sys/kern/kern_descrip.c

   1 /*
   2  * Copyright (c) 2005 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Jeffrey Hsu.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  *
  35  * Copyright (c) 1982, 1986, 1989, 1991, 1993
  36  *      The Regents of the University of California.  All rights reserved.
  37  * (c) UNIX System Laboratories, Inc.
  38  * All or some portions of this file are derived from material licensed
  39  * to the University of California by American Telephone and Telegraph
  40  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
  41  * the permission of UNIX System Laboratories, Inc.
  42  *
  43  * Redistribution and use in source and binary forms, with or without
  44  * modification, are permitted provided that the following conditions
  45  * are met:
  46  * 1. Redistributions of source code must retain the above copyright
  47  *    notice, this list of conditions and the following disclaimer.
  48  * 2. Redistributions in binary form must reproduce the above copyright
  49  *    notice, this list of conditions and the following disclaimer in the
  50  *    documentation and/or other materials provided with the distribution.
  51  * 3. All advertising materials mentioning features or use of this software
  52  *    must display the following acknowledgement:
  53  *      This product includes software developed by the University of
  54  *      California, Berkeley and its contributors.
  55  * 4. Neither the name of the University nor the names of its contributors
  56  *    may be used to endorse or promote products derived from this software
  57  *    without specific prior written permission.
  58  *
  59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  69  * SUCH DAMAGE.
  70  *
  71  *      @(#)kern_descrip.c      8.6 (Berkeley) 4/19/94
  72  * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
  73  * $DragonFly: src/sys/kern/kern_descrip.c,v 1.79 2008/08/31 13:18:28 aggelos Exp $
  74  */
  75
  76 #include "opt_compat.h"
  77 #include <sys/param.h>
  78 #include <sys/systm.h>
  79 #include <sys/malloc.h>
  80 #include <sys/sysproto.h>
  81 #include <sys/conf.h>
  82 #include <sys/device.h>
  83 #include <sys/filedesc.h>
  84 #include <sys/kernel.h>
  85 #include <sys/sysctl.h>
  86 #include <sys/vnode.h>
  87 #include <sys/proc.h>
  88 #include <sys/nlookup.h>
  89 #include <sys/file.h>
  90 #include <sys/stat.h>
  91 #include <sys/filio.h>
  92 #include <sys/fcntl.h>
  93 #include <sys/unistd.h>
  94 #include <sys/resourcevar.h>
  95 #include <sys/event.h>
  96 #include <sys/kern_syscall.h>
  97 #include <sys/kcore.h>
  98 #include <sys/kinfo.h>
  99 #include <sys/un.h>
 100
 101 #include <vm/vm.h>
 102 #include <vm/vm_extern.h>
 103
 104 #include <sys/thread2.h>
 105 #include <sys/file2.h>
 106 #include <sys/spinlock2.h>
 107
 108 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
 109 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
 110 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
 111 static int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp);
 112 static void ffree(struct file *fp);
 113
 114 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
 115 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
 116                      "file desc to leader structures");
 117 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
 118 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
 119
 120 static   d_open_t  fdopen;
 121 #define NUMFDESC 64
 122
 123 #define CDEV_MAJOR 22
 124 static struct dev_ops fildesc_ops = {
 125         { "FD", CDEV_MAJOR, 0 },
 126         .d_open =       fdopen,
 127 };
 128
 129 static int badfo_readwrite (struct file *fp, struct uio *uio,
 130     struct ucred *cred, int flags);
 131 static int badfo_ioctl (struct file *fp, u_long com, caddr_t data,
 132     struct ucred *cred);
 133 static int badfo_poll (struct file *fp, int events, struct ucred *cred);
 134 static int badfo_kqfilter (struct file *fp, struct knote *kn);
 135 static int badfo_stat (struct file *fp, struct stat *sb, struct ucred *cred);
 136 static int badfo_close (struct file *fp);
 137 static int badfo_shutdown (struct file *fp, int how);
 138
 139 /*
 140  * Descriptor management.
 141  */
 142 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
 143 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
 144 static int nfiles;              /* actual number of open files */
 145 extern int cmask;
 146
 147 /*
 148  * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
 149  *
 150  * MPSAFE - must be called with fdp->fd_spin exclusively held
 151  */
 152 static __inline
 153 void
 154 fdfixup_locked(struct filedesc *fdp, int fd)
 155 {
 156         if (fd < fdp->fd_freefile) {
 157                fdp->fd_freefile = fd;
 158         }
 159         while (fdp->fd_lastfile >= 0 &&
 160                fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
 161                fdp->fd_files[fdp->fd_lastfile].reserved == 0
 162         ) {
 163                 --fdp->fd_lastfile;
 164         }
 165 }
 166
 167 /*
 168  * System calls on descriptors.
 169  *
 170  * MPSAFE
 171  */
 172 int
 173 sys_getdtablesize(struct getdtablesize_args *uap)
 174 {
 175         struct proc *p = curproc;
 176         struct plimit *limit = p->p_limit;
 177
 178         spin_lock_rd(&limit->p_spin);
 179         uap->sysmsg_result =
 180             min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
 181         spin_unlock_rd(&limit->p_spin);
 182         return (0);
 183 }
 184
 185 /*
 186  * Duplicate a file descriptor to a particular value.
 187  *
 188  * note: keep in mind that a potential race condition exists when closing
 189  * descriptors from a shared descriptor table (via rfork).
 190  *
 191  * MPSAFE
 192  */
 193 int
 194 sys_dup2(struct dup2_args *uap)
 195 {
 196         int error;
 197         int fd = 0;
 198
 199         error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
 200         uap->sysmsg_fds[0] = fd;
 201
 202         return (error);
 203 }
 204
 205 /*
 206  * Duplicate a file descriptor.
 207  *
 208  * MPSAFE
 209  */
 210 int
 211 sys_dup(struct dup_args *uap)
 212 {
 213         int error;
 214         int fd = 0;
 215
 216         error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
 217         uap->sysmsg_fds[0] = fd;
 218
 219         return (error);
 220 }
 221
 222 /*
 223  * MPALMOSTSAFE - acquires mplock for fp operations
 224  */
 225 int
 226 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
 227 {
 228         struct thread *td = curthread;
 229         struct proc *p = td->td_proc;
 230         struct file *fp;
 231         struct vnode *vp;
 232         u_int newmin;
 233         u_int oflags;
 234         u_int nflags;
 235         int tmp, error, flg = F_POSIX;
 236
 237         KKASSERT(p);
 238
 239         /*
 240          * Operations on file descriptors that do not require a file pointer.
 241          */
 242         switch (cmd) {
 243         case F_GETFD:
 244                 error = fgetfdflags(p->p_fd, fd, &tmp);
 245                 if (error == 0)
 246                         dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
 247                 return (error);
 248
 249         case F_SETFD:
 250                 if (dat->fc_cloexec & FD_CLOEXEC)
 251                         error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
 252                 else
 253                         error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
 254                 return (error);
 255         case F_DUPFD:
 256                 newmin = dat->fc_fd;
 257                 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
 258                 return (error);
 259         default:
 260                 break;
 261         }
 262
 263         /*
 264          * Operations on file pointers
 265          */
 266         if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
 267                 return (EBADF);
 268
 269         get_mplock();
 270         switch (cmd) {
 271         case F_GETFL:
 272                 dat->fc_flags = OFLAGS(fp->f_flag);
 273                 error = 0;
 274                 break;
 275
 276         case F_SETFL:
 277                 oflags = fp->f_flag;
 278                 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
 279                 nflags |= oflags & ~FCNTLFLAGS;
 280
 281                 error = 0;
 282                 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
 283                         error = EINVAL;
 284                 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
 285                         tmp = nflags & FASYNC;
 286                         error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred);
 287                 }
 288                 if (error == 0)
 289                         fp->f_flag = nflags;
 290                 break;
 291
 292         case F_GETOWN:
 293                 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, cred);
 294                 break;
 295
 296         case F_SETOWN:
 297                 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, cred);
 298                 break;
 299
 300         case F_SETLKW:
 301                 flg |= F_WAIT;
 302                 /* Fall into F_SETLK */
 303
 304         case F_SETLK:
 305                 if (fp->f_type != DTYPE_VNODE) {
 306                         error = EBADF;
 307                         break;
 308                 }
 309                 vp = (struct vnode *)fp->f_data;
 310
 311                 /*
 312                  * copyin/lockop may block
 313                  */
 314                 if (dat->fc_flock.l_whence == SEEK_CUR)
 315                         dat->fc_flock.l_start += fp->f_offset;
 316
 317                 switch (dat->fc_flock.l_type) {
 318                 case F_RDLCK:
 319                         if ((fp->f_flag & FREAD) == 0) {
 320                                 error = EBADF;
 321                                 break;
 322                         }
 323                         p->p_leader->p_flag |= P_ADVLOCK;
 324                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
 325                             &dat->fc_flock, flg);
 326                         break;
 327                 case F_WRLCK:
 328                         if ((fp->f_flag & FWRITE) == 0) {
 329                                 error = EBADF;
 330                                 break;
 331                         }
 332                         p->p_leader->p_flag |= P_ADVLOCK;
 333                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
 334                             &dat->fc_flock, flg);
 335                         break;
 336                 case F_UNLCK:
 337                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
 338                                 &dat->fc_flock, F_POSIX);
 339                         break;
 340                 default:
 341                         error = EINVAL;
 342                         break;
 343                 }
 344
 345                 /*
 346                  * It is possible to race a close() on the descriptor while
 347                  * we were blocked getting the lock.  If this occurs the
 348                  * close might not have caught the lock.
 349                  */
 350                 if (checkfpclosed(p->p_fd, fd, fp)) {
 351                         dat->fc_flock.l_whence = SEEK_SET;
 352                         dat->fc_flock.l_start = 0;
 353                         dat->fc_flock.l_len = 0;
 354                         dat->fc_flock.l_type = F_UNLCK;
 355                         (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
 356                                            F_UNLCK, &dat->fc_flock, F_POSIX);
 357                 }
 358                 break;
 359
 360         case F_GETLK:
 361                 if (fp->f_type != DTYPE_VNODE) {
 362                         error = EBADF;
 363                         break;
 364                 }
 365                 vp = (struct vnode *)fp->f_data;
 366                 /*
 367                  * copyin/lockop may block
 368                  */
 369                 if (dat->fc_flock.l_type != F_RDLCK &&
 370                     dat->fc_flock.l_type != F_WRLCK &&
 371                     dat->fc_flock.l_type != F_UNLCK) {
 372                         error = EINVAL;
 373                         break;
 374                 }
 375                 if (dat->fc_flock.l_whence == SEEK_CUR)
 376                         dat->fc_flock.l_start += fp->f_offset;
 377                 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
 378                             &dat->fc_flock, F_POSIX);
 379                 break;
 380         default:
 381                 error = EINVAL;
 382                 break;
 383         }
 384         rel_mplock();
 385
 386         fdrop(fp);
 387         return (error);
 388 }
 389
 390 /*
 391  * The file control system call.
 392  *
 393  * MPSAFE
 394  */
 395 int
 396 sys_fcntl(struct fcntl_args *uap)
 397 {
 398         union fcntl_dat dat;
 399         int error;
 400
 401         switch (uap->cmd) {
 402         case F_DUPFD:
 403                 dat.fc_fd = uap->arg;
 404                 break;
 405         case F_SETFD:
 406                 dat.fc_cloexec = uap->arg;
 407                 break;
 408         case F_SETFL:
 409                 dat.fc_flags = uap->arg;
 410                 break;
 411         case F_SETOWN:
 412                 dat.fc_owner = uap->arg;
 413                 break;
 414         case F_SETLKW:
 415         case F_SETLK:
 416         case F_GETLK:
 417                 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
 418                                sizeof(struct flock));
 419                 if (error)
 420                         return (error);
 421                 break;
 422         }
 423
 424         error = kern_fcntl(uap->fd, uap->cmd, &dat, curproc->p_ucred);
 425
 426         if (error == 0) {
 427                 switch (uap->cmd) {
 428                 case F_DUPFD:
 429                         uap->sysmsg_result = dat.fc_fd;
 430                         break;
 431                 case F_GETFD:
 432                         uap->sysmsg_result = dat.fc_cloexec;
 433                         break;
 434                 case F_GETFL:
 435                         uap->sysmsg_result = dat.fc_flags;
 436                         break;
 437                 case F_GETOWN:
 438                         uap->sysmsg_result = dat.fc_owner;
 439                 case F_GETLK:
 440                         error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
 441                             sizeof(struct flock));
 442                         break;
 443                 }
 444         }
 445
 446         return (error);
 447 }
 448
 449 /*
 450  * Common code for dup, dup2, and fcntl(F_DUPFD).
 451  *
 452  * The type flag can be either DUP_FIXED or DUP_VARIABLE.  DUP_FIXED tells
 453  * kern_dup() to destructively dup over an existing file descriptor if new
 454  * is already open.  DUP_VARIABLE tells kern_dup() to find the lowest
 455  * unused file descriptor that is greater than or equal to new.
 456  *
 457  * MPSAFE
 458  */
 459 int
 460 kern_dup(enum dup_type type, int old, int new, int *res)
 461 {
 462         struct thread *td = curthread;
 463         struct proc *p = td->td_proc;
 464         struct filedesc *fdp = p->p_fd;
 465         struct file *fp;
 466         struct file *delfp;
 467         int oldflags;
 468         int holdleaders;
 469         int error, newfd;
 470
 471         /*
 472          * Verify that we have a valid descriptor to dup from and
 473          * possibly to dup to.
 474          */
 475 retry:
 476         spin_lock_wr(&fdp->fd_spin);
 477         if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
 478             new >= maxfilesperproc) {
 479                 spin_unlock_wr(&fdp->fd_spin);
 480                 return (EINVAL);
 481         }
 482         if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
 483                 spin_unlock_wr(&fdp->fd_spin);
 484                 return (EBADF);
 485         }
 486         if (type == DUP_FIXED && old == new) {
 487                 *res = new;
 488                 spin_unlock_wr(&fdp->fd_spin);
 489                 return (0);
 490         }
 491         fp = fdp->fd_files[old].fp;
 492         oldflags = fdp->fd_files[old].fileflags;
 493         fhold(fp);      /* MPSAFE - can be called with a spinlock held */
 494
 495         /*
 496          * Allocate a new descriptor if DUP_VARIABLE, or expand the table
 497          * if the requested descriptor is beyond the current table size.
 498          *
 499          * This can block.  Retry if the source descriptor no longer matches
 500          * or if our expectation in the expansion case races.
 501          *
 502          * If we are not expanding or allocating a new decriptor, then reset
 503          * the target descriptor to a reserved state so we have a uniform
 504          * setup for the next code block.
 505          */
 506         if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
 507                 spin_unlock_wr(&fdp->fd_spin);
 508                 error = fdalloc(p, new, &newfd);
 509                 spin_lock_wr(&fdp->fd_spin);
 510                 if (error) {
 511                         spin_unlock_wr(&fdp->fd_spin);
 512                         fdrop(fp);
 513                         return (error);
 514                 }
 515                 /*
 516                  * Check for ripout
 517                  */
 518                 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
 519                         fsetfd_locked(fdp, NULL, newfd);
 520                         spin_unlock_wr(&fdp->fd_spin);
 521                         fdrop(fp);
 522                         goto retry;
 523                 }
 524                 /*
 525                  * Check for expansion race
 526                  */
 527                 if (type != DUP_VARIABLE && new != newfd) {
 528                         fsetfd_locked(fdp, NULL, newfd);
 529                         spin_unlock_wr(&fdp->fd_spin);
 530                         fdrop(fp);
 531                         goto retry;
 532                 }
 533                 /*
 534                  * Check for ripout, newfd reused old (this case probably
 535                  * can't occur).
 536                  */
 537                 if (old == newfd) {
 538                         fsetfd_locked(fdp, NULL, newfd);
 539                         spin_unlock_wr(&fdp->fd_spin);
 540                         fdrop(fp);
 541                         goto retry;
 542                 }
 543                 new = newfd;
 544                 delfp = NULL;
 545         } else {
 546                 if (fdp->fd_files[new].reserved) {
 547                         spin_unlock_wr(&fdp->fd_spin);
 548                         fdrop(fp);
 549                         kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
 550                         tsleep(fdp, 0, "fdres", hz);
 551                         goto retry;
 552                 }
 553
 554                 /*
 555                  * If the target descriptor was never allocated we have
 556                  * to allocate it.  If it was we have to clean out the
 557                  * old descriptor.  delfp inherits the ref from the
 558                  * descriptor table.
 559                  */
 560                 delfp = fdp->fd_files[new].fp;
 561                 fdp->fd_files[new].fp = NULL;
 562                 fdp->fd_files[new].reserved = 1;
 563                 if (delfp == NULL) {
 564                         fdreserve_locked(fdp, new, 1);
 565                         if (new > fdp->fd_lastfile)
 566                                 fdp->fd_lastfile = new;
 567                 }
 568
 569         }
 570
 571         /*
 572          * NOTE: still holding an exclusive spinlock
 573          */
 574
 575         /*
 576          * If a descriptor is being overwritten we may hve to tell
 577          * fdfree() to sleep to ensure that all relevant process
 578          * leaders can be traversed in closef().
 579          */
 580         if (delfp != NULL && p->p_fdtol != NULL) {
 581                 fdp->fd_holdleaderscount++;
 582                 holdleaders = 1;
 583         } else {
 584                 holdleaders = 0;
 585         }
 586         KASSERT(delfp == NULL || type == DUP_FIXED,
 587                 ("dup() picked an open file"));
 588
 589         /*
 590          * Duplicate the source descriptor, update lastfile.  If the new
 591          * descriptor was not allocated and we aren't replacing an existing
 592          * descriptor we have to mark the descriptor as being in use.
 593          *
 594          * The fd_files[] array inherits fp's hold reference.
 595          */
 596         fsetfd_locked(fdp, fp, new);
 597         fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
 598         spin_unlock_wr(&fdp->fd_spin);
 599         fdrop(fp);
 600         *res = new;
 601
 602         /*
 603          * If we dup'd over a valid file, we now own the reference to it
 604          * and must dispose of it using closef() semantics (as if a
 605          * close() were performed on it).
 606          */
 607         if (delfp) {
 608                 closef(delfp, p);
 609                 if (holdleaders) {
 610                         spin_lock_wr(&fdp->fd_spin);
 611                         fdp->fd_holdleaderscount--;
 612                         if (fdp->fd_holdleaderscount == 0 &&
 613                             fdp->fd_holdleaderswakeup != 0) {
 614                                 fdp->fd_holdleaderswakeup = 0;
 615                                 spin_unlock_wr(&fdp->fd_spin);
 616                                 wakeup(&fdp->fd_holdleaderscount);
 617                         } else {
 618                                 spin_unlock_wr(&fdp->fd_spin);
 619                         }
 620                 }
 621         }
 622         return (0);
 623 }
 624
 625 /*
 626  * If sigio is on the list associated with a process or process group,
 627  * disable signalling from the device, remove sigio from the list and
 628  * free sigio.
 629  */
 630 void
 631 funsetown(struct sigio *sigio)
 632 {
 633         if (sigio == NULL)
 634                 return;
 635         crit_enter();
 636         *(sigio->sio_myref) = NULL;
 637         crit_exit();
 638         if (sigio->sio_pgid < 0) {
 639                 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
 640                              sigio, sio_pgsigio);
 641         } else /* if ((*sigiop)->sio_pgid > 0) */ {
 642                 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
 643                              sigio, sio_pgsigio);
 644         }
 645         crfree(sigio->sio_ucred);
 646         kfree(sigio, M_SIGIO);
 647 }
 648
 649 /* Free a list of sigio structures. */
 650 void
 651 funsetownlst(struct sigiolst *sigiolst)
 652 {
 653         struct sigio *sigio;
 654
 655         while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
 656                 funsetown(sigio);
 657 }
 658
 659 /*
 660  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
 661  *
 662  * After permission checking, add a sigio structure to the sigio list for
 663  * the process or process group.
 664  */
 665 int
 666 fsetown(pid_t pgid, struct sigio **sigiop)
 667 {
 668         struct proc *proc;
 669         struct pgrp *pgrp;
 670         struct sigio *sigio;
 671
 672         if (pgid == 0) {
 673                 funsetown(*sigiop);
 674                 return (0);
 675         }
 676         if (pgid > 0) {
 677                 proc = pfind(pgid);
 678                 if (proc == NULL)
 679                         return (ESRCH);
 680
 681                 /*
 682                  * Policy - Don't allow a process to FSETOWN a process
 683                  * in another session.
 684                  *
 685                  * Remove this test to allow maximum flexibility or
 686                  * restrict FSETOWN to the current process or process
 687                  * group for maximum safety.
 688                  */
 689                 if (proc->p_session != curproc->p_session)
 690                         return (EPERM);
 691
 692                 pgrp = NULL;
 693         } else /* if (pgid < 0) */ {
 694                 pgrp = pgfind(-pgid);
 695                 if (pgrp == NULL)
 696                         return (ESRCH);
 697
 698                 /*
 699                  * Policy - Don't allow a process to FSETOWN a process
 700                  * in another session.
 701                  *
 702                  * Remove this test to allow maximum flexibility or
 703                  * restrict FSETOWN to the current process or process
 704                  * group for maximum safety.
 705                  */
 706                 if (pgrp->pg_session != curproc->p_session)
 707                         return (EPERM);
 708
 709                 proc = NULL;
 710         }
 711         funsetown(*sigiop);
 712         sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
 713         if (pgid > 0) {
 714                 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
 715                 sigio->sio_proc = proc;
 716         } else {
 717                 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
 718                 sigio->sio_pgrp = pgrp;
 719         }
 720         sigio->sio_pgid = pgid;
 721         sigio->sio_ucred = crhold(curproc->p_ucred);
 722         /* It would be convenient if p_ruid was in ucred. */
 723         sigio->sio_ruid = curproc->p_ucred->cr_ruid;
 724         sigio->sio_myref = sigiop;
 725         crit_enter();
 726         *sigiop = sigio;
 727         crit_exit();
 728         return (0);
 729 }
 730
 731 /*
 732  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
 733  */
 734 pid_t
 735 fgetown(struct sigio *sigio)
 736 {
 737         return (sigio != NULL ? sigio->sio_pgid : 0);
 738 }
 739
 740 /*
 741  * Close many file descriptors.
 742  *
 743  * MPSAFE
 744  */
 745 int
 746 sys_closefrom(struct closefrom_args *uap)
 747 {
 748         return(kern_closefrom(uap->fd));
 749 }
 750
 751 /*
 752  * Close all file descriptors greater then or equal to fd
 753  *
 754  * MPSAFE
 755  */
 756 int
 757 kern_closefrom(int fd)
 758 {
 759         struct thread *td = curthread;
 760         struct proc *p = td->td_proc;
 761         struct filedesc *fdp;
 762
 763         KKASSERT(p);
 764         fdp = p->p_fd;
 765
 766         if (fd < 0)
 767                 return (EINVAL);
 768
 769         /*
 770          * NOTE: This function will skip unassociated descriptors and
 771          * reserved descriptors that have not yet been assigned.
 772          * fd_lastfile can change as a side effect of kern_close().
 773          */
 774         spin_lock_wr(&fdp->fd_spin);
 775         while (fd <= fdp->fd_lastfile) {
 776                 if (fdp->fd_files[fd].fp != NULL) {
 777                         spin_unlock_wr(&fdp->fd_spin);
 778                         /* ok if this races another close */
 779                         if (kern_close(fd) == EINTR)
 780                                 return (EINTR);
 781                         spin_lock_wr(&fdp->fd_spin);
 782                 }
 783                 ++fd;
 784         }
 785         spin_unlock_wr(&fdp->fd_spin);
 786         return (0);
 787 }
 788
 789 /*
 790  * Close a file descriptor.
 791  *
 792  * MPSAFE
 793  */
 794 int
 795 sys_close(struct close_args *uap)
 796 {
 797         return(kern_close(uap->fd));
 798 }
 799
 800 /*
 801  * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
 802  */
 803 int
 804 kern_close(int fd)
 805 {
 806         struct thread *td = curthread;
 807         struct proc *p = td->td_proc;
 808         struct filedesc *fdp;
 809         struct file *fp;
 810         int error;
 811         int holdleaders;
 812
 813         KKASSERT(p);
 814         fdp = p->p_fd;
 815
 816         spin_lock_wr(&fdp->fd_spin);
 817         if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
 818                 spin_unlock_wr(&fdp->fd_spin);
 819                 return (EBADF);
 820         }
 821         holdleaders = 0;
 822         if (p->p_fdtol != NULL) {
 823                 /*
 824                  * Ask fdfree() to sleep to ensure that all relevant
 825                  * process leaders can be traversed in closef().
 826                  */
 827                 fdp->fd_holdleaderscount++;
 828                 holdleaders = 1;
 829         }
 830
 831         /*
 832          * we now hold the fp reference that used to be owned by the descriptor
 833          * array.
 834          */
 835         spin_unlock_wr(&fdp->fd_spin);
 836         if (fd < fdp->fd_knlistsize) {
 837                 get_mplock();
 838                 if (fd < fdp->fd_knlistsize)
 839                         knote_fdclose(p, fd);
 840                 rel_mplock();
 841         }
 842         error = closef(fp, p);
 843         if (holdleaders) {
 844                 spin_lock_wr(&fdp->fd_spin);
 845                 fdp->fd_holdleaderscount--;
 846                 if (fdp->fd_holdleaderscount == 0 &&
 847                     fdp->fd_holdleaderswakeup != 0) {
 848                         fdp->fd_holdleaderswakeup = 0;
 849                         spin_unlock_wr(&fdp->fd_spin);
 850                         wakeup(&fdp->fd_holdleaderscount);
 851                 } else {
 852                         spin_unlock_wr(&fdp->fd_spin);
 853                 }
 854         }
 855         return (error);
 856 }
 857
 858 /*
 859  * shutdown_args(int fd, int how)
 860  */
 861 int
 862 kern_shutdown(int fd, int how)
 863 {
 864         struct thread *td = curthread;
 865         struct proc *p = td->td_proc;
 866         struct file *fp;
 867         int error;
 868
 869         KKASSERT(p);
 870
 871         if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
 872                 return (EBADF);
 873         error = fo_shutdown(fp, how);
 874         fdrop(fp);
 875
 876         return (error);
 877 }
 878
 879 int
 880 sys_shutdown(struct shutdown_args *uap)
 881 {
 882         int error;
 883
 884         error = kern_shutdown(uap->s, uap->how);
 885
 886         return (error);
 887 }
 888
 889 int
 890 kern_fstat(int fd, struct stat *ub)
 891 {
 892         struct thread *td = curthread;
 893         struct proc *p = td->td_proc;
 894         struct file *fp;
 895         int error;
 896
 897         KKASSERT(p);
 898
 899         if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
 900                 return (EBADF);
 901         error = fo_stat(fp, ub, p->p_ucred);
 902         fdrop(fp);
 903
 904         return (error);
 905 }
 906
 907 /*
 908  * Return status information about a file descriptor.
 909  */
 910 int
 911 sys_fstat(struct fstat_args *uap)
 912 {
 913         struct stat st;
 914         int error;
 915
 916         error = kern_fstat(uap->fd, &st);
 917
 918         if (error == 0)
 919                 error = copyout(&st, uap->sb, sizeof(st));
 920         return (error);
 921 }
 922
 923 /*
 924  * Return pathconf information about a file descriptor.
 925  */
 926 /* ARGSUSED */
 927 int
 928 sys_fpathconf(struct fpathconf_args *uap)
 929 {
 930         struct thread *td = curthread;
 931         struct proc *p = td->td_proc;
 932         struct file *fp;
 933         struct vnode *vp;
 934         int error = 0;
 935
 936         KKASSERT(p);
 937
 938         if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
 939                 return (EBADF);
 940
 941         switch (fp->f_type) {
 942         case DTYPE_PIPE:
 943         case DTYPE_SOCKET:
 944                 if (uap->name != _PC_PIPE_BUF) {
 945                         error = EINVAL;
 946                 } else {
 947                         uap->sysmsg_result = PIPE_BUF;
 948                         error = 0;
 949                 }
 950                 break;
 951         case DTYPE_FIFO:
 952         case DTYPE_VNODE:
 953                 vp = (struct vnode *)fp->f_data;
 954                 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
 955                 break;
 956         default:
 957                 error = EOPNOTSUPP;
 958                 break;
 959         }
 960         fdrop(fp);
 961         return(error);
 962 }
 963
 964 static int fdexpand;
 965 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
 966
 967 /*
 968  * Grow the file table so it can hold through descriptor (want).
 969  *
 970  * The fdp's spinlock must be held exclusively on entry and may be held
 971  * exclusively on return.  The spinlock may be cycled by the routine.
 972  *
 973  * MPSAFE
 974  */
 975 static void
 976 fdgrow_locked(struct filedesc *fdp, int want)
 977 {
 978         struct fdnode *newfiles;
 979         struct fdnode *oldfiles;
 980         int nf, extra;
 981
 982         nf = fdp->fd_nfiles;
 983         do {
 984                 /* nf has to be of the form 2^n - 1 */
 985                 nf = 2 * nf + 1;
 986         } while (nf <= want);
 987
 988         spin_unlock_wr(&fdp->fd_spin);
 989         newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
 990         spin_lock_wr(&fdp->fd_spin);
 991
 992         /*
 993          * We could have raced another extend while we were not holding
 994          * the spinlock.
 995          */
 996         if (fdp->fd_nfiles >= nf) {
 997                 spin_unlock_wr(&fdp->fd_spin);
 998                 kfree(newfiles, M_FILEDESC);
 999                 spin_lock_wr(&fdp->fd_spin);
1000                 return;
1001         }
1002         /*
1003          * Copy the existing ofile and ofileflags arrays
1004          * and zero the new portion of each array.
1005          */
1006         extra = nf - fdp->fd_nfiles;
1007         bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1008         bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1009
1010         oldfiles = fdp->fd_files;
1011         fdp->fd_files = newfiles;
1012         fdp->fd_nfiles = nf;
1013
1014         if (oldfiles != fdp->fd_builtin_files) {
1015                 spin_unlock_wr(&fdp->fd_spin);
1016                 kfree(oldfiles, M_FILEDESC);
1017                 spin_lock_wr(&fdp->fd_spin);
1018         }
1019         fdexpand++;
1020 }
1021
1022 /*
1023  * Number of nodes in right subtree, including the root.
1024  */
1025 static __inline int
1026 right_subtree_size(int n)
1027 {
1028         return (n ^ (n | (n + 1)));
1029 }
1030
1031 /*
1032  * Bigger ancestor.
1033  */
1034 static __inline int
1035 right_ancestor(int n)
1036 {
1037         return (n | (n + 1));
1038 }
1039
1040 /*
1041  * Smaller ancestor.
1042  */
1043 static __inline int
1044 left_ancestor(int n)
1045 {
1046         return ((n & (n + 1)) - 1);
1047 }
1048
1049 /*
1050  * Traverse the in-place binary tree buttom-up adjusting the allocation
1051  * count so scans can determine where free descriptors are located.
1052  *
1053  * MPSAFE - caller must be holding an exclusive spinlock on fdp
1054  */
1055 static
1056 void
1057 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1058 {
1059         while (fd >= 0) {
1060                 fdp->fd_files[fd].allocated += incr;
1061                 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1062                 fd = left_ancestor(fd);
1063         }
1064 }
1065
1066 /*
1067  * Reserve a file descriptor for the process.  If no error occurs, the
1068  * caller MUST at some point call fsetfd() or assign a file pointer
1069  * or dispose of the reservation.
1070  *
1071  * MPSAFE
1072  */
1073 int
1074 fdalloc(struct proc *p, int want, int *result)
1075 {
1076         struct filedesc *fdp = p->p_fd;
1077         int fd, rsize, rsum, node, lim;
1078
1079         spin_lock_rd(&p->p_limit->p_spin);
1080         lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1081         spin_unlock_rd(&p->p_limit->p_spin);
1082         if (want >= lim)
1083                 return (EMFILE);
1084         spin_lock_wr(&fdp->fd_spin);
1085         if (want >= fdp->fd_nfiles)
1086                 fdgrow_locked(fdp, want);
1087
1088         /*
1089          * Search for a free descriptor starting at the higher
1090          * of want or fd_freefile.  If that fails, consider
1091          * expanding the ofile array.
1092          *
1093          * NOTE! the 'allocated' field is a cumulative recursive allocation
1094          * count.  If we happen to see a value of 0 then we can shortcut
1095          * our search.  Otherwise we run through through the tree going
1096          * down branches we know have free descriptor(s) until we hit a
1097          * leaf node.  The leaf node will be free but will not necessarily
1098          * have an allocated field of 0.
1099          */
1100 retry:
1101         /* move up the tree looking for a subtree with a free node */
1102         for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1103              fd = right_ancestor(fd)) {
1104                 if (fdp->fd_files[fd].allocated == 0)
1105                         goto found;
1106
1107                 rsize = right_subtree_size(fd);
1108                 if (fdp->fd_files[fd].allocated == rsize)
1109                         continue;       /* right subtree full */
1110
1111                 /*
1112                  * Free fd is in the right subtree of the tree rooted at fd.
1113                  * Call that subtree R.  Look for the smallest (leftmost)
1114                  * subtree of R with an unallocated fd: continue moving
1115                  * down the left branch until encountering a full left
1116                  * subtree, then move to the right.
1117                  */
1118                 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1119                         node = fd + rsize;
1120                         rsum += fdp->fd_files[node].allocated;
1121                         if (fdp->fd_files[fd].allocated == rsum + rsize) {
1122                                 fd = node;      /* move to the right */
1123                                 if (fdp->fd_files[node].allocated == 0)
1124                                         goto found;
1125                                 rsum = 0;
1126                         }
1127                 }
1128                 goto found;
1129         }
1130
1131         /*
1132          * No space in current array.  Expand?
1133          */
1134         if (fdp->fd_nfiles >= lim) {
1135                 spin_unlock_wr(&fdp->fd_spin);
1136                 return (EMFILE);
1137         }
1138         fdgrow_locked(fdp, want);
1139         goto retry;
1140
1141 found:
1142         KKASSERT(fd < fdp->fd_nfiles);
1143         if (fd > fdp->fd_lastfile)
1144                 fdp->fd_lastfile = fd;
1145         if (want <= fdp->fd_freefile)
1146                 fdp->fd_freefile = fd;
1147         *result = fd;
1148         KKASSERT(fdp->fd_files[fd].fp == NULL);
1149         KKASSERT(fdp->fd_files[fd].reserved == 0);
1150         fdp->fd_files[fd].fileflags = 0;
1151         fdp->fd_files[fd].reserved = 1;
1152         fdreserve_locked(fdp, fd, 1);
1153         spin_unlock_wr(&fdp->fd_spin);
1154         return (0);
1155 }
1156
1157 /*
1158  * Check to see whether n user file descriptors
1159  * are available to the process p.
1160  *
1161  * MPSAFE
1162  */
1163 int
1164 fdavail(struct proc *p, int n)
1165 {
1166         struct filedesc *fdp = p->p_fd;
1167         struct fdnode *fdnode;
1168         int i, lim, last;
1169
1170         spin_lock_rd(&p->p_limit->p_spin);
1171         lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1172         spin_unlock_rd(&p->p_limit->p_spin);
1173
1174         spin_lock_rd(&fdp->fd_spin);
1175         if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1176                 spin_unlock_rd(&fdp->fd_spin);
1177                 return (1);
1178         }
1179         last = min(fdp->fd_nfiles, lim);
1180         fdnode = &fdp->fd_files[fdp->fd_freefile];
1181         for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1182                 if (fdnode->fp == NULL && --n <= 0) {
1183                         spin_unlock_rd(&fdp->fd_spin);
1184                         return (1);
1185                 }
1186         }
1187         spin_unlock_rd(&fdp->fd_spin);
1188         return (0);
1189 }
1190
1191 /*
1192  * Revoke open descriptors referencing (f_data, f_type)
1193  *
1194  * Any revoke executed within a prison is only able to
1195  * revoke descriptors for processes within that prison.
1196  *
1197  * Returns 0 on success or an error code.
1198  */
1199 struct fdrevoke_info {
1200         void *data;
1201         short type;
1202         short unused;
1203         int count;
1204         int intransit;
1205         struct ucred *cred;
1206         struct file *nfp;
1207 };
1208
1209 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1210 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1211
1212 int
1213 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1214 {
1215         struct fdrevoke_info info;
1216         int error;
1217
1218         bzero(&info, sizeof(info));
1219         info.data = f_data;
1220         info.type = f_type;
1221         info.cred = cred;
1222         error = falloc(NULL, &info.nfp, NULL);
1223         if (error)
1224                 return (error);
1225
1226         /*
1227          * Scan the file pointer table once.  dups do not dup file pointers,
1228          * only descriptors, so there is no leak.  Set FREVOKED on the fps
1229          * being revoked.
1230          */
1231         allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1232
1233         /*
1234          * If any fps were marked track down the related descriptors
1235          * and close them.  Any dup()s at this point will notice
1236          * the FREVOKED already set in the fp and do the right thing.
1237          *
1238          * Any fps with non-zero msgcounts (aka sent over a unix-domain
1239          * socket) bumped the intransit counter and will require a
1240          * scan.  Races against fps leaving the socket are closed by
1241          * the socket code checking for FREVOKED.
1242          */
1243         if (info.count)
1244                 allproc_scan(fdrevoke_proc_callback, &info);
1245         if (info.intransit)
1246                 unp_revoke_gc(info.nfp);
1247         fdrop(info.nfp);
1248         return(0);
1249 }
1250
1251 /*
1252  * Locate matching file pointers directly.
1253  */
1254 static int
1255 fdrevoke_check_callback(struct file *fp, void *vinfo)
1256 {
1257         struct fdrevoke_info *info = vinfo;
1258
1259         /*
1260          * File pointers already flagged for revokation are skipped.
1261          */
1262         if (fp->f_flag & FREVOKED)
1263                 return(0);
1264
1265         /*
1266          * If revoking from a prison file pointers created outside of
1267          * that prison, or file pointers without creds, cannot be revoked.
1268          */
1269         if (info->cred->cr_prison &&
1270             (fp->f_cred == NULL ||
1271              info->cred->cr_prison != fp->f_cred->cr_prison)) {
1272                 return(0);
1273         }
1274
1275         /*
1276          * If the file pointer matches then mark it for revocation.  The
1277          * flag is currently only used by unp_revoke_gc().
1278          *
1279          * info->count is a heuristic and can race in a SMP environment.
1280          */
1281         if (info->data == fp->f_data && info->type == fp->f_type) {
1282                 atomic_set_int(&fp->f_flag, FREVOKED);
1283                 info->count += fp->f_count;
1284                 if (fp->f_msgcount)
1285                         ++info->intransit;
1286         }
1287         return(0);
1288 }
1289
1290 /*
1291  * Locate matching file pointers via process descriptor tables.
1292  */
1293 static int
1294 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1295 {
1296         struct fdrevoke_info *info = vinfo;
1297         struct filedesc *fdp;
1298         struct file *fp;
1299         int n;
1300
1301         if (p->p_stat == SIDL || p->p_stat == SZOMB)
1302                 return(0);
1303         if (info->cred->cr_prison &&
1304             info->cred->cr_prison != p->p_ucred->cr_prison) {
1305                 return(0);
1306         }
1307
1308         /*
1309          * If the controlling terminal of the process matches the
1310          * vnode being revoked we clear the controlling terminal.
1311          *
1312          * The normal spec_close() may not catch this because it
1313          * uses curproc instead of p.
1314          */
1315         if (p->p_session && info->type == DTYPE_VNODE &&
1316             info->data == p->p_session->s_ttyvp) {
1317                 p->p_session->s_ttyvp = NULL;
1318                 vrele(info->data);
1319         }
1320
1321         /*
1322          * Locate and close any matching file descriptors.
1323          */
1324         if ((fdp = p->p_fd) == NULL)
1325                 return(0);
1326         spin_lock_wr(&fdp->fd_spin);
1327         for (n = 0; n < fdp->fd_nfiles; ++n) {
1328                 if ((fp = fdp->fd_files[n].fp) == NULL)
1329                         continue;
1330                 if (fp->f_flag & FREVOKED) {
1331                         fhold(info->nfp);
1332                         fdp->fd_files[n].fp = info->nfp;
1333                         spin_unlock_wr(&fdp->fd_spin);
1334                         closef(fp, p);
1335                         spin_lock_wr(&fdp->fd_spin);
1336                         --info->count;
1337                 }
1338         }
1339         spin_unlock_wr(&fdp->fd_spin);
1340         return(0);
1341 }
1342
1343 /*
1344  * falloc:
1345  *      Create a new open file structure and reserve a file decriptor
1346  *      for the process that refers to it.
1347  *
1348  *      Root creds are checked using p, or assumed if p is NULL.  If
1349  *      resultfd is non-NULL then p must also be non-NULL.  No file
1350  *      descriptor is reserved if resultfd is NULL.
1351  *
1352  *      A file pointer with a refcount of 1 is returned.  Note that the
1353  *      file pointer is NOT associated with the descriptor.  If falloc
1354  *      returns success, fsetfd() MUST be called to either associate the
1355  *      file pointer or clear the reservation.
1356  *
1357  * MPSAFE
1358  */
1359 int
1360 falloc(struct proc *p, struct file **resultfp, int *resultfd)
1361 {
1362         static struct timeval lastfail;
1363         static int curfail;
1364         struct file *fp;
1365         int error;
1366
1367         fp = NULL;
1368
1369         /*
1370          * Handle filetable full issues and root overfill.
1371          */
1372         if (nfiles >= maxfiles - maxfilesrootres &&
1373             ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) {
1374                 if (ppsratecheck(&lastfail, &curfail, 1)) {
1375                         kprintf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n",
1376                                 (p ? p->p_ucred->cr_ruid : -1));
1377                 }
1378                 error = ENFILE;
1379                 goto done;
1380         }
1381
1382         /*
1383          * Allocate a new file descriptor.
1384          */
1385         fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1386         spin_init(&fp->f_spin);
1387         fp->f_count = 1;
1388         fp->f_ops = &badfileops;
1389         fp->f_seqcount = 1;
1390         if (p)
1391                 fp->f_cred = crhold(p->p_ucred);
1392         else
1393                 fp->f_cred = crhold(proc0.p_ucred);
1394         spin_lock_wr(&filehead_spin);
1395         nfiles++;
1396         LIST_INSERT_HEAD(&filehead, fp, f_list);
1397         spin_unlock_wr(&filehead_spin);
1398         if (resultfd) {
1399                 if ((error = fdalloc(p, 0, resultfd)) != 0) {
1400                         fdrop(fp);
1401                         fp = NULL;
1402                 }
1403         } else {
1404                 error = 0;
1405         }
1406 done:
1407         *resultfp = fp;
1408         return (error);
1409 }
1410
1411 /*
1412  * MPSAFE
1413  */
1414 static
1415 int
1416 checkfpclosed(struct filedesc *fdp, int fd, struct file *fp)
1417 {
1418         int error;
1419
1420         spin_lock_rd(&fdp->fd_spin);
1421         if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1422                 error = EBADF;
1423         else
1424                 error = 0;
1425         spin_unlock_rd(&fdp->fd_spin);
1426         return (error);
1427 }
1428
1429 /*
1430  * Associate a file pointer with a previously reserved file descriptor.
1431  * This function always succeeds.
1432  *
1433  * If fp is NULL, the file descriptor is returned to the pool.
1434  */
1435
1436 /*
1437  * MPSAFE (exclusive spinlock must be held on call)
1438  */
1439 static void
1440 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1441 {
1442         KKASSERT((unsigned)fd < fdp->fd_nfiles);
1443         KKASSERT(fdp->fd_files[fd].reserved != 0);
1444         if (fp) {
1445                 fhold(fp);
1446                 fdp->fd_files[fd].fp = fp;
1447                 fdp->fd_files[fd].reserved = 0;
1448                 if (fp->f_type == DTYPE_KQUEUE) {
1449                         if (fdp->fd_knlistsize < 0)
1450                                 fdp->fd_knlistsize = 0;
1451                 }
1452         } else {
1453                 fdp->fd_files[fd].reserved = 0;
1454                 fdreserve_locked(fdp, fd, -1);
1455                 fdfixup_locked(fdp, fd);
1456         }
1457 }
1458
1459 /*
1460  * MPSAFE
1461  */
1462 void
1463 fsetfd(struct proc *p, struct file *fp, int fd)
1464 {
1465         struct filedesc *fdp = p->p_fd;
1466
1467         spin_lock_wr(&fdp->fd_spin);
1468         fsetfd_locked(fdp, fp, fd);
1469         spin_unlock_wr(&fdp->fd_spin);
1470 }
1471
1472 /*
1473  * MPSAFE (exclusive spinlock must be held on call)
1474  */
1475 static
1476 struct file *
1477 funsetfd_locked(struct filedesc *fdp, int fd)
1478 {
1479         struct file *fp;
1480
1481         if ((unsigned)fd >= fdp->fd_nfiles)
1482                 return (NULL);
1483         if ((fp = fdp->fd_files[fd].fp) == NULL)
1484                 return (NULL);
1485         fdp->fd_files[fd].fp = NULL;
1486         fdp->fd_files[fd].fileflags = 0;
1487
1488         fdreserve_locked(fdp, fd, -1);
1489         fdfixup_locked(fdp, fd);
1490         return(fp);
1491 }
1492
1493 /*
1494  * MPSAFE
1495  */
1496 int
1497 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1498 {
1499         int error;
1500
1501         spin_lock_rd(&fdp->fd_spin);
1502         if (((u_int)fd) >= fdp->fd_nfiles) {
1503                 error = EBADF;
1504         } else if (fdp->fd_files[fd].fp == NULL) {
1505                 error = EBADF;
1506         } else {
1507                 *flagsp = fdp->fd_files[fd].fileflags;
1508                 error = 0;
1509         }
1510         spin_unlock_rd(&fdp->fd_spin);
1511         return (error);
1512 }
1513
1514 /*
1515  * MPSAFE
1516  */
1517 int
1518 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1519 {
1520         int error;
1521
1522         spin_lock_wr(&fdp->fd_spin);
1523         if (((u_int)fd) >= fdp->fd_nfiles) {
1524                 error = EBADF;
1525         } else if (fdp->fd_files[fd].fp == NULL) {
1526                 error = EBADF;
1527         } else {
1528                 fdp->fd_files[fd].fileflags |= add_flags;
1529                 error = 0;
1530         }
1531         spin_unlock_wr(&fdp->fd_spin);
1532         return (error);
1533 }
1534
1535 /*
1536  * MPSAFE
1537  */
1538 int
1539 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1540 {
1541         int error;
1542
1543         spin_lock_wr(&fdp->fd_spin);
1544         if (((u_int)fd) >= fdp->fd_nfiles) {
1545                 error = EBADF;
1546         } else if (fdp->fd_files[fd].fp == NULL) {
1547                 error = EBADF;
1548         } else {
1549                 fdp->fd_files[fd].fileflags &= ~rem_flags;
1550                 error = 0;
1551         }
1552         spin_unlock_wr(&fdp->fd_spin);
1553         return (error);
1554 }
1555
1556 void
1557 fsetcred(struct file *fp, struct ucred *cr)
1558 {
1559         crhold(cr);
1560         crfree(fp->f_cred);
1561         fp->f_cred = cr;
1562 }
1563
1564 /*
1565  * Free a file descriptor.
1566  */
1567 static
1568 void
1569 ffree(struct file *fp)
1570 {
1571         KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1572         spin_lock_wr(&filehead_spin);
1573         LIST_REMOVE(fp, f_list);
1574         nfiles--;
1575         spin_unlock_wr(&filehead_spin);
1576         crfree(fp->f_cred);
1577         if (fp->f_nchandle.ncp)
1578             cache_drop(&fp->f_nchandle);
1579         kfree(fp, M_FILE);
1580 }
1581
1582 /*
1583  * called from init_main, initialize filedesc0 for proc0.
1584  */
1585 void
1586 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1587 {
1588         p0->p_fd = fdp0;
1589         p0->p_fdtol = NULL;
1590         fdp0->fd_refcnt = 1;
1591         fdp0->fd_cmask = cmask;
1592         fdp0->fd_files = fdp0->fd_builtin_files;
1593         fdp0->fd_nfiles = NDFILE;
1594         fdp0->fd_lastfile = -1;
1595         spin_init(&fdp0->fd_spin);
1596 }
1597
1598 /*
1599  * Build a new filedesc structure.
1600  *
1601  * NOT MPSAFE (vref)
1602  */
1603 struct filedesc *
1604 fdinit(struct proc *p)
1605 {
1606         struct filedesc *newfdp;
1607         struct filedesc *fdp = p->p_fd;
1608
1609         newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1610         spin_lock_rd(&fdp->fd_spin);
1611         if (fdp->fd_cdir) {
1612                 newfdp->fd_cdir = fdp->fd_cdir;
1613                 vref(newfdp->fd_cdir);
1614                 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1615         }
1616
1617         /*
1618          * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1619          * proc0, but should unconditionally exist in other processes.
1620          */
1621         if (fdp->fd_rdir) {
1622                 newfdp->fd_rdir = fdp->fd_rdir;
1623                 vref(newfdp->fd_rdir);
1624                 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1625         }
1626         if (fdp->fd_jdir) {
1627                 newfdp->fd_jdir = fdp->fd_jdir;
1628                 vref(newfdp->fd_jdir);
1629                 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1630         }
1631         spin_unlock_rd(&fdp->fd_spin);
1632
1633         /* Create the file descriptor table. */
1634         newfdp->fd_refcnt = 1;
1635         newfdp->fd_cmask = cmask;
1636         newfdp->fd_files = newfdp->fd_builtin_files;
1637         newfdp->fd_nfiles = NDFILE;
1638         newfdp->fd_knlistsize = -1;
1639         newfdp->fd_lastfile = -1;
1640         spin_init(&newfdp->fd_spin);
1641
1642         return (newfdp);
1643 }
1644
1645 /*
1646  * Share a filedesc structure.
1647  *
1648  * MPSAFE
1649  */
1650 struct filedesc *
1651 fdshare(struct proc *p)
1652 {
1653         struct filedesc *fdp;
1654
1655         fdp = p->p_fd;
1656         spin_lock_wr(&fdp->fd_spin);
1657         fdp->fd_refcnt++;
1658         spin_unlock_wr(&fdp->fd_spin);
1659         return (fdp);
1660 }
1661
1662 /*
1663  * Copy a filedesc structure.
1664  *
1665  * MPSAFE
1666  */
1667 struct filedesc *
1668 fdcopy(struct proc *p)
1669 {
1670         struct filedesc *fdp = p->p_fd;
1671         struct filedesc *newfdp;
1672         struct fdnode *fdnode;
1673         int i;
1674         int ni;
1675
1676         /*
1677          * Certain daemons might not have file descriptors.
1678          */
1679         if (fdp == NULL)
1680                 return (NULL);
1681
1682         /*
1683          * Allocate the new filedesc and fd_files[] array.  This can race
1684          * with operations by other threads on the fdp so we have to be
1685          * careful.
1686          */
1687         newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1688 again:
1689         spin_lock_rd(&fdp->fd_spin);
1690         if (fdp->fd_lastfile < NDFILE) {
1691                 newfdp->fd_files = newfdp->fd_builtin_files;
1692                 i = NDFILE;
1693         } else {
1694                 /*
1695                  * We have to allocate (N^2-1) entries for our in-place
1696                  * binary tree.  Allow the table to shrink.
1697                  */
1698                 i = fdp->fd_nfiles;
1699                 ni = (i - 1) / 2;
1700                 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1701                         i = ni;
1702                         ni = (i - 1) / 2;
1703                 }
1704                 spin_unlock_rd(&fdp->fd_spin);
1705                 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1706                                           M_FILEDESC, M_WAITOK | M_ZERO);
1707
1708                 /*
1709                  * Check for race, retry
1710                  */
1711                 spin_lock_rd(&fdp->fd_spin);
1712                 if (i <= fdp->fd_lastfile) {
1713                         spin_unlock_rd(&fdp->fd_spin);
1714                         kfree(newfdp->fd_files, M_FILEDESC);
1715                         goto again;
1716                 }
1717         }
1718
1719         /*
1720          * Dup the remaining fields. vref() and cache_hold() can be
1721          * safely called while holding the read spinlock on fdp.
1722          *
1723          * The read spinlock on fdp is still being held.
1724          *
1725          * NOTE: vref and cache_hold calls for the case where the vnode
1726          * or cache entry already has at least one ref may be called
1727          * while holding spin locks.
1728          */
1729         if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1730                 vref(newfdp->fd_cdir);
1731                 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1732         }
1733         /*
1734          * We must check for fd_rdir here, at least for now because
1735          * the init process is created before we have access to the
1736          * rootvode to take a reference to it.
1737          */
1738         if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1739                 vref(newfdp->fd_rdir);
1740                 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1741         }
1742         if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1743                 vref(newfdp->fd_jdir);
1744                 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1745         }
1746         newfdp->fd_refcnt = 1;
1747         newfdp->fd_nfiles = i;
1748         newfdp->fd_lastfile = fdp->fd_lastfile;
1749         newfdp->fd_freefile = fdp->fd_freefile;
1750         newfdp->fd_cmask = fdp->fd_cmask;
1751         newfdp->fd_knlist = NULL;
1752         newfdp->fd_knlistsize = -1;
1753         newfdp->fd_knhash = NULL;
1754         newfdp->fd_knhashmask = 0;
1755         spin_init(&newfdp->fd_spin);
1756
1757         /*
1758          * Copy the descriptor table through (i).  This also copies the
1759          * allocation state.   Then go through and ref the file pointers
1760          * and clean up any KQ descriptors.
1761          *
1762          * kq descriptors cannot be copied.  Since we haven't ref'd the
1763          * copied files yet we can ignore the return value from funsetfd().
1764          *
1765          * The read spinlock on fdp is still being held.
1766          */
1767         bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1768         for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1769                 fdnode = &newfdp->fd_files[i];
1770                 if (fdnode->reserved) {
1771                         fdreserve_locked(newfdp, i, -1);
1772                         fdnode->reserved = 0;
1773                         fdfixup_locked(newfdp, i);
1774                 } else if (fdnode->fp) {
1775                         if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1776                                 (void)funsetfd_locked(newfdp, i);
1777                         } else {
1778                                 fhold(fdnode->fp);
1779                         }
1780                 }
1781         }
1782         spin_unlock_rd(&fdp->fd_spin);
1783         return (newfdp);
1784 }
1785
1786 /*
1787  * Release a filedesc structure.
1788  *
1789  * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1790  */
1791 void
1792 fdfree(struct proc *p)
1793 {
1794         struct filedesc *fdp = p->p_fd;
1795         struct fdnode *fdnode;
1796         int i;
1797         struct filedesc_to_leader *fdtol;
1798         struct file *fp;
1799         struct vnode *vp;
1800         struct flock lf;
1801
1802         /* Certain daemons might not have file descriptors. */
1803         if (fdp == NULL)
1804                 return;
1805
1806         /*
1807          * Severe messing around to follow
1808          */
1809         spin_lock_wr(&fdp->fd_spin);
1810
1811         /* Check for special need to clear POSIX style locks */
1812         fdtol = p->p_fdtol;
1813         if (fdtol != NULL) {
1814                 KASSERT(fdtol->fdl_refcount > 0,
1815                         ("filedesc_to_refcount botch: fdl_refcount=%d",
1816                          fdtol->fdl_refcount));
1817                 if (fdtol->fdl_refcount == 1 &&
1818                     (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1819                         for (i = 0; i <= fdp->fd_lastfile; ++i) {
1820                                 fdnode = &fdp->fd_files[i];
1821                                 if (fdnode->fp == NULL ||
1822                                     fdnode->fp->f_type != DTYPE_VNODE) {
1823                                         continue;
1824                                 }
1825                                 fp = fdnode->fp;
1826                                 fhold(fp);
1827                                 spin_unlock_wr(&fdp->fd_spin);
1828
1829                                 lf.l_whence = SEEK_SET;
1830                                 lf.l_start = 0;
1831                                 lf.l_len = 0;
1832                                 lf.l_type = F_UNLCK;
1833                                 vp = (struct vnode *)fp->f_data;
1834                                 (void) VOP_ADVLOCK(vp,
1835                                                    (caddr_t)p->p_leader,
1836                                                    F_UNLCK,
1837                                                    &lf,
1838                                                    F_POSIX);
1839                                 fdrop(fp);
1840                                 spin_lock_wr(&fdp->fd_spin);
1841                         }
1842                 }
1843         retry:
1844                 if (fdtol->fdl_refcount == 1) {
1845                         if (fdp->fd_holdleaderscount > 0 &&
1846                             (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1847                                 /*
1848                                  * close() or do_dup() has cleared a reference
1849                                  * in a shared file descriptor table.
1850                                  */
1851                                 fdp->fd_holdleaderswakeup = 1;
1852                                 msleep(&fdp->fd_holdleaderscount,
1853                                        &fdp->fd_spin, 0, "fdlhold", 0);
1854                                 goto retry;
1855                         }
1856                         if (fdtol->fdl_holdcount > 0) {
1857                                 /*
1858                                  * Ensure that fdtol->fdl_leader
1859                                  * remains valid in closef().
1860                                  */
1861                                 fdtol->fdl_wakeup = 1;
1862                                 msleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1863                                 goto retry;
1864                         }
1865                 }
1866                 fdtol->fdl_refcount--;
1867                 if (fdtol->fdl_refcount == 0 &&
1868                     fdtol->fdl_holdcount == 0) {
1869                         fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1870                         fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1871                 } else {
1872                         fdtol = NULL;
1873                 }
1874                 p->p_fdtol = NULL;
1875                 if (fdtol != NULL) {
1876                         spin_unlock_wr(&fdp->fd_spin);
1877                         kfree(fdtol, M_FILEDESC_TO_LEADER);
1878                         spin_lock_wr(&fdp->fd_spin);
1879                 }
1880         }
1881         if (--fdp->fd_refcnt > 0) {
1882                 spin_unlock_wr(&fdp->fd_spin);
1883                 return;
1884         }
1885         spin_unlock_wr(&fdp->fd_spin);
1886
1887         /*
1888          * we are the last reference to the structure, we can
1889          * safely assume it will not change out from under us.
1890          */
1891         for (i = 0; i <= fdp->fd_lastfile; ++i) {
1892                 if (fdp->fd_files[i].fp)
1893                         closef(fdp->fd_files[i].fp, p);
1894         }
1895         if (fdp->fd_files != fdp->fd_builtin_files)
1896                 kfree(fdp->fd_files, M_FILEDESC);
1897         if (fdp->fd_cdir) {
1898                 cache_drop(&fdp->fd_ncdir);
1899                 vrele(fdp->fd_cdir);
1900         }
1901         if (fdp->fd_rdir) {
1902                 cache_drop(&fdp->fd_nrdir);
1903                 vrele(fdp->fd_rdir);
1904         }
1905         if (fdp->fd_jdir) {
1906                 cache_drop(&fdp->fd_njdir);
1907                 vrele(fdp->fd_jdir);
1908         }
1909         if (fdp->fd_knlist)
1910                 kfree(fdp->fd_knlist, M_KQUEUE);
1911         if (fdp->fd_knhash)
1912                 kfree(fdp->fd_knhash, M_KQUEUE);
1913         kfree(fdp, M_FILEDESC);
1914 }
1915
1916 /*
1917  * Retrieve and reference the file pointer associated with a descriptor.
1918  *
1919  * MPSAFE
1920  */
1921 struct file *
1922 holdfp(struct filedesc *fdp, int fd, int flag)
1923 {
1924         struct file* fp;
1925
1926         spin_lock_rd(&fdp->fd_spin);
1927         if (((u_int)fd) >= fdp->fd_nfiles) {
1928                 fp = NULL;
1929                 goto done;
1930         }
1931         if ((fp = fdp->fd_files[fd].fp) == NULL)
1932                 goto done;
1933         if ((fp->f_flag & flag) == 0 && flag != -1) {
1934                 fp = NULL;
1935                 goto done;
1936         }
1937         fhold(fp);
1938 done:
1939         spin_unlock_rd(&fdp->fd_spin);
1940         return (fp);
1941 }
1942
1943 /*
1944  * holdsock() - load the struct file pointer associated
1945  * with a socket into *fpp.  If an error occurs, non-zero
1946  * will be returned and *fpp will be set to NULL.
1947  *
1948  * MPSAFE
1949  */
1950 int
1951 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
1952 {
1953         struct file *fp;
1954         int error;
1955
1956         spin_lock_rd(&fdp->fd_spin);
1957         if ((unsigned)fd >= fdp->fd_nfiles) {
1958                 error = EBADF;
1959                 fp = NULL;
1960                 goto done;
1961         }
1962         if ((fp = fdp->fd_files[fd].fp) == NULL) {
1963                 error = EBADF;
1964                 goto done;
1965         }
1966         if (fp->f_type != DTYPE_SOCKET) {
1967                 error = ENOTSOCK;
1968                 goto done;
1969         }
1970         fhold(fp);
1971         error = 0;
1972 done:
1973         spin_unlock_rd(&fdp->fd_spin);
1974         *fpp = fp;
1975         return (error);
1976 }
1977
1978 /*
1979  * Convert a user file descriptor to a held file pointer.
1980  *
1981  * MPSAFE
1982  */
1983 int
1984 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
1985 {
1986         struct file *fp;
1987         int error;
1988
1989         spin_lock_rd(&fdp->fd_spin);
1990         if ((unsigned)fd >= fdp->fd_nfiles) {
1991                 error = EBADF;
1992                 fp = NULL;
1993                 goto done;
1994         }
1995         if ((fp = fdp->fd_files[fd].fp) == NULL) {
1996                 error = EBADF;
1997                 goto done;
1998         }
1999         if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2000                 fp = NULL;
2001                 error = EINVAL;
2002                 goto done;
2003         }
2004         fhold(fp);
2005         error = 0;
2006 done:
2007         spin_unlock_rd(&fdp->fd_spin);
2008         *fpp = fp;
2009         return (error);
2010 }
2011
2012 /*
2013  * For setugid programs, we don't want to people to use that setugidness
2014  * to generate error messages which write to a file which otherwise would
2015  * otherwise be off-limits to the process.
2016  *
2017  * This is a gross hack to plug the hole.  A better solution would involve
2018  * a special vop or other form of generalized access control mechanism.  We
2019  * go ahead and just reject all procfs file systems accesses as dangerous.
2020  *
2021  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2022  * sufficient.  We also don't for check setugidness since we know we are.
2023  */
2024 static int
2025 is_unsafe(struct file *fp)
2026 {
2027         if (fp->f_type == DTYPE_VNODE &&
2028             ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2029                 return (1);
2030         return (0);
2031 }
2032
2033 /*
2034  * Make this setguid thing safe, if at all possible.
2035  *
2036  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2037  */
2038 void
2039 setugidsafety(struct proc *p)
2040 {
2041         struct filedesc *fdp = p->p_fd;
2042         int i;
2043
2044         /* Certain daemons might not have file descriptors. */
2045         if (fdp == NULL)
2046                 return;
2047
2048         /*
2049          * note: fdp->fd_files may be reallocated out from under us while
2050          * we are blocked in a close.  Be careful!
2051          */
2052         for (i = 0; i <= fdp->fd_lastfile; i++) {
2053                 if (i > 2)
2054                         break;
2055                 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2056                         struct file *fp;
2057
2058                         if (i < fdp->fd_knlistsize)
2059                                 knote_fdclose(p, i);
2060                         /*
2061                          * NULL-out descriptor prior to close to avoid
2062                          * a race while close blocks.
2063                          */
2064                         if ((fp = funsetfd_locked(fdp, i)) != NULL)
2065                                 closef(fp, p);
2066                 }
2067         }
2068 }
2069
2070 /*
2071  * Close any files on exec?
2072  *
2073  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2074  */
2075 void
2076 fdcloseexec(struct proc *p)
2077 {
2078         struct filedesc *fdp = p->p_fd;
2079         int i;
2080
2081         /* Certain daemons might not have file descriptors. */
2082         if (fdp == NULL)
2083                 return;
2084
2085         /*
2086          * We cannot cache fd_files since operations may block and rip
2087          * them out from under us.
2088          */
2089         for (i = 0; i <= fdp->fd_lastfile; i++) {
2090                 if (fdp->fd_files[i].fp != NULL &&
2091                     (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2092                         struct file *fp;
2093
2094                         if (i < fdp->fd_knlistsize)
2095                                 knote_fdclose(p, i);
2096                         /*
2097                          * NULL-out descriptor prior to close to avoid
2098                          * a race while close blocks.
2099                          */
2100                         if ((fp = funsetfd_locked(fdp, i)) != NULL)
2101                                 closef(fp, p);
2102                 }
2103         }
2104 }
2105
2106 /*
2107  * It is unsafe for set[ug]id processes to be started with file
2108  * descriptors 0..2 closed, as these descriptors are given implicit
2109  * significance in the Standard C library.  fdcheckstd() will create a
2110  * descriptor referencing /dev/null for each of stdin, stdout, and
2111  * stderr that is not already open.
2112  *
2113  * NOT MPSAFE - calls falloc, vn_open, etc
2114  */
2115 int
2116 fdcheckstd(struct proc *p)
2117 {
2118         struct nlookupdata nd;
2119         struct filedesc *fdp;
2120         struct file *fp;
2121         int retval;
2122         int i, error, flags, devnull;
2123
2124         fdp = p->p_fd;
2125         if (fdp == NULL)
2126                 return (0);
2127         devnull = -1;
2128         error = 0;
2129         for (i = 0; i < 3; i++) {
2130                 if (fdp->fd_files[i].fp != NULL)
2131                         continue;
2132                 if (devnull < 0) {
2133                         if ((error = falloc(p, &fp, &devnull)) != 0)
2134                                 break;
2135
2136                         error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2137                                                 NLC_FOLLOW|NLC_LOCKVP);
2138                         flags = FREAD | FWRITE;
2139                         if (error == 0)
2140                                 error = vn_open(&nd, fp, flags, 0);
2141                         if (error == 0)
2142                                 fsetfd(p, fp, devnull);
2143                         else
2144                                 fsetfd(p, NULL, devnull);
2145                         fdrop(fp);
2146                         nlookup_done(&nd);
2147                         if (error)
2148                                 break;
2149                         KKASSERT(i == devnull);
2150                 } else {
2151                         error = kern_dup(DUP_FIXED, devnull, i, &retval);
2152                         if (error != 0)
2153                                 break;
2154                 }
2155         }
2156         return (error);
2157 }
2158
2159 /*
2160  * Internal form of close.
2161  * Decrement reference count on file structure.
2162  * Note: td and/or p may be NULL when closing a file
2163  * that was being passed in a message.
2164  *
2165  * MPALMOSTSAFE - acquires mplock for VOP operations
2166  */
2167 int
2168 closef(struct file *fp, struct proc *p)
2169 {
2170         struct vnode *vp;
2171         struct flock lf;
2172         struct filedesc_to_leader *fdtol;
2173
2174         if (fp == NULL)
2175                 return (0);
2176
2177         /*
2178          * POSIX record locking dictates that any close releases ALL
2179          * locks owned by this process.  This is handled by setting
2180          * a flag in the unlock to free ONLY locks obeying POSIX
2181          * semantics, and not to free BSD-style file locks.
2182          * If the descriptor was in a message, POSIX-style locks
2183          * aren't passed with the descriptor.
2184          */
2185         if (p != NULL && fp->f_type == DTYPE_VNODE &&
2186             (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2187         ) {
2188                 get_mplock();
2189                 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2190                         lf.l_whence = SEEK_SET;
2191                         lf.l_start = 0;
2192                         lf.l_len = 0;
2193                         lf.l_type = F_UNLCK;
2194                         vp = (struct vnode *)fp->f_data;
2195                         (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2196                                            &lf, F_POSIX);
2197                 }
2198                 fdtol = p->p_fdtol;
2199                 if (fdtol != NULL) {
2200                         /*
2201                          * Handle special case where file descriptor table
2202                          * is shared between multiple process leaders.
2203                          */
2204                         for (fdtol = fdtol->fdl_next;
2205                              fdtol != p->p_fdtol;
2206                              fdtol = fdtol->fdl_next) {
2207                                 if ((fdtol->fdl_leader->p_flag &
2208                                      P_ADVLOCK) == 0)
2209                                         continue;
2210                                 fdtol->fdl_holdcount++;
2211                                 lf.l_whence = SEEK_SET;
2212                                 lf.l_start = 0;
2213                                 lf.l_len = 0;
2214                                 lf.l_type = F_UNLCK;
2215                                 vp = (struct vnode *)fp->f_data;
2216                                 (void) VOP_ADVLOCK(vp,
2217                                                    (caddr_t)fdtol->fdl_leader,
2218                                                    F_UNLCK, &lf, F_POSIX);
2219                                 fdtol->fdl_holdcount--;
2220                                 if (fdtol->fdl_holdcount == 0 &&
2221                                     fdtol->fdl_wakeup != 0) {
2222                                         fdtol->fdl_wakeup = 0;
2223                                         wakeup(fdtol);
2224                                 }
2225                         }
2226                 }
2227                 rel_mplock();
2228         }
2229         return (fdrop(fp));
2230 }
2231
2232 /*
2233  * MPSAFE
2234  *
2235  * fhold() can only be called if f_count is already at least 1 (i.e. the
2236  * caller of fhold() already has a reference to the file pointer in some
2237  * manner or other).
2238  *
2239  * f_count is not spin-locked.  Instead, atomic ops are used for
2240  * incrementing, decrementing, and handling the 1->0 transition.
2241  */
2242 void
2243 fhold(struct file *fp)
2244 {
2245         atomic_add_int(&fp->f_count, 1);
2246 }
2247
2248 /*
2249  * fdrop() - drop a reference to a descriptor
2250  *
2251  * MPALMOSTSAFE - acquires mplock for final close sequence
2252  */
2253 int
2254 fdrop(struct file *fp)
2255 {
2256         struct flock lf;
2257         struct vnode *vp;
2258         int error;
2259
2260         /*
2261          * A combined fetch and subtract is needed to properly detect
2262          * 1->0 transitions, otherwise two cpus dropping from a ref
2263          * count of 2 might both try to run the 1->0 code.
2264          */
2265         if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2266                 return (0);
2267
2268         get_mplock();
2269
2270         /*
2271          * The last reference has gone away, we own the fp structure free
2272          * and clear.
2273          */
2274         if (fp->f_count < 0)
2275                 panic("fdrop: count < 0");
2276         if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2277             (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2278         ) {
2279                 lf.l_whence = SEEK_SET;
2280                 lf.l_start = 0;
2281                 lf.l_len = 0;
2282                 lf.l_type = F_UNLCK;
2283                 vp = (struct vnode *)fp->f_data;
2284                 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2285         }
2286         if (fp->f_ops != &badfileops)
2287                 error = fo_close(fp);
2288         else
2289                 error = 0;
2290         ffree(fp);
2291         rel_mplock();
2292         return (error);
2293 }
2294
2295 /*
2296  * Apply an advisory lock on a file descriptor.
2297  *
2298  * Just attempt to get a record lock of the requested type on
2299  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2300  */
2301 int
2302 sys_flock(struct flock_args *uap)
2303 {
2304         struct proc *p = curproc;
2305         struct file *fp;
2306         struct vnode *vp;
2307         struct flock lf;
2308         int error;
2309
2310         if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2311                 return (EBADF);
2312         if (fp->f_type != DTYPE_VNODE) {
2313                 error = EOPNOTSUPP;
2314                 goto done;
2315         }
2316         vp = (struct vnode *)fp->f_data;
2317         lf.l_whence = SEEK_SET;
2318         lf.l_start = 0;
2319         lf.l_len = 0;
2320         if (uap->how & LOCK_UN) {
2321                 lf.l_type = F_UNLCK;
2322                 fp->f_flag &= ~FHASLOCK;
2323                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2324                 goto done;
2325         }
2326         if (uap->how & LOCK_EX)
2327                 lf.l_type = F_WRLCK;
2328         else if (uap->how & LOCK_SH)
2329                 lf.l_type = F_RDLCK;
2330         else {
2331                 error = EBADF;
2332                 goto done;
2333         }
2334         fp->f_flag |= FHASLOCK;
2335         if (uap->how & LOCK_NB)
2336                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2337         else
2338                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2339 done:
2340         fdrop(fp);
2341         return (error);
2342 }
2343
2344 /*
2345  * File Descriptor pseudo-device driver (/dev/fd/).
2346  *
2347  * Opening minor device N dup()s the file (if any) connected to file
2348  * descriptor N belonging to the calling process.  Note that this driver
2349  * consists of only the ``open()'' routine, because all subsequent
2350  * references to this file will be direct to the other driver.
2351  */
2352 /* ARGSUSED */
2353 static int
2354 fdopen(struct dev_open_args *ap)
2355 {
2356         thread_t td = curthread;
2357
2358         KKASSERT(td->td_lwp != NULL);
2359
2360         /*
2361          * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2362          * the file descriptor being sought for duplication. The error
2363          * return ensures that the vnode for this device will be released
2364          * by vn_open. Open will detect this special error and take the
2365          * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2366          * will simply report the error.
2367          */
2368         td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2369         return (ENODEV);
2370 }
2371
2372 /*
2373  * The caller has reserved the file descriptor dfd for us.  On success we
2374  * must fsetfd() it.  On failure the caller will clean it up.
2375  *
2376  * NOT MPSAFE - isn't getting spinlocks, possibly other things
2377  */
2378 int
2379 dupfdopen(struct proc *p, int dfd, int sfd, int mode, int error)
2380 {
2381         struct filedesc *fdp = p->p_fd;
2382         struct file *wfp;
2383         struct file *xfp;
2384         int werror;
2385
2386         if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2387                 return (EBADF);
2388
2389         /*
2390          * Close a revoke/dup race.  Duping a descriptor marked as revoked
2391          * will dup a dummy descriptor instead of the real one.
2392          */
2393         if (wfp->f_flag & FREVOKED) {
2394                 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2395                 fdrop(wfp);
2396                 wfp = NULL;
2397                 werror = falloc(NULL, &wfp, NULL);
2398                 if (werror)
2399                         return (werror);
2400         }
2401
2402         /*
2403          * There are two cases of interest here.
2404          *
2405          * For ENODEV simply dup sfd to file descriptor dfd and return.
2406          *
2407          * For ENXIO steal away the file structure from sfd and store it
2408          * dfd.  sfd is effectively closed by this operation.
2409          *
2410          * Any other error code is just returned.
2411          */
2412         switch (error) {
2413         case ENODEV:
2414                 /*
2415                  * Check that the mode the file is being opened for is a
2416                  * subset of the mode of the existing descriptor.
2417                  */
2418                 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2419                         error = EACCES;
2420                         break;
2421                 }
2422                 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2423                 fsetfd(p, wfp, dfd);
2424                 error = 0;
2425                 break;
2426         case ENXIO:
2427                 /*
2428                  * Steal away the file pointer from dfd, and stuff it into indx.
2429                  */
2430                 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2431                 fsetfd(p, wfp, dfd);
2432                 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL)
2433                         fdrop(xfp);
2434                 error = 0;
2435                 break;
2436         default:
2437                 break;
2438         }
2439         fdrop(wfp);
2440         return (error);
2441 }
2442
2443 /*
2444  * NOT MPSAFE - I think these refer to a common file descriptor table
2445  * and we need to spinlock that to link fdtol in.
2446  */
2447 struct filedesc_to_leader *
2448 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2449                          struct proc *leader)
2450 {
2451         struct filedesc_to_leader *fdtol;
2452
2453         fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2454                         M_FILEDESC_TO_LEADER, M_WAITOK);
2455         fdtol->fdl_refcount = 1;
2456         fdtol->fdl_holdcount = 0;
2457         fdtol->fdl_wakeup = 0;
2458         fdtol->fdl_leader = leader;
2459         if (old != NULL) {
2460                 fdtol->fdl_next = old->fdl_next;
2461                 fdtol->fdl_prev = old;
2462                 old->fdl_next = fdtol;
2463                 fdtol->fdl_next->fdl_prev = fdtol;
2464         } else {
2465                 fdtol->fdl_next = fdtol;
2466                 fdtol->fdl_prev = fdtol;
2467         }
2468         return fdtol;
2469 }
2470
2471 /*
2472  * Scan all file pointers in the system.  The callback is made with
2473  * the master list spinlock held exclusively.
2474  *
2475  * MPSAFE
2476  */
2477 void
2478 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2479 {
2480         struct file *fp;
2481         int res;
2482
2483         spin_lock_wr(&filehead_spin);
2484         LIST_FOREACH(fp, &filehead, f_list) {
2485                 res = callback(fp, data);
2486                 if (res < 0)
2487                         break;
2488         }
2489         spin_unlock_wr(&filehead_spin);
2490 }
2491
2492 /*
2493  * Get file structures.
2494  *
2495  * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2496  */
2497
2498 struct sysctl_kern_file_info {
2499         int count;
2500         int error;
2501         struct sysctl_req *req;
2502 };
2503
2504 static int sysctl_kern_file_callback(struct proc *p, void *data);
2505
2506 static int
2507 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2508 {
2509         struct sysctl_kern_file_info info;
2510
2511         /*
2512          * Note: because the number of file descriptors is calculated
2513          * in different ways for sizing vs returning the data,
2514          * there is information leakage from the first loop.  However,
2515          * it is of a similar order of magnitude to the leakage from
2516          * global system statistics such as kern.openfiles.
2517          *
2518          * When just doing a count, note that we cannot just count
2519          * the elements and add f_count via the filehead list because
2520          * threaded processes share their descriptor table and f_count might
2521          * still be '1' in that case.
2522          *
2523          * Since the SYSCTL op can block, we must hold the process to
2524          * prevent it being ripped out from under us either in the
2525          * file descriptor loop or in the greater LIST_FOREACH.  The
2526          * process may be in varying states of disrepair.  If the process
2527          * is in SZOMB we may have caught it just as it is being removed
2528          * from the allproc list, we must skip it in that case to maintain
2529          * an unbroken chain through the allproc list.
2530          */
2531         info.count = 0;
2532         info.error = 0;
2533         info.req = req;
2534         allproc_scan(sysctl_kern_file_callback, &info);
2535
2536         /*
2537          * When just calculating the size, overestimate a bit to try to
2538          * prevent system activity from causing the buffer-fill call
2539          * to fail later on.
2540          */
2541         if (req->oldptr == NULL) {
2542                 info.count = (info.count + 16) + (info.count / 10);
2543                 info.error = SYSCTL_OUT(req, NULL,
2544                                         info.count * sizeof(struct kinfo_file));
2545         }
2546         return (info.error);
2547 }
2548
2549 static int
2550 sysctl_kern_file_callback(struct proc *p, void *data)
2551 {
2552         struct sysctl_kern_file_info *info = data;
2553         struct kinfo_file kf;
2554         struct filedesc *fdp;
2555         struct file *fp;
2556         uid_t uid;
2557         int n;
2558
2559         if (p->p_stat == SIDL || p->p_stat == SZOMB)
2560                 return(0);
2561         if (!PRISON_CHECK(info->req->td->td_proc->p_ucred, p->p_ucred) != 0)
2562                 return(0);
2563         if ((fdp = p->p_fd) == NULL)
2564                 return(0);
2565         spin_lock_rd(&fdp->fd_spin);
2566         for (n = 0; n < fdp->fd_nfiles; ++n) {
2567                 if ((fp = fdp->fd_files[n].fp) == NULL)
2568                         continue;
2569                 if (info->req->oldptr == NULL) {
2570                         ++info->count;
2571                 } else {
2572                         uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2573                         kcore_make_file(&kf, fp, p->p_pid, uid, n);
2574                         spin_unlock_rd(&fdp->fd_spin);
2575                         info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2576                         spin_lock_rd(&fdp->fd_spin);
2577                         if (info->error)
2578                                 break;
2579                 }
2580         }
2581         spin_unlock_rd(&fdp->fd_spin);
2582         if (info->error)
2583                 return(-1);
2584         return(0);
2585 }
2586
2587 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2588     0, 0, sysctl_kern_file, "S,file", "Entire file table");
2589
2590 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2591     &maxfilesperproc, 0, "Maximum files allowed open per process");
2592
2593 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2594     &maxfiles, 0, "Maximum number of files");
2595
2596 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2597     &maxfilesrootres, 0, "Descriptors reserved for root use");
2598
2599 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2600         &nfiles, 0, "System-wide number of open files");
2601
2602 static void
2603 fildesc_drvinit(void *unused)
2604 {
2605         int fd;
2606
2607         dev_ops_add(&fildesc_ops, 0, 0);
2608         for (fd = 0; fd < NUMFDESC; fd++) {
2609                 make_dev(&fildesc_ops, fd,
2610                     UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2611         }
2612         make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2613         make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2614         make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2615 }
2616
2617 /*
2618  * MPSAFE
2619  */
2620 struct fileops badfileops = {
2621         .fo_read = badfo_readwrite,
2622         .fo_write = badfo_readwrite,
2623         .fo_ioctl = badfo_ioctl,
2624         .fo_poll = badfo_poll,
2625         .fo_kqfilter = badfo_kqfilter,
2626         .fo_stat = badfo_stat,
2627         .fo_close = badfo_close,
2628         .fo_shutdown = badfo_shutdown
2629 };
2630
2631 /*
2632  * MPSAFE
2633  */
2634 static int
2635 badfo_readwrite(
2636         struct file *fp,
2637         struct uio *uio,
2638         struct ucred *cred,
2639         int flags
2640 ) {
2641         return (EBADF);
2642 }
2643
2644 /*
2645  * MPSAFE
2646  */
2647 static int
2648 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *cred)
2649 {
2650         return (EBADF);
2651 }
2652
2653 /*
2654  * MPSAFE
2655  */
2656 static int
2657 badfo_poll(struct file *fp, int events, struct ucred *cred)
2658 {
2659         return (0);
2660 }
2661
2662 /*
2663  * MPSAFE
2664  */
2665 static int
2666 badfo_kqfilter(struct file *fp, struct knote *kn)
2667 {
2668         return (0);
2669 }
2670
2671 static int
2672 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2673 {
2674         return (EBADF);
2675 }
2676
2677 /*
2678  * MPSAFE
2679  */
2680 static int
2681 badfo_close(struct file *fp)
2682 {
2683         return (EBADF);
2684 }
2685
2686 /*
2687  * MPSAFE
2688  */
2689 static int
2690 badfo_shutdown(struct file *fp, int how)
2691 {
2692         return (EBADF);
2693 }
2694
2695 /*
2696  * MPSAFE
2697  */
2698 int
2699 nofo_shutdown(struct file *fp, int how)
2700 {
2701         return (EOPNOTSUPP);
2702 }
2703
2704 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2705                                         fildesc_drvinit,NULL)