sys/kern/kern_descrip.c

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