29
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>
35#include <sys/lock.h>
36#include <sys/mutex.h>
37#include <sys/bio.h>
38#include <sys/sysctl.h>
39#include <sys/malloc.h>
40#include <sys/kthread.h>
41#include <sys/proc.h>
42#include <sys/sched.h>
43#include <sys/smp.h>
44#include <sys/uio.h>
45#include <sys/vnode.h>
46
47#include <vm/uma.h>
48
49#include <geom/geom.h>
50#include <geom/eli/g_eli.h>
51#include <geom/eli/pkcs5v2.h>
52
53/*
54 * The data layout description when integrity verification is configured.
55 *
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
58 * it work reliable.
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.
76 *
77 * And here is the picture:
78 *
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|
85 * +----------+
86 * da0.eli: +----+----+----+----+----+----+----+----+----+
87 * |480b|480b|480b|480b|480b|480b|480b|480b|256b|
88 * +----+----+----+----+----+----+----+----+----+
89 * | 4096 bytes |
90 * +--------------------------------------------+
91 *
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.
95 */
96
97/*
98 * Code paths:
99 * BIO_READ:
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
101 * BIO_WRITE:
102 * g_eli_start -> g_eli_auth_run -> g_eli_auth_write_done -> g_io_request -> g_eli_write_done -> g_io_deliver
103 */
104
105MALLOC_DECLARE(M_ELI);
106
107/*
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.
112 */
113static void
114g_eli_auth_keygen(struct g_eli_softc *sc, off_t offset, u_char *key)
115{
116 SHA256_CTX ctx;
117
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);
122}
123
124/*
125 * The function is called after we read and decrypt data.
126 *
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
128 */
129static int
130g_eli_auth_read_done(struct cryptop *crp)
131{
132 struct g_eli_softc *sc;
133 struct bio *bp;
134
135 if (crp->crp_etype == EAGAIN) {
136 if (g_eli_crypto_rerun(crp) == 0)
137 return (0);
138 }
139 bp = (struct bio *)crp->crp_opaque;
140 bp->bio_inbed++;
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);
145 } else {
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;
150 }
151 sc = bp->bio_to->geom->softc;
152 g_eli_key_drop(sc, crp->crp_desc->crd_next->crd_key);
153 /*
154 * Do we have all sectors already?
155 */
156 if (bp->bio_inbed < bp->bio_children)
157 return (0);
158 if (bp->bio_error == 0) {
159 u_int i, lsec, nsec, data_secsize, decr_secsize, encr_secsize;
160 u_char *srcdata, *dstdata, *auth;
161 off_t coroff, corsize;
162
163 /*
164 * Verify data integrity based on calculated and read HMACs.
165 */
166 /* Sectorsize of decrypted provider eg. 4096. */
167 decr_secsize = bp->bio_to->sectorsize;
168 /* The real sectorsize of encrypted provider, eg. 512. */
169 encr_secsize = LIST_FIRST(&sc->sc_geom->consumer)->provider->sectorsize;
170 /* Number of data bytes in one encrypted sector, eg. 480. */
171 data_secsize = sc->sc_data_per_sector;
172 /* Number of sectors from decrypted provider, eg. 2. */
173 nsec = bp->bio_length / decr_secsize;
174 /* Number of sectors from encrypted provider, eg. 18. */
175 nsec = (nsec * sc->sc_bytes_per_sector) / encr_secsize;
176 /* Last sector number in every big sector, eg. 9. */
177 lsec = sc->sc_bytes_per_sector / encr_secsize;
178
179 srcdata = bp->bio_driver2;
180 dstdata = bp->bio_data;
181 auth = srcdata + encr_secsize * nsec;
182 coroff = -1;
183 corsize = 0;
184
185 for (i = 1; i <= nsec; i++) {
186 data_secsize = sc->sc_data_per_sector;
187 if ((i % lsec) == 0)
188 data_secsize = decr_secsize % data_secsize;
189 if (bcmp(srcdata, auth, sc->sc_alen) != 0) {
190 /*
191 * Curruption detected, remember the offset if
192 * this is the first corrupted sector and
193 * increase size.
194 */
195 if (bp->bio_error == 0)
196 bp->bio_error = -1;
197 if (coroff == -1) {
198 coroff = bp->bio_offset +
199 (dstdata - (u_char *)bp->bio_data);
200 }
201 corsize += data_secsize;
202 } else {
203 /*
204 * No curruption, good.
205 * Report previous corruption if there was one.
206 */
207 if (coroff != -1) {
208 G_ELI_DEBUG(0, "%s: %jd bytes "
209 "corrupted at offset %jd.",
210 sc->sc_name, (intmax_t)corsize,
211 (intmax_t)coroff);
212 coroff = -1;
213 corsize = 0;
214 }
215 bcopy(srcdata + sc->sc_alen, dstdata,
216 data_secsize);
217 }
218 srcdata += encr_secsize;
219 dstdata += data_secsize;
220 auth += sc->sc_alen;
221 }
222 /* Report previous corruption if there was one. */
223 if (coroff != -1) {
224 G_ELI_DEBUG(0, "%s: %jd bytes corrupted at offset %jd.",
225 sc->sc_name, (intmax_t)corsize, (intmax_t)coroff);
226 }
227 }
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;
233 else {
234 G_ELI_LOGREQ(0, bp,
235 "Crypto READ request failed (error=%d).",
236 bp->bio_error);
237 }
238 bp->bio_completed = 0;
239 }
240 /*
241 * Read is finished, send it up.
242 */
243 g_io_deliver(bp, bp->bio_error);
244 atomic_subtract_int(&sc->sc_inflight, 1);
245 return (0);
246}
247
248/*
249 * The function is called after data encryption.
250 *
251 * g_eli_start -> g_eli_auth_run -> G_ELI_AUTH_WRITE_DONE -> g_io_request -> g_eli_write_done -> g_io_deliver
252 */
253static int
254g_eli_auth_write_done(struct cryptop *crp)
255{
256 struct g_eli_softc *sc;
257 struct g_consumer *cp;
258 struct bio *bp, *cbp, *cbp2;
259 u_int nsec;
260
261 if (crp->crp_etype == EAGAIN) {
262 if (g_eli_crypto_rerun(crp) == 0)
263 return (0);
264 }
265 bp = (struct bio *)crp->crp_opaque;
266 bp->bio_inbed++;
267 if (crp->crp_etype == 0) {
268 G_ELI_DEBUG(3, "Crypto WRITE request done (%d/%d).",
269 bp->bio_inbed, bp->bio_children);
270 } else {
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;
275 }
276 sc = bp->bio_to->geom->softc;
277 g_eli_key_drop(sc, crp->crp_desc->crd_key);
278 /*
279 * All sectors are already encrypted?
280 */
281 if (bp->bio_inbed < bp->bio_children)
282 return (0);
283 if (bp->bio_error != 0) {
284 G_ELI_LOGREQ(0, bp, "Crypto WRITE request failed (error=%d).",
285 bp->bio_error);
286 free(bp->bio_driver2, M_ELI);
287 bp->bio_driver2 = NULL;
288 cbp = bp->bio_driver1;
289 bp->bio_driver1 = NULL;
290 g_destroy_bio(cbp);
291 g_io_deliver(bp, bp->bio_error);
292 atomic_subtract_int(&sc->sc_inflight, 1);
293 return (0);
294 }
295 cp = LIST_FIRST(&sc->sc_geom->consumer);
296 cbp = bp->bio_driver1;
297 bp->bio_driver1 = NULL;
298 cbp->bio_to = cp->provider;
299 cbp->bio_done = g_eli_write_done;
300
301 /* Number of sectors from decrypted provider, eg. 1. */
302 nsec = bp->bio_length / bp->bio_to->sectorsize;
303 /* Number of sectors from encrypted provider, eg. 9. */
304 nsec = (nsec * sc->sc_bytes_per_sector) / cp->provider->sectorsize;
305
306 cbp->bio_length = cp->provider->sectorsize * nsec;
307 cbp->bio_offset = (bp->bio_offset / bp->bio_to->sectorsize) * sc->sc_bytes_per_sector;
308 cbp->bio_data = bp->bio_driver2;
309
310 /*
311 * We write more than what is requested, so we have to be ready to write
312 * more than MAXPHYS.
313 */
314 cbp2 = NULL;
315 if (cbp->bio_length > MAXPHYS) {
316 cbp2 = g_duplicate_bio(bp);
317 cbp2->bio_length = cbp->bio_length - MAXPHYS;
318 cbp2->bio_data = cbp->bio_data + MAXPHYS;
319 cbp2->bio_offset = cbp->bio_offset + MAXPHYS;
320 cbp2->bio_to = cp->provider;
321 cbp2->bio_done = g_eli_write_done;
322 cbp->bio_length = MAXPHYS;
323 }
324 /*
325 * Send encrypted data to the provider.
326 */
327 G_ELI_LOGREQ(2, cbp, "Sending request.");
328 bp->bio_inbed = 0;
329 bp->bio_children = (cbp2 != NULL ? 2 : 1);
330 g_io_request(cbp, cp);
331 if (cbp2 != NULL) {
332 G_ELI_LOGREQ(2, cbp2, "Sending request.");
333 g_io_request(cbp2, cp);
334 }
335 return (0);
336}
337
338void
339g_eli_auth_read(struct g_eli_softc *sc, struct bio *bp)
340{
341 struct g_consumer *cp;
342 struct bio *cbp, *cbp2;
343 size_t size;
344 off_t nsec;
345
346 bp->bio_pflags = 0;
347
348 cp = LIST_FIRST(&sc->sc_geom->consumer);
349 cbp = bp->bio_driver1;
350 bp->bio_driver1 = NULL;
351 cbp->bio_to = cp->provider;
352 cbp->bio_done = g_eli_read_done;
353
354 /* Number of sectors from decrypted provider, eg. 1. */
355 nsec = bp->bio_length / bp->bio_to->sectorsize;
356 /* Number of sectors from encrypted provider, eg. 9. */
357 nsec = (nsec * sc->sc_bytes_per_sector) / cp->provider->sectorsize;
358
359 cbp->bio_length = cp->provider->sectorsize * nsec;
360 size = cbp->bio_length;
361 size += sc->sc_alen * nsec;
362 size += sizeof(struct cryptop) * nsec;
363 size += sizeof(struct cryptodesc) * nsec * 2;
364 size += G_ELI_AUTH_SECKEYLEN * nsec;
365 size += sizeof(struct uio) * nsec;
366 size += sizeof(struct iovec) * nsec;
367 cbp->bio_offset = (bp->bio_offset / bp->bio_to->sectorsize) * sc->sc_bytes_per_sector;
368 bp->bio_driver2 = malloc(size, M_ELI, M_WAITOK);
369 cbp->bio_data = bp->bio_driver2;
370
371 /*
372 * We read more than what is requested, so we have to be ready to read
373 * more than MAXPHYS.
374 */
375 cbp2 = NULL;
376 if (cbp->bio_length > MAXPHYS) {
377 cbp2 = g_duplicate_bio(bp);
378 cbp2->bio_length = cbp->bio_length - MAXPHYS;
379 cbp2->bio_data = cbp->bio_data + MAXPHYS;
380 cbp2->bio_offset = cbp->bio_offset + MAXPHYS;
381 cbp2->bio_to = cp->provider;
382 cbp2->bio_done = g_eli_read_done;
383 cbp->bio_length = MAXPHYS;
384 }
385 /*
386 * Read encrypted data from provider.
387 */
388 G_ELI_LOGREQ(2, cbp, "Sending request.");
389 g_io_request(cbp, cp);
390 if (cbp2 != NULL) {
391 G_ELI_LOGREQ(2, cbp2, "Sending request.");
392 g_io_request(cbp2, cp);
393 }
394}
395
396/*
397 * This is the main function responsible for cryptography (ie. communication
398 * with crypto(9) subsystem).
399 *
400 * BIO_READ:
401 * 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 * BIO_WRITE:
403 * g_eli_start -> G_ELI_AUTH_RUN -> g_eli_auth_write_done -> g_io_request -> g_eli_write_done -> g_io_deliver
404 */
405void
406g_eli_auth_run(struct g_eli_worker *wr, struct bio *bp)
407{
408 struct g_eli_softc *sc;
409 struct cryptop *crp;
410 struct cryptodesc *crde, *crda;
411 struct uio *uio;
412 struct iovec *iov;
413 u_int i, lsec, nsec, data_secsize, decr_secsize, encr_secsize;
414 off_t dstoff;
415 int err, error;
416 u_char *p, *data, *auth, *authkey, *plaindata;
417
418 G_ELI_LOGREQ(3, bp, "%s", __func__);
419
420 bp->bio_pflags = wr->w_number;
421 sc = wr->w_softc;
422 /* Sectorsize of decrypted provider eg. 4096. */
423 decr_secsize = bp->bio_to->sectorsize;
424 /* The real sectorsize of encrypted provider, eg. 512. */
425 encr_secsize = LIST_FIRST(&sc->sc_geom->consumer)->provider->sectorsize;
426 /* Number of data bytes in one encrypted sector, eg. 480. */
427 data_secsize = sc->sc_data_per_sector;
428 /* Number of sectors from decrypted provider, eg. 2. */
429 nsec = bp->bio_length / decr_secsize;
430 /* Number of sectors from encrypted provider, eg. 18. */
431 nsec = (nsec * sc->sc_bytes_per_sector) / encr_secsize;
432 /* Last sector number in every big sector, eg. 9. */
433 lsec = sc->sc_bytes_per_sector / encr_secsize;
434 /* Destination offset, used for IV generation. */
435 dstoff = (bp->bio_offset / bp->bio_to->sectorsize) * sc->sc_bytes_per_sector;
436
437 auth = NULL; /* Silence compiler warning. */
438 plaindata = bp->bio_data;
439 if (bp->bio_cmd == BIO_READ) {
440 data = bp->bio_driver2;
441 auth = data + encr_secsize * nsec;
442 p = auth + sc->sc_alen * nsec;
443 } else {
444 size_t size;
445
446 size = encr_secsize * nsec;
447 size += sizeof(*crp) * nsec;
448 size += sizeof(*crde) * nsec;
449 size += sizeof(*crda) * nsec;
450 size += G_ELI_AUTH_SECKEYLEN * nsec;
451 size += sizeof(*uio) * nsec;
452 size += sizeof(*iov) * nsec;
453 data = malloc(size, M_ELI, M_WAITOK);
454 bp->bio_driver2 = data;
455 p = data + encr_secsize * nsec;
456 }
457 bp->bio_inbed = 0;
458 bp->bio_children = nsec;
459
460 error = 0;
461 for (i = 1; i <= nsec; i++, dstoff += encr_secsize) {
462 crp = (struct cryptop *)p; p += sizeof(*crp);
463 crde = (struct cryptodesc *)p; p += sizeof(*crde);
464 crda = (struct cryptodesc *)p; p += sizeof(*crda);
465 authkey = (u_char *)p; p += G_ELI_AUTH_SECKEYLEN;
466 uio = (struct uio *)p; p += sizeof(*uio);
467 iov = (struct iovec *)p; p += sizeof(*iov);
468
469 data_secsize = sc->sc_data_per_sector;
470 if ((i % lsec) == 0)
471 data_secsize = decr_secsize % data_secsize;
472
473 if (bp->bio_cmd == BIO_READ) {
474 /* Remember read HMAC. */
475 bcopy(data, auth, sc->sc_alen);
476 auth += sc->sc_alen;
477 /* TODO: bzero(9) can be commented out later. */
478 bzero(data, sc->sc_alen);
479 } else {
480 bcopy(plaindata, data + sc->sc_alen, data_secsize);
481 plaindata += data_secsize;
482 }
483
484 iov->iov_len = sc->sc_alen + data_secsize;
485 iov->iov_base = data;
486 data += encr_secsize;
487
488 uio->uio_iov = iov;
489 uio->uio_iovcnt = 1;
490 uio->uio_segflg = UIO_SYSSPACE;
491 uio->uio_resid = iov->iov_len;
492
493 crp->crp_sid = wr->w_sid;
494 crp->crp_ilen = uio->uio_resid;
495 crp->crp_olen = data_secsize;
496 crp->crp_opaque = (void *)bp;
497 crp->crp_buf = (void *)uio;
498 crp->crp_flags = CRYPTO_F_IOV | CRYPTO_F_CBIFSYNC | CRYPTO_F_REL;
499 if (g_eli_batch)
500 crp->crp_flags |= CRYPTO_F_BATCH;
501 if (bp->bio_cmd == BIO_WRITE) {
502 crp->crp_callback = g_eli_auth_write_done;
503 crp->crp_desc = crde;
504 crde->crd_next = crda;
505 crda->crd_next = NULL;
506 } else {
507 crp->crp_callback = g_eli_auth_read_done;
508 crp->crp_desc = crda;
509 crda->crd_next = crde;
510 crde->crd_next = NULL;
511 }
512
513 crde->crd_skip = sc->sc_alen;
514 crde->crd_len = data_secsize;
515 crde->crd_flags = CRD_F_IV_EXPLICIT | CRD_F_IV_PRESENT;
|