xref: /linux/block/bio-integrity.c (revision b7019ac550eb3916f34d79db583e9b7ea2524afa)
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/blkdev.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 #define BIP_INLINE_VECS	4
18 
19 static struct kmem_cache *bip_slab;
20 static struct workqueue_struct *kintegrityd_wq;
21 
22 void blk_flush_integrity(void)
23 {
24 	flush_workqueue(kintegrityd_wq);
25 }
26 
27 /**
28  * bio_integrity_alloc - Allocate integrity payload and attach it to bio
29  * @bio:	bio to attach integrity metadata to
30  * @gfp_mask:	Memory allocation mask
31  * @nr_vecs:	Number of integrity metadata scatter-gather elements
32  *
33  * Description: This function prepares a bio for attaching integrity
34  * metadata.  nr_vecs specifies the maximum number of pages containing
35  * integrity metadata that can be attached.
36  */
37 struct bio_integrity_payload *bio_integrity_alloc(struct bio *bio,
38 						  gfp_t gfp_mask,
39 						  unsigned int nr_vecs)
40 {
41 	struct bio_integrity_payload *bip;
42 	struct bio_set *bs = bio->bi_pool;
43 	unsigned inline_vecs;
44 
45 	if (!bs || !mempool_initialized(&bs->bio_integrity_pool)) {
46 		bip = kmalloc(struct_size(bip, bip_inline_vecs, nr_vecs), gfp_mask);
47 		inline_vecs = nr_vecs;
48 	} else {
49 		bip = mempool_alloc(&bs->bio_integrity_pool, gfp_mask);
50 		inline_vecs = BIP_INLINE_VECS;
51 	}
52 
53 	if (unlikely(!bip))
54 		return ERR_PTR(-ENOMEM);
55 
56 	memset(bip, 0, sizeof(*bip));
57 
58 	if (nr_vecs > inline_vecs) {
59 		unsigned long idx = 0;
60 
61 		bip->bip_vec = bvec_alloc(gfp_mask, nr_vecs, &idx,
62 					  &bs->bvec_integrity_pool);
63 		if (!bip->bip_vec)
64 			goto err;
65 		bip->bip_max_vcnt = bvec_nr_vecs(idx);
66 		bip->bip_slab = idx;
67 	} else {
68 		bip->bip_vec = bip->bip_inline_vecs;
69 		bip->bip_max_vcnt = inline_vecs;
70 	}
71 
72 	bip->bip_bio = bio;
73 	bio->bi_integrity = bip;
74 	bio->bi_opf |= REQ_INTEGRITY;
75 
76 	return bip;
77 err:
78 	mempool_free(bip, &bs->bio_integrity_pool);
79 	return ERR_PTR(-ENOMEM);
80 }
81 EXPORT_SYMBOL(bio_integrity_alloc);
82 
83 /**
84  * bio_integrity_free - Free bio integrity payload
85  * @bio:	bio containing bip to be freed
86  *
87  * Description: Used to free the integrity portion of a bio. Usually
88  * called from bio_free().
89  */
90 static void bio_integrity_free(struct bio *bio)
91 {
92 	struct bio_integrity_payload *bip = bio_integrity(bio);
93 	struct bio_set *bs = bio->bi_pool;
94 
95 	if (bip->bip_flags & BIP_BLOCK_INTEGRITY)
96 		kfree(page_address(bip->bip_vec->bv_page) +
97 		      bip->bip_vec->bv_offset);
98 
99 	if (bs && mempool_initialized(&bs->bio_integrity_pool)) {
100 		bvec_free(&bs->bvec_integrity_pool, bip->bip_vec, bip->bip_slab);
101 
102 		mempool_free(bip, &bs->bio_integrity_pool);
103 	} else {
104 		kfree(bip);
105 	}
106 
107 	bio->bi_integrity = NULL;
108 	bio->bi_opf &= ~REQ_INTEGRITY;
109 }
110 
111 /**
112  * bio_integrity_add_page - Attach integrity metadata
113  * @bio:	bio to update
114  * @page:	page containing integrity metadata
115  * @len:	number of bytes of integrity metadata in page
116  * @offset:	start offset within page
117  *
118  * Description: Attach a page containing integrity metadata to bio.
119  */
120 int bio_integrity_add_page(struct bio *bio, struct page *page,
121 			   unsigned int len, unsigned int offset)
122 {
123 	struct bio_integrity_payload *bip = bio_integrity(bio);
124 	struct bio_vec *iv;
125 
126 	if (bip->bip_vcnt >= bip->bip_max_vcnt) {
127 		printk(KERN_ERR "%s: bip_vec full\n", __func__);
128 		return 0;
129 	}
130 
131 	iv = bip->bip_vec + bip->bip_vcnt;
132 
133 	if (bip->bip_vcnt &&
134 	    bvec_gap_to_prev(bio->bi_disk->queue,
135 			     &bip->bip_vec[bip->bip_vcnt - 1], offset))
136 		return 0;
137 
138 	iv->bv_page = page;
139 	iv->bv_len = len;
140 	iv->bv_offset = offset;
141 	bip->bip_vcnt++;
142 
143 	return len;
144 }
145 EXPORT_SYMBOL(bio_integrity_add_page);
146 
147 /**
148  * bio_integrity_process - Process integrity metadata for a bio
149  * @bio:	bio to generate/verify integrity metadata for
150  * @proc_iter:  iterator to process
151  * @proc_fn:	Pointer to the relevant processing function
152  */
153 static blk_status_t bio_integrity_process(struct bio *bio,
154 		struct bvec_iter *proc_iter, integrity_processing_fn *proc_fn)
155 {
156 	struct blk_integrity *bi = blk_get_integrity(bio->bi_disk);
157 	struct blk_integrity_iter iter;
158 	struct bvec_iter bviter;
159 	struct bio_vec bv;
160 	struct bio_integrity_payload *bip = bio_integrity(bio);
161 	blk_status_t ret = BLK_STS_OK;
162 	void *prot_buf = page_address(bip->bip_vec->bv_page) +
163 		bip->bip_vec->bv_offset;
164 
165 	iter.disk_name = bio->bi_disk->disk_name;
166 	iter.interval = 1 << bi->interval_exp;
167 	iter.seed = proc_iter->bi_sector;
168 	iter.prot_buf = prot_buf;
169 
170 	__bio_for_each_segment(bv, bio, bviter, *proc_iter) {
171 		void *kaddr = kmap_atomic(bv.bv_page);
172 
173 		iter.data_buf = kaddr + bv.bv_offset;
174 		iter.data_size = bv.bv_len;
175 
176 		ret = proc_fn(&iter);
177 		if (ret) {
178 			kunmap_atomic(kaddr);
179 			return ret;
180 		}
181 
182 		kunmap_atomic(kaddr);
183 	}
184 	return ret;
185 }
186 
187 /**
188  * bio_integrity_prep - Prepare bio for integrity I/O
189  * @bio:	bio to prepare
190  *
191  * Description:  Checks if the bio already has an integrity payload attached.
192  * If it does, the payload has been generated by another kernel subsystem,
193  * and we just pass it through. Otherwise allocates integrity payload.
194  * The bio must have data direction, target device and start sector set priot
195  * to calling.  In the WRITE case, integrity metadata will be generated using
196  * the block device's integrity function.  In the READ case, the buffer
197  * will be prepared for DMA and a suitable end_io handler set up.
198  */
199 bool bio_integrity_prep(struct bio *bio)
200 {
201 	struct bio_integrity_payload *bip;
202 	struct blk_integrity *bi = blk_get_integrity(bio->bi_disk);
203 	struct request_queue *q = bio->bi_disk->queue;
204 	void *buf;
205 	unsigned long start, end;
206 	unsigned int len, nr_pages;
207 	unsigned int bytes, offset, i;
208 	unsigned int intervals;
209 	blk_status_t status;
210 
211 	if (!bi)
212 		return true;
213 
214 	if (bio_op(bio) != REQ_OP_READ && bio_op(bio) != REQ_OP_WRITE)
215 		return true;
216 
217 	if (!bio_sectors(bio))
218 		return true;
219 
220 	/* Already protected? */
221 	if (bio_integrity(bio))
222 		return true;
223 
224 	if (bio_data_dir(bio) == READ) {
225 		if (!bi->profile->verify_fn ||
226 		    !(bi->flags & BLK_INTEGRITY_VERIFY))
227 			return true;
228 	} else {
229 		if (!bi->profile->generate_fn ||
230 		    !(bi->flags & BLK_INTEGRITY_GENERATE))
231 			return true;
232 	}
233 	intervals = bio_integrity_intervals(bi, bio_sectors(bio));
234 
235 	/* Allocate kernel buffer for protection data */
236 	len = intervals * bi->tuple_size;
237 	buf = kmalloc(len, GFP_NOIO | q->bounce_gfp);
238 	status = BLK_STS_RESOURCE;
239 	if (unlikely(buf == NULL)) {
240 		printk(KERN_ERR "could not allocate integrity buffer\n");
241 		goto err_end_io;
242 	}
243 
244 	end = (((unsigned long) buf) + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
245 	start = ((unsigned long) buf) >> PAGE_SHIFT;
246 	nr_pages = end - start;
247 
248 	/* Allocate bio integrity payload and integrity vectors */
249 	bip = bio_integrity_alloc(bio, GFP_NOIO, nr_pages);
250 	if (IS_ERR(bip)) {
251 		printk(KERN_ERR "could not allocate data integrity bioset\n");
252 		kfree(buf);
253 		status = BLK_STS_RESOURCE;
254 		goto err_end_io;
255 	}
256 
257 	bip->bip_flags |= BIP_BLOCK_INTEGRITY;
258 	bip->bip_iter.bi_size = len;
259 	bip_set_seed(bip, bio->bi_iter.bi_sector);
260 
261 	if (bi->flags & BLK_INTEGRITY_IP_CHECKSUM)
262 		bip->bip_flags |= BIP_IP_CHECKSUM;
263 
264 	/* Map it */
265 	offset = offset_in_page(buf);
266 	for (i = 0 ; i < nr_pages ; i++) {
267 		int ret;
268 		bytes = PAGE_SIZE - offset;
269 
270 		if (len <= 0)
271 			break;
272 
273 		if (bytes > len)
274 			bytes = len;
275 
276 		ret = bio_integrity_add_page(bio, virt_to_page(buf),
277 					     bytes, offset);
278 
279 		if (ret == 0)
280 			return false;
281 
282 		if (ret < bytes)
283 			break;
284 
285 		buf += bytes;
286 		len -= bytes;
287 		offset = 0;
288 	}
289 
290 	/* Auto-generate integrity metadata if this is a write */
291 	if (bio_data_dir(bio) == WRITE) {
292 		bio_integrity_process(bio, &bio->bi_iter,
293 				      bi->profile->generate_fn);
294 	} else {
295 		bip->bio_iter = bio->bi_iter;
296 	}
297 	return true;
298 
299 err_end_io:
300 	bio->bi_status = status;
301 	bio_endio(bio);
302 	return false;
303 
304 }
305 EXPORT_SYMBOL(bio_integrity_prep);
306 
307 /**
308  * bio_integrity_verify_fn - Integrity I/O completion worker
309  * @work:	Work struct stored in bio to be verified
310  *
311  * Description: This workqueue function is called to complete a READ
312  * request.  The function verifies the transferred integrity metadata
313  * and then calls the original bio end_io function.
314  */
315 static void bio_integrity_verify_fn(struct work_struct *work)
316 {
317 	struct bio_integrity_payload *bip =
318 		container_of(work, struct bio_integrity_payload, bip_work);
319 	struct bio *bio = bip->bip_bio;
320 	struct blk_integrity *bi = blk_get_integrity(bio->bi_disk);
321 
322 	/*
323 	 * At the moment verify is called bio's iterator was advanced
324 	 * during split and completion, we need to rewind iterator to
325 	 * it's original position.
326 	 */
327 	bio->bi_status = bio_integrity_process(bio, &bip->bio_iter,
328 						bi->profile->verify_fn);
329 	bio_integrity_free(bio);
330 	bio_endio(bio);
331 }
332 
333 /**
334  * __bio_integrity_endio - Integrity I/O completion function
335  * @bio:	Protected bio
336  *
337  * Description: Completion for integrity I/O
338  *
339  * Normally I/O completion is done in interrupt context.  However,
340  * verifying I/O integrity is a time-consuming task which must be run
341  * in process context.	This function postpones completion
342  * accordingly.
343  */
344 bool __bio_integrity_endio(struct bio *bio)
345 {
346 	struct blk_integrity *bi = blk_get_integrity(bio->bi_disk);
347 	struct bio_integrity_payload *bip = bio_integrity(bio);
348 
349 	if (bio_op(bio) == REQ_OP_READ && !bio->bi_status &&
350 	    (bip->bip_flags & BIP_BLOCK_INTEGRITY) && bi->profile->verify_fn) {
351 		INIT_WORK(&bip->bip_work, bio_integrity_verify_fn);
352 		queue_work(kintegrityd_wq, &bip->bip_work);
353 		return false;
354 	}
355 
356 	bio_integrity_free(bio);
357 	return true;
358 }
359 
360 /**
361  * bio_integrity_advance - Advance integrity vector
362  * @bio:	bio whose integrity vector to update
363  * @bytes_done:	number of data bytes that have been completed
364  *
365  * Description: This function calculates how many integrity bytes the
366  * number of completed data bytes correspond to and advances the
367  * integrity vector accordingly.
368  */
369 void bio_integrity_advance(struct bio *bio, unsigned int bytes_done)
370 {
371 	struct bio_integrity_payload *bip = bio_integrity(bio);
372 	struct blk_integrity *bi = blk_get_integrity(bio->bi_disk);
373 	unsigned bytes = bio_integrity_bytes(bi, bytes_done >> 9);
374 
375 	bip->bip_iter.bi_sector += bytes_done >> 9;
376 	bvec_iter_advance(bip->bip_vec, &bip->bip_iter, bytes);
377 }
378 
379 /**
380  * bio_integrity_trim - Trim integrity vector
381  * @bio:	bio whose integrity vector to update
382  *
383  * Description: Used to trim the integrity vector in a cloned bio.
384  */
385 void bio_integrity_trim(struct bio *bio)
386 {
387 	struct bio_integrity_payload *bip = bio_integrity(bio);
388 	struct blk_integrity *bi = blk_get_integrity(bio->bi_disk);
389 
390 	bip->bip_iter.bi_size = bio_integrity_bytes(bi, bio_sectors(bio));
391 }
392 EXPORT_SYMBOL(bio_integrity_trim);
393 
394 /**
395  * bio_integrity_clone - Callback for cloning bios with integrity metadata
396  * @bio:	New bio
397  * @bio_src:	Original bio
398  * @gfp_mask:	Memory allocation mask
399  *
400  * Description:	Called to allocate a bip when cloning a bio
401  */
402 int bio_integrity_clone(struct bio *bio, struct bio *bio_src,
403 			gfp_t gfp_mask)
404 {
405 	struct bio_integrity_payload *bip_src = bio_integrity(bio_src);
406 	struct bio_integrity_payload *bip;
407 
408 	BUG_ON(bip_src == NULL);
409 
410 	bip = bio_integrity_alloc(bio, gfp_mask, bip_src->bip_vcnt);
411 	if (IS_ERR(bip))
412 		return PTR_ERR(bip);
413 
414 	memcpy(bip->bip_vec, bip_src->bip_vec,
415 	       bip_src->bip_vcnt * sizeof(struct bio_vec));
416 
417 	bip->bip_vcnt = bip_src->bip_vcnt;
418 	bip->bip_iter = bip_src->bip_iter;
419 
420 	return 0;
421 }
422 EXPORT_SYMBOL(bio_integrity_clone);
423 
424 int bioset_integrity_create(struct bio_set *bs, int pool_size)
425 {
426 	if (mempool_initialized(&bs->bio_integrity_pool))
427 		return 0;
428 
429 	if (mempool_init_slab_pool(&bs->bio_integrity_pool,
430 				   pool_size, bip_slab))
431 		return -1;
432 
433 	if (biovec_init_pool(&bs->bvec_integrity_pool, pool_size)) {
434 		mempool_exit(&bs->bio_integrity_pool);
435 		return -1;
436 	}
437 
438 	return 0;
439 }
440 EXPORT_SYMBOL(bioset_integrity_create);
441 
442 void bioset_integrity_free(struct bio_set *bs)
443 {
444 	mempool_exit(&bs->bio_integrity_pool);
445 	mempool_exit(&bs->bvec_integrity_pool);
446 }
447 
448 void __init bio_integrity_init(void)
449 {
450 	/*
451 	 * kintegrityd won't block much but may burn a lot of CPU cycles.
452 	 * Make it highpri CPU intensive wq with max concurrency of 1.
453 	 */
454 	kintegrityd_wq = alloc_workqueue("kintegrityd", WQ_MEM_RECLAIM |
455 					 WQ_HIGHPRI | WQ_CPU_INTENSIVE, 1);
456 	if (!kintegrityd_wq)
457 		panic("Failed to create kintegrityd\n");
458 
459 	bip_slab = kmem_cache_create("bio_integrity_payload",
460 				     sizeof(struct bio_integrity_payload) +
461 				     sizeof(struct bio_vec) * BIP_INLINE_VECS,
462 				     0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
463 }
464