xref: /illumos-gate/usr/src/uts/common/fs/zfs/zio_checksum.c (revision 9a686fbc186e8e2a64e9a5094d44c7d6fa0ea167)
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, 2015 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 <zfs_fletcher.h>
35 
36 /*
37  * Checksum vectors.
38  *
39  * In the SPA, everything is checksummed.  We support checksum vectors
40  * for three distinct reasons:
41  *
42  *   1. Different kinds of data need different levels of protection.
43  *	For SPA metadata, we always want a very strong checksum.
44  *	For user data, we let users make the trade-off between speed
45  *	and checksum strength.
46  *
47  *   2. Cryptographic hash and MAC algorithms are an area of active research.
48  *	It is likely that in future hash functions will be at least as strong
49  *	as current best-of-breed, and may be substantially faster as well.
50  *	We want the ability to take advantage of these new hashes as soon as
51  *	they become available.
52  *
53  *   3. If someone develops hardware that can compute a strong hash quickly,
54  *	we want the ability to take advantage of that hardware.
55  *
56  * Of course, we don't want a checksum upgrade to invalidate existing
57  * data, so we store the checksum *function* in eight bits of the bp.
58  * This gives us room for up to 256 different checksum functions.
59  *
60  * When writing a block, we always checksum it with the latest-and-greatest
61  * checksum function of the appropriate strength.  When reading a block,
62  * we compare the expected checksum against the actual checksum, which we
63  * compute via the checksum function specified by BP_GET_CHECKSUM(bp).
64  *
65  * SALTED CHECKSUMS
66  *
67  * To enable the use of less secure hash algorithms with dedup, we
68  * introduce the notion of salted checksums (MACs, really).  A salted
69  * checksum is fed both a random 256-bit value (the salt) and the data
70  * to be checksummed.  This salt is kept secret (stored on the pool, but
71  * never shown to the user).  Thus even if an attacker knew of collision
72  * weaknesses in the hash algorithm, they won't be able to mount a known
73  * plaintext attack on the DDT, since the actual hash value cannot be
74  * known ahead of time.  How the salt is used is algorithm-specific
75  * (some might simply prefix it to the data block, others might need to
76  * utilize a full-blown HMAC).  On disk the salt is stored in a ZAP
77  * object in the MOS (DMU_POOL_CHECKSUM_SALT).
78  *
79  * CONTEXT TEMPLATES
80  *
81  * Some hashing algorithms need to perform a substantial amount of
82  * initialization work (e.g. salted checksums above may need to pre-hash
83  * the salt) before being able to process data.  Performing this
84  * redundant work for each block would be wasteful, so we instead allow
85  * a checksum algorithm to do the work once (the first time it's used)
86  * and then keep this pre-initialized context as a template inside the
87  * spa_t (spa_cksum_tmpls).  If the zio_checksum_info_t contains
88  * non-NULL ci_tmpl_init and ci_tmpl_free callbacks, they are used to
89  * construct and destruct the pre-initialized checksum context.  The
90  * pre-initialized context is then reused during each checksum
91  * invocation and passed to the checksum function.
92  */
93 
94 /*ARGSUSED*/
95 static void
96 zio_checksum_off(const void *buf, uint64_t size,
97     const void *ctx_template, zio_cksum_t *zcp)
98 {
99 	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
100 }
101 
102 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
103 	{{NULL, NULL}, NULL, NULL, 0, "inherit"},
104 	{{NULL, NULL}, NULL, NULL, 0, "on"},
105 	{{zio_checksum_off,		zio_checksum_off},
106 	    NULL, NULL, 0, "off"},
107 	{{zio_checksum_SHA256,		zio_checksum_SHA256},
108 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
109 	    "label"},
110 	{{zio_checksum_SHA256,		zio_checksum_SHA256},
111 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
112 	    "gang_header"},
113 	{{fletcher_2_native,		fletcher_2_byteswap},
114 	    NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog"},
115 	{{fletcher_2_native,		fletcher_2_byteswap},
116 	    NULL, NULL, 0, "fletcher2"},
117 	{{fletcher_4_native,		fletcher_4_byteswap},
118 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA, "fletcher4"},
119 	{{zio_checksum_SHA256,		zio_checksum_SHA256},
120 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
121 	    ZCHECKSUM_FLAG_NOPWRITE, "sha256"},
122 	{{fletcher_4_native,		fletcher_4_byteswap},
123 	    NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog2"},
124 	{{zio_checksum_off,		zio_checksum_off},
125 	    NULL, NULL, 0, "noparity"},
126 	{{zio_checksum_SHA512_native,	zio_checksum_SHA512_byteswap},
127 	    NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
128 	    ZCHECKSUM_FLAG_NOPWRITE, "sha512"},
129 	{{zio_checksum_skein_native,	zio_checksum_skein_byteswap},
130 	    zio_checksum_skein_tmpl_init, zio_checksum_skein_tmpl_free,
131 	    ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
132 	    ZCHECKSUM_FLAG_SALTED | ZCHECKSUM_FLAG_NOPWRITE, "skein"},
133 	{{zio_checksum_edonr_native,	zio_checksum_edonr_byteswap},
134 	    zio_checksum_edonr_tmpl_init, zio_checksum_edonr_tmpl_free,
135 	    ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_SALTED |
136 	    ZCHECKSUM_FLAG_NOPWRITE, "edonr"},
137 };
138 
139 spa_feature_t
140 zio_checksum_to_feature(enum zio_checksum cksum)
141 {
142 	switch (cksum) {
143 	case ZIO_CHECKSUM_SHA512:
144 		return (SPA_FEATURE_SHA512);
145 	case ZIO_CHECKSUM_SKEIN:
146 		return (SPA_FEATURE_SKEIN);
147 	case ZIO_CHECKSUM_EDONR:
148 		return (SPA_FEATURE_EDONR);
149 	}
150 	return (SPA_FEATURE_NONE);
151 }
152 
153 enum zio_checksum
154 zio_checksum_select(enum zio_checksum child, enum zio_checksum parent)
155 {
156 	ASSERT(child < ZIO_CHECKSUM_FUNCTIONS);
157 	ASSERT(parent < ZIO_CHECKSUM_FUNCTIONS);
158 	ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
159 
160 	if (child == ZIO_CHECKSUM_INHERIT)
161 		return (parent);
162 
163 	if (child == ZIO_CHECKSUM_ON)
164 		return (ZIO_CHECKSUM_ON_VALUE);
165 
166 	return (child);
167 }
168 
169 enum zio_checksum
170 zio_checksum_dedup_select(spa_t *spa, enum zio_checksum child,
171     enum zio_checksum parent)
172 {
173 	ASSERT((child & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
174 	ASSERT((parent & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
175 	ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
176 
177 	if (child == ZIO_CHECKSUM_INHERIT)
178 		return (parent);
179 
180 	if (child == ZIO_CHECKSUM_ON)
181 		return (spa_dedup_checksum(spa));
182 
183 	if (child == (ZIO_CHECKSUM_ON | ZIO_CHECKSUM_VERIFY))
184 		return (spa_dedup_checksum(spa) | ZIO_CHECKSUM_VERIFY);
185 
186 	ASSERT((zio_checksum_table[child & ZIO_CHECKSUM_MASK].ci_flags &
187 	    ZCHECKSUM_FLAG_DEDUP) ||
188 	    (child & ZIO_CHECKSUM_VERIFY) || child == ZIO_CHECKSUM_OFF);
189 
190 	return (child);
191 }
192 
193 /*
194  * Set the external verifier for a gang block based on <vdev, offset, txg>,
195  * a tuple which is guaranteed to be unique for the life of the pool.
196  */
197 static void
198 zio_checksum_gang_verifier(zio_cksum_t *zcp, blkptr_t *bp)
199 {
200 	dva_t *dva = BP_IDENTITY(bp);
201 	uint64_t txg = BP_PHYSICAL_BIRTH(bp);
202 
203 	ASSERT(BP_IS_GANG(bp));
204 
205 	ZIO_SET_CHECKSUM(zcp, DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), txg, 0);
206 }
207 
208 /*
209  * Set the external verifier for a label block based on its offset.
210  * The vdev is implicit, and the txg is unknowable at pool open time --
211  * hence the logic in vdev_uberblock_load() to find the most recent copy.
212  */
213 static void
214 zio_checksum_label_verifier(zio_cksum_t *zcp, uint64_t offset)
215 {
216 	ZIO_SET_CHECKSUM(zcp, offset, 0, 0, 0);
217 }
218 
219 /*
220  * Calls the template init function of a checksum which supports context
221  * templates and installs the template into the spa_t.
222  */
223 static void
224 zio_checksum_template_init(enum zio_checksum checksum, spa_t *spa)
225 {
226 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
227 
228 	if (ci->ci_tmpl_init == NULL)
229 		return;
230 	if (spa->spa_cksum_tmpls[checksum] != NULL)
231 		return;
232 
233 	VERIFY(ci->ci_tmpl_free != NULL);
234 	mutex_enter(&spa->spa_cksum_tmpls_lock);
235 	if (spa->spa_cksum_tmpls[checksum] == NULL) {
236 		spa->spa_cksum_tmpls[checksum] =
237 		    ci->ci_tmpl_init(&spa->spa_cksum_salt);
238 		VERIFY(spa->spa_cksum_tmpls[checksum] != NULL);
239 	}
240 	mutex_exit(&spa->spa_cksum_tmpls_lock);
241 }
242 
243 /*
244  * Generate the checksum.
245  */
246 void
247 zio_checksum_compute(zio_t *zio, enum zio_checksum checksum,
248     void *data, uint64_t size)
249 {
250 	blkptr_t *bp = zio->io_bp;
251 	uint64_t offset = zio->io_offset;
252 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
253 	zio_cksum_t cksum;
254 	spa_t *spa = zio->io_spa;
255 
256 	ASSERT((uint_t)checksum < ZIO_CHECKSUM_FUNCTIONS);
257 	ASSERT(ci->ci_func[0] != NULL);
258 
259 	zio_checksum_template_init(checksum, spa);
260 
261 	if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
262 		zio_eck_t *eck;
263 
264 		if (checksum == ZIO_CHECKSUM_ZILOG2) {
265 			zil_chain_t *zilc = data;
266 
267 			size = P2ROUNDUP_TYPED(zilc->zc_nused, ZIL_MIN_BLKSZ,
268 			    uint64_t);
269 			eck = &zilc->zc_eck;
270 		} else {
271 			eck = (zio_eck_t *)((char *)data + size) - 1;
272 		}
273 		if (checksum == ZIO_CHECKSUM_GANG_HEADER)
274 			zio_checksum_gang_verifier(&eck->zec_cksum, bp);
275 		else if (checksum == ZIO_CHECKSUM_LABEL)
276 			zio_checksum_label_verifier(&eck->zec_cksum, offset);
277 		else
278 			bp->blk_cksum = eck->zec_cksum;
279 		eck->zec_magic = ZEC_MAGIC;
280 		ci->ci_func[0](data, size, spa->spa_cksum_tmpls[checksum],
281 		    &cksum);
282 		eck->zec_cksum = cksum;
283 	} else {
284 		ci->ci_func[0](data, size, spa->spa_cksum_tmpls[checksum],
285 		    &bp->blk_cksum);
286 	}
287 }
288 
289 int
290 zio_checksum_error(zio_t *zio, zio_bad_cksum_t *info)
291 {
292 	blkptr_t *bp = zio->io_bp;
293 	uint_t checksum = (bp == NULL ? zio->io_prop.zp_checksum :
294 	    (BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : BP_GET_CHECKSUM(bp)));
295 	int byteswap;
296 	int error;
297 	uint64_t size = (bp == NULL ? zio->io_size :
298 	    (BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(bp)));
299 	uint64_t offset = zio->io_offset;
300 	void *data = zio->io_data;
301 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
302 	zio_cksum_t actual_cksum, expected_cksum, verifier;
303 	spa_t *spa = zio->io_spa;
304 
305 	if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL)
306 		return (SET_ERROR(EINVAL));
307 
308 	zio_checksum_template_init(checksum, spa);
309 
310 	if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
311 		zio_eck_t *eck;
312 
313 		if (checksum == ZIO_CHECKSUM_ZILOG2) {
314 			zil_chain_t *zilc = data;
315 			uint64_t nused;
316 
317 			eck = &zilc->zc_eck;
318 			if (eck->zec_magic == ZEC_MAGIC)
319 				nused = zilc->zc_nused;
320 			else if (eck->zec_magic == BSWAP_64(ZEC_MAGIC))
321 				nused = BSWAP_64(zilc->zc_nused);
322 			else
323 				return (SET_ERROR(ECKSUM));
324 
325 			if (nused > size)
326 				return (SET_ERROR(ECKSUM));
327 
328 			size = P2ROUNDUP_TYPED(nused, ZIL_MIN_BLKSZ, uint64_t);
329 		} else {
330 			eck = (zio_eck_t *)((char *)data + size) - 1;
331 		}
332 
333 		if (checksum == ZIO_CHECKSUM_GANG_HEADER)
334 			zio_checksum_gang_verifier(&verifier, bp);
335 		else if (checksum == ZIO_CHECKSUM_LABEL)
336 			zio_checksum_label_verifier(&verifier, offset);
337 		else
338 			verifier = bp->blk_cksum;
339 
340 		byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
341 
342 		if (byteswap)
343 			byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
344 
345 		expected_cksum = eck->zec_cksum;
346 		eck->zec_cksum = verifier;
347 		ci->ci_func[byteswap](data, size,
348 		    spa->spa_cksum_tmpls[checksum], &actual_cksum);
349 		eck->zec_cksum = expected_cksum;
350 
351 		if (byteswap)
352 			byteswap_uint64_array(&expected_cksum,
353 			    sizeof (zio_cksum_t));
354 	} else {
355 		ASSERT(!BP_IS_GANG(bp));
356 		byteswap = BP_SHOULD_BYTESWAP(bp);
357 		expected_cksum = bp->blk_cksum;
358 		ci->ci_func[byteswap](data, size,
359 		    spa->spa_cksum_tmpls[checksum], &actual_cksum);
360 	}
361 
362 	info->zbc_expected = expected_cksum;
363 	info->zbc_actual = actual_cksum;
364 	info->zbc_checksum_name = ci->ci_name;
365 	info->zbc_byteswapped = byteswap;
366 	info->zbc_injected = 0;
367 	info->zbc_has_cksum = 1;
368 
369 	if (!ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
370 		return (SET_ERROR(ECKSUM));
371 
372 	if (zio_injection_enabled && !zio->io_error &&
373 	    (error = zio_handle_fault_injection(zio, ECKSUM)) != 0) {
374 
375 		info->zbc_injected = 1;
376 		return (error);
377 	}
378 
379 	return (0);
380 }
381 
382 /*
383  * Called by a spa_t that's about to be deallocated. This steps through
384  * all of the checksum context templates and deallocates any that were
385  * initialized using the algorithm-specific template init function.
386  */
387 void
388 zio_checksum_templates_free(spa_t *spa)
389 {
390 	for (enum zio_checksum checksum = 0;
391 	    checksum < ZIO_CHECKSUM_FUNCTIONS; checksum++) {
392 		if (spa->spa_cksum_tmpls[checksum] != NULL) {
393 			zio_checksum_info_t *ci = &zio_checksum_table[checksum];
394 
395 			VERIFY(ci->ci_tmpl_free != NULL);
396 			ci->ci_tmpl_free(spa->spa_cksum_tmpls[checksum]);
397 			spa->spa_cksum_tmpls[checksum] = NULL;
398 		}
399 	}
400 }
401