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 execve_nosetid(struct image_params *imgp)
361 imgp->credential_setid = false;
362 if (imgp->newcred != NULL) {
363 crfree(imgp->newcred);
364 imgp->newcred = NULL;
369 * In-kernel implementation of execve(). All arguments are assumed to be
370 * userspace pointers from the passed thread.
373 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
374 struct vmspace *oldvmspace)
376 struct proc *p = td->td_proc;
378 struct ucred *oldcred;
379 struct uidinfo *euip = NULL;
380 uintptr_t stack_base;
381 struct image_params image_params, *imgp;
383 int (*img_first)(struct image_params *);
384 struct pargs *oldargs = NULL, *newargs = NULL;
385 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
387 struct ktr_io_params *kiop;
389 struct vnode *oldtextvp = NULL, *newtextvp;
390 int credential_changing;
392 struct label *interpvplabel = NULL;
396 struct pmckern_procexec pe;
398 int error, i, orig_osrel;
400 Elf_Brandinfo *orig_brandinfo;
401 static const char fexecv_proc_title[] = "(fexecv)";
403 imgp = &image_params;
409 * Lock the process and set the P_INEXEC flag to indicate that
410 * it should be left alone until we're done here. This is
411 * necessary to avoid race conditions - e.g. in ptrace() -
412 * that might allow a local user to illicitly obtain elevated
416 KASSERT((p->p_flag & P_INEXEC) == 0,
417 ("%s(): process already has P_INEXEC flag", __func__));
418 p->p_flag |= P_INEXEC;
422 * Initialize part of the common data
424 bzero(imgp, sizeof(*imgp));
428 oldcred = p->p_ucred;
429 orig_osrel = p->p_osrel;
430 orig_fctl0 = p->p_fctl0;
431 orig_brandinfo = p->p_elf_brandinfo;
434 error = mac_execve_enter(imgp, mac_p);
440 * Translate the file name. namei() returns a vnode pointer
441 * in ni_vp among other things.
443 * XXXAUDIT: It would be desirable to also audit the name of the
444 * interpreter if this is an interpreted binary.
446 if (args->fname != NULL) {
447 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
448 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
451 SDT_PROBE1(proc, , , exec, args->fname);
454 if (args->fname != NULL) {
455 #ifdef CAPABILITY_MODE
457 * While capability mode can't reach this point via direct
458 * path arguments to execve(), we also don't allow
459 * interpreters to be used in capability mode (for now).
460 * Catch indirect lookups and return a permissions error.
462 if (IN_CAPABILITY_MODE(td)) {
471 newtextvp = nd.ni_vp;
472 imgp->vp = newtextvp;
474 AUDIT_ARG_FD(args->fd);
476 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
478 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
481 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
482 AUDIT_ARG_VNODE1(newtextvp);
483 imgp->vp = newtextvp;
487 * Check file permissions. Also 'opens' file and sets its vnode to
490 error = exec_check_permissions(imgp);
492 goto exec_fail_dealloc;
494 imgp->object = imgp->vp->v_object;
495 if (imgp->object != NULL)
496 vm_object_reference(imgp->object);
498 error = exec_map_first_page(imgp);
500 goto exec_fail_dealloc;
502 imgp->proc->p_osrel = 0;
503 imgp->proc->p_fctl0 = 0;
504 imgp->proc->p_elf_brandinfo = NULL;
507 * Implement image setuid/setgid.
509 * Determine new credentials before attempting image activators
510 * so that it can be used by process_exec handlers to determine
511 * credential/setid changes.
513 * Don't honor setuid/setgid if the filesystem prohibits it or if
514 * the process is being traced.
516 * We disable setuid/setgid/etc in capability mode on the basis
517 * that most setugid applications are not written with that
518 * environment in mind, and will therefore almost certainly operate
519 * incorrectly. In principle there's no reason that setugid
520 * applications might not be useful in capability mode, so we may want
521 * to reconsider this conservative design choice in the future.
523 * XXXMAC: For the time being, use NOSUID to also prohibit
524 * transitions on the file system.
526 credential_changing = 0;
527 credential_changing |= (attr.va_mode & S_ISUID) &&
528 oldcred->cr_uid != attr.va_uid;
529 credential_changing |= (attr.va_mode & S_ISGID) &&
530 oldcred->cr_gid != attr.va_gid;
532 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
533 interpvplabel, imgp);
534 credential_changing |= will_transition;
537 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
538 if (credential_changing)
539 imgp->proc->p_pdeathsig = 0;
541 if (credential_changing &&
542 #ifdef CAPABILITY_MODE
543 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
545 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
546 (p->p_flag & P_TRACED) == 0) {
547 imgp->credential_setid = true;
548 VOP_UNLOCK(imgp->vp);
549 imgp->newcred = crdup(oldcred);
550 if (attr.va_mode & S_ISUID) {
551 euip = uifind(attr.va_uid);
552 change_euid(imgp->newcred, euip);
554 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
555 if (attr.va_mode & S_ISGID)
556 change_egid(imgp->newcred, attr.va_gid);
558 * Implement correct POSIX saved-id behavior.
560 * XXXMAC: Note that the current logic will save the
561 * uid and gid if a MAC domain transition occurs, even
562 * though maybe it shouldn't.
564 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
565 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
568 * Implement correct POSIX saved-id behavior.
570 * XXX: It's not clear that the existing behavior is
571 * POSIX-compliant. A number of sources indicate that the
572 * saved uid/gid should only be updated if the new ruid is
573 * not equal to the old ruid, or the new euid is not equal
574 * to the old euid and the new euid is not equal to the old
575 * ruid. The FreeBSD code always updates the saved uid/gid.
576 * Also, this code uses the new (replaced) euid and egid as
577 * the source, which may or may not be the right ones to use.
579 if (oldcred->cr_svuid != oldcred->cr_uid ||
580 oldcred->cr_svgid != oldcred->cr_gid) {
581 VOP_UNLOCK(imgp->vp);
582 imgp->newcred = crdup(oldcred);
583 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
584 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
585 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
588 /* The new credentials are installed into the process later. */
591 * Do the best to calculate the full path to the image file.
593 if (args->fname != NULL && args->fname[0] == '/')
594 imgp->execpath = args->fname;
596 VOP_UNLOCK(imgp->vp);
597 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0)
598 imgp->execpath = args->fname;
599 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
603 * If the current process has a special image activator it
604 * wants to try first, call it. For example, emulating shell
605 * scripts differently.
608 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
609 error = img_first(imgp);
612 * Loop through the list of image activators, calling each one.
613 * An activator returns -1 if there is no match, 0 on success,
614 * and an error otherwise.
616 for (i = 0; error == -1 && execsw[i]; ++i) {
617 if (execsw[i]->ex_imgact == NULL ||
618 execsw[i]->ex_imgact == img_first) {
621 error = (*execsw[i]->ex_imgact)(imgp);
627 goto exec_fail_dealloc;
631 * Special interpreter operation, cleanup and loop up to try to
632 * activate the interpreter.
634 if (imgp->interpreted) {
635 exec_unmap_first_page(imgp);
637 * The text reference needs to be removed for scripts.
638 * There is a short period before we determine that
639 * something is a script where text reference is active.
640 * The vnode lock is held over this entire period
641 * so nothing should illegitimately be blocked.
643 MPASS(imgp->textset);
644 VOP_UNSET_TEXT_CHECKED(newtextvp);
645 imgp->textset = false;
646 /* free name buffer and old vnode */
647 if (args->fname != NULL)
648 NDFREE(&nd, NDF_ONLY_PNBUF);
650 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
653 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
657 vm_object_deallocate(imgp->object);
659 execve_nosetid(imgp);
660 imgp->execpath = NULL;
661 free(imgp->freepath, M_TEMP);
662 imgp->freepath = NULL;
663 /* set new name to that of the interpreter */
664 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
665 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
666 args->fname = imgp->interpreter_name;
671 * NB: We unlock the vnode here because it is believed that none
672 * of the sv_copyout_strings/sv_fixup operations require the vnode.
674 VOP_UNLOCK(imgp->vp);
676 if (disallow_high_osrel &&
677 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
679 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
680 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
681 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
682 goto exec_fail_dealloc;
685 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
686 if (SV_PROC_FLAG(p, SV_CAPSICUM))
687 sys_cap_enter(td, NULL);
690 * Copy out strings (args and env) and initialize stack base.
692 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
694 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
695 goto exec_fail_dealloc;
701 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
703 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
704 goto exec_fail_dealloc;
707 if (args->fdp != NULL) {
708 /* Install a brand new file descriptor table. */
709 fdinstall_remapped(td, args->fdp);
713 * Keep on using the existing file descriptor table. For
714 * security and other reasons, the file descriptor table
715 * cannot be shared after an exec.
719 /* close files on exec */
724 * Malloc things before we need locks.
726 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
727 /* Cache arguments if they fit inside our allowance */
728 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
729 newargs = pargs_alloc(i);
730 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
734 * For security and other reasons, signal handlers cannot
735 * be shared after an exec. The new process gets a copy of the old
736 * handlers. In execsigs(), the new process will have its signals
739 if (sigacts_shared(p->p_sigacts)) {
740 oldsigacts = p->p_sigacts;
741 newsigacts = sigacts_alloc();
742 sigacts_copy(newsigacts, oldsigacts);
745 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
749 p->p_sigacts = newsigacts;
753 /* reset caught signals */
756 /* name this process - nameiexec(p, ndp) */
757 bzero(p->p_comm, sizeof(p->p_comm));
759 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
760 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
761 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
762 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
763 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
765 sched_clear_tdname(td);
769 * mark as execed, wakeup the process that vforked (if any) and tell
770 * it that it now has its own resources back
773 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
774 p->p_flag2 &= ~P2_NOTRACE;
775 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
776 p->p_flag2 &= ~P2_STKGAP_DISABLE;
777 if (p->p_flag & P_PPWAIT) {
778 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
779 cv_broadcast(&p->p_pwait);
780 /* STOPs are no longer ignored, arrange for AST */
784 if ((imgp->sysent->sv_setid_allowed != NULL &&
785 !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
786 (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
787 execve_nosetid(imgp);
790 * Implement image setuid/setgid installation.
792 if (imgp->credential_setid) {
794 * Turn off syscall tracing for set-id programs, except for
795 * root. Record any set-id flags first to make sure that
796 * we do not regain any tracing during a possible block.
800 kiop = ktrprocexec(p);
803 * Close any file descriptors 0..2 that reference procfs,
804 * then make sure file descriptors 0..2 are in use.
806 * Both fdsetugidsafety() and fdcheckstd() may call functions
807 * taking sleepable locks, so temporarily drop our locks.
810 VOP_UNLOCK(imgp->vp);
812 error = fdcheckstd(td);
813 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
815 goto exec_fail_dealloc;
818 if (will_transition) {
819 mac_vnode_execve_transition(oldcred, imgp->newcred,
820 imgp->vp, interpvplabel, imgp);
824 if (oldcred->cr_uid == oldcred->cr_ruid &&
825 oldcred->cr_gid == oldcred->cr_rgid)
826 p->p_flag &= ~P_SUGID;
829 * Set the new credentials.
831 if (imgp->newcred != NULL) {
832 proc_set_cred(p, imgp->newcred);
838 * Store the vp for use in procfs. This vnode was referenced by namei
841 oldtextvp = p->p_textvp;
842 p->p_textvp = newtextvp;
846 * Tell the DTrace fasttrap provider about the exec if it
847 * has declared an interest.
849 if (dtrace_fasttrap_exec)
850 dtrace_fasttrap_exec(p);
854 * Notify others that we exec'd, and clear the P_INEXEC flag
855 * as we're now a bona fide freshly-execed process.
857 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
858 p->p_flag &= ~P_INEXEC;
860 /* clear "fork but no exec" flag, as we _are_ execing */
861 p->p_acflag &= ~AFORK;
864 * Free any previous argument cache and replace it with
865 * the new argument cache, if any.
875 * Check if system-wide sampling is in effect or if the
876 * current process is using PMCs. If so, do exec() time
877 * processing. This processing needs to happen AFTER the
878 * P_INEXEC flag is cleared.
880 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
881 VOP_UNLOCK(imgp->vp);
882 pe.pm_credentialschanged = credential_changing;
883 pe.pm_entryaddr = imgp->entry_addr;
885 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
886 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
890 /* Set values passed into the program in registers. */
891 (*p->p_sysent->sv_setregs)(td, imgp, stack_base);
893 VOP_MMAPPED(imgp->vp);
895 SDT_PROBE1(proc, , , exec__success, args->fname);
899 p->p_osrel = orig_osrel;
900 p->p_fctl0 = orig_fctl0;
901 p->p_elf_brandinfo = orig_brandinfo;
904 if (imgp->firstpage != NULL)
905 exec_unmap_first_page(imgp);
907 if (imgp->vp != NULL) {
909 NDFREE(&nd, NDF_ONLY_PNBUF);
911 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
913 VOP_UNSET_TEXT_CHECKED(imgp->vp);
917 VOP_UNLOCK(imgp->vp);
920 if (imgp->object != NULL)
921 vm_object_deallocate(imgp->object);
923 free(imgp->freepath, M_TEMP);
926 if (p->p_ptevents & PTRACE_EXEC) {
928 if (p->p_ptevents & PTRACE_EXEC)
929 td->td_dbgflags |= TDB_EXEC;
934 /* we're done here, clear P_INEXEC */
936 p->p_flag &= ~P_INEXEC;
939 SDT_PROBE1(proc, , , exec__failure, error);
942 if (imgp->newcred != NULL && oldcred != NULL)
943 crfree(imgp->newcred);
946 mac_execve_exit(imgp);
947 mac_execve_interpreter_exit(interpvplabel);
949 exec_free_args(args);
952 * Handle deferred decrement of ref counts.
954 if (oldtextvp != NULL)
957 ktr_io_params_free(kiop);
961 if (oldsigacts != NULL)
962 sigacts_free(oldsigacts);
966 if (error && imgp->vmspace_destroyed) {
967 /* sorry, no more process anymore. exit gracefully */
968 exec_cleanup(td, oldvmspace);
969 exit1(td, 0, SIGABRT);
979 * We don't want cpu_set_syscall_retval() to overwrite any of
980 * the register values put in place by exec_setregs().
981 * Implementations of cpu_set_syscall_retval() will leave
982 * registers unmodified when returning EJUSTRETURN.
984 return (error == 0 ? EJUSTRETURN : error);
988 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
990 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
991 KASSERT(td->td_proc->p_vmspace != oldvmspace,
992 ("oldvmspace still used"));
993 vmspace_free(oldvmspace);
994 td->td_pflags &= ~TDP_EXECVMSPC;
999 exec_map_first_page(struct image_params *imgp)
1005 if (imgp->firstpage != NULL)
1006 exec_unmap_first_page(imgp);
1008 object = imgp->vp->v_object;
1011 #if VM_NRESERVLEVEL > 0
1012 if ((object->flags & OBJ_COLORED) == 0) {
1013 VM_OBJECT_WLOCK(object);
1014 vm_object_color(object, 0);
1015 VM_OBJECT_WUNLOCK(object);
1018 error = vm_page_grab_valid_unlocked(&m, object, 0,
1019 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1020 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1022 if (error != VM_PAGER_OK)
1024 imgp->firstpage = sf_buf_alloc(m, 0);
1025 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1031 exec_unmap_first_page(struct image_params *imgp)
1035 if (imgp->firstpage != NULL) {
1036 m = sf_buf_page(imgp->firstpage);
1037 sf_buf_free(imgp->firstpage);
1038 imgp->firstpage = NULL;
1039 vm_page_unwire(m, PQ_ACTIVE);
1044 exec_onexec_old(struct thread *td)
1046 sigfastblock_clear(td);
1047 umtx_exec(td->td_proc);
1051 * Destroy old address space, and allocate a new stack.
1052 * The new stack is only sgrowsiz large because it is grown
1053 * automatically on a page fault.
1056 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1059 struct proc *p = imgp->proc;
1060 struct vmspace *vmspace = p->p_vmspace;
1061 struct thread *td = curthread;
1063 struct rlimit rlim_stack;
1064 vm_offset_t sv_minuser, stack_addr;
1066 vm_prot_t stack_prot;
1069 imgp->vmspace_destroyed = 1;
1072 if (p->p_sysent->sv_onexec_old != NULL)
1073 p->p_sysent->sv_onexec_old(td);
1076 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1079 * Blow away entire process VM, if address space not shared,
1080 * otherwise, create a new VM space so that other threads are
1083 map = &vmspace->vm_map;
1085 sv_minuser = sv->sv_minuser;
1087 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1088 if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1089 vm_map_min(map) == sv_minuser &&
1090 vm_map_max(map) == sv->sv_maxuser &&
1091 cpu_exec_vmspace_reuse(p, map)) {
1093 pmap_remove_pages(vmspace_pmap(vmspace));
1094 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1096 * An exec terminates mlockall(MCL_FUTURE).
1097 * ASLR and W^X states must be re-evaluated.
1100 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1101 MAP_ASLR_IGNSTART | MAP_WXORX);
1104 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1107 vmspace = p->p_vmspace;
1108 map = &vmspace->vm_map;
1110 map->flags |= imgp->map_flags;
1112 /* Map a shared page */
1113 obj = sv->sv_shared_page_obj;
1115 vm_object_reference(obj);
1116 error = vm_map_fixed(map, obj, 0,
1117 sv->sv_shared_page_base, sv->sv_shared_page_len,
1118 VM_PROT_READ | VM_PROT_EXECUTE,
1119 VM_PROT_READ | VM_PROT_EXECUTE,
1120 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1121 if (error != KERN_SUCCESS) {
1122 vm_object_deallocate(obj);
1123 return (vm_mmap_to_errno(error));
1127 /* Allocate a new stack */
1128 if (imgp->stack_sz != 0) {
1129 ssiz = trunc_page(imgp->stack_sz);
1131 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1133 if (ssiz > rlim_stack.rlim_max)
1134 ssiz = rlim_stack.rlim_max;
1135 if (ssiz > rlim_stack.rlim_cur) {
1136 rlim_stack.rlim_cur = ssiz;
1137 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1139 } else if (sv->sv_maxssiz != NULL) {
1140 ssiz = *sv->sv_maxssiz;
1144 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1145 if (ssiz < imgp->eff_stack_sz)
1146 imgp->eff_stack_sz = ssiz;
1147 stack_addr = sv->sv_usrstack - ssiz;
1148 stack_prot = obj != NULL && imgp->stack_prot != 0 ?
1149 imgp->stack_prot : sv->sv_stackprot;
1150 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot,
1151 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1152 if (error != KERN_SUCCESS) {
1153 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1154 "failed mach error %d errno %d\n", (uintmax_t)ssiz,
1155 stack_prot, error, vm_mmap_to_errno(error));
1156 return (vm_mmap_to_errno(error));
1160 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1161 * are still used to enforce the stack rlimit on the process stack.
1163 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1164 vmspace->vm_maxsaddr = (char *)stack_addr;
1166 return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1170 * Copy out argument and environment strings from the old process address
1171 * space into the temporary string buffer.
1174 exec_copyin_args(struct image_args *args, const char *fname,
1175 enum uio_seg segflg, char **argv, char **envv)
1180 bzero(args, sizeof(*args));
1185 * Allocate demand-paged memory for the file name, argument, and
1186 * environment strings.
1188 error = exec_alloc_args(args);
1193 * Copy the file name.
1195 error = exec_args_add_fname(args, fname, segflg);
1200 * extract arguments first
1203 error = fueword(argv++, &arg);
1210 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1217 * extract environment strings
1221 error = fueword(envv++, &env);
1228 error = exec_args_add_env(args,
1229 (char *)(uintptr_t)env, UIO_USERSPACE);
1238 exec_free_args(args);
1243 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1244 const void *data, size_t datalen, const int *fds, size_t fdslen)
1246 struct filedesc *ofdp;
1251 memset(args, '\0', sizeof(*args));
1252 ofdp = td->td_proc->p_fd;
1253 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1255 error = exec_alloc_args(args);
1259 args->begin_argv = args->buf;
1260 args->stringspace = ARG_MAX;
1264 * Argument buffer has been provided. Copy it into the
1265 * kernel as a single string and add a terminating null
1268 error = copyin(data, args->begin_argv, datalen);
1271 args->begin_argv[datalen] = '\0';
1272 args->endp = args->begin_argv + datalen + 1;
1273 args->stringspace -= datalen + 1;
1276 * Traditional argument counting. Count the number of
1279 for (p = args->begin_argv; p < args->endp; ++p)
1283 /* No argument buffer provided. */
1284 args->endp = args->begin_argv;
1287 /* Create new file descriptor table. */
1288 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1289 error = copyin(fds, kfds, fdslen * sizeof(int));
1294 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1301 exec_free_args(args);
1305 struct exec_args_kva {
1308 SLIST_ENTRY(exec_args_kva) next;
1311 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1313 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1314 static struct mtx exec_args_kva_mtx;
1315 static u_int exec_args_gen;
1318 exec_prealloc_args_kva(void *arg __unused)
1320 struct exec_args_kva *argkva;
1323 SLIST_INIT(&exec_args_kva_freelist);
1324 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1325 for (i = 0; i < exec_map_entries; i++) {
1326 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1327 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1328 argkva->gen = exec_args_gen;
1329 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1332 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1335 exec_alloc_args_kva(void **cookie)
1337 struct exec_args_kva *argkva;
1339 argkva = (void *)atomic_readandclear_ptr(
1340 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1341 if (argkva == NULL) {
1342 mtx_lock(&exec_args_kva_mtx);
1343 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1344 (void)mtx_sleep(&exec_args_kva_freelist,
1345 &exec_args_kva_mtx, 0, "execkva", 0);
1346 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1347 mtx_unlock(&exec_args_kva_mtx);
1349 kasan_mark((void *)argkva->addr, exec_map_entry_size,
1350 exec_map_entry_size, 0);
1351 *(struct exec_args_kva **)cookie = argkva;
1352 return (argkva->addr);
1356 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1360 base = argkva->addr;
1361 kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1362 KASAN_EXEC_ARGS_FREED);
1363 if (argkva->gen != gen) {
1364 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1368 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1369 (uintptr_t)NULL, (uintptr_t)argkva)) {
1370 mtx_lock(&exec_args_kva_mtx);
1371 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1372 wakeup_one(&exec_args_kva_freelist);
1373 mtx_unlock(&exec_args_kva_mtx);
1378 exec_free_args_kva(void *cookie)
1381 exec_release_args_kva(cookie, exec_args_gen);
1385 exec_args_kva_lowmem(void *arg __unused)
1387 SLIST_HEAD(, exec_args_kva) head;
1388 struct exec_args_kva *argkva;
1392 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1395 * Force an madvise of each KVA range. Any currently allocated ranges
1396 * will have MADV_FREE applied once they are freed.
1399 mtx_lock(&exec_args_kva_mtx);
1400 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1401 mtx_unlock(&exec_args_kva_mtx);
1402 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1403 SLIST_REMOVE_HEAD(&head, next);
1404 exec_release_args_kva(argkva, gen);
1408 argkva = (void *)atomic_readandclear_ptr(
1409 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1411 exec_release_args_kva(argkva, gen);
1414 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1415 EVENTHANDLER_PRI_ANY);
1418 * Allocate temporary demand-paged, zero-filled memory for the file name,
1419 * argument, and environment strings.
1422 exec_alloc_args(struct image_args *args)
1425 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1430 exec_free_args(struct image_args *args)
1433 if (args->buf != NULL) {
1434 exec_free_args_kva(args->bufkva);
1437 if (args->fname_buf != NULL) {
1438 free(args->fname_buf, M_TEMP);
1439 args->fname_buf = NULL;
1441 if (args->fdp != NULL)
1442 fdescfree_remapped(args->fdp);
1446 * A set to functions to fill struct image args.
1448 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1449 * fname) before the other functions. All exec_args_add_arg() calls must
1450 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1451 * may be called any time after exec_args_add_fname().
1453 * exec_args_add_fname() - install path to be executed
1454 * exec_args_add_arg() - append an argument string
1455 * exec_args_add_env() - append an env string
1456 * exec_args_adjust_args() - adjust location of the argument list to
1457 * allow new arguments to be prepended
1460 exec_args_add_fname(struct image_args *args, const char *fname,
1461 enum uio_seg segflg)
1466 KASSERT(args->fname == NULL, ("fname already appended"));
1467 KASSERT(args->endp == NULL, ("already appending to args"));
1469 if (fname != NULL) {
1470 args->fname = args->buf;
1471 error = segflg == UIO_SYSSPACE ?
1472 copystr(fname, args->fname, PATH_MAX, &length) :
1473 copyinstr(fname, args->fname, PATH_MAX, &length);
1475 return (error == ENAMETOOLONG ? E2BIG : error);
1479 /* Set up for _arg_*()/_env_*() */
1480 args->endp = args->buf + length;
1481 /* begin_argv must be set and kept updated */
1482 args->begin_argv = args->endp;
1483 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1484 ("too little space remaining for arguments %zu < %zu",
1485 exec_map_entry_size - length, (size_t)ARG_MAX));
1486 args->stringspace = ARG_MAX;
1492 exec_args_add_str(struct image_args *args, const char *str,
1493 enum uio_seg segflg, int *countp)
1498 KASSERT(args->endp != NULL, ("endp not initialized"));
1499 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1501 error = (segflg == UIO_SYSSPACE) ?
1502 copystr(str, args->endp, args->stringspace, &length) :
1503 copyinstr(str, args->endp, args->stringspace, &length);
1505 return (error == ENAMETOOLONG ? E2BIG : error);
1506 args->stringspace -= length;
1507 args->endp += length;
1514 exec_args_add_arg(struct image_args *args, const char *argp,
1515 enum uio_seg segflg)
1518 KASSERT(args->envc == 0, ("appending args after env"));
1520 return (exec_args_add_str(args, argp, segflg, &args->argc));
1524 exec_args_add_env(struct image_args *args, const char *envp,
1525 enum uio_seg segflg)
1528 if (args->envc == 0)
1529 args->begin_envv = args->endp;
1531 return (exec_args_add_str(args, envp, segflg, &args->envc));
1535 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1539 KASSERT(args->endp != NULL, ("endp not initialized"));
1540 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1542 offset = extend - consume;
1543 if (args->stringspace < offset)
1545 memmove(args->begin_argv + extend, args->begin_argv + consume,
1546 args->endp - args->begin_argv + consume);
1548 args->begin_envv += offset;
1549 args->endp += offset;
1550 args->stringspace -= offset;
1555 exec_args_get_begin_envv(struct image_args *args)
1558 KASSERT(args->endp != NULL, ("endp not initialized"));
1561 return (args->begin_envv);
1562 return (args->endp);
1566 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1568 if (imgp->sysent->sv_stackgap == NULL ||
1569 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1570 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1571 (imgp->map_flags & MAP_ASLR) == 0)
1573 imgp->sysent->sv_stackgap(imgp, dp);
1577 * Copy strings out to the new process address space, constructing new arg
1578 * and env vector tables. Return a pointer to the base so that it can be used
1579 * as the initial stack pointer.
1582 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1587 uintptr_t destp, ustringp;
1588 struct ps_strings *arginfo;
1590 size_t execpath_len;
1591 int error, szsigcode, szps;
1592 char canary[sizeof(long) * 8];
1594 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1596 * Calculate string base and vector table pointers.
1597 * Also deal with signal trampoline code for this exec type.
1599 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1600 execpath_len = strlen(imgp->execpath) + 1;
1605 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1606 imgp->ps_strings = arginfo;
1607 if (p->p_sysent->sv_sigcode_base == 0) {
1608 if (p->p_sysent->sv_szsigcode != NULL)
1609 szsigcode = *(p->p_sysent->sv_szsigcode);
1611 destp = (uintptr_t)arginfo;
1616 if (szsigcode != 0) {
1618 destp = rounddown2(destp, sizeof(void *));
1619 error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1626 * Copy the image path for the rtld.
1628 if (execpath_len != 0) {
1629 destp -= execpath_len;
1630 destp = rounddown2(destp, sizeof(void *));
1631 imgp->execpathp = (void *)destp;
1632 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1638 * Prepare the canary for SSP.
1640 arc4rand(canary, sizeof(canary), 0);
1641 destp -= sizeof(canary);
1642 imgp->canary = (void *)destp;
1643 error = copyout(canary, imgp->canary, sizeof(canary));
1646 imgp->canarylen = sizeof(canary);
1649 * Prepare the pagesizes array.
1652 destp = rounddown2(destp, sizeof(void *));
1653 imgp->pagesizes = (void *)destp;
1654 error = copyout(pagesizes, imgp->pagesizes, szps);
1657 imgp->pagesizeslen = szps;
1660 * Allocate room for the argument and environment strings.
1662 destp -= ARG_MAX - imgp->args->stringspace;
1663 destp = rounddown2(destp, sizeof(void *));
1666 exec_stackgap(imgp, &destp);
1668 if (imgp->auxargs) {
1670 * Allocate room on the stack for the ELF auxargs
1671 * array. It has up to AT_COUNT entries.
1673 destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1674 destp = rounddown2(destp, sizeof(void *));
1677 vectp = (char **)destp;
1680 * Allocate room for the argv[] and env vectors including the
1681 * terminating NULL pointers.
1683 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1686 * vectp also becomes our initial stack base
1688 *stack_base = (uintptr_t)vectp;
1690 stringp = imgp->args->begin_argv;
1691 argc = imgp->args->argc;
1692 envc = imgp->args->envc;
1695 * Copy out strings - arguments and environment.
1697 error = copyout(stringp, (void *)ustringp,
1698 ARG_MAX - imgp->args->stringspace);
1703 * Fill in "ps_strings" struct for ps, w, etc.
1706 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1707 suword32(&arginfo->ps_nargvstr, argc) != 0)
1711 * Fill in argument portion of vector table.
1713 for (; argc > 0; --argc) {
1714 if (suword(vectp++, ustringp) != 0)
1716 while (*stringp++ != 0)
1721 /* a null vector table pointer separates the argp's from the envp's */
1722 if (suword(vectp++, 0) != 0)
1726 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1727 suword32(&arginfo->ps_nenvstr, envc) != 0)
1731 * Fill in environment portion of vector table.
1733 for (; envc > 0; --envc) {
1734 if (suword(vectp++, ustringp) != 0)
1736 while (*stringp++ != 0)
1741 /* end of vector table is a null pointer */
1742 if (suword(vectp, 0) != 0)
1745 if (imgp->auxargs) {
1747 error = imgp->sysent->sv_copyout_auxargs(imgp,
1757 * Check permissions of file to execute.
1758 * Called with imgp->vp locked.
1759 * Return 0 for success or error code on failure.
1762 exec_check_permissions(struct image_params *imgp)
1764 struct vnode *vp = imgp->vp;
1765 struct vattr *attr = imgp->attr;
1771 /* Get file attributes */
1772 error = VOP_GETATTR(vp, attr, td->td_ucred);
1777 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1783 * 1) Check if file execution is disabled for the filesystem that
1784 * this file resides on.
1785 * 2) Ensure that at least one execute bit is on. Otherwise, a
1786 * privileged user will always succeed, and we don't want this
1787 * to happen unless the file really is executable.
1788 * 3) Ensure that the file is a regular file.
1790 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1791 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1792 (attr->va_type != VREG))
1796 * Zero length files can't be exec'd
1798 if (attr->va_size == 0)
1802 * Check for execute permission to file based on current credentials.
1804 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1809 * Check number of open-for-writes on the file and deny execution
1812 * Add a text reference now so no one can write to the
1813 * executable while we're activating it.
1815 * Remember if this was set before and unset it in case this is not
1816 * actually an executable image.
1818 error = VOP_SET_TEXT(vp);
1821 imgp->textset = true;
1824 * Call filesystem specific open routine (which does nothing in the
1827 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1834 * Exec handler registration
1837 exec_register(const struct execsw *execsw_arg)
1839 const struct execsw **es, **xs, **newexecsw;
1840 u_int count = 2; /* New slot and trailing NULL */
1843 for (es = execsw; *es; es++)
1845 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1848 for (es = execsw; *es; es++)
1853 free(execsw, M_TEMP);
1859 exec_unregister(const struct execsw *execsw_arg)
1861 const struct execsw **es, **xs, **newexecsw;
1865 panic("unregister with no handlers left?\n");
1867 for (es = execsw; *es; es++) {
1868 if (*es == execsw_arg)
1873 for (es = execsw; *es; es++)
1874 if (*es != execsw_arg)
1876 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1878 for (es = execsw; *es; es++)
1879 if (*es != execsw_arg)
1883 free(execsw, M_TEMP);
1889 * Write out a core segment to the compression stream.
1892 compress_chunk(struct coredump_params *cp, char *base, char *buf, u_int len)
1898 chunk_len = MIN(len, CORE_BUF_SIZE);
1901 * We can get EFAULT error here.
1902 * In that case zero out the current chunk of the segment.
1904 error = copyin(base, buf, chunk_len);
1906 bzero(buf, chunk_len);
1907 error = compressor_write(cp->comp, buf, chunk_len);
1917 core_write(struct coredump_params *cp, const void *base, size_t len,
1918 off_t offset, enum uio_seg seg, size_t *resid)
1921 return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1922 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1923 cp->active_cred, cp->file_cred, resid, cp->td));
1927 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1932 size_t resid, runlen;
1936 KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1937 ("%s: user address %p is not page-aligned", __func__, base));
1939 if (cp->comp != NULL)
1940 return (compress_chunk(cp, base, tmpbuf, len));
1942 map = &cp->td->td_proc->p_vmspace->vm_map;
1943 for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1945 * Attempt to page in all virtual pages in the range. If a
1946 * virtual page is not backed by the pager, it is represented as
1947 * a hole in the file. This can occur with zero-filled
1948 * anonymous memory or truncated files, for example.
1950 for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
1951 error = vm_fault(map, (uintptr_t)base + runlen,
1952 VM_PROT_READ, VM_FAULT_NOFILL, NULL);
1954 success = error == KERN_SUCCESS;
1955 else if ((error == KERN_SUCCESS) != success)
1960 error = core_write(cp, base, runlen, offset,
1961 UIO_USERSPACE, &resid);
1963 if (error != EFAULT)
1967 * EFAULT may be returned if the user mapping
1968 * could not be accessed, e.g., because a mapped
1969 * file has been truncated. Skip the page if no
1970 * progress was made, to protect against a
1971 * hypothetical scenario where vm_fault() was
1972 * successful but core_write() returns EFAULT
1983 error = vn_start_write(cp->vp, &mp, V_WAIT);
1986 vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
1987 error = vn_truncate_locked(cp->vp, offset + runlen,
1988 false, cp->td->td_ucred);
1990 vn_finished_write(mp);
1999 * Drain into a core file.
2002 sbuf_drain_core_output(void *arg, const char *data, int len)
2004 struct coredump_params *cp;
2009 p = cp->td->td_proc;
2012 * Some kern_proc out routines that print to this sbuf may
2013 * call us with the process lock held. Draining with the
2014 * non-sleepable lock held is unsafe. The lock is needed for
2015 * those routines when dumping a live process. In our case we
2016 * can safely release the lock before draining and acquire
2019 locked = PROC_LOCKED(p);
2022 if (cp->comp != NULL)
2023 error = compressor_write(cp->comp, __DECONST(char *, data), len);
2025 error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2026 UIO_SYSSPACE, NULL);