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