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