1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/zfs_context.h> 29 #include <sys/spa.h> 30 #include <sys/refcount.h> 31 #include <sys/vdev_disk.h> 32 #include <sys/vdev_impl.h> 33 #include <sys/fs/zfs.h> 34 #include <sys/zio.h> 35 #include <sys/sunldi.h> 36 #include <sys/fm/fs/zfs.h> 37 38 /* 39 * Virtual device vector for disks. 40 */ 41 42 extern ldi_ident_t zfs_li; 43 44 typedef struct vdev_disk_buf { 45 buf_t vdb_buf; 46 zio_t *vdb_io; 47 } vdev_disk_buf_t; 48 49 static int 50 vdev_disk_open_common(vdev_t *vd) 51 { 52 vdev_disk_t *dvd; 53 dev_t dev; 54 int error; 55 56 /* 57 * We must have a pathname, and it must be absolute. 58 */ 59 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') { 60 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; 61 return (EINVAL); 62 } 63 64 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP); 65 66 /* 67 * When opening a disk device, we want to preserve the user's original 68 * intent. We always want to open the device by the path the user gave 69 * us, even if it is one of multiple paths to the save device. But we 70 * also want to be able to survive disks being removed/recabled. 71 * Therefore the sequence of opening devices is: 72 * 73 * 1. Try opening the device by path. For legacy pools without the 74 * 'whole_disk' property, attempt to fix the path by appending 's0'. 75 * 76 * 2. If the devid of the device matches the stored value, return 77 * success. 78 * 79 * 3. Otherwise, the device may have moved. Try opening the device 80 * by the devid instead. 81 * 82 * If the vdev is part of the root pool, we avoid opening it by path. 83 * We do this because there is no /dev path available early in boot, 84 * and if we try to open the device by path at a later point, we can 85 * deadlock when devfsadm attempts to open the underlying backing store 86 * file. 87 */ 88 if (vd->vdev_devid != NULL) { 89 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid, 90 &dvd->vd_minor) != 0) { 91 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; 92 return (EINVAL); 93 } 94 } 95 96 error = EINVAL; /* presume failure */ 97 98 if (vd->vdev_path != NULL && !spa_is_root(vd->vdev_spa)) { 99 ddi_devid_t devid; 100 101 if (vd->vdev_wholedisk == -1ULL) { 102 size_t len = strlen(vd->vdev_path) + 3; 103 char *buf = kmem_alloc(len, KM_SLEEP); 104 ldi_handle_t lh; 105 106 (void) snprintf(buf, len, "%ss0", vd->vdev_path); 107 108 if (ldi_open_by_name(buf, spa_mode, kcred, 109 &lh, zfs_li) == 0) { 110 spa_strfree(vd->vdev_path); 111 vd->vdev_path = buf; 112 vd->vdev_wholedisk = 1ULL; 113 (void) ldi_close(lh, spa_mode, kcred); 114 } else { 115 kmem_free(buf, len); 116 } 117 } 118 119 error = ldi_open_by_name(vd->vdev_path, spa_mode, kcred, 120 &dvd->vd_lh, zfs_li); 121 122 /* 123 * Compare the devid to the stored value. 124 */ 125 if (error == 0 && vd->vdev_devid != NULL && 126 ldi_get_devid(dvd->vd_lh, &devid) == 0) { 127 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) { 128 error = EINVAL; 129 (void) ldi_close(dvd->vd_lh, spa_mode, kcred); 130 dvd->vd_lh = NULL; 131 } 132 ddi_devid_free(devid); 133 } 134 135 /* 136 * If we succeeded in opening the device, but 'vdev_wholedisk' 137 * is not yet set, then this must be a slice. 138 */ 139 if (error == 0 && vd->vdev_wholedisk == -1ULL) 140 vd->vdev_wholedisk = 0; 141 } 142 143 /* 144 * If we were unable to open by path, or the devid check fails, open by 145 * devid instead. 146 */ 147 if (error != 0 && vd->vdev_devid != NULL) 148 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor, 149 spa_mode, kcred, &dvd->vd_lh, zfs_li); 150 151 /* 152 * If all else fails, then try opening by physical path (if available) 153 * or the logical path (if we failed due to the devid check). While not 154 * as reliable as the devid, this will give us something, and the higher 155 * level vdev validation will prevent us from opening the wrong device. 156 */ 157 if (error) { 158 if (vd->vdev_physpath != NULL && 159 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != ENODEV) 160 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode, 161 kcred, &dvd->vd_lh, zfs_li); 162 163 /* 164 * Note that we don't support the legacy auto-wholedisk support 165 * as above. This hasn't been used in a very long time and we 166 * don't need to propagate its oddities to this edge condition. 167 */ 168 if (error && vd->vdev_path != NULL && 169 !spa_is_root(vd->vdev_spa)) 170 error = ldi_open_by_name(vd->vdev_path, spa_mode, kcred, 171 &dvd->vd_lh, zfs_li); 172 } 173 174 if (error) 175 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; 176 177 return (error); 178 } 179 180 static int 181 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift) 182 { 183 vdev_disk_t *dvd; 184 struct dk_minfo dkm; 185 int error; 186 dev_t dev; 187 int otyp; 188 189 error = vdev_disk_open_common(vd); 190 if (error) 191 return (error); 192 193 dvd = vd->vdev_tsd; 194 /* 195 * Once a device is opened, verify that the physical device path (if 196 * available) is up to date. 197 */ 198 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 && 199 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) { 200 char *physpath, *minorname; 201 202 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 203 minorname = NULL; 204 if (ddi_dev_pathname(dev, otyp, physpath) == 0 && 205 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 && 206 (vd->vdev_physpath == NULL || 207 strcmp(vd->vdev_physpath, physpath) != 0)) { 208 if (vd->vdev_physpath) 209 spa_strfree(vd->vdev_physpath); 210 (void) strlcat(physpath, ":", MAXPATHLEN); 211 (void) strlcat(physpath, minorname, MAXPATHLEN); 212 vd->vdev_physpath = spa_strdup(physpath); 213 } 214 if (minorname) 215 kmem_free(minorname, strlen(minorname) + 1); 216 kmem_free(physpath, MAXPATHLEN); 217 } 218 219 /* 220 * Determine the actual size of the device. 221 */ 222 if (ldi_get_size(dvd->vd_lh, psize) != 0) { 223 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; 224 return (EINVAL); 225 } 226 227 /* 228 * If we own the whole disk, try to enable disk write caching. 229 * We ignore errors because it's OK if we can't do it. 230 */ 231 if (vd->vdev_wholedisk == 1) { 232 int wce = 1; 233 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce, 234 FKIOCTL, kcred, NULL); 235 } 236 237 /* 238 * Determine the device's minimum transfer size. 239 * If the ioctl isn't supported, assume DEV_BSIZE. 240 */ 241 if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO, (intptr_t)&dkm, 242 FKIOCTL, kcred, NULL) != 0) 243 dkm.dki_lbsize = DEV_BSIZE; 244 245 *ashift = highbit(MAX(dkm.dki_lbsize, SPA_MINBLOCKSIZE)) - 1; 246 247 /* 248 * Clear the nowritecache bit, so that on a vdev_reopen() we will 249 * try again. 250 */ 251 vd->vdev_nowritecache = B_FALSE; 252 253 return (0); 254 } 255 256 static void 257 vdev_disk_close(vdev_t *vd) 258 { 259 vdev_disk_t *dvd = vd->vdev_tsd; 260 261 if (dvd == NULL) 262 return; 263 264 if (dvd->vd_minor != NULL) 265 ddi_devid_str_free(dvd->vd_minor); 266 267 if (dvd->vd_devid != NULL) 268 ddi_devid_free(dvd->vd_devid); 269 270 if (dvd->vd_lh != NULL) 271 (void) ldi_close(dvd->vd_lh, spa_mode, kcred); 272 273 kmem_free(dvd, sizeof (vdev_disk_t)); 274 vd->vdev_tsd = NULL; 275 } 276 277 int 278 vdev_disk_physio(ldi_handle_t vd_lh, caddr_t data, size_t size, 279 uint64_t offset, int flags) 280 { 281 buf_t *bp; 282 int error = 0; 283 284 if (vd_lh == NULL) 285 return (EINVAL); 286 287 ASSERT(flags & B_READ || flags & B_WRITE); 288 289 bp = getrbuf(KM_SLEEP); 290 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST; 291 bp->b_bcount = size; 292 bp->b_un.b_addr = (void *)data; 293 bp->b_lblkno = lbtodb(offset); 294 bp->b_bufsize = size; 295 296 error = ldi_strategy(vd_lh, bp); 297 ASSERT(error == 0); 298 if ((error = biowait(bp)) == 0 && bp->b_resid != 0) 299 error = EIO; 300 freerbuf(bp); 301 302 return (error); 303 } 304 305 static int 306 vdev_disk_probe_io(vdev_t *vd, caddr_t data, size_t size, uint64_t offset, 307 int flags) 308 { 309 int error = 0; 310 vdev_disk_t *dvd = vd ? vd->vdev_tsd : NULL; 311 312 if (vd == NULL || dvd == NULL || dvd->vd_lh == NULL) 313 return (EINVAL); 314 315 error = vdev_disk_physio(dvd->vd_lh, data, size, offset, flags); 316 317 if (zio_injection_enabled && error == 0) 318 error = zio_handle_device_injection(vd, EIO); 319 320 return (error); 321 } 322 323 /* 324 * Determine if the underlying device is accessible by reading and writing 325 * to a known location. We must be able to do this during syncing context 326 * and thus we cannot set the vdev state directly. 327 */ 328 static int 329 vdev_disk_probe(vdev_t *vd) 330 { 331 uint64_t offset; 332 vdev_t *nvd; 333 int l, error = 0, retries = 0; 334 char *vl_pad; 335 336 if (vd == NULL) 337 return (EINVAL); 338 339 /* Hijack the current vdev */ 340 nvd = vd; 341 342 /* 343 * Pick a random label to rewrite. 344 */ 345 l = spa_get_random(VDEV_LABELS); 346 ASSERT(l < VDEV_LABELS); 347 348 offset = vdev_label_offset(vd->vdev_psize, l, 349 offsetof(vdev_label_t, vl_pad)); 350 351 vl_pad = kmem_alloc(VDEV_SKIP_SIZE, KM_SLEEP); 352 353 /* 354 * Try to read and write to a special location on the 355 * label. We use the existing vdev initially and only 356 * try to create and reopen it if we encounter a failure. 357 */ 358 while ((error = vdev_disk_probe_io(nvd, vl_pad, VDEV_SKIP_SIZE, 359 offset, B_READ)) != 0 && retries == 0) { 360 361 nvd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 362 if (vd->vdev_path) 363 nvd->vdev_path = spa_strdup(vd->vdev_path); 364 if (vd->vdev_physpath) 365 nvd->vdev_physpath = spa_strdup(vd->vdev_physpath); 366 if (vd->vdev_devid) 367 nvd->vdev_devid = spa_strdup(vd->vdev_devid); 368 nvd->vdev_wholedisk = vd->vdev_wholedisk; 369 nvd->vdev_guid = vd->vdev_guid; 370 nvd->vdev_spa = vd->vdev_spa; 371 retries++; 372 373 error = vdev_disk_open_common(nvd); 374 if (error) 375 break; 376 } 377 378 if (!error) { 379 error = vdev_disk_probe_io(nvd, vl_pad, VDEV_SKIP_SIZE, 380 offset, B_WRITE); 381 } 382 383 /* Clean up if we allocated a new vdev */ 384 if (retries) { 385 vdev_disk_close(nvd); 386 if (nvd->vdev_path) 387 spa_strfree(nvd->vdev_path); 388 if (nvd->vdev_physpath) 389 spa_strfree(nvd->vdev_physpath); 390 if (nvd->vdev_devid) 391 spa_strfree(nvd->vdev_devid); 392 kmem_free(nvd, sizeof (vdev_t)); 393 } 394 kmem_free(vl_pad, VDEV_SKIP_SIZE); 395 396 /* Reset the failing flag */ 397 if (!error) 398 vd->vdev_is_failing = B_FALSE; 399 400 return (error); 401 } 402 403 static void 404 vdev_disk_io_intr(buf_t *bp) 405 { 406 vdev_disk_buf_t *vdb = (vdev_disk_buf_t *)bp; 407 zio_t *zio = vdb->vdb_io; 408 409 /* 410 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO. 411 * Rather than teach the rest of the stack about other error 412 * possibilities (EFAULT, etc), we normalize the error value here. 413 */ 414 zio->io_error = (geterror(bp) != 0 ? EIO : 0); 415 416 if (zio->io_error == 0 && bp->b_resid != 0) 417 zio->io_error = EIO; 418 419 kmem_free(vdb, sizeof (vdev_disk_buf_t)); 420 421 zio_interrupt(zio); 422 } 423 424 static void 425 vdev_disk_ioctl_done(void *zio_arg, int error) 426 { 427 zio_t *zio = zio_arg; 428 429 zio->io_error = error; 430 431 zio_interrupt(zio); 432 } 433 434 static int 435 vdev_disk_io_start(zio_t *zio) 436 { 437 vdev_t *vd = zio->io_vd; 438 vdev_disk_t *dvd = vd->vdev_tsd; 439 vdev_disk_buf_t *vdb; 440 buf_t *bp; 441 int flags, error; 442 443 if (zio->io_type == ZIO_TYPE_IOCTL) { 444 zio_vdev_io_bypass(zio); 445 446 /* XXPOLICY */ 447 if (!vdev_readable(vd)) { 448 zio->io_error = ENXIO; 449 return (ZIO_PIPELINE_CONTINUE); 450 } 451 452 switch (zio->io_cmd) { 453 454 case DKIOCFLUSHWRITECACHE: 455 456 if (zfs_nocacheflush) 457 break; 458 459 if (vd->vdev_nowritecache) { 460 zio->io_error = ENOTSUP; 461 break; 462 } 463 464 zio->io_dk_callback.dkc_callback = vdev_disk_ioctl_done; 465 zio->io_dk_callback.dkc_flag = FLUSH_VOLATILE; 466 zio->io_dk_callback.dkc_cookie = zio; 467 468 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd, 469 (uintptr_t)&zio->io_dk_callback, 470 FKIOCTL, kcred, NULL); 471 472 if (error == 0) { 473 /* 474 * The ioctl will be done asychronously, 475 * and will call vdev_disk_ioctl_done() 476 * upon completion. 477 */ 478 return (ZIO_PIPELINE_STOP); 479 } 480 481 if (error == ENOTSUP || error == ENOTTY) { 482 /* 483 * If we get ENOTSUP or ENOTTY, we know that 484 * no future attempts will ever succeed. 485 * In this case we set a persistent bit so 486 * that we don't bother with the ioctl in the 487 * future. 488 */ 489 vd->vdev_nowritecache = B_TRUE; 490 } 491 zio->io_error = error; 492 493 break; 494 495 default: 496 zio->io_error = ENOTSUP; 497 } 498 499 return (ZIO_PIPELINE_CONTINUE); 500 } 501 502 if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0) 503 return (ZIO_PIPELINE_STOP); 504 505 if ((zio = vdev_queue_io(zio)) == NULL) 506 return (ZIO_PIPELINE_STOP); 507 508 if (zio->io_type == ZIO_TYPE_WRITE) 509 error = vdev_writeable(vd) ? vdev_error_inject(vd, zio) : ENXIO; 510 else 511 error = vdev_readable(vd) ? vdev_error_inject(vd, zio) : ENXIO; 512 error = (vd->vdev_remove_wanted || vd->vdev_is_failing) ? ENXIO : error; 513 514 if (error) { 515 zio->io_error = error; 516 zio_interrupt(zio); 517 return (ZIO_PIPELINE_STOP); 518 } 519 520 flags = (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE); 521 flags |= B_BUSY | B_NOCACHE; 522 if (zio->io_flags & ZIO_FLAG_FAILFAST) 523 flags |= B_FAILFAST; 524 525 vdb = kmem_alloc(sizeof (vdev_disk_buf_t), KM_SLEEP); 526 527 vdb->vdb_io = zio; 528 bp = &vdb->vdb_buf; 529 530 bioinit(bp); 531 bp->b_flags = flags; 532 bp->b_bcount = zio->io_size; 533 bp->b_un.b_addr = zio->io_data; 534 bp->b_lblkno = lbtodb(zio->io_offset); 535 bp->b_bufsize = zio->io_size; 536 bp->b_iodone = (int (*)())vdev_disk_io_intr; 537 538 error = ldi_strategy(dvd->vd_lh, bp); 539 /* ldi_strategy() will return non-zero only on programming errors */ 540 ASSERT(error == 0); 541 542 return (ZIO_PIPELINE_STOP); 543 } 544 545 static int 546 vdev_disk_io_done(zio_t *zio) 547 { 548 vdev_queue_io_done(zio); 549 550 if (zio->io_type == ZIO_TYPE_WRITE) 551 vdev_cache_write(zio); 552 553 if (zio_injection_enabled && zio->io_error == 0) 554 zio->io_error = zio_handle_device_injection(zio->io_vd, EIO); 555 556 /* 557 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if 558 * the device has been removed. If this is the case, then we trigger an 559 * asynchronous removal of the device. Otherwise, probe the device and 560 * make sure it's still accessible. 561 */ 562 if (zio->io_error == EIO) { 563 vdev_t *vd = zio->io_vd; 564 vdev_disk_t *dvd = vd->vdev_tsd; 565 int state; 566 567 state = DKIO_NONE; 568 if (dvd && ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state, 569 FKIOCTL, kcred, NULL) == 0 && 570 state != DKIO_INSERTED) { 571 vd->vdev_remove_wanted = B_TRUE; 572 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE); 573 } else if (vdev_probe(vd) != 0) { 574 ASSERT(vd->vdev_ops->vdev_op_leaf); 575 if (!vd->vdev_is_failing) { 576 vd->vdev_is_failing = B_TRUE; 577 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 578 vd->vdev_spa, vd, zio, 0, 0); 579 } 580 } 581 } 582 583 if (zio_injection_enabled && zio->io_error == 0) 584 zio->io_error = zio_handle_label_injection(zio, EIO); 585 586 return (ZIO_PIPELINE_CONTINUE); 587 } 588 589 vdev_ops_t vdev_disk_ops = { 590 vdev_disk_open, 591 vdev_disk_close, 592 vdev_disk_probe, 593 vdev_default_asize, 594 vdev_disk_io_start, 595 vdev_disk_io_done, 596 NULL, 597 VDEV_TYPE_DISK, /* name of this vdev type */ 598 B_TRUE /* leaf vdev */ 599 }; 600 601 /* 602 * Given the root disk device devid or pathname, read the label from 603 * the device, and construct a configuration nvlist. 604 */ 605 nvlist_t * 606 vdev_disk_read_rootlabel(char *devpath, char *devid) 607 { 608 nvlist_t *config = NULL; 609 ldi_handle_t vd_lh; 610 vdev_label_t *label; 611 uint64_t s, size; 612 int l; 613 ddi_devid_t tmpdevid; 614 int error = -1; 615 char *minor_name; 616 617 /* 618 * Read the device label and build the nvlist. 619 */ 620 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid, 621 &minor_name) == 0) { 622 error = ldi_open_by_devid(tmpdevid, minor_name, 623 spa_mode, kcred, &vd_lh, zfs_li); 624 ddi_devid_free(tmpdevid); 625 ddi_devid_str_free(minor_name); 626 } 627 628 if (error && ldi_open_by_name(devpath, FREAD, kcred, &vd_lh, 629 zfs_li)) 630 return (NULL); 631 632 if (ldi_get_size(vd_lh, &s)) { 633 (void) ldi_close(vd_lh, FREAD, kcred); 634 return (NULL); 635 } 636 637 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t); 638 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP); 639 640 for (l = 0; l < VDEV_LABELS; l++) { 641 uint64_t offset, state, txg = 0; 642 643 /* read vdev label */ 644 offset = vdev_label_offset(size, l, 0); 645 if (vdev_disk_physio(vd_lh, (caddr_t)label, 646 VDEV_SKIP_SIZE + VDEV_BOOT_HEADER_SIZE + 647 VDEV_PHYS_SIZE, offset, B_READ) != 0) 648 continue; 649 650 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist, 651 sizeof (label->vl_vdev_phys.vp_nvlist), &config, 0) != 0) { 652 config = NULL; 653 continue; 654 } 655 656 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, 657 &state) != 0 || state >= POOL_STATE_DESTROYED) { 658 nvlist_free(config); 659 config = NULL; 660 continue; 661 } 662 663 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 664 &txg) != 0 || txg == 0) { 665 nvlist_free(config); 666 config = NULL; 667 continue; 668 } 669 670 break; 671 } 672 673 kmem_free(label, sizeof (vdev_label_t)); 674 (void) ldi_close(vd_lh, FREAD, kcred); 675 676 return (config); 677 } 678