1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2013 by Delphix. All rights reserved.
  24  * Copyright 2017 Nexenta Systems, Inc.  All rights reserved.
  25  */
  26 
  27 #include <sys/types.h>
  28 #include <sys/param.h>
  29 #include <sys/time.h>
  30 #include <sys/systm.h>
  31 #include <sys/sysmacros.h>
  32 #include <sys/resource.h>
  33 #include <sys/vfs.h>
  34 #include <sys/vnode.h>
  35 #include <sys/sid.h>
  36 #include <sys/file.h>
  37 #include <sys/stat.h>
  38 #include <sys/kmem.h>
  39 #include <sys/cmn_err.h>
  40 #include <sys/errno.h>
  41 #include <sys/unistd.h>
  42 #include <sys/sdt.h>
  43 #include <sys/fs/zfs.h>
  44 #include <sys/mode.h>
  45 #include <sys/policy.h>
  46 #include <sys/zfs_znode.h>
  47 #include <sys/zfs_fuid.h>
  48 #include <sys/zfs_acl.h>
  49 #include <sys/zfs_dir.h>
  50 #include <sys/zfs_vfsops.h>
  51 #include <sys/dmu.h>
  52 #include <sys/dnode.h>
  53 #include <sys/zap.h>
  54 #include <sys/sa.h>
  55 #include "fs/fs_subr.h"
  56 #include <acl/acl_common.h>
  57 
  58 #define ALLOW   ACE_ACCESS_ALLOWED_ACE_TYPE
  59 #define DENY    ACE_ACCESS_DENIED_ACE_TYPE
  60 #define MAX_ACE_TYPE    ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
  61 #define MIN_ACE_TYPE    ALLOW
  62 
  63 #define OWNING_GROUP            (ACE_GROUP|ACE_IDENTIFIER_GROUP)
  64 #define EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
  65     ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
  66 #define EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
  67     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
  68 #define OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
  69     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
  70 
  71 #define ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
  72     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
  73     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
  74     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
  75 
  76 #define WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
  77 #define WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
  78     ACE_DELETE|ACE_DELETE_CHILD)
  79 #define WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
  80 
  81 #define OGE_CLEAR       (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
  82     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
  83 
  84 #define OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
  85     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
  86 
  87 #define ALL_INHERIT     (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
  88     ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
  89 
  90 #define RESTRICTED_CLEAR        (ACE_WRITE_ACL|ACE_WRITE_OWNER)
  91 
  92 #define V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
  93     ZFS_ACL_PROTECTED)
  94 
  95 #define ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
  96     ZFS_ACL_OBJ_ACE)
  97 
  98 #define ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
  99 
 100 static uint16_t
 101 zfs_ace_v0_get_type(void *acep)
 102 {
 103         return (((zfs_oldace_t *)acep)->z_type);
 104 }
 105 
 106 static uint16_t
 107 zfs_ace_v0_get_flags(void *acep)
 108 {
 109         return (((zfs_oldace_t *)acep)->z_flags);
 110 }
 111 
 112 static uint32_t
 113 zfs_ace_v0_get_mask(void *acep)
 114 {
 115         return (((zfs_oldace_t *)acep)->z_access_mask);
 116 }
 117 
 118 static uint64_t
 119 zfs_ace_v0_get_who(void *acep)
 120 {
 121         return (((zfs_oldace_t *)acep)->z_fuid);
 122 }
 123 
 124 static void
 125 zfs_ace_v0_set_type(void *acep, uint16_t type)
 126 {
 127         ((zfs_oldace_t *)acep)->z_type = type;
 128 }
 129 
 130 static void
 131 zfs_ace_v0_set_flags(void *acep, uint16_t flags)
 132 {
 133         ((zfs_oldace_t *)acep)->z_flags = flags;
 134 }
 135 
 136 static void
 137 zfs_ace_v0_set_mask(void *acep, uint32_t mask)
 138 {
 139         ((zfs_oldace_t *)acep)->z_access_mask = mask;
 140 }
 141 
 142 static void
 143 zfs_ace_v0_set_who(void *acep, uint64_t who)
 144 {
 145         ((zfs_oldace_t *)acep)->z_fuid = who;
 146 }
 147 
 148 /*ARGSUSED*/
 149 static size_t
 150 zfs_ace_v0_size(void *acep)
 151 {
 152         return (sizeof (zfs_oldace_t));
 153 }
 154 
 155 static size_t
 156 zfs_ace_v0_abstract_size(void)
 157 {
 158         return (sizeof (zfs_oldace_t));
 159 }
 160 
 161 static int
 162 zfs_ace_v0_mask_off(void)
 163 {
 164         return (offsetof(zfs_oldace_t, z_access_mask));
 165 }
 166 
 167 /*ARGSUSED*/
 168 static int
 169 zfs_ace_v0_data(void *acep, void **datap)
 170 {
 171         *datap = NULL;
 172         return (0);
 173 }
 174 
 175 static acl_ops_t zfs_acl_v0_ops = {
 176         zfs_ace_v0_get_mask,
 177         zfs_ace_v0_set_mask,
 178         zfs_ace_v0_get_flags,
 179         zfs_ace_v0_set_flags,
 180         zfs_ace_v0_get_type,
 181         zfs_ace_v0_set_type,
 182         zfs_ace_v0_get_who,
 183         zfs_ace_v0_set_who,
 184         zfs_ace_v0_size,
 185         zfs_ace_v0_abstract_size,
 186         zfs_ace_v0_mask_off,
 187         zfs_ace_v0_data
 188 };
 189 
 190 static uint16_t
 191 zfs_ace_fuid_get_type(void *acep)
 192 {
 193         return (((zfs_ace_hdr_t *)acep)->z_type);
 194 }
 195 
 196 static uint16_t
 197 zfs_ace_fuid_get_flags(void *acep)
 198 {
 199         return (((zfs_ace_hdr_t *)acep)->z_flags);
 200 }
 201 
 202 static uint32_t
 203 zfs_ace_fuid_get_mask(void *acep)
 204 {
 205         return (((zfs_ace_hdr_t *)acep)->z_access_mask);
 206 }
 207 
 208 static uint64_t
 209 zfs_ace_fuid_get_who(void *args)
 210 {
 211         uint16_t entry_type;
 212         zfs_ace_t *acep = args;
 213 
 214         entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
 215 
 216         if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
 217             entry_type == ACE_EVERYONE)
 218                 return (-1);
 219         return (((zfs_ace_t *)acep)->z_fuid);
 220 }
 221 
 222 static void
 223 zfs_ace_fuid_set_type(void *acep, uint16_t type)
 224 {
 225         ((zfs_ace_hdr_t *)acep)->z_type = type;
 226 }
 227 
 228 static void
 229 zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
 230 {
 231         ((zfs_ace_hdr_t *)acep)->z_flags = flags;
 232 }
 233 
 234 static void
 235 zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
 236 {
 237         ((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
 238 }
 239 
 240 static void
 241 zfs_ace_fuid_set_who(void *arg, uint64_t who)
 242 {
 243         zfs_ace_t *acep = arg;
 244 
 245         uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
 246 
 247         if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
 248             entry_type == ACE_EVERYONE)
 249                 return;
 250         acep->z_fuid = who;
 251 }
 252 
 253 static size_t
 254 zfs_ace_fuid_size(void *acep)
 255 {
 256         zfs_ace_hdr_t *zacep = acep;
 257         uint16_t entry_type;
 258 
 259         switch (zacep->z_type) {
 260         case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
 261         case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
 262         case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
 263         case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
 264                 return (sizeof (zfs_object_ace_t));
 265         case ALLOW:
 266         case DENY:
 267                 entry_type =
 268                     (((zfs_ace_hdr_t *)acep)->z_flags & ACE_TYPE_FLAGS);
 269                 if (entry_type == ACE_OWNER ||
 270                     entry_type == OWNING_GROUP ||
 271                     entry_type == ACE_EVERYONE)
 272                         return (sizeof (zfs_ace_hdr_t));
 273                 /*FALLTHROUGH*/
 274         default:
 275                 return (sizeof (zfs_ace_t));
 276         }
 277 }
 278 
 279 static size_t
 280 zfs_ace_fuid_abstract_size(void)
 281 {
 282         return (sizeof (zfs_ace_hdr_t));
 283 }
 284 
 285 static int
 286 zfs_ace_fuid_mask_off(void)
 287 {
 288         return (offsetof(zfs_ace_hdr_t, z_access_mask));
 289 }
 290 
 291 static int
 292 zfs_ace_fuid_data(void *acep, void **datap)
 293 {
 294         zfs_ace_t *zacep = acep;
 295         zfs_object_ace_t *zobjp;
 296 
 297         switch (zacep->z_hdr.z_type) {
 298         case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
 299         case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
 300         case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
 301         case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
 302                 zobjp = acep;
 303                 *datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
 304                 return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
 305         default:
 306                 *datap = NULL;
 307                 return (0);
 308         }
 309 }
 310 
 311 static acl_ops_t zfs_acl_fuid_ops = {
 312         zfs_ace_fuid_get_mask,
 313         zfs_ace_fuid_set_mask,
 314         zfs_ace_fuid_get_flags,
 315         zfs_ace_fuid_set_flags,
 316         zfs_ace_fuid_get_type,
 317         zfs_ace_fuid_set_type,
 318         zfs_ace_fuid_get_who,
 319         zfs_ace_fuid_set_who,
 320         zfs_ace_fuid_size,
 321         zfs_ace_fuid_abstract_size,
 322         zfs_ace_fuid_mask_off,
 323         zfs_ace_fuid_data
 324 };
 325 
 326 /*
 327  * The following three functions are provided for compatibility with
 328  * older ZPL version in order to determine if the file use to have
 329  * an external ACL and what version of ACL previously existed on the
 330  * file.  Would really be nice to not need this, sigh.
 331  */
 332 uint64_t
 333 zfs_external_acl(znode_t *zp)
 334 {
 335         zfs_acl_phys_t acl_phys;
 336         int error;
 337 
 338         if (zp->z_is_sa)
 339                 return (0);
 340 
 341         /*
 342          * Need to deal with a potential
 343          * race where zfs_sa_upgrade could cause
 344          * z_isa_sa to change.
 345          *
 346          * If the lookup fails then the state of z_is_sa should have
 347          * changed.
 348          */
 349 
 350         if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
 351             &acl_phys, sizeof (acl_phys))) == 0)
 352                 return (acl_phys.z_acl_extern_obj);
 353         else {
 354                 /*
 355                  * after upgrade the SA_ZPL_ZNODE_ACL should have been
 356                  * removed
 357                  */
 358                 VERIFY(zp->z_is_sa && error == ENOENT);
 359                 return (0);
 360         }
 361 }
 362 
 363 /*
 364  * Determine size of ACL in bytes
 365  *
 366  * This is more complicated than it should be since we have to deal
 367  * with old external ACLs.
 368  */
 369 static int
 370 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
 371     zfs_acl_phys_t *aclphys)
 372 {
 373         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 374         uint64_t acl_count;
 375         int size;
 376         int error;
 377 
 378         ASSERT(MUTEX_HELD(&zp->z_acl_lock));
 379         if (zp->z_is_sa) {
 380                 if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
 381                     &size)) != 0)
 382                         return (error);
 383                 *aclsize = size;
 384                 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
 385                     &acl_count, sizeof (acl_count))) != 0)
 386                         return (error);
 387                 *aclcount = acl_count;
 388         } else {
 389                 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
 390                     aclphys, sizeof (*aclphys))) != 0)
 391                         return (error);
 392 
 393                 if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
 394                         *aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
 395                         *aclcount = aclphys->z_acl_size;
 396                 } else {
 397                         *aclsize = aclphys->z_acl_size;
 398                         *aclcount = aclphys->z_acl_count;
 399                 }
 400         }
 401         return (0);
 402 }
 403 
 404 int
 405 zfs_znode_acl_version(znode_t *zp)
 406 {
 407         zfs_acl_phys_t acl_phys;
 408 
 409         if (zp->z_is_sa)
 410                 return (ZFS_ACL_VERSION_FUID);
 411         else {
 412                 int error;
 413 
 414                 /*
 415                  * Need to deal with a potential
 416                  * race where zfs_sa_upgrade could cause
 417                  * z_isa_sa to change.
 418                  *
 419                  * If the lookup fails then the state of z_is_sa should have
 420                  * changed.
 421                  */
 422                 if ((error = sa_lookup(zp->z_sa_hdl,
 423                     SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
 424                     &acl_phys, sizeof (acl_phys))) == 0)
 425                         return (acl_phys.z_acl_version);
 426                 else {
 427                         /*
 428                          * After upgrade SA_ZPL_ZNODE_ACL should have
 429                          * been removed.
 430                          */
 431                         VERIFY(zp->z_is_sa && error == ENOENT);
 432                         return (ZFS_ACL_VERSION_FUID);
 433                 }
 434         }
 435 }
 436 
 437 static int
 438 zfs_acl_version(int version)
 439 {
 440         if (version < ZPL_VERSION_FUID)
 441                 return (ZFS_ACL_VERSION_INITIAL);
 442         else
 443                 return (ZFS_ACL_VERSION_FUID);
 444 }
 445 
 446 static int
 447 zfs_acl_version_zp(znode_t *zp)
 448 {
 449         return (zfs_acl_version(zp->z_zfsvfs->z_version));
 450 }
 451 
 452 zfs_acl_t *
 453 zfs_acl_alloc(int vers)
 454 {
 455         zfs_acl_t *aclp;
 456 
 457         aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
 458         list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
 459             offsetof(zfs_acl_node_t, z_next));
 460         aclp->z_version = vers;
 461         if (vers == ZFS_ACL_VERSION_FUID)
 462                 aclp->z_ops = zfs_acl_fuid_ops;
 463         else
 464                 aclp->z_ops = zfs_acl_v0_ops;
 465         return (aclp);
 466 }
 467 
 468 zfs_acl_node_t *
 469 zfs_acl_node_alloc(size_t bytes)
 470 {
 471         zfs_acl_node_t *aclnode;
 472 
 473         aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
 474         if (bytes) {
 475                 aclnode->z_acldata = kmem_alloc(bytes, KM_SLEEP);
 476                 aclnode->z_allocdata = aclnode->z_acldata;
 477                 aclnode->z_allocsize = bytes;
 478                 aclnode->z_size = bytes;
 479         }
 480 
 481         return (aclnode);
 482 }
 483 
 484 static void
 485 zfs_acl_node_free(zfs_acl_node_t *aclnode)
 486 {
 487         if (aclnode->z_allocsize)
 488                 kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
 489         kmem_free(aclnode, sizeof (zfs_acl_node_t));
 490 }
 491 
 492 static void
 493 zfs_acl_release_nodes(zfs_acl_t *aclp)
 494 {
 495         zfs_acl_node_t *aclnode;
 496 
 497         while (aclnode = list_head(&aclp->z_acl)) {
 498                 list_remove(&aclp->z_acl, aclnode);
 499                 zfs_acl_node_free(aclnode);
 500         }
 501         aclp->z_acl_count = 0;
 502         aclp->z_acl_bytes = 0;
 503 }
 504 
 505 void
 506 zfs_acl_free(zfs_acl_t *aclp)
 507 {
 508         zfs_acl_release_nodes(aclp);
 509         list_destroy(&aclp->z_acl);
 510         kmem_free(aclp, sizeof (zfs_acl_t));
 511 }
 512 
 513 static boolean_t
 514 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
 515 {
 516         uint16_t entry_type;
 517 
 518         switch (type) {
 519         case ALLOW:
 520         case DENY:
 521         case ACE_SYSTEM_AUDIT_ACE_TYPE:
 522         case ACE_SYSTEM_ALARM_ACE_TYPE:
 523                 entry_type = flags & ACE_TYPE_FLAGS;
 524                 return (entry_type == ACE_OWNER ||
 525                     entry_type == OWNING_GROUP ||
 526                     entry_type == ACE_EVERYONE || entry_type == 0 ||
 527                     entry_type == ACE_IDENTIFIER_GROUP);
 528         default:
 529                 if (type >= MIN_ACE_TYPE && type <= MAX_ACE_TYPE)
 530                         return (B_TRUE);
 531         }
 532         return (B_FALSE);
 533 }
 534 
 535 static boolean_t
 536 zfs_ace_valid(vtype_t obj_type, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
 537 {
 538         /*
 539          * first check type of entry
 540          */
 541 
 542         if (!zfs_acl_valid_ace_type(type, iflags))
 543                 return (B_FALSE);
 544 
 545         switch (type) {
 546         case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
 547         case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
 548         case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
 549         case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
 550                 if (aclp->z_version < ZFS_ACL_VERSION_FUID)
 551                         return (B_FALSE);
 552                 aclp->z_hints |= ZFS_ACL_OBJ_ACE;
 553         }
 554 
 555         /*
 556          * next check inheritance level flags
 557          */
 558 
 559         if (obj_type == VDIR &&
 560             (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
 561                 aclp->z_hints |= ZFS_INHERIT_ACE;
 562 
 563         if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
 564                 if ((iflags & (ACE_FILE_INHERIT_ACE|
 565                     ACE_DIRECTORY_INHERIT_ACE)) == 0) {
 566                         return (B_FALSE);
 567                 }
 568         }
 569 
 570         return (B_TRUE);
 571 }
 572 
 573 static void *
 574 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
 575     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
 576 {
 577         zfs_acl_node_t *aclnode;
 578 
 579         ASSERT(aclp);
 580 
 581         if (start == NULL) {
 582                 aclnode = list_head(&aclp->z_acl);
 583                 if (aclnode == NULL)
 584                         return (NULL);
 585 
 586                 aclp->z_next_ace = aclnode->z_acldata;
 587                 aclp->z_curr_node = aclnode;
 588                 aclnode->z_ace_idx = 0;
 589         }
 590 
 591         aclnode = aclp->z_curr_node;
 592 
 593         if (aclnode == NULL)
 594                 return (NULL);
 595 
 596         if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
 597                 aclnode = list_next(&aclp->z_acl, aclnode);
 598                 if (aclnode == NULL)
 599                         return (NULL);
 600                 else {
 601                         aclp->z_curr_node = aclnode;
 602                         aclnode->z_ace_idx = 0;
 603                         aclp->z_next_ace = aclnode->z_acldata;
 604                 }
 605         }
 606 
 607         if (aclnode->z_ace_idx < aclnode->z_ace_count) {
 608                 void *acep = aclp->z_next_ace;
 609                 size_t ace_size;
 610 
 611                 /*
 612                  * Make sure we don't overstep our bounds
 613                  */
 614                 ace_size = aclp->z_ops.ace_size(acep);
 615 
 616                 if (((caddr_t)acep + ace_size) >
 617                     ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
 618                         return (NULL);
 619                 }
 620 
 621                 *iflags = aclp->z_ops.ace_flags_get(acep);
 622                 *type = aclp->z_ops.ace_type_get(acep);
 623                 *access_mask = aclp->z_ops.ace_mask_get(acep);
 624                 *who = aclp->z_ops.ace_who_get(acep);
 625                 aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
 626                 aclnode->z_ace_idx++;
 627 
 628                 return ((void *)acep);
 629         }
 630         return (NULL);
 631 }
 632 
 633 /*ARGSUSED*/
 634 static uint64_t
 635 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
 636     uint16_t *flags, uint16_t *type, uint32_t *mask)
 637 {
 638         zfs_acl_t *aclp = datap;
 639         zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
 640         uint64_t who;
 641 
 642         acep = zfs_acl_next_ace(aclp, acep, &who, mask,
 643             flags, type);
 644         return ((uint64_t)(uintptr_t)acep);
 645 }
 646 
 647 static zfs_acl_node_t *
 648 zfs_acl_curr_node(zfs_acl_t *aclp)
 649 {
 650         ASSERT(aclp->z_curr_node);
 651         return (aclp->z_curr_node);
 652 }
 653 
 654 /*
 655  * Copy ACE to internal ZFS format.
 656  * While processing the ACL each ACE will be validated for correctness.
 657  * ACE FUIDs will be created later.
 658  */
 659 int
 660 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, vtype_t obj_type, zfs_acl_t *aclp,
 661     void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
 662     zfs_fuid_info_t **fuidp, cred_t *cr)
 663 {
 664         int i;
 665         uint16_t entry_type;
 666         zfs_ace_t *aceptr = z_acl;
 667         ace_t *acep = datap;
 668         zfs_object_ace_t *zobjacep;
 669         ace_object_t *aceobjp;
 670 
 671         for (i = 0; i != aclcnt; i++) {
 672                 aceptr->z_hdr.z_access_mask = acep->a_access_mask;
 673                 aceptr->z_hdr.z_flags = acep->a_flags;
 674                 aceptr->z_hdr.z_type = acep->a_type;
 675                 entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
 676                 if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
 677                     entry_type != ACE_EVERYONE) {
 678                         aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
 679                             cr, (entry_type == 0) ?
 680                             ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
 681                 }
 682 
 683                 /*
 684                  * Make sure ACE is valid
 685                  */
 686                 if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
 687                     aceptr->z_hdr.z_flags) != B_TRUE)
 688                         return (SET_ERROR(EINVAL));
 689 
 690                 switch (acep->a_type) {
 691                 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
 692                 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
 693                 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
 694                 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
 695                         zobjacep = (zfs_object_ace_t *)aceptr;
 696                         aceobjp = (ace_object_t *)acep;
 697 
 698                         bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
 699                             sizeof (aceobjp->a_obj_type));
 700                         bcopy(aceobjp->a_inherit_obj_type,
 701                             zobjacep->z_inherit_type,
 702                             sizeof (aceobjp->a_inherit_obj_type));
 703                         acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
 704                         break;
 705                 default:
 706                         acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
 707                 }
 708 
 709                 aceptr = (zfs_ace_t *)((caddr_t)aceptr +
 710                     aclp->z_ops.ace_size(aceptr));
 711         }
 712 
 713         *size = (caddr_t)aceptr - (caddr_t)z_acl;
 714 
 715         return (0);
 716 }
 717 
 718 /*
 719  * Copy ZFS ACEs to fixed size ace_t layout
 720  */
 721 static void
 722 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
 723     void *datap, int filter)
 724 {
 725         uint64_t who;
 726         uint32_t access_mask;
 727         uint16_t iflags, type;
 728         zfs_ace_hdr_t *zacep = NULL;
 729         ace_t *acep = datap;
 730         ace_object_t *objacep;
 731         zfs_object_ace_t *zobjacep;
 732         size_t ace_size;
 733         uint16_t entry_type;
 734 
 735         while (zacep = zfs_acl_next_ace(aclp, zacep,
 736             &who, &access_mask, &iflags, &type)) {
 737 
 738                 switch (type) {
 739                 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
 740                 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
 741                 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
 742                 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
 743                         if (filter) {
 744                                 continue;
 745                         }
 746                         zobjacep = (zfs_object_ace_t *)zacep;
 747                         objacep = (ace_object_t *)acep;
 748                         bcopy(zobjacep->z_object_type,
 749                             objacep->a_obj_type,
 750                             sizeof (zobjacep->z_object_type));
 751                         bcopy(zobjacep->z_inherit_type,
 752                             objacep->a_inherit_obj_type,
 753                             sizeof (zobjacep->z_inherit_type));
 754                         ace_size = sizeof (ace_object_t);
 755                         break;
 756                 default:
 757                         ace_size = sizeof (ace_t);
 758                         break;
 759                 }
 760 
 761                 entry_type = (iflags & ACE_TYPE_FLAGS);
 762                 if ((entry_type != ACE_OWNER &&
 763                     entry_type != OWNING_GROUP &&
 764                     entry_type != ACE_EVERYONE)) {
 765                         acep->a_who = zfs_fuid_map_id(zfsvfs, who,
 766                             cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
 767                             ZFS_ACE_GROUP : ZFS_ACE_USER);
 768                 } else {
 769                         acep->a_who = (uid_t)(int64_t)who;
 770                 }
 771                 acep->a_access_mask = access_mask;
 772                 acep->a_flags = iflags;
 773                 acep->a_type = type;
 774                 acep = (ace_t *)((caddr_t)acep + ace_size);
 775         }
 776 }
 777 
 778 static int
 779 zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
 780     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
 781 {
 782         int i;
 783         zfs_oldace_t *aceptr = z_acl;
 784 
 785         for (i = 0; i != aclcnt; i++, aceptr++) {
 786                 aceptr->z_access_mask = acep[i].a_access_mask;
 787                 aceptr->z_type = acep[i].a_type;
 788                 aceptr->z_flags = acep[i].a_flags;
 789                 aceptr->z_fuid = acep[i].a_who;
 790                 /*
 791                  * Make sure ACE is valid
 792                  */
 793                 if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
 794                     aceptr->z_flags) != B_TRUE)
 795                         return (SET_ERROR(EINVAL));
 796         }
 797         *size = (caddr_t)aceptr - (caddr_t)z_acl;
 798         return (0);
 799 }
 800 
 801 /*
 802  * convert old ACL format to new
 803  */
 804 void
 805 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
 806 {
 807         zfs_oldace_t *oldaclp;
 808         int i;
 809         uint16_t type, iflags;
 810         uint32_t access_mask;
 811         uint64_t who;
 812         void *cookie = NULL;
 813         zfs_acl_node_t *newaclnode;
 814 
 815         ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
 816         /*
 817          * First create the ACE in a contiguous piece of memory
 818          * for zfs_copy_ace_2_fuid().
 819          *
 820          * We only convert an ACL once, so this won't happen
 821          * everytime.
 822          */
 823         oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
 824             KM_SLEEP);
 825         i = 0;
 826         while (cookie = zfs_acl_next_ace(aclp, cookie, &who,
 827             &access_mask, &iflags, &type)) {
 828                 oldaclp[i].z_flags = iflags;
 829                 oldaclp[i].z_type = type;
 830                 oldaclp[i].z_fuid = who;
 831                 oldaclp[i++].z_access_mask = access_mask;
 832         }
 833 
 834         newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
 835             sizeof (zfs_object_ace_t));
 836         aclp->z_ops = zfs_acl_fuid_ops;
 837         VERIFY(zfs_copy_ace_2_fuid(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
 838             oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
 839             &newaclnode->z_size, NULL, cr) == 0);
 840         newaclnode->z_ace_count = aclp->z_acl_count;
 841         aclp->z_version = ZFS_ACL_VERSION;
 842         kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
 843 
 844         /*
 845          * Release all previous ACL nodes
 846          */
 847 
 848         zfs_acl_release_nodes(aclp);
 849 
 850         list_insert_head(&aclp->z_acl, newaclnode);
 851 
 852         aclp->z_acl_bytes = newaclnode->z_size;
 853         aclp->z_acl_count = newaclnode->z_ace_count;
 854 
 855 }
 856 
 857 /*
 858  * Convert unix access mask to v4 access mask
 859  */
 860 static uint32_t
 861 zfs_unix_to_v4(uint32_t access_mask)
 862 {
 863         uint32_t new_mask = 0;
 864 
 865         if (access_mask & S_IXOTH)
 866                 new_mask |= ACE_EXECUTE;
 867         if (access_mask & S_IWOTH)
 868                 new_mask |= ACE_WRITE_DATA;
 869         if (access_mask & S_IROTH)
 870                 new_mask |= ACE_READ_DATA;
 871         return (new_mask);
 872 }
 873 
 874 static void
 875 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
 876     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
 877 {
 878         uint16_t type = entry_type & ACE_TYPE_FLAGS;
 879 
 880         aclp->z_ops.ace_mask_set(acep, access_mask);
 881         aclp->z_ops.ace_type_set(acep, access_type);
 882         aclp->z_ops.ace_flags_set(acep, entry_type);
 883         if ((type != ACE_OWNER && type != OWNING_GROUP &&
 884             type != ACE_EVERYONE))
 885                 aclp->z_ops.ace_who_set(acep, fuid);
 886 }
 887 
 888 /*
 889  * Determine mode of file based on ACL.
 890  */
 891 uint64_t
 892 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
 893     uint64_t *pflags, uint64_t fuid, uint64_t fgid)
 894 {
 895         int             entry_type;
 896         mode_t          mode;
 897         mode_t          seen = 0;
 898         zfs_ace_hdr_t   *acep = NULL;
 899         uint64_t        who;
 900         uint16_t        iflags, type;
 901         uint32_t        access_mask;
 902         boolean_t       an_exec_denied = B_FALSE;
 903 
 904         mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
 905 
 906         while (acep = zfs_acl_next_ace(aclp, acep, &who,
 907             &access_mask, &iflags, &type)) {
 908 
 909                 if (!zfs_acl_valid_ace_type(type, iflags))
 910                         continue;
 911 
 912                 entry_type = (iflags & ACE_TYPE_FLAGS);
 913 
 914                 /*
 915                  * Skip over any inherit_only ACEs
 916                  */
 917                 if (iflags & ACE_INHERIT_ONLY_ACE)
 918                         continue;
 919 
 920                 if (entry_type == ACE_OWNER || (entry_type == 0 &&
 921                     who == fuid)) {
 922                         if ((access_mask & ACE_READ_DATA) &&
 923                             (!(seen & S_IRUSR))) {
 924                                 seen |= S_IRUSR;
 925                                 if (type == ALLOW) {
 926                                         mode |= S_IRUSR;
 927                                 }
 928                         }
 929                         if ((access_mask & ACE_WRITE_DATA) &&
 930                             (!(seen & S_IWUSR))) {
 931                                 seen |= S_IWUSR;
 932                                 if (type == ALLOW) {
 933                                         mode |= S_IWUSR;
 934                                 }
 935                         }
 936                         if ((access_mask & ACE_EXECUTE) &&
 937                             (!(seen & S_IXUSR))) {
 938                                 seen |= S_IXUSR;
 939                                 if (type == ALLOW) {
 940                                         mode |= S_IXUSR;
 941                                 }
 942                         }
 943                 } else if (entry_type == OWNING_GROUP ||
 944                     (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
 945                         if ((access_mask & ACE_READ_DATA) &&
 946                             (!(seen & S_IRGRP))) {
 947                                 seen |= S_IRGRP;
 948                                 if (type == ALLOW) {
 949                                         mode |= S_IRGRP;
 950                                 }
 951                         }
 952                         if ((access_mask & ACE_WRITE_DATA) &&
 953                             (!(seen & S_IWGRP))) {
 954                                 seen |= S_IWGRP;
 955                                 if (type == ALLOW) {
 956                                         mode |= S_IWGRP;
 957                                 }
 958                         }
 959                         if ((access_mask & ACE_EXECUTE) &&
 960                             (!(seen & S_IXGRP))) {
 961                                 seen |= S_IXGRP;
 962                                 if (type == ALLOW) {
 963                                         mode |= S_IXGRP;
 964                                 }
 965                         }
 966                 } else if (entry_type == ACE_EVERYONE) {
 967                         if ((access_mask & ACE_READ_DATA)) {
 968                                 if (!(seen & S_IRUSR)) {
 969                                         seen |= S_IRUSR;
 970                                         if (type == ALLOW) {
 971                                                 mode |= S_IRUSR;
 972                                         }
 973                                 }
 974                                 if (!(seen & S_IRGRP)) {
 975                                         seen |= S_IRGRP;
 976                                         if (type == ALLOW) {
 977                                                 mode |= S_IRGRP;
 978                                         }
 979                                 }
 980                                 if (!(seen & S_IROTH)) {
 981                                         seen |= S_IROTH;
 982                                         if (type == ALLOW) {
 983                                                 mode |= S_IROTH;
 984                                         }
 985                                 }
 986                         }
 987                         if ((access_mask & ACE_WRITE_DATA)) {
 988                                 if (!(seen & S_IWUSR)) {
 989                                         seen |= S_IWUSR;
 990                                         if (type == ALLOW) {
 991                                                 mode |= S_IWUSR;
 992                                         }
 993                                 }
 994                                 if (!(seen & S_IWGRP)) {
 995                                         seen |= S_IWGRP;
 996                                         if (type == ALLOW) {
 997                                                 mode |= S_IWGRP;
 998                                         }
 999                                 }
1000                                 if (!(seen & S_IWOTH)) {
1001                                         seen |= S_IWOTH;
1002                                         if (type == ALLOW) {
1003                                                 mode |= S_IWOTH;
1004                                         }
1005                                 }
1006                         }
1007                         if ((access_mask & ACE_EXECUTE)) {
1008                                 if (!(seen & S_IXUSR)) {
1009                                         seen |= S_IXUSR;
1010                                         if (type == ALLOW) {
1011                                                 mode |= S_IXUSR;
1012                                         }
1013                                 }
1014                                 if (!(seen & S_IXGRP)) {
1015                                         seen |= S_IXGRP;
1016                                         if (type == ALLOW) {
1017                                                 mode |= S_IXGRP;
1018                                         }
1019                                 }
1020                                 if (!(seen & S_IXOTH)) {
1021                                         seen |= S_IXOTH;
1022                                         if (type == ALLOW) {
1023                                                 mode |= S_IXOTH;
1024                                         }
1025                                 }
1026                         }
1027                 } else {
1028                         /*
1029                          * Only care if this IDENTIFIER_GROUP or
1030                          * USER ACE denies execute access to someone,
1031                          * mode is not affected
1032                          */
1033                         if ((access_mask & ACE_EXECUTE) && type == DENY)
1034                                 an_exec_denied = B_TRUE;
1035                 }
1036         }
1037 
1038         /*
1039          * Failure to allow is effectively a deny, so execute permission
1040          * is denied if it was never mentioned or if we explicitly
1041          * weren't allowed it.
1042          */
1043         if (!an_exec_denied &&
1044             ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1045             (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1046                 an_exec_denied = B_TRUE;
1047 
1048         if (an_exec_denied)
1049                 *pflags &= ~ZFS_NO_EXECS_DENIED;
1050         else
1051                 *pflags |= ZFS_NO_EXECS_DENIED;
1052 
1053         return (mode);
1054 }
1055 
1056 /*
1057  * Read an external acl object.  If the intent is to modify, always
1058  * create a new acl and leave any cached acl in place.
1059  */
1060 static int
1061 zfs_acl_node_read(znode_t *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1062     boolean_t will_modify)
1063 {
1064         zfs_acl_t       *aclp;
1065         int             aclsize;
1066         int             acl_count;
1067         zfs_acl_node_t  *aclnode;
1068         zfs_acl_phys_t  znode_acl;
1069         int             version;
1070         int             error;
1071         boolean_t       drop_lock = B_FALSE;
1072 
1073         ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1074 
1075         if (zp->z_acl_cached && !will_modify) {
1076                 *aclpp = zp->z_acl_cached;
1077                 return (0);
1078         }
1079 
1080         /*
1081          * close race where znode could be upgrade while trying to
1082          * read the znode attributes.
1083          *
1084          * But this could only happen if the file isn't already an SA
1085          * znode
1086          */
1087         if (!zp->z_is_sa && !have_lock) {
1088                 mutex_enter(&zp->z_lock);
1089                 drop_lock = B_TRUE;
1090         }
1091         version = zfs_znode_acl_version(zp);
1092 
1093         if ((error = zfs_acl_znode_info(zp, &aclsize,
1094             &acl_count, &znode_acl)) != 0) {
1095                 goto done;
1096         }
1097 
1098         aclp = zfs_acl_alloc(version);
1099 
1100         aclp->z_acl_count = acl_count;
1101         aclp->z_acl_bytes = aclsize;
1102 
1103         aclnode = zfs_acl_node_alloc(aclsize);
1104         aclnode->z_ace_count = aclp->z_acl_count;
1105         aclnode->z_size = aclsize;
1106 
1107         if (!zp->z_is_sa) {
1108                 if (znode_acl.z_acl_extern_obj) {
1109                         error = dmu_read(zp->z_zfsvfs->z_os,
1110                             znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1111                             aclnode->z_acldata, DMU_READ_PREFETCH);
1112                 } else {
1113                         bcopy(znode_acl.z_ace_data, aclnode->z_acldata,
1114                             aclnode->z_size);
1115                 }
1116         } else {
1117                 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zp->z_zfsvfs),
1118                     aclnode->z_acldata, aclnode->z_size);
1119         }
1120 
1121         if (error != 0) {
1122                 zfs_acl_free(aclp);
1123                 zfs_acl_node_free(aclnode);
1124                 /* convert checksum errors into IO errors */
1125                 if (error == ECKSUM)
1126                         error = SET_ERROR(EIO);
1127                 goto done;
1128         }
1129 
1130         list_insert_head(&aclp->z_acl, aclnode);
1131 
1132         *aclpp = aclp;
1133         if (!will_modify)
1134                 zp->z_acl_cached = aclp;
1135 done:
1136         if (drop_lock)
1137                 mutex_exit(&zp->z_lock);
1138         return (error);
1139 }
1140 
1141 /*ARGSUSED*/
1142 void
1143 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1144     boolean_t start, void *userdata)
1145 {
1146         zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1147 
1148         if (start) {
1149                 cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1150         } else {
1151                 cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1152                     cb->cb_acl_node);
1153         }
1154         *dataptr = cb->cb_acl_node->z_acldata;
1155         *length = cb->cb_acl_node->z_size;
1156 }
1157 
1158 int
1159 zfs_acl_chown_setattr(znode_t *zp)
1160 {
1161         int error;
1162         zfs_acl_t *aclp;
1163 
1164         ASSERT(MUTEX_HELD(&zp->z_lock));
1165         ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1166 
1167         if ((error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE)) == 0)
1168                 zp->z_mode = zfs_mode_compute(zp->z_mode, aclp,
1169                     &zp->z_pflags, zp->z_uid, zp->z_gid);
1170         return (error);
1171 }
1172 
1173 /*
1174  * common code for setting ACLs.
1175  *
1176  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1177  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1178  * already checked the acl and knows whether to inherit.
1179  */
1180 int
1181 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1182 {
1183         int                     error;
1184         zfsvfs_t                *zfsvfs = zp->z_zfsvfs;
1185         dmu_object_type_t       otype;
1186         zfs_acl_locator_cb_t    locate = { 0 };
1187         uint64_t                mode;
1188         sa_bulk_attr_t          bulk[5];
1189         uint64_t                ctime[2];
1190         int                     count = 0;
1191         zfs_acl_phys_t          acl_phys;
1192 
1193         mode = zp->z_mode;
1194 
1195         mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1196             zp->z_uid, zp->z_gid);
1197 
1198         zp->z_mode = mode;
1199         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1200             &mode, sizeof (mode));
1201         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1202             &zp->z_pflags, sizeof (zp->z_pflags));
1203         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1204             &ctime, sizeof (ctime));
1205 
1206         if (zp->z_acl_cached) {
1207                 zfs_acl_free(zp->z_acl_cached);
1208                 zp->z_acl_cached = NULL;
1209         }
1210 
1211         /*
1212          * Upgrade needed?
1213          */
1214         if (!zfsvfs->z_use_fuids) {
1215                 otype = DMU_OT_OLDACL;
1216         } else {
1217                 if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1218                     (zfsvfs->z_version >= ZPL_VERSION_FUID))
1219                         zfs_acl_xform(zp, aclp, cr);
1220                 ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1221                 otype = DMU_OT_ACL;
1222         }
1223 
1224         /*
1225          * Arrgh, we have to handle old on disk format
1226          * as well as newer (preferred) SA format.
1227          */
1228 
1229         if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1230                 locate.cb_aclp = aclp;
1231                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1232                     zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1233                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1234                     NULL, &aclp->z_acl_count, sizeof (uint64_t));
1235         } else { /* Painful legacy way */
1236                 zfs_acl_node_t *aclnode;
1237                 uint64_t off = 0;
1238                 uint64_t aoid;
1239 
1240                 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1241                     &acl_phys, sizeof (acl_phys))) != 0)
1242                         return (error);
1243 
1244                 aoid = acl_phys.z_acl_extern_obj;
1245 
1246                 if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1247                         /*
1248                          * If ACL was previously external and we are now
1249                          * converting to new ACL format then release old
1250                          * ACL object and create a new one.
1251                          */
1252                         if (aoid &&
1253                             aclp->z_version != acl_phys.z_acl_version) {
1254                                 error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1255                                 if (error)
1256                                         return (error);
1257                                 aoid = 0;
1258                         }
1259                         if (aoid == 0) {
1260                                 aoid = dmu_object_alloc(zfsvfs->z_os,
1261                                     otype, aclp->z_acl_bytes,
1262                                     otype == DMU_OT_ACL ?
1263                                     DMU_OT_SYSACL : DMU_OT_NONE,
1264                                     otype == DMU_OT_ACL ?
1265                                     DN_MAX_BONUSLEN : 0, tx);
1266                         } else {
1267                                 (void) dmu_object_set_blocksize(zfsvfs->z_os,
1268                                     aoid, aclp->z_acl_bytes, 0, tx);
1269                         }
1270                         acl_phys.z_acl_extern_obj = aoid;
1271                         for (aclnode = list_head(&aclp->z_acl); aclnode;
1272                             aclnode = list_next(&aclp->z_acl, aclnode)) {
1273                                 if (aclnode->z_ace_count == 0)
1274                                         continue;
1275                                 dmu_write(zfsvfs->z_os, aoid, off,
1276                                     aclnode->z_size, aclnode->z_acldata, tx);
1277                                 off += aclnode->z_size;
1278                         }
1279                 } else {
1280                         void *start = acl_phys.z_ace_data;
1281                         /*
1282                          * Migrating back embedded?
1283                          */
1284                         if (acl_phys.z_acl_extern_obj) {
1285                                 error = dmu_object_free(zfsvfs->z_os,
1286                                     acl_phys.z_acl_extern_obj, tx);
1287                                 if (error)
1288                                         return (error);
1289                                 acl_phys.z_acl_extern_obj = 0;
1290                         }
1291 
1292                         for (aclnode = list_head(&aclp->z_acl); aclnode;
1293                             aclnode = list_next(&aclp->z_acl, aclnode)) {
1294                                 if (aclnode->z_ace_count == 0)
1295                                         continue;
1296                                 bcopy(aclnode->z_acldata, start,
1297                                     aclnode->z_size);
1298                                 start = (caddr_t)start + aclnode->z_size;
1299                         }
1300                 }
1301                 /*
1302                  * If Old version then swap count/bytes to match old
1303                  * layout of znode_acl_phys_t.
1304                  */
1305                 if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1306                         acl_phys.z_acl_size = aclp->z_acl_count;
1307                         acl_phys.z_acl_count = aclp->z_acl_bytes;
1308                 } else {
1309                         acl_phys.z_acl_size = aclp->z_acl_bytes;
1310                         acl_phys.z_acl_count = aclp->z_acl_count;
1311                 }
1312                 acl_phys.z_acl_version = aclp->z_version;
1313 
1314                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1315                     &acl_phys, sizeof (acl_phys));
1316         }
1317 
1318         /*
1319          * Replace ACL wide bits, but first clear them.
1320          */
1321         zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1322 
1323         zp->z_pflags |= aclp->z_hints;
1324 
1325         if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1326                 zp->z_pflags |= ZFS_ACL_TRIVIAL;
1327 
1328         zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime, B_TRUE);
1329         return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1330 }
1331 
1332 static void
1333 zfs_acl_chmod(vtype_t vtype, uint64_t mode, boolean_t split, boolean_t trim,
1334     zfs_acl_t *aclp)
1335 {
1336         void            *acep = NULL;
1337         uint64_t        who;
1338         int             new_count, new_bytes;
1339         int             ace_size;
1340         int             entry_type;
1341         uint16_t        iflags, type;
1342         uint32_t        access_mask;
1343         zfs_acl_node_t  *newnode;
1344         size_t          abstract_size = aclp->z_ops.ace_abstract_size();
1345         void            *zacep;
1346         boolean_t       isdir;
1347         trivial_acl_t   masks;
1348 
1349         new_count = new_bytes = 0;
1350 
1351         isdir = (vtype == VDIR);
1352 
1353         acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1354 
1355         newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1356 
1357         zacep = newnode->z_acldata;
1358         if (masks.allow0) {
1359                 zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1360                 zacep = (void *)((uintptr_t)zacep + abstract_size);
1361                 new_count++;
1362                 new_bytes += abstract_size;
1363         }
1364         if (masks.deny1) {
1365                 zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1366                 zacep = (void *)((uintptr_t)zacep + abstract_size);
1367                 new_count++;
1368                 new_bytes += abstract_size;
1369         }
1370         if (masks.deny2) {
1371                 zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1372                 zacep = (void *)((uintptr_t)zacep + abstract_size);
1373                 new_count++;
1374                 new_bytes += abstract_size;
1375         }
1376 
1377         while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1378             &iflags, &type)) {
1379                 entry_type = (iflags & ACE_TYPE_FLAGS);
1380                 /*
1381                  * ACEs used to represent the file mode may be divided
1382                  * into an equivalent pair of inherit-only and regular
1383                  * ACEs, if they are inheritable.
1384                  * Skip regular ACEs, which are replaced by the new mode.
1385                  */
1386                 if (split && (entry_type == ACE_OWNER ||
1387                     entry_type == OWNING_GROUP ||
1388                     entry_type == ACE_EVERYONE)) {
1389                         if (!isdir || !(iflags &
1390                             (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1391                                 continue;
1392                         /*
1393                          * We preserve owner@, group@, or @everyone
1394                          * permissions, if they are inheritable, by
1395                          * copying them to inherit_only ACEs. This
1396                          * prevents inheritable permissions from being
1397                          * altered along with the file mode.
1398                          */
1399                         iflags |= ACE_INHERIT_ONLY_ACE;
1400                 }
1401 
1402                 /*
1403                  * If this ACL has any inheritable ACEs, mark that in
1404                  * the hints (which are later masked into the pflags)
1405                  * so create knows to do inheritance.
1406                  */
1407                 if (isdir && (iflags &
1408                     (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1409                         aclp->z_hints |= ZFS_INHERIT_ACE;
1410 
1411                 if ((type != ALLOW && type != DENY) ||
1412                     (iflags & ACE_INHERIT_ONLY_ACE)) {
1413                         switch (type) {
1414                         case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1415                         case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1416                         case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1417                         case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1418                                 aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1419                                 break;
1420                         }
1421                 } else {
1422                         /*
1423                          * Limit permissions granted by ACEs to be no greater
1424                          * than permissions of the requested group mode.
1425                          * Applies when the "aclmode" property is set to
1426                          * "groupmask".
1427                          */
1428                         if ((type == ALLOW) && trim)
1429                                 access_mask &= masks.group;
1430                 }
1431                 zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1432                 ace_size = aclp->z_ops.ace_size(acep);
1433                 zacep = (void *)((uintptr_t)zacep + ace_size);
1434                 new_count++;
1435                 new_bytes += ace_size;
1436         }
1437         zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1438         zacep = (void *)((uintptr_t)zacep + abstract_size);
1439         zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1440         zacep = (void *)((uintptr_t)zacep + abstract_size);
1441         zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1442 
1443         new_count += 3;
1444         new_bytes += abstract_size * 3;
1445         zfs_acl_release_nodes(aclp);
1446         aclp->z_acl_count = new_count;
1447         aclp->z_acl_bytes = new_bytes;
1448         newnode->z_ace_count = new_count;
1449         newnode->z_size = new_bytes;
1450         list_insert_tail(&aclp->z_acl, newnode);
1451 }
1452 
1453 int
1454 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1455 {
1456         int error = 0;
1457 
1458         mutex_enter(&zp->z_acl_lock);
1459         mutex_enter(&zp->z_lock);
1460         if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1461                 *aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1462         else
1463                 error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1464 
1465         if (error == 0) {
1466                 (*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1467                 zfs_acl_chmod(ZTOV(zp)->v_type, mode, B_TRUE,
1468                     (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1469         }
1470         mutex_exit(&zp->z_lock);
1471         mutex_exit(&zp->z_acl_lock);
1472 
1473         return (error);
1474 }
1475 
1476 /*
1477  * Should ACE be inherited?
1478  */
1479 static int
1480 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1481 {
1482         int     iflags = (acep_flags & 0xf);
1483 
1484         if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1485                 return (1);
1486         else if (iflags & ACE_FILE_INHERIT_ACE)
1487                 return (!((vtype == VDIR) &&
1488                     (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1489         return (0);
1490 }
1491 
1492 /*
1493  * inherit inheritable ACEs from parent
1494  */
1495 static zfs_acl_t *
1496 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1497     uint64_t mode)
1498 {
1499         void            *pacep = NULL;
1500         void            *acep;
1501         zfs_acl_node_t  *aclnode;
1502         zfs_acl_t       *aclp = NULL;
1503         uint64_t        who;
1504         uint32_t        access_mask;
1505         uint16_t        iflags, newflags, type;
1506         size_t          ace_size;
1507         void            *data1, *data2;
1508         size_t          data1sz, data2sz;
1509         uint_t          aclinherit;
1510         boolean_t       isdir = (vtype == VDIR);
1511 
1512         aclp = zfs_acl_alloc(paclp->z_version);
1513         aclinherit = zfsvfs->z_acl_inherit;
1514         if (aclinherit == ZFS_ACL_DISCARD || vtype == VLNK)
1515                 return (aclp);
1516 
1517         while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1518             &access_mask, &iflags, &type)) {
1519 
1520                 /*
1521                  * don't inherit bogus ACEs
1522                  */
1523                 if (!zfs_acl_valid_ace_type(type, iflags))
1524                         continue;
1525 
1526                 /*
1527                  * Check if ACE is inheritable by this vnode
1528                  */
1529                 if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1530                     !zfs_ace_can_use(vtype, iflags))
1531                         continue;
1532 
1533                 /*
1534                  * Strip inherited execute permission from file if
1535                  * not in mode
1536                  */
1537                 if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1538                     !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1539                         access_mask &= ~ACE_EXECUTE;
1540                 }
1541 
1542                 /*
1543                  * Strip write_acl and write_owner from permissions
1544                  * when inheriting an ACE
1545                  */
1546                 if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1547                         access_mask &= ~RESTRICTED_CLEAR;
1548                 }
1549 
1550                 ace_size = aclp->z_ops.ace_size(pacep);
1551                 aclnode = zfs_acl_node_alloc(ace_size);
1552                 list_insert_tail(&aclp->z_acl, aclnode);
1553                 acep = aclnode->z_acldata;
1554 
1555                 zfs_set_ace(aclp, acep, access_mask, type,
1556                     who, iflags|ACE_INHERITED_ACE);
1557 
1558                 /*
1559                  * Copy special opaque data if any
1560                  */
1561                 if ((data1sz = paclp->z_ops.ace_data(pacep, &data1)) != 0) {
1562                         VERIFY((data2sz = aclp->z_ops.ace_data(acep,
1563                             &data2)) == data1sz);
1564                         bcopy(data1, data2, data2sz);
1565                 }
1566 
1567                 aclp->z_acl_count++;
1568                 aclnode->z_ace_count++;
1569                 aclp->z_acl_bytes += aclnode->z_size;
1570                 newflags = aclp->z_ops.ace_flags_get(acep);
1571 
1572                 /*
1573                  * If ACE is not to be inherited further, or if the vnode is
1574                  * not a directory, remove all inheritance flags
1575                  */
1576                 if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1577                         newflags &= ~ALL_INHERIT;
1578                         aclp->z_ops.ace_flags_set(acep,
1579                             newflags|ACE_INHERITED_ACE);
1580                         continue;
1581                 }
1582 
1583                 /*
1584                  * This directory has an inheritable ACE
1585                  */
1586                 aclp->z_hints |= ZFS_INHERIT_ACE;
1587 
1588                 /*
1589                  * If only FILE_INHERIT is set then turn on
1590                  * inherit_only
1591                  */
1592                 if ((iflags & (ACE_FILE_INHERIT_ACE |
1593                     ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1594                         newflags |= ACE_INHERIT_ONLY_ACE;
1595                         aclp->z_ops.ace_flags_set(acep,
1596                             newflags|ACE_INHERITED_ACE);
1597                 } else {
1598                         newflags &= ~ACE_INHERIT_ONLY_ACE;
1599                         aclp->z_ops.ace_flags_set(acep,
1600                             newflags|ACE_INHERITED_ACE);
1601                 }
1602         }
1603 
1604         return (aclp);
1605 }
1606 
1607 /*
1608  * Create file system object initial permissions
1609  * including inheritable ACEs.
1610  * Also, create FUIDs for owner and group.
1611  */
1612 int
1613 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1614     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1615 {
1616         int             error;
1617         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1618         zfs_acl_t       *paclp;
1619         gid_t           gid;
1620         boolean_t       trim = B_FALSE;
1621         boolean_t       inherited = B_FALSE;
1622 
1623         bzero(acl_ids, sizeof (zfs_acl_ids_t));
1624         acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1625 
1626         if (vsecp)
1627                 if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1628                     &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1629                         return (error);
1630         /*
1631          * Determine uid and gid.
1632          */
1633         if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1634             ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1635                 acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1636                     (uint64_t)vap->va_uid, cr,
1637                     ZFS_OWNER, &acl_ids->z_fuidp);
1638                 acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1639                     (uint64_t)vap->va_gid, cr,
1640                     ZFS_GROUP, &acl_ids->z_fuidp);
1641                 gid = vap->va_gid;
1642         } else {
1643                 acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1644                     cr, &acl_ids->z_fuidp);
1645                 acl_ids->z_fgid = 0;
1646                 if (vap->va_mask & AT_GID)  {
1647                         acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1648                             (uint64_t)vap->va_gid,
1649                             cr, ZFS_GROUP, &acl_ids->z_fuidp);
1650                         gid = vap->va_gid;
1651                         if (acl_ids->z_fgid != dzp->z_gid &&
1652                             !groupmember(vap->va_gid, cr) &&
1653                             secpolicy_vnode_create_gid(cr) != 0)
1654                                 acl_ids->z_fgid = 0;
1655                 }
1656                 if (acl_ids->z_fgid == 0) {
1657                         if (dzp->z_mode & S_ISGID) {
1658                                 char            *domain;
1659                                 uint32_t        rid;
1660 
1661                                 acl_ids->z_fgid = dzp->z_gid;
1662                                 gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1663                                     cr, ZFS_GROUP);
1664 
1665                                 if (zfsvfs->z_use_fuids &&
1666                                     IS_EPHEMERAL(acl_ids->z_fgid)) {
1667                                         domain = zfs_fuid_idx_domain(
1668                                             &zfsvfs->z_fuid_idx,
1669                                             FUID_INDEX(acl_ids->z_fgid));
1670                                         rid = FUID_RID(acl_ids->z_fgid);
1671                                         zfs_fuid_node_add(&acl_ids->z_fuidp,
1672                                             domain, rid,
1673                                             FUID_INDEX(acl_ids->z_fgid),
1674                                             acl_ids->z_fgid, ZFS_GROUP);
1675                                 }
1676                         } else {
1677                                 acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1678                                     ZFS_GROUP, cr, &acl_ids->z_fuidp);
1679                                 gid = crgetgid(cr);
1680                         }
1681                 }
1682         }
1683 
1684         /*
1685          * If we're creating a directory, and the parent directory has the
1686          * set-GID bit set, set in on the new directory.
1687          * Otherwise, if the user is neither privileged nor a member of the
1688          * file's new group, clear the file's set-GID bit.
1689          */
1690 
1691         if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1692             (vap->va_type == VDIR)) {
1693                 acl_ids->z_mode |= S_ISGID;
1694         } else {
1695                 if ((acl_ids->z_mode & S_ISGID) &&
1696                     secpolicy_vnode_setids_setgids(cr, gid) != 0)
1697                         acl_ids->z_mode &= ~S_ISGID;
1698         }
1699 
1700         if (acl_ids->z_aclp == NULL) {
1701                 mutex_enter(&dzp->z_acl_lock);
1702                 mutex_enter(&dzp->z_lock);
1703                 if (!(flag & IS_ROOT_NODE) &&
1704                     (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1705                     !(dzp->z_pflags & ZFS_XATTR)) {
1706                         VERIFY(0 == zfs_acl_node_read(dzp, B_TRUE,
1707                             &paclp, B_FALSE));
1708                         acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1709                             vap->va_type, paclp, acl_ids->z_mode);
1710                         inherited = B_TRUE;
1711                 } else {
1712                         acl_ids->z_aclp =
1713                             zfs_acl_alloc(zfs_acl_version_zp(dzp));
1714                         acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1715                 }
1716                 mutex_exit(&dzp->z_lock);
1717                 mutex_exit(&dzp->z_acl_lock);
1718 
1719                 if (vap->va_type == VDIR)
1720                         acl_ids->z_aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1721 
1722                 if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK &&
1723                     zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH &&
1724                     zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH_X)
1725                         trim = B_TRUE;
1726                 zfs_acl_chmod(vap->va_type, acl_ids->z_mode, B_FALSE, trim,
1727                     acl_ids->z_aclp);
1728         }
1729 
1730         if (inherited || vsecp) {
1731                 acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1732                     acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1733                     acl_ids->z_fuid, acl_ids->z_fgid);
1734                 if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1735                         acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1736         }
1737 
1738         return (0);
1739 }
1740 
1741 /*
1742  * Free ACL and fuid_infop, but not the acl_ids structure
1743  */
1744 void
1745 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1746 {
1747         if (acl_ids->z_aclp)
1748                 zfs_acl_free(acl_ids->z_aclp);
1749         if (acl_ids->z_fuidp)
1750                 zfs_fuid_info_free(acl_ids->z_fuidp);
1751         acl_ids->z_aclp = NULL;
1752         acl_ids->z_fuidp = NULL;
1753 }
1754 
1755 boolean_t
1756 zfs_acl_ids_overquota(zfsvfs_t *zfsvfs, zfs_acl_ids_t *acl_ids)
1757 {
1758         return (zfs_fuid_overquota(zfsvfs, B_FALSE, acl_ids->z_fuid) ||
1759             zfs_fuid_overquota(zfsvfs, B_TRUE, acl_ids->z_fgid));
1760 }
1761 
1762 /*
1763  * Retrieve a file's ACL
1764  */
1765 int
1766 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1767 {
1768         zfs_acl_t       *aclp;
1769         ulong_t         mask;
1770         int             error;
1771         int             count = 0;
1772         int             largeace = 0;
1773 
1774         mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1775             VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1776 
1777         if (mask == 0)
1778                 return (SET_ERROR(ENOSYS));
1779 
1780         if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
1781                 return (error);
1782 
1783         mutex_enter(&zp->z_acl_lock);
1784 
1785         error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1786         if (error != 0) {
1787                 mutex_exit(&zp->z_acl_lock);
1788                 return (error);
1789         }
1790 
1791         /*
1792          * Scan ACL to determine number of ACEs
1793          */
1794         if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1795                 void *zacep = NULL;
1796                 uint64_t who;
1797                 uint32_t access_mask;
1798                 uint16_t type, iflags;
1799 
1800                 while (zacep = zfs_acl_next_ace(aclp, zacep,
1801                     &who, &access_mask, &iflags, &type)) {
1802                         switch (type) {
1803                         case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1804                         case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1805                         case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1806                         case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1807                                 largeace++;
1808                                 continue;
1809                         default:
1810                                 count++;
1811                         }
1812                 }
1813                 vsecp->vsa_aclcnt = count;
1814         } else
1815                 count = (int)aclp->z_acl_count;
1816 
1817         if (mask & VSA_ACECNT) {
1818                 vsecp->vsa_aclcnt = count;
1819         }
1820 
1821         if (mask & VSA_ACE) {
1822                 size_t aclsz;
1823 
1824                 aclsz = count * sizeof (ace_t) +
1825                     sizeof (ace_object_t) * largeace;
1826 
1827                 vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1828                 vsecp->vsa_aclentsz = aclsz;
1829 
1830                 if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1831                         zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1832                             vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1833                 else {
1834                         zfs_acl_node_t *aclnode;
1835                         void *start = vsecp->vsa_aclentp;
1836 
1837                         for (aclnode = list_head(&aclp->z_acl); aclnode;
1838                             aclnode = list_next(&aclp->z_acl, aclnode)) {
1839                                 bcopy(aclnode->z_acldata, start,
1840                                     aclnode->z_size);
1841                                 start = (caddr_t)start + aclnode->z_size;
1842                         }
1843                         ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
1844                             aclp->z_acl_bytes);
1845                 }
1846         }
1847         if (mask & VSA_ACE_ACLFLAGS) {
1848                 vsecp->vsa_aclflags = 0;
1849                 if (zp->z_pflags & ZFS_ACL_DEFAULTED)
1850                         vsecp->vsa_aclflags |= ACL_DEFAULTED;
1851                 if (zp->z_pflags & ZFS_ACL_PROTECTED)
1852                         vsecp->vsa_aclflags |= ACL_PROTECTED;
1853                 if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
1854                         vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1855         }
1856 
1857         mutex_exit(&zp->z_acl_lock);
1858 
1859         return (0);
1860 }
1861 
1862 int
1863 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
1864     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
1865 {
1866         zfs_acl_t *aclp;
1867         zfs_acl_node_t *aclnode;
1868         int aclcnt = vsecp->vsa_aclcnt;
1869         int error;
1870 
1871         if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1872                 return (SET_ERROR(EINVAL));
1873 
1874         aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1875 
1876         aclp->z_hints = 0;
1877         aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
1878         if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1879                 if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
1880                     (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
1881                     aclcnt, &aclnode->z_size)) != 0) {
1882                         zfs_acl_free(aclp);
1883                         zfs_acl_node_free(aclnode);
1884                         return (error);
1885                 }
1886         } else {
1887                 if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
1888                     vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
1889                     &aclnode->z_size, fuidp, cr)) != 0) {
1890                         zfs_acl_free(aclp);
1891                         zfs_acl_node_free(aclnode);
1892                         return (error);
1893                 }
1894         }
1895         aclp->z_acl_bytes = aclnode->z_size;
1896         aclnode->z_ace_count = aclcnt;
1897         aclp->z_acl_count = aclcnt;
1898         list_insert_head(&aclp->z_acl, aclnode);
1899 
1900         /*
1901          * If flags are being set then add them to z_hints
1902          */
1903         if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
1904                 if (vsecp->vsa_aclflags & ACL_PROTECTED)
1905                         aclp->z_hints |= ZFS_ACL_PROTECTED;
1906                 if (vsecp->vsa_aclflags & ACL_DEFAULTED)
1907                         aclp->z_hints |= ZFS_ACL_DEFAULTED;
1908                 if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
1909                         aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1910         }
1911 
1912         *zaclp = aclp;
1913 
1914         return (0);
1915 }
1916 
1917 /*
1918  * Set a file's ACL
1919  */
1920 int
1921 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1922 {
1923         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
1924         zilog_t         *zilog = zfsvfs->z_log;
1925         ulong_t         mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
1926         dmu_tx_t        *tx;
1927         int             error;
1928         zfs_acl_t       *aclp;
1929         zfs_fuid_info_t *fuidp = NULL;
1930         boolean_t       fuid_dirtied;
1931         uint64_t        acl_obj;
1932 
1933         if (mask == 0)
1934                 return (SET_ERROR(ENOSYS));
1935 
1936         if (zp->z_pflags & ZFS_IMMUTABLE)
1937                 return (SET_ERROR(EPERM));
1938 
1939         if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
1940                 return (error);
1941 
1942         error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
1943             &aclp);
1944         if (error)
1945                 return (error);
1946 
1947         /*
1948          * If ACL wide flags aren't being set then preserve any
1949          * existing flags.
1950          */
1951         if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
1952                 aclp->z_hints |=
1953                     (zp->z_pflags & V4_ACL_WIDE_FLAGS);
1954         }
1955 top:
1956         mutex_enter(&zp->z_acl_lock);
1957         mutex_enter(&zp->z_lock);
1958 
1959         tx = dmu_tx_create(zfsvfs->z_os);
1960 
1961         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1962 
1963         fuid_dirtied = zfsvfs->z_fuid_dirty;
1964         if (fuid_dirtied)
1965                 zfs_fuid_txhold(zfsvfs, tx);
1966 
1967         /*
1968          * If old version and ACL won't fit in bonus and we aren't
1969          * upgrading then take out necessary DMU holds
1970          */
1971 
1972         if ((acl_obj = zfs_external_acl(zp)) != 0) {
1973                 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
1974                     zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
1975                         dmu_tx_hold_free(tx, acl_obj, 0,
1976                             DMU_OBJECT_END);
1977                         dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1978                             aclp->z_acl_bytes);
1979                 } else {
1980                         dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
1981                 }
1982         } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1983                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
1984         }
1985 
1986         zfs_sa_upgrade_txholds(tx, zp);
1987         error = dmu_tx_assign(tx, TXG_NOWAIT);
1988         if (error) {
1989                 mutex_exit(&zp->z_acl_lock);
1990                 mutex_exit(&zp->z_lock);
1991 
1992                 if (error == ERESTART) {
1993                         dmu_tx_wait(tx);
1994                         dmu_tx_abort(tx);
1995                         goto top;
1996                 }
1997                 dmu_tx_abort(tx);
1998                 zfs_acl_free(aclp);
1999                 return (error);
2000         }
2001 
2002         error = zfs_aclset_common(zp, aclp, cr, tx);
2003         ASSERT(error == 0);
2004         ASSERT(zp->z_acl_cached == NULL);
2005         zp->z_acl_cached = aclp;
2006 
2007         if (fuid_dirtied)
2008                 zfs_fuid_sync(zfsvfs, tx);
2009 
2010         zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2011 
2012         if (fuidp)
2013                 zfs_fuid_info_free(fuidp);
2014         dmu_tx_commit(tx);
2015 done:
2016         mutex_exit(&zp->z_lock);
2017         mutex_exit(&zp->z_acl_lock);
2018 
2019         return (error);
2020 }
2021 
2022 /*
2023  * Check accesses of interest (AoI) against attributes of the dataset
2024  * such as read-only.  Returns zero if no AoI conflict with dataset
2025  * attributes, otherwise an appropriate errno is returned.
2026  */
2027 static int
2028 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2029 {
2030         if ((v4_mode & WRITE_MASK) &&
2031             (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2032             (!IS_DEVVP(ZTOV(zp)) ||
2033             (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2034                 return (SET_ERROR(EROFS));
2035         }
2036 
2037         /*
2038          * Intentionally allow ZFS_READONLY through here.
2039          * See zfs_zaccess_common().
2040          */
2041         if ((v4_mode & WRITE_MASK_DATA) &&
2042             (zp->z_pflags & ZFS_IMMUTABLE)) {
2043                 return (SET_ERROR(EPERM));
2044         }
2045 
2046         if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2047             (zp->z_pflags & ZFS_NOUNLINK)) {
2048                 return (SET_ERROR(EPERM));
2049         }
2050 
2051         if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2052             (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2053                 return (SET_ERROR(EACCES));
2054         }
2055 
2056         return (0);
2057 }
2058 
2059 /*
2060  * The primary usage of this function is to loop through all of the
2061  * ACEs in the znode, determining what accesses of interest (AoI) to
2062  * the caller are allowed or denied.  The AoI are expressed as bits in
2063  * the working_mode parameter.  As each ACE is processed, bits covered
2064  * by that ACE are removed from the working_mode.  This removal
2065  * facilitates two things.  The first is that when the working mode is
2066  * empty (= 0), we know we've looked at all the AoI. The second is
2067  * that the ACE interpretation rules don't allow a later ACE to undo
2068  * something granted or denied by an earlier ACE.  Removing the
2069  * discovered access or denial enforces this rule.  At the end of
2070  * processing the ACEs, all AoI that were found to be denied are
2071  * placed into the working_mode, giving the caller a mask of denied
2072  * accesses.  Returns:
2073  *      0               if all AoI granted
2074  *      EACCES          if the denied mask is non-zero
2075  *      other error     if abnormal failure (e.g., IO error)
2076  *
2077  * A secondary usage of the function is to determine if any of the
2078  * AoI are granted.  If an ACE grants any access in
2079  * the working_mode, we immediately short circuit out of the function.
2080  * This mode is chosen by setting anyaccess to B_TRUE.  The
2081  * working_mode is not a denied access mask upon exit if the function
2082  * is used in this manner.
2083  */
2084 static int
2085 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2086     boolean_t anyaccess, cred_t *cr)
2087 {
2088         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
2089         zfs_acl_t       *aclp;
2090         int             error;
2091         uid_t           uid = crgetuid(cr);
2092         uint64_t        who;
2093         uint16_t        type, iflags;
2094         uint16_t        entry_type;
2095         uint32_t        access_mask;
2096         uint32_t        deny_mask = 0;
2097         zfs_ace_hdr_t   *acep = NULL;
2098         boolean_t       checkit;
2099         uid_t           gowner;
2100         uid_t           fowner;
2101 
2102         zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2103 
2104         mutex_enter(&zp->z_acl_lock);
2105 
2106         error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
2107         if (error != 0) {
2108                 mutex_exit(&zp->z_acl_lock);
2109                 return (error);
2110         }
2111 
2112         ASSERT(zp->z_acl_cached);
2113 
2114         while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2115             &iflags, &type)) {
2116                 uint32_t mask_matched;
2117 
2118                 if (!zfs_acl_valid_ace_type(type, iflags))
2119                         continue;
2120 
2121                 if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2122                         continue;
2123 
2124                 /* Skip ACE if it does not affect any AoI */
2125                 mask_matched = (access_mask & *working_mode);
2126                 if (!mask_matched)
2127                         continue;
2128 
2129                 entry_type = (iflags & ACE_TYPE_FLAGS);
2130 
2131                 checkit = B_FALSE;
2132 
2133                 switch (entry_type) {
2134                 case ACE_OWNER:
2135                         if (uid == fowner)
2136                                 checkit = B_TRUE;
2137                         break;
2138                 case OWNING_GROUP:
2139                         who = gowner;
2140                         /*FALLTHROUGH*/
2141                 case ACE_IDENTIFIER_GROUP:
2142                         checkit = zfs_groupmember(zfsvfs, who, cr);
2143                         break;
2144                 case ACE_EVERYONE:
2145                         checkit = B_TRUE;
2146                         break;
2147 
2148                 /* USER Entry */
2149                 default:
2150                         if (entry_type == 0) {
2151                                 uid_t newid;
2152 
2153                                 newid = zfs_fuid_map_id(zfsvfs, who, cr,
2154                                     ZFS_ACE_USER);
2155                                 if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2156                                     uid == newid)
2157                                         checkit = B_TRUE;
2158                                 break;
2159                         } else {
2160                                 mutex_exit(&zp->z_acl_lock);
2161                                 return (SET_ERROR(EIO));
2162                         }
2163                 }
2164 
2165                 if (checkit) {
2166                         if (type == DENY) {
2167                                 DTRACE_PROBE3(zfs__ace__denies,
2168                                     znode_t *, zp,
2169                                     zfs_ace_hdr_t *, acep,
2170                                     uint32_t, mask_matched);
2171                                 deny_mask |= mask_matched;
2172                         } else {
2173                                 DTRACE_PROBE3(zfs__ace__allows,
2174                                     znode_t *, zp,
2175                                     zfs_ace_hdr_t *, acep,
2176                                     uint32_t, mask_matched);
2177                                 if (anyaccess) {
2178                                         mutex_exit(&zp->z_acl_lock);
2179                                         return (0);
2180                                 }
2181                         }
2182                         *working_mode &= ~mask_matched;
2183                 }
2184 
2185                 /* Are we done? */
2186                 if (*working_mode == 0)
2187                         break;
2188         }
2189 
2190         mutex_exit(&zp->z_acl_lock);
2191 
2192         /* Put the found 'denies' back on the working mode */
2193         if (deny_mask) {
2194                 *working_mode |= deny_mask;
2195                 return (SET_ERROR(EACCES));
2196         } else if (*working_mode) {
2197                 return (-1);
2198         }
2199 
2200         return (0);
2201 }
2202 
2203 /*
2204  * Return true if any access whatsoever granted, we don't actually
2205  * care what access is granted.
2206  */
2207 boolean_t
2208 zfs_has_access(znode_t *zp, cred_t *cr)
2209 {
2210         uint32_t have = ACE_ALL_PERMS;
2211 
2212         if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2213                 uid_t owner;
2214 
2215                 owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2216                 return (secpolicy_vnode_any_access(cr, ZTOV(zp), owner) == 0);
2217         }
2218         return (B_TRUE);
2219 }
2220 
2221 static int
2222 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2223     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2224 {
2225         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2226         int err;
2227 
2228         *working_mode = v4_mode;
2229         *check_privs = B_TRUE;
2230 
2231         /*
2232          * Short circuit empty requests
2233          */
2234         if (v4_mode == 0 || zfsvfs->z_replay) {
2235                 *working_mode = 0;
2236                 return (0);
2237         }
2238 
2239         if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2240                 *check_privs = B_FALSE;
2241                 return (err);
2242         }
2243 
2244         /*
2245          * The caller requested that the ACL check be skipped.  This
2246          * would only happen if the caller checked VOP_ACCESS() with a
2247          * 32 bit ACE mask and already had the appropriate permissions.
2248          */
2249         if (skipaclchk) {
2250                 *working_mode = 0;
2251                 return (0);
2252         }
2253 
2254         /*
2255          * Note: ZFS_READONLY represents the "DOS R/O" attribute.
2256          * When that flag is set, we should behave as if write access
2257          * were not granted by anything in the ACL.  In particular:
2258          * We _must_ allow writes after opening the file r/w, then
2259          * setting the DOS R/O attribute, and writing some more.
2260          * (Similar to how you can write after fchmod(fd, 0444).)
2261          *
2262          * Therefore ZFS_READONLY is ignored in the dataset check
2263          * above, and checked here as if part of the ACL check.
2264          * Also note: DOS R/O is ignored for directories.
2265          */
2266         if ((v4_mode & WRITE_MASK_DATA) &&
2267             (ZTOV(zp)->v_type != VDIR) &&
2268             (zp->z_pflags & ZFS_READONLY)) {
2269                 return (SET_ERROR(EPERM));
2270         }
2271 
2272         return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2273 }
2274 
2275 static int
2276 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2277     cred_t *cr)
2278 {
2279         if (*working_mode != ACE_WRITE_DATA)
2280                 return (SET_ERROR(EACCES));
2281 
2282         return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2283             check_privs, B_FALSE, cr));
2284 }
2285 
2286 int
2287 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2288 {
2289         boolean_t owner = B_FALSE;
2290         boolean_t groupmbr = B_FALSE;
2291         boolean_t is_attr;
2292         uid_t uid = crgetuid(cr);
2293         int error;
2294 
2295         if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2296                 return (SET_ERROR(EACCES));
2297 
2298         is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2299             (ZTOV(zdp)->v_type == VDIR));
2300         if (is_attr)
2301                 goto slow;
2302 
2303 
2304         mutex_enter(&zdp->z_acl_lock);
2305 
2306         if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2307                 mutex_exit(&zdp->z_acl_lock);
2308                 return (0);
2309         }
2310 
2311         if (FUID_INDEX(zdp->z_uid) != 0 || FUID_INDEX(zdp->z_gid) != 0) {
2312                 mutex_exit(&zdp->z_acl_lock);
2313                 goto slow;
2314         }
2315 
2316         if (uid == zdp->z_uid) {
2317                 owner = B_TRUE;
2318                 if (zdp->z_mode & S_IXUSR) {
2319                         mutex_exit(&zdp->z_acl_lock);
2320                         return (0);
2321                 } else {
2322                         mutex_exit(&zdp->z_acl_lock);
2323                         goto slow;
2324                 }
2325         }
2326         if (groupmember(zdp->z_gid, cr)) {
2327                 groupmbr = B_TRUE;
2328                 if (zdp->z_mode & S_IXGRP) {
2329                         mutex_exit(&zdp->z_acl_lock);
2330                         return (0);
2331                 } else {
2332                         mutex_exit(&zdp->z_acl_lock);
2333                         goto slow;
2334                 }
2335         }
2336         if (!owner && !groupmbr) {
2337                 if (zdp->z_mode & S_IXOTH) {
2338                         mutex_exit(&zdp->z_acl_lock);
2339                         return (0);
2340                 }
2341         }
2342 
2343         mutex_exit(&zdp->z_acl_lock);
2344 
2345 slow:
2346         DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2347         ZFS_ENTER(zdp->z_zfsvfs);
2348         error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
2349         ZFS_EXIT(zdp->z_zfsvfs);
2350         return (error);
2351 }
2352 
2353 /*
2354  * Determine whether Access should be granted/denied.
2355  *
2356  * The least priv subsytem is always consulted as a basic privilege
2357  * can define any form of access.
2358  */
2359 int
2360 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2361 {
2362         uint32_t        working_mode;
2363         int             error;
2364         int             is_attr;
2365         boolean_t       check_privs;
2366         znode_t         *xzp;
2367         znode_t         *check_zp = zp;
2368         mode_t          needed_bits;
2369         uid_t           owner;
2370 
2371         is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2372 
2373         /*
2374          * If attribute then validate against base file
2375          */
2376         if (is_attr) {
2377                 uint64_t        parent;
2378 
2379                 if ((error = sa_lookup(zp->z_sa_hdl,
2380                     SA_ZPL_PARENT(zp->z_zfsvfs), &parent,
2381                     sizeof (parent))) != 0)
2382                         return (error);
2383 
2384                 if ((error = zfs_zget(zp->z_zfsvfs,
2385                     parent, &xzp)) != 0)    {
2386                         return (error);
2387                 }
2388 
2389                 check_zp = xzp;
2390 
2391                 /*
2392                  * fixup mode to map to xattr perms
2393                  */
2394 
2395                 if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2396                         mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2397                         mode |= ACE_WRITE_NAMED_ATTRS;
2398                 }
2399 
2400                 if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2401                         mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2402                         mode |= ACE_READ_NAMED_ATTRS;
2403                 }
2404         }
2405 
2406         owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2407         /*
2408          * Map the bits required to the standard vnode flags VREAD|VWRITE|VEXEC
2409          * in needed_bits.  Map the bits mapped by working_mode (currently
2410          * missing) in missing_bits.
2411          * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2412          * needed_bits.
2413          */
2414         needed_bits = 0;
2415 
2416         working_mode = mode;
2417         if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2418             owner == crgetuid(cr))
2419                 working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2420 
2421         if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2422             ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2423                 needed_bits |= VREAD;
2424         if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2425             ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2426                 needed_bits |= VWRITE;
2427         if (working_mode & ACE_EXECUTE)
2428                 needed_bits |= VEXEC;
2429 
2430         if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2431             &check_privs, skipaclchk, cr)) == 0) {
2432                 if (is_attr)
2433                         VN_RELE(ZTOV(xzp));
2434                 return (secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2435                     needed_bits, needed_bits));
2436         }
2437 
2438         if (error && !check_privs) {
2439                 if (is_attr)
2440                         VN_RELE(ZTOV(xzp));
2441                 return (error);
2442         }
2443 
2444         if (error && (flags & V_APPEND)) {
2445                 error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2446         }
2447 
2448         if (error && check_privs) {
2449                 mode_t          checkmode = 0;
2450 
2451                 /*
2452                  * First check for implicit owner permission on
2453                  * read_acl/read_attributes
2454                  */
2455 
2456                 error = 0;
2457                 ASSERT(working_mode != 0);
2458 
2459                 if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2460                     owner == crgetuid(cr)))
2461                         working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2462 
2463                 if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2464                     ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2465                         checkmode |= VREAD;
2466                 if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2467                     ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2468                         checkmode |= VWRITE;
2469                 if (working_mode & ACE_EXECUTE)
2470                         checkmode |= VEXEC;
2471 
2472                 error = secpolicy_vnode_access2(cr, ZTOV(check_zp), owner,
2473                     needed_bits & ~checkmode, needed_bits);
2474 
2475                 if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2476                         error = secpolicy_vnode_chown(cr, owner);
2477                 if (error == 0 && (working_mode & ACE_WRITE_ACL))
2478                         error = secpolicy_vnode_setdac(cr, owner);
2479 
2480                 if (error == 0 && (working_mode &
2481                     (ACE_DELETE|ACE_DELETE_CHILD)))
2482                         error = secpolicy_vnode_remove(cr);
2483 
2484                 if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2485                         error = secpolicy_vnode_chown(cr, owner);
2486                 }
2487                 if (error == 0) {
2488                         /*
2489                          * See if any bits other than those already checked
2490                          * for are still present.  If so then return EACCES
2491                          */
2492                         if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2493                                 error = SET_ERROR(EACCES);
2494                         }
2495                 }
2496         } else if (error == 0) {
2497                 error = secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2498                     needed_bits, needed_bits);
2499         }
2500 
2501 
2502         if (is_attr)
2503                 VN_RELE(ZTOV(xzp));
2504 
2505         return (error);
2506 }
2507 
2508 /*
2509  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2510  * native ACL format and call zfs_zaccess()
2511  */
2512 int
2513 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2514 {
2515         return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2516 }
2517 
2518 /*
2519  * Access function for secpolicy_vnode_setattr
2520  */
2521 int
2522 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2523 {
2524         int v4_mode = zfs_unix_to_v4(mode >> 6);
2525 
2526         return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2527 }
2528 
2529 /* See zfs_zaccess_delete() */
2530 int zfs_write_implies_delete_child = 1;
2531 
2532 /*
2533  * Determine whether delete access should be granted.
2534  *
2535  * The following chart outlines how we handle delete permissions which is
2536  * how recent versions of windows (Windows 2008) handles it.  The efficiency
2537  * comes from not having to check the parent ACL where the object itself grants
2538  * delete:
2539  *
2540  *      -------------------------------------------------------
2541  *      |   Parent Dir  |      Target Object Permissions      |
2542  *      |  permissions  |                                     |
2543  *      -------------------------------------------------------
2544  *      |               | ACL Allows | ACL Denies| Delete     |
2545  *      |               |  Delete    |  Delete   | unspecified|
2546  *      -------------------------------------------------------
2547  *      | ACL Allows    | Permit     | Deny *    | Permit     |
2548  *      | DELETE_CHILD  |            |           |            |
2549  *      -------------------------------------------------------
2550  *      | ACL Denies    | Permit     | Deny      | Deny       |
2551  *      | DELETE_CHILD  |            |           |            |
2552  *      -------------------------------------------------------
2553  *      | ACL specifies |            |           |            |
2554  *      | only allow    | Permit     | Deny *    | Permit     |
2555  *      | write and     |            |           |            |
2556  *      | execute       |            |           |            |
2557  *      -------------------------------------------------------
2558  *      | ACL denies    |            |           |            |
2559  *      | write and     | Permit     | Deny      | Deny       |
2560  *      | execute       |            |           |            |
2561  *      -------------------------------------------------------
2562  *         ^
2563  *         |
2564  *         Re. execute permission on the directory:  if that's missing,
2565  *         the vnode lookup of the target will fail before we get here.
2566  *
2567  * Re [*] in the table above:  NFSv4 would normally Permit delete for
2568  * these two cells of the matrix.
2569  * See acl.h for notes on which ACE_... flags should be checked for which
2570  * operations.  Specifically, the NFSv4 committee recommendation is in
2571  * conflict with the Windows interpretation of DENY ACEs, where DENY ACEs
2572  * should take precedence ahead of ALLOW ACEs.
2573  *
2574  * This implementation always consults the target object's ACL first.
2575  * If a DENY ACE is present on the target object that specifies ACE_DELETE,
2576  * delete access is denied.  If an ALLOW ACE with ACE_DELETE is present on
2577  * the target object, access is allowed.  If and only if no entries with
2578  * ACE_DELETE are present in the object's ACL, check the container's ACL
2579  * for entries with ACE_DELETE_CHILD.
2580  *
2581  * A summary of the logic implemented from the table above is as follows:
2582  *
2583  * First check for DENY ACEs that apply.
2584  * If either target or container has a deny, EACCES.
2585  *
2586  * Delete access can then be summarized as follows:
2587  * 1: The object to be deleted grants ACE_DELETE, or
2588  * 2: The containing directory grants ACE_DELETE_CHILD.
2589  * In a Windows system, that would be the end of the story.
2590  * In this system, (2) has some complications...
2591  * 2a: "sticky" bit on a directory adds restrictions, and
2592  * 2b: existing ACEs from previous versions of ZFS may
2593  * not carry ACE_DELETE_CHILD where they should, so we
2594  * also allow delete when ACE_WRITE_DATA is granted.
2595  *
2596  * Note: 2b is technically a work-around for a prior bug,
2597  * which hopefully can go away some day.  For those who
2598  * no longer need the work around, and for testing, this
2599  * work-around is made conditional via the tunable:
2600  * zfs_write_implies_delete_child
2601  */
2602 int
2603 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2604 {
2605         uint32_t wanted_dirperms;
2606         uint32_t dzp_working_mode = 0;
2607         uint32_t zp_working_mode = 0;
2608         int dzp_error, zp_error;
2609         boolean_t dzpcheck_privs;
2610         boolean_t zpcheck_privs;
2611 
2612         if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2613                 return (SET_ERROR(EPERM));
2614 
2615         /*
2616          * Case 1:
2617          * If target object grants ACE_DELETE then we are done.  This is
2618          * indicated by a return value of 0.  For this case we don't worry
2619          * about the sticky bit because sticky only applies to the parent
2620          * directory and this is the child access result.
2621          *
2622          * If we encounter a DENY ACE here, we're also done (EACCES).
2623          * Note that if we hit a DENY ACE here (on the target) it should
2624          * take precedence over a DENY ACE on the container, so that when
2625          * we have more complete auditing support we will be able to
2626          * report an access failure against the specific target.
2627          * (This is part of why we're checking the target first.)
2628          */
2629         zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2630             &zpcheck_privs, B_FALSE, cr);
2631         if (zp_error == EACCES) {
2632                 /* We hit a DENY ACE. */
2633                 if (!zpcheck_privs)
2634                         return (SET_ERROR(zp_error));
2635                 return (secpolicy_vnode_remove(cr));
2636 
2637         }
2638         if (zp_error == 0)
2639                 return (0);
2640 
2641         /*
2642          * Case 2:
2643          * If the containing directory grants ACE_DELETE_CHILD,
2644          * or we're in backward compatibility mode and the
2645          * containing directory has ACE_WRITE_DATA, allow.
2646          * Case 2b is handled with wanted_dirperms.
2647          */
2648         wanted_dirperms = ACE_DELETE_CHILD;
2649         if (zfs_write_implies_delete_child)
2650                 wanted_dirperms |= ACE_WRITE_DATA;
2651         dzp_error = zfs_zaccess_common(dzp, wanted_dirperms,
2652             &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2653         if (dzp_error == EACCES) {
2654                 /* We hit a DENY ACE. */
2655                 if (!dzpcheck_privs)
2656                         return (SET_ERROR(dzp_error));
2657                 return (secpolicy_vnode_remove(cr));
2658         }
2659 
2660         /*
2661          * Cases 2a, 2b (continued)
2662          *
2663          * Note: dzp_working_mode now contains any permissions
2664          * that were NOT granted.  Therefore, if any of the
2665          * wanted_dirperms WERE granted, we will have:
2666          *   dzp_working_mode != wanted_dirperms
2667          * We're really asking if ANY of those permissions
2668          * were granted, and if so, grant delete access.
2669          */
2670         if (dzp_working_mode != wanted_dirperms)
2671                 dzp_error = 0;
2672 
2673         /*
2674          * dzp_error is 0 if the container granted us permissions to "modify".
2675          * If we do not have permission via one or more ACEs, our current
2676          * privileges may still permit us to modify the container.
2677          *
2678          * dzpcheck_privs is false when i.e. the FS is read-only.
2679          * Otherwise, do privilege checks for the container.
2680          */
2681         if (dzp_error != 0 && dzpcheck_privs) {
2682                 uid_t owner;
2683 
2684                 /*
2685                  * The secpolicy call needs the requested access and
2686                  * the current access mode of the container, but it
2687                  * only knows about Unix-style modes (VEXEC, VWRITE),
2688                  * so this must condense the fine-grained ACE bits into
2689                  * Unix modes.
2690                  *
2691                  * The VEXEC flag is easy, because we know that has
2692                  * always been checked before we get here (during the
2693                  * lookup of the target vnode).  The container has not
2694                  * granted us permissions to "modify", so we do not set
2695                  * the VWRITE flag in the current access mode.
2696                  */
2697                 owner = zfs_fuid_map_id(dzp->z_zfsvfs, dzp->z_uid, cr,
2698                     ZFS_OWNER);
2699                 dzp_error = secpolicy_vnode_access2(cr, ZTOV(dzp),
2700                     owner, VEXEC, VWRITE|VEXEC);
2701         }
2702         if (dzp_error != 0) {
2703                 /*
2704                  * Note: We may have dzp_error = -1 here (from
2705                  * zfs_zacess_common).  Don't return that.
2706                  */
2707                 return (SET_ERROR(EACCES));
2708         }
2709 
2710         /*
2711          * At this point, we know that the directory permissions allow
2712          * us to modify, but we still need to check for the additional
2713          * restrictions that apply when the "sticky bit" is set.
2714          *
2715          * Yes, zfs_sticky_remove_access() also checks this bit, but
2716          * checking it here and skipping the call below is nice when
2717          * you're watching all of this with dtrace.
2718          */
2719         if ((dzp->z_mode & S_ISVTX) == 0)
2720                 return (0);
2721 
2722         /*
2723          * zfs_sticky_remove_access will succeed if:
2724          * 1. The sticky bit is absent.
2725          * 2. We pass the sticky bit restrictions.
2726          * 3. We have privileges that always allow file removal.
2727          */
2728         return (zfs_sticky_remove_access(dzp, zp, cr));
2729 }
2730 
2731 int
2732 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2733     znode_t *tzp, cred_t *cr)
2734 {
2735         int add_perm;
2736         int error;
2737 
2738         if (szp->z_pflags & ZFS_AV_QUARANTINED)
2739                 return (SET_ERROR(EACCES));
2740 
2741         add_perm = (ZTOV(szp)->v_type == VDIR) ?
2742             ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2743 
2744         /*
2745          * Rename permissions are combination of delete permission +
2746          * add file/subdir permission.
2747          */
2748 
2749         /*
2750          * first make sure we do the delete portion.
2751          *
2752          * If that succeeds then check for add_file/add_subdir permissions
2753          */
2754 
2755         if (error = zfs_zaccess_delete(sdzp, szp, cr))
2756                 return (error);
2757 
2758         /*
2759          * If we have a tzp, see if we can delete it?
2760          */
2761         if (tzp) {
2762                 if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2763                         return (error);
2764         }
2765 
2766         /*
2767          * Now check for add permissions
2768          */
2769         error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2770 
2771         return (error);
2772 }