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 2016, Joyent, Inc.
  29  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  30  * Copyright 2016 RackTop Systems.
  31  */
  32 
  33 #include <sys/types.h>
  34 #include <sys/param.h>
  35 #include <sys/t_lock.h>
  36 #include <sys/systm.h>
  37 #include <sys/sysmacros.h>
  38 #include <sys/user.h>
  39 #include <sys/time.h>
  40 #include <sys/vfs.h>
  41 #include <sys/vfs_opreg.h>
  42 #include <sys/vnode.h>
  43 #include <sys/file.h>
  44 #include <sys/fcntl.h>
  45 #include <sys/flock.h>
  46 #include <sys/kmem.h>
  47 #include <sys/uio.h>
  48 #include <sys/errno.h>
  49 #include <sys/stat.h>
  50 #include <sys/cred.h>
  51 #include <sys/dirent.h>
  52 #include <sys/pathname.h>
  53 #include <sys/vmsystm.h>
  54 #include <sys/fs/tmp.h>
  55 #include <sys/fs/tmpnode.h>
  56 #include <sys/mman.h>
  57 #include <vm/hat.h>
  58 #include <vm/seg_vn.h>
  59 #include <vm/seg_map.h>
  60 #include <vm/seg.h>
  61 #include <vm/anon.h>
  62 #include <vm/as.h>
  63 #include <vm/page.h>
  64 #include <vm/pvn.h>
  65 #include <sys/cmn_err.h>
  66 #include <sys/debug.h>
  67 #include <sys/swap.h>
  68 #include <sys/buf.h>
  69 #include <sys/vm.h>
  70 #include <sys/vtrace.h>
  71 #include <sys/policy.h>
  72 #include <fs/fs_subr.h>
  73 
  74 static int      tmp_getapage(struct vnode *, u_offset_t, size_t, uint_t *,
  75         page_t **, size_t, struct seg *, caddr_t, enum seg_rw, struct cred *);
  76 static int      tmp_putapage(struct vnode *, page_t *, u_offset_t *, size_t *,
  77         int, struct cred *);
  78 
  79 /* ARGSUSED1 */
  80 static int
  81 tmp_open(struct vnode **vpp, int flag, struct cred *cred, caller_context_t *ct)
  82 {
  83         /*
  84          * swapon to a tmpfs file is not supported so access
  85          * is denied on open if VISSWAP is set.
  86          */
  87         if ((*vpp)->v_flag & VISSWAP)
  88                 return (EINVAL);
  89         return (0);
  90 }
  91 
  92 /* ARGSUSED1 */
  93 static int
  94 tmp_close(
  95         struct vnode *vp,
  96         int flag,
  97         int count,
  98         offset_t offset,
  99         struct cred *cred,
 100         caller_context_t *ct)
 101 {
 102         cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
 103         cleanshares(vp, ttoproc(curthread)->p_pid);
 104         return (0);
 105 }
 106 
 107 /*
 108  * wrtmp does the real work of write requests for tmpfs.
 109  */
 110 static int
 111 wrtmp(
 112         struct tmount *tm,
 113         struct tmpnode *tp,
 114         struct uio *uio,
 115         struct cred *cr,
 116         struct caller_context *ct)
 117 {
 118         pgcnt_t pageoffset;     /* offset in pages */
 119         ulong_t segmap_offset;  /* pagesize byte offset into segmap */
 120         caddr_t base;           /* base of segmap */
 121         ssize_t bytes;          /* bytes to uiomove */
 122         pfn_t pagenumber;       /* offset in pages into tmp file */
 123         struct vnode *vp;
 124         int error = 0;
 125         int     pagecreate;     /* == 1 if we allocated a page */
 126         int     newpage;
 127         rlim64_t limit = uio->uio_llimit;
 128         long oresid = uio->uio_resid;
 129         timestruc_t now;
 130 
 131         long tn_size_changed = 0;
 132         long old_tn_size;
 133         long new_tn_size;
 134 
 135         vp = TNTOV(tp);
 136         ASSERT(vp->v_type == VREG);
 137 
 138         TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START,
 139             "tmp_wrtmp_start:vp %p", vp);
 140 
 141         ASSERT(RW_WRITE_HELD(&tp->tn_contents));
 142         ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
 143 
 144         if (MANDLOCK(vp, tp->tn_mode)) {
 145                 rw_exit(&tp->tn_contents);
 146                 /*
 147                  * tmp_getattr ends up being called by chklock
 148                  */
 149                 error = chklock(vp, FWRITE, uio->uio_loffset, uio->uio_resid,
 150                     uio->uio_fmode, ct);
 151                 rw_enter(&tp->tn_contents, RW_WRITER);
 152                 if (error != 0) {
 153                         TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 154                             "tmp_wrtmp_end:vp %p error %d", vp, error);
 155                         return (error);
 156                 }
 157         }
 158 
 159         if (uio->uio_loffset < 0)
 160                 return (EINVAL);
 161 
 162         if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
 163                 limit = MAXOFFSET_T;
 164 
 165         if (uio->uio_loffset >= limit) {
 166                 proc_t *p = ttoproc(curthread);
 167 
 168                 mutex_enter(&p->p_lock);
 169                 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
 170                     p, RCA_UNSAFE_SIGINFO);
 171                 mutex_exit(&p->p_lock);
 172                 return (EFBIG);
 173         }
 174 
 175         if (uio->uio_loffset >= MAXOFF_T) {
 176                 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 177                     "tmp_wrtmp_end:vp %p error %d", vp, EINVAL);
 178                 return (EFBIG);
 179         }
 180 
 181         if (uio->uio_resid == 0) {
 182                 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 183                     "tmp_wrtmp_end:vp %p error %d", vp, 0);
 184                 return (0);
 185         }
 186 
 187         if (limit > MAXOFF_T)
 188                 limit = MAXOFF_T;
 189 
 190         do {
 191                 long    offset;
 192                 long    delta;
 193 
 194                 offset = (long)uio->uio_offset;
 195                 pageoffset = offset & PAGEOFFSET;
 196                 /*
 197                  * A maximum of PAGESIZE bytes of data is transferred
 198                  * each pass through this loop
 199                  */
 200                 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
 201 
 202                 if (offset + bytes >= limit) {
 203                         if (offset >= limit) {
 204                                 error = EFBIG;
 205                                 goto out;
 206                         }
 207                         bytes = limit - offset;
 208                 }
 209                 pagenumber = btop(offset);
 210 
 211                 /*
 212                  * delta is the amount of anonymous memory
 213                  * to reserve for the file.
 214                  * We always reserve in pagesize increments so
 215                  * unless we're extending the file into a new page,
 216                  * we don't need to call tmp_resv.
 217                  */
 218                 delta = offset + bytes -
 219                     P2ROUNDUP_TYPED(tp->tn_size, PAGESIZE, u_offset_t);
 220                 if (delta > 0) {
 221                         pagecreate = 1;
 222                         if (tmp_resv(tm, tp, delta, pagecreate)) {
 223                                 /*
 224                                  * Log file system full in the zone that owns
 225                                  * the tmpfs mount, as well as in the global
 226                                  * zone if necessary.
 227                                  */
 228                                 zcmn_err(tm->tm_vfsp->vfs_zone->zone_id,
 229                                     CE_WARN, "%s: File system full, "
 230                                     "swap space limit exceeded",
 231                                     tm->tm_mntpath);
 232 
 233                                 if (tm->tm_vfsp->vfs_zone->zone_id !=
 234                                     GLOBAL_ZONEID) {
 235 
 236                                         vfs_t *vfs = tm->tm_vfsp;
 237 
 238                                         zcmn_err(GLOBAL_ZONEID,
 239                                             CE_WARN, "%s: File system full, "
 240                                             "swap space limit exceeded",
 241                                             vfs->vfs_vnodecovered->v_path);
 242                                 }
 243                                 error = ENOSPC;
 244                                 break;
 245                         }
 246                         tmpnode_growmap(tp, (ulong_t)offset + bytes);
 247                 }
 248                 /* grow the file to the new length */
 249                 if (offset + bytes > tp->tn_size) {
 250                         tn_size_changed = 1;
 251                         old_tn_size = tp->tn_size;
 252                         /*
 253                          * Postpone updating tp->tn_size until uiomove() is
 254                          * done.
 255                          */
 256                         new_tn_size = offset + bytes;
 257                 }
 258                 if (bytes == PAGESIZE) {
 259                         /*
 260                          * Writing whole page so reading from disk
 261                          * is a waste
 262                          */
 263                         pagecreate = 1;
 264                 } else {
 265                         pagecreate = 0;
 266                 }
 267                 /*
 268                  * If writing past EOF or filling in a hole
 269                  * we need to allocate an anon slot.
 270                  */
 271                 if (anon_get_ptr(tp->tn_anon, pagenumber) == NULL) {
 272                         (void) anon_set_ptr(tp->tn_anon, pagenumber,
 273                             anon_alloc(vp, ptob(pagenumber)), ANON_SLEEP);
 274                         pagecreate = 1;
 275                         tp->tn_nblocks++;
 276                 }
 277 
 278                 /*
 279                  * We have to drop the contents lock to allow the VM
 280                  * system to reacquire it in tmp_getpage()
 281                  */
 282                 rw_exit(&tp->tn_contents);
 283 
 284                 /*
 285                  * Touch the page and fault it in if it is not in core
 286                  * before segmap_getmapflt or vpm_data_copy can lock it.
 287                  * This is to avoid the deadlock if the buffer is mapped
 288                  * to the same file through mmap which we want to write.
 289                  */
 290                 uio_prefaultpages((long)bytes, uio);
 291 
 292                 newpage = 0;
 293                 if (vpm_enable) {
 294                         /*
 295                          * Copy data. If new pages are created, part of
 296                          * the page that is not written will be initizliazed
 297                          * with zeros.
 298                          */
 299                         error = vpm_data_copy(vp, offset, bytes, uio,
 300                             !pagecreate, &newpage, 1, S_WRITE);
 301                 } else {
 302                         /* Get offset within the segmap mapping */
 303                         segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
 304                         base = segmap_getmapflt(segkmap, vp,
 305                             (offset &  MAXBMASK), PAGESIZE, !pagecreate,
 306                             S_WRITE);
 307                 }
 308 
 309 
 310                 if (!vpm_enable && pagecreate) {
 311                         /*
 312                          * segmap_pagecreate() returns 1 if it calls
 313                          * page_create_va() to allocate any pages.
 314                          */
 315                         newpage = segmap_pagecreate(segkmap,
 316                             base + segmap_offset, (size_t)PAGESIZE, 0);
 317                         /*
 318                          * Clear from the beginning of the page to the starting
 319                          * offset of the data.
 320                          */
 321                         if (pageoffset != 0)
 322                                 (void) kzero(base + segmap_offset,
 323                                     (size_t)pageoffset);
 324                 }
 325 
 326                 if (!vpm_enable) {
 327                         error = uiomove(base + segmap_offset + pageoffset,
 328                             (long)bytes, UIO_WRITE, uio);
 329                 }
 330 
 331                 if (!vpm_enable && pagecreate &&
 332                     uio->uio_offset < P2ROUNDUP(offset + bytes, PAGESIZE)) {
 333                         long    zoffset; /* zero from offset into page */
 334                         /*
 335                          * We created pages w/o initializing them completely,
 336                          * thus we need to zero the part that wasn't set up.
 337                          * This happens on most EOF write cases and if
 338                          * we had some sort of error during the uiomove.
 339                          */
 340                         long nmoved;
 341 
 342                         nmoved = uio->uio_offset - offset;
 343                         ASSERT((nmoved + pageoffset) <= PAGESIZE);
 344 
 345                         /*
 346                          * Zero from the end of data in the page to the
 347                          * end of the page.
 348                          */
 349                         if ((zoffset = pageoffset + nmoved) < PAGESIZE)
 350                                 (void) kzero(base + segmap_offset + zoffset,
 351                                     (size_t)PAGESIZE - zoffset);
 352                 }
 353 
 354                 /*
 355                  * Unlock the pages which have been allocated by
 356                  * page_create_va() in segmap_pagecreate()
 357                  */
 358                 if (!vpm_enable && newpage) {
 359                         segmap_pageunlock(segkmap, base + segmap_offset,
 360                             (size_t)PAGESIZE, S_WRITE);
 361                 }
 362 
 363                 if (error) {
 364                         /*
 365                          * If we failed on a write, we must
 366                          * be sure to invalidate any pages that may have
 367                          * been allocated.
 368                          */
 369                         if (vpm_enable) {
 370                                 (void) vpm_sync_pages(vp, offset, PAGESIZE,
 371                                     SM_INVAL);
 372                         } else {
 373                                 (void) segmap_release(segkmap, base, SM_INVAL);
 374                         }
 375                 } else {
 376                         if (vpm_enable) {
 377                                 error = vpm_sync_pages(vp, offset, PAGESIZE,
 378                                     0);
 379                         } else {
 380                                 error = segmap_release(segkmap, base, 0);
 381                         }
 382                 }
 383 
 384                 /*
 385                  * Re-acquire contents lock.
 386                  */
 387                 rw_enter(&tp->tn_contents, RW_WRITER);
 388 
 389                 /*
 390                  * Update tn_size.
 391                  */
 392                 if (tn_size_changed)
 393                         tp->tn_size = new_tn_size;
 394 
 395                 /*
 396                  * If the uiomove failed, fix up tn_size.
 397                  */
 398                 if (error) {
 399                         if (tn_size_changed) {
 400                                 /*
 401                                  * The uiomove failed, and we
 402                                  * allocated blocks,so get rid
 403                                  * of them.
 404                                  */
 405                                 (void) tmpnode_trunc(tm, tp,
 406                                     (ulong_t)old_tn_size);
 407                         }
 408                 } else {
 409                         /*
 410                          * XXX - Can this be out of the loop?
 411                          */
 412                         if ((tp->tn_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) &&
 413                             (tp->tn_mode & (S_ISUID | S_ISGID)) &&
 414                             secpolicy_vnode_setid_retain(cr,
 415                             (tp->tn_mode & S_ISUID) != 0 && tp->tn_uid == 0)) {
 416                                 /*
 417                                  * Clear Set-UID & Set-GID bits on
 418                                  * successful write if not privileged
 419                                  * and at least one of the execute bits
 420                                  * is set.  If we always clear Set-GID,
 421                                  * mandatory file and record locking is
 422                                  * unuseable.
 423                                  */
 424                                 tp->tn_mode &= ~(S_ISUID | S_ISGID);
 425                         }
 426                         gethrestime(&now);
 427                         tp->tn_mtime = now;
 428                         tp->tn_ctime = now;
 429                 }
 430         } while (error == 0 && uio->uio_resid > 0 && bytes != 0);
 431 
 432 out:
 433         /*
 434          * If we've already done a partial-write, terminate
 435          * the write but return no error.
 436          */
 437         if (oresid != uio->uio_resid)
 438                 error = 0;
 439         TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 440             "tmp_wrtmp_end:vp %p error %d", vp, error);
 441         return (error);
 442 }
 443 
 444 /*
 445  * rdtmp does the real work of read requests for tmpfs.
 446  */
 447 static int
 448 rdtmp(
 449         struct tmount *tm,
 450         struct tmpnode *tp,
 451         struct uio *uio,
 452         struct caller_context *ct)
 453 {
 454         ulong_t pageoffset;     /* offset in tmpfs file (uio_offset) */
 455         ulong_t segmap_offset;  /* pagesize byte offset into segmap */
 456         caddr_t base;           /* base of segmap */
 457         ssize_t bytes;          /* bytes to uiomove */
 458         struct vnode *vp;
 459         int error;
 460         long oresid = uio->uio_resid;
 461 
 462 #if defined(lint)
 463         tm = tm;
 464 #endif
 465         vp = TNTOV(tp);
 466 
 467         TRACE_1(TR_FAC_TMPFS, TR_TMPFS_RWTMP_START, "tmp_rdtmp_start:vp %p",
 468             vp);
 469 
 470         ASSERT(RW_LOCK_HELD(&tp->tn_contents));
 471 
 472         if (MANDLOCK(vp, tp->tn_mode)) {
 473                 rw_exit(&tp->tn_contents);
 474                 /*
 475                  * tmp_getattr ends up being called by chklock
 476                  */
 477                 error = chklock(vp, FREAD, uio->uio_loffset, uio->uio_resid,
 478                     uio->uio_fmode, ct);
 479                 rw_enter(&tp->tn_contents, RW_READER);
 480                 if (error != 0) {
 481                         TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 482                             "tmp_rdtmp_end:vp %p error %d", vp, error);
 483                         return (error);
 484                 }
 485         }
 486         ASSERT(tp->tn_type == VREG);
 487 
 488         if (uio->uio_loffset >= MAXOFF_T) {
 489                 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 490                     "tmp_rdtmp_end:vp %p error %d", vp, EINVAL);
 491                 return (0);
 492         }
 493         if (uio->uio_loffset < 0)
 494                 return (EINVAL);
 495         if (uio->uio_resid == 0) {
 496                 TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 497                     "tmp_rdtmp_end:vp %p error %d", vp, 0);
 498                 return (0);
 499         }
 500 
 501         vp = TNTOV(tp);
 502 
 503         do {
 504                 long diff;
 505                 long offset;
 506 
 507                 offset = uio->uio_offset;
 508                 pageoffset = offset & PAGEOFFSET;
 509                 bytes = MIN(PAGESIZE - pageoffset, uio->uio_resid);
 510 
 511                 diff = tp->tn_size - offset;
 512 
 513                 if (diff <= 0) {
 514                         error = 0;
 515                         goto out;
 516                 }
 517                 if (diff < bytes)
 518                         bytes = diff;
 519 
 520                 /*
 521                  * We have to drop the contents lock to allow the VM system
 522                  * to reacquire it in tmp_getpage() should the uiomove cause a
 523                  * pagefault.
 524                  */
 525                 rw_exit(&tp->tn_contents);
 526 
 527                 if (vpm_enable) {
 528                         /*
 529                          * Copy data.
 530                          */
 531                         error = vpm_data_copy(vp, offset, bytes, uio, 1, NULL,
 532                             0, S_READ);
 533                 } else {
 534                         segmap_offset = (offset & PAGEMASK) & MAXBOFFSET;
 535                         base = segmap_getmapflt(segkmap, vp, offset & MAXBMASK,
 536                             bytes, 1, S_READ);
 537 
 538                         error = uiomove(base + segmap_offset + pageoffset,
 539                             (long)bytes, UIO_READ, uio);
 540                 }
 541 
 542                 if (error) {
 543                         if (vpm_enable) {
 544                                 (void) vpm_sync_pages(vp, offset, PAGESIZE, 0);
 545                         } else {
 546                                 (void) segmap_release(segkmap, base, 0);
 547                         }
 548                 } else {
 549                         if (vpm_enable) {
 550                                 error = vpm_sync_pages(vp, offset, PAGESIZE,
 551                                     0);
 552                         } else {
 553                                 error = segmap_release(segkmap, base, 0);
 554                         }
 555                 }
 556 
 557                 /*
 558                  * Re-acquire contents lock.
 559                  */
 560                 rw_enter(&tp->tn_contents, RW_READER);
 561 
 562         } while (error == 0 && uio->uio_resid > 0);
 563 
 564 out:
 565         gethrestime(&tp->tn_atime);
 566 
 567         /*
 568          * If we've already done a partial read, terminate
 569          * the read but return no error.
 570          */
 571         if (oresid != uio->uio_resid)
 572                 error = 0;
 573 
 574         TRACE_2(TR_FAC_TMPFS, TR_TMPFS_RWTMP_END,
 575             "tmp_rdtmp_end:vp %x error %d", vp, error);
 576         return (error);
 577 }
 578 
 579 /* ARGSUSED2 */
 580 static int
 581 tmp_read(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cred,
 582     struct caller_context *ct)
 583 {
 584         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 585         struct tmount *tm = (struct tmount *)VTOTM(vp);
 586         int error;
 587 
 588         /*
 589          * We don't currently support reading non-regular files
 590          */
 591         if (vp->v_type == VDIR)
 592                 return (EISDIR);
 593         if (vp->v_type != VREG)
 594                 return (EINVAL);
 595         /*
 596          * tmp_rwlock should have already been called from layers above
 597          */
 598         ASSERT(RW_READ_HELD(&tp->tn_rwlock));
 599 
 600         rw_enter(&tp->tn_contents, RW_READER);
 601 
 602         error = rdtmp(tm, tp, uiop, ct);
 603 
 604         rw_exit(&tp->tn_contents);
 605 
 606         return (error);
 607 }
 608 
 609 static int
 610 tmp_write(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cred,
 611     struct caller_context *ct)
 612 {
 613         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 614         struct tmount *tm = (struct tmount *)VTOTM(vp);
 615         int error;
 616 
 617         /*
 618          * We don't currently support writing to non-regular files
 619          */
 620         if (vp->v_type != VREG)
 621                 return (EINVAL);        /* XXX EISDIR? */
 622 
 623         /*
 624          * tmp_rwlock should have already been called from layers above
 625          */
 626         ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
 627 
 628         rw_enter(&tp->tn_contents, RW_WRITER);
 629 
 630         if (ioflag & FAPPEND) {
 631                 /*
 632                  * In append mode start at end of file.
 633                  */
 634                 uiop->uio_loffset = tp->tn_size;
 635         }
 636 
 637         error = wrtmp(tm, tp, uiop, cred, ct);
 638 
 639         rw_exit(&tp->tn_contents);
 640 
 641         return (error);
 642 }
 643 
 644 /* ARGSUSED */
 645 static int
 646 tmp_ioctl(
 647         struct vnode *vp,
 648         int com,
 649         intptr_t data,
 650         int flag,
 651         struct cred *cred,
 652         int *rvalp,
 653         caller_context_t *ct)
 654 {
 655         return (ENOTTY);
 656 }
 657 
 658 /* ARGSUSED2 */
 659 static int
 660 tmp_getattr(
 661         struct vnode *vp,
 662         struct vattr *vap,
 663         int flags,
 664         struct cred *cred,
 665         caller_context_t *ct)
 666 {
 667         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 668         struct vnode *mvp;
 669         struct vattr va;
 670         int attrs = 1;
 671 
 672         /*
 673          * A special case to handle the root tnode on a diskless nfs
 674          * client who may have had its uid and gid inherited
 675          * from an nfs vnode with nobody ownership.  Likely the
 676          * root filesystem. After nfs is fully functional the uid/gid
 677          * may be mapable so ask again.
 678          * vfsp can't get unmounted because we hold vp.
 679          */
 680         if (vp->v_flag & VROOT &&
 681             (mvp = vp->v_vfsp->vfs_vnodecovered) != NULL) {
 682                 mutex_enter(&tp->tn_tlock);
 683                 if (tp->tn_uid == UID_NOBODY || tp->tn_gid == GID_NOBODY) {
 684                         mutex_exit(&tp->tn_tlock);
 685                         bzero(&va, sizeof (struct vattr));
 686                         va.va_mask = AT_UID|AT_GID;
 687                         attrs = VOP_GETATTR(mvp, &va, 0, cred, ct);
 688                 } else {
 689                         mutex_exit(&tp->tn_tlock);
 690                 }
 691         }
 692         mutex_enter(&tp->tn_tlock);
 693         if (attrs == 0) {
 694                 tp->tn_uid = va.va_uid;
 695                 tp->tn_gid = va.va_gid;
 696         }
 697         vap->va_type = vp->v_type;
 698         vap->va_mode = tp->tn_mode & MODEMASK;
 699         vap->va_uid = tp->tn_uid;
 700         vap->va_gid = tp->tn_gid;
 701         vap->va_fsid = tp->tn_fsid;
 702         vap->va_nodeid = (ino64_t)tp->tn_nodeid;
 703         vap->va_nlink = tp->tn_nlink;
 704         vap->va_size = (u_offset_t)tp->tn_size;
 705         vap->va_atime = tp->tn_atime;
 706         vap->va_mtime = tp->tn_mtime;
 707         vap->va_ctime = tp->tn_ctime;
 708         vap->va_blksize = PAGESIZE;
 709         vap->va_rdev = tp->tn_rdev;
 710         vap->va_seq = tp->tn_seq;
 711 
 712         /*
 713          * XXX Holes are not taken into account.  We could take the time to
 714          * run through the anon array looking for allocated slots...
 715          */
 716         vap->va_nblocks = (fsblkcnt64_t)btodb(ptob(btopr(vap->va_size)));
 717         mutex_exit(&tp->tn_tlock);
 718         return (0);
 719 }
 720 
 721 /*ARGSUSED4*/
 722 static int
 723 tmp_setattr(
 724         struct vnode *vp,
 725         struct vattr *vap,
 726         int flags,
 727         struct cred *cred,
 728         caller_context_t *ct)
 729 {
 730         struct tmount *tm = (struct tmount *)VTOTM(vp);
 731         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 732         int error = 0;
 733         struct vattr *get;
 734         long mask;
 735 
 736         /*
 737          * Cannot set these attributes
 738          */
 739         if ((vap->va_mask & AT_NOSET) || (vap->va_mask & AT_XVATTR))
 740                 return (EINVAL);
 741 
 742         mutex_enter(&tp->tn_tlock);
 743 
 744         get = &tp->tn_attr;
 745         /*
 746          * Change file access modes. Must be owner or have sufficient
 747          * privileges.
 748          */
 749         error = secpolicy_vnode_setattr(cred, vp, vap, get, flags, tmp_taccess,
 750             tp);
 751 
 752         if (error)
 753                 goto out;
 754 
 755         mask = vap->va_mask;
 756 
 757         if (mask & AT_MODE) {
 758                 get->va_mode &= S_IFMT;
 759                 get->va_mode |= vap->va_mode & ~S_IFMT;
 760         }
 761 
 762         if (mask & AT_UID)
 763                 get->va_uid = vap->va_uid;
 764         if (mask & AT_GID)
 765                 get->va_gid = vap->va_gid;
 766         if (mask & AT_ATIME)
 767                 get->va_atime = vap->va_atime;
 768         if (mask & AT_MTIME)
 769                 get->va_mtime = vap->va_mtime;
 770 
 771         if (mask & (AT_UID | AT_GID | AT_MODE | AT_MTIME))
 772                 gethrestime(&tp->tn_ctime);
 773 
 774         if (mask & AT_SIZE) {
 775                 ASSERT(vp->v_type != VDIR);
 776 
 777                 /* Don't support large files. */
 778                 if (vap->va_size > MAXOFF_T) {
 779                         error = EFBIG;
 780                         goto out;
 781                 }
 782                 mutex_exit(&tp->tn_tlock);
 783 
 784                 rw_enter(&tp->tn_rwlock, RW_WRITER);
 785                 rw_enter(&tp->tn_contents, RW_WRITER);
 786                 error = tmpnode_trunc(tm, tp, (ulong_t)vap->va_size);
 787                 rw_exit(&tp->tn_contents);
 788                 rw_exit(&tp->tn_rwlock);
 789 
 790                 if (error == 0) {
 791                         if (vap->va_size == 0) {
 792                                 vnevent_truncate(vp, ct);
 793                         } else {
 794                                 vnevent_resize(vp, ct);
 795                         }
 796                 }
 797 
 798                 goto out1;
 799         }
 800 out:
 801         mutex_exit(&tp->tn_tlock);
 802 out1:
 803         return (error);
 804 }
 805 
 806 /* ARGSUSED2 */
 807 static int
 808 tmp_access(
 809         struct vnode *vp,
 810         int mode,
 811         int flags,
 812         struct cred *cred,
 813         caller_context_t *ct)
 814 {
 815         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
 816         int error;
 817 
 818         mutex_enter(&tp->tn_tlock);
 819         error = tmp_taccess(tp, mode, cred);
 820         mutex_exit(&tp->tn_tlock);
 821         return (error);
 822 }
 823 
 824 /* ARGSUSED3 */
 825 static int
 826 tmp_lookup(
 827         struct vnode *dvp,
 828         char *nm,
 829         struct vnode **vpp,
 830         struct pathname *pnp,
 831         int flags,
 832         struct vnode *rdir,
 833         struct cred *cred,
 834         caller_context_t *ct,
 835         int *direntflags,
 836         pathname_t *realpnp)
 837 {
 838         struct tmpnode *tp = (struct tmpnode *)VTOTN(dvp);
 839         struct tmpnode *ntp = NULL;
 840         int error;
 841 
 842 
 843         /* allow cd into @ dir */
 844         if (flags & LOOKUP_XATTR) {
 845                 struct tmpnode *xdp;
 846                 struct tmount *tm;
 847 
 848                 /*
 849                  * don't allow attributes if not mounted XATTR support
 850                  */
 851                 if (!(dvp->v_vfsp->vfs_flag & VFS_XATTR))
 852                         return (EINVAL);
 853 
 854                 if (tp->tn_flags & ISXATTR)
 855                         /* No attributes on attributes */
 856                         return (EINVAL);
 857 
 858                 rw_enter(&tp->tn_rwlock, RW_WRITER);
 859                 if (tp->tn_xattrdp == NULL) {
 860                         if (!(flags & CREATE_XATTR_DIR)) {
 861                                 rw_exit(&tp->tn_rwlock);
 862                                 return (ENOENT);
 863                         }
 864 
 865                         /*
 866                          * No attribute directory exists for this
 867                          * node - create the attr dir as a side effect
 868                          * of this lookup.
 869                          */
 870 
 871                         /*
 872                          * Make sure we have adequate permission...
 873                          */
 874 
 875                         if ((error = tmp_taccess(tp, VWRITE, cred)) != 0) {
 876                                 rw_exit(&tp->tn_rwlock);
 877                                 return (error);
 878                         }
 879 
 880                         xdp = tmp_memalloc(sizeof (struct tmpnode),
 881                             TMP_MUSTHAVE);
 882                         tm = VTOTM(dvp);
 883                         tmpnode_init(tm, xdp, &tp->tn_attr, NULL);
 884                         /*
 885                          * Fix-up fields unique to attribute directories.
 886                          */
 887                         xdp->tn_flags = ISXATTR;
 888                         xdp->tn_type = VDIR;
 889                         if (tp->tn_type == VDIR) {
 890                                 xdp->tn_mode = tp->tn_attr.va_mode;
 891                         } else {
 892                                 xdp->tn_mode = 0700;
 893                                 if (tp->tn_attr.va_mode & 0040)
 894                                         xdp->tn_mode |= 0750;
 895                                 if (tp->tn_attr.va_mode & 0004)
 896                                         xdp->tn_mode |= 0705;
 897                         }
 898                         xdp->tn_vnode->v_type = VDIR;
 899                         xdp->tn_vnode->v_flag |= V_XATTRDIR;
 900                         tdirinit(tp, xdp);
 901                         tp->tn_xattrdp = xdp;
 902                 } else {
 903                         VN_HOLD(tp->tn_xattrdp->tn_vnode);
 904                 }
 905                 *vpp = TNTOV(tp->tn_xattrdp);
 906                 rw_exit(&tp->tn_rwlock);
 907                 return (0);
 908         }
 909 
 910         /*
 911          * Null component name is a synonym for directory being searched.
 912          */
 913         if (*nm == '\0') {
 914                 VN_HOLD(dvp);
 915                 *vpp = dvp;
 916                 return (0);
 917         }
 918         ASSERT(tp);
 919 
 920         error = tdirlookup(tp, nm, &ntp, cred);
 921 
 922         if (error == 0) {
 923                 ASSERT(ntp);
 924                 *vpp = TNTOV(ntp);
 925                 /*
 926                  * If vnode is a device return special vnode instead
 927                  */
 928                 if (IS_DEVVP(*vpp)) {
 929                         struct vnode *newvp;
 930 
 931                         newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
 932                             cred);
 933                         VN_RELE(*vpp);
 934                         *vpp = newvp;
 935                 }
 936         }
 937         TRACE_4(TR_FAC_TMPFS, TR_TMPFS_LOOKUP,
 938             "tmpfs lookup:vp %p name %s vpp %p error %d",
 939             dvp, nm, vpp, error);
 940         return (error);
 941 }
 942 
 943 /*ARGSUSED7*/
 944 static int
 945 tmp_create(
 946         struct vnode *dvp,
 947         char *nm,
 948         struct vattr *vap,
 949         enum vcexcl exclusive,
 950         int mode,
 951         struct vnode **vpp,
 952         struct cred *cred,
 953         int flag,
 954         caller_context_t *ct,
 955         vsecattr_t *vsecp)
 956 {
 957         struct tmpnode *parent;
 958         struct tmount *tm;
 959         struct tmpnode *self;
 960         int error;
 961         struct tmpnode *oldtp;
 962 
 963 again:
 964         parent = (struct tmpnode *)VTOTN(dvp);
 965         tm = (struct tmount *)VTOTM(dvp);
 966         self = NULL;
 967         error = 0;
 968         oldtp = NULL;
 969 
 970         /* device files not allowed in ext. attr dirs */
 971         if ((parent->tn_flags & ISXATTR) &&
 972             (vap->va_type == VBLK || vap->va_type == VCHR ||
 973             vap->va_type == VFIFO || vap->va_type == VDOOR ||
 974             vap->va_type == VSOCK || vap->va_type == VPORT))
 975                         return (EINVAL);
 976 
 977         if (vap->va_type == VREG && (vap->va_mode & VSVTX)) {
 978                 /* Must be privileged to set sticky bit */
 979                 if (secpolicy_vnode_stky_modify(cred))
 980                         vap->va_mode &= ~VSVTX;
 981         } else if (vap->va_type == VNON) {
 982                 return (EINVAL);
 983         }
 984 
 985         /*
 986          * Null component name is a synonym for directory being searched.
 987          */
 988         if (*nm == '\0') {
 989                 VN_HOLD(dvp);
 990                 oldtp = parent;
 991         } else {
 992                 error = tdirlookup(parent, nm, &oldtp, cred);
 993         }
 994 
 995         if (error == 0) {       /* name found */
 996                 boolean_t trunc = B_FALSE;
 997 
 998                 ASSERT(oldtp);
 999 
1000                 rw_enter(&oldtp->tn_rwlock, RW_WRITER);
1001 
1002                 /*
1003                  * if create/read-only an existing
1004                  * directory, allow it
1005                  */
1006                 if (exclusive == EXCL)
1007                         error = EEXIST;
1008                 else if ((oldtp->tn_type == VDIR) && (mode & VWRITE))
1009                         error = EISDIR;
1010                 else {
1011                         error = tmp_taccess(oldtp, mode, cred);
1012                 }
1013 
1014                 if (error) {
1015                         rw_exit(&oldtp->tn_rwlock);
1016                         tmpnode_rele(oldtp);
1017                         return (error);
1018                 }
1019                 *vpp = TNTOV(oldtp);
1020                 if ((*vpp)->v_type == VREG && (vap->va_mask & AT_SIZE) &&
1021                     vap->va_size == 0) {
1022                         rw_enter(&oldtp->tn_contents, RW_WRITER);
1023                         (void) tmpnode_trunc(tm, oldtp, 0);
1024                         rw_exit(&oldtp->tn_contents);
1025                         trunc = B_TRUE;
1026                 }
1027                 rw_exit(&oldtp->tn_rwlock);
1028                 if (IS_DEVVP(*vpp)) {
1029                         struct vnode *newvp;
1030 
1031                         newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
1032                             cred);
1033                         VN_RELE(*vpp);
1034                         if (newvp == NULL) {
1035                                 return (ENOSYS);
1036                         }
1037                         *vpp = newvp;
1038                 }
1039 
1040                 if (trunc)
1041                         vnevent_create(*vpp, ct);
1042 
1043                 return (0);
1044         }
1045 
1046         if (error != ENOENT)
1047                 return (error);
1048 
1049         rw_enter(&parent->tn_rwlock, RW_WRITER);
1050         error = tdirenter(tm, parent, nm, DE_CREATE,
1051             (struct tmpnode *)NULL, (struct tmpnode *)NULL,
1052             vap, &self, cred, ct);
1053         rw_exit(&parent->tn_rwlock);
1054 
1055         if (error) {
1056                 if (self)
1057                         tmpnode_rele(self);
1058 
1059                 if (error == EEXIST) {
1060                         /*
1061                          * This means that the file was created sometime
1062                          * after we checked and did not find it and when
1063                          * we went to create it.
1064                          * Since creat() is supposed to truncate a file
1065                          * that already exits go back to the begining
1066                          * of the function. This time we will find it
1067                          * and go down the tmp_trunc() path
1068                          */
1069                         goto again;
1070                 }
1071                 return (error);
1072         }
1073 
1074         *vpp = TNTOV(self);
1075 
1076         if (!error && IS_DEVVP(*vpp)) {
1077                 struct vnode *newvp;
1078 
1079                 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cred);
1080                 VN_RELE(*vpp);
1081                 if (newvp == NULL)
1082                         return (ENOSYS);
1083                 *vpp = newvp;
1084         }
1085         TRACE_3(TR_FAC_TMPFS, TR_TMPFS_CREATE,
1086             "tmpfs create:dvp %p nm %s vpp %p", dvp, nm, vpp);
1087         return (0);
1088 }
1089 
1090 /* ARGSUSED3 */
1091 static int
1092 tmp_remove(
1093         struct vnode *dvp,
1094         char *nm,
1095         struct cred *cred,
1096         caller_context_t *ct,
1097         int flags)
1098 {
1099         struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1100         int error;
1101         struct tmpnode *tp = NULL;
1102 
1103         error = tdirlookup(parent, nm, &tp, cred);
1104         if (error)
1105                 return (error);
1106 
1107         ASSERT(tp);
1108         rw_enter(&parent->tn_rwlock, RW_WRITER);
1109         rw_enter(&tp->tn_rwlock, RW_WRITER);
1110 
1111         if (tp->tn_type != VDIR ||
1112             (error = secpolicy_fs_linkdir(cred, dvp->v_vfsp)) == 0)
1113                 error = tdirdelete(parent, tp, nm, tp->tn_type == VDIR ?
1114                     DR_RMDIR : DR_REMOVE, cred);
1115 
1116         rw_exit(&tp->tn_rwlock);
1117         rw_exit(&parent->tn_rwlock);
1118         vnevent_remove(TNTOV(tp), dvp, nm, ct);
1119         tmpnode_rele(tp);
1120 
1121         TRACE_3(TR_FAC_TMPFS, TR_TMPFS_REMOVE,
1122             "tmpfs remove:dvp %p nm %s error %d", dvp, nm, error);
1123         return (error);
1124 }
1125 
1126 /* ARGSUSED4 */
1127 static int
1128 tmp_link(
1129         struct vnode *dvp,
1130         struct vnode *srcvp,
1131         char *tnm,
1132         struct cred *cred,
1133         caller_context_t *ct,
1134         int flags)
1135 {
1136         struct tmpnode *parent;
1137         struct tmpnode *from;
1138         struct tmount *tm = (struct tmount *)VTOTM(dvp);
1139         int error;
1140         struct tmpnode *found = NULL;
1141         struct vnode *realvp;
1142 
1143         if (VOP_REALVP(srcvp, &realvp, ct) == 0)
1144                 srcvp = realvp;
1145 
1146         parent = (struct tmpnode *)VTOTN(dvp);
1147         from = (struct tmpnode *)VTOTN(srcvp);
1148 
1149         if ((srcvp->v_type == VDIR &&
1150             secpolicy_fs_linkdir(cred, dvp->v_vfsp)) ||
1151             (from->tn_uid != crgetuid(cred) && secpolicy_basic_link(cred)))
1152                 return (EPERM);
1153 
1154         /*
1155          * Make sure link for extended attributes is valid
1156          * We only support hard linking of xattr's in xattrdir to an xattrdir
1157          */
1158         if ((from->tn_flags & ISXATTR) != (parent->tn_flags & ISXATTR))
1159                 return (EINVAL);
1160 
1161         error = tdirlookup(parent, tnm, &found, cred);
1162         if (error == 0) {
1163                 ASSERT(found);
1164                 tmpnode_rele(found);
1165                 return (EEXIST);
1166         }
1167 
1168         if (error != ENOENT)
1169                 return (error);
1170 
1171         rw_enter(&parent->tn_rwlock, RW_WRITER);
1172         error = tdirenter(tm, parent, tnm, DE_LINK, (struct tmpnode *)NULL,
1173             from, NULL, (struct tmpnode **)NULL, cred, ct);
1174         rw_exit(&parent->tn_rwlock);
1175         if (error == 0) {
1176                 vnevent_link(srcvp, ct);
1177         }
1178         return (error);
1179 }
1180 
1181 /* ARGSUSED5 */
1182 static int
1183 tmp_rename(
1184         struct vnode *odvp,     /* source parent vnode */
1185         char *onm,              /* source name */
1186         struct vnode *ndvp,     /* destination parent vnode */
1187         char *nnm,              /* destination name */
1188         struct cred *cred,
1189         caller_context_t *ct,
1190         int flags)
1191 {
1192         struct tmpnode *fromparent;
1193         struct tmpnode *toparent;
1194         struct tmpnode *fromtp = NULL;  /* source tmpnode */
1195         struct tmpnode *totp;           /* target tmpnode */
1196         struct tmount *tm = (struct tmount *)VTOTM(odvp);
1197         int error;
1198         int samedir = 0;        /* set if odvp == ndvp */
1199         struct vnode *realvp;
1200 
1201         if (VOP_REALVP(ndvp, &realvp, ct) == 0)
1202                 ndvp = realvp;
1203 
1204         fromparent = (struct tmpnode *)VTOTN(odvp);
1205         toparent = (struct tmpnode *)VTOTN(ndvp);
1206 
1207         if ((fromparent->tn_flags & ISXATTR) != (toparent->tn_flags & ISXATTR))
1208                 return (EINVAL);
1209 
1210         mutex_enter(&tm->tm_renamelck);
1211 
1212         /*
1213          * Look up tmpnode of file we're supposed to rename.
1214          */
1215         error = tdirlookup(fromparent, onm, &fromtp, cred);
1216         if (error) {
1217                 mutex_exit(&tm->tm_renamelck);
1218                 return (error);
1219         }
1220 
1221         /*
1222          * Make sure we can delete the old (source) entry.  This
1223          * requires write permission on the containing directory.  If
1224          * that directory is "sticky" it requires further checks.
1225          */
1226         if (((error = tmp_taccess(fromparent, VWRITE, cred)) != 0) ||
1227             (error = tmp_sticky_remove_access(fromparent, fromtp, cred)) != 0)
1228                 goto done;
1229 
1230         /*
1231          * Check for renaming to or from '.' or '..' or that
1232          * fromtp == fromparent
1233          */
1234         if ((onm[0] == '.' &&
1235             (onm[1] == '\0' || (onm[1] == '.' && onm[2] == '\0'))) ||
1236             (nnm[0] == '.' &&
1237             (nnm[1] == '\0' || (nnm[1] == '.' && nnm[2] == '\0'))) ||
1238             (fromparent == fromtp)) {
1239                 error = EINVAL;
1240                 goto done;
1241         }
1242 
1243         samedir = (fromparent == toparent);
1244         /*
1245          * Make sure we can search and rename into the new
1246          * (destination) directory.
1247          */
1248         if (!samedir) {
1249                 error = tmp_taccess(toparent, VEXEC|VWRITE, cred);
1250                 if (error)
1251                         goto done;
1252         }
1253 
1254         if (tdirlookup(toparent, nnm, &totp, cred) == 0) {
1255                 vnevent_pre_rename_dest(TNTOV(totp), ndvp, nnm, ct);
1256                 tmpnode_rele(totp);
1257         }
1258 
1259         /* Notify the target dir. if not the same as the source dir. */
1260         if (ndvp != odvp) {
1261                 vnevent_pre_rename_dest_dir(ndvp, TNTOV(fromtp), nnm, ct);
1262         }
1263 
1264         vnevent_pre_rename_src(TNTOV(fromtp), odvp, onm, ct);
1265 
1266         /*
1267          * Link source to new target
1268          */
1269         rw_enter(&toparent->tn_rwlock, RW_WRITER);
1270         error = tdirenter(tm, toparent, nnm, DE_RENAME,
1271             fromparent, fromtp, (struct vattr *)NULL,
1272             (struct tmpnode **)NULL, cred, ct);
1273         rw_exit(&toparent->tn_rwlock);
1274 
1275         if (error) {
1276                 /*
1277                  * ESAME isn't really an error; it indicates that the
1278                  * operation should not be done because the source and target
1279                  * are the same file, but that no error should be reported.
1280                  */
1281                 if (error == ESAME)
1282                         error = 0;
1283                 goto done;
1284         }
1285 
1286         /*
1287          * Unlink from source.
1288          */
1289         rw_enter(&fromparent->tn_rwlock, RW_WRITER);
1290         rw_enter(&fromtp->tn_rwlock, RW_WRITER);
1291 
1292         error = tdirdelete(fromparent, fromtp, onm, DR_RENAME, cred);
1293 
1294         /*
1295          * The following handles the case where our source tmpnode was
1296          * removed before we got to it.
1297          *
1298          * XXX We should also cleanup properly in the case where tdirdelete
1299          * fails for some other reason.  Currently this case shouldn't happen.
1300          * (see 1184991).
1301          */
1302         if (error == ENOENT)
1303                 error = 0;
1304 
1305         rw_exit(&fromtp->tn_rwlock);
1306         rw_exit(&fromparent->tn_rwlock);
1307 
1308         if (error == 0) {
1309                 vnevent_rename_src(TNTOV(fromtp), odvp, onm, ct);
1310                 /*
1311                  * vnevent_rename_dest is called in tdirenter().
1312                  */
1313                 vnevent_rename_dest_dir(ndvp, TNTOV(fromtp), nnm, ct);
1314         }
1315 
1316 done:
1317         tmpnode_rele(fromtp);
1318         mutex_exit(&tm->tm_renamelck);
1319 
1320         TRACE_5(TR_FAC_TMPFS, TR_TMPFS_RENAME,
1321             "tmpfs rename:ovp %p onm %s nvp %p nnm %s error %d", odvp, onm,
1322             ndvp, nnm, error);
1323         return (error);
1324 }
1325 
1326 /* ARGSUSED5 */
1327 static int
1328 tmp_mkdir(
1329         struct vnode *dvp,
1330         char *nm,
1331         struct vattr *va,
1332         struct vnode **vpp,
1333         struct cred *cred,
1334         caller_context_t *ct,
1335         int flags,
1336         vsecattr_t *vsecp)
1337 {
1338         struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1339         struct tmpnode *self = NULL;
1340         struct tmount *tm = (struct tmount *)VTOTM(dvp);
1341         int error;
1342 
1343         /* no new dirs allowed in xattr dirs */
1344         if (parent->tn_flags & ISXATTR)
1345                 return (EINVAL);
1346 
1347         /*
1348          * Might be dangling directory.  Catch it here,
1349          * because a ENOENT return from tdirlookup() is
1350          * an "o.k. return".
1351          */
1352         if (parent->tn_nlink == 0)
1353                 return (ENOENT);
1354 
1355         error = tdirlookup(parent, nm, &self, cred);
1356         if (error == 0) {
1357                 ASSERT(self);
1358                 tmpnode_rele(self);
1359                 return (EEXIST);
1360         }
1361         if (error != ENOENT)
1362                 return (error);
1363 
1364         rw_enter(&parent->tn_rwlock, RW_WRITER);
1365         error = tdirenter(tm, parent, nm, DE_MKDIR, (struct tmpnode *)NULL,
1366             (struct tmpnode *)NULL, va, &self, cred, ct);
1367         if (error) {
1368                 rw_exit(&parent->tn_rwlock);
1369                 if (self)
1370                         tmpnode_rele(self);
1371                 return (error);
1372         }
1373         rw_exit(&parent->tn_rwlock);
1374         *vpp = TNTOV(self);
1375         return (0);
1376 }
1377 
1378 /* ARGSUSED4 */
1379 static int
1380 tmp_rmdir(
1381         struct vnode *dvp,
1382         char *nm,
1383         struct vnode *cdir,
1384         struct cred *cred,
1385         caller_context_t *ct,
1386         int flags)
1387 {
1388         struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1389         struct tmpnode *self = NULL;
1390         struct vnode *vp;
1391         int error = 0;
1392 
1393         /*
1394          * Return error when removing . and ..
1395          */
1396         if (strcmp(nm, ".") == 0)
1397                 return (EINVAL);
1398         if (strcmp(nm, "..") == 0)
1399                 return (EEXIST); /* Should be ENOTEMPTY */
1400         error = tdirlookup(parent, nm, &self, cred);
1401         if (error)
1402                 return (error);
1403 
1404         rw_enter(&parent->tn_rwlock, RW_WRITER);
1405         rw_enter(&self->tn_rwlock, RW_WRITER);
1406 
1407         vp = TNTOV(self);
1408         if (vp == dvp || vp == cdir) {
1409                 error = EINVAL;
1410                 goto done1;
1411         }
1412         if (self->tn_type != VDIR) {
1413                 error = ENOTDIR;
1414                 goto done1;
1415         }
1416 
1417         mutex_enter(&self->tn_tlock);
1418         if (self->tn_nlink > 2) {
1419                 mutex_exit(&self->tn_tlock);
1420                 error = EEXIST;
1421                 goto done1;
1422         }
1423         mutex_exit(&self->tn_tlock);
1424 
1425         if (vn_vfswlock(vp)) {
1426                 error = EBUSY;
1427                 goto done1;
1428         }
1429         if (vn_mountedvfs(vp) != NULL) {
1430                 error = EBUSY;
1431                 goto done;
1432         }
1433 
1434         /*
1435          * Check for an empty directory
1436          * i.e. only includes entries for "." and ".."
1437          */
1438         if (self->tn_dirents > 2) {
1439                 error = EEXIST;         /* SIGH should be ENOTEMPTY */
1440                 /*
1441                  * Update atime because checking tn_dirents is logically
1442                  * equivalent to reading the directory
1443                  */
1444                 gethrestime(&self->tn_atime);
1445                 goto done;
1446         }
1447 
1448         error = tdirdelete(parent, self, nm, DR_RMDIR, cred);
1449 done:
1450         vn_vfsunlock(vp);
1451 done1:
1452         rw_exit(&self->tn_rwlock);
1453         rw_exit(&parent->tn_rwlock);
1454         vnevent_rmdir(TNTOV(self), dvp, nm, ct);
1455         tmpnode_rele(self);
1456 
1457         return (error);
1458 }
1459 
1460 /* ARGSUSED2 */
1461 static int
1462 tmp_readdir(
1463         struct vnode *vp,
1464         struct uio *uiop,
1465         struct cred *cred,
1466         int *eofp,
1467         caller_context_t *ct,
1468         int flags)
1469 {
1470         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1471         struct tdirent *tdp;
1472         int error = 0;
1473         size_t namelen;
1474         struct dirent64 *dp;
1475         ulong_t offset;
1476         ulong_t total_bytes_wanted;
1477         long outcount = 0;
1478         long bufsize;
1479         int reclen;
1480         caddr_t outbuf;
1481 
1482         if (uiop->uio_loffset >= MAXOFF_T) {
1483                 if (eofp)
1484                         *eofp = 1;
1485                 return (0);
1486         }
1487         /*
1488          * assuming system call has already called tmp_rwlock
1489          */
1490         ASSERT(RW_READ_HELD(&tp->tn_rwlock));
1491 
1492         if (uiop->uio_iovcnt != 1)
1493                 return (EINVAL);
1494 
1495         if (vp->v_type != VDIR)
1496                 return (ENOTDIR);
1497 
1498         /*
1499          * There's a window here where someone could have removed
1500          * all the entries in the directory after we put a hold on the
1501          * vnode but before we grabbed the rwlock.  Just return.
1502          */
1503         if (tp->tn_dir == NULL) {
1504                 if (tp->tn_nlink) {
1505                         panic("empty directory 0x%p", (void *)tp);
1506                         /*NOTREACHED*/
1507                 }
1508                 return (0);
1509         }
1510 
1511         /*
1512          * Get space for multiple directory entries
1513          */
1514         total_bytes_wanted = uiop->uio_iov->iov_len;
1515         bufsize = total_bytes_wanted + sizeof (struct dirent64);
1516         outbuf = kmem_alloc(bufsize, KM_SLEEP);
1517 
1518         dp = (struct dirent64 *)outbuf;
1519 
1520 
1521         offset = 0;
1522         tdp = tp->tn_dir;
1523         while (tdp) {
1524                 namelen = strlen(tdp->td_name);      /* no +1 needed */
1525                 offset = tdp->td_offset;
1526                 if (offset >= uiop->uio_offset) {
1527                         reclen = (int)DIRENT64_RECLEN(namelen);
1528                         if (outcount + reclen > total_bytes_wanted) {
1529                                 if (!outcount)
1530                                         /*
1531                                          * Buffer too small for any entries.
1532                                          */
1533                                         error = EINVAL;
1534                                 break;
1535                         }
1536                         ASSERT(tdp->td_tmpnode != NULL);
1537 
1538                         /* use strncpy(9f) to zero out uninitialized bytes */
1539 
1540                         (void) strncpy(dp->d_name, tdp->td_name,
1541                             DIRENT64_NAMELEN(reclen));
1542                         dp->d_reclen = (ushort_t)reclen;
1543                         dp->d_ino = (ino64_t)tdp->td_tmpnode->tn_nodeid;
1544                         dp->d_off = (offset_t)tdp->td_offset + 1;
1545                         dp = (struct dirent64 *)
1546                             ((uintptr_t)dp + dp->d_reclen);
1547                         outcount += reclen;
1548                         ASSERT(outcount <= bufsize);
1549                 }
1550                 tdp = tdp->td_next;
1551         }
1552 
1553         if (!error)
1554                 error = uiomove(outbuf, outcount, UIO_READ, uiop);
1555 
1556         if (!error) {
1557                 /* If we reached the end of the list our offset */
1558                 /* should now be just past the end. */
1559                 if (!tdp) {
1560                         offset += 1;
1561                         if (eofp)
1562                                 *eofp = 1;
1563                 } else if (eofp)
1564                         *eofp = 0;
1565                 uiop->uio_offset = offset;
1566         }
1567         gethrestime(&tp->tn_atime);
1568         kmem_free(outbuf, bufsize);
1569         return (error);
1570 }
1571 
1572 /* ARGSUSED5 */
1573 static int
1574 tmp_symlink(
1575         struct vnode *dvp,
1576         char *lnm,
1577         struct vattr *tva,
1578         char *tnm,
1579         struct cred *cred,
1580         caller_context_t *ct,
1581         int flags)
1582 {
1583         struct tmpnode *parent = (struct tmpnode *)VTOTN(dvp);
1584         struct tmpnode *self = (struct tmpnode *)NULL;
1585         struct tmount *tm = (struct tmount *)VTOTM(dvp);
1586         char *cp = NULL;
1587         int error;
1588         size_t len;
1589 
1590         /* no symlinks allowed to files in xattr dirs */
1591         if (parent->tn_flags & ISXATTR)
1592                 return (EINVAL);
1593 
1594         error = tdirlookup(parent, lnm, &self, cred);
1595         if (error == 0) {
1596                 /*
1597                  * The entry already exists
1598                  */
1599                 tmpnode_rele(self);
1600                 return (EEXIST);        /* was 0 */
1601         }
1602 
1603         if (error != ENOENT) {
1604                 if (self != NULL)
1605                         tmpnode_rele(self);
1606                 return (error);
1607         }
1608 
1609         rw_enter(&parent->tn_rwlock, RW_WRITER);
1610         error = tdirenter(tm, parent, lnm, DE_CREATE, (struct tmpnode *)NULL,
1611             (struct tmpnode *)NULL, tva, &self, cred, ct);
1612         rw_exit(&parent->tn_rwlock);
1613 
1614         if (error) {
1615                 if (self)
1616                         tmpnode_rele(self);
1617                 return (error);
1618         }
1619         len = strlen(tnm) + 1;
1620         cp = tmp_memalloc(len, 0);
1621         if (cp == NULL) {
1622                 tmpnode_rele(self);
1623                 return (ENOSPC);
1624         }
1625         (void) strcpy(cp, tnm);
1626 
1627         self->tn_symlink = cp;
1628         self->tn_size = len - 1;
1629         tmpnode_rele(self);
1630         return (error);
1631 }
1632 
1633 /* ARGSUSED2 */
1634 static int
1635 tmp_readlink(
1636         struct vnode *vp,
1637         struct uio *uiop,
1638         struct cred *cred,
1639         caller_context_t *ct)
1640 {
1641         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1642         int error = 0;
1643 
1644         if (vp->v_type != VLNK)
1645                 return (EINVAL);
1646 
1647         rw_enter(&tp->tn_rwlock, RW_READER);
1648         rw_enter(&tp->tn_contents, RW_READER);
1649         error = uiomove(tp->tn_symlink, tp->tn_size, UIO_READ, uiop);
1650         gethrestime(&tp->tn_atime);
1651         rw_exit(&tp->tn_contents);
1652         rw_exit(&tp->tn_rwlock);
1653         return (error);
1654 }
1655 
1656 /* ARGSUSED */
1657 static int
1658 tmp_fsync(
1659         struct vnode *vp,
1660         int syncflag,
1661         struct cred *cred,
1662         caller_context_t *ct)
1663 {
1664         return (0);
1665 }
1666 
1667 /* ARGSUSED */
1668 static void
1669 tmp_inactive(struct vnode *vp, struct cred *cred, caller_context_t *ct)
1670 {
1671         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1672         struct tmount *tm = (struct tmount *)VFSTOTM(vp->v_vfsp);
1673 
1674         rw_enter(&tp->tn_rwlock, RW_WRITER);
1675 top:
1676         mutex_enter(&tp->tn_tlock);
1677         mutex_enter(&vp->v_lock);
1678         ASSERT(vp->v_count >= 1);
1679 
1680         /*
1681          * If we don't have the last hold or the link count is non-zero,
1682          * there's little to do -- just drop our hold.
1683          */
1684         if (vp->v_count > 1 || tp->tn_nlink != 0) {
1685                 vp->v_count--;
1686                 mutex_exit(&vp->v_lock);
1687                 mutex_exit(&tp->tn_tlock);
1688                 rw_exit(&tp->tn_rwlock);
1689                 return;
1690         }
1691 
1692         /*
1693          * We have the last hold *and* the link count is zero, so this
1694          * tmpnode is dead from the filesystem's viewpoint.  However,
1695          * if the tmpnode has any pages associated with it (i.e. if it's
1696          * a normal file with non-zero size), the tmpnode can still be
1697          * discovered by pageout or fsflush via the page vnode pointers.
1698          * In this case we must drop all our locks, truncate the tmpnode,
1699          * and try the whole dance again.
1700          */
1701         if (tp->tn_size != 0) {
1702                 if (tp->tn_type == VREG) {
1703                         mutex_exit(&vp->v_lock);
1704                         mutex_exit(&tp->tn_tlock);
1705                         rw_enter(&tp->tn_contents, RW_WRITER);
1706                         (void) tmpnode_trunc(tm, tp, 0);
1707                         rw_exit(&tp->tn_contents);
1708                         ASSERT(tp->tn_size == 0);
1709                         ASSERT(tp->tn_nblocks == 0);
1710                         goto top;
1711                 }
1712                 if (tp->tn_type == VLNK)
1713                         tmp_memfree(tp->tn_symlink, tp->tn_size + 1);
1714         }
1715 
1716         /*
1717          * Remove normal file/dir's xattr dir and xattrs.
1718          */
1719         if (tp->tn_xattrdp) {
1720                 struct tmpnode *xtp = tp->tn_xattrdp;
1721 
1722                 ASSERT(xtp->tn_flags & ISXATTR);
1723                 tmpnode_hold(xtp);
1724                 rw_enter(&xtp->tn_rwlock, RW_WRITER);
1725                 tdirtrunc(xtp);
1726                 DECR_COUNT(&xtp->tn_nlink, &xtp->tn_tlock);
1727                 tp->tn_xattrdp = NULL;
1728                 rw_exit(&xtp->tn_rwlock);
1729                 tmpnode_rele(xtp);
1730         }
1731 
1732         mutex_exit(&vp->v_lock);
1733         mutex_exit(&tp->tn_tlock);
1734         /* Here's our chance to send invalid event while we're between locks */
1735         vn_invalid(TNTOV(tp));
1736         mutex_enter(&tm->tm_contents);
1737         if (tp->tn_forw == NULL)
1738                 tm->tm_rootnode->tn_back = tp->tn_back;
1739         else
1740                 tp->tn_forw->tn_back = tp->tn_back;
1741         tp->tn_back->tn_forw = tp->tn_forw;
1742         mutex_exit(&tm->tm_contents);
1743         rw_exit(&tp->tn_rwlock);
1744         rw_destroy(&tp->tn_rwlock);
1745         mutex_destroy(&tp->tn_tlock);
1746         vn_free(TNTOV(tp));
1747         tmp_memfree(tp, sizeof (struct tmpnode));
1748 }
1749 
1750 /* ARGSUSED2 */
1751 static int
1752 tmp_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
1753 {
1754         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
1755         struct tfid *tfid;
1756 
1757         if (fidp->fid_len < (sizeof (struct tfid) - sizeof (ushort_t))) {
1758                 fidp->fid_len = sizeof (struct tfid) - sizeof (ushort_t);
1759                 return (ENOSPC);
1760         }
1761 
1762         tfid = (struct tfid *)fidp;
1763         bzero(tfid, sizeof (struct tfid));
1764         tfid->tfid_len = (int)sizeof (struct tfid) - sizeof (ushort_t);
1765 
1766         tfid->tfid_ino = tp->tn_nodeid;
1767         tfid->tfid_gen = tp->tn_gen;
1768 
1769         return (0);
1770 }
1771 
1772 
1773 /*
1774  * Return all the pages from [off..off+len] in given file
1775  */
1776 /* ARGSUSED */
1777 static int
1778 tmp_getpage(
1779         struct vnode *vp,
1780         offset_t off,
1781         size_t len,
1782         uint_t *protp,
1783         page_t *pl[],
1784         size_t plsz,
1785         struct seg *seg,
1786         caddr_t addr,
1787         enum seg_rw rw,
1788         struct cred *cr,
1789         caller_context_t *ct)
1790 {
1791         int err = 0;
1792         struct tmpnode *tp = VTOTN(vp);
1793         anoff_t toff = (anoff_t)off;
1794         size_t tlen = len;
1795         u_offset_t tmpoff;
1796         timestruc_t now;
1797 
1798         rw_enter(&tp->tn_contents, RW_READER);
1799 
1800         if (off + len  > tp->tn_size + PAGEOFFSET) {
1801                 err = EFAULT;
1802                 goto out;
1803         }
1804         /*
1805          * Look for holes (no anon slot) in faulting range. If there are
1806          * holes we have to switch to a write lock and fill them in. Swap
1807          * space for holes was already reserved when the file was grown.
1808          */
1809         tmpoff = toff;
1810         if (non_anon(tp->tn_anon, btop(off), &tmpoff, &tlen)) {
1811                 if (!rw_tryupgrade(&tp->tn_contents)) {
1812                         rw_exit(&tp->tn_contents);
1813                         rw_enter(&tp->tn_contents, RW_WRITER);
1814                         /* Size may have changed when lock was dropped */
1815                         if (off + len  > tp->tn_size + PAGEOFFSET) {
1816                                 err = EFAULT;
1817                                 goto out;
1818                         }
1819                 }
1820                 for (toff = (anoff_t)off; toff < (anoff_t)off + len;
1821                     toff += PAGESIZE) {
1822                         if (anon_get_ptr(tp->tn_anon, btop(toff)) == NULL) {
1823                                 /* XXX - may allocate mem w. write lock held */
1824                                 (void) anon_set_ptr(tp->tn_anon, btop(toff),
1825                                     anon_alloc(vp, toff), ANON_SLEEP);
1826                                 tp->tn_nblocks++;
1827                         }
1828                 }
1829                 rw_downgrade(&tp->tn_contents);
1830         }
1831 
1832 
1833         err = pvn_getpages(tmp_getapage, vp, (u_offset_t)off, len, protp,
1834             pl, plsz, seg, addr, rw, cr);
1835 
1836         gethrestime(&now);
1837         tp->tn_atime = now;
1838         if (rw == S_WRITE)
1839                 tp->tn_mtime = now;
1840 
1841 out:
1842         rw_exit(&tp->tn_contents);
1843         return (err);
1844 }
1845 
1846 /*
1847  * Called from pvn_getpages to get a particular page.
1848  */
1849 /*ARGSUSED*/
1850 static int
1851 tmp_getapage(
1852         struct vnode *vp,
1853         u_offset_t off,
1854         size_t len,
1855         uint_t *protp,
1856         page_t *pl[],
1857         size_t plsz,
1858         struct seg *seg,
1859         caddr_t addr,
1860         enum seg_rw rw,
1861         struct cred *cr)
1862 {
1863         struct page *pp;
1864         int flags;
1865         int err = 0;
1866         struct vnode *pvp;
1867         u_offset_t poff;
1868 
1869         if (protp != NULL)
1870                 *protp = PROT_ALL;
1871 again:
1872         if (pp = page_lookup(vp, off, rw == S_CREATE ? SE_EXCL : SE_SHARED)) {
1873                 if (pl) {
1874                         pl[0] = pp;
1875                         pl[1] = NULL;
1876                 } else {
1877                         page_unlock(pp);
1878                 }
1879         } else {
1880                 pp = page_create_va(vp, off, PAGESIZE,
1881                     PG_WAIT | PG_EXCL, seg, addr);
1882                 /*
1883                  * Someone raced in and created the page after we did the
1884                  * lookup but before we did the create, so go back and
1885                  * try to look it up again.
1886                  */
1887                 if (pp == NULL)
1888                         goto again;
1889                 /*
1890                  * Fill page from backing store, if any. If none, then
1891                  * either this is a newly filled hole or page must have
1892                  * been unmodified and freed so just zero it out.
1893                  */
1894                 err = swap_getphysname(vp, off, &pvp, &poff);
1895                 if (err) {
1896                         panic("tmp_getapage: no anon slot vp %p "
1897                             "off %llx pp %p\n", (void *)vp, off, (void *)pp);
1898                 }
1899                 if (pvp) {
1900                         flags = (pl == NULL ? B_ASYNC|B_READ : B_READ);
1901                         err = VOP_PAGEIO(pvp, pp, (u_offset_t)poff, PAGESIZE,
1902                             flags, cr, NULL);
1903                         if (flags & B_ASYNC)
1904                                 pp = NULL;
1905                 } else if (rw != S_CREATE) {
1906                         pagezero(pp, 0, PAGESIZE);
1907                 }
1908                 if (err && pp)
1909                         pvn_read_done(pp, B_ERROR);
1910                 if (err == 0) {
1911                         if (pl)
1912                                 pvn_plist_init(pp, pl, plsz, off, PAGESIZE, rw);
1913                         else
1914                                 pvn_io_done(pp);
1915                 }
1916         }
1917         return (err);
1918 }
1919 
1920 
1921 /*
1922  * Flags are composed of {B_INVAL, B_DIRTY B_FREE, B_DONTNEED}.
1923  * If len == 0, do from off to EOF.
1924  */
1925 static int tmp_nopage = 0;      /* Don't do tmp_putpage's if set */
1926 
1927 /* ARGSUSED */
1928 int
1929 tmp_putpage(
1930         register struct vnode *vp,
1931         offset_t off,
1932         size_t len,
1933         int flags,
1934         struct cred *cr,
1935         caller_context_t *ct)
1936 {
1937         register page_t *pp;
1938         u_offset_t io_off;
1939         size_t io_len = 0;
1940         int err = 0;
1941         struct tmpnode *tp = VTOTN(vp);
1942         int dolock;
1943 
1944         if (tmp_nopage)
1945                 return (0);
1946 
1947         ASSERT(vp->v_count != 0);
1948 
1949         if (vp->v_flag & VNOMAP)
1950                 return (ENOSYS);
1951 
1952         /*
1953          * This being tmpfs, we don't ever do i/o unless we really
1954          * have to (when we're low on memory and pageout calls us
1955          * with B_ASYNC | B_FREE or the user explicitly asks for it with
1956          * B_DONTNEED).
1957          * XXX to approximately track the mod time like ufs we should
1958          * update the times here. The problem is, once someone does a
1959          * store we never clear the mod bit and do i/o, thus fsflush
1960          * will keep calling us every 30 seconds to do the i/o and we'll
1961          * continually update the mod time. At least we update the mod
1962          * time on the first store because this results in a call to getpage.
1963          */
1964         if (flags != (B_ASYNC | B_FREE) && (flags & B_INVAL) == 0 &&
1965             (flags & B_DONTNEED) == 0)
1966                 return (0);
1967         /*
1968          * If this thread owns the lock, i.e., this thread grabbed it
1969          * as writer somewhere above, then we don't need to grab the
1970          * lock as reader in this routine.
1971          */
1972         dolock = (rw_owner(&tp->tn_contents) != curthread);
1973 
1974         /*
1975          * If this is pageout don't block on the lock as you could deadlock
1976          * when freemem == 0 (another thread has the read lock and is blocked
1977          * creating a page, and a third thread is waiting to get the writers
1978          * lock - waiting writers priority blocks us from getting the read
1979          * lock). Of course, if the only freeable pages are on this tmpnode
1980          * we're hosed anyways. A better solution might be a new lock type.
1981          * Note: ufs has the same problem.
1982          */
1983         if (curproc == proc_pageout) {
1984                 if (!rw_tryenter(&tp->tn_contents, RW_READER))
1985                         return (ENOMEM);
1986         } else if (dolock)
1987                 rw_enter(&tp->tn_contents, RW_READER);
1988 
1989         if (!vn_has_cached_data(vp))
1990                 goto out;
1991 
1992         if (len == 0) {
1993                 if (curproc == proc_pageout) {
1994                         panic("tmp: pageout can't block");
1995                         /*NOTREACHED*/
1996                 }
1997 
1998                 /* Search the entire vp list for pages >= off. */
1999                 err = pvn_vplist_dirty(vp, (u_offset_t)off, tmp_putapage,
2000                     flags, cr);
2001         } else {
2002                 u_offset_t eoff;
2003 
2004                 /*
2005                  * Loop over all offsets in the range [off...off + len]
2006                  * looking for pages to deal with.
2007                  */
2008                 eoff = MIN(off + len, tp->tn_size);
2009                 for (io_off = off; io_off < eoff; io_off += io_len) {
2010                         /*
2011                          * If we are not invalidating, synchronously
2012                          * freeing or writing pages use the routine
2013                          * page_lookup_nowait() to prevent reclaiming
2014                          * them from the free list.
2015                          */
2016                         if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2017                                 pp = page_lookup(vp, io_off,
2018                                     (flags & (B_INVAL | B_FREE)) ?
2019                                     SE_EXCL : SE_SHARED);
2020                         } else {
2021                                 pp = page_lookup_nowait(vp, io_off,
2022                                     (flags & B_FREE) ? SE_EXCL : SE_SHARED);
2023                         }
2024 
2025                         if (pp == NULL || pvn_getdirty(pp, flags) == 0)
2026                                 io_len = PAGESIZE;
2027                         else {
2028                                 err = tmp_putapage(vp, pp, &io_off, &io_len,
2029                                     flags, cr);
2030                                 if (err != 0)
2031                                         break;
2032                         }
2033                 }
2034         }
2035         /* If invalidating, verify all pages on vnode list are gone. */
2036         if (err == 0 && off == 0 && len == 0 &&
2037             (flags & B_INVAL) && vn_has_cached_data(vp)) {
2038                 panic("tmp_putpage: B_INVAL, pages not gone");
2039                 /*NOTREACHED*/
2040         }
2041 out:
2042         if ((curproc == proc_pageout) || dolock)
2043                 rw_exit(&tp->tn_contents);
2044         /*
2045          * Only reason putapage is going to give us SE_NOSWAP as error
2046          * is when we ask a page to be written to physical backing store
2047          * and there is none. Ignore this because we might be dealing
2048          * with a swap page which does not have any backing store
2049          * on disk. In any other case we won't get this error over here.
2050          */
2051         if (err == SE_NOSWAP)
2052                 err = 0;
2053         return (err);
2054 }
2055 
2056 long tmp_putpagecnt, tmp_pagespushed;
2057 
2058 /*
2059  * Write out a single page.
2060  * For tmpfs this means choose a physical swap slot and write the page
2061  * out using VOP_PAGEIO. For performance, we attempt to kluster; i.e.,
2062  * we try to find a bunch of other dirty pages adjacent in the file
2063  * and a bunch of contiguous swap slots, and then write all the pages
2064  * out in a single i/o.
2065  */
2066 /*ARGSUSED*/
2067 static int
2068 tmp_putapage(
2069         struct vnode *vp,
2070         page_t *pp,
2071         u_offset_t *offp,
2072         size_t *lenp,
2073         int flags,
2074         struct cred *cr)
2075 {
2076         int err;
2077         ulong_t klstart, kllen;
2078         page_t *pplist, *npplist;
2079         extern int klustsize;
2080         long tmp_klustsize;
2081         struct tmpnode *tp;
2082         size_t pp_off, pp_len;
2083         u_offset_t io_off;
2084         size_t io_len;
2085         struct vnode *pvp;
2086         u_offset_t pstart;
2087         u_offset_t offset;
2088         u_offset_t tmpoff;
2089 
2090         ASSERT(PAGE_LOCKED(pp));
2091 
2092         /* Kluster in tmp_klustsize chunks */
2093         tp = VTOTN(vp);
2094         tmp_klustsize = klustsize;
2095         offset = pp->p_offset;
2096         klstart = (offset / tmp_klustsize) * tmp_klustsize;
2097         kllen = MIN(tmp_klustsize, tp->tn_size - klstart);
2098 
2099         /* Get a kluster of pages */
2100         pplist =
2101             pvn_write_kluster(vp, pp, &tmpoff, &pp_len, klstart, kllen, flags);
2102 
2103         pp_off = (size_t)tmpoff;
2104 
2105         /*
2106          * Get a cluster of physical offsets for the pages; the amount we
2107          * get may be some subrange of what we ask for (io_off, io_len).
2108          */
2109         io_off = pp_off;
2110         io_len = pp_len;
2111         err = swap_newphysname(vp, offset, &io_off, &io_len, &pvp, &pstart);
2112         ASSERT(err != SE_NOANON); /* anon slot must have been filled */
2113         if (err) {
2114                 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2115                 /*
2116                  * If this routine is called as a result of segvn_sync
2117                  * operation and we have no physical swap then we can get an
2118                  * error here. In such case we would return SE_NOSWAP as error.
2119                  * At this point, we expect only SE_NOSWAP.
2120                  */
2121                 ASSERT(err == SE_NOSWAP);
2122                 if (flags & B_INVAL)
2123                         err = ENOMEM;
2124                 goto out;
2125         }
2126         ASSERT(pp_off <= io_off && io_off + io_len <= pp_off + pp_len);
2127         ASSERT(io_off <= offset && offset < io_off + io_len);
2128 
2129         /* Toss pages at front/rear that we couldn't get physical backing for */
2130         if (io_off != pp_off) {
2131                 npplist = NULL;
2132                 page_list_break(&pplist, &npplist, btop(io_off - pp_off));
2133                 ASSERT(pplist->p_offset == pp_off);
2134                 ASSERT(pplist->p_prev->p_offset == io_off - PAGESIZE);
2135                 pvn_write_done(pplist, B_ERROR | B_WRITE | flags);
2136                 pplist = npplist;
2137         }
2138         if (io_off + io_len < pp_off + pp_len) {
2139                 npplist = NULL;
2140                 page_list_break(&pplist, &npplist, btop(io_len));
2141                 ASSERT(npplist->p_offset == io_off + io_len);
2142                 ASSERT(npplist->p_prev->p_offset == pp_off + pp_len - PAGESIZE);
2143                 pvn_write_done(npplist, B_ERROR | B_WRITE | flags);
2144         }
2145 
2146         ASSERT(pplist->p_offset == io_off);
2147         ASSERT(pplist->p_prev->p_offset == io_off + io_len - PAGESIZE);
2148         ASSERT(btopr(io_len) <= btopr(kllen));
2149 
2150         /* Do i/o on the remaining kluster */
2151         err = VOP_PAGEIO(pvp, pplist, (u_offset_t)pstart, io_len,
2152             B_WRITE | flags, cr, NULL);
2153 
2154         if ((flags & B_ASYNC) == 0) {
2155                 pvn_write_done(pplist, ((err) ? B_ERROR : 0) | B_WRITE | flags);
2156         }
2157 out:
2158         if (!err) {
2159                 if (offp)
2160                         *offp = io_off;
2161                 if (lenp)
2162                         *lenp = io_len;
2163                 tmp_putpagecnt++;
2164                 tmp_pagespushed += btop(io_len);
2165         }
2166         if (err && err != ENOMEM && err != SE_NOSWAP)
2167                 cmn_err(CE_WARN, "tmp_putapage: err %d\n", err);
2168         return (err);
2169 }
2170 
2171 /* ARGSUSED */
2172 static int
2173 tmp_map(
2174         struct vnode *vp,
2175         offset_t off,
2176         struct as *as,
2177         caddr_t *addrp,
2178         size_t len,
2179         uchar_t prot,
2180         uchar_t maxprot,
2181         uint_t flags,
2182         struct cred *cred,
2183         caller_context_t *ct)
2184 {
2185         struct segvn_crargs vn_a;
2186         struct tmpnode *tp = (struct tmpnode *)VTOTN(vp);
2187         int error;
2188 
2189 #ifdef _ILP32
2190         if (len > MAXOFF_T)
2191                 return (ENOMEM);
2192 #endif
2193 
2194         if (vp->v_flag & VNOMAP)
2195                 return (ENOSYS);
2196 
2197         if (off < 0 || (offset_t)(off + len) < 0 ||
2198             off > MAXOFF_T || (off + len) > MAXOFF_T)
2199                 return (ENXIO);
2200 
2201         if (vp->v_type != VREG)
2202                 return (ENODEV);
2203 
2204         /*
2205          * Don't allow mapping to locked file
2206          */
2207         if (vn_has_mandatory_locks(vp, tp->tn_mode)) {
2208                 return (EAGAIN);
2209         }
2210 
2211         as_rangelock(as);
2212         error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
2213         if (error != 0) {
2214                 as_rangeunlock(as);
2215                 return (error);
2216         }
2217 
2218         vn_a.vp = vp;
2219         vn_a.offset = (u_offset_t)off;
2220         vn_a.type = flags & MAP_TYPE;
2221         vn_a.prot = prot;
2222         vn_a.maxprot = maxprot;
2223         vn_a.flags = flags & ~MAP_TYPE;
2224         vn_a.cred = cred;
2225         vn_a.amp = NULL;
2226         vn_a.szc = 0;
2227         vn_a.lgrp_mem_policy_flags = 0;
2228 
2229         error = as_map(as, *addrp, len, segvn_create, &vn_a);
2230         as_rangeunlock(as);
2231         return (error);
2232 }
2233 
2234 /*
2235  * tmp_addmap and tmp_delmap can't be called since the vp
2236  * maintained in the segvn mapping is NULL.
2237  */
2238 /* ARGSUSED */
2239 static int
2240 tmp_addmap(
2241         struct vnode *vp,
2242         offset_t off,
2243         struct as *as,
2244         caddr_t addr,
2245         size_t len,
2246         uchar_t prot,
2247         uchar_t maxprot,
2248         uint_t flags,
2249         struct cred *cred,
2250         caller_context_t *ct)
2251 {
2252         return (0);
2253 }
2254 
2255 /* ARGSUSED */
2256 static int
2257 tmp_delmap(
2258         struct vnode *vp,
2259         offset_t off,
2260         struct as *as,
2261         caddr_t addr,
2262         size_t len,
2263         uint_t prot,
2264         uint_t maxprot,
2265         uint_t flags,
2266         struct cred *cred,
2267         caller_context_t *ct)
2268 {
2269         return (0);
2270 }
2271 
2272 static int
2273 tmp_freesp(struct vnode *vp, struct flock64 *lp, int flag)
2274 {
2275         register int i;
2276         register struct tmpnode *tp = VTOTN(vp);
2277         int error;
2278 
2279         ASSERT(vp->v_type == VREG);
2280         ASSERT(lp->l_start >= 0);
2281 
2282         if (lp->l_len != 0)
2283                 return (EINVAL);
2284 
2285         rw_enter(&tp->tn_rwlock, RW_WRITER);
2286         if (tp->tn_size == lp->l_start) {
2287                 rw_exit(&tp->tn_rwlock);
2288                 return (0);
2289         }
2290 
2291         /*
2292          * Check for any mandatory locks on the range
2293          */
2294         if (MANDLOCK(vp, tp->tn_mode)) {
2295                 long save_start;
2296 
2297                 save_start = lp->l_start;
2298 
2299                 if (tp->tn_size < lp->l_start) {
2300                         /*
2301                          * "Truncate up" case: need to make sure there
2302                          * is no lock beyond current end-of-file. To
2303                          * do so, we need to set l_start to the size
2304                          * of the file temporarily.
2305                          */
2306                         lp->l_start = tp->tn_size;
2307                 }
2308                 lp->l_type = F_WRLCK;
2309                 lp->l_sysid = 0;
2310                 lp->l_pid = ttoproc(curthread)->p_pid;
2311                 i = (flag & (FNDELAY|FNONBLOCK)) ? 0 : SLPFLCK;
2312                 if ((i = reclock(vp, lp, i, 0, lp->l_start, NULL)) != 0 ||
2313                     lp->l_type != F_UNLCK) {
2314                         rw_exit(&tp->tn_rwlock);
2315                         return (i ? i : EAGAIN);
2316                 }
2317 
2318                 lp->l_start = save_start;
2319         }
2320         VFSTOTM(vp->v_vfsp);
2321 
2322         rw_enter(&tp->tn_contents, RW_WRITER);
2323         error = tmpnode_trunc((struct tmount *)VFSTOTM(vp->v_vfsp),
2324             tp, (ulong_t)lp->l_start);
2325         rw_exit(&tp->tn_contents);
2326         rw_exit(&tp->tn_rwlock);
2327         return (error);
2328 }
2329 
2330 /* ARGSUSED */
2331 static int
2332 tmp_space(
2333         struct vnode *vp,
2334         int cmd,
2335         struct flock64 *bfp,
2336         int flag,
2337         offset_t offset,
2338         cred_t *cred,
2339         caller_context_t *ct)
2340 {
2341         int error;
2342 
2343         if (cmd != F_FREESP)
2344                 return (EINVAL);
2345         if ((error = convoff(vp, bfp, 0, (offset_t)offset)) == 0) {
2346                 if ((bfp->l_start > MAXOFF_T) || (bfp->l_len > MAXOFF_T))
2347                         return (EFBIG);
2348                 error = tmp_freesp(vp, bfp, flag);
2349 
2350                 if (error == 0) {
2351                         if (bfp->l_start == 0) {
2352                                 vnevent_truncate(vp, ct);
2353                         } else {
2354                                 vnevent_resize(vp, ct);
2355                         }
2356                 }
2357         }
2358         return (error);
2359 }
2360 
2361 /* ARGSUSED */
2362 static int
2363 tmp_seek(
2364         struct vnode *vp,
2365         offset_t ooff,
2366         offset_t *noffp,
2367         caller_context_t *ct)
2368 {
2369         return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
2370 }
2371 
2372 /* ARGSUSED2 */
2373 static int
2374 tmp_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2375 {
2376         struct tmpnode *tp = VTOTN(vp);
2377 
2378         if (write_lock) {
2379                 rw_enter(&tp->tn_rwlock, RW_WRITER);
2380         } else {
2381                 rw_enter(&tp->tn_rwlock, RW_READER);
2382         }
2383         return (write_lock);
2384 }
2385 
2386 /* ARGSUSED1 */
2387 static void
2388 tmp_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
2389 {
2390         struct tmpnode *tp = VTOTN(vp);
2391 
2392         rw_exit(&tp->tn_rwlock);
2393 }
2394 
2395 static int
2396 tmp_pathconf(
2397         struct vnode *vp,
2398         int cmd,
2399         ulong_t *valp,
2400         cred_t *cr,
2401         caller_context_t *ct)
2402 {
2403         struct tmpnode *tp = NULL;
2404         int error;
2405 
2406         switch (cmd) {
2407         case _PC_XATTR_EXISTS:
2408                 if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
2409                         *valp = 0;      /* assume no attributes */
2410                         error = 0;      /* okay to ask */
2411                         tp = VTOTN(vp);
2412                         rw_enter(&tp->tn_rwlock, RW_READER);
2413                         if (tp->tn_xattrdp) {
2414                                 rw_enter(&tp->tn_xattrdp->tn_rwlock, RW_READER);
2415                                 /* do not count "." and ".." */
2416                                 if (tp->tn_xattrdp->tn_dirents > 2)
2417                                         *valp = 1;
2418                                 rw_exit(&tp->tn_xattrdp->tn_rwlock);
2419                         }
2420                         rw_exit(&tp->tn_rwlock);
2421                 } else {
2422                         error = EINVAL;
2423                 }
2424                 break;
2425         case _PC_SATTR_ENABLED:
2426         case _PC_SATTR_EXISTS:
2427                 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2428                     (vp->v_type == VREG || vp->v_type == VDIR);
2429                 error = 0;
2430                 break;
2431         case _PC_TIMESTAMP_RESOLUTION:
2432                 /* nanosecond timestamp resolution */
2433                 *valp = 1L;
2434                 error = 0;
2435                 break;
2436         default:
2437                 error = fs_pathconf(vp, cmd, valp, cr, ct);
2438         }
2439         return (error);
2440 }
2441 
2442 
2443 struct vnodeops *tmp_vnodeops;
2444 
2445 const fs_operation_def_t tmp_vnodeops_template[] = {
2446         VOPNAME_OPEN,           { .vop_open = tmp_open },
2447         VOPNAME_CLOSE,          { .vop_close = tmp_close },
2448         VOPNAME_READ,           { .vop_read = tmp_read },
2449         VOPNAME_WRITE,          { .vop_write = tmp_write },
2450         VOPNAME_IOCTL,          { .vop_ioctl = tmp_ioctl },
2451         VOPNAME_GETATTR,        { .vop_getattr = tmp_getattr },
2452         VOPNAME_SETATTR,        { .vop_setattr = tmp_setattr },
2453         VOPNAME_ACCESS,         { .vop_access = tmp_access },
2454         VOPNAME_LOOKUP,         { .vop_lookup = tmp_lookup },
2455         VOPNAME_CREATE,         { .vop_create = tmp_create },
2456         VOPNAME_REMOVE,         { .vop_remove = tmp_remove },
2457         VOPNAME_LINK,           { .vop_link = tmp_link },
2458         VOPNAME_RENAME,         { .vop_rename = tmp_rename },
2459         VOPNAME_MKDIR,          { .vop_mkdir = tmp_mkdir },
2460         VOPNAME_RMDIR,          { .vop_rmdir = tmp_rmdir },
2461         VOPNAME_READDIR,        { .vop_readdir = tmp_readdir },
2462         VOPNAME_SYMLINK,        { .vop_symlink = tmp_symlink },
2463         VOPNAME_READLINK,       { .vop_readlink = tmp_readlink },
2464         VOPNAME_FSYNC,          { .vop_fsync = tmp_fsync },
2465         VOPNAME_INACTIVE,       { .vop_inactive = tmp_inactive },
2466         VOPNAME_FID,            { .vop_fid = tmp_fid },
2467         VOPNAME_RWLOCK,         { .vop_rwlock = tmp_rwlock },
2468         VOPNAME_RWUNLOCK,       { .vop_rwunlock = tmp_rwunlock },
2469         VOPNAME_SEEK,           { .vop_seek = tmp_seek },
2470         VOPNAME_SPACE,          { .vop_space = tmp_space },
2471         VOPNAME_GETPAGE,        { .vop_getpage = tmp_getpage },
2472         VOPNAME_PUTPAGE,        { .vop_putpage = tmp_putpage },
2473         VOPNAME_MAP,            { .vop_map = tmp_map },
2474         VOPNAME_ADDMAP,         { .vop_addmap = tmp_addmap },
2475         VOPNAME_DELMAP,         { .vop_delmap = tmp_delmap },
2476         VOPNAME_PATHCONF,       { .vop_pathconf = tmp_pathconf },
2477         VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
2478         NULL,                   NULL
2479 };