2 * Copyright (c) 2002 Poul-Henning Kamp
3 * Copyright (c) 2002 Networks Associates Technology, Inc.
6 * This software was developed for the FreeBSD Project by Poul-Henning Kamp
7 * and NAI Labs, the Security Research Division of Network Associates, Inc.
8 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9 * DARPA CHATS research program.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The names of the authors may not be used to endorse or promote
20 * products derived from this software without specific prior written
23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * This is the method for dealing with BSD disklabels. It has been
38 * extensively (by my standards at least) commented, in the vain hope that
39 * it will serve as the source in future copy&paste operations.
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
45 #include <sys/param.h>
46 #include <sys/endian.h>
47 #include <sys/systm.h>
48 #include <sys/sysctl.h>
49 #include <sys/kernel.h>
50 #include <sys/fcntl.h>
53 #include <sys/malloc.h>
55 #include <sys/mutex.h>
57 #include <sys/errno.h>
58 #include <sys/disklabel.h>
63 #include <geom/geom.h>
64 #include <geom/geom_slice.h>
66 FEATURE(geom_bsd, "GEOM BSD disklabels support");
68 #define BSD_CLASS_NAME "BSD"
70 #define ALPHA_LABEL_OFFSET 64
71 #define HISTORIC_LABEL_OFFSET 512
73 #define LABELSIZE (148 + 16 * MAXPARTITIONS)
75 static void g_bsd_hotwrite(void *arg, int flag);
77 * Our private data about one instance. All the rest is handled by the
78 * slice code and stored in its softc, so this is just the stuff
79 * specific to BSD disklabels.
85 struct disklabel ondisk;
86 u_char label[LABELSIZE];
91 * Modify our slicer to match proposed disklabel, if possible.
92 * This is where we make sure we don't do something stupid.
95 g_bsd_modify(struct g_geom *gp, u_char *label)
98 struct partition *ppp;
100 struct g_consumer *cp;
101 struct g_bsd_softc *ms;
111 error = bsd_disklabel_le_dec(label, &dl, MAXPARTITIONS);
116 /* Get dimensions of our device. */
117 cp = LIST_FIRST(&gp->consumer);
118 secsize = cp->provider->sectorsize;
120 /* ... or a smaller sector size. */
121 if (dl.d_secsize < secsize) {
125 /* ... or a non-multiple sector size. */
126 if (dl.d_secsize % secsize != 0) {
130 /* Historical braindamage... */
131 rawoffset = (off_t)dl.d_partitions[RAW_PART].p_offset * dl.d_secsize;
133 for (i = 0; i < dl.d_npartitions; i++) {
134 ppp = &dl.d_partitions[i];
135 if (ppp->p_size == 0)
137 o = (off_t)ppp->p_offset * dl.d_secsize;
143 if (rawoffset != 0 && (off_t)rawoffset != ms->mbroffset)
144 printf("WARNING: %s expected rawoffset %jd, found %jd\n",
146 (intmax_t)ms->mbroffset/dl.d_secsize,
147 (intmax_t)rawoffset/dl.d_secsize);
149 /* Don't munge open partitions. */
150 for (i = 0; i < dl.d_npartitions; i++) {
151 ppp = &dl.d_partitions[i];
153 o = (off_t)ppp->p_offset * dl.d_secsize;
156 error = g_slice_config(gp, i, G_SLICE_CONFIG_CHECK,
158 (off_t)ppp->p_size * dl.d_secsize,
160 "%s%c", gp->name, 'a' + i);
165 /* Look good, go for it... */
166 for (u = 0; u < gsp->nslice; u++) {
167 ppp = &dl.d_partitions[u];
168 o = (off_t)ppp->p_offset * dl.d_secsize;
171 g_slice_config(gp, u, G_SLICE_CONFIG_SET,
173 (off_t)ppp->p_size * dl.d_secsize,
175 "%s%c", gp->name, 'a' + u);
178 /* Update our softc */
180 if (label != ms->label)
181 bcopy(label, ms->label, LABELSIZE);
182 ms->rawoffset = rawoffset;
185 * In order to avoid recursively attaching to the same
186 * on-disk label (it's usually visible through the 'c'
187 * partition) we calculate an MD5 and ask if other BSD's
188 * below us love that label. If they do, we don't.
191 MD5Update(&md5sum, ms->label, sizeof(ms->label));
192 MD5Final(ms->labelsum, &md5sum);
198 * This is an internal helper function, called multiple times from the taste
199 * function to try to locate a disklabel on the disk. More civilized formats
200 * will not need this, as there is only one possible place on disk to look
201 * for the magic spot.
205 g_bsd_try(struct g_geom *gp, struct g_slicer *gsp, struct g_consumer *cp, int secsize, struct g_bsd_softc *ms, off_t offset)
209 struct disklabel *dl;
213 * We need to read entire aligned sectors, and we assume that the
214 * disklabel does not span sectors, so one sector is enough.
216 secoff = offset % secsize;
217 buf = g_read_data(cp, offset - secoff, secsize, NULL);
221 /* Decode into our native format. */
223 error = bsd_disklabel_le_dec(buf + secoff, dl, MAXPARTITIONS);
225 bcopy(buf + secoff, ms->label, LABELSIZE);
227 /* Remember to free the buffer g_read_data() gave us. */
230 ms->labeloffset = offset;
235 * This function writes the current label to disk, possibly updating
236 * the alpha SRM checksum.
240 g_bsd_writelabel(struct g_geom *gp, u_char *bootcode)
244 struct g_consumer *cp;
245 struct g_slicer *gsp;
246 struct g_bsd_softc *ms;
253 cp = LIST_FIRST(&gp->consumer);
254 /* Get sector size, we need it to read data. */
255 secsize = cp->provider->sectorsize;
256 secoff = ms->labeloffset % secsize;
257 if (bootcode == NULL) {
258 buf = g_read_data(cp, ms->labeloffset - secoff, secsize, &error);
261 bcopy(ms->label, buf + secoff, sizeof(ms->label));
264 bcopy(ms->label, buf + ms->labeloffset, sizeof(ms->label));
266 if (ms->labeloffset == ALPHA_LABEL_OFFSET) {
268 for (i = 0; i < 63; i++)
269 sum += le64dec(buf + i * 8);
270 le64enc(buf + 504, sum);
272 if (bootcode == NULL) {
273 error = g_write_data(cp, ms->labeloffset - secoff, buf, secsize);
276 error = g_write_data(cp, 0, bootcode, BBSIZE);
282 * If the user tries to overwrite our disklabel through an open partition
283 * or via a magicwrite config call, we end up here and try to prevent
284 * footshooting as best we can.
287 g_bsd_hotwrite(void *arg, int flag)
291 struct g_slicer *gsp;
293 struct g_bsd_softc *ms;
299 * We should never get canceled, because that would amount to a removal
300 * of the geom while there was outstanding I/O requests.
302 KASSERT(flag != EV_CANCEL, ("g_bsd_hotwrite cancelled"));
304 gp = bp->bio_to->geom;
307 gsl = &gsp->slices[bp->bio_to->index];
308 p = (u_char*)bp->bio_data + ms->labeloffset -
309 (bp->bio_offset + gsl->offset);
310 error = g_bsd_modify(gp, p);
312 g_io_deliver(bp, EPERM);
315 g_slice_finish_hot(bp);
319 g_bsd_start(struct bio *bp)
322 struct g_bsd_softc *ms;
323 struct g_slicer *gsp;
325 gp = bp->bio_to->geom;
328 if (bp->bio_cmd == BIO_GETATTR) {
329 if (g_handleattr(bp, "BSD::labelsum", ms->labelsum,
330 sizeof(ms->labelsum)))
337 * Dump configuration information in XML format.
338 * Notice that the function is called once for the geom and once for each
339 * consumer and provider. We let g_slice_dumpconf() do most of the work.
342 g_bsd_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
344 struct g_bsd_softc *ms;
345 struct g_slicer *gsp;
349 g_slice_dumpconf(sb, indent, gp, cp, pp);
350 if (indent != NULL && pp == NULL && cp == NULL) {
351 sbuf_printf(sb, "%s<labeloffset>%jd</labeloffset>\n",
352 indent, (intmax_t)ms->labeloffset);
353 sbuf_printf(sb, "%s<rawoffset>%jd</rawoffset>\n",
354 indent, (intmax_t)ms->rawoffset);
355 sbuf_printf(sb, "%s<mbroffset>%jd</mbroffset>\n",
356 indent, (intmax_t)ms->mbroffset);
357 } else if (pp != NULL) {
359 sbuf_printf(sb, " ty %d",
360 ms->ondisk.d_partitions[pp->index].p_fstype);
362 sbuf_printf(sb, "%s<type>%d</type>\n", indent,
363 ms->ondisk.d_partitions[pp->index].p_fstype);
368 * The taste function is called from the event-handler, with the topology
369 * lock already held and a provider to examine. The flags are unused.
371 * If flags == G_TF_NORMAL, the idea is to take a bite of the provider and
372 * if we find valid, consistent magic on it, build a geom on it.
374 * There may be cases where the operator would like to put a BSD-geom on
375 * providers which do not meet all of the requirements. This can be done
376 * by instead passing the G_TF_INSIST flag, which will override these
379 * The final flags value is G_TF_TRANSPARENT, which instructs the method
380 * to put a geom on top of the provider and configure it to be as transparent
381 * as possible. This is not really relevant to the BSD method and therefore
382 * not implemented here.
385 static struct uuid freebsd_slice = GPT_ENT_TYPE_FREEBSD;
387 static struct g_geom *
388 g_bsd_taste(struct g_class *mp, struct g_provider *pp, int flags)
391 struct g_consumer *cp;
393 struct g_bsd_softc *ms;
395 struct g_slicer *gsp;
400 g_trace(G_T_TOPOLOGY, "bsd_taste(%s,%s)", mp->name, pp->name);
403 /* We don't implement transparent inserts. */
404 if (flags == G_TF_TRANSPARENT)
408 * BSD labels are a subclass of the general "slicing" topology so
409 * a lot of the work can be done by the common "slice" code.
410 * Create a geom with space for MAXPARTITIONS providers, one consumer
411 * and a softc structure for us. Specify the provider to attach
412 * the consumer to and our "start" routine for special requests.
413 * The provider is opened with mode (1,0,0) so we can do reads
416 gp = g_slice_new(mp, MAXPARTITIONS, pp, &cp, &ms,
417 sizeof(*ms), g_bsd_start);
421 /* Get the geom_slicer softc from the geom. */
425 * The do...while loop here allows us to have multiple escapes
426 * using a simple "break". This improves code clarity without
427 * ending up in deep nesting and without using goto or come from.
431 * If the provider is an MBR we will only auto attach
432 * to type 165 slices in the G_TF_NORMAL case. We will
433 * attach to any other type.
435 error = g_getattr("MBR::type", cp, &i);
437 if (i != 165 && flags == G_TF_NORMAL)
439 error = g_getattr("MBR::offset", cp, &ms->mbroffset);
444 /* Same thing if we are inside a PC98 */
445 error = g_getattr("PC98::type", cp, &i);
447 if (i != 0xc494 && flags == G_TF_NORMAL)
449 error = g_getattr("PC98::offset", cp, &ms->mbroffset);
454 /* Same thing if we are inside a GPT */
455 error = g_getattr("GPT::type", cp, &uuid);
457 if (memcmp(&uuid, &freebsd_slice, sizeof(uuid)) != 0 &&
458 flags == G_TF_NORMAL)
462 /* Get sector size, we need it to read data. */
463 secsize = cp->provider->sectorsize;
467 /* First look for a label at the start of the second sector. */
468 error = g_bsd_try(gp, gsp, cp, secsize, ms, secsize);
471 * If sector size is not 512 the label still can be at
472 * offset 512, not at the start of the second sector. At least
473 * it's true for labels created by the FreeBSD's bsdlabel(8).
475 if (error && secsize != HISTORIC_LABEL_OFFSET)
476 error = g_bsd_try(gp, gsp, cp, secsize, ms,
477 HISTORIC_LABEL_OFFSET);
479 /* Next, look for alpha labels */
481 error = g_bsd_try(gp, gsp, cp, secsize, ms,
484 /* If we didn't find a label, punt. */
489 * In order to avoid recursively attaching to the same
490 * on-disk label (it's usually visible through the 'c'
491 * partition) we calculate an MD5 and ask if other BSD's
492 * below us love that label. If they do, we don't.
495 MD5Update(&md5sum, ms->label, sizeof(ms->label));
496 MD5Final(ms->labelsum, &md5sum);
498 error = g_getattr("BSD::labelsum", cp, &hash);
499 if (!error && !bcmp(ms->labelsum, hash, sizeof(hash)))
503 * Process the found disklabel, and modify our "slice"
504 * instance to match it, if possible.
506 error = g_bsd_modify(gp, ms->label);
509 /* Success or failure, we can close our provider now. */
510 g_access(cp, -1, 0, 0);
512 /* If we have configured any providers, return the new geom. */
513 if (gsp->nprovider > 0) {
514 g_slice_conf_hot(gp, 0, ms->labeloffset, LABELSIZE,
515 G_SLICE_HOT_ALLOW, G_SLICE_HOT_DENY, G_SLICE_HOT_CALL);
516 gsp->hot = g_bsd_hotwrite;
520 * ...else push the "self-destruct" button, by spoiling our own
521 * consumer. This triggers a call to g_slice_spoiled which will
522 * dismantle what was setup.
530 struct g_bsd_softc *ms;
536 g_bsd_callconfig(void *arg, int flag)
541 hp->error = g_bsd_modify(hp->gp, hp->label);
543 hp->error = g_bsd_writelabel(hp->gp, NULL);
547 * NB! curthread is user process which GCTL'ed.
550 g_bsd_config(struct gctl_req *req, struct g_class *mp, char const *verb)
556 struct g_slicer *gsp;
557 struct g_consumer *cp;
558 struct g_bsd_softc *ms;
561 gp = gctl_get_geom(req, mp, "geom");
564 cp = LIST_FIRST(&gp->consumer);
567 if (!strcmp(verb, "read mbroffset")) {
568 gctl_set_param_err(req, "mbroffset", &ms->mbroffset,
569 sizeof(ms->mbroffset));
571 } else if (!strcmp(verb, "write label")) {
572 label = gctl_get_paraml(req, "label", LABELSIZE);
576 h0h0.ms = gsp->softc;
579 /* XXX: Does this reference register with our selfdestruct code ? */
580 error = g_access(cp, 1, 1, 1);
582 gctl_error(req, "could not access consumer");
585 g_bsd_callconfig(&h0h0, 0);
587 g_access(cp, -1, -1, -1);
588 } else if (!strcmp(verb, "write bootcode")) {
589 label = gctl_get_paraml(req, "bootcode", BBSIZE);
592 /* XXX: Does this reference register with our selfdestruct code ? */
593 error = g_access(cp, 1, 1, 1);
595 gctl_error(req, "could not access consumer");
598 error = g_bsd_writelabel(gp, label);
599 g_access(cp, -1, -1, -1);
601 gctl_error(req, "Unknown verb parameter");
607 /* Finally, register with GEOM infrastructure. */
608 static struct g_class g_bsd_class = {
609 .name = BSD_CLASS_NAME,
610 .version = G_VERSION,
611 .taste = g_bsd_taste,
612 .ctlreq = g_bsd_config,
613 .dumpconf = g_bsd_dumpconf,
616 DECLARE_GEOM_CLASS(g_bsd_class, g_bsd);