1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2004-2005 Pawel Jakub Dawidek <pjd@FreeBSD.org> 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 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/kernel.h> 35 #include <sys/module.h> 36 #include <sys/lock.h> 37 #include <sys/mutex.h> 38 #include <sys/bio.h> 39 #include <sys/sbuf.h> 40 #include <sys/sysctl.h> 41 #include <sys/malloc.h> 42 #include <vm/uma.h> 43 #include <geom/geom.h> 44 #include <geom/geom_dbg.h> 45 #include <geom/stripe/g_stripe.h> 46 47 FEATURE(geom_stripe, "GEOM striping support"); 48 49 static MALLOC_DEFINE(M_STRIPE, "stripe_data", "GEOM_STRIPE Data"); 50 51 static uma_zone_t g_stripe_zone; 52 53 static int g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force); 54 static int g_stripe_destroy_geom(struct gctl_req *req, struct g_class *mp, 55 struct g_geom *gp); 56 57 static g_taste_t g_stripe_taste; 58 static g_ctl_req_t g_stripe_config; 59 static g_dumpconf_t g_stripe_dumpconf; 60 static g_init_t g_stripe_init; 61 static g_fini_t g_stripe_fini; 62 63 struct g_class g_stripe_class = { 64 .name = G_STRIPE_CLASS_NAME, 65 .version = G_VERSION, 66 .ctlreq = g_stripe_config, 67 .taste = g_stripe_taste, 68 .destroy_geom = g_stripe_destroy_geom, 69 .init = g_stripe_init, 70 .fini = g_stripe_fini 71 }; 72 73 SYSCTL_DECL(_kern_geom); 74 static SYSCTL_NODE(_kern_geom, OID_AUTO, stripe, CTLFLAG_RW, 0, 75 "GEOM_STRIPE stuff"); 76 static u_int g_stripe_debug = 0; 77 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, debug, CTLFLAG_RWTUN, &g_stripe_debug, 0, 78 "Debug level"); 79 static int g_stripe_fast = 0; 80 static int 81 g_sysctl_stripe_fast(SYSCTL_HANDLER_ARGS) 82 { 83 int error, fast; 84 85 fast = g_stripe_fast; 86 error = sysctl_handle_int(oidp, &fast, 0, req); 87 if (error == 0 && req->newptr != NULL) 88 g_stripe_fast = fast; 89 return (error); 90 } 91 SYSCTL_PROC(_kern_geom_stripe, OID_AUTO, fast, CTLTYPE_INT | CTLFLAG_RWTUN, 92 NULL, 0, g_sysctl_stripe_fast, "I", "Fast, but memory-consuming, mode"); 93 static u_int g_stripe_maxmem = MAXPHYS * 100; 94 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, maxmem, CTLFLAG_RDTUN, &g_stripe_maxmem, 95 0, "Maximum memory that can be allocated in \"fast\" mode (in bytes)"); 96 static u_int g_stripe_fast_failed = 0; 97 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, fast_failed, CTLFLAG_RD, 98 &g_stripe_fast_failed, 0, "How many times \"fast\" mode failed"); 99 100 /* 101 * Greatest Common Divisor. 102 */ 103 static u_int 104 gcd(u_int a, u_int b) 105 { 106 u_int c; 107 108 while (b != 0) { 109 c = a; 110 a = b; 111 b = (c % b); 112 } 113 return (a); 114 } 115 116 /* 117 * Least Common Multiple. 118 */ 119 static u_int 120 lcm(u_int a, u_int b) 121 { 122 123 return ((a * b) / gcd(a, b)); 124 } 125 126 static void 127 g_stripe_init(struct g_class *mp __unused) 128 { 129 130 g_stripe_zone = uma_zcreate("g_stripe_zone", MAXPHYS, NULL, NULL, 131 NULL, NULL, 0, 0); 132 g_stripe_maxmem -= g_stripe_maxmem % MAXPHYS; 133 uma_zone_set_max(g_stripe_zone, g_stripe_maxmem / MAXPHYS); 134 } 135 136 static void 137 g_stripe_fini(struct g_class *mp __unused) 138 { 139 140 uma_zdestroy(g_stripe_zone); 141 } 142 143 /* 144 * Return the number of valid disks. 145 */ 146 static u_int 147 g_stripe_nvalid(struct g_stripe_softc *sc) 148 { 149 u_int i, no; 150 151 no = 0; 152 for (i = 0; i < sc->sc_ndisks; i++) { 153 if (sc->sc_disks[i] != NULL) 154 no++; 155 } 156 157 return (no); 158 } 159 160 static void 161 g_stripe_remove_disk(struct g_consumer *cp) 162 { 163 struct g_stripe_softc *sc; 164 165 g_topology_assert(); 166 KASSERT(cp != NULL, ("Non-valid disk in %s.", __func__)); 167 sc = (struct g_stripe_softc *)cp->geom->softc; 168 KASSERT(sc != NULL, ("NULL sc in %s.", __func__)); 169 170 if (cp->private == NULL) { 171 G_STRIPE_DEBUG(0, "Disk %s removed from %s.", 172 cp->provider->name, sc->sc_name); 173 cp->private = (void *)(uintptr_t)-1; 174 } 175 176 if (sc->sc_provider != NULL) { 177 G_STRIPE_DEBUG(0, "Device %s deactivated.", 178 sc->sc_provider->name); 179 g_wither_provider(sc->sc_provider, ENXIO); 180 sc->sc_provider = NULL; 181 } 182 183 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 184 return; 185 sc->sc_disks[cp->index] = NULL; 186 cp->index = 0; 187 g_detach(cp); 188 g_destroy_consumer(cp); 189 /* If there are no valid disks anymore, remove device. */ 190 if (LIST_EMPTY(&sc->sc_geom->consumer)) 191 g_stripe_destroy(sc, 1); 192 } 193 194 static void 195 g_stripe_orphan(struct g_consumer *cp) 196 { 197 struct g_stripe_softc *sc; 198 struct g_geom *gp; 199 200 g_topology_assert(); 201 gp = cp->geom; 202 sc = gp->softc; 203 if (sc == NULL) 204 return; 205 206 g_stripe_remove_disk(cp); 207 } 208 209 static int 210 g_stripe_access(struct g_provider *pp, int dr, int dw, int de) 211 { 212 struct g_consumer *cp1, *cp2, *tmp; 213 struct g_stripe_softc *sc; 214 struct g_geom *gp; 215 int error; 216 217 g_topology_assert(); 218 gp = pp->geom; 219 sc = gp->softc; 220 KASSERT(sc != NULL, ("NULL sc in %s.", __func__)); 221 222 /* On first open, grab an extra "exclusive" bit */ 223 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 224 de++; 225 /* ... and let go of it on last close */ 226 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 && (pp->ace + de) == 0) 227 de--; 228 229 LIST_FOREACH_SAFE(cp1, &gp->consumer, consumer, tmp) { 230 error = g_access(cp1, dr, dw, de); 231 if (error != 0) 232 goto fail; 233 if (cp1->acr == 0 && cp1->acw == 0 && cp1->ace == 0 && 234 cp1->private != NULL) { 235 g_stripe_remove_disk(cp1); /* May destroy geom. */ 236 } 237 } 238 return (0); 239 240 fail: 241 LIST_FOREACH(cp2, &gp->consumer, consumer) { 242 if (cp1 == cp2) 243 break; 244 g_access(cp2, -dr, -dw, -de); 245 } 246 return (error); 247 } 248 249 static void 250 g_stripe_copy(struct g_stripe_softc *sc, char *src, char *dst, off_t offset, 251 off_t length, int mode) 252 { 253 off_t stripesize; 254 size_t len; 255 256 stripesize = sc->sc_stripesize; 257 len = (size_t)(stripesize - (offset & (stripesize - 1))); 258 do { 259 bcopy(src, dst, len); 260 if (mode) { 261 dst += len + stripesize * (sc->sc_ndisks - 1); 262 src += len; 263 } else { 264 dst += len; 265 src += len + stripesize * (sc->sc_ndisks - 1); 266 } 267 length -= len; 268 KASSERT(length >= 0, 269 ("Length < 0 (stripesize=%ju, offset=%ju, length=%jd).", 270 (uintmax_t)stripesize, (uintmax_t)offset, (intmax_t)length)); 271 if (length > stripesize) 272 len = stripesize; 273 else 274 len = length; 275 } while (length > 0); 276 } 277 278 static void 279 g_stripe_done(struct bio *bp) 280 { 281 struct g_stripe_softc *sc; 282 struct bio *pbp; 283 284 pbp = bp->bio_parent; 285 sc = pbp->bio_to->geom->softc; 286 if (bp->bio_cmd == BIO_READ && bp->bio_caller1 != NULL) { 287 g_stripe_copy(sc, bp->bio_data, bp->bio_caller1, bp->bio_offset, 288 bp->bio_length, 1); 289 bp->bio_data = bp->bio_caller1; 290 bp->bio_caller1 = NULL; 291 } 292 mtx_lock(&sc->sc_lock); 293 if (pbp->bio_error == 0) 294 pbp->bio_error = bp->bio_error; 295 pbp->bio_completed += bp->bio_completed; 296 pbp->bio_inbed++; 297 if (pbp->bio_children == pbp->bio_inbed) { 298 mtx_unlock(&sc->sc_lock); 299 if (pbp->bio_driver1 != NULL) 300 uma_zfree(g_stripe_zone, pbp->bio_driver1); 301 if (bp->bio_cmd == BIO_SPEEDUP) 302 pbp->bio_completed = pbp->bio_length; 303 g_io_deliver(pbp, pbp->bio_error); 304 } else 305 mtx_unlock(&sc->sc_lock); 306 g_destroy_bio(bp); 307 } 308 309 static int 310 g_stripe_start_fast(struct bio *bp, u_int no, off_t offset, off_t length) 311 { 312 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 313 struct g_stripe_softc *sc; 314 char *addr, *data = NULL; 315 struct bio *cbp; 316 off_t stripesize; 317 u_int nparts = 0; 318 int error; 319 320 sc = bp->bio_to->geom->softc; 321 322 addr = bp->bio_data; 323 stripesize = sc->sc_stripesize; 324 325 cbp = g_clone_bio(bp); 326 if (cbp == NULL) { 327 error = ENOMEM; 328 goto failure; 329 } 330 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 331 nparts++; 332 /* 333 * Fill in the component buf structure. 334 */ 335 cbp->bio_done = g_stripe_done; 336 cbp->bio_offset = offset; 337 cbp->bio_data = addr; 338 cbp->bio_caller1 = NULL; 339 cbp->bio_length = length; 340 cbp->bio_caller2 = sc->sc_disks[no]; 341 342 /* offset -= offset % stripesize; */ 343 offset -= offset & (stripesize - 1); 344 addr += length; 345 length = bp->bio_length - length; 346 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) { 347 if (no > sc->sc_ndisks - 1) { 348 no = 0; 349 offset += stripesize; 350 } 351 if (nparts >= sc->sc_ndisks) { 352 cbp = TAILQ_NEXT(cbp, bio_queue); 353 if (cbp == NULL) 354 cbp = TAILQ_FIRST(&queue); 355 nparts++; 356 /* 357 * Update bio structure. 358 */ 359 /* 360 * MIN() is in case when 361 * (bp->bio_length % sc->sc_stripesize) != 0. 362 */ 363 cbp->bio_length += MIN(stripesize, length); 364 if (cbp->bio_caller1 == NULL) { 365 cbp->bio_caller1 = cbp->bio_data; 366 cbp->bio_data = NULL; 367 if (data == NULL) { 368 data = uma_zalloc(g_stripe_zone, 369 M_NOWAIT); 370 if (data == NULL) { 371 error = ENOMEM; 372 goto failure; 373 } 374 } 375 } 376 } else { 377 cbp = g_clone_bio(bp); 378 if (cbp == NULL) { 379 error = ENOMEM; 380 goto failure; 381 } 382 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 383 nparts++; 384 /* 385 * Fill in the component buf structure. 386 */ 387 cbp->bio_done = g_stripe_done; 388 cbp->bio_offset = offset; 389 cbp->bio_data = addr; 390 cbp->bio_caller1 = NULL; 391 /* 392 * MIN() is in case when 393 * (bp->bio_length % sc->sc_stripesize) != 0. 394 */ 395 cbp->bio_length = MIN(stripesize, length); 396 cbp->bio_caller2 = sc->sc_disks[no]; 397 } 398 } 399 if (data != NULL) 400 bp->bio_driver1 = data; 401 /* 402 * Fire off all allocated requests! 403 */ 404 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 405 struct g_consumer *cp; 406 407 TAILQ_REMOVE(&queue, cbp, bio_queue); 408 cp = cbp->bio_caller2; 409 cbp->bio_caller2 = NULL; 410 cbp->bio_to = cp->provider; 411 if (cbp->bio_caller1 != NULL) { 412 cbp->bio_data = data; 413 if (bp->bio_cmd == BIO_WRITE) { 414 g_stripe_copy(sc, cbp->bio_caller1, data, 415 cbp->bio_offset, cbp->bio_length, 0); 416 } 417 data += cbp->bio_length; 418 } 419 G_STRIPE_LOGREQ(cbp, "Sending request."); 420 g_io_request(cbp, cp); 421 } 422 return (0); 423 failure: 424 if (data != NULL) 425 uma_zfree(g_stripe_zone, data); 426 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 427 TAILQ_REMOVE(&queue, cbp, bio_queue); 428 if (cbp->bio_caller1 != NULL) { 429 cbp->bio_data = cbp->bio_caller1; 430 cbp->bio_caller1 = NULL; 431 } 432 bp->bio_children--; 433 g_destroy_bio(cbp); 434 } 435 return (error); 436 } 437 438 static int 439 g_stripe_start_economic(struct bio *bp, u_int no, off_t offset, off_t length) 440 { 441 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); 442 struct g_stripe_softc *sc; 443 off_t stripesize; 444 struct bio *cbp; 445 char *addr; 446 int error; 447 448 sc = bp->bio_to->geom->softc; 449 450 stripesize = sc->sc_stripesize; 451 452 cbp = g_clone_bio(bp); 453 if (cbp == NULL) { 454 error = ENOMEM; 455 goto failure; 456 } 457 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 458 /* 459 * Fill in the component buf structure. 460 */ 461 if (bp->bio_length == length) 462 cbp->bio_done = g_std_done; /* Optimized lockless case. */ 463 else 464 cbp->bio_done = g_stripe_done; 465 cbp->bio_offset = offset; 466 cbp->bio_length = length; 467 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 468 bp->bio_ma_n = round_page(bp->bio_ma_offset + 469 bp->bio_length) / PAGE_SIZE; 470 addr = NULL; 471 } else 472 addr = bp->bio_data; 473 cbp->bio_caller2 = sc->sc_disks[no]; 474 475 /* offset -= offset % stripesize; */ 476 offset -= offset & (stripesize - 1); 477 if (bp->bio_cmd != BIO_DELETE) 478 addr += length; 479 length = bp->bio_length - length; 480 for (no++; length > 0; no++, length -= stripesize) { 481 if (no > sc->sc_ndisks - 1) { 482 no = 0; 483 offset += stripesize; 484 } 485 cbp = g_clone_bio(bp); 486 if (cbp == NULL) { 487 error = ENOMEM; 488 goto failure; 489 } 490 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 491 492 /* 493 * Fill in the component buf structure. 494 */ 495 cbp->bio_done = g_stripe_done; 496 cbp->bio_offset = offset; 497 /* 498 * MIN() is in case when 499 * (bp->bio_length % sc->sc_stripesize) != 0. 500 */ 501 cbp->bio_length = MIN(stripesize, length); 502 if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 503 cbp->bio_ma_offset += (uintptr_t)addr; 504 cbp->bio_ma += cbp->bio_ma_offset / PAGE_SIZE; 505 cbp->bio_ma_offset %= PAGE_SIZE; 506 cbp->bio_ma_n = round_page(cbp->bio_ma_offset + 507 cbp->bio_length) / PAGE_SIZE; 508 } else 509 cbp->bio_data = addr; 510 511 cbp->bio_caller2 = sc->sc_disks[no]; 512 513 if (bp->bio_cmd != BIO_DELETE) 514 addr += stripesize; 515 } 516 /* 517 * Fire off all allocated requests! 518 */ 519 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 520 struct g_consumer *cp; 521 522 TAILQ_REMOVE(&queue, cbp, bio_queue); 523 cp = cbp->bio_caller2; 524 cbp->bio_caller2 = NULL; 525 cbp->bio_to = cp->provider; 526 G_STRIPE_LOGREQ(cbp, "Sending request."); 527 g_io_request(cbp, cp); 528 } 529 return (0); 530 failure: 531 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 532 TAILQ_REMOVE(&queue, cbp, bio_queue); 533 bp->bio_children--; 534 g_destroy_bio(cbp); 535 } 536 return (error); 537 } 538 539 static void 540 g_stripe_pushdown(struct g_stripe_softc *sc, struct bio *bp) 541 { 542 struct bio_queue_head queue; 543 struct g_consumer *cp; 544 struct bio *cbp; 545 u_int no; 546 547 bioq_init(&queue); 548 for (no = 0; no < sc->sc_ndisks; no++) { 549 cbp = g_clone_bio(bp); 550 if (cbp == NULL) { 551 for (cbp = bioq_first(&queue); cbp != NULL; 552 cbp = bioq_first(&queue)) { 553 bioq_remove(&queue, cbp); 554 g_destroy_bio(cbp); 555 } 556 if (bp->bio_error == 0) 557 bp->bio_error = ENOMEM; 558 g_io_deliver(bp, bp->bio_error); 559 return; 560 } 561 bioq_insert_tail(&queue, cbp); 562 cbp->bio_done = g_stripe_done; 563 cbp->bio_caller2 = sc->sc_disks[no]; 564 cbp->bio_to = sc->sc_disks[no]->provider; 565 } 566 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 567 bioq_remove(&queue, cbp); 568 G_STRIPE_LOGREQ(cbp, "Sending request."); 569 cp = cbp->bio_caller2; 570 cbp->bio_caller2 = NULL; 571 g_io_request(cbp, cp); 572 } 573 } 574 575 static void 576 g_stripe_start(struct bio *bp) 577 { 578 off_t offset, start, length, nstripe, stripesize; 579 struct g_stripe_softc *sc; 580 u_int no; 581 int error, fast = 0; 582 583 sc = bp->bio_to->geom->softc; 584 /* 585 * If sc == NULL, provider's error should be set and g_stripe_start() 586 * should not be called at all. 587 */ 588 KASSERT(sc != NULL, 589 ("Provider's error should be set (error=%d)(device=%s).", 590 bp->bio_to->error, bp->bio_to->name)); 591 592 G_STRIPE_LOGREQ(bp, "Request received."); 593 594 switch (bp->bio_cmd) { 595 case BIO_READ: 596 case BIO_WRITE: 597 case BIO_DELETE: 598 break; 599 case BIO_SPEEDUP: 600 case BIO_FLUSH: 601 g_stripe_pushdown(sc, bp); 602 return; 603 case BIO_GETATTR: 604 /* To which provider it should be delivered? */ 605 default: 606 g_io_deliver(bp, EOPNOTSUPP); 607 return; 608 } 609 610 stripesize = sc->sc_stripesize; 611 612 /* 613 * Calculations are quite messy, but fast I hope. 614 */ 615 616 /* Stripe number. */ 617 /* nstripe = bp->bio_offset / stripesize; */ 618 nstripe = bp->bio_offset >> (off_t)sc->sc_stripebits; 619 /* Disk number. */ 620 no = nstripe % sc->sc_ndisks; 621 /* Start position in stripe. */ 622 /* start = bp->bio_offset % stripesize; */ 623 start = bp->bio_offset & (stripesize - 1); 624 /* Start position in disk. */ 625 /* offset = (nstripe / sc->sc_ndisks) * stripesize + start; */ 626 offset = ((nstripe / sc->sc_ndisks) << sc->sc_stripebits) + start; 627 /* Length of data to operate. */ 628 length = MIN(bp->bio_length, stripesize - start); 629 630 /* 631 * Do use "fast" mode when: 632 * 1. "Fast" mode is ON. 633 * and 634 * 2. Request size is less than or equal to MAXPHYS, 635 * which should always be true. 636 * and 637 * 3. Request size is bigger than stripesize * ndisks. If it isn't, 638 * there will be no need to send more than one I/O request to 639 * a provider, so there is nothing to optmize. 640 * and 641 * 4. Request is not unmapped. 642 * and 643 * 5. It is not a BIO_DELETE. 644 */ 645 if (g_stripe_fast && bp->bio_length <= MAXPHYS && 646 bp->bio_length >= stripesize * sc->sc_ndisks && 647 (bp->bio_flags & BIO_UNMAPPED) == 0 && 648 bp->bio_cmd != BIO_DELETE) { 649 fast = 1; 650 } 651 error = 0; 652 if (fast) { 653 error = g_stripe_start_fast(bp, no, offset, length); 654 if (error != 0) 655 g_stripe_fast_failed++; 656 } 657 /* 658 * Do use "economic" when: 659 * 1. "Economic" mode is ON. 660 * or 661 * 2. "Fast" mode failed. It can only fail if there is no memory. 662 */ 663 if (!fast || error != 0) 664 error = g_stripe_start_economic(bp, no, offset, length); 665 if (error != 0) { 666 if (bp->bio_error == 0) 667 bp->bio_error = error; 668 g_io_deliver(bp, bp->bio_error); 669 } 670 } 671 672 static void 673 g_stripe_check_and_run(struct g_stripe_softc *sc) 674 { 675 struct g_provider *dp; 676 off_t mediasize, ms; 677 u_int no, sectorsize = 0; 678 679 g_topology_assert(); 680 if (g_stripe_nvalid(sc) != sc->sc_ndisks) 681 return; 682 683 sc->sc_provider = g_new_providerf(sc->sc_geom, "stripe/%s", 684 sc->sc_name); 685 sc->sc_provider->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE; 686 if (g_stripe_fast == 0) 687 sc->sc_provider->flags |= G_PF_ACCEPT_UNMAPPED; 688 /* 689 * Find the smallest disk. 690 */ 691 mediasize = sc->sc_disks[0]->provider->mediasize; 692 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) 693 mediasize -= sc->sc_disks[0]->provider->sectorsize; 694 mediasize -= mediasize % sc->sc_stripesize; 695 sectorsize = sc->sc_disks[0]->provider->sectorsize; 696 for (no = 1; no < sc->sc_ndisks; no++) { 697 dp = sc->sc_disks[no]->provider; 698 ms = dp->mediasize; 699 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) 700 ms -= dp->sectorsize; 701 ms -= ms % sc->sc_stripesize; 702 if (ms < mediasize) 703 mediasize = ms; 704 sectorsize = lcm(sectorsize, dp->sectorsize); 705 706 /* A provider underneath us doesn't support unmapped */ 707 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 708 G_STRIPE_DEBUG(1, "Cancelling unmapped " 709 "because of %s.", dp->name); 710 sc->sc_provider->flags &= ~G_PF_ACCEPT_UNMAPPED; 711 } 712 } 713 sc->sc_provider->sectorsize = sectorsize; 714 sc->sc_provider->mediasize = mediasize * sc->sc_ndisks; 715 sc->sc_provider->stripesize = sc->sc_stripesize; 716 sc->sc_provider->stripeoffset = 0; 717 g_error_provider(sc->sc_provider, 0); 718 719 G_STRIPE_DEBUG(0, "Device %s activated.", sc->sc_provider->name); 720 } 721 722 static int 723 g_stripe_read_metadata(struct g_consumer *cp, struct g_stripe_metadata *md) 724 { 725 struct g_provider *pp; 726 u_char *buf; 727 int error; 728 729 g_topology_assert(); 730 731 error = g_access(cp, 1, 0, 0); 732 if (error != 0) 733 return (error); 734 pp = cp->provider; 735 g_topology_unlock(); 736 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 737 &error); 738 g_topology_lock(); 739 g_access(cp, -1, 0, 0); 740 if (buf == NULL) 741 return (error); 742 743 /* Decode metadata. */ 744 stripe_metadata_decode(buf, md); 745 g_free(buf); 746 747 return (0); 748 } 749 750 /* 751 * Add disk to given device. 752 */ 753 static int 754 g_stripe_add_disk(struct g_stripe_softc *sc, struct g_provider *pp, u_int no) 755 { 756 struct g_consumer *cp, *fcp; 757 struct g_geom *gp; 758 int error; 759 760 g_topology_assert(); 761 /* Metadata corrupted? */ 762 if (no >= sc->sc_ndisks) 763 return (EINVAL); 764 765 /* Check if disk is not already attached. */ 766 if (sc->sc_disks[no] != NULL) 767 return (EEXIST); 768 769 gp = sc->sc_geom; 770 fcp = LIST_FIRST(&gp->consumer); 771 772 cp = g_new_consumer(gp); 773 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 774 cp->private = NULL; 775 cp->index = no; 776 error = g_attach(cp, pp); 777 if (error != 0) { 778 g_destroy_consumer(cp); 779 return (error); 780 } 781 782 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) { 783 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace); 784 if (error != 0) { 785 g_detach(cp); 786 g_destroy_consumer(cp); 787 return (error); 788 } 789 } 790 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) { 791 struct g_stripe_metadata md; 792 793 /* Reread metadata. */ 794 error = g_stripe_read_metadata(cp, &md); 795 if (error != 0) 796 goto fail; 797 798 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0 || 799 strcmp(md.md_name, sc->sc_name) != 0 || 800 md.md_id != sc->sc_id) { 801 G_STRIPE_DEBUG(0, "Metadata on %s changed.", pp->name); 802 goto fail; 803 } 804 } 805 806 sc->sc_disks[no] = cp; 807 G_STRIPE_DEBUG(0, "Disk %s attached to %s.", pp->name, sc->sc_name); 808 g_stripe_check_and_run(sc); 809 810 return (0); 811 fail: 812 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) 813 g_access(cp, -fcp->acr, -fcp->acw, -fcp->ace); 814 g_detach(cp); 815 g_destroy_consumer(cp); 816 return (error); 817 } 818 819 static struct g_geom * 820 g_stripe_create(struct g_class *mp, const struct g_stripe_metadata *md, 821 u_int type) 822 { 823 struct g_stripe_softc *sc; 824 struct g_geom *gp; 825 u_int no; 826 827 g_topology_assert(); 828 G_STRIPE_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 829 md->md_id); 830 831 /* Two disks is minimum. */ 832 if (md->md_all < 2) { 833 G_STRIPE_DEBUG(0, "Too few disks defined for %s.", md->md_name); 834 return (NULL); 835 } 836 #if 0 837 /* Stripe size have to be grater than or equal to sector size. */ 838 if (md->md_stripesize < sectorsize) { 839 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.", md->md_name); 840 return (NULL); 841 } 842 #endif 843 /* Stripe size have to be power of 2. */ 844 if (!powerof2(md->md_stripesize)) { 845 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.", md->md_name); 846 return (NULL); 847 } 848 849 /* Check for duplicate unit */ 850 LIST_FOREACH(gp, &mp->geom, geom) { 851 sc = gp->softc; 852 if (sc != NULL && strcmp(sc->sc_name, md->md_name) == 0) { 853 G_STRIPE_DEBUG(0, "Device %s already configured.", 854 sc->sc_name); 855 return (NULL); 856 } 857 } 858 gp = g_new_geomf(mp, "%s", md->md_name); 859 sc = malloc(sizeof(*sc), M_STRIPE, M_WAITOK | M_ZERO); 860 gp->start = g_stripe_start; 861 gp->spoiled = g_stripe_orphan; 862 gp->orphan = g_stripe_orphan; 863 gp->access = g_stripe_access; 864 gp->dumpconf = g_stripe_dumpconf; 865 866 sc->sc_id = md->md_id; 867 sc->sc_stripesize = md->md_stripesize; 868 sc->sc_stripebits = bitcount32(sc->sc_stripesize - 1); 869 sc->sc_ndisks = md->md_all; 870 sc->sc_disks = malloc(sizeof(struct g_consumer *) * sc->sc_ndisks, 871 M_STRIPE, M_WAITOK | M_ZERO); 872 for (no = 0; no < sc->sc_ndisks; no++) 873 sc->sc_disks[no] = NULL; 874 sc->sc_type = type; 875 mtx_init(&sc->sc_lock, "gstripe lock", NULL, MTX_DEF); 876 877 gp->softc = sc; 878 sc->sc_geom = gp; 879 sc->sc_provider = NULL; 880 881 G_STRIPE_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 882 883 return (gp); 884 } 885 886 static int 887 g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force) 888 { 889 struct g_provider *pp; 890 struct g_consumer *cp, *cp1; 891 struct g_geom *gp; 892 893 g_topology_assert(); 894 895 if (sc == NULL) 896 return (ENXIO); 897 898 pp = sc->sc_provider; 899 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 900 if (force) { 901 G_STRIPE_DEBUG(0, "Device %s is still open, so it " 902 "can't be definitely removed.", pp->name); 903 } else { 904 G_STRIPE_DEBUG(1, 905 "Device %s is still open (r%dw%de%d).", pp->name, 906 pp->acr, pp->acw, pp->ace); 907 return (EBUSY); 908 } 909 } 910 911 gp = sc->sc_geom; 912 LIST_FOREACH_SAFE(cp, &gp->consumer, consumer, cp1) { 913 g_stripe_remove_disk(cp); 914 if (cp1 == NULL) 915 return (0); /* Recursion happened. */ 916 } 917 if (!LIST_EMPTY(&gp->consumer)) 918 return (EINPROGRESS); 919 920 gp->softc = NULL; 921 KASSERT(sc->sc_provider == NULL, ("Provider still exists? (device=%s)", 922 gp->name)); 923 free(sc->sc_disks, M_STRIPE); 924 mtx_destroy(&sc->sc_lock); 925 free(sc, M_STRIPE); 926 G_STRIPE_DEBUG(0, "Device %s destroyed.", gp->name); 927 g_wither_geom(gp, ENXIO); 928 return (0); 929 } 930 931 static int 932 g_stripe_destroy_geom(struct gctl_req *req __unused, 933 struct g_class *mp __unused, struct g_geom *gp) 934 { 935 struct g_stripe_softc *sc; 936 937 sc = gp->softc; 938 return (g_stripe_destroy(sc, 0)); 939 } 940 941 static struct g_geom * 942 g_stripe_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 943 { 944 struct g_stripe_metadata md; 945 struct g_stripe_softc *sc; 946 struct g_consumer *cp; 947 struct g_geom *gp; 948 int error; 949 950 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 951 g_topology_assert(); 952 953 /* Skip providers that are already open for writing. */ 954 if (pp->acw > 0) 955 return (NULL); 956 957 G_STRIPE_DEBUG(3, "Tasting %s.", pp->name); 958 959 gp = g_new_geomf(mp, "stripe:taste"); 960 gp->start = g_stripe_start; 961 gp->access = g_stripe_access; 962 gp->orphan = g_stripe_orphan; 963 cp = g_new_consumer(gp); 964 g_attach(cp, pp); 965 error = g_stripe_read_metadata(cp, &md); 966 g_detach(cp); 967 g_destroy_consumer(cp); 968 g_destroy_geom(gp); 969 if (error != 0) 970 return (NULL); 971 gp = NULL; 972 973 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0) 974 return (NULL); 975 if (md.md_version > G_STRIPE_VERSION) { 976 printf("geom_stripe.ko module is too old to handle %s.\n", 977 pp->name); 978 return (NULL); 979 } 980 /* 981 * Backward compatibility: 982 */ 983 /* There was no md_provider field in earlier versions of metadata. */ 984 if (md.md_version < 2) 985 bzero(md.md_provider, sizeof(md.md_provider)); 986 /* There was no md_provsize field in earlier versions of metadata. */ 987 if (md.md_version < 3) 988 md.md_provsize = pp->mediasize; 989 990 if (md.md_provider[0] != '\0' && 991 !g_compare_names(md.md_provider, pp->name)) 992 return (NULL); 993 if (md.md_provsize != pp->mediasize) 994 return (NULL); 995 996 /* 997 * Let's check if device already exists. 998 */ 999 sc = NULL; 1000 LIST_FOREACH(gp, &mp->geom, geom) { 1001 sc = gp->softc; 1002 if (sc == NULL) 1003 continue; 1004 if (sc->sc_type != G_STRIPE_TYPE_AUTOMATIC) 1005 continue; 1006 if (strcmp(md.md_name, sc->sc_name) != 0) 1007 continue; 1008 if (md.md_id != sc->sc_id) 1009 continue; 1010 break; 1011 } 1012 if (gp != NULL) { 1013 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 1014 error = g_stripe_add_disk(sc, pp, md.md_no); 1015 if (error != 0) { 1016 G_STRIPE_DEBUG(0, 1017 "Cannot add disk %s to %s (error=%d).", pp->name, 1018 gp->name, error); 1019 return (NULL); 1020 } 1021 } else { 1022 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_AUTOMATIC); 1023 if (gp == NULL) { 1024 G_STRIPE_DEBUG(0, "Cannot create device %s.", 1025 md.md_name); 1026 return (NULL); 1027 } 1028 sc = gp->softc; 1029 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 1030 error = g_stripe_add_disk(sc, pp, md.md_no); 1031 if (error != 0) { 1032 G_STRIPE_DEBUG(0, 1033 "Cannot add disk %s to %s (error=%d).", pp->name, 1034 gp->name, error); 1035 g_stripe_destroy(sc, 1); 1036 return (NULL); 1037 } 1038 } 1039 1040 return (gp); 1041 } 1042 1043 static void 1044 g_stripe_ctl_create(struct gctl_req *req, struct g_class *mp) 1045 { 1046 u_int attached, no; 1047 struct g_stripe_metadata md; 1048 struct g_provider *pp; 1049 struct g_stripe_softc *sc; 1050 struct g_geom *gp; 1051 struct sbuf *sb; 1052 off_t *stripesize; 1053 const char *name; 1054 char param[16]; 1055 int *nargs; 1056 1057 g_topology_assert(); 1058 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); 1059 if (nargs == NULL) { 1060 gctl_error(req, "No '%s' argument.", "nargs"); 1061 return; 1062 } 1063 if (*nargs <= 2) { 1064 gctl_error(req, "Too few arguments."); 1065 return; 1066 } 1067 1068 strlcpy(md.md_magic, G_STRIPE_MAGIC, sizeof(md.md_magic)); 1069 md.md_version = G_STRIPE_VERSION; 1070 name = gctl_get_asciiparam(req, "arg0"); 1071 if (name == NULL) { 1072 gctl_error(req, "No 'arg%u' argument.", 0); 1073 return; 1074 } 1075 strlcpy(md.md_name, name, sizeof(md.md_name)); 1076 md.md_id = arc4random(); 1077 md.md_no = 0; 1078 md.md_all = *nargs - 1; 1079 stripesize = gctl_get_paraml(req, "stripesize", sizeof(*stripesize)); 1080 if (stripesize == NULL) { 1081 gctl_error(req, "No '%s' argument.", "stripesize"); 1082 return; 1083 } 1084 md.md_stripesize = (uint32_t)*stripesize; 1085 bzero(md.md_provider, sizeof(md.md_provider)); 1086 /* This field is not important here. */ 1087 md.md_provsize = 0; 1088 1089 /* Check all providers are valid */ 1090 for (no = 1; no < *nargs; no++) { 1091 snprintf(param, sizeof(param), "arg%u", no); 1092 name = gctl_get_asciiparam(req, param); 1093 if (name == NULL) { 1094 gctl_error(req, "No 'arg%u' argument.", no); 1095 return; 1096 } 1097 if (strncmp(name, "/dev/", strlen("/dev/")) == 0) 1098 name += strlen("/dev/"); 1099 pp = g_provider_by_name(name); 1100 if (pp == NULL) { 1101 G_STRIPE_DEBUG(1, "Disk %s is invalid.", name); 1102 gctl_error(req, "Disk %s is invalid.", name); 1103 return; 1104 } 1105 } 1106 1107 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_MANUAL); 1108 if (gp == NULL) { 1109 gctl_error(req, "Can't configure %s.", md.md_name); 1110 return; 1111 } 1112 1113 sc = gp->softc; 1114 sb = sbuf_new_auto(); 1115 sbuf_printf(sb, "Can't attach disk(s) to %s:", gp->name); 1116 for (attached = 0, no = 1; no < *nargs; no++) { 1117 snprintf(param, sizeof(param), "arg%u", no); 1118 name = gctl_get_asciiparam(req, param); 1119 if (name == NULL) { 1120 gctl_error(req, "No 'arg%u' argument.", no); 1121 continue; 1122 } 1123 if (strncmp(name, "/dev/", strlen("/dev/")) == 0) 1124 name += strlen("/dev/"); 1125 pp = g_provider_by_name(name); 1126 KASSERT(pp != NULL, ("Provider %s disappear?!", name)); 1127 if (g_stripe_add_disk(sc, pp, no - 1) != 0) { 1128 G_STRIPE_DEBUG(1, "Disk %u (%s) not attached to %s.", 1129 no, pp->name, gp->name); 1130 sbuf_printf(sb, " %s", pp->name); 1131 continue; 1132 } 1133 attached++; 1134 } 1135 sbuf_finish(sb); 1136 if (md.md_all != attached) { 1137 g_stripe_destroy(gp->softc, 1); 1138 gctl_error(req, "%s", sbuf_data(sb)); 1139 } 1140 sbuf_delete(sb); 1141 } 1142 1143 static struct g_stripe_softc * 1144 g_stripe_find_device(struct g_class *mp, const char *name) 1145 { 1146 struct g_stripe_softc *sc; 1147 struct g_geom *gp; 1148 1149 LIST_FOREACH(gp, &mp->geom, geom) { 1150 sc = gp->softc; 1151 if (sc == NULL) 1152 continue; 1153 if (strcmp(sc->sc_name, name) == 0) 1154 return (sc); 1155 } 1156 return (NULL); 1157 } 1158 1159 static void 1160 g_stripe_ctl_destroy(struct gctl_req *req, struct g_class *mp) 1161 { 1162 struct g_stripe_softc *sc; 1163 int *force, *nargs, error; 1164 const char *name; 1165 char param[16]; 1166 u_int i; 1167 1168 g_topology_assert(); 1169 1170 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); 1171 if (nargs == NULL) { 1172 gctl_error(req, "No '%s' argument.", "nargs"); 1173 return; 1174 } 1175 if (*nargs <= 0) { 1176 gctl_error(req, "Missing device(s)."); 1177 return; 1178 } 1179 force = gctl_get_paraml(req, "force", sizeof(*force)); 1180 if (force == NULL) { 1181 gctl_error(req, "No '%s' argument.", "force"); 1182 return; 1183 } 1184 1185 for (i = 0; i < (u_int)*nargs; i++) { 1186 snprintf(param, sizeof(param), "arg%u", i); 1187 name = gctl_get_asciiparam(req, param); 1188 if (name == NULL) { 1189 gctl_error(req, "No 'arg%u' argument.", i); 1190 return; 1191 } 1192 sc = g_stripe_find_device(mp, name); 1193 if (sc == NULL) { 1194 gctl_error(req, "No such device: %s.", name); 1195 return; 1196 } 1197 error = g_stripe_destroy(sc, *force); 1198 if (error != 0) { 1199 gctl_error(req, "Cannot destroy device %s (error=%d).", 1200 sc->sc_name, error); 1201 return; 1202 } 1203 } 1204 } 1205 1206 static void 1207 g_stripe_config(struct gctl_req *req, struct g_class *mp, const char *verb) 1208 { 1209 uint32_t *version; 1210 1211 g_topology_assert(); 1212 1213 version = gctl_get_paraml(req, "version", sizeof(*version)); 1214 if (version == NULL) { 1215 gctl_error(req, "No '%s' argument.", "version"); 1216 return; 1217 } 1218 if (*version != G_STRIPE_VERSION) { 1219 gctl_error(req, "Userland and kernel parts are out of sync."); 1220 return; 1221 } 1222 1223 if (strcmp(verb, "create") == 0) { 1224 g_stripe_ctl_create(req, mp); 1225 return; 1226 } else if (strcmp(verb, "destroy") == 0 || 1227 strcmp(verb, "stop") == 0) { 1228 g_stripe_ctl_destroy(req, mp); 1229 return; 1230 } 1231 1232 gctl_error(req, "Unknown verb."); 1233 } 1234 1235 static void 1236 g_stripe_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 1237 struct g_consumer *cp, struct g_provider *pp) 1238 { 1239 struct g_stripe_softc *sc; 1240 1241 sc = gp->softc; 1242 if (sc == NULL) 1243 return; 1244 if (pp != NULL) { 1245 /* Nothing here. */ 1246 } else if (cp != NULL) { 1247 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent, 1248 (u_int)cp->index); 1249 } else { 1250 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 1251 sbuf_printf(sb, "%s<Stripesize>%ju</Stripesize>\n", indent, 1252 (uintmax_t)sc->sc_stripesize); 1253 sbuf_printf(sb, "%s<Type>", indent); 1254 switch (sc->sc_type) { 1255 case G_STRIPE_TYPE_AUTOMATIC: 1256 sbuf_cat(sb, "AUTOMATIC"); 1257 break; 1258 case G_STRIPE_TYPE_MANUAL: 1259 sbuf_cat(sb, "MANUAL"); 1260 break; 1261 default: 1262 sbuf_cat(sb, "UNKNOWN"); 1263 break; 1264 } 1265 sbuf_cat(sb, "</Type>\n"); 1266 sbuf_printf(sb, "%s<Status>Total=%u, Online=%u</Status>\n", 1267 indent, sc->sc_ndisks, g_stripe_nvalid(sc)); 1268 sbuf_printf(sb, "%s<State>", indent); 1269 if (sc->sc_provider != NULL && sc->sc_provider->error == 0) 1270 sbuf_cat(sb, "UP"); 1271 else 1272 sbuf_cat(sb, "DOWN"); 1273 sbuf_cat(sb, "</State>\n"); 1274 } 1275 } 1276 1277 DECLARE_GEOM_CLASS(g_stripe_class, g_stripe); 1278 MODULE_VERSION(geom_stripe, 0); 1279