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) 2012, 2014 by Delphix. All rights reserved.
  24  * Copyright 2014 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) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
 115  *      dmu_tx_assign().  This is critical because we don't want to block
 116  *      while holding locks.
 117  *
 118  *      If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT.  This
 119  *      reduces lock contention and CPU usage when we must wait (note that if
 120  *      throughput is constrained by the storage, nearly every transaction
 121  *      must wait).
 122  *
 123  *      Note, in particular, that if a lock is sometimes acquired before
 124  *      the tx assigns, and sometimes after (e.g. z_lock), then failing
 125  *      to use a non-blocking assign can deadlock the system.  The scenario:
 126  *
 127  *      Thread A has grabbed a lock before calling dmu_tx_assign().
 128  *      Thread B is in an already-assigned tx, and blocks for this lock.
 129  *      Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
 130  *      forever, because the previous txg can't quiesce until B's tx commits.
 131  *
 132  *      If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
 133  *      then drop all locks, call dmu_tx_wait(), and try again.  On subsequent
 134  *      calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
 135  *      to indicate that this operation has already called dmu_tx_wait().
 136  *      This will ensure that we don't retry forever, waiting a short bit
 137  *      each time.
 138  *
 139  *  (5) If the operation succeeded, generate the intent log entry for it
 140  *      before dropping locks.  This ensures that the ordering of events
 141  *      in the intent log matches the order in which they actually occurred.
 142  *      During ZIL replay the zfs_log_* functions will update the sequence
 143  *      number to indicate the zil transaction has replayed.
 144  *
 145  *  (6) At the end of each vnode op, the DMU tx must always commit,
 146  *      regardless of whether there were any errors.
 147  *
 148  *  (7) After dropping all locks, invoke zil_commit(zilog, foid)
 149  *      to ensure that synchronous semantics are provided when necessary.
 150  *
 151  * In general, this is how things should be ordered in each vnode op:
 152  *
 153  *      ZFS_ENTER(zfsvfs);              // exit if unmounted
 154  * top:
 155  *      zfs_dirent_lock(&dl, ...)   // lock directory entry (may VN_HOLD())
 156  *      rw_enter(...);                  // grab any other locks you need
 157  *      tx = dmu_tx_create(...);        // get DMU tx
 158  *      dmu_tx_hold_*();                // hold each object you might modify
 159  *      error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
 160  *      if (error) {
 161  *              rw_exit(...);           // drop locks
 162  *              zfs_dirent_unlock(dl);  // unlock directory entry
 163  *              VN_RELE(...);           // release held vnodes
 164  *              if (error == ERESTART) {
 165  *                      waited = B_TRUE;
 166  *                      dmu_tx_wait(tx);
 167  *                      dmu_tx_abort(tx);
 168  *                      goto top;
 169  *              }
 170  *              dmu_tx_abort(tx);       // abort DMU tx
 171  *              ZFS_EXIT(zfsvfs);       // finished in zfs
 172  *              return (error);         // really out of space
 173  *      }
 174  *      error = do_real_work();         // do whatever this VOP does
 175  *      if (error == 0)
 176  *              zfs_log_*(...);         // on success, make ZIL entry
 177  *      dmu_tx_commit(tx);              // commit DMU tx -- error or not
 178  *      rw_exit(...);                   // drop locks
 179  *      zfs_dirent_unlock(dl);          // unlock directory entry
 180  *      VN_RELE(...);                   // release held vnodes
 181  *      zil_commit(zilog, foid);        // synchronous when necessary
 182  *      ZFS_EXIT(zfsvfs);               // finished in zfs
 183  *      return (error);                 // done, report error
 184  */
 185 
 186 /* ARGSUSED */
 187 static int
 188 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
 189 {
 190         znode_t *zp = VTOZ(*vpp);
 191         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 192 
 193         ZFS_ENTER(zfsvfs);
 194         ZFS_VERIFY_ZP(zp);
 195 
 196         if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
 197             ((flag & FAPPEND) == 0)) {
 198                 ZFS_EXIT(zfsvfs);
 199                 return (SET_ERROR(EPERM));
 200         }
 201 
 202         if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
 203             ZTOV(zp)->v_type == VREG &&
 204             !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
 205                 if (fs_vscan(*vpp, cr, 0) != 0) {
 206                         ZFS_EXIT(zfsvfs);
 207                         return (SET_ERROR(EACCES));
 208                 }
 209         }
 210 
 211         /* Keep a count of the synchronous opens in the znode */
 212         if (flag & (FSYNC | FDSYNC))
 213                 atomic_inc_32(&zp->z_sync_cnt);
 214 
 215         ZFS_EXIT(zfsvfs);
 216         return (0);
 217 }
 218 
 219 /* ARGSUSED */
 220 static int
 221 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
 222     caller_context_t *ct)
 223 {
 224         znode_t *zp = VTOZ(vp);
 225         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 226 
 227         /*
 228          * Clean up any locks held by this process on the vp.
 229          */
 230         cleanlocks(vp, ddi_get_pid(), 0);
 231         cleanshares(vp, ddi_get_pid());
 232 
 233         ZFS_ENTER(zfsvfs);
 234         ZFS_VERIFY_ZP(zp);
 235 
 236         /* Decrement the synchronous opens in the znode */
 237         if ((flag & (FSYNC | FDSYNC)) && (count == 1))
 238                 atomic_dec_32(&zp->z_sync_cnt);
 239 
 240         if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
 241             ZTOV(zp)->v_type == VREG &&
 242             !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
 243                 VERIFY(fs_vscan(vp, cr, 1) == 0);
 244 
 245         ZFS_EXIT(zfsvfs);
 246         return (0);
 247 }
 248 
 249 /*
 250  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
 251  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
 252  */
 253 static int
 254 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
 255 {
 256         znode_t *zp = VTOZ(vp);
 257         uint64_t noff = (uint64_t)*off; /* new offset */
 258         uint64_t file_sz;
 259         int error;
 260         boolean_t hole;
 261 
 262         file_sz = zp->z_size;
 263         if (noff >= file_sz)  {
 264                 return (SET_ERROR(ENXIO));
 265         }
 266 
 267         if (cmd == _FIO_SEEK_HOLE)
 268                 hole = B_TRUE;
 269         else
 270                 hole = B_FALSE;
 271 
 272         error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
 273 
 274         if (error == ESRCH)
 275                 return (SET_ERROR(ENXIO));
 276 
 277         /*
 278          * We could find a hole that begins after the logical end-of-file,
 279          * because dmu_offset_next() only works on whole blocks.  If the
 280          * EOF falls mid-block, then indicate that the "virtual hole"
 281          * at the end of the file begins at the logical EOF, rather than
 282          * at the end of the last block.
 283          */
 284         if (noff > file_sz) {
 285                 ASSERT(hole);
 286                 noff = file_sz;
 287         }
 288 
 289         if (noff < *off)
 290                 return (error);
 291         *off = noff;
 292         return (error);
 293 }
 294 
 295 /* ARGSUSED */
 296 static int
 297 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
 298     int *rvalp, caller_context_t *ct)
 299 {
 300         offset_t off;
 301         offset_t ndata;
 302         dmu_object_info_t doi;
 303         int error;
 304         zfsvfs_t *zfsvfs;
 305         znode_t *zp;
 306 
 307         switch (com) {
 308         case _FIOFFS:
 309         {
 310                 return (zfs_sync(vp->v_vfsp, 0, cred));
 311 
 312                 /*
 313                  * The following two ioctls are used by bfu.  Faking out,
 314                  * necessary to avoid bfu errors.
 315                  */
 316         }
 317         case _FIOGDIO:
 318         case _FIOSDIO:
 319         {
 320                 return (0);
 321         }
 322 
 323         case _FIO_SEEK_DATA:
 324         case _FIO_SEEK_HOLE:
 325         {
 326                 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
 327                         return (SET_ERROR(EFAULT));
 328 
 329                 zp = VTOZ(vp);
 330                 zfsvfs = zp->z_zfsvfs;
 331                 ZFS_ENTER(zfsvfs);
 332                 ZFS_VERIFY_ZP(zp);
 333 
 334                 /* offset parameter is in/out */
 335                 error = zfs_holey(vp, com, &off);
 336                 ZFS_EXIT(zfsvfs);
 337                 if (error)
 338                         return (error);
 339                 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
 340                         return (SET_ERROR(EFAULT));
 341                 return (0);
 342         }
 343         case _FIO_COUNT_FILLED:
 344         {
 345                 /*
 346                  * _FIO_COUNT_FILLED adds a new ioctl command which
 347                  * exposes the number of filled blocks in a
 348                  * ZFS object.
 349                  */
 350                 zp = VTOZ(vp);
 351                 zfsvfs = zp->z_zfsvfs;
 352                 ZFS_ENTER(zfsvfs);
 353                 ZFS_VERIFY_ZP(zp);
 354 
 355                 /*
 356                  * Wait for all dirty blocks for this object
 357                  * to get synced out to disk, and the DMU info
 358                  * updated.
 359                  */
 360                 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
 361                 if (error) {
 362                         ZFS_EXIT(zfsvfs);
 363                         return (error);
 364                 }
 365 
 366                 /*
 367                  * Retrieve fill count from DMU object.
 368                  */
 369                 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
 370                 if (error) {
 371                         ZFS_EXIT(zfsvfs);
 372                         return (error);
 373                 }
 374 
 375                 ndata = doi.doi_fill_count;
 376 
 377                 ZFS_EXIT(zfsvfs);
 378                 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
 379                         return (SET_ERROR(EFAULT));
 380                 return (0);
 381         }
 382         }
 383         return (SET_ERROR(ENOTTY));
 384 }
 385 
 386 /*
 387  * Utility functions to map and unmap a single physical page.  These
 388  * are used to manage the mappable copies of ZFS file data, and therefore
 389  * do not update ref/mod bits.
 390  */
 391 caddr_t
 392 zfs_map_page(page_t *pp, enum seg_rw rw)
 393 {
 394         if (kpm_enable)
 395                 return (hat_kpm_mapin(pp, 0));
 396         ASSERT(rw == S_READ || rw == S_WRITE);
 397         return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
 398             (caddr_t)-1));
 399 }
 400 
 401 void
 402 zfs_unmap_page(page_t *pp, caddr_t addr)
 403 {
 404         if (kpm_enable) {
 405                 hat_kpm_mapout(pp, 0, addr);
 406         } else {
 407                 ppmapout(addr);
 408         }
 409 }
 410 
 411 /*
 412  * When a file is memory mapped, we must keep the IO data synchronized
 413  * between the DMU cache and the memory mapped pages.  What this means:
 414  *
 415  * On Write:    If we find a memory mapped page, we write to *both*
 416  *              the page and the dmu buffer.
 417  */
 418 static void
 419 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
 420 {
 421         int64_t off;
 422 
 423         off = start & PAGEOFFSET;
 424         for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
 425                 page_t *pp;
 426                 uint64_t nbytes = MIN(PAGESIZE - off, len);
 427 
 428                 if (pp = page_lookup(vp, start, SE_SHARED)) {
 429                         caddr_t va;
 430 
 431                         va = zfs_map_page(pp, S_WRITE);
 432                         (void) dmu_read(os, oid, start+off, nbytes, va+off,
 433                             DMU_READ_PREFETCH);
 434                         zfs_unmap_page(pp, va);
 435                         page_unlock(pp);
 436                 }
 437                 len -= nbytes;
 438                 off = 0;
 439         }
 440 }
 441 
 442 /*
 443  * When a file is memory mapped, we must keep the IO data synchronized
 444  * between the DMU cache and the memory mapped pages.  What this means:
 445  *
 446  * On Read:     We "read" preferentially from memory mapped pages,
 447  *              else we default from the dmu buffer.
 448  *
 449  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
 450  *       the file is memory mapped.
 451  */
 452 static int
 453 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
 454 {
 455         znode_t *zp = VTOZ(vp);
 456         int64_t start, off;
 457         int len = nbytes;
 458         int error = 0;
 459 
 460         start = uio->uio_loffset;
 461         off = start & PAGEOFFSET;
 462         for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
 463                 page_t *pp;
 464                 uint64_t bytes = MIN(PAGESIZE - off, len);
 465 
 466                 if (pp = page_lookup(vp, start, SE_SHARED)) {
 467                         caddr_t va;
 468 
 469                         va = zfs_map_page(pp, S_READ);
 470                         error = uiomove(va + off, bytes, UIO_READ, uio);
 471                         zfs_unmap_page(pp, va);
 472                         page_unlock(pp);
 473                 } else {
 474                         error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
 475                             uio, bytes);
 476                 }
 477                 len -= bytes;
 478                 off = 0;
 479                 if (error)
 480                         break;
 481         }
 482         return (error);
 483 }
 484 
 485 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
 486 
 487 /*
 488  * Read bytes from specified file into supplied buffer.
 489  *
 490  *      IN:     vp      - vnode of file to be read from.
 491  *              uio     - structure supplying read location, range info,
 492  *                        and return buffer.
 493  *              ioflag  - SYNC flags; used to provide FRSYNC semantics.
 494  *              cr      - credentials of caller.
 495  *              ct      - caller context
 496  *
 497  *      OUT:    uio     - updated offset and range, buffer filled.
 498  *
 499  *      RETURN: 0 on success, error code on failure.
 500  *
 501  * Side Effects:
 502  *      vp - atime updated if byte count > 0
 503  */
 504 /* ARGSUSED */
 505 static int
 506 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 507 {
 508         znode_t         *zp = VTOZ(vp);
 509         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
 510         ssize_t         n, nbytes;
 511         int             error = 0;
 512         rl_t            *rl;
 513         xuio_t          *xuio = NULL;
 514 
 515         ZFS_ENTER(zfsvfs);
 516         ZFS_VERIFY_ZP(zp);
 517 
 518         if (zp->z_pflags & ZFS_AV_QUARANTINED) {
 519                 ZFS_EXIT(zfsvfs);
 520                 return (SET_ERROR(EACCES));
 521         }
 522 
 523         /*
 524          * Validate file offset
 525          */
 526         if (uio->uio_loffset < (offset_t)0) {
 527                 ZFS_EXIT(zfsvfs);
 528                 return (SET_ERROR(EINVAL));
 529         }
 530 
 531         /*
 532          * Fasttrack empty reads
 533          */
 534         if (uio->uio_resid == 0) {
 535                 ZFS_EXIT(zfsvfs);
 536                 return (0);
 537         }
 538 
 539         /*
 540          * Check for mandatory locks
 541          */
 542         if (MANDMODE(zp->z_mode)) {
 543                 if (error = chklock(vp, FREAD,
 544                     uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
 545                         ZFS_EXIT(zfsvfs);
 546                         return (error);
 547                 }
 548         }
 549 
 550         /*
 551          * If we're in FRSYNC mode, sync out this znode before reading it.
 552          */
 553         if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
 554                 zil_commit(zfsvfs->z_log, zp->z_id);
 555 
 556         /*
 557          * Lock the range against changes.
 558          */
 559         rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
 560 
 561         /*
 562          * If we are reading past end-of-file we can skip
 563          * to the end; but we might still need to set atime.
 564          */
 565         if (uio->uio_loffset >= zp->z_size) {
 566                 error = 0;
 567                 goto out;
 568         }
 569 
 570         ASSERT(uio->uio_loffset < zp->z_size);
 571         n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
 572 
 573         if ((uio->uio_extflg == UIO_XUIO) &&
 574             (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
 575                 int nblk;
 576                 int blksz = zp->z_blksz;
 577                 uint64_t offset = uio->uio_loffset;
 578 
 579                 xuio = (xuio_t *)uio;
 580                 if ((ISP2(blksz))) {
 581                         nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
 582                             blksz)) / blksz;
 583                 } else {
 584                         ASSERT(offset + n <= blksz);
 585                         nblk = 1;
 586                 }
 587                 (void) dmu_xuio_init(xuio, nblk);
 588 
 589                 if (vn_has_cached_data(vp)) {
 590                         /*
 591                          * For simplicity, we always allocate a full buffer
 592                          * even if we only expect to read a portion of a block.
 593                          */
 594                         while (--nblk >= 0) {
 595                                 (void) dmu_xuio_add(xuio,
 596                                     dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
 597                                     blksz), 0, blksz);
 598                         }
 599                 }
 600         }
 601 
 602         while (n > 0) {
 603                 nbytes = MIN(n, zfs_read_chunk_size -
 604                     P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
 605 
 606                 if (vn_has_cached_data(vp)) {
 607                         error = mappedread(vp, nbytes, uio);
 608                 } else {
 609                         error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
 610                             uio, nbytes);
 611                 }
 612                 if (error) {
 613                         /* convert checksum errors into IO errors */
 614                         if (error == ECKSUM)
 615                                 error = SET_ERROR(EIO);
 616                         break;
 617                 }
 618 
 619                 n -= nbytes;
 620         }
 621 out:
 622         zfs_range_unlock(rl);
 623 
 624         ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
 625         ZFS_EXIT(zfsvfs);
 626         return (error);
 627 }
 628 
 629 /*
 630  * Write the bytes to a file.
 631  *
 632  *      IN:     vp      - vnode of file to be written to.
 633  *              uio     - structure supplying write location, range info,
 634  *                        and data buffer.
 635  *              ioflag  - FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
 636  *                        set if in append mode.
 637  *              cr      - credentials of caller.
 638  *              ct      - caller context (NFS/CIFS fem monitor only)
 639  *
 640  *      OUT:    uio     - updated offset and range.
 641  *
 642  *      RETURN: 0 on success, error code on failure.
 643  *
 644  * Timestamps:
 645  *      vp - ctime|mtime updated if byte count > 0
 646  */
 647 
 648 /* ARGSUSED */
 649 static int
 650 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 651 {
 652         znode_t         *zp = VTOZ(vp);
 653         rlim64_t        limit = uio->uio_llimit;
 654         ssize_t         start_resid = uio->uio_resid;
 655         ssize_t         tx_bytes;
 656         uint64_t        end_size;
 657         dmu_tx_t        *tx;
 658         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
 659         zilog_t         *zilog;
 660         offset_t        woff;
 661         ssize_t         n, nbytes;
 662         rl_t            *rl;
 663         int             max_blksz = zfsvfs->z_max_blksz;
 664         int             error = 0;
 665         arc_buf_t       *abuf;
 666         iovec_t         *aiov = NULL;
 667         xuio_t          *xuio = NULL;
 668         int             i_iov = 0;
 669         int             iovcnt = uio->uio_iovcnt;
 670         iovec_t         *iovp = uio->uio_iov;
 671         int             write_eof;
 672         int             count = 0;
 673         sa_bulk_attr_t  bulk[4];
 674         uint64_t        mtime[2], ctime[2];
 675 
 676         /*
 677          * Fasttrack empty write
 678          */
 679         n = start_resid;
 680         if (n == 0)
 681                 return (0);
 682 
 683         if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
 684                 limit = MAXOFFSET_T;
 685 
 686         ZFS_ENTER(zfsvfs);
 687         ZFS_VERIFY_ZP(zp);
 688 
 689         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
 690         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
 691         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
 692             &zp->z_size, 8);
 693         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
 694             &zp->z_pflags, 8);
 695 
 696         /*
 697          * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
 698          * callers might not be able to detect properly that we are read-only,
 699          * so check it explicitly here.
 700          */
 701         if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
 702                 ZFS_EXIT(zfsvfs);
 703                 return (SET_ERROR(EROFS));
 704         }
 705 
 706         /*
 707          * If immutable or not appending then return EPERM
 708          */
 709         if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
 710             ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
 711             (uio->uio_loffset < zp->z_size))) {
 712                 ZFS_EXIT(zfsvfs);
 713                 return (SET_ERROR(EPERM));
 714         }
 715 
 716         zilog = zfsvfs->z_log;
 717 
 718         /*
 719          * Validate file offset
 720          */
 721         woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
 722         if (woff < 0) {
 723                 ZFS_EXIT(zfsvfs);
 724                 return (SET_ERROR(EINVAL));
 725         }
 726 
 727         /*
 728          * Check for mandatory locks before calling zfs_range_lock()
 729          * in order to prevent a deadlock with locks set via fcntl().
 730          */
 731         if (MANDMODE((mode_t)zp->z_mode) &&
 732             (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
 733                 ZFS_EXIT(zfsvfs);
 734                 return (error);
 735         }
 736 
 737         /*
 738          * Pre-fault the pages to ensure slow (eg NFS) pages
 739          * don't hold up txg.
 740          * Skip this if uio contains loaned arc_buf.
 741          */
 742         if ((uio->uio_extflg == UIO_XUIO) &&
 743             (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
 744                 xuio = (xuio_t *)uio;
 745         else
 746                 uio_prefaultpages(MIN(n, max_blksz), uio);
 747 
 748         /*
 749          * If in append mode, set the io offset pointer to eof.
 750          */
 751         if (ioflag & FAPPEND) {
 752                 /*
 753                  * Obtain an appending range lock to guarantee file append
 754                  * semantics.  We reset the write offset once we have the lock.
 755                  */
 756                 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
 757                 woff = rl->r_off;
 758                 if (rl->r_len == UINT64_MAX) {
 759                         /*
 760                          * We overlocked the file because this write will cause
 761                          * the file block size to increase.
 762                          * Note that zp_size cannot change with this lock held.
 763                          */
 764                         woff = zp->z_size;
 765                 }
 766                 uio->uio_loffset = woff;
 767         } else {
 768                 /*
 769                  * Note that if the file block size will change as a result of
 770                  * this write, then this range lock will lock the entire file
 771                  * so that we can re-write the block safely.
 772                  */
 773                 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
 774         }
 775 
 776         if (woff >= limit) {
 777                 zfs_range_unlock(rl);
 778                 ZFS_EXIT(zfsvfs);
 779                 return (SET_ERROR(EFBIG));
 780         }
 781 
 782         if ((woff + n) > limit || woff > (limit - n))
 783                 n = limit - woff;
 784 
 785         /* Will this write extend the file length? */
 786         write_eof = (woff + n > zp->z_size);
 787 
 788         end_size = MAX(zp->z_size, woff + n);
 789 
 790         /*
 791          * Write the file in reasonable size chunks.  Each chunk is written
 792          * in a separate transaction; this keeps the intent log records small
 793          * and allows us to do more fine-grained space accounting.
 794          */
 795         while (n > 0) {
 796                 abuf = NULL;
 797                 woff = uio->uio_loffset;
 798                 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
 799                     zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
 800                         if (abuf != NULL)
 801                                 dmu_return_arcbuf(abuf);
 802                         error = SET_ERROR(EDQUOT);
 803                         break;
 804                 }
 805 
 806                 if (xuio && abuf == NULL) {
 807                         ASSERT(i_iov < iovcnt);
 808                         aiov = &iovp[i_iov];
 809                         abuf = dmu_xuio_arcbuf(xuio, i_iov);
 810                         dmu_xuio_clear(xuio, i_iov);
 811                         DTRACE_PROBE3(zfs_cp_write, int, i_iov,
 812                             iovec_t *, aiov, arc_buf_t *, abuf);
 813                         ASSERT((aiov->iov_base == abuf->b_data) ||
 814                             ((char *)aiov->iov_base - (char *)abuf->b_data +
 815                             aiov->iov_len == arc_buf_size(abuf)));
 816                         i_iov++;
 817                 } else if (abuf == NULL && n >= max_blksz &&
 818                     woff >= zp->z_size &&
 819                     P2PHASE(woff, max_blksz) == 0 &&
 820                     zp->z_blksz == max_blksz) {
 821                         /*
 822                          * This write covers a full block.  "Borrow" a buffer
 823                          * from the dmu so that we can fill it before we enter
 824                          * a transaction.  This avoids the possibility of
 825                          * holding up the transaction if the data copy hangs
 826                          * up on a pagefault (e.g., from an NFS server mapping).
 827                          */
 828                         size_t cbytes;
 829 
 830                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
 831                             max_blksz);
 832                         ASSERT(abuf != NULL);
 833                         ASSERT(arc_buf_size(abuf) == max_blksz);
 834                         if (error = uiocopy(abuf->b_data, max_blksz,
 835                             UIO_WRITE, uio, &cbytes)) {
 836                                 dmu_return_arcbuf(abuf);
 837                                 break;
 838                         }
 839                         ASSERT(cbytes == max_blksz);
 840                 }
 841 
 842                 /*
 843                  * Start a transaction.
 844                  */
 845                 tx = dmu_tx_create(zfsvfs->z_os);
 846                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 847                 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
 848                 zfs_sa_upgrade_txholds(tx, zp);
 849                 error = dmu_tx_assign(tx, TXG_WAIT);
 850                 if (error) {
 851                         dmu_tx_abort(tx);
 852                         if (abuf != NULL)
 853                                 dmu_return_arcbuf(abuf);
 854                         break;
 855                 }
 856 
 857                 /*
 858                  * If zfs_range_lock() over-locked we grow the blocksize
 859                  * and then reduce the lock range.  This will only happen
 860                  * on the first iteration since zfs_range_reduce() will
 861                  * shrink down r_len to the appropriate size.
 862                  */
 863                 if (rl->r_len == UINT64_MAX) {
 864                         uint64_t new_blksz;
 865 
 866                         if (zp->z_blksz > max_blksz) {
 867                                 /*
 868                                  * File's blocksize is already larger than the
 869                                  * "recordsize" property.  Only let it grow to
 870                                  * the next power of 2.
 871                                  */
 872                                 ASSERT(!ISP2(zp->z_blksz));
 873                                 new_blksz = MIN(end_size,
 874                                     1 << highbit64(zp->z_blksz));
 875                         } else {
 876                                 new_blksz = MIN(end_size, max_blksz);
 877                         }
 878                         zfs_grow_blocksize(zp, new_blksz, tx);
 879                         zfs_range_reduce(rl, woff, n);
 880                 }
 881 
 882                 /*
 883                  * XXX - should we really limit each write to z_max_blksz?
 884                  * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
 885                  */
 886                 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
 887 
 888                 if (abuf == NULL) {
 889                         tx_bytes = uio->uio_resid;
 890                         error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
 891                             uio, nbytes, tx);
 892                         tx_bytes -= uio->uio_resid;
 893                 } else {
 894                         tx_bytes = nbytes;
 895                         ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
 896                         /*
 897                          * If this is not a full block write, but we are
 898                          * extending the file past EOF and this data starts
 899                          * block-aligned, use assign_arcbuf().  Otherwise,
 900                          * write via dmu_write().
 901                          */
 902                         if (tx_bytes < max_blksz && (!write_eof ||
 903                             aiov->iov_base != abuf->b_data)) {
 904                                 ASSERT(xuio);
 905                                 dmu_write(zfsvfs->z_os, zp->z_id, woff,
 906                                     aiov->iov_len, aiov->iov_base, tx);
 907                                 dmu_return_arcbuf(abuf);
 908                                 xuio_stat_wbuf_copied();
 909                         } else {
 910                                 ASSERT(xuio || tx_bytes == max_blksz);
 911                                 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
 912                                     woff, abuf, tx);
 913                         }
 914                         ASSERT(tx_bytes <= uio->uio_resid);
 915                         uioskip(uio, tx_bytes);
 916                 }
 917                 if (tx_bytes && vn_has_cached_data(vp)) {
 918                         update_pages(vp, woff,
 919                             tx_bytes, zfsvfs->z_os, zp->z_id);
 920                 }
 921 
 922                 /*
 923                  * If we made no progress, we're done.  If we made even
 924                  * partial progress, update the znode and ZIL accordingly.
 925                  */
 926                 if (tx_bytes == 0) {
 927                         (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
 928                             (void *)&zp->z_size, sizeof (uint64_t), tx);
 929                         dmu_tx_commit(tx);
 930                         ASSERT(error != 0);
 931                         break;
 932                 }
 933 
 934                 /*
 935                  * Clear Set-UID/Set-GID bits on successful write if not
 936                  * privileged and at least one of the excute bits is set.
 937                  *
 938                  * It would be nice to to this after all writes have
 939                  * been done, but that would still expose the ISUID/ISGID
 940                  * to another app after the partial write is committed.
 941                  *
 942                  * Note: we don't call zfs_fuid_map_id() here because
 943                  * user 0 is not an ephemeral uid.
 944                  */
 945                 mutex_enter(&zp->z_acl_lock);
 946                 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
 947                     (S_IXUSR >> 6))) != 0 &&
 948                     (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
 949                     secpolicy_vnode_setid_retain(cr,
 950                     (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
 951                         uint64_t newmode;
 952                         zp->z_mode &= ~(S_ISUID | S_ISGID);
 953                         newmode = zp->z_mode;
 954                         (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
 955                             (void *)&newmode, sizeof (uint64_t), tx);
 956                 }
 957                 mutex_exit(&zp->z_acl_lock);
 958 
 959                 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
 960                     B_TRUE);
 961 
 962                 /*
 963                  * Update the file size (zp_size) if it has changed;
 964                  * account for possible concurrent updates.
 965                  */
 966                 while ((end_size = zp->z_size) < uio->uio_loffset) {
 967                         (void) atomic_cas_64(&zp->z_size, end_size,
 968                             uio->uio_loffset);
 969                         ASSERT(error == 0);
 970                 }
 971                 /*
 972                  * If we are replaying and eof is non zero then force
 973                  * the file size to the specified eof. Note, there's no
 974                  * concurrency during replay.
 975                  */
 976                 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
 977                         zp->z_size = zfsvfs->z_replay_eof;
 978 
 979                 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
 980 
 981                 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
 982                 dmu_tx_commit(tx);
 983 
 984                 if (error != 0)
 985                         break;
 986                 ASSERT(tx_bytes == nbytes);
 987                 n -= nbytes;
 988 
 989                 if (!xuio && n > 0)
 990                         uio_prefaultpages(MIN(n, max_blksz), uio);
 991         }
 992 
 993         zfs_range_unlock(rl);
 994 
 995         /*
 996          * If we're in replay mode, or we made no progress, return error.
 997          * Otherwise, it's at least a partial write, so it's successful.
 998          */
 999         if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1000                 ZFS_EXIT(zfsvfs);
1001                 return (error);
1002         }
1003 
1004         if (ioflag & (FSYNC | FDSYNC) ||
1005             zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1006                 zil_commit(zilog, zp->z_id);
1007 
1008         ZFS_EXIT(zfsvfs);
1009         return (0);
1010 }
1011 
1012 void
1013 zfs_get_done(zgd_t *zgd, int error)
1014 {
1015         znode_t *zp = zgd->zgd_private;
1016         objset_t *os = zp->z_zfsvfs->z_os;
1017 
1018         if (zgd->zgd_db)
1019                 dmu_buf_rele(zgd->zgd_db, zgd);
1020 
1021         zfs_range_unlock(zgd->zgd_rl);
1022 
1023         /*
1024          * Release the vnode asynchronously as we currently have the
1025          * txg stopped from syncing.
1026          */
1027         VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1028 
1029         if (error == 0 && zgd->zgd_bp)
1030                 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1031 
1032         kmem_free(zgd, sizeof (zgd_t));
1033 }
1034 
1035 #ifdef DEBUG
1036 static int zil_fault_io = 0;
1037 #endif
1038 
1039 /*
1040  * Get data to generate a TX_WRITE intent log record.
1041  */
1042 int
1043 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1044 {
1045         zfsvfs_t *zfsvfs = arg;
1046         objset_t *os = zfsvfs->z_os;
1047         znode_t *zp;
1048         uint64_t object = lr->lr_foid;
1049         uint64_t offset = lr->lr_offset;
1050         uint64_t size = lr->lr_length;
1051         blkptr_t *bp = &lr->lr_blkptr;
1052         dmu_buf_t *db;
1053         zgd_t *zgd;
1054         int error = 0;
1055 
1056         ASSERT(zio != NULL);
1057         ASSERT(size != 0);
1058 
1059         /*
1060          * Nothing to do if the file has been removed
1061          */
1062         if (zfs_zget(zfsvfs, object, &zp) != 0)
1063                 return (SET_ERROR(ENOENT));
1064         if (zp->z_unlinked) {
1065                 /*
1066                  * Release the vnode asynchronously as we currently have the
1067                  * txg stopped from syncing.
1068                  */
1069                 VN_RELE_ASYNC(ZTOV(zp),
1070                     dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1071                 return (SET_ERROR(ENOENT));
1072         }
1073 
1074         zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1075         zgd->zgd_zilog = zfsvfs->z_log;
1076         zgd->zgd_private = zp;
1077 
1078         /*
1079          * Write records come in two flavors: immediate and indirect.
1080          * For small writes it's cheaper to store the data with the
1081          * log record (immediate); for large writes it's cheaper to
1082          * sync the data and get a pointer to it (indirect) so that
1083          * we don't have to write the data twice.
1084          */
1085         if (buf != NULL) { /* immediate write */
1086                 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1087                 /* test for truncation needs to be done while range locked */
1088                 if (offset >= zp->z_size) {
1089                         error = SET_ERROR(ENOENT);
1090                 } else {
1091                         error = dmu_read(os, object, offset, size, buf,
1092                             DMU_READ_NO_PREFETCH);
1093                 }
1094                 ASSERT(error == 0 || error == ENOENT);
1095         } else { /* indirect write */
1096                 /*
1097                  * Have to lock the whole block to ensure when it's
1098                  * written out and it's checksum is being calculated
1099                  * that no one can change the data. We need to re-check
1100                  * blocksize after we get the lock in case it's changed!
1101                  */
1102                 for (;;) {
1103                         uint64_t blkoff;
1104                         size = zp->z_blksz;
1105                         blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1106                         offset -= blkoff;
1107                         zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1108                             RL_READER);
1109                         if (zp->z_blksz == size)
1110                                 break;
1111                         offset += blkoff;
1112                         zfs_range_unlock(zgd->zgd_rl);
1113                 }
1114                 /* test for truncation needs to be done while range locked */
1115                 if (lr->lr_offset >= zp->z_size)
1116                         error = SET_ERROR(ENOENT);
1117 #ifdef DEBUG
1118                 if (zil_fault_io) {
1119                         error = SET_ERROR(EIO);
1120                         zil_fault_io = 0;
1121                 }
1122 #endif
1123                 if (error == 0)
1124                         error = dmu_buf_hold(os, object, offset, zgd, &db,
1125                             DMU_READ_NO_PREFETCH);
1126 
1127                 if (error == 0) {
1128                         blkptr_t *obp = dmu_buf_get_blkptr(db);
1129                         if (obp) {
1130                                 ASSERT(BP_IS_HOLE(bp));
1131                                 *bp = *obp;
1132                         }
1133 
1134                         zgd->zgd_db = db;
1135                         zgd->zgd_bp = bp;
1136 
1137                         ASSERT(db->db_offset == offset);
1138                         ASSERT(db->db_size == size);
1139 
1140                         error = dmu_sync(zio, lr->lr_common.lrc_txg,
1141                             zfs_get_done, zgd);
1142                         ASSERT(error || lr->lr_length <= zp->z_blksz);
1143 
1144                         /*
1145                          * On success, we need to wait for the write I/O
1146                          * initiated by dmu_sync() to complete before we can
1147                          * release this dbuf.  We will finish everything up
1148                          * in the zfs_get_done() callback.
1149                          */
1150                         if (error == 0)
1151                                 return (0);
1152 
1153                         if (error == EALREADY) {
1154                                 lr->lr_common.lrc_txtype = TX_WRITE2;
1155                                 error = 0;
1156                         }
1157                 }
1158         }
1159 
1160         zfs_get_done(zgd, error);
1161 
1162         return (error);
1163 }
1164 
1165 /*ARGSUSED*/
1166 static int
1167 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1168     caller_context_t *ct)
1169 {
1170         znode_t *zp = VTOZ(vp);
1171         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1172         int error;
1173 
1174         ZFS_ENTER(zfsvfs);
1175         ZFS_VERIFY_ZP(zp);
1176 
1177         if (flag & V_ACE_MASK)
1178                 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1179         else
1180                 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1181 
1182         ZFS_EXIT(zfsvfs);
1183         return (error);
1184 }
1185 
1186 /*
1187  * If vnode is for a device return a specfs vnode instead.
1188  */
1189 static int
1190 specvp_check(vnode_t **vpp, cred_t *cr)
1191 {
1192         int error = 0;
1193 
1194         if (IS_DEVVP(*vpp)) {
1195                 struct vnode *svp;
1196 
1197                 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1198                 VN_RELE(*vpp);
1199                 if (svp == NULL)
1200                         error = SET_ERROR(ENOSYS);
1201                 *vpp = svp;
1202         }
1203         return (error);
1204 }
1205 
1206 
1207 /*
1208  * Lookup an entry in a directory, or an extended attribute directory.
1209  * If it exists, return a held vnode reference for it.
1210  *
1211  *      IN:     dvp     - vnode of directory to search.
1212  *              nm      - name of entry to lookup.
1213  *              pnp     - full pathname to lookup [UNUSED].
1214  *              flags   - LOOKUP_XATTR set if looking for an attribute.
1215  *              rdir    - root directory vnode [UNUSED].
1216  *              cr      - credentials of caller.
1217  *              ct      - caller context
1218  *              direntflags - directory lookup flags
1219  *              realpnp - returned pathname.
1220  *
1221  *      OUT:    vpp     - vnode of located entry, NULL if not found.
1222  *
1223  *      RETURN: 0 on success, error code on failure.
1224  *
1225  * Timestamps:
1226  *      NA
1227  */
1228 /* ARGSUSED */
1229 static int
1230 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1231     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1232     int *direntflags, pathname_t *realpnp)
1233 {
1234         znode_t *zdp = VTOZ(dvp);
1235         zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1236         int     error = 0;
1237 
1238         /* fast path */
1239         if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1240 
1241                 if (dvp->v_type != VDIR) {
1242                         return (SET_ERROR(ENOTDIR));
1243                 } else if (zdp->z_sa_hdl == NULL) {
1244                         return (SET_ERROR(EIO));
1245                 }
1246 
1247                 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1248                         error = zfs_fastaccesschk_execute(zdp, cr);
1249                         if (!error) {
1250                                 *vpp = dvp;
1251                                 VN_HOLD(*vpp);
1252                                 return (0);
1253                         }
1254                         return (error);
1255                 } else {
1256                         vnode_t *tvp = dnlc_lookup(dvp, nm);
1257 
1258                         if (tvp) {
1259                                 error = zfs_fastaccesschk_execute(zdp, cr);
1260                                 if (error) {
1261                                         VN_RELE(tvp);
1262                                         return (error);
1263                                 }
1264                                 if (tvp == DNLC_NO_VNODE) {
1265                                         VN_RELE(tvp);
1266                                         return (SET_ERROR(ENOENT));
1267                                 } else {
1268                                         *vpp = tvp;
1269                                         return (specvp_check(vpp, cr));
1270                                 }
1271                         }
1272                 }
1273         }
1274 
1275         DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1276 
1277         ZFS_ENTER(zfsvfs);
1278         ZFS_VERIFY_ZP(zdp);
1279 
1280         *vpp = NULL;
1281 
1282         if (flags & LOOKUP_XATTR) {
1283                 /*
1284                  * If the xattr property is off, refuse the lookup request.
1285                  */
1286                 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1287                         ZFS_EXIT(zfsvfs);
1288                         return (SET_ERROR(EINVAL));
1289                 }
1290 
1291                 /*
1292                  * We don't allow recursive attributes..
1293                  * Maybe someday we will.
1294                  */
1295                 if (zdp->z_pflags & ZFS_XATTR) {
1296                         ZFS_EXIT(zfsvfs);
1297                         return (SET_ERROR(EINVAL));
1298                 }
1299 
1300                 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1301                         ZFS_EXIT(zfsvfs);
1302                         return (error);
1303                 }
1304 
1305                 /*
1306                  * Do we have permission to get into attribute directory?
1307                  */
1308 
1309                 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1310                     B_FALSE, cr)) {
1311                         VN_RELE(*vpp);
1312                         *vpp = NULL;
1313                 }
1314 
1315                 ZFS_EXIT(zfsvfs);
1316                 return (error);
1317         }
1318 
1319         if (dvp->v_type != VDIR) {
1320                 ZFS_EXIT(zfsvfs);
1321                 return (SET_ERROR(ENOTDIR));
1322         }
1323 
1324         /*
1325          * Check accessibility of directory.
1326          */
1327 
1328         if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1329                 ZFS_EXIT(zfsvfs);
1330                 return (error);
1331         }
1332 
1333         if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1334             NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1335                 ZFS_EXIT(zfsvfs);
1336                 return (SET_ERROR(EILSEQ));
1337         }
1338 
1339         error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1340         if (error == 0)
1341                 error = specvp_check(vpp, cr);
1342 
1343         ZFS_EXIT(zfsvfs);
1344         return (error);
1345 }
1346 
1347 /*
1348  * Attempt to create a new entry in a directory.  If the entry
1349  * already exists, truncate the file if permissible, else return
1350  * an error.  Return the vp of the created or trunc'd file.
1351  *
1352  *      IN:     dvp     - vnode of directory to put new file entry in.
1353  *              name    - name of new file entry.
1354  *              vap     - attributes of new file.
1355  *              excl    - flag indicating exclusive or non-exclusive mode.
1356  *              mode    - mode to open file with.
1357  *              cr      - credentials of caller.
1358  *              flag    - large file flag [UNUSED].
1359  *              ct      - caller context
1360  *              vsecp   - ACL to be set
1361  *
1362  *      OUT:    vpp     - vnode of created or trunc'd entry.
1363  *
1364  *      RETURN: 0 on success, error code on failure.
1365  *
1366  * Timestamps:
1367  *      dvp - ctime|mtime updated if new entry created
1368  *       vp - ctime|mtime always, atime if new
1369  */
1370 
1371 /* ARGSUSED */
1372 static int
1373 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1374     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1375     vsecattr_t *vsecp)
1376 {
1377         znode_t         *zp, *dzp = VTOZ(dvp);
1378         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1379         zilog_t         *zilog;
1380         objset_t        *os;
1381         zfs_dirlock_t   *dl;
1382         dmu_tx_t        *tx;
1383         int             error;
1384         ksid_t          *ksid;
1385         uid_t           uid;
1386         gid_t           gid = crgetgid(cr);
1387         zfs_acl_ids_t   acl_ids;
1388         boolean_t       fuid_dirtied;
1389         boolean_t       have_acl = B_FALSE;
1390         boolean_t       waited = B_FALSE;
1391 
1392         /*
1393          * If we have an ephemeral id, ACL, or XVATTR then
1394          * make sure file system is at proper version
1395          */
1396 
1397         ksid = crgetsid(cr, KSID_OWNER);
1398         if (ksid)
1399                 uid = ksid_getid(ksid);
1400         else
1401                 uid = crgetuid(cr);
1402 
1403         if (zfsvfs->z_use_fuids == B_FALSE &&
1404             (vsecp || (vap->va_mask & AT_XVATTR) ||
1405             IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1406                 return (SET_ERROR(EINVAL));
1407 
1408         ZFS_ENTER(zfsvfs);
1409         ZFS_VERIFY_ZP(dzp);
1410         os = zfsvfs->z_os;
1411         zilog = zfsvfs->z_log;
1412 
1413         if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1414             NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1415                 ZFS_EXIT(zfsvfs);
1416                 return (SET_ERROR(EILSEQ));
1417         }
1418 
1419         if (vap->va_mask & AT_XVATTR) {
1420                 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1421                     crgetuid(cr), cr, vap->va_type)) != 0) {
1422                         ZFS_EXIT(zfsvfs);
1423                         return (error);
1424                 }
1425         }
1426 top:
1427         *vpp = NULL;
1428 
1429         if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1430                 vap->va_mode &= ~VSVTX;
1431 
1432         if (*name == '\0') {
1433                 /*
1434                  * Null component name refers to the directory itself.
1435                  */
1436                 VN_HOLD(dvp);
1437                 zp = dzp;
1438                 dl = NULL;
1439                 error = 0;
1440         } else {
1441                 /* possible VN_HOLD(zp) */
1442                 int zflg = 0;
1443 
1444                 if (flag & FIGNORECASE)
1445                         zflg |= ZCILOOK;
1446 
1447                 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1448                     NULL, NULL);
1449                 if (error) {
1450                         if (have_acl)
1451                                 zfs_acl_ids_free(&acl_ids);
1452                         if (strcmp(name, "..") == 0)
1453                                 error = SET_ERROR(EISDIR);
1454                         ZFS_EXIT(zfsvfs);
1455                         return (error);
1456                 }
1457         }
1458 
1459         if (zp == NULL) {
1460                 uint64_t txtype;
1461 
1462                 /*
1463                  * Create a new file object and update the directory
1464                  * to reference it.
1465                  */
1466                 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1467                         if (have_acl)
1468                                 zfs_acl_ids_free(&acl_ids);
1469                         goto out;
1470                 }
1471 
1472                 /*
1473                  * We only support the creation of regular files in
1474                  * extended attribute directories.
1475                  */
1476 
1477                 if ((dzp->z_pflags & ZFS_XATTR) &&
1478                     (vap->va_type != VREG)) {
1479                         if (have_acl)
1480                                 zfs_acl_ids_free(&acl_ids);
1481                         error = SET_ERROR(EINVAL);
1482                         goto out;
1483                 }
1484 
1485                 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1486                     cr, vsecp, &acl_ids)) != 0)
1487                         goto out;
1488                 have_acl = B_TRUE;
1489 
1490                 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1491                         zfs_acl_ids_free(&acl_ids);
1492                         error = SET_ERROR(EDQUOT);
1493                         goto out;
1494                 }
1495 
1496                 tx = dmu_tx_create(os);
1497 
1498                 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1499                     ZFS_SA_BASE_ATTR_SIZE);
1500 
1501                 fuid_dirtied = zfsvfs->z_fuid_dirty;
1502                 if (fuid_dirtied)
1503                         zfs_fuid_txhold(zfsvfs, tx);
1504                 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1505                 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1506                 if (!zfsvfs->z_use_sa &&
1507                     acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1508                         dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1509                             0, acl_ids.z_aclp->z_acl_bytes);
1510                 }
1511                 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1512                 if (error) {
1513                         zfs_dirent_unlock(dl);
1514                         if (error == ERESTART) {
1515                                 waited = B_TRUE;
1516                                 dmu_tx_wait(tx);
1517                                 dmu_tx_abort(tx);
1518                                 goto top;
1519                         }
1520                         zfs_acl_ids_free(&acl_ids);
1521                         dmu_tx_abort(tx);
1522                         ZFS_EXIT(zfsvfs);
1523                         return (error);
1524                 }
1525                 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1526 
1527                 if (fuid_dirtied)
1528                         zfs_fuid_sync(zfsvfs, tx);
1529 
1530                 (void) zfs_link_create(dl, zp, tx, ZNEW);
1531                 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1532                 if (flag & FIGNORECASE)
1533                         txtype |= TX_CI;
1534                 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1535                     vsecp, acl_ids.z_fuidp, vap);
1536                 zfs_acl_ids_free(&acl_ids);
1537                 dmu_tx_commit(tx);
1538         } else {
1539                 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1540 
1541                 if (have_acl)
1542                         zfs_acl_ids_free(&acl_ids);
1543                 have_acl = B_FALSE;
1544 
1545                 /*
1546                  * A directory entry already exists for this name.
1547                  */
1548                 /*
1549                  * Can't truncate an existing file if in exclusive mode.
1550                  */
1551                 if (excl == EXCL) {
1552                         error = SET_ERROR(EEXIST);
1553                         goto out;
1554                 }
1555                 /*
1556                  * Can't open a directory for writing.
1557                  */
1558                 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1559                         error = SET_ERROR(EISDIR);
1560                         goto out;
1561                 }
1562                 /*
1563                  * Verify requested access to file.
1564                  */
1565                 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1566                         goto out;
1567                 }
1568 
1569                 mutex_enter(&dzp->z_lock);
1570                 dzp->z_seq++;
1571                 mutex_exit(&dzp->z_lock);
1572 
1573                 /*
1574                  * Truncate regular files if requested.
1575                  */
1576                 if ((ZTOV(zp)->v_type == VREG) &&
1577                     (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1578                         /* we can't hold any locks when calling zfs_freesp() */
1579                         zfs_dirent_unlock(dl);
1580                         dl = NULL;
1581                         error = zfs_freesp(zp, 0, 0, mode, TRUE);
1582                         if (error == 0) {
1583                                 vnevent_create(ZTOV(zp), ct);
1584                         }
1585                 }
1586         }
1587 out:
1588 
1589         if (dl)
1590                 zfs_dirent_unlock(dl);
1591 
1592         if (error) {
1593                 if (zp)
1594                         VN_RELE(ZTOV(zp));
1595         } else {
1596                 *vpp = ZTOV(zp);
1597                 error = specvp_check(vpp, cr);
1598         }
1599 
1600         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1601                 zil_commit(zilog, 0);
1602 
1603         ZFS_EXIT(zfsvfs);
1604         return (error);
1605 }
1606 
1607 /*
1608  * Remove an entry from a directory.
1609  *
1610  *      IN:     dvp     - vnode of directory to remove entry from.
1611  *              name    - name of entry to remove.
1612  *              cr      - credentials of caller.
1613  *              ct      - caller context
1614  *              flags   - case flags
1615  *
1616  *      RETURN: 0 on success, error code on failure.
1617  *
1618  * Timestamps:
1619  *      dvp - ctime|mtime
1620  *       vp - ctime (if nlink > 0)
1621  */
1622 
1623 uint64_t null_xattr = 0;
1624 
1625 /*ARGSUSED*/
1626 static int
1627 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1628     int flags)
1629 {
1630         znode_t         *zp, *dzp = VTOZ(dvp);
1631         znode_t         *xzp;
1632         vnode_t         *vp;
1633         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1634         zilog_t         *zilog;
1635         uint64_t        acl_obj, xattr_obj;
1636         uint64_t        xattr_obj_unlinked = 0;
1637         uint64_t        obj = 0;
1638         zfs_dirlock_t   *dl;
1639         dmu_tx_t        *tx;
1640         boolean_t       may_delete_now, delete_now = FALSE;
1641         boolean_t       unlinked, toobig = FALSE;
1642         uint64_t        txtype;
1643         pathname_t      *realnmp = NULL;
1644         pathname_t      realnm;
1645         int             error;
1646         int             zflg = ZEXISTS;
1647         boolean_t       waited = B_FALSE;
1648 
1649         ZFS_ENTER(zfsvfs);
1650         ZFS_VERIFY_ZP(dzp);
1651         zilog = zfsvfs->z_log;
1652 
1653         if (flags & FIGNORECASE) {
1654                 zflg |= ZCILOOK;
1655                 pn_alloc(&realnm);
1656                 realnmp = &realnm;
1657         }
1658 
1659 top:
1660         xattr_obj = 0;
1661         xzp = NULL;
1662         /*
1663          * Attempt to lock directory; fail if entry doesn't exist.
1664          */
1665         if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1666             NULL, realnmp)) {
1667                 if (realnmp)
1668                         pn_free(realnmp);
1669                 ZFS_EXIT(zfsvfs);
1670                 return (error);
1671         }
1672 
1673         vp = ZTOV(zp);
1674 
1675         if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1676                 goto out;
1677         }
1678 
1679         /*
1680          * Need to use rmdir for removing directories.
1681          */
1682         if (vp->v_type == VDIR) {
1683                 error = SET_ERROR(EPERM);
1684                 goto out;
1685         }
1686 
1687         vnevent_remove(vp, dvp, name, ct);
1688 
1689         if (realnmp)
1690                 dnlc_remove(dvp, realnmp->pn_buf);
1691         else
1692                 dnlc_remove(dvp, name);
1693 
1694         mutex_enter(&vp->v_lock);
1695         may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1696         mutex_exit(&vp->v_lock);
1697 
1698         /*
1699          * We may delete the znode now, or we may put it in the unlinked set;
1700          * it depends on whether we're the last link, and on whether there are
1701          * other holds on the vnode.  So we dmu_tx_hold() the right things to
1702          * allow for either case.
1703          */
1704         obj = zp->z_id;
1705         tx = dmu_tx_create(zfsvfs->z_os);
1706         dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1707         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1708         zfs_sa_upgrade_txholds(tx, zp);
1709         zfs_sa_upgrade_txholds(tx, dzp);
1710         if (may_delete_now) {
1711                 toobig =
1712                     zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1713                 /* if the file is too big, only hold_free a token amount */
1714                 dmu_tx_hold_free(tx, zp->z_id, 0,
1715                     (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1716         }
1717 
1718         /* are there any extended attributes? */
1719         error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1720             &xattr_obj, sizeof (xattr_obj));
1721         if (error == 0 && xattr_obj) {
1722                 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1723                 ASSERT0(error);
1724                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1725                 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1726         }
1727 
1728         mutex_enter(&zp->z_lock);
1729         if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1730                 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1731         mutex_exit(&zp->z_lock);
1732 
1733         /* charge as an update -- would be nice not to charge at all */
1734         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1735 
1736         /*
1737          * Mark this transaction as typically resulting in a net free of space
1738          */
1739         dmu_tx_mark_netfree(tx);
1740 
1741         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1742         if (error) {
1743                 zfs_dirent_unlock(dl);
1744                 VN_RELE(vp);
1745                 if (xzp)
1746                         VN_RELE(ZTOV(xzp));
1747                 if (error == ERESTART) {
1748                         waited = B_TRUE;
1749                         dmu_tx_wait(tx);
1750                         dmu_tx_abort(tx);
1751                         goto top;
1752                 }
1753                 if (realnmp)
1754                         pn_free(realnmp);
1755                 dmu_tx_abort(tx);
1756                 ZFS_EXIT(zfsvfs);
1757                 return (error);
1758         }
1759 
1760         /*
1761          * Remove the directory entry.
1762          */
1763         error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1764 
1765         if (error) {
1766                 dmu_tx_commit(tx);
1767                 goto out;
1768         }
1769 
1770         if (unlinked) {
1771                 /*
1772                  * Hold z_lock so that we can make sure that the ACL obj
1773                  * hasn't changed.  Could have been deleted due to
1774                  * zfs_sa_upgrade().
1775                  */
1776                 mutex_enter(&zp->z_lock);
1777                 mutex_enter(&vp->v_lock);
1778                 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1779                     &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1780                 delete_now = may_delete_now && !toobig &&
1781                     vp->v_count == 1 && !vn_has_cached_data(vp) &&
1782                     xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1783                     acl_obj;
1784                 mutex_exit(&vp->v_lock);
1785         }
1786 
1787         if (delete_now) {
1788                 if (xattr_obj_unlinked) {
1789                         ASSERT3U(xzp->z_links, ==, 2);
1790                         mutex_enter(&xzp->z_lock);
1791                         xzp->z_unlinked = 1;
1792                         xzp->z_links = 0;
1793                         error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1794                             &xzp->z_links, sizeof (xzp->z_links), tx);
1795                         ASSERT3U(error,  ==,  0);
1796                         mutex_exit(&xzp->z_lock);
1797                         zfs_unlinked_add(xzp, tx);
1798 
1799                         if (zp->z_is_sa)
1800                                 error = sa_remove(zp->z_sa_hdl,
1801                                     SA_ZPL_XATTR(zfsvfs), tx);
1802                         else
1803                                 error = sa_update(zp->z_sa_hdl,
1804                                     SA_ZPL_XATTR(zfsvfs), &null_xattr,
1805                                     sizeof (uint64_t), tx);
1806                         ASSERT0(error);
1807                 }
1808                 mutex_enter(&vp->v_lock);
1809                 vp->v_count--;
1810                 ASSERT0(vp->v_count);
1811                 mutex_exit(&vp->v_lock);
1812                 mutex_exit(&zp->z_lock);
1813                 zfs_znode_delete(zp, tx);
1814         } else if (unlinked) {
1815                 mutex_exit(&zp->z_lock);
1816                 zfs_unlinked_add(zp, tx);
1817         }
1818 
1819         txtype = TX_REMOVE;
1820         if (flags & FIGNORECASE)
1821                 txtype |= TX_CI;
1822         zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1823 
1824         dmu_tx_commit(tx);
1825 out:
1826         if (realnmp)
1827                 pn_free(realnmp);
1828 
1829         zfs_dirent_unlock(dl);
1830 
1831         if (!delete_now)
1832                 VN_RELE(vp);
1833         if (xzp)
1834                 VN_RELE(ZTOV(xzp));
1835 
1836         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1837                 zil_commit(zilog, 0);
1838 
1839         ZFS_EXIT(zfsvfs);
1840         return (error);
1841 }
1842 
1843 /*
1844  * Create a new directory and insert it into dvp using the name
1845  * provided.  Return a pointer to the inserted directory.
1846  *
1847  *      IN:     dvp     - vnode of directory to add subdir to.
1848  *              dirname - name of new directory.
1849  *              vap     - attributes of new directory.
1850  *              cr      - credentials of caller.
1851  *              ct      - caller context
1852  *              flags   - case flags
1853  *              vsecp   - ACL to be set
1854  *
1855  *      OUT:    vpp     - vnode of created directory.
1856  *
1857  *      RETURN: 0 on success, error code on failure.
1858  *
1859  * Timestamps:
1860  *      dvp - ctime|mtime updated
1861  *       vp - ctime|mtime|atime updated
1862  */
1863 /*ARGSUSED*/
1864 static int
1865 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1866     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1867 {
1868         znode_t         *zp, *dzp = VTOZ(dvp);
1869         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
1870         zilog_t         *zilog;
1871         zfs_dirlock_t   *dl;
1872         uint64_t        txtype;
1873         dmu_tx_t        *tx;
1874         int             error;
1875         int             zf = ZNEW;
1876         ksid_t          *ksid;
1877         uid_t           uid;
1878         gid_t           gid = crgetgid(cr);
1879         zfs_acl_ids_t   acl_ids;
1880         boolean_t       fuid_dirtied;
1881         boolean_t       waited = B_FALSE;
1882 
1883         ASSERT(vap->va_type == VDIR);
1884 
1885         /*
1886          * If we have an ephemeral id, ACL, or XVATTR then
1887          * make sure file system is at proper version
1888          */
1889 
1890         ksid = crgetsid(cr, KSID_OWNER);
1891         if (ksid)
1892                 uid = ksid_getid(ksid);
1893         else
1894                 uid = crgetuid(cr);
1895         if (zfsvfs->z_use_fuids == B_FALSE &&
1896             (vsecp || (vap->va_mask & AT_XVATTR) ||
1897             IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1898                 return (SET_ERROR(EINVAL));
1899 
1900         ZFS_ENTER(zfsvfs);
1901         ZFS_VERIFY_ZP(dzp);
1902         zilog = zfsvfs->z_log;
1903 
1904         if (dzp->z_pflags & ZFS_XATTR) {
1905                 ZFS_EXIT(zfsvfs);
1906                 return (SET_ERROR(EINVAL));
1907         }
1908 
1909         if (zfsvfs->z_utf8 && u8_validate(dirname,
1910             strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1911                 ZFS_EXIT(zfsvfs);
1912                 return (SET_ERROR(EILSEQ));
1913         }
1914         if (flags & FIGNORECASE)
1915                 zf |= ZCILOOK;
1916 
1917         if (vap->va_mask & AT_XVATTR) {
1918                 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1919                     crgetuid(cr), cr, vap->va_type)) != 0) {
1920                         ZFS_EXIT(zfsvfs);
1921                         return (error);
1922                 }
1923         }
1924 
1925         if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1926             vsecp, &acl_ids)) != 0) {
1927                 ZFS_EXIT(zfsvfs);
1928                 return (error);
1929         }
1930         /*
1931          * First make sure the new directory doesn't exist.
1932          *
1933          * Existence is checked first to make sure we don't return
1934          * EACCES instead of EEXIST which can cause some applications
1935          * to fail.
1936          */
1937 top:
1938         *vpp = NULL;
1939 
1940         if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1941             NULL, NULL)) {
1942                 zfs_acl_ids_free(&acl_ids);
1943                 ZFS_EXIT(zfsvfs);
1944                 return (error);
1945         }
1946 
1947         if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1948                 zfs_acl_ids_free(&acl_ids);
1949                 zfs_dirent_unlock(dl);
1950                 ZFS_EXIT(zfsvfs);
1951                 return (error);
1952         }
1953 
1954         if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1955                 zfs_acl_ids_free(&acl_ids);
1956                 zfs_dirent_unlock(dl);
1957                 ZFS_EXIT(zfsvfs);
1958                 return (SET_ERROR(EDQUOT));
1959         }
1960 
1961         /*
1962          * Add a new entry to the directory.
1963          */
1964         tx = dmu_tx_create(zfsvfs->z_os);
1965         dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1966         dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1967         fuid_dirtied = zfsvfs->z_fuid_dirty;
1968         if (fuid_dirtied)
1969                 zfs_fuid_txhold(zfsvfs, tx);
1970         if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1971                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1972                     acl_ids.z_aclp->z_acl_bytes);
1973         }
1974 
1975         dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1976             ZFS_SA_BASE_ATTR_SIZE);
1977 
1978         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1979         if (error) {
1980                 zfs_dirent_unlock(dl);
1981                 if (error == ERESTART) {
1982                         waited = B_TRUE;
1983                         dmu_tx_wait(tx);
1984                         dmu_tx_abort(tx);
1985                         goto top;
1986                 }
1987                 zfs_acl_ids_free(&acl_ids);
1988                 dmu_tx_abort(tx);
1989                 ZFS_EXIT(zfsvfs);
1990                 return (error);
1991         }
1992 
1993         /*
1994          * Create new node.
1995          */
1996         zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1997 
1998         if (fuid_dirtied)
1999                 zfs_fuid_sync(zfsvfs, tx);
2000 
2001         /*
2002          * Now put new name in parent dir.
2003          */
2004         (void) zfs_link_create(dl, zp, tx, ZNEW);
2005 
2006         *vpp = ZTOV(zp);
2007 
2008         txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2009         if (flags & FIGNORECASE)
2010                 txtype |= TX_CI;
2011         zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2012             acl_ids.z_fuidp, vap);
2013 
2014         zfs_acl_ids_free(&acl_ids);
2015 
2016         dmu_tx_commit(tx);
2017 
2018         zfs_dirent_unlock(dl);
2019 
2020         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2021                 zil_commit(zilog, 0);
2022 
2023         ZFS_EXIT(zfsvfs);
2024         return (0);
2025 }
2026 
2027 /*
2028  * Remove a directory subdir entry.  If the current working
2029  * directory is the same as the subdir to be removed, the
2030  * remove will fail.
2031  *
2032  *      IN:     dvp     - vnode of directory to remove from.
2033  *              name    - name of directory to be removed.
2034  *              cwd     - vnode of current working directory.
2035  *              cr      - credentials of caller.
2036  *              ct      - caller context
2037  *              flags   - case flags
2038  *
2039  *      RETURN: 0 on success, error code on failure.
2040  *
2041  * Timestamps:
2042  *      dvp - ctime|mtime updated
2043  */
2044 /*ARGSUSED*/
2045 static int
2046 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2047     caller_context_t *ct, int flags)
2048 {
2049         znode_t         *dzp = VTOZ(dvp);
2050         znode_t         *zp;
2051         vnode_t         *vp;
2052         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
2053         zilog_t         *zilog;
2054         zfs_dirlock_t   *dl;
2055         dmu_tx_t        *tx;
2056         int             error;
2057         int             zflg = ZEXISTS;
2058         boolean_t       waited = B_FALSE;
2059 
2060         ZFS_ENTER(zfsvfs);
2061         ZFS_VERIFY_ZP(dzp);
2062         zilog = zfsvfs->z_log;
2063 
2064         if (flags & FIGNORECASE)
2065                 zflg |= ZCILOOK;
2066 top:
2067         zp = NULL;
2068 
2069         /*
2070          * Attempt to lock directory; fail if entry doesn't exist.
2071          */
2072         if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2073             NULL, NULL)) {
2074                 ZFS_EXIT(zfsvfs);
2075                 return (error);
2076         }
2077 
2078         vp = ZTOV(zp);
2079 
2080         if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2081                 goto out;
2082         }
2083 
2084         if (vp->v_type != VDIR) {
2085                 error = SET_ERROR(ENOTDIR);
2086                 goto out;
2087         }
2088 
2089         if (vp == cwd) {
2090                 error = SET_ERROR(EINVAL);
2091                 goto out;
2092         }
2093 
2094         vnevent_rmdir(vp, dvp, name, ct);
2095 
2096         /*
2097          * Grab a lock on the directory to make sure that noone is
2098          * trying to add (or lookup) entries while we are removing it.
2099          */
2100         rw_enter(&zp->z_name_lock, RW_WRITER);
2101 
2102         /*
2103          * Grab a lock on the parent pointer to make sure we play well
2104          * with the treewalk and directory rename code.
2105          */
2106         rw_enter(&zp->z_parent_lock, RW_WRITER);
2107 
2108         tx = dmu_tx_create(zfsvfs->z_os);
2109         dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2110         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2111         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2112         zfs_sa_upgrade_txholds(tx, zp);
2113         zfs_sa_upgrade_txholds(tx, dzp);
2114         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2115         if (error) {
2116                 rw_exit(&zp->z_parent_lock);
2117                 rw_exit(&zp->z_name_lock);
2118                 zfs_dirent_unlock(dl);
2119                 VN_RELE(vp);
2120                 if (error == ERESTART) {
2121                         waited = B_TRUE;
2122                         dmu_tx_wait(tx);
2123                         dmu_tx_abort(tx);
2124                         goto top;
2125                 }
2126                 dmu_tx_abort(tx);
2127                 ZFS_EXIT(zfsvfs);
2128                 return (error);
2129         }
2130 
2131         error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2132 
2133         if (error == 0) {
2134                 uint64_t txtype = TX_RMDIR;
2135                 if (flags & FIGNORECASE)
2136                         txtype |= TX_CI;
2137                 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2138         }
2139 
2140         dmu_tx_commit(tx);
2141 
2142         rw_exit(&zp->z_parent_lock);
2143         rw_exit(&zp->z_name_lock);
2144 out:
2145         zfs_dirent_unlock(dl);
2146 
2147         VN_RELE(vp);
2148 
2149         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2150                 zil_commit(zilog, 0);
2151 
2152         ZFS_EXIT(zfsvfs);
2153         return (error);
2154 }
2155 
2156 /*
2157  * Read as many directory entries as will fit into the provided
2158  * buffer from the given directory cursor position (specified in
2159  * the uio structure).
2160  *
2161  *      IN:     vp      - vnode of directory to read.
2162  *              uio     - structure supplying read location, range info,
2163  *                        and return buffer.
2164  *              cr      - credentials of caller.
2165  *              ct      - caller context
2166  *              flags   - case flags
2167  *
2168  *      OUT:    uio     - updated offset and range, buffer filled.
2169  *              eofp    - set to true if end-of-file detected.
2170  *
2171  *      RETURN: 0 on success, error code on failure.
2172  *
2173  * Timestamps:
2174  *      vp - atime updated
2175  *
2176  * Note that the low 4 bits of the cookie returned by zap is always zero.
2177  * This allows us to use the low range for "special" directory entries:
2178  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2179  * we use the offset 2 for the '.zfs' directory.
2180  */
2181 /* ARGSUSED */
2182 static int
2183 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2184     caller_context_t *ct, int flags)
2185 {
2186         znode_t         *zp = VTOZ(vp);
2187         iovec_t         *iovp;
2188         edirent_t       *eodp;
2189         dirent64_t      *odp;
2190         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
2191         objset_t        *os;
2192         caddr_t         outbuf;
2193         size_t          bufsize;
2194         zap_cursor_t    zc;
2195         zap_attribute_t zap;
2196         uint_t          bytes_wanted;
2197         uint64_t        offset; /* must be unsigned; checks for < 1 */
2198         uint64_t        parent;
2199         int             local_eof;
2200         int             outcount;
2201         int             error;
2202         uint8_t         prefetch;
2203         boolean_t       check_sysattrs;
2204 
2205         ZFS_ENTER(zfsvfs);
2206         ZFS_VERIFY_ZP(zp);
2207 
2208         if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2209             &parent, sizeof (parent))) != 0) {
2210                 ZFS_EXIT(zfsvfs);
2211                 return (error);
2212         }
2213 
2214         /*
2215          * If we are not given an eof variable,
2216          * use a local one.
2217          */
2218         if (eofp == NULL)
2219                 eofp = &local_eof;
2220 
2221         /*
2222          * Check for valid iov_len.
2223          */
2224         if (uio->uio_iov->iov_len <= 0) {
2225                 ZFS_EXIT(zfsvfs);
2226                 return (SET_ERROR(EINVAL));
2227         }
2228 
2229         /*
2230          * Quit if directory has been removed (posix)
2231          */
2232         if ((*eofp = zp->z_unlinked) != 0) {
2233                 ZFS_EXIT(zfsvfs);
2234                 return (0);
2235         }
2236 
2237         error = 0;
2238         os = zfsvfs->z_os;
2239         offset = uio->uio_loffset;
2240         prefetch = zp->z_zn_prefetch;
2241 
2242         /*
2243          * Initialize the iterator cursor.
2244          */
2245         if (offset <= 3) {
2246                 /*
2247                  * Start iteration from the beginning of the directory.
2248                  */
2249                 zap_cursor_init(&zc, os, zp->z_id);
2250         } else {
2251                 /*
2252                  * The offset is a serialized cursor.
2253                  */
2254                 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2255         }
2256 
2257         /*
2258          * Get space to change directory entries into fs independent format.
2259          */
2260         iovp = uio->uio_iov;
2261         bytes_wanted = iovp->iov_len;
2262         if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2263                 bufsize = bytes_wanted;
2264                 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2265                 odp = (struct dirent64 *)outbuf;
2266         } else {
2267                 bufsize = bytes_wanted;
2268                 outbuf = NULL;
2269                 odp = (struct dirent64 *)iovp->iov_base;
2270         }
2271         eodp = (struct edirent *)odp;
2272 
2273         /*
2274          * If this VFS supports the system attribute view interface; and
2275          * we're looking at an extended attribute directory; and we care
2276          * about normalization conflicts on this vfs; then we must check
2277          * for normalization conflicts with the sysattr name space.
2278          */
2279         check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2280             (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2281             (flags & V_RDDIR_ENTFLAGS);
2282 
2283         /*
2284          * Transform to file-system independent format
2285          */
2286         outcount = 0;
2287         while (outcount < bytes_wanted) {
2288                 ino64_t objnum;
2289                 ushort_t reclen;
2290                 off64_t *next = NULL;
2291 
2292                 /*
2293                  * Special case `.', `..', and `.zfs'.
2294                  */
2295                 if (offset == 0) {
2296                         (void) strcpy(zap.za_name, ".");
2297                         zap.za_normalization_conflict = 0;
2298                         objnum = zp->z_id;
2299                 } else if (offset == 1) {
2300                         (void) strcpy(zap.za_name, "..");
2301                         zap.za_normalization_conflict = 0;
2302                         objnum = parent;
2303                 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2304                         (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2305                         zap.za_normalization_conflict = 0;
2306                         objnum = ZFSCTL_INO_ROOT;
2307                 } else {
2308                         /*
2309                          * Grab next entry.
2310                          */
2311                         if (error = zap_cursor_retrieve(&zc, &zap)) {
2312                                 if ((*eofp = (error == ENOENT)) != 0)
2313                                         break;
2314                                 else
2315                                         goto update;
2316                         }
2317 
2318                         if (zap.za_integer_length != 8 ||
2319                             zap.za_num_integers != 1) {
2320                                 cmn_err(CE_WARN, "zap_readdir: bad directory "
2321                                     "entry, obj = %lld, offset = %lld\n",
2322                                     (u_longlong_t)zp->z_id,
2323                                     (u_longlong_t)offset);
2324                                 error = SET_ERROR(ENXIO);
2325                                 goto update;
2326                         }
2327 
2328                         objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2329                         /*
2330                          * MacOS X can extract the object type here such as:
2331                          * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2332                          */
2333 
2334                         if (check_sysattrs && !zap.za_normalization_conflict) {
2335                                 zap.za_normalization_conflict =
2336                                     xattr_sysattr_casechk(zap.za_name);
2337                         }
2338                 }
2339 
2340                 if (flags & V_RDDIR_ACCFILTER) {
2341                         /*
2342                          * If we have no access at all, don't include
2343                          * this entry in the returned information
2344                          */
2345                         znode_t *ezp;
2346                         if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2347                                 goto skip_entry;
2348                         if (!zfs_has_access(ezp, cr)) {
2349                                 VN_RELE(ZTOV(ezp));
2350                                 goto skip_entry;
2351                         }
2352                         VN_RELE(ZTOV(ezp));
2353                 }
2354 
2355                 if (flags & V_RDDIR_ENTFLAGS)
2356                         reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2357                 else
2358                         reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2359 
2360                 /*
2361                  * Will this entry fit in the buffer?
2362                  */
2363                 if (outcount + reclen > bufsize) {
2364                         /*
2365                          * Did we manage to fit anything in the buffer?
2366                          */
2367                         if (!outcount) {
2368                                 error = SET_ERROR(EINVAL);
2369                                 goto update;
2370                         }
2371                         break;
2372                 }
2373                 if (flags & V_RDDIR_ENTFLAGS) {
2374                         /*
2375                          * Add extended flag entry:
2376                          */
2377                         eodp->ed_ino = objnum;
2378                         eodp->ed_reclen = reclen;
2379                         /* NOTE: ed_off is the offset for the *next* entry */
2380                         next = &(eodp->ed_off);
2381                         eodp->ed_eflags = zap.za_normalization_conflict ?
2382                             ED_CASE_CONFLICT : 0;
2383                         (void) strncpy(eodp->ed_name, zap.za_name,
2384                             EDIRENT_NAMELEN(reclen));
2385                         eodp = (edirent_t *)((intptr_t)eodp + reclen);
2386                 } else {
2387                         /*
2388                          * Add normal entry:
2389                          */
2390                         odp->d_ino = objnum;
2391                         odp->d_reclen = reclen;
2392                         /* NOTE: d_off is the offset for the *next* entry */
2393                         next = &(odp->d_off);
2394                         (void) strncpy(odp->d_name, zap.za_name,
2395                             DIRENT64_NAMELEN(reclen));
2396                         odp = (dirent64_t *)((intptr_t)odp + reclen);
2397                 }
2398                 outcount += reclen;
2399 
2400                 ASSERT(outcount <= bufsize);
2401 
2402                 /* Prefetch znode */
2403                 if (prefetch)
2404                         dmu_prefetch(os, objnum, 0, 0);
2405 
2406         skip_entry:
2407                 /*
2408                  * Move to the next entry, fill in the previous offset.
2409                  */
2410                 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2411                         zap_cursor_advance(&zc);
2412                         offset = zap_cursor_serialize(&zc);
2413                 } else {
2414                         offset += 1;
2415                 }
2416                 if (next)
2417                         *next = offset;
2418         }
2419         zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2420 
2421         if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2422                 iovp->iov_base += outcount;
2423                 iovp->iov_len -= outcount;
2424                 uio->uio_resid -= outcount;
2425         } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2426                 /*
2427                  * Reset the pointer.
2428                  */
2429                 offset = uio->uio_loffset;
2430         }
2431 
2432 update:
2433         zap_cursor_fini(&zc);
2434         if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2435                 kmem_free(outbuf, bufsize);
2436 
2437         if (error == ENOENT)
2438                 error = 0;
2439 
2440         ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2441 
2442         uio->uio_loffset = offset;
2443         ZFS_EXIT(zfsvfs);
2444         return (error);
2445 }
2446 
2447 ulong_t zfs_fsync_sync_cnt = 4;
2448 
2449 static int
2450 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2451 {
2452         znode_t *zp = VTOZ(vp);
2453         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2454 
2455         /*
2456          * Regardless of whether this is required for standards conformance,
2457          * this is the logical behavior when fsync() is called on a file with
2458          * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2459          * going to be pushed out as part of the zil_commit().
2460          */
2461         if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2462             (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2463                 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2464 
2465         (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2466 
2467         if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2468                 ZFS_ENTER(zfsvfs);
2469                 ZFS_VERIFY_ZP(zp);
2470                 zil_commit(zfsvfs->z_log, zp->z_id);
2471                 ZFS_EXIT(zfsvfs);
2472         }
2473         return (0);
2474 }
2475 
2476 
2477 /*
2478  * Get the requested file attributes and place them in the provided
2479  * vattr structure.
2480  *
2481  *      IN:     vp      - vnode of file.
2482  *              vap     - va_mask identifies requested attributes.
2483  *                        If AT_XVATTR set, then optional attrs are requested
2484  *              flags   - ATTR_NOACLCHECK (CIFS server context)
2485  *              cr      - credentials of caller.
2486  *              ct      - caller context
2487  *
2488  *      OUT:    vap     - attribute values.
2489  *
2490  *      RETURN: 0 (always succeeds).
2491  */
2492 /* ARGSUSED */
2493 static int
2494 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2495     caller_context_t *ct)
2496 {
2497         znode_t *zp = VTOZ(vp);
2498         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2499         int     error = 0;
2500         uint64_t links;
2501         uint64_t mtime[2], ctime[2];
2502         xvattr_t *xvap = (xvattr_t *)vap;       /* vap may be an xvattr_t * */
2503         xoptattr_t *xoap = NULL;
2504         boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2505         sa_bulk_attr_t bulk[2];
2506         int count = 0;
2507 
2508         ZFS_ENTER(zfsvfs);
2509         ZFS_VERIFY_ZP(zp);
2510 
2511         zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2512 
2513         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2514         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2515 
2516         if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2517                 ZFS_EXIT(zfsvfs);
2518                 return (error);
2519         }
2520 
2521         /*
2522          * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2523          * Also, if we are the owner don't bother, since owner should
2524          * always be allowed to read basic attributes of file.
2525          */
2526         if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2527             (vap->va_uid != crgetuid(cr))) {
2528                 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2529                     skipaclchk, cr)) {
2530                         ZFS_EXIT(zfsvfs);
2531                         return (error);
2532                 }
2533         }
2534 
2535         /*
2536          * Return all attributes.  It's cheaper to provide the answer
2537          * than to determine whether we were asked the question.
2538          */
2539 
2540         mutex_enter(&zp->z_lock);
2541         vap->va_type = vp->v_type;
2542         vap->va_mode = zp->z_mode & MODEMASK;
2543         vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2544         vap->va_nodeid = zp->z_id;
2545         if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2546                 links = zp->z_links + 1;
2547         else
2548                 links = zp->z_links;
2549         vap->va_nlink = MIN(links, UINT32_MAX);      /* nlink_t limit! */
2550         vap->va_size = zp->z_size;
2551         vap->va_rdev = vp->v_rdev;
2552         vap->va_seq = zp->z_seq;
2553 
2554         /*
2555          * Add in any requested optional attributes and the create time.
2556          * Also set the corresponding bits in the returned attribute bitmap.
2557          */
2558         if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2559                 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2560                         xoap->xoa_archive =
2561                             ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2562                         XVA_SET_RTN(xvap, XAT_ARCHIVE);
2563                 }
2564 
2565                 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2566                         xoap->xoa_readonly =
2567                             ((zp->z_pflags & ZFS_READONLY) != 0);
2568                         XVA_SET_RTN(xvap, XAT_READONLY);
2569                 }
2570 
2571                 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2572                         xoap->xoa_system =
2573                             ((zp->z_pflags & ZFS_SYSTEM) != 0);
2574                         XVA_SET_RTN(xvap, XAT_SYSTEM);
2575                 }
2576 
2577                 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2578                         xoap->xoa_hidden =
2579                             ((zp->z_pflags & ZFS_HIDDEN) != 0);
2580                         XVA_SET_RTN(xvap, XAT_HIDDEN);
2581                 }
2582 
2583                 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2584                         xoap->xoa_nounlink =
2585                             ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2586                         XVA_SET_RTN(xvap, XAT_NOUNLINK);
2587                 }
2588 
2589                 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2590                         xoap->xoa_immutable =
2591                             ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2592                         XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2593                 }
2594 
2595                 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2596                         xoap->xoa_appendonly =
2597                             ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2598                         XVA_SET_RTN(xvap, XAT_APPENDONLY);
2599                 }
2600 
2601                 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2602                         xoap->xoa_nodump =
2603                             ((zp->z_pflags & ZFS_NODUMP) != 0);
2604                         XVA_SET_RTN(xvap, XAT_NODUMP);
2605                 }
2606 
2607                 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2608                         xoap->xoa_opaque =
2609                             ((zp->z_pflags & ZFS_OPAQUE) != 0);
2610                         XVA_SET_RTN(xvap, XAT_OPAQUE);
2611                 }
2612 
2613                 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2614                         xoap->xoa_av_quarantined =
2615                             ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2616                         XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2617                 }
2618 
2619                 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2620                         xoap->xoa_av_modified =
2621                             ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2622                         XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2623                 }
2624 
2625                 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2626                     vp->v_type == VREG) {
2627                         zfs_sa_get_scanstamp(zp, xvap);
2628                 }
2629 
2630                 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2631                         uint64_t times[2];
2632 
2633                         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2634                             times, sizeof (times));
2635                         ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2636                         XVA_SET_RTN(xvap, XAT_CREATETIME);
2637                 }
2638 
2639                 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2640                         xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2641                         XVA_SET_RTN(xvap, XAT_REPARSE);
2642                 }
2643                 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2644                         xoap->xoa_generation = zp->z_gen;
2645                         XVA_SET_RTN(xvap, XAT_GEN);
2646                 }
2647 
2648                 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2649                         xoap->xoa_offline =
2650                             ((zp->z_pflags & ZFS_OFFLINE) != 0);
2651                         XVA_SET_RTN(xvap, XAT_OFFLINE);
2652                 }
2653 
2654                 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2655                         xoap->xoa_sparse =
2656                             ((zp->z_pflags & ZFS_SPARSE) != 0);
2657                         XVA_SET_RTN(xvap, XAT_SPARSE);
2658                 }
2659         }
2660 
2661         ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2662         ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2663         ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2664 
2665         mutex_exit(&zp->z_lock);
2666 
2667         sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2668 
2669         if (zp->z_blksz == 0) {
2670                 /*
2671                  * Block size hasn't been set; suggest maximal I/O transfers.
2672                  */
2673                 vap->va_blksize = zfsvfs->z_max_blksz;
2674         }
2675 
2676         ZFS_EXIT(zfsvfs);
2677         return (0);
2678 }
2679 
2680 /*
2681  * Set the file attributes to the values contained in the
2682  * vattr structure.
2683  *
2684  *      IN:     vp      - vnode of file to be modified.
2685  *              vap     - new attribute values.
2686  *                        If AT_XVATTR set, then optional attrs are being set
2687  *              flags   - ATTR_UTIME set if non-default time values provided.
2688  *                      - ATTR_NOACLCHECK (CIFS context only).
2689  *              cr      - credentials of caller.
2690  *              ct      - caller context
2691  *
2692  *      RETURN: 0 on success, error code on failure.
2693  *
2694  * Timestamps:
2695  *      vp - ctime updated, mtime updated if size changed.
2696  */
2697 /* ARGSUSED */
2698 static int
2699 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2700     caller_context_t *ct)
2701 {
2702         znode_t         *zp = VTOZ(vp);
2703         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
2704         zilog_t         *zilog;
2705         dmu_tx_t        *tx;
2706         vattr_t         oldva;
2707         xvattr_t        tmpxvattr;
2708         uint_t          mask = vap->va_mask;
2709         uint_t          saved_mask = 0;
2710         int             trim_mask = 0;
2711         uint64_t        new_mode;
2712         uint64_t        new_uid, new_gid;
2713         uint64_t        xattr_obj;
2714         uint64_t        mtime[2], ctime[2];
2715         znode_t         *attrzp;
2716         int             need_policy = FALSE;
2717         int             err, err2;
2718         zfs_fuid_info_t *fuidp = NULL;
2719         xvattr_t *xvap = (xvattr_t *)vap;       /* vap may be an xvattr_t * */
2720         xoptattr_t      *xoap;
2721         zfs_acl_t       *aclp;
2722         boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2723         boolean_t       fuid_dirtied = B_FALSE;
2724         sa_bulk_attr_t  bulk[7], xattr_bulk[7];
2725         int             count = 0, xattr_count = 0;
2726 
2727         if (mask == 0)
2728                 return (0);
2729 
2730         if (mask & AT_NOSET)
2731                 return (SET_ERROR(EINVAL));
2732 
2733         ZFS_ENTER(zfsvfs);
2734         ZFS_VERIFY_ZP(zp);
2735 
2736         zilog = zfsvfs->z_log;
2737 
2738         /*
2739          * Make sure that if we have ephemeral uid/gid or xvattr specified
2740          * that file system is at proper version level
2741          */
2742 
2743         if (zfsvfs->z_use_fuids == B_FALSE &&
2744             (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2745             ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2746             (mask & AT_XVATTR))) {
2747                 ZFS_EXIT(zfsvfs);
2748                 return (SET_ERROR(EINVAL));
2749         }
2750 
2751         if (mask & AT_SIZE && vp->v_type == VDIR) {
2752                 ZFS_EXIT(zfsvfs);
2753                 return (SET_ERROR(EISDIR));
2754         }
2755 
2756         if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2757                 ZFS_EXIT(zfsvfs);
2758                 return (SET_ERROR(EINVAL));
2759         }
2760 
2761         /*
2762          * If this is an xvattr_t, then get a pointer to the structure of
2763          * optional attributes.  If this is NULL, then we have a vattr_t.
2764          */
2765         xoap = xva_getxoptattr(xvap);
2766 
2767         xva_init(&tmpxvattr);
2768 
2769         /*
2770          * Immutable files can only alter immutable bit and atime
2771          */
2772         if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2773             ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2774             ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2775                 ZFS_EXIT(zfsvfs);
2776                 return (SET_ERROR(EPERM));
2777         }
2778 
2779         if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2780                 ZFS_EXIT(zfsvfs);
2781                 return (SET_ERROR(EPERM));
2782         }
2783 
2784         /*
2785          * Verify timestamps doesn't overflow 32 bits.
2786          * ZFS can handle large timestamps, but 32bit syscalls can't
2787          * handle times greater than 2039.  This check should be removed
2788          * once large timestamps are fully supported.
2789          */
2790         if (mask & (AT_ATIME | AT_MTIME)) {
2791                 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2792                     ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2793                         ZFS_EXIT(zfsvfs);
2794                         return (SET_ERROR(EOVERFLOW));
2795                 }
2796         }
2797 
2798 top:
2799         attrzp = NULL;
2800         aclp = NULL;
2801 
2802         /* Can this be moved to before the top label? */
2803         if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2804                 ZFS_EXIT(zfsvfs);
2805                 return (SET_ERROR(EROFS));
2806         }
2807 
2808         /*
2809          * First validate permissions
2810          */
2811 
2812         if (mask & AT_SIZE) {
2813                 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2814                 if (err) {
2815                         ZFS_EXIT(zfsvfs);
2816                         return (err);
2817                 }
2818                 /*
2819                  * XXX - Note, we are not providing any open
2820                  * mode flags here (like FNDELAY), so we may
2821                  * block if there are locks present... this
2822                  * should be addressed in openat().
2823                  */
2824                 /* XXX - would it be OK to generate a log record here? */
2825                 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2826                 if (err) {
2827                         ZFS_EXIT(zfsvfs);
2828                         return (err);
2829                 }
2830 
2831                 if (vap->va_size == 0)
2832                         vnevent_truncate(ZTOV(zp), ct);
2833         }
2834 
2835         if (mask & (AT_ATIME|AT_MTIME) ||
2836             ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2837             XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2838             XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2839             XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2840             XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2841             XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2842             XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2843                 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2844                     skipaclchk, cr);
2845         }
2846 
2847         if (mask & (AT_UID|AT_GID)) {
2848                 int     idmask = (mask & (AT_UID|AT_GID));
2849                 int     take_owner;
2850                 int     take_group;
2851 
2852                 /*
2853                  * NOTE: even if a new mode is being set,
2854                  * we may clear S_ISUID/S_ISGID bits.
2855                  */
2856 
2857                 if (!(mask & AT_MODE))
2858                         vap->va_mode = zp->z_mode;
2859 
2860                 /*
2861                  * Take ownership or chgrp to group we are a member of
2862                  */
2863 
2864                 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2865                 take_group = (mask & AT_GID) &&
2866                     zfs_groupmember(zfsvfs, vap->va_gid, cr);
2867 
2868                 /*
2869                  * If both AT_UID and AT_GID are set then take_owner and
2870                  * take_group must both be set in order to allow taking
2871                  * ownership.
2872                  *
2873                  * Otherwise, send the check through secpolicy_vnode_setattr()
2874                  *
2875                  */
2876 
2877                 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2878                     ((idmask == AT_UID) && take_owner) ||
2879                     ((idmask == AT_GID) && take_group)) {
2880                         if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2881                             skipaclchk, cr) == 0) {
2882                                 /*
2883                                  * Remove setuid/setgid for non-privileged users
2884                                  */
2885                                 secpolicy_setid_clear(vap, cr);
2886                                 trim_mask = (mask & (AT_UID|AT_GID));
2887                         } else {
2888                                 need_policy =  TRUE;
2889                         }
2890                 } else {
2891                         need_policy =  TRUE;
2892                 }
2893         }
2894 
2895         mutex_enter(&zp->z_lock);
2896         oldva.va_mode = zp->z_mode;
2897         zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2898         if (mask & AT_XVATTR) {
2899                 /*
2900                  * Update xvattr mask to include only those attributes
2901                  * that are actually changing.
2902                  *
2903                  * the bits will be restored prior to actually setting
2904                  * the attributes so the caller thinks they were set.
2905                  */
2906                 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2907                         if (xoap->xoa_appendonly !=
2908                             ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2909                                 need_policy = TRUE;
2910                         } else {
2911                                 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2912                                 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2913                         }
2914                 }
2915 
2916                 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2917                         if (xoap->xoa_nounlink !=
2918                             ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2919                                 need_policy = TRUE;
2920                         } else {
2921                                 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2922                                 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2923                         }
2924                 }
2925 
2926                 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2927                         if (xoap->xoa_immutable !=
2928                             ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2929                                 need_policy = TRUE;
2930                         } else {
2931                                 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2932                                 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2933                         }
2934                 }
2935 
2936                 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2937                         if (xoap->xoa_nodump !=
2938                             ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2939                                 need_policy = TRUE;
2940                         } else {
2941                                 XVA_CLR_REQ(xvap, XAT_NODUMP);
2942                                 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2943                         }
2944                 }
2945 
2946                 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2947                         if (xoap->xoa_av_modified !=
2948                             ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2949                                 need_policy = TRUE;
2950                         } else {
2951                                 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2952                                 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2953                         }
2954                 }
2955 
2956                 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2957                         if ((vp->v_type != VREG &&
2958                             xoap->xoa_av_quarantined) ||
2959                             xoap->xoa_av_quarantined !=
2960                             ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2961                                 need_policy = TRUE;
2962                         } else {
2963                                 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2964                                 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2965                         }
2966                 }
2967 
2968                 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2969                         mutex_exit(&zp->z_lock);
2970                         ZFS_EXIT(zfsvfs);
2971                         return (SET_ERROR(EPERM));
2972                 }
2973 
2974                 if (need_policy == FALSE &&
2975                     (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2976                     XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2977                         need_policy = TRUE;
2978                 }
2979         }
2980 
2981         mutex_exit(&zp->z_lock);
2982 
2983         if (mask & AT_MODE) {
2984                 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2985                         err = secpolicy_setid_setsticky_clear(vp, vap,
2986                             &oldva, cr);
2987                         if (err) {
2988                                 ZFS_EXIT(zfsvfs);
2989                                 return (err);
2990                         }
2991                         trim_mask |= AT_MODE;
2992                 } else {
2993                         need_policy = TRUE;
2994                 }
2995         }
2996 
2997         if (need_policy) {
2998                 /*
2999                  * If trim_mask is set then take ownership
3000                  * has been granted or write_acl is present and user
3001                  * has the ability to modify mode.  In that case remove
3002                  * UID|GID and or MODE from mask so that
3003                  * secpolicy_vnode_setattr() doesn't revoke it.
3004                  */
3005 
3006                 if (trim_mask) {
3007                         saved_mask = vap->va_mask;
3008                         vap->va_mask &= ~trim_mask;
3009                 }
3010                 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3011                     (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3012                 if (err) {
3013                         ZFS_EXIT(zfsvfs);
3014                         return (err);
3015                 }
3016 
3017                 if (trim_mask)
3018                         vap->va_mask |= saved_mask;
3019         }
3020 
3021         /*
3022          * secpolicy_vnode_setattr, or take ownership may have
3023          * changed va_mask
3024          */
3025         mask = vap->va_mask;
3026 
3027         if ((mask & (AT_UID | AT_GID))) {
3028                 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3029                     &xattr_obj, sizeof (xattr_obj));
3030 
3031                 if (err == 0 && xattr_obj) {
3032                         err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3033                         if (err)
3034                                 goto out2;
3035                 }
3036                 if (mask & AT_UID) {
3037                         new_uid = zfs_fuid_create(zfsvfs,
3038                             (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3039                         if (new_uid != zp->z_uid &&
3040                             zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3041                                 if (attrzp)
3042                                         VN_RELE(ZTOV(attrzp));
3043                                 err = SET_ERROR(EDQUOT);
3044                                 goto out2;
3045                         }
3046                 }
3047 
3048                 if (mask & AT_GID) {
3049                         new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3050                             cr, ZFS_GROUP, &fuidp);
3051                         if (new_gid != zp->z_gid &&
3052                             zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3053                                 if (attrzp)
3054                                         VN_RELE(ZTOV(attrzp));
3055                                 err = SET_ERROR(EDQUOT);
3056                                 goto out2;
3057                         }
3058                 }
3059         }
3060         tx = dmu_tx_create(zfsvfs->z_os);
3061 
3062         if (mask & AT_MODE) {
3063                 uint64_t pmode = zp->z_mode;
3064                 uint64_t acl_obj;
3065                 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3066 
3067                 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3068                     !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3069                         err = SET_ERROR(EPERM);
3070                         goto out;
3071                 }
3072 
3073                 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3074                         goto out;
3075 
3076                 mutex_enter(&zp->z_lock);
3077                 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3078                         /*
3079                          * Are we upgrading ACL from old V0 format
3080                          * to V1 format?
3081                          */
3082                         if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3083                             zfs_znode_acl_version(zp) ==
3084                             ZFS_ACL_VERSION_INITIAL) {
3085                                 dmu_tx_hold_free(tx, acl_obj, 0,
3086                                     DMU_OBJECT_END);
3087                                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3088                                     0, aclp->z_acl_bytes);
3089                         } else {
3090                                 dmu_tx_hold_write(tx, acl_obj, 0,
3091                                     aclp->z_acl_bytes);
3092                         }
3093                 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3094                         dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3095                             0, aclp->z_acl_bytes);
3096                 }
3097                 mutex_exit(&zp->z_lock);
3098                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3099         } else {
3100                 if ((mask & AT_XVATTR) &&
3101                     XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3102                         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3103                 else
3104                         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3105         }
3106 
3107         if (attrzp) {
3108                 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3109         }
3110 
3111         fuid_dirtied = zfsvfs->z_fuid_dirty;
3112         if (fuid_dirtied)
3113                 zfs_fuid_txhold(zfsvfs, tx);
3114 
3115         zfs_sa_upgrade_txholds(tx, zp);
3116 
3117         err = dmu_tx_assign(tx, TXG_WAIT);
3118         if (err)
3119                 goto out;
3120 
3121         count = 0;
3122         /*
3123          * Set each attribute requested.
3124          * We group settings according to the locks they need to acquire.
3125          *
3126          * Note: you cannot set ctime directly, although it will be
3127          * updated as a side-effect of calling this function.
3128          */
3129 
3130 
3131         if (mask & (AT_UID|AT_GID|AT_MODE))
3132                 mutex_enter(&zp->z_acl_lock);
3133         mutex_enter(&zp->z_lock);
3134 
3135         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3136             &zp->z_pflags, sizeof (zp->z_pflags));
3137 
3138         if (attrzp) {
3139                 if (mask & (AT_UID|AT_GID|AT_MODE))
3140                         mutex_enter(&attrzp->z_acl_lock);
3141                 mutex_enter(&attrzp->z_lock);
3142                 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3143                     SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3144                     sizeof (attrzp->z_pflags));
3145         }
3146 
3147         if (mask & (AT_UID|AT_GID)) {
3148 
3149                 if (mask & AT_UID) {
3150                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3151                             &new_uid, sizeof (new_uid));
3152                         zp->z_uid = new_uid;
3153                         if (attrzp) {
3154                                 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3155                                     SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3156                                     sizeof (new_uid));
3157                                 attrzp->z_uid = new_uid;
3158                         }
3159                 }
3160 
3161                 if (mask & AT_GID) {
3162                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3163                             NULL, &new_gid, sizeof (new_gid));
3164                         zp->z_gid = new_gid;
3165                         if (attrzp) {
3166                                 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3167                                     SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3168                                     sizeof (new_gid));
3169                                 attrzp->z_gid = new_gid;
3170                         }
3171                 }
3172                 if (!(mask & AT_MODE)) {
3173                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3174                             NULL, &new_mode, sizeof (new_mode));
3175                         new_mode = zp->z_mode;
3176                 }
3177                 err = zfs_acl_chown_setattr(zp);
3178                 ASSERT(err == 0);
3179                 if (attrzp) {
3180                         err = zfs_acl_chown_setattr(attrzp);
3181                         ASSERT(err == 0);
3182                 }
3183         }
3184 
3185         if (mask & AT_MODE) {
3186                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3187                     &new_mode, sizeof (new_mode));
3188                 zp->z_mode = new_mode;
3189                 ASSERT3U((uintptr_t)aclp, !=, NULL);
3190                 err = zfs_aclset_common(zp, aclp, cr, tx);
3191                 ASSERT0(err);
3192                 if (zp->z_acl_cached)
3193                         zfs_acl_free(zp->z_acl_cached);
3194                 zp->z_acl_cached = aclp;
3195                 aclp = NULL;
3196         }
3197 
3198 
3199         if (mask & AT_ATIME) {
3200                 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3201                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3202                     &zp->z_atime, sizeof (zp->z_atime));
3203         }
3204 
3205         if (mask & AT_MTIME) {
3206                 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3207                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3208                     mtime, sizeof (mtime));
3209         }
3210 
3211         /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3212         if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3213                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3214                     NULL, mtime, sizeof (mtime));
3215                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3216                     &ctime, sizeof (ctime));
3217                 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3218                     B_TRUE);
3219         } else if (mask != 0) {
3220                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3221                     &ctime, sizeof (ctime));
3222                 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3223                     B_TRUE);
3224                 if (attrzp) {
3225                         SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3226                             SA_ZPL_CTIME(zfsvfs), NULL,
3227                             &ctime, sizeof (ctime));
3228                         zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3229                             mtime, ctime, B_TRUE);
3230                 }
3231         }
3232         /*
3233          * Do this after setting timestamps to prevent timestamp
3234          * update from toggling bit
3235          */
3236 
3237         if (xoap && (mask & AT_XVATTR)) {
3238 
3239                 /*
3240                  * restore trimmed off masks
3241                  * so that return masks can be set for caller.
3242                  */
3243 
3244                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3245                         XVA_SET_REQ(xvap, XAT_APPENDONLY);
3246                 }
3247                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3248                         XVA_SET_REQ(xvap, XAT_NOUNLINK);
3249                 }
3250                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3251                         XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3252                 }
3253                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3254                         XVA_SET_REQ(xvap, XAT_NODUMP);
3255                 }
3256                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3257                         XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3258                 }
3259                 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3260                         XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3261                 }
3262 
3263                 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3264                         ASSERT(vp->v_type == VREG);
3265 
3266                 zfs_xvattr_set(zp, xvap, tx);
3267         }
3268 
3269         if (fuid_dirtied)
3270                 zfs_fuid_sync(zfsvfs, tx);
3271 
3272         if (mask != 0)
3273                 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3274 
3275         mutex_exit(&zp->z_lock);
3276         if (mask & (AT_UID|AT_GID|AT_MODE))
3277                 mutex_exit(&zp->z_acl_lock);
3278 
3279         if (attrzp) {
3280                 if (mask & (AT_UID|AT_GID|AT_MODE))
3281                         mutex_exit(&attrzp->z_acl_lock);
3282                 mutex_exit(&attrzp->z_lock);
3283         }
3284 out:
3285         if (err == 0 && attrzp) {
3286                 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3287                     xattr_count, tx);
3288                 ASSERT(err2 == 0);
3289         }
3290 
3291         if (attrzp)
3292                 VN_RELE(ZTOV(attrzp));
3293 
3294         if (aclp)
3295                 zfs_acl_free(aclp);
3296 
3297         if (fuidp) {
3298                 zfs_fuid_info_free(fuidp);
3299                 fuidp = NULL;
3300         }
3301 
3302         if (err) {
3303                 dmu_tx_abort(tx);
3304                 if (err == ERESTART)
3305                         goto top;
3306         } else {
3307                 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3308                 dmu_tx_commit(tx);
3309         }
3310 
3311 out2:
3312         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3313                 zil_commit(zilog, 0);
3314 
3315         ZFS_EXIT(zfsvfs);
3316         return (err);
3317 }
3318 
3319 typedef struct zfs_zlock {
3320         krwlock_t       *zl_rwlock;     /* lock we acquired */
3321         znode_t         *zl_znode;      /* znode we held */
3322         struct zfs_zlock *zl_next;      /* next in list */
3323 } zfs_zlock_t;
3324 
3325 /*
3326  * Drop locks and release vnodes that were held by zfs_rename_lock().
3327  */
3328 static void
3329 zfs_rename_unlock(zfs_zlock_t **zlpp)
3330 {
3331         zfs_zlock_t *zl;
3332 
3333         while ((zl = *zlpp) != NULL) {
3334                 if (zl->zl_znode != NULL)
3335                         VN_RELE(ZTOV(zl->zl_znode));
3336                 rw_exit(zl->zl_rwlock);
3337                 *zlpp = zl->zl_next;
3338                 kmem_free(zl, sizeof (*zl));
3339         }
3340 }
3341 
3342 /*
3343  * Search back through the directory tree, using the ".." entries.
3344  * Lock each directory in the chain to prevent concurrent renames.
3345  * Fail any attempt to move a directory into one of its own descendants.
3346  * XXX - z_parent_lock can overlap with map or grow locks
3347  */
3348 static int
3349 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3350 {
3351         zfs_zlock_t     *zl;
3352         znode_t         *zp = tdzp;
3353         uint64_t        rootid = zp->z_zfsvfs->z_root;
3354         uint64_t        oidp = zp->z_id;
3355         krwlock_t       *rwlp = &szp->z_parent_lock;
3356         krw_t           rw = RW_WRITER;
3357 
3358         /*
3359          * First pass write-locks szp and compares to zp->z_id.
3360          * Later passes read-lock zp and compare to zp->z_parent.
3361          */
3362         do {
3363                 if (!rw_tryenter(rwlp, rw)) {
3364                         /*
3365                          * Another thread is renaming in this path.
3366                          * Note that if we are a WRITER, we don't have any
3367                          * parent_locks held yet.
3368                          */
3369                         if (rw == RW_READER && zp->z_id > szp->z_id) {
3370                                 /*
3371                                  * Drop our locks and restart
3372                                  */
3373                                 zfs_rename_unlock(&zl);
3374                                 *zlpp = NULL;
3375                                 zp = tdzp;
3376                                 oidp = zp->z_id;
3377                                 rwlp = &szp->z_parent_lock;
3378                                 rw = RW_WRITER;
3379                                 continue;
3380                         } else {
3381                                 /*
3382                                  * Wait for other thread to drop its locks
3383                                  */
3384                                 rw_enter(rwlp, rw);
3385                         }
3386                 }
3387 
3388                 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3389                 zl->zl_rwlock = rwlp;
3390                 zl->zl_znode = NULL;
3391                 zl->zl_next = *zlpp;
3392                 *zlpp = zl;
3393 
3394                 if (oidp == szp->z_id)               /* We're a descendant of szp */
3395                         return (SET_ERROR(EINVAL));
3396 
3397                 if (oidp == rootid)             /* We've hit the top */
3398                         return (0);
3399 
3400                 if (rw == RW_READER) {          /* i.e. not the first pass */
3401                         int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3402                         if (error)
3403                                 return (error);
3404                         zl->zl_znode = zp;
3405                 }
3406                 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3407                     &oidp, sizeof (oidp));
3408                 rwlp = &zp->z_parent_lock;
3409                 rw = RW_READER;
3410 
3411         } while (zp->z_id != sdzp->z_id);
3412 
3413         return (0);
3414 }
3415 
3416 /*
3417  * Move an entry from the provided source directory to the target
3418  * directory.  Change the entry name as indicated.
3419  *
3420  *      IN:     sdvp    - Source directory containing the "old entry".
3421  *              snm     - Old entry name.
3422  *              tdvp    - Target directory to contain the "new entry".
3423  *              tnm     - New entry name.
3424  *              cr      - credentials of caller.
3425  *              ct      - caller context
3426  *              flags   - case flags
3427  *
3428  *      RETURN: 0 on success, error code on failure.
3429  *
3430  * Timestamps:
3431  *      sdvp,tdvp - ctime|mtime updated
3432  */
3433 /*ARGSUSED*/
3434 static int
3435 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3436     caller_context_t *ct, int flags)
3437 {
3438         znode_t         *tdzp, *szp, *tzp;
3439         znode_t         *sdzp = VTOZ(sdvp);
3440         zfsvfs_t        *zfsvfs = sdzp->z_zfsvfs;
3441         zilog_t         *zilog;
3442         vnode_t         *realvp;
3443         zfs_dirlock_t   *sdl, *tdl;
3444         dmu_tx_t        *tx;
3445         zfs_zlock_t     *zl;
3446         int             cmp, serr, terr;
3447         int             error = 0;
3448         int             zflg = 0;
3449         boolean_t       waited = B_FALSE;
3450 
3451         ZFS_ENTER(zfsvfs);
3452         ZFS_VERIFY_ZP(sdzp);
3453         zilog = zfsvfs->z_log;
3454 
3455         /*
3456          * Make sure we have the real vp for the target directory.
3457          */
3458         if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3459                 tdvp = realvp;
3460 
3461         tdzp = VTOZ(tdvp);
3462         ZFS_VERIFY_ZP(tdzp);
3463 
3464         /*
3465          * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3466          * ctldir appear to have the same v_vfsp.
3467          */
3468         if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3469                 ZFS_EXIT(zfsvfs);
3470                 return (SET_ERROR(EXDEV));
3471         }
3472 
3473         if (zfsvfs->z_utf8 && u8_validate(tnm,
3474             strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3475                 ZFS_EXIT(zfsvfs);
3476                 return (SET_ERROR(EILSEQ));
3477         }
3478 
3479         if (flags & FIGNORECASE)
3480                 zflg |= ZCILOOK;
3481 
3482 top:
3483         szp = NULL;
3484         tzp = NULL;
3485         zl = NULL;
3486 
3487         /*
3488          * This is to prevent the creation of links into attribute space
3489          * by renaming a linked file into/outof an attribute directory.
3490          * See the comment in zfs_link() for why this is considered bad.
3491          */
3492         if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3493                 ZFS_EXIT(zfsvfs);
3494                 return (SET_ERROR(EINVAL));
3495         }
3496 
3497         /*
3498          * Lock source and target directory entries.  To prevent deadlock,
3499          * a lock ordering must be defined.  We lock the directory with
3500          * the smallest object id first, or if it's a tie, the one with
3501          * the lexically first name.
3502          */
3503         if (sdzp->z_id < tdzp->z_id) {
3504                 cmp = -1;
3505         } else if (sdzp->z_id > tdzp->z_id) {
3506                 cmp = 1;
3507         } else {
3508                 /*
3509                  * First compare the two name arguments without
3510                  * considering any case folding.
3511                  */
3512                 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3513 
3514                 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3515                 ASSERT(error == 0 || !zfsvfs->z_utf8);
3516                 if (cmp == 0) {
3517                         /*
3518                          * POSIX: "If the old argument and the new argument
3519                          * both refer to links to the same existing file,
3520                          * the rename() function shall return successfully
3521                          * and perform no other action."
3522                          */
3523                         ZFS_EXIT(zfsvfs);
3524                         return (0);
3525                 }
3526                 /*
3527                  * If the file system is case-folding, then we may
3528                  * have some more checking to do.  A case-folding file
3529                  * system is either supporting mixed case sensitivity
3530                  * access or is completely case-insensitive.  Note
3531                  * that the file system is always case preserving.
3532                  *
3533                  * In mixed sensitivity mode case sensitive behavior
3534                  * is the default.  FIGNORECASE must be used to
3535                  * explicitly request case insensitive behavior.
3536                  *
3537                  * If the source and target names provided differ only
3538                  * by case (e.g., a request to rename 'tim' to 'Tim'),
3539                  * we will treat this as a special case in the
3540                  * case-insensitive mode: as long as the source name
3541                  * is an exact match, we will allow this to proceed as
3542                  * a name-change request.
3543                  */
3544                 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3545                     (zfsvfs->z_case == ZFS_CASE_MIXED &&
3546                     flags & FIGNORECASE)) &&
3547                     u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3548                     &error) == 0) {
3549                         /*
3550                          * case preserving rename request, require exact
3551                          * name matches
3552                          */
3553                         zflg |= ZCIEXACT;
3554                         zflg &= ~ZCILOOK;
3555                 }
3556         }
3557 
3558         /*
3559          * If the source and destination directories are the same, we should
3560          * grab the z_name_lock of that directory only once.
3561          */
3562         if (sdzp == tdzp) {
3563                 zflg |= ZHAVELOCK;
3564                 rw_enter(&sdzp->z_name_lock, RW_READER);
3565         }
3566 
3567         if (cmp < 0) {
3568                 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3569                     ZEXISTS | zflg, NULL, NULL);
3570                 terr = zfs_dirent_lock(&tdl,
3571                     tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3572         } else {
3573                 terr = zfs_dirent_lock(&tdl,
3574                     tdzp, tnm, &tzp, zflg, NULL, NULL);
3575                 serr = zfs_dirent_lock(&sdl,
3576                     sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3577                     NULL, NULL);
3578         }
3579 
3580         if (serr) {
3581                 /*
3582                  * Source entry invalid or not there.
3583                  */
3584                 if (!terr) {
3585                         zfs_dirent_unlock(tdl);
3586                         if (tzp)
3587                                 VN_RELE(ZTOV(tzp));
3588                 }
3589 
3590                 if (sdzp == tdzp)
3591                         rw_exit(&sdzp->z_name_lock);
3592 
3593                 if (strcmp(snm, "..") == 0)
3594                         serr = SET_ERROR(EINVAL);
3595                 ZFS_EXIT(zfsvfs);
3596                 return (serr);
3597         }
3598         if (terr) {
3599                 zfs_dirent_unlock(sdl);
3600                 VN_RELE(ZTOV(szp));
3601 
3602                 if (sdzp == tdzp)
3603                         rw_exit(&sdzp->z_name_lock);
3604 
3605                 if (strcmp(tnm, "..") == 0)
3606                         terr = SET_ERROR(EINVAL);
3607                 ZFS_EXIT(zfsvfs);
3608                 return (terr);
3609         }
3610 
3611         /*
3612          * Must have write access at the source to remove the old entry
3613          * and write access at the target to create the new entry.
3614          * Note that if target and source are the same, this can be
3615          * done in a single check.
3616          */
3617 
3618         if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3619                 goto out;
3620 
3621         if (ZTOV(szp)->v_type == VDIR) {
3622                 /*
3623                  * Check to make sure rename is valid.
3624                  * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3625                  */
3626                 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3627                         goto out;
3628         }
3629 
3630         /*
3631          * Does target exist?
3632          */
3633         if (tzp) {
3634                 /*
3635                  * Source and target must be the same type.
3636                  */
3637                 if (ZTOV(szp)->v_type == VDIR) {
3638                         if (ZTOV(tzp)->v_type != VDIR) {
3639                                 error = SET_ERROR(ENOTDIR);
3640                                 goto out;
3641                         }
3642                 } else {
3643                         if (ZTOV(tzp)->v_type == VDIR) {
3644                                 error = SET_ERROR(EISDIR);
3645                                 goto out;
3646                         }
3647                 }
3648                 /*
3649                  * POSIX dictates that when the source and target
3650                  * entries refer to the same file object, rename
3651                  * must do nothing and exit without error.
3652                  */
3653                 if (szp->z_id == tzp->z_id) {
3654                         error = 0;
3655                         goto out;
3656                 }
3657         }
3658 
3659         vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3660         if (tzp)
3661                 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3662 
3663         /*
3664          * notify the target directory if it is not the same
3665          * as source directory.
3666          */
3667         if (tdvp != sdvp) {
3668                 vnevent_rename_dest_dir(tdvp, ct);
3669         }
3670 
3671         tx = dmu_tx_create(zfsvfs->z_os);
3672         dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3673         dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3674         dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3675         dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3676         if (sdzp != tdzp) {
3677                 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3678                 zfs_sa_upgrade_txholds(tx, tdzp);
3679         }
3680         if (tzp) {
3681                 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3682                 zfs_sa_upgrade_txholds(tx, tzp);
3683         }
3684 
3685         zfs_sa_upgrade_txholds(tx, szp);
3686         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3687         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3688         if (error) {
3689                 if (zl != NULL)
3690                         zfs_rename_unlock(&zl);
3691                 zfs_dirent_unlock(sdl);
3692                 zfs_dirent_unlock(tdl);
3693 
3694                 if (sdzp == tdzp)
3695                         rw_exit(&sdzp->z_name_lock);
3696 
3697                 VN_RELE(ZTOV(szp));
3698                 if (tzp)
3699                         VN_RELE(ZTOV(tzp));
3700                 if (error == ERESTART) {
3701                         waited = B_TRUE;
3702                         dmu_tx_wait(tx);
3703                         dmu_tx_abort(tx);
3704                         goto top;
3705                 }
3706                 dmu_tx_abort(tx);
3707                 ZFS_EXIT(zfsvfs);
3708                 return (error);
3709         }
3710 
3711         if (tzp)        /* Attempt to remove the existing target */
3712                 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3713 
3714         if (error == 0) {
3715                 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3716                 if (error == 0) {
3717                         szp->z_pflags |= ZFS_AV_MODIFIED;
3718 
3719                         error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3720                             (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3721                         ASSERT0(error);
3722 
3723                         error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3724                         if (error == 0) {
3725                                 zfs_log_rename(zilog, tx, TX_RENAME |
3726                                     (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3727                                     sdl->dl_name, tdzp, tdl->dl_name, szp);
3728 
3729                                 /*
3730                                  * Update path information for the target vnode
3731                                  */
3732                                 vn_renamepath(tdvp, ZTOV(szp), tnm,
3733                                     strlen(tnm));
3734                         } else {
3735                                 /*
3736                                  * At this point, we have successfully created
3737                                  * the target name, but have failed to remove
3738                                  * the source name.  Since the create was done
3739                                  * with the ZRENAMING flag, there are
3740                                  * complications; for one, the link count is
3741                                  * wrong.  The easiest way to deal with this
3742                                  * is to remove the newly created target, and
3743                                  * return the original error.  This must
3744                                  * succeed; fortunately, it is very unlikely to
3745                                  * fail, since we just created it.
3746                                  */
3747                                 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3748                                     ZRENAMING, NULL), ==, 0);
3749                         }
3750                 }
3751         }
3752 
3753         dmu_tx_commit(tx);
3754 out:
3755         if (zl != NULL)
3756                 zfs_rename_unlock(&zl);
3757 
3758         zfs_dirent_unlock(sdl);
3759         zfs_dirent_unlock(tdl);
3760 
3761         if (sdzp == tdzp)
3762                 rw_exit(&sdzp->z_name_lock);
3763 
3764 
3765         VN_RELE(ZTOV(szp));
3766         if (tzp)
3767                 VN_RELE(ZTOV(tzp));
3768 
3769         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3770                 zil_commit(zilog, 0);
3771 
3772         ZFS_EXIT(zfsvfs);
3773         return (error);
3774 }
3775 
3776 /*
3777  * Insert the indicated symbolic reference entry into the directory.
3778  *
3779  *      IN:     dvp     - Directory to contain new symbolic link.
3780  *              link    - Name for new symlink entry.
3781  *              vap     - Attributes of new entry.
3782  *              cr      - credentials of caller.
3783  *              ct      - caller context
3784  *              flags   - case flags
3785  *
3786  *      RETURN: 0 on success, error code on failure.
3787  *
3788  * Timestamps:
3789  *      dvp - ctime|mtime updated
3790  */
3791 /*ARGSUSED*/
3792 static int
3793 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3794     caller_context_t *ct, int flags)
3795 {
3796         znode_t         *zp, *dzp = VTOZ(dvp);
3797         zfs_dirlock_t   *dl;
3798         dmu_tx_t        *tx;
3799         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
3800         zilog_t         *zilog;
3801         uint64_t        len = strlen(link);
3802         int             error;
3803         int             zflg = ZNEW;
3804         zfs_acl_ids_t   acl_ids;
3805         boolean_t       fuid_dirtied;
3806         uint64_t        txtype = TX_SYMLINK;
3807         boolean_t       waited = B_FALSE;
3808 
3809         ASSERT(vap->va_type == VLNK);
3810 
3811         ZFS_ENTER(zfsvfs);
3812         ZFS_VERIFY_ZP(dzp);
3813         zilog = zfsvfs->z_log;
3814 
3815         if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3816             NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3817                 ZFS_EXIT(zfsvfs);
3818                 return (SET_ERROR(EILSEQ));
3819         }
3820         if (flags & FIGNORECASE)
3821                 zflg |= ZCILOOK;
3822 
3823         if (len > MAXPATHLEN) {
3824                 ZFS_EXIT(zfsvfs);
3825                 return (SET_ERROR(ENAMETOOLONG));
3826         }
3827 
3828         if ((error = zfs_acl_ids_create(dzp, 0,
3829             vap, cr, NULL, &acl_ids)) != 0) {
3830                 ZFS_EXIT(zfsvfs);
3831                 return (error);
3832         }
3833 top:
3834         /*
3835          * Attempt to lock directory; fail if entry already exists.
3836          */
3837         error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3838         if (error) {
3839                 zfs_acl_ids_free(&acl_ids);
3840                 ZFS_EXIT(zfsvfs);
3841                 return (error);
3842         }
3843 
3844         if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3845                 zfs_acl_ids_free(&acl_ids);
3846                 zfs_dirent_unlock(dl);
3847                 ZFS_EXIT(zfsvfs);
3848                 return (error);
3849         }
3850 
3851         if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3852                 zfs_acl_ids_free(&acl_ids);
3853                 zfs_dirent_unlock(dl);
3854                 ZFS_EXIT(zfsvfs);
3855                 return (SET_ERROR(EDQUOT));
3856         }
3857         tx = dmu_tx_create(zfsvfs->z_os);
3858         fuid_dirtied = zfsvfs->z_fuid_dirty;
3859         dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3860         dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3861         dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3862             ZFS_SA_BASE_ATTR_SIZE + len);
3863         dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3864         if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3865                 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3866                     acl_ids.z_aclp->z_acl_bytes);
3867         }
3868         if (fuid_dirtied)
3869                 zfs_fuid_txhold(zfsvfs, tx);
3870         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3871         if (error) {
3872                 zfs_dirent_unlock(dl);
3873                 if (error == ERESTART) {
3874                         waited = B_TRUE;
3875                         dmu_tx_wait(tx);
3876                         dmu_tx_abort(tx);
3877                         goto top;
3878                 }
3879                 zfs_acl_ids_free(&acl_ids);
3880                 dmu_tx_abort(tx);
3881                 ZFS_EXIT(zfsvfs);
3882                 return (error);
3883         }
3884 
3885         /*
3886          * Create a new object for the symlink.
3887          * for version 4 ZPL datsets the symlink will be an SA attribute
3888          */
3889         zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3890 
3891         if (fuid_dirtied)
3892                 zfs_fuid_sync(zfsvfs, tx);
3893 
3894         mutex_enter(&zp->z_lock);
3895         if (zp->z_is_sa)
3896                 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3897                     link, len, tx);
3898         else
3899                 zfs_sa_symlink(zp, link, len, tx);
3900         mutex_exit(&zp->z_lock);
3901 
3902         zp->z_size = len;
3903         (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3904             &zp->z_size, sizeof (zp->z_size), tx);
3905         /*
3906          * Insert the new object into the directory.
3907          */
3908         (void) zfs_link_create(dl, zp, tx, ZNEW);
3909 
3910         if (flags & FIGNORECASE)
3911                 txtype |= TX_CI;
3912         zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3913 
3914         zfs_acl_ids_free(&acl_ids);
3915 
3916         dmu_tx_commit(tx);
3917 
3918         zfs_dirent_unlock(dl);
3919 
3920         VN_RELE(ZTOV(zp));
3921 
3922         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3923                 zil_commit(zilog, 0);
3924 
3925         ZFS_EXIT(zfsvfs);
3926         return (error);
3927 }
3928 
3929 /*
3930  * Return, in the buffer contained in the provided uio structure,
3931  * the symbolic path referred to by vp.
3932  *
3933  *      IN:     vp      - vnode of symbolic link.
3934  *              uio     - structure to contain the link path.
3935  *              cr      - credentials of caller.
3936  *              ct      - caller context
3937  *
3938  *      OUT:    uio     - structure containing the link path.
3939  *
3940  *      RETURN: 0 on success, error code on failure.
3941  *
3942  * Timestamps:
3943  *      vp - atime updated
3944  */
3945 /* ARGSUSED */
3946 static int
3947 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3948 {
3949         znode_t         *zp = VTOZ(vp);
3950         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
3951         int             error;
3952 
3953         ZFS_ENTER(zfsvfs);
3954         ZFS_VERIFY_ZP(zp);
3955 
3956         mutex_enter(&zp->z_lock);
3957         if (zp->z_is_sa)
3958                 error = sa_lookup_uio(zp->z_sa_hdl,
3959                     SA_ZPL_SYMLINK(zfsvfs), uio);
3960         else
3961                 error = zfs_sa_readlink(zp, uio);
3962         mutex_exit(&zp->z_lock);
3963 
3964         ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3965 
3966         ZFS_EXIT(zfsvfs);
3967         return (error);
3968 }
3969 
3970 /*
3971  * Insert a new entry into directory tdvp referencing svp.
3972  *
3973  *      IN:     tdvp    - Directory to contain new entry.
3974  *              svp     - vnode of new entry.
3975  *              name    - name of new entry.
3976  *              cr      - credentials of caller.
3977  *              ct      - caller context
3978  *
3979  *      RETURN: 0 on success, error code on failure.
3980  *
3981  * Timestamps:
3982  *      tdvp - ctime|mtime updated
3983  *       svp - ctime updated
3984  */
3985 /* ARGSUSED */
3986 static int
3987 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3988     caller_context_t *ct, int flags)
3989 {
3990         znode_t         *dzp = VTOZ(tdvp);
3991         znode_t         *tzp, *szp;
3992         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
3993         zilog_t         *zilog;
3994         zfs_dirlock_t   *dl;
3995         dmu_tx_t        *tx;
3996         vnode_t         *realvp;
3997         int             error;
3998         int             zf = ZNEW;
3999         uint64_t        parent;
4000         uid_t           owner;
4001         boolean_t       waited = B_FALSE;
4002 
4003         ASSERT(tdvp->v_type == VDIR);
4004 
4005         ZFS_ENTER(zfsvfs);
4006         ZFS_VERIFY_ZP(dzp);
4007         zilog = zfsvfs->z_log;
4008 
4009         if (VOP_REALVP(svp, &realvp, ct) == 0)
4010                 svp = realvp;
4011 
4012         /*
4013          * POSIX dictates that we return EPERM here.
4014          * Better choices include ENOTSUP or EISDIR.
4015          */
4016         if (svp->v_type == VDIR) {
4017                 ZFS_EXIT(zfsvfs);
4018                 return (SET_ERROR(EPERM));
4019         }
4020 
4021         szp = VTOZ(svp);
4022         ZFS_VERIFY_ZP(szp);
4023 
4024         /*
4025          * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4026          * ctldir appear to have the same v_vfsp.
4027          */
4028         if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4029                 ZFS_EXIT(zfsvfs);
4030                 return (SET_ERROR(EXDEV));
4031         }
4032 
4033         /* Prevent links to .zfs/shares files */
4034 
4035         if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4036             &parent, sizeof (uint64_t))) != 0) {
4037                 ZFS_EXIT(zfsvfs);
4038                 return (error);
4039         }
4040         if (parent == zfsvfs->z_shares_dir) {
4041                 ZFS_EXIT(zfsvfs);
4042                 return (SET_ERROR(EPERM));
4043         }
4044 
4045         if (zfsvfs->z_utf8 && u8_validate(name,
4046             strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4047                 ZFS_EXIT(zfsvfs);
4048                 return (SET_ERROR(EILSEQ));
4049         }
4050         if (flags & FIGNORECASE)
4051                 zf |= ZCILOOK;
4052 
4053         /*
4054          * We do not support links between attributes and non-attributes
4055          * because of the potential security risk of creating links
4056          * into "normal" file space in order to circumvent restrictions
4057          * imposed in attribute space.
4058          */
4059         if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4060                 ZFS_EXIT(zfsvfs);
4061                 return (SET_ERROR(EINVAL));
4062         }
4063 
4064 
4065         owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4066         if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4067                 ZFS_EXIT(zfsvfs);
4068                 return (SET_ERROR(EPERM));
4069         }
4070 
4071         if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4072                 ZFS_EXIT(zfsvfs);
4073                 return (error);
4074         }
4075 
4076 top:
4077         /*
4078          * Attempt to lock directory; fail if entry already exists.
4079          */
4080         error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4081         if (error) {
4082                 ZFS_EXIT(zfsvfs);
4083                 return (error);
4084         }
4085 
4086         tx = dmu_tx_create(zfsvfs->z_os);
4087         dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4088         dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4089         zfs_sa_upgrade_txholds(tx, szp);
4090         zfs_sa_upgrade_txholds(tx, dzp);
4091         error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4092         if (error) {
4093                 zfs_dirent_unlock(dl);
4094                 if (error == ERESTART) {
4095                         waited = B_TRUE;
4096                         dmu_tx_wait(tx);
4097                         dmu_tx_abort(tx);
4098                         goto top;
4099                 }
4100                 dmu_tx_abort(tx);
4101                 ZFS_EXIT(zfsvfs);
4102                 return (error);
4103         }
4104 
4105         error = zfs_link_create(dl, szp, tx, 0);
4106 
4107         if (error == 0) {
4108                 uint64_t txtype = TX_LINK;
4109                 if (flags & FIGNORECASE)
4110                         txtype |= TX_CI;
4111                 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4112         }
4113 
4114         dmu_tx_commit(tx);
4115 
4116         zfs_dirent_unlock(dl);
4117 
4118         if (error == 0) {
4119                 vnevent_link(svp, ct);
4120         }
4121 
4122         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4123                 zil_commit(zilog, 0);
4124 
4125         ZFS_EXIT(zfsvfs);
4126         return (error);
4127 }
4128 
4129 /*
4130  * zfs_null_putapage() is used when the file system has been force
4131  * unmounted. It just drops the pages.
4132  */
4133 /* ARGSUSED */
4134 static int
4135 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4136                 size_t *lenp, int flags, cred_t *cr)
4137 {
4138         pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4139         return (0);
4140 }
4141 
4142 /*
4143  * Push a page out to disk, klustering if possible.
4144  *
4145  *      IN:     vp      - file to push page to.
4146  *              pp      - page to push.
4147  *              flags   - additional flags.
4148  *              cr      - credentials of caller.
4149  *
4150  *      OUT:    offp    - start of range pushed.
4151  *              lenp    - len of range pushed.
4152  *
4153  *      RETURN: 0 on success, error code on failure.
4154  *
4155  * NOTE: callers must have locked the page to be pushed.  On
4156  * exit, the page (and all other pages in the kluster) must be
4157  * unlocked.
4158  */
4159 /* ARGSUSED */
4160 static int
4161 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4162                 size_t *lenp, int flags, cred_t *cr)
4163 {
4164         znode_t         *zp = VTOZ(vp);
4165         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4166         dmu_tx_t        *tx;
4167         u_offset_t      off, koff;
4168         size_t          len, klen;
4169         int             err;
4170 
4171         off = pp->p_offset;
4172         len = PAGESIZE;
4173         /*
4174          * If our blocksize is bigger than the page size, try to kluster
4175          * multiple pages so that we write a full block (thus avoiding
4176          * a read-modify-write).
4177          */
4178         if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4179                 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4180                 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4181                 ASSERT(koff <= zp->z_size);
4182                 if (koff + klen > zp->z_size)
4183                         klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4184                 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4185         }
4186         ASSERT3U(btop(len), ==, btopr(len));
4187 
4188         /*
4189          * Can't push pages past end-of-file.
4190          */
4191         if (off >= zp->z_size) {
4192                 /* ignore all pages */
4193                 err = 0;
4194                 goto out;
4195         } else if (off + len > zp->z_size) {
4196                 int npages = btopr(zp->z_size - off);
4197                 page_t *trunc;
4198 
4199                 page_list_break(&pp, &trunc, npages);
4200                 /* ignore pages past end of file */
4201                 if (trunc)
4202                         pvn_write_done(trunc, flags);
4203                 len = zp->z_size - off;
4204         }
4205 
4206         if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4207             zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4208                 err = SET_ERROR(EDQUOT);
4209                 goto out;
4210         }
4211         tx = dmu_tx_create(zfsvfs->z_os);
4212         dmu_tx_hold_write(tx, zp->z_id, off, len);
4213 
4214         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4215         zfs_sa_upgrade_txholds(tx, zp);
4216         err = dmu_tx_assign(tx, TXG_WAIT);
4217         if (err != 0) {
4218                 dmu_tx_abort(tx);
4219                 goto out;
4220         }
4221 
4222         if (zp->z_blksz <= PAGESIZE) {
4223                 caddr_t va = zfs_map_page(pp, S_READ);
4224                 ASSERT3U(len, <=, PAGESIZE);
4225                 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4226                 zfs_unmap_page(pp, va);
4227         } else {
4228                 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4229         }
4230 
4231         if (err == 0) {
4232                 uint64_t mtime[2], ctime[2];
4233                 sa_bulk_attr_t bulk[3];
4234                 int count = 0;
4235 
4236                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4237                     &mtime, 16);
4238                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4239                     &ctime, 16);
4240                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4241                     &zp->z_pflags, 8);
4242                 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4243                     B_TRUE);
4244                 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4245         }
4246         dmu_tx_commit(tx);
4247 
4248 out:
4249         pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4250         if (offp)
4251                 *offp = off;
4252         if (lenp)
4253                 *lenp = len;
4254 
4255         return (err);
4256 }
4257 
4258 /*
4259  * Copy the portion of the file indicated from pages into the file.
4260  * The pages are stored in a page list attached to the files vnode.
4261  *
4262  *      IN:     vp      - vnode of file to push page data to.
4263  *              off     - position in file to put data.
4264  *              len     - amount of data to write.
4265  *              flags   - flags to control the operation.
4266  *              cr      - credentials of caller.
4267  *              ct      - caller context.
4268  *
4269  *      RETURN: 0 on success, error code on failure.
4270  *
4271  * Timestamps:
4272  *      vp - ctime|mtime updated
4273  */
4274 /*ARGSUSED*/
4275 static int
4276 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4277     caller_context_t *ct)
4278 {
4279         znode_t         *zp = VTOZ(vp);
4280         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4281         page_t          *pp;
4282         size_t          io_len;
4283         u_offset_t      io_off;
4284         uint_t          blksz;
4285         rl_t            *rl;
4286         int             error = 0;
4287 
4288         ZFS_ENTER(zfsvfs);
4289         ZFS_VERIFY_ZP(zp);
4290 
4291         /*
4292          * There's nothing to do if no data is cached.
4293          */
4294         if (!vn_has_cached_data(vp)) {
4295                 ZFS_EXIT(zfsvfs);
4296                 return (0);
4297         }
4298 
4299         /*
4300          * Align this request to the file block size in case we kluster.
4301          * XXX - this can result in pretty aggresive locking, which can
4302          * impact simultanious read/write access.  One option might be
4303          * to break up long requests (len == 0) into block-by-block
4304          * operations to get narrower locking.
4305          */
4306         blksz = zp->z_blksz;
4307         if (ISP2(blksz))
4308                 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4309         else
4310                 io_off = 0;
4311         if (len > 0 && ISP2(blksz))
4312                 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4313         else
4314                 io_len = 0;
4315 
4316         if (io_len == 0) {
4317                 /*
4318                  * Search the entire vp list for pages >= io_off.
4319                  */
4320                 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4321                 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4322                 goto out;
4323         }
4324         rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4325 
4326         if (off > zp->z_size) {
4327                 /* past end of file */
4328                 zfs_range_unlock(rl);
4329                 ZFS_EXIT(zfsvfs);
4330                 return (0);
4331         }
4332 
4333         len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4334 
4335         for (off = io_off; io_off < off + len; io_off += io_len) {
4336                 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4337                         pp = page_lookup(vp, io_off,
4338                             (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4339                 } else {
4340                         pp = page_lookup_nowait(vp, io_off,
4341                             (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4342                 }
4343 
4344                 if (pp != NULL && pvn_getdirty(pp, flags)) {
4345                         int err;
4346 
4347                         /*
4348                          * Found a dirty page to push
4349                          */
4350                         err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4351                         if (err)
4352                                 error = err;
4353                 } else {
4354                         io_len = PAGESIZE;
4355                 }
4356         }
4357 out:
4358         zfs_range_unlock(rl);
4359         if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4360                 zil_commit(zfsvfs->z_log, zp->z_id);
4361         ZFS_EXIT(zfsvfs);
4362         return (error);
4363 }
4364 
4365 /*ARGSUSED*/
4366 void
4367 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4368 {
4369         znode_t *zp = VTOZ(vp);
4370         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4371         int error;
4372 
4373         rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4374         if (zp->z_sa_hdl == NULL) {
4375                 /*
4376                  * The fs has been unmounted, or we did a
4377                  * suspend/resume and this file no longer exists.
4378                  */
4379                 if (vn_has_cached_data(vp)) {
4380                         (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4381                             B_INVAL, cr);
4382                 }
4383 
4384                 mutex_enter(&zp->z_lock);
4385                 mutex_enter(&vp->v_lock);
4386                 ASSERT(vp->v_count == 1);
4387                 vp->v_count = 0;
4388                 mutex_exit(&vp->v_lock);
4389                 mutex_exit(&zp->z_lock);
4390                 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4391                 zfs_znode_free(zp);
4392                 return;
4393         }
4394 
4395         /*
4396          * Attempt to push any data in the page cache.  If this fails
4397          * we will get kicked out later in zfs_zinactive().
4398          */
4399         if (vn_has_cached_data(vp)) {
4400                 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4401                     cr);
4402         }
4403 
4404         if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4405                 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4406 
4407                 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4408                 zfs_sa_upgrade_txholds(tx, zp);
4409                 error = dmu_tx_assign(tx, TXG_WAIT);
4410                 if (error) {
4411                         dmu_tx_abort(tx);
4412                 } else {
4413                         mutex_enter(&zp->z_lock);
4414                         (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4415                             (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4416                         zp->z_atime_dirty = 0;
4417                         mutex_exit(&zp->z_lock);
4418                         dmu_tx_commit(tx);
4419                 }
4420         }
4421 
4422         zfs_zinactive(zp);
4423         rw_exit(&zfsvfs->z_teardown_inactive_lock);
4424 }
4425 
4426 /*
4427  * Bounds-check the seek operation.
4428  *
4429  *      IN:     vp      - vnode seeking within
4430  *              ooff    - old file offset
4431  *              noffp   - pointer to new file offset
4432  *              ct      - caller context
4433  *
4434  *      RETURN: 0 on success, EINVAL if new offset invalid.
4435  */
4436 /* ARGSUSED */
4437 static int
4438 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4439     caller_context_t *ct)
4440 {
4441         if (vp->v_type == VDIR)
4442                 return (0);
4443         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4444 }
4445 
4446 /*
4447  * Pre-filter the generic locking function to trap attempts to place
4448  * a mandatory lock on a memory mapped file.
4449  */
4450 static int
4451 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4452     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4453 {
4454         znode_t *zp = VTOZ(vp);
4455         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4456 
4457         ZFS_ENTER(zfsvfs);
4458         ZFS_VERIFY_ZP(zp);
4459 
4460         /*
4461          * We are following the UFS semantics with respect to mapcnt
4462          * here: If we see that the file is mapped already, then we will
4463          * return an error, but we don't worry about races between this
4464          * function and zfs_map().
4465          */
4466         if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4467                 ZFS_EXIT(zfsvfs);
4468                 return (SET_ERROR(EAGAIN));
4469         }
4470         ZFS_EXIT(zfsvfs);
4471         return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4472 }
4473 
4474 /*
4475  * If we can't find a page in the cache, we will create a new page
4476  * and fill it with file data.  For efficiency, we may try to fill
4477  * multiple pages at once (klustering) to fill up the supplied page
4478  * list.  Note that the pages to be filled are held with an exclusive
4479  * lock to prevent access by other threads while they are being filled.
4480  */
4481 static int
4482 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4483     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4484 {
4485         znode_t *zp = VTOZ(vp);
4486         page_t *pp, *cur_pp;
4487         objset_t *os = zp->z_zfsvfs->z_os;
4488         u_offset_t io_off, total;
4489         size_t io_len;
4490         int err;
4491 
4492         if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4493                 /*
4494                  * We only have a single page, don't bother klustering
4495                  */
4496                 io_off = off;
4497                 io_len = PAGESIZE;
4498                 pp = page_create_va(vp, io_off, io_len,
4499                     PG_EXCL | PG_WAIT, seg, addr);
4500         } else {
4501                 /*
4502                  * Try to find enough pages to fill the page list
4503                  */
4504                 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4505                     &io_len, off, plsz, 0);
4506         }
4507         if (pp == NULL) {
4508                 /*
4509                  * The page already exists, nothing to do here.
4510                  */
4511                 *pl = NULL;
4512                 return (0);
4513         }
4514 
4515         /*
4516          * Fill the pages in the kluster.
4517          */
4518         cur_pp = pp;
4519         for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4520                 caddr_t va;
4521 
4522                 ASSERT3U(io_off, ==, cur_pp->p_offset);
4523                 va = zfs_map_page(cur_pp, S_WRITE);
4524                 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4525                     DMU_READ_PREFETCH);
4526                 zfs_unmap_page(cur_pp, va);
4527                 if (err) {
4528                         /* On error, toss the entire kluster */
4529                         pvn_read_done(pp, B_ERROR);
4530                         /* convert checksum errors into IO errors */
4531                         if (err == ECKSUM)
4532                                 err = SET_ERROR(EIO);
4533                         return (err);
4534                 }
4535                 cur_pp = cur_pp->p_next;
4536         }
4537 
4538         /*
4539          * Fill in the page list array from the kluster starting
4540          * from the desired offset `off'.
4541          * NOTE: the page list will always be null terminated.
4542          */
4543         pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4544         ASSERT(pl == NULL || (*pl)->p_offset == off);
4545 
4546         return (0);
4547 }
4548 
4549 /*
4550  * Return pointers to the pages for the file region [off, off + len]
4551  * in the pl array.  If plsz is greater than len, this function may
4552  * also return page pointers from after the specified region
4553  * (i.e. the region [off, off + plsz]).  These additional pages are
4554  * only returned if they are already in the cache, or were created as
4555  * part of a klustered read.
4556  *
4557  *      IN:     vp      - vnode of file to get data from.
4558  *              off     - position in file to get data from.
4559  *              len     - amount of data to retrieve.
4560  *              plsz    - length of provided page list.
4561  *              seg     - segment to obtain pages for.
4562  *              addr    - virtual address of fault.
4563  *              rw      - mode of created pages.
4564  *              cr      - credentials of caller.
4565  *              ct      - caller context.
4566  *
4567  *      OUT:    protp   - protection mode of created pages.
4568  *              pl      - list of pages created.
4569  *
4570  *      RETURN: 0 on success, error code on failure.
4571  *
4572  * Timestamps:
4573  *      vp - atime updated
4574  */
4575 /* ARGSUSED */
4576 static int
4577 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4578     page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4579     enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4580 {
4581         znode_t         *zp = VTOZ(vp);
4582         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4583         page_t          **pl0 = pl;
4584         int             err = 0;
4585 
4586         /* we do our own caching, faultahead is unnecessary */
4587         if (pl == NULL)
4588                 return (0);
4589         else if (len > plsz)
4590                 len = plsz;
4591         else
4592                 len = P2ROUNDUP(len, PAGESIZE);
4593         ASSERT(plsz >= len);
4594 
4595         ZFS_ENTER(zfsvfs);
4596         ZFS_VERIFY_ZP(zp);
4597 
4598         if (protp)
4599                 *protp = PROT_ALL;
4600 
4601         /*
4602          * Loop through the requested range [off, off + len) looking
4603          * for pages.  If we don't find a page, we will need to create
4604          * a new page and fill it with data from the file.
4605          */
4606         while (len > 0) {
4607                 if (*pl = page_lookup(vp, off, SE_SHARED))
4608                         *(pl+1) = NULL;
4609                 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4610                         goto out;
4611                 while (*pl) {
4612                         ASSERT3U((*pl)->p_offset, ==, off);
4613                         off += PAGESIZE;
4614                         addr += PAGESIZE;
4615                         if (len > 0) {
4616                                 ASSERT3U(len, >=, PAGESIZE);
4617                                 len -= PAGESIZE;
4618                         }
4619                         ASSERT3U(plsz, >=, PAGESIZE);
4620                         plsz -= PAGESIZE;
4621                         pl++;
4622                 }
4623         }
4624 
4625         /*
4626          * Fill out the page array with any pages already in the cache.
4627          */
4628         while (plsz > 0 &&
4629             (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4630                         off += PAGESIZE;
4631                         plsz -= PAGESIZE;
4632         }
4633 out:
4634         if (err) {
4635                 /*
4636                  * Release any pages we have previously locked.
4637                  */
4638                 while (pl > pl0)
4639                         page_unlock(*--pl);
4640         } else {
4641                 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4642         }
4643 
4644         *pl = NULL;
4645 
4646         ZFS_EXIT(zfsvfs);
4647         return (err);
4648 }
4649 
4650 /*
4651  * Request a memory map for a section of a file.  This code interacts
4652  * with common code and the VM system as follows:
4653  *
4654  * - common code calls mmap(), which ends up in smmap_common()
4655  * - this calls VOP_MAP(), which takes you into (say) zfs
4656  * - zfs_map() calls as_map(), passing segvn_create() as the callback
4657  * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4658  * - zfs_addmap() updates z_mapcnt
4659  */
4660 /*ARGSUSED*/
4661 static int
4662 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4663     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4664     caller_context_t *ct)
4665 {
4666         znode_t *zp = VTOZ(vp);
4667         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4668         segvn_crargs_t  vn_a;
4669         int             error;
4670 
4671         ZFS_ENTER(zfsvfs);
4672         ZFS_VERIFY_ZP(zp);
4673 
4674         if ((prot & PROT_WRITE) && (zp->z_pflags &
4675             (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4676                 ZFS_EXIT(zfsvfs);
4677                 return (SET_ERROR(EPERM));
4678         }
4679 
4680         if ((prot & (PROT_READ | PROT_EXEC)) &&
4681             (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4682                 ZFS_EXIT(zfsvfs);
4683                 return (SET_ERROR(EACCES));
4684         }
4685 
4686         if (vp->v_flag & VNOMAP) {
4687                 ZFS_EXIT(zfsvfs);
4688                 return (SET_ERROR(ENOSYS));
4689         }
4690 
4691         if (off < 0 || len > MAXOFFSET_T - off) {
4692                 ZFS_EXIT(zfsvfs);
4693                 return (SET_ERROR(ENXIO));
4694         }
4695 
4696         if (vp->v_type != VREG) {
4697                 ZFS_EXIT(zfsvfs);
4698                 return (SET_ERROR(ENODEV));
4699         }
4700 
4701         /*
4702          * If file is locked, disallow mapping.
4703          */
4704         if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4705                 ZFS_EXIT(zfsvfs);
4706                 return (SET_ERROR(EAGAIN));
4707         }
4708 
4709         as_rangelock(as);
4710         error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4711         if (error != 0) {
4712                 as_rangeunlock(as);
4713                 ZFS_EXIT(zfsvfs);
4714                 return (error);
4715         }
4716 
4717         vn_a.vp = vp;
4718         vn_a.offset = (u_offset_t)off;
4719         vn_a.type = flags & MAP_TYPE;
4720         vn_a.prot = prot;
4721         vn_a.maxprot = maxprot;
4722         vn_a.cred = cr;
4723         vn_a.amp = NULL;
4724         vn_a.flags = flags & ~MAP_TYPE;
4725         vn_a.szc = 0;
4726         vn_a.lgrp_mem_policy_flags = 0;
4727 
4728         error = as_map(as, *addrp, len, segvn_create, &vn_a);
4729 
4730         as_rangeunlock(as);
4731         ZFS_EXIT(zfsvfs);
4732         return (error);
4733 }
4734 
4735 /* ARGSUSED */
4736 static int
4737 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4738     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4739     caller_context_t *ct)
4740 {
4741         uint64_t pages = btopr(len);
4742 
4743         atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4744         return (0);
4745 }
4746 
4747 /*
4748  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4749  * more accurate mtime for the associated file.  Since we don't have a way of
4750  * detecting when the data was actually modified, we have to resort to
4751  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4752  * last page is pushed.  The problem occurs when the msync() call is omitted,
4753  * which by far the most common case:
4754  *
4755  *      open()
4756  *      mmap()
4757  *      <modify memory>
4758  *      munmap()
4759  *      close()
4760  *      <time lapse>
4761  *      putpage() via fsflush
4762  *
4763  * If we wait until fsflush to come along, we can have a modification time that
4764  * is some arbitrary point in the future.  In order to prevent this in the
4765  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4766  * torn down.
4767  */
4768 /* ARGSUSED */
4769 static int
4770 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4771     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4772     caller_context_t *ct)
4773 {
4774         uint64_t pages = btopr(len);
4775 
4776         ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4777         atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4778 
4779         if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4780             vn_has_cached_data(vp))
4781                 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4782 
4783         return (0);
4784 }
4785 
4786 /*
4787  * Free or allocate space in a file.  Currently, this function only
4788  * supports the `F_FREESP' command.  However, this command is somewhat
4789  * misnamed, as its functionality includes the ability to allocate as
4790  * well as free space.
4791  *
4792  *      IN:     vp      - vnode of file to free data in.
4793  *              cmd     - action to take (only F_FREESP supported).
4794  *              bfp     - section of file to free/alloc.
4795  *              flag    - current file open mode flags.
4796  *              offset  - current file offset.
4797  *              cr      - credentials of caller [UNUSED].
4798  *              ct      - caller context.
4799  *
4800  *      RETURN: 0 on success, error code on failure.
4801  *
4802  * Timestamps:
4803  *      vp - ctime|mtime updated
4804  */
4805 /* ARGSUSED */
4806 static int
4807 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4808     offset_t offset, cred_t *cr, caller_context_t *ct)
4809 {
4810         znode_t         *zp = VTOZ(vp);
4811         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4812         uint64_t        off, len;
4813         int             error;
4814 
4815         ZFS_ENTER(zfsvfs);
4816         ZFS_VERIFY_ZP(zp);
4817 
4818         if (cmd != F_FREESP) {
4819                 ZFS_EXIT(zfsvfs);
4820                 return (SET_ERROR(EINVAL));
4821         }
4822 
4823         /*
4824          * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4825          * callers might not be able to detect properly that we are read-only,
4826          * so check it explicitly here.
4827          */
4828         if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
4829                 ZFS_EXIT(zfsvfs);
4830                 return (SET_ERROR(EROFS));
4831         }
4832 
4833         if (error = convoff(vp, bfp, 0, offset)) {
4834                 ZFS_EXIT(zfsvfs);
4835                 return (error);
4836         }
4837 
4838         if (bfp->l_len < 0) {
4839                 ZFS_EXIT(zfsvfs);
4840                 return (SET_ERROR(EINVAL));
4841         }
4842 
4843         off = bfp->l_start;
4844         len = bfp->l_len; /* 0 means from off to end of file */
4845 
4846         error = zfs_freesp(zp, off, len, flag, TRUE);
4847 
4848         if (error == 0 && off == 0 && len == 0)
4849                 vnevent_truncate(ZTOV(zp), ct);
4850 
4851         ZFS_EXIT(zfsvfs);
4852         return (error);
4853 }
4854 
4855 /*ARGSUSED*/
4856 static int
4857 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4858 {
4859         znode_t         *zp = VTOZ(vp);
4860         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
4861         uint32_t        gen;
4862         uint64_t        gen64;
4863         uint64_t        object = zp->z_id;
4864         zfid_short_t    *zfid;
4865         int             size, i, error;
4866 
4867         ZFS_ENTER(zfsvfs);
4868         ZFS_VERIFY_ZP(zp);
4869 
4870         if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4871             &gen64, sizeof (uint64_t))) != 0) {
4872                 ZFS_EXIT(zfsvfs);
4873                 return (error);
4874         }
4875 
4876         gen = (uint32_t)gen64;
4877 
4878         size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4879         if (fidp->fid_len < size) {
4880                 fidp->fid_len = size;
4881                 ZFS_EXIT(zfsvfs);
4882                 return (SET_ERROR(ENOSPC));
4883         }
4884 
4885         zfid = (zfid_short_t *)fidp;
4886 
4887         zfid->zf_len = size;
4888 
4889         for (i = 0; i < sizeof (zfid->zf_object); i++)
4890                 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4891 
4892         /* Must have a non-zero generation number to distinguish from .zfs */
4893         if (gen == 0)
4894                 gen = 1;
4895         for (i = 0; i < sizeof (zfid->zf_gen); i++)
4896                 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4897 
4898         if (size == LONG_FID_LEN) {
4899                 uint64_t        objsetid = dmu_objset_id(zfsvfs->z_os);
4900                 zfid_long_t     *zlfid;
4901 
4902                 zlfid = (zfid_long_t *)fidp;
4903 
4904                 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4905                         zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4906 
4907                 /* XXX - this should be the generation number for the objset */
4908                 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4909                         zlfid->zf_setgen[i] = 0;
4910         }
4911 
4912         ZFS_EXIT(zfsvfs);
4913         return (0);
4914 }
4915 
4916 static int
4917 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4918     caller_context_t *ct)
4919 {
4920         znode_t         *zp, *xzp;
4921         zfsvfs_t        *zfsvfs;
4922         zfs_dirlock_t   *dl;
4923         int             error;
4924 
4925         switch (cmd) {
4926         case _PC_LINK_MAX:
4927                 *valp = ULONG_MAX;
4928                 return (0);
4929 
4930         case _PC_FILESIZEBITS:
4931                 *valp = 64;
4932                 return (0);
4933 
4934         case _PC_XATTR_EXISTS:
4935                 zp = VTOZ(vp);
4936                 zfsvfs = zp->z_zfsvfs;
4937                 ZFS_ENTER(zfsvfs);
4938                 ZFS_VERIFY_ZP(zp);
4939                 *valp = 0;
4940                 error = zfs_dirent_lock(&dl, zp, "", &xzp,
4941                     ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4942                 if (error == 0) {
4943                         zfs_dirent_unlock(dl);
4944                         if (!zfs_dirempty(xzp))
4945                                 *valp = 1;
4946                         VN_RELE(ZTOV(xzp));
4947                 } else if (error == ENOENT) {
4948                         /*
4949                          * If there aren't extended attributes, it's the
4950                          * same as having zero of them.
4951                          */
4952                         error = 0;
4953                 }
4954                 ZFS_EXIT(zfsvfs);
4955                 return (error);
4956 
4957         case _PC_SATTR_ENABLED:
4958         case _PC_SATTR_EXISTS:
4959                 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4960                     (vp->v_type == VREG || vp->v_type == VDIR);
4961                 return (0);
4962 
4963         case _PC_ACCESS_FILTERING:
4964                 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4965                     vp->v_type == VDIR;
4966                 return (0);
4967 
4968         case _PC_ACL_ENABLED:
4969                 *valp = _ACL_ACE_ENABLED;
4970                 return (0);
4971 
4972         case _PC_MIN_HOLE_SIZE:
4973                 *valp = (ulong_t)SPA_MINBLOCKSIZE;
4974                 return (0);
4975 
4976         case _PC_TIMESTAMP_RESOLUTION:
4977                 /* nanosecond timestamp resolution */
4978                 *valp = 1L;
4979                 return (0);
4980 
4981         default:
4982                 return (fs_pathconf(vp, cmd, valp, cr, ct));
4983         }
4984 }
4985 
4986 /*ARGSUSED*/
4987 static int
4988 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4989     caller_context_t *ct)
4990 {
4991         znode_t *zp = VTOZ(vp);
4992         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4993         int error;
4994         boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4995 
4996         ZFS_ENTER(zfsvfs);
4997         ZFS_VERIFY_ZP(zp);
4998         error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4999         ZFS_EXIT(zfsvfs);
5000 
5001         return (error);
5002 }
5003 
5004 /*ARGSUSED*/
5005 static int
5006 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5007     caller_context_t *ct)
5008 {
5009         znode_t *zp = VTOZ(vp);
5010         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5011         int error;
5012         boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5013         zilog_t *zilog = zfsvfs->z_log;
5014 
5015         ZFS_ENTER(zfsvfs);
5016         ZFS_VERIFY_ZP(zp);
5017 
5018         error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5019 
5020         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5021                 zil_commit(zilog, 0);
5022 
5023         ZFS_EXIT(zfsvfs);
5024         return (error);
5025 }
5026 
5027 /*
5028  * The smallest read we may consider to loan out an arcbuf.
5029  * This must be a power of 2.
5030  */
5031 int zcr_blksz_min = (1 << 10);    /* 1K */
5032 /*
5033  * If set to less than the file block size, allow loaning out of an
5034  * arcbuf for a partial block read.  This must be a power of 2.
5035  */
5036 int zcr_blksz_max = (1 << 17);    /* 128K */
5037 
5038 /*ARGSUSED*/
5039 static int
5040 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5041     caller_context_t *ct)
5042 {
5043         znode_t *zp = VTOZ(vp);
5044         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5045         int max_blksz = zfsvfs->z_max_blksz;
5046         uio_t *uio = &xuio->xu_uio;
5047         ssize_t size = uio->uio_resid;
5048         offset_t offset = uio->uio_loffset;
5049         int blksz;
5050         int fullblk, i;
5051         arc_buf_t *abuf;
5052         ssize_t maxsize;
5053         int preamble, postamble;
5054 
5055         if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5056                 return (SET_ERROR(EINVAL));
5057 
5058         ZFS_ENTER(zfsvfs);
5059         ZFS_VERIFY_ZP(zp);
5060         switch (ioflag) {
5061         case UIO_WRITE:
5062                 /*
5063                  * Loan out an arc_buf for write if write size is bigger than
5064                  * max_blksz, and the file's block size is also max_blksz.
5065                  */
5066                 blksz = max_blksz;
5067                 if (size < blksz || zp->z_blksz != blksz) {
5068                         ZFS_EXIT(zfsvfs);
5069                         return (SET_ERROR(EINVAL));
5070                 }
5071                 /*
5072                  * Caller requests buffers for write before knowing where the
5073                  * write offset might be (e.g. NFS TCP write).
5074                  */
5075                 if (offset == -1) {
5076                         preamble = 0;
5077                 } else {
5078                         preamble = P2PHASE(offset, blksz);
5079                         if (preamble) {
5080                                 preamble = blksz - preamble;
5081                                 size -= preamble;
5082                         }
5083                 }
5084 
5085                 postamble = P2PHASE(size, blksz);
5086                 size -= postamble;
5087 
5088                 fullblk = size / blksz;
5089                 (void) dmu_xuio_init(xuio,
5090                     (preamble != 0) + fullblk + (postamble != 0));
5091                 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5092                     int, postamble, int,
5093                     (preamble != 0) + fullblk + (postamble != 0));
5094 
5095                 /*
5096                  * Have to fix iov base/len for partial buffers.  They
5097                  * currently represent full arc_buf's.
5098                  */
5099                 if (preamble) {
5100                         /* data begins in the middle of the arc_buf */
5101                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5102                             blksz);
5103                         ASSERT(abuf);
5104                         (void) dmu_xuio_add(xuio, abuf,
5105                             blksz - preamble, preamble);
5106                 }
5107 
5108                 for (i = 0; i < fullblk; i++) {
5109                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5110                             blksz);
5111                         ASSERT(abuf);
5112                         (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5113                 }
5114 
5115                 if (postamble) {
5116                         /* data ends in the middle of the arc_buf */
5117                         abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5118                             blksz);
5119                         ASSERT(abuf);
5120                         (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5121                 }
5122                 break;
5123         case UIO_READ:
5124                 /*
5125                  * Loan out an arc_buf for read if the read size is larger than
5126                  * the current file block size.  Block alignment is not
5127                  * considered.  Partial arc_buf will be loaned out for read.
5128                  */
5129                 blksz = zp->z_blksz;
5130                 if (blksz < zcr_blksz_min)
5131                         blksz = zcr_blksz_min;
5132                 if (blksz > zcr_blksz_max)
5133                         blksz = zcr_blksz_max;
5134                 /* avoid potential complexity of dealing with it */
5135                 if (blksz > max_blksz) {
5136                         ZFS_EXIT(zfsvfs);
5137                         return (SET_ERROR(EINVAL));
5138                 }
5139 
5140                 maxsize = zp->z_size - uio->uio_loffset;
5141                 if (size > maxsize)
5142                         size = maxsize;
5143 
5144                 if (size < blksz || vn_has_cached_data(vp)) {
5145                         ZFS_EXIT(zfsvfs);
5146                         return (SET_ERROR(EINVAL));
5147                 }
5148                 break;
5149         default:
5150                 ZFS_EXIT(zfsvfs);
5151                 return (SET_ERROR(EINVAL));
5152         }
5153 
5154         uio->uio_extflg = UIO_XUIO;
5155         XUIO_XUZC_RW(xuio) = ioflag;
5156         ZFS_EXIT(zfsvfs);
5157         return (0);
5158 }
5159 
5160 /*ARGSUSED*/
5161 static int
5162 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5163 {
5164         int i;
5165         arc_buf_t *abuf;
5166         int ioflag = XUIO_XUZC_RW(xuio);
5167 
5168         ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5169 
5170         i = dmu_xuio_cnt(xuio);
5171         while (i-- > 0) {
5172                 abuf = dmu_xuio_arcbuf(xuio, i);
5173                 /*
5174                  * if abuf == NULL, it must be a write buffer
5175                  * that has been returned in zfs_write().
5176                  */
5177                 if (abuf)
5178                         dmu_return_arcbuf(abuf);
5179                 ASSERT(abuf || ioflag == UIO_WRITE);
5180         }
5181 
5182         dmu_xuio_fini(xuio);
5183         return (0);
5184 }
5185 
5186 /*
5187  * Predeclare these here so that the compiler assumes that
5188  * this is an "old style" function declaration that does
5189  * not include arguments => we won't get type mismatch errors
5190  * in the initializations that follow.
5191  */
5192 static int zfs_inval();
5193 static int zfs_isdir();
5194 
5195 static int
5196 zfs_inval()
5197 {
5198         return (SET_ERROR(EINVAL));
5199 }
5200 
5201 static int
5202 zfs_isdir()
5203 {
5204         return (SET_ERROR(EISDIR));
5205 }
5206 /*
5207  * Directory vnode operations template
5208  */
5209 vnodeops_t *zfs_dvnodeops;
5210 const fs_operation_def_t zfs_dvnodeops_template[] = {
5211         VOPNAME_OPEN,           { .vop_open = zfs_open },
5212         VOPNAME_CLOSE,          { .vop_close = zfs_close },
5213         VOPNAME_READ,           { .error = zfs_isdir },
5214         VOPNAME_WRITE,          { .error = zfs_isdir },
5215         VOPNAME_IOCTL,          { .vop_ioctl = zfs_ioctl },
5216         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5217         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5218         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5219         VOPNAME_LOOKUP,         { .vop_lookup = zfs_lookup },
5220         VOPNAME_CREATE,         { .vop_create = zfs_create },
5221         VOPNAME_REMOVE,         { .vop_remove = zfs_remove },
5222         VOPNAME_LINK,           { .vop_link = zfs_link },
5223         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5224         VOPNAME_MKDIR,          { .vop_mkdir = zfs_mkdir },
5225         VOPNAME_RMDIR,          { .vop_rmdir = zfs_rmdir },
5226         VOPNAME_READDIR,        { .vop_readdir = zfs_readdir },
5227         VOPNAME_SYMLINK,        { .vop_symlink = zfs_symlink },
5228         VOPNAME_FSYNC,          { .vop_fsync = zfs_fsync },
5229         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5230         VOPNAME_FID,            { .vop_fid = zfs_fid },
5231         VOPNAME_SEEK,           { .vop_seek = zfs_seek },
5232         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5233         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5234         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5235         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5236         NULL,                   NULL
5237 };
5238 
5239 /*
5240  * Regular file vnode operations template
5241  */
5242 vnodeops_t *zfs_fvnodeops;
5243 const fs_operation_def_t zfs_fvnodeops_template[] = {
5244         VOPNAME_OPEN,           { .vop_open = zfs_open },
5245         VOPNAME_CLOSE,          { .vop_close = zfs_close },
5246         VOPNAME_READ,           { .vop_read = zfs_read },
5247         VOPNAME_WRITE,          { .vop_write = zfs_write },
5248         VOPNAME_IOCTL,          { .vop_ioctl = zfs_ioctl },
5249         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5250         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5251         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5252         VOPNAME_LOOKUP,         { .vop_lookup = zfs_lookup },
5253         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5254         VOPNAME_FSYNC,          { .vop_fsync = zfs_fsync },
5255         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5256         VOPNAME_FID,            { .vop_fid = zfs_fid },
5257         VOPNAME_SEEK,           { .vop_seek = zfs_seek },
5258         VOPNAME_FRLOCK,         { .vop_frlock = zfs_frlock },
5259         VOPNAME_SPACE,          { .vop_space = zfs_space },
5260         VOPNAME_GETPAGE,        { .vop_getpage = zfs_getpage },
5261         VOPNAME_PUTPAGE,        { .vop_putpage = zfs_putpage },
5262         VOPNAME_MAP,            { .vop_map = zfs_map },
5263         VOPNAME_ADDMAP,         { .vop_addmap = zfs_addmap },
5264         VOPNAME_DELMAP,         { .vop_delmap = zfs_delmap },
5265         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5266         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5267         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5268         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5269         VOPNAME_REQZCBUF,       { .vop_reqzcbuf = zfs_reqzcbuf },
5270         VOPNAME_RETZCBUF,       { .vop_retzcbuf = zfs_retzcbuf },
5271         NULL,                   NULL
5272 };
5273 
5274 /*
5275  * Symbolic link vnode operations template
5276  */
5277 vnodeops_t *zfs_symvnodeops;
5278 const fs_operation_def_t zfs_symvnodeops_template[] = {
5279         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5280         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5281         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5282         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5283         VOPNAME_READLINK,       { .vop_readlink = zfs_readlink },
5284         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5285         VOPNAME_FID,            { .vop_fid = zfs_fid },
5286         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5287         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5288         NULL,                   NULL
5289 };
5290 
5291 /*
5292  * special share hidden files vnode operations template
5293  */
5294 vnodeops_t *zfs_sharevnodeops;
5295 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5296         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5297         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5298         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5299         VOPNAME_FID,            { .vop_fid = zfs_fid },
5300         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5301         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5302         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5303         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5304         NULL,                   NULL
5305 };
5306 
5307 /*
5308  * Extended attribute directory vnode operations template
5309  *
5310  * This template is identical to the directory vnodes
5311  * operation template except for restricted operations:
5312  *      VOP_MKDIR()
5313  *      VOP_SYMLINK()
5314  *
5315  * Note that there are other restrictions embedded in:
5316  *      zfs_create()    - restrict type to VREG
5317  *      zfs_link()      - no links into/out of attribute space
5318  *      zfs_rename()    - no moves into/out of attribute space
5319  */
5320 vnodeops_t *zfs_xdvnodeops;
5321 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5322         VOPNAME_OPEN,           { .vop_open = zfs_open },
5323         VOPNAME_CLOSE,          { .vop_close = zfs_close },
5324         VOPNAME_IOCTL,          { .vop_ioctl = zfs_ioctl },
5325         VOPNAME_GETATTR,        { .vop_getattr = zfs_getattr },
5326         VOPNAME_SETATTR,        { .vop_setattr = zfs_setattr },
5327         VOPNAME_ACCESS,         { .vop_access = zfs_access },
5328         VOPNAME_LOOKUP,         { .vop_lookup = zfs_lookup },
5329         VOPNAME_CREATE,         { .vop_create = zfs_create },
5330         VOPNAME_REMOVE,         { .vop_remove = zfs_remove },
5331         VOPNAME_LINK,           { .vop_link = zfs_link },
5332         VOPNAME_RENAME,         { .vop_rename = zfs_rename },
5333         VOPNAME_MKDIR,          { .error = zfs_inval },
5334         VOPNAME_RMDIR,          { .vop_rmdir = zfs_rmdir },
5335         VOPNAME_READDIR,        { .vop_readdir = zfs_readdir },
5336         VOPNAME_SYMLINK,        { .error = zfs_inval },
5337         VOPNAME_FSYNC,          { .vop_fsync = zfs_fsync },
5338         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5339         VOPNAME_FID,            { .vop_fid = zfs_fid },
5340         VOPNAME_SEEK,           { .vop_seek = zfs_seek },
5341         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5342         VOPNAME_GETSECATTR,     { .vop_getsecattr = zfs_getsecattr },
5343         VOPNAME_SETSECATTR,     { .vop_setsecattr = zfs_setsecattr },
5344         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
5345         NULL,                   NULL
5346 };
5347 
5348 /*
5349  * Error vnode operations template
5350  */
5351 vnodeops_t *zfs_evnodeops;
5352 const fs_operation_def_t zfs_evnodeops_template[] = {
5353         VOPNAME_INACTIVE,       { .vop_inactive = zfs_inactive },
5354         VOPNAME_PATHCONF,       { .vop_pathconf = zfs_pathconf },
5355         NULL,                   NULL
5356 };