2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 1993, David Greenman
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
37 #include <sys/param.h>
38 #include <sys/systm.h>
41 #include <sys/capsicum.h>
42 #include <sys/compressor.h>
43 #include <sys/eventhandler.h>
45 #include <sys/fcntl.h>
46 #include <sys/filedesc.h>
47 #include <sys/imgact.h>
48 #include <sys/imgact_elf.h>
49 #include <sys/kernel.h>
51 #include <sys/malloc.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
58 #include <sys/ptrace.h>
59 #include <sys/resourcevar.h>
60 #include <sys/rwlock.h>
61 #include <sys/sched.h>
63 #include <sys/sf_buf.h>
65 #include <sys/signalvar.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/sysproto.h>
72 #include <sys/timers.h>
74 #include <sys/vnode.h>
77 #include <sys/ktrace.h>
81 #include <vm/vm_param.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_kern.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_pager.h>
91 #include <sys/pmckern.h>
94 #include <machine/reg.h>
96 #include <security/audit/audit.h>
97 #include <security/mac/mac_framework.h>
100 #include <sys/dtrace_bsd.h>
101 dtrace_execexit_func_t dtrace_fasttrap_exec;
104 SDT_PROVIDER_DECLARE(proc);
105 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
106 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
107 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
109 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
111 int coredump_pack_fileinfo = 1;
112 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
113 &coredump_pack_fileinfo, 0,
114 "Enable file path packing in 'procstat -f' coredump notes");
116 int coredump_pack_vmmapinfo = 1;
117 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
118 &coredump_pack_vmmapinfo, 0,
119 "Enable file path packing in 'procstat -v' coredump notes");
121 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
122 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
123 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
124 static int do_execve(struct thread *td, struct image_args *args,
125 struct mac *mac_p, struct vmspace *oldvmspace);
127 /* XXX This should be vm_size_t. */
128 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
129 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
130 "Location of process' ps_strings structure");
132 /* XXX This should be vm_size_t. */
133 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
134 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
135 "Top of process stack");
137 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
138 NULL, 0, sysctl_kern_stackprot, "I",
139 "Stack memory permissions");
141 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
142 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
143 &ps_arg_cache_limit, 0,
144 "Process' command line characters cache limit");
146 static int disallow_high_osrel;
147 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
148 &disallow_high_osrel, 0,
149 "Disallow execution of binaries built for higher version of the world");
151 static int map_at_zero = 0;
152 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
153 "Permit processes to map an object at virtual address 0.");
156 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
163 if (req->flags & SCTL_MASK32) {
165 val = (unsigned int)p->p_sysent->sv_psstrings;
166 error = SYSCTL_OUT(req, &val, sizeof(val));
169 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
170 sizeof(p->p_sysent->sv_psstrings));
175 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
182 if (req->flags & SCTL_MASK32) {
184 val = (unsigned int)p->p_sysent->sv_usrstack;
185 error = SYSCTL_OUT(req, &val, sizeof(val));
188 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
189 sizeof(p->p_sysent->sv_usrstack));
194 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
199 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
200 sizeof(p->p_sysent->sv_stackprot)));
204 * Each of the items is a pointer to a `const struct execsw', hence the
205 * double pointer here.
207 static const struct execsw **execsw;
209 #ifndef _SYS_SYSPROTO_H_
218 sys_execve(struct thread *td, struct execve_args *uap)
220 struct image_args args;
221 struct vmspace *oldvmspace;
224 error = pre_execve(td, &oldvmspace);
227 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
228 uap->argv, uap->envv);
230 error = kern_execve(td, &args, NULL, oldvmspace);
231 post_execve(td, error, oldvmspace);
232 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
236 #ifndef _SYS_SYSPROTO_H_
237 struct fexecve_args {
244 sys_fexecve(struct thread *td, struct fexecve_args *uap)
246 struct image_args args;
247 struct vmspace *oldvmspace;
250 error = pre_execve(td, &oldvmspace);
253 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
254 uap->argv, uap->envv);
257 error = kern_execve(td, &args, NULL, oldvmspace);
259 post_execve(td, error, oldvmspace);
260 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
264 #ifndef _SYS_SYSPROTO_H_
265 struct __mac_execve_args {
274 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
277 struct image_args args;
278 struct vmspace *oldvmspace;
281 error = pre_execve(td, &oldvmspace);
284 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
285 uap->argv, uap->envv);
287 error = kern_execve(td, &args, uap->mac_p, oldvmspace);
288 post_execve(td, error, oldvmspace);
289 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
297 pre_execve(struct thread *td, struct vmspace **oldvmspace)
302 KASSERT(td == curthread, ("non-current thread %p", td));
305 if ((p->p_flag & P_HADTHREADS) != 0) {
307 if (thread_single(p, SINGLE_BOUNDARY) != 0)
311 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
313 *oldvmspace = p->p_vmspace;
318 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
322 KASSERT(td == curthread, ("non-current thread %p", td));
324 if ((p->p_flag & P_HADTHREADS) != 0) {
327 * If success, we upgrade to SINGLE_EXIT state to
328 * force other threads to suicide.
330 if (error == EJUSTRETURN)
331 thread_single(p, SINGLE_EXIT);
333 thread_single_end(p, SINGLE_BOUNDARY);
336 exec_cleanup(td, oldvmspace);
340 * kern_execve() has the astonishing property of not always returning to
341 * the caller. If sufficiently bad things happen during the call to
342 * do_execve(), it can end up calling exit1(); as a result, callers must
343 * avoid doing anything which they might need to undo (e.g., allocating
347 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
348 struct vmspace *oldvmspace)
351 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
352 exec_args_get_begin_envv(args) - args->begin_argv);
353 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
354 args->endp - exec_args_get_begin_envv(args));
355 return (do_execve(td, args, mac_p, oldvmspace));
359 * In-kernel implementation of execve(). All arguments are assumed to be
360 * userspace pointers from the passed thread.
363 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
364 struct vmspace *oldvmspace)
366 struct proc *p = td->td_proc;
368 struct ucred *oldcred;
369 struct uidinfo *euip = NULL;
370 uintptr_t stack_base;
371 struct image_params image_params, *imgp;
373 int (*img_first)(struct image_params *);
374 struct pargs *oldargs = NULL, *newargs = NULL;
375 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
377 struct ktr_io_params *kiop;
379 struct vnode *oldtextvp = NULL, *newtextvp;
380 int credential_changing;
382 struct label *interpvplabel = NULL;
386 struct pmckern_procexec pe;
388 int error, i, orig_osrel;
390 static const char fexecv_proc_title[] = "(fexecv)";
392 imgp = &image_params;
396 * Lock the process and set the P_INEXEC flag to indicate that
397 * it should be left alone until we're done here. This is
398 * necessary to avoid race conditions - e.g. in ptrace() -
399 * that might allow a local user to illicitly obtain elevated
403 KASSERT((p->p_flag & P_INEXEC) == 0,
404 ("%s(): process already has P_INEXEC flag", __func__));
405 p->p_flag |= P_INEXEC;
409 * Initialize part of the common data
411 bzero(imgp, sizeof(*imgp));
415 oldcred = p->p_ucred;
416 orig_osrel = p->p_osrel;
417 orig_fctl0 = p->p_fctl0;
420 error = mac_execve_enter(imgp, mac_p);
426 * Translate the file name. namei() returns a vnode pointer
427 * in ni_vp among other things.
429 * XXXAUDIT: It would be desirable to also audit the name of the
430 * interpreter if this is an interpreted binary.
432 if (args->fname != NULL) {
433 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
434 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
437 SDT_PROBE1(proc, , , exec, args->fname);
440 if (args->fname != NULL) {
441 #ifdef CAPABILITY_MODE
443 * While capability mode can't reach this point via direct
444 * path arguments to execve(), we also don't allow
445 * interpreters to be used in capability mode (for now).
446 * Catch indirect lookups and return a permissions error.
448 if (IN_CAPABILITY_MODE(td)) {
457 newtextvp = nd.ni_vp;
458 imgp->vp = newtextvp;
460 AUDIT_ARG_FD(args->fd);
462 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
464 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
467 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
468 AUDIT_ARG_VNODE1(newtextvp);
469 imgp->vp = newtextvp;
473 * Check file permissions. Also 'opens' file and sets its vnode to
476 error = exec_check_permissions(imgp);
478 goto exec_fail_dealloc;
480 imgp->object = imgp->vp->v_object;
481 if (imgp->object != NULL)
482 vm_object_reference(imgp->object);
484 error = exec_map_first_page(imgp);
486 goto exec_fail_dealloc;
488 imgp->proc->p_osrel = 0;
489 imgp->proc->p_fctl0 = 0;
492 * Implement image setuid/setgid.
494 * Determine new credentials before attempting image activators
495 * so that it can be used by process_exec handlers to determine
496 * credential/setid changes.
498 * Don't honor setuid/setgid if the filesystem prohibits it or if
499 * the process is being traced.
501 * We disable setuid/setgid/etc in capability mode on the basis
502 * that most setugid applications are not written with that
503 * environment in mind, and will therefore almost certainly operate
504 * incorrectly. In principle there's no reason that setugid
505 * applications might not be useful in capability mode, so we may want
506 * to reconsider this conservative design choice in the future.
508 * XXXMAC: For the time being, use NOSUID to also prohibit
509 * transitions on the file system.
511 credential_changing = 0;
512 credential_changing |= (attr.va_mode & S_ISUID) &&
513 oldcred->cr_uid != attr.va_uid;
514 credential_changing |= (attr.va_mode & S_ISGID) &&
515 oldcred->cr_gid != attr.va_gid;
517 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
518 interpvplabel, imgp);
519 credential_changing |= will_transition;
522 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
523 if (credential_changing)
524 imgp->proc->p_pdeathsig = 0;
526 if (credential_changing &&
527 #ifdef CAPABILITY_MODE
528 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
530 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
531 (p->p_flag & P_TRACED) == 0) {
532 imgp->credential_setid = true;
533 VOP_UNLOCK(imgp->vp);
534 imgp->newcred = crdup(oldcred);
535 if (attr.va_mode & S_ISUID) {
536 euip = uifind(attr.va_uid);
537 change_euid(imgp->newcred, euip);
539 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
540 if (attr.va_mode & S_ISGID)
541 change_egid(imgp->newcred, attr.va_gid);
543 * Implement correct POSIX saved-id behavior.
545 * XXXMAC: Note that the current logic will save the
546 * uid and gid if a MAC domain transition occurs, even
547 * though maybe it shouldn't.
549 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
550 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
553 * Implement correct POSIX saved-id behavior.
555 * XXX: It's not clear that the existing behavior is
556 * POSIX-compliant. A number of sources indicate that the
557 * saved uid/gid should only be updated if the new ruid is
558 * not equal to the old ruid, or the new euid is not equal
559 * to the old euid and the new euid is not equal to the old
560 * ruid. The FreeBSD code always updates the saved uid/gid.
561 * Also, this code uses the new (replaced) euid and egid as
562 * the source, which may or may not be the right ones to use.
564 if (oldcred->cr_svuid != oldcred->cr_uid ||
565 oldcred->cr_svgid != oldcred->cr_gid) {
566 VOP_UNLOCK(imgp->vp);
567 imgp->newcred = crdup(oldcred);
568 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
569 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
570 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
573 /* The new credentials are installed into the process later. */
576 * Do the best to calculate the full path to the image file.
578 if (args->fname != NULL && args->fname[0] == '/')
579 imgp->execpath = args->fname;
581 VOP_UNLOCK(imgp->vp);
582 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0)
583 imgp->execpath = args->fname;
584 vn_lock(imgp->vp, LK_SHARED | 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 goto exec_fail_dealloc;
616 * Special interpreter operation, cleanup and loop up to try to
617 * activate the interpreter.
619 if (imgp->interpreted) {
620 exec_unmap_first_page(imgp);
622 * The text reference needs to be removed for scripts.
623 * There is a short period before we determine that
624 * something is a script where text reference is active.
625 * The vnode lock is held over this entire period
626 * so nothing should illegitimately be blocked.
628 MPASS(imgp->textset);
629 VOP_UNSET_TEXT_CHECKED(newtextvp);
630 imgp->textset = false;
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, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
654 SAVENAME, 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);
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;
674 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
675 if (SV_PROC_FLAG(p, SV_CAPSICUM))
676 sys_cap_enter(td, NULL);
679 * Copy out strings (args and env) and initialize stack base.
681 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
683 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
684 goto exec_fail_dealloc;
690 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
692 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
693 goto exec_fail_dealloc;
696 if (args->fdp != NULL) {
697 /* Install a brand new file descriptor table. */
698 fdinstall_remapped(td, args->fdp);
702 * Keep on using the existing file descriptor table. For
703 * security and other reasons, the file descriptor table
704 * cannot be shared after an exec.
708 /* close files on exec */
713 * Malloc things before we need locks.
715 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
716 /* Cache arguments if they fit inside our allowance */
717 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
718 newargs = pargs_alloc(i);
719 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
723 * For security and other reasons, signal handlers cannot
724 * be shared after an exec. The new process gets a copy of the old
725 * handlers. In execsigs(), the new process will have its signals
728 if (sigacts_shared(p->p_sigacts)) {
729 oldsigacts = p->p_sigacts;
730 newsigacts = sigacts_alloc();
731 sigacts_copy(newsigacts, oldsigacts);
734 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
738 p->p_sigacts = newsigacts;
742 /* reset caught signals */
745 /* name this process - nameiexec(p, ndp) */
746 bzero(p->p_comm, sizeof(p->p_comm));
748 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
749 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
750 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
751 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
752 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
754 sched_clear_tdname(td);
758 * mark as execed, wakeup the process that vforked (if any) and tell
759 * it that it now has its own resources back
762 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
763 p->p_flag2 &= ~P2_NOTRACE;
764 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
765 p->p_flag2 &= ~P2_STKGAP_DISABLE;
766 if (p->p_flag & P_PPWAIT) {
767 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
768 cv_broadcast(&p->p_pwait);
769 /* STOPs are no longer ignored, arrange for AST */
774 * Implement image setuid/setgid installation.
776 if (imgp->credential_setid) {
778 * Turn off syscall tracing for set-id programs, except for
779 * root. Record any set-id flags first to make sure that
780 * we do not regain any tracing during a possible block.
786 kiop = ktrprocexec(p);
789 * Close any file descriptors 0..2 that reference procfs,
790 * then make sure file descriptors 0..2 are in use.
792 * Both fdsetugidsafety() and fdcheckstd() may call functions
793 * taking sleepable locks, so temporarily drop our locks.
796 VOP_UNLOCK(imgp->vp);
798 error = fdcheckstd(td);
799 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
801 goto exec_fail_dealloc;
804 if (will_transition) {
805 mac_vnode_execve_transition(oldcred, imgp->newcred,
806 imgp->vp, interpvplabel, imgp);
810 if (oldcred->cr_uid == oldcred->cr_ruid &&
811 oldcred->cr_gid == oldcred->cr_rgid)
812 p->p_flag &= ~P_SUGID;
815 * Set the new credentials.
817 if (imgp->newcred != NULL) {
818 proc_set_cred(p, imgp->newcred);
824 * Store the vp for use in procfs. This vnode was referenced by namei
827 oldtextvp = p->p_textvp;
828 p->p_textvp = newtextvp;
832 * Tell the DTrace fasttrap provider about the exec if it
833 * has declared an interest.
835 if (dtrace_fasttrap_exec)
836 dtrace_fasttrap_exec(p);
840 * Notify others that we exec'd, and clear the P_INEXEC flag
841 * as we're now a bona fide freshly-execed process.
843 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
844 p->p_flag &= ~P_INEXEC;
846 /* clear "fork but no exec" flag, as we _are_ execing */
847 p->p_acflag &= ~AFORK;
850 * Free any previous argument cache and replace it with
851 * the new argument cache, if any.
861 * Check if system-wide sampling is in effect or if the
862 * current process is using PMCs. If so, do exec() time
863 * processing. This processing needs to happen AFTER the
864 * P_INEXEC flag is cleared.
866 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
867 VOP_UNLOCK(imgp->vp);
868 pe.pm_credentialschanged = credential_changing;
869 pe.pm_entryaddr = imgp->entry_addr;
871 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
872 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
876 /* Set values passed into the program in registers. */
877 (*p->p_sysent->sv_setregs)(td, imgp, stack_base);
879 VOP_MMAPPED(imgp->vp);
881 SDT_PROBE1(proc, , , exec__success, args->fname);
885 p->p_osrel = orig_osrel;
886 p->p_fctl0 = orig_fctl0;
889 if (imgp->firstpage != NULL)
890 exec_unmap_first_page(imgp);
892 if (imgp->vp != NULL) {
894 NDFREE(&nd, NDF_ONLY_PNBUF);
896 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
898 VOP_UNSET_TEXT_CHECKED(imgp->vp);
902 VOP_UNLOCK(imgp->vp);
905 if (imgp->object != NULL)
906 vm_object_deallocate(imgp->object);
908 free(imgp->freepath, M_TEMP);
911 if (p->p_ptevents & PTRACE_EXEC) {
913 if (p->p_ptevents & PTRACE_EXEC)
914 td->td_dbgflags |= TDB_EXEC;
919 /* we're done here, clear P_INEXEC */
921 p->p_flag &= ~P_INEXEC;
924 SDT_PROBE1(proc, , , exec__failure, error);
927 if (imgp->newcred != NULL && oldcred != NULL)
928 crfree(imgp->newcred);
931 mac_execve_exit(imgp);
932 mac_execve_interpreter_exit(interpvplabel);
934 exec_free_args(args);
937 * Handle deferred decrement of ref counts.
939 if (oldtextvp != NULL)
941 ktr_io_params_free(kiop);
944 if (oldsigacts != NULL)
945 sigacts_free(oldsigacts);
949 if (error && imgp->vmspace_destroyed) {
950 /* sorry, no more process anymore. exit gracefully */
951 exec_cleanup(td, oldvmspace);
952 exit1(td, 0, SIGABRT);
962 * We don't want cpu_set_syscall_retval() to overwrite any of
963 * the register values put in place by exec_setregs().
964 * Implementations of cpu_set_syscall_retval() will leave
965 * registers unmodified when returning EJUSTRETURN.
967 return (error == 0 ? EJUSTRETURN : error);
971 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
973 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
974 KASSERT(td->td_proc->p_vmspace != oldvmspace,
975 ("oldvmspace still used"));
976 vmspace_free(oldvmspace);
977 td->td_pflags &= ~TDP_EXECVMSPC;
982 exec_map_first_page(struct image_params *imgp)
988 if (imgp->firstpage != NULL)
989 exec_unmap_first_page(imgp);
991 object = imgp->vp->v_object;
994 #if VM_NRESERVLEVEL > 0
995 if ((object->flags & OBJ_COLORED) == 0) {
996 VM_OBJECT_WLOCK(object);
997 vm_object_color(object, 0);
998 VM_OBJECT_WUNLOCK(object);
1001 error = vm_page_grab_valid_unlocked(&m, object, 0,
1002 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1003 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1005 if (error != VM_PAGER_OK)
1007 imgp->firstpage = sf_buf_alloc(m, 0);
1008 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1014 exec_unmap_first_page(struct image_params *imgp)
1018 if (imgp->firstpage != NULL) {
1019 m = sf_buf_page(imgp->firstpage);
1020 sf_buf_free(imgp->firstpage);
1021 imgp->firstpage = NULL;
1022 vm_page_unwire(m, PQ_ACTIVE);
1027 * Destroy old address space, and allocate a new stack.
1028 * The new stack is only sgrowsiz large because it is grown
1029 * automatically on a page fault.
1032 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1035 struct proc *p = imgp->proc;
1036 struct vmspace *vmspace = p->p_vmspace;
1037 struct thread *td = curthread;
1039 struct rlimit rlim_stack;
1040 vm_offset_t sv_minuser, stack_addr;
1042 vm_prot_t stack_prot;
1045 imgp->vmspace_destroyed = 1;
1048 sigfastblock_clear(td);
1051 if (sv->sv_onexec != NULL)
1052 sv->sv_onexec(p, imgp);
1054 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1057 * Blow away entire process VM, if address space not shared,
1058 * otherwise, create a new VM space so that other threads are
1061 map = &vmspace->vm_map;
1063 sv_minuser = sv->sv_minuser;
1065 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1066 if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1067 vm_map_min(map) == sv_minuser &&
1068 vm_map_max(map) == sv->sv_maxuser &&
1069 cpu_exec_vmspace_reuse(p, map)) {
1071 pmap_remove_pages(vmspace_pmap(vmspace));
1072 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1074 * An exec terminates mlockall(MCL_FUTURE).
1075 * ASLR and W^X states must be re-evaluated.
1078 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1079 MAP_ASLR_IGNSTART | MAP_WXORX);
1082 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1085 vmspace = p->p_vmspace;
1086 map = &vmspace->vm_map;
1088 map->flags |= imgp->map_flags;
1090 /* Map a shared page */
1091 obj = sv->sv_shared_page_obj;
1093 vm_object_reference(obj);
1094 error = vm_map_fixed(map, obj, 0,
1095 sv->sv_shared_page_base, sv->sv_shared_page_len,
1096 VM_PROT_READ | VM_PROT_EXECUTE,
1097 VM_PROT_READ | VM_PROT_EXECUTE,
1098 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1099 if (error != KERN_SUCCESS) {
1100 vm_object_deallocate(obj);
1101 return (vm_mmap_to_errno(error));
1105 /* Allocate a new stack */
1106 if (imgp->stack_sz != 0) {
1107 ssiz = trunc_page(imgp->stack_sz);
1109 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1111 if (ssiz > rlim_stack.rlim_max)
1112 ssiz = rlim_stack.rlim_max;
1113 if (ssiz > rlim_stack.rlim_cur) {
1114 rlim_stack.rlim_cur = ssiz;
1115 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1117 } else if (sv->sv_maxssiz != NULL) {
1118 ssiz = *sv->sv_maxssiz;
1122 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1123 if (ssiz < imgp->eff_stack_sz)
1124 imgp->eff_stack_sz = ssiz;
1125 stack_addr = sv->sv_usrstack - ssiz;
1126 stack_prot = obj != NULL && imgp->stack_prot != 0 ?
1127 imgp->stack_prot : sv->sv_stackprot;
1128 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot,
1129 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1130 if (error != KERN_SUCCESS) {
1131 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1132 "failed mach error %d errno %d\n", (uintmax_t)ssiz,
1133 stack_prot, error, vm_mmap_to_errno(error));
1134 return (vm_mmap_to_errno(error));
1138 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1139 * are still used to enforce the stack rlimit on the process stack.
1141 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1142 vmspace->vm_maxsaddr = (char *)stack_addr;
1148 * Copy out argument and environment strings from the old process address
1149 * space into the temporary string buffer.
1152 exec_copyin_args(struct image_args *args, const char *fname,
1153 enum uio_seg segflg, char **argv, char **envv)
1158 bzero(args, sizeof(*args));
1163 * Allocate demand-paged memory for the file name, argument, and
1164 * environment strings.
1166 error = exec_alloc_args(args);
1171 * Copy the file name.
1173 error = exec_args_add_fname(args, fname, segflg);
1178 * extract arguments first
1181 error = fueword(argv++, &arg);
1188 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1195 * extract environment strings
1199 error = fueword(envv++, &env);
1206 error = exec_args_add_env(args,
1207 (char *)(uintptr_t)env, UIO_USERSPACE);
1216 exec_free_args(args);
1221 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1222 const void *data, size_t datalen, const int *fds, size_t fdslen)
1224 struct filedesc *ofdp;
1229 memset(args, '\0', sizeof(*args));
1230 ofdp = td->td_proc->p_fd;
1231 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1233 error = exec_alloc_args(args);
1237 args->begin_argv = args->buf;
1238 args->stringspace = ARG_MAX;
1242 * Argument buffer has been provided. Copy it into the
1243 * kernel as a single string and add a terminating null
1246 error = copyin(data, args->begin_argv, datalen);
1249 args->begin_argv[datalen] = '\0';
1250 args->endp = args->begin_argv + datalen + 1;
1251 args->stringspace -= datalen + 1;
1254 * Traditional argument counting. Count the number of
1257 for (p = args->begin_argv; p < args->endp; ++p)
1261 /* No argument buffer provided. */
1262 args->endp = args->begin_argv;
1265 /* Create new file descriptor table. */
1266 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1267 error = copyin(fds, kfds, fdslen * sizeof(int));
1272 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1279 exec_free_args(args);
1283 struct exec_args_kva {
1286 SLIST_ENTRY(exec_args_kva) next;
1289 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1291 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1292 static struct mtx exec_args_kva_mtx;
1293 static u_int exec_args_gen;
1296 exec_prealloc_args_kva(void *arg __unused)
1298 struct exec_args_kva *argkva;
1301 SLIST_INIT(&exec_args_kva_freelist);
1302 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1303 for (i = 0; i < exec_map_entries; i++) {
1304 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1305 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1306 argkva->gen = exec_args_gen;
1307 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1310 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1313 exec_alloc_args_kva(void **cookie)
1315 struct exec_args_kva *argkva;
1317 argkva = (void *)atomic_readandclear_ptr(
1318 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1319 if (argkva == NULL) {
1320 mtx_lock(&exec_args_kva_mtx);
1321 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1322 (void)mtx_sleep(&exec_args_kva_freelist,
1323 &exec_args_kva_mtx, 0, "execkva", 0);
1324 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1325 mtx_unlock(&exec_args_kva_mtx);
1327 kasan_mark((void *)argkva->addr, exec_map_entry_size,
1328 exec_map_entry_size, 0);
1329 *(struct exec_args_kva **)cookie = argkva;
1330 return (argkva->addr);
1334 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1338 base = argkva->addr;
1339 kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1340 KASAN_EXEC_ARGS_FREED);
1341 if (argkva->gen != gen) {
1342 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1346 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1347 (uintptr_t)NULL, (uintptr_t)argkva)) {
1348 mtx_lock(&exec_args_kva_mtx);
1349 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1350 wakeup_one(&exec_args_kva_freelist);
1351 mtx_unlock(&exec_args_kva_mtx);
1356 exec_free_args_kva(void *cookie)
1359 exec_release_args_kva(cookie, exec_args_gen);
1363 exec_args_kva_lowmem(void *arg __unused)
1365 SLIST_HEAD(, exec_args_kva) head;
1366 struct exec_args_kva *argkva;
1370 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1373 * Force an madvise of each KVA range. Any currently allocated ranges
1374 * will have MADV_FREE applied once they are freed.
1377 mtx_lock(&exec_args_kva_mtx);
1378 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1379 mtx_unlock(&exec_args_kva_mtx);
1380 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1381 SLIST_REMOVE_HEAD(&head, next);
1382 exec_release_args_kva(argkva, gen);
1386 argkva = (void *)atomic_readandclear_ptr(
1387 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1389 exec_release_args_kva(argkva, gen);
1392 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1393 EVENTHANDLER_PRI_ANY);
1396 * Allocate temporary demand-paged, zero-filled memory for the file name,
1397 * argument, and environment strings.
1400 exec_alloc_args(struct image_args *args)
1403 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1408 exec_free_args(struct image_args *args)
1411 if (args->buf != NULL) {
1412 exec_free_args_kva(args->bufkva);
1415 if (args->fname_buf != NULL) {
1416 free(args->fname_buf, M_TEMP);
1417 args->fname_buf = NULL;
1419 if (args->fdp != NULL)
1420 fdescfree_remapped(args->fdp);
1424 * A set to functions to fill struct image args.
1426 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1427 * fname) before the other functions. All exec_args_add_arg() calls must
1428 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1429 * may be called any time after exec_args_add_fname().
1431 * exec_args_add_fname() - install path to be executed
1432 * exec_args_add_arg() - append an argument string
1433 * exec_args_add_env() - append an env string
1434 * exec_args_adjust_args() - adjust location of the argument list to
1435 * allow new arguments to be prepended
1438 exec_args_add_fname(struct image_args *args, const char *fname,
1439 enum uio_seg segflg)
1444 KASSERT(args->fname == NULL, ("fname already appended"));
1445 KASSERT(args->endp == NULL, ("already appending to args"));
1447 if (fname != NULL) {
1448 args->fname = args->buf;
1449 error = segflg == UIO_SYSSPACE ?
1450 copystr(fname, args->fname, PATH_MAX, &length) :
1451 copyinstr(fname, args->fname, PATH_MAX, &length);
1453 return (error == ENAMETOOLONG ? E2BIG : error);
1457 /* Set up for _arg_*()/_env_*() */
1458 args->endp = args->buf + length;
1459 /* begin_argv must be set and kept updated */
1460 args->begin_argv = args->endp;
1461 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1462 ("too little space remaining for arguments %zu < %zu",
1463 exec_map_entry_size - length, (size_t)ARG_MAX));
1464 args->stringspace = ARG_MAX;
1470 exec_args_add_str(struct image_args *args, const char *str,
1471 enum uio_seg segflg, int *countp)
1476 KASSERT(args->endp != NULL, ("endp not initialized"));
1477 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1479 error = (segflg == UIO_SYSSPACE) ?
1480 copystr(str, args->endp, args->stringspace, &length) :
1481 copyinstr(str, args->endp, args->stringspace, &length);
1483 return (error == ENAMETOOLONG ? E2BIG : error);
1484 args->stringspace -= length;
1485 args->endp += length;
1492 exec_args_add_arg(struct image_args *args, const char *argp,
1493 enum uio_seg segflg)
1496 KASSERT(args->envc == 0, ("appending args after env"));
1498 return (exec_args_add_str(args, argp, segflg, &args->argc));
1502 exec_args_add_env(struct image_args *args, const char *envp,
1503 enum uio_seg segflg)
1506 if (args->envc == 0)
1507 args->begin_envv = args->endp;
1509 return (exec_args_add_str(args, envp, segflg, &args->envc));
1513 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1517 KASSERT(args->endp != NULL, ("endp not initialized"));
1518 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1520 offset = extend - consume;
1521 if (args->stringspace < offset)
1523 memmove(args->begin_argv + extend, args->begin_argv + consume,
1524 args->endp - args->begin_argv + consume);
1526 args->begin_envv += offset;
1527 args->endp += offset;
1528 args->stringspace -= offset;
1533 exec_args_get_begin_envv(struct image_args *args)
1536 KASSERT(args->endp != NULL, ("endp not initialized"));
1539 return (args->begin_envv);
1540 return (args->endp);
1544 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1546 if (imgp->sysent->sv_stackgap == NULL ||
1547 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1548 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1549 (imgp->map_flags & MAP_ASLR) == 0)
1551 imgp->sysent->sv_stackgap(imgp, dp);
1555 * Copy strings out to the new process address space, constructing new arg
1556 * and env vector tables. Return a pointer to the base so that it can be used
1557 * as the initial stack pointer.
1560 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1565 uintptr_t destp, ustringp;
1566 struct ps_strings *arginfo;
1568 size_t execpath_len;
1569 int error, szsigcode, szps;
1570 char canary[sizeof(long) * 8];
1572 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1574 * Calculate string base and vector table pointers.
1575 * Also deal with signal trampoline code for this exec type.
1577 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1578 execpath_len = strlen(imgp->execpath) + 1;
1583 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1584 imgp->ps_strings = arginfo;
1585 if (p->p_sysent->sv_sigcode_base == 0) {
1586 if (p->p_sysent->sv_szsigcode != NULL)
1587 szsigcode = *(p->p_sysent->sv_szsigcode);
1589 destp = (uintptr_t)arginfo;
1594 if (szsigcode != 0) {
1596 destp = rounddown2(destp, sizeof(void *));
1597 error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1604 * Copy the image path for the rtld.
1606 if (execpath_len != 0) {
1607 destp -= execpath_len;
1608 destp = rounddown2(destp, sizeof(void *));
1609 imgp->execpathp = (void *)destp;
1610 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1616 * Prepare the canary for SSP.
1618 arc4rand(canary, sizeof(canary), 0);
1619 destp -= sizeof(canary);
1620 imgp->canary = (void *)destp;
1621 error = copyout(canary, imgp->canary, sizeof(canary));
1624 imgp->canarylen = sizeof(canary);
1627 * Prepare the pagesizes array.
1630 destp = rounddown2(destp, sizeof(void *));
1631 imgp->pagesizes = (void *)destp;
1632 error = copyout(pagesizes, imgp->pagesizes, szps);
1635 imgp->pagesizeslen = szps;
1638 * Allocate room for the argument and environment strings.
1640 destp -= ARG_MAX - imgp->args->stringspace;
1641 destp = rounddown2(destp, sizeof(void *));
1644 exec_stackgap(imgp, &destp);
1646 if (imgp->auxargs) {
1648 * Allocate room on the stack for the ELF auxargs
1649 * array. It has up to AT_COUNT entries.
1651 destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1652 destp = rounddown2(destp, sizeof(void *));
1655 vectp = (char **)destp;
1658 * Allocate room for the argv[] and env vectors including the
1659 * terminating NULL pointers.
1661 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1664 * vectp also becomes our initial stack base
1666 *stack_base = (uintptr_t)vectp;
1668 stringp = imgp->args->begin_argv;
1669 argc = imgp->args->argc;
1670 envc = imgp->args->envc;
1673 * Copy out strings - arguments and environment.
1675 error = copyout(stringp, (void *)ustringp,
1676 ARG_MAX - imgp->args->stringspace);
1681 * Fill in "ps_strings" struct for ps, w, etc.
1684 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1685 suword32(&arginfo->ps_nargvstr, argc) != 0)
1689 * Fill in argument portion of vector table.
1691 for (; argc > 0; --argc) {
1692 if (suword(vectp++, ustringp) != 0)
1694 while (*stringp++ != 0)
1699 /* a null vector table pointer separates the argp's from the envp's */
1700 if (suword(vectp++, 0) != 0)
1704 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1705 suword32(&arginfo->ps_nenvstr, envc) != 0)
1709 * Fill in environment portion of vector table.
1711 for (; envc > 0; --envc) {
1712 if (suword(vectp++, ustringp) != 0)
1714 while (*stringp++ != 0)
1719 /* end of vector table is a null pointer */
1720 if (suword(vectp, 0) != 0)
1723 if (imgp->auxargs) {
1725 error = imgp->sysent->sv_copyout_auxargs(imgp,
1735 * Check permissions of file to execute.
1736 * Called with imgp->vp locked.
1737 * Return 0 for success or error code on failure.
1740 exec_check_permissions(struct image_params *imgp)
1742 struct vnode *vp = imgp->vp;
1743 struct vattr *attr = imgp->attr;
1749 /* Get file attributes */
1750 error = VOP_GETATTR(vp, attr, td->td_ucred);
1755 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1761 * 1) Check if file execution is disabled for the filesystem that
1762 * this file resides on.
1763 * 2) Ensure that at least one execute bit is on. Otherwise, a
1764 * privileged user will always succeed, and we don't want this
1765 * to happen unless the file really is executable.
1766 * 3) Ensure that the file is a regular file.
1768 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1769 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1770 (attr->va_type != VREG))
1774 * Zero length files can't be exec'd
1776 if (attr->va_size == 0)
1780 * Check for execute permission to file based on current credentials.
1782 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1787 * Check number of open-for-writes on the file and deny execution
1790 * Add a text reference now so no one can write to the
1791 * executable while we're activating it.
1793 * Remember if this was set before and unset it in case this is not
1794 * actually an executable image.
1796 error = VOP_SET_TEXT(vp);
1799 imgp->textset = true;
1802 * Call filesystem specific open routine (which does nothing in the
1805 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1812 * Exec handler registration
1815 exec_register(const struct execsw *execsw_arg)
1817 const struct execsw **es, **xs, **newexecsw;
1818 u_int count = 2; /* New slot and trailing NULL */
1821 for (es = execsw; *es; es++)
1823 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1826 for (es = execsw; *es; es++)
1831 free(execsw, M_TEMP);
1837 exec_unregister(const struct execsw *execsw_arg)
1839 const struct execsw **es, **xs, **newexecsw;
1843 panic("unregister with no handlers left?\n");
1845 for (es = execsw; *es; es++) {
1846 if (*es == execsw_arg)
1851 for (es = execsw; *es; es++)
1852 if (*es != execsw_arg)
1854 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1856 for (es = execsw; *es; es++)
1857 if (*es != execsw_arg)
1861 free(execsw, M_TEMP);
1867 * Write out a core segment to the compression stream.
1870 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1876 chunk_len = MIN(len, CORE_BUF_SIZE);
1879 * We can get EFAULT error here.
1880 * In that case zero out the current chunk of the segment.
1882 error = copyin(base, buf, chunk_len);
1884 bzero(buf, chunk_len);
1885 error = compressor_write(p->comp, buf, chunk_len);
1895 core_write(struct coredump_params *p, const void *base, size_t len,
1896 off_t offset, enum uio_seg seg, size_t *resid)
1899 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1900 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1901 p->active_cred, p->file_cred, resid, p->td));
1905 core_output(char *base, size_t len, off_t offset, struct coredump_params *p,
1910 size_t resid, runlen;
1914 KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1915 ("%s: user address %p is not page-aligned", __func__, base));
1917 if (p->comp != NULL)
1918 return (compress_chunk(p, base, tmpbuf, len));
1920 map = &p->td->td_proc->p_vmspace->vm_map;
1921 for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1923 * Attempt to page in all virtual pages in the range. If a
1924 * virtual page is not backed by the pager, it is represented as
1925 * a hole in the file. This can occur with zero-filled
1926 * anonymous memory or truncated files, for example.
1928 for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
1929 error = vm_fault(map, (uintptr_t)base + runlen,
1930 VM_PROT_READ, VM_FAULT_NOFILL, NULL);
1932 success = error == KERN_SUCCESS;
1933 else if ((error == KERN_SUCCESS) != success)
1938 error = core_write(p, base, runlen, offset,
1939 UIO_USERSPACE, &resid);
1941 if (error != EFAULT)
1945 * EFAULT may be returned if the user mapping
1946 * could not be accessed, e.g., because a mapped
1947 * file has been truncated. Skip the page if no
1948 * progress was made, to protect against a
1949 * hypothetical scenario where vm_fault() was
1950 * successful but core_write() returns EFAULT
1961 error = vn_start_write(p->vp, &mp, V_WAIT);
1964 vn_lock(p->vp, LK_EXCLUSIVE | LK_RETRY);
1965 error = vn_truncate_locked(p->vp, offset + runlen,
1966 false, p->td->td_ucred);
1968 vn_finished_write(mp);
1977 * Drain into a core file.
1980 sbuf_drain_core_output(void *arg, const char *data, int len)
1982 struct coredump_params *p;
1985 p = (struct coredump_params *)arg;
1988 * Some kern_proc out routines that print to this sbuf may
1989 * call us with the process lock held. Draining with the
1990 * non-sleepable lock held is unsafe. The lock is needed for
1991 * those routines when dumping a live process. In our case we
1992 * can safely release the lock before draining and acquire
1995 locked = PROC_LOCKED(p->td->td_proc);
1997 PROC_UNLOCK(p->td->td_proc);
1998 if (p->comp != NULL)
1999 error = compressor_write(p->comp, __DECONST(char *, data), len);
2001 error = core_write(p, __DECONST(void *, data), len, p->offset,
2002 UIO_SYSSPACE, NULL);
2004 PROC_LOCK(p->td->td_proc);