1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2005-2019 Pawel Jakub Dawidek <pawel@dawidek.net>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/cons.h>
33 #include <sys/kenv.h>
34 #include <sys/kernel.h>
35 #include <sys/linker.h>
36 #include <sys/module.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/bio.h>
40 #include <sys/sbuf.h>
41 #include <sys/sysctl.h>
42 #include <sys/malloc.h>
43 #include <sys/eventhandler.h>
44 #include <sys/kthread.h>
45 #include <sys/proc.h>
46 #include <sys/sched.h>
47 #include <sys/smp.h>
48 #include <sys/uio.h>
49 #include <sys/vnode.h>
50
51 #include <machine/vmparam.h>
52
53 #include <vm/uma.h>
54 #include <vm/vm.h>
55 #include <vm/swap_pager.h>
56
57 #include <geom/geom.h>
58 #include <geom/geom_dbg.h>
59 #include <geom/eli/g_eli.h>
60 #include <geom/eli/pkcs5v2.h>
61
62 #include <crypto/intake.h>
63
64 FEATURE(geom_eli, "GEOM crypto module");
65
66 MALLOC_DEFINE(M_ELI, "eli_data", "GEOM_ELI Data");
67
68 SYSCTL_DECL(_kern_geom);
69 SYSCTL_NODE(_kern_geom, OID_AUTO, eli, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
70 "GEOM_ELI stuff");
71 static int g_eli_version = G_ELI_VERSION;
72 SYSCTL_INT(_kern_geom_eli, OID_AUTO, version, CTLFLAG_RD, &g_eli_version, 0,
73 "GELI version");
74 int g_eli_debug = 0;
75 SYSCTL_INT(_kern_geom_eli, OID_AUTO, debug, CTLFLAG_RWTUN, &g_eli_debug, 0,
76 "Debug level");
77 static u_int g_eli_tries = 3;
78 SYSCTL_UINT(_kern_geom_eli, OID_AUTO, tries, CTLFLAG_RWTUN, &g_eli_tries, 0,
79 "Number of tries for entering the passphrase");
80 static u_int g_eli_visible_passphrase = GETS_NOECHO;
81 SYSCTL_UINT(_kern_geom_eli, OID_AUTO, visible_passphrase, CTLFLAG_RWTUN,
82 &g_eli_visible_passphrase, 0,
83 "Visibility of passphrase prompt (0 = invisible, 1 = visible, 2 = asterisk)");
84 u_int g_eli_overwrites = G_ELI_OVERWRITES;
85 SYSCTL_UINT(_kern_geom_eli, OID_AUTO, overwrites, CTLFLAG_RWTUN, &g_eli_overwrites,
86 0, "Number of times on-disk keys should be overwritten when destroying them");
87 static u_int g_eli_threads = 0;
88 SYSCTL_UINT(_kern_geom_eli, OID_AUTO, threads, CTLFLAG_RWTUN, &g_eli_threads, 0,
89 "Number of threads doing crypto work");
90 u_int g_eli_batch = 0;
91 SYSCTL_UINT(_kern_geom_eli, OID_AUTO, batch, CTLFLAG_RWTUN, &g_eli_batch, 0,
92 "Use crypto operations batching");
93 static u_int g_eli_minbufs = 16;
94 static int sysctl_g_eli_minbufs(SYSCTL_HANDLER_ARGS);
95 SYSCTL_PROC(_kern_geom_eli, OID_AUTO, minbufs, CTLTYPE_UINT | CTLFLAG_RW |
96 CTLFLAG_MPSAFE, NULL, 0, sysctl_g_eli_minbufs, "IU",
97 "Number of GELI bufs reserved for swap transactions");
98 static bool g_eli_blocking_malloc = false;
99 SYSCTL_BOOL(_kern_geom_eli, OID_AUTO, blocking_malloc, CTLFLAG_RWTUN,
100 &g_eli_blocking_malloc, 0, "Use blocking malloc calls for GELI buffers");
101 static bool g_eli_unmapped_io = true;
102 SYSCTL_BOOL(_kern_geom_eli, OID_AUTO, unmapped_io, CTLFLAG_RDTUN,
103 &g_eli_unmapped_io, 0, "Enable support for unmapped I/O");
104 static int g_eli_alloc_sz;
105 SYSCTL_UINT(_kern_geom_eli, OID_AUTO, use_uma_bytes, CTLFLAG_RD,
106 &g_eli_alloc_sz, 0, "Use uma(9) for allocations of this size or smaller.");
107
108 static struct sx g_eli_umalock; /* Controls changes to UMA zone. */
109 SX_SYSINIT(g_eli_umalock, &g_eli_umalock, "GELI UMA");
110 static uma_zone_t g_eli_uma = NULL;
111 static volatile int g_eli_umaoutstanding;
112 static volatile int g_eli_devs;
113
114 /*
115 * Control the number of reserved entries in the GELI zone.
116 * If the GELI zone has already been allocated, update the zone. Otherwise,
117 * simply update the variable for use the next time the zone is created.
118 */
119 static int
sysctl_g_eli_minbufs(SYSCTL_HANDLER_ARGS)120 sysctl_g_eli_minbufs(SYSCTL_HANDLER_ARGS)
121 {
122 int error;
123 u_int new;
124
125 new = g_eli_minbufs;
126 error = sysctl_handle_int(oidp, &new, 0, req);
127 if (error != 0 || req->newptr == NULL)
128 return (error);
129 sx_xlock(&g_eli_umalock);
130 if (g_eli_uma != NULL) {
131 if (new != g_eli_minbufs)
132 uma_zone_reserve(g_eli_uma, new);
133 if (new > g_eli_minbufs)
134 uma_prealloc(g_eli_uma, new - g_eli_minbufs);
135 }
136 if (new != g_eli_minbufs)
137 g_eli_minbufs = new;
138 sx_xunlock(&g_eli_umalock);
139 return (0);
140 }
141
142 /*
143 * Passphrase cached during boot, in order to be more user-friendly if
144 * there are multiple providers using the same passphrase.
145 */
146 static char cached_passphrase[256];
147 static u_int g_eli_boot_passcache = 1;
148 TUNABLE_INT("kern.geom.eli.boot_passcache", &g_eli_boot_passcache);
149 SYSCTL_UINT(_kern_geom_eli, OID_AUTO, boot_passcache, CTLFLAG_RD,
150 &g_eli_boot_passcache, 0,
151 "Passphrases are cached during boot process for possible reuse");
152 static void
fetch_loader_passphrase(void * dummy)153 fetch_loader_passphrase(void * dummy)
154 {
155 char * env_passphrase;
156
157 KASSERT(dynamic_kenv, ("need dynamic kenv"));
158
159 if ((env_passphrase = kern_getenv("kern.geom.eli.passphrase")) != NULL) {
160 /* Extract passphrase from the environment. */
161 strlcpy(cached_passphrase, env_passphrase,
162 sizeof(cached_passphrase));
163 freeenv(env_passphrase);
164
165 /* Wipe the passphrase from the environment. */
166 kern_unsetenv("kern.geom.eli.passphrase");
167 }
168 }
169 SYSINIT(geli_fetch_loader_passphrase, SI_SUB_KMEM + 1, SI_ORDER_ANY,
170 fetch_loader_passphrase, NULL);
171
172 static void
zero_boot_passcache(void)173 zero_boot_passcache(void)
174 {
175
176 explicit_bzero(cached_passphrase, sizeof(cached_passphrase));
177 }
178
179 static void
zero_geli_intake_keys(void)180 zero_geli_intake_keys(void)
181 {
182 struct keybuf *keybuf;
183 int i;
184
185 if ((keybuf = get_keybuf()) != NULL) {
186 /* Scan the key buffer, clear all GELI keys. */
187 for (i = 0; i < keybuf->kb_nents; i++) {
188 if (keybuf->kb_ents[i].ke_type == KEYBUF_TYPE_GELI) {
189 explicit_bzero(keybuf->kb_ents[i].ke_data,
190 sizeof(keybuf->kb_ents[i].ke_data));
191 keybuf->kb_ents[i].ke_type = KEYBUF_TYPE_NONE;
192 }
193 }
194 }
195 }
196
197 static void
zero_intake_passcache(void * dummy)198 zero_intake_passcache(void *dummy)
199 {
200 zero_boot_passcache();
201 zero_geli_intake_keys();
202 }
203 EVENTHANDLER_DEFINE(mountroot, zero_intake_passcache, NULL, 0);
204
205 static eventhandler_tag g_eli_pre_sync = NULL;
206
207 static int g_eli_read_metadata_offset(struct g_class *mp, struct g_provider *pp,
208 off_t offset, struct g_eli_metadata *md);
209
210 static int g_eli_destroy_geom(struct gctl_req *req, struct g_class *mp,
211 struct g_geom *gp);
212 static void g_eli_init(struct g_class *mp);
213 static void g_eli_fini(struct g_class *mp);
214
215 static g_taste_t g_eli_taste;
216 static g_dumpconf_t g_eli_dumpconf;
217
218 struct g_class g_eli_class = {
219 .name = G_ELI_CLASS_NAME,
220 .version = G_VERSION,
221 .ctlreq = g_eli_config,
222 .taste = g_eli_taste,
223 .destroy_geom = g_eli_destroy_geom,
224 .init = g_eli_init,
225 .fini = g_eli_fini
226 };
227
228 /*
229 * Code paths:
230 * BIO_READ:
231 * g_eli_start -> g_eli_crypto_read -> g_io_request -> g_eli_read_done -> g_eli_crypto_run -> g_eli_crypto_read_done -> g_io_deliver
232 * BIO_WRITE:
233 * g_eli_start -> g_eli_crypto_run -> g_eli_crypto_write_done -> g_io_request -> g_eli_write_done -> g_io_deliver
234 */
235
236 /*
237 * EAGAIN from crypto(9) means, that we were probably balanced to another crypto
238 * accelerator or something like this.
239 * The function updates the SID and rerun the operation.
240 */
241 int
g_eli_crypto_rerun(struct cryptop * crp)242 g_eli_crypto_rerun(struct cryptop *crp)
243 {
244 struct g_eli_softc *sc;
245 struct g_eli_worker *wr;
246 struct bio *bp;
247 int error;
248
249 bp = (struct bio *)crp->crp_opaque;
250 sc = bp->bio_to->geom->softc;
251 LIST_FOREACH(wr, &sc->sc_workers, w_next) {
252 if (wr->w_number == G_ELI_WORKER(bp->bio_pflags))
253 break;
254 }
255 KASSERT(wr != NULL, ("Invalid worker (%u).",
256 G_ELI_WORKER(bp->bio_pflags)));
257 G_ELI_DEBUG(1, "Rerunning crypto %s request (sid: %p -> %p).",
258 bp->bio_cmd == BIO_READ ? "READ" : "WRITE", wr->w_sid,
259 crp->crp_session);
260 wr->w_sid = crp->crp_session;
261 crp->crp_etype = 0;
262 error = crypto_dispatch(crp);
263 if (error == 0)
264 return (0);
265 G_ELI_DEBUG(1, "%s: crypto_dispatch() returned %d.", __func__, error);
266 crp->crp_etype = error;
267 return (error);
268 }
269
270 static void
g_eli_getattr_done(struct bio * bp)271 g_eli_getattr_done(struct bio *bp)
272 {
273 if (bp->bio_error == 0 &&
274 !strcmp(bp->bio_attribute, "GEOM::physpath")) {
275 strlcat(bp->bio_data, "/eli", bp->bio_length);
276 }
277 g_std_done(bp);
278 }
279
280 /*
281 * The function is called afer reading encrypted data from the provider.
282 *
283 * g_eli_start -> g_eli_crypto_read -> g_io_request -> G_ELI_READ_DONE -> g_eli_crypto_run -> g_eli_crypto_read_done -> g_io_deliver
284 */
285 void
g_eli_read_done(struct bio * bp)286 g_eli_read_done(struct bio *bp)
287 {
288 struct g_eli_softc *sc;
289 struct bio *pbp;
290
291 G_ELI_LOGREQ(2, bp, "Request done.");
292 pbp = bp->bio_parent;
293 if (pbp->bio_error == 0 && bp->bio_error != 0)
294 pbp->bio_error = bp->bio_error;
295 g_destroy_bio(bp);
296 /*
297 * Do we have all sectors already?
298 */
299 pbp->bio_inbed++;
300 if (pbp->bio_inbed < pbp->bio_children)
301 return;
302 sc = pbp->bio_to->geom->softc;
303 if (pbp->bio_error != 0) {
304 G_ELI_LOGREQ(0, pbp, "%s() failed (error=%d)", __func__,
305 pbp->bio_error);
306 pbp->bio_completed = 0;
307 g_eli_free_data(pbp);
308 g_io_deliver(pbp, pbp->bio_error);
309 if (sc != NULL)
310 atomic_subtract_int(&sc->sc_inflight, 1);
311 return;
312 }
313 mtx_lock(&sc->sc_queue_mtx);
314 bioq_insert_tail(&sc->sc_queue, pbp);
315 mtx_unlock(&sc->sc_queue_mtx);
316 wakeup(sc);
317 }
318
319 /*
320 * The function is called after we encrypt and write data.
321 *
322 * g_eli_start -> g_eli_crypto_run -> g_eli_crypto_write_done -> g_io_request -> G_ELI_WRITE_DONE -> g_io_deliver
323 */
324 void
g_eli_write_done(struct bio * bp)325 g_eli_write_done(struct bio *bp)
326 {
327 struct g_eli_softc *sc;
328 struct bio *pbp;
329
330 G_ELI_LOGREQ(2, bp, "Request done.");
331 pbp = bp->bio_parent;
332 if (pbp->bio_error == 0 && bp->bio_error != 0)
333 pbp->bio_error = bp->bio_error;
334 g_destroy_bio(bp);
335 /*
336 * Do we have all sectors already?
337 */
338 pbp->bio_inbed++;
339 if (pbp->bio_inbed < pbp->bio_children)
340 return;
341 sc = pbp->bio_to->geom->softc;
342 g_eli_free_data(pbp);
343 if (pbp->bio_error != 0) {
344 G_ELI_LOGREQ(0, pbp, "%s() failed (error=%d)", __func__,
345 pbp->bio_error);
346 pbp->bio_completed = 0;
347 } else
348 pbp->bio_completed = pbp->bio_length;
349
350 /*
351 * Write is finished, send it up.
352 */
353 g_io_deliver(pbp, pbp->bio_error);
354 if (sc != NULL)
355 atomic_subtract_int(&sc->sc_inflight, 1);
356 }
357
358 /*
359 * This function should never be called, but GEOM made as it set ->orphan()
360 * method for every geom.
361 */
362 static void
g_eli_orphan_spoil_assert(struct g_consumer * cp)363 g_eli_orphan_spoil_assert(struct g_consumer *cp)
364 {
365
366 panic("Function %s() called for %s.", __func__, cp->geom->name);
367 }
368
369 static void
g_eli_orphan(struct g_consumer * cp)370 g_eli_orphan(struct g_consumer *cp)
371 {
372 struct g_eli_softc *sc;
373
374 g_topology_assert();
375 sc = cp->geom->softc;
376 if (sc == NULL)
377 return;
378 g_eli_destroy(sc, TRUE);
379 }
380
381 static void
g_eli_resize(struct g_consumer * cp)382 g_eli_resize(struct g_consumer *cp)
383 {
384 struct g_eli_softc *sc;
385 struct g_provider *epp, *pp;
386 off_t oldsize;
387
388 g_topology_assert();
389 sc = cp->geom->softc;
390 if (sc == NULL)
391 return;
392
393 if ((sc->sc_flags & G_ELI_FLAG_AUTORESIZE) == 0) {
394 G_ELI_DEBUG(0, "Autoresize is turned off, old size: %jd.",
395 (intmax_t)sc->sc_provsize);
396 return;
397 }
398
399 pp = cp->provider;
400
401 if ((sc->sc_flags & G_ELI_FLAG_ONETIME) == 0) {
402 struct g_eli_metadata md;
403 u_char *sector;
404 int error;
405
406 sector = NULL;
407
408 error = g_eli_read_metadata_offset(cp->geom->class, pp,
409 sc->sc_provsize - pp->sectorsize, &md);
410 if (error != 0) {
411 G_ELI_DEBUG(0, "Cannot read metadata from %s (error=%d).",
412 pp->name, error);
413 goto iofail;
414 }
415
416 md.md_provsize = pp->mediasize;
417
418 sector = malloc(pp->sectorsize, M_ELI, M_WAITOK | M_ZERO);
419 eli_metadata_encode(&md, sector);
420 error = g_write_data(cp, pp->mediasize - pp->sectorsize, sector,
421 pp->sectorsize);
422 if (error != 0) {
423 G_ELI_DEBUG(0, "Cannot store metadata on %s (error=%d).",
424 pp->name, error);
425 goto iofail;
426 }
427 explicit_bzero(sector, pp->sectorsize);
428 error = g_write_data(cp, sc->sc_provsize - pp->sectorsize,
429 sector, pp->sectorsize);
430 if (error != 0) {
431 G_ELI_DEBUG(0, "Cannot clear old metadata from %s (error=%d).",
432 pp->name, error);
433 goto iofail;
434 }
435 iofail:
436 explicit_bzero(&md, sizeof(md));
437 zfree(sector, M_ELI);
438 }
439
440 oldsize = sc->sc_mediasize;
441 sc->sc_mediasize = eli_mediasize(sc, pp->mediasize, pp->sectorsize);
442 g_eli_key_resize(sc);
443 sc->sc_provsize = pp->mediasize;
444
445 epp = LIST_FIRST(&sc->sc_geom->provider);
446 g_resize_provider(epp, sc->sc_mediasize);
447 G_ELI_DEBUG(0, "Device %s size changed from %jd to %jd.", epp->name,
448 (intmax_t)oldsize, (intmax_t)sc->sc_mediasize);
449 }
450
451 /*
452 * BIO_READ:
453 * G_ELI_START -> g_eli_crypto_read -> g_io_request -> g_eli_read_done -> g_eli_crypto_run -> g_eli_crypto_read_done -> g_io_deliver
454 * BIO_WRITE:
455 * G_ELI_START -> g_eli_crypto_run -> g_eli_crypto_write_done -> g_io_request -> g_eli_write_done -> g_io_deliver
456 */
457 static void
g_eli_start(struct bio * bp)458 g_eli_start(struct bio *bp)
459 {
460 struct g_eli_softc *sc;
461 struct g_consumer *cp;
462 struct bio *cbp;
463
464 sc = bp->bio_to->geom->softc;
465 KASSERT(sc != NULL,
466 ("Provider's error should be set (error=%d)(device=%s).",
467 bp->bio_to->error, bp->bio_to->name));
468 G_ELI_LOGREQ(2, bp, "Request received.");
469
470 switch (bp->bio_cmd) {
471 case BIO_READ:
472 case BIO_WRITE:
473 case BIO_GETATTR:
474 case BIO_FLUSH:
475 case BIO_ZONE:
476 case BIO_SPEEDUP:
477 break;
478 case BIO_DELETE:
479 /*
480 * If the user hasn't set the NODELETE flag, we just pass
481 * it down the stack and let the layers beneath us do (or
482 * not) whatever they do with it. If they have, we
483 * reject it. A possible extension would be an
484 * additional flag to take it as a hint to shred the data
485 * with [multiple?] overwrites.
486 */
487 if (!(sc->sc_flags & G_ELI_FLAG_NODELETE))
488 break;
489 default:
490 g_io_deliver(bp, EOPNOTSUPP);
491 return;
492 }
493 cbp = g_clone_bio(bp);
494 if (cbp == NULL) {
495 g_io_deliver(bp, ENOMEM);
496 return;
497 }
498 bp->bio_driver1 = cbp;
499 bp->bio_pflags = 0;
500 G_ELI_SET_NEW_BIO(bp->bio_pflags);
501 switch (bp->bio_cmd) {
502 case BIO_READ:
503 if (!(sc->sc_flags & G_ELI_FLAG_AUTH)) {
504 g_eli_crypto_read(sc, bp, 0);
505 break;
506 }
507 /* FALLTHROUGH */
508 case BIO_WRITE:
509 mtx_lock(&sc->sc_queue_mtx);
510 bioq_insert_tail(&sc->sc_queue, bp);
511 mtx_unlock(&sc->sc_queue_mtx);
512 wakeup(sc);
513 break;
514 case BIO_GETATTR:
515 case BIO_FLUSH:
516 case BIO_DELETE:
517 case BIO_SPEEDUP:
518 case BIO_ZONE:
519 if (bp->bio_cmd == BIO_GETATTR)
520 cbp->bio_done = g_eli_getattr_done;
521 else
522 cbp->bio_done = g_std_done;
523 cp = LIST_FIRST(&sc->sc_geom->consumer);
524 cbp->bio_to = cp->provider;
525 G_ELI_LOGREQ(2, cbp, "Sending request.");
526 g_io_request(cbp, cp);
527 break;
528 }
529 }
530
531 static int
g_eli_newsession(struct g_eli_worker * wr)532 g_eli_newsession(struct g_eli_worker *wr)
533 {
534 struct g_eli_softc *sc;
535 struct crypto_session_params csp;
536 uint32_t caps;
537 int error, new_crypto;
538 void *key;
539
540 sc = wr->w_softc;
541
542 memset(&csp, 0, sizeof(csp));
543 csp.csp_mode = CSP_MODE_CIPHER;
544 csp.csp_cipher_alg = sc->sc_ealgo;
545 csp.csp_ivlen = g_eli_ivlen(sc->sc_ealgo);
546 csp.csp_cipher_klen = sc->sc_ekeylen / 8;
547 if (sc->sc_ealgo == CRYPTO_AES_XTS)
548 csp.csp_cipher_klen <<= 1;
549 if ((sc->sc_flags & G_ELI_FLAG_FIRST_KEY) != 0) {
550 key = g_eli_key_hold(sc, 0,
551 LIST_FIRST(&sc->sc_geom->consumer)->provider->sectorsize);
552 csp.csp_cipher_key = key;
553 } else {
554 key = NULL;
555 csp.csp_cipher_key = sc->sc_ekey;
556 }
557 if (sc->sc_flags & G_ELI_FLAG_AUTH) {
558 csp.csp_mode = CSP_MODE_ETA;
559 csp.csp_auth_alg = sc->sc_aalgo;
560 csp.csp_auth_klen = G_ELI_AUTH_SECKEYLEN;
561 }
562
563 switch (sc->sc_crypto) {
564 case G_ELI_CRYPTO_SW_ACCEL:
565 case G_ELI_CRYPTO_SW:
566 error = crypto_newsession(&wr->w_sid, &csp,
567 CRYPTOCAP_F_SOFTWARE);
568 break;
569 case G_ELI_CRYPTO_HW:
570 error = crypto_newsession(&wr->w_sid, &csp,
571 CRYPTOCAP_F_HARDWARE);
572 break;
573 case G_ELI_CRYPTO_UNKNOWN:
574 error = crypto_newsession(&wr->w_sid, &csp,
575 CRYPTOCAP_F_HARDWARE | CRYPTOCAP_F_SOFTWARE);
576 if (error == 0) {
577 caps = crypto_ses2caps(wr->w_sid);
578 if (caps & CRYPTOCAP_F_HARDWARE)
579 new_crypto = G_ELI_CRYPTO_HW;
580 else if (caps & CRYPTOCAP_F_ACCEL_SOFTWARE)
581 new_crypto = G_ELI_CRYPTO_SW_ACCEL;
582 else
583 new_crypto = G_ELI_CRYPTO_SW;
584 mtx_lock(&sc->sc_queue_mtx);
585 if (sc->sc_crypto == G_ELI_CRYPTO_UNKNOWN)
586 sc->sc_crypto = new_crypto;
587 mtx_unlock(&sc->sc_queue_mtx);
588 }
589 break;
590 default:
591 panic("%s: invalid condition", __func__);
592 }
593
594 if ((sc->sc_flags & G_ELI_FLAG_FIRST_KEY) != 0) {
595 if (error)
596 g_eli_key_drop(sc, key);
597 else
598 wr->w_first_key = key;
599 }
600
601 return (error);
602 }
603
604 static void
g_eli_freesession(struct g_eli_worker * wr)605 g_eli_freesession(struct g_eli_worker *wr)
606 {
607 struct g_eli_softc *sc;
608
609 crypto_freesession(wr->w_sid);
610 if (wr->w_first_key != NULL) {
611 sc = wr->w_softc;
612 g_eli_key_drop(sc, wr->w_first_key);
613 wr->w_first_key = NULL;
614 }
615 }
616
617 static void
g_eli_cancel(struct g_eli_softc * sc)618 g_eli_cancel(struct g_eli_softc *sc)
619 {
620 struct bio *bp;
621
622 mtx_assert(&sc->sc_queue_mtx, MA_OWNED);
623
624 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) {
625 KASSERT(G_ELI_IS_NEW_BIO(bp->bio_pflags),
626 ("Not new bio when canceling (bp=%p).", bp));
627 g_io_deliver(bp, ENXIO);
628 }
629 }
630
631 static struct bio *
g_eli_takefirst(struct g_eli_softc * sc)632 g_eli_takefirst(struct g_eli_softc *sc)
633 {
634 struct bio *bp;
635
636 mtx_assert(&sc->sc_queue_mtx, MA_OWNED);
637
638 if (!(sc->sc_flags & G_ELI_FLAG_SUSPEND))
639 return (bioq_takefirst(&sc->sc_queue));
640 /*
641 * Device suspended, so we skip new I/O requests.
642 */
643 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
644 if (!G_ELI_IS_NEW_BIO(bp->bio_pflags))
645 break;
646 }
647 if (bp != NULL)
648 bioq_remove(&sc->sc_queue, bp);
649 return (bp);
650 }
651
652 /*
653 * This is the main function for kernel worker thread when we don't have
654 * hardware acceleration and we have to do cryptography in software.
655 * Dedicated thread is needed, so we don't slow down g_up/g_down GEOM
656 * threads with crypto work.
657 */
658 static void
g_eli_worker(void * arg)659 g_eli_worker(void *arg)
660 {
661 struct g_eli_softc *sc;
662 struct g_eli_worker *wr;
663 struct bio *bp;
664 int error __diagused;
665
666 wr = arg;
667 sc = wr->w_softc;
668 #ifdef EARLY_AP_STARTUP
669 MPASS(!sc->sc_cpubind || smp_started);
670 #elif defined(SMP)
671 /* Before sched_bind() to a CPU, wait for all CPUs to go on-line. */
672 if (sc->sc_cpubind) {
673 while (!smp_started)
674 tsleep(wr, 0, "geli:smp", hz / 4);
675 }
676 #endif
677 thread_lock(curthread);
678 sched_prio(curthread, PUSER);
679 if (sc->sc_cpubind)
680 sched_bind(curthread, wr->w_number % mp_ncpus);
681 thread_unlock(curthread);
682
683 G_ELI_DEBUG(1, "Thread %s started.", curthread->td_proc->p_comm);
684
685 for (;;) {
686 mtx_lock(&sc->sc_queue_mtx);
687 again:
688 bp = g_eli_takefirst(sc);
689 if (bp == NULL) {
690 if (sc->sc_flags & G_ELI_FLAG_DESTROY) {
691 g_eli_cancel(sc);
692 LIST_REMOVE(wr, w_next);
693 g_eli_freesession(wr);
694 free(wr, M_ELI);
695 G_ELI_DEBUG(1, "Thread %s exiting.",
696 curthread->td_proc->p_comm);
697 wakeup(&sc->sc_workers);
698 mtx_unlock(&sc->sc_queue_mtx);
699 kproc_exit(0);
700 }
701 while (sc->sc_flags & G_ELI_FLAG_SUSPEND) {
702 if (sc->sc_inflight > 0) {
703 G_ELI_DEBUG(0, "inflight=%d",
704 sc->sc_inflight);
705 /*
706 * We still have inflight BIOs, so
707 * sleep and retry.
708 */
709 msleep(sc, &sc->sc_queue_mtx, PRIBIO,
710 "geli:inf", hz / 5);
711 goto again;
712 }
713 /*
714 * Suspend requested, mark the worker as
715 * suspended and go to sleep.
716 */
717 if (wr->w_active) {
718 g_eli_freesession(wr);
719 wr->w_active = FALSE;
720 }
721 wakeup(&sc->sc_workers);
722 msleep(sc, &sc->sc_queue_mtx, PRIBIO,
723 "geli:suspend", 0);
724 if (!wr->w_active &&
725 !(sc->sc_flags & G_ELI_FLAG_SUSPEND)) {
726 error = g_eli_newsession(wr);
727 KASSERT(error == 0,
728 ("g_eli_newsession() failed on resume (error=%d)",
729 error));
730 wr->w_active = TRUE;
731 }
732 goto again;
733 }
734 msleep(sc, &sc->sc_queue_mtx, PDROP, "geli:w", 0);
735 continue;
736 }
737 if (G_ELI_IS_NEW_BIO(bp->bio_pflags))
738 atomic_add_int(&sc->sc_inflight, 1);
739 mtx_unlock(&sc->sc_queue_mtx);
740 if (G_ELI_IS_NEW_BIO(bp->bio_pflags)) {
741 G_ELI_SETWORKER(bp->bio_pflags, 0);
742 if (sc->sc_flags & G_ELI_FLAG_AUTH) {
743 if (bp->bio_cmd == BIO_READ)
744 g_eli_auth_read(sc, bp);
745 else
746 g_eli_auth_run(wr, bp);
747 } else {
748 if (bp->bio_cmd == BIO_READ)
749 g_eli_crypto_read(sc, bp, 1);
750 else
751 g_eli_crypto_run(wr, bp);
752 }
753 } else {
754 if (sc->sc_flags & G_ELI_FLAG_AUTH)
755 g_eli_auth_run(wr, bp);
756 else
757 g_eli_crypto_run(wr, bp);
758 }
759 }
760 }
761
762 static int
g_eli_read_metadata_offset(struct g_class * mp,struct g_provider * pp,off_t offset,struct g_eli_metadata * md)763 g_eli_read_metadata_offset(struct g_class *mp, struct g_provider *pp,
764 off_t offset, struct g_eli_metadata *md)
765 {
766 struct g_geom *gp;
767 struct g_consumer *cp;
768 u_char *buf = NULL;
769 int error;
770
771 g_topology_assert();
772
773 gp = g_new_geomf(mp, "eli:taste");
774 gp->start = g_eli_start;
775 gp->access = g_std_access;
776 /*
777 * g_eli_read_metadata() is always called from the event thread.
778 * Our geom is created and destroyed in the same event, so there
779 * could be no orphan nor spoil event in the meantime.
780 */
781 gp->orphan = g_eli_orphan_spoil_assert;
782 gp->spoiled = g_eli_orphan_spoil_assert;
783 cp = g_new_consumer(gp);
784 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
785 error = g_attach(cp, pp);
786 if (error != 0)
787 goto end;
788 error = g_access(cp, 1, 0, 0);
789 if (error != 0)
790 goto end;
791 g_topology_unlock();
792 buf = g_read_data(cp, offset, pp->sectorsize, &error);
793 g_topology_lock();
794 if (buf == NULL)
795 goto end;
796 error = eli_metadata_decode(buf, md);
797 if (error != 0)
798 goto end;
799 /* Metadata was read and decoded successfully. */
800 end:
801 g_free(buf);
802 if (cp->provider != NULL) {
803 if (cp->acr == 1)
804 g_access(cp, -1, 0, 0);
805 g_detach(cp);
806 }
807 g_destroy_consumer(cp);
808 g_destroy_geom(gp);
809 return (error);
810 }
811
812 int
g_eli_read_metadata(struct g_class * mp,struct g_provider * pp,struct g_eli_metadata * md)813 g_eli_read_metadata(struct g_class *mp, struct g_provider *pp,
814 struct g_eli_metadata *md)
815 {
816
817 return (g_eli_read_metadata_offset(mp, pp,
818 pp->mediasize - pp->sectorsize, md));
819 }
820
821 /*
822 * The function is called when we had last close on provider and user requested
823 * to close it when this situation occur.
824 */
825 static void
g_eli_last_close(void * arg,int flags __unused)826 g_eli_last_close(void *arg, int flags __unused)
827 {
828 struct g_geom *gp;
829 char gpname[64];
830 int error __diagused;
831
832 g_topology_assert();
833 gp = arg;
834 strlcpy(gpname, gp->name, sizeof(gpname));
835 error = g_eli_destroy(gp->softc, TRUE);
836 KASSERT(error == 0, ("Cannot detach %s on last close (error=%d).",
837 gpname, error));
838 G_ELI_DEBUG(0, "Detached %s on last close.", gpname);
839 }
840
841 int
g_eli_access(struct g_provider * pp,int dr,int dw,int de)842 g_eli_access(struct g_provider *pp, int dr, int dw, int de)
843 {
844 struct g_eli_softc *sc;
845 struct g_geom *gp;
846
847 gp = pp->geom;
848 sc = gp->softc;
849
850 if (dw > 0) {
851 if (sc->sc_flags & G_ELI_FLAG_RO) {
852 /* Deny write attempts. */
853 return (EROFS);
854 }
855 /* Someone is opening us for write, we need to remember that. */
856 sc->sc_flags |= G_ELI_FLAG_WOPEN;
857 return (0);
858 }
859 /* Is this the last close? */
860 if (pp->acr + dr > 0 || pp->acw + dw > 0 || pp->ace + de > 0)
861 return (0);
862
863 /*
864 * Automatically detach on last close if requested.
865 */
866 if ((sc->sc_flags & G_ELI_FLAG_RW_DETACH) ||
867 (sc->sc_flags & G_ELI_FLAG_WOPEN)) {
868 g_post_event(g_eli_last_close, gp, M_WAITOK, NULL);
869 }
870 return (0);
871 }
872
873 static int
g_eli_cpu_is_disabled(int cpu)874 g_eli_cpu_is_disabled(int cpu)
875 {
876 #ifdef SMP
877 return (CPU_ISSET(cpu, &hlt_cpus_mask));
878 #else
879 return (0);
880 #endif
881 }
882
883 static void
g_eli_init_uma(void)884 g_eli_init_uma(void)
885 {
886
887 atomic_add_int(&g_eli_devs, 1);
888 sx_xlock(&g_eli_umalock);
889 if (g_eli_uma == NULL) {
890 /*
891 * Calculate the maximum-sized swap buffer we are
892 * likely to see.
893 */
894 g_eli_alloc_sz = roundup2((PAGE_SIZE + sizeof(int) +
895 G_ELI_AUTH_SECKEYLEN) * nsw_cluster_max +
896 sizeof(uintptr_t), PAGE_SIZE);
897
898 g_eli_uma = uma_zcreate("GELI buffers", g_eli_alloc_sz,
899 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
900
901 /* Reserve and pre-allocate pages, as appropriate. */
902 uma_zone_reserve(g_eli_uma, g_eli_minbufs);
903 uma_prealloc(g_eli_uma, g_eli_minbufs);
904 }
905 sx_xunlock(&g_eli_umalock);
906 }
907
908 /*
909 * Try to destroy the UMA pool. This will do nothing if there are existing
910 * GELI devices or existing UMA allocations.
911 */
912 static void
g_eli_destroy_uma(void)913 g_eli_destroy_uma(void)
914 {
915 uma_zone_t oldzone;
916
917 sx_xlock(&g_eli_umalock);
918 /* Ensure we really should be destroying this. */
919 if (atomic_load_int(&g_eli_devs) == 0 &&
920 atomic_load_int(&g_eli_umaoutstanding) == 0) {
921 oldzone = g_eli_uma;
922 g_eli_uma = NULL;
923 } else
924 oldzone = NULL;
925 sx_xunlock(&g_eli_umalock);
926
927 if (oldzone != NULL)
928 uma_zdestroy(oldzone);
929 }
930
931 static void
g_eli_fini_uma(void)932 g_eli_fini_uma(void)
933 {
934
935 /*
936 * If this is the last outstanding GELI device, try to
937 * destroy the UMA pool.
938 */
939 if (atomic_fetchadd_int(&g_eli_devs, -1) == 1)
940 g_eli_destroy_uma();
941 }
942
943 /*
944 * Allocate a data buffer. If the size fits within our swap-sized buffers,
945 * try to allocate a swap-sized buffer from the UMA pool. Otherwise, fall
946 * back to using malloc.
947 *
948 * Swap-related requests are special: they can only use the UMA pool, they
949 * use M_USE_RESERVE to let them dip farther into system resources, and
950 * they always use M_NOWAIT to prevent swap operations from deadlocking.
951 */
952 bool
g_eli_alloc_data(struct bio * bp,int sz)953 g_eli_alloc_data(struct bio *bp, int sz)
954 {
955
956 KASSERT(sz <= g_eli_alloc_sz || (bp->bio_flags & BIO_SWAP) == 0,
957 ("BIO_SWAP request for %d bytes exceeds the precalculated buffer"
958 " size (%d)", sz, g_eli_alloc_sz));
959 if (sz <= g_eli_alloc_sz) {
960 bp->bio_driver2 = uma_zalloc(g_eli_uma, M_NOWAIT |
961 ((bp->bio_flags & BIO_SWAP) != 0 ? M_USE_RESERVE : 0));
962 if (bp->bio_driver2 != NULL) {
963 bp->bio_pflags |= G_ELI_UMA_ALLOC;
964 atomic_add_int(&g_eli_umaoutstanding, 1);
965 }
966 if (bp->bio_driver2 != NULL || (bp->bio_flags & BIO_SWAP) != 0)
967 return (bp->bio_driver2 != NULL);
968 }
969 bp->bio_pflags &= ~(G_ELI_UMA_ALLOC);
970 bp->bio_driver2 = malloc(sz, M_ELI, g_eli_blocking_malloc ? M_WAITOK :
971 M_NOWAIT);
972 return (bp->bio_driver2 != NULL);
973 }
974
975 /*
976 * Free a buffer from bp->bio_driver2 which was allocated with
977 * g_eli_alloc_data(). This function makes sure that the memory is freed
978 * to the correct place.
979 *
980 * Additionally, if this function frees the last outstanding UMA request
981 * and there are no open GELI devices, this will destroy the UMA pool.
982 */
983 void
g_eli_free_data(struct bio * bp)984 g_eli_free_data(struct bio *bp)
985 {
986
987 /*
988 * Mimic the free(9) behavior of allowing a NULL pointer to be
989 * freed.
990 */
991 if (bp->bio_driver2 == NULL)
992 return;
993
994 if ((bp->bio_pflags & G_ELI_UMA_ALLOC) != 0) {
995 uma_zfree(g_eli_uma, bp->bio_driver2);
996 if (atomic_fetchadd_int(&g_eli_umaoutstanding, -1) == 1 &&
997 atomic_load_int(&g_eli_devs) == 0)
998 g_eli_destroy_uma();
999 } else
1000 free(bp->bio_driver2, M_ELI);
1001 bp->bio_driver2 = NULL;
1002 }
1003
1004 struct g_geom *
g_eli_create(struct gctl_req * req,struct g_class * mp,struct g_provider * bpp,const struct g_eli_metadata * md,const u_char * mkey,int nkey)1005 g_eli_create(struct gctl_req *req, struct g_class *mp, struct g_provider *bpp,
1006 const struct g_eli_metadata *md, const u_char *mkey, int nkey)
1007 {
1008 struct g_eli_softc *sc;
1009 struct g_eli_worker *wr;
1010 struct g_geom *gp;
1011 struct g_provider *pp;
1012 struct g_consumer *cp;
1013 struct g_geom_alias *gap;
1014 u_int i, threads;
1015 int dcw, error;
1016
1017 G_ELI_DEBUG(1, "Creating device %s%s.", bpp->name, G_ELI_SUFFIX);
1018 KASSERT(eli_metadata_crypto_supported(md),
1019 ("%s: unsupported crypto for %s", __func__, bpp->name));
1020
1021 gp = g_new_geomf(mp, "%s%s", bpp->name, G_ELI_SUFFIX);
1022 sc = malloc(sizeof(*sc), M_ELI, M_WAITOK | M_ZERO);
1023 gp->start = g_eli_start;
1024 /*
1025 * Spoiling can happen even though we have the provider open
1026 * exclusively, e.g. through media change events.
1027 */
1028 gp->spoiled = g_eli_orphan;
1029 gp->orphan = g_eli_orphan;
1030 gp->resize = g_eli_resize;
1031 gp->dumpconf = g_eli_dumpconf;
1032 /*
1033 * If detach-on-last-close feature is not enabled and we don't operate
1034 * on read-only provider, we can simply use g_std_access().
1035 */
1036 if (md->md_flags & (G_ELI_FLAG_WO_DETACH | G_ELI_FLAG_RO))
1037 gp->access = g_eli_access;
1038 else
1039 gp->access = g_std_access;
1040
1041 eli_metadata_softc(sc, md, bpp->sectorsize, bpp->mediasize);
1042 sc->sc_nkey = nkey;
1043
1044 gp->softc = sc;
1045 sc->sc_geom = gp;
1046
1047 bioq_init(&sc->sc_queue);
1048 mtx_init(&sc->sc_queue_mtx, "geli:queue", NULL, MTX_DEF);
1049 mtx_init(&sc->sc_ekeys_lock, "geli:ekeys", NULL, MTX_DEF);
1050 g_eli_init_uma();
1051
1052 pp = NULL;
1053 cp = g_new_consumer(gp);
1054 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
1055
1056 error = g_attach(cp, bpp);
1057 if (error != 0) {
1058 if (req != NULL) {
1059 gctl_error(req, "Cannot attach to %s (error=%d).",
1060 bpp->name, error);
1061 } else {
1062 G_ELI_DEBUG(1, "Cannot attach to %s (error=%d).",
1063 bpp->name, error);
1064 }
1065 goto failed;
1066 }
1067 /*
1068 * Keep provider open all the time, so we can run critical tasks,
1069 * like Master Keys deletion, without wondering if we can open
1070 * provider or not.
1071 * We don't open provider for writing only when user requested read-only
1072 * access.
1073 */
1074 dcw = (sc->sc_flags & G_ELI_FLAG_RO) ? 0 : 1;
1075 error = g_access(cp, 1, dcw, 1);
1076 if (error != 0) {
1077 if (req != NULL) {
1078 gctl_error(req, "Cannot access %s (error=%d).",
1079 bpp->name, error);
1080 } else {
1081 G_ELI_DEBUG(1, "Cannot access %s (error=%d).",
1082 bpp->name, error);
1083 }
1084 goto failed;
1085 }
1086
1087 /*
1088 * Remember the keys in our softc structure.
1089 */
1090 g_eli_mkey_propagate(sc, mkey);
1091
1092 LIST_INIT(&sc->sc_workers);
1093
1094 threads = g_eli_threads;
1095 if (threads == 0)
1096 threads = mp_ncpus;
1097 sc->sc_cpubind = (mp_ncpus > 1 && threads == mp_ncpus);
1098 for (i = 0; i < threads; i++) {
1099 if (g_eli_cpu_is_disabled(i)) {
1100 G_ELI_DEBUG(1, "%s: CPU %u disabled, skipping.",
1101 bpp->name, i);
1102 continue;
1103 }
1104 wr = malloc(sizeof(*wr), M_ELI, M_WAITOK | M_ZERO);
1105 wr->w_softc = sc;
1106 wr->w_number = i;
1107 wr->w_active = TRUE;
1108
1109 error = g_eli_newsession(wr);
1110 if (error != 0) {
1111 free(wr, M_ELI);
1112 if (req != NULL) {
1113 gctl_error(req, "Cannot set up crypto session "
1114 "for %s (error=%d).", bpp->name, error);
1115 } else {
1116 G_ELI_DEBUG(1, "Cannot set up crypto session "
1117 "for %s (error=%d).", bpp->name, error);
1118 }
1119 goto failed;
1120 }
1121
1122 error = kproc_create(g_eli_worker, wr, &wr->w_proc, 0, 0,
1123 "g_eli[%u] %s", i, bpp->name);
1124 if (error != 0) {
1125 g_eli_freesession(wr);
1126 free(wr, M_ELI);
1127 if (req != NULL) {
1128 gctl_error(req, "Cannot create kernel thread "
1129 "for %s (error=%d).", bpp->name, error);
1130 } else {
1131 G_ELI_DEBUG(1, "Cannot create kernel thread "
1132 "for %s (error=%d).", bpp->name, error);
1133 }
1134 goto failed;
1135 }
1136 LIST_INSERT_HEAD(&sc->sc_workers, wr, w_next);
1137 }
1138
1139 /*
1140 * Create decrypted provider.
1141 */
1142 pp = g_new_providerf(gp, "%s%s", bpp->name, G_ELI_SUFFIX);
1143 pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1144 if (g_eli_unmapped_io && CRYPTO_HAS_VMPAGE) {
1145 /*
1146 * On DMAP architectures we can use unmapped I/O. But don't
1147 * use it with data integrity verification. That code hasn't
1148 * been written yet.
1149 */
1150 if ((sc->sc_flags & G_ELI_FLAG_AUTH) == 0)
1151 pp->flags |= G_PF_ACCEPT_UNMAPPED;
1152 }
1153 pp->mediasize = sc->sc_mediasize;
1154 pp->sectorsize = sc->sc_sectorsize;
1155 LIST_FOREACH(gap, &bpp->aliases, ga_next)
1156 g_provider_add_alias(pp, "%s%s", gap->ga_alias, G_ELI_SUFFIX);
1157
1158 g_error_provider(pp, 0);
1159
1160 G_ELI_DEBUG(0, "Device %s created.", pp->name);
1161 G_ELI_DEBUG(0, "Encryption: %s %u", g_eli_algo2str(sc->sc_ealgo),
1162 sc->sc_ekeylen);
1163 if (sc->sc_flags & G_ELI_FLAG_AUTH)
1164 G_ELI_DEBUG(0, " Integrity: %s", g_eli_algo2str(sc->sc_aalgo));
1165 G_ELI_DEBUG(0, " Crypto: %s",
1166 sc->sc_crypto == G_ELI_CRYPTO_SW_ACCEL ? "accelerated software" :
1167 sc->sc_crypto == G_ELI_CRYPTO_SW ? "software" : "hardware");
1168 return (gp);
1169
1170 failed:
1171 mtx_lock(&sc->sc_queue_mtx);
1172 sc->sc_flags |= G_ELI_FLAG_DESTROY;
1173 wakeup(sc);
1174 /*
1175 * Wait for kernel threads self destruction.
1176 */
1177 while (!LIST_EMPTY(&sc->sc_workers)) {
1178 msleep(&sc->sc_workers, &sc->sc_queue_mtx, PRIBIO,
1179 "geli:destroy", 0);
1180 }
1181 mtx_destroy(&sc->sc_queue_mtx);
1182 if (cp->provider != NULL) {
1183 if (cp->acr == 1)
1184 g_access(cp, -1, -dcw, -1);
1185 g_detach(cp);
1186 }
1187 g_destroy_consumer(cp);
1188 g_destroy_geom(gp);
1189 g_eli_key_destroy(sc);
1190 g_eli_fini_uma();
1191 zfree(sc, M_ELI);
1192 return (NULL);
1193 }
1194
1195 int
g_eli_destroy(struct g_eli_softc * sc,boolean_t force)1196 g_eli_destroy(struct g_eli_softc *sc, boolean_t force)
1197 {
1198 struct g_geom *gp;
1199 struct g_provider *pp;
1200
1201 g_topology_assert();
1202
1203 if (sc == NULL)
1204 return (ENXIO);
1205
1206 gp = sc->sc_geom;
1207 pp = LIST_FIRST(&gp->provider);
1208 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
1209 if (force) {
1210 G_ELI_DEBUG(1, "Device %s is still open, so it "
1211 "cannot be definitely removed.", pp->name);
1212 sc->sc_flags |= G_ELI_FLAG_RW_DETACH;
1213 gp->access = g_eli_access;
1214 g_wither_provider(pp, ENXIO);
1215 return (EBUSY);
1216 } else {
1217 G_ELI_DEBUG(1,
1218 "Device %s is still open (r%dw%de%d).", pp->name,
1219 pp->acr, pp->acw, pp->ace);
1220 return (EBUSY);
1221 }
1222 }
1223
1224 mtx_lock(&sc->sc_queue_mtx);
1225 sc->sc_flags |= G_ELI_FLAG_DESTROY;
1226 wakeup(sc);
1227 while (!LIST_EMPTY(&sc->sc_workers)) {
1228 msleep(&sc->sc_workers, &sc->sc_queue_mtx, PRIBIO,
1229 "geli:destroy", 0);
1230 }
1231 mtx_destroy(&sc->sc_queue_mtx);
1232 gp->softc = NULL;
1233 g_eli_key_destroy(sc);
1234 g_eli_fini_uma();
1235 zfree(sc, M_ELI);
1236
1237 G_ELI_DEBUG(0, "Device %s destroyed.", gp->name);
1238 g_wither_geom_close(gp, ENXIO);
1239
1240 return (0);
1241 }
1242
1243 static int
g_eli_destroy_geom(struct gctl_req * req __unused,struct g_class * mp __unused,struct g_geom * gp)1244 g_eli_destroy_geom(struct gctl_req *req __unused,
1245 struct g_class *mp __unused, struct g_geom *gp)
1246 {
1247 struct g_eli_softc *sc;
1248
1249 sc = gp->softc;
1250 return (g_eli_destroy(sc, FALSE));
1251 }
1252
1253 static int
g_eli_keyfiles_load(struct hmac_ctx * ctx,const char * provider)1254 g_eli_keyfiles_load(struct hmac_ctx *ctx, const char *provider)
1255 {
1256 u_char *keyfile, *data;
1257 char *file, name[64];
1258 size_t size;
1259 int i;
1260
1261 for (i = 0; ; i++) {
1262 snprintf(name, sizeof(name), "%s:geli_keyfile%d", provider, i);
1263 keyfile = preload_search_by_type(name);
1264 if (keyfile == NULL && i == 0) {
1265 /*
1266 * If there is only one keyfile, allow simpler name.
1267 */
1268 snprintf(name, sizeof(name), "%s:geli_keyfile", provider);
1269 keyfile = preload_search_by_type(name);
1270 }
1271 if (keyfile == NULL)
1272 return (i); /* Return number of loaded keyfiles. */
1273 data = preload_fetch_addr(keyfile);
1274 if (data == NULL) {
1275 G_ELI_DEBUG(0, "Cannot find key file data for %s.",
1276 name);
1277 return (0);
1278 }
1279 size = preload_fetch_size(keyfile);
1280 if (size == 0) {
1281 G_ELI_DEBUG(0, "Cannot find key file size for %s.",
1282 name);
1283 return (0);
1284 }
1285 file = preload_search_info(keyfile, MODINFO_NAME);
1286 if (file == NULL) {
1287 G_ELI_DEBUG(0, "Cannot find key file name for %s.",
1288 name);
1289 return (0);
1290 }
1291 G_ELI_DEBUG(1, "Loaded keyfile %s for %s (type: %s).", file,
1292 provider, name);
1293 g_eli_crypto_hmac_update(ctx, data, size);
1294 }
1295 }
1296
1297 static void
g_eli_keyfiles_clear(const char * provider)1298 g_eli_keyfiles_clear(const char *provider)
1299 {
1300 u_char *keyfile, *data;
1301 char name[64];
1302 size_t size;
1303 int i;
1304
1305 for (i = 0; ; i++) {
1306 snprintf(name, sizeof(name), "%s:geli_keyfile%d", provider, i);
1307 keyfile = preload_search_by_type(name);
1308 if (keyfile == NULL)
1309 return;
1310 data = preload_fetch_addr(keyfile);
1311 size = preload_fetch_size(keyfile);
1312 if (data != NULL && size != 0)
1313 explicit_bzero(data, size);
1314 }
1315 }
1316
1317 /*
1318 * Tasting is only made on boot.
1319 * We detect providers which should be attached before root is mounted.
1320 */
1321 static struct g_geom *
g_eli_taste(struct g_class * mp,struct g_provider * pp,int flags __unused)1322 g_eli_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1323 {
1324 struct g_eli_metadata md;
1325 struct g_geom *gp;
1326 struct hmac_ctx ctx;
1327 char passphrase[256];
1328 u_char key[G_ELI_USERKEYLEN], mkey[G_ELI_DATAIVKEYLEN];
1329 u_int i, nkey, nkeyfiles, tries, showpass;
1330 int error;
1331 struct keybuf *keybuf;
1332
1333 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
1334 g_topology_assert();
1335
1336 if (root_mounted() || g_eli_tries == 0)
1337 return (NULL);
1338
1339 G_ELI_DEBUG(3, "Tasting %s.", pp->name);
1340
1341 error = g_eli_read_metadata(mp, pp, &md);
1342 if (error != 0)
1343 return (NULL);
1344 gp = NULL;
1345
1346 if (strcmp(md.md_magic, G_ELI_MAGIC) != 0)
1347 return (NULL);
1348 if (md.md_version > G_ELI_VERSION) {
1349 printf("geom_eli.ko module is too old to handle %s.\n",
1350 pp->name);
1351 return (NULL);
1352 }
1353 if (md.md_provsize != pp->mediasize)
1354 return (NULL);
1355 /* Should we attach it on boot? */
1356 if (!(md.md_flags & G_ELI_FLAG_BOOT) &&
1357 !(md.md_flags & G_ELI_FLAG_GELIBOOT))
1358 return (NULL);
1359 if (md.md_keys == 0x00) {
1360 G_ELI_DEBUG(0, "No valid keys on %s.", pp->name);
1361 return (NULL);
1362 }
1363 if (!eli_metadata_crypto_supported(&md)) {
1364 G_ELI_DEBUG(0, "%s uses invalid or unsupported algorithms\n",
1365 pp->name);
1366 return (NULL);
1367 }
1368 if (md.md_iterations == -1) {
1369 /* If there is no passphrase, we try only once. */
1370 tries = 1;
1371 } else {
1372 /* Ask for the passphrase no more than g_eli_tries times. */
1373 tries = g_eli_tries;
1374 }
1375
1376 if ((keybuf = get_keybuf()) != NULL) {
1377 /* Scan the key buffer, try all GELI keys. */
1378 for (i = 0; i < keybuf->kb_nents; i++) {
1379 if (keybuf->kb_ents[i].ke_type == KEYBUF_TYPE_GELI) {
1380 memcpy(key, keybuf->kb_ents[i].ke_data,
1381 sizeof(key));
1382
1383 if (g_eli_mkey_decrypt_any(&md, key,
1384 mkey, &nkey) == 0 ) {
1385 explicit_bzero(key, sizeof(key));
1386 goto have_key;
1387 }
1388 }
1389 }
1390 }
1391
1392 for (i = 0; i <= tries; i++) {
1393 g_eli_crypto_hmac_init(&ctx, NULL, 0);
1394
1395 /*
1396 * Load all key files.
1397 */
1398 nkeyfiles = g_eli_keyfiles_load(&ctx, pp->name);
1399
1400 if (nkeyfiles == 0 && md.md_iterations == -1) {
1401 /*
1402 * No key files and no passphrase, something is
1403 * definitely wrong here.
1404 * geli(8) doesn't allow for such situation, so assume
1405 * that there was really no passphrase and in that case
1406 * key files are no properly defined in loader.conf.
1407 */
1408 G_ELI_DEBUG(0,
1409 "Found no key files in loader.conf for %s.",
1410 pp->name);
1411 return (NULL);
1412 }
1413
1414 /* Ask for the passphrase if defined. */
1415 if (md.md_iterations >= 0) {
1416 /* Try first with cached passphrase. */
1417 if (i == 0) {
1418 if (!g_eli_boot_passcache)
1419 continue;
1420 memcpy(passphrase, cached_passphrase,
1421 sizeof(passphrase));
1422 } else {
1423 printf("Enter passphrase for %s: ", pp->name);
1424 showpass = g_eli_visible_passphrase;
1425 if ((md.md_flags & G_ELI_FLAG_GELIDISPLAYPASS) != 0)
1426 showpass = GETS_ECHOPASS;
1427 cngets(passphrase, sizeof(passphrase),
1428 showpass);
1429 memcpy(cached_passphrase, passphrase,
1430 sizeof(passphrase));
1431 }
1432 }
1433
1434 /*
1435 * Prepare Derived-Key from the user passphrase.
1436 */
1437 if (md.md_iterations == 0) {
1438 g_eli_crypto_hmac_update(&ctx, md.md_salt,
1439 sizeof(md.md_salt));
1440 g_eli_crypto_hmac_update(&ctx, passphrase,
1441 strlen(passphrase));
1442 explicit_bzero(passphrase, sizeof(passphrase));
1443 } else if (md.md_iterations > 0) {
1444 u_char dkey[G_ELI_USERKEYLEN];
1445
1446 pkcs5v2_genkey(dkey, sizeof(dkey), md.md_salt,
1447 sizeof(md.md_salt), passphrase, md.md_iterations);
1448 explicit_bzero(passphrase, sizeof(passphrase));
1449 g_eli_crypto_hmac_update(&ctx, dkey, sizeof(dkey));
1450 explicit_bzero(dkey, sizeof(dkey));
1451 }
1452
1453 g_eli_crypto_hmac_final(&ctx, key, 0);
1454
1455 /*
1456 * Decrypt Master-Key.
1457 */
1458 error = g_eli_mkey_decrypt_any(&md, key, mkey, &nkey);
1459 explicit_bzero(key, sizeof(key));
1460 if (error == -1) {
1461 if (i == tries) {
1462 G_ELI_DEBUG(0,
1463 "Wrong key for %s. No tries left.",
1464 pp->name);
1465 g_eli_keyfiles_clear(pp->name);
1466 return (NULL);
1467 }
1468 if (i > 0) {
1469 G_ELI_DEBUG(0,
1470 "Wrong key for %s. Tries left: %u.",
1471 pp->name, tries - i);
1472 }
1473 /* Try again. */
1474 continue;
1475 } else if (error > 0) {
1476 G_ELI_DEBUG(0,
1477 "Cannot decrypt Master Key for %s (error=%d).",
1478 pp->name, error);
1479 g_eli_keyfiles_clear(pp->name);
1480 return (NULL);
1481 }
1482 g_eli_keyfiles_clear(pp->name);
1483 G_ELI_DEBUG(1, "Using Master Key %u for %s.", nkey, pp->name);
1484 break;
1485 }
1486 have_key:
1487
1488 /*
1489 * We have correct key, let's attach provider.
1490 */
1491 gp = g_eli_create(NULL, mp, pp, &md, mkey, nkey);
1492 explicit_bzero(mkey, sizeof(mkey));
1493 explicit_bzero(&md, sizeof(md));
1494 if (gp == NULL) {
1495 G_ELI_DEBUG(0, "Cannot create device %s%s.", pp->name,
1496 G_ELI_SUFFIX);
1497 return (NULL);
1498 }
1499 return (gp);
1500 }
1501
1502 static void
g_eli_dumpconf(struct sbuf * sb,const char * indent,struct g_geom * gp,struct g_consumer * cp,struct g_provider * pp)1503 g_eli_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1504 struct g_consumer *cp, struct g_provider *pp)
1505 {
1506 struct g_eli_softc *sc;
1507
1508 g_topology_assert();
1509 sc = gp->softc;
1510 if (sc == NULL)
1511 return;
1512 if (pp != NULL || cp != NULL)
1513 return; /* Nothing here. */
1514
1515 sbuf_printf(sb, "%s<KeysTotal>%ju</KeysTotal>\n", indent,
1516 (uintmax_t)sc->sc_ekeys_total);
1517 sbuf_printf(sb, "%s<KeysAllocated>%ju</KeysAllocated>\n", indent,
1518 (uintmax_t)sc->sc_ekeys_allocated);
1519 sbuf_printf(sb, "%s<Flags>", indent);
1520 if (sc->sc_flags == 0)
1521 sbuf_cat(sb, "NONE");
1522 else {
1523 int first = 1;
1524
1525 #define ADD_FLAG(flag, name) do { \
1526 if (sc->sc_flags & (flag)) { \
1527 if (!first) \
1528 sbuf_cat(sb, ", "); \
1529 else \
1530 first = 0; \
1531 sbuf_cat(sb, name); \
1532 } \
1533 } while (0)
1534 ADD_FLAG(G_ELI_FLAG_SUSPEND, "SUSPEND");
1535 ADD_FLAG(G_ELI_FLAG_SINGLE_KEY, "SINGLE-KEY");
1536 ADD_FLAG(G_ELI_FLAG_NATIVE_BYTE_ORDER, "NATIVE-BYTE-ORDER");
1537 ADD_FLAG(G_ELI_FLAG_ONETIME, "ONETIME");
1538 ADD_FLAG(G_ELI_FLAG_BOOT, "BOOT");
1539 ADD_FLAG(G_ELI_FLAG_WO_DETACH, "W-DETACH");
1540 ADD_FLAG(G_ELI_FLAG_RW_DETACH, "RW-DETACH");
1541 ADD_FLAG(G_ELI_FLAG_AUTH, "AUTH");
1542 ADD_FLAG(G_ELI_FLAG_WOPEN, "W-OPEN");
1543 ADD_FLAG(G_ELI_FLAG_DESTROY, "DESTROY");
1544 ADD_FLAG(G_ELI_FLAG_RO, "READ-ONLY");
1545 ADD_FLAG(G_ELI_FLAG_NODELETE, "NODELETE");
1546 ADD_FLAG(G_ELI_FLAG_GELIBOOT, "GELIBOOT");
1547 ADD_FLAG(G_ELI_FLAG_GELIDISPLAYPASS, "GELIDISPLAYPASS");
1548 ADD_FLAG(G_ELI_FLAG_AUTORESIZE, "AUTORESIZE");
1549 #undef ADD_FLAG
1550 }
1551 sbuf_cat(sb, "</Flags>\n");
1552
1553 if (!(sc->sc_flags & G_ELI_FLAG_ONETIME)) {
1554 sbuf_printf(sb, "%s<UsedKey>%u</UsedKey>\n", indent,
1555 sc->sc_nkey);
1556 }
1557 sbuf_printf(sb, "%s<Version>%u</Version>\n", indent, sc->sc_version);
1558 sbuf_printf(sb, "%s<Crypto>", indent);
1559 switch (sc->sc_crypto) {
1560 case G_ELI_CRYPTO_HW:
1561 sbuf_cat(sb, "hardware");
1562 break;
1563 case G_ELI_CRYPTO_SW:
1564 sbuf_cat(sb, "software");
1565 break;
1566 case G_ELI_CRYPTO_SW_ACCEL:
1567 sbuf_cat(sb, "accelerated software");
1568 break;
1569 default:
1570 sbuf_cat(sb, "UNKNOWN");
1571 break;
1572 }
1573 sbuf_cat(sb, "</Crypto>\n");
1574 if (sc->sc_flags & G_ELI_FLAG_AUTH) {
1575 sbuf_printf(sb,
1576 "%s<AuthenticationAlgorithm>%s</AuthenticationAlgorithm>\n",
1577 indent, g_eli_algo2str(sc->sc_aalgo));
1578 }
1579 sbuf_printf(sb, "%s<KeyLength>%u</KeyLength>\n", indent,
1580 sc->sc_ekeylen);
1581 sbuf_printf(sb, "%s<EncryptionAlgorithm>%s</EncryptionAlgorithm>\n",
1582 indent, g_eli_algo2str(sc->sc_ealgo));
1583 sbuf_printf(sb, "%s<State>%s</State>\n", indent,
1584 (sc->sc_flags & G_ELI_FLAG_SUSPEND) ? "SUSPENDED" : "ACTIVE");
1585 }
1586
1587 static void
g_eli_shutdown_pre_sync(void * arg,int howto)1588 g_eli_shutdown_pre_sync(void *arg, int howto)
1589 {
1590 struct g_class *mp;
1591 struct g_geom *gp, *gp2;
1592 struct g_provider *pp;
1593 struct g_eli_softc *sc;
1594
1595 mp = arg;
1596 g_topology_lock();
1597 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
1598 sc = gp->softc;
1599 if (sc == NULL)
1600 continue;
1601 pp = LIST_FIRST(&gp->provider);
1602 KASSERT(pp != NULL, ("No provider? gp=%p (%s)", gp, gp->name));
1603 if (pp->acr != 0 || pp->acw != 0 || pp->ace != 0 ||
1604 SCHEDULER_STOPPED())
1605 {
1606 sc->sc_flags |= G_ELI_FLAG_RW_DETACH;
1607 gp->access = g_eli_access;
1608 } else {
1609 (void) g_eli_destroy(sc, TRUE);
1610 }
1611 }
1612 g_topology_unlock();
1613 }
1614
1615 static void
g_eli_init(struct g_class * mp)1616 g_eli_init(struct g_class *mp)
1617 {
1618
1619 g_eli_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync,
1620 g_eli_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST);
1621 if (g_eli_pre_sync == NULL)
1622 G_ELI_DEBUG(0, "Warning! Cannot register shutdown event.");
1623 }
1624
1625 static void
g_eli_fini(struct g_class * mp)1626 g_eli_fini(struct g_class *mp)
1627 {
1628
1629 if (g_eli_pre_sync != NULL)
1630 EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_eli_pre_sync);
1631 }
1632
1633 DECLARE_GEOM_CLASS(g_eli_class, g_eli);
1634 MODULE_DEPEND(g_eli, crypto, 1, 1, 1);
1635 MODULE_VERSION(geom_eli, 0);
1636