sys/kern/vfs_vnops.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  * (c) UNIX System Laboratories, Inc.
   5  * All or some portions of this file are derived from material licensed
   6  * to the University of California by American Telephone and Telegraph
   7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
   8  * the permission of UNIX System Laboratories, Inc.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in the
  17  *    documentation and/or other materials provided with the distribution.
  18  * 3. All advertising materials mentioning features or use of this software
  19  *    must display the following acknowledgement:
  20  *      This product includes software developed by the University of
  21  *      California, Berkeley and its contributors.
  22  * 4. Neither the name of the University nor the names of its contributors
  23  *    may be used to endorse or promote products derived from this software
  24  *    without specific prior written permission.
  25  *
  26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  36  * SUCH DAMAGE.
  37  *
  38  *      @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
  39  * $FreeBSD: src/sys/kern/vfs_vnops.c,v 1.87.2.13 2002/12/29 18:19:53 dillon Exp $
  40  * $DragonFly: src/sys/kern/vfs_vnops.c,v 1.58 2008/06/28 17:59:49 dillon Exp $
  41  */
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/fcntl.h>
  46 #include <sys/file.h>
  47 #include <sys/stat.h>
  48 #include <sys/proc.h>
  49 #include <sys/priv.h>
  50 #include <sys/mount.h>
  51 #include <sys/nlookup.h>
  52 #include <sys/vnode.h>
  53 #include <sys/buf.h>
  54 #include <sys/filio.h>
  55 #include <sys/ttycom.h>
  56 #include <sys/conf.h>
  57 #include <sys/syslog.h>
  58
  59 static int vn_closefile (struct file *fp);
  60 static int vn_ioctl (struct file *fp, u_long com, caddr_t data,
  61                 struct ucred *cred);
  62 static int vn_read (struct file *fp, struct uio *uio,
  63                 struct ucred *cred, int flags);
  64 static int svn_read (struct file *fp, struct uio *uio,
  65                 struct ucred *cred, int flags);
  66 static int vn_poll (struct file *fp, int events, struct ucred *cred);
  67 static int vn_kqfilter (struct file *fp, struct knote *kn);
  68 static int vn_statfile (struct file *fp, struct stat *sb, struct ucred *cred);
  69 static int vn_write (struct file *fp, struct uio *uio,
  70                 struct ucred *cred, int flags);
  71 static int svn_write (struct file *fp, struct uio *uio,
  72                 struct ucred *cred, int flags);
  73
  74 struct fileops vnode_fileops = {
  75         .fo_read = vn_read,
  76         .fo_write = vn_write,
  77         .fo_ioctl = vn_ioctl,
  78         .fo_poll = vn_poll,
  79         .fo_kqfilter = vn_kqfilter,
  80         .fo_stat = vn_statfile,
  81         .fo_close = vn_closefile,
  82         .fo_shutdown = nofo_shutdown
  83 };
  84
  85 struct fileops specvnode_fileops = {
  86         .fo_read = svn_read,
  87         .fo_write = svn_write,
  88         .fo_ioctl = vn_ioctl,
  89         .fo_poll = vn_poll,
  90         .fo_kqfilter = vn_kqfilter,
  91         .fo_stat = vn_statfile,
  92         .fo_close = vn_closefile,
  93         .fo_shutdown = nofo_shutdown
  94 };
  95
  96 /*
  97  * Shortcut the device read/write.  This avoids a lot of vnode junk.
  98  * Basically the specfs vnops for read and write take the locked vnode,
  99  * unlock it (because we can't hold the vnode locked while reading or writing
 100  * a device which may block indefinitely), issues the device operation, then
 101  * relock the vnode before returning, plus other junk.  This bypasses all
 102  * of that and just does the device operation.
 103  */
 104 void
 105 vn_setspecops(struct file *fp)
 106 {
 107         if (vfs_fastdev && fp->f_ops == &vnode_fileops) {
 108                 fp->f_ops = &specvnode_fileops;
 109         }
 110 }
 111
 112 /*
 113  * Common code for vnode open operations.  Check permissions, and call
 114  * the VOP_NOPEN or VOP_NCREATE routine.
 115  *
 116  * The caller is responsible for setting up nd with nlookup_init() and
 117  * for cleaning it up with nlookup_done(), whether we return an error
 118  * or not.
 119  *
 120  * On success nd->nl_open_vp will hold a referenced and, if requested,
 121  * locked vnode.  A locked vnode is requested via NLC_LOCKVP.  If fp
 122  * is non-NULL the vnode will be installed in the file pointer.
 123  *
 124  * NOTE: The vnode is referenced just once on return whether or not it
 125  * is also installed in the file pointer.
 126  */
 127 int
 128 vn_open(struct nlookupdata *nd, struct file *fp, int fmode, int cmode)
 129 {
 130         struct vnode *vp;
 131         struct ucred *cred = nd->nl_cred;
 132         struct vattr vat;
 133         struct vattr *vap = &vat;
 134         int mode, error;
 135
 136         /*
 137          * Lookup the path and create or obtain the vnode.  After a
 138          * successful lookup a locked nd->nl_nch will be returned.
 139          *
 140          * The result of this section should be a locked vnode.
 141          *
 142          * XXX with only a little work we should be able to avoid locking
 143          * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set.
 144          */
 145         if (fmode & O_CREAT) {
 146                 /*
 147                  * CONDITIONAL CREATE FILE CASE
 148                  *
 149                  * Setting NLC_CREATE causes a negative hit to store
 150                  * the negative hit ncp and not return an error.  Then
 151                  * nc_error or nc_vp may be checked to see if the ncp
 152                  * represents a negative hit.  NLC_CREATE also requires
 153                  * write permission on the governing directory or EPERM
 154                  * is returned.
 155                  */
 156                 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
 157                         nd->nl_flags |= NLC_FOLLOW;
 158                 nd->nl_flags |= NLC_CREATE;
 159                 nd->nl_flags |= NLC_REFDVP;
 160                 bwillinode(1);
 161                 error = nlookup(nd);
 162         } else {
 163                 /*
 164                  * NORMAL OPEN FILE CASE
 165                  */
 166                 error = nlookup(nd);
 167         }
 168
 169         if (error)
 170                 return (error);
 171
 172         /*
 173          * split case to allow us to re-resolve and retry the ncp in case
 174          * we get ESTALE.
 175          */
 176 again:
 177         if (fmode & O_CREAT) {
 178                 if (nd->nl_nch.ncp->nc_vp == NULL) {
 179                         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
 180                                 return (error);
 181                         VATTR_NULL(vap);
 182                         vap->va_type = VREG;
 183                         vap->va_mode = cmode;
 184                         if (fmode & O_EXCL)
 185                                 vap->va_vaflags |= VA_EXCLUSIVE;
 186                         error = VOP_NCREATE(&nd->nl_nch, nd->nl_dvp, &vp,
 187                                             nd->nl_cred, vap);
 188                         if (error)
 189                                 return (error);
 190                         fmode &= ~O_TRUNC;
 191                         /* locked vnode is returned */
 192                 } else {
 193                         if (fmode & O_EXCL) {
 194                                 error = EEXIST;
 195                         } else {
 196                                 error = cache_vget(&nd->nl_nch, cred,
 197                                                     LK_EXCLUSIVE, &vp);
 198                         }
 199                         if (error)
 200                                 return (error);
 201                         fmode &= ~O_CREAT;
 202                 }
 203         } else {
 204                 error = cache_vget(&nd->nl_nch, cred, LK_EXCLUSIVE, &vp);
 205                 if (error)
 206                         return (error);
 207         }
 208
 209         /*
 210          * We have a locked vnode and ncp now.  Note that the ncp will
 211          * be cleaned up by the caller if nd->nl_nch is left intact.
 212          */
 213         if (vp->v_type == VLNK) {
 214                 error = EMLINK;
 215                 goto bad;
 216         }
 217         if (vp->v_type == VSOCK) {
 218                 error = EOPNOTSUPP;
 219                 goto bad;
 220         }
 221         if ((fmode & O_CREAT) == 0) {
 222                 mode = 0;
 223                 if (fmode & (FWRITE | O_TRUNC)) {
 224                         if (vp->v_type == VDIR) {
 225                                 error = EISDIR;
 226                                 goto bad;
 227                         }
 228                         error = vn_writechk(vp, &nd->nl_nch);
 229                         if (error) {
 230                                 /*
 231                                  * Special stale handling, re-resolve the
 232                                  * vnode.
 233                                  */
 234                                 if (error == ESTALE) {
 235                                         vput(vp);
 236                                         vp = NULL;
 237                                         cache_setunresolved(&nd->nl_nch);
 238                                         error = cache_resolve(&nd->nl_nch, cred);
 239                                         if (error == 0)
 240                                                 goto again;
 241                                 }
 242                                 goto bad;
 243                         }
 244                         mode |= VWRITE;
 245                 }
 246                 if (fmode & FREAD)
 247                         mode |= VREAD;
 248                 if (mode) {
 249                         error = VOP_ACCESS(vp, mode, cred);
 250                         if (error) {
 251                                 /*
 252                                  * Special stale handling, re-resolve the
 253                                  * vnode.
 254                                  */
 255                                 if (error == ESTALE) {
 256                                         vput(vp);
 257                                         vp = NULL;
 258                                         cache_setunresolved(&nd->nl_nch);
 259                                         error = cache_resolve(&nd->nl_nch, cred);
 260                                         if (error == 0)
 261                                                 goto again;
 262                                 }
 263                                 goto bad;
 264                         }
 265                 }
 266         }
 267         if (fmode & O_TRUNC) {
 268                 vn_unlock(vp);                          /* XXX */
 269                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);   /* XXX */
 270                 VATTR_NULL(vap);
 271                 vap->va_size = 0;
 272                 error = VOP_SETATTR(vp, vap, cred);
 273                 if (error)
 274                         goto bad;
 275         }
 276
 277         /*
 278          * Setup the fp so VOP_OPEN can override it.  No descriptor has been
 279          * associated with the fp yet so we own it clean.
 280          *
 281          * f_nchandle inherits nl_nch.  This used to be necessary only for
 282          * directories but now we do it unconditionally so f*() ops
 283          * such as fchmod() can access the actual namespace that was
 284          * used to open the file.
 285          */
 286         if (fp) {
 287                 fp->f_nchandle = nd->nl_nch;
 288                 cache_zero(&nd->nl_nch);
 289                 cache_unlock(&fp->f_nchandle);
 290         }
 291
 292         /*
 293          * Get rid of nl_nch.  vn_open does not return it (it returns the
 294          * vnode or the file pointer).  Note: we can't leave nl_nch locked
 295          * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g.
 296          * on /dev/ttyd0
 297          */
 298         if (nd->nl_nch.ncp)
 299                 cache_put(&nd->nl_nch);
 300
 301         error = VOP_OPEN(vp, fmode, cred, fp);
 302         if (error) {
 303                 /*
 304                  * setting f_ops to &badfileops will prevent the descriptor
 305                  * code from trying to close and release the vnode, since
 306                  * the open failed we do not want to call close.
 307                  */
 308                 if (fp) {
 309                         fp->f_data = NULL;
 310                         fp->f_ops = &badfileops;
 311                 }
 312                 goto bad;
 313         }
 314
 315 #if 0
 316         /*
 317          * Assert that VREG files have been setup for vmio.
 318          */
 319         KASSERT(vp->v_type != VREG || vp->v_object != NULL,
 320                 ("vn_open: regular file was not VMIO enabled!"));
 321 #endif
 322
 323         /*
 324          * Return the vnode.  XXX needs some cleaning up.  The vnode is
 325          * only returned in the fp == NULL case.
 326          */
 327         if (fp == NULL) {
 328                 nd->nl_open_vp = vp;
 329                 nd->nl_vp_fmode = fmode;
 330                 if ((nd->nl_flags & NLC_LOCKVP) == 0)
 331                         vn_unlock(vp);
 332         } else {
 333                 vput(vp);
 334         }
 335         return (0);
 336 bad:
 337         if (vp)
 338                 vput(vp);
 339         return (error);
 340 }
 341
 342 int
 343 vn_opendisk(const char *devname, int fmode, struct vnode **vpp)
 344 {
 345         struct vnode *vp;
 346         int error;
 347
 348         if (strncmp(devname, "/dev/", 5) == 0)
 349                 devname += 5;
 350         if ((vp = getsynthvnode(devname)) == NULL) {
 351                 error = ENODEV;
 352         } else {
 353                 error = VOP_OPEN(vp, fmode, proc0.p_ucred, NULL);
 354                 vn_unlock(vp);
 355                 if (error) {
 356                         vrele(vp);
 357                         vp = NULL;
 358                 }
 359         }
 360         *vpp = vp;
 361         return (error);
 362 }
 363
 364 /*
 365  * Check for write permissions on the specified vnode.  nch may be NULL.
 366  */
 367 int
 368 vn_writechk(struct vnode *vp, struct nchandle *nch)
 369 {
 370         /*
 371          * If there's shared text associated with
 372          * the vnode, try to free it up once.  If
 373          * we fail, we can't allow writing.
 374          */
 375         if (vp->v_flag & VTEXT)
 376                 return (ETXTBSY);
 377
 378         /*
 379          * If the vnode represents a regular file, check the mount
 380          * point via the nch.  This may be a different mount point
 381          * then the one embedded in the vnode (e.g. nullfs).
 382          *
 383          * We can still write to non-regular files (e.g. devices)
 384          * via read-only mounts.
 385          */
 386         if (nch && nch->ncp && vp->v_type == VREG)
 387                 return (ncp_writechk(nch));
 388         return (0);
 389 }
 390
 391 /*
 392  * Check whether the underlying mount is read-only.  The mount point
 393  * referenced by the namecache may be different from the mount point
 394  * used by the underlying vnode in the case of NULLFS, so a separate
 395  * check is needed.
 396  */
 397 int
 398 ncp_writechk(struct nchandle *nch)
 399 {
 400         if (nch->mount && (nch->mount->mnt_flag & MNT_RDONLY))
 401                 return (EROFS);
 402         return(0);
 403 }
 404
 405 /*
 406  * Vnode close call
 407  */
 408 int
 409 vn_close(struct vnode *vp, int flags)
 410 {
 411         int error;
 412
 413         error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 414         if (error == 0) {
 415                 error = VOP_CLOSE(vp, flags);
 416                 vn_unlock(vp);
 417         }
 418         vrele(vp);
 419         return (error);
 420 }
 421
 422 static __inline
 423 int
 424 sequential_heuristic(struct uio *uio, struct file *fp)
 425 {
 426         /*
 427          * Sequential heuristic - detect sequential operation
 428          */
 429         if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
 430             uio->uio_offset == fp->f_nextoff) {
 431                 int tmpseq = fp->f_seqcount;
 432                 /*
 433                  * XXX we assume that the filesystem block size is
 434                  * the default.  Not true, but still gives us a pretty
 435                  * good indicator of how sequential the read operations
 436                  * are.
 437                  */
 438                 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
 439                 if (tmpseq > IO_SEQMAX)
 440                         tmpseq = IO_SEQMAX;
 441                 fp->f_seqcount = tmpseq;
 442                 return(fp->f_seqcount << IO_SEQSHIFT);
 443         }
 444
 445         /*
 446          * Not sequential, quick draw-down of seqcount
 447          */
 448         if (fp->f_seqcount > 1)
 449                 fp->f_seqcount = 1;
 450         else
 451                 fp->f_seqcount = 0;
 452         return(0);
 453 }
 454
 455 /*
 456  * Package up an I/O request on a vnode into a uio and do it.
 457  */
 458 int
 459 vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, int len,
 460         off_t offset, enum uio_seg segflg, int ioflg,
 461         struct ucred *cred, int *aresid)
 462 {
 463         struct uio auio;
 464         struct iovec aiov;
 465         struct ccms_lock ccms_lock;
 466         int error;
 467
 468         if ((ioflg & IO_NODELOCKED) == 0)
 469                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 470         auio.uio_iov = &aiov;
 471         auio.uio_iovcnt = 1;
 472         aiov.iov_base = base;
 473         aiov.iov_len = len;
 474         auio.uio_resid = len;
 475         auio.uio_offset = offset;
 476         auio.uio_segflg = segflg;
 477         auio.uio_rw = rw;
 478         auio.uio_td = curthread;
 479         ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, &auio);
 480         if (rw == UIO_READ) {
 481                 error = VOP_READ(vp, &auio, ioflg, cred);
 482         } else {
 483                 error = VOP_WRITE(vp, &auio, ioflg, cred);
 484         }
 485         ccms_lock_put(&vp->v_ccms, &ccms_lock);
 486         if (aresid)
 487                 *aresid = auio.uio_resid;
 488         else
 489                 if (auio.uio_resid && error == 0)
 490                         error = EIO;
 491         if ((ioflg & IO_NODELOCKED) == 0)
 492                 vn_unlock(vp);
 493         return (error);
 494 }
 495
 496 /*
 497  * Package up an I/O request on a vnode into a uio and do it.  The I/O
 498  * request is split up into smaller chunks and we try to avoid saturating
 499  * the buffer cache while potentially holding a vnode locked, so we
 500  * check bwillwrite() before calling vn_rdwr().  We also call uio_yield()
 501  * to give other processes a chance to lock the vnode (either other processes
 502  * core'ing the same binary, or unrelated processes scanning the directory).
 503  */
 504 int
 505 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, caddr_t base, int len,
 506                  off_t offset, enum uio_seg segflg, int ioflg,
 507                  struct ucred *cred, int *aresid)
 508 {
 509         int error = 0;
 510
 511         do {
 512                 int chunk;
 513
 514                 /*
 515                  * Force `offset' to a multiple of MAXBSIZE except possibly
 516                  * for the first chunk, so that filesystems only need to
 517                  * write full blocks except possibly for the first and last
 518                  * chunks.
 519                  */
 520                 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
 521
 522                 if (chunk > len)
 523                         chunk = len;
 524                 if (vp->v_type == VREG) {
 525                         switch(rw) {
 526                         case UIO_READ:
 527                                 bwillread(chunk);
 528                                 break;
 529                         case UIO_WRITE:
 530                                 bwillwrite(chunk);
 531                                 break;
 532                         }
 533                 }
 534                 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
 535                             ioflg, cred, aresid);
 536                 len -= chunk;   /* aresid calc already includes length */
 537                 if (error)
 538                         break;
 539                 offset += chunk;
 540                 base += chunk;
 541                 uio_yield();
 542         } while (len);
 543         if (aresid)
 544                 *aresid += len;
 545         return (error);
 546 }
 547
 548 /*
 549  * MPALMOSTSAFE - acquires mplock
 550  */
 551 static int
 552 vn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 553 {
 554         struct ccms_lock ccms_lock;
 555         struct vnode *vp;
 556         int error, ioflag;
 557
 558         get_mplock();
 559         KASSERT(uio->uio_td == curthread,
 560                 ("uio_td %p is not td %p", uio->uio_td, curthread));
 561         vp = (struct vnode *)fp->f_data;
 562
 563         ioflag = 0;
 564         if (flags & O_FBLOCKING) {
 565                 /* ioflag &= ~IO_NDELAY; */
 566         } else if (flags & O_FNONBLOCKING) {
 567                 ioflag |= IO_NDELAY;
 568         } else if (fp->f_flag & FNONBLOCK) {
 569                 ioflag |= IO_NDELAY;
 570         }
 571         if (flags & O_FBUFFERED) {
 572                 /* ioflag &= ~IO_DIRECT; */
 573         } else if (flags & O_FUNBUFFERED) {
 574                 ioflag |= IO_DIRECT;
 575         } else if (fp->f_flag & O_DIRECT) {
 576                 ioflag |= IO_DIRECT;
 577         }
 578         vn_lock(vp, LK_SHARED | LK_RETRY);
 579         if ((flags & O_FOFFSET) == 0)
 580                 uio->uio_offset = fp->f_offset;
 581         ioflag |= sequential_heuristic(uio, fp);
 582
 583         ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio);
 584         error = VOP_READ(vp, uio, ioflag, cred);
 585         ccms_lock_put(&vp->v_ccms, &ccms_lock);
 586         if ((flags & O_FOFFSET) == 0)
 587                 fp->f_offset = uio->uio_offset;
 588         fp->f_nextoff = uio->uio_offset;
 589         vn_unlock(vp);
 590         rel_mplock();
 591         return (error);
 592 }
 593
 594 /*
 595  * Device-optimized file table vnode read routine.
 596  *
 597  * This bypasses the VOP table and talks directly to the device.  Most
 598  * filesystems just route to specfs and can make this optimization.
 599  *
 600  * MPALMOSTSAFE - acquires mplock
 601  */
 602 static int
 603 svn_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 604 {
 605         struct vnode *vp;
 606         int ioflag;
 607         int error;
 608         cdev_t dev;
 609
 610         get_mplock();
 611         KASSERT(uio->uio_td == curthread,
 612                 ("uio_td %p is not td %p", uio->uio_td, curthread));
 613
 614         vp = (struct vnode *)fp->f_data;
 615         if (vp == NULL || vp->v_type == VBAD) {
 616                 error = EBADF;
 617                 goto done;
 618         }
 619
 620         if ((dev = vp->v_rdev) == NULL) {
 621                 error = EBADF;
 622                 goto done;
 623         }
 624         reference_dev(dev);
 625
 626         if (uio->uio_resid == 0) {
 627                 error = 0;
 628                 goto done;
 629         }
 630         if ((flags & O_FOFFSET) == 0)
 631                 uio->uio_offset = fp->f_offset;
 632
 633         ioflag = 0;
 634         if (flags & O_FBLOCKING) {
 635                 /* ioflag &= ~IO_NDELAY; */
 636         } else if (flags & O_FNONBLOCKING) {
 637                 ioflag |= IO_NDELAY;
 638         } else if (fp->f_flag & FNONBLOCK) {
 639                 ioflag |= IO_NDELAY;
 640         }
 641         if (flags & O_FBUFFERED) {
 642                 /* ioflag &= ~IO_DIRECT; */
 643         } else if (flags & O_FUNBUFFERED) {
 644                 ioflag |= IO_DIRECT;
 645         } else if (fp->f_flag & O_DIRECT) {
 646                 ioflag |= IO_DIRECT;
 647         }
 648         ioflag |= sequential_heuristic(uio, fp);
 649
 650         error = dev_dread(dev, uio, ioflag);
 651
 652         release_dev(dev);
 653         if ((flags & O_FOFFSET) == 0)
 654                 fp->f_offset = uio->uio_offset;
 655         fp->f_nextoff = uio->uio_offset;
 656 done:
 657         rel_mplock();
 658         return (error);
 659 }
 660
 661 /*
 662  * MPALMOSTSAFE - acquires mplock
 663  */
 664 static int
 665 vn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 666 {
 667         struct ccms_lock ccms_lock;
 668         struct vnode *vp;
 669         int error, ioflag;
 670
 671         get_mplock();
 672         KASSERT(uio->uio_td == curthread,
 673                 ("uio_td %p is not p %p", uio->uio_td, curthread));
 674         vp = (struct vnode *)fp->f_data;
 675 #if 0
 676         /* VOP_WRITE should handle this now */
 677         if (vp->v_type == VREG || vp->v_type == VDATABASE)
 678                 bwillwrite();
 679 #endif
 680         vp = (struct vnode *)fp->f_data;        /* XXX needed? */
 681
 682         ioflag = IO_UNIT;
 683         if (vp->v_type == VREG &&
 684            ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
 685                 ioflag |= IO_APPEND;
 686         }
 687
 688         if (flags & O_FBLOCKING) {
 689                 /* ioflag &= ~IO_NDELAY; */
 690         } else if (flags & O_FNONBLOCKING) {
 691                 ioflag |= IO_NDELAY;
 692         } else if (fp->f_flag & FNONBLOCK) {
 693                 ioflag |= IO_NDELAY;
 694         }
 695         if (flags & O_FBUFFERED) {
 696                 /* ioflag &= ~IO_DIRECT; */
 697         } else if (flags & O_FUNBUFFERED) {
 698                 ioflag |= IO_DIRECT;
 699         } else if (fp->f_flag & O_DIRECT) {
 700                 ioflag |= IO_DIRECT;
 701         }
 702         if (flags & O_FASYNCWRITE) {
 703                 /* ioflag &= ~IO_SYNC; */
 704         } else if (flags & O_FSYNCWRITE) {
 705                 ioflag |= IO_SYNC;
 706         } else if (fp->f_flag & O_FSYNC) {
 707                 ioflag |= IO_SYNC;
 708         }
 709
 710         if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
 711                 ioflag |= IO_SYNC;
 712         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 713         if ((flags & O_FOFFSET) == 0)
 714                 uio->uio_offset = fp->f_offset;
 715         ioflag |= sequential_heuristic(uio, fp);
 716         ccms_lock_get_uio(&vp->v_ccms, &ccms_lock, uio);
 717         error = VOP_WRITE(vp, uio, ioflag, cred);
 718         ccms_lock_put(&vp->v_ccms, &ccms_lock);
 719         if ((flags & O_FOFFSET) == 0)
 720                 fp->f_offset = uio->uio_offset;
 721         fp->f_nextoff = uio->uio_offset;
 722         vn_unlock(vp);
 723         rel_mplock();
 724         return (error);
 725 }
 726
 727 /*
 728  * Device-optimized file table vnode write routine.
 729  *
 730  * This bypasses the VOP table and talks directly to the device.  Most
 731  * filesystems just route to specfs and can make this optimization.
 732  *
 733  * MPALMOSTSAFE - acquires mplock
 734  */
 735 static int
 736 svn_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 737 {
 738         struct vnode *vp;
 739         int ioflag;
 740         int error;
 741         cdev_t dev;
 742
 743         get_mplock();
 744         KASSERT(uio->uio_td == curthread,
 745                 ("uio_td %p is not p %p", uio->uio_td, curthread));
 746
 747         vp = (struct vnode *)fp->f_data;
 748         if (vp == NULL || vp->v_type == VBAD) {
 749                 error = EBADF;
 750                 goto done;
 751         }
 752         if (vp->v_type == VREG)
 753                 bwillwrite(uio->uio_resid);
 754         vp = (struct vnode *)fp->f_data;        /* XXX needed? */
 755
 756         if ((dev = vp->v_rdev) == NULL) {
 757                 error = EBADF;
 758                 goto done;
 759         }
 760         reference_dev(dev);
 761
 762         if ((flags & O_FOFFSET) == 0)
 763                 uio->uio_offset = fp->f_offset;
 764
 765         ioflag = IO_UNIT;
 766         if (vp->v_type == VREG &&
 767            ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
 768                 ioflag |= IO_APPEND;
 769         }
 770
 771         if (flags & O_FBLOCKING) {
 772                 /* ioflag &= ~IO_NDELAY; */
 773         } else if (flags & O_FNONBLOCKING) {
 774                 ioflag |= IO_NDELAY;
 775         } else if (fp->f_flag & FNONBLOCK) {
 776                 ioflag |= IO_NDELAY;
 777         }
 778         if (flags & O_FBUFFERED) {
 779                 /* ioflag &= ~IO_DIRECT; */
 780         } else if (flags & O_FUNBUFFERED) {
 781                 ioflag |= IO_DIRECT;
 782         } else if (fp->f_flag & O_DIRECT) {
 783                 ioflag |= IO_DIRECT;
 784         }
 785         if (flags & O_FASYNCWRITE) {
 786                 /* ioflag &= ~IO_SYNC; */
 787         } else if (flags & O_FSYNCWRITE) {
 788                 ioflag |= IO_SYNC;
 789         } else if (fp->f_flag & O_FSYNC) {
 790                 ioflag |= IO_SYNC;
 791         }
 792
 793         if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
 794                 ioflag |= IO_SYNC;
 795         ioflag |= sequential_heuristic(uio, fp);
 796
 797         error = dev_dwrite(dev, uio, ioflag);
 798
 799         release_dev(dev);
 800         if ((flags & O_FOFFSET) == 0)
 801                 fp->f_offset = uio->uio_offset;
 802         fp->f_nextoff = uio->uio_offset;
 803 done:
 804         rel_mplock();
 805         return (error);
 806 }
 807
 808 /*
 809  * MPALMOSTSAFE - acquires mplock
 810  */
 811 static int
 812 vn_statfile(struct file *fp, struct stat *sb, struct ucred *cred)
 813 {
 814         struct vnode *vp;
 815         int error;
 816
 817         get_mplock();
 818         vp = (struct vnode *)fp->f_data;
 819         error = vn_stat(vp, sb, cred);
 820         rel_mplock();
 821         return (error);
 822 }
 823
 824 int
 825 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *cred)
 826 {
 827         struct vattr vattr;
 828         struct vattr *vap;
 829         int error;
 830         u_short mode;
 831         cdev_t dev;
 832
 833         vap = &vattr;
 834         error = VOP_GETATTR(vp, vap);
 835         if (error)
 836                 return (error);
 837
 838         /*
 839          * Zero the spare stat fields
 840          */
 841         sb->st_lspare = 0;
 842         sb->st_qspare = 0;
 843
 844         /*
 845          * Copy from vattr table
 846          */
 847         if (vap->va_fsid != VNOVAL)
 848                 sb->st_dev = vap->va_fsid;
 849         else
 850                 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
 851         sb->st_ino = vap->va_fileid;
 852         mode = vap->va_mode;
 853         switch (vap->va_type) {
 854         case VREG:
 855                 mode |= S_IFREG;
 856                 break;
 857         case VDATABASE:
 858                 mode |= S_IFDB;
 859                 break;
 860         case VDIR:
 861                 mode |= S_IFDIR;
 862                 break;
 863         case VBLK:
 864                 mode |= S_IFBLK;
 865                 break;
 866         case VCHR:
 867                 mode |= S_IFCHR;
 868                 break;
 869         case VLNK:
 870                 mode |= S_IFLNK;
 871                 /* This is a cosmetic change, symlinks do not have a mode. */
 872                 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW)
 873                         sb->st_mode &= ~ACCESSPERMS;    /* 0000 */
 874                 else
 875                         sb->st_mode |= ACCESSPERMS;     /* 0777 */
 876                 break;
 877         case VSOCK:
 878                 mode |= S_IFSOCK;
 879                 break;
 880         case VFIFO:
 881                 mode |= S_IFIFO;
 882                 break;
 883         default:
 884                 return (EBADF);
 885         }
 886         sb->st_mode = mode;
 887         if (vap->va_nlink > (nlink_t)-1)
 888                 sb->st_nlink = (nlink_t)-1;
 889         else
 890                 sb->st_nlink = vap->va_nlink;
 891         sb->st_uid = vap->va_uid;
 892         sb->st_gid = vap->va_gid;
 893         sb->st_rdev = makeudev(vap->va_rmajor, vap->va_rminor);
 894         sb->st_size = vap->va_size;
 895         sb->st_atimespec = vap->va_atime;
 896         sb->st_mtimespec = vap->va_mtime;
 897         sb->st_ctimespec = vap->va_ctime;
 898
 899         /*
 900          * A VCHR and VBLK device may track the last access and last modified
 901          * time independantly of the filesystem.  This is particularly true
 902          * because device read and write calls may bypass the filesystem.
 903          */
 904         if (vp->v_type == VCHR || vp->v_type == VBLK) {
 905                 dev = vp->v_rdev;
 906                 if (dev != NULL) {
 907                         if (dev->si_lastread) {
 908                                 sb->st_atimespec.tv_sec = dev->si_lastread;
 909                                 sb->st_atimespec.tv_nsec = 0;
 910                         }
 911                         if (dev->si_lastwrite) {
 912                                 sb->st_atimespec.tv_sec = dev->si_lastwrite;
 913                                 sb->st_atimespec.tv_nsec = 0;
 914                         }
 915                 }
 916         }
 917
 918         /*
 919          * According to www.opengroup.org, the meaning of st_blksize is
 920          *   "a filesystem-specific preferred I/O block size for this
 921          *    object.  In some filesystem types, this may vary from file
 922          *    to file"
 923          * Default to PAGE_SIZE after much discussion.
 924          */
 925
 926         if (vap->va_type == VREG) {
 927                 sb->st_blksize = vap->va_blocksize;
 928         } else if (vn_isdisk(vp, NULL)) {
 929                 /*
 930                  * XXX this is broken.  If the device is not yet open (aka
 931                  * stat() call, aka v_rdev == NULL), how are we supposed
 932                  * to get a valid block size out of it?
 933                  */
 934                 dev = vp->v_rdev;
 935                 if (dev == NULL && vp->v_type == VCHR) {
 936                         dev = get_dev(vp->v_umajor, vp->v_uminor);
 937                 }
 938                 sb->st_blksize = dev->si_bsize_best;
 939                 if (sb->st_blksize < dev->si_bsize_phys)
 940                         sb->st_blksize = dev->si_bsize_phys;
 941                 if (sb->st_blksize < BLKDEV_IOSIZE)
 942                         sb->st_blksize = BLKDEV_IOSIZE;
 943         } else {
 944                 sb->st_blksize = PAGE_SIZE;
 945         }
 946
 947         sb->st_flags = vap->va_flags;
 948
 949         error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
 950         if (error)
 951                 sb->st_gen = 0;
 952         else
 953                 sb->st_gen = (u_int32_t)vap->va_gen;
 954
 955         sb->st_blocks = vap->va_bytes / S_BLKSIZE;
 956         sb->st_fsmid = vap->va_fsmid;
 957         return (0);
 958 }
 959
 960 /*
 961  * MPALMOSTSAFE - acquires mplock
 962  */
 963 static int
 964 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *ucred)
 965 {
 966         struct vnode *vp = ((struct vnode *)fp->f_data);
 967         struct vnode *ovp;
 968         struct vattr vattr;
 969         int error;
 970
 971         get_mplock();
 972
 973         switch (vp->v_type) {
 974         case VREG:
 975         case VDIR:
 976                 if (com == FIONREAD) {
 977                         error = VOP_GETATTR(vp, &vattr);
 978                         if (error)
 979                                 break;
 980                         *(int *)data = vattr.va_size - fp->f_offset;
 981                         error = 0;
 982                         break;
 983                 }
 984                 if (com == FIOASYNC) {                          /* XXX */
 985                         error = 0;                              /* XXX */
 986                         break;
 987                 }
 988                 /* fall into ... */
 989         default:
 990 #if 0
 991                 return (ENOTTY);
 992 #endif
 993         case VFIFO:
 994         case VCHR:
 995         case VBLK:
 996                 if (com == FIODTYPE) {
 997                         if (vp->v_type != VCHR && vp->v_type != VBLK) {
 998                                 error = ENOTTY;
 999                                 break;
1000                         }
1001                         *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK;
1002                         error = 0;
1003                         break;
1004                 }
1005                 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred);
1006                 if (error == 0 && com == TIOCSCTTY) {
1007                         struct proc *p = curthread->td_proc;
1008                         struct session *sess;
1009
1010                         if (p == NULL) {
1011                                 error = ENOTTY;
1012                                 break;
1013                         }
1014
1015                         sess = p->p_session;
1016                         /* Do nothing if reassigning same control tty */
1017                         if (sess->s_ttyvp == vp) {
1018                                 error = 0;
1019                                 break;
1020                         }
1021
1022                         /* Get rid of reference to old control tty */
1023                         ovp = sess->s_ttyvp;
1024                         vref(vp);
1025                         sess->s_ttyvp = vp;
1026                         if (ovp)
1027                                 vrele(ovp);
1028                 }
1029                 break;
1030         }
1031         rel_mplock();
1032         return (error);
1033 }
1034
1035 /*
1036  * MPALMOSTSAFE - acquires mplock
1037  */
1038 static int
1039 vn_poll(struct file *fp, int events, struct ucred *cred)
1040 {
1041         int error;
1042
1043         get_mplock();
1044         error = VOP_POLL(((struct vnode *)fp->f_data), events, cred);
1045         rel_mplock();
1046         return (error);
1047 }
1048
1049 /*
1050  * Check that the vnode is still valid, and if so
1051  * acquire requested lock.
1052  */
1053 int
1054 #ifndef DEBUG_LOCKS
1055 vn_lock(struct vnode *vp, int flags)
1056 #else
1057 debug_vn_lock(struct vnode *vp, int flags, const char *filename, int line)
1058 #endif
1059 {
1060         int error;
1061
1062         do {
1063 #ifdef  DEBUG_LOCKS
1064                 vp->filename = filename;
1065                 vp->line = line;
1066                 error = debuglockmgr(&vp->v_lock, flags,
1067                                      "vn_lock", filename, line);
1068 #else
1069                 error = lockmgr(&vp->v_lock, flags);
1070 #endif
1071                 if (error == 0)
1072                         break;
1073         } while (flags & LK_RETRY);
1074
1075         /*
1076          * Because we (had better!) have a ref on the vnode, once it
1077          * goes to VRECLAIMED state it will not be recycled until all
1078          * refs go away.  So we can just check the flag.
1079          */
1080         if (error == 0 && (vp->v_flag & VRECLAIMED)) {
1081                 lockmgr(&vp->v_lock, LK_RELEASE);
1082                 error = ENOENT;
1083         }
1084         return (error);
1085 }
1086
1087 void
1088 vn_unlock(struct vnode *vp)
1089 {
1090         lockmgr(&vp->v_lock, LK_RELEASE);
1091 }
1092
1093 int
1094 vn_islocked(struct vnode *vp)
1095 {
1096         return (lockstatus(&vp->v_lock, curthread));
1097 }
1098
1099 /*
1100  * MPALMOSTSAFE - acquires mplock
1101  */
1102 static int
1103 vn_closefile(struct file *fp)
1104 {
1105         int error;
1106
1107         get_mplock();
1108         fp->f_ops = &badfileops;
1109         error = vn_close(((struct vnode *)fp->f_data), fp->f_flag);
1110         rel_mplock();
1111         return (error);
1112 }
1113
1114 /*
1115  * MPALMOSTSAFE - acquires mplock
1116  */
1117 static int
1118 vn_kqfilter(struct file *fp, struct knote *kn)
1119 {
1120         int error;
1121
1122         get_mplock();
1123         error = VOP_KQFILTER(((struct vnode *)fp->f_data), kn);
1124         rel_mplock();
1125         return (error);
1126 }