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-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
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/10.1/sys/kern/kern_descrip.c 273137 2014-10-15 16:54:18Z mjg $");
   39 
   40 #include "opt_capsicum.h"
   41 #include "opt_compat.h"
   42 #include "opt_ddb.h"
   43 #include "opt_ktrace.h"
   44 #include "opt_procdesc.h"
   45 
   46 #include <sys/param.h>
   47 #include <sys/systm.h>
   48 
   49 #include <sys/capability.h>
   50 #include <sys/conf.h>
   51 #include <sys/domain.h>
   52 #include <sys/fcntl.h>
   53 #include <sys/file.h>
   54 #include <sys/filedesc.h>
   55 #include <sys/filio.h>
   56 #include <sys/jail.h>
   57 #include <sys/kernel.h>
   58 #include <sys/ksem.h>
   59 #include <sys/limits.h>
   60 #include <sys/lock.h>
   61 #include <sys/malloc.h>
   62 #include <sys/mman.h>
   63 #include <sys/mount.h>
   64 #include <sys/mqueue.h>
   65 #include <sys/mutex.h>
   66 #include <sys/namei.h>
   67 #include <sys/selinfo.h>
   68 #include <sys/pipe.h>
   69 #include <sys/priv.h>
   70 #include <sys/proc.h>
   71 #include <sys/procdesc.h>
   72 #include <sys/protosw.h>
   73 #include <sys/racct.h>
   74 #include <sys/resourcevar.h>
   75 #include <sys/sbuf.h>
   76 #include <sys/signalvar.h>
   77 #include <sys/socketvar.h>
   78 #include <sys/stat.h>
   79 #include <sys/sx.h>
   80 #include <sys/syscallsubr.h>
   81 #include <sys/sysctl.h>
   82 #include <sys/sysproto.h>
   83 #include <sys/tty.h>
   84 #include <sys/unistd.h>
   85 #include <sys/un.h>
   86 #include <sys/unpcb.h>
   87 #include <sys/user.h>
   88 #include <sys/vnode.h>
   89 #ifdef KTRACE
   90 #include <sys/ktrace.h>
   91 #endif
   92 
   93 #include <net/vnet.h>
   94 
   95 #include <netinet/in.h>
   96 #include <netinet/in_pcb.h>
   97 
   98 #include <security/audit/audit.h>
   99 
  100 #include <vm/uma.h>
  101 #include <vm/vm.h>
  102 
  103 #include <ddb/ddb.h>
  104 
  105 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
  106 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
  107     "file desc to leader structures");
  108 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
  109 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
  110 
  111 MALLOC_DECLARE(M_FADVISE);
  112 
  113 static uma_zone_t file_zone;
  114 
  115 void    (*ksem_info)(struct ksem *ks, char *path, size_t size, uint32_t *value);
  116 
  117 static int      closefp(struct filedesc *fdp, int fd, struct file *fp,
  118                     struct thread *td, int holdleaders);
  119 static int      fd_first_free(struct filedesc *fdp, int low, int size);
  120 static int      fd_last_used(struct filedesc *fdp, int size);
  121 static void     fdgrowtable(struct filedesc *fdp, int nfd);
  122 static void     fdgrowtable_exp(struct filedesc *fdp, int nfd);
  123 static void     fdunused(struct filedesc *fdp, int fd);
  124 static void     fdused(struct filedesc *fdp, int fd);
  125 static int      fill_pipe_info(struct pipe *pi, struct kinfo_file *kif);
  126 static int      fill_procdesc_info(struct procdesc *pdp,
  127                     struct kinfo_file *kif);
  128 static int      fill_pts_info(struct tty *tp, struct kinfo_file *kif);
  129 static int      fill_sem_info(struct file *fp, struct kinfo_file *kif);
  130 static int      fill_shm_info(struct file *fp, struct kinfo_file *kif);
  131 static int      fill_socket_info(struct socket *so, struct kinfo_file *kif);
  132 static int      fill_vnode_info(struct vnode *vp, struct kinfo_file *kif);
  133 static int      getmaxfd(struct proc *p);
  134 
  135 /*
  136  * Each process has:
  137  *
  138  * - An array of open file descriptors (fd_ofiles)
  139  * - An array of file flags (fd_ofileflags)
  140  * - A bitmap recording which descriptors are in use (fd_map)
  141  *
  142  * A process starts out with NDFILE descriptors.  The value of NDFILE has
  143  * been selected based the historical limit of 20 open files, and an
  144  * assumption that the majority of processes, especially short-lived
  145  * processes like shells, will never need more.
  146  *
  147  * If this initial allocation is exhausted, a larger descriptor table and
  148  * map are allocated dynamically, and the pointers in the process's struct
  149  * filedesc are updated to point to those.  This is repeated every time
  150  * the process runs out of file descriptors (provided it hasn't hit its
  151  * resource limit).
  152  *
  153  * Since threads may hold references to individual descriptor table
  154  * entries, the tables are never freed.  Instead, they are placed on a
  155  * linked list and freed only when the struct filedesc is released.
  156  */
  157 #define NDFILE          20
  158 #define NDSLOTSIZE      sizeof(NDSLOTTYPE)
  159 #define NDENTRIES       (NDSLOTSIZE * __CHAR_BIT)
  160 #define NDSLOT(x)       ((x) / NDENTRIES)
  161 #define NDBIT(x)        ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
  162 #define NDSLOTS(x)      (((x) + NDENTRIES - 1) / NDENTRIES)
  163 
  164 /*
  165  * SLIST entry used to keep track of ofiles which must be reclaimed when
  166  * the process exits.
  167  */
  168 struct freetable {
  169         struct filedescent *ft_table;
  170         SLIST_ENTRY(freetable) ft_next;
  171 };
  172 
  173 /*
  174  * Initial allocation: a filedesc structure + the head of SLIST used to
  175  * keep track of old ofiles + enough space for NDFILE descriptors.
  176  */
  177 struct filedesc0 {
  178         struct filedesc fd_fd;
  179         SLIST_HEAD(, freetable) fd_free;
  180         struct  filedescent fd_dfiles[NDFILE];
  181         NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
  182 };
  183 
  184 /*
  185  * Descriptor management.
  186  */
  187 volatile int openfiles;                 /* actual number of open files */
  188 struct mtx sigio_lock;          /* mtx to protect pointers to sigio */
  189 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
  190 
  191 /* A mutex to protect the association between a proc and filedesc. */
  192 static struct mtx fdesc_mtx;
  193 
  194 /*
  195  * If low >= size, just return low. Otherwise find the first zero bit in the
  196  * given bitmap, starting at low and not exceeding size - 1. Return size if
  197  * not found.
  198  */
  199 static int
  200 fd_first_free(struct filedesc *fdp, int low, int size)
  201 {
  202         NDSLOTTYPE *map = fdp->fd_map;
  203         NDSLOTTYPE mask;
  204         int off, maxoff;
  205 
  206         if (low >= size)
  207                 return (low);
  208 
  209         off = NDSLOT(low);
  210         if (low % NDENTRIES) {
  211                 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
  212                 if ((mask &= ~map[off]) != 0UL)
  213                         return (off * NDENTRIES + ffsl(mask) - 1);
  214                 ++off;
  215         }
  216         for (maxoff = NDSLOTS(size); off < maxoff; ++off)
  217                 if (map[off] != ~0UL)
  218                         return (off * NDENTRIES + ffsl(~map[off]) - 1);
  219         return (size);
  220 }
  221 
  222 /*
  223  * Find the highest non-zero bit in the given bitmap, starting at 0 and
  224  * not exceeding size - 1. Return -1 if not found.
  225  */
  226 static int
  227 fd_last_used(struct filedesc *fdp, int size)
  228 {
  229         NDSLOTTYPE *map = fdp->fd_map;
  230         NDSLOTTYPE mask;
  231         int off, minoff;
  232 
  233         off = NDSLOT(size);
  234         if (size % NDENTRIES) {
  235                 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
  236                 if ((mask &= map[off]) != 0)
  237                         return (off * NDENTRIES + flsl(mask) - 1);
  238                 --off;
  239         }
  240         for (minoff = NDSLOT(0); off >= minoff; --off)
  241                 if (map[off] != 0)
  242                         return (off * NDENTRIES + flsl(map[off]) - 1);
  243         return (-1);
  244 }
  245 
  246 static int
  247 fdisused(struct filedesc *fdp, int fd)
  248 {
  249 
  250         FILEDESC_LOCK_ASSERT(fdp);
  251 
  252         KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
  253             ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
  254 
  255         return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
  256 }
  257 
  258 /*
  259  * Mark a file descriptor as used.
  260  */
  261 static void
  262 fdused(struct filedesc *fdp, int fd)
  263 {
  264 
  265         FILEDESC_XLOCK_ASSERT(fdp);
  266 
  267         KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
  268 
  269         fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
  270         if (fd > fdp->fd_lastfile)
  271                 fdp->fd_lastfile = fd;
  272         if (fd == fdp->fd_freefile)
  273                 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
  274 }
  275 
  276 /*
  277  * Mark a file descriptor as unused.
  278  */
  279 static void
  280 fdunused(struct filedesc *fdp, int fd)
  281 {
  282 
  283         FILEDESC_XLOCK_ASSERT(fdp);
  284 
  285         KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
  286         KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
  287             ("fd=%d is still in use", fd));
  288 
  289         fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
  290         if (fd < fdp->fd_freefile)
  291                 fdp->fd_freefile = fd;
  292         if (fd == fdp->fd_lastfile)
  293                 fdp->fd_lastfile = fd_last_used(fdp, fd);
  294 }
  295 
  296 /*
  297  * Free a file descriptor.
  298  *
  299  * Avoid some work if fdp is about to be destroyed.
  300  */
  301 static inline void
  302 _fdfree(struct filedesc *fdp, int fd, int last)
  303 {
  304         struct filedescent *fde;
  305 
  306         fde = &fdp->fd_ofiles[fd];
  307 #ifdef CAPABILITIES
  308         if (!last)
  309                 seq_write_begin(&fde->fde_seq);
  310 #endif
  311         filecaps_free(&fde->fde_caps);
  312         if (last)
  313                 return;
  314         bzero(fde, fde_change_size);
  315         fdunused(fdp, fd);
  316 #ifdef CAPABILITIES
  317         seq_write_end(&fde->fde_seq);
  318 #endif
  319 }
  320 
  321 static inline void
  322 fdfree(struct filedesc *fdp, int fd)
  323 {
  324 
  325         _fdfree(fdp, fd, 0);
  326 }
  327 
  328 static inline void
  329 fdfree_last(struct filedesc *fdp, int fd)
  330 {
  331 
  332         _fdfree(fdp, fd, 1);
  333 }
  334 
  335 /*
  336  * System calls on descriptors.
  337  */
  338 #ifndef _SYS_SYSPROTO_H_
  339 struct getdtablesize_args {
  340         int     dummy;
  341 };
  342 #endif
  343 /* ARGSUSED */
  344 int
  345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
  346 {
  347         struct proc *p = td->td_proc;
  348         uint64_t lim;
  349 
  350         PROC_LOCK(p);
  351         td->td_retval[0] =
  352             min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
  353         lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
  354         PROC_UNLOCK(p);
  355         if (lim < td->td_retval[0])
  356                 td->td_retval[0] = lim;
  357         return (0);
  358 }
  359 
  360 /*
  361  * Duplicate a file descriptor to a particular value.
  362  *
  363  * Note: keep in mind that a potential race condition exists when closing
  364  * descriptors from a shared descriptor table (via rfork).
  365  */
  366 #ifndef _SYS_SYSPROTO_H_
  367 struct dup2_args {
  368         u_int   from;
  369         u_int   to;
  370 };
  371 #endif
  372 /* ARGSUSED */
  373 int
  374 sys_dup2(struct thread *td, struct dup2_args *uap)
  375 {
  376 
  377         return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
  378                     td->td_retval));
  379 }
  380 
  381 /*
  382  * Duplicate a file descriptor.
  383  */
  384 #ifndef _SYS_SYSPROTO_H_
  385 struct dup_args {
  386         u_int   fd;
  387 };
  388 #endif
  389 /* ARGSUSED */
  390 int
  391 sys_dup(struct thread *td, struct dup_args *uap)
  392 {
  393 
  394         return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
  395 }
  396 
  397 /*
  398  * The file control system call.
  399  */
  400 #ifndef _SYS_SYSPROTO_H_
  401 struct fcntl_args {
  402         int     fd;
  403         int     cmd;
  404         long    arg;
  405 };
  406 #endif
  407 /* ARGSUSED */
  408 int
  409 sys_fcntl(struct thread *td, struct fcntl_args *uap)
  410 {
  411 
  412         return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
  413 }
  414 
  415 int
  416 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
  417 {
  418         struct flock fl;
  419         struct __oflock ofl;
  420         intptr_t arg1;
  421         int error;
  422 
  423         error = 0;
  424         switch (cmd) {
  425         case F_OGETLK:
  426         case F_OSETLK:
  427         case F_OSETLKW:
  428                 /*
  429                  * Convert old flock structure to new.
  430                  */
  431                 error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
  432                 fl.l_start = ofl.l_start;
  433                 fl.l_len = ofl.l_len;
  434                 fl.l_pid = ofl.l_pid;
  435                 fl.l_type = ofl.l_type;
  436                 fl.l_whence = ofl.l_whence;
  437                 fl.l_sysid = 0;
  438 
  439                 switch (cmd) {
  440                 case F_OGETLK:
  441                     cmd = F_GETLK;
  442                     break;
  443                 case F_OSETLK:
  444                     cmd = F_SETLK;
  445                     break;
  446                 case F_OSETLKW:
  447                     cmd = F_SETLKW;
  448                     break;
  449                 }
  450                 arg1 = (intptr_t)&fl;
  451                 break;
  452         case F_GETLK:
  453         case F_SETLK:
  454         case F_SETLKW:
  455         case F_SETLK_REMOTE:
  456                 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
  457                 arg1 = (intptr_t)&fl;
  458                 break;
  459         default:
  460                 arg1 = arg;
  461                 break;
  462         }
  463         if (error)
  464                 return (error);
  465         error = kern_fcntl(td, fd, cmd, arg1);
  466         if (error)
  467                 return (error);
  468         if (cmd == F_OGETLK) {
  469                 ofl.l_start = fl.l_start;
  470                 ofl.l_len = fl.l_len;
  471                 ofl.l_pid = fl.l_pid;
  472                 ofl.l_type = fl.l_type;
  473                 ofl.l_whence = fl.l_whence;
  474                 error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
  475         } else if (cmd == F_GETLK) {
  476                 error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
  477         }
  478         return (error);
  479 }
  480 
  481 int
  482 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
  483 {
  484         struct filedesc *fdp;
  485         struct flock *flp;
  486         struct file *fp, *fp2;
  487         struct filedescent *fde;
  488         struct proc *p;
  489         struct vnode *vp;
  490         cap_rights_t rights;
  491         int error, flg, tmp;
  492         uint64_t bsize;
  493         off_t foffset;
  494 
  495         error = 0;
  496         flg = F_POSIX;
  497         p = td->td_proc;
  498         fdp = p->p_fd;
  499 
  500         switch (cmd) {
  501         case F_DUPFD:
  502                 tmp = arg;
  503                 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
  504                 break;
  505 
  506         case F_DUPFD_CLOEXEC:
  507                 tmp = arg;
  508                 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp,
  509                     td->td_retval);
  510                 break;
  511 
  512         case F_DUP2FD:
  513                 tmp = arg;
  514                 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
  515                 break;
  516 
  517         case F_DUP2FD_CLOEXEC:
  518                 tmp = arg;
  519                 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp,
  520                     td->td_retval);
  521                 break;
  522 
  523         case F_GETFD:
  524                 FILEDESC_SLOCK(fdp);
  525                 if ((fp = fget_locked(fdp, fd)) == NULL) {
  526                         FILEDESC_SUNLOCK(fdp);
  527                         error = EBADF;
  528                         break;
  529                 }
  530                 fde = &fdp->fd_ofiles[fd];
  531                 td->td_retval[0] =
  532                     (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
  533                 FILEDESC_SUNLOCK(fdp);
  534                 break;
  535 
  536         case F_SETFD:
  537                 FILEDESC_XLOCK(fdp);
  538                 if ((fp = fget_locked(fdp, fd)) == NULL) {
  539                         FILEDESC_XUNLOCK(fdp);
  540                         error = EBADF;
  541                         break;
  542                 }
  543                 fde = &fdp->fd_ofiles[fd];
  544                 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
  545                     (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
  546                 FILEDESC_XUNLOCK(fdp);
  547                 break;
  548 
  549         case F_GETFL:
  550                 error = fget_unlocked(fdp, fd,
  551                     cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL);
  552                 if (error != 0)
  553                         break;
  554                 td->td_retval[0] = OFLAGS(fp->f_flag);
  555                 fdrop(fp, td);
  556                 break;
  557 
  558         case F_SETFL:
  559                 error = fget_unlocked(fdp, fd,
  560                     cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL);
  561                 if (error != 0)
  562                         break;
  563                 do {
  564                         tmp = flg = fp->f_flag;
  565                         tmp &= ~FCNTLFLAGS;
  566                         tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
  567                 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
  568                 tmp = fp->f_flag & FNONBLOCK;
  569                 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
  570                 if (error != 0) {
  571                         fdrop(fp, td);
  572                         break;
  573                 }
  574                 tmp = fp->f_flag & FASYNC;
  575                 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
  576                 if (error == 0) {
  577                         fdrop(fp, td);
  578                         break;
  579                 }
  580                 atomic_clear_int(&fp->f_flag, FNONBLOCK);
  581                 tmp = 0;
  582                 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
  583                 fdrop(fp, td);
  584                 break;
  585 
  586         case F_GETOWN:
  587                 error = fget_unlocked(fdp, fd,
  588                     cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL);
  589                 if (error != 0)
  590                         break;
  591                 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
  592                 if (error == 0)
  593                         td->td_retval[0] = tmp;
  594                 fdrop(fp, td);
  595                 break;
  596 
  597         case F_SETOWN:
  598                 error = fget_unlocked(fdp, fd,
  599                     cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL);
  600                 if (error != 0)
  601                         break;
  602                 tmp = arg;
  603                 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
  604                 fdrop(fp, td);
  605                 break;
  606 
  607         case F_SETLK_REMOTE:
  608                 error = priv_check(td, PRIV_NFS_LOCKD);
  609                 if (error)
  610                         return (error);
  611                 flg = F_REMOTE;
  612                 goto do_setlk;
  613 
  614         case F_SETLKW:
  615                 flg |= F_WAIT;
  616                 /* FALLTHROUGH F_SETLK */
  617 
  618         case F_SETLK:
  619         do_setlk:
  620                 cap_rights_init(&rights, CAP_FLOCK);
  621                 error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL);
  622                 if (error != 0)
  623                         break;
  624                 if (fp->f_type != DTYPE_VNODE) {
  625                         error = EBADF;
  626                         fdrop(fp, td);
  627                         break;
  628                 }
  629 
  630                 flp = (struct flock *)arg;
  631                 if (flp->l_whence == SEEK_CUR) {
  632                         foffset = foffset_get(fp);
  633                         if (foffset < 0 ||
  634                             (flp->l_start > 0 &&
  635                              foffset > OFF_MAX - flp->l_start)) {
  636                                 FILEDESC_SUNLOCK(fdp);
  637                                 error = EOVERFLOW;
  638                                 fdrop(fp, td);
  639                                 break;
  640                         }
  641                         flp->l_start += foffset;
  642                 }
  643 
  644                 vp = fp->f_vnode;
  645                 switch (flp->l_type) {
  646                 case F_RDLCK:
  647                         if ((fp->f_flag & FREAD) == 0) {
  648                                 error = EBADF;
  649                                 break;
  650                         }
  651                         PROC_LOCK(p->p_leader);
  652                         p->p_leader->p_flag |= P_ADVLOCK;
  653                         PROC_UNLOCK(p->p_leader);
  654                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
  655                             flp, flg);
  656                         break;
  657                 case F_WRLCK:
  658                         if ((fp->f_flag & FWRITE) == 0) {
  659                                 error = EBADF;
  660                                 break;
  661                         }
  662                         PROC_LOCK(p->p_leader);
  663                         p->p_leader->p_flag |= P_ADVLOCK;
  664                         PROC_UNLOCK(p->p_leader);
  665                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
  666                             flp, flg);
  667                         break;
  668                 case F_UNLCK:
  669                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
  670                             flp, flg);
  671                         break;
  672                 case F_UNLCKSYS:
  673                         /*
  674                          * Temporary api for testing remote lock
  675                          * infrastructure.
  676                          */
  677                         if (flg != F_REMOTE) {
  678                                 error = EINVAL;
  679                                 break;
  680                         }
  681                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
  682                             F_UNLCKSYS, flp, flg);
  683                         break;
  684                 default:
  685                         error = EINVAL;
  686                         break;
  687                 }
  688                 if (error != 0 || flp->l_type == F_UNLCK ||
  689                     flp->l_type == F_UNLCKSYS) {
  690                         fdrop(fp, td);
  691                         break;
  692                 }
  693 
  694                 /*
  695                  * Check for a race with close.
  696                  *
  697                  * The vnode is now advisory locked (or unlocked, but this case
  698                  * is not really important) as the caller requested.
  699                  * We had to drop the filedesc lock, so we need to recheck if
  700                  * the descriptor is still valid, because if it was closed
  701                  * in the meantime we need to remove advisory lock from the
  702                  * vnode - close on any descriptor leading to an advisory
  703                  * locked vnode, removes that lock.
  704                  * We will return 0 on purpose in that case, as the result of
  705                  * successful advisory lock might have been externally visible
  706                  * already. This is fine - effectively we pretend to the caller
  707                  * that the closing thread was a bit slower and that the
  708                  * advisory lock succeeded before the close.
  709                  */
  710                 error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL);
  711                 if (error != 0) {
  712                         fdrop(fp, td);
  713                         break;
  714                 }
  715                 if (fp != fp2) {
  716                         flp->l_whence = SEEK_SET;
  717                         flp->l_start = 0;
  718                         flp->l_len = 0;
  719                         flp->l_type = F_UNLCK;
  720                         (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
  721                             F_UNLCK, flp, F_POSIX);
  722                 }
  723                 fdrop(fp, td);
  724                 fdrop(fp2, td);
  725                 break;
  726 
  727         case F_GETLK:
  728                 error = fget_unlocked(fdp, fd,
  729                     cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL);
  730                 if (error != 0)
  731                         break;
  732                 if (fp->f_type != DTYPE_VNODE) {
  733                         error = EBADF;
  734                         fdrop(fp, td);
  735                         break;
  736                 }
  737                 flp = (struct flock *)arg;
  738                 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
  739                     flp->l_type != F_UNLCK) {
  740                         error = EINVAL;
  741                         fdrop(fp, td);
  742                         break;
  743                 }
  744                 if (flp->l_whence == SEEK_CUR) {
  745                         foffset = foffset_get(fp);
  746                         if ((flp->l_start > 0 &&
  747                             foffset > OFF_MAX - flp->l_start) ||
  748                             (flp->l_start < 0 &&
  749                              foffset < OFF_MIN - flp->l_start)) {
  750                                 FILEDESC_SUNLOCK(fdp);
  751                                 error = EOVERFLOW;
  752                                 fdrop(fp, td);
  753                                 break;
  754                         }
  755                         flp->l_start += foffset;
  756                 }
  757                 vp = fp->f_vnode;
  758                 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
  759                     F_POSIX);
  760                 fdrop(fp, td);
  761                 break;
  762 
  763         case F_RDAHEAD:
  764                 arg = arg ? 128 * 1024: 0;
  765                 /* FALLTHROUGH */
  766         case F_READAHEAD:
  767                 error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL);
  768                 if (error != 0)
  769                         break;
  770                 if (fp->f_type != DTYPE_VNODE) {
  771                         fdrop(fp, td);
  772                         error = EBADF;
  773                         break;
  774                 }
  775                 vp = fp->f_vnode;
  776                 /*
  777                  * Exclusive lock synchronizes against f_seqcount reads and
  778                  * writes in sequential_heuristic().
  779                  */
  780                 error = vn_lock(vp, LK_EXCLUSIVE);
  781                 if (error != 0) {
  782                         fdrop(fp, td);
  783                         break;
  784                 }
  785                 if (arg >= 0) {
  786                         bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
  787                         fp->f_seqcount = (arg + bsize - 1) / bsize;
  788                         atomic_set_int(&fp->f_flag, FRDAHEAD);
  789                 } else {
  790                         atomic_clear_int(&fp->f_flag, FRDAHEAD);
  791                 }
  792                 VOP_UNLOCK(vp, 0);
  793                 fdrop(fp, td);
  794                 break;
  795 
  796         default:
  797                 error = EINVAL;
  798                 break;
  799         }
  800         return (error);
  801 }
  802 
  803 static int
  804 getmaxfd(struct proc *p)
  805 {
  806         int maxfd;
  807 
  808         PROC_LOCK(p);
  809         maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
  810         PROC_UNLOCK(p);
  811 
  812         return (maxfd);
  813 }
  814 
  815 /*
  816  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
  817  */
  818 int
  819 do_dup(struct thread *td, int flags, int old, int new,
  820     register_t *retval)
  821 {
  822         struct filedesc *fdp;
  823         struct filedescent *oldfde, *newfde;
  824         struct proc *p;
  825         struct file *fp;
  826         struct file *delfp;
  827         int error, maxfd;
  828 
  829         p = td->td_proc;
  830         fdp = p->p_fd;
  831 
  832         /*
  833          * Verify we have a valid descriptor to dup from and possibly to
  834          * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
  835          * return EINVAL when the new descriptor is out of bounds.
  836          */
  837         if (old < 0)
  838                 return (EBADF);
  839         if (new < 0)
  840                 return (flags & DUP_FCNTL ? EINVAL : EBADF);
  841         maxfd = getmaxfd(p);
  842         if (new >= maxfd)
  843                 return (flags & DUP_FCNTL ? EINVAL : EBADF);
  844 
  845         FILEDESC_XLOCK(fdp);
  846         if (fget_locked(fdp, old) == NULL) {
  847                 FILEDESC_XUNLOCK(fdp);
  848                 return (EBADF);
  849         }
  850         oldfde = &fdp->fd_ofiles[old];
  851         if (flags & DUP_FIXED && old == new) {
  852                 *retval = new;
  853                 if (flags & DUP_CLOEXEC)
  854                         fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
  855                 FILEDESC_XUNLOCK(fdp);
  856                 return (0);
  857         }
  858         fp = oldfde->fde_file;
  859         fhold(fp);
  860 
  861         /*
  862          * If the caller specified a file descriptor, make sure the file
  863          * table is large enough to hold it, and grab it.  Otherwise, just
  864          * allocate a new descriptor the usual way.
  865          */
  866         if (flags & DUP_FIXED) {
  867                 if (new >= fdp->fd_nfiles) {
  868                         /*
  869                          * The resource limits are here instead of e.g.
  870                          * fdalloc(), because the file descriptor table may be
  871                          * shared between processes, so we can't really use
  872                          * racct_add()/racct_sub().  Instead of counting the
  873                          * number of actually allocated descriptors, just put
  874                          * the limit on the size of the file descriptor table.
  875                          */
  876 #ifdef RACCT
  877                         PROC_LOCK(p);
  878                         error = racct_set(p, RACCT_NOFILE, new + 1);
  879                         PROC_UNLOCK(p);
  880                         if (error != 0) {
  881                                 FILEDESC_XUNLOCK(fdp);
  882                                 fdrop(fp, td);
  883                                 return (EMFILE);
  884                         }
  885 #endif
  886                         fdgrowtable_exp(fdp, new + 1);
  887                         oldfde = &fdp->fd_ofiles[old];
  888                 }
  889                 newfde = &fdp->fd_ofiles[new];
  890                 if (newfde->fde_file == NULL)
  891                         fdused(fdp, new);
  892         } else {
  893                 if ((error = fdalloc(td, new, &new)) != 0) {
  894                         FILEDESC_XUNLOCK(fdp);
  895                         fdrop(fp, td);
  896                         return (error);
  897                 }
  898                 newfde = &fdp->fd_ofiles[new];
  899         }
  900 
  901         KASSERT(fp == oldfde->fde_file, ("old fd has been modified"));
  902         KASSERT(old != new, ("new fd is same as old"));
  903 
  904         delfp = newfde->fde_file;
  905 
  906         /*
  907          * Duplicate the source descriptor.
  908          */
  909 #ifdef CAPABILITIES
  910         seq_write_begin(&newfde->fde_seq);
  911 #endif
  912         filecaps_free(&newfde->fde_caps);
  913         memcpy(newfde, oldfde, fde_change_size);
  914         filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
  915         if ((flags & DUP_CLOEXEC) != 0)
  916                 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
  917         else
  918                 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
  919 #ifdef CAPABILITIES
  920         seq_write_end(&newfde->fde_seq);
  921 #endif
  922         *retval = new;
  923 
  924         if (delfp != NULL) {
  925                 (void) closefp(fdp, new, delfp, td, 1);
  926                 /* closefp() drops the FILEDESC lock for us. */
  927         } else {
  928                 FILEDESC_XUNLOCK(fdp);
  929         }
  930 
  931         return (0);
  932 }
  933 
  934 /*
  935  * If sigio is on the list associated with a process or process group,
  936  * disable signalling from the device, remove sigio from the list and
  937  * free sigio.
  938  */
  939 void
  940 funsetown(struct sigio **sigiop)
  941 {
  942         struct sigio *sigio;
  943 
  944         SIGIO_LOCK();
  945         sigio = *sigiop;
  946         if (sigio == NULL) {
  947                 SIGIO_UNLOCK();
  948                 return;
  949         }
  950         *(sigio->sio_myref) = NULL;
  951         if ((sigio)->sio_pgid < 0) {
  952                 struct pgrp *pg = (sigio)->sio_pgrp;
  953                 PGRP_LOCK(pg);
  954                 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
  955                              sigio, sio_pgsigio);
  956                 PGRP_UNLOCK(pg);
  957         } else {
  958                 struct proc *p = (sigio)->sio_proc;
  959                 PROC_LOCK(p);
  960                 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
  961                              sigio, sio_pgsigio);
  962                 PROC_UNLOCK(p);
  963         }
  964         SIGIO_UNLOCK();
  965         crfree(sigio->sio_ucred);
  966         free(sigio, M_SIGIO);
  967 }
  968 
  969 /*
  970  * Free a list of sigio structures.
  971  * We only need to lock the SIGIO_LOCK because we have made ourselves
  972  * inaccessible to callers of fsetown and therefore do not need to lock
  973  * the proc or pgrp struct for the list manipulation.
  974  */
  975 void
  976 funsetownlst(struct sigiolst *sigiolst)
  977 {
  978         struct proc *p;
  979         struct pgrp *pg;
  980         struct sigio *sigio;
  981 
  982         sigio = SLIST_FIRST(sigiolst);
  983         if (sigio == NULL)
  984                 return;
  985         p = NULL;
  986         pg = NULL;
  987 
  988         /*
  989          * Every entry of the list should belong
  990          * to a single proc or pgrp.
  991          */
  992         if (sigio->sio_pgid < 0) {
  993                 pg = sigio->sio_pgrp;
  994                 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
  995         } else /* if (sigio->sio_pgid > 0) */ {
  996                 p = sigio->sio_proc;
  997                 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
  998         }
  999 
 1000         SIGIO_LOCK();
 1001         while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
 1002                 *(sigio->sio_myref) = NULL;
 1003                 if (pg != NULL) {
 1004                         KASSERT(sigio->sio_pgid < 0,
 1005                             ("Proc sigio in pgrp sigio list"));
 1006                         KASSERT(sigio->sio_pgrp == pg,
 1007                             ("Bogus pgrp in sigio list"));
 1008                         PGRP_LOCK(pg);
 1009                         SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
 1010                             sio_pgsigio);
 1011                         PGRP_UNLOCK(pg);
 1012                 } else /* if (p != NULL) */ {
 1013                         KASSERT(sigio->sio_pgid > 0,
 1014                             ("Pgrp sigio in proc sigio list"));
 1015                         KASSERT(sigio->sio_proc == p,
 1016                             ("Bogus proc in sigio list"));
 1017                         PROC_LOCK(p);
 1018                         SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
 1019                             sio_pgsigio);
 1020                         PROC_UNLOCK(p);
 1021                 }
 1022                 SIGIO_UNLOCK();
 1023                 crfree(sigio->sio_ucred);
 1024                 free(sigio, M_SIGIO);
 1025                 SIGIO_LOCK();
 1026         }
 1027         SIGIO_UNLOCK();
 1028 }
 1029 
 1030 /*
 1031  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
 1032  *
 1033  * After permission checking, add a sigio structure to the sigio list for
 1034  * the process or process group.
 1035  */
 1036 int
 1037 fsetown(pid_t pgid, struct sigio **sigiop)
 1038 {
 1039         struct proc *proc;
 1040         struct pgrp *pgrp;
 1041         struct sigio *sigio;
 1042         int ret;
 1043 
 1044         if (pgid == 0) {
 1045                 funsetown(sigiop);
 1046                 return (0);
 1047         }
 1048 
 1049         ret = 0;
 1050 
 1051         /* Allocate and fill in the new sigio out of locks. */
 1052         sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
 1053         sigio->sio_pgid = pgid;
 1054         sigio->sio_ucred = crhold(curthread->td_ucred);
 1055         sigio->sio_myref = sigiop;
 1056 
 1057         sx_slock(&proctree_lock);
 1058         if (pgid > 0) {
 1059                 proc = pfind(pgid);
 1060                 if (proc == NULL) {
 1061                         ret = ESRCH;
 1062                         goto fail;
 1063                 }
 1064 
 1065                 /*
 1066                  * Policy - Don't allow a process to FSETOWN a process
 1067                  * in another session.
 1068                  *
 1069                  * Remove this test to allow maximum flexibility or
 1070                  * restrict FSETOWN to the current process or process
 1071                  * group for maximum safety.
 1072                  */
 1073                 PROC_UNLOCK(proc);
 1074                 if (proc->p_session != curthread->td_proc->p_session) {
 1075                         ret = EPERM;
 1076                         goto fail;
 1077                 }
 1078 
 1079                 pgrp = NULL;
 1080         } else /* if (pgid < 0) */ {
 1081                 pgrp = pgfind(-pgid);
 1082                 if (pgrp == NULL) {
 1083                         ret = ESRCH;
 1084                         goto fail;
 1085                 }
 1086                 PGRP_UNLOCK(pgrp);
 1087 
 1088                 /*
 1089                  * Policy - Don't allow a process to FSETOWN a process
 1090                  * in another session.
 1091                  *
 1092                  * Remove this test to allow maximum flexibility or
 1093                  * restrict FSETOWN to the current process or process
 1094                  * group for maximum safety.
 1095                  */
 1096                 if (pgrp->pg_session != curthread->td_proc->p_session) {
 1097                         ret = EPERM;
 1098                         goto fail;
 1099                 }
 1100 
 1101                 proc = NULL;
 1102         }
 1103         funsetown(sigiop);
 1104         if (pgid > 0) {
 1105                 PROC_LOCK(proc);
 1106                 /*
 1107                  * Since funsetownlst() is called without the proctree
 1108                  * locked, we need to check for P_WEXIT.
 1109                  * XXX: is ESRCH correct?
 1110                  */
 1111                 if ((proc->p_flag & P_WEXIT) != 0) {
 1112                         PROC_UNLOCK(proc);
 1113                         ret = ESRCH;
 1114                         goto fail;
 1115                 }
 1116                 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
 1117                 sigio->sio_proc = proc;
 1118                 PROC_UNLOCK(proc);
 1119         } else {
 1120                 PGRP_LOCK(pgrp);
 1121                 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
 1122                 sigio->sio_pgrp = pgrp;
 1123                 PGRP_UNLOCK(pgrp);
 1124         }
 1125         sx_sunlock(&proctree_lock);
 1126         SIGIO_LOCK();
 1127         *sigiop = sigio;
 1128         SIGIO_UNLOCK();
 1129         return (0);
 1130 
 1131 fail:
 1132         sx_sunlock(&proctree_lock);
 1133         crfree(sigio->sio_ucred);
 1134         free(sigio, M_SIGIO);
 1135         return (ret);
 1136 }
 1137 
 1138 /*
 1139  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
 1140  */
 1141 pid_t
 1142 fgetown(sigiop)
 1143         struct sigio **sigiop;
 1144 {
 1145         pid_t pgid;
 1146 
 1147         SIGIO_LOCK();
 1148         pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
 1149         SIGIO_UNLOCK();
 1150         return (pgid);
 1151 }
 1152 
 1153 /*
 1154  * Function drops the filedesc lock on return.
 1155  */
 1156 static int
 1157 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
 1158     int holdleaders)
 1159 {
 1160         int error;
 1161 
 1162         FILEDESC_XLOCK_ASSERT(fdp);
 1163 
 1164         if (holdleaders) {
 1165                 if (td->td_proc->p_fdtol != NULL) {
 1166                         /*
 1167                          * Ask fdfree() to sleep to ensure that all relevant
 1168                          * process leaders can be traversed in closef().
 1169                          */
 1170                         fdp->fd_holdleaderscount++;
 1171                 } else {
 1172                         holdleaders = 0;
 1173                 }
 1174         }
 1175 
 1176         /*
 1177          * We now hold the fp reference that used to be owned by the
 1178          * descriptor array.  We have to unlock the FILEDESC *AFTER*
 1179          * knote_fdclose to prevent a race of the fd getting opened, a knote
 1180          * added, and deleteing a knote for the new fd.
 1181          */
 1182         knote_fdclose(td, fd);
 1183 
 1184         /*
 1185          * We need to notify mqueue if the object is of type mqueue.
 1186          */
 1187         if (fp->f_type == DTYPE_MQUEUE)
 1188                 mq_fdclose(td, fd, fp);
 1189         FILEDESC_XUNLOCK(fdp);
 1190 
 1191         error = closef(fp, td);
 1192         if (holdleaders) {
 1193                 FILEDESC_XLOCK(fdp);
 1194                 fdp->fd_holdleaderscount--;
 1195                 if (fdp->fd_holdleaderscount == 0 &&
 1196                     fdp->fd_holdleaderswakeup != 0) {
 1197                         fdp->fd_holdleaderswakeup = 0;
 1198                         wakeup(&fdp->fd_holdleaderscount);
 1199                 }
 1200                 FILEDESC_XUNLOCK(fdp);
 1201         }
 1202         return (error);
 1203 }
 1204 
 1205 /*
 1206  * Close a file descriptor.
 1207  */
 1208 #ifndef _SYS_SYSPROTO_H_
 1209 struct close_args {
 1210         int     fd;
 1211 };
 1212 #endif
 1213 /* ARGSUSED */
 1214 int
 1215 sys_close(td, uap)
 1216         struct thread *td;
 1217         struct close_args *uap;
 1218 {
 1219 
 1220         return (kern_close(td, uap->fd));
 1221 }
 1222 
 1223 int
 1224 kern_close(td, fd)
 1225         struct thread *td;
 1226         int fd;
 1227 {
 1228         struct filedesc *fdp;
 1229         struct file *fp;
 1230 
 1231         fdp = td->td_proc->p_fd;
 1232 
 1233         AUDIT_SYSCLOSE(td, fd);
 1234 
 1235         FILEDESC_XLOCK(fdp);
 1236         if ((fp = fget_locked(fdp, fd)) == NULL) {
 1237                 FILEDESC_XUNLOCK(fdp);
 1238                 return (EBADF);
 1239         }
 1240         fdfree(fdp, fd);
 1241 
 1242         /* closefp() drops the FILEDESC lock for us. */
 1243         return (closefp(fdp, fd, fp, td, 1));
 1244 }
 1245 
 1246 /*
 1247  * Close open file descriptors.
 1248  */
 1249 #ifndef _SYS_SYSPROTO_H_
 1250 struct closefrom_args {
 1251         int     lowfd;
 1252 };
 1253 #endif
 1254 /* ARGSUSED */
 1255 int
 1256 sys_closefrom(struct thread *td, struct closefrom_args *uap)
 1257 {
 1258         struct filedesc *fdp;
 1259         int fd;
 1260 
 1261         fdp = td->td_proc->p_fd;
 1262         AUDIT_ARG_FD(uap->lowfd);
 1263 
 1264         /*
 1265          * Treat negative starting file descriptor values identical to
 1266          * closefrom(0) which closes all files.
 1267          */
 1268         if (uap->lowfd < 0)
 1269                 uap->lowfd = 0;
 1270         FILEDESC_SLOCK(fdp);
 1271         for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
 1272                 if (fdp->fd_ofiles[fd].fde_file != NULL) {
 1273                         FILEDESC_SUNLOCK(fdp);
 1274                         (void)kern_close(td, fd);
 1275                         FILEDESC_SLOCK(fdp);
 1276                 }
 1277         }
 1278         FILEDESC_SUNLOCK(fdp);
 1279         return (0);
 1280 }
 1281 
 1282 #if defined(COMPAT_43)
 1283 /*
 1284  * Return status information about a file descriptor.
 1285  */
 1286 #ifndef _SYS_SYSPROTO_H_
 1287 struct ofstat_args {
 1288         int     fd;
 1289         struct  ostat *sb;
 1290 };
 1291 #endif
 1292 /* ARGSUSED */
 1293 int
 1294 ofstat(struct thread *td, struct ofstat_args *uap)
 1295 {
 1296         struct ostat oub;
 1297         struct stat ub;
 1298         int error;
 1299 
 1300         error = kern_fstat(td, uap->fd, &ub);
 1301         if (error == 0) {
 1302                 cvtstat(&ub, &oub);
 1303                 error = copyout(&oub, uap->sb, sizeof(oub));
 1304         }
 1305         return (error);
 1306 }
 1307 #endif /* COMPAT_43 */
 1308 
 1309 /*
 1310  * Return status information about a file descriptor.
 1311  */
 1312 #ifndef _SYS_SYSPROTO_H_
 1313 struct fstat_args {
 1314         int     fd;
 1315         struct  stat *sb;
 1316 };
 1317 #endif
 1318 /* ARGSUSED */
 1319 int
 1320 sys_fstat(struct thread *td, struct fstat_args *uap)
 1321 {
 1322         struct stat ub;
 1323         int error;
 1324 
 1325         error = kern_fstat(td, uap->fd, &ub);
 1326         if (error == 0)
 1327                 error = copyout(&ub, uap->sb, sizeof(ub));
 1328         return (error);
 1329 }
 1330 
 1331 int
 1332 kern_fstat(struct thread *td, int fd, struct stat *sbp)
 1333 {
 1334         struct file *fp;
 1335         cap_rights_t rights;
 1336         int error;
 1337 
 1338         AUDIT_ARG_FD(fd);
 1339 
 1340         error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
 1341         if (error != 0)
 1342                 return (error);
 1343 
 1344         AUDIT_ARG_FILE(td->td_proc, fp);
 1345 
 1346         error = fo_stat(fp, sbp, td->td_ucred, td);
 1347         fdrop(fp, td);
 1348 #ifdef KTRACE
 1349         if (error == 0 && KTRPOINT(td, KTR_STRUCT))
 1350                 ktrstat(sbp);
 1351 #endif
 1352         return (error);
 1353 }
 1354 
 1355 /*
 1356  * Return status information about a file descriptor.
 1357  */
 1358 #ifndef _SYS_SYSPROTO_H_
 1359 struct nfstat_args {
 1360         int     fd;
 1361         struct  nstat *sb;
 1362 };
 1363 #endif
 1364 /* ARGSUSED */
 1365 int
 1366 sys_nfstat(struct thread *td, struct nfstat_args *uap)
 1367 {
 1368         struct nstat nub;
 1369         struct stat ub;
 1370         int error;
 1371 
 1372         error = kern_fstat(td, uap->fd, &ub);
 1373         if (error == 0) {
 1374                 cvtnstat(&ub, &nub);
 1375                 error = copyout(&nub, uap->sb, sizeof(nub));
 1376         }
 1377         return (error);
 1378 }
 1379 
 1380 /*
 1381  * Return pathconf information about a file descriptor.
 1382  */
 1383 #ifndef _SYS_SYSPROTO_H_
 1384 struct fpathconf_args {
 1385         int     fd;
 1386         int     name;
 1387 };
 1388 #endif
 1389 /* ARGSUSED */
 1390 int
 1391 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
 1392 {
 1393         struct file *fp;
 1394         struct vnode *vp;
 1395         cap_rights_t rights;
 1396         int error;
 1397 
 1398         error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
 1399         if (error != 0)
 1400                 return (error);
 1401 
 1402         /* If asynchronous I/O is available, it works for all descriptors. */
 1403         if (uap->name == _PC_ASYNC_IO) {
 1404                 td->td_retval[0] = async_io_version;
 1405                 goto out;
 1406         }
 1407         vp = fp->f_vnode;
 1408         if (vp != NULL) {
 1409                 vn_lock(vp, LK_SHARED | LK_RETRY);
 1410                 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
 1411                 VOP_UNLOCK(vp, 0);
 1412         } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
 1413                 if (uap->name != _PC_PIPE_BUF) {
 1414                         error = EINVAL;
 1415                 } else {
 1416                         td->td_retval[0] = PIPE_BUF;
 1417                         error = 0;
 1418                 }
 1419         } else {
 1420                 error = EOPNOTSUPP;
 1421         }
 1422 out:
 1423         fdrop(fp, td);
 1424         return (error);
 1425 }
 1426 
 1427 /*
 1428  * Initialize filecaps structure.
 1429  */
 1430 void
 1431 filecaps_init(struct filecaps *fcaps)
 1432 {
 1433 
 1434         bzero(fcaps, sizeof(*fcaps));
 1435         fcaps->fc_nioctls = -1;
 1436 }
 1437 
 1438 /*
 1439  * Copy filecaps structure allocating memory for ioctls array if needed.
 1440  */
 1441 void
 1442 filecaps_copy(const struct filecaps *src, struct filecaps *dst)
 1443 {
 1444         size_t size;
 1445 
 1446         *dst = *src;
 1447         if (src->fc_ioctls != NULL) {
 1448                 KASSERT(src->fc_nioctls > 0,
 1449                     ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
 1450 
 1451                 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
 1452                 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
 1453                 bcopy(src->fc_ioctls, dst->fc_ioctls, size);
 1454         }
 1455 }
 1456 
 1457 /*
 1458  * Move filecaps structure to the new place and clear the old place.
 1459  */
 1460 void
 1461 filecaps_move(struct filecaps *src, struct filecaps *dst)
 1462 {
 1463 
 1464         *dst = *src;
 1465         bzero(src, sizeof(*src));
 1466 }
 1467 
 1468 /*
 1469  * Fill the given filecaps structure with full rights.
 1470  */
 1471 static void
 1472 filecaps_fill(struct filecaps *fcaps)
 1473 {
 1474 
 1475         CAP_ALL(&fcaps->fc_rights);
 1476         fcaps->fc_ioctls = NULL;
 1477         fcaps->fc_nioctls = -1;
 1478         fcaps->fc_fcntls = CAP_FCNTL_ALL;
 1479 }
 1480 
 1481 /*
 1482  * Free memory allocated within filecaps structure.
 1483  */
 1484 void
 1485 filecaps_free(struct filecaps *fcaps)
 1486 {
 1487 
 1488         free(fcaps->fc_ioctls, M_FILECAPS);
 1489         bzero(fcaps, sizeof(*fcaps));
 1490 }
 1491 
 1492 /*
 1493  * Validate the given filecaps structure.
 1494  */
 1495 static void
 1496 filecaps_validate(const struct filecaps *fcaps, const char *func)
 1497 {
 1498 
 1499         KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
 1500             ("%s: invalid rights", func));
 1501         KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
 1502             ("%s: invalid fcntls", func));
 1503         KASSERT(fcaps->fc_fcntls == 0 ||
 1504             cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
 1505             ("%s: fcntls without CAP_FCNTL", func));
 1506         KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
 1507             (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
 1508             ("%s: invalid ioctls", func));
 1509         KASSERT(fcaps->fc_nioctls == 0 ||
 1510             cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
 1511             ("%s: ioctls without CAP_IOCTL", func));
 1512 }
 1513 
 1514 static void
 1515 fdgrowtable_exp(struct filedesc *fdp, int nfd)
 1516 {
 1517         int nfd1;
 1518 
 1519         FILEDESC_XLOCK_ASSERT(fdp);
 1520 
 1521         nfd1 = fdp->fd_nfiles * 2;
 1522         if (nfd1 < nfd)
 1523                 nfd1 = nfd;
 1524         fdgrowtable(fdp, nfd1);
 1525 }
 1526 
 1527 /*
 1528  * Grow the file table to accomodate (at least) nfd descriptors.
 1529  */
 1530 static void
 1531 fdgrowtable(struct filedesc *fdp, int nfd)
 1532 {
 1533         struct filedesc0 *fdp0;
 1534         struct freetable *ft;
 1535         struct filedescent *ntable;
 1536         struct filedescent *otable;
 1537         int nnfiles, onfiles;
 1538         NDSLOTTYPE *nmap, *omap;
 1539 
 1540         FILEDESC_XLOCK_ASSERT(fdp);
 1541 
 1542         KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
 1543 
 1544         /* save old values */
 1545         onfiles = fdp->fd_nfiles;
 1546         otable = fdp->fd_ofiles;
 1547         omap = fdp->fd_map;
 1548 
 1549         /* compute the size of the new table */
 1550         nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
 1551         if (nnfiles <= onfiles)
 1552                 /* the table is already large enough */
 1553                 return;
 1554 
 1555         /*
 1556          * Allocate a new table.  We need enough space for the
 1557          * file entries themselves and the struct freetable we will use
 1558          * when we decommission the table and place it on the freelist.
 1559          * We place the struct freetable in the middle so we don't have
 1560          * to worry about padding.
 1561          */
 1562         ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable),
 1563             M_FILEDESC, M_ZERO | M_WAITOK);
 1564         /* copy the old data over and point at the new tables */
 1565         memcpy(ntable, otable, onfiles * sizeof(*otable));
 1566         fdp->fd_ofiles = ntable;
 1567 
 1568         /*
 1569          * Allocate a new map only if the old is not large enough.  It will
 1570          * grow at a slower rate than the table as it can map more
 1571          * entries than the table can hold.
 1572          */
 1573         if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
 1574                 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
 1575                     M_ZERO | M_WAITOK);
 1576                 /* copy over the old data and update the pointer */
 1577                 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
 1578                 fdp->fd_map = nmap;
 1579         }
 1580 
 1581         /*
 1582          * In order to have a valid pattern for fget_unlocked()
 1583          * fdp->fd_nfiles must be the last member to be updated, otherwise
 1584          * fget_unlocked() consumers may reference a new, higher value for
 1585          * fdp->fd_nfiles before to access the fdp->fd_ofiles array,
 1586          * resulting in OOB accesses.
 1587          */
 1588         atomic_store_rel_int(&fdp->fd_nfiles, nnfiles);
 1589 
 1590         /*
 1591          * Do not free the old file table, as some threads may still
 1592          * reference entries within it.  Instead, place it on a freelist
 1593          * which will be processed when the struct filedesc is released.
 1594          *
 1595          * Note that if onfiles == NDFILE, we're dealing with the original
 1596          * static allocation contained within (struct filedesc0 *)fdp,
 1597          * which must not be freed.
 1598          */
 1599         if (onfiles > NDFILE) {
 1600                 ft = (struct freetable *)&otable[onfiles];
 1601                 fdp0 = (struct filedesc0 *)fdp;
 1602                 ft->ft_table = otable;
 1603                 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
 1604         }
 1605         /*
 1606          * The map does not have the same possibility of threads still
 1607          * holding references to it.  So always free it as long as it
 1608          * does not reference the original static allocation.
 1609          */
 1610         if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
 1611                 free(omap, M_FILEDESC);
 1612 }
 1613 
 1614 /*
 1615  * Allocate a file descriptor for the process.
 1616  */
 1617 int
 1618 fdalloc(struct thread *td, int minfd, int *result)
 1619 {
 1620         struct proc *p = td->td_proc;
 1621         struct filedesc *fdp = p->p_fd;
 1622         int fd = -1, maxfd, allocfd;
 1623 #ifdef RACCT
 1624         int error;
 1625 #endif
 1626 
 1627         FILEDESC_XLOCK_ASSERT(fdp);
 1628 
 1629         if (fdp->fd_freefile > minfd)
 1630                 minfd = fdp->fd_freefile;
 1631 
 1632         maxfd = getmaxfd(p);
 1633 
 1634         /*
 1635          * Search the bitmap for a free descriptor starting at minfd.
 1636          * If none is found, grow the file table.
 1637          */
 1638         fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
 1639         if (fd >= maxfd)
 1640                 return (EMFILE);
 1641         if (fd >= fdp->fd_nfiles) {
 1642                 allocfd = min(fd * 2, maxfd);
 1643 #ifdef RACCT
 1644                 PROC_LOCK(p);
 1645                 error = racct_set(p, RACCT_NOFILE, allocfd);
 1646                 PROC_UNLOCK(p);
 1647                 if (error != 0)
 1648                         return (EMFILE);
 1649 #endif
 1650                 /*
 1651                  * fd is already equal to first free descriptor >= minfd, so
 1652                  * we only need to grow the table and we are done.
 1653                  */
 1654                 fdgrowtable_exp(fdp, allocfd);
 1655         }
 1656 
 1657         /*
 1658          * Perform some sanity checks, then mark the file descriptor as
 1659          * used and return it to the caller.
 1660          */
 1661         KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
 1662             ("invalid descriptor %d", fd));
 1663         KASSERT(!fdisused(fdp, fd),
 1664             ("fd_first_free() returned non-free descriptor"));
 1665         KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
 1666             ("file descriptor isn't free"));
 1667         KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set"));
 1668         fdused(fdp, fd);
 1669         *result = fd;
 1670         return (0);
 1671 }
 1672 
 1673 /*
 1674  * Allocate n file descriptors for the process.
 1675  */
 1676 int
 1677 fdallocn(struct thread *td, int minfd, int *fds, int n)
 1678 {
 1679         struct proc *p = td->td_proc;
 1680         struct filedesc *fdp = p->p_fd;
 1681         int i;
 1682 
 1683         FILEDESC_XLOCK_ASSERT(fdp);
 1684 
 1685         if (!fdavail(td, n))
 1686                 return (EMFILE);
 1687 
 1688         for (i = 0; i < n; i++)
 1689                 if (fdalloc(td, 0, &fds[i]) != 0)
 1690                         break;
 1691 
 1692         if (i < n) {
 1693                 for (i--; i >= 0; i--)
 1694                         fdunused(fdp, fds[i]);
 1695                 return (EMFILE);
 1696         }
 1697 
 1698         return (0);
 1699 }
 1700 
 1701 /*
 1702  * Check to see whether n user file descriptors are available to the process
 1703  * p.
 1704  */
 1705 int
 1706 fdavail(struct thread *td, int n)
 1707 {
 1708         struct proc *p = td->td_proc;
 1709         struct filedesc *fdp = td->td_proc->p_fd;
 1710         int i, lim, last;
 1711 
 1712         FILEDESC_LOCK_ASSERT(fdp);
 1713 
 1714         /*
 1715          * XXX: This is only called from uipc_usrreq.c:unp_externalize();
 1716          *      call racct_add() from there instead of dealing with containers
 1717          *      here.
 1718          */
 1719         lim = getmaxfd(p);
 1720         if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
 1721                 return (1);
 1722         last = min(fdp->fd_nfiles, lim);
 1723         for (i = fdp->fd_freefile; i < last; i++) {
 1724                 if (fdp->fd_ofiles[i].fde_file == NULL && --n <= 0)
 1725                         return (1);
 1726         }
 1727         return (0);
 1728 }
 1729 
 1730 /*
 1731  * Create a new open file structure and allocate a file decriptor for the
 1732  * process that refers to it.  We add one reference to the file for the
 1733  * descriptor table and one reference for resultfp. This is to prevent us
 1734  * being preempted and the entry in the descriptor table closed after we
 1735  * release the FILEDESC lock.
 1736  */
 1737 int
 1738 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
 1739 {
 1740         struct file *fp;
 1741         int error, fd;
 1742 
 1743         error = falloc_noinstall(td, &fp);
 1744         if (error)
 1745                 return (error);         /* no reference held on error */
 1746 
 1747         error = finstall(td, fp, &fd, flags, NULL);
 1748         if (error) {
 1749                 fdrop(fp, td);          /* one reference (fp only) */
 1750                 return (error);
 1751         }
 1752 
 1753         if (resultfp != NULL)
 1754                 *resultfp = fp;         /* copy out result */
 1755         else
 1756                 fdrop(fp, td);          /* release local reference */
 1757 
 1758         if (resultfd != NULL)
 1759                 *resultfd = fd;
 1760 
 1761         return (0);
 1762 }
 1763 
 1764 /*
 1765  * Create a new open file structure without allocating a file descriptor.
 1766  */
 1767 int
 1768 falloc_noinstall(struct thread *td, struct file **resultfp)
 1769 {
 1770         struct file *fp;
 1771         int maxuserfiles = maxfiles - (maxfiles / 20);
 1772         static struct timeval lastfail;
 1773         static int curfail;
 1774 
 1775         KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
 1776 
 1777         if ((openfiles >= maxuserfiles &&
 1778             priv_check(td, PRIV_MAXFILES) != 0) ||
 1779             openfiles >= maxfiles) {
 1780                 if (ppsratecheck(&lastfail, &curfail, 1)) {
 1781                         printf("kern.maxfiles limit exceeded by uid %i, "
 1782                             "please see tuning(7).\n", td->td_ucred->cr_ruid);
 1783                 }
 1784                 return (ENFILE);
 1785         }
 1786         atomic_add_int(&openfiles, 1);
 1787         fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
 1788         refcount_init(&fp->f_count, 1);
 1789         fp->f_cred = crhold(td->td_ucred);
 1790         fp->f_ops = &badfileops;
 1791         fp->f_data = NULL;
 1792         fp->f_vnode = NULL;
 1793         *resultfp = fp;
 1794         return (0);
 1795 }
 1796 
 1797 /*
 1798  * Install a file in a file descriptor table.
 1799  */
 1800 int
 1801 finstall(struct thread *td, struct file *fp, int *fd, int flags,
 1802     struct filecaps *fcaps)
 1803 {
 1804         struct filedesc *fdp = td->td_proc->p_fd;
 1805         struct filedescent *fde;
 1806         int error;
 1807 
 1808         KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
 1809         KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
 1810         if (fcaps != NULL)
 1811                 filecaps_validate(fcaps, __func__);
 1812 
 1813         FILEDESC_XLOCK(fdp);
 1814         if ((error = fdalloc(td, 0, fd))) {
 1815                 FILEDESC_XUNLOCK(fdp);
 1816                 return (error);
 1817         }
 1818         fhold(fp);
 1819         fde = &fdp->fd_ofiles[*fd];
 1820 #ifdef CAPABILITIES
 1821         seq_write_begin(&fde->fde_seq);
 1822 #endif
 1823         fde->fde_file = fp;
 1824         if ((flags & O_CLOEXEC) != 0)
 1825                 fde->fde_flags |= UF_EXCLOSE;
 1826         if (fcaps != NULL)
 1827                 filecaps_move(fcaps, &fde->fde_caps);
 1828         else
 1829                 filecaps_fill(&fde->fde_caps);
 1830 #ifdef CAPABILITIES
 1831         seq_write_end(&fde->fde_seq);
 1832 #endif
 1833         FILEDESC_XUNLOCK(fdp);
 1834         return (0);
 1835 }
 1836 
 1837 /*
 1838  * Build a new filedesc structure from another.
 1839  * Copy the current, root, and jail root vnode references.
 1840  */
 1841 struct filedesc *
 1842 fdinit(struct filedesc *fdp)
 1843 {
 1844         struct filedesc0 *newfdp;
 1845 
 1846         newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
 1847         FILEDESC_LOCK_INIT(&newfdp->fd_fd);
 1848         if (fdp != NULL) {
 1849                 FILEDESC_SLOCK(fdp);
 1850                 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
 1851                 if (newfdp->fd_fd.fd_cdir)
 1852                         VREF(newfdp->fd_fd.fd_cdir);
 1853                 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
 1854                 if (newfdp->fd_fd.fd_rdir)
 1855                         VREF(newfdp->fd_fd.fd_rdir);
 1856                 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
 1857                 if (newfdp->fd_fd.fd_jdir)
 1858                         VREF(newfdp->fd_fd.fd_jdir);
 1859                 FILEDESC_SUNLOCK(fdp);
 1860         }
 1861 
 1862         /* Create the file descriptor table. */
 1863         newfdp->fd_fd.fd_refcnt = 1;
 1864         newfdp->fd_fd.fd_holdcnt = 1;
 1865         newfdp->fd_fd.fd_cmask = CMASK;
 1866         newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
 1867         newfdp->fd_fd.fd_nfiles = NDFILE;
 1868         newfdp->fd_fd.fd_map = newfdp->fd_dmap;
 1869         newfdp->fd_fd.fd_lastfile = -1;
 1870         return (&newfdp->fd_fd);
 1871 }
 1872 
 1873 static struct filedesc *
 1874 fdhold(struct proc *p)
 1875 {
 1876         struct filedesc *fdp;
 1877 
 1878         mtx_lock(&fdesc_mtx);
 1879         fdp = p->p_fd;
 1880         if (fdp != NULL)
 1881                 fdp->fd_holdcnt++;
 1882         mtx_unlock(&fdesc_mtx);
 1883         return (fdp);
 1884 }
 1885 
 1886 static void
 1887 fddrop(struct filedesc *fdp)
 1888 {
 1889         struct filedesc0 *fdp0;
 1890         struct freetable *ft;
 1891         int i;
 1892 
 1893         mtx_lock(&fdesc_mtx);
 1894         i = --fdp->fd_holdcnt;
 1895         mtx_unlock(&fdesc_mtx);
 1896         if (i > 0)
 1897                 return;
 1898 
 1899         FILEDESC_LOCK_DESTROY(fdp);
 1900         fdp0 = (struct filedesc0 *)fdp;
 1901         while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
 1902                 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
 1903                 free(ft->ft_table, M_FILEDESC);
 1904         }
 1905         free(fdp, M_FILEDESC);
 1906 }
 1907 
 1908 /*
 1909  * Share a filedesc structure.
 1910  */
 1911 struct filedesc *
 1912 fdshare(struct filedesc *fdp)
 1913 {
 1914 
 1915         FILEDESC_XLOCK(fdp);
 1916         fdp->fd_refcnt++;
 1917         FILEDESC_XUNLOCK(fdp);
 1918         return (fdp);
 1919 }
 1920 
 1921 /*
 1922  * Unshare a filedesc structure, if necessary by making a copy
 1923  */
 1924 void
 1925 fdunshare(struct thread *td)
 1926 {
 1927         struct filedesc *tmp;
 1928         struct proc *p = td->td_proc;
 1929 
 1930         if (p->p_fd->fd_refcnt == 1)
 1931                 return;
 1932 
 1933         tmp = fdcopy(p->p_fd);
 1934         fdescfree(td);
 1935         p->p_fd = tmp;
 1936 }
 1937 
 1938 /*
 1939  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
 1940  * this is to ease callers, not catch errors.
 1941  */
 1942 struct filedesc *
 1943 fdcopy(struct filedesc *fdp)
 1944 {
 1945         struct filedesc *newfdp;
 1946         struct filedescent *nfde, *ofde;
 1947         int i;
 1948 
 1949         /* Certain daemons might not have file descriptors. */
 1950         if (fdp == NULL)
 1951                 return (NULL);
 1952 
 1953         newfdp = fdinit(fdp);
 1954         FILEDESC_SLOCK(fdp);
 1955         while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
 1956                 FILEDESC_SUNLOCK(fdp);
 1957                 FILEDESC_XLOCK(newfdp);
 1958                 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
 1959                 FILEDESC_XUNLOCK(newfdp);
 1960                 FILEDESC_SLOCK(fdp);
 1961         }
 1962         /* copy all passable descriptors (i.e. not kqueue) */
 1963         newfdp->fd_freefile = -1;
 1964         for (i = 0; i <= fdp->fd_lastfile; ++i) {
 1965                 ofde = &fdp->fd_ofiles[i];
 1966                 if (fdisused(fdp, i) &&
 1967                     (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) &&
 1968                     ofde->fde_file->f_ops != &badfileops) {
 1969                         nfde = &newfdp->fd_ofiles[i];
 1970                         *nfde = *ofde;
 1971                         filecaps_copy(&ofde->fde_caps, &nfde->fde_caps);
 1972                         fhold(nfde->fde_file);
 1973                         newfdp->fd_lastfile = i;
 1974                 } else {
 1975                         if (newfdp->fd_freefile == -1)
 1976                                 newfdp->fd_freefile = i;
 1977                 }
 1978         }
 1979         newfdp->fd_cmask = fdp->fd_cmask;
 1980         FILEDESC_SUNLOCK(fdp);
 1981         FILEDESC_XLOCK(newfdp);
 1982         for (i = 0; i <= newfdp->fd_lastfile; ++i) {
 1983                 if (newfdp->fd_ofiles[i].fde_file != NULL)
 1984                         fdused(newfdp, i);
 1985         }
 1986         if (newfdp->fd_freefile == -1)
 1987                 newfdp->fd_freefile = i;
 1988         FILEDESC_XUNLOCK(newfdp);
 1989         return (newfdp);
 1990 }
 1991 
 1992 /*
 1993  * Release a filedesc structure.
 1994  */
 1995 void
 1996 fdescfree(struct thread *td)
 1997 {
 1998         struct filedesc *fdp;
 1999         int i;
 2000         struct filedesc_to_leader *fdtol;
 2001         struct file *fp;
 2002         struct vnode *cdir, *jdir, *rdir, *vp;
 2003         struct flock lf;
 2004 
 2005         /* Certain daemons might not have file descriptors. */
 2006         fdp = td->td_proc->p_fd;
 2007         if (fdp == NULL)
 2008                 return;
 2009 
 2010 #ifdef RACCT
 2011         PROC_LOCK(td->td_proc);
 2012         racct_set(td->td_proc, RACCT_NOFILE, 0);
 2013         PROC_UNLOCK(td->td_proc);
 2014 #endif
 2015 
 2016         /* Check for special need to clear POSIX style locks */
 2017         fdtol = td->td_proc->p_fdtol;
 2018         if (fdtol != NULL) {
 2019                 FILEDESC_XLOCK(fdp);
 2020                 KASSERT(fdtol->fdl_refcount > 0,
 2021                     ("filedesc_to_refcount botch: fdl_refcount=%d",
 2022                     fdtol->fdl_refcount));
 2023                 if (fdtol->fdl_refcount == 1 &&
 2024                     (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 2025                         for (i = 0; i <= fdp->fd_lastfile; i++) {
 2026                                 fp = fdp->fd_ofiles[i].fde_file;
 2027                                 if (fp == NULL || fp->f_type != DTYPE_VNODE)
 2028                                         continue;
 2029                                 fhold(fp);
 2030                                 FILEDESC_XUNLOCK(fdp);
 2031                                 lf.l_whence = SEEK_SET;
 2032                                 lf.l_start = 0;
 2033                                 lf.l_len = 0;
 2034                                 lf.l_type = F_UNLCK;
 2035                                 vp = fp->f_vnode;
 2036                                 (void) VOP_ADVLOCK(vp,
 2037                                     (caddr_t)td->td_proc->p_leader, F_UNLCK,
 2038                                     &lf, F_POSIX);
 2039                                 FILEDESC_XLOCK(fdp);
 2040                                 fdrop(fp, td);
 2041                         }
 2042                 }
 2043         retry:
 2044                 if (fdtol->fdl_refcount == 1) {
 2045                         if (fdp->fd_holdleaderscount > 0 &&
 2046                             (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 2047                                 /*
 2048                                  * close() or do_dup() has cleared a reference
 2049                                  * in a shared file descriptor table.
 2050                                  */
 2051                                 fdp->fd_holdleaderswakeup = 1;
 2052                                 sx_sleep(&fdp->fd_holdleaderscount,
 2053                                     FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
 2054                                 goto retry;
 2055                         }
 2056                         if (fdtol->fdl_holdcount > 0) {
 2057                                 /*
 2058                                  * Ensure that fdtol->fdl_leader remains
 2059                                  * valid in closef().
 2060                                  */
 2061                                 fdtol->fdl_wakeup = 1;
 2062                                 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
 2063                                     "fdlhold", 0);
 2064                                 goto retry;
 2065                         }
 2066                 }
 2067                 fdtol->fdl_refcount--;
 2068                 if (fdtol->fdl_refcount == 0 &&
 2069                     fdtol->fdl_holdcount == 0) {
 2070                         fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
 2071                         fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
 2072                 } else
 2073                         fdtol = NULL;
 2074                 td->td_proc->p_fdtol = NULL;
 2075                 FILEDESC_XUNLOCK(fdp);
 2076                 if (fdtol != NULL)
 2077                         free(fdtol, M_FILEDESC_TO_LEADER);
 2078         }
 2079 
 2080         mtx_lock(&fdesc_mtx);
 2081         td->td_proc->p_fd = NULL;
 2082         mtx_unlock(&fdesc_mtx);
 2083 
 2084         FILEDESC_XLOCK(fdp);
 2085         i = --fdp->fd_refcnt;
 2086         if (i > 0) {
 2087                 FILEDESC_XUNLOCK(fdp);
 2088                 return;
 2089         }
 2090 
 2091         cdir = fdp->fd_cdir;
 2092         fdp->fd_cdir = NULL;
 2093         rdir = fdp->fd_rdir;
 2094         fdp->fd_rdir = NULL;
 2095         jdir = fdp->fd_jdir;
 2096         fdp->fd_jdir = NULL;
 2097         FILEDESC_XUNLOCK(fdp);
 2098 
 2099         for (i = 0; i <= fdp->fd_lastfile; i++) {
 2100                 fp = fdp->fd_ofiles[i].fde_file;
 2101                 if (fp != NULL) {
 2102                         fdfree_last(fdp, i);
 2103                         (void) closef(fp, td);
 2104                 }
 2105         }
 2106 
 2107         if (fdp->fd_nfiles > NDFILE)
 2108                 free(fdp->fd_ofiles, M_FILEDESC);
 2109         if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
 2110                 free(fdp->fd_map, M_FILEDESC);
 2111 
 2112         if (cdir != NULL)
 2113                 vrele(cdir);
 2114         if (rdir != NULL)
 2115                 vrele(rdir);
 2116         if (jdir != NULL)
 2117                 vrele(jdir);
 2118 
 2119         fddrop(fdp);
 2120 }
 2121 
 2122 /*
 2123  * For setugid programs, we don't want to people to use that setugidness
 2124  * to generate error messages which write to a file which otherwise would
 2125  * otherwise be off-limits to the process.  We check for filesystems where
 2126  * the vnode can change out from under us after execve (like [lin]procfs).
 2127  *
 2128  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
 2129  * sufficient.  We also don't check for setugidness since we know we are.
 2130  */
 2131 static int
 2132 is_unsafe(struct file *fp)
 2133 {
 2134         if (fp->f_type == DTYPE_VNODE) {
 2135                 struct vnode *vp = fp->f_vnode;
 2136 
 2137                 if ((vp->v_vflag & VV_PROCDEP) != 0)
 2138                         return (1);
 2139         }
 2140         return (0);
 2141 }
 2142 
 2143 /*
 2144  * Make this setguid thing safe, if at all possible.
 2145  */
 2146 void
 2147 setugidsafety(struct thread *td)
 2148 {
 2149         struct filedesc *fdp;
 2150         struct file *fp;
 2151         int i;
 2152 
 2153         fdp = td->td_proc->p_fd;
 2154         KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
 2155         FILEDESC_XLOCK(fdp);
 2156         for (i = 0; i <= fdp->fd_lastfile; i++) {
 2157                 if (i > 2)
 2158                         break;
 2159                 fp = fdp->fd_ofiles[i].fde_file;
 2160                 if (fp != NULL && is_unsafe(fp)) {
 2161                         knote_fdclose(td, i);
 2162                         /*
 2163                          * NULL-out descriptor prior to close to avoid
 2164                          * a race while close blocks.
 2165                          */
 2166                         fdfree(fdp, i);
 2167                         FILEDESC_XUNLOCK(fdp);
 2168                         (void) closef(fp, td);
 2169                         FILEDESC_XLOCK(fdp);
 2170                 }
 2171         }
 2172         FILEDESC_XUNLOCK(fdp);
 2173 }
 2174 
 2175 /*
 2176  * If a specific file object occupies a specific file descriptor, close the
 2177  * file descriptor entry and drop a reference on the file object.  This is a
 2178  * convenience function to handle a subsequent error in a function that calls
 2179  * falloc() that handles the race that another thread might have closed the
 2180  * file descriptor out from under the thread creating the file object.
 2181  */
 2182 void
 2183 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
 2184 {
 2185 
 2186         FILEDESC_XLOCK(fdp);
 2187         if (fdp->fd_ofiles[idx].fde_file == fp) {
 2188                 fdfree(fdp, idx);
 2189                 FILEDESC_XUNLOCK(fdp);
 2190                 fdrop(fp, td);
 2191         } else
 2192                 FILEDESC_XUNLOCK(fdp);
 2193 }
 2194 
 2195 /*
 2196  * Close any files on exec?
 2197  */
 2198 void
 2199 fdcloseexec(struct thread *td)
 2200 {
 2201         struct filedesc *fdp;
 2202         struct filedescent *fde;
 2203         struct file *fp;
 2204         int i;
 2205 
 2206         fdp = td->td_proc->p_fd;
 2207         KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
 2208         FILEDESC_XLOCK(fdp);
 2209         for (i = 0; i <= fdp->fd_lastfile; i++) {
 2210                 fde = &fdp->fd_ofiles[i];
 2211                 fp = fde->fde_file;
 2212                 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
 2213                     (fde->fde_flags & UF_EXCLOSE))) {
 2214                         fdfree(fdp, i);
 2215                         (void) closefp(fdp, i, fp, td, 0);
 2216                         /* closefp() drops the FILEDESC lock. */
 2217                         FILEDESC_XLOCK(fdp);
 2218                 }
 2219         }
 2220         FILEDESC_XUNLOCK(fdp);
 2221 }
 2222 
 2223 /*
 2224  * It is unsafe for set[ug]id processes to be started with file
 2225  * descriptors 0..2 closed, as these descriptors are given implicit
 2226  * significance in the Standard C library.  fdcheckstd() will create a
 2227  * descriptor referencing /dev/null for each of stdin, stdout, and
 2228  * stderr that is not already open.
 2229  */
 2230 int
 2231 fdcheckstd(struct thread *td)
 2232 {
 2233         struct filedesc *fdp;
 2234         register_t retval, save;
 2235         int i, error, devnull;
 2236 
 2237         fdp = td->td_proc->p_fd;
 2238         KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
 2239         devnull = -1;
 2240         error = 0;
 2241         for (i = 0; i < 3; i++) {
 2242                 if (fdp->fd_ofiles[i].fde_file != NULL)
 2243                         continue;
 2244                 if (devnull < 0) {
 2245                         save = td->td_retval[0];
 2246                         error = kern_open(td, "/dev/null", UIO_SYSSPACE,
 2247                             O_RDWR, 0);
 2248                         devnull = td->td_retval[0];
 2249                         td->td_retval[0] = save;
 2250                         if (error)
 2251                                 break;
 2252                         KASSERT(devnull == i, ("oof, we didn't get our fd"));
 2253                 } else {
 2254                         error = do_dup(td, DUP_FIXED, devnull, i, &retval);
 2255                         if (error != 0)
 2256                                 break;
 2257                 }
 2258         }
 2259         return (error);
 2260 }
 2261 
 2262 /*
 2263  * Internal form of close.  Decrement reference count on file structure.
 2264  * Note: td may be NULL when closing a file that was being passed in a
 2265  * message.
 2266  *
 2267  * XXXRW: Giant is not required for the caller, but often will be held; this
 2268  * makes it moderately likely the Giant will be recursed in the VFS case.
 2269  */
 2270 int
 2271 closef(struct file *fp, struct thread *td)
 2272 {
 2273         struct vnode *vp;
 2274         struct flock lf;
 2275         struct filedesc_to_leader *fdtol;
 2276         struct filedesc *fdp;
 2277 
 2278         /*
 2279          * POSIX record locking dictates that any close releases ALL
 2280          * locks owned by this process.  This is handled by setting
 2281          * a flag in the unlock to free ONLY locks obeying POSIX
 2282          * semantics, and not to free BSD-style file locks.
 2283          * If the descriptor was in a message, POSIX-style locks
 2284          * aren't passed with the descriptor, and the thread pointer
 2285          * will be NULL.  Callers should be careful only to pass a
 2286          * NULL thread pointer when there really is no owning
 2287          * context that might have locks, or the locks will be
 2288          * leaked.
 2289          */
 2290         if (fp->f_type == DTYPE_VNODE && td != NULL) {
 2291                 vp = fp->f_vnode;
 2292                 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 2293                         lf.l_whence = SEEK_SET;
 2294                         lf.l_start = 0;
 2295                         lf.l_len = 0;
 2296                         lf.l_type = F_UNLCK;
 2297                         (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
 2298                             F_UNLCK, &lf, F_POSIX);
 2299                 }
 2300                 fdtol = td->td_proc->p_fdtol;
 2301                 if (fdtol != NULL) {
 2302                         /*
 2303                          * Handle special case where file descriptor table is
 2304                          * shared between multiple process leaders.
 2305                          */
 2306                         fdp = td->td_proc->p_fd;
 2307                         FILEDESC_XLOCK(fdp);
 2308                         for (fdtol = fdtol->fdl_next;
 2309                              fdtol != td->td_proc->p_fdtol;
 2310                              fdtol = fdtol->fdl_next) {
 2311                                 if ((fdtol->fdl_leader->p_flag &
 2312                                      P_ADVLOCK) == 0)
 2313                                         continue;
 2314                                 fdtol->fdl_holdcount++;
 2315                                 FILEDESC_XUNLOCK(fdp);
 2316                                 lf.l_whence = SEEK_SET;
 2317                                 lf.l_start = 0;
 2318                                 lf.l_len = 0;
 2319                                 lf.l_type = F_UNLCK;
 2320                                 vp = fp->f_vnode;
 2321                                 (void) VOP_ADVLOCK(vp,
 2322                                     (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
 2323                                     F_POSIX);
 2324                                 FILEDESC_XLOCK(fdp);
 2325                                 fdtol->fdl_holdcount--;
 2326                                 if (fdtol->fdl_holdcount == 0 &&
 2327                                     fdtol->fdl_wakeup != 0) {
 2328                                         fdtol->fdl_wakeup = 0;
 2329                                         wakeup(fdtol);
 2330                                 }
 2331                         }
 2332                         FILEDESC_XUNLOCK(fdp);
 2333                 }
 2334         }
 2335         return (fdrop(fp, td));
 2336 }
 2337 
 2338 /*
 2339  * Initialize the file pointer with the specified properties.
 2340  *
 2341  * The ops are set with release semantics to be certain that the flags, type,
 2342  * and data are visible when ops is.  This is to prevent ops methods from being
 2343  * called with bad data.
 2344  */
 2345 void
 2346 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
 2347 {
 2348         fp->f_data = data;
 2349         fp->f_flag = flag;
 2350         fp->f_type = type;
 2351         atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
 2352 }
 2353 
 2354 int
 2355 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
 2356     int needfcntl, struct file **fpp, cap_rights_t *haverightsp)
 2357 {
 2358 #ifdef CAPABILITIES
 2359         struct filedescent fde;
 2360 #endif
 2361         struct file *fp;
 2362         u_int count;
 2363 #ifdef CAPABILITIES
 2364         seq_t seq;
 2365         cap_rights_t haverights;
 2366         int error;
 2367 #endif
 2368 
 2369         /*
 2370          * Avoid reads reordering and then a first access to the
 2371          * fdp->fd_ofiles table which could result in OOB operation.
 2372          */
 2373         if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles))
 2374                 return (EBADF);
 2375         /*
 2376          * Fetch the descriptor locklessly.  We avoid fdrop() races by
 2377          * never raising a refcount above 0.  To accomplish this we have
 2378          * to use a cmpset loop rather than an atomic_add.  The descriptor
 2379          * must be re-verified once we acquire a reference to be certain
 2380          * that the identity is still correct and we did not lose a race
 2381          * due to preemption.
 2382          */
 2383         for (;;) {
 2384 #ifdef CAPABILITIES
 2385                 seq = seq_read(fd_seq(fdp, fd));
 2386                 fde = fdp->fd_ofiles[fd];
 2387                 if (!seq_consistent(fd_seq(fdp, fd), seq)) {
 2388                         cpu_spinwait();
 2389                         continue;
 2390                 }
 2391                 fp = fde.fde_file;
 2392 #else
 2393                 fp = fdp->fd_ofiles[fd].fde_file;
 2394 #endif
 2395                 if (fp == NULL)
 2396                         return (EBADF);
 2397 #ifdef CAPABILITIES
 2398                 haverights = *cap_rights_fde(&fde);
 2399                 if (needrightsp != NULL) {
 2400                         error = cap_check(&haverights, needrightsp);
 2401                         if (error != 0)
 2402                                 return (error);
 2403                         if (cap_rights_is_set(needrightsp, CAP_FCNTL)) {
 2404                                 error = cap_fcntl_check_fde(&fde, needfcntl);
 2405                                 if (error != 0)
 2406                                         return (error);
 2407                         }
 2408                 }
 2409 #endif
 2410                 count = fp->f_count;
 2411                 if (count == 0)
 2412                         continue;
 2413                 /*
 2414                  * Use an acquire barrier to prevent caching of fd_ofiles
 2415                  * so it is refreshed for verification.
 2416                  */
 2417                 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
 2418                         continue;
 2419 #ifdef  CAPABILITIES
 2420                 if (seq_consistent_nomb(fd_seq(fdp, fd), seq))
 2421 #else
 2422                 if (fp == fdp->fd_ofiles[fd].fde_file)
 2423 #endif
 2424                         break;
 2425                 fdrop(fp, curthread);
 2426         }
 2427         *fpp = fp;
 2428         if (haverightsp != NULL) {
 2429 #ifdef CAPABILITIES
 2430                 *haverightsp = haverights;
 2431 #else
 2432                 CAP_ALL(haverightsp);
 2433 #endif
 2434         }
 2435         return (0);
 2436 }
 2437 
 2438 /*
 2439  * Extract the file pointer associated with the specified descriptor for the
 2440  * current user process.
 2441  *
 2442  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
 2443  * returned.
 2444  *
 2445  * File's rights will be checked against the capability rights mask.
 2446  *
 2447  * If an error occured the non-zero error is returned and *fpp is set to
 2448  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
 2449  * responsible for fdrop().
 2450  */
 2451 static __inline int
 2452 _fget(struct thread *td, int fd, struct file **fpp, int flags,
 2453     cap_rights_t *needrightsp, u_char *maxprotp)
 2454 {
 2455         struct filedesc *fdp;
 2456         struct file *fp;
 2457         cap_rights_t haverights, needrights;
 2458         int error;
 2459 
 2460         *fpp = NULL;
 2461         if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
 2462                 return (EBADF);
 2463         if (needrightsp != NULL)
 2464                 needrights = *needrightsp;
 2465         else
 2466                 cap_rights_init(&needrights);
 2467         if (maxprotp != NULL)
 2468                 cap_rights_set(&needrights, CAP_MMAP);
 2469         error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights);
 2470         if (error != 0)
 2471                 return (error);
 2472         if (fp->f_ops == &badfileops) {
 2473                 fdrop(fp, td);
 2474                 return (EBADF);
 2475         }
 2476 
 2477 #ifdef CAPABILITIES
 2478         /*
 2479          * If requested, convert capability rights to access flags.
 2480          */
 2481         if (maxprotp != NULL)
 2482                 *maxprotp = cap_rights_to_vmprot(&haverights);
 2483 #else /* !CAPABILITIES */
 2484         if (maxprotp != NULL)
 2485                 *maxprotp = VM_PROT_ALL;
 2486 #endif /* CAPABILITIES */
 2487 
 2488         /*
 2489          * FREAD and FWRITE failure return EBADF as per POSIX.
 2490          */
 2491         error = 0;
 2492         switch (flags) {
 2493         case FREAD:
 2494         case FWRITE:
 2495                 if ((fp->f_flag & flags) == 0)
 2496                         error = EBADF;
 2497                 break;
 2498         case FEXEC:
 2499                 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
 2500                     ((fp->f_flag & FWRITE) != 0))
 2501                         error = EBADF;
 2502                 break;
 2503         case 0:
 2504                 break;
 2505         default:
 2506                 KASSERT(0, ("wrong flags"));
 2507         }
 2508 
 2509         if (error != 0) {
 2510                 fdrop(fp, td);
 2511                 return (error);
 2512         }
 2513 
 2514         *fpp = fp;
 2515         return (0);
 2516 }
 2517 
 2518 int
 2519 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
 2520 {
 2521 
 2522         return(_fget(td, fd, fpp, 0, rightsp, NULL));
 2523 }
 2524 
 2525 int
 2526 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
 2527     struct file **fpp)
 2528 {
 2529 
 2530         return (_fget(td, fd, fpp, 0, rightsp, maxprotp));
 2531 }
 2532 
 2533 int
 2534 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
 2535 {
 2536 
 2537         return(_fget(td, fd, fpp, FREAD, rightsp, NULL));
 2538 }
 2539 
 2540 int
 2541 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
 2542 {
 2543 
 2544         return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
 2545 }
 2546 
 2547 /*
 2548  * Like fget() but loads the underlying vnode, or returns an error if the
 2549  * descriptor does not represent a vnode.  Note that pipes use vnodes but
 2550  * never have VM objects.  The returned vnode will be vref()'d.
 2551  *
 2552  * XXX: what about the unused flags ?
 2553  */
 2554 static __inline int
 2555 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
 2556     struct vnode **vpp)
 2557 {
 2558         struct file *fp;
 2559         int error;
 2560 
 2561         *vpp = NULL;
 2562         error = _fget(td, fd, &fp, flags, needrightsp, NULL);
 2563         if (error != 0)
 2564                 return (error);
 2565         if (fp->f_vnode == NULL) {
 2566                 error = EINVAL;
 2567         } else {
 2568                 *vpp = fp->f_vnode;
 2569                 vref(*vpp);
 2570         }
 2571         fdrop(fp, td);
 2572 
 2573         return (error);
 2574 }
 2575 
 2576 int
 2577 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
 2578 {
 2579 
 2580         return (_fgetvp(td, fd, 0, rightsp, vpp));
 2581 }
 2582 
 2583 int
 2584 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
 2585     struct filecaps *havecaps, struct vnode **vpp)
 2586 {
 2587         struct filedesc *fdp;
 2588         struct file *fp;
 2589 #ifdef CAPABILITIES
 2590         int error;
 2591 #endif
 2592 
 2593         if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
 2594                 return (EBADF);
 2595 
 2596         fp = fget_locked(fdp, fd);
 2597         if (fp == NULL || fp->f_ops == &badfileops)
 2598                 return (EBADF);
 2599 
 2600 #ifdef CAPABILITIES
 2601         if (needrightsp != NULL) {
 2602                 error = cap_check(cap_rights(fdp, fd), needrightsp);
 2603                 if (error != 0)
 2604                         return (error);
 2605         }
 2606 #endif
 2607 
 2608         if (fp->f_vnode == NULL)
 2609                 return (EINVAL);
 2610 
 2611         *vpp = fp->f_vnode;
 2612         vref(*vpp);
 2613         filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps);
 2614 
 2615         return (0);
 2616 }
 2617 
 2618 int
 2619 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
 2620 {
 2621 
 2622         return (_fgetvp(td, fd, FREAD, rightsp, vpp));
 2623 }
 2624 
 2625 int
 2626 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
 2627 {
 2628 
 2629         return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
 2630 }
 2631 
 2632 #ifdef notyet
 2633 int
 2634 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
 2635     struct vnode **vpp)
 2636 {
 2637 
 2638         return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
 2639 }
 2640 #endif
 2641 
 2642 /*
 2643  * Like fget() but loads the underlying socket, or returns an error if the
 2644  * descriptor does not represent a socket.
 2645  *
 2646  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
 2647  * in the future.
 2648  *
 2649  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
 2650  * on their file descriptor reference to prevent the socket from being free'd
 2651  * during use.
 2652  */
 2653 int
 2654 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp,
 2655     u_int *fflagp)
 2656 {
 2657         struct file *fp;
 2658         int error;
 2659 
 2660         *spp = NULL;
 2661         if (fflagp != NULL)
 2662                 *fflagp = 0;
 2663         if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0)
 2664                 return (error);
 2665         if (fp->f_type != DTYPE_SOCKET) {
 2666                 error = ENOTSOCK;
 2667         } else {
 2668                 *spp = fp->f_data;
 2669                 if (fflagp)
 2670                         *fflagp = fp->f_flag;
 2671                 SOCK_LOCK(*spp);
 2672                 soref(*spp);
 2673                 SOCK_UNLOCK(*spp);
 2674         }
 2675         fdrop(fp, td);
 2676 
 2677         return (error);
 2678 }
 2679 
 2680 /*
 2681  * Drop the reference count on the socket and XXX release the SX lock in the
 2682  * future.  The last reference closes the socket.
 2683  *
 2684  * Note: fputsock() is deprecated, see comment for fgetsock().
 2685  */
 2686 void
 2687 fputsock(struct socket *so)
 2688 {
 2689 
 2690         ACCEPT_LOCK();
 2691         SOCK_LOCK(so);
 2692         CURVNET_SET(so->so_vnet);
 2693         sorele(so);
 2694         CURVNET_RESTORE();
 2695 }
 2696 
 2697 /*
 2698  * Handle the last reference to a file being closed.
 2699  */
 2700 int
 2701 _fdrop(struct file *fp, struct thread *td)
 2702 {
 2703         int error;
 2704 
 2705         error = 0;
 2706         if (fp->f_count != 0)
 2707                 panic("fdrop: count %d", fp->f_count);
 2708         if (fp->f_ops != &badfileops)
 2709                 error = fo_close(fp, td);
 2710         atomic_subtract_int(&openfiles, 1);
 2711         crfree(fp->f_cred);
 2712         free(fp->f_advice, M_FADVISE);
 2713         uma_zfree(file_zone, fp);
 2714 
 2715         return (error);
 2716 }
 2717 
 2718 /*
 2719  * Apply an advisory lock on a file descriptor.
 2720  *
 2721  * Just attempt to get a record lock of the requested type on the entire file
 2722  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
 2723  */
 2724 #ifndef _SYS_SYSPROTO_H_
 2725 struct flock_args {
 2726         int     fd;
 2727         int     how;
 2728 };
 2729 #endif
 2730 /* ARGSUSED */
 2731 int
 2732 sys_flock(struct thread *td, struct flock_args *uap)
 2733 {
 2734         struct file *fp;
 2735         struct vnode *vp;
 2736         struct flock lf;
 2737         cap_rights_t rights;
 2738         int error;
 2739 
 2740         error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
 2741         if (error != 0)
 2742                 return (error);
 2743         if (fp->f_type != DTYPE_VNODE) {
 2744                 fdrop(fp, td);
 2745                 return (EOPNOTSUPP);
 2746         }
 2747 
 2748         vp = fp->f_vnode;
 2749         lf.l_whence = SEEK_SET;
 2750         lf.l_start = 0;
 2751         lf.l_len = 0;
 2752         if (uap->how & LOCK_UN) {
 2753                 lf.l_type = F_UNLCK;
 2754                 atomic_clear_int(&fp->f_flag, FHASLOCK);
 2755                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
 2756                 goto done2;
 2757         }
 2758         if (uap->how & LOCK_EX)
 2759                 lf.l_type = F_WRLCK;
 2760         else if (uap->how & LOCK_SH)
 2761                 lf.l_type = F_RDLCK;
 2762         else {
 2763                 error = EBADF;
 2764                 goto done2;
 2765         }
 2766         atomic_set_int(&fp->f_flag, FHASLOCK);
 2767         error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
 2768             (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
 2769 done2:
 2770         fdrop(fp, td);
 2771         return (error);
 2772 }
 2773 /*
 2774  * Duplicate the specified descriptor to a free descriptor.
 2775  */
 2776 int
 2777 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
 2778     int openerror, int *indxp)
 2779 {
 2780         struct filedescent *newfde, *oldfde;
 2781         struct file *fp;
 2782         int error, indx;
 2783 
 2784         KASSERT(openerror == ENODEV || openerror == ENXIO,
 2785             ("unexpected error %d in %s", openerror, __func__));
 2786 
 2787         /*
 2788          * If the to-be-dup'd fd number is greater than the allowed number
 2789          * of file descriptors, or the fd to be dup'd has already been
 2790          * closed, then reject.
 2791          */
 2792         FILEDESC_XLOCK(fdp);
 2793         if ((fp = fget_locked(fdp, dfd)) == NULL) {
 2794                 FILEDESC_XUNLOCK(fdp);
 2795                 return (EBADF);
 2796         }
 2797 
 2798         error = fdalloc(td, 0, &indx);
 2799         if (error != 0) {
 2800                 FILEDESC_XUNLOCK(fdp);
 2801                 return (error);
 2802         }
 2803 
 2804         /*
 2805          * There are two cases of interest here.
 2806          *
 2807          * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
 2808          *
 2809          * For ENXIO steal away the file structure from (dfd) and store it in
 2810          * (indx).  (dfd) is effectively closed by this operation.
 2811          */
 2812         switch (openerror) {
 2813         case ENODEV:
 2814                 /*
 2815                  * Check that the mode the file is being opened for is a
 2816                  * subset of the mode of the existing descriptor.
 2817                  */
 2818                 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
 2819                         fdunused(fdp, indx);
 2820                         FILEDESC_XUNLOCK(fdp);
 2821                         return (EACCES);
 2822                 }
 2823                 fhold(fp);
 2824                 newfde = &fdp->fd_ofiles[indx];
 2825                 oldfde = &fdp->fd_ofiles[dfd];
 2826 #ifdef CAPABILITIES
 2827                 seq_write_begin(&newfde->fde_seq);
 2828 #endif
 2829                 memcpy(newfde, oldfde, fde_change_size);
 2830                 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
 2831 #ifdef CAPABILITIES
 2832                 seq_write_end(&newfde->fde_seq);
 2833 #endif
 2834                 break;
 2835         case ENXIO:
 2836                 /*
 2837                  * Steal away the file pointer from dfd and stuff it into indx.
 2838                  */
 2839                 newfde = &fdp->fd_ofiles[indx];
 2840                 oldfde = &fdp->fd_ofiles[dfd];
 2841 #ifdef CAPABILITIES
 2842                 seq_write_begin(&newfde->fde_seq);
 2843 #endif
 2844                 memcpy(newfde, oldfde, fde_change_size);
 2845                 bzero(oldfde, fde_change_size);
 2846                 fdunused(fdp, dfd);
 2847 #ifdef CAPABILITIES
 2848                 seq_write_end(&newfde->fde_seq);
 2849 #endif
 2850                 break;
 2851         }
 2852         FILEDESC_XUNLOCK(fdp);
 2853         *indxp = indx;
 2854         return (0);
 2855 }
 2856 
 2857 /*
 2858  * Scan all active processes and prisons to see if any of them have a current
 2859  * or root directory of `olddp'. If so, replace them with the new mount point.
 2860  */
 2861 void
 2862 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
 2863 {
 2864         struct filedesc *fdp;
 2865         struct prison *pr;
 2866         struct proc *p;
 2867         int nrele;
 2868 
 2869         if (vrefcnt(olddp) == 1)
 2870                 return;
 2871         nrele = 0;
 2872         sx_slock(&allproc_lock);
 2873         FOREACH_PROC_IN_SYSTEM(p) {
 2874                 fdp = fdhold(p);
 2875                 if (fdp == NULL)
 2876                         continue;
 2877                 FILEDESC_XLOCK(fdp);
 2878                 if (fdp->fd_cdir == olddp) {
 2879                         vref(newdp);
 2880                         fdp->fd_cdir = newdp;
 2881                         nrele++;
 2882                 }
 2883                 if (fdp->fd_rdir == olddp) {
 2884                         vref(newdp);
 2885                         fdp->fd_rdir = newdp;
 2886                         nrele++;
 2887                 }
 2888                 if (fdp->fd_jdir == olddp) {
 2889                         vref(newdp);
 2890                         fdp->fd_jdir = newdp;
 2891                         nrele++;
 2892                 }
 2893                 FILEDESC_XUNLOCK(fdp);
 2894                 fddrop(fdp);
 2895         }
 2896         sx_sunlock(&allproc_lock);
 2897         if (rootvnode == olddp) {
 2898                 vref(newdp);
 2899                 rootvnode = newdp;
 2900                 nrele++;
 2901         }
 2902         mtx_lock(&prison0.pr_mtx);
 2903         if (prison0.pr_root == olddp) {
 2904                 vref(newdp);
 2905                 prison0.pr_root = newdp;
 2906                 nrele++;
 2907         }
 2908         mtx_unlock(&prison0.pr_mtx);
 2909         sx_slock(&allprison_lock);
 2910         TAILQ_FOREACH(pr, &allprison, pr_list) {
 2911                 mtx_lock(&pr->pr_mtx);
 2912                 if (pr->pr_root == olddp) {
 2913                         vref(newdp);
 2914                         pr->pr_root = newdp;
 2915                         nrele++;
 2916                 }
 2917                 mtx_unlock(&pr->pr_mtx);
 2918         }
 2919         sx_sunlock(&allprison_lock);
 2920         while (nrele--)
 2921                 vrele(olddp);
 2922 }
 2923 
 2924 struct filedesc_to_leader *
 2925 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
 2926 {
 2927         struct filedesc_to_leader *fdtol;
 2928 
 2929         fdtol = malloc(sizeof(struct filedesc_to_leader),
 2930                M_FILEDESC_TO_LEADER,
 2931                M_WAITOK);
 2932         fdtol->fdl_refcount = 1;
 2933         fdtol->fdl_holdcount = 0;
 2934         fdtol->fdl_wakeup = 0;
 2935         fdtol->fdl_leader = leader;
 2936         if (old != NULL) {
 2937                 FILEDESC_XLOCK(fdp);
 2938                 fdtol->fdl_next = old->fdl_next;
 2939                 fdtol->fdl_prev = old;
 2940                 old->fdl_next = fdtol;
 2941                 fdtol->fdl_next->fdl_prev = fdtol;
 2942                 FILEDESC_XUNLOCK(fdp);
 2943         } else {
 2944                 fdtol->fdl_next = fdtol;
 2945                 fdtol->fdl_prev = fdtol;
 2946         }
 2947         return (fdtol);
 2948 }
 2949 
 2950 /*
 2951  * Get file structures globally.
 2952  */
 2953 static int
 2954 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
 2955 {
 2956         struct xfile xf;
 2957         struct filedesc *fdp;
 2958         struct file *fp;
 2959         struct proc *p;
 2960         int error, n;
 2961 
 2962         error = sysctl_wire_old_buffer(req, 0);
 2963         if (error != 0)
 2964                 return (error);
 2965         if (req->oldptr == NULL) {
 2966                 n = 0;
 2967                 sx_slock(&allproc_lock);
 2968                 FOREACH_PROC_IN_SYSTEM(p) {
 2969                         if (p->p_state == PRS_NEW)
 2970                                 continue;
 2971                         fdp = fdhold(p);
 2972                         if (fdp == NULL)
 2973                                 continue;
 2974                         /* overestimates sparse tables. */
 2975                         if (fdp->fd_lastfile > 0)
 2976                                 n += fdp->fd_lastfile;
 2977                         fddrop(fdp);
 2978                 }
 2979                 sx_sunlock(&allproc_lock);
 2980                 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
 2981         }
 2982         error = 0;
 2983         bzero(&xf, sizeof(xf));
 2984         xf.xf_size = sizeof(xf);
 2985         sx_slock(&allproc_lock);
 2986         FOREACH_PROC_IN_SYSTEM(p) {
 2987                 PROC_LOCK(p);
 2988                 if (p->p_state == PRS_NEW) {
 2989                         PROC_UNLOCK(p);
 2990                         continue;
 2991                 }
 2992                 if (p_cansee(req->td, p) != 0) {
 2993                         PROC_UNLOCK(p);
 2994                         continue;
 2995                 }
 2996                 xf.xf_pid = p->p_pid;
 2997                 xf.xf_uid = p->p_ucred->cr_uid;
 2998                 PROC_UNLOCK(p);
 2999                 fdp = fdhold(p);
 3000                 if (fdp == NULL)
 3001                         continue;
 3002                 FILEDESC_SLOCK(fdp);
 3003                 for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
 3004                         if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
 3005                                 continue;
 3006                         xf.xf_fd = n;
 3007                         xf.xf_file = fp;
 3008                         xf.xf_data = fp->f_data;
 3009                         xf.xf_vnode = fp->f_vnode;
 3010                         xf.xf_type = fp->f_type;
 3011                         xf.xf_count = fp->f_count;
 3012                         xf.xf_msgcount = 0;
 3013                         xf.xf_offset = foffset_get(fp);
 3014                         xf.xf_flag = fp->f_flag;
 3015                         error = SYSCTL_OUT(req, &xf, sizeof(xf));
 3016                         if (error)
 3017                                 break;
 3018                 }
 3019                 FILEDESC_SUNLOCK(fdp);
 3020                 fddrop(fdp);
 3021                 if (error)
 3022                         break;
 3023         }
 3024         sx_sunlock(&allproc_lock);
 3025         return (error);
 3026 }
 3027 
 3028 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
 3029     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
 3030 
 3031 #ifdef KINFO_OFILE_SIZE
 3032 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
 3033 #endif
 3034 
 3035 #ifdef COMPAT_FREEBSD7
 3036 static int
 3037 export_vnode_for_osysctl(struct vnode *vp, int type,
 3038     struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
 3039 {
 3040         int error;
 3041         char *fullpath, *freepath;
 3042 
 3043         bzero(kif, sizeof(*kif));
 3044         kif->kf_structsize = sizeof(*kif);
 3045 
 3046         vref(vp);
 3047         kif->kf_fd = type;
 3048         kif->kf_type = KF_TYPE_VNODE;
 3049         /* This function only handles directories. */
 3050         if (vp->v_type != VDIR) {
 3051                 vrele(vp);
 3052                 return (ENOTDIR);
 3053         }
 3054         kif->kf_vnode_type = KF_VTYPE_VDIR;
 3055 
 3056         /*
 3057          * This is not a true file descriptor, so we set a bogus refcount
 3058          * and offset to indicate these fields should be ignored.
 3059          */
 3060         kif->kf_ref_count = -1;
 3061         kif->kf_offset = -1;
 3062 
 3063         freepath = NULL;
 3064         fullpath = "-";
 3065         FILEDESC_SUNLOCK(fdp);
 3066         vn_fullpath(curthread, vp, &fullpath, &freepath);
 3067         vrele(vp);
 3068         strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
 3069         if (freepath != NULL)
 3070                 free(freepath, M_TEMP);
 3071         error = SYSCTL_OUT(req, kif, sizeof(*kif));
 3072         FILEDESC_SLOCK(fdp);
 3073         return (error);
 3074 }
 3075 
 3076 /*
 3077  * Get per-process file descriptors for use by procstat(1), et al.
 3078  */
 3079 static int
 3080 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
 3081 {
 3082         char *fullpath, *freepath;
 3083         struct kinfo_ofile *kif;
 3084         struct filedesc *fdp;
 3085         int error, i, *name;
 3086         struct shmfd *shmfd;
 3087         struct socket *so;
 3088         struct vnode *vp;
 3089         struct ksem *ks;
 3090         struct file *fp;
 3091         struct proc *p;
 3092         struct tty *tp;
 3093 
 3094         name = (int *)arg1;
 3095         error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
 3096         if (error != 0)
 3097                 return (error);
 3098         fdp = fdhold(p);
 3099         PROC_UNLOCK(p);
 3100         if (fdp == NULL)
 3101                 return (ENOENT);
 3102         kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
 3103         FILEDESC_SLOCK(fdp);
 3104         if (fdp->fd_cdir != NULL)
 3105                 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
 3106                                 fdp, req);
 3107         if (fdp->fd_rdir != NULL)
 3108                 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
 3109                                 fdp, req);
 3110         if (fdp->fd_jdir != NULL)
 3111                 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
 3112                                 fdp, req);
 3113         for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
 3114                 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
 3115                         continue;
 3116                 bzero(kif, sizeof(*kif));
 3117                 kif->kf_structsize = sizeof(*kif);
 3118                 ks = NULL;
 3119                 vp = NULL;
 3120                 so = NULL;
 3121                 tp = NULL;
 3122                 shmfd = NULL;
 3123                 kif->kf_fd = i;
 3124 
 3125                 switch (fp->f_type) {
 3126                 case DTYPE_VNODE:
 3127                         kif->kf_type = KF_TYPE_VNODE;
 3128                         vp = fp->f_vnode;
 3129                         break;
 3130 
 3131                 case DTYPE_SOCKET:
 3132                         kif->kf_type = KF_TYPE_SOCKET;
 3133                         so = fp->f_data;
 3134                         break;
 3135 
 3136                 case DTYPE_PIPE:
 3137                         kif->kf_type = KF_TYPE_PIPE;
 3138                         break;
 3139 
 3140                 case DTYPE_FIFO:
 3141                         kif->kf_type = KF_TYPE_FIFO;
 3142                         vp = fp->f_vnode;
 3143                         break;
 3144 
 3145                 case DTYPE_KQUEUE:
 3146                         kif->kf_type = KF_TYPE_KQUEUE;
 3147                         break;
 3148 
 3149                 case DTYPE_CRYPTO:
 3150                         kif->kf_type = KF_TYPE_CRYPTO;
 3151                         break;
 3152 
 3153                 case DTYPE_MQUEUE:
 3154                         kif->kf_type = KF_TYPE_MQUEUE;
 3155                         break;
 3156 
 3157                 case DTYPE_SHM:
 3158                         kif->kf_type = KF_TYPE_SHM;
 3159                         shmfd = fp->f_data;
 3160                         break;
 3161 
 3162                 case DTYPE_SEM:
 3163                         kif->kf_type = KF_TYPE_SEM;
 3164                         ks = fp->f_data;
 3165                         break;
 3166 
 3167                 case DTYPE_PTS:
 3168                         kif->kf_type = KF_TYPE_PTS;
 3169                         tp = fp->f_data;
 3170                         break;
 3171 
 3172 #ifdef PROCDESC
 3173                 case DTYPE_PROCDESC:
 3174                         kif->kf_type = KF_TYPE_PROCDESC;
 3175                         break;
 3176 #endif
 3177 
 3178                 default:
 3179                         kif->kf_type = KF_TYPE_UNKNOWN;
 3180                         break;
 3181                 }
 3182                 kif->kf_ref_count = fp->f_count;
 3183                 if (fp->f_flag & FREAD)
 3184                         kif->kf_flags |= KF_FLAG_READ;
 3185                 if (fp->f_flag & FWRITE)
 3186                         kif->kf_flags |= KF_FLAG_WRITE;
 3187                 if (fp->f_flag & FAPPEND)
 3188                         kif->kf_flags |= KF_FLAG_APPEND;
 3189                 if (fp->f_flag & FASYNC)
 3190                         kif->kf_flags |= KF_FLAG_ASYNC;
 3191                 if (fp->f_flag & FFSYNC)
 3192                         kif->kf_flags |= KF_FLAG_FSYNC;
 3193                 if (fp->f_flag & FNONBLOCK)
 3194                         kif->kf_flags |= KF_FLAG_NONBLOCK;
 3195                 if (fp->f_flag & O_DIRECT)
 3196                         kif->kf_flags |= KF_FLAG_DIRECT;
 3197                 if (fp->f_flag & FHASLOCK)
 3198                         kif->kf_flags |= KF_FLAG_HASLOCK;
 3199                 kif->kf_offset = foffset_get(fp);
 3200                 if (vp != NULL) {
 3201                         vref(vp);
 3202                         switch (vp->v_type) {
 3203                         case VNON:
 3204                                 kif->kf_vnode_type = KF_VTYPE_VNON;
 3205                                 break;
 3206                         case VREG:
 3207                                 kif->kf_vnode_type = KF_VTYPE_VREG;
 3208                                 break;
 3209                         case VDIR:
 3210                                 kif->kf_vnode_type = KF_VTYPE_VDIR;
 3211                                 break;
 3212                         case VBLK:
 3213                                 kif->kf_vnode_type = KF_VTYPE_VBLK;
 3214                                 break;
 3215                         case VCHR:
 3216                                 kif->kf_vnode_type = KF_VTYPE_VCHR;
 3217                                 break;
 3218                         case VLNK:
 3219                                 kif->kf_vnode_type = KF_VTYPE_VLNK;
 3220                                 break;
 3221                         case VSOCK:
 3222                                 kif->kf_vnode_type = KF_VTYPE_VSOCK;
 3223                                 break;
 3224                         case VFIFO:
 3225                                 kif->kf_vnode_type = KF_VTYPE_VFIFO;
 3226                                 break;
 3227                         case VBAD:
 3228                                 kif->kf_vnode_type = KF_VTYPE_VBAD;
 3229                                 break;
 3230                         default:
 3231                                 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
 3232                                 break;
 3233                         }
 3234                         /*
 3235                          * It is OK to drop the filedesc lock here as we will
 3236                          * re-validate and re-evaluate its properties when
 3237                          * the loop continues.
 3238                          */
 3239                         freepath = NULL;
 3240                         fullpath = "-";
 3241                         FILEDESC_SUNLOCK(fdp);
 3242                         vn_fullpath(curthread, vp, &fullpath, &freepath);
 3243                         vrele(vp);
 3244                         strlcpy(kif->kf_path, fullpath,
 3245                             sizeof(kif->kf_path));
 3246                         if (freepath != NULL)
 3247                                 free(freepath, M_TEMP);
 3248                         FILEDESC_SLOCK(fdp);
 3249                 }
 3250                 if (so != NULL) {
 3251                         struct sockaddr *sa;
 3252 
 3253                         if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
 3254                             == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
 3255                                 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
 3256                                 free(sa, M_SONAME);
 3257                         }
 3258                         if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
 3259                             == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
 3260                                 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
 3261                                 free(sa, M_SONAME);
 3262                         }
 3263                         kif->kf_sock_domain =
 3264                             so->so_proto->pr_domain->dom_family;
 3265                         kif->kf_sock_type = so->so_type;
 3266                         kif->kf_sock_protocol = so->so_proto->pr_protocol;
 3267                 }
 3268                 if (tp != NULL) {
 3269                         strlcpy(kif->kf_path, tty_devname(tp),
 3270                             sizeof(kif->kf_path));
 3271                 }
 3272                 if (shmfd != NULL)
 3273                         shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path));
 3274                 if (ks != NULL && ksem_info != NULL)
 3275                         ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL);
 3276                 error = SYSCTL_OUT(req, kif, sizeof(*kif));
 3277                 if (error)
 3278                         break;
 3279         }
 3280         FILEDESC_SUNLOCK(fdp);
 3281         fddrop(fdp);
 3282         free(kif, M_TEMP);
 3283         return (0);
 3284 }
 3285 
 3286 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
 3287     CTLFLAG_RD||CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
 3288     "Process ofiledesc entries");
 3289 #endif  /* COMPAT_FREEBSD7 */
 3290 
 3291 #ifdef KINFO_FILE_SIZE
 3292 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
 3293 #endif
 3294 
 3295 struct export_fd_buf {
 3296         struct filedesc         *fdp;
 3297         struct sbuf             *sb;
 3298         ssize_t                 remainder;
 3299         struct kinfo_file       kif;
 3300 };
 3301 
 3302 static int
 3303 export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt,
 3304     int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf)
 3305 {
 3306         struct {
 3307                 int     fflag;
 3308                 int     kf_fflag;
 3309         } fflags_table[] = {
 3310                 { FAPPEND, KF_FLAG_APPEND },
 3311                 { FASYNC, KF_FLAG_ASYNC },
 3312                 { FFSYNC, KF_FLAG_FSYNC },
 3313                 { FHASLOCK, KF_FLAG_HASLOCK },
 3314                 { FNONBLOCK, KF_FLAG_NONBLOCK },
 3315                 { FREAD, KF_FLAG_READ },
 3316                 { FWRITE, KF_FLAG_WRITE },
 3317                 { O_CREAT, KF_FLAG_CREAT },
 3318                 { O_DIRECT, KF_FLAG_DIRECT },
 3319                 { O_EXCL, KF_FLAG_EXCL },
 3320                 { O_EXEC, KF_FLAG_EXEC },
 3321                 { O_EXLOCK, KF_FLAG_EXLOCK },
 3322                 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
 3323                 { O_SHLOCK, KF_FLAG_SHLOCK },
 3324                 { O_TRUNC, KF_FLAG_TRUNC }
 3325         };
 3326 #define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table))
 3327         struct kinfo_file *kif;
 3328         struct vnode *vp;
 3329         int error, locked;
 3330         unsigned int i;
 3331 
 3332         if (efbuf->remainder == 0)
 3333                 return (0);
 3334         kif = &efbuf->kif;
 3335         bzero(kif, sizeof(*kif));
 3336         locked = efbuf->fdp != NULL;
 3337         switch (type) {
 3338         case KF_TYPE_FIFO:
 3339         case KF_TYPE_VNODE:
 3340                 if (locked) {
 3341                         FILEDESC_SUNLOCK(efbuf->fdp);
 3342                         locked = 0;
 3343                 }
 3344                 vp = (struct vnode *)data;
 3345                 error = fill_vnode_info(vp, kif);
 3346                 vrele(vp);
 3347                 break;
 3348         case KF_TYPE_SOCKET:
 3349                 error = fill_socket_info((struct socket *)data, kif);
 3350                 break;
 3351         case KF_TYPE_PIPE:
 3352                 error = fill_pipe_info((struct pipe *)data, kif);
 3353                 break;
 3354         case KF_TYPE_PTS:
 3355                 error = fill_pts_info((struct tty *)data, kif);
 3356                 break;
 3357         case KF_TYPE_PROCDESC:
 3358                 error = fill_procdesc_info((struct procdesc *)data, kif);
 3359                 break;
 3360         case KF_TYPE_SEM:
 3361                 error = fill_sem_info((struct file *)data, kif);
 3362                 break;
 3363         case KF_TYPE_SHM:
 3364                 error = fill_shm_info((struct file *)data, kif);
 3365                 break;
 3366         default:
 3367                 error = 0;
 3368         }
 3369         if (error == 0)
 3370                 kif->kf_status |= KF_ATTR_VALID;
 3371 
 3372         /*
 3373          * Translate file access flags.
 3374          */
 3375         for (i = 0; i < NFFLAGS; i++)
 3376                 if (fflags & fflags_table[i].fflag)
 3377                         kif->kf_flags |=  fflags_table[i].kf_fflag;
 3378         if (rightsp != NULL)
 3379                 kif->kf_cap_rights = *rightsp;
 3380         else
 3381                 cap_rights_init(&kif->kf_cap_rights);
 3382         kif->kf_fd = fd;
 3383         kif->kf_type = type;
 3384         kif->kf_ref_count = refcnt;
 3385         kif->kf_offset = offset;
 3386         /* Pack record size down */
 3387         kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
 3388             strlen(kif->kf_path) + 1;
 3389         kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
 3390         if (efbuf->remainder != -1) {
 3391                 if (efbuf->remainder < kif->kf_structsize) {
 3392                         /* Terminate export. */
 3393                         efbuf->remainder = 0;
 3394                         if (efbuf->fdp != NULL && !locked)
 3395                                 FILEDESC_SLOCK(efbuf->fdp);
 3396                         return (0);
 3397                 }
 3398                 efbuf->remainder -= kif->kf_structsize;
 3399         }
 3400         if (locked)
 3401                 FILEDESC_SUNLOCK(efbuf->fdp);
 3402         error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize);
 3403         if (efbuf->fdp != NULL)
 3404                 FILEDESC_SLOCK(efbuf->fdp);
 3405         return (error);
 3406 }
 3407 
 3408 /*
 3409  * Store a process file descriptor information to sbuf.
 3410  *
 3411  * Takes a locked proc as argument, and returns with the proc unlocked.
 3412  */
 3413 int
 3414 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
 3415 {
 3416         struct file *fp;
 3417         struct filedesc *fdp;
 3418         struct export_fd_buf *efbuf;
 3419         struct vnode *cttyvp, *textvp, *tracevp;
 3420         int64_t offset;
 3421         void *data;
 3422         int error, i;
 3423         int type, refcnt, fflags;
 3424         cap_rights_t rights;
 3425 
 3426         PROC_LOCK_ASSERT(p, MA_OWNED);
 3427 
 3428         /* ktrace vnode */
 3429         tracevp = p->p_tracevp;
 3430         if (tracevp != NULL)
 3431                 vref(tracevp);
 3432         /* text vnode */
 3433         textvp = p->p_textvp;
 3434         if (textvp != NULL)
 3435                 vref(textvp);
 3436         /* Controlling tty. */
 3437         cttyvp = NULL;
 3438         if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
 3439                 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
 3440                 if (cttyvp != NULL)
 3441                         vref(cttyvp);
 3442         }
 3443         fdp = fdhold(p);
 3444         PROC_UNLOCK(p);
 3445         efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
 3446         efbuf->fdp = NULL;
 3447         efbuf->sb = sb;
 3448         efbuf->remainder = maxlen;
 3449         if (tracevp != NULL)
 3450                 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE,
 3451                     FREAD | FWRITE, -1, -1, NULL, efbuf);
 3452         if (textvp != NULL)
 3453                 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT,
 3454                     FREAD, -1, -1, NULL, efbuf);
 3455         if (cttyvp != NULL)
 3456                 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY,
 3457                     FREAD | FWRITE, -1, -1, NULL, efbuf);
 3458         error = 0;
 3459         if (fdp == NULL)
 3460                 goto fail;
 3461         efbuf->fdp = fdp;
 3462         FILEDESC_SLOCK(fdp);
 3463         /* working directory */
 3464         if (fdp->fd_cdir != NULL) {
 3465                 vref(fdp->fd_cdir);
 3466                 data = fdp->fd_cdir;
 3467                 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD,
 3468                     FREAD, -1, -1, NULL, efbuf);
 3469         }
 3470         /* root directory */
 3471         if (fdp->fd_rdir != NULL) {
 3472                 vref(fdp->fd_rdir);
 3473                 data = fdp->fd_rdir;
 3474                 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT,
 3475                     FREAD, -1, -1, NULL, efbuf);
 3476         }
 3477         /* jail directory */
 3478         if (fdp->fd_jdir != NULL) {
 3479                 vref(fdp->fd_jdir);
 3480                 data = fdp->fd_jdir;
 3481                 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL,
 3482                     FREAD, -1, -1, NULL, efbuf);
 3483         }
 3484         for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
 3485                 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
 3486                         continue;
 3487                 data = NULL;
 3488 #ifdef CAPABILITIES
 3489                 rights = *cap_rights(fdp, i);
 3490 #else /* !CAPABILITIES */
 3491                 cap_rights_init(&rights);
 3492 #endif
 3493                 switch (fp->f_type) {
 3494                 case DTYPE_VNODE:
 3495                         type = KF_TYPE_VNODE;
 3496                         vref(fp->f_vnode);
 3497                         data = fp->f_vnode;
 3498                         break;
 3499 
 3500                 case DTYPE_SOCKET:
 3501                         type = KF_TYPE_SOCKET;
 3502                         data = fp->f_data;
 3503                         break;
 3504 
 3505                 case DTYPE_PIPE:
 3506                         type = KF_TYPE_PIPE;
 3507                         data = fp->f_data;
 3508                         break;
 3509 
 3510                 case DTYPE_FIFO:
 3511                         type = KF_TYPE_FIFO;
 3512                         vref(fp->f_vnode);
 3513                         data = fp->f_vnode;
 3514                         break;
 3515 
 3516                 case DTYPE_KQUEUE:
 3517                         type = KF_TYPE_KQUEUE;
 3518                         break;
 3519 
 3520                 case DTYPE_CRYPTO:
 3521                         type = KF_TYPE_CRYPTO;
 3522                         break;
 3523 
 3524                 case DTYPE_MQUEUE:
 3525                         type = KF_TYPE_MQUEUE;
 3526                         break;
 3527 
 3528                 case DTYPE_SHM:
 3529                         type = KF_TYPE_SHM;
 3530                         data = fp;
 3531                         break;
 3532 
 3533                 case DTYPE_SEM:
 3534                         type = KF_TYPE_SEM;
 3535                         data = fp;
 3536                         break;
 3537 
 3538                 case DTYPE_PTS:
 3539                         type = KF_TYPE_PTS;
 3540                         data = fp->f_data;
 3541                         break;
 3542 
 3543 #ifdef PROCDESC
 3544                 case DTYPE_PROCDESC:
 3545                         type = KF_TYPE_PROCDESC;
 3546                         data = fp->f_data;
 3547                         break;
 3548 #endif
 3549 
 3550                 default:
 3551                         type = KF_TYPE_UNKNOWN;
 3552                         break;
 3553                 }
 3554                 refcnt = fp->f_count;
 3555                 fflags = fp->f_flag;
 3556                 offset = foffset_get(fp);
 3557 
 3558                 /*
 3559                  * Create sysctl entry.
 3560                  * It is OK to drop the filedesc lock here as we will
 3561                  * re-validate and re-evaluate its properties when
 3562                  * the loop continues.
 3563                  */
 3564                 error = export_fd_to_sb(data, type, i, fflags, refcnt,
 3565                     offset, &rights, efbuf);
 3566                 if (error != 0)
 3567                         break;
 3568         }
 3569         FILEDESC_SUNLOCK(fdp);
 3570         fddrop(fdp);
 3571 fail:
 3572         free(efbuf, M_TEMP);
 3573         return (error);
 3574 }
 3575 
 3576 #define FILEDESC_SBUF_SIZE      (sizeof(struct kinfo_file) * 5)
 3577 
 3578 /*
 3579  * Get per-process file descriptors for use by procstat(1), et al.
 3580  */
 3581 static int
 3582 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
 3583 {
 3584         struct sbuf sb;
 3585         struct proc *p;
 3586         ssize_t maxlen;
 3587         int error, error2, *name;
 3588 
 3589         name = (int *)arg1;
 3590 
 3591         sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
 3592         error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
 3593         if (error != 0) {
 3594                 sbuf_delete(&sb);
 3595                 return (error);
 3596         }
 3597         maxlen = req->oldptr != NULL ? req->oldlen : -1;
 3598         error = kern_proc_filedesc_out(p, &sb, maxlen);
 3599         error2 = sbuf_finish(&sb);
 3600         sbuf_delete(&sb);
 3601         return (error != 0 ? error : error2);
 3602 }
 3603 
 3604 int
 3605 vntype_to_kinfo(int vtype)
 3606 {
 3607         struct {
 3608                 int     vtype;
 3609                 int     kf_vtype;
 3610         } vtypes_table[] = {
 3611                 { VBAD, KF_VTYPE_VBAD },
 3612                 { VBLK, KF_VTYPE_VBLK },
 3613                 { VCHR, KF_VTYPE_VCHR },
 3614                 { VDIR, KF_VTYPE_VDIR },
 3615                 { VFIFO, KF_VTYPE_VFIFO },
 3616                 { VLNK, KF_VTYPE_VLNK },
 3617                 { VNON, KF_VTYPE_VNON },
 3618                 { VREG, KF_VTYPE_VREG },
 3619                 { VSOCK, KF_VTYPE_VSOCK }
 3620         };
 3621 #define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table))
 3622         unsigned int i;
 3623 
 3624         /*
 3625          * Perform vtype translation.
 3626          */
 3627         for (i = 0; i < NVTYPES; i++)
 3628                 if (vtypes_table[i].vtype == vtype)
 3629                         break;
 3630         if (i < NVTYPES)
 3631                 return (vtypes_table[i].kf_vtype);
 3632 
 3633         return (KF_VTYPE_UNKNOWN);
 3634 }
 3635 
 3636 static int
 3637 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif)
 3638 {
 3639         struct vattr va;
 3640         char *fullpath, *freepath;
 3641         int error;
 3642 
 3643         if (vp == NULL)
 3644                 return (1);
 3645         kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
 3646         freepath = NULL;
 3647         fullpath = "-";
 3648         error = vn_fullpath(curthread, vp, &fullpath, &freepath);
 3649         if (error == 0) {
 3650                 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
 3651         }
 3652         if (freepath != NULL)
 3653                 free(freepath, M_TEMP);
 3654 
 3655         /*
 3656          * Retrieve vnode attributes.
 3657          */
 3658         va.va_fsid = VNOVAL;
 3659         va.va_rdev = NODEV;
 3660         vn_lock(vp, LK_SHARED | LK_RETRY);
 3661         error = VOP_GETATTR(vp, &va, curthread->td_ucred);
 3662         VOP_UNLOCK(vp, 0);
 3663         if (error != 0)
 3664                 return (error);
 3665         if (va.va_fsid != VNOVAL)
 3666                 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
 3667         else
 3668                 kif->kf_un.kf_file.kf_file_fsid =
 3669                     vp->v_mount->mnt_stat.f_fsid.val[0];
 3670         kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
 3671         kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
 3672         kif->kf_un.kf_file.kf_file_size = va.va_size;
 3673         kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
 3674         return (0);
 3675 }
 3676 
 3677 static int
 3678 fill_socket_info(struct socket *so, struct kinfo_file *kif)
 3679 {
 3680         struct sockaddr *sa;
 3681         struct inpcb *inpcb;
 3682         struct unpcb *unpcb;
 3683         int error;
 3684 
 3685         if (so == NULL)
 3686                 return (1);
 3687         kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
 3688         kif->kf_sock_type = so->so_type;
 3689         kif->kf_sock_protocol = so->so_proto->pr_protocol;
 3690         kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
 3691         switch(kif->kf_sock_domain) {
 3692         case AF_INET:
 3693         case AF_INET6:
 3694                 if (kif->kf_sock_protocol == IPPROTO_TCP) {
 3695                         if (so->so_pcb != NULL) {
 3696                                 inpcb = (struct inpcb *)(so->so_pcb);
 3697                                 kif->kf_un.kf_sock.kf_sock_inpcb =
 3698                                     (uintptr_t)inpcb->inp_ppcb;
 3699                         }
 3700                 }
 3701                 break;
 3702         case AF_UNIX:
 3703                 if (so->so_pcb != NULL) {
 3704                         unpcb = (struct unpcb *)(so->so_pcb);
 3705                         if (unpcb->unp_conn) {
 3706                                 kif->kf_un.kf_sock.kf_sock_unpconn =
 3707                                     (uintptr_t)unpcb->unp_conn;
 3708                                 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
 3709                                     so->so_rcv.sb_state;
 3710                                 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
 3711                                     so->so_snd.sb_state;
 3712                         }
 3713                 }
 3714                 break;
 3715         }
 3716         error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
 3717         if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
 3718                 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
 3719                 free(sa, M_SONAME);
 3720         }
 3721         error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
 3722         if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
 3723                 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
 3724                 free(sa, M_SONAME);
 3725         }
 3726         strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
 3727             sizeof(kif->kf_path));
 3728         return (0);
 3729 }
 3730 
 3731 static int
 3732 fill_pts_info(struct tty *tp, struct kinfo_file *kif)
 3733 {
 3734 
 3735         if (tp == NULL)
 3736                 return (1);
 3737         kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp);
 3738         strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path));
 3739         return (0);
 3740 }
 3741 
 3742 static int
 3743 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
 3744 {
 3745 
 3746         if (pi == NULL)
 3747                 return (1);
 3748         kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
 3749         kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
 3750         kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
 3751         return (0);
 3752 }
 3753 
 3754 static int
 3755 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
 3756 {
 3757 
 3758         if (pdp == NULL)
 3759                 return (1);
 3760         kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
 3761         return (0);
 3762 }
 3763 
 3764 static int
 3765 fill_sem_info(struct file *fp, struct kinfo_file *kif)
 3766 {
 3767         struct thread *td;
 3768         struct stat sb;
 3769 
 3770         td = curthread;
 3771         if (fp->f_data == NULL)
 3772                 return (1);
 3773         if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
 3774                 return (1);
 3775         if (ksem_info == NULL)
 3776                 return (1);
 3777         ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path),
 3778             &kif->kf_un.kf_sem.kf_sem_value);
 3779         kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode;
 3780         return (0);
 3781 }
 3782 
 3783 static int
 3784 fill_shm_info(struct file *fp, struct kinfo_file *kif)
 3785 {
 3786         struct thread *td;
 3787         struct stat sb;
 3788 
 3789         td = curthread;
 3790         if (fp->f_data == NULL)
 3791                 return (1);
 3792         if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
 3793                 return (1);
 3794         shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path));
 3795         kif->kf_un.kf_file.kf_file_mode = sb.st_mode;
 3796         kif->kf_un.kf_file.kf_file_size = sb.st_size;
 3797         return (0);
 3798 }
 3799 
 3800 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
 3801     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
 3802     "Process filedesc entries");
 3803 
 3804 #ifdef DDB
 3805 /*
 3806  * For the purposes of debugging, generate a human-readable string for the
 3807  * file type.
 3808  */
 3809 static const char *
 3810 file_type_to_name(short type)
 3811 {
 3812 
 3813         switch (type) {
 3814         case 0:
 3815                 return ("zero");
 3816         case DTYPE_VNODE:
 3817                 return ("vnod");
 3818         case DTYPE_SOCKET:
 3819                 return ("sock");
 3820         case DTYPE_PIPE:
 3821                 return ("pipe");
 3822         case DTYPE_FIFO:
 3823                 return ("fifo");
 3824         case DTYPE_KQUEUE:
 3825                 return ("kque");
 3826         case DTYPE_CRYPTO:
 3827                 return ("crpt");
 3828         case DTYPE_MQUEUE:
 3829                 return ("mque");
 3830         case DTYPE_SHM:
 3831                 return ("shm");
 3832         case DTYPE_SEM:
 3833                 return ("ksem");
 3834         default:
 3835                 return ("unkn");
 3836         }
 3837 }
 3838 
 3839 /*
 3840  * For the purposes of debugging, identify a process (if any, perhaps one of
 3841  * many) that references the passed file in its file descriptor array. Return
 3842  * NULL if none.
 3843  */
 3844 static struct proc *
 3845 file_to_first_proc(struct file *fp)
 3846 {
 3847         struct filedesc *fdp;
 3848         struct proc *p;
 3849         int n;
 3850 
 3851         FOREACH_PROC_IN_SYSTEM(p) {
 3852                 if (p->p_state == PRS_NEW)
 3853                         continue;
 3854                 fdp = p->p_fd;
 3855                 if (fdp == NULL)
 3856                         continue;
 3857                 for (n = 0; n <= fdp->fd_lastfile; n++) {
 3858                         if (fp == fdp->fd_ofiles[n].fde_file)
 3859                                 return (p);
 3860                 }
 3861         }
 3862         return (NULL);
 3863 }
 3864 
 3865 static void
 3866 db_print_file(struct file *fp, int header)
 3867 {
 3868         struct proc *p;
 3869 
 3870         if (header)
 3871                 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
 3872                     "File", "Type", "Data", "Flag", "GCFl", "Count",
 3873                     "MCount", "Vnode", "FPID", "FCmd");
 3874         p = file_to_first_proc(fp);
 3875         db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
 3876             file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
 3877             0, fp->f_count, 0, fp->f_vnode,
 3878             p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
 3879 }
 3880 
 3881 DB_SHOW_COMMAND(file, db_show_file)
 3882 {
 3883         struct file *fp;
 3884 
 3885         if (!have_addr) {
 3886                 db_printf("usage: show file <addr>\n");
 3887                 return;
 3888         }
 3889         fp = (struct file *)addr;
 3890         db_print_file(fp, 1);
 3891 }
 3892 
 3893 DB_SHOW_COMMAND(files, db_show_files)
 3894 {
 3895         struct filedesc *fdp;
 3896         struct file *fp;
 3897         struct proc *p;
 3898         int header;
 3899         int n;
 3900 
 3901         header = 1;
 3902         FOREACH_PROC_IN_SYSTEM(p) {
 3903                 if (p->p_state == PRS_NEW)
 3904                         continue;
 3905                 if ((fdp = p->p_fd) == NULL)
 3906                         continue;
 3907                 for (n = 0; n <= fdp->fd_lastfile; ++n) {
 3908                         if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
 3909                                 continue;
 3910                         db_print_file(fp, header);
 3911                         header = 0;
 3912                 }
 3913         }
 3914 }
 3915 #endif
 3916 
 3917 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
 3918     &maxfilesperproc, 0, "Maximum files allowed open per process");
 3919 
 3920 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
 3921     &maxfiles, 0, "Maximum number of files");
 3922 
 3923 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
 3924     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
 3925 
 3926 /* ARGSUSED*/
 3927 static void
 3928 filelistinit(void *dummy)
 3929 {
 3930 
 3931         file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
 3932             NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
 3933         mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
 3934         mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
 3935 }
 3936 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
 3937 
 3938 /*-------------------------------------------------------------------*/
 3939 
 3940 static int
 3941 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
 3942     int flags, struct thread *td)
 3943 {
 3944 
 3945         return (EBADF);
 3946 }
 3947 
 3948 static int
 3949 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
 3950     struct thread *td)
 3951 {
 3952 
 3953         return (EINVAL);
 3954 }
 3955 
 3956 static int
 3957 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
 3958     struct thread *td)
 3959 {
 3960 
 3961         return (EBADF);
 3962 }
 3963 
 3964 static int
 3965 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
 3966     struct thread *td)
 3967 {
 3968 
 3969         return (0);
 3970 }
 3971 
 3972 static int
 3973 badfo_kqfilter(struct file *fp, struct knote *kn)
 3974 {
 3975 
 3976         return (EBADF);
 3977 }
 3978 
 3979 static int
 3980 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
 3981     struct thread *td)
 3982 {
 3983 
 3984         return (EBADF);
 3985 }
 3986 
 3987 static int
 3988 badfo_close(struct file *fp, struct thread *td)
 3989 {
 3990 
 3991         return (EBADF);
 3992 }
 3993 
 3994 static int
 3995 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 3996     struct thread *td)
 3997 {
 3998 
 3999         return (EBADF);
 4000 }
 4001 
 4002 static int
 4003 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 4004     struct thread *td)
 4005 {
 4006 
 4007         return (EBADF);
 4008 }
 4009 
 4010 static int
 4011 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
 4012     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
 4013     int kflags, struct thread *td)
 4014 {
 4015 
 4016         return (EBADF);
 4017 }
 4018 
 4019 struct fileops badfileops = {
 4020         .fo_read = badfo_readwrite,
 4021         .fo_write = badfo_readwrite,
 4022         .fo_truncate = badfo_truncate,
 4023         .fo_ioctl = badfo_ioctl,
 4024         .fo_poll = badfo_poll,
 4025         .fo_kqfilter = badfo_kqfilter,
 4026         .fo_stat = badfo_stat,
 4027         .fo_close = badfo_close,
 4028         .fo_chmod = badfo_chmod,
 4029         .fo_chown = badfo_chown,
 4030         .fo_sendfile = badfo_sendfile,
 4031 };
 4032 
 4033 int
 4034 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 4035     struct thread *td)
 4036 {
 4037 
 4038         return (EINVAL);
 4039 }
 4040 
 4041 int
 4042 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 4043     struct thread *td)
 4044 {
 4045 
 4046         return (EINVAL);
 4047 }
 4048 
 4049 int
 4050 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
 4051     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
 4052     int kflags, struct thread *td)
 4053 {
 4054 
 4055         return (EINVAL);
 4056 }
 4057 
 4058 /*-------------------------------------------------------------------*/
 4059 
 4060 /*
 4061  * File Descriptor pseudo-device driver (/dev/fd/).
 4062  *
 4063  * Opening minor device N dup()s the file (if any) connected to file
 4064  * descriptor N belonging to the calling process.  Note that this driver
 4065  * consists of only the ``open()'' routine, because all subsequent
 4066  * references to this file will be direct to the other driver.
 4067  *
 4068  * XXX: we could give this one a cloning event handler if necessary.
 4069  */
 4070 
 4071 /* ARGSUSED */
 4072 static int
 4073 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
 4074 {
 4075 
 4076         /*
 4077          * XXX Kludge: set curthread->td_dupfd to contain the value of the
 4078          * the file descriptor being sought for duplication. The error
 4079          * return ensures that the vnode for this device will be released
 4080          * by vn_open. Open will detect this special error and take the
 4081          * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
 4082          * will simply report the error.
 4083          */
 4084         td->td_dupfd = dev2unit(dev);
 4085         return (ENODEV);
 4086 }
 4087 
 4088 static struct cdevsw fildesc_cdevsw = {
 4089         .d_version =    D_VERSION,
 4090         .d_open =       fdopen,
 4091         .d_name =       "FD",
 4092 };
 4093 
 4094 static void
 4095 fildesc_drvinit(void *unused)
 4096 {
 4097         struct cdev *dev;
 4098 
 4099         dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
 4100             UID_ROOT, GID_WHEEL, 0666, "fd/0");
 4101         make_dev_alias(dev, "stdin");
 4102         dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
 4103             UID_ROOT, GID_WHEEL, 0666, "fd/1");
 4104         make_dev_alias(dev, "stdout");
 4105         dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
 4106             UID_ROOT, GID_WHEEL, 0666, "fd/2");
 4107         make_dev_alias(dev, "stderr");
 4108 }
 4109 
 4110 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);

Cache object: 949598e299a91e8fc61b20ded81ff8e3


[ 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.