The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_descrip.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  * (c) UNIX System Laboratories, Inc.
    5  * All or some portions of this file are derived from material licensed
    6  * to the University of California by American Telephone and Telegraph
    7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
    8  * the permission of UNIX System Laboratories, Inc.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 4. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)kern_descrip.c      8.6 (Berkeley) 4/19/94
   35  */
   36 
   37 #include <sys/cdefs.h>
   38 __FBSDID("$FreeBSD: releng/8.3/sys/kern/kern_descrip.c 229506 2012-01-04 16:51:04Z jhb $");
   39 
   40 #include "opt_compat.h"
   41 #include "opt_ddb.h"
   42 #include "opt_ktrace.h"
   43 
   44 #include <sys/param.h>
   45 #include <sys/systm.h>
   46 
   47 #include <sys/conf.h>
   48 #include <sys/domain.h>
   49 #include <sys/fcntl.h>
   50 #include <sys/file.h>
   51 #include <sys/filedesc.h>
   52 #include <sys/filio.h>
   53 #include <sys/jail.h>
   54 #include <sys/kernel.h>
   55 #include <sys/limits.h>
   56 #include <sys/lock.h>
   57 #include <sys/malloc.h>
   58 #include <sys/mount.h>
   59 #include <sys/mqueue.h>
   60 #include <sys/mutex.h>
   61 #include <sys/namei.h>
   62 #include <sys/priv.h>
   63 #include <sys/proc.h>
   64 #include <sys/protosw.h>
   65 #include <sys/resourcevar.h>
   66 #include <sys/signalvar.h>
   67 #include <sys/socketvar.h>
   68 #include <sys/stat.h>
   69 #include <sys/sx.h>
   70 #include <sys/syscallsubr.h>
   71 #include <sys/sysctl.h>
   72 #include <sys/sysproto.h>
   73 #include <sys/tty.h>
   74 #include <sys/unistd.h>
   75 #include <sys/user.h>
   76 #include <sys/vnode.h>
   77 #ifdef KTRACE
   78 #include <sys/ktrace.h>
   79 #endif
   80 
   81 #include <net/vnet.h>
   82 
   83 #include <security/audit/audit.h>
   84 
   85 #include <vm/uma.h>
   86 
   87 #include <ddb/ddb.h>
   88 
   89 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
   90 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
   91                      "file desc to leader structures");
   92 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
   93 
   94 static uma_zone_t file_zone;
   95 
   96 
   97 /* Flags for do_dup() */
   98 #define DUP_FIXED       0x1     /* Force fixed allocation */
   99 #define DUP_FCNTL       0x2     /* fcntl()-style errors */
  100 
  101 static int do_dup(struct thread *td, int flags, int old, int new,
  102     register_t *retval);
  103 static int      fd_first_free(struct filedesc *, int, int);
  104 static int      fd_last_used(struct filedesc *, int, int);
  105 static void     fdgrowtable(struct filedesc *, int);
  106 static void     fdunused(struct filedesc *fdp, int fd);
  107 static void     fdused(struct filedesc *fdp, int fd);
  108 
  109 /*
  110  * A process is initially started out with NDFILE descriptors stored within
  111  * this structure, selected to be enough for typical applications based on
  112  * the historical limit of 20 open files (and the usage of descriptors by
  113  * shells).  If these descriptors are exhausted, a larger descriptor table
  114  * may be allocated, up to a process' resource limit; the internal arrays
  115  * are then unused.
  116  */
  117 #define NDFILE          20
  118 #define NDSLOTSIZE      sizeof(NDSLOTTYPE)
  119 #define NDENTRIES       (NDSLOTSIZE * __CHAR_BIT)
  120 #define NDSLOT(x)       ((x) / NDENTRIES)
  121 #define NDBIT(x)        ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
  122 #define NDSLOTS(x)      (((x) + NDENTRIES - 1) / NDENTRIES)
  123 
  124 /*
  125  * Storage required per open file descriptor.
  126  */
  127 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
  128 
  129 /*
  130  * Storage to hold unused ofiles that need to be reclaimed.
  131  */
  132 struct freetable {
  133         struct file     **ft_table;
  134         SLIST_ENTRY(freetable) ft_next;
  135 };
  136 
  137 /*
  138  * Basic allocation of descriptors:
  139  * one of the above, plus arrays for NDFILE descriptors.
  140  */
  141 struct filedesc0 {
  142         struct  filedesc fd_fd;
  143         /*
  144          * ofiles which need to be reclaimed on free.
  145          */
  146         SLIST_HEAD(,freetable) fd_free;
  147         /*
  148          * These arrays are used when the number of open files is
  149          * <= NDFILE, and are then pointed to by the pointers above.
  150          */
  151         struct  file *fd_dfiles[NDFILE];
  152         char    fd_dfileflags[NDFILE];
  153         NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
  154 };
  155 
  156 /*
  157  * Descriptor management.
  158  */
  159 volatile int openfiles;                 /* actual number of open files */
  160 struct mtx sigio_lock;          /* mtx to protect pointers to sigio */
  161 void    (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
  162 
  163 /* A mutex to protect the association between a proc and filedesc. */
  164 static struct mtx       fdesc_mtx;
  165 
  166 /*
  167  * Find the first zero bit in the given bitmap, starting at low and not
  168  * exceeding size - 1.
  169  */
  170 static int
  171 fd_first_free(struct filedesc *fdp, int low, int size)
  172 {
  173         NDSLOTTYPE *map = fdp->fd_map;
  174         NDSLOTTYPE mask;
  175         int off, maxoff;
  176 
  177         if (low >= size)
  178                 return (low);
  179 
  180         off = NDSLOT(low);
  181         if (low % NDENTRIES) {
  182                 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
  183                 if ((mask &= ~map[off]) != 0UL)
  184                         return (off * NDENTRIES + ffsl(mask) - 1);
  185                 ++off;
  186         }
  187         for (maxoff = NDSLOTS(size); off < maxoff; ++off)
  188                 if (map[off] != ~0UL)
  189                         return (off * NDENTRIES + ffsl(~map[off]) - 1);
  190         return (size);
  191 }
  192 
  193 /*
  194  * Find the highest non-zero bit in the given bitmap, starting at low and
  195  * not exceeding size - 1.
  196  */
  197 static int
  198 fd_last_used(struct filedesc *fdp, int low, int size)
  199 {
  200         NDSLOTTYPE *map = fdp->fd_map;
  201         NDSLOTTYPE mask;
  202         int off, minoff;
  203 
  204         if (low >= size)
  205                 return (-1);
  206 
  207         off = NDSLOT(size);
  208         if (size % NDENTRIES) {
  209                 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
  210                 if ((mask &= map[off]) != 0)
  211                         return (off * NDENTRIES + flsl(mask) - 1);
  212                 --off;
  213         }
  214         for (minoff = NDSLOT(low); off >= minoff; --off)
  215                 if (map[off] != 0)
  216                         return (off * NDENTRIES + flsl(map[off]) - 1);
  217         return (low - 1);
  218 }
  219 
  220 static int
  221 fdisused(struct filedesc *fdp, int fd)
  222 {
  223         KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
  224             ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
  225         return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
  226 }
  227 
  228 /*
  229  * Mark a file descriptor as used.
  230  */
  231 static void
  232 fdused(struct filedesc *fdp, int fd)
  233 {
  234 
  235         FILEDESC_XLOCK_ASSERT(fdp);
  236         KASSERT(!fdisused(fdp, fd),
  237             ("fd already used"));
  238 
  239         fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
  240         if (fd > fdp->fd_lastfile)
  241                 fdp->fd_lastfile = fd;
  242         if (fd == fdp->fd_freefile)
  243                 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
  244 }
  245 
  246 /*
  247  * Mark a file descriptor as unused.
  248  */
  249 static void
  250 fdunused(struct filedesc *fdp, int fd)
  251 {
  252 
  253         FILEDESC_XLOCK_ASSERT(fdp);
  254         KASSERT(fdisused(fdp, fd),
  255             ("fd is already unused"));
  256         KASSERT(fdp->fd_ofiles[fd] == NULL,
  257             ("fd is still in use"));
  258 
  259         fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
  260         if (fd < fdp->fd_freefile)
  261                 fdp->fd_freefile = fd;
  262         if (fd == fdp->fd_lastfile)
  263                 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
  264 }
  265 
  266 /*
  267  * System calls on descriptors.
  268  */
  269 #ifndef _SYS_SYSPROTO_H_
  270 struct getdtablesize_args {
  271         int     dummy;
  272 };
  273 #endif
  274 /* ARGSUSED */
  275 int
  276 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
  277 {
  278         struct proc *p = td->td_proc;
  279 
  280         PROC_LOCK(p);
  281         td->td_retval[0] =
  282             min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
  283         PROC_UNLOCK(p);
  284         return (0);
  285 }
  286 
  287 /*
  288  * Duplicate a file descriptor to a particular value.
  289  *
  290  * Note: keep in mind that a potential race condition exists when closing
  291  * descriptors from a shared descriptor table (via rfork).
  292  */
  293 #ifndef _SYS_SYSPROTO_H_
  294 struct dup2_args {
  295         u_int   from;
  296         u_int   to;
  297 };
  298 #endif
  299 /* ARGSUSED */
  300 int
  301 dup2(struct thread *td, struct dup2_args *uap)
  302 {
  303 
  304         return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
  305                     td->td_retval));
  306 }
  307 
  308 /*
  309  * Duplicate a file descriptor.
  310  */
  311 #ifndef _SYS_SYSPROTO_H_
  312 struct dup_args {
  313         u_int   fd;
  314 };
  315 #endif
  316 /* ARGSUSED */
  317 int
  318 dup(struct thread *td, struct dup_args *uap)
  319 {
  320 
  321         return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
  322 }
  323 
  324 /*
  325  * The file control system call.
  326  */
  327 #ifndef _SYS_SYSPROTO_H_
  328 struct fcntl_args {
  329         int     fd;
  330         int     cmd;
  331         long    arg;
  332 };
  333 #endif
  334 /* ARGSUSED */
  335 int
  336 fcntl(struct thread *td, struct fcntl_args *uap)
  337 {
  338         struct flock fl;
  339         struct oflock ofl;
  340         intptr_t arg;
  341         int error;
  342         int cmd;
  343 
  344         error = 0;
  345         cmd = uap->cmd;
  346         switch (uap->cmd) {
  347         case F_OGETLK:
  348         case F_OSETLK:
  349         case F_OSETLKW:
  350                 /*
  351                  * Convert old flock structure to new.
  352                  */
  353                 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
  354                 fl.l_start = ofl.l_start;
  355                 fl.l_len = ofl.l_len;
  356                 fl.l_pid = ofl.l_pid;
  357                 fl.l_type = ofl.l_type;
  358                 fl.l_whence = ofl.l_whence;
  359                 fl.l_sysid = 0;
  360 
  361                 switch (uap->cmd) {
  362                 case F_OGETLK:
  363                     cmd = F_GETLK;
  364                     break;
  365                 case F_OSETLK:
  366                     cmd = F_SETLK;
  367                     break;
  368                 case F_OSETLKW:
  369                     cmd = F_SETLKW;
  370                     break;
  371                 }
  372                 arg = (intptr_t)&fl;
  373                 break;
  374         case F_GETLK:
  375         case F_SETLK:
  376         case F_SETLKW:
  377         case F_SETLK_REMOTE:
  378                 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
  379                 arg = (intptr_t)&fl;
  380                 break;
  381         default:
  382                 arg = uap->arg;
  383                 break;
  384         }
  385         if (error)
  386                 return (error);
  387         error = kern_fcntl(td, uap->fd, cmd, arg);
  388         if (error)
  389                 return (error);
  390         if (uap->cmd == F_OGETLK) {
  391                 ofl.l_start = fl.l_start;
  392                 ofl.l_len = fl.l_len;
  393                 ofl.l_pid = fl.l_pid;
  394                 ofl.l_type = fl.l_type;
  395                 ofl.l_whence = fl.l_whence;
  396                 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
  397         } else if (uap->cmd == F_GETLK) {
  398                 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
  399         }
  400         return (error);
  401 }
  402 
  403 static inline struct file *
  404 fdtofp(int fd, struct filedesc *fdp)
  405 {
  406         struct file *fp;
  407 
  408         FILEDESC_LOCK_ASSERT(fdp);
  409         if ((unsigned)fd >= fdp->fd_nfiles ||
  410             (fp = fdp->fd_ofiles[fd]) == NULL)
  411                 return (NULL);
  412         return (fp);
  413 }
  414 
  415 int
  416 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
  417 {
  418         struct filedesc *fdp;
  419         struct flock *flp;
  420         struct file *fp;
  421         struct proc *p;
  422         char *pop;
  423         struct vnode *vp;
  424         int error, flg, tmp;
  425         int vfslocked;
  426         u_int old, new;
  427         uint64_t bsize;
  428 
  429         vfslocked = 0;
  430         error = 0;
  431         flg = F_POSIX;
  432         p = td->td_proc;
  433         fdp = p->p_fd;
  434 
  435         switch (cmd) {
  436         case F_DUPFD:
  437                 tmp = arg;
  438                 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
  439                 break;
  440 
  441         case F_DUP2FD:
  442                 tmp = arg;
  443                 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
  444                 break;
  445 
  446         case F_GETFD:
  447                 FILEDESC_SLOCK(fdp);
  448                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  449                         FILEDESC_SUNLOCK(fdp);
  450                         error = EBADF;
  451                         break;
  452                 }
  453                 pop = &fdp->fd_ofileflags[fd];
  454                 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
  455                 FILEDESC_SUNLOCK(fdp);
  456                 break;
  457 
  458         case F_SETFD:
  459                 FILEDESC_XLOCK(fdp);
  460                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  461                         FILEDESC_XUNLOCK(fdp);
  462                         error = EBADF;
  463                         break;
  464                 }
  465                 pop = &fdp->fd_ofileflags[fd];
  466                 *pop = (*pop &~ UF_EXCLOSE) |
  467                     (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
  468                 FILEDESC_XUNLOCK(fdp);
  469                 break;
  470 
  471         case F_GETFL:
  472                 FILEDESC_SLOCK(fdp);
  473                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  474                         FILEDESC_SUNLOCK(fdp);
  475                         error = EBADF;
  476                         break;
  477                 }
  478                 td->td_retval[0] = OFLAGS(fp->f_flag);
  479                 FILEDESC_SUNLOCK(fdp);
  480                 break;
  481 
  482         case F_SETFL:
  483                 FILEDESC_SLOCK(fdp);
  484                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  485                         FILEDESC_SUNLOCK(fdp);
  486                         error = EBADF;
  487                         break;
  488                 }
  489                 fhold(fp);
  490                 FILEDESC_SUNLOCK(fdp);
  491                 do {
  492                         tmp = flg = fp->f_flag;
  493                         tmp &= ~FCNTLFLAGS;
  494                         tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
  495                 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
  496                 tmp = fp->f_flag & FNONBLOCK;
  497                 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
  498                 if (error) {
  499                         fdrop(fp, td);
  500                         break;
  501                 }
  502                 tmp = fp->f_flag & FASYNC;
  503                 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
  504                 if (error == 0) {
  505                         fdrop(fp, td);
  506                         break;
  507                 }
  508                 atomic_clear_int(&fp->f_flag, FNONBLOCK);
  509                 tmp = 0;
  510                 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
  511                 fdrop(fp, td);
  512                 break;
  513 
  514         case F_GETOWN:
  515                 FILEDESC_SLOCK(fdp);
  516                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  517                         FILEDESC_SUNLOCK(fdp);
  518                         error = EBADF;
  519                         break;
  520                 }
  521                 fhold(fp);
  522                 FILEDESC_SUNLOCK(fdp);
  523                 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
  524                 if (error == 0)
  525                         td->td_retval[0] = tmp;
  526                 fdrop(fp, td);
  527                 break;
  528 
  529         case F_SETOWN:
  530                 FILEDESC_SLOCK(fdp);
  531                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  532                         FILEDESC_SUNLOCK(fdp);
  533                         error = EBADF;
  534                         break;
  535                 }
  536                 fhold(fp);
  537                 FILEDESC_SUNLOCK(fdp);
  538                 tmp = arg;
  539                 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
  540                 fdrop(fp, td);
  541                 break;
  542 
  543         case F_SETLK_REMOTE:
  544                 error = priv_check(td, PRIV_NFS_LOCKD);
  545                 if (error)
  546                         return (error);
  547                 flg = F_REMOTE;
  548                 goto do_setlk;
  549 
  550         case F_SETLKW:
  551                 flg |= F_WAIT;
  552                 /* FALLTHROUGH F_SETLK */
  553 
  554         case F_SETLK:
  555         do_setlk:
  556                 FILEDESC_SLOCK(fdp);
  557                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  558                         FILEDESC_SUNLOCK(fdp);
  559                         error = EBADF;
  560                         break;
  561                 }
  562                 if (fp->f_type != DTYPE_VNODE) {
  563                         FILEDESC_SUNLOCK(fdp);
  564                         error = EBADF;
  565                         break;
  566                 }
  567                 flp = (struct flock *)arg;
  568                 if (flp->l_whence == SEEK_CUR) {
  569                         if (fp->f_offset < 0 ||
  570                             (flp->l_start > 0 &&
  571                              fp->f_offset > OFF_MAX - flp->l_start)) {
  572                                 FILEDESC_SUNLOCK(fdp);
  573                                 error = EOVERFLOW;
  574                                 break;
  575                         }
  576                         flp->l_start += fp->f_offset;
  577                 }
  578 
  579                 /*
  580                  * VOP_ADVLOCK() may block.
  581                  */
  582                 fhold(fp);
  583                 FILEDESC_SUNLOCK(fdp);
  584                 vp = fp->f_vnode;
  585                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  586                 switch (flp->l_type) {
  587                 case F_RDLCK:
  588                         if ((fp->f_flag & FREAD) == 0) {
  589                                 error = EBADF;
  590                                 break;
  591                         }
  592                         PROC_LOCK(p->p_leader);
  593                         p->p_leader->p_flag |= P_ADVLOCK;
  594                         PROC_UNLOCK(p->p_leader);
  595                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
  596                             flp, flg);
  597                         break;
  598                 case F_WRLCK:
  599                         if ((fp->f_flag & FWRITE) == 0) {
  600                                 error = EBADF;
  601                                 break;
  602                         }
  603                         PROC_LOCK(p->p_leader);
  604                         p->p_leader->p_flag |= P_ADVLOCK;
  605                         PROC_UNLOCK(p->p_leader);
  606                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
  607                             flp, flg);
  608                         break;
  609                 case F_UNLCK:
  610                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
  611                             flp, flg);
  612                         break;
  613                 case F_UNLCKSYS:
  614                         /*
  615                          * Temporary api for testing remote lock
  616                          * infrastructure.
  617                          */
  618                         if (flg != F_REMOTE) {
  619                                 error = EINVAL;
  620                                 break;
  621                         }
  622                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
  623                             F_UNLCKSYS, flp, flg);
  624                         break;
  625                 default:
  626                         error = EINVAL;
  627                         break;
  628                 }
  629                 VFS_UNLOCK_GIANT(vfslocked);
  630                 vfslocked = 0;
  631                 /* Check for race with close */
  632                 FILEDESC_SLOCK(fdp);
  633                 if ((unsigned) fd >= fdp->fd_nfiles ||
  634                     fp != fdp->fd_ofiles[fd]) {
  635                         FILEDESC_SUNLOCK(fdp);
  636                         flp->l_whence = SEEK_SET;
  637                         flp->l_start = 0;
  638                         flp->l_len = 0;
  639                         flp->l_type = F_UNLCK;
  640                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  641                         (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
  642                                            F_UNLCK, flp, F_POSIX);
  643                         VFS_UNLOCK_GIANT(vfslocked);
  644                         vfslocked = 0;
  645                 } else
  646                         FILEDESC_SUNLOCK(fdp);
  647                 fdrop(fp, td);
  648                 break;
  649 
  650         case F_GETLK:
  651                 FILEDESC_SLOCK(fdp);
  652                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  653                         FILEDESC_SUNLOCK(fdp);
  654                         error = EBADF;
  655                         break;
  656                 }
  657                 if (fp->f_type != DTYPE_VNODE) {
  658                         FILEDESC_SUNLOCK(fdp);
  659                         error = EBADF;
  660                         break;
  661                 }
  662                 flp = (struct flock *)arg;
  663                 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
  664                     flp->l_type != F_UNLCK) {
  665                         FILEDESC_SUNLOCK(fdp);
  666                         error = EINVAL;
  667                         break;
  668                 }
  669                 if (flp->l_whence == SEEK_CUR) {
  670                         if ((flp->l_start > 0 &&
  671                             fp->f_offset > OFF_MAX - flp->l_start) ||
  672                             (flp->l_start < 0 &&
  673                              fp->f_offset < OFF_MIN - flp->l_start)) {
  674                                 FILEDESC_SUNLOCK(fdp);
  675                                 error = EOVERFLOW;
  676                                 break;
  677                         }
  678                         flp->l_start += fp->f_offset;
  679                 }
  680                 /*
  681                  * VOP_ADVLOCK() may block.
  682                  */
  683                 fhold(fp);
  684                 FILEDESC_SUNLOCK(fdp);
  685                 vp = fp->f_vnode;
  686                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  687                 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
  688                     F_POSIX);
  689                 VFS_UNLOCK_GIANT(vfslocked);
  690                 vfslocked = 0;
  691                 fdrop(fp, td);
  692                 break;
  693 
  694         case F_RDAHEAD:
  695                 arg = arg ? 128 * 1024: 0;
  696                 /* FALLTHROUGH */
  697         case F_READAHEAD:
  698                 FILEDESC_SLOCK(fdp);
  699                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  700                         FILEDESC_SUNLOCK(fdp);
  701                         error = EBADF;
  702                         break;
  703                 }
  704                 if (fp->f_type != DTYPE_VNODE) {
  705                         FILEDESC_SUNLOCK(fdp);
  706                         error = EBADF;
  707                         break;
  708                 }
  709                 fhold(fp);
  710                 FILEDESC_SUNLOCK(fdp);
  711                 if (arg != 0) {
  712                         vp = fp->f_vnode;
  713                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  714                         error = vn_lock(vp, LK_SHARED);
  715                         if (error != 0)
  716                                 goto readahead_vnlock_fail;
  717                         bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
  718                         VOP_UNLOCK(vp, 0);
  719                         fp->f_seqcount = (arg + bsize - 1) / bsize;
  720                         do {
  721                                 new = old = fp->f_flag;
  722                                 new |= FRDAHEAD;
  723                         } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
  724 readahead_vnlock_fail:
  725                         VFS_UNLOCK_GIANT(vfslocked);
  726                         vfslocked = 0;
  727                 } else {
  728                         do {
  729                                 new = old = fp->f_flag;
  730                                 new &= ~FRDAHEAD;
  731                         } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
  732                 }
  733                 fdrop(fp, td);
  734                 break;
  735 
  736         default:
  737                 error = EINVAL;
  738                 break;
  739         }
  740         VFS_UNLOCK_GIANT(vfslocked);
  741         return (error);
  742 }
  743 
  744 /*
  745  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
  746  */
  747 static int
  748 do_dup(struct thread *td, int flags, int old, int new,
  749     register_t *retval)
  750 {
  751         struct filedesc *fdp;
  752         struct proc *p;
  753         struct file *fp;
  754         struct file *delfp;
  755         int error, holdleaders, maxfd;
  756 
  757         p = td->td_proc;
  758         fdp = p->p_fd;
  759 
  760         /*
  761          * Verify we have a valid descriptor to dup from and possibly to
  762          * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
  763          * return EINVAL when the new descriptor is out of bounds.
  764          */
  765         if (old < 0)
  766                 return (EBADF);
  767         if (new < 0)
  768                 return (flags & DUP_FCNTL ? EINVAL : EBADF);
  769         PROC_LOCK(p);
  770         maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
  771         PROC_UNLOCK(p);
  772         if (new >= maxfd)
  773                 return (flags & DUP_FCNTL ? EINVAL : EMFILE);
  774 
  775         FILEDESC_XLOCK(fdp);
  776         if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
  777                 FILEDESC_XUNLOCK(fdp);
  778                 return (EBADF);
  779         }
  780         if (flags & DUP_FIXED && old == new) {
  781                 *retval = new;
  782                 FILEDESC_XUNLOCK(fdp);
  783                 return (0);
  784         }
  785         fp = fdp->fd_ofiles[old];
  786         fhold(fp);
  787 
  788         /*
  789          * If the caller specified a file descriptor, make sure the file
  790          * table is large enough to hold it, and grab it.  Otherwise, just
  791          * allocate a new descriptor the usual way.  Since the filedesc
  792          * lock may be temporarily dropped in the process, we have to look
  793          * out for a race.
  794          */
  795         if (flags & DUP_FIXED) {
  796                 if (new >= fdp->fd_nfiles)
  797                         fdgrowtable(fdp, new + 1);
  798                 if (fdp->fd_ofiles[new] == NULL)
  799                         fdused(fdp, new);
  800         } else {
  801                 if ((error = fdalloc(td, new, &new)) != 0) {
  802                         FILEDESC_XUNLOCK(fdp);
  803                         fdrop(fp, td);
  804                         return (error);
  805                 }
  806         }
  807 
  808         /*
  809          * If the old file changed out from under us then treat it as a
  810          * bad file descriptor.  Userland should do its own locking to
  811          * avoid this case.
  812          */
  813         if (fdp->fd_ofiles[old] != fp) {
  814                 /* we've allocated a descriptor which we won't use */
  815                 if (fdp->fd_ofiles[new] == NULL)
  816                         fdunused(fdp, new);
  817                 FILEDESC_XUNLOCK(fdp);
  818                 fdrop(fp, td);
  819                 return (EBADF);
  820         }
  821         KASSERT(old != new,
  822             ("new fd is same as old"));
  823 
  824         /*
  825          * Save info on the descriptor being overwritten.  We cannot close
  826          * it without introducing an ownership race for the slot, since we
  827          * need to drop the filedesc lock to call closef().
  828          *
  829          * XXX this duplicates parts of close().
  830          */
  831         delfp = fdp->fd_ofiles[new];
  832         holdleaders = 0;
  833         if (delfp != NULL) {
  834                 if (td->td_proc->p_fdtol != NULL) {
  835                         /*
  836                          * Ask fdfree() to sleep to ensure that all relevant
  837                          * process leaders can be traversed in closef().
  838                          */
  839                         fdp->fd_holdleaderscount++;
  840                         holdleaders = 1;
  841                 }
  842         }
  843 
  844         /*
  845          * Duplicate the source descriptor
  846          */
  847         fdp->fd_ofiles[new] = fp;
  848         fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
  849         if (new > fdp->fd_lastfile)
  850                 fdp->fd_lastfile = new;
  851         *retval = new;
  852 
  853         /*
  854          * If we dup'd over a valid file, we now own the reference to it
  855          * and must dispose of it using closef() semantics (as if a
  856          * close() were performed on it).
  857          *
  858          * XXX this duplicates parts of close().
  859          */
  860         if (delfp != NULL) {
  861                 knote_fdclose(td, new);
  862                 if (delfp->f_type == DTYPE_MQUEUE)
  863                         mq_fdclose(td, new, delfp);
  864                 FILEDESC_XUNLOCK(fdp);
  865                 (void) closef(delfp, td);
  866                 if (holdleaders) {
  867                         FILEDESC_XLOCK(fdp);
  868                         fdp->fd_holdleaderscount--;
  869                         if (fdp->fd_holdleaderscount == 0 &&
  870                             fdp->fd_holdleaderswakeup != 0) {
  871                                 fdp->fd_holdleaderswakeup = 0;
  872                                 wakeup(&fdp->fd_holdleaderscount);
  873                         }
  874                         FILEDESC_XUNLOCK(fdp);
  875                 }
  876         } else {
  877                 FILEDESC_XUNLOCK(fdp);
  878         }
  879         return (0);
  880 }
  881 
  882 /*
  883  * If sigio is on the list associated with a process or process group,
  884  * disable signalling from the device, remove sigio from the list and
  885  * free sigio.
  886  */
  887 void
  888 funsetown(struct sigio **sigiop)
  889 {
  890         struct sigio *sigio;
  891 
  892         SIGIO_LOCK();
  893         sigio = *sigiop;
  894         if (sigio == NULL) {
  895                 SIGIO_UNLOCK();
  896                 return;
  897         }
  898         *(sigio->sio_myref) = NULL;
  899         if ((sigio)->sio_pgid < 0) {
  900                 struct pgrp *pg = (sigio)->sio_pgrp;
  901                 PGRP_LOCK(pg);
  902                 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
  903                              sigio, sio_pgsigio);
  904                 PGRP_UNLOCK(pg);
  905         } else {
  906                 struct proc *p = (sigio)->sio_proc;
  907                 PROC_LOCK(p);
  908                 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
  909                              sigio, sio_pgsigio);
  910                 PROC_UNLOCK(p);
  911         }
  912         SIGIO_UNLOCK();
  913         crfree(sigio->sio_ucred);
  914         free(sigio, M_SIGIO);
  915 }
  916 
  917 /*
  918  * Free a list of sigio structures.
  919  * We only need to lock the SIGIO_LOCK because we have made ourselves
  920  * inaccessible to callers of fsetown and therefore do not need to lock
  921  * the proc or pgrp struct for the list manipulation.
  922  */
  923 void
  924 funsetownlst(struct sigiolst *sigiolst)
  925 {
  926         struct proc *p;
  927         struct pgrp *pg;
  928         struct sigio *sigio;
  929 
  930         sigio = SLIST_FIRST(sigiolst);
  931         if (sigio == NULL)
  932                 return;
  933         p = NULL;
  934         pg = NULL;
  935 
  936         /*
  937          * Every entry of the list should belong
  938          * to a single proc or pgrp.
  939          */
  940         if (sigio->sio_pgid < 0) {
  941                 pg = sigio->sio_pgrp;
  942                 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
  943         } else /* if (sigio->sio_pgid > 0) */ {
  944                 p = sigio->sio_proc;
  945                 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
  946         }
  947 
  948         SIGIO_LOCK();
  949         while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
  950                 *(sigio->sio_myref) = NULL;
  951                 if (pg != NULL) {
  952                         KASSERT(sigio->sio_pgid < 0,
  953                             ("Proc sigio in pgrp sigio list"));
  954                         KASSERT(sigio->sio_pgrp == pg,
  955                             ("Bogus pgrp in sigio list"));
  956                         PGRP_LOCK(pg);
  957                         SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
  958                             sio_pgsigio);
  959                         PGRP_UNLOCK(pg);
  960                 } else /* if (p != NULL) */ {
  961                         KASSERT(sigio->sio_pgid > 0,
  962                             ("Pgrp sigio in proc sigio list"));
  963                         KASSERT(sigio->sio_proc == p,
  964                             ("Bogus proc in sigio list"));
  965                         PROC_LOCK(p);
  966                         SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
  967                             sio_pgsigio);
  968                         PROC_UNLOCK(p);
  969                 }
  970                 SIGIO_UNLOCK();
  971                 crfree(sigio->sio_ucred);
  972                 free(sigio, M_SIGIO);
  973                 SIGIO_LOCK();
  974         }
  975         SIGIO_UNLOCK();
  976 }
  977 
  978 /*
  979  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
  980  *
  981  * After permission checking, add a sigio structure to the sigio list for
  982  * the process or process group.
  983  */
  984 int
  985 fsetown(pid_t pgid, struct sigio **sigiop)
  986 {
  987         struct proc *proc;
  988         struct pgrp *pgrp;
  989         struct sigio *sigio;
  990         int ret;
  991 
  992         if (pgid == 0) {
  993                 funsetown(sigiop);
  994                 return (0);
  995         }
  996 
  997         ret = 0;
  998 
  999         /* Allocate and fill in the new sigio out of locks. */
 1000         sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
 1001         sigio->sio_pgid = pgid;
 1002         sigio->sio_ucred = crhold(curthread->td_ucred);
 1003         sigio->sio_myref = sigiop;
 1004 
 1005         sx_slock(&proctree_lock);
 1006         if (pgid > 0) {
 1007                 proc = pfind(pgid);
 1008                 if (proc == NULL) {
 1009                         ret = ESRCH;
 1010                         goto fail;
 1011                 }
 1012 
 1013                 /*
 1014                  * Policy - Don't allow a process to FSETOWN a process
 1015                  * in another session.
 1016                  *
 1017                  * Remove this test to allow maximum flexibility or
 1018                  * restrict FSETOWN to the current process or process
 1019                  * group for maximum safety.
 1020                  */
 1021                 PROC_UNLOCK(proc);
 1022                 if (proc->p_session != curthread->td_proc->p_session) {
 1023                         ret = EPERM;
 1024                         goto fail;
 1025                 }
 1026 
 1027                 pgrp = NULL;
 1028         } else /* if (pgid < 0) */ {
 1029                 pgrp = pgfind(-pgid);
 1030                 if (pgrp == NULL) {
 1031                         ret = ESRCH;
 1032                         goto fail;
 1033                 }
 1034                 PGRP_UNLOCK(pgrp);
 1035 
 1036                 /*
 1037                  * Policy - Don't allow a process to FSETOWN a process
 1038                  * in another session.
 1039                  *
 1040                  * Remove this test to allow maximum flexibility or
 1041                  * restrict FSETOWN to the current process or process
 1042                  * group for maximum safety.
 1043                  */
 1044                 if (pgrp->pg_session != curthread->td_proc->p_session) {
 1045                         ret = EPERM;
 1046                         goto fail;
 1047                 }
 1048 
 1049                 proc = NULL;
 1050         }
 1051         funsetown(sigiop);
 1052         if (pgid > 0) {
 1053                 PROC_LOCK(proc);
 1054                 /*
 1055                  * Since funsetownlst() is called without the proctree
 1056                  * locked, we need to check for P_WEXIT.
 1057                  * XXX: is ESRCH correct?
 1058                  */
 1059                 if ((proc->p_flag & P_WEXIT) != 0) {
 1060                         PROC_UNLOCK(proc);
 1061                         ret = ESRCH;
 1062                         goto fail;
 1063                 }
 1064                 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
 1065                 sigio->sio_proc = proc;
 1066                 PROC_UNLOCK(proc);
 1067         } else {
 1068                 PGRP_LOCK(pgrp);
 1069                 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
 1070                 sigio->sio_pgrp = pgrp;
 1071                 PGRP_UNLOCK(pgrp);
 1072         }
 1073         sx_sunlock(&proctree_lock);
 1074         SIGIO_LOCK();
 1075         *sigiop = sigio;
 1076         SIGIO_UNLOCK();
 1077         return (0);
 1078 
 1079 fail:
 1080         sx_sunlock(&proctree_lock);
 1081         crfree(sigio->sio_ucred);
 1082         free(sigio, M_SIGIO);
 1083         return (ret);
 1084 }
 1085 
 1086 /*
 1087  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
 1088  */
 1089 pid_t
 1090 fgetown(sigiop)
 1091         struct sigio **sigiop;
 1092 {
 1093         pid_t pgid;
 1094 
 1095         SIGIO_LOCK();
 1096         pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
 1097         SIGIO_UNLOCK();
 1098         return (pgid);
 1099 }
 1100 
 1101 /*
 1102  * Close a file descriptor.
 1103  */
 1104 #ifndef _SYS_SYSPROTO_H_
 1105 struct close_args {
 1106         int     fd;
 1107 };
 1108 #endif
 1109 /* ARGSUSED */
 1110 int
 1111 close(td, uap)
 1112         struct thread *td;
 1113         struct close_args *uap;
 1114 {
 1115 
 1116         return (kern_close(td, uap->fd));
 1117 }
 1118 
 1119 int
 1120 kern_close(td, fd)
 1121         struct thread *td;
 1122         int fd;
 1123 {
 1124         struct filedesc *fdp;
 1125         struct file *fp;
 1126         int error;
 1127         int holdleaders;
 1128 
 1129         error = 0;
 1130         holdleaders = 0;
 1131         fdp = td->td_proc->p_fd;
 1132 
 1133         AUDIT_SYSCLOSE(td, fd);
 1134 
 1135         FILEDESC_XLOCK(fdp);
 1136         if ((unsigned)fd >= fdp->fd_nfiles ||
 1137             (fp = fdp->fd_ofiles[fd]) == NULL) {
 1138                 FILEDESC_XUNLOCK(fdp);
 1139                 return (EBADF);
 1140         }
 1141         fdp->fd_ofiles[fd] = NULL;
 1142         fdp->fd_ofileflags[fd] = 0;
 1143         fdunused(fdp, fd);
 1144         if (td->td_proc->p_fdtol != NULL) {
 1145                 /*
 1146                  * Ask fdfree() to sleep to ensure that all relevant
 1147                  * process leaders can be traversed in closef().
 1148                  */
 1149                 fdp->fd_holdleaderscount++;
 1150                 holdleaders = 1;
 1151         }
 1152 
 1153         /*
 1154          * We now hold the fp reference that used to be owned by the
 1155          * descriptor array.  We have to unlock the FILEDESC *AFTER*
 1156          * knote_fdclose to prevent a race of the fd getting opened, a knote
 1157          * added, and deleteing a knote for the new fd.
 1158          */
 1159         knote_fdclose(td, fd);
 1160         if (fp->f_type == DTYPE_MQUEUE)
 1161                 mq_fdclose(td, fd, fp);
 1162         FILEDESC_XUNLOCK(fdp);
 1163 
 1164         error = closef(fp, td);
 1165         if (holdleaders) {
 1166                 FILEDESC_XLOCK(fdp);
 1167                 fdp->fd_holdleaderscount--;
 1168                 if (fdp->fd_holdleaderscount == 0 &&
 1169                     fdp->fd_holdleaderswakeup != 0) {
 1170                         fdp->fd_holdleaderswakeup = 0;
 1171                         wakeup(&fdp->fd_holdleaderscount);
 1172                 }
 1173                 FILEDESC_XUNLOCK(fdp);
 1174         }
 1175         return (error);
 1176 }
 1177 
 1178 /*
 1179  * Close open file descriptors.
 1180  */
 1181 #ifndef _SYS_SYSPROTO_H_
 1182 struct closefrom_args {
 1183         int     lowfd;
 1184 };
 1185 #endif
 1186 /* ARGSUSED */
 1187 int
 1188 closefrom(struct thread *td, struct closefrom_args *uap)
 1189 {
 1190         struct filedesc *fdp;
 1191         int fd;
 1192 
 1193         fdp = td->td_proc->p_fd;
 1194         AUDIT_ARG_FD(uap->lowfd);
 1195 
 1196         /*
 1197          * Treat negative starting file descriptor values identical to
 1198          * closefrom(0) which closes all files.
 1199          */
 1200         if (uap->lowfd < 0)
 1201                 uap->lowfd = 0;
 1202         FILEDESC_SLOCK(fdp);
 1203         for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) {
 1204                 if (fdp->fd_ofiles[fd] != NULL) {
 1205                         FILEDESC_SUNLOCK(fdp);
 1206                         (void)kern_close(td, fd);
 1207                         FILEDESC_SLOCK(fdp);
 1208                 }
 1209         }
 1210         FILEDESC_SUNLOCK(fdp);
 1211         return (0);
 1212 }
 1213 
 1214 #if defined(COMPAT_43)
 1215 /*
 1216  * Return status information about a file descriptor.
 1217  */
 1218 #ifndef _SYS_SYSPROTO_H_
 1219 struct ofstat_args {
 1220         int     fd;
 1221         struct  ostat *sb;
 1222 };
 1223 #endif
 1224 /* ARGSUSED */
 1225 int
 1226 ofstat(struct thread *td, struct ofstat_args *uap)
 1227 {
 1228         struct ostat oub;
 1229         struct stat ub;
 1230         int error;
 1231 
 1232         error = kern_fstat(td, uap->fd, &ub);
 1233         if (error == 0) {
 1234                 cvtstat(&ub, &oub);
 1235                 error = copyout(&oub, uap->sb, sizeof(oub));
 1236         }
 1237         return (error);
 1238 }
 1239 #endif /* COMPAT_43 */
 1240 
 1241 /*
 1242  * Return status information about a file descriptor.
 1243  */
 1244 #ifndef _SYS_SYSPROTO_H_
 1245 struct fstat_args {
 1246         int     fd;
 1247         struct  stat *sb;
 1248 };
 1249 #endif
 1250 /* ARGSUSED */
 1251 int
 1252 fstat(struct thread *td, struct fstat_args *uap)
 1253 {
 1254         struct stat ub;
 1255         int error;
 1256 
 1257         error = kern_fstat(td, uap->fd, &ub);
 1258         if (error == 0)
 1259                 error = copyout(&ub, uap->sb, sizeof(ub));
 1260         return (error);
 1261 }
 1262 
 1263 int
 1264 kern_fstat(struct thread *td, int fd, struct stat *sbp)
 1265 {
 1266         struct file *fp;
 1267         int error;
 1268 
 1269         AUDIT_ARG_FD(fd);
 1270 
 1271         if ((error = fget(td, fd, &fp)) != 0)
 1272                 return (error);
 1273 
 1274         AUDIT_ARG_FILE(td->td_proc, fp);
 1275 
 1276         error = fo_stat(fp, sbp, td->td_ucred, td);
 1277         fdrop(fp, td);
 1278 #ifdef KTRACE
 1279         if (error == 0 && KTRPOINT(td, KTR_STRUCT))
 1280                 ktrstat(sbp);
 1281 #endif
 1282         return (error);
 1283 }
 1284 
 1285 /*
 1286  * Return status information about a file descriptor.
 1287  */
 1288 #ifndef _SYS_SYSPROTO_H_
 1289 struct nfstat_args {
 1290         int     fd;
 1291         struct  nstat *sb;
 1292 };
 1293 #endif
 1294 /* ARGSUSED */
 1295 int
 1296 nfstat(struct thread *td, struct nfstat_args *uap)
 1297 {
 1298         struct nstat nub;
 1299         struct stat ub;
 1300         int error;
 1301 
 1302         error = kern_fstat(td, uap->fd, &ub);
 1303         if (error == 0) {
 1304                 cvtnstat(&ub, &nub);
 1305                 error = copyout(&nub, uap->sb, sizeof(nub));
 1306         }
 1307         return (error);
 1308 }
 1309 
 1310 /*
 1311  * Return pathconf information about a file descriptor.
 1312  */
 1313 #ifndef _SYS_SYSPROTO_H_
 1314 struct fpathconf_args {
 1315         int     fd;
 1316         int     name;
 1317 };
 1318 #endif
 1319 /* ARGSUSED */
 1320 int
 1321 fpathconf(struct thread *td, struct fpathconf_args *uap)
 1322 {
 1323         struct file *fp;
 1324         struct vnode *vp;
 1325         int error;
 1326 
 1327         if ((error = fget(td, uap->fd, &fp)) != 0)
 1328                 return (error);
 1329 
 1330         /* If asynchronous I/O is available, it works for all descriptors. */
 1331         if (uap->name == _PC_ASYNC_IO) {
 1332                 td->td_retval[0] = async_io_version;
 1333                 goto out;
 1334         }
 1335         vp = fp->f_vnode;
 1336         if (vp != NULL) {
 1337                 int vfslocked;
 1338                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 1339                 vn_lock(vp, LK_SHARED | LK_RETRY);
 1340                 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
 1341                 VOP_UNLOCK(vp, 0);
 1342                 VFS_UNLOCK_GIANT(vfslocked);
 1343         } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
 1344                 if (uap->name != _PC_PIPE_BUF) {
 1345                         error = EINVAL;
 1346                 } else {
 1347                         td->td_retval[0] = PIPE_BUF;
 1348                 error = 0;
 1349                 }
 1350         } else {
 1351                 error = EOPNOTSUPP;
 1352         }
 1353 out:
 1354         fdrop(fp, td);
 1355         return (error);
 1356 }
 1357 
 1358 /*
 1359  * Grow the file table to accomodate (at least) nfd descriptors.  This may
 1360  * block and drop the filedesc lock, but it will reacquire it before
 1361  * returning.
 1362  */
 1363 static void
 1364 fdgrowtable(struct filedesc *fdp, int nfd)
 1365 {
 1366         struct filedesc0 *fdp0;
 1367         struct freetable *fo;
 1368         struct file **ntable;
 1369         struct file **otable;
 1370         char *nfileflags;
 1371         int nnfiles, onfiles;
 1372         NDSLOTTYPE *nmap;
 1373 
 1374         FILEDESC_XLOCK_ASSERT(fdp);
 1375 
 1376         KASSERT(fdp->fd_nfiles > 0,
 1377             ("zero-length file table"));
 1378 
 1379         /* compute the size of the new table */
 1380         onfiles = fdp->fd_nfiles;
 1381         nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
 1382         if (nnfiles <= onfiles)
 1383                 /* the table is already large enough */
 1384                 return;
 1385 
 1386         /* allocate a new table and (if required) new bitmaps */
 1387         FILEDESC_XUNLOCK(fdp);
 1388         ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable),
 1389             M_FILEDESC, M_ZERO | M_WAITOK);
 1390         nfileflags = (char *)&ntable[nnfiles];
 1391         if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
 1392                 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE,
 1393                     M_FILEDESC, M_ZERO | M_WAITOK);
 1394         else
 1395                 nmap = NULL;
 1396         FILEDESC_XLOCK(fdp);
 1397 
 1398         /*
 1399          * We now have new tables ready to go.  Since we dropped the
 1400          * filedesc lock to call malloc(), watch out for a race.
 1401          */
 1402         onfiles = fdp->fd_nfiles;
 1403         if (onfiles >= nnfiles) {
 1404                 /* we lost the race, but that's OK */
 1405                 free(ntable, M_FILEDESC);
 1406                 if (nmap != NULL)
 1407                         free(nmap, M_FILEDESC);
 1408                 return;
 1409         }
 1410         bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
 1411         bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
 1412         otable = fdp->fd_ofiles;
 1413         fdp->fd_ofileflags = nfileflags;
 1414         fdp->fd_ofiles = ntable;
 1415         /*
 1416          * We must preserve ofiles until the process exits because we can't
 1417          * be certain that no threads have references to the old table via
 1418          * _fget().
 1419          */
 1420         if (onfiles > NDFILE) {
 1421                 fo = (struct freetable *)&otable[onfiles];
 1422                 fdp0 = (struct filedesc0 *)fdp;
 1423                 fo->ft_table = otable;
 1424                 SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next);
 1425         }
 1426         if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
 1427                 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
 1428                 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
 1429                         free(fdp->fd_map, M_FILEDESC);
 1430                 fdp->fd_map = nmap;
 1431         }
 1432         fdp->fd_nfiles = nnfiles;
 1433 }
 1434 
 1435 /*
 1436  * Allocate a file descriptor for the process.
 1437  */
 1438 int
 1439 fdalloc(struct thread *td, int minfd, int *result)
 1440 {
 1441         struct proc *p = td->td_proc;
 1442         struct filedesc *fdp = p->p_fd;
 1443         int fd = -1, maxfd;
 1444 
 1445         FILEDESC_XLOCK_ASSERT(fdp);
 1446 
 1447         if (fdp->fd_freefile > minfd)
 1448                 minfd = fdp->fd_freefile;          
 1449 
 1450         PROC_LOCK(p);
 1451         maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
 1452         PROC_UNLOCK(p);
 1453 
 1454         /*
 1455          * Search the bitmap for a free descriptor.  If none is found, try
 1456          * to grow the file table.  Keep at it until we either get a file
 1457          * descriptor or run into process or system limits; fdgrowtable()
 1458          * may drop the filedesc lock, so we're in a race.
 1459          */
 1460         for (;;) {
 1461                 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
 1462                 if (fd >= maxfd)
 1463                         return (EMFILE);
 1464                 if (fd < fdp->fd_nfiles)
 1465                         break;
 1466                 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
 1467         }
 1468 
 1469         /*
 1470          * Perform some sanity checks, then mark the file descriptor as
 1471          * used and return it to the caller.
 1472          */
 1473         KASSERT(!fdisused(fdp, fd),
 1474             ("fd_first_free() returned non-free descriptor"));
 1475         KASSERT(fdp->fd_ofiles[fd] == NULL,
 1476             ("free descriptor isn't"));
 1477         fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
 1478         fdused(fdp, fd);
 1479         *result = fd;
 1480         return (0);
 1481 }
 1482 
 1483 /*
 1484  * Check to see whether n user file descriptors are available to the process
 1485  * p.
 1486  */
 1487 int
 1488 fdavail(struct thread *td, int n)
 1489 {
 1490         struct proc *p = td->td_proc;
 1491         struct filedesc *fdp = td->td_proc->p_fd;
 1492         struct file **fpp;
 1493         int i, lim, last;
 1494 
 1495         FILEDESC_LOCK_ASSERT(fdp);
 1496 
 1497         PROC_LOCK(p);
 1498         lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
 1499         PROC_UNLOCK(p);
 1500         if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
 1501                 return (1);
 1502         last = min(fdp->fd_nfiles, lim);
 1503         fpp = &fdp->fd_ofiles[fdp->fd_freefile];
 1504         for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
 1505                 if (*fpp == NULL && --n <= 0)
 1506                         return (1);
 1507         }
 1508         return (0);
 1509 }
 1510 
 1511 /*
 1512  * Create a new open file structure and allocate a file decriptor for the
 1513  * process that refers to it.  We add one reference to the file for the
 1514  * descriptor table and one reference for resultfp. This is to prevent us
 1515  * being preempted and the entry in the descriptor table closed after we
 1516  * release the FILEDESC lock.
 1517  */
 1518 int
 1519 fallocf(struct thread *td, struct file **resultfp, int *resultfd, int flags)
 1520 {
 1521         struct proc *p = td->td_proc;
 1522         struct file *fp;
 1523         int error, i;
 1524         int maxuserfiles = maxfiles - (maxfiles / 20);
 1525         static struct timeval lastfail;
 1526         static int curfail;
 1527 
 1528         fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
 1529         if ((openfiles >= maxuserfiles &&
 1530             priv_check(td, PRIV_MAXFILES) != 0) ||
 1531             openfiles >= maxfiles) {
 1532                 if (ppsratecheck(&lastfail, &curfail, 1)) {
 1533                         printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
 1534                                 td->td_ucred->cr_ruid);
 1535                 }
 1536                 uma_zfree(file_zone, fp);
 1537                 return (ENFILE);
 1538         }
 1539         atomic_add_int(&openfiles, 1);
 1540 
 1541         /*
 1542          * If the process has file descriptor zero open, add the new file
 1543          * descriptor to the list of open files at that point, otherwise
 1544          * put it at the front of the list of open files.
 1545          */
 1546         refcount_init(&fp->f_count, 1);
 1547         if (resultfp)
 1548                 fhold(fp);
 1549         fp->f_cred = crhold(td->td_ucred);
 1550         fp->f_ops = &badfileops;
 1551         fp->f_data = NULL;
 1552         fp->f_vnode = NULL;
 1553         FILEDESC_XLOCK(p->p_fd);
 1554         if ((error = fdalloc(td, 0, &i))) {
 1555                 FILEDESC_XUNLOCK(p->p_fd);
 1556                 fdrop(fp, td);
 1557                 if (resultfp)
 1558                         fdrop(fp, td);
 1559                 return (error);
 1560         }
 1561         p->p_fd->fd_ofiles[i] = fp;
 1562         if ((flags & O_CLOEXEC) != 0)
 1563                 p->p_fd->fd_ofileflags[i] |= UF_EXCLOSE;
 1564         FILEDESC_XUNLOCK(p->p_fd);
 1565         if (resultfp)
 1566                 *resultfp = fp;
 1567         if (resultfd)
 1568                 *resultfd = i;
 1569         return (0);
 1570 }
 1571 
 1572 int
 1573 falloc(struct thread *td, struct file **resultfp, int *resultfd)
 1574 {
 1575 
 1576         return (fallocf(td, resultfp, resultfd, 0));
 1577 }
 1578 
 1579 /*
 1580  * Build a new filedesc structure from another.
 1581  * Copy the current, root, and jail root vnode references.
 1582  */
 1583 struct filedesc *
 1584 fdinit(struct filedesc *fdp)
 1585 {
 1586         struct filedesc0 *newfdp;
 1587 
 1588         newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
 1589         FILEDESC_LOCK_INIT(&newfdp->fd_fd);
 1590         if (fdp != NULL) {
 1591                 FILEDESC_XLOCK(fdp);
 1592                 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
 1593                 if (newfdp->fd_fd.fd_cdir)
 1594                         VREF(newfdp->fd_fd.fd_cdir);
 1595                 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
 1596                 if (newfdp->fd_fd.fd_rdir)
 1597                         VREF(newfdp->fd_fd.fd_rdir);
 1598                 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
 1599                 if (newfdp->fd_fd.fd_jdir)
 1600                         VREF(newfdp->fd_fd.fd_jdir);
 1601                 FILEDESC_XUNLOCK(fdp);
 1602         }
 1603 
 1604         /* Create the file descriptor table. */
 1605         newfdp->fd_fd.fd_refcnt = 1;
 1606         newfdp->fd_fd.fd_holdcnt = 1;
 1607         newfdp->fd_fd.fd_cmask = CMASK;
 1608         newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
 1609         newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
 1610         newfdp->fd_fd.fd_nfiles = NDFILE;
 1611         newfdp->fd_fd.fd_map = newfdp->fd_dmap;
 1612         newfdp->fd_fd.fd_lastfile = -1;
 1613         return (&newfdp->fd_fd);
 1614 }
 1615 
 1616 static struct filedesc *
 1617 fdhold(struct proc *p)
 1618 {
 1619         struct filedesc *fdp;
 1620 
 1621         mtx_lock(&fdesc_mtx);
 1622         fdp = p->p_fd;
 1623         if (fdp != NULL)
 1624                 fdp->fd_holdcnt++;
 1625         mtx_unlock(&fdesc_mtx);
 1626         return (fdp);
 1627 }
 1628 
 1629 static void
 1630 fddrop(struct filedesc *fdp)
 1631 {
 1632         struct filedesc0 *fdp0;
 1633         struct freetable *ft;
 1634         int i;
 1635 
 1636         mtx_lock(&fdesc_mtx);
 1637         i = --fdp->fd_holdcnt;
 1638         mtx_unlock(&fdesc_mtx);
 1639         if (i > 0)
 1640                 return;
 1641 
 1642         FILEDESC_LOCK_DESTROY(fdp);
 1643         fdp0 = (struct filedesc0 *)fdp;
 1644         while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
 1645                 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
 1646                 free(ft->ft_table, M_FILEDESC);
 1647         }
 1648         free(fdp, M_FILEDESC);
 1649 }
 1650 
 1651 /*
 1652  * Share a filedesc structure.
 1653  */
 1654 struct filedesc *
 1655 fdshare(struct filedesc *fdp)
 1656 {
 1657 
 1658         FILEDESC_XLOCK(fdp);
 1659         fdp->fd_refcnt++;
 1660         FILEDESC_XUNLOCK(fdp);
 1661         return (fdp);
 1662 }
 1663 
 1664 /*
 1665  * Unshare a filedesc structure, if necessary by making a copy
 1666  */
 1667 void
 1668 fdunshare(struct proc *p, struct thread *td)
 1669 {
 1670 
 1671         FILEDESC_XLOCK(p->p_fd);
 1672         if (p->p_fd->fd_refcnt > 1) {
 1673                 struct filedesc *tmp;
 1674 
 1675                 FILEDESC_XUNLOCK(p->p_fd);
 1676                 tmp = fdcopy(p->p_fd);
 1677                 fdfree(td);
 1678                 p->p_fd = tmp;
 1679         } else
 1680                 FILEDESC_XUNLOCK(p->p_fd);
 1681 }
 1682 
 1683 /*
 1684  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
 1685  * this is to ease callers, not catch errors.
 1686  */
 1687 struct filedesc *
 1688 fdcopy(struct filedesc *fdp)
 1689 {
 1690         struct filedesc *newfdp;
 1691         int i;
 1692 
 1693         /* Certain daemons might not have file descriptors. */
 1694         if (fdp == NULL)
 1695                 return (NULL);
 1696 
 1697         newfdp = fdinit(fdp);
 1698         FILEDESC_SLOCK(fdp);
 1699         while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
 1700                 FILEDESC_SUNLOCK(fdp);
 1701                 FILEDESC_XLOCK(newfdp);
 1702                 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
 1703                 FILEDESC_XUNLOCK(newfdp);
 1704                 FILEDESC_SLOCK(fdp);
 1705         }
 1706         /* copy everything except kqueue descriptors */
 1707         newfdp->fd_freefile = -1;
 1708         for (i = 0; i <= fdp->fd_lastfile; ++i) {
 1709                 if (fdisused(fdp, i) &&
 1710                     fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE &&
 1711                     fdp->fd_ofiles[i]->f_ops != &badfileops) {
 1712                         newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
 1713                         newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
 1714                         fhold(newfdp->fd_ofiles[i]);
 1715                         newfdp->fd_lastfile = i;
 1716                 } else {
 1717                         if (newfdp->fd_freefile == -1)
 1718                                 newfdp->fd_freefile = i;
 1719                 }
 1720         }
 1721         newfdp->fd_cmask = fdp->fd_cmask;
 1722         FILEDESC_SUNLOCK(fdp);
 1723         FILEDESC_XLOCK(newfdp);
 1724         for (i = 0; i <= newfdp->fd_lastfile; ++i)
 1725                 if (newfdp->fd_ofiles[i] != NULL)
 1726                         fdused(newfdp, i);
 1727         if (newfdp->fd_freefile == -1)
 1728                 newfdp->fd_freefile = i;
 1729         FILEDESC_XUNLOCK(newfdp);
 1730         return (newfdp);
 1731 }
 1732 
 1733 /*
 1734  * Release a filedesc structure.
 1735  */
 1736 void
 1737 fdfree(struct thread *td)
 1738 {
 1739         struct filedesc *fdp;
 1740         struct file **fpp;
 1741         int i, locked;
 1742         struct filedesc_to_leader *fdtol;
 1743         struct file *fp;
 1744         struct vnode *cdir, *jdir, *rdir, *vp;
 1745         struct flock lf;
 1746 
 1747         /* Certain daemons might not have file descriptors. */
 1748         fdp = td->td_proc->p_fd;
 1749         if (fdp == NULL)
 1750                 return;
 1751 
 1752         /* Check for special need to clear POSIX style locks */
 1753         fdtol = td->td_proc->p_fdtol;
 1754         if (fdtol != NULL) {
 1755                 FILEDESC_XLOCK(fdp);
 1756                 KASSERT(fdtol->fdl_refcount > 0,
 1757                         ("filedesc_to_refcount botch: fdl_refcount=%d",
 1758                          fdtol->fdl_refcount));
 1759                 if (fdtol->fdl_refcount == 1 &&
 1760                     (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 1761                         for (i = 0, fpp = fdp->fd_ofiles;
 1762                              i <= fdp->fd_lastfile;
 1763                              i++, fpp++) {
 1764                                 if (*fpp == NULL ||
 1765                                     (*fpp)->f_type != DTYPE_VNODE)
 1766                                         continue;
 1767                                 fp = *fpp;
 1768                                 fhold(fp);
 1769                                 FILEDESC_XUNLOCK(fdp);
 1770                                 lf.l_whence = SEEK_SET;
 1771                                 lf.l_start = 0;
 1772                                 lf.l_len = 0;
 1773                                 lf.l_type = F_UNLCK;
 1774                                 vp = fp->f_vnode;
 1775                                 locked = VFS_LOCK_GIANT(vp->v_mount);
 1776                                 (void) VOP_ADVLOCK(vp,
 1777                                                    (caddr_t)td->td_proc->
 1778                                                    p_leader,
 1779                                                    F_UNLCK,
 1780                                                    &lf,
 1781                                                    F_POSIX);
 1782                                 VFS_UNLOCK_GIANT(locked);
 1783                                 FILEDESC_XLOCK(fdp);
 1784                                 fdrop(fp, td);
 1785                                 fpp = fdp->fd_ofiles + i;
 1786                         }
 1787                 }
 1788         retry:
 1789                 if (fdtol->fdl_refcount == 1) {
 1790                         if (fdp->fd_holdleaderscount > 0 &&
 1791                             (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 1792                                 /*
 1793                                  * close() or do_dup() has cleared a reference
 1794                                  * in a shared file descriptor table.
 1795                                  */
 1796                                 fdp->fd_holdleaderswakeup = 1;
 1797                                 sx_sleep(&fdp->fd_holdleaderscount,
 1798                                     FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
 1799                                 goto retry;
 1800                         }
 1801                         if (fdtol->fdl_holdcount > 0) {
 1802                                 /*
 1803                                  * Ensure that fdtol->fdl_leader remains
 1804                                  * valid in closef().
 1805                                  */
 1806                                 fdtol->fdl_wakeup = 1;
 1807                                 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
 1808                                     "fdlhold", 0);
 1809                                 goto retry;
 1810                         }
 1811                 }
 1812                 fdtol->fdl_refcount--;
 1813                 if (fdtol->fdl_refcount == 0 &&
 1814                     fdtol->fdl_holdcount == 0) {
 1815                         fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
 1816                         fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
 1817                 } else
 1818                         fdtol = NULL;
 1819                 td->td_proc->p_fdtol = NULL;
 1820                 FILEDESC_XUNLOCK(fdp);
 1821                 if (fdtol != NULL)
 1822                         free(fdtol, M_FILEDESC_TO_LEADER);
 1823         }
 1824         FILEDESC_XLOCK(fdp);
 1825         i = --fdp->fd_refcnt;
 1826         FILEDESC_XUNLOCK(fdp);
 1827         if (i > 0)
 1828                 return;
 1829 
 1830         fpp = fdp->fd_ofiles;
 1831         for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
 1832                 if (*fpp) {
 1833                         FILEDESC_XLOCK(fdp);
 1834                         fp = *fpp;
 1835                         *fpp = NULL;
 1836                         FILEDESC_XUNLOCK(fdp);
 1837                         (void) closef(fp, td);
 1838                 }
 1839         }
 1840         FILEDESC_XLOCK(fdp);
 1841 
 1842         /* XXX This should happen earlier. */
 1843         mtx_lock(&fdesc_mtx);
 1844         td->td_proc->p_fd = NULL;
 1845         mtx_unlock(&fdesc_mtx);
 1846 
 1847         if (fdp->fd_nfiles > NDFILE)
 1848                 free(fdp->fd_ofiles, M_FILEDESC);
 1849         if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
 1850                 free(fdp->fd_map, M_FILEDESC);
 1851 
 1852         fdp->fd_nfiles = 0;
 1853 
 1854         cdir = fdp->fd_cdir;
 1855         fdp->fd_cdir = NULL;
 1856         rdir = fdp->fd_rdir;
 1857         fdp->fd_rdir = NULL;
 1858         jdir = fdp->fd_jdir;
 1859         fdp->fd_jdir = NULL;
 1860         FILEDESC_XUNLOCK(fdp);
 1861 
 1862         if (cdir) {
 1863                 locked = VFS_LOCK_GIANT(cdir->v_mount);
 1864                 vrele(cdir);
 1865                 VFS_UNLOCK_GIANT(locked);
 1866         }
 1867         if (rdir) {
 1868                 locked = VFS_LOCK_GIANT(rdir->v_mount);
 1869                 vrele(rdir);
 1870                 VFS_UNLOCK_GIANT(locked);
 1871         }
 1872         if (jdir) {
 1873                 locked = VFS_LOCK_GIANT(jdir->v_mount);
 1874                 vrele(jdir);
 1875                 VFS_UNLOCK_GIANT(locked);
 1876         }
 1877 
 1878         fddrop(fdp);
 1879 }
 1880 
 1881 /*
 1882  * For setugid programs, we don't want to people to use that setugidness
 1883  * to generate error messages which write to a file which otherwise would
 1884  * otherwise be off-limits to the process.  We check for filesystems where
 1885  * the vnode can change out from under us after execve (like [lin]procfs).
 1886  *
 1887  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
 1888  * sufficient.  We also don't check for setugidness since we know we are.
 1889  */
 1890 static int
 1891 is_unsafe(struct file *fp)
 1892 {
 1893         if (fp->f_type == DTYPE_VNODE) {
 1894                 struct vnode *vp = fp->f_vnode;
 1895 
 1896                 if ((vp->v_vflag & VV_PROCDEP) != 0)
 1897                         return (1);
 1898         }
 1899         return (0);
 1900 }
 1901 
 1902 /*
 1903  * Make this setguid thing safe, if at all possible.
 1904  */
 1905 void
 1906 setugidsafety(struct thread *td)
 1907 {
 1908         struct filedesc *fdp;
 1909         int i;
 1910 
 1911         /* Certain daemons might not have file descriptors. */
 1912         fdp = td->td_proc->p_fd;
 1913         if (fdp == NULL)
 1914                 return;
 1915 
 1916         /*
 1917          * Note: fdp->fd_ofiles may be reallocated out from under us while
 1918          * we are blocked in a close.  Be careful!
 1919          */
 1920         FILEDESC_XLOCK(fdp);
 1921         for (i = 0; i <= fdp->fd_lastfile; i++) {
 1922                 if (i > 2)
 1923                         break;
 1924                 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
 1925                         struct file *fp;
 1926 
 1927                         knote_fdclose(td, i);
 1928                         /*
 1929                          * NULL-out descriptor prior to close to avoid
 1930                          * a race while close blocks.
 1931                          */
 1932                         fp = fdp->fd_ofiles[i];
 1933                         fdp->fd_ofiles[i] = NULL;
 1934                         fdp->fd_ofileflags[i] = 0;
 1935                         fdunused(fdp, i);
 1936                         FILEDESC_XUNLOCK(fdp);
 1937                         (void) closef(fp, td);
 1938                         FILEDESC_XLOCK(fdp);
 1939                 }
 1940         }
 1941         FILEDESC_XUNLOCK(fdp);
 1942 }
 1943 
 1944 /*
 1945  * If a specific file object occupies a specific file descriptor, close the
 1946  * file descriptor entry and drop a reference on the file object.  This is a
 1947  * convenience function to handle a subsequent error in a function that calls
 1948  * falloc() that handles the race that another thread might have closed the
 1949  * file descriptor out from under the thread creating the file object.
 1950  */
 1951 void
 1952 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
 1953 {
 1954 
 1955         FILEDESC_XLOCK(fdp);
 1956         if (fdp->fd_ofiles[idx] == fp) {
 1957                 fdp->fd_ofiles[idx] = NULL;
 1958                 fdunused(fdp, idx);
 1959                 FILEDESC_XUNLOCK(fdp);
 1960                 fdrop(fp, td);
 1961         } else
 1962                 FILEDESC_XUNLOCK(fdp);
 1963 }
 1964 
 1965 /*
 1966  * Close any files on exec?
 1967  */
 1968 void
 1969 fdcloseexec(struct thread *td)
 1970 {
 1971         struct filedesc *fdp;
 1972         int i;
 1973 
 1974         /* Certain daemons might not have file descriptors. */
 1975         fdp = td->td_proc->p_fd;
 1976         if (fdp == NULL)
 1977                 return;
 1978 
 1979         FILEDESC_XLOCK(fdp);
 1980 
 1981         /*
 1982          * We cannot cache fd_ofiles or fd_ofileflags since operations
 1983          * may block and rip them out from under us.
 1984          */
 1985         for (i = 0; i <= fdp->fd_lastfile; i++) {
 1986                 if (fdp->fd_ofiles[i] != NULL &&
 1987                     (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
 1988                     (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
 1989                         struct file *fp;
 1990 
 1991                         knote_fdclose(td, i);
 1992                         /*
 1993                          * NULL-out descriptor prior to close to avoid
 1994                          * a race while close blocks.
 1995                          */
 1996                         fp = fdp->fd_ofiles[i];
 1997                         fdp->fd_ofiles[i] = NULL;
 1998                         fdp->fd_ofileflags[i] = 0;
 1999                         fdunused(fdp, i);
 2000                         if (fp->f_type == DTYPE_MQUEUE)
 2001                                 mq_fdclose(td, i, fp);
 2002                         FILEDESC_XUNLOCK(fdp);
 2003                         (void) closef(fp, td);
 2004                         FILEDESC_XLOCK(fdp);
 2005                 }
 2006         }
 2007         FILEDESC_XUNLOCK(fdp);
 2008 }
 2009 
 2010 /*
 2011  * It is unsafe for set[ug]id processes to be started with file
 2012  * descriptors 0..2 closed, as these descriptors are given implicit
 2013  * significance in the Standard C library.  fdcheckstd() will create a
 2014  * descriptor referencing /dev/null for each of stdin, stdout, and
 2015  * stderr that is not already open.
 2016  */
 2017 int
 2018 fdcheckstd(struct thread *td)
 2019 {
 2020         struct filedesc *fdp;
 2021         register_t retval, save;
 2022         int i, error, devnull;
 2023 
 2024         fdp = td->td_proc->p_fd;
 2025         if (fdp == NULL)
 2026                 return (0);
 2027         KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
 2028         devnull = -1;
 2029         error = 0;
 2030         for (i = 0; i < 3; i++) {
 2031                 if (fdp->fd_ofiles[i] != NULL)
 2032                         continue;
 2033                 if (devnull < 0) {
 2034                         save = td->td_retval[0];
 2035                         error = kern_open(td, "/dev/null", UIO_SYSSPACE,
 2036                             O_RDWR, 0);
 2037                         devnull = td->td_retval[0];
 2038                         td->td_retval[0] = save;
 2039                         if (error)
 2040                                 break;
 2041                         KASSERT(devnull == i, ("oof, we didn't get our fd"));
 2042                 } else {
 2043                         error = do_dup(td, DUP_FIXED, devnull, i, &retval);
 2044                         if (error != 0)
 2045                                 break;
 2046                 }
 2047         }
 2048         return (error);
 2049 }
 2050 
 2051 /*
 2052  * Internal form of close.  Decrement reference count on file structure.
 2053  * Note: td may be NULL when closing a file that was being passed in a
 2054  * message.
 2055  *
 2056  * XXXRW: Giant is not required for the caller, but often will be held; this
 2057  * makes it moderately likely the Giant will be recursed in the VFS case.
 2058  */
 2059 int
 2060 closef(struct file *fp, struct thread *td)
 2061 {
 2062         struct vnode *vp;
 2063         struct flock lf;
 2064         struct filedesc_to_leader *fdtol;
 2065         struct filedesc *fdp;
 2066 
 2067         /*
 2068          * POSIX record locking dictates that any close releases ALL
 2069          * locks owned by this process.  This is handled by setting
 2070          * a flag in the unlock to free ONLY locks obeying POSIX
 2071          * semantics, and not to free BSD-style file locks.
 2072          * If the descriptor was in a message, POSIX-style locks
 2073          * aren't passed with the descriptor, and the thread pointer
 2074          * will be NULL.  Callers should be careful only to pass a
 2075          * NULL thread pointer when there really is no owning
 2076          * context that might have locks, or the locks will be
 2077          * leaked.
 2078          */
 2079         if (fp->f_type == DTYPE_VNODE && td != NULL) {
 2080                 int vfslocked;
 2081 
 2082                 vp = fp->f_vnode;
 2083                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 2084                 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 2085                         lf.l_whence = SEEK_SET;
 2086                         lf.l_start = 0;
 2087                         lf.l_len = 0;
 2088                         lf.l_type = F_UNLCK;
 2089                         (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
 2090                                            F_UNLCK, &lf, F_POSIX);
 2091                 }
 2092                 fdtol = td->td_proc->p_fdtol;
 2093                 if (fdtol != NULL) {
 2094                         /*
 2095                          * Handle special case where file descriptor table is
 2096                          * shared between multiple process leaders.
 2097                          */
 2098                         fdp = td->td_proc->p_fd;
 2099                         FILEDESC_XLOCK(fdp);
 2100                         for (fdtol = fdtol->fdl_next;
 2101                              fdtol != td->td_proc->p_fdtol;
 2102                              fdtol = fdtol->fdl_next) {
 2103                                 if ((fdtol->fdl_leader->p_flag &
 2104                                      P_ADVLOCK) == 0)
 2105                                         continue;
 2106                                 fdtol->fdl_holdcount++;
 2107                                 FILEDESC_XUNLOCK(fdp);
 2108                                 lf.l_whence = SEEK_SET;
 2109                                 lf.l_start = 0;
 2110                                 lf.l_len = 0;
 2111                                 lf.l_type = F_UNLCK;
 2112                                 vp = fp->f_vnode;
 2113                                 (void) VOP_ADVLOCK(vp,
 2114                                                    (caddr_t)fdtol->fdl_leader,
 2115                                                    F_UNLCK, &lf, F_POSIX);
 2116                                 FILEDESC_XLOCK(fdp);
 2117                                 fdtol->fdl_holdcount--;
 2118                                 if (fdtol->fdl_holdcount == 0 &&
 2119                                     fdtol->fdl_wakeup != 0) {
 2120                                         fdtol->fdl_wakeup = 0;
 2121                                         wakeup(fdtol);
 2122                                 }
 2123                         }
 2124                         FILEDESC_XUNLOCK(fdp);
 2125                 }
 2126                 VFS_UNLOCK_GIANT(vfslocked);
 2127         }
 2128         return (fdrop(fp, td));
 2129 }
 2130 
 2131 /*
 2132  * Initialize the file pointer with the specified properties.
 2133  * 
 2134  * The ops are set with release semantics to be certain that the flags, type,
 2135  * and data are visible when ops is.  This is to prevent ops methods from being
 2136  * called with bad data.
 2137  */
 2138 void
 2139 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
 2140 {
 2141         fp->f_data = data;
 2142         fp->f_flag = flag;
 2143         fp->f_type = type;
 2144         atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
 2145 }
 2146 
 2147 struct file *
 2148 fget_unlocked(struct filedesc *fdp, int fd)
 2149 {
 2150         struct file *fp;
 2151         u_int count;
 2152 
 2153         if (fd < 0 || fd >= fdp->fd_nfiles)
 2154                 return (NULL);
 2155         /*
 2156          * Fetch the descriptor locklessly.  We avoid fdrop() races by
 2157          * never raising a refcount above 0.  To accomplish this we have
 2158          * to use a cmpset loop rather than an atomic_add.  The descriptor
 2159          * must be re-verified once we acquire a reference to be certain
 2160          * that the identity is still correct and we did not lose a race
 2161          * due to preemption.
 2162          */
 2163         for (;;) {
 2164                 fp = fdp->fd_ofiles[fd];
 2165                 if (fp == NULL)
 2166                         break;
 2167                 count = fp->f_count;
 2168                 if (count == 0)
 2169                         continue;
 2170                 /*
 2171                  * Use an acquire barrier to prevent caching of fd_ofiles
 2172                  * so it is refreshed for verification.
 2173                  */
 2174                 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
 2175                         continue;
 2176                 if (fp == fdp->fd_ofiles[fd])
 2177                         break;
 2178                 fdrop(fp, curthread);
 2179         }
 2180 
 2181         return (fp);
 2182 }
 2183 
 2184 /*
 2185  * Extract the file pointer associated with the specified descriptor for the
 2186  * current user process.
 2187  *
 2188  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
 2189  * returned.
 2190  *
 2191  * If an error occured the non-zero error is returned and *fpp is set to
 2192  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
 2193  * responsible for fdrop().
 2194  */
 2195 static __inline int
 2196 _fget(struct thread *td, int fd, struct file **fpp, int flags)
 2197 {
 2198         struct filedesc *fdp;
 2199         struct file *fp;
 2200 
 2201         *fpp = NULL;
 2202         if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
 2203                 return (EBADF);
 2204         if ((fp = fget_unlocked(fdp, fd)) == NULL)
 2205                 return (EBADF);
 2206         if (fp->f_ops == &badfileops) {
 2207                 fdrop(fp, td);
 2208                 return (EBADF);
 2209         }
 2210         /*
 2211          * FREAD and FWRITE failure return EBADF as per POSIX.
 2212          *
 2213          * Only one flag, or 0, may be specified.
 2214          */
 2215         if ((flags == FREAD && (fp->f_flag & FREAD) == 0) ||
 2216             (flags == FWRITE && (fp->f_flag & FWRITE) == 0)) {
 2217                 fdrop(fp, td);
 2218                 return (EBADF);
 2219         }
 2220         *fpp = fp;
 2221         return (0);
 2222 }
 2223 
 2224 int
 2225 fget(struct thread *td, int fd, struct file **fpp)
 2226 {
 2227 
 2228         return(_fget(td, fd, fpp, 0));
 2229 }
 2230 
 2231 int
 2232 fget_read(struct thread *td, int fd, struct file **fpp)
 2233 {
 2234 
 2235         return(_fget(td, fd, fpp, FREAD));
 2236 }
 2237 
 2238 int
 2239 fget_write(struct thread *td, int fd, struct file **fpp)
 2240 {
 2241 
 2242         return(_fget(td, fd, fpp, FWRITE));
 2243 }
 2244 
 2245 /*
 2246  * Like fget() but loads the underlying vnode, or returns an error if the
 2247  * descriptor does not represent a vnode.  Note that pipes use vnodes but
 2248  * never have VM objects.  The returned vnode will be vref()'d.
 2249  *
 2250  * XXX: what about the unused flags ?
 2251  */
 2252 static __inline int
 2253 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
 2254 {
 2255         struct file *fp;
 2256         int error;
 2257 
 2258         *vpp = NULL;
 2259         if ((error = _fget(td, fd, &fp, flags)) != 0)
 2260                 return (error);
 2261         if (fp->f_vnode == NULL) {
 2262                 error = EINVAL;
 2263         } else {
 2264                 *vpp = fp->f_vnode;
 2265                 vref(*vpp);
 2266         }
 2267         fdrop(fp, td);
 2268 
 2269         return (error);
 2270 }
 2271 
 2272 int
 2273 fgetvp(struct thread *td, int fd, struct vnode **vpp)
 2274 {
 2275 
 2276         return (_fgetvp(td, fd, vpp, 0));
 2277 }
 2278 
 2279 int
 2280 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
 2281 {
 2282 
 2283         return (_fgetvp(td, fd, vpp, FREAD));
 2284 }
 2285 
 2286 #ifdef notyet
 2287 int
 2288 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
 2289 {
 2290 
 2291         return (_fgetvp(td, fd, vpp, FWRITE));
 2292 }
 2293 #endif
 2294 
 2295 /*
 2296  * Like fget() but loads the underlying socket, or returns an error if the
 2297  * descriptor does not represent a socket.
 2298  *
 2299  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
 2300  * in the future.
 2301  *
 2302  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
 2303  * on their file descriptor reference to prevent the socket from being free'd
 2304  * during use.
 2305  */
 2306 int
 2307 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
 2308 {
 2309         struct file *fp;
 2310         int error;
 2311 
 2312         *spp = NULL;
 2313         if (fflagp != NULL)
 2314                 *fflagp = 0;
 2315         if ((error = _fget(td, fd, &fp, 0)) != 0)
 2316                 return (error);
 2317         if (fp->f_type != DTYPE_SOCKET) {
 2318                 error = ENOTSOCK;
 2319         } else {
 2320                 *spp = fp->f_data;
 2321                 if (fflagp)
 2322                         *fflagp = fp->f_flag;
 2323                 SOCK_LOCK(*spp);
 2324                 soref(*spp);
 2325                 SOCK_UNLOCK(*spp);
 2326         }
 2327         fdrop(fp, td);
 2328 
 2329         return (error);
 2330 }
 2331 
 2332 /*
 2333  * Drop the reference count on the socket and XXX release the SX lock in the
 2334  * future.  The last reference closes the socket.
 2335  *
 2336  * Note: fputsock() is deprecated, see comment for fgetsock().
 2337  */
 2338 void
 2339 fputsock(struct socket *so)
 2340 {
 2341 
 2342         ACCEPT_LOCK();
 2343         SOCK_LOCK(so);
 2344         CURVNET_SET(so->so_vnet);
 2345         sorele(so);
 2346         CURVNET_RESTORE();
 2347 }
 2348 
 2349 /*
 2350  * Handle the last reference to a file being closed.
 2351  */
 2352 int
 2353 _fdrop(struct file *fp, struct thread *td)
 2354 {
 2355         int error;
 2356 
 2357         error = 0;
 2358         if (fp->f_count != 0)
 2359                 panic("fdrop: count %d", fp->f_count);
 2360         if (fp->f_ops != &badfileops)
 2361                 error = fo_close(fp, td);
 2362         atomic_subtract_int(&openfiles, 1);
 2363         crfree(fp->f_cred);
 2364         uma_zfree(file_zone, fp);
 2365 
 2366         return (error);
 2367 }
 2368 
 2369 /*
 2370  * Apply an advisory lock on a file descriptor.
 2371  *
 2372  * Just attempt to get a record lock of the requested type on the entire file
 2373  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
 2374  */
 2375 #ifndef _SYS_SYSPROTO_H_
 2376 struct flock_args {
 2377         int     fd;
 2378         int     how;
 2379 };
 2380 #endif
 2381 /* ARGSUSED */
 2382 int
 2383 flock(struct thread *td, struct flock_args *uap)
 2384 {
 2385         struct file *fp;
 2386         struct vnode *vp;
 2387         struct flock lf;
 2388         int vfslocked;
 2389         int error;
 2390 
 2391         if ((error = fget(td, uap->fd, &fp)) != 0)
 2392                 return (error);
 2393         if (fp->f_type != DTYPE_VNODE) {
 2394                 fdrop(fp, td);
 2395                 return (EOPNOTSUPP);
 2396         }
 2397 
 2398         vp = fp->f_vnode;
 2399         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 2400         lf.l_whence = SEEK_SET;
 2401         lf.l_start = 0;
 2402         lf.l_len = 0;
 2403         if (uap->how & LOCK_UN) {
 2404                 lf.l_type = F_UNLCK;
 2405                 atomic_clear_int(&fp->f_flag, FHASLOCK);
 2406                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
 2407                 goto done2;
 2408         }
 2409         if (uap->how & LOCK_EX)
 2410                 lf.l_type = F_WRLCK;
 2411         else if (uap->how & LOCK_SH)
 2412                 lf.l_type = F_RDLCK;
 2413         else {
 2414                 error = EBADF;
 2415                 goto done2;
 2416         }
 2417         atomic_set_int(&fp->f_flag, FHASLOCK);
 2418         error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
 2419             (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
 2420 done2:
 2421         fdrop(fp, td);
 2422         VFS_UNLOCK_GIANT(vfslocked);
 2423         return (error);
 2424 }
 2425 /*
 2426  * Duplicate the specified descriptor to a free descriptor.
 2427  */
 2428 int
 2429 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
 2430 {
 2431         struct file *wfp;
 2432         struct file *fp;
 2433 
 2434         /*
 2435          * If the to-be-dup'd fd number is greater than the allowed number
 2436          * of file descriptors, or the fd to be dup'd has already been
 2437          * closed, then reject.
 2438          */
 2439         FILEDESC_XLOCK(fdp);
 2440         if (dfd < 0 || dfd >= fdp->fd_nfiles ||
 2441             (wfp = fdp->fd_ofiles[dfd]) == NULL) {
 2442                 FILEDESC_XUNLOCK(fdp);
 2443                 return (EBADF);
 2444         }
 2445 
 2446         /*
 2447          * There are two cases of interest here.
 2448          *
 2449          * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
 2450          *
 2451          * For ENXIO steal away the file structure from (dfd) and store it in
 2452          * (indx).  (dfd) is effectively closed by this operation.
 2453          *
 2454          * Any other error code is just returned.
 2455          */
 2456         switch (error) {
 2457         case ENODEV:
 2458                 /*
 2459                  * Check that the mode the file is being opened for is a
 2460                  * subset of the mode of the existing descriptor.
 2461                  */
 2462                 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
 2463                         FILEDESC_XUNLOCK(fdp);
 2464                         return (EACCES);
 2465                 }
 2466                 fp = fdp->fd_ofiles[indx];
 2467                 fdp->fd_ofiles[indx] = wfp;
 2468                 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
 2469                 if (fp == NULL)
 2470                         fdused(fdp, indx);
 2471                 fhold(wfp);
 2472                 FILEDESC_XUNLOCK(fdp);
 2473                 if (fp != NULL)
 2474                         /*
 2475                          * We now own the reference to fp that the ofiles[]
 2476                          * array used to own.  Release it.
 2477                          */
 2478                         fdrop(fp, td);
 2479                 return (0);
 2480 
 2481         case ENXIO:
 2482                 /*
 2483                  * Steal away the file pointer from dfd and stuff it into indx.
 2484                  */
 2485                 fp = fdp->fd_ofiles[indx];
 2486                 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
 2487                 fdp->fd_ofiles[dfd] = NULL;
 2488                 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
 2489                 fdp->fd_ofileflags[dfd] = 0;
 2490                 fdunused(fdp, dfd);
 2491                 if (fp == NULL)
 2492                         fdused(fdp, indx);
 2493                 FILEDESC_XUNLOCK(fdp);
 2494 
 2495                 /*
 2496                  * We now own the reference to fp that the ofiles[] array
 2497                  * used to own.  Release it.
 2498                  */
 2499                 if (fp != NULL)
 2500                         fdrop(fp, td);
 2501                 return (0);
 2502 
 2503         default:
 2504                 FILEDESC_XUNLOCK(fdp);
 2505                 return (error);
 2506         }
 2507         /* NOTREACHED */
 2508 }
 2509 
 2510 /*
 2511  * Scan all active processes and prisons to see if any of them have a current
 2512  * or root directory of `olddp'. If so, replace them with the new mount point.
 2513  */
 2514 void
 2515 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
 2516 {
 2517         struct filedesc *fdp;
 2518         struct prison *pr;
 2519         struct proc *p;
 2520         int nrele;
 2521 
 2522         if (vrefcnt(olddp) == 1)
 2523                 return;
 2524         nrele = 0;
 2525         sx_slock(&allproc_lock);
 2526         FOREACH_PROC_IN_SYSTEM(p) {
 2527                 fdp = fdhold(p);
 2528                 if (fdp == NULL)
 2529                         continue;
 2530                 FILEDESC_XLOCK(fdp);
 2531                 if (fdp->fd_cdir == olddp) {
 2532                         vref(newdp);
 2533                         fdp->fd_cdir = newdp;
 2534                         nrele++;
 2535                 }
 2536                 if (fdp->fd_rdir == olddp) {
 2537                         vref(newdp);
 2538                         fdp->fd_rdir = newdp;
 2539                         nrele++;
 2540                 }
 2541                 if (fdp->fd_jdir == olddp) {
 2542                         vref(newdp);
 2543                         fdp->fd_jdir = newdp;
 2544                         nrele++;
 2545                 }
 2546                 FILEDESC_XUNLOCK(fdp);
 2547                 fddrop(fdp);
 2548         }
 2549         sx_sunlock(&allproc_lock);
 2550         if (rootvnode == olddp) {
 2551                 vref(newdp);
 2552                 rootvnode = newdp;
 2553                 nrele++;
 2554         }
 2555         mtx_lock(&prison0.pr_mtx);
 2556         if (prison0.pr_root == olddp) {
 2557                 vref(newdp);
 2558                 prison0.pr_root = newdp;
 2559                 nrele++;
 2560         }
 2561         mtx_unlock(&prison0.pr_mtx);
 2562         sx_slock(&allprison_lock);
 2563         TAILQ_FOREACH(pr, &allprison, pr_list) {
 2564                 mtx_lock(&pr->pr_mtx);
 2565                 if (pr->pr_root == olddp) {
 2566                         vref(newdp);
 2567                         pr->pr_root = newdp;
 2568                         nrele++;
 2569                 }
 2570                 mtx_unlock(&pr->pr_mtx);
 2571         }
 2572         sx_sunlock(&allprison_lock);
 2573         while (nrele--)
 2574                 vrele(olddp);
 2575 }
 2576 
 2577 struct filedesc_to_leader *
 2578 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
 2579 {
 2580         struct filedesc_to_leader *fdtol;
 2581 
 2582         fdtol = malloc(sizeof(struct filedesc_to_leader),
 2583                M_FILEDESC_TO_LEADER,
 2584                M_WAITOK);
 2585         fdtol->fdl_refcount = 1;
 2586         fdtol->fdl_holdcount = 0;
 2587         fdtol->fdl_wakeup = 0;
 2588         fdtol->fdl_leader = leader;
 2589         if (old != NULL) {
 2590                 FILEDESC_XLOCK(fdp);
 2591                 fdtol->fdl_next = old->fdl_next;
 2592                 fdtol->fdl_prev = old;
 2593                 old->fdl_next = fdtol;
 2594                 fdtol->fdl_next->fdl_prev = fdtol;
 2595                 FILEDESC_XUNLOCK(fdp);
 2596         } else {
 2597                 fdtol->fdl_next = fdtol;
 2598                 fdtol->fdl_prev = fdtol;
 2599         }
 2600         return (fdtol);
 2601 }
 2602 
 2603 /*
 2604  * Get file structures globally.
 2605  */
 2606 static int
 2607 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
 2608 {
 2609         struct xfile xf;
 2610         struct filedesc *fdp;
 2611         struct file *fp;
 2612         struct proc *p;
 2613         int error, n;
 2614 
 2615         error = sysctl_wire_old_buffer(req, 0);
 2616         if (error != 0)
 2617                 return (error);
 2618         if (req->oldptr == NULL) {
 2619                 n = 0;
 2620                 sx_slock(&allproc_lock);
 2621                 FOREACH_PROC_IN_SYSTEM(p) {
 2622                         if (p->p_state == PRS_NEW)
 2623                                 continue;
 2624                         fdp = fdhold(p);
 2625                         if (fdp == NULL)
 2626                                 continue;
 2627                         /* overestimates sparse tables. */
 2628                         if (fdp->fd_lastfile > 0)
 2629                                 n += fdp->fd_lastfile;
 2630                         fddrop(fdp);
 2631                 }
 2632                 sx_sunlock(&allproc_lock);
 2633                 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
 2634         }
 2635         error = 0;
 2636         bzero(&xf, sizeof(xf));
 2637         xf.xf_size = sizeof(xf);
 2638         sx_slock(&allproc_lock);
 2639         FOREACH_PROC_IN_SYSTEM(p) {
 2640                 PROC_LOCK(p);
 2641                 if (p->p_state == PRS_NEW) {
 2642                         PROC_UNLOCK(p);
 2643                         continue;
 2644                 }
 2645                 if (p_cansee(req->td, p) != 0) {
 2646                         PROC_UNLOCK(p);
 2647                         continue;
 2648                 }
 2649                 xf.xf_pid = p->p_pid;
 2650                 xf.xf_uid = p->p_ucred->cr_uid;
 2651                 PROC_UNLOCK(p);
 2652                 fdp = fdhold(p);
 2653                 if (fdp == NULL)
 2654                         continue;
 2655                 FILEDESC_SLOCK(fdp);
 2656                 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
 2657                         if ((fp = fdp->fd_ofiles[n]) == NULL)
 2658                                 continue;
 2659                         xf.xf_fd = n;
 2660                         xf.xf_file = fp;
 2661                         xf.xf_data = fp->f_data;
 2662                         xf.xf_vnode = fp->f_vnode;
 2663                         xf.xf_type = fp->f_type;
 2664                         xf.xf_count = fp->f_count;
 2665                         xf.xf_msgcount = 0;
 2666                         xf.xf_offset = fp->f_offset;
 2667                         xf.xf_flag = fp->f_flag;
 2668                         error = SYSCTL_OUT(req, &xf, sizeof(xf));
 2669                         if (error)
 2670                                 break;
 2671                 }
 2672                 FILEDESC_SUNLOCK(fdp);
 2673                 fddrop(fdp);
 2674                 if (error)
 2675                         break;
 2676         }
 2677         sx_sunlock(&allproc_lock);
 2678         return (error);
 2679 }
 2680 
 2681 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
 2682     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
 2683 
 2684 #ifdef KINFO_OFILE_SIZE
 2685 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
 2686 #endif
 2687 
 2688 #ifdef COMPAT_FREEBSD7
 2689 static int
 2690 export_vnode_for_osysctl(struct vnode *vp, int type,
 2691     struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
 2692 {
 2693         int error;
 2694         char *fullpath, *freepath;
 2695         int vfslocked;
 2696 
 2697         bzero(kif, sizeof(*kif));
 2698         kif->kf_structsize = sizeof(*kif);
 2699 
 2700         vref(vp);
 2701         kif->kf_fd = type;
 2702         kif->kf_type = KF_TYPE_VNODE;
 2703         /* This function only handles directories. */
 2704         if (vp->v_type != VDIR) {
 2705                 vrele(vp);
 2706                 return (ENOTDIR);
 2707         }
 2708         kif->kf_vnode_type = KF_VTYPE_VDIR;
 2709 
 2710         /*
 2711          * This is not a true file descriptor, so we set a bogus refcount
 2712          * and offset to indicate these fields should be ignored.
 2713          */
 2714         kif->kf_ref_count = -1;
 2715         kif->kf_offset = -1;
 2716 
 2717         freepath = NULL;
 2718         fullpath = "-";
 2719         FILEDESC_SUNLOCK(fdp);
 2720         vn_fullpath(curthread, vp, &fullpath, &freepath);
 2721         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 2722         vrele(vp);
 2723         VFS_UNLOCK_GIANT(vfslocked);
 2724         strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
 2725         if (freepath != NULL)
 2726                 free(freepath, M_TEMP);
 2727         error = SYSCTL_OUT(req, kif, sizeof(*kif));
 2728         FILEDESC_SLOCK(fdp);
 2729         return (error);
 2730 }
 2731 
 2732 /*
 2733  * Get per-process file descriptors for use by procstat(1), et al.
 2734  */
 2735 static int
 2736 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
 2737 {
 2738         char *fullpath, *freepath;
 2739         struct kinfo_ofile *kif;
 2740         struct filedesc *fdp;
 2741         int error, i, *name;
 2742         struct socket *so;
 2743         struct vnode *vp;
 2744         struct file *fp;
 2745         struct proc *p;
 2746         struct tty *tp;
 2747         int vfslocked;
 2748 
 2749         name = (int *)arg1;
 2750         if ((p = pfind((pid_t)name[0])) == NULL)
 2751                 return (ESRCH);
 2752         if ((error = p_candebug(curthread, p))) {
 2753                 PROC_UNLOCK(p);
 2754                 return (error);
 2755         }
 2756         fdp = fdhold(p);
 2757         PROC_UNLOCK(p);
 2758         if (fdp == NULL)
 2759                 return (ENOENT);
 2760         kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
 2761         FILEDESC_SLOCK(fdp);
 2762         if (fdp->fd_cdir != NULL)
 2763                 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
 2764                                 fdp, req);
 2765         if (fdp->fd_rdir != NULL)
 2766                 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
 2767                                 fdp, req);
 2768         if (fdp->fd_jdir != NULL)
 2769                 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
 2770                                 fdp, req);
 2771         for (i = 0; i < fdp->fd_nfiles; i++) {
 2772                 if ((fp = fdp->fd_ofiles[i]) == NULL)
 2773                         continue;
 2774                 bzero(kif, sizeof(*kif));
 2775                 kif->kf_structsize = sizeof(*kif);
 2776                 vp = NULL;
 2777                 so = NULL;
 2778                 tp = NULL;
 2779                 kif->kf_fd = i;
 2780                 switch (fp->f_type) {
 2781                 case DTYPE_VNODE:
 2782                         kif->kf_type = KF_TYPE_VNODE;
 2783                         vp = fp->f_vnode;
 2784                         break;
 2785 
 2786                 case DTYPE_SOCKET:
 2787                         kif->kf_type = KF_TYPE_SOCKET;
 2788                         so = fp->f_data;
 2789                         break;
 2790 
 2791                 case DTYPE_PIPE:
 2792                         kif->kf_type = KF_TYPE_PIPE;
 2793                         break;
 2794 
 2795                 case DTYPE_FIFO:
 2796                         kif->kf_type = KF_TYPE_FIFO;
 2797                         vp = fp->f_vnode;
 2798                         break;
 2799 
 2800                 case DTYPE_KQUEUE:
 2801                         kif->kf_type = KF_TYPE_KQUEUE;
 2802                         break;
 2803 
 2804                 case DTYPE_CRYPTO:
 2805                         kif->kf_type = KF_TYPE_CRYPTO;
 2806                         break;
 2807 
 2808                 case DTYPE_MQUEUE:
 2809                         kif->kf_type = KF_TYPE_MQUEUE;
 2810                         break;
 2811 
 2812                 case DTYPE_SHM:
 2813                         kif->kf_type = KF_TYPE_SHM;
 2814                         break;
 2815 
 2816                 case DTYPE_SEM:
 2817                         kif->kf_type = KF_TYPE_SEM;
 2818                         break;
 2819 
 2820                 case DTYPE_PTS:
 2821                         kif->kf_type = KF_TYPE_PTS;
 2822                         tp = fp->f_data;
 2823                         break;
 2824 
 2825                 default:
 2826                         kif->kf_type = KF_TYPE_UNKNOWN;
 2827                         break;
 2828                 }
 2829                 kif->kf_ref_count = fp->f_count;
 2830                 if (fp->f_flag & FREAD)
 2831                         kif->kf_flags |= KF_FLAG_READ;
 2832                 if (fp->f_flag & FWRITE)
 2833                         kif->kf_flags |= KF_FLAG_WRITE;
 2834                 if (fp->f_flag & FAPPEND)
 2835                         kif->kf_flags |= KF_FLAG_APPEND;
 2836                 if (fp->f_flag & FASYNC)
 2837                         kif->kf_flags |= KF_FLAG_ASYNC;
 2838                 if (fp->f_flag & FFSYNC)
 2839                         kif->kf_flags |= KF_FLAG_FSYNC;
 2840                 if (fp->f_flag & FNONBLOCK)
 2841                         kif->kf_flags |= KF_FLAG_NONBLOCK;
 2842                 if (fp->f_flag & O_DIRECT)
 2843                         kif->kf_flags |= KF_FLAG_DIRECT;
 2844                 if (fp->f_flag & FHASLOCK)
 2845                         kif->kf_flags |= KF_FLAG_HASLOCK;
 2846                 kif->kf_offset = fp->f_offset;
 2847                 if (vp != NULL) {
 2848                         vref(vp);
 2849                         switch (vp->v_type) {
 2850                         case VNON:
 2851                                 kif->kf_vnode_type = KF_VTYPE_VNON;
 2852                                 break;
 2853                         case VREG:
 2854                                 kif->kf_vnode_type = KF_VTYPE_VREG;
 2855                                 break;
 2856                         case VDIR:
 2857                                 kif->kf_vnode_type = KF_VTYPE_VDIR;
 2858                                 break;
 2859                         case VBLK:
 2860                                 kif->kf_vnode_type = KF_VTYPE_VBLK;
 2861                                 break;
 2862                         case VCHR:
 2863                                 kif->kf_vnode_type = KF_VTYPE_VCHR;
 2864                                 break;
 2865                         case VLNK:
 2866                                 kif->kf_vnode_type = KF_VTYPE_VLNK;
 2867                                 break;
 2868                         case VSOCK:
 2869                                 kif->kf_vnode_type = KF_VTYPE_VSOCK;
 2870                                 break;
 2871                         case VFIFO:
 2872                                 kif->kf_vnode_type = KF_VTYPE_VFIFO;
 2873                                 break;
 2874                         case VBAD:
 2875                                 kif->kf_vnode_type = KF_VTYPE_VBAD;
 2876                                 break;
 2877                         default:
 2878                                 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
 2879                                 break;
 2880                         }
 2881                         /*
 2882                          * It is OK to drop the filedesc lock here as we will
 2883                          * re-validate and re-evaluate its properties when
 2884                          * the loop continues.
 2885                          */
 2886                         freepath = NULL;
 2887                         fullpath = "-";
 2888                         FILEDESC_SUNLOCK(fdp);
 2889                         vn_fullpath(curthread, vp, &fullpath, &freepath);
 2890                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 2891                         vrele(vp);
 2892                         VFS_UNLOCK_GIANT(vfslocked);
 2893                         strlcpy(kif->kf_path, fullpath,
 2894                             sizeof(kif->kf_path));
 2895                         if (freepath != NULL)
 2896                                 free(freepath, M_TEMP);
 2897                         FILEDESC_SLOCK(fdp);
 2898                 }
 2899                 if (so != NULL) {
 2900                         struct sockaddr *sa;
 2901 
 2902                         if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
 2903                             == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
 2904                                 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
 2905                                 free(sa, M_SONAME);
 2906                         }
 2907                         if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
 2908                             == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
 2909                                 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
 2910                                 free(sa, M_SONAME);
 2911                         }
 2912                         kif->kf_sock_domain =
 2913                             so->so_proto->pr_domain->dom_family;
 2914                         kif->kf_sock_type = so->so_type;
 2915                         kif->kf_sock_protocol = so->so_proto->pr_protocol;
 2916                 }
 2917                 if (tp != NULL) {
 2918                         strlcpy(kif->kf_path, tty_devname(tp),
 2919                             sizeof(kif->kf_path));
 2920                 }
 2921                 error = SYSCTL_OUT(req, kif, sizeof(*kif));
 2922                 if (error)
 2923                         break;
 2924         }
 2925         FILEDESC_SUNLOCK(fdp);
 2926         fddrop(fdp);
 2927         free(kif, M_TEMP);
 2928         return (0);
 2929 }
 2930 
 2931 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
 2932     sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
 2933 #endif  /* COMPAT_FREEBSD7 */
 2934 
 2935 #ifdef KINFO_FILE_SIZE
 2936 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
 2937 #endif
 2938 
 2939 static int
 2940 export_vnode_for_sysctl(struct vnode *vp, int type,
 2941     struct kinfo_file *kif, struct filedesc *fdp, struct sysctl_req *req)
 2942 {
 2943         int error;
 2944         char *fullpath, *freepath;
 2945         int vfslocked;
 2946 
 2947         bzero(kif, sizeof(*kif));
 2948 
 2949         vref(vp);
 2950         kif->kf_fd = type;
 2951         kif->kf_type = KF_TYPE_VNODE;
 2952         /* This function only handles directories. */
 2953         if (vp->v_type != VDIR) {
 2954                 vrele(vp);
 2955                 return (ENOTDIR);
 2956         }
 2957         kif->kf_vnode_type = KF_VTYPE_VDIR;
 2958 
 2959         /*
 2960          * This is not a true file descriptor, so we set a bogus refcount
 2961          * and offset to indicate these fields should be ignored.
 2962          */
 2963         kif->kf_ref_count = -1;
 2964         kif->kf_offset = -1;
 2965 
 2966         freepath = NULL;
 2967         fullpath = "-";
 2968         FILEDESC_SUNLOCK(fdp);
 2969         vn_fullpath(curthread, vp, &fullpath, &freepath);
 2970         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 2971         vrele(vp);
 2972         VFS_UNLOCK_GIANT(vfslocked);
 2973         strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
 2974         if (freepath != NULL)
 2975                 free(freepath, M_TEMP);
 2976         /* Pack record size down */
 2977         kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
 2978             strlen(kif->kf_path) + 1;
 2979         kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
 2980         error = SYSCTL_OUT(req, kif, kif->kf_structsize);
 2981         FILEDESC_SLOCK(fdp);
 2982         return (error);
 2983 }
 2984 
 2985 /*
 2986  * Get per-process file descriptors for use by procstat(1), et al.
 2987  */
 2988 static int
 2989 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
 2990 {
 2991         char *fullpath, *freepath;
 2992         struct kinfo_file *kif;
 2993         struct filedesc *fdp;
 2994         int error, i, *name;
 2995         struct socket *so;
 2996         struct vnode *vp;
 2997         struct file *fp;
 2998         struct proc *p;
 2999         struct tty *tp;
 3000         int vfslocked;
 3001         size_t oldidx;
 3002 
 3003         name = (int *)arg1;
 3004         if ((p = pfind((pid_t)name[0])) == NULL)
 3005                 return (ESRCH);
 3006         if ((error = p_candebug(curthread, p))) {
 3007                 PROC_UNLOCK(p);
 3008                 return (error);
 3009         }
 3010         fdp = fdhold(p);
 3011         PROC_UNLOCK(p);
 3012         if (fdp == NULL)
 3013                 return (ENOENT);
 3014         kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
 3015         FILEDESC_SLOCK(fdp);
 3016         if (fdp->fd_cdir != NULL)
 3017                 export_vnode_for_sysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
 3018                                 fdp, req);
 3019         if (fdp->fd_rdir != NULL)
 3020                 export_vnode_for_sysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
 3021                                 fdp, req);
 3022         if (fdp->fd_jdir != NULL)
 3023                 export_vnode_for_sysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
 3024                                 fdp, req);
 3025         for (i = 0; i < fdp->fd_nfiles; i++) {
 3026                 if ((fp = fdp->fd_ofiles[i]) == NULL)
 3027                         continue;
 3028                 bzero(kif, sizeof(*kif));
 3029                 vp = NULL;
 3030                 so = NULL;
 3031                 tp = NULL;
 3032                 kif->kf_fd = i;
 3033                 switch (fp->f_type) {
 3034                 case DTYPE_VNODE:
 3035                         kif->kf_type = KF_TYPE_VNODE;
 3036                         vp = fp->f_vnode;
 3037                         break;
 3038 
 3039                 case DTYPE_SOCKET:
 3040                         kif->kf_type = KF_TYPE_SOCKET;
 3041                         so = fp->f_data;
 3042                         break;
 3043 
 3044                 case DTYPE_PIPE:
 3045                         kif->kf_type = KF_TYPE_PIPE;
 3046                         break;
 3047 
 3048                 case DTYPE_FIFO:
 3049                         kif->kf_type = KF_TYPE_FIFO;
 3050                         vp = fp->f_vnode;
 3051                         break;
 3052 
 3053                 case DTYPE_KQUEUE:
 3054                         kif->kf_type = KF_TYPE_KQUEUE;
 3055                         break;
 3056 
 3057                 case DTYPE_CRYPTO:
 3058                         kif->kf_type = KF_TYPE_CRYPTO;
 3059                         break;
 3060 
 3061                 case DTYPE_MQUEUE:
 3062                         kif->kf_type = KF_TYPE_MQUEUE;
 3063                         break;
 3064 
 3065                 case DTYPE_SHM:
 3066                         kif->kf_type = KF_TYPE_SHM;
 3067                         break;
 3068 
 3069                 case DTYPE_SEM:
 3070                         kif->kf_type = KF_TYPE_SEM;
 3071                         break;
 3072 
 3073                 case DTYPE_PTS:
 3074                         kif->kf_type = KF_TYPE_PTS;
 3075                         tp = fp->f_data;
 3076                         break;
 3077 
 3078                 default:
 3079                         kif->kf_type = KF_TYPE_UNKNOWN;
 3080                         break;
 3081                 }
 3082                 kif->kf_ref_count = fp->f_count;
 3083                 if (fp->f_flag & FREAD)
 3084                         kif->kf_flags |= KF_FLAG_READ;
 3085                 if (fp->f_flag & FWRITE)
 3086                         kif->kf_flags |= KF_FLAG_WRITE;
 3087                 if (fp->f_flag & FAPPEND)
 3088                         kif->kf_flags |= KF_FLAG_APPEND;
 3089                 if (fp->f_flag & FASYNC)
 3090                         kif->kf_flags |= KF_FLAG_ASYNC;
 3091                 if (fp->f_flag & FFSYNC)
 3092                         kif->kf_flags |= KF_FLAG_FSYNC;
 3093                 if (fp->f_flag & FNONBLOCK)
 3094                         kif->kf_flags |= KF_FLAG_NONBLOCK;
 3095                 if (fp->f_flag & O_DIRECT)
 3096                         kif->kf_flags |= KF_FLAG_DIRECT;
 3097                 if (fp->f_flag & FHASLOCK)
 3098                         kif->kf_flags |= KF_FLAG_HASLOCK;
 3099                 kif->kf_offset = fp->f_offset;
 3100                 if (vp != NULL) {
 3101                         vref(vp);
 3102                         switch (vp->v_type) {
 3103                         case VNON:
 3104                                 kif->kf_vnode_type = KF_VTYPE_VNON;
 3105                                 break;
 3106                         case VREG:
 3107                                 kif->kf_vnode_type = KF_VTYPE_VREG;
 3108                                 break;
 3109                         case VDIR:
 3110                                 kif->kf_vnode_type = KF_VTYPE_VDIR;
 3111                                 break;
 3112                         case VBLK:
 3113                                 kif->kf_vnode_type = KF_VTYPE_VBLK;
 3114                                 break;
 3115                         case VCHR:
 3116                                 kif->kf_vnode_type = KF_VTYPE_VCHR;
 3117                                 break;
 3118                         case VLNK:
 3119                                 kif->kf_vnode_type = KF_VTYPE_VLNK;
 3120                                 break;
 3121                         case VSOCK:
 3122                                 kif->kf_vnode_type = KF_VTYPE_VSOCK;
 3123                                 break;
 3124                         case VFIFO:
 3125                                 kif->kf_vnode_type = KF_VTYPE_VFIFO;
 3126                                 break;
 3127                         case VBAD:
 3128                                 kif->kf_vnode_type = KF_VTYPE_VBAD;
 3129                                 break;
 3130                         default:
 3131                                 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
 3132                                 break;
 3133                         }
 3134                         /*
 3135                          * It is OK to drop the filedesc lock here as we will
 3136                          * re-validate and re-evaluate its properties when
 3137                          * the loop continues.
 3138                          */
 3139                         freepath = NULL;
 3140                         fullpath = "-";
 3141                         FILEDESC_SUNLOCK(fdp);
 3142                         vn_fullpath(curthread, vp, &fullpath, &freepath);
 3143                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 3144                         vrele(vp);
 3145                         VFS_UNLOCK_GIANT(vfslocked);
 3146                         strlcpy(kif->kf_path, fullpath,
 3147                             sizeof(kif->kf_path));
 3148                         if (freepath != NULL)
 3149                                 free(freepath, M_TEMP);
 3150                         FILEDESC_SLOCK(fdp);
 3151                 }
 3152                 if (so != NULL) {
 3153                         struct sockaddr *sa;
 3154 
 3155                         if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
 3156                             == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
 3157                                 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
 3158                                 free(sa, M_SONAME);
 3159                         }
 3160                         if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
 3161                             == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
 3162                                 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
 3163                                 free(sa, M_SONAME);
 3164                         }
 3165                         kif->kf_sock_domain =
 3166                             so->so_proto->pr_domain->dom_family;
 3167                         kif->kf_sock_type = so->so_type;
 3168                         kif->kf_sock_protocol = so->so_proto->pr_protocol;
 3169                 }
 3170                 if (tp != NULL) {
 3171                         strlcpy(kif->kf_path, tty_devname(tp),
 3172                             sizeof(kif->kf_path));
 3173                 }
 3174                 /* Pack record size down */
 3175                 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
 3176                     strlen(kif->kf_path) + 1;
 3177                 kif->kf_structsize = roundup(kif->kf_structsize,
 3178                     sizeof(uint64_t));
 3179                 oldidx = req->oldidx;
 3180                 error = SYSCTL_OUT(req, kif, kif->kf_structsize);
 3181                 if (error) {
 3182                         if (error == ENOMEM) {
 3183                                 /*
 3184                                  * The hack to keep the ABI of sysctl
 3185                                  * kern.proc.filedesc intact, but not
 3186                                  * to account a partially copied
 3187                                  * kinfo_file into the oldidx.
 3188                                  */
 3189                                 req->oldidx = oldidx;
 3190                                 error = 0;
 3191                         }
 3192                         break;
 3193                 }
 3194         }
 3195         FILEDESC_SUNLOCK(fdp);
 3196         fddrop(fdp);
 3197         free(kif, M_TEMP);
 3198         return (error);
 3199 }
 3200 
 3201 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
 3202     sysctl_kern_proc_filedesc, "Process filedesc entries");
 3203 
 3204 #ifdef DDB
 3205 /*
 3206  * For the purposes of debugging, generate a human-readable string for the
 3207  * file type.
 3208  */
 3209 static const char *
 3210 file_type_to_name(short type)
 3211 {
 3212 
 3213         switch (type) {
 3214         case 0:
 3215                 return ("zero");
 3216         case DTYPE_VNODE:
 3217                 return ("vnod");
 3218         case DTYPE_SOCKET:
 3219                 return ("sock");
 3220         case DTYPE_PIPE:
 3221                 return ("pipe");
 3222         case DTYPE_FIFO:
 3223                 return ("fifo");
 3224         case DTYPE_KQUEUE:
 3225                 return ("kque");
 3226         case DTYPE_CRYPTO:
 3227                 return ("crpt");
 3228         case DTYPE_MQUEUE:
 3229                 return ("mque");
 3230         case DTYPE_SHM:
 3231                 return ("shm");
 3232         case DTYPE_SEM:
 3233                 return ("ksem");
 3234         default:
 3235                 return ("unkn");
 3236         }
 3237 }
 3238 
 3239 /*
 3240  * For the purposes of debugging, identify a process (if any, perhaps one of
 3241  * many) that references the passed file in its file descriptor array. Return
 3242  * NULL if none.
 3243  */
 3244 static struct proc *
 3245 file_to_first_proc(struct file *fp)
 3246 {
 3247         struct filedesc *fdp;
 3248         struct proc *p;
 3249         int n;
 3250 
 3251         FOREACH_PROC_IN_SYSTEM(p) {
 3252                 if (p->p_state == PRS_NEW)
 3253                         continue;
 3254                 fdp = p->p_fd;
 3255                 if (fdp == NULL)
 3256                         continue;
 3257                 for (n = 0; n < fdp->fd_nfiles; n++) {
 3258                         if (fp == fdp->fd_ofiles[n])
 3259                                 return (p);
 3260                 }
 3261         }
 3262         return (NULL);
 3263 }
 3264 
 3265 static void
 3266 db_print_file(struct file *fp, int header)
 3267 {
 3268         struct proc *p;
 3269 
 3270         if (header)
 3271                 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
 3272                     "File", "Type", "Data", "Flag", "GCFl", "Count",
 3273                     "MCount", "Vnode", "FPID", "FCmd");
 3274         p = file_to_first_proc(fp);
 3275         db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
 3276             file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
 3277             0, fp->f_count, 0, fp->f_vnode,
 3278             p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
 3279 }
 3280 
 3281 DB_SHOW_COMMAND(file, db_show_file)
 3282 {
 3283         struct file *fp;
 3284 
 3285         if (!have_addr) {
 3286                 db_printf("usage: show file <addr>\n");
 3287                 return;
 3288         }
 3289         fp = (struct file *)addr;
 3290         db_print_file(fp, 1);
 3291 }
 3292 
 3293 DB_SHOW_COMMAND(files, db_show_files)
 3294 {
 3295         struct filedesc *fdp;
 3296         struct file *fp;
 3297         struct proc *p;
 3298         int header;
 3299         int n;
 3300 
 3301         header = 1;
 3302         FOREACH_PROC_IN_SYSTEM(p) {
 3303                 if (p->p_state == PRS_NEW)
 3304                         continue;
 3305                 if ((fdp = p->p_fd) == NULL)
 3306                         continue;
 3307                 for (n = 0; n < fdp->fd_nfiles; ++n) {
 3308                         if ((fp = fdp->fd_ofiles[n]) == NULL)
 3309                                 continue;
 3310                         db_print_file(fp, header);
 3311                         header = 0;
 3312                 }
 3313         }
 3314 }
 3315 #endif
 3316 
 3317 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
 3318     &maxfilesperproc, 0, "Maximum files allowed open per process");
 3319 
 3320 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
 3321     &maxfiles, 0, "Maximum number of files");
 3322 
 3323 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
 3324     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
 3325 
 3326 /* ARGSUSED*/
 3327 static void
 3328 filelistinit(void *dummy)
 3329 {
 3330 
 3331         file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
 3332             NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
 3333         mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
 3334         mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
 3335 }
 3336 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
 3337 
 3338 /*-------------------------------------------------------------------*/
 3339 
 3340 static int
 3341 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
 3342 {
 3343 
 3344         return (EBADF);
 3345 }
 3346 
 3347 static int
 3348 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td)
 3349 {
 3350 
 3351         return (EINVAL);
 3352 }
 3353 
 3354 static int
 3355 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
 3356 {
 3357 
 3358         return (EBADF);
 3359 }
 3360 
 3361 static int
 3362 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
 3363 {
 3364 
 3365         return (0);
 3366 }
 3367 
 3368 static int
 3369 badfo_kqfilter(struct file *fp, struct knote *kn)
 3370 {
 3371 
 3372         return (EBADF);
 3373 }
 3374 
 3375 static int
 3376 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
 3377 {
 3378 
 3379         return (EBADF);
 3380 }
 3381 
 3382 static int
 3383 badfo_close(struct file *fp, struct thread *td)
 3384 {
 3385 
 3386         return (EBADF);
 3387 }
 3388 
 3389 struct fileops badfileops = {
 3390         .fo_read = badfo_readwrite,
 3391         .fo_write = badfo_readwrite,
 3392         .fo_truncate = badfo_truncate,
 3393         .fo_ioctl = badfo_ioctl,
 3394         .fo_poll = badfo_poll,
 3395         .fo_kqfilter = badfo_kqfilter,
 3396         .fo_stat = badfo_stat,
 3397         .fo_close = badfo_close,
 3398 };
 3399 
 3400 
 3401 /*-------------------------------------------------------------------*/
 3402 
 3403 /*
 3404  * File Descriptor pseudo-device driver (/dev/fd/).
 3405  *
 3406  * Opening minor device N dup()s the file (if any) connected to file
 3407  * descriptor N belonging to the calling process.  Note that this driver
 3408  * consists of only the ``open()'' routine, because all subsequent
 3409  * references to this file will be direct to the other driver.
 3410  *
 3411  * XXX: we could give this one a cloning event handler if necessary.
 3412  */
 3413 
 3414 /* ARGSUSED */
 3415 static int
 3416 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
 3417 {
 3418 
 3419         /*
 3420          * XXX Kludge: set curthread->td_dupfd to contain the value of the
 3421          * the file descriptor being sought for duplication. The error
 3422          * return ensures that the vnode for this device will be released
 3423          * by vn_open. Open will detect this special error and take the
 3424          * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
 3425          * will simply report the error.
 3426          */
 3427         td->td_dupfd = dev2unit(dev);
 3428         return (ENODEV);
 3429 }
 3430 
 3431 static struct cdevsw fildesc_cdevsw = {
 3432         .d_version =    D_VERSION,
 3433         .d_open =       fdopen,
 3434         .d_name =       "FD",
 3435 };
 3436 
 3437 static void
 3438 fildesc_drvinit(void *unused)
 3439 {
 3440         struct cdev *dev;
 3441 
 3442         dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
 3443             UID_ROOT, GID_WHEEL, 0666, "fd/0");
 3444         make_dev_alias(dev, "stdin");
 3445         dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
 3446             UID_ROOT, GID_WHEEL, 0666, "fd/1");
 3447         make_dev_alias(dev, "stdout");
 3448         dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
 3449             UID_ROOT, GID_WHEEL, 0666, "fd/2");
 3450         make_dev_alias(dev, "stderr");
 3451 }
 3452 
 3453 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);

Cache object: 5867b5ad9fec8b5dcea638b6e62d4d8c


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.