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_exec.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) 1993, David Greenman
    3  * All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    8  * 1. Redistributions of source code must retain the above copyright
    9  *    notice, this list of conditions and the following disclaimer.
   10  * 2. Redistributions in binary form must reproduce the above copyright
   11  *    notice, this list of conditions and the following disclaimer in the
   12  *    documentation and/or other materials provided with the distribution.
   13  *
   14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   24  * SUCH DAMAGE.
   25  */
   26 
   27 #include <sys/cdefs.h>
   28 __FBSDID("$FreeBSD: releng/9.2/sys/kern/kern_exec.c 267018 2014-06-03 19:03:11Z delphij $");
   29 
   30 #include "opt_capsicum.h"
   31 #include "opt_hwpmc_hooks.h"
   32 #include "opt_kdtrace.h"
   33 #include "opt_ktrace.h"
   34 #include "opt_vm.h"
   35 
   36 #include <sys/param.h>
   37 #include <sys/capability.h>
   38 #include <sys/systm.h>
   39 #include <sys/capability.h>
   40 #include <sys/eventhandler.h>
   41 #include <sys/lock.h>
   42 #include <sys/mutex.h>
   43 #include <sys/sysproto.h>
   44 #include <sys/signalvar.h>
   45 #include <sys/kernel.h>
   46 #include <sys/mount.h>
   47 #include <sys/filedesc.h>
   48 #include <sys/fcntl.h>
   49 #include <sys/acct.h>
   50 #include <sys/exec.h>
   51 #include <sys/imgact.h>
   52 #include <sys/imgact_elf.h>
   53 #include <sys/wait.h>
   54 #include <sys/malloc.h>
   55 #include <sys/priv.h>
   56 #include <sys/proc.h>
   57 #include <sys/pioctl.h>
   58 #include <sys/namei.h>
   59 #include <sys/resourcevar.h>
   60 #include <sys/sched.h>
   61 #include <sys/sdt.h>
   62 #include <sys/sf_buf.h>
   63 #include <sys/syscallsubr.h>
   64 #include <sys/sysent.h>
   65 #include <sys/shm.h>
   66 #include <sys/sysctl.h>
   67 #include <sys/vnode.h>
   68 #include <sys/stat.h>
   69 #ifdef KTRACE
   70 #include <sys/ktrace.h>
   71 #endif
   72 
   73 #include <vm/vm.h>
   74 #include <vm/vm_param.h>
   75 #include <vm/pmap.h>
   76 #include <vm/vm_page.h>
   77 #include <vm/vm_map.h>
   78 #include <vm/vm_kern.h>
   79 #include <vm/vm_extern.h>
   80 #include <vm/vm_object.h>
   81 #include <vm/vm_pager.h>
   82 
   83 #ifdef  HWPMC_HOOKS
   84 #include <sys/pmckern.h>
   85 #endif
   86 
   87 #include <machine/reg.h>
   88 
   89 #include <security/audit/audit.h>
   90 #include <security/mac/mac_framework.h>
   91 
   92 #ifdef KDTRACE_HOOKS
   93 #include <sys/dtrace_bsd.h>
   94 dtrace_execexit_func_t  dtrace_fasttrap_exec;
   95 #endif
   96 
   97 SDT_PROVIDER_DECLARE(proc);
   98 SDT_PROBE_DEFINE(proc, kernel, , exec, exec);
   99 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
  100 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
  101 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
  102 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
  103 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
  104 
  105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
  106 
  107 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
  108 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
  109 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
  110 static int do_execve(struct thread *td, struct image_args *args,
  111     struct mac *mac_p);
  112 
  113 /* XXX This should be vm_size_t. */
  114 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
  115     NULL, 0, sysctl_kern_ps_strings, "LU", "");
  116 
  117 /* XXX This should be vm_size_t. */
  118 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
  119     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
  120 
  121 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
  122     NULL, 0, sysctl_kern_stackprot, "I", "");
  123 
  124 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
  125 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 
  126     &ps_arg_cache_limit, 0, "");
  127 
  128 static int map_at_zero = 0;
  129 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
  130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
  131     "Permit processes to map an object at virtual address 0.");
  132 
  133 static int
  134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
  135 {
  136         struct proc *p;
  137         int error;
  138 
  139         p = curproc;
  140 #ifdef SCTL_MASK32
  141         if (req->flags & SCTL_MASK32) {
  142                 unsigned int val;
  143                 val = (unsigned int)p->p_sysent->sv_psstrings;
  144                 error = SYSCTL_OUT(req, &val, sizeof(val));
  145         } else
  146 #endif
  147                 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
  148                    sizeof(p->p_sysent->sv_psstrings));
  149         return error;
  150 }
  151 
  152 static int
  153 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
  154 {
  155         struct proc *p;
  156         int error;
  157 
  158         p = curproc;
  159 #ifdef SCTL_MASK32
  160         if (req->flags & SCTL_MASK32) {
  161                 unsigned int val;
  162                 val = (unsigned int)p->p_sysent->sv_usrstack;
  163                 error = SYSCTL_OUT(req, &val, sizeof(val));
  164         } else
  165 #endif
  166                 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
  167                     sizeof(p->p_sysent->sv_usrstack));
  168         return error;
  169 }
  170 
  171 static int
  172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
  173 {
  174         struct proc *p;
  175 
  176         p = curproc;
  177         return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
  178             sizeof(p->p_sysent->sv_stackprot)));
  179 }
  180 
  181 /*
  182  * Each of the items is a pointer to a `const struct execsw', hence the
  183  * double pointer here.
  184  */
  185 static const struct execsw **execsw;
  186 
  187 #ifndef _SYS_SYSPROTO_H_
  188 struct execve_args {
  189         char    *fname; 
  190         char    **argv;
  191         char    **envv; 
  192 };
  193 #endif
  194 
  195 int
  196 sys_execve(td, uap)
  197         struct thread *td;
  198         struct execve_args /* {
  199                 char *fname;
  200                 char **argv;
  201                 char **envv;
  202         } */ *uap;
  203 {
  204         int error;
  205         struct image_args args;
  206 
  207         error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
  208             uap->argv, uap->envv);
  209         if (error == 0)
  210                 error = kern_execve(td, &args, NULL);
  211         return (error);
  212 }
  213 
  214 #ifndef _SYS_SYSPROTO_H_
  215 struct fexecve_args {
  216         int     fd;
  217         char    **argv;
  218         char    **envv;
  219 }
  220 #endif
  221 int
  222 sys_fexecve(struct thread *td, struct fexecve_args *uap)
  223 {
  224         int error;
  225         struct image_args args;
  226 
  227         error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
  228             uap->argv, uap->envv);
  229         if (error == 0) {
  230                 args.fd = uap->fd;
  231                 error = kern_execve(td, &args, NULL);
  232         }
  233         return (error);
  234 }
  235 
  236 #ifndef _SYS_SYSPROTO_H_
  237 struct __mac_execve_args {
  238         char    *fname;
  239         char    **argv;
  240         char    **envv;
  241         struct mac      *mac_p;
  242 };
  243 #endif
  244 
  245 int
  246 sys___mac_execve(td, uap)
  247         struct thread *td;
  248         struct __mac_execve_args /* {
  249                 char *fname;
  250                 char **argv;
  251                 char **envv;
  252                 struct mac *mac_p;
  253         } */ *uap;
  254 {
  255 #ifdef MAC
  256         int error;
  257         struct image_args args;
  258 
  259         error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
  260             uap->argv, uap->envv);
  261         if (error == 0)
  262                 error = kern_execve(td, &args, uap->mac_p);
  263         return (error);
  264 #else
  265         return (ENOSYS);
  266 #endif
  267 }
  268 
  269 /*
  270  * XXX: kern_execve has the astonishing property of not always returning to
  271  * the caller.  If sufficiently bad things happen during the call to
  272  * do_execve(), it can end up calling exit1(); as a result, callers must
  273  * avoid doing anything which they might need to undo (e.g., allocating
  274  * memory).
  275  */
  276 int
  277 kern_execve(td, args, mac_p)
  278         struct thread *td;
  279         struct image_args *args;
  280         struct mac *mac_p;
  281 {
  282         struct proc *p = td->td_proc;
  283         struct vmspace *oldvmspace;
  284         int error;
  285 
  286         AUDIT_ARG_ARGV(args->begin_argv, args->argc,
  287             args->begin_envv - args->begin_argv);
  288         AUDIT_ARG_ENVV(args->begin_envv, args->envc,
  289             args->endp - args->begin_envv);
  290         if (p->p_flag & P_HADTHREADS) {
  291                 PROC_LOCK(p);
  292                 if (thread_single(SINGLE_BOUNDARY)) {
  293                         PROC_UNLOCK(p);
  294                         exec_free_args(args);
  295                         return (ERESTART);      /* Try again later. */
  296                 }
  297                 PROC_UNLOCK(p);
  298         }
  299 
  300         KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve"));
  301         oldvmspace = td->td_proc->p_vmspace;
  302         error = do_execve(td, args, mac_p);
  303 
  304         if (p->p_flag & P_HADTHREADS) {
  305                 PROC_LOCK(p);
  306                 /*
  307                  * If success, we upgrade to SINGLE_EXIT state to
  308                  * force other threads to suicide.
  309                  */
  310                 if (error == 0)
  311                         thread_single(SINGLE_EXIT);
  312                 else
  313                         thread_single_end();
  314                 PROC_UNLOCK(p);
  315         }
  316         if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
  317                 KASSERT(td->td_proc->p_vmspace != oldvmspace,
  318                     ("oldvmspace still used"));
  319                 vmspace_free(oldvmspace);
  320                 td->td_pflags &= ~TDP_EXECVMSPC;
  321         }
  322 
  323         return (error);
  324 }
  325 
  326 /*
  327  * In-kernel implementation of execve().  All arguments are assumed to be
  328  * userspace pointers from the passed thread.
  329  */
  330 static int
  331 do_execve(td, args, mac_p)
  332         struct thread *td;
  333         struct image_args *args;
  334         struct mac *mac_p;
  335 {
  336         struct proc *p = td->td_proc;
  337         struct nameidata nd;
  338         struct ucred *newcred = NULL, *oldcred;
  339         struct uidinfo *euip;
  340         register_t *stack_base;
  341         int error, i;
  342         struct image_params image_params, *imgp;
  343         struct vattr attr;
  344         int (*img_first)(struct image_params *);
  345         struct pargs *oldargs = NULL, *newargs = NULL;
  346         struct sigacts *oldsigacts, *newsigacts;
  347 #ifdef KTRACE
  348         struct vnode *tracevp = NULL;
  349         struct ucred *tracecred = NULL;
  350 #endif
  351         struct vnode *textvp = NULL, *binvp = NULL;
  352         int credential_changing;
  353         int vfslocked;
  354         int textset;
  355 #ifdef MAC
  356         struct label *interpvplabel = NULL;
  357         int will_transition;
  358 #endif
  359 #ifdef HWPMC_HOOKS
  360         struct pmckern_procexec pe;
  361 #endif
  362         static const char fexecv_proc_title[] = "(fexecv)";
  363 
  364         vfslocked = 0;
  365         imgp = &image_params;
  366 
  367         /*
  368          * Lock the process and set the P_INEXEC flag to indicate that
  369          * it should be left alone until we're done here.  This is
  370          * necessary to avoid race conditions - e.g. in ptrace() -
  371          * that might allow a local user to illicitly obtain elevated
  372          * privileges.
  373          */
  374         PROC_LOCK(p);
  375         KASSERT((p->p_flag & P_INEXEC) == 0,
  376             ("%s(): process already has P_INEXEC flag", __func__));
  377         p->p_flag |= P_INEXEC;
  378         PROC_UNLOCK(p);
  379 
  380         /*
  381          * Initialize part of the common data
  382          */
  383         imgp->proc = p;
  384         imgp->execlabel = NULL;
  385         imgp->attr = &attr;
  386         imgp->entry_addr = 0;
  387         imgp->reloc_base = 0;
  388         imgp->vmspace_destroyed = 0;
  389         imgp->interpreted = 0;
  390         imgp->opened = 0;
  391         imgp->interpreter_name = NULL;
  392         imgp->auxargs = NULL;
  393         imgp->vp = NULL;
  394         imgp->object = NULL;
  395         imgp->firstpage = NULL;
  396         imgp->ps_strings = 0;
  397         imgp->auxarg_size = 0;
  398         imgp->args = args;
  399         imgp->execpath = imgp->freepath = NULL;
  400         imgp->execpathp = 0;
  401         imgp->canary = 0;
  402         imgp->canarylen = 0;
  403         imgp->pagesizes = 0;
  404         imgp->pagesizeslen = 0;
  405         imgp->stack_prot = 0;
  406 
  407 #ifdef MAC
  408         error = mac_execve_enter(imgp, mac_p);
  409         if (error)
  410                 goto exec_fail;
  411 #endif
  412 
  413         imgp->image_header = NULL;
  414 
  415         /*
  416          * Translate the file name. namei() returns a vnode pointer
  417          *      in ni_vp amoung other things.
  418          *
  419          * XXXAUDIT: It would be desirable to also audit the name of the
  420          * interpreter if this is an interpreted binary.
  421          */
  422         if (args->fname != NULL) {
  423                 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
  424                     | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
  425         }
  426 
  427         SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
  428 
  429 interpret:
  430         if (args->fname != NULL) {
  431 #ifdef CAPABILITY_MODE
  432                 /*
  433                  * While capability mode can't reach this point via direct
  434                  * path arguments to execve(), we also don't allow
  435                  * interpreters to be used in capability mode (for now).
  436                  * Catch indirect lookups and return a permissions error.
  437                  */
  438                 if (IN_CAPABILITY_MODE(td)) {
  439                         error = ECAPMODE;
  440                         goto exec_fail;
  441                 }
  442 #endif
  443                 error = namei(&nd);
  444                 if (error)
  445                         goto exec_fail;
  446 
  447                 vfslocked = NDHASGIANT(&nd);
  448                 binvp  = nd.ni_vp;
  449                 imgp->vp = binvp;
  450         } else {
  451                 AUDIT_ARG_FD(args->fd);
  452                 /*
  453                  * Some might argue that CAP_READ and/or CAP_MMAP should also
  454                  * be required here; such arguments will be entertained.
  455                  *
  456                  * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
  457                  */
  458                 error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
  459                 if (error)
  460                         goto exec_fail;
  461                 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
  462                 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
  463                 AUDIT_ARG_VNODE1(binvp);
  464                 imgp->vp = binvp;
  465         }
  466 
  467         /*
  468          * Check file permissions (also 'opens' file)
  469          */
  470         error = exec_check_permissions(imgp);
  471         if (error)
  472                 goto exec_fail_dealloc;
  473 
  474         imgp->object = imgp->vp->v_object;
  475         if (imgp->object != NULL)
  476                 vm_object_reference(imgp->object);
  477 
  478         /*
  479          * Set VV_TEXT now so no one can write to the executable while we're
  480          * activating it.
  481          *
  482          * Remember if this was set before and unset it in case this is not
  483          * actually an executable image.
  484          */
  485         textset = VOP_IS_TEXT(imgp->vp);
  486         VOP_SET_TEXT(imgp->vp);
  487 
  488         error = exec_map_first_page(imgp);
  489         if (error)
  490                 goto exec_fail_dealloc;
  491 
  492         imgp->proc->p_osrel = 0;
  493         /*
  494          *      If the current process has a special image activator it
  495          *      wants to try first, call it.   For example, emulating shell
  496          *      scripts differently.
  497          */
  498         error = -1;
  499         if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
  500                 error = img_first(imgp);
  501 
  502         /*
  503          *      Loop through the list of image activators, calling each one.
  504          *      An activator returns -1 if there is no match, 0 on success,
  505          *      and an error otherwise.
  506          */
  507         for (i = 0; error == -1 && execsw[i]; ++i) {
  508                 if (execsw[i]->ex_imgact == NULL ||
  509                     execsw[i]->ex_imgact == img_first) {
  510                         continue;
  511                 }
  512                 error = (*execsw[i]->ex_imgact)(imgp);
  513         }
  514 
  515         if (error) {
  516                 if (error == -1) {
  517                         if (textset == 0)
  518                                 VOP_UNSET_TEXT(imgp->vp);
  519                         error = ENOEXEC;
  520                 }
  521                 goto exec_fail_dealloc;
  522         }
  523 
  524         /*
  525          * Special interpreter operation, cleanup and loop up to try to
  526          * activate the interpreter.
  527          */
  528         if (imgp->interpreted) {
  529                 exec_unmap_first_page(imgp);
  530                 /*
  531                  * VV_TEXT needs to be unset for scripts.  There is a short
  532                  * period before we determine that something is a script where
  533                  * VV_TEXT will be set. The vnode lock is held over this
  534                  * entire period so nothing should illegitimately be blocked.
  535                  */
  536                 VOP_UNSET_TEXT(imgp->vp);
  537                 /* free name buffer and old vnode */
  538                 if (args->fname != NULL)
  539                         NDFREE(&nd, NDF_ONLY_PNBUF);
  540 #ifdef MAC
  541                 mac_execve_interpreter_enter(binvp, &interpvplabel);
  542 #endif
  543                 if (imgp->opened) {
  544                         VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
  545                         imgp->opened = 0;
  546                 }
  547                 vput(binvp);
  548                 vm_object_deallocate(imgp->object);
  549                 imgp->object = NULL;
  550                 VFS_UNLOCK_GIANT(vfslocked);
  551                 vfslocked = 0;
  552                 /* set new name to that of the interpreter */
  553                 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
  554                     UIO_SYSSPACE, imgp->interpreter_name, td);
  555                 args->fname = imgp->interpreter_name;
  556                 goto interpret;
  557         }
  558 
  559         /*
  560          * NB: We unlock the vnode here because it is believed that none
  561          * of the sv_copyout_strings/sv_fixup operations require the vnode.
  562          */
  563         VOP_UNLOCK(imgp->vp, 0);
  564 
  565         /*
  566          * Do the best to calculate the full path to the image file.
  567          */
  568         if (imgp->auxargs != NULL &&
  569             ((args->fname != NULL && args->fname[0] == '/') ||
  570              vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
  571                 imgp->execpath = args->fname;
  572 
  573         /*
  574          * Copy out strings (args and env) and initialize stack base
  575          */
  576         if (p->p_sysent->sv_copyout_strings)
  577                 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
  578         else
  579                 stack_base = exec_copyout_strings(imgp);
  580 
  581         /*
  582          * If custom stack fixup routine present for this process
  583          * let it do the stack setup.
  584          * Else stuff argument count as first item on stack
  585          */
  586         if (p->p_sysent->sv_fixup != NULL)
  587                 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
  588         else
  589                 suword(--stack_base, imgp->args->argc);
  590 
  591         /*
  592          * For security and other reasons, the file descriptor table cannot
  593          * be shared after an exec.
  594          */
  595         fdunshare(p, td);
  596 
  597         /*
  598          * Malloc things before we need locks.
  599          */
  600         newcred = crget();
  601         euip = uifind(attr.va_uid);
  602         i = imgp->args->begin_envv - imgp->args->begin_argv;
  603         /* Cache arguments if they fit inside our allowance */
  604         if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
  605                 newargs = pargs_alloc(i);
  606                 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
  607         }
  608 
  609         /* close files on exec */
  610         fdcloseexec(td);
  611         vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
  612 
  613         /* Get a reference to the vnode prior to locking the proc */
  614         VREF(binvp);
  615 
  616         /*
  617          * For security and other reasons, signal handlers cannot
  618          * be shared after an exec. The new process gets a copy of the old
  619          * handlers. In execsigs(), the new process will have its signals
  620          * reset.
  621          */
  622         PROC_LOCK(p);
  623         oldcred = crcopysafe(p, newcred);
  624         if (sigacts_shared(p->p_sigacts)) {
  625                 oldsigacts = p->p_sigacts;
  626                 PROC_UNLOCK(p);
  627                 newsigacts = sigacts_alloc();
  628                 sigacts_copy(newsigacts, oldsigacts);
  629                 PROC_LOCK(p);
  630                 p->p_sigacts = newsigacts;
  631         } else
  632                 oldsigacts = NULL;
  633 
  634         /* Stop profiling */
  635         stopprofclock(p);
  636 
  637         /* reset caught signals */
  638         execsigs(p);
  639 
  640         /* name this process - nameiexec(p, ndp) */
  641         bzero(p->p_comm, sizeof(p->p_comm));
  642         if (args->fname)
  643                 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
  644                     min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
  645         else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
  646                 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
  647         bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
  648 #ifdef KTR
  649         sched_clear_tdname(td);
  650 #endif
  651 
  652         /*
  653          * mark as execed, wakeup the process that vforked (if any) and tell
  654          * it that it now has its own resources back
  655          */
  656         p->p_flag |= P_EXEC;
  657         if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
  658                 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
  659                 cv_broadcast(&p->p_pwait);
  660         }
  661 
  662         /*
  663          * Implement image setuid/setgid.
  664          *
  665          * Don't honor setuid/setgid if the filesystem prohibits it or if
  666          * the process is being traced.
  667          *
  668          * We disable setuid/setgid/etc in compatibility mode on the basis
  669          * that most setugid applications are not written with that
  670          * environment in mind, and will therefore almost certainly operate
  671          * incorrectly. In principle there's no reason that setugid
  672          * applications might not be useful in capability mode, so we may want
  673          * to reconsider this conservative design choice in the future.
  674          *
  675          * XXXMAC: For the time being, use NOSUID to also prohibit
  676          * transitions on the file system.
  677          */
  678         credential_changing = 0;
  679         credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
  680             attr.va_uid;
  681         credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
  682             attr.va_gid;
  683 #ifdef MAC
  684         will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
  685             interpvplabel, imgp);
  686         credential_changing |= will_transition;
  687 #endif
  688 
  689         if (credential_changing &&
  690 #ifdef CAPABILITY_MODE
  691             ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
  692 #endif
  693             (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
  694             (p->p_flag & P_TRACED) == 0) {
  695                 /*
  696                  * Turn off syscall tracing for set-id programs, except for
  697                  * root.  Record any set-id flags first to make sure that
  698                  * we do not regain any tracing during a possible block.
  699                  */
  700                 setsugid(p);
  701 
  702 #ifdef KTRACE
  703                 if (p->p_tracecred != NULL &&
  704                     priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
  705                         ktrprocexec(p, &tracecred, &tracevp);
  706 #endif
  707                 /*
  708                  * Close any file descriptors 0..2 that reference procfs,
  709                  * then make sure file descriptors 0..2 are in use.
  710                  *
  711                  * setugidsafety() may call closef() and then pfind()
  712                  * which may grab the process lock.
  713                  * fdcheckstd() may call falloc() which may block to
  714                  * allocate memory, so temporarily drop the process lock.
  715                  */
  716                 PROC_UNLOCK(p);
  717                 VOP_UNLOCK(imgp->vp, 0);
  718                 setugidsafety(td);
  719                 error = fdcheckstd(td);
  720                 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
  721                 if (error != 0)
  722                         goto done1;
  723                 PROC_LOCK(p);
  724                 /*
  725                  * Set the new credentials.
  726                  */
  727                 if (attr.va_mode & S_ISUID)
  728                         change_euid(newcred, euip);
  729                 if (attr.va_mode & S_ISGID)
  730                         change_egid(newcred, attr.va_gid);
  731 #ifdef MAC
  732                 if (will_transition) {
  733                         mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
  734                             interpvplabel, imgp);
  735                 }
  736 #endif
  737                 /*
  738                  * Implement correct POSIX saved-id behavior.
  739                  *
  740                  * XXXMAC: Note that the current logic will save the
  741                  * uid and gid if a MAC domain transition occurs, even
  742                  * though maybe it shouldn't.
  743                  */
  744                 change_svuid(newcred, newcred->cr_uid);
  745                 change_svgid(newcred, newcred->cr_gid);
  746                 p->p_ucred = newcred;
  747                 newcred = NULL;
  748         } else {
  749                 if (oldcred->cr_uid == oldcred->cr_ruid &&
  750                     oldcred->cr_gid == oldcred->cr_rgid)
  751                         p->p_flag &= ~P_SUGID;
  752                 /*
  753                  * Implement correct POSIX saved-id behavior.
  754                  *
  755                  * XXX: It's not clear that the existing behavior is
  756                  * POSIX-compliant.  A number of sources indicate that the
  757                  * saved uid/gid should only be updated if the new ruid is
  758                  * not equal to the old ruid, or the new euid is not equal
  759                  * to the old euid and the new euid is not equal to the old
  760                  * ruid.  The FreeBSD code always updates the saved uid/gid.
  761                  * Also, this code uses the new (replaced) euid and egid as
  762                  * the source, which may or may not be the right ones to use.
  763                  */
  764                 if (oldcred->cr_svuid != oldcred->cr_uid ||
  765                     oldcred->cr_svgid != oldcred->cr_gid) {
  766                         change_svuid(newcred, newcred->cr_uid);
  767                         change_svgid(newcred, newcred->cr_gid);
  768                         p->p_ucred = newcred;
  769                         newcred = NULL;
  770                 }
  771         }
  772 
  773         /*
  774          * Store the vp for use in procfs.  This vnode was referenced prior
  775          * to locking the proc lock.
  776          */
  777         textvp = p->p_textvp;
  778         p->p_textvp = binvp;
  779 
  780 #ifdef KDTRACE_HOOKS
  781         /*
  782          * Tell the DTrace fasttrap provider about the exec if it
  783          * has declared an interest.
  784          */
  785         if (dtrace_fasttrap_exec)
  786                 dtrace_fasttrap_exec(p);
  787 #endif
  788 
  789         /*
  790          * Notify others that we exec'd, and clear the P_INEXEC flag
  791          * as we're now a bona fide freshly-execed process.
  792          */
  793         KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
  794         p->p_flag &= ~P_INEXEC;
  795 
  796         /* clear "fork but no exec" flag, as we _are_ execing */
  797         p->p_acflag &= ~AFORK;
  798 
  799         /*
  800          * Free any previous argument cache and replace it with
  801          * the new argument cache, if any.
  802          */
  803         oldargs = p->p_args;
  804         p->p_args = newargs;
  805         newargs = NULL;
  806 
  807 #ifdef  HWPMC_HOOKS
  808         /*
  809          * Check if system-wide sampling is in effect or if the
  810          * current process is using PMCs.  If so, do exec() time
  811          * processing.  This processing needs to happen AFTER the
  812          * P_INEXEC flag is cleared.
  813          *
  814          * The proc lock needs to be released before taking the PMC
  815          * SX.
  816          */
  817         if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
  818                 PROC_UNLOCK(p);
  819                 VOP_UNLOCK(imgp->vp, 0);
  820                 pe.pm_credentialschanged = credential_changing;
  821                 pe.pm_entryaddr = imgp->entry_addr;
  822 
  823                 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
  824                 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
  825         } else
  826                 PROC_UNLOCK(p);
  827 #else  /* !HWPMC_HOOKS */
  828         PROC_UNLOCK(p);
  829 #endif
  830 
  831         /* Set values passed into the program in registers. */
  832         if (p->p_sysent->sv_setregs)
  833                 (*p->p_sysent->sv_setregs)(td, imgp, 
  834                     (u_long)(uintptr_t)stack_base);
  835         else
  836                 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
  837 
  838         vfs_mark_atime(imgp->vp, td->td_ucred);
  839 
  840         SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
  841 
  842 done1:
  843         /*
  844          * Free any resources malloc'd earlier that we didn't use.
  845          */
  846         uifree(euip);
  847         if (newcred == NULL)
  848                 crfree(oldcred);
  849         else
  850                 crfree(newcred);
  851         VOP_UNLOCK(imgp->vp, 0);
  852 
  853         /*
  854          * Handle deferred decrement of ref counts.
  855          */
  856         if (textvp != NULL) {
  857                 int tvfslocked;
  858 
  859                 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
  860                 vrele(textvp);
  861                 VFS_UNLOCK_GIANT(tvfslocked);
  862         }
  863         if (binvp && error != 0)
  864                 vrele(binvp);
  865 #ifdef KTRACE
  866         if (tracevp != NULL) {
  867                 int tvfslocked;
  868 
  869                 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
  870                 vrele(tracevp);
  871                 VFS_UNLOCK_GIANT(tvfslocked);
  872         }
  873         if (tracecred != NULL)
  874                 crfree(tracecred);
  875 #endif
  876         vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
  877         pargs_drop(oldargs);
  878         pargs_drop(newargs);
  879         if (oldsigacts != NULL)
  880                 sigacts_free(oldsigacts);
  881 
  882 exec_fail_dealloc:
  883 
  884         /*
  885          * free various allocated resources
  886          */
  887         if (imgp->firstpage != NULL)
  888                 exec_unmap_first_page(imgp);
  889 
  890         if (imgp->vp != NULL) {
  891                 if (args->fname)
  892                         NDFREE(&nd, NDF_ONLY_PNBUF);
  893                 if (imgp->opened)
  894                         VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
  895                 vput(imgp->vp);
  896         }
  897 
  898         if (imgp->object != NULL)
  899                 vm_object_deallocate(imgp->object);
  900 
  901         free(imgp->freepath, M_TEMP);
  902 
  903         if (error == 0) {
  904                 PROC_LOCK(p);
  905                 td->td_dbgflags |= TDB_EXEC;
  906                 PROC_UNLOCK(p);
  907 
  908                 /*
  909                  * Stop the process here if its stop event mask has
  910                  * the S_EXEC bit set.
  911                  */
  912                 STOPEVENT(p, S_EXEC, 0);
  913                 goto done2;
  914         }
  915 
  916 exec_fail:
  917         /* we're done here, clear P_INEXEC */
  918         PROC_LOCK(p);
  919         p->p_flag &= ~P_INEXEC;
  920         PROC_UNLOCK(p);
  921 
  922         SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
  923 
  924 done2:
  925 #ifdef MAC
  926         mac_execve_exit(imgp);
  927         mac_execve_interpreter_exit(interpvplabel);
  928 #endif
  929         VFS_UNLOCK_GIANT(vfslocked);
  930         exec_free_args(args);
  931 
  932         if (error && imgp->vmspace_destroyed) {
  933                 /* sorry, no more process anymore. exit gracefully */
  934                 exit1(td, W_EXITCODE(0, SIGABRT));
  935                 /* NOT REACHED */
  936         }
  937 
  938 #ifdef KTRACE
  939         if (error == 0)
  940                 ktrprocctor(p);
  941 #endif
  942 
  943         return (error);
  944 }
  945 
  946 int
  947 exec_map_first_page(imgp)
  948         struct image_params *imgp;
  949 {
  950         int rv, i;
  951         int initial_pagein;
  952         vm_page_t ma[VM_INITIAL_PAGEIN];
  953         vm_object_t object;
  954 
  955         if (imgp->firstpage != NULL)
  956                 exec_unmap_first_page(imgp);
  957 
  958         object = imgp->vp->v_object;
  959         if (object == NULL)
  960                 return (EACCES);
  961         VM_OBJECT_LOCK(object);
  962 #if VM_NRESERVLEVEL > 0
  963         if ((object->flags & OBJ_COLORED) == 0) {
  964                 object->flags |= OBJ_COLORED;
  965                 object->pg_color = 0;
  966         }
  967 #endif
  968         ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
  969         if (ma[0]->valid != VM_PAGE_BITS_ALL) {
  970                 initial_pagein = VM_INITIAL_PAGEIN;
  971                 if (initial_pagein > object->size)
  972                         initial_pagein = object->size;
  973                 for (i = 1; i < initial_pagein; i++) {
  974                         if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
  975                                 if (ma[i]->valid)
  976                                         break;
  977                                 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
  978                                         break;
  979                                 vm_page_busy(ma[i]);
  980                         } else {
  981                                 ma[i] = vm_page_alloc(object, i,
  982                                     VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
  983                                 if (ma[i] == NULL)
  984                                         break;
  985                         }
  986                 }
  987                 initial_pagein = i;
  988                 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
  989                 ma[0] = vm_page_lookup(object, 0);
  990                 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
  991                         if (ma[0] != NULL) {
  992                                 vm_page_lock(ma[0]);
  993                                 vm_page_free(ma[0]);
  994                                 vm_page_unlock(ma[0]);
  995                         }
  996                         VM_OBJECT_UNLOCK(object);
  997                         return (EIO);
  998                 }
  999         }
 1000         vm_page_lock(ma[0]);
 1001         vm_page_hold(ma[0]);
 1002         vm_page_unlock(ma[0]);
 1003         vm_page_wakeup(ma[0]);
 1004         VM_OBJECT_UNLOCK(object);
 1005 
 1006         imgp->firstpage = sf_buf_alloc(ma[0], 0);
 1007         imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
 1008 
 1009         return (0);
 1010 }
 1011 
 1012 void
 1013 exec_unmap_first_page(imgp)
 1014         struct image_params *imgp;
 1015 {
 1016         vm_page_t m;
 1017 
 1018         if (imgp->firstpage != NULL) {
 1019                 m = sf_buf_page(imgp->firstpage);
 1020                 sf_buf_free(imgp->firstpage);
 1021                 imgp->firstpage = NULL;
 1022                 vm_page_lock(m);
 1023                 vm_page_unhold(m);
 1024                 vm_page_unlock(m);
 1025         }
 1026 }
 1027 
 1028 /*
 1029  * Destroy old address space, and allocate a new stack
 1030  *      The new stack is only SGROWSIZ large because it is grown
 1031  *      automatically in trap.c.
 1032  */
 1033 int
 1034 exec_new_vmspace(imgp, sv)
 1035         struct image_params *imgp;
 1036         struct sysentvec *sv;
 1037 {
 1038         int error;
 1039         struct proc *p = imgp->proc;
 1040         struct vmspace *vmspace = p->p_vmspace;
 1041         vm_object_t obj;
 1042         vm_offset_t sv_minuser, stack_addr;
 1043         vm_map_t map;
 1044         u_long ssiz;
 1045 
 1046         imgp->vmspace_destroyed = 1;
 1047         imgp->sysent = sv;
 1048 
 1049         /* May be called with Giant held */
 1050         EVENTHANDLER_INVOKE(process_exec, p, imgp);
 1051 
 1052         /*
 1053          * Blow away entire process VM, if address space not shared,
 1054          * otherwise, create a new VM space so that other threads are
 1055          * not disrupted
 1056          */
 1057         map = &vmspace->vm_map;
 1058         if (map_at_zero)
 1059                 sv_minuser = sv->sv_minuser;
 1060         else
 1061                 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
 1062         if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
 1063             vm_map_max(map) == sv->sv_maxuser) {
 1064                 shmexit(vmspace);
 1065                 pmap_remove_pages(vmspace_pmap(vmspace));
 1066                 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
 1067         } else {
 1068                 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
 1069                 if (error)
 1070                         return (error);
 1071                 vmspace = p->p_vmspace;
 1072                 map = &vmspace->vm_map;
 1073         }
 1074 
 1075         /* Map a shared page */
 1076         obj = sv->sv_shared_page_obj;
 1077         if (obj != NULL) {
 1078                 vm_object_reference(obj);
 1079                 error = vm_map_fixed(map, obj, 0,
 1080                     sv->sv_shared_page_base, sv->sv_shared_page_len,
 1081                     VM_PROT_READ | VM_PROT_EXECUTE,
 1082                     VM_PROT_READ | VM_PROT_EXECUTE,
 1083                     MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
 1084                 if (error) {
 1085                         vm_object_deallocate(obj);
 1086                         return (error);
 1087                 }
 1088         }
 1089 
 1090         /* Allocate a new stack */
 1091         if (sv->sv_maxssiz != NULL)
 1092                 ssiz = *sv->sv_maxssiz;
 1093         else
 1094                 ssiz = maxssiz;
 1095         stack_addr = sv->sv_usrstack - ssiz;
 1096         error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
 1097             obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
 1098                 sv->sv_stackprot,
 1099             VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
 1100         if (error)
 1101                 return (error);
 1102 
 1103 #ifdef __ia64__
 1104         /* Allocate a new register stack */
 1105         stack_addr = IA64_BACKINGSTORE;
 1106         error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
 1107             sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
 1108         if (error)
 1109                 return (error);
 1110 #endif
 1111 
 1112         /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
 1113          * VM_STACK case, but they are still used to monitor the size of the
 1114          * process stack so we can check the stack rlimit.
 1115          */
 1116         vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
 1117         vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
 1118 
 1119         return (0);
 1120 }
 1121 
 1122 /*
 1123  * Copy out argument and environment strings from the old process address
 1124  * space into the temporary string buffer.
 1125  */
 1126 int
 1127 exec_copyin_args(struct image_args *args, char *fname,
 1128     enum uio_seg segflg, char **argv, char **envv)
 1129 {
 1130         char *argp, *envp;
 1131         int error;
 1132         size_t length;
 1133 
 1134         bzero(args, sizeof(*args));
 1135         if (argv == NULL)
 1136                 return (EFAULT);
 1137 
 1138         /*
 1139          * Allocate demand-paged memory for the file name, argument, and
 1140          * environment strings.
 1141          */
 1142         error = exec_alloc_args(args);
 1143         if (error != 0)
 1144                 return (error);
 1145 
 1146         /*
 1147          * Copy the file name.
 1148          */
 1149         if (fname != NULL) {
 1150                 args->fname = args->buf;
 1151                 error = (segflg == UIO_SYSSPACE) ?
 1152                     copystr(fname, args->fname, PATH_MAX, &length) :
 1153                     copyinstr(fname, args->fname, PATH_MAX, &length);
 1154                 if (error != 0)
 1155                         goto err_exit;
 1156         } else
 1157                 length = 0;
 1158 
 1159         args->begin_argv = args->buf + length;
 1160         args->endp = args->begin_argv;
 1161         args->stringspace = ARG_MAX;
 1162 
 1163         /*
 1164          * extract arguments first
 1165          */
 1166         while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
 1167                 if (argp == (caddr_t) -1) {
 1168                         error = EFAULT;
 1169                         goto err_exit;
 1170                 }
 1171                 if ((error = copyinstr(argp, args->endp,
 1172                     args->stringspace, &length))) {
 1173                         if (error == ENAMETOOLONG) 
 1174                                 error = E2BIG;
 1175                         goto err_exit;
 1176                 }
 1177                 args->stringspace -= length;
 1178                 args->endp += length;
 1179                 args->argc++;
 1180         }
 1181 
 1182         args->begin_envv = args->endp;
 1183 
 1184         /*
 1185          * extract environment strings
 1186          */
 1187         if (envv) {
 1188                 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
 1189                         if (envp == (caddr_t)-1) {
 1190                                 error = EFAULT;
 1191                                 goto err_exit;
 1192                         }
 1193                         if ((error = copyinstr(envp, args->endp,
 1194                             args->stringspace, &length))) {
 1195                                 if (error == ENAMETOOLONG)
 1196                                         error = E2BIG;
 1197                                 goto err_exit;
 1198                         }
 1199                         args->stringspace -= length;
 1200                         args->endp += length;
 1201                         args->envc++;
 1202                 }
 1203         }
 1204 
 1205         return (0);
 1206 
 1207 err_exit:
 1208         exec_free_args(args);
 1209         return (error);
 1210 }
 1211 
 1212 /*
 1213  * Allocate temporary demand-paged, zero-filled memory for the file name,
 1214  * argument, and environment strings.  Returns zero if the allocation succeeds
 1215  * and ENOMEM otherwise.
 1216  */
 1217 int
 1218 exec_alloc_args(struct image_args *args)
 1219 {
 1220 
 1221         args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
 1222         return (args->buf != NULL ? 0 : ENOMEM);
 1223 }
 1224 
 1225 void
 1226 exec_free_args(struct image_args *args)
 1227 {
 1228 
 1229         if (args->buf != NULL) {
 1230                 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
 1231                     PATH_MAX + ARG_MAX);
 1232                 args->buf = NULL;
 1233         }
 1234         if (args->fname_buf != NULL) {
 1235                 free(args->fname_buf, M_TEMP);
 1236                 args->fname_buf = NULL;
 1237         }
 1238 }
 1239 
 1240 /*
 1241  * Copy strings out to the new process address space, constructing new arg
 1242  * and env vector tables. Return a pointer to the base so that it can be used
 1243  * as the initial stack pointer.
 1244  */
 1245 register_t *
 1246 exec_copyout_strings(imgp)
 1247         struct image_params *imgp;
 1248 {
 1249         int argc, envc;
 1250         char **vectp;
 1251         char *stringp, *destp;
 1252         register_t *stack_base;
 1253         struct ps_strings *arginfo;
 1254         struct proc *p;
 1255         size_t execpath_len;
 1256         int szsigcode, szps;
 1257         char canary[sizeof(long) * 8];
 1258 
 1259         szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
 1260         /*
 1261          * Calculate string base and vector table pointers.
 1262          * Also deal with signal trampoline code for this exec type.
 1263          */
 1264         if (imgp->execpath != NULL && imgp->auxargs != NULL)
 1265                 execpath_len = strlen(imgp->execpath) + 1;
 1266         else
 1267                 execpath_len = 0;
 1268         p = imgp->proc;
 1269         szsigcode = 0;
 1270         arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
 1271         if (p->p_sysent->sv_sigcode_base == 0) {
 1272                 if (p->p_sysent->sv_szsigcode != NULL)
 1273                         szsigcode = *(p->p_sysent->sv_szsigcode);
 1274         }
 1275         destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
 1276             roundup(execpath_len, sizeof(char *)) -
 1277             roundup(sizeof(canary), sizeof(char *)) -
 1278             roundup(szps, sizeof(char *)) -
 1279             roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
 1280 
 1281         /*
 1282          * install sigcode
 1283          */
 1284         if (szsigcode != 0)
 1285                 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
 1286                     szsigcode), szsigcode);
 1287 
 1288         /*
 1289          * Copy the image path for the rtld.
 1290          */
 1291         if (execpath_len != 0) {
 1292                 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
 1293                 copyout(imgp->execpath, (void *)imgp->execpathp,
 1294                     execpath_len);
 1295         }
 1296 
 1297         /*
 1298          * Prepare the canary for SSP.
 1299          */
 1300         arc4rand(canary, sizeof(canary), 0);
 1301         imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
 1302             sizeof(canary);
 1303         copyout(canary, (void *)imgp->canary, sizeof(canary));
 1304         imgp->canarylen = sizeof(canary);
 1305 
 1306         /*
 1307          * Prepare the pagesizes array.
 1308          */
 1309         imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
 1310             roundup(sizeof(canary), sizeof(char *)) - szps;
 1311         copyout(pagesizes, (void *)imgp->pagesizes, szps);
 1312         imgp->pagesizeslen = szps;
 1313 
 1314         /*
 1315          * If we have a valid auxargs ptr, prepare some room
 1316          * on the stack.
 1317          */
 1318         if (imgp->auxargs) {
 1319                 /*
 1320                  * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
 1321                  * lower compatibility.
 1322                  */
 1323                 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
 1324                     (AT_COUNT * 2);
 1325                 /*
 1326                  * The '+ 2' is for the null pointers at the end of each of
 1327                  * the arg and env vector sets,and imgp->auxarg_size is room
 1328                  * for argument of Runtime loader.
 1329                  */
 1330                 vectp = (char **)(destp - (imgp->args->argc +
 1331                     imgp->args->envc + 2 + imgp->auxarg_size)
 1332                     * sizeof(char *));
 1333         } else {
 1334                 /*
 1335                  * The '+ 2' is for the null pointers at the end of each of
 1336                  * the arg and env vector sets
 1337                  */
 1338                 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
 1339                     sizeof(char *));
 1340         }
 1341 
 1342         /*
 1343          * vectp also becomes our initial stack base
 1344          */
 1345         stack_base = (register_t *)vectp;
 1346 
 1347         stringp = imgp->args->begin_argv;
 1348         argc = imgp->args->argc;
 1349         envc = imgp->args->envc;
 1350 
 1351         /*
 1352          * Copy out strings - arguments and environment.
 1353          */
 1354         copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
 1355 
 1356         /*
 1357          * Fill in "ps_strings" struct for ps, w, etc.
 1358          */
 1359         suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
 1360         suword32(&arginfo->ps_nargvstr, argc);
 1361 
 1362         /*
 1363          * Fill in argument portion of vector table.
 1364          */
 1365         for (; argc > 0; --argc) {
 1366                 suword(vectp++, (long)(intptr_t)destp);
 1367                 while (*stringp++ != 0)
 1368                         destp++;
 1369                 destp++;
 1370         }
 1371 
 1372         /* a null vector table pointer separates the argp's from the envp's */
 1373         suword(vectp++, 0);
 1374 
 1375         suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
 1376         suword32(&arginfo->ps_nenvstr, envc);
 1377 
 1378         /*
 1379          * Fill in environment portion of vector table.
 1380          */
 1381         for (; envc > 0; --envc) {
 1382                 suword(vectp++, (long)(intptr_t)destp);
 1383                 while (*stringp++ != 0)
 1384                         destp++;
 1385                 destp++;
 1386         }
 1387 
 1388         /* end of vector table is a null pointer */
 1389         suword(vectp, 0);
 1390 
 1391         return (stack_base);
 1392 }
 1393 
 1394 /*
 1395  * Check permissions of file to execute.
 1396  *      Called with imgp->vp locked.
 1397  *      Return 0 for success or error code on failure.
 1398  */
 1399 int
 1400 exec_check_permissions(imgp)
 1401         struct image_params *imgp;
 1402 {
 1403         struct vnode *vp = imgp->vp;
 1404         struct vattr *attr = imgp->attr;
 1405         struct thread *td;
 1406         int error, writecount;
 1407 
 1408         td = curthread;
 1409 
 1410         /* Get file attributes */
 1411         error = VOP_GETATTR(vp, attr, td->td_ucred);
 1412         if (error)
 1413                 return (error);
 1414 
 1415 #ifdef MAC
 1416         error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
 1417         if (error)
 1418                 return (error);
 1419 #endif
 1420 
 1421         /*
 1422          * 1) Check if file execution is disabled for the filesystem that
 1423          *    this file resides on.
 1424          * 2) Ensure that at least one execute bit is on. Otherwise, a
 1425          *    privileged user will always succeed, and we don't want this
 1426          *    to happen unless the file really is executable.
 1427          * 3) Ensure that the file is a regular file.
 1428          */
 1429         if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
 1430             (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
 1431             (attr->va_type != VREG))
 1432                 return (EACCES);
 1433 
 1434         /*
 1435          * Zero length files can't be exec'd
 1436          */
 1437         if (attr->va_size == 0)
 1438                 return (ENOEXEC);
 1439 
 1440         /*
 1441          *  Check for execute permission to file based on current credentials.
 1442          */
 1443         error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
 1444         if (error)
 1445                 return (error);
 1446 
 1447         /*
 1448          * Check number of open-for-writes on the file and deny execution
 1449          * if there are any.
 1450          */
 1451         error = VOP_GET_WRITECOUNT(vp, &writecount);
 1452         if (error != 0)
 1453                 return (error);
 1454         if (writecount != 0)
 1455                 return (ETXTBSY);
 1456 
 1457         /*
 1458          * Call filesystem specific open routine (which does nothing in the
 1459          * general case).
 1460          */
 1461         error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
 1462         if (error == 0)
 1463                 imgp->opened = 1;
 1464         return (error);
 1465 }
 1466 
 1467 /*
 1468  * Exec handler registration
 1469  */
 1470 int
 1471 exec_register(execsw_arg)
 1472         const struct execsw *execsw_arg;
 1473 {
 1474         const struct execsw **es, **xs, **newexecsw;
 1475         int count = 2;  /* New slot and trailing NULL */
 1476 
 1477         if (execsw)
 1478                 for (es = execsw; *es; es++)
 1479                         count++;
 1480         newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
 1481         if (newexecsw == NULL)
 1482                 return (ENOMEM);
 1483         xs = newexecsw;
 1484         if (execsw)
 1485                 for (es = execsw; *es; es++)
 1486                         *xs++ = *es;
 1487         *xs++ = execsw_arg;
 1488         *xs = NULL;
 1489         if (execsw)
 1490                 free(execsw, M_TEMP);
 1491         execsw = newexecsw;
 1492         return (0);
 1493 }
 1494 
 1495 int
 1496 exec_unregister(execsw_arg)
 1497         const struct execsw *execsw_arg;
 1498 {
 1499         const struct execsw **es, **xs, **newexecsw;
 1500         int count = 1;
 1501 
 1502         if (execsw == NULL)
 1503                 panic("unregister with no handlers left?\n");
 1504 
 1505         for (es = execsw; *es; es++) {
 1506                 if (*es == execsw_arg)
 1507                         break;
 1508         }
 1509         if (*es == NULL)
 1510                 return (ENOENT);
 1511         for (es = execsw; *es; es++)
 1512                 if (*es != execsw_arg)
 1513                         count++;
 1514         newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
 1515         if (newexecsw == NULL)
 1516                 return (ENOMEM);
 1517         xs = newexecsw;
 1518         for (es = execsw; *es; es++)
 1519                 if (*es != execsw_arg)
 1520                         *xs++ = *es;
 1521         *xs = NULL;
 1522         if (execsw)
 1523                 free(execsw, M_TEMP);
 1524         execsw = newexecsw;
 1525         return (0);
 1526 }

Cache object: 1fa9007df7a9cf250c91d45eebf160e8


[ 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.