xref: /illumos-gate/usr/src/uts/common/fs/zfs/zio_checksum.c (revision e07d85f87c3920e032adb855fdc500e4616c7718)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2013, 2016 by Delphix. All rights reserved.
24  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
25  * Copyright 2013 Saso Kiselkov. All rights reserved.
26  */
27 
28 #include <sys/zfs_context.h>
29 #include <sys/spa.h>
30 #include <sys/spa_impl.h>
31 #include <sys/zio.h>
32 #include <sys/zio_checksum.h>
33 #include <sys/zil.h>
34 #include <sys/abd.h>
35 #include <zfs_fletcher.h>
36 
37 /*
38  * Checksum vectors.
39  *
40  * In the SPA, everything is checksummed.  We support checksum vectors
41  * for three distinct reasons:
42  *
43  *   1. Different kinds of data need different levels of protection.
44  *	For SPA metadata, we always want a very strong checksum.
45  *	For user data, we let users make the trade-off between speed
46  *	and checksum strength.
47  *
48  *   2. Cryptographic hash and MAC algorithms are an area of active research.
49  *	It is likely that in future hash functions will be at least as strong
50  *	as current best-of-breed, and may be substantially faster as well.
51  *	We want the ability to take advantage of these new hashes as soon as
52  *	they become available.
53  *
54  *   3. If someone develops hardware that can compute a strong hash quickly,
55  *	we want the ability to take advantage of that hardware.
56  *
57  * Of course, we don't want a checksum upgrade to invalidate existing
58  * data, so we store the checksum *function* in eight bits of the bp.
59  * This gives us room for up to 256 different checksum functions.
60  *
61  * When writing a block, we always checksum it with the latest-and-greatest
62  * checksum function of the appropriate strength.  When reading a block,
63  * we compare the expected checksum against the actual checksum, which we
64  * compute via the checksum function specified by BP_GET_CHECKSUM(bp).
65  *
66  * SALTED CHECKSUMS
67  *
68  * To enable the use of less secure hash algorithms with dedup, we
69  * introduce the notion of salted checksums (MACs, really).  A salted
70  * checksum is fed both a random 256-bit value (the salt) and the data
71  * to be checksummed.  This salt is kept secret (stored on the pool, but
72  * never shown to the user).  Thus even if an attacker knew of collision
73  * weaknesses in the hash algorithm, they won't be able to mount a known
74  * plaintext attack on the DDT, since the actual hash value cannot be
75  * known ahead of time.  How the salt is used is algorithm-specific
76  * (some might simply prefix it to the data block, others might need to
77  * utilize a full-blown HMAC).  On disk the salt is stored in a ZAP
78  * object in the MOS (DMU_POOL_CHECKSUM_SALT).
79  *
80  * CONTEXT TEMPLATES
81  *
82  * Some hashing algorithms need to perform a substantial amount of
83  * initialization work (e.g. salted checksums above may need to pre-hash
84  * the salt) before being able to process data.  Performing this
85  * redundant work for each block would be wasteful, so we instead allow
86  * a checksum algorithm to do the work once (the first time it's used)
87  * and then keep this pre-initialized context as a template inside the
88  * spa_t (spa_cksum_tmpls).  If the zio_checksum_info_t contains
89  * non-NULL ci_tmpl_init and ci_tmpl_free callbacks, they are used to
90  * construct and destruct the pre-initialized checksum context.  The
91  * pre-initialized context is then reused during each checksum
92  * invocation and passed to the checksum function.
93  */
94 
95 /*ARGSUSED*/
96 static void
97 abd_checksum_off(abd_t *abd, uint64_t size,
98     const void *ctx_template, zio_cksum_t *zcp)
99 {
100 	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
101 }
102 
103 /*ARGSUSED*/
104 void
105 abd_fletcher_2_native(abd_t *abd, uint64_t size,
106     const void *ctx_template, zio_cksum_t *zcp)
107 {
108 	fletcher_init(zcp);
109 	(void) abd_iterate_func(abd, 0, size,
110 	    fletcher_2_incremental_native, zcp);
111 }
112 
113 /*ARGSUSED*/
114 void
115 abd_fletcher_2_byteswap(abd_t *abd, uint64_t size,
116     const void *ctx_template, zio_cksum_t *zcp)
117 {
118 	fletcher_init(zcp);
119 	(void) abd_iterate_func(abd, 0, size,
120 	    fletcher_2_incremental_byteswap, zcp);
121 }
122 
123 /*ARGSUSED*/
124 void
125 abd_fletcher_4_native(abd_t *abd, uint64_t size,
126     const void *ctx_template, zio_cksum_t *zcp)
127 {
128 	fletcher_init(zcp);
129 	(void) abd_iterate_func(abd, 0, size,
130 	    fletcher_4_incremental_native, zcp);
131 }
132 
133 /*ARGSUSED*/
134 void
135 abd_fletcher_4_byteswap(abd_t *abd, uint64_t size,
136     const void *ctx_template, zio_cksum_t *zcp)
137 {
138 	fletcher_init(zcp);
139 	(void) abd_iterate_func(abd, 0, size,
140 	    fletcher_4_incremental_byteswap, zcp);
141 }
142 
143 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
144 	{{NULL, NULL}, NULL, NULL, 0, "inherit"},
145 	{{NULL, NULL}, NULL, NULL, 0, "on"},
146 	{{abd_checksum_off,		abd_checksum_off},
147 	    NULL, NULL, 0, "off"},
148 	{{abd_checksum_SHA256,		abd_checksum_SHA256},
149 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
150 	    "label"},
151 	{{abd_checksum_SHA256,		abd_checksum_SHA256},
152 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
153 	    "gang_header"},
154 	{{abd_fletcher_2_native,	abd_fletcher_2_byteswap},
155 	    NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog"},
156 	{{abd_fletcher_2_native,	abd_fletcher_2_byteswap},
157 	    NULL, NULL, 0, "fletcher2"},
158 	{{abd_fletcher_4_native,	abd_fletcher_4_byteswap},
159 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA, "fletcher4"},
160 	{{abd_checksum_SHA256,		abd_checksum_SHA256},
161 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
162 	    ZCHECKSUM_FLAG_NOPWRITE, "sha256"},
163 	{{abd_fletcher_4_native,	abd_fletcher_4_byteswap},
164 	    NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog2"},
165 	{{abd_checksum_off,		abd_checksum_off},
166 	    NULL, NULL, 0, "noparity"},
167 	{{abd_checksum_SHA512_native,	abd_checksum_SHA512_byteswap},
168 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
169 	    ZCHECKSUM_FLAG_NOPWRITE, "sha512"},
170 	{{abd_checksum_skein_native,	abd_checksum_skein_byteswap},
171 	    abd_checksum_skein_tmpl_init, abd_checksum_skein_tmpl_free,
172 	    ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
173 	    ZCHECKSUM_FLAG_SALTED | ZCHECKSUM_FLAG_NOPWRITE, "skein"},
174 	{{abd_checksum_edonr_native,	abd_checksum_edonr_byteswap},
175 	    abd_checksum_edonr_tmpl_init, abd_checksum_edonr_tmpl_free,
176 	    ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_SALTED |
177 	    ZCHECKSUM_FLAG_NOPWRITE, "edonr"},
178 };
179 
180 /*
181  * The flag corresponding to the "verify" in dedup=[checksum,]verify
182  * must be cleared first, so callers should use ZIO_CHECKSUM_MASK.
183  */
184 spa_feature_t
185 zio_checksum_to_feature(enum zio_checksum cksum)
186 {
187 	VERIFY((cksum & ~ZIO_CHECKSUM_MASK) == 0);
188 
189 	switch (cksum) {
190 	case ZIO_CHECKSUM_SHA512:
191 		return (SPA_FEATURE_SHA512);
192 	case ZIO_CHECKSUM_SKEIN:
193 		return (SPA_FEATURE_SKEIN);
194 	case ZIO_CHECKSUM_EDONR:
195 		return (SPA_FEATURE_EDONR);
196 	}
197 	return (SPA_FEATURE_NONE);
198 }
199 
200 enum zio_checksum
201 zio_checksum_select(enum zio_checksum child, enum zio_checksum parent)
202 {
203 	ASSERT(child < ZIO_CHECKSUM_FUNCTIONS);
204 	ASSERT(parent < ZIO_CHECKSUM_FUNCTIONS);
205 	ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
206 
207 	if (child == ZIO_CHECKSUM_INHERIT)
208 		return (parent);
209 
210 	if (child == ZIO_CHECKSUM_ON)
211 		return (ZIO_CHECKSUM_ON_VALUE);
212 
213 	return (child);
214 }
215 
216 enum zio_checksum
217 zio_checksum_dedup_select(spa_t *spa, enum zio_checksum child,
218     enum zio_checksum parent)
219 {
220 	ASSERT((child & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
221 	ASSERT((parent & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
222 	ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
223 
224 	if (child == ZIO_CHECKSUM_INHERIT)
225 		return (parent);
226 
227 	if (child == ZIO_CHECKSUM_ON)
228 		return (spa_dedup_checksum(spa));
229 
230 	if (child == (ZIO_CHECKSUM_ON | ZIO_CHECKSUM_VERIFY))
231 		return (spa_dedup_checksum(spa) | ZIO_CHECKSUM_VERIFY);
232 
233 	ASSERT((zio_checksum_table[child & ZIO_CHECKSUM_MASK].ci_flags &
234 	    ZCHECKSUM_FLAG_DEDUP) ||
235 	    (child & ZIO_CHECKSUM_VERIFY) || child == ZIO_CHECKSUM_OFF);
236 
237 	return (child);
238 }
239 
240 /*
241  * Set the external verifier for a gang block based on <vdev, offset, txg>,
242  * a tuple which is guaranteed to be unique for the life of the pool.
243  */
244 static void
245 zio_checksum_gang_verifier(zio_cksum_t *zcp, blkptr_t *bp)
246 {
247 	dva_t *dva = BP_IDENTITY(bp);
248 	uint64_t txg = BP_PHYSICAL_BIRTH(bp);
249 
250 	ASSERT(BP_IS_GANG(bp));
251 
252 	ZIO_SET_CHECKSUM(zcp, DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), txg, 0);
253 }
254 
255 /*
256  * Set the external verifier for a label block based on its offset.
257  * The vdev is implicit, and the txg is unknowable at pool open time --
258  * hence the logic in vdev_uberblock_load() to find the most recent copy.
259  */
260 static void
261 zio_checksum_label_verifier(zio_cksum_t *zcp, uint64_t offset)
262 {
263 	ZIO_SET_CHECKSUM(zcp, offset, 0, 0, 0);
264 }
265 
266 /*
267  * Calls the template init function of a checksum which supports context
268  * templates and installs the template into the spa_t.
269  */
270 static void
271 zio_checksum_template_init(enum zio_checksum checksum, spa_t *spa)
272 {
273 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
274 
275 	if (ci->ci_tmpl_init == NULL)
276 		return;
277 	if (spa->spa_cksum_tmpls[checksum] != NULL)
278 		return;
279 
280 	VERIFY(ci->ci_tmpl_free != NULL);
281 	mutex_enter(&spa->spa_cksum_tmpls_lock);
282 	if (spa->spa_cksum_tmpls[checksum] == NULL) {
283 		spa->spa_cksum_tmpls[checksum] =
284 		    ci->ci_tmpl_init(&spa->spa_cksum_salt);
285 		VERIFY(spa->spa_cksum_tmpls[checksum] != NULL);
286 	}
287 	mutex_exit(&spa->spa_cksum_tmpls_lock);
288 }
289 
290 /*
291  * Generate the checksum.
292  */
293 void
294 zio_checksum_compute(zio_t *zio, enum zio_checksum checksum,
295     abd_t *abd, uint64_t size)
296 {
297 	blkptr_t *bp = zio->io_bp;
298 	uint64_t offset = zio->io_offset;
299 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
300 	zio_cksum_t cksum;
301 	spa_t *spa = zio->io_spa;
302 
303 	ASSERT((uint_t)checksum < ZIO_CHECKSUM_FUNCTIONS);
304 	ASSERT(ci->ci_func[0] != NULL);
305 
306 	zio_checksum_template_init(checksum, spa);
307 
308 	if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
309 		zio_eck_t *eck;
310 		void *data = abd_to_buf(abd);
311 
312 		if (checksum == ZIO_CHECKSUM_ZILOG2) {
313 			zil_chain_t *zilc = data;
314 
315 			size = P2ROUNDUP_TYPED(zilc->zc_nused, ZIL_MIN_BLKSZ,
316 			    uint64_t);
317 			eck = &zilc->zc_eck;
318 		} else {
319 			eck = (zio_eck_t *)((char *)data + size) - 1;
320 		}
321 		if (checksum == ZIO_CHECKSUM_GANG_HEADER)
322 			zio_checksum_gang_verifier(&eck->zec_cksum, bp);
323 		else if (checksum == ZIO_CHECKSUM_LABEL)
324 			zio_checksum_label_verifier(&eck->zec_cksum, offset);
325 		else
326 			bp->blk_cksum = eck->zec_cksum;
327 		eck->zec_magic = ZEC_MAGIC;
328 		ci->ci_func[0](abd, size, spa->spa_cksum_tmpls[checksum],
329 		    &cksum);
330 		eck->zec_cksum = cksum;
331 	} else {
332 		ci->ci_func[0](abd, size, spa->spa_cksum_tmpls[checksum],
333 		    &bp->blk_cksum);
334 	}
335 }
336 
337 int
338 zio_checksum_error_impl(spa_t *spa, blkptr_t *bp, enum zio_checksum checksum,
339     abd_t *abd, uint64_t size, uint64_t offset, zio_bad_cksum_t *info)
340 {
341 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
342 	zio_cksum_t actual_cksum, expected_cksum;
343 	int byteswap;
344 
345 	if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL)
346 		return (SET_ERROR(EINVAL));
347 
348 	zio_checksum_template_init(checksum, spa);
349 
350 	if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
351 		zio_eck_t *eck;
352 		zio_cksum_t verifier;
353 		uint64_t data_size = size;
354 		void *data = abd_borrow_buf_copy(abd, data_size);
355 
356 		if (checksum == ZIO_CHECKSUM_ZILOG2) {
357 			zil_chain_t *zilc = data;
358 			uint64_t nused;
359 
360 			eck = &zilc->zc_eck;
361 			if (eck->zec_magic == ZEC_MAGIC) {
362 				nused = zilc->zc_nused;
363 			} else if (eck->zec_magic == BSWAP_64(ZEC_MAGIC)) {
364 				nused = BSWAP_64(zilc->zc_nused);
365 			} else {
366 				abd_return_buf(abd, data, data_size);
367 				return (SET_ERROR(ECKSUM));
368 			}
369 
370 			if (nused > data_size) {
371 				abd_return_buf(abd, data, data_size);
372 				return (SET_ERROR(ECKSUM));
373 			}
374 
375 			size = P2ROUNDUP_TYPED(nused, ZIL_MIN_BLKSZ, uint64_t);
376 		} else {
377 			eck = (zio_eck_t *)((char *)data + data_size) - 1;
378 		}
379 
380 		if (checksum == ZIO_CHECKSUM_GANG_HEADER)
381 			zio_checksum_gang_verifier(&verifier, bp);
382 		else if (checksum == ZIO_CHECKSUM_LABEL)
383 			zio_checksum_label_verifier(&verifier, offset);
384 		else
385 			verifier = bp->blk_cksum;
386 
387 		byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
388 
389 		if (byteswap)
390 			byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
391 
392 		size_t eck_offset = (size_t)(&eck->zec_cksum) - (size_t)data;
393 		expected_cksum = eck->zec_cksum;
394 		eck->zec_cksum = verifier;
395 		abd_return_buf_copy(abd, data, data_size);
396 
397 		ci->ci_func[byteswap](abd, size,
398 		    spa->spa_cksum_tmpls[checksum], &actual_cksum);
399 		abd_copy_from_buf_off(abd, &expected_cksum,
400 		    eck_offset, sizeof (zio_cksum_t));
401 
402 		if (byteswap) {
403 			byteswap_uint64_array(&expected_cksum,
404 			    sizeof (zio_cksum_t));
405 		}
406 	} else {
407 		byteswap = BP_SHOULD_BYTESWAP(bp);
408 		expected_cksum = bp->blk_cksum;
409 		ci->ci_func[byteswap](abd, size,
410 		    spa->spa_cksum_tmpls[checksum], &actual_cksum);
411 	}
412 
413 	if (info != NULL) {
414 		info->zbc_expected = expected_cksum;
415 		info->zbc_actual = actual_cksum;
416 		info->zbc_checksum_name = ci->ci_name;
417 		info->zbc_byteswapped = byteswap;
418 		info->zbc_injected = 0;
419 		info->zbc_has_cksum = 1;
420 	}
421 
422 	if (!ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
423 		return (SET_ERROR(ECKSUM));
424 
425 	return (0);
426 }
427 
428 int
429 zio_checksum_error(zio_t *zio, zio_bad_cksum_t *info)
430 {
431 	blkptr_t *bp = zio->io_bp;
432 	uint_t checksum = (bp == NULL ? zio->io_prop.zp_checksum :
433 	    (BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : BP_GET_CHECKSUM(bp)));
434 	int error;
435 	uint64_t size = (bp == NULL ? zio->io_size :
436 	    (BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(bp)));
437 	uint64_t offset = zio->io_offset;
438 	abd_t *data = zio->io_abd;
439 	spa_t *spa = zio->io_spa;
440 
441 	error = zio_checksum_error_impl(spa, bp, checksum, data, size,
442 	    offset, info);
443 
444 	if (zio_injection_enabled && error == 0 && zio->io_error == 0) {
445 		error = zio_handle_fault_injection(zio, ECKSUM);
446 		if (error != 0)
447 			info->zbc_injected = 1;
448 	}
449 
450 	return (error);
451 }
452 
453 /*
454  * Called by a spa_t that's about to be deallocated. This steps through
455  * all of the checksum context templates and deallocates any that were
456  * initialized using the algorithm-specific template init function.
457  */
458 void
459 zio_checksum_templates_free(spa_t *spa)
460 {
461 	for (enum zio_checksum checksum = 0;
462 	    checksum < ZIO_CHECKSUM_FUNCTIONS; checksum++) {
463 		if (spa->spa_cksum_tmpls[checksum] != NULL) {
464 			zio_checksum_info_t *ci = &zio_checksum_table[checksum];
465 
466 			VERIFY(ci->ci_tmpl_free != NULL);
467 			ci->ci_tmpl_free(spa->spa_cksum_tmpls[checksum]);
468 			spa->spa_cksum_tmpls[checksum] = NULL;
469 		}
470 	}
471 }
472