7938 Port ZOL #3712 disable LBA weighting on files and SSDs

   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, 2015 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/fs/zfs.h>
  34 #include <sys/zio.h>
  35 #include <sys/sunldi.h>
  36 #include <sys/efi_partition.h>
  37 #include <sys/fm/fs/zfs.h>
  38 
  39 /*
  40  * Virtual device vector for disks.
  41  */
  42 
  43 extern ldi_ident_t zfs_li;
  44 
  45 static void vdev_disk_close(vdev_t *);
  46 
  47 typedef struct vdev_disk_ldi_cb {
  48         list_node_t             lcb_next;
  49         ldi_callback_id_t       lcb_id;
  50 } vdev_disk_ldi_cb_t;
  51 
  52 static void
  53 vdev_disk_alloc(vdev_t *vd)
  54 {
  55         vdev_disk_t *dvd;
  56 
  57         dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
  58         /*
  59          * Create the LDI event callback list.
  60          */
  61         list_create(&dvd->vd_ldi_cbs, sizeof (vdev_disk_ldi_cb_t),
  62             offsetof(vdev_disk_ldi_cb_t, lcb_next));
  63 }
  64 
  65 static void
  66 vdev_disk_free(vdev_t *vd)
  67 {
  68         vdev_disk_t *dvd = vd->vdev_tsd;
  69         vdev_disk_ldi_cb_t *lcb;
  70 
  71         if (dvd == NULL)
  72                 return;
  73 
  74         /*
  75          * We have already closed the LDI handle. Clean up the LDI event
  76          * callbacks and free vd->vdev_tsd.
  77          */
  78         while ((lcb = list_head(&dvd->vd_ldi_cbs)) != NULL) {
  79                 list_remove(&dvd->vd_ldi_cbs, lcb);
  80                 (void) ldi_ev_remove_callbacks(lcb->lcb_id);
  81                 kmem_free(lcb, sizeof (vdev_disk_ldi_cb_t));
  82         }
  83         list_destroy(&dvd->vd_ldi_cbs);
  84         kmem_free(dvd, sizeof (vdev_disk_t));
  85         vd->vdev_tsd = NULL;
  86 }
  87 
  88 /* ARGSUSED */
  89 static int
  90 vdev_disk_off_notify(ldi_handle_t lh, ldi_ev_cookie_t ecookie, void *arg,
  91     void *ev_data)
  92 {
  93         vdev_t *vd = (vdev_t *)arg;
  94         vdev_disk_t *dvd = vd->vdev_tsd;
  95 
  96         /*
  97          * Ignore events other than offline.
  98          */
  99         if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
 100                 return (LDI_EV_SUCCESS);
 101 
 102         /*
 103          * All LDI handles must be closed for the state change to succeed, so
 104          * call on vdev_disk_close() to do this.
 105          *
 106          * We inform vdev_disk_close that it is being called from offline
 107          * notify context so it will defer cleanup of LDI event callbacks and
 108          * freeing of vd->vdev_tsd to the offline finalize or a reopen.
 109          */
 110         dvd->vd_ldi_offline = B_TRUE;
 111         vdev_disk_close(vd);
 112 
 113         /*
 114          * Now that the device is closed, request that the spa_async_thread
 115          * mark the device as REMOVED and notify FMA of the removal.
 116          */
 117         zfs_post_remove(vd->vdev_spa, vd);
 118         vd->vdev_remove_wanted = B_TRUE;
 119         spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
 120 
 121         return (LDI_EV_SUCCESS);
 122 }
 123 
 124 /* ARGSUSED */
 125 static void
 126 vdev_disk_off_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
 127     int ldi_result, void *arg, void *ev_data)
 128 {
 129         vdev_t *vd = (vdev_t *)arg;
 130 
 131         /*
 132          * Ignore events other than offline.
 133          */
 134         if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
 135                 return;
 136 
 137         /*
 138          * We have already closed the LDI handle in notify.
 139          * Clean up the LDI event callbacks and free vd->vdev_tsd.
 140          */
 141         vdev_disk_free(vd);
 142 
 143         /*
 144          * Request that the vdev be reopened if the offline state change was
 145          * unsuccessful.
 146          */
 147         if (ldi_result != LDI_EV_SUCCESS) {
 148                 vd->vdev_probe_wanted = B_TRUE;
 149                 spa_async_request(vd->vdev_spa, SPA_ASYNC_PROBE);
 150         }
 151 }
 152 
 153 static ldi_ev_callback_t vdev_disk_off_callb = {
 154         .cb_vers = LDI_EV_CB_VERS,
 155         .cb_notify = vdev_disk_off_notify,
 156         .cb_finalize = vdev_disk_off_finalize
 157 };
 158 
 159 /* ARGSUSED */
 160 static void
 161 vdev_disk_dgrd_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
 162     int ldi_result, void *arg, void *ev_data)
 163 {
 164         vdev_t *vd = (vdev_t *)arg;
 165 
 166         /*
 167          * Ignore events other than degrade.
 168          */
 169         if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_DEGRADE) != 0)
 170                 return;
 171 
 172         /*
 173          * Degrade events always succeed. Mark the vdev as degraded.
 174          * This status is purely informative for the user.
 175          */
 176         (void) vdev_degrade(vd->vdev_spa, vd->vdev_guid, 0);
 177 }
 178 
 179 static ldi_ev_callback_t vdev_disk_dgrd_callb = {
 180         .cb_vers = LDI_EV_CB_VERS,
 181         .cb_notify = NULL,
 182         .cb_finalize = vdev_disk_dgrd_finalize
 183 };
 184 
 185 static void
 186 vdev_disk_hold(vdev_t *vd)
 187 {
 188         ddi_devid_t devid;
 189         char *minor;
 190 
 191         ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
 192 
 193         /*
 194          * We must have a pathname, and it must be absolute.
 195          */
 196         if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
 197                 return;
 198 
 199         /*
 200          * Only prefetch path and devid info if the device has
 201          * never been opened.
 202          */
 203         if (vd->vdev_tsd != NULL)
 204                 return;
 205 
 206         if (vd->vdev_wholedisk == -1ULL) {
 207                 size_t len = strlen(vd->vdev_path) + 3;
 208                 char *buf = kmem_alloc(len, KM_SLEEP);
 209 
 210                 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
 211 
 212                 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
 213                 kmem_free(buf, len);
 214         }
 215 
 216         if (vd->vdev_name_vp == NULL)
 217                 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
 218 
 219         if (vd->vdev_devid != NULL &&
 220             ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
 221                 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
 222                 ddi_devid_str_free(minor);
 223                 ddi_devid_free(devid);
 224         }
 225 }
 226 
 227 static void
 228 vdev_disk_rele(vdev_t *vd)
 229 {
 230         ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
 231 
 232         if (vd->vdev_name_vp) {
 233                 VN_RELE_ASYNC(vd->vdev_name_vp,
 234                     dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
 235                 vd->vdev_name_vp = NULL;
 236         }
 237         if (vd->vdev_devid_vp) {
 238                 VN_RELE_ASYNC(vd->vdev_devid_vp,
 239                     dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
 240                 vd->vdev_devid_vp = NULL;
 241         }
 242 }
 243 
 244 /*
 245  * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
 246  * even a fallback to DKIOCGMEDIAINFO fails.
 247  */
 248 #ifdef DEBUG
 249 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__)
 250 #else
 251 #define VDEV_DEBUG(...) /* Nothing... */
 252 #endif
 253 
 254 static int
 255 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
 256     uint64_t *ashift)
 257 {
 258         spa_t *spa = vd->vdev_spa;
 259         vdev_disk_t *dvd = vd->vdev_tsd;
 260         ldi_ev_cookie_t ecookie;
 261         vdev_disk_ldi_cb_t *lcb;
 262         union {
 263                 struct dk_minfo_ext ude;
 264                 struct dk_minfo ud;
 265         } dks;
 266         struct dk_minfo_ext *dkmext = &dks.ude;
 267         struct dk_minfo *dkm = &dks.ud;
 268         int error;
 269         dev_t dev;
 270         int otyp;
 271         boolean_t validate_devid = B_FALSE;
 272         ddi_devid_t devid;
 273         uint64_t capacity = 0, blksz = 0, pbsize;


 274 
 275         /*
 276          * We must have a pathname, and it must be absolute.
 277          */
 278         if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
 279                 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
 280                 return (SET_ERROR(EINVAL));
 281         }
 282 
 283         /*
 284          * Reopen the device if it's not currently open. Otherwise,
 285          * just update the physical size of the device.
 286          */
 287         if (dvd != NULL) {
 288                 if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) {
 289                         /*
 290                          * If we are opening a device in its offline notify
 291                          * context, the LDI handle was just closed. Clean
 292                          * up the LDI event callbacks and free vd->vdev_tsd.
 293                          */
 294                         vdev_disk_free(vd);
 295                 } else {
 296                         ASSERT(vd->vdev_reopening);
 297                         goto skip_open;
 298                 }
 299         }
 300 
 301         /*
 302          * Create vd->vdev_tsd.
 303          */
 304         vdev_disk_alloc(vd);
 305         dvd = vd->vdev_tsd;
 306 
 307         /*
 308          * When opening a disk device, we want to preserve the user's original
 309          * intent.  We always want to open the device by the path the user gave
 310          * us, even if it is one of multiple paths to the same device.  But we
 311          * also want to be able to survive disks being removed/recabled.
 312          * Therefore the sequence of opening devices is:
 313          *
 314          * 1. Try opening the device by path.  For legacy pools without the
 315          *    'whole_disk' property, attempt to fix the path by appending 's0'.
 316          *
 317          * 2. If the devid of the device matches the stored value, return
 318          *    success.
 319          *
 320          * 3. Otherwise, the device may have moved.  Try opening the device
 321          *    by the devid instead.
 322          */
 323         if (vd->vdev_devid != NULL) {
 324                 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
 325                     &dvd->vd_minor) != 0) {
 326                         vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
 327                         return (SET_ERROR(EINVAL));
 328                 }
 329         }
 330 
 331         error = EINVAL;         /* presume failure */
 332 
 333         if (vd->vdev_path != NULL) {
 334 
 335                 if (vd->vdev_wholedisk == -1ULL) {
 336                         size_t len = strlen(vd->vdev_path) + 3;
 337                         char *buf = kmem_alloc(len, KM_SLEEP);
 338 
 339                         (void) snprintf(buf, len, "%ss0", vd->vdev_path);
 340 
 341                         error = ldi_open_by_name(buf, spa_mode(spa), kcred,
 342                             &dvd->vd_lh, zfs_li);
 343                         if (error == 0) {
 344                                 spa_strfree(vd->vdev_path);
 345                                 vd->vdev_path = buf;
 346                                 vd->vdev_wholedisk = 1ULL;
 347                         } else {
 348                                 kmem_free(buf, len);
 349                         }
 350                 }
 351 
 352                 /*
 353                  * If we have not yet opened the device, try to open it by the
 354                  * specified path.
 355                  */
 356                 if (error != 0) {
 357                         error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
 358                             kcred, &dvd->vd_lh, zfs_li);
 359                 }
 360 
 361                 /*
 362                  * Compare the devid to the stored value.
 363                  */
 364                 if (error == 0 && vd->vdev_devid != NULL &&
 365                     ldi_get_devid(dvd->vd_lh, &devid) == 0) {
 366                         if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
 367                                 error = SET_ERROR(EINVAL);
 368                                 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
 369                                     kcred);
 370                                 dvd->vd_lh = NULL;
 371                         }
 372                         ddi_devid_free(devid);
 373                 }
 374 
 375                 /*
 376                  * If we succeeded in opening the device, but 'vdev_wholedisk'
 377                  * is not yet set, then this must be a slice.
 378                  */
 379                 if (error == 0 && vd->vdev_wholedisk == -1ULL)
 380                         vd->vdev_wholedisk = 0;
 381         }
 382 
 383         /*
 384          * If we were unable to open by path, or the devid check fails, open by
 385          * devid instead.
 386          */
 387         if (error != 0 && vd->vdev_devid != NULL) {
 388                 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
 389                     spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
 390         }
 391 
 392         /*
 393          * If all else fails, then try opening by physical path (if available)
 394          * or the logical path (if we failed due to the devid check).  While not
 395          * as reliable as the devid, this will give us something, and the higher
 396          * level vdev validation will prevent us from opening the wrong device.
 397          */
 398         if (error) {
 399                 if (vd->vdev_devid != NULL)
 400                         validate_devid = B_TRUE;
 401 
 402                 if (vd->vdev_physpath != NULL &&
 403                     (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
 404                         error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
 405                             kcred, &dvd->vd_lh, zfs_li);
 406 
 407                 /*
 408                  * Note that we don't support the legacy auto-wholedisk support
 409                  * as above.  This hasn't been used in a very long time and we
 410                  * don't need to propagate its oddities to this edge condition.
 411                  */
 412                 if (error && vd->vdev_path != NULL)
 413                         error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
 414                             kcred, &dvd->vd_lh, zfs_li);
 415         }
 416 
 417         if (error) {
 418                 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
 419                 return (error);
 420         }
 421 
 422         /*
 423          * Now that the device has been successfully opened, update the devid
 424          * if necessary.
 425          */
 426         if (validate_devid && spa_writeable(spa) &&
 427             ldi_get_devid(dvd->vd_lh, &devid) == 0) {
 428                 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
 429                         char *vd_devid;
 430 
 431                         vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
 432                         zfs_dbgmsg("vdev %s: update devid from %s, "
 433                             "to %s", vd->vdev_path, vd->vdev_devid, vd_devid);
 434                         spa_strfree(vd->vdev_devid);
 435                         vd->vdev_devid = spa_strdup(vd_devid);
 436                         ddi_devid_str_free(vd_devid);
 437                 }
 438                 ddi_devid_free(devid);
 439         }
 440 
 441         /*
 442          * Once a device is opened, verify that the physical device path (if
 443          * available) is up to date.
 444          */
 445         if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
 446             ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
 447                 char *physpath, *minorname;
 448 
 449                 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
 450                 minorname = NULL;
 451                 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
 452                     ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
 453                     (vd->vdev_physpath == NULL ||
 454                     strcmp(vd->vdev_physpath, physpath) != 0)) {
 455                         if (vd->vdev_physpath)
 456                                 spa_strfree(vd->vdev_physpath);
 457                         (void) strlcat(physpath, ":", MAXPATHLEN);
 458                         (void) strlcat(physpath, minorname, MAXPATHLEN);
 459                         vd->vdev_physpath = spa_strdup(physpath);
 460                 }
 461                 if (minorname)
 462                         kmem_free(minorname, strlen(minorname) + 1);
 463                 kmem_free(physpath, MAXPATHLEN);
 464         }
 465 
 466         /*
 467          * Register callbacks for the LDI offline event.
 468          */
 469         if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
 470             LDI_EV_SUCCESS) {
 471                 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
 472                 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
 473                 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
 474                     &vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
 475         }
 476 
 477         /*
 478          * Register callbacks for the LDI degrade event.
 479          */
 480         if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
 481             LDI_EV_SUCCESS) {
 482                 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
 483                 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
 484                 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
 485                     &vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
 486         }
 487 skip_open:
 488         /*
 489          * Determine the actual size of the device.
 490          */
 491         if (ldi_get_size(dvd->vd_lh, psize) != 0) {
 492                 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
 493                 return (SET_ERROR(EINVAL));
 494         }
 495 
 496         *max_psize = *psize;
 497 
 498         /*
 499          * Determine the device's minimum transfer size.
 500          * If the ioctl isn't supported, assume DEV_BSIZE.
 501          */
 502         if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
 503             (intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
 504                 capacity = dkmext->dki_capacity - 1;
 505                 blksz = dkmext->dki_lbsize;
 506                 pbsize = dkmext->dki_pbsize;
 507         } else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
 508             (intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
 509                 VDEV_DEBUG(
 510                     "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
 511                     vd->vdev_path);
 512                 capacity = dkm->dki_capacity - 1;
 513                 blksz = dkm->dki_lbsize;
 514                 pbsize = blksz;
 515         } else {
 516                 VDEV_DEBUG("vdev_disk_open(\"%s\"): "
 517                     "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
 518                     vd->vdev_path, error);
 519                 pbsize = DEV_BSIZE;
 520         }
 521 
 522         *ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
 523 
 524         if (vd->vdev_wholedisk == 1) {
 525                 int wce = 1;
 526 
 527                 if (error == 0) {
 528                         /*
 529                          * If we have the capability to expand, we'd have
 530                          * found out via success from DKIOCGMEDIAINFO{,EXT}.
 531                          * Adjust max_psize upward accordingly since we know
 532                          * we own the whole disk now.
 533                          */
 534                         *max_psize = capacity * blksz;
 535                 }
 536 
 537                 /*
 538                  * Since we own the whole disk, try to enable disk write
 539                  * caching.  We ignore errors because it's OK if we can't do it.
 540                  */
 541                 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
 542                     FKIOCTL, kcred, NULL);
 543         }
 544 
 545         /*






















 546          * Clear the nowritecache bit, so that on a vdev_reopen() we will
 547          * try again.
 548          */
 549         vd->vdev_nowritecache = B_FALSE;
 550 
 551         return (0);
 552 }
 553 
 554 static void
 555 vdev_disk_close(vdev_t *vd)
 556 {
 557         vdev_disk_t *dvd = vd->vdev_tsd;
 558 
 559         if (vd->vdev_reopening || dvd == NULL)
 560                 return;
 561 
 562         if (dvd->vd_minor != NULL) {
 563                 ddi_devid_str_free(dvd->vd_minor);
 564                 dvd->vd_minor = NULL;
 565         }
 566 
 567         if (dvd->vd_devid != NULL) {
 568                 ddi_devid_free(dvd->vd_devid);
 569                 dvd->vd_devid = NULL;
 570         }
 571 
 572         if (dvd->vd_lh != NULL) {
 573                 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
 574                 dvd->vd_lh = NULL;
 575         }
 576 
 577         vd->vdev_delayed_close = B_FALSE;
 578         /*
 579          * If we closed the LDI handle due to an offline notify from LDI,
 580          * don't free vd->vdev_tsd or unregister the callbacks here;
 581          * the offline finalize callback or a reopen will take care of it.
 582          */
 583         if (dvd->vd_ldi_offline)
 584                 return;
 585 
 586         vdev_disk_free(vd);
 587 }
 588 
 589 int
 590 vdev_disk_physio(vdev_t *vd, caddr_t data,
 591     size_t size, uint64_t offset, int flags, boolean_t isdump)
 592 {
 593         vdev_disk_t *dvd = vd->vdev_tsd;
 594 
 595         /*
 596          * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
 597          * Nothing to be done here but return failure.
 598          */
 599         if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL))
 600                 return (EIO);
 601 
 602         ASSERT(vd->vdev_ops == &vdev_disk_ops);
 603 
 604         /*
 605          * If in the context of an active crash dump, use the ldi_dump(9F)
 606          * call instead of ldi_strategy(9F) as usual.
 607          */
 608         if (isdump) {
 609                 ASSERT3P(dvd, !=, NULL);
 610                 return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
 611                     lbtodb(size)));
 612         }
 613 
 614         return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
 615 }
 616 
 617 int
 618 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
 619     size_t size, uint64_t offset, int flags)
 620 {
 621         buf_t *bp;
 622         int error = 0;
 623 
 624         if (vd_lh == NULL)
 625                 return (SET_ERROR(EINVAL));
 626 
 627         ASSERT(flags & B_READ || flags & B_WRITE);
 628 
 629         bp = getrbuf(KM_SLEEP);
 630         bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
 631         bp->b_bcount = size;
 632         bp->b_un.b_addr = (void *)data;
 633         bp->b_lblkno = lbtodb(offset);
 634         bp->b_bufsize = size;
 635 
 636         error = ldi_strategy(vd_lh, bp);
 637         ASSERT(error == 0);
 638         if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
 639                 error = SET_ERROR(EIO);
 640         freerbuf(bp);
 641 
 642         return (error);
 643 }
 644 
 645 static void
 646 vdev_disk_io_intr(buf_t *bp)
 647 {
 648         vdev_buf_t *vb = (vdev_buf_t *)bp;
 649         zio_t *zio = vb->vb_io;
 650 
 651         /*
 652          * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
 653          * Rather than teach the rest of the stack about other error
 654          * possibilities (EFAULT, etc), we normalize the error value here.
 655          */
 656         zio->io_error = (geterror(bp) != 0 ? EIO : 0);
 657 
 658         if (zio->io_error == 0 && bp->b_resid != 0)
 659                 zio->io_error = SET_ERROR(EIO);
 660 
 661         kmem_free(vb, sizeof (vdev_buf_t));
 662 
 663         zio_delay_interrupt(zio);
 664 }
 665 
 666 static void
 667 vdev_disk_ioctl_free(zio_t *zio)
 668 {
 669         kmem_free(zio->io_vsd, sizeof (struct dk_callback));
 670 }
 671 
 672 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
 673         vdev_disk_ioctl_free,
 674         zio_vsd_default_cksum_report
 675 };
 676 
 677 static void
 678 vdev_disk_ioctl_done(void *zio_arg, int error)
 679 {
 680         zio_t *zio = zio_arg;
 681 
 682         zio->io_error = error;
 683 
 684         zio_interrupt(zio);
 685 }
 686 
 687 static void
 688 vdev_disk_io_start(zio_t *zio)
 689 {
 690         vdev_t *vd = zio->io_vd;
 691         vdev_disk_t *dvd = vd->vdev_tsd;
 692         vdev_buf_t *vb;
 693         struct dk_callback *dkc;
 694         buf_t *bp;
 695         int error;
 696 
 697         /*
 698          * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
 699          * Nothing to be done here but return failure.
 700          */
 701         if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) {
 702                 zio->io_error = ENXIO;
 703                 zio_interrupt(zio);
 704                 return;
 705         }
 706 
 707         if (zio->io_type == ZIO_TYPE_IOCTL) {
 708                 /* XXPOLICY */
 709                 if (!vdev_readable(vd)) {
 710                         zio->io_error = SET_ERROR(ENXIO);
 711                         zio_interrupt(zio);
 712                         return;
 713                 }
 714 
 715                 switch (zio->io_cmd) {
 716 
 717                 case DKIOCFLUSHWRITECACHE:
 718 
 719                         if (zfs_nocacheflush)
 720                                 break;
 721 
 722                         if (vd->vdev_nowritecache) {
 723                                 zio->io_error = SET_ERROR(ENOTSUP);
 724                                 break;
 725                         }
 726 
 727                         zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
 728                         zio->io_vsd_ops = &vdev_disk_vsd_ops;
 729 
 730                         dkc->dkc_callback = vdev_disk_ioctl_done;
 731                         dkc->dkc_flag = FLUSH_VOLATILE;
 732                         dkc->dkc_cookie = zio;
 733 
 734                         error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
 735                             (uintptr_t)dkc, FKIOCTL, kcred, NULL);
 736 
 737                         if (error == 0) {
 738                                 /*
 739                                  * The ioctl will be done asychronously,
 740                                  * and will call vdev_disk_ioctl_done()
 741                                  * upon completion.
 742                                  */
 743                                 return;
 744                         }
 745 
 746                         zio->io_error = error;
 747 
 748                         break;
 749 
 750                 default:
 751                         zio->io_error = SET_ERROR(ENOTSUP);
 752                 }
 753 
 754                 zio_execute(zio);
 755                 return;
 756         }
 757 
 758         ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
 759         zio->io_target_timestamp = zio_handle_io_delay(zio);
 760 
 761         vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
 762 
 763         vb->vb_io = zio;
 764         bp = &vb->vb_buf;
 765 
 766         bioinit(bp);
 767         bp->b_flags = B_BUSY | B_NOCACHE |
 768             (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
 769         if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
 770                 bp->b_flags |= B_FAILFAST;
 771         bp->b_bcount = zio->io_size;
 772         bp->b_un.b_addr = zio->io_data;
 773         bp->b_lblkno = lbtodb(zio->io_offset);
 774         bp->b_bufsize = zio->io_size;
 775         bp->b_iodone = (int (*)())vdev_disk_io_intr;
 776 
 777         /* ldi_strategy() will return non-zero only on programming errors */
 778         VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
 779 }
 780 
 781 static void
 782 vdev_disk_io_done(zio_t *zio)
 783 {
 784         vdev_t *vd = zio->io_vd;
 785 
 786         /*
 787          * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
 788          * the device has been removed.  If this is the case, then we trigger an
 789          * asynchronous removal of the device. Otherwise, probe the device and
 790          * make sure it's still accessible.
 791          */
 792         if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
 793                 vdev_disk_t *dvd = vd->vdev_tsd;
 794                 int state = DKIO_NONE;
 795 
 796                 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
 797                     FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
 798                         /*
 799                          * We post the resource as soon as possible, instead of
 800                          * when the async removal actually happens, because the
 801                          * DE is using this information to discard previous I/O
 802                          * errors.
 803                          */
 804                         zfs_post_remove(zio->io_spa, vd);
 805                         vd->vdev_remove_wanted = B_TRUE;
 806                         spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
 807                 } else if (!vd->vdev_delayed_close) {
 808                         vd->vdev_delayed_close = B_TRUE;
 809                 }
 810         }
 811 }
 812 
 813 vdev_ops_t vdev_disk_ops = {
 814         vdev_disk_open,
 815         vdev_disk_close,
 816         vdev_default_asize,
 817         vdev_disk_io_start,
 818         vdev_disk_io_done,
 819         NULL,
 820         vdev_disk_hold,
 821         vdev_disk_rele,
 822         VDEV_TYPE_DISK,         /* name of this vdev type */
 823         B_TRUE                  /* leaf vdev */
 824 };
 825 
 826 /*
 827  * Given the root disk device devid or pathname, read the label from
 828  * the device, and construct a configuration nvlist.
 829  */
 830 int
 831 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
 832 {
 833         ldi_handle_t vd_lh;
 834         vdev_label_t *label;
 835         uint64_t s, size;
 836         int l;
 837         ddi_devid_t tmpdevid;
 838         int error = -1;
 839         char *minor_name;
 840 
 841         /*
 842          * Read the device label and build the nvlist.
 843          */
 844         if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
 845             &minor_name) == 0) {
 846                 error = ldi_open_by_devid(tmpdevid, minor_name,
 847                     FREAD, kcred, &vd_lh, zfs_li);
 848                 ddi_devid_free(tmpdevid);
 849                 ddi_devid_str_free(minor_name);
 850         }
 851 
 852         if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
 853             zfs_li)))
 854                 return (error);
 855 
 856         if (ldi_get_size(vd_lh, &s)) {
 857                 (void) ldi_close(vd_lh, FREAD, kcred);
 858                 return (SET_ERROR(EIO));
 859         }
 860 
 861         size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
 862         label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
 863 
 864         *config = NULL;
 865         for (l = 0; l < VDEV_LABELS; l++) {
 866                 uint64_t offset, state, txg = 0;
 867 
 868                 /* read vdev label */
 869                 offset = vdev_label_offset(size, l, 0);
 870                 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
 871                     VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
 872                         continue;
 873 
 874                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 875                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
 876                         *config = NULL;
 877                         continue;
 878                 }
 879 
 880                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 881                     &state) != 0 || state >= POOL_STATE_DESTROYED) {
 882                         nvlist_free(*config);
 883                         *config = NULL;
 884                         continue;
 885                 }
 886 
 887                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 888                     &txg) != 0 || txg == 0) {
 889                         nvlist_free(*config);
 890                         *config = NULL;
 891                         continue;
 892                 }
 893 
 894                 break;
 895         }
 896 
 897         kmem_free(label, sizeof (vdev_label_t));
 898         (void) ldi_close(vd_lh, FREAD, kcred);
 899         if (*config == NULL)
 900                 error = SET_ERROR(EIDRM);
 901 
 902         return (error);
 903 }
--- EOF ---