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  */
  74
  75 #include "opt_compat.h"
  76 #include <sys/param.h>
  77 #include <sys/systm.h>
  78 #include <sys/malloc.h>
  79 #include <sys/sysproto.h>
  80 #include <sys/conf.h>
  81 #include <sys/device.h>
  82 #include <sys/file.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 void ffree(struct file *fp);
 112
 113 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
 114 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
 115                      "file desc to leader structures");
 116 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
 117 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
 118
 119 static struct krate krate_uidinfo = { .freq = 1 };
 120
 121 static   d_open_t  fdopen;
 122 #define NUMFDESC 64
 123
 124 #define CDEV_MAJOR 22
 125 static struct dev_ops fildesc_ops = {
 126         { "FD", 0, 0 },
 127         .d_open =       fdopen,
 128 };
 129
 130 /*
 131  * Descriptor management.
 132  */
 133 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
 134 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
 135 static int nfiles;              /* actual number of open files */
 136 extern int cmask;
 137
 138 /*
 139  * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
 140  *
 141  * MPSAFE - must be called with fdp->fd_spin exclusively held
 142  */
 143 static __inline
 144 void
 145 fdfixup_locked(struct filedesc *fdp, int fd)
 146 {
 147         if (fd < fdp->fd_freefile) {
 148                fdp->fd_freefile = fd;
 149         }
 150         while (fdp->fd_lastfile >= 0 &&
 151                fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
 152                fdp->fd_files[fdp->fd_lastfile].reserved == 0
 153         ) {
 154                 --fdp->fd_lastfile;
 155         }
 156 }
 157
 158 /*
 159  * System calls on descriptors.
 160  *
 161  * MPSAFE
 162  */
 163 int
 164 sys_getdtablesize(struct getdtablesize_args *uap)
 165 {
 166         struct proc *p = curproc;
 167         struct plimit *limit = p->p_limit;
 168         int dtsize;
 169
 170         spin_lock(&limit->p_spin);
 171         if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
 172                 dtsize = INT_MAX;
 173         else
 174                 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
 175         spin_unlock(&limit->p_spin);
 176
 177         if (dtsize > maxfilesperproc)
 178                 dtsize = maxfilesperproc;
 179         if (dtsize < minfilesperproc)
 180                 dtsize = minfilesperproc;
 181         if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
 182                 dtsize = maxfilesperuser;
 183         uap->sysmsg_result = dtsize;
 184         return (0);
 185 }
 186
 187 /*
 188  * Duplicate a file descriptor to a particular value.
 189  *
 190  * note: keep in mind that a potential race condition exists when closing
 191  * descriptors from a shared descriptor table (via rfork).
 192  *
 193  * MPSAFE
 194  */
 195 int
 196 sys_dup2(struct dup2_args *uap)
 197 {
 198         int error;
 199         int fd = 0;
 200
 201         error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
 202         uap->sysmsg_fds[0] = fd;
 203
 204         return (error);
 205 }
 206
 207 /*
 208  * Duplicate a file descriptor.
 209  *
 210  * MPSAFE
 211  */
 212 int
 213 sys_dup(struct dup_args *uap)
 214 {
 215         int error;
 216         int fd = 0;
 217
 218         error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
 219         uap->sysmsg_fds[0] = fd;
 220
 221         return (error);
 222 }
 223
 224 /*
 225  * MPALMOSTSAFE - acquires mplock for fp operations
 226  */
 227 int
 228 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
 229 {
 230         struct thread *td = curthread;
 231         struct proc *p = td->td_proc;
 232         struct file *fp;
 233         struct vnode *vp;
 234         u_int newmin;
 235         u_int oflags;
 236         u_int nflags;
 237         int tmp, error, flg = F_POSIX;
 238
 239         KKASSERT(p);
 240
 241         /*
 242          * Operations on file descriptors that do not require a file pointer.
 243          */
 244         switch (cmd) {
 245         case F_GETFD:
 246                 error = fgetfdflags(p->p_fd, fd, &tmp);
 247                 if (error == 0)
 248                         dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
 249                 return (error);
 250
 251         case F_SETFD:
 252                 if (dat->fc_cloexec & FD_CLOEXEC)
 253                         error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
 254                 else
 255                         error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
 256                 return (error);
 257         case F_DUPFD:
 258                 newmin = dat->fc_fd;
 259                 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
 260                 return (error);
 261         default:
 262                 break;
 263         }
 264
 265         /*
 266          * Operations on file pointers
 267          */
 268         if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
 269                 return (EBADF);
 270
 271         switch (cmd) {
 272         case F_GETFL:
 273                 dat->fc_flags = OFLAGS(fp->f_flag);
 274                 error = 0;
 275                 break;
 276
 277         case F_SETFL:
 278                 oflags = fp->f_flag;
 279                 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
 280                 nflags |= oflags & ~FCNTLFLAGS;
 281
 282                 error = 0;
 283                 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
 284                         error = EINVAL;
 285                 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
 286                         tmp = nflags & FASYNC;
 287                         error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
 288                                          cred, NULL);
 289                 }
 290                 if (error == 0)
 291                         fp->f_flag = nflags;
 292                 break;
 293
 294         case F_GETOWN:
 295                 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
 296                                  cred, NULL);
 297                 break;
 298
 299         case F_SETOWN:
 300                 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
 301                                  cred, NULL);
 302                 break;
 303
 304         case F_SETLKW:
 305                 flg |= F_WAIT;
 306                 /* Fall into F_SETLK */
 307
 308         case F_SETLK:
 309                 if (fp->f_type != DTYPE_VNODE) {
 310                         error = EBADF;
 311                         break;
 312                 }
 313                 vp = (struct vnode *)fp->f_data;
 314
 315                 /*
 316                  * copyin/lockop may block
 317                  */
 318                 if (dat->fc_flock.l_whence == SEEK_CUR)
 319                         dat->fc_flock.l_start += fp->f_offset;
 320
 321                 switch (dat->fc_flock.l_type) {
 322                 case F_RDLCK:
 323                         if ((fp->f_flag & FREAD) == 0) {
 324                                 error = EBADF;
 325                                 break;
 326                         }
 327                         p->p_leader->p_flag |= P_ADVLOCK;
 328                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
 329                             &dat->fc_flock, flg);
 330                         break;
 331                 case F_WRLCK:
 332                         if ((fp->f_flag & FWRITE) == 0) {
 333                                 error = EBADF;
 334                                 break;
 335                         }
 336                         p->p_leader->p_flag |= P_ADVLOCK;
 337                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
 338                             &dat->fc_flock, flg);
 339                         break;
 340                 case F_UNLCK:
 341                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
 342                                 &dat->fc_flock, F_POSIX);
 343                         break;
 344                 default:
 345                         error = EINVAL;
 346                         break;
 347                 }
 348
 349                 /*
 350                  * It is possible to race a close() on the descriptor while
 351                  * we were blocked getting the lock.  If this occurs the
 352                  * close might not have caught the lock.
 353                  */
 354                 if (checkfdclosed(p->p_fd, fd, fp)) {
 355                         dat->fc_flock.l_whence = SEEK_SET;
 356                         dat->fc_flock.l_start = 0;
 357                         dat->fc_flock.l_len = 0;
 358                         dat->fc_flock.l_type = F_UNLCK;
 359                         (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
 360                                            F_UNLCK, &dat->fc_flock, F_POSIX);
 361                 }
 362                 break;
 363
 364         case F_GETLK:
 365                 if (fp->f_type != DTYPE_VNODE) {
 366                         error = EBADF;
 367                         break;
 368                 }
 369                 vp = (struct vnode *)fp->f_data;
 370                 /*
 371                  * copyin/lockop may block
 372                  */
 373                 if (dat->fc_flock.l_type != F_RDLCK &&
 374                     dat->fc_flock.l_type != F_WRLCK &&
 375                     dat->fc_flock.l_type != F_UNLCK) {
 376                         error = EINVAL;
 377                         break;
 378                 }
 379                 if (dat->fc_flock.l_whence == SEEK_CUR)
 380                         dat->fc_flock.l_start += fp->f_offset;
 381                 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
 382                             &dat->fc_flock, F_POSIX);
 383                 break;
 384         default:
 385                 error = EINVAL;
 386                 break;
 387         }
 388
 389         fdrop(fp);
 390         return (error);
 391 }
 392
 393 /*
 394  * The file control system call.
 395  *
 396  * MPSAFE
 397  */
 398 int
 399 sys_fcntl(struct fcntl_args *uap)
 400 {
 401         union fcntl_dat dat;
 402         int error;
 403
 404         switch (uap->cmd) {
 405         case F_DUPFD:
 406                 dat.fc_fd = uap->arg;
 407                 break;
 408         case F_SETFD:
 409                 dat.fc_cloexec = uap->arg;
 410                 break;
 411         case F_SETFL:
 412                 dat.fc_flags = uap->arg;
 413                 break;
 414         case F_SETOWN:
 415                 dat.fc_owner = uap->arg;
 416                 break;
 417         case F_SETLKW:
 418         case F_SETLK:
 419         case F_GETLK:
 420                 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
 421                                sizeof(struct flock));
 422                 if (error)
 423                         return (error);
 424                 break;
 425         }
 426
 427         error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
 428
 429         if (error == 0) {
 430                 switch (uap->cmd) {
 431                 case F_DUPFD:
 432                         uap->sysmsg_result = dat.fc_fd;
 433                         break;
 434                 case F_GETFD:
 435                         uap->sysmsg_result = dat.fc_cloexec;
 436                         break;
 437                 case F_GETFL:
 438                         uap->sysmsg_result = dat.fc_flags;
 439                         break;
 440                 case F_GETOWN:
 441                         uap->sysmsg_result = dat.fc_owner;
 442                 case F_GETLK:
 443                         error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
 444                             sizeof(struct flock));
 445                         break;
 446                 }
 447         }
 448
 449         return (error);
 450 }
 451
 452 /*
 453  * Common code for dup, dup2, and fcntl(F_DUPFD).
 454  *
 455  * The type flag can be either DUP_FIXED or DUP_VARIABLE.  DUP_FIXED tells
 456  * kern_dup() to destructively dup over an existing file descriptor if new
 457  * is already open.  DUP_VARIABLE tells kern_dup() to find the lowest
 458  * unused file descriptor that is greater than or equal to new.
 459  *
 460  * MPSAFE
 461  */
 462 int
 463 kern_dup(enum dup_type type, int old, int new, int *res)
 464 {
 465         struct thread *td = curthread;
 466         struct proc *p = td->td_proc;
 467         struct filedesc *fdp = p->p_fd;
 468         struct file *fp;
 469         struct file *delfp;
 470         int oldflags;
 471         int holdleaders;
 472         int dtsize;
 473         int error, newfd;
 474
 475         /*
 476          * Verify that we have a valid descriptor to dup from and
 477          * possibly to dup to.
 478          *
 479          * NOTE: maxfilesperuser is not applicable to dup()
 480          */
 481 retry:
 482         if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
 483                 dtsize = INT_MAX;
 484         else
 485                 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
 486         if (dtsize > maxfilesperproc)
 487                 dtsize = maxfilesperproc;
 488         if (dtsize < minfilesperproc)
 489                 dtsize = minfilesperproc;
 490
 491         if (new < 0 || new > dtsize)
 492                 return (EINVAL);
 493
 494         spin_lock(&fdp->fd_spin);
 495         if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
 496                 spin_unlock(&fdp->fd_spin);
 497                 return (EBADF);
 498         }
 499         if (type == DUP_FIXED && old == new) {
 500                 *res = new;
 501                 spin_unlock(&fdp->fd_spin);
 502                 return (0);
 503         }
 504         fp = fdp->fd_files[old].fp;
 505         oldflags = fdp->fd_files[old].fileflags;
 506         fhold(fp);      /* MPSAFE - can be called with a spinlock held */
 507
 508         /*
 509          * Allocate a new descriptor if DUP_VARIABLE, or expand the table
 510          * if the requested descriptor is beyond the current table size.
 511          *
 512          * This can block.  Retry if the source descriptor no longer matches
 513          * or if our expectation in the expansion case races.
 514          *
 515          * If we are not expanding or allocating a new decriptor, then reset
 516          * the target descriptor to a reserved state so we have a uniform
 517          * setup for the next code block.
 518          */
 519         if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
 520                 spin_unlock(&fdp->fd_spin);
 521                 error = fdalloc(p, new, &newfd);
 522                 spin_lock(&fdp->fd_spin);
 523                 if (error) {
 524                         spin_unlock(&fdp->fd_spin);
 525                         fdrop(fp);
 526                         return (error);
 527                 }
 528                 /*
 529                  * Check for ripout
 530                  */
 531                 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
 532                         fsetfd_locked(fdp, NULL, newfd);
 533                         spin_unlock(&fdp->fd_spin);
 534                         fdrop(fp);
 535                         goto retry;
 536                 }
 537                 /*
 538                  * Check for expansion race
 539                  */
 540                 if (type != DUP_VARIABLE && new != newfd) {
 541                         fsetfd_locked(fdp, NULL, newfd);
 542                         spin_unlock(&fdp->fd_spin);
 543                         fdrop(fp);
 544                         goto retry;
 545                 }
 546                 /*
 547                  * Check for ripout, newfd reused old (this case probably
 548                  * can't occur).
 549                  */
 550                 if (old == newfd) {
 551                         fsetfd_locked(fdp, NULL, newfd);
 552                         spin_unlock(&fdp->fd_spin);
 553                         fdrop(fp);
 554                         goto retry;
 555                 }
 556                 new = newfd;
 557                 delfp = NULL;
 558         } else {
 559                 if (fdp->fd_files[new].reserved) {
 560                         spin_unlock(&fdp->fd_spin);
 561                         fdrop(fp);
 562                         kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
 563                         tsleep(fdp, 0, "fdres", hz);
 564                         goto retry;
 565                 }
 566
 567                 /*
 568                  * If the target descriptor was never allocated we have
 569                  * to allocate it.  If it was we have to clean out the
 570                  * old descriptor.  delfp inherits the ref from the
 571                  * descriptor table.
 572                  */
 573                 delfp = fdp->fd_files[new].fp;
 574                 fdp->fd_files[new].fp = NULL;
 575                 fdp->fd_files[new].reserved = 1;
 576                 if (delfp == NULL) {
 577                         fdreserve_locked(fdp, new, 1);
 578                         if (new > fdp->fd_lastfile)
 579                                 fdp->fd_lastfile = new;
 580                 }
 581
 582         }
 583
 584         /*
 585          * NOTE: still holding an exclusive spinlock
 586          */
 587
 588         /*
 589          * If a descriptor is being overwritten we may hve to tell
 590          * fdfree() to sleep to ensure that all relevant process
 591          * leaders can be traversed in closef().
 592          */
 593         if (delfp != NULL && p->p_fdtol != NULL) {
 594                 fdp->fd_holdleaderscount++;
 595                 holdleaders = 1;
 596         } else {
 597                 holdleaders = 0;
 598         }
 599         KASSERT(delfp == NULL || type == DUP_FIXED,
 600                 ("dup() picked an open file"));
 601
 602         /*
 603          * Duplicate the source descriptor, update lastfile.  If the new
 604          * descriptor was not allocated and we aren't replacing an existing
 605          * descriptor we have to mark the descriptor as being in use.
 606          *
 607          * The fd_files[] array inherits fp's hold reference.
 608          */
 609         fsetfd_locked(fdp, fp, new);
 610         fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
 611         spin_unlock(&fdp->fd_spin);
 612         fdrop(fp);
 613         *res = new;
 614
 615         /*
 616          * If we dup'd over a valid file, we now own the reference to it
 617          * and must dispose of it using closef() semantics (as if a
 618          * close() were performed on it).
 619          */
 620         if (delfp) {
 621                 if (SLIST_FIRST(&delfp->f_klist))
 622                         knote_fdclose(delfp, fdp, new);
 623                 closef(delfp, p);
 624                 if (holdleaders) {
 625                         spin_lock(&fdp->fd_spin);
 626                         fdp->fd_holdleaderscount--;
 627                         if (fdp->fd_holdleaderscount == 0 &&
 628                             fdp->fd_holdleaderswakeup != 0) {
 629                                 fdp->fd_holdleaderswakeup = 0;
 630                                 spin_unlock(&fdp->fd_spin);
 631                                 wakeup(&fdp->fd_holdleaderscount);
 632                         } else {
 633                                 spin_unlock(&fdp->fd_spin);
 634                         }
 635                 }
 636         }
 637         return (0);
 638 }
 639
 640 /*
 641  * If sigio is on the list associated with a process or process group,
 642  * disable signalling from the device, remove sigio from the list and
 643  * free sigio.
 644  *
 645  * MPSAFE
 646  */
 647 void
 648 funsetown(struct sigio **sigiop)
 649 {
 650         struct pgrp *pgrp;
 651         struct proc *p;
 652         struct sigio *sigio;
 653
 654         if ((sigio = *sigiop) != NULL) {
 655                 lwkt_gettoken(&proc_token);     /* protect sigio */
 656                 KKASSERT(sigiop == sigio->sio_myref);
 657                 sigio = *sigiop;
 658                 *sigiop = NULL;
 659                 lwkt_reltoken(&proc_token);
 660         }
 661         if (sigio == NULL)
 662                 return;
 663
 664         if (sigio->sio_pgid < 0) {
 665                 pgrp = sigio->sio_pgrp;
 666                 sigio->sio_pgrp = NULL;
 667                 lwkt_gettoken(&pgrp->pg_token);
 668                 SLIST_REMOVE(&pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio);
 669                 lwkt_reltoken(&pgrp->pg_token);
 670                 pgrel(pgrp);
 671         } else /* if ((*sigiop)->sio_pgid > 0) */ {
 672                 p = sigio->sio_proc;
 673                 sigio->sio_proc = NULL;
 674                 PHOLD(p);
 675                 lwkt_gettoken(&p->p_token);
 676                 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
 677                 lwkt_reltoken(&p->p_token);
 678                 PRELE(p);
 679         }
 680         crfree(sigio->sio_ucred);
 681         sigio->sio_ucred = NULL;
 682         kfree(sigio, M_SIGIO);
 683 }
 684
 685 /*
 686  * Free a list of sigio structures.  Caller is responsible for ensuring
 687  * that the list is MPSAFE.
 688  *
 689  * MPSAFE
 690  */
 691 void
 692 funsetownlst(struct sigiolst *sigiolst)
 693 {
 694         struct sigio *sigio;
 695
 696         while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
 697                 funsetown(sigio->sio_myref);
 698 }
 699
 700 /*
 701  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
 702  *
 703  * After permission checking, add a sigio structure to the sigio list for
 704  * the process or process group.
 705  *
 706  * MPSAFE
 707  */
 708 int
 709 fsetown(pid_t pgid, struct sigio **sigiop)
 710 {
 711         struct proc *proc = NULL;
 712         struct pgrp *pgrp = NULL;
 713         struct sigio *sigio;
 714         int error;
 715
 716         if (pgid == 0) {
 717                 funsetown(sigiop);
 718                 return (0);
 719         }
 720
 721         if (pgid > 0) {
 722                 proc = pfind(pgid);
 723                 if (proc == NULL) {
 724                         error = ESRCH;
 725                         goto done;
 726                 }
 727
 728                 /*
 729                  * Policy - Don't allow a process to FSETOWN a process
 730                  * in another session.
 731                  *
 732                  * Remove this test to allow maximum flexibility or
 733                  * restrict FSETOWN to the current process or process
 734                  * group for maximum safety.
 735                  */
 736                 if (proc->p_session != curproc->p_session) {
 737                         error = EPERM;
 738                         goto done;
 739                 }
 740         } else /* if (pgid < 0) */ {
 741                 pgrp = pgfind(-pgid);
 742                 if (pgrp == NULL) {
 743                         error = ESRCH;
 744                         goto done;
 745                 }
 746
 747                 /*
 748                  * Policy - Don't allow a process to FSETOWN a process
 749                  * in another session.
 750                  *
 751                  * Remove this test to allow maximum flexibility or
 752                  * restrict FSETOWN to the current process or process
 753                  * group for maximum safety.
 754                  */
 755                 if (pgrp->pg_session != curproc->p_session) {
 756                         error = EPERM;
 757                         goto done;
 758                 }
 759         }
 760         sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK | M_ZERO);
 761         if (pgid > 0) {
 762                 KKASSERT(pgrp == NULL);
 763                 lwkt_gettoken(&proc->p_token);
 764                 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
 765                 sigio->sio_proc = proc;
 766                 lwkt_reltoken(&proc->p_token);
 767         } else {
 768                 KKASSERT(proc == NULL);
 769                 lwkt_gettoken(&pgrp->pg_token);
 770                 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
 771                 sigio->sio_pgrp = pgrp;
 772                 lwkt_reltoken(&pgrp->pg_token);
 773                 pgrp = NULL;
 774         }
 775         sigio->sio_pgid = pgid;
 776         sigio->sio_ucred = crhold(curthread->td_ucred);
 777         /* It would be convenient if p_ruid was in ucred. */
 778         sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
 779         sigio->sio_myref = sigiop;
 780
 781         lwkt_gettoken(&proc_token);
 782         while (*sigiop)
 783                 funsetown(sigiop);
 784         *sigiop = sigio;
 785         lwkt_reltoken(&proc_token);
 786         error = 0;
 787 done:
 788         if (pgrp)
 789                 pgrel(pgrp);
 790         if (proc)
 791                 PRELE(proc);
 792         return (error);
 793 }
 794
 795 /*
 796  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
 797  *
 798  * MPSAFE
 799  */
 800 pid_t
 801 fgetown(struct sigio **sigiop)
 802 {
 803         struct sigio *sigio;
 804         pid_t own;
 805
 806         lwkt_gettoken(&proc_token);
 807         sigio = *sigiop;
 808         own = (sigio != NULL ? sigio->sio_pgid : 0);
 809         lwkt_reltoken(&proc_token);
 810
 811         return (own);
 812 }
 813
 814 /*
 815  * Close many file descriptors.
 816  *
 817  * MPSAFE
 818  */
 819 int
 820 sys_closefrom(struct closefrom_args *uap)
 821 {
 822         return(kern_closefrom(uap->fd));
 823 }
 824
 825 /*
 826  * Close all file descriptors greater then or equal to fd
 827  *
 828  * MPSAFE
 829  */
 830 int
 831 kern_closefrom(int fd)
 832 {
 833         struct thread *td = curthread;
 834         struct proc *p = td->td_proc;
 835         struct filedesc *fdp;
 836
 837         KKASSERT(p);
 838         fdp = p->p_fd;
 839
 840         if (fd < 0)
 841                 return (EINVAL);
 842
 843         /*
 844          * NOTE: This function will skip unassociated descriptors and
 845          * reserved descriptors that have not yet been assigned.
 846          * fd_lastfile can change as a side effect of kern_close().
 847          */
 848         spin_lock(&fdp->fd_spin);
 849         while (fd <= fdp->fd_lastfile) {
 850                 if (fdp->fd_files[fd].fp != NULL) {
 851                         spin_unlock(&fdp->fd_spin);
 852                         /* ok if this races another close */
 853                         if (kern_close(fd) == EINTR)
 854                                 return (EINTR);
 855                         spin_lock(&fdp->fd_spin);
 856                 }
 857                 ++fd;
 858         }
 859         spin_unlock(&fdp->fd_spin);
 860         return (0);
 861 }
 862
 863 /*
 864  * Close a file descriptor.
 865  *
 866  * MPSAFE
 867  */
 868 int
 869 sys_close(struct close_args *uap)
 870 {
 871         return(kern_close(uap->fd));
 872 }
 873
 874 /*
 875  * MPSAFE
 876  */
 877 int
 878 kern_close(int fd)
 879 {
 880         struct thread *td = curthread;
 881         struct proc *p = td->td_proc;
 882         struct filedesc *fdp;
 883         struct file *fp;
 884         int error;
 885         int holdleaders;
 886
 887         KKASSERT(p);
 888         fdp = p->p_fd;
 889
 890         spin_lock(&fdp->fd_spin);
 891         if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
 892                 spin_unlock(&fdp->fd_spin);
 893                 return (EBADF);
 894         }
 895         holdleaders = 0;
 896         if (p->p_fdtol != NULL) {
 897                 /*
 898                  * Ask fdfree() to sleep to ensure that all relevant
 899                  * process leaders can be traversed in closef().
 900                  */
 901                 fdp->fd_holdleaderscount++;
 902                 holdleaders = 1;
 903         }
 904
 905         /*
 906          * we now hold the fp reference that used to be owned by the descriptor
 907          * array.
 908          */
 909         spin_unlock(&fdp->fd_spin);
 910         if (SLIST_FIRST(&fp->f_klist))
 911                 knote_fdclose(fp, fdp, fd);
 912         error = closef(fp, p);
 913         if (holdleaders) {
 914                 spin_lock(&fdp->fd_spin);
 915                 fdp->fd_holdleaderscount--;
 916                 if (fdp->fd_holdleaderscount == 0 &&
 917                     fdp->fd_holdleaderswakeup != 0) {
 918                         fdp->fd_holdleaderswakeup = 0;
 919                         spin_unlock(&fdp->fd_spin);
 920                         wakeup(&fdp->fd_holdleaderscount);
 921                 } else {
 922                         spin_unlock(&fdp->fd_spin);
 923                 }
 924         }
 925         return (error);
 926 }
 927
 928 /*
 929  * shutdown_args(int fd, int how)
 930  */
 931 int
 932 kern_shutdown(int fd, int how)
 933 {
 934         struct thread *td = curthread;
 935         struct proc *p = td->td_proc;
 936         struct file *fp;
 937         int error;
 938
 939         KKASSERT(p);
 940
 941         if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
 942                 return (EBADF);
 943         error = fo_shutdown(fp, how);
 944         fdrop(fp);
 945
 946         return (error);
 947 }
 948
 949 /*
 950  * MPALMOSTSAFE
 951  */
 952 int
 953 sys_shutdown(struct shutdown_args *uap)
 954 {
 955         int error;
 956
 957         error = kern_shutdown(uap->s, uap->how);
 958
 959         return (error);
 960 }
 961
 962 /*
 963  * MPSAFE
 964  */
 965 int
 966 kern_fstat(int fd, struct stat *ub)
 967 {
 968         struct thread *td = curthread;
 969         struct proc *p = td->td_proc;
 970         struct file *fp;
 971         int error;
 972
 973         KKASSERT(p);
 974
 975         if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
 976                 return (EBADF);
 977         error = fo_stat(fp, ub, td->td_ucred);
 978         fdrop(fp);
 979
 980         return (error);
 981 }
 982
 983 /*
 984  * Return status information about a file descriptor.
 985  *
 986  * MPSAFE
 987  */
 988 int
 989 sys_fstat(struct fstat_args *uap)
 990 {
 991         struct stat st;
 992         int error;
 993
 994         error = kern_fstat(uap->fd, &st);
 995
 996         if (error == 0)
 997                 error = copyout(&st, uap->sb, sizeof(st));
 998         return (error);
 999 }
1000
1001 /*
1002  * Return pathconf information about a file descriptor.
1003  *
1004  * MPALMOSTSAFE
1005  */
1006 int
1007 sys_fpathconf(struct fpathconf_args *uap)
1008 {
1009         struct thread *td = curthread;
1010         struct proc *p = td->td_proc;
1011         struct file *fp;
1012         struct vnode *vp;
1013         int error = 0;
1014
1015         if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
1016                 return (EBADF);
1017
1018         switch (fp->f_type) {
1019         case DTYPE_PIPE:
1020         case DTYPE_SOCKET:
1021                 if (uap->name != _PC_PIPE_BUF) {
1022                         error = EINVAL;
1023                 } else {
1024                         uap->sysmsg_result = PIPE_BUF;
1025                         error = 0;
1026                 }
1027                 break;
1028         case DTYPE_FIFO:
1029         case DTYPE_VNODE:
1030                 vp = (struct vnode *)fp->f_data;
1031                 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
1032                 break;
1033         default:
1034                 error = EOPNOTSUPP;
1035                 break;
1036         }
1037         fdrop(fp);
1038         return(error);
1039 }
1040
1041 static int fdexpand;
1042 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0,
1043     "Number of times a file table has been expanded");
1044
1045 /*
1046  * Grow the file table so it can hold through descriptor (want).
1047  *
1048  * The fdp's spinlock must be held exclusively on entry and may be held
1049  * exclusively on return.  The spinlock may be cycled by the routine.
1050  *
1051  * MPSAFE
1052  */
1053 static void
1054 fdgrow_locked(struct filedesc *fdp, int want)
1055 {
1056         struct fdnode *newfiles;
1057         struct fdnode *oldfiles;
1058         int nf, extra;
1059
1060         nf = fdp->fd_nfiles;
1061         do {
1062                 /* nf has to be of the form 2^n - 1 */
1063                 nf = 2 * nf + 1;
1064         } while (nf <= want);
1065
1066         spin_unlock(&fdp->fd_spin);
1067         newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
1068         spin_lock(&fdp->fd_spin);
1069
1070         /*
1071          * We could have raced another extend while we were not holding
1072          * the spinlock.
1073          */
1074         if (fdp->fd_nfiles >= nf) {
1075                 spin_unlock(&fdp->fd_spin);
1076                 kfree(newfiles, M_FILEDESC);
1077                 spin_lock(&fdp->fd_spin);
1078                 return;
1079         }
1080         /*
1081          * Copy the existing ofile and ofileflags arrays
1082          * and zero the new portion of each array.
1083          */
1084         extra = nf - fdp->fd_nfiles;
1085         bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1086         bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1087
1088         oldfiles = fdp->fd_files;
1089         fdp->fd_files = newfiles;
1090         fdp->fd_nfiles = nf;
1091
1092         if (oldfiles != fdp->fd_builtin_files) {
1093                 spin_unlock(&fdp->fd_spin);
1094                 kfree(oldfiles, M_FILEDESC);
1095                 spin_lock(&fdp->fd_spin);
1096         }
1097         fdexpand++;
1098 }
1099
1100 /*
1101  * Number of nodes in right subtree, including the root.
1102  */
1103 static __inline int
1104 right_subtree_size(int n)
1105 {
1106         return (n ^ (n | (n + 1)));
1107 }
1108
1109 /*
1110  * Bigger ancestor.
1111  */
1112 static __inline int
1113 right_ancestor(int n)
1114 {
1115         return (n | (n + 1));
1116 }
1117
1118 /*
1119  * Smaller ancestor.
1120  */
1121 static __inline int
1122 left_ancestor(int n)
1123 {
1124         return ((n & (n + 1)) - 1);
1125 }
1126
1127 /*
1128  * Traverse the in-place binary tree buttom-up adjusting the allocation
1129  * count so scans can determine where free descriptors are located.
1130  *
1131  * MPSAFE - caller must be holding an exclusive spinlock on fdp
1132  */
1133 static
1134 void
1135 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1136 {
1137         while (fd >= 0) {
1138                 fdp->fd_files[fd].allocated += incr;
1139                 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1140                 fd = left_ancestor(fd);
1141         }
1142 }
1143
1144 /*
1145  * Reserve a file descriptor for the process.  If no error occurs, the
1146  * caller MUST at some point call fsetfd() or assign a file pointer
1147  * or dispose of the reservation.
1148  *
1149  * MPSAFE
1150  */
1151 int
1152 fdalloc(struct proc *p, int want, int *result)
1153 {
1154         struct filedesc *fdp = p->p_fd;
1155         struct uidinfo *uip;
1156         int fd, rsize, rsum, node, lim;
1157
1158         /*
1159          * Check dtable size limit
1160          */
1161         spin_lock(&p->p_limit->p_spin);
1162         if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1163                 lim = INT_MAX;
1164         else
1165                 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1166         spin_unlock(&p->p_limit->p_spin);
1167
1168         if (lim > maxfilesperproc)
1169                 lim = maxfilesperproc;
1170         if (lim < minfilesperproc)
1171                 lim = minfilesperproc;
1172         if (want >= lim)
1173                 return (EMFILE);
1174
1175         /*
1176          * Check that the user has not run out of descriptors (non-root only).
1177          * As a safety measure the dtable is allowed to have at least
1178          * minfilesperproc open fds regardless of the maxfilesperuser limit.
1179          */
1180         if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1181                 uip = p->p_ucred->cr_uidinfo;
1182                 if (uip->ui_openfiles > maxfilesperuser) {
1183                         krateprintf(&krate_uidinfo,
1184                                     "Warning: user %d pid %d (%s) ran out of "
1185                                     "file descriptors (%d/%d)\n",
1186                                     p->p_ucred->cr_uid, (int)p->p_pid,
1187                                     p->p_comm,
1188                                     uip->ui_openfiles, maxfilesperuser);
1189                         return(ENFILE);
1190                 }
1191         }
1192
1193         /*
1194          * Grow the dtable if necessary
1195          */
1196         spin_lock(&fdp->fd_spin);
1197         if (want >= fdp->fd_nfiles)
1198                 fdgrow_locked(fdp, want);
1199
1200         /*
1201          * Search for a free descriptor starting at the higher
1202          * of want or fd_freefile.  If that fails, consider
1203          * expanding the ofile array.
1204          *
1205          * NOTE! the 'allocated' field is a cumulative recursive allocation
1206          * count.  If we happen to see a value of 0 then we can shortcut
1207          * our search.  Otherwise we run through through the tree going
1208          * down branches we know have free descriptor(s) until we hit a
1209          * leaf node.  The leaf node will be free but will not necessarily
1210          * have an allocated field of 0.
1211          */
1212 retry:
1213         /* move up the tree looking for a subtree with a free node */
1214         for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1215              fd = right_ancestor(fd)) {
1216                 if (fdp->fd_files[fd].allocated == 0)
1217                         goto found;
1218
1219                 rsize = right_subtree_size(fd);
1220                 if (fdp->fd_files[fd].allocated == rsize)
1221                         continue;       /* right subtree full */
1222
1223                 /*
1224                  * Free fd is in the right subtree of the tree rooted at fd.
1225                  * Call that subtree R.  Look for the smallest (leftmost)
1226                  * subtree of R with an unallocated fd: continue moving
1227                  * down the left branch until encountering a full left
1228                  * subtree, then move to the right.
1229                  */
1230                 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1231                         node = fd + rsize;
1232                         rsum += fdp->fd_files[node].allocated;
1233                         if (fdp->fd_files[fd].allocated == rsum + rsize) {
1234                                 fd = node;      /* move to the right */
1235                                 if (fdp->fd_files[node].allocated == 0)
1236                                         goto found;
1237                                 rsum = 0;
1238                         }
1239                 }
1240                 goto found;
1241         }
1242
1243         /*
1244          * No space in current array.  Expand?
1245          */
1246         if (fdp->fd_nfiles >= lim) {
1247                 spin_unlock(&fdp->fd_spin);
1248                 return (EMFILE);
1249         }
1250         fdgrow_locked(fdp, want);
1251         goto retry;
1252
1253 found:
1254         KKASSERT(fd < fdp->fd_nfiles);
1255         if (fd > fdp->fd_lastfile)
1256                 fdp->fd_lastfile = fd;
1257         if (want <= fdp->fd_freefile)
1258                 fdp->fd_freefile = fd;
1259         *result = fd;
1260         KKASSERT(fdp->fd_files[fd].fp == NULL);
1261         KKASSERT(fdp->fd_files[fd].reserved == 0);
1262         fdp->fd_files[fd].fileflags = 0;
1263         fdp->fd_files[fd].reserved = 1;
1264         fdreserve_locked(fdp, fd, 1);
1265         spin_unlock(&fdp->fd_spin);
1266         return (0);
1267 }
1268
1269 /*
1270  * Check to see whether n user file descriptors
1271  * are available to the process p.
1272  *
1273  * MPSAFE
1274  */
1275 int
1276 fdavail(struct proc *p, int n)
1277 {
1278         struct filedesc *fdp = p->p_fd;
1279         struct fdnode *fdnode;
1280         int i, lim, last;
1281
1282         spin_lock(&p->p_limit->p_spin);
1283         if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1284                 lim = INT_MAX;
1285         else
1286                 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1287         spin_unlock(&p->p_limit->p_spin);
1288
1289         if (lim > maxfilesperproc)
1290                 lim = maxfilesperproc;
1291         if (lim < minfilesperproc)
1292                 lim = minfilesperproc;
1293
1294         spin_lock(&fdp->fd_spin);
1295         if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1296                 spin_unlock(&fdp->fd_spin);
1297                 return (1);
1298         }
1299         last = min(fdp->fd_nfiles, lim);
1300         fdnode = &fdp->fd_files[fdp->fd_freefile];
1301         for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1302                 if (fdnode->fp == NULL && --n <= 0) {
1303                         spin_unlock(&fdp->fd_spin);
1304                         return (1);
1305                 }
1306         }
1307         spin_unlock(&fdp->fd_spin);
1308         return (0);
1309 }
1310
1311 /*
1312  * Revoke open descriptors referencing (f_data, f_type)
1313  *
1314  * Any revoke executed within a prison is only able to
1315  * revoke descriptors for processes within that prison.
1316  *
1317  * Returns 0 on success or an error code.
1318  */
1319 struct fdrevoke_info {
1320         void *data;
1321         short type;
1322         short unused;
1323         int count;
1324         int intransit;
1325         struct ucred *cred;
1326         struct file *nfp;
1327 };
1328
1329 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1330 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1331
1332 int
1333 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1334 {
1335         struct fdrevoke_info info;
1336         int error;
1337
1338         bzero(&info, sizeof(info));
1339         info.data = f_data;
1340         info.type = f_type;
1341         info.cred = cred;
1342         error = falloc(NULL, &info.nfp, NULL);
1343         if (error)
1344                 return (error);
1345
1346         /*
1347          * Scan the file pointer table once.  dups do not dup file pointers,
1348          * only descriptors, so there is no leak.  Set FREVOKED on the fps
1349          * being revoked.
1350          */
1351         allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1352
1353         /*
1354          * If any fps were marked track down the related descriptors
1355          * and close them.  Any dup()s at this point will notice
1356          * the FREVOKED already set in the fp and do the right thing.
1357          *
1358          * Any fps with non-zero msgcounts (aka sent over a unix-domain
1359          * socket) bumped the intransit counter and will require a
1360          * scan.  Races against fps leaving the socket are closed by
1361          * the socket code checking for FREVOKED.
1362          */
1363         if (info.count)
1364                 allproc_scan(fdrevoke_proc_callback, &info);
1365         if (info.intransit)
1366                 unp_revoke_gc(info.nfp);
1367         fdrop(info.nfp);
1368         return(0);
1369 }
1370
1371 /*
1372  * Locate matching file pointers directly.
1373  *
1374  * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
1375  */
1376 static int
1377 fdrevoke_check_callback(struct file *fp, void *vinfo)
1378 {
1379         struct fdrevoke_info *info = vinfo;
1380
1381         /*
1382          * File pointers already flagged for revokation are skipped.
1383          */
1384         if (fp->f_flag & FREVOKED)
1385                 return(0);
1386
1387         /*
1388          * If revoking from a prison file pointers created outside of
1389          * that prison, or file pointers without creds, cannot be revoked.
1390          */
1391         if (info->cred->cr_prison &&
1392             (fp->f_cred == NULL ||
1393              info->cred->cr_prison != fp->f_cred->cr_prison)) {
1394                 return(0);
1395         }
1396
1397         /*
1398          * If the file pointer matches then mark it for revocation.  The
1399          * flag is currently only used by unp_revoke_gc().
1400          *
1401          * info->count is a heuristic and can race in a SMP environment.
1402          */
1403         if (info->data == fp->f_data && info->type == fp->f_type) {
1404                 atomic_set_int(&fp->f_flag, FREVOKED);
1405                 info->count += fp->f_count;
1406                 if (fp->f_msgcount)
1407                         ++info->intransit;
1408         }
1409         return(0);
1410 }
1411
1412 /*
1413  * Locate matching file pointers via process descriptor tables.
1414  */
1415 static int
1416 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1417 {
1418         struct fdrevoke_info *info = vinfo;
1419         struct filedesc *fdp;
1420         struct file *fp;
1421         int n;
1422
1423         if (p->p_stat == SIDL || p->p_stat == SZOMB)
1424                 return(0);
1425         if (info->cred->cr_prison &&
1426             info->cred->cr_prison != p->p_ucred->cr_prison) {
1427                 return(0);
1428         }
1429
1430         /*
1431          * If the controlling terminal of the process matches the
1432          * vnode being revoked we clear the controlling terminal.
1433          *
1434          * The normal spec_close() may not catch this because it
1435          * uses curproc instead of p.
1436          */
1437         if (p->p_session && info->type == DTYPE_VNODE &&
1438             info->data == p->p_session->s_ttyvp) {
1439                 p->p_session->s_ttyvp = NULL;
1440                 vrele(info->data);
1441         }
1442
1443         /*
1444          * Softref the fdp to prevent it from being destroyed
1445          */
1446         spin_lock(&p->p_spin);
1447         if ((fdp = p->p_fd) == NULL) {
1448                 spin_unlock(&p->p_spin);
1449                 return(0);
1450         }
1451         atomic_add_int(&fdp->fd_softrefs, 1);
1452         spin_unlock(&p->p_spin);
1453
1454         /*
1455          * Locate and close any matching file descriptors.
1456          */
1457         spin_lock(&fdp->fd_spin);
1458         for (n = 0; n < fdp->fd_nfiles; ++n) {
1459                 if ((fp = fdp->fd_files[n].fp) == NULL)
1460                         continue;
1461                 if (fp->f_flag & FREVOKED) {
1462                         fhold(info->nfp);
1463                         fdp->fd_files[n].fp = info->nfp;
1464                         spin_unlock(&fdp->fd_spin);
1465                         knote_fdclose(fp, fdp, n);      /* XXX */
1466                         closef(fp, p);
1467                         spin_lock(&fdp->fd_spin);
1468                         --info->count;
1469                 }
1470         }
1471         spin_unlock(&fdp->fd_spin);
1472         atomic_subtract_int(&fdp->fd_softrefs, 1);
1473         return(0);
1474 }
1475
1476 /*
1477  * falloc:
1478  *      Create a new open file structure and reserve a file decriptor
1479  *      for the process that refers to it.
1480  *
1481  *      Root creds are checked using lp, or assumed if lp is NULL.  If
1482  *      resultfd is non-NULL then lp must also be non-NULL.  No file
1483  *      descriptor is reserved (and no process context is needed) if
1484  *      resultfd is NULL.
1485  *
1486  *      A file pointer with a refcount of 1 is returned.  Note that the
1487  *      file pointer is NOT associated with the descriptor.  If falloc
1488  *      returns success, fsetfd() MUST be called to either associate the
1489  *      file pointer or clear the reservation.
1490  *
1491  * MPSAFE
1492  */
1493 int
1494 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1495 {
1496         static struct timeval lastfail;
1497         static int curfail;
1498         struct file *fp;
1499         struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1500         int error;
1501
1502         fp = NULL;
1503
1504         /*
1505          * Handle filetable full issues and root overfill.
1506          */
1507         if (nfiles >= maxfiles - maxfilesrootres &&
1508             (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1509                 if (ppsratecheck(&lastfail, &curfail, 1)) {
1510                         kprintf("kern.maxfiles limit exceeded by uid %d, "
1511                                 "please see tuning(7).\n",
1512                                 cred->cr_ruid);
1513                 }
1514                 error = ENFILE;
1515                 goto done;
1516         }
1517
1518         /*
1519          * Allocate a new file descriptor.
1520          */
1521         fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1522         spin_init(&fp->f_spin);
1523         SLIST_INIT(&fp->f_klist);
1524         fp->f_count = 1;
1525         fp->f_ops = &badfileops;
1526         fp->f_seqcount = 1;
1527         fsetcred(fp, cred);
1528         spin_lock(&filehead_spin);
1529         nfiles++;
1530         LIST_INSERT_HEAD(&filehead, fp, f_list);
1531         spin_unlock(&filehead_spin);
1532         if (resultfd) {
1533                 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1534                         fdrop(fp);
1535                         fp = NULL;
1536                 }
1537         } else {
1538                 error = 0;
1539         }
1540 done:
1541         *resultfp = fp;
1542         return (error);
1543 }
1544
1545 /*
1546  * Check for races against a file descriptor by determining that the
1547  * file pointer is still associated with the specified file descriptor,
1548  * and a close is not currently in progress.
1549  *
1550  * MPSAFE
1551  */
1552 int
1553 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp)
1554 {
1555         int error;
1556
1557         spin_lock(&fdp->fd_spin);
1558         if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1559                 error = EBADF;
1560         else
1561                 error = 0;
1562         spin_unlock(&fdp->fd_spin);
1563         return (error);
1564 }
1565
1566 /*
1567  * Associate a file pointer with a previously reserved file descriptor.
1568  * This function always succeeds.
1569  *
1570  * If fp is NULL, the file descriptor is returned to the pool.
1571  */
1572
1573 /*
1574  * MPSAFE (exclusive spinlock must be held on call)
1575  */
1576 static void
1577 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1578 {
1579         KKASSERT((unsigned)fd < fdp->fd_nfiles);
1580         KKASSERT(fdp->fd_files[fd].reserved != 0);
1581         if (fp) {
1582                 fhold(fp);
1583                 fdp->fd_files[fd].fp = fp;
1584                 fdp->fd_files[fd].reserved = 0;
1585         } else {
1586                 fdp->fd_files[fd].reserved = 0;
1587                 fdreserve_locked(fdp, fd, -1);
1588                 fdfixup_locked(fdp, fd);
1589         }
1590 }
1591
1592 /*
1593  * MPSAFE
1594  */
1595 void
1596 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1597 {
1598         spin_lock(&fdp->fd_spin);
1599         fsetfd_locked(fdp, fp, fd);
1600         spin_unlock(&fdp->fd_spin);
1601 }
1602
1603 /*
1604  * MPSAFE (exclusive spinlock must be held on call)
1605  */
1606 static
1607 struct file *
1608 funsetfd_locked(struct filedesc *fdp, int fd)
1609 {
1610         struct file *fp;
1611
1612         if ((unsigned)fd >= fdp->fd_nfiles)
1613                 return (NULL);
1614         if ((fp = fdp->fd_files[fd].fp) == NULL)
1615                 return (NULL);
1616         fdp->fd_files[fd].fp = NULL;
1617         fdp->fd_files[fd].fileflags = 0;
1618
1619         fdreserve_locked(fdp, fd, -1);
1620         fdfixup_locked(fdp, fd);
1621         return(fp);
1622 }
1623
1624 /*
1625  * MPSAFE
1626  */
1627 int
1628 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1629 {
1630         int error;
1631
1632         spin_lock(&fdp->fd_spin);
1633         if (((u_int)fd) >= fdp->fd_nfiles) {
1634                 error = EBADF;
1635         } else if (fdp->fd_files[fd].fp == NULL) {
1636                 error = EBADF;
1637         } else {
1638                 *flagsp = fdp->fd_files[fd].fileflags;
1639                 error = 0;
1640         }
1641         spin_unlock(&fdp->fd_spin);
1642         return (error);
1643 }
1644
1645 /*
1646  * MPSAFE
1647  */
1648 int
1649 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1650 {
1651         int error;
1652
1653         spin_lock(&fdp->fd_spin);
1654         if (((u_int)fd) >= fdp->fd_nfiles) {
1655                 error = EBADF;
1656         } else if (fdp->fd_files[fd].fp == NULL) {
1657                 error = EBADF;
1658         } else {
1659                 fdp->fd_files[fd].fileflags |= add_flags;
1660                 error = 0;
1661         }
1662         spin_unlock(&fdp->fd_spin);
1663         return (error);
1664 }
1665
1666 /*
1667  * MPSAFE
1668  */
1669 int
1670 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1671 {
1672         int error;
1673
1674         spin_lock(&fdp->fd_spin);
1675         if (((u_int)fd) >= fdp->fd_nfiles) {
1676                 error = EBADF;
1677         } else if (fdp->fd_files[fd].fp == NULL) {
1678                 error = EBADF;
1679         } else {
1680                 fdp->fd_files[fd].fileflags &= ~rem_flags;
1681                 error = 0;
1682         }
1683         spin_unlock(&fdp->fd_spin);
1684         return (error);
1685 }
1686
1687 /*
1688  * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1689  */
1690 void
1691 fsetcred(struct file *fp, struct ucred *ncr)
1692 {
1693         struct ucred *ocr;
1694         struct uidinfo *uip;
1695
1696         ocr = fp->f_cred;
1697         if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1698                 if (ocr) {
1699                         uip = ocr->cr_uidinfo;
1700                         atomic_add_int(&uip->ui_openfiles, -1);
1701                 }
1702                 if (ncr) {
1703                         uip = ncr->cr_uidinfo;
1704                         atomic_add_int(&uip->ui_openfiles, 1);
1705                 }
1706         }
1707         if (ncr)
1708                 crhold(ncr);
1709         fp->f_cred = ncr;
1710         if (ocr)
1711                 crfree(ocr);
1712 }
1713
1714 /*
1715  * Free a file descriptor.
1716  */
1717 static
1718 void
1719 ffree(struct file *fp)
1720 {
1721         KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1722         spin_lock(&filehead_spin);
1723         LIST_REMOVE(fp, f_list);
1724         nfiles--;
1725         spin_unlock(&filehead_spin);
1726         fsetcred(fp, NULL);
1727         if (fp->f_nchandle.ncp)
1728             cache_drop(&fp->f_nchandle);
1729         kfree(fp, M_FILE);
1730 }
1731
1732 /*
1733  * called from init_main, initialize filedesc0 for proc0.
1734  */
1735 void
1736 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1737 {
1738         p0->p_fd = fdp0;
1739         p0->p_fdtol = NULL;
1740         fdp0->fd_refcnt = 1;
1741         fdp0->fd_cmask = cmask;
1742         fdp0->fd_files = fdp0->fd_builtin_files;
1743         fdp0->fd_nfiles = NDFILE;
1744         fdp0->fd_lastfile = -1;
1745         spin_init(&fdp0->fd_spin);
1746 }
1747
1748 /*
1749  * Build a new filedesc structure.
1750  *
1751  * NOT MPSAFE (vref)
1752  */
1753 struct filedesc *
1754 fdinit(struct proc *p)
1755 {
1756         struct filedesc *newfdp;
1757         struct filedesc *fdp = p->p_fd;
1758
1759         newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1760         spin_lock(&fdp->fd_spin);
1761         if (fdp->fd_cdir) {
1762                 newfdp->fd_cdir = fdp->fd_cdir;
1763                 vref(newfdp->fd_cdir);
1764                 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1765         }
1766
1767         /*
1768          * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1769          * proc0, but should unconditionally exist in other processes.
1770          */
1771         if (fdp->fd_rdir) {
1772                 newfdp->fd_rdir = fdp->fd_rdir;
1773                 vref(newfdp->fd_rdir);
1774                 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1775         }
1776         if (fdp->fd_jdir) {
1777                 newfdp->fd_jdir = fdp->fd_jdir;
1778                 vref(newfdp->fd_jdir);
1779                 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1780         }
1781         spin_unlock(&fdp->fd_spin);
1782
1783         /* Create the file descriptor table. */
1784         newfdp->fd_refcnt = 1;
1785         newfdp->fd_cmask = cmask;
1786         newfdp->fd_files = newfdp->fd_builtin_files;
1787         newfdp->fd_nfiles = NDFILE;
1788         newfdp->fd_lastfile = -1;
1789         spin_init(&newfdp->fd_spin);
1790
1791         return (newfdp);
1792 }
1793
1794 /*
1795  * Share a filedesc structure.
1796  *
1797  * MPSAFE
1798  */
1799 struct filedesc *
1800 fdshare(struct proc *p)
1801 {
1802         struct filedesc *fdp;
1803
1804         fdp = p->p_fd;
1805         spin_lock(&fdp->fd_spin);
1806         fdp->fd_refcnt++;
1807         spin_unlock(&fdp->fd_spin);
1808         return (fdp);
1809 }
1810
1811 /*
1812  * Copy a filedesc structure.
1813  *
1814  * MPSAFE
1815  */
1816 int
1817 fdcopy(struct proc *p, struct filedesc **fpp)
1818 {
1819         struct filedesc *fdp = p->p_fd;
1820         struct filedesc *newfdp;
1821         struct fdnode *fdnode;
1822         int i;
1823         int ni;
1824
1825         /*
1826          * Certain daemons might not have file descriptors.
1827          */
1828         if (fdp == NULL)
1829                 return (0);
1830
1831         /*
1832          * Allocate the new filedesc and fd_files[] array.  This can race
1833          * with operations by other threads on the fdp so we have to be
1834          * careful.
1835          */
1836         newfdp = kmalloc(sizeof(struct filedesc),
1837                          M_FILEDESC, M_WAITOK | M_ZERO | M_NULLOK);
1838         if (newfdp == NULL) {
1839                 *fpp = NULL;
1840                 return (-1);
1841         }
1842 again:
1843         spin_lock(&fdp->fd_spin);
1844         if (fdp->fd_lastfile < NDFILE) {
1845                 newfdp->fd_files = newfdp->fd_builtin_files;
1846                 i = NDFILE;
1847         } else {
1848                 /*
1849                  * We have to allocate (N^2-1) entries for our in-place
1850                  * binary tree.  Allow the table to shrink.
1851                  */
1852                 i = fdp->fd_nfiles;
1853                 ni = (i - 1) / 2;
1854                 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1855                         i = ni;
1856                         ni = (i - 1) / 2;
1857                 }
1858                 spin_unlock(&fdp->fd_spin);
1859                 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1860                                           M_FILEDESC, M_WAITOK | M_ZERO);
1861
1862                 /*
1863                  * Check for race, retry
1864                  */
1865                 spin_lock(&fdp->fd_spin);
1866                 if (i <= fdp->fd_lastfile) {
1867                         spin_unlock(&fdp->fd_spin);
1868                         kfree(newfdp->fd_files, M_FILEDESC);
1869                         goto again;
1870                 }
1871         }
1872
1873         /*
1874          * Dup the remaining fields. vref() and cache_hold() can be
1875          * safely called while holding the read spinlock on fdp.
1876          *
1877          * The read spinlock on fdp is still being held.
1878          *
1879          * NOTE: vref and cache_hold calls for the case where the vnode
1880          * or cache entry already has at least one ref may be called
1881          * while holding spin locks.
1882          */
1883         if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1884                 vref(newfdp->fd_cdir);
1885                 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1886         }
1887         /*
1888          * We must check for fd_rdir here, at least for now because
1889          * the init process is created before we have access to the
1890          * rootvode to take a reference to it.
1891          */
1892         if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1893                 vref(newfdp->fd_rdir);
1894                 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1895         }
1896         if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1897                 vref(newfdp->fd_jdir);
1898                 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1899         }
1900         newfdp->fd_refcnt = 1;
1901         newfdp->fd_nfiles = i;
1902         newfdp->fd_lastfile = fdp->fd_lastfile;
1903         newfdp->fd_freefile = fdp->fd_freefile;
1904         newfdp->fd_cmask = fdp->fd_cmask;
1905         spin_init(&newfdp->fd_spin);
1906
1907         /*
1908          * Copy the descriptor table through (i).  This also copies the
1909          * allocation state.   Then go through and ref the file pointers
1910          * and clean up any KQ descriptors.
1911          *
1912          * kq descriptors cannot be copied.  Since we haven't ref'd the
1913          * copied files yet we can ignore the return value from funsetfd().
1914          *
1915          * The read spinlock on fdp is still being held.
1916          */
1917         bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1918         for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1919                 fdnode = &newfdp->fd_files[i];
1920                 if (fdnode->reserved) {
1921                         fdreserve_locked(newfdp, i, -1);
1922                         fdnode->reserved = 0;
1923                         fdfixup_locked(newfdp, i);
1924                 } else if (fdnode->fp) {
1925                         if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1926                                 (void)funsetfd_locked(newfdp, i);
1927                         } else {
1928                                 fhold(fdnode->fp);
1929                         }
1930                 }
1931         }
1932         spin_unlock(&fdp->fd_spin);
1933         *fpp = newfdp;
1934         return (0);
1935 }
1936
1937 /*
1938  * Release a filedesc structure.
1939  *
1940  * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1941  */
1942 void
1943 fdfree(struct proc *p, struct filedesc *repl)
1944 {
1945         struct filedesc *fdp;
1946         struct fdnode *fdnode;
1947         int i;
1948         struct filedesc_to_leader *fdtol;
1949         struct file *fp;
1950         struct vnode *vp;
1951         struct flock lf;
1952
1953         /*
1954          * Certain daemons might not have file descriptors.
1955          */
1956         fdp = p->p_fd;
1957         if (fdp == NULL) {
1958                 p->p_fd = repl;
1959                 return;
1960         }
1961
1962         /*
1963          * Severe messing around to follow.
1964          */
1965         spin_lock(&fdp->fd_spin);
1966
1967         /* Check for special need to clear POSIX style locks */
1968         fdtol = p->p_fdtol;
1969         if (fdtol != NULL) {
1970                 KASSERT(fdtol->fdl_refcount > 0,
1971                         ("filedesc_to_refcount botch: fdl_refcount=%d",
1972                          fdtol->fdl_refcount));
1973                 if (fdtol->fdl_refcount == 1 &&
1974                     (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1975                         for (i = 0; i <= fdp->fd_lastfile; ++i) {
1976                                 fdnode = &fdp->fd_files[i];
1977                                 if (fdnode->fp == NULL ||
1978                                     fdnode->fp->f_type != DTYPE_VNODE) {
1979                                         continue;
1980                                 }
1981                                 fp = fdnode->fp;
1982                                 fhold(fp);
1983                                 spin_unlock(&fdp->fd_spin);
1984
1985                                 lf.l_whence = SEEK_SET;
1986                                 lf.l_start = 0;
1987                                 lf.l_len = 0;
1988                                 lf.l_type = F_UNLCK;
1989                                 vp = (struct vnode *)fp->f_data;
1990                                 (void) VOP_ADVLOCK(vp,
1991                                                    (caddr_t)p->p_leader,
1992                                                    F_UNLCK,
1993                                                    &lf,
1994                                                    F_POSIX);
1995                                 fdrop(fp);
1996                                 spin_lock(&fdp->fd_spin);
1997                         }
1998                 }
1999         retry:
2000                 if (fdtol->fdl_refcount == 1) {
2001                         if (fdp->fd_holdleaderscount > 0 &&
2002                             (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2003                                 /*
2004                                  * close() or do_dup() has cleared a reference
2005                                  * in a shared file descriptor table.
2006                                  */
2007                                 fdp->fd_holdleaderswakeup = 1;
2008                                 ssleep(&fdp->fd_holdleaderscount,
2009                                        &fdp->fd_spin, 0, "fdlhold", 0);
2010                                 goto retry;
2011                         }
2012                         if (fdtol->fdl_holdcount > 0) {
2013                                 /*
2014                                  * Ensure that fdtol->fdl_leader
2015                                  * remains valid in closef().
2016                                  */
2017                                 fdtol->fdl_wakeup = 1;
2018                                 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
2019                                 goto retry;
2020                         }
2021                 }
2022                 fdtol->fdl_refcount--;
2023                 if (fdtol->fdl_refcount == 0 &&
2024                     fdtol->fdl_holdcount == 0) {
2025                         fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2026                         fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2027                 } else {
2028                         fdtol = NULL;
2029                 }
2030                 p->p_fdtol = NULL;
2031                 if (fdtol != NULL) {
2032                         spin_unlock(&fdp->fd_spin);
2033                         kfree(fdtol, M_FILEDESC_TO_LEADER);
2034                         spin_lock(&fdp->fd_spin);
2035                 }
2036         }
2037         if (--fdp->fd_refcnt > 0) {
2038                 spin_unlock(&fdp->fd_spin);
2039                 spin_lock(&p->p_spin);
2040                 p->p_fd = repl;
2041                 spin_unlock(&p->p_spin);
2042                 return;
2043         }
2044
2045         /*
2046          * Even though we are the last reference to the structure allproc
2047          * scans may still reference the structure.  Maintain proper
2048          * locks until we can replace p->p_fd.
2049          *
2050          * Also note that kqueue's closef still needs to reference the
2051          * fdp via p->p_fd, so we have to close the descriptors before
2052          * we replace p->p_fd.
2053          */
2054         for (i = 0; i <= fdp->fd_lastfile; ++i) {
2055                 if (fdp->fd_files[i].fp) {
2056                         fp = funsetfd_locked(fdp, i);
2057                         if (fp) {
2058                                 spin_unlock(&fdp->fd_spin);
2059                                 if (SLIST_FIRST(&fp->f_klist))
2060                                         knote_fdclose(fp, fdp, i);
2061                                 closef(fp, p);
2062                                 spin_lock(&fdp->fd_spin);
2063                         }
2064                 }
2065         }
2066         spin_unlock(&fdp->fd_spin);
2067
2068         /*
2069          * Interlock against an allproc scan operations (typically frevoke).
2070          */
2071         spin_lock(&p->p_spin);
2072         p->p_fd = repl;
2073         spin_unlock(&p->p_spin);
2074
2075         /*
2076          * Wait for any softrefs to go away.  This race rarely occurs so
2077          * we can use a non-critical-path style poll/sleep loop.  The
2078          * race only occurs against allproc scans.
2079          *
2080          * No new softrefs can occur with the fdp disconnected from the
2081          * process.
2082          */
2083         if (fdp->fd_softrefs) {
2084                 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2085                 while (fdp->fd_softrefs)
2086                         tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2087         }
2088
2089         if (fdp->fd_files != fdp->fd_builtin_files)
2090                 kfree(fdp->fd_files, M_FILEDESC);
2091         if (fdp->fd_cdir) {
2092                 cache_drop(&fdp->fd_ncdir);
2093                 vrele(fdp->fd_cdir);
2094         }
2095         if (fdp->fd_rdir) {
2096                 cache_drop(&fdp->fd_nrdir);
2097                 vrele(fdp->fd_rdir);
2098         }
2099         if (fdp->fd_jdir) {
2100                 cache_drop(&fdp->fd_njdir);
2101                 vrele(fdp->fd_jdir);
2102         }
2103         kfree(fdp, M_FILEDESC);
2104 }
2105
2106 /*
2107  * Retrieve and reference the file pointer associated with a descriptor.
2108  *
2109  * MPSAFE
2110  */
2111 struct file *
2112 holdfp(struct filedesc *fdp, int fd, int flag)
2113 {
2114         struct file* fp;
2115
2116         spin_lock(&fdp->fd_spin);
2117         if (((u_int)fd) >= fdp->fd_nfiles) {
2118                 fp = NULL;
2119                 goto done;
2120         }
2121         if ((fp = fdp->fd_files[fd].fp) == NULL)
2122                 goto done;
2123         if ((fp->f_flag & flag) == 0 && flag != -1) {
2124                 fp = NULL;
2125                 goto done;
2126         }
2127         fhold(fp);
2128 done:
2129         spin_unlock(&fdp->fd_spin);
2130         return (fp);
2131 }
2132
2133 /*
2134  * holdsock() - load the struct file pointer associated
2135  * with a socket into *fpp.  If an error occurs, non-zero
2136  * will be returned and *fpp will be set to NULL.
2137  *
2138  * MPSAFE
2139  */
2140 int
2141 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2142 {
2143         struct file *fp;
2144         int error;
2145
2146         spin_lock(&fdp->fd_spin);
2147         if ((unsigned)fd >= fdp->fd_nfiles) {
2148                 error = EBADF;
2149                 fp = NULL;
2150                 goto done;
2151         }
2152         if ((fp = fdp->fd_files[fd].fp) == NULL) {
2153                 error = EBADF;
2154                 goto done;
2155         }
2156         if (fp->f_type != DTYPE_SOCKET) {
2157                 error = ENOTSOCK;
2158                 goto done;
2159         }
2160         fhold(fp);
2161         error = 0;
2162 done:
2163         spin_unlock(&fdp->fd_spin);
2164         *fpp = fp;
2165         return (error);
2166 }
2167
2168 /*
2169  * Convert a user file descriptor to a held file pointer.
2170  *
2171  * MPSAFE
2172  */
2173 int
2174 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2175 {
2176         struct file *fp;
2177         int error;
2178
2179         spin_lock(&fdp->fd_spin);
2180         if ((unsigned)fd >= fdp->fd_nfiles) {
2181                 error = EBADF;
2182                 fp = NULL;
2183                 goto done;
2184         }
2185         if ((fp = fdp->fd_files[fd].fp) == NULL) {
2186                 error = EBADF;
2187                 goto done;
2188         }
2189         if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2190                 fp = NULL;
2191                 error = EINVAL;
2192                 goto done;
2193         }
2194         fhold(fp);
2195         error = 0;
2196 done:
2197         spin_unlock(&fdp->fd_spin);
2198         *fpp = fp;
2199         return (error);
2200 }
2201
2202 /*
2203  * For setugid programs, we don't want to people to use that setugidness
2204  * to generate error messages which write to a file which otherwise would
2205  * otherwise be off-limits to the process.
2206  *
2207  * This is a gross hack to plug the hole.  A better solution would involve
2208  * a special vop or other form of generalized access control mechanism.  We
2209  * go ahead and just reject all procfs file systems accesses as dangerous.
2210  *
2211  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2212  * sufficient.  We also don't for check setugidness since we know we are.
2213  */
2214 static int
2215 is_unsafe(struct file *fp)
2216 {
2217         if (fp->f_type == DTYPE_VNODE &&
2218             ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2219                 return (1);
2220         return (0);
2221 }
2222
2223 /*
2224  * Make this setguid thing safe, if at all possible.
2225  *
2226  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2227  */
2228 void
2229 setugidsafety(struct proc *p)
2230 {
2231         struct filedesc *fdp = p->p_fd;
2232         int i;
2233
2234         /* Certain daemons might not have file descriptors. */
2235         if (fdp == NULL)
2236                 return;
2237
2238         /*
2239          * note: fdp->fd_files may be reallocated out from under us while
2240          * we are blocked in a close.  Be careful!
2241          */
2242         for (i = 0; i <= fdp->fd_lastfile; i++) {
2243                 if (i > 2)
2244                         break;
2245                 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2246                         struct file *fp;
2247
2248                         /*
2249                          * NULL-out descriptor prior to close to avoid
2250                          * a race while close blocks.
2251                          */
2252                         if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2253                                 knote_fdclose(fp, fdp, i);
2254                                 closef(fp, p);
2255                         }
2256                 }
2257         }
2258 }
2259
2260 /*
2261  * Close any files on exec?
2262  *
2263  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2264  */
2265 void
2266 fdcloseexec(struct proc *p)
2267 {
2268         struct filedesc *fdp = p->p_fd;
2269         int i;
2270
2271         /* Certain daemons might not have file descriptors. */
2272         if (fdp == NULL)
2273                 return;
2274
2275         /*
2276          * We cannot cache fd_files since operations may block and rip
2277          * them out from under us.
2278          */
2279         for (i = 0; i <= fdp->fd_lastfile; i++) {
2280                 if (fdp->fd_files[i].fp != NULL &&
2281                     (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2282                         struct file *fp;
2283
2284                         /*
2285                          * NULL-out descriptor prior to close to avoid
2286                          * a race while close blocks.
2287                          */
2288                         if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2289                                 knote_fdclose(fp, fdp, i);
2290                                 closef(fp, p);
2291                         }
2292                 }
2293         }
2294 }
2295
2296 /*
2297  * It is unsafe for set[ug]id processes to be started with file
2298  * descriptors 0..2 closed, as these descriptors are given implicit
2299  * significance in the Standard C library.  fdcheckstd() will create a
2300  * descriptor referencing /dev/null for each of stdin, stdout, and
2301  * stderr that is not already open.
2302  *
2303  * NOT MPSAFE - calls falloc, vn_open, etc
2304  */
2305 int
2306 fdcheckstd(struct lwp *lp)
2307 {
2308         struct nlookupdata nd;
2309         struct filedesc *fdp;
2310         struct file *fp;
2311         int retval;
2312         int i, error, flags, devnull;
2313
2314         fdp = lp->lwp_proc->p_fd;
2315         if (fdp == NULL)
2316                 return (0);
2317         devnull = -1;
2318         error = 0;
2319         for (i = 0; i < 3; i++) {
2320                 if (fdp->fd_files[i].fp != NULL)
2321                         continue;
2322                 if (devnull < 0) {
2323                         if ((error = falloc(lp, &fp, &devnull)) != 0)
2324                                 break;
2325
2326                         error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2327                                                 NLC_FOLLOW|NLC_LOCKVP);
2328                         flags = FREAD | FWRITE;
2329                         if (error == 0)
2330                                 error = vn_open(&nd, fp, flags, 0);
2331                         if (error == 0)
2332                                 fsetfd(fdp, fp, devnull);
2333                         else
2334                                 fsetfd(fdp, NULL, devnull);
2335                         fdrop(fp);
2336                         nlookup_done(&nd);
2337                         if (error)
2338                                 break;
2339                         KKASSERT(i == devnull);
2340                 } else {
2341                         error = kern_dup(DUP_FIXED, devnull, i, &retval);
2342                         if (error != 0)
2343                                 break;
2344                 }
2345         }
2346         return (error);
2347 }
2348
2349 /*
2350  * Internal form of close.
2351  * Decrement reference count on file structure.
2352  * Note: td and/or p may be NULL when closing a file
2353  * that was being passed in a message.
2354  *
2355  * MPALMOSTSAFE - acquires mplock for VOP operations
2356  */
2357 int
2358 closef(struct file *fp, struct proc *p)
2359 {
2360         struct vnode *vp;
2361         struct flock lf;
2362         struct filedesc_to_leader *fdtol;
2363
2364         if (fp == NULL)
2365                 return (0);
2366
2367         /*
2368          * POSIX record locking dictates that any close releases ALL
2369          * locks owned by this process.  This is handled by setting
2370          * a flag in the unlock to free ONLY locks obeying POSIX
2371          * semantics, and not to free BSD-style file locks.
2372          * If the descriptor was in a message, POSIX-style locks
2373          * aren't passed with the descriptor.
2374          */
2375         if (p != NULL && fp->f_type == DTYPE_VNODE &&
2376             (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2377         ) {
2378                 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2379                         lf.l_whence = SEEK_SET;
2380                         lf.l_start = 0;
2381                         lf.l_len = 0;
2382                         lf.l_type = F_UNLCK;
2383                         vp = (struct vnode *)fp->f_data;
2384                         (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2385                                            &lf, F_POSIX);
2386                 }
2387                 fdtol = p->p_fdtol;
2388                 if (fdtol != NULL) {
2389                         lwkt_gettoken(&p->p_token);
2390                         /*
2391                          * Handle special case where file descriptor table
2392                          * is shared between multiple process leaders.
2393                          */
2394                         for (fdtol = fdtol->fdl_next;
2395                              fdtol != p->p_fdtol;
2396                              fdtol = fdtol->fdl_next) {
2397                                 if ((fdtol->fdl_leader->p_flag &
2398                                      P_ADVLOCK) == 0)
2399                                         continue;
2400                                 fdtol->fdl_holdcount++;
2401                                 lf.l_whence = SEEK_SET;
2402                                 lf.l_start = 0;
2403                                 lf.l_len = 0;
2404                                 lf.l_type = F_UNLCK;
2405                                 vp = (struct vnode *)fp->f_data;
2406                                 (void) VOP_ADVLOCK(vp,
2407                                                    (caddr_t)fdtol->fdl_leader,
2408                                                    F_UNLCK, &lf, F_POSIX);
2409                                 fdtol->fdl_holdcount--;
2410                                 if (fdtol->fdl_holdcount == 0 &&
2411                                     fdtol->fdl_wakeup != 0) {
2412                                         fdtol->fdl_wakeup = 0;
2413                                         wakeup(fdtol);
2414                                 }
2415                         }
2416                         lwkt_reltoken(&p->p_token);
2417                 }
2418         }
2419         return (fdrop(fp));
2420 }
2421
2422 /*
2423  * MPSAFE
2424  *
2425  * fhold() can only be called if f_count is already at least 1 (i.e. the
2426  * caller of fhold() already has a reference to the file pointer in some
2427  * manner or other).
2428  *
2429  * f_count is not spin-locked.  Instead, atomic ops are used for
2430  * incrementing, decrementing, and handling the 1->0 transition.
2431  */
2432 void
2433 fhold(struct file *fp)
2434 {
2435         atomic_add_int(&fp->f_count, 1);
2436 }
2437
2438 /*
2439  * fdrop() - drop a reference to a descriptor
2440  *
2441  * MPALMOSTSAFE - acquires mplock for final close sequence
2442  */
2443 int
2444 fdrop(struct file *fp)
2445 {
2446         struct flock lf;
2447         struct vnode *vp;
2448         int error;
2449
2450         /*
2451          * A combined fetch and subtract is needed to properly detect
2452          * 1->0 transitions, otherwise two cpus dropping from a ref
2453          * count of 2 might both try to run the 1->0 code.
2454          */
2455         if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2456                 return (0);
2457
2458         KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
2459
2460         /*
2461          * The last reference has gone away, we own the fp structure free
2462          * and clear.
2463          */
2464         if (fp->f_count < 0)
2465                 panic("fdrop: count < 0");
2466         if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2467             (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2468         ) {
2469                 lf.l_whence = SEEK_SET;
2470                 lf.l_start = 0;
2471                 lf.l_len = 0;
2472                 lf.l_type = F_UNLCK;
2473                 vp = (struct vnode *)fp->f_data;
2474                 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2475         }
2476         if (fp->f_ops != &badfileops)
2477                 error = fo_close(fp);
2478         else
2479                 error = 0;
2480         ffree(fp);
2481         return (error);
2482 }
2483
2484 /*
2485  * Apply an advisory lock on a file descriptor.
2486  *
2487  * Just attempt to get a record lock of the requested type on
2488  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2489  *
2490  * MPALMOSTSAFE
2491  */
2492 int
2493 sys_flock(struct flock_args *uap)
2494 {
2495         struct proc *p = curproc;
2496         struct file *fp;
2497         struct vnode *vp;
2498         struct flock lf;
2499         int error;
2500
2501         if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2502                 return (EBADF);
2503         if (fp->f_type != DTYPE_VNODE) {
2504                 error = EOPNOTSUPP;
2505                 goto done;
2506         }
2507         vp = (struct vnode *)fp->f_data;
2508         lf.l_whence = SEEK_SET;
2509         lf.l_start = 0;
2510         lf.l_len = 0;
2511         if (uap->how & LOCK_UN) {
2512                 lf.l_type = F_UNLCK;
2513                 fp->f_flag &= ~FHASLOCK;
2514                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2515                 goto done;
2516         }
2517         if (uap->how & LOCK_EX)
2518                 lf.l_type = F_WRLCK;
2519         else if (uap->how & LOCK_SH)
2520                 lf.l_type = F_RDLCK;
2521         else {
2522                 error = EBADF;
2523                 goto done;
2524         }
2525         fp->f_flag |= FHASLOCK;
2526         if (uap->how & LOCK_NB)
2527                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2528         else
2529                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2530 done:
2531         fdrop(fp);
2532         return (error);
2533 }
2534
2535 /*
2536  * File Descriptor pseudo-device driver (/dev/fd/).
2537  *
2538  * Opening minor device N dup()s the file (if any) connected to file
2539  * descriptor N belonging to the calling process.  Note that this driver
2540  * consists of only the ``open()'' routine, because all subsequent
2541  * references to this file will be direct to the other driver.
2542  */
2543 static int
2544 fdopen(struct dev_open_args *ap)
2545 {
2546         thread_t td = curthread;
2547
2548         KKASSERT(td->td_lwp != NULL);
2549
2550         /*
2551          * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2552          * the file descriptor being sought for duplication. The error
2553          * return ensures that the vnode for this device will be released
2554          * by vn_open. Open will detect this special error and take the
2555          * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2556          * will simply report the error.
2557          */
2558         td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2559         return (ENODEV);
2560 }
2561
2562 /*
2563  * The caller has reserved the file descriptor dfd for us.  On success we
2564  * must fsetfd() it.  On failure the caller will clean it up.
2565  *
2566  * MPSAFE
2567  */
2568 int
2569 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2570 {
2571         struct file *wfp;
2572         struct file *xfp;
2573         int werror;
2574
2575         if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2576                 return (EBADF);
2577
2578         /*
2579          * Close a revoke/dup race.  Duping a descriptor marked as revoked
2580          * will dup a dummy descriptor instead of the real one.
2581          */
2582         if (wfp->f_flag & FREVOKED) {
2583                 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2584                 fdrop(wfp);
2585                 wfp = NULL;
2586                 werror = falloc(NULL, &wfp, NULL);
2587                 if (werror)
2588                         return (werror);
2589         }
2590
2591         /*
2592          * There are two cases of interest here.
2593          *
2594          * For ENODEV simply dup sfd to file descriptor dfd and return.
2595          *
2596          * For ENXIO steal away the file structure from sfd and store it
2597          * dfd.  sfd is effectively closed by this operation.
2598          *
2599          * Any other error code is just returned.
2600          */
2601         switch (error) {
2602         case ENODEV:
2603                 /*
2604                  * Check that the mode the file is being opened for is a
2605                  * subset of the mode of the existing descriptor.
2606                  */
2607                 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2608                         error = EACCES;
2609                         break;
2610                 }
2611                 spin_lock(&fdp->fd_spin);
2612                 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2613                 fsetfd_locked(fdp, wfp, dfd);
2614                 spin_unlock(&fdp->fd_spin);
2615                 error = 0;
2616                 break;
2617         case ENXIO:
2618                 /*
2619                  * Steal away the file pointer from dfd, and stuff it into indx.
2620                  */
2621                 spin_lock(&fdp->fd_spin);
2622                 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2623                 fsetfd(fdp, wfp, dfd);
2624                 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
2625                         spin_unlock(&fdp->fd_spin);
2626                         fdrop(xfp);
2627                 } else {
2628                         spin_unlock(&fdp->fd_spin);
2629                 }
2630                 error = 0;
2631                 break;
2632         default:
2633                 break;
2634         }
2635         fdrop(wfp);
2636         return (error);
2637 }
2638
2639 /*
2640  * NOT MPSAFE - I think these refer to a common file descriptor table
2641  * and we need to spinlock that to link fdtol in.
2642  */
2643 struct filedesc_to_leader *
2644 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2645                          struct proc *leader)
2646 {
2647         struct filedesc_to_leader *fdtol;
2648
2649         fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2650                         M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO);
2651         fdtol->fdl_refcount = 1;
2652         fdtol->fdl_holdcount = 0;
2653         fdtol->fdl_wakeup = 0;
2654         fdtol->fdl_leader = leader;
2655         if (old != NULL) {
2656                 fdtol->fdl_next = old->fdl_next;
2657                 fdtol->fdl_prev = old;
2658                 old->fdl_next = fdtol;
2659                 fdtol->fdl_next->fdl_prev = fdtol;
2660         } else {
2661                 fdtol->fdl_next = fdtol;
2662                 fdtol->fdl_prev = fdtol;
2663         }
2664         return fdtol;
2665 }
2666
2667 /*
2668  * Scan all file pointers in the system.  The callback is made with
2669  * the master list spinlock held exclusively.
2670  *
2671  * MPSAFE
2672  */
2673 void
2674 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2675 {
2676         struct file *fp;
2677         int res;
2678
2679         spin_lock(&filehead_spin);
2680         LIST_FOREACH(fp, &filehead, f_list) {
2681                 res = callback(fp, data);
2682                 if (res < 0)
2683                         break;
2684         }
2685         spin_unlock(&filehead_spin);
2686 }
2687
2688 /*
2689  * Get file structures.
2690  *
2691  * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2692  */
2693
2694 struct sysctl_kern_file_info {
2695         int count;
2696         int error;
2697         struct sysctl_req *req;
2698 };
2699
2700 static int sysctl_kern_file_callback(struct proc *p, void *data);
2701
2702 static int
2703 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2704 {
2705         struct sysctl_kern_file_info info;
2706
2707         /*
2708          * Note: because the number of file descriptors is calculated
2709          * in different ways for sizing vs returning the data,
2710          * there is information leakage from the first loop.  However,
2711          * it is of a similar order of magnitude to the leakage from
2712          * global system statistics such as kern.openfiles.
2713          *
2714          * When just doing a count, note that we cannot just count
2715          * the elements and add f_count via the filehead list because
2716          * threaded processes share their descriptor table and f_count might
2717          * still be '1' in that case.
2718          *
2719          * Since the SYSCTL op can block, we must hold the process to
2720          * prevent it being ripped out from under us either in the
2721          * file descriptor loop or in the greater LIST_FOREACH.  The
2722          * process may be in varying states of disrepair.  If the process
2723          * is in SZOMB we may have caught it just as it is being removed
2724          * from the allproc list, we must skip it in that case to maintain
2725          * an unbroken chain through the allproc list.
2726          */
2727         info.count = 0;
2728         info.error = 0;
2729         info.req = req;
2730         allproc_scan(sysctl_kern_file_callback, &info);
2731
2732         /*
2733          * When just calculating the size, overestimate a bit to try to
2734          * prevent system activity from causing the buffer-fill call
2735          * to fail later on.
2736          */
2737         if (req->oldptr == NULL) {
2738                 info.count = (info.count + 16) + (info.count / 10);
2739                 info.error = SYSCTL_OUT(req, NULL,
2740                                         info.count * sizeof(struct kinfo_file));
2741         }
2742         return (info.error);
2743 }
2744
2745 static int
2746 sysctl_kern_file_callback(struct proc *p, void *data)
2747 {
2748         struct sysctl_kern_file_info *info = data;
2749         struct kinfo_file kf;
2750         struct filedesc *fdp;
2751         struct file *fp;
2752         uid_t uid;
2753         int n;
2754
2755         if (p->p_stat == SIDL || p->p_stat == SZOMB)
2756                 return(0);
2757         if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2758                 return(0);
2759
2760         /*
2761          * Softref the fdp to prevent it from being destroyed
2762          */
2763         spin_lock(&p->p_spin);
2764         if ((fdp = p->p_fd) == NULL) {
2765                 spin_unlock(&p->p_spin);
2766                 return(0);
2767         }
2768         atomic_add_int(&fdp->fd_softrefs, 1);
2769         spin_unlock(&p->p_spin);
2770
2771         /*
2772          * The fdp's own spinlock prevents the contents from being
2773          * modified.
2774          */
2775         spin_lock(&fdp->fd_spin);
2776         for (n = 0; n < fdp->fd_nfiles; ++n) {
2777                 if ((fp = fdp->fd_files[n].fp) == NULL)
2778                         continue;
2779                 if (info->req->oldptr == NULL) {
2780                         ++info->count;
2781                 } else {
2782                         uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2783                         kcore_make_file(&kf, fp, p->p_pid, uid, n);
2784                         spin_unlock(&fdp->fd_spin);
2785                         info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2786                         spin_lock(&fdp->fd_spin);
2787                         if (info->error)
2788                                 break;
2789                 }
2790         }
2791         spin_unlock(&fdp->fd_spin);
2792         atomic_subtract_int(&fdp->fd_softrefs, 1);
2793         if (info->error)
2794                 return(-1);
2795         return(0);
2796 }
2797
2798 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2799     0, 0, sysctl_kern_file, "S,file", "Entire file table");
2800
2801 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2802     &minfilesperproc, 0, "Minimum files allowed open per process");
2803 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2804     &maxfilesperproc, 0, "Maximum files allowed open per process");
2805 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2806     &maxfilesperuser, 0, "Maximum files allowed open per user");
2807
2808 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2809     &maxfiles, 0, "Maximum number of files");
2810
2811 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2812     &maxfilesrootres, 0, "Descriptors reserved for root use");
2813
2814 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2815         &nfiles, 0, "System-wide number of open files");
2816
2817 static void
2818 fildesc_drvinit(void *unused)
2819 {
2820         int fd;
2821
2822         for (fd = 0; fd < NUMFDESC; fd++) {
2823                 make_dev(&fildesc_ops, fd,
2824                          UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2825         }
2826
2827         make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2828         make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2829         make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2830 }
2831
2832 /*
2833  * MPSAFE
2834  */
2835 struct fileops badfileops = {
2836         .fo_read = badfo_readwrite,
2837         .fo_write = badfo_readwrite,
2838         .fo_ioctl = badfo_ioctl,
2839         .fo_kqfilter = badfo_kqfilter,
2840         .fo_stat = badfo_stat,
2841         .fo_close = badfo_close,
2842         .fo_shutdown = badfo_shutdown
2843 };
2844
2845 int
2846 badfo_readwrite(
2847         struct file *fp,
2848         struct uio *uio,
2849         struct ucred *cred,
2850         int flags
2851 ) {
2852         return (EBADF);
2853 }
2854
2855 int
2856 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2857             struct ucred *cred, struct sysmsg *msgv)
2858 {
2859         return (EBADF);
2860 }
2861
2862 /*
2863  * Must return an error to prevent registration, typically
2864  * due to a revoked descriptor (file_filtops assigned).
2865  */
2866 int
2867 badfo_kqfilter(struct file *fp, struct knote *kn)
2868 {
2869         return (EOPNOTSUPP);
2870 }
2871
2872 /*
2873  * MPSAFE
2874  */
2875 int
2876 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2877 {
2878         return (EBADF);
2879 }
2880
2881 /*
2882  * MPSAFE
2883  */
2884 int
2885 badfo_close(struct file *fp)
2886 {
2887         return (EBADF);
2888 }
2889
2890 /*
2891  * MPSAFE
2892  */
2893 int
2894 badfo_shutdown(struct file *fp, int how)
2895 {
2896         return (EBADF);
2897 }
2898
2899 /*
2900  * MPSAFE
2901  */
2902 int
2903 nofo_shutdown(struct file *fp, int how)
2904 {
2905         return (EOPNOTSUPP);
2906 }
2907
2908 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2909                                         fildesc_drvinit,NULL)