity(bio);
522 	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
523 	unsigned bytes = bio_integrity_bytes(bi, bytes_done >> 9);
524 
525 	bip->bip_iter.bi_sector += bio_integrity_intervals(bi, bytes_done >> 9);
526 	bvec_iter_advance(bip->bip_vec, &bip->bip_iter, bytes);
527 }
528 
529 /**
530  * bio_integrity_trim - Trim integrity vector
531  * @bio:	bio whose integrity vector to update
532  *
533  * Description: Used to trim the integrity vector in a cloned bio.
534  */
535 void bio_integrity_trim(struct bio *bio)
536 {
537 	struct bio_integrity_payload *bip = bio_integrity(bio);
538 	struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);
539 
540 	bip->bip_iter.bi_size = bio_integrity_bytes(bi, bio_sectors(bio));
541 }
542 EXPORT_SYMBOL(bio_integrity_trim);
543 
544 /**
545  * bio_integrity_clone - Callback for cloning bios with integrity metadata
546  * @bio:	New bio
547  * @bio_src:	Original bio
548  * @gfp_mask:	Memory allocation mask
549  *
550  * Description:	Called to allocate a bip when cloning a bio
551  */
552 int bio_integrity_clone(struct bio *bio, struct bio *bio_src,
553 			gfp_t gfp_mask)
554 {
555 	struct bio_integrity_payload *bip_src = bio_integrity(bio_src);
556 	struct bio_integrity_payload *bip;
557 
558 	BUG_ON(bip_src == NULL);
559 
560 	bip = bio_integrity_alloc(bio, gfp_mask, 0);
561 	if (IS_ERR(bip))
562 		return PTR_ERR(bip);
563 
564 	bip->bip_vec = bip_src->bip_vec;
565 	bip->bip_iter = bip_src->bip_iter;
566 	bip->bip_flags = bip_src->bip_flags & ~BIP_BLOCK_INTEGRITY;
567 
568 	return 0;
569 }
570 
571 int bioset_integrity_create(struct bio_set *bs, int pool_size)
572 {
573 	if (mempool_initialized(&bs->bio_integrity_pool))
574 		return 0;
575 
576 	if (mempool_init_slab_pool(&bs->bio_integrity_pool,
577 				   pool_size, bip_slab))
578 		return -1;
579 
580 	if (biovec_init_pool(&bs->bvec_integrity_pool, pool_size)) {
581 		mempool_exit(&bs->bio_integrity_pool);
582 		return -1;
583 	}
584 
585 	return 0;
586 }
587 EXPORT_SYMBOL(bioset_integrity_create);
588 
589 void bioset_integrity_free(struct bio_set *bs)
590 {
591 	mempool_exit(&bs->bio_integrity_pool);
592 	mempool_exit(&bs->bvec_integrity_pool);
593 }
594 
595 void __init bio_integrity_init(void)
596 {
597 	/*
598 	 * kintegrityd won't block much but may burn a lot of CPU cycles.
599 	 * Make it highpri CPU intensive wq with max concurrency of 1.
600 	 */
601 	kintegrityd_wq = alloc_workqueue("kintegrityd", WQ_MEM_RECLAIM |
602 					 WQ_HIGHPRI | WQ_CPU_INTENSIVE, 1);
603 	if (!kintegrityd_wq)
604 		panic("Failed to create kintegrityd\n");
605 
606 	bip_slab = kmem_cache_create("bio_integrity_payload",
607 				     sizeof(struct bio_integrity_payload) +
608 				     sizeof(struct bio_vec) * BIO_INLINE_VECS,
609 				     0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
610 }
611