1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2013 by Delphix. All rights reserved.
  24  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
  25  */
  26 
  27 /* Portions Copyright 2007 Jeremy Teo */
  28 /* Portions Copyright 2010 Robert Milkowski */
  29 
  30 #include <sys/types.h>
  31 #include <sys/param.h>
  32 #include <sys/time.h>
  33 #include <sys/systm.h>
  34 #include <sys/sysmacros.h>
  35 #include <sys/resource.h>
  36 #include <sys/vfs.h>
  37 #include <sys/vfs_opreg.h>
  38 #include <sys/vnode.h>
  39 #include <sys/file.h>
  40 #include <sys/stat.h>
  41 #include <sys/kmem.h>
  42 #include <sys/taskq.h>
  43 #include <sys/uio.h>
  44 #include <sys/vmsystm.h>
  45 #include <sys/atomic.h>
  46 #include <sys/vm.h>
  47 #include <vm/seg_vn.h>
  48 #include <vm/pvn.h>
  49 #include <vm/as.h>
  50 #include <vm/kpm.h>
  51 #include <vm/seg_kpm.h>
  52 #include <sys/mman.h>
  53 #include <sys/pathname.h>
  54 #include <sys/cmn_err.h>
  55 #include <sys/errno.h>
  56 #include <sys/unistd.h>
  57 #include <sys/zfs_dir.h>
  58 #include <sys/zfs_acl.h>
  59 #include <sys/zfs_ioctl.h>
  60 #include <sys/fs/zfs.h>
  61 #include <sys/dmu.h>
  62 #include <sys/dmu_objset.h>
  63 #include <sys/spa.h>
  64 #include <sys/txg.h>
  65 #include <sys/dbuf.h>
  66 #include <sys/zap.h>
  67 #include <sys/sa.h>
  68 #include <sys/dirent.h>
  69 #include <sys/policy.h>
  70 #include <sys/sunddi.h>
  71 #include <sys/filio.h>
  72 #include <sys/sid.h>
  73 #include "fs/fs_subr.h"
  74 #include <sys/zfs_ctldir.h>
  75 #include <sys/zfs_fuid.h>
  76 #include <sys/zfs_sa.h>
  77 #include <sys/dnlc.h>
  78 #include <sys/zfs_rlock.h>
  79 #include <sys/extdirent.h>
  80 #include <sys/kidmap.h>
  81 #include <sys/cred.h>
  82 #include <sys/attr.h>
  83 
  84 /*
  85  * Programming rules.
  86  *
  87  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
  88  * properly lock its in-core state, create a DMU transaction, do the work,
  89  * record this work in the intent log (ZIL), commit the DMU transaction,
  90  * and wait for the intent log to commit if it is a synchronous operation.
  91  * Moreover, the vnode ops must work in both normal and log replay context.
  92  * The ordering of events is important to avoid deadlocks and references
  93  * to freed memory.  The example below illustrates the following Big Rules:
  94  *
  95  *  (1) A check must be made in each zfs thread for a mounted file system.
  96  *      This is done avoiding races using ZFS_ENTER(zfsvfs).
  97  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
  98  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
  99  *      can return EIO from the calling function.
 100  *
 101  *  (2) VN_RELE() should always be the last thing except for zil_commit()
 102  *      (if necessary) and ZFS_EXIT(). This is for 3 reasons:
 103  *      First, if it's the last reference, the vnode/znode
 104  *      can be freed, so the zp may point to freed memory.  Second, the last
 105  *      reference will call zfs_zinactive(), which may induce a lot of work --
 106  *      pushing cached pages (which acquires range locks) and syncing out
 107  *      cached atime changes.  Third, zfs_zinactive() may require a new tx,
 108  *      which could deadlock the system if you were already holding one.
 109  *      If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
 110  *
 111  *  (3) All range locks must be grabbed before calling dmu_tx_assign(),
 112  *      as they can span dmu_tx_assign() calls.
 113  *
 114  *  (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
 115  *      This is critical because we don't want to block while holding locks.
 116  *      Note, in particular, that if a lock is sometimes acquired before
 117  *      the tx assigns, and sometimes after (e.g. z_lock), then failing to
 118  *      use a non-blocking assign can deadlock the system.  The scenario:
 119  *
 120  *      Thread A has grabbed a lock before calling dmu_tx_assign().
 121  *      Thread B is in an already-assigned tx, and blocks for this lock.
 122  *      Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
 123  *      forever, because the previous txg can't quiesce until B's tx commits.
 124  *
 125  *      If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
 126  *      then drop all locks, call dmu_tx_wait(), and try again.  On subsequent
 127  *      calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
 128  *      to indicate that this operation has already called dmu_tx_wait().
 129  *      This will ensure that we don't retry forever, waiting a short bit
 130  *      each time.
 131  *
 132  *  (5) If the operation succeeded, generate the intent log entry for it
 133  *      before dropping locks.  This ensures that the ordering of events
 134  *      in the intent log matches the order in which they actually occurred.
 135  *      During ZIL replay the zfs_log_* functions will update the sequence
 136  *      number to indicate the zil transaction has replayed.
 137  *
 138  *  (6) At the end of each vnode op, the DMU tx must always commit,
 139  *      regardless of whether there were any errors.
 140  *
 141  *  (7) After dropping all locks, invoke zil_commit(zilog, foid)
 142  *      to ensure that synchronous semantics are provided when necessary.
 143  *
 144  * In general, this is how things should be ordered in each vnode op:
 145  *
 146  *      ZFS_ENTER(zfsvfs);              // exit if unmounted
 147  * top:
 148  *      zfs_dirent_lock(&dl, ...)   // lock directory entry (may VN_HOLD())
 149  *      rw_enter(...);                  // grab any other locks you need
 150  *      tx = dmu_tx_create(...);        // get DMU tx
 151  *      dmu_tx_hold_*();                // hold each object you might modify
 152  *      error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
 153  *      if (error) {
 154  *              rw_exit(...);           // drop locks
 155  *              zfs_dirent_unlock(dl);  // unlock directory entry
 156  *              VN_RELE(...);           // release held vnodes
 157  *              if (error == ERESTART) {
 158  *                      waited = B_TRUE;
 159  *                      dmu_tx_wait(tx);
 160  *                      dmu_tx_abort(tx);
 161  *                      goto top;
 162  *              }
 163  *              dmu_tx_abort(tx);       // abort DMU tx
 164  *              ZFS_EXIT(zfsvfs);       // finished in zfs
 165  *              return (error);         // really out of space
 166  *      }
 167  *      error = do_real_work();         // do whatever this VOP does
 168  *      if (error == 0)
 169  *              zfs_log_*(...);         // on success, make ZIL entry
 170  *      dmu_tx_commit(tx);              // commit DMU tx -- error or not
 171  *      rw_exit(...);                   // drop locks
 172  *      zfs_dirent_unlock(dl);          // unlock directory entry
 173  *      VN_RELE(...);                   // release held vnodes
 174  *      zil_commit(zilog, foid);        // synchronous when necessary
 175  *      ZFS_EXIT(zfsvfs);               // finished in zfs
 176  *      return (error);                 // done, report error
 177  */
 178 
 179 /* ARGSUSED */
 180 static int
 181 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
 182 {
 183         znode_t *zp = VTOZ(*vpp);
 184         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 185 
 186         ZFS_ENTER(zfsvfs);
 187         ZFS_VERIFY_ZP(zp);
 188 
 189         if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
 190             ((flag & FAPPEND) == 0)) {
 191                 ZFS_EXIT(zfsvfs);
 192                 return (SET_ERROR(EPERM));
 193         }
 194 
 195         if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
 196             ZTOV(zp)->v_type == VREG &&
 197             !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
 198                 if (fs_vscan(*vpp, cr, 0) != 0) {
 199                         ZFS_EXIT(zfsvfs);
 200                         return (SET_ERROR(EACCES));
 201                 }
 202         }
 203 
 204         /* Keep a count of the synchronous opens in the znode */
 205         if (flag & (FSYNC | FDSYNC))
 206                 atomic_inc_32(&zp->z_sync_cnt);
 207 
 208         ZFS_EXIT(zfsvfs);
 209         return (0);
 210 }
 211 
 212 /* ARGSUSED */
 213 static int
 214 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
 215     caller_context_t *ct)
 216 {
 217         znode_t *zp = VTOZ(vp);
 218         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 219 
 220         /*
 221          * Clean up any locks held by this process on the vp.
 222          */
 223         cleanlocks(vp, ddi_get_pid(), 0);
 224         cleanshares(vp, ddi_get_pid());
 225 
 226         ZFS_ENTER(zfsvfs);
 227         ZFS_VERIFY_ZP(zp);
 228 
 229         /* Decrement the synchronous opens in the znode */
 230         if ((flag & (FSYNC | FDSYNC)) && (count == 1))
 231                 atomic_dec_32(&zp->z_sync_cnt);
 232 
 233         if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
 234             ZTOV(zp)->v_type == VREG &&
 235             !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
 236                 VERIFY(fs_vscan(vp, cr, 1) == 0);
 237 
 238         ZFS_EXIT(zfsvfs);
 239         return (0);
 240 }
 241 
 242 /*
 243  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
 244  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
 245  */
 246 static int
 247 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
 248 {
 249         znode_t *zp = VTOZ(vp);
 250         uint64_t noff = (uint64_t)*off; /* new offset */
 251         uint64_t file_sz;
 252         int error;
 253         boolean_t hole;
 254 
 255         file_sz = zp->z_size;
 256         if (noff >= file_sz)  {
 257                 return (SET_ERROR(ENXIO));
 258         }
 259 
 260         if (cmd == _FIO_SEEK_HOLE)
 261                 hole = B_TRUE;
 262         else
 263                 hole = B_FALSE;
 264 
 265         error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
 266 
 267         /* end of file? */
 268         if ((error == ESRCH) || (noff > file_sz)) {
 269                 /*
 270                  * Handle the virtual hole at the end of file.
 271                  */
 272                 if (hole) {
 273                         *off = file_sz;
 274                         return (0);
 275                 }
 276                 return (SET_ERROR(ENXIO));
 277         }
 278 
 279         if (noff < *off)
 280                 return (error);
 281         *off = noff;
 282         return (error);
 283 }
 284 
 285 /* ARGSUSED */
 286 static int
 287 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
 288     int *rvalp, caller_context_t *ct)
 289 {
 290         offset_t off;
 291         int error;
 292         zfsvfs_t *zfsvfs;
 293         znode_t *zp;
 294 
 295         switch (com) {
 296         case _FIOFFS:
 297                 return (zfs_sync(vp->v_vfsp, 0, cred));
 298 
 299                 /*
 300                  * The following two ioctls are used by bfu.  Faking out,
 301                  * necessary to avoid bfu errors.
 302                  */
 303         case _FIOGDIO:
 304         case _FIOSDIO:
 305                 return (0);
 306 
 307         case _FIO_SEEK_DATA:
 308         case _FIO_SEEK_HOLE:
 309                 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
 310                         return (SET_ERROR(EFAULT));
 311 
 312                 zp = VTOZ(vp);
 313                 zfsvfs = zp->z_zfsvfs;
 314                 ZFS_ENTER(zfsvfs);
 315                 ZFS_VERIFY_ZP(zp);
 316 
 317                 /* offset parameter is in/out */
 318                 error = zfs_holey(vp, com, &off);
 319                 ZFS_EXIT(zfsvfs);
 320                 if (error)
 321                         return (error);
 322                 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
 323                         return (SET_ERROR(EFAULT));
 324                 return (0);
 325         }
 326         return (SET_ERROR(ENOTTY));
 327 }
 328 
 329 /*
 330  * Utility functions to map and unmap a single physical page.  These
 331  * are used to manage the mappable copies of ZFS file data, and therefore
 332  * do not update ref/mod bits.
 333  */
 334 caddr_t
 335 zfs_map_page(page_t *pp, enum seg_rw rw)
 336 {
 337         if (kpm_enable)
 338                 return (hat_kpm_mapin(pp, 0));
 339         ASSERT(rw == S_READ || rw == S_WRITE);
 340         return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
 341             (caddr_t)-1));
 342 }
 343 
 344 void
 345 zfs_unmap_page(page_t *pp, caddr_t addr)
 346 {
 347         if (kpm_enable) {
 348                 hat_kpm_mapout(pp, 0, addr);
 349         } else {
 350                 ppmapout(addr);
 351         }
 352 }
 353 
 354 /*
 355  * When a file is memory mapped, we must keep the IO data synchronized
 356  * between the DMU cache and the memory mapped pages.  What this means:
 357  *
 358  * On Write:    If we find a memory mapped page, we write to *both*
 359  *              the page and the dmu buffer.
 360  */
 361 static void
 362 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
 363 {
 364         int64_t off;
 365 
 366         off = start & PAGEOFFSET;
 367         for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
 368                 page_t *pp;
 369                 uint64_t nbytes = MIN(PAGESIZE - off, len);
 370 
 371                 if (pp = page_lookup(vp, start, SE_SHARED)) {
 372                         caddr_t va;
 373 
 374                         va = zfs_map_page(pp, S_WRITE);
 375                         (void) dmu_read(os, oid, start+off, nbytes, va+off,
 376                             DMU_READ_PREFETCH);
 377                         zfs_unmap_page(pp, va);
 378                         page_unlock(pp);
 379                 }
 380                 len -= nbytes;
 381                 off = 0;
 382         }
 383 }
 384 
 385 /*
 386  * When a file is memory mapped, we must keep the IO data synchronized
 387  * between the DMU cache and the memory mapped pages.  What this means:
 388  *
 389  * On Read:     We "read" preferentially from memory mapped pages,
 390  *              else we default from the dmu buffer.
 391  *
 392  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
 393  *       the file is memory mapped.
 394  */
 395 static int
 396 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
 397 {
 398         znode_t *zp = VTOZ(vp);
 399         objset_t *os = zp->z_zfsvfs->z_os;
 400         int64_t start, off;
 401         int len = nbytes;
 402         int error = 0;
 403 
 404         start = uio->uio_loffset;
 405         off = start & PAGEOFFSET;
 406         for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
 407                 page_t *pp;
 408                 uint64_t bytes = MIN(PAGESIZE - off, len);
 409 
 410                 if (pp = page_lookup(vp, start, SE_SHARED)) {
 411                         caddr_t va;
 412 
 413                         va = zfs_map_page(pp, S_READ);
 414                         error = uiomove(va + off, bytes, UIO_READ, uio);
 415                         zfs_unmap_page(pp, va);
 416                         page_unlock(pp);
 417                 } else {
 418                         error = dmu_read_uio(os, zp->z_id, uio, bytes);
 419                 }
 420                 len -= bytes;
 421                 off = 0;
 422                 if (error)
 423                         break;
 424         }
 425         return (error);
 426 }
 427 
 428 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
 429 
 430 /*
 431  * Read bytes from specified file into supplied buffer.
 432  *
 433  *      IN:     vp      - vnode of file to be read from.
 434  *              uio     - structure supplying read location, range info,
 435  *                        and return buffer.
 436  *              ioflag  - SYNC flags; used to provide FRSYNC semantics.
 437  *              cr      - credentials of caller.
 438  *              ct      - caller context
 439  *
 440  *      OUT:    uio     - updated offset and range, buffer filled.
 441  *
 442  *      RETURN: 0 on success, error code on failure.
 443  *
 444  * Side Effects:
 445  *      vp - atime updated if byte count > 0
 446  */
 447 /* ARGSUSED */
 448 static int
 449 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 450 {
 451         znode_t         *zp = VTOZ(vp);
 452         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
 453         objset_t        *os;
 454         ssize_t         n, nbytes;
 455         int             error = 0;
 456         rl_t            *rl;
 457         xuio_t          *xuio = NULL;
 458 
 459         ZFS_ENTER(zfsvfs);
 460         ZFS_VERIFY_ZP(zp);
 461         os = zfsvfs->z_os;
 462 
 463         if (zp->z_pflags & ZFS_AV_QUARANTINED) {
 464                 ZFS_EXIT(zfsvfs);
 465                 return (SET_ERROR(EACCES));
 466         }
 467 
 468         /*
 469          * Validate file offset
 470          */
 471         if (uio->uio_loffset < (offset_t)0) {
 472                 ZFS_EXIT(zfsvfs);
 473                 return (SET_ERROR(EINVAL));
 474         }
 475 
 476         /*
 477          * Fasttrack empty reads
 478          */
 479         if (uio->uio_resid == 0) {
 480                 ZFS_EXIT(zfsvfs);
 481                 return (0);
 482         }
 483 
 484         /*
 485          * Check for mandatory locks
 486          */
 487         if (MANDMODE(zp->z_mode)) {
 488                 if (error = chklock(vp, FREAD,
 489                     uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
 490                         ZFS_EXIT(zfsvfs);
 491                         return (error);
 492                 }
 493         }
 494 
 495         /*
 496          * If we're in FRSYNC mode, sync out this znode before reading it.
 497          */
 498         if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
 499                 zil_commit(zfsvfs->z_log, zp->z_id);
 500 
 501         /*
 502          * Lock the range against changes.
 503          */
 504         rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
 505 
 506         /*
 507          * If we are reading past end-of-file we can skip
 508          * to the end; but we might still need to set atime.
 509          */
 510         if (uio->uio_loffset >= zp->z_size) {
 511                 error = 0;
 512                 goto out;
 513         }
 514 
 515         ASSERT(uio->uio_loffset < zp->z_size);
 516         n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
 517 
 518         if ((uio->uio_extflg == UIO_XUIO) &&
 519             (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
 520                 int nblk;
 521                 int blksz = zp->z_blksz;
 522                 uint64_t offset = uio->uio_loffset;
 523 
 524                 xuio = (xuio_t *)uio;
 525                 if ((ISP2(blksz))) {
 526                         nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
 527                             blksz)) / blksz;
 528                 } else {
 529                         ASSERT(offset + n <= blksz);
 530                         nblk = 1;
 531                 }
 532                 (void) dmu_xuio_init(xuio, nblk);
 533 
 534                 if (vn_has_cached_data(vp)) {
 535                         /*
 536                          * For simplicity, we always allocate a full buffer
 537                          * even if we only expect to read a portion of a block.
 538                          */
 539                         while (--nblk >= 0) {
 540                                 (void) dmu_xuio_add(xuio,
 541                                     dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
 542                                     blksz), 0, blksz);
 543                         }
 544                 }
 545         }
 546 
 547         while (n > 0) {
 548                 nbytes = MIN(n, zfs_read_chunk_size -
 549                     P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
 550 
 551                 if (vn_has_cached_data(vp))
 552                         error = mappedread(vp, nbytes, uio);
 553                 else
 554                         error = dmu_read_uio(os, zp->z_id, uio, nbytes);
 555                 if (error) {
 556                         /* convert checksum errors into IO errors */
 557                         if (error == ECKSUM)
 558                                 error = SET_ERROR(EIO);
 559                         break;
 560                 }
 561 
 562                 n -= nbytes;
 563         }
 564 out:
 565         zfs_range_unlock(rl);
 566 
 567         ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
 568         ZFS_EXIT(zfsvfs);
 569         return (error);
 570 }
 571 
 572 /*
 573  * Write the bytes to a file.
 574  *
 575  *      IN:     vp      - vnode of file to be written to.
 576  *              uio     - structure supplying write location, range info,
 577  *                        and data buffer.
 578  *              ioflag  - FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
 579  *                        set if in append mode.
 580  *              cr      - credentials of caller.
 581  *              ct      - caller context (NFS/CIFS fem monitor only)
 582  *
 583  *      OUT:    uio     - updated offset and range.
 584  *
 585  *      RETURN: 0 on success, error code on failure.
 586  *
 587  * Timestamps:
 588  *      vp - ctime|mtime updated if byte count > 0
 589  */
 590 
 591 /* ARGSUSED */
 592 static int
 593 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 594 {
 595         znode_t         *zp = VTOZ(vp);
 596         rlim64_t        limit = uio->uio_llimit;
 597         ssize_t         start_resid = uio->uio_resid;
 598         ssize_t         tx_bytes;
 599         uint64_t        end_size;
 600         dmu_tx_t        *tx;
 601         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
 602         zilog_t         *zilog;
 603         offset_t        woff;
 604         ssize_t         n, nbytes;
 605         rl_t            *rl;
 606         int             max_blksz = zfsvfs->z_max_blksz;
 607         int             error = 0;
 608         arc_buf_t       *abuf;
 609         iovec_t         *aiov = NULL;
 610         xuio_t          *xuio = NULL;
 611         int             i_iov = 0;
 612         int             iovcnt = uio->uio_iovcnt;
 613         iovec_t         *iovp = uio->uio_iov;
 614         int             write_eof;
 615         int             count = 0;
 616         sa_bulk_attr_t  bulk[4];
 617         uint64_t        mtime[2], ctime[2];
 618 
 619         /*
 620          * Fasttrack empty write
 621          */
 622         n = start_resid;
 623         if (n == 0)
 624                 return (0);
 625 
 626         if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
 627                 limit = MAXOFFSET_T;
 628 
 629         ZFS_ENTER(zfsvfs);
 630         ZFS_VERIFY_ZP(zp);
 631 
 632         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
 633         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
 634         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
 635             &zp->z_size, 8);
 636         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
 637             &zp->z_pflags, 8);
 638 
 639         /*
 640          * If immutable or not appending then return EPERM
 641          */
 642         if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
 643             ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
 644             (uio->uio_loffset < zp->z_size))) {
 645                 ZFS_EXIT(zfsvfs);
 646                 return (SET_ERROR(EPERM));
 647         }
 648 
 649         zilog = zfsvfs->z_log;
 650 
 651         /*
 652          * Validate file offset
 653          */
 654         woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
 655         if (woff < 0) {
 656                 ZFS_EXIT(zfsvfs);
 657                 return (SET_ERROR(EINVAL));
 658         }
 659 
 660         /*
 661          * Check for mandatory locks before calling zfs_range_lock()
 662          * in order to prevent a deadlock with locks set via fcntl().
 663          */
 664         if (MANDMODE((mode_t)zp->z_mode) &&
 665             (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
 666                 ZFS_EXIT(zfsvfs);
 667                 return (error);
 668         }
 669 
 670         /*
 671          * Pre-fault the pages to ensure slow (eg NFS) pages
 672          * don't hold up txg.
 673          * Skip this if uio contains loaned arc_buf.
 674          */
 675         if ((uio->uio_extflg == UIO_XUIO) &&
 676             (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
 677                 xuio = (xuio_t *)uio;
 678         else
 679                 uio_prefaultpages(MIN(n, max_blksz), uio);
 680 
 681         /*
 682          * If in append mode, set the io offset pointer to eof.
 683          */
 684         if (ioflag & FAPPEND) {
 685                 /*
 686                  * Obtain an appending range lock to guarantee file append
 687                  * semantics.  We reset the write offset once we have the lock.
 688                  */
 689                 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
 690                 woff = rl->r_off;
 691                 if (rl->r_len == UINT64_MAX) {
 692                         /*
 693                          * We overlocked the file because this write will cause
 694                          * the file block size to increase.
 695                          * Note that zp_size cannot change with this lock held.
 696                          */
 697                         woff = zp->z_size;
 698                 }
 699                 uio->uio_loffset = woff;
 700         } else {
 701                 /*
 702                  * Note that if the file block size will change as a result of
 703                  * this write, then this range lock will lock the entire file
 704                  * so that we can re-write the block safely.
 705                  */
 706                 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
 707         }
 708 
 709         if (woff >= limit) {
 710                 zfs_range_unlock(rl);
 711                 ZFS_EXIT(zfsvfs);
 712                 return (SET_ERROR(EFBIG));
 713         }
 714 
 715         if ((woff + n) > limit || woff > (limit - n))
 716                 n = limit - woff;
 717 
 718         /* Will this write extend the file length? */
 719         write_eof = (woff + n > zp->z_size);
 720 
 721         end_size = MAX(zp->z_size, woff + n);
 722 
 723         /*
 724          * Write the file in reasonable size chunks.  Each chunk is written
 725          * in a separate transaction; this keeps the intent log records small
 726          * and allows us to do more fine-grained space accounting.
 727          */
 728         while (n > 0) {
 729                 abuf = NULL;
 730                 woff = uio->uio_loffset;
 731 again:
 732                 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
 733                     zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
 734                         if (abuf != NULL)
 735                                 dmu_return_arcbuf(abuf);
 736                         error = SET_ERROR(EDQUOT);
 737                         break;
 738                 }
 739 
 740                 if (xuio && abuf == NULL) {
 741                         ASSERT(i_iov < iovcnt);
 742                         aiov = &iovp[i_iov];
 743                         abuf = dmu_xuio_arcbuf(xuio, i_iov);
 744                         dmu_xuio_clear(xuio, i_iov);
 745                         DTRACE_PROBE3(zfs_cp_write, int, i_iov,
 746                             iovec_t *, aiov, arc_buf_t *, abuf);
 747                         ASSERT((aiov->iov_base == abuf->b_data) ||
 748                             ((char *)aiov->iov_base - (char *)abuf->b_data +
 749                             aiov->iov_len == arc_buf_size(abuf)));
 750                         i_iov++;
 751                 } else if (abuf == NULL && n >= max_blksz &&
 752                     woff >= zp->z_size &&
 753                     P2PHASE(woff, max_blksz) == 0 &&
 754                     zp->z_blksz == max_blksz) {
 755                         /*
 756                          * This write covers a full block.  "Borrow" a buffer
 757                          * from the dmu so that we can fill it before we enter
 758                          * a transaction.  This avoids the possibility of
 759                          * holding up the transaction if the data copy hangs
 760                          * up on a pagefault (e.g., from an NFS server mapping).
 761                          */
 762                         size_t cbytes;
 763 
 764                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
 765                             max_blksz);
 766                         ASSERT(abuf != NULL);
 767                         ASSERT(arc_buf_size(abuf) == max_blksz);
 768                         if (error = uiocopy(abuf->b_data, max_blksz,
 769                             UIO_WRITE, uio, &cbytes)) {
 770                                 dmu_return_arcbuf(abuf);
 771                                 break;
 772                         }
 773                         ASSERT(cbytes == max_blksz);
 774                 }
 775 
 776                 /*
 777                  * Start a transaction.
 778                  */
 779                 tx = dmu_tx_create(zfsvfs->z_os);
 780                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 781                 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
 782                 zfs_sa_upgrade_txholds(tx, zp);
 783                 error = dmu_tx_assign(tx, TXG_NOWAIT);
 784                 if (error) {
 785                         if (error == ERESTART) {
 786                                 dmu_tx_wait(tx);
 787                                 dmu_tx_abort(tx);
 788                                 goto again;
 789                         }
 790                         dmu_tx_abort(tx);
 791                         if (abuf != NULL)
 792                                 dmu_return_arcbuf(abuf);
 793                         break;
 794                 }
 795 
 796                 /*
 797                  * If zfs_range_lock() over-locked we grow the blocksize
 798                  * and then reduce the lock range.  This will only happen
 799                  * on the first iteration since zfs_range_reduce() will
 800                  * shrink down r_len to the appropriate size.
 801                  */
 802                 if (rl->r_len == UINT64_MAX) {
 803                         uint64_t new_blksz;
 804 
 805                         if (zp->z_blksz > max_blksz) {
 806                                 ASSERT(!ISP2(zp->z_blksz));
 807                                 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
 808                         } else {
 809                                 new_blksz = MIN(end_size, max_blksz);
 810                         }
 811                         zfs_grow_blocksize(zp, new_blksz, tx);
 812                         zfs_range_reduce(rl, woff, n);
 813                 }
 814 
 815                 /*
 816                  * XXX - should we really limit each write to z_max_blksz?
 817                  * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
 818                  */
 819                 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
 820 
 821                 if (abuf == NULL) {
 822                         tx_bytes = uio->uio_resid;
 823                         error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
 824                             uio, nbytes, tx);
 825                         tx_bytes -= uio->uio_resid;
 826                 } else {
 827                         tx_bytes = nbytes;
 828                         ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
 829                         /*
 830                          * If this is not a full block write, but we are
 831                          * extending the file past EOF and this data starts
 832                          * block-aligned, use assign_arcbuf().  Otherwise,
 833                          * write via dmu_write().
 834                          */
 835                         if (tx_bytes < max_blksz && (!write_eof ||
 836                             aiov->iov_base != abuf->b_data)) {
 837                                 ASSERT(xuio);
 838                                 dmu_write(zfsvfs->z_os, zp->z_id, woff,
 839                                     aiov->iov_len, aiov->iov_base, tx);
 840                                 dmu_return_arcbuf(abuf);
 841                                 xuio_stat_wbuf_copied();
 842                         } else {
 843                                 ASSERT(xuio || tx_bytes == max_blksz);
 844                                 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
 845                                     woff, abuf, tx);
 846                         }
 847                         ASSERT(tx_bytes <= uio->uio_resid);
 848                         uioskip(uio, tx_bytes);
 849                 }
 850                 if (tx_bytes && vn_has_cached_data(vp)) {
 851                         update_pages(vp, woff,
 852                             tx_bytes, zfsvfs->z_os, zp->z_id);
 853                 }
 854 
 855                 /*
 856                  * If we made no progress, we're done.  If we made even
 857                  * partial progress, update the znode and ZIL accordingly.
 858                  */
 859                 if (tx_bytes == 0) {
 860                         (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
 861                             (void *)&zp->z_size, sizeof (uint64_t), tx);
 862                         dmu_tx_commit(tx);
 863                         ASSERT(error != 0);
 864                         break;
 865                 }
 866 
 867                 /*
 868                  * Clear Set-UID/Set-GID bits on successful write if not
 869                  * privileged and at least one of the excute bits is set.
 870                  *
 871                  * It would be nice to to this after all writes have
 872                  * been done, but that would still expose the ISUID/ISGID
 873                  * to another app after the partial write is committed.
 874                  *
 875                  * Note: we don't call zfs_fuid_map_id() here because
 876                  * user 0 is not an ephemeral uid.
 877                  */
 878                 mutex_enter(&zp->z_acl_lock);
 879                 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
 880                     (S_IXUSR >> 6))) != 0 &&
 881                     (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
 882                     secpolicy_vnode_setid_retain(cr,
 883                     (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
 884                         uint64_t newmode;
 885                         zp->z_mode &= ~(S_ISUID | S_ISGID);
 886                         newmode = zp->z_mode;
 887                         (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
 888                             (void *)&newmode, sizeof (uint64_t), tx);
 889                 }
 890                 mutex_exit(&zp->z_acl_lock);
 891 
 892                 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
 893                     B_TRUE);
 894 
 895                 /*
 896                  * Update the file size (zp_size) if it has changed;
 897                  * account for possible concurrent updates.
 898                  */
 899                 while ((end_size = zp->z_size) < uio->uio_loffset) {
 900                         (void) atomic_cas_64(&zp->z_size, end_size,
 901                             uio->uio_loffset);
 902                         ASSERT(error == 0);
 903                 }
 904                 /*
 905                  * If we are replaying and eof is non zero then force
 906                  * the file size to the specified eof. Note, there's no
 907                  * concurrency during replay.
 908                  */
 909                 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
 910                         zp->z_size = zfsvfs->z_replay_eof;
 911 
 912                 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
 913 
 914                 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
 915                 dmu_tx_commit(tx);
 916 
 917                 if (error != 0)
 918                         break;
 919                 ASSERT(tx_bytes == nbytes);
 920                 n -= nbytes;
 921 
 922                 if (!xuio && n > 0)
 923                         uio_prefaultpages(MIN(n, max_blksz), uio);
 924         }
 925 
 926         zfs_range_unlock(rl);
 927 
 928         /*
 929          * If we're in replay mode, or we made no progress, return error.
 930          * Otherwise, it's at least a partial write, so it's successful.
 931          */
 932         if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
 933                 ZFS_EXIT(zfsvfs);
 934                 return (error);
 935         }
 936 
 937         if (ioflag & (FSYNC | FDSYNC) ||
 938             zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
 939                 zil_commit(zilog, zp->z_id);
 940 
 941         ZFS_EXIT(zfsvfs);
 942         return (0);
 943 }
 944 
 945 void
 946 zfs_get_done(zgd_t *zgd, int error)
 947 {
 948         znode_t *zp = zgd->zgd_private;
 949         objset_t *os = zp->z_zfsvfs->z_os;
 950 
 951         if (zgd->zgd_db)
 952                 dmu_buf_rele(zgd->zgd_db, zgd);
 953 
 954         zfs_range_unlock(zgd->zgd_rl);
 955 
 956         /*
 957          * Release the vnode asynchronously as we currently have the
 958          * txg stopped from syncing.
 959          */
 960         VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
 961 
 962         if (error == 0 && zgd->zgd_bp)
 963                 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
 964 
 965         kmem_free(zgd, sizeof (zgd_t));
 966 }
 967 
 968 #ifdef DEBUG
 969 static int zil_fault_io = 0;
 970 #endif
 971 
 972 /*
 973  * Get data to generate a TX_WRITE intent log record.
 974  */
 975 int
 976 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
 977 {
 978         zfsvfs_t *zfsvfs = arg;
 979         objset_t *os = zfsvfs->z_os;
 980         znode_t *zp;
 981         uint64_t object = lr->lr_foid;
 982         uint64_t offset = lr->lr_offset;
 983         uint64_t size = lr->lr_length;
 984         blkptr_t *bp = &lr->lr_blkptr;
 985         dmu_buf_t *db;
 986         zgd_t *zgd;
 987         int error = 0;
 988 
 989         ASSERT(zio != NULL);
 990         ASSERT(size != 0);
 991 
 992         /*
 993          * Nothing to do if the file has been removed
 994          */
 995         if (zfs_zget(zfsvfs, object, &zp) != 0)
 996                 return (SET_ERROR(ENOENT));
 997         if (zp->z_unlinked) {
 998                 /*
 999                  * Release the vnode asynchronously as we currently have the
1000                  * txg stopped from syncing.
1001                  */
1002                 VN_RELE_ASYNC(ZTOV(zp),
1003                     dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1004                 return (SET_ERROR(ENOENT));
1005         }
1006 
1007         zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1008         zgd->zgd_zilog = zfsvfs->z_log;
1009         zgd->zgd_private = zp;
1010 
1011         /*
1012          * Write records come in two flavors: immediate and indirect.
1013          * For small writes it's cheaper to store the data with the
1014          * log record (immediate); for large writes it's cheaper to
1015          * sync the data and get a pointer to it (indirect) so that
1016          * we don't have to write the data twice.
1017          */
1018         if (buf != NULL) { /* immediate write */
1019                 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1020                 /* test for truncation needs to be done while range locked */
1021                 if (offset >= zp->z_size) {
1022                         error = SET_ERROR(ENOENT);
1023                 } else {
1024                         error = dmu_read(os, object, offset, size, buf,
1025                             DMU_READ_NO_PREFETCH);
1026                 }
1027                 ASSERT(error == 0 || error == ENOENT);
1028         } else { /* indirect write */
1029                 /*
1030                  * Have to lock the whole block to ensure when it's
1031                  * written out and it's checksum is being calculated
1032                  * that no one can change the data. We need to re-check
1033                  * blocksize after we get the lock in case it's changed!
1034                  */
1035                 for (;;) {
1036                         uint64_t blkoff;
1037                         size = zp->z_blksz;
1038                         blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1039                         offset -= blkoff;
1040                         zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1041                             RL_READER);
1042                         if (zp->z_blksz == size)
1043                                 break;
1044                         offset += blkoff;
1045                         zfs_range_unlock(zgd->zgd_rl);
1046                 }
1047                 /* test for truncation needs to be done while range locked */
1048                 if (lr->lr_offset >= zp->z_size)
1049                         error = SET_ERROR(ENOENT);
1050 #ifdef DEBUG
1051                 if (zil_fault_io) {
1052                         error = SET_ERROR(EIO);
1053                         zil_fault_io = 0;
1054                 }
1055 #endif
1056                 if (error == 0)
1057                         error = dmu_buf_hold(os, object, offset, zgd, &db,
1058                             DMU_READ_NO_PREFETCH);
1059 
1060                 if (error == 0) {
1061                         blkptr_t *obp = dmu_buf_get_blkptr(db);
1062                         if (obp) {
1063                                 ASSERT(BP_IS_HOLE(bp));
1064                                 *bp = *obp;
1065                         }
1066 
1067                         zgd->zgd_db = db;
1068                         zgd->zgd_bp = bp;
1069 
1070                         ASSERT(db->db_offset == offset);
1071                         ASSERT(db->db_size == size);
1072 
1073                         error = dmu_sync(zio, lr->lr_common.lrc_txg,
1074                             zfs_get_done, zgd);
1075                         ASSERT(error || lr->lr_length <= zp->z_blksz);
1076 
1077                         /*
1078                          * On success, we need to wait for the write I/O
1079                          * initiated by dmu_sync() to complete before we can
1080                          * release this dbuf.  We will finish everything up
1081                          * in the zfs_get_done() callback.
1082                          */
1083                         if (error == 0)
1084                                 return (0);
1085 
1086                         if (error == EALREADY) {
1087                                 lr->lr_common.lrc_txtype = TX_WRITE2;
1088                                 error = 0;
1089                         }
1090                 }
1091         }
1092 
1093         zfs_get_done(zgd, error);
1094 
1095         return (error);
1096 }
1097 
1098 /*ARGSUSED*/
1099 static int
1100 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1101     caller_context_t *ct)
1102 {
1103         znode_t *zp = VTOZ(vp);
1104         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1105         int error;
1106 
1107         ZFS_ENTER(zfsvfs);
1108         ZFS_VERIFY_ZP(zp);
1109 
1110         if (flag & V_ACE_MASK)
1111                 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1112         else
1113                 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1114 
1115         ZFS_EXIT(zfsvfs);
1116         return (error);
1117 }
1118 
1119 /*
1120  * If vnode is for a device return a specfs vnode instead.
1121  */
1122 static int
1123 specvp_check(vnode_t **vpp, cred_t *cr)
1124 {
1125         int error = 0;
1126 
1127         if (IS_DEVVP(*vpp)) {
1128                 struct vnode *svp;
1129 
1130                 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1131                 VN_RELE(*vpp);
1132                 if (svp == NULL)
1133                         error = SET_ERROR(ENOSYS);
1134                 *vpp = svp;
1135         }
1136         return (error);
1137 }
1138 
1139 
1140 /*
1141  * Lookup an entry in a directory, or an extended attribute directory.
1142  * If it exists, return a held vnode reference for it.
1143  *
1144  *      IN:     dvp     - vnode of directory to search.
1145  *              nm      - name of entry to lookup.
1146  *              pnp     - full pathname to lookup [UNUSED].
1147  *              flags   - LOOKUP_XATTR set if looking for an attribute.
1148  *              rdir    - root directory vnode [UNUSED].
1149  *              cr      - credentials of caller.
1150  *              ct      - caller context
1151  *              direntflags - directory lookup flags
1152  *              realpnp - returned pathname.
1153  *
1154  *      OUT:    vpp     - vnode of located entry, NULL if not found.
1155  *
1156  *      RETURN: 0 on success, error code on failure.
1157  *
1158  * Timestamps:
1159  *      NA
1160  */
1161 /* ARGSUSED */
1162 static int
1163 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1164     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1165     int *direntflags, pathname_t *realpnp)
1166 {
1167         znode_t *zdp = VTOZ(dvp);
1168         zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1169         int     error = 0;
1170 
1171         /* fast path */
1172         if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1173 
1174                 if (dvp->v_type != VDIR) {
1175                         return (SET_ERROR(ENOTDIR));
1176                 } else if (zdp->z_sa_hdl == NULL) {
1177                         return (SET_ERROR(EIO));
1178                 }
1179 
1180                 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1181                         error = zfs_fastaccesschk_execute(zdp, cr);
1182                         if (!error) {
1183                                 *vpp = dvp;
1184                                 VN_HOLD(*vpp);
1185                                 return (0);
1186                         }
1187                         return (error);
1188                 } else {
1189                         vnode_t *tvp = dnlc_lookup(dvp, nm);
1190 
1191                         if (tvp) {
1192                                 error = zfs_fastaccesschk_execute(zdp, cr);
1193                                 if (error) {
1194                                         VN_RELE(tvp);
1195                                         return (error);
1196                                 }
1197                                 if (tvp == DNLC_NO_VNODE) {
1198                                         VN_RELE(tvp);
1199                                         return (SET_ERROR(ENOENT));
1200                                 } else {
1201                                         *vpp = tvp;
1202                                         return (specvp_check(vpp, cr));
1203                                 }
1204                         }
1205                 }
1206         }
1207 
1208         DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1209 
1210         ZFS_ENTER(zfsvfs);
1211         ZFS_VERIFY_ZP(zdp);
1212 
1213         *vpp = NULL;
1214 
1215         if (flags & LOOKUP_XATTR) {
1216                 /*
1217                  * If the xattr property is off, refuse the lookup request.
1218                  */
1219                 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1220                         ZFS_EXIT(zfsvfs);
1221                         return (SET_ERROR(EINVAL));
1222                 }
1223 
1224                 /*
1225                  * We don't allow recursive attributes..
1226                  * Maybe someday we will.
1227                  */
1228                 if (zdp->z_pflags & ZFS_XATTR) {
1229                         ZFS_EXIT(zfsvfs);
1230                         return (SET_ERROR(EINVAL));
1231                 }
1232 
1233                 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1234                         ZFS_EXIT(zfsvfs);
1235                         return (error);
1236                 }
1237 
1238                 /*
1239                  * Do we have permission to get into attribute directory?
1240                  */
1241 
1242                 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1243                     B_FALSE, cr)) {
1244                         VN_RELE(*vpp);
1245                         *vpp = NULL;
1246                 }
1247 
1248                 ZFS_EXIT(zfsvfs);
1249                 return (error);
1250         }
1251 
1252         if (dvp->v_type != VDIR) {
1253                 ZFS_EXIT(zfsvfs);
1254                 return (SET_ERROR(ENOTDIR));
1255         }
1256 
1257         /*
1258          * Check accessibility of directory.
1259          */
1260 
1261         if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1262                 ZFS_EXIT(zfsvfs);
1263                 return (error);
1264         }
1265 
1266         if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1267             NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1268                 ZFS_EXIT(zfsvfs);
1269                 return (SET_ERROR(EILSEQ));
1270         }
1271 
1272         error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1273         if (error == 0)
1274                 error = specvp_check(vpp, cr);
1275 
1276         ZFS_EXIT(zfsvfs);
1277         return (error);
1278 }
1279 
1280 /*
1281  * Attempt to create a new entry in a directory.  If the entry
1282  * already exists, truncate the file if permissible, else return
1283  * an error.  Return the vp of the created or trunc'd file.
1284  *
1285  *      IN:     dvp     - vnode of directory to put new file entry in.
1286  *              name    - name of new file entry.
1287  *              vap     - attributes of new file.
1288  *              excl    - flag indicating exclusive or non-exclusive mode.
1289  *              mode    - mode to open file with.
1290  *              cr      - credentials of caller.
1291  *              flag    - large file flag [UNUSED].
1292  *              ct      - caller context
1293  *              vsecp   - ACL to be set
1294  *
1295  *      OUT:    vpp     - vnode of created or trunc'd entry.
1296  *
1297  *      RETURN: 0 on success, error code on failure.
1298  *
1299  * Timestamps:
1300  *      dvp - ctime|mtime updated if new entry created
1301  *       vp - ctime|mtime always, atime if new
1302  */
1303 
1304 /* ARGSUSED */
1305 static int
1306 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1307     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1308     vsecattr_t *vsecp)
1309 {
1310         znode_t         *zp, *dzp = VTOZ(dvp);
1311         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1312         zilog_t         *zilog;
1313         objset_t        *os;
1314         zfs_dirlock_t   *dl;
1315         dmu_tx_t        *tx;
1316         int             error;
1317         ksid_t          *ksid;
1318         uid_t           uid;
1319         gid_t           gid = crgetgid(cr);
1320         zfs_acl_ids_t   acl_ids;
1321         boolean_t       fuid_dirtied;
1322         boolean_t       have_acl = B_FALSE;
1323         boolean_t       waited = B_FALSE;
1324 
1325         /*
1326          * If we have an ephemeral id, ACL, or XVATTR then
1327          * make sure file system is at proper version
1328          */
1329 
1330         ksid = crgetsid(cr, KSID_OWNER);
1331         if (ksid)
1332                 uid = ksid_getid(ksid);
1333         else
1334                 uid = crgetuid(cr);
1335 
1336         if (zfsvfs->z_use_fuids == B_FALSE &&
1337             (vsecp || (vap->va_mask & AT_XVATTR) ||
1338             IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1339                 return (SET_ERROR(EINVAL));
1340 
1341         ZFS_ENTER(zfsvfs);
1342         ZFS_VERIFY_ZP(dzp);
1343         os = zfsvfs->z_os;
1344         zilog = zfsvfs->z_log;
1345 
1346         if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1347             NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1348                 ZFS_EXIT(zfsvfs);
1349                 return (SET_ERROR(EILSEQ));
1350         }
1351 
1352         if (vap->va_mask & AT_XVATTR) {
1353                 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1354                     crgetuid(cr), cr, vap->va_type)) != 0) {
1355                         ZFS_EXIT(zfsvfs);
1356                         return (error);
1357                 }
1358         }
1359 top:
1360         *vpp = NULL;
1361 
1362         if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1363                 vap->va_mode &= ~VSVTX;
1364 
1365         if (*name == '\0') {
1366                 /*
1367                  * Null component name refers to the directory itself.
1368                  */
1369                 VN_HOLD(dvp);
1370                 zp = dzp;
1371                 dl = NULL;
1372                 error = 0;
1373         } else {
1374                 /* possible VN_HOLD(zp) */
1375                 int zflg = 0;
1376 
1377                 if (flag & FIGNORECASE)
1378                         zflg |= ZCILOOK;
1379 
1380                 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1381                     NULL, NULL);
1382                 if (error) {
1383                         if (have_acl)
1384                                 zfs_acl_ids_free(&acl_ids);
1385                         if (strcmp(name, "..") == 0)
1386                                 error = SET_ERROR(EISDIR);
1387                         ZFS_EXIT(zfsvfs);
1388                         return (error);
1389                 }
1390         }
1391 
1392         if (zp == NULL) {
1393                 uint64_t txtype;
1394 
1395                 /*
1396                  * Create a new file object and update the directory
1397                  * to reference it.
1398                  */
1399                 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1400                         if (have_acl)
1401                                 zfs_acl_ids_free(&acl_ids);
1402                         goto out;
1403                 }
1404 
1405                 /*
1406                  * We only support the creation of regular files in
1407                  * extended attribute directories.
1408                  */
1409 
1410                 if ((dzp->z_pflags & ZFS_XATTR) &&
1411                     (vap->va_type != VREG)) {
1412                         if (have_acl)
1413                                 zfs_acl_ids_free(&acl_ids);
1414                         error = SET_ERROR(EINVAL);
1415                         goto out;
1416                 }
1417 
1418                 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1419                     cr, vsecp, &acl_ids)) != 0)
1420                         goto out;
1421                 have_acl = B_TRUE;
1422 
1423                 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1424                         zfs_acl_ids_free(&acl_ids);
1425                         error = SET_ERROR(EDQUOT);
1426                         goto out;
1427                 }
1428 
1429                 tx = dmu_tx_create(os);
1430 
1431                 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1432                     ZFS_SA_BASE_ATTR_SIZE);
1433 
1434                 fuid_dirtied = zfsvfs->z_fuid_dirty;
1435                 if (fuid_dirtied)
1436                         zfs_fuid_txhold(zfsvfs, tx);
1437                 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1438                 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1439                 if (!zfsvfs->z_use_sa &&
1440                     acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1441                         dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1442                             0, acl_ids.z_aclp->z_acl_bytes);
1443                 }
1444                 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1445                 if (error) {
1446                         zfs_dirent_unlock(dl);
1447                         if (error == ERESTART) {
1448                                 waited = B_TRUE;
1449                                 dmu_tx_wait(tx);
1450                                 dmu_tx_abort(tx);
1451                                 goto top;
1452                         }
1453                         zfs_acl_ids_free(&acl_ids);
1454                         dmu_tx_abort(tx);
1455                         ZFS_EXIT(zfsvfs);
1456                         return (error);
1457                 }
1458                 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1459 
1460                 if (fuid_dirtied)
1461                         zfs_fuid_sync(zfsvfs, tx);
1462 
1463                 (void) zfs_link_create(dl, zp, tx, ZNEW);
1464                 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1465                 if (flag & FIGNORECASE)
1466                         txtype |= TX_CI;
1467                 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1468                     vsecp, acl_ids.z_fuidp, vap);
1469                 zfs_acl_ids_free(&acl_ids);
1470                 dmu_tx_commit(tx);
1471         } else {
1472                 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1473 
1474                 if (have_acl)
1475                         zfs_acl_ids_free(&acl_ids);
1476                 have_acl = B_FALSE;
1477 
1478                 /*
1479                  * A directory entry already exists for this name.
1480                  */
1481                 /*
1482                  * Can't truncate an existing file if in exclusive mode.
1483                  */
1484                 if (excl == EXCL) {
1485                         error = SET_ERROR(EEXIST);
1486                         goto out;
1487                 }
1488                 /*
1489                  * Can't open a directory for writing.
1490                  */
1491                 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1492                         error = SET_ERROR(EISDIR);
1493                         goto out;
1494                 }
1495                 /*
1496                  * Verify requested access to file.
1497                  */
1498                 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1499                         goto out;
1500                 }
1501 
1502                 mutex_enter(&dzp->z_lock);
1503                 dzp->z_seq++;
1504                 mutex_exit(&dzp->z_lock);
1505 
1506                 /*
1507                  * Truncate regular files if requested.
1508                  */
1509                 if ((ZTOV(zp)->v_type == VREG) &&
1510                     (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1511                         /* we can't hold any locks when calling zfs_freesp() */
1512                         zfs_dirent_unlock(dl);
1513                         dl = NULL;
1514                         error = zfs_freesp(zp, 0, 0, mode, TRUE);
1515                         if (error == 0) {
1516                                 vnevent_create(ZTOV(zp), ct);
1517                         }
1518                 }
1519         }
1520 out:
1521 
1522         if (dl)
1523                 zfs_dirent_unlock(dl);
1524 
1525         if (error) {
1526                 if (zp)
1527                         VN_RELE(ZTOV(zp));
1528         } else {
1529                 *vpp = ZTOV(zp);
1530                 error = specvp_check(vpp, cr);
1531         }
1532 
1533         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1534                 zil_commit(zilog, 0);
1535 
1536         ZFS_EXIT(zfsvfs);
1537         return (error);
1538 }
1539 
1540 /*
1541  * Remove an entry from a directory.
1542  *
1543  *      IN:     dvp     - vnode of directory to remove entry from.
1544  *              name    - name of entry to remove.
1545  *              cr      - credentials of caller.
1546  *              ct      - caller context
1547  *              flags   - case flags
1548  *
1549  *      RETURN: 0 on success, error code on failure.
1550  *
1551  * Timestamps:
1552  *      dvp - ctime|mtime
1553  *       vp - ctime (if nlink > 0)
1554  */
1555 
1556 uint64_t null_xattr = 0;
1557 
1558 /*ARGSUSED*/
1559 static int
1560 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1561     int flags)
1562 {
1563         znode_t         *zp, *dzp = VTOZ(dvp);
1564         znode_t         *xzp;
1565         vnode_t         *vp;
1566         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1567         zilog_t         *zilog;
1568         uint64_t        acl_obj, xattr_obj;
1569         uint64_t        xattr_obj_unlinked = 0;
1570         uint64_t        obj = 0;
1571         zfs_dirlock_t   *dl;
1572         dmu_tx_t        *tx;
1573         boolean_t       may_delete_now, delete_now = FALSE;
1574         boolean_t       unlinked, toobig = FALSE;
1575         uint64_t        txtype;
1576         pathname_t      *realnmp = NULL;
1577         pathname_t      realnm;
1578         int             error;
1579         int             zflg = ZEXISTS;
1580         boolean_t       waited = B_FALSE;
1581 
1582         ZFS_ENTER(zfsvfs);
1583         ZFS_VERIFY_ZP(dzp);
1584         zilog = zfsvfs->z_log;
1585 
1586         if (flags & FIGNORECASE) {
1587                 zflg |= ZCILOOK;
1588                 pn_alloc(&realnm);
1589                 realnmp = &realnm;
1590         }
1591 
1592 top:
1593         xattr_obj = 0;
1594         xzp = NULL;
1595         /*
1596          * Attempt to lock directory; fail if entry doesn't exist.
1597          */
1598         if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1599             NULL, realnmp)) {
1600                 if (realnmp)
1601                         pn_free(realnmp);
1602                 ZFS_EXIT(zfsvfs);
1603                 return (error);
1604         }
1605 
1606         vp = ZTOV(zp);
1607 
1608         if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1609                 goto out;
1610         }
1611 
1612         /*
1613          * Need to use rmdir for removing directories.
1614          */
1615         if (vp->v_type == VDIR) {
1616                 error = SET_ERROR(EPERM);
1617                 goto out;
1618         }
1619 
1620         vnevent_remove(vp, dvp, name, ct);
1621 
1622         if (realnmp)
1623                 dnlc_remove(dvp, realnmp->pn_buf);
1624         else
1625                 dnlc_remove(dvp, name);
1626 
1627         mutex_enter(&vp->v_lock);
1628         may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1629         mutex_exit(&vp->v_lock);
1630 
1631         /*
1632          * We may delete the znode now, or we may put it in the unlinked set;
1633          * it depends on whether we're the last link, and on whether there are
1634          * other holds on the vnode.  So we dmu_tx_hold() the right things to
1635          * allow for either case.
1636          */
1637         obj = zp->z_id;
1638         tx = dmu_tx_create(zfsvfs->z_os);
1639         dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1640         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1641         zfs_sa_upgrade_txholds(tx, zp);
1642         zfs_sa_upgrade_txholds(tx, dzp);
1643         if (may_delete_now) {
1644                 toobig =
1645                     zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1646                 /* if the file is too big, only hold_free a token amount */
1647                 dmu_tx_hold_free(tx, zp->z_id, 0,
1648                     (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1649         }
1650 
1651         /* are there any extended attributes? */
1652         error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1653             &xattr_obj, sizeof (xattr_obj));
1654         if (error == 0 && xattr_obj) {
1655                 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1656                 ASSERT0(error);
1657                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1658                 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1659         }
1660 
1661         mutex_enter(&zp->z_lock);
1662         if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1663                 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1664         mutex_exit(&zp->z_lock);
1665 
1666         /* charge as an update -- would be nice not to charge at all */
1667         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1668 
1669         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1670         if (error) {
1671                 zfs_dirent_unlock(dl);
1672                 VN_RELE(vp);
1673                 if (xzp)
1674                         VN_RELE(ZTOV(xzp));
1675                 if (error == ERESTART) {
1676                         waited = B_TRUE;
1677                         dmu_tx_wait(tx);
1678                         dmu_tx_abort(tx);
1679                         goto top;
1680                 }
1681                 if (realnmp)
1682                         pn_free(realnmp);
1683                 dmu_tx_abort(tx);
1684                 ZFS_EXIT(zfsvfs);
1685                 return (error);
1686         }
1687 
1688         /*
1689          * Remove the directory entry.
1690          */
1691         error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1692 
1693         if (error) {
1694                 dmu_tx_commit(tx);
1695                 goto out;
1696         }
1697 
1698         if (unlinked) {
1699 
1700                 /*
1701                  * Hold z_lock so that we can make sure that the ACL obj
1702                  * hasn't changed.  Could have been deleted due to
1703                  * zfs_sa_upgrade().
1704                  */
1705                 mutex_enter(&zp->z_lock);
1706                 mutex_enter(&vp->v_lock);
1707                 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1708                     &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1709                 delete_now = may_delete_now && !toobig &&
1710                     vp->v_count == 1 && !vn_has_cached_data(vp) &&
1711                     xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1712                     acl_obj;
1713                 mutex_exit(&vp->v_lock);
1714         }
1715 
1716         if (delete_now) {
1717                 if (xattr_obj_unlinked) {
1718                         ASSERT3U(xzp->z_links, ==, 2);
1719                         mutex_enter(&xzp->z_lock);
1720                         xzp->z_unlinked = 1;
1721                         xzp->z_links = 0;
1722                         error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1723                             &xzp->z_links, sizeof (xzp->z_links), tx);
1724                         ASSERT3U(error,  ==,  0);
1725                         mutex_exit(&xzp->z_lock);
1726                         zfs_unlinked_add(xzp, tx);
1727 
1728                         if (zp->z_is_sa)
1729                                 error = sa_remove(zp->z_sa_hdl,
1730                                     SA_ZPL_XATTR(zfsvfs), tx);
1731                         else
1732                                 error = sa_update(zp->z_sa_hdl,
1733                                     SA_ZPL_XATTR(zfsvfs), &null_xattr,
1734                                     sizeof (uint64_t), tx);
1735                         ASSERT0(error);
1736                 }
1737                 mutex_enter(&vp->v_lock);
1738                 vp->v_count--;
1739                 ASSERT0(vp->v_count);
1740                 mutex_exit(&vp->v_lock);
1741                 mutex_exit(&zp->z_lock);
1742                 zfs_znode_delete(zp, tx);
1743         } else if (unlinked) {
1744                 mutex_exit(&zp->z_lock);
1745                 zfs_unlinked_add(zp, tx);
1746         }
1747 
1748         txtype = TX_REMOVE;
1749         if (flags & FIGNORECASE)
1750                 txtype |= TX_CI;
1751         zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1752 
1753         dmu_tx_commit(tx);
1754 out:
1755         if (realnmp)
1756                 pn_free(realnmp);
1757 
1758         zfs_dirent_unlock(dl);
1759 
1760         if (!delete_now)
1761                 VN_RELE(vp);
1762         if (xzp)
1763                 VN_RELE(ZTOV(xzp));
1764 
1765         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1766                 zil_commit(zilog, 0);
1767 
1768         ZFS_EXIT(zfsvfs);
1769         return (error);
1770 }
1771 
1772 /*
1773  * Create a new directory and insert it into dvp using the name
1774  * provided.  Return a pointer to the inserted directory.
1775  *
1776  *      IN:     dvp     - vnode of directory to add subdir to.
1777  *              dirname - name of new directory.
1778  *              vap     - attributes of new directory.
1779  *              cr      - credentials of caller.
1780  *              ct      - caller context
1781  *              flags   - case flags
1782  *              vsecp   - ACL to be set
1783  *
1784  *      OUT:    vpp     - vnode of created directory.
1785  *
1786  *      RETURN: 0 on success, error code on failure.
1787  *
1788  * Timestamps:
1789  *      dvp - ctime|mtime updated
1790  *       vp - ctime|mtime|atime updated
1791  */
1792 /*ARGSUSED*/
1793 static int
1794 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1795     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1796 {
1797         znode_t         *zp, *dzp = VTOZ(dvp);
1798         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1799         zilog_t         *zilog;
1800         zfs_dirlock_t   *dl;
1801         uint64_t        txtype;
1802         dmu_tx_t        *tx;
1803         int             error;
1804         int             zf = ZNEW;
1805         ksid_t          *ksid;
1806         uid_t           uid;
1807         gid_t           gid = crgetgid(cr);
1808         zfs_acl_ids_t   acl_ids;
1809         boolean_t       fuid_dirtied;
1810         boolean_t       waited = B_FALSE;
1811 
1812         ASSERT(vap->va_type == VDIR);
1813 
1814         /*
1815          * If we have an ephemeral id, ACL, or XVATTR then
1816          * make sure file system is at proper version
1817          */
1818 
1819         ksid = crgetsid(cr, KSID_OWNER);
1820         if (ksid)
1821                 uid = ksid_getid(ksid);
1822         else
1823                 uid = crgetuid(cr);
1824         if (zfsvfs->z_use_fuids == B_FALSE &&
1825             (vsecp || (vap->va_mask & AT_XVATTR) ||
1826             IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1827                 return (SET_ERROR(EINVAL));
1828 
1829         ZFS_ENTER(zfsvfs);
1830         ZFS_VERIFY_ZP(dzp);
1831         zilog = zfsvfs->z_log;
1832 
1833         if (dzp->z_pflags & ZFS_XATTR) {
1834                 ZFS_EXIT(zfsvfs);
1835                 return (SET_ERROR(EINVAL));
1836         }
1837 
1838         if (zfsvfs->z_utf8 && u8_validate(dirname,
1839             strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1840                 ZFS_EXIT(zfsvfs);
1841                 return (SET_ERROR(EILSEQ));
1842         }
1843         if (flags & FIGNORECASE)
1844                 zf |= ZCILOOK;
1845 
1846         if (vap->va_mask & AT_XVATTR) {
1847                 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1848                     crgetuid(cr), cr, vap->va_type)) != 0) {
1849                         ZFS_EXIT(zfsvfs);
1850                         return (error);
1851                 }
1852         }
1853 
1854         if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1855             vsecp, &acl_ids)) != 0) {
1856                 ZFS_EXIT(zfsvfs);
1857                 return (error);
1858         }
1859         /*
1860          * First make sure the new directory doesn't exist.
1861          *
1862          * Existence is checked first to make sure we don't return
1863          * EACCES instead of EEXIST which can cause some applications
1864          * to fail.
1865          */
1866 top:
1867         *vpp = NULL;
1868 
1869         if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1870             NULL, NULL)) {
1871                 zfs_acl_ids_free(&acl_ids);
1872                 ZFS_EXIT(zfsvfs);
1873                 return (error);
1874         }
1875 
1876         if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1877                 zfs_acl_ids_free(&acl_ids);
1878                 zfs_dirent_unlock(dl);
1879                 ZFS_EXIT(zfsvfs);
1880                 return (error);
1881         }
1882 
1883         if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1884                 zfs_acl_ids_free(&acl_ids);
1885                 zfs_dirent_unlock(dl);
1886                 ZFS_EXIT(zfsvfs);
1887                 return (SET_ERROR(EDQUOT));
1888         }
1889 
1890         /*
1891          * Add a new entry to the directory.
1892          */
1893         tx = dmu_tx_create(zfsvfs->z_os);
1894         dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1895         dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1896         fuid_dirtied = zfsvfs->z_fuid_dirty;
1897         if (fuid_dirtied)
1898                 zfs_fuid_txhold(zfsvfs, tx);
1899         if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1900                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1901                     acl_ids.z_aclp->z_acl_bytes);
1902         }
1903 
1904         dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1905             ZFS_SA_BASE_ATTR_SIZE);
1906 
1907         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1908         if (error) {
1909                 zfs_dirent_unlock(dl);
1910                 if (error == ERESTART) {
1911                         waited = B_TRUE;
1912                         dmu_tx_wait(tx);
1913                         dmu_tx_abort(tx);
1914                         goto top;
1915                 }
1916                 zfs_acl_ids_free(&acl_ids);
1917                 dmu_tx_abort(tx);
1918                 ZFS_EXIT(zfsvfs);
1919                 return (error);
1920         }
1921 
1922         /*
1923          * Create new node.
1924          */
1925         zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1926 
1927         if (fuid_dirtied)
1928                 zfs_fuid_sync(zfsvfs, tx);
1929 
1930         /*
1931          * Now put new name in parent dir.
1932          */
1933         (void) zfs_link_create(dl, zp, tx, ZNEW);
1934 
1935         *vpp = ZTOV(zp);
1936 
1937         txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1938         if (flags & FIGNORECASE)
1939                 txtype |= TX_CI;
1940         zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1941             acl_ids.z_fuidp, vap);
1942 
1943         zfs_acl_ids_free(&acl_ids);
1944 
1945         dmu_tx_commit(tx);
1946 
1947         zfs_dirent_unlock(dl);
1948 
1949         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1950                 zil_commit(zilog, 0);
1951 
1952         ZFS_EXIT(zfsvfs);
1953         return (0);
1954 }
1955 
1956 /*
1957  * Remove a directory subdir entry.  If the current working
1958  * directory is the same as the subdir to be removed, the
1959  * remove will fail.
1960  *
1961  *      IN:     dvp     - vnode of directory to remove from.
1962  *              name    - name of directory to be removed.
1963  *              cwd     - vnode of current working directory.
1964  *              cr      - credentials of caller.
1965  *              ct      - caller context
1966  *              flags   - case flags
1967  *
1968  *      RETURN: 0 on success, error code on failure.
1969  *
1970  * Timestamps:
1971  *      dvp - ctime|mtime updated
1972  */
1973 /*ARGSUSED*/
1974 static int
1975 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1976     caller_context_t *ct, int flags)
1977 {
1978         znode_t         *dzp = VTOZ(dvp);
1979         znode_t         *zp;
1980         vnode_t         *vp;
1981         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1982         zilog_t         *zilog;
1983         zfs_dirlock_t   *dl;
1984         dmu_tx_t        *tx;
1985         int             error;
1986         int             zflg = ZEXISTS;
1987         boolean_t       waited = B_FALSE;
1988 
1989         ZFS_ENTER(zfsvfs);
1990         ZFS_VERIFY_ZP(dzp);
1991         zilog = zfsvfs->z_log;
1992 
1993         if (flags & FIGNORECASE)
1994                 zflg |= ZCILOOK;
1995 top:
1996         zp = NULL;
1997 
1998         /*
1999          * Attempt to lock directory; fail if entry doesn't exist.
2000          */
2001         if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2002             NULL, NULL)) {
2003                 ZFS_EXIT(zfsvfs);
2004                 return (error);
2005         }
2006 
2007         vp = ZTOV(zp);
2008 
2009         if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2010                 goto out;
2011         }
2012 
2013         if (vp->v_type != VDIR) {
2014                 error = SET_ERROR(ENOTDIR);
2015                 goto out;
2016         }
2017 
2018         if (vp == cwd) {
2019                 error = SET_ERROR(EINVAL);
2020                 goto out;
2021         }
2022 
2023         vnevent_rmdir(vp, dvp, name, ct);
2024 
2025         /*
2026          * Grab a lock on the directory to make sure that noone is
2027          * trying to add (or lookup) entries while we are removing it.
2028          */
2029         rw_enter(&zp->z_name_lock, RW_WRITER);
2030 
2031         /*
2032          * Grab a lock on the parent pointer to make sure we play well
2033          * with the treewalk and directory rename code.
2034          */
2035         rw_enter(&zp->z_parent_lock, RW_WRITER);
2036 
2037         tx = dmu_tx_create(zfsvfs->z_os);
2038         dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2039         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2040         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2041         zfs_sa_upgrade_txholds(tx, zp);
2042         zfs_sa_upgrade_txholds(tx, dzp);
2043         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2044         if (error) {
2045                 rw_exit(&zp->z_parent_lock);
2046                 rw_exit(&zp->z_name_lock);
2047                 zfs_dirent_unlock(dl);
2048                 VN_RELE(vp);
2049                 if (error == ERESTART) {
2050                         waited = B_TRUE;
2051                         dmu_tx_wait(tx);
2052                         dmu_tx_abort(tx);
2053                         goto top;
2054                 }
2055                 dmu_tx_abort(tx);
2056                 ZFS_EXIT(zfsvfs);
2057                 return (error);
2058         }
2059 
2060         error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2061 
2062         if (error == 0) {
2063                 uint64_t txtype = TX_RMDIR;
2064                 if (flags & FIGNORECASE)
2065                         txtype |= TX_CI;
2066                 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2067         }
2068 
2069         dmu_tx_commit(tx);
2070 
2071         rw_exit(&zp->z_parent_lock);
2072         rw_exit(&zp->z_name_lock);
2073 out:
2074         zfs_dirent_unlock(dl);
2075 
2076         VN_RELE(vp);
2077 
2078         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2079                 zil_commit(zilog, 0);
2080 
2081         ZFS_EXIT(zfsvfs);
2082         return (error);
2083 }
2084 
2085 /*
2086  * Read as many directory entries as will fit into the provided
2087  * buffer from the given directory cursor position (specified in
2088  * the uio structure).
2089  *
2090  *      IN:     vp      - vnode of directory to read.
2091  *              uio     - structure supplying read location, range info,
2092  *                        and return buffer.
2093  *              cr      - credentials of caller.
2094  *              ct      - caller context
2095  *              flags   - case flags
2096  *
2097  *      OUT:    uio     - updated offset and range, buffer filled.
2098  *              eofp    - set to true if end-of-file detected.
2099  *
2100  *      RETURN: 0 on success, error code on failure.
2101  *
2102  * Timestamps:
2103  *      vp - atime updated
2104  *
2105  * Note that the low 4 bits of the cookie returned by zap is always zero.
2106  * This allows us to use the low range for "special" directory entries:
2107  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2108  * we use the offset 2 for the '.zfs' directory.
2109  */
2110 /* ARGSUSED */
2111 static int
2112 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2113     caller_context_t *ct, int flags)
2114 {
2115         znode_t         *zp = VTOZ(vp);
2116         iovec_t         *iovp;
2117         edirent_t       *eodp;
2118         dirent64_t      *odp;
2119         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
2120         objset_t        *os;
2121         caddr_t         outbuf;
2122         size_t          bufsize;
2123         zap_cursor_t    zc;
2124         zap_attribute_t zap;
2125         uint_t          bytes_wanted;
2126         uint64_t        offset; /* must be unsigned; checks for < 1 */
2127         uint64_t        parent;
2128         int             local_eof;
2129         int             outcount;
2130         int             error;
2131         uint8_t         prefetch;
2132         boolean_t       check_sysattrs;
2133 
2134         ZFS_ENTER(zfsvfs);
2135         ZFS_VERIFY_ZP(zp);
2136 
2137         if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2138             &parent, sizeof (parent))) != 0) {
2139                 ZFS_EXIT(zfsvfs);
2140                 return (error);
2141         }
2142 
2143         /*
2144          * If we are not given an eof variable,
2145          * use a local one.
2146          */
2147         if (eofp == NULL)
2148                 eofp = &local_eof;
2149 
2150         /*
2151          * Check for valid iov_len.
2152          */
2153         if (uio->uio_iov->iov_len <= 0) {
2154                 ZFS_EXIT(zfsvfs);
2155                 return (SET_ERROR(EINVAL));
2156         }
2157 
2158         /*
2159          * Quit if directory has been removed (posix)
2160          */
2161         if ((*eofp = zp->z_unlinked) != 0) {
2162                 ZFS_EXIT(zfsvfs);
2163                 return (0);
2164         }
2165 
2166         error = 0;
2167         os = zfsvfs->z_os;
2168         offset = uio->uio_loffset;
2169         prefetch = zp->z_zn_prefetch;
2170 
2171         /*
2172          * Initialize the iterator cursor.
2173          */
2174         if (offset <= 3) {
2175                 /*
2176                  * Start iteration from the beginning of the directory.
2177                  */
2178                 zap_cursor_init(&zc, os, zp->z_id);
2179         } else {
2180                 /*
2181                  * The offset is a serialized cursor.
2182                  */
2183                 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2184         }
2185 
2186         /*
2187          * Get space to change directory entries into fs independent format.
2188          */
2189         iovp = uio->uio_iov;
2190         bytes_wanted = iovp->iov_len;
2191         if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2192                 bufsize = bytes_wanted;
2193                 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2194                 odp = (struct dirent64 *)outbuf;
2195         } else {
2196                 bufsize = bytes_wanted;
2197                 outbuf = NULL;
2198                 odp = (struct dirent64 *)iovp->iov_base;
2199         }
2200         eodp = (struct edirent *)odp;
2201 
2202         /*
2203          * If this VFS supports the system attribute view interface; and
2204          * we're looking at an extended attribute directory; and we care
2205          * about normalization conflicts on this vfs; then we must check
2206          * for normalization conflicts with the sysattr name space.
2207          */
2208         check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2209             (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2210             (flags & V_RDDIR_ENTFLAGS);
2211 
2212         /*
2213          * Transform to file-system independent format
2214          */
2215         outcount = 0;
2216         while (outcount < bytes_wanted) {
2217                 ino64_t objnum;
2218                 ushort_t reclen;
2219                 off64_t *next = NULL;
2220 
2221                 /*
2222                  * Special case `.', `..', and `.zfs'.
2223                  */
2224                 if (offset == 0) {
2225                         (void) strcpy(zap.za_name, ".");
2226                         zap.za_normalization_conflict = 0;
2227                         objnum = zp->z_id;
2228                 } else if (offset == 1) {
2229                         (void) strcpy(zap.za_name, "..");
2230                         zap.za_normalization_conflict = 0;
2231                         objnum = parent;
2232                 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2233                         (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2234                         zap.za_normalization_conflict = 0;
2235                         objnum = ZFSCTL_INO_ROOT;
2236                 } else {
2237                         /*
2238                          * Grab next entry.
2239                          */
2240                         if (error = zap_cursor_retrieve(&zc, &zap)) {
2241                                 if ((*eofp = (error == ENOENT)) != 0)
2242                                         break;
2243                                 else
2244                                         goto update;
2245                         }
2246 
2247                         if (zap.za_integer_length != 8 ||
2248                             zap.za_num_integers != 1) {
2249                                 cmn_err(CE_WARN, "zap_readdir: bad directory "
2250                                     "entry, obj = %lld, offset = %lld\n",
2251                                     (u_longlong_t)zp->z_id,
2252                                     (u_longlong_t)offset);
2253                                 error = SET_ERROR(ENXIO);
2254                                 goto update;
2255                         }
2256 
2257                         objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2258                         /*
2259                          * MacOS X can extract the object type here such as:
2260                          * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2261                          */
2262 
2263                         if (check_sysattrs && !zap.za_normalization_conflict) {
2264                                 zap.za_normalization_conflict =
2265                                     xattr_sysattr_casechk(zap.za_name);
2266                         }
2267                 }
2268 
2269                 if (flags & V_RDDIR_ACCFILTER) {
2270                         /*
2271                          * If we have no access at all, don't include
2272                          * this entry in the returned information
2273                          */
2274                         znode_t *ezp;
2275                         if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2276                                 goto skip_entry;
2277                         if (!zfs_has_access(ezp, cr)) {
2278                                 VN_RELE(ZTOV(ezp));
2279                                 goto skip_entry;
2280                         }
2281                         VN_RELE(ZTOV(ezp));
2282                 }
2283 
2284                 if (flags & V_RDDIR_ENTFLAGS)
2285                         reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2286                 else
2287                         reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2288 
2289                 /*
2290                  * Will this entry fit in the buffer?
2291                  */
2292                 if (outcount + reclen > bufsize) {
2293                         /*
2294                          * Did we manage to fit anything in the buffer?
2295                          */
2296                         if (!outcount) {
2297                                 error = SET_ERROR(EINVAL);
2298                                 goto update;
2299                         }
2300                         break;
2301                 }
2302                 if (flags & V_RDDIR_ENTFLAGS) {
2303                         /*
2304                          * Add extended flag entry:
2305                          */
2306                         eodp->ed_ino = objnum;
2307                         eodp->ed_reclen = reclen;
2308                         /* NOTE: ed_off is the offset for the *next* entry */
2309                         next = &(eodp->ed_off);
2310                         eodp->ed_eflags = zap.za_normalization_conflict ?
2311                             ED_CASE_CONFLICT : 0;
2312                         (void) strncpy(eodp->ed_name, zap.za_name,
2313                             EDIRENT_NAMELEN(reclen));
2314                         eodp = (edirent_t *)((intptr_t)eodp + reclen);
2315                 } else {
2316                         /*
2317                          * Add normal entry:
2318                          */
2319                         odp->d_ino = objnum;
2320                         odp->d_reclen = reclen;
2321                         /* NOTE: d_off is the offset for the *next* entry */
2322                         next = &(odp->d_off);
2323                         (void) strncpy(odp->d_name, zap.za_name,
2324                             DIRENT64_NAMELEN(reclen));
2325                         odp = (dirent64_t *)((intptr_t)odp + reclen);
2326                 }
2327                 outcount += reclen;
2328 
2329                 ASSERT(outcount <= bufsize);
2330 
2331                 /* Prefetch znode */
2332                 if (prefetch)
2333                         dmu_prefetch(os, objnum, 0, 0);
2334 
2335         skip_entry:
2336                 /*
2337                  * Move to the next entry, fill in the previous offset.
2338                  */
2339                 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2340                         zap_cursor_advance(&zc);
2341                         offset = zap_cursor_serialize(&zc);
2342                 } else {
2343                         offset += 1;
2344                 }
2345                 if (next)
2346                         *next = offset;
2347         }
2348         zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2349 
2350         if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2351                 iovp->iov_base += outcount;
2352                 iovp->iov_len -= outcount;
2353                 uio->uio_resid -= outcount;
2354         } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2355                 /*
2356                  * Reset the pointer.
2357                  */
2358                 offset = uio->uio_loffset;
2359         }
2360 
2361 update:
2362         zap_cursor_fini(&zc);
2363         if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2364                 kmem_free(outbuf, bufsize);
2365 
2366         if (error == ENOENT)
2367                 error = 0;
2368 
2369         ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2370 
2371         uio->uio_loffset = offset;
2372         ZFS_EXIT(zfsvfs);
2373         return (error);
2374 }
2375 
2376 ulong_t zfs_fsync_sync_cnt = 4;
2377 
2378 static int
2379 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2380 {
2381         znode_t *zp = VTOZ(vp);
2382         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2383 
2384         /*
2385          * Regardless of whether this is required for standards conformance,
2386          * this is the logical behavior when fsync() is called on a file with
2387          * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2388          * going to be pushed out as part of the zil_commit().
2389          */
2390         if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2391             (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2392                 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2393 
2394         (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2395 
2396         if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2397                 ZFS_ENTER(zfsvfs);
2398                 ZFS_VERIFY_ZP(zp);
2399                 zil_commit(zfsvfs->z_log, zp->z_id);
2400                 ZFS_EXIT(zfsvfs);
2401         }
2402         return (0);
2403 }
2404 
2405 
2406 /*
2407  * Get the requested file attributes and place them in the provided
2408  * vattr structure.
2409  *
2410  *      IN:     vp      - vnode of file.
2411  *              vap     - va_mask identifies requested attributes.
2412  *                        If AT_XVATTR set, then optional attrs are requested
2413  *              flags   - ATTR_NOACLCHECK (CIFS server context)
2414  *              cr      - credentials of caller.
2415  *              ct      - caller context
2416  *
2417  *      OUT:    vap     - attribute values.
2418  *
2419  *      RETURN: 0 (always succeeds).
2420  */
2421 /* ARGSUSED */
2422 static int
2423 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2424     caller_context_t *ct)
2425 {
2426         znode_t *zp = VTOZ(vp);
2427         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2428         int     error = 0;
2429         uint64_t links;
2430         uint64_t mtime[2], ctime[2];
2431         xvattr_t *xvap = (xvattr_t *)vap;       /* vap may be an xvattr_t * */
2432         xoptattr_t *xoap = NULL;
2433         boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2434         sa_bulk_attr_t bulk[2];
2435         int count = 0;
2436 
2437         ZFS_ENTER(zfsvfs);
2438         ZFS_VERIFY_ZP(zp);
2439 
2440         zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2441 
2442         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2443         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2444 
2445         if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2446                 ZFS_EXIT(zfsvfs);
2447                 return (error);
2448         }
2449 
2450         /*
2451          * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2452          * Also, if we are the owner don't bother, since owner should
2453          * always be allowed to read basic attributes of file.
2454          */
2455         if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2456             (vap->va_uid != crgetuid(cr))) {
2457                 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2458                     skipaclchk, cr)) {
2459                         ZFS_EXIT(zfsvfs);
2460                         return (error);
2461                 }
2462         }
2463 
2464         /*
2465          * Return all attributes.  It's cheaper to provide the answer
2466          * than to determine whether we were asked the question.
2467          */
2468 
2469         mutex_enter(&zp->z_lock);
2470         vap->va_type = vp->v_type;
2471         vap->va_mode = zp->z_mode & MODEMASK;
2472         vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2473         vap->va_nodeid = zp->z_id;
2474         if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2475                 links = zp->z_links + 1;
2476         else
2477                 links = zp->z_links;
2478         vap->va_nlink = MIN(links, UINT32_MAX);      /* nlink_t limit! */
2479         vap->va_size = zp->z_size;
2480         vap->va_rdev = vp->v_rdev;
2481         vap->va_seq = zp->z_seq;
2482 
2483         /*
2484          * Add in any requested optional attributes and the create time.
2485          * Also set the corresponding bits in the returned attribute bitmap.
2486          */
2487         if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2488                 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2489                         xoap->xoa_archive =
2490                             ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2491                         XVA_SET_RTN(xvap, XAT_ARCHIVE);
2492                 }
2493 
2494                 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2495                         xoap->xoa_readonly =
2496                             ((zp->z_pflags & ZFS_READONLY) != 0);
2497                         XVA_SET_RTN(xvap, XAT_READONLY);
2498                 }
2499 
2500                 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2501                         xoap->xoa_system =
2502                             ((zp->z_pflags & ZFS_SYSTEM) != 0);
2503                         XVA_SET_RTN(xvap, XAT_SYSTEM);
2504                 }
2505 
2506                 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2507                         xoap->xoa_hidden =
2508                             ((zp->z_pflags & ZFS_HIDDEN) != 0);
2509                         XVA_SET_RTN(xvap, XAT_HIDDEN);
2510                 }
2511 
2512                 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2513                         xoap->xoa_nounlink =
2514                             ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2515                         XVA_SET_RTN(xvap, XAT_NOUNLINK);
2516                 }
2517 
2518                 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2519                         xoap->xoa_immutable =
2520                             ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2521                         XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2522                 }
2523 
2524                 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2525                         xoap->xoa_appendonly =
2526                             ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2527                         XVA_SET_RTN(xvap, XAT_APPENDONLY);
2528                 }
2529 
2530                 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2531                         xoap->xoa_nodump =
2532                             ((zp->z_pflags & ZFS_NODUMP) != 0);
2533                         XVA_SET_RTN(xvap, XAT_NODUMP);
2534                 }
2535 
2536                 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2537                         xoap->xoa_opaque =
2538                             ((zp->z_pflags & ZFS_OPAQUE) != 0);
2539                         XVA_SET_RTN(xvap, XAT_OPAQUE);
2540                 }
2541 
2542                 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2543                         xoap->xoa_av_quarantined =
2544                             ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2545                         XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2546                 }
2547 
2548                 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2549                         xoap->xoa_av_modified =
2550                             ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2551                         XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2552                 }
2553 
2554                 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2555                     vp->v_type == VREG) {
2556                         zfs_sa_get_scanstamp(zp, xvap);
2557                 }
2558 
2559                 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2560                         uint64_t times[2];
2561 
2562                         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2563                             times, sizeof (times));
2564                         ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2565                         XVA_SET_RTN(xvap, XAT_CREATETIME);
2566                 }
2567 
2568                 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2569                         xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2570                         XVA_SET_RTN(xvap, XAT_REPARSE);
2571                 }
2572                 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2573                         xoap->xoa_generation = zp->z_gen;
2574                         XVA_SET_RTN(xvap, XAT_GEN);
2575                 }
2576 
2577                 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2578                         xoap->xoa_offline =
2579                             ((zp->z_pflags & ZFS_OFFLINE) != 0);
2580                         XVA_SET_RTN(xvap, XAT_OFFLINE);
2581                 }
2582 
2583                 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2584                         xoap->xoa_sparse =
2585                             ((zp->z_pflags & ZFS_SPARSE) != 0);
2586                         XVA_SET_RTN(xvap, XAT_SPARSE);
2587                 }
2588         }
2589 
2590         ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2591         ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2592         ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2593 
2594         mutex_exit(&zp->z_lock);
2595 
2596         sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2597 
2598         if (zp->z_blksz == 0) {
2599                 /*
2600                  * Block size hasn't been set; suggest maximal I/O transfers.
2601                  */
2602                 vap->va_blksize = zfsvfs->z_max_blksz;
2603         }
2604 
2605         ZFS_EXIT(zfsvfs);
2606         return (0);
2607 }
2608 
2609 /*
2610  * Set the file attributes to the values contained in the
2611  * vattr structure.
2612  *
2613  *      IN:     vp      - vnode of file to be modified.
2614  *              vap     - new attribute values.
2615  *                        If AT_XVATTR set, then optional attrs are being set
2616  *              flags   - ATTR_UTIME set if non-default time values provided.
2617  *                      - ATTR_NOACLCHECK (CIFS context only).
2618  *              cr      - credentials of caller.
2619  *              ct      - caller context
2620  *
2621  *      RETURN: 0 on success, error code on failure.
2622  *
2623  * Timestamps:
2624  *      vp - ctime updated, mtime updated if size changed.
2625  */
2626 /* ARGSUSED */
2627 static int
2628 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2629     caller_context_t *ct)
2630 {
2631         znode_t         *zp = VTOZ(vp);
2632         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
2633         zilog_t         *zilog;
2634         dmu_tx_t        *tx;
2635         vattr_t         oldva;
2636         xvattr_t        tmpxvattr;
2637         uint_t          mask = vap->va_mask;
2638         uint_t          saved_mask = 0;
2639         int             trim_mask = 0;
2640         uint64_t        new_mode;
2641         uint64_t        new_uid, new_gid;
2642         uint64_t        xattr_obj;
2643         uint64_t        mtime[2], ctime[2];
2644         znode_t         *attrzp;
2645         int             need_policy = FALSE;
2646         int             err, err2;
2647         zfs_fuid_info_t *fuidp = NULL;
2648         xvattr_t *xvap = (xvattr_t *)vap;       /* vap may be an xvattr_t * */
2649         xoptattr_t      *xoap;
2650         zfs_acl_t       *aclp;
2651         boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2652         boolean_t       fuid_dirtied = B_FALSE;
2653         sa_bulk_attr_t  bulk[7], xattr_bulk[7];
2654         int             count = 0, xattr_count = 0;
2655 
2656         if (mask == 0)
2657                 return (0);
2658 
2659         if (mask & AT_NOSET)
2660                 return (SET_ERROR(EINVAL));
2661 
2662         ZFS_ENTER(zfsvfs);
2663         ZFS_VERIFY_ZP(zp);
2664 
2665         zilog = zfsvfs->z_log;
2666 
2667         /*
2668          * Make sure that if we have ephemeral uid/gid or xvattr specified
2669          * that file system is at proper version level
2670          */
2671 
2672         if (zfsvfs->z_use_fuids == B_FALSE &&
2673             (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2674             ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2675             (mask & AT_XVATTR))) {
2676                 ZFS_EXIT(zfsvfs);
2677                 return (SET_ERROR(EINVAL));
2678         }
2679 
2680         if (mask & AT_SIZE && vp->v_type == VDIR) {
2681                 ZFS_EXIT(zfsvfs);
2682                 return (SET_ERROR(EISDIR));
2683         }
2684 
2685         if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2686                 ZFS_EXIT(zfsvfs);
2687                 return (SET_ERROR(EINVAL));
2688         }
2689 
2690         /*
2691          * If this is an xvattr_t, then get a pointer to the structure of
2692          * optional attributes.  If this is NULL, then we have a vattr_t.
2693          */
2694         xoap = xva_getxoptattr(xvap);
2695 
2696         xva_init(&tmpxvattr);
2697 
2698         /*
2699          * Immutable files can only alter immutable bit and atime
2700          */
2701         if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2702             ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2703             ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2704                 ZFS_EXIT(zfsvfs);
2705                 return (SET_ERROR(EPERM));
2706         }
2707 
2708         if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2709                 ZFS_EXIT(zfsvfs);
2710                 return (SET_ERROR(EPERM));
2711         }
2712 
2713         /*
2714          * Verify timestamps doesn't overflow 32 bits.
2715          * ZFS can handle large timestamps, but 32bit syscalls can't
2716          * handle times greater than 2039.  This check should be removed
2717          * once large timestamps are fully supported.
2718          */
2719         if (mask & (AT_ATIME | AT_MTIME)) {
2720                 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2721                     ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2722                         ZFS_EXIT(zfsvfs);
2723                         return (SET_ERROR(EOVERFLOW));
2724                 }
2725         }
2726 
2727 top:
2728         attrzp = NULL;
2729         aclp = NULL;
2730 
2731         /* Can this be moved to before the top label? */
2732         if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2733                 ZFS_EXIT(zfsvfs);
2734                 return (SET_ERROR(EROFS));
2735         }
2736 
2737         /*
2738          * First validate permissions
2739          */
2740 
2741         if (mask & AT_SIZE) {
2742                 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2743                 if (err) {
2744                         ZFS_EXIT(zfsvfs);
2745                         return (err);
2746                 }
2747                 /*
2748                  * XXX - Note, we are not providing any open
2749                  * mode flags here (like FNDELAY), so we may
2750                  * block if there are locks present... this
2751                  * should be addressed in openat().
2752                  */
2753                 /* XXX - would it be OK to generate a log record here? */
2754                 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2755                 if (err) {
2756                         ZFS_EXIT(zfsvfs);
2757                         return (err);
2758                 }
2759         }
2760 
2761         if (mask & (AT_ATIME|AT_MTIME) ||
2762             ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2763             XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2764             XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2765             XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2766             XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2767             XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2768             XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2769                 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2770                     skipaclchk, cr);
2771         }
2772 
2773         if (mask & (AT_UID|AT_GID)) {
2774                 int     idmask = (mask & (AT_UID|AT_GID));
2775                 int     take_owner;
2776                 int     take_group;
2777 
2778                 /*
2779                  * NOTE: even if a new mode is being set,
2780                  * we may clear S_ISUID/S_ISGID bits.
2781                  */
2782 
2783                 if (!(mask & AT_MODE))
2784                         vap->va_mode = zp->z_mode;
2785 
2786                 /*
2787                  * Take ownership or chgrp to group we are a member of
2788                  */
2789 
2790                 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2791                 take_group = (mask & AT_GID) &&
2792                     zfs_groupmember(zfsvfs, vap->va_gid, cr);
2793 
2794                 /*
2795                  * If both AT_UID and AT_GID are set then take_owner and
2796                  * take_group must both be set in order to allow taking
2797                  * ownership.
2798                  *
2799                  * Otherwise, send the check through secpolicy_vnode_setattr()
2800                  *
2801                  */
2802 
2803                 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2804                     ((idmask == AT_UID) && take_owner) ||
2805                     ((idmask == AT_GID) && take_group)) {
2806                         if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2807                             skipaclchk, cr) == 0) {
2808                                 /*
2809                                  * Remove setuid/setgid for non-privileged users
2810                                  */
2811                                 secpolicy_setid_clear(vap, cr);
2812                                 trim_mask = (mask & (AT_UID|AT_GID));
2813                         } else {
2814                                 need_policy =  TRUE;
2815                         }
2816                 } else {
2817                         need_policy =  TRUE;
2818                 }
2819         }
2820 
2821         mutex_enter(&zp->z_lock);
2822         oldva.va_mode = zp->z_mode;
2823         zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2824         if (mask & AT_XVATTR) {
2825                 /*
2826                  * Update xvattr mask to include only those attributes
2827                  * that are actually changing.
2828                  *
2829                  * the bits will be restored prior to actually setting
2830                  * the attributes so the caller thinks they were set.
2831                  */
2832                 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2833                         if (xoap->xoa_appendonly !=
2834                             ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2835                                 need_policy = TRUE;
2836                         } else {
2837                                 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2838                                 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2839                         }
2840                 }
2841 
2842                 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2843                         if (xoap->xoa_nounlink !=
2844                             ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2845                                 need_policy = TRUE;
2846                         } else {
2847                                 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2848                                 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2849                         }
2850                 }
2851 
2852                 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2853                         if (xoap->xoa_immutable !=
2854                             ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2855                                 need_policy = TRUE;
2856                         } else {
2857                                 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2858                                 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2859                         }
2860                 }
2861 
2862                 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2863                         if (xoap->xoa_nodump !=
2864                             ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2865                                 need_policy = TRUE;
2866                         } else {
2867                                 XVA_CLR_REQ(xvap, XAT_NODUMP);
2868                                 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2869                         }
2870                 }
2871 
2872                 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2873                         if (xoap->xoa_av_modified !=
2874                             ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2875                                 need_policy = TRUE;
2876                         } else {
2877                                 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2878                                 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2879                         }
2880                 }
2881 
2882                 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2883                         if ((vp->v_type != VREG &&
2884                             xoap->xoa_av_quarantined) ||
2885                             xoap->xoa_av_quarantined !=
2886                             ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2887                                 need_policy = TRUE;
2888                         } else {
2889                                 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2890                                 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2891                         }
2892                 }
2893 
2894                 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2895                         mutex_exit(&zp->z_lock);
2896                         ZFS_EXIT(zfsvfs);
2897                         return (SET_ERROR(EPERM));
2898                 }
2899 
2900                 if (need_policy == FALSE &&
2901                     (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2902                     XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2903                         need_policy = TRUE;
2904                 }
2905         }
2906 
2907         mutex_exit(&zp->z_lock);
2908 
2909         if (mask & AT_MODE) {
2910                 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2911                         err = secpolicy_setid_setsticky_clear(vp, vap,
2912                             &oldva, cr);
2913                         if (err) {
2914                                 ZFS_EXIT(zfsvfs);
2915                                 return (err);
2916                         }
2917                         trim_mask |= AT_MODE;
2918                 } else {
2919                         need_policy = TRUE;
2920                 }
2921         }
2922 
2923         if (need_policy) {
2924                 /*
2925                  * If trim_mask is set then take ownership
2926                  * has been granted or write_acl is present and user
2927                  * has the ability to modify mode.  In that case remove
2928                  * UID|GID and or MODE from mask so that
2929                  * secpolicy_vnode_setattr() doesn't revoke it.
2930                  */
2931 
2932                 if (trim_mask) {
2933                         saved_mask = vap->va_mask;
2934                         vap->va_mask &= ~trim_mask;
2935                 }
2936                 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2937                     (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2938                 if (err) {
2939                         ZFS_EXIT(zfsvfs);
2940                         return (err);
2941                 }
2942 
2943                 if (trim_mask)
2944                         vap->va_mask |= saved_mask;
2945         }
2946 
2947         /*
2948          * secpolicy_vnode_setattr, or take ownership may have
2949          * changed va_mask
2950          */
2951         mask = vap->va_mask;
2952 
2953         if ((mask & (AT_UID | AT_GID))) {
2954                 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2955                     &xattr_obj, sizeof (xattr_obj));
2956 
2957                 if (err == 0 && xattr_obj) {
2958                         err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2959                         if (err)
2960                                 goto out2;
2961                 }
2962                 if (mask & AT_UID) {
2963                         new_uid = zfs_fuid_create(zfsvfs,
2964                             (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2965                         if (new_uid != zp->z_uid &&
2966                             zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
2967                                 if (attrzp)
2968                                         VN_RELE(ZTOV(attrzp));
2969                                 err = SET_ERROR(EDQUOT);
2970                                 goto out2;
2971                         }
2972                 }
2973 
2974                 if (mask & AT_GID) {
2975                         new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2976                             cr, ZFS_GROUP, &fuidp);
2977                         if (new_gid != zp->z_gid &&
2978                             zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
2979                                 if (attrzp)
2980                                         VN_RELE(ZTOV(attrzp));
2981                                 err = SET_ERROR(EDQUOT);
2982                                 goto out2;
2983                         }
2984                 }
2985         }
2986         tx = dmu_tx_create(zfsvfs->z_os);
2987 
2988         if (mask & AT_MODE) {
2989                 uint64_t pmode = zp->z_mode;
2990                 uint64_t acl_obj;
2991                 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2992 
2993                 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
2994                     !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
2995                         err = SET_ERROR(EPERM);
2996                         goto out;
2997                 }
2998 
2999                 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3000                         goto out;
3001 
3002                 mutex_enter(&zp->z_lock);
3003                 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3004                         /*
3005                          * Are we upgrading ACL from old V0 format
3006                          * to V1 format?
3007                          */
3008                         if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3009                             zfs_znode_acl_version(zp) ==
3010                             ZFS_ACL_VERSION_INITIAL) {
3011                                 dmu_tx_hold_free(tx, acl_obj, 0,
3012                                     DMU_OBJECT_END);
3013                                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3014                                     0, aclp->z_acl_bytes);
3015                         } else {
3016                                 dmu_tx_hold_write(tx, acl_obj, 0,
3017                                     aclp->z_acl_bytes);
3018                         }
3019                 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3020                         dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3021                             0, aclp->z_acl_bytes);
3022                 }
3023                 mutex_exit(&zp->z_lock);
3024                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3025         } else {
3026                 if ((mask & AT_XVATTR) &&
3027                     XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3028                         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3029                 else
3030                         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3031         }
3032 
3033         if (attrzp) {
3034                 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3035         }
3036 
3037         fuid_dirtied = zfsvfs->z_fuid_dirty;
3038         if (fuid_dirtied)
3039                 zfs_fuid_txhold(zfsvfs, tx);
3040 
3041         zfs_sa_upgrade_txholds(tx, zp);
3042 
3043         err = dmu_tx_assign(tx, TXG_NOWAIT);
3044         if (err) {
3045                 if (err == ERESTART)
3046                         dmu_tx_wait(tx);
3047                 goto out;
3048         }
3049 
3050         count = 0;
3051         /*
3052          * Set each attribute requested.
3053          * We group settings according to the locks they need to acquire.
3054          *
3055          * Note: you cannot set ctime directly, although it will be
3056          * updated as a side-effect of calling this function.
3057          */
3058 
3059 
3060         if (mask & (AT_UID|AT_GID|AT_MODE))
3061                 mutex_enter(&zp->z_acl_lock);
3062         mutex_enter(&zp->z_lock);
3063 
3064         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3065             &zp->z_pflags, sizeof (zp->z_pflags));
3066 
3067         if (attrzp) {
3068                 if (mask & (AT_UID|AT_GID|AT_MODE))
3069                         mutex_enter(&attrzp->z_acl_lock);
3070                 mutex_enter(&attrzp->z_lock);
3071                 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3072                     SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3073                     sizeof (attrzp->z_pflags));
3074         }
3075 
3076         if (mask & (AT_UID|AT_GID)) {
3077 
3078                 if (mask & AT_UID) {
3079                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3080                             &new_uid, sizeof (new_uid));
3081                         zp->z_uid = new_uid;
3082                         if (attrzp) {
3083                                 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3084                                     SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3085                                     sizeof (new_uid));
3086                                 attrzp->z_uid = new_uid;
3087                         }
3088                 }
3089 
3090                 if (mask & AT_GID) {
3091                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3092                             NULL, &new_gid, sizeof (new_gid));
3093                         zp->z_gid = new_gid;
3094                         if (attrzp) {
3095                                 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3096                                     SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3097                                     sizeof (new_gid));
3098                                 attrzp->z_gid = new_gid;
3099                         }
3100                 }
3101                 if (!(mask & AT_MODE)) {
3102                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3103                             NULL, &new_mode, sizeof (new_mode));
3104                         new_mode = zp->z_mode;
3105                 }
3106                 err = zfs_acl_chown_setattr(zp);
3107                 ASSERT(err == 0);
3108                 if (attrzp) {
3109                         err = zfs_acl_chown_setattr(attrzp);
3110                         ASSERT(err == 0);
3111                 }
3112         }
3113 
3114         if (mask & AT_MODE) {
3115                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3116                     &new_mode, sizeof (new_mode));
3117                 zp->z_mode = new_mode;
3118                 ASSERT3U((uintptr_t)aclp, !=, NULL);
3119                 err = zfs_aclset_common(zp, aclp, cr, tx);
3120                 ASSERT0(err);
3121                 if (zp->z_acl_cached)
3122                         zfs_acl_free(zp->z_acl_cached);
3123                 zp->z_acl_cached = aclp;
3124                 aclp = NULL;
3125         }
3126 
3127 
3128         if (mask & AT_ATIME) {
3129                 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3130                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3131                     &zp->z_atime, sizeof (zp->z_atime));
3132         }
3133 
3134         if (mask & AT_MTIME) {
3135                 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3136                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3137                     mtime, sizeof (mtime));
3138         }
3139 
3140         /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3141         if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3142                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3143                     NULL, mtime, sizeof (mtime));
3144                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3145                     &ctime, sizeof (ctime));
3146                 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3147                     B_TRUE);
3148         } else if (mask != 0) {
3149                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3150                     &ctime, sizeof (ctime));
3151                 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3152                     B_TRUE);
3153                 if (attrzp) {
3154                         SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3155                             SA_ZPL_CTIME(zfsvfs), NULL,
3156                             &ctime, sizeof (ctime));
3157                         zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3158                             mtime, ctime, B_TRUE);
3159                 }
3160         }
3161         /*
3162          * Do this after setting timestamps to prevent timestamp
3163          * update from toggling bit
3164          */
3165 
3166         if (xoap && (mask & AT_XVATTR)) {
3167 
3168                 /*
3169                  * restore trimmed off masks
3170                  * so that return masks can be set for caller.
3171                  */
3172 
3173                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3174                         XVA_SET_REQ(xvap, XAT_APPENDONLY);
3175                 }
3176                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3177                         XVA_SET_REQ(xvap, XAT_NOUNLINK);
3178                 }
3179                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3180                         XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3181                 }
3182                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3183                         XVA_SET_REQ(xvap, XAT_NODUMP);
3184                 }
3185                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3186                         XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3187                 }
3188                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3189                         XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3190                 }
3191 
3192                 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3193                         ASSERT(vp->v_type == VREG);
3194 
3195                 zfs_xvattr_set(zp, xvap, tx);
3196         }
3197 
3198         if (fuid_dirtied)
3199                 zfs_fuid_sync(zfsvfs, tx);
3200 
3201         if (mask != 0)
3202                 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3203 
3204         mutex_exit(&zp->z_lock);
3205         if (mask & (AT_UID|AT_GID|AT_MODE))
3206                 mutex_exit(&zp->z_acl_lock);
3207 
3208         if (attrzp) {
3209                 if (mask & (AT_UID|AT_GID|AT_MODE))
3210                         mutex_exit(&attrzp->z_acl_lock);
3211                 mutex_exit(&attrzp->z_lock);
3212         }
3213 out:
3214         if (err == 0 && attrzp) {
3215                 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3216                     xattr_count, tx);
3217                 ASSERT(err2 == 0);
3218         }
3219 
3220         if (attrzp)
3221                 VN_RELE(ZTOV(attrzp));
3222 
3223         if (aclp)
3224                 zfs_acl_free(aclp);
3225 
3226         if (fuidp) {
3227                 zfs_fuid_info_free(fuidp);
3228                 fuidp = NULL;
3229         }
3230 
3231         if (err) {
3232                 dmu_tx_abort(tx);
3233                 if (err == ERESTART)
3234                         goto top;
3235         } else {
3236                 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3237                 dmu_tx_commit(tx);
3238         }
3239 
3240 out2:
3241         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3242                 zil_commit(zilog, 0);
3243 
3244         ZFS_EXIT(zfsvfs);
3245         return (err);
3246 }
3247 
3248 typedef struct zfs_zlock {
3249         krwlock_t       *zl_rwlock;     /* lock we acquired */
3250         znode_t         *zl_znode;      /* znode we held */
3251         struct zfs_zlock *zl_next;      /* next in list */
3252 } zfs_zlock_t;
3253 
3254 /*
3255  * Drop locks and release vnodes that were held by zfs_rename_lock().
3256  */
3257 static void
3258 zfs_rename_unlock(zfs_zlock_t **zlpp)
3259 {
3260         zfs_zlock_t *zl;
3261 
3262         while ((zl = *zlpp) != NULL) {
3263                 if (zl->zl_znode != NULL)
3264                         VN_RELE(ZTOV(zl->zl_znode));
3265                 rw_exit(zl->zl_rwlock);
3266                 *zlpp = zl->zl_next;
3267                 kmem_free(zl, sizeof (*zl));
3268         }
3269 }
3270 
3271 /*
3272  * Search back through the directory tree, using the ".." entries.
3273  * Lock each directory in the chain to prevent concurrent renames.
3274  * Fail any attempt to move a directory into one of its own descendants.
3275  * XXX - z_parent_lock can overlap with map or grow locks
3276  */
3277 static int
3278 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3279 {
3280         zfs_zlock_t     *zl;
3281         znode_t         *zp = tdzp;
3282         uint64_t        rootid = zp->z_zfsvfs->z_root;
3283         uint64_t        oidp = zp->z_id;
3284         krwlock_t       *rwlp = &szp->z_parent_lock;
3285         krw_t           rw = RW_WRITER;
3286 
3287         /*
3288          * First pass write-locks szp and compares to zp->z_id.
3289          * Later passes read-lock zp and compare to zp->z_parent.
3290          */
3291         do {
3292                 if (!rw_tryenter(rwlp, rw)) {
3293                         /*
3294                          * Another thread is renaming in this path.
3295                          * Note that if we are a WRITER, we don't have any
3296                          * parent_locks held yet.
3297                          */
3298                         if (rw == RW_READER && zp->z_id > szp->z_id) {
3299                                 /*
3300                                  * Drop our locks and restart
3301                                  */
3302                                 zfs_rename_unlock(&zl);
3303                                 *zlpp = NULL;
3304                                 zp = tdzp;
3305                                 oidp = zp->z_id;
3306                                 rwlp = &szp->z_parent_lock;
3307                                 rw = RW_WRITER;
3308                                 continue;
3309                         } else {
3310                                 /*
3311                                  * Wait for other thread to drop its locks
3312                                  */
3313                                 rw_enter(rwlp, rw);
3314                         }
3315                 }
3316 
3317                 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3318                 zl->zl_rwlock = rwlp;
3319                 zl->zl_znode = NULL;
3320                 zl->zl_next = *zlpp;
3321                 *zlpp = zl;
3322 
3323                 if (oidp == szp->z_id)               /* We're a descendant of szp */
3324                         return (SET_ERROR(EINVAL));
3325 
3326                 if (oidp == rootid)             /* We've hit the top */
3327                         return (0);
3328 
3329                 if (rw == RW_READER) {          /* i.e. not the first pass */
3330                         int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3331                         if (error)
3332                                 return (error);
3333                         zl->zl_znode = zp;
3334                 }
3335                 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3336                     &oidp, sizeof (oidp));
3337                 rwlp = &zp->z_parent_lock;
3338                 rw = RW_READER;
3339 
3340         } while (zp->z_id != sdzp->z_id);
3341 
3342         return (0);
3343 }
3344 
3345 /*
3346  * Move an entry from the provided source directory to the target
3347  * directory.  Change the entry name as indicated.
3348  *
3349  *      IN:     sdvp    - Source directory containing the "old entry".
3350  *              snm     - Old entry name.
3351  *              tdvp    - Target directory to contain the "new entry".
3352  *              tnm     - New entry name.
3353  *              cr      - credentials of caller.
3354  *              ct      - caller context
3355  *              flags   - case flags
3356  *
3357  *      RETURN: 0 on success, error code on failure.
3358  *
3359  * Timestamps:
3360  *      sdvp,tdvp - ctime|mtime updated
3361  */
3362 /*ARGSUSED*/
3363 static int
3364 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3365     caller_context_t *ct, int flags)
3366 {
3367         znode_t         *tdzp, *szp, *tzp;
3368         znode_t         *sdzp = VTOZ(sdvp);
3369         zfsvfs_t        *zfsvfs = sdzp->z_zfsvfs;
3370         zilog_t         *zilog;
3371         vnode_t         *realvp;
3372         zfs_dirlock_t   *sdl, *tdl;
3373         dmu_tx_t        *tx;
3374         zfs_zlock_t     *zl;
3375         int             cmp, serr, terr;
3376         int             error = 0;
3377         int             zflg = 0;
3378         boolean_t       waited = B_FALSE;
3379 
3380         ZFS_ENTER(zfsvfs);
3381         ZFS_VERIFY_ZP(sdzp);
3382         zilog = zfsvfs->z_log;
3383 
3384         /*
3385          * Make sure we have the real vp for the target directory.
3386          */
3387         if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3388                 tdvp = realvp;
3389 
3390         tdzp = VTOZ(tdvp);
3391         ZFS_VERIFY_ZP(tdzp);
3392 
3393         /*
3394          * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3395          * ctldir appear to have the same v_vfsp.
3396          */
3397         if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3398                 ZFS_EXIT(zfsvfs);
3399                 return (SET_ERROR(EXDEV));
3400         }
3401 
3402         if (zfsvfs->z_utf8 && u8_validate(tnm,
3403             strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3404                 ZFS_EXIT(zfsvfs);
3405                 return (SET_ERROR(EILSEQ));
3406         }
3407 
3408         if (flags & FIGNORECASE)
3409                 zflg |= ZCILOOK;
3410 
3411 top:
3412         szp = NULL;
3413         tzp = NULL;
3414         zl = NULL;
3415 
3416         /*
3417          * This is to prevent the creation of links into attribute space
3418          * by renaming a linked file into/outof an attribute directory.
3419          * See the comment in zfs_link() for why this is considered bad.
3420          */
3421         if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3422                 ZFS_EXIT(zfsvfs);
3423                 return (SET_ERROR(EINVAL));
3424         }
3425 
3426         /*
3427          * Lock source and target directory entries.  To prevent deadlock,
3428          * a lock ordering must be defined.  We lock the directory with
3429          * the smallest object id first, or if it's a tie, the one with
3430          * the lexically first name.
3431          */
3432         if (sdzp->z_id < tdzp->z_id) {
3433                 cmp = -1;
3434         } else if (sdzp->z_id > tdzp->z_id) {
3435                 cmp = 1;
3436         } else {
3437                 /*
3438                  * First compare the two name arguments without
3439                  * considering any case folding.
3440                  */
3441                 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3442 
3443                 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3444                 ASSERT(error == 0 || !zfsvfs->z_utf8);
3445                 if (cmp == 0) {
3446                         /*
3447                          * POSIX: "If the old argument and the new argument
3448                          * both refer to links to the same existing file,
3449                          * the rename() function shall return successfully
3450                          * and perform no other action."
3451                          */
3452                         ZFS_EXIT(zfsvfs);
3453                         return (0);
3454                 }
3455                 /*
3456                  * If the file system is case-folding, then we may
3457                  * have some more checking to do.  A case-folding file
3458                  * system is either supporting mixed case sensitivity
3459                  * access or is completely case-insensitive.  Note
3460                  * that the file system is always case preserving.
3461                  *
3462                  * In mixed sensitivity mode case sensitive behavior
3463                  * is the default.  FIGNORECASE must be used to
3464                  * explicitly request case insensitive behavior.
3465                  *
3466                  * If the source and target names provided differ only
3467                  * by case (e.g., a request to rename 'tim' to 'Tim'),
3468                  * we will treat this as a special case in the
3469                  * case-insensitive mode: as long as the source name
3470                  * is an exact match, we will allow this to proceed as
3471                  * a name-change request.
3472                  */
3473                 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3474                     (zfsvfs->z_case == ZFS_CASE_MIXED &&
3475                     flags & FIGNORECASE)) &&
3476                     u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3477                     &error) == 0) {
3478                         /*
3479                          * case preserving rename request, require exact
3480                          * name matches
3481                          */
3482                         zflg |= ZCIEXACT;
3483                         zflg &= ~ZCILOOK;
3484                 }
3485         }
3486 
3487         /*
3488          * If the source and destination directories are the same, we should
3489          * grab the z_name_lock of that directory only once.
3490          */
3491         if (sdzp == tdzp) {
3492                 zflg |= ZHAVELOCK;
3493                 rw_enter(&sdzp->z_name_lock, RW_READER);
3494         }
3495 
3496         if (cmp < 0) {
3497                 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3498                     ZEXISTS | zflg, NULL, NULL);
3499                 terr = zfs_dirent_lock(&tdl,
3500                     tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3501         } else {
3502                 terr = zfs_dirent_lock(&tdl,
3503                     tdzp, tnm, &tzp, zflg, NULL, NULL);
3504                 serr = zfs_dirent_lock(&sdl,
3505                     sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3506                     NULL, NULL);
3507         }
3508 
3509         if (serr) {
3510                 /*
3511                  * Source entry invalid or not there.
3512                  */
3513                 if (!terr) {
3514                         zfs_dirent_unlock(tdl);
3515                         if (tzp)
3516                                 VN_RELE(ZTOV(tzp));
3517                 }
3518 
3519                 if (sdzp == tdzp)
3520                         rw_exit(&sdzp->z_name_lock);
3521 
3522                 if (strcmp(snm, "..") == 0)
3523                         serr = SET_ERROR(EINVAL);
3524                 ZFS_EXIT(zfsvfs);
3525                 return (serr);
3526         }
3527         if (terr) {
3528                 zfs_dirent_unlock(sdl);
3529                 VN_RELE(ZTOV(szp));
3530 
3531                 if (sdzp == tdzp)
3532                         rw_exit(&sdzp->z_name_lock);
3533 
3534                 if (strcmp(tnm, "..") == 0)
3535                         terr = SET_ERROR(EINVAL);
3536                 ZFS_EXIT(zfsvfs);
3537                 return (terr);
3538         }
3539 
3540         /*
3541          * Must have write access at the source to remove the old entry
3542          * and write access at the target to create the new entry.
3543          * Note that if target and source are the same, this can be
3544          * done in a single check.
3545          */
3546 
3547         if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3548                 goto out;
3549 
3550         if (ZTOV(szp)->v_type == VDIR) {
3551                 /*
3552                  * Check to make sure rename is valid.
3553                  * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3554                  */
3555                 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3556                         goto out;
3557         }
3558 
3559         /*
3560          * Does target exist?
3561          */
3562         if (tzp) {
3563                 /*
3564                  * Source and target must be the same type.
3565                  */
3566                 if (ZTOV(szp)->v_type == VDIR) {
3567                         if (ZTOV(tzp)->v_type != VDIR) {
3568                                 error = SET_ERROR(ENOTDIR);
3569                                 goto out;
3570                         }
3571                 } else {
3572                         if (ZTOV(tzp)->v_type == VDIR) {
3573                                 error = SET_ERROR(EISDIR);
3574                                 goto out;
3575                         }
3576                 }
3577                 /*
3578                  * POSIX dictates that when the source and target
3579                  * entries refer to the same file object, rename
3580                  * must do nothing and exit without error.
3581                  */
3582                 if (szp->z_id == tzp->z_id) {
3583                         error = 0;
3584                         goto out;
3585                 }
3586         }
3587 
3588         vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3589         if (tzp)
3590                 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3591 
3592         /*
3593          * notify the target directory if it is not the same
3594          * as source directory.
3595          */
3596         if (tdvp != sdvp) {
3597                 vnevent_rename_dest_dir(tdvp, ct);
3598         }
3599 
3600         tx = dmu_tx_create(zfsvfs->z_os);
3601         dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3602         dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3603         dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3604         dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3605         if (sdzp != tdzp) {
3606                 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3607                 zfs_sa_upgrade_txholds(tx, tdzp);
3608         }
3609         if (tzp) {
3610                 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3611                 zfs_sa_upgrade_txholds(tx, tzp);
3612         }
3613 
3614         zfs_sa_upgrade_txholds(tx, szp);
3615         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3616         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3617         if (error) {
3618                 if (zl != NULL)
3619                         zfs_rename_unlock(&zl);
3620                 zfs_dirent_unlock(sdl);
3621                 zfs_dirent_unlock(tdl);
3622 
3623                 if (sdzp == tdzp)
3624                         rw_exit(&sdzp->z_name_lock);
3625 
3626                 VN_RELE(ZTOV(szp));
3627                 if (tzp)
3628                         VN_RELE(ZTOV(tzp));
3629                 if (error == ERESTART) {
3630                         waited = B_TRUE;
3631                         dmu_tx_wait(tx);
3632                         dmu_tx_abort(tx);
3633                         goto top;
3634                 }
3635                 dmu_tx_abort(tx);
3636                 ZFS_EXIT(zfsvfs);
3637                 return (error);
3638         }
3639 
3640         if (tzp)        /* Attempt to remove the existing target */
3641                 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3642 
3643         if (error == 0) {
3644                 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3645                 if (error == 0) {
3646                         szp->z_pflags |= ZFS_AV_MODIFIED;
3647 
3648                         error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3649                             (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3650                         ASSERT0(error);
3651 
3652                         error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3653                         if (error == 0) {
3654                                 zfs_log_rename(zilog, tx, TX_RENAME |
3655                                     (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3656                                     sdl->dl_name, tdzp, tdl->dl_name, szp);
3657 
3658                                 /*
3659                                  * Update path information for the target vnode
3660                                  */
3661                                 vn_renamepath(tdvp, ZTOV(szp), tnm,
3662                                     strlen(tnm));
3663                         } else {
3664                                 /*
3665                                  * At this point, we have successfully created
3666                                  * the target name, but have failed to remove
3667                                  * the source name.  Since the create was done
3668                                  * with the ZRENAMING flag, there are
3669                                  * complications; for one, the link count is
3670                                  * wrong.  The easiest way to deal with this
3671                                  * is to remove the newly created target, and
3672                                  * return the original error.  This must
3673                                  * succeed; fortunately, it is very unlikely to
3674                                  * fail, since we just created it.
3675                                  */
3676                                 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3677                                     ZRENAMING, NULL), ==, 0);
3678                         }
3679                 }
3680         }
3681 
3682         dmu_tx_commit(tx);
3683 out:
3684         if (zl != NULL)
3685                 zfs_rename_unlock(&zl);
3686 
3687         zfs_dirent_unlock(sdl);
3688         zfs_dirent_unlock(tdl);
3689 
3690         if (sdzp == tdzp)
3691                 rw_exit(&sdzp->z_name_lock);
3692 
3693 
3694         VN_RELE(ZTOV(szp));
3695         if (tzp)
3696                 VN_RELE(ZTOV(tzp));
3697 
3698         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3699                 zil_commit(zilog, 0);
3700 
3701         ZFS_EXIT(zfsvfs);
3702         return (error);
3703 }
3704 
3705 /*
3706  * Insert the indicated symbolic reference entry into the directory.
3707  *
3708  *      IN:     dvp     - Directory to contain new symbolic link.
3709  *              link    - Name for new symlink entry.
3710  *              vap     - Attributes of new entry.
3711  *              cr      - credentials of caller.
3712  *              ct      - caller context
3713  *              flags   - case flags
3714  *
3715  *      RETURN: 0 on success, error code on failure.
3716  *
3717  * Timestamps:
3718  *      dvp - ctime|mtime updated
3719  */
3720 /*ARGSUSED*/
3721 static int
3722 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3723     caller_context_t *ct, int flags)
3724 {
3725         znode_t         *zp, *dzp = VTOZ(dvp);
3726         zfs_dirlock_t   *dl;
3727         dmu_tx_t        *tx;
3728         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
3729         zilog_t         *zilog;
3730         uint64_t        len = strlen(link);
3731         int             error;
3732         int             zflg = ZNEW;
3733         zfs_acl_ids_t   acl_ids;
3734         boolean_t       fuid_dirtied;
3735         uint64_t        txtype = TX_SYMLINK;
3736         boolean_t       waited = B_FALSE;
3737 
3738         ASSERT(vap->va_type == VLNK);
3739 
3740         ZFS_ENTER(zfsvfs);
3741         ZFS_VERIFY_ZP(dzp);
3742         zilog = zfsvfs->z_log;
3743 
3744         if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3745             NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3746                 ZFS_EXIT(zfsvfs);
3747                 return (SET_ERROR(EILSEQ));
3748         }
3749         if (flags & FIGNORECASE)
3750                 zflg |= ZCILOOK;
3751 
3752         if (len > MAXPATHLEN) {
3753                 ZFS_EXIT(zfsvfs);
3754                 return (SET_ERROR(ENAMETOOLONG));
3755         }
3756 
3757         if ((error = zfs_acl_ids_create(dzp, 0,
3758             vap, cr, NULL, &acl_ids)) != 0) {
3759                 ZFS_EXIT(zfsvfs);
3760                 return (error);
3761         }
3762 top:
3763         /*
3764          * Attempt to lock directory; fail if entry already exists.
3765          */
3766         error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3767         if (error) {
3768                 zfs_acl_ids_free(&acl_ids);
3769                 ZFS_EXIT(zfsvfs);
3770                 return (error);
3771         }
3772 
3773         if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3774                 zfs_acl_ids_free(&acl_ids);
3775                 zfs_dirent_unlock(dl);
3776                 ZFS_EXIT(zfsvfs);
3777                 return (error);
3778         }
3779 
3780         if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3781                 zfs_acl_ids_free(&acl_ids);
3782                 zfs_dirent_unlock(dl);
3783                 ZFS_EXIT(zfsvfs);
3784                 return (SET_ERROR(EDQUOT));
3785         }
3786         tx = dmu_tx_create(zfsvfs->z_os);
3787         fuid_dirtied = zfsvfs->z_fuid_dirty;
3788         dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3789         dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3790         dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3791             ZFS_SA_BASE_ATTR_SIZE + len);
3792         dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3793         if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3794                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3795                     acl_ids.z_aclp->z_acl_bytes);
3796         }
3797         if (fuid_dirtied)
3798                 zfs_fuid_txhold(zfsvfs, tx);
3799         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3800         if (error) {
3801                 zfs_dirent_unlock(dl);
3802                 if (error == ERESTART) {
3803                         waited = B_TRUE;
3804                         dmu_tx_wait(tx);
3805                         dmu_tx_abort(tx);
3806                         goto top;
3807                 }
3808                 zfs_acl_ids_free(&acl_ids);
3809                 dmu_tx_abort(tx);
3810                 ZFS_EXIT(zfsvfs);
3811                 return (error);
3812         }
3813 
3814         /*
3815          * Create a new object for the symlink.
3816          * for version 4 ZPL datsets the symlink will be an SA attribute
3817          */
3818         zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3819 
3820         if (fuid_dirtied)
3821                 zfs_fuid_sync(zfsvfs, tx);
3822 
3823         mutex_enter(&zp->z_lock);
3824         if (zp->z_is_sa)
3825                 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3826                     link, len, tx);
3827         else
3828                 zfs_sa_symlink(zp, link, len, tx);
3829         mutex_exit(&zp->z_lock);
3830 
3831         zp->z_size = len;
3832         (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3833             &zp->z_size, sizeof (zp->z_size), tx);
3834         /*
3835          * Insert the new object into the directory.
3836          */
3837         (void) zfs_link_create(dl, zp, tx, ZNEW);
3838 
3839         if (flags & FIGNORECASE)
3840                 txtype |= TX_CI;
3841         zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3842 
3843         zfs_acl_ids_free(&acl_ids);
3844 
3845         dmu_tx_commit(tx);
3846 
3847         zfs_dirent_unlock(dl);
3848 
3849         VN_RELE(ZTOV(zp));
3850 
3851         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3852                 zil_commit(zilog, 0);
3853 
3854         ZFS_EXIT(zfsvfs);
3855         return (error);
3856 }
3857 
3858 /*
3859  * Return, in the buffer contained in the provided uio structure,
3860  * the symbolic path referred to by vp.
3861  *
3862  *      IN:     vp      - vnode of symbolic link.
3863  *              uio     - structure to contain the link path.
3864  *              cr      - credentials of caller.
3865  *              ct      - caller context
3866  *
3867  *      OUT:    uio     - structure containing the link path.
3868  *
3869  *      RETURN: 0 on success, error code on failure.
3870  *
3871  * Timestamps:
3872  *      vp - atime updated
3873  */
3874 /* ARGSUSED */
3875 static int
3876 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3877 {
3878         znode_t         *zp = VTOZ(vp);
3879         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
3880         int             error;
3881 
3882         ZFS_ENTER(zfsvfs);
3883         ZFS_VERIFY_ZP(zp);
3884 
3885         mutex_enter(&zp->z_lock);
3886         if (zp->z_is_sa)
3887                 error = sa_lookup_uio(zp->z_sa_hdl,
3888                     SA_ZPL_SYMLINK(zfsvfs), uio);
3889         else
3890                 error = zfs_sa_readlink(zp, uio);
3891         mutex_exit(&zp->z_lock);
3892 
3893         ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3894 
3895         ZFS_EXIT(zfsvfs);
3896         return (error);
3897 }
3898 
3899 /*
3900  * Insert a new entry into directory tdvp referencing svp.
3901  *
3902  *      IN:     tdvp    - Directory to contain new entry.
3903  *              svp     - vnode of new entry.
3904  *              name    - name of new entry.
3905  *              cr      - credentials of caller.
3906  *              ct      - caller context
3907  *
3908  *      RETURN: 0 on success, error code on failure.
3909  *
3910  * Timestamps:
3911  *      tdvp - ctime|mtime updated
3912  *       svp - ctime updated
3913  */
3914 /* ARGSUSED */
3915 static int
3916 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3917     caller_context_t *ct, int flags)
3918 {
3919         znode_t         *dzp = VTOZ(tdvp);
3920         znode_t         *tzp, *szp;
3921         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
3922         zilog_t         *zilog;
3923         zfs_dirlock_t   *dl;
3924         dmu_tx_t        *tx;
3925         vnode_t         *realvp;
3926         int             error;
3927         int             zf = ZNEW;
3928         uint64_t        parent;
3929         uid_t           owner;
3930         boolean_t       waited = B_FALSE;
3931 
3932         ASSERT(tdvp->v_type == VDIR);
3933 
3934         ZFS_ENTER(zfsvfs);
3935         ZFS_VERIFY_ZP(dzp);
3936         zilog = zfsvfs->z_log;
3937 
3938         if (VOP_REALVP(svp, &realvp, ct) == 0)
3939                 svp = realvp;
3940 
3941         /*
3942          * POSIX dictates that we return EPERM here.
3943          * Better choices include ENOTSUP or EISDIR.
3944          */
3945         if (svp->v_type == VDIR) {
3946                 ZFS_EXIT(zfsvfs);
3947                 return (SET_ERROR(EPERM));
3948         }
3949 
3950         szp = VTOZ(svp);
3951         ZFS_VERIFY_ZP(szp);
3952 
3953         /*
3954          * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3955          * ctldir appear to have the same v_vfsp.
3956          */
3957         if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
3958                 ZFS_EXIT(zfsvfs);
3959                 return (SET_ERROR(EXDEV));
3960         }
3961 
3962         /* Prevent links to .zfs/shares files */
3963 
3964         if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3965             &parent, sizeof (uint64_t))) != 0) {
3966                 ZFS_EXIT(zfsvfs);
3967                 return (error);
3968         }
3969         if (parent == zfsvfs->z_shares_dir) {
3970                 ZFS_EXIT(zfsvfs);
3971                 return (SET_ERROR(EPERM));
3972         }
3973 
3974         if (zfsvfs->z_utf8 && u8_validate(name,
3975             strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3976                 ZFS_EXIT(zfsvfs);
3977                 return (SET_ERROR(EILSEQ));
3978         }
3979         if (flags & FIGNORECASE)
3980                 zf |= ZCILOOK;
3981 
3982         /*
3983          * We do not support links between attributes and non-attributes
3984          * because of the potential security risk of creating links
3985          * into "normal" file space in order to circumvent restrictions
3986          * imposed in attribute space.
3987          */
3988         if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3989                 ZFS_EXIT(zfsvfs);
3990                 return (SET_ERROR(EINVAL));
3991         }
3992 
3993 
3994         owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
3995         if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3996                 ZFS_EXIT(zfsvfs);
3997                 return (SET_ERROR(EPERM));
3998         }
3999 
4000         if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4001                 ZFS_EXIT(zfsvfs);
4002                 return (error);
4003         }
4004 
4005 top:
4006         /*
4007          * Attempt to lock directory; fail if entry already exists.
4008          */
4009         error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4010         if (error) {
4011                 ZFS_EXIT(zfsvfs);
4012                 return (error);
4013         }
4014 
4015         tx = dmu_tx_create(zfsvfs->z_os);
4016         dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4017         dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4018         zfs_sa_upgrade_txholds(tx, szp);
4019         zfs_sa_upgrade_txholds(tx, dzp);
4020         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4021         if (error) {
4022                 zfs_dirent_unlock(dl);
4023                 if (error == ERESTART) {
4024                         waited = B_TRUE;
4025                         dmu_tx_wait(tx);
4026                         dmu_tx_abort(tx);
4027                         goto top;
4028                 }
4029                 dmu_tx_abort(tx);
4030                 ZFS_EXIT(zfsvfs);
4031                 return (error);
4032         }
4033 
4034         error = zfs_link_create(dl, szp, tx, 0);
4035 
4036         if (error == 0) {
4037                 uint64_t txtype = TX_LINK;
4038                 if (flags & FIGNORECASE)
4039                         txtype |= TX_CI;
4040                 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4041         }
4042 
4043         dmu_tx_commit(tx);
4044 
4045         zfs_dirent_unlock(dl);
4046 
4047         if (error == 0) {
4048                 vnevent_link(svp, ct);
4049         }
4050 
4051         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4052                 zil_commit(zilog, 0);
4053 
4054         ZFS_EXIT(zfsvfs);
4055         return (error);
4056 }
4057 
4058 /*
4059  * zfs_null_putapage() is used when the file system has been force
4060  * unmounted. It just drops the pages.
4061  */
4062 /* ARGSUSED */
4063 static int
4064 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4065                 size_t *lenp, int flags, cred_t *cr)
4066 {
4067         pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4068         return (0);
4069 }
4070 
4071 /*
4072  * Push a page out to disk, klustering if possible.
4073  *
4074  *      IN:     vp      - file to push page to.
4075  *              pp      - page to push.
4076  *              flags   - additional flags.
4077  *              cr      - credentials of caller.
4078  *
4079  *      OUT:    offp    - start of range pushed.
4080  *              lenp    - len of range pushed.
4081  *
4082  *      RETURN: 0 on success, error code on failure.
4083  *
4084  * NOTE: callers must have locked the page to be pushed.  On
4085  * exit, the page (and all other pages in the kluster) must be
4086  * unlocked.
4087  */
4088 /* ARGSUSED */
4089 static int
4090 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4091                 size_t *lenp, int flags, cred_t *cr)
4092 {
4093         znode_t         *zp = VTOZ(vp);
4094         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4095         dmu_tx_t        *tx;
4096         u_offset_t      off, koff;
4097         size_t          len, klen;
4098         int             err;
4099 
4100         off = pp->p_offset;
4101         len = PAGESIZE;
4102         /*
4103          * If our blocksize is bigger than the page size, try to kluster
4104          * multiple pages so that we write a full block (thus avoiding
4105          * a read-modify-write).
4106          */
4107         if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4108                 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4109                 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4110                 ASSERT(koff <= zp->z_size);
4111                 if (koff + klen > zp->z_size)
4112                         klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4113                 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4114         }
4115         ASSERT3U(btop(len), ==, btopr(len));
4116 
4117         /*
4118          * Can't push pages past end-of-file.
4119          */
4120         if (off >= zp->z_size) {
4121                 /* ignore all pages */
4122                 err = 0;
4123                 goto out;
4124         } else if (off + len > zp->z_size) {
4125                 int npages = btopr(zp->z_size - off);
4126                 page_t *trunc;
4127 
4128                 page_list_break(&pp, &trunc, npages);
4129                 /* ignore pages past end of file */
4130                 if (trunc)
4131                         pvn_write_done(trunc, flags);
4132                 len = zp->z_size - off;
4133         }
4134 
4135         if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4136             zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4137                 err = SET_ERROR(EDQUOT);
4138                 goto out;
4139         }
4140 top:
4141         tx = dmu_tx_create(zfsvfs->z_os);
4142         dmu_tx_hold_write(tx, zp->z_id, off, len);
4143 
4144         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4145         zfs_sa_upgrade_txholds(tx, zp);
4146         err = dmu_tx_assign(tx, TXG_NOWAIT);
4147         if (err != 0) {
4148                 if (err == ERESTART) {
4149                         dmu_tx_wait(tx);
4150                         dmu_tx_abort(tx);
4151                         goto top;
4152                 }
4153                 dmu_tx_abort(tx);
4154                 goto out;
4155         }
4156 
4157         if (zp->z_blksz <= PAGESIZE) {
4158                 caddr_t va = zfs_map_page(pp, S_READ);
4159                 ASSERT3U(len, <=, PAGESIZE);
4160                 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4161                 zfs_unmap_page(pp, va);
4162         } else {
4163                 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4164         }
4165 
4166         if (err == 0) {
4167                 uint64_t mtime[2], ctime[2];
4168                 sa_bulk_attr_t bulk[3];
4169                 int count = 0;
4170 
4171                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4172                     &mtime, 16);
4173                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4174                     &ctime, 16);
4175                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4176                     &zp->z_pflags, 8);
4177                 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4178                     B_TRUE);
4179                 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4180         }
4181         dmu_tx_commit(tx);
4182 
4183 out:
4184         pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4185         if (offp)
4186                 *offp = off;
4187         if (lenp)
4188                 *lenp = len;
4189 
4190         return (err);
4191 }
4192 
4193 /*
4194  * Copy the portion of the file indicated from pages into the file.
4195  * The pages are stored in a page list attached to the files vnode.
4196  *
4197  *      IN:     vp      - vnode of file to push page data to.
4198  *              off     - position in file to put data.
4199  *              len     - amount of data to write.
4200  *              flags   - flags to control the operation.
4201  *              cr      - credentials of caller.
4202  *              ct      - caller context.
4203  *
4204  *      RETURN: 0 on success, error code on failure.
4205  *
4206  * Timestamps:
4207  *      vp - ctime|mtime updated
4208  */
4209 /*ARGSUSED*/
4210 static int
4211 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4212     caller_context_t *ct)
4213 {
4214         znode_t         *zp = VTOZ(vp);
4215         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4216         page_t          *pp;
4217         size_t          io_len;
4218         u_offset_t      io_off;
4219         uint_t          blksz;
4220         rl_t            *rl;
4221         int             error = 0;
4222 
4223         ZFS_ENTER(zfsvfs);
4224         ZFS_VERIFY_ZP(zp);
4225 
4226         /*
4227          * There's nothing to do if no data is cached.
4228          */
4229         if (!vn_has_cached_data(vp)) {
4230                 ZFS_EXIT(zfsvfs);
4231                 return (0);
4232         }
4233 
4234         /*
4235          * Align this request to the file block size in case we kluster.
4236          * XXX - this can result in pretty aggresive locking, which can
4237          * impact simultanious read/write access.  One option might be
4238          * to break up long requests (len == 0) into block-by-block
4239          * operations to get narrower locking.
4240          */
4241         blksz = zp->z_blksz;
4242         if (ISP2(blksz))
4243                 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4244         else
4245                 io_off = 0;
4246         if (len > 0 && ISP2(blksz))
4247                 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4248         else
4249                 io_len = 0;
4250 
4251         if (io_len == 0) {
4252                 /*
4253                  * Search the entire vp list for pages >= io_off.
4254                  */
4255                 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4256                 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4257                 goto out;
4258         }
4259         rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4260 
4261         if (off > zp->z_size) {
4262                 /* past end of file */
4263                 zfs_range_unlock(rl);
4264                 ZFS_EXIT(zfsvfs);
4265                 return (0);
4266         }
4267 
4268         len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4269 
4270         for (off = io_off; io_off < off + len; io_off += io_len) {
4271                 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4272                         pp = page_lookup(vp, io_off,
4273                             (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4274                 } else {
4275                         pp = page_lookup_nowait(vp, io_off,
4276                             (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4277                 }
4278 
4279                 if (pp != NULL && pvn_getdirty(pp, flags)) {
4280                         int err;
4281 
4282                         /*
4283                          * Found a dirty page to push
4284                          */
4285                         err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4286                         if (err)
4287                                 error = err;
4288                 } else {
4289                         io_len = PAGESIZE;
4290                 }
4291         }
4292 out:
4293         zfs_range_unlock(rl);
4294         if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4295                 zil_commit(zfsvfs->z_log, zp->z_id);
4296         ZFS_EXIT(zfsvfs);
4297         return (error);
4298 }
4299 
4300 /*ARGSUSED*/
4301 void
4302 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4303 {
4304         znode_t *zp = VTOZ(vp);
4305         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4306         int error;
4307 
4308         rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4309         if (zp->z_sa_hdl == NULL) {
4310                 /*
4311                  * The fs has been unmounted, or we did a
4312                  * suspend/resume and this file no longer exists.
4313                  */
4314                 if (vn_has_cached_data(vp)) {
4315                         (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4316                             B_INVAL, cr);
4317                 }
4318 
4319                 mutex_enter(&zp->z_lock);
4320                 mutex_enter(&vp->v_lock);
4321                 ASSERT(vp->v_count == 1);
4322                 vp->v_count = 0;
4323                 mutex_exit(&vp->v_lock);
4324                 mutex_exit(&zp->z_lock);
4325                 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4326                 zfs_znode_free(zp);
4327                 return;
4328         }
4329 
4330         /*
4331          * Attempt to push any data in the page cache.  If this fails
4332          * we will get kicked out later in zfs_zinactive().
4333          */
4334         if (vn_has_cached_data(vp)) {
4335                 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4336                     cr);
4337         }
4338 
4339         if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4340                 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4341 
4342                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4343                 zfs_sa_upgrade_txholds(tx, zp);
4344                 error = dmu_tx_assign(tx, TXG_WAIT);
4345                 if (error) {
4346                         dmu_tx_abort(tx);
4347                 } else {
4348                         mutex_enter(&zp->z_lock);
4349                         (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4350                             (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4351                         zp->z_atime_dirty = 0;
4352                         mutex_exit(&zp->z_lock);
4353                         dmu_tx_commit(tx);
4354                 }
4355         }
4356 
4357         zfs_zinactive(zp);
4358         rw_exit(&zfsvfs->z_teardown_inactive_lock);
4359 }
4360 
4361 /*
4362  * Bounds-check the seek operation.
4363  *
4364  *      IN:     vp      - vnode seeking within
4365  *              ooff    - old file offset
4366  *              noffp   - pointer to new file offset
4367  *              ct      - caller context
4368  *
4369  *      RETURN: 0 on success, EINVAL if new offset invalid.
4370  */
4371 /* ARGSUSED */
4372 static int
4373 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4374     caller_context_t *ct)
4375 {
4376         if (vp->v_type == VDIR)
4377                 return (0);
4378         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4379 }
4380 
4381 /*
4382  * Pre-filter the generic locking function to trap attempts to place
4383  * a mandatory lock on a memory mapped file.
4384  */
4385 static int
4386 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4387     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4388 {
4389         znode_t *zp = VTOZ(vp);
4390         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4391 
4392         ZFS_ENTER(zfsvfs);
4393         ZFS_VERIFY_ZP(zp);
4394 
4395         /*
4396          * We are following the UFS semantics with respect to mapcnt
4397          * here: If we see that the file is mapped already, then we will
4398          * return an error, but we don't worry about races between this
4399          * function and zfs_map().
4400          */
4401         if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4402                 ZFS_EXIT(zfsvfs);
4403                 return (SET_ERROR(EAGAIN));
4404         }
4405         ZFS_EXIT(zfsvfs);
4406         return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4407 }
4408 
4409 /*
4410  * If we can't find a page in the cache, we will create a new page
4411  * and fill it with file data.  For efficiency, we may try to fill
4412  * multiple pages at once (klustering) to fill up the supplied page
4413  * list.  Note that the pages to be filled are held with an exclusive
4414  * lock to prevent access by other threads while they are being filled.
4415  */
4416 static int
4417 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4418     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4419 {
4420         znode_t *zp = VTOZ(vp);
4421         page_t *pp, *cur_pp;
4422         objset_t *os = zp->z_zfsvfs->z_os;
4423         u_offset_t io_off, total;
4424         size_t io_len;
4425         int err;
4426 
4427         if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4428                 /*
4429                  * We only have a single page, don't bother klustering
4430                  */
4431                 io_off = off;
4432                 io_len = PAGESIZE;
4433                 pp = page_create_va(vp, io_off, io_len,
4434                     PG_EXCL | PG_WAIT, seg, addr);
4435         } else {
4436                 /*
4437                  * Try to find enough pages to fill the page list
4438                  */
4439                 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4440                     &io_len, off, plsz, 0);
4441         }
4442         if (pp == NULL) {
4443                 /*
4444                  * The page already exists, nothing to do here.
4445                  */
4446                 *pl = NULL;
4447                 return (0);
4448         }
4449 
4450         /*
4451          * Fill the pages in the kluster.
4452          */
4453         cur_pp = pp;
4454         for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4455                 caddr_t va;
4456 
4457                 ASSERT3U(io_off, ==, cur_pp->p_offset);
4458                 va = zfs_map_page(cur_pp, S_WRITE);
4459                 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4460                     DMU_READ_PREFETCH);
4461                 zfs_unmap_page(cur_pp, va);
4462                 if (err) {
4463                         /* On error, toss the entire kluster */
4464                         pvn_read_done(pp, B_ERROR);
4465                         /* convert checksum errors into IO errors */
4466                         if (err == ECKSUM)
4467                                 err = SET_ERROR(EIO);
4468                         return (err);
4469                 }
4470                 cur_pp = cur_pp->p_next;
4471         }
4472 
4473         /*
4474          * Fill in the page list array from the kluster starting
4475          * from the desired offset `off'.
4476          * NOTE: the page list will always be null terminated.
4477          */
4478         pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4479         ASSERT(pl == NULL || (*pl)->p_offset == off);
4480 
4481         return (0);
4482 }
4483 
4484 /*
4485  * Return pointers to the pages for the file region [off, off + len]
4486  * in the pl array.  If plsz is greater than len, this function may
4487  * also return page pointers from after the specified region
4488  * (i.e. the region [off, off + plsz]).  These additional pages are
4489  * only returned if they are already in the cache, or were created as
4490  * part of a klustered read.
4491  *
4492  *      IN:     vp      - vnode of file to get data from.
4493  *              off     - position in file to get data from.
4494  *              len     - amount of data to retrieve.
4495  *              plsz    - length of provided page list.
4496  *              seg     - segment to obtain pages for.
4497  *              addr    - virtual address of fault.
4498  *              rw      - mode of created pages.
4499  *              cr      - credentials of caller.
4500  *              ct      - caller context.
4501  *
4502  *      OUT:    protp   - protection mode of created pages.
4503  *              pl      - list of pages created.
4504  *
4505  *      RETURN: 0 on success, error code on failure.
4506  *
4507  * Timestamps:
4508  *      vp - atime updated
4509  */
4510 /* ARGSUSED */
4511 static int
4512 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4513     page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4514     enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4515 {
4516         znode_t         *zp = VTOZ(vp);
4517         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4518         page_t          **pl0 = pl;
4519         int             err = 0;
4520 
4521         /* we do our own caching, faultahead is unnecessary */
4522         if (pl == NULL)
4523                 return (0);
4524         else if (len > plsz)
4525                 len = plsz;
4526         else
4527                 len = P2ROUNDUP(len, PAGESIZE);
4528         ASSERT(plsz >= len);
4529 
4530         ZFS_ENTER(zfsvfs);
4531         ZFS_VERIFY_ZP(zp);
4532 
4533         if (protp)
4534                 *protp = PROT_ALL;
4535 
4536         /*
4537          * Loop through the requested range [off, off + len) looking
4538          * for pages.  If we don't find a page, we will need to create
4539          * a new page and fill it with data from the file.
4540          */
4541         while (len > 0) {
4542                 if (*pl = page_lookup(vp, off, SE_SHARED))
4543                         *(pl+1) = NULL;
4544                 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4545                         goto out;
4546                 while (*pl) {
4547                         ASSERT3U((*pl)->p_offset, ==, off);
4548                         off += PAGESIZE;
4549                         addr += PAGESIZE;
4550                         if (len > 0) {
4551                                 ASSERT3U(len, >=, PAGESIZE);
4552                                 len -= PAGESIZE;
4553                         }
4554                         ASSERT3U(plsz, >=, PAGESIZE);
4555                         plsz -= PAGESIZE;
4556                         pl++;
4557                 }
4558         }
4559 
4560         /*
4561          * Fill out the page array with any pages already in the cache.
4562          */
4563         while (plsz > 0 &&
4564             (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4565                         off += PAGESIZE;
4566                         plsz -= PAGESIZE;
4567         }
4568 out:
4569         if (err) {
4570                 /*
4571                  * Release any pages we have previously locked.
4572                  */
4573                 while (pl > pl0)
4574                         page_unlock(*--pl);
4575         } else {
4576                 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4577         }
4578 
4579         *pl = NULL;
4580 
4581         ZFS_EXIT(zfsvfs);
4582         return (err);
4583 }
4584 
4585 /*
4586  * Request a memory map for a section of a file.  This code interacts
4587  * with common code and the VM system as follows:
4588  *
4589  * - common code calls mmap(), which ends up in smmap_common()
4590  * - this calls VOP_MAP(), which takes you into (say) zfs
4591  * - zfs_map() calls as_map(), passing segvn_create() as the callback
4592  * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4593  * - zfs_addmap() updates z_mapcnt
4594  */
4595 /*ARGSUSED*/
4596 static int
4597 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4598     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4599     caller_context_t *ct)
4600 {
4601         znode_t *zp = VTOZ(vp);
4602         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4603         segvn_crargs_t  vn_a;
4604         int             error;
4605 
4606         ZFS_ENTER(zfsvfs);
4607         ZFS_VERIFY_ZP(zp);
4608 
4609         if ((prot & PROT_WRITE) && (zp->z_pflags &
4610             (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4611                 ZFS_EXIT(zfsvfs);
4612                 return (SET_ERROR(EPERM));
4613         }
4614 
4615         if ((prot & (PROT_READ | PROT_EXEC)) &&
4616             (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4617                 ZFS_EXIT(zfsvfs);
4618                 return (SET_ERROR(EACCES));
4619         }
4620 
4621         if (vp->v_flag & VNOMAP) {
4622                 ZFS_EXIT(zfsvfs);
4623                 return (SET_ERROR(ENOSYS));
4624         }
4625 
4626         if (off < 0 || len > MAXOFFSET_T - off) {
4627                 ZFS_EXIT(zfsvfs);
4628                 return (SET_ERROR(ENXIO));
4629         }
4630 
4631         if (vp->v_type != VREG) {
4632                 ZFS_EXIT(zfsvfs);
4633                 return (SET_ERROR(ENODEV));
4634         }
4635 
4636         /*
4637          * If file is locked, disallow mapping.
4638          */
4639         if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4640                 ZFS_EXIT(zfsvfs);
4641                 return (SET_ERROR(EAGAIN));
4642         }
4643 
4644         as_rangelock(as);
4645         error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4646         if (error != 0) {
4647                 as_rangeunlock(as);
4648                 ZFS_EXIT(zfsvfs);
4649                 return (error);
4650         }
4651 
4652         vn_a.vp = vp;
4653         vn_a.offset = (u_offset_t)off;
4654         vn_a.type = flags & MAP_TYPE;
4655         vn_a.prot = prot;
4656         vn_a.maxprot = maxprot;
4657         vn_a.cred = cr;
4658         vn_a.amp = NULL;
4659         vn_a.flags = flags & ~MAP_TYPE;
4660         vn_a.szc = 0;
4661         vn_a.lgrp_mem_policy_flags = 0;
4662 
4663         error = as_map(as, *addrp, len, segvn_create, &vn_a);
4664 
4665         as_rangeunlock(as);
4666         ZFS_EXIT(zfsvfs);
4667         return (error);
4668 }
4669 
4670 /* ARGSUSED */
4671 static int
4672 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4673     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4674     caller_context_t *ct)
4675 {
4676         uint64_t pages = btopr(len);
4677 
4678         atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4679         return (0);
4680 }
4681 
4682 /*
4683  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4684  * more accurate mtime for the associated file.  Since we don't have a way of
4685  * detecting when the data was actually modified, we have to resort to
4686  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4687  * last page is pushed.  The problem occurs when the msync() call is omitted,
4688  * which by far the most common case:
4689  *
4690  *      open()
4691  *      mmap()
4692  *      <modify memory>
4693  *      munmap()
4694  *      close()
4695  *      <time lapse>
4696  *      putpage() via fsflush
4697  *
4698  * If we wait until fsflush to come along, we can have a modification time that
4699  * is some arbitrary point in the future.  In order to prevent this in the
4700  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4701  * torn down.
4702  */
4703 /* ARGSUSED */
4704 static int
4705 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4706     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4707     caller_context_t *ct)
4708 {
4709         uint64_t pages = btopr(len);
4710 
4711         ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4712         atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4713 
4714         if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4715             vn_has_cached_data(vp))
4716                 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4717 
4718         return (0);
4719 }
4720 
4721 /*
4722  * Free or allocate space in a file.  Currently, this function only
4723  * supports the `F_FREESP' command.  However, this command is somewhat
4724  * misnamed, as its functionality includes the ability to allocate as
4725  * well as free space.
4726  *
4727  *      IN:     vp      - vnode of file to free data in.
4728  *              cmd     - action to take (only F_FREESP supported).
4729  *              bfp     - section of file to free/alloc.
4730  *              flag    - current file open mode flags.
4731  *              offset  - current file offset.
4732  *              cr      - credentials of caller [UNUSED].
4733  *              ct      - caller context.
4734  *
4735  *      RETURN: 0 on success, error code on failure.
4736  *
4737  * Timestamps:
4738  *      vp - ctime|mtime updated
4739  */
4740 /* ARGSUSED */
4741 static int
4742 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4743     offset_t offset, cred_t *cr, caller_context_t *ct)
4744 {
4745         znode_t         *zp = VTOZ(vp);
4746         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4747         uint64_t        off, len;
4748         int             error;
4749 
4750         ZFS_ENTER(zfsvfs);
4751         ZFS_VERIFY_ZP(zp);
4752 
4753         if (cmd != F_FREESP) {
4754                 ZFS_EXIT(zfsvfs);
4755                 return (SET_ERROR(EINVAL));
4756         }
4757 
4758         if (error = convoff(vp, bfp, 0, offset)) {
4759                 ZFS_EXIT(zfsvfs);
4760                 return (error);
4761         }
4762 
4763         if (bfp->l_len < 0) {
4764                 ZFS_EXIT(zfsvfs);
4765                 return (SET_ERROR(EINVAL));
4766         }
4767 
4768         off = bfp->l_start;
4769         len = bfp->l_len; /* 0 means from off to end of file */
4770 
4771         error = zfs_freesp(zp, off, len, flag, TRUE);
4772 
4773         ZFS_EXIT(zfsvfs);
4774         return (error);
4775 }
4776 
4777 /*ARGSUSED*/
4778 static int
4779 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4780 {
4781         znode_t         *zp = VTOZ(vp);
4782         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4783         uint32_t        gen;
4784         uint64_t        gen64;
4785         uint64_t        object = zp->z_id;
4786         zfid_short_t    *zfid;
4787         int             size, i, error;
4788 
4789         ZFS_ENTER(zfsvfs);
4790         ZFS_VERIFY_ZP(zp);
4791 
4792         if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4793             &gen64, sizeof (uint64_t))) != 0) {
4794                 ZFS_EXIT(zfsvfs);
4795                 return (error);
4796         }
4797 
4798         gen = (uint32_t)gen64;
4799 
4800         size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4801         if (fidp->fid_len < size) {
4802                 fidp->fid_len = size;
4803                 ZFS_EXIT(zfsvfs);
4804                 return (SET_ERROR(ENOSPC));
4805         }
4806 
4807         zfid = (zfid_short_t *)fidp;
4808 
4809         zfid->zf_len = size;
4810 
4811         for (i = 0; i < sizeof (zfid->zf_object); i++)
4812                 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4813 
4814         /* Must have a non-zero generation number to distinguish from .zfs */
4815         if (gen == 0)
4816                 gen = 1;
4817         for (i = 0; i < sizeof (zfid->zf_gen); i++)
4818                 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4819 
4820         if (size == LONG_FID_LEN) {
4821                 uint64_t        objsetid = dmu_objset_id(zfsvfs->z_os);
4822                 zfid_long_t     *zlfid;
4823 
4824                 zlfid = (zfid_long_t *)fidp;
4825 
4826                 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4827                         zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4828 
4829                 /* XXX - this should be the generation number for the objset */
4830                 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4831                         zlfid->zf_setgen[i] = 0;
4832         }
4833 
4834         ZFS_EXIT(zfsvfs);
4835         return (0);
4836 }
4837 
4838 static int
4839 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4840     caller_context_t *ct)
4841 {
4842         znode_t         *zp, *xzp;
4843         zfsvfs_t        *zfsvfs;
4844         zfs_dirlock_t   *dl;
4845         int             error;
4846 
4847         switch (cmd) {
4848         case _PC_LINK_MAX:
4849                 *valp = ULONG_MAX;
4850                 return (0);
4851 
4852         case _PC_FILESIZEBITS:
4853                 *valp = 64;
4854                 return (0);
4855 
4856         case _PC_XATTR_EXISTS:
4857                 zp = VTOZ(vp);
4858                 zfsvfs = zp->z_zfsvfs;
4859                 ZFS_ENTER(zfsvfs);
4860                 ZFS_VERIFY_ZP(zp);
4861                 *valp = 0;
4862                 error = zfs_dirent_lock(&dl, zp, "", &xzp,
4863                     ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4864                 if (error == 0) {
4865                         zfs_dirent_unlock(dl);
4866                         if (!zfs_dirempty(xzp))
4867                                 *valp = 1;
4868                         VN_RELE(ZTOV(xzp));
4869                 } else if (error == ENOENT) {
4870                         /*
4871                          * If there aren't extended attributes, it's the
4872                          * same as having zero of them.
4873                          */
4874                         error = 0;
4875                 }
4876                 ZFS_EXIT(zfsvfs);
4877                 return (error);
4878 
4879         case _PC_SATTR_ENABLED:
4880         case _PC_SATTR_EXISTS:
4881                 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4882                     (vp->v_type == VREG || vp->v_type == VDIR);
4883                 return (0);
4884 
4885         case _PC_ACCESS_FILTERING:
4886                 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4887                     vp->v_type == VDIR;
4888                 return (0);
4889 
4890         case _PC_ACL_ENABLED:
4891                 *valp = _ACL_ACE_ENABLED;
4892                 return (0);
4893 
4894         case _PC_MIN_HOLE_SIZE:
4895                 *valp = (ulong_t)SPA_MINBLOCKSIZE;
4896                 return (0);
4897 
4898         case _PC_TIMESTAMP_RESOLUTION:
4899                 /* nanosecond timestamp resolution */
4900                 *valp = 1L;
4901                 return (0);
4902 
4903         default:
4904                 return (fs_pathconf(vp, cmd, valp, cr, ct));
4905         }
4906 }
4907 
4908 /*ARGSUSED*/
4909 static int
4910 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4911     caller_context_t *ct)
4912 {
4913         znode_t *zp = VTOZ(vp);
4914         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4915         int error;
4916         boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4917 
4918         ZFS_ENTER(zfsvfs);
4919         ZFS_VERIFY_ZP(zp);
4920         error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4921         ZFS_EXIT(zfsvfs);
4922 
4923         return (error);
4924 }
4925 
4926 /*ARGSUSED*/
4927 static int
4928 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4929     caller_context_t *ct)
4930 {
4931         znode_t *zp = VTOZ(vp);
4932         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4933         int error;
4934         boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4935         zilog_t *zilog = zfsvfs->z_log;
4936 
4937         ZFS_ENTER(zfsvfs);
4938         ZFS_VERIFY_ZP(zp);
4939 
4940         error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4941 
4942         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4943                 zil_commit(zilog, 0);
4944 
4945         ZFS_EXIT(zfsvfs);
4946         return (error);
4947 }
4948 
4949 /*
4950  * The smallest read we may consider to loan out an arcbuf.
4951  * This must be a power of 2.
4952  */
4953 int zcr_blksz_min = (1 << 10);    /* 1K */
4954 /*
4955  * If set to less than the file block size, allow loaning out of an
4956  * arcbuf for a partial block read.  This must be a power of 2.
4957  */
4958 int zcr_blksz_max = (1 << 17);    /* 128K */
4959 
4960 /*ARGSUSED*/
4961 static int
4962 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
4963     caller_context_t *ct)
4964 {
4965         znode_t *zp = VTOZ(vp);
4966         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4967         int max_blksz = zfsvfs->z_max_blksz;
4968         uio_t *uio = &xuio->xu_uio;
4969         ssize_t size = uio->uio_resid;
4970         offset_t offset = uio->uio_loffset;
4971         int blksz;
4972         int fullblk, i;
4973         arc_buf_t *abuf;
4974         ssize_t maxsize;
4975         int preamble, postamble;
4976 
4977         if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4978                 return (SET_ERROR(EINVAL));
4979 
4980         ZFS_ENTER(zfsvfs);
4981         ZFS_VERIFY_ZP(zp);
4982         switch (ioflag) {
4983         case UIO_WRITE:
4984                 /*
4985                  * Loan out an arc_buf for write if write size is bigger than
4986                  * max_blksz, and the file's block size is also max_blksz.
4987                  */
4988                 blksz = max_blksz;
4989                 if (size < blksz || zp->z_blksz != blksz) {
4990                         ZFS_EXIT(zfsvfs);
4991                         return (SET_ERROR(EINVAL));
4992                 }
4993                 /*
4994                  * Caller requests buffers for write before knowing where the
4995                  * write offset might be (e.g. NFS TCP write).
4996                  */
4997                 if (offset == -1) {
4998                         preamble = 0;
4999                 } else {
5000                         preamble = P2PHASE(offset, blksz);
5001                         if (preamble) {
5002                                 preamble = blksz - preamble;
5003                                 size -= preamble;
5004                         }
5005                 }
5006 
5007                 postamble = P2PHASE(size, blksz);
5008                 size -= postamble;
5009 
5010                 fullblk = size / blksz;
5011                 (void) dmu_xuio_init(xuio,
5012                     (preamble != 0) + fullblk + (postamble != 0));
5013                 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5014                     int, postamble, int,
5015                     (preamble != 0) + fullblk + (postamble != 0));
5016 
5017                 /*
5018                  * Have to fix iov base/len for partial buffers.  They
5019                  * currently represent full arc_buf's.
5020                  */
5021                 if (preamble) {
5022                         /* data begins in the middle of the arc_buf */
5023                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5024                             blksz);
5025                         ASSERT(abuf);
5026                         (void) dmu_xuio_add(xuio, abuf,
5027                             blksz - preamble, preamble);
5028                 }
5029 
5030                 for (i = 0; i < fullblk; i++) {
5031                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5032                             blksz);
5033                         ASSERT(abuf);
5034                         (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5035                 }
5036 
5037                 if (postamble) {
5038                         /* data ends in the middle of the arc_buf */
5039                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5040                             blksz);
5041                         ASSERT(abuf);
5042                         (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5043                 }
5044                 break;
5045         case UIO_READ:
5046                 /*
5047                  * Loan out an arc_buf for read if the read size is larger than
5048                  * the current file block size.  Block alignment is not
5049                  * considered.  Partial arc_buf will be loaned out for read.
5050                  */
5051                 blksz = zp->z_blksz;
5052                 if (blksz < zcr_blksz_min)
5053                         blksz = zcr_blksz_min;
5054                 if (blksz > zcr_blksz_max)
5055                         blksz = zcr_blksz_max;
5056                 /* avoid potential complexity of dealing with it */
5057                 if (blksz > max_blksz) {
5058                         ZFS_EXIT(zfsvfs);
5059                         return (SET_ERROR(EINVAL));
5060                 }
5061 
5062                 maxsize = zp->z_size - uio->uio_loffset;
5063                 if (size > maxsize)
5064                         size = maxsize;
5065 
5066                 if (size < blksz || vn_has_cached_data(vp)) {
5067                         ZFS_EXIT(zfsvfs);
5068                         return (SET_ERROR(EINVAL));
5069                 }
5070                 break;
5071         default:
5072                 ZFS_EXIT(zfsvfs);
5073                 return (SET_ERROR(EINVAL));
5074         }
5075 
5076         uio->uio_extflg = UIO_XUIO;
5077         XUIO_XUZC_RW(xuio) = ioflag;
5078         ZFS_EXIT(zfsvfs);
5079         return (0);
5080 }
5081 
5082 /*ARGSUSED*/
5083 static int
5084 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5085 {
5086         int i;
5087         arc_buf_t *abuf;
5088         int ioflag = XUIO_XUZC_RW(xuio);
5089 
5090         ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5091 
5092         i = dmu_xuio_cnt(xuio);
5093         while (i-- > 0) {
5094                 abuf = dmu_xuio_arcbuf(xuio, i);
5095                 /*
5096                  * if abuf == NULL, it must be a write buffer
5097                  * that has been returned in zfs_write().
5098                  */
5099                 if (abuf)
5100                         dmu_return_arcbuf(abuf);
5101                 ASSERT(abuf || ioflag == UIO_WRITE);
5102         }
5103 
5104         dmu_xuio_fini(xuio);
5105         return (0);
5106 }
5107 
5108 /*
5109  * Predeclare these here so that the compiler assumes that
5110  * this is an "old style" function declaration that does
5111  * not include arguments => we won't get type mismatch errors
5112  * in the initializations that follow.
5113  */
5114 static int zfs_inval();
5115 static int zfs_isdir();
5116 
5117 static int
5118 zfs_inval()
5119 {
5120         return (SET_ERROR(EINVAL));
5121 }
5122 
5123 static int
5124 zfs_isdir()
5125 {
5126         return (SET_ERROR(EISDIR));
5127 }
5128 /*
5129  * Directory vnode operations template
5130  */
5131 vnodeops_t *zfs_dvnodeops;
5132 const fs_operation_def_t zfs_dvnodeops_template[] = {
5133         VOPNAME_OPEN,           { .vop_open = zfs_open },
5134         VOPNAME_CLOSE,          { .vop_close = zfs_close },
5135         VOPNAME_READ,           { .error = zfs_isdir },
5136         VOPNAME_WRITE,          { .error = zfs_isdir },
5137         VOPNAME_IOCTL,          { .vop_ioctl = zfs_ioctl },
5138         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5139         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5140         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5141         VOPNAME_LOOKUP,         { .vop_lookup = zfs_lookup },
5142         VOPNAME_CREATE,         { .vop_create = zfs_create },
5143         VOPNAME_REMOVE,         { .vop_remove = zfs_remove },
5144         VOPNAME_LINK,           { .vop_link = zfs_link },
5145         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5146         VOPNAME_MKDIR,          { .vop_mkdir = zfs_mkdir },
5147         VOPNAME_RMDIR,          { .vop_rmdir = zfs_rmdir },
5148         VOPNAME_READDIR,        { .vop_readdir = zfs_readdir },
5149         VOPNAME_SYMLINK,        { .vop_symlink = zfs_symlink },
5150         VOPNAME_FSYNC,          { .vop_fsync = zfs_fsync },
5151         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5152         VOPNAME_FID,            { .vop_fid = zfs_fid },
5153         VOPNAME_SEEK,           { .vop_seek = zfs_seek },
5154         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5155         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5156         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5157         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5158         NULL,                   NULL
5159 };
5160 
5161 /*
5162  * Regular file vnode operations template
5163  */
5164 vnodeops_t *zfs_fvnodeops;
5165 const fs_operation_def_t zfs_fvnodeops_template[] = {
5166         VOPNAME_OPEN,           { .vop_open = zfs_open },
5167         VOPNAME_CLOSE,          { .vop_close = zfs_close },
5168         VOPNAME_READ,           { .vop_read = zfs_read },
5169         VOPNAME_WRITE,          { .vop_write = zfs_write },
5170         VOPNAME_IOCTL,          { .vop_ioctl = zfs_ioctl },
5171         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5172         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5173         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5174         VOPNAME_LOOKUP,         { .vop_lookup = zfs_lookup },
5175         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5176         VOPNAME_FSYNC,          { .vop_fsync = zfs_fsync },
5177         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5178         VOPNAME_FID,            { .vop_fid = zfs_fid },
5179         VOPNAME_SEEK,           { .vop_seek = zfs_seek },
5180         VOPNAME_FRLOCK,         { .vop_frlock = zfs_frlock },
5181         VOPNAME_SPACE,          { .vop_space = zfs_space },
5182         VOPNAME_GETPAGE,        { .vop_getpage = zfs_getpage },
5183         VOPNAME_PUTPAGE,        { .vop_putpage = zfs_putpage },
5184         VOPNAME_MAP,            { .vop_map = zfs_map },
5185         VOPNAME_ADDMAP,         { .vop_addmap = zfs_addmap },
5186         VOPNAME_DELMAP,         { .vop_delmap = zfs_delmap },
5187         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5188         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5189         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5190         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5191         VOPNAME_REQZCBUF,       { .vop_reqzcbuf = zfs_reqzcbuf },
5192         VOPNAME_RETZCBUF,       { .vop_retzcbuf = zfs_retzcbuf },
5193         NULL,                   NULL
5194 };
5195 
5196 /*
5197  * Symbolic link vnode operations template
5198  */
5199 vnodeops_t *zfs_symvnodeops;
5200 const fs_operation_def_t zfs_symvnodeops_template[] = {
5201         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5202         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5203         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5204         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5205         VOPNAME_READLINK,       { .vop_readlink = zfs_readlink },
5206         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5207         VOPNAME_FID,            { .vop_fid = zfs_fid },
5208         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5209         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5210         NULL,                   NULL
5211 };
5212 
5213 /*
5214  * special share hidden files vnode operations template
5215  */
5216 vnodeops_t *zfs_sharevnodeops;
5217 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5218         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5219         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5220         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5221         VOPNAME_FID,            { .vop_fid = zfs_fid },
5222         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5223         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5224         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5225         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5226         NULL,                   NULL
5227 };
5228 
5229 /*
5230  * Extended attribute directory vnode operations template
5231  *
5232  * This template is identical to the directory vnodes
5233  * operation template except for restricted operations:
5234  *      VOP_MKDIR()
5235  *      VOP_SYMLINK()
5236  *
5237  * Note that there are other restrictions embedded in:
5238  *      zfs_create()    - restrict type to VREG
5239  *      zfs_link()      - no links into/out of attribute space
5240  *      zfs_rename()    - no moves into/out of attribute space
5241  */
5242 vnodeops_t *zfs_xdvnodeops;
5243 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5244         VOPNAME_OPEN,           { .vop_open = zfs_open },
5245         VOPNAME_CLOSE,          { .vop_close = zfs_close },
5246         VOPNAME_IOCTL,          { .vop_ioctl = zfs_ioctl },
5247         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5248         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5249         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5250         VOPNAME_LOOKUP,         { .vop_lookup = zfs_lookup },
5251         VOPNAME_CREATE,         { .vop_create = zfs_create },
5252         VOPNAME_REMOVE,         { .vop_remove = zfs_remove },
5253         VOPNAME_LINK,           { .vop_link = zfs_link },
5254         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5255         VOPNAME_MKDIR,          { .error = zfs_inval },
5256         VOPNAME_RMDIR,          { .vop_rmdir = zfs_rmdir },
5257         VOPNAME_READDIR,        { .vop_readdir = zfs_readdir },
5258         VOPNAME_SYMLINK,        { .error = zfs_inval },
5259         VOPNAME_FSYNC,          { .vop_fsync = zfs_fsync },
5260         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5261         VOPNAME_FID,            { .vop_fid = zfs_fid },
5262         VOPNAME_SEEK,           { .vop_seek = zfs_seek },
5263         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5264         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5265         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5266         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5267         NULL,                   NULL
5268 };
5269 
5270 /*
5271  * Error vnode operations template
5272  */
5273 vnodeops_t *zfs_evnodeops;
5274 const fs_operation_def_t zfs_evnodeops_template[] = {
5275         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5276         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5277         NULL,                   NULL
5278 };