kern_exec.c revision 1.460 1 /* $NetBSD: kern_exec.c,v 1.460 2018/08/10 21:44:59 pgoyette 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 1.460 2018/08/10 21:44:59 pgoyette 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
123 struct execve_data;
124
125 extern int user_va0_disable;
126
127 static size_t calcargs(struct execve_data * restrict, const size_t);
128 static size_t calcstack(struct execve_data * restrict, const size_t);
129 static int copyoutargs(struct execve_data * restrict, struct lwp *,
130 char * const);
131 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
132 static int copyinargs(struct execve_data * restrict, char * const *,
133 char * const *, execve_fetch_element_t, char **);
134 static int copyinargstrs(struct execve_data * restrict, char * const *,
135 execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
136 static 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))
145 static 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 */
156 SDT_PROVIDER_DECLARE(proc);
157 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
158 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
159 SDT_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 */
171 static const struct execsw **execsw = NULL;
172 static int nexecs;
173
174 u_int exec_maxhdrsz; /* must not be static - used by netbsd32 */
175
176 /* list of dynamically loaded execsw entries */
177 static LIST_HEAD(execlist_head, exec_entry) ex_head =
178 LIST_HEAD_INITIALIZER(ex_head);
179 struct 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
186 void 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 */
195 struct 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 .e_nomodbits = sysent_nomodbits,
213 #ifdef SYSCALL_DEBUG
214 .e_syscallnames = syscallnames,
215 #else
216 .e_syscallnames = NULL,
217 #endif
218 .e_sendsig = sendsig,
219 .e_trapsignal = trapsignal,
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_vm_default_addr = uvm_default_mapaddr,
236 .e_usertrap = NULL,
237 .e_ucsize = sizeof(ucontext_t),
238 .e_startlwp = startlwp
239 };
240
241 /*
242 * Exec lock. Used to control access to execsw[] structures.
243 * This must not be static so that netbsd32 can access it, too.
244 */
245 krwlock_t exec_lock;
246
247 static kmutex_t sigobject_lock;
248
249 /*
250 * Data used between a loadvm and execve part of an "exec" operation
251 */
252 struct execve_data {
253 struct exec_package ed_pack;
254 struct pathbuf *ed_pathbuf;
255 struct vattr ed_attr;
256 struct ps_strings ed_arginfo;
257 char *ed_argp;
258 const char *ed_pathstring;
259 char *ed_resolvedpathbuf;
260 size_t ed_ps_strings_sz;
261 int ed_szsigcode;
262 size_t ed_argslen;
263 long ed_argc;
264 long ed_envc;
265 };
266
267 /*
268 * data passed from parent lwp to child during a posix_spawn()
269 */
270 struct spawn_exec_data {
271 struct execve_data sed_exec;
272 struct posix_spawn_file_actions
273 *sed_actions;
274 struct posix_spawnattr *sed_attrs;
275 struct proc *sed_parent;
276 kcondvar_t sed_cv_child_ready;
277 kmutex_t sed_mtx_child;
278 int sed_error;
279 volatile uint32_t sed_refcnt;
280 };
281
282 static struct vm_map *exec_map;
283 static struct pool exec_pool;
284
285 static void *
286 exec_pool_alloc(struct pool *pp, int flags)
287 {
288
289 return (void *)uvm_km_alloc(exec_map, NCARGS, 0,
290 UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
291 }
292
293 static void
294 exec_pool_free(struct pool *pp, void *addr)
295 {
296
297 uvm_km_free(exec_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
298 }
299
300 static 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 */
331 int
332 /*ARGSUSED*/
333 check_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 #if 1
341 // grab the absolute pathbuf here before namei() trashes it.
342 pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX);
343 #endif
344 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
345
346 /* first get the vnode */
347 if ((error = namei(&nd)) != 0)
348 return error;
349 epp->ep_vp = vp = nd.ni_vp;
350 #if 0
351 /*
352 * XXX: can't use nd.ni_pnbuf, because although pb contains an
353 * absolute path, nd.ni_pnbuf does not if the path contains symlinks.
354 */
355 /* normally this can't fail */
356 error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL);
357 KASSERT(error == 0);
358 #endif
359
360 #ifdef DIAGNOSTIC
361 /* paranoia (take this out once namei stuff stabilizes) */
362 memset(nd.ni_pnbuf, '~', PATH_MAX);
363 #endif
364
365 /* check access and type */
366 if (vp->v_type != VREG) {
367 error = EACCES;
368 goto bad1;
369 }
370 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
371 goto bad1;
372
373 /* get attributes */
374 if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
375 goto bad1;
376
377 /* Check mount point */
378 if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
379 error = EACCES;
380 goto bad1;
381 }
382 if (vp->v_mount->mnt_flag & MNT_NOSUID)
383 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
384
385 /* try to open it */
386 if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
387 goto bad1;
388
389 /* unlock vp, since we need it unlocked from here on out. */
390 VOP_UNLOCK(vp);
391
392 #if NVERIEXEC > 0
393 error = veriexec_verify(l, vp, epp->ep_resolvedname,
394 epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
395 NULL);
396 if (error)
397 goto bad2;
398 #endif /* NVERIEXEC > 0 */
399
400 #ifdef PAX_SEGVGUARD
401 error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
402 if (error)
403 goto bad2;
404 #endif /* PAX_SEGVGUARD */
405
406 /* now we have the file, get the exec header */
407 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
408 UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
409 if (error)
410 goto bad2;
411 epp->ep_hdrvalid = epp->ep_hdrlen - resid;
412
413 /*
414 * Set up default address space limits. Can be overridden
415 * by individual exec packages.
416 */
417 epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS);
418 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
419
420 /*
421 * set up the vmcmds for creation of the process
422 * address space
423 */
424 error = ENOEXEC;
425 for (i = 0; i < nexecs; i++) {
426 int newerror;
427
428 epp->ep_esch = execsw[i];
429 newerror = (*execsw[i]->es_makecmds)(l, epp);
430
431 if (!newerror) {
432 /* Seems ok: check that entry point is not too high */
433 if (epp->ep_entry >= epp->ep_vm_maxaddr) {
434 #ifdef DIAGNOSTIC
435 printf("%s: rejecting %p due to "
436 "too high entry address (>= %p)\n",
437 __func__, (void *)epp->ep_entry,
438 (void *)epp->ep_vm_maxaddr);
439 #endif
440 error = ENOEXEC;
441 break;
442 }
443 /* Seems ok: check that entry point is not too low */
444 if (epp->ep_entry < epp->ep_vm_minaddr) {
445 #ifdef DIAGNOSTIC
446 printf("%s: rejecting %p due to "
447 "too low entry address (< %p)\n",
448 __func__, (void *)epp->ep_entry,
449 (void *)epp->ep_vm_minaddr);
450 #endif
451 error = ENOEXEC;
452 break;
453 }
454
455 /* check limits */
456 if ((epp->ep_tsize > MAXTSIZ) ||
457 (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
458 [RLIMIT_DATA].rlim_cur)) {
459 #ifdef DIAGNOSTIC
460 printf("%s: rejecting due to "
461 "limits (t=%llu > %llu || d=%llu > %llu)\n",
462 __func__,
463 (unsigned long long)epp->ep_tsize,
464 (unsigned long long)MAXTSIZ,
465 (unsigned long long)epp->ep_dsize,
466 (unsigned long long)
467 l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
468 #endif
469 error = ENOMEM;
470 break;
471 }
472 return 0;
473 }
474
475 /*
476 * Reset all the fields that may have been modified by the
477 * loader.
478 */
479 KASSERT(epp->ep_emul_arg == NULL);
480 if (epp->ep_emul_root != NULL) {
481 vrele(epp->ep_emul_root);
482 epp->ep_emul_root = NULL;
483 }
484 if (epp->ep_interp != NULL) {
485 vrele(epp->ep_interp);
486 epp->ep_interp = NULL;
487 }
488 epp->ep_pax_flags = 0;
489
490 /* make sure the first "interesting" error code is saved. */
491 if (error == ENOEXEC)
492 error = newerror;
493
494 if (epp->ep_flags & EXEC_DESTR)
495 /* Error from "#!" code, tidied up by recursive call */
496 return error;
497 }
498
499 /* not found, error */
500
501 /*
502 * free any vmspace-creation commands,
503 * and release their references
504 */
505 kill_vmcmds(&epp->ep_vmcmds);
506
507 bad2:
508 /*
509 * close and release the vnode, restore the old one, free the
510 * pathname buf, and punt.
511 */
512 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
513 VOP_CLOSE(vp, FREAD, l->l_cred);
514 vput(vp);
515 return error;
516
517 bad1:
518 /*
519 * free the namei pathname buffer, and put the vnode
520 * (which we don't yet have open).
521 */
522 vput(vp); /* was still locked */
523 return error;
524 }
525
526 #ifdef __MACHINE_STACK_GROWS_UP
527 #define STACK_PTHREADSPACE NBPG
528 #else
529 #define STACK_PTHREADSPACE 0
530 #endif
531
532 static int
533 execve_fetch_element(char * const *array, size_t index, char **value)
534 {
535 return copyin(array + index, value, sizeof(*value));
536 }
537
538 /*
539 * exec system call
540 */
541 int
542 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
543 {
544 /* {
545 syscallarg(const char *) path;
546 syscallarg(char * const *) argp;
547 syscallarg(char * const *) envp;
548 } */
549
550 return execve1(l, SCARG(uap, path), SCARG(uap, argp),
551 SCARG(uap, envp), execve_fetch_element);
552 }
553
554 int
555 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
556 register_t *retval)
557 {
558 /* {
559 syscallarg(int) fd;
560 syscallarg(char * const *) argp;
561 syscallarg(char * const *) envp;
562 } */
563
564 return ENOSYS;
565 }
566
567 /*
568 * Load modules to try and execute an image that we do not understand.
569 * If no execsw entries are present, we load those likely to be needed
570 * in order to run native images only. Otherwise, we autoload all
571 * possible modules that could let us run the binary. XXX lame
572 */
573 static void
574 exec_autoload(void)
575 {
576 #ifdef MODULAR
577 static const char * const native[] = {
578 "exec_elf32",
579 "exec_elf64",
580 "exec_script",
581 NULL
582 };
583 static const char * const compat[] = {
584 "exec_elf32",
585 "exec_elf64",
586 "exec_script",
587 "exec_aout",
588 "exec_coff",
589 "exec_ecoff",
590 "compat_aoutm68k",
591 "compat_netbsd32",
592 "compat_sunos",
593 "compat_sunos32",
594 "compat_ultrix",
595 NULL
596 };
597 char const * const *list;
598 int i;
599
600 list = (nexecs == 0 ? native : compat);
601 for (i = 0; list[i] != NULL; i++) {
602 if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
603 continue;
604 }
605 yield();
606 }
607 #endif
608 }
609
610 int
611 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg,
612 struct pathbuf **pbp, size_t *offs)
613 {
614 char *path, *bp;
615 size_t len, tlen;
616 int error;
617 struct cwdinfo *cwdi;
618
619 path = PNBUF_GET();
620 if (seg == UIO_SYSSPACE) {
621 error = copystr(upath, path, MAXPATHLEN, &len);
622 } else {
623 error = copyinstr(upath, path, MAXPATHLEN, &len);
624 }
625 if (error) {
626 PNBUF_PUT(path);
627 DPRINTF(("%s: copyin path @%p %d\n", __func__, upath, error));
628 return error;
629 }
630
631 if (path[0] == '/') {
632 if (offs)
633 *offs = 0;
634 goto out;
635 }
636
637 len++;
638 if (len + 1 >= MAXPATHLEN)
639 goto out;
640 bp = path + MAXPATHLEN - len;
641 memmove(bp, path, len);
642 *(--bp) = '/';
643
644 cwdi = l->l_proc->p_cwdi;
645 rw_enter(&cwdi->cwdi_lock, RW_READER);
646 error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
647 GETCWD_CHECK_ACCESS, l);
648 rw_exit(&cwdi->cwdi_lock);
649
650 if (error) {
651 DPRINTF(("%s: getcwd_common path %s %d\n", __func__, path,
652 error));
653 goto out;
654 }
655 tlen = path + MAXPATHLEN - bp;
656
657 memmove(path, bp, tlen);
658 path[tlen] = '\0';
659 if (offs)
660 *offs = tlen - len;
661 out:
662 *pbp = pathbuf_assimilate(path);
663 return 0;
664 }
665
666 vaddr_t
667 exec_vm_minaddr(vaddr_t va_min)
668 {
669 /*
670 * Increase va_min if we don't want NULL to be mappable by the
671 * process.
672 */
673 #define VM_MIN_GUARD PAGE_SIZE
674 if (user_va0_disable && (va_min < VM_MIN_GUARD))
675 return VM_MIN_GUARD;
676 return va_min;
677 }
678
679 static int
680 execve_loadvm(struct lwp *l, const char *path, char * const *args,
681 char * const *envs, execve_fetch_element_t fetch_element,
682 struct execve_data * restrict data)
683 {
684 struct exec_package * const epp = &data->ed_pack;
685 int error;
686 struct proc *p;
687 char *dp;
688 u_int modgen;
689 size_t offs = 0; // XXX: GCC
690
691 KASSERT(data != NULL);
692
693 p = l->l_proc;
694 modgen = 0;
695
696 SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
697
698 /*
699 * Check if we have exceeded our number of processes limit.
700 * This is so that we handle the case where a root daemon
701 * forked, ran setuid to become the desired user and is trying
702 * to exec. The obvious place to do the reference counting check
703 * is setuid(), but we don't do the reference counting check there
704 * like other OS's do because then all the programs that use setuid()
705 * must be modified to check the return code of setuid() and exit().
706 * It is dangerous to make setuid() fail, because it fails open and
707 * the program will continue to run as root. If we make it succeed
708 * and return an error code, again we are not enforcing the limit.
709 * The best place to enforce the limit is here, when the process tries
710 * to execute a new image, because eventually the process will need
711 * to call exec in order to do something useful.
712 */
713 retry:
714 if (p->p_flag & PK_SUGID) {
715 if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
716 p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
717 &p->p_rlimit[RLIMIT_NPROC],
718 KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
719 chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
720 p->p_rlimit[RLIMIT_NPROC].rlim_cur)
721 return EAGAIN;
722 }
723
724 /*
725 * Drain existing references and forbid new ones. The process
726 * should be left alone until we're done here. This is necessary
727 * to avoid race conditions - e.g. in ptrace() - that might allow
728 * a local user to illicitly obtain elevated privileges.
729 */
730 rw_enter(&p->p_reflock, RW_WRITER);
731
732 /*
733 * Init the namei data to point the file user's program name.
734 * This is done here rather than in check_exec(), so that it's
735 * possible to override this settings if any of makecmd/probe
736 * functions call check_exec() recursively - for example,
737 * see exec_script_makecmds().
738 */
739 if ((error = exec_makepathbuf(l, path, UIO_USERSPACE,
740 &data->ed_pathbuf, &offs)) != 0)
741 goto clrflg;
742 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
743 data->ed_resolvedpathbuf = PNBUF_GET();
744
745 /*
746 * initialize the fields of the exec package.
747 */
748 epp->ep_kname = data->ed_pathstring + offs;
749 epp->ep_resolvedname = data->ed_resolvedpathbuf;
750 epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
751 epp->ep_hdrlen = exec_maxhdrsz;
752 epp->ep_hdrvalid = 0;
753 epp->ep_emul_arg = NULL;
754 epp->ep_emul_arg_free = NULL;
755 memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
756 epp->ep_vap = &data->ed_attr;
757 epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
758 MD_TOPDOWN_INIT(epp);
759 epp->ep_emul_root = NULL;
760 epp->ep_interp = NULL;
761 epp->ep_esch = NULL;
762 epp->ep_pax_flags = 0;
763 memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
764
765 rw_enter(&exec_lock, RW_READER);
766
767 /* see if we can run it. */
768 if ((error = check_exec(l, epp, data->ed_pathbuf)) != 0) {
769 if (error != ENOENT && error != EACCES && error != ENOEXEC) {
770 DPRINTF(("%s: check exec failed for %s, error %d\n",
771 __func__, epp->ep_kname, error));
772 }
773 goto freehdr;
774 }
775
776 /* allocate an argument buffer */
777 data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
778 KASSERT(data->ed_argp != NULL);
779 dp = data->ed_argp;
780
781 if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
782 goto bad;
783 }
784
785 /*
786 * Calculate the new stack size.
787 */
788
789 #ifdef __MACHINE_STACK_GROWS_UP
790 /*
791 * copyargs() fills argc/argv/envp from the lower address even on
792 * __MACHINE_STACK_GROWS_UP machines. Reserve a few words just below the SP
793 * so that _rtld() use it.
794 */
795 #define RTLD_GAP 32
796 #else
797 #define RTLD_GAP 0
798 #endif
799
800 const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
801
802 data->ed_argslen = calcargs(data, argenvstrlen);
803
804 const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
805
806 if (len > epp->ep_ssize) {
807 /* in effect, compare to initial limit */
808 DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
809 error = ENOMEM;
810 goto bad;
811 }
812 /* adjust "active stack depth" for process VSZ */
813 epp->ep_ssize = len;
814
815 return 0;
816
817 bad:
818 /* free the vmspace-creation commands, and release their references */
819 kill_vmcmds(&epp->ep_vmcmds);
820 /* kill any opened file descriptor, if necessary */
821 if (epp->ep_flags & EXEC_HASFD) {
822 epp->ep_flags &= ~EXEC_HASFD;
823 fd_close(epp->ep_fd);
824 }
825 /* close and put the exec'd file */
826 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
827 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
828 vput(epp->ep_vp);
829 pool_put(&exec_pool, data->ed_argp);
830
831 freehdr:
832 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
833 if (epp->ep_emul_root != NULL)
834 vrele(epp->ep_emul_root);
835 if (epp->ep_interp != NULL)
836 vrele(epp->ep_interp);
837
838 rw_exit(&exec_lock);
839
840 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
841 pathbuf_destroy(data->ed_pathbuf);
842 PNBUF_PUT(data->ed_resolvedpathbuf);
843
844 clrflg:
845 rw_exit(&p->p_reflock);
846
847 if (modgen != module_gen && error == ENOEXEC) {
848 modgen = module_gen;
849 exec_autoload();
850 goto retry;
851 }
852
853 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
854 return error;
855 }
856
857 static int
858 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
859 {
860 struct exec_package * const epp = &data->ed_pack;
861 struct proc *p = l->l_proc;
862 struct exec_vmcmd *base_vcp;
863 int error = 0;
864 size_t i;
865
866 /* record proc's vnode, for use by procfs and others */
867 if (p->p_textvp)
868 vrele(p->p_textvp);
869 vref(epp->ep_vp);
870 p->p_textvp = epp->ep_vp;
871
872 /* create the new process's VM space by running the vmcmds */
873 KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
874
875 #ifdef TRACE_EXEC
876 DUMPVMCMDS(epp, 0, 0);
877 #endif
878
879 base_vcp = NULL;
880
881 for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
882 struct exec_vmcmd *vcp;
883
884 vcp = &epp->ep_vmcmds.evs_cmds[i];
885 if (vcp->ev_flags & VMCMD_RELATIVE) {
886 KASSERTMSG(base_vcp != NULL,
887 "%s: relative vmcmd with no base", __func__);
888 KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
889 "%s: illegal base & relative vmcmd", __func__);
890 vcp->ev_addr += base_vcp->ev_addr;
891 }
892 error = (*vcp->ev_proc)(l, vcp);
893 if (error)
894 DUMPVMCMDS(epp, i, error);
895 if (vcp->ev_flags & VMCMD_BASE)
896 base_vcp = vcp;
897 }
898
899 /* free the vmspace-creation commands, and release their references */
900 kill_vmcmds(&epp->ep_vmcmds);
901
902 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
903 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
904 vput(epp->ep_vp);
905
906 /* if an error happened, deallocate and punt */
907 if (error != 0) {
908 DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
909 }
910 return error;
911 }
912
913 static void
914 execve_free_data(struct execve_data *data)
915 {
916 struct exec_package * const epp = &data->ed_pack;
917
918 /* free the vmspace-creation commands, and release their references */
919 kill_vmcmds(&epp->ep_vmcmds);
920 /* kill any opened file descriptor, if necessary */
921 if (epp->ep_flags & EXEC_HASFD) {
922 epp->ep_flags &= ~EXEC_HASFD;
923 fd_close(epp->ep_fd);
924 }
925
926 /* close and put the exec'd file */
927 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
928 VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
929 vput(epp->ep_vp);
930 pool_put(&exec_pool, data->ed_argp);
931
932 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
933 if (epp->ep_emul_root != NULL)
934 vrele(epp->ep_emul_root);
935 if (epp->ep_interp != NULL)
936 vrele(epp->ep_interp);
937
938 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
939 pathbuf_destroy(data->ed_pathbuf);
940 PNBUF_PUT(data->ed_resolvedpathbuf);
941 }
942
943 static void
944 pathexec(struct proc *p, const char *resolvedname)
945 {
946 KASSERT(resolvedname[0] == '/');
947
948 /* set command name & other accounting info */
949 strlcpy(p->p_comm, strrchr(resolvedname, '/') + 1, sizeof(p->p_comm));
950
951 kmem_strfree(p->p_path);
952 p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP);
953 }
954
955 /* XXX elsewhere */
956 static int
957 credexec(struct lwp *l, struct vattr *attr)
958 {
959 struct proc *p = l->l_proc;
960 int error;
961
962 /*
963 * Deal with set[ug]id. MNT_NOSUID has already been used to disable
964 * s[ug]id. It's OK to check for PSL_TRACED here as we have blocked
965 * out additional references on the process for the moment.
966 */
967 if ((p->p_slflag & PSL_TRACED) == 0 &&
968
969 (((attr->va_mode & S_ISUID) != 0 &&
970 kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
971
972 ((attr->va_mode & S_ISGID) != 0 &&
973 kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
974 /*
975 * Mark the process as SUGID before we do
976 * anything that might block.
977 */
978 proc_crmod_enter();
979 proc_crmod_leave(NULL, NULL, true);
980
981 /* Make sure file descriptors 0..2 are in use. */
982 if ((error = fd_checkstd()) != 0) {
983 DPRINTF(("%s: fdcheckstd failed %d\n",
984 __func__, error));
985 return error;
986 }
987
988 /*
989 * Copy the credential so other references don't see our
990 * changes.
991 */
992 l->l_cred = kauth_cred_copy(l->l_cred);
993 #ifdef KTRACE
994 /*
995 * If the persistent trace flag isn't set, turn off.
996 */
997 if (p->p_tracep) {
998 mutex_enter(&ktrace_lock);
999 if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1000 ktrderef(p);
1001 mutex_exit(&ktrace_lock);
1002 }
1003 #endif
1004 if (attr->va_mode & S_ISUID)
1005 kauth_cred_seteuid(l->l_cred, attr->va_uid);
1006 if (attr->va_mode & S_ISGID)
1007 kauth_cred_setegid(l->l_cred, attr->va_gid);
1008 } else {
1009 if (kauth_cred_geteuid(l->l_cred) ==
1010 kauth_cred_getuid(l->l_cred) &&
1011 kauth_cred_getegid(l->l_cred) ==
1012 kauth_cred_getgid(l->l_cred))
1013 p->p_flag &= ~PK_SUGID;
1014 }
1015
1016 /*
1017 * Copy the credential so other references don't see our changes.
1018 * Test to see if this is necessary first, since in the common case
1019 * we won't need a private reference.
1020 */
1021 if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1022 kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1023 l->l_cred = kauth_cred_copy(l->l_cred);
1024 kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1025 kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1026 }
1027
1028 /* Update the master credentials. */
1029 if (l->l_cred != p->p_cred) {
1030 kauth_cred_t ocred;
1031
1032 kauth_cred_hold(l->l_cred);
1033 mutex_enter(p->p_lock);
1034 ocred = p->p_cred;
1035 p->p_cred = l->l_cred;
1036 mutex_exit(p->p_lock);
1037 kauth_cred_free(ocred);
1038 }
1039
1040 return 0;
1041 }
1042
1043 static void
1044 emulexec(struct lwp *l, struct exec_package *epp)
1045 {
1046 struct proc *p = l->l_proc;
1047
1048 /* The emulation root will usually have been found when we looked
1049 * for the elf interpreter (or similar), if not look now. */
1050 if (epp->ep_esch->es_emul->e_path != NULL &&
1051 epp->ep_emul_root == NULL)
1052 emul_find_root(l, epp);
1053
1054 /* Any old emulation root got removed by fdcloseexec */
1055 rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1056 p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1057 rw_exit(&p->p_cwdi->cwdi_lock);
1058 epp->ep_emul_root = NULL;
1059 if (epp->ep_interp != NULL)
1060 vrele(epp->ep_interp);
1061
1062 /*
1063 * Call emulation specific exec hook. This can setup per-process
1064 * p->p_emuldata or do any other per-process stuff an emulation needs.
1065 *
1066 * If we are executing process of different emulation than the
1067 * original forked process, call e_proc_exit() of the old emulation
1068 * first, then e_proc_exec() of new emulation. If the emulation is
1069 * same, the exec hook code should deallocate any old emulation
1070 * resources held previously by this process.
1071 */
1072 if (p->p_emul && p->p_emul->e_proc_exit
1073 && p->p_emul != epp->ep_esch->es_emul)
1074 (*p->p_emul->e_proc_exit)(p);
1075
1076 /*
1077 * This is now LWP 1.
1078 */
1079 /* XXX elsewhere */
1080 mutex_enter(p->p_lock);
1081 p->p_nlwpid = 1;
1082 l->l_lid = 1;
1083 mutex_exit(p->p_lock);
1084
1085 /*
1086 * Call exec hook. Emulation code may NOT store reference to anything
1087 * from &pack.
1088 */
1089 if (epp->ep_esch->es_emul->e_proc_exec)
1090 (*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1091
1092 /* update p_emul, the old value is no longer needed */
1093 p->p_emul = epp->ep_esch->es_emul;
1094
1095 /* ...and the same for p_execsw */
1096 p->p_execsw = epp->ep_esch;
1097
1098 #ifdef __HAVE_SYSCALL_INTERN
1099 (*p->p_emul->e_syscall_intern)(p);
1100 #endif
1101 ktremul();
1102 }
1103
1104 static int
1105 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1106 bool no_local_exec_lock, bool is_spawn)
1107 {
1108 struct exec_package * const epp = &data->ed_pack;
1109 int error = 0;
1110 struct proc *p;
1111
1112 /*
1113 * In case of a posix_spawn operation, the child doing the exec
1114 * might not hold the reader lock on exec_lock, but the parent
1115 * will do this instead.
1116 */
1117 KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1118 KASSERT(!no_local_exec_lock || is_spawn);
1119 KASSERT(data != NULL);
1120
1121 p = l->l_proc;
1122
1123 /* Get rid of other LWPs. */
1124 if (p->p_nlwps > 1) {
1125 mutex_enter(p->p_lock);
1126 exit_lwps(l);
1127 mutex_exit(p->p_lock);
1128 }
1129 KDASSERT(p->p_nlwps == 1);
1130
1131 /* Destroy any lwpctl info. */
1132 if (p->p_lwpctl != NULL)
1133 lwp_ctl_exit();
1134
1135 /* Remove POSIX timers */
1136 timers_free(p, TIMERS_POSIX);
1137
1138 /* Set the PaX flags. */
1139 pax_set_flags(epp, p);
1140
1141 /*
1142 * Do whatever is necessary to prepare the address space
1143 * for remapping. Note that this might replace the current
1144 * vmspace with another!
1145 */
1146 if (is_spawn)
1147 uvmspace_spawn(l, epp->ep_vm_minaddr,
1148 epp->ep_vm_maxaddr,
1149 epp->ep_flags & EXEC_TOPDOWN_VM);
1150 else
1151 uvmspace_exec(l, epp->ep_vm_minaddr,
1152 epp->ep_vm_maxaddr,
1153 epp->ep_flags & EXEC_TOPDOWN_VM);
1154
1155 struct vmspace *vm;
1156 vm = p->p_vmspace;
1157 vm->vm_taddr = (void *)epp->ep_taddr;
1158 vm->vm_tsize = btoc(epp->ep_tsize);
1159 vm->vm_daddr = (void*)epp->ep_daddr;
1160 vm->vm_dsize = btoc(epp->ep_dsize);
1161 vm->vm_ssize = btoc(epp->ep_ssize);
1162 vm->vm_issize = 0;
1163 vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1164 vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1165
1166 pax_aslr_init_vm(l, vm, epp);
1167
1168 /* Now map address space. */
1169 error = execve_dovmcmds(l, data);
1170 if (error != 0)
1171 goto exec_abort;
1172
1173 pathexec(p, epp->ep_resolvedname);
1174
1175 char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1176
1177 error = copyoutargs(data, l, newstack);
1178 if (error != 0)
1179 goto exec_abort;
1180
1181 cwdexec(p);
1182 fd_closeexec(); /* handle close on exec */
1183
1184 if (__predict_false(ktrace_on))
1185 fd_ktrexecfd();
1186
1187 execsigs(p); /* reset caught signals */
1188
1189 mutex_enter(p->p_lock);
1190 l->l_ctxlink = NULL; /* reset ucontext link */
1191 p->p_acflag &= ~AFORK;
1192 p->p_flag |= PK_EXEC;
1193 mutex_exit(p->p_lock);
1194
1195 /*
1196 * Stop profiling.
1197 */
1198 if ((p->p_stflag & PST_PROFIL) != 0) {
1199 mutex_spin_enter(&p->p_stmutex);
1200 stopprofclock(p);
1201 mutex_spin_exit(&p->p_stmutex);
1202 }
1203
1204 /*
1205 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1206 * exited and exec()/exit() are the only places it will be cleared.
1207 */
1208 if ((p->p_lflag & PL_PPWAIT) != 0) {
1209 mutex_enter(proc_lock);
1210 l->l_lwpctl = NULL; /* was on loan from blocked parent */
1211 p->p_lflag &= ~PL_PPWAIT;
1212 cv_broadcast(&p->p_pptr->p_waitcv);
1213 mutex_exit(proc_lock);
1214 }
1215
1216 error = credexec(l, &data->ed_attr);
1217 if (error)
1218 goto exec_abort;
1219
1220 #if defined(__HAVE_RAS)
1221 /*
1222 * Remove all RASs from the address space.
1223 */
1224 ras_purgeall();
1225 #endif
1226
1227 doexechooks(p);
1228
1229 /*
1230 * Set initial SP at the top of the stack.
1231 *
1232 * Note that on machines where stack grows up (e.g. hppa), SP points to
1233 * the end of arg/env strings. Userland guesses the address of argc
1234 * via ps_strings::ps_argvstr.
1235 */
1236
1237 /* Setup new registers and do misc. setup. */
1238 (*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1239 if (epp->ep_esch->es_setregs)
1240 (*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1241
1242 /* Provide a consistent LWP private setting */
1243 (void)lwp_setprivate(l, NULL);
1244
1245 /* Discard all PCU state; need to start fresh */
1246 pcu_discard_all(l);
1247
1248 /* map the process's signal trampoline code */
1249 if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1250 DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1251 goto exec_abort;
1252 }
1253
1254 pool_put(&exec_pool, data->ed_argp);
1255
1256 /* notify others that we exec'd */
1257 KNOTE(&p->p_klist, NOTE_EXEC);
1258
1259 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1260
1261 SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1262
1263 emulexec(l, epp);
1264
1265 /* Allow new references from the debugger/procfs. */
1266 rw_exit(&p->p_reflock);
1267 if (!no_local_exec_lock)
1268 rw_exit(&exec_lock);
1269
1270 mutex_enter(proc_lock);
1271
1272 if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1273 mutex_enter(p->p_lock);
1274 p->p_xsig = SIGTRAP;
1275 p->p_sigctx.ps_faked = true; // XXX
1276 p->p_sigctx.ps_info._signo = p->p_xsig;
1277 p->p_sigctx.ps_info._code = TRAP_EXEC;
1278 sigswitch(0, SIGTRAP, false);
1279 // XXX ktrpoint(KTR_PSIG)
1280 mutex_exit(p->p_lock);
1281 mutex_enter(proc_lock);
1282 }
1283
1284 if (p->p_sflag & PS_STOPEXEC) {
1285 ksiginfoq_t kq;
1286
1287 KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1288 p->p_pptr->p_nstopchild++;
1289 p->p_waited = 0;
1290 mutex_enter(p->p_lock);
1291 ksiginfo_queue_init(&kq);
1292 sigclearall(p, &contsigmask, &kq);
1293 lwp_lock(l);
1294 l->l_stat = LSSTOP;
1295 p->p_stat = SSTOP;
1296 p->p_nrlwps--;
1297 lwp_unlock(l);
1298 mutex_exit(p->p_lock);
1299 mutex_exit(proc_lock);
1300 lwp_lock(l);
1301 mi_switch(l);
1302 ksiginfo_queue_drain(&kq);
1303 KERNEL_LOCK(l->l_biglocks, l);
1304 } else {
1305 mutex_exit(proc_lock);
1306 }
1307
1308 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1309 pathbuf_destroy(data->ed_pathbuf);
1310 PNBUF_PUT(data->ed_resolvedpathbuf);
1311 #ifdef TRACE_EXEC
1312 DPRINTF(("%s finished\n", __func__));
1313 #endif
1314 return EJUSTRETURN;
1315
1316 exec_abort:
1317 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1318 rw_exit(&p->p_reflock);
1319 if (!no_local_exec_lock)
1320 rw_exit(&exec_lock);
1321
1322 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1323 pathbuf_destroy(data->ed_pathbuf);
1324 PNBUF_PUT(data->ed_resolvedpathbuf);
1325
1326 /*
1327 * the old process doesn't exist anymore. exit gracefully.
1328 * get rid of the (new) address space we have created, if any, get rid
1329 * of our namei data and vnode, and exit noting failure
1330 */
1331 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1332 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1333
1334 exec_free_emul_arg(epp);
1335 pool_put(&exec_pool, data->ed_argp);
1336 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1337 if (epp->ep_emul_root != NULL)
1338 vrele(epp->ep_emul_root);
1339 if (epp->ep_interp != NULL)
1340 vrele(epp->ep_interp);
1341
1342 /* Acquire the sched-state mutex (exit1() will release it). */
1343 if (!is_spawn) {
1344 mutex_enter(p->p_lock);
1345 exit1(l, error, SIGABRT);
1346 }
1347
1348 return error;
1349 }
1350
1351 int
1352 execve1(struct lwp *l, const char *path, char * const *args,
1353 char * const *envs, execve_fetch_element_t fetch_element)
1354 {
1355 struct execve_data data;
1356 int error;
1357
1358 error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1359 if (error)
1360 return error;
1361 error = execve_runproc(l, &data, false, false);
1362 return error;
1363 }
1364
1365 static size_t
1366 fromptrsz(const struct exec_package *epp)
1367 {
1368 return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1369 }
1370
1371 static size_t
1372 ptrsz(const struct exec_package *epp)
1373 {
1374 return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1375 }
1376
1377 static size_t
1378 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1379 {
1380 struct exec_package * const epp = &data->ed_pack;
1381
1382 const size_t nargenvptrs =
1383 1 + /* long argc */
1384 data->ed_argc + /* char *argv[] */
1385 1 + /* \0 */
1386 data->ed_envc + /* char *env[] */
1387 1; /* \0 */
1388
1389 return (nargenvptrs * ptrsz(epp)) /* pointers */
1390 + argenvstrlen /* strings */
1391 + epp->ep_esch->es_arglen; /* auxinfo */
1392 }
1393
1394 static size_t
1395 calcstack(struct execve_data * restrict data, const size_t gaplen)
1396 {
1397 struct exec_package * const epp = &data->ed_pack;
1398
1399 data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1400 epp->ep_esch->es_emul->e_sigcode;
1401
1402 data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1403 sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1404
1405 const size_t sigcode_psstr_sz =
1406 data->ed_szsigcode + /* sigcode */
1407 data->ed_ps_strings_sz + /* ps_strings */
1408 STACK_PTHREADSPACE; /* pthread space */
1409
1410 const size_t stacklen =
1411 data->ed_argslen +
1412 gaplen +
1413 sigcode_psstr_sz;
1414
1415 /* make the stack "safely" aligned */
1416 return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1417 }
1418
1419 static int
1420 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1421 char * const newstack)
1422 {
1423 struct exec_package * const epp = &data->ed_pack;
1424 struct proc *p = l->l_proc;
1425 int error;
1426
1427 /* remember information about the process */
1428 data->ed_arginfo.ps_nargvstr = data->ed_argc;
1429 data->ed_arginfo.ps_nenvstr = data->ed_envc;
1430
1431 /*
1432 * Allocate the stack address passed to the newly execve()'ed process.
1433 *
1434 * The new stack address will be set to the SP (stack pointer) register
1435 * in setregs().
1436 */
1437
1438 char *newargs = STACK_ALLOC(
1439 STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1440
1441 error = (*epp->ep_esch->es_copyargs)(l, epp,
1442 &data->ed_arginfo, &newargs, data->ed_argp);
1443
1444 if (error) {
1445 DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1446 return error;
1447 }
1448
1449 error = copyoutpsstrs(data, p);
1450 if (error != 0)
1451 return error;
1452
1453 return 0;
1454 }
1455
1456 static int
1457 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1458 {
1459 struct exec_package * const epp = &data->ed_pack;
1460 struct ps_strings32 arginfo32;
1461 void *aip;
1462 int error;
1463
1464 /* fill process ps_strings info */
1465 p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1466 STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1467
1468 if (epp->ep_flags & EXEC_32) {
1469 aip = &arginfo32;
1470 arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1471 arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1472 arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1473 arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1474 } else
1475 aip = &data->ed_arginfo;
1476
1477 /* copy out the process's ps_strings structure */
1478 if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1479 != 0) {
1480 DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1481 __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1482 return error;
1483 }
1484
1485 return 0;
1486 }
1487
1488 static int
1489 copyinargs(struct execve_data * restrict data, char * const *args,
1490 char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1491 {
1492 struct exec_package * const epp = &data->ed_pack;
1493 char *dp;
1494 size_t i;
1495 int error;
1496
1497 dp = *dpp;
1498
1499 data->ed_argc = 0;
1500
1501 /* copy the fake args list, if there's one, freeing it as we go */
1502 if (epp->ep_flags & EXEC_HASARGL) {
1503 struct exec_fakearg *fa = epp->ep_fa;
1504
1505 while (fa->fa_arg != NULL) {
1506 const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1507 size_t len;
1508
1509 len = strlcpy(dp, fa->fa_arg, maxlen);
1510 /* Count NUL into len. */
1511 if (len < maxlen)
1512 len++;
1513 else {
1514 while (fa->fa_arg != NULL) {
1515 kmem_free(fa->fa_arg, fa->fa_len);
1516 fa++;
1517 }
1518 kmem_free(epp->ep_fa, epp->ep_fa_len);
1519 epp->ep_flags &= ~EXEC_HASARGL;
1520 return E2BIG;
1521 }
1522 ktrexecarg(fa->fa_arg, len - 1);
1523 dp += len;
1524
1525 kmem_free(fa->fa_arg, fa->fa_len);
1526 fa++;
1527 data->ed_argc++;
1528 }
1529 kmem_free(epp->ep_fa, epp->ep_fa_len);
1530 epp->ep_flags &= ~EXEC_HASARGL;
1531 }
1532
1533 /*
1534 * Read and count argument strings from user.
1535 */
1536
1537 if (args == NULL) {
1538 DPRINTF(("%s: null args\n", __func__));
1539 return EINVAL;
1540 }
1541 if (epp->ep_flags & EXEC_SKIPARG)
1542 args = (const void *)((const char *)args + fromptrsz(epp));
1543 i = 0;
1544 error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1545 if (error != 0) {
1546 DPRINTF(("%s: copyin arg %d\n", __func__, error));
1547 return error;
1548 }
1549 data->ed_argc += i;
1550
1551 /*
1552 * Read and count environment strings from user.
1553 */
1554
1555 data->ed_envc = 0;
1556 /* environment need not be there */
1557 if (envs == NULL)
1558 goto done;
1559 i = 0;
1560 error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1561 if (error != 0) {
1562 DPRINTF(("%s: copyin env %d\n", __func__, error));
1563 return error;
1564 }
1565 data->ed_envc += i;
1566
1567 done:
1568 *dpp = dp;
1569
1570 return 0;
1571 }
1572
1573 static int
1574 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1575 execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1576 void (*ktr)(const void *, size_t))
1577 {
1578 char *dp, *sp;
1579 size_t i;
1580 int error;
1581
1582 dp = *dpp;
1583
1584 i = 0;
1585 while (1) {
1586 const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1587 size_t len;
1588
1589 if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1590 return error;
1591 }
1592 if (!sp)
1593 break;
1594 if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1595 if (error == ENAMETOOLONG)
1596 error = E2BIG;
1597 return error;
1598 }
1599 if (__predict_false(ktrace_on))
1600 (*ktr)(dp, len - 1);
1601 dp += len;
1602 i++;
1603 }
1604
1605 *dpp = dp;
1606 *ip = i;
1607
1608 return 0;
1609 }
1610
1611 /*
1612 * Copy argv and env strings from kernel buffer (argp) to the new stack.
1613 * Those strings are located just after auxinfo.
1614 */
1615 int
1616 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1617 char **stackp, void *argp)
1618 {
1619 char **cpp, *dp, *sp;
1620 size_t len;
1621 void *nullp;
1622 long argc, envc;
1623 int error;
1624
1625 cpp = (char **)*stackp;
1626 nullp = NULL;
1627 argc = arginfo->ps_nargvstr;
1628 envc = arginfo->ps_nenvstr;
1629
1630 /* argc on stack is long */
1631 CTASSERT(sizeof(*cpp) == sizeof(argc));
1632
1633 dp = (char *)(cpp +
1634 1 + /* long argc */
1635 argc + /* char *argv[] */
1636 1 + /* \0 */
1637 envc + /* char *env[] */
1638 1) + /* \0 */
1639 pack->ep_esch->es_arglen; /* auxinfo */
1640 sp = argp;
1641
1642 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1643 COPYPRINTF("", cpp - 1, sizeof(argc));
1644 return error;
1645 }
1646
1647 /* XXX don't copy them out, remap them! */
1648 arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1649
1650 for (; --argc >= 0; sp += len, dp += len) {
1651 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1652 COPYPRINTF("", cpp - 1, sizeof(dp));
1653 return error;
1654 }
1655 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1656 COPYPRINTF("str", dp, (size_t)ARG_MAX);
1657 return error;
1658 }
1659 }
1660
1661 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1662 COPYPRINTF("", cpp - 1, sizeof(nullp));
1663 return error;
1664 }
1665
1666 arginfo->ps_envstr = cpp; /* remember location of envp for later */
1667
1668 for (; --envc >= 0; sp += len, dp += len) {
1669 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1670 COPYPRINTF("", cpp - 1, sizeof(dp));
1671 return error;
1672 }
1673 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1674 COPYPRINTF("str", dp, (size_t)ARG_MAX);
1675 return error;
1676 }
1677
1678 }
1679
1680 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1681 COPYPRINTF("", cpp - 1, sizeof(nullp));
1682 return error;
1683 }
1684
1685 *stackp = (char *)cpp;
1686 return 0;
1687 }
1688
1689
1690 /*
1691 * Add execsw[] entries.
1692 */
1693 int
1694 exec_add(struct execsw *esp, int count)
1695 {
1696 struct exec_entry *it;
1697 int i;
1698
1699 if (count == 0) {
1700 return 0;
1701 }
1702
1703 /* Check for duplicates. */
1704 rw_enter(&exec_lock, RW_WRITER);
1705 for (i = 0; i < count; i++) {
1706 LIST_FOREACH(it, &ex_head, ex_list) {
1707 /* assume unique (makecmds, probe_func, emulation) */
1708 if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1709 it->ex_sw->u.elf_probe_func ==
1710 esp[i].u.elf_probe_func &&
1711 it->ex_sw->es_emul == esp[i].es_emul) {
1712 rw_exit(&exec_lock);
1713 return EEXIST;
1714 }
1715 }
1716 }
1717
1718 /* Allocate new entries. */
1719 for (i = 0; i < count; i++) {
1720 it = kmem_alloc(sizeof(*it), KM_SLEEP);
1721 it->ex_sw = &esp[i];
1722 LIST_INSERT_HEAD(&ex_head, it, ex_list);
1723 }
1724
1725 /* update execsw[] */
1726 exec_init(0);
1727 rw_exit(&exec_lock);
1728 return 0;
1729 }
1730
1731 /*
1732 * Remove execsw[] entry.
1733 */
1734 int
1735 exec_remove(struct execsw *esp, int count)
1736 {
1737 struct exec_entry *it, *next;
1738 int i;
1739 const struct proclist_desc *pd;
1740 proc_t *p;
1741
1742 if (count == 0) {
1743 return 0;
1744 }
1745
1746 /* Abort if any are busy. */
1747 rw_enter(&exec_lock, RW_WRITER);
1748 for (i = 0; i < count; i++) {
1749 mutex_enter(proc_lock);
1750 for (pd = proclists; pd->pd_list != NULL; pd++) {
1751 PROCLIST_FOREACH(p, pd->pd_list) {
1752 if (p->p_execsw == &esp[i]) {
1753 mutex_exit(proc_lock);
1754 rw_exit(&exec_lock);
1755 return EBUSY;
1756 }
1757 }
1758 }
1759 mutex_exit(proc_lock);
1760 }
1761
1762 /* None are busy, so remove them all. */
1763 for (i = 0; i < count; i++) {
1764 for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1765 next = LIST_NEXT(it, ex_list);
1766 if (it->ex_sw == &esp[i]) {
1767 LIST_REMOVE(it, ex_list);
1768 kmem_free(it, sizeof(*it));
1769 break;
1770 }
1771 }
1772 }
1773
1774 /* update execsw[] */
1775 exec_init(0);
1776 rw_exit(&exec_lock);
1777 return 0;
1778 }
1779
1780 /*
1781 * Initialize exec structures. If init_boot is true, also does necessary
1782 * one-time initialization (it's called from main() that way).
1783 * Once system is multiuser, this should be called with exec_lock held,
1784 * i.e. via exec_{add|remove}().
1785 */
1786 int
1787 exec_init(int init_boot)
1788 {
1789 const struct execsw **sw;
1790 struct exec_entry *ex;
1791 SLIST_HEAD(,exec_entry) first;
1792 SLIST_HEAD(,exec_entry) any;
1793 SLIST_HEAD(,exec_entry) last;
1794 int i, sz;
1795
1796 if (init_boot) {
1797 /* do one-time initializations */
1798 vaddr_t vmin = 0, vmax;
1799
1800 rw_init(&exec_lock);
1801 mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1802 exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax,
1803 maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL);
1804 pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1805 "execargs", &exec_palloc, IPL_NONE);
1806 pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1807 } else {
1808 KASSERT(rw_write_held(&exec_lock));
1809 }
1810
1811 /* Sort each entry onto the appropriate queue. */
1812 SLIST_INIT(&first);
1813 SLIST_INIT(&any);
1814 SLIST_INIT(&last);
1815 sz = 0;
1816 LIST_FOREACH(ex, &ex_head, ex_list) {
1817 switch(ex->ex_sw->es_prio) {
1818 case EXECSW_PRIO_FIRST:
1819 SLIST_INSERT_HEAD(&first, ex, ex_slist);
1820 break;
1821 case EXECSW_PRIO_ANY:
1822 SLIST_INSERT_HEAD(&any, ex, ex_slist);
1823 break;
1824 case EXECSW_PRIO_LAST:
1825 SLIST_INSERT_HEAD(&last, ex, ex_slist);
1826 break;
1827 default:
1828 panic("%s", __func__);
1829 break;
1830 }
1831 sz++;
1832 }
1833
1834 /*
1835 * Create new execsw[]. Ensure we do not try a zero-sized
1836 * allocation.
1837 */
1838 sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1839 i = 0;
1840 SLIST_FOREACH(ex, &first, ex_slist) {
1841 sw[i++] = ex->ex_sw;
1842 }
1843 SLIST_FOREACH(ex, &any, ex_slist) {
1844 sw[i++] = ex->ex_sw;
1845 }
1846 SLIST_FOREACH(ex, &last, ex_slist) {
1847 sw[i++] = ex->ex_sw;
1848 }
1849
1850 /* Replace old execsw[] and free used memory. */
1851 if (execsw != NULL) {
1852 kmem_free(__UNCONST(execsw),
1853 nexecs * sizeof(struct execsw *) + 1);
1854 }
1855 execsw = sw;
1856 nexecs = sz;
1857
1858 /* Figure out the maximum size of an exec header. */
1859 exec_maxhdrsz = sizeof(int);
1860 for (i = 0; i < nexecs; i++) {
1861 if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1862 exec_maxhdrsz = execsw[i]->es_hdrsz;
1863 }
1864
1865 return 0;
1866 }
1867
1868 static int
1869 exec_sigcode_map(struct proc *p, const struct emul *e)
1870 {
1871 vaddr_t va;
1872 vsize_t sz;
1873 int error;
1874 struct uvm_object *uobj;
1875
1876 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1877
1878 if (e->e_sigobject == NULL || sz == 0) {
1879 return 0;
1880 }
1881
1882 /*
1883 * If we don't have a sigobject for this emulation, create one.
1884 *
1885 * sigobject is an anonymous memory object (just like SYSV shared
1886 * memory) that we keep a permanent reference to and that we map
1887 * in all processes that need this sigcode. The creation is simple,
1888 * we create an object, add a permanent reference to it, map it in
1889 * kernel space, copy out the sigcode to it and unmap it.
1890 * We map it with PROT_READ|PROT_EXEC into the process just
1891 * the way sys_mmap() would map it.
1892 */
1893
1894 uobj = *e->e_sigobject;
1895 if (uobj == NULL) {
1896 mutex_enter(&sigobject_lock);
1897 if ((uobj = *e->e_sigobject) == NULL) {
1898 uobj = uao_create(sz, 0);
1899 (*uobj->pgops->pgo_reference)(uobj);
1900 va = vm_map_min(kernel_map);
1901 if ((error = uvm_map(kernel_map, &va, round_page(sz),
1902 uobj, 0, 0,
1903 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1904 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1905 printf("kernel mapping failed %d\n", error);
1906 (*uobj->pgops->pgo_detach)(uobj);
1907 mutex_exit(&sigobject_lock);
1908 return error;
1909 }
1910 memcpy((void *)va, e->e_sigcode, sz);
1911 #ifdef PMAP_NEED_PROCWR
1912 pmap_procwr(&proc0, va, sz);
1913 #endif
1914 uvm_unmap(kernel_map, va, va + round_page(sz));
1915 *e->e_sigobject = uobj;
1916 }
1917 mutex_exit(&sigobject_lock);
1918 }
1919
1920 /* Just a hint to uvm_map where to put it. */
1921 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1922 round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
1923
1924 #ifdef __alpha__
1925 /*
1926 * Tru64 puts /sbin/loader at the end of user virtual memory,
1927 * which causes the above calculation to put the sigcode at
1928 * an invalid address. Put it just below the text instead.
1929 */
1930 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1931 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1932 }
1933 #endif
1934
1935 (*uobj->pgops->pgo_reference)(uobj);
1936 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1937 uobj, 0, 0,
1938 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1939 UVM_ADV_RANDOM, 0));
1940 if (error) {
1941 DPRINTF(("%s, %d: map %p "
1942 "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1943 __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1944 va, error));
1945 (*uobj->pgops->pgo_detach)(uobj);
1946 return error;
1947 }
1948 p->p_sigctx.ps_sigcode = (void *)va;
1949 return 0;
1950 }
1951
1952 /*
1953 * Release a refcount on spawn_exec_data and destroy memory, if this
1954 * was the last one.
1955 */
1956 static void
1957 spawn_exec_data_release(struct spawn_exec_data *data)
1958 {
1959 if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1960 return;
1961
1962 cv_destroy(&data->sed_cv_child_ready);
1963 mutex_destroy(&data->sed_mtx_child);
1964
1965 if (data->sed_actions)
1966 posix_spawn_fa_free(data->sed_actions,
1967 data->sed_actions->len);
1968 if (data->sed_attrs)
1969 kmem_free(data->sed_attrs,
1970 sizeof(*data->sed_attrs));
1971 kmem_free(data, sizeof(*data));
1972 }
1973
1974 /*
1975 * A child lwp of a posix_spawn operation starts here and ends up in
1976 * cpu_spawn_return, dealing with all filedescriptor and scheduler
1977 * manipulations in between.
1978 * The parent waits for the child, as it is not clear whether the child
1979 * will be able to acquire its own exec_lock. If it can, the parent can
1980 * be released early and continue running in parallel. If not (or if the
1981 * magic debug flag is passed in the scheduler attribute struct), the
1982 * child rides on the parent's exec lock until it is ready to return to
1983 * to userland - and only then releases the parent. This method loses
1984 * concurrency, but improves error reporting.
1985 */
1986 static void
1987 spawn_return(void *arg)
1988 {
1989 struct spawn_exec_data *spawn_data = arg;
1990 struct lwp *l = curlwp;
1991 int error, newfd;
1992 int ostat;
1993 size_t i;
1994 const struct posix_spawn_file_actions_entry *fae;
1995 pid_t ppid;
1996 register_t retval;
1997 bool have_reflock;
1998 bool parent_is_waiting = true;
1999
2000 /*
2001 * Check if we can release parent early.
2002 * We either need to have no sed_attrs, or sed_attrs does not
2003 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2004 * safe access to the parent proc (passed in sed_parent).
2005 * We then try to get the exec_lock, and only if that works, we can
2006 * release the parent here already.
2007 */
2008 ppid = spawn_data->sed_parent->p_pid;
2009 if ((!spawn_data->sed_attrs
2010 || (spawn_data->sed_attrs->sa_flags
2011 & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2012 && rw_tryenter(&exec_lock, RW_READER)) {
2013 parent_is_waiting = false;
2014 mutex_enter(&spawn_data->sed_mtx_child);
2015 cv_signal(&spawn_data->sed_cv_child_ready);
2016 mutex_exit(&spawn_data->sed_mtx_child);
2017 }
2018
2019 /* don't allow debugger access yet */
2020 rw_enter(&l->l_proc->p_reflock, RW_WRITER);
2021 have_reflock = true;
2022
2023 error = 0;
2024 /* handle posix_spawn_file_actions */
2025 if (spawn_data->sed_actions != NULL) {
2026 for (i = 0; i < spawn_data->sed_actions->len; i++) {
2027 fae = &spawn_data->sed_actions->fae[i];
2028 switch (fae->fae_action) {
2029 case FAE_OPEN:
2030 if (fd_getfile(fae->fae_fildes) != NULL) {
2031 error = fd_close(fae->fae_fildes);
2032 if (error)
2033 break;
2034 }
2035 error = fd_open(fae->fae_path, fae->fae_oflag,
2036 fae->fae_mode, &newfd);
2037 if (error)
2038 break;
2039 if (newfd != fae->fae_fildes) {
2040 error = dodup(l, newfd,
2041 fae->fae_fildes, 0, &retval);
2042 if (fd_getfile(newfd) != NULL)
2043 fd_close(newfd);
2044 }
2045 break;
2046 case FAE_DUP2:
2047 error = dodup(l, fae->fae_fildes,
2048 fae->fae_newfildes, 0, &retval);
2049 break;
2050 case FAE_CLOSE:
2051 if (fd_getfile(fae->fae_fildes) == NULL) {
2052 error = EBADF;
2053 break;
2054 }
2055 error = fd_close(fae->fae_fildes);
2056 break;
2057 }
2058 if (error)
2059 goto report_error;
2060 }
2061 }
2062
2063 /* handle posix_spawnattr */
2064 if (spawn_data->sed_attrs != NULL) {
2065 struct sigaction sigact;
2066 sigact._sa_u._sa_handler = SIG_DFL;
2067 sigact.sa_flags = 0;
2068
2069 /*
2070 * set state to SSTOP so that this proc can be found by pid.
2071 * see proc_enterprp, do_sched_setparam below
2072 */
2073 mutex_enter(proc_lock);
2074 /*
2075 * p_stat should be SACTIVE, so we need to adjust the
2076 * parent's p_nstopchild here. For safety, just make
2077 * we're on the good side of SDEAD before we adjust.
2078 */
2079 ostat = l->l_proc->p_stat;
2080 KASSERT(ostat < SSTOP);
2081 l->l_proc->p_stat = SSTOP;
2082 l->l_proc->p_waited = 0;
2083 l->l_proc->p_pptr->p_nstopchild++;
2084 mutex_exit(proc_lock);
2085
2086 /* Set process group */
2087 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2088 pid_t mypid = l->l_proc->p_pid,
2089 pgrp = spawn_data->sed_attrs->sa_pgroup;
2090
2091 if (pgrp == 0)
2092 pgrp = mypid;
2093
2094 error = proc_enterpgrp(spawn_data->sed_parent,
2095 mypid, pgrp, false);
2096 if (error)
2097 goto report_error_stopped;
2098 }
2099
2100 /* Set scheduler policy */
2101 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2102 error = do_sched_setparam(l->l_proc->p_pid, 0,
2103 spawn_data->sed_attrs->sa_schedpolicy,
2104 &spawn_data->sed_attrs->sa_schedparam);
2105 else if (spawn_data->sed_attrs->sa_flags
2106 & POSIX_SPAWN_SETSCHEDPARAM) {
2107 error = do_sched_setparam(ppid, 0,
2108 SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2109 }
2110 if (error)
2111 goto report_error_stopped;
2112
2113 /* Reset user ID's */
2114 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2115 error = do_setresuid(l, -1,
2116 kauth_cred_getgid(l->l_cred), -1,
2117 ID_E_EQ_R | ID_E_EQ_S);
2118 if (error)
2119 goto report_error_stopped;
2120 error = do_setresuid(l, -1,
2121 kauth_cred_getuid(l->l_cred), -1,
2122 ID_E_EQ_R | ID_E_EQ_S);
2123 if (error)
2124 goto report_error_stopped;
2125 }
2126
2127 /* Set signal masks/defaults */
2128 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2129 mutex_enter(l->l_proc->p_lock);
2130 error = sigprocmask1(l, SIG_SETMASK,
2131 &spawn_data->sed_attrs->sa_sigmask, NULL);
2132 mutex_exit(l->l_proc->p_lock);
2133 if (error)
2134 goto report_error_stopped;
2135 }
2136
2137 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2138 /*
2139 * The following sigaction call is using a sigaction
2140 * version 0 trampoline which is in the compatibility
2141 * code only. This is not a problem because for SIG_DFL
2142 * and SIG_IGN, the trampolines are now ignored. If they
2143 * were not, this would be a problem because we are
2144 * holding the exec_lock, and the compat code needs
2145 * to do the same in order to replace the trampoline
2146 * code of the process.
2147 */
2148 for (i = 1; i <= NSIG; i++) {
2149 if (sigismember(
2150 &spawn_data->sed_attrs->sa_sigdefault, i))
2151 sigaction1(l, i, &sigact, NULL, NULL,
2152 0);
2153 }
2154 }
2155 mutex_enter(proc_lock);
2156 l->l_proc->p_stat = ostat;
2157 l->l_proc->p_pptr->p_nstopchild--;
2158 mutex_exit(proc_lock);
2159 }
2160
2161 /* now do the real exec */
2162 error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2163 true);
2164 have_reflock = false;
2165 if (error == EJUSTRETURN)
2166 error = 0;
2167 else if (error)
2168 goto report_error;
2169
2170 if (parent_is_waiting) {
2171 mutex_enter(&spawn_data->sed_mtx_child);
2172 cv_signal(&spawn_data->sed_cv_child_ready);
2173 mutex_exit(&spawn_data->sed_mtx_child);
2174 }
2175
2176 /* release our refcount on the data */
2177 spawn_exec_data_release(spawn_data);
2178
2179 /* and finally: leave to userland for the first time */
2180 cpu_spawn_return(l);
2181
2182 /* NOTREACHED */
2183 return;
2184
2185 report_error_stopped:
2186 mutex_enter(proc_lock);
2187 l->l_proc->p_stat = ostat;
2188 l->l_proc->p_pptr->p_nstopchild--;
2189 mutex_exit(proc_lock);
2190 report_error:
2191 if (have_reflock) {
2192 /*
2193 * We have not passed through execve_runproc(),
2194 * which would have released the p_reflock and also
2195 * taken ownership of the sed_exec part of spawn_data,
2196 * so release/free both here.
2197 */
2198 rw_exit(&l->l_proc->p_reflock);
2199 execve_free_data(&spawn_data->sed_exec);
2200 }
2201
2202 if (parent_is_waiting) {
2203 /* pass error to parent */
2204 mutex_enter(&spawn_data->sed_mtx_child);
2205 spawn_data->sed_error = error;
2206 cv_signal(&spawn_data->sed_cv_child_ready);
2207 mutex_exit(&spawn_data->sed_mtx_child);
2208 } else {
2209 rw_exit(&exec_lock);
2210 }
2211
2212 /* release our refcount on the data */
2213 spawn_exec_data_release(spawn_data);
2214
2215 /* done, exit */
2216 mutex_enter(l->l_proc->p_lock);
2217 /*
2218 * Posix explicitly asks for an exit code of 127 if we report
2219 * errors from the child process - so, unfortunately, there
2220 * is no way to report a more exact error code.
2221 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2222 * flag bit in the attrp argument to posix_spawn(2), see above.
2223 */
2224 exit1(l, 127, 0);
2225 }
2226
2227 void
2228 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2229 {
2230
2231 for (size_t i = 0; i < len; i++) {
2232 struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2233 if (fae->fae_action != FAE_OPEN)
2234 continue;
2235 kmem_strfree(fae->fae_path);
2236 }
2237 if (fa->len > 0)
2238 kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2239 kmem_free(fa, sizeof(*fa));
2240 }
2241
2242 static int
2243 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2244 const struct posix_spawn_file_actions *ufa, rlim_t lim)
2245 {
2246 struct posix_spawn_file_actions *fa;
2247 struct posix_spawn_file_actions_entry *fae;
2248 char *pbuf = NULL;
2249 int error;
2250 size_t i = 0;
2251
2252 fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2253 error = copyin(ufa, fa, sizeof(*fa));
2254 if (error || fa->len == 0) {
2255 kmem_free(fa, sizeof(*fa));
2256 return error; /* 0 if not an error, and len == 0 */
2257 }
2258
2259 if (fa->len > lim) {
2260 kmem_free(fa, sizeof(*fa));
2261 return EINVAL;
2262 }
2263
2264 fa->size = fa->len;
2265 size_t fal = fa->len * sizeof(*fae);
2266 fae = fa->fae;
2267 fa->fae = kmem_alloc(fal, KM_SLEEP);
2268 error = copyin(fae, fa->fae, fal);
2269 if (error)
2270 goto out;
2271
2272 pbuf = PNBUF_GET();
2273 for (; i < fa->len; i++) {
2274 fae = &fa->fae[i];
2275 if (fae->fae_action != FAE_OPEN)
2276 continue;
2277 error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2278 if (error)
2279 goto out;
2280 fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2281 memcpy(fae->fae_path, pbuf, fal);
2282 }
2283 PNBUF_PUT(pbuf);
2284
2285 *fap = fa;
2286 return 0;
2287 out:
2288 if (pbuf)
2289 PNBUF_PUT(pbuf);
2290 posix_spawn_fa_free(fa, i);
2291 return error;
2292 }
2293
2294 int
2295 check_posix_spawn(struct lwp *l1)
2296 {
2297 int error, tnprocs, count;
2298 uid_t uid;
2299 struct proc *p1;
2300
2301 p1 = l1->l_proc;
2302 uid = kauth_cred_getuid(l1->l_cred);
2303 tnprocs = atomic_inc_uint_nv(&nprocs);
2304
2305 /*
2306 * Although process entries are dynamically created, we still keep
2307 * a global limit on the maximum number we will create.
2308 */
2309 if (__predict_false(tnprocs >= maxproc))
2310 error = -1;
2311 else
2312 error = kauth_authorize_process(l1->l_cred,
2313 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2314
2315 if (error) {
2316 atomic_dec_uint(&nprocs);
2317 return EAGAIN;
2318 }
2319
2320 /*
2321 * Enforce limits.
2322 */
2323 count = chgproccnt(uid, 1);
2324 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2325 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2326 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2327 __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2328 (void)chgproccnt(uid, -1);
2329 atomic_dec_uint(&nprocs);
2330 return EAGAIN;
2331 }
2332
2333 return 0;
2334 }
2335
2336 int
2337 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2338 struct posix_spawn_file_actions *fa,
2339 struct posix_spawnattr *sa,
2340 char *const *argv, char *const *envp,
2341 execve_fetch_element_t fetch)
2342 {
2343
2344 struct proc *p1, *p2;
2345 struct lwp *l2;
2346 int error;
2347 struct spawn_exec_data *spawn_data;
2348 vaddr_t uaddr;
2349 pid_t pid;
2350 bool have_exec_lock = false;
2351
2352 p1 = l1->l_proc;
2353
2354 /* Allocate and init spawn_data */
2355 spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2356 spawn_data->sed_refcnt = 1; /* only parent so far */
2357 cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2358 mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2359 mutex_enter(&spawn_data->sed_mtx_child);
2360
2361 /*
2362 * Do the first part of the exec now, collect state
2363 * in spawn_data.
2364 */
2365 error = execve_loadvm(l1, path, argv,
2366 envp, fetch, &spawn_data->sed_exec);
2367 if (error == EJUSTRETURN)
2368 error = 0;
2369 else if (error)
2370 goto error_exit;
2371
2372 have_exec_lock = true;
2373
2374 /*
2375 * Allocate virtual address space for the U-area now, while it
2376 * is still easy to abort the fork operation if we're out of
2377 * kernel virtual address space.
2378 */
2379 uaddr = uvm_uarea_alloc();
2380 if (__predict_false(uaddr == 0)) {
2381 error = ENOMEM;
2382 goto error_exit;
2383 }
2384
2385 /*
2386 * Allocate new proc. Borrow proc0 vmspace for it, we will
2387 * replace it with its own before returning to userland
2388 * in the child.
2389 * This is a point of no return, we will have to go through
2390 * the child proc to properly clean it up past this point.
2391 */
2392 p2 = proc_alloc();
2393 pid = p2->p_pid;
2394
2395 /*
2396 * Make a proc table entry for the new process.
2397 * Start by zeroing the section of proc that is zero-initialized,
2398 * then copy the section that is copied directly from the parent.
2399 */
2400 memset(&p2->p_startzero, 0,
2401 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2402 memcpy(&p2->p_startcopy, &p1->p_startcopy,
2403 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2404 p2->p_vmspace = proc0.p_vmspace;
2405
2406 TAILQ_INIT(&p2->p_sigpend.sp_info);
2407
2408 LIST_INIT(&p2->p_lwps);
2409 LIST_INIT(&p2->p_sigwaiters);
2410
2411 /*
2412 * Duplicate sub-structures as needed.
2413 * Increase reference counts on shared objects.
2414 * Inherit flags we want to keep. The flags related to SIGCHLD
2415 * handling are important in order to keep a consistent behaviour
2416 * for the child after the fork. If we are a 32-bit process, the
2417 * child will be too.
2418 */
2419 p2->p_flag =
2420 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2421 p2->p_emul = p1->p_emul;
2422 p2->p_execsw = p1->p_execsw;
2423
2424 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2425 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2426 rw_init(&p2->p_reflock);
2427 cv_init(&p2->p_waitcv, "wait");
2428 cv_init(&p2->p_lwpcv, "lwpwait");
2429
2430 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2431
2432 kauth_proc_fork(p1, p2);
2433
2434 p2->p_raslist = NULL;
2435 p2->p_fd = fd_copy();
2436
2437 /* XXX racy */
2438 p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2439
2440 p2->p_cwdi = cwdinit();
2441
2442 /*
2443 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2444 * we just need increase pl_refcnt.
2445 */
2446 if (!p1->p_limit->pl_writeable) {
2447 lim_addref(p1->p_limit);
2448 p2->p_limit = p1->p_limit;
2449 } else {
2450 p2->p_limit = lim_copy(p1->p_limit);
2451 }
2452
2453 p2->p_lflag = 0;
2454 p2->p_sflag = 0;
2455 p2->p_slflag = 0;
2456 p2->p_pptr = p1;
2457 p2->p_ppid = p1->p_pid;
2458 LIST_INIT(&p2->p_children);
2459
2460 p2->p_aio = NULL;
2461
2462 #ifdef KTRACE
2463 /*
2464 * Copy traceflag and tracefile if enabled.
2465 * If not inherited, these were zeroed above.
2466 */
2467 if (p1->p_traceflag & KTRFAC_INHERIT) {
2468 mutex_enter(&ktrace_lock);
2469 p2->p_traceflag = p1->p_traceflag;
2470 if ((p2->p_tracep = p1->p_tracep) != NULL)
2471 ktradref(p2);
2472 mutex_exit(&ktrace_lock);
2473 }
2474 #endif
2475
2476 /*
2477 * Create signal actions for the child process.
2478 */
2479 p2->p_sigacts = sigactsinit(p1, 0);
2480 mutex_enter(p1->p_lock);
2481 p2->p_sflag |=
2482 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2483 sched_proc_fork(p1, p2);
2484 mutex_exit(p1->p_lock);
2485
2486 p2->p_stflag = p1->p_stflag;
2487
2488 /*
2489 * p_stats.
2490 * Copy parts of p_stats, and zero out the rest.
2491 */
2492 p2->p_stats = pstatscopy(p1->p_stats);
2493
2494 /* copy over machdep flags to the new proc */
2495 cpu_proc_fork(p1, p2);
2496
2497 /*
2498 * Prepare remaining parts of spawn data
2499 */
2500 spawn_data->sed_actions = fa;
2501 spawn_data->sed_attrs = sa;
2502
2503 spawn_data->sed_parent = p1;
2504
2505 /* create LWP */
2506 lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2507 &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
2508 l2->l_ctxlink = NULL; /* reset ucontext link */
2509
2510 /*
2511 * Copy the credential so other references don't see our changes.
2512 * Test to see if this is necessary first, since in the common case
2513 * we won't need a private reference.
2514 */
2515 if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2516 kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2517 l2->l_cred = kauth_cred_copy(l2->l_cred);
2518 kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2519 kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2520 }
2521
2522 /* Update the master credentials. */
2523 if (l2->l_cred != p2->p_cred) {
2524 kauth_cred_t ocred;
2525
2526 kauth_cred_hold(l2->l_cred);
2527 mutex_enter(p2->p_lock);
2528 ocred = p2->p_cred;
2529 p2->p_cred = l2->l_cred;
2530 mutex_exit(p2->p_lock);
2531 kauth_cred_free(ocred);
2532 }
2533
2534 *child_ok = true;
2535 spawn_data->sed_refcnt = 2; /* child gets it as well */
2536 #if 0
2537 l2->l_nopreempt = 1; /* start it non-preemptable */
2538 #endif
2539
2540 /*
2541 * It's now safe for the scheduler and other processes to see the
2542 * child process.
2543 */
2544 mutex_enter(proc_lock);
2545
2546 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2547 p2->p_lflag |= PL_CONTROLT;
2548
2549 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2550 p2->p_exitsig = SIGCHLD; /* signal for parent on exit */
2551
2552 LIST_INSERT_AFTER(p1, p2, p_pglist);
2553 LIST_INSERT_HEAD(&allproc, p2, p_list);
2554
2555 p2->p_trace_enabled = trace_is_enabled(p2);
2556 #ifdef __HAVE_SYSCALL_INTERN
2557 (*p2->p_emul->e_syscall_intern)(p2);
2558 #endif
2559
2560 /*
2561 * Make child runnable, set start time, and add to run queue except
2562 * if the parent requested the child to start in SSTOP state.
2563 */
2564 mutex_enter(p2->p_lock);
2565
2566 getmicrotime(&p2->p_stats->p_start);
2567
2568 lwp_lock(l2);
2569 KASSERT(p2->p_nrlwps == 1);
2570 p2->p_nrlwps = 1;
2571 p2->p_stat = SACTIVE;
2572 l2->l_stat = LSRUN;
2573 sched_enqueue(l2, false);
2574 lwp_unlock(l2);
2575
2576 mutex_exit(p2->p_lock);
2577 mutex_exit(proc_lock);
2578
2579 cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2580 error = spawn_data->sed_error;
2581 mutex_exit(&spawn_data->sed_mtx_child);
2582 spawn_exec_data_release(spawn_data);
2583
2584 rw_exit(&p1->p_reflock);
2585 rw_exit(&exec_lock);
2586 have_exec_lock = false;
2587
2588 *pid_res = pid;
2589 return error;
2590
2591 error_exit:
2592 if (have_exec_lock) {
2593 execve_free_data(&spawn_data->sed_exec);
2594 rw_exit(&p1->p_reflock);
2595 rw_exit(&exec_lock);
2596 }
2597 mutex_exit(&spawn_data->sed_mtx_child);
2598 spawn_exec_data_release(spawn_data);
2599
2600 return error;
2601 }
2602
2603 int
2604 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2605 register_t *retval)
2606 {
2607 /* {
2608 syscallarg(pid_t *) pid;
2609 syscallarg(const char *) path;
2610 syscallarg(const struct posix_spawn_file_actions *) file_actions;
2611 syscallarg(const struct posix_spawnattr *) attrp;
2612 syscallarg(char *const *) argv;
2613 syscallarg(char *const *) envp;
2614 } */
2615
2616 int error;
2617 struct posix_spawn_file_actions *fa = NULL;
2618 struct posix_spawnattr *sa = NULL;
2619 pid_t pid;
2620 bool child_ok = false;
2621 rlim_t max_fileactions;
2622 proc_t *p = l1->l_proc;
2623
2624 error = check_posix_spawn(l1);
2625 if (error) {
2626 *retval = error;
2627 return 0;
2628 }
2629
2630 /* copy in file_actions struct */
2631 if (SCARG(uap, file_actions) != NULL) {
2632 max_fileactions = 2 * min(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2633 maxfiles);
2634 error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2635 max_fileactions);
2636 if (error)
2637 goto error_exit;
2638 }
2639
2640 /* copyin posix_spawnattr struct */
2641 if (SCARG(uap, attrp) != NULL) {
2642 sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2643 error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2644 if (error)
2645 goto error_exit;
2646 }
2647
2648 /*
2649 * Do the spawn
2650 */
2651 error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2652 SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2653 if (error)
2654 goto error_exit;
2655
2656 if (error == 0 && SCARG(uap, pid) != NULL)
2657 error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2658
2659 *retval = error;
2660 return 0;
2661
2662 error_exit:
2663 if (!child_ok) {
2664 (void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2665 atomic_dec_uint(&nprocs);
2666
2667 if (sa)
2668 kmem_free(sa, sizeof(*sa));
2669 if (fa)
2670 posix_spawn_fa_free(fa, fa->len);
2671 }
2672
2673 *retval = error;
2674 return 0;
2675 }
2676
2677 void
2678 exec_free_emul_arg(struct exec_package *epp)
2679 {
2680 if (epp->ep_emul_arg_free != NULL) {
2681 KASSERT(epp->ep_emul_arg != NULL);
2682 (*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2683 epp->ep_emul_arg_free = NULL;
2684 epp->ep_emul_arg = NULL;
2685 } else {
2686 KASSERT(epp->ep_emul_arg == NULL);
2687 }
2688 }
2689
2690 #ifdef DEBUG_EXEC
2691 static void
2692 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2693 {
2694 struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2695 size_t j;
2696
2697 if (error == 0)
2698 DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2699 else
2700 DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2701 epp->ep_vmcmds.evs_used, error));
2702
2703 for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2704 DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2705 PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2706 PRIxVSIZE" prot=0%o flags=%d\n", j,
2707 vp[j].ev_proc == vmcmd_map_pagedvn ?
2708 "pagedvn" :
2709 vp[j].ev_proc == vmcmd_map_readvn ?
2710 "readvn" :
2711 vp[j].ev_proc == vmcmd_map_zero ?
2712 "zero" : "*unknown*",
2713 vp[j].ev_addr, vp[j].ev_len,
2714 vp[j].ev_offset, vp[j].ev_prot,
2715 vp[j].ev_flags));
2716 if (error != 0 && j == x)
2717 DPRINTF((" ^--- failed\n"));
2718 }
2719 }
2720 #endif
2721