2 * Copyright (c) 1993, David Greenman
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kdtrace.h"
33 #include "opt_ktrace.h"
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>
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>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
54 #include <sys/malloc.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>
64 #include <sys/sf_buf.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/sysent.h>
68 #include <sys/sysctl.h>
69 #include <sys/vnode.h>
72 #include <sys/ktrace.h>
76 #include <vm/vm_param.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>
86 #include <sys/pmckern.h>
89 #include <machine/reg.h>
91 #include <security/audit/audit.h>
92 #include <security/mac/mac_framework.h>
95 #include <sys/dtrace_bsd.h>
96 dtrace_execexit_func_t dtrace_fasttrap_exec;
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 *");
104 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
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");
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");
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,
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", "");
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", "");
130 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
131 NULL, 0, sysctl_kern_stackprot, "I", "");
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, "");
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");
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.");
148 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
155 if (req->flags & SCTL_MASK32) {
157 val = (unsigned int)p->p_sysent->sv_psstrings;
158 error = SYSCTL_OUT(req, &val, sizeof(val));
161 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
162 sizeof(p->p_sysent->sv_psstrings));
167 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
174 if (req->flags & SCTL_MASK32) {
176 val = (unsigned int)p->p_sysent->sv_usrstack;
177 error = SYSCTL_OUT(req, &val, sizeof(val));
180 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
181 sizeof(p->p_sysent->sv_usrstack));
186 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
191 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
192 sizeof(p->p_sysent->sv_stackprot)));
196 * Each of the items is a pointer to a `const struct execsw', hence the
197 * double pointer here.
199 static const struct execsw **execsw;
201 #ifndef _SYS_SYSPROTO_H_
210 sys_execve(struct thread *td, struct execve_args *uap)
212 struct image_args args;
213 struct vmspace *oldvmspace;
216 error = pre_execve(td, &oldvmspace);
219 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
220 uap->argv, uap->envv);
222 error = kern_execve(td, &args, NULL);
223 post_execve(td, error, oldvmspace);
227 #ifndef _SYS_SYSPROTO_H_
228 struct fexecve_args {
235 sys_fexecve(struct thread *td, struct fexecve_args *uap)
237 struct image_args args;
238 struct vmspace *oldvmspace;
241 error = pre_execve(td, &oldvmspace);
244 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
245 uap->argv, uap->envv);
248 error = kern_execve(td, &args, NULL);
250 post_execve(td, error, oldvmspace);
254 #ifndef _SYS_SYSPROTO_H_
255 struct __mac_execve_args {
264 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
267 struct image_args args;
268 struct vmspace *oldvmspace;
271 error = pre_execve(td, &oldvmspace);
274 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
275 uap->argv, uap->envv);
277 error = kern_execve(td, &args, uap->mac_p);
278 post_execve(td, error, oldvmspace);
286 pre_execve(struct thread *td, struct vmspace **oldvmspace)
291 KASSERT(td == curthread, ("non-current thread %p", td));
294 if ((p->p_flag & P_HADTHREADS) != 0) {
296 if (thread_single(p, SINGLE_BOUNDARY) != 0)
300 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
302 *oldvmspace = p->p_vmspace;
307 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
311 KASSERT(td == curthread, ("non-current thread %p", td));
313 if ((p->p_flag & P_HADTHREADS) != 0) {
316 * If success, we upgrade to SINGLE_EXIT state to
317 * force other threads to suicide.
320 thread_single(p, SINGLE_EXIT);
322 thread_single_end(p, SINGLE_BOUNDARY);
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;
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
341 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
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));
352 * In-kernel implementation of execve(). All arguments are assumed to be
353 * userspace pointers from the passed thread.
356 do_execve(td, args, mac_p)
358 struct image_args *args;
361 struct proc *p = td->td_proc;
363 struct ucred *oldcred;
364 struct uidinfo *euip = NULL;
365 register_t *stack_base;
367 struct image_params image_params, *imgp;
369 int (*img_first)(struct image_params *);
370 struct pargs *oldargs = NULL, *newargs = NULL;
371 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
373 struct vnode *tracevp = NULL;
374 struct ucred *tracecred = NULL;
376 struct vnode *oldtextvp = NULL, *newtextvp;
378 int credential_changing;
381 struct label *interpvplabel = NULL;
385 struct pmckern_procexec pe;
387 static const char fexecv_proc_title[] = "(fexecv)";
389 imgp = &image_params;
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
399 KASSERT((p->p_flag & P_INEXEC) == 0,
400 ("%s(): process already has P_INEXEC flag", __func__));
401 p->p_flag |= P_INEXEC;
405 * Initialize part of the common data
407 bzero(imgp, sizeof(*imgp));
411 oldcred = p->p_ucred;
414 error = mac_execve_enter(imgp, mac_p);
420 * Translate the file name. namei() returns a vnode pointer
421 * in ni_vp among other things.
423 * XXXAUDIT: It would be desirable to also audit the name of the
424 * interpreter if this is an interpreted binary.
426 if (args->fname != NULL) {
427 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
428 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
431 SDT_PROBE1(proc, , , exec, args->fname);
434 if (args->fname != NULL) {
435 #ifdef CAPABILITY_MODE
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.
442 if (IN_CAPABILITY_MODE(td)) {
451 newtextvp = nd.ni_vp;
452 imgp->vp = newtextvp;
454 AUDIT_ARG_FD(args->fd);
456 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
458 error = fgetvp_exec(td, args->fd,
459 cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
462 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
463 AUDIT_ARG_VNODE1(newtextvp);
464 imgp->vp = newtextvp;
468 * Check file permissions (also 'opens' file)
470 error = exec_check_permissions(imgp);
472 goto exec_fail_dealloc;
474 imgp->object = imgp->vp->v_object;
475 if (imgp->object != NULL)
476 vm_object_reference(imgp->object);
479 * Set VV_TEXT now so no one can write to the executable while we're
482 * Remember if this was set before and unset it in case this is not
483 * actually an executable image.
485 textset = VOP_IS_TEXT(imgp->vp);
486 VOP_SET_TEXT(imgp->vp);
488 error = exec_map_first_page(imgp);
490 goto exec_fail_dealloc;
492 imgp->proc->p_osrel = 0;
495 * Implement image setuid/setgid.
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.
501 * Don't honor setuid/setgid if the filesystem prohibits it or if
502 * the process is being traced.
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.
511 * XXXMAC: For the time being, use NOSUID to also prohibit
512 * transitions on the file system.
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;
520 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
521 interpvplabel, imgp);
522 credential_changing |= will_transition;
525 if (credential_changing &&
526 #ifdef CAPABILITY_MODE
527 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
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);
538 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
539 if (attr.va_mode & S_ISGID)
540 change_egid(imgp->newcred, attr.va_gid);
542 * Implement correct POSIX saved-id behavior.
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.
548 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
549 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
552 * Implement correct POSIX saved-id behavior.
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.
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);
572 /* The new credentials are installed into the process later. */
575 * Do the best to calculate the full path to the image file.
577 if (args->fname != NULL && args->fname[0] == '/')
578 imgp->execpath = args->fname;
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);
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.
593 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
594 error = img_first(imgp);
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.
601 for (i = 0; error == -1 && execsw[i]; ++i) {
602 if (execsw[i]->ex_imgact == NULL ||
603 execsw[i]->ex_imgact == img_first) {
606 error = (*execsw[i]->ex_imgact)(imgp);
612 VOP_UNSET_TEXT(imgp->vp);
615 goto exec_fail_dealloc;
619 * Special interpreter operation, cleanup and loop up to try to
620 * activate the interpreter.
622 if (imgp->interpreted) {
623 exec_unmap_first_page(imgp);
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.
630 VOP_UNSET_TEXT(imgp->vp);
631 /* free name buffer and old vnode */
632 if (args->fname != NULL)
633 NDFREE(&nd, NDF_ONLY_PNBUF);
635 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
638 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
642 vm_object_deallocate(imgp->object);
644 imgp->credential_setid = false;
645 if (imgp->newcred != NULL) {
646 crfree(imgp->newcred);
647 imgp->newcred = NULL;
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;
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.
663 VOP_UNLOCK(imgp->vp, 0);
665 if (disallow_high_osrel &&
666 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
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;
675 * Copy out strings (args and env) and initialize stack base
677 if (p->p_sysent->sv_copyout_strings)
678 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
680 stack_base = exec_copyout_strings(imgp);
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
687 if (p->p_sysent->sv_fixup != NULL)
688 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
690 suword(--stack_base, imgp->args->argc);
693 * For security and other reasons, the file descriptor table cannot
694 * be shared after an exec.
697 /* close files on exec */
701 * Malloc things before we need locks.
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);
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
716 if (sigacts_shared(p->p_sigacts)) {
717 oldsigacts = p->p_sigacts;
718 newsigacts = sigacts_alloc();
719 sigacts_copy(newsigacts, oldsigacts);
722 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
726 p->p_sigacts = newsigacts;
730 /* reset caught signals */
733 /* name this process - nameiexec(p, ndp) */
734 bzero(p->p_comm, sizeof(p->p_comm));
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));
742 sched_clear_tdname(td);
746 * mark as execed, wakeup the process that vforked (if any) and tell
747 * it that it now has its own resources back
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 */
760 * Implement image setuid/setgid installation.
762 if (imgp->credential_setid) {
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.
771 if (p->p_tracecred != NULL &&
772 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
773 ktrprocexec(p, &tracecred, &tracevp);
776 * Close any file descriptors 0..2 that reference procfs,
777 * then make sure file descriptors 0..2 are in use.
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.
785 VOP_UNLOCK(imgp->vp, 0);
787 error = fdcheckstd(td);
788 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
790 goto exec_fail_dealloc;
793 if (will_transition) {
794 mac_vnode_execve_transition(oldcred, imgp->newcred,
795 imgp->vp, interpvplabel, imgp);
799 if (oldcred->cr_uid == oldcred->cr_ruid &&
800 oldcred->cr_gid == oldcred->cr_rgid)
801 p->p_flag &= ~P_SUGID;
804 * Set the new credentials.
806 if (imgp->newcred != NULL) {
807 proc_set_cred(p, imgp->newcred);
813 * Store the vp for use in procfs. This vnode was referenced by namei
816 oldtextvp = p->p_textvp;
817 p->p_textvp = newtextvp;
821 * Tell the DTrace fasttrap provider about the exec if it
822 * has declared an interest.
824 if (dtrace_fasttrap_exec)
825 dtrace_fasttrap_exec(p);
829 * Notify others that we exec'd, and clear the P_INEXEC flag
830 * as we're now a bona fide freshly-execed process.
832 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
833 p->p_flag &= ~P_INEXEC;
835 /* clear "fork but no exec" flag, as we _are_ execing */
836 p->p_acflag &= ~AFORK;
839 * Free any previous argument cache and replace it with
840 * the new argument cache, if any.
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.
853 * The proc lock needs to be released before taking the PMC
856 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
858 VOP_UNLOCK(imgp->vp, 0);
859 pe.pm_credentialschanged = credential_changing;
860 pe.pm_entryaddr = imgp->entry_addr;
862 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
863 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
866 #else /* !HWPMC_HOOKS */
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);
875 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
877 vfs_mark_atime(imgp->vp, td->td_ucred);
879 SDT_PROBE1(proc, , , exec__success, args->fname);
882 if (imgp->firstpage != NULL)
883 exec_unmap_first_page(imgp);
885 if (imgp->vp != NULL) {
887 NDFREE(&nd, NDF_ONLY_PNBUF);
889 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
893 VOP_UNLOCK(imgp->vp, 0);
896 if (imgp->object != NULL)
897 vm_object_deallocate(imgp->object);
899 free(imgp->freepath, M_TEMP);
903 if (p->p_ptevents & PTRACE_EXEC)
904 td->td_dbgflags |= TDB_EXEC;
908 * Stop the process here if its stop event mask has
909 * the S_EXEC bit set.
911 STOPEVENT(p, S_EXEC, 0);
914 /* we're done here, clear P_INEXEC */
916 p->p_flag &= ~P_INEXEC;
919 SDT_PROBE1(proc, , , exec__failure, error);
922 if (imgp->newcred != NULL && oldcred != NULL)
923 crfree(imgp->newcred);
926 mac_execve_exit(imgp);
927 mac_execve_interpreter_exit(interpvplabel);
929 exec_free_args(args);
932 * Handle deferred decrement of ref counts.
934 if (oldtextvp != NULL)
939 if (tracecred != NULL)
944 if (oldsigacts != NULL)
945 sigacts_free(oldsigacts);
949 if (error && imgp->vmspace_destroyed) {
950 /* sorry, no more process anymore. exit gracefully */
951 exit1(td, W_EXITCODE(0, SIGABRT));
964 exec_map_first_page(imgp)
965 struct image_params *imgp;
969 vm_page_t ma[VM_INITIAL_PAGEIN];
972 if (imgp->firstpage != NULL)
973 exec_unmap_first_page(imgp);
975 object = imgp->vp->v_object;
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;
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) {
994 if (vm_page_tryxbusy(ma[i]))
997 ma[i] = vm_page_alloc(object, i,
998 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
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]);
1012 VM_OBJECT_WUNLOCK(object);
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);
1023 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1024 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1030 exec_unmap_first_page(imgp)
1031 struct image_params *imgp;
1035 if (imgp->firstpage != NULL) {
1036 m = sf_buf_page(imgp->firstpage);
1037 sf_buf_free(imgp->firstpage);
1038 imgp->firstpage = NULL;
1046 * Destroy old address space, and allocate a new stack
1047 * The new stack is only SGROWSIZ large because it is grown
1048 * automatically in trap.c.
1051 exec_new_vmspace(imgp, sv)
1052 struct image_params *imgp;
1053 struct sysentvec *sv;
1056 struct proc *p = imgp->proc;
1057 struct vmspace *vmspace = p->p_vmspace;
1059 struct rlimit rlim_stack;
1060 vm_offset_t sv_minuser, stack_addr;
1064 imgp->vmspace_destroyed = 1;
1067 /* May be called with Giant held */
1068 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1071 * Blow away entire process VM, if address space not shared,
1072 * otherwise, create a new VM space so that other threads are
1075 map = &vmspace->vm_map;
1077 sv_minuser = sv->sv_minuser;
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) {
1083 pmap_remove_pages(vmspace_pmap(vmspace));
1084 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1086 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1089 vmspace = p->p_vmspace;
1090 map = &vmspace->vm_map;
1093 /* Map a shared page */
1094 obj = sv->sv_shared_page_obj;
1096 vm_object_reference(obj);
1097 error = vm_map_fixed(map, obj, 0,
1098 sv->sv_shared_page_base, sv->sv_shared_page_len,
1099 VM_PROT_READ | VM_PROT_EXECUTE,
1100 VM_PROT_READ | VM_PROT_EXECUTE,
1101 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1103 vm_object_deallocate(obj);
1108 /* Allocate a new stack */
1109 if (imgp->stack_sz != 0) {
1110 ssiz = trunc_page(imgp->stack_sz);
1112 lim_rlimit(p, RLIMIT_STACK, &rlim_stack);
1114 if (ssiz > rlim_stack.rlim_max)
1115 ssiz = rlim_stack.rlim_max;
1116 if (ssiz > rlim_stack.rlim_cur) {
1117 rlim_stack.rlim_cur = ssiz;
1118 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1120 } else if (sv->sv_maxssiz != NULL) {
1121 ssiz = *sv->sv_maxssiz;
1125 stack_addr = sv->sv_usrstack - ssiz;
1126 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1127 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1129 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1134 /* Allocate a new register stack */
1135 error = vm_map_stack(map, IA64_BACKINGSTORE, (vm_size_t)ssiz,
1136 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1141 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1142 * VM_STACK case, but they are still used to monitor the size of the
1143 * process stack so we can check the stack rlimit.
1145 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1146 vmspace->vm_maxsaddr = (char *)stack_addr;
1152 * Copy out argument and environment strings from the old process address
1153 * space into the temporary string buffer.
1156 exec_copyin_args(struct image_args *args, char *fname,
1157 enum uio_seg segflg, char **argv, char **envv)
1163 bzero(args, sizeof(*args));
1168 * Allocate demand-paged memory for the file name, argument, and
1169 * environment strings.
1171 error = exec_alloc_args(args);
1176 * Copy the file name.
1178 if (fname != NULL) {
1179 args->fname = args->buf;
1180 error = (segflg == UIO_SYSSPACE) ?
1181 copystr(fname, args->fname, PATH_MAX, &length) :
1182 copyinstr(fname, args->fname, PATH_MAX, &length);
1188 args->begin_argv = args->buf + length;
1189 args->endp = args->begin_argv;
1190 args->stringspace = ARG_MAX;
1193 * extract arguments first
1196 error = fueword(argv++, &argp);
1203 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1204 args->stringspace, &length);
1206 if (error == ENAMETOOLONG)
1210 args->stringspace -= length;
1211 args->endp += length;
1215 args->begin_envv = args->endp;
1218 * extract environment strings
1222 error = fueword(envv++, &envp);
1229 error = copyinstr((void *)(uintptr_t)envp,
1230 args->endp, args->stringspace, &length);
1232 if (error == ENAMETOOLONG)
1236 args->stringspace -= length;
1237 args->endp += length;
1245 exec_free_args(args);
1250 * Allocate temporary demand-paged, zero-filled memory for the file name,
1251 * argument, and environment strings. Returns zero if the allocation succeeds
1252 * and ENOMEM otherwise.
1255 exec_alloc_args(struct image_args *args)
1258 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1259 return (args->buf != NULL ? 0 : ENOMEM);
1263 exec_free_args(struct image_args *args)
1266 if (args->buf != NULL) {
1267 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1268 PATH_MAX + ARG_MAX);
1271 if (args->fname_buf != NULL) {
1272 free(args->fname_buf, M_TEMP);
1273 args->fname_buf = NULL;
1278 * Copy strings out to the new process address space, constructing new arg
1279 * and env vector tables. Return a pointer to the base so that it can be used
1280 * as the initial stack pointer.
1283 exec_copyout_strings(imgp)
1284 struct image_params *imgp;
1290 register_t *stack_base;
1291 struct ps_strings *arginfo;
1293 size_t execpath_len;
1294 int szsigcode, szps;
1295 char canary[sizeof(long) * 8];
1297 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1299 * Calculate string base and vector table pointers.
1300 * Also deal with signal trampoline code for this exec type.
1302 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1303 execpath_len = strlen(imgp->execpath) + 1;
1308 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1309 if (p->p_sysent->sv_sigcode_base == 0) {
1310 if (p->p_sysent->sv_szsigcode != NULL)
1311 szsigcode = *(p->p_sysent->sv_szsigcode);
1313 destp = (uintptr_t)arginfo;
1318 if (szsigcode != 0) {
1320 destp = rounddown2(destp, sizeof(void *));
1321 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1325 * Copy the image path for the rtld.
1327 if (execpath_len != 0) {
1328 destp -= execpath_len;
1329 imgp->execpathp = destp;
1330 copyout(imgp->execpath, (void *)destp, execpath_len);
1334 * Prepare the canary for SSP.
1336 arc4rand(canary, sizeof(canary), 0);
1337 destp -= sizeof(canary);
1338 imgp->canary = destp;
1339 copyout(canary, (void *)destp, sizeof(canary));
1340 imgp->canarylen = sizeof(canary);
1343 * Prepare the pagesizes array.
1346 destp = rounddown2(destp, sizeof(void *));
1347 imgp->pagesizes = destp;
1348 copyout(pagesizes, (void *)destp, szps);
1349 imgp->pagesizeslen = szps;
1351 destp -= ARG_MAX - imgp->args->stringspace;
1352 destp = rounddown2(destp, sizeof(void *));
1355 * If we have a valid auxargs ptr, prepare some room
1358 if (imgp->auxargs) {
1360 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1361 * lower compatibility.
1363 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1366 * The '+ 2' is for the null pointers at the end of each of
1367 * the arg and env vector sets,and imgp->auxarg_size is room
1368 * for argument of Runtime loader.
1370 vectp = (char **)(destp - (imgp->args->argc +
1371 imgp->args->envc + 2 + imgp->auxarg_size)
1375 * The '+ 2' is for the null pointers at the end of each of
1376 * the arg and env vector sets
1378 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1379 + 2) * sizeof(char *));
1383 * vectp also becomes our initial stack base
1385 stack_base = (register_t *)vectp;
1387 stringp = imgp->args->begin_argv;
1388 argc = imgp->args->argc;
1389 envc = imgp->args->envc;
1392 * Copy out strings - arguments and environment.
1394 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1397 * Fill in "ps_strings" struct for ps, w, etc.
1399 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1400 suword32(&arginfo->ps_nargvstr, argc);
1403 * Fill in argument portion of vector table.
1405 for (; argc > 0; --argc) {
1406 suword(vectp++, (long)(intptr_t)destp);
1407 while (*stringp++ != 0)
1412 /* a null vector table pointer separates the argp's from the envp's */
1415 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1416 suword32(&arginfo->ps_nenvstr, envc);
1419 * Fill in environment portion of vector table.
1421 for (; envc > 0; --envc) {
1422 suword(vectp++, (long)(intptr_t)destp);
1423 while (*stringp++ != 0)
1428 /* end of vector table is a null pointer */
1431 return (stack_base);
1435 * Check permissions of file to execute.
1436 * Called with imgp->vp locked.
1437 * Return 0 for success or error code on failure.
1440 exec_check_permissions(imgp)
1441 struct image_params *imgp;
1443 struct vnode *vp = imgp->vp;
1444 struct vattr *attr = imgp->attr;
1446 int error, writecount;
1450 /* Get file attributes */
1451 error = VOP_GETATTR(vp, attr, td->td_ucred);
1456 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1462 * 1) Check if file execution is disabled for the filesystem that
1463 * this file resides on.
1464 * 2) Ensure that at least one execute bit is on. Otherwise, a
1465 * privileged user will always succeed, and we don't want this
1466 * to happen unless the file really is executable.
1467 * 3) Ensure that the file is a regular file.
1469 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1470 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1471 (attr->va_type != VREG))
1475 * Zero length files can't be exec'd
1477 if (attr->va_size == 0)
1481 * Check for execute permission to file based on current credentials.
1483 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1488 * Check number of open-for-writes on the file and deny execution
1491 error = VOP_GET_WRITECOUNT(vp, &writecount);
1494 if (writecount != 0)
1498 * Call filesystem specific open routine (which does nothing in the
1501 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1508 * Exec handler registration
1511 exec_register(execsw_arg)
1512 const struct execsw *execsw_arg;
1514 const struct execsw **es, **xs, **newexecsw;
1515 int count = 2; /* New slot and trailing NULL */
1518 for (es = execsw; *es; es++)
1520 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1523 for (es = execsw; *es; es++)
1528 free(execsw, M_TEMP);
1534 exec_unregister(execsw_arg)
1535 const struct execsw *execsw_arg;
1537 const struct execsw **es, **xs, **newexecsw;
1541 panic("unregister with no handlers left?\n");
1543 for (es = execsw; *es; es++) {
1544 if (*es == execsw_arg)
1549 for (es = execsw; *es; es++)
1550 if (*es != execsw_arg)
1552 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1554 for (es = execsw; *es; es++)
1555 if (*es != execsw_arg)
1559 free(execsw, M_TEMP);