2 * Copyright (c) 2006-2007 Ivan Voras <ivoras@freebsd.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 /* Implementation notes:
28 * - "Components" are wrappers around providers that make up the
29 * virtual storage (i.e. a virstor has "physical" components)
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/module.h>
40 #include <sys/mutex.h>
43 #include <sys/sysctl.h>
44 #include <sys/malloc.h>
47 #include <sys/kthread.h>
48 #include <sys/mutex.h>
50 #include <geom/geom.h>
52 #include <geom/virstor/g_virstor.h>
53 #include <geom/virstor/g_virstor_md.h>
55 /* Declare malloc(9) label */
56 static MALLOC_DEFINE(M_GVIRSTOR, "gvirstor", "GEOM_VIRSTOR Data");
58 /* GEOM class methods */
59 static g_init_t g_virstor_init;
60 static g_fini_t g_virstor_fini;
61 static g_taste_t g_virstor_taste;
62 static g_ctl_req_t g_virstor_config;
63 static g_ctl_destroy_geom_t g_virstor_destroy_geom;
65 /* Declare & initialize class structure ("geom class") */
66 struct g_class g_virstor_class = {
67 .name = G_VIRSTOR_CLASS_NAME,
69 .init = g_virstor_init,
70 .fini = g_virstor_fini,
71 .taste = g_virstor_taste,
72 .ctlreq = g_virstor_config,
73 .destroy_geom = g_virstor_destroy_geom
74 /* The .dumpconf and the rest are only usable for a geom instance, so
75 * they will be set when such instance is created. */
78 /* Declare sysctl's and loader tunables */
79 SYSCTL_DECL(_kern_geom);
80 SYSCTL_NODE(_kern_geom, OID_AUTO, virstor, CTLFLAG_RW, 0, "GEOM_GVIRSTOR information");
82 static u_int g_virstor_debug = 2; /* XXX: lower to 2 when released to public */
83 TUNABLE_INT("kern.geom.virstor.debug", &g_virstor_debug);
84 SYSCTL_UINT(_kern_geom_virstor, OID_AUTO, debug, CTLFLAG_RW, &g_virstor_debug,
85 0, "Debug level (2=production, 5=normal, 15=excessive)");
87 static u_int g_virstor_chunk_watermark = 100;
88 TUNABLE_INT("kern.geom.virstor.chunk_watermark", &g_virstor_chunk_watermark);
89 SYSCTL_UINT(_kern_geom_virstor, OID_AUTO, chunk_watermark, CTLFLAG_RW,
90 &g_virstor_chunk_watermark, 0,
91 "Minimum number of free chunks before issuing administrative warning");
93 static u_int g_virstor_component_watermark = 1;
94 TUNABLE_INT("kern.geom.virstor.component_watermark",
95 &g_virstor_component_watermark);
96 SYSCTL_UINT(_kern_geom_virstor, OID_AUTO, component_watermark, CTLFLAG_RW,
97 &g_virstor_component_watermark, 0,
98 "Minimum number of free components before issuing administrative warning");
100 static int read_metadata(struct g_consumer *, struct g_virstor_metadata *);
101 static int write_metadata(struct g_consumer *, struct g_virstor_metadata *);
102 static int clear_metadata(struct g_virstor_component *);
103 static int add_provider_to_geom(struct g_virstor_softc *, struct g_provider *,
104 struct g_virstor_metadata *);
105 static struct g_geom *create_virstor_geom(struct g_class *,
106 struct g_virstor_metadata *);
107 static void virstor_check_and_run(struct g_virstor_softc *);
108 static u_int virstor_valid_components(struct g_virstor_softc *);
109 static int virstor_geom_destroy(struct g_virstor_softc *, boolean_t,
111 static void remove_component(struct g_virstor_softc *,
112 struct g_virstor_component *, boolean_t);
113 static void bioq_dismantle(struct bio_queue_head *);
114 static int allocate_chunk(struct g_virstor_softc *,
115 struct g_virstor_component **, u_int *, u_int *);
116 static void delay_destroy_consumer(void *, int);
117 static void dump_component(struct g_virstor_component *comp);
119 static void dump_me(struct virstor_map_entry *me, unsigned int nr);
122 static void virstor_ctl_stop(struct gctl_req *, struct g_class *);
123 static void virstor_ctl_add(struct gctl_req *, struct g_class *);
124 static void virstor_ctl_remove(struct gctl_req *, struct g_class *);
125 static struct g_virstor_softc * virstor_find_geom(const struct g_class *,
127 static void update_metadata(struct g_virstor_softc *);
128 static void fill_metadata(struct g_virstor_softc *, struct g_virstor_metadata *,
131 static void g_virstor_orphan(struct g_consumer *);
132 static int g_virstor_access(struct g_provider *, int, int, int);
133 static void g_virstor_start(struct bio *);
134 static void g_virstor_dumpconf(struct sbuf *, const char *, struct g_geom *,
135 struct g_consumer *, struct g_provider *);
136 static void g_virstor_done(struct bio *);
138 static void invalid_call(void);
140 * Initialise GEOM class (per-class callback)
143 g_virstor_init(struct g_class *mp __unused)
146 /* Catch map struct size mismatch at compile time; Map entries must
147 * fit into MAXPHYS exactly, with no wasted space. */
148 CTASSERT(VIRSTOR_MAP_BLOCK_ENTRIES*VIRSTOR_MAP_ENTRY_SIZE == MAXPHYS);
150 /* Init UMA zones, TAILQ's, other global vars */
154 * Finalise GEOM class (per-class callback)
157 g_virstor_fini(struct g_class *mp __unused)
160 /* Deinit UMA zones & global vars */
164 * Config (per-class callback)
167 g_virstor_config(struct gctl_req *req, struct g_class *cp, char const *verb)
173 version = gctl_get_paraml(req, "version", sizeof(*version));
174 if (version == NULL) {
175 gctl_error(req, "Failed to get 'version' argument");
178 if (*version != G_VIRSTOR_VERSION) {
179 gctl_error(req, "Userland and kernel versions out of sync");
184 if (strcmp(verb, "add") == 0)
185 virstor_ctl_add(req, cp);
186 else if (strcmp(verb, "stop") == 0 || strcmp(verb, "destroy") == 0)
187 virstor_ctl_stop(req, cp);
188 else if (strcmp(verb, "remove") == 0)
189 virstor_ctl_remove(req, cp);
191 gctl_error(req, "unknown verb: '%s'", verb);
196 * "stop" verb from userland
199 virstor_ctl_stop(struct gctl_req *req, struct g_class *cp)
204 nargs = gctl_get_paraml(req, "nargs", sizeof *nargs);
206 gctl_error(req, "Error fetching argument '%s'", "nargs");
210 gctl_error(req, "Invalid number of arguments");
213 force = gctl_get_paraml(req, "force", sizeof *force);
215 gctl_error(req, "Error fetching argument '%s'", "force");
220 for (i = 0; i < *nargs; i++) {
223 struct g_virstor_softc *sc;
226 sprintf(param, "arg%d", i);
227 name = gctl_get_asciiparam(req, param);
229 gctl_error(req, "No 'arg%d' argument", i);
233 sc = virstor_find_geom(cp, name);
234 LOG_MSG(LVL_INFO, "Stopping %s by the userland command",
237 if ((error = virstor_geom_destroy(sc, TRUE, TRUE)) != 0) {
238 LOG_MSG(LVL_ERROR, "Cannot destroy %s: %d",
239 sc->geom->name, error);
246 * "add" verb from userland - add new component(s) to the structure.
247 * This will be done all at once in here, without going through the
248 * .taste function for new components.
251 virstor_ctl_add(struct gctl_req *req, struct g_class *cp)
253 /* Note: while this is going on, I/O is being done on
254 * the g_up and g_down threads. The idea is to make changes
255 * to softc members in a way that can atomically activate
256 * them all at once. */
257 struct g_virstor_softc *sc;
258 int *hardcode, *nargs;
259 const char *geom_name; /* geom to add a component to */
260 struct g_consumer *fcp;
261 struct g_virstor_bio_q *bq;
266 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
268 gctl_error(req, "Error fetching argument '%s'", "nargs");
272 gctl_error(req, "Invalid number of arguments");
275 hardcode = gctl_get_paraml(req, "hardcode", sizeof(*hardcode));
276 if (hardcode == NULL) {
277 gctl_error(req, "Error fetching argument '%s'", "hardcode");
281 /* Find "our" geom */
282 geom_name = gctl_get_asciiparam(req, "arg0");
283 if (geom_name == NULL) {
284 gctl_error(req, "Error fetching argument '%s'", "geom_name (arg0)");
287 sc = virstor_find_geom(cp, geom_name);
289 gctl_error(req, "Don't know anything about '%s'", geom_name);
293 if (virstor_valid_components(sc) != sc->n_components) {
294 LOG_MSG(LVL_ERROR, "Cannot add components to incomplete "
295 "virstor %s", sc->geom->name);
296 gctl_error(req, "Virstor %s is incomplete", sc->geom->name);
300 fcp = sc->components[0].gcons;
303 for (i = 1; i < *nargs; i++) {
304 struct g_virstor_metadata md;
306 const char *prov_name;
307 struct g_provider *pp;
308 struct g_consumer *cp;
312 snprintf(aname, sizeof aname, "arg%d", i);
313 prov_name = gctl_get_asciiparam(req, aname);
314 if (prov_name == NULL) {
315 gctl_error(req, "Error fetching argument '%s'", aname);
319 if (strncmp(prov_name, _PATH_DEV, strlen(_PATH_DEV)) == 0)
320 prov_name += strlen(_PATH_DEV);
322 pp = g_provider_by_name(prov_name);
324 /* This is the most common error so be verbose about it */
326 gctl_error(req, "Invalid provider: '%s' (added"
327 " %u components)", prov_name, added);
330 gctl_error(req, "Invalid provider: '%s'",
336 cp = g_new_consumer(sc->geom);
338 gctl_error(req, "Cannot create consumer");
342 error = g_attach(cp, pp);
344 gctl_error(req, "Cannot attach a consumer to %s",
346 g_destroy_consumer(cp);
350 if (fcp->acr != 0 || fcp->acw != 0 || fcp->ace != 0) {
351 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace);
353 gctl_error(req, "Access request failed for %s",
355 g_destroy_consumer(cp);
360 if (fcp->provider->sectorsize != pp->sectorsize) {
361 gctl_error(req, "Sector size doesn't fit for %s",
363 g_destroy_consumer(cp);
367 for (j = 0; j < sc->n_components; j++) {
368 if (strcmp(sc->components[j].gcons->provider->name,
370 gctl_error(req, "Component %s already in %s",
371 pp->name, sc->geom->name);
372 g_destroy_consumer(cp);
377 sc->components = realloc(sc->components,
378 sizeof(*sc->components) * (sc->n_components + 1),
379 M_GVIRSTOR, M_WAITOK);
381 nc = sc->n_components;
382 sc->components[nc].gcons = cp;
383 sc->components[nc].sc = sc;
384 sc->components[nc].index = nc;
385 sc->components[nc].chunk_count = cp->provider->mediasize /
387 sc->components[nc].chunk_next = 0;
388 sc->components[nc].chunk_reserved = 0;
390 if (sc->components[nc].chunk_count < 4) {
391 gctl_error(req, "Provider too small: %s",
393 g_destroy_consumer(cp);
397 fill_metadata(sc, &md, nc, *hardcode);
398 write_metadata(cp, &md);
399 /* The new component becomes visible when n_components is
405 /* This call to update_metadata() is critical. In case there's a
406 * power failure in the middle of it and some components are updated
407 * while others are not, there will be trouble on next .taste() iff
408 * a non-updated component is detected first */
411 LOG_MSG(LVL_INFO, "Added %d component(s) to %s", added,
413 /* Fire off BIOs previously queued because there wasn't any
414 * physical space left. If the BIOs still can't be satisfied
415 * they will again be added to the end of the queue (during
416 * which the mutex will be recursed) */
417 bq = malloc(sizeof(*bq), M_GVIRSTOR, M_WAITOK);
419 mtx_lock(&sc->delayed_bio_q_mtx);
420 /* First, insert a sentinel to the queue end, so we don't
421 * end up in an infinite loop if there's still no free
422 * space available. */
423 STAILQ_INSERT_TAIL(&sc->delayed_bio_q, bq, linkage);
424 while (!STAILQ_EMPTY(&sc->delayed_bio_q)) {
425 bq = STAILQ_FIRST(&sc->delayed_bio_q);
426 if (bq->bio != NULL) {
427 g_virstor_start(bq->bio);
428 STAILQ_REMOVE_HEAD(&sc->delayed_bio_q, linkage);
429 free(bq, M_GVIRSTOR);
431 STAILQ_REMOVE_HEAD(&sc->delayed_bio_q, linkage);
432 free(bq, M_GVIRSTOR);
436 mtx_unlock(&sc->delayed_bio_q_mtx);
441 * Find a geom handled by the class
443 static struct g_virstor_softc *
444 virstor_find_geom(const struct g_class *cp, const char *name)
448 LIST_FOREACH(gp, &cp->geom, geom) {
449 if (strcmp(name, gp->name) == 0)
456 * Update metadata on all components to reflect the current state
461 * Expects things to be set up so write_metadata() can work, i.e.
462 * the topology lock must be held.
465 update_metadata(struct g_virstor_softc *sc)
467 struct g_virstor_metadata md;
470 if (virstor_valid_components(sc) != sc->n_components)
471 return; /* Incomplete device */
472 LOG_MSG(LVL_DEBUG, "Updating metadata on components for %s",
474 /* Update metadata on components */
475 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__,
476 sc->geom->class->name, sc->geom->name);
478 for (n = 0; n < sc->n_components; n++) {
479 read_metadata(sc->components[n].gcons, &md);
480 md.chunk_next = sc->components[n].chunk_next;
481 md.flags = sc->components[n].flags;
482 md.md_count = sc->n_components;
483 write_metadata(sc->components[n].gcons, &md);
488 * Fills metadata (struct md) from information stored in softc and the nc'th
489 * component of virstor
492 fill_metadata(struct g_virstor_softc *sc, struct g_virstor_metadata *md,
493 u_int nc, u_int hardcode)
495 struct g_virstor_component *c;
497 bzero(md, sizeof *md);
498 c = &sc->components[nc];
500 strncpy(md->md_magic, G_VIRSTOR_MAGIC, sizeof md->md_magic);
501 md->md_version = G_VIRSTOR_VERSION;
502 strncpy(md->md_name, sc->geom->name, sizeof md->md_name);
504 md->md_virsize = sc->virsize;
505 md->md_chunk_size = sc->chunk_size;
506 md->md_count = sc->n_components;
509 strncpy(md->provider, c->gcons->provider->name,
510 sizeof md->provider);
513 md->provsize = c->gcons->provider->mediasize;
514 md->chunk_count = c->chunk_count;
515 md->chunk_next = c->chunk_next;
516 md->chunk_reserved = c->chunk_reserved;
517 md->flags = c->flags;
521 * Remove a component from virstor device.
522 * Can only be done if the component is unallocated.
525 virstor_ctl_remove(struct gctl_req *req, struct g_class *cp)
527 /* As this is executed in parallel to I/O, operations on virstor
528 * structures must be as atomic as possible. */
529 struct g_virstor_softc *sc;
531 const char *geom_name;
535 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
537 gctl_error(req, "Error fetching argument '%s'", "nargs");
541 gctl_error(req, "Invalid number of arguments");
544 /* Find "our" geom */
545 geom_name = gctl_get_asciiparam(req, "arg0");
546 if (geom_name == NULL) {
547 gctl_error(req, "Error fetching argument '%s'",
551 sc = virstor_find_geom(cp, geom_name);
553 gctl_error(req, "Don't know anything about '%s'", geom_name);
557 if (virstor_valid_components(sc) != sc->n_components) {
558 LOG_MSG(LVL_ERROR, "Cannot remove components from incomplete "
559 "virstor %s", sc->geom->name);
560 gctl_error(req, "Virstor %s is incomplete", sc->geom->name);
565 for (i = 1; i < *nargs; i++) {
567 const char *prov_name;
569 struct g_virstor_component *newcomp, *compbak;
571 sprintf(param, "arg%d", i);
572 prov_name = gctl_get_asciiparam(req, param);
573 if (prov_name == NULL) {
574 gctl_error(req, "Error fetching argument '%s'", param);
577 if (strncmp(prov_name, _PATH_DEV, strlen(_PATH_DEV)) == 0)
578 prov_name += strlen(_PATH_DEV);
581 for (j = 0; j < sc->n_components; j++) {
582 if (strcmp(sc->components[j].gcons->provider->name,
589 LOG_MSG(LVL_ERROR, "No %s component in %s",
590 prov_name, sc->geom->name);
594 compbak = sc->components;
595 newcomp = malloc(sc->n_components * sizeof(*sc->components),
596 M_GVIRSTOR, M_WAITOK | M_ZERO);
597 bcopy(sc->components, newcomp, found * sizeof(*sc->components));
598 bcopy(&sc->components[found + 1], newcomp + found,
599 found * sizeof(*sc->components));
600 if ((sc->components[j].flags & VIRSTOR_PROVIDER_ALLOCATED) != 0) {
601 LOG_MSG(LVL_ERROR, "Allocated provider %s cannot be "
603 prov_name, sc->geom->name);
604 free(newcomp, M_GVIRSTOR);
605 /* We'll consider this non-fatal error */
608 /* Renumerate unallocated components */
609 for (j = 0; j < sc->n_components-1; j++) {
610 if ((sc->components[j].flags &
611 VIRSTOR_PROVIDER_ALLOCATED) == 0) {
612 sc->components[j].index = j;
615 /* This is the critical section. If a component allocation
616 * event happens while both variables are not yet set,
617 * there will be trouble. Something will panic on encountering
618 * NULL sc->components[x].gcomp member.
619 * Luckily, component allocation happens very rarely and
620 * removing components is an abnormal action in any case. */
621 sc->components = newcomp;
623 /* End critical section */
626 if (clear_metadata(&compbak[found]) != 0) {
627 LOG_MSG(LVL_WARNING, "Trouble ahead: cannot clear "
628 "metadata on %s", prov_name);
630 g_detach(compbak[found].gcons);
631 g_destroy_consumer(compbak[found].gcons);
634 free(compbak, M_GVIRSTOR);
639 /* This call to update_metadata() is critical. In case there's a
640 * power failure in the middle of it and some components are updated
641 * while others are not, there will be trouble on next .taste() iff
642 * a non-updated component is detected first */
646 LOG_MSG(LVL_INFO, "Removed %d component(s) from %s", removed,
651 * Clear metadata sector on component
654 clear_metadata(struct g_virstor_component *comp)
659 LOG_MSG(LVL_INFO, "Clearing metadata on %s",
660 comp->gcons->provider->name);
662 error = g_access(comp->gcons, 0, 1, 0);
665 buf = malloc(comp->gcons->provider->sectorsize, M_GVIRSTOR,
667 error = g_write_data(comp->gcons,
668 comp->gcons->provider->mediasize -
669 comp->gcons->provider->sectorsize,
671 comp->gcons->provider->sectorsize);
672 free(buf, M_GVIRSTOR);
673 g_access(comp->gcons, 0, -1, 0);
678 * Destroy geom forcibly.
681 g_virstor_destroy_geom(struct gctl_req *req __unused, struct g_class *mp,
684 struct g_virstor_softc *sc;
688 KASSERT(sc != NULL, ("%s: NULL sc", __func__));
691 LOG_MSG(LVL_DEBUG, "%s called for %s, sc=%p", __func__, gp->name,
702 LOG_MSG(LVL_INFO, "INVARIANTS detected");
703 LOG_MSG(LVL_INFO, "Verifying allocation "
704 "table for %s", sc->geom->name);
706 for (n = 0; n < sc->chunk_count; n++) {
707 if (sc->map[n].flags || VIRSTOR_MAP_ALLOCATED != 0)
710 LOG_MSG(LVL_INFO, "Device %s has %d allocated chunks",
711 sc->geom->name, count);
714 if (virstor_valid_components(sc) != sc->n_components) {
715 /* This is a incomplete virstor device (not all
716 * components have been found) */
717 LOG_MSG(LVL_ERROR, "Device %s is incomplete",
721 error = g_access(sc->components[0].gcons, 1, 0, 0);
722 KASSERT(error == 0, ("%s: g_access failed (%d)", __func__,
724 /* Compare the whole on-disk allocation table with what's
725 * currently in memory */
726 while (n < sc->chunk_count) {
727 buf = g_read_data(sc->components[0].gcons, off,
728 sc->sectorsize, &error);
729 KASSERT(buf != NULL, ("g_read_data returned NULL (%d) "
730 "for read at %jd", error, off));
731 if (bcmp(buf, &sc->map[n], sc->sectorsize) != 0) {
732 LOG_MSG(LVL_ERROR, "ERROR in allocation table, "
733 "entry %d, offset %jd", n, off);
737 n += sc->me_per_sector;
738 off += sc->sectorsize;
741 error = g_access(sc->components[0].gcons, -1, 0, 0);
742 KASSERT(error == 0, ("%s: g_access failed (%d) on exit",
745 LOG_MSG(LVL_ERROR, "ALLOCATION TABLE CORRUPTED FOR %s "
746 "(%d sectors don't match, max %zu allocations)",
747 sc->geom->name, count,
748 count * sc->me_per_sector);
750 LOG_MSG(LVL_INFO, "Allocation table ok for %s",
756 virstor_geom_destroy(sc, FALSE, FALSE);
764 * Taste event (per-class callback)
765 * Examines a provider and creates geom instances if needed
767 static struct g_geom *
768 g_virstor_taste(struct g_class *mp, struct g_provider *pp, int flags)
770 struct g_virstor_metadata md;
772 struct g_consumer *cp;
773 struct g_virstor_softc *sc;
776 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
778 LOG_MSG(LVL_DEBUG, "Tasting %s", pp->name);
780 /* We need a dummy geom to attach a consumer to the given provider */
781 gp = g_new_geomf(mp, "virstor:taste.helper");
782 gp->start = (void *)invalid_call; /* XXX: hacked up so the */
783 gp->access = (void *)invalid_call; /* compiler doesn't complain. */
784 gp->orphan = (void *)invalid_call; /* I really want these to fail. */
786 cp = g_new_consumer(gp);
788 error = read_metadata(cp, &md);
790 g_destroy_consumer(cp);
796 if (strcmp(md.md_magic, G_VIRSTOR_MAGIC) != 0)
798 if (md.md_version != G_VIRSTOR_VERSION) {
799 LOG_MSG(LVL_ERROR, "Kernel module version invalid "
800 "to handle %s (%s) : %d should be %d",
801 md.md_name, pp->name, md.md_version, G_VIRSTOR_VERSION);
804 if (md.provsize != pp->mediasize)
807 /* If the provider name is hardcoded, use the offered provider only
808 * if it's been offered with its proper name (the one used in
809 * the label command). */
810 if (md.provider[0] != '\0') {
811 if (strcmp(md.provider, pp->name) != 0)
815 /* Iterate all geoms this class already knows about to see if a new
816 * geom instance of this class needs to be created (in case the provider
817 * is first from a (possibly) multi-consumer geom) or it just needs
818 * to be added to an existing instance. */
821 LIST_FOREACH(gp, &mp->geom, geom) {
825 if (strcmp(md.md_name, sc->geom->name) != 0)
827 if (md.md_id != sc->id)
831 if (gp != NULL) { /* We found an existing geom instance; add to it */
832 LOG_MSG(LVL_INFO, "Adding %s to %s", pp->name, md.md_name);
833 error = add_provider_to_geom(sc, pp, &md);
835 LOG_MSG(LVL_ERROR, "Error adding %s to %s (error %d)",
836 pp->name, md.md_name, error);
839 } else { /* New geom instance needs to be created */
840 gp = create_virstor_geom(mp, &md);
842 LOG_MSG(LVL_ERROR, "Error creating new instance of "
843 "class %s: %s", mp->name, md.md_name);
844 LOG_MSG(LVL_DEBUG, "Error creating %s at %s",
845 md.md_name, pp->name);
849 LOG_MSG(LVL_INFO, "Adding %s to %s (first found)", pp->name,
851 error = add_provider_to_geom(sc, pp, &md);
853 LOG_MSG(LVL_ERROR, "Error adding %s to %s (error %d)",
854 pp->name, md.md_name, error);
855 virstor_geom_destroy(sc, TRUE, FALSE);
864 * Destroyes consumer passed to it in arguments. Used as a callback
868 delay_destroy_consumer(void *arg, int flags __unused)
870 struct g_consumer *c = arg;
871 KASSERT(c != NULL, ("%s: invalid consumer", __func__));
872 LOG_MSG(LVL_DEBUG, "Consumer %s destroyed with delay",
875 g_destroy_consumer(c);
879 * Remove a component (consumer) from geom instance; If it's the first
880 * component being removed, orphan the provider to announce geom's being
884 remove_component(struct g_virstor_softc *sc, struct g_virstor_component *comp,
887 struct g_consumer *c;
889 KASSERT(comp->gcons != NULL, ("Component with no consumer in %s",
894 KASSERT(c->provider != NULL, ("%s: no provider", __func__));
895 LOG_MSG(LVL_DEBUG, "Component %s removed from %s", c->provider->name,
897 if (sc->provider != NULL) {
899 sc->provider->flags |= G_PF_WITHER;
900 g_orphan_provider(sc->provider, ENXIO);
902 LOG_MSG(LVL_INFO, "Removing provider %s", sc->geom->name);
905 if (c->acr > 0 || c->acw > 0 || c->ace > 0)
906 g_access(c, -c->acr, -c->acw, -c->ace);
908 /* Destroy consumer after it's tasted */
909 g_post_event(delay_destroy_consumer, c, M_WAITOK, NULL);
912 g_destroy_consumer(c);
917 * Destroy geom - called internally
918 * See g_virstor_destroy_geom for the other one
921 virstor_geom_destroy(struct g_virstor_softc *sc, boolean_t force,
924 struct g_provider *pp;
934 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
935 LOG_MSG(force ? LVL_WARNING : LVL_ERROR,
936 "Device %s is still open.", pp->name);
941 for (n = 0; n < sc->n_components; n++) {
942 if (sc->components[n].gcons != NULL)
943 remove_component(sc, &sc->components[n], delay);
949 KASSERT(sc->provider == NULL, ("Provider still exists for %s",
952 /* XXX: This might or might not work, since we're called with
953 * the topology lock held. Also, it might panic the kernel if
954 * the error'd BIO is in softupdates code. */
955 mtx_lock(&sc->delayed_bio_q_mtx);
956 while (!STAILQ_EMPTY(&sc->delayed_bio_q)) {
957 struct g_virstor_bio_q *bq;
958 bq = STAILQ_FIRST(&sc->delayed_bio_q);
959 bq->bio->bio_error = ENOSPC;
960 g_io_deliver(bq->bio, EIO);
961 STAILQ_REMOVE_HEAD(&sc->delayed_bio_q, linkage);
962 free(bq, M_GVIRSTOR);
964 mtx_unlock(&sc->delayed_bio_q_mtx);
965 mtx_destroy(&sc->delayed_bio_q_mtx);
967 free(sc->map, M_GVIRSTOR);
968 free(sc->components, M_GVIRSTOR);
969 bzero(sc, sizeof *sc);
970 free(sc, M_GVIRSTOR);
972 pp = LIST_FIRST(&gp->provider); /* We only offer one provider */
973 if (pp == NULL || (pp->acr == 0 && pp->acw == 0 && pp->ace == 0))
974 LOG_MSG(LVL_DEBUG, "Device %s destroyed", gp->name);
976 g_wither_geom(gp, ENXIO);
982 * Utility function: read metadata & decode. Wants topology lock to be
986 read_metadata(struct g_consumer *cp, struct g_virstor_metadata *md)
988 struct g_provider *pp;
993 error = g_access(cp, 1, 0, 0);
998 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
1001 g_access(cp, -1, 0, 0);
1005 virstor_metadata_decode(buf, md);
1012 * Utility function: encode & write metadata. Assumes topology lock is
1016 write_metadata(struct g_consumer *cp, struct g_virstor_metadata *md)
1018 struct g_provider *pp;
1022 KASSERT(cp != NULL && md != NULL && cp->provider != NULL,
1023 ("Something's fishy in %s", __func__));
1024 LOG_MSG(LVL_DEBUG, "Writing metadata on %s", cp->provider->name);
1025 g_topology_assert();
1026 error = g_access(cp, 0, 1, 0);
1031 buf = malloc(pp->sectorsize, M_GVIRSTOR, M_WAITOK);
1032 virstor_metadata_encode(md, buf);
1033 g_topology_unlock();
1034 error = g_write_data(cp, pp->mediasize - pp->sectorsize, buf,
1037 g_access(cp, 0, -1, 0);
1039 free(buf, M_GVIRSTOR);
1044 * Creates a new instance of this GEOM class, initialise softc
1046 static struct g_geom *
1047 create_virstor_geom(struct g_class *mp, struct g_virstor_metadata *md)
1050 struct g_virstor_softc *sc;
1052 LOG_MSG(LVL_DEBUG, "Creating geom instance for %s (id=%u)",
1053 md->md_name, md->md_id);
1055 if (md->md_count < 1 || md->md_chunk_size < 1 ||
1056 md->md_virsize < md->md_chunk_size) {
1057 /* This is bogus configuration, and probably means data is
1058 * somehow corrupted. Panic, maybe? */
1059 LOG_MSG(LVL_ERROR, "Nonsensical metadata information for %s",
1064 /* Check if it's already created */
1065 LIST_FOREACH(gp, &mp->geom, geom) {
1067 if (sc != NULL && strcmp(sc->geom->name, md->md_name) == 0) {
1068 LOG_MSG(LVL_WARNING, "Geom %s already exists",
1070 if (sc->id != md->md_id) {
1072 "Some stale or invalid components "
1073 "exist for virstor device named %s. "
1074 "You will need to <CLEAR> all stale "
1075 "components and maybe reconfigure "
1076 "the virstor device. Tune "
1077 "kern.geom.virstor.debug sysctl up "
1078 "for more information.",
1084 gp = g_new_geomf(mp, "%s", md->md_name);
1085 gp->softc = NULL; /* to circumevent races that test softc */
1087 gp->start = g_virstor_start;
1088 gp->spoiled = g_virstor_orphan;
1089 gp->orphan = g_virstor_orphan;
1090 gp->access = g_virstor_access;
1091 gp->dumpconf = g_virstor_dumpconf;
1093 sc = malloc(sizeof(*sc), M_GVIRSTOR, M_WAITOK | M_ZERO);
1095 sc->n_components = md->md_count;
1096 sc->components = malloc(sizeof(struct g_virstor_component) * md->md_count,
1097 M_GVIRSTOR, M_WAITOK | M_ZERO);
1098 sc->chunk_size = md->md_chunk_size;
1099 sc->virsize = md->md_virsize;
1100 STAILQ_INIT(&sc->delayed_bio_q);
1101 mtx_init(&sc->delayed_bio_q_mtx, "gvirstor_delayed_bio_q_mtx",
1102 "gvirstor", MTX_DEF | MTX_RECURSE);
1105 sc->provider = NULL; /* virstor_check_and_run will create it */
1108 LOG_MSG(LVL_ANNOUNCE, "Device %s created", sc->geom->name);
1114 * Add provider to a GEOM class instance
1117 add_provider_to_geom(struct g_virstor_softc *sc, struct g_provider *pp,
1118 struct g_virstor_metadata *md)
1120 struct g_virstor_component *component;
1121 struct g_consumer *cp, *fcp;
1125 if (md->no >= sc->n_components)
1128 /* "Current" compontent */
1129 component = &(sc->components[md->no]);
1130 if (component->gcons != NULL)
1134 fcp = LIST_FIRST(&gp->consumer);
1136 cp = g_new_consumer(gp);
1137 error = g_attach(cp, pp);
1140 g_destroy_consumer(cp);
1145 if (fcp->provider->sectorsize != pp->sectorsize) {
1146 /* TODO: this can be made to work */
1147 LOG_MSG(LVL_ERROR, "Provider %s of %s has invalid "
1148 "sector size (%d)", pp->name, sc->geom->name,
1152 if (fcp->acr > 0 || fcp->acw || fcp->ace > 0) {
1153 /* Replicate access permissions from first "live" consumer
1155 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace);
1158 g_destroy_consumer(cp);
1164 /* Bring up a new component */
1165 cp->private = component;
1166 component->gcons = cp;
1168 component->index = md->no;
1169 component->chunk_count = md->chunk_count;
1170 component->chunk_next = md->chunk_next;
1171 component->chunk_reserved = md->chunk_reserved;
1172 component->flags = md->flags;
1174 LOG_MSG(LVL_DEBUG, "%s attached to %s", pp->name, sc->geom->name);
1176 virstor_check_and_run(sc);
1181 * Check if everything's ready to create the geom provider & device entry,
1182 * create and start provider.
1183 * Called ultimately by .taste, from g_event thread
1186 virstor_check_and_run(struct g_virstor_softc *sc)
1193 if (virstor_valid_components(sc) != sc->n_components)
1196 if (virstor_valid_components(sc) == 0) {
1197 /* This is actually a candidate for panic() */
1198 LOG_MSG(LVL_ERROR, "No valid components for %s?",
1199 sc->provider->name);
1203 sc->sectorsize = sc->components[0].gcons->provider->sectorsize;
1205 /* Initialise allocation map from the first consumer */
1206 sc->chunk_count = sc->virsize / sc->chunk_size;
1207 if (sc->chunk_count * (off_t)sc->chunk_size != sc->virsize) {
1208 LOG_MSG(LVL_WARNING, "Device %s truncated to %ju bytes",
1210 sc->chunk_count * (off_t)sc->chunk_size);
1212 sc->map_size = sc->chunk_count * sizeof *(sc->map);
1213 /* The following allocation is in order of 4MB - 8MB */
1214 sc->map = malloc(sc->map_size, M_GVIRSTOR, M_WAITOK);
1215 KASSERT(sc->map != NULL, ("%s: Memory allocation error (%zu bytes) for %s",
1216 __func__, sc->map_size, sc->provider->name));
1217 sc->map_sectors = sc->map_size / sc->sectorsize;
1220 for (n = 0; n < sc->n_components; n++)
1221 count += sc->components[n].chunk_count;
1222 LOG_MSG(LVL_INFO, "Device %s has %zu physical chunks and %zu virtual "
1224 sc->geom->name, count, sc->chunk_count, sc->chunk_size / 1024);
1226 error = g_access(sc->components[0].gcons, 1, 0, 0);
1228 LOG_MSG(LVL_ERROR, "Cannot acquire read access for %s to "
1229 "read allocation map for %s",
1230 sc->components[0].gcons->provider->name,
1234 /* Read in the allocation map */
1235 LOG_MSG(LVL_DEBUG, "Reading map for %s from %s", sc->geom->name,
1236 sc->components[0].gcons->provider->name);
1237 off = count = n = 0;
1238 while (count < sc->map_size) {
1239 struct g_virstor_map_entry *mapbuf;
1242 bs = MIN(MAXPHYS, sc->map_size - count);
1243 if (bs % sc->sectorsize != 0) {
1244 /* Check for alignment errors */
1245 bs = (bs / sc->sectorsize) * sc->sectorsize;
1248 LOG_MSG(LVL_ERROR, "Trouble: map is not sector-aligned "
1249 "for %s on %s", sc->geom->name,
1250 sc->components[0].gcons->provider->name);
1252 mapbuf = g_read_data(sc->components[0].gcons, off, bs, &error);
1253 if (mapbuf == NULL) {
1254 free(sc->map, M_GVIRSTOR);
1255 LOG_MSG(LVL_ERROR, "Error reading allocation map "
1256 "for %s from %s (offset %ju) (error %d)",
1258 sc->components[0].gcons->provider->name,
1263 bcopy(mapbuf, &sc->map[n], bs);
1266 n += bs / sizeof *(sc->map);
1269 g_access(sc->components[0].gcons, -1, 0, 0);
1270 LOG_MSG(LVL_DEBUG, "Read map for %s", sc->geom->name);
1272 /* find first component with allocatable chunks */
1274 for (n = 0; n < sc->n_components; n++) {
1275 if (sc->components[n].chunk_next <
1276 sc->components[n].chunk_count) {
1282 /* not found? set it to the last component and handle it
1284 index = sc->n_components - 1;
1286 if (index >= sc->n_components - g_virstor_component_watermark - 1) {
1287 LOG_MSG(LVL_WARNING, "Device %s running out of components "
1288 "(%d/%u: %s)", sc->geom->name,
1291 sc->components[index].gcons->provider->name);
1293 sc->curr_component = index;
1295 if (sc->components[index].chunk_next >=
1296 sc->components[index].chunk_count - g_virstor_chunk_watermark) {
1297 LOG_MSG(LVL_WARNING,
1298 "Component %s of %s is running out of free space "
1300 sc->components[index].gcons->provider->name,
1301 sc->geom->name, sc->components[index].chunk_count -
1302 sc->components[index].chunk_next);
1305 sc->me_per_sector = sc->sectorsize / sizeof *(sc->map);
1306 if (sc->sectorsize % sizeof *(sc->map) != 0) {
1308 "%s: Map entries don't fit exactly in a sector (%s)",
1309 __func__, sc->geom->name);
1313 /* Recalculate allocated chunks in components & at the same time
1314 * verify map data is sane. We could trust metadata on this, but
1315 * we want to make sure. */
1316 for (n = 0; n < sc->n_components; n++)
1317 sc->components[n].chunk_next = sc->components[n].chunk_reserved;
1319 for (n = 0; n < sc->chunk_count; n++) {
1320 if (sc->map[n].provider_no >= sc->n_components ||
1321 sc->map[n].provider_chunk >=
1322 sc->components[sc->map[n].provider_no].chunk_count) {
1323 LOG_MSG(LVL_ERROR, "%s: Invalid entry %u in map for %s",
1324 __func__, (u_int)n, sc->geom->name);
1325 LOG_MSG(LVL_ERROR, "%s: provider_no: %u, n_components: %u"
1326 " provider_chunk: %u, chunk_count: %u", __func__,
1327 sc->map[n].provider_no, sc->n_components,
1328 sc->map[n].provider_chunk,
1329 sc->components[sc->map[n].provider_no].chunk_count);
1332 if (sc->map[n].flags & VIRSTOR_MAP_ALLOCATED)
1333 sc->components[sc->map[n].provider_no].chunk_next++;
1336 sc->provider = g_new_providerf(sc->geom, "virstor/%s",
1339 sc->provider->sectorsize = sc->sectorsize;
1340 sc->provider->mediasize = sc->virsize;
1341 g_error_provider(sc->provider, 0);
1343 LOG_MSG(LVL_INFO, "%s activated", sc->provider->name);
1344 LOG_MSG(LVL_DEBUG, "%s starting with current component %u, starting "
1345 "chunk %u", sc->provider->name, sc->curr_component,
1346 sc->components[sc->curr_component].chunk_next);
1350 * Returns count of active providers in this geom instance
1353 virstor_valid_components(struct g_virstor_softc *sc)
1358 KASSERT(sc != NULL, ("%s: softc is NULL", __func__));
1359 KASSERT(sc->components != NULL, ("%s: sc->components is NULL", __func__));
1360 for (i = 0; i < sc->n_components; i++)
1361 if (sc->components[i].gcons != NULL)
1367 * Called when the consumer gets orphaned (?)
1370 g_virstor_orphan(struct g_consumer *cp)
1372 struct g_virstor_softc *sc;
1373 struct g_virstor_component *comp;
1376 g_topology_assert();
1383 KASSERT(comp != NULL, ("%s: No component in private part of consumer",
1385 remove_component(sc, comp, FALSE);
1386 if (virstor_valid_components(sc) == 0)
1387 virstor_geom_destroy(sc, TRUE, FALSE);
1391 * Called to notify geom when it's been opened, and for what intent
1394 g_virstor_access(struct g_provider *pp, int dr, int dw, int de)
1396 struct g_consumer *c;
1397 struct g_virstor_softc *sc;
1401 KASSERT(pp != NULL, ("%s: NULL provider", __func__));
1403 KASSERT(gp != NULL, ("%s: NULL geom", __func__));
1407 /* It seems that .access can be called with negative dr,dw,dx
1408 * in this case but I want to check for myself */
1409 LOG_MSG(LVL_WARNING, "access(%d, %d, %d) for %s",
1410 dr, dw, de, pp->name);
1411 /* This should only happen when geom is withered so
1412 * allow only negative requests */
1413 KASSERT(dr <= 0 && dw <= 0 && de <= 0,
1414 ("%s: Positive access for %s", __func__, pp->name));
1415 if (pp->acr + dr == 0 && pp->acw + dw == 0 && pp->ace + de == 0)
1416 LOG_MSG(LVL_DEBUG, "Device %s definitely destroyed",
1421 /* Grab an exclusive bit to propagate on our consumers on first open */
1422 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
1424 /* ... drop it on close */
1425 if (pp->acr + dr == 0 && pp->acw + dw == 0 && pp->ace + de == 0) {
1427 update_metadata(sc); /* Writes statistical information */
1431 LIST_FOREACH(c, &gp->consumer, consumer) {
1432 KASSERT(c != NULL, ("%s: consumer is NULL", __func__));
1433 error = g_access(c, dr, dw, de);
1435 struct g_consumer *c2;
1437 /* Backout earlier changes */
1438 LIST_FOREACH(c2, &gp->consumer, consumer) {
1439 if (c2 == c) /* all eariler components fixed */
1441 g_access(c2, -dr, -dw, -de);
1450 * Generate XML dump of current state
1453 g_virstor_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1454 struct g_consumer *cp, struct g_provider *pp)
1456 struct g_virstor_softc *sc;
1458 g_topology_assert();
1461 if (sc == NULL || pp != NULL)
1465 /* For each component */
1466 struct g_virstor_component *comp;
1471 sbuf_printf(sb, "%s<ComponentIndex>%u</ComponentIndex>\n",
1472 indent, comp->index);
1473 sbuf_printf(sb, "%s<ChunkCount>%u</ChunkCount>\n",
1474 indent, comp->chunk_count);
1475 sbuf_printf(sb, "%s<ChunksUsed>%u</ChunksUsed>\n",
1476 indent, comp->chunk_next);
1477 sbuf_printf(sb, "%s<ChunksReserved>%u</ChunksReserved>\n",
1478 indent, comp->chunk_reserved);
1479 sbuf_printf(sb, "%s<StorageFree>%u%%</StorageFree>\n",
1481 comp->chunk_next > 0 ? 100 -
1482 ((comp->chunk_next + comp->chunk_reserved) * 100) /
1483 comp->chunk_count : 100);
1485 /* For the whole thing */
1486 u_int count, used, i;
1489 count = used = size = 0;
1490 for (i = 0; i < sc->n_components; i++) {
1491 if (sc->components[i].gcons != NULL) {
1492 count += sc->components[i].chunk_count;
1493 used += sc->components[i].chunk_next +
1494 sc->components[i].chunk_reserved;
1495 size += sc->components[i].gcons->
1496 provider->mediasize;
1500 sbuf_printf(sb, "%s<Status>"
1501 "Components=%u, Online=%u</Status>\n", indent,
1502 sc->n_components, virstor_valid_components(sc));
1503 sbuf_printf(sb, "%s<State>%u%% physical free</State>\n",
1504 indent, 100-(used * 100) / count);
1505 sbuf_printf(sb, "%s<ChunkSize>%zu</ChunkSize>\n", indent,
1507 sbuf_printf(sb, "%s<PhysicalFree>%u%%</PhysicalFree>\n",
1508 indent, used > 0 ? 100 - (used * 100) / count : 100);
1509 sbuf_printf(sb, "%s<ChunkPhysicalCount>%u</ChunkPhysicalCount>\n",
1511 sbuf_printf(sb, "%s<ChunkVirtualCount>%zu</ChunkVirtualCount>\n",
1512 indent, sc->chunk_count);
1513 sbuf_printf(sb, "%s<PhysicalBacking>%zu%%</PhysicalBacking>\n",
1515 (count * 100) / sc->chunk_count);
1516 sbuf_printf(sb, "%s<PhysicalBackingSize>%jd</PhysicalBackingSize>\n",
1518 sbuf_printf(sb, "%s<VirtualSize>%jd</VirtualSize>\n", indent,
1524 * GEOM .done handler
1525 * Can't use standard handler because one requested IO may
1526 * fork into additional data IOs
1529 g_virstor_done(struct bio *b)
1531 struct g_virstor_softc *sc;
1532 struct bio *parent_b;
1534 parent_b = b->bio_parent;
1535 sc = parent_b->bio_to->geom->softc;
1537 if (b->bio_error != 0) {
1538 LOG_MSG(LVL_ERROR, "Error %d for offset=%ju, length=%ju, %s",
1539 b->bio_error, b->bio_offset, b->bio_length,
1541 if (parent_b->bio_error == 0)
1542 parent_b->bio_error = b->bio_error;
1545 parent_b->bio_inbed++;
1546 parent_b->bio_completed += b->bio_completed;
1548 if (parent_b->bio_children == parent_b->bio_inbed) {
1549 parent_b->bio_completed = parent_b->bio_length;
1550 g_io_deliver(parent_b, parent_b->bio_error);
1557 * Called in g_down thread
1560 g_virstor_start(struct bio *b)
1562 struct g_virstor_softc *sc;
1563 struct g_virstor_component *comp;
1565 struct g_provider *pp;
1567 off_t offset, length;
1568 struct bio_queue_head bq;
1569 size_t chunk_size; /* cached for convenience */
1573 sc = pp->geom->softc;
1574 KASSERT(sc != NULL, ("%s: no softc (error=%d, device=%s)", __func__,
1575 b->bio_to->error, b->bio_to->name));
1577 LOG_REQ(LVL_MOREDEBUG, b, "%s", __func__);
1579 switch (b->bio_cmd) {
1585 g_io_deliver(b, EOPNOTSUPP);
1589 LOG_MSG(LVL_DEBUG2, "BIO arrived, size=%ju", b->bio_length);
1592 chunk_size = sc->chunk_size;
1594 offset = b->bio_offset; /* virtual offset and length */
1595 length = b->bio_length;
1597 while (length > 0) {
1598 size_t chunk_index, in_chunk_offset, in_chunk_length;
1599 struct virstor_map_entry *me;
1601 chunk_index = offset / chunk_size; /* round downwards */
1602 in_chunk_offset = offset % chunk_size;
1603 in_chunk_length = min(length, chunk_size - in_chunk_offset);
1604 LOG_MSG(LVL_DEBUG, "Mapped %s(%ju, %ju) to (%zu,%zu,%zu)",
1605 b->bio_cmd == BIO_READ ? "R" : "W",
1607 chunk_index, in_chunk_offset, in_chunk_length);
1608 me = &sc->map[chunk_index];
1610 if (b->bio_cmd == BIO_READ || b->bio_cmd == BIO_DELETE) {
1611 if ((me->flags & VIRSTOR_MAP_ALLOCATED) == 0) {
1612 /* Reads from unallocated chunks return zeroed
1614 if (b->bio_cmd == BIO_READ)
1615 bzero(addr, in_chunk_length);
1617 comp = &sc->components[me->provider_no];
1619 cb = g_clone_bio(b);
1621 bioq_dismantle(&bq);
1622 if (b->bio_error == 0)
1623 b->bio_error = ENOMEM;
1624 g_io_deliver(b, b->bio_error);
1627 cb->bio_to = comp->gcons->provider;
1628 cb->bio_done = g_virstor_done;
1630 (off_t)me->provider_chunk * (off_t)chunk_size
1632 cb->bio_length = in_chunk_length;
1633 cb->bio_data = addr;
1634 cb->bio_caller1 = comp;
1635 bioq_disksort(&bq, cb);
1637 } else { /* handle BIO_WRITE */
1638 KASSERT(b->bio_cmd == BIO_WRITE,
1639 ("%s: Unknown command %d", __func__,
1642 if ((me->flags & VIRSTOR_MAP_ALLOCATED) == 0) {
1643 /* We have a virtual chunk, represented by
1644 * the "me" entry, but it's not yet allocated
1645 * (tied to) a physical chunk. So do it now. */
1646 struct virstor_map_entry *data_me;
1647 u_int phys_chunk, comp_no;
1651 error = allocate_chunk(sc, &comp, &comp_no,
1654 /* We cannot allocate a physical chunk
1655 * to satisfy this request, so we'll
1656 * delay it to when we can...
1657 * XXX: this will prevent the fs from
1658 * being umounted! */
1659 struct g_virstor_bio_q *biq;
1660 biq = malloc(sizeof *biq, M_GVIRSTOR,
1663 bioq_dismantle(&bq);
1664 if (b->bio_error == 0)
1665 b->bio_error = ENOMEM;
1666 g_io_deliver(b, b->bio_error);
1670 mtx_lock(&sc->delayed_bio_q_mtx);
1671 STAILQ_INSERT_TAIL(&sc->delayed_bio_q,
1673 mtx_unlock(&sc->delayed_bio_q_mtx);
1674 LOG_MSG(LVL_WARNING, "Delaying BIO "
1675 "(size=%ju) until free physical "
1676 "space can be found on %s",
1678 sc->provider->name);
1681 LOG_MSG(LVL_DEBUG, "Allocated chunk %u on %s "
1684 comp->gcons->provider->name,
1685 sc->provider->name);
1687 me->provider_no = comp_no;
1688 me->provider_chunk = phys_chunk;
1689 me->flags |= VIRSTOR_MAP_ALLOCATED;
1691 cb = g_clone_bio(b);
1693 me->flags &= ~VIRSTOR_MAP_ALLOCATED;
1694 me->provider_no = 0;
1695 me->provider_chunk = 0;
1696 bioq_dismantle(&bq);
1697 if (b->bio_error == 0)
1698 b->bio_error = ENOMEM;
1699 g_io_deliver(b, b->bio_error);
1703 /* The allocation table is stored continuously
1704 * at the start of the drive. We need to
1705 * calculate the offset of the sector that holds
1706 * this map entry both on the drive and in the
1708 * sc_offset will end up pointing to the drive
1710 s_offset = chunk_index * sizeof *me;
1711 s_offset = (s_offset / sc->sectorsize) *
1714 /* data_me points to map entry sector
1715 * in memory (analoguos to offset) */
1716 data_me = &sc->map[(chunk_index /
1717 sc->me_per_sector) * sc->me_per_sector];
1719 /* Commit sector with map entry to storage */
1720 cb->bio_to = sc->components[0].gcons->provider;
1721 cb->bio_done = g_virstor_done;
1722 cb->bio_offset = s_offset;
1723 cb->bio_data = (char *)data_me;
1724 cb->bio_length = sc->sectorsize;
1725 cb->bio_caller1 = &sc->components[0];
1726 bioq_disksort(&bq, cb);
1729 comp = &sc->components[me->provider_no];
1730 cb = g_clone_bio(b);
1732 bioq_dismantle(&bq);
1733 if (b->bio_error == 0)
1734 b->bio_error = ENOMEM;
1735 g_io_deliver(b, b->bio_error);
1738 /* Finally, handle the data */
1739 cb->bio_to = comp->gcons->provider;
1740 cb->bio_done = g_virstor_done;
1741 cb->bio_offset = (off_t)me->provider_chunk*(off_t)chunk_size +
1743 cb->bio_length = in_chunk_length;
1744 cb->bio_data = addr;
1745 cb->bio_caller1 = comp;
1746 bioq_disksort(&bq, cb);
1748 addr += in_chunk_length;
1749 length -= in_chunk_length;
1750 offset += in_chunk_length;
1753 /* Fire off bio's here */
1755 for (cb = bioq_first(&bq); cb != NULL; cb = bioq_first(&bq)) {
1756 bioq_remove(&bq, cb);
1757 LOG_REQ(LVL_MOREDEBUG, cb, "Firing request");
1758 comp = cb->bio_caller1;
1759 cb->bio_caller1 = NULL;
1760 LOG_MSG(LVL_DEBUG, " firing bio, offset=%ju, length=%ju",
1761 cb->bio_offset, cb->bio_length);
1762 g_io_request(cb, comp->gcons);
1765 if (count == 0) { /* We handled everything locally */
1766 b->bio_completed = b->bio_length;
1773 * Allocate a chunk from a physical provider. Returns physical component,
1774 * chunk index relative to the component and the component's index.
1777 allocate_chunk(struct g_virstor_softc *sc, struct g_virstor_component **comp,
1778 u_int *comp_no_p, u_int *chunk)
1782 KASSERT(sc->curr_component < sc->n_components,
1783 ("%s: Invalid curr_component: %u", __func__, sc->curr_component));
1785 comp_no = sc->curr_component;
1786 *comp = &sc->components[comp_no];
1787 dump_component(*comp);
1788 if ((*comp)->chunk_next >= (*comp)->chunk_count) {
1789 /* This component is full. Allocate next component */
1790 if (comp_no >= sc->n_components-1) {
1791 LOG_MSG(LVL_ERROR, "All physical space allocated for %s",
1795 (*comp)->flags &= ~VIRSTOR_PROVIDER_CURRENT;
1796 sc->curr_component = ++comp_no;
1798 *comp = &sc->components[comp_no];
1799 if (comp_no >= sc->n_components - g_virstor_component_watermark-1)
1800 LOG_MSG(LVL_WARNING, "Device %s running out of components "
1801 "(switching to %u/%u: %s)", sc->geom->name,
1802 comp_no+1, sc->n_components,
1803 (*comp)->gcons->provider->name);
1804 /* Take care not to overwrite reserved chunks */
1805 if ( (*comp)->chunk_reserved > 0 &&
1806 (*comp)->chunk_next < (*comp)->chunk_reserved)
1807 (*comp)->chunk_next = (*comp)->chunk_reserved;
1810 VIRSTOR_PROVIDER_ALLOCATED | VIRSTOR_PROVIDER_CURRENT;
1811 dump_component(*comp);
1812 *comp_no_p = comp_no;
1813 *chunk = (*comp)->chunk_next++;
1815 *comp_no_p = comp_no;
1816 *chunk = (*comp)->chunk_next++;
1821 /* Dump a component */
1823 dump_component(struct g_virstor_component *comp)
1826 if (g_virstor_debug < LVL_DEBUG2)
1828 printf("Component %d: %s\n", comp->index, comp->gcons->provider->name);
1829 printf(" chunk_count: %u\n", comp->chunk_count);
1830 printf(" chunk_next: %u\n", comp->chunk_next);
1831 printf(" flags: %u\n", comp->flags);
1835 /* Dump a map entry */
1837 dump_me(struct virstor_map_entry *me, unsigned int nr)
1839 if (g_virstor_debug < LVL_DEBUG)
1841 printf("VIRT. CHUNK #%d: ", nr);
1842 if ((me->flags & VIRSTOR_MAP_ALLOCATED) == 0)
1843 printf("(unallocated)\n");
1845 printf("allocated at provider %u, provider_chunk %u\n",
1846 me->provider_no, me->provider_chunk);
1851 * Dismantle bio_queue and destroy its components
1854 bioq_dismantle(struct bio_queue_head *bq)
1858 for (b = bioq_first(bq); b != NULL; b = bioq_first(bq)) {
1865 * The function that shouldn't be called.
1866 * When this is called, the stack is already garbled because of
1867 * argument mismatch. There's nothing to do now but panic, which is
1868 * accidentally the whole purpose of this function.
1869 * Motivation: to guard from accidentally calling geom methods when
1870 * they shouldn't be called. (see g_..._taste)
1875 panic("invalid_call() has just been called. Something's fishy here.");
1878 DECLARE_GEOM_CLASS(g_virstor_class, g_virstor); /* Let there be light */