1/*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2003 Poul-Henning Kamp 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. The names of the authors may not be used to endorse or promote 16 * products derived from this software without specific prior written 17 * permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * $FreeBSD$ 32 */ 33 34/* This is a GEOM module for handling path selection for multi-path 35 * storage devices. It is named "fox" because it, like they, prefer 36 * to have multiple exits to choose from. 37 * 38 */ 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/sysctl.h> 43#include <sys/kernel.h> 44#include <sys/conf.h> 45#include <sys/bio.h> 46#include <sys/malloc.h> 47#include <sys/lock.h> 48#include <sys/mutex.h> 49#include <sys/libkern.h> 50#include <sys/endian.h> 51#include <sys/md5.h> 52#include <sys/errno.h> 53#include <geom/geom.h> 54 55#define FOX_CLASS_NAME "FOX" 56#define FOX_MAGIC "GEOM::FOX" 57 58static int g_fox_once; 59 60FEATURE(geom_fox, "GEOM FOX redundant path mitigation support"); 61 62struct g_fox_softc { 63 off_t mediasize; 64 u_int sectorsize; 65 TAILQ_HEAD(, bio) queue; 66 struct mtx lock; 67 u_char magic[16]; 68 struct g_consumer *path; 69 struct g_consumer *opath; 70 int waiting; 71 int cr, cw, ce; 72}; 73 74/* 75 * This function is called whenever we need to select a new path. 76 */ 77static void 78g_fox_select_path(void *arg, int flag) 79{ 80 struct g_geom *gp; 81 struct g_fox_softc *sc; 82 struct g_consumer *cp1; 83 struct bio *bp; 84 int error; 85 86 g_topology_assert(); 87 if (flag == EV_CANCEL) 88 return; 89 gp = arg; 90 sc = gp->softc; 91 92 if (sc->opath != NULL) { 93 /* 94 * First, close the old path entirely. 95 */ 96 printf("Closing old path (%s) on fox (%s)\n", 97 sc->opath->provider->name, gp->name); 98 99 cp1 = LIST_NEXT(sc->opath, consumer); 100 101 g_access(sc->opath, -sc->cr, -sc->cw, -(sc->ce + 1)); 102 103 /* 104 * The attempt to reopen it with a exclusive count 105 */ 106 error = g_access(sc->opath, 0, 0, 1); 107 if (error) { 108 /* 109 * Ok, ditch this consumer, we can't use it. 110 */ 111 printf("Drop old path (%s) on fox (%s)\n", 112 sc->opath->provider->name, gp->name); 113 g_detach(sc->opath); 114 g_destroy_consumer(sc->opath); 115 if (LIST_EMPTY(&gp->consumer)) { 116 /* No consumers left */ 117 g_wither_geom(gp, ENXIO); 118 for (;;) { 119 bp = TAILQ_FIRST(&sc->queue); 120 if (bp == NULL) 121 break; 122 TAILQ_REMOVE(&sc->queue, bp, bio_queue); 123 bp->bio_error = ENXIO; 124 g_std_done(bp); 125 } 126 return; 127 } 128 } else { 129 printf("Got e-bit on old path (%s) on fox (%s)\n", 130 sc->opath->provider->name, gp->name); 131 } 132 sc->opath = NULL; 133 } else { 134 cp1 = LIST_FIRST(&gp->consumer); 135 } 136 if (cp1 == NULL) 137 cp1 = LIST_FIRST(&gp->consumer); 138 printf("Open new path (%s) on fox (%s)\n", 139 cp1->provider->name, gp->name); 140 error = g_access(cp1, sc->cr, sc->cw, sc->ce); 141 if (error) { 142 /* 143 * If we failed, we take another trip through here 144 */ 145 printf("Open new path (%s) on fox (%s) failed, reselect.\n", 146 cp1->provider->name, gp->name); 147 sc->opath = cp1; 148 g_post_event(g_fox_select_path, gp, M_WAITOK, gp, NULL); 149 } else { 150 printf("Open new path (%s) on fox (%s) succeeded\n", 151 cp1->provider->name, gp->name); 152 mtx_lock(&sc->lock); 153 sc->path = cp1; 154 sc->waiting = 0; 155 for (;;) { 156 bp = TAILQ_FIRST(&sc->queue); 157 if (bp == NULL) 158 break; 159 TAILQ_REMOVE(&sc->queue, bp, bio_queue); 160 g_io_request(bp, sc->path); 161 } 162 mtx_unlock(&sc->lock); 163 } 164} 165 166static void 167g_fox_orphan(struct g_consumer *cp) 168{ 169 struct g_geom *gp; 170 struct g_fox_softc *sc; 171 int error, mark; 172 173 g_topology_assert(); 174 gp = cp->geom; 175 sc = gp->softc; 176 printf("Removing path (%s) from fox (%s)\n", 177 cp->provider->name, gp->name); 178 mtx_lock(&sc->lock); 179 if (cp == sc->path) { 180 sc->opath = NULL; 181 sc->path = NULL; 182 sc->waiting = 1; 183 mark = 1; 184 } else { 185 mark = 0; 186 } 187 mtx_unlock(&sc->lock); 188 189 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 190 error = cp->provider->error; 191 g_detach(cp); 192 g_destroy_consumer(cp); 193 if (!LIST_EMPTY(&gp->consumer)) { 194 if (mark) 195 g_post_event(g_fox_select_path, gp, M_WAITOK, gp, NULL); 196 return; 197 } 198 199 mtx_destroy(&sc->lock); 200 g_free(gp->softc); 201 gp->softc = NULL; 202 g_wither_geom(gp, ENXIO); 203} 204 205static void 206g_fox_done(struct bio *bp) 207{ 208 struct g_geom *gp; 209 struct g_fox_softc *sc; 210 int error; 211 212 if (bp->bio_error == 0) { 213 g_std_done(bp); 214 return; 215 } 216 gp = bp->bio_from->geom; 217 sc = gp->softc; 218 if (bp->bio_from != sc->path) { 219 g_io_request(bp, sc->path); 220 return; 221 } 222 mtx_lock(&sc->lock); 223 sc->opath = sc->path; 224 sc->path = NULL; 225 error = g_post_event(g_fox_select_path, gp, M_NOWAIT, gp, NULL); 226 if (error) { 227 bp->bio_error = ENOMEM; 228 g_std_done(bp); 229 } else { 230 sc->waiting = 1; 231 TAILQ_INSERT_TAIL(&sc->queue, bp, bio_queue); 232 } 233 mtx_unlock(&sc->lock); 234} 235 236static void 237g_fox_start(struct bio *bp) 238{ 239 struct g_geom *gp; 240 struct bio *bp2; 241 struct g_fox_softc *sc; 242 int error; 243 244 gp = bp->bio_to->geom; 245 sc = gp->softc; 246 if (sc == NULL) { 247 g_io_deliver(bp, ENXIO); 248 return; 249 } 250 switch(bp->bio_cmd) { 251 case BIO_READ: 252 case BIO_WRITE: 253 case BIO_DELETE: 254 bp2 = g_clone_bio(bp); 255 if (bp2 == NULL) { 256 g_io_deliver(bp, ENOMEM); 257 break; 258 } 259 bp2->bio_offset += sc->sectorsize; 260 bp2->bio_done = g_fox_done; 261 mtx_lock(&sc->lock); 262 if (sc->path == NULL || !TAILQ_EMPTY(&sc->queue)) { 263 if (sc->waiting == 0) { 264 error = g_post_event(g_fox_select_path, gp, 265 M_NOWAIT, gp, NULL); 266 if (error) { 267 g_destroy_bio(bp2); 268 bp2 = NULL; 269 g_io_deliver(bp, error); 270 } else { 271 sc->waiting = 1; 272 } 273 } 274 if (bp2 != NULL) 275 TAILQ_INSERT_TAIL(&sc->queue, bp2, 276 bio_queue); 277 } else { 278 g_io_request(bp2, sc->path); 279 } 280 mtx_unlock(&sc->lock); 281 break; 282 default: 283 g_io_deliver(bp, EOPNOTSUPP); 284 break; 285 } 286 return; 287} 288 289static int 290g_fox_access(struct g_provider *pp, int dr, int dw, int de) 291{ 292 struct g_geom *gp; 293 struct g_fox_softc *sc; 294 struct g_consumer *cp1; 295 int error; 296 297 g_topology_assert(); 298 gp = pp->geom; 299 sc = gp->softc; 300 if (sc == NULL) { 301 if (dr <= 0 && dw <= 0 && de <= 0) 302 return (0); 303 else 304 return (ENXIO); 305 } 306 307 if (sc->cr == 0 && sc->cw == 0 && sc->ce == 0) { 308 /* 309 * First open, open all consumers with an exclusive bit 310 */ 311 error = 0; 312 LIST_FOREACH(cp1, &gp->consumer, consumer) { 313 error = g_access(cp1, 0, 0, 1); 314 if (error) { 315 printf("FOX: access(%s,0,0,1) = %d\n", 316 cp1->provider->name, error); 317 break; 318 } 319 } 320 if (error) { 321 LIST_FOREACH(cp1, &gp->consumer, consumer) { 322 if (cp1->ace) 323 g_access(cp1, 0, 0, -1); 324 } 325 return (error); 326 } 327 } 328 if (sc->path == NULL) 329 g_fox_select_path(gp, 0); 330 if (sc->path == NULL) 331 error = ENXIO; 332 else 333 error = g_access(sc->path, dr, dw, de); 334 if (error == 0) { 335 sc->cr += dr; 336 sc->cw += dw; 337 sc->ce += de; 338 if (sc->cr == 0 && sc->cw == 0 && sc->ce == 0) { 339 /* 340 * Last close, remove e-bit on all consumers 341 */ 342 LIST_FOREACH(cp1, &gp->consumer, consumer) 343 g_access(cp1, 0, 0, -1); 344 } 345 } 346 return (error); 347} 348 349static struct g_geom * 350g_fox_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 351{ 352 struct g_geom *gp, *gp2; 353 struct g_provider *pp2; 354 struct g_consumer *cp, *cp2; 355 struct g_fox_softc *sc, *sc2; 356 int error; 357 u_int sectorsize; 358 u_char *buf; 359 360 g_trace(G_T_TOPOLOGY, "fox_taste(%s, %s)", mp->name, pp->name); 361 g_topology_assert(); 362 if (!strcmp(pp->geom->class->name, mp->name)) 363 return (NULL); 364 gp = g_new_geomf(mp, "%s.fox", pp->name); 365 gp->softc = g_malloc(sizeof(struct g_fox_softc), M_WAITOK | M_ZERO); 366 sc = gp->softc; 367 368 cp = g_new_consumer(gp); 369 g_attach(cp, pp); 370 error = g_access(cp, 1, 0, 0); 371 if (error) { 372 g_free(sc); 373 g_detach(cp); 374 g_destroy_consumer(cp); 375 g_destroy_geom(gp); 376 return(NULL); 377 } 378 do { 379 sectorsize = cp->provider->sectorsize; 380 g_topology_unlock(); 381 buf = g_read_data(cp, 0, sectorsize, NULL); 382 g_topology_lock(); 383 if (buf == NULL) 384 break; 385 if (memcmp(buf, FOX_MAGIC, strlen(FOX_MAGIC))) 386 break; 387 388 /* 389 * First we need to see if this a new path for an existing fox. 390 */ 391 LIST_FOREACH(gp2, &mp->geom, geom) { 392 sc2 = gp2->softc; 393 if (sc2 == NULL) 394 continue; 395 if (memcmp(buf + 16, sc2->magic, sizeof sc2->magic)) 396 continue; 397 break; 398 } 399 if (gp2 != NULL) { 400 /* 401 * It was. Create a new consumer for that fox, 402 * attach it, and if the fox is open, open this 403 * path with an exclusive count of one. 404 */ 405 printf("Adding path (%s) to fox (%s)\n", 406 pp->name, gp2->name); 407 cp2 = g_new_consumer(gp2); 408 g_attach(cp2, pp); 409 pp2 = LIST_FIRST(&gp2->provider); 410 if (pp2->acr > 0 || pp2->acw > 0 || pp2->ace > 0) { 411 error = g_access(cp2, 0, 0, 1); 412 if (error) { 413 /* 414 * This is bad, or more likely, 415 * the user is doing something stupid 416 */ 417 printf( 418 "WARNING: New path (%s) to fox(%s) not added: %s\n%s", 419 cp2->provider->name, gp2->name, 420 "Could not get exclusive bit.", 421 "WARNING: This indicates a risk of data inconsistency." 422 ); 423 g_detach(cp2); 424 g_destroy_consumer(cp2); 425 } 426 } 427 break; 428 } 429 printf("Creating new fox (%s)\n", pp->name); 430 sc->path = cp; 431 memcpy(sc->magic, buf + 16, sizeof sc->magic); 432 pp2 = g_new_providerf(gp, "%s", gp->name); 433 pp2->mediasize = sc->mediasize = pp->mediasize - pp->sectorsize; 434 pp2->sectorsize = sc->sectorsize = pp->sectorsize; 435printf("fox %s lock %p\n", gp->name, &sc->lock); 436 437 mtx_init(&sc->lock, "fox queue", NULL, MTX_DEF); 438 TAILQ_INIT(&sc->queue); 439 g_error_provider(pp2, 0); 440 } while (0); 441 if (buf != NULL) 442 g_free(buf); 443 g_access(cp, -1, 0, 0); 444 445 if (!LIST_EMPTY(&gp->provider)) { 446 if (!g_fox_once) { 447 g_fox_once = 1; 448 printf( 449 "WARNING: geom_fox (geom %s) is deprecated, " 450 "use gmultipath instead.\n", gp->name); 451 } 452 return (gp); 453 } 454 455 g_free(gp->softc); 456 g_detach(cp); 457 g_destroy_consumer(cp); 458 g_destroy_geom(gp); 459 return (NULL); 460} 461 462static int 463g_fox_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp) 464{ 465 struct g_fox_softc *sc; 466 467 g_topology_assert(); 468 sc = gp->softc; 469 mtx_destroy(&sc->lock); 470 g_free(gp->softc); 471 gp->softc = NULL; 472 g_wither_geom(gp, ENXIO); 473 return (0); 474} 475 476static struct g_class g_fox_class = { 477 .name = FOX_CLASS_NAME, 478 .version = G_VERSION, 479 .taste = g_fox_taste, 480 .destroy_geom = g_fox_destroy_geom, 481 .start = g_fox_start, 482 .spoiled = g_fox_orphan, 483 .orphan = g_fox_orphan, 484 .access= g_fox_access, 485}; 486 487DECLARE_GEOM_CLASS(g_fox_class, g_fox); 488MODULE_VERSION(geom_fox, 0); 489