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 /*
  23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2013 by Delphix. All rights reserved.
  26  */
  27 
  28 #include <sys/conf.h>
  29 #include <sys/file.h>
  30 #include <sys/ddi.h>
  31 #include <sys/sunddi.h>
  32 #include <sys/modctl.h>
  33 #include <sys/scsi/scsi.h>
  34 #include <sys/scsi/impl/scsi_reset_notify.h>
  35 #include <sys/disp.h>
  36 #include <sys/byteorder.h>
  37 #include <sys/pathname.h>
  38 #include <sys/atomic.h>
  39 #include <sys/nvpair.h>
  40 #include <sys/fs/zfs.h>
  41 #include <sys/sdt.h>
  42 #include <sys/dkio.h>
  43 #include <sys/zfs_ioctl.h>
  44 
  45 #include <sys/stmf.h>
  46 #include <sys/lpif.h>
  47 #include <sys/stmf_ioctl.h>
  48 #include <sys/stmf_sbd_ioctl.h>
  49 
  50 #include "stmf_sbd.h"
  51 #include "sbd_impl.h"
  52 
  53 #define SBD_IS_ZVOL(zvol)       (strncmp("/dev/zvol", zvol, 9))
  54 
  55 extern sbd_status_t sbd_pgr_meta_init(sbd_lu_t *sl);
  56 extern sbd_status_t sbd_pgr_meta_load(sbd_lu_t *sl);
  57 extern void sbd_pgr_reset(sbd_lu_t *sl);
  58 
  59 static int sbd_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg,
  60     void **result);
  61 static int sbd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
  62 static int sbd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
  63 static int sbd_open(dev_t *devp, int flag, int otype, cred_t *credp);
  64 static int sbd_close(dev_t dev, int flag, int otype, cred_t *credp);
  65 static int stmf_sbd_ioctl(dev_t dev, int cmd, intptr_t data, int mode,
  66     cred_t *credp, int *rval);
  67 void sbd_lp_cb(stmf_lu_provider_t *lp, int cmd, void *arg, uint32_t flags);
  68 stmf_status_t sbd_proxy_reg_lu(uint8_t *luid, void *proxy_reg_arg,
  69     uint32_t proxy_reg_arg_len);
  70 stmf_status_t sbd_proxy_dereg_lu(uint8_t *luid, void *proxy_reg_arg,
  71     uint32_t proxy_reg_arg_len);
  72 stmf_status_t sbd_proxy_msg(uint8_t *luid, void *proxy_arg,
  73     uint32_t proxy_arg_len, uint32_t type);
  74 int sbd_create_register_lu(sbd_create_and_reg_lu_t *slu, int struct_sz,
  75     uint32_t *err_ret);
  76 int sbd_create_standby_lu(sbd_create_standby_lu_t *slu, uint32_t *err_ret);
  77 int sbd_set_lu_standby(sbd_set_lu_standby_t *stlu, uint32_t *err_ret);
  78 int sbd_import_lu(sbd_import_lu_t *ilu, int struct_sz, uint32_t *err_ret,
  79     int no_register, sbd_lu_t **slr);
  80 int sbd_import_active_lu(sbd_import_lu_t *ilu, sbd_lu_t *sl, uint32_t *err_ret);
  81 int sbd_delete_lu(sbd_delete_lu_t *dlu, int struct_sz, uint32_t *err_ret);
  82 int sbd_modify_lu(sbd_modify_lu_t *mlu, int struct_sz, uint32_t *err_ret);
  83 int sbd_set_global_props(sbd_global_props_t *mlu, int struct_sz,
  84     uint32_t *err_ret);
  85 int sbd_get_global_props(sbd_global_props_t *oslp, uint32_t oslp_sz,
  86     uint32_t *err_ret);
  87 int sbd_get_lu_props(sbd_lu_props_t *islp, uint32_t islp_sz,
  88     sbd_lu_props_t *oslp, uint32_t oslp_sz, uint32_t *err_ret);
  89 static char *sbd_get_zvol_name(sbd_lu_t *);
  90 static int sbd_get_unmap_props(sbd_unmap_props_t *sup, sbd_unmap_props_t *osup,
  91     uint32_t *err_ret);
  92 sbd_status_t sbd_create_zfs_meta_object(sbd_lu_t *sl);
  93 sbd_status_t sbd_open_zfs_meta(sbd_lu_t *sl);
  94 sbd_status_t sbd_read_zfs_meta(sbd_lu_t *sl, uint8_t *buf, uint64_t sz,
  95     uint64_t off);
  96 sbd_status_t sbd_write_zfs_meta(sbd_lu_t *sl, uint8_t *buf, uint64_t sz,
  97     uint64_t off);
  98 sbd_status_t sbd_update_zfs_prop(sbd_lu_t *sl);
  99 int sbd_is_zvol(char *path);
 100 int sbd_zvolget(char *zvol_name, char **comstarprop);
 101 int sbd_zvolset(char *zvol_name, char *comstarprop);
 102 char sbd_ctoi(char c);
 103 void sbd_close_lu(sbd_lu_t *sl);
 104 
 105 static ldi_ident_t      sbd_zfs_ident;
 106 static stmf_lu_provider_t *sbd_lp;
 107 static sbd_lu_t         *sbd_lu_list = NULL;
 108 static kmutex_t         sbd_lock;
 109 static dev_info_t       *sbd_dip;
 110 static uint32_t         sbd_lu_count = 0;
 111 
 112 /* Global property settings for the logical unit */
 113 char sbd_vendor_id[]    = "SUN     ";
 114 char sbd_product_id[]   = "COMSTAR         ";
 115 char sbd_revision[]     = "1.0 ";
 116 char *sbd_mgmt_url = NULL;
 117 uint16_t sbd_mgmt_url_alloc_size = 0;
 118 krwlock_t sbd_global_prop_lock;
 119 
 120 static char sbd_name[] = "sbd";
 121 
 122 static struct cb_ops sbd_cb_ops = {
 123         sbd_open,                       /* open */
 124         sbd_close,                      /* close */
 125         nodev,                          /* strategy */
 126         nodev,                          /* print */
 127         nodev,                          /* dump */
 128         nodev,                          /* read */
 129         nodev,                          /* write */
 130         stmf_sbd_ioctl,                 /* ioctl */
 131         nodev,                          /* devmap */
 132         nodev,                          /* mmap */
 133         nodev,                          /* segmap */
 134         nochpoll,                       /* chpoll */
 135         ddi_prop_op,                    /* cb_prop_op */
 136         0,                              /* streamtab */
 137         D_NEW | D_MP,                   /* cb_flag */
 138         CB_REV,                         /* rev */
 139         nodev,                          /* aread */
 140         nodev                           /* awrite */
 141 };
 142 
 143 static struct dev_ops sbd_ops = {
 144         DEVO_REV,
 145         0,
 146         sbd_getinfo,
 147         nulldev,                /* identify */
 148         nulldev,                /* probe */
 149         sbd_attach,
 150         sbd_detach,
 151         nodev,                  /* reset */
 152         &sbd_cb_ops,
 153         NULL,                   /* bus_ops */
 154         NULL                    /* power */
 155 };
 156 
 157 #define SBD_NAME        "COMSTAR SBD"
 158 
 159 static struct modldrv modldrv = {
 160         &mod_driverops,
 161         SBD_NAME,
 162         &sbd_ops
 163 };
 164 
 165 static struct modlinkage modlinkage = {
 166         MODREV_1,
 167         &modldrv,
 168         NULL
 169 };
 170 
 171 int
 172 _init(void)
 173 {
 174         int ret;
 175 
 176         ret = mod_install(&modlinkage);
 177         if (ret)
 178                 return (ret);
 179         sbd_lp = (stmf_lu_provider_t *)stmf_alloc(STMF_STRUCT_LU_PROVIDER,
 180             0, 0);
 181         sbd_lp->lp_lpif_rev = LPIF_REV_2;
 182         sbd_lp->lp_instance = 0;
 183         sbd_lp->lp_name = sbd_name;
 184         sbd_lp->lp_cb = sbd_lp_cb;
 185         sbd_lp->lp_alua_support = 1;
 186         sbd_lp->lp_proxy_msg = sbd_proxy_msg;
 187         sbd_zfs_ident = ldi_ident_from_anon();
 188 
 189         if (stmf_register_lu_provider(sbd_lp) != STMF_SUCCESS) {
 190                 (void) mod_remove(&modlinkage);
 191                 stmf_free(sbd_lp);
 192                 return (EINVAL);
 193         }
 194         mutex_init(&sbd_lock, NULL, MUTEX_DRIVER, NULL);
 195         rw_init(&sbd_global_prop_lock, NULL, RW_DRIVER, NULL);
 196         return (0);
 197 }
 198 
 199 int
 200 _fini(void)
 201 {
 202         int ret;
 203 
 204         /*
 205          * If we have registered lus, then make sure they are all offline
 206          * if so then deregister them. This should drop the sbd_lu_count
 207          * to zero.
 208          */
 209         if (sbd_lu_count) {
 210                 sbd_lu_t *slu;
 211 
 212                 /* See if all of them are offline */
 213                 mutex_enter(&sbd_lock);
 214                 for (slu = sbd_lu_list; slu != NULL; slu = slu->sl_next) {
 215                         if ((slu->sl_state != STMF_STATE_OFFLINE) ||
 216                             slu->sl_state_not_acked) {
 217                                 mutex_exit(&sbd_lock);
 218                                 return (EBUSY);
 219                         }
 220                 }
 221                 mutex_exit(&sbd_lock);
 222 
 223 #if 0
 224                 /* ok start deregistering them */
 225                 while (sbd_lu_list) {
 226                         sbd_store_t *sst = sbd_lu_list->sl_sst;
 227                         if (sst->sst_deregister_lu(sst) != STMF_SUCCESS)
 228                                 return (EBUSY);
 229                 }
 230 #endif
 231                 return (EBUSY);
 232         }
 233         if (stmf_deregister_lu_provider(sbd_lp) != STMF_SUCCESS)
 234                 return (EBUSY);
 235         ret = mod_remove(&modlinkage);
 236         if (ret != 0) {
 237                 (void) stmf_register_lu_provider(sbd_lp);
 238                 return (ret);
 239         }
 240         stmf_free(sbd_lp);
 241         mutex_destroy(&sbd_lock);
 242         rw_destroy(&sbd_global_prop_lock);
 243         ldi_ident_release(sbd_zfs_ident);
 244         return (0);
 245 }
 246 
 247 int
 248 _info(struct modinfo *modinfop)
 249 {
 250         return (mod_info(&modlinkage, modinfop));
 251 }
 252 
 253 /* ARGSUSED */
 254 static int
 255 sbd_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **result)
 256 {
 257         switch (cmd) {
 258         case DDI_INFO_DEVT2DEVINFO:
 259                 *result = sbd_dip;
 260                 break;
 261         case DDI_INFO_DEVT2INSTANCE:
 262                 *result = (void *)(uintptr_t)ddi_get_instance(sbd_dip);
 263                 break;
 264         default:
 265                 return (DDI_FAILURE);
 266         }
 267 
 268         return (DDI_SUCCESS);
 269 }
 270 
 271 static int
 272 sbd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
 273 {
 274         switch (cmd) {
 275         case DDI_ATTACH:
 276                 sbd_dip = dip;
 277 
 278                 if (ddi_create_minor_node(dip, "admin", S_IFCHR, 0,
 279                     DDI_NT_STMF_LP, 0) != DDI_SUCCESS) {
 280                         break;
 281                 }
 282                 ddi_report_dev(dip);
 283                 return (DDI_SUCCESS);
 284         }
 285 
 286         return (DDI_FAILURE);
 287 }
 288 
 289 static int
 290 sbd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
 291 {
 292         switch (cmd) {
 293         case DDI_DETACH:
 294                 ddi_remove_minor_node(dip, 0);
 295                 return (DDI_SUCCESS);
 296         }
 297 
 298         return (DDI_FAILURE);
 299 }
 300 
 301 /* ARGSUSED */
 302 static int
 303 sbd_open(dev_t *devp, int flag, int otype, cred_t *credp)
 304 {
 305         if (otype != OTYP_CHR)
 306                 return (EINVAL);
 307         return (0);
 308 }
 309 
 310 /* ARGSUSED */
 311 static int
 312 sbd_close(dev_t dev, int flag, int otype, cred_t *credp)
 313 {
 314         return (0);
 315 }
 316 
 317 /* ARGSUSED */
 318 static int
 319 stmf_sbd_ioctl(dev_t dev, int cmd, intptr_t data, int mode,
 320         cred_t *credp, int *rval)
 321 {
 322         stmf_iocdata_t          *iocd;
 323         void                    *ibuf   = NULL;
 324         void                    *obuf   = NULL;
 325         sbd_lu_t                *nsl;
 326         int                     i;
 327         int                     ret;
 328 
 329         if (drv_priv(credp) != 0) {
 330                 return (EPERM);
 331         }
 332 
 333         ret = stmf_copyin_iocdata(data, mode, &iocd, &ibuf, &obuf);
 334         if (ret)
 335                 return (ret);
 336         iocd->stmf_error = 0;
 337 
 338         switch (cmd) {
 339         case SBD_IOCTL_CREATE_AND_REGISTER_LU:
 340                 if (iocd->stmf_ibuf_size <
 341                     (sizeof (sbd_create_and_reg_lu_t) - 8)) {
 342                         ret = EFAULT;
 343                         break;
 344                 }
 345                 if ((iocd->stmf_obuf_size == 0) ||
 346                     (iocd->stmf_obuf_size > iocd->stmf_ibuf_size)) {
 347                         ret = EINVAL;
 348                         break;
 349                 }
 350                 ret = sbd_create_register_lu((sbd_create_and_reg_lu_t *)
 351                     ibuf, iocd->stmf_ibuf_size, &iocd->stmf_error);
 352                 bcopy(ibuf, obuf, iocd->stmf_obuf_size);
 353                 break;
 354         case SBD_IOCTL_SET_LU_STANDBY:
 355                 if (iocd->stmf_ibuf_size < sizeof (sbd_set_lu_standby_t)) {
 356                         ret = EFAULT;
 357                         break;
 358                 }
 359                 if (iocd->stmf_obuf_size) {
 360                         ret = EINVAL;
 361                         break;
 362                 }
 363                 ret = sbd_set_lu_standby((sbd_set_lu_standby_t *)ibuf,
 364                     &iocd->stmf_error);
 365                 break;
 366         case SBD_IOCTL_IMPORT_LU:
 367                 if (iocd->stmf_ibuf_size <
 368                     (sizeof (sbd_import_lu_t) - 8)) {
 369                         ret = EFAULT;
 370                         break;
 371                 }
 372                 if ((iocd->stmf_obuf_size == 0) ||
 373                     (iocd->stmf_obuf_size > iocd->stmf_ibuf_size)) {
 374                         ret = EINVAL;
 375                         break;
 376                 }
 377                 ret = sbd_import_lu((sbd_import_lu_t *)ibuf,
 378                     iocd->stmf_ibuf_size, &iocd->stmf_error, 0, NULL);
 379                 bcopy(ibuf, obuf, iocd->stmf_obuf_size);
 380                 break;
 381         case SBD_IOCTL_DELETE_LU:
 382                 if (iocd->stmf_ibuf_size < (sizeof (sbd_delete_lu_t) - 8)) {
 383                         ret = EFAULT;
 384                         break;
 385                 }
 386                 if (iocd->stmf_obuf_size) {
 387                         ret = EINVAL;
 388                         break;
 389                 }
 390                 ret = sbd_delete_lu((sbd_delete_lu_t *)ibuf,
 391                     iocd->stmf_ibuf_size, &iocd->stmf_error);
 392                 break;
 393         case SBD_IOCTL_MODIFY_LU:
 394                 if (iocd->stmf_ibuf_size < (sizeof (sbd_modify_lu_t) - 8)) {
 395                         ret = EFAULT;
 396                         break;
 397                 }
 398                 if (iocd->stmf_obuf_size) {
 399                         ret = EINVAL;
 400                         break;
 401                 }
 402                 ret = sbd_modify_lu((sbd_modify_lu_t *)ibuf,
 403                     iocd->stmf_ibuf_size, &iocd->stmf_error);
 404                 break;
 405         case SBD_IOCTL_SET_GLOBAL_LU:
 406                 if (iocd->stmf_ibuf_size < (sizeof (sbd_global_props_t) - 8)) {
 407                         ret = EFAULT;
 408                         break;
 409                 }
 410                 if (iocd->stmf_obuf_size) {
 411                         ret = EINVAL;
 412                         break;
 413                 }
 414                 ret = sbd_set_global_props((sbd_global_props_t *)ibuf,
 415                     iocd->stmf_ibuf_size, &iocd->stmf_error);
 416                 break;
 417         case SBD_IOCTL_GET_GLOBAL_LU:
 418                 if (iocd->stmf_ibuf_size) {
 419                         ret = EINVAL;
 420                         break;
 421                 }
 422                 if (iocd->stmf_obuf_size < sizeof (sbd_global_props_t)) {
 423                         ret = EINVAL;
 424                         break;
 425                 }
 426                 ret = sbd_get_global_props((sbd_global_props_t *)obuf,
 427                     iocd->stmf_obuf_size, &iocd->stmf_error);
 428                 break;
 429         case SBD_IOCTL_GET_LU_PROPS:
 430                 if (iocd->stmf_ibuf_size < (sizeof (sbd_lu_props_t) - 8)) {
 431                         ret = EFAULT;
 432                         break;
 433                 }
 434                 if (iocd->stmf_obuf_size < sizeof (sbd_lu_props_t)) {
 435                         ret = EINVAL;
 436                         break;
 437                 }
 438                 ret = sbd_get_lu_props((sbd_lu_props_t *)ibuf,
 439                     iocd->stmf_ibuf_size, (sbd_lu_props_t *)obuf,
 440                     iocd->stmf_obuf_size, &iocd->stmf_error);
 441                 break;
 442         case SBD_IOCTL_GET_LU_LIST:
 443                 mutex_enter(&sbd_lock);
 444                 iocd->stmf_obuf_max_nentries = sbd_lu_count;
 445                 iocd->stmf_obuf_nentries = min((iocd->stmf_obuf_size >> 4),
 446                     sbd_lu_count);
 447                 for (nsl = sbd_lu_list, i = 0; nsl &&
 448                     (i < iocd->stmf_obuf_nentries); i++, nsl = nsl->sl_next) {
 449                         bcopy(nsl->sl_device_id + 4,
 450                             &(((uint8_t *)obuf)[i << 4]), 16);
 451                 }
 452                 mutex_exit(&sbd_lock);
 453                 ret = 0;
 454                 iocd->stmf_error = 0;
 455                 break;
 456         case SBD_IOCTL_GET_UNMAP_PROPS:
 457                 if (iocd->stmf_ibuf_size < sizeof (sbd_unmap_props_t)) {
 458                         ret = EFAULT;
 459                         break;
 460                 }
 461                 if (iocd->stmf_obuf_size < sizeof (sbd_unmap_props_t)) {
 462                         ret = EINVAL;
 463                         break;
 464                 }
 465                 ret = sbd_get_unmap_props((sbd_unmap_props_t *)ibuf,
 466                     (sbd_unmap_props_t *)obuf, &iocd->stmf_error);
 467                 break;
 468         default:
 469                 ret = ENOTTY;
 470         }
 471 
 472         if (ret == 0) {
 473                 ret = stmf_copyout_iocdata(data, mode, iocd, obuf);
 474         } else if (iocd->stmf_error) {
 475                 (void) stmf_copyout_iocdata(data, mode, iocd, obuf);
 476         }
 477         if (obuf) {
 478                 kmem_free(obuf, iocd->stmf_obuf_size);
 479                 obuf = NULL;
 480         }
 481         if (ibuf) {
 482                 kmem_free(ibuf, iocd->stmf_ibuf_size);
 483                 ibuf = NULL;
 484         }
 485         kmem_free(iocd, sizeof (stmf_iocdata_t));
 486         return (ret);
 487 }
 488 
 489 /* ARGSUSED */
 490 void
 491 sbd_lp_cb(stmf_lu_provider_t *lp, int cmd, void *arg, uint32_t flags)
 492 {
 493         nvpair_t        *np;
 494         char            *s;
 495         sbd_import_lu_t *ilu;
 496         uint32_t        ilu_sz;
 497         uint32_t        struct_sz;
 498         uint32_t        err_ret;
 499         int             iret;
 500 
 501         if ((cmd != STMF_PROVIDER_DATA_UPDATED) || (arg == NULL)) {
 502                 return;
 503         }
 504 
 505         if ((flags & (STMF_PCB_STMF_ONLINING | STMF_PCB_PREG_COMPLETE)) == 0) {
 506                 return;
 507         }
 508 
 509         np = NULL;
 510         ilu_sz = 1024;
 511         ilu = (sbd_import_lu_t *)kmem_zalloc(ilu_sz, KM_SLEEP);
 512         while ((np = nvlist_next_nvpair((nvlist_t *)arg, np)) != NULL) {
 513                 if (nvpair_type(np) != DATA_TYPE_STRING) {
 514                         continue;
 515                 }
 516                 if (nvpair_value_string(np, &s) != 0) {
 517                         continue;
 518                 }
 519                 struct_sz = max(8, strlen(s) + 1);
 520                 struct_sz += sizeof (sbd_import_lu_t) - 8;
 521                 if (struct_sz > ilu_sz) {
 522                         kmem_free(ilu, ilu_sz);
 523                         ilu_sz = struct_sz + 32;
 524                         ilu = (sbd_import_lu_t *)kmem_zalloc(ilu_sz, KM_SLEEP);
 525                 }
 526                 ilu->ilu_struct_size = struct_sz;
 527                 (void) strcpy(ilu->ilu_meta_fname, s);
 528                 iret = sbd_import_lu(ilu, struct_sz, &err_ret, 0, NULL);
 529                 if (iret) {
 530                         stmf_trace(0, "sbd_lp_cb: import_lu failed, ret = %d, "
 531                             "err_ret = %d", iret, err_ret);
 532                 } else {
 533                         stmf_trace(0, "Imported the LU %s", nvpair_name(np));
 534                 }
 535         }
 536 
 537         if (ilu) {
 538                 kmem_free(ilu, ilu_sz);
 539                 ilu = NULL;
 540         }
 541 }
 542 
 543 sbd_status_t
 544 sbd_link_lu(sbd_lu_t *sl)
 545 {
 546         sbd_lu_t *nsl;
 547 
 548         mutex_enter(&sbd_lock);
 549         mutex_enter(&sl->sl_lock);
 550         ASSERT(sl->sl_trans_op != SL_OP_NONE);
 551 
 552         if (sl->sl_flags & SL_LINKED) {
 553                 mutex_exit(&sbd_lock);
 554                 mutex_exit(&sl->sl_lock);
 555                 return (SBD_ALREADY);
 556         }
 557         for (nsl = sbd_lu_list; nsl; nsl = nsl->sl_next) {
 558                 if (strcmp(nsl->sl_name, sl->sl_name) == 0)
 559                         break;
 560         }
 561         if (nsl) {
 562                 mutex_exit(&sbd_lock);
 563                 mutex_exit(&sl->sl_lock);
 564                 return (SBD_ALREADY);
 565         }
 566         sl->sl_next = sbd_lu_list;
 567         sbd_lu_list = sl;
 568         sl->sl_flags |= SL_LINKED;
 569         mutex_exit(&sbd_lock);
 570         mutex_exit(&sl->sl_lock);
 571         return (SBD_SUCCESS);
 572 }
 573 
 574 void
 575 sbd_unlink_lu(sbd_lu_t *sl)
 576 {
 577         sbd_lu_t **ppnsl;
 578 
 579         mutex_enter(&sbd_lock);
 580         mutex_enter(&sl->sl_lock);
 581         ASSERT(sl->sl_trans_op != SL_OP_NONE);
 582 
 583         ASSERT(sl->sl_flags & SL_LINKED);
 584         for (ppnsl = &sbd_lu_list; *ppnsl; ppnsl = &((*ppnsl)->sl_next)) {
 585                 if (*ppnsl == sl)
 586                         break;
 587         }
 588         ASSERT(*ppnsl);
 589         *ppnsl = (*ppnsl)->sl_next;
 590         sl->sl_flags &= ~SL_LINKED;
 591         mutex_exit(&sbd_lock);
 592         mutex_exit(&sl->sl_lock);
 593 }
 594 
 595 sbd_status_t
 596 sbd_find_and_lock_lu(uint8_t *guid, uint8_t *meta_name, uint8_t op,
 597     sbd_lu_t **ppsl)
 598 {
 599         sbd_lu_t *sl;
 600         int found = 0;
 601         sbd_status_t sret;
 602 
 603         mutex_enter(&sbd_lock);
 604         for (sl = sbd_lu_list; sl; sl = sl->sl_next) {
 605                 if (guid) {
 606                         found = bcmp(sl->sl_device_id + 4, guid, 16) == 0;
 607                 } else {
 608                         found = strcmp(sl->sl_name, (char *)meta_name) == 0;
 609                 }
 610                 if (found)
 611                         break;
 612         }
 613         if (!found) {
 614                 mutex_exit(&sbd_lock);
 615                 return (SBD_NOT_FOUND);
 616         }
 617         mutex_enter(&sl->sl_lock);
 618         if (sl->sl_trans_op == SL_OP_NONE) {
 619                 sl->sl_trans_op = op;
 620                 *ppsl = sl;
 621                 sret = SBD_SUCCESS;
 622         } else {
 623                 sret = SBD_BUSY;
 624         }
 625         mutex_exit(&sl->sl_lock);
 626         mutex_exit(&sbd_lock);
 627         return (sret);
 628 }
 629 
 630 sbd_status_t
 631 sbd_read_meta(sbd_lu_t *sl, uint64_t offset, uint64_t size, uint8_t *buf)
 632 {
 633         uint64_t        meta_align;
 634         uint64_t        starting_off;
 635         uint64_t        data_off;
 636         uint64_t        ending_off;
 637         uint64_t        io_size;
 638         uint8_t         *io_buf;
 639         vnode_t         *vp;
 640         sbd_status_t    ret;
 641         ssize_t         resid;
 642         int             vret;
 643 
 644         ASSERT(sl->sl_flags & SL_META_OPENED);
 645         if (sl->sl_flags & SL_SHARED_META) {
 646                 meta_align = (((uint64_t)1) << sl->sl_data_blocksize_shift) - 1;
 647                 vp = sl->sl_data_vp;
 648                 ASSERT(vp);
 649         } else {
 650                 meta_align = (((uint64_t)1) << sl->sl_meta_blocksize_shift) - 1;
 651                 if ((sl->sl_flags & SL_ZFS_META) == 0) {
 652                         vp = sl->sl_meta_vp;
 653                         ASSERT(vp);
 654                 }
 655         }
 656         starting_off = offset & ~(meta_align);
 657         data_off = offset & meta_align;
 658         ending_off = (offset + size + meta_align) & (~meta_align);
 659         if (ending_off > sl->sl_meta_size_used) {
 660                 bzero(buf, size);
 661                 if (starting_off >= sl->sl_meta_size_used) {
 662                         return (SBD_SUCCESS);
 663                 }
 664                 ending_off = (sl->sl_meta_size_used + meta_align) &
 665                     (~meta_align);
 666                 if (size > (ending_off - (starting_off + data_off))) {
 667                         size = ending_off - (starting_off + data_off);
 668                 }
 669         }
 670         io_size = ending_off - starting_off;
 671         io_buf = (uint8_t *)kmem_zalloc(io_size, KM_SLEEP);
 672         ASSERT((starting_off + io_size) <= sl->sl_total_meta_size);
 673 
 674         /*
 675          * Don't proceed if the device has been closed
 676          * This can occur on an access state change to standby or
 677          * a delete. The writer lock is acquired before closing the
 678          * lu. If importing, reading the metadata is valid, hence
 679          * the check on SL_OP_IMPORT_LU.
 680          */
 681         rw_enter(&sl->sl_access_state_lock, RW_READER);
 682         if ((sl->sl_flags & SL_MEDIA_LOADED) == 0 &&
 683             sl->sl_trans_op != SL_OP_IMPORT_LU) {
 684                 rw_exit(&sl->sl_access_state_lock);
 685                 ret = SBD_FILEIO_FAILURE;
 686                 goto sbd_read_meta_failure;
 687         }
 688         if (sl->sl_flags & SL_ZFS_META) {
 689                 if ((ret = sbd_read_zfs_meta(sl, io_buf, io_size,
 690                     starting_off)) != SBD_SUCCESS) {
 691                         rw_exit(&sl->sl_access_state_lock);
 692                         goto sbd_read_meta_failure;
 693                 }
 694         } else {
 695                 vret = vn_rdwr(UIO_READ, vp, (caddr_t)io_buf, (ssize_t)io_size,
 696                     (offset_t)starting_off, UIO_SYSSPACE, FRSYNC,
 697                     RLIM64_INFINITY, CRED(), &resid);
 698 
 699                 if (vret || resid) {
 700                         ret = SBD_FILEIO_FAILURE | vret;
 701                         rw_exit(&sl->sl_access_state_lock);
 702                         goto sbd_read_meta_failure;
 703                 }
 704         }
 705         rw_exit(&sl->sl_access_state_lock);
 706 
 707         bcopy(io_buf + data_off, buf, size);
 708         ret = SBD_SUCCESS;
 709 
 710 sbd_read_meta_failure:
 711         kmem_free(io_buf, io_size);
 712         return (ret);
 713 }
 714 
 715 sbd_status_t
 716 sbd_write_meta(sbd_lu_t *sl, uint64_t offset, uint64_t size, uint8_t *buf)
 717 {
 718         uint64_t        meta_align;
 719         uint64_t        starting_off;
 720         uint64_t        data_off;
 721         uint64_t        ending_off;
 722         uint64_t        io_size;
 723         uint8_t         *io_buf;
 724         vnode_t         *vp;
 725         sbd_status_t    ret;
 726         ssize_t         resid;
 727         int             vret;
 728 
 729         ASSERT(sl->sl_flags & SL_META_OPENED);
 730         if (sl->sl_flags & SL_SHARED_META) {
 731                 meta_align = (((uint64_t)1) << sl->sl_data_blocksize_shift) - 1;
 732                 vp = sl->sl_data_vp;
 733                 ASSERT(vp);
 734         } else {
 735                 meta_align = (((uint64_t)1) << sl->sl_meta_blocksize_shift) - 1;
 736                 if ((sl->sl_flags & SL_ZFS_META) == 0) {
 737                         vp = sl->sl_meta_vp;
 738                         ASSERT(vp);
 739                 }
 740         }
 741         starting_off = offset & ~(meta_align);
 742         data_off = offset & meta_align;
 743         ending_off = (offset + size + meta_align) & (~meta_align);
 744         io_size = ending_off - starting_off;
 745         io_buf = (uint8_t *)kmem_zalloc(io_size, KM_SLEEP);
 746         ret = sbd_read_meta(sl, starting_off, io_size, io_buf);
 747         if (ret != SBD_SUCCESS) {
 748                 goto sbd_write_meta_failure;
 749         }
 750         bcopy(buf, io_buf + data_off, size);
 751         /*
 752          * Don't proceed if the device has been closed
 753          * This can occur on an access state change to standby or
 754          * a delete. The writer lock is acquired before closing the
 755          * lu. If importing, reading the metadata is valid, hence
 756          * the check on SL_OP_IMPORT_LU.
 757          */
 758         rw_enter(&sl->sl_access_state_lock, RW_READER);
 759         if ((sl->sl_flags & SL_MEDIA_LOADED) == 0 &&
 760             sl->sl_trans_op != SL_OP_IMPORT_LU) {
 761                 rw_exit(&sl->sl_access_state_lock);
 762                 ret = SBD_FILEIO_FAILURE;
 763                 goto sbd_write_meta_failure;
 764         }
 765         if (sl->sl_flags & SL_ZFS_META) {
 766                 if ((ret = sbd_write_zfs_meta(sl, io_buf, io_size,
 767                     starting_off)) != SBD_SUCCESS) {
 768                         rw_exit(&sl->sl_access_state_lock);
 769                         goto sbd_write_meta_failure;
 770                 }
 771         } else {
 772                 vret = vn_rdwr(UIO_WRITE, vp, (caddr_t)io_buf, (ssize_t)io_size,
 773                     (offset_t)starting_off, UIO_SYSSPACE, FDSYNC,
 774                     RLIM64_INFINITY, CRED(), &resid);
 775 
 776                 if (vret || resid) {
 777                         ret = SBD_FILEIO_FAILURE | vret;
 778                         rw_exit(&sl->sl_access_state_lock);
 779                         goto sbd_write_meta_failure;
 780                 }
 781         }
 782         rw_exit(&sl->sl_access_state_lock);
 783 
 784         ret = SBD_SUCCESS;
 785 
 786 sbd_write_meta_failure:
 787         kmem_free(io_buf, io_size);
 788         return (ret);
 789 }
 790 
 791 uint8_t
 792 sbd_calc_sum(uint8_t *buf, int size)
 793 {
 794         uint8_t s = 0;
 795 
 796         while (size > 0)
 797                 s += buf[--size];
 798 
 799         return (s);
 800 }
 801 
 802 uint8_t
 803 sbd_calc_section_sum(sm_section_hdr_t *sm, uint32_t sz)
 804 {
 805         uint8_t s, o;
 806 
 807         o = sm->sms_chksum;
 808         sm->sms_chksum = 0;
 809         s = sbd_calc_sum((uint8_t *)sm, sz);
 810         sm->sms_chksum = o;
 811 
 812         return (s);
 813 }
 814 
 815 uint32_t
 816 sbd_strlen(char *str, uint32_t maxlen)
 817 {
 818         uint32_t i;
 819 
 820         for (i = 0; i < maxlen; i++) {
 821                 if (str[i] == 0)
 822                         return (i);
 823         }
 824         return (i);
 825 }
 826 
 827 void
 828 sbd_swap_meta_start(sbd_meta_start_t *sm)
 829 {
 830         if (sm->sm_magic == SBD_MAGIC)
 831                 return;
 832         sm->sm_magic         = BSWAP_64(sm->sm_magic);
 833         sm->sm_meta_size     = BSWAP_64(sm->sm_meta_size);
 834         sm->sm_meta_size_used        = BSWAP_64(sm->sm_meta_size_used);
 835         sm->sm_ver_major     = BSWAP_16(sm->sm_ver_major);
 836         sm->sm_ver_minor     = BSWAP_16(sm->sm_ver_minor);
 837         sm->sm_ver_subminor  = BSWAP_16(sm->sm_ver_subminor);
 838 }
 839 
 840 void
 841 sbd_swap_section_hdr(sm_section_hdr_t *sm)
 842 {
 843         if (sm->sms_data_order == SMS_DATA_ORDER)
 844                 return;
 845         sm->sms_offset               = BSWAP_64(sm->sms_offset);
 846         sm->sms_size         = BSWAP_32(sm->sms_size);
 847         sm->sms_id           = BSWAP_16(sm->sms_id);
 848         sm->sms_chksum               += SMS_DATA_ORDER - sm->sms_data_order;
 849         sm->sms_data_order   = SMS_DATA_ORDER;
 850 }
 851 
 852 void
 853 sbd_swap_lu_info_1_0(sbd_lu_info_1_0_t *sli)
 854 {
 855         sbd_swap_section_hdr(&sli->sli_sms_header);
 856         if (sli->sli_data_order == SMS_DATA_ORDER)
 857                 return;
 858         sli->sli_sms_header.sms_chksum       += SMS_DATA_ORDER - sli->sli_data_order;
 859         sli->sli_data_order          = SMS_DATA_ORDER;
 860         sli->sli_total_store_size    = BSWAP_64(sli->sli_total_store_size);
 861         sli->sli_total_meta_size     = BSWAP_64(sli->sli_total_meta_size);
 862         sli->sli_lu_data_offset              = BSWAP_64(sli->sli_lu_data_offset);
 863         sli->sli_lu_data_size                = BSWAP_64(sli->sli_lu_data_size);
 864         sli->sli_flags                       = BSWAP_32(sli->sli_flags);
 865         sli->sli_blocksize           = BSWAP_16(sli->sli_blocksize);
 866 }
 867 
 868 void
 869 sbd_swap_lu_info_1_1(sbd_lu_info_1_1_t *sli)
 870 {
 871         sbd_swap_section_hdr(&sli->sli_sms_header);
 872         if (sli->sli_data_order == SMS_DATA_ORDER)
 873                 return;
 874         sli->sli_sms_header.sms_chksum       += SMS_DATA_ORDER - sli->sli_data_order;
 875         sli->sli_data_order          = SMS_DATA_ORDER;
 876         sli->sli_flags                       = BSWAP_32(sli->sli_flags);
 877         sli->sli_lu_size             = BSWAP_64(sli->sli_lu_size);
 878         sli->sli_meta_fname_offset   = BSWAP_64(sli->sli_meta_fname_offset);
 879         sli->sli_data_fname_offset   = BSWAP_64(sli->sli_data_fname_offset);
 880         sli->sli_serial_offset               = BSWAP_64(sli->sli_serial_offset);
 881         sli->sli_alias_offset                = BSWAP_64(sli->sli_alias_offset);
 882         sli->sli_mgmt_url_offset     = BSWAP_64(sli->sli_mgmt_url_offset);
 883 }
 884 
 885 sbd_status_t
 886 sbd_load_section_hdr(sbd_lu_t *sl, sm_section_hdr_t *sms)
 887 {
 888         sm_section_hdr_t        h;
 889         uint64_t                st;
 890         sbd_status_t            ret;
 891 
 892         for (st = sl->sl_meta_offset + sizeof (sbd_meta_start_t);
 893             st < sl->sl_meta_size_used; st += h.sms_size) {
 894                 if ((ret = sbd_read_meta(sl, st, sizeof (sm_section_hdr_t),
 895                     (uint8_t *)&h)) != SBD_SUCCESS) {
 896                         return (ret);
 897                 }
 898                 if (h.sms_data_order != SMS_DATA_ORDER) {
 899                         sbd_swap_section_hdr(&h);
 900                 }
 901                 if ((h.sms_data_order != SMS_DATA_ORDER) ||
 902                     (h.sms_offset != st) || (h.sms_size < sizeof (h)) ||
 903                     ((st + h.sms_size) > sl->sl_meta_size_used)) {
 904                         return (SBD_META_CORRUPTED);
 905                 }
 906                 if (h.sms_id == sms->sms_id) {
 907                         bcopy(&h, sms, sizeof (h));
 908                         return (SBD_SUCCESS);
 909                 }
 910         }
 911 
 912         return (SBD_NOT_FOUND);
 913 }
 914 
 915 sbd_status_t
 916 sbd_load_meta_start(sbd_lu_t *sl)
 917 {
 918         sbd_meta_start_t *sm;
 919         sbd_status_t ret;
 920 
 921         /* Fake meta params initially */
 922         sl->sl_total_meta_size = (uint64_t)-1;
 923         sl->sl_meta_size_used = sl->sl_meta_offset + sizeof (sbd_meta_start_t);
 924 
 925         sm = kmem_zalloc(sizeof (*sm), KM_SLEEP);
 926         ret = sbd_read_meta(sl, sl->sl_meta_offset, sizeof (*sm),
 927             (uint8_t *)sm);
 928         if (ret != SBD_SUCCESS) {
 929                 goto load_meta_start_failed;
 930         }
 931 
 932         if (sm->sm_magic != SBD_MAGIC) {
 933                 sbd_swap_meta_start(sm);
 934         }
 935 
 936         if ((sm->sm_magic != SBD_MAGIC) || (sbd_calc_sum((uint8_t *)sm,
 937             sizeof (*sm) - 1) != sm->sm_chksum)) {
 938                 ret = SBD_META_CORRUPTED;
 939                 goto load_meta_start_failed;
 940         }
 941 
 942         if (sm->sm_ver_major != SBD_VER_MAJOR) {
 943                 ret = SBD_NOT_SUPPORTED;
 944                 goto load_meta_start_failed;
 945         }
 946 
 947         sl->sl_total_meta_size = sm->sm_meta_size;
 948         sl->sl_meta_size_used = sm->sm_meta_size_used;
 949         ret = SBD_SUCCESS;
 950 
 951 load_meta_start_failed:
 952         kmem_free(sm, sizeof (*sm));
 953         return (ret);
 954 }
 955 
 956 sbd_status_t
 957 sbd_write_meta_start(sbd_lu_t *sl, uint64_t meta_size, uint64_t meta_size_used)
 958 {
 959         sbd_meta_start_t *sm;
 960         sbd_status_t ret;
 961 
 962         sm = (sbd_meta_start_t *)kmem_zalloc(sizeof (sbd_meta_start_t),
 963             KM_SLEEP);
 964 
 965         sm->sm_magic = SBD_MAGIC;
 966         sm->sm_meta_size = meta_size;
 967         sm->sm_meta_size_used = meta_size_used;
 968         sm->sm_ver_major = SBD_VER_MAJOR;
 969         sm->sm_ver_minor = SBD_VER_MINOR;
 970         sm->sm_ver_subminor = SBD_VER_SUBMINOR;
 971         sm->sm_chksum = sbd_calc_sum((uint8_t *)sm, sizeof (*sm) - 1);
 972 
 973         ret = sbd_write_meta(sl, sl->sl_meta_offset, sizeof (*sm),
 974             (uint8_t *)sm);
 975         kmem_free(sm, sizeof (*sm));
 976 
 977         return (ret);
 978 }
 979 
 980 sbd_status_t
 981 sbd_read_meta_section(sbd_lu_t *sl, sm_section_hdr_t **ppsms, uint16_t sms_id)
 982 {
 983         sbd_status_t ret;
 984         sm_section_hdr_t sms;
 985         int alloced = 0;
 986 
 987         mutex_enter(&sl->sl_metadata_lock);
 988         if (((*ppsms) == NULL) || ((*ppsms)->sms_offset == 0)) {
 989                 bzero(&sms, sizeof (sm_section_hdr_t));
 990                 sms.sms_id = sms_id;
 991                 if ((ret = sbd_load_section_hdr(sl, &sms)) != SBD_SUCCESS) {
 992                         mutex_exit(&sl->sl_metadata_lock);
 993                         return (ret);
 994                 } else {
 995                         if ((*ppsms) == NULL) {
 996                                 *ppsms = (sm_section_hdr_t *)kmem_zalloc(
 997                                     sms.sms_size, KM_SLEEP);
 998                                 alloced = 1;
 999                         }
1000                         bcopy(&sms, *ppsms, sizeof (sm_section_hdr_t));
1001                 }
1002         }
1003 
1004         ret = sbd_read_meta(sl, (*ppsms)->sms_offset, (*ppsms)->sms_size,
1005             (uint8_t *)(*ppsms));
1006         if (ret == SBD_SUCCESS) {
1007                 uint8_t s;
1008                 if ((*ppsms)->sms_data_order != SMS_DATA_ORDER)
1009                         sbd_swap_section_hdr(*ppsms);
1010                 if ((*ppsms)->sms_id != SMS_ID_UNUSED) {
1011                         s = sbd_calc_section_sum(*ppsms, (*ppsms)->sms_size);
1012                         if (s != (*ppsms)->sms_chksum)
1013                                 ret = SBD_META_CORRUPTED;
1014                 }
1015         }
1016         mutex_exit(&sl->sl_metadata_lock);
1017 
1018         if ((ret != SBD_SUCCESS) && alloced)
1019                 kmem_free(*ppsms, sms.sms_size);
1020         return (ret);
1021 }
1022 
1023 sbd_status_t
1024 sbd_load_section_hdr_unbuffered(sbd_lu_t *sl, sm_section_hdr_t *sms)
1025 {
1026         sbd_status_t    ret;
1027 
1028         /*
1029          * Bypass buffering and re-read the meta data from permanent storage.
1030          */
1031         if (sl->sl_flags & SL_ZFS_META) {
1032                 if ((ret = sbd_open_zfs_meta(sl)) != SBD_SUCCESS) {
1033                         return (ret);
1034                 }
1035         }
1036         /* Re-get the meta sizes into sl */
1037         if ((ret = sbd_load_meta_start(sl)) != SBD_SUCCESS) {
1038                 return (ret);
1039         }
1040         return (sbd_load_section_hdr(sl, sms));
1041 }
1042 
1043 sbd_status_t
1044 sbd_write_meta_section(sbd_lu_t *sl, sm_section_hdr_t *sms)
1045 {
1046         sm_section_hdr_t t;
1047         uint64_t off, s;
1048         uint64_t unused_start;
1049         sbd_status_t ret;
1050         sbd_status_t write_meta_ret = SBD_SUCCESS;
1051         uint8_t *cb;
1052         int meta_size_changed = 0;
1053         sm_section_hdr_t sms_before_unused = {0};
1054 
1055         mutex_enter(&sl->sl_metadata_lock);
1056 write_meta_section_again:
1057         if (sms->sms_offset) {
1058                 /*
1059                  * If the section already exists and the size is the
1060                  * same as this new data then overwrite in place. If
1061                  * the sizes are different then mark the existing as
1062                  * unused and look for free space.
1063                  */
1064                 ret = sbd_read_meta(sl, sms->sms_offset, sizeof (t),
1065                     (uint8_t *)&t);
1066                 if (ret != SBD_SUCCESS) {
1067                         mutex_exit(&sl->sl_metadata_lock);
1068                         return (ret);
1069                 }
1070                 if (t.sms_data_order != SMS_DATA_ORDER) {
1071                         sbd_swap_section_hdr(&t);
1072                 }
1073                 if (t.sms_id != sms->sms_id) {
1074                         mutex_exit(&sl->sl_metadata_lock);
1075                         return (SBD_INVALID_ARG);
1076                 }
1077                 if (t.sms_size == sms->sms_size) {
1078                         ret = sbd_write_meta(sl, sms->sms_offset,
1079                             sms->sms_size, (uint8_t *)sms);
1080                         mutex_exit(&sl->sl_metadata_lock);
1081                         return (ret);
1082                 }
1083                 sms_before_unused = t;
1084 
1085                 t.sms_id = SMS_ID_UNUSED;
1086                 /*
1087                  * For unused sections we only use chksum of the header. for
1088                  * all other sections, the chksum is for the entire section.
1089                  */
1090                 t.sms_chksum = sbd_calc_section_sum(&t, sizeof (t));
1091                 ret = sbd_write_meta(sl, t.sms_offset, sizeof (t),
1092                     (uint8_t *)&t);
1093                 if (ret != SBD_SUCCESS) {
1094                         mutex_exit(&sl->sl_metadata_lock);
1095                         return (ret);
1096                 }
1097                 sms->sms_offset = 0;
1098         } else {
1099                 /* Section location is unknown, search for it. */
1100                 t.sms_id = sms->sms_id;
1101                 t.sms_data_order = SMS_DATA_ORDER;
1102                 ret = sbd_load_section_hdr(sl, &t);
1103                 if (ret == SBD_SUCCESS) {
1104                         sms->sms_offset = t.sms_offset;
1105                         sms->sms_chksum =
1106                             sbd_calc_section_sum(sms, sms->sms_size);
1107                         goto write_meta_section_again;
1108                 } else if (ret != SBD_NOT_FOUND) {
1109                         mutex_exit(&sl->sl_metadata_lock);
1110                         return (ret);
1111                 }
1112         }
1113 
1114         /*
1115          * At this point we know that section does not already exist.
1116          * Find space large enough to hold the section or grow meta if
1117          * possible.
1118          */
1119         unused_start = 0;
1120         s = 0;  /* size of space found */
1121 
1122         /*
1123          * Search all sections for unused space of sufficient size.
1124          * The first one found is taken. Contiguous unused sections
1125          * will be combined.
1126          */
1127         for (off = sl->sl_meta_offset + sizeof (sbd_meta_start_t);
1128             off < sl->sl_meta_size_used; off += t.sms_size) {
1129                 ret = sbd_read_meta(sl, off, sizeof (t), (uint8_t *)&t);
1130                 if (ret != SBD_SUCCESS) {
1131                         mutex_exit(&sl->sl_metadata_lock);
1132                         return (ret);
1133                 }
1134                 if (t.sms_data_order != SMS_DATA_ORDER)
1135                         sbd_swap_section_hdr(&t);
1136                 if (t.sms_size == 0) {
1137                         mutex_exit(&sl->sl_metadata_lock);
1138                         return (SBD_META_CORRUPTED);
1139                 }
1140                 if (t.sms_id == SMS_ID_UNUSED) {
1141                         if (unused_start == 0)
1142                                 unused_start = off;
1143                         /*
1144                          * Calculate size of the unused space, break out
1145                          * if it satisfies the requirement.
1146                          */
1147                         s = t.sms_size - unused_start + off;
1148                         if ((s == sms->sms_size) || (s >= (sms->sms_size +
1149                             sizeof (t)))) {
1150                                 break;
1151                         } else {
1152                                 s = 0;
1153                         }
1154                 } else {
1155                         unused_start = 0;
1156                 }
1157         }
1158 
1159         off = (unused_start == 0) ? sl->sl_meta_size_used : unused_start;
1160         /*
1161          * If none found, how much room is at the end?
1162          * See if the data can be expanded.
1163          */
1164         if (s == 0) {
1165                 s = sl->sl_total_meta_size - off;
1166                 if (s >= sms->sms_size || !(sl->sl_flags & SL_SHARED_META)) {
1167                         s = sms->sms_size;
1168                         meta_size_changed = 1;
1169                 } else {
1170                         s = 0;
1171                 }
1172         }
1173 
1174         if (s == 0) {
1175                 mutex_exit(&sl->sl_metadata_lock);
1176                 return (SBD_ALLOC_FAILURE);
1177         }
1178 
1179         sms->sms_offset = off;
1180         sms->sms_chksum = sbd_calc_section_sum(sms, sms->sms_size);
1181         /*
1182          * Since we may have to write more than one section (current +
1183          * any unused), use a combined buffer.
1184          */
1185         cb = kmem_zalloc(s, KM_SLEEP);
1186         bcopy(sms, cb, sms->sms_size);
1187         if (s > sms->sms_size) {
1188                 t.sms_offset = off + sms->sms_size;
1189                 t.sms_size = s - sms->sms_size;
1190                 t.sms_id = SMS_ID_UNUSED;
1191                 t.sms_data_order = SMS_DATA_ORDER;
1192                 t.sms_chksum = sbd_calc_section_sum(&t, sizeof (t));
1193                 bcopy(&t, cb + sms->sms_size, sizeof (t));
1194         }
1195         /*
1196          * Two write events & statuses take place. Failure writing the
1197          * meta section takes precedence, can possibly be rolled back,
1198          * & gets reported. Else return status from writing the meta start.
1199          */
1200         ret = SBD_SUCCESS; /* Set a default, it's not always loaded below. */
1201         if (meta_size_changed) {
1202                 uint64_t old_meta_size;
1203                 uint64_t old_sz_used = sl->sl_meta_size_used; /* save a copy */
1204                 old_meta_size = sl->sl_total_meta_size; /* save a copy */
1205 
1206                 write_meta_ret = sbd_write_meta(sl, off, s, cb);
1207                 if (write_meta_ret == SBD_SUCCESS) {
1208                         sl->sl_meta_size_used = off + s;
1209                         if (sl->sl_total_meta_size < sl->sl_meta_size_used) {
1210                                 uint64_t meta_align =
1211                                     (((uint64_t)1) <<
1212                                     sl->sl_meta_blocksize_shift) - 1;
1213                                 sl->sl_total_meta_size =
1214                                     (sl->sl_meta_size_used + meta_align) &
1215                                     (~meta_align);
1216                         }
1217                         ret = sbd_write_meta_start(sl, sl->sl_total_meta_size,
1218                             sl->sl_meta_size_used);
1219                         if (ret != SBD_SUCCESS) {
1220                                 sl->sl_meta_size_used = old_sz_used;
1221                                 sl->sl_total_meta_size = old_meta_size;
1222                         }
1223                 } else {
1224                         sl->sl_meta_size_used = old_sz_used;
1225                         sl->sl_total_meta_size = old_meta_size;
1226                 }
1227         } else {
1228                 write_meta_ret = sbd_write_meta(sl, off, s, cb);
1229         }
1230         if ((write_meta_ret != SBD_SUCCESS) &&
1231             (sms_before_unused.sms_offset != 0)) {
1232                 sm_section_hdr_t new_sms;
1233                 sm_section_hdr_t *unused_sms;
1234                 /*
1235                  * On failure writing the meta section attempt to undo
1236                  * the change to unused.
1237                  * Re-read the meta data from permanent storage.
1238                  * The section id can't exist for undo to be possible.
1239                  * Read what should be the entire old section data and
1240                  * insure the old data's still present by validating
1241                  * against it's old checksum.
1242                  */
1243                 new_sms.sms_id = sms->sms_id;
1244                 new_sms.sms_data_order = SMS_DATA_ORDER;
1245                 if (sbd_load_section_hdr_unbuffered(sl, &new_sms) !=
1246                     SBD_NOT_FOUND) {
1247                         goto done;
1248                 }
1249                 unused_sms = kmem_zalloc(sms_before_unused.sms_size, KM_SLEEP);
1250                 if (sbd_read_meta(sl, sms_before_unused.sms_offset,
1251                     sms_before_unused.sms_size,
1252                     (uint8_t *)unused_sms) != SBD_SUCCESS) {
1253                         goto done;
1254                 }
1255                 if (unused_sms->sms_data_order != SMS_DATA_ORDER) {
1256                         sbd_swap_section_hdr(unused_sms);
1257                 }
1258                 if (unused_sms->sms_id != SMS_ID_UNUSED) {
1259                         goto done;
1260                 }
1261                 if (unused_sms->sms_offset != sms_before_unused.sms_offset) {
1262                         goto done;
1263                 }
1264                 if (unused_sms->sms_size != sms_before_unused.sms_size) {
1265                         goto done;
1266                 }
1267                 unused_sms->sms_id = sms_before_unused.sms_id;
1268                 if (sbd_calc_section_sum(unused_sms,
1269                     sizeof (sm_section_hdr_t)) !=
1270                     sbd_calc_section_sum(&sms_before_unused,
1271                     sizeof (sm_section_hdr_t))) {
1272                         goto done;
1273                 }
1274                 unused_sms->sms_chksum =
1275                     sbd_calc_section_sum(unused_sms, unused_sms->sms_size);
1276                 if (unused_sms->sms_chksum != sms_before_unused.sms_chksum) {
1277                         goto done;
1278                 }
1279                 (void) sbd_write_meta(sl, unused_sms->sms_offset,
1280                     sizeof (sm_section_hdr_t), (uint8_t *)unused_sms);
1281         }
1282 done:
1283         mutex_exit(&sl->sl_metadata_lock);
1284         kmem_free(cb, s);
1285         if (write_meta_ret != SBD_SUCCESS) {
1286                 return (write_meta_ret);
1287         }
1288         return (ret);
1289 }
1290 
1291 sbd_status_t
1292 sbd_write_lu_info(sbd_lu_t *sl)
1293 {
1294         sbd_lu_info_1_1_t *sli;
1295         int s;
1296         uint8_t *p;
1297         char *zvol_name = NULL;
1298         sbd_status_t ret;
1299 
1300         mutex_enter(&sl->sl_lock);
1301 
1302         s = sl->sl_serial_no_size;
1303         if ((sl->sl_flags & (SL_SHARED_META | SL_ZFS_META)) == 0) {
1304                 if (sl->sl_data_filename) {
1305                         s += strlen(sl->sl_data_filename) + 1;
1306                 }
1307         }
1308         if (sl->sl_flags & SL_ZFS_META) {
1309                 zvol_name = sbd_get_zvol_name(sl);
1310                 s += strlen(zvol_name) + 1;
1311         }
1312         if (sl->sl_alias) {
1313                 s += strlen(sl->sl_alias) + 1;
1314         }
1315         if (sl->sl_mgmt_url) {
1316                 s += strlen(sl->sl_mgmt_url) + 1;
1317         }
1318         sli = (sbd_lu_info_1_1_t *)kmem_zalloc(sizeof (*sli) + s, KM_SLEEP);
1319         p = sli->sli_buf;
1320         if ((sl->sl_flags & (SL_SHARED_META | SL_ZFS_META)) == 0) {
1321                 sli->sli_flags |= SLI_SEPARATE_META;
1322                 (void) strcpy((char *)p, sl->sl_data_filename);
1323                 sli->sli_data_fname_offset =
1324                     (uintptr_t)p - (uintptr_t)sli->sli_buf;
1325                 sli->sli_flags |= SLI_DATA_FNAME_VALID;
1326                 p += strlen(sl->sl_data_filename) + 1;
1327         }
1328         if (sl->sl_flags & SL_ZFS_META) {
1329                 (void) strcpy((char *)p, zvol_name);
1330                 sli->sli_meta_fname_offset =
1331                     (uintptr_t)p - (uintptr_t)sli->sli_buf;
1332                 sli->sli_flags |= SLI_META_FNAME_VALID | SLI_ZFS_META;
1333                 p += strlen(zvol_name) + 1;
1334                 kmem_free(zvol_name, strlen(zvol_name) + 1);
1335                 zvol_name = NULL;
1336         }
1337         if (sl->sl_alias) {
1338                 (void) strcpy((char *)p, sl->sl_alias);
1339                 sli->sli_alias_offset =
1340                     (uintptr_t)p - (uintptr_t)sli->sli_buf;
1341                 sli->sli_flags |= SLI_ALIAS_VALID;
1342                 p += strlen(sl->sl_alias) + 1;
1343         }
1344         if (sl->sl_mgmt_url) {
1345                 (void) strcpy((char *)p, sl->sl_mgmt_url);
1346                 sli->sli_mgmt_url_offset =
1347                     (uintptr_t)p - (uintptr_t)sli->sli_buf;
1348                 sli->sli_flags |= SLI_MGMT_URL_VALID;
1349                 p += strlen(sl->sl_mgmt_url) + 1;
1350         }
1351         if (sl->sl_flags & SL_WRITE_PROTECTED) {
1352                 sli->sli_flags |= SLI_WRITE_PROTECTED;
1353         }
1354         if (sl->sl_flags & SL_SAVED_WRITE_CACHE_DISABLE) {
1355                 sli->sli_flags |= SLI_WRITEBACK_CACHE_DISABLE;
1356         }
1357         if (sl->sl_flags & SL_VID_VALID) {
1358                 bcopy(sl->sl_vendor_id, sli->sli_vid, 8);
1359                 sli->sli_flags |= SLI_VID_VALID;
1360         }
1361         if (sl->sl_flags & SL_PID_VALID) {
1362                 bcopy(sl->sl_product_id, sli->sli_pid, 16);
1363                 sli->sli_flags |= SLI_PID_VALID;
1364         }
1365         if (sl->sl_flags & SL_REV_VALID) {
1366                 bcopy(sl->sl_revision, sli->sli_rev, 4);
1367                 sli->sli_flags |= SLI_REV_VALID;
1368         }
1369         if (sl->sl_serial_no_size) {
1370                 bcopy(sl->sl_serial_no, p, sl->sl_serial_no_size);
1371                 sli->sli_serial_size = sl->sl_serial_no_size;
1372                 sli->sli_serial_offset =
1373                     (uintptr_t)p - (uintptr_t)sli->sli_buf;
1374                 sli->sli_flags |= SLI_SERIAL_VALID;
1375                 p += sli->sli_serial_size;
1376         }
1377         sli->sli_lu_size = sl->sl_lu_size;
1378         sli->sli_data_blocksize_shift = sl->sl_data_blocksize_shift;
1379         sli->sli_data_order = SMS_DATA_ORDER;
1380         bcopy(sl->sl_device_id, sli->sli_device_id, 20);
1381 
1382         sli->sli_sms_header.sms_size = sizeof (*sli) + s;
1383         sli->sli_sms_header.sms_id = SMS_ID_LU_INFO_1_1;
1384         sli->sli_sms_header.sms_data_order = SMS_DATA_ORDER;
1385 
1386         mutex_exit(&sl->sl_lock);
1387         ret = sbd_write_meta_section(sl, (sm_section_hdr_t *)sli);
1388         kmem_free(sli, sizeof (*sli) + s);
1389         return (ret);
1390 }
1391 
1392 /*
1393  * Will scribble SL_UNMAP_ENABLED into sl_flags if we succeed.
1394  */
1395 static void
1396 do_unmap_setup(sbd_lu_t *sl)
1397 {
1398         ASSERT((sl->sl_flags & SL_UNMAP_ENABLED) == 0);
1399 
1400         if ((sl->sl_flags & SL_ZFS_META) == 0)
1401                 return; /* No UNMAP for you. */
1402 
1403         sl->sl_flags |= SL_UNMAP_ENABLED;
1404 }
1405 
1406 int
1407 sbd_populate_and_register_lu(sbd_lu_t *sl, uint32_t *err_ret)
1408 {
1409         stmf_lu_t *lu = sl->sl_lu;
1410         stmf_status_t ret;
1411 
1412         do_unmap_setup(sl);
1413 
1414         lu->lu_id = (scsi_devid_desc_t *)sl->sl_device_id;
1415         if (sl->sl_alias) {
1416                 lu->lu_alias = sl->sl_alias;
1417         } else {
1418                 lu->lu_alias = sl->sl_name;
1419         }
1420         if (sl->sl_access_state == SBD_LU_STANDBY) {
1421                 /* call set access state */
1422                 ret = stmf_set_lu_access(lu, STMF_LU_STANDBY);
1423                 if (ret != STMF_SUCCESS) {
1424                         *err_ret = SBD_RET_ACCESS_STATE_FAILED;
1425                         return (EIO);
1426                 }
1427         }
1428         /* set proxy_reg_cb_arg to meta filename */
1429         if (sl->sl_meta_filename) {
1430                 lu->lu_proxy_reg_arg = sl->sl_meta_filename;
1431                 lu->lu_proxy_reg_arg_len = strlen(sl->sl_meta_filename) + 1;
1432         } else {
1433                 lu->lu_proxy_reg_arg = sl->sl_data_filename;
1434                 lu->lu_proxy_reg_arg_len = strlen(sl->sl_data_filename) + 1;
1435         }
1436         lu->lu_lp = sbd_lp;
1437         lu->lu_task_alloc = sbd_task_alloc;
1438         lu->lu_new_task = sbd_new_task;
1439         lu->lu_dbuf_xfer_done = sbd_dbuf_xfer_done;
1440         lu->lu_send_status_done = sbd_send_status_done;
1441         lu->lu_task_free = sbd_task_free;
1442         lu->lu_abort = sbd_abort;
1443         lu->lu_dbuf_free = sbd_dbuf_free;
1444         lu->lu_ctl = sbd_ctl;
1445         lu->lu_info = sbd_info;
1446         sl->sl_state = STMF_STATE_OFFLINE;
1447 
1448         if ((ret = stmf_register_lu(lu)) != STMF_SUCCESS) {
1449                 stmf_trace(0, "Failed to register with framework, ret=%llx",
1450                     ret);
1451                 if (ret == STMF_ALREADY) {
1452                         *err_ret = SBD_RET_GUID_ALREADY_REGISTERED;
1453                 }
1454                 return (EIO);
1455         }
1456 
1457         *err_ret = 0;
1458         return (0);
1459 }
1460 
1461 int
1462 sbd_open_data_file(sbd_lu_t *sl, uint32_t *err_ret, int lu_size_valid,
1463     int vp_valid, int keep_open)
1464 {
1465         int ret;
1466         int flag;
1467         ulong_t nbits;
1468         uint64_t supported_size;
1469         vattr_t vattr;
1470         enum vtype vt;
1471         struct dk_cinfo dki;
1472         int unused;
1473 
1474         mutex_enter(&sl->sl_lock);
1475         if (vp_valid) {
1476                 goto odf_over_open;
1477         }
1478         if (sl->sl_data_filename[0] != '/') {
1479                 *err_ret = SBD_RET_DATA_PATH_NOT_ABSOLUTE;
1480                 mutex_exit(&sl->sl_lock);
1481                 return (EINVAL);
1482         }
1483         if ((ret = lookupname(sl->sl_data_filename, UIO_SYSSPACE, FOLLOW,
1484             NULLVPP, &sl->sl_data_vp)) != 0) {
1485                 *err_ret = SBD_RET_DATA_FILE_LOOKUP_FAILED;
1486                 mutex_exit(&sl->sl_lock);
1487                 return (ret);
1488         }
1489         sl->sl_data_vtype = vt = sl->sl_data_vp->v_type;
1490         VN_RELE(sl->sl_data_vp);
1491         if ((vt != VREG) && (vt != VCHR) && (vt != VBLK)) {
1492                 *err_ret = SBD_RET_WRONG_DATA_FILE_TYPE;
1493                 mutex_exit(&sl->sl_lock);
1494                 return (EINVAL);
1495         }
1496         if (sl->sl_flags & SL_WRITE_PROTECTED) {
1497                 flag = FREAD | FOFFMAX;
1498         } else {
1499                 flag = FREAD | FWRITE | FOFFMAX | FEXCL;
1500         }
1501         if ((ret = vn_open(sl->sl_data_filename, UIO_SYSSPACE, flag, 0,
1502             &sl->sl_data_vp, 0, 0)) != 0) {
1503                 *err_ret = SBD_RET_DATA_FILE_OPEN_FAILED;
1504                 mutex_exit(&sl->sl_lock);
1505                 return (ret);
1506         }
1507 odf_over_open:
1508         vattr.va_mask = AT_SIZE;
1509         if ((ret = VOP_GETATTR(sl->sl_data_vp, &vattr, 0, CRED(), NULL)) != 0) {
1510                 *err_ret = SBD_RET_DATA_FILE_GETATTR_FAILED;
1511                 goto odf_close_data_and_exit;
1512         }
1513         if ((vt != VREG) && (vattr.va_size == 0)) {
1514                 /*
1515                  * Its a zero byte block or char device. This cannot be
1516                  * a raw disk.
1517                  */
1518                 *err_ret = SBD_RET_WRONG_DATA_FILE_TYPE;
1519                 ret = EINVAL;
1520                 goto odf_close_data_and_exit;
1521         }
1522         /* sl_data_readable size includes any metadata. */
1523         sl->sl_data_readable_size = vattr.va_size;
1524 
1525         if (VOP_PATHCONF(sl->sl_data_vp, _PC_FILESIZEBITS, &nbits,
1526             CRED(), NULL) != 0) {
1527                 nbits = 0;
1528         }
1529         /* nbits cannot be greater than 64 */
1530         sl->sl_data_fs_nbits = (uint8_t)nbits;
1531         if (lu_size_valid) {
1532                 sl->sl_total_data_size = sl->sl_lu_size;
1533                 if (sl->sl_flags & SL_SHARED_META) {
1534                         sl->sl_total_data_size += SHARED_META_DATA_SIZE;
1535                 }
1536                 if ((nbits > 0) && (nbits < 64)) {
1537                         /*
1538                          * The expression below is correct only if nbits is
1539                          * positive and less than 64.
1540                          */
1541                         supported_size = (((uint64_t)1) << nbits) - 1;
1542                         if (sl->sl_total_data_size > supported_size) {
1543                                 *err_ret = SBD_RET_SIZE_NOT_SUPPORTED_BY_FS;
1544                                 ret = EINVAL;
1545                                 goto odf_close_data_and_exit;
1546                         }
1547                 }
1548         } else {
1549                 sl->sl_total_data_size = vattr.va_size;
1550                 if (sl->sl_flags & SL_SHARED_META) {
1551                         if (vattr.va_size > SHARED_META_DATA_SIZE) {
1552                                 sl->sl_lu_size = vattr.va_size -
1553                                     SHARED_META_DATA_SIZE;
1554                         } else {
1555                                 *err_ret = SBD_RET_FILE_SIZE_ERROR;
1556                                 ret = EINVAL;
1557                                 goto odf_close_data_and_exit;
1558                         }
1559                 } else {
1560                         sl->sl_lu_size = vattr.va_size;
1561                 }
1562         }
1563         if (sl->sl_lu_size < SBD_MIN_LU_SIZE) {
1564                 *err_ret = SBD_RET_FILE_SIZE_ERROR;
1565                 ret = EINVAL;
1566                 goto odf_close_data_and_exit;
1567         }
1568         if (sl->sl_lu_size &
1569             ((((uint64_t)1) << sl->sl_data_blocksize_shift) - 1)) {
1570                 *err_ret = SBD_RET_FILE_ALIGN_ERROR;
1571                 ret = EINVAL;
1572                 goto odf_close_data_and_exit;
1573         }
1574         /*
1575          * Get the minor device for direct zvol access
1576          */
1577         if (sl->sl_flags & SL_ZFS_META) {
1578                 if ((ret = VOP_IOCTL(sl->sl_data_vp, DKIOCINFO, (intptr_t)&dki,
1579                     FKIOCTL, kcred, &unused, NULL)) != 0) {
1580                         cmn_err(CE_WARN, "ioctl(DKIOCINFO) failed %d", ret);
1581                         /* zvol reserves 0, so this would fail later */
1582                         sl->sl_zvol_minor = 0;
1583                 } else {
1584                         sl->sl_zvol_minor = dki.dki_unit;
1585                         if (sbd_zvol_get_volume_params(sl) == 0)
1586                                 sl->sl_flags |= SL_CALL_ZVOL;
1587                 }
1588         }
1589         sl->sl_flags |= SL_MEDIA_LOADED;
1590         mutex_exit(&sl->sl_lock);
1591         return (0);
1592 
1593 odf_close_data_and_exit:
1594         if (!keep_open) {
1595                 (void) VOP_CLOSE(sl->sl_data_vp, flag, 1, 0, CRED(), NULL);
1596                 VN_RELE(sl->sl_data_vp);
1597         }
1598         mutex_exit(&sl->sl_lock);
1599         return (ret);
1600 }
1601 
1602 void
1603 sbd_close_lu(sbd_lu_t *sl)
1604 {
1605         int flag;
1606 
1607         if (((sl->sl_flags & SL_SHARED_META) == 0) &&
1608             (sl->sl_flags & SL_META_OPENED)) {
1609                 if (sl->sl_flags & SL_ZFS_META) {
1610                         rw_destroy(&sl->sl_zfs_meta_lock);
1611                         if (sl->sl_zfs_meta) {
1612                                 kmem_free(sl->sl_zfs_meta, ZAP_MAXVALUELEN / 2);
1613                                 sl->sl_zfs_meta = NULL;
1614                         }
1615                 } else {
1616                         flag = FREAD | FWRITE | FOFFMAX | FEXCL;
1617                         (void) VOP_CLOSE(sl->sl_meta_vp, flag, 1, 0,
1618                             CRED(), NULL);
1619                         VN_RELE(sl->sl_meta_vp);
1620                 }
1621                 sl->sl_flags &= ~SL_META_OPENED;
1622         }
1623         if (sl->sl_flags & SL_MEDIA_LOADED) {
1624                 if (sl->sl_flags & SL_WRITE_PROTECTED) {
1625                         flag = FREAD | FOFFMAX;
1626                 } else {
1627                         flag = FREAD | FWRITE | FOFFMAX | FEXCL;
1628                 }
1629                 (void) VOP_CLOSE(sl->sl_data_vp, flag, 1, 0, CRED(), NULL);
1630                 VN_RELE(sl->sl_data_vp);
1631                 sl->sl_flags &= ~SL_MEDIA_LOADED;
1632                 if (sl->sl_flags & SL_SHARED_META) {
1633                         sl->sl_flags &= ~SL_META_OPENED;
1634                 }
1635         }
1636 }
1637 
1638 int
1639 sbd_set_lu_standby(sbd_set_lu_standby_t *stlu, uint32_t *err_ret)
1640 {
1641         sbd_lu_t *sl;
1642         sbd_status_t sret;
1643         stmf_status_t stret;
1644         uint8_t old_access_state;
1645 
1646         sret = sbd_find_and_lock_lu(stlu->stlu_guid, NULL,
1647             SL_OP_MODIFY_LU, &sl);
1648         if (sret != SBD_SUCCESS) {
1649                 if (sret == SBD_BUSY) {
1650                         *err_ret = SBD_RET_LU_BUSY;
1651                         return (EBUSY);
1652                 } else if (sret == SBD_NOT_FOUND) {
1653                         *err_ret = SBD_RET_NOT_FOUND;
1654                         return (ENOENT);
1655                 }
1656                 *err_ret = SBD_RET_ACCESS_STATE_FAILED;
1657                 return (EIO);
1658         }
1659 
1660         old_access_state = sl->sl_access_state;
1661         sl->sl_access_state = SBD_LU_TRANSITION_TO_STANDBY;
1662         stret = stmf_set_lu_access((stmf_lu_t *)sl->sl_lu, STMF_LU_STANDBY);
1663         if (stret != STMF_SUCCESS) {
1664                 sl->sl_trans_op = SL_OP_NONE;
1665                 *err_ret = SBD_RET_ACCESS_STATE_FAILED;
1666                 sl->sl_access_state = old_access_state;
1667                 return (EIO);
1668         }
1669 
1670         /*
1671          * acquire the writer lock here to ensure we're not pulling
1672          * the rug from the vn_rdwr to the backing store
1673          */
1674         rw_enter(&sl->sl_access_state_lock, RW_WRITER);
1675         sbd_close_lu(sl);
1676         rw_exit(&sl->sl_access_state_lock);
1677 
1678         sl->sl_trans_op = SL_OP_NONE;
1679         return (0);
1680 }
1681 
1682 int
1683 sbd_close_delete_lu(sbd_lu_t *sl, int ret)
1684 {
1685 
1686         /*
1687          * acquire the writer lock here to ensure we're not pulling
1688          * the rug from the vn_rdwr to the backing store
1689          */
1690         rw_enter(&sl->sl_access_state_lock, RW_WRITER);
1691         sbd_close_lu(sl);
1692         rw_exit(&sl->sl_access_state_lock);
1693 
1694         if (sl->sl_flags & SL_LINKED)
1695                 sbd_unlink_lu(sl);
1696         mutex_destroy(&sl->sl_metadata_lock);
1697         mutex_destroy(&sl->sl_lock);
1698         rw_destroy(&sl->sl_pgr->pgr_lock);
1699         rw_destroy(&sl->sl_access_state_lock);
1700         if (sl->sl_serial_no_alloc_size) {
1701                 kmem_free(sl->sl_serial_no, sl->sl_serial_no_alloc_size);
1702         }
1703         if (sl->sl_data_fname_alloc_size) {
1704                 kmem_free(sl->sl_data_filename, sl->sl_data_fname_alloc_size);
1705         }
1706         if (sl->sl_alias_alloc_size) {
1707                 kmem_free(sl->sl_alias, sl->sl_alias_alloc_size);
1708         }
1709         if (sl->sl_mgmt_url_alloc_size) {
1710                 kmem_free(sl->sl_mgmt_url, sl->sl_mgmt_url_alloc_size);
1711         }
1712         stmf_free(sl->sl_lu);
1713         return (ret);
1714 }
1715 
1716 int
1717 sbd_create_register_lu(sbd_create_and_reg_lu_t *slu, int struct_sz,
1718     uint32_t *err_ret)
1719 {
1720         char *namebuf;
1721         sbd_lu_t *sl;
1722         stmf_lu_t *lu;
1723         char *p;
1724         int sz;
1725         int alloc_sz;
1726         int ret = EIO;
1727         int flag;
1728         int wcd = 0;
1729         uint32_t hid = 0;
1730         enum vtype vt;
1731 
1732         sz = struct_sz - sizeof (sbd_create_and_reg_lu_t) + 8 + 1;
1733 
1734         *err_ret = 0;
1735 
1736         /* Lets validate various offsets */
1737         if (((slu->slu_meta_fname_valid) &&
1738             (slu->slu_meta_fname_off >= sz)) ||
1739             (slu->slu_data_fname_off >= sz) ||
1740             ((slu->slu_alias_valid) &&
1741             (slu->slu_alias_off >= sz)) ||
1742             ((slu->slu_mgmt_url_valid) &&
1743             (slu->slu_mgmt_url_off >= sz)) ||
1744             ((slu->slu_serial_valid) &&
1745             ((slu->slu_serial_off + slu->slu_serial_size) >= sz))) {
1746                 return (EINVAL);
1747         }
1748 
1749         namebuf = kmem_zalloc(sz, KM_SLEEP);
1750         bcopy(slu->slu_buf, namebuf, sz - 1);
1751         namebuf[sz - 1] = 0;
1752 
1753         alloc_sz = sizeof (sbd_lu_t) + sizeof (sbd_pgr_t);
1754         if (slu->slu_meta_fname_valid) {
1755                 alloc_sz += strlen(namebuf + slu->slu_meta_fname_off) + 1;
1756         }
1757         alloc_sz += strlen(namebuf + slu->slu_data_fname_off) + 1;
1758         if (slu->slu_alias_valid) {
1759                 alloc_sz += strlen(namebuf + slu->slu_alias_off) + 1;
1760         }
1761         if (slu->slu_mgmt_url_valid) {
1762                 alloc_sz += strlen(namebuf + slu->slu_mgmt_url_off) + 1;
1763         }
1764         if (slu->slu_serial_valid) {
1765                 alloc_sz += slu->slu_serial_size;
1766         }
1767 
1768         lu = (stmf_lu_t *)stmf_alloc(STMF_STRUCT_STMF_LU, alloc_sz, 0);
1769         if (lu == NULL) {
1770                 kmem_free(namebuf, sz);
1771                 return (ENOMEM);
1772         }
1773         sl = (sbd_lu_t *)lu->lu_provider_private;
1774         bzero(sl, alloc_sz);
1775         sl->sl_lu = lu;
1776         sl->sl_alloc_size = alloc_sz;
1777         sl->sl_pgr = (sbd_pgr_t *)(sl + 1);
1778         rw_init(&sl->sl_pgr->pgr_lock, NULL, RW_DRIVER, NULL);
1779         mutex_init(&sl->sl_lock, NULL, MUTEX_DRIVER, NULL);
1780         mutex_init(&sl->sl_metadata_lock, NULL, MUTEX_DRIVER, NULL);
1781         rw_init(&sl->sl_access_state_lock, NULL, RW_DRIVER, NULL);
1782         p = ((char *)sl) + sizeof (sbd_lu_t) + sizeof (sbd_pgr_t);
1783         sl->sl_data_filename = p;
1784         (void) strcpy(sl->sl_data_filename, namebuf + slu->slu_data_fname_off);
1785         p += strlen(sl->sl_data_filename) + 1;
1786         sl->sl_meta_offset = SBD_META_OFFSET;
1787         sl->sl_access_state = SBD_LU_ACTIVE;
1788         if (slu->slu_meta_fname_valid) {
1789                 sl->sl_alias = sl->sl_name = sl->sl_meta_filename = p;
1790                 (void) strcpy(sl->sl_meta_filename, namebuf +
1791                     slu->slu_meta_fname_off);
1792                 p += strlen(sl->sl_meta_filename) + 1;
1793         } else {
1794                 sl->sl_alias = sl->sl_name = sl->sl_data_filename;
1795                 if (sbd_is_zvol(sl->sl_data_filename)) {
1796                         sl->sl_flags |= SL_ZFS_META;
1797                         sl->sl_meta_offset = 0;
1798                 } else {
1799                         sl->sl_flags |= SL_SHARED_META;
1800                         sl->sl_data_offset = SHARED_META_DATA_SIZE;
1801                         sl->sl_total_meta_size = SHARED_META_DATA_SIZE;
1802                         sl->sl_meta_size_used = 0;
1803                 }
1804         }
1805         if (slu->slu_alias_valid) {
1806                 sl->sl_alias = p;
1807                 (void) strcpy(p, namebuf + slu->slu_alias_off);
1808                 p += strlen(sl->sl_alias) + 1;
1809         }
1810         if (slu->slu_mgmt_url_valid) {
1811                 sl->sl_mgmt_url = p;
1812                 (void) strcpy(p, namebuf + slu->slu_mgmt_url_off);
1813                 p += strlen(sl->sl_mgmt_url) + 1;
1814         }
1815         if (slu->slu_serial_valid) {
1816                 sl->sl_serial_no = (uint8_t *)p;
1817                 bcopy(namebuf + slu->slu_serial_off, sl->sl_serial_no,
1818                     slu->slu_serial_size);
1819                 sl->sl_serial_no_size = slu->slu_serial_size;
1820                 p += slu->slu_serial_size;
1821         }
1822         kmem_free(namebuf, sz);
1823         if (slu->slu_vid_valid) {
1824                 bcopy(slu->slu_vid, sl->sl_vendor_id, 8);
1825                 sl->sl_flags |= SL_VID_VALID;
1826         }
1827         if (slu->slu_pid_valid) {
1828                 bcopy(slu->slu_pid, sl->sl_product_id, 16);
1829                 sl->sl_flags |= SL_PID_VALID;
1830         }
1831         if (slu->slu_rev_valid) {
1832                 bcopy(slu->slu_rev, sl->sl_revision, 4);
1833                 sl->sl_flags |= SL_REV_VALID;
1834         }
1835         if (slu->slu_write_protected) {
1836                 sl->sl_flags |= SL_WRITE_PROTECTED;
1837         }
1838         if (slu->slu_blksize_valid) {
1839                 if ((slu->slu_blksize & (slu->slu_blksize - 1)) ||
1840                     (slu->slu_blksize > (32 * 1024)) ||
1841                     (slu->slu_blksize == 0)) {
1842                         *err_ret = SBD_RET_INVALID_BLKSIZE;
1843                         ret = EINVAL;
1844                         goto scm_err_out;
1845                 }
1846                 while ((1 << sl->sl_data_blocksize_shift) != slu->slu_blksize) {
1847                         sl->sl_data_blocksize_shift++;
1848                 }
1849         } else {
1850                 sl->sl_data_blocksize_shift = 9;     /* 512 by default */
1851                 slu->slu_blksize = 512;
1852         }
1853 
1854         /* Now lets start creating meta */
1855         sl->sl_trans_op = SL_OP_CREATE_REGISTER_LU;
1856         if (sbd_link_lu(sl) != SBD_SUCCESS) {
1857                 *err_ret = SBD_RET_FILE_ALREADY_REGISTERED;
1858                 ret = EALREADY;
1859                 goto scm_err_out;
1860         }
1861 
1862         /* 1st focus on the data store */
1863         if (slu->slu_lu_size_valid) {
1864                 sl->sl_lu_size = slu->slu_lu_size;
1865         }
1866         ret = sbd_open_data_file(sl, err_ret, slu->slu_lu_size_valid, 0, 0);
1867         slu->slu_ret_filesize_nbits = sl->sl_data_fs_nbits;
1868         slu->slu_lu_size = sl->sl_lu_size;
1869         if (ret) {
1870                 goto scm_err_out;
1871         }
1872 
1873         /*
1874          * Check if we were explicitly asked to disable/enable write
1875          * cache on the device, otherwise get current device setting.
1876          */
1877         if (slu->slu_writeback_cache_disable_valid) {
1878                 if (slu->slu_writeback_cache_disable) {
1879                         /*
1880                          * Set write cache disable on the device. If it fails,
1881                          * we'll support it using sync/flush.
1882                          */
1883                         (void) sbd_wcd_set(1, sl);
1884                         wcd = 1;
1885                 } else {
1886                         /*
1887                          * Set write cache enable on the device. If it fails,
1888                          * return an error.
1889                          */
1890                         if (sbd_wcd_set(0, sl) != SBD_SUCCESS) {
1891                                 *err_ret = SBD_RET_WRITE_CACHE_SET_FAILED;
1892                                 ret = EFAULT;
1893                                 goto scm_err_out;
1894                         }
1895                 }
1896         } else {
1897                 sbd_wcd_get(&wcd, sl);
1898         }
1899 
1900         if (wcd) {
1901                 sl->sl_flags |= SL_WRITEBACK_CACHE_DISABLE |
1902                     SL_SAVED_WRITE_CACHE_DISABLE;
1903         }
1904 
1905         if (sl->sl_flags & SL_SHARED_META) {
1906                 goto over_meta_open;
1907         }
1908         if (sl->sl_flags & SL_ZFS_META) {
1909                 if (sbd_create_zfs_meta_object(sl) != SBD_SUCCESS) {
1910                         *err_ret = SBD_RET_ZFS_META_CREATE_FAILED;
1911                         ret = ENOMEM;
1912                         goto scm_err_out;
1913                 }
1914                 sl->sl_meta_blocksize_shift = 0;
1915                 goto over_meta_create;
1916         }
1917         if ((ret = lookupname(sl->sl_meta_filename, UIO_SYSSPACE, FOLLOW,
1918             NULLVPP, &sl->sl_meta_vp)) != 0) {
1919                 *err_ret = SBD_RET_META_FILE_LOOKUP_FAILED;
1920                 goto scm_err_out;
1921         }
1922         sl->sl_meta_vtype = vt = sl->sl_meta_vp->v_type;
1923         VN_RELE(sl->sl_meta_vp);
1924         if ((vt != VREG) && (vt != VCHR) && (vt != VBLK)) {
1925                 *err_ret = SBD_RET_WRONG_META_FILE_TYPE;
1926                 ret = EINVAL;
1927                 goto scm_err_out;
1928         }
1929         if (vt == VREG) {
1930                 sl->sl_meta_blocksize_shift = 0;
1931         } else {
1932                 sl->sl_meta_blocksize_shift = 9;
1933         }
1934         flag = FREAD | FWRITE | FOFFMAX | FEXCL;
1935         if ((ret = vn_open(sl->sl_meta_filename, UIO_SYSSPACE, flag, 0,
1936             &sl->sl_meta_vp, 0, 0)) != 0) {
1937                 *err_ret = SBD_RET_META_FILE_OPEN_FAILED;
1938                 goto scm_err_out;
1939         }
1940 over_meta_create:
1941         sl->sl_total_meta_size = sl->sl_meta_offset + sizeof (sbd_meta_start_t);
1942         sl->sl_total_meta_size +=
1943             (((uint64_t)1) << sl->sl_meta_blocksize_shift) - 1;
1944         sl->sl_total_meta_size &=
1945             ~((((uint64_t)1) << sl->sl_meta_blocksize_shift) - 1);
1946         sl->sl_meta_size_used = 0;
1947 over_meta_open:
1948         sl->sl_flags |= SL_META_OPENED;
1949 
1950         sl->sl_device_id[3] = 16;
1951         if (slu->slu_guid_valid) {
1952                 sl->sl_device_id[0] = 0xf1;
1953                 sl->sl_device_id[1] = 3;
1954                 sl->sl_device_id[2] = 0;
1955                 bcopy(slu->slu_guid, sl->sl_device_id + 4, 16);
1956         } else {
1957                 if (slu->slu_host_id_valid)
1958                         hid = slu->slu_host_id;
1959                 if (!slu->slu_company_id_valid)
1960                         slu->slu_company_id = COMPANY_ID_SUN;
1961                 if (stmf_scsilib_uniq_lu_id2(slu->slu_company_id, hid,
1962                     (scsi_devid_desc_t *)&sl->sl_device_id[0]) !=
1963                     STMF_SUCCESS) {
1964                         *err_ret = SBD_RET_META_CREATION_FAILED;
1965                         ret = EIO;
1966                         goto scm_err_out;
1967                 }
1968                 bcopy(sl->sl_device_id + 4, slu->slu_guid, 16);
1969         }
1970 
1971         /* Lets create the meta now */
1972         mutex_enter(&sl->sl_metadata_lock);
1973         if (sbd_write_meta_start(sl, sl->sl_total_meta_size,
1974             sizeof (sbd_meta_start_t)) != SBD_SUCCESS) {
1975                 mutex_exit(&sl->sl_metadata_lock);
1976                 *err_ret = SBD_RET_META_CREATION_FAILED;
1977                 ret = EIO;
1978                 goto scm_err_out;
1979         }
1980         mutex_exit(&sl->sl_metadata_lock);
1981         sl->sl_meta_size_used = sl->sl_meta_offset + sizeof (sbd_meta_start_t);
1982 
1983         if (sbd_write_lu_info(sl) != SBD_SUCCESS) {
1984                 *err_ret = SBD_RET_META_CREATION_FAILED;
1985                 ret = EIO;
1986                 goto scm_err_out;
1987         }
1988 
1989         if (sbd_pgr_meta_init(sl) != SBD_SUCCESS) {
1990                 *err_ret = SBD_RET_META_CREATION_FAILED;
1991                 ret = EIO;
1992                 goto scm_err_out;
1993         }
1994 
1995         /*
1996          * Update the zvol separately as this need only be called upon
1997          * completion of the metadata initialization.
1998          */
1999         if (sl->sl_flags & SL_ZFS_META) {
2000                 if (sbd_update_zfs_prop(sl) != SBD_SUCCESS) {
2001                         *err_ret = SBD_RET_META_CREATION_FAILED;
2002                         ret = EIO;
2003                         goto scm_err_out;
2004                 }
2005         }
2006 
2007         ret = sbd_populate_and_register_lu(sl, err_ret);
2008         if (ret) {
2009                 goto scm_err_out;
2010         }
2011 
2012         sl->sl_trans_op = SL_OP_NONE;
2013         atomic_inc_32(&sbd_lu_count);
2014         return (0);
2015 
2016 scm_err_out:
2017         return (sbd_close_delete_lu(sl, ret));
2018 }
2019 
2020 stmf_status_t
2021 sbd_proxy_msg(uint8_t *luid, void *proxy_arg, uint32_t proxy_arg_len,
2022     uint32_t type)
2023 {
2024         switch (type) {
2025                 case STMF_MSG_LU_ACTIVE:
2026                         return (sbd_proxy_reg_lu(luid, proxy_arg,
2027                             proxy_arg_len));
2028                 case STMF_MSG_LU_REGISTER:
2029                         return (sbd_proxy_reg_lu(luid, proxy_arg,
2030                             proxy_arg_len));
2031                 case STMF_MSG_LU_DEREGISTER:
2032                         return (sbd_proxy_dereg_lu(luid, proxy_arg,
2033                             proxy_arg_len));
2034                 default:
2035                         return (STMF_INVALID_ARG);
2036         }
2037 }
2038 
2039 
2040 /*
2041  * register a standby logical unit
2042  * proxy_reg_arg contains the meta filename
2043  */
2044 stmf_status_t
2045 sbd_proxy_reg_lu(uint8_t *luid, void *proxy_reg_arg, uint32_t proxy_reg_arg_len)
2046 {
2047         sbd_lu_t *sl;
2048         sbd_status_t sret;
2049         sbd_create_standby_lu_t *stlu;
2050         int alloc_sz;
2051         uint32_t err_ret = 0;
2052         stmf_status_t stret = STMF_SUCCESS;
2053 
2054         if (luid == NULL) {
2055                 return (STMF_INVALID_ARG);
2056         }
2057 
2058         do {
2059                 sret = sbd_find_and_lock_lu(luid, NULL, SL_OP_MODIFY_LU, &sl);
2060         } while (sret == SBD_BUSY);
2061 
2062         if (sret == SBD_NOT_FOUND) {
2063                 alloc_sz = sizeof (*stlu) + proxy_reg_arg_len - 8;
2064                 stlu = (sbd_create_standby_lu_t *)kmem_zalloc(alloc_sz,
2065                     KM_SLEEP);
2066                 bcopy(luid, stlu->stlu_guid, 16);
2067                 if (proxy_reg_arg_len) {
2068                         bcopy(proxy_reg_arg, stlu->stlu_meta_fname,
2069                             proxy_reg_arg_len);
2070                         stlu->stlu_meta_fname_size = proxy_reg_arg_len;
2071                 }
2072                 if (sbd_create_standby_lu(stlu, &err_ret) != 0) {
2073                         cmn_err(CE_WARN,
2074                             "Unable to create standby logical unit for %s",
2075                             stlu->stlu_meta_fname);
2076                         stret = STMF_FAILURE;
2077                 }
2078                 kmem_free(stlu, alloc_sz);
2079                 return (stret);
2080         } else if (sret == SBD_SUCCESS) {
2081                 /*
2082                  * if the lu is already registered, then the lu should now
2083                  * be in standby mode
2084                  */
2085                 sbd_it_data_t *it;
2086                 if (sl->sl_access_state != SBD_LU_STANDBY) {
2087                         mutex_enter(&sl->sl_lock);
2088                         sl->sl_access_state = SBD_LU_STANDBY;
2089                         for (it = sl->sl_it_list; it != NULL;
2090                             it = it->sbd_it_next) {
2091                                 it->sbd_it_ua_conditions |=
2092                                     SBD_UA_ASYMMETRIC_ACCESS_CHANGED;
2093                                 it->sbd_it_flags &=
2094                                     ~SBD_IT_HAS_SCSI2_RESERVATION;
2095                                 sl->sl_flags &= ~SL_LU_HAS_SCSI2_RESERVATION;
2096                         }
2097                         mutex_exit(&sl->sl_lock);
2098                         sbd_pgr_reset(sl);
2099                 }
2100                 sl->sl_trans_op = SL_OP_NONE;
2101         } else {
2102                 cmn_err(CE_WARN, "could not find and lock logical unit");
2103                 stret = STMF_FAILURE;
2104         }
2105 out:
2106         return (stret);
2107 }
2108 
2109 /* ARGSUSED */
2110 stmf_status_t
2111 sbd_proxy_dereg_lu(uint8_t *luid, void *proxy_reg_arg,
2112     uint32_t proxy_reg_arg_len)
2113 {
2114         sbd_delete_lu_t dlu = {0};
2115         uint32_t err_ret;
2116 
2117         if (luid == NULL) {
2118                 cmn_err(CE_WARN, "de-register lu request had null luid");
2119                 return (STMF_INVALID_ARG);
2120         }
2121 
2122         bcopy(luid, &dlu.dlu_guid, 16);
2123 
2124         if (sbd_delete_lu(&dlu, (int)sizeof (dlu), &err_ret) != 0) {
2125                 cmn_err(CE_WARN, "failed to delete de-register lu request");
2126                 return (STMF_FAILURE);
2127         }
2128 
2129         return (STMF_SUCCESS);
2130 }
2131 
2132 int
2133 sbd_create_standby_lu(sbd_create_standby_lu_t *slu, uint32_t *err_ret)
2134 {
2135         sbd_lu_t *sl;
2136         stmf_lu_t *lu;
2137         int ret = EIO;
2138         int alloc_sz;
2139 
2140         alloc_sz = sizeof (sbd_lu_t) + sizeof (sbd_pgr_t) +
2141             slu->stlu_meta_fname_size;
2142         lu = (stmf_lu_t *)stmf_alloc(STMF_STRUCT_STMF_LU, alloc_sz, 0);
2143         if (lu == NULL) {
2144                 return (ENOMEM);
2145         }
2146         sl = (sbd_lu_t *)lu->lu_provider_private;
2147         bzero(sl, alloc_sz);
2148         sl->sl_lu = lu;
2149         sl->sl_alloc_size = alloc_sz;
2150 
2151         sl->sl_pgr = (sbd_pgr_t *)(sl + 1);
2152         sl->sl_meta_filename = ((char *)sl) + sizeof (sbd_lu_t) +
2153             sizeof (sbd_pgr_t);
2154 
2155         if (slu->stlu_meta_fname_size > 0) {
2156                 (void) strcpy(sl->sl_meta_filename, slu->stlu_meta_fname);
2157         }
2158         sl->sl_name = sl->sl_meta_filename;
2159 
2160         sl->sl_device_id[3] = 16;
2161         sl->sl_device_id[0] = 0xf1;
2162         sl->sl_device_id[1] = 3;
2163         sl->sl_device_id[2] = 0;
2164         bcopy(slu->stlu_guid, sl->sl_device_id + 4, 16);
2165         lu->lu_id = (scsi_devid_desc_t *)sl->sl_device_id;
2166         sl->sl_access_state = SBD_LU_STANDBY;
2167 
2168         rw_init(&sl->sl_pgr->pgr_lock, NULL, RW_DRIVER, NULL);
2169         mutex_init(&sl->sl_lock, NULL, MUTEX_DRIVER, NULL);
2170         mutex_init(&sl->sl_metadata_lock, NULL, MUTEX_DRIVER, NULL);
2171         rw_init(&sl->sl_access_state_lock, NULL, RW_DRIVER, NULL);
2172 
2173         sl->sl_trans_op = SL_OP_CREATE_REGISTER_LU;
2174 
2175         if (sbd_link_lu(sl) != SBD_SUCCESS) {
2176                 *err_ret = SBD_RET_FILE_ALREADY_REGISTERED;
2177                 ret = EALREADY;
2178                 goto scs_err_out;
2179         }
2180 
2181         ret = sbd_populate_and_register_lu(sl, err_ret);
2182         if (ret) {
2183                 goto scs_err_out;
2184         }
2185 
2186         sl->sl_trans_op = SL_OP_NONE;
2187         atomic_inc_32(&sbd_lu_count);
2188         return (0);
2189 
2190 scs_err_out:
2191         return (sbd_close_delete_lu(sl, ret));
2192 }
2193 
2194 int
2195 sbd_load_sli_1_0(sbd_lu_t *sl, uint32_t *err_ret)
2196 {
2197         sbd_lu_info_1_0_t *sli = NULL;
2198         sbd_status_t sret;
2199 
2200         sret = sbd_read_meta_section(sl, (sm_section_hdr_t **)&sli,
2201             SMS_ID_LU_INFO_1_0);
2202 
2203         if (sret != SBD_SUCCESS) {
2204                 *err_ret = SBD_RET_NO_META;
2205                 return (EIO);
2206         }
2207         if (sli->sli_data_order != SMS_DATA_ORDER) {
2208                 sbd_swap_lu_info_1_0(sli);
2209                 if (sli->sli_data_order != SMS_DATA_ORDER) {
2210                         kmem_free(sli, sli->sli_sms_header.sms_size);
2211                         *err_ret = SBD_RET_NO_META;
2212                         return (EIO);
2213                 }
2214         }
2215 
2216         sl->sl_flags |= SL_SHARED_META;
2217         sl->sl_data_blocksize_shift = 9;
2218         sl->sl_data_offset = SHARED_META_DATA_SIZE;
2219         sl->sl_lu_size = sli->sli_total_store_size - SHARED_META_DATA_SIZE;
2220         sl->sl_total_data_size = SHARED_META_DATA_SIZE + sl->sl_lu_size;
2221         bcopy(sli->sli_lu_devid, sl->sl_device_id, 20);
2222 
2223         kmem_free(sli, sli->sli_sms_header.sms_size);
2224         return (0);
2225 }
2226 
2227 int
2228 sbd_import_lu(sbd_import_lu_t *ilu, int struct_sz, uint32_t *err_ret,
2229     int no_register, sbd_lu_t **slr)
2230 {
2231         stmf_lu_t *lu;
2232         sbd_lu_t *sl;
2233         sbd_lu_info_1_1_t *sli = NULL;
2234         int asz;
2235         int ret = 0;
2236         stmf_status_t stret;
2237         int flag;
2238         int wcd = 0;
2239         int data_opened;
2240         uint16_t sli_buf_sz;
2241         uint8_t *sli_buf_copy = NULL;
2242         enum vtype vt;
2243         int standby = 0;
2244         sbd_status_t sret;
2245 
2246         if (no_register && slr == NULL) {
2247                 return (EINVAL);
2248         }
2249         ilu->ilu_meta_fname[struct_sz - sizeof (*ilu) + 8 - 1] = 0;
2250         /*
2251          * check whether logical unit is already registered ALUA
2252          * For a standby logical unit, the meta filename is set. Use
2253          * that to search for an existing logical unit.
2254          */
2255         sret = sbd_find_and_lock_lu(NULL, (uint8_t *)&(ilu->ilu_meta_fname),
2256             SL_OP_IMPORT_LU, &sl);
2257 
2258         if (sret == SBD_SUCCESS) {
2259                 if (sl->sl_access_state != SBD_LU_ACTIVE) {
2260                         no_register = 1;
2261                         standby = 1;
2262                         lu = sl->sl_lu;
2263                         if (sl->sl_alias_alloc_size) {
2264                                 kmem_free(sl->sl_alias,
2265                                     sl->sl_alias_alloc_size);
2266                                 sl->sl_alias_alloc_size = 0;
2267                                 sl->sl_alias = NULL;
2268                                 lu->lu_alias = NULL;
2269                         }
2270                         if (sl->sl_meta_filename == NULL) {
2271                                 sl->sl_meta_filename = sl->sl_data_filename;
2272                         } else if (sl->sl_data_fname_alloc_size) {
2273                                 kmem_free(sl->sl_data_filename,
2274                                     sl->sl_data_fname_alloc_size);
2275                                 sl->sl_data_fname_alloc_size = 0;
2276                         }
2277                         if (sl->sl_serial_no_alloc_size) {
2278                                 kmem_free(sl->sl_serial_no,
2279                                     sl->sl_serial_no_alloc_size);
2280                                 sl->sl_serial_no_alloc_size = 0;
2281                         }
2282                         if (sl->sl_mgmt_url_alloc_size) {
2283                                 kmem_free(sl->sl_mgmt_url,
2284                                     sl->sl_mgmt_url_alloc_size);
2285                                 sl->sl_mgmt_url_alloc_size = 0;
2286                         }
2287                 } else {
2288                         *err_ret = SBD_RET_FILE_ALREADY_REGISTERED;
2289                         bcopy(sl->sl_device_id + 4, ilu->ilu_ret_guid, 16);
2290                         sl->sl_trans_op = SL_OP_NONE;
2291                         return (EALREADY);
2292                 }
2293         } else if (sret == SBD_NOT_FOUND) {
2294                 asz = strlen(ilu->ilu_meta_fname) + 1;
2295 
2296                 lu = (stmf_lu_t *)stmf_alloc(STMF_STRUCT_STMF_LU,
2297                     sizeof (sbd_lu_t) + sizeof (sbd_pgr_t) + asz, 0);
2298                 if (lu == NULL) {
2299                         return (ENOMEM);
2300                 }
2301                 sl = (sbd_lu_t *)lu->lu_provider_private;
2302                 bzero(sl, sizeof (*sl));
2303                 sl->sl_lu = lu;
2304                 sl->sl_pgr = (sbd_pgr_t *)(sl + 1);
2305                 sl->sl_meta_filename = ((char *)sl) + sizeof (*sl) +
2306                     sizeof (sbd_pgr_t);
2307                 (void) strcpy(sl->sl_meta_filename, ilu->ilu_meta_fname);
2308                 sl->sl_name = sl->sl_meta_filename;
2309                 rw_init(&sl->sl_pgr->pgr_lock, NULL, RW_DRIVER, NULL);
2310                 rw_init(&sl->sl_access_state_lock, NULL, RW_DRIVER, NULL);
2311                 mutex_init(&sl->sl_lock, NULL, MUTEX_DRIVER, NULL);
2312                 mutex_init(&sl->sl_metadata_lock, NULL, MUTEX_DRIVER, NULL);
2313                 sl->sl_trans_op = SL_OP_IMPORT_LU;
2314         } else {
2315                 *err_ret = SBD_RET_META_FILE_LOOKUP_FAILED;
2316                 return (EIO);
2317         }
2318 
2319         /* we're only loading the metadata */
2320         if (!no_register) {
2321                 if (sbd_link_lu(sl) != SBD_SUCCESS) {
2322                         *err_ret = SBD_RET_FILE_ALREADY_REGISTERED;
2323                         bcopy(sl->sl_device_id + 4, ilu->ilu_ret_guid, 16);
2324                         ret = EALREADY;
2325                         goto sim_err_out;
2326                 }
2327         }
2328         if ((ret = lookupname(sl->sl_meta_filename, UIO_SYSSPACE, FOLLOW,
2329             NULLVPP, &sl->sl_meta_vp)) != 0) {
2330                 *err_ret = SBD_RET_META_FILE_LOOKUP_FAILED;
2331                 goto sim_err_out;
2332         }
2333         if (sbd_is_zvol(sl->sl_meta_filename)) {
2334                 sl->sl_flags |= SL_ZFS_META;
2335                 sl->sl_data_filename = sl->sl_meta_filename;
2336         }
2337         sl->sl_meta_vtype = vt = sl->sl_meta_vp->v_type;
2338         VN_RELE(sl->sl_meta_vp);
2339         if ((vt != VREG) && (vt != VCHR) && (vt != VBLK)) {
2340                 *err_ret = SBD_RET_WRONG_META_FILE_TYPE;
2341                 ret = EINVAL;
2342                 goto sim_err_out;
2343         }
2344         if (sl->sl_flags & SL_ZFS_META) {
2345                 if (sbd_open_zfs_meta(sl) != SBD_SUCCESS) {
2346                         /* let see if metadata is in the 64k block */
2347                         sl->sl_flags &= ~SL_ZFS_META;
2348                 }
2349         }
2350         if (!(sl->sl_flags & SL_ZFS_META)) {
2351                 /* metadata is always writable */
2352                 flag = FREAD | FWRITE | FOFFMAX | FEXCL;
2353                 if ((ret = vn_open(sl->sl_meta_filename, UIO_SYSSPACE, flag, 0,
2354                     &sl->sl_meta_vp, 0, 0)) != 0) {
2355                         *err_ret = SBD_RET_META_FILE_OPEN_FAILED;
2356                         goto sim_err_out;
2357                 }
2358         }
2359         if ((sl->sl_flags & SL_ZFS_META) || (vt == VREG)) {
2360                 sl->sl_meta_blocksize_shift = 0;
2361         } else {
2362                 sl->sl_meta_blocksize_shift = 9;
2363         }
2364         sl->sl_meta_offset = (sl->sl_flags & SL_ZFS_META) ? 0 : SBD_META_OFFSET;
2365         sl->sl_flags |= SL_META_OPENED;
2366 
2367         mutex_enter(&sl->sl_metadata_lock);
2368         sret = sbd_load_meta_start(sl);
2369         mutex_exit(&sl->sl_metadata_lock);
2370         if (sret != SBD_SUCCESS) {
2371                 if (sret == SBD_META_CORRUPTED) {
2372                         *err_ret = SBD_RET_NO_META;
2373                 } else if (sret == SBD_NOT_SUPPORTED) {
2374                         *err_ret = SBD_RET_VERSION_NOT_SUPPORTED;
2375                 } else {
2376                         *err_ret = SBD_RET_NO_META;
2377                 }
2378                 ret = EINVAL;
2379                 goto sim_err_out;
2380         }
2381 
2382         /* Now lets see if we can read the most recent LU info */
2383         sret = sbd_read_meta_section(sl, (sm_section_hdr_t **)&sli,
2384             SMS_ID_LU_INFO_1_1);
2385         if ((sret == SBD_NOT_FOUND) && ((sl->sl_flags & SL_ZFS_META) == 0)) {
2386                 ret = sbd_load_sli_1_0(sl, err_ret);
2387                 if (ret) {
2388                         goto sim_err_out;
2389                 }
2390                 goto sim_sli_loaded;
2391         }
2392         if (sret != SBD_SUCCESS) {
2393                 *err_ret = SBD_RET_NO_META;
2394                 ret = EIO;
2395                 goto sim_err_out;
2396         }
2397         /* load sli 1.1 */
2398         if (sli->sli_data_order != SMS_DATA_ORDER) {
2399                 sbd_swap_lu_info_1_1(sli);
2400                 if (sli->sli_data_order != SMS_DATA_ORDER) {
2401                         *err_ret = SBD_RET_NO_META;
2402                         ret = EIO;
2403                         goto sim_err_out;
2404                 }
2405         }
2406 
2407         sli_buf_sz = sli->sli_sms_header.sms_size -
2408             sizeof (sbd_lu_info_1_1_t) + 8;
2409         sli_buf_copy = kmem_alloc(sli_buf_sz + 1, KM_SLEEP);
2410         bcopy(sli->sli_buf, sli_buf_copy, sli_buf_sz);
2411         sli_buf_copy[sli_buf_sz] = 0;
2412 
2413         /* Make sure all the offsets are within limits */
2414         if (((sli->sli_flags & SLI_META_FNAME_VALID) &&
2415             (sli->sli_meta_fname_offset > sli_buf_sz)) ||
2416             ((sli->sli_flags & SLI_DATA_FNAME_VALID) &&
2417             (sli->sli_data_fname_offset > sli_buf_sz)) ||
2418             ((sli->sli_flags & SLI_MGMT_URL_VALID) &&
2419             (sli->sli_mgmt_url_offset > sli_buf_sz)) ||
2420             ((sli->sli_flags & SLI_SERIAL_VALID) &&
2421             ((sli->sli_serial_offset + sli->sli_serial_size) > sli_buf_sz)) ||
2422             ((sli->sli_flags & SLI_ALIAS_VALID) &&
2423             (sli->sli_alias_offset > sli_buf_sz))) {
2424                 *err_ret = SBD_RET_NO_META;
2425                 ret = EIO;
2426                 goto sim_err_out;
2427         }
2428 
2429         sl->sl_lu_size = sli->sli_lu_size;
2430         sl->sl_data_blocksize_shift = sli->sli_data_blocksize_shift;
2431         bcopy(sli->sli_device_id, sl->sl_device_id, 20);
2432         if (sli->sli_flags & SLI_SERIAL_VALID) {
2433                 sl->sl_serial_no_size = sl->sl_serial_no_alloc_size =
2434                     sli->sli_serial_size;
2435                 sl->sl_serial_no = kmem_zalloc(sli->sli_serial_size, KM_SLEEP);
2436                 bcopy(sli_buf_copy + sli->sli_serial_offset, sl->sl_serial_no,
2437                     sl->sl_serial_no_size);
2438         }
2439         if (sli->sli_flags & SLI_SEPARATE_META) {
2440                 sl->sl_total_data_size = sl->sl_lu_size;
2441                 if (sli->sli_flags & SLI_DATA_FNAME_VALID) {
2442                         sl->sl_data_fname_alloc_size = strlen((char *)
2443                             sli_buf_copy + sli->sli_data_fname_offset) + 1;
2444                         sl->sl_data_filename = kmem_zalloc(
2445                             sl->sl_data_fname_alloc_size, KM_SLEEP);
2446                         (void) strcpy(sl->sl_data_filename,
2447                             (char *)sli_buf_copy + sli->sli_data_fname_offset);
2448                 }
2449         } else {
2450                 if (sl->sl_flags & SL_ZFS_META) {
2451                         sl->sl_total_data_size = sl->sl_lu_size;
2452                         sl->sl_data_offset = 0;
2453                 } else {
2454                         sl->sl_total_data_size =
2455                             sl->sl_lu_size + SHARED_META_DATA_SIZE;
2456                         sl->sl_data_offset = SHARED_META_DATA_SIZE;
2457                         sl->sl_flags |= SL_SHARED_META;
2458                 }
2459         }
2460         if (sli->sli_flags & SLI_ALIAS_VALID) {
2461                 sl->sl_alias_alloc_size = strlen((char *)sli_buf_copy +
2462                     sli->sli_alias_offset) + 1;
2463                 sl->sl_alias = kmem_alloc(sl->sl_alias_alloc_size, KM_SLEEP);
2464                 (void) strcpy(sl->sl_alias, (char *)sli_buf_copy +
2465                     sli->sli_alias_offset);
2466         }
2467         if (sli->sli_flags & SLI_MGMT_URL_VALID) {
2468                 sl->sl_mgmt_url_alloc_size = strlen((char *)sli_buf_copy +
2469                     sli->sli_mgmt_url_offset) + 1;
2470                 sl->sl_mgmt_url = kmem_alloc(sl->sl_mgmt_url_alloc_size,
2471                     KM_SLEEP);
2472                 (void) strcpy(sl->sl_mgmt_url, (char *)sli_buf_copy +
2473                     sli->sli_mgmt_url_offset);
2474         }
2475         if (sli->sli_flags & SLI_WRITE_PROTECTED) {
2476                 sl->sl_flags |= SL_WRITE_PROTECTED;
2477         }
2478         if (sli->sli_flags & SLI_VID_VALID) {
2479                 sl->sl_flags |= SL_VID_VALID;
2480                 bcopy(sli->sli_vid, sl->sl_vendor_id, 8);
2481         }
2482         if (sli->sli_flags & SLI_PID_VALID) {
2483                 sl->sl_flags |= SL_PID_VALID;
2484                 bcopy(sli->sli_pid, sl->sl_product_id, 16);
2485         }
2486         if (sli->sli_flags & SLI_REV_VALID) {
2487                 sl->sl_flags |= SL_REV_VALID;
2488                 bcopy(sli->sli_rev, sl->sl_revision, 4);
2489         }
2490         if (sli->sli_flags & SLI_WRITEBACK_CACHE_DISABLE) {
2491                 sl->sl_flags |= SL_WRITEBACK_CACHE_DISABLE;
2492         }
2493 sim_sli_loaded:
2494         if ((sl->sl_flags & SL_SHARED_META) == 0) {
2495                 data_opened = 0;
2496         } else {
2497                 data_opened = 1;
2498                 sl->sl_data_filename = sl->sl_meta_filename;
2499                 sl->sl_data_vp = sl->sl_meta_vp;
2500                 sl->sl_data_vtype = sl->sl_meta_vtype;
2501         }
2502 
2503         sret = sbd_pgr_meta_load(sl);
2504         if (sret != SBD_SUCCESS) {
2505                 *err_ret = SBD_RET_NO_META;
2506                 ret = EIO;
2507                 goto sim_err_out;
2508         }
2509 
2510         ret = sbd_open_data_file(sl, err_ret, 1, data_opened, 0);
2511         if (ret) {
2512                 goto sim_err_out;
2513         }
2514 
2515         /*
2516          * set write cache disable on the device
2517          * Note: this shouldn't fail on import unless the cache capabilities
2518          * of the device changed. If that happened, modify will need to
2519          * be used to set the cache flag appropriately after import is done.
2520          */
2521         if (sl->sl_flags & SL_WRITEBACK_CACHE_DISABLE) {
2522                 (void) sbd_wcd_set(1, sl);
2523                 wcd = 1;
2524         /*
2525          * if not explicitly set, attempt to set it to enable, if that fails
2526          * get the current setting and use that
2527          */
2528         } else {
2529                 sret = sbd_wcd_set(0, sl);
2530                 if (sret != SBD_SUCCESS) {
2531                         sbd_wcd_get(&wcd, sl);
2532                 }
2533         }
2534 
2535         if (wcd) {
2536                 sl->sl_flags |= SL_WRITEBACK_CACHE_DISABLE |
2537                     SL_SAVED_WRITE_CACHE_DISABLE;
2538         }
2539 
2540         /* we're only loading the metadata */
2541         if (!no_register) {
2542                 ret = sbd_populate_and_register_lu(sl, err_ret);
2543                 if (ret) {
2544                         goto sim_err_out;
2545                 }
2546                 atomic_inc_32(&sbd_lu_count);
2547         }
2548 
2549         bcopy(sl->sl_device_id + 4, ilu->ilu_ret_guid, 16);
2550         sl->sl_trans_op = SL_OP_NONE;
2551 
2552         if (sli) {
2553                 kmem_free(sli, sli->sli_sms_header.sms_size);
2554                 sli = NULL;
2555         }
2556         if (sli_buf_copy) {
2557                 kmem_free(sli_buf_copy, sli_buf_sz + 1);
2558                 sli_buf_copy = NULL;
2559         }
2560         if (no_register && !standby) {
2561                 *slr = sl;
2562         }
2563 
2564         /*
2565          * if this was imported from standby, set the access state
2566          * to active.
2567          */
2568         if (standby) {
2569                 sbd_it_data_t *it;
2570                 mutex_enter(&sl->sl_lock);
2571                 sl->sl_access_state = SBD_LU_ACTIVE;
2572                 for (it = sl->sl_it_list; it != NULL;
2573                     it = it->sbd_it_next) {
2574                         it->sbd_it_ua_conditions |=
2575                             SBD_UA_ASYMMETRIC_ACCESS_CHANGED;
2576                         it->sbd_it_ua_conditions |= SBD_UA_POR;
2577                         it->sbd_it_flags |=  SBD_IT_PGR_CHECK_FLAG;
2578                 }
2579                 mutex_exit(&sl->sl_lock);
2580                 /* call set access state */
2581                 stret = stmf_set_lu_access(lu, STMF_LU_ACTIVE);
2582                 if (stret != STMF_SUCCESS) {
2583                         *err_ret = SBD_RET_ACCESS_STATE_FAILED;
2584                         sl->sl_access_state = SBD_LU_STANDBY;
2585                         goto sim_err_out;
2586                 }
2587                 if (sl->sl_alias) {
2588                         lu->lu_alias = sl->sl_alias;
2589                 } else {
2590                         lu->lu_alias = sl->sl_name;
2591                 }
2592         }
2593         sl->sl_access_state = SBD_LU_ACTIVE;
2594         return (0);
2595 
2596 sim_err_out:
2597         if (sli) {
2598                 kmem_free(sli, sli->sli_sms_header.sms_size);
2599                 sli = NULL;
2600         }
2601         if (sli_buf_copy) {
2602                 kmem_free(sli_buf_copy, sli_buf_sz + 1);
2603                 sli_buf_copy = NULL;
2604         }
2605 
2606         if (standby) {
2607                 *err_ret = SBD_RET_ACCESS_STATE_FAILED;
2608                 sl->sl_trans_op = SL_OP_NONE;
2609                 return (EIO);
2610         } else {
2611                 return (sbd_close_delete_lu(sl, ret));
2612         }
2613 }
2614 
2615 int
2616 sbd_modify_lu(sbd_modify_lu_t *mlu, int struct_sz, uint32_t *err_ret)
2617 {
2618         sbd_lu_t *sl = NULL;
2619         uint16_t alias_sz;
2620         int ret = 0;
2621         sbd_it_data_t *it;
2622         sbd_status_t sret;
2623         uint64_t old_size;
2624         int modify_unregistered = 0;
2625         int ua = 0;
2626         sbd_import_lu_t *ilu;
2627         stmf_lu_t *lu;
2628         uint32_t ilu_sz;
2629         uint32_t sz;
2630 
2631         sz = struct_sz - sizeof (*mlu) + 8 + 1;
2632 
2633         /* if there is data in the buf, null terminate it */
2634         if (struct_sz > sizeof (*mlu)) {
2635                 mlu->mlu_buf[struct_sz - sizeof (*mlu) + 8 - 1] = 0;
2636         }
2637 
2638         *err_ret = 0;
2639 
2640         /* Lets validate offsets */
2641         if (((mlu->mlu_alias_valid) &&
2642             (mlu->mlu_alias_off >= sz)) ||
2643             ((mlu->mlu_mgmt_url_valid) &&
2644             (mlu->mlu_mgmt_url_off >= sz)) ||
2645             (mlu->mlu_by_fname) &&
2646             (mlu->mlu_fname_off >= sz)) {
2647                 return (EINVAL);
2648         }
2649 
2650         /*
2651          * We'll look for the device but if we don't find it registered,
2652          * we'll still try to modify the unregistered device.
2653          */
2654         if (mlu->mlu_by_guid) {
2655                 sret = sbd_find_and_lock_lu(mlu->mlu_input_guid, NULL,
2656                     SL_OP_MODIFY_LU, &sl);
2657         } else if (mlu->mlu_by_fname) {
2658                 sret = sbd_find_and_lock_lu(NULL,
2659                     (uint8_t *)&(mlu->mlu_buf[mlu->mlu_fname_off]),
2660                     SL_OP_MODIFY_LU, &sl);
2661         } else {
2662                 return (EINVAL);
2663         }
2664 
2665 
2666         if (sret != SBD_SUCCESS) {
2667                 if (sret == SBD_BUSY) {
2668                         *err_ret = SBD_RET_LU_BUSY;
2669                         return (EBUSY);
2670                 } else if (sret != SBD_NOT_FOUND) {
2671                         return (EIO);
2672                 } else if (!mlu->mlu_by_fname) {
2673                         return (EINVAL);
2674                 }
2675                 /* Okay, try to import the device */
2676                 struct_sz = max(8, strlen(&(mlu->mlu_buf[mlu->mlu_fname_off]))
2677                     + 1);
2678                 struct_sz += sizeof (sbd_import_lu_t) - 8;
2679                 ilu_sz = struct_sz;
2680                 ilu = (sbd_import_lu_t *)kmem_zalloc(ilu_sz, KM_SLEEP);
2681                 ilu->ilu_struct_size = struct_sz;
2682                 (void) strcpy(ilu->ilu_meta_fname,
2683                     &(mlu->mlu_buf[mlu->mlu_fname_off]));
2684                 ret = sbd_import_lu(ilu, struct_sz, err_ret, 1, &sl);
2685                 kmem_free(ilu, ilu_sz);
2686                 if (ret != SBD_SUCCESS) {
2687                         return (ENOENT);
2688                 }
2689                 modify_unregistered = 1;
2690         }
2691 
2692         if (sl->sl_access_state != SBD_LU_ACTIVE) {
2693                 *err_ret = SBD_RET_ACCESS_STATE_FAILED;
2694                 ret = EINVAL;
2695                 goto smm_err_out;
2696         }
2697 
2698         /* check for write cache change */
2699         if (mlu->mlu_writeback_cache_disable_valid) {
2700                 /* set wce on device */
2701                 sret = sbd_wcd_set(mlu->mlu_writeback_cache_disable, sl);
2702                 if (!mlu->mlu_writeback_cache_disable && sret != SBD_SUCCESS) {
2703                         *err_ret = SBD_RET_WRITE_CACHE_SET_FAILED;
2704                         ret = EFAULT;
2705                         goto smm_err_out;
2706                 }
2707                 mutex_enter(&sl->sl_lock);
2708                 if (!mlu->mlu_writeback_cache_disable) {
2709                         if (sl->sl_flags & SL_WRITEBACK_CACHE_DISABLE) {
2710                                 ua = 1;
2711                                 sl->sl_flags &= ~SL_WRITEBACK_CACHE_DISABLE;
2712                                 sl->sl_flags &= ~SL_SAVED_WRITE_CACHE_DISABLE;
2713                         }
2714                 } else {
2715                         if ((sl->sl_flags & SL_WRITEBACK_CACHE_DISABLE) == 0) {
2716                                 ua = 1;
2717                                 sl->sl_flags |= SL_WRITEBACK_CACHE_DISABLE;
2718                                 sl->sl_flags |= SL_SAVED_WRITE_CACHE_DISABLE;
2719                         }
2720                 }
2721                 for (it = sl->sl_it_list; ua && it != NULL;
2722                     it = it->sbd_it_next) {
2723                         it->sbd_it_ua_conditions |=
2724                             SBD_UA_MODE_PARAMETERS_CHANGED;
2725                 }
2726                 mutex_exit(&sl->sl_lock);
2727         }
2728         ua = 0;
2729 
2730         if (mlu->mlu_alias_valid) {
2731                 alias_sz = strlen((char *)mlu->mlu_buf +
2732                     mlu->mlu_alias_off) + 1;
2733                 /*
2734                  * Use the allocated buffer or alloc a new one.
2735                  * Don't copy into sl_alias if sl_alias_alloc_size is 0
2736                  * otherwise or you'll be writing over the data/metadata
2737                  * filename.
2738                  */
2739                 mutex_enter(&sl->sl_lock);
2740                 if (sl->sl_alias_alloc_size > 0 &&
2741                     sl->sl_alias_alloc_size < alias_sz) {
2742                         kmem_free(sl->sl_alias,
2743                             sl->sl_alias_alloc_size);
2744                         sl->sl_alias_alloc_size = 0;
2745                 }
2746                 if (sl->sl_alias_alloc_size == 0) {
2747                         sl->sl_alias = kmem_alloc(alias_sz, KM_SLEEP);
2748                         sl->sl_alias_alloc_size = alias_sz;
2749                 }
2750                 (void) strcpy(sl->sl_alias, (char *)mlu->mlu_buf +
2751                     mlu->mlu_alias_off);
2752                 lu = sl->sl_lu;
2753                 lu->lu_alias = sl->sl_alias;
2754                 mutex_exit(&sl->sl_lock);
2755         }
2756 
2757         if (mlu->mlu_mgmt_url_valid) {
2758                 uint16_t url_sz;
2759 
2760                 url_sz = strlen((char *)mlu->mlu_buf + mlu->mlu_mgmt_url_off);
2761                 if (url_sz > 0)
2762                         url_sz++;
2763 
2764                 mutex_enter(&sl->sl_lock);
2765                 if (sl->sl_mgmt_url_alloc_size > 0 &&
2766                     (url_sz == 0 || sl->sl_mgmt_url_alloc_size < url_sz)) {
2767                         kmem_free(sl->sl_mgmt_url, sl->sl_mgmt_url_alloc_size);
2768                         sl->sl_mgmt_url = NULL;
2769                         sl->sl_mgmt_url_alloc_size = 0;
2770                 }
2771                 if (url_sz > 0) {
2772                         if (sl->sl_mgmt_url_alloc_size == 0) {
2773                                 sl->sl_mgmt_url = kmem_alloc(url_sz, KM_SLEEP);
2774                                 sl->sl_mgmt_url_alloc_size = url_sz;
2775                         }
2776                         (void) strcpy(sl->sl_mgmt_url, (char *)mlu->mlu_buf +
2777                             mlu->mlu_mgmt_url_off);
2778                 }
2779                 for (it = sl->sl_it_list; it != NULL;
2780                     it = it->sbd_it_next) {
2781                         it->sbd_it_ua_conditions |=
2782                             SBD_UA_MODE_PARAMETERS_CHANGED;
2783                 }
2784                 mutex_exit(&sl->sl_lock);
2785         }
2786 
2787         if (mlu->mlu_write_protected_valid) {
2788                 mutex_enter(&sl->sl_lock);
2789                 if (mlu->mlu_write_protected) {
2790                         if ((sl->sl_flags & SL_WRITE_PROTECTED) == 0) {
2791                                 ua = 1;
2792                                 sl->sl_flags |= SL_WRITE_PROTECTED;
2793                         }
2794                 } else {
2795                         if (sl->sl_flags & SL_WRITE_PROTECTED) {
2796                                 ua = 1;
2797                                 sl->sl_flags &= ~SL_WRITE_PROTECTED;
2798                         }
2799                 }
2800                 for (it = sl->sl_it_list; ua && it != NULL;
2801                     it = it->sbd_it_next) {
2802                         it->sbd_it_ua_conditions |=
2803                             SBD_UA_MODE_PARAMETERS_CHANGED;
2804                 }
2805                 mutex_exit(&sl->sl_lock);
2806         }
2807 
2808         if (mlu->mlu_lu_size_valid) {
2809                 /*
2810                  * validate lu size and set
2811                  * For open file only (registered lu)
2812                  */
2813                 mutex_enter(&sl->sl_lock);
2814                 old_size = sl->sl_lu_size;
2815                 sl->sl_lu_size = mlu->mlu_lu_size;
2816                 mutex_exit(&sl->sl_lock);
2817                 ret = sbd_open_data_file(sl, err_ret, 1, 1, 1);
2818                 if (ret) {
2819                         mutex_enter(&sl->sl_lock);
2820                         sl->sl_lu_size = old_size;
2821                         mutex_exit(&sl->sl_lock);
2822                         goto smm_err_out;
2823                 }
2824                 if (old_size != mlu->mlu_lu_size) {
2825                         mutex_enter(&sl->sl_lock);
2826                         for (it = sl->sl_it_list; it != NULL;
2827                             it = it->sbd_it_next) {
2828                                 it->sbd_it_ua_conditions |=
2829                                     SBD_UA_CAPACITY_CHANGED;
2830                         }
2831                         mutex_exit(&sl->sl_lock);
2832                 }
2833         }
2834 
2835         if (sbd_write_lu_info(sl) != SBD_SUCCESS) {
2836                 *err_ret = SBD_RET_META_CREATION_FAILED;
2837                 ret = EIO;
2838         }
2839 
2840 smm_err_out:
2841         if (modify_unregistered) {
2842                 (void) sbd_close_delete_lu(sl, 0);
2843         } else {
2844                 sl->sl_trans_op = SL_OP_NONE;
2845         }
2846         return (ret);
2847 }
2848 
2849 int
2850 sbd_set_global_props(sbd_global_props_t *mlu, int struct_sz,
2851     uint32_t *err_ret)
2852 {
2853         sbd_lu_t *sl = NULL;
2854         int ret = 0;
2855         sbd_it_data_t *it;
2856         uint32_t sz;
2857 
2858         sz = struct_sz - sizeof (*mlu) + 8 + 1;
2859 
2860         /* if there is data in the buf, null terminate it */
2861         if (struct_sz > sizeof (*mlu)) {
2862                 mlu->mlu_buf[struct_sz - sizeof (*mlu) + 8 - 1] = 0;
2863         }
2864 
2865         *err_ret = 0;
2866 
2867         /* Lets validate offsets */
2868         if (((mlu->mlu_mgmt_url_valid) &&
2869             (mlu->mlu_mgmt_url_off >= sz))) {
2870                 return (EINVAL);
2871         }
2872 
2873         if (mlu->mlu_mgmt_url_valid) {
2874                 uint16_t url_sz;
2875 
2876                 url_sz = strlen((char *)mlu->mlu_buf + mlu->mlu_mgmt_url_off);
2877                 if (url_sz > 0)
2878                         url_sz++;
2879 
2880                 rw_enter(&sbd_global_prop_lock, RW_WRITER);
2881                 if (sbd_mgmt_url_alloc_size > 0 &&
2882                     (url_sz == 0 || sbd_mgmt_url_alloc_size < url_sz)) {
2883                         kmem_free(sbd_mgmt_url, sbd_mgmt_url_alloc_size);
2884                         sbd_mgmt_url = NULL;
2885                         sbd_mgmt_url_alloc_size = 0;
2886                 }
2887                 if (url_sz > 0) {
2888                         if (sbd_mgmt_url_alloc_size == 0) {
2889                                 sbd_mgmt_url = kmem_alloc(url_sz, KM_SLEEP);
2890                                 sbd_mgmt_url_alloc_size = url_sz;
2891                         }
2892                         (void) strcpy(sbd_mgmt_url, (char *)mlu->mlu_buf +
2893                             mlu->mlu_mgmt_url_off);
2894                 }
2895                 /*
2896                  * check each lu to determine whether a UA is needed.
2897                  */
2898                 mutex_enter(&sbd_lock);
2899                 for (sl = sbd_lu_list; sl; sl = sl->sl_next) {
2900                         if (sl->sl_mgmt_url) {
2901                                 continue;
2902                         }
2903                         mutex_enter(&sl->sl_lock);
2904                         for (it = sl->sl_it_list; it != NULL;
2905                             it = it->sbd_it_next) {
2906                                 it->sbd_it_ua_conditions |=
2907                                     SBD_UA_MODE_PARAMETERS_CHANGED;
2908                         }
2909                         mutex_exit(&sl->sl_lock);
2910                 }
2911                 mutex_exit(&sbd_lock);
2912                 rw_exit(&sbd_global_prop_lock);
2913         }
2914         return (ret);
2915 }
2916 
2917 /* ARGSUSED */
2918 int
2919 sbd_delete_locked_lu(sbd_lu_t *sl, uint32_t *err_ret,
2920     stmf_state_change_info_t *ssi)
2921 {
2922         int i;
2923         stmf_status_t ret;
2924 
2925         if ((sl->sl_state == STMF_STATE_OFFLINE) &&
2926             !sl->sl_state_not_acked) {
2927                 goto sdl_do_dereg;
2928         }
2929 
2930         if ((sl->sl_state != STMF_STATE_ONLINE) ||
2931             sl->sl_state_not_acked) {
2932                 return (EBUSY);
2933         }
2934 
2935         ret = stmf_ctl(STMF_CMD_LU_OFFLINE, sl->sl_lu, ssi);
2936         if ((ret != STMF_SUCCESS) && (ret != STMF_ALREADY)) {
2937                 return (EBUSY);
2938         }
2939 
2940         for (i = 0; i < 500; i++) {
2941                 if ((sl->sl_state == STMF_STATE_OFFLINE) &&
2942                     !sl->sl_state_not_acked) {
2943                         goto sdl_do_dereg;
2944                 }
2945                 delay(drv_usectohz(10000));
2946         }
2947         return (EBUSY);
2948 
2949 sdl_do_dereg:;
2950         if (stmf_deregister_lu(sl->sl_lu) != STMF_SUCCESS)
2951                 return (EBUSY);
2952         atomic_dec_32(&sbd_lu_count);
2953 
2954         return (sbd_close_delete_lu(sl, 0));
2955 }
2956 
2957 int
2958 sbd_delete_lu(sbd_delete_lu_t *dlu, int struct_sz, uint32_t *err_ret)
2959 {
2960         sbd_lu_t *sl;
2961         sbd_status_t sret;
2962         stmf_state_change_info_t ssi;
2963         int ret;
2964 
2965         if (dlu->dlu_by_meta_name) {
2966                 ((char *)dlu)[struct_sz - 1] = 0;
2967                 sret = sbd_find_and_lock_lu(NULL, dlu->dlu_meta_name,
2968                     SL_OP_DELETE_LU, &sl);
2969         } else {
2970                 sret = sbd_find_and_lock_lu(dlu->dlu_guid, NULL,
2971                     SL_OP_DELETE_LU, &sl);
2972         }
2973         if (sret != SBD_SUCCESS) {
2974                 if (sret == SBD_BUSY) {
2975                         *err_ret = SBD_RET_LU_BUSY;
2976                         return (EBUSY);
2977                 } else if (sret == SBD_NOT_FOUND) {
2978                         *err_ret = SBD_RET_NOT_FOUND;
2979                         return (ENOENT);
2980                 }
2981                 return (EIO);
2982         }
2983 
2984         ssi.st_rflags = STMF_RFLAG_USER_REQUEST;
2985         ssi.st_additional_info = "sbd_delete_lu call (ioctl)";
2986         ret = sbd_delete_locked_lu(sl, err_ret, &ssi);
2987 
2988         if (ret) {
2989                 /* Once its locked, no need to grab mutex again */
2990                 sl->sl_trans_op = SL_OP_NONE;
2991         }
2992         return (ret);
2993 }
2994 
2995 sbd_status_t
2996 sbd_data_read(sbd_lu_t *sl, struct scsi_task *task,
2997     uint64_t offset, uint64_t size, uint8_t *buf)
2998 {
2999         int ret;
3000         long resid;
3001 
3002         if ((offset + size) > sl->sl_lu_size) {
3003                 return (SBD_IO_PAST_EOF);
3004         }
3005 
3006         offset += sl->sl_data_offset;
3007 
3008         if ((offset + size) > sl->sl_data_readable_size) {
3009                 uint64_t store_end;
3010                 if (offset > sl->sl_data_readable_size) {
3011                         bzero(buf, size);
3012                         return (SBD_SUCCESS);
3013                 }
3014                 store_end = sl->sl_data_readable_size - offset;
3015                 bzero(buf + store_end, size - store_end);
3016                 size = store_end;
3017         }
3018 
3019         DTRACE_PROBE5(backing__store__read__start, sbd_lu_t *, sl,
3020             uint8_t *, buf, uint64_t, size, uint64_t, offset,
3021             scsi_task_t *, task);
3022 
3023         /*
3024          * Don't proceed if the device has been closed
3025          * This can occur on an access state change to standby or
3026          * a delete. The writer lock is acquired before closing the
3027          * lu.
3028          */
3029         rw_enter(&sl->sl_access_state_lock, RW_READER);
3030         if ((sl->sl_flags & SL_MEDIA_LOADED) == 0) {
3031                 rw_exit(&sl->sl_access_state_lock);
3032                 return (SBD_FAILURE);
3033         }
3034         ret = vn_rdwr(UIO_READ, sl->sl_data_vp, (caddr_t)buf, (ssize_t)size,
3035             (offset_t)offset, UIO_SYSSPACE, 0, RLIM64_INFINITY, CRED(),
3036             &resid);
3037         rw_exit(&sl->sl_access_state_lock);
3038 
3039         DTRACE_PROBE6(backing__store__read__end, sbd_lu_t *, sl,
3040             uint8_t *, buf, uint64_t, size, uint64_t, offset,
3041             int, ret, scsi_task_t *, task);
3042 
3043 over_sl_data_read:
3044         if (ret || resid) {
3045                 stmf_trace(0, "UIO_READ failed, ret = %d, resid = %d", ret,
3046                     resid);
3047                 return (SBD_FAILURE);
3048         }
3049 
3050         return (SBD_SUCCESS);
3051 }
3052 
3053 sbd_status_t
3054 sbd_data_write(sbd_lu_t *sl, struct scsi_task *task,
3055     uint64_t offset, uint64_t size, uint8_t *buf)
3056 {
3057         int ret;
3058         long resid;
3059         sbd_status_t sret = SBD_SUCCESS;
3060         int ioflag;
3061 
3062         if ((offset + size) > sl->sl_lu_size) {
3063                 return (SBD_IO_PAST_EOF);
3064         }
3065 
3066         offset += sl->sl_data_offset;
3067 
3068         if ((sl->sl_flags & SL_WRITEBACK_CACHE_DISABLE) &&
3069             (sl->sl_flags & SL_FLUSH_ON_DISABLED_WRITECACHE)) {
3070                 ioflag = FSYNC;
3071         } else {
3072                 ioflag = 0;
3073         }
3074 
3075         DTRACE_PROBE5(backing__store__write__start, sbd_lu_t *, sl,
3076             uint8_t *, buf, uint64_t, size, uint64_t, offset,
3077             scsi_task_t *, task);
3078 
3079         /*
3080          * Don't proceed if the device has been closed
3081          * This can occur on an access state change to standby or
3082          * a delete. The writer lock is acquired before closing the
3083          * lu.
3084          */
3085         rw_enter(&sl->sl_access_state_lock, RW_READER);
3086         if ((sl->sl_flags & SL_MEDIA_LOADED) == 0) {
3087                 rw_exit(&sl->sl_access_state_lock);
3088                 return (SBD_FAILURE);
3089         }
3090         ret = vn_rdwr(UIO_WRITE, sl->sl_data_vp, (caddr_t)buf, (ssize_t)size,
3091             (offset_t)offset, UIO_SYSSPACE, ioflag, RLIM64_INFINITY, CRED(),
3092             &resid);
3093         rw_exit(&sl->sl_access_state_lock);
3094 
3095         DTRACE_PROBE6(backing__store__write__end, sbd_lu_t *, sl,
3096             uint8_t *, buf, uint64_t, size, uint64_t, offset,
3097             int, ret, scsi_task_t *, task);
3098 
3099         if ((ret == 0) && (resid == 0) &&
3100             (sl->sl_flags & SL_WRITEBACK_CACHE_DISABLE) &&
3101             (sl->sl_flags & SL_FLUSH_ON_DISABLED_WRITECACHE)) {
3102                 sret = sbd_flush_data_cache(sl, 1);
3103         }
3104 over_sl_data_write:
3105 
3106         if ((ret || resid) || (sret != SBD_SUCCESS)) {
3107                 return (SBD_FAILURE);
3108         } else if ((offset + size) > sl->sl_data_readable_size) {
3109                 uint64_t old_size, new_size;
3110 
3111                 do {
3112                         old_size = sl->sl_data_readable_size;
3113                         if ((offset + size) <= old_size)
3114                                 break;
3115                         new_size = offset + size;
3116                 } while (atomic_cas_64(&sl->sl_data_readable_size, old_size,
3117                     new_size) != old_size);
3118         }
3119 
3120         return (SBD_SUCCESS);
3121 }
3122 
3123 int
3124 sbd_get_global_props(sbd_global_props_t *oslp, uint32_t oslp_sz,
3125     uint32_t *err_ret)
3126 {
3127         uint32_t sz = 0;
3128         uint16_t off;
3129 
3130         rw_enter(&sbd_global_prop_lock, RW_READER);
3131         if (sbd_mgmt_url) {
3132                 sz += strlen(sbd_mgmt_url) + 1;
3133         }
3134         bzero(oslp, sizeof (*oslp) - 8);
3135         oslp->mlu_buf_size_needed = sz;
3136 
3137         if (sz > (oslp_sz - sizeof (*oslp) + 8)) {
3138                 *err_ret = SBD_RET_INSUFFICIENT_BUF_SPACE;
3139                 rw_exit(&sbd_global_prop_lock);
3140                 return (ENOMEM);
3141         }
3142 
3143         off = 0;
3144         if (sbd_mgmt_url) {
3145                 oslp->mlu_mgmt_url_valid = 1;
3146                 oslp->mlu_mgmt_url_off = off;
3147                 (void) strcpy((char *)&oslp->mlu_buf[off], sbd_mgmt_url);
3148                 off += strlen(sbd_mgmt_url) + 1;
3149         }
3150 
3151         rw_exit(&sbd_global_prop_lock);
3152         return (0);
3153 }
3154 
3155 static int
3156 sbd_get_unmap_props(sbd_unmap_props_t *sup,
3157     sbd_unmap_props_t *osup, uint32_t *err_ret)
3158 {
3159         sbd_status_t sret;
3160         sbd_lu_t *sl = NULL;
3161 
3162         if (sup->sup_guid_valid) {
3163                 sret = sbd_find_and_lock_lu(sup->sup_guid,
3164                     NULL, SL_OP_LU_PROPS, &sl);
3165         } else {
3166                 sret = sbd_find_and_lock_lu(NULL,
3167                     (uint8_t *)sup->sup_zvol_path, SL_OP_LU_PROPS,
3168                     &sl);
3169         }
3170         if (sret != SBD_SUCCESS) {
3171                 if (sret == SBD_BUSY) {
3172                         *err_ret = SBD_RET_LU_BUSY;
3173                         return (EBUSY);
3174                 } else if (sret == SBD_NOT_FOUND) {
3175                         *err_ret = SBD_RET_NOT_FOUND;
3176                         return (ENOENT);
3177                 }
3178                 return (EIO);
3179         }
3180 
3181         sup->sup_found_lu = 1;
3182         sup->sup_guid_valid = 1;
3183         bcopy(sl->sl_device_id + 4, sup->sup_guid, 16);
3184         if (sl->sl_flags & SL_UNMAP_ENABLED)
3185                 sup->sup_unmap_enabled = 1;
3186         else
3187                 sup->sup_unmap_enabled = 0;
3188 
3189         *osup = *sup;
3190         sl->sl_trans_op = SL_OP_NONE;
3191 
3192         return (0);
3193 }
3194 
3195 int
3196 sbd_get_lu_props(sbd_lu_props_t *islp, uint32_t islp_sz,
3197     sbd_lu_props_t *oslp, uint32_t oslp_sz, uint32_t *err_ret)
3198 {
3199         sbd_status_t sret;
3200         sbd_lu_t *sl = NULL;
3201         uint32_t sz;
3202         uint16_t off;
3203 
3204         if (islp->slp_input_guid) {
3205                 sret = sbd_find_and_lock_lu(islp->slp_guid, NULL,
3206                     SL_OP_LU_PROPS, &sl);
3207         } else {
3208                 ((char *)islp)[islp_sz - 1] = 0;
3209                 sret = sbd_find_and_lock_lu(NULL, islp->slp_buf,
3210                     SL_OP_LU_PROPS, &sl);
3211         }
3212         if (sret != SBD_SUCCESS) {
3213                 if (sret == SBD_BUSY) {
3214                         *err_ret = SBD_RET_LU_BUSY;
3215                         return (EBUSY);
3216                 } else if (sret == SBD_NOT_FOUND) {
3217                         *err_ret = SBD_RET_NOT_FOUND;
3218                         return (ENOENT);
3219                 }
3220                 return (EIO);
3221         }
3222 
3223         sz = strlen(sl->sl_name) + 1;
3224         if ((sl->sl_flags & (SL_ZFS_META | SL_SHARED_META)) == 0) {
3225                 if (sl->sl_data_filename) {
3226                         sz += strlen(sl->sl_data_filename) + 1;
3227                 }
3228         }
3229         sz += sl->sl_serial_no_size;
3230         if (sl->sl_alias) {
3231                 sz += strlen(sl->sl_alias) + 1;
3232         }
3233 
3234         rw_enter(&sbd_global_prop_lock, RW_READER);
3235         if (sl->sl_mgmt_url) {
3236                 sz += strlen(sl->sl_mgmt_url) + 1;
3237         } else if (sbd_mgmt_url) {
3238                 sz += strlen(sbd_mgmt_url) + 1;
3239         }
3240         bzero(oslp, sizeof (*oslp) - 8);
3241         oslp->slp_buf_size_needed = sz;
3242 
3243         if (sz > (oslp_sz - sizeof (*oslp) + 8)) {
3244                 sl->sl_trans_op = SL_OP_NONE;
3245                 *err_ret = SBD_RET_INSUFFICIENT_BUF_SPACE;
3246                 rw_exit(&sbd_global_prop_lock);
3247                 return (ENOMEM);
3248         }
3249 
3250         off = 0;
3251         (void) strcpy((char *)oslp->slp_buf, sl->sl_name);
3252         oslp->slp_meta_fname_off = off;
3253         off += strlen(sl->sl_name) + 1;
3254         if ((sl->sl_flags & (SL_ZFS_META | SL_SHARED_META)) == 0) {
3255                 oslp->slp_meta_fname_valid = 1;
3256                 oslp->slp_separate_meta = 1;
3257                 if (sl->sl_data_filename) {
3258                         oslp->slp_data_fname_valid = 1;
3259                         oslp->slp_data_fname_off = off;
3260                         (void) strcpy((char *)&oslp->slp_buf[off],
3261                             sl->sl_data_filename);
3262                         off += strlen(sl->sl_data_filename) + 1;
3263                 }
3264         } else {
3265                 oslp->slp_data_fname_valid = 1;
3266                 oslp->slp_data_fname_off = oslp->slp_meta_fname_off;
3267                 if (sl->sl_flags & SL_ZFS_META) {
3268                         oslp->slp_zfs_meta = 1;
3269                 }
3270         }
3271         if (sl->sl_alias) {
3272                 oslp->slp_alias_valid = 1;
3273                 oslp->slp_alias_off = off;
3274                 (void) strcpy((char *)&oslp->slp_buf[off], sl->sl_alias);
3275                 off += strlen(sl->sl_alias) + 1;
3276         }
3277         if (sl->sl_mgmt_url) {
3278                 oslp->slp_mgmt_url_valid = 1;
3279                 oslp->slp_mgmt_url_off = off;
3280                 (void) strcpy((char *)&oslp->slp_buf[off], sl->sl_mgmt_url);
3281                 off += strlen(sl->sl_mgmt_url) + 1;
3282         } else if (sbd_mgmt_url) {
3283                 oslp->slp_mgmt_url_valid = 1;
3284                 oslp->slp_mgmt_url_off = off;
3285                 (void) strcpy((char *)&oslp->slp_buf[off], sbd_mgmt_url);
3286                 off += strlen(sbd_mgmt_url) + 1;
3287         }
3288         if (sl->sl_serial_no_size) {
3289                 oslp->slp_serial_off = off;
3290                 bcopy(sl->sl_serial_no, &oslp->slp_buf[off],
3291                     sl->sl_serial_no_size);
3292                 oslp->slp_serial_size = sl->sl_serial_no_size;
3293                 oslp->slp_serial_valid = 1;
3294                 off += sl->sl_serial_no_size;
3295         }
3296 
3297         oslp->slp_lu_size = sl->sl_lu_size;
3298         oslp->slp_blksize = ((uint16_t)1) << sl->sl_data_blocksize_shift;
3299 
3300         oslp->slp_access_state = sl->sl_access_state;
3301 
3302         if (sl->sl_flags & SL_VID_VALID) {
3303                 oslp->slp_lu_vid = 1;
3304                 bcopy(sl->sl_vendor_id, oslp->slp_vid, 8);
3305         } else {
3306                 bcopy(sbd_vendor_id, oslp->slp_vid, 8);
3307         }
3308         if (sl->sl_flags & SL_PID_VALID) {
3309                 oslp->slp_lu_pid = 1;
3310                 bcopy(sl->sl_product_id, oslp->slp_pid, 16);
3311         } else {
3312                 bcopy(sbd_product_id, oslp->slp_pid, 16);
3313         }
3314         if (sl->sl_flags & SL_REV_VALID) {
3315                 oslp->slp_lu_rev = 1;
3316                 bcopy(sl->sl_revision, oslp->slp_rev, 4);
3317         } else {
3318                 bcopy(sbd_revision, oslp->slp_rev, 4);
3319         }
3320         bcopy(sl->sl_device_id + 4, oslp->slp_guid, 16);
3321 
3322         if (sl->sl_flags & SL_WRITEBACK_CACHE_DISABLE)
3323                 oslp->slp_writeback_cache_disable_cur = 1;
3324         if (sl->sl_flags & SL_SAVED_WRITE_CACHE_DISABLE)
3325                 oslp->slp_writeback_cache_disable_saved = 1;
3326         if (sl->sl_flags & SL_WRITE_PROTECTED)
3327                 oslp->slp_write_protected = 1;
3328 
3329         sl->sl_trans_op = SL_OP_NONE;
3330 
3331         rw_exit(&sbd_global_prop_lock);
3332         return (0);
3333 }
3334 
3335 /*
3336  * Returns an allocated string with the "<pool>/..." form of the zvol name.
3337  */
3338 static char *
3339 sbd_get_zvol_name(sbd_lu_t *sl)
3340 {
3341         char *src;
3342         char *p;
3343 
3344         if (sl->sl_data_filename)
3345                 src = sl->sl_data_filename;
3346         else
3347                 src = sl->sl_meta_filename;
3348         /* There has to be a better way */
3349         if (SBD_IS_ZVOL(src) != 0) {
3350                 ASSERT(0);
3351         }
3352         src += 14;      /* Past /dev/zvol/dsk/ */
3353         if (*src == '/')
3354                 src++;  /* or /dev/zvol/rdsk/ */
3355         p = (char *)kmem_alloc(strlen(src) + 1, KM_SLEEP);
3356         (void) strcpy(p, src);
3357         return (p);
3358 }
3359 
3360 /*
3361  * this function creates a local metadata zvol property
3362  */
3363 sbd_status_t
3364 sbd_create_zfs_meta_object(sbd_lu_t *sl)
3365 {
3366         /*
3367          * -allocate 1/2 the property size, the zfs property
3368          *  is 8k in size and stored as ascii hex string, all
3369          *  we needed is 4k buffer to store the binary data.
3370          * -initialize reader/write lock
3371          */
3372         sl->sl_zfs_meta = kmem_zalloc(ZAP_MAXVALUELEN / 2, KM_SLEEP);
3373         rw_init(&sl->sl_zfs_meta_lock, NULL, RW_DRIVER, NULL);
3374         return (SBD_SUCCESS);
3375 }
3376 
3377 char
3378 sbd_ctoi(char c)
3379 {
3380         if ((c >= '0') && (c <= '9'))
3381                 c -= '0';
3382         else if ((c >= 'A') && (c <= 'F'))
3383                 c = c - 'A' + 10;
3384         else if ((c >= 'a') && (c <= 'f'))
3385                 c = c - 'a' + 10;
3386         else
3387                 c = -1;
3388         return (c);
3389 }
3390 
3391 /*
3392  * read zvol property and convert to binary
3393  */
3394 sbd_status_t
3395 sbd_open_zfs_meta(sbd_lu_t *sl)
3396 {
3397         char            *meta = NULL, cl, ch;
3398         int             i;
3399         char            *tmp, *ptr;
3400         uint64_t        rc = SBD_SUCCESS;
3401         int             len;
3402         char            *file;
3403 
3404         if (sl->sl_zfs_meta == NULL) {
3405                 if (sbd_create_zfs_meta_object(sl) == SBD_FAILURE)
3406                         return (SBD_FAILURE);
3407         } else {
3408                 bzero(sl->sl_zfs_meta, (ZAP_MAXVALUELEN / 2));
3409         }
3410 
3411         rw_enter(&sl->sl_zfs_meta_lock, RW_WRITER);
3412         file = sbd_get_zvol_name(sl);
3413         if (sbd_zvolget(file, &meta)) {
3414                 rc = SBD_FAILURE;
3415                 goto done;
3416         }
3417         tmp = meta;
3418         /* convert ascii hex to binary meta */
3419         len = strlen(meta);
3420         ptr = sl->sl_zfs_meta;
3421         for (i = 0; i < len; i += 2) {
3422                 ch = sbd_ctoi(*tmp++);
3423                 cl = sbd_ctoi(*tmp++);
3424                 if (ch == -1 || cl == -1) {
3425                         rc = SBD_FAILURE;
3426                         break;
3427                 }
3428                 *ptr++ = (ch << 4) + cl;
3429         }
3430 done:
3431         rw_exit(&sl->sl_zfs_meta_lock);
3432         if (meta)
3433                 kmem_free(meta, len + 1);
3434         kmem_free(file, strlen(file) + 1);
3435         return (rc);
3436 }
3437 
3438 sbd_status_t
3439 sbd_read_zfs_meta(sbd_lu_t *sl, uint8_t *buf, uint64_t sz, uint64_t off)
3440 {
3441         ASSERT(sl->sl_zfs_meta);
3442         rw_enter(&sl->sl_zfs_meta_lock, RW_READER);
3443         bcopy(&sl->sl_zfs_meta[off], buf, sz);
3444         rw_exit(&sl->sl_zfs_meta_lock);
3445         return (SBD_SUCCESS);
3446 }
3447 
3448 sbd_status_t
3449 sbd_write_zfs_meta(sbd_lu_t *sl, uint8_t *buf, uint64_t sz, uint64_t off)
3450 {
3451         ASSERT(sl->sl_zfs_meta);
3452         if ((off + sz) > (ZAP_MAXVALUELEN / 2 - 1)) {
3453                 return (SBD_META_CORRUPTED);
3454         }
3455         if ((off + sz) > sl->sl_meta_size_used) {
3456                 sl->sl_meta_size_used = off + sz;
3457                 if (sl->sl_total_meta_size < sl->sl_meta_size_used) {
3458                         uint64_t meta_align =
3459                             (((uint64_t)1) << sl->sl_meta_blocksize_shift) - 1;
3460                         sl->sl_total_meta_size = (sl->sl_meta_size_used +
3461                             meta_align) & (~meta_align);
3462                 }
3463         }
3464         rw_enter(&sl->sl_zfs_meta_lock, RW_WRITER);
3465         bcopy(buf, &sl->sl_zfs_meta[off], sz);
3466         rw_exit(&sl->sl_zfs_meta_lock);
3467         /*
3468          * During creation of a logical unit, sbd_update_zfs_prop will be
3469          * called separately to avoid multiple calls as each meta section
3470          * create/update will result in a call to sbd_write_zfs_meta().
3471          * We only need to update the zvol once during create.
3472          */
3473         mutex_enter(&sl->sl_lock);
3474         if (sl->sl_trans_op != SL_OP_CREATE_REGISTER_LU) {
3475                 mutex_exit(&sl->sl_lock);
3476                 return (sbd_update_zfs_prop(sl));
3477         }
3478         mutex_exit(&sl->sl_lock);
3479         return (SBD_SUCCESS);
3480 }
3481 
3482 sbd_status_t
3483 sbd_update_zfs_prop(sbd_lu_t *sl)
3484 {
3485         char    *ptr, *ah_meta;
3486         char    *dp = NULL;
3487         int     i, num;
3488         char    *file;
3489         sbd_status_t ret = SBD_SUCCESS;
3490 
3491         ASSERT(sl->sl_zfs_meta);
3492         ptr = ah_meta = kmem_zalloc(ZAP_MAXVALUELEN, KM_SLEEP);
3493         rw_enter(&sl->sl_zfs_meta_lock, RW_READER);
3494         /* convert local copy to ascii hex */
3495         dp = sl->sl_zfs_meta;
3496         for (i = 0; i < sl->sl_total_meta_size; i++, dp++) {
3497                 num = ((*dp) >> 4) & 0xF;
3498                 *ah_meta++ = (num < 10) ? (num + '0') : (num + ('a' - 10));
3499                 num = (*dp) & 0xF;
3500                 *ah_meta++ = (num < 10) ? (num + '0') : (num + ('a' - 10));
3501         }
3502         *ah_meta = NULL;
3503         file = sbd_get_zvol_name(sl);
3504         if (sbd_zvolset(file, (char *)ptr)) {
3505                 ret = SBD_META_CORRUPTED;
3506         }
3507         rw_exit(&sl->sl_zfs_meta_lock);
3508         kmem_free(ptr, ZAP_MAXVALUELEN);
3509         kmem_free(file, strlen(file) + 1);
3510         return (ret);
3511 }
3512 
3513 int
3514 sbd_is_zvol(char *path)
3515 {
3516         int is_zfs = 0;
3517 
3518         if (SBD_IS_ZVOL(path) == 0)
3519                 is_zfs = 1;
3520 
3521         return (is_zfs);
3522 }
3523 
3524 /*
3525  * set write cache disable
3526  * wcd - 1 = disable, 0 = enable
3527  */
3528 sbd_status_t
3529 sbd_wcd_set(int wcd, sbd_lu_t *sl)
3530 {
3531         /* translate to wce bit */
3532         int wce = wcd ? 0 : 1;
3533         int ret;
3534         sbd_status_t sret = SBD_SUCCESS;
3535 
3536         mutex_enter(&sl->sl_lock);
3537         sl->sl_flags &= ~SL_WRITEBACK_CACHE_SET_UNSUPPORTED;
3538 
3539         if (sl->sl_data_vp->v_type == VREG) {
3540                 sl->sl_flags |= SL_FLUSH_ON_DISABLED_WRITECACHE;
3541                 goto done;
3542         }
3543 
3544         ret = VOP_IOCTL(sl->sl_data_vp, DKIOCSETWCE, (intptr_t)&wce, FKIOCTL,
3545             kcred, NULL, NULL);
3546         if (ret == 0) {
3547                 sl->sl_flags &= ~SL_WRITEBACK_CACHE_SET_UNSUPPORTED;
3548                 sl->sl_flags &= ~SL_FLUSH_ON_DISABLED_WRITECACHE;
3549         } else {
3550                 sl->sl_flags |= SL_WRITEBACK_CACHE_SET_UNSUPPORTED;
3551                 sl->sl_flags |= SL_FLUSH_ON_DISABLED_WRITECACHE;
3552                 sret = SBD_FAILURE;
3553                 goto done;
3554         }
3555 
3556 done:
3557         mutex_exit(&sl->sl_lock);
3558         return (sret);
3559 }
3560 
3561 /*
3562  * get write cache disable
3563  * wcd - 1 = disable, 0 = enable
3564  */
3565 void
3566 sbd_wcd_get(int *wcd, sbd_lu_t *sl)
3567 {
3568         int wce;
3569         int ret;
3570 
3571         if (sl->sl_data_vp->v_type == VREG) {
3572                 *wcd = 0;
3573                 return;
3574         }
3575 
3576         ret = VOP_IOCTL(sl->sl_data_vp, DKIOCGETWCE, (intptr_t)&wce, FKIOCTL,
3577             kcred, NULL, NULL);
3578         /* if write cache get failed, assume disabled */
3579         if (ret) {
3580                 *wcd = 1;
3581         } else {
3582                 /* translate to wcd bit */
3583                 *wcd = wce ? 0 : 1;
3584         }
3585 }
3586 
3587 int
3588 sbd_zvolget(char *zvol_name, char **comstarprop)
3589 {
3590         ldi_handle_t    zfs_lh;
3591         nvlist_t        *nv = NULL, *nv2;
3592         zfs_cmd_t       *zc;
3593         char            *ptr;
3594         int size = 1024;
3595         int unused;
3596         int rc;
3597 
3598         if ((rc = ldi_open_by_name("/dev/zfs", FREAD | FWRITE, kcred,
3599             &zfs_lh, sbd_zfs_ident)) != 0) {
3600                 cmn_err(CE_WARN, "ldi_open %d", rc);
3601                 return (ENXIO);
3602         }
3603 
3604         zc = kmem_zalloc(sizeof (zfs_cmd_t), KM_SLEEP);
3605         (void) strlcpy(zc->zc_name, zvol_name, sizeof (zc->zc_name));
3606 again:
3607         zc->zc_nvlist_dst = (uint64_t)(intptr_t)kmem_alloc(size,
3608             KM_SLEEP);
3609         zc->zc_nvlist_dst_size = size;
3610         rc = ldi_ioctl(zfs_lh, ZFS_IOC_OBJSET_STATS, (intptr_t)zc,
3611             FKIOCTL, kcred, &unused);
3612         /*
3613          * ENOMEM means the list is larger than what we've allocated
3614          * ldi_ioctl will fail with ENOMEM only once
3615          */
3616         if (rc == ENOMEM) {
3617                 int newsize;
3618                 newsize = zc->zc_nvlist_dst_size;
3619                 kmem_free((void *)(uintptr_t)zc->zc_nvlist_dst, size);
3620                 size = newsize;
3621                 goto again;
3622         } else if (rc != 0) {
3623                 goto out;
3624         }
3625         rc = nvlist_unpack((char *)(uintptr_t)zc->zc_nvlist_dst,
3626             zc->zc_nvlist_dst_size, &nv, 0);
3627         ASSERT(rc == 0);        /* nvlist_unpack should not fail */
3628         if ((rc = nvlist_lookup_nvlist(nv, "stmf_sbd_lu", &nv2)) == 0) {
3629                 rc = nvlist_lookup_string(nv2, ZPROP_VALUE, &ptr);
3630                 if (rc != 0) {
3631                         cmn_err(CE_WARN, "couldn't get value");
3632                 } else {
3633                         *comstarprop = kmem_alloc(strlen(ptr) + 1,
3634                             KM_SLEEP);
3635                         (void) strcpy(*comstarprop, ptr);
3636                 }
3637         }
3638 out:
3639         if (nv != NULL)
3640                 nvlist_free(nv);
3641         kmem_free((void *)(uintptr_t)zc->zc_nvlist_dst, size);
3642         kmem_free(zc, sizeof (zfs_cmd_t));
3643         (void) ldi_close(zfs_lh, FREAD|FWRITE, kcred);
3644 
3645         return (rc);
3646 }
3647 
3648 int
3649 sbd_zvolset(char *zvol_name, char *comstarprop)
3650 {
3651         ldi_handle_t    zfs_lh;
3652         nvlist_t        *nv;
3653         char            *packed = NULL;
3654         size_t          len;
3655         zfs_cmd_t       *zc;
3656         int unused;
3657         int rc;
3658 
3659         if ((rc = ldi_open_by_name("/dev/zfs", FREAD | FWRITE, kcred,
3660             &zfs_lh, sbd_zfs_ident)) != 0) {
3661                 cmn_err(CE_WARN, "ldi_open %d", rc);
3662                 return (ENXIO);
3663         }
3664         (void) nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP);
3665         (void) nvlist_add_string(nv, "stmf_sbd_lu", comstarprop);
3666         if ((rc = nvlist_pack(nv, &packed, &len, NV_ENCODE_NATIVE, KM_SLEEP))) {
3667                 goto out;
3668         }
3669 
3670         zc = kmem_zalloc(sizeof (zfs_cmd_t), KM_SLEEP);
3671         (void) strlcpy(zc->zc_name, zvol_name, sizeof (zc->zc_name));
3672         zc->zc_nvlist_src = (uint64_t)(intptr_t)packed;
3673         zc->zc_nvlist_src_size = len;
3674         rc = ldi_ioctl(zfs_lh, ZFS_IOC_SET_PROP, (intptr_t)zc,
3675             FKIOCTL, kcred, &unused);
3676         if (rc != 0) {
3677                 cmn_err(CE_NOTE, "ioctl failed %d", rc);
3678         }
3679         kmem_free(zc, sizeof (zfs_cmd_t));
3680         if (packed)
3681                 kmem_free(packed, len);
3682 out:
3683         nvlist_free(nv);
3684         (void) ldi_close(zfs_lh, FREAD|FWRITE, kcred);
3685         return (rc);
3686 }
3687 
3688 /*
3689  * Unmap a region in a volume.  Currently only supported for zvols.
3690  */
3691 int
3692 sbd_unmap(sbd_lu_t *sl, uint64_t offset, uint64_t length)
3693 {
3694         vnode_t *vp;
3695         int unused;
3696         dkioc_free_t df;
3697 
3698         /* Right now, we only support UNMAP on zvols. */
3699         if (!(sl->sl_flags & SL_ZFS_META))
3700                 return (EIO);
3701 
3702         df.df_flags = (sl->sl_flags & SL_WRITEBACK_CACHE_DISABLE) ?
3703             DF_WAIT_SYNC : 0;
3704         df.df_start = offset;
3705         df.df_length = length;
3706 
3707         /* Use the data vnode we have to send a fop_ioctl(). */
3708         vp = sl->sl_data_vp;
3709         if (vp == NULL) {
3710                 cmn_err(CE_WARN, "Cannot unmap - no vnode pointer.");
3711                 return (EIO);
3712         }
3713 
3714         return (VOP_IOCTL(vp, DKIOCFREE, (intptr_t)(&df), FKIOCTL, kcred,
3715             &unused, NULL));
3716 }