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 /* Portions Copyright 2010 Robert Milkowski */
  28 
  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/vfs.h>
  36 #include <sys/vfs_opreg.h>
  37 #include <sys/vnode.h>
  38 #include <sys/file.h>
  39 #include <sys/stat.h>
  40 #include <sys/kmem.h>
  41 #include <sys/taskq.h>
  42 #include <sys/uio.h>
  43 #include <sys/vmsystm.h>
  44 #include <sys/atomic.h>
  45 #include <sys/vm.h>
  46 #include <vm/seg_vn.h>
  47 #include <vm/pvn.h>
  48 #include <vm/as.h>
  49 #include <vm/kpm.h>
  50 #include <vm/seg_kpm.h>
  51 #include <sys/mman.h>
  52 #include <sys/pathname.h>
  53 #include <sys/cmn_err.h>
  54 #include <sys/errno.h>
  55 #include <sys/unistd.h>
  56 #include <sys/zfs_dir.h>
  57 #include <sys/zfs_acl.h>
  58 #include <sys/zfs_ioctl.h>
  59 #include <sys/fs/zfs.h>
  60 #include <sys/dmu.h>
  61 #include <sys/dmu_objset.h>
  62 #include <sys/spa.h>
  63 #include <sys/txg.h>
  64 #include <sys/dbuf.h>
  65 #include <sys/zap.h>
  66 #include <sys/sa.h>
  67 #include <sys/dirent.h>
  68 #include <sys/policy.h>
  69 #include <sys/sunddi.h>
  70 #include <sys/filio.h>
  71 #include <sys/sid.h>
  72 #include "fs/fs_subr.h"
  73 #include <sys/zfs_ctldir.h>
  74 #include <sys/zfs_fuid.h>
  75 #include <sys/zfs_sa.h>
  76 #include <sys/dnlc.h>
  77 #include <sys/zfs_rlock.h>
  78 #include <sys/extdirent.h>
  79 #include <sys/kidmap.h>
  80 #include <sys/cred.h>
  81 #include <sys/attr.h>
  82 
  83 /*
  84  * Programming rules.
  85  *
  86  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
  87  * properly lock its in-core state, create a DMU transaction, do the work,
  88  * record this work in the intent log (ZIL), commit the DMU transaction,
  89  * and wait for the intent log to commit if it is a synchronous operation.
  90  * Moreover, the vnode ops must work in both normal and log replay context.
  91  * The ordering of events is important to avoid deadlocks and references
  92  * to freed memory.  The example below illustrates the following Big Rules:
  93  *
  94  *  (1) A check must be made in each zfs thread for a mounted file system.
  95  *      This is done avoiding races using ZFS_ENTER(zfsvfs).
  96  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
  97  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
  98  *      can return EIO from the calling function.
  99  *
 100  *  (2) VN_RELE() should always be the last thing except for zil_commit()
 101  *      (if necessary) and ZFS_EXIT(). This is for 3 reasons:
 102  *      First, if it's the last reference, the vnode/znode
 103  *      can be freed, so the zp may point to freed memory.  Second, the last
 104  *      reference will call zfs_zinactive(), which may induce a lot of work --
 105  *      pushing cached pages (which acquires range locks) and syncing out
 106  *      cached atime changes.  Third, zfs_zinactive() may require a new tx,
 107  *      which could deadlock the system if you were already holding one.
 108  *      If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
 109  *
 110  *  (3) All range locks must be grabbed before calling dmu_tx_assign(),
 111  *      as they can span dmu_tx_assign() calls.
 112  *
 113  *  (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
 114  *      This is critical because we don't want to block while holding locks.
 115  *      Note, in particular, that if a lock is sometimes acquired before
 116  *      the tx assigns, and sometimes after (e.g. z_lock), then failing to
 117  *      use a non-blocking assign can deadlock the system.  The scenario:
 118  *
 119  *      Thread A has grabbed a lock before calling dmu_tx_assign().
 120  *      Thread B is in an already-assigned tx, and blocks for this lock.
 121  *      Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
 122  *      forever, because the previous txg can't quiesce until B's tx commits.
 123  *
 124  *      If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
 125  *      then drop all locks, call dmu_tx_wait(), and try again.
 126  *
 127  *  (5) If the operation succeeded, generate the intent log entry for it
 128  *      before dropping locks.  This ensures that the ordering of events
 129  *      in the intent log matches the order in which they actually occurred.
 130  *      During ZIL replay the zfs_log_* functions will update the sequence
 131  *      number to indicate the zil transaction has replayed.
 132  *
 133  *  (6) At the end of each vnode op, the DMU tx must always commit,
 134  *      regardless of whether there were any errors.
 135  *
 136  *  (7) After dropping all locks, invoke zil_commit(zilog, foid)
 137  *      to ensure that synchronous semantics are provided when necessary.
 138  *
 139  * In general, this is how things should be ordered in each vnode op:
 140  *
 141  *      ZFS_ENTER(zfsvfs);              // exit if unmounted
 142  * top:
 143  *      zfs_dirent_lock(&dl, ...)   // lock directory entry (may VN_HOLD())
 144  *      rw_enter(...);                  // grab any other locks you need
 145  *      tx = dmu_tx_create(...);        // get DMU tx
 146  *      dmu_tx_hold_*();                // hold each object you might modify
 147  *      error = dmu_tx_assign(tx, TXG_NOWAIT);  // try to assign
 148  *      if (error) {
 149  *              rw_exit(...);           // drop locks
 150  *              zfs_dirent_unlock(dl);  // unlock directory entry
 151  *              VN_RELE(...);           // release held vnodes
 152  *              if (error == ERESTART) {
 153  *                      dmu_tx_wait(tx);
 154  *                      dmu_tx_abort(tx);
 155  *                      goto top;
 156  *              }
 157  *              dmu_tx_abort(tx);       // abort DMU tx
 158  *              ZFS_EXIT(zfsvfs);       // finished in zfs
 159  *              return (error);         // really out of space
 160  *      }
 161  *      error = do_real_work();         // do whatever this VOP does
 162  *      if (error == 0)
 163  *              zfs_log_*(...);         // on success, make ZIL entry
 164  *      dmu_tx_commit(tx);              // commit DMU tx -- error or not
 165  *      rw_exit(...);                   // drop locks
 166  *      zfs_dirent_unlock(dl);          // unlock directory entry
 167  *      VN_RELE(...);                   // release held vnodes
 168  *      zil_commit(zilog, foid);        // synchronous when necessary
 169  *      ZFS_EXIT(zfsvfs);               // finished in zfs
 170  *      return (error);                 // done, report error
 171  */
 172 
 173 /* ARGSUSED */
 174 static int
 175 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
 176 {
 177         znode_t *zp = VTOZ(*vpp);
 178         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 179 
 180         ZFS_ENTER(zfsvfs);
 181         ZFS_VERIFY_ZP(zp);
 182 
 183         if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
 184             ((flag & FAPPEND) == 0)) {
 185                 ZFS_EXIT(zfsvfs);
 186                 return (SET_ERROR(EPERM));
 187         }
 188 
 189         if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
 190             ZTOV(zp)->v_type == VREG &&
 191             !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
 192                 if (fs_vscan(*vpp, cr, 0) != 0) {
 193                         ZFS_EXIT(zfsvfs);
 194                         return (SET_ERROR(EACCES));
 195                 }
 196         }
 197 
 198         /* Keep a count of the synchronous opens in the znode */
 199         if (flag & (FSYNC | FDSYNC))
 200                 atomic_inc_32(&zp->z_sync_cnt);
 201 
 202         ZFS_EXIT(zfsvfs);
 203         return (0);
 204 }
 205 
 206 /* ARGSUSED */
 207 static int
 208 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
 209     caller_context_t *ct)
 210 {
 211         znode_t *zp = VTOZ(vp);
 212         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 213 
 214         /*
 215          * Clean up any locks held by this process on the vp.
 216          */
 217         cleanlocks(vp, ddi_get_pid(), 0);
 218         cleanshares(vp, ddi_get_pid());
 219 
 220         ZFS_ENTER(zfsvfs);
 221         ZFS_VERIFY_ZP(zp);
 222 
 223         /* Decrement the synchronous opens in the znode */
 224         if ((flag & (FSYNC | FDSYNC)) && (count == 1))
 225                 atomic_dec_32(&zp->z_sync_cnt);
 226 
 227         if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
 228             ZTOV(zp)->v_type == VREG &&
 229             !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
 230                 VERIFY(fs_vscan(vp, cr, 1) == 0);
 231 
 232         ZFS_EXIT(zfsvfs);
 233         return (0);
 234 }
 235 
 236 /*
 237  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
 238  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
 239  */
 240 static int
 241 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
 242 {
 243         znode_t *zp = VTOZ(vp);
 244         uint64_t noff = (uint64_t)*off; /* new offset */
 245         uint64_t file_sz;
 246         int error;
 247         boolean_t hole;
 248 
 249         file_sz = zp->z_size;
 250         if (noff >= file_sz)  {
 251                 return (SET_ERROR(ENXIO));
 252         }
 253 
 254         if (cmd == _FIO_SEEK_HOLE)
 255                 hole = B_TRUE;
 256         else
 257                 hole = B_FALSE;
 258 
 259         error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
 260 
 261         /* end of file? */
 262         if ((error == ESRCH) || (noff > file_sz)) {
 263                 /*
 264                  * Handle the virtual hole at the end of file.
 265                  */
 266                 if (hole) {
 267                         *off = file_sz;
 268                         return (0);
 269                 }
 270                 return (SET_ERROR(ENXIO));
 271         }
 272 
 273         if (noff < *off)
 274                 return (error);
 275         *off = noff;
 276         return (error);
 277 }
 278 
 279 /* ARGSUSED */
 280 static int
 281 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
 282     int *rvalp, caller_context_t *ct)
 283 {
 284         offset_t off;
 285         int error;
 286         zfsvfs_t *zfsvfs;
 287         znode_t *zp;
 288 
 289         switch (com) {
 290         case _FIOFFS:
 291                 return (zfs_sync(vp->v_vfsp, 0, cred));
 292 
 293                 /*
 294                  * The following two ioctls are used by bfu.  Faking out,
 295                  * necessary to avoid bfu errors.
 296                  */
 297         case _FIOGDIO:
 298         case _FIOSDIO:
 299                 return (0);
 300 
 301         case _FIO_SEEK_DATA:
 302         case _FIO_SEEK_HOLE:
 303                 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
 304                         return (SET_ERROR(EFAULT));
 305 
 306                 zp = VTOZ(vp);
 307                 zfsvfs = zp->z_zfsvfs;
 308                 ZFS_ENTER(zfsvfs);
 309                 ZFS_VERIFY_ZP(zp);
 310 
 311                 /* offset parameter is in/out */
 312                 error = zfs_holey(vp, com, &off);
 313                 ZFS_EXIT(zfsvfs);
 314                 if (error)
 315                         return (error);
 316                 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
 317                         return (SET_ERROR(EFAULT));
 318                 return (0);
 319         }
 320         return (SET_ERROR(ENOTTY));
 321 }
 322 
 323 /*
 324  * Utility functions to map and unmap a single physical page.  These
 325  * are used to manage the mappable copies of ZFS file data, and therefore
 326  * do not update ref/mod bits.
 327  */
 328 caddr_t
 329 zfs_map_page(page_t *pp, enum seg_rw rw)
 330 {
 331         if (kpm_enable)
 332                 return (hat_kpm_mapin(pp, 0));
 333         ASSERT(rw == S_READ || rw == S_WRITE);
 334         return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
 335             (caddr_t)-1));
 336 }
 337 
 338 void
 339 zfs_unmap_page(page_t *pp, caddr_t addr)
 340 {
 341         if (kpm_enable) {
 342                 hat_kpm_mapout(pp, 0, addr);
 343         } else {
 344                 ppmapout(addr);
 345         }
 346 }
 347 
 348 /*
 349  * When a file is memory mapped, we must keep the IO data synchronized
 350  * between the DMU cache and the memory mapped pages.  What this means:
 351  *
 352  * On Write:    If we find a memory mapped page, we write to *both*
 353  *              the page and the dmu buffer.
 354  */
 355 static void
 356 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
 357 {
 358         int64_t off;
 359 
 360         off = start & PAGEOFFSET;
 361         for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
 362                 page_t *pp;
 363                 uint64_t nbytes = MIN(PAGESIZE - off, len);
 364 
 365                 if (pp = page_lookup(vp, start, SE_SHARED)) {
 366                         caddr_t va;
 367 
 368                         va = zfs_map_page(pp, S_WRITE);
 369                         (void) dmu_read(os, oid, start+off, nbytes, va+off,
 370                             DMU_READ_PREFETCH);
 371                         zfs_unmap_page(pp, va);
 372                         page_unlock(pp);
 373                 }
 374                 len -= nbytes;
 375                 off = 0;
 376         }
 377 }
 378 
 379 /*
 380  * When a file is memory mapped, we must keep the IO data synchronized
 381  * between the DMU cache and the memory mapped pages.  What this means:
 382  *
 383  * On Read:     We "read" preferentially from memory mapped pages,
 384  *              else we default from the dmu buffer.
 385  *
 386  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
 387  *       the file is memory mapped.
 388  */
 389 static int
 390 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
 391 {
 392         znode_t *zp = VTOZ(vp);
 393         objset_t *os = zp->z_zfsvfs->z_os;
 394         int64_t start, off;
 395         int len = nbytes;
 396         int error = 0;
 397 
 398         start = uio->uio_loffset;
 399         off = start & PAGEOFFSET;
 400         for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
 401                 page_t *pp;
 402                 uint64_t bytes = MIN(PAGESIZE - off, len);
 403 
 404                 if (pp = page_lookup(vp, start, SE_SHARED)) {
 405                         caddr_t va;
 406 
 407                         va = zfs_map_page(pp, S_READ);
 408                         error = uiomove(va + off, bytes, UIO_READ, uio);
 409                         zfs_unmap_page(pp, va);
 410                         page_unlock(pp);
 411                 } else {
 412                         error = dmu_read_uio(os, zp->z_id, uio, bytes);
 413                 }
 414                 len -= bytes;
 415                 off = 0;
 416                 if (error)
 417                         break;
 418         }
 419         return (error);
 420 }
 421 
 422 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
 423 
 424 /*
 425  * Read bytes from specified file into supplied buffer.
 426  *
 427  *      IN:     vp      - vnode of file to be read from.
 428  *              uio     - structure supplying read location, range info,
 429  *                        and return buffer.
 430  *              ioflag  - SYNC flags; used to provide FRSYNC semantics.
 431  *              cr      - credentials of caller.
 432  *              ct      - caller context
 433  *
 434  *      OUT:    uio     - updated offset and range, buffer filled.
 435  *
 436  *      RETURN: 0 on success, error code on failure.
 437  *
 438  * Side Effects:
 439  *      vp - atime updated if byte count > 0
 440  */
 441 /* ARGSUSED */
 442 static int
 443 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 444 {
 445         znode_t         *zp = VTOZ(vp);
 446         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
 447         objset_t        *os;
 448         ssize_t         n, nbytes;
 449         int             error = 0;
 450         rl_t            *rl;
 451         xuio_t          *xuio = NULL;
 452 
 453         ZFS_ENTER(zfsvfs);
 454         ZFS_VERIFY_ZP(zp);
 455         os = zfsvfs->z_os;
 456 
 457         if (zp->z_pflags & ZFS_AV_QUARANTINED) {
 458                 ZFS_EXIT(zfsvfs);
 459                 return (SET_ERROR(EACCES));
 460         }
 461 
 462         /*
 463          * Validate file offset
 464          */
 465         if (uio->uio_loffset < (offset_t)0) {
 466                 ZFS_EXIT(zfsvfs);
 467                 return (SET_ERROR(EINVAL));
 468         }
 469 
 470         /*
 471          * Fasttrack empty reads
 472          */
 473         if (uio->uio_resid == 0) {
 474                 ZFS_EXIT(zfsvfs);
 475                 return (0);
 476         }
 477 
 478         /*
 479          * Check for mandatory locks
 480          */
 481         if (MANDMODE(zp->z_mode)) {
 482                 if (error = chklock(vp, FREAD,
 483                     uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
 484                         ZFS_EXIT(zfsvfs);
 485                         return (error);
 486                 }
 487         }
 488 
 489         /*
 490          * If we're in FRSYNC mode, sync out this znode before reading it.
 491          */
 492         if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
 493                 zil_commit(zfsvfs->z_log, zp->z_id);
 494 
 495         /*
 496          * Lock the range against changes.
 497          */
 498         rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
 499 
 500         /*
 501          * If we are reading past end-of-file we can skip
 502          * to the end; but we might still need to set atime.
 503          */
 504         if (uio->uio_loffset >= zp->z_size) {
 505                 error = 0;
 506                 goto out;
 507         }
 508 
 509         ASSERT(uio->uio_loffset < zp->z_size);
 510         n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
 511 
 512         if ((uio->uio_extflg == UIO_XUIO) &&
 513             (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
 514                 int nblk;
 515                 int blksz = zp->z_blksz;
 516                 uint64_t offset = uio->uio_loffset;
 517 
 518                 xuio = (xuio_t *)uio;
 519                 if ((ISP2(blksz))) {
 520                         nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
 521                             blksz)) / blksz;
 522                 } else {
 523                         ASSERT(offset + n <= blksz);
 524                         nblk = 1;
 525                 }
 526                 (void) dmu_xuio_init(xuio, nblk);
 527 
 528                 if (vn_has_cached_data(vp)) {
 529                         /*
 530                          * For simplicity, we always allocate a full buffer
 531                          * even if we only expect to read a portion of a block.
 532                          */
 533                         while (--nblk >= 0) {
 534                                 (void) dmu_xuio_add(xuio,
 535                                     dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
 536                                     blksz), 0, blksz);
 537                         }
 538                 }
 539         }
 540 
 541         while (n > 0) {
 542                 nbytes = MIN(n, zfs_read_chunk_size -
 543                     P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
 544 
 545                 if (vn_has_cached_data(vp))
 546                         error = mappedread(vp, nbytes, uio);
 547                 else
 548                         error = dmu_read_uio(os, zp->z_id, uio, nbytes);
 549                 if (error) {
 550                         /* convert checksum errors into IO errors */
 551                         if (error == ECKSUM)
 552                                 error = SET_ERROR(EIO);
 553                         break;
 554                 }
 555 
 556                 n -= nbytes;
 557         }
 558 out:
 559         zfs_range_unlock(rl);
 560 
 561         ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
 562         ZFS_EXIT(zfsvfs);
 563         return (error);
 564 }
 565 
 566 /*
 567  * Write the bytes to a file.
 568  *
 569  *      IN:     vp      - vnode of file to be written to.
 570  *              uio     - structure supplying write location, range info,
 571  *                        and data buffer.
 572  *              ioflag  - FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
 573  *                        set if in append mode.
 574  *              cr      - credentials of caller.
 575  *              ct      - caller context (NFS/CIFS fem monitor only)
 576  *
 577  *      OUT:    uio     - updated offset and range.
 578  *
 579  *      RETURN: 0 on success, error code on failure.
 580  *
 581  * Timestamps:
 582  *      vp - ctime|mtime updated if byte count > 0
 583  */
 584 
 585 /* ARGSUSED */
 586 static int
 587 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 588 {
 589         znode_t         *zp = VTOZ(vp);
 590         rlim64_t        limit = uio->uio_llimit;
 591         ssize_t         start_resid = uio->uio_resid;
 592         ssize_t         tx_bytes;
 593         uint64_t        end_size;
 594         dmu_tx_t        *tx;
 595         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
 596         zilog_t         *zilog;
 597         offset_t        woff;
 598         ssize_t         n, nbytes;
 599         rl_t            *rl;
 600         int             max_blksz = zfsvfs->z_max_blksz;
 601         int             error = 0;
 602         arc_buf_t       *abuf;
 603         iovec_t         *aiov = NULL;
 604         xuio_t          *xuio = NULL;
 605         int             i_iov = 0;
 606         int             iovcnt = uio->uio_iovcnt;
 607         iovec_t         *iovp = uio->uio_iov;
 608         int             write_eof;
 609         int             count = 0;
 610         sa_bulk_attr_t  bulk[4];
 611         uint64_t        mtime[2], ctime[2];
 612 
 613         /*
 614          * Fasttrack empty write
 615          */
 616         n = start_resid;
 617         if (n == 0)
 618                 return (0);
 619 
 620         if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
 621                 limit = MAXOFFSET_T;
 622 
 623         ZFS_ENTER(zfsvfs);
 624         ZFS_VERIFY_ZP(zp);
 625 
 626         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
 627         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
 628         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
 629             &zp->z_size, 8);
 630         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
 631             &zp->z_pflags, 8);
 632 
 633         /*
 634          * If immutable or not appending then return EPERM
 635          */
 636         if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
 637             ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
 638             (uio->uio_loffset < zp->z_size))) {
 639                 ZFS_EXIT(zfsvfs);
 640                 return (SET_ERROR(EPERM));
 641         }
 642 
 643         zilog = zfsvfs->z_log;
 644 
 645         /*
 646          * Validate file offset
 647          */
 648         woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
 649         if (woff < 0) {
 650                 ZFS_EXIT(zfsvfs);
 651                 return (SET_ERROR(EINVAL));
 652         }
 653 
 654         /*
 655          * Check for mandatory locks before calling zfs_range_lock()
 656          * in order to prevent a deadlock with locks set via fcntl().
 657          */
 658         if (MANDMODE((mode_t)zp->z_mode) &&
 659             (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
 660                 ZFS_EXIT(zfsvfs);
 661                 return (error);
 662         }
 663 
 664         /*
 665          * Pre-fault the pages to ensure slow (eg NFS) pages
 666          * don't hold up txg.
 667          * Skip this if uio contains loaned arc_buf.
 668          */
 669         if ((uio->uio_extflg == UIO_XUIO) &&
 670             (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
 671                 xuio = (xuio_t *)uio;
 672         else
 673                 uio_prefaultpages(MIN(n, max_blksz), uio);
 674 
 675         /*
 676          * If in append mode, set the io offset pointer to eof.
 677          */
 678         if (ioflag & FAPPEND) {
 679                 /*
 680                  * Obtain an appending range lock to guarantee file append
 681                  * semantics.  We reset the write offset once we have the lock.
 682                  */
 683                 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
 684                 woff = rl->r_off;
 685                 if (rl->r_len == UINT64_MAX) {
 686                         /*
 687                          * We overlocked the file because this write will cause
 688                          * the file block size to increase.
 689                          * Note that zp_size cannot change with this lock held.
 690                          */
 691                         woff = zp->z_size;
 692                 }
 693                 uio->uio_loffset = woff;
 694         } else {
 695                 /*
 696                  * Note that if the file block size will change as a result of
 697                  * this write, then this range lock will lock the entire file
 698                  * so that we can re-write the block safely.
 699                  */
 700                 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
 701         }
 702 
 703         if (woff >= limit) {
 704                 zfs_range_unlock(rl);
 705                 ZFS_EXIT(zfsvfs);
 706                 return (SET_ERROR(EFBIG));
 707         }
 708 
 709         if ((woff + n) > limit || woff > (limit - n))
 710                 n = limit - woff;
 711 
 712         /* Will this write extend the file length? */
 713         write_eof = (woff + n > zp->z_size);
 714 
 715         end_size = MAX(zp->z_size, woff + n);
 716 
 717         /*
 718          * Write the file in reasonable size chunks.  Each chunk is written
 719          * in a separate transaction; this keeps the intent log records small
 720          * and allows us to do more fine-grained space accounting.
 721          */
 722         while (n > 0) {
 723                 abuf = NULL;
 724                 woff = uio->uio_loffset;
 725 again:
 726                 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
 727                     zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
 728                         if (abuf != NULL)
 729                                 dmu_return_arcbuf(abuf);
 730                         error = SET_ERROR(EDQUOT);
 731                         break;
 732                 }
 733 
 734                 if (xuio && abuf == NULL) {
 735                         ASSERT(i_iov < iovcnt);
 736                         aiov = &iovp[i_iov];
 737                         abuf = dmu_xuio_arcbuf(xuio, i_iov);
 738                         dmu_xuio_clear(xuio, i_iov);
 739                         DTRACE_PROBE3(zfs_cp_write, int, i_iov,
 740                             iovec_t *, aiov, arc_buf_t *, abuf);
 741                         ASSERT((aiov->iov_base == abuf->b_data) ||
 742                             ((char *)aiov->iov_base - (char *)abuf->b_data +
 743                             aiov->iov_len == arc_buf_size(abuf)));
 744                         i_iov++;
 745                 } else if (abuf == NULL && n >= max_blksz &&
 746                     woff >= zp->z_size &&
 747                     P2PHASE(woff, max_blksz) == 0 &&
 748                     zp->z_blksz == max_blksz) {
 749                         /*
 750                          * This write covers a full block.  "Borrow" a buffer
 751                          * from the dmu so that we can fill it before we enter
 752                          * a transaction.  This avoids the possibility of
 753                          * holding up the transaction if the data copy hangs
 754                          * up on a pagefault (e.g., from an NFS server mapping).
 755                          */
 756                         size_t cbytes;
 757 
 758                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
 759                             max_blksz);
 760                         ASSERT(abuf != NULL);
 761                         ASSERT(arc_buf_size(abuf) == max_blksz);
 762                         if (error = uiocopy(abuf->b_data, max_blksz,
 763                             UIO_WRITE, uio, &cbytes)) {
 764                                 dmu_return_arcbuf(abuf);
 765                                 break;
 766                         }
 767                         ASSERT(cbytes == max_blksz);
 768                 }
 769 
 770                 /*
 771                  * Start a transaction.
 772                  */
 773                 tx = dmu_tx_create(zfsvfs->z_os);
 774                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 775                 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
 776                 zfs_sa_upgrade_txholds(tx, zp);
 777                 error = dmu_tx_assign(tx, TXG_NOWAIT);
 778                 if (error) {
 779                         if (error == ERESTART) {
 780                                 dmu_tx_wait(tx);
 781                                 dmu_tx_abort(tx);
 782                                 goto again;
 783                         }
 784                         dmu_tx_abort(tx);
 785                         if (abuf != NULL)
 786                                 dmu_return_arcbuf(abuf);
 787                         break;
 788                 }
 789 
 790                 /*
 791                  * If zfs_range_lock() over-locked we grow the blocksize
 792                  * and then reduce the lock range.  This will only happen
 793                  * on the first iteration since zfs_range_reduce() will
 794                  * shrink down r_len to the appropriate size.
 795                  */
 796                 if (rl->r_len == UINT64_MAX) {
 797                         uint64_t new_blksz;
 798 
 799                         if (zp->z_blksz > max_blksz) {
 800                                 ASSERT(!ISP2(zp->z_blksz));
 801                                 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
 802                         } else {
 803                                 new_blksz = MIN(end_size, max_blksz);
 804                         }
 805                         zfs_grow_blocksize(zp, new_blksz, tx);
 806                         zfs_range_reduce(rl, woff, n);
 807                 }
 808 
 809                 /*
 810                  * XXX - should we really limit each write to z_max_blksz?
 811                  * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
 812                  */
 813                 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
 814 
 815                 if (abuf == NULL) {
 816                         tx_bytes = uio->uio_resid;
 817                         error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
 818                             uio, nbytes, tx);
 819                         tx_bytes -= uio->uio_resid;
 820                 } else {
 821                         tx_bytes = nbytes;
 822                         ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
 823                         /*
 824                          * If this is not a full block write, but we are
 825                          * extending the file past EOF and this data starts
 826                          * block-aligned, use assign_arcbuf().  Otherwise,
 827                          * write via dmu_write().
 828                          */
 829                         if (tx_bytes < max_blksz && (!write_eof ||
 830                             aiov->iov_base != abuf->b_data)) {
 831                                 ASSERT(xuio);
 832                                 dmu_write(zfsvfs->z_os, zp->z_id, woff,
 833                                     aiov->iov_len, aiov->iov_base, tx);
 834                                 dmu_return_arcbuf(abuf);
 835                                 xuio_stat_wbuf_copied();
 836                         } else {
 837                                 ASSERT(xuio || tx_bytes == max_blksz);
 838                                 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
 839                                     woff, abuf, tx);
 840                         }
 841                         ASSERT(tx_bytes <= uio->uio_resid);
 842                         uioskip(uio, tx_bytes);
 843                 }
 844                 if (tx_bytes && vn_has_cached_data(vp)) {
 845                         update_pages(vp, woff,
 846                             tx_bytes, zfsvfs->z_os, zp->z_id);
 847                 }
 848 
 849                 /*
 850                  * If we made no progress, we're done.  If we made even
 851                  * partial progress, update the znode and ZIL accordingly.
 852                  */
 853                 if (tx_bytes == 0) {
 854                         (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
 855                             (void *)&zp->z_size, sizeof (uint64_t), tx);
 856                         dmu_tx_commit(tx);
 857                         ASSERT(error != 0);
 858                         break;
 859                 }
 860 
 861                 /*
 862                  * Clear Set-UID/Set-GID bits on successful write if not
 863                  * privileged and at least one of the excute bits is set.
 864                  *
 865                  * It would be nice to to this after all writes have
 866                  * been done, but that would still expose the ISUID/ISGID
 867                  * to another app after the partial write is committed.
 868                  *
 869                  * Note: we don't call zfs_fuid_map_id() here because
 870                  * user 0 is not an ephemeral uid.
 871                  */
 872                 mutex_enter(&zp->z_acl_lock);
 873                 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
 874                     (S_IXUSR >> 6))) != 0 &&
 875                     (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
 876                     secpolicy_vnode_setid_retain(cr,
 877                     (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
 878                         uint64_t newmode;
 879                         zp->z_mode &= ~(S_ISUID | S_ISGID);
 880                         newmode = zp->z_mode;
 881                         (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
 882                             (void *)&newmode, sizeof (uint64_t), tx);
 883                 }
 884                 mutex_exit(&zp->z_acl_lock);
 885 
 886                 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
 887                     B_TRUE);
 888 
 889                 /*
 890                  * Update the file size (zp_size) if it has changed;
 891                  * account for possible concurrent updates.
 892                  */
 893                 while ((end_size = zp->z_size) < uio->uio_loffset) {
 894                         (void) atomic_cas_64(&zp->z_size, end_size,
 895                             uio->uio_loffset);
 896                         ASSERT(error == 0);
 897                 }
 898                 /*
 899                  * If we are replaying and eof is non zero then force
 900                  * the file size to the specified eof. Note, there's no
 901                  * concurrency during replay.
 902                  */
 903                 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
 904                         zp->z_size = zfsvfs->z_replay_eof;
 905 
 906                 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
 907 
 908                 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
 909                 dmu_tx_commit(tx);
 910 
 911                 if (error != 0)
 912                         break;
 913                 ASSERT(tx_bytes == nbytes);
 914                 n -= nbytes;
 915 
 916                 if (!xuio && n > 0)
 917                         uio_prefaultpages(MIN(n, max_blksz), uio);
 918         }
 919 
 920         zfs_range_unlock(rl);
 921 
 922         /*
 923          * If we're in replay mode, or we made no progress, return error.
 924          * Otherwise, it's at least a partial write, so it's successful.
 925          */
 926         if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
 927                 ZFS_EXIT(zfsvfs);
 928                 return (error);
 929         }
 930 
 931         if (ioflag & (FSYNC | FDSYNC) ||
 932             zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
 933                 zil_commit(zilog, zp->z_id);
 934 
 935         ZFS_EXIT(zfsvfs);
 936         return (0);
 937 }
 938 
 939 void
 940 zfs_get_done(zgd_t *zgd, int error)
 941 {
 942         znode_t *zp = zgd->zgd_private;
 943         objset_t *os = zp->z_zfsvfs->z_os;
 944 
 945         if (zgd->zgd_db)
 946                 dmu_buf_rele(zgd->zgd_db, zgd);
 947 
 948         zfs_range_unlock(zgd->zgd_rl);
 949 
 950         /*
 951          * Release the vnode asynchronously as we currently have the
 952          * txg stopped from syncing.
 953          */
 954         VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
 955 
 956         if (error == 0 && zgd->zgd_bp)
 957                 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
 958 
 959         kmem_free(zgd, sizeof (zgd_t));
 960 }
 961 
 962 #ifdef DEBUG
 963 static int zil_fault_io = 0;
 964 #endif
 965 
 966 /*
 967  * Get data to generate a TX_WRITE intent log record.
 968  */
 969 int
 970 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
 971 {
 972         zfsvfs_t *zfsvfs = arg;
 973         objset_t *os = zfsvfs->z_os;
 974         znode_t *zp;
 975         uint64_t object = lr->lr_foid;
 976         uint64_t offset = lr->lr_offset;
 977         uint64_t size = lr->lr_length;
 978         blkptr_t *bp = &lr->lr_blkptr;
 979         dmu_buf_t *db;
 980         zgd_t *zgd;
 981         int error = 0;
 982 
 983         ASSERT(zio != NULL);
 984         ASSERT(size != 0);
 985 
 986         /*
 987          * Nothing to do if the file has been removed
 988          */
 989         if (zfs_zget(zfsvfs, object, &zp) != 0)
 990                 return (SET_ERROR(ENOENT));
 991         if (zp->z_unlinked) {
 992                 /*
 993                  * Release the vnode asynchronously as we currently have the
 994                  * txg stopped from syncing.
 995                  */
 996                 VN_RELE_ASYNC(ZTOV(zp),
 997                     dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
 998                 return (SET_ERROR(ENOENT));
 999         }
1000 
1001         zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1002         zgd->zgd_zilog = zfsvfs->z_log;
1003         zgd->zgd_private = zp;
1004 
1005         /*
1006          * Write records come in two flavors: immediate and indirect.
1007          * For small writes it's cheaper to store the data with the
1008          * log record (immediate); for large writes it's cheaper to
1009          * sync the data and get a pointer to it (indirect) so that
1010          * we don't have to write the data twice.
1011          */
1012         if (buf != NULL) { /* immediate write */
1013                 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1014                 /* test for truncation needs to be done while range locked */
1015                 if (offset >= zp->z_size) {
1016                         error = SET_ERROR(ENOENT);
1017                 } else {
1018                         error = dmu_read(os, object, offset, size, buf,
1019                             DMU_READ_NO_PREFETCH);
1020                 }
1021                 ASSERT(error == 0 || error == ENOENT);
1022         } else { /* indirect write */
1023                 /*
1024                  * Have to lock the whole block to ensure when it's
1025                  * written out and it's checksum is being calculated
1026                  * that no one can change the data. We need to re-check
1027                  * blocksize after we get the lock in case it's changed!
1028                  */
1029                 for (;;) {
1030                         uint64_t blkoff;
1031                         size = zp->z_blksz;
1032                         blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1033                         offset -= blkoff;
1034                         zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1035                             RL_READER);
1036                         if (zp->z_blksz == size)
1037                                 break;
1038                         offset += blkoff;
1039                         zfs_range_unlock(zgd->zgd_rl);
1040                 }
1041                 /* test for truncation needs to be done while range locked */
1042                 if (lr->lr_offset >= zp->z_size)
1043                         error = SET_ERROR(ENOENT);
1044 #ifdef DEBUG
1045                 if (zil_fault_io) {
1046                         error = SET_ERROR(EIO);
1047                         zil_fault_io = 0;
1048                 }
1049 #endif
1050                 if (error == 0)
1051                         error = dmu_buf_hold(os, object, offset, zgd, &db,
1052                             DMU_READ_NO_PREFETCH);
1053 
1054                 if (error == 0) {
1055                         blkptr_t *obp = dmu_buf_get_blkptr(db);
1056                         if (obp) {
1057                                 ASSERT(BP_IS_HOLE(bp));
1058                                 *bp = *obp;
1059                         }
1060 
1061                         zgd->zgd_db = db;
1062                         zgd->zgd_bp = bp;
1063 
1064                         ASSERT(db->db_offset == offset);
1065                         ASSERT(db->db_size == size);
1066 
1067                         error = dmu_sync(zio, lr->lr_common.lrc_txg,
1068                             zfs_get_done, zgd);
1069                         ASSERT(error || lr->lr_length <= zp->z_blksz);
1070 
1071                         /*
1072                          * On success, we need to wait for the write I/O
1073                          * initiated by dmu_sync() to complete before we can
1074                          * release this dbuf.  We will finish everything up
1075                          * in the zfs_get_done() callback.
1076                          */
1077                         if (error == 0)
1078                                 return (0);
1079 
1080                         if (error == EALREADY) {
1081                                 lr->lr_common.lrc_txtype = TX_WRITE2;
1082                                 error = 0;
1083                         }
1084                 }
1085         }
1086 
1087         zfs_get_done(zgd, error);
1088 
1089         return (error);
1090 }
1091 
1092 /*ARGSUSED*/
1093 static int
1094 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1095     caller_context_t *ct)
1096 {
1097         znode_t *zp = VTOZ(vp);
1098         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1099         int error;
1100 
1101         ZFS_ENTER(zfsvfs);
1102         ZFS_VERIFY_ZP(zp);
1103 
1104         if (flag & V_ACE_MASK)
1105                 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1106         else
1107                 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1108 
1109         ZFS_EXIT(zfsvfs);
1110         return (error);
1111 }
1112 
1113 /*
1114  * If vnode is for a device return a specfs vnode instead.
1115  */
1116 static int
1117 specvp_check(vnode_t **vpp, cred_t *cr)
1118 {
1119         int error = 0;
1120 
1121         if (IS_DEVVP(*vpp)) {
1122                 struct vnode *svp;
1123 
1124                 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1125                 VN_RELE(*vpp);
1126                 if (svp == NULL)
1127                         error = SET_ERROR(ENOSYS);
1128                 *vpp = svp;
1129         }
1130         return (error);
1131 }
1132 
1133 
1134 /*
1135  * Lookup an entry in a directory, or an extended attribute directory.
1136  * If it exists, return a held vnode reference for it.
1137  *
1138  *      IN:     dvp     - vnode of directory to search.
1139  *              nm      - name of entry to lookup.
1140  *              pnp     - full pathname to lookup [UNUSED].
1141  *              flags   - LOOKUP_XATTR set if looking for an attribute.
1142  *              rdir    - root directory vnode [UNUSED].
1143  *              cr      - credentials of caller.
1144  *              ct      - caller context
1145  *              direntflags - directory lookup flags
1146  *              realpnp - returned pathname.
1147  *
1148  *      OUT:    vpp     - vnode of located entry, NULL if not found.
1149  *
1150  *      RETURN: 0 on success, error code on failure.
1151  *
1152  * Timestamps:
1153  *      NA
1154  */
1155 /* ARGSUSED */
1156 static int
1157 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1158     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1159     int *direntflags, pathname_t *realpnp)
1160 {
1161         znode_t *zdp = VTOZ(dvp);
1162         zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1163         int     error = 0;
1164 
1165         /* fast path */
1166         if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1167 
1168                 if (dvp->v_type != VDIR) {
1169                         return (SET_ERROR(ENOTDIR));
1170                 } else if (zdp->z_sa_hdl == NULL) {
1171                         return (SET_ERROR(EIO));
1172                 }
1173 
1174                 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1175                         error = zfs_fastaccesschk_execute(zdp, cr);
1176                         if (!error) {
1177                                 *vpp = dvp;
1178                                 VN_HOLD(*vpp);
1179                                 return (0);
1180                         }
1181                         return (error);
1182                 } else {
1183                         vnode_t *tvp = dnlc_lookup(dvp, nm);
1184 
1185                         if (tvp) {
1186                                 error = zfs_fastaccesschk_execute(zdp, cr);
1187                                 if (error) {
1188                                         VN_RELE(tvp);
1189                                         return (error);
1190                                 }
1191                                 if (tvp == DNLC_NO_VNODE) {
1192                                         VN_RELE(tvp);
1193                                         return (SET_ERROR(ENOENT));
1194                                 } else {
1195                                         *vpp = tvp;
1196                                         return (specvp_check(vpp, cr));
1197                                 }
1198                         }
1199                 }
1200         }
1201 
1202         DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1203 
1204         ZFS_ENTER(zfsvfs);
1205         ZFS_VERIFY_ZP(zdp);
1206 
1207         *vpp = NULL;
1208 
1209         if (flags & LOOKUP_XATTR) {
1210                 /*
1211                  * If the xattr property is off, refuse the lookup request.
1212                  */
1213                 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1214                         ZFS_EXIT(zfsvfs);
1215                         return (SET_ERROR(EINVAL));
1216                 }
1217 
1218                 /*
1219                  * We don't allow recursive attributes..
1220                  * Maybe someday we will.
1221                  */
1222                 if (zdp->z_pflags & ZFS_XATTR) {
1223                         ZFS_EXIT(zfsvfs);
1224                         return (SET_ERROR(EINVAL));
1225                 }
1226 
1227                 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1228                         ZFS_EXIT(zfsvfs);
1229                         return (error);
1230                 }
1231 
1232                 /*
1233                  * Do we have permission to get into attribute directory?
1234                  */
1235 
1236                 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1237                     B_FALSE, cr)) {
1238                         VN_RELE(*vpp);
1239                         *vpp = NULL;
1240                 }
1241 
1242                 ZFS_EXIT(zfsvfs);
1243                 return (error);
1244         }
1245 
1246         if (dvp->v_type != VDIR) {
1247                 ZFS_EXIT(zfsvfs);
1248                 return (SET_ERROR(ENOTDIR));
1249         }
1250 
1251         /*
1252          * Check accessibility of directory.
1253          */
1254 
1255         if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1256                 ZFS_EXIT(zfsvfs);
1257                 return (error);
1258         }
1259 
1260         if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1261             NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1262                 ZFS_EXIT(zfsvfs);
1263                 return (SET_ERROR(EILSEQ));
1264         }
1265 
1266         error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1267         if (error == 0)
1268                 error = specvp_check(vpp, cr);
1269 
1270         ZFS_EXIT(zfsvfs);
1271         return (error);
1272 }
1273 
1274 /*
1275  * Attempt to create a new entry in a directory.  If the entry
1276  * already exists, truncate the file if permissible, else return
1277  * an error.  Return the vp of the created or trunc'd file.
1278  *
1279  *      IN:     dvp     - vnode of directory to put new file entry in.
1280  *              name    - name of new file entry.
1281  *              vap     - attributes of new file.
1282  *              excl    - flag indicating exclusive or non-exclusive mode.
1283  *              mode    - mode to open file with.
1284  *              cr      - credentials of caller.
1285  *              flag    - large file flag [UNUSED].
1286  *              ct      - caller context
1287  *              vsecp   - ACL to be set
1288  *
1289  *      OUT:    vpp     - vnode of created or trunc'd entry.
1290  *
1291  *      RETURN: 0 on success, error code on failure.
1292  *
1293  * Timestamps:
1294  *      dvp - ctime|mtime updated if new entry created
1295  *       vp - ctime|mtime always, atime if new
1296  */
1297 
1298 /* ARGSUSED */
1299 static int
1300 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1301     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1302     vsecattr_t *vsecp)
1303 {
1304         znode_t         *zp, *dzp = VTOZ(dvp);
1305         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1306         zilog_t         *zilog;
1307         objset_t        *os;
1308         zfs_dirlock_t   *dl;
1309         dmu_tx_t        *tx;
1310         int             error;
1311         ksid_t          *ksid;
1312         uid_t           uid;
1313         gid_t           gid = crgetgid(cr);
1314         zfs_acl_ids_t   acl_ids;
1315         boolean_t       fuid_dirtied;
1316         boolean_t       have_acl = B_FALSE;
1317 
1318         /*
1319          * If we have an ephemeral id, ACL, or XVATTR then
1320          * make sure file system is at proper version
1321          */
1322 
1323         ksid = crgetsid(cr, KSID_OWNER);
1324         if (ksid)
1325                 uid = ksid_getid(ksid);
1326         else
1327                 uid = crgetuid(cr);
1328 
1329         if (zfsvfs->z_use_fuids == B_FALSE &&
1330             (vsecp || (vap->va_mask & AT_XVATTR) ||
1331             IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1332                 return (SET_ERROR(EINVAL));
1333 
1334         ZFS_ENTER(zfsvfs);
1335         ZFS_VERIFY_ZP(dzp);
1336         os = zfsvfs->z_os;
1337         zilog = zfsvfs->z_log;
1338 
1339         if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1340             NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1341                 ZFS_EXIT(zfsvfs);
1342                 return (SET_ERROR(EILSEQ));
1343         }
1344 
1345         if (vap->va_mask & AT_XVATTR) {
1346                 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1347                     crgetuid(cr), cr, vap->va_type)) != 0) {
1348                         ZFS_EXIT(zfsvfs);
1349                         return (error);
1350                 }
1351         }
1352 top:
1353         *vpp = NULL;
1354 
1355         if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1356                 vap->va_mode &= ~VSVTX;
1357 
1358         if (*name == '\0') {
1359                 /*
1360                  * Null component name refers to the directory itself.
1361                  */
1362                 VN_HOLD(dvp);
1363                 zp = dzp;
1364                 dl = NULL;
1365                 error = 0;
1366         } else {
1367                 /* possible VN_HOLD(zp) */
1368                 int zflg = 0;
1369 
1370                 if (flag & FIGNORECASE)
1371                         zflg |= ZCILOOK;
1372 
1373                 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1374                     NULL, NULL);
1375                 if (error) {
1376                         if (have_acl)
1377                                 zfs_acl_ids_free(&acl_ids);
1378                         if (strcmp(name, "..") == 0)
1379                                 error = SET_ERROR(EISDIR);
1380                         ZFS_EXIT(zfsvfs);
1381                         return (error);
1382                 }
1383         }
1384 
1385         if (zp == NULL) {
1386                 uint64_t txtype;
1387 
1388                 /*
1389                  * Create a new file object and update the directory
1390                  * to reference it.
1391                  */
1392                 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1393                         if (have_acl)
1394                                 zfs_acl_ids_free(&acl_ids);
1395                         goto out;
1396                 }
1397 
1398                 /*
1399                  * We only support the creation of regular files in
1400                  * extended attribute directories.
1401                  */
1402 
1403                 if ((dzp->z_pflags & ZFS_XATTR) &&
1404                     (vap->va_type != VREG)) {
1405                         if (have_acl)
1406                                 zfs_acl_ids_free(&acl_ids);
1407                         error = SET_ERROR(EINVAL);
1408                         goto out;
1409                 }
1410 
1411                 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1412                     cr, vsecp, &acl_ids)) != 0)
1413                         goto out;
1414                 have_acl = B_TRUE;
1415 
1416                 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1417                         zfs_acl_ids_free(&acl_ids);
1418                         error = SET_ERROR(EDQUOT);
1419                         goto out;
1420                 }
1421 
1422                 tx = dmu_tx_create(os);
1423 
1424                 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1425                     ZFS_SA_BASE_ATTR_SIZE);
1426 
1427                 fuid_dirtied = zfsvfs->z_fuid_dirty;
1428                 if (fuid_dirtied)
1429                         zfs_fuid_txhold(zfsvfs, tx);
1430                 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1431                 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1432                 if (!zfsvfs->z_use_sa &&
1433                     acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1434                         dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1435                             0, acl_ids.z_aclp->z_acl_bytes);
1436                 }
1437                 error = dmu_tx_assign(tx, TXG_NOWAIT);
1438                 if (error) {
1439                         zfs_dirent_unlock(dl);
1440                         if (error == ERESTART) {
1441                                 dmu_tx_wait(tx);
1442                                 dmu_tx_abort(tx);
1443                                 goto top;
1444                         }
1445                         zfs_acl_ids_free(&acl_ids);
1446                         dmu_tx_abort(tx);
1447                         ZFS_EXIT(zfsvfs);
1448                         return (error);
1449                 }
1450                 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1451 
1452                 if (fuid_dirtied)
1453                         zfs_fuid_sync(zfsvfs, tx);
1454 
1455                 (void) zfs_link_create(dl, zp, tx, ZNEW);
1456                 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1457                 if (flag & FIGNORECASE)
1458                         txtype |= TX_CI;
1459                 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1460                     vsecp, acl_ids.z_fuidp, vap);
1461                 zfs_acl_ids_free(&acl_ids);
1462                 dmu_tx_commit(tx);
1463         } else {
1464                 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1465 
1466                 if (have_acl)
1467                         zfs_acl_ids_free(&acl_ids);
1468                 have_acl = B_FALSE;
1469 
1470                 /*
1471                  * A directory entry already exists for this name.
1472                  */
1473                 /*
1474                  * Can't truncate an existing file if in exclusive mode.
1475                  */
1476                 if (excl == EXCL) {
1477                         error = SET_ERROR(EEXIST);
1478                         goto out;
1479                 }
1480                 /*
1481                  * Can't open a directory for writing.
1482                  */
1483                 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1484                         error = SET_ERROR(EISDIR);
1485                         goto out;
1486                 }
1487                 /*
1488                  * Verify requested access to file.
1489                  */
1490                 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1491                         goto out;
1492                 }
1493 
1494                 mutex_enter(&dzp->z_lock);
1495                 dzp->z_seq++;
1496                 mutex_exit(&dzp->z_lock);
1497 
1498                 /*
1499                  * Truncate regular files if requested.
1500                  */
1501                 if ((ZTOV(zp)->v_type == VREG) &&
1502                     (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1503                         /* we can't hold any locks when calling zfs_freesp() */
1504                         zfs_dirent_unlock(dl);
1505                         dl = NULL;
1506                         error = zfs_freesp(zp, 0, 0, mode, TRUE);
1507                         if (error == 0) {
1508                                 vnevent_create(ZTOV(zp), ct);
1509                         }
1510                 }
1511         }
1512 out:
1513 
1514         if (dl)
1515                 zfs_dirent_unlock(dl);
1516 
1517         if (error) {
1518                 if (zp)
1519                         VN_RELE(ZTOV(zp));
1520         } else {
1521                 *vpp = ZTOV(zp);
1522                 error = specvp_check(vpp, cr);
1523         }
1524 
1525         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1526                 zil_commit(zilog, 0);
1527 
1528         ZFS_EXIT(zfsvfs);
1529         return (error);
1530 }
1531 
1532 /*
1533  * Remove an entry from a directory.
1534  *
1535  *      IN:     dvp     - vnode of directory to remove entry from.
1536  *              name    - name of entry to remove.
1537  *              cr      - credentials of caller.
1538  *              ct      - caller context
1539  *              flags   - case flags
1540  *
1541  *      RETURN: 0 on success, error code on failure.
1542  *
1543  * Timestamps:
1544  *      dvp - ctime|mtime
1545  *       vp - ctime (if nlink > 0)
1546  */
1547 
1548 uint64_t null_xattr = 0;
1549 
1550 /*ARGSUSED*/
1551 static int
1552 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1553     int flags)
1554 {
1555         znode_t         *zp, *dzp = VTOZ(dvp);
1556         znode_t         *xzp;
1557         vnode_t         *vp;
1558         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1559         zilog_t         *zilog;
1560         uint64_t        acl_obj, xattr_obj;
1561         uint64_t        xattr_obj_unlinked = 0;
1562         uint64_t        obj = 0;
1563         zfs_dirlock_t   *dl;
1564         dmu_tx_t        *tx;
1565         boolean_t       may_delete_now, delete_now = FALSE;
1566         boolean_t       unlinked, toobig = FALSE;
1567         uint64_t        txtype;
1568         pathname_t      *realnmp = NULL;
1569         pathname_t      realnm;
1570         int             error;
1571         int             zflg = ZEXISTS;
1572 
1573         ZFS_ENTER(zfsvfs);
1574         ZFS_VERIFY_ZP(dzp);
1575         zilog = zfsvfs->z_log;
1576 
1577         if (flags & FIGNORECASE) {
1578                 zflg |= ZCILOOK;
1579                 pn_alloc(&realnm);
1580                 realnmp = &realnm;
1581         }
1582 
1583 top:
1584         xattr_obj = 0;
1585         xzp = NULL;
1586         /*
1587          * Attempt to lock directory; fail if entry doesn't exist.
1588          */
1589         if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1590             NULL, realnmp)) {
1591                 if (realnmp)
1592                         pn_free(realnmp);
1593                 ZFS_EXIT(zfsvfs);
1594                 return (error);
1595         }
1596 
1597         vp = ZTOV(zp);
1598 
1599         if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1600                 goto out;
1601         }
1602 
1603         /*
1604          * Need to use rmdir for removing directories.
1605          */
1606         if (vp->v_type == VDIR) {
1607                 error = SET_ERROR(EPERM);
1608                 goto out;
1609         }
1610 
1611         vnevent_remove(vp, dvp, name, ct);
1612 
1613         if (realnmp)
1614                 dnlc_remove(dvp, realnmp->pn_buf);
1615         else
1616                 dnlc_remove(dvp, name);
1617 
1618         mutex_enter(&vp->v_lock);
1619         may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1620         mutex_exit(&vp->v_lock);
1621 
1622         /*
1623          * We may delete the znode now, or we may put it in the unlinked set;
1624          * it depends on whether we're the last link, and on whether there are
1625          * other holds on the vnode.  So we dmu_tx_hold() the right things to
1626          * allow for either case.
1627          */
1628         obj = zp->z_id;
1629         tx = dmu_tx_create(zfsvfs->z_os);
1630         dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1631         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1632         zfs_sa_upgrade_txholds(tx, zp);
1633         zfs_sa_upgrade_txholds(tx, dzp);
1634         if (may_delete_now) {
1635                 toobig =
1636                     zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1637                 /* if the file is too big, only hold_free a token amount */
1638                 dmu_tx_hold_free(tx, zp->z_id, 0,
1639                     (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1640         }
1641 
1642         /* are there any extended attributes? */
1643         error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1644             &xattr_obj, sizeof (xattr_obj));
1645         if (error == 0 && xattr_obj) {
1646                 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1647                 ASSERT0(error);
1648                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1649                 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1650         }
1651 
1652         mutex_enter(&zp->z_lock);
1653         if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1654                 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1655         mutex_exit(&zp->z_lock);
1656 
1657         /* charge as an update -- would be nice not to charge at all */
1658         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1659 
1660         error = dmu_tx_assign(tx, TXG_NOWAIT);
1661         if (error) {
1662                 zfs_dirent_unlock(dl);
1663                 VN_RELE(vp);
1664                 if (xzp)
1665                         VN_RELE(ZTOV(xzp));
1666                 if (error == ERESTART) {
1667                         dmu_tx_wait(tx);
1668                         dmu_tx_abort(tx);
1669                         goto top;
1670                 }
1671                 if (realnmp)
1672                         pn_free(realnmp);
1673                 dmu_tx_abort(tx);
1674                 ZFS_EXIT(zfsvfs);
1675                 return (error);
1676         }
1677 
1678         /*
1679          * Remove the directory entry.
1680          */
1681         error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1682 
1683         if (error) {
1684                 dmu_tx_commit(tx);
1685                 goto out;
1686         }
1687 
1688         if (unlinked) {
1689 
1690                 /*
1691                  * Hold z_lock so that we can make sure that the ACL obj
1692                  * hasn't changed.  Could have been deleted due to
1693                  * zfs_sa_upgrade().
1694                  */
1695                 mutex_enter(&zp->z_lock);
1696                 mutex_enter(&vp->v_lock);
1697                 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1698                     &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1699                 delete_now = may_delete_now && !toobig &&
1700                     vp->v_count == 1 && !vn_has_cached_data(vp) &&
1701                     xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1702                     acl_obj;
1703                 mutex_exit(&vp->v_lock);
1704         }
1705 
1706         if (delete_now) {
1707                 if (xattr_obj_unlinked) {
1708                         ASSERT3U(xzp->z_links, ==, 2);
1709                         mutex_enter(&xzp->z_lock);
1710                         xzp->z_unlinked = 1;
1711                         xzp->z_links = 0;
1712                         error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1713                             &xzp->z_links, sizeof (xzp->z_links), tx);
1714                         ASSERT3U(error,  ==,  0);
1715                         mutex_exit(&xzp->z_lock);
1716                         zfs_unlinked_add(xzp, tx);
1717 
1718                         if (zp->z_is_sa)
1719                                 error = sa_remove(zp->z_sa_hdl,
1720                                     SA_ZPL_XATTR(zfsvfs), tx);
1721                         else
1722                                 error = sa_update(zp->z_sa_hdl,
1723                                     SA_ZPL_XATTR(zfsvfs), &null_xattr,
1724                                     sizeof (uint64_t), tx);
1725                         ASSERT0(error);
1726                 }
1727                 mutex_enter(&vp->v_lock);
1728                 vp->v_count--;
1729                 ASSERT0(vp->v_count);
1730                 mutex_exit(&vp->v_lock);
1731                 mutex_exit(&zp->z_lock);
1732                 zfs_znode_delete(zp, tx);
1733         } else if (unlinked) {
1734                 mutex_exit(&zp->z_lock);
1735                 zfs_unlinked_add(zp, tx);
1736         }
1737 
1738         txtype = TX_REMOVE;
1739         if (flags & FIGNORECASE)
1740                 txtype |= TX_CI;
1741         zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1742 
1743         dmu_tx_commit(tx);
1744 out:
1745         if (realnmp)
1746                 pn_free(realnmp);
1747 
1748         zfs_dirent_unlock(dl);
1749 
1750         if (!delete_now)
1751                 VN_RELE(vp);
1752         if (xzp)
1753                 VN_RELE(ZTOV(xzp));
1754 
1755         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1756                 zil_commit(zilog, 0);
1757 
1758         ZFS_EXIT(zfsvfs);
1759         return (error);
1760 }
1761 
1762 /*
1763  * Create a new directory and insert it into dvp using the name
1764  * provided.  Return a pointer to the inserted directory.
1765  *
1766  *      IN:     dvp     - vnode of directory to add subdir to.
1767  *              dirname - name of new directory.
1768  *              vap     - attributes of new directory.
1769  *              cr      - credentials of caller.
1770  *              ct      - caller context
1771  *              flags   - case flags
1772  *              vsecp   - ACL to be set
1773  *
1774  *      OUT:    vpp     - vnode of created directory.
1775  *
1776  *      RETURN: 0 on success, error code on failure.
1777  *
1778  * Timestamps:
1779  *      dvp - ctime|mtime updated
1780  *       vp - ctime|mtime|atime updated
1781  */
1782 /*ARGSUSED*/
1783 static int
1784 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1785     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1786 {
1787         znode_t         *zp, *dzp = VTOZ(dvp);
1788         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1789         zilog_t         *zilog;
1790         zfs_dirlock_t   *dl;
1791         uint64_t        txtype;
1792         dmu_tx_t        *tx;
1793         int             error;
1794         int             zf = ZNEW;
1795         ksid_t          *ksid;
1796         uid_t           uid;
1797         gid_t           gid = crgetgid(cr);
1798         zfs_acl_ids_t   acl_ids;
1799         boolean_t       fuid_dirtied;
1800 
1801         ASSERT(vap->va_type == VDIR);
1802 
1803         /*
1804          * If we have an ephemeral id, ACL, or XVATTR then
1805          * make sure file system is at proper version
1806          */
1807 
1808         ksid = crgetsid(cr, KSID_OWNER);
1809         if (ksid)
1810                 uid = ksid_getid(ksid);
1811         else
1812                 uid = crgetuid(cr);
1813         if (zfsvfs->z_use_fuids == B_FALSE &&
1814             (vsecp || (vap->va_mask & AT_XVATTR) ||
1815             IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1816                 return (SET_ERROR(EINVAL));
1817 
1818         ZFS_ENTER(zfsvfs);
1819         ZFS_VERIFY_ZP(dzp);
1820         zilog = zfsvfs->z_log;
1821 
1822         if (dzp->z_pflags & ZFS_XATTR) {
1823                 ZFS_EXIT(zfsvfs);
1824                 return (SET_ERROR(EINVAL));
1825         }
1826 
1827         if (zfsvfs->z_utf8 && u8_validate(dirname,
1828             strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1829                 ZFS_EXIT(zfsvfs);
1830                 return (SET_ERROR(EILSEQ));
1831         }
1832         if (flags & FIGNORECASE)
1833                 zf |= ZCILOOK;
1834 
1835         if (vap->va_mask & AT_XVATTR) {
1836                 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1837                     crgetuid(cr), cr, vap->va_type)) != 0) {
1838                         ZFS_EXIT(zfsvfs);
1839                         return (error);
1840                 }
1841         }
1842 
1843         if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1844             vsecp, &acl_ids)) != 0) {
1845                 ZFS_EXIT(zfsvfs);
1846                 return (error);
1847         }
1848         /*
1849          * First make sure the new directory doesn't exist.
1850          *
1851          * Existence is checked first to make sure we don't return
1852          * EACCES instead of EEXIST which can cause some applications
1853          * to fail.
1854          */
1855 top:
1856         *vpp = NULL;
1857 
1858         if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1859             NULL, NULL)) {
1860                 zfs_acl_ids_free(&acl_ids);
1861                 ZFS_EXIT(zfsvfs);
1862                 return (error);
1863         }
1864 
1865         if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1866                 zfs_acl_ids_free(&acl_ids);
1867                 zfs_dirent_unlock(dl);
1868                 ZFS_EXIT(zfsvfs);
1869                 return (error);
1870         }
1871 
1872         if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1873                 zfs_acl_ids_free(&acl_ids);
1874                 zfs_dirent_unlock(dl);
1875                 ZFS_EXIT(zfsvfs);
1876                 return (SET_ERROR(EDQUOT));
1877         }
1878 
1879         /*
1880          * Add a new entry to the directory.
1881          */
1882         tx = dmu_tx_create(zfsvfs->z_os);
1883         dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1884         dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1885         fuid_dirtied = zfsvfs->z_fuid_dirty;
1886         if (fuid_dirtied)
1887                 zfs_fuid_txhold(zfsvfs, tx);
1888         if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1889                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1890                     acl_ids.z_aclp->z_acl_bytes);
1891         }
1892 
1893         dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1894             ZFS_SA_BASE_ATTR_SIZE);
1895 
1896         error = dmu_tx_assign(tx, TXG_NOWAIT);
1897         if (error) {
1898                 zfs_dirent_unlock(dl);
1899                 if (error == ERESTART) {
1900                         dmu_tx_wait(tx);
1901                         dmu_tx_abort(tx);
1902                         goto top;
1903                 }
1904                 zfs_acl_ids_free(&acl_ids);
1905                 dmu_tx_abort(tx);
1906                 ZFS_EXIT(zfsvfs);
1907                 return (error);
1908         }
1909 
1910         /*
1911          * Create new node.
1912          */
1913         zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1914 
1915         if (fuid_dirtied)
1916                 zfs_fuid_sync(zfsvfs, tx);
1917 
1918         /*
1919          * Now put new name in parent dir.
1920          */
1921         (void) zfs_link_create(dl, zp, tx, ZNEW);
1922 
1923         *vpp = ZTOV(zp);
1924 
1925         txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1926         if (flags & FIGNORECASE)
1927                 txtype |= TX_CI;
1928         zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1929             acl_ids.z_fuidp, vap);
1930 
1931         zfs_acl_ids_free(&acl_ids);
1932 
1933         dmu_tx_commit(tx);
1934 
1935         zfs_dirent_unlock(dl);
1936 
1937         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1938                 zil_commit(zilog, 0);
1939 
1940         ZFS_EXIT(zfsvfs);
1941         return (0);
1942 }
1943 
1944 /*
1945  * Remove a directory subdir entry.  If the current working
1946  * directory is the same as the subdir to be removed, the
1947  * remove will fail.
1948  *
1949  *      IN:     dvp     - vnode of directory to remove from.
1950  *              name    - name of directory to be removed.
1951  *              cwd     - vnode of current working directory.
1952  *              cr      - credentials of caller.
1953  *              ct      - caller context
1954  *              flags   - case flags
1955  *
1956  *      RETURN: 0 on success, error code on failure.
1957  *
1958  * Timestamps:
1959  *      dvp - ctime|mtime updated
1960  */
1961 /*ARGSUSED*/
1962 static int
1963 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1964     caller_context_t *ct, int flags)
1965 {
1966         znode_t         *dzp = VTOZ(dvp);
1967         znode_t         *zp;
1968         vnode_t         *vp;
1969         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1970         zilog_t         *zilog;
1971         zfs_dirlock_t   *dl;
1972         dmu_tx_t        *tx;
1973         int             error;
1974         int             zflg = ZEXISTS;
1975 
1976         ZFS_ENTER(zfsvfs);
1977         ZFS_VERIFY_ZP(dzp);
1978         zilog = zfsvfs->z_log;
1979 
1980         if (flags & FIGNORECASE)
1981                 zflg |= ZCILOOK;
1982 top:
1983         zp = NULL;
1984 
1985         /*
1986          * Attempt to lock directory; fail if entry doesn't exist.
1987          */
1988         if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1989             NULL, NULL)) {
1990                 ZFS_EXIT(zfsvfs);
1991                 return (error);
1992         }
1993 
1994         vp = ZTOV(zp);
1995 
1996         if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1997                 goto out;
1998         }
1999 
2000         if (vp->v_type != VDIR) {
2001                 error = SET_ERROR(ENOTDIR);
2002                 goto out;
2003         }
2004 
2005         if (vp == cwd) {
2006                 error = SET_ERROR(EINVAL);
2007                 goto out;
2008         }
2009 
2010         vnevent_rmdir(vp, dvp, name, ct);
2011 
2012         /*
2013          * Grab a lock on the directory to make sure that noone is
2014          * trying to add (or lookup) entries while we are removing it.
2015          */
2016         rw_enter(&zp->z_name_lock, RW_WRITER);
2017 
2018         /*
2019          * Grab a lock on the parent pointer to make sure we play well
2020          * with the treewalk and directory rename code.
2021          */
2022         rw_enter(&zp->z_parent_lock, RW_WRITER);
2023 
2024         tx = dmu_tx_create(zfsvfs->z_os);
2025         dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2026         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2027         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2028         zfs_sa_upgrade_txholds(tx, zp);
2029         zfs_sa_upgrade_txholds(tx, dzp);
2030         error = dmu_tx_assign(tx, TXG_NOWAIT);
2031         if (error) {
2032                 rw_exit(&zp->z_parent_lock);
2033                 rw_exit(&zp->z_name_lock);
2034                 zfs_dirent_unlock(dl);
2035                 VN_RELE(vp);
2036                 if (error == ERESTART) {
2037                         dmu_tx_wait(tx);
2038                         dmu_tx_abort(tx);
2039                         goto top;
2040                 }
2041                 dmu_tx_abort(tx);
2042                 ZFS_EXIT(zfsvfs);
2043                 return (error);
2044         }
2045 
2046         error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2047 
2048         if (error == 0) {
2049                 uint64_t txtype = TX_RMDIR;
2050                 if (flags & FIGNORECASE)
2051                         txtype |= TX_CI;
2052                 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2053         }
2054 
2055         dmu_tx_commit(tx);
2056 
2057         rw_exit(&zp->z_parent_lock);
2058         rw_exit(&zp->z_name_lock);
2059 out:
2060         zfs_dirent_unlock(dl);
2061 
2062         VN_RELE(vp);
2063 
2064         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2065                 zil_commit(zilog, 0);
2066 
2067         ZFS_EXIT(zfsvfs);
2068         return (error);
2069 }
2070 
2071 /*
2072  * Read as many directory entries as will fit into the provided
2073  * buffer from the given directory cursor position (specified in
2074  * the uio structure).
2075  *
2076  *      IN:     vp      - vnode of directory to read.
2077  *              uio     - structure supplying read location, range info,
2078  *                        and return buffer.
2079  *              cr      - credentials of caller.
2080  *              ct      - caller context
2081  *              flags   - case flags
2082  *
2083  *      OUT:    uio     - updated offset and range, buffer filled.
2084  *              eofp    - set to true if end-of-file detected.
2085  *
2086  *      RETURN: 0 on success, error code on failure.
2087  *
2088  * Timestamps:
2089  *      vp - atime updated
2090  *
2091  * Note that the low 4 bits of the cookie returned by zap is always zero.
2092  * This allows us to use the low range for "special" directory entries:
2093  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2094  * we use the offset 2 for the '.zfs' directory.
2095  */
2096 /* ARGSUSED */
2097 static int
2098 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2099     caller_context_t *ct, int flags)
2100 {
2101         znode_t         *zp = VTOZ(vp);
2102         iovec_t         *iovp;
2103         edirent_t       *eodp;
2104         dirent64_t      *odp;
2105         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
2106         objset_t        *os;
2107         caddr_t         outbuf;
2108         size_t          bufsize;
2109         zap_cursor_t    zc;
2110         zap_attribute_t zap;
2111         uint_t          bytes_wanted;
2112         uint64_t        offset; /* must be unsigned; checks for < 1 */
2113         uint64_t        parent;
2114         int             local_eof;
2115         int             outcount;
2116         int             error;
2117         uint8_t         prefetch;
2118         boolean_t       check_sysattrs;
2119 
2120         ZFS_ENTER(zfsvfs);
2121         ZFS_VERIFY_ZP(zp);
2122 
2123         if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2124             &parent, sizeof (parent))) != 0) {
2125                 ZFS_EXIT(zfsvfs);
2126                 return (error);
2127         }
2128 
2129         /*
2130          * If we are not given an eof variable,
2131          * use a local one.
2132          */
2133         if (eofp == NULL)
2134                 eofp = &local_eof;
2135 
2136         /*
2137          * Check for valid iov_len.
2138          */
2139         if (uio->uio_iov->iov_len <= 0) {
2140                 ZFS_EXIT(zfsvfs);
2141                 return (SET_ERROR(EINVAL));
2142         }
2143 
2144         /*
2145          * Quit if directory has been removed (posix)
2146          */
2147         if ((*eofp = zp->z_unlinked) != 0) {
2148                 ZFS_EXIT(zfsvfs);
2149                 return (0);
2150         }
2151 
2152         error = 0;
2153         os = zfsvfs->z_os;
2154         offset = uio->uio_loffset;
2155         prefetch = zp->z_zn_prefetch;
2156 
2157         /*
2158          * Initialize the iterator cursor.
2159          */
2160         if (offset <= 3) {
2161                 /*
2162                  * Start iteration from the beginning of the directory.
2163                  */
2164                 zap_cursor_init(&zc, os, zp->z_id);
2165         } else {
2166                 /*
2167                  * The offset is a serialized cursor.
2168                  */
2169                 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2170         }
2171 
2172         /*
2173          * Get space to change directory entries into fs independent format.
2174          */
2175         iovp = uio->uio_iov;
2176         bytes_wanted = iovp->iov_len;
2177         if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2178                 bufsize = bytes_wanted;
2179                 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2180                 odp = (struct dirent64 *)outbuf;
2181         } else {
2182                 bufsize = bytes_wanted;
2183                 outbuf = NULL;
2184                 odp = (struct dirent64 *)iovp->iov_base;
2185         }
2186         eodp = (struct edirent *)odp;
2187 
2188         /*
2189          * If this VFS supports the system attribute view interface; and
2190          * we're looking at an extended attribute directory; and we care
2191          * about normalization conflicts on this vfs; then we must check
2192          * for normalization conflicts with the sysattr name space.
2193          */
2194         check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2195             (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2196             (flags & V_RDDIR_ENTFLAGS);
2197 
2198         /*
2199          * Transform to file-system independent format
2200          */
2201         outcount = 0;
2202         while (outcount < bytes_wanted) {
2203                 ino64_t objnum;
2204                 ushort_t reclen;
2205                 off64_t *next = NULL;
2206 
2207                 /*
2208                  * Special case `.', `..', and `.zfs'.
2209                  */
2210                 if (offset == 0) {
2211                         (void) strcpy(zap.za_name, ".");
2212                         zap.za_normalization_conflict = 0;
2213                         objnum = zp->z_id;
2214                 } else if (offset == 1) {
2215                         (void) strcpy(zap.za_name, "..");
2216                         zap.za_normalization_conflict = 0;
2217                         objnum = parent;
2218                 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2219                         (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2220                         zap.za_normalization_conflict = 0;
2221                         objnum = ZFSCTL_INO_ROOT;
2222                 } else {
2223                         /*
2224                          * Grab next entry.
2225                          */
2226                         if (error = zap_cursor_retrieve(&zc, &zap)) {
2227                                 if ((*eofp = (error == ENOENT)) != 0)
2228                                         break;
2229                                 else
2230                                         goto update;
2231                         }
2232 
2233                         if (zap.za_integer_length != 8 ||
2234                             zap.za_num_integers != 1) {
2235                                 cmn_err(CE_WARN, "zap_readdir: bad directory "
2236                                     "entry, obj = %lld, offset = %lld\n",
2237                                     (u_longlong_t)zp->z_id,
2238                                     (u_longlong_t)offset);
2239                                 error = SET_ERROR(ENXIO);
2240                                 goto update;
2241                         }
2242 
2243                         objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2244                         /*
2245                          * MacOS X can extract the object type here such as:
2246                          * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2247                          */
2248 
2249                         if (check_sysattrs && !zap.za_normalization_conflict) {
2250                                 zap.za_normalization_conflict =
2251                                     xattr_sysattr_casechk(zap.za_name);
2252                         }
2253                 }
2254 
2255                 if (flags & V_RDDIR_ACCFILTER) {
2256                         /*
2257                          * If we have no access at all, don't include
2258                          * this entry in the returned information
2259                          */
2260                         znode_t *ezp;
2261                         if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2262                                 goto skip_entry;
2263                         if (!zfs_has_access(ezp, cr)) {
2264                                 VN_RELE(ZTOV(ezp));
2265                                 goto skip_entry;
2266                         }
2267                         VN_RELE(ZTOV(ezp));
2268                 }
2269 
2270                 if (flags & V_RDDIR_ENTFLAGS)
2271                         reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2272                 else
2273                         reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2274 
2275                 /*
2276                  * Will this entry fit in the buffer?
2277                  */
2278                 if (outcount + reclen > bufsize) {
2279                         /*
2280                          * Did we manage to fit anything in the buffer?
2281                          */
2282                         if (!outcount) {
2283                                 error = SET_ERROR(EINVAL);
2284                                 goto update;
2285                         }
2286                         break;
2287                 }
2288                 if (flags & V_RDDIR_ENTFLAGS) {
2289                         /*
2290                          * Add extended flag entry:
2291                          */
2292                         eodp->ed_ino = objnum;
2293                         eodp->ed_reclen = reclen;
2294                         /* NOTE: ed_off is the offset for the *next* entry */
2295                         next = &(eodp->ed_off);
2296                         eodp->ed_eflags = zap.za_normalization_conflict ?
2297                             ED_CASE_CONFLICT : 0;
2298                         (void) strncpy(eodp->ed_name, zap.za_name,
2299                             EDIRENT_NAMELEN(reclen));
2300                         eodp = (edirent_t *)((intptr_t)eodp + reclen);
2301                 } else {
2302                         /*
2303                          * Add normal entry:
2304                          */
2305                         odp->d_ino = objnum;
2306                         odp->d_reclen = reclen;
2307                         /* NOTE: d_off is the offset for the *next* entry */
2308                         next = &(odp->d_off);
2309                         (void) strncpy(odp->d_name, zap.za_name,
2310                             DIRENT64_NAMELEN(reclen));
2311                         odp = (dirent64_t *)((intptr_t)odp + reclen);
2312                 }
2313                 outcount += reclen;
2314 
2315                 ASSERT(outcount <= bufsize);
2316 
2317                 /* Prefetch znode */
2318                 if (prefetch)
2319                         dmu_prefetch(os, objnum, 0, 0);
2320 
2321         skip_entry:
2322                 /*
2323                  * Move to the next entry, fill in the previous offset.
2324                  */
2325                 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2326                         zap_cursor_advance(&zc);
2327                         offset = zap_cursor_serialize(&zc);
2328                 } else {
2329                         offset += 1;
2330                 }
2331                 if (next)
2332                         *next = offset;
2333         }
2334         zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2335 
2336         if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2337                 iovp->iov_base += outcount;
2338                 iovp->iov_len -= outcount;
2339                 uio->uio_resid -= outcount;
2340         } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2341                 /*
2342                  * Reset the pointer.
2343                  */
2344                 offset = uio->uio_loffset;
2345         }
2346 
2347 update:
2348         zap_cursor_fini(&zc);
2349         if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2350                 kmem_free(outbuf, bufsize);
2351 
2352         if (error == ENOENT)
2353                 error = 0;
2354 
2355         ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2356 
2357         uio->uio_loffset = offset;
2358         ZFS_EXIT(zfsvfs);
2359         return (error);
2360 }
2361 
2362 ulong_t zfs_fsync_sync_cnt = 4;
2363 
2364 static int
2365 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2366 {
2367         znode_t *zp = VTOZ(vp);
2368         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2369 
2370         /*
2371          * Regardless of whether this is required for standards conformance,
2372          * this is the logical behavior when fsync() is called on a file with
2373          * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2374          * going to be pushed out as part of the zil_commit().
2375          */
2376         if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2377             (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2378                 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2379 
2380         (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2381 
2382         if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2383                 ZFS_ENTER(zfsvfs);
2384                 ZFS_VERIFY_ZP(zp);
2385                 zil_commit(zfsvfs->z_log, zp->z_id);
2386                 ZFS_EXIT(zfsvfs);
2387         }
2388         return (0);
2389 }
2390 
2391 
2392 /*
2393  * Get the requested file attributes and place them in the provided
2394  * vattr structure.
2395  *
2396  *      IN:     vp      - vnode of file.
2397  *              vap     - va_mask identifies requested attributes.
2398  *                        If AT_XVATTR set, then optional attrs are requested
2399  *              flags   - ATTR_NOACLCHECK (CIFS server context)
2400  *              cr      - credentials of caller.
2401  *              ct      - caller context
2402  *
2403  *      OUT:    vap     - attribute values.
2404  *
2405  *      RETURN: 0 (always succeeds).
2406  */
2407 /* ARGSUSED */
2408 static int
2409 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2410     caller_context_t *ct)
2411 {
2412         znode_t *zp = VTOZ(vp);
2413         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2414         int     error = 0;
2415         uint64_t links;
2416         uint64_t mtime[2], ctime[2];
2417         xvattr_t *xvap = (xvattr_t *)vap;       /* vap may be an xvattr_t * */
2418         xoptattr_t *xoap = NULL;
2419         boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2420         sa_bulk_attr_t bulk[2];
2421         int count = 0;
2422 
2423         ZFS_ENTER(zfsvfs);
2424         ZFS_VERIFY_ZP(zp);
2425 
2426         zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2427 
2428         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2429         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2430 
2431         if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2432                 ZFS_EXIT(zfsvfs);
2433                 return (error);
2434         }
2435 
2436         /*
2437          * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2438          * Also, if we are the owner don't bother, since owner should
2439          * always be allowed to read basic attributes of file.
2440          */
2441         if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2442             (vap->va_uid != crgetuid(cr))) {
2443                 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2444                     skipaclchk, cr)) {
2445                         ZFS_EXIT(zfsvfs);
2446                         return (error);
2447                 }
2448         }
2449 
2450         /*
2451          * Return all attributes.  It's cheaper to provide the answer
2452          * than to determine whether we were asked the question.
2453          */
2454 
2455         mutex_enter(&zp->z_lock);
2456         vap->va_type = vp->v_type;
2457         vap->va_mode = zp->z_mode & MODEMASK;
2458         vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2459         vap->va_nodeid = zp->z_id;
2460         if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2461                 links = zp->z_links + 1;
2462         else
2463                 links = zp->z_links;
2464         vap->va_nlink = MIN(links, UINT32_MAX);      /* nlink_t limit! */
2465         vap->va_size = zp->z_size;
2466         vap->va_rdev = vp->v_rdev;
2467         vap->va_seq = zp->z_seq;
2468 
2469         /*
2470          * Add in any requested optional attributes and the create time.
2471          * Also set the corresponding bits in the returned attribute bitmap.
2472          */
2473         if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2474                 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2475                         xoap->xoa_archive =
2476                             ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2477                         XVA_SET_RTN(xvap, XAT_ARCHIVE);
2478                 }
2479 
2480                 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2481                         xoap->xoa_readonly =
2482                             ((zp->z_pflags & ZFS_READONLY) != 0);
2483                         XVA_SET_RTN(xvap, XAT_READONLY);
2484                 }
2485 
2486                 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2487                         xoap->xoa_system =
2488                             ((zp->z_pflags & ZFS_SYSTEM) != 0);
2489                         XVA_SET_RTN(xvap, XAT_SYSTEM);
2490                 }
2491 
2492                 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2493                         xoap->xoa_hidden =
2494                             ((zp->z_pflags & ZFS_HIDDEN) != 0);
2495                         XVA_SET_RTN(xvap, XAT_HIDDEN);
2496                 }
2497 
2498                 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2499                         xoap->xoa_nounlink =
2500                             ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2501                         XVA_SET_RTN(xvap, XAT_NOUNLINK);
2502                 }
2503 
2504                 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2505                         xoap->xoa_immutable =
2506                             ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2507                         XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2508                 }
2509 
2510                 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2511                         xoap->xoa_appendonly =
2512                             ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2513                         XVA_SET_RTN(xvap, XAT_APPENDONLY);
2514                 }
2515 
2516                 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2517                         xoap->xoa_nodump =
2518                             ((zp->z_pflags & ZFS_NODUMP) != 0);
2519                         XVA_SET_RTN(xvap, XAT_NODUMP);
2520                 }
2521 
2522                 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2523                         xoap->xoa_opaque =
2524                             ((zp->z_pflags & ZFS_OPAQUE) != 0);
2525                         XVA_SET_RTN(xvap, XAT_OPAQUE);
2526                 }
2527 
2528                 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2529                         xoap->xoa_av_quarantined =
2530                             ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2531                         XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2532                 }
2533 
2534                 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2535                         xoap->xoa_av_modified =
2536                             ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2537                         XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2538                 }
2539 
2540                 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2541                     vp->v_type == VREG) {
2542                         zfs_sa_get_scanstamp(zp, xvap);
2543                 }
2544 
2545                 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2546                         uint64_t times[2];
2547 
2548                         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2549                             times, sizeof (times));
2550                         ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2551                         XVA_SET_RTN(xvap, XAT_CREATETIME);
2552                 }
2553 
2554                 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2555                         xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2556                         XVA_SET_RTN(xvap, XAT_REPARSE);
2557                 }
2558                 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2559                         xoap->xoa_generation = zp->z_gen;
2560                         XVA_SET_RTN(xvap, XAT_GEN);
2561                 }
2562 
2563                 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2564                         xoap->xoa_offline =
2565                             ((zp->z_pflags & ZFS_OFFLINE) != 0);
2566                         XVA_SET_RTN(xvap, XAT_OFFLINE);
2567                 }
2568 
2569                 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2570                         xoap->xoa_sparse =
2571                             ((zp->z_pflags & ZFS_SPARSE) != 0);
2572                         XVA_SET_RTN(xvap, XAT_SPARSE);
2573                 }
2574         }
2575 
2576         ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2577         ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2578         ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2579 
2580         mutex_exit(&zp->z_lock);
2581 
2582         sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2583 
2584         if (zp->z_blksz == 0) {
2585                 /*
2586                  * Block size hasn't been set; suggest maximal I/O transfers.
2587                  */
2588                 vap->va_blksize = zfsvfs->z_max_blksz;
2589         }
2590 
2591         ZFS_EXIT(zfsvfs);
2592         return (0);
2593 }
2594 
2595 /*
2596  * Set the file attributes to the values contained in the
2597  * vattr structure.
2598  *
2599  *      IN:     vp      - vnode of file to be modified.
2600  *              vap     - new attribute values.
2601  *                        If AT_XVATTR set, then optional attrs are being set
2602  *              flags   - ATTR_UTIME set if non-default time values provided.
2603  *                      - ATTR_NOACLCHECK (CIFS context only).
2604  *              cr      - credentials of caller.
2605  *              ct      - caller context
2606  *
2607  *      RETURN: 0 on success, error code on failure.
2608  *
2609  * Timestamps:
2610  *      vp - ctime updated, mtime updated if size changed.
2611  */
2612 /* ARGSUSED */
2613 static int
2614 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2615     caller_context_t *ct)
2616 {
2617         znode_t         *zp = VTOZ(vp);
2618         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
2619         zilog_t         *zilog;
2620         dmu_tx_t        *tx;
2621         vattr_t         oldva;
2622         xvattr_t        tmpxvattr;
2623         uint_t          mask = vap->va_mask;
2624         uint_t          saved_mask = 0;
2625         int             trim_mask = 0;
2626         uint64_t        new_mode;
2627         uint64_t        new_uid, new_gid;
2628         uint64_t        xattr_obj;
2629         uint64_t        mtime[2], ctime[2];
2630         znode_t         *attrzp;
2631         int             need_policy = FALSE;
2632         int             err, err2;
2633         zfs_fuid_info_t *fuidp = NULL;
2634         xvattr_t *xvap = (xvattr_t *)vap;       /* vap may be an xvattr_t * */
2635         xoptattr_t      *xoap;
2636         zfs_acl_t       *aclp;
2637         boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2638         boolean_t       fuid_dirtied = B_FALSE;
2639         sa_bulk_attr_t  bulk[7], xattr_bulk[7];
2640         int             count = 0, xattr_count = 0;
2641 
2642         if (mask == 0)
2643                 return (0);
2644 
2645         if (mask & AT_NOSET)
2646                 return (SET_ERROR(EINVAL));
2647 
2648         ZFS_ENTER(zfsvfs);
2649         ZFS_VERIFY_ZP(zp);
2650 
2651         zilog = zfsvfs->z_log;
2652 
2653         /*
2654          * Make sure that if we have ephemeral uid/gid or xvattr specified
2655          * that file system is at proper version level
2656          */
2657 
2658         if (zfsvfs->z_use_fuids == B_FALSE &&
2659             (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2660             ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2661             (mask & AT_XVATTR))) {
2662                 ZFS_EXIT(zfsvfs);
2663                 return (SET_ERROR(EINVAL));
2664         }
2665 
2666         if (mask & AT_SIZE && vp->v_type == VDIR) {
2667                 ZFS_EXIT(zfsvfs);
2668                 return (SET_ERROR(EISDIR));
2669         }
2670 
2671         if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2672                 ZFS_EXIT(zfsvfs);
2673                 return (SET_ERROR(EINVAL));
2674         }
2675 
2676         /*
2677          * If this is an xvattr_t, then get a pointer to the structure of
2678          * optional attributes.  If this is NULL, then we have a vattr_t.
2679          */
2680         xoap = xva_getxoptattr(xvap);
2681 
2682         xva_init(&tmpxvattr);
2683 
2684         /*
2685          * Immutable files can only alter immutable bit and atime
2686          */
2687         if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2688             ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2689             ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2690                 ZFS_EXIT(zfsvfs);
2691                 return (SET_ERROR(EPERM));
2692         }
2693 
2694         if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2695                 ZFS_EXIT(zfsvfs);
2696                 return (SET_ERROR(EPERM));
2697         }
2698 
2699         /*
2700          * Verify timestamps doesn't overflow 32 bits.
2701          * ZFS can handle large timestamps, but 32bit syscalls can't
2702          * handle times greater than 2039.  This check should be removed
2703          * once large timestamps are fully supported.
2704          */
2705         if (mask & (AT_ATIME | AT_MTIME)) {
2706                 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2707                     ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2708                         ZFS_EXIT(zfsvfs);
2709                         return (SET_ERROR(EOVERFLOW));
2710                 }
2711         }
2712 
2713 top:
2714         attrzp = NULL;
2715         aclp = NULL;
2716 
2717         /* Can this be moved to before the top label? */
2718         if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2719                 ZFS_EXIT(zfsvfs);
2720                 return (SET_ERROR(EROFS));
2721         }
2722 
2723         /*
2724          * First validate permissions
2725          */
2726 
2727         if (mask & AT_SIZE) {
2728                 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2729                 if (err) {
2730                         ZFS_EXIT(zfsvfs);
2731                         return (err);
2732                 }
2733                 /*
2734                  * XXX - Note, we are not providing any open
2735                  * mode flags here (like FNDELAY), so we may
2736                  * block if there are locks present... this
2737                  * should be addressed in openat().
2738                  */
2739                 /* XXX - would it be OK to generate a log record here? */
2740                 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2741                 if (err) {
2742                         ZFS_EXIT(zfsvfs);
2743                         return (err);
2744                 }
2745         }
2746 
2747         if (mask & (AT_ATIME|AT_MTIME) ||
2748             ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2749             XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2750             XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2751             XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2752             XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2753             XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2754             XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2755                 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2756                     skipaclchk, cr);
2757         }
2758 
2759         if (mask & (AT_UID|AT_GID)) {
2760                 int     idmask = (mask & (AT_UID|AT_GID));
2761                 int     take_owner;
2762                 int     take_group;
2763 
2764                 /*
2765                  * NOTE: even if a new mode is being set,
2766                  * we may clear S_ISUID/S_ISGID bits.
2767                  */
2768 
2769                 if (!(mask & AT_MODE))
2770                         vap->va_mode = zp->z_mode;
2771 
2772                 /*
2773                  * Take ownership or chgrp to group we are a member of
2774                  */
2775 
2776                 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2777                 take_group = (mask & AT_GID) &&
2778                     zfs_groupmember(zfsvfs, vap->va_gid, cr);
2779 
2780                 /*
2781                  * If both AT_UID and AT_GID are set then take_owner and
2782                  * take_group must both be set in order to allow taking
2783                  * ownership.
2784                  *
2785                  * Otherwise, send the check through secpolicy_vnode_setattr()
2786                  *
2787                  */
2788 
2789                 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2790                     ((idmask == AT_UID) && take_owner) ||
2791                     ((idmask == AT_GID) && take_group)) {
2792                         if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2793                             skipaclchk, cr) == 0) {
2794                                 /*
2795                                  * Remove setuid/setgid for non-privileged users
2796                                  */
2797                                 secpolicy_setid_clear(vap, cr);
2798                                 trim_mask = (mask & (AT_UID|AT_GID));
2799                         } else {
2800                                 need_policy =  TRUE;
2801                         }
2802                 } else {
2803                         need_policy =  TRUE;
2804                 }
2805         }
2806 
2807         mutex_enter(&zp->z_lock);
2808         oldva.va_mode = zp->z_mode;
2809         zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2810         if (mask & AT_XVATTR) {
2811                 /*
2812                  * Update xvattr mask to include only those attributes
2813                  * that are actually changing.
2814                  *
2815                  * the bits will be restored prior to actually setting
2816                  * the attributes so the caller thinks they were set.
2817                  */
2818                 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2819                         if (xoap->xoa_appendonly !=
2820                             ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2821                                 need_policy = TRUE;
2822                         } else {
2823                                 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2824                                 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2825                         }
2826                 }
2827 
2828                 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2829                         if (xoap->xoa_nounlink !=
2830                             ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2831                                 need_policy = TRUE;
2832                         } else {
2833                                 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2834                                 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2835                         }
2836                 }
2837 
2838                 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2839                         if (xoap->xoa_immutable !=
2840                             ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2841                                 need_policy = TRUE;
2842                         } else {
2843                                 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2844                                 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2845                         }
2846                 }
2847 
2848                 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2849                         if (xoap->xoa_nodump !=
2850                             ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2851                                 need_policy = TRUE;
2852                         } else {
2853                                 XVA_CLR_REQ(xvap, XAT_NODUMP);
2854                                 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2855                         }
2856                 }
2857 
2858                 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2859                         if (xoap->xoa_av_modified !=
2860                             ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2861                                 need_policy = TRUE;
2862                         } else {
2863                                 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2864                                 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2865                         }
2866                 }
2867 
2868                 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2869                         if ((vp->v_type != VREG &&
2870                             xoap->xoa_av_quarantined) ||
2871                             xoap->xoa_av_quarantined !=
2872                             ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2873                                 need_policy = TRUE;
2874                         } else {
2875                                 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2876                                 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2877                         }
2878                 }
2879 
2880                 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2881                         mutex_exit(&zp->z_lock);
2882                         ZFS_EXIT(zfsvfs);
2883                         return (SET_ERROR(EPERM));
2884                 }
2885 
2886                 if (need_policy == FALSE &&
2887                     (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2888                     XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2889                         need_policy = TRUE;
2890                 }
2891         }
2892 
2893         mutex_exit(&zp->z_lock);
2894 
2895         if (mask & AT_MODE) {
2896                 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2897                         err = secpolicy_setid_setsticky_clear(vp, vap,
2898                             &oldva, cr);
2899                         if (err) {
2900                                 ZFS_EXIT(zfsvfs);
2901                                 return (err);
2902                         }
2903                         trim_mask |= AT_MODE;
2904                 } else {
2905                         need_policy = TRUE;
2906                 }
2907         }
2908 
2909         if (need_policy) {
2910                 /*
2911                  * If trim_mask is set then take ownership
2912                  * has been granted or write_acl is present and user
2913                  * has the ability to modify mode.  In that case remove
2914                  * UID|GID and or MODE from mask so that
2915                  * secpolicy_vnode_setattr() doesn't revoke it.
2916                  */
2917 
2918                 if (trim_mask) {
2919                         saved_mask = vap->va_mask;
2920                         vap->va_mask &= ~trim_mask;
2921                 }
2922                 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2923                     (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2924                 if (err) {
2925                         ZFS_EXIT(zfsvfs);
2926                         return (err);
2927                 }
2928 
2929                 if (trim_mask)
2930                         vap->va_mask |= saved_mask;
2931         }
2932 
2933         /*
2934          * secpolicy_vnode_setattr, or take ownership may have
2935          * changed va_mask
2936          */
2937         mask = vap->va_mask;
2938 
2939         if ((mask & (AT_UID | AT_GID))) {
2940                 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2941                     &xattr_obj, sizeof (xattr_obj));
2942 
2943                 if (err == 0 && xattr_obj) {
2944                         err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2945                         if (err)
2946                                 goto out2;
2947                 }
2948                 if (mask & AT_UID) {
2949                         new_uid = zfs_fuid_create(zfsvfs,
2950                             (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2951                         if (new_uid != zp->z_uid &&
2952                             zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
2953                                 if (attrzp)
2954                                         VN_RELE(ZTOV(attrzp));
2955                                 err = SET_ERROR(EDQUOT);
2956                                 goto out2;
2957                         }
2958                 }
2959 
2960                 if (mask & AT_GID) {
2961                         new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2962                             cr, ZFS_GROUP, &fuidp);
2963                         if (new_gid != zp->z_gid &&
2964                             zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
2965                                 if (attrzp)
2966                                         VN_RELE(ZTOV(attrzp));
2967                                 err = SET_ERROR(EDQUOT);
2968                                 goto out2;
2969                         }
2970                 }
2971         }
2972         tx = dmu_tx_create(zfsvfs->z_os);
2973 
2974         if (mask & AT_MODE) {
2975                 uint64_t pmode = zp->z_mode;
2976                 uint64_t acl_obj;
2977                 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2978 
2979                 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
2980                     !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
2981                         err = SET_ERROR(EPERM);
2982                         goto out;
2983                 }
2984 
2985                 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
2986                         goto out;
2987 
2988                 mutex_enter(&zp->z_lock);
2989                 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2990                         /*
2991                          * Are we upgrading ACL from old V0 format
2992                          * to V1 format?
2993                          */
2994                         if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2995                             zfs_znode_acl_version(zp) ==
2996                             ZFS_ACL_VERSION_INITIAL) {
2997                                 dmu_tx_hold_free(tx, acl_obj, 0,
2998                                     DMU_OBJECT_END);
2999                                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3000                                     0, aclp->z_acl_bytes);
3001                         } else {
3002                                 dmu_tx_hold_write(tx, acl_obj, 0,
3003                                     aclp->z_acl_bytes);
3004                         }
3005                 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3006                         dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3007                             0, aclp->z_acl_bytes);
3008                 }
3009                 mutex_exit(&zp->z_lock);
3010                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3011         } else {
3012                 if ((mask & AT_XVATTR) &&
3013                     XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3014                         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3015                 else
3016                         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3017         }
3018 
3019         if (attrzp) {
3020                 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3021         }
3022 
3023         fuid_dirtied = zfsvfs->z_fuid_dirty;
3024         if (fuid_dirtied)
3025                 zfs_fuid_txhold(zfsvfs, tx);
3026 
3027         zfs_sa_upgrade_txholds(tx, zp);
3028 
3029         err = dmu_tx_assign(tx, TXG_NOWAIT);
3030         if (err) {
3031                 if (err == ERESTART)
3032                         dmu_tx_wait(tx);
3033                 goto out;
3034         }
3035 
3036         count = 0;
3037         /*
3038          * Set each attribute requested.
3039          * We group settings according to the locks they need to acquire.
3040          *
3041          * Note: you cannot set ctime directly, although it will be
3042          * updated as a side-effect of calling this function.
3043          */
3044 
3045 
3046         if (mask & (AT_UID|AT_GID|AT_MODE))
3047                 mutex_enter(&zp->z_acl_lock);
3048         mutex_enter(&zp->z_lock);
3049 
3050         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3051             &zp->z_pflags, sizeof (zp->z_pflags));
3052 
3053         if (attrzp) {
3054                 if (mask & (AT_UID|AT_GID|AT_MODE))
3055                         mutex_enter(&attrzp->z_acl_lock);
3056                 mutex_enter(&attrzp->z_lock);
3057                 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3058                     SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3059                     sizeof (attrzp->z_pflags));
3060         }
3061 
3062         if (mask & (AT_UID|AT_GID)) {
3063 
3064                 if (mask & AT_UID) {
3065                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3066                             &new_uid, sizeof (new_uid));
3067                         zp->z_uid = new_uid;
3068                         if (attrzp) {
3069                                 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3070                                     SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3071                                     sizeof (new_uid));
3072                                 attrzp->z_uid = new_uid;
3073                         }
3074                 }
3075 
3076                 if (mask & AT_GID) {
3077                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3078                             NULL, &new_gid, sizeof (new_gid));
3079                         zp->z_gid = new_gid;
3080                         if (attrzp) {
3081                                 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3082                                     SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3083                                     sizeof (new_gid));
3084                                 attrzp->z_gid = new_gid;
3085                         }
3086                 }
3087                 if (!(mask & AT_MODE)) {
3088                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3089                             NULL, &new_mode, sizeof (new_mode));
3090                         new_mode = zp->z_mode;
3091                 }
3092                 err = zfs_acl_chown_setattr(zp);
3093                 ASSERT(err == 0);
3094                 if (attrzp) {
3095                         err = zfs_acl_chown_setattr(attrzp);
3096                         ASSERT(err == 0);
3097                 }
3098         }
3099 
3100         if (mask & AT_MODE) {
3101                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3102                     &new_mode, sizeof (new_mode));
3103                 zp->z_mode = new_mode;
3104                 ASSERT3U((uintptr_t)aclp, !=, NULL);
3105                 err = zfs_aclset_common(zp, aclp, cr, tx);
3106                 ASSERT0(err);
3107                 if (zp->z_acl_cached)
3108                         zfs_acl_free(zp->z_acl_cached);
3109                 zp->z_acl_cached = aclp;
3110                 aclp = NULL;
3111         }
3112 
3113 
3114         if (mask & AT_ATIME) {
3115                 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3116                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3117                     &zp->z_atime, sizeof (zp->z_atime));
3118         }
3119 
3120         if (mask & AT_MTIME) {
3121                 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3122                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3123                     mtime, sizeof (mtime));
3124         }
3125 
3126         /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3127         if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3128                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3129                     NULL, mtime, sizeof (mtime));
3130                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3131                     &ctime, sizeof (ctime));
3132                 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3133                     B_TRUE);
3134         } else if (mask != 0) {
3135                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3136                     &ctime, sizeof (ctime));
3137                 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3138                     B_TRUE);
3139                 if (attrzp) {
3140                         SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3141                             SA_ZPL_CTIME(zfsvfs), NULL,
3142                             &ctime, sizeof (ctime));
3143                         zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3144                             mtime, ctime, B_TRUE);
3145                 }
3146         }
3147         /*
3148          * Do this after setting timestamps to prevent timestamp
3149          * update from toggling bit
3150          */
3151 
3152         if (xoap && (mask & AT_XVATTR)) {
3153 
3154                 /*
3155                  * restore trimmed off masks
3156                  * so that return masks can be set for caller.
3157                  */
3158 
3159                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3160                         XVA_SET_REQ(xvap, XAT_APPENDONLY);
3161                 }
3162                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3163                         XVA_SET_REQ(xvap, XAT_NOUNLINK);
3164                 }
3165                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3166                         XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3167                 }
3168                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3169                         XVA_SET_REQ(xvap, XAT_NODUMP);
3170                 }
3171                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3172                         XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3173                 }
3174                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3175                         XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3176                 }
3177 
3178                 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3179                         ASSERT(vp->v_type == VREG);
3180 
3181                 zfs_xvattr_set(zp, xvap, tx);
3182         }
3183 
3184         if (fuid_dirtied)
3185                 zfs_fuid_sync(zfsvfs, tx);
3186 
3187         if (mask != 0)
3188                 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3189 
3190         mutex_exit(&zp->z_lock);
3191         if (mask & (AT_UID|AT_GID|AT_MODE))
3192                 mutex_exit(&zp->z_acl_lock);
3193 
3194         if (attrzp) {
3195                 if (mask & (AT_UID|AT_GID|AT_MODE))
3196                         mutex_exit(&attrzp->z_acl_lock);
3197                 mutex_exit(&attrzp->z_lock);
3198         }
3199 out:
3200         if (err == 0 && attrzp) {
3201                 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3202                     xattr_count, tx);
3203                 ASSERT(err2 == 0);
3204         }
3205 
3206         if (attrzp)
3207                 VN_RELE(ZTOV(attrzp));
3208 
3209         if (aclp)
3210                 zfs_acl_free(aclp);
3211 
3212         if (fuidp) {
3213                 zfs_fuid_info_free(fuidp);
3214                 fuidp = NULL;
3215         }
3216 
3217         if (err) {
3218                 dmu_tx_abort(tx);
3219                 if (err == ERESTART)
3220                         goto top;
3221         } else {
3222                 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3223                 dmu_tx_commit(tx);
3224         }
3225 
3226 out2:
3227         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3228                 zil_commit(zilog, 0);
3229 
3230         ZFS_EXIT(zfsvfs);
3231         return (err);
3232 }
3233 
3234 typedef struct zfs_zlock {
3235         krwlock_t       *zl_rwlock;     /* lock we acquired */
3236         znode_t         *zl_znode;      /* znode we held */
3237         struct zfs_zlock *zl_next;      /* next in list */
3238 } zfs_zlock_t;
3239 
3240 /*
3241  * Drop locks and release vnodes that were held by zfs_rename_lock().
3242  */
3243 static void
3244 zfs_rename_unlock(zfs_zlock_t **zlpp)
3245 {
3246         zfs_zlock_t *zl;
3247 
3248         while ((zl = *zlpp) != NULL) {
3249                 if (zl->zl_znode != NULL)
3250                         VN_RELE(ZTOV(zl->zl_znode));
3251                 rw_exit(zl->zl_rwlock);
3252                 *zlpp = zl->zl_next;
3253                 kmem_free(zl, sizeof (*zl));
3254         }
3255 }
3256 
3257 /*
3258  * Search back through the directory tree, using the ".." entries.
3259  * Lock each directory in the chain to prevent concurrent renames.
3260  * Fail any attempt to move a directory into one of its own descendants.
3261  * XXX - z_parent_lock can overlap with map or grow locks
3262  */
3263 static int
3264 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3265 {
3266         zfs_zlock_t     *zl;
3267         znode_t         *zp = tdzp;
3268         uint64_t        rootid = zp->z_zfsvfs->z_root;
3269         uint64_t        oidp = zp->z_id;
3270         krwlock_t       *rwlp = &szp->z_parent_lock;
3271         krw_t           rw = RW_WRITER;
3272 
3273         /*
3274          * First pass write-locks szp and compares to zp->z_id.
3275          * Later passes read-lock zp and compare to zp->z_parent.
3276          */
3277         do {
3278                 if (!rw_tryenter(rwlp, rw)) {
3279                         /*
3280                          * Another thread is renaming in this path.
3281                          * Note that if we are a WRITER, we don't have any
3282                          * parent_locks held yet.
3283                          */
3284                         if (rw == RW_READER && zp->z_id > szp->z_id) {
3285                                 /*
3286                                  * Drop our locks and restart
3287                                  */
3288                                 zfs_rename_unlock(&zl);
3289                                 *zlpp = NULL;
3290                                 zp = tdzp;
3291                                 oidp = zp->z_id;
3292                                 rwlp = &szp->z_parent_lock;
3293                                 rw = RW_WRITER;
3294                                 continue;
3295                         } else {
3296                                 /*
3297                                  * Wait for other thread to drop its locks
3298                                  */
3299                                 rw_enter(rwlp, rw);
3300                         }
3301                 }
3302 
3303                 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3304                 zl->zl_rwlock = rwlp;
3305                 zl->zl_znode = NULL;
3306                 zl->zl_next = *zlpp;
3307                 *zlpp = zl;
3308 
3309                 if (oidp == szp->z_id)               /* We're a descendant of szp */
3310                         return (SET_ERROR(EINVAL));
3311 
3312                 if (oidp == rootid)             /* We've hit the top */
3313                         return (0);
3314 
3315                 if (rw == RW_READER) {          /* i.e. not the first pass */
3316                         int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3317                         if (error)
3318                                 return (error);
3319                         zl->zl_znode = zp;
3320                 }
3321                 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3322                     &oidp, sizeof (oidp));
3323                 rwlp = &zp->z_parent_lock;
3324                 rw = RW_READER;
3325 
3326         } while (zp->z_id != sdzp->z_id);
3327 
3328         return (0);
3329 }
3330 
3331 /*
3332  * Move an entry from the provided source directory to the target
3333  * directory.  Change the entry name as indicated.
3334  *
3335  *      IN:     sdvp    - Source directory containing the "old entry".
3336  *              snm     - Old entry name.
3337  *              tdvp    - Target directory to contain the "new entry".
3338  *              tnm     - New entry name.
3339  *              cr      - credentials of caller.
3340  *              ct      - caller context
3341  *              flags   - case flags
3342  *
3343  *      RETURN: 0 on success, error code on failure.
3344  *
3345  * Timestamps:
3346  *      sdvp,tdvp - ctime|mtime updated
3347  */
3348 /*ARGSUSED*/
3349 static int
3350 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3351     caller_context_t *ct, int flags)
3352 {
3353         znode_t         *tdzp, *szp, *tzp;
3354         znode_t         *sdzp = VTOZ(sdvp);
3355         zfsvfs_t        *zfsvfs = sdzp->z_zfsvfs;
3356         zilog_t         *zilog;
3357         vnode_t         *realvp;
3358         zfs_dirlock_t   *sdl, *tdl;
3359         dmu_tx_t        *tx;
3360         zfs_zlock_t     *zl;
3361         int             cmp, serr, terr;
3362         int             error = 0;
3363         int             zflg = 0;
3364 
3365         ZFS_ENTER(zfsvfs);
3366         ZFS_VERIFY_ZP(sdzp);
3367         zilog = zfsvfs->z_log;
3368 
3369         /*
3370          * Make sure we have the real vp for the target directory.
3371          */
3372         if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3373                 tdvp = realvp;
3374 
3375         if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3376                 ZFS_EXIT(zfsvfs);
3377                 return (SET_ERROR(EXDEV));
3378         }
3379 
3380         tdzp = VTOZ(tdvp);
3381         ZFS_VERIFY_ZP(tdzp);
3382         if (zfsvfs->z_utf8 && u8_validate(tnm,
3383             strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3384                 ZFS_EXIT(zfsvfs);
3385                 return (SET_ERROR(EILSEQ));
3386         }
3387 
3388         if (flags & FIGNORECASE)
3389                 zflg |= ZCILOOK;
3390 
3391 top:
3392         szp = NULL;
3393         tzp = NULL;
3394         zl = NULL;
3395 
3396         /*
3397          * This is to prevent the creation of links into attribute space
3398          * by renaming a linked file into/outof an attribute directory.
3399          * See the comment in zfs_link() for why this is considered bad.
3400          */
3401         if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3402                 ZFS_EXIT(zfsvfs);
3403                 return (SET_ERROR(EINVAL));
3404         }
3405 
3406         /*
3407          * Lock source and target directory entries.  To prevent deadlock,
3408          * a lock ordering must be defined.  We lock the directory with
3409          * the smallest object id first, or if it's a tie, the one with
3410          * the lexically first name.
3411          */
3412         if (sdzp->z_id < tdzp->z_id) {
3413                 cmp = -1;
3414         } else if (sdzp->z_id > tdzp->z_id) {
3415                 cmp = 1;
3416         } else {
3417                 /*
3418                  * First compare the two name arguments without
3419                  * considering any case folding.
3420                  */
3421                 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3422 
3423                 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3424                 ASSERT(error == 0 || !zfsvfs->z_utf8);
3425                 if (cmp == 0) {
3426                         /*
3427                          * POSIX: "If the old argument and the new argument
3428                          * both refer to links to the same existing file,
3429                          * the rename() function shall return successfully
3430                          * and perform no other action."
3431                          */
3432                         ZFS_EXIT(zfsvfs);
3433                         return (0);
3434                 }
3435                 /*
3436                  * If the file system is case-folding, then we may
3437                  * have some more checking to do.  A case-folding file
3438                  * system is either supporting mixed case sensitivity
3439                  * access or is completely case-insensitive.  Note
3440                  * that the file system is always case preserving.
3441                  *
3442                  * In mixed sensitivity mode case sensitive behavior
3443                  * is the default.  FIGNORECASE must be used to
3444                  * explicitly request case insensitive behavior.
3445                  *
3446                  * If the source and target names provided differ only
3447                  * by case (e.g., a request to rename 'tim' to 'Tim'),
3448                  * we will treat this as a special case in the
3449                  * case-insensitive mode: as long as the source name
3450                  * is an exact match, we will allow this to proceed as
3451                  * a name-change request.
3452                  */
3453                 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3454                     (zfsvfs->z_case == ZFS_CASE_MIXED &&
3455                     flags & FIGNORECASE)) &&
3456                     u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3457                     &error) == 0) {
3458                         /*
3459                          * case preserving rename request, require exact
3460                          * name matches
3461                          */
3462                         zflg |= ZCIEXACT;
3463                         zflg &= ~ZCILOOK;
3464                 }
3465         }
3466 
3467         /*
3468          * If the source and destination directories are the same, we should
3469          * grab the z_name_lock of that directory only once.
3470          */
3471         if (sdzp == tdzp) {
3472                 zflg |= ZHAVELOCK;
3473                 rw_enter(&sdzp->z_name_lock, RW_READER);
3474         }
3475 
3476         if (cmp < 0) {
3477                 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3478                     ZEXISTS | zflg, NULL, NULL);
3479                 terr = zfs_dirent_lock(&tdl,
3480                     tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3481         } else {
3482                 terr = zfs_dirent_lock(&tdl,
3483                     tdzp, tnm, &tzp, zflg, NULL, NULL);
3484                 serr = zfs_dirent_lock(&sdl,
3485                     sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3486                     NULL, NULL);
3487         }
3488 
3489         if (serr) {
3490                 /*
3491                  * Source entry invalid or not there.
3492                  */
3493                 if (!terr) {
3494                         zfs_dirent_unlock(tdl);
3495                         if (tzp)
3496                                 VN_RELE(ZTOV(tzp));
3497                 }
3498 
3499                 if (sdzp == tdzp)
3500                         rw_exit(&sdzp->z_name_lock);
3501 
3502                 if (strcmp(snm, "..") == 0)
3503                         serr = SET_ERROR(EINVAL);
3504                 ZFS_EXIT(zfsvfs);
3505                 return (serr);
3506         }
3507         if (terr) {
3508                 zfs_dirent_unlock(sdl);
3509                 VN_RELE(ZTOV(szp));
3510 
3511                 if (sdzp == tdzp)
3512                         rw_exit(&sdzp->z_name_lock);
3513 
3514                 if (strcmp(tnm, "..") == 0)
3515                         terr = SET_ERROR(EINVAL);
3516                 ZFS_EXIT(zfsvfs);
3517                 return (terr);
3518         }
3519 
3520         /*
3521          * Must have write access at the source to remove the old entry
3522          * and write access at the target to create the new entry.
3523          * Note that if target and source are the same, this can be
3524          * done in a single check.
3525          */
3526 
3527         if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3528                 goto out;
3529 
3530         if (ZTOV(szp)->v_type == VDIR) {
3531                 /*
3532                  * Check to make sure rename is valid.
3533                  * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3534                  */
3535                 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3536                         goto out;
3537         }
3538 
3539         /*
3540          * Does target exist?
3541          */
3542         if (tzp) {
3543                 /*
3544                  * Source and target must be the same type.
3545                  */
3546                 if (ZTOV(szp)->v_type == VDIR) {
3547                         if (ZTOV(tzp)->v_type != VDIR) {
3548                                 error = SET_ERROR(ENOTDIR);
3549                                 goto out;
3550                         }
3551                 } else {
3552                         if (ZTOV(tzp)->v_type == VDIR) {
3553                                 error = SET_ERROR(EISDIR);
3554                                 goto out;
3555                         }
3556                 }
3557                 /*
3558                  * POSIX dictates that when the source and target
3559                  * entries refer to the same file object, rename
3560                  * must do nothing and exit without error.
3561                  */
3562                 if (szp->z_id == tzp->z_id) {
3563                         error = 0;
3564                         goto out;
3565                 }
3566         }
3567 
3568         vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3569         if (tzp)
3570                 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3571 
3572         /*
3573          * notify the target directory if it is not the same
3574          * as source directory.
3575          */
3576         if (tdvp != sdvp) {
3577                 vnevent_rename_dest_dir(tdvp, ct);
3578         }
3579 
3580         tx = dmu_tx_create(zfsvfs->z_os);
3581         dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3582         dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3583         dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3584         dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3585         if (sdzp != tdzp) {
3586                 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3587                 zfs_sa_upgrade_txholds(tx, tdzp);
3588         }
3589         if (tzp) {
3590                 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3591                 zfs_sa_upgrade_txholds(tx, tzp);
3592         }
3593 
3594         zfs_sa_upgrade_txholds(tx, szp);
3595         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3596         error = dmu_tx_assign(tx, TXG_NOWAIT);
3597         if (error) {
3598                 if (zl != NULL)
3599                         zfs_rename_unlock(&zl);
3600                 zfs_dirent_unlock(sdl);
3601                 zfs_dirent_unlock(tdl);
3602 
3603                 if (sdzp == tdzp)
3604                         rw_exit(&sdzp->z_name_lock);
3605 
3606                 VN_RELE(ZTOV(szp));
3607                 if (tzp)
3608                         VN_RELE(ZTOV(tzp));
3609                 if (error == ERESTART) {
3610                         dmu_tx_wait(tx);
3611                         dmu_tx_abort(tx);
3612                         goto top;
3613                 }
3614                 dmu_tx_abort(tx);
3615                 ZFS_EXIT(zfsvfs);
3616                 return (error);
3617         }
3618 
3619         if (tzp)        /* Attempt to remove the existing target */
3620                 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3621 
3622         if (error == 0) {
3623                 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3624                 if (error == 0) {
3625                         szp->z_pflags |= ZFS_AV_MODIFIED;
3626 
3627                         error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3628                             (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3629                         ASSERT0(error);
3630 
3631                         error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3632                         if (error == 0) {
3633                                 zfs_log_rename(zilog, tx, TX_RENAME |
3634                                     (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3635                                     sdl->dl_name, tdzp, tdl->dl_name, szp);
3636 
3637                                 /*
3638                                  * Update path information for the target vnode
3639                                  */
3640                                 vn_renamepath(tdvp, ZTOV(szp), tnm,
3641                                     strlen(tnm));
3642                         } else {
3643                                 /*
3644                                  * At this point, we have successfully created
3645                                  * the target name, but have failed to remove
3646                                  * the source name.  Since the create was done
3647                                  * with the ZRENAMING flag, there are
3648                                  * complications; for one, the link count is
3649                                  * wrong.  The easiest way to deal with this
3650                                  * is to remove the newly created target, and
3651                                  * return the original error.  This must
3652                                  * succeed; fortunately, it is very unlikely to
3653                                  * fail, since we just created it.
3654                                  */
3655                                 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3656                                     ZRENAMING, NULL), ==, 0);
3657                         }
3658                 }
3659         }
3660 
3661         dmu_tx_commit(tx);
3662 out:
3663         if (zl != NULL)
3664                 zfs_rename_unlock(&zl);
3665 
3666         zfs_dirent_unlock(sdl);
3667         zfs_dirent_unlock(tdl);
3668 
3669         if (sdzp == tdzp)
3670                 rw_exit(&sdzp->z_name_lock);
3671 
3672 
3673         VN_RELE(ZTOV(szp));
3674         if (tzp)
3675                 VN_RELE(ZTOV(tzp));
3676 
3677         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3678                 zil_commit(zilog, 0);
3679 
3680         ZFS_EXIT(zfsvfs);
3681         return (error);
3682 }
3683 
3684 /*
3685  * Insert the indicated symbolic reference entry into the directory.
3686  *
3687  *      IN:     dvp     - Directory to contain new symbolic link.
3688  *              link    - Name for new symlink entry.
3689  *              vap     - Attributes of new entry.
3690  *              cr      - credentials of caller.
3691  *              ct      - caller context
3692  *              flags   - case flags
3693  *
3694  *      RETURN: 0 on success, error code on failure.
3695  *
3696  * Timestamps:
3697  *      dvp - ctime|mtime updated
3698  */
3699 /*ARGSUSED*/
3700 static int
3701 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3702     caller_context_t *ct, int flags)
3703 {
3704         znode_t         *zp, *dzp = VTOZ(dvp);
3705         zfs_dirlock_t   *dl;
3706         dmu_tx_t        *tx;
3707         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
3708         zilog_t         *zilog;
3709         uint64_t        len = strlen(link);
3710         int             error;
3711         int             zflg = ZNEW;
3712         zfs_acl_ids_t   acl_ids;
3713         boolean_t       fuid_dirtied;
3714         uint64_t        txtype = TX_SYMLINK;
3715 
3716         ASSERT(vap->va_type == VLNK);
3717 
3718         ZFS_ENTER(zfsvfs);
3719         ZFS_VERIFY_ZP(dzp);
3720         zilog = zfsvfs->z_log;
3721 
3722         if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3723             NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3724                 ZFS_EXIT(zfsvfs);
3725                 return (SET_ERROR(EILSEQ));
3726         }
3727         if (flags & FIGNORECASE)
3728                 zflg |= ZCILOOK;
3729 
3730         if (len > MAXPATHLEN) {
3731                 ZFS_EXIT(zfsvfs);
3732                 return (SET_ERROR(ENAMETOOLONG));
3733         }
3734 
3735         if ((error = zfs_acl_ids_create(dzp, 0,
3736             vap, cr, NULL, &acl_ids)) != 0) {
3737                 ZFS_EXIT(zfsvfs);
3738                 return (error);
3739         }
3740 top:
3741         /*
3742          * Attempt to lock directory; fail if entry already exists.
3743          */
3744         error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3745         if (error) {
3746                 zfs_acl_ids_free(&acl_ids);
3747                 ZFS_EXIT(zfsvfs);
3748                 return (error);
3749         }
3750 
3751         if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3752                 zfs_acl_ids_free(&acl_ids);
3753                 zfs_dirent_unlock(dl);
3754                 ZFS_EXIT(zfsvfs);
3755                 return (error);
3756         }
3757 
3758         if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3759                 zfs_acl_ids_free(&acl_ids);
3760                 zfs_dirent_unlock(dl);
3761                 ZFS_EXIT(zfsvfs);
3762                 return (SET_ERROR(EDQUOT));
3763         }
3764         tx = dmu_tx_create(zfsvfs->z_os);
3765         fuid_dirtied = zfsvfs->z_fuid_dirty;
3766         dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3767         dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3768         dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3769             ZFS_SA_BASE_ATTR_SIZE + len);
3770         dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3771         if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3772                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3773                     acl_ids.z_aclp->z_acl_bytes);
3774         }
3775         if (fuid_dirtied)
3776                 zfs_fuid_txhold(zfsvfs, tx);
3777         error = dmu_tx_assign(tx, TXG_NOWAIT);
3778         if (error) {
3779                 zfs_dirent_unlock(dl);
3780                 if (error == ERESTART) {
3781                         dmu_tx_wait(tx);
3782                         dmu_tx_abort(tx);
3783                         goto top;
3784                 }
3785                 zfs_acl_ids_free(&acl_ids);
3786                 dmu_tx_abort(tx);
3787                 ZFS_EXIT(zfsvfs);
3788                 return (error);
3789         }
3790 
3791         /*
3792          * Create a new object for the symlink.
3793          * for version 4 ZPL datsets the symlink will be an SA attribute
3794          */
3795         zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3796 
3797         if (fuid_dirtied)
3798                 zfs_fuid_sync(zfsvfs, tx);
3799 
3800         mutex_enter(&zp->z_lock);
3801         if (zp->z_is_sa)
3802                 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3803                     link, len, tx);
3804         else
3805                 zfs_sa_symlink(zp, link, len, tx);
3806         mutex_exit(&zp->z_lock);
3807 
3808         zp->z_size = len;
3809         (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3810             &zp->z_size, sizeof (zp->z_size), tx);
3811         /*
3812          * Insert the new object into the directory.
3813          */
3814         (void) zfs_link_create(dl, zp, tx, ZNEW);
3815 
3816         if (flags & FIGNORECASE)
3817                 txtype |= TX_CI;
3818         zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3819 
3820         zfs_acl_ids_free(&acl_ids);
3821 
3822         dmu_tx_commit(tx);
3823 
3824         zfs_dirent_unlock(dl);
3825 
3826         VN_RELE(ZTOV(zp));
3827 
3828         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3829                 zil_commit(zilog, 0);
3830 
3831         ZFS_EXIT(zfsvfs);
3832         return (error);
3833 }
3834 
3835 /*
3836  * Return, in the buffer contained in the provided uio structure,
3837  * the symbolic path referred to by vp.
3838  *
3839  *      IN:     vp      - vnode of symbolic link.
3840  *              uio     - structure to contain the link path.
3841  *              cr      - credentials of caller.
3842  *              ct      - caller context
3843  *
3844  *      OUT:    uio     - structure containing the link path.
3845  *
3846  *      RETURN: 0 on success, error code on failure.
3847  *
3848  * Timestamps:
3849  *      vp - atime updated
3850  */
3851 /* ARGSUSED */
3852 static int
3853 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3854 {
3855         znode_t         *zp = VTOZ(vp);
3856         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
3857         int             error;
3858 
3859         ZFS_ENTER(zfsvfs);
3860         ZFS_VERIFY_ZP(zp);
3861 
3862         mutex_enter(&zp->z_lock);
3863         if (zp->z_is_sa)
3864                 error = sa_lookup_uio(zp->z_sa_hdl,
3865                     SA_ZPL_SYMLINK(zfsvfs), uio);
3866         else
3867                 error = zfs_sa_readlink(zp, uio);
3868         mutex_exit(&zp->z_lock);
3869 
3870         ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3871 
3872         ZFS_EXIT(zfsvfs);
3873         return (error);
3874 }
3875 
3876 /*
3877  * Insert a new entry into directory tdvp referencing svp.
3878  *
3879  *      IN:     tdvp    - Directory to contain new entry.
3880  *              svp     - vnode of new entry.
3881  *              name    - name of new entry.
3882  *              cr      - credentials of caller.
3883  *              ct      - caller context
3884  *
3885  *      RETURN: 0 on success, error code on failure.
3886  *
3887  * Timestamps:
3888  *      tdvp - ctime|mtime updated
3889  *       svp - ctime updated
3890  */
3891 /* ARGSUSED */
3892 static int
3893 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3894     caller_context_t *ct, int flags)
3895 {
3896         znode_t         *dzp = VTOZ(tdvp);
3897         znode_t         *tzp, *szp;
3898         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
3899         zilog_t         *zilog;
3900         zfs_dirlock_t   *dl;
3901         dmu_tx_t        *tx;
3902         vnode_t         *realvp;
3903         int             error;
3904         int             zf = ZNEW;
3905         uint64_t        parent;
3906         uid_t           owner;
3907 
3908         ASSERT(tdvp->v_type == VDIR);
3909 
3910         ZFS_ENTER(zfsvfs);
3911         ZFS_VERIFY_ZP(dzp);
3912         zilog = zfsvfs->z_log;
3913 
3914         if (VOP_REALVP(svp, &realvp, ct) == 0)
3915                 svp = realvp;
3916 
3917         /*
3918          * POSIX dictates that we return EPERM here.
3919          * Better choices include ENOTSUP or EISDIR.
3920          */
3921         if (svp->v_type == VDIR) {
3922                 ZFS_EXIT(zfsvfs);
3923                 return (SET_ERROR(EPERM));
3924         }
3925 
3926         if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
3927                 ZFS_EXIT(zfsvfs);
3928                 return (SET_ERROR(EXDEV));
3929         }
3930 
3931         szp = VTOZ(svp);
3932         ZFS_VERIFY_ZP(szp);
3933 
3934         /* Prevent links to .zfs/shares files */
3935 
3936         if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3937             &parent, sizeof (uint64_t))) != 0) {
3938                 ZFS_EXIT(zfsvfs);
3939                 return (error);
3940         }
3941         if (parent == zfsvfs->z_shares_dir) {
3942                 ZFS_EXIT(zfsvfs);
3943                 return (SET_ERROR(EPERM));
3944         }
3945 
3946         if (zfsvfs->z_utf8 && u8_validate(name,
3947             strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3948                 ZFS_EXIT(zfsvfs);
3949                 return (SET_ERROR(EILSEQ));
3950         }
3951         if (flags & FIGNORECASE)
3952                 zf |= ZCILOOK;
3953 
3954         /*
3955          * We do not support links between attributes and non-attributes
3956          * because of the potential security risk of creating links
3957          * into "normal" file space in order to circumvent restrictions
3958          * imposed in attribute space.
3959          */
3960         if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3961                 ZFS_EXIT(zfsvfs);
3962                 return (SET_ERROR(EINVAL));
3963         }
3964 
3965 
3966         owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
3967         if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3968                 ZFS_EXIT(zfsvfs);
3969                 return (SET_ERROR(EPERM));
3970         }
3971 
3972         if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3973                 ZFS_EXIT(zfsvfs);
3974                 return (error);
3975         }
3976 
3977 top:
3978         /*
3979          * Attempt to lock directory; fail if entry already exists.
3980          */
3981         error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3982         if (error) {
3983                 ZFS_EXIT(zfsvfs);
3984                 return (error);
3985         }
3986 
3987         tx = dmu_tx_create(zfsvfs->z_os);
3988         dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3989         dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3990         zfs_sa_upgrade_txholds(tx, szp);
3991         zfs_sa_upgrade_txholds(tx, dzp);
3992         error = dmu_tx_assign(tx, TXG_NOWAIT);
3993         if (error) {
3994                 zfs_dirent_unlock(dl);
3995                 if (error == ERESTART) {
3996                         dmu_tx_wait(tx);
3997                         dmu_tx_abort(tx);
3998                         goto top;
3999                 }
4000                 dmu_tx_abort(tx);
4001                 ZFS_EXIT(zfsvfs);
4002                 return (error);
4003         }
4004 
4005         error = zfs_link_create(dl, szp, tx, 0);
4006 
4007         if (error == 0) {
4008                 uint64_t txtype = TX_LINK;
4009                 if (flags & FIGNORECASE)
4010                         txtype |= TX_CI;
4011                 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4012         }
4013 
4014         dmu_tx_commit(tx);
4015 
4016         zfs_dirent_unlock(dl);
4017 
4018         if (error == 0) {
4019                 vnevent_link(svp, ct);
4020         }
4021 
4022         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4023                 zil_commit(zilog, 0);
4024 
4025         ZFS_EXIT(zfsvfs);
4026         return (error);
4027 }
4028 
4029 /*
4030  * zfs_null_putapage() is used when the file system has been force
4031  * unmounted. It just drops the pages.
4032  */
4033 /* ARGSUSED */
4034 static int
4035 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4036                 size_t *lenp, int flags, cred_t *cr)
4037 {
4038         pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4039         return (0);
4040 }
4041 
4042 /*
4043  * Push a page out to disk, klustering if possible.
4044  *
4045  *      IN:     vp      - file to push page to.
4046  *              pp      - page to push.
4047  *              flags   - additional flags.
4048  *              cr      - credentials of caller.
4049  *
4050  *      OUT:    offp    - start of range pushed.
4051  *              lenp    - len of range pushed.
4052  *
4053  *      RETURN: 0 on success, error code on failure.
4054  *
4055  * NOTE: callers must have locked the page to be pushed.  On
4056  * exit, the page (and all other pages in the kluster) must be
4057  * unlocked.
4058  */
4059 /* ARGSUSED */
4060 static int
4061 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4062                 size_t *lenp, int flags, cred_t *cr)
4063 {
4064         znode_t         *zp = VTOZ(vp);
4065         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4066         dmu_tx_t        *tx;
4067         u_offset_t      off, koff;
4068         size_t          len, klen;
4069         int             err;
4070 
4071         off = pp->p_offset;
4072         len = PAGESIZE;
4073         /*
4074          * If our blocksize is bigger than the page size, try to kluster
4075          * multiple pages so that we write a full block (thus avoiding
4076          * a read-modify-write).
4077          */
4078         if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4079                 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4080                 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4081                 ASSERT(koff <= zp->z_size);
4082                 if (koff + klen > zp->z_size)
4083                         klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4084                 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4085         }
4086         ASSERT3U(btop(len), ==, btopr(len));
4087 
4088         /*
4089          * Can't push pages past end-of-file.
4090          */
4091         if (off >= zp->z_size) {
4092                 /* ignore all pages */
4093                 err = 0;
4094                 goto out;
4095         } else if (off + len > zp->z_size) {
4096                 int npages = btopr(zp->z_size - off);
4097                 page_t *trunc;
4098 
4099                 page_list_break(&pp, &trunc, npages);
4100                 /* ignore pages past end of file */
4101                 if (trunc)
4102                         pvn_write_done(trunc, flags);
4103                 len = zp->z_size - off;
4104         }
4105 
4106         if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4107             zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4108                 err = SET_ERROR(EDQUOT);
4109                 goto out;
4110         }
4111 top:
4112         tx = dmu_tx_create(zfsvfs->z_os);
4113         dmu_tx_hold_write(tx, zp->z_id, off, len);
4114 
4115         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4116         zfs_sa_upgrade_txholds(tx, zp);
4117         err = dmu_tx_assign(tx, TXG_NOWAIT);
4118         if (err != 0) {
4119                 if (err == ERESTART) {
4120                         dmu_tx_wait(tx);
4121                         dmu_tx_abort(tx);
4122                         goto top;
4123                 }
4124                 dmu_tx_abort(tx);
4125                 goto out;
4126         }
4127 
4128         if (zp->z_blksz <= PAGESIZE) {
4129                 caddr_t va = zfs_map_page(pp, S_READ);
4130                 ASSERT3U(len, <=, PAGESIZE);
4131                 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4132                 zfs_unmap_page(pp, va);
4133         } else {
4134                 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4135         }
4136 
4137         if (err == 0) {
4138                 uint64_t mtime[2], ctime[2];
4139                 sa_bulk_attr_t bulk[3];
4140                 int count = 0;
4141 
4142                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4143                     &mtime, 16);
4144                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4145                     &ctime, 16);
4146                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4147                     &zp->z_pflags, 8);
4148                 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4149                     B_TRUE);
4150                 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4151         }
4152         dmu_tx_commit(tx);
4153 
4154 out:
4155         pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4156         if (offp)
4157                 *offp = off;
4158         if (lenp)
4159                 *lenp = len;
4160 
4161         return (err);
4162 }
4163 
4164 /*
4165  * Copy the portion of the file indicated from pages into the file.
4166  * The pages are stored in a page list attached to the files vnode.
4167  *
4168  *      IN:     vp      - vnode of file to push page data to.
4169  *              off     - position in file to put data.
4170  *              len     - amount of data to write.
4171  *              flags   - flags to control the operation.
4172  *              cr      - credentials of caller.
4173  *              ct      - caller context.
4174  *
4175  *      RETURN: 0 on success, error code on failure.
4176  *
4177  * Timestamps:
4178  *      vp - ctime|mtime updated
4179  */
4180 /*ARGSUSED*/
4181 static int
4182 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4183     caller_context_t *ct)
4184 {
4185         znode_t         *zp = VTOZ(vp);
4186         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4187         page_t          *pp;
4188         size_t          io_len;
4189         u_offset_t      io_off;
4190         uint_t          blksz;
4191         rl_t            *rl;
4192         int             error = 0;
4193 
4194         ZFS_ENTER(zfsvfs);
4195         ZFS_VERIFY_ZP(zp);
4196 
4197         /*
4198          * There's nothing to do if no data is cached.
4199          */
4200         if (!vn_has_cached_data(vp)) {
4201                 ZFS_EXIT(zfsvfs);
4202                 return (0);
4203         }
4204 
4205         /*
4206          * Align this request to the file block size in case we kluster.
4207          * XXX - this can result in pretty aggresive locking, which can
4208          * impact simultanious read/write access.  One option might be
4209          * to break up long requests (len == 0) into block-by-block
4210          * operations to get narrower locking.
4211          */
4212         blksz = zp->z_blksz;
4213         if (ISP2(blksz))
4214                 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4215         else
4216                 io_off = 0;
4217         if (len > 0 && ISP2(blksz))
4218                 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4219         else
4220                 io_len = 0;
4221 
4222         if (io_len == 0) {
4223                 /*
4224                  * Search the entire vp list for pages >= io_off.
4225                  */
4226                 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4227                 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4228                 goto out;
4229         }
4230         rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4231 
4232         if (off > zp->z_size) {
4233                 /* past end of file */
4234                 zfs_range_unlock(rl);
4235                 ZFS_EXIT(zfsvfs);
4236                 return (0);
4237         }
4238 
4239         len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4240 
4241         for (off = io_off; io_off < off + len; io_off += io_len) {
4242                 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4243                         pp = page_lookup(vp, io_off,
4244                             (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4245                 } else {
4246                         pp = page_lookup_nowait(vp, io_off,
4247                             (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4248                 }
4249 
4250                 if (pp != NULL && pvn_getdirty(pp, flags)) {
4251                         int err;
4252 
4253                         /*
4254                          * Found a dirty page to push
4255                          */
4256                         err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4257                         if (err)
4258                                 error = err;
4259                 } else {
4260                         io_len = PAGESIZE;
4261                 }
4262         }
4263 out:
4264         zfs_range_unlock(rl);
4265         if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4266                 zil_commit(zfsvfs->z_log, zp->z_id);
4267         ZFS_EXIT(zfsvfs);
4268         return (error);
4269 }
4270 
4271 /*ARGSUSED*/
4272 void
4273 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4274 {
4275         znode_t *zp = VTOZ(vp);
4276         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4277         int error;
4278 
4279         rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4280         if (zp->z_sa_hdl == NULL) {
4281                 /*
4282                  * The fs has been unmounted, or we did a
4283                  * suspend/resume and this file no longer exists.
4284                  */
4285                 if (vn_has_cached_data(vp)) {
4286                         (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4287                             B_INVAL, cr);
4288                 }
4289 
4290                 mutex_enter(&zp->z_lock);
4291                 mutex_enter(&vp->v_lock);
4292                 ASSERT(vp->v_count == 1);
4293                 vp->v_count = 0;
4294                 mutex_exit(&vp->v_lock);
4295                 mutex_exit(&zp->z_lock);
4296                 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4297                 zfs_znode_free(zp);
4298                 return;
4299         }
4300 
4301         /*
4302          * Attempt to push any data in the page cache.  If this fails
4303          * we will get kicked out later in zfs_zinactive().
4304          */
4305         if (vn_has_cached_data(vp)) {
4306                 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4307                     cr);
4308         }
4309 
4310         if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4311                 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4312 
4313                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4314                 zfs_sa_upgrade_txholds(tx, zp);
4315                 error = dmu_tx_assign(tx, TXG_WAIT);
4316                 if (error) {
4317                         dmu_tx_abort(tx);
4318                 } else {
4319                         mutex_enter(&zp->z_lock);
4320                         (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4321                             (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4322                         zp->z_atime_dirty = 0;
4323                         mutex_exit(&zp->z_lock);
4324                         dmu_tx_commit(tx);
4325                 }
4326         }
4327 
4328         zfs_zinactive(zp);
4329         rw_exit(&zfsvfs->z_teardown_inactive_lock);
4330 }
4331 
4332 /*
4333  * Bounds-check the seek operation.
4334  *
4335  *      IN:     vp      - vnode seeking within
4336  *              ooff    - old file offset
4337  *              noffp   - pointer to new file offset
4338  *              ct      - caller context
4339  *
4340  *      RETURN: 0 on success, EINVAL if new offset invalid.
4341  */
4342 /* ARGSUSED */
4343 static int
4344 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4345     caller_context_t *ct)
4346 {
4347         if (vp->v_type == VDIR)
4348                 return (0);
4349         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4350 }
4351 
4352 /*
4353  * Pre-filter the generic locking function to trap attempts to place
4354  * a mandatory lock on a memory mapped file.
4355  */
4356 static int
4357 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4358     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4359 {
4360         znode_t *zp = VTOZ(vp);
4361         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4362 
4363         ZFS_ENTER(zfsvfs);
4364         ZFS_VERIFY_ZP(zp);
4365 
4366         /*
4367          * We are following the UFS semantics with respect to mapcnt
4368          * here: If we see that the file is mapped already, then we will
4369          * return an error, but we don't worry about races between this
4370          * function and zfs_map().
4371          */
4372         if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4373                 ZFS_EXIT(zfsvfs);
4374                 return (SET_ERROR(EAGAIN));
4375         }
4376         ZFS_EXIT(zfsvfs);
4377         return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4378 }
4379 
4380 /*
4381  * If we can't find a page in the cache, we will create a new page
4382  * and fill it with file data.  For efficiency, we may try to fill
4383  * multiple pages at once (klustering) to fill up the supplied page
4384  * list.  Note that the pages to be filled are held with an exclusive
4385  * lock to prevent access by other threads while they are being filled.
4386  */
4387 static int
4388 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4389     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4390 {
4391         znode_t *zp = VTOZ(vp);
4392         page_t *pp, *cur_pp;
4393         objset_t *os = zp->z_zfsvfs->z_os;
4394         u_offset_t io_off, total;
4395         size_t io_len;
4396         int err;
4397 
4398         if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4399                 /*
4400                  * We only have a single page, don't bother klustering
4401                  */
4402                 io_off = off;
4403                 io_len = PAGESIZE;
4404                 pp = page_create_va(vp, io_off, io_len,
4405                     PG_EXCL | PG_WAIT, seg, addr);
4406         } else {
4407                 /*
4408                  * Try to find enough pages to fill the page list
4409                  */
4410                 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4411                     &io_len, off, plsz, 0);
4412         }
4413         if (pp == NULL) {
4414                 /*
4415                  * The page already exists, nothing to do here.
4416                  */
4417                 *pl = NULL;
4418                 return (0);
4419         }
4420 
4421         /*
4422          * Fill the pages in the kluster.
4423          */
4424         cur_pp = pp;
4425         for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4426                 caddr_t va;
4427 
4428                 ASSERT3U(io_off, ==, cur_pp->p_offset);
4429                 va = zfs_map_page(cur_pp, S_WRITE);
4430                 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4431                     DMU_READ_PREFETCH);
4432                 zfs_unmap_page(cur_pp, va);
4433                 if (err) {
4434                         /* On error, toss the entire kluster */
4435                         pvn_read_done(pp, B_ERROR);
4436                         /* convert checksum errors into IO errors */
4437                         if (err == ECKSUM)
4438                                 err = SET_ERROR(EIO);
4439                         return (err);
4440                 }
4441                 cur_pp = cur_pp->p_next;
4442         }
4443 
4444         /*
4445          * Fill in the page list array from the kluster starting
4446          * from the desired offset `off'.
4447          * NOTE: the page list will always be null terminated.
4448          */
4449         pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4450         ASSERT(pl == NULL || (*pl)->p_offset == off);
4451 
4452         return (0);
4453 }
4454 
4455 /*
4456  * Return pointers to the pages for the file region [off, off + len]
4457  * in the pl array.  If plsz is greater than len, this function may
4458  * also return page pointers from after the specified region
4459  * (i.e. the region [off, off + plsz]).  These additional pages are
4460  * only returned if they are already in the cache, or were created as
4461  * part of a klustered read.
4462  *
4463  *      IN:     vp      - vnode of file to get data from.
4464  *              off     - position in file to get data from.
4465  *              len     - amount of data to retrieve.
4466  *              plsz    - length of provided page list.
4467  *              seg     - segment to obtain pages for.
4468  *              addr    - virtual address of fault.
4469  *              rw      - mode of created pages.
4470  *              cr      - credentials of caller.
4471  *              ct      - caller context.
4472  *
4473  *      OUT:    protp   - protection mode of created pages.
4474  *              pl      - list of pages created.
4475  *
4476  *      RETURN: 0 on success, error code on failure.
4477  *
4478  * Timestamps:
4479  *      vp - atime updated
4480  */
4481 /* ARGSUSED */
4482 static int
4483 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4484     page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4485     enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4486 {
4487         znode_t         *zp = VTOZ(vp);
4488         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4489         page_t          **pl0 = pl;
4490         int             err = 0;
4491 
4492         /* we do our own caching, faultahead is unnecessary */
4493         if (pl == NULL)
4494                 return (0);
4495         else if (len > plsz)
4496                 len = plsz;
4497         else
4498                 len = P2ROUNDUP(len, PAGESIZE);
4499         ASSERT(plsz >= len);
4500 
4501         ZFS_ENTER(zfsvfs);
4502         ZFS_VERIFY_ZP(zp);
4503 
4504         if (protp)
4505                 *protp = PROT_ALL;
4506 
4507         /*
4508          * Loop through the requested range [off, off + len) looking
4509          * for pages.  If we don't find a page, we will need to create
4510          * a new page and fill it with data from the file.
4511          */
4512         while (len > 0) {
4513                 if (*pl = page_lookup(vp, off, SE_SHARED))
4514                         *(pl+1) = NULL;
4515                 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4516                         goto out;
4517                 while (*pl) {
4518                         ASSERT3U((*pl)->p_offset, ==, off);
4519                         off += PAGESIZE;
4520                         addr += PAGESIZE;
4521                         if (len > 0) {
4522                                 ASSERT3U(len, >=, PAGESIZE);
4523                                 len -= PAGESIZE;
4524                         }
4525                         ASSERT3U(plsz, >=, PAGESIZE);
4526                         plsz -= PAGESIZE;
4527                         pl++;
4528                 }
4529         }
4530 
4531         /*
4532          * Fill out the page array with any pages already in the cache.
4533          */
4534         while (plsz > 0 &&
4535             (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4536                         off += PAGESIZE;
4537                         plsz -= PAGESIZE;
4538         }
4539 out:
4540         if (err) {
4541                 /*
4542                  * Release any pages we have previously locked.
4543                  */
4544                 while (pl > pl0)
4545                         page_unlock(*--pl);
4546         } else {
4547                 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4548         }
4549 
4550         *pl = NULL;
4551 
4552         ZFS_EXIT(zfsvfs);
4553         return (err);
4554 }
4555 
4556 /*
4557  * Request a memory map for a section of a file.  This code interacts
4558  * with common code and the VM system as follows:
4559  *
4560  * - common code calls mmap(), which ends up in smmap_common()
4561  * - this calls VOP_MAP(), which takes you into (say) zfs
4562  * - zfs_map() calls as_map(), passing segvn_create() as the callback
4563  * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4564  * - zfs_addmap() updates z_mapcnt
4565  */
4566 /*ARGSUSED*/
4567 static int
4568 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4569     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4570     caller_context_t *ct)
4571 {
4572         znode_t *zp = VTOZ(vp);
4573         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4574         segvn_crargs_t  vn_a;
4575         int             error;
4576 
4577         ZFS_ENTER(zfsvfs);
4578         ZFS_VERIFY_ZP(zp);
4579 
4580         if ((prot & PROT_WRITE) && (zp->z_pflags &
4581             (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4582                 ZFS_EXIT(zfsvfs);
4583                 return (SET_ERROR(EPERM));
4584         }
4585 
4586         if ((prot & (PROT_READ | PROT_EXEC)) &&
4587             (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4588                 ZFS_EXIT(zfsvfs);
4589                 return (SET_ERROR(EACCES));
4590         }
4591 
4592         if (vp->v_flag & VNOMAP) {
4593                 ZFS_EXIT(zfsvfs);
4594                 return (SET_ERROR(ENOSYS));
4595         }
4596 
4597         if (off < 0 || len > MAXOFFSET_T - off) {
4598                 ZFS_EXIT(zfsvfs);
4599                 return (SET_ERROR(ENXIO));
4600         }
4601 
4602         if (vp->v_type != VREG) {
4603                 ZFS_EXIT(zfsvfs);
4604                 return (SET_ERROR(ENODEV));
4605         }
4606 
4607         /*
4608          * If file is locked, disallow mapping.
4609          */
4610         if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4611                 ZFS_EXIT(zfsvfs);
4612                 return (SET_ERROR(EAGAIN));
4613         }
4614 
4615         as_rangelock(as);
4616         error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4617         if (error != 0) {
4618                 as_rangeunlock(as);
4619                 ZFS_EXIT(zfsvfs);
4620                 return (error);
4621         }
4622 
4623         vn_a.vp = vp;
4624         vn_a.offset = (u_offset_t)off;
4625         vn_a.type = flags & MAP_TYPE;
4626         vn_a.prot = prot;
4627         vn_a.maxprot = maxprot;
4628         vn_a.cred = cr;
4629         vn_a.amp = NULL;
4630         vn_a.flags = flags & ~MAP_TYPE;
4631         vn_a.szc = 0;
4632         vn_a.lgrp_mem_policy_flags = 0;
4633 
4634         error = as_map(as, *addrp, len, segvn_create, &vn_a);
4635 
4636         as_rangeunlock(as);
4637         ZFS_EXIT(zfsvfs);
4638         return (error);
4639 }
4640 
4641 /* ARGSUSED */
4642 static int
4643 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4644     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4645     caller_context_t *ct)
4646 {
4647         uint64_t pages = btopr(len);
4648 
4649         atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4650         return (0);
4651 }
4652 
4653 /*
4654  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4655  * more accurate mtime for the associated file.  Since we don't have a way of
4656  * detecting when the data was actually modified, we have to resort to
4657  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4658  * last page is pushed.  The problem occurs when the msync() call is omitted,
4659  * which by far the most common case:
4660  *
4661  *      open()
4662  *      mmap()
4663  *      <modify memory>
4664  *      munmap()
4665  *      close()
4666  *      <time lapse>
4667  *      putpage() via fsflush
4668  *
4669  * If we wait until fsflush to come along, we can have a modification time that
4670  * is some arbitrary point in the future.  In order to prevent this in the
4671  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4672  * torn down.
4673  */
4674 /* ARGSUSED */
4675 static int
4676 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4677     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4678     caller_context_t *ct)
4679 {
4680         uint64_t pages = btopr(len);
4681 
4682         ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4683         atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4684 
4685         if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4686             vn_has_cached_data(vp))
4687                 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4688 
4689         return (0);
4690 }
4691 
4692 /*
4693  * Free or allocate space in a file.  Currently, this function only
4694  * supports the `F_FREESP' command.  However, this command is somewhat
4695  * misnamed, as its functionality includes the ability to allocate as
4696  * well as free space.
4697  *
4698  *      IN:     vp      - vnode of file to free data in.
4699  *              cmd     - action to take (only F_FREESP supported).
4700  *              bfp     - section of file to free/alloc.
4701  *              flag    - current file open mode flags.
4702  *              offset  - current file offset.
4703  *              cr      - credentials of caller [UNUSED].
4704  *              ct      - caller context.
4705  *
4706  *      RETURN: 0 on success, error code on failure.
4707  *
4708  * Timestamps:
4709  *      vp - ctime|mtime updated
4710  */
4711 /* ARGSUSED */
4712 static int
4713 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4714     offset_t offset, cred_t *cr, caller_context_t *ct)
4715 {
4716         znode_t         *zp = VTOZ(vp);
4717         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4718         uint64_t        off, len;
4719         int             error;
4720 
4721         ZFS_ENTER(zfsvfs);
4722         ZFS_VERIFY_ZP(zp);
4723 
4724         if (cmd != F_FREESP) {
4725                 ZFS_EXIT(zfsvfs);
4726                 return (SET_ERROR(EINVAL));
4727         }
4728 
4729         if (error = convoff(vp, bfp, 0, offset)) {
4730                 ZFS_EXIT(zfsvfs);
4731                 return (error);
4732         }
4733 
4734         if (bfp->l_len < 0) {
4735                 ZFS_EXIT(zfsvfs);
4736                 return (SET_ERROR(EINVAL));
4737         }
4738 
4739         off = bfp->l_start;
4740         len = bfp->l_len; /* 0 means from off to end of file */
4741 
4742         error = zfs_freesp(zp, off, len, flag, TRUE);
4743 
4744         ZFS_EXIT(zfsvfs);
4745         return (error);
4746 }
4747 
4748 /*ARGSUSED*/
4749 static int
4750 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4751 {
4752         znode_t         *zp = VTOZ(vp);
4753         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4754         uint32_t        gen;
4755         uint64_t        gen64;
4756         uint64_t        object = zp->z_id;
4757         zfid_short_t    *zfid;
4758         int             size, i, error;
4759 
4760         ZFS_ENTER(zfsvfs);
4761         ZFS_VERIFY_ZP(zp);
4762 
4763         if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4764             &gen64, sizeof (uint64_t))) != 0) {
4765                 ZFS_EXIT(zfsvfs);
4766                 return (error);
4767         }
4768 
4769         gen = (uint32_t)gen64;
4770 
4771         size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4772         if (fidp->fid_len < size) {
4773                 fidp->fid_len = size;
4774                 ZFS_EXIT(zfsvfs);
4775                 return (SET_ERROR(ENOSPC));
4776         }
4777 
4778         zfid = (zfid_short_t *)fidp;
4779 
4780         zfid->zf_len = size;
4781 
4782         for (i = 0; i < sizeof (zfid->zf_object); i++)
4783                 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4784 
4785         /* Must have a non-zero generation number to distinguish from .zfs */
4786         if (gen == 0)
4787                 gen = 1;
4788         for (i = 0; i < sizeof (zfid->zf_gen); i++)
4789                 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4790 
4791         if (size == LONG_FID_LEN) {
4792                 uint64_t        objsetid = dmu_objset_id(zfsvfs->z_os);
4793                 zfid_long_t     *zlfid;
4794 
4795                 zlfid = (zfid_long_t *)fidp;
4796 
4797                 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4798                         zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4799 
4800                 /* XXX - this should be the generation number for the objset */
4801                 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4802                         zlfid->zf_setgen[i] = 0;
4803         }
4804 
4805         ZFS_EXIT(zfsvfs);
4806         return (0);
4807 }
4808 
4809 static int
4810 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4811     caller_context_t *ct)
4812 {
4813         znode_t         *zp, *xzp;
4814         zfsvfs_t        *zfsvfs;
4815         zfs_dirlock_t   *dl;
4816         int             error;
4817 
4818         switch (cmd) {
4819         case _PC_LINK_MAX:
4820                 *valp = ULONG_MAX;
4821                 return (0);
4822 
4823         case _PC_FILESIZEBITS:
4824                 *valp = 64;
4825                 return (0);
4826 
4827         case _PC_XATTR_EXISTS:
4828                 zp = VTOZ(vp);
4829                 zfsvfs = zp->z_zfsvfs;
4830                 ZFS_ENTER(zfsvfs);
4831                 ZFS_VERIFY_ZP(zp);
4832                 *valp = 0;
4833                 error = zfs_dirent_lock(&dl, zp, "", &xzp,
4834                     ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4835                 if (error == 0) {
4836                         zfs_dirent_unlock(dl);
4837                         if (!zfs_dirempty(xzp))
4838                                 *valp = 1;
4839                         VN_RELE(ZTOV(xzp));
4840                 } else if (error == ENOENT) {
4841                         /*
4842                          * If there aren't extended attributes, it's the
4843                          * same as having zero of them.
4844                          */
4845                         error = 0;
4846                 }
4847                 ZFS_EXIT(zfsvfs);
4848                 return (error);
4849 
4850         case _PC_SATTR_ENABLED:
4851         case _PC_SATTR_EXISTS:
4852                 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4853                     (vp->v_type == VREG || vp->v_type == VDIR);
4854                 return (0);
4855 
4856         case _PC_ACCESS_FILTERING:
4857                 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4858                     vp->v_type == VDIR;
4859                 return (0);
4860 
4861         case _PC_ACL_ENABLED:
4862                 *valp = _ACL_ACE_ENABLED;
4863                 return (0);
4864 
4865         case _PC_MIN_HOLE_SIZE:
4866                 *valp = (ulong_t)SPA_MINBLOCKSIZE;
4867                 return (0);
4868 
4869         case _PC_TIMESTAMP_RESOLUTION:
4870                 /* nanosecond timestamp resolution */
4871                 *valp = 1L;
4872                 return (0);
4873 
4874         default:
4875                 return (fs_pathconf(vp, cmd, valp, cr, ct));
4876         }
4877 }
4878 
4879 /*ARGSUSED*/
4880 static int
4881 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4882     caller_context_t *ct)
4883 {
4884         znode_t *zp = VTOZ(vp);
4885         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4886         int error;
4887         boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4888 
4889         ZFS_ENTER(zfsvfs);
4890         ZFS_VERIFY_ZP(zp);
4891         error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4892         ZFS_EXIT(zfsvfs);
4893 
4894         return (error);
4895 }
4896 
4897 /*ARGSUSED*/
4898 static int
4899 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4900     caller_context_t *ct)
4901 {
4902         znode_t *zp = VTOZ(vp);
4903         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4904         int error;
4905         boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4906         zilog_t *zilog = zfsvfs->z_log;
4907 
4908         ZFS_ENTER(zfsvfs);
4909         ZFS_VERIFY_ZP(zp);
4910 
4911         error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4912 
4913         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4914                 zil_commit(zilog, 0);
4915 
4916         ZFS_EXIT(zfsvfs);
4917         return (error);
4918 }
4919 
4920 /*
4921  * The smallest read we may consider to loan out an arcbuf.
4922  * This must be a power of 2.
4923  */
4924 int zcr_blksz_min = (1 << 10);    /* 1K */
4925 /*
4926  * If set to less than the file block size, allow loaning out of an
4927  * arcbuf for a partial block read.  This must be a power of 2.
4928  */
4929 int zcr_blksz_max = (1 << 17);    /* 128K */
4930 
4931 /*ARGSUSED*/
4932 static int
4933 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
4934     caller_context_t *ct)
4935 {
4936         znode_t *zp = VTOZ(vp);
4937         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4938         int max_blksz = zfsvfs->z_max_blksz;
4939         uio_t *uio = &xuio->xu_uio;
4940         ssize_t size = uio->uio_resid;
4941         offset_t offset = uio->uio_loffset;
4942         int blksz;
4943         int fullblk, i;
4944         arc_buf_t *abuf;
4945         ssize_t maxsize;
4946         int preamble, postamble;
4947 
4948         if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4949                 return (SET_ERROR(EINVAL));
4950 
4951         ZFS_ENTER(zfsvfs);
4952         ZFS_VERIFY_ZP(zp);
4953         switch (ioflag) {
4954         case UIO_WRITE:
4955                 /*
4956                  * Loan out an arc_buf for write if write size is bigger than
4957                  * max_blksz, and the file's block size is also max_blksz.
4958                  */
4959                 blksz = max_blksz;
4960                 if (size < blksz || zp->z_blksz != blksz) {
4961                         ZFS_EXIT(zfsvfs);
4962                         return (SET_ERROR(EINVAL));
4963                 }
4964                 /*
4965                  * Caller requests buffers for write before knowing where the
4966                  * write offset might be (e.g. NFS TCP write).
4967                  */
4968                 if (offset == -1) {
4969                         preamble = 0;
4970                 } else {
4971                         preamble = P2PHASE(offset, blksz);
4972                         if (preamble) {
4973                                 preamble = blksz - preamble;
4974                                 size -= preamble;
4975                         }
4976                 }
4977 
4978                 postamble = P2PHASE(size, blksz);
4979                 size -= postamble;
4980 
4981                 fullblk = size / blksz;
4982                 (void) dmu_xuio_init(xuio,
4983                     (preamble != 0) + fullblk + (postamble != 0));
4984                 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
4985                     int, postamble, int,
4986                     (preamble != 0) + fullblk + (postamble != 0));
4987 
4988                 /*
4989                  * Have to fix iov base/len for partial buffers.  They
4990                  * currently represent full arc_buf's.
4991                  */
4992                 if (preamble) {
4993                         /* data begins in the middle of the arc_buf */
4994                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4995                             blksz);
4996                         ASSERT(abuf);
4997                         (void) dmu_xuio_add(xuio, abuf,
4998                             blksz - preamble, preamble);
4999                 }
5000 
5001                 for (i = 0; i < fullblk; i++) {
5002                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5003                             blksz);
5004                         ASSERT(abuf);
5005                         (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5006                 }
5007 
5008                 if (postamble) {
5009                         /* data ends in the middle of the arc_buf */
5010                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5011                             blksz);
5012                         ASSERT(abuf);
5013                         (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5014                 }
5015                 break;
5016         case UIO_READ:
5017                 /*
5018                  * Loan out an arc_buf for read if the read size is larger than
5019                  * the current file block size.  Block alignment is not
5020                  * considered.  Partial arc_buf will be loaned out for read.
5021                  */
5022                 blksz = zp->z_blksz;
5023                 if (blksz < zcr_blksz_min)
5024                         blksz = zcr_blksz_min;
5025                 if (blksz > zcr_blksz_max)
5026                         blksz = zcr_blksz_max;
5027                 /* avoid potential complexity of dealing with it */
5028                 if (blksz > max_blksz) {
5029                         ZFS_EXIT(zfsvfs);
5030                         return (SET_ERROR(EINVAL));
5031                 }
5032 
5033                 maxsize = zp->z_size - uio->uio_loffset;
5034                 if (size > maxsize)
5035                         size = maxsize;
5036 
5037                 if (size < blksz || vn_has_cached_data(vp)) {
5038                         ZFS_EXIT(zfsvfs);
5039                         return (SET_ERROR(EINVAL));
5040                 }
5041                 break;
5042         default:
5043                 ZFS_EXIT(zfsvfs);
5044                 return (SET_ERROR(EINVAL));
5045         }
5046 
5047         uio->uio_extflg = UIO_XUIO;
5048         XUIO_XUZC_RW(xuio) = ioflag;
5049         ZFS_EXIT(zfsvfs);
5050         return (0);
5051 }
5052 
5053 /*ARGSUSED*/
5054 static int
5055 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5056 {
5057         int i;
5058         arc_buf_t *abuf;
5059         int ioflag = XUIO_XUZC_RW(xuio);
5060 
5061         ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5062 
5063         i = dmu_xuio_cnt(xuio);
5064         while (i-- > 0) {
5065                 abuf = dmu_xuio_arcbuf(xuio, i);
5066                 /*
5067                  * if abuf == NULL, it must be a write buffer
5068                  * that has been returned in zfs_write().
5069                  */
5070                 if (abuf)
5071                         dmu_return_arcbuf(abuf);
5072                 ASSERT(abuf || ioflag == UIO_WRITE);
5073         }
5074 
5075         dmu_xuio_fini(xuio);
5076         return (0);
5077 }
5078 
5079 /*
5080  * Predeclare these here so that the compiler assumes that
5081  * this is an "old style" function declaration that does
5082  * not include arguments => we won't get type mismatch errors
5083  * in the initializations that follow.
5084  */
5085 static int zfs_inval();
5086 static int zfs_isdir();
5087 
5088 static int
5089 zfs_inval()
5090 {
5091         return (SET_ERROR(EINVAL));
5092 }
5093 
5094 static int
5095 zfs_isdir()
5096 {
5097         return (SET_ERROR(EISDIR));
5098 }
5099 /*
5100  * Directory vnode operations template
5101  */
5102 vnodeops_t *zfs_dvnodeops;
5103 const fs_operation_def_t zfs_dvnodeops_template[] = {
5104         VOPNAME_OPEN,           { .vop_open = zfs_open },
5105         VOPNAME_CLOSE,          { .vop_close = zfs_close },
5106         VOPNAME_READ,           { .error = zfs_isdir },
5107         VOPNAME_WRITE,          { .error = zfs_isdir },
5108         VOPNAME_IOCTL,          { .vop_ioctl = zfs_ioctl },
5109         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5110         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5111         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5112         VOPNAME_LOOKUP,         { .vop_lookup = zfs_lookup },
5113         VOPNAME_CREATE,         { .vop_create = zfs_create },
5114         VOPNAME_REMOVE,         { .vop_remove = zfs_remove },
5115         VOPNAME_LINK,           { .vop_link = zfs_link },
5116         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5117         VOPNAME_MKDIR,          { .vop_mkdir = zfs_mkdir },
5118         VOPNAME_RMDIR,          { .vop_rmdir = zfs_rmdir },
5119         VOPNAME_READDIR,        { .vop_readdir = zfs_readdir },
5120         VOPNAME_SYMLINK,        { .vop_symlink = zfs_symlink },
5121         VOPNAME_FSYNC,          { .vop_fsync = zfs_fsync },
5122         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5123         VOPNAME_FID,            { .vop_fid = zfs_fid },
5124         VOPNAME_SEEK,           { .vop_seek = zfs_seek },
5125         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5126         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5127         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5128         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5129         NULL,                   NULL
5130 };
5131 
5132 /*
5133  * Regular file vnode operations template
5134  */
5135 vnodeops_t *zfs_fvnodeops;
5136 const fs_operation_def_t zfs_fvnodeops_template[] = {
5137         VOPNAME_OPEN,           { .vop_open = zfs_open },
5138         VOPNAME_CLOSE,          { .vop_close = zfs_close },
5139         VOPNAME_READ,           { .vop_read = zfs_read },
5140         VOPNAME_WRITE,          { .vop_write = zfs_write },
5141         VOPNAME_IOCTL,          { .vop_ioctl = zfs_ioctl },
5142         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5143         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5144         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5145         VOPNAME_LOOKUP,         { .vop_lookup = zfs_lookup },
5146         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5147         VOPNAME_FSYNC,          { .vop_fsync = zfs_fsync },
5148         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5149         VOPNAME_FID,            { .vop_fid = zfs_fid },
5150         VOPNAME_SEEK,           { .vop_seek = zfs_seek },
5151         VOPNAME_FRLOCK,         { .vop_frlock = zfs_frlock },
5152         VOPNAME_SPACE,          { .vop_space = zfs_space },
5153         VOPNAME_GETPAGE,        { .vop_getpage = zfs_getpage },
5154         VOPNAME_PUTPAGE,        { .vop_putpage = zfs_putpage },
5155         VOPNAME_MAP,            { .vop_map = zfs_map },
5156         VOPNAME_ADDMAP,         { .vop_addmap = zfs_addmap },
5157         VOPNAME_DELMAP,         { .vop_delmap = zfs_delmap },
5158         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5159         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5160         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5161         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5162         VOPNAME_REQZCBUF,       { .vop_reqzcbuf = zfs_reqzcbuf },
5163         VOPNAME_RETZCBUF,       { .vop_retzcbuf = zfs_retzcbuf },
5164         NULL,                   NULL
5165 };
5166 
5167 /*
5168  * Symbolic link vnode operations template
5169  */
5170 vnodeops_t *zfs_symvnodeops;
5171 const fs_operation_def_t zfs_symvnodeops_template[] = {
5172         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5173         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5174         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5175         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5176         VOPNAME_READLINK,       { .vop_readlink = zfs_readlink },
5177         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5178         VOPNAME_FID,            { .vop_fid = zfs_fid },
5179         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5180         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5181         NULL,                   NULL
5182 };
5183 
5184 /*
5185  * special share hidden files vnode operations template
5186  */
5187 vnodeops_t *zfs_sharevnodeops;
5188 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5189         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5190         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5191         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5192         VOPNAME_FID,            { .vop_fid = zfs_fid },
5193         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5194         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5195         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5196         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5197         NULL,                   NULL
5198 };
5199 
5200 /*
5201  * Extended attribute directory vnode operations template
5202  *
5203  * This template is identical to the directory vnodes
5204  * operation template except for restricted operations:
5205  *      VOP_MKDIR()
5206  *      VOP_SYMLINK()
5207  *
5208  * Note that there are other restrictions embedded in:
5209  *      zfs_create()    - restrict type to VREG
5210  *      zfs_link()      - no links into/out of attribute space
5211  *      zfs_rename()    - no moves into/out of attribute space
5212  */
5213 vnodeops_t *zfs_xdvnodeops;
5214 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5215         VOPNAME_OPEN,           { .vop_open = zfs_open },
5216         VOPNAME_CLOSE,          { .vop_close = zfs_close },
5217         VOPNAME_IOCTL,          { .vop_ioctl = zfs_ioctl },
5218         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5219         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5220         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5221         VOPNAME_LOOKUP,         { .vop_lookup = zfs_lookup },
5222         VOPNAME_CREATE,         { .vop_create = zfs_create },
5223         VOPNAME_REMOVE,         { .vop_remove = zfs_remove },
5224         VOPNAME_LINK,           { .vop_link = zfs_link },
5225         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5226         VOPNAME_MKDIR,          { .error = zfs_inval },
5227         VOPNAME_RMDIR,          { .vop_rmdir = zfs_rmdir },
5228         VOPNAME_READDIR,        { .vop_readdir = zfs_readdir },
5229         VOPNAME_SYMLINK,        { .error = zfs_inval },
5230         VOPNAME_FSYNC,          { .vop_fsync = zfs_fsync },
5231         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5232         VOPNAME_FID,            { .vop_fid = zfs_fid },
5233         VOPNAME_SEEK,           { .vop_seek = zfs_seek },
5234         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5235         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5236         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5237         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5238         NULL,                   NULL
5239 };
5240 
5241 /*
5242  * Error vnode operations template
5243  */
5244 vnodeops_t *zfs_evnodeops;
5245 const fs_operation_def_t zfs_evnodeops_template[] = {
5246         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5247         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5248         NULL,                   NULL
5249 };