2 * SPDX-License-Identifier: BSD-2-Clause
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 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_ktrace.h"
35 #include <sys/param.h>
36 #include <sys/systm.h>
39 #include <sys/capsicum.h>
40 #include <sys/compressor.h>
41 #include <sys/eventhandler.h>
43 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
56 #include <sys/ptrace.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
62 #include <sys/sf_buf.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/timers.h>
72 #include <sys/umtxvar.h>
73 #include <sys/vnode.h>
76 #include <sys/ktrace.h>
80 #include <vm/vm_param.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_pager.h>
90 #include <sys/pmckern.h>
93 #include <security/audit/audit.h>
94 #include <security/mac/mac_framework.h>
97 #include <sys/dtrace_bsd.h>
98 dtrace_execexit_func_t dtrace_fasttrap_exec;
101 SDT_PROVIDER_DECLARE(proc);
102 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
108 int coredump_pack_fileinfo = 1;
109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
110 &coredump_pack_fileinfo, 0,
111 "Enable file path packing in 'procstat -f' coredump notes");
113 int coredump_pack_vmmapinfo = 1;
114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
115 &coredump_pack_vmmapinfo, 0,
116 "Enable file path packing in 'procstat -v' coredump notes");
118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
121 static int do_execve(struct thread *td, struct image_args *args,
122 struct mac *mac_p, struct vmspace *oldvmspace);
124 /* XXX This should be vm_size_t. */
125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
126 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
127 "Location of process' ps_strings structure");
129 /* XXX This should be vm_size_t. */
130 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
131 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
132 "Top of process stack");
134 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
135 NULL, 0, sysctl_kern_stackprot, "I",
136 "Stack memory permissions");
138 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
139 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
140 &ps_arg_cache_limit, 0,
141 "Process' command line characters cache limit");
143 static int disallow_high_osrel;
144 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
145 &disallow_high_osrel, 0,
146 "Disallow execution of binaries built for higher version of the world");
148 static int map_at_zero = 0;
149 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
150 "Permit processes to map an object at virtual address 0.");
152 static int core_dump_can_intr = 1;
153 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN,
154 &core_dump_can_intr, 0,
155 "Core dumping interruptible with SIGKILL");
158 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
161 vm_offset_t ps_strings;
165 if (req->flags & SCTL_MASK32) {
167 val = (unsigned int)PROC_PS_STRINGS(p);
168 return (SYSCTL_OUT(req, &val, sizeof(val)));
171 ps_strings = PROC_PS_STRINGS(p);
172 return (SYSCTL_OUT(req, &ps_strings, sizeof(ps_strings)));
176 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
183 if (req->flags & SCTL_MASK32) {
186 val32 = round_page((unsigned int)p->p_vmspace->vm_stacktop);
187 return (SYSCTL_OUT(req, &val32, sizeof(val32)));
190 val = round_page(p->p_vmspace->vm_stacktop);
191 return (SYSCTL_OUT(req, &val, sizeof(val)));
195 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
200 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
201 sizeof(p->p_sysent->sv_stackprot)));
205 * Each of the items is a pointer to a `const struct execsw', hence the
206 * double pointer here.
208 static const struct execsw **execsw;
210 #ifndef _SYS_SYSPROTO_H_
219 sys_execve(struct thread *td, struct execve_args *uap)
221 struct image_args args;
222 struct vmspace *oldvmspace;
225 error = pre_execve(td, &oldvmspace);
228 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
229 uap->argv, uap->envv);
231 error = kern_execve(td, &args, NULL, oldvmspace);
232 post_execve(td, error, oldvmspace);
233 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
237 #ifndef _SYS_SYSPROTO_H_
238 struct fexecve_args {
245 sys_fexecve(struct thread *td, struct fexecve_args *uap)
247 struct image_args args;
248 struct vmspace *oldvmspace;
251 error = pre_execve(td, &oldvmspace);
254 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
255 uap->argv, uap->envv);
258 error = kern_execve(td, &args, NULL, oldvmspace);
260 post_execve(td, error, oldvmspace);
261 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
265 #ifndef _SYS_SYSPROTO_H_
266 struct __mac_execve_args {
275 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
278 struct image_args args;
279 struct vmspace *oldvmspace;
282 error = pre_execve(td, &oldvmspace);
285 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
286 uap->argv, uap->envv);
288 error = kern_execve(td, &args, uap->mac_p, oldvmspace);
289 post_execve(td, error, oldvmspace);
290 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
298 pre_execve(struct thread *td, struct vmspace **oldvmspace)
303 KASSERT(td == curthread, ("non-current thread %p", td));
306 if ((p->p_flag & P_HADTHREADS) != 0) {
308 if (thread_single(p, SINGLE_BOUNDARY) != 0)
312 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
314 *oldvmspace = p->p_vmspace;
319 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
323 KASSERT(td == curthread, ("non-current thread %p", td));
325 if ((p->p_flag & P_HADTHREADS) != 0) {
328 * If success, we upgrade to SINGLE_EXIT state to
329 * force other threads to suicide.
331 if (error == EJUSTRETURN)
332 thread_single(p, SINGLE_EXIT);
334 thread_single_end(p, SINGLE_BOUNDARY);
337 exec_cleanup(td, oldvmspace);
341 * kern_execve() has the astonishing property of not always returning to
342 * the caller. If sufficiently bad things happen during the call to
343 * do_execve(), it can end up calling exit1(); as a result, callers must
344 * avoid doing anything which they might need to undo (e.g., allocating
348 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
349 struct vmspace *oldvmspace)
352 TSEXEC(td->td_proc->p_pid, args->begin_argv);
353 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
354 exec_args_get_begin_envv(args) - args->begin_argv);
355 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
356 args->endp - exec_args_get_begin_envv(args));
358 /* Must have at least one argument. */
359 if (args->argc == 0) {
360 exec_free_args(args);
363 return (do_execve(td, args, mac_p, oldvmspace));
367 execve_nosetid(struct image_params *imgp)
369 imgp->credential_setid = false;
370 if (imgp->newcred != NULL) {
371 crfree(imgp->newcred);
372 imgp->newcred = NULL;
377 * In-kernel implementation of execve(). All arguments are assumed to be
378 * userspace pointers from the passed thread.
381 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
382 struct vmspace *oldvmspace)
384 struct proc *p = td->td_proc;
386 struct ucred *oldcred;
387 struct uidinfo *euip = NULL;
388 uintptr_t stack_base;
389 struct image_params image_params, *imgp;
391 int (*img_first)(struct image_params *);
392 struct pargs *oldargs = NULL, *newargs = NULL;
393 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
395 struct ktr_io_params *kiop;
397 struct vnode *oldtextvp, *newtextvp;
398 struct vnode *oldtextdvp, *newtextdvp;
399 char *oldbinname, *newbinname;
400 bool credential_changing;
402 struct label *interpvplabel = NULL;
403 bool will_transition;
406 struct pmckern_procexec pe;
408 int error, i, orig_osrel;
410 Elf_Brandinfo *orig_brandinfo;
411 size_t freepath_size;
412 static const char fexecv_proc_title[] = "(fexecv)";
414 imgp = &image_params;
415 oldtextvp = oldtextdvp = NULL;
416 newtextvp = newtextdvp = NULL;
417 newbinname = oldbinname = NULL;
423 * Lock the process and set the P_INEXEC flag to indicate that
424 * it should be left alone until we're done here. This is
425 * necessary to avoid race conditions - e.g. in ptrace() -
426 * that might allow a local user to illicitly obtain elevated
430 KASSERT((p->p_flag & P_INEXEC) == 0,
431 ("%s(): process already has P_INEXEC flag", __func__));
432 p->p_flag |= P_INEXEC;
436 * Initialize part of the common data
438 bzero(imgp, sizeof(*imgp));
442 oldcred = p->p_ucred;
443 orig_osrel = p->p_osrel;
444 orig_fctl0 = p->p_fctl0;
445 orig_brandinfo = p->p_elf_brandinfo;
448 error = mac_execve_enter(imgp, mac_p);
453 SDT_PROBE1(proc, , , exec, args->fname);
456 if (args->fname != NULL) {
457 #ifdef CAPABILITY_MODE
459 * While capability mode can't reach this point via direct
460 * path arguments to execve(), we also don't allow
461 * interpreters to be used in capability mode (for now).
462 * Catch indirect lookups and return a permissions error.
464 if (IN_CAPABILITY_MODE(td)) {
471 * Translate the file name. namei() returns a vnode
472 * pointer in ni_vp among other things.
474 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
475 SAVENAME | AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
482 newtextvp = nd.ni_vp;
483 newtextdvp = nd.ni_dvp;
485 newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
487 memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
488 newbinname[nd.ni_cnd.cn_namelen] = '\0';
489 imgp->vp = newtextvp;
492 * Do the best to calculate the full path to the image file.
494 if (args->fname[0] == '/') {
495 imgp->execpath = args->fname;
497 VOP_UNLOCK(imgp->vp);
498 freepath_size = MAXPATHLEN;
499 if (vn_fullpath_hardlink(newtextvp, newtextdvp,
500 newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
501 &imgp->freepath, &freepath_size) != 0)
502 imgp->execpath = args->fname;
503 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
505 } else if (imgp->interpreter_vp) {
507 * An image activator has already provided an open vnode
509 newtextvp = imgp->interpreter_vp;
510 imgp->interpreter_vp = NULL;
511 if (vn_fullpath(newtextvp, &imgp->execpath,
512 &imgp->freepath) != 0)
513 imgp->execpath = args->fname;
514 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
515 AUDIT_ARG_VNODE1(newtextvp);
516 imgp->vp = newtextvp;
518 AUDIT_ARG_FD(args->fd);
521 * If the descriptors was not opened with O_PATH, then
522 * we require that it was opened with O_EXEC or
523 * O_RDONLY. In either case, exec_check_permissions()
524 * below checks _current_ file access mode regardless
525 * of the permissions additionally checked at the
528 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
533 if (vn_fullpath(newtextvp, &imgp->execpath,
534 &imgp->freepath) != 0)
535 imgp->execpath = args->fname;
536 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
537 AUDIT_ARG_VNODE1(newtextvp);
538 imgp->vp = newtextvp;
542 * Check file permissions. Also 'opens' file and sets its vnode to
545 error = exec_check_permissions(imgp);
547 goto exec_fail_dealloc;
549 imgp->object = imgp->vp->v_object;
550 if (imgp->object != NULL)
551 vm_object_reference(imgp->object);
553 error = exec_map_first_page(imgp);
555 goto exec_fail_dealloc;
557 imgp->proc->p_osrel = 0;
558 imgp->proc->p_fctl0 = 0;
559 imgp->proc->p_elf_brandinfo = NULL;
562 * Implement image setuid/setgid.
564 * Determine new credentials before attempting image activators
565 * so that it can be used by process_exec handlers to determine
566 * credential/setid changes.
568 * Don't honor setuid/setgid if the filesystem prohibits it or if
569 * the process is being traced.
571 * We disable setuid/setgid/etc in capability mode on the basis
572 * that most setugid applications are not written with that
573 * environment in mind, and will therefore almost certainly operate
574 * incorrectly. In principle there's no reason that setugid
575 * applications might not be useful in capability mode, so we may want
576 * to reconsider this conservative design choice in the future.
578 * XXXMAC: For the time being, use NOSUID to also prohibit
579 * transitions on the file system.
581 credential_changing = false;
582 credential_changing |= (attr.va_mode & S_ISUID) &&
583 oldcred->cr_uid != attr.va_uid;
584 credential_changing |= (attr.va_mode & S_ISGID) &&
585 oldcred->cr_gid != attr.va_gid;
587 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
588 interpvplabel, imgp) != 0;
589 credential_changing |= will_transition;
592 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
593 if (credential_changing)
594 imgp->proc->p_pdeathsig = 0;
596 if (credential_changing &&
597 #ifdef CAPABILITY_MODE
598 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
600 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
601 (p->p_flag & P_TRACED) == 0) {
602 imgp->credential_setid = true;
603 VOP_UNLOCK(imgp->vp);
604 imgp->newcred = crdup(oldcred);
605 if (attr.va_mode & S_ISUID) {
606 euip = uifind(attr.va_uid);
607 change_euid(imgp->newcred, euip);
609 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
610 if (attr.va_mode & S_ISGID)
611 change_egid(imgp->newcred, attr.va_gid);
613 * Implement correct POSIX saved-id behavior.
615 * XXXMAC: Note that the current logic will save the
616 * uid and gid if a MAC domain transition occurs, even
617 * though maybe it shouldn't.
619 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
620 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
623 * Implement correct POSIX saved-id behavior.
625 * XXX: It's not clear that the existing behavior is
626 * POSIX-compliant. A number of sources indicate that the
627 * saved uid/gid should only be updated if the new ruid is
628 * not equal to the old ruid, or the new euid is not equal
629 * to the old euid and the new euid is not equal to the old
630 * ruid. The FreeBSD code always updates the saved uid/gid.
631 * Also, this code uses the new (replaced) euid and egid as
632 * the source, which may or may not be the right ones to use.
634 if (oldcred->cr_svuid != oldcred->cr_uid ||
635 oldcred->cr_svgid != oldcred->cr_gid) {
636 VOP_UNLOCK(imgp->vp);
637 imgp->newcred = crdup(oldcred);
638 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
639 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
640 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
643 /* The new credentials are installed into the process later. */
646 * If the current process has a special image activator it
647 * wants to try first, call it. For example, emulating shell
648 * scripts differently.
651 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
652 error = img_first(imgp);
655 * Loop through the list of image activators, calling each one.
656 * An activator returns -1 if there is no match, 0 on success,
657 * and an error otherwise.
659 for (i = 0; error == -1 && execsw[i]; ++i) {
660 if (execsw[i]->ex_imgact == NULL ||
661 execsw[i]->ex_imgact == img_first) {
664 error = (*execsw[i]->ex_imgact)(imgp);
670 goto exec_fail_dealloc;
674 * Special interpreter operation, cleanup and loop up to try to
675 * activate the interpreter.
677 if (imgp->interpreted) {
678 exec_unmap_first_page(imgp);
680 * The text reference needs to be removed for scripts.
681 * There is a short period before we determine that
682 * something is a script where text reference is active.
683 * The vnode lock is held over this entire period
684 * so nothing should illegitimately be blocked.
686 MPASS(imgp->textset);
687 VOP_UNSET_TEXT_CHECKED(newtextvp);
688 imgp->textset = false;
689 /* free name buffer and old vnode */
691 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
694 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
695 imgp->opened = false;
698 imgp->vp = newtextvp = NULL;
699 if (args->fname != NULL) {
700 if (newtextdvp != NULL) {
704 NDFREE(&nd, NDF_ONLY_PNBUF);
705 free(newbinname, M_PARGS);
708 vm_object_deallocate(imgp->object);
710 execve_nosetid(imgp);
711 imgp->execpath = NULL;
712 free(imgp->freepath, M_TEMP);
713 imgp->freepath = NULL;
714 /* set new name to that of the interpreter */
715 if (imgp->interpreter_vp) {
718 args->fname = imgp->interpreter_name;
724 * NB: We unlock the vnode here because it is believed that none
725 * of the sv_copyout_strings/sv_fixup operations require the vnode.
727 VOP_UNLOCK(imgp->vp);
729 if (disallow_high_osrel &&
730 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
732 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
733 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
734 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
735 goto exec_fail_dealloc;
738 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
739 if (SV_PROC_FLAG(p, SV_CAPSICUM))
740 sys_cap_enter(td, NULL);
743 * Copy out strings (args and env) and initialize stack base.
745 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
747 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
748 goto exec_fail_dealloc;
754 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
756 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
757 goto exec_fail_dealloc;
760 if (args->fdp != NULL) {
761 /* Install a brand new file descriptor table. */
762 fdinstall_remapped(td, args->fdp);
766 * Keep on using the existing file descriptor table. For
767 * security and other reasons, the file descriptor table
768 * cannot be shared after an exec.
772 /* close files on exec */
777 * Malloc things before we need locks.
779 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
780 /* Cache arguments if they fit inside our allowance */
781 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
782 newargs = pargs_alloc(i);
783 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
787 * For security and other reasons, signal handlers cannot
788 * be shared after an exec. The new process gets a copy of the old
789 * handlers. In execsigs(), the new process will have its signals
792 if (sigacts_shared(p->p_sigacts)) {
793 oldsigacts = p->p_sigacts;
794 newsigacts = sigacts_alloc();
795 sigacts_copy(newsigacts, oldsigacts);
798 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
802 p->p_sigacts = newsigacts;
806 /* reset caught signals */
809 /* name this process - nameiexec(p, ndp) */
810 bzero(p->p_comm, sizeof(p->p_comm));
812 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
813 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
814 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
815 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
816 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
818 sched_clear_tdname(td);
822 * mark as execed, wakeup the process that vforked (if any) and tell
823 * it that it now has its own resources back
826 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
827 p->p_flag2 &= ~P2_NOTRACE;
828 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
829 p->p_flag2 &= ~P2_STKGAP_DISABLE;
830 if (p->p_flag & P_PPWAIT) {
831 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
832 cv_broadcast(&p->p_pwait);
833 /* STOPs are no longer ignored, arrange for AST */
837 if ((imgp->sysent->sv_setid_allowed != NULL &&
838 !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
839 (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
840 execve_nosetid(imgp);
843 * Implement image setuid/setgid installation.
845 if (imgp->credential_setid) {
847 * Turn off syscall tracing for set-id programs, except for
848 * root. Record any set-id flags first to make sure that
849 * we do not regain any tracing during a possible block.
853 kiop = ktrprocexec(p);
856 * Close any file descriptors 0..2 that reference procfs,
857 * then make sure file descriptors 0..2 are in use.
859 * Both fdsetugidsafety() and fdcheckstd() may call functions
860 * taking sleepable locks, so temporarily drop our locks.
863 VOP_UNLOCK(imgp->vp);
865 error = fdcheckstd(td);
866 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
868 goto exec_fail_dealloc;
871 if (will_transition) {
872 mac_vnode_execve_transition(oldcred, imgp->newcred,
873 imgp->vp, interpvplabel, imgp);
877 if (oldcred->cr_uid == oldcred->cr_ruid &&
878 oldcred->cr_gid == oldcred->cr_rgid)
879 p->p_flag &= ~P_SUGID;
882 * Set the new credentials.
884 if (imgp->newcred != NULL) {
885 proc_set_cred(p, imgp->newcred);
891 * Store the vp for use in kern.proc.pathname. This vnode was
892 * referenced by namei() or by fexecve variant of fname handling.
894 oldtextvp = p->p_textvp;
895 p->p_textvp = newtextvp;
896 oldtextdvp = p->p_textdvp;
897 p->p_textdvp = newtextdvp;
899 oldbinname = p->p_binname;
900 p->p_binname = newbinname;
905 * Tell the DTrace fasttrap provider about the exec if it
906 * has declared an interest.
908 if (dtrace_fasttrap_exec)
909 dtrace_fasttrap_exec(p);
913 * Notify others that we exec'd, and clear the P_INEXEC flag
914 * as we're now a bona fide freshly-execed process.
916 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
917 p->p_flag &= ~P_INEXEC;
919 /* clear "fork but no exec" flag, as we _are_ execing */
920 p->p_acflag &= ~AFORK;
923 * Free any previous argument cache and replace it with
924 * the new argument cache, if any.
934 * Check if system-wide sampling is in effect or if the
935 * current process is using PMCs. If so, do exec() time
936 * processing. This processing needs to happen AFTER the
937 * P_INEXEC flag is cleared.
939 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
940 VOP_UNLOCK(imgp->vp);
941 pe.pm_credentialschanged = credential_changing;
942 pe.pm_entryaddr = imgp->entry_addr;
944 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
945 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
949 /* Set values passed into the program in registers. */
950 (*p->p_sysent->sv_setregs)(td, imgp, stack_base);
952 VOP_MMAPPED(imgp->vp);
954 SDT_PROBE1(proc, , , exec__success, args->fname);
958 p->p_osrel = orig_osrel;
959 p->p_fctl0 = orig_fctl0;
960 p->p_elf_brandinfo = orig_brandinfo;
963 if (imgp->firstpage != NULL)
964 exec_unmap_first_page(imgp);
966 if (imgp->vp != NULL) {
968 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
970 VOP_UNSET_TEXT_CHECKED(imgp->vp);
974 VOP_UNLOCK(imgp->vp);
975 if (args->fname != NULL)
976 NDFREE(&nd, NDF_ONLY_PNBUF);
977 if (newtextdvp != NULL)
979 free(newbinname, M_PARGS);
982 if (imgp->object != NULL)
983 vm_object_deallocate(imgp->object);
985 free(imgp->freepath, M_TEMP);
988 if (p->p_ptevents & PTRACE_EXEC) {
990 if (p->p_ptevents & PTRACE_EXEC)
991 td->td_dbgflags |= TDB_EXEC;
996 /* we're done here, clear P_INEXEC */
998 p->p_flag &= ~P_INEXEC;
1001 SDT_PROBE1(proc, , , exec__failure, error);
1004 if (imgp->newcred != NULL && oldcred != NULL)
1005 crfree(imgp->newcred);
1008 mac_execve_exit(imgp);
1009 mac_execve_interpreter_exit(interpvplabel);
1011 exec_free_args(args);
1014 * Handle deferred decrement of ref counts.
1016 if (oldtextvp != NULL)
1018 if (oldtextdvp != NULL)
1020 free(oldbinname, M_PARGS);
1022 ktr_io_params_free(kiop);
1024 pargs_drop(oldargs);
1025 pargs_drop(newargs);
1026 if (oldsigacts != NULL)
1027 sigacts_free(oldsigacts);
1031 if (error && imgp->vmspace_destroyed) {
1032 /* sorry, no more process anymore. exit gracefully */
1033 exec_cleanup(td, oldvmspace);
1034 exit1(td, 0, SIGABRT);
1044 * We don't want cpu_set_syscall_retval() to overwrite any of
1045 * the register values put in place by exec_setregs().
1046 * Implementations of cpu_set_syscall_retval() will leave
1047 * registers unmodified when returning EJUSTRETURN.
1049 return (error == 0 ? EJUSTRETURN : error);
1053 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1055 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1056 KASSERT(td->td_proc->p_vmspace != oldvmspace,
1057 ("oldvmspace still used"));
1058 vmspace_free(oldvmspace);
1059 td->td_pflags &= ~TDP_EXECVMSPC;
1064 exec_map_first_page(struct image_params *imgp)
1070 if (imgp->firstpage != NULL)
1071 exec_unmap_first_page(imgp);
1073 object = imgp->vp->v_object;
1076 #if VM_NRESERVLEVEL > 0
1077 if ((object->flags & OBJ_COLORED) == 0) {
1078 VM_OBJECT_WLOCK(object);
1079 vm_object_color(object, 0);
1080 VM_OBJECT_WUNLOCK(object);
1083 error = vm_page_grab_valid_unlocked(&m, object, 0,
1084 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1085 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1087 if (error != VM_PAGER_OK)
1089 imgp->firstpage = sf_buf_alloc(m, 0);
1090 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1096 exec_unmap_first_page(struct image_params *imgp)
1100 if (imgp->firstpage != NULL) {
1101 m = sf_buf_page(imgp->firstpage);
1102 sf_buf_free(imgp->firstpage);
1103 imgp->firstpage = NULL;
1104 vm_page_unwire(m, PQ_ACTIVE);
1109 exec_onexec_old(struct thread *td)
1111 sigfastblock_clear(td);
1112 umtx_exec(td->td_proc);
1116 * This is an optimization which removes the unmanaged shared page
1117 * mapping. In combination with pmap_remove_pages(), which cleans all
1118 * managed mappings in the process' vmspace pmap, no work will be left
1119 * for pmap_remove(min, max).
1122 exec_free_abi_mappings(struct proc *p)
1124 struct vmspace *vmspace;
1125 struct sysentvec *sv;
1127 vmspace = p->p_vmspace;
1128 if (refcount_load(&vmspace->vm_refcnt) != 1)
1132 if (sv->sv_shared_page_obj == NULL)
1135 pmap_remove(vmspace_pmap(vmspace), sv->sv_shared_page_base,
1136 sv->sv_shared_page_base + sv->sv_shared_page_len);
1140 * Run down the current address space and install a new one. Map the shared
1144 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1147 struct proc *p = imgp->proc;
1148 struct vmspace *vmspace = p->p_vmspace;
1149 struct thread *td = curthread;
1151 vm_offset_t sv_minuser;
1154 imgp->vmspace_destroyed = true;
1157 if (p->p_sysent->sv_onexec_old != NULL)
1158 p->p_sysent->sv_onexec_old(td);
1161 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1164 * Blow away entire process VM, if address space not shared,
1165 * otherwise, create a new VM space so that other threads are
1168 map = &vmspace->vm_map;
1170 sv_minuser = sv->sv_minuser;
1172 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1173 if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1174 vm_map_min(map) == sv_minuser &&
1175 vm_map_max(map) == sv->sv_maxuser &&
1176 cpu_exec_vmspace_reuse(p, map)) {
1177 exec_free_abi_mappings(p);
1179 pmap_remove_pages(vmspace_pmap(vmspace));
1180 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1182 * An exec terminates mlockall(MCL_FUTURE).
1183 * ASLR and W^X states must be re-evaluated.
1186 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1187 MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1190 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1193 vmspace = p->p_vmspace;
1194 map = &vmspace->vm_map;
1196 map->flags |= imgp->map_flags;
1198 /* Map a shared page */
1199 obj = sv->sv_shared_page_obj;
1201 vm_object_reference(obj);
1202 error = vm_map_fixed(map, obj, 0,
1203 sv->sv_shared_page_base, sv->sv_shared_page_len,
1204 VM_PROT_READ | VM_PROT_EXECUTE,
1205 VM_PROT_READ | VM_PROT_EXECUTE,
1206 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1207 if (error != KERN_SUCCESS) {
1208 vm_object_deallocate(obj);
1209 return (vm_mmap_to_errno(error));
1213 return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1217 * Compute the stack size limit and map the main process stack.
1220 exec_map_stack(struct image_params *imgp)
1222 struct rlimit rlim_stack;
1223 struct sysentvec *sv;
1226 struct vmspace *vmspace;
1227 vm_offset_t stack_addr, stack_top;
1229 int error, find_space, stack_off;
1230 vm_prot_t stack_prot;
1235 if (imgp->stack_sz != 0) {
1236 ssiz = trunc_page(imgp->stack_sz);
1238 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1240 if (ssiz > rlim_stack.rlim_max)
1241 ssiz = rlim_stack.rlim_max;
1242 if (ssiz > rlim_stack.rlim_cur) {
1243 rlim_stack.rlim_cur = ssiz;
1244 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1246 } else if (sv->sv_maxssiz != NULL) {
1247 ssiz = *sv->sv_maxssiz;
1252 vmspace = p->p_vmspace;
1253 map = &vmspace->vm_map;
1255 stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1256 imgp->stack_prot : sv->sv_stackprot;
1257 if ((map->flags & MAP_ASLR_STACK) != 0) {
1258 stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1259 lim_max(curthread, RLIMIT_DATA));
1260 find_space = VMFS_ANY_SPACE;
1262 stack_addr = sv->sv_usrstack - ssiz;
1263 find_space = VMFS_NO_SPACE;
1265 error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1266 sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1267 MAP_STACK_GROWS_DOWN);
1268 if (error != KERN_SUCCESS) {
1269 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1270 "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1271 stack_prot, error, vm_mmap_to_errno(error));
1272 return (vm_mmap_to_errno(error));
1275 stack_top = stack_addr + ssiz;
1276 if ((map->flags & MAP_ASLR_STACK) != 0) {
1277 /* Randomize within the first page of the stack. */
1278 arc4rand(&stack_off, sizeof(stack_off), 0);
1279 stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1283 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1284 * are still used to enforce the stack rlimit on the process stack.
1286 vmspace->vm_maxsaddr = (char *)stack_addr;
1287 vmspace->vm_stacktop = stack_top;
1288 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1294 * Copy out argument and environment strings from the old process address
1295 * space into the temporary string buffer.
1298 exec_copyin_args(struct image_args *args, const char *fname,
1299 enum uio_seg segflg, char **argv, char **envv)
1304 bzero(args, sizeof(*args));
1309 * Allocate demand-paged memory for the file name, argument, and
1310 * environment strings.
1312 error = exec_alloc_args(args);
1317 * Copy the file name.
1319 error = exec_args_add_fname(args, fname, segflg);
1324 * extract arguments first
1327 error = fueword(argv++, &arg);
1334 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1341 * extract environment strings
1345 error = fueword(envv++, &env);
1352 error = exec_args_add_env(args,
1353 (char *)(uintptr_t)env, UIO_USERSPACE);
1362 exec_free_args(args);
1367 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1368 const void *data, size_t datalen, const int *fds, size_t fdslen)
1370 struct filedesc *ofdp;
1375 memset(args, '\0', sizeof(*args));
1376 ofdp = td->td_proc->p_fd;
1377 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1379 error = exec_alloc_args(args);
1383 args->begin_argv = args->buf;
1384 args->stringspace = ARG_MAX;
1388 * Argument buffer has been provided. Copy it into the
1389 * kernel as a single string and add a terminating null
1392 error = copyin(data, args->begin_argv, datalen);
1395 args->begin_argv[datalen] = '\0';
1396 args->endp = args->begin_argv + datalen + 1;
1397 args->stringspace -= datalen + 1;
1400 * Traditional argument counting. Count the number of
1403 for (p = args->begin_argv; p < args->endp; ++p)
1407 /* No argument buffer provided. */
1408 args->endp = args->begin_argv;
1411 /* Create new file descriptor table. */
1412 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1413 error = copyin(fds, kfds, fdslen * sizeof(int));
1418 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1425 exec_free_args(args);
1429 struct exec_args_kva {
1432 SLIST_ENTRY(exec_args_kva) next;
1435 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1437 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1438 static struct mtx exec_args_kva_mtx;
1439 static u_int exec_args_gen;
1442 exec_prealloc_args_kva(void *arg __unused)
1444 struct exec_args_kva *argkva;
1447 SLIST_INIT(&exec_args_kva_freelist);
1448 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1449 for (i = 0; i < exec_map_entries; i++) {
1450 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1451 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1452 argkva->gen = exec_args_gen;
1453 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1456 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1459 exec_alloc_args_kva(void **cookie)
1461 struct exec_args_kva *argkva;
1463 argkva = (void *)atomic_readandclear_ptr(
1464 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1465 if (argkva == NULL) {
1466 mtx_lock(&exec_args_kva_mtx);
1467 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1468 (void)mtx_sleep(&exec_args_kva_freelist,
1469 &exec_args_kva_mtx, 0, "execkva", 0);
1470 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1471 mtx_unlock(&exec_args_kva_mtx);
1473 kasan_mark((void *)argkva->addr, exec_map_entry_size,
1474 exec_map_entry_size, 0);
1475 *(struct exec_args_kva **)cookie = argkva;
1476 return (argkva->addr);
1480 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1484 base = argkva->addr;
1485 kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1486 KASAN_EXEC_ARGS_FREED);
1487 if (argkva->gen != gen) {
1488 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1492 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1493 (uintptr_t)NULL, (uintptr_t)argkva)) {
1494 mtx_lock(&exec_args_kva_mtx);
1495 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1496 wakeup_one(&exec_args_kva_freelist);
1497 mtx_unlock(&exec_args_kva_mtx);
1502 exec_free_args_kva(void *cookie)
1505 exec_release_args_kva(cookie, exec_args_gen);
1509 exec_args_kva_lowmem(void *arg __unused)
1511 SLIST_HEAD(, exec_args_kva) head;
1512 struct exec_args_kva *argkva;
1516 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1519 * Force an madvise of each KVA range. Any currently allocated ranges
1520 * will have MADV_FREE applied once they are freed.
1523 mtx_lock(&exec_args_kva_mtx);
1524 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1525 mtx_unlock(&exec_args_kva_mtx);
1526 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1527 SLIST_REMOVE_HEAD(&head, next);
1528 exec_release_args_kva(argkva, gen);
1532 argkva = (void *)atomic_readandclear_ptr(
1533 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1535 exec_release_args_kva(argkva, gen);
1538 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1539 EVENTHANDLER_PRI_ANY);
1542 * Allocate temporary demand-paged, zero-filled memory for the file name,
1543 * argument, and environment strings.
1546 exec_alloc_args(struct image_args *args)
1549 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1554 exec_free_args(struct image_args *args)
1557 if (args->buf != NULL) {
1558 exec_free_args_kva(args->bufkva);
1561 if (args->fname_buf != NULL) {
1562 free(args->fname_buf, M_TEMP);
1563 args->fname_buf = NULL;
1565 if (args->fdp != NULL)
1566 fdescfree_remapped(args->fdp);
1570 * A set to functions to fill struct image args.
1572 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1573 * fname) before the other functions. All exec_args_add_arg() calls must
1574 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1575 * may be called any time after exec_args_add_fname().
1577 * exec_args_add_fname() - install path to be executed
1578 * exec_args_add_arg() - append an argument string
1579 * exec_args_add_env() - append an env string
1580 * exec_args_adjust_args() - adjust location of the argument list to
1581 * allow new arguments to be prepended
1584 exec_args_add_fname(struct image_args *args, const char *fname,
1585 enum uio_seg segflg)
1590 KASSERT(args->fname == NULL, ("fname already appended"));
1591 KASSERT(args->endp == NULL, ("already appending to args"));
1593 if (fname != NULL) {
1594 args->fname = args->buf;
1595 error = segflg == UIO_SYSSPACE ?
1596 copystr(fname, args->fname, PATH_MAX, &length) :
1597 copyinstr(fname, args->fname, PATH_MAX, &length);
1599 return (error == ENAMETOOLONG ? E2BIG : error);
1603 /* Set up for _arg_*()/_env_*() */
1604 args->endp = args->buf + length;
1605 /* begin_argv must be set and kept updated */
1606 args->begin_argv = args->endp;
1607 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1608 ("too little space remaining for arguments %zu < %zu",
1609 exec_map_entry_size - length, (size_t)ARG_MAX));
1610 args->stringspace = ARG_MAX;
1616 exec_args_add_str(struct image_args *args, const char *str,
1617 enum uio_seg segflg, int *countp)
1622 KASSERT(args->endp != NULL, ("endp not initialized"));
1623 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1625 error = (segflg == UIO_SYSSPACE) ?
1626 copystr(str, args->endp, args->stringspace, &length) :
1627 copyinstr(str, args->endp, args->stringspace, &length);
1629 return (error == ENAMETOOLONG ? E2BIG : error);
1630 args->stringspace -= length;
1631 args->endp += length;
1638 exec_args_add_arg(struct image_args *args, const char *argp,
1639 enum uio_seg segflg)
1642 KASSERT(args->envc == 0, ("appending args after env"));
1644 return (exec_args_add_str(args, argp, segflg, &args->argc));
1648 exec_args_add_env(struct image_args *args, const char *envp,
1649 enum uio_seg segflg)
1652 if (args->envc == 0)
1653 args->begin_envv = args->endp;
1655 return (exec_args_add_str(args, envp, segflg, &args->envc));
1659 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1663 KASSERT(args->endp != NULL, ("endp not initialized"));
1664 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1666 offset = extend - consume;
1667 if (args->stringspace < offset)
1669 memmove(args->begin_argv + extend, args->begin_argv + consume,
1670 args->endp - args->begin_argv + consume);
1672 args->begin_envv += offset;
1673 args->endp += offset;
1674 args->stringspace -= offset;
1679 exec_args_get_begin_envv(struct image_args *args)
1682 KASSERT(args->endp != NULL, ("endp not initialized"));
1685 return (args->begin_envv);
1686 return (args->endp);
1690 * Copy strings out to the new process address space, constructing new arg
1691 * and env vector tables. Return a pointer to the base so that it can be used
1692 * as the initial stack pointer.
1695 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1700 uintptr_t destp, ustringp;
1701 struct ps_strings *arginfo;
1703 struct sysentvec *sysent;
1704 size_t execpath_len;
1705 int error, szsigcode;
1706 char canary[sizeof(long) * 8];
1709 sysent = p->p_sysent;
1711 destp = PROC_PS_STRINGS(p);
1712 arginfo = imgp->ps_strings = (void *)destp;
1717 if (sysent->sv_sigcode_base == 0 && sysent->sv_szsigcode != NULL) {
1718 szsigcode = *(sysent->sv_szsigcode);
1720 destp = rounddown2(destp, sizeof(void *));
1721 error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode);
1727 * Copy the image path for the rtld.
1729 if (imgp->execpath != NULL && imgp->auxargs != NULL) {
1730 execpath_len = strlen(imgp->execpath) + 1;
1731 destp -= execpath_len;
1732 destp = rounddown2(destp, sizeof(void *));
1733 imgp->execpathp = (void *)destp;
1734 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1740 * Prepare the canary for SSP.
1742 arc4rand(canary, sizeof(canary), 0);
1743 destp -= sizeof(canary);
1744 imgp->canary = (void *)destp;
1745 error = copyout(canary, imgp->canary, sizeof(canary));
1748 imgp->canarylen = sizeof(canary);
1751 * Prepare the pagesizes array.
1753 imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES;
1754 destp -= imgp->pagesizeslen;
1755 destp = rounddown2(destp, sizeof(void *));
1756 imgp->pagesizes = (void *)destp;
1757 error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen);
1762 * Allocate room for the argument and environment strings.
1764 destp -= ARG_MAX - imgp->args->stringspace;
1765 destp = rounddown2(destp, sizeof(void *));
1768 if (imgp->auxargs) {
1770 * Allocate room on the stack for the ELF auxargs
1771 * array. It has up to AT_COUNT entries.
1773 destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1774 destp = rounddown2(destp, sizeof(void *));
1777 vectp = (char **)destp;
1780 * Allocate room for the argv[] and env vectors including the
1781 * terminating NULL pointers.
1783 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1786 * vectp also becomes our initial stack base
1788 *stack_base = (uintptr_t)vectp;
1790 stringp = imgp->args->begin_argv;
1791 argc = imgp->args->argc;
1792 envc = imgp->args->envc;
1795 * Copy out strings - arguments and environment.
1797 error = copyout(stringp, (void *)ustringp,
1798 ARG_MAX - imgp->args->stringspace);
1803 * Fill in "ps_strings" struct for ps, w, etc.
1806 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1807 suword32(&arginfo->ps_nargvstr, argc) != 0)
1811 * Fill in argument portion of vector table.
1813 for (; argc > 0; --argc) {
1814 if (suword(vectp++, ustringp) != 0)
1816 while (*stringp++ != 0)
1821 /* a null vector table pointer separates the argp's from the envp's */
1822 if (suword(vectp++, 0) != 0)
1826 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1827 suword32(&arginfo->ps_nenvstr, envc) != 0)
1831 * Fill in environment portion of vector table.
1833 for (; envc > 0; --envc) {
1834 if (suword(vectp++, ustringp) != 0)
1836 while (*stringp++ != 0)
1841 /* end of vector table is a null pointer */
1842 if (suword(vectp, 0) != 0)
1845 if (imgp->auxargs) {
1847 error = imgp->sysent->sv_copyout_auxargs(imgp,
1857 * Check permissions of file to execute.
1858 * Called with imgp->vp locked.
1859 * Return 0 for success or error code on failure.
1862 exec_check_permissions(struct image_params *imgp)
1864 struct vnode *vp = imgp->vp;
1865 struct vattr *attr = imgp->attr;
1871 /* Get file attributes */
1872 error = VOP_GETATTR(vp, attr, td->td_ucred);
1877 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1883 * 1) Check if file execution is disabled for the filesystem that
1884 * this file resides on.
1885 * 2) Ensure that at least one execute bit is on. Otherwise, a
1886 * privileged user will always succeed, and we don't want this
1887 * to happen unless the file really is executable.
1888 * 3) Ensure that the file is a regular file.
1890 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1891 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1892 (attr->va_type != VREG))
1896 * Zero length files can't be exec'd
1898 if (attr->va_size == 0)
1902 * Check for execute permission to file based on current credentials.
1904 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1909 * Check number of open-for-writes on the file and deny execution
1912 * Add a text reference now so no one can write to the
1913 * executable while we're activating it.
1915 * Remember if this was set before and unset it in case this is not
1916 * actually an executable image.
1918 error = VOP_SET_TEXT(vp);
1921 imgp->textset = true;
1924 * Call filesystem specific open routine (which does nothing in the
1927 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1929 imgp->opened = true;
1934 * Exec handler registration
1937 exec_register(const struct execsw *execsw_arg)
1939 const struct execsw **es, **xs, **newexecsw;
1940 u_int count = 2; /* New slot and trailing NULL */
1943 for (es = execsw; *es; es++)
1945 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1948 for (es = execsw; *es; es++)
1953 free(execsw, M_TEMP);
1959 exec_unregister(const struct execsw *execsw_arg)
1961 const struct execsw **es, **xs, **newexecsw;
1965 panic("unregister with no handlers left?\n");
1967 for (es = execsw; *es; es++) {
1968 if (*es == execsw_arg)
1973 for (es = execsw; *es; es++)
1974 if (*es != execsw_arg)
1976 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1978 for (es = execsw; *es; es++)
1979 if (*es != execsw_arg)
1983 free(execsw, M_TEMP);
1989 * Write out a core segment to the compression stream.
1992 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1998 chunk_len = MIN(len, CORE_BUF_SIZE);
2001 * We can get EFAULT error here.
2002 * In that case zero out the current chunk of the segment.
2004 error = copyin(base, buf, chunk_len);
2006 bzero(buf, chunk_len);
2007 error = compressor_write(cp->comp, buf, chunk_len);
2017 core_write(struct coredump_params *cp, const void *base, size_t len,
2018 off_t offset, enum uio_seg seg, size_t *resid)
2021 return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
2022 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
2023 cp->active_cred, cp->file_cred, resid, cp->td));
2027 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
2032 size_t resid, runlen;
2036 KASSERT((uintptr_t)base % PAGE_SIZE == 0,
2037 ("%s: user address %p is not page-aligned", __func__, base));
2039 if (cp->comp != NULL)
2040 return (compress_chunk(cp, base, tmpbuf, len));
2042 map = &cp->td->td_proc->p_vmspace->vm_map;
2043 for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
2045 * Attempt to page in all virtual pages in the range. If a
2046 * virtual page is not backed by the pager, it is represented as
2047 * a hole in the file. This can occur with zero-filled
2048 * anonymous memory or truncated files, for example.
2050 for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
2051 if (core_dump_can_intr && curproc_sigkilled())
2053 error = vm_fault(map, (uintptr_t)base + runlen,
2054 VM_PROT_READ, VM_FAULT_NOFILL, NULL);
2056 success = error == KERN_SUCCESS;
2057 else if ((error == KERN_SUCCESS) != success)
2062 error = core_write(cp, base, runlen, offset,
2063 UIO_USERSPACE, &resid);
2065 if (error != EFAULT)
2069 * EFAULT may be returned if the user mapping
2070 * could not be accessed, e.g., because a mapped
2071 * file has been truncated. Skip the page if no
2072 * progress was made, to protect against a
2073 * hypothetical scenario where vm_fault() was
2074 * successful but core_write() returns EFAULT
2085 error = vn_start_write(cp->vp, &mp, V_WAIT);
2088 vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
2089 error = vn_truncate_locked(cp->vp, offset + runlen,
2090 false, cp->td->td_ucred);
2092 vn_finished_write(mp);
2101 * Drain into a core file.
2104 sbuf_drain_core_output(void *arg, const char *data, int len)
2106 struct coredump_params *cp;
2111 p = cp->td->td_proc;
2114 * Some kern_proc out routines that print to this sbuf may
2115 * call us with the process lock held. Draining with the
2116 * non-sleepable lock held is unsafe. The lock is needed for
2117 * those routines when dumping a live process. In our case we
2118 * can safely release the lock before draining and acquire
2121 locked = PROC_LOCKED(p);
2124 if (cp->comp != NULL)
2125 error = compressor_write(cp->comp, __DECONST(char *, data), len);
2127 error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2128 UIO_SYSSPACE, NULL);