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 /*
  23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26 
  27 /*
  28  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  29  */
  30 
  31 #include <sys/types.h>
  32 #include <sys/param.h>
  33 #include <sys/t_lock.h>
  34 #include <sys/systm.h>
  35 #include <sys/sysmacros.h>
  36 #include <sys/user.h>
  37 #include <sys/time.h>
  38 #include <sys/vfs.h>
  39 #include <sys/vfs_opreg.h>
  40 #include <sys/vnode.h>
  41 #include <sys/file.h>
  42 #include <sys/fcntl.h>
  43 #include <sys/flock.h>
  44 #include <sys/kmem.h>
  45 #include <sys/uio.h>
  46 #include <sys/errno.h>
  47 #include <sys/stat.h>
  48 #include <sys/cred.h>
  49 #include <sys/dirent.h>
  50 #include <sys/pathname.h>
  51 #include <sys/vmsystm.h>
  52 #include <sys/fs/tmp.h>
  53 #include <sys/fs/tmpnode.h>
  54 #include <sys/mman.h>
  55 #include <vm/hat.h>
  56 #include <vm/seg_vn.h>
  57 #include <vm/seg_map.h>
  58 #include <vm/seg.h>
  59 #include <vm/anon.h>
  60 #include <vm/as.h>
  61 #include <vm/page.h>
  62 #include <vm/pvn.h>
  63 #include <sys/cmn_err.h>
  64 #include <sys/debug.h>
  65 #include <sys/swap.h>
  66 #include <sys/buf.h>
  67 #include <sys/vm.h>
  68 #include <sys/vtrace.h>
  69 #include <sys/policy.h>
  70 #include <fs/fs_subr.h>
  71 
  72 static int      tmp_getapage(struct vnode *, u_offset_t, size_t, uint_t *,
  73         page_t **, size_t, struct seg *, caddr_t, enum seg_rw, struct cred *);
  74 static int      tmp_putapage(struct vnode *, page_t *, u_offset_t *, size_t *,
  75         int, struct cred *);
  76 
  77 /* ARGSUSED1 */
  78 static int
  79 tmp_open(struct vnode **vpp, int flag, struct cred *cred, caller_context_t *ct)
  80 {
  81         /*
  82          * swapon to a tmpfs file is not supported so access
  83          * is denied on open if VISSWAP is set.
  84          */
  85         if ((*vpp)->v_flag & VISSWAP)
  86                 return (EINVAL);
  87         return (0);
  88 }
  89 
  90 /* ARGSUSED1 */
  91 static int
  92 tmp_close(
  93         struct vnode *vp,
  94         int flag,
  95         int count,
  96         offset_t offset,
  97         struct cred *cred,
  98         caller_context_t *ct)
  99 {
 100         cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
 101         cleanshares(vp, ttoproc(curthread)->p_pid);
 102         return (0);
 103 }
 104 
 105 /*
 106  * wrtmp does the real work of write requests for tmpfs.
 107  */
 108 static int
 109 wrtmp(
 110         struct tmount *tm,
 111         struct tmpnode *tp,
 112         struct uio *uio,
 113         struct cred *cr,
 114         struct caller_context *ct)
 115 {
 116         pgcnt_t pageoffset;     /* offset in pages */
 117         ulong_t segmap_offset;  /* pagesize byte offset into segmap */
 118         caddr_t base;           /* base of segmap */
 119         ssize_t bytes;          /* bytes to uiomove */
 120         pfn_t pagenumber;       /* offset in pages into tmp file */
 121         struct vnode *vp;
 122         int error = 0;
 123         int     pagecreate;     /* == 1 if we allocated a page */
 124         int     newpage;
 125         rlim64_t limit = uio->uio_llimit;
 126         long oresid = uio->uio_resid;
 127         timestruc_t now;
 128 
 129         long tn_size_changed = 0;
 130         long old_tn_size;
 131         long new_tn_size;
 132 
 133         vp = TNTOV(tp);
 134         ASSERT(vp->v_type == VREG);
 135 
 136         TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START,
 137             "tmp_wrtmp_start:vp %p", vp);
 138 
 139         ASSERT(RW_WRITE_HELD(&tp->tn_contents));
 140         ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
 141 
 142         if (MANDLOCK(vp, tp->tn_mode)) {
 143                 rw_exit(&tp->tn_contents);
 144                 /*
 145                  * tmp_getattr ends up being called by chklock
 146                  */
 147                 error = chklock(vp, FWRITE, uio->uio_loffset, uio->uio_resid,
 148                     uio->uio_fmode, ct);
 149                 rw_enter(&tp->tn_contents, RW_WRITER);
 150                 if (error != 0) {
 151                         TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 152                             "tmp_wrtmp_end:vp %p error %d", vp, error);
 153                         return (error);
 154                 }
 155         }
 156 
 157         if (uio->uio_loffset < 0)
 158                 return (EINVAL);
 159 
 160         if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
 161                 limit = MAXOFFSET_T;
 162 
 163         if (uio->uio_loffset >= limit) {
 164                 proc_t *p = ttoproc(curthread);
 165 
 166                 mutex_enter(&p->p_lock);
 167                 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
 168                     p, RCA_UNSAFE_SIGINFO);
 169                 mutex_exit(&p->p_lock);
 170                 return (EFBIG);
 171         }
 172 
 173         if (uio->uio_loffset >= MAXOFF_T) {
 174                 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 175                     "tmp_wrtmp_end:vp %p error %d", vp, EINVAL);
 176                 return (EFBIG);
 177         }
 178 
 179         if (uio->uio_resid == 0) {
 180                 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 181                     "tmp_wrtmp_end:vp %p error %d", vp, 0);
 182                 return (0);
 183         }
 184 
 185         if (limit > MAXOFF_T)
 186                 limit = MAXOFF_T;
 187 
 188         do {
 189                 long    offset;
 190                 long    delta;
 191 
 192                 offset = (long)uio->uio_offset;
 193                 pageoffset = offset & PAGEOFFSET;
 194                 /*
 195                  * A maximum of PAGESIZE bytes of data is transferred
 196                  * each pass through this loop
 197                  */
 198                 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
 199 
 200                 if (offset + bytes >= limit) {
 201                         if (offset >= limit) {
 202                                 error = EFBIG;
 203                                 goto out;
 204                         }
 205                         bytes = limit - offset;
 206                 }
 207                 pagenumber = btop(offset);
 208 
 209                 /*
 210                  * delta is the amount of anonymous memory
 211                  * to reserve for the file.
 212                  * We always reserve in pagesize increments so
 213                  * unless we're extending the file into a new page,
 214                  * we don't need to call tmp_resv.
 215                  */
 216                 delta = offset + bytes -
 217                     P2ROUNDUP_TYPED(tp->tn_size, PAGESIZE, u_offset_t);
 218                 if (delta > 0) {
 219                         pagecreate = 1;
 220                         if (tmp_resv(tm, tp, delta, pagecreate)) {
 221                                 /*
 222                                  * Log file system full in the zone that owns
 223                                  * the tmpfs mount, as well as in the global
 224                                  * zone if necessary.
 225                                  */
 226                                 zcmn_err(tm->tm_vfsp->vfs_zone->zone_id,
 227                                     CE_WARN, "%s: File system full, "
 228                                     "swap space limit exceeded",
 229                                     tm->tm_mntpath);
 230 
 231                                 if (tm->tm_vfsp->vfs_zone->zone_id !=
 232                                     GLOBAL_ZONEID) {
 233 
 234                                         vfs_t *vfs = tm->tm_vfsp;
 235 
 236                                         zcmn_err(GLOBAL_ZONEID,
 237                                             CE_WARN, "%s: File system full, "
 238                                             "swap space limit exceeded",
 239                                             vfs->vfs_vnodecovered->v_path);
 240                                 }
 241                                 error = ENOSPC;
 242                                 break;
 243                         }
 244                         tmpnode_growmap(tp, (ulong_t)offset + bytes);
 245                 }
 246                 /* grow the file to the new length */
 247                 if (offset + bytes > tp->tn_size) {
 248                         tn_size_changed = 1;
 249                         old_tn_size = tp->tn_size;
 250                         /*
 251                          * Postpone updating tp->tn_size until uiomove() is
 252                          * done.
 253                          */
 254                         new_tn_size = offset + bytes;
 255                 }
 256                 if (bytes == PAGESIZE) {
 257                         /*
 258                          * Writing whole page so reading from disk
 259                          * is a waste
 260                          */
 261                         pagecreate = 1;
 262                 } else {
 263                         pagecreate = 0;
 264                 }
 265                 /*
 266                  * If writing past EOF or filling in a hole
 267                  * we need to allocate an anon slot.
 268                  */
 269                 if (anon_get_ptr(tp->tn_anon, pagenumber) == NULL) {
 270                         (void) anon_set_ptr(tp->tn_anon, pagenumber,
 271                             anon_alloc(vp, ptob(pagenumber)), ANON_SLEEP);
 272                         pagecreate = 1;
 273                         tp->tn_nblocks++;
 274                 }
 275 
 276                 /*
 277                  * We have to drop the contents lock to allow the VM
 278                  * system to reacquire it in tmp_getpage()
 279                  */
 280                 rw_exit(&tp->tn_contents);
 281 
 282                 /*
 283                  * Touch the page and fault it in if it is not in core
 284                  * before segmap_getmapflt or vpm_data_copy can lock it.
 285                  * This is to avoid the deadlock if the buffer is mapped
 286                  * to the same file through mmap which we want to write.
 287                  */
 288                 uio_prefaultpages((long)bytes, uio);
 289 
 290                 newpage = 0;
 291                 if (vpm_enable) {
 292                         /*
 293                          * Copy data. If new pages are created, part of
 294                          * the page that is not written will be initizliazed
 295                          * with zeros.
 296                          */
 297                         error = vpm_data_copy(vp, offset, bytes, uio,
 298                             !pagecreate, &newpage, 1, S_WRITE);
 299                 } else {
 300                         /* Get offset within the segmap mapping */
 301                         segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
 302                         base = segmap_getmapflt(segkmap, vp,
 303                             (offset &  MAXBMASK), PAGESIZE, !pagecreate,
 304                             S_WRITE);
 305                 }
 306 
 307 
 308                 if (!vpm_enable && pagecreate) {
 309                         /*
 310                          * segmap_pagecreate() returns 1 if it calls
 311                          * page_create_va() to allocate any pages.
 312                          */
 313                         newpage = segmap_pagecreate(segkmap,
 314                             base + segmap_offset, (size_t)PAGESIZE, 0);
 315                         /*
 316                          * Clear from the beginning of the page to the starting
 317                          * offset of the data.
 318                          */
 319                         if (pageoffset != 0)
 320                                 (void) kzero(base + segmap_offset,
 321                                     (size_t)pageoffset);
 322                 }
 323 
 324                 if (!vpm_enable) {
 325                         error = uiomove(base + segmap_offset + pageoffset,
 326                             (long)bytes, UIO_WRITE, uio);
 327                 }
 328 
 329                 if (!vpm_enable && pagecreate &&
 330                     uio->uio_offset < P2ROUNDUP(offset + bytes, PAGESIZE)) {
 331                         long    zoffset; /* zero from offset into page */
 332                         /*
 333                          * We created pages w/o initializing them completely,
 334                          * thus we need to zero the part that wasn't set up.
 335                          * This happens on most EOF write cases and if
 336                          * we had some sort of error during the uiomove.
 337                          */
 338                         long nmoved;
 339 
 340                         nmoved = uio->uio_offset - offset;
 341                         ASSERT((nmoved + pageoffset) <= PAGESIZE);
 342 
 343                         /*
 344                          * Zero from the end of data in the page to the
 345                          * end of the page.
 346                          */
 347                         if ((zoffset = pageoffset + nmoved) < PAGESIZE)
 348                                 (void) kzero(base + segmap_offset + zoffset,
 349                                     (size_t)PAGESIZE - zoffset);
 350                 }
 351 
 352                 /*
 353                  * Unlock the pages which have been allocated by
 354                  * page_create_va() in segmap_pagecreate()
 355                  */
 356                 if (!vpm_enable && newpage) {
 357                         segmap_pageunlock(segkmap, base + segmap_offset,
 358                             (size_t)PAGESIZE, S_WRITE);
 359                 }
 360 
 361                 if (error) {
 362                         /*
 363                          * If we failed on a write, we must
 364                          * be sure to invalidate any pages that may have
 365                          * been allocated.
 366                          */
 367                         if (vpm_enable) {
 368                                 (void) vpm_sync_pages(vp, offset, PAGESIZE,
 369                                     SM_INVAL);
 370                         } else {
 371                                 (void) segmap_release(segkmap, base, SM_INVAL);
 372                         }
 373                 } else {
 374                         if (vpm_enable) {
 375                                 error = vpm_sync_pages(vp, offset, PAGESIZE,
 376                                     0);
 377                         } else {
 378                                 error = segmap_release(segkmap, base, 0);
 379                         }
 380                 }
 381 
 382                 /*
 383                  * Re-acquire contents lock.
 384                  */
 385                 rw_enter(&tp->tn_contents, RW_WRITER);
 386 
 387                 /*
 388                  * Update tn_size.
 389                  */
 390                 if (tn_size_changed)
 391                         tp->tn_size = new_tn_size;
 392 
 393                 /*
 394                  * If the uiomove failed, fix up tn_size.
 395                  */
 396                 if (error) {
 397                         if (tn_size_changed) {
 398                                 /*
 399                                  * The uiomove failed, and we
 400                                  * allocated blocks,so get rid
 401                                  * of them.
 402                                  */
 403                                 (void) tmpnode_trunc(tm, tp,
 404                                     (ulong_t)old_tn_size);
 405                         }
 406                 } else {
 407                         /*
 408                          * XXX - Can this be out of the loop?
 409                          */
 410                         if ((tp->tn_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) &&
 411                             (tp->tn_mode & (S_ISUID | S_ISGID)) &&
 412                             secpolicy_vnode_setid_retain(cr,
 413                             (tp->tn_mode & S_ISUID) != 0 && tp->tn_uid == 0)) {
 414                                 /*
 415                                  * Clear Set-UID & Set-GID bits on
 416                                  * successful write if not privileged
 417                                  * and at least one of the execute bits
 418                                  * is set.  If we always clear Set-GID,
 419                                  * mandatory file and record locking is
 420                                  * unuseable.
 421                                  */
 422                                 tp->tn_mode &= ~(S_ISUID | S_ISGID);
 423                         }
 424                         gethrestime(&now);
 425                         tp->tn_mtime = now;
 426                         tp->tn_ctime = now;
 427                 }
 428         } while (error == 0 && uio->uio_resid > 0 && bytes != 0);
 429 
 430 out:
 431         /*
 432          * If we've already done a partial-write, terminate
 433          * the write but return no error.
 434          */
 435         if (oresid != uio->uio_resid)
 436                 error = 0;
 437         TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 438             "tmp_wrtmp_end:vp %p error %d", vp, error);
 439         return (error);
 440 }
 441 
 442 /*
 443  * rdtmp does the real work of read requests for tmpfs.
 444  */
 445 static int
 446 rdtmp(
 447         struct tmount *tm,
 448         struct tmpnode *tp,
 449         struct uio *uio,
 450         struct caller_context *ct)
 451 {
 452         ulong_t pageoffset;     /* offset in tmpfs file (uio_offset) */
 453         ulong_t segmap_offset;  /* pagesize byte offset into segmap */
 454         caddr_t base;           /* base of segmap */
 455         ssize_t bytes;          /* bytes to uiomove */
 456         struct vnode *vp;
 457         int error;
 458         long oresid = uio->uio_resid;
 459 
 460 #if defined(lint)
 461         tm = tm;
 462 #endif
 463         vp = TNTOV(tp);
 464 
 465         TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START, "tmp_rdtmp_start:vp %p",
 466             vp);
 467 
 468         ASSERT(RW_LOCK_HELD(&tp->tn_contents));
 469 
 470         if (MANDLOCK(vp, tp->tn_mode)) {
 471                 rw_exit(&tp->tn_contents);
 472                 /*
 473                  * tmp_getattr ends up being called by chklock
 474                  */
 475                 error = chklock(vp, FREAD, uio->uio_loffset, uio->uio_resid,
 476                     uio->uio_fmode, ct);
 477                 rw_enter(&tp->tn_contents, RW_READER);
 478                 if (error != 0) {
 479                         TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 480                             "tmp_rdtmp_end:vp %p error %d", vp, error);
 481                         return (error);
 482                 }
 483         }
 484         ASSERT(tp->tn_type == VREG);
 485 
 486         if (uio->uio_loffset >= MAXOFF_T) {
 487                 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 488                     "tmp_rdtmp_end:vp %p error %d", vp, EINVAL);
 489                 return (0);
 490         }
 491         if (uio->uio_loffset < 0)
 492                 return (EINVAL);
 493         if (uio->uio_resid == 0) {
 494                 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 495                     "tmp_rdtmp_end:vp %p error %d", vp, 0);
 496                 return (0);
 497         }
 498 
 499         vp = TNTOV(tp);
 500 
 501         do {
 502                 long diff;
 503                 long offset;
 504 
 505                 offset = uio->uio_offset;
 506                 pageoffset = offset & PAGEOFFSET;
 507                 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
 508 
 509                 diff = tp->tn_size - offset;
 510 
 511                 if (diff <= 0) {
 512                         error = 0;
 513                         goto out;
 514                 }
 515                 if (diff < bytes)
 516                         bytes = diff;
 517 
 518                 /*
 519                  * We have to drop the contents lock to allow the VM system
 520                  * to reacquire it in tmp_getpage() should the uiomove cause a
 521                  * pagefault.
 522                  */
 523                 rw_exit(&tp->tn_contents);
 524 
 525                 if (vpm_enable) {
 526                         /*
 527                          * Copy data.
 528                          */
 529                         error = vpm_data_copy(vp, offset, bytes, uio, 1, NULL,
 530                             0, S_READ);
 531                 } else {
 532                         segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
 533                         base = segmap_getmapflt(segkmap, vp, offset & MAXBMASK,
 534                             bytes, 1, S_READ);
 535 
 536                         error = uiomove(base + segmap_offset + pageoffset,
 537                             (long)bytes, UIO_READ, uio);
 538                 }
 539 
 540                 if (error) {
 541                         if (vpm_enable) {
 542                                 (void) vpm_sync_pages(vp, offset, PAGESIZE, 0);
 543                         } else {
 544                                 (void) segmap_release(segkmap, base, 0);
 545                         }
 546                 } else {
 547                         if (vpm_enable) {
 548                                 error = vpm_sync_pages(vp, offset, PAGESIZE,
 549                                     0);
 550                         } else {
 551                                 error = segmap_release(segkmap, base, 0);
 552                         }
 553                 }
 554 
 555                 /*
 556                  * Re-acquire contents lock.
 557                  */
 558                 rw_enter(&tp->tn_contents, RW_READER);
 559 
 560         } while (error == 0 && uio->uio_resid > 0);
 561 
 562 out:
 563         gethrestime(&tp->tn_atime);
 564 
 565         /*
 566          * If we've already done a partial read, terminate
 567          * the read but return no error.
 568          */
 569         if (oresid != uio->uio_resid)
 570                 error = 0;
 571 
 572         TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 573             "tmp_rdtmp_end:vp %x error %d", vp, error);
 574         return (error);
 575 }
 576 
 577 /* ARGSUSED2 */
 578 static int
 579 tmp_read(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cred,
 580     struct caller_context *ct)
 581 {
 582         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 583         struct tmount *tm = (struct tmount *)VTOTM(vp);
 584         int error;
 585 
 586         /*
 587          * We don't currently support reading non-regular files
 588          */
 589         if (vp->v_type == VDIR)
 590                 return (EISDIR);
 591         if (vp->v_type != VREG)
 592                 return (EINVAL);
 593         /*
 594          * tmp_rwlock should have already been called from layers above
 595          */
 596         ASSERT(RW_READ_HELD(&tp->tn_rwlock));
 597 
 598         rw_enter(&tp->tn_contents, RW_READER);
 599 
 600         error = rdtmp(tm, tp, uiop, ct);
 601 
 602         rw_exit(&tp->tn_contents);
 603 
 604         return (error);
 605 }
 606 
 607 static int
 608 tmp_write(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cred,
 609     struct caller_context *ct)
 610 {
 611         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 612         struct tmount *tm = (struct tmount *)VTOTM(vp);
 613         int error;
 614 
 615         /*
 616          * We don't currently support writing to non-regular files
 617          */
 618         if (vp->v_type != VREG)
 619                 return (EINVAL);        /* XXX EISDIR? */
 620 
 621         /*
 622          * tmp_rwlock should have already been called from layers above
 623          */
 624         ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
 625 
 626         rw_enter(&tp->tn_contents, RW_WRITER);
 627 
 628         if (ioflag & FAPPEND) {
 629                 /*
 630                  * In append mode start at end of file.
 631                  */
 632                 uiop->uio_loffset = tp->tn_size;
 633         }
 634 
 635         error = wrtmp(tm, tp, uiop, cred, ct);
 636 
 637         rw_exit(&tp->tn_contents);
 638 
 639         return (error);
 640 }
 641 
 642 /* ARGSUSED */
 643 static int
 644 tmp_ioctl(
 645         struct vnode *vp,
 646         int com,
 647         intptr_t data,
 648         int flag,
 649         struct cred *cred,
 650         int *rvalp,
 651         caller_context_t *ct)
 652 {
 653         return (ENOTTY);
 654 }
 655 
 656 /* ARGSUSED2 */
 657 static int
 658 tmp_getattr(
 659         struct vnode *vp,
 660         struct vattr *vap,
 661         int flags,
 662         struct cred *cred,
 663         caller_context_t *ct)
 664 {
 665         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 666         struct vnode *mvp;
 667         struct vattr va;
 668         int attrs = 1;
 669 
 670         /*
 671          * A special case to handle the root tnode on a diskless nfs
 672          * client who may have had its uid and gid inherited
 673          * from an nfs vnode with nobody ownership.  Likely the
 674          * root filesystem. After nfs is fully functional the uid/gid
 675          * may be mapable so ask again.
 676          * vfsp can't get unmounted because we hold vp.
 677          */
 678         if (vp->v_flag & VROOT &&
 679             (mvp = vp->v_vfsp->vfs_vnodecovered) != NULL) {
 680                 mutex_enter(&tp->tn_tlock);
 681                 if (tp->tn_uid == UID_NOBODY || tp->tn_gid == GID_NOBODY) {
 682                         mutex_exit(&tp->tn_tlock);
 683                         bzero(&va, sizeof (struct vattr));
 684                         va.va_mask = AT_UID|AT_GID;
 685                         attrs = VOP_GETATTR(mvp, &va, 0, cred, ct);
 686                 } else {
 687                         mutex_exit(&tp->tn_tlock);
 688                 }
 689         }
 690         mutex_enter(&tp->tn_tlock);
 691         if (attrs == 0) {
 692                 tp->tn_uid = va.va_uid;
 693                 tp->tn_gid = va.va_gid;
 694         }
 695         vap->va_type = vp->v_type;
 696         vap->va_mode = tp->tn_mode & MODEMASK;
 697         vap->va_uid = tp->tn_uid;
 698         vap->va_gid = tp->tn_gid;
 699         vap->va_fsid = tp->tn_fsid;
 700         vap->va_nodeid = (ino64_t)tp->tn_nodeid;
 701         vap->va_nlink = tp->tn_nlink;
 702         vap->va_size = (u_offset_t)tp->tn_size;
 703         vap->va_atime = tp->tn_atime;
 704         vap->va_mtime = tp->tn_mtime;
 705         vap->va_ctime = tp->tn_ctime;
 706         vap->va_blksize = PAGESIZE;
 707         vap->va_rdev = tp->tn_rdev;
 708         vap->va_seq = tp->tn_seq;
 709 
 710         /*
 711          * XXX Holes are not taken into account.  We could take the time to
 712          * run through the anon array looking for allocated slots...
 713          */
 714         vap->va_nblocks = (fsblkcnt64_t)btodb(ptob(btopr(vap->va_size)));
 715         mutex_exit(&tp->tn_tlock);
 716         return (0);
 717 }
 718 
 719 /*ARGSUSED4*/
 720 static int
 721 tmp_setattr(
 722         struct vnode *vp,
 723         struct vattr *vap,
 724         int flags,
 725         struct cred *cred,
 726         caller_context_t *ct)
 727 {
 728         struct tmount *tm = (struct tmount *)VTOTM(vp);
 729         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 730         int error = 0;
 731         struct vattr *get;
 732         long mask;
 733 
 734         /*
 735          * Cannot set these attributes
 736          */
 737         if ((vap->va_mask & AT_NOSET) || (vap->va_mask & AT_XVATTR))
 738                 return (EINVAL);
 739 
 740         mutex_enter(&tp->tn_tlock);
 741 
 742         get = &tp->tn_attr;
 743         /*
 744          * Change file access modes. Must be owner or have sufficient
 745          * privileges.
 746          */
 747         error = secpolicy_vnode_setattr(cred, vp, vap, get, flags, tmp_taccess,
 748             tp);
 749 
 750         if (error)
 751                 goto out;
 752 
 753         mask = vap->va_mask;
 754 
 755         if (mask & AT_MODE) {
 756                 get->va_mode &= S_IFMT;
 757                 get->va_mode |= vap->va_mode & ~S_IFMT;
 758         }
 759 
 760         if (mask & AT_UID)
 761                 get->va_uid = vap->va_uid;
 762         if (mask & AT_GID)
 763                 get->va_gid = vap->va_gid;
 764         if (mask & AT_ATIME)
 765                 get->va_atime = vap->va_atime;
 766         if (mask & AT_MTIME)
 767                 get->va_mtime = vap->va_mtime;
 768 
 769         if (mask & (AT_UID | AT_GID | AT_MODE | AT_MTIME))
 770                 gethrestime(&tp->tn_ctime);
 771 
 772         if (mask & AT_SIZE) {
 773                 ASSERT(vp->v_type != VDIR);
 774 
 775                 /* Don't support large files. */
 776                 if (vap->va_size > MAXOFF_T) {
 777                         error = EFBIG;
 778                         goto out;
 779                 }
 780                 mutex_exit(&tp->tn_tlock);
 781 
 782                 rw_enter(&tp->tn_rwlock, RW_WRITER);
 783                 rw_enter(&tp->tn_contents, RW_WRITER);
 784                 error = tmpnode_trunc(tm, tp, (ulong_t)vap->va_size);
 785                 rw_exit(&tp->tn_contents);
 786                 rw_exit(&tp->tn_rwlock);
 787                 goto out1;
 788         }
 789 out:
 790         mutex_exit(&tp->tn_tlock);
 791 out1:
 792         return (error);
 793 }
 794 
 795 /* ARGSUSED2 */
 796 static int
 797 tmp_access(
 798         struct vnode *vp,
 799         int mode,
 800         int flags,
 801         struct cred *cred,
 802         caller_context_t *ct)
 803 {
 804         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 805         int error;
 806 
 807         mutex_enter(&tp->tn_tlock);
 808         error = tmp_taccess(tp, mode, cred);
 809         mutex_exit(&tp->tn_tlock);
 810         return (error);
 811 }
 812 
 813 /* ARGSUSED3 */
 814 static int
 815 tmp_lookup(
 816         struct vnode *dvp,
 817         char *nm,
 818         struct vnode **vpp,
 819         struct pathname *pnp,
 820         int flags,
 821         struct vnode *rdir,
 822         struct cred *cred,
 823         caller_context_t *ct,
 824         int *direntflags,
 825         pathname_t *realpnp)
 826 {
 827         struct tmpnode *tp = (struct tmpnode *)VTOTN(dvp);
 828         struct tmpnode *ntp = NULL;
 829         int error;
 830 
 831 
 832         /* allow cd into @ dir */
 833         if (flags & LOOKUP_XATTR) {
 834                 struct tmpnode *xdp;
 835                 struct tmount *tm;
 836 
 837                 /*
 838                  * don't allow attributes if not mounted XATTR support
 839                  */
 840                 if (!(dvp->v_vfsp->vfs_flag & VFS_XATTR))
 841                         return (EINVAL);
 842 
 843                 if (tp->tn_flags & ISXATTR)
 844                         /* No attributes on attributes */
 845                         return (EINVAL);
 846 
 847                 rw_enter(&tp->tn_rwlock, RW_WRITER);
 848                 if (tp->tn_xattrdp == NULL) {
 849                         if (!(flags & CREATE_XATTR_DIR)) {
 850                                 rw_exit(&tp->tn_rwlock);
 851                                 return (ENOENT);
 852                         }
 853 
 854                         /*
 855                          * No attribute directory exists for this
 856                          * node - create the attr dir as a side effect
 857                          * of this lookup.
 858                          */
 859 
 860                         /*
 861                          * Make sure we have adequate permission...
 862                          */
 863 
 864                         if ((error = tmp_taccess(tp, VWRITE, cred)) != 0) {
 865                                 rw_exit(&tp->tn_rwlock);
 866                                 return (error);
 867                         }
 868 
 869                         xdp = tmp_memalloc(sizeof (struct tmpnode),
 870                             TMP_MUSTHAVE);
 871                         tm = VTOTM(dvp);
 872                         tmpnode_init(tm, xdp, &tp->tn_attr, NULL);
 873                         /*
 874                          * Fix-up fields unique to attribute directories.
 875                          */
 876                         xdp->tn_flags = ISXATTR;
 877                         xdp->tn_type = VDIR;
 878                         if (tp->tn_type == VDIR) {
 879                                 xdp->tn_mode = tp->tn_attr.va_mode;
 880                         } else {
 881                                 xdp->tn_mode = 0700;
 882                                 if (tp->tn_attr.va_mode & 0040)
 883                                         xdp->tn_mode |= 0750;
 884                                 if (tp->tn_attr.va_mode & 0004)
 885                                         xdp->tn_mode |= 0705;
 886                         }
 887                         xdp->tn_vnode->v_type = VDIR;
 888                         xdp->tn_vnode->v_flag |= V_XATTRDIR;
 889                         tdirinit(tp, xdp);
 890                         tp->tn_xattrdp = xdp;
 891                 } else {
 892                         VN_HOLD(tp->tn_xattrdp->tn_vnode);
 893                 }
 894                 *vpp = TNTOV(tp->tn_xattrdp);
 895                 rw_exit(&tp->tn_rwlock);
 896                 return (0);
 897         }
 898 
 899         /*
 900          * Null component name is a synonym for directory being searched.
 901          */
 902         if (*nm == '\0') {
 903                 VN_HOLD(dvp);
 904                 *vpp = dvp;
 905                 return (0);
 906         }
 907         ASSERT(tp);
 908 
 909         error = tdirlookup(tp, nm, &ntp, cred);
 910 
 911         if (error == 0) {
 912                 ASSERT(ntp);
 913                 *vpp = TNTOV(ntp);
 914                 /*
 915                  * If vnode is a device return special vnode instead
 916                  */
 917                 if (IS_DEVVP(*vpp)) {
 918                         struct vnode *newvp;
 919 
 920                         newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
 921                             cred);
 922                         VN_RELE(*vpp);
 923                         *vpp = newvp;
 924                 }
 925         }
 926         TRACE_4(TR_FAC_TMPFS, TR_TMPFS_LOOKUP,
 927             "tmpfs lookup:vp %p name %s vpp %p error %d",
 928             dvp, nm, vpp, error);
 929         return (error);
 930 }
 931 
 932 /*ARGSUSED7*/
 933 static int
 934 tmp_create(
 935         struct vnode *dvp,
 936         char *nm,
 937         struct vattr *vap,
 938         enum vcexcl exclusive,
 939         int mode,
 940         struct vnode **vpp,
 941         struct cred *cred,
 942         int flag,
 943         caller_context_t *ct,
 944         vsecattr_t *vsecp)
 945 {
 946         struct tmpnode *parent;
 947         struct tmount *tm;
 948         struct tmpnode *self;
 949         int error;
 950         struct tmpnode *oldtp;
 951 
 952 again:
 953         parent = (struct tmpnode *)VTOTN(dvp);
 954         tm = (struct tmount *)VTOTM(dvp);
 955         self = NULL;
 956         error = 0;
 957         oldtp = NULL;
 958 
 959         /* device files not allowed in ext. attr dirs */
 960         if ((parent->tn_flags & ISXATTR) &&
 961             (vap->va_type == VBLK || vap->va_type == VCHR ||
 962             vap->va_type == VFIFO || vap->va_type == VDOOR ||
 963             vap->va_type == VSOCK || vap->va_type == VPORT))
 964                         return (EINVAL);
 965 
 966         if (vap->va_type == VREG && (vap->va_mode & VSVTX)) {
 967                 /* Must be privileged to set sticky bit */
 968                 if (secpolicy_vnode_stky_modify(cred))
 969                         vap->va_mode &= ~VSVTX;
 970         } else if (vap->va_type == VNON) {
 971                 return (EINVAL);
 972         }
 973 
 974         /*
 975          * Null component name is a synonym for directory being searched.
 976          */
 977         if (*nm == '\0') {
 978                 VN_HOLD(dvp);
 979                 oldtp = parent;
 980         } else {
 981                 error = tdirlookup(parent, nm, &oldtp, cred);
 982         }
 983 
 984         if (error == 0) {       /* name found */
 985                 boolean_t trunc = B_FALSE;
 986 
 987                 ASSERT(oldtp);
 988 
 989                 rw_enter(&oldtp->tn_rwlock, RW_WRITER);
 990 
 991                 /*
 992                  * if create/read-only an existing
 993                  * directory, allow it
 994                  */
 995                 if (exclusive == EXCL)
 996                         error = EEXIST;
 997                 else if ((oldtp->tn_type == VDIR) && (mode & VWRITE))
 998                         error = EISDIR;
 999                 else {
1000                         error = tmp_taccess(oldtp, mode, cred);
1001                 }
1002 
1003                 if (error) {
1004                         rw_exit(&oldtp->tn_rwlock);
1005                         tmpnode_rele(oldtp);
1006                         return (error);
1007                 }
1008                 *vpp = TNTOV(oldtp);
1009                 if ((*vpp)->v_type == VREG && (vap->va_mask & AT_SIZE) &&
1010                     vap->va_size == 0) {
1011                         rw_enter(&oldtp->tn_contents, RW_WRITER);
1012                         (void) tmpnode_trunc(tm, oldtp, 0);
1013                         rw_exit(&oldtp->tn_contents);
1014                         trunc = B_TRUE;
1015                 }
1016                 rw_exit(&oldtp->tn_rwlock);
1017                 if (IS_DEVVP(*vpp)) {
1018                         struct vnode *newvp;
1019 
1020                         newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
1021                             cred);
1022                         VN_RELE(*vpp);
1023                         if (newvp == NULL) {
1024                                 return (ENOSYS);
1025                         }
1026                         *vpp = newvp;
1027                 }
1028 
1029                 if (trunc)
1030                         vnevent_create(*vpp, ct);
1031 
1032                 return (0);
1033         }
1034 
1035         if (error != ENOENT)
1036                 return (error);
1037 
1038         rw_enter(&parent->tn_rwlock, RW_WRITER);
1039         error = tdirenter(tm, parent, nm, DE_CREATE,
1040             (struct tmpnode *)NULL, (struct tmpnode *)NULL,
1041             vap, &self, cred, ct);
1042         rw_exit(&parent->tn_rwlock);
1043 
1044         if (error) {
1045                 if (self)
1046                         tmpnode_rele(self);
1047 
1048                 if (error == EEXIST) {
1049                         /*
1050                          * This means that the file was created sometime
1051                          * after we checked and did not find it and when
1052                          * we went to create it.
1053                          * Since creat() is supposed to truncate a file
1054                          * that already exits go back to the begining
1055                          * of the function. This time we will find it
1056                          * and go down the tmp_trunc() path
1057                          */
1058                         goto again;
1059                 }
1060                 return (error);
1061         }
1062 
1063         *vpp = TNTOV(self);
1064 
1065         if (!error && IS_DEVVP(*vpp)) {
1066                 struct vnode *newvp;
1067 
1068                 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cred);
1069                 VN_RELE(*vpp);
1070                 if (newvp == NULL)
1071                         return (ENOSYS);
1072                 *vpp = newvp;
1073         }
1074         TRACE_3(TR_FAC_TMPFS, TR_TMPFS_CREATE,
1075             "tmpfs create:dvp %p nm %s vpp %p", dvp, nm, vpp);
1076         return (0);
1077 }
1078 
1079 /* ARGSUSED3 */
1080 static int
1081 tmp_remove(
1082         struct vnode *dvp,
1083         char *nm,
1084         struct cred *cred,
1085         caller_context_t *ct,
1086         int flags)
1087 {
1088         struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1089         int error;
1090         struct tmpnode *tp = NULL;
1091 
1092         error = tdirlookup(parent, nm, &tp, cred);
1093         if (error)
1094                 return (error);
1095 
1096         ASSERT(tp);
1097         rw_enter(&parent->tn_rwlock, RW_WRITER);
1098         rw_enter(&tp->tn_rwlock, RW_WRITER);
1099 
1100         if (tp->tn_type != VDIR ||
1101             (error = secpolicy_fs_linkdir(cred, dvp->v_vfsp)) == 0)
1102                 error = tdirdelete(parent, tp, nm, DR_REMOVE, cred);
1103 
1104         rw_exit(&tp->tn_rwlock);
1105         rw_exit(&parent->tn_rwlock);
1106         vnevent_remove(TNTOV(tp), dvp, nm, ct);
1107         tmpnode_rele(tp);
1108 
1109         TRACE_3(TR_FAC_TMPFS, TR_TMPFS_REMOVE,
1110             "tmpfs remove:dvp %p nm %s error %d", dvp, nm, error);
1111         return (error);
1112 }
1113 
1114 /* ARGSUSED4 */
1115 static int
1116 tmp_link(
1117         struct vnode *dvp,
1118         struct vnode *srcvp,
1119         char *tnm,
1120         struct cred *cred,
1121         caller_context_t *ct,
1122         int flags)
1123 {
1124         struct tmpnode *parent;
1125         struct tmpnode *from;
1126         struct tmount *tm = (struct tmount *)VTOTM(dvp);
1127         int error;
1128         struct tmpnode *found = NULL;
1129         struct vnode *realvp;
1130 
1131         if (VOP_REALVP(srcvp, &realvp, ct) == 0)
1132                 srcvp = realvp;
1133 
1134         parent = (struct tmpnode *)VTOTN(dvp);
1135         from = (struct tmpnode *)VTOTN(srcvp);
1136 
1137         if ((srcvp->v_type == VDIR &&
1138             secpolicy_fs_linkdir(cred, dvp->v_vfsp)) ||
1139             (from->tn_uid != crgetuid(cred) && secpolicy_basic_link(cred)))
1140                 return (EPERM);
1141 
1142         /*
1143          * Make sure link for extended attributes is valid
1144          * We only support hard linking of xattr's in xattrdir to an xattrdir
1145          */
1146         if ((from->tn_flags & ISXATTR) != (parent->tn_flags & ISXATTR))
1147                 return (EINVAL);
1148 
1149         error = tdirlookup(parent, tnm, &found, cred);
1150         if (error == 0) {
1151                 ASSERT(found);
1152                 tmpnode_rele(found);
1153                 return (EEXIST);
1154         }
1155 
1156         if (error != ENOENT)
1157                 return (error);
1158 
1159         rw_enter(&parent->tn_rwlock, RW_WRITER);
1160         error = tdirenter(tm, parent, tnm, DE_LINK, (struct tmpnode *)NULL,
1161             from, NULL, (struct tmpnode **)NULL, cred, ct);
1162         rw_exit(&parent->tn_rwlock);
1163         if (error == 0) {
1164                 vnevent_link(srcvp, ct);
1165         }
1166         return (error);
1167 }
1168 
1169 /* ARGSUSED5 */
1170 static int
1171 tmp_rename(
1172         struct vnode *odvp,     /* source parent vnode */
1173         char *onm,              /* source name */
1174         struct vnode *ndvp,     /* destination parent vnode */
1175         char *nnm,              /* destination name */
1176         struct cred *cred,
1177         caller_context_t *ct,
1178         int flags)
1179 {
1180         struct tmpnode *fromparent;
1181         struct tmpnode *toparent;
1182         struct tmpnode *fromtp = NULL;  /* source tmpnode */
1183         struct tmount *tm = (struct tmount *)VTOTM(odvp);
1184         int error;
1185         int samedir = 0;        /* set if odvp == ndvp */
1186         struct vnode *realvp;
1187 
1188         if (VOP_REALVP(ndvp, &realvp, ct) == 0)
1189                 ndvp = realvp;
1190 
1191         fromparent = (struct tmpnode *)VTOTN(odvp);
1192         toparent = (struct tmpnode *)VTOTN(ndvp);
1193 
1194         if ((fromparent->tn_flags & ISXATTR) != (toparent->tn_flags & ISXATTR))
1195                 return (EINVAL);
1196 
1197         mutex_enter(&tm->tm_renamelck);
1198 
1199         /*
1200          * Look up tmpnode of file we're supposed to rename.
1201          */
1202         error = tdirlookup(fromparent, onm, &fromtp, cred);
1203         if (error) {
1204                 mutex_exit(&tm->tm_renamelck);
1205                 return (error);
1206         }
1207 
1208         /*
1209          * Make sure we can delete the old (source) entry.  This
1210          * requires write permission on the containing directory.  If
1211          * that directory is "sticky" it requires further checks.
1212          */
1213         if (((error = tmp_taccess(fromparent, VWRITE, cred)) != 0) ||
1214             (error = tmp_sticky_remove_access(fromparent, fromtp, cred)) != 0)
1215                 goto done;
1216 
1217         /*
1218          * Check for renaming to or from '.' or '..' or that
1219          * fromtp == fromparent
1220          */
1221         if ((onm[0] == '.' &&
1222             (onm[1] == '\0' || (onm[1] == '.' && onm[2] == '\0'))) ||
1223             (nnm[0] == '.' &&
1224             (nnm[1] == '\0' || (nnm[1] == '.' && nnm[2] == '\0'))) ||
1225             (fromparent == fromtp)) {
1226                 error = EINVAL;
1227                 goto done;
1228         }
1229 
1230         samedir = (fromparent == toparent);
1231         /*
1232          * Make sure we can search and rename into the new
1233          * (destination) directory.
1234          */
1235         if (!samedir) {
1236                 error = tmp_taccess(toparent, VEXEC|VWRITE, cred);
1237                 if (error)
1238                         goto done;
1239         }
1240 
1241         /*
1242          * Link source to new target
1243          */
1244         rw_enter(&toparent->tn_rwlock, RW_WRITER);
1245         error = tdirenter(tm, toparent, nnm, DE_RENAME,
1246             fromparent, fromtp, (struct vattr *)NULL,
1247             (struct tmpnode **)NULL, cred, ct);
1248         rw_exit(&toparent->tn_rwlock);
1249 
1250         if (error) {
1251                 /*
1252                  * ESAME isn't really an error; it indicates that the
1253                  * operation should not be done because the source and target
1254                  * are the same file, but that no error should be reported.
1255                  */
1256                 if (error == ESAME)
1257                         error = 0;
1258                 goto done;
1259         }
1260         vnevent_rename_src(TNTOV(fromtp), odvp, onm, ct);
1261 
1262         /*
1263          * Notify the target directory if not same as
1264          * source directory.
1265          */
1266         if (ndvp != odvp) {
1267                 vnevent_rename_dest_dir(ndvp, ct);
1268         }
1269 
1270         /*
1271          * Unlink from source.
1272          */
1273         rw_enter(&fromparent->tn_rwlock, RW_WRITER);
1274         rw_enter(&fromtp->tn_rwlock, RW_WRITER);
1275 
1276         error = tdirdelete(fromparent, fromtp, onm, DR_RENAME, cred);
1277 
1278         /*
1279          * The following handles the case where our source tmpnode was
1280          * removed before we got to it.
1281          *
1282          * XXX We should also cleanup properly in the case where tdirdelete
1283          * fails for some other reason.  Currently this case shouldn't happen.
1284          * (see 1184991).
1285          */
1286         if (error == ENOENT)
1287                 error = 0;
1288 
1289         rw_exit(&fromtp->tn_rwlock);
1290         rw_exit(&fromparent->tn_rwlock);
1291 done:
1292         tmpnode_rele(fromtp);
1293         mutex_exit(&tm->tm_renamelck);
1294 
1295         TRACE_5(TR_FAC_TMPFS, TR_TMPFS_RENAME,
1296             "tmpfs rename:ovp %p onm %s nvp %p nnm %s error %d", odvp, onm,
1297             ndvp, nnm, error);
1298         return (error);
1299 }
1300 
1301 /* ARGSUSED5 */
1302 static int
1303 tmp_mkdir(
1304         struct vnode *dvp,
1305         char *nm,
1306         struct vattr *va,
1307         struct vnode **vpp,
1308         struct cred *cred,
1309         caller_context_t *ct,
1310         int flags,
1311         vsecattr_t *vsecp)
1312 {
1313         struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1314         struct tmpnode *self = NULL;
1315         struct tmount *tm = (struct tmount *)VTOTM(dvp);
1316         int error;
1317 
1318         /* no new dirs allowed in xattr dirs */
1319         if (parent->tn_flags & ISXATTR)
1320                 return (EINVAL);
1321 
1322         /*
1323          * Might be dangling directory.  Catch it here,
1324          * because a ENOENT return from tdirlookup() is
1325          * an "o.k. return".
1326          */
1327         if (parent->tn_nlink == 0)
1328                 return (ENOENT);
1329 
1330         error = tdirlookup(parent, nm, &self, cred);
1331         if (error == 0) {
1332                 ASSERT(self);
1333                 tmpnode_rele(self);
1334                 return (EEXIST);
1335         }
1336         if (error != ENOENT)
1337                 return (error);
1338 
1339         rw_enter(&parent->tn_rwlock, RW_WRITER);
1340         error = tdirenter(tm, parent, nm, DE_MKDIR, (struct tmpnode *)NULL,
1341             (struct tmpnode *)NULL, va, &self, cred, ct);
1342         if (error) {
1343                 rw_exit(&parent->tn_rwlock);
1344                 if (self)
1345                         tmpnode_rele(self);
1346                 return (error);
1347         }
1348         rw_exit(&parent->tn_rwlock);
1349         *vpp = TNTOV(self);
1350         return (0);
1351 }
1352 
1353 /* ARGSUSED4 */
1354 static int
1355 tmp_rmdir(
1356         struct vnode *dvp,
1357         char *nm,
1358         struct vnode *cdir,
1359         struct cred *cred,
1360         caller_context_t *ct,
1361         int flags)
1362 {
1363         struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1364         struct tmpnode *self = NULL;
1365         struct vnode *vp;
1366         int error = 0;
1367 
1368         /*
1369          * Return error when removing . and ..
1370          */
1371         if (strcmp(nm, ".") == 0)
1372                 return (EINVAL);
1373         if (strcmp(nm, "..") == 0)
1374                 return (EEXIST); /* Should be ENOTEMPTY */
1375         error = tdirlookup(parent, nm, &self, cred);
1376         if (error)
1377                 return (error);
1378 
1379         rw_enter(&parent->tn_rwlock, RW_WRITER);
1380         rw_enter(&self->tn_rwlock, RW_WRITER);
1381 
1382         vp = TNTOV(self);
1383         if (vp == dvp || vp == cdir) {
1384                 error = EINVAL;
1385                 goto done1;
1386         }
1387         if (self->tn_type != VDIR) {
1388                 error = ENOTDIR;
1389                 goto done1;
1390         }
1391 
1392         mutex_enter(&self->tn_tlock);
1393         if (self->tn_nlink > 2) {
1394                 mutex_exit(&self->tn_tlock);
1395                 error = EEXIST;
1396                 goto done1;
1397         }
1398         mutex_exit(&self->tn_tlock);
1399 
1400         if (vn_vfswlock(vp)) {
1401                 error = EBUSY;
1402                 goto done1;
1403         }
1404         if (vn_mountedvfs(vp) != NULL) {
1405                 error = EBUSY;
1406                 goto done;
1407         }
1408 
1409         /*
1410          * Check for an empty directory
1411          * i.e. only includes entries for "." and ".."
1412          */
1413         if (self->tn_dirents > 2) {
1414                 error = EEXIST;         /* SIGH should be ENOTEMPTY */
1415                 /*
1416                  * Update atime because checking tn_dirents is logically
1417                  * equivalent to reading the directory
1418                  */
1419                 gethrestime(&self->tn_atime);
1420                 goto done;
1421         }
1422 
1423         error = tdirdelete(parent, self, nm, DR_RMDIR, cred);
1424 done:
1425         vn_vfsunlock(vp);
1426 done1:
1427         rw_exit(&self->tn_rwlock);
1428         rw_exit(&parent->tn_rwlock);
1429         vnevent_rmdir(TNTOV(self), dvp, nm, ct);
1430         tmpnode_rele(self);
1431 
1432         return (error);
1433 }
1434 
1435 /* ARGSUSED2 */
1436 static int
1437 tmp_readdir(
1438         struct vnode *vp,
1439         struct uio *uiop,
1440         struct cred *cred,
1441         int *eofp,
1442         caller_context_t *ct,
1443         int flags)
1444 {
1445         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1446         struct tdirent *tdp;
1447         int error = 0;
1448         size_t namelen;
1449         struct dirent64 *dp;
1450         ulong_t offset;
1451         ulong_t total_bytes_wanted;
1452         long outcount = 0;
1453         long bufsize;
1454         int reclen;
1455         caddr_t outbuf;
1456 
1457         if (uiop->uio_loffset >= MAXOFF_T) {
1458                 if (eofp)
1459                         *eofp = 1;
1460                 return (0);
1461         }
1462         /*
1463          * assuming system call has already called tmp_rwlock
1464          */
1465         ASSERT(RW_READ_HELD(&tp->tn_rwlock));
1466 
1467         if (uiop->uio_iovcnt != 1)
1468                 return (EINVAL);
1469 
1470         if (vp->v_type != VDIR)
1471                 return (ENOTDIR);
1472 
1473         /*
1474          * There's a window here where someone could have removed
1475          * all the entries in the directory after we put a hold on the
1476          * vnode but before we grabbed the rwlock.  Just return.
1477          */
1478         if (tp->tn_dir == NULL) {
1479                 if (tp->tn_nlink) {
1480                         panic("empty directory 0x%p", (void *)tp);
1481                         /*NOTREACHED*/
1482                 }
1483                 return (0);
1484         }
1485 
1486         /*
1487          * Get space for multiple directory entries
1488          */
1489         total_bytes_wanted = uiop->uio_iov->iov_len;
1490         bufsize = total_bytes_wanted + sizeof (struct dirent64);
1491         outbuf = kmem_alloc(bufsize, KM_SLEEP);
1492 
1493         dp = (struct dirent64 *)outbuf;
1494 
1495 
1496         offset = 0;
1497         tdp = tp->tn_dir;
1498         while (tdp) {
1499                 namelen = strlen(tdp->td_name);      /* no +1 needed */
1500                 offset = tdp->td_offset;
1501                 if (offset >= uiop->uio_offset) {
1502                         reclen = (int)DIRENT64_RECLEN(namelen);
1503                         if (outcount + reclen > total_bytes_wanted) {
1504                                 if (!outcount)
1505                                         /*
1506                                          * Buffer too small for any entries.
1507                                          */
1508                                         error = EINVAL;
1509                                 break;
1510                         }
1511                         ASSERT(tdp->td_tmpnode != NULL);
1512 
1513                         /* use strncpy(9f) to zero out uninitialized bytes */
1514 
1515                         (void) strncpy(dp->d_name, tdp->td_name,
1516                             DIRENT64_NAMELEN(reclen));
1517                         dp->d_reclen = (ushort_t)reclen;
1518                         dp->d_ino = (ino64_t)tdp->td_tmpnode->tn_nodeid;
1519                         dp->d_off = (offset_t)tdp->td_offset + 1;
1520                         dp = (struct dirent64 *)
1521                             ((uintptr_t)dp + dp->d_reclen);
1522                         outcount += reclen;
1523                         ASSERT(outcount <= bufsize);
1524                 }
1525                 tdp = tdp->td_next;
1526         }
1527 
1528         if (!error)
1529                 error = uiomove(outbuf, outcount, UIO_READ, uiop);
1530 
1531         if (!error) {
1532                 /* If we reached the end of the list our offset */
1533                 /* should now be just past the end. */
1534                 if (!tdp) {
1535                         offset += 1;
1536                         if (eofp)
1537                                 *eofp = 1;
1538                 } else if (eofp)
1539                         *eofp = 0;
1540                 uiop->uio_offset = offset;
1541         }
1542         gethrestime(&tp->tn_atime);
1543         kmem_free(outbuf, bufsize);
1544         return (error);
1545 }
1546 
1547 /* ARGSUSED5 */
1548 static int
1549 tmp_symlink(
1550         struct vnode *dvp,
1551         char *lnm,
1552         struct vattr *tva,
1553         char *tnm,
1554         struct cred *cred,
1555         caller_context_t *ct,
1556         int flags)
1557 {
1558         struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1559         struct tmpnode *self = (struct tmpnode *)NULL;
1560         struct tmount *tm = (struct tmount *)VTOTM(dvp);
1561         char *cp = NULL;
1562         int error;
1563         size_t len;
1564 
1565         /* no symlinks allowed to files in xattr dirs */
1566         if (parent->tn_flags & ISXATTR)
1567                 return (EINVAL);
1568 
1569         error = tdirlookup(parent, lnm, &self, cred);
1570         if (error == 0) {
1571                 /*
1572                  * The entry already exists
1573                  */
1574                 tmpnode_rele(self);
1575                 return (EEXIST);        /* was 0 */
1576         }
1577 
1578         if (error != ENOENT) {
1579                 if (self != NULL)
1580                         tmpnode_rele(self);
1581                 return (error);
1582         }
1583 
1584         rw_enter(&parent->tn_rwlock, RW_WRITER);
1585         error = tdirenter(tm, parent, lnm, DE_CREATE, (struct tmpnode *)NULL,
1586             (struct tmpnode *)NULL, tva, &self, cred, ct);
1587         rw_exit(&parent->tn_rwlock);
1588 
1589         if (error) {
1590                 if (self)
1591                         tmpnode_rele(self);
1592                 return (error);
1593         }
1594         len = strlen(tnm) + 1;
1595         cp = tmp_memalloc(len, 0);
1596         if (cp == NULL) {
1597                 tmpnode_rele(self);
1598                 return (ENOSPC);
1599         }
1600         (void) strcpy(cp, tnm);
1601 
1602         self->tn_symlink = cp;
1603         self->tn_size = len - 1;
1604         tmpnode_rele(self);
1605         return (error);
1606 }
1607 
1608 /* ARGSUSED2 */
1609 static int
1610 tmp_readlink(
1611         struct vnode *vp,
1612         struct uio *uiop,
1613         struct cred *cred,
1614         caller_context_t *ct)
1615 {
1616         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1617         int error = 0;
1618 
1619         if (vp->v_type != VLNK)
1620                 return (EINVAL);
1621 
1622         rw_enter(&tp->tn_rwlock, RW_READER);
1623         rw_enter(&tp->tn_contents, RW_READER);
1624         error = uiomove(tp->tn_symlink, tp->tn_size, UIO_READ, uiop);
1625         gethrestime(&tp->tn_atime);
1626         rw_exit(&tp->tn_contents);
1627         rw_exit(&tp->tn_rwlock);
1628         return (error);
1629 }
1630 
1631 /* ARGSUSED */
1632 static int
1633 tmp_fsync(
1634         struct vnode *vp,
1635         int syncflag,
1636         struct cred *cred,
1637         caller_context_t *ct)
1638 {
1639         return (0);
1640 }
1641 
1642 /* ARGSUSED */
1643 static void
1644 tmp_inactive(struct vnode *vp, struct cred *cred, caller_context_t *ct)
1645 {
1646         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1647         struct tmount *tm = (struct tmount *)VFSTOTM(vp->v_vfsp);
1648 
1649         rw_enter(&tp->tn_rwlock, RW_WRITER);
1650 top:
1651         mutex_enter(&tp->tn_tlock);
1652         mutex_enter(&vp->v_lock);
1653         ASSERT(vp->v_count >= 1);
1654 
1655         /*
1656          * If we don't have the last hold or the link count is non-zero,
1657          * there's little to do -- just drop our hold.
1658          */
1659         if (vp->v_count > 1 || tp->tn_nlink != 0) {
1660                 vp->v_count--;
1661                 mutex_exit(&vp->v_lock);
1662                 mutex_exit(&tp->tn_tlock);
1663                 rw_exit(&tp->tn_rwlock);
1664                 return;
1665         }
1666 
1667         /*
1668          * We have the last hold *and* the link count is zero, so this
1669          * tmpnode is dead from the filesystem's viewpoint.  However,
1670          * if the tmpnode has any pages associated with it (i.e. if it's
1671          * a normal file with non-zero size), the tmpnode can still be
1672          * discovered by pageout or fsflush via the page vnode pointers.
1673          * In this case we must drop all our locks, truncate the tmpnode,
1674          * and try the whole dance again.
1675          */
1676         if (tp->tn_size != 0) {
1677                 if (tp->tn_type == VREG) {
1678                         mutex_exit(&vp->v_lock);
1679                         mutex_exit(&tp->tn_tlock);
1680                         rw_enter(&tp->tn_contents, RW_WRITER);
1681                         (void) tmpnode_trunc(tm, tp, 0);
1682                         rw_exit(&tp->tn_contents);
1683                         ASSERT(tp->tn_size == 0);
1684                         ASSERT(tp->tn_nblocks == 0);
1685                         goto top;
1686                 }
1687                 if (tp->tn_type == VLNK)
1688                         tmp_memfree(tp->tn_symlink, tp->tn_size + 1);
1689         }
1690 
1691         /*
1692          * Remove normal file/dir's xattr dir and xattrs.
1693          */
1694         if (tp->tn_xattrdp) {
1695                 struct tmpnode *xtp = tp->tn_xattrdp;
1696 
1697                 ASSERT(xtp->tn_flags & ISXATTR);
1698                 tmpnode_hold(xtp);
1699                 rw_enter(&xtp->tn_rwlock, RW_WRITER);
1700                 tdirtrunc(xtp);
1701                 DECR_COUNT(&xtp->tn_nlink, &xtp->tn_tlock);
1702                 tp->tn_xattrdp = NULL;
1703                 rw_exit(&xtp->tn_rwlock);
1704                 tmpnode_rele(xtp);
1705         }
1706 
1707         mutex_exit(&vp->v_lock);
1708         mutex_exit(&tp->tn_tlock);
1709         /* Here's our chance to send invalid event while we're between locks */
1710         vn_invalid(TNTOV(tp));
1711         mutex_enter(&tm->tm_contents);
1712         if (tp->tn_forw == NULL)
1713                 tm->tm_rootnode->tn_back = tp->tn_back;
1714         else
1715                 tp->tn_forw->tn_back = tp->tn_back;
1716         tp->tn_back->tn_forw = tp->tn_forw;
1717         mutex_exit(&tm->tm_contents);
1718         rw_exit(&tp->tn_rwlock);
1719         rw_destroy(&tp->tn_rwlock);
1720         mutex_destroy(&tp->tn_tlock);
1721         vn_free(TNTOV(tp));
1722         tmp_memfree(tp, sizeof (struct tmpnode));
1723 }
1724 
1725 /* ARGSUSED2 */
1726 static int
1727 tmp_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
1728 {
1729         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1730         struct tfid *tfid;
1731 
1732         if (fidp->fid_len < (sizeof (struct tfid) - sizeof (ushort_t))) {
1733                 fidp->fid_len = sizeof (struct tfid) - sizeof (ushort_t);
1734                 return (ENOSPC);
1735         }
1736 
1737         tfid = (struct tfid *)fidp;
1738         bzero(tfid, sizeof (struct tfid));
1739         tfid->tfid_len = (int)sizeof (struct tfid) - sizeof (ushort_t);
1740 
1741         tfid->tfid_ino = tp->tn_nodeid;
1742         tfid->tfid_gen = tp->tn_gen;
1743 
1744         return (0);
1745 }
1746 
1747 
1748 /*
1749  * Return all the pages from [off..off+len] in given file
1750  */
1751 /* ARGSUSED */
1752 static int
1753 tmp_getpage(
1754         struct vnode *vp,
1755         offset_t off,
1756         size_t len,
1757         uint_t *protp,
1758         page_t *pl[],
1759         size_t plsz,
1760         struct seg *seg,
1761         caddr_t addr,
1762         enum seg_rw rw,
1763         struct cred *cr,
1764         caller_context_t *ct)
1765 {
1766         int err = 0;
1767         struct tmpnode *tp = VTOTN(vp);
1768         anoff_t toff = (anoff_t)off;
1769         size_t tlen = len;
1770         u_offset_t tmpoff;
1771         timestruc_t now;
1772 
1773         rw_enter(&tp->tn_contents, RW_READER);
1774 
1775         if (off + len  > tp->tn_size + PAGEOFFSET) {
1776                 err = EFAULT;
1777                 goto out;
1778         }
1779         /*
1780          * Look for holes (no anon slot) in faulting range. If there are
1781          * holes we have to switch to a write lock and fill them in. Swap
1782          * space for holes was already reserved when the file was grown.
1783          */
1784         tmpoff = toff;
1785         if (non_anon(tp->tn_anon, btop(off), &tmpoff, &tlen)) {
1786                 if (!rw_tryupgrade(&tp->tn_contents)) {
1787                         rw_exit(&tp->tn_contents);
1788                         rw_enter(&tp->tn_contents, RW_WRITER);
1789                         /* Size may have changed when lock was dropped */
1790                         if (off + len  > tp->tn_size + PAGEOFFSET) {
1791                                 err = EFAULT;
1792                                 goto out;
1793                         }
1794                 }
1795                 for (toff = (anoff_t)off; toff < (anoff_t)off + len;
1796                     toff += PAGESIZE) {
1797                         if (anon_get_ptr(tp->tn_anon, btop(toff)) == NULL) {
1798                                 /* XXX - may allocate mem w. write lock held */
1799                                 (void) anon_set_ptr(tp->tn_anon, btop(toff),
1800                                     anon_alloc(vp, toff), ANON_SLEEP);
1801                                 tp->tn_nblocks++;
1802                         }
1803                 }
1804                 rw_downgrade(&tp->tn_contents);
1805         }
1806 
1807 
1808         if (len <= PAGESIZE)
1809                 err = tmp_getapage(vp, (u_offset_t)off, len, protp, pl, plsz,
1810                     seg, addr, rw, cr);
1811         else
1812                 err = pvn_getpages(tmp_getapage, vp, (u_offset_t)off, len,
1813                     protp, pl, plsz, seg, addr, rw, cr);
1814 
1815         gethrestime(&now);
1816         tp->tn_atime = now;
1817         if (rw == S_WRITE)
1818                 tp->tn_mtime = now;
1819 
1820 out:
1821         rw_exit(&tp->tn_contents);
1822         return (err);
1823 }
1824 
1825 /*
1826  * Called from pvn_getpages or swap_getpage to get a particular page.
1827  */
1828 /*ARGSUSED*/
1829 static int
1830 tmp_getapage(
1831         struct vnode *vp,
1832         u_offset_t off,
1833         size_t len,
1834         uint_t *protp,
1835         page_t *pl[],
1836         size_t plsz,
1837         struct seg *seg,
1838         caddr_t addr,
1839         enum seg_rw rw,
1840         struct cred *cr)
1841 {
1842         struct page *pp;
1843         int flags;
1844         int err = 0;
1845         struct vnode *pvp;
1846         u_offset_t poff;
1847 
1848         if (protp != NULL)
1849                 *protp = PROT_ALL;
1850 again:
1851         if (pp = page_lookup(vp, off, rw == S_CREATE ? SE_EXCL : SE_SHARED)) {
1852                 if (pl) {
1853                         pl[0] = pp;
1854                         pl[1] = NULL;
1855                 } else {
1856                         page_unlock(pp);
1857                 }
1858         } else {
1859                 pp = page_create_va(vp, off, PAGESIZE,
1860                     PG_WAIT | PG_EXCL, seg, addr);
1861                 /*
1862                  * Someone raced in and created the page after we did the
1863                  * lookup but before we did the create, so go back and
1864                  * try to look it up again.
1865                  */
1866                 if (pp == NULL)
1867                         goto again;
1868                 /*
1869                  * Fill page from backing store, if any. If none, then
1870                  * either this is a newly filled hole or page must have
1871                  * been unmodified and freed so just zero it out.
1872                  */
1873                 err = swap_getphysname(vp, off, &pvp, &poff);
1874                 if (err) {
1875                         panic("tmp_getapage: no anon slot vp %p "
1876                             "off %llx pp %p\n", (void *)vp, off, (void *)pp);
1877                 }
1878                 if (pvp) {
1879                         flags = (pl == NULL ? B_ASYNC|B_READ : B_READ);
1880                         err = VOP_PAGEIO(pvp, pp, (u_offset_t)poff, PAGESIZE,
1881                             flags, cr, NULL);
1882                         if (flags & B_ASYNC)
1883                                 pp = NULL;
1884                 } else if (rw != S_CREATE) {
1885                         pagezero(pp, 0, PAGESIZE);
1886                 }
1887                 if (err && pp)
1888                         pvn_read_done(pp, B_ERROR);
1889                 if (err == 0) {
1890                         if (pl)
1891                                 pvn_plist_init(pp, pl, plsz, off, PAGESIZE, rw);
1892                         else
1893                                 pvn_io_done(pp);
1894                 }
1895         }
1896         return (err);
1897 }
1898 
1899 
1900 /*
1901  * Flags are composed of {B_INVAL, B_DIRTY B_FREE, B_DONTNEED}.
1902  * If len == 0, do from off to EOF.
1903  */
1904 static int tmp_nopage = 0;      /* Don't do tmp_putpage's if set */
1905 
1906 /* ARGSUSED */
1907 int
1908 tmp_putpage(
1909         register struct vnode *vp,
1910         offset_t off,
1911         size_t len,
1912         int flags,
1913         struct cred *cr,
1914         caller_context_t *ct)
1915 {
1916         register page_t *pp;
1917         u_offset_t io_off;
1918         size_t io_len = 0;
1919         int err = 0;
1920         struct tmpnode *tp = VTOTN(vp);
1921         int dolock;
1922 
1923         if (tmp_nopage)
1924                 return (0);
1925 
1926         ASSERT(vp->v_count != 0);
1927 
1928         if (vp->v_flag & VNOMAP)
1929                 return (ENOSYS);
1930 
1931         /*
1932          * This being tmpfs, we don't ever do i/o unless we really
1933          * have to (when we're low on memory and pageout calls us
1934          * with B_ASYNC | B_FREE or the user explicitly asks for it with
1935          * B_DONTNEED).
1936          * XXX to approximately track the mod time like ufs we should
1937          * update the times here. The problem is, once someone does a
1938          * store we never clear the mod bit and do i/o, thus fsflush
1939          * will keep calling us every 30 seconds to do the i/o and we'll
1940          * continually update the mod time. At least we update the mod
1941          * time on the first store because this results in a call to getpage.
1942          */
1943         if (flags != (B_ASYNC | B_FREE) && (flags & B_INVAL) == 0 &&
1944             (flags & B_DONTNEED) == 0)
1945                 return (0);
1946         /*
1947          * If this thread owns the lock, i.e., this thread grabbed it
1948          * as writer somewhere above, then we don't need to grab the
1949          * lock as reader in this routine.
1950          */
1951         dolock = (rw_owner(&tp->tn_contents) != curthread);
1952 
1953         /*
1954          * If this is pageout don't block on the lock as you could deadlock
1955          * when freemem == 0 (another thread has the read lock and is blocked
1956          * creating a page, and a third thread is waiting to get the writers
1957          * lock - waiting writers priority blocks us from getting the read
1958          * lock). Of course, if the only freeable pages are on this tmpnode
1959          * we're hosed anyways. A better solution might be a new lock type.
1960          * Note: ufs has the same problem.
1961          */
1962         if (curproc == proc_pageout) {
1963                 if (!rw_tryenter(&tp->tn_contents, RW_READER))
1964                         return (ENOMEM);
1965         } else if (dolock)
1966                 rw_enter(&tp->tn_contents, RW_READER);
1967 
1968         if (!vn_has_cached_data(vp))
1969                 goto out;
1970 
1971         if (len == 0) {
1972                 if (curproc == proc_pageout) {
1973                         panic("tmp: pageout can't block");
1974                         /*NOTREACHED*/
1975                 }
1976 
1977                 /* Search the entire vp list for pages >= off. */
1978                 err = pvn_vplist_dirty(vp, (u_offset_t)off, tmp_putapage,
1979                     flags, cr);
1980         } else {
1981                 u_offset_t eoff;
1982 
1983                 /*
1984                  * Loop over all offsets in the range [off...off + len]
1985                  * looking for pages to deal with.
1986                  */
1987                 eoff = MIN(off + len, tp->tn_size);
1988                 for (io_off = off; io_off < eoff; io_off += io_len) {
1989                         /*
1990                          * If we are not invalidating, synchronously
1991                          * freeing or writing pages use the routine
1992                          * page_lookup_nowait() to prevent reclaiming
1993                          * them from the free list.
1994                          */
1995                         if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
1996                                 pp = page_lookup(vp, io_off,
1997                                     (flags & (B_INVAL | B_FREE)) ?
1998                                     SE_EXCL : SE_SHARED);
1999                         } else {
2000                                 pp = page_lookup_nowait(vp, io_off,
2001                                     (flags & B_FREE) ? SE_EXCL : SE_SHARED);
2002                         }
2003 
2004                         if (pp == NULL || pvn_getdirty(pp, flags) == 0)
2005                                 io_len = PAGESIZE;
2006                         else {
2007                                 err = tmp_putapage(vp, pp, &io_off, &io_len,
2008                                     flags, cr);
2009                                 if (err != 0)
2010                                         break;
2011                         }
2012                 }
2013         }
2014         /* If invalidating, verify all pages on vnode list are gone. */
2015         if (err == 0 && off == 0 && len == 0 &&
2016             (flags & B_INVAL) && vn_has_cached_data(vp)) {
2017                 panic("tmp_putpage: B_INVAL, pages not gone");
2018                 /*NOTREACHED*/
2019         }
2020 out:
2021         if ((curproc == proc_pageout) || dolock)
2022                 rw_exit(&tp->tn_contents);
2023         /*
2024          * Only reason putapage is going to give us SE_NOSWAP as error
2025          * is when we ask a page to be written to physical backing store
2026          * and there is none. Ignore this because we might be dealing
2027          * with a swap page which does not have any backing store
2028          * on disk. In any other case we won't get this error over here.
2029          */
2030         if (err == SE_NOSWAP)
2031                 err = 0;
2032         return (err);
2033 }
2034 
2035 long tmp_putpagecnt, tmp_pagespushed;
2036 
2037 /*
2038  * Write out a single page.
2039  * For tmpfs this means choose a physical swap slot and write the page
2040  * out using VOP_PAGEIO. For performance, we attempt to kluster; i.e.,
2041  * we try to find a bunch of other dirty pages adjacent in the file
2042  * and a bunch of contiguous swap slots, and then write all the pages
2043  * out in a single i/o.
2044  */
2045 /*ARGSUSED*/
2046 static int
2047 tmp_putapage(
2048         struct vnode *vp,
2049         page_t *pp,
2050         u_offset_t *offp,
2051         size_t *lenp,
2052         int flags,
2053         struct cred *cr)
2054 {
2055         int err;
2056         ulong_t klstart, kllen;
2057         page_t *pplist, *npplist;
2058         extern int klustsize;
2059         long tmp_klustsize;
2060         struct tmpnode *tp;
2061         size_t pp_off, pp_len;
2062         u_offset_t io_off;
2063         size_t io_len;
2064         struct vnode *pvp;
2065         u_offset_t pstart;
2066         u_offset_t offset;
2067         u_offset_t tmpoff;
2068 
2069         ASSERT(PAGE_LOCKED(pp));
2070 
2071         /* Kluster in tmp_klustsize chunks */
2072         tp = VTOTN(vp);
2073         tmp_klustsize = klustsize;
2074         offset = pp->p_offset;
2075         klstart = (offset / tmp_klustsize) * tmp_klustsize;
2076         kllen = MIN(tmp_klustsize, tp->tn_size - klstart);
2077 
2078         /* Get a kluster of pages */
2079         pplist =
2080             pvn_write_kluster(vp, pp, &tmpoff, &pp_len, klstart, kllen, flags);
2081 
2082         pp_off = (size_t)tmpoff;
2083 
2084         /*
2085          * Get a cluster of physical offsets for the pages; the amount we
2086          * get may be some subrange of what we ask for (io_off, io_len).
2087          */
2088         io_off = pp_off;
2089         io_len = pp_len;
2090         err = swap_newphysname(vp, offset, &io_off, &io_len, &pvp, &pstart);
2091         ASSERT(err != SE_NOANON); /* anon slot must have been filled */
2092         if (err) {
2093                 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2094                 /*
2095                  * If this routine is called as a result of segvn_sync
2096                  * operation and we have no physical swap then we can get an
2097                  * error here. In such case we would return SE_NOSWAP as error.
2098                  * At this point, we expect only SE_NOSWAP.
2099                  */
2100                 ASSERT(err == SE_NOSWAP);
2101                 if (flags & B_INVAL)
2102                         err = ENOMEM;
2103                 goto out;
2104         }
2105         ASSERT(pp_off <= io_off && io_off + io_len <= pp_off + pp_len);
2106         ASSERT(io_off <= offset && offset < io_off + io_len);
2107 
2108         /* Toss pages at front/rear that we couldn't get physical backing for */
2109         if (io_off != pp_off) {
2110                 npplist = NULL;
2111                 page_list_break(&pplist, &npplist, btop(io_off - pp_off));
2112                 ASSERT(pplist->p_offset == pp_off);
2113                 ASSERT(pplist->p_prev->p_offset == io_off - PAGESIZE);
2114                 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2115                 pplist = npplist;
2116         }
2117         if (io_off + io_len < pp_off + pp_len) {
2118                 npplist = NULL;
2119                 page_list_break(&pplist, &npplist, btop(io_len));
2120                 ASSERT(npplist->p_offset == io_off + io_len);
2121                 ASSERT(npplist->p_prev->p_offset == pp_off + pp_len - PAGESIZE);
2122                 pvn_write_done(npplist, B_ERROR | B_WRITE | flags);
2123         }
2124 
2125         ASSERT(pplist->p_offset == io_off);
2126         ASSERT(pplist->p_prev->p_offset == io_off + io_len - PAGESIZE);
2127         ASSERT(btopr(io_len) <= btopr(kllen));
2128 
2129         /* Do i/o on the remaining kluster */
2130         err = VOP_PAGEIO(pvp, pplist, (u_offset_t)pstart, io_len,
2131             B_WRITE | flags, cr, NULL);
2132 
2133         if ((flags & B_ASYNC) == 0) {
2134                 pvn_write_done(pplist, ((err) ? B_ERROR : 0) | B_WRITE | flags);
2135         }
2136 out:
2137         if (!err) {
2138                 if (offp)
2139                         *offp = io_off;
2140                 if (lenp)
2141                         *lenp = io_len;
2142                 tmp_putpagecnt++;
2143                 tmp_pagespushed += btop(io_len);
2144         }
2145         if (err && err != ENOMEM && err != SE_NOSWAP)
2146                 cmn_err(CE_WARN, "tmp_putapage: err %d\n", err);
2147         return (err);
2148 }
2149 
2150 /* ARGSUSED */
2151 static int
2152 tmp_map(
2153         struct vnode *vp,
2154         offset_t off,
2155         struct as *as,
2156         caddr_t *addrp,
2157         size_t len,
2158         uchar_t prot,
2159         uchar_t maxprot,
2160         uint_t flags,
2161         struct cred *cred,
2162         caller_context_t *ct)
2163 {
2164         struct segvn_crargs vn_a;
2165         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
2166         int error;
2167 
2168 #ifdef _ILP32
2169         if (len > MAXOFF_T)
2170                 return (ENOMEM);
2171 #endif
2172 
2173         if (vp->v_flag & VNOMAP)
2174                 return (ENOSYS);
2175 
2176         if (off < 0 || (offset_t)(off + len) < 0 ||
2177             off > MAXOFF_T || (off + len) > MAXOFF_T)
2178                 return (ENXIO);
2179 
2180         if (vp->v_type != VREG)
2181                 return (ENODEV);
2182 
2183         /*
2184          * Don't allow mapping to locked file
2185          */
2186         if (vn_has_mandatory_locks(vp, tp->tn_mode)) {
2187                 return (EAGAIN);
2188         }
2189 
2190         as_rangelock(as);
2191         error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
2192         if (error != 0) {
2193                 as_rangeunlock(as);
2194                 return (error);
2195         }
2196 
2197         vn_a.vp = vp;
2198         vn_a.offset = (u_offset_t)off;
2199         vn_a.type = flags & MAP_TYPE;
2200         vn_a.prot = prot;
2201         vn_a.maxprot = maxprot;
2202         vn_a.flags = flags & ~MAP_TYPE;
2203         vn_a.cred = cred;
2204         vn_a.amp = NULL;
2205         vn_a.szc = 0;
2206         vn_a.lgrp_mem_policy_flags = 0;
2207 
2208         error = as_map(as, *addrp, len, segvn_create, &vn_a);
2209         as_rangeunlock(as);
2210         return (error);
2211 }
2212 
2213 /*
2214  * tmp_addmap and tmp_delmap can't be called since the vp
2215  * maintained in the segvn mapping is NULL.
2216  */
2217 /* ARGSUSED */
2218 static int
2219 tmp_addmap(
2220         struct vnode *vp,
2221         offset_t off,
2222         struct as *as,
2223         caddr_t addr,
2224         size_t len,
2225         uchar_t prot,
2226         uchar_t maxprot,
2227         uint_t flags,
2228         struct cred *cred,
2229         caller_context_t *ct)
2230 {
2231         return (0);
2232 }
2233 
2234 /* ARGSUSED */
2235 static int
2236 tmp_delmap(
2237         struct vnode *vp,
2238         offset_t off,
2239         struct as *as,
2240         caddr_t addr,
2241         size_t len,
2242         uint_t prot,
2243         uint_t maxprot,
2244         uint_t flags,
2245         struct cred *cred,
2246         caller_context_t *ct)
2247 {
2248         return (0);
2249 }
2250 
2251 static int
2252 tmp_freesp(struct vnode *vp, struct flock64 *lp, int flag)
2253 {
2254         register int i;
2255         register struct tmpnode *tp = VTOTN(vp);
2256         int error;
2257 
2258         ASSERT(vp->v_type == VREG);
2259         ASSERT(lp->l_start >= 0);
2260 
2261         if (lp->l_len != 0)
2262                 return (EINVAL);
2263 
2264         rw_enter(&tp->tn_rwlock, RW_WRITER);
2265         if (tp->tn_size == lp->l_start) {
2266                 rw_exit(&tp->tn_rwlock);
2267                 return (0);
2268         }
2269 
2270         /*
2271          * Check for any mandatory locks on the range
2272          */
2273         if (MANDLOCK(vp, tp->tn_mode)) {
2274                 long save_start;
2275 
2276                 save_start = lp->l_start;
2277 
2278                 if (tp->tn_size < lp->l_start) {
2279                         /*
2280                          * "Truncate up" case: need to make sure there
2281                          * is no lock beyond current end-of-file. To
2282                          * do so, we need to set l_start to the size
2283                          * of the file temporarily.
2284                          */
2285                         lp->l_start = tp->tn_size;
2286                 }
2287                 lp->l_type = F_WRLCK;
2288                 lp->l_sysid = 0;
2289                 lp->l_pid = ttoproc(curthread)->p_pid;
2290                 i = (flag & (FNDELAY|FNONBLOCK)) ? 0 : SLPFLCK;
2291                 if ((i = reclock(vp, lp, i, 0, lp->l_start, NULL)) != 0 ||
2292                     lp->l_type != F_UNLCK) {
2293                         rw_exit(&tp->tn_rwlock);
2294                         return (i ? i : EAGAIN);
2295                 }
2296 
2297                 lp->l_start = save_start;
2298         }
2299         VFSTOTM(vp->v_vfsp);
2300 
2301         rw_enter(&tp->tn_contents, RW_WRITER);
2302         error = tmpnode_trunc((struct tmount *)VFSTOTM(vp->v_vfsp),
2303             tp, (ulong_t)lp->l_start);
2304         rw_exit(&tp->tn_contents);
2305         rw_exit(&tp->tn_rwlock);
2306         return (error);
2307 }
2308 
2309 /* ARGSUSED */
2310 static int
2311 tmp_space(
2312         struct vnode *vp,
2313         int cmd,
2314         struct flock64 *bfp,
2315         int flag,
2316         offset_t offset,
2317         cred_t *cred,
2318         caller_context_t *ct)
2319 {
2320         int error;
2321 
2322         if (cmd != F_FREESP)
2323                 return (EINVAL);
2324         if ((error = convoff(vp, bfp, 0, (offset_t)offset)) == 0) {
2325                 if ((bfp->l_start > MAXOFF_T) || (bfp->l_len > MAXOFF_T))
2326                         return (EFBIG);
2327                 error = tmp_freesp(vp, bfp, flag);
2328         }
2329         return (error);
2330 }
2331 
2332 /* ARGSUSED */
2333 static int
2334 tmp_seek(
2335         struct vnode *vp,
2336         offset_t ooff,
2337         offset_t *noffp,
2338         caller_context_t *ct)
2339 {
2340         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
2341 }
2342 
2343 /* ARGSUSED2 */
2344 static int
2345 tmp_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2346 {
2347         struct tmpnode *tp = VTOTN(vp);
2348 
2349         if (write_lock) {
2350                 rw_enter(&tp->tn_rwlock, RW_WRITER);
2351         } else {
2352                 rw_enter(&tp->tn_rwlock, RW_READER);
2353         }
2354         return (write_lock);
2355 }
2356 
2357 /* ARGSUSED1 */
2358 static void
2359 tmp_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2360 {
2361         struct tmpnode *tp = VTOTN(vp);
2362 
2363         rw_exit(&tp->tn_rwlock);
2364 }
2365 
2366 static int
2367 tmp_pathconf(
2368         struct vnode *vp,
2369         int cmd,
2370         ulong_t *valp,
2371         cred_t *cr,
2372         caller_context_t *ct)
2373 {
2374         struct tmpnode *tp = NULL;
2375         int error;
2376 
2377         switch (cmd) {
2378         case _PC_XATTR_EXISTS:
2379                 if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
2380                         *valp = 0;      /* assume no attributes */
2381                         error = 0;      /* okay to ask */
2382                         tp = VTOTN(vp);
2383                         rw_enter(&tp->tn_rwlock, RW_READER);
2384                         if (tp->tn_xattrdp) {
2385                                 rw_enter(&tp->tn_xattrdp->tn_rwlock, RW_READER);
2386                                 /* do not count "." and ".." */
2387                                 if (tp->tn_xattrdp->tn_dirents > 2)
2388                                         *valp = 1;
2389                                 rw_exit(&tp->tn_xattrdp->tn_rwlock);
2390                         }
2391                         rw_exit(&tp->tn_rwlock);
2392                 } else {
2393                         error = EINVAL;
2394                 }
2395                 break;
2396         case _PC_SATTR_ENABLED:
2397         case _PC_SATTR_EXISTS:
2398                 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2399                     (vp->v_type == VREG || vp->v_type == VDIR);
2400                 error = 0;
2401                 break;
2402         case _PC_TIMESTAMP_RESOLUTION:
2403                 /* nanosecond timestamp resolution */
2404                 *valp = 1L;
2405                 error = 0;
2406                 break;
2407         default:
2408                 error = fs_pathconf(vp, cmd, valp, cr, ct);
2409         }
2410         return (error);
2411 }
2412 
2413 
2414 struct vnodeops *tmp_vnodeops;
2415 
2416 const fs_operation_def_t tmp_vnodeops_template[] = {
2417         VOPNAME_OPEN,           { .vop_open = tmp_open },
2418         VOPNAME_CLOSE,          { .vop_close = tmp_close },
2419         VOPNAME_READ,           { .vop_read = tmp_read },
2420         VOPNAME_WRITE,          { .vop_write = tmp_write },
2421         VOPNAME_IOCTL,          { .vop_ioctl = tmp_ioctl },
2422         VOPNAME_GETATTR,        { .vop_getattr = tmp_getattr },
2423         VOPNAME_SETATTR,        { .vop_setattr = tmp_setattr },
2424         VOPNAME_ACCESS,         { .vop_access = tmp_access },
2425         VOPNAME_LOOKUP,         { .vop_lookup = tmp_lookup },
2426         VOPNAME_CREATE,         { .vop_create = tmp_create },
2427         VOPNAME_REMOVE,         { .vop_remove = tmp_remove },
2428         VOPNAME_LINK,           { .vop_link = tmp_link },
2429         VOPNAME_RENAME,         { .vop_rename = tmp_rename },
2430         VOPNAME_MKDIR,          { .vop_mkdir = tmp_mkdir },
2431         VOPNAME_RMDIR,          { .vop_rmdir = tmp_rmdir },
2432         VOPNAME_READDIR,        { .vop_readdir = tmp_readdir },
2433         VOPNAME_SYMLINK,        { .vop_symlink = tmp_symlink },
2434         VOPNAME_READLINK,       { .vop_readlink = tmp_readlink },
2435         VOPNAME_FSYNC,          { .vop_fsync = tmp_fsync },
2436         VOPNAME_INACTIVE,       { .vop_inactive = tmp_inactive },
2437         VOPNAME_FID,            { .vop_fid = tmp_fid },
2438         VOPNAME_RWLOCK,         { .vop_rwlock = tmp_rwlock },
2439         VOPNAME_RWUNLOCK,       { .vop_rwunlock = tmp_rwunlock },
2440         VOPNAME_SEEK,           { .vop_seek = tmp_seek },
2441         VOPNAME_SPACE,          { .vop_space = tmp_space },
2442         VOPNAME_GETPAGE,        { .vop_getpage = tmp_getpage },
2443         VOPNAME_PUTPAGE,        { .vop_putpage = tmp_putpage },
2444         VOPNAME_MAP,            { .vop_map = tmp_map },
2445         VOPNAME_ADDMAP,         { .vop_addmap = tmp_addmap },
2446         VOPNAME_DELMAP,         { .vop_delmap = tmp_delmap },
2447         VOPNAME_PATHCONF,       { .vop_pathconf = tmp_pathconf },
2448         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
2449         NULL,                   NULL
2450 };