1 /*- 2 * Copyright (c) 2004-2005 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 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. 13 * 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 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/kernel.h> 33 #include <sys/module.h> 34 #include <sys/lock.h> 35 #include <sys/mutex.h> 36 #include <sys/bio.h> 37 #include <sys/sysctl.h> 38 #include <sys/malloc.h> 39 #include <vm/uma.h> 40 #include <geom/geom.h> 41 #include <geom/stripe/g_stripe.h> 42 43 FEATURE(geom_stripe, "GEOM striping support"); 44 45 static MALLOC_DEFINE(M_STRIPE, "stripe_data", "GEOM_STRIPE Data"); 46 47 static uma_zone_t g_stripe_zone; 48 49 static int g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force); 50 static int g_stripe_destroy_geom(struct gctl_req *req, struct g_class *mp, 51 struct g_geom *gp); 52 53 static g_taste_t g_stripe_taste; 54 static g_ctl_req_t g_stripe_config; 55 static g_dumpconf_t g_stripe_dumpconf; 56 static g_init_t g_stripe_init; 57 static g_fini_t g_stripe_fini; 58 59 struct g_class g_stripe_class = { 60 .name = G_STRIPE_CLASS_NAME, 61 .version = G_VERSION, 62 .ctlreq = g_stripe_config, 63 .taste = g_stripe_taste, 64 .destroy_geom = g_stripe_destroy_geom, 65 .init = g_stripe_init, 66 .fini = g_stripe_fini 67 }; 68 69 SYSCTL_DECL(_kern_geom); 70 SYSCTL_NODE(_kern_geom, OID_AUTO, stripe, CTLFLAG_RW, 0, "GEOM_STRIPE stuff"); 71 static u_int g_stripe_debug = 0; 72 TUNABLE_INT("kern.geom.stripe.debug", &g_stripe_debug); 73 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, debug, CTLFLAG_RW, &g_stripe_debug, 0, 74 "Debug level"); 75 static int g_stripe_fast = 0; 76 TUNABLE_INT("kern.geom.stripe.fast", &g_stripe_fast); 77 static int 78 g_sysctl_stripe_fast(SYSCTL_HANDLER_ARGS) 79 { 80 int error, fast; 81 82 fast = g_stripe_fast; 83 error = sysctl_handle_int(oidp, &fast, 0, req); 84 if (error == 0 && req->newptr != NULL) 85 g_stripe_fast = fast; 86 return (error); 87 } 88 SYSCTL_PROC(_kern_geom_stripe, OID_AUTO, fast, CTLTYPE_INT | CTLFLAG_RW, 89 NULL, 0, g_sysctl_stripe_fast, "I", "Fast, but memory-consuming, mode"); 90 static u_int g_stripe_maxmem = MAXPHYS * 100; 91 TUNABLE_INT("kern.geom.stripe.maxmem", &g_stripe_maxmem); 92 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, maxmem, CTLFLAG_RD, &g_stripe_maxmem, 93 0, "Maximum memory that can be allocated in \"fast\" mode (in bytes)"); 94 static u_int g_stripe_fast_failed = 0; 95 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, fast_failed, CTLFLAG_RD, 96 &g_stripe_fast_failed, 0, "How many times \"fast\" mode failed"); 97 98 /* 99 * Greatest Common Divisor. 100 */ 101 static u_int 102 gcd(u_int a, u_int b) 103 { 104 u_int c; 105 106 while (b != 0) { 107 c = a; 108 a = b; 109 b = (c % b); 110 } 111 return (a); 112 } 113 114 /* 115 * Least Common Multiple. 116 */ 117 static u_int 118 lcm(u_int a, u_int b) 119 { 120 121 return ((a * b) / gcd(a, b)); 122 } 123 124 static void 125 g_stripe_init(struct g_class *mp __unused) 126 { 127 128 g_stripe_zone = uma_zcreate("g_stripe_zone", MAXPHYS, NULL, NULL, 129 NULL, NULL, 0, 0); 130 g_stripe_maxmem -= g_stripe_maxmem % MAXPHYS; 131 uma_zone_set_max(g_stripe_zone, g_stripe_maxmem / MAXPHYS); 132 } 133 134 static void 135 g_stripe_fini(struct g_class *mp __unused) 136 { 137 138 uma_zdestroy(g_stripe_zone); 139 } 140 141 /* 142 * Return the number of valid disks. 143 */ 144 static u_int 145 g_stripe_nvalid(struct g_stripe_softc *sc) 146 { 147 u_int i, no; 148 149 no = 0; 150 for (i = 0; i < sc->sc_ndisks; i++) { 151 if (sc->sc_disks[i] != NULL) 152 no++; 153 } 154 155 return (no); 156 } 157 158 static void 159 g_stripe_remove_disk(struct g_consumer *cp) 160 { 161 struct g_stripe_softc *sc; 162 u_int no; 163 164 KASSERT(cp != NULL, ("Non-valid disk in %s.", __func__)); 165 sc = (struct g_stripe_softc *)cp->private; 166 KASSERT(sc != NULL, ("NULL sc in %s.", __func__)); 167 no = cp->index; 168 169 G_STRIPE_DEBUG(0, "Disk %s removed from %s.", cp->provider->name, 170 sc->sc_name); 171 172 sc->sc_disks[no] = NULL; 173 if (sc->sc_provider != NULL) { 174 sc->sc_provider->flags |= G_PF_WITHER; 175 g_orphan_provider(sc->sc_provider, ENXIO); 176 sc->sc_provider = NULL; 177 G_STRIPE_DEBUG(0, "Device %s removed.", sc->sc_name); 178 } 179 180 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 181 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 182 g_detach(cp); 183 g_destroy_consumer(cp); 184 } 185 186 static void 187 g_stripe_orphan(struct g_consumer *cp) 188 { 189 struct g_stripe_softc *sc; 190 struct g_geom *gp; 191 192 g_topology_assert(); 193 gp = cp->geom; 194 sc = gp->softc; 195 if (sc == NULL) 196 return; 197 198 g_stripe_remove_disk(cp); 199 /* If there are no valid disks anymore, remove device. */ 200 if (g_stripe_nvalid(sc) == 0) 201 g_stripe_destroy(sc, 1); 202 } 203 204 static int 205 g_stripe_access(struct g_provider *pp, int dr, int dw, int de) 206 { 207 struct g_consumer *cp1, *cp2; 208 struct g_stripe_softc *sc; 209 struct g_geom *gp; 210 int error; 211 212 gp = pp->geom; 213 sc = gp->softc; 214 215 if (sc == NULL) { 216 /* 217 * It looks like geom is being withered. 218 * In that case we allow only negative requests. 219 */ 220 KASSERT(dr <= 0 && dw <= 0 && de <= 0, 221 ("Positive access request (device=%s).", pp->name)); 222 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 && 223 (pp->ace + de) == 0) { 224 G_STRIPE_DEBUG(0, "Device %s definitely destroyed.", 225 gp->name); 226 } 227 return (0); 228 } 229 230 /* On first open, grab an extra "exclusive" bit */ 231 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 232 de++; 233 /* ... and let go of it on last close */ 234 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 && (pp->ace + de) == 0) 235 de--; 236 237 error = ENXIO; 238 LIST_FOREACH(cp1, &gp->consumer, consumer) { 239 error = g_access(cp1, dr, dw, de); 240 if (error == 0) 241 continue; 242 /* 243 * If we fail here, backout all previous changes. 244 */ 245 LIST_FOREACH(cp2, &gp->consumer, consumer) { 246 if (cp1 == cp2) 247 return (error); 248 g_access(cp2, -dr, -dw, -de); 249 } 250 /* NOTREACHED */ 251 } 252 253 return (error); 254 } 255 256 static void 257 g_stripe_copy(struct g_stripe_softc *sc, char *src, char *dst, off_t offset, 258 off_t length, int mode) 259 { 260 u_int stripesize; 261 size_t len; 262 263 stripesize = sc->sc_stripesize; 264 len = (size_t)(stripesize - (offset & (stripesize - 1))); 265 do { 266 bcopy(src, dst, len); 267 if (mode) { 268 dst += len + stripesize * (sc->sc_ndisks - 1); 269 src += len; 270 } else { 271 dst += len; 272 src += len + stripesize * (sc->sc_ndisks - 1); 273 } 274 length -= len; 275 KASSERT(length >= 0, 276 ("Length < 0 (stripesize=%zu, offset=%jd, length=%jd).", 277 (size_t)stripesize, (intmax_t)offset, (intmax_t)length)); 278 if (length > stripesize) 279 len = stripesize; 280 else 281 len = length; 282 } while (length > 0); 283 } 284 285 static void 286 g_stripe_done(struct bio *bp) 287 { 288 struct g_stripe_softc *sc; 289 struct bio *pbp; 290 291 pbp = bp->bio_parent; 292 sc = pbp->bio_to->geom->softc; 293 if (pbp->bio_error == 0) 294 pbp->bio_error = bp->bio_error; 295 pbp->bio_completed += bp->bio_completed; 296 if (bp->bio_cmd == BIO_READ && bp->bio_caller1 != NULL) { 297 g_stripe_copy(sc, bp->bio_data, bp->bio_caller1, bp->bio_offset, 298 bp->bio_length, 1); 299 bp->bio_data = bp->bio_caller1; 300 bp->bio_caller1 = NULL; 301 } 302 g_destroy_bio(bp); 303 pbp->bio_inbed++; 304 if (pbp->bio_children == pbp->bio_inbed) { 305 if (pbp->bio_driver1 != NULL) 306 uma_zfree(g_stripe_zone, pbp->bio_driver1); 307 g_io_deliver(pbp, pbp->bio_error); 308 } 309 } 310 311 static int 312 g_stripe_start_fast(struct bio *bp, u_int no, off_t offset, off_t length) 313 { 314 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 315 u_int nparts = 0, stripesize; 316 struct g_stripe_softc *sc; 317 char *addr, *data = NULL; 318 struct bio *cbp; 319 int error; 320 321 sc = bp->bio_to->geom->softc; 322 323 addr = bp->bio_data; 324 stripesize = sc->sc_stripesize; 325 326 cbp = g_clone_bio(bp); 327 if (cbp == NULL) { 328 error = ENOMEM; 329 goto failure; 330 } 331 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 332 nparts++; 333 /* 334 * Fill in the component buf structure. 335 */ 336 cbp->bio_done = g_stripe_done; 337 cbp->bio_offset = offset; 338 cbp->bio_data = addr; 339 cbp->bio_caller1 = NULL; 340 cbp->bio_length = length; 341 cbp->bio_caller2 = sc->sc_disks[no]; 342 343 /* offset -= offset % stripesize; */ 344 offset -= offset & (stripesize - 1); 345 addr += length; 346 length = bp->bio_length - length; 347 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) { 348 if (no > sc->sc_ndisks - 1) { 349 no = 0; 350 offset += stripesize; 351 } 352 if (nparts >= sc->sc_ndisks) { 353 cbp = TAILQ_NEXT(cbp, bio_queue); 354 if (cbp == NULL) 355 cbp = TAILQ_FIRST(&queue); 356 nparts++; 357 /* 358 * Update bio structure. 359 */ 360 /* 361 * MIN() is in case when 362 * (bp->bio_length % sc->sc_stripesize) != 0. 363 */ 364 cbp->bio_length += MIN(stripesize, length); 365 if (cbp->bio_caller1 == NULL) { 366 cbp->bio_caller1 = cbp->bio_data; 367 cbp->bio_data = NULL; 368 if (data == NULL) { 369 data = uma_zalloc(g_stripe_zone, 370 M_NOWAIT); 371 if (data == NULL) { 372 error = ENOMEM; 373 goto failure; 374 } 375 } 376 } 377 } else { 378 cbp = g_clone_bio(bp); 379 if (cbp == NULL) { 380 error = ENOMEM; 381 goto failure; 382 } 383 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 384 nparts++; 385 /* 386 * Fill in the component buf structure. 387 */ 388 cbp->bio_done = g_stripe_done; 389 cbp->bio_offset = offset; 390 cbp->bio_data = addr; 391 cbp->bio_caller1 = NULL; 392 /* 393 * MIN() is in case when 394 * (bp->bio_length % sc->sc_stripesize) != 0. 395 */ 396 cbp->bio_length = MIN(stripesize, length); 397 cbp->bio_caller2 = sc->sc_disks[no]; 398 } 399 } 400 if (data != NULL) 401 bp->bio_driver1 = data; 402 /* 403 * Fire off all allocated requests! 404 */ 405 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 406 struct g_consumer *cp; 407 408 TAILQ_REMOVE(&queue, cbp, bio_queue); 409 cp = cbp->bio_caller2; 410 cbp->bio_caller2 = NULL; 411 cbp->bio_to = cp->provider; 412 if (cbp->bio_caller1 != NULL) { 413 cbp->bio_data = data; 414 if (bp->bio_cmd == BIO_WRITE) { 415 g_stripe_copy(sc, cbp->bio_caller1, data, 416 cbp->bio_offset, cbp->bio_length, 0); 417 } 418 data += cbp->bio_length; 419 } 420 G_STRIPE_LOGREQ(cbp, "Sending request."); 421 g_io_request(cbp, cp); 422 } 423 return (0); 424 failure: 425 if (data != NULL) 426 uma_zfree(g_stripe_zone, data); 427 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 428 TAILQ_REMOVE(&queue, cbp, bio_queue); 429 if (cbp->bio_caller1 != NULL) { 430 cbp->bio_data = cbp->bio_caller1; 431 cbp->bio_caller1 = NULL; 432 } 433 bp->bio_children--; 434 g_destroy_bio(cbp); 435 } 436 return (error); 437 } 438 439 static int 440 g_stripe_start_economic(struct bio *bp, u_int no, off_t offset, off_t length) 441 { 442 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 443 struct g_stripe_softc *sc; 444 uint32_t stripesize; 445 struct bio *cbp; 446 char *addr; 447 int error; 448 449 sc = bp->bio_to->geom->softc; 450 451 addr = bp->bio_data; 452 stripesize = sc->sc_stripesize; 453 454 cbp = g_clone_bio(bp); 455 if (cbp == NULL) { 456 error = ENOMEM; 457 goto failure; 458 } 459 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 460 /* 461 * Fill in the component buf structure. 462 */ 463 cbp->bio_done = g_std_done; 464 cbp->bio_offset = offset; 465 cbp->bio_data = addr; 466 cbp->bio_length = length; 467 cbp->bio_caller2 = sc->sc_disks[no]; 468 469 /* offset -= offset % stripesize; */ 470 offset -= offset & (stripesize - 1); 471 addr += length; 472 length = bp->bio_length - length; 473 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) { 474 if (no > sc->sc_ndisks - 1) { 475 no = 0; 476 offset += stripesize; 477 } 478 cbp = g_clone_bio(bp); 479 if (cbp == NULL) { 480 error = ENOMEM; 481 goto failure; 482 } 483 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 484 485 /* 486 * Fill in the component buf structure. 487 */ 488 cbp->bio_done = g_std_done; 489 cbp->bio_offset = offset; 490 cbp->bio_data = addr; 491 /* 492 * MIN() is in case when 493 * (bp->bio_length % sc->sc_stripesize) != 0. 494 */ 495 cbp->bio_length = MIN(stripesize, length); 496 497 cbp->bio_caller2 = sc->sc_disks[no]; 498 } 499 /* 500 * Fire off all allocated requests! 501 */ 502 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 503 struct g_consumer *cp; 504 505 TAILQ_REMOVE(&queue, cbp, bio_queue); 506 cp = cbp->bio_caller2; 507 cbp->bio_caller2 = NULL; 508 cbp->bio_to = cp->provider; 509 G_STRIPE_LOGREQ(cbp, "Sending request."); 510 g_io_request(cbp, cp); 511 } 512 return (0); 513 failure: 514 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 515 TAILQ_REMOVE(&queue, cbp, bio_queue); 516 bp->bio_children--; 517 g_destroy_bio(cbp); 518 } 519 return (error); 520 } 521 522 static void 523 g_stripe_flush(struct g_stripe_softc *sc, struct bio *bp) 524 { 525 struct bio_queue_head queue; 526 struct g_consumer *cp; 527 struct bio *cbp; 528 u_int no; 529 530 bioq_init(&queue); 531 for (no = 0; no < sc->sc_ndisks; no++) { 532 cbp = g_clone_bio(bp); 533 if (cbp == NULL) { 534 for (cbp = bioq_first(&queue); cbp != NULL; 535 cbp = bioq_first(&queue)) { 536 bioq_remove(&queue, cbp); 537 g_destroy_bio(cbp); 538 } 539 if (bp->bio_error == 0) 540 bp->bio_error = ENOMEM; 541 g_io_deliver(bp, bp->bio_error); 542 return; 543 } 544 bioq_insert_tail(&queue, cbp); 545 cbp->bio_done = g_std_done; 546 cbp->bio_caller1 = sc->sc_disks[no]; 547 cbp->bio_to = sc->sc_disks[no]->provider; 548 } 549 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 550 bioq_remove(&queue, cbp); 551 G_STRIPE_LOGREQ(cbp, "Sending request."); 552 cp = cbp->bio_caller1; 553 cbp->bio_caller1 = NULL; 554 g_io_request(cbp, cp); 555 } 556 } 557 558 static void 559 g_stripe_start(struct bio *bp) 560 { 561 off_t offset, start, length, nstripe; 562 struct g_stripe_softc *sc; 563 u_int no, stripesize; 564 int error, fast = 0; 565 566 sc = bp->bio_to->geom->softc; 567 /* 568 * If sc == NULL, provider's error should be set and g_stripe_start() 569 * should not be called at all. 570 */ 571 KASSERT(sc != NULL, 572 ("Provider's error should be set (error=%d)(device=%s).", 573 bp->bio_to->error, bp->bio_to->name)); 574 575 G_STRIPE_LOGREQ(bp, "Request received."); 576 577 switch (bp->bio_cmd) { 578 case BIO_READ: 579 case BIO_WRITE: 580 case BIO_DELETE: 581 break; 582 case BIO_FLUSH: 583 g_stripe_flush(sc, bp); 584 return; 585 case BIO_GETATTR: 586 /* To which provider it should be delivered? */ 587 default: 588 g_io_deliver(bp, EOPNOTSUPP); 589 return; 590 } 591 592 stripesize = sc->sc_stripesize; 593 594 /* 595 * Calculations are quite messy, but fast I hope. 596 */ 597 598 /* Stripe number. */ 599 /* nstripe = bp->bio_offset / stripesize; */ 600 nstripe = bp->bio_offset >> (off_t)sc->sc_stripebits; 601 /* Disk number. */ 602 no = nstripe % sc->sc_ndisks; 603 /* Start position in stripe. */ 604 /* start = bp->bio_offset % stripesize; */ 605 start = bp->bio_offset & (stripesize - 1); 606 /* Start position in disk. */ 607 /* offset = (nstripe / sc->sc_ndisks) * stripesize + start; */ 608 offset = ((nstripe / sc->sc_ndisks) << sc->sc_stripebits) + start; 609 /* Length of data to operate. */ 610 length = MIN(bp->bio_length, stripesize - start); 611 612 /* 613 * Do use "fast" mode when: 614 * 1. "Fast" mode is ON. 615 * and 616 * 2. Request size is less than or equal to MAXPHYS, 617 * which should always be true. 618 * and 619 * 3. Request size is bigger than stripesize * ndisks. If it isn't, 620 * there will be no need to send more than one I/O request to 621 * a provider, so there is nothing to optmize. 622 */ 623 if (g_stripe_fast && bp->bio_length <= MAXPHYS && 624 bp->bio_length >= stripesize * sc->sc_ndisks) { 625 fast = 1; 626 } 627 error = 0; 628 if (fast) { 629 error = g_stripe_start_fast(bp, no, offset, length); 630 if (error != 0) 631 g_stripe_fast_failed++; 632 } 633 /* 634 * Do use "economic" when: 635 * 1. "Economic" mode is ON. 636 * or 637 * 2. "Fast" mode failed. It can only fail if there is no memory. 638 */ 639 if (!fast || error != 0) 640 error = g_stripe_start_economic(bp, no, offset, length); 641 if (error != 0) { 642 if (bp->bio_error == 0) 643 bp->bio_error = error; 644 g_io_deliver(bp, bp->bio_error); 645 } 646 } 647 648 static void 649 g_stripe_check_and_run(struct g_stripe_softc *sc) 650 { 651 off_t mediasize, ms; 652 u_int no, sectorsize = 0; 653 654 if (g_stripe_nvalid(sc) != sc->sc_ndisks) 655 return; 656 657 sc->sc_provider = g_new_providerf(sc->sc_geom, "stripe/%s", 658 sc->sc_name); 659 /* 660 * Find the smallest disk. 661 */ 662 mediasize = sc->sc_disks[0]->provider->mediasize; 663 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) 664 mediasize -= sc->sc_disks[0]->provider->sectorsize; 665 mediasize -= mediasize % sc->sc_stripesize; 666 sectorsize = sc->sc_disks[0]->provider->sectorsize; 667 for (no = 1; no < sc->sc_ndisks; no++) { 668 ms = sc->sc_disks[no]->provider->mediasize; 669 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) 670 ms -= sc->sc_disks[no]->provider->sectorsize; 671 ms -= ms % sc->sc_stripesize; 672 if (ms < mediasize) 673 mediasize = ms; 674 sectorsize = lcm(sectorsize, 675 sc->sc_disks[no]->provider->sectorsize); 676 } 677 sc->sc_provider->sectorsize = sectorsize; 678 sc->sc_provider->mediasize = mediasize * sc->sc_ndisks; 679 sc->sc_provider->stripesize = sc->sc_stripesize; 680 sc->sc_provider->stripeoffset = 0; 681 g_error_provider(sc->sc_provider, 0); 682 683 G_STRIPE_DEBUG(0, "Device %s activated.", sc->sc_name); 684 } 685 686 static int 687 g_stripe_read_metadata(struct g_consumer *cp, struct g_stripe_metadata *md) 688 { 689 struct g_provider *pp; 690 u_char *buf; 691 int error; 692 693 g_topology_assert(); 694 695 error = g_access(cp, 1, 0, 0); 696 if (error != 0) 697 return (error); 698 pp = cp->provider; 699 g_topology_unlock(); 700 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 701 &error); 702 g_topology_lock(); 703 g_access(cp, -1, 0, 0); 704 if (buf == NULL) 705 return (error); 706 707 /* Decode metadata. */ 708 stripe_metadata_decode(buf, md); 709 g_free(buf); 710 711 return (0); 712 } 713 714 /* 715 * Add disk to given device. 716 */ 717 static int 718 g_stripe_add_disk(struct g_stripe_softc *sc, struct g_provider *pp, u_int no) 719 { 720 struct g_consumer *cp, *fcp; 721 struct g_geom *gp; 722 int error; 723 724 /* Metadata corrupted? */ 725 if (no >= sc->sc_ndisks) 726 return (EINVAL); 727 728 /* Check if disk is not already attached. */ 729 if (sc->sc_disks[no] != NULL) 730 return (EEXIST); 731 732 gp = sc->sc_geom; 733 fcp = LIST_FIRST(&gp->consumer); 734 735 cp = g_new_consumer(gp); 736 error = g_attach(cp, pp); 737 if (error != 0) { 738 g_destroy_consumer(cp); 739 return (error); 740 } 741 742 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) { 743 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace); 744 if (error != 0) { 745 g_detach(cp); 746 g_destroy_consumer(cp); 747 return (error); 748 } 749 } 750 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) { 751 struct g_stripe_metadata md; 752 753 /* Reread metadata. */ 754 error = g_stripe_read_metadata(cp, &md); 755 if (error != 0) 756 goto fail; 757 758 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0 || 759 strcmp(md.md_name, sc->sc_name) != 0 || 760 md.md_id != sc->sc_id) { 761 G_STRIPE_DEBUG(0, "Metadata on %s changed.", pp->name); 762 goto fail; 763 } 764 } 765 766 cp->private = sc; 767 cp->index = no; 768 sc->sc_disks[no] = cp; 769 770 G_STRIPE_DEBUG(0, "Disk %s attached to %s.", pp->name, sc->sc_name); 771 772 g_stripe_check_and_run(sc); 773 774 return (0); 775 fail: 776 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) 777 g_access(cp, -fcp->acr, -fcp->acw, -fcp->ace); 778 g_detach(cp); 779 g_destroy_consumer(cp); 780 return (error); 781 } 782 783 static struct g_geom * 784 g_stripe_create(struct g_class *mp, const struct g_stripe_metadata *md, 785 u_int type) 786 { 787 struct g_stripe_softc *sc; 788 struct g_geom *gp; 789 u_int no; 790 791 G_STRIPE_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 792 md->md_id); 793 794 /* Two disks is minimum. */ 795 if (md->md_all < 2) { 796 G_STRIPE_DEBUG(0, "Too few disks defined for %s.", md->md_name); 797 return (NULL); 798 } 799 #if 0 800 /* Stripe size have to be grater than or equal to sector size. */ 801 if (md->md_stripesize < sectorsize) { 802 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.", md->md_name); 803 return (NULL); 804 } 805 #endif 806 /* Stripe size have to be power of 2. */ 807 if (!powerof2(md->md_stripesize)) { 808 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.", md->md_name); 809 return (NULL); 810 } 811 812 /* Check for duplicate unit */ 813 LIST_FOREACH(gp, &mp->geom, geom) { 814 sc = gp->softc; 815 if (sc != NULL && strcmp(sc->sc_name, md->md_name) == 0) { 816 G_STRIPE_DEBUG(0, "Device %s already configured.", 817 sc->sc_name); 818 return (NULL); 819 } 820 } 821 gp = g_new_geomf(mp, "%s", md->md_name); 822 gp->softc = NULL; /* for a moment */ 823 824 sc = malloc(sizeof(*sc), M_STRIPE, M_WAITOK | M_ZERO); 825 gp->start = g_stripe_start; 826 gp->spoiled = g_stripe_orphan; 827 gp->orphan = g_stripe_orphan; 828 gp->access = g_stripe_access; 829 gp->dumpconf = g_stripe_dumpconf; 830 831 sc->sc_id = md->md_id; 832 sc->sc_stripesize = md->md_stripesize; 833 sc->sc_stripebits = bitcount32(sc->sc_stripesize - 1); 834 sc->sc_ndisks = md->md_all; 835 sc->sc_disks = malloc(sizeof(struct g_consumer *) * sc->sc_ndisks, 836 M_STRIPE, M_WAITOK | M_ZERO); 837 for (no = 0; no < sc->sc_ndisks; no++) 838 sc->sc_disks[no] = NULL; 839 sc->sc_type = type; 840 841 gp->softc = sc; 842 sc->sc_geom = gp; 843 sc->sc_provider = NULL; 844 845 G_STRIPE_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 846 847 return (gp); 848 } 849 850 static int 851 g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force) 852 { 853 struct g_provider *pp; 854 struct g_geom *gp; 855 u_int no; 856 857 g_topology_assert(); 858 859 if (sc == NULL) 860 return (ENXIO); 861 862 pp = sc->sc_provider; 863 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 864 if (force) { 865 G_STRIPE_DEBUG(0, "Device %s is still open, so it " 866 "can't be definitely removed.", pp->name); 867 } else { 868 G_STRIPE_DEBUG(1, 869 "Device %s is still open (r%dw%de%d).", pp->name, 870 pp->acr, pp->acw, pp->ace); 871 return (EBUSY); 872 } 873 } 874 875 for (no = 0; no < sc->sc_ndisks; no++) { 876 if (sc->sc_disks[no] != NULL) 877 g_stripe_remove_disk(sc->sc_disks[no]); 878 } 879 880 gp = sc->sc_geom; 881 gp->softc = NULL; 882 KASSERT(sc->sc_provider == NULL, ("Provider still exists? (device=%s)", 883 gp->name)); 884 free(sc->sc_disks, M_STRIPE); 885 free(sc, M_STRIPE); 886 887 pp = LIST_FIRST(&gp->provider); 888 if (pp == NULL || (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)) 889 G_STRIPE_DEBUG(0, "Device %s destroyed.", gp->name); 890 891 g_wither_geom(gp, ENXIO); 892 893 return (0); 894 } 895 896 static int 897 g_stripe_destroy_geom(struct gctl_req *req __unused, 898 struct g_class *mp __unused, struct g_geom *gp) 899 { 900 struct g_stripe_softc *sc; 901 902 sc = gp->softc; 903 return (g_stripe_destroy(sc, 0)); 904 } 905 906 static struct g_geom * 907 g_stripe_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 908 { 909 struct g_stripe_metadata md; 910 struct g_stripe_softc *sc; 911 struct g_consumer *cp; 912 struct g_geom *gp; 913 int error; 914 915 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 916 g_topology_assert(); 917 918 /* Skip providers that are already open for writing. */ 919 if (pp->acw > 0) 920 return (NULL); 921 922 G_STRIPE_DEBUG(3, "Tasting %s.", pp->name); 923 924 gp = g_new_geomf(mp, "stripe:taste"); 925 gp->start = g_stripe_start; 926 gp->access = g_stripe_access; 927 gp->orphan = g_stripe_orphan; 928 cp = g_new_consumer(gp); 929 g_attach(cp, pp); 930 error = g_stripe_read_metadata(cp, &md); 931 g_detach(cp); 932 g_destroy_consumer(cp); 933 g_destroy_geom(gp); 934 if (error != 0) 935 return (NULL); 936 gp = NULL; 937 938 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0) 939 return (NULL); 940 if (md.md_version > G_STRIPE_VERSION) { 941 printf("geom_stripe.ko module is too old to handle %s.\n", 942 pp->name); 943 return (NULL); 944 } 945 /* 946 * Backward compatibility: 947 */ 948 /* There was no md_provider field in earlier versions of metadata. */ 949 if (md.md_version < 2) 950 bzero(md.md_provider, sizeof(md.md_provider)); 951 /* There was no md_provsize field in earlier versions of metadata. */ 952 if (md.md_version < 3) 953 md.md_provsize = pp->mediasize; 954 955 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 956 return (NULL); 957 if (md.md_provsize != pp->mediasize) 958 return (NULL); 959 960 /* 961 * Let's check if device already exists. 962 */ 963 sc = NULL; 964 LIST_FOREACH(gp, &mp->geom, geom) { 965 sc = gp->softc; 966 if (sc == NULL) 967 continue; 968 if (sc->sc_type != G_STRIPE_TYPE_AUTOMATIC) 969 continue; 970 if (strcmp(md.md_name, sc->sc_name) != 0) 971 continue; 972 if (md.md_id != sc->sc_id) 973 continue; 974 break; 975 } 976 if (gp != NULL) { 977 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 978 error = g_stripe_add_disk(sc, pp, md.md_no); 979 if (error != 0) { 980 G_STRIPE_DEBUG(0, 981 "Cannot add disk %s to %s (error=%d).", pp->name, 982 gp->name, error); 983 return (NULL); 984 } 985 } else { 986 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_AUTOMATIC); 987 if (gp == NULL) { 988 G_STRIPE_DEBUG(0, "Cannot create device %s.", 989 md.md_name); 990 return (NULL); 991 } 992 sc = gp->softc; 993 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 994 error = g_stripe_add_disk(sc, pp, md.md_no); 995 if (error != 0) { 996 G_STRIPE_DEBUG(0, 997 "Cannot add disk %s to %s (error=%d).", pp->name, 998 gp->name, error); 999 g_stripe_destroy(sc, 1); 1000 return (NULL); 1001 } 1002 } 1003 1004 return (gp); 1005 } 1006 1007 static void 1008 g_stripe_ctl_create(struct gctl_req *req, struct g_class *mp) 1009 { 1010 u_int attached, no; 1011 struct g_stripe_metadata md; 1012 struct g_provider *pp; 1013 struct g_stripe_softc *sc; 1014 struct g_geom *gp; 1015 struct sbuf *sb; 1016 intmax_t *stripesize; 1017 const char *name; 1018 char param[16]; 1019 int *nargs; 1020 1021 g_topology_assert(); 1022 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); 1023 if (nargs == NULL) { 1024 gctl_error(req, "No '%s' argument.", "nargs"); 1025 return; 1026 } 1027 if (*nargs <= 2) { 1028 gctl_error(req, "Too few arguments."); 1029 return; 1030 } 1031 1032 strlcpy(md.md_magic, G_STRIPE_MAGIC, sizeof(md.md_magic)); 1033 md.md_version = G_STRIPE_VERSION; 1034 name = gctl_get_asciiparam(req, "arg0"); 1035 if (name == NULL) { 1036 gctl_error(req, "No 'arg%u' argument.", 0); 1037 return; 1038 } 1039 strlcpy(md.md_name, name, sizeof(md.md_name)); 1040 md.md_id = arc4random(); 1041 md.md_no = 0; 1042 md.md_all = *nargs - 1; 1043 stripesize = gctl_get_paraml(req, "stripesize", sizeof(*stripesize)); 1044 if (stripesize == NULL) { 1045 gctl_error(req, "No '%s' argument.", "stripesize"); 1046 return; 1047 } 1048 md.md_stripesize = *stripesize; 1049 bzero(md.md_provider, sizeof(md.md_provider)); 1050 /* This field is not important here. */ 1051 md.md_provsize = 0; 1052 1053 /* Check all providers are valid */ 1054 for (no = 1; no < *nargs; no++) { 1055 snprintf(param, sizeof(param), "arg%u", no); 1056 name = gctl_get_asciiparam(req, param); 1057 if (name == NULL) { 1058 gctl_error(req, "No 'arg%u' argument.", no); 1059 return; 1060 } 1061 if (strncmp(name, "/dev/", strlen("/dev/")) == 0) 1062 name += strlen("/dev/"); 1063 pp = g_provider_by_name(name); 1064 if (pp == NULL) { 1065 G_STRIPE_DEBUG(1, "Disk %s is invalid.", name); 1066 gctl_error(req, "Disk %s is invalid.", name); 1067 return; 1068 } 1069 } 1070 1071 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_MANUAL); 1072 if (gp == NULL) { 1073 gctl_error(req, "Can't configure %s.", md.md_name); 1074 return; 1075 } 1076 1077 sc = gp->softc; 1078 sb = sbuf_new_auto(); 1079 sbuf_printf(sb, "Can't attach disk(s) to %s:", gp->name); 1080 for (attached = 0, no = 1; no < *nargs; no++) { 1081 snprintf(param, sizeof(param), "arg%u", no); 1082 name = gctl_get_asciiparam(req, param); 1083 if (name == NULL) { 1084 gctl_error(req, "No 'arg%u' argument.", no); 1085 continue; 1086 } 1087 if (strncmp(name, "/dev/", strlen("/dev/")) == 0) 1088 name += strlen("/dev/"); 1089 pp = g_provider_by_name(name); 1090 KASSERT(pp != NULL, ("Provider %s disappear?!", name)); 1091 if (g_stripe_add_disk(sc, pp, no - 1) != 0) { 1092 G_STRIPE_DEBUG(1, "Disk %u (%s) not attached to %s.", 1093 no, pp->name, gp->name); 1094 sbuf_printf(sb, " %s", pp->name); 1095 continue; 1096 } 1097 attached++; 1098 } 1099 sbuf_finish(sb); 1100 if (md.md_all != attached) { 1101 g_stripe_destroy(gp->softc, 1); 1102 gctl_error(req, "%s", sbuf_data(sb)); 1103 } 1104 sbuf_delete(sb); 1105 } 1106 1107 static struct g_stripe_softc * 1108 g_stripe_find_device(struct g_class *mp, const char *name) 1109 { 1110 struct g_stripe_softc *sc; 1111 struct g_geom *gp; 1112 1113 LIST_FOREACH(gp, &mp->geom, geom) { 1114 sc = gp->softc; 1115 if (sc == NULL) 1116 continue; 1117 if (strcmp(sc->sc_name, name) == 0) 1118 return (sc); 1119 } 1120 return (NULL); 1121 } 1122 1123 static void 1124 g_stripe_ctl_destroy(struct gctl_req *req, struct g_class *mp) 1125 { 1126 struct g_stripe_softc *sc; 1127 int *force, *nargs, error; 1128 const char *name; 1129 char param[16]; 1130 u_int i; 1131 1132 g_topology_assert(); 1133 1134 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); 1135 if (nargs == NULL) { 1136 gctl_error(req, "No '%s' argument.", "nargs"); 1137 return; 1138 } 1139 if (*nargs <= 0) { 1140 gctl_error(req, "Missing device(s)."); 1141 return; 1142 } 1143 force = gctl_get_paraml(req, "force", sizeof(*force)); 1144 if (force == NULL) { 1145 gctl_error(req, "No '%s' argument.", "force"); 1146 return; 1147 } 1148 1149 for (i = 0; i < (u_int)*nargs; i++) { 1150 snprintf(param, sizeof(param), "arg%u", i); 1151 name = gctl_get_asciiparam(req, param); 1152 if (name == NULL) { 1153 gctl_error(req, "No 'arg%u' argument.", i); 1154 return; 1155 } 1156 sc = g_stripe_find_device(mp, name); 1157 if (sc == NULL) { 1158 gctl_error(req, "No such device: %s.", name); 1159 return; 1160 } 1161 error = g_stripe_destroy(sc, *force); 1162 if (error != 0) { 1163 gctl_error(req, "Cannot destroy device %s (error=%d).", 1164 sc->sc_name, error); 1165 return; 1166 } 1167 } 1168 } 1169 1170 static void 1171 g_stripe_config(struct gctl_req *req, struct g_class *mp, const char *verb) 1172 { 1173 uint32_t *version; 1174 1175 g_topology_assert(); 1176 1177 version = gctl_get_paraml(req, "version", sizeof(*version)); 1178 if (version == NULL) { 1179 gctl_error(req, "No '%s' argument.", "version"); 1180 return; 1181 } 1182 if (*version != G_STRIPE_VERSION) { 1183 gctl_error(req, "Userland and kernel parts are out of sync."); 1184 return; 1185 } 1186 1187 if (strcmp(verb, "create") == 0) { 1188 g_stripe_ctl_create(req, mp); 1189 return; 1190 } else if (strcmp(verb, "destroy") == 0 || 1191 strcmp(verb, "stop") == 0) { 1192 g_stripe_ctl_destroy(req, mp); 1193 return; 1194 } 1195 1196 gctl_error(req, "Unknown verb."); 1197 } 1198 1199 static void 1200 g_stripe_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1201 struct g_consumer *cp, struct g_provider *pp) 1202 { 1203 struct g_stripe_softc *sc; 1204 1205 sc = gp->softc; 1206 if (sc == NULL) 1207 return; 1208 if (pp != NULL) { 1209 /* Nothing here. */ 1210 } else if (cp != NULL) { 1211 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent, 1212 (u_int)cp->index); 1213 } else { 1214 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 1215 sbuf_printf(sb, "%s<Stripesize>%u</Stripesize>\n", indent, 1216 (u_int)sc->sc_stripesize); 1217 sbuf_printf(sb, "%s<Type>", indent); 1218 switch (sc->sc_type) { 1219 case G_STRIPE_TYPE_AUTOMATIC: 1220 sbuf_printf(sb, "AUTOMATIC"); 1221 break; 1222 case G_STRIPE_TYPE_MANUAL: 1223 sbuf_printf(sb, "MANUAL"); 1224 break; 1225 default: 1226 sbuf_printf(sb, "UNKNOWN"); 1227 break; 1228 } 1229 sbuf_printf(sb, "</Type>\n"); 1230 sbuf_printf(sb, "%s<Status>Total=%u, Online=%u</Status>\n", 1231 indent, sc->sc_ndisks, g_stripe_nvalid(sc)); 1232 sbuf_printf(sb, "%s<State>", indent); 1233 if (sc->sc_provider != NULL && sc->sc_provider->error == 0) 1234 sbuf_printf(sb, "UP"); 1235 else 1236 sbuf_printf(sb, "DOWN"); 1237 sbuf_printf(sb, "</State>\n"); 1238 } 1239 } 1240 1241 DECLARE_GEOM_CLASS(g_stripe_class, g_stripe); 1242