2 * Copyright (c) 2005-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/linker.h>
34 #include <sys/module.h>
36 #include <sys/mutex.h>
38 #include <sys/sysctl.h>
39 #include <sys/malloc.h>
40 #include <sys/kthread.h>
42 #include <sys/sched.h>
45 #include <sys/vnode.h>
49 #include <geom/geom.h>
50 #include <geom/eli/g_eli.h>
51 #include <geom/eli/pkcs5v2.h>
54 * The data layout description when integrity verification is configured.
56 * One of the most important assumption here is that authenticated data and its
57 * HMAC has to be stored in the same place (namely in the same sector) to make
59 * The problem is that file systems work only with sectors that are multiple of
60 * 512 bytes and a power of two number.
61 * My idea to implement it is as follows.
62 * Let's store HMAC in sector. This is a must. This leaves us 480 bytes for
63 * data. We can't use that directly (ie. we can't create provider with 480 bytes
64 * sector size). We need another sector from where we take only 32 bytes of data
65 * and we store HMAC of this data as well. This takes two sectors from the
66 * original provider at the input and leaves us one sector of authenticated data
67 * at the output. Not very efficient, but you got the idea.
68 * Now, let's assume, we want to create provider with 4096 bytes sector.
69 * To output 4096 bytes of authenticated data we need 8x480 plus 1x256, so we
70 * need nine 512-bytes sectors at the input to get one 4096-bytes sector at the
71 * output. That's better. With 4096 bytes sector we can use 89% of size of the
72 * original provider. I find it as an acceptable cost.
73 * The reliability comes from the fact, that every HMAC stored inside the sector
74 * is calculated only for the data in the same sector, so its impossible to
75 * write new data and leave old HMAC or vice versa.
77 * And here is the picture:
79 * da0: +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+-----+
80 * |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |480b| |32b |256b |
81 * |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data| |HMAC|Data |
82 * +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+----+ +----+-----+
83 * |512 bytes| |512 bytes| |512 bytes| |512 bytes| |512 bytes| |512 bytes| |512 bytes| |512 bytes| |288 bytes |
84 * +---------+ +---------+ +---------+ +---------+ +---------+ +---------+ +---------+ +---------+ |224 unused|
86 * da0.eli: +----+----+----+----+----+----+----+----+----+
87 * |480b|480b|480b|480b|480b|480b|480b|480b|256b|
88 * +----+----+----+----+----+----+----+----+----+
90 * +--------------------------------------------+
92 * PS. You can use any sector size with geli(8). My example is using 4kB,
93 * because it's most efficient. For 8kB sectors you need 2 extra sectors,
94 * so the cost is the same as for 4kB sectors.
100 * g_eli_start -> g_eli_auth_read -> g_io_request -> g_eli_read_done -> g_eli_auth_run -> g_eli_auth_read_done -> g_io_deliver
102 * g_eli_start -> g_eli_auth_run -> g_eli_auth_write_done -> g_io_request -> g_eli_write_done -> g_io_deliver
105 MALLOC_DECLARE(M_ELI);
108 * Here we generate key for HMAC. Every sector has its own HMAC key, so it is
109 * not possible to copy sectors.
110 * We cannot depend on fact, that every sector has its own IV, because different
111 * IV doesn't change HMAC, when we use encrypt-then-authenticate method.
114 g_eli_auth_keygen(struct g_eli_softc *sc, off_t offset, u_char *key)
118 /* Copy precalculated SHA256 context. */
119 bcopy(&sc->sc_akeyctx, &ctx, sizeof(ctx));
120 SHA256_Update(&ctx, (uint8_t *)&offset, sizeof(offset));
121 SHA256_Final(key, &ctx);
125 * The function is called after we read and decrypt data.
127 * g_eli_start -> g_eli_auth_read -> g_io_request -> g_eli_read_done -> g_eli_auth_run -> G_ELI_AUTH_READ_DONE -> g_io_deliver
130 g_eli_auth_read_done(struct cryptop *crp)
132 struct g_eli_softc *sc;
135 if (crp->crp_etype == EAGAIN) {
136 if (g_eli_crypto_rerun(crp) == 0)
139 bp = (struct bio *)crp->crp_opaque;
141 if (crp->crp_etype == 0) {
142 bp->bio_completed += crp->crp_olen;
143 G_ELI_DEBUG(3, "Crypto READ request done (%d/%d) (add=%jd completed=%jd).",
144 bp->bio_inbed, bp->bio_children, (intmax_t)crp->crp_olen, (intmax_t)bp->bio_completed);
146 G_ELI_DEBUG(1, "Crypto READ request failed (%d/%d) error=%d.",
147 bp->bio_inbed, bp->bio_children, crp->crp_etype);
148 if (bp->bio_error == 0)
149 bp->bio_error = crp->crp_etype;
152 * Do we have all sectors already?
154 if (bp->bio_inbed < bp->bio_children)
156 sc = bp->bio_to->geom->softc;
157 if (bp->bio_error == 0) {
158 u_int i, lsec, nsec, data_secsize, decr_secsize, encr_secsize;
159 u_char *srcdata, *dstdata, *auth;
160 off_t coroff, corsize;
163 * Verify data integrity based on calculated and read HMACs.
165 /* Sectorsize of decrypted provider eg. 4096. */
166 decr_secsize = bp->bio_to->sectorsize;
167 /* The real sectorsize of encrypted provider, eg. 512. */
168 encr_secsize = LIST_FIRST(&sc->sc_geom->consumer)->provider->sectorsize;
169 /* Number of data bytes in one encrypted sector, eg. 480. */
170 data_secsize = sc->sc_data_per_sector;
171 /* Number of sectors from decrypted provider, eg. 2. */
172 nsec = bp->bio_length / decr_secsize;
173 /* Number of sectors from encrypted provider, eg. 18. */
174 nsec = (nsec * sc->sc_bytes_per_sector) / encr_secsize;
175 /* Last sector number in every big sector, eg. 9. */
176 lsec = sc->sc_bytes_per_sector / encr_secsize;
178 srcdata = bp->bio_driver2;
179 dstdata = bp->bio_data;
180 auth = srcdata + encr_secsize * nsec;
184 for (i = 1; i <= nsec; i++) {
185 data_secsize = sc->sc_data_per_sector;
187 data_secsize = decr_secsize % data_secsize;
188 if (bcmp(srcdata, auth, sc->sc_alen) != 0) {
190 * Curruption detected, remember the offset if
191 * this is the first corrupted sector and
194 if (bp->bio_error == 0)
197 coroff = bp->bio_offset +
198 (dstdata - (u_char *)bp->bio_data);
200 corsize += data_secsize;
203 * No curruption, good.
204 * Report previous corruption if there was one.
207 G_ELI_DEBUG(0, "%s: Failed to authenticate %jd "
208 "bytes of data at offset %jd.",
209 sc->sc_name, (intmax_t)corsize,
214 bcopy(srcdata + sc->sc_alen, dstdata,
217 srcdata += encr_secsize;
218 dstdata += data_secsize;
221 /* Report previous corruption if there was one. */
223 G_ELI_DEBUG(0, "%s: Failed to authenticate %jd "
224 "bytes of data at offset %jd.",
225 sc->sc_name, (intmax_t)corsize, (intmax_t)coroff);
228 free(bp->bio_driver2, M_ELI);
229 bp->bio_driver2 = NULL;
230 if (bp->bio_error != 0) {
231 if (bp->bio_error == -1)
232 bp->bio_error = EINVAL;
235 "Crypto READ request failed (error=%d).",
238 bp->bio_completed = 0;
241 * Read is finished, send it up.
243 g_io_deliver(bp, bp->bio_error);
244 atomic_subtract_int(&sc->sc_inflight, 1);
249 * The function is called after data encryption.
251 * g_eli_start -> g_eli_auth_run -> G_ELI_AUTH_WRITE_DONE -> g_io_request -> g_eli_write_done -> g_io_deliver
254 g_eli_auth_write_done(struct cryptop *crp)
256 struct g_eli_softc *sc;
257 struct g_consumer *cp;
258 struct bio *bp, *cbp, *cbp2;
261 if (crp->crp_etype == EAGAIN) {
262 if (g_eli_crypto_rerun(crp) == 0)
265 bp = (struct bio *)crp->crp_opaque;
267 if (crp->crp_etype == 0) {
268 G_ELI_DEBUG(3, "Crypto WRITE request done (%d/%d).",
269 bp->bio_inbed, bp->bio_children);
271 G_ELI_DEBUG(1, "Crypto WRITE request failed (%d/%d) error=%d.",
272 bp->bio_inbed, bp->bio_children, crp->crp_etype);
273 if (bp->bio_error == 0)
274 bp->bio_error = crp->crp_etype;
277 * All sectors are already encrypted?
279 if (bp->bio_inbed < bp->bio_children)
281 sc = bp->bio_to->geom->softc;
282 if (bp->bio_error != 0) {
283 G_ELI_LOGREQ(0, bp, "Crypto WRITE request failed (error=%d).",
285 free(bp->bio_driver2, M_ELI);
286 bp->bio_driver2 = NULL;
287 cbp = bp->bio_driver1;
288 bp->bio_driver1 = NULL;
290 g_io_deliver(bp, bp->bio_error);
291 atomic_subtract_int(&sc->sc_inflight, 1);
294 cp = LIST_FIRST(&sc->sc_geom->consumer);
295 cbp = bp->bio_driver1;
296 bp->bio_driver1 = NULL;
297 cbp->bio_to = cp->provider;
298 cbp->bio_done = g_eli_write_done;
300 /* Number of sectors from decrypted provider, eg. 1. */
301 nsec = bp->bio_length / bp->bio_to->sectorsize;
302 /* Number of sectors from encrypted provider, eg. 9. */
303 nsec = (nsec * sc->sc_bytes_per_sector) / cp->provider->sectorsize;
305 cbp->bio_length = cp->provider->sectorsize * nsec;
306 cbp->bio_offset = (bp->bio_offset / bp->bio_to->sectorsize) * sc->sc_bytes_per_sector;
307 cbp->bio_data = bp->bio_driver2;
310 * We write more than what is requested, so we have to be ready to write
314 if (cbp->bio_length > MAXPHYS) {
315 cbp2 = g_duplicate_bio(bp);
316 cbp2->bio_length = cbp->bio_length - MAXPHYS;
317 cbp2->bio_data = cbp->bio_data + MAXPHYS;
318 cbp2->bio_offset = cbp->bio_offset + MAXPHYS;
319 cbp2->bio_to = cp->provider;
320 cbp2->bio_done = g_eli_write_done;
321 cbp->bio_length = MAXPHYS;
324 * Send encrypted data to the provider.
326 G_ELI_LOGREQ(2, cbp, "Sending request.");
328 bp->bio_children = (cbp2 != NULL ? 2 : 1);
329 g_io_request(cbp, cp);
331 G_ELI_LOGREQ(2, cbp2, "Sending request.");
332 g_io_request(cbp2, cp);
338 g_eli_auth_read(struct g_eli_softc *sc, struct bio *bp)
340 struct g_consumer *cp;
341 struct bio *cbp, *cbp2;
347 cp = LIST_FIRST(&sc->sc_geom->consumer);
348 cbp = bp->bio_driver1;
349 bp->bio_driver1 = NULL;
350 cbp->bio_to = cp->provider;
351 cbp->bio_done = g_eli_read_done;
353 /* Number of sectors from decrypted provider, eg. 1. */
354 nsec = bp->bio_length / bp->bio_to->sectorsize;
355 /* Number of sectors from encrypted provider, eg. 9. */
356 nsec = (nsec * sc->sc_bytes_per_sector) / cp->provider->sectorsize;
358 cbp->bio_length = cp->provider->sectorsize * nsec;
359 size = cbp->bio_length;
360 size += sc->sc_alen * nsec;
361 size += sizeof(struct cryptop) * nsec;
362 size += sizeof(struct cryptodesc) * nsec * 2;
363 size += G_ELI_AUTH_SECKEYLEN * nsec;
364 size += sizeof(struct uio) * nsec;
365 size += sizeof(struct iovec) * nsec;
366 cbp->bio_offset = (bp->bio_offset / bp->bio_to->sectorsize) * sc->sc_bytes_per_sector;
367 bp->bio_driver2 = malloc(size, M_ELI, M_WAITOK);
368 cbp->bio_data = bp->bio_driver2;
371 * We read more than what is requested, so we have to be ready to read
375 if (cbp->bio_length > MAXPHYS) {
376 cbp2 = g_duplicate_bio(bp);
377 cbp2->bio_length = cbp->bio_length - MAXPHYS;
378 cbp2->bio_data = cbp->bio_data + MAXPHYS;
379 cbp2->bio_offset = cbp->bio_offset + MAXPHYS;
380 cbp2->bio_to = cp->provider;
381 cbp2->bio_done = g_eli_read_done;
382 cbp->bio_length = MAXPHYS;
385 * Read encrypted data from provider.
387 G_ELI_LOGREQ(2, cbp, "Sending request.");
388 g_io_request(cbp, cp);
390 G_ELI_LOGREQ(2, cbp2, "Sending request.");
391 g_io_request(cbp2, cp);
396 * This is the main function responsible for cryptography (ie. communication
397 * with crypto(9) subsystem).
400 * g_eli_start -> g_eli_auth_read -> g_io_request -> g_eli_read_done -> G_ELI_AUTH_RUN -> g_eli_auth_read_done -> g_io_deliver
402 * g_eli_start -> G_ELI_AUTH_RUN -> g_eli_auth_write_done -> g_io_request -> g_eli_write_done -> g_io_deliver
405 g_eli_auth_run(struct g_eli_worker *wr, struct bio *bp)
407 struct g_eli_softc *sc;
409 struct cryptodesc *crde, *crda;
412 u_int i, lsec, nsec, data_secsize, decr_secsize, encr_secsize;
415 u_char *p, *data, *auth, *authkey, *plaindata;
417 G_ELI_LOGREQ(3, bp, "%s", __func__);
419 bp->bio_pflags = wr->w_number;
421 /* Sectorsize of decrypted provider eg. 4096. */
422 decr_secsize = bp->bio_to->sectorsize;
423 /* The real sectorsize of encrypted provider, eg. 512. */
424 encr_secsize = LIST_FIRST(&sc->sc_geom->consumer)->provider->sectorsize;
425 /* Number of data bytes in one encrypted sector, eg. 480. */
426 data_secsize = sc->sc_data_per_sector;
427 /* Number of sectors from decrypted provider, eg. 2. */
428 nsec = bp->bio_length / decr_secsize;
429 /* Number of sectors from encrypted provider, eg. 18. */
430 nsec = (nsec * sc->sc_bytes_per_sector) / encr_secsize;
431 /* Last sector number in every big sector, eg. 9. */
432 lsec = sc->sc_bytes_per_sector / encr_secsize;
433 /* Destination offset, used for IV generation. */
434 dstoff = (bp->bio_offset / bp->bio_to->sectorsize) * sc->sc_bytes_per_sector;
436 auth = NULL; /* Silence compiler warning. */
437 plaindata = bp->bio_data;
438 if (bp->bio_cmd == BIO_READ) {
439 data = bp->bio_driver2;
440 auth = data + encr_secsize * nsec;
441 p = auth + sc->sc_alen * nsec;
445 size = encr_secsize * nsec;
446 size += sizeof(*crp) * nsec;
447 size += sizeof(*crde) * nsec;
448 size += sizeof(*crda) * nsec;
449 size += G_ELI_AUTH_SECKEYLEN * nsec;
450 size += sizeof(*uio) * nsec;
451 size += sizeof(*iov) * nsec;
452 data = malloc(size, M_ELI, M_WAITOK);
453 bp->bio_driver2 = data;
454 p = data + encr_secsize * nsec;
457 bp->bio_children = nsec;
460 for (i = 1; i <= nsec; i++, dstoff += encr_secsize) {
461 crp = (struct cryptop *)p; p += sizeof(*crp);
462 crde = (struct cryptodesc *)p; p += sizeof(*crde);
463 crda = (struct cryptodesc *)p; p += sizeof(*crda);
464 authkey = (u_char *)p; p += G_ELI_AUTH_SECKEYLEN;
465 uio = (struct uio *)p; p += sizeof(*uio);
466 iov = (struct iovec *)p; p += sizeof(*iov);
468 data_secsize = sc->sc_data_per_sector;
470 data_secsize = decr_secsize % data_secsize;
472 if (bp->bio_cmd == BIO_READ) {
473 /* Remember read HMAC. */
474 bcopy(data, auth, sc->sc_alen);
476 /* TODO: bzero(9) can be commented out later. */
477 bzero(data, sc->sc_alen);
479 bcopy(plaindata, data + sc->sc_alen, data_secsize);
480 plaindata += data_secsize;
483 iov->iov_len = sc->sc_alen + data_secsize;
484 iov->iov_base = data;
485 data += encr_secsize;
489 uio->uio_segflg = UIO_SYSSPACE;
490 uio->uio_resid = iov->iov_len;
492 crp->crp_sid = wr->w_sid;
493 crp->crp_ilen = uio->uio_resid;
494 crp->crp_olen = data_secsize;
495 crp->crp_opaque = (void *)bp;
496 crp->crp_buf = (void *)uio;
497 crp->crp_flags = CRYPTO_F_IOV | CRYPTO_F_CBIFSYNC | CRYPTO_F_REL;
499 crp->crp_flags |= CRYPTO_F_BATCH;
500 if (bp->bio_cmd == BIO_WRITE) {
501 crp->crp_callback = g_eli_auth_write_done;
502 crp->crp_desc = crde;
503 crde->crd_next = crda;
504 crda->crd_next = NULL;
506 crp->crp_callback = g_eli_auth_read_done;
507 crp->crp_desc = crda;
508 crda->crd_next = crde;
509 crde->crd_next = NULL;
512 crde->crd_skip = sc->sc_alen;
513 crde->crd_len = data_secsize;
514 crde->crd_flags = CRD_F_IV_EXPLICIT | CRD_F_IV_PRESENT;
515 if (bp->bio_cmd == BIO_WRITE)
516 crde->crd_flags |= CRD_F_ENCRYPT;
517 crde->crd_alg = sc->sc_ealgo;
518 crde->crd_key = g_eli_crypto_key(sc, dstoff, encr_secsize);
519 crde->crd_klen = sc->sc_ekeylen;
520 if (sc->sc_ealgo == CRYPTO_AES_XTS)
521 crde->crd_klen <<= 1;
522 g_eli_crypto_ivgen(sc, dstoff, crde->crd_iv,
523 sizeof(crde->crd_iv));
525 crda->crd_skip = sc->sc_alen;
526 crda->crd_len = data_secsize;
527 crda->crd_inject = 0;
528 crda->crd_flags = CRD_F_KEY_EXPLICIT;
529 crda->crd_alg = sc->sc_aalgo;
530 g_eli_auth_keygen(sc, dstoff, authkey);
531 crda->crd_key = authkey;
532 crda->crd_klen = G_ELI_AUTH_SECKEYLEN * 8;
535 err = crypto_dispatch(crp);
536 if (err != 0 && error == 0)
539 if (bp->bio_error == 0)
540 bp->bio_error = error;