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_kdtrace.h"
32 #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>
58 #include <sys/sf_buf.h>
59 #include <sys/syscallsubr.h>
60 #include <sys/sysent.h>
62 #include <sys/sysctl.h>
63 #include <sys/vnode.h>
66 #include <sys/ktrace.h>
70 #include <vm/vm_param.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_pager.h>
80 #include <sys/pmckern.h>
83 #include <machine/reg.h>
85 #include <security/audit/audit.h>
86 #include <security/mac/mac_framework.h>
89 #include <sys/dtrace_bsd.h>
90 dtrace_execexit_func_t dtrace_fasttrap_exec;
93 SDT_PROVIDER_DECLARE(proc);
94 SDT_PROBE_DEFINE(proc, kernel, , exec);
95 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
96 SDT_PROBE_DEFINE(proc, kernel, , exec_failure);
97 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
98 SDT_PROBE_DEFINE(proc, kernel, , exec_success);
99 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
103 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
104 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106 static int do_execve(struct thread *td, struct image_args *args,
108 static void exec_free_args(struct image_args *);
110 /* XXX This should be vm_size_t. */
111 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
112 NULL, 0, sysctl_kern_ps_strings, "LU", "");
114 /* XXX This should be vm_size_t. */
115 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
116 NULL, 0, sysctl_kern_usrstack, "LU", "");
118 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
119 NULL, 0, sysctl_kern_stackprot, "I", "");
121 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
122 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
123 &ps_arg_cache_limit, 0, "");
126 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
133 if (req->flags & SCTL_MASK32) {
135 val = (unsigned int)p->p_sysent->sv_psstrings;
136 error = SYSCTL_OUT(req, &val, sizeof(val));
139 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
140 sizeof(p->p_sysent->sv_psstrings));
145 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
152 if (req->flags & SCTL_MASK32) {
154 val = (unsigned int)p->p_sysent->sv_usrstack;
155 error = SYSCTL_OUT(req, &val, sizeof(val));
158 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
159 sizeof(p->p_sysent->sv_usrstack));
164 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
169 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
170 sizeof(p->p_sysent->sv_stackprot)));
174 * Each of the items is a pointer to a `const struct execsw', hence the
175 * double pointer here.
177 static const struct execsw **execsw;
179 #ifndef _SYS_SYSPROTO_H_
190 struct execve_args /* {
197 struct image_args args;
199 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
200 uap->argv, uap->envv);
202 error = kern_execve(td, &args, NULL);
206 #ifndef _SYS_SYSPROTO_H_
207 struct fexecve_args {
214 fexecve(struct thread *td, struct fexecve_args *uap)
217 struct image_args args;
219 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
220 uap->argv, uap->envv);
223 error = kern_execve(td, &args, NULL);
228 #ifndef _SYS_SYSPROTO_H_
229 struct __mac_execve_args {
238 __mac_execve(td, uap)
240 struct __mac_execve_args /* {
249 struct image_args args;
251 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
252 uap->argv, uap->envv);
254 error = kern_execve(td, &args, uap->mac_p);
262 * XXX: kern_execve has the astonishing property of not always returning to
263 * the caller. If sufficiently bad things happen during the call to
264 * do_execve(), it can end up calling exit1(); as a result, callers must
265 * avoid doing anything which they might need to undo (e.g., allocating
269 kern_execve(td, args, mac_p)
271 struct image_args *args;
274 struct proc *p = td->td_proc;
277 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
278 args->begin_envv - args->begin_argv);
279 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
280 args->endp - args->begin_envv);
281 if (p->p_flag & P_HADTHREADS) {
283 if (thread_single(SINGLE_BOUNDARY)) {
285 exec_free_args(args);
286 return (ERESTART); /* Try again later. */
291 error = do_execve(td, args, mac_p);
293 if (p->p_flag & P_HADTHREADS) {
296 * If success, we upgrade to SINGLE_EXIT state to
297 * force other threads to suicide.
300 thread_single(SINGLE_EXIT);
310 * In-kernel implementation of execve(). All arguments are assumed to be
311 * userspace pointers from the passed thread.
314 do_execve(td, args, mac_p)
316 struct image_args *args;
319 struct proc *p = td->td_proc;
321 struct ucred *newcred = NULL, *oldcred;
322 struct uidinfo *euip;
323 register_t *stack_base;
324 int error, len = 0, i;
325 struct image_params image_params, *imgp;
327 int (*img_first)(struct image_params *);
328 struct pargs *oldargs = NULL, *newargs = NULL;
329 struct sigacts *oldsigacts, *newsigacts;
331 struct vnode *tracevp = NULL;
332 struct ucred *tracecred = NULL;
334 struct vnode *textvp = NULL, *binvp = NULL;
335 int credential_changing;
339 struct label *interpvplabel = NULL;
343 struct pmckern_procexec pe;
345 static const char fexecv_proc_title[] = "(fexecv)";
348 imgp = &image_params;
351 * Lock the process and set the P_INEXEC flag to indicate that
352 * it should be left alone until we're done here. This is
353 * necessary to avoid race conditions - e.g. in ptrace() -
354 * that might allow a local user to illicitly obtain elevated
358 KASSERT((p->p_flag & P_INEXEC) == 0,
359 ("%s(): process already has P_INEXEC flag", __func__));
360 p->p_flag |= P_INEXEC;
364 * Initialize part of the common data
367 imgp->execlabel = NULL;
369 imgp->entry_addr = 0;
370 imgp->vmspace_destroyed = 0;
371 imgp->interpreted = 0;
373 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
374 imgp->auxargs = NULL;
377 imgp->firstpage = NULL;
378 imgp->ps_strings = 0;
379 imgp->auxarg_size = 0;
381 imgp->execpath = imgp->freepath = NULL;
385 error = mac_execve_enter(imgp, mac_p);
390 imgp->image_header = NULL;
393 * Translate the file name. namei() returns a vnode pointer
394 * in ni_vp amoung other things.
396 * XXXAUDIT: It would be desirable to also audit the name of the
397 * interpreter if this is an interpreted binary.
399 if (args->fname != NULL) {
400 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
401 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
404 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
407 if (args->fname != NULL) {
412 vfslocked = NDHASGIANT(&nd);
416 AUDIT_ARG_FD(args->fd);
417 error = fgetvp(td, args->fd, &binvp);
420 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
421 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
422 AUDIT_ARG_VNODE1(binvp);
427 * Check file permissions (also 'opens' file)
429 error = exec_check_permissions(imgp);
431 goto exec_fail_dealloc;
433 imgp->object = imgp->vp->v_object;
434 if (imgp->object != NULL)
435 vm_object_reference(imgp->object);
438 * Set VV_TEXT now so no one can write to the executable while we're
441 * Remember if this was set before and unset it in case this is not
442 * actually an executable image.
444 textset = imgp->vp->v_vflag & VV_TEXT;
445 imgp->vp->v_vflag |= VV_TEXT;
447 error = exec_map_first_page(imgp);
449 goto exec_fail_dealloc;
451 imgp->proc->p_osrel = 0;
453 * If the current process has a special image activator it
454 * wants to try first, call it. For example, emulating shell
455 * scripts differently.
458 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
459 error = img_first(imgp);
462 * Loop through the list of image activators, calling each one.
463 * An activator returns -1 if there is no match, 0 on success,
464 * and an error otherwise.
466 for (i = 0; error == -1 && execsw[i]; ++i) {
467 if (execsw[i]->ex_imgact == NULL ||
468 execsw[i]->ex_imgact == img_first) {
471 error = (*execsw[i]->ex_imgact)(imgp);
477 imgp->vp->v_vflag &= ~VV_TEXT;
480 goto exec_fail_dealloc;
484 * Special interpreter operation, cleanup and loop up to try to
485 * activate the interpreter.
487 if (imgp->interpreted) {
488 exec_unmap_first_page(imgp);
490 * VV_TEXT needs to be unset for scripts. There is a short
491 * period before we determine that something is a script where
492 * VV_TEXT will be set. The vnode lock is held over this
493 * entire period so nothing should illegitimately be blocked.
495 imgp->vp->v_vflag &= ~VV_TEXT;
496 /* free name buffer and old vnode */
497 if (args->fname != NULL)
498 NDFREE(&nd, NDF_ONLY_PNBUF);
500 mac_execve_interpreter_enter(binvp, &interpvplabel);
503 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
507 vm_object_deallocate(imgp->object);
509 VFS_UNLOCK_GIANT(vfslocked);
511 /* set new name to that of the interpreter */
512 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
513 UIO_SYSSPACE, imgp->interpreter_name, td);
514 args->fname = imgp->interpreter_name;
519 * NB: We unlock the vnode here because it is believed that none
520 * of the sv_copyout_strings/sv_fixup operations require the vnode.
522 VOP_UNLOCK(imgp->vp, 0);
525 * Do the best to calculate the full path to the image file.
527 if (imgp->auxargs != NULL &&
528 ((args->fname != NULL && args->fname[0] == '/') ||
529 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
530 imgp->execpath = args->fname;
533 * Copy out strings (args and env) and initialize stack base
535 if (p->p_sysent->sv_copyout_strings)
536 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
538 stack_base = exec_copyout_strings(imgp);
541 * If custom stack fixup routine present for this process
542 * let it do the stack setup.
543 * Else stuff argument count as first item on stack
545 if (p->p_sysent->sv_fixup != NULL)
546 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
548 suword(--stack_base, imgp->args->argc);
551 * For security and other reasons, the file descriptor table cannot
552 * be shared after an exec.
557 * Malloc things before we need locks.
560 euip = uifind(attr.va_uid);
561 i = imgp->args->begin_envv - imgp->args->begin_argv;
562 /* Cache arguments if they fit inside our allowance */
563 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
564 newargs = pargs_alloc(i);
565 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
568 /* close files on exec */
570 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
572 /* Get a reference to the vnode prior to locking the proc */
576 * For security and other reasons, signal handlers cannot
577 * be shared after an exec. The new process gets a copy of the old
578 * handlers. In execsigs(), the new process will have its signals
582 oldcred = crcopysafe(p, newcred);
583 if (sigacts_shared(p->p_sigacts)) {
584 oldsigacts = p->p_sigacts;
586 newsigacts = sigacts_alloc();
587 sigacts_copy(newsigacts, oldsigacts);
589 p->p_sigacts = newsigacts;
596 /* reset caught signals */
599 /* name this process - nameiexec(p, ndp) */
601 len = min(nd.ni_cnd.cn_namelen,MAXCOMLEN);
602 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, len);
604 if (vn_commname(binvp, p->p_comm, MAXCOMLEN + 1) == 0)
607 len = sizeof(fexecv_proc_title);
608 bcopy(fexecv_proc_title, p->p_comm, len);
612 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
615 * mark as execed, wakeup the process that vforked (if any) and tell
616 * it that it now has its own resources back
619 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
620 p->p_flag &= ~P_PPWAIT;
621 cv_broadcast(&p->p_pwait);
625 * Implement image setuid/setgid.
627 * Don't honor setuid/setgid if the filesystem prohibits it or if
628 * the process is being traced.
630 * XXXMAC: For the time being, use NOSUID to also prohibit
631 * transitions on the file system.
633 credential_changing = 0;
634 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
636 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
639 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
640 interpvplabel, imgp);
641 credential_changing |= will_transition;
644 if (credential_changing &&
645 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
646 (p->p_flag & P_TRACED) == 0) {
648 * Turn off syscall tracing for set-id programs, except for
649 * root. Record any set-id flags first to make sure that
650 * we do not regain any tracing during a possible block.
655 if (p->p_tracevp != NULL &&
656 priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) {
657 mtx_lock(&ktrace_mtx);
659 tracevp = p->p_tracevp;
661 tracecred = p->p_tracecred;
662 p->p_tracecred = NULL;
663 mtx_unlock(&ktrace_mtx);
667 * Close any file descriptors 0..2 that reference procfs,
668 * then make sure file descriptors 0..2 are in use.
670 * setugidsafety() may call closef() and then pfind()
671 * which may grab the process lock.
672 * fdcheckstd() may call falloc() which may block to
673 * allocate memory, so temporarily drop the process lock.
676 VOP_UNLOCK(imgp->vp, 0);
678 error = fdcheckstd(td);
679 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
684 * Set the new credentials.
686 if (attr.va_mode & S_ISUID)
687 change_euid(newcred, euip);
688 if (attr.va_mode & S_ISGID)
689 change_egid(newcred, attr.va_gid);
691 if (will_transition) {
692 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
693 interpvplabel, imgp);
697 * Implement correct POSIX saved-id behavior.
699 * XXXMAC: Note that the current logic will save the
700 * uid and gid if a MAC domain transition occurs, even
701 * though maybe it shouldn't.
703 change_svuid(newcred, newcred->cr_uid);
704 change_svgid(newcred, newcred->cr_gid);
705 p->p_ucred = newcred;
708 if (oldcred->cr_uid == oldcred->cr_ruid &&
709 oldcred->cr_gid == oldcred->cr_rgid)
710 p->p_flag &= ~P_SUGID;
712 * Implement correct POSIX saved-id behavior.
714 * XXX: It's not clear that the existing behavior is
715 * POSIX-compliant. A number of sources indicate that the
716 * saved uid/gid should only be updated if the new ruid is
717 * not equal to the old ruid, or the new euid is not equal
718 * to the old euid and the new euid is not equal to the old
719 * ruid. The FreeBSD code always updates the saved uid/gid.
720 * Also, this code uses the new (replaced) euid and egid as
721 * the source, which may or may not be the right ones to use.
723 if (oldcred->cr_svuid != oldcred->cr_uid ||
724 oldcred->cr_svgid != oldcred->cr_gid) {
725 change_svuid(newcred, newcred->cr_uid);
726 change_svgid(newcred, newcred->cr_gid);
727 p->p_ucred = newcred;
733 * Store the vp for use in procfs. This vnode was referenced prior
734 * to locking the proc lock.
736 textvp = p->p_textvp;
741 * Tell the DTrace fasttrap provider about the exec if it
742 * has declared an interest.
744 if (dtrace_fasttrap_exec)
745 dtrace_fasttrap_exec(p);
749 * Notify others that we exec'd, and clear the P_INEXEC flag
750 * as we're now a bona fide freshly-execed process.
752 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
753 p->p_flag &= ~P_INEXEC;
756 * If tracing the process, trap to debugger so breakpoints
757 * can be set before the program executes.
758 * Use tdsignal to deliver signal to current thread, use
759 * psignal may cause the signal to be delivered to wrong thread
760 * because that thread will exit, remember we are going to enter
761 * single thread mode.
763 if (p->p_flag & P_TRACED)
764 tdsignal(p, td, SIGTRAP, NULL);
766 /* clear "fork but no exec" flag, as we _are_ execing */
767 p->p_acflag &= ~AFORK;
770 * Free any previous argument cache and replace it with
771 * the new argument cache, if any.
779 * Check if system-wide sampling is in effect or if the
780 * current process is using PMCs. If so, do exec() time
781 * processing. This processing needs to happen AFTER the
782 * P_INEXEC flag is cleared.
784 * The proc lock needs to be released before taking the PMC
787 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
789 pe.pm_credentialschanged = credential_changing;
790 pe.pm_entryaddr = imgp->entry_addr;
792 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
795 #else /* !HWPMC_HOOKS */
799 /* Set values passed into the program in registers. */
800 if (p->p_sysent->sv_setregs)
801 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
802 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
804 exec_setregs(td, imgp->entry_addr,
805 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
807 vfs_mark_atime(imgp->vp, td->td_ucred);
809 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
813 * Free any resources malloc'd earlier that we didn't use.
820 VOP_UNLOCK(imgp->vp, 0);
823 * Handle deferred decrement of ref counts.
825 if (textvp != NULL) {
828 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
830 VFS_UNLOCK_GIANT(tvfslocked);
832 if (binvp && error != 0)
835 if (tracevp != NULL) {
838 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
840 VFS_UNLOCK_GIANT(tvfslocked);
842 if (tracecred != NULL)
845 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
848 if (oldsigacts != NULL)
849 sigacts_free(oldsigacts);
854 * free various allocated resources
856 if (imgp->firstpage != NULL)
857 exec_unmap_first_page(imgp);
859 if (imgp->vp != NULL) {
861 NDFREE(&nd, NDF_ONLY_PNBUF);
863 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
867 if (imgp->object != NULL)
868 vm_object_deallocate(imgp->object);
870 free(imgp->freepath, M_TEMP);
874 * Stop the process here if its stop event mask has
875 * the S_EXEC bit set.
877 STOPEVENT(p, S_EXEC, 0);
882 /* we're done here, clear P_INEXEC */
884 p->p_flag &= ~P_INEXEC;
887 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
891 mac_execve_exit(imgp);
892 mac_execve_interpreter_exit(interpvplabel);
894 VFS_UNLOCK_GIANT(vfslocked);
895 exec_free_args(args);
897 if (error && imgp->vmspace_destroyed) {
898 /* sorry, no more process anymore. exit gracefully */
899 exit1(td, W_EXITCODE(0, SIGABRT));
906 exec_map_first_page(imgp)
907 struct image_params *imgp;
911 vm_page_t ma[VM_INITIAL_PAGEIN];
914 if (imgp->firstpage != NULL)
915 exec_unmap_first_page(imgp);
917 object = imgp->vp->v_object;
920 VM_OBJECT_LOCK(object);
921 #if VM_NRESERVLEVEL > 0
922 if ((object->flags & OBJ_COLORED) == 0) {
923 object->flags |= OBJ_COLORED;
924 object->pg_color = 0;
927 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
928 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
929 initial_pagein = VM_INITIAL_PAGEIN;
930 if (initial_pagein > object->size)
931 initial_pagein = object->size;
932 for (i = 1; i < initial_pagein; i++) {
933 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
936 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
940 ma[i] = vm_page_alloc(object, i,
941 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
947 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
948 ma[0] = vm_page_lookup(object, 0);
949 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
951 vm_page_lock_queues();
953 vm_page_unlock_queues();
955 VM_OBJECT_UNLOCK(object);
959 vm_page_lock_queues();
961 vm_page_unlock_queues();
962 vm_page_wakeup(ma[0]);
963 VM_OBJECT_UNLOCK(object);
965 imgp->firstpage = sf_buf_alloc(ma[0], 0);
966 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
972 exec_unmap_first_page(imgp)
973 struct image_params *imgp;
977 if (imgp->firstpage != NULL) {
978 m = sf_buf_page(imgp->firstpage);
979 sf_buf_free(imgp->firstpage);
980 imgp->firstpage = NULL;
981 vm_page_lock_queues();
983 vm_page_unlock_queues();
988 * Destroy old address space, and allocate a new stack
989 * The new stack is only SGROWSIZ large because it is grown
990 * automatically in trap.c.
993 exec_new_vmspace(imgp, sv)
994 struct image_params *imgp;
995 struct sysentvec *sv;
998 struct proc *p = imgp->proc;
999 struct vmspace *vmspace = p->p_vmspace;
1000 vm_offset_t stack_addr;
1004 imgp->vmspace_destroyed = 1;
1007 /* May be called with Giant held */
1008 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1011 * Blow away entire process VM, if address space not shared,
1012 * otherwise, create a new VM space so that other threads are
1015 map = &vmspace->vm_map;
1016 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
1017 vm_map_max(map) == sv->sv_maxuser) {
1019 pmap_remove_pages(vmspace_pmap(vmspace));
1020 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1022 error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
1025 vmspace = p->p_vmspace;
1026 map = &vmspace->vm_map;
1029 /* Allocate a new stack */
1030 if (sv->sv_maxssiz != NULL)
1031 ssiz = *sv->sv_maxssiz;
1034 stack_addr = sv->sv_usrstack - ssiz;
1035 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1036 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1041 /* Allocate a new register stack */
1042 stack_addr = IA64_BACKINGSTORE;
1043 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1044 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1049 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1050 * VM_STACK case, but they are still used to monitor the size of the
1051 * process stack so we can check the stack rlimit.
1053 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1054 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1060 * Copy out argument and environment strings from the old process address
1061 * space into the temporary string buffer.
1064 exec_copyin_args(struct image_args *args, char *fname,
1065 enum uio_seg segflg, char **argv, char **envv)
1071 bzero(args, sizeof(*args));
1075 * Allocate temporary demand zeroed space for argument and
1076 * environment strings:
1078 * o ARG_MAX for argument and environment;
1079 * o MAXSHELLCMDLEN for the name of interpreters.
1081 args->buf = (char *) kmem_alloc_wait(exec_map,
1082 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1083 if (args->buf == NULL)
1085 args->begin_argv = args->buf;
1086 args->endp = args->begin_argv;
1087 args->stringspace = ARG_MAX;
1089 * Copy the file name.
1091 if (fname != NULL) {
1092 args->fname = args->buf + ARG_MAX;
1093 error = (segflg == UIO_SYSSPACE) ?
1094 copystr(fname, args->fname, PATH_MAX, &length) :
1095 copyinstr(fname, args->fname, PATH_MAX, &length);
1102 * extract arguments first
1104 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1105 if (argp == (caddr_t) -1) {
1109 if ((error = copyinstr(argp, args->endp,
1110 args->stringspace, &length))) {
1111 if (error == ENAMETOOLONG)
1115 args->stringspace -= length;
1116 args->endp += length;
1120 args->begin_envv = args->endp;
1123 * extract environment strings
1126 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1127 if (envp == (caddr_t)-1) {
1131 if ((error = copyinstr(envp, args->endp,
1132 args->stringspace, &length))) {
1133 if (error == ENAMETOOLONG)
1137 args->stringspace -= length;
1138 args->endp += length;
1146 exec_free_args(args);
1151 exec_free_args(struct image_args *args)
1155 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1156 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1162 * Copy strings out to the new process address space, constructing new arg
1163 * and env vector tables. Return a pointer to the base so that it can be used
1164 * as the initial stack pointer.
1167 exec_copyout_strings(imgp)
1168 struct image_params *imgp;
1172 char *stringp, *destp;
1173 register_t *stack_base;
1174 struct ps_strings *arginfo;
1176 size_t execpath_len;
1180 * Calculate string base and vector table pointers.
1181 * Also deal with signal trampoline code for this exec type.
1183 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1184 execpath_len = strlen(imgp->execpath) + 1;
1189 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1190 if (p->p_sysent->sv_szsigcode != NULL)
1191 szsigcode = *(p->p_sysent->sv_szsigcode);
1192 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1193 roundup(execpath_len, sizeof(char *)) -
1194 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1200 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1201 szsigcode), szsigcode);
1204 * Copy the image path for the rtld.
1206 if (execpath_len != 0) {
1207 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1208 copyout(imgp->execpath, (void *)imgp->execpathp,
1213 * If we have a valid auxargs ptr, prepare some room
1216 if (imgp->auxargs) {
1218 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1219 * lower compatibility.
1221 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1224 * The '+ 2' is for the null pointers at the end of each of
1225 * the arg and env vector sets,and imgp->auxarg_size is room
1226 * for argument of Runtime loader.
1228 vectp = (char **)(destp - (imgp->args->argc +
1229 imgp->args->envc + 2 + imgp->auxarg_size + execpath_len) *
1233 * The '+ 2' is for the null pointers at the end of each of
1234 * the arg and env vector sets
1236 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1241 * vectp also becomes our initial stack base
1243 stack_base = (register_t *)vectp;
1245 stringp = imgp->args->begin_argv;
1246 argc = imgp->args->argc;
1247 envc = imgp->args->envc;
1250 * Copy out strings - arguments and environment.
1252 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1255 * Fill in "ps_strings" struct for ps, w, etc.
1257 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1258 suword(&arginfo->ps_nargvstr, argc);
1261 * Fill in argument portion of vector table.
1263 for (; argc > 0; --argc) {
1264 suword(vectp++, (long)(intptr_t)destp);
1265 while (*stringp++ != 0)
1270 /* a null vector table pointer separates the argp's from the envp's */
1273 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1274 suword(&arginfo->ps_nenvstr, envc);
1277 * Fill in environment portion of vector table.
1279 for (; envc > 0; --envc) {
1280 suword(vectp++, (long)(intptr_t)destp);
1281 while (*stringp++ != 0)
1286 /* end of vector table is a null pointer */
1289 return (stack_base);
1293 * Check permissions of file to execute.
1294 * Called with imgp->vp locked.
1295 * Return 0 for success or error code on failure.
1298 exec_check_permissions(imgp)
1299 struct image_params *imgp;
1301 struct vnode *vp = imgp->vp;
1302 struct vattr *attr = imgp->attr;
1308 /* Get file attributes */
1309 error = VOP_GETATTR(vp, attr, td->td_ucred);
1314 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1320 * 1) Check if file execution is disabled for the filesystem that this
1322 * 2) Insure that at least one execute bit is on - otherwise root
1323 * will always succeed, and we don't want to happen unless the
1324 * file really is executable.
1325 * 3) Insure that the file is a regular file.
1327 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1328 ((attr->va_mode & 0111) == 0) ||
1329 (attr->va_type != VREG))
1333 * Zero length files can't be exec'd
1335 if (attr->va_size == 0)
1339 * Check for execute permission to file based on current credentials.
1341 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1346 * Check number of open-for-writes on the file and deny execution
1349 if (vp->v_writecount)
1353 * Call filesystem specific open routine (which does nothing in the
1356 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1363 * Exec handler registration
1366 exec_register(execsw_arg)
1367 const struct execsw *execsw_arg;
1369 const struct execsw **es, **xs, **newexecsw;
1370 int count = 2; /* New slot and trailing NULL */
1373 for (es = execsw; *es; es++)
1375 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1376 if (newexecsw == NULL)
1380 for (es = execsw; *es; es++)
1385 free(execsw, M_TEMP);
1391 exec_unregister(execsw_arg)
1392 const struct execsw *execsw_arg;
1394 const struct execsw **es, **xs, **newexecsw;
1398 panic("unregister with no handlers left?\n");
1400 for (es = execsw; *es; es++) {
1401 if (*es == execsw_arg)
1406 for (es = execsw; *es; es++)
1407 if (*es != execsw_arg)
1409 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1410 if (newexecsw == NULL)
1413 for (es = execsw; *es; es++)
1414 if (*es != execsw_arg)
1418 free(execsw, M_TEMP);