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/9.2/sys/kern/kern_descrip.c 252829 2013-07-05 19:38:57Z trociny $");
   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 
  110 MALLOC_DECLARE(M_FADVISE);
  111 
  112 static uma_zone_t file_zone;
  113 
  114 void    (*ksem_info)(struct ksem *ks, char *path, size_t size, uint32_t *value);
  115 
  116 /* Flags for do_dup() */
  117 #define DUP_FIXED       0x1     /* Force fixed allocation */
  118 #define DUP_FCNTL       0x2     /* fcntl()-style errors */
  119 #define DUP_CLOEXEC     0x4     /* Atomically set FD_CLOEXEC. */
  120 
  121 static int do_dup(struct thread *td, int flags, int old, int new,
  122     register_t *retval);
  123 static int      fd_first_free(struct filedesc *, int, int);
  124 static int      fd_last_used(struct filedesc *, int, int);
  125 static void     fdgrowtable(struct filedesc *, int);
  126 static void     fdunused(struct filedesc *fdp, int fd);
  127 static void     fdused(struct filedesc *fdp, int fd);
  128 static int      fill_vnode_info(struct vnode *vp, struct kinfo_file *kif);
  129 static int      fill_socket_info(struct socket *so, struct kinfo_file *kif);
  130 static int      fill_pts_info(struct tty *tp, struct kinfo_file *kif);
  131 static int      fill_pipe_info(struct pipe *pi, struct kinfo_file *kif);
  132 static int      fill_procdesc_info(struct procdesc *pdp,
  133     struct kinfo_file *kif);
  134 static int      fill_sem_info(struct file *fp, struct kinfo_file *kif);
  135 static int      fill_shm_info(struct file *fp, struct kinfo_file *kif);
  136 
  137 /*
  138  * A process is initially started out with NDFILE descriptors stored within
  139  * this structure, selected to be enough for typical applications based on
  140  * the historical limit of 20 open files (and the usage of descriptors by
  141  * shells).  If these descriptors are exhausted, a larger descriptor table
  142  * may be allocated, up to a process' resource limit; the internal arrays
  143  * are then unused.
  144  */
  145 #define NDFILE          20
  146 #define NDSLOTSIZE      sizeof(NDSLOTTYPE)
  147 #define NDENTRIES       (NDSLOTSIZE * __CHAR_BIT)
  148 #define NDSLOT(x)       ((x) / NDENTRIES)
  149 #define NDBIT(x)        ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
  150 #define NDSLOTS(x)      (((x) + NDENTRIES - 1) / NDENTRIES)
  151 
  152 /*
  153  * Storage required per open file descriptor.
  154  */
  155 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
  156 
  157 /*
  158  * Storage to hold unused ofiles that need to be reclaimed.
  159  */
  160 struct freetable {
  161         struct file     **ft_table;
  162         SLIST_ENTRY(freetable) ft_next;
  163 };
  164 
  165 /*
  166  * Basic allocation of descriptors:
  167  * one of the above, plus arrays for NDFILE descriptors.
  168  */
  169 struct filedesc0 {
  170         struct  filedesc fd_fd;
  171         /*
  172          * ofiles which need to be reclaimed on free.
  173          */
  174         SLIST_HEAD(,freetable) fd_free;
  175         /*
  176          * These arrays are used when the number of open files is
  177          * <= NDFILE, and are then pointed to by the pointers above.
  178          */
  179         struct  file *fd_dfiles[NDFILE];
  180         char    fd_dfileflags[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  * Find the first zero bit in the given bitmap, starting at low and not
  196  * exceeding size - 1.
  197  */
  198 static int
  199 fd_first_free(struct filedesc *fdp, int low, int size)
  200 {
  201         NDSLOTTYPE *map = fdp->fd_map;
  202         NDSLOTTYPE mask;
  203         int off, maxoff;
  204 
  205         if (low >= size)
  206                 return (low);
  207 
  208         off = NDSLOT(low);
  209         if (low % NDENTRIES) {
  210                 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
  211                 if ((mask &= ~map[off]) != 0UL)
  212                         return (off * NDENTRIES + ffsl(mask) - 1);
  213                 ++off;
  214         }
  215         for (maxoff = NDSLOTS(size); off < maxoff; ++off)
  216                 if (map[off] != ~0UL)
  217                         return (off * NDENTRIES + ffsl(~map[off]) - 1);
  218         return (size);
  219 }
  220 
  221 /*
  222  * Find the highest non-zero bit in the given bitmap, starting at low and
  223  * not exceeding size - 1.
  224  */
  225 static int
  226 fd_last_used(struct filedesc *fdp, int low, int size)
  227 {
  228         NDSLOTTYPE *map = fdp->fd_map;
  229         NDSLOTTYPE mask;
  230         int off, minoff;
  231 
  232         if (low >= size)
  233                 return (-1);
  234 
  235         off = NDSLOT(size);
  236         if (size % NDENTRIES) {
  237                 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
  238                 if ((mask &= map[off]) != 0)
  239                         return (off * NDENTRIES + flsl(mask) - 1);
  240                 --off;
  241         }
  242         for (minoff = NDSLOT(low); off >= minoff; --off)
  243                 if (map[off] != 0)
  244                         return (off * NDENTRIES + flsl(map[off]) - 1);
  245         return (low - 1);
  246 }
  247 
  248 static int
  249 fdisused(struct filedesc *fdp, int fd)
  250 {
  251         KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
  252             ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
  253         return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
  254 }
  255 
  256 /*
  257  * Mark a file descriptor as used.
  258  */
  259 static void
  260 fdused(struct filedesc *fdp, int fd)
  261 {
  262 
  263         FILEDESC_XLOCK_ASSERT(fdp);
  264         KASSERT(!fdisused(fdp, fd),
  265             ("fd already used"));
  266 
  267         fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
  268         if (fd > fdp->fd_lastfile)
  269                 fdp->fd_lastfile = fd;
  270         if (fd == fdp->fd_freefile)
  271                 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
  272 }
  273 
  274 /*
  275  * Mark a file descriptor as unused.
  276  */
  277 static void
  278 fdunused(struct filedesc *fdp, int fd)
  279 {
  280 
  281         FILEDESC_XLOCK_ASSERT(fdp);
  282         KASSERT(fdisused(fdp, fd),
  283             ("fd is already unused"));
  284         KASSERT(fdp->fd_ofiles[fd] == NULL,
  285             ("fd is still in use"));
  286 
  287         fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
  288         if (fd < fdp->fd_freefile)
  289                 fdp->fd_freefile = fd;
  290         if (fd == fdp->fd_lastfile)
  291                 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
  292 }
  293 
  294 /*
  295  * System calls on descriptors.
  296  */
  297 #ifndef _SYS_SYSPROTO_H_
  298 struct getdtablesize_args {
  299         int     dummy;
  300 };
  301 #endif
  302 /* ARGSUSED */
  303 int
  304 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
  305 {
  306         struct proc *p = td->td_proc;
  307         uint64_t lim;
  308 
  309         PROC_LOCK(p);
  310         td->td_retval[0] =
  311             min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
  312         lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
  313         PROC_UNLOCK(p);
  314         if (lim < td->td_retval[0])
  315                 td->td_retval[0] = lim;
  316         return (0);
  317 }
  318 
  319 /*
  320  * Duplicate a file descriptor to a particular value.
  321  *
  322  * Note: keep in mind that a potential race condition exists when closing
  323  * descriptors from a shared descriptor table (via rfork).
  324  */
  325 #ifndef _SYS_SYSPROTO_H_
  326 struct dup2_args {
  327         u_int   from;
  328         u_int   to;
  329 };
  330 #endif
  331 /* ARGSUSED */
  332 int
  333 sys_dup2(struct thread *td, struct dup2_args *uap)
  334 {
  335 
  336         return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
  337                     td->td_retval));
  338 }
  339 
  340 /*
  341  * Duplicate a file descriptor.
  342  */
  343 #ifndef _SYS_SYSPROTO_H_
  344 struct dup_args {
  345         u_int   fd;
  346 };
  347 #endif
  348 /* ARGSUSED */
  349 int
  350 sys_dup(struct thread *td, struct dup_args *uap)
  351 {
  352 
  353         return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
  354 }
  355 
  356 /*
  357  * The file control system call.
  358  */
  359 #ifndef _SYS_SYSPROTO_H_
  360 struct fcntl_args {
  361         int     fd;
  362         int     cmd;
  363         long    arg;
  364 };
  365 #endif
  366 /* ARGSUSED */
  367 int
  368 sys_fcntl(struct thread *td, struct fcntl_args *uap)
  369 {
  370         struct flock fl;
  371         struct __oflock ofl;
  372         intptr_t arg;
  373         int error;
  374         int cmd;
  375 
  376         error = 0;
  377         cmd = uap->cmd;
  378         switch (uap->cmd) {
  379         case F_OGETLK:
  380         case F_OSETLK:
  381         case F_OSETLKW:
  382                 /*
  383                  * Convert old flock structure to new.
  384                  */
  385                 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
  386                 fl.l_start = ofl.l_start;
  387                 fl.l_len = ofl.l_len;
  388                 fl.l_pid = ofl.l_pid;
  389                 fl.l_type = ofl.l_type;
  390                 fl.l_whence = ofl.l_whence;
  391                 fl.l_sysid = 0;
  392 
  393                 switch (uap->cmd) {
  394                 case F_OGETLK:
  395                     cmd = F_GETLK;
  396                     break;
  397                 case F_OSETLK:
  398                     cmd = F_SETLK;
  399                     break;
  400                 case F_OSETLKW:
  401                     cmd = F_SETLKW;
  402                     break;
  403                 }
  404                 arg = (intptr_t)&fl;
  405                 break;
  406         case F_GETLK:
  407         case F_SETLK:
  408         case F_SETLKW:
  409         case F_SETLK_REMOTE:
  410                 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
  411                 arg = (intptr_t)&fl;
  412                 break;
  413         default:
  414                 arg = uap->arg;
  415                 break;
  416         }
  417         if (error)
  418                 return (error);
  419         error = kern_fcntl(td, uap->fd, cmd, arg);
  420         if (error)
  421                 return (error);
  422         if (uap->cmd == F_OGETLK) {
  423                 ofl.l_start = fl.l_start;
  424                 ofl.l_len = fl.l_len;
  425                 ofl.l_pid = fl.l_pid;
  426                 ofl.l_type = fl.l_type;
  427                 ofl.l_whence = fl.l_whence;
  428                 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
  429         } else if (uap->cmd == F_GETLK) {
  430                 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
  431         }
  432         return (error);
  433 }
  434 
  435 static inline struct file *
  436 fdtofp(int fd, struct filedesc *fdp)
  437 {
  438         struct file *fp;
  439 
  440         FILEDESC_LOCK_ASSERT(fdp);
  441         if ((unsigned)fd >= fdp->fd_nfiles ||
  442             (fp = fdp->fd_ofiles[fd]) == NULL)
  443                 return (NULL);
  444         return (fp);
  445 }
  446 
  447 static inline int
  448 fdunwrap(int fd, cap_rights_t rights, struct filedesc *fdp, struct file **fpp)
  449 {
  450 
  451         *fpp = fdtofp(fd, fdp);
  452         if (*fpp == NULL)
  453                 return (EBADF);
  454 
  455 #ifdef CAPABILITIES
  456         if ((*fpp)->f_type == DTYPE_CAPABILITY) {
  457                 int err = cap_funwrap(*fpp, rights, fpp);
  458                 if (err != 0) {
  459                         *fpp = NULL;
  460                         return (err);
  461                 }
  462         }
  463 #endif /* CAPABILITIES */
  464         return (0);
  465 }
  466 
  467 int
  468 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
  469 {
  470         struct filedesc *fdp;
  471         struct flock *flp;
  472         struct file *fp;
  473         struct proc *p;
  474         char *pop;
  475         struct vnode *vp;
  476         int error, flg, tmp;
  477         int vfslocked;
  478         u_int old, new;
  479         uint64_t bsize;
  480         off_t foffset;
  481 
  482         vfslocked = 0;
  483         error = 0;
  484         flg = F_POSIX;
  485         p = td->td_proc;
  486         fdp = p->p_fd;
  487 
  488         switch (cmd) {
  489         case F_DUPFD:
  490                 tmp = arg;
  491                 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
  492                 break;
  493 
  494         case F_DUPFD_CLOEXEC:
  495                 tmp = arg;
  496                 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp,
  497                     td->td_retval);
  498                 break;
  499 
  500         case F_DUP2FD:
  501                 tmp = arg;
  502                 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
  503                 break;
  504 
  505         case F_DUP2FD_CLOEXEC:
  506                 tmp = arg;
  507                 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp,
  508                     td->td_retval);
  509                 break;
  510 
  511         case F_GETFD:
  512                 FILEDESC_SLOCK(fdp);
  513                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  514                         FILEDESC_SUNLOCK(fdp);
  515                         error = EBADF;
  516                         break;
  517                 }
  518                 pop = &fdp->fd_ofileflags[fd];
  519                 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
  520                 FILEDESC_SUNLOCK(fdp);
  521                 break;
  522 
  523         case F_SETFD:
  524                 FILEDESC_XLOCK(fdp);
  525                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  526                         FILEDESC_XUNLOCK(fdp);
  527                         error = EBADF;
  528                         break;
  529                 }
  530                 pop = &fdp->fd_ofileflags[fd];
  531                 *pop = (*pop &~ UF_EXCLOSE) |
  532                     (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
  533                 FILEDESC_XUNLOCK(fdp);
  534                 break;
  535 
  536         case F_GETFL:
  537                 FILEDESC_SLOCK(fdp);
  538                 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
  539                 if (error != 0) {
  540                         FILEDESC_SUNLOCK(fdp);
  541                         break;
  542                 }
  543                 td->td_retval[0] = OFLAGS(fp->f_flag);
  544                 FILEDESC_SUNLOCK(fdp);
  545                 break;
  546 
  547         case F_SETFL:
  548                 FILEDESC_SLOCK(fdp);
  549                 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
  550                 if (error != 0) {
  551                         FILEDESC_SUNLOCK(fdp);
  552                         break;
  553                 }
  554                 fhold(fp);
  555                 FILEDESC_SUNLOCK(fdp);
  556                 do {
  557                         tmp = flg = fp->f_flag;
  558                         tmp &= ~FCNTLFLAGS;
  559                         tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
  560                 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
  561                 tmp = fp->f_flag & FNONBLOCK;
  562                 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
  563                 if (error) {
  564                         fdrop(fp, td);
  565                         break;
  566                 }
  567                 tmp = fp->f_flag & FASYNC;
  568                 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
  569                 if (error == 0) {
  570                         fdrop(fp, td);
  571                         break;
  572                 }
  573                 atomic_clear_int(&fp->f_flag, FNONBLOCK);
  574                 tmp = 0;
  575                 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
  576                 fdrop(fp, td);
  577                 break;
  578 
  579         case F_GETOWN:
  580                 FILEDESC_SLOCK(fdp);
  581                 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
  582                 if (error != 0) {
  583                         FILEDESC_SUNLOCK(fdp);
  584                         break;
  585                 }
  586                 fhold(fp);
  587                 FILEDESC_SUNLOCK(fdp);
  588                 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
  589                 if (error == 0)
  590                         td->td_retval[0] = tmp;
  591                 fdrop(fp, td);
  592                 break;
  593 
  594         case F_SETOWN:
  595                 FILEDESC_SLOCK(fdp);
  596                 error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
  597                 if (error != 0) {
  598                         FILEDESC_SUNLOCK(fdp);
  599                         break;
  600                 }
  601                 fhold(fp);
  602                 FILEDESC_SUNLOCK(fdp);
  603                 tmp = arg;
  604                 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
  605                 fdrop(fp, td);
  606                 break;
  607 
  608         case F_SETLK_REMOTE:
  609                 error = priv_check(td, PRIV_NFS_LOCKD);
  610                 if (error)
  611                         return (error);
  612                 flg = F_REMOTE;
  613                 goto do_setlk;
  614 
  615         case F_SETLKW:
  616                 flg |= F_WAIT;
  617                 /* FALLTHROUGH F_SETLK */
  618 
  619         case F_SETLK:
  620         do_setlk:
  621                 FILEDESC_SLOCK(fdp);
  622                 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp);
  623                 if (error != 0) {
  624                         FILEDESC_SUNLOCK(fdp);
  625                         break;
  626                 }
  627                 if (fp->f_type != DTYPE_VNODE) {
  628                         FILEDESC_SUNLOCK(fdp);
  629                         error = EBADF;
  630                         break;
  631                 }
  632                 flp = (struct flock *)arg;
  633                 if (flp->l_whence == SEEK_CUR) {
  634                         foffset = foffset_get(fp);
  635                         if (foffset < 0 ||
  636                             (flp->l_start > 0 &&
  637                              foffset > OFF_MAX - flp->l_start)) {
  638                                 FILEDESC_SUNLOCK(fdp);
  639                                 error = EOVERFLOW;
  640                                 break;
  641                         }
  642                         flp->l_start += foffset;
  643                 }
  644 
  645                 /*
  646                  * VOP_ADVLOCK() may block.
  647                  */
  648                 fhold(fp);
  649                 FILEDESC_SUNLOCK(fdp);
  650                 vp = fp->f_vnode;
  651                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  652                 switch (flp->l_type) {
  653                 case F_RDLCK:
  654                         if ((fp->f_flag & FREAD) == 0) {
  655                                 error = EBADF;
  656                                 break;
  657                         }
  658                         PROC_LOCK(p->p_leader);
  659                         p->p_leader->p_flag |= P_ADVLOCK;
  660                         PROC_UNLOCK(p->p_leader);
  661                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
  662                             flp, flg);
  663                         break;
  664                 case F_WRLCK:
  665                         if ((fp->f_flag & FWRITE) == 0) {
  666                                 error = EBADF;
  667                                 break;
  668                         }
  669                         PROC_LOCK(p->p_leader);
  670                         p->p_leader->p_flag |= P_ADVLOCK;
  671                         PROC_UNLOCK(p->p_leader);
  672                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
  673                             flp, flg);
  674                         break;
  675                 case F_UNLCK:
  676                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
  677                             flp, flg);
  678                         break;
  679                 case F_UNLCKSYS:
  680                         /*
  681                          * Temporary api for testing remote lock
  682                          * infrastructure.
  683                          */
  684                         if (flg != F_REMOTE) {
  685                                 error = EINVAL;
  686                                 break;
  687                         }
  688                         error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
  689                             F_UNLCKSYS, flp, flg);
  690                         break;
  691                 default:
  692                         error = EINVAL;
  693                         break;
  694                 }
  695                 VFS_UNLOCK_GIANT(vfslocked);
  696                 vfslocked = 0;
  697                 /* Check for race with close */
  698                 FILEDESC_SLOCK(fdp);
  699                 if ((unsigned) fd >= fdp->fd_nfiles ||
  700                     fp != fdp->fd_ofiles[fd]) {
  701                         FILEDESC_SUNLOCK(fdp);
  702                         flp->l_whence = SEEK_SET;
  703                         flp->l_start = 0;
  704                         flp->l_len = 0;
  705                         flp->l_type = F_UNLCK;
  706                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  707                         (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
  708                                            F_UNLCK, flp, F_POSIX);
  709                         VFS_UNLOCK_GIANT(vfslocked);
  710                         vfslocked = 0;
  711                 } else
  712                         FILEDESC_SUNLOCK(fdp);
  713                 fdrop(fp, td);
  714                 break;
  715 
  716         case F_GETLK:
  717                 FILEDESC_SLOCK(fdp);
  718                 error = fdunwrap(fd, CAP_FLOCK, fdp, &fp);
  719                 if (error != 0) {
  720                         FILEDESC_SUNLOCK(fdp);
  721                         break;
  722                 }
  723                 if (fp->f_type != DTYPE_VNODE) {
  724                         FILEDESC_SUNLOCK(fdp);
  725                         error = EBADF;
  726                         break;
  727                 }
  728                 flp = (struct flock *)arg;
  729                 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
  730                     flp->l_type != F_UNLCK) {
  731                         FILEDESC_SUNLOCK(fdp);
  732                         error = EINVAL;
  733                         break;
  734                 }
  735                 if (flp->l_whence == SEEK_CUR) {
  736                         foffset = foffset_get(fp);
  737                         if ((flp->l_start > 0 &&
  738                             foffset > OFF_MAX - flp->l_start) ||
  739                             (flp->l_start < 0 &&
  740                              foffset < OFF_MIN - flp->l_start)) {
  741                                 FILEDESC_SUNLOCK(fdp);
  742                                 error = EOVERFLOW;
  743                                 break;
  744                         }
  745                         flp->l_start += foffset;
  746                 }
  747                 /*
  748                  * VOP_ADVLOCK() may block.
  749                  */
  750                 fhold(fp);
  751                 FILEDESC_SUNLOCK(fdp);
  752                 vp = fp->f_vnode;
  753                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  754                 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
  755                     F_POSIX);
  756                 VFS_UNLOCK_GIANT(vfslocked);
  757                 vfslocked = 0;
  758                 fdrop(fp, td);
  759                 break;
  760 
  761         case F_RDAHEAD:
  762                 arg = arg ? 128 * 1024: 0;
  763                 /* FALLTHROUGH */
  764         case F_READAHEAD:
  765                 FILEDESC_SLOCK(fdp);
  766                 if ((fp = fdtofp(fd, fdp)) == NULL) {
  767                         FILEDESC_SUNLOCK(fdp);
  768                         error = EBADF;
  769                         break;
  770                 }
  771                 if (fp->f_type != DTYPE_VNODE) {
  772                         FILEDESC_SUNLOCK(fdp);
  773                         error = EBADF;
  774                         break;
  775                 }
  776                 fhold(fp);
  777                 FILEDESC_SUNLOCK(fdp);
  778                 if (arg != 0) {
  779                         vp = fp->f_vnode;
  780                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  781                         error = vn_lock(vp, LK_SHARED);
  782                         if (error != 0)
  783                                 goto readahead_vnlock_fail;
  784                         bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
  785                         VOP_UNLOCK(vp, 0);
  786                         fp->f_seqcount = (arg + bsize - 1) / bsize;
  787                         do {
  788                                 new = old = fp->f_flag;
  789                                 new |= FRDAHEAD;
  790                         } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
  791 readahead_vnlock_fail:
  792                         VFS_UNLOCK_GIANT(vfslocked);
  793                         vfslocked = 0;
  794                 } else {
  795                         do {
  796                                 new = old = fp->f_flag;
  797                                 new &= ~FRDAHEAD;
  798                         } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
  799                 }
  800                 fdrop(fp, td);
  801                 break;
  802 
  803         default:
  804                 error = EINVAL;
  805                 break;
  806         }
  807         VFS_UNLOCK_GIANT(vfslocked);
  808         return (error);
  809 }
  810 
  811 /*
  812  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
  813  */
  814 static int
  815 do_dup(struct thread *td, int flags, int old, int new,
  816     register_t *retval)
  817 {
  818         struct filedesc *fdp;
  819         struct proc *p;
  820         struct file *fp;
  821         struct file *delfp;
  822         int error, holdleaders, maxfd;
  823 
  824         p = td->td_proc;
  825         fdp = p->p_fd;
  826 
  827         /*
  828          * Verify we have a valid descriptor to dup from and possibly to
  829          * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
  830          * return EINVAL when the new descriptor is out of bounds.
  831          */
  832         if (old < 0)
  833                 return (EBADF);
  834         if (new < 0)
  835                 return (flags & DUP_FCNTL ? EINVAL : EBADF);
  836         PROC_LOCK(p);
  837         maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
  838         PROC_UNLOCK(p);
  839         if (new >= maxfd)
  840                 return (flags & DUP_FCNTL ? EINVAL : EBADF);
  841 
  842         FILEDESC_XLOCK(fdp);
  843         if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
  844                 FILEDESC_XUNLOCK(fdp);
  845                 return (EBADF);
  846         }
  847         if (flags & DUP_FIXED && old == new) {
  848                 *retval = new;
  849                 if (flags & DUP_CLOEXEC)
  850                         fdp->fd_ofileflags[new] |= UF_EXCLOSE;
  851                 FILEDESC_XUNLOCK(fdp);
  852                 return (0);
  853         }
  854         fp = fdp->fd_ofiles[old];
  855         fhold(fp);
  856 
  857         /*
  858          * If the caller specified a file descriptor, make sure the file
  859          * table is large enough to hold it, and grab it.  Otherwise, just
  860          * allocate a new descriptor the usual way.  Since the filedesc
  861          * lock may be temporarily dropped in the process, we have to look
  862          * out for a race.
  863          */
  864         if (flags & DUP_FIXED) {
  865                 if (new >= fdp->fd_nfiles) {
  866                         /*
  867                          * The resource limits are here instead of e.g.
  868                          * fdalloc(), because the file descriptor table may be
  869                          * shared between processes, so we can't really use
  870                          * racct_add()/racct_sub().  Instead of counting the
  871                          * number of actually allocated descriptors, just put
  872                          * the limit on the size of the file descriptor table.
  873                          */
  874 #ifdef RACCT
  875                         PROC_LOCK(p);
  876                         error = racct_set(p, RACCT_NOFILE, new + 1);
  877                         PROC_UNLOCK(p);
  878                         if (error != 0) {
  879                                 FILEDESC_XUNLOCK(fdp);
  880                                 fdrop(fp, td);
  881                                 return (EMFILE);
  882                         }
  883 #endif
  884                         fdgrowtable(fdp, new + 1);
  885                 }
  886                 if (fdp->fd_ofiles[new] == NULL)
  887                         fdused(fdp, new);
  888         } else {
  889                 if ((error = fdalloc(td, new, &new)) != 0) {
  890                         FILEDESC_XUNLOCK(fdp);
  891                         fdrop(fp, td);
  892                         return (error);
  893                 }
  894         }
  895 
  896         /*
  897          * If the old file changed out from under us then treat it as a
  898          * bad file descriptor.  Userland should do its own locking to
  899          * avoid this case.
  900          */
  901         if (fdp->fd_ofiles[old] != fp) {
  902                 /* we've allocated a descriptor which we won't use */
  903                 if (fdp->fd_ofiles[new] == NULL)
  904                         fdunused(fdp, new);
  905                 FILEDESC_XUNLOCK(fdp);
  906                 fdrop(fp, td);
  907                 return (EBADF);
  908         }
  909         KASSERT(old != new,
  910             ("new fd is same as old"));
  911 
  912         /*
  913          * Save info on the descriptor being overwritten.  We cannot close
  914          * it without introducing an ownership race for the slot, since we
  915          * need to drop the filedesc lock to call closef().
  916          *
  917          * XXX this duplicates parts of close().
  918          */
  919         delfp = fdp->fd_ofiles[new];
  920         holdleaders = 0;
  921         if (delfp != NULL) {
  922                 if (td->td_proc->p_fdtol != NULL) {
  923                         /*
  924                          * Ask fdfree() to sleep to ensure that all relevant
  925                          * process leaders can be traversed in closef().
  926                          */
  927                         fdp->fd_holdleaderscount++;
  928                         holdleaders = 1;
  929                 }
  930         }
  931 
  932         /*
  933          * Duplicate the source descriptor
  934          */
  935         fdp->fd_ofiles[new] = fp;
  936         if ((flags & DUP_CLOEXEC) != 0)
  937                 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] | UF_EXCLOSE;
  938         else
  939                 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] & ~UF_EXCLOSE;
  940         if (new > fdp->fd_lastfile)
  941                 fdp->fd_lastfile = new;
  942         *retval = new;
  943 
  944         /*
  945          * If we dup'd over a valid file, we now own the reference to it
  946          * and must dispose of it using closef() semantics (as if a
  947          * close() were performed on it).
  948          *
  949          * XXX this duplicates parts of close().
  950          */
  951         if (delfp != NULL) {
  952                 knote_fdclose(td, new);
  953                 if (delfp->f_type == DTYPE_MQUEUE)
  954                         mq_fdclose(td, new, delfp);
  955                 FILEDESC_XUNLOCK(fdp);
  956                 (void) closef(delfp, td);
  957                 if (holdleaders) {
  958                         FILEDESC_XLOCK(fdp);
  959                         fdp->fd_holdleaderscount--;
  960                         if (fdp->fd_holdleaderscount == 0 &&
  961                             fdp->fd_holdleaderswakeup != 0) {
  962                                 fdp->fd_holdleaderswakeup = 0;
  963                                 wakeup(&fdp->fd_holdleaderscount);
  964                         }
  965                         FILEDESC_XUNLOCK(fdp);
  966                 }
  967         } else {
  968                 FILEDESC_XUNLOCK(fdp);
  969         }
  970         return (0);
  971 }
  972 
  973 /*
  974  * If sigio is on the list associated with a process or process group,
  975  * disable signalling from the device, remove sigio from the list and
  976  * free sigio.
  977  */
  978 void
  979 funsetown(struct sigio **sigiop)
  980 {
  981         struct sigio *sigio;
  982 
  983         SIGIO_LOCK();
  984         sigio = *sigiop;
  985         if (sigio == NULL) {
  986                 SIGIO_UNLOCK();
  987                 return;
  988         }
  989         *(sigio->sio_myref) = NULL;
  990         if ((sigio)->sio_pgid < 0) {
  991                 struct pgrp *pg = (sigio)->sio_pgrp;
  992                 PGRP_LOCK(pg);
  993                 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
  994                              sigio, sio_pgsigio);
  995                 PGRP_UNLOCK(pg);
  996         } else {
  997                 struct proc *p = (sigio)->sio_proc;
  998                 PROC_LOCK(p);
  999                 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
 1000                              sigio, sio_pgsigio);
 1001                 PROC_UNLOCK(p);
 1002         }
 1003         SIGIO_UNLOCK();
 1004         crfree(sigio->sio_ucred);
 1005         free(sigio, M_SIGIO);
 1006 }
 1007 
 1008 /*
 1009  * Free a list of sigio structures.
 1010  * We only need to lock the SIGIO_LOCK because we have made ourselves
 1011  * inaccessible to callers of fsetown and therefore do not need to lock
 1012  * the proc or pgrp struct for the list manipulation.
 1013  */
 1014 void
 1015 funsetownlst(struct sigiolst *sigiolst)
 1016 {
 1017         struct proc *p;
 1018         struct pgrp *pg;
 1019         struct sigio *sigio;
 1020 
 1021         sigio = SLIST_FIRST(sigiolst);
 1022         if (sigio == NULL)
 1023                 return;
 1024         p = NULL;
 1025         pg = NULL;
 1026 
 1027         /*
 1028          * Every entry of the list should belong
 1029          * to a single proc or pgrp.
 1030          */
 1031         if (sigio->sio_pgid < 0) {
 1032                 pg = sigio->sio_pgrp;
 1033                 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
 1034         } else /* if (sigio->sio_pgid > 0) */ {
 1035                 p = sigio->sio_proc;
 1036                 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
 1037         }
 1038 
 1039         SIGIO_LOCK();
 1040         while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
 1041                 *(sigio->sio_myref) = NULL;
 1042                 if (pg != NULL) {
 1043                         KASSERT(sigio->sio_pgid < 0,
 1044                             ("Proc sigio in pgrp sigio list"));
 1045                         KASSERT(sigio->sio_pgrp == pg,
 1046                             ("Bogus pgrp in sigio list"));
 1047                         PGRP_LOCK(pg);
 1048                         SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
 1049                             sio_pgsigio);
 1050                         PGRP_UNLOCK(pg);
 1051                 } else /* if (p != NULL) */ {
 1052                         KASSERT(sigio->sio_pgid > 0,
 1053                             ("Pgrp sigio in proc sigio list"));
 1054                         KASSERT(sigio->sio_proc == p,
 1055                             ("Bogus proc in sigio list"));
 1056                         PROC_LOCK(p);
 1057                         SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
 1058                             sio_pgsigio);
 1059                         PROC_UNLOCK(p);
 1060                 }
 1061                 SIGIO_UNLOCK();
 1062                 crfree(sigio->sio_ucred);
 1063                 free(sigio, M_SIGIO);
 1064                 SIGIO_LOCK();
 1065         }
 1066         SIGIO_UNLOCK();
 1067 }
 1068 
 1069 /*
 1070  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
 1071  *
 1072  * After permission checking, add a sigio structure to the sigio list for
 1073  * the process or process group.
 1074  */
 1075 int
 1076 fsetown(pid_t pgid, struct sigio **sigiop)
 1077 {
 1078         struct proc *proc;
 1079         struct pgrp *pgrp;
 1080         struct sigio *sigio;
 1081         int ret;
 1082 
 1083         if (pgid == 0) {
 1084                 funsetown(sigiop);
 1085                 return (0);
 1086         }
 1087 
 1088         ret = 0;
 1089 
 1090         /* Allocate and fill in the new sigio out of locks. */
 1091         sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
 1092         sigio->sio_pgid = pgid;
 1093         sigio->sio_ucred = crhold(curthread->td_ucred);
 1094         sigio->sio_myref = sigiop;
 1095 
 1096         sx_slock(&proctree_lock);
 1097         if (pgid > 0) {
 1098                 proc = pfind(pgid);
 1099                 if (proc == NULL) {
 1100                         ret = ESRCH;
 1101                         goto fail;
 1102                 }
 1103 
 1104                 /*
 1105                  * Policy - Don't allow a process to FSETOWN a process
 1106                  * in another session.
 1107                  *
 1108                  * Remove this test to allow maximum flexibility or
 1109                  * restrict FSETOWN to the current process or process
 1110                  * group for maximum safety.
 1111                  */
 1112                 PROC_UNLOCK(proc);
 1113                 if (proc->p_session != curthread->td_proc->p_session) {
 1114                         ret = EPERM;
 1115                         goto fail;
 1116                 }
 1117 
 1118                 pgrp = NULL;
 1119         } else /* if (pgid < 0) */ {
 1120                 pgrp = pgfind(-pgid);
 1121                 if (pgrp == NULL) {
 1122                         ret = ESRCH;
 1123                         goto fail;
 1124                 }
 1125                 PGRP_UNLOCK(pgrp);
 1126 
 1127                 /*
 1128                  * Policy - Don't allow a process to FSETOWN a process
 1129                  * in another session.
 1130                  *
 1131                  * Remove this test to allow maximum flexibility or
 1132                  * restrict FSETOWN to the current process or process
 1133                  * group for maximum safety.
 1134                  */
 1135                 if (pgrp->pg_session != curthread->td_proc->p_session) {
 1136                         ret = EPERM;
 1137                         goto fail;
 1138                 }
 1139 
 1140                 proc = NULL;
 1141         }
 1142         funsetown(sigiop);
 1143         if (pgid > 0) {
 1144                 PROC_LOCK(proc);
 1145                 /*
 1146                  * Since funsetownlst() is called without the proctree
 1147                  * locked, we need to check for P_WEXIT.
 1148                  * XXX: is ESRCH correct?
 1149                  */
 1150                 if ((proc->p_flag & P_WEXIT) != 0) {
 1151                         PROC_UNLOCK(proc);
 1152                         ret = ESRCH;
 1153                         goto fail;
 1154                 }
 1155                 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
 1156                 sigio->sio_proc = proc;
 1157                 PROC_UNLOCK(proc);
 1158         } else {
 1159                 PGRP_LOCK(pgrp);
 1160                 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
 1161                 sigio->sio_pgrp = pgrp;
 1162                 PGRP_UNLOCK(pgrp);
 1163         }
 1164         sx_sunlock(&proctree_lock);
 1165         SIGIO_LOCK();
 1166         *sigiop = sigio;
 1167         SIGIO_UNLOCK();
 1168         return (0);
 1169 
 1170 fail:
 1171         sx_sunlock(&proctree_lock);
 1172         crfree(sigio->sio_ucred);
 1173         free(sigio, M_SIGIO);
 1174         return (ret);
 1175 }
 1176 
 1177 /*
 1178  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
 1179  */
 1180 pid_t
 1181 fgetown(sigiop)
 1182         struct sigio **sigiop;
 1183 {
 1184         pid_t pgid;
 1185 
 1186         SIGIO_LOCK();
 1187         pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
 1188         SIGIO_UNLOCK();
 1189         return (pgid);
 1190 }
 1191 
 1192 /*
 1193  * Close a file descriptor.
 1194  */
 1195 #ifndef _SYS_SYSPROTO_H_
 1196 struct close_args {
 1197         int     fd;
 1198 };
 1199 #endif
 1200 /* ARGSUSED */
 1201 int
 1202 sys_close(td, uap)
 1203         struct thread *td;
 1204         struct close_args *uap;
 1205 {
 1206 
 1207         return (kern_close(td, uap->fd));
 1208 }
 1209 
 1210 int
 1211 kern_close(td, fd)
 1212         struct thread *td;
 1213         int fd;
 1214 {
 1215         struct filedesc *fdp;
 1216         struct file *fp, *fp_object;
 1217         int error;
 1218         int holdleaders;
 1219 
 1220         error = 0;
 1221         holdleaders = 0;
 1222         fdp = td->td_proc->p_fd;
 1223 
 1224         AUDIT_SYSCLOSE(td, fd);
 1225 
 1226         FILEDESC_XLOCK(fdp);
 1227         if ((unsigned)fd >= fdp->fd_nfiles ||
 1228             (fp = fdp->fd_ofiles[fd]) == NULL) {
 1229                 FILEDESC_XUNLOCK(fdp);
 1230                 return (EBADF);
 1231         }
 1232         fdp->fd_ofiles[fd] = NULL;
 1233         fdp->fd_ofileflags[fd] = 0;
 1234         fdunused(fdp, fd);
 1235         if (td->td_proc->p_fdtol != NULL) {
 1236                 /*
 1237                  * Ask fdfree() to sleep to ensure that all relevant
 1238                  * process leaders can be traversed in closef().
 1239                  */
 1240                 fdp->fd_holdleaderscount++;
 1241                 holdleaders = 1;
 1242         }
 1243 
 1244         /*
 1245          * We now hold the fp reference that used to be owned by the
 1246          * descriptor array.  We have to unlock the FILEDESC *AFTER*
 1247          * knote_fdclose to prevent a race of the fd getting opened, a knote
 1248          * added, and deleteing a knote for the new fd.
 1249          */
 1250         knote_fdclose(td, fd);
 1251 
 1252         /*
 1253          * When we're closing an fd with a capability, we need to notify
 1254          * mqueue if the underlying object is of type mqueue.
 1255          */
 1256         (void)cap_funwrap(fp, 0, &fp_object);
 1257         if (fp_object->f_type == DTYPE_MQUEUE)
 1258                 mq_fdclose(td, fd, fp_object);
 1259         FILEDESC_XUNLOCK(fdp);
 1260 
 1261         error = closef(fp, td);
 1262         if (holdleaders) {
 1263                 FILEDESC_XLOCK(fdp);
 1264                 fdp->fd_holdleaderscount--;
 1265                 if (fdp->fd_holdleaderscount == 0 &&
 1266                     fdp->fd_holdleaderswakeup != 0) {
 1267                         fdp->fd_holdleaderswakeup = 0;
 1268                         wakeup(&fdp->fd_holdleaderscount);
 1269                 }
 1270                 FILEDESC_XUNLOCK(fdp);
 1271         }
 1272         return (error);
 1273 }
 1274 
 1275 /*
 1276  * Close open file descriptors.
 1277  */
 1278 #ifndef _SYS_SYSPROTO_H_
 1279 struct closefrom_args {
 1280         int     lowfd;
 1281 };
 1282 #endif
 1283 /* ARGSUSED */
 1284 int
 1285 sys_closefrom(struct thread *td, struct closefrom_args *uap)
 1286 {
 1287         struct filedesc *fdp;
 1288         int fd;
 1289 
 1290         fdp = td->td_proc->p_fd;
 1291         AUDIT_ARG_FD(uap->lowfd);
 1292 
 1293         /*
 1294          * Treat negative starting file descriptor values identical to
 1295          * closefrom(0) which closes all files.
 1296          */
 1297         if (uap->lowfd < 0)
 1298                 uap->lowfd = 0;
 1299         FILEDESC_SLOCK(fdp);
 1300         for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) {
 1301                 if (fdp->fd_ofiles[fd] != NULL) {
 1302                         FILEDESC_SUNLOCK(fdp);
 1303                         (void)kern_close(td, fd);
 1304                         FILEDESC_SLOCK(fdp);
 1305                 }
 1306         }
 1307         FILEDESC_SUNLOCK(fdp);
 1308         return (0);
 1309 }
 1310 
 1311 #if defined(COMPAT_43)
 1312 /*
 1313  * Return status information about a file descriptor.
 1314  */
 1315 #ifndef _SYS_SYSPROTO_H_
 1316 struct ofstat_args {
 1317         int     fd;
 1318         struct  ostat *sb;
 1319 };
 1320 #endif
 1321 /* ARGSUSED */
 1322 int
 1323 ofstat(struct thread *td, struct ofstat_args *uap)
 1324 {
 1325         struct ostat oub;
 1326         struct stat ub;
 1327         int error;
 1328 
 1329         error = kern_fstat(td, uap->fd, &ub);
 1330         if (error == 0) {
 1331                 cvtstat(&ub, &oub);
 1332                 error = copyout(&oub, uap->sb, sizeof(oub));
 1333         }
 1334         return (error);
 1335 }
 1336 #endif /* COMPAT_43 */
 1337 
 1338 /*
 1339  * Return status information about a file descriptor.
 1340  */
 1341 #ifndef _SYS_SYSPROTO_H_
 1342 struct fstat_args {
 1343         int     fd;
 1344         struct  stat *sb;
 1345 };
 1346 #endif
 1347 /* ARGSUSED */
 1348 int
 1349 sys_fstat(struct thread *td, struct fstat_args *uap)
 1350 {
 1351         struct stat ub;
 1352         int error;
 1353 
 1354         error = kern_fstat(td, uap->fd, &ub);
 1355         if (error == 0)
 1356                 error = copyout(&ub, uap->sb, sizeof(ub));
 1357         return (error);
 1358 }
 1359 
 1360 int
 1361 kern_fstat(struct thread *td, int fd, struct stat *sbp)
 1362 {
 1363         struct file *fp;
 1364         int error;
 1365 
 1366         AUDIT_ARG_FD(fd);
 1367 
 1368         if ((error = fget(td, fd, CAP_FSTAT, &fp)) != 0)
 1369                 return (error);
 1370 
 1371         AUDIT_ARG_FILE(td->td_proc, fp);
 1372 
 1373         error = fo_stat(fp, sbp, td->td_ucred, td);
 1374         fdrop(fp, td);
 1375 #ifdef KTRACE
 1376         if (error == 0 && KTRPOINT(td, KTR_STRUCT))
 1377                 ktrstat(sbp);
 1378 #endif
 1379         return (error);
 1380 }
 1381 
 1382 /*
 1383  * Return status information about a file descriptor.
 1384  */
 1385 #ifndef _SYS_SYSPROTO_H_
 1386 struct nfstat_args {
 1387         int     fd;
 1388         struct  nstat *sb;
 1389 };
 1390 #endif
 1391 /* ARGSUSED */
 1392 int
 1393 sys_nfstat(struct thread *td, struct nfstat_args *uap)
 1394 {
 1395         struct nstat nub;
 1396         struct stat ub;
 1397         int error;
 1398 
 1399         error = kern_fstat(td, uap->fd, &ub);
 1400         if (error == 0) {
 1401                 cvtnstat(&ub, &nub);
 1402                 error = copyout(&nub, uap->sb, sizeof(nub));
 1403         }
 1404         return (error);
 1405 }
 1406 
 1407 /*
 1408  * Return pathconf information about a file descriptor.
 1409  */
 1410 #ifndef _SYS_SYSPROTO_H_
 1411 struct fpathconf_args {
 1412         int     fd;
 1413         int     name;
 1414 };
 1415 #endif
 1416 /* ARGSUSED */
 1417 int
 1418 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
 1419 {
 1420         struct file *fp;
 1421         struct vnode *vp;
 1422         int error;
 1423 
 1424         if ((error = fget(td, uap->fd, CAP_FPATHCONF, &fp)) != 0)
 1425                 return (error);
 1426 
 1427         /* If asynchronous I/O is available, it works for all descriptors. */
 1428         if (uap->name == _PC_ASYNC_IO) {
 1429                 td->td_retval[0] = async_io_version;
 1430                 goto out;
 1431         }
 1432         vp = fp->f_vnode;
 1433         if (vp != NULL) {
 1434                 int vfslocked;
 1435                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 1436                 vn_lock(vp, LK_SHARED | LK_RETRY);
 1437                 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
 1438                 VOP_UNLOCK(vp, 0);
 1439                 VFS_UNLOCK_GIANT(vfslocked);
 1440         } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
 1441                 if (uap->name != _PC_PIPE_BUF) {
 1442                         error = EINVAL;
 1443                 } else {
 1444                         td->td_retval[0] = PIPE_BUF;
 1445                 error = 0;
 1446                 }
 1447         } else {
 1448                 error = EOPNOTSUPP;
 1449         }
 1450 out:
 1451         fdrop(fp, td);
 1452         return (error);
 1453 }
 1454 
 1455 /*
 1456  * Grow the file table to accomodate (at least) nfd descriptors.  This may
 1457  * block and drop the filedesc lock, but it will reacquire it before
 1458  * returning.
 1459  */
 1460 static void
 1461 fdgrowtable(struct filedesc *fdp, int nfd)
 1462 {
 1463         struct filedesc0 *fdp0;
 1464         struct freetable *fo;
 1465         struct file **ntable;
 1466         struct file **otable;
 1467         char *nfileflags;
 1468         int nnfiles, onfiles;
 1469         NDSLOTTYPE *nmap;
 1470 
 1471         FILEDESC_XLOCK_ASSERT(fdp);
 1472 
 1473         KASSERT(fdp->fd_nfiles > 0,
 1474             ("zero-length file table"));
 1475 
 1476         /* compute the size of the new table */
 1477         onfiles = fdp->fd_nfiles;
 1478         nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
 1479         if (nnfiles <= onfiles)
 1480                 /* the table is already large enough */
 1481                 return;
 1482 
 1483         /* allocate a new table and (if required) new bitmaps */
 1484         FILEDESC_XUNLOCK(fdp);
 1485         ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable),
 1486             M_FILEDESC, M_ZERO | M_WAITOK);
 1487         nfileflags = (char *)&ntable[nnfiles];
 1488         if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
 1489                 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE,
 1490                     M_FILEDESC, M_ZERO | M_WAITOK);
 1491         else
 1492                 nmap = NULL;
 1493         FILEDESC_XLOCK(fdp);
 1494 
 1495         /*
 1496          * We now have new tables ready to go.  Since we dropped the
 1497          * filedesc lock to call malloc(), watch out for a race.
 1498          */
 1499         onfiles = fdp->fd_nfiles;
 1500         if (onfiles >= nnfiles) {
 1501                 /* we lost the race, but that's OK */
 1502                 free(ntable, M_FILEDESC);
 1503                 if (nmap != NULL)
 1504                         free(nmap, M_FILEDESC);
 1505                 return;
 1506         }
 1507         bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
 1508         bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
 1509         otable = fdp->fd_ofiles;
 1510         fdp->fd_ofileflags = nfileflags;
 1511         fdp->fd_ofiles = ntable;
 1512         /*
 1513          * We must preserve ofiles until the process exits because we can't
 1514          * be certain that no threads have references to the old table via
 1515          * _fget().
 1516          */
 1517         if (onfiles > NDFILE) {
 1518                 fo = (struct freetable *)&otable[onfiles];
 1519                 fdp0 = (struct filedesc0 *)fdp;
 1520                 fo->ft_table = otable;
 1521                 SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next);
 1522         }
 1523         if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
 1524                 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
 1525                 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
 1526                         free(fdp->fd_map, M_FILEDESC);
 1527                 fdp->fd_map = nmap;
 1528         }
 1529         fdp->fd_nfiles = nnfiles;
 1530 }
 1531 
 1532 /*
 1533  * Allocate a file descriptor for the process.
 1534  */
 1535 int
 1536 fdalloc(struct thread *td, int minfd, int *result)
 1537 {
 1538         struct proc *p = td->td_proc;
 1539         struct filedesc *fdp = p->p_fd;
 1540         int fd = -1, maxfd;
 1541 #ifdef RACCT
 1542         int error;
 1543 #endif
 1544 
 1545         FILEDESC_XLOCK_ASSERT(fdp);
 1546 
 1547         if (fdp->fd_freefile > minfd)
 1548                 minfd = fdp->fd_freefile;
 1549 
 1550         PROC_LOCK(p);
 1551         maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
 1552         PROC_UNLOCK(p);
 1553 
 1554         /*
 1555          * Search the bitmap for a free descriptor.  If none is found, try
 1556          * to grow the file table.  Keep at it until we either get a file
 1557          * descriptor or run into process or system limits; fdgrowtable()
 1558          * may drop the filedesc lock, so we're in a race.
 1559          */
 1560         for (;;) {
 1561                 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
 1562                 if (fd >= maxfd)
 1563                         return (EMFILE);
 1564                 if (fd < fdp->fd_nfiles)
 1565                         break;
 1566 #ifdef RACCT
 1567                 PROC_LOCK(p);
 1568                 error = racct_set(p, RACCT_NOFILE, min(fdp->fd_nfiles * 2, maxfd));
 1569                 PROC_UNLOCK(p);
 1570                 if (error != 0)
 1571                         return (EMFILE);
 1572 #endif
 1573                 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
 1574         }
 1575 
 1576         /*
 1577          * Perform some sanity checks, then mark the file descriptor as
 1578          * used and return it to the caller.
 1579          */
 1580         KASSERT(!fdisused(fdp, fd),
 1581             ("fd_first_free() returned non-free descriptor"));
 1582         KASSERT(fdp->fd_ofiles[fd] == NULL,
 1583             ("free descriptor isn't"));
 1584         fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
 1585         fdused(fdp, fd);
 1586         *result = fd;
 1587         return (0);
 1588 }
 1589 
 1590 /*
 1591  * Allocate n file descriptors for the process.
 1592  */
 1593 int
 1594 fdallocn(struct thread *td, int minfd, int *fds, int n)
 1595 {
 1596         struct proc *p = td->td_proc;
 1597         struct filedesc *fdp = p->p_fd;
 1598         int i;
 1599 
 1600         FILEDESC_XLOCK_ASSERT(fdp);
 1601 
 1602         if (!fdavail(td, n))
 1603                 return (EMFILE);
 1604 
 1605         for (i = 0; i < n; i++)
 1606                 if (fdalloc(td, 0, &fds[i]) != 0)
 1607                         break;
 1608 
 1609         if (i < n) {
 1610                 for (i--; i >= 0; i--)
 1611                         fdunused(fdp, fds[i]);
 1612                 return (EMFILE);
 1613         }
 1614 
 1615         return (0);
 1616 }
 1617 
 1618 /*
 1619  * Check to see whether n user file descriptors are available to the process
 1620  * p.
 1621  */
 1622 int
 1623 fdavail(struct thread *td, int n)
 1624 {
 1625         struct proc *p = td->td_proc;
 1626         struct filedesc *fdp = td->td_proc->p_fd;
 1627         struct file **fpp;
 1628         int i, lim, last;
 1629 
 1630         FILEDESC_LOCK_ASSERT(fdp);
 1631 
 1632         /*
 1633          * XXX: This is only called from uipc_usrreq.c:unp_externalize();
 1634          *      call racct_add() from there instead of dealing with containers
 1635          *      here.
 1636          */
 1637         PROC_LOCK(p);
 1638         lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
 1639         PROC_UNLOCK(p);
 1640         if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
 1641                 return (1);
 1642         last = min(fdp->fd_nfiles, lim);
 1643         fpp = &fdp->fd_ofiles[fdp->fd_freefile];
 1644         for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
 1645                 if (*fpp == NULL && --n <= 0)
 1646                         return (1);
 1647         }
 1648         return (0);
 1649 }
 1650 
 1651 /*
 1652  * Create a new open file structure and allocate a file decriptor for the
 1653  * process that refers to it.  We add one reference to the file for the
 1654  * descriptor table and one reference for resultfp. This is to prevent us
 1655  * being preempted and the entry in the descriptor table closed after we
 1656  * release the FILEDESC lock.
 1657  */
 1658 int
 1659 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
 1660 {
 1661         struct file *fp;
 1662         int error, fd;
 1663 
 1664         error = falloc_noinstall(td, &fp);
 1665         if (error)
 1666                 return (error);         /* no reference held on error */
 1667 
 1668         error = finstall(td, fp, &fd, flags);
 1669         if (error) {
 1670                 fdrop(fp, td);          /* one reference (fp only) */
 1671                 return (error);
 1672         }
 1673 
 1674         if (resultfp != NULL)
 1675                 *resultfp = fp;         /* copy out result */
 1676         else
 1677                 fdrop(fp, td);          /* release local reference */
 1678 
 1679         if (resultfd != NULL)
 1680                 *resultfd = fd;
 1681 
 1682         return (0);
 1683 }
 1684 
 1685 /*
 1686  * Create a new open file structure without allocating a file descriptor.
 1687  */
 1688 int
 1689 falloc_noinstall(struct thread *td, struct file **resultfp)
 1690 {
 1691         struct file *fp;
 1692         int maxuserfiles = maxfiles - (maxfiles / 20);
 1693         static struct timeval lastfail;
 1694         static int curfail;
 1695 
 1696         KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
 1697 
 1698         if ((openfiles >= maxuserfiles &&
 1699             priv_check(td, PRIV_MAXFILES) != 0) ||
 1700             openfiles >= maxfiles) {
 1701                 if (ppsratecheck(&lastfail, &curfail, 1)) {
 1702                         printf("kern.maxfiles limit exceeded by uid %i, "
 1703                             "please see tuning(7).\n", td->td_ucred->cr_ruid);
 1704                 }
 1705                 return (ENFILE);
 1706         }
 1707         atomic_add_int(&openfiles, 1);
 1708         fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
 1709         refcount_init(&fp->f_count, 1);
 1710         fp->f_cred = crhold(td->td_ucred);
 1711         fp->f_ops = &badfileops;
 1712         fp->f_data = NULL;
 1713         fp->f_vnode = NULL;
 1714         *resultfp = fp;
 1715         return (0);
 1716 }
 1717 
 1718 /*
 1719  * Install a file in a file descriptor table.
 1720  */
 1721 int
 1722 finstall(struct thread *td, struct file *fp, int *fd, int flags)
 1723 {
 1724         struct filedesc *fdp = td->td_proc->p_fd;
 1725         int error;
 1726 
 1727         KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
 1728         KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
 1729 
 1730         FILEDESC_XLOCK(fdp);
 1731         if ((error = fdalloc(td, 0, fd))) {
 1732                 FILEDESC_XUNLOCK(fdp);
 1733                 return (error);
 1734         }
 1735         fhold(fp);
 1736         fdp->fd_ofiles[*fd] = fp;
 1737         if ((flags & O_CLOEXEC) != 0)
 1738                 fdp->fd_ofileflags[*fd] |= UF_EXCLOSE;
 1739         FILEDESC_XUNLOCK(fdp);
 1740         return (0);
 1741 }
 1742 
 1743 /*
 1744  * Build a new filedesc structure from another.
 1745  * Copy the current, root, and jail root vnode references.
 1746  */
 1747 struct filedesc *
 1748 fdinit(struct filedesc *fdp)
 1749 {
 1750         struct filedesc0 *newfdp;
 1751 
 1752         newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
 1753         FILEDESC_LOCK_INIT(&newfdp->fd_fd);
 1754         if (fdp != NULL) {
 1755                 FILEDESC_XLOCK(fdp);
 1756                 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
 1757                 if (newfdp->fd_fd.fd_cdir)
 1758                         VREF(newfdp->fd_fd.fd_cdir);
 1759                 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
 1760                 if (newfdp->fd_fd.fd_rdir)
 1761                         VREF(newfdp->fd_fd.fd_rdir);
 1762                 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
 1763                 if (newfdp->fd_fd.fd_jdir)
 1764                         VREF(newfdp->fd_fd.fd_jdir);
 1765                 FILEDESC_XUNLOCK(fdp);
 1766         }
 1767 
 1768         /* Create the file descriptor table. */
 1769         newfdp->fd_fd.fd_refcnt = 1;
 1770         newfdp->fd_fd.fd_holdcnt = 1;
 1771         newfdp->fd_fd.fd_cmask = CMASK;
 1772         newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
 1773         newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
 1774         newfdp->fd_fd.fd_nfiles = NDFILE;
 1775         newfdp->fd_fd.fd_map = newfdp->fd_dmap;
 1776         newfdp->fd_fd.fd_lastfile = -1;
 1777         return (&newfdp->fd_fd);
 1778 }
 1779 
 1780 static struct filedesc *
 1781 fdhold(struct proc *p)
 1782 {
 1783         struct filedesc *fdp;
 1784 
 1785         mtx_lock(&fdesc_mtx);
 1786         fdp = p->p_fd;
 1787         if (fdp != NULL)
 1788                 fdp->fd_holdcnt++;
 1789         mtx_unlock(&fdesc_mtx);
 1790         return (fdp);
 1791 }
 1792 
 1793 static void
 1794 fddrop(struct filedesc *fdp)
 1795 {
 1796         struct filedesc0 *fdp0;
 1797         struct freetable *ft;
 1798         int i;
 1799 
 1800         mtx_lock(&fdesc_mtx);
 1801         i = --fdp->fd_holdcnt;
 1802         mtx_unlock(&fdesc_mtx);
 1803         if (i > 0)
 1804                 return;
 1805 
 1806         FILEDESC_LOCK_DESTROY(fdp);
 1807         fdp0 = (struct filedesc0 *)fdp;
 1808         while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
 1809                 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
 1810                 free(ft->ft_table, M_FILEDESC);
 1811         }
 1812         free(fdp, M_FILEDESC);
 1813 }
 1814 
 1815 /*
 1816  * Share a filedesc structure.
 1817  */
 1818 struct filedesc *
 1819 fdshare(struct filedesc *fdp)
 1820 {
 1821 
 1822         FILEDESC_XLOCK(fdp);
 1823         fdp->fd_refcnt++;
 1824         FILEDESC_XUNLOCK(fdp);
 1825         return (fdp);
 1826 }
 1827 
 1828 /*
 1829  * Unshare a filedesc structure, if necessary by making a copy
 1830  */
 1831 void
 1832 fdunshare(struct proc *p, struct thread *td)
 1833 {
 1834 
 1835         FILEDESC_XLOCK(p->p_fd);
 1836         if (p->p_fd->fd_refcnt > 1) {
 1837                 struct filedesc *tmp;
 1838 
 1839                 FILEDESC_XUNLOCK(p->p_fd);
 1840                 tmp = fdcopy(p->p_fd);
 1841                 fdfree(td);
 1842                 p->p_fd = tmp;
 1843         } else
 1844                 FILEDESC_XUNLOCK(p->p_fd);
 1845 }
 1846 
 1847 /*
 1848  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
 1849  * this is to ease callers, not catch errors.
 1850  */
 1851 struct filedesc *
 1852 fdcopy(struct filedesc *fdp)
 1853 {
 1854         struct filedesc *newfdp;
 1855         int i;
 1856 
 1857         /* Certain daemons might not have file descriptors. */
 1858         if (fdp == NULL)
 1859                 return (NULL);
 1860 
 1861         newfdp = fdinit(fdp);
 1862         FILEDESC_SLOCK(fdp);
 1863         while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
 1864                 FILEDESC_SUNLOCK(fdp);
 1865                 FILEDESC_XLOCK(newfdp);
 1866                 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
 1867                 FILEDESC_XUNLOCK(newfdp);
 1868                 FILEDESC_SLOCK(fdp);
 1869         }
 1870         /* copy all passable descriptors (i.e. not kqueue) */
 1871         newfdp->fd_freefile = -1;
 1872         for (i = 0; i <= fdp->fd_lastfile; ++i) {
 1873                 if (fdisused(fdp, i) &&
 1874                     (fdp->fd_ofiles[i]->f_ops->fo_flags & DFLAG_PASSABLE) &&
 1875                     fdp->fd_ofiles[i]->f_ops != &badfileops) {
 1876                         newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
 1877                         newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
 1878                         fhold(newfdp->fd_ofiles[i]);
 1879                         newfdp->fd_lastfile = i;
 1880                 } else {
 1881                         if (newfdp->fd_freefile == -1)
 1882                                 newfdp->fd_freefile = i;
 1883                 }
 1884         }
 1885         newfdp->fd_cmask = fdp->fd_cmask;
 1886         FILEDESC_SUNLOCK(fdp);
 1887         FILEDESC_XLOCK(newfdp);
 1888         for (i = 0; i <= newfdp->fd_lastfile; ++i)
 1889                 if (newfdp->fd_ofiles[i] != NULL)
 1890                         fdused(newfdp, i);
 1891         if (newfdp->fd_freefile == -1)
 1892                 newfdp->fd_freefile = i;
 1893         FILEDESC_XUNLOCK(newfdp);
 1894         return (newfdp);
 1895 }
 1896 
 1897 /*
 1898  * Release a filedesc structure.
 1899  */
 1900 void
 1901 fdfree(struct thread *td)
 1902 {
 1903         struct filedesc *fdp;
 1904         struct file **fpp;
 1905         int i, locked;
 1906         struct filedesc_to_leader *fdtol;
 1907         struct file *fp;
 1908         struct vnode *cdir, *jdir, *rdir, *vp;
 1909         struct flock lf;
 1910 
 1911         /* Certain daemons might not have file descriptors. */
 1912         fdp = td->td_proc->p_fd;
 1913         if (fdp == NULL)
 1914                 return;
 1915 
 1916 #ifdef RACCT
 1917         PROC_LOCK(td->td_proc);
 1918         racct_set(td->td_proc, RACCT_NOFILE, 0);
 1919         PROC_UNLOCK(td->td_proc);
 1920 #endif
 1921 
 1922         /* Check for special need to clear POSIX style locks */
 1923         fdtol = td->td_proc->p_fdtol;
 1924         if (fdtol != NULL) {
 1925                 FILEDESC_XLOCK(fdp);
 1926                 KASSERT(fdtol->fdl_refcount > 0,
 1927                         ("filedesc_to_refcount botch: fdl_refcount=%d",
 1928                          fdtol->fdl_refcount));
 1929                 if (fdtol->fdl_refcount == 1 &&
 1930                     (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 1931                         for (i = 0, fpp = fdp->fd_ofiles;
 1932                              i <= fdp->fd_lastfile;
 1933                              i++, fpp++) {
 1934                                 if (*fpp == NULL ||
 1935                                     (*fpp)->f_type != DTYPE_VNODE)
 1936                                         continue;
 1937                                 fp = *fpp;
 1938                                 fhold(fp);
 1939                                 FILEDESC_XUNLOCK(fdp);
 1940                                 lf.l_whence = SEEK_SET;
 1941                                 lf.l_start = 0;
 1942                                 lf.l_len = 0;
 1943                                 lf.l_type = F_UNLCK;
 1944                                 vp = fp->f_vnode;
 1945                                 locked = VFS_LOCK_GIANT(vp->v_mount);
 1946                                 (void) VOP_ADVLOCK(vp,
 1947                                                    (caddr_t)td->td_proc->
 1948                                                    p_leader,
 1949                                                    F_UNLCK,
 1950                                                    &lf,
 1951                                                    F_POSIX);
 1952                                 VFS_UNLOCK_GIANT(locked);
 1953                                 FILEDESC_XLOCK(fdp);
 1954                                 fdrop(fp, td);
 1955                                 fpp = fdp->fd_ofiles + i;
 1956                         }
 1957                 }
 1958         retry:
 1959                 if (fdtol->fdl_refcount == 1) {
 1960                         if (fdp->fd_holdleaderscount > 0 &&
 1961                             (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 1962                                 /*
 1963                                  * close() or do_dup() has cleared a reference
 1964                                  * in a shared file descriptor table.
 1965                                  */
 1966                                 fdp->fd_holdleaderswakeup = 1;
 1967                                 sx_sleep(&fdp->fd_holdleaderscount,
 1968                                     FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
 1969                                 goto retry;
 1970                         }
 1971                         if (fdtol->fdl_holdcount > 0) {
 1972                                 /*
 1973                                  * Ensure that fdtol->fdl_leader remains
 1974                                  * valid in closef().
 1975                                  */
 1976                                 fdtol->fdl_wakeup = 1;
 1977                                 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
 1978                                     "fdlhold", 0);
 1979                                 goto retry;
 1980                         }
 1981                 }
 1982                 fdtol->fdl_refcount--;
 1983                 if (fdtol->fdl_refcount == 0 &&
 1984                     fdtol->fdl_holdcount == 0) {
 1985                         fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
 1986                         fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
 1987                 } else
 1988                         fdtol = NULL;
 1989                 td->td_proc->p_fdtol = NULL;
 1990                 FILEDESC_XUNLOCK(fdp);
 1991                 if (fdtol != NULL)
 1992                         free(fdtol, M_FILEDESC_TO_LEADER);
 1993         }
 1994         FILEDESC_XLOCK(fdp);
 1995         i = --fdp->fd_refcnt;
 1996         FILEDESC_XUNLOCK(fdp);
 1997         if (i > 0)
 1998                 return;
 1999 
 2000         fpp = fdp->fd_ofiles;
 2001         for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
 2002                 if (*fpp) {
 2003                         FILEDESC_XLOCK(fdp);
 2004                         fp = *fpp;
 2005                         *fpp = NULL;
 2006                         FILEDESC_XUNLOCK(fdp);
 2007                         (void) closef(fp, td);
 2008                 }
 2009         }
 2010         FILEDESC_XLOCK(fdp);
 2011 
 2012         /* XXX This should happen earlier. */
 2013         mtx_lock(&fdesc_mtx);
 2014         td->td_proc->p_fd = NULL;
 2015         mtx_unlock(&fdesc_mtx);
 2016 
 2017         if (fdp->fd_nfiles > NDFILE)
 2018                 free(fdp->fd_ofiles, M_FILEDESC);
 2019         if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
 2020                 free(fdp->fd_map, M_FILEDESC);
 2021 
 2022         fdp->fd_nfiles = 0;
 2023 
 2024         cdir = fdp->fd_cdir;
 2025         fdp->fd_cdir = NULL;
 2026         rdir = fdp->fd_rdir;
 2027         fdp->fd_rdir = NULL;
 2028         jdir = fdp->fd_jdir;
 2029         fdp->fd_jdir = NULL;
 2030         FILEDESC_XUNLOCK(fdp);
 2031 
 2032         if (cdir) {
 2033                 locked = VFS_LOCK_GIANT(cdir->v_mount);
 2034                 vrele(cdir);
 2035                 VFS_UNLOCK_GIANT(locked);
 2036         }
 2037         if (rdir) {
 2038                 locked = VFS_LOCK_GIANT(rdir->v_mount);
 2039                 vrele(rdir);
 2040                 VFS_UNLOCK_GIANT(locked);
 2041         }
 2042         if (jdir) {
 2043                 locked = VFS_LOCK_GIANT(jdir->v_mount);
 2044                 vrele(jdir);
 2045                 VFS_UNLOCK_GIANT(locked);
 2046         }
 2047 
 2048         fddrop(fdp);
 2049 }
 2050 
 2051 /*
 2052  * For setugid programs, we don't want to people to use that setugidness
 2053  * to generate error messages which write to a file which otherwise would
 2054  * otherwise be off-limits to the process.  We check for filesystems where
 2055  * the vnode can change out from under us after execve (like [lin]procfs).
 2056  *
 2057  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
 2058  * sufficient.  We also don't check for setugidness since we know we are.
 2059  */
 2060 static int
 2061 is_unsafe(struct file *fp)
 2062 {
 2063         if (fp->f_type == DTYPE_VNODE) {
 2064                 struct vnode *vp = fp->f_vnode;
 2065 
 2066                 if ((vp->v_vflag & VV_PROCDEP) != 0)
 2067                         return (1);
 2068         }
 2069         return (0);
 2070 }
 2071 
 2072 /*
 2073  * Make this setguid thing safe, if at all possible.
 2074  */
 2075 void
 2076 setugidsafety(struct thread *td)
 2077 {
 2078         struct filedesc *fdp;
 2079         int i;
 2080 
 2081         /* Certain daemons might not have file descriptors. */
 2082         fdp = td->td_proc->p_fd;
 2083         if (fdp == NULL)
 2084                 return;
 2085 
 2086         /*
 2087          * Note: fdp->fd_ofiles may be reallocated out from under us while
 2088          * we are blocked in a close.  Be careful!
 2089          */
 2090         FILEDESC_XLOCK(fdp);
 2091         for (i = 0; i <= fdp->fd_lastfile; i++) {
 2092                 if (i > 2)
 2093                         break;
 2094                 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
 2095                         struct file *fp;
 2096 
 2097                         knote_fdclose(td, i);
 2098                         /*
 2099                          * NULL-out descriptor prior to close to avoid
 2100                          * a race while close blocks.
 2101                          */
 2102                         fp = fdp->fd_ofiles[i];
 2103                         fdp->fd_ofiles[i] = NULL;
 2104                         fdp->fd_ofileflags[i] = 0;
 2105                         fdunused(fdp, i);
 2106                         FILEDESC_XUNLOCK(fdp);
 2107                         (void) closef(fp, td);
 2108                         FILEDESC_XLOCK(fdp);
 2109                 }
 2110         }
 2111         FILEDESC_XUNLOCK(fdp);
 2112 }
 2113 
 2114 /*
 2115  * If a specific file object occupies a specific file descriptor, close the
 2116  * file descriptor entry and drop a reference on the file object.  This is a
 2117  * convenience function to handle a subsequent error in a function that calls
 2118  * falloc() that handles the race that another thread might have closed the
 2119  * file descriptor out from under the thread creating the file object.
 2120  */
 2121 void
 2122 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
 2123 {
 2124 
 2125         FILEDESC_XLOCK(fdp);
 2126         if (fdp->fd_ofiles[idx] == fp) {
 2127                 fdp->fd_ofiles[idx] = NULL;
 2128                 fdunused(fdp, idx);
 2129                 FILEDESC_XUNLOCK(fdp);
 2130                 fdrop(fp, td);
 2131         } else
 2132                 FILEDESC_XUNLOCK(fdp);
 2133 }
 2134 
 2135 /*
 2136  * Close any files on exec?
 2137  */
 2138 void
 2139 fdcloseexec(struct thread *td)
 2140 {
 2141         struct filedesc *fdp;
 2142         int i;
 2143 
 2144         /* Certain daemons might not have file descriptors. */
 2145         fdp = td->td_proc->p_fd;
 2146         if (fdp == NULL)
 2147                 return;
 2148 
 2149         FILEDESC_XLOCK(fdp);
 2150 
 2151         /*
 2152          * We cannot cache fd_ofiles or fd_ofileflags since operations
 2153          * may block and rip them out from under us.
 2154          */
 2155         for (i = 0; i <= fdp->fd_lastfile; i++) {
 2156                 if (fdp->fd_ofiles[i] != NULL &&
 2157                     (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
 2158                     (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
 2159                         struct file *fp;
 2160 
 2161                         knote_fdclose(td, i);
 2162                         /*
 2163                          * NULL-out descriptor prior to close to avoid
 2164                          * a race while close blocks.
 2165                          */
 2166                         fp = fdp->fd_ofiles[i];
 2167                         fdp->fd_ofiles[i] = NULL;
 2168                         fdp->fd_ofileflags[i] = 0;
 2169                         fdunused(fdp, i);
 2170                         if (fp->f_type == DTYPE_MQUEUE)
 2171                                 mq_fdclose(td, i, fp);
 2172                         FILEDESC_XUNLOCK(fdp);
 2173                         (void) closef(fp, td);
 2174                         FILEDESC_XLOCK(fdp);
 2175                 }
 2176         }
 2177         FILEDESC_XUNLOCK(fdp);
 2178 }
 2179 
 2180 /*
 2181  * It is unsafe for set[ug]id processes to be started with file
 2182  * descriptors 0..2 closed, as these descriptors are given implicit
 2183  * significance in the Standard C library.  fdcheckstd() will create a
 2184  * descriptor referencing /dev/null for each of stdin, stdout, and
 2185  * stderr that is not already open.
 2186  */
 2187 int
 2188 fdcheckstd(struct thread *td)
 2189 {
 2190         struct filedesc *fdp;
 2191         register_t retval, save;
 2192         int i, error, devnull;
 2193 
 2194         fdp = td->td_proc->p_fd;
 2195         if (fdp == NULL)
 2196                 return (0);
 2197         KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
 2198         devnull = -1;
 2199         error = 0;
 2200         for (i = 0; i < 3; i++) {
 2201                 if (fdp->fd_ofiles[i] != NULL)
 2202                         continue;
 2203                 if (devnull < 0) {
 2204                         save = td->td_retval[0];
 2205                         error = kern_open(td, "/dev/null", UIO_SYSSPACE,
 2206                             O_RDWR, 0);
 2207                         devnull = td->td_retval[0];
 2208                         td->td_retval[0] = save;
 2209                         if (error)
 2210                                 break;
 2211                         KASSERT(devnull == i, ("oof, we didn't get our fd"));
 2212                 } else {
 2213                         error = do_dup(td, DUP_FIXED, devnull, i, &retval);
 2214                         if (error != 0)
 2215                                 break;
 2216                 }
 2217         }
 2218         return (error);
 2219 }
 2220 
 2221 /*
 2222  * Internal form of close.  Decrement reference count on file structure.
 2223  * Note: td may be NULL when closing a file that was being passed in a
 2224  * message.
 2225  *
 2226  * XXXRW: Giant is not required for the caller, but often will be held; this
 2227  * makes it moderately likely the Giant will be recursed in the VFS case.
 2228  */
 2229 int
 2230 closef(struct file *fp, struct thread *td)
 2231 {
 2232         struct vnode *vp;
 2233         struct flock lf;
 2234         struct filedesc_to_leader *fdtol;
 2235         struct filedesc *fdp;
 2236         struct file *fp_object;
 2237 
 2238         /*
 2239          * POSIX record locking dictates that any close releases ALL
 2240          * locks owned by this process.  This is handled by setting
 2241          * a flag in the unlock to free ONLY locks obeying POSIX
 2242          * semantics, and not to free BSD-style file locks.
 2243          * If the descriptor was in a message, POSIX-style locks
 2244          * aren't passed with the descriptor, and the thread pointer
 2245          * will be NULL.  Callers should be careful only to pass a
 2246          * NULL thread pointer when there really is no owning
 2247          * context that might have locks, or the locks will be
 2248          * leaked.
 2249          *
 2250          * If this is a capability, we do lock processing under the underlying
 2251          * node, not the capability itself.
 2252          */
 2253         (void)cap_funwrap(fp, 0, &fp_object);
 2254         if ((fp_object->f_type == DTYPE_VNODE) && (td != NULL)) {
 2255                 int vfslocked;
 2256 
 2257                 vp = fp_object->f_vnode;
 2258                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 2259                 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 2260                         lf.l_whence = SEEK_SET;
 2261                         lf.l_start = 0;
 2262                         lf.l_len = 0;
 2263                         lf.l_type = F_UNLCK;
 2264                         (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
 2265                                            F_UNLCK, &lf, F_POSIX);
 2266                 }
 2267                 fdtol = td->td_proc->p_fdtol;
 2268                 if (fdtol != NULL) {
 2269                         /*
 2270                          * Handle special case where file descriptor table is
 2271                          * shared between multiple process leaders.
 2272                          */
 2273                         fdp = td->td_proc->p_fd;
 2274                         FILEDESC_XLOCK(fdp);
 2275                         for (fdtol = fdtol->fdl_next;
 2276                              fdtol != td->td_proc->p_fdtol;
 2277                              fdtol = fdtol->fdl_next) {
 2278                                 if ((fdtol->fdl_leader->p_flag &
 2279                                      P_ADVLOCK) == 0)
 2280                                         continue;
 2281                                 fdtol->fdl_holdcount++;
 2282                                 FILEDESC_XUNLOCK(fdp);
 2283                                 lf.l_whence = SEEK_SET;
 2284                                 lf.l_start = 0;
 2285                                 lf.l_len = 0;
 2286                                 lf.l_type = F_UNLCK;
 2287                                 vp = fp_object->f_vnode;
 2288                                 (void) VOP_ADVLOCK(vp,
 2289                                                    (caddr_t)fdtol->fdl_leader,
 2290                                                    F_UNLCK, &lf, F_POSIX);
 2291                                 FILEDESC_XLOCK(fdp);
 2292                                 fdtol->fdl_holdcount--;
 2293                                 if (fdtol->fdl_holdcount == 0 &&
 2294                                     fdtol->fdl_wakeup != 0) {
 2295                                         fdtol->fdl_wakeup = 0;
 2296                                         wakeup(fdtol);
 2297                                 }
 2298                         }
 2299                         FILEDESC_XUNLOCK(fdp);
 2300                 }
 2301                 VFS_UNLOCK_GIANT(vfslocked);
 2302         }
 2303         return (fdrop(fp, td));
 2304 }
 2305 
 2306 /*
 2307  * Initialize the file pointer with the specified properties.
 2308  *
 2309  * The ops are set with release semantics to be certain that the flags, type,
 2310  * and data are visible when ops is.  This is to prevent ops methods from being
 2311  * called with bad data.
 2312  */
 2313 void
 2314 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
 2315 {
 2316         fp->f_data = data;
 2317         fp->f_flag = flag;
 2318         fp->f_type = type;
 2319         atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
 2320 }
 2321 
 2322 struct file *
 2323 fget_unlocked(struct filedesc *fdp, int fd)
 2324 {
 2325         struct file *fp;
 2326         u_int count;
 2327 
 2328         if (fd < 0 || fd >= fdp->fd_nfiles)
 2329                 return (NULL);
 2330         /*
 2331          * Fetch the descriptor locklessly.  We avoid fdrop() races by
 2332          * never raising a refcount above 0.  To accomplish this we have
 2333          * to use a cmpset loop rather than an atomic_add.  The descriptor
 2334          * must be re-verified once we acquire a reference to be certain
 2335          * that the identity is still correct and we did not lose a race
 2336          * due to preemption.
 2337          */
 2338         for (;;) {
 2339                 fp = fdp->fd_ofiles[fd];
 2340                 if (fp == NULL)
 2341                         break;
 2342                 count = fp->f_count;
 2343                 if (count == 0)
 2344                         continue;
 2345                 /*
 2346                  * Use an acquire barrier to prevent caching of fd_ofiles
 2347                  * so it is refreshed for verification.
 2348                  */
 2349                 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
 2350                         continue;
 2351                 if (fp == fdp->fd_ofiles[fd])
 2352                         break;
 2353                 fdrop(fp, curthread);
 2354         }
 2355 
 2356         return (fp);
 2357 }
 2358 
 2359 /*
 2360  * Extract the file pointer associated with the specified descriptor for the
 2361  * current user process.
 2362  *
 2363  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
 2364  * returned.
 2365  *
 2366  * If the FGET_GETCAP flag is set, the capability itself will be returned.
 2367  * Calling _fget() with FGET_GETCAP on a non-capability will return EINVAL.
 2368  * Otherwise, if the file is a capability, its rights will be checked against
 2369  * the capability rights mask, and if successful, the object will be unwrapped.
 2370  *
 2371  * If an error occured the non-zero error is returned and *fpp is set to
 2372  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
 2373  * responsible for fdrop().
 2374  */
 2375 #define FGET_GETCAP     0x00000001
 2376 static __inline int
 2377 _fget(struct thread *td, int fd, struct file **fpp, int flags,
 2378     cap_rights_t needrights, cap_rights_t *haverightsp, u_char *maxprotp,
 2379     int fget_flags)
 2380 {
 2381         struct filedesc *fdp;
 2382         struct file *fp;
 2383 #ifdef CAPABILITIES
 2384         struct file *fp_fromcap;
 2385 #endif
 2386         int error;
 2387 
 2388         *fpp = NULL;
 2389         if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
 2390                 return (EBADF);
 2391         if ((fp = fget_unlocked(fdp, fd)) == NULL)
 2392                 return (EBADF);
 2393         if (fp->f_ops == &badfileops) {
 2394                 fdrop(fp, td);
 2395                 return (EBADF);
 2396         }
 2397 
 2398 #ifdef CAPABILITIES
 2399         /*
 2400          * If this is a capability, what rights does it have?
 2401          */
 2402         if (haverightsp != NULL) {
 2403                 if (fp->f_type == DTYPE_CAPABILITY)
 2404                         *haverightsp = cap_rights(fp);
 2405                 else
 2406                         *haverightsp = CAP_MASK_VALID;
 2407         }
 2408 
 2409         /*
 2410          * If a capability has been requested, return the capability directly.
 2411          * Otherwise, check capability rights, extract the underlying object,
 2412          * and check its access flags.
 2413          */
 2414         if (fget_flags & FGET_GETCAP) {
 2415                 if (fp->f_type != DTYPE_CAPABILITY) {
 2416                         fdrop(fp, td);
 2417                         return (EINVAL);
 2418                 }
 2419         } else {
 2420                 if (maxprotp == NULL)
 2421                         error = cap_funwrap(fp, needrights, &fp_fromcap);
 2422                 else
 2423                         error = cap_funwrap_mmap(fp, needrights, maxprotp,
 2424                             &fp_fromcap);
 2425                 if (error != 0) {
 2426                         fdrop(fp, td);
 2427                         return (error);
 2428                 }
 2429 
 2430                 /*
 2431                  * If we've unwrapped a file, drop the original capability
 2432                  * and hold the new descriptor.  fp after this point refers to
 2433                  * the actual (unwrapped) object, not the capability.
 2434                  */
 2435                 if (fp != fp_fromcap) {
 2436                         fhold(fp_fromcap);
 2437                         fdrop(fp, td);
 2438                         fp = fp_fromcap;
 2439                 }
 2440         }
 2441 #else /* !CAPABILITIES */
 2442         KASSERT(fp->f_type != DTYPE_CAPABILITY,
 2443             ("%s: saw capability", __func__));
 2444         if (maxprotp != NULL)
 2445                 *maxprotp = VM_PROT_ALL;
 2446 #endif /* CAPABILITIES */
 2447 
 2448         /*
 2449          * FREAD and FWRITE failure return EBADF as per POSIX.
 2450          */
 2451         error = 0;
 2452         switch (flags) {
 2453         case FREAD:
 2454         case FWRITE:
 2455                 if ((fp->f_flag & flags) == 0)
 2456                         error = EBADF;
 2457                 break;
 2458         case FEXEC:
 2459                 if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
 2460                     ((fp->f_flag & FWRITE) != 0))
 2461                         error = EBADF;
 2462                 break;
 2463         case 0:
 2464                 break;
 2465         default:
 2466                 KASSERT(0, ("wrong flags"));
 2467         }
 2468 
 2469         if (error != 0) {
 2470                 fdrop(fp, td);
 2471                 return (error);
 2472         }
 2473 
 2474         *fpp = fp;
 2475         return (0);
 2476 }
 2477 
 2478 int
 2479 fget(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
 2480 {
 2481 
 2482         return(_fget(td, fd, fpp, 0, rights, NULL, NULL, 0));
 2483 }
 2484 
 2485 int
 2486 fget_mmap(struct thread *td, int fd, cap_rights_t rights, u_char *maxprotp,
 2487     struct file **fpp)
 2488 {
 2489 
 2490         return (_fget(td, fd, fpp, 0, rights, NULL, maxprotp, 0));
 2491 }
 2492 
 2493 int
 2494 fget_read(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
 2495 {
 2496 
 2497         return(_fget(td, fd, fpp, FREAD, rights, NULL, NULL, 0));
 2498 }
 2499 
 2500 int
 2501 fget_write(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
 2502 {
 2503 
 2504         return (_fget(td, fd, fpp, FWRITE, rights, NULL, NULL, 0));
 2505 }
 2506 
 2507 /*
 2508  * Unlike the other fget() calls, which accept and check capability rights
 2509  * but never return capabilities, fgetcap() returns the capability but doesn't
 2510  * check capability rights.
 2511  */
 2512 int
 2513 fgetcap(struct thread *td, int fd, struct file **fpp)
 2514 {
 2515 
 2516         return (_fget(td, fd, fpp, 0, 0, NULL, NULL, FGET_GETCAP));
 2517 }
 2518 
 2519 
 2520 /*
 2521  * Like fget() but loads the underlying vnode, or returns an error if the
 2522  * descriptor does not represent a vnode.  Note that pipes use vnodes but
 2523  * never have VM objects.  The returned vnode will be vref()'d.
 2524  *
 2525  * XXX: what about the unused flags ?
 2526  */
 2527 static __inline int
 2528 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t needrights,
 2529     cap_rights_t *haverightsp, struct vnode **vpp)
 2530 {
 2531         struct file *fp;
 2532         int error;
 2533 
 2534         *vpp = NULL;
 2535         if ((error = _fget(td, fd, &fp, flags, needrights, haverightsp,
 2536             NULL, 0)) != 0)
 2537                 return (error);
 2538         if (fp->f_vnode == NULL) {
 2539                 error = EINVAL;
 2540         } else {
 2541                 *vpp = fp->f_vnode;
 2542                 vref(*vpp);
 2543         }
 2544         fdrop(fp, td);
 2545 
 2546         return (error);
 2547 }
 2548 
 2549 int
 2550 fgetvp(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
 2551 {
 2552 
 2553         return (_fgetvp(td, fd, 0, rights, NULL, vpp));
 2554 }
 2555 
 2556 int
 2557 fgetvp_rights(struct thread *td, int fd, cap_rights_t need, cap_rights_t *have,
 2558     struct vnode **vpp)
 2559 {
 2560         return (_fgetvp(td, fd, 0, need, have, vpp));
 2561 }
 2562 
 2563 int
 2564 fgetvp_read(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
 2565 {
 2566 
 2567         return (_fgetvp(td, fd, FREAD, rights, NULL, vpp));
 2568 }
 2569 
 2570 int
 2571 fgetvp_exec(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
 2572 {
 2573 
 2574         return (_fgetvp(td, fd, FEXEC, rights, NULL, vpp));
 2575 }
 2576 
 2577 #ifdef notyet
 2578 int
 2579 fgetvp_write(struct thread *td, int fd, cap_rights_t rights,
 2580     struct vnode **vpp)
 2581 {
 2582 
 2583         return (_fgetvp(td, fd, FWRITE, rights, NULL, vpp));
 2584 }
 2585 #endif
 2586 
 2587 /*
 2588  * Like fget() but loads the underlying socket, or returns an error if the
 2589  * descriptor does not represent a socket.
 2590  *
 2591  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
 2592  * in the future.
 2593  *
 2594  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
 2595  * on their file descriptor reference to prevent the socket from being free'd
 2596  * during use.
 2597  */
 2598 int
 2599 fgetsock(struct thread *td, int fd, cap_rights_t rights, struct socket **spp,
 2600     u_int *fflagp)
 2601 {
 2602         struct file *fp;
 2603         int error;
 2604 
 2605         *spp = NULL;
 2606         if (fflagp != NULL)
 2607                 *fflagp = 0;
 2608         if ((error = _fget(td, fd, &fp, 0, rights, NULL, NULL, 0)) != 0)
 2609                 return (error);
 2610         if (fp->f_type != DTYPE_SOCKET) {
 2611                 error = ENOTSOCK;
 2612         } else {
 2613                 *spp = fp->f_data;
 2614                 if (fflagp)
 2615                         *fflagp = fp->f_flag;
 2616                 SOCK_LOCK(*spp);
 2617                 soref(*spp);
 2618                 SOCK_UNLOCK(*spp);
 2619         }
 2620         fdrop(fp, td);
 2621 
 2622         return (error);
 2623 }
 2624 
 2625 /*
 2626  * Drop the reference count on the socket and XXX release the SX lock in the
 2627  * future.  The last reference closes the socket.
 2628  *
 2629  * Note: fputsock() is deprecated, see comment for fgetsock().
 2630  */
 2631 void
 2632 fputsock(struct socket *so)
 2633 {
 2634 
 2635         ACCEPT_LOCK();
 2636         SOCK_LOCK(so);
 2637         CURVNET_SET(so->so_vnet);
 2638         sorele(so);
 2639         CURVNET_RESTORE();
 2640 }
 2641 
 2642 /*
 2643  * Handle the last reference to a file being closed.
 2644  *
 2645  * No special capability handling here, as the capability's fo_close will run
 2646  * instead of the object here, and perform any necessary drop on the object.
 2647  */
 2648 int
 2649 _fdrop(struct file *fp, struct thread *td)
 2650 {
 2651         int error;
 2652 
 2653         error = 0;
 2654         if (fp->f_count != 0)
 2655                 panic("fdrop: count %d", fp->f_count);
 2656         if (fp->f_ops != &badfileops)
 2657                 error = fo_close(fp, td);
 2658         atomic_subtract_int(&openfiles, 1);
 2659         crfree(fp->f_cred);
 2660         free(fp->f_advice, M_FADVISE);
 2661         uma_zfree(file_zone, fp);
 2662 
 2663         return (error);
 2664 }
 2665 
 2666 /*
 2667  * Apply an advisory lock on a file descriptor.
 2668  *
 2669  * Just attempt to get a record lock of the requested type on the entire file
 2670  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
 2671  */
 2672 #ifndef _SYS_SYSPROTO_H_
 2673 struct flock_args {
 2674         int     fd;
 2675         int     how;
 2676 };
 2677 #endif
 2678 /* ARGSUSED */
 2679 int
 2680 sys_flock(struct thread *td, struct flock_args *uap)
 2681 {
 2682         struct file *fp;
 2683         struct vnode *vp;
 2684         struct flock lf;
 2685         int vfslocked;
 2686         int error;
 2687 
 2688         if ((error = fget(td, uap->fd, CAP_FLOCK, &fp)) != 0)
 2689                 return (error);
 2690         if (fp->f_type != DTYPE_VNODE) {
 2691                 fdrop(fp, td);
 2692                 return (EOPNOTSUPP);
 2693         }
 2694 
 2695         vp = fp->f_vnode;
 2696         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 2697         lf.l_whence = SEEK_SET;
 2698         lf.l_start = 0;
 2699         lf.l_len = 0;
 2700         if (uap->how & LOCK_UN) {
 2701                 lf.l_type = F_UNLCK;
 2702                 atomic_clear_int(&fp->f_flag, FHASLOCK);
 2703                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
 2704                 goto done2;
 2705         }
 2706         if (uap->how & LOCK_EX)
 2707                 lf.l_type = F_WRLCK;
 2708         else if (uap->how & LOCK_SH)
 2709                 lf.l_type = F_RDLCK;
 2710         else {
 2711                 error = EBADF;
 2712                 goto done2;
 2713         }
 2714         atomic_set_int(&fp->f_flag, FHASLOCK);
 2715         error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
 2716             (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
 2717 done2:
 2718         fdrop(fp, td);
 2719         VFS_UNLOCK_GIANT(vfslocked);
 2720         return (error);
 2721 }
 2722 /*
 2723  * Duplicate the specified descriptor to a free descriptor.
 2724  */
 2725 int
 2726 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
 2727 {
 2728         struct file *wfp;
 2729         struct file *fp;
 2730 
 2731         /*
 2732          * If the to-be-dup'd fd number is greater than the allowed number
 2733          * of file descriptors, or the fd to be dup'd has already been
 2734          * closed, then reject.
 2735          */
 2736         FILEDESC_XLOCK(fdp);
 2737         if (dfd < 0 || dfd >= fdp->fd_nfiles ||
 2738             (wfp = fdp->fd_ofiles[dfd]) == NULL) {
 2739                 FILEDESC_XUNLOCK(fdp);
 2740                 return (EBADF);
 2741         }
 2742 
 2743         /*
 2744          * There are two cases of interest here.
 2745          *
 2746          * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
 2747          *
 2748          * For ENXIO steal away the file structure from (dfd) and store it in
 2749          * (indx).  (dfd) is effectively closed by this operation.
 2750          *
 2751          * Any other error code is just returned.
 2752          */
 2753         switch (error) {
 2754         case ENODEV:
 2755                 /*
 2756                  * Check that the mode the file is being opened for is a
 2757                  * subset of the mode of the existing descriptor.
 2758                  */
 2759                 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
 2760                         FILEDESC_XUNLOCK(fdp);
 2761                         return (EACCES);
 2762                 }
 2763                 fp = fdp->fd_ofiles[indx];
 2764                 fdp->fd_ofiles[indx] = wfp;
 2765                 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
 2766                 if (fp == NULL)
 2767                         fdused(fdp, indx);
 2768                 fhold(wfp);
 2769                 FILEDESC_XUNLOCK(fdp);
 2770                 if (fp != NULL)
 2771                         /*
 2772                          * We now own the reference to fp that the ofiles[]
 2773                          * array used to own.  Release it.
 2774                          */
 2775                         fdrop(fp, td);
 2776                 return (0);
 2777 
 2778         case ENXIO:
 2779                 /*
 2780                  * Steal away the file pointer from dfd and stuff it into indx.
 2781                  */
 2782                 fp = fdp->fd_ofiles[indx];
 2783                 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
 2784                 fdp->fd_ofiles[dfd] = NULL;
 2785                 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
 2786                 fdp->fd_ofileflags[dfd] = 0;
 2787                 fdunused(fdp, dfd);
 2788                 if (fp == NULL)
 2789                         fdused(fdp, indx);
 2790                 FILEDESC_XUNLOCK(fdp);
 2791 
 2792                 /*
 2793                  * We now own the reference to fp that the ofiles[] array
 2794                  * used to own.  Release it.
 2795                  */
 2796                 if (fp != NULL)
 2797                         fdrop(fp, td);
 2798                 return (0);
 2799 
 2800         default:
 2801                 FILEDESC_XUNLOCK(fdp);
 2802                 return (error);
 2803         }
 2804         /* NOTREACHED */
 2805 }
 2806 
 2807 /*
 2808  * Scan all active processes and prisons to see if any of them have a current
 2809  * or root directory of `olddp'. If so, replace them with the new mount point.
 2810  */
 2811 void
 2812 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
 2813 {
 2814         struct filedesc *fdp;
 2815         struct prison *pr;
 2816         struct proc *p;
 2817         int nrele;
 2818 
 2819         if (vrefcnt(olddp) == 1)
 2820                 return;
 2821         nrele = 0;
 2822         sx_slock(&allproc_lock);
 2823         FOREACH_PROC_IN_SYSTEM(p) {
 2824                 fdp = fdhold(p);
 2825                 if (fdp == NULL)
 2826                         continue;
 2827                 FILEDESC_XLOCK(fdp);
 2828                 if (fdp->fd_cdir == olddp) {
 2829                         vref(newdp);
 2830                         fdp->fd_cdir = newdp;
 2831                         nrele++;
 2832                 }
 2833                 if (fdp->fd_rdir == olddp) {
 2834                         vref(newdp);
 2835                         fdp->fd_rdir = newdp;
 2836                         nrele++;
 2837                 }
 2838                 if (fdp->fd_jdir == olddp) {
 2839                         vref(newdp);
 2840                         fdp->fd_jdir = newdp;
 2841                         nrele++;
 2842                 }
 2843                 FILEDESC_XUNLOCK(fdp);
 2844                 fddrop(fdp);
 2845         }
 2846         sx_sunlock(&allproc_lock);
 2847         if (rootvnode == olddp) {
 2848                 vref(newdp);
 2849                 rootvnode = newdp;
 2850                 nrele++;
 2851         }
 2852         mtx_lock(&prison0.pr_mtx);
 2853         if (prison0.pr_root == olddp) {
 2854                 vref(newdp);
 2855                 prison0.pr_root = newdp;
 2856                 nrele++;
 2857         }
 2858         mtx_unlock(&prison0.pr_mtx);
 2859         sx_slock(&allprison_lock);
 2860         TAILQ_FOREACH(pr, &allprison, pr_list) {
 2861                 mtx_lock(&pr->pr_mtx);
 2862                 if (pr->pr_root == olddp) {
 2863                         vref(newdp);
 2864                         pr->pr_root = newdp;
 2865                         nrele++;
 2866                 }
 2867                 mtx_unlock(&pr->pr_mtx);
 2868         }
 2869         sx_sunlock(&allprison_lock);
 2870         while (nrele--)
 2871                 vrele(olddp);
 2872 }
 2873 
 2874 struct filedesc_to_leader *
 2875 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
 2876 {
 2877         struct filedesc_to_leader *fdtol;
 2878 
 2879         fdtol = malloc(sizeof(struct filedesc_to_leader),
 2880                M_FILEDESC_TO_LEADER,
 2881                M_WAITOK);
 2882         fdtol->fdl_refcount = 1;
 2883         fdtol->fdl_holdcount = 0;
 2884         fdtol->fdl_wakeup = 0;
 2885         fdtol->fdl_leader = leader;
 2886         if (old != NULL) {
 2887                 FILEDESC_XLOCK(fdp);
 2888                 fdtol->fdl_next = old->fdl_next;
 2889                 fdtol->fdl_prev = old;
 2890                 old->fdl_next = fdtol;
 2891                 fdtol->fdl_next->fdl_prev = fdtol;
 2892                 FILEDESC_XUNLOCK(fdp);
 2893         } else {
 2894                 fdtol->fdl_next = fdtol;
 2895                 fdtol->fdl_prev = fdtol;
 2896         }
 2897         return (fdtol);
 2898 }
 2899 
 2900 /*
 2901  * Get file structures globally.
 2902  */
 2903 static int
 2904 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
 2905 {
 2906         struct xfile xf;
 2907         struct filedesc *fdp;
 2908         struct file *fp;
 2909         struct proc *p;
 2910         int error, n;
 2911 
 2912         error = sysctl_wire_old_buffer(req, 0);
 2913         if (error != 0)
 2914                 return (error);
 2915         if (req->oldptr == NULL) {
 2916                 n = 0;
 2917                 sx_slock(&allproc_lock);
 2918                 FOREACH_PROC_IN_SYSTEM(p) {
 2919                         if (p->p_state == PRS_NEW)
 2920                                 continue;
 2921                         fdp = fdhold(p);
 2922                         if (fdp == NULL)
 2923                                 continue;
 2924                         /* overestimates sparse tables. */
 2925                         if (fdp->fd_lastfile > 0)
 2926                                 n += fdp->fd_lastfile;
 2927                         fddrop(fdp);
 2928                 }
 2929                 sx_sunlock(&allproc_lock);
 2930                 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
 2931         }
 2932         error = 0;
 2933         bzero(&xf, sizeof(xf));
 2934         xf.xf_size = sizeof(xf);
 2935         sx_slock(&allproc_lock);
 2936         FOREACH_PROC_IN_SYSTEM(p) {
 2937                 PROC_LOCK(p);
 2938                 if (p->p_state == PRS_NEW) {
 2939                         PROC_UNLOCK(p);
 2940                         continue;
 2941                 }
 2942                 if (p_cansee(req->td, p) != 0) {
 2943                         PROC_UNLOCK(p);
 2944                         continue;
 2945                 }
 2946                 xf.xf_pid = p->p_pid;
 2947                 xf.xf_uid = p->p_ucred->cr_uid;
 2948                 PROC_UNLOCK(p);
 2949                 fdp = fdhold(p);
 2950                 if (fdp == NULL)
 2951                         continue;
 2952                 FILEDESC_SLOCK(fdp);
 2953                 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
 2954                         if ((fp = fdp->fd_ofiles[n]) == NULL)
 2955                                 continue;
 2956                         xf.xf_fd = n;
 2957                         xf.xf_file = fp;
 2958                         xf.xf_data = fp->f_data;
 2959                         xf.xf_vnode = fp->f_vnode;
 2960                         xf.xf_type = fp->f_type;
 2961                         xf.xf_count = fp->f_count;
 2962                         xf.xf_msgcount = 0;
 2963                         xf.xf_offset = foffset_get(fp);
 2964                         xf.xf_flag = fp->f_flag;
 2965                         error = SYSCTL_OUT(req, &xf, sizeof(xf));
 2966                         if (error)
 2967                                 break;
 2968                 }
 2969                 FILEDESC_SUNLOCK(fdp);
 2970                 fddrop(fdp);
 2971                 if (error)
 2972                         break;
 2973         }
 2974         sx_sunlock(&allproc_lock);
 2975         return (error);
 2976 }
 2977 
 2978 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
 2979     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
 2980 
 2981 #ifdef KINFO_OFILE_SIZE
 2982 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
 2983 #endif
 2984 
 2985 #ifdef COMPAT_FREEBSD7
 2986 static int
 2987 export_vnode_for_osysctl(struct vnode *vp, int type,
 2988     struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
 2989 {
 2990         int error;
 2991         char *fullpath, *freepath;
 2992         int vfslocked;
 2993 
 2994         bzero(kif, sizeof(*kif));
 2995         kif->kf_structsize = sizeof(*kif);
 2996 
 2997         vref(vp);
 2998         kif->kf_fd = type;
 2999         kif->kf_type = KF_TYPE_VNODE;
 3000         /* This function only handles directories. */
 3001         if (vp->v_type != VDIR) {
 3002                 vrele(vp);
 3003                 return (ENOTDIR);
 3004         }
 3005         kif->kf_vnode_type = KF_VTYPE_VDIR;
 3006 
 3007         /*
 3008          * This is not a true file descriptor, so we set a bogus refcount
 3009          * and offset to indicate these fields should be ignored.
 3010          */
 3011         kif->kf_ref_count = -1;
 3012         kif->kf_offset = -1;
 3013 
 3014         freepath = NULL;
 3015         fullpath = "-";
 3016         FILEDESC_SUNLOCK(fdp);
 3017         vn_fullpath(curthread, vp, &fullpath, &freepath);
 3018         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 3019         vrele(vp);
 3020         VFS_UNLOCK_GIANT(vfslocked);
 3021         strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
 3022         if (freepath != NULL)
 3023                 free(freepath, M_TEMP);
 3024         error = SYSCTL_OUT(req, kif, sizeof(*kif));
 3025         FILEDESC_SLOCK(fdp);
 3026         return (error);
 3027 }
 3028 
 3029 /*
 3030  * Get per-process file descriptors for use by procstat(1), et al.
 3031  */
 3032 static int
 3033 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
 3034 {
 3035         char *fullpath, *freepath;
 3036         struct kinfo_ofile *kif;
 3037         struct filedesc *fdp;
 3038         int error, i, *name;
 3039         struct shmfd *shmfd;
 3040         struct socket *so;
 3041         struct vnode *vp;
 3042         struct ksem *ks;
 3043         struct file *fp;
 3044         struct proc *p;
 3045         struct tty *tp;
 3046         int vfslocked;
 3047 
 3048         name = (int *)arg1;
 3049         error = pget((pid_t)name[0], PGET_CANDEBUG, &p);
 3050         if (error != 0)
 3051                 return (error);
 3052         fdp = fdhold(p);
 3053         PROC_UNLOCK(p);
 3054         if (fdp == NULL)
 3055                 return (ENOENT);
 3056         kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
 3057         FILEDESC_SLOCK(fdp);
 3058         if (fdp->fd_cdir != NULL)
 3059                 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
 3060                                 fdp, req);
 3061         if (fdp->fd_rdir != NULL)
 3062                 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
 3063                                 fdp, req);
 3064         if (fdp->fd_jdir != NULL)
 3065                 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
 3066                                 fdp, req);
 3067         for (i = 0; i < fdp->fd_nfiles; i++) {
 3068                 if ((fp = fdp->fd_ofiles[i]) == NULL)
 3069                         continue;
 3070                 bzero(kif, sizeof(*kif));
 3071                 kif->kf_structsize = sizeof(*kif);
 3072                 ks = NULL;
 3073                 vp = NULL;
 3074                 so = NULL;
 3075                 tp = NULL;
 3076                 shmfd = NULL;
 3077                 kif->kf_fd = i;
 3078 
 3079 #ifdef CAPABILITIES
 3080                 /*
 3081                  * When reporting a capability, most fields will be from the
 3082                  * underlying object, but do mark as a capability. With
 3083                  * ofiledesc, we don't have a field to export the cap_rights_t,
 3084                  * but we do with the new filedesc.
 3085                  */
 3086                 if (fp->f_type == DTYPE_CAPABILITY) {
 3087                         kif->kf_flags |= KF_FLAG_CAPABILITY;
 3088                         (void)cap_funwrap(fp, 0, &fp);
 3089                 }
 3090 #else
 3091                 KASSERT(fp->f_type != DTYPE_CAPABILITY,
 3092                     ("sysctl_kern_proc_ofiledesc: saw capability"));
 3093 #endif
 3094                 switch (fp->f_type) {
 3095                 case DTYPE_VNODE:
 3096                         kif->kf_type = KF_TYPE_VNODE;
 3097                         vp = fp->f_vnode;
 3098                         break;
 3099 
 3100                 case DTYPE_SOCKET:
 3101                         kif->kf_type = KF_TYPE_SOCKET;
 3102                         so = fp->f_data;
 3103                         break;
 3104 
 3105                 case DTYPE_PIPE:
 3106                         kif->kf_type = KF_TYPE_PIPE;
 3107                         break;
 3108 
 3109                 case DTYPE_FIFO:
 3110                         kif->kf_type = KF_TYPE_FIFO;
 3111                         vp = fp->f_vnode;
 3112                         break;
 3113 
 3114                 case DTYPE_KQUEUE:
 3115                         kif->kf_type = KF_TYPE_KQUEUE;
 3116                         break;
 3117 
 3118                 case DTYPE_CRYPTO:
 3119                         kif->kf_type = KF_TYPE_CRYPTO;
 3120                         break;
 3121 
 3122                 case DTYPE_MQUEUE:
 3123                         kif->kf_type = KF_TYPE_MQUEUE;
 3124                         break;
 3125 
 3126                 case DTYPE_SHM:
 3127                         kif->kf_type = KF_TYPE_SHM;
 3128                         shmfd = fp->f_data;
 3129                         break;
 3130 
 3131                 case DTYPE_SEM:
 3132                         kif->kf_type = KF_TYPE_SEM;
 3133                         ks = fp->f_data;
 3134                         break;
 3135 
 3136                 case DTYPE_PTS:
 3137                         kif->kf_type = KF_TYPE_PTS;
 3138                         tp = fp->f_data;
 3139                         break;
 3140 
 3141 #ifdef PROCDESC
 3142                 case DTYPE_PROCDESC:
 3143                         kif->kf_type = KF_TYPE_PROCDESC;
 3144                         break;
 3145 #endif
 3146 
 3147                 default:
 3148                         kif->kf_type = KF_TYPE_UNKNOWN;
 3149                         break;
 3150                 }
 3151                 kif->kf_ref_count = fp->f_count;
 3152                 if (fp->f_flag & FREAD)
 3153                         kif->kf_flags |= KF_FLAG_READ;
 3154                 if (fp->f_flag & FWRITE)
 3155                         kif->kf_flags |= KF_FLAG_WRITE;
 3156                 if (fp->f_flag & FAPPEND)
 3157                         kif->kf_flags |= KF_FLAG_APPEND;
 3158                 if (fp->f_flag & FASYNC)
 3159                         kif->kf_flags |= KF_FLAG_ASYNC;
 3160                 if (fp->f_flag & FFSYNC)
 3161                         kif->kf_flags |= KF_FLAG_FSYNC;
 3162                 if (fp->f_flag & FNONBLOCK)
 3163                         kif->kf_flags |= KF_FLAG_NONBLOCK;
 3164                 if (fp->f_flag & O_DIRECT)
 3165                         kif->kf_flags |= KF_FLAG_DIRECT;
 3166                 if (fp->f_flag & FHASLOCK)
 3167                         kif->kf_flags |= KF_FLAG_HASLOCK;
 3168                 kif->kf_offset = foffset_get(fp);
 3169                 if (vp != NULL) {
 3170                         vref(vp);
 3171                         switch (vp->v_type) {
 3172                         case VNON:
 3173                                 kif->kf_vnode_type = KF_VTYPE_VNON;
 3174                                 break;
 3175                         case VREG:
 3176                                 kif->kf_vnode_type = KF_VTYPE_VREG;
 3177                                 break;
 3178                         case VDIR:
 3179                                 kif->kf_vnode_type = KF_VTYPE_VDIR;
 3180                                 break;
 3181                         case VBLK:
 3182                                 kif->kf_vnode_type = KF_VTYPE_VBLK;
 3183                                 break;
 3184                         case VCHR:
 3185                                 kif->kf_vnode_type = KF_VTYPE_VCHR;
 3186                                 break;
 3187                         case VLNK:
 3188                                 kif->kf_vnode_type = KF_VTYPE_VLNK;
 3189                                 break;
 3190                         case VSOCK:
 3191                                 kif->kf_vnode_type = KF_VTYPE_VSOCK;
 3192                                 break;
 3193                         case VFIFO:
 3194                                 kif->kf_vnode_type = KF_VTYPE_VFIFO;
 3195                                 break;
 3196                         case VBAD:
 3197                                 kif->kf_vnode_type = KF_VTYPE_VBAD;
 3198                                 break;
 3199                         default:
 3200                                 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
 3201                                 break;
 3202                         }
 3203                         /*
 3204                          * It is OK to drop the filedesc lock here as we will
 3205                          * re-validate and re-evaluate its properties when
 3206                          * the loop continues.
 3207                          */
 3208                         freepath = NULL;
 3209                         fullpath = "-";
 3210                         FILEDESC_SUNLOCK(fdp);
 3211                         vn_fullpath(curthread, vp, &fullpath, &freepath);
 3212                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 3213                         vrele(vp);
 3214                         VFS_UNLOCK_GIANT(vfslocked);
 3215                         strlcpy(kif->kf_path, fullpath,
 3216                             sizeof(kif->kf_path));
 3217                         if (freepath != NULL)
 3218                                 free(freepath, M_TEMP);
 3219                         FILEDESC_SLOCK(fdp);
 3220                 }
 3221                 if (so != NULL) {
 3222                         struct sockaddr *sa;
 3223 
 3224                         if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
 3225                             == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
 3226                                 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
 3227                                 free(sa, M_SONAME);
 3228                         }
 3229                         if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
 3230                             == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
 3231                                 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
 3232                                 free(sa, M_SONAME);
 3233                         }
 3234                         kif->kf_sock_domain =
 3235                             so->so_proto->pr_domain->dom_family;
 3236                         kif->kf_sock_type = so->so_type;
 3237                         kif->kf_sock_protocol = so->so_proto->pr_protocol;
 3238                 }
 3239                 if (tp != NULL) {
 3240                         strlcpy(kif->kf_path, tty_devname(tp),
 3241                             sizeof(kif->kf_path));
 3242                 }
 3243                 if (shmfd != NULL)
 3244                         shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path));
 3245                 if (ks != NULL && ksem_info != NULL)
 3246                         ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL);
 3247                 error = SYSCTL_OUT(req, kif, sizeof(*kif));
 3248                 if (error)
 3249                         break;
 3250         }
 3251         FILEDESC_SUNLOCK(fdp);
 3252         fddrop(fdp);
 3253         free(kif, M_TEMP);
 3254         return (0);
 3255 }
 3256 
 3257 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
 3258     sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
 3259 #endif  /* COMPAT_FREEBSD7 */
 3260 
 3261 #ifdef KINFO_FILE_SIZE
 3262 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
 3263 #endif
 3264 
 3265 struct export_fd_buf {
 3266         struct filedesc         *fdp;
 3267         struct sbuf             *sb;
 3268         ssize_t                 remainder;
 3269         struct kinfo_file       kif;
 3270 };
 3271 
 3272 static int
 3273 export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt,
 3274     int64_t offset, int fd_is_cap, cap_rights_t fd_cap_rights,
 3275     struct export_fd_buf *efbuf)
 3276 {
 3277         struct {
 3278                 int     fflag;
 3279                 int     kf_fflag;
 3280         } fflags_table[] = {
 3281                 { FAPPEND, KF_FLAG_APPEND },
 3282                 { FASYNC, KF_FLAG_ASYNC },
 3283                 { FFSYNC, KF_FLAG_FSYNC },
 3284                 { FHASLOCK, KF_FLAG_HASLOCK },
 3285                 { FNONBLOCK, KF_FLAG_NONBLOCK },
 3286                 { FREAD, KF_FLAG_READ },
 3287                 { FWRITE, KF_FLAG_WRITE },
 3288                 { O_CREAT, KF_FLAG_CREAT },
 3289                 { O_DIRECT, KF_FLAG_DIRECT },
 3290                 { O_EXCL, KF_FLAG_EXCL },
 3291                 { O_EXEC, KF_FLAG_EXEC },
 3292                 { O_EXLOCK, KF_FLAG_EXLOCK },
 3293                 { O_NOFOLLOW, KF_FLAG_NOFOLLOW },
 3294                 { O_SHLOCK, KF_FLAG_SHLOCK },
 3295                 { O_TRUNC, KF_FLAG_TRUNC }
 3296         };
 3297 #define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table))
 3298         struct kinfo_file *kif;
 3299         struct vnode *vp;
 3300         int error, locked, vfslocked;
 3301         unsigned int i;
 3302 
 3303         if (efbuf->remainder == 0)
 3304                 return (0);
 3305         kif = &efbuf->kif;
 3306         bzero(kif, sizeof(*kif));
 3307         locked = efbuf->fdp != NULL;
 3308         switch (type) {
 3309         case KF_TYPE_FIFO:
 3310         case KF_TYPE_VNODE:
 3311                 if (locked) {
 3312                         FILEDESC_SUNLOCK(efbuf->fdp);
 3313                         locked = 0;
 3314                 }
 3315                 vp = (struct vnode *)data;
 3316                 error = fill_vnode_info(vp, kif);
 3317                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 3318                 vrele(vp);
 3319                 VFS_UNLOCK_GIANT(vfslocked);
 3320                 break;
 3321         case KF_TYPE_SOCKET:
 3322                 error = fill_socket_info((struct socket *)data, kif);
 3323                 break;
 3324         case KF_TYPE_PIPE:
 3325                 error = fill_pipe_info((struct pipe *)data, kif);
 3326                 break;
 3327         case KF_TYPE_PTS:
 3328                 error = fill_pts_info((struct tty *)data, kif);
 3329                 break;
 3330         case KF_TYPE_PROCDESC:
 3331                 error = fill_procdesc_info((struct procdesc *)data, kif);
 3332                 break;
 3333         case KF_TYPE_SEM:
 3334                 error = fill_sem_info((struct file *)data, kif);
 3335                 break;
 3336         case KF_TYPE_SHM:
 3337                 error = fill_shm_info((struct file *)data, kif);
 3338                 break;
 3339         default:
 3340                 error = 0;
 3341         }
 3342         if (error == 0)
 3343                 kif->kf_status |= KF_ATTR_VALID;
 3344 
 3345         /*
 3346          * Translate file access flags.
 3347          */
 3348         for (i = 0; i < NFFLAGS; i++)
 3349                 if (fflags & fflags_table[i].fflag)
 3350                         kif->kf_flags |=  fflags_table[i].kf_fflag;
 3351         if (fd_is_cap)
 3352                 kif->kf_flags |= KF_FLAG_CAPABILITY;
 3353         if (fd_is_cap)
 3354                 kif->kf_cap_rights = fd_cap_rights;
 3355         kif->kf_fd = fd;
 3356         kif->kf_type = type;
 3357         kif->kf_ref_count = refcnt;
 3358         kif->kf_offset = offset;
 3359         /* Pack record size down */
 3360         kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
 3361             strlen(kif->kf_path) + 1;
 3362         kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
 3363         if (efbuf->remainder != -1) {
 3364                 if (efbuf->remainder < kif->kf_structsize) {
 3365                         /* Terminate export. */
 3366                         efbuf->remainder = 0;
 3367                         if (efbuf->fdp != NULL && !locked)
 3368                                 FILEDESC_SLOCK(efbuf->fdp);
 3369                         return (0);
 3370                 }
 3371                 efbuf->remainder -= kif->kf_structsize;
 3372         }
 3373         if (locked)
 3374                 FILEDESC_SUNLOCK(efbuf->fdp);
 3375         error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize);
 3376         if (efbuf->fdp != NULL)
 3377                 FILEDESC_SLOCK(efbuf->fdp);
 3378         return (error);
 3379 }
 3380 
 3381 /*
 3382  * Store a process file descriptor information to sbuf.
 3383  *
 3384  * Takes a locked proc as argument, and returns with the proc unlocked.
 3385  */
 3386 int
 3387 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
 3388 {
 3389         struct file *fp;
 3390         struct filedesc *fdp;
 3391         struct export_fd_buf *efbuf;
 3392         struct vnode *cttyvp, *textvp, *tracevp;
 3393         int64_t offset;
 3394         void *data;
 3395         int error, i;
 3396         int fd_is_cap, type, refcnt, fflags;
 3397         cap_rights_t fd_cap_rights;
 3398 
 3399         PROC_LOCK_ASSERT(p, MA_OWNED);
 3400 
 3401         /* ktrace vnode */
 3402         tracevp = p->p_tracevp;
 3403         if (tracevp != NULL)
 3404                 vref(tracevp);
 3405         /* text vnode */
 3406         textvp = p->p_textvp;
 3407         if (textvp != NULL)
 3408                 vref(textvp);
 3409         /* Controlling tty. */
 3410         cttyvp = NULL;
 3411         if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
 3412                 cttyvp = p->p_pgrp->pg_session->s_ttyvp;
 3413                 if (cttyvp != NULL)
 3414                         vref(cttyvp);
 3415         }
 3416         fdp = fdhold(p);
 3417         PROC_UNLOCK(p);
 3418         efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
 3419         efbuf->fdp = NULL;
 3420         efbuf->sb = sb;
 3421         efbuf->remainder = maxlen;
 3422         if (tracevp != NULL)
 3423                 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE,
 3424                     FREAD | FWRITE, -1, -1, 0, 0, efbuf);
 3425         if (textvp != NULL)
 3426                 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT,
 3427                     FREAD, -1, -1, 0, 0, efbuf);
 3428         if (cttyvp != NULL)
 3429                 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY,
 3430                     FREAD | FWRITE, -1, -1, 0, 0, efbuf);
 3431         error = 0;
 3432         if (fdp == NULL)
 3433                 goto fail;
 3434         efbuf->fdp = fdp;
 3435         FILEDESC_SLOCK(fdp);
 3436         /* working directory */
 3437         if (fdp->fd_cdir != NULL) {
 3438                 vref(fdp->fd_cdir);
 3439                 data = fdp->fd_cdir;
 3440                 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD,
 3441                     FREAD, -1, -1, 0, 0, efbuf);
 3442         }
 3443         /* root directory */
 3444         if (fdp->fd_rdir != NULL) {
 3445                 vref(fdp->fd_rdir);
 3446                 data = fdp->fd_rdir;
 3447                 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT,
 3448                     FREAD, -1, -1, 0, 0, efbuf);
 3449         }
 3450         /* jail directory */
 3451         if (fdp->fd_jdir != NULL) {
 3452                 vref(fdp->fd_jdir);
 3453                 data = fdp->fd_jdir;
 3454                 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL,
 3455                     FREAD, -1, -1, 0, 0, efbuf);
 3456         }
 3457         for (i = 0; i < fdp->fd_nfiles; i++) {
 3458                 if ((fp = fdp->fd_ofiles[i]) == NULL)
 3459                         continue;
 3460                 data = NULL;
 3461                 fd_is_cap = 0;
 3462                 fd_cap_rights = 0;
 3463 
 3464 #ifdef CAPABILITIES
 3465                 /*
 3466                  * When reporting a capability, most fields will be from the
 3467                  * underlying object, but do mark as a capability and export
 3468                  * the capability rights mask.
 3469                  */
 3470                 if (fp->f_type == DTYPE_CAPABILITY) {
 3471                         fd_is_cap = 1;
 3472                         fd_cap_rights = cap_rights(fp);
 3473                         (void)cap_funwrap(fp, 0, &fp);
 3474                 }
 3475 #else /* !CAPABILITIES */
 3476                 KASSERT(fp->f_type != DTYPE_CAPABILITY,
 3477                     ("sysctl_kern_proc_filedesc: saw capability"));
 3478 #endif
 3479                 switch (fp->f_type) {
 3480                 case DTYPE_VNODE:
 3481                         type = KF_TYPE_VNODE;
 3482                         vref(fp->f_vnode);
 3483                         data = fp->f_vnode;
 3484                         break;
 3485 
 3486                 case DTYPE_SOCKET:
 3487                         type = KF_TYPE_SOCKET;
 3488                         data = fp->f_data;
 3489                         break;
 3490 
 3491                 case DTYPE_PIPE:
 3492                         type = KF_TYPE_PIPE;
 3493                         data = fp->f_data;
 3494                         break;
 3495 
 3496                 case DTYPE_FIFO:
 3497                         type = KF_TYPE_FIFO;
 3498                         vref(fp->f_vnode);
 3499                         data = fp->f_vnode;
 3500                         break;
 3501 
 3502                 case DTYPE_KQUEUE:
 3503                         type = KF_TYPE_KQUEUE;
 3504                         break;
 3505 
 3506                 case DTYPE_CRYPTO:
 3507                         type = KF_TYPE_CRYPTO;
 3508                         break;
 3509 
 3510                 case DTYPE_MQUEUE:
 3511                         type = KF_TYPE_MQUEUE;
 3512                         break;
 3513 
 3514                 case DTYPE_SHM:
 3515                         type = KF_TYPE_SHM;
 3516                         data = fp;
 3517                         break;
 3518 
 3519                 case DTYPE_SEM:
 3520                         type = KF_TYPE_SEM;
 3521                         data = fp;
 3522                         break;
 3523 
 3524                 case DTYPE_PTS:
 3525                         type = KF_TYPE_PTS;
 3526                         data = fp->f_data;
 3527                         break;
 3528 
 3529 #ifdef PROCDESC
 3530                 case DTYPE_PROCDESC:
 3531                         type = KF_TYPE_PROCDESC;
 3532                         data = fp->f_data;
 3533                         break;
 3534 #endif
 3535 
 3536                 default:
 3537                         type = KF_TYPE_UNKNOWN;
 3538                         break;
 3539                 }
 3540                 refcnt = fp->f_count;
 3541                 fflags = fp->f_flag;
 3542                 offset = foffset_get(fp);
 3543 
 3544                 /*
 3545                  * Create sysctl entry.
 3546                  * It is OK to drop the filedesc lock here as we will
 3547                  * re-validate and re-evaluate its properties when
 3548                  * the loop continues.
 3549                  */
 3550                 error = export_fd_to_sb(data, type, i, fflags, refcnt,
 3551                     offset, fd_is_cap, fd_cap_rights, efbuf);
 3552                 if (error)
 3553                         break;
 3554         }
 3555         FILEDESC_SUNLOCK(fdp);
 3556         fddrop(fdp);
 3557 fail:
 3558         free(efbuf, M_TEMP);
 3559         return (error);
 3560 }
 3561 
 3562 #define FILEDESC_SBUF_SIZE      (sizeof(struct kinfo_file) * 5)
 3563 
 3564 /*
 3565  * Get per-process file descriptors for use by procstat(1), et al.
 3566  */
 3567 static int
 3568 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
 3569 {
 3570         struct sbuf sb;
 3571         struct proc *p;
 3572         ssize_t maxlen;
 3573         int error, error2, *name;
 3574 
 3575         name = (int *)arg1;
 3576 
 3577         sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
 3578         error = pget((pid_t)name[0], PGET_CANDEBUG, &p);
 3579         if (error != 0) {
 3580                 sbuf_delete(&sb);
 3581                 return (error);
 3582         }
 3583         maxlen = req->oldptr != NULL ? req->oldlen : -1;
 3584         error = kern_proc_filedesc_out(p, &sb, maxlen);
 3585         error2 = sbuf_finish(&sb);
 3586         sbuf_delete(&sb);
 3587         return (error != 0 ? error : error2);
 3588 }
 3589 
 3590 int
 3591 vntype_to_kinfo(int vtype)
 3592 {
 3593         struct {
 3594                 int     vtype;
 3595                 int     kf_vtype;
 3596         } vtypes_table[] = {
 3597                 { VBAD, KF_VTYPE_VBAD },
 3598                 { VBLK, KF_VTYPE_VBLK },
 3599                 { VCHR, KF_VTYPE_VCHR },
 3600                 { VDIR, KF_VTYPE_VDIR },
 3601                 { VFIFO, KF_VTYPE_VFIFO },
 3602                 { VLNK, KF_VTYPE_VLNK },
 3603                 { VNON, KF_VTYPE_VNON },
 3604                 { VREG, KF_VTYPE_VREG },
 3605                 { VSOCK, KF_VTYPE_VSOCK }
 3606         };
 3607 #define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table))
 3608         unsigned int i;
 3609 
 3610         /*
 3611          * Perform vtype translation.
 3612          */
 3613         for (i = 0; i < NVTYPES; i++)
 3614                 if (vtypes_table[i].vtype == vtype)
 3615                         break;
 3616         if (i < NVTYPES)
 3617                 return (vtypes_table[i].kf_vtype);
 3618 
 3619         return (KF_VTYPE_UNKNOWN);
 3620 }
 3621 
 3622 static int
 3623 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif)
 3624 {
 3625         struct vattr va;
 3626         char *fullpath, *freepath;
 3627         int error, vfslocked;
 3628 
 3629         if (vp == NULL)
 3630                 return (1);
 3631         kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
 3632         freepath = NULL;
 3633         fullpath = "-";
 3634         error = vn_fullpath(curthread, vp, &fullpath, &freepath);
 3635         if (error == 0) {
 3636                 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
 3637         }
 3638         if (freepath != NULL)
 3639                 free(freepath, M_TEMP);
 3640 
 3641         /*
 3642          * Retrieve vnode attributes.
 3643          */
 3644         va.va_fsid = VNOVAL;
 3645         va.va_rdev = NODEV;
 3646         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 3647         vn_lock(vp, LK_SHARED | LK_RETRY);
 3648         error = VOP_GETATTR(vp, &va, curthread->td_ucred);
 3649         VOP_UNLOCK(vp, 0);
 3650         VFS_UNLOCK_GIANT(vfslocked);
 3651         if (error != 0)
 3652                 return (error);
 3653         if (va.va_fsid != VNOVAL)
 3654                 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
 3655         else
 3656                 kif->kf_un.kf_file.kf_file_fsid =
 3657                     vp->v_mount->mnt_stat.f_fsid.val[0];
 3658         kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
 3659         kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
 3660         kif->kf_un.kf_file.kf_file_size = va.va_size;
 3661         kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
 3662         return (0);
 3663 }
 3664 
 3665 static int
 3666 fill_socket_info(struct socket *so, struct kinfo_file *kif)
 3667 {
 3668         struct sockaddr *sa;
 3669         struct inpcb *inpcb;
 3670         struct unpcb *unpcb;
 3671         int error;
 3672 
 3673         if (so == NULL)
 3674                 return (1);
 3675         kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
 3676         kif->kf_sock_type = so->so_type;
 3677         kif->kf_sock_protocol = so->so_proto->pr_protocol;
 3678         kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
 3679         switch(kif->kf_sock_domain) {
 3680         case AF_INET:
 3681         case AF_INET6:
 3682                 if (kif->kf_sock_protocol == IPPROTO_TCP) {
 3683                         if (so->so_pcb != NULL) {
 3684                                 inpcb = (struct inpcb *)(so->so_pcb);
 3685                                 kif->kf_un.kf_sock.kf_sock_inpcb =
 3686                                     (uintptr_t)inpcb->inp_ppcb;
 3687                         }
 3688                 }
 3689                 break;
 3690         case AF_UNIX:
 3691                 if (so->so_pcb != NULL) {
 3692                         unpcb = (struct unpcb *)(so->so_pcb);
 3693                         if (unpcb->unp_conn) {
 3694                                 kif->kf_un.kf_sock.kf_sock_unpconn =
 3695                                     (uintptr_t)unpcb->unp_conn;
 3696                                 kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
 3697                                     so->so_rcv.sb_state;
 3698                                 kif->kf_un.kf_sock.kf_sock_snd_sb_state =
 3699                                     so->so_snd.sb_state;
 3700                         }
 3701                 }
 3702                 break;
 3703         }
 3704         error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
 3705         if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
 3706                 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
 3707                 free(sa, M_SONAME);
 3708         }
 3709         error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
 3710         if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
 3711                 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
 3712                 free(sa, M_SONAME);
 3713         }
 3714         strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
 3715             sizeof(kif->kf_path));
 3716         return (0);
 3717 }
 3718 
 3719 static int
 3720 fill_pts_info(struct tty *tp, struct kinfo_file *kif)
 3721 {
 3722 
 3723         if (tp == NULL)
 3724                 return (1);
 3725         kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp);
 3726         strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path));
 3727         return (0);
 3728 }
 3729 
 3730 static int
 3731 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
 3732 {
 3733 
 3734         if (pi == NULL)
 3735                 return (1);
 3736         kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
 3737         kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
 3738         kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
 3739         return (0);
 3740 }
 3741 
 3742 static int
 3743 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
 3744 {
 3745 
 3746         if (pdp == NULL)
 3747                 return (1);
 3748         kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
 3749         return (0);
 3750 }
 3751 
 3752 static int
 3753 fill_sem_info(struct file *fp, struct kinfo_file *kif)
 3754 {
 3755         struct thread *td;
 3756         struct stat sb;
 3757 
 3758         td = curthread;
 3759         if (fp->f_data == NULL)
 3760                 return (1);
 3761         if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
 3762                 return (1);
 3763         if (ksem_info == NULL)
 3764                 return (1);
 3765         ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path),
 3766             &kif->kf_un.kf_sem.kf_sem_value);
 3767         kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode;
 3768         return (0);
 3769 }
 3770 
 3771 static int
 3772 fill_shm_info(struct file *fp, struct kinfo_file *kif)
 3773 {
 3774         struct thread *td;
 3775         struct stat sb;
 3776 
 3777         td = curthread;
 3778         if (fp->f_data == NULL)
 3779                 return (1);
 3780         if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
 3781                 return (1);
 3782         shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path));
 3783         kif->kf_un.kf_file.kf_file_mode = sb.st_mode;
 3784         kif->kf_un.kf_file.kf_file_size = sb.st_size;
 3785         return (0);
 3786 }
 3787 
 3788 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
 3789     sysctl_kern_proc_filedesc, "Process filedesc entries");
 3790 
 3791 #ifdef DDB
 3792 /*
 3793  * For the purposes of debugging, generate a human-readable string for the
 3794  * file type.
 3795  */
 3796 static const char *
 3797 file_type_to_name(short type)
 3798 {
 3799 
 3800         switch (type) {
 3801         case 0:
 3802                 return ("zero");
 3803         case DTYPE_VNODE:
 3804                 return ("vnod");
 3805         case DTYPE_SOCKET:
 3806                 return ("sock");
 3807         case DTYPE_PIPE:
 3808                 return ("pipe");
 3809         case DTYPE_FIFO:
 3810                 return ("fifo");
 3811         case DTYPE_KQUEUE:
 3812                 return ("kque");
 3813         case DTYPE_CRYPTO:
 3814                 return ("crpt");
 3815         case DTYPE_MQUEUE:
 3816                 return ("mque");
 3817         case DTYPE_SHM:
 3818                 return ("shm");
 3819         case DTYPE_SEM:
 3820                 return ("ksem");
 3821         default:
 3822                 return ("unkn");
 3823         }
 3824 }
 3825 
 3826 /*
 3827  * For the purposes of debugging, identify a process (if any, perhaps one of
 3828  * many) that references the passed file in its file descriptor array. Return
 3829  * NULL if none.
 3830  */
 3831 static struct proc *
 3832 file_to_first_proc(struct file *fp)
 3833 {
 3834         struct filedesc *fdp;
 3835         struct proc *p;
 3836         int n;
 3837 
 3838         FOREACH_PROC_IN_SYSTEM(p) {
 3839                 if (p->p_state == PRS_NEW)
 3840                         continue;
 3841                 fdp = p->p_fd;
 3842                 if (fdp == NULL)
 3843                         continue;
 3844                 for (n = 0; n < fdp->fd_nfiles; n++) {
 3845                         if (fp == fdp->fd_ofiles[n])
 3846                                 return (p);
 3847                 }
 3848         }
 3849         return (NULL);
 3850 }
 3851 
 3852 static void
 3853 db_print_file(struct file *fp, int header)
 3854 {
 3855         struct proc *p;
 3856 
 3857         if (header)
 3858                 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
 3859                     "File", "Type", "Data", "Flag", "GCFl", "Count",
 3860                     "MCount", "Vnode", "FPID", "FCmd");
 3861         p = file_to_first_proc(fp);
 3862         db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
 3863             file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
 3864             0, fp->f_count, 0, fp->f_vnode,
 3865             p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
 3866 }
 3867 
 3868 DB_SHOW_COMMAND(file, db_show_file)
 3869 {
 3870         struct file *fp;
 3871 
 3872         if (!have_addr) {
 3873                 db_printf("usage: show file <addr>\n");
 3874                 return;
 3875         }
 3876         fp = (struct file *)addr;
 3877         db_print_file(fp, 1);
 3878 }
 3879 
 3880 DB_SHOW_COMMAND(files, db_show_files)
 3881 {
 3882         struct filedesc *fdp;
 3883         struct file *fp;
 3884         struct proc *p;
 3885         int header;
 3886         int n;
 3887 
 3888         header = 1;
 3889         FOREACH_PROC_IN_SYSTEM(p) {
 3890                 if (p->p_state == PRS_NEW)
 3891                         continue;
 3892                 if ((fdp = p->p_fd) == NULL)
 3893                         continue;
 3894                 for (n = 0; n < fdp->fd_nfiles; ++n) {
 3895                         if ((fp = fdp->fd_ofiles[n]) == NULL)
 3896                                 continue;
 3897                         db_print_file(fp, header);
 3898                         header = 0;
 3899                 }
 3900         }
 3901 }
 3902 #endif
 3903 
 3904 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
 3905     &maxfilesperproc, 0, "Maximum files allowed open per process");
 3906 
 3907 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
 3908     &maxfiles, 0, "Maximum number of files");
 3909 
 3910 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
 3911     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
 3912 
 3913 /* ARGSUSED*/
 3914 static void
 3915 filelistinit(void *dummy)
 3916 {
 3917 
 3918         file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
 3919             NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
 3920         mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
 3921         mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
 3922 }
 3923 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
 3924 
 3925 /*-------------------------------------------------------------------*/
 3926 
 3927 static int
 3928 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
 3929     int flags, struct thread *td)
 3930 {
 3931 
 3932         return (EBADF);
 3933 }
 3934 
 3935 static int
 3936 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
 3937     struct thread *td)
 3938 {
 3939 
 3940         return (EINVAL);
 3941 }
 3942 
 3943 static int
 3944 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
 3945     struct thread *td)
 3946 {
 3947 
 3948         return (EBADF);
 3949 }
 3950 
 3951 static int
 3952 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
 3953     struct thread *td)
 3954 {
 3955 
 3956         return (0);
 3957 }
 3958 
 3959 static int
 3960 badfo_kqfilter(struct file *fp, struct knote *kn)
 3961 {
 3962 
 3963         return (EBADF);
 3964 }
 3965 
 3966 static int
 3967 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
 3968     struct thread *td)
 3969 {
 3970 
 3971         return (EBADF);
 3972 }
 3973 
 3974 static int
 3975 badfo_close(struct file *fp, struct thread *td)
 3976 {
 3977 
 3978         return (EBADF);
 3979 }
 3980 
 3981 static int
 3982 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 3983     struct thread *td)
 3984 {
 3985 
 3986         return (EBADF);
 3987 }
 3988 
 3989 static int
 3990 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 3991     struct thread *td)
 3992 {
 3993 
 3994         return (EBADF);
 3995 }
 3996 
 3997 struct fileops badfileops = {
 3998         .fo_read = badfo_readwrite,
 3999         .fo_write = badfo_readwrite,
 4000         .fo_truncate = badfo_truncate,
 4001         .fo_ioctl = badfo_ioctl,
 4002         .fo_poll = badfo_poll,
 4003         .fo_kqfilter = badfo_kqfilter,
 4004         .fo_stat = badfo_stat,
 4005         .fo_close = badfo_close,
 4006         .fo_chmod = badfo_chmod,
 4007         .fo_chown = badfo_chown,
 4008 };
 4009 
 4010 int
 4011 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 4012     struct thread *td)
 4013 {
 4014 
 4015         return (EINVAL);
 4016 }
 4017 
 4018 int
 4019 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 4020     struct thread *td)
 4021 {
 4022 
 4023         return (EINVAL);
 4024 }
 4025 
 4026 /*-------------------------------------------------------------------*/
 4027 
 4028 /*
 4029  * File Descriptor pseudo-device driver (/dev/fd/).
 4030  *
 4031  * Opening minor device N dup()s the file (if any) connected to file
 4032  * descriptor N belonging to the calling process.  Note that this driver
 4033  * consists of only the ``open()'' routine, because all subsequent
 4034  * references to this file will be direct to the other driver.
 4035  *
 4036  * XXX: we could give this one a cloning event handler if necessary.
 4037  */
 4038 
 4039 /* ARGSUSED */
 4040 static int
 4041 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
 4042 {
 4043 
 4044         /*
 4045          * XXX Kludge: set curthread->td_dupfd to contain the value of the
 4046          * the file descriptor being sought for duplication. The error
 4047          * return ensures that the vnode for this device will be released
 4048          * by vn_open. Open will detect this special error and take the
 4049          * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
 4050          * will simply report the error.
 4051          */
 4052         td->td_dupfd = dev2unit(dev);
 4053         return (ENODEV);
 4054 }
 4055 
 4056 static struct cdevsw fildesc_cdevsw = {
 4057         .d_version =    D_VERSION,
 4058         .d_open =       fdopen,
 4059         .d_name =       "FD",
 4060 };
 4061 
 4062 static void
 4063 fildesc_drvinit(void *unused)
 4064 {
 4065         struct cdev *dev;
 4066 
 4067         dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
 4068             UID_ROOT, GID_WHEEL, 0666, "fd/0");
 4069         make_dev_alias(dev, "stdin");
 4070         dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
 4071             UID_ROOT, GID_WHEEL, 0666, "fd/1");
 4072         make_dev_alias(dev, "stdout");
 4073         dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
 4074             UID_ROOT, GID_WHEEL, 0666, "fd/2");
 4075         make_dev_alias(dev, "stderr");
 4076 }
 4077 
 4078 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);

Cache object: f1b27c3693541509738683dbabd0caab


[ 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.