1/* $NetBSD: kern_exec.c,v 1.438 2016/11/03 22:08:30 kamil Exp $ */
2
3/*-
4 * Copyright (c) 2008 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
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.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26 * POSSIBILITY OF SUCH DAMAGE.
27 */
28
29/*-
30 * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
31 * Copyright (C) 1992 Wolfgang Solfrank.
32 * Copyright (C) 1992 TooLs GmbH.
33 * All rights reserved.
34 *
35 * Redistribution and use in source and binary forms, with or without
36 * modification, are permitted provided that the following conditions
37 * are met:
38 * 1. Redistributions of source code must retain the above copyright
39 * notice, this list of conditions and the following disclaimer.
40 * 2. Redistributions in binary form must reproduce the above copyright
41 * notice, this list of conditions and the following disclaimer in the
42 * documentation and/or other materials provided with the distribution.
43 * 3. All advertising materials mentioning features or use of this software
44 * must display the following acknowledgement:
45 * This product includes software developed by TooLs GmbH.
46 * 4. The name of TooLs GmbH may not be used to endorse or promote products
47 * derived from this software without specific prior written permission.
48 *
49 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
50 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
51 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
52 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
53 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
54 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
55 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
56 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
57 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
58 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
59 */
60
61#include <sys/cdefs.h>
62__KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.438 2016/11/03 22:08:30 kamil Exp $");
63
64#include "opt_exec.h"
65#include "opt_execfmt.h"
66#include "opt_ktrace.h"
67#include "opt_modular.h"
68#include "opt_syscall_debug.h"
69#include "veriexec.h"
70#include "opt_pax.h"
71
72#include <sys/param.h>
73#include <sys/systm.h>
74#include <sys/filedesc.h>
75#include <sys/kernel.h>
76#include <sys/proc.h>
77#include <sys/mount.h>
78#include <sys/kmem.h>
79#include <sys/namei.h>
80#include <sys/vnode.h>
81#include <sys/file.h>
82#include <sys/filedesc.h>
83#include <sys/acct.h>
84#include <sys/atomic.h>
85#include <sys/exec.h>
86#include <sys/ktrace.h>
87#include <sys/uidinfo.h>
88#include <sys/wait.h>
89#include <sys/mman.h>
90#include <sys/ras.h>
91#include <sys/signalvar.h>
92#include <sys/stat.h>
93#include <sys/syscall.h>
94#include <sys/kauth.h>
95#include <sys/lwpctl.h>
96#include <sys/pax.h>
97#include <sys/cpu.h>
98#include <sys/module.h>
99#include <sys/syscallvar.h>
100#include <sys/syscallargs.h>
101#if NVERIEXEC > 0
102#include <sys/verified_exec.h>
103#endif /* NVERIEXEC > 0 */
104#include <sys/sdt.h>
105#include <sys/spawn.h>
106#include <sys/prot.h>
107#include <sys/cprng.h>
108
109#include <uvm/uvm_extern.h>
110
111#include <machine/reg.h>
112
113#include <compat/common/compat_util.h>
114
115#ifndef MD_TOPDOWN_INIT
116#ifdef __USE_TOPDOWN_VM
117#define MD_TOPDOWN_INIT(epp) (epp)->ep_flags |= EXEC_TOPDOWN_VM
118#else
119#define MD_TOPDOWN_INIT(epp)
120#endif
121#endif
122
123struct execve_data;
124
125extern int user_va0_disable;
126
127static size_t calcargs(struct execve_data * restrict, const size_t);
128static size_t calcstack(struct execve_data * restrict, const size_t);
129static int copyoutargs(struct execve_data * restrict, struct lwp *,
130 char * const);
131static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
132static int copyinargs(struct execve_data * restrict, char * const *,
133 char * const *, execve_fetch_element_t, char **);
134static int copyinargstrs(struct execve_data * restrict, char * const *,
135 execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
136static int exec_sigcode_map(struct proc *, const struct emul *);
137
138#if defined(DEBUG) && !defined(DEBUG_EXEC)
139#define DEBUG_EXEC
140#endif
141#ifdef DEBUG_EXEC
142#define DPRINTF(a) printf a
143#define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
144 __LINE__, (s), (a), (b))
145static void dump_vmcmds(const struct exec_package * const, size_t, int);
146#define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
147#else
148#define DPRINTF(a)
149#define COPYPRINTF(s, a, b)
150#define DUMPVMCMDS(p, x, e) do {} while (0)
151#endif /* DEBUG_EXEC */
152
153/*
154 * DTrace SDT provider definitions
155 */
156SDT_PROVIDER_DECLARE(proc);
157SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
158SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
159SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
160
161/*
162 * Exec function switch:
163 *
164 * Note that each makecmds function is responsible for loading the
165 * exec package with the necessary functions for any exec-type-specific
166 * handling.
167 *
168 * Functions for specific exec types should be defined in their own
169 * header file.
170 */
171static const struct execsw **execsw = NULL;
172static int nexecs;
173
174u_int exec_maxhdrsz; /* must not be static - used by netbsd32 */
175
176/* list of dynamically loaded execsw entries */
177static LIST_HEAD(execlist_head, exec_entry) ex_head =
178 LIST_HEAD_INITIALIZER(ex_head);
179struct exec_entry {
180 LIST_ENTRY(exec_entry) ex_list;
181 SLIST_ENTRY(exec_entry) ex_slist;
182 const struct execsw *ex_sw;
183};
184
185#ifndef __HAVE_SYSCALL_INTERN
186void syscall(void);
187#endif
188
189/* NetBSD autoloadable syscalls */
190#ifdef MODULAR
191#include <kern/syscalls_autoload.c>
192#endif
193
194/* NetBSD emul struct */
195struct emul emul_netbsd = {
196 .e_name = "netbsd",
197#ifdef EMUL_NATIVEROOT
198 .e_path = EMUL_NATIVEROOT,
199#else
200 .e_path = NULL,
201#endif
202#ifndef __HAVE_MINIMAL_EMUL
203 .e_flags = EMUL_HAS_SYS___syscall,
204 .e_errno = NULL,
205 .e_nosys = SYS_syscall,
206 .e_nsysent = SYS_NSYSENT,
207#endif
208#ifdef MODULAR
209 .e_sc_autoload = netbsd_syscalls_autoload,
210#endif
211 .e_sysent = sysent,
212#ifdef SYSCALL_DEBUG
213 .e_syscallnames = syscallnames,
214#else
215 .e_syscallnames = NULL,
216#endif
217 .e_sendsig = sendsig,
218 .e_trapsignal = trapsignal,
219 .e_tracesig = NULL,
220 .e_sigcode = NULL,
221 .e_esigcode = NULL,
222 .e_sigobject = NULL,
223 .e_setregs = setregs,
224 .e_proc_exec = NULL,
225 .e_proc_fork = NULL,
226 .e_proc_exit = NULL,
227 .e_lwp_fork = NULL,
228 .e_lwp_exit = NULL,
229#ifdef __HAVE_SYSCALL_INTERN
230 .e_syscall_intern = syscall_intern,
231#else
232 .e_syscall = syscall,
233#endif
234 .e_sysctlovly = NULL,
235 .e_fault = NULL,
236 .e_vm_default_addr = uvm_default_mapaddr,
237 .e_usertrap = NULL,
238 .e_ucsize = sizeof(ucontext_t),
239 .e_startlwp = startlwp
240};
241
242/*
243 * Exec lock. Used to control access to execsw[] structures.
244 * This must not be static so that netbsd32 can access it, too.
245 */
246krwlock_t exec_lock;
247
248static kmutex_t sigobject_lock;
249
250/*
251 * Data used between a loadvm and execve part of an "exec" operation
252 */
253struct execve_data {
254 struct exec_package ed_pack;
255 struct pathbuf *ed_pathbuf;
256 struct vattr ed_attr;
257 struct ps_strings ed_arginfo;
258 char *ed_argp;
259 const char *ed_pathstring;
260 char *ed_resolvedpathbuf;
261 size_t ed_ps_strings_sz;
262 int ed_szsigcode;
263 size_t ed_argslen;
264 long ed_argc;
265 long ed_envc;
266};
267
268/*
269 * data passed from parent lwp to child during a posix_spawn()
270 */
271struct spawn_exec_data {
272 struct execve_data sed_exec;
273 struct posix_spawn_file_actions
274 *sed_actions;
275 struct posix_spawnattr *sed_attrs;
276 struct proc *sed_parent;
277 kcondvar_t sed_cv_child_ready;
278 kmutex_t sed_mtx_child;
279 int sed_error;
280 volatile uint32_t sed_refcnt;
281};
282
283static void *
284exec_pool_alloc(struct pool *pp, int flags)
285{
286
287 return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
288 UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
289}
290
291static void
292exec_pool_free(struct pool *pp, void *addr)
293{
294
295 uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
296}
297
298static struct pool exec_pool;
299
300static struct pool_allocator exec_palloc = {
301 .pa_alloc = exec_pool_alloc,
302 .pa_free = exec_pool_free,
303 .pa_pagesz = NCARGS
304};
305
306/*
307 * check exec:
308 * given an "executable" described in the exec package's namei info,
309 * see what we can do with it.
310 *
311 * ON ENTRY:
312 * exec package with appropriate namei info
313 * lwp pointer of exec'ing lwp
314 * NO SELF-LOCKED VNODES
315 *
316 * ON EXIT:
317 * error: nothing held, etc. exec header still allocated.
318 * ok: filled exec package, executable's vnode (unlocked).
319 *
320 * EXEC SWITCH ENTRY:
321 * Locked vnode to check, exec package, proc.
322 *
323 * EXEC SWITCH EXIT:
324 * ok: return 0, filled exec package, executable's vnode (unlocked).
325 * error: destructive:
326 * everything deallocated execept exec header.
327 * non-destructive:
328 * error code, executable's vnode (unlocked),
329 * exec header unmodified.
330 */
331int
332/*ARGSUSED*/
333check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
334{
335 int error, i;
336 struct vnode *vp;
337 struct nameidata nd;
338 size_t resid;
339
340 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
341
342 /* first get the vnode */
343 if ((error = namei(&nd)) != 0)
344 return error;
345 epp->ep_vp = vp = nd.ni_vp;
346 /* normally this can't fail */
347 error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL);
348 KASSERT(error == 0);
349
350#ifdef DIAGNOSTIC
351 /* paranoia (take this out once namei stuff stabilizes) */
352 memset(nd.ni_pnbuf, '~', PATH_MAX);
353#endif
354
355 /* check access and type */
356 if (vp->v_type != VREG) {
357 error = EACCES;
358 goto bad1;
359 }
360 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
361 goto bad1;
362
363 /* get attributes */
364 if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
365 goto bad1;
366
367 /* Check mount point */
368 if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
369 error = EACCES;
370 goto bad1;
371 }
372 if (vp->v_mount->mnt_flag & MNT_NOSUID)
373 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
374
375 /* try to open it */
376 if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
377 goto bad1;
378
379 /* unlock vp, since we need it unlocked from here on out. */
380 VOP_UNLOCK(vp);
381
382#if NVERIEXEC > 0
383 error = veriexec_verify(l, vp, epp->ep_resolvedname,
384 epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
385 NULL);
386 if (error)
387 goto bad2;
388#endif /* NVERIEXEC > 0 */
389
390#ifdef PAX_SEGVGUARD
391 error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
392 if (error)
393 goto bad2;
394#endif /* PAX_SEGVGUARD */
395
396 /* now we have the file, get the exec header */
397 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
398 UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
399 if (error)
400 goto bad2;
401 epp->ep_hdrvalid = epp->ep_hdrlen - resid;
402
403 /*
404 * Set up default address space limits. Can be overridden
405 * by individual exec packages.
406 */
407 epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS);
408 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
409
410 /*
411 * set up the vmcmds for creation of the process
412 * address space
413 */
414 error = ENOEXEC;
415 for (i = 0; i < nexecs; i++) {
416 int newerror;
417
418 epp->ep_esch = execsw[i];
419 newerror = (*execsw[i]->es_makecmds)(l, epp);
420
421 if (!newerror) {
422 /* Seems ok: check that entry point is not too high */
423 if (epp->ep_entry > epp->ep_vm_maxaddr) {
424#ifdef DIAGNOSTIC
425 printf("%s: rejecting %p due to "
426 "too high entry address (> %p)\n",
427 __func__, (void *)epp->ep_entry,
428 (void *)epp->ep_vm_maxaddr);
429#endif
430 error = ENOEXEC;
431 break;
432 }
433 /* Seems ok: check that entry point is not too low */
434 if (epp->ep_entry < epp->ep_vm_minaddr) {
435#ifdef DIAGNOSTIC
436 printf("%s: rejecting %p due to "
437 "too low entry address (< %p)\n",
438 __func__, (void *)epp->ep_entry,
439 (void *)epp->ep_vm_minaddr);
440#endif
441 error = ENOEXEC;
442 break;
443 }
444
445 /* check limits */
446 if ((epp->ep_tsize > MAXTSIZ) ||
447 (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
448 [RLIMIT_DATA].rlim_cur)) {
449#ifdef DIAGNOSTIC
450 printf("%s: rejecting due to "
451 "limits (t=%llu > %llu || d=%llu > %llu)\n",
452 __func__,
453 (unsigned long long)epp->ep_tsize,
454 (unsigned long long)MAXTSIZ,
455 (unsigned long long)epp->ep_dsize,
456 (unsigned long long)
457 l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
458#endif
459 error = ENOMEM;
460 break;
461 }
462 return 0;
463 }
464
465 /*
466 * Reset all the fields that may have been modified by the
467 * loader.
468 */
469 KASSERT(epp->ep_emul_arg == NULL);
470 if (epp->ep_emul_root != NULL) {
471 vrele(epp->ep_emul_root);
472 epp->ep_emul_root = NULL;
473 }
474 if (epp->ep_interp != NULL) {
475 vrele(epp->ep_interp);
476 epp->ep_interp = NULL;
477 }
478 epp->ep_pax_flags = 0;
479
480 /* make sure the first "interesting" error code is saved. */
481 if (error == ENOEXEC)
482 error = newerror;
483
484 if (epp->ep_flags & EXEC_DESTR)
485 /* Error from "#!" code, tidied up by recursive call */
486 return error;
487 }
488
489 /* not found, error */
490
491 /*
492 * free any vmspace-creation commands,
493 * and release their references
494 */
495 kill_vmcmds(&epp->ep_vmcmds);
496
497bad2:
498 /*
499 * close and release the vnode, restore the old one, free the
500 * pathname buf, and punt.
501 */
502 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
503 VOP_CLOSE(vp, FREAD, l->l_cred);
504 vput(vp);
505 return error;
506
507bad1:
508 /*
509 * free the namei pathname buffer, and put the vnode
510 * (which we don't yet have open).
511 */
512 vput(vp); /* was still locked */
513 return error;
514}
515
516#ifdef __MACHINE_STACK_GROWS_UP
517#define STACK_PTHREADSPACE NBPG
518#else
519#define STACK_PTHREADSPACE 0
520#endif
521
522static int
523execve_fetch_element(char * const *array, size_t index, char **value)
524{
525 return copyin(array + index, value, sizeof(*value));
526}
527
528/*
529 * exec system call
530 */
531int
532sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
533{
534 /* {
535 syscallarg(const char *) path;
536 syscallarg(char * const *) argp;
537 syscallarg(char * const *) envp;
538 } */
539
540 return execve1(l, SCARG(uap, path), SCARG(uap, argp),
541 SCARG(uap, envp), execve_fetch_element);
542}
543
544int
545sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
546 register_t *retval)
547{
548 /* {
549 syscallarg(int) fd;
550 syscallarg(char * const *) argp;
551 syscallarg(char * const *) envp;
552 } */
553
554 return ENOSYS;
555}
556
557/*
558 * Load modules to try and execute an image that we do not understand.
559 * If no execsw entries are present, we load those likely to be needed
560 * in order to run native images only. Otherwise, we autoload all
561 * possible modules that could let us run the binary. XXX lame
562 */
563static void
564exec_autoload(void)
565{
566#ifdef MODULAR
567 static const char * const native[] = {
568 "exec_elf32",
569 "exec_elf64",
570 "exec_script",
571 NULL
572 };
573 static const char * const compat[] = {
574 "exec_elf32",
575 "exec_elf64",
576 "exec_script",
577 "exec_aout",
578 "exec_coff",
579 "exec_ecoff",
580 "compat_aoutm68k",
581 "compat_freebsd",
582 "compat_ibcs2",
583 "compat_linux",
584 "compat_linux32",
585 "compat_netbsd32",
586 "compat_sunos",
587 "compat_sunos32",
588 "compat_svr4",
589 "compat_svr4_32",
590 "compat_ultrix",
591 NULL
592 };
593 char const * const *list;
594 int i;
595
596 list = (nexecs == 0 ? native : compat);
597 for (i = 0; list[i] != NULL; i++) {
598 if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
599 continue;
600 }
601 yield();
602 }
603#endif
604}
605
606static int
607makepathbuf(struct lwp *l, const char *upath, struct pathbuf **pbp,
608 size_t *offs)
609{
610 char *path, *bp;
611 size_t len, tlen;
612 int error;
613 struct cwdinfo *cwdi;
614
615 path = PNBUF_GET();
616 error = copyinstr(upath, path, MAXPATHLEN, &len);
617 if (error) {
618 PNBUF_PUT(path);
619 DPRINTF(("%s: copyin path @%p %d\n", __func__, upath, error));
620 return error;
621 }
622
623 if (path[0] == '/') {
624 *offs = 0;
625 goto out;
626 }
627
628 len++;
629 if (len + 1 >= MAXPATHLEN)
630 goto out;
631 bp = path + MAXPATHLEN - len;
632 memmove(bp, path, len);
633 *(--bp) = '/';
634
635 cwdi = l->l_proc->p_cwdi;
636 rw_enter(&cwdi->cwdi_lock, RW_READER);
637 error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
638 GETCWD_CHECK_ACCESS, l);
639 rw_exit(&cwdi->cwdi_lock);
640
641 if (error) {
642 DPRINTF(("%s: getcwd_common path %s %d\n", __func__, path,
643 error));
644 goto out;
645 }
646 tlen = path + MAXPATHLEN - bp;
647
648 memmove(path, bp, tlen);
649 path[tlen] = '\0';
650 *offs = tlen - len;
651out:
652 *pbp = pathbuf_assimilate(path);
653 return 0;
654}
655
656vaddr_t
657exec_vm_minaddr(vaddr_t va_min)
658{
659 /*
660 * Increase va_min if we don't want NULL to be mappable by the
661 * process.
662 */
663#define VM_MIN_GUARD PAGE_SIZE
664 if (user_va0_disable && (va_min < VM_MIN_GUARD))
665 return VM_MIN_GUARD;
666 return va_min;
667}
668
669static int
670execve_loadvm(struct lwp *l, const char *path, char * const *args,
671 char * const *envs, execve_fetch_element_t fetch_element,
672 struct execve_data * restrict data)
673{
674 struct exec_package * const epp = &data->ed_pack;
675 int error;
676 struct proc *p;
677 char *dp;
678 u_int modgen;
679 size_t offs = 0; // XXX: GCC
680
681 KASSERT(data != NULL);
682
683 p = l->l_proc;
684 modgen = 0;
685
686 SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
687
688 /*
689 * Check if we have exceeded our number of processes limit.
690 * This is so that we handle the case where a root daemon
691 * forked, ran setuid to become the desired user and is trying
692 * to exec. The obvious place to do the reference counting check
693 * is setuid(), but we don't do the reference counting check there
694 * like other OS's do because then all the programs that use setuid()
695 * must be modified to check the return code of setuid() and exit().
696 * It is dangerous to make setuid() fail, because it fails open and
697 * the program will continue to run as root. If we make it succeed
698 * and return an error code, again we are not enforcing the limit.
699 * The best place to enforce the limit is here, when the process tries
700 * to execute a new image, because eventually the process will need
701 * to call exec in order to do something useful.
702 */
703 retry:
704 if (p->p_flag & PK_SUGID) {
705 if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
706 p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
707 &p->p_rlimit[RLIMIT_NPROC],
708 KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
709 chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
710 p->p_rlimit[RLIMIT_NPROC].rlim_cur)
711 return EAGAIN;
712 }
713
714 /*
715 * Drain existing references and forbid new ones. The process
716 * should be left alone until we're done here. This is necessary
717 * to avoid race conditions - e.g. in ptrace() - that might allow
718 * a local user to illicitly obtain elevated privileges.
719 */
720 rw_enter(&p->p_reflock, RW_WRITER);
721
722 /*
723 * Init the namei data to point the file user's program name.
724 * This is done here rather than in check_exec(), so that it's
725 * possible to override this settings if any of makecmd/probe
726 * functions call check_exec() recursively - for example,
727 * see exec_script_makecmds().
728 */
729 if ((error = makepathbuf(l, path, &data->ed_pathbuf, &offs)) != 0)
730 goto clrflg;
731 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
732 data->ed_resolvedpathbuf = PNBUF_GET();
733
734 /*
735 * initialize the fields of the exec package.
736 */
737 epp->ep_kname = data->ed_pathstring + offs;
738 epp->ep_resolvedname = data->ed_resolvedpathbuf;
739 epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
740 epp->ep_hdrlen = exec_maxhdrsz;
741 epp->ep_hdrvalid = 0;
742 epp->ep_emul_arg = NULL;
743 epp->ep_emul_arg_free = NULL;
744 memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
745 epp->ep_vap = &data->ed_attr;
746 epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
747 MD_TOPDOWN_INIT(epp);
748 epp->ep_emul_root = NULL;
749 epp->ep_interp = NULL;
750 epp->ep_esch = NULL;
751 epp->ep_pax_flags = 0;
752 memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
753
754 rw_enter(&exec_lock, RW_READER);
755
756 /* see if we can run it. */
757 if ((error = check_exec(l, epp, data->ed_pathbuf)) != 0) {
758 if (error != ENOENT && error != EACCES) {
759 DPRINTF(("%s: check exec failed %d\n",
760 __func__, error));
761 }
762 goto freehdr;
763 }
764
765 /* allocate an argument buffer */
766 data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
767 KASSERT(data->ed_argp != NULL);
768 dp = data->ed_argp;
769
770 if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
771 goto bad;
772 }
773
774 /*
775 * Calculate the new stack size.
776 */
777
778#ifdef __MACHINE_STACK_GROWS_UP
779/*
780 * copyargs() fills argc/argv/envp from the lower address even on
781 * __MACHINE_STACK_GROWS_UP machines. Reserve a few words just below the SP
782 * so that _rtld() use it.
783 */
784#define RTLD_GAP 32
785#else
786#define RTLD_GAP 0
787#endif
788
789 const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
790
791 data->ed_argslen = calcargs(data, argenvstrlen);
792
793 const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
794
795 if (len > epp->ep_ssize) {
796 /* in effect, compare to initial limit */
797 DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
798 error = ENOMEM;
799 goto bad;
800 }
801 /* adjust "active stack depth" for process VSZ */
802 epp->ep_ssize = len;
803
804 return 0;
805
806 bad:
807 /* free the vmspace-creation commands, and release their references */
808 kill_vmcmds(&epp->ep_vmcmds);
809 /* kill any opened file descriptor, if necessary */
810 if (epp->ep_flags & EXEC_HASFD) {
811 epp->ep_flags &= ~EXEC_HASFD;
812 fd_close(epp->ep_fd);
813 }
814 /* close and put the exec'd file */
815 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
816 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
817 vput(epp->ep_vp);
818 pool_put(&exec_pool, data->ed_argp);
819
820 freehdr:
821 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
822 if (epp->ep_emul_root != NULL)
823 vrele(epp->ep_emul_root);
824 if (epp->ep_interp != NULL)
825 vrele(epp->ep_interp);
826
827 rw_exit(&exec_lock);
828
829 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
830 pathbuf_destroy(data->ed_pathbuf);
831 PNBUF_PUT(data->ed_resolvedpathbuf);
832
833 clrflg:
834 rw_exit(&p->p_reflock);
835
836 if (modgen != module_gen && error == ENOEXEC) {
837 modgen = module_gen;
838 exec_autoload();
839 goto retry;
840 }
841
842 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
843 return error;
844}
845
846static int
847execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
848{
849 struct exec_package * const epp = &data->ed_pack;
850 struct proc *p = l->l_proc;
851 struct exec_vmcmd *base_vcp;
852 int error = 0;
853 size_t i;
854
855 /* record proc's vnode, for use by procfs and others */
856 if (p->p_textvp)
857 vrele(p->p_textvp);
858 vref(epp->ep_vp);
859 p->p_textvp = epp->ep_vp;
860
861 /* create the new process's VM space by running the vmcmds */
862 KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
863
864#ifdef TRACE_EXEC
865 DUMPVMCMDS(epp, 0, 0);
866#endif
867
868 base_vcp = NULL;
869
870 for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
871 struct exec_vmcmd *vcp;
872
873 vcp = &epp->ep_vmcmds.evs_cmds[i];
874 if (vcp->ev_flags & VMCMD_RELATIVE) {
875 KASSERTMSG(base_vcp != NULL,
876 "%s: relative vmcmd with no base", __func__);
877 KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
878 "%s: illegal base & relative vmcmd", __func__);
879 vcp->ev_addr += base_vcp->ev_addr;
880 }
881 error = (*vcp->ev_proc)(l, vcp);
882 if (error)
883 DUMPVMCMDS(epp, i, error);
884 if (vcp->ev_flags & VMCMD_BASE)
885 base_vcp = vcp;
886 }
887
888 /* free the vmspace-creation commands, and release their references */
889 kill_vmcmds(&epp->ep_vmcmds);
890
891 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
892 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
893 vput(epp->ep_vp);
894
895 /* if an error happened, deallocate and punt */
896 if (error != 0) {
897 DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
898 }
899 return error;
900}
901
902static void
903execve_free_data(struct execve_data *data)
904{
905 struct exec_package * const epp = &data->ed_pack;
906
907 /* free the vmspace-creation commands, and release their references */
908 kill_vmcmds(&epp->ep_vmcmds);
909 /* kill any opened file descriptor, if necessary */
910 if (epp->ep_flags & EXEC_HASFD) {
911 epp->ep_flags &= ~EXEC_HASFD;
912 fd_close(epp->ep_fd);
913 }
914
915 /* close and put the exec'd file */
916 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
917 VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
918 vput(epp->ep_vp);
919 pool_put(&exec_pool, data->ed_argp);
920
921 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
922 if (epp->ep_emul_root != NULL)
923 vrele(epp->ep_emul_root);
924 if (epp->ep_interp != NULL)
925 vrele(epp->ep_interp);
926
927 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
928 pathbuf_destroy(data->ed_pathbuf);
929 PNBUF_PUT(data->ed_resolvedpathbuf);
930}
931
932static void
933pathexec(struct exec_package *epp, struct lwp *l, const char *pathstring)
934{
935 const char *commandname;
936 size_t commandlen;
937 char *path;
938 struct proc *p = l->l_proc;
939
940 /* set command name & other accounting info */
941 commandname = strrchr(epp->ep_resolvedname, '/');
942 if (commandname != NULL) {
943 commandname++;
944 } else {
945 commandname = epp->ep_resolvedname;
946 }
947 commandlen = min(strlen(commandname), MAXCOMLEN);
948 (void)memcpy(p->p_comm, commandname, commandlen);
949 p->p_comm[commandlen] = '\0';
950
951
952 /*
953 * If the path starts with /, we don't need to do any work.
954 * This handles the majority of the cases.
955 * In the future perhaps we could canonicalize it?
956 */
957 path = PNBUF_GET();
958 if (pathstring[0] == '/') {
959 (void)strlcpy(path, pathstring, MAXPATHLEN);
960 epp->ep_path = path;
961 }
962#ifdef notyet
963 /*
964 * Although this works most of the time [since the entry was just
965 * entered in the cache] we don't use it because it will fail for
966 * entries that are not placed in the cache because their name is
967 * longer than NCHNAMLEN and it is not the cleanest interface,
968 * because there could be races. When the namei cache is re-written,
969 * this can be changed to use the appropriate function.
970 */
971 else if (!(error = vnode_to_path(path, MAXPATHLEN, p->p_textvp, l, p)))
972 epp->ep_path = path;
973#endif
974 else {
975#ifdef notyet
976 printf("Cannot get path for pid %d [%s] (error %d)\n",
977 (int)p->p_pid, p->p_comm, error);
978#endif
979 PNBUF_PUT(path);
980 epp->ep_path = NULL;
981 }
982}
983
984/* XXX elsewhere */
985static int
986credexec(struct lwp *l, struct vattr *attr)
987{
988 struct proc *p = l->l_proc;
989 int error;
990
991 /*
992 * Deal with set[ug]id. MNT_NOSUID has already been used to disable
993 * s[ug]id. It's OK to check for PSL_TRACED here as we have blocked
994 * out additional references on the process for the moment.
995 */
996 if ((p->p_slflag & PSL_TRACED) == 0 &&
997
998 (((attr->va_mode & S_ISUID) != 0 &&
999 kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
1000
1001 ((attr->va_mode & S_ISGID) != 0 &&
1002 kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
1003 /*
1004 * Mark the process as SUGID before we do
1005 * anything that might block.
1006 */
1007 proc_crmod_enter();
1008 proc_crmod_leave(NULL, NULL, true);
1009
1010 /* Make sure file descriptors 0..2 are in use. */
1011 if ((error = fd_checkstd()) != 0) {
1012 DPRINTF(("%s: fdcheckstd failed %d\n",
1013 __func__, error));
1014 return error;
1015 }
1016
1017 /*
1018 * Copy the credential so other references don't see our
1019 * changes.
1020 */
1021 l->l_cred = kauth_cred_copy(l->l_cred);
1022#ifdef KTRACE
1023 /*
1024 * If the persistent trace flag isn't set, turn off.
1025 */
1026 if (p->p_tracep) {
1027 mutex_enter(&ktrace_lock);
1028 if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1029 ktrderef(p);
1030 mutex_exit(&ktrace_lock);
1031 }
1032#endif
1033 if (attr->va_mode & S_ISUID)
1034 kauth_cred_seteuid(l->l_cred, attr->va_uid);
1035 if (attr->va_mode & S_ISGID)
1036 kauth_cred_setegid(l->l_cred, attr->va_gid);
1037 } else {
1038 if (kauth_cred_geteuid(l->l_cred) ==
1039 kauth_cred_getuid(l->l_cred) &&
1040 kauth_cred_getegid(l->l_cred) ==
1041 kauth_cred_getgid(l->l_cred))
1042 p->p_flag &= ~PK_SUGID;
1043 }
1044
1045 /*
1046 * Copy the credential so other references don't see our changes.
1047 * Test to see if this is necessary first, since in the common case
1048 * we won't need a private reference.
1049 */
1050 if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1051 kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1052 l->l_cred = kauth_cred_copy(l->l_cred);
1053 kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1054 kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1055 }
1056
1057 /* Update the master credentials. */
1058 if (l->l_cred != p->p_cred) {
1059 kauth_cred_t ocred;
1060
1061 kauth_cred_hold(l->l_cred);
1062 mutex_enter(p->p_lock);
1063 ocred = p->p_cred;
1064 p->p_cred = l->l_cred;
1065 mutex_exit(p->p_lock);
1066 kauth_cred_free(ocred);
1067 }
1068
1069 return 0;
1070}
1071
1072static void
1073emulexec(struct lwp *l, struct exec_package *epp)
1074{
1075 struct proc *p = l->l_proc;
1076
1077 /* The emulation root will usually have been found when we looked
1078 * for the elf interpreter (or similar), if not look now. */
1079 if (epp->ep_esch->es_emul->e_path != NULL &&
1080 epp->ep_emul_root == NULL)
1081 emul_find_root(l, epp);
1082
1083 /* Any old emulation root got removed by fdcloseexec */
1084 rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1085 p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1086 rw_exit(&p->p_cwdi->cwdi_lock);
1087 epp->ep_emul_root = NULL;
1088 if (epp->ep_interp != NULL)
1089 vrele(epp->ep_interp);
1090
1091 /*
1092 * Call emulation specific exec hook. This can setup per-process
1093 * p->p_emuldata or do any other per-process stuff an emulation needs.
1094 *
1095 * If we are executing process of different emulation than the
1096 * original forked process, call e_proc_exit() of the old emulation
1097 * first, then e_proc_exec() of new emulation. If the emulation is
1098 * same, the exec hook code should deallocate any old emulation
1099 * resources held previously by this process.
1100 */
1101 if (p->p_emul && p->p_emul->e_proc_exit
1102 && p->p_emul != epp->ep_esch->es_emul)
1103 (*p->p_emul->e_proc_exit)(p);
1104
1105 /*
1106 * This is now LWP 1.
1107 */
1108 /* XXX elsewhere */
1109 mutex_enter(p->p_lock);
1110 p->p_nlwpid = 1;
1111 l->l_lid = 1;
1112 mutex_exit(p->p_lock);
1113
1114 /*
1115 * Call exec hook. Emulation code may NOT store reference to anything
1116 * from &pack.
1117 */
1118 if (epp->ep_esch->es_emul->e_proc_exec)
1119 (*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1120
1121 /* update p_emul, the old value is no longer needed */
1122 p->p_emul = epp->ep_esch->es_emul;
1123
1124 /* ...and the same for p_execsw */
1125 p->p_execsw = epp->ep_esch;
1126
1127#ifdef __HAVE_SYSCALL_INTERN
1128 (*p->p_emul->e_syscall_intern)(p);
1129#endif
1130 ktremul();
1131}
1132
1133static int
1134execve_runproc(struct lwp *l, struct execve_data * restrict data,
1135 bool no_local_exec_lock, bool is_spawn)
1136{
1137 struct exec_package * const epp = &data->ed_pack;
1138 int error = 0;
1139 struct proc *p;
1140
1141 /*
1142 * In case of a posix_spawn operation, the child doing the exec
1143 * might not hold the reader lock on exec_lock, but the parent
1144 * will do this instead.
1145 */
1146 KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1147 KASSERT(!no_local_exec_lock || is_spawn);
1148 KASSERT(data != NULL);
1149
1150 p = l->l_proc;
1151
1152 /* Get rid of other LWPs. */
1153 if (p->p_nlwps > 1) {
1154 mutex_enter(p->p_lock);
1155 exit_lwps(l);
1156 mutex_exit(p->p_lock);
1157 }
1158 KDASSERT(p->p_nlwps == 1);
1159
1160 /* Destroy any lwpctl info. */
1161 if (p->p_lwpctl != NULL)
1162 lwp_ctl_exit();
1163
1164 /* Remove POSIX timers */
1165 timers_free(p, TIMERS_POSIX);
1166
1167 /* Set the PaX flags. */
1168 pax_set_flags(epp, p);
1169
1170 /*
1171 * Do whatever is necessary to prepare the address space
1172 * for remapping. Note that this might replace the current
1173 * vmspace with another!
1174 */
1175 if (is_spawn)
1176 uvmspace_spawn(l, epp->ep_vm_minaddr,
1177 epp->ep_vm_maxaddr,
1178 epp->ep_flags & EXEC_TOPDOWN_VM);
1179 else
1180 uvmspace_exec(l, epp->ep_vm_minaddr,
1181 epp->ep_vm_maxaddr,
1182 epp->ep_flags & EXEC_TOPDOWN_VM);
1183
1184 struct vmspace *vm;
1185 vm = p->p_vmspace;
1186 vm->vm_taddr = (void *)epp->ep_taddr;
1187 vm->vm_tsize = btoc(epp->ep_tsize);
1188 vm->vm_daddr = (void*)epp->ep_daddr;
1189 vm->vm_dsize = btoc(epp->ep_dsize);
1190 vm->vm_ssize = btoc(epp->ep_ssize);
1191 vm->vm_issize = 0;
1192 vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1193 vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1194
1195 pax_aslr_init_vm(l, vm, epp);
1196
1197 /* Now map address space. */
1198 error = execve_dovmcmds(l, data);
1199 if (error != 0)
1200 goto exec_abort;
1201
1202 pathexec(epp, l, data->ed_pathstring);
1203
1204 char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1205
1206 error = copyoutargs(data, l, newstack);
1207 if (error != 0)
1208 goto exec_abort;
1209
1210 cwdexec(p);
1211 fd_closeexec(); /* handle close on exec */
1212
1213 if (__predict_false(ktrace_on))
1214 fd_ktrexecfd();
1215
1216 execsigs(p); /* reset caught signals */
1217
1218 mutex_enter(p->p_lock);
1219 l->l_ctxlink = NULL; /* reset ucontext link */
1220 p->p_acflag &= ~AFORK;
1221 p->p_flag |= PK_EXEC;
1222 mutex_exit(p->p_lock);
1223
1224 /*
1225 * Stop profiling.
1226 */
1227 if ((p->p_stflag & PST_PROFIL) != 0) {
1228 mutex_spin_enter(&p->p_stmutex);
1229 stopprofclock(p);
1230 mutex_spin_exit(&p->p_stmutex);
1231 }
1232
1233 /*
1234 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1235 * exited and exec()/exit() are the only places it will be cleared.
1236 */
1237 if ((p->p_lflag & PL_PPWAIT) != 0) {
1238#if 0
1239 lwp_t *lp;
1240
1241 mutex_enter(proc_lock);
1242 lp = p->p_vforklwp;
1243 p->p_vforklwp = NULL;
1244
1245 l->l_lwpctl = NULL; /* was on loan from blocked parent */
1246 p->p_lflag &= ~PL_PPWAIT;
1247
1248 lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
1249 cv_broadcast(&lp->l_waitcv);
1250 mutex_exit(proc_lock);
1251#else
1252 mutex_enter(proc_lock);
1253 l->l_lwpctl = NULL; /* was on loan from blocked parent */
1254 p->p_lflag &= ~PL_PPWAIT;
1255 cv_broadcast(&p->p_pptr->p_waitcv);
1256 mutex_exit(proc_lock);
1257#endif
1258 }
1259
1260 error = credexec(l, &data->ed_attr);
1261 if (error)
1262 goto exec_abort;
1263
1264#if defined(__HAVE_RAS)
1265 /*
1266 * Remove all RASs from the address space.
1267 */
1268 ras_purgeall();
1269#endif
1270
1271 doexechooks(p);
1272
1273 /*
1274 * Set initial SP at the top of the stack.
1275 *
1276 * Note that on machines where stack grows up (e.g. hppa), SP points to
1277 * the end of arg/env strings. Userland guesses the address of argc
1278 * via ps_strings::ps_argvstr.
1279 */
1280
1281 /* Setup new registers and do misc. setup. */
1282 (*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1283 if (epp->ep_esch->es_setregs)
1284 (*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1285
1286 /* Provide a consistent LWP private setting */
1287 (void)lwp_setprivate(l, NULL);
1288
1289 /* Discard all PCU state; need to start fresh */
1290 pcu_discard_all(l);
1291
1292 /* map the process's signal trampoline code */
1293 if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1294 DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1295 goto exec_abort;
1296 }
1297
1298 pool_put(&exec_pool, data->ed_argp);
1299
1300 /* notify others that we exec'd */
1301 KNOTE(&p->p_klist, NOTE_EXEC);
1302
1303 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1304
1305 SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1306
1307 emulexec(l, epp);
1308
1309 /* Allow new references from the debugger/procfs. */
1310 rw_exit(&p->p_reflock);
1311 if (!no_local_exec_lock)
1312 rw_exit(&exec_lock);
1313
1314 mutex_enter(proc_lock);
1315
1316 if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1317 ksiginfo_t ksi;
1318
1319 KSI_INIT_EMPTY(&ksi);
1320 ksi.ksi_signo = SIGTRAP;
1321 ksi.ksi_lid = l->l_lid;
1322 kpsignal(p, &ksi, NULL);
1323 }
1324
1325 if (p->p_sflag & PS_STOPEXEC) {
1326 ksiginfoq_t kq;
1327
1328 KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1329 p->p_pptr->p_nstopchild++;
1330 p->p_waited = 0;
1331 mutex_enter(p->p_lock);
1332 ksiginfo_queue_init(&kq);
1333 sigclearall(p, &contsigmask, &kq);
1334 lwp_lock(l);
1335 l->l_stat = LSSTOP;
1336 p->p_stat = SSTOP;
1337 p->p_nrlwps--;
1338 lwp_unlock(l);
1339 mutex_exit(p->p_lock);
1340 mutex_exit(proc_lock);
1341 lwp_lock(l);
1342 mi_switch(l);
1343 ksiginfo_queue_drain(&kq);
1344 KERNEL_LOCK(l->l_biglocks, l);
1345 } else {
1346 mutex_exit(proc_lock);
1347 }
1348
1349 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1350 pathbuf_destroy(data->ed_pathbuf);
1351 PNBUF_PUT(data->ed_resolvedpathbuf);
1352#ifdef TRACE_EXEC
1353 DPRINTF(("%s finished\n", __func__));
1354#endif
1355 return EJUSTRETURN;
1356
1357 exec_abort:
1358 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1359 rw_exit(&p->p_reflock);
1360 if (!no_local_exec_lock)
1361 rw_exit(&exec_lock);
1362
1363 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1364 pathbuf_destroy(data->ed_pathbuf);
1365 PNBUF_PUT(data->ed_resolvedpathbuf);
1366
1367 /*
1368 * the old process doesn't exist anymore. exit gracefully.
1369 * get rid of the (new) address space we have created, if any, get rid
1370 * of our namei data and vnode, and exit noting failure
1371 */
1372 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1373 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1374
1375 exec_free_emul_arg(epp);
1376 pool_put(&exec_pool, data->ed_argp);
1377 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1378 if (epp->ep_emul_root != NULL)
1379 vrele(epp->ep_emul_root);
1380 if (epp->ep_interp != NULL)
1381 vrele(epp->ep_interp);
1382
1383 /* Acquire the sched-state mutex (exit1() will release it). */
1384 if (!is_spawn) {
1385 mutex_enter(p->p_lock);
1386 exit1(l, error, SIGABRT);
1387 }
1388
1389 return error;
1390}
1391
1392int
1393execve1(struct lwp *l, const char *path, char * const *args,
1394 char * const *envs, execve_fetch_element_t fetch_element)
1395{
1396 struct execve_data data;
1397 int error;
1398
1399 error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1400 if (error)
1401 return error;
1402 error = execve_runproc(l, &data, false, false);
1403 return error;
1404}
1405
1406static size_t
1407fromptrsz(const struct exec_package *epp)
1408{
1409 return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1410}
1411
1412static size_t
1413ptrsz(const struct exec_package *epp)
1414{
1415 return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1416}
1417
1418static size_t
1419calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1420{
1421 struct exec_package * const epp = &data->ed_pack;
1422
1423 const size_t nargenvptrs =
1424 1 + /* long argc */
1425 data->ed_argc + /* char *argv[] */
1426 1 + /* \0 */
1427 data->ed_envc + /* char *env[] */
1428 1 + /* \0 */
1429 epp->ep_esch->es_arglen; /* auxinfo */
1430
1431 return (nargenvptrs * ptrsz(epp)) + argenvstrlen;
1432}
1433
1434static size_t
1435calcstack(struct execve_data * restrict data, const size_t gaplen)
1436{
1437 struct exec_package * const epp = &data->ed_pack;
1438
1439 data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1440 epp->ep_esch->es_emul->e_sigcode;
1441
1442 data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1443 sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1444
1445 const size_t sigcode_psstr_sz =
1446 data->ed_szsigcode + /* sigcode */
1447 data->ed_ps_strings_sz + /* ps_strings */
1448 STACK_PTHREADSPACE; /* pthread space */
1449
1450 const size_t stacklen =
1451 data->ed_argslen +
1452 gaplen +
1453 sigcode_psstr_sz;
1454
1455 /* make the stack "safely" aligned */
1456 return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1457}
1458
1459static int
1460copyoutargs(struct execve_data * restrict data, struct lwp *l,
1461 char * const newstack)
1462{
1463 struct exec_package * const epp = &data->ed_pack;
1464 struct proc *p = l->l_proc;
1465 int error;
1466
1467 /* remember information about the process */
1468 data->ed_arginfo.ps_nargvstr = data->ed_argc;
1469 data->ed_arginfo.ps_nenvstr = data->ed_envc;
1470
1471 /*
1472 * Allocate the stack address passed to the newly execve()'ed process.
1473 *
1474 * The new stack address will be set to the SP (stack pointer) register
1475 * in setregs().
1476 */
1477
1478 char *newargs = STACK_ALLOC(
1479 STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1480
1481 error = (*epp->ep_esch->es_copyargs)(l, epp,
1482 &data->ed_arginfo, &newargs, data->ed_argp);
1483
1484 if (epp->ep_path) {
1485 PNBUF_PUT(epp->ep_path);
1486 epp->ep_path = NULL;
1487 }
1488 if (error) {
1489 DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1490 return error;
1491 }
1492
1493 error = copyoutpsstrs(data, p);
1494 if (error != 0)
1495 return error;
1496
1497 return 0;
1498}
1499
1500static int
1501copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1502{
1503 struct exec_package * const epp = &data->ed_pack;
1504 struct ps_strings32 arginfo32;
1505 void *aip;
1506 int error;
1507
1508 /* fill process ps_strings info */
1509 p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1510 STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1511
1512 if (epp->ep_flags & EXEC_32) {
1513 aip = &arginfo32;
1514 arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1515 arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1516 arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1517 arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1518 } else
1519 aip = &data->ed_arginfo;
1520
1521 /* copy out the process's ps_strings structure */
1522 if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1523 != 0) {
1524 DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1525 __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1526 return error;
1527 }
1528
1529 return 0;
1530}
1531
1532static int
1533copyinargs(struct execve_data * restrict data, char * const *args,
1534 char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1535{
1536 struct exec_package * const epp = &data->ed_pack;
1537 char *dp;
1538 size_t i;
1539 int error;
1540
1541 dp = *dpp;
1542
1543 data->ed_argc = 0;
1544
1545 /* copy the fake args list, if there's one, freeing it as we go */
1546 if (epp->ep_flags & EXEC_HASARGL) {
1547 struct exec_fakearg *fa = epp->ep_fa;
1548
1549 while (fa->fa_arg != NULL) {
1550 const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1551 size_t len;
1552
1553 len = strlcpy(dp, fa->fa_arg, maxlen);
1554 /* Count NUL into len. */
1555 if (len < maxlen)
1556 len++;
1557 else {
1558 while (fa->fa_arg != NULL) {
1559 kmem_free(fa->fa_arg, fa->fa_len);
1560 fa++;
1561 }
1562 kmem_free(epp->ep_fa, epp->ep_fa_len);
1563 epp->ep_flags &= ~EXEC_HASARGL;
1564 return E2BIG;
1565 }
1566 ktrexecarg(fa->fa_arg, len - 1);
1567 dp += len;
1568
1569 kmem_free(fa->fa_arg, fa->fa_len);
1570 fa++;
1571 data->ed_argc++;
1572 }
1573 kmem_free(epp->ep_fa, epp->ep_fa_len);
1574 epp->ep_flags &= ~EXEC_HASARGL;
1575 }
1576
1577 /*
1578 * Read and count argument strings from user.
1579 */
1580
1581 if (args == NULL) {
1582 DPRINTF(("%s: null args\n", __func__));
1583 return EINVAL;
1584 }
1585 if (epp->ep_flags & EXEC_SKIPARG)
1586 args = (const void *)((const char *)args + fromptrsz(epp));
1587 i = 0;
1588 error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1589 if (error != 0) {
1590 DPRINTF(("%s: copyin arg %d\n", __func__, error));
1591 return error;
1592 }
1593 data->ed_argc += i;
1594
1595 /*
1596 * Read and count environment strings from user.
1597 */
1598
1599 data->ed_envc = 0;
1600 /* environment need not be there */
1601 if (envs == NULL)
1602 goto done;
1603 i = 0;
1604 error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1605 if (error != 0) {
1606 DPRINTF(("%s: copyin env %d\n", __func__, error));
1607 return error;
1608 }
1609 data->ed_envc += i;
1610
1611done:
1612 *dpp = dp;
1613
1614 return 0;
1615}
1616
1617static int
1618copyinargstrs(struct execve_data * restrict data, char * const *strs,
1619 execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1620 void (*ktr)(const void *, size_t))
1621{
1622 char *dp, *sp;
1623 size_t i;
1624 int error;
1625
1626 dp = *dpp;
1627
1628 i = 0;
1629 while (1) {
1630 const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1631 size_t len;
1632
1633 if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1634 return error;
1635 }
1636 if (!sp)
1637 break;
1638 if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1639 if (error == ENAMETOOLONG)
1640 error = E2BIG;
1641 return error;
1642 }
1643 if (__predict_false(ktrace_on))
1644 (*ktr)(dp, len - 1);
1645 dp += len;
1646 i++;
1647 }
1648
1649 *dpp = dp;
1650 *ip = i;
1651
1652 return 0;
1653}
1654
1655/*
1656 * Copy argv and env strings from kernel buffer (argp) to the new stack.
1657 * Those strings are located just after auxinfo.
1658 */
1659int
1660copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1661 char **stackp, void *argp)
1662{
1663 char **cpp, *dp, *sp;
1664 size_t len;
1665 void *nullp;
1666 long argc, envc;
1667 int error;
1668
1669 cpp = (char **)*stackp;
1670 nullp = NULL;
1671 argc = arginfo->ps_nargvstr;
1672 envc = arginfo->ps_nenvstr;
1673
1674 /* argc on stack is long */
1675 CTASSERT(sizeof(*cpp) == sizeof(argc));
1676
1677 dp = (char *)(cpp +
1678 1 + /* long argc */
1679 argc + /* char *argv[] */
1680 1 + /* \0 */
1681 envc + /* char *env[] */
1682 1 + /* \0 */
1683 /* XXX auxinfo multiplied by ptr size? */
1684 pack->ep_esch->es_arglen); /* auxinfo */
1685 sp = argp;
1686
1687 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1688 COPYPRINTF("", cpp - 1, sizeof(argc));
1689 return error;
1690 }
1691
1692 /* XXX don't copy them out, remap them! */
1693 arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1694
1695 for (; --argc >= 0; sp += len, dp += len) {
1696 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1697 COPYPRINTF("", cpp - 1, sizeof(dp));
1698 return error;
1699 }
1700 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1701 COPYPRINTF("str", dp, (size_t)ARG_MAX);
1702 return error;
1703 }
1704 }
1705
1706 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1707 COPYPRINTF("", cpp - 1, sizeof(nullp));
1708 return error;
1709 }
1710
1711 arginfo->ps_envstr = cpp; /* remember location of envp for later */
1712
1713 for (; --envc >= 0; sp += len, dp += len) {
1714 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1715 COPYPRINTF("", cpp - 1, sizeof(dp));
1716 return error;
1717 }
1718 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1719 COPYPRINTF("str", dp, (size_t)ARG_MAX);
1720 return error;
1721 }
1722
1723 }
1724
1725 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1726 COPYPRINTF("", cpp - 1, sizeof(nullp));
1727 return error;
1728 }
1729
1730 *stackp = (char *)cpp;
1731 return 0;
1732}
1733
1734
1735/*
1736 * Add execsw[] entries.
1737 */
1738int
1739exec_add(struct execsw *esp, int count)
1740{
1741 struct exec_entry *it;
1742 int i;
1743
1744 if (count == 0) {
1745 return 0;
1746 }
1747
1748 /* Check for duplicates. */
1749 rw_enter(&exec_lock, RW_WRITER);
1750 for (i = 0; i < count; i++) {
1751 LIST_FOREACH(it, &ex_head, ex_list) {
1752 /* assume unique (makecmds, probe_func, emulation) */
1753 if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1754 it->ex_sw->u.elf_probe_func ==
1755 esp[i].u.elf_probe_func &&
1756 it->ex_sw->es_emul == esp[i].es_emul) {
1757 rw_exit(&exec_lock);
1758 return EEXIST;
1759 }
1760 }
1761 }
1762
1763 /* Allocate new entries. */
1764 for (i = 0; i < count; i++) {
1765 it = kmem_alloc(sizeof(*it), KM_SLEEP);
1766 it->ex_sw = &esp[i];
1767 LIST_INSERT_HEAD(&ex_head, it, ex_list);
1768 }
1769
1770 /* update execsw[] */
1771 exec_init(0);
1772 rw_exit(&exec_lock);
1773 return 0;
1774}
1775
1776/*
1777 * Remove execsw[] entry.
1778 */
1779int
1780exec_remove(struct execsw *esp, int count)
1781{
1782 struct exec_entry *it, *next;
1783 int i;
1784 const struct proclist_desc *pd;
1785 proc_t *p;
1786
1787 if (count == 0) {
1788 return 0;
1789 }
1790
1791 /* Abort if any are busy. */
1792 rw_enter(&exec_lock, RW_WRITER);
1793 for (i = 0; i < count; i++) {
1794 mutex_enter(proc_lock);
1795 for (pd = proclists; pd->pd_list != NULL; pd++) {
1796 PROCLIST_FOREACH(p, pd->pd_list) {
1797 if (p->p_execsw == &esp[i]) {
1798 mutex_exit(proc_lock);
1799 rw_exit(&exec_lock);
1800 return EBUSY;
1801 }
1802 }
1803 }
1804 mutex_exit(proc_lock);
1805 }
1806
1807 /* None are busy, so remove them all. */
1808 for (i = 0; i < count; i++) {
1809 for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1810 next = LIST_NEXT(it, ex_list);
1811 if (it->ex_sw == &esp[i]) {
1812 LIST_REMOVE(it, ex_list);
1813 kmem_free(it, sizeof(*it));
1814 break;
1815 }
1816 }
1817 }
1818
1819 /* update execsw[] */
1820 exec_init(0);
1821 rw_exit(&exec_lock);
1822 return 0;
1823}
1824
1825/*
1826 * Initialize exec structures. If init_boot is true, also does necessary
1827 * one-time initialization (it's called from main() that way).
1828 * Once system is multiuser, this should be called with exec_lock held,
1829 * i.e. via exec_{add|remove}().
1830 */
1831int
1832exec_init(int init_boot)
1833{
1834 const struct execsw **sw;
1835 struct exec_entry *ex;
1836 SLIST_HEAD(,exec_entry) first;
1837 SLIST_HEAD(,exec_entry) any;
1838 SLIST_HEAD(,exec_entry) last;
1839 int i, sz;
1840
1841 if (init_boot) {
1842 /* do one-time initializations */
1843 rw_init(&exec_lock);
1844 mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1845 pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1846 "execargs", &exec_palloc, IPL_NONE);
1847 pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1848 } else {
1849 KASSERT(rw_write_held(&exec_lock));
1850 }
1851
1852 /* Sort each entry onto the appropriate queue. */
1853 SLIST_INIT(&first);
1854 SLIST_INIT(&any);
1855 SLIST_INIT(&last);
1856 sz = 0;
1857 LIST_FOREACH(ex, &ex_head, ex_list) {
1858 switch(ex->ex_sw->es_prio) {
1859 case EXECSW_PRIO_FIRST:
1860 SLIST_INSERT_HEAD(&first, ex, ex_slist);
1861 break;
1862 case EXECSW_PRIO_ANY:
1863 SLIST_INSERT_HEAD(&any, ex, ex_slist);
1864 break;
1865 case EXECSW_PRIO_LAST:
1866 SLIST_INSERT_HEAD(&last, ex, ex_slist);
1867 break;
1868 default:
1869 panic("%s", __func__);
1870 break;
1871 }
1872 sz++;
1873 }
1874
1875 /*
1876 * Create new execsw[]. Ensure we do not try a zero-sized
1877 * allocation.
1878 */
1879 sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1880 i = 0;
1881 SLIST_FOREACH(ex, &first, ex_slist) {
1882 sw[i++] = ex->ex_sw;
1883 }
1884 SLIST_FOREACH(ex, &any, ex_slist) {
1885 sw[i++] = ex->ex_sw;
1886 }
1887 SLIST_FOREACH(ex, &last, ex_slist) {
1888 sw[i++] = ex->ex_sw;
1889 }
1890
1891 /* Replace old execsw[] and free used memory. */
1892 if (execsw != NULL) {
1893 kmem_free(__UNCONST(execsw),
1894 nexecs * sizeof(struct execsw *) + 1);
1895 }
1896 execsw = sw;
1897 nexecs = sz;
1898
1899 /* Figure out the maximum size of an exec header. */
1900 exec_maxhdrsz = sizeof(int);
1901 for (i = 0; i < nexecs; i++) {
1902 if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1903 exec_maxhdrsz = execsw[i]->es_hdrsz;
1904 }
1905
1906 return 0;
1907}
1908
1909static int
1910exec_sigcode_map(struct proc *p, const struct emul *e)
1911{
1912 vaddr_t va;
1913 vsize_t sz;
1914 int error;
1915 struct uvm_object *uobj;
1916
1917 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1918
1919 if (e->e_sigobject == NULL || sz == 0) {
1920 return 0;
1921 }
1922
1923 /*
1924 * If we don't have a sigobject for this emulation, create one.
1925 *
1926 * sigobject is an anonymous memory object (just like SYSV shared
1927 * memory) that we keep a permanent reference to and that we map
1928 * in all processes that need this sigcode. The creation is simple,
1929 * we create an object, add a permanent reference to it, map it in
1930 * kernel space, copy out the sigcode to it and unmap it.
1931 * We map it with PROT_READ|PROT_EXEC into the process just
1932 * the way sys_mmap() would map it.
1933 */
1934
1935 uobj = *e->e_sigobject;
1936 if (uobj == NULL) {
1937 mutex_enter(&sigobject_lock);
1938 if ((uobj = *e->e_sigobject) == NULL) {
1939 uobj = uao_create(sz, 0);
1940 (*uobj->pgops->pgo_reference)(uobj);
1941 va = vm_map_min(kernel_map);
1942 if ((error = uvm_map(kernel_map, &va, round_page(sz),
1943 uobj, 0, 0,
1944 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1945 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1946 printf("kernel mapping failed %d\n", error);
1947 (*uobj->pgops->pgo_detach)(uobj);
1948 mutex_exit(&sigobject_lock);
1949 return error;
1950 }
1951 memcpy((void *)va, e->e_sigcode, sz);
1952#ifdef PMAP_NEED_PROCWR
1953 pmap_procwr(&proc0, va, sz);
1954#endif
1955 uvm_unmap(kernel_map, va, va + round_page(sz));
1956 *e->e_sigobject = uobj;
1957 }
1958 mutex_exit(&sigobject_lock);
1959 }
1960
1961 /* Just a hint to uvm_map where to put it. */
1962 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1963 round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
1964
1965#ifdef __alpha__
1966 /*
1967 * Tru64 puts /sbin/loader at the end of user virtual memory,
1968 * which causes the above calculation to put the sigcode at
1969 * an invalid address. Put it just below the text instead.
1970 */
1971 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1972 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1973 }
1974#endif
1975
1976 (*uobj->pgops->pgo_reference)(uobj);
1977 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1978 uobj, 0, 0,
1979 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1980 UVM_ADV_RANDOM, 0));
1981 if (error) {
1982 DPRINTF(("%s, %d: map %p "
1983 "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1984 __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1985 va, error));
1986 (*uobj->pgops->pgo_detach)(uobj);
1987 return error;
1988 }
1989 p->p_sigctx.ps_sigcode = (void *)va;
1990 return 0;
1991}
1992
1993/*
1994 * Release a refcount on spawn_exec_data and destroy memory, if this
1995 * was the last one.
1996 */
1997static void
1998spawn_exec_data_release(struct spawn_exec_data *data)
1999{
2000 if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
2001 return;
2002
2003 cv_destroy(&data->sed_cv_child_ready);
2004 mutex_destroy(&data->sed_mtx_child);
2005
2006 if (data->sed_actions)
2007 posix_spawn_fa_free(data->sed_actions,
2008 data->sed_actions->len);
2009 if (data->sed_attrs)
2010 kmem_free(data->sed_attrs,
2011 sizeof(*data->sed_attrs));
2012 kmem_free(data, sizeof(*data));
2013}
2014
2015/*
2016 * A child lwp of a posix_spawn operation starts here and ends up in
2017 * cpu_spawn_return, dealing with all filedescriptor and scheduler
2018 * manipulations in between.
2019 * The parent waits for the child, as it is not clear whether the child
2020 * will be able to acquire its own exec_lock. If it can, the parent can
2021 * be released early and continue running in parallel. If not (or if the
2022 * magic debug flag is passed in the scheduler attribute struct), the
2023 * child rides on the parent's exec lock until it is ready to return to
2024 * to userland - and only then releases the parent. This method loses
2025 * concurrency, but improves error reporting.
2026 */
2027static void
2028spawn_return(void *arg)
2029{
2030 struct spawn_exec_data *spawn_data = arg;
2031 struct lwp *l = curlwp;
2032 int error, newfd;
2033 int ostat;
2034 size_t i;
2035 const struct posix_spawn_file_actions_entry *fae;
2036 pid_t ppid;
2037 register_t retval;
2038 bool have_reflock;
2039 bool parent_is_waiting = true;
2040
2041 /*
2042 * Check if we can release parent early.
2043 * We either need to have no sed_attrs, or sed_attrs does not
2044 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2045 * safe access to the parent proc (passed in sed_parent).
2046 * We then try to get the exec_lock, and only if that works, we can
2047 * release the parent here already.
2048 */
2049 ppid = spawn_data->sed_parent->p_pid;
2050 if ((!spawn_data->sed_attrs
2051 || (spawn_data->sed_attrs->sa_flags
2052 & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2053 && rw_tryenter(&exec_lock, RW_READER)) {
2054 parent_is_waiting = false;
2055 mutex_enter(&spawn_data->sed_mtx_child);
2056 cv_signal(&spawn_data->sed_cv_child_ready);
2057 mutex_exit(&spawn_data->sed_mtx_child);
2058 }
2059
2060 /* don't allow debugger access yet */
2061 rw_enter(&l->l_proc->p_reflock, RW_WRITER);
2062 have_reflock = true;
2063
2064 error = 0;
2065 /* handle posix_spawn_file_actions */
2066 if (spawn_data->sed_actions != NULL) {
2067 for (i = 0; i < spawn_data->sed_actions->len; i++) {
2068 fae = &spawn_data->sed_actions->fae[i];
2069 switch (fae->fae_action) {
2070 case FAE_OPEN:
2071 if (fd_getfile(fae->fae_fildes) != NULL) {
2072 error = fd_close(fae->fae_fildes);
2073 if (error)
2074 break;
2075 }
2076 error = fd_open(fae->fae_path, fae->fae_oflag,
2077 fae->fae_mode, &newfd);
2078 if (error)
2079 break;
2080 if (newfd != fae->fae_fildes) {
2081 error = dodup(l, newfd,
2082 fae->fae_fildes, 0, &retval);
2083 if (fd_getfile(newfd) != NULL)
2084 fd_close(newfd);
2085 }
2086 break;
2087 case FAE_DUP2:
2088 error = dodup(l, fae->fae_fildes,
2089 fae->fae_newfildes, 0, &retval);
2090 break;
2091 case FAE_CLOSE:
2092 if (fd_getfile(fae->fae_fildes) == NULL) {
2093 error = EBADF;
2094 break;
2095 }
2096 error = fd_close(fae->fae_fildes);
2097 break;
2098 }
2099 if (error)
2100 goto report_error;
2101 }
2102 }
2103
2104 /* handle posix_spawnattr */
2105 if (spawn_data->sed_attrs != NULL) {
2106 struct sigaction sigact;
2107 sigact._sa_u._sa_handler = SIG_DFL;
2108 sigact.sa_flags = 0;
2109
2110 /*
2111 * set state to SSTOP so that this proc can be found by pid.
2112 * see proc_enterprp, do_sched_setparam below
2113 */
2114 mutex_enter(proc_lock);
2115 /*
2116 * p_stat should be SACTIVE, so we need to adjust the
2117 * parent's p_nstopchild here. For safety, just make
2118 * we're on the good side of SDEAD before we adjust.
2119 */
2120 ostat = l->l_proc->p_stat;
2121 KASSERT(ostat < SSTOP);
2122 l->l_proc->p_stat = SSTOP;
2123 l->l_proc->p_waited = 0;
2124 l->l_proc->p_pptr->p_nstopchild++;
2125 mutex_exit(proc_lock);
2126
2127 /* Set process group */
2128 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2129 pid_t mypid = l->l_proc->p_pid,
2130 pgrp = spawn_data->sed_attrs->sa_pgroup;
2131
2132 if (pgrp == 0)
2133 pgrp = mypid;
2134
2135 error = proc_enterpgrp(spawn_data->sed_parent,
2136 mypid, pgrp, false);
2137 if (error)
2138 goto report_error_stopped;
2139 }
2140
2141 /* Set scheduler policy */
2142 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2143 error = do_sched_setparam(l->l_proc->p_pid, 0,
2144 spawn_data->sed_attrs->sa_schedpolicy,
2145 &spawn_data->sed_attrs->sa_schedparam);
2146 else if (spawn_data->sed_attrs->sa_flags
2147 & POSIX_SPAWN_SETSCHEDPARAM) {
2148 error = do_sched_setparam(ppid, 0,
2149 SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2150 }
2151 if (error)
2152 goto report_error_stopped;
2153
2154 /* Reset user ID's */
2155 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2156 error = do_setresuid(l, -1,
2157 kauth_cred_getgid(l->l_cred), -1,
2158 ID_E_EQ_R | ID_E_EQ_S);
2159 if (error)
2160 goto report_error_stopped;
2161 error = do_setresuid(l, -1,
2162 kauth_cred_getuid(l->l_cred), -1,
2163 ID_E_EQ_R | ID_E_EQ_S);
2164 if (error)
2165 goto report_error_stopped;
2166 }
2167
2168 /* Set signal masks/defaults */
2169 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2170 mutex_enter(l->l_proc->p_lock);
2171 error = sigprocmask1(l, SIG_SETMASK,
2172 &spawn_data->sed_attrs->sa_sigmask, NULL);
2173 mutex_exit(l->l_proc->p_lock);
2174 if (error)
2175 goto report_error_stopped;
2176 }
2177
2178 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2179 /*
2180 * The following sigaction call is using a sigaction
2181 * version 0 trampoline which is in the compatibility
2182 * code only. This is not a problem because for SIG_DFL
2183 * and SIG_IGN, the trampolines are now ignored. If they
2184 * were not, this would be a problem because we are
2185 * holding the exec_lock, and the compat code needs
2186 * to do the same in order to replace the trampoline
2187 * code of the process.
2188 */
2189 for (i = 1; i <= NSIG; i++) {
2190 if (sigismember(
2191 &spawn_data->sed_attrs->sa_sigdefault, i))
2192 sigaction1(l, i, &sigact, NULL, NULL,
2193 0);
2194 }
2195 }
2196 mutex_enter(proc_lock);
2197 l->l_proc->p_stat = ostat;
2198 l->l_proc->p_pptr->p_nstopchild--;
2199 mutex_exit(proc_lock);
2200 }
2201
2202 /* now do the real exec */
2203 error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2204 true);
2205 have_reflock = false;
2206 if (error == EJUSTRETURN)
2207 error = 0;
2208 else if (error)
2209 goto report_error;
2210
2211 if (parent_is_waiting) {
2212 mutex_enter(&spawn_data->sed_mtx_child);
2213 cv_signal(&spawn_data->sed_cv_child_ready);
2214 mutex_exit(&spawn_data->sed_mtx_child);
2215 }
2216
2217 /* release our refcount on the data */
2218 spawn_exec_data_release(spawn_data);
2219
2220 /* and finally: leave to userland for the first time */
2221 cpu_spawn_return(l);
2222
2223 /* NOTREACHED */
2224 return;
2225
2226 report_error_stopped:
2227 mutex_enter(proc_lock);
2228 l->l_proc->p_stat = ostat;
2229 l->l_proc->p_pptr->p_nstopchild--;
2230 mutex_exit(proc_lock);
2231 report_error:
2232 if (have_reflock) {
2233 /*
2234 * We have not passed through execve_runproc(),
2235 * which would have released the p_reflock and also
2236 * taken ownership of the sed_exec part of spawn_data,
2237 * so release/free both here.
2238 */
2239 rw_exit(&l->l_proc->p_reflock);
2240 execve_free_data(&spawn_data->sed_exec);
2241 }
2242
2243 if (parent_is_waiting) {
2244 /* pass error to parent */
2245 mutex_enter(&spawn_data->sed_mtx_child);
2246 spawn_data->sed_error = error;
2247 cv_signal(&spawn_data->sed_cv_child_ready);
2248 mutex_exit(&spawn_data->sed_mtx_child);
2249 } else {
2250 rw_exit(&exec_lock);
2251 }
2252
2253 /* release our refcount on the data */
2254 spawn_exec_data_release(spawn_data);
2255
2256 /* done, exit */
2257 mutex_enter(l->l_proc->p_lock);
2258 /*
2259 * Posix explicitly asks for an exit code of 127 if we report
2260 * errors from the child process - so, unfortunately, there
2261 * is no way to report a more exact error code.
2262 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2263 * flag bit in the attrp argument to posix_spawn(2), see above.
2264 */
2265 exit1(l, 127, 0);
2266}
2267
2268void
2269posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2270{
2271
2272 for (size_t i = 0; i < len; i++) {
2273 struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2274 if (fae->fae_action != FAE_OPEN)
2275 continue;
2276 kmem_free(fae->fae_path, strlen(fae->fae_path) + 1);
2277 }
2278 if (fa->len > 0)
2279 kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2280 kmem_free(fa, sizeof(*fa));
2281}
2282
2283static int
2284posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2285 const struct posix_spawn_file_actions *ufa, rlim_t lim)
2286{
2287 struct posix_spawn_file_actions *fa;
2288 struct posix_spawn_file_actions_entry *fae;
2289 char *pbuf = NULL;
2290 int error;
2291 size_t i = 0;
2292
2293 fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2294 error = copyin(ufa, fa, sizeof(*fa));
2295 if (error || fa->len == 0) {
2296 kmem_free(fa, sizeof(*fa));
2297 return error; /* 0 if not an error, and len == 0 */
2298 }
2299
2300 if (fa->len > lim) {
2301 kmem_free(fa, sizeof(*fa));
2302 return EINVAL;
2303 }
2304
2305 fa->size = fa->len;
2306 size_t fal = fa->len * sizeof(*fae);
2307 fae = fa->fae;
2308 fa->fae = kmem_alloc(fal, KM_SLEEP);
2309 error = copyin(fae, fa->fae, fal);
2310 if (error)
2311 goto out;
2312
2313 pbuf = PNBUF_GET();
2314 for (; i < fa->len; i++) {
2315 fae = &fa->fae[i];
2316 if (fae->fae_action != FAE_OPEN)
2317 continue;
2318 error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2319 if (error)
2320 goto out;
2321 fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2322 memcpy(fae->fae_path, pbuf, fal);
2323 }
2324 PNBUF_PUT(pbuf);
2325
2326 *fap = fa;
2327 return 0;
2328out:
2329 if (pbuf)
2330 PNBUF_PUT(pbuf);
2331 posix_spawn_fa_free(fa, i);
2332 return error;
2333}
2334
2335int
2336check_posix_spawn(struct lwp *l1)
2337{
2338 int error, tnprocs, count;
2339 uid_t uid;
2340 struct proc *p1;
2341
2342 p1 = l1->l_proc;
2343 uid = kauth_cred_getuid(l1->l_cred);
2344 tnprocs = atomic_inc_uint_nv(&nprocs);
2345
2346 /*
2347 * Although process entries are dynamically created, we still keep
2348 * a global limit on the maximum number we will create.
2349 */
2350 if (__predict_false(tnprocs >= maxproc))
2351 error = -1;
2352 else
2353 error = kauth_authorize_process(l1->l_cred,
2354 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2355
2356 if (error) {
2357 atomic_dec_uint(&nprocs);
2358 return EAGAIN;
2359 }
2360
2361 /*
2362 * Enforce limits.
2363 */
2364 count = chgproccnt(uid, 1);
2365 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2366 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2367 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2368 __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2369 (void)chgproccnt(uid, -1);
2370 atomic_dec_uint(&nprocs);
2371 return EAGAIN;
2372 }
2373
2374 return 0;
2375}
2376
2377int
2378do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2379 struct posix_spawn_file_actions *fa,
2380 struct posix_spawnattr *sa,
2381 char *const *argv, char *const *envp,
2382 execve_fetch_element_t fetch)
2383{
2384
2385 struct proc *p1, *p2;
2386 struct lwp *l2;
2387 int error;
2388 struct spawn_exec_data *spawn_data;
2389 vaddr_t uaddr;
2390 pid_t pid;
2391 bool have_exec_lock = false;
2392
2393 p1 = l1->l_proc;
2394
2395 /* Allocate and init spawn_data */
2396 spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2397 spawn_data->sed_refcnt = 1; /* only parent so far */
2398 cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2399 mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2400 mutex_enter(&spawn_data->sed_mtx_child);
2401
2402 /*
2403 * Do the first part of the exec now, collect state
2404 * in spawn_data.
2405 */
2406 error = execve_loadvm(l1, path, argv,
2407 envp, fetch, &spawn_data->sed_exec);
2408 if (error == EJUSTRETURN)
2409 error = 0;
2410 else if (error)
2411 goto error_exit;
2412
2413 have_exec_lock = true;
2414
2415 /*
2416 * Allocate virtual address space for the U-area now, while it
2417 * is still easy to abort the fork operation if we're out of
2418 * kernel virtual address space.
2419 */
2420 uaddr = uvm_uarea_alloc();
2421 if (__predict_false(uaddr == 0)) {
2422 error = ENOMEM;
2423 goto error_exit;
2424 }
2425
2426 /*
2427 * Allocate new proc. Borrow proc0 vmspace for it, we will
2428 * replace it with its own before returning to userland
2429 * in the child.
2430 * This is a point of no return, we will have to go through
2431 * the child proc to properly clean it up past this point.
2432 */
2433 p2 = proc_alloc();
2434 pid = p2->p_pid;
2435
2436 /*
2437 * Make a proc table entry for the new process.
2438 * Start by zeroing the section of proc that is zero-initialized,
2439 * then copy the section that is copied directly from the parent.
2440 */
2441 memset(&p2->p_startzero, 0,
2442 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2443 memcpy(&p2->p_startcopy, &p1->p_startcopy,
2444 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2445 p2->p_vmspace = proc0.p_vmspace;
2446
2447 TAILQ_INIT(&p2->p_sigpend.sp_info);
2448
2449 LIST_INIT(&p2->p_lwps);
2450 LIST_INIT(&p2->p_sigwaiters);
2451
2452 /*
2453 * Duplicate sub-structures as needed.
2454 * Increase reference counts on shared objects.
2455 * Inherit flags we want to keep. The flags related to SIGCHLD
2456 * handling are important in order to keep a consistent behaviour
2457 * for the child after the fork. If we are a 32-bit process, the
2458 * child will be too.
2459 */
2460 p2->p_flag =
2461 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2462 p2->p_emul = p1->p_emul;
2463 p2->p_execsw = p1->p_execsw;
2464
2465 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2466 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2467 rw_init(&p2->p_reflock);
2468 cv_init(&p2->p_waitcv, "wait");
2469 cv_init(&p2->p_lwpcv, "lwpwait");
2470
2471 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2472
2473 kauth_proc_fork(p1, p2);
2474
2475 p2->p_raslist = NULL;
2476 p2->p_fd = fd_copy();
2477
2478 /* XXX racy */
2479 p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2480
2481 p2->p_cwdi = cwdinit();
2482
2483 /*
2484 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2485 * we just need increase pl_refcnt.
2486 */
2487 if (!p1->p_limit->pl_writeable) {
2488 lim_addref(p1->p_limit);
2489 p2->p_limit = p1->p_limit;
2490 } else {
2491 p2->p_limit = lim_copy(p1->p_limit);
2492 }
2493
2494 p2->p_lflag = 0;
2495 p2->p_sflag = 0;
2496 p2->p_slflag = 0;
2497 p2->p_pptr = p1;
2498 p2->p_ppid = p1->p_pid;
2499 LIST_INIT(&p2->p_children);
2500
2501 p2->p_aio = NULL;
2502
2503#ifdef KTRACE
2504 /*
2505 * Copy traceflag and tracefile if enabled.
2506 * If not inherited, these were zeroed above.
2507 */
2508 if (p1->p_traceflag & KTRFAC_INHERIT) {
2509 mutex_enter(&ktrace_lock);
2510 p2->p_traceflag = p1->p_traceflag;
2511 if ((p2->p_tracep = p1->p_tracep) != NULL)
2512 ktradref(p2);
2513 mutex_exit(&ktrace_lock);
2514 }
2515#endif
2516
2517 /*
2518 * Create signal actions for the child process.
2519 */
2520 p2->p_sigacts = sigactsinit(p1, 0);
2521 mutex_enter(p1->p_lock);
2522 p2->p_sflag |=
2523 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2524 sched_proc_fork(p1, p2);
2525 mutex_exit(p1->p_lock);
2526
2527 p2->p_stflag = p1->p_stflag;
2528
2529 /*
2530 * p_stats.
2531 * Copy parts of p_stats, and zero out the rest.
2532 */
2533 p2->p_stats = pstatscopy(p1->p_stats);
2534
2535 /* copy over machdep flags to the new proc */
2536 cpu_proc_fork(p1, p2);
2537
2538 /*
2539 * Prepare remaining parts of spawn data
2540 */
2541 spawn_data->sed_actions = fa;
2542 spawn_data->sed_attrs = sa;
2543
2544 spawn_data->sed_parent = p1;
2545
2546 /* create LWP */
2547 lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2548 &l2, l1->l_class);
2549 l2->l_ctxlink = NULL; /* reset ucontext link */
2550
2551 /*
2552 * Copy the credential so other references don't see our changes.
2553 * Test to see if this is necessary first, since in the common case
2554 * we won't need a private reference.
2555 */
2556 if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2557 kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2558 l2->l_cred = kauth_cred_copy(l2->l_cred);
2559 kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2560 kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2561 }
2562
2563 /* Update the master credentials. */
2564 if (l2->l_cred != p2->p_cred) {
2565 kauth_cred_t ocred;
2566
2567 kauth_cred_hold(l2->l_cred);
2568 mutex_enter(p2->p_lock);
2569 ocred = p2->p_cred;
2570 p2->p_cred = l2->l_cred;
2571 mutex_exit(p2->p_lock);
2572 kauth_cred_free(ocred);
2573 }
2574
2575 *child_ok = true;
2576 spawn_data->sed_refcnt = 2; /* child gets it as well */
2577#if 0
2578 l2->l_nopreempt = 1; /* start it non-preemptable */
2579#endif
2580
2581 /*
2582 * It's now safe for the scheduler and other processes to see the
2583 * child process.
2584 */
2585 mutex_enter(proc_lock);
2586
2587 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2588 p2->p_lflag |= PL_CONTROLT;
2589
2590 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2591 p2->p_exitsig = SIGCHLD; /* signal for parent on exit */
2592
2593 LIST_INSERT_AFTER(p1, p2, p_pglist);
2594 LIST_INSERT_HEAD(&allproc, p2, p_list);
2595
2596 p2->p_trace_enabled = trace_is_enabled(p2);
2597#ifdef __HAVE_SYSCALL_INTERN
2598 (*p2->p_emul->e_syscall_intern)(p2);
2599#endif
2600
2601 /*
2602 * Make child runnable, set start time, and add to run queue except
2603 * if the parent requested the child to start in SSTOP state.
2604 */
2605 mutex_enter(p2->p_lock);
2606
2607 getmicrotime(&p2->p_stats->p_start);
2608
2609 lwp_lock(l2);
2610 KASSERT(p2->p_nrlwps == 1);
2611 p2->p_nrlwps = 1;
2612 p2->p_stat = SACTIVE;
2613 l2->l_stat = LSRUN;
2614 sched_enqueue(l2, false);
2615 lwp_unlock(l2);
2616
2617 mutex_exit(p2->p_lock);
2618 mutex_exit(proc_lock);
2619
2620 cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2621 error = spawn_data->sed_error;
2622 mutex_exit(&spawn_data->sed_mtx_child);
2623 spawn_exec_data_release(spawn_data);
2624
2625 rw_exit(&p1->p_reflock);
2626 rw_exit(&exec_lock);
2627 have_exec_lock = false;
2628
2629 *pid_res = pid;
2630 return error;
2631
2632 error_exit:
2633 if (have_exec_lock) {
2634 execve_free_data(&spawn_data->sed_exec);
2635 rw_exit(&p1->p_reflock);
2636 rw_exit(&exec_lock);
2637 }
2638 mutex_exit(&spawn_data->sed_mtx_child);
2639 spawn_exec_data_release(spawn_data);
2640
2641 return error;
2642}
2643
2644int
2645sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2646 register_t *retval)
2647{
2648 /* {
2649 syscallarg(pid_t *) pid;
2650 syscallarg(const char *) path;
2651 syscallarg(const struct posix_spawn_file_actions *) file_actions;
2652 syscallarg(const struct posix_spawnattr *) attrp;
2653 syscallarg(char *const *) argv;
2654 syscallarg(char *const *) envp;
2655 } */
2656
2657 int error;
2658 struct posix_spawn_file_actions *fa = NULL;
2659 struct posix_spawnattr *sa = NULL;
2660 pid_t pid;
2661 bool child_ok = false;
2662 rlim_t max_fileactions;
2663 proc_t *p = l1->l_proc;
2664
2665 error = check_posix_spawn(l1);
2666 if (error) {
2667 *retval = error;
2668 return 0;
2669 }
2670
2671 /* copy in file_actions struct */
2672 if (SCARG(uap, file_actions) != NULL) {
2673 max_fileactions = 2 * min(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2674 maxfiles);
2675 error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2676 max_fileactions);
2677 if (error)
2678 goto error_exit;
2679 }
2680
2681 /* copyin posix_spawnattr struct */
2682 if (SCARG(uap, attrp) != NULL) {
2683 sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2684 error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2685 if (error)
2686 goto error_exit;
2687 }
2688
2689 /*
2690 * Do the spawn
2691 */
2692 error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2693 SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2694 if (error)
2695 goto error_exit;
2696
2697 if (error == 0 && SCARG(uap, pid) != NULL)
2698 error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2699
2700 *retval = error;
2701 return 0;
2702
2703 error_exit:
2704 if (!child_ok) {
2705 (void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2706 atomic_dec_uint(&nprocs);
2707
2708 if (sa)
2709 kmem_free(sa, sizeof(*sa));
2710 if (fa)
2711 posix_spawn_fa_free(fa, fa->len);
2712 }
2713
2714 *retval = error;
2715 return 0;
2716}
2717
2718void
2719exec_free_emul_arg(struct exec_package *epp)
2720{
2721 if (epp->ep_emul_arg_free != NULL) {
2722 KASSERT(epp->ep_emul_arg != NULL);
2723 (*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2724 epp->ep_emul_arg_free = NULL;
2725 epp->ep_emul_arg = NULL;
2726 } else {
2727 KASSERT(epp->ep_emul_arg == NULL);
2728 }
2729}
2730
2731#ifdef DEBUG_EXEC
2732static void
2733dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2734{
2735 struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2736 size_t j;
2737
2738 if (error == 0)
2739 DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2740 else
2741 DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2742 epp->ep_vmcmds.evs_used, error));
2743
2744 for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2745 DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2746 PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2747 PRIxVSIZE" prot=0%o flags=%d\n", j,
2748 vp[j].ev_proc == vmcmd_map_pagedvn ?
2749 "pagedvn" :
2750 vp[j].ev_proc == vmcmd_map_readvn ?
2751 "readvn" :
2752 vp[j].ev_proc == vmcmd_map_zero ?
2753 "zero" : "*unknown*",
2754 vp[j].ev_addr, vp[j].ev_len,
2755 vp[j].ev_offset, vp[j].ev_prot,
2756 vp[j].ev_flags));
2757 if (error != 0 && j == x)
2758 DPRINTF((" ^--- failed\n"));
2759 }
2760}
2761#endif
2762