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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2016 by Delphix. All rights reserved. 24 * Copyright 2016 Nexenta Systems, Inc. All rights reserved. 25 * Copyright (c) 2013 Joyent, Inc. All rights reserved. 26 */ 27 28 #include <sys/zfs_context.h> 29 #include <sys/spa_impl.h> 30 #include <sys/refcount.h> 31 #include <sys/vdev_disk.h> 32 #include <sys/vdev_impl.h> 33 #include <sys/abd.h> 34 #include <sys/fs/zfs.h> 35 #include <sys/zio.h> 36 #include <sys/sunldi.h> 37 #include <sys/efi_partition.h> 38 #include <sys/fm/fs/zfs.h> 39 40 /* 41 * Tunable parameter for debugging or performance analysis. Setting this 42 * will cause pool corruption on power loss if a volatile out-of-order 43 * write cache is enabled. 44 */ 45 boolean_t zfs_nocacheflush = B_FALSE; 46 47 /* 48 * Virtual device vector for disks. 49 */ 50 51 extern ldi_ident_t zfs_li; 52 53 static void vdev_disk_close(vdev_t *); 54 55 typedef struct vdev_disk_ldi_cb { 56 list_node_t lcb_next; 57 ldi_callback_id_t lcb_id; 58 } vdev_disk_ldi_cb_t; 59 60 static void 61 vdev_disk_alloc(vdev_t *vd) 62 { 63 vdev_disk_t *dvd; 64 65 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP); 66 /* 67 * Create the LDI event callback list. 68 */ 69 list_create(&dvd->vd_ldi_cbs, sizeof (vdev_disk_ldi_cb_t), 70 offsetof(vdev_disk_ldi_cb_t, lcb_next)); 71 } 72 73 static void 74 vdev_disk_free(vdev_t *vd) 75 { 76 vdev_disk_t *dvd = vd->vdev_tsd; 77 vdev_disk_ldi_cb_t *lcb; 78 79 if (dvd == NULL) 80 return; 81 82 /* 83 * We have already closed the LDI handle. Clean up the LDI event 84 * callbacks and free vd->vdev_tsd. 85 */ 86 while ((lcb = list_head(&dvd->vd_ldi_cbs)) != NULL) { 87 list_remove(&dvd->vd_ldi_cbs, lcb); 88 (void) ldi_ev_remove_callbacks(lcb->lcb_id); 89 kmem_free(lcb, sizeof (vdev_disk_ldi_cb_t)); 90 } 91 list_destroy(&dvd->vd_ldi_cbs); 92 kmem_free(dvd, sizeof (vdev_disk_t)); 93 vd->vdev_tsd = NULL; 94 } 95 96 /* ARGSUSED */ 97 static int 98 vdev_disk_off_notify(ldi_handle_t lh, ldi_ev_cookie_t ecookie, void *arg, 99 void *ev_data) 100 { 101 vdev_t *vd = (vdev_t *)arg; 102 vdev_disk_t *dvd = vd->vdev_tsd; 103 104 /* 105 * Ignore events other than offline. 106 */ 107 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0) 108 return (LDI_EV_SUCCESS); 109 110 /* 111 * All LDI handles must be closed for the state change to succeed, so 112 * call on vdev_disk_close() to do this. 113 * 114 * We inform vdev_disk_close that it is being called from offline 115 * notify context so it will defer cleanup of LDI event callbacks and 116 * freeing of vd->vdev_tsd to the offline finalize or a reopen. 117 */ 118 dvd->vd_ldi_offline = B_TRUE; 119 vdev_disk_close(vd); 120 121 /* 122 * Now that the device is closed, request that the spa_async_thread 123 * mark the device as REMOVED and notify FMA of the removal. 124 */ 125 zfs_post_remove(vd->vdev_spa, vd); 126 vd->vdev_remove_wanted = B_TRUE; 127 spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE); 128 129 return (LDI_EV_SUCCESS); 130 } 131 132 /* ARGSUSED */ 133 static void 134 vdev_disk_off_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie, 135 int ldi_result, void *arg, void *ev_data) 136 { 137 vdev_t *vd = (vdev_t *)arg; 138 139 /* 140 * Ignore events other than offline. 141 */ 142 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0) 143 return; 144 145 /* 146 * We have already closed the LDI handle in notify. 147 * Clean up the LDI event callbacks and free vd->vdev_tsd. 148 */ 149 vdev_disk_free(vd); 150 151 /* 152 * Request that the vdev be reopened if the offline state change was 153 * unsuccessful. 154 */ 155 if (ldi_result != LDI_EV_SUCCESS) { 156 vd->vdev_probe_wanted = B_TRUE; 157 spa_async_request(vd->vdev_spa, SPA_ASYNC_PROBE); 158 } 159 } 160 161 static ldi_ev_callback_t vdev_disk_off_callb = { 162 .cb_vers = LDI_EV_CB_VERS, 163 .cb_notify = vdev_disk_off_notify, 164 .cb_finalize = vdev_disk_off_finalize 165 }; 166 167 /* ARGSUSED */ 168 static void 169 vdev_disk_dgrd_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie, 170 int ldi_result, void *arg, void *ev_data) 171 { 172 vdev_t *vd = (vdev_t *)arg; 173 174 /* 175 * Ignore events other than degrade. 176 */ 177 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_DEGRADE) != 0) 178 return; 179 180 /* 181 * Degrade events always succeed. Mark the vdev as degraded. 182 * This status is purely informative for the user. 183 */ 184 (void) vdev_degrade(vd->vdev_spa, vd->vdev_guid, 0); 185 } 186 187 static ldi_ev_callback_t vdev_disk_dgrd_callb = { 188 .cb_vers = LDI_EV_CB_VERS, 189 .cb_notify = NULL, 190 .cb_finalize = vdev_disk_dgrd_finalize 191 }; 192 193 static void 194 vdev_disk_hold(vdev_t *vd) 195 { 196 ddi_devid_t devid; 197 char *minor; 198 199 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER)); 200 201 /* 202 * We must have a pathname, and it must be absolute. 203 */ 204 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') 205 return; 206 207 /* 208 * Only prefetch path and devid info if the device has 209 * never been opened. 210 */ 211 if (vd->vdev_tsd != NULL) 212 return; 213 214 if (vd->vdev_wholedisk == -1ULL) { 215 size_t len = strlen(vd->vdev_path) + 3; 216 char *buf = kmem_alloc(len, KM_SLEEP); 217 218 (void) snprintf(buf, len, "%ss0", vd->vdev_path); 219 220 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp); 221 kmem_free(buf, len); 222 } 223 224 if (vd->vdev_name_vp == NULL) 225 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp); 226 227 if (vd->vdev_devid != NULL && 228 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) { 229 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp); 230 ddi_devid_str_free(minor); 231 ddi_devid_free(devid); 232 } 233 } 234 235 static void 236 vdev_disk_rele(vdev_t *vd) 237 { 238 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER)); 239 240 if (vd->vdev_name_vp) { 241 VN_RELE_ASYNC(vd->vdev_name_vp, 242 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool)); 243 vd->vdev_name_vp = NULL; 244 } 245 if (vd->vdev_devid_vp) { 246 VN_RELE_ASYNC(vd->vdev_devid_vp, 247 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool)); 248 vd->vdev_devid_vp = NULL; 249 } 250 } 251 252 /* 253 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when 254 * even a fallback to DKIOCGMEDIAINFO fails. 255 */ 256 #ifdef DEBUG 257 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__) 258 #else 259 #define VDEV_DEBUG(...) /* Nothing... */ 260 #endif 261 262 static int 263 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize, 264 uint64_t *ashift) 265 { 266 spa_t *spa = vd->vdev_spa; 267 vdev_disk_t *dvd = vd->vdev_tsd; 268 ldi_ev_cookie_t ecookie; 269 vdev_disk_ldi_cb_t *lcb; 270 union { 271 struct dk_minfo_ext ude; 272 struct dk_minfo ud; 273 } dks; 274 struct dk_minfo_ext *dkmext = &dks.ude; 275 struct dk_minfo *dkm = &dks.ud; 276 int error; 277 dev_t dev; 278 int otyp; 279 boolean_t validate_devid = B_FALSE; 280 ddi_devid_t devid; 281 uint64_t capacity = 0, blksz = 0, pbsize; 282 283 /* 284 * We must have a pathname, and it must be absolute. 285 */ 286 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') { 287 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; 288 return (SET_ERROR(EINVAL)); 289 } 290 291 /* 292 * Reopen the device if it's not currently open. Otherwise, 293 * just update the physical size of the device. 294 */ 295 if (dvd != NULL) { 296 if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) { 297 /* 298 * If we are opening a device in its offline notify 299 * context, the LDI handle was just closed. Clean 300 * up the LDI event callbacks and free vd->vdev_tsd. 301 */ 302 vdev_disk_free(vd); 303 } else { 304 ASSERT(vd->vdev_reopening); 305 goto skip_open; 306 } 307 } 308 309 /* 310 * Create vd->vdev_tsd. 311 */ 312 vdev_disk_alloc(vd); 313 dvd = vd->vdev_tsd; 314 315 /* 316 * When opening a disk device, we want to preserve the user's original 317 * intent. We always want to open the device by the path the user gave 318 * us, even if it is one of multiple paths to the same device. But we 319 * also want to be able to survive disks being removed/recabled. 320 * Therefore the sequence of opening devices is: 321 * 322 * 1. Try opening the device by path. For legacy pools without the 323 * 'whole_disk' property, attempt to fix the path by appending 's0'. 324 * 325 * 2. If the devid of the device matches the stored value, return 326 * success. 327 * 328 * 3. Otherwise, the device may have moved. Try opening the device 329 * by the devid instead. 330 */ 331 if (vd->vdev_devid != NULL) { 332 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid, 333 &dvd->vd_minor) != 0) { 334 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; 335 vdev_dbgmsg(vd, "vdev_disk_open: invalid " 336 "vdev_devid '%s'", vd->vdev_devid); 337 return (SET_ERROR(EINVAL)); 338 } 339 } 340 341 error = EINVAL; /* presume failure */ 342 343 if (vd->vdev_path != NULL) { 344 345 if (vd->vdev_wholedisk == -1ULL) { 346 size_t len = strlen(vd->vdev_path) + 3; 347 char *buf = kmem_alloc(len, KM_SLEEP); 348 349 (void) snprintf(buf, len, "%ss0", vd->vdev_path); 350 351 error = ldi_open_by_name(buf, spa_mode(spa), kcred, 352 &dvd->vd_lh, zfs_li); 353 if (error == 0) { 354 spa_strfree(vd->vdev_path); 355 vd->vdev_path = buf; 356 vd->vdev_wholedisk = 1ULL; 357 } else { 358 kmem_free(buf, len); 359 } 360 } 361 362 /* 363 * If we have not yet opened the device, try to open it by the 364 * specified path. 365 */ 366 if (error != 0) { 367 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), 368 kcred, &dvd->vd_lh, zfs_li); 369 } 370 371 /* 372 * Compare the devid to the stored value. 373 */ 374 if (error == 0 && vd->vdev_devid != NULL && 375 ldi_get_devid(dvd->vd_lh, &devid) == 0) { 376 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) { 377 error = SET_ERROR(EINVAL); 378 (void) ldi_close(dvd->vd_lh, spa_mode(spa), 379 kcred); 380 dvd->vd_lh = NULL; 381 } 382 ddi_devid_free(devid); 383 } 384 385 /* 386 * If we succeeded in opening the device, but 'vdev_wholedisk' 387 * is not yet set, then this must be a slice. 388 */ 389 if (error == 0 && vd->vdev_wholedisk == -1ULL) 390 vd->vdev_wholedisk = 0; 391 } 392 393 /* 394 * If we were unable to open by path, or the devid check fails, open by 395 * devid instead. 396 */ 397 if (error != 0 && vd->vdev_devid != NULL) { 398 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor, 399 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li); 400 } 401 402 /* 403 * If all else fails, then try opening by physical path (if available) 404 * or the logical path (if we failed due to the devid check). While not 405 * as reliable as the devid, this will give us something, and the higher 406 * level vdev validation will prevent us from opening the wrong device. 407 */ 408 if (error) { 409 if (vd->vdev_devid != NULL) 410 validate_devid = B_TRUE; 411 412 if (vd->vdev_physpath != NULL && 413 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV) 414 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa), 415 kcred, &dvd->vd_lh, zfs_li); 416 417 /* 418 * Note that we don't support the legacy auto-wholedisk support 419 * as above. This hasn't been used in a very long time and we 420 * don't need to propagate its oddities to this edge condition. 421 */ 422 if (error && vd->vdev_path != NULL) 423 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), 424 kcred, &dvd->vd_lh, zfs_li); 425 } 426 427 if (error) { 428 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; 429 vdev_dbgmsg(vd, "vdev_disk_open: failed to open [error=%d]", 430 error); 431 return (error); 432 } 433 434 /* 435 * Now that the device has been successfully opened, update the devid 436 * if necessary. 437 */ 438 if (validate_devid && spa_writeable(spa) && 439 ldi_get_devid(dvd->vd_lh, &devid) == 0) { 440 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) { 441 char *vd_devid; 442 443 vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor); 444 vdev_dbgmsg(vd, "vdev_disk_open: update devid from " 445 "'%s' to '%s'", vd->vdev_devid, vd_devid); 446 spa_strfree(vd->vdev_devid); 447 vd->vdev_devid = spa_strdup(vd_devid); 448 ddi_devid_str_free(vd_devid); 449 } 450 ddi_devid_free(devid); 451 } 452 453 /* 454 * Once a device is opened, verify that the physical device path (if 455 * available) is up to date. 456 */ 457 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 && 458 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) { 459 char *physpath, *minorname; 460 461 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP); 462 minorname = NULL; 463 if (ddi_dev_pathname(dev, otyp, physpath) == 0 && 464 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 && 465 (vd->vdev_physpath == NULL || 466 strcmp(vd->vdev_physpath, physpath) != 0)) { 467 if (vd->vdev_physpath) 468 spa_strfree(vd->vdev_physpath); 469 (void) strlcat(physpath, ":", MAXPATHLEN); 470 (void) strlcat(physpath, minorname, MAXPATHLEN); 471 vd->vdev_physpath = spa_strdup(physpath); 472 } 473 if (minorname) 474 kmem_free(minorname, strlen(minorname) + 1); 475 kmem_free(physpath, MAXPATHLEN); 476 } 477 478 /* 479 * Register callbacks for the LDI offline event. 480 */ 481 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) == 482 LDI_EV_SUCCESS) { 483 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP); 484 list_insert_tail(&dvd->vd_ldi_cbs, lcb); 485 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie, 486 &vdev_disk_off_callb, (void *) vd, &lcb->lcb_id); 487 } 488 489 /* 490 * Register callbacks for the LDI degrade event. 491 */ 492 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) == 493 LDI_EV_SUCCESS) { 494 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP); 495 list_insert_tail(&dvd->vd_ldi_cbs, lcb); 496 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie, 497 &vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id); 498 } 499 skip_open: 500 /* 501 * Determine the actual size of the device. 502 */ 503 if (ldi_get_size(dvd->vd_lh, psize) != 0) { 504 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; 505 vdev_dbgmsg(vd, "vdev_disk_open: failed to get size"); 506 return (SET_ERROR(EINVAL)); 507 } 508 509 *max_psize = *psize; 510 511 /* 512 * Determine the device's minimum transfer size. 513 * If the ioctl isn't supported, assume DEV_BSIZE. 514 */ 515 if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, 516 (intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) { 517 capacity = dkmext->dki_capacity - 1; 518 blksz = dkmext->dki_lbsize; 519 pbsize = dkmext->dki_pbsize; 520 } else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO, 521 (intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) { 522 VDEV_DEBUG( 523 "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n", 524 vd->vdev_path); 525 capacity = dkm->dki_capacity - 1; 526 blksz = dkm->dki_lbsize; 527 pbsize = blksz; 528 } else { 529 VDEV_DEBUG("vdev_disk_open(\"%s\"): " 530 "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n", 531 vd->vdev_path, error); 532 pbsize = DEV_BSIZE; 533 } 534 535 *ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1; 536 537 if (vd->vdev_wholedisk == 1) { 538 int wce = 1; 539 540 if (error == 0) { 541 /* 542 * If we have the capability to expand, we'd have 543 * found out via success from DKIOCGMEDIAINFO{,EXT}. 544 * Adjust max_psize upward accordingly since we know 545 * we own the whole disk now. 546 */ 547 *max_psize = capacity * blksz; 548 } 549 550 /* 551 * Since we own the whole disk, try to enable disk write 552 * caching. We ignore errors because it's OK if we can't do it. 553 */ 554 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce, 555 FKIOCTL, kcred, NULL); 556 } 557 558 /* 559 * Clear the nowritecache bit, so that on a vdev_reopen() we will 560 * try again. 561 */ 562 vd->vdev_nowritecache = B_FALSE; 563 564 return (0); 565 } 566 567 static void 568 vdev_disk_close(vdev_t *vd) 569 { 570 vdev_disk_t *dvd = vd->vdev_tsd; 571 572 if (vd->vdev_reopening || dvd == NULL) 573 return; 574 575 if (dvd->vd_minor != NULL) { 576 ddi_devid_str_free(dvd->vd_minor); 577 dvd->vd_minor = NULL; 578 } 579 580 if (dvd->vd_devid != NULL) { 581 ddi_devid_free(dvd->vd_devid); 582 dvd->vd_devid = NULL; 583 } 584 585 if (dvd->vd_lh != NULL) { 586 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred); 587 dvd->vd_lh = NULL; 588 } 589 590 vd->vdev_delayed_close = B_FALSE; 591 /* 592 * If we closed the LDI handle due to an offline notify from LDI, 593 * don't free vd->vdev_tsd or unregister the callbacks here; 594 * the offline finalize callback or a reopen will take care of it. 595 */ 596 if (dvd->vd_ldi_offline) 597 return; 598 599 vdev_disk_free(vd); 600 } 601 602 int 603 vdev_disk_physio(vdev_t *vd, caddr_t data, 604 size_t size, uint64_t offset, int flags, boolean_t isdump) 605 { 606 vdev_disk_t *dvd = vd->vdev_tsd; 607 608 /* 609 * If the vdev is closed, it's likely in the REMOVED or FAULTED state. 610 * Nothing to be done here but return failure. 611 */ 612 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) 613 return (EIO); 614 615 ASSERT(vd->vdev_ops == &vdev_disk_ops); 616 617 /* 618 * If in the context of an active crash dump, use the ldi_dump(9F) 619 * call instead of ldi_strategy(9F) as usual. 620 */ 621 if (isdump) { 622 ASSERT3P(dvd, !=, NULL); 623 return (ldi_dump(dvd->vd_lh, data, lbtodb(offset), 624 lbtodb(size))); 625 } 626 627 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags)); 628 } 629 630 int 631 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data, 632 size_t size, uint64_t offset, int flags) 633 { 634 buf_t *bp; 635 int error = 0; 636 637 if (vd_lh == NULL) 638 return (SET_ERROR(EINVAL)); 639 640 ASSERT(flags & B_READ || flags & B_WRITE); 641 642 bp = getrbuf(KM_SLEEP); 643 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST; 644 bp->b_bcount = size; 645 bp->b_un.b_addr = (void *)data; 646 bp->b_lblkno = lbtodb(offset); 647 bp->b_bufsize = size; 648 649 error = ldi_strategy(vd_lh, bp); 650 ASSERT(error == 0); 651 if ((error = biowait(bp)) == 0 && bp->b_resid != 0) 652 error = SET_ERROR(EIO); 653 freerbuf(bp); 654 655 return (error); 656 } 657 658 static void 659 vdev_disk_io_intr(buf_t *bp) 660 { 661 vdev_buf_t *vb = (vdev_buf_t *)bp; 662 zio_t *zio = vb->vb_io; 663 664 /* 665 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO. 666 * Rather than teach the rest of the stack about other error 667 * possibilities (EFAULT, etc), we normalize the error value here. 668 */ 669 zio->io_error = (geterror(bp) != 0 ? EIO : 0); 670 671 if (zio->io_error == 0 && bp->b_resid != 0) 672 zio->io_error = SET_ERROR(EIO); 673 674 if (zio->io_type == ZIO_TYPE_READ) { 675 abd_return_buf_copy(zio->io_abd, bp->b_un.b_addr, zio->io_size); 676 } else { 677 abd_return_buf(zio->io_abd, bp->b_un.b_addr, zio->io_size); 678 } 679 680 kmem_free(vb, sizeof (vdev_buf_t)); 681 682 zio_delay_interrupt(zio); 683 } 684 685 static void 686 vdev_disk_ioctl_free(zio_t *zio) 687 { 688 kmem_free(zio->io_vsd, sizeof (struct dk_callback)); 689 } 690 691 static const zio_vsd_ops_t vdev_disk_vsd_ops = { 692 vdev_disk_ioctl_free, 693 zio_vsd_default_cksum_report 694 }; 695 696 static void 697 vdev_disk_ioctl_done(void *zio_arg, int error) 698 { 699 zio_t *zio = zio_arg; 700 701 zio->io_error = error; 702 703 zio_interrupt(zio); 704 } 705 706 static void 707 vdev_disk_io_start(zio_t *zio) 708 { 709 vdev_t *vd = zio->io_vd; 710 vdev_disk_t *dvd = vd->vdev_tsd; 711 vdev_buf_t *vb; 712 struct dk_callback *dkc; 713 buf_t *bp; 714 int error; 715 716 /* 717 * If the vdev is closed, it's likely in the REMOVED or FAULTED state. 718 * Nothing to be done here but return failure. 719 */ 720 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) { 721 zio->io_error = ENXIO; 722 zio_interrupt(zio); 723 return; 724 } 725 726 if (zio->io_type == ZIO_TYPE_IOCTL) { 727 /* XXPOLICY */ 728 if (!vdev_readable(vd)) { 729 zio->io_error = SET_ERROR(ENXIO); 730 zio_interrupt(zio); 731 return; 732 } 733 734 switch (zio->io_cmd) { 735 736 case DKIOCFLUSHWRITECACHE: 737 738 if (zfs_nocacheflush) 739 break; 740 741 if (vd->vdev_nowritecache) { 742 zio->io_error = SET_ERROR(ENOTSUP); 743 break; 744 } 745 746 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP); 747 zio->io_vsd_ops = &vdev_disk_vsd_ops; 748 749 dkc->dkc_callback = vdev_disk_ioctl_done; 750 dkc->dkc_flag = FLUSH_VOLATILE; 751 dkc->dkc_cookie = zio; 752 753 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd, 754 (uintptr_t)dkc, FKIOCTL, kcred, NULL); 755 756 if (error == 0) { 757 /* 758 * The ioctl will be done asychronously, 759 * and will call vdev_disk_ioctl_done() 760 * upon completion. 761 */ 762 return; 763 } 764 765 zio->io_error = error; 766 767 break; 768 769 default: 770 zio->io_error = SET_ERROR(ENOTSUP); 771 } 772 773 zio_execute(zio); 774 return; 775 } 776 777 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); 778 zio->io_target_timestamp = zio_handle_io_delay(zio); 779 780 vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP); 781 782 vb->vb_io = zio; 783 bp = &vb->vb_buf; 784 785 bioinit(bp); 786 bp->b_flags = B_BUSY | B_NOCACHE | 787 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE); 788 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD))) 789 bp->b_flags |= B_FAILFAST; 790 bp->b_bcount = zio->io_size; 791 792 if (zio->io_type == ZIO_TYPE_READ) { 793 bp->b_un.b_addr = 794 abd_borrow_buf(zio->io_abd, zio->io_size); 795 } else { 796 bp->b_un.b_addr = 797 abd_borrow_buf_copy(zio->io_abd, zio->io_size); 798 } 799 800 bp->b_lblkno = lbtodb(zio->io_offset); 801 bp->b_bufsize = zio->io_size; 802 bp->b_iodone = (int (*)())vdev_disk_io_intr; 803 804 /* ldi_strategy() will return non-zero only on programming errors */ 805 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0); 806 } 807 808 static void 809 vdev_disk_io_done(zio_t *zio) 810 { 811 vdev_t *vd = zio->io_vd; 812 813 /* 814 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if 815 * the device has been removed. If this is the case, then we trigger an 816 * asynchronous removal of the device. Otherwise, probe the device and 817 * make sure it's still accessible. 818 */ 819 if (zio->io_error == EIO && !vd->vdev_remove_wanted) { 820 vdev_disk_t *dvd = vd->vdev_tsd; 821 int state = DKIO_NONE; 822 823 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state, 824 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) { 825 /* 826 * We post the resource as soon as possible, instead of 827 * when the async removal actually happens, because the 828 * DE is using this information to discard previous I/O 829 * errors. 830 */ 831 zfs_post_remove(zio->io_spa, vd); 832 vd->vdev_remove_wanted = B_TRUE; 833 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE); 834 } else if (!vd->vdev_delayed_close) { 835 vd->vdev_delayed_close = B_TRUE; 836 } 837 } 838 } 839 840 vdev_ops_t vdev_disk_ops = { 841 vdev_disk_open, 842 vdev_disk_close, 843 vdev_default_asize, 844 vdev_disk_io_start, 845 vdev_disk_io_done, 846 NULL, 847 vdev_disk_hold, 848 vdev_disk_rele, 849 NULL, 850 vdev_default_xlate, 851 VDEV_TYPE_DISK, /* name of this vdev type */ 852 B_TRUE /* leaf vdev */ 853 }; 854 855 /* 856 * Given the root disk device devid or pathname, read the label from 857 * the device, and construct a configuration nvlist. 858 */ 859 int 860 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config) 861 { 862 ldi_handle_t vd_lh; 863 vdev_label_t *label; 864 uint64_t s, size; 865 int l; 866 ddi_devid_t tmpdevid; 867 int error = -1; 868 char *minor_name; 869 870 /* 871 * Read the device label and build the nvlist. 872 */ 873 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid, 874 &minor_name) == 0) { 875 error = ldi_open_by_devid(tmpdevid, minor_name, 876 FREAD, kcred, &vd_lh, zfs_li); 877 ddi_devid_free(tmpdevid); 878 ddi_devid_str_free(minor_name); 879 } 880 881 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh, 882 zfs_li))) 883 return (error); 884 885 if (ldi_get_size(vd_lh, &s)) { 886 (void) ldi_close(vd_lh, FREAD, kcred); 887 return (SET_ERROR(EIO)); 888 } 889 890 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t); 891 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP); 892 893 *config = NULL; 894 for (l = 0; l < VDEV_LABELS; l++) { 895 uint64_t offset, state, txg = 0; 896 897 /* read vdev label */ 898 offset = vdev_label_offset(size, l, 0); 899 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label, 900 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0) 901 continue; 902 903 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist, 904 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) { 905 *config = NULL; 906 continue; 907 } 908 909 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE, 910 &state) != 0 || state >= POOL_STATE_DESTROYED) { 911 nvlist_free(*config); 912 *config = NULL; 913 continue; 914 } 915 916 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG, 917 &txg) != 0 || txg == 0) { 918 nvlist_free(*config); 919 *config = NULL; 920 continue; 921 } 922 923 break; 924 } 925 926 kmem_free(label, sizeof (vdev_label_t)); 927 (void) ldi_close(vd_lh, FREAD, kcred); 928 if (*config == NULL) 929 error = SET_ERROR(EIDRM); 930 931 return (error); 932 } 933