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_hwpmc_hooks.h"
31 #include "opt_ktrace.h"
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/eventhandler.h>
39 #include <sys/mutex.h>
40 #include <sys/sysproto.h>
41 #include <sys/signalvar.h>
42 #include <sys/kernel.h>
43 #include <sys/mount.h>
44 #include <sys/filedesc.h>
45 #include <sys/fcntl.h>
48 #include <sys/imgact.h>
49 #include <sys/imgact_elf.h>
51 #include <sys/malloc.h>
54 #include <sys/pioctl.h>
55 #include <sys/namei.h>
56 #include <sys/resourcevar.h>
57 #include <sys/sf_buf.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/sysent.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
64 #include <sys/ktrace.h>
68 #include <vm/vm_param.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_kern.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_pager.h>
78 #include <sys/pmckern.h>
81 #include <machine/reg.h>
83 #include <security/audit/audit.h>
84 #include <security/mac/mac_framework.h>
86 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
88 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
89 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
90 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
91 static int do_execve(struct thread *td, struct image_args *args,
93 static void exec_free_args(struct image_args *);
95 /* XXX This should be vm_size_t. */
96 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
97 NULL, 0, sysctl_kern_ps_strings, "LU", "");
99 /* XXX This should be vm_size_t. */
100 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
101 NULL, 0, sysctl_kern_usrstack, "LU", "");
103 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
104 NULL, 0, sysctl_kern_stackprot, "I", "");
106 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
107 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
108 &ps_arg_cache_limit, 0, "");
111 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
118 if (req->flags & SCTL_MASK32) {
120 val = (unsigned int)p->p_sysent->sv_psstrings;
121 error = SYSCTL_OUT(req, &val, sizeof(val));
124 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
125 sizeof(p->p_sysent->sv_psstrings));
130 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
137 if (req->flags & SCTL_MASK32) {
139 val = (unsigned int)p->p_sysent->sv_usrstack;
140 error = SYSCTL_OUT(req, &val, sizeof(val));
143 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
144 sizeof(p->p_sysent->sv_usrstack));
149 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
154 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
155 sizeof(p->p_sysent->sv_stackprot)));
159 * Each of the items is a pointer to a `const struct execsw', hence the
160 * double pointer here.
162 static const struct execsw **execsw;
164 #ifndef _SYS_SYSPROTO_H_
175 struct execve_args /* {
182 struct image_args args;
184 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
185 uap->argv, uap->envv);
187 error = kern_execve(td, &args, NULL);
191 #ifndef _SYS_SYSPROTO_H_
192 struct __mac_execve_args {
201 __mac_execve(td, uap)
203 struct __mac_execve_args /* {
212 struct image_args args;
214 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
215 uap->argv, uap->envv);
217 error = kern_execve(td, &args, uap->mac_p);
225 * XXX: kern_execve has the astonishing property of not always returning to
226 * the caller. If sufficiently bad things happen during the call to
227 * do_execve(), it can end up calling exit1(); as a result, callers must
228 * avoid doing anything which they might need to undo (e.g., allocating
232 kern_execve(td, args, mac_p)
234 struct image_args *args;
237 struct proc *p = td->td_proc;
240 AUDIT_ARG(argv, args->begin_argv, args->argc,
241 args->begin_envv - args->begin_argv);
242 AUDIT_ARG(envv, args->begin_envv, args->envc,
243 args->endp - args->begin_envv);
244 if (p->p_flag & P_HADTHREADS) {
246 if (thread_single(SINGLE_BOUNDARY)) {
248 exec_free_args(args);
249 return (ERESTART); /* Try again later. */
254 error = do_execve(td, args, mac_p);
256 if (p->p_flag & P_HADTHREADS) {
259 * If success, we upgrade to SINGLE_EXIT state to
260 * force other threads to suicide.
263 thread_single(SINGLE_EXIT);
273 * In-kernel implementation of execve(). All arguments are assumed to be
274 * userspace pointers from the passed thread.
277 do_execve(td, args, mac_p)
279 struct image_args *args;
282 struct proc *p = td->td_proc;
283 struct nameidata nd, *ndp;
284 struct ucred *newcred = NULL, *oldcred;
285 struct uidinfo *euip;
286 register_t *stack_base;
288 struct image_params image_params, *imgp;
290 int (*img_first)(struct image_params *);
291 struct pargs *oldargs = NULL, *newargs = NULL;
292 struct sigacts *oldsigacts, *newsigacts;
294 struct vnode *tracevp = NULL;
295 struct ucred *tracecred = NULL;
297 struct vnode *textvp = NULL;
298 int credential_changing;
302 struct label *interplabel = NULL;
306 struct pmckern_procexec pe;
310 imgp = &image_params;
313 * Lock the process and set the P_INEXEC flag to indicate that
314 * it should be left alone until we're done here. This is
315 * necessary to avoid race conditions - e.g. in ptrace() -
316 * that might allow a local user to illicitly obtain elevated
320 KASSERT((p->p_flag & P_INEXEC) == 0,
321 ("%s(): process already has P_INEXEC flag", __func__));
322 p->p_flag |= P_INEXEC;
326 * Initialize part of the common data
329 imgp->execlabel = NULL;
331 imgp->entry_addr = 0;
332 imgp->vmspace_destroyed = 0;
333 imgp->interpreted = 0;
334 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
335 imgp->auxargs = NULL;
338 imgp->firstpage = NULL;
339 imgp->ps_strings = 0;
340 imgp->auxarg_size = 0;
344 error = mac_execve_enter(imgp, mac_p);
349 imgp->image_header = NULL;
352 * Translate the file name. namei() returns a vnode pointer
353 * in ni_vp amoung other things.
355 * XXXAUDIT: It would be desirable to also audit the name of the
356 * interpreter if this is an interpreted binary.
359 NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | MPSAFE |
360 AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
367 vfslocked = NDHASGIANT(ndp);
368 imgp->vp = ndp->ni_vp;
371 * Check file permissions (also 'opens' file)
373 error = exec_check_permissions(imgp);
375 goto exec_fail_dealloc;
377 imgp->object = imgp->vp->v_object;
378 if (imgp->object != NULL)
379 vm_object_reference(imgp->object);
382 * Set VV_TEXT now so no one can write to the executable while we're
385 * Remember if this was set before and unset it in case this is not
386 * actually an executable image.
388 textset = imgp->vp->v_vflag & VV_TEXT;
389 imgp->vp->v_vflag |= VV_TEXT;
391 error = exec_map_first_page(imgp);
393 goto exec_fail_dealloc;
395 imgp->proc->p_osrel = 0;
397 * If the current process has a special image activator it
398 * wants to try first, call it. For example, emulating shell
399 * scripts differently.
402 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
403 error = img_first(imgp);
406 * Loop through the list of image activators, calling each one.
407 * An activator returns -1 if there is no match, 0 on success,
408 * and an error otherwise.
410 for (i = 0; error == -1 && execsw[i]; ++i) {
411 if (execsw[i]->ex_imgact == NULL ||
412 execsw[i]->ex_imgact == img_first) {
415 error = (*execsw[i]->ex_imgact)(imgp);
421 imgp->vp->v_vflag &= ~VV_TEXT;
424 goto exec_fail_dealloc;
428 * Special interpreter operation, cleanup and loop up to try to
429 * activate the interpreter.
431 if (imgp->interpreted) {
432 exec_unmap_first_page(imgp);
434 * VV_TEXT needs to be unset for scripts. There is a short
435 * period before we determine that something is a script where
436 * VV_TEXT will be set. The vnode lock is held over this
437 * entire period so nothing should illegitimately be blocked.
439 imgp->vp->v_vflag &= ~VV_TEXT;
440 /* free name buffer and old vnode */
441 NDFREE(ndp, NDF_ONLY_PNBUF);
443 interplabel = mac_vnode_label_alloc();
444 mac_vnode_copy_label(ndp->ni_vp->v_label, interplabel);
447 vm_object_deallocate(imgp->object);
449 VFS_UNLOCK_GIANT(vfslocked);
451 /* set new name to that of the interpreter */
452 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
453 UIO_SYSSPACE, imgp->interpreter_name, td);
458 * Copy out strings (args and env) and initialize stack base
460 if (p->p_sysent->sv_copyout_strings)
461 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
463 stack_base = exec_copyout_strings(imgp);
466 * If custom stack fixup routine present for this process
467 * let it do the stack setup.
468 * Else stuff argument count as first item on stack
470 if (p->p_sysent->sv_fixup != NULL)
471 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
473 suword(--stack_base, imgp->args->argc);
476 * For security and other reasons, the file descriptor table cannot
477 * be shared after an exec.
482 * Malloc things before we need locks.
485 euip = uifind(attr.va_uid);
486 i = imgp->args->begin_envv - imgp->args->begin_argv;
487 /* Cache arguments if they fit inside our allowance */
488 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
489 newargs = pargs_alloc(i);
490 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
493 /* close files on exec */
494 VOP_UNLOCK(imgp->vp, 0);
496 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
498 /* Get a reference to the vnode prior to locking the proc */
502 * For security and other reasons, signal handlers cannot
503 * be shared after an exec. The new process gets a copy of the old
504 * handlers. In execsigs(), the new process will have its signals
508 if (sigacts_shared(p->p_sigacts)) {
509 oldsigacts = p->p_sigacts;
511 newsigacts = sigacts_alloc();
512 sigacts_copy(newsigacts, oldsigacts);
514 p->p_sigacts = newsigacts;
521 /* reset caught signals */
524 /* name this process - nameiexec(p, ndp) */
525 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
526 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
528 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
531 * mark as execed, wakeup the process that vforked (if any) and tell
532 * it that it now has its own resources back
535 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
536 p->p_flag &= ~P_PPWAIT;
541 * Implement image setuid/setgid.
543 * Don't honor setuid/setgid if the filesystem prohibits it or if
544 * the process is being traced.
546 * XXXMAC: For the time being, use NOSUID to also prohibit
547 * transitions on the file system.
549 oldcred = p->p_ucred;
550 credential_changing = 0;
551 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
553 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
556 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
558 credential_changing |= will_transition;
561 if (credential_changing &&
562 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
563 (p->p_flag & P_TRACED) == 0) {
565 * Turn off syscall tracing for set-id programs, except for
566 * root. Record any set-id flags first to make sure that
567 * we do not regain any tracing during a possible block.
572 if (p->p_tracevp != NULL &&
573 priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) {
574 mtx_lock(&ktrace_mtx);
576 tracevp = p->p_tracevp;
578 tracecred = p->p_tracecred;
579 p->p_tracecred = NULL;
580 mtx_unlock(&ktrace_mtx);
584 * Close any file descriptors 0..2 that reference procfs,
585 * then make sure file descriptors 0..2 are in use.
587 * setugidsafety() may call closef() and then pfind()
588 * which may grab the process lock.
589 * fdcheckstd() may call falloc() which may block to
590 * allocate memory, so temporarily drop the process lock.
594 VOP_UNLOCK(imgp->vp, 0);
595 error = fdcheckstd(td);
596 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
601 * Set the new credentials.
603 crcopy(newcred, oldcred);
604 if (attr.va_mode & VSUID)
605 change_euid(newcred, euip);
606 if (attr.va_mode & VSGID)
607 change_egid(newcred, attr.va_gid);
609 if (will_transition) {
610 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
615 * Implement correct POSIX saved-id behavior.
617 * XXXMAC: Note that the current logic will save the
618 * uid and gid if a MAC domain transition occurs, even
619 * though maybe it shouldn't.
621 change_svuid(newcred, newcred->cr_uid);
622 change_svgid(newcred, newcred->cr_gid);
623 p->p_ucred = newcred;
626 if (oldcred->cr_uid == oldcred->cr_ruid &&
627 oldcred->cr_gid == oldcred->cr_rgid)
628 p->p_flag &= ~P_SUGID;
630 * Implement correct POSIX saved-id behavior.
632 * XXX: It's not clear that the existing behavior is
633 * POSIX-compliant. A number of sources indicate that the
634 * saved uid/gid should only be updated if the new ruid is
635 * not equal to the old ruid, or the new euid is not equal
636 * to the old euid and the new euid is not equal to the old
637 * ruid. The FreeBSD code always updates the saved uid/gid.
638 * Also, this code uses the new (replaced) euid and egid as
639 * the source, which may or may not be the right ones to use.
641 if (oldcred->cr_svuid != oldcred->cr_uid ||
642 oldcred->cr_svgid != oldcred->cr_gid) {
643 crcopy(newcred, oldcred);
644 change_svuid(newcred, newcred->cr_uid);
645 change_svgid(newcred, newcred->cr_gid);
646 p->p_ucred = newcred;
652 * Store the vp for use in procfs. This vnode was referenced prior
653 * to locking the proc lock.
655 textvp = p->p_textvp;
656 p->p_textvp = ndp->ni_vp;
659 * Notify others that we exec'd, and clear the P_INEXEC flag
660 * as we're now a bona fide freshly-execed process.
662 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
663 p->p_flag &= ~P_INEXEC;
666 * If tracing the process, trap to debugger so breakpoints
667 * can be set before the program executes.
668 * Use tdsignal to deliver signal to current thread, use
669 * psignal may cause the signal to be delivered to wrong thread
670 * because that thread will exit, remember we are going to enter
671 * single thread mode.
673 if (p->p_flag & P_TRACED)
674 tdsignal(p, td, SIGTRAP, NULL);
676 /* clear "fork but no exec" flag, as we _are_ execing */
677 p->p_acflag &= ~AFORK;
680 * Free any previous argument cache and replace it with
681 * the new argument cache, if any.
689 * Check if system-wide sampling is in effect or if the
690 * current process is using PMCs. If so, do exec() time
691 * processing. This processing needs to happen AFTER the
692 * P_INEXEC flag is cleared.
694 * The proc lock needs to be released before taking the PMC
697 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
699 pe.pm_credentialschanged = credential_changing;
700 pe.pm_entryaddr = imgp->entry_addr;
702 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
705 #else /* !HWPMC_HOOKS */
709 /* Set values passed into the program in registers. */
710 if (p->p_sysent->sv_setregs)
711 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
712 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
714 exec_setregs(td, imgp->entry_addr,
715 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
717 vfs_mark_atime(imgp->vp, td);
721 * Free any resources malloc'd earlier that we didn't use.
728 VOP_UNLOCK(imgp->vp, 0);
730 * Handle deferred decrement of ref counts.
732 if (textvp != NULL) {
735 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
737 VFS_UNLOCK_GIANT(tvfslocked);
739 if (ndp->ni_vp && error != 0)
742 if (tracevp != NULL) {
745 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
747 VFS_UNLOCK_GIANT(tvfslocked);
749 if (tracecred != NULL)
752 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
757 if (oldsigacts != NULL)
758 sigacts_free(oldsigacts);
763 * free various allocated resources
765 if (imgp->firstpage != NULL)
766 exec_unmap_first_page(imgp);
768 if (imgp->vp != NULL) {
769 NDFREE(ndp, NDF_ONLY_PNBUF);
773 if (imgp->object != NULL)
774 vm_object_deallocate(imgp->object);
778 * Stop the process here if its stop event mask has
779 * the S_EXEC bit set.
781 STOPEVENT(p, S_EXEC, 0);
786 /* we're done here, clear P_INEXEC */
788 p->p_flag &= ~P_INEXEC;
793 mac_execve_exit(imgp);
794 if (interplabel != NULL)
795 mac_vnode_label_free(interplabel);
797 VFS_UNLOCK_GIANT(vfslocked);
798 exec_free_args(args);
800 if (error && imgp->vmspace_destroyed) {
801 /* sorry, no more process anymore. exit gracefully */
802 exit1(td, W_EXITCODE(0, SIGABRT));
809 exec_map_first_page(imgp)
810 struct image_params *imgp;
814 vm_page_t ma[VM_INITIAL_PAGEIN];
817 if (imgp->firstpage != NULL)
818 exec_unmap_first_page(imgp);
820 object = imgp->vp->v_object;
823 VM_OBJECT_LOCK(object);
824 #if VM_NRESERVLEVEL > 0
825 if ((object->flags & OBJ_COLORED) == 0) {
826 object->flags |= OBJ_COLORED;
827 object->pg_color = 0;
830 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
831 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
832 initial_pagein = VM_INITIAL_PAGEIN;
833 if (initial_pagein > object->size)
834 initial_pagein = object->size;
835 for (i = 1; i < initial_pagein; i++) {
836 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
839 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
843 ma[i] = vm_page_alloc(object, i,
844 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
850 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
851 ma[0] = vm_page_lookup(object, 0);
852 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
853 (ma[0]->valid == 0)) {
855 vm_page_lock_queues();
857 vm_page_unlock_queues();
859 VM_OBJECT_UNLOCK(object);
863 vm_page_lock_queues();
865 vm_page_unlock_queues();
866 vm_page_wakeup(ma[0]);
867 VM_OBJECT_UNLOCK(object);
869 imgp->firstpage = sf_buf_alloc(ma[0], 0);
870 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
876 exec_unmap_first_page(imgp)
877 struct image_params *imgp;
881 if (imgp->firstpage != NULL) {
882 m = sf_buf_page(imgp->firstpage);
883 sf_buf_free(imgp->firstpage);
884 imgp->firstpage = NULL;
885 vm_page_lock_queues();
887 vm_page_unlock_queues();
892 * Destroy old address space, and allocate a new stack
893 * The new stack is only SGROWSIZ large because it is grown
894 * automatically in trap.c.
897 exec_new_vmspace(imgp, sv)
898 struct image_params *imgp;
899 struct sysentvec *sv;
902 struct proc *p = imgp->proc;
903 struct vmspace *vmspace = p->p_vmspace;
904 vm_offset_t stack_addr;
908 imgp->vmspace_destroyed = 1;
911 /* May be called with Giant held */
912 EVENTHANDLER_INVOKE(process_exec, p, imgp);
915 * Blow away entire process VM, if address space not shared,
916 * otherwise, create a new VM space so that other threads are
919 map = &vmspace->vm_map;
920 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
921 vm_map_max(map) == sv->sv_maxuser) {
923 pmap_remove_pages(vmspace_pmap(vmspace));
924 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
926 error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
929 vmspace = p->p_vmspace;
930 map = &vmspace->vm_map;
933 /* Allocate a new stack */
934 if (sv->sv_maxssiz != NULL)
935 ssiz = *sv->sv_maxssiz;
938 stack_addr = sv->sv_usrstack - ssiz;
939 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
940 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
945 /* Allocate a new register stack */
946 stack_addr = IA64_BACKINGSTORE;
947 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
948 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
953 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
954 * VM_STACK case, but they are still used to monitor the size of the
955 * process stack so we can check the stack rlimit.
957 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
958 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
964 * Copy out argument and environment strings from the old process address
965 * space into the temporary string buffer.
968 exec_copyin_args(struct image_args *args, char *fname,
969 enum uio_seg segflg, char **argv, char **envv)
977 bzero(args, sizeof(*args));
981 * Allocate temporary demand zeroed space for argument and
982 * environment strings:
984 * o ARG_MAX for argument and environment;
985 * o MAXSHELLCMDLEN for the name of interpreters.
987 args->buf = (char *) kmem_alloc_wait(exec_map,
988 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
989 if (args->buf == NULL)
991 args->begin_argv = args->buf;
992 args->endp = args->begin_argv;
993 args->stringspace = ARG_MAX;
995 args->fname = args->buf + ARG_MAX;
998 * Copy the file name.
1000 error = (segflg == UIO_SYSSPACE) ?
1001 copystr(fname, args->fname, PATH_MAX, &length) :
1002 copyinstr(fname, args->fname, PATH_MAX, &length);
1007 * extract arguments first
1009 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1010 if (argp == (caddr_t) -1) {
1014 if ((error = copyinstr(argp, args->endp,
1015 args->stringspace, &length))) {
1016 if (error == ENAMETOOLONG)
1020 args->stringspace -= length;
1021 args->endp += length;
1025 args->begin_envv = args->endp;
1028 * extract environment strings
1031 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1032 if (envp == (caddr_t)-1) {
1036 if ((error = copyinstr(envp, args->endp,
1037 args->stringspace, &length))) {
1038 if (error == ENAMETOOLONG)
1042 args->stringspace -= length;
1043 args->endp += length;
1051 exec_free_args(args);
1056 exec_free_args(struct image_args *args)
1060 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1061 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1067 * Copy strings out to the new process address space, constructing new arg
1068 * and env vector tables. Return a pointer to the base so that it can be used
1069 * as the initial stack pointer.
1072 exec_copyout_strings(imgp)
1073 struct image_params *imgp;
1077 char *stringp, *destp;
1078 register_t *stack_base;
1079 struct ps_strings *arginfo;
1084 * Calculate string base and vector table pointers.
1085 * Also deal with signal trampoline code for this exec type.
1089 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1090 if (p->p_sysent->sv_szsigcode != NULL)
1091 szsigcode = *(p->p_sysent->sv_szsigcode);
1092 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1093 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1099 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1100 szsigcode), szsigcode);
1103 * If we have a valid auxargs ptr, prepare some room
1106 if (imgp->auxargs) {
1108 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1109 * lower compatibility.
1111 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1114 * The '+ 2' is for the null pointers at the end of each of
1115 * the arg and env vector sets,and imgp->auxarg_size is room
1116 * for argument of Runtime loader.
1118 vectp = (char **)(destp - (imgp->args->argc +
1119 imgp->args->envc + 2 + imgp->auxarg_size) *
1124 * The '+ 2' is for the null pointers at the end of each of
1125 * the arg and env vector sets
1127 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1132 * vectp also becomes our initial stack base
1134 stack_base = (register_t *)vectp;
1136 stringp = imgp->args->begin_argv;
1137 argc = imgp->args->argc;
1138 envc = imgp->args->envc;
1141 * Copy out strings - arguments and environment.
1143 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1146 * Fill in "ps_strings" struct for ps, w, etc.
1148 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1149 suword(&arginfo->ps_nargvstr, argc);
1152 * Fill in argument portion of vector table.
1154 for (; argc > 0; --argc) {
1155 suword(vectp++, (long)(intptr_t)destp);
1156 while (*stringp++ != 0)
1161 /* a null vector table pointer separates the argp's from the envp's */
1164 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1165 suword(&arginfo->ps_nenvstr, envc);
1168 * Fill in environment portion of vector table.
1170 for (; envc > 0; --envc) {
1171 suword(vectp++, (long)(intptr_t)destp);
1172 while (*stringp++ != 0)
1177 /* end of vector table is a null pointer */
1180 return (stack_base);
1184 * Check permissions of file to execute.
1185 * Called with imgp->vp locked.
1186 * Return 0 for success or error code on failure.
1189 exec_check_permissions(imgp)
1190 struct image_params *imgp;
1192 struct vnode *vp = imgp->vp;
1193 struct vattr *attr = imgp->attr;
1199 /* Get file attributes */
1200 error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1205 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1211 * 1) Check if file execution is disabled for the filesystem that this
1213 * 2) Insure that at least one execute bit is on - otherwise root
1214 * will always succeed, and we don't want to happen unless the
1215 * file really is executable.
1216 * 3) Insure that the file is a regular file.
1218 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1219 ((attr->va_mode & 0111) == 0) ||
1220 (attr->va_type != VREG))
1224 * Zero length files can't be exec'd
1226 if (attr->va_size == 0)
1230 * Check for execute permission to file based on current credentials.
1232 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1237 * Check number of open-for-writes on the file and deny execution
1240 if (vp->v_writecount)
1244 * Call filesystem specific open routine (which does nothing in the
1247 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1252 * Exec handler registration
1255 exec_register(execsw_arg)
1256 const struct execsw *execsw_arg;
1258 const struct execsw **es, **xs, **newexecsw;
1259 int count = 2; /* New slot and trailing NULL */
1262 for (es = execsw; *es; es++)
1264 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1265 if (newexecsw == NULL)
1269 for (es = execsw; *es; es++)
1274 free(execsw, M_TEMP);
1280 exec_unregister(execsw_arg)
1281 const struct execsw *execsw_arg;
1283 const struct execsw **es, **xs, **newexecsw;
1287 panic("unregister with no handlers left?\n");
1289 for (es = execsw; *es; es++) {
1290 if (*es == execsw_arg)
1295 for (es = execsw; *es; es++)
1296 if (*es != execsw_arg)
1298 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1299 if (newexecsw == NULL)
1302 for (es = execsw; *es; es++)
1303 if (*es != execsw_arg)
1307 free(execsw, M_TEMP);