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  -  FREEBSD11  -  FREEBSD10  -  FREEBSD9  -  FREEBSD92  -  FREEBSD91  -  FREEBSD90  -  FREEBSD8  -  FREEBSD82  -  FREEBSD81  -  FREEBSD80  -  FREEBSD7  -  FREEBSD74  -  FREEBSD73  -  FREEBSD72  -  FREEBSD71  -  FREEBSD70  -  FREEBSD6  -  FREEBSD64  -  FREEBSD63  -  FREEBSD62  -  FREEBSD61  -  FREEBSD60  -  FREEBSD5  -  FREEBSD55  -  FREEBSD54  -  FREEBSD53  -  FREEBSD52  -  FREEBSD51  -  FREEBSD50  -  FREEBSD4  -  FREEBSD3  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 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: stable/10/sys/kern/kern_exec.c 321359 2017-07-22 04:57:51Z alc $");
   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/capsicum.h>
   38 #include <sys/systm.h>
   39 #include <sys/capsicum.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/ptrace.h>
   59 #include <sys/namei.h>
   60 #include <sys/resourcevar.h>
   61 #include <sys/rwlock.h>
   62 #include <sys/sched.h>
   63 #include <sys/sdt.h>
   64 #include <sys/sf_buf.h>
   65 #include <sys/syscallsubr.h>
   66 #include <sys/sysent.h>
   67 #include <sys/shm.h>
   68 #include <sys/sysctl.h>
   69 #include <sys/vnode.h>
   70 #include <sys/stat.h>
   71 #ifdef KTRACE
   72 #include <sys/ktrace.h>
   73 #endif
   74 
   75 #include <vm/vm.h>
   76 #include <vm/vm_param.h>
   77 #include <vm/pmap.h>
   78 #include <vm/vm_page.h>
   79 #include <vm/vm_map.h>
   80 #include <vm/vm_kern.h>
   81 #include <vm/vm_extern.h>
   82 #include <vm/vm_object.h>
   83 #include <vm/vm_pager.h>
   84 
   85 #ifdef  HWPMC_HOOKS
   86 #include <sys/pmckern.h>
   87 #endif
   88 
   89 #include <machine/reg.h>
   90 
   91 #include <security/audit/audit.h>
   92 #include <security/mac/mac_framework.h>
   93 
   94 #ifdef KDTRACE_HOOKS
   95 #include <sys/dtrace_bsd.h>
   96 dtrace_execexit_func_t  dtrace_fasttrap_exec;
   97 #endif
   98 
   99 SDT_PROVIDER_DECLARE(proc);
  100 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
  101 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
  102 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
  103 
  104 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
  105 
  106 int coredump_pack_fileinfo = 1;
  107 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
  108     &coredump_pack_fileinfo, 0,
  109     "Enable file path packing in 'procstat -f' coredump notes");
  110 
  111 int coredump_pack_vmmapinfo = 1;
  112 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
  113     &coredump_pack_vmmapinfo, 0,
  114     "Enable file path packing in 'procstat -v' coredump notes");
  115 
  116 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
  117 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
  118 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
  119 static int do_execve(struct thread *td, struct image_args *args,
  120     struct mac *mac_p);
  121 
  122 /* XXX This should be vm_size_t. */
  123 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
  124     NULL, 0, sysctl_kern_ps_strings, "LU", "");
  125 
  126 /* XXX This should be vm_size_t. */
  127 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
  128     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
  129 
  130 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
  131     NULL, 0, sysctl_kern_stackprot, "I", "");
  132 
  133 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
  134 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 
  135     &ps_arg_cache_limit, 0, "");
  136 
  137 static int disallow_high_osrel;
  138 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
  139     &disallow_high_osrel, 0,
  140     "Disallow execution of binaries built for higher version of the world");
  141 
  142 static int map_at_zero = 0;
  143 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
  144 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
  145     "Permit processes to map an object at virtual address 0.");
  146 
  147 static int
  148 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
  149 {
  150         struct proc *p;
  151         int error;
  152 
  153         p = curproc;
  154 #ifdef SCTL_MASK32
  155         if (req->flags & SCTL_MASK32) {
  156                 unsigned int val;
  157                 val = (unsigned int)p->p_sysent->sv_psstrings;
  158                 error = SYSCTL_OUT(req, &val, sizeof(val));
  159         } else
  160 #endif
  161                 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
  162                    sizeof(p->p_sysent->sv_psstrings));
  163         return error;
  164 }
  165 
  166 static int
  167 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
  168 {
  169         struct proc *p;
  170         int error;
  171 
  172         p = curproc;
  173 #ifdef SCTL_MASK32
  174         if (req->flags & SCTL_MASK32) {
  175                 unsigned int val;
  176                 val = (unsigned int)p->p_sysent->sv_usrstack;
  177                 error = SYSCTL_OUT(req, &val, sizeof(val));
  178         } else
  179 #endif
  180                 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
  181                     sizeof(p->p_sysent->sv_usrstack));
  182         return error;
  183 }
  184 
  185 static int
  186 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
  187 {
  188         struct proc *p;
  189 
  190         p = curproc;
  191         return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
  192             sizeof(p->p_sysent->sv_stackprot)));
  193 }
  194 
  195 /*
  196  * Each of the items is a pointer to a `const struct execsw', hence the
  197  * double pointer here.
  198  */
  199 static const struct execsw **execsw;
  200 
  201 #ifndef _SYS_SYSPROTO_H_
  202 struct execve_args {
  203         char    *fname; 
  204         char    **argv;
  205         char    **envv; 
  206 };
  207 #endif
  208 
  209 int
  210 sys_execve(struct thread *td, struct execve_args *uap)
  211 {
  212         struct image_args args;
  213         struct vmspace *oldvmspace;
  214         int error;
  215 
  216         error = pre_execve(td, &oldvmspace);
  217         if (error != 0)
  218                 return (error);
  219         error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
  220             uap->argv, uap->envv);
  221         if (error == 0)
  222                 error = kern_execve(td, &args, NULL);
  223         post_execve(td, error, oldvmspace);
  224         return (error);
  225 }
  226 
  227 #ifndef _SYS_SYSPROTO_H_
  228 struct fexecve_args {
  229         int     fd;
  230         char    **argv;
  231         char    **envv;
  232 }
  233 #endif
  234 int
  235 sys_fexecve(struct thread *td, struct fexecve_args *uap)
  236 {
  237         struct image_args args;
  238         struct vmspace *oldvmspace;
  239         int error;
  240 
  241         error = pre_execve(td, &oldvmspace);
  242         if (error != 0)
  243                 return (error);
  244         error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
  245             uap->argv, uap->envv);
  246         if (error == 0) {
  247                 args.fd = uap->fd;
  248                 error = kern_execve(td, &args, NULL);
  249         }
  250         post_execve(td, error, oldvmspace);
  251         return (error);
  252 }
  253 
  254 #ifndef _SYS_SYSPROTO_H_
  255 struct __mac_execve_args {
  256         char    *fname;
  257         char    **argv;
  258         char    **envv;
  259         struct mac      *mac_p;
  260 };
  261 #endif
  262 
  263 int
  264 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
  265 {
  266 #ifdef MAC
  267         struct image_args args;
  268         struct vmspace *oldvmspace;
  269         int error;
  270 
  271         error = pre_execve(td, &oldvmspace);
  272         if (error != 0)
  273                 return (error);
  274         error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
  275             uap->argv, uap->envv);
  276         if (error == 0)
  277                 error = kern_execve(td, &args, uap->mac_p);
  278         post_execve(td, error, oldvmspace);
  279         return (error);
  280 #else
  281         return (ENOSYS);
  282 #endif
  283 }
  284 
  285 int
  286 pre_execve(struct thread *td, struct vmspace **oldvmspace)
  287 {
  288         struct proc *p;
  289         int error;
  290 
  291         KASSERT(td == curthread, ("non-current thread %p", td));
  292         error = 0;
  293         p = td->td_proc;
  294         if ((p->p_flag & P_HADTHREADS) != 0) {
  295                 PROC_LOCK(p);
  296                 if (thread_single(p, SINGLE_BOUNDARY) != 0)
  297                         error = ERESTART;
  298                 PROC_UNLOCK(p);
  299         }
  300         KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
  301             ("nested execve"));
  302         *oldvmspace = p->p_vmspace;
  303         return (error);
  304 }
  305 
  306 void
  307 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
  308 {
  309         struct proc *p;
  310 
  311         KASSERT(td == curthread, ("non-current thread %p", td));
  312         p = td->td_proc;
  313         if ((p->p_flag & P_HADTHREADS) != 0) {
  314                 PROC_LOCK(p);
  315                 /*
  316                  * If success, we upgrade to SINGLE_EXIT state to
  317                  * force other threads to suicide.
  318                  */
  319                 if (error == 0)
  320                         thread_single(p, SINGLE_EXIT);
  321                 else
  322                         thread_single_end(p, SINGLE_BOUNDARY);
  323                 PROC_UNLOCK(p);
  324         }
  325         if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
  326                 KASSERT(p->p_vmspace != oldvmspace,
  327                     ("oldvmspace still used"));
  328                 vmspace_free(oldvmspace);
  329                 td->td_pflags &= ~TDP_EXECVMSPC;
  330         }
  331 }
  332 
  333 /*
  334  * XXX: kern_execve has the astonishing property of not always returning to
  335  * the caller.  If sufficiently bad things happen during the call to
  336  * do_execve(), it can end up calling exit1(); as a result, callers must
  337  * avoid doing anything which they might need to undo (e.g., allocating
  338  * memory).
  339  */
  340 int
  341 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
  342 {
  343 
  344         AUDIT_ARG_ARGV(args->begin_argv, args->argc,
  345             args->begin_envv - args->begin_argv);
  346         AUDIT_ARG_ENVV(args->begin_envv, args->envc,
  347             args->endp - args->begin_envv);
  348         return (do_execve(td, args, mac_p));
  349 }
  350 
  351 /*
  352  * In-kernel implementation of execve().  All arguments are assumed to be
  353  * userspace pointers from the passed thread.
  354  */
  355 static int
  356 do_execve(td, args, mac_p)
  357         struct thread *td;
  358         struct image_args *args;
  359         struct mac *mac_p;
  360 {
  361         struct proc *p = td->td_proc;
  362         struct nameidata nd;
  363         struct ucred *oldcred;
  364         struct uidinfo *euip = NULL;
  365         register_t *stack_base;
  366         int error, i;
  367         struct image_params image_params, *imgp;
  368         struct vattr attr;
  369         int (*img_first)(struct image_params *);
  370         struct pargs *oldargs = NULL, *newargs = NULL;
  371         struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
  372 #ifdef KTRACE
  373         struct vnode *tracevp = NULL;
  374         struct ucred *tracecred = NULL;
  375 #endif
  376         struct vnode *oldtextvp = NULL, *newtextvp;
  377         cap_rights_t rights;
  378         int credential_changing;
  379         int textset;
  380 #ifdef MAC
  381         struct label *interpvplabel = NULL;
  382         int will_transition;
  383 #endif
  384 #ifdef HWPMC_HOOKS
  385         struct pmckern_procexec pe;
  386 #endif
  387         static const char fexecv_proc_title[] = "(fexecv)";
  388 
  389         imgp = &image_params;
  390 
  391         /*
  392          * Lock the process and set the P_INEXEC flag to indicate that
  393          * it should be left alone until we're done here.  This is
  394          * necessary to avoid race conditions - e.g. in ptrace() -
  395          * that might allow a local user to illicitly obtain elevated
  396          * privileges.
  397          */
  398         PROC_LOCK(p);
  399         KASSERT((p->p_flag & P_INEXEC) == 0,
  400             ("%s(): process already has P_INEXEC flag", __func__));
  401         p->p_flag |= P_INEXEC;
  402         PROC_UNLOCK(p);
  403 
  404         /*
  405          * Initialize part of the common data
  406          */
  407         bzero(imgp, sizeof(*imgp));
  408         imgp->proc = p;
  409         imgp->attr = &attr;
  410         imgp->args = args;
  411         oldcred = p->p_ucred;
  412 
  413 #ifdef MAC
  414         error = mac_execve_enter(imgp, mac_p);
  415         if (error)
  416                 goto exec_fail;
  417 #endif
  418 
  419         /*
  420          * Translate the file name. namei() returns a vnode pointer
  421          *      in ni_vp among other things.
  422          *
  423          * XXXAUDIT: It would be desirable to also audit the name of the
  424          * interpreter if this is an interpreted binary.
  425          */
  426         if (args->fname != NULL) {
  427                 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
  428                     | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
  429         }
  430 
  431         SDT_PROBE1(proc, , , exec, args->fname);
  432 
  433 interpret:
  434         if (args->fname != NULL) {
  435 #ifdef CAPABILITY_MODE
  436                 /*
  437                  * While capability mode can't reach this point via direct
  438                  * path arguments to execve(), we also don't allow
  439                  * interpreters to be used in capability mode (for now).
  440                  * Catch indirect lookups and return a permissions error.
  441                  */
  442                 if (IN_CAPABILITY_MODE(td)) {
  443                         error = ECAPMODE;
  444                         goto exec_fail;
  445                 }
  446 #endif
  447                 error = namei(&nd);
  448                 if (error)
  449                         goto exec_fail;
  450 
  451                 newtextvp = nd.ni_vp;
  452                 imgp->vp = newtextvp;
  453         } else {
  454                 AUDIT_ARG_FD(args->fd);
  455                 /*
  456                  * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
  457                  */
  458                 error = fgetvp_exec(td, args->fd,
  459                     cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
  460                 if (error)
  461                         goto exec_fail;
  462                 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
  463                 AUDIT_ARG_VNODE1(newtextvp);
  464                 imgp->vp = newtextvp;
  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         /*
  495          * Implement image setuid/setgid.
  496          *
  497          * Determine new credentials before attempting image activators
  498          * so that it can be used by process_exec handlers to determine
  499          * credential/setid changes.
  500          *
  501          * Don't honor setuid/setgid if the filesystem prohibits it or if
  502          * the process is being traced.
  503          *
  504          * We disable setuid/setgid/etc in capability mode on the basis
  505          * that most setugid applications are not written with that
  506          * environment in mind, and will therefore almost certainly operate
  507          * incorrectly. In principle there's no reason that setugid
  508          * applications might not be useful in capability mode, so we may want
  509          * to reconsider this conservative design choice in the future.
  510          *
  511          * XXXMAC: For the time being, use NOSUID to also prohibit
  512          * transitions on the file system.
  513          */
  514         credential_changing = 0;
  515         credential_changing |= (attr.va_mode & S_ISUID) &&
  516             oldcred->cr_uid != attr.va_uid;
  517         credential_changing |= (attr.va_mode & S_ISGID) &&
  518             oldcred->cr_gid != attr.va_gid;
  519 #ifdef MAC
  520         will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
  521             interpvplabel, imgp);
  522         credential_changing |= will_transition;
  523 #endif
  524 
  525         if (credential_changing &&
  526 #ifdef CAPABILITY_MODE
  527             ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
  528 #endif
  529             (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
  530             (p->p_flag & P_TRACED) == 0) {
  531                 imgp->credential_setid = true;
  532                 VOP_UNLOCK(imgp->vp, 0);
  533                 imgp->newcred = crdup(oldcred);
  534                 if (attr.va_mode & S_ISUID) {
  535                         euip = uifind(attr.va_uid);
  536                         change_euid(imgp->newcred, euip);
  537                 }
  538                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
  539                 if (attr.va_mode & S_ISGID)
  540                         change_egid(imgp->newcred, attr.va_gid);
  541                 /*
  542                  * Implement correct POSIX saved-id behavior.
  543                  *
  544                  * XXXMAC: Note that the current logic will save the
  545                  * uid and gid if a MAC domain transition occurs, even
  546                  * though maybe it shouldn't.
  547                  */
  548                 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
  549                 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
  550         } else {
  551                 /*
  552                  * Implement correct POSIX saved-id behavior.
  553                  *
  554                  * XXX: It's not clear that the existing behavior is
  555                  * POSIX-compliant.  A number of sources indicate that the
  556                  * saved uid/gid should only be updated if the new ruid is
  557                  * not equal to the old ruid, or the new euid is not equal
  558                  * to the old euid and the new euid is not equal to the old
  559                  * ruid.  The FreeBSD code always updates the saved uid/gid.
  560                  * Also, this code uses the new (replaced) euid and egid as
  561                  * the source, which may or may not be the right ones to use.
  562                  */
  563                 if (oldcred->cr_svuid != oldcred->cr_uid ||
  564                     oldcred->cr_svgid != oldcred->cr_gid) {
  565                         VOP_UNLOCK(imgp->vp, 0);
  566                         imgp->newcred = crdup(oldcred);
  567                         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
  568                         change_svuid(imgp->newcred, imgp->newcred->cr_uid);
  569                         change_svgid(imgp->newcred, imgp->newcred->cr_gid);
  570                 }
  571         }
  572         /* The new credentials are installed into the process later. */
  573 
  574         /*
  575          * Do the best to calculate the full path to the image file.
  576          */
  577         if (args->fname != NULL && args->fname[0] == '/')
  578                 imgp->execpath = args->fname;
  579         else {
  580                 VOP_UNLOCK(imgp->vp, 0);
  581                 if (vn_fullpath(td, imgp->vp, &imgp->execpath,
  582                     &imgp->freepath) != 0)
  583                         imgp->execpath = args->fname;
  584                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
  585         }
  586 
  587         /*
  588          *      If the current process has a special image activator it
  589          *      wants to try first, call it.   For example, emulating shell
  590          *      scripts differently.
  591          */
  592         error = -1;
  593         if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
  594                 error = img_first(imgp);
  595 
  596         /*
  597          *      Loop through the list of image activators, calling each one.
  598          *      An activator returns -1 if there is no match, 0 on success,
  599          *      and an error otherwise.
  600          */
  601         for (i = 0; error == -1 && execsw[i]; ++i) {
  602                 if (execsw[i]->ex_imgact == NULL ||
  603                     execsw[i]->ex_imgact == img_first) {
  604                         continue;
  605                 }
  606                 error = (*execsw[i]->ex_imgact)(imgp);
  607         }
  608 
  609         if (error) {
  610                 if (error == -1) {
  611                         if (textset == 0)
  612                                 VOP_UNSET_TEXT(imgp->vp);
  613                         error = ENOEXEC;
  614                 }
  615                 goto exec_fail_dealloc;
  616         }
  617 
  618         /*
  619          * Special interpreter operation, cleanup and loop up to try to
  620          * activate the interpreter.
  621          */
  622         if (imgp->interpreted) {
  623                 exec_unmap_first_page(imgp);
  624                 /*
  625                  * VV_TEXT needs to be unset for scripts.  There is a short
  626                  * period before we determine that something is a script where
  627                  * VV_TEXT will be set. The vnode lock is held over this
  628                  * entire period so nothing should illegitimately be blocked.
  629                  */
  630                 VOP_UNSET_TEXT(imgp->vp);
  631                 /* free name buffer and old vnode */
  632                 if (args->fname != NULL)
  633                         NDFREE(&nd, NDF_ONLY_PNBUF);
  634 #ifdef MAC
  635                 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
  636 #endif
  637                 if (imgp->opened) {
  638                         VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
  639                         imgp->opened = 0;
  640                 }
  641                 vput(newtextvp);
  642                 vm_object_deallocate(imgp->object);
  643                 imgp->object = NULL;
  644                 imgp->credential_setid = false;
  645                 if (imgp->newcred != NULL) {
  646                         crfree(imgp->newcred);
  647                         imgp->newcred = NULL;
  648                 }
  649                 imgp->execpath = NULL;
  650                 free(imgp->freepath, M_TEMP);
  651                 imgp->freepath = NULL;
  652                 /* set new name to that of the interpreter */
  653                 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
  654                     UIO_SYSSPACE, imgp->interpreter_name, td);
  655                 args->fname = imgp->interpreter_name;
  656                 goto interpret;
  657         }
  658 
  659         /*
  660          * NB: We unlock the vnode here because it is believed that none
  661          * of the sv_copyout_strings/sv_fixup operations require the vnode.
  662          */
  663         VOP_UNLOCK(imgp->vp, 0);
  664 
  665         if (disallow_high_osrel &&
  666             P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
  667                 error = ENOEXEC;
  668                 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
  669                     imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
  670                 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
  671                 goto exec_fail_dealloc;
  672         }
  673 
  674         /*
  675          * Copy out strings (args and env) and initialize stack base
  676          */
  677         if (p->p_sysent->sv_copyout_strings)
  678                 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
  679         else
  680                 stack_base = exec_copyout_strings(imgp);
  681 
  682         /*
  683          * If custom stack fixup routine present for this process
  684          * let it do the stack setup.
  685          * Else stuff argument count as first item on stack
  686          */
  687         if (p->p_sysent->sv_fixup != NULL)
  688                 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
  689         else
  690                 suword(--stack_base, imgp->args->argc);
  691 
  692         /*
  693          * For security and other reasons, the file descriptor table cannot
  694          * be shared after an exec.
  695          */
  696         fdunshare(td);
  697         /* close files on exec */
  698         fdcloseexec(td);
  699 
  700         /*
  701          * Malloc things before we need locks.
  702          */
  703         i = imgp->args->begin_envv - imgp->args->begin_argv;
  704         /* Cache arguments if they fit inside our allowance */
  705         if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
  706                 newargs = pargs_alloc(i);
  707                 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
  708         }
  709 
  710         /*
  711          * For security and other reasons, signal handlers cannot
  712          * be shared after an exec. The new process gets a copy of the old
  713          * handlers. In execsigs(), the new process will have its signals
  714          * reset.
  715          */
  716         if (sigacts_shared(p->p_sigacts)) {
  717                 oldsigacts = p->p_sigacts;
  718                 newsigacts = sigacts_alloc();
  719                 sigacts_copy(newsigacts, oldsigacts);
  720         }
  721 
  722         vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
  723 
  724         PROC_LOCK(p);
  725         if (oldsigacts)
  726                 p->p_sigacts = newsigacts;
  727         /* Stop profiling */
  728         stopprofclock(p);
  729 
  730         /* reset caught signals */
  731         execsigs(p);
  732 
  733         /* name this process - nameiexec(p, ndp) */
  734         bzero(p->p_comm, sizeof(p->p_comm));
  735         if (args->fname)
  736                 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
  737                     min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
  738         else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
  739                 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
  740         bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
  741 #ifdef KTR
  742         sched_clear_tdname(td);
  743 #endif
  744 
  745         /*
  746          * mark as execed, wakeup the process that vforked (if any) and tell
  747          * it that it now has its own resources back
  748          */
  749         p->p_flag |= P_EXEC;
  750         if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
  751                 p->p_flag2 &= ~P2_NOTRACE;
  752         if (p->p_flag & P_PPWAIT) {
  753                 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
  754                 cv_broadcast(&p->p_pwait);
  755                 /* STOPs are no longer ignored, arrange for AST */
  756                 signotify(td);
  757         }
  758 
  759         /*
  760          * Implement image setuid/setgid installation.
  761          */
  762         if (imgp->credential_setid) {
  763                 /*
  764                  * Turn off syscall tracing for set-id programs, except for
  765                  * root.  Record any set-id flags first to make sure that
  766                  * we do not regain any tracing during a possible block.
  767                  */
  768                 setsugid(p);
  769 
  770 #ifdef KTRACE
  771                 if (p->p_tracecred != NULL &&
  772                     priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
  773                         ktrprocexec(p, &tracecred, &tracevp);
  774 #endif
  775                 /*
  776                  * Close any file descriptors 0..2 that reference procfs,
  777                  * then make sure file descriptors 0..2 are in use.
  778                  *
  779                  * setugidsafety() may call closef() and then pfind()
  780                  * which may grab the process lock.
  781                  * fdcheckstd() may call falloc() which may block to
  782                  * allocate memory, so temporarily drop the process lock.
  783                  */
  784                 PROC_UNLOCK(p);
  785                 VOP_UNLOCK(imgp->vp, 0);
  786                 setugidsafety(td);
  787                 error = fdcheckstd(td);
  788                 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
  789                 if (error != 0)
  790                         goto exec_fail_dealloc;
  791                 PROC_LOCK(p);
  792 #ifdef MAC
  793                 if (will_transition) {
  794                         mac_vnode_execve_transition(oldcred, imgp->newcred,
  795                             imgp->vp, interpvplabel, imgp);
  796                 }
  797 #endif
  798         } else {
  799                 if (oldcred->cr_uid == oldcred->cr_ruid &&
  800                     oldcred->cr_gid == oldcred->cr_rgid)
  801                         p->p_flag &= ~P_SUGID;
  802         }
  803         /*
  804          * Set the new credentials.
  805          */
  806         if (imgp->newcred != NULL) {
  807                 proc_set_cred(p, imgp->newcred);
  808                 crfree(oldcred);
  809                 oldcred = NULL;
  810         }
  811 
  812         /*
  813          * Store the vp for use in procfs.  This vnode was referenced by namei
  814          * or fgetvp_exec.
  815          */
  816         oldtextvp = p->p_textvp;
  817         p->p_textvp = newtextvp;
  818 
  819 #ifdef KDTRACE_HOOKS
  820         /*
  821          * Tell the DTrace fasttrap provider about the exec if it
  822          * has declared an interest.
  823          */
  824         if (dtrace_fasttrap_exec)
  825                 dtrace_fasttrap_exec(p);
  826 #endif
  827 
  828         /*
  829          * Notify others that we exec'd, and clear the P_INEXEC flag
  830          * as we're now a bona fide freshly-execed process.
  831          */
  832         KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
  833         p->p_flag &= ~P_INEXEC;
  834 
  835         /* clear "fork but no exec" flag, as we _are_ execing */
  836         p->p_acflag &= ~AFORK;
  837 
  838         /*
  839          * Free any previous argument cache and replace it with
  840          * the new argument cache, if any.
  841          */
  842         oldargs = p->p_args;
  843         p->p_args = newargs;
  844         newargs = NULL;
  845 
  846 #ifdef  HWPMC_HOOKS
  847         /*
  848          * Check if system-wide sampling is in effect or if the
  849          * current process is using PMCs.  If so, do exec() time
  850          * processing.  This processing needs to happen AFTER the
  851          * P_INEXEC flag is cleared.
  852          *
  853          * The proc lock needs to be released before taking the PMC
  854          * SX.
  855          */
  856         if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
  857                 PROC_UNLOCK(p);
  858                 VOP_UNLOCK(imgp->vp, 0);
  859                 pe.pm_credentialschanged = credential_changing;
  860                 pe.pm_entryaddr = imgp->entry_addr;
  861 
  862                 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
  863                 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
  864         } else
  865                 PROC_UNLOCK(p);
  866 #else  /* !HWPMC_HOOKS */
  867         PROC_UNLOCK(p);
  868 #endif
  869 
  870         /* Set values passed into the program in registers. */
  871         if (p->p_sysent->sv_setregs)
  872                 (*p->p_sysent->sv_setregs)(td, imgp, 
  873                     (u_long)(uintptr_t)stack_base);
  874         else
  875                 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
  876 
  877         vfs_mark_atime(imgp->vp, td->td_ucred);
  878 
  879         SDT_PROBE1(proc, , , exec__success, args->fname);
  880 
  881 exec_fail_dealloc:
  882         if (imgp->firstpage != NULL)
  883                 exec_unmap_first_page(imgp);
  884 
  885         if (imgp->vp != NULL) {
  886                 if (args->fname)
  887                         NDFREE(&nd, NDF_ONLY_PNBUF);
  888                 if (imgp->opened)
  889                         VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
  890                 if (error != 0)
  891                         vput(imgp->vp);
  892                 else
  893                         VOP_UNLOCK(imgp->vp, 0);
  894         }
  895 
  896         if (imgp->object != NULL)
  897                 vm_object_deallocate(imgp->object);
  898 
  899         free(imgp->freepath, M_TEMP);
  900 
  901         if (error == 0) {
  902                 PROC_LOCK(p);
  903                 if (p->p_ptevents & PTRACE_EXEC)
  904                         td->td_dbgflags |= TDB_EXEC;
  905                 PROC_UNLOCK(p);
  906 
  907                 /*
  908                  * Stop the process here if its stop event mask has
  909                  * the S_EXEC bit set.
  910                  */
  911                 STOPEVENT(p, S_EXEC, 0);
  912         } else {
  913 exec_fail:
  914                 /* we're done here, clear P_INEXEC */
  915                 PROC_LOCK(p);
  916                 p->p_flag &= ~P_INEXEC;
  917                 PROC_UNLOCK(p);
  918 
  919                 SDT_PROBE1(proc, , , exec__failure, error);
  920         }
  921 
  922         if (imgp->newcred != NULL && oldcred != NULL)
  923                 crfree(imgp->newcred);
  924 
  925 #ifdef MAC
  926         mac_execve_exit(imgp);
  927         mac_execve_interpreter_exit(interpvplabel);
  928 #endif
  929         exec_free_args(args);
  930 
  931         /*
  932          * Handle deferred decrement of ref counts.
  933          */
  934         if (oldtextvp != NULL)
  935                 vrele(oldtextvp);
  936 #ifdef KTRACE
  937         if (tracevp != NULL)
  938                 vrele(tracevp);
  939         if (tracecred != NULL)
  940                 crfree(tracecred);
  941 #endif
  942         pargs_drop(oldargs);
  943         pargs_drop(newargs);
  944         if (oldsigacts != NULL)
  945                 sigacts_free(oldsigacts);
  946         if (euip != NULL)
  947                 uifree(euip);
  948 
  949         if (error && imgp->vmspace_destroyed) {
  950                 /* sorry, no more process anymore. exit gracefully */
  951                 exit1(td, W_EXITCODE(0, SIGABRT));
  952                 /* NOT REACHED */
  953         }
  954 
  955 #ifdef KTRACE
  956         if (error == 0)
  957                 ktrprocctor(p);
  958 #endif
  959 
  960         return (error);
  961 }
  962 
  963 int
  964 exec_map_first_page(imgp)
  965         struct image_params *imgp;
  966 {
  967         int rv, i;
  968         int initial_pagein;
  969         vm_page_t ma[VM_INITIAL_PAGEIN];
  970         vm_object_t object;
  971 
  972         if (imgp->firstpage != NULL)
  973                 exec_unmap_first_page(imgp);
  974 
  975         object = imgp->vp->v_object;
  976         if (object == NULL)
  977                 return (EACCES);
  978         VM_OBJECT_WLOCK(object);
  979 #if VM_NRESERVLEVEL > 0
  980         if ((object->flags & OBJ_COLORED) == 0) {
  981                 object->flags |= OBJ_COLORED;
  982                 object->pg_color = 0;
  983         }
  984 #endif
  985         ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
  986         if (ma[0]->valid != VM_PAGE_BITS_ALL) {
  987                 initial_pagein = VM_INITIAL_PAGEIN;
  988                 if (initial_pagein > object->size)
  989                         initial_pagein = object->size;
  990                 for (i = 1; i < initial_pagein; i++) {
  991                         if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
  992                                 if (ma[i]->valid)
  993                                         break;
  994                                 if (vm_page_tryxbusy(ma[i]))
  995                                         break;
  996                         } else {
  997                                 ma[i] = vm_page_alloc(object, i,
  998                                     VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
  999                                 if (ma[i] == NULL)
 1000                                         break;
 1001                         }
 1002                 }
 1003                 initial_pagein = i;
 1004                 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
 1005                 ma[0] = vm_page_lookup(object, 0);
 1006                 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
 1007                         if (ma[0] != NULL) {
 1008                                 vm_page_lock(ma[0]);
 1009                                 vm_page_free(ma[0]);
 1010                                 vm_page_unlock(ma[0]);
 1011                         }
 1012                         VM_OBJECT_WUNLOCK(object);
 1013                         return (EIO);
 1014                 }
 1015         }
 1016         vm_page_xunbusy(ma[0]);
 1017         vm_page_lock(ma[0]);
 1018         vm_page_hold(ma[0]);
 1019         vm_page_activate(ma[0]);
 1020         vm_page_unlock(ma[0]);
 1021         VM_OBJECT_WUNLOCK(object);
 1022 
 1023         imgp->firstpage = sf_buf_alloc(ma[0], 0);
 1024         imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
 1025 
 1026         return (0);
 1027 }
 1028 
 1029 void
 1030 exec_unmap_first_page(imgp)
 1031         struct image_params *imgp;
 1032 {
 1033         vm_page_t m;
 1034 
 1035         if (imgp->firstpage != NULL) {
 1036                 m = sf_buf_page(imgp->firstpage);
 1037                 sf_buf_free(imgp->firstpage);
 1038                 imgp->firstpage = NULL;
 1039                 vm_page_lock(m);
 1040                 vm_page_unhold(m);
 1041                 vm_page_unlock(m);
 1042         }
 1043 }
 1044 
 1045 /*
 1046  * Destroy old address space, and allocate a new stack.
 1047  *      The new stack is only sgrowsiz large because it is grown
 1048  *      automatically on a page fault.
 1049  */
 1050 int
 1051 exec_new_vmspace(imgp, sv)
 1052         struct image_params *imgp;
 1053         struct sysentvec *sv;
 1054 {
 1055         int error;
 1056         struct proc *p = imgp->proc;
 1057         struct vmspace *vmspace = p->p_vmspace;
 1058         vm_object_t obj;
 1059         struct rlimit rlim_stack;
 1060         vm_offset_t sv_minuser, stack_addr;
 1061         vm_map_t map;
 1062         u_long ssiz;
 1063 
 1064         imgp->vmspace_destroyed = 1;
 1065         imgp->sysent = sv;
 1066 
 1067         /* May be called with Giant held */
 1068         EVENTHANDLER_INVOKE(process_exec, p, imgp);
 1069 
 1070         /*
 1071          * Blow away entire process VM, if address space not shared,
 1072          * otherwise, create a new VM space so that other threads are
 1073          * not disrupted
 1074          */
 1075         map = &vmspace->vm_map;
 1076         if (map_at_zero)
 1077                 sv_minuser = sv->sv_minuser;
 1078         else
 1079                 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
 1080         if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
 1081             vm_map_max(map) == sv->sv_maxuser) {
 1082                 shmexit(vmspace);
 1083                 pmap_remove_pages(vmspace_pmap(vmspace));
 1084                 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
 1085                 /* An exec terminates mlockall(MCL_FUTURE). */
 1086                 vm_map_lock(map);
 1087                 vm_map_modflags(map, 0, MAP_WIREFUTURE);
 1088                 vm_map_unlock(map);
 1089         } else {
 1090                 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
 1091                 if (error)
 1092                         return (error);
 1093                 vmspace = p->p_vmspace;
 1094                 map = &vmspace->vm_map;
 1095         }
 1096 
 1097         /* Map a shared page */
 1098         obj = sv->sv_shared_page_obj;
 1099         if (obj != NULL) {
 1100                 vm_object_reference(obj);
 1101                 error = vm_map_fixed(map, obj, 0,
 1102                     sv->sv_shared_page_base, sv->sv_shared_page_len,
 1103                     VM_PROT_READ | VM_PROT_EXECUTE,
 1104                     VM_PROT_READ | VM_PROT_EXECUTE,
 1105                     MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
 1106                 if (error != KERN_SUCCESS) {
 1107                         vm_object_deallocate(obj);
 1108                         return (vm_mmap_to_errno(error));
 1109                 }
 1110         }
 1111 
 1112         /* Allocate a new stack */
 1113         if (imgp->stack_sz != 0) {
 1114                 ssiz = trunc_page(imgp->stack_sz);
 1115                 PROC_LOCK(p);
 1116                 lim_rlimit(p, RLIMIT_STACK, &rlim_stack);
 1117                 PROC_UNLOCK(p);
 1118                 if (ssiz > rlim_stack.rlim_max)
 1119                         ssiz = rlim_stack.rlim_max;
 1120                 if (ssiz > rlim_stack.rlim_cur) {
 1121                         rlim_stack.rlim_cur = ssiz;
 1122                         kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
 1123                 }
 1124         } else if (sv->sv_maxssiz != NULL) {
 1125                 ssiz = *sv->sv_maxssiz;
 1126         } else {
 1127                 ssiz = maxssiz;
 1128         }
 1129         stack_addr = sv->sv_usrstack - ssiz;
 1130         error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
 1131             obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
 1132             sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
 1133         if (error != KERN_SUCCESS)
 1134                 return (vm_mmap_to_errno(error));
 1135 
 1136 #ifdef __ia64__
 1137         /* Allocate a new register stack */
 1138         error = vm_map_stack(map, IA64_BACKINGSTORE, (vm_size_t)ssiz,
 1139             sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
 1140         if (error)
 1141                 return (error);
 1142 #endif
 1143 
 1144         /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
 1145          * VM_STACK case, but they are still used to monitor the size of the
 1146          * process stack so we can check the stack rlimit.
 1147          */
 1148         vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
 1149         vmspace->vm_maxsaddr = (char *)stack_addr;
 1150 
 1151         return (0);
 1152 }
 1153 
 1154 /*
 1155  * Copy out argument and environment strings from the old process address
 1156  * space into the temporary string buffer.
 1157  */
 1158 int
 1159 exec_copyin_args(struct image_args *args, char *fname,
 1160     enum uio_seg segflg, char **argv, char **envv)
 1161 {
 1162         u_long argp, envp;
 1163         int error;
 1164         size_t length;
 1165 
 1166         bzero(args, sizeof(*args));
 1167         if (argv == NULL)
 1168                 return (EFAULT);
 1169 
 1170         /*
 1171          * Allocate demand-paged memory for the file name, argument, and
 1172          * environment strings.
 1173          */
 1174         error = exec_alloc_args(args);
 1175         if (error != 0)
 1176                 return (error);
 1177 
 1178         /*
 1179          * Copy the file name.
 1180          */
 1181         if (fname != NULL) {
 1182                 args->fname = args->buf;
 1183                 error = (segflg == UIO_SYSSPACE) ?
 1184                     copystr(fname, args->fname, PATH_MAX, &length) :
 1185                     copyinstr(fname, args->fname, PATH_MAX, &length);
 1186                 if (error != 0)
 1187                         goto err_exit;
 1188         } else
 1189                 length = 0;
 1190 
 1191         args->begin_argv = args->buf + length;
 1192         args->endp = args->begin_argv;
 1193         args->stringspace = ARG_MAX;
 1194 
 1195         /*
 1196          * extract arguments first
 1197          */
 1198         for (;;) {
 1199                 error = fueword(argv++, &argp);
 1200                 if (error == -1) {
 1201                         error = EFAULT;
 1202                         goto err_exit;
 1203                 }
 1204                 if (argp == 0)
 1205                         break;
 1206                 error = copyinstr((void *)(uintptr_t)argp, args->endp,
 1207                     args->stringspace, &length);
 1208                 if (error != 0) {
 1209                         if (error == ENAMETOOLONG) 
 1210                                 error = E2BIG;
 1211                         goto err_exit;
 1212                 }
 1213                 args->stringspace -= length;
 1214                 args->endp += length;
 1215                 args->argc++;
 1216         }
 1217 
 1218         args->begin_envv = args->endp;
 1219 
 1220         /*
 1221          * extract environment strings
 1222          */
 1223         if (envv) {
 1224                 for (;;) {
 1225                         error = fueword(envv++, &envp);
 1226                         if (error == -1) {
 1227                                 error = EFAULT;
 1228                                 goto err_exit;
 1229                         }
 1230                         if (envp == 0)
 1231                                 break;
 1232                         error = copyinstr((void *)(uintptr_t)envp,
 1233                             args->endp, args->stringspace, &length);
 1234                         if (error != 0) {
 1235                                 if (error == ENAMETOOLONG)
 1236                                         error = E2BIG;
 1237                                 goto err_exit;
 1238                         }
 1239                         args->stringspace -= length;
 1240                         args->endp += length;
 1241                         args->envc++;
 1242                 }
 1243         }
 1244 
 1245         return (0);
 1246 
 1247 err_exit:
 1248         exec_free_args(args);
 1249         return (error);
 1250 }
 1251 
 1252 /*
 1253  * Allocate temporary demand-paged, zero-filled memory for the file name,
 1254  * argument, and environment strings.  Returns zero if the allocation succeeds
 1255  * and ENOMEM otherwise.
 1256  */
 1257 int
 1258 exec_alloc_args(struct image_args *args)
 1259 {
 1260 
 1261         args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
 1262         return (args->buf != NULL ? 0 : ENOMEM);
 1263 }
 1264 
 1265 void
 1266 exec_free_args(struct image_args *args)
 1267 {
 1268 
 1269         if (args->buf != NULL) {
 1270                 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
 1271                     PATH_MAX + ARG_MAX);
 1272                 args->buf = NULL;
 1273         }
 1274         if (args->fname_buf != NULL) {
 1275                 free(args->fname_buf, M_TEMP);
 1276                 args->fname_buf = NULL;
 1277         }
 1278 }
 1279 
 1280 /*
 1281  * Copy strings out to the new process address space, constructing new arg
 1282  * and env vector tables. Return a pointer to the base so that it can be used
 1283  * as the initial stack pointer.
 1284  */
 1285 register_t *
 1286 exec_copyout_strings(imgp)
 1287         struct image_params *imgp;
 1288 {
 1289         int argc, envc;
 1290         char **vectp;
 1291         char *stringp;
 1292         uintptr_t destp;
 1293         register_t *stack_base;
 1294         struct ps_strings *arginfo;
 1295         struct proc *p;
 1296         size_t execpath_len;
 1297         int szsigcode, szps;
 1298         char canary[sizeof(long) * 8];
 1299 
 1300         szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
 1301         /*
 1302          * Calculate string base and vector table pointers.
 1303          * Also deal with signal trampoline code for this exec type.
 1304          */
 1305         if (imgp->execpath != NULL && imgp->auxargs != NULL)
 1306                 execpath_len = strlen(imgp->execpath) + 1;
 1307         else
 1308                 execpath_len = 0;
 1309         p = imgp->proc;
 1310         szsigcode = 0;
 1311         arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
 1312         if (p->p_sysent->sv_sigcode_base == 0) {
 1313                 if (p->p_sysent->sv_szsigcode != NULL)
 1314                         szsigcode = *(p->p_sysent->sv_szsigcode);
 1315         }
 1316         destp = (uintptr_t)arginfo;
 1317 
 1318         /*
 1319          * install sigcode
 1320          */
 1321         if (szsigcode != 0) {
 1322                 destp -= szsigcode;
 1323                 destp = rounddown2(destp, sizeof(void *));
 1324                 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
 1325         }
 1326 
 1327         /*
 1328          * Copy the image path for the rtld.
 1329          */
 1330         if (execpath_len != 0) {
 1331                 destp -= execpath_len;
 1332                 imgp->execpathp = destp;
 1333                 copyout(imgp->execpath, (void *)destp, execpath_len);
 1334         }
 1335 
 1336         /*
 1337          * Prepare the canary for SSP.
 1338          */
 1339         arc4rand(canary, sizeof(canary), 0);
 1340         destp -= sizeof(canary);
 1341         imgp->canary = destp;
 1342         copyout(canary, (void *)destp, sizeof(canary));
 1343         imgp->canarylen = sizeof(canary);
 1344 
 1345         /*
 1346          * Prepare the pagesizes array.
 1347          */
 1348         destp -= szps;
 1349         destp = rounddown2(destp, sizeof(void *));
 1350         imgp->pagesizes = destp;
 1351         copyout(pagesizes, (void *)destp, szps);
 1352         imgp->pagesizeslen = szps;
 1353 
 1354         destp -= ARG_MAX - imgp->args->stringspace;
 1355         destp = rounddown2(destp, sizeof(void *));
 1356 
 1357         /*
 1358          * If we have a valid auxargs ptr, prepare some room
 1359          * on the stack.
 1360          */
 1361         if (imgp->auxargs) {
 1362                 /*
 1363                  * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
 1364                  * lower compatibility.
 1365                  */
 1366                 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
 1367                     (AT_COUNT * 2);
 1368                 /*
 1369                  * The '+ 2' is for the null pointers at the end of each of
 1370                  * the arg and env vector sets,and imgp->auxarg_size is room
 1371                  * for argument of Runtime loader.
 1372                  */
 1373                 vectp = (char **)(destp - (imgp->args->argc +
 1374                     imgp->args->envc + 2 + imgp->auxarg_size)
 1375                     * sizeof(char *));
 1376         } else {
 1377                 /*
 1378                  * The '+ 2' is for the null pointers at the end of each of
 1379                  * the arg and env vector sets
 1380                  */
 1381                 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
 1382                     + 2) * sizeof(char *));
 1383         }
 1384 
 1385         /*
 1386          * vectp also becomes our initial stack base
 1387          */
 1388         stack_base = (register_t *)vectp;
 1389 
 1390         stringp = imgp->args->begin_argv;
 1391         argc = imgp->args->argc;
 1392         envc = imgp->args->envc;
 1393 
 1394         /*
 1395          * Copy out strings - arguments and environment.
 1396          */
 1397         copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
 1398 
 1399         /*
 1400          * Fill in "ps_strings" struct for ps, w, etc.
 1401          */
 1402         suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
 1403         suword32(&arginfo->ps_nargvstr, argc);
 1404 
 1405         /*
 1406          * Fill in argument portion of vector table.
 1407          */
 1408         for (; argc > 0; --argc) {
 1409                 suword(vectp++, (long)(intptr_t)destp);
 1410                 while (*stringp++ != 0)
 1411                         destp++;
 1412                 destp++;
 1413         }
 1414 
 1415         /* a null vector table pointer separates the argp's from the envp's */
 1416         suword(vectp++, 0);
 1417 
 1418         suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
 1419         suword32(&arginfo->ps_nenvstr, envc);
 1420 
 1421         /*
 1422          * Fill in environment portion of vector table.
 1423          */
 1424         for (; envc > 0; --envc) {
 1425                 suword(vectp++, (long)(intptr_t)destp);
 1426                 while (*stringp++ != 0)
 1427                         destp++;
 1428                 destp++;
 1429         }
 1430 
 1431         /* end of vector table is a null pointer */
 1432         suword(vectp, 0);
 1433 
 1434         return (stack_base);
 1435 }
 1436 
 1437 /*
 1438  * Check permissions of file to execute.
 1439  *      Called with imgp->vp locked.
 1440  *      Return 0 for success or error code on failure.
 1441  */
 1442 int
 1443 exec_check_permissions(imgp)
 1444         struct image_params *imgp;
 1445 {
 1446         struct vnode *vp = imgp->vp;
 1447         struct vattr *attr = imgp->attr;
 1448         struct thread *td;
 1449         int error, writecount;
 1450 
 1451         td = curthread;
 1452 
 1453         /* Get file attributes */
 1454         error = VOP_GETATTR(vp, attr, td->td_ucred);
 1455         if (error)
 1456                 return (error);
 1457 
 1458 #ifdef MAC
 1459         error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
 1460         if (error)
 1461                 return (error);
 1462 #endif
 1463 
 1464         /*
 1465          * 1) Check if file execution is disabled for the filesystem that
 1466          *    this file resides on.
 1467          * 2) Ensure that at least one execute bit is on. Otherwise, a
 1468          *    privileged user will always succeed, and we don't want this
 1469          *    to happen unless the file really is executable.
 1470          * 3) Ensure that the file is a regular file.
 1471          */
 1472         if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
 1473             (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
 1474             (attr->va_type != VREG))
 1475                 return (EACCES);
 1476 
 1477         /*
 1478          * Zero length files can't be exec'd
 1479          */
 1480         if (attr->va_size == 0)
 1481                 return (ENOEXEC);
 1482 
 1483         /*
 1484          *  Check for execute permission to file based on current credentials.
 1485          */
 1486         error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
 1487         if (error)
 1488                 return (error);
 1489 
 1490         /*
 1491          * Check number of open-for-writes on the file and deny execution
 1492          * if there are any.
 1493          */
 1494         error = VOP_GET_WRITECOUNT(vp, &writecount);
 1495         if (error != 0)
 1496                 return (error);
 1497         if (writecount != 0)
 1498                 return (ETXTBSY);
 1499 
 1500         /*
 1501          * Call filesystem specific open routine (which does nothing in the
 1502          * general case).
 1503          */
 1504         error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
 1505         if (error == 0)
 1506                 imgp->opened = 1;
 1507         return (error);
 1508 }
 1509 
 1510 /*
 1511  * Exec handler registration
 1512  */
 1513 int
 1514 exec_register(execsw_arg)
 1515         const struct execsw *execsw_arg;
 1516 {
 1517         const struct execsw **es, **xs, **newexecsw;
 1518         int count = 2;  /* New slot and trailing NULL */
 1519 
 1520         if (execsw)
 1521                 for (es = execsw; *es; es++)
 1522                         count++;
 1523         newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
 1524         xs = newexecsw;
 1525         if (execsw)
 1526                 for (es = execsw; *es; es++)
 1527                         *xs++ = *es;
 1528         *xs++ = execsw_arg;
 1529         *xs = NULL;
 1530         if (execsw)
 1531                 free(execsw, M_TEMP);
 1532         execsw = newexecsw;
 1533         return (0);
 1534 }
 1535 
 1536 int
 1537 exec_unregister(execsw_arg)
 1538         const struct execsw *execsw_arg;
 1539 {
 1540         const struct execsw **es, **xs, **newexecsw;
 1541         int count = 1;
 1542 
 1543         if (execsw == NULL)
 1544                 panic("unregister with no handlers left?\n");
 1545 
 1546         for (es = execsw; *es; es++) {
 1547                 if (*es == execsw_arg)
 1548                         break;
 1549         }
 1550         if (*es == NULL)
 1551                 return (ENOENT);
 1552         for (es = execsw; *es; es++)
 1553                 if (*es != execsw_arg)
 1554                         count++;
 1555         newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
 1556         xs = newexecsw;
 1557         for (es = execsw; *es; es++)
 1558                 if (*es != execsw_arg)
 1559                         *xs++ = *es;
 1560         *xs = NULL;
 1561         if (execsw)
 1562                 free(execsw, M_TEMP);
 1563         execsw = newexecsw;
 1564         return (0);
 1565 }

Cache object: 1f3cc4ab256c4e99ea5557ec79b12d76


[ 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.