2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2002 Poul-Henning Kamp
5 * Copyright (c) 2002 Networks Associates Technology, Inc.
8 * This software was developed for the FreeBSD Project by Poul-Henning Kamp
9 * and NAI Labs, the Security Research Division of Network Associates, Inc.
10 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
11 * DARPA CHATS research program.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. The names of the authors may not be used to endorse or promote
22 * products derived from this software without specific prior written
25 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/devicestat.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
49 #include <sys/sysctl.h>
51 #include <sys/kthread.h>
53 #include <sys/mutex.h>
54 #include <sys/errno.h>
57 #include <geom/geom.h>
58 #include <geom/geom_dbg.h>
59 #include <geom/geom_int.h>
60 #include <machine/stdarg.h>
70 SDT_PROVIDER_DEFINE(geom);
72 struct class_list_head g_classes = LIST_HEAD_INITIALIZER(g_classes);
73 static struct g_tailq_head geoms = TAILQ_HEAD_INITIALIZER(geoms);
74 char *g_wait_event, *g_wait_up, *g_wait_down, *g_wait_sim;
78 struct g_provider *pp;
85 g_dbg_printf(const char *classname, int lvl, struct bio *bp,
89 #ifndef PRINTF_BUFR_SIZE
90 #define PRINTF_BUFR_SIZE 64
92 char bufr[PRINTF_BUFR_SIZE];
93 struct sbuf sb, *sbp __unused;
96 sbp = sbuf_new(&sb, bufr, sizeof(bufr), SBUF_FIXEDLEN);
97 KASSERT(sbp != NULL, ("sbuf_new misused?"));
99 sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
101 sbuf_cat(&sb, classname);
103 sbuf_printf(&sb, "[%d]", lvl);
105 va_start(ap, format);
106 sbuf_vprintf(&sb, format, ap);
111 g_format_bio(&sb, bp);
114 /* Terminate the debug line with a single '\n'. */
115 sbuf_nl_terminate(&sb);
117 /* Flush line to printf. */
123 * This event offers a new class a chance to taste all preexisting providers.
126 g_load_class(void *arg, int flag)
129 struct g_class *mp2, *mp;
131 struct g_provider *pp;
134 if (flag == EV_CANCEL) /* XXX: can't happen ? */
146 g_trace(G_T_TOPOLOGY, "g_load_class(%s)", mp->name);
147 KASSERT(mp->name != NULL && *mp->name != '\0',
148 ("GEOM class has no name"));
149 LIST_FOREACH(mp2, &g_classes, class) {
151 printf("The GEOM class %s is already loaded.\n",
156 } else if (strcmp(mp2->name, mp->name) == 0) {
157 printf("A GEOM class %s is already loaded.\n",
165 LIST_INIT(&mp->geom);
166 LIST_INSERT_HEAD(&g_classes, mp, class);
167 if (mp->init != NULL)
169 if (mp->taste == NULL)
171 LIST_FOREACH(mp2, &g_classes, class) {
174 LIST_FOREACH(gp, &mp2->geom, geom) {
175 LIST_FOREACH(pp, &gp->provider, provider) {
176 mp->taste(mp, pp, 0);
184 g_unload_class(struct g_class *mp)
187 struct g_provider *pp;
188 struct g_consumer *cp;
192 g_trace(G_T_TOPOLOGY, "g_unload_class(%s)", mp->name);
195 LIST_FOREACH(gp, &mp->geom, geom) {
196 /* We refuse to unload if anything is open */
197 LIST_FOREACH(pp, &gp->provider, provider)
198 if (pp->acr || pp->acw || pp->ace) {
202 LIST_FOREACH(cp, &gp->consumer, consumer)
203 if (cp->acr || cp->acw || cp->ace) {
207 /* If the geom is withering, wait for it to finish. */
208 if (gp->flags & G_GEOM_WITHER) {
209 g_topology_sleep(mp, 1);
215 * We allow unloading if we have no geoms, or a class
216 * method we can use to get rid of them.
218 if (!LIST_EMPTY(&mp->geom) && mp->destroy_geom == NULL) {
223 /* Bar new entries */
227 LIST_FOREACH(gp, &mp->geom, geom) {
228 error = mp->destroy_geom(NULL, mp, gp);
234 /* Wait for withering to finish. */
236 gp = LIST_FIRST(&mp->geom);
239 KASSERT(gp->flags & G_GEOM_WITHER,
240 ("Non-withering geom in class %s", mp->name));
241 g_topology_sleep(mp, 1);
244 if (mp->fini != NULL)
246 LIST_REMOVE(mp, class);
253 g_modevent(module_t mod, int type, void *data)
257 static int g_ignition;
261 if (mp->version != G_VERSION) {
262 printf("GEOM class %s has Wrong version %x\n",
263 mp->name, mp->version);
273 g_trace(G_T_TOPOLOGY, "g_modevent(%s, LOAD)", mp->name);
274 hh = g_malloc(sizeof *hh, M_WAITOK | M_ZERO);
277 * Once the system is not cold, MOD_LOAD calls will be
278 * from the userland and the g_event thread will be able
279 * to acknowledge their completion.
283 error = g_post_event(g_load_class, hh, M_WAITOK, NULL);
285 error = g_waitfor_event(g_load_class, hh, M_WAITOK,
293 g_trace(G_T_TOPOLOGY, "g_modevent(%s, UNLOAD)", mp->name);
294 error = g_unload_class(mp);
296 KASSERT(LIST_EMPTY(&mp->geom),
297 ("Unloaded class (%s) still has geom", mp->name));
305 g_retaste_event(void *arg, int flag)
307 struct g_class *mp, *mp2;
310 struct g_provider *pp;
311 struct g_consumer *cp;
314 if (flag == EV_CANCEL) /* XXX: can't happen ? */
316 if (g_shutdown || g_notaste)
326 g_trace(G_T_TOPOLOGY, "g_retaste(%s)", mp->name);
328 LIST_FOREACH(mp2, &g_classes, class) {
329 LIST_FOREACH(gp, &mp2->geom, geom) {
330 LIST_FOREACH(pp, &gp->provider, provider) {
331 if (pp->acr || pp->acw || pp->ace)
333 LIST_FOREACH(cp, &pp->consumers, consumers) {
334 if (cp->geom->class == mp &&
335 (cp->flags & G_CF_ORPHAN) == 0)
339 cp->flags |= G_CF_ORPHAN;
340 g_wither_geom(cp->geom, ENXIO);
342 mp->taste(mp, pp, 0);
350 g_retaste(struct g_class *mp)
355 if (mp->taste == NULL)
358 hh = g_malloc(sizeof *hh, M_WAITOK | M_ZERO);
363 error = g_post_event(g_retaste_event, hh, M_WAITOK, NULL);
365 error = g_waitfor_event(g_retaste_event, hh, M_WAITOK, NULL);
375 g_new_geomf(struct g_class *mp, const char *fmt, ...)
383 sb = sbuf_new_auto();
385 sbuf_vprintf(sb, fmt, ap);
388 gp = g_malloc(sizeof *gp, M_WAITOK | M_ZERO);
389 gp->name = g_malloc(sbuf_len(sb) + 1, M_WAITOK | M_ZERO);
392 LIST_INIT(&gp->consumer);
393 LIST_INIT(&gp->provider);
394 LIST_INSERT_HEAD(&mp->geom, gp, geom);
395 TAILQ_INSERT_HEAD(&geoms, gp, geoms);
396 strcpy(gp->name, sbuf_data(sb));
398 /* Fill in defaults from class */
399 gp->start = mp->start;
400 gp->spoiled = mp->spoiled;
401 gp->attrchanged = mp->attrchanged;
402 gp->providergone = mp->providergone;
403 gp->dumpconf = mp->dumpconf;
404 gp->access = mp->access;
405 gp->orphan = mp->orphan;
406 gp->ioctl = mp->ioctl;
407 gp->resize = mp->resize;
412 g_destroy_geom(struct g_geom *gp)
417 g_trace(G_T_TOPOLOGY, "g_destroy_geom(%p(%s))", gp, gp->name);
418 KASSERT(LIST_EMPTY(&gp->consumer),
419 ("g_destroy_geom(%s) with consumer(s) [%p]",
420 gp->name, LIST_FIRST(&gp->consumer)));
421 KASSERT(LIST_EMPTY(&gp->provider),
422 ("g_destroy_geom(%s) with provider(s) [%p]",
423 gp->name, LIST_FIRST(&gp->provider)));
425 LIST_REMOVE(gp, geom);
426 TAILQ_REMOVE(&geoms, gp, geoms);
432 * This function is called (repeatedly) until the geom has withered away.
435 g_wither_geom(struct g_geom *gp, int error)
437 struct g_provider *pp;
441 g_trace(G_T_TOPOLOGY, "g_wither_geom(%p(%s))", gp, gp->name);
442 if (!(gp->flags & G_GEOM_WITHER)) {
443 gp->flags |= G_GEOM_WITHER;
444 LIST_FOREACH(pp, &gp->provider, provider)
445 if (!(pp->flags & G_PF_ORPHAN))
446 g_orphan_provider(pp, error);
452 * Convenience function to destroy a particular provider.
455 g_wither_provider(struct g_provider *pp, int error)
458 pp->flags |= G_PF_WITHER;
459 if (!(pp->flags & G_PF_ORPHAN))
460 g_orphan_provider(pp, error);
464 * This function is called (repeatedly) until the has withered away.
467 g_wither_geom_close(struct g_geom *gp, int error)
469 struct g_consumer *cp;
473 g_trace(G_T_TOPOLOGY, "g_wither_geom_close(%p(%s))", gp, gp->name);
474 LIST_FOREACH(cp, &gp->consumer, consumer)
475 if (cp->acr || cp->acw || cp->ace)
476 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
477 g_wither_geom(gp, error);
481 * This function is called (repeatedly) until we cant wash away more
482 * withered bits at present.
488 struct g_geom *gp, *gp2;
489 struct g_provider *pp, *pp2;
490 struct g_consumer *cp, *cp2;
493 LIST_FOREACH(mp, &g_classes, class) {
494 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
495 LIST_FOREACH_SAFE(pp, &gp->provider, provider, pp2) {
496 if (!(pp->flags & G_PF_WITHER))
498 if (LIST_EMPTY(&pp->consumers))
499 g_destroy_provider(pp);
501 if (!(gp->flags & G_GEOM_WITHER))
503 LIST_FOREACH_SAFE(pp, &gp->provider, provider, pp2) {
504 if (LIST_EMPTY(&pp->consumers))
505 g_destroy_provider(pp);
507 LIST_FOREACH_SAFE(cp, &gp->consumer, consumer, cp2) {
508 if (cp->acr || cp->acw || cp->ace)
510 if (cp->provider != NULL)
512 g_destroy_consumer(cp);
514 if (LIST_EMPTY(&gp->provider) &&
515 LIST_EMPTY(&gp->consumer))
522 g_new_consumer(struct g_geom *gp)
524 struct g_consumer *cp;
528 KASSERT(!(gp->flags & G_GEOM_WITHER),
529 ("g_new_consumer on WITHERing geom(%s) (class %s)",
530 gp->name, gp->class->name));
531 KASSERT(gp->orphan != NULL,
532 ("g_new_consumer on geom(%s) (class %s) without orphan",
533 gp->name, gp->class->name));
535 cp = g_malloc(sizeof *cp, M_WAITOK | M_ZERO);
537 cp->stat = devstat_new_entry(cp, -1, 0, DEVSTAT_ALL_SUPPORTED,
538 DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
539 LIST_INSERT_HEAD(&gp->consumer, cp, consumer);
544 g_destroy_consumer(struct g_consumer *cp)
549 G_VALID_CONSUMER(cp);
550 g_trace(G_T_TOPOLOGY, "g_destroy_consumer(%p)", cp);
551 KASSERT (cp->provider == NULL, ("g_destroy_consumer but attached"));
552 KASSERT (cp->acr == 0, ("g_destroy_consumer with acr"));
553 KASSERT (cp->acw == 0, ("g_destroy_consumer with acw"));
554 KASSERT (cp->ace == 0, ("g_destroy_consumer with ace"));
557 LIST_REMOVE(cp, consumer);
558 devstat_remove_entry(cp->stat);
560 if (gp->flags & G_GEOM_WITHER)
565 g_new_provider_event(void *arg, int flag)
568 struct g_provider *pp;
569 struct g_consumer *cp, *next_cp;
572 if (flag == EV_CANCEL)
577 G_VALID_PROVIDER(pp);
578 KASSERT(!(pp->flags & G_PF_WITHER),
579 ("g_new_provider_event but withered"));
580 LIST_FOREACH_SAFE(cp, &pp->consumers, consumers, next_cp) {
581 if ((cp->flags & G_CF_ORPHAN) == 0 &&
582 cp->geom->attrchanged != NULL)
583 cp->geom->attrchanged(cp, "GEOM::media");
587 LIST_FOREACH(mp, &g_classes, class) {
588 if (mp->taste == NULL)
590 LIST_FOREACH(cp, &pp->consumers, consumers)
591 if (cp->geom->class == mp &&
592 (cp->flags & G_CF_ORPHAN) == 0)
596 mp->taste(mp, pp, 0);
603 g_new_providerf(struct g_geom *gp, const char *fmt, ...)
605 struct g_provider *pp;
611 KASSERT(gp->access != NULL,
612 ("new provider on geom(%s) without ->access (class %s)",
613 gp->name, gp->class->name));
614 KASSERT(gp->start != NULL,
615 ("new provider on geom(%s) without ->start (class %s)",
616 gp->name, gp->class->name));
617 KASSERT(!(gp->flags & G_GEOM_WITHER),
618 ("new provider on WITHERing geom(%s) (class %s)",
619 gp->name, gp->class->name));
620 sb = sbuf_new_auto();
622 sbuf_vprintf(sb, fmt, ap);
625 pp = g_malloc(sizeof *pp + sbuf_len(sb) + 1, M_WAITOK | M_ZERO);
626 pp->name = (char *)(pp + 1);
627 strcpy(pp->name, sbuf_data(sb));
629 LIST_INIT(&pp->consumers);
630 LIST_INIT(&pp->aliases);
633 pp->stat = devstat_new_entry(pp, -1, 0, DEVSTAT_ALL_SUPPORTED,
634 DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
635 LIST_INSERT_HEAD(&gp->provider, pp, provider);
636 g_post_event(g_new_provider_event, pp, M_WAITOK, pp, gp, NULL);
641 g_provider_add_alias(struct g_provider *pp, const char *fmt, ...)
644 struct g_geom_alias *gap;
648 * Generate the alias string and save it in the list.
650 sb = sbuf_new_auto();
652 sbuf_vprintf(sb, fmt, ap);
656 LIST_FOREACH(gap, &pp->aliases, ga_next) {
657 if (strcmp(gap->ga_alias, sbuf_data(sb)) != 0)
659 /* Don't re-add the same alias. */
664 gap = g_malloc(sizeof(*gap) + sbuf_len(sb) + 1, M_WAITOK | M_ZERO);
665 memcpy((char *)(gap + 1), sbuf_data(sb), sbuf_len(sb));
667 gap->ga_alias = (const char *)(gap + 1);
668 LIST_INSERT_HEAD(&pp->aliases, gap, ga_next);
672 g_error_provider(struct g_provider *pp, int error)
675 /* G_VALID_PROVIDER(pp); We may not have g_topology */
680 g_resize_provider_event(void *arg, int flag)
685 struct g_provider *pp;
686 struct g_consumer *cp, *cp2;
698 G_VALID_PROVIDER(pp);
699 KASSERT(!(pp->flags & G_PF_WITHER),
700 ("g_resize_provider_event but withered"));
701 g_trace(G_T_TOPOLOGY, "g_resize_provider_event(%p)", pp);
703 LIST_FOREACH_SAFE(cp, &pp->consumers, consumers, cp2) {
705 if (gp->resize == NULL && size < pp->mediasize) {
707 * XXX: g_dev_orphan method does deferred destroying
708 * and it is possible, that other event could already
709 * call the orphan method. Check consumer's flags to
710 * do not schedule it twice.
712 if (cp->flags & G_CF_ORPHAN)
714 cp->flags |= G_CF_ORPHAN;
715 cp->geom->orphan(cp);
719 pp->mediasize = size;
721 LIST_FOREACH_SAFE(cp, &pp->consumers, consumers, cp2) {
723 if ((gp->flags & G_GEOM_WITHER) == 0 && gp->resize != NULL)
728 * After resizing, the previously invalid GEOM class metadata
729 * might become valid. This means we should retaste.
731 LIST_FOREACH(mp, &g_classes, class) {
732 if (mp->taste == NULL)
734 LIST_FOREACH(cp, &pp->consumers, consumers)
735 if (cp->geom->class == mp &&
736 (cp->flags & G_CF_ORPHAN) == 0)
740 mp->taste(mp, pp, 0);
746 g_resize_provider(struct g_provider *pp, off_t size)
750 G_VALID_PROVIDER(pp);
751 if (pp->flags & G_PF_WITHER)
754 if (size == pp->mediasize)
757 hh = g_malloc(sizeof *hh, M_WAITOK | M_ZERO);
760 g_post_event(g_resize_provider_event, hh, M_WAITOK, NULL);
764 #define _PATH_DEV "/dev/"
768 g_provider_by_name(char const *arg)
772 struct g_provider *pp, *wpp;
774 if (strncmp(arg, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
775 arg += sizeof(_PATH_DEV) - 1;
778 LIST_FOREACH(cp, &g_classes, class) {
779 LIST_FOREACH(gp, &cp->geom, geom) {
780 LIST_FOREACH(pp, &gp->provider, provider) {
781 if (strcmp(arg, pp->name) != 0)
783 if ((gp->flags & G_GEOM_WITHER) == 0 &&
784 (pp->flags & G_PF_WITHER) == 0)
796 g_destroy_provider(struct g_provider *pp)
799 struct g_geom_alias *gap, *gaptmp;
802 G_VALID_PROVIDER(pp);
803 KASSERT(LIST_EMPTY(&pp->consumers),
804 ("g_destroy_provider but attached"));
805 KASSERT (pp->acr == 0, ("g_destroy_provider with acr"));
806 KASSERT (pp->acw == 0, ("g_destroy_provider with acw"));
807 KASSERT (pp->ace == 0, ("g_destroy_provider with ace"));
809 LIST_REMOVE(pp, provider);
811 devstat_remove_entry(pp->stat);
813 * If a callback was provided, send notification that the provider
816 if (gp->providergone != NULL)
817 gp->providergone(pp);
818 LIST_FOREACH_SAFE(gap, &pp->aliases, ga_next, gaptmp)
821 if ((gp->flags & G_GEOM_WITHER))
826 * We keep the "geoms" list sorted by topological order (== increasing
827 * numerical rank) at all times.
828 * When an attach is done, the attaching geoms rank is invalidated
829 * and it is moved to the tail of the list.
830 * All geoms later in the sequence has their ranks reevaluated in
831 * sequence. If we cannot assign rank to a geom because it's
832 * prerequisites do not have rank, we move that element to the tail
833 * of the sequence with invalid rank as well.
834 * At some point we encounter our original geom and if we stil fail
835 * to assign it a rank, there must be a loop and we fail back to
836 * g_attach() which detach again and calls redo_rank again
837 * to fix up the damage.
838 * It would be much simpler code wise to do it recursively, but we
839 * can't risk that on the kernel stack.
843 redo_rank(struct g_geom *gp)
845 struct g_consumer *cp;
846 struct g_geom *gp1, *gp2;
852 /* Invalidate this geoms rank and move it to the tail */
853 gp1 = TAILQ_NEXT(gp, geoms);
856 TAILQ_REMOVE(&geoms, gp, geoms);
857 TAILQ_INSERT_TAIL(&geoms, gp, geoms);
862 /* re-rank the rest of the sequence */
863 for (; gp1 != NULL; gp1 = gp2) {
866 LIST_FOREACH(cp, &gp1->consumer, consumer) {
867 if (cp->provider == NULL)
869 n = cp->provider->geom->rank;
877 gp2 = TAILQ_NEXT(gp1, geoms);
879 /* got a rank, moving on */
883 /* no rank to original geom means loop */
887 /* no rank, put it at the end move on */
888 TAILQ_REMOVE(&geoms, gp1, geoms);
889 TAILQ_INSERT_TAIL(&geoms, gp1, geoms);
895 g_attach(struct g_consumer *cp, struct g_provider *pp)
900 G_VALID_CONSUMER(cp);
901 G_VALID_PROVIDER(pp);
902 g_trace(G_T_TOPOLOGY, "g_attach(%p, %p)", cp, pp);
903 KASSERT(cp->provider == NULL, ("attach but attached"));
905 cp->flags &= ~G_CF_ORPHAN;
906 LIST_INSERT_HEAD(&pp->consumers, cp, consumers);
907 error = redo_rank(cp->geom);
909 LIST_REMOVE(cp, consumers);
917 g_detach(struct g_consumer *cp)
919 struct g_provider *pp;
922 G_VALID_CONSUMER(cp);
923 g_trace(G_T_TOPOLOGY, "g_detach(%p)", cp);
924 KASSERT(cp->provider != NULL, ("detach but not attached"));
925 KASSERT(cp->acr == 0, ("detach but nonzero acr"));
926 KASSERT(cp->acw == 0, ("detach but nonzero acw"));
927 KASSERT(cp->ace == 0, ("detach but nonzero ace"));
928 KASSERT(cp->nstart == cp->nend,
929 ("detach with active requests"));
931 LIST_REMOVE(cp, consumers);
933 if ((cp->geom->flags & G_GEOM_WITHER) ||
934 (pp->geom->flags & G_GEOM_WITHER) ||
935 (pp->flags & G_PF_WITHER))
943 * Access-check with delta values. The question asked is "can provider
944 * "cp" change the access counters by the relative amounts dc[rwe] ?"
948 g_access(struct g_consumer *cp, int dcr, int dcw, int dce)
950 struct g_provider *pp;
959 G_VALID_CONSUMER(cp);
961 KASSERT(pp != NULL, ("access but not attached"));
962 G_VALID_PROVIDER(pp);
965 g_trace(G_T_ACCESS, "g_access(%p(%s), %d, %d, %d)",
966 cp, pp->name, dcr, dcw, dce);
968 KASSERT(cp->acr + dcr >= 0, ("access resulting in negative acr"));
969 KASSERT(cp->acw + dcw >= 0, ("access resulting in negative acw"));
970 KASSERT(cp->ace + dce >= 0, ("access resulting in negative ace"));
971 KASSERT(dcr != 0 || dcw != 0 || dce != 0, ("NOP access request"));
972 KASSERT(cp->acr + dcr != 0 || cp->acw + dcw != 0 ||
973 cp->ace + dce != 0 || cp->nstart == cp->nend,
974 ("Last close with active requests"));
975 KASSERT(gp->access != NULL, ("NULL geom->access"));
978 * If our class cares about being spoiled, and we have been, we
979 * are probably just ahead of the event telling us that. Fail
980 * now rather than having to unravel this later.
982 if (cp->geom->spoiled != NULL && (cp->flags & G_CF_SPOILED) &&
983 (dcr > 0 || dcw > 0 || dce > 0))
987 * A number of GEOM classes either need to perform an I/O on the first
988 * open or to acquire a different subsystem's lock. To do that they
989 * may have to drop the topology lock.
990 * Other GEOM classes perform special actions when opening a lower rank
991 * geom for the first time. As a result, more than one thread may
992 * end up performing the special actions.
993 * So, we prevent concurrent "first" opens by marking the consumer with
996 * Note that if the geom's access method never drops the topology lock,
997 * then we will never see G_GEOM_IN_ACCESS here.
999 while ((gp->flags & G_GEOM_IN_ACCESS) != 0) {
1001 "%s: race on geom %s via provider %s and consumer of %s",
1002 __func__, gp->name, pp->name, cp->geom->name);
1003 gp->flags |= G_GEOM_ACCESS_WAIT;
1004 g_topology_sleep(gp, 0);
1008 * Figure out what counts the provider would have had, if this
1009 * consumer had (r0w0e0) at this time.
1011 pw = pp->acw - cp->acw;
1012 pe = pp->ace - cp->ace;
1015 "open delta:[r%dw%de%d] old:[r%dw%de%d] provider:[r%dw%de%d] %p(%s)",
1017 cp->acr, cp->acw, cp->ace,
1018 pp->acr, pp->acw, pp->ace,
1021 /* If foot-shooting is enabled, any open on rank#1 is OK */
1022 if ((g_debugflags & G_F_FOOTSHOOTING) && gp->rank == 1)
1024 /* If we try exclusive but already write: fail */
1025 else if (dce > 0 && pw > 0)
1027 /* If we try write but already exclusive: fail */
1028 else if (dcw > 0 && pe > 0)
1030 /* If we try to open more but provider is error'ed: fail */
1031 else if ((dcr > 0 || dcw > 0 || dce > 0) && pp->error != 0) {
1032 printf("%s(%d): provider %s has error %d set\n",
1033 __func__, __LINE__, pp->name, pp->error);
1044 gp->flags |= G_GEOM_IN_ACCESS;
1045 error = gp->access(pp, dcr, dcw, dce);
1046 KASSERT(dcr > 0 || dcw > 0 || dce > 0 || error == 0,
1047 ("Geom provider %s::%s dcr=%d dcw=%d dce=%d error=%d failed "
1048 "closing ->access()", gp->class->name, pp->name, dcr, dcw,
1051 g_topology_assert();
1052 gp->flags &= ~G_GEOM_IN_ACCESS;
1053 KASSERT(cp->acr == sr && cp->acw == sw && cp->ace == se,
1054 ("Access counts changed during geom->access"));
1055 if ((gp->flags & G_GEOM_ACCESS_WAIT) != 0) {
1056 gp->flags &= ~G_GEOM_ACCESS_WAIT;
1062 * If we open first write, spoil any partner consumers.
1063 * If we close last write and provider is not errored,
1066 if (pp->acw == 0 && dcw != 0)
1068 else if (pp->acw != 0 && pp->acw == -dcw && pp->error == 0 &&
1069 !(gp->flags & G_GEOM_WITHER))
1070 g_post_event(g_new_provider_event, pp, M_WAITOK,
1079 if (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)
1080 KASSERT(pp->sectorsize > 0,
1081 ("Provider %s lacks sectorsize", pp->name));
1082 if ((cp->geom->flags & G_GEOM_WITHER) &&
1083 cp->acr == 0 && cp->acw == 0 && cp->ace == 0)
1090 g_handleattr_int(struct bio *bp, const char *attribute, int val)
1093 return (g_handleattr(bp, attribute, &val, sizeof val));
1097 g_handleattr_uint16_t(struct bio *bp, const char *attribute, uint16_t val)
1100 return (g_handleattr(bp, attribute, &val, sizeof val));
1104 g_handleattr_off_t(struct bio *bp, const char *attribute, off_t val)
1107 return (g_handleattr(bp, attribute, &val, sizeof val));
1111 g_handleattr_str(struct bio *bp, const char *attribute, const char *str)
1114 return (g_handleattr(bp, attribute, str, 0));
1118 g_handleattr(struct bio *bp, const char *attribute, const void *val, int len)
1122 if (strcmp(bp->bio_attribute, attribute))
1125 bzero(bp->bio_data, bp->bio_length);
1126 if (strlcpy(bp->bio_data, val, bp->bio_length) >=
1128 printf("%s: %s %s bio_length %jd strlen %zu -> EFAULT\n",
1129 __func__, bp->bio_to->name, attribute,
1130 (intmax_t)bp->bio_length, strlen(val));
1133 } else if (bp->bio_length == len) {
1134 bcopy(val, bp->bio_data, len);
1136 printf("%s: %s %s bio_length %jd len %d -> EFAULT\n", __func__,
1137 bp->bio_to->name, attribute, (intmax_t)bp->bio_length, len);
1141 bp->bio_completed = bp->bio_length;
1142 g_io_deliver(bp, error);
1147 g_std_access(struct g_provider *pp,
1148 int dr __unused, int dw __unused, int de __unused)
1151 g_topology_assert();
1152 G_VALID_PROVIDER(pp);
1157 g_std_done(struct bio *bp)
1161 bp2 = bp->bio_parent;
1162 if (bp2->bio_error == 0)
1163 bp2->bio_error = bp->bio_error;
1164 bp2->bio_completed += bp->bio_completed;
1167 if (bp2->bio_children == bp2->bio_inbed) {
1168 if (bp2->bio_cmd == BIO_SPEEDUP)
1169 bp2->bio_completed = bp2->bio_length;
1170 g_io_deliver(bp2, bp2->bio_error);
1174 /* XXX: maybe this is only g_slice_spoiled */
1177 g_std_spoiled(struct g_consumer *cp)
1180 struct g_provider *pp;
1182 g_topology_assert();
1183 G_VALID_CONSUMER(cp);
1184 g_trace(G_T_TOPOLOGY, "g_std_spoiled(%p)", cp);
1185 cp->flags |= G_CF_ORPHAN;
1188 LIST_FOREACH(pp, &gp->provider, provider)
1189 g_orphan_provider(pp, ENXIO);
1190 g_destroy_consumer(cp);
1191 if (LIST_EMPTY(&gp->provider) && LIST_EMPTY(&gp->consumer))
1194 gp->flags |= G_GEOM_WITHER;
1198 * Spoiling happens when a provider is opened for writing, but consumers
1199 * which are configured by in-band data are attached (slicers for instance).
1200 * Since the write might potentially change the in-band data, such consumers
1201 * need to re-evaluate their existence after the writing session closes.
1202 * We do this by (offering to) tear them down when the open for write happens
1203 * in return for a re-taste when it closes again.
1204 * Together with the fact that such consumers grab an 'e' bit whenever they
1205 * are open, regardless of mode, this ends up DTRT.
1209 g_spoil_event(void *arg, int flag)
1211 struct g_provider *pp;
1212 struct g_consumer *cp, *cp2;
1214 g_topology_assert();
1215 if (flag == EV_CANCEL)
1218 G_VALID_PROVIDER(pp);
1219 g_trace(G_T_TOPOLOGY, "%s %p(%s:%s:%s)", __func__, pp,
1220 pp->geom->class->name, pp->geom->name, pp->name);
1221 for (cp = LIST_FIRST(&pp->consumers); cp != NULL; cp = cp2) {
1222 cp2 = LIST_NEXT(cp, consumers);
1223 if ((cp->flags & G_CF_SPOILED) == 0)
1225 cp->flags &= ~G_CF_SPOILED;
1226 if (cp->geom->spoiled == NULL)
1228 cp->geom->spoiled(cp);
1229 g_topology_assert();
1234 g_spoil(struct g_provider *pp, struct g_consumer *cp)
1236 struct g_consumer *cp2;
1238 g_topology_assert();
1239 G_VALID_PROVIDER(pp);
1240 G_VALID_CONSUMER(cp);
1242 LIST_FOREACH(cp2, &pp->consumers, consumers) {
1246 KASSERT(cp2->acr == 0, ("spoiling cp->acr = %d", cp2->acr));
1247 KASSERT(cp2->acw == 0, ("spoiling cp->acw = %d", cp2->acw));
1249 KASSERT(cp2->ace == 0, ("spoiling cp->ace = %d", cp2->ace));
1250 cp2->flags |= G_CF_SPOILED;
1252 g_post_event(g_spoil_event, pp, M_WAITOK, pp, NULL);
1256 g_media_changed_event(void *arg, int flag)
1258 struct g_provider *pp;
1261 g_topology_assert();
1262 if (flag == EV_CANCEL)
1265 G_VALID_PROVIDER(pp);
1268 * If provider was not open for writing, queue retaste after spoiling.
1269 * If it was, retaste will happen automatically on close.
1271 retaste = (pp->acw == 0 && pp->error == 0 &&
1272 !(pp->geom->flags & G_GEOM_WITHER));
1273 g_spoil_event(arg, flag);
1275 g_post_event(g_new_provider_event, pp, M_WAITOK, pp, NULL);
1279 g_media_changed(struct g_provider *pp, int flag)
1281 struct g_consumer *cp;
1283 LIST_FOREACH(cp, &pp->consumers, consumers)
1284 cp->flags |= G_CF_SPOILED;
1285 return (g_post_event(g_media_changed_event, pp, flag, pp, NULL));
1289 g_media_gone(struct g_provider *pp, int flag)
1291 struct g_consumer *cp;
1293 LIST_FOREACH(cp, &pp->consumers, consumers)
1294 cp->flags |= G_CF_SPOILED;
1295 return (g_post_event(g_spoil_event, pp, flag, pp, NULL));
1299 g_getattr__(const char *attr, struct g_consumer *cp, void *var, int len)
1304 error = g_io_getattr(attr, cp, &i, var);
1313 g_get_device_prefix_len(const char *name)
1317 if (strncmp(name, "ada", 3) == 0)
1319 else if (strncmp(name, "ad", 2) == 0)
1323 if (name[len] < '0' || name[len] > '9')
1327 } while (name[len] >= '0' && name[len] <= '9');
1332 g_compare_names(const char *namea, const char *nameb)
1336 if (strcmp(namea, nameb) == 0)
1338 deva = g_get_device_prefix_len(namea);
1341 devb = g_get_device_prefix_len(nameb);
1344 if (strcmp(namea + deva, nameb + devb) == 0)
1349 #if defined(DIAGNOSTIC) || defined(DDB)
1351 * This function walks the mesh and returns a non-zero integer if it
1352 * finds the argument pointer is an object. The return value indicates
1353 * which type of object it is believed to be. If topology is not locked,
1354 * this function is potentially dangerous, but we don't assert that the
1355 * topology lock is held when called from debugger.
1358 g_valid_obj(void const *ptr)
1362 struct g_consumer *cp;
1363 struct g_provider *pp;
1366 if (kdb_active == 0)
1368 g_topology_assert();
1370 LIST_FOREACH(mp, &g_classes, class) {
1373 LIST_FOREACH(gp, &mp->geom, geom) {
1376 LIST_FOREACH(cp, &gp->consumer, consumer)
1379 LIST_FOREACH(pp, &gp->provider, provider)
1390 #define gprintf(...) do { \
1391 db_printf("%*s", indent, ""); \
1392 db_printf(__VA_ARGS__); \
1394 #define gprintln(...) do { \
1395 gprintf(__VA_ARGS__); \
1399 #define ADDFLAG(obj, flag, sflag) do { \
1400 if ((obj)->flags & (flag)) { \
1402 strlcat(str, ",", size); \
1403 strlcat(str, (sflag), size); \
1409 provider_flags_to_string(struct g_provider *pp, char *str, size_t size)
1414 if (pp->flags == 0) {
1415 strlcpy(str, "NONE", size);
1418 ADDFLAG(pp, G_PF_WITHER, "G_PF_WITHER");
1419 ADDFLAG(pp, G_PF_ORPHAN, "G_PF_ORPHAN");
1424 geom_flags_to_string(struct g_geom *gp, char *str, size_t size)
1429 if (gp->flags == 0) {
1430 strlcpy(str, "NONE", size);
1433 ADDFLAG(gp, G_GEOM_WITHER, "G_GEOM_WITHER");
1437 db_show_geom_consumer(int indent, struct g_consumer *cp)
1441 gprintln("consumer: %p", cp);
1442 gprintln(" class: %s (%p)", cp->geom->class->name,
1444 gprintln(" geom: %s (%p)", cp->geom->name, cp->geom);
1445 if (cp->provider == NULL)
1446 gprintln(" provider: none");
1448 gprintln(" provider: %s (%p)", cp->provider->name,
1451 gprintln(" access: r%dw%de%d", cp->acr, cp->acw, cp->ace);
1452 gprintln(" flags: 0x%04x", cp->flags);
1454 gprintln(" nstart: %u", cp->nstart);
1455 gprintln(" nend: %u", cp->nend);
1458 gprintf("consumer: %p (%s), access=r%dw%de%d", cp,
1459 cp->provider != NULL ? cp->provider->name : "none",
1460 cp->acr, cp->acw, cp->ace);
1462 db_printf(", flags=0x%04x", cp->flags);
1468 db_show_geom_provider(int indent, struct g_provider *pp)
1470 struct g_consumer *cp;
1474 gprintln("provider: %s (%p)", pp->name, pp);
1475 gprintln(" class: %s (%p)", pp->geom->class->name,
1477 gprintln(" geom: %s (%p)", pp->geom->name, pp->geom);
1478 gprintln(" mediasize: %jd", (intmax_t)pp->mediasize);
1479 gprintln(" sectorsize: %u", pp->sectorsize);
1480 gprintln(" stripesize: %ju", (uintmax_t)pp->stripesize);
1481 gprintln(" stripeoffset: %ju", (uintmax_t)pp->stripeoffset);
1482 gprintln(" access: r%dw%de%d", pp->acr, pp->acw,
1484 gprintln(" flags: %s (0x%04x)",
1485 provider_flags_to_string(pp, flags, sizeof(flags)),
1487 gprintln(" error: %d", pp->error);
1488 if (LIST_EMPTY(&pp->consumers))
1489 gprintln(" consumers: none");
1491 gprintf("provider: %s (%p), access=r%dw%de%d",
1492 pp->name, pp, pp->acr, pp->acw, pp->ace);
1493 if (pp->flags != 0) {
1494 db_printf(", flags=%s (0x%04x)",
1495 provider_flags_to_string(pp, flags, sizeof(flags)),
1500 if (!LIST_EMPTY(&pp->consumers)) {
1501 LIST_FOREACH(cp, &pp->consumers, consumers) {
1502 db_show_geom_consumer(indent + 2, cp);
1510 db_show_geom_geom(int indent, struct g_geom *gp)
1512 struct g_provider *pp;
1513 struct g_consumer *cp;
1517 gprintln("geom: %s (%p)", gp->name, gp);
1518 gprintln(" class: %s (%p)", gp->class->name, gp->class);
1519 gprintln(" flags: %s (0x%04x)",
1520 geom_flags_to_string(gp, flags, sizeof(flags)), gp->flags);
1521 gprintln(" rank: %d", gp->rank);
1522 if (LIST_EMPTY(&gp->provider))
1523 gprintln(" providers: none");
1524 if (LIST_EMPTY(&gp->consumer))
1525 gprintln(" consumers: none");
1527 gprintf("geom: %s (%p), rank=%d", gp->name, gp, gp->rank);
1528 if (gp->flags != 0) {
1529 db_printf(", flags=%s (0x%04x)",
1530 geom_flags_to_string(gp, flags, sizeof(flags)),
1535 if (!LIST_EMPTY(&gp->provider)) {
1536 LIST_FOREACH(pp, &gp->provider, provider) {
1537 db_show_geom_provider(indent + 2, pp);
1542 if (!LIST_EMPTY(&gp->consumer)) {
1543 LIST_FOREACH(cp, &gp->consumer, consumer) {
1544 db_show_geom_consumer(indent + 2, cp);
1552 db_show_geom_class(struct g_class *mp)
1556 db_printf("class: %s (%p)\n", mp->name, mp);
1557 LIST_FOREACH(gp, &mp->geom, geom) {
1558 db_show_geom_geom(2, gp);
1565 * Print the GEOM topology or the given object.
1567 DB_SHOW_COMMAND(geom, db_show_geom)
1572 /* No address given, print the entire topology. */
1573 LIST_FOREACH(mp, &g_classes, class) {
1574 db_show_geom_class(mp);
1580 switch (g_valid_obj((void *)addr)) {
1582 db_show_geom_class((struct g_class *)addr);
1585 db_show_geom_geom(0, (struct g_geom *)addr);
1588 db_show_geom_consumer(0, (struct g_consumer *)addr);
1591 db_show_geom_provider(0, (struct g_provider *)addr);
1594 db_printf("Not a GEOM object.\n");
1601 db_print_bio_cmd(struct bio *bp)
1603 db_printf(" cmd: ");
1604 switch (bp->bio_cmd) {
1605 case BIO_READ: db_printf("BIO_READ"); break;
1606 case BIO_WRITE: db_printf("BIO_WRITE"); break;
1607 case BIO_DELETE: db_printf("BIO_DELETE"); break;
1608 case BIO_GETATTR: db_printf("BIO_GETATTR"); break;
1609 case BIO_FLUSH: db_printf("BIO_FLUSH"); break;
1610 case BIO_CMD0: db_printf("BIO_CMD0"); break;
1611 case BIO_CMD1: db_printf("BIO_CMD1"); break;
1612 case BIO_CMD2: db_printf("BIO_CMD2"); break;
1613 case BIO_ZONE: db_printf("BIO_ZONE"); break;
1614 default: db_printf("UNKNOWN"); break;
1620 db_print_bio_flags(struct bio *bp)
1625 db_printf(" flags: ");
1626 if (bp->bio_flags & BIO_ERROR) {
1627 db_printf("BIO_ERROR");
1630 if (bp->bio_flags & BIO_DONE) {
1631 db_printf("%sBIO_DONE", (comma ? ", " : ""));
1634 if (bp->bio_flags & BIO_ONQUEUE)
1635 db_printf("%sBIO_ONQUEUE", (comma ? ", " : ""));
1640 * Print useful information in a BIO
1642 DB_SHOW_COMMAND(bio, db_show_bio)
1647 bp = (struct bio *)addr;
1648 db_printf("BIO %p\n", bp);
1649 db_print_bio_cmd(bp);
1650 db_print_bio_flags(bp);
1651 db_printf(" cflags: 0x%hx\n", bp->bio_cflags);
1652 db_printf(" pflags: 0x%hx\n", bp->bio_pflags);
1653 db_printf(" offset: %jd\n", (intmax_t)bp->bio_offset);
1654 db_printf(" length: %jd\n", (intmax_t)bp->bio_length);
1655 db_printf(" bcount: %ld\n", bp->bio_bcount);
1656 db_printf(" resid: %ld\n", bp->bio_resid);
1657 db_printf(" completed: %jd\n", (intmax_t)bp->bio_completed);
1658 db_printf(" children: %u\n", bp->bio_children);
1659 db_printf(" inbed: %u\n", bp->bio_inbed);
1660 db_printf(" error: %d\n", bp->bio_error);
1661 db_printf(" parent: %p\n", bp->bio_parent);
1662 db_printf(" driver1: %p\n", bp->bio_driver1);
1663 db_printf(" driver2: %p\n", bp->bio_driver2);
1664 db_printf(" caller1: %p\n", bp->bio_caller1);
1665 db_printf(" caller2: %p\n", bp->bio_caller2);
1666 db_printf(" bio_from: %p\n", bp->bio_from);
1667 db_printf(" bio_to: %p\n", bp->bio_to);
1669 #if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING)
1670 db_printf(" bio_track_bp: %p\n", bp->bio_track_bp);