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  */
  25 
  26 /* Portions Copyright 2007 Jeremy Teo */
  27 
  28 #ifdef _KERNEL
  29 #include <sys/types.h>
  30 #include <sys/param.h>
  31 #include <sys/time.h>
  32 #include <sys/systm.h>
  33 #include <sys/sysmacros.h>
  34 #include <sys/resource.h>
  35 #include <sys/mntent.h>
  36 #include <sys/mkdev.h>
  37 #include <sys/u8_textprep.h>
  38 #include <sys/dsl_dataset.h>
  39 #include <sys/vfs.h>
  40 #include <sys/vfs_opreg.h>
  41 #include <sys/vnode.h>
  42 #include <sys/file.h>
  43 #include <sys/kmem.h>
  44 #include <sys/errno.h>
  45 #include <sys/unistd.h>
  46 #include <sys/mode.h>
  47 #include <sys/atomic.h>
  48 #include <vm/pvn.h>
  49 #include "fs/fs_subr.h"
  50 #include <sys/zfs_dir.h>
  51 #include <sys/zfs_acl.h>
  52 #include <sys/zfs_ioctl.h>
  53 #include <sys/zfs_rlock.h>
  54 #include <sys/zfs_fuid.h>
  55 #include <sys/dnode.h>
  56 #include <sys/fs/zfs.h>
  57 #include <sys/kidmap.h>
  58 #endif /* _KERNEL */
  59 
  60 #include <sys/dmu.h>
  61 #include <sys/refcount.h>
  62 #include <sys/stat.h>
  63 #include <sys/zap.h>
  64 #include <sys/zfs_znode.h>
  65 #include <sys/sa.h>
  66 #include <sys/zfs_sa.h>
  67 #include <sys/zfs_stat.h>
  68 
  69 #include "zfs_prop.h"
  70 #include "zfs_comutil.h"
  71 
  72 /*
  73  * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
  74  * turned on when DEBUG is also defined.
  75  */
  76 #ifdef  DEBUG
  77 #define ZNODE_STATS
  78 #endif  /* DEBUG */
  79 
  80 #ifdef  ZNODE_STATS
  81 #define ZNODE_STAT_ADD(stat)                    ((stat)++)
  82 #else
  83 #define ZNODE_STAT_ADD(stat)                    /* nothing */
  84 #endif  /* ZNODE_STATS */
  85 
  86 /*
  87  * Functions needed for userland (ie: libzpool) are not put under
  88  * #ifdef_KERNEL; the rest of the functions have dependencies
  89  * (such as VFS logic) that will not compile easily in userland.
  90  */
  91 #ifdef _KERNEL
  92 /*
  93  * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
  94  * be freed before it can be safely accessed.
  95  */
  96 krwlock_t zfsvfs_lock;
  97 
  98 static kmem_cache_t *znode_cache = NULL;
  99 
 100 /*ARGSUSED*/
 101 static void
 102 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
 103 {
 104         /*
 105          * We should never drop all dbuf refs without first clearing
 106          * the eviction callback.
 107          */
 108         panic("evicting znode %p\n", user_ptr);
 109 }
 110 
 111 /*ARGSUSED*/
 112 static int
 113 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
 114 {
 115         znode_t *zp = buf;
 116 
 117         ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
 118 
 119         zp->z_vnode = vn_alloc(kmflags);
 120         if (zp->z_vnode == NULL) {
 121                 return (-1);
 122         }
 123         ZTOV(zp)->v_data = zp;
 124 
 125         list_link_init(&zp->z_link_node);
 126 
 127         mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
 128         rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
 129         rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
 130         mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
 131 
 132         mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
 133         avl_create(&zp->z_range_avl, zfs_range_compare,
 134             sizeof (rl_t), offsetof(rl_t, r_node));
 135 
 136         zp->z_dirlocks = NULL;
 137         zp->z_acl_cached = NULL;
 138         zp->z_moved = 0;
 139         return (0);
 140 }
 141 
 142 /*ARGSUSED*/
 143 static void
 144 zfs_znode_cache_destructor(void *buf, void *arg)
 145 {
 146         znode_t *zp = buf;
 147 
 148         ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
 149         ASSERT(ZTOV(zp)->v_data == zp);
 150         vn_free(ZTOV(zp));
 151         ASSERT(!list_link_active(&zp->z_link_node));
 152         mutex_destroy(&zp->z_lock);
 153         rw_destroy(&zp->z_parent_lock);
 154         rw_destroy(&zp->z_name_lock);
 155         mutex_destroy(&zp->z_acl_lock);
 156         avl_destroy(&zp->z_range_avl);
 157         mutex_destroy(&zp->z_range_lock);
 158 
 159         ASSERT(zp->z_dirlocks == NULL);
 160         ASSERT(zp->z_acl_cached == NULL);
 161 }
 162 
 163 #ifdef  ZNODE_STATS
 164 static struct {
 165         uint64_t zms_zfsvfs_invalid;
 166         uint64_t zms_zfsvfs_recheck1;
 167         uint64_t zms_zfsvfs_unmounted;
 168         uint64_t zms_zfsvfs_recheck2;
 169         uint64_t zms_obj_held;
 170         uint64_t zms_vnode_locked;
 171         uint64_t zms_not_only_dnlc;
 172 } znode_move_stats;
 173 #endif  /* ZNODE_STATS */
 174 
 175 static void
 176 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
 177 {
 178         vnode_t *vp;
 179 
 180         /* Copy fields. */
 181         nzp->z_zfsvfs = ozp->z_zfsvfs;
 182 
 183         /* Swap vnodes. */
 184         vp = nzp->z_vnode;
 185         nzp->z_vnode = ozp->z_vnode;
 186         ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
 187         ZTOV(ozp)->v_data = ozp;
 188         ZTOV(nzp)->v_data = nzp;
 189 
 190         nzp->z_id = ozp->z_id;
 191         ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
 192         ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
 193         nzp->z_unlinked = ozp->z_unlinked;
 194         nzp->z_atime_dirty = ozp->z_atime_dirty;
 195         nzp->z_zn_prefetch = ozp->z_zn_prefetch;
 196         nzp->z_blksz = ozp->z_blksz;
 197         nzp->z_seq = ozp->z_seq;
 198         nzp->z_mapcnt = ozp->z_mapcnt;
 199         nzp->z_gen = ozp->z_gen;
 200         nzp->z_sync_cnt = ozp->z_sync_cnt;
 201         nzp->z_is_sa = ozp->z_is_sa;
 202         nzp->z_sa_hdl = ozp->z_sa_hdl;
 203         bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
 204         nzp->z_links = ozp->z_links;
 205         nzp->z_size = ozp->z_size;
 206         nzp->z_pflags = ozp->z_pflags;
 207         nzp->z_uid = ozp->z_uid;
 208         nzp->z_gid = ozp->z_gid;
 209         nzp->z_mode = ozp->z_mode;
 210 
 211         /*
 212          * Since this is just an idle znode and kmem is already dealing with
 213          * memory pressure, release any cached ACL.
 214          */
 215         if (ozp->z_acl_cached) {
 216                 zfs_acl_free(ozp->z_acl_cached);
 217                 ozp->z_acl_cached = NULL;
 218         }
 219 
 220         sa_set_userp(nzp->z_sa_hdl, nzp);
 221 
 222         /*
 223          * Invalidate the original znode by clearing fields that provide a
 224          * pointer back to the znode. Set the low bit of the vfs pointer to
 225          * ensure that zfs_znode_move() recognizes the znode as invalid in any
 226          * subsequent callback.
 227          */
 228         ozp->z_sa_hdl = NULL;
 229         POINTER_INVALIDATE(&ozp->z_zfsvfs);
 230 
 231         /*
 232          * Mark the znode.
 233          */
 234         nzp->z_moved = 1;
 235         ozp->z_moved = (uint8_t)-1;
 236 }
 237 
 238 /*ARGSUSED*/
 239 static kmem_cbrc_t
 240 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
 241 {
 242         znode_t *ozp = buf, *nzp = newbuf;
 243         zfsvfs_t *zfsvfs;
 244         vnode_t *vp;
 245 
 246         /*
 247          * The znode is on the file system's list of known znodes if the vfs
 248          * pointer is valid. We set the low bit of the vfs pointer when freeing
 249          * the znode to invalidate it, and the memory patterns written by kmem
 250          * (baddcafe and deadbeef) set at least one of the two low bits. A newly
 251          * created znode sets the vfs pointer last of all to indicate that the
 252          * znode is known and in a valid state to be moved by this function.
 253          */
 254         zfsvfs = ozp->z_zfsvfs;
 255         if (!POINTER_IS_VALID(zfsvfs)) {
 256                 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
 257                 return (KMEM_CBRC_DONT_KNOW);
 258         }
 259 
 260         /*
 261          * Close a small window in which it's possible that the filesystem could
 262          * be unmounted and freed, and zfsvfs, though valid in the previous
 263          * statement, could point to unrelated memory by the time we try to
 264          * prevent the filesystem from being unmounted.
 265          */
 266         rw_enter(&zfsvfs_lock, RW_WRITER);
 267         if (zfsvfs != ozp->z_zfsvfs) {
 268                 rw_exit(&zfsvfs_lock);
 269                 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
 270                 return (KMEM_CBRC_DONT_KNOW);
 271         }
 272 
 273         /*
 274          * If the znode is still valid, then so is the file system. We know that
 275          * no valid file system can be freed while we hold zfsvfs_lock, so we
 276          * can safely ensure that the filesystem is not and will not be
 277          * unmounted. The next statement is equivalent to ZFS_ENTER().
 278          */
 279         rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
 280         if (zfsvfs->z_unmounted) {
 281                 ZFS_EXIT(zfsvfs);
 282                 rw_exit(&zfsvfs_lock);
 283                 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
 284                 return (KMEM_CBRC_DONT_KNOW);
 285         }
 286         rw_exit(&zfsvfs_lock);
 287 
 288         mutex_enter(&zfsvfs->z_znodes_lock);
 289         /*
 290          * Recheck the vfs pointer in case the znode was removed just before
 291          * acquiring the lock.
 292          */
 293         if (zfsvfs != ozp->z_zfsvfs) {
 294                 mutex_exit(&zfsvfs->z_znodes_lock);
 295                 ZFS_EXIT(zfsvfs);
 296                 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
 297                 return (KMEM_CBRC_DONT_KNOW);
 298         }
 299 
 300         /*
 301          * At this point we know that as long as we hold z_znodes_lock, the
 302          * znode cannot be freed and fields within the znode can be safely
 303          * accessed. Now, prevent a race with zfs_zget().
 304          */
 305         if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
 306                 mutex_exit(&zfsvfs->z_znodes_lock);
 307                 ZFS_EXIT(zfsvfs);
 308                 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
 309                 return (KMEM_CBRC_LATER);
 310         }
 311 
 312         vp = ZTOV(ozp);
 313         if (mutex_tryenter(&vp->v_lock) == 0) {
 314                 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
 315                 mutex_exit(&zfsvfs->z_znodes_lock);
 316                 ZFS_EXIT(zfsvfs);
 317                 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
 318                 return (KMEM_CBRC_LATER);
 319         }
 320 
 321         /* Only move znodes that are referenced _only_ by the DNLC. */
 322         if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
 323                 mutex_exit(&vp->v_lock);
 324                 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
 325                 mutex_exit(&zfsvfs->z_znodes_lock);
 326                 ZFS_EXIT(zfsvfs);
 327                 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
 328                 return (KMEM_CBRC_LATER);
 329         }
 330 
 331         /*
 332          * The znode is known and in a valid state to move. We're holding the
 333          * locks needed to execute the critical section.
 334          */
 335         zfs_znode_move_impl(ozp, nzp);
 336         mutex_exit(&vp->v_lock);
 337         ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
 338 
 339         list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
 340         mutex_exit(&zfsvfs->z_znodes_lock);
 341         ZFS_EXIT(zfsvfs);
 342 
 343         return (KMEM_CBRC_YES);
 344 }
 345 
 346 void
 347 zfs_znode_init(void)
 348 {
 349         /*
 350          * Initialize zcache
 351          */
 352         rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
 353         ASSERT(znode_cache == NULL);
 354         znode_cache = kmem_cache_create("zfs_znode_cache",
 355             sizeof (znode_t), 0, zfs_znode_cache_constructor,
 356             zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
 357         kmem_cache_set_move(znode_cache, zfs_znode_move);
 358 }
 359 
 360 void
 361 zfs_znode_fini(void)
 362 {
 363         /*
 364          * Cleanup vfs & vnode ops
 365          */
 366         zfs_remove_op_tables();
 367 
 368         /*
 369          * Cleanup zcache
 370          */
 371         if (znode_cache)
 372                 kmem_cache_destroy(znode_cache);
 373         znode_cache = NULL;
 374         rw_destroy(&zfsvfs_lock);
 375 }
 376 
 377 struct vnodeops *zfs_dvnodeops;
 378 struct vnodeops *zfs_fvnodeops;
 379 struct vnodeops *zfs_symvnodeops;
 380 struct vnodeops *zfs_xdvnodeops;
 381 struct vnodeops *zfs_evnodeops;
 382 struct vnodeops *zfs_sharevnodeops;
 383 
 384 void
 385 zfs_remove_op_tables()
 386 {
 387         /*
 388          * Remove vfs ops
 389          */
 390         ASSERT(zfsfstype);
 391         (void) vfs_freevfsops_by_type(zfsfstype);
 392         zfsfstype = 0;
 393 
 394         /*
 395          * Remove vnode ops
 396          */
 397         if (zfs_dvnodeops)
 398                 vn_freevnodeops(zfs_dvnodeops);
 399         if (zfs_fvnodeops)
 400                 vn_freevnodeops(zfs_fvnodeops);
 401         if (zfs_symvnodeops)
 402                 vn_freevnodeops(zfs_symvnodeops);
 403         if (zfs_xdvnodeops)
 404                 vn_freevnodeops(zfs_xdvnodeops);
 405         if (zfs_evnodeops)
 406                 vn_freevnodeops(zfs_evnodeops);
 407         if (zfs_sharevnodeops)
 408                 vn_freevnodeops(zfs_sharevnodeops);
 409 
 410         zfs_dvnodeops = NULL;
 411         zfs_fvnodeops = NULL;
 412         zfs_symvnodeops = NULL;
 413         zfs_xdvnodeops = NULL;
 414         zfs_evnodeops = NULL;
 415         zfs_sharevnodeops = NULL;
 416 }
 417 
 418 extern const fs_operation_def_t zfs_dvnodeops_template[];
 419 extern const fs_operation_def_t zfs_fvnodeops_template[];
 420 extern const fs_operation_def_t zfs_xdvnodeops_template[];
 421 extern const fs_operation_def_t zfs_symvnodeops_template[];
 422 extern const fs_operation_def_t zfs_evnodeops_template[];
 423 extern const fs_operation_def_t zfs_sharevnodeops_template[];
 424 
 425 int
 426 zfs_create_op_tables()
 427 {
 428         int error;
 429 
 430         /*
 431          * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
 432          * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
 433          * In this case we just return as the ops vectors are already set up.
 434          */
 435         if (zfs_dvnodeops)
 436                 return (0);
 437 
 438         error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
 439             &zfs_dvnodeops);
 440         if (error)
 441                 return (error);
 442 
 443         error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
 444             &zfs_fvnodeops);
 445         if (error)
 446                 return (error);
 447 
 448         error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
 449             &zfs_symvnodeops);
 450         if (error)
 451                 return (error);
 452 
 453         error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
 454             &zfs_xdvnodeops);
 455         if (error)
 456                 return (error);
 457 
 458         error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
 459             &zfs_evnodeops);
 460         if (error)
 461                 return (error);
 462 
 463         error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
 464             &zfs_sharevnodeops);
 465 
 466         return (error);
 467 }
 468 
 469 int
 470 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
 471 {
 472         zfs_acl_ids_t acl_ids;
 473         vattr_t vattr;
 474         znode_t *sharezp;
 475         vnode_t *vp;
 476         znode_t *zp;
 477         int error;
 478 
 479         vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
 480         vattr.va_type = VDIR;
 481         vattr.va_mode = S_IFDIR|0555;
 482         vattr.va_uid = crgetuid(kcred);
 483         vattr.va_gid = crgetgid(kcred);
 484 
 485         sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
 486         ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
 487         sharezp->z_moved = 0;
 488         sharezp->z_unlinked = 0;
 489         sharezp->z_atime_dirty = 0;
 490         sharezp->z_zfsvfs = zfsvfs;
 491         sharezp->z_is_sa = zfsvfs->z_use_sa;
 492 
 493         vp = ZTOV(sharezp);
 494         vn_reinit(vp);
 495         vp->v_type = VDIR;
 496 
 497         VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
 498             kcred, NULL, &acl_ids));
 499         zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
 500         ASSERT3P(zp, ==, sharezp);
 501         ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
 502         POINTER_INVALIDATE(&sharezp->z_zfsvfs);
 503         error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
 504             ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
 505         zfsvfs->z_shares_dir = sharezp->z_id;
 506 
 507         zfs_acl_ids_free(&acl_ids);
 508         ZTOV(sharezp)->v_count = 0;
 509         sa_handle_destroy(sharezp->z_sa_hdl);
 510         kmem_cache_free(znode_cache, sharezp);
 511 
 512         return (error);
 513 }
 514 
 515 /*
 516  * define a couple of values we need available
 517  * for both 64 and 32 bit environments.
 518  */
 519 #ifndef NBITSMINOR64
 520 #define NBITSMINOR64    32
 521 #endif
 522 #ifndef MAXMAJ64
 523 #define MAXMAJ64        0xffffffffUL
 524 #endif
 525 #ifndef MAXMIN64
 526 #define MAXMIN64        0xffffffffUL
 527 #endif
 528 
 529 /*
 530  * Create special expldev for ZFS private use.
 531  * Can't use standard expldev since it doesn't do
 532  * what we want.  The standard expldev() takes a
 533  * dev32_t in LP64 and expands it to a long dev_t.
 534  * We need an interface that takes a dev32_t in ILP32
 535  * and expands it to a long dev_t.
 536  */
 537 static uint64_t
 538 zfs_expldev(dev_t dev)
 539 {
 540 #ifndef _LP64
 541         major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32;
 542         return (((uint64_t)major << NBITSMINOR64) |
 543             ((minor_t)dev & MAXMIN32));
 544 #else
 545         return (dev);
 546 #endif
 547 }
 548 
 549 /*
 550  * Special cmpldev for ZFS private use.
 551  * Can't use standard cmpldev since it takes
 552  * a long dev_t and compresses it to dev32_t in
 553  * LP64.  We need to do a compaction of a long dev_t
 554  * to a dev32_t in ILP32.
 555  */
 556 dev_t
 557 zfs_cmpldev(uint64_t dev)
 558 {
 559 #ifndef _LP64
 560         minor_t minor = (minor_t)dev & MAXMIN64;
 561         major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64;
 562 
 563         if (major > MAXMAJ32 || minor > MAXMIN32)
 564                 return (NODEV32);
 565 
 566         return (((dev32_t)major << NBITSMINOR32) | minor);
 567 #else
 568         return (dev);
 569 #endif
 570 }
 571 
 572 static void
 573 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
 574     dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
 575 {
 576         ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
 577         ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
 578 
 579         mutex_enter(&zp->z_lock);
 580 
 581         ASSERT(zp->z_sa_hdl == NULL);
 582         ASSERT(zp->z_acl_cached == NULL);
 583         if (sa_hdl == NULL) {
 584                 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
 585                     SA_HDL_SHARED, &zp->z_sa_hdl));
 586         } else {
 587                 zp->z_sa_hdl = sa_hdl;
 588                 sa_set_userp(sa_hdl, zp);
 589         }
 590 
 591         zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
 592 
 593         /*
 594          * Slap on VROOT if we are the root znode
 595          */
 596         if (zp->z_id == zfsvfs->z_root)
 597                 ZTOV(zp)->v_flag |= VROOT;
 598 
 599         mutex_exit(&zp->z_lock);
 600         vn_exists(ZTOV(zp));
 601 }
 602 
 603 void
 604 zfs_znode_dmu_fini(znode_t *zp)
 605 {
 606         ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
 607             zp->z_unlinked ||
 608             RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
 609 
 610         sa_handle_destroy(zp->z_sa_hdl);
 611         zp->z_sa_hdl = NULL;
 612 }
 613 
 614 /*
 615  * Construct a new znode/vnode and intialize.
 616  *
 617  * This does not do a call to dmu_set_user() that is
 618  * up to the caller to do, in case you don't want to
 619  * return the znode
 620  */
 621 static znode_t *
 622 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
 623     dmu_object_type_t obj_type, sa_handle_t *hdl)
 624 {
 625         znode_t *zp;
 626         vnode_t *vp;
 627         uint64_t mode;
 628         uint64_t parent;
 629         sa_bulk_attr_t bulk[9];
 630         int count = 0;
 631 
 632         zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
 633 
 634         ASSERT(zp->z_dirlocks == NULL);
 635         ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
 636         zp->z_moved = 0;
 637 
 638         /*
 639          * Defer setting z_zfsvfs until the znode is ready to be a candidate for
 640          * the zfs_znode_move() callback.
 641          */
 642         zp->z_sa_hdl = NULL;
 643         zp->z_unlinked = 0;
 644         zp->z_atime_dirty = 0;
 645         zp->z_mapcnt = 0;
 646         zp->z_id = db->db_object;
 647         zp->z_blksz = blksz;
 648         zp->z_seq = 0x7A4653;
 649         zp->z_sync_cnt = 0;
 650 
 651         vp = ZTOV(zp);
 652         vn_reinit(vp);
 653 
 654         zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
 655 
 656         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
 657         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
 658         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
 659             &zp->z_size, 8);
 660         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
 661             &zp->z_links, 8);
 662         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
 663             &zp->z_pflags, 8);
 664         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
 665         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
 666             &zp->z_atime, 16);
 667         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
 668             &zp->z_uid, 8);
 669         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
 670             &zp->z_gid, 8);
 671 
 672         if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
 673                 if (hdl == NULL)
 674                         sa_handle_destroy(zp->z_sa_hdl);
 675                 kmem_cache_free(znode_cache, zp);
 676                 return (NULL);
 677         }
 678 
 679         zp->z_mode = mode;
 680         vp->v_vfsp = zfsvfs->z_parent->z_vfs;
 681 
 682         vp->v_type = IFTOVT((mode_t)mode);
 683 
 684         switch (vp->v_type) {
 685         case VDIR:
 686                 if (zp->z_pflags & ZFS_XATTR) {
 687                         vn_setops(vp, zfs_xdvnodeops);
 688                         vp->v_flag |= V_XATTRDIR;
 689                 } else {
 690                         vn_setops(vp, zfs_dvnodeops);
 691                 }
 692                 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
 693                 break;
 694         case VBLK:
 695         case VCHR:
 696                 {
 697                         uint64_t rdev;
 698                         VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
 699                             &rdev, sizeof (rdev)) == 0);
 700 
 701                         vp->v_rdev = zfs_cmpldev(rdev);
 702                 }
 703                 /*FALLTHROUGH*/
 704         case VFIFO:
 705         case VSOCK:
 706         case VDOOR:
 707                 vn_setops(vp, zfs_fvnodeops);
 708                 break;
 709         case VREG:
 710                 vp->v_flag |= VMODSORT;
 711                 if (parent == zfsvfs->z_shares_dir) {
 712                         ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
 713                         vn_setops(vp, zfs_sharevnodeops);
 714                 } else {
 715                         vn_setops(vp, zfs_fvnodeops);
 716                 }
 717                 break;
 718         case VLNK:
 719                 vn_setops(vp, zfs_symvnodeops);
 720                 break;
 721         default:
 722                 vn_setops(vp, zfs_evnodeops);
 723                 break;
 724         }
 725 
 726         mutex_enter(&zfsvfs->z_znodes_lock);
 727         list_insert_tail(&zfsvfs->z_all_znodes, zp);
 728         membar_producer();
 729         /*
 730          * Everything else must be valid before assigning z_zfsvfs makes the
 731          * znode eligible for zfs_znode_move().
 732          */
 733         zp->z_zfsvfs = zfsvfs;
 734         mutex_exit(&zfsvfs->z_znodes_lock);
 735 
 736         VFS_HOLD(zfsvfs->z_vfs);
 737         return (zp);
 738 }
 739 
 740 static uint64_t empty_xattr;
 741 static uint64_t pad[4];
 742 static zfs_acl_phys_t acl_phys;
 743 /*
 744  * Create a new DMU object to hold a zfs znode.
 745  *
 746  *      IN:     dzp     - parent directory for new znode
 747  *              vap     - file attributes for new znode
 748  *              tx      - dmu transaction id for zap operations
 749  *              cr      - credentials of caller
 750  *              flag    - flags:
 751  *                        IS_ROOT_NODE  - new object will be root
 752  *                        IS_XATTR      - new object is an attribute
 753  *              bonuslen - length of bonus buffer
 754  *              setaclp  - File/Dir initial ACL
 755  *              fuidp    - Tracks fuid allocation.
 756  *
 757  *      OUT:    zpp     - allocated znode
 758  *
 759  */
 760 void
 761 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
 762     uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
 763 {
 764         uint64_t        crtime[2], atime[2], mtime[2], ctime[2];
 765         uint64_t        mode, size, links, parent, pflags;
 766         uint64_t        dzp_pflags = 0;
 767         uint64_t        rdev = 0;
 768         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
 769         dmu_buf_t       *db;
 770         timestruc_t     now;
 771         uint64_t        gen, obj;
 772         int             err;
 773         int             bonuslen;
 774         sa_handle_t     *sa_hdl;
 775         dmu_object_type_t obj_type;
 776         sa_bulk_attr_t  sa_attrs[ZPL_END];
 777         int             cnt = 0;
 778         zfs_acl_locator_cb_t locate = { 0 };
 779 
 780         ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
 781 
 782         if (zfsvfs->z_replay) {
 783                 obj = vap->va_nodeid;
 784                 now = vap->va_ctime;         /* see zfs_replay_create() */
 785                 gen = vap->va_nblocks;               /* ditto */
 786         } else {
 787                 obj = 0;
 788                 gethrestime(&now);
 789                 gen = dmu_tx_get_txg(tx);
 790         }
 791 
 792         obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
 793         bonuslen = (obj_type == DMU_OT_SA) ?
 794             DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
 795 
 796         /*
 797          * Create a new DMU object.
 798          */
 799         /*
 800          * There's currently no mechanism for pre-reading the blocks that will
 801          * be needed to allocate a new object, so we accept the small chance
 802          * that there will be an i/o error and we will fail one of the
 803          * assertions below.
 804          */
 805         if (vap->va_type == VDIR) {
 806                 if (zfsvfs->z_replay) {
 807                         err = zap_create_claim_norm(zfsvfs->z_os, obj,
 808                             zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
 809                             obj_type, bonuslen, tx);
 810                         ASSERT0(err);
 811                 } else {
 812                         obj = zap_create_norm(zfsvfs->z_os,
 813                             zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
 814                             obj_type, bonuslen, tx);
 815                 }
 816         } else {
 817                 if (zfsvfs->z_replay) {
 818                         err = dmu_object_claim(zfsvfs->z_os, obj,
 819                             DMU_OT_PLAIN_FILE_CONTENTS, 0,
 820                             obj_type, bonuslen, tx);
 821                         ASSERT0(err);
 822                 } else {
 823                         obj = dmu_object_alloc(zfsvfs->z_os,
 824                             DMU_OT_PLAIN_FILE_CONTENTS, 0,
 825                             obj_type, bonuslen, tx);
 826                 }
 827         }
 828 
 829         ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
 830         VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
 831 
 832         /*
 833          * If this is the root, fix up the half-initialized parent pointer
 834          * to reference the just-allocated physical data area.
 835          */
 836         if (flag & IS_ROOT_NODE) {
 837                 dzp->z_id = obj;
 838         } else {
 839                 dzp_pflags = dzp->z_pflags;
 840         }
 841 
 842         /*
 843          * If parent is an xattr, so am I.
 844          */
 845         if (dzp_pflags & ZFS_XATTR) {
 846                 flag |= IS_XATTR;
 847         }
 848 
 849         if (zfsvfs->z_use_fuids)
 850                 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
 851         else
 852                 pflags = 0;
 853 
 854         if (vap->va_type == VDIR) {
 855                 size = 2;               /* contents ("." and "..") */
 856                 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
 857         } else {
 858                 size = links = 0;
 859         }
 860 
 861         if (vap->va_type == VBLK || vap->va_type == VCHR) {
 862                 rdev = zfs_expldev(vap->va_rdev);
 863         }
 864 
 865         parent = dzp->z_id;
 866         mode = acl_ids->z_mode;
 867         if (flag & IS_XATTR)
 868                 pflags |= ZFS_XATTR;
 869 
 870         /*
 871          * No execs denied will be deterimed when zfs_mode_compute() is called.
 872          */
 873         pflags |= acl_ids->z_aclp->z_hints &
 874             (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
 875             ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
 876 
 877         ZFS_TIME_ENCODE(&now, crtime);
 878         ZFS_TIME_ENCODE(&now, ctime);
 879 
 880         if (vap->va_mask & AT_ATIME) {
 881                 ZFS_TIME_ENCODE(&vap->va_atime, atime);
 882         } else {
 883                 ZFS_TIME_ENCODE(&now, atime);
 884         }
 885 
 886         if (vap->va_mask & AT_MTIME) {
 887                 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
 888         } else {
 889                 ZFS_TIME_ENCODE(&now, mtime);
 890         }
 891 
 892         /* Now add in all of the "SA" attributes */
 893         VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
 894             &sa_hdl));
 895 
 896         /*
 897          * Setup the array of attributes to be replaced/set on the new file
 898          *
 899          * order for  DMU_OT_ZNODE is critical since it needs to be constructed
 900          * in the old znode_phys_t format.  Don't change this ordering
 901          */
 902 
 903         if (obj_type == DMU_OT_ZNODE) {
 904                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
 905                     NULL, &atime, 16);
 906                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
 907                     NULL, &mtime, 16);
 908                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
 909                     NULL, &ctime, 16);
 910                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
 911                     NULL, &crtime, 16);
 912                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
 913                     NULL, &gen, 8);
 914                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
 915                     NULL, &mode, 8);
 916                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
 917                     NULL, &size, 8);
 918                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
 919                     NULL, &parent, 8);
 920         } else {
 921                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
 922                     NULL, &mode, 8);
 923                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
 924                     NULL, &size, 8);
 925                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
 926                     NULL, &gen, 8);
 927                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
 928                     &acl_ids->z_fuid, 8);
 929                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
 930                     &acl_ids->z_fgid, 8);
 931                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
 932                     NULL, &parent, 8);
 933                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
 934                     NULL, &pflags, 8);
 935                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
 936                     NULL, &atime, 16);
 937                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
 938                     NULL, &mtime, 16);
 939                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
 940                     NULL, &ctime, 16);
 941                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
 942                     NULL, &crtime, 16);
 943         }
 944 
 945         SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
 946 
 947         if (obj_type == DMU_OT_ZNODE) {
 948                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
 949                     &empty_xattr, 8);
 950         }
 951         if (obj_type == DMU_OT_ZNODE ||
 952             (vap->va_type == VBLK || vap->va_type == VCHR)) {
 953                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
 954                     NULL, &rdev, 8);
 955 
 956         }
 957         if (obj_type == DMU_OT_ZNODE) {
 958                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
 959                     NULL, &pflags, 8);
 960                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
 961                     &acl_ids->z_fuid, 8);
 962                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
 963                     &acl_ids->z_fgid, 8);
 964                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
 965                     sizeof (uint64_t) * 4);
 966                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
 967                     &acl_phys, sizeof (zfs_acl_phys_t));
 968         } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
 969                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
 970                     &acl_ids->z_aclp->z_acl_count, 8);
 971                 locate.cb_aclp = acl_ids->z_aclp;
 972                 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
 973                     zfs_acl_data_locator, &locate,
 974                     acl_ids->z_aclp->z_acl_bytes);
 975                 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
 976                     acl_ids->z_fuid, acl_ids->z_fgid);
 977         }
 978 
 979         VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
 980 
 981         if (!(flag & IS_ROOT_NODE)) {
 982                 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
 983                 ASSERT(*zpp != NULL);
 984         } else {
 985                 /*
 986                  * If we are creating the root node, the "parent" we
 987                  * passed in is the znode for the root.
 988                  */
 989                 *zpp = dzp;
 990 
 991                 (*zpp)->z_sa_hdl = sa_hdl;
 992         }
 993 
 994         (*zpp)->z_pflags = pflags;
 995         (*zpp)->z_mode = mode;
 996 
 997         if (vap->va_mask & AT_XVATTR)
 998                 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
 999 
1000         if (obj_type == DMU_OT_ZNODE ||
1001             acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1002                 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx);
1003                 ASSERT0(err);
1004         }
1005         ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1006 }
1007 
1008 /*
1009  * zfs_xvattr_set only updates the in-core attributes
1010  * it is assumed the caller will be doing an sa_bulk_update
1011  * to push the changes out
1012  */
1013 void
1014 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1015 {
1016         xoptattr_t *xoap;
1017 
1018         xoap = xva_getxoptattr(xvap);
1019         ASSERT(xoap);
1020 
1021         if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1022                 uint64_t times[2];
1023                 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1024                 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1025                     &times, sizeof (times), tx);
1026                 XVA_SET_RTN(xvap, XAT_CREATETIME);
1027         }
1028         if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1029                 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1030                     zp->z_pflags, tx);
1031                 XVA_SET_RTN(xvap, XAT_READONLY);
1032         }
1033         if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1034                 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1035                     zp->z_pflags, tx);
1036                 XVA_SET_RTN(xvap, XAT_HIDDEN);
1037         }
1038         if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1039                 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1040                     zp->z_pflags, tx);
1041                 XVA_SET_RTN(xvap, XAT_SYSTEM);
1042         }
1043         if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1044                 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1045                     zp->z_pflags, tx);
1046                 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1047         }
1048         if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1049                 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1050                     zp->z_pflags, tx);
1051                 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1052         }
1053         if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1054                 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1055                     zp->z_pflags, tx);
1056                 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1057         }
1058         if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1059                 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1060                     zp->z_pflags, tx);
1061                 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1062         }
1063         if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1064                 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1065                     zp->z_pflags, tx);
1066                 XVA_SET_RTN(xvap, XAT_NODUMP);
1067         }
1068         if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1069                 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1070                     zp->z_pflags, tx);
1071                 XVA_SET_RTN(xvap, XAT_OPAQUE);
1072         }
1073         if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1074                 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1075                     xoap->xoa_av_quarantined, zp->z_pflags, tx);
1076                 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1077         }
1078         if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1079                 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1080                     zp->z_pflags, tx);
1081                 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1082         }
1083         if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1084                 zfs_sa_set_scanstamp(zp, xvap, tx);
1085                 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1086         }
1087         if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1088                 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1089                     zp->z_pflags, tx);
1090                 XVA_SET_RTN(xvap, XAT_REPARSE);
1091         }
1092         if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1093                 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1094                     zp->z_pflags, tx);
1095                 XVA_SET_RTN(xvap, XAT_OFFLINE);
1096         }
1097         if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1098                 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1099                     zp->z_pflags, tx);
1100                 XVA_SET_RTN(xvap, XAT_SPARSE);
1101         }
1102 }
1103 
1104 int
1105 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1106 {
1107         dmu_object_info_t doi;
1108         dmu_buf_t       *db;
1109         znode_t         *zp;
1110         int err;
1111         sa_handle_t     *hdl;
1112 
1113         *zpp = NULL;
1114 
1115         ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1116 
1117         err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1118         if (err) {
1119                 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1120                 return (err);
1121         }
1122 
1123         dmu_object_info_from_db(db, &doi);
1124         if (doi.doi_bonus_type != DMU_OT_SA &&
1125             (doi.doi_bonus_type != DMU_OT_ZNODE ||
1126             (doi.doi_bonus_type == DMU_OT_ZNODE &&
1127             doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1128                 sa_buf_rele(db, NULL);
1129                 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1130                 return (SET_ERROR(EINVAL));
1131         }
1132 
1133         hdl = (sa_handle_t *)dmu_buf_get_user(db);
1134         if (hdl != NULL) {
1135                 zp  = sa_get_userdata(hdl);
1136 
1137 
1138                 /*
1139                  * Since "SA" does immediate eviction we
1140                  * should never find a sa handle that doesn't
1141                  * know about the znode.
1142                  */
1143 
1144                 ASSERT3P(zp, !=, NULL);
1145 
1146                 mutex_enter(&zp->z_lock);
1147                 ASSERT3U(zp->z_id, ==, obj_num);
1148                 if (zp->z_unlinked) {
1149                         err = SET_ERROR(ENOENT);
1150                 } else {
1151                         VN_HOLD(ZTOV(zp));
1152                         *zpp = zp;
1153                         err = 0;
1154                 }
1155                 sa_buf_rele(db, NULL);
1156                 mutex_exit(&zp->z_lock);
1157                 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1158                 return (err);
1159         }
1160 
1161         /*
1162          * Not found create new znode/vnode
1163          * but only if file exists.
1164          *
1165          * There is a small window where zfs_vget() could
1166          * find this object while a file create is still in
1167          * progress.  This is checked for in zfs_znode_alloc()
1168          *
1169          * if zfs_znode_alloc() fails it will drop the hold on the
1170          * bonus buffer.
1171          */
1172         zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1173             doi.doi_bonus_type, NULL);
1174         if (zp == NULL) {
1175                 err = SET_ERROR(ENOENT);
1176         } else {
1177                 *zpp = zp;
1178         }
1179         ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1180         return (err);
1181 }
1182 
1183 int
1184 zfs_rezget(znode_t *zp)
1185 {
1186         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1187         dmu_object_info_t doi;
1188         dmu_buf_t *db;
1189         uint64_t obj_num = zp->z_id;
1190         uint64_t mode;
1191         sa_bulk_attr_t bulk[8];
1192         int err;
1193         int count = 0;
1194         uint64_t gen;
1195 
1196         ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1197 
1198         mutex_enter(&zp->z_acl_lock);
1199         if (zp->z_acl_cached) {
1200                 zfs_acl_free(zp->z_acl_cached);
1201                 zp->z_acl_cached = NULL;
1202         }
1203 
1204         mutex_exit(&zp->z_acl_lock);
1205         ASSERT(zp->z_sa_hdl == NULL);
1206         err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1207         if (err) {
1208                 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1209                 return (err);
1210         }
1211 
1212         dmu_object_info_from_db(db, &doi);
1213         if (doi.doi_bonus_type != DMU_OT_SA &&
1214             (doi.doi_bonus_type != DMU_OT_ZNODE ||
1215             (doi.doi_bonus_type == DMU_OT_ZNODE &&
1216             doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1217                 sa_buf_rele(db, NULL);
1218                 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1219                 return (SET_ERROR(EINVAL));
1220         }
1221 
1222         zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1223 
1224         /* reload cached values */
1225         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1226             &gen, sizeof (gen));
1227         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1228             &zp->z_size, sizeof (zp->z_size));
1229         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1230             &zp->z_links, sizeof (zp->z_links));
1231         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1232             &zp->z_pflags, sizeof (zp->z_pflags));
1233         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1234             &zp->z_atime, sizeof (zp->z_atime));
1235         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1236             &zp->z_uid, sizeof (zp->z_uid));
1237         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1238             &zp->z_gid, sizeof (zp->z_gid));
1239         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1240             &mode, sizeof (mode));
1241 
1242         if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1243                 zfs_znode_dmu_fini(zp);
1244                 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1245                 return (SET_ERROR(EIO));
1246         }
1247 
1248         zp->z_mode = mode;
1249 
1250         if (gen != zp->z_gen) {
1251                 zfs_znode_dmu_fini(zp);
1252                 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1253                 return (SET_ERROR(EIO));
1254         }
1255 
1256         zp->z_unlinked = (zp->z_links == 0);
1257         zp->z_blksz = doi.doi_data_block_size;
1258 
1259         ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1260 
1261         return (0);
1262 }
1263 
1264 void
1265 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1266 {
1267         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1268         objset_t *os = zfsvfs->z_os;
1269         uint64_t obj = zp->z_id;
1270         uint64_t acl_obj = zfs_external_acl(zp);
1271 
1272         ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1273         if (acl_obj) {
1274                 VERIFY(!zp->z_is_sa);
1275                 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1276         }
1277         VERIFY(0 == dmu_object_free(os, obj, tx));
1278         zfs_znode_dmu_fini(zp);
1279         ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1280         zfs_znode_free(zp);
1281 }
1282 
1283 void
1284 zfs_zinactive(znode_t *zp)
1285 {
1286         vnode_t *vp = ZTOV(zp);
1287         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1288         uint64_t z_id = zp->z_id;
1289 
1290         ASSERT(zp->z_sa_hdl);
1291 
1292         /*
1293          * Don't allow a zfs_zget() while were trying to release this znode
1294          */
1295         ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1296 
1297         mutex_enter(&zp->z_lock);
1298         mutex_enter(&vp->v_lock);
1299         vp->v_count--;
1300         if (vp->v_count > 0 || vn_has_cached_data(vp)) {
1301                 /*
1302                  * If the hold count is greater than zero, somebody has
1303                  * obtained a new reference on this znode while we were
1304                  * processing it here, so we are done.  If we still have
1305                  * mapped pages then we are also done, since we don't
1306                  * want to inactivate the znode until the pages get pushed.
1307                  *
1308                  * XXX - if vn_has_cached_data(vp) is true, but count == 0,
1309                  * this seems like it would leave the znode hanging with
1310                  * no chance to go inactive...
1311                  */
1312                 mutex_exit(&vp->v_lock);
1313                 mutex_exit(&zp->z_lock);
1314                 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1315                 return;
1316         }
1317         mutex_exit(&vp->v_lock);
1318 
1319         /*
1320          * If this was the last reference to a file with no links,
1321          * remove the file from the file system.
1322          */
1323         if (zp->z_unlinked) {
1324                 mutex_exit(&zp->z_lock);
1325                 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1326                 zfs_rmnode(zp);
1327                 return;
1328         }
1329 
1330         mutex_exit(&zp->z_lock);
1331         zfs_znode_dmu_fini(zp);
1332         ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1333         zfs_znode_free(zp);
1334 }
1335 
1336 void
1337 zfs_znode_free(znode_t *zp)
1338 {
1339         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1340 
1341         vn_invalid(ZTOV(zp));
1342 
1343         ASSERT(ZTOV(zp)->v_count == 0);
1344 
1345         mutex_enter(&zfsvfs->z_znodes_lock);
1346         POINTER_INVALIDATE(&zp->z_zfsvfs);
1347         list_remove(&zfsvfs->z_all_znodes, zp);
1348         mutex_exit(&zfsvfs->z_znodes_lock);
1349 
1350         if (zp->z_acl_cached) {
1351                 zfs_acl_free(zp->z_acl_cached);
1352                 zp->z_acl_cached = NULL;
1353         }
1354 
1355         kmem_cache_free(znode_cache, zp);
1356 
1357         VFS_RELE(zfsvfs->z_vfs);
1358 }
1359 
1360 void
1361 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1362     uint64_t ctime[2], boolean_t have_tx)
1363 {
1364         timestruc_t     now;
1365 
1366         gethrestime(&now);
1367 
1368         if (have_tx) {  /* will sa_bulk_update happen really soon? */
1369                 zp->z_atime_dirty = 0;
1370                 zp->z_seq++;
1371         } else {
1372                 zp->z_atime_dirty = 1;
1373         }
1374 
1375         if (flag & AT_ATIME) {
1376                 ZFS_TIME_ENCODE(&now, zp->z_atime);
1377         }
1378 
1379         if (flag & AT_MTIME) {
1380                 ZFS_TIME_ENCODE(&now, mtime);
1381                 if (zp->z_zfsvfs->z_use_fuids) {
1382                         zp->z_pflags |= (ZFS_ARCHIVE |
1383                             ZFS_AV_MODIFIED);
1384                 }
1385         }
1386 
1387         if (flag & AT_CTIME) {
1388                 ZFS_TIME_ENCODE(&now, ctime);
1389                 if (zp->z_zfsvfs->z_use_fuids)
1390                         zp->z_pflags |= ZFS_ARCHIVE;
1391         }
1392 }
1393 
1394 /*
1395  * Grow the block size for a file.
1396  *
1397  *      IN:     zp      - znode of file to free data in.
1398  *              size    - requested block size
1399  *              tx      - open transaction.
1400  *
1401  * NOTE: this function assumes that the znode is write locked.
1402  */
1403 void
1404 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1405 {
1406         int             error;
1407         u_longlong_t    dummy;
1408 
1409         if (size <= zp->z_blksz)
1410                 return;
1411         /*
1412          * If the file size is already greater than the current blocksize,
1413          * we will not grow.  If there is more than one block in a file,
1414          * the blocksize cannot change.
1415          */
1416         if (zp->z_blksz && zp->z_size > zp->z_blksz)
1417                 return;
1418 
1419         error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1420             size, 0, tx);
1421 
1422         if (error == ENOTSUP)
1423                 return;
1424         ASSERT0(error);
1425 
1426         /* What blocksize did we actually get? */
1427         dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1428 }
1429 
1430 /*
1431  * This is a dummy interface used when pvn_vplist_dirty() should *not*
1432  * be calling back into the fs for a putpage().  E.g.: when truncating
1433  * a file, the pages being "thrown away* don't need to be written out.
1434  */
1435 /* ARGSUSED */
1436 static int
1437 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1438     int flags, cred_t *cr)
1439 {
1440         ASSERT(0);
1441         return (0);
1442 }
1443 
1444 /*
1445  * Increase the file length
1446  *
1447  *      IN:     zp      - znode of file to free data in.
1448  *              end     - new end-of-file
1449  *
1450  *      RETURN: 0 if success
1451  *              error code if failure
1452  */
1453 static int
1454 zfs_extend(znode_t *zp, uint64_t end)
1455 {
1456         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1457         dmu_tx_t *tx;
1458         rl_t *rl;
1459         uint64_t newblksz;
1460         int error;
1461 
1462         /*
1463          * We will change zp_size, lock the whole file.
1464          */
1465         rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1466 
1467         /*
1468          * Nothing to do if file already at desired length.
1469          */
1470         if (end <= zp->z_size) {
1471                 zfs_range_unlock(rl);
1472                 return (0);
1473         }
1474 top:
1475         tx = dmu_tx_create(zfsvfs->z_os);
1476         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1477         zfs_sa_upgrade_txholds(tx, zp);
1478         if (end > zp->z_blksz &&
1479             (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1480                 /*
1481                  * We are growing the file past the current block size.
1482                  */
1483                 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1484                         ASSERT(!ISP2(zp->z_blksz));
1485                         newblksz = MIN(end, SPA_MAXBLOCKSIZE);
1486                 } else {
1487                         newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1488                 }
1489                 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1490         } else {
1491                 newblksz = 0;
1492         }
1493 
1494         error = dmu_tx_assign(tx, TXG_NOWAIT);
1495         if (error) {
1496                 if (error == ERESTART) {
1497                         dmu_tx_wait(tx);
1498                         dmu_tx_abort(tx);
1499                         goto top;
1500                 }
1501                 dmu_tx_abort(tx);
1502                 zfs_range_unlock(rl);
1503                 return (error);
1504         }
1505 
1506         if (newblksz)
1507                 zfs_grow_blocksize(zp, newblksz, tx);
1508 
1509         zp->z_size = end;
1510 
1511         VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1512             &zp->z_size, sizeof (zp->z_size), tx));
1513 
1514         zfs_range_unlock(rl);
1515 
1516         dmu_tx_commit(tx);
1517 
1518         return (0);
1519 }
1520 
1521 /*
1522  * Free space in a file.
1523  *
1524  *      IN:     zp      - znode of file to free data in.
1525  *              off     - start of section to free.
1526  *              len     - length of section to free.
1527  *
1528  *      RETURN: 0 if success
1529  *              error code if failure
1530  */
1531 static int
1532 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1533 {
1534         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1535         rl_t *rl;
1536         int error;
1537 
1538         /*
1539          * Lock the range being freed.
1540          */
1541         rl = zfs_range_lock(zp, off, len, RL_WRITER);
1542 
1543         /*
1544          * Nothing to do if file already at desired length.
1545          */
1546         if (off >= zp->z_size) {
1547                 zfs_range_unlock(rl);
1548                 return (0);
1549         }
1550 
1551         if (off + len > zp->z_size)
1552                 len = zp->z_size - off;
1553 
1554         error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1555 
1556         zfs_range_unlock(rl);
1557 
1558         return (error);
1559 }
1560 
1561 /*
1562  * Truncate a file
1563  *
1564  *      IN:     zp      - znode of file to free data in.
1565  *              end     - new end-of-file.
1566  *
1567  *      RETURN: 0 if success
1568  *              error code if failure
1569  */
1570 static int
1571 zfs_trunc(znode_t *zp, uint64_t end)
1572 {
1573         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1574         vnode_t *vp = ZTOV(zp);
1575         dmu_tx_t *tx;
1576         rl_t *rl;
1577         int error;
1578         sa_bulk_attr_t bulk[2];
1579         int count = 0;
1580 
1581         /*
1582          * We will change zp_size, lock the whole file.
1583          */
1584         rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1585 
1586         /*
1587          * Nothing to do if file already at desired length.
1588          */
1589         if (end >= zp->z_size) {
1590                 zfs_range_unlock(rl);
1591                 return (0);
1592         }
1593 
1594         error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,  -1);
1595         if (error) {
1596                 zfs_range_unlock(rl);
1597                 return (error);
1598         }
1599 top:
1600         tx = dmu_tx_create(zfsvfs->z_os);
1601         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1602         zfs_sa_upgrade_txholds(tx, zp);
1603         error = dmu_tx_assign(tx, TXG_NOWAIT);
1604         if (error) {
1605                 if (error == ERESTART) {
1606                         dmu_tx_wait(tx);
1607                         dmu_tx_abort(tx);
1608                         goto top;
1609                 }
1610                 dmu_tx_abort(tx);
1611                 zfs_range_unlock(rl);
1612                 return (error);
1613         }
1614 
1615         zp->z_size = end;
1616         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1617             NULL, &zp->z_size, sizeof (zp->z_size));
1618 
1619         if (end == 0) {
1620                 zp->z_pflags &= ~ZFS_SPARSE;
1621                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1622                     NULL, &zp->z_pflags, 8);
1623         }
1624         VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1625 
1626         dmu_tx_commit(tx);
1627 
1628         /*
1629          * Clear any mapped pages in the truncated region.  This has to
1630          * happen outside of the transaction to avoid the possibility of
1631          * a deadlock with someone trying to push a page that we are
1632          * about to invalidate.
1633          */
1634         if (vn_has_cached_data(vp)) {
1635                 page_t *pp;
1636                 uint64_t start = end & PAGEMASK;
1637                 int poff = end & PAGEOFFSET;
1638 
1639                 if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) {
1640                         /*
1641                          * We need to zero a partial page.
1642                          */
1643                         pagezero(pp, poff, PAGESIZE - poff);
1644                         start += PAGESIZE;
1645                         page_unlock(pp);
1646                 }
1647                 error = pvn_vplist_dirty(vp, start, zfs_no_putpage,
1648                     B_INVAL | B_TRUNC, NULL);
1649                 ASSERT(error == 0);
1650         }
1651 
1652         zfs_range_unlock(rl);
1653 
1654         return (0);
1655 }
1656 
1657 /*
1658  * Free space in a file
1659  *
1660  *      IN:     zp      - znode of file to free data in.
1661  *              off     - start of range
1662  *              len     - end of range (0 => EOF)
1663  *              flag    - current file open mode flags.
1664  *              log     - TRUE if this action should be logged
1665  *
1666  *      RETURN: 0 if success
1667  *              error code if failure
1668  */
1669 int
1670 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1671 {
1672         vnode_t *vp = ZTOV(zp);
1673         dmu_tx_t *tx;
1674         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1675         zilog_t *zilog = zfsvfs->z_log;
1676         uint64_t mode;
1677         uint64_t mtime[2], ctime[2];
1678         sa_bulk_attr_t bulk[3];
1679         int count = 0;
1680         int error;
1681 
1682         if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1683             sizeof (mode))) != 0)
1684                 return (error);
1685 
1686         if (off > zp->z_size) {
1687                 error =  zfs_extend(zp, off+len);
1688                 if (error == 0 && log)
1689                         goto log;
1690                 else
1691                         return (error);
1692         }
1693 
1694         /*
1695          * Check for any locks in the region to be freed.
1696          */
1697 
1698         if (MANDLOCK(vp, (mode_t)mode)) {
1699                 uint64_t length = (len ? len : zp->z_size - off);
1700                 if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1701                         return (error);
1702         }
1703 
1704         if (len == 0) {
1705                 error = zfs_trunc(zp, off);
1706         } else {
1707                 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1708                     off + len > zp->z_size)
1709                         error = zfs_extend(zp, off+len);
1710         }
1711         if (error || !log)
1712                 return (error);
1713 log:
1714         tx = dmu_tx_create(zfsvfs->z_os);
1715         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1716         zfs_sa_upgrade_txholds(tx, zp);
1717         error = dmu_tx_assign(tx, TXG_NOWAIT);
1718         if (error) {
1719                 if (error == ERESTART) {
1720                         dmu_tx_wait(tx);
1721                         dmu_tx_abort(tx);
1722                         goto log;
1723                 }
1724                 dmu_tx_abort(tx);
1725                 return (error);
1726         }
1727 
1728         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1729         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1730         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1731             NULL, &zp->z_pflags, 8);
1732         zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1733         error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1734         ASSERT(error == 0);
1735 
1736         zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1737 
1738         dmu_tx_commit(tx);
1739         return (0);
1740 }
1741 
1742 void
1743 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1744 {
1745         zfsvfs_t        zfsvfs;
1746         uint64_t        moid, obj, sa_obj, version;
1747         uint64_t        sense = ZFS_CASE_SENSITIVE;
1748         uint64_t        norm = 0;
1749         nvpair_t        *elem;
1750         int             error;
1751         int             i;
1752         znode_t         *rootzp = NULL;
1753         vnode_t         *vp;
1754         vattr_t         vattr;
1755         znode_t         *zp;
1756         zfs_acl_ids_t   acl_ids;
1757 
1758         /*
1759          * First attempt to create master node.
1760          */
1761         /*
1762          * In an empty objset, there are no blocks to read and thus
1763          * there can be no i/o errors (which we assert below).
1764          */
1765         moid = MASTER_NODE_OBJ;
1766         error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1767             DMU_OT_NONE, 0, tx);
1768         ASSERT(error == 0);
1769 
1770         /*
1771          * Set starting attributes.
1772          */
1773         version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1774         elem = NULL;
1775         while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1776                 /* For the moment we expect all zpl props to be uint64_ts */
1777                 uint64_t val;
1778                 char *name;
1779 
1780                 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1781                 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1782                 name = nvpair_name(elem);
1783                 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1784                         if (val < version)
1785                                 version = val;
1786                 } else {
1787                         error = zap_update(os, moid, name, 8, 1, &val, tx);
1788                 }
1789                 ASSERT(error == 0);
1790                 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1791                         norm = val;
1792                 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1793                         sense = val;
1794         }
1795         ASSERT(version != 0);
1796         error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1797 
1798         /*
1799          * Create zap object used for SA attribute registration
1800          */
1801 
1802         if (version >= ZPL_VERSION_SA) {
1803                 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1804                     DMU_OT_NONE, 0, tx);
1805                 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1806                 ASSERT(error == 0);
1807         } else {
1808                 sa_obj = 0;
1809         }
1810         /*
1811          * Create a delete queue.
1812          */
1813         obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1814 
1815         error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1816         ASSERT(error == 0);
1817 
1818         /*
1819          * Create root znode.  Create minimal znode/vnode/zfsvfs
1820          * to allow zfs_mknode to work.
1821          */
1822         vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1823         vattr.va_type = VDIR;
1824         vattr.va_mode = S_IFDIR|0755;
1825         vattr.va_uid = crgetuid(cr);
1826         vattr.va_gid = crgetgid(cr);
1827 
1828         rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1829         ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1830         rootzp->z_moved = 0;
1831         rootzp->z_unlinked = 0;
1832         rootzp->z_atime_dirty = 0;
1833         rootzp->z_is_sa = USE_SA(version, os);
1834 
1835         vp = ZTOV(rootzp);
1836         vn_reinit(vp);
1837         vp->v_type = VDIR;
1838 
1839         bzero(&zfsvfs, sizeof (zfsvfs_t));
1840 
1841         zfsvfs.z_os = os;
1842         zfsvfs.z_parent = &zfsvfs;
1843         zfsvfs.z_version = version;
1844         zfsvfs.z_use_fuids = USE_FUIDS(version, os);
1845         zfsvfs.z_use_sa = USE_SA(version, os);
1846         zfsvfs.z_norm = norm;
1847 
1848         error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1849             &zfsvfs.z_attr_table);
1850 
1851         ASSERT(error == 0);
1852 
1853         /*
1854          * Fold case on file systems that are always or sometimes case
1855          * insensitive.
1856          */
1857         if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1858                 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER;
1859 
1860         mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1861         list_create(&zfsvfs.z_all_znodes, sizeof (znode_t),
1862             offsetof(znode_t, z_link_node));
1863 
1864         for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1865                 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1866 
1867         rootzp->z_zfsvfs = &zfsvfs;
1868         VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1869             cr, NULL, &acl_ids));
1870         zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1871         ASSERT3P(zp, ==, rootzp);
1872         ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */
1873         error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1874         ASSERT(error == 0);
1875         zfs_acl_ids_free(&acl_ids);
1876         POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1877 
1878         ZTOV(rootzp)->v_count = 0;
1879         sa_handle_destroy(rootzp->z_sa_hdl);
1880         kmem_cache_free(znode_cache, rootzp);
1881 
1882         /*
1883          * Create shares directory
1884          */
1885 
1886         error = zfs_create_share_dir(&zfsvfs, tx);
1887 
1888         ASSERT(error == 0);
1889 
1890         for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1891                 mutex_destroy(&zfsvfs.z_hold_mtx[i]);
1892 }
1893 
1894 #endif /* _KERNEL */
1895 
1896 static int
1897 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1898 {
1899         uint64_t sa_obj = 0;
1900         int error;
1901 
1902         error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1903         if (error != 0 && error != ENOENT)
1904                 return (error);
1905 
1906         error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1907         return (error);
1908 }
1909 
1910 static int
1911 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1912     dmu_buf_t **db, void *tag)
1913 {
1914         dmu_object_info_t doi;
1915         int error;
1916 
1917         if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1918                 return (error);
1919 
1920         dmu_object_info_from_db(*db, &doi);
1921         if ((doi.doi_bonus_type != DMU_OT_SA &&
1922             doi.doi_bonus_type != DMU_OT_ZNODE) ||
1923             doi.doi_bonus_type == DMU_OT_ZNODE &&
1924             doi.doi_bonus_size < sizeof (znode_phys_t)) {
1925                 sa_buf_rele(*db, tag);
1926                 return (SET_ERROR(ENOTSUP));
1927         }
1928 
1929         error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1930         if (error != 0) {
1931                 sa_buf_rele(*db, tag);
1932                 return (error);
1933         }
1934 
1935         return (0);
1936 }
1937 
1938 void
1939 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1940 {
1941         sa_handle_destroy(hdl);
1942         sa_buf_rele(db, tag);
1943 }
1944 
1945 /*
1946  * Given an object number, return its parent object number and whether
1947  * or not the object is an extended attribute directory.
1948  */
1949 static int
1950 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
1951     uint64_t *pobjp, int *is_xattrdir)
1952 {
1953         uint64_t parent;
1954         uint64_t pflags;
1955         uint64_t mode;
1956         uint64_t parent_mode;
1957         sa_bulk_attr_t bulk[3];
1958         sa_handle_t *sa_hdl;
1959         dmu_buf_t *sa_db;
1960         int count = 0;
1961         int error;
1962 
1963         SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
1964             &parent, sizeof (parent));
1965         SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
1966             &pflags, sizeof (pflags));
1967         SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1968             &mode, sizeof (mode));
1969 
1970         if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
1971                 return (error);
1972 
1973         /*
1974          * When a link is removed its parent pointer is not changed and will
1975          * be invalid.  There are two cases where a link is removed but the
1976          * file stays around, when it goes to the delete queue and when there
1977          * are additional links.
1978          */
1979         error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
1980         if (error != 0)
1981                 return (error);
1982 
1983         error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
1984         zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
1985         if (error != 0)
1986                 return (error);
1987 
1988         *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
1989 
1990         /*
1991          * Extended attributes can be applied to files, directories, etc.
1992          * Otherwise the parent must be a directory.
1993          */
1994         if (!*is_xattrdir && !S_ISDIR(parent_mode))
1995                 return (SET_ERROR(EINVAL));
1996 
1997         *pobjp = parent;
1998 
1999         return (0);
2000 }
2001 
2002 /*
2003  * Given an object number, return some zpl level statistics
2004  */
2005 static int
2006 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2007     zfs_stat_t *sb)
2008 {
2009         sa_bulk_attr_t bulk[4];
2010         int count = 0;
2011 
2012         SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2013             &sb->zs_mode, sizeof (sb->zs_mode));
2014         SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2015             &sb->zs_gen, sizeof (sb->zs_gen));
2016         SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2017             &sb->zs_links, sizeof (sb->zs_links));
2018         SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2019             &sb->zs_ctime, sizeof (sb->zs_ctime));
2020 
2021         return (sa_bulk_lookup(hdl, bulk, count));
2022 }
2023 
2024 static int
2025 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2026     sa_attr_type_t *sa_table, char *buf, int len)
2027 {
2028         sa_handle_t *sa_hdl;
2029         sa_handle_t *prevhdl = NULL;
2030         dmu_buf_t *prevdb = NULL;
2031         dmu_buf_t *sa_db = NULL;
2032         char *path = buf + len - 1;
2033         int error;
2034 
2035         *path = '\0';
2036         sa_hdl = hdl;
2037 
2038         for (;;) {
2039                 uint64_t pobj;
2040                 char component[MAXNAMELEN + 2];
2041                 size_t complen;
2042                 int is_xattrdir;
2043 
2044                 if (prevdb)
2045                         zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2046 
2047                 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2048                     &is_xattrdir)) != 0)
2049                         break;
2050 
2051                 if (pobj == obj) {
2052                         if (path[0] != '/')
2053                                 *--path = '/';
2054                         break;
2055                 }
2056 
2057                 component[0] = '/';
2058                 if (is_xattrdir) {
2059                         (void) sprintf(component + 1, "<xattrdir>");
2060                 } else {
2061                         error = zap_value_search(osp, pobj, obj,
2062                             ZFS_DIRENT_OBJ(-1ULL), component + 1);
2063                         if (error != 0)
2064                                 break;
2065                 }
2066 
2067                 complen = strlen(component);
2068                 path -= complen;
2069                 ASSERT(path >= buf);
2070                 bcopy(component, path, complen);
2071                 obj = pobj;
2072 
2073                 if (sa_hdl != hdl) {
2074                         prevhdl = sa_hdl;
2075                         prevdb = sa_db;
2076                 }
2077                 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2078                 if (error != 0) {
2079                         sa_hdl = prevhdl;
2080                         sa_db = prevdb;
2081                         break;
2082                 }
2083         }
2084 
2085         if (sa_hdl != NULL && sa_hdl != hdl) {
2086                 ASSERT(sa_db != NULL);
2087                 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2088         }
2089 
2090         if (error == 0)
2091                 (void) memmove(buf, path, buf + len - path);
2092 
2093         return (error);
2094 }
2095 
2096 int
2097 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2098 {
2099         sa_attr_type_t *sa_table;
2100         sa_handle_t *hdl;
2101         dmu_buf_t *db;
2102         int error;
2103 
2104         error = zfs_sa_setup(osp, &sa_table);
2105         if (error != 0)
2106                 return (error);
2107 
2108         error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2109         if (error != 0)
2110                 return (error);
2111 
2112         error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2113 
2114         zfs_release_sa_handle(hdl, db, FTAG);
2115         return (error);
2116 }
2117 
2118 int
2119 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2120     char *buf, int len)
2121 {
2122         char *path = buf + len - 1;
2123         sa_attr_type_t *sa_table;
2124         sa_handle_t *hdl;
2125         dmu_buf_t *db;
2126         int error;
2127 
2128         *path = '\0';
2129 
2130         error = zfs_sa_setup(osp, &sa_table);
2131         if (error != 0)
2132                 return (error);
2133 
2134         error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2135         if (error != 0)
2136                 return (error);
2137 
2138         error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2139         if (error != 0) {
2140                 zfs_release_sa_handle(hdl, db, FTAG);
2141                 return (error);
2142         }
2143 
2144         error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2145 
2146         zfs_release_sa_handle(hdl, db, FTAG);
2147         return (error);
2148 }