1 /* $NetBSD: kern_exec.c,v 1.435 2016/07/07 06:55:43 msaitoh 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 1.435 2016/07/07 06:55:43 msaitoh 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 static size_t calcargs(struct execve_data * restrict, const size_t);
126 static size_t calcstack(struct execve_data * restrict, const size_t);
127 static int copyoutargs(struct execve_data * restrict, struct lwp *,
128 char * const);
129 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
130 static int copyinargs(struct execve_data * restrict, char * const *,
131 char * const *, execve_fetch_element_t, char **);
132 static int copyinargstrs(struct execve_data * restrict, char * const *,
133 execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
134 static int exec_sigcode_map(struct proc *, const struct emul *);
135
136 #if defined(DEBUG) && !defined(DEBUG_EXEC)
137 #define DEBUG_EXEC
138 #endif
139 #ifdef DEBUG_EXEC
140 #define DPRINTF(a) printf a
141 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
142 __LINE__, (s), (a), (b))
143 static void dump_vmcmds(const struct exec_package * const, size_t, int);
144 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
145 #else
146 #define DPRINTF(a)
147 #define COPYPRINTF(s, a, b)
148 #define DUMPVMCMDS(p, x, e) do {} while (0)
149 #endif /* DEBUG_EXEC */
150
151 /*
152 * DTrace SDT provider definitions
153 */
154 SDT_PROVIDER_DECLARE(proc);
155 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
156 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
157 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
158
159 /*
160 * Exec function switch:
161 *
162 * Note that each makecmds function is responsible for loading the
163 * exec package with the necessary functions for any exec-type-specific
164 * handling.
165 *
166 * Functions for specific exec types should be defined in their own
167 * header file.
168 */
169 static const struct execsw **execsw = NULL;
170 static int nexecs;
171
172 u_int exec_maxhdrsz; /* must not be static - used by netbsd32 */
173
174 /* list of dynamically loaded execsw entries */
175 static LIST_HEAD(execlist_head, exec_entry) ex_head =
176 LIST_HEAD_INITIALIZER(ex_head);
177 struct exec_entry {
178 LIST_ENTRY(exec_entry) ex_list;
179 SLIST_ENTRY(exec_entry) ex_slist;
180 const struct execsw *ex_sw;
181 };
182
183 #ifndef __HAVE_SYSCALL_INTERN
184 void syscall(void);
185 #endif
186
187 /* NetBSD autoloadable syscalls */
188 #ifdef MODULAR
189 #include <kern/syscalls_autoload.c>
190 #endif
191
192 /* NetBSD emul struct */
193 struct emul emul_netbsd = {
194 .e_name = "netbsd",
195 #ifdef EMUL_NATIVEROOT
196 .e_path = EMUL_NATIVEROOT,
197 #else
198 .e_path = NULL,
199 #endif
200 #ifndef __HAVE_MINIMAL_EMUL
201 .e_flags = EMUL_HAS_SYS___syscall,
202 .e_errno = NULL,
203 .e_nosys = SYS_syscall,
204 .e_nsysent = SYS_NSYSENT,
205 #endif
206 #ifdef MODULAR
207 .e_sc_autoload = netbsd_syscalls_autoload,
208 #endif
209 .e_sysent = sysent,
210 #ifdef SYSCALL_DEBUG
211 .e_syscallnames = syscallnames,
212 #else
213 .e_syscallnames = NULL,
214 #endif
215 .e_sendsig = sendsig,
216 .e_trapsignal = trapsignal,
217 .e_tracesig = NULL,
218 .e_sigcode = NULL,
219 .e_esigcode = NULL,
220 .e_sigobject = NULL,
221 .e_setregs = setregs,
222 .e_proc_exec = NULL,
223 .e_proc_fork = NULL,
224 .e_proc_exit = NULL,
225 .e_lwp_fork = NULL,
226 .e_lwp_exit = NULL,
227 #ifdef __HAVE_SYSCALL_INTERN
228 .e_syscall_intern = syscall_intern,
229 #else
230 .e_syscall = syscall,
231 #endif
232 .e_sysctlovly = NULL,
233 .e_fault = NULL,
234 .e_vm_default_addr = uvm_default_mapaddr,
235 .e_usertrap = NULL,
236 .e_ucsize = sizeof(ucontext_t),
237 .e_startlwp = startlwp
238 };
239
240 /*
241 * Exec lock. Used to control access to execsw[] structures.
242 * This must not be static so that netbsd32 can access it, too.
243 */
244 krwlock_t exec_lock;
245
246 static kmutex_t sigobject_lock;
247
248 /*
249 * Data used between a loadvm and execve part of an "exec" operation
250 */
251 struct execve_data {
252 struct exec_package ed_pack;
253 struct pathbuf *ed_pathbuf;
254 struct vattr ed_attr;
255 struct ps_strings ed_arginfo;
256 char *ed_argp;
257 const char *ed_pathstring;
258 char *ed_resolvedpathbuf;
259 size_t ed_ps_strings_sz;
260 int ed_szsigcode;
261 size_t ed_argslen;
262 long ed_argc;
263 long ed_envc;
264 };
265
266 /*
267 * data passed from parent lwp to child during a posix_spawn()
268 */
269 struct spawn_exec_data {
270 struct execve_data sed_exec;
271 struct posix_spawn_file_actions
272 *sed_actions;
273 struct posix_spawnattr *sed_attrs;
274 struct proc *sed_parent;
275 kcondvar_t sed_cv_child_ready;
276 kmutex_t sed_mtx_child;
277 int sed_error;
278 volatile uint32_t sed_refcnt;
279 };
280
281 static void *
exec_pool_alloc(struct pool * pp,int flags)282 exec_pool_alloc(struct pool *pp, int flags)
283 {
284
285 return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
286 UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
287 }
288
289 static void
exec_pool_free(struct pool * pp,void * addr)290 exec_pool_free(struct pool *pp, void *addr)
291 {
292
293 uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
294 }
295
296 static struct pool exec_pool;
297
298 static struct pool_allocator exec_palloc = {
299 .pa_alloc = exec_pool_alloc,
300 .pa_free = exec_pool_free,
301 .pa_pagesz = NCARGS
302 };
303
304 /*
305 * check exec:
306 * given an "executable" described in the exec package's namei info,
307 * see what we can do with it.
308 *
309 * ON ENTRY:
310 * exec package with appropriate namei info
311 * lwp pointer of exec'ing lwp
312 * NO SELF-LOCKED VNODES
313 *
314 * ON EXIT:
315 * error: nothing held, etc. exec header still allocated.
316 * ok: filled exec package, executable's vnode (unlocked).
317 *
318 * EXEC SWITCH ENTRY:
319 * Locked vnode to check, exec package, proc.
320 *
321 * EXEC SWITCH EXIT:
322 * ok: return 0, filled exec package, executable's vnode (unlocked).
323 * error: destructive:
324 * everything deallocated execept exec header.
325 * non-destructive:
326 * error code, executable's vnode (unlocked),
327 * exec header unmodified.
328 */
329 int
330 /*ARGSUSED*/
check_exec(struct lwp * l,struct exec_package * epp,struct pathbuf * pb)331 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
332 {
333 int error, i;
334 struct vnode *vp;
335 struct nameidata nd;
336 size_t resid;
337
338 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
339
340 /* first get the vnode */
341 if ((error = namei(&nd)) != 0)
342 return error;
343 epp->ep_vp = vp = nd.ni_vp;
344 /* normally this can't fail */
345 error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL);
346 KASSERT(error == 0);
347
348 #ifdef DIAGNOSTIC
349 /* paranoia (take this out once namei stuff stabilizes) */
350 memset(nd.ni_pnbuf, '~', PATH_MAX);
351 #endif
352
353 /* check access and type */
354 if (vp->v_type != VREG) {
355 error = EACCES;
356 goto bad1;
357 }
358 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
359 goto bad1;
360
361 /* get attributes */
362 if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
363 goto bad1;
364
365 /* Check mount point */
366 if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
367 error = EACCES;
368 goto bad1;
369 }
370 if (vp->v_mount->mnt_flag & MNT_NOSUID)
371 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
372
373 /* try to open it */
374 if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
375 goto bad1;
376
377 /* unlock vp, since we need it unlocked from here on out. */
378 VOP_UNLOCK(vp);
379
380 #if NVERIEXEC > 0
381 error = veriexec_verify(l, vp, epp->ep_resolvedname,
382 epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
383 NULL);
384 if (error)
385 goto bad2;
386 #endif /* NVERIEXEC > 0 */
387
388 #ifdef PAX_SEGVGUARD
389 error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
390 if (error)
391 goto bad2;
392 #endif /* PAX_SEGVGUARD */
393
394 /* now we have the file, get the exec header */
395 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
396 UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
397 if (error)
398 goto bad2;
399 epp->ep_hdrvalid = epp->ep_hdrlen - resid;
400
401 /*
402 * Set up default address space limits. Can be overridden
403 * by individual exec packages.
404 *
405 * XXX probably should be all done in the exec packages.
406 */
407 epp->ep_vm_minaddr = VM_MIN_ADDRESS;
408 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
409 /*
410 * set up the vmcmds for creation of the process
411 * address space
412 */
413 error = ENOEXEC;
414 for (i = 0; i < nexecs; i++) {
415 int newerror;
416
417 epp->ep_esch = execsw[i];
418 newerror = (*execsw[i]->es_makecmds)(l, epp);
419
420 if (!newerror) {
421 /* Seems ok: check that entry point is not too high */
422 if (epp->ep_entry > epp->ep_vm_maxaddr) {
423 #ifdef DIAGNOSTIC
424 printf("%s: rejecting %p due to "
425 "too high entry address (> %p)\n",
426 __func__, (void *)epp->ep_entry,
427 (void *)epp->ep_vm_maxaddr);
428 #endif
429 error = ENOEXEC;
430 break;
431 }
432 /* Seems ok: check that entry point is not too low */
433 if (epp->ep_entry < epp->ep_vm_minaddr) {
434 #ifdef DIAGNOSTIC
435 printf("%s: rejecting %p due to "
436 "too low entry address (< %p)\n",
437 __func__, (void *)epp->ep_entry,
438 (void *)epp->ep_vm_minaddr);
439 #endif
440 error = ENOEXEC;
441 break;
442 }
443
444 /* check limits */
445 if ((epp->ep_tsize > MAXTSIZ) ||
446 (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
447 [RLIMIT_DATA].rlim_cur)) {
448 #ifdef DIAGNOSTIC
449 printf("%s: rejecting due to "
450 "limits (t=%llu > %llu || d=%llu > %llu)\n",
451 __func__,
452 (unsigned long long)epp->ep_tsize,
453 (unsigned long long)MAXTSIZ,
454 (unsigned long long)epp->ep_dsize,
455 (unsigned long long)
456 l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
457 #endif
458 error = ENOMEM;
459 break;
460 }
461 return 0;
462 }
463
464 /*
465 * Reset all the fields that may have been modified by the
466 * loader.
467 */
468 KASSERT(epp->ep_emul_arg == NULL);
469 if (epp->ep_emul_root != NULL) {
470 vrele(epp->ep_emul_root);
471 epp->ep_emul_root = NULL;
472 }
473 if (epp->ep_interp != NULL) {
474 vrele(epp->ep_interp);
475 epp->ep_interp = NULL;
476 }
477 epp->ep_pax_flags = 0;
478
479 /* make sure the first "interesting" error code is saved. */
480 if (error == ENOEXEC)
481 error = newerror;
482
483 if (epp->ep_flags & EXEC_DESTR)
484 /* Error from "#!" code, tidied up by recursive call */
485 return error;
486 }
487
488 /* not found, error */
489
490 /*
491 * free any vmspace-creation commands,
492 * and release their references
493 */
494 kill_vmcmds(&epp->ep_vmcmds);
495
496 bad2:
497 /*
498 * close and release the vnode, restore the old one, free the
499 * pathname buf, and punt.
500 */
501 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
502 VOP_CLOSE(vp, FREAD, l->l_cred);
503 vput(vp);
504 return error;
505
506 bad1:
507 /*
508 * free the namei pathname buffer, and put the vnode
509 * (which we don't yet have open).
510 */
511 vput(vp); /* was still locked */
512 return error;
513 }
514
515 #ifdef __MACHINE_STACK_GROWS_UP
516 #define STACK_PTHREADSPACE NBPG
517 #else
518 #define STACK_PTHREADSPACE 0
519 #endif
520
521 static int
execve_fetch_element(char * const * array,size_t index,char ** value)522 execve_fetch_element(char * const *array, size_t index, char **value)
523 {
524 return copyin(array + index, value, sizeof(*value));
525 }
526
527 /*
528 * exec system call
529 */
530 int
sys_execve(struct lwp * l,const struct sys_execve_args * uap,register_t * retval)531 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
532 {
533 /* {
534 syscallarg(const char *) path;
535 syscallarg(char * const *) argp;
536 syscallarg(char * const *) envp;
537 } */
538
539 return execve1(l, SCARG(uap, path), SCARG(uap, argp),
540 SCARG(uap, envp), execve_fetch_element);
541 }
542
543 int
sys_fexecve(struct lwp * l,const struct sys_fexecve_args * uap,register_t * retval)544 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
545 register_t *retval)
546 {
547 /* {
548 syscallarg(int) fd;
549 syscallarg(char * const *) argp;
550 syscallarg(char * const *) envp;
551 } */
552
553 return ENOSYS;
554 }
555
556 /*
557 * Load modules to try and execute an image that we do not understand.
558 * If no execsw entries are present, we load those likely to be needed
559 * in order to run native images only. Otherwise, we autoload all
560 * possible modules that could let us run the binary. XXX lame
561 */
562 static void
exec_autoload(void)563 exec_autoload(void)
564 {
565 #ifdef MODULAR
566 static const char * const native[] = {
567 "exec_elf32",
568 "exec_elf64",
569 "exec_script",
570 NULL
571 };
572 static const char * const compat[] = {
573 "exec_elf32",
574 "exec_elf64",
575 "exec_script",
576 "exec_aout",
577 "exec_coff",
578 "exec_ecoff",
579 "compat_aoutm68k",
580 "compat_freebsd",
581 "compat_ibcs2",
582 "compat_linux",
583 "compat_linux32",
584 "compat_netbsd32",
585 "compat_sunos",
586 "compat_sunos32",
587 "compat_svr4",
588 "compat_svr4_32",
589 "compat_ultrix",
590 NULL
591 };
592 char const * const *list;
593 int i;
594
595 list = (nexecs == 0 ? native : compat);
596 for (i = 0; list[i] != NULL; i++) {
597 if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
598 continue;
599 }
600 yield();
601 }
602 #endif
603 }
604
605 static int
makepathbuf(struct lwp * l,const char * upath,struct pathbuf ** pbp,size_t * offs)606 makepathbuf(struct lwp *l, const char *upath, struct pathbuf **pbp,
607 size_t *offs)
608 {
609 char *path, *bp;
610 size_t len, tlen;
611 int error;
612 struct cwdinfo *cwdi;
613
614 path = PNBUF_GET();
615 error = copyinstr(upath, path, MAXPATHLEN, &len);
616 if (error) {
617 PNBUF_PUT(path);
618 DPRINTF(("%s: copyin path @%p %d\n", __func__, upath, error));
619 return error;
620 }
621
622 if (path[0] == '/') {
623 *offs = 0;
624 goto out;
625 }
626
627 len++;
628 if (len + 1 >= MAXPATHLEN)
629 goto out;
630 bp = path + MAXPATHLEN - len;
631 memmove(bp, path, len);
632 *(--bp) = '/';
633
634 cwdi = l->l_proc->p_cwdi;
635 rw_enter(&cwdi->cwdi_lock, RW_READER);
636 error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
637 GETCWD_CHECK_ACCESS, l);
638 rw_exit(&cwdi->cwdi_lock);
639
640 if (error) {
641 DPRINTF(("%s: getcwd_common path %s %d\n", __func__, path,
642 error));
643 goto out;
644 }
645 tlen = path + MAXPATHLEN - bp;
646
647 memmove(path, bp, tlen);
648 path[tlen] = '\0';
649 *offs = tlen - len;
650 out:
651 *pbp = pathbuf_assimilate(path);
652 return 0;
653 }
654
655 static int
execve_loadvm(struct lwp * l,const char * path,char * const * args,char * const * envs,execve_fetch_element_t fetch_element,struct execve_data * restrict data)656 execve_loadvm(struct lwp *l, const char *path, char * const *args,
657 char * const *envs, execve_fetch_element_t fetch_element,
658 struct execve_data * restrict data)
659 {
660 struct exec_package * const epp = &data->ed_pack;
661 int error;
662 struct proc *p;
663 char *dp;
664 u_int modgen;
665 size_t offs = 0; // XXX: GCC
666
667 KASSERT(data != NULL);
668
669 p = l->l_proc;
670 modgen = 0;
671
672 SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
673
674 /*
675 * Check if we have exceeded our number of processes limit.
676 * This is so that we handle the case where a root daemon
677 * forked, ran setuid to become the desired user and is trying
678 * to exec. The obvious place to do the reference counting check
679 * is setuid(), but we don't do the reference counting check there
680 * like other OS's do because then all the programs that use setuid()
681 * must be modified to check the return code of setuid() and exit().
682 * It is dangerous to make setuid() fail, because it fails open and
683 * the program will continue to run as root. If we make it succeed
684 * and return an error code, again we are not enforcing the limit.
685 * The best place to enforce the limit is here, when the process tries
686 * to execute a new image, because eventually the process will need
687 * to call exec in order to do something useful.
688 */
689 retry:
690 if (p->p_flag & PK_SUGID) {
691 if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
692 p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
693 &p->p_rlimit[RLIMIT_NPROC],
694 KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
695 chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
696 p->p_rlimit[RLIMIT_NPROC].rlim_cur)
697 return EAGAIN;
698 }
699
700 /*
701 * Drain existing references and forbid new ones. The process
702 * should be left alone until we're done here. This is necessary
703 * to avoid race conditions - e.g. in ptrace() - that might allow
704 * a local user to illicitly obtain elevated privileges.
705 */
706 rw_enter(&p->p_reflock, RW_WRITER);
707
708 /*
709 * Init the namei data to point the file user's program name.
710 * This is done here rather than in check_exec(), so that it's
711 * possible to override this settings if any of makecmd/probe
712 * functions call check_exec() recursively - for example,
713 * see exec_script_makecmds().
714 */
715 if ((error = makepathbuf(l, path, &data->ed_pathbuf, &offs)) != 0)
716 goto clrflg;
717 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
718 data->ed_resolvedpathbuf = PNBUF_GET();
719
720 /*
721 * initialize the fields of the exec package.
722 */
723 epp->ep_kname = data->ed_pathstring + offs;
724 epp->ep_resolvedname = data->ed_resolvedpathbuf;
725 epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
726 epp->ep_hdrlen = exec_maxhdrsz;
727 epp->ep_hdrvalid = 0;
728 epp->ep_emul_arg = NULL;
729 epp->ep_emul_arg_free = NULL;
730 memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
731 epp->ep_vap = &data->ed_attr;
732 epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
733 MD_TOPDOWN_INIT(epp);
734 epp->ep_emul_root = NULL;
735 epp->ep_interp = NULL;
736 epp->ep_esch = NULL;
737 epp->ep_pax_flags = 0;
738 memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
739
740 rw_enter(&exec_lock, RW_READER);
741
742 /* see if we can run it. */
743 if ((error = check_exec(l, epp, data->ed_pathbuf)) != 0) {
744 if (error != ENOENT && error != EACCES) {
745 DPRINTF(("%s: check exec failed %d\n",
746 __func__, error));
747 }
748 goto freehdr;
749 }
750
751 /* allocate an argument buffer */
752 data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
753 KASSERT(data->ed_argp != NULL);
754 dp = data->ed_argp;
755
756 if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
757 goto bad;
758 }
759
760 /*
761 * Calculate the new stack size.
762 */
763
764 #ifdef __MACHINE_STACK_GROWS_UP
765 /*
766 * copyargs() fills argc/argv/envp from the lower address even on
767 * __MACHINE_STACK_GROWS_UP machines. Reserve a few words just below the SP
768 * so that _rtld() use it.
769 */
770 #define RTLD_GAP 32
771 #else
772 #define RTLD_GAP 0
773 #endif
774
775 const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
776
777 data->ed_argslen = calcargs(data, argenvstrlen);
778
779 const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
780
781 if (len > epp->ep_ssize) {
782 /* in effect, compare to initial limit */
783 DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
784 error = ENOMEM;
785 goto bad;
786 }
787 /* adjust "active stack depth" for process VSZ */
788 epp->ep_ssize = len;
789
790 return 0;
791
792 bad:
793 /* free the vmspace-creation commands, and release their references */
794 kill_vmcmds(&epp->ep_vmcmds);
795 /* kill any opened file descriptor, if necessary */
796 if (epp->ep_flags & EXEC_HASFD) {
797 epp->ep_flags &= ~EXEC_HASFD;
798 fd_close(epp->ep_fd);
799 }
800 /* close and put the exec'd file */
801 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
802 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
803 vput(epp->ep_vp);
804 pool_put(&exec_pool, data->ed_argp);
805
806 freehdr:
807 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
808 if (epp->ep_emul_root != NULL)
809 vrele(epp->ep_emul_root);
810 if (epp->ep_interp != NULL)
811 vrele(epp->ep_interp);
812
813 rw_exit(&exec_lock);
814
815 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
816 pathbuf_destroy(data->ed_pathbuf);
817 PNBUF_PUT(data->ed_resolvedpathbuf);
818
819 clrflg:
820 rw_exit(&p->p_reflock);
821
822 if (modgen != module_gen && error == ENOEXEC) {
823 modgen = module_gen;
824 exec_autoload();
825 goto retry;
826 }
827
828 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
829 return error;
830 }
831
832 static int
execve_dovmcmds(struct lwp * l,struct execve_data * restrict data)833 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
834 {
835 struct exec_package * const epp = &data->ed_pack;
836 struct proc *p = l->l_proc;
837 struct exec_vmcmd *base_vcp;
838 int error = 0;
839 size_t i;
840
841 /* record proc's vnode, for use by procfs and others */
842 if (p->p_textvp)
843 vrele(p->p_textvp);
844 vref(epp->ep_vp);
845 p->p_textvp = epp->ep_vp;
846
847 /* create the new process's VM space by running the vmcmds */
848 KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
849
850 #ifdef TRACE_EXEC
851 DUMPVMCMDS(epp, 0, 0);
852 #endif
853
854 base_vcp = NULL;
855
856 for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
857 struct exec_vmcmd *vcp;
858
859 vcp = &epp->ep_vmcmds.evs_cmds[i];
860 if (vcp->ev_flags & VMCMD_RELATIVE) {
861 KASSERTMSG(base_vcp != NULL,
862 "%s: relative vmcmd with no base", __func__);
863 KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
864 "%s: illegal base & relative vmcmd", __func__);
865 vcp->ev_addr += base_vcp->ev_addr;
866 }
867 error = (*vcp->ev_proc)(l, vcp);
868 if (error)
869 DUMPVMCMDS(epp, i, error);
870 if (vcp->ev_flags & VMCMD_BASE)
871 base_vcp = vcp;
872 }
873
874 /* free the vmspace-creation commands, and release their references */
875 kill_vmcmds(&epp->ep_vmcmds);
876
877 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
878 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
879 vput(epp->ep_vp);
880
881 /* if an error happened, deallocate and punt */
882 if (error != 0) {
883 DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
884 }
885 return error;
886 }
887
888 static void
execve_free_data(struct execve_data * data)889 execve_free_data(struct execve_data *data)
890 {
891 struct exec_package * const epp = &data->ed_pack;
892
893 /* free the vmspace-creation commands, and release their references */
894 kill_vmcmds(&epp->ep_vmcmds);
895 /* kill any opened file descriptor, if necessary */
896 if (epp->ep_flags & EXEC_HASFD) {
897 epp->ep_flags &= ~EXEC_HASFD;
898 fd_close(epp->ep_fd);
899 }
900
901 /* close and put the exec'd file */
902 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
903 VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
904 vput(epp->ep_vp);
905 pool_put(&exec_pool, data->ed_argp);
906
907 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
908 if (epp->ep_emul_root != NULL)
909 vrele(epp->ep_emul_root);
910 if (epp->ep_interp != NULL)
911 vrele(epp->ep_interp);
912
913 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
914 pathbuf_destroy(data->ed_pathbuf);
915 PNBUF_PUT(data->ed_resolvedpathbuf);
916 }
917
918 static void
pathexec(struct exec_package * epp,struct lwp * l,const char * pathstring)919 pathexec(struct exec_package *epp, struct lwp *l, const char *pathstring)
920 {
921 const char *commandname;
922 size_t commandlen;
923 char *path;
924 struct proc *p = l->l_proc;
925
926 /* set command name & other accounting info */
927 commandname = strrchr(epp->ep_resolvedname, '/');
928 if (commandname != NULL) {
929 commandname++;
930 } else {
931 commandname = epp->ep_resolvedname;
932 }
933 commandlen = min(strlen(commandname), MAXCOMLEN);
934 (void)memcpy(p->p_comm, commandname, commandlen);
935 p->p_comm[commandlen] = '\0';
936
937
938 /*
939 * If the path starts with /, we don't need to do any work.
940 * This handles the majority of the cases.
941 * In the future perhaps we could canonicalize it?
942 */
943 path = PNBUF_GET();
944 if (pathstring[0] == '/') {
945 (void)strlcpy(path, pathstring, MAXPATHLEN);
946 epp->ep_path = path;
947 }
948 #ifdef notyet
949 /*
950 * Although this works most of the time [since the entry was just
951 * entered in the cache] we don't use it because it will fail for
952 * entries that are not placed in the cache because their name is
953 * longer than NCHNAMLEN and it is not the cleanest interface,
954 * because there could be races. When the namei cache is re-written,
955 * this can be changed to use the appropriate function.
956 */
957 else if (!(error = vnode_to_path(path, MAXPATHLEN, p->p_textvp, l, p)))
958 epp->ep_path = path;
959 #endif
960 else {
961 #ifdef notyet
962 printf("Cannot get path for pid %d [%s] (error %d)\n",
963 (int)p->p_pid, p->p_comm, error);
964 #endif
965 PNBUF_PUT(path);
966 epp->ep_path = NULL;
967 }
968 }
969
970 /* XXX elsewhere */
971 static int
credexec(struct lwp * l,struct vattr * attr)972 credexec(struct lwp *l, struct vattr *attr)
973 {
974 struct proc *p = l->l_proc;
975 int error;
976
977 /*
978 * Deal with set[ug]id. MNT_NOSUID has already been used to disable
979 * s[ug]id. It's OK to check for PSL_TRACED here as we have blocked
980 * out additional references on the process for the moment.
981 */
982 if ((p->p_slflag & PSL_TRACED) == 0 &&
983
984 (((attr->va_mode & S_ISUID) != 0 &&
985 kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
986
987 ((attr->va_mode & S_ISGID) != 0 &&
988 kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
989 /*
990 * Mark the process as SUGID before we do
991 * anything that might block.
992 */
993 proc_crmod_enter();
994 proc_crmod_leave(NULL, NULL, true);
995
996 /* Make sure file descriptors 0..2 are in use. */
997 if ((error = fd_checkstd()) != 0) {
998 DPRINTF(("%s: fdcheckstd failed %d\n",
999 __func__, error));
1000 return error;
1001 }
1002
1003 /*
1004 * Copy the credential so other references don't see our
1005 * changes.
1006 */
1007 l->l_cred = kauth_cred_copy(l->l_cred);
1008 #ifdef KTRACE
1009 /*
1010 * If the persistent trace flag isn't set, turn off.
1011 */
1012 if (p->p_tracep) {
1013 mutex_enter(&ktrace_lock);
1014 if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1015 ktrderef(p);
1016 mutex_exit(&ktrace_lock);
1017 }
1018 #endif
1019 if (attr->va_mode & S_ISUID)
1020 kauth_cred_seteuid(l->l_cred, attr->va_uid);
1021 if (attr->va_mode & S_ISGID)
1022 kauth_cred_setegid(l->l_cred, attr->va_gid);
1023 } else {
1024 if (kauth_cred_geteuid(l->l_cred) ==
1025 kauth_cred_getuid(l->l_cred) &&
1026 kauth_cred_getegid(l->l_cred) ==
1027 kauth_cred_getgid(l->l_cred))
1028 p->p_flag &= ~PK_SUGID;
1029 }
1030
1031 /*
1032 * Copy the credential so other references don't see our changes.
1033 * Test to see if this is necessary first, since in the common case
1034 * we won't need a private reference.
1035 */
1036 if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1037 kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1038 l->l_cred = kauth_cred_copy(l->l_cred);
1039 kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1040 kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1041 }
1042
1043 /* Update the master credentials. */
1044 if (l->l_cred != p->p_cred) {
1045 kauth_cred_t ocred;
1046
1047 kauth_cred_hold(l->l_cred);
1048 mutex_enter(p->p_lock);
1049 ocred = p->p_cred;
1050 p->p_cred = l->l_cred;
1051 mutex_exit(p->p_lock);
1052 kauth_cred_free(ocred);
1053 }
1054
1055 return 0;
1056 }
1057
1058 static void
emulexec(struct lwp * l,struct exec_package * epp)1059 emulexec(struct lwp *l, struct exec_package *epp)
1060 {
1061 struct proc *p = l->l_proc;
1062
1063 /* The emulation root will usually have been found when we looked
1064 * for the elf interpreter (or similar), if not look now. */
1065 if (epp->ep_esch->es_emul->e_path != NULL &&
1066 epp->ep_emul_root == NULL)
1067 emul_find_root(l, epp);
1068
1069 /* Any old emulation root got removed by fdcloseexec */
1070 rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1071 p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1072 rw_exit(&p->p_cwdi->cwdi_lock);
1073 epp->ep_emul_root = NULL;
1074 if (epp->ep_interp != NULL)
1075 vrele(epp->ep_interp);
1076
1077 /*
1078 * Call emulation specific exec hook. This can setup per-process
1079 * p->p_emuldata or do any other per-process stuff an emulation needs.
1080 *
1081 * If we are executing process of different emulation than the
1082 * original forked process, call e_proc_exit() of the old emulation
1083 * first, then e_proc_exec() of new emulation. If the emulation is
1084 * same, the exec hook code should deallocate any old emulation
1085 * resources held previously by this process.
1086 */
1087 if (p->p_emul && p->p_emul->e_proc_exit
1088 && p->p_emul != epp->ep_esch->es_emul)
1089 (*p->p_emul->e_proc_exit)(p);
1090
1091 /*
1092 * This is now LWP 1.
1093 */
1094 /* XXX elsewhere */
1095 mutex_enter(p->p_lock);
1096 p->p_nlwpid = 1;
1097 l->l_lid = 1;
1098 mutex_exit(p->p_lock);
1099
1100 /*
1101 * Call exec hook. Emulation code may NOT store reference to anything
1102 * from &pack.
1103 */
1104 if (epp->ep_esch->es_emul->e_proc_exec)
1105 (*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1106
1107 /* update p_emul, the old value is no longer needed */
1108 p->p_emul = epp->ep_esch->es_emul;
1109
1110 /* ...and the same for p_execsw */
1111 p->p_execsw = epp->ep_esch;
1112
1113 #ifdef __HAVE_SYSCALL_INTERN
1114 (*p->p_emul->e_syscall_intern)(p);
1115 #endif
1116 ktremul();
1117 }
1118
1119 static int
execve_runproc(struct lwp * l,struct execve_data * restrict data,bool no_local_exec_lock,bool is_spawn)1120 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1121 bool no_local_exec_lock, bool is_spawn)
1122 {
1123 struct exec_package * const epp = &data->ed_pack;
1124 int error = 0;
1125 struct proc *p;
1126
1127 /*
1128 * In case of a posix_spawn operation, the child doing the exec
1129 * might not hold the reader lock on exec_lock, but the parent
1130 * will do this instead.
1131 */
1132 KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1133 KASSERT(!no_local_exec_lock || is_spawn);
1134 KASSERT(data != NULL);
1135
1136 p = l->l_proc;
1137
1138 /* Get rid of other LWPs. */
1139 if (p->p_nlwps > 1) {
1140 mutex_enter(p->p_lock);
1141 exit_lwps(l);
1142 mutex_exit(p->p_lock);
1143 }
1144 KDASSERT(p->p_nlwps == 1);
1145
1146 /* Destroy any lwpctl info. */
1147 if (p->p_lwpctl != NULL)
1148 lwp_ctl_exit();
1149
1150 /* Remove POSIX timers */
1151 timers_free(p, TIMERS_POSIX);
1152
1153 /* Set the PaX flags. */
1154 pax_set_flags(epp, p);
1155
1156 /*
1157 * Do whatever is necessary to prepare the address space
1158 * for remapping. Note that this might replace the current
1159 * vmspace with another!
1160 */
1161 if (is_spawn)
1162 uvmspace_spawn(l, epp->ep_vm_minaddr,
1163 epp->ep_vm_maxaddr,
1164 epp->ep_flags & EXEC_TOPDOWN_VM);
1165 else
1166 uvmspace_exec(l, epp->ep_vm_minaddr,
1167 epp->ep_vm_maxaddr,
1168 epp->ep_flags & EXEC_TOPDOWN_VM);
1169
1170 struct vmspace *vm;
1171 vm = p->p_vmspace;
1172 vm->vm_taddr = (void *)epp->ep_taddr;
1173 vm->vm_tsize = btoc(epp->ep_tsize);
1174 vm->vm_daddr = (void*)epp->ep_daddr;
1175 vm->vm_dsize = btoc(epp->ep_dsize);
1176 vm->vm_ssize = btoc(epp->ep_ssize);
1177 vm->vm_issize = 0;
1178 vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1179 vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1180
1181 pax_aslr_init_vm(l, vm, epp);
1182
1183 /* Now map address space. */
1184 error = execve_dovmcmds(l, data);
1185 if (error != 0)
1186 goto exec_abort;
1187
1188 pathexec(epp, l, data->ed_pathstring);
1189
1190 char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1191
1192 error = copyoutargs(data, l, newstack);
1193 if (error != 0)
1194 goto exec_abort;
1195
1196 cwdexec(p);
1197 fd_closeexec(); /* handle close on exec */
1198
1199 if (__predict_false(ktrace_on))
1200 fd_ktrexecfd();
1201
1202 execsigs(p); /* reset catched signals */
1203
1204 mutex_enter(p->p_lock);
1205 l->l_ctxlink = NULL; /* reset ucontext link */
1206 p->p_acflag &= ~AFORK;
1207 p->p_flag |= PK_EXEC;
1208 mutex_exit(p->p_lock);
1209
1210 /*
1211 * Stop profiling.
1212 */
1213 if ((p->p_stflag & PST_PROFIL) != 0) {
1214 mutex_spin_enter(&p->p_stmutex);
1215 stopprofclock(p);
1216 mutex_spin_exit(&p->p_stmutex);
1217 }
1218
1219 /*
1220 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1221 * exited and exec()/exit() are the only places it will be cleared.
1222 */
1223 if ((p->p_lflag & PL_PPWAIT) != 0) {
1224 #if 0
1225 lwp_t *lp;
1226
1227 mutex_enter(proc_lock);
1228 lp = p->p_vforklwp;
1229 p->p_vforklwp = NULL;
1230
1231 l->l_lwpctl = NULL; /* was on loan from blocked parent */
1232 p->p_lflag &= ~PL_PPWAIT;
1233
1234 lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
1235 cv_broadcast(&lp->l_waitcv);
1236 mutex_exit(proc_lock);
1237 #else
1238 mutex_enter(proc_lock);
1239 l->l_lwpctl = NULL; /* was on loan from blocked parent */
1240 p->p_lflag &= ~PL_PPWAIT;
1241 cv_broadcast(&p->p_pptr->p_waitcv);
1242 mutex_exit(proc_lock);
1243 #endif
1244 }
1245
1246 error = credexec(l, &data->ed_attr);
1247 if (error)
1248 goto exec_abort;
1249
1250 #if defined(__HAVE_RAS)
1251 /*
1252 * Remove all RASs from the address space.
1253 */
1254 ras_purgeall();
1255 #endif
1256
1257 doexechooks(p);
1258
1259 /*
1260 * Set initial SP at the top of the stack.
1261 *
1262 * Note that on machines where stack grows up (e.g. hppa), SP points to
1263 * the end of arg/env strings. Userland guesses the address of argc
1264 * via ps_strings::ps_argvstr.
1265 */
1266
1267 /* Setup new registers and do misc. setup. */
1268 (*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1269 if (epp->ep_esch->es_setregs)
1270 (*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1271
1272 /* Provide a consistent LWP private setting */
1273 (void)lwp_setprivate(l, NULL);
1274
1275 /* Discard all PCU state; need to start fresh */
1276 pcu_discard_all(l);
1277
1278 /* map the process's signal trampoline code */
1279 if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1280 DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1281 goto exec_abort;
1282 }
1283
1284 pool_put(&exec_pool, data->ed_argp);
1285
1286 /* notify others that we exec'd */
1287 KNOTE(&p->p_klist, NOTE_EXEC);
1288
1289 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1290
1291 SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1292
1293 emulexec(l, epp);
1294
1295 /* Allow new references from the debugger/procfs. */
1296 rw_exit(&p->p_reflock);
1297 if (!no_local_exec_lock)
1298 rw_exit(&exec_lock);
1299
1300 mutex_enter(proc_lock);
1301
1302 if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1303 ksiginfo_t ksi;
1304
1305 KSI_INIT_EMPTY(&ksi);
1306 ksi.ksi_signo = SIGTRAP;
1307 ksi.ksi_lid = l->l_lid;
1308 kpsignal(p, &ksi, NULL);
1309 }
1310
1311 if (p->p_sflag & PS_STOPEXEC) {
1312 ksiginfoq_t kq;
1313
1314 KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1315 p->p_pptr->p_nstopchild++;
1316 p->p_waited = 0;
1317 mutex_enter(p->p_lock);
1318 ksiginfo_queue_init(&kq);
1319 sigclearall(p, &contsigmask, &kq);
1320 lwp_lock(l);
1321 l->l_stat = LSSTOP;
1322 p->p_stat = SSTOP;
1323 p->p_nrlwps--;
1324 lwp_unlock(l);
1325 mutex_exit(p->p_lock);
1326 mutex_exit(proc_lock);
1327 lwp_lock(l);
1328 mi_switch(l);
1329 ksiginfo_queue_drain(&kq);
1330 KERNEL_LOCK(l->l_biglocks, l);
1331 } else {
1332 mutex_exit(proc_lock);
1333 }
1334
1335 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1336 pathbuf_destroy(data->ed_pathbuf);
1337 PNBUF_PUT(data->ed_resolvedpathbuf);
1338 #ifdef TRACE_EXEC
1339 DPRINTF(("%s finished\n", __func__));
1340 #endif
1341 return EJUSTRETURN;
1342
1343 exec_abort:
1344 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1345 rw_exit(&p->p_reflock);
1346 if (!no_local_exec_lock)
1347 rw_exit(&exec_lock);
1348
1349 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1350 pathbuf_destroy(data->ed_pathbuf);
1351 PNBUF_PUT(data->ed_resolvedpathbuf);
1352
1353 /*
1354 * the old process doesn't exist anymore. exit gracefully.
1355 * get rid of the (new) address space we have created, if any, get rid
1356 * of our namei data and vnode, and exit noting failure
1357 */
1358 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1359 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1360
1361 exec_free_emul_arg(epp);
1362 pool_put(&exec_pool, data->ed_argp);
1363 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1364 if (epp->ep_emul_root != NULL)
1365 vrele(epp->ep_emul_root);
1366 if (epp->ep_interp != NULL)
1367 vrele(epp->ep_interp);
1368
1369 /* Acquire the sched-state mutex (exit1() will release it). */
1370 if (!is_spawn) {
1371 mutex_enter(p->p_lock);
1372 exit1(l, error, SIGABRT);
1373 }
1374
1375 return error;
1376 }
1377
1378 int
execve1(struct lwp * l,const char * path,char * const * args,char * const * envs,execve_fetch_element_t fetch_element)1379 execve1(struct lwp *l, const char *path, char * const *args,
1380 char * const *envs, execve_fetch_element_t fetch_element)
1381 {
1382 struct execve_data data;
1383 int error;
1384
1385 error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1386 if (error)
1387 return error;
1388 error = execve_runproc(l, &data, false, false);
1389 return error;
1390 }
1391
1392 static size_t
fromptrsz(const struct exec_package * epp)1393 fromptrsz(const struct exec_package *epp)
1394 {
1395 return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1396 }
1397
1398 static size_t
ptrsz(const struct exec_package * epp)1399 ptrsz(const struct exec_package *epp)
1400 {
1401 return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1402 }
1403
1404 static size_t
calcargs(struct execve_data * restrict data,const size_t argenvstrlen)1405 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1406 {
1407 struct exec_package * const epp = &data->ed_pack;
1408
1409 const size_t nargenvptrs =
1410 1 + /* long argc */
1411 data->ed_argc + /* char *argv[] */
1412 1 + /* \0 */
1413 data->ed_envc + /* char *env[] */
1414 1 + /* \0 */
1415 epp->ep_esch->es_arglen; /* auxinfo */
1416
1417 return (nargenvptrs * ptrsz(epp)) + argenvstrlen;
1418 }
1419
1420 static size_t
calcstack(struct execve_data * restrict data,const size_t gaplen)1421 calcstack(struct execve_data * restrict data, const size_t gaplen)
1422 {
1423 struct exec_package * const epp = &data->ed_pack;
1424
1425 data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1426 epp->ep_esch->es_emul->e_sigcode;
1427
1428 data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1429 sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1430
1431 const size_t sigcode_psstr_sz =
1432 data->ed_szsigcode + /* sigcode */
1433 data->ed_ps_strings_sz + /* ps_strings */
1434 STACK_PTHREADSPACE; /* pthread space */
1435
1436 const size_t stacklen =
1437 data->ed_argslen +
1438 gaplen +
1439 sigcode_psstr_sz;
1440
1441 /* make the stack "safely" aligned */
1442 return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1443 }
1444
1445 static int
copyoutargs(struct execve_data * restrict data,struct lwp * l,char * const newstack)1446 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1447 char * const newstack)
1448 {
1449 struct exec_package * const epp = &data->ed_pack;
1450 struct proc *p = l->l_proc;
1451 int error;
1452
1453 /* remember information about the process */
1454 data->ed_arginfo.ps_nargvstr = data->ed_argc;
1455 data->ed_arginfo.ps_nenvstr = data->ed_envc;
1456
1457 /*
1458 * Allocate the stack address passed to the newly execve()'ed process.
1459 *
1460 * The new stack address will be set to the SP (stack pointer) register
1461 * in setregs().
1462 */
1463
1464 char *newargs = STACK_ALLOC(
1465 STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1466
1467 error = (*epp->ep_esch->es_copyargs)(l, epp,
1468 &data->ed_arginfo, &newargs, data->ed_argp);
1469
1470 if (epp->ep_path) {
1471 PNBUF_PUT(epp->ep_path);
1472 epp->ep_path = NULL;
1473 }
1474 if (error) {
1475 DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1476 return error;
1477 }
1478
1479 error = copyoutpsstrs(data, p);
1480 if (error != 0)
1481 return error;
1482
1483 return 0;
1484 }
1485
1486 static int
copyoutpsstrs(struct execve_data * restrict data,struct proc * p)1487 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1488 {
1489 struct exec_package * const epp = &data->ed_pack;
1490 struct ps_strings32 arginfo32;
1491 void *aip;
1492 int error;
1493
1494 /* fill process ps_strings info */
1495 p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1496 STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1497
1498 if (epp->ep_flags & EXEC_32) {
1499 aip = &arginfo32;
1500 arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1501 arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1502 arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1503 arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1504 } else
1505 aip = &data->ed_arginfo;
1506
1507 /* copy out the process's ps_strings structure */
1508 if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1509 != 0) {
1510 DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1511 __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1512 return error;
1513 }
1514
1515 return 0;
1516 }
1517
1518 static int
copyinargs(struct execve_data * restrict data,char * const * args,char * const * envs,execve_fetch_element_t fetch_element,char ** dpp)1519 copyinargs(struct execve_data * restrict data, char * const *args,
1520 char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1521 {
1522 struct exec_package * const epp = &data->ed_pack;
1523 char *dp;
1524 size_t i;
1525 int error;
1526
1527 dp = *dpp;
1528
1529 data->ed_argc = 0;
1530
1531 /* copy the fake args list, if there's one, freeing it as we go */
1532 if (epp->ep_flags & EXEC_HASARGL) {
1533 struct exec_fakearg *fa = epp->ep_fa;
1534
1535 while (fa->fa_arg != NULL) {
1536 const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1537 size_t len;
1538
1539 len = strlcpy(dp, fa->fa_arg, maxlen);
1540 /* Count NUL into len. */
1541 if (len < maxlen)
1542 len++;
1543 else {
1544 while (fa->fa_arg != NULL) {
1545 kmem_free(fa->fa_arg, fa->fa_len);
1546 fa++;
1547 }
1548 kmem_free(epp->ep_fa, epp->ep_fa_len);
1549 epp->ep_flags &= ~EXEC_HASARGL;
1550 return E2BIG;
1551 }
1552 ktrexecarg(fa->fa_arg, len - 1);
1553 dp += len;
1554
1555 kmem_free(fa->fa_arg, fa->fa_len);
1556 fa++;
1557 data->ed_argc++;
1558 }
1559 kmem_free(epp->ep_fa, epp->ep_fa_len);
1560 epp->ep_flags &= ~EXEC_HASARGL;
1561 }
1562
1563 /*
1564 * Read and count argument strings from user.
1565 */
1566
1567 if (args == NULL) {
1568 DPRINTF(("%s: null args\n", __func__));
1569 return EINVAL;
1570 }
1571 if (epp->ep_flags & EXEC_SKIPARG)
1572 args = (const void *)((const char *)args + fromptrsz(epp));
1573 i = 0;
1574 error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1575 if (error != 0) {
1576 DPRINTF(("%s: copyin arg %d\n", __func__, error));
1577 return error;
1578 }
1579 data->ed_argc += i;
1580
1581 /*
1582 * Read and count environment strings from user.
1583 */
1584
1585 data->ed_envc = 0;
1586 /* environment need not be there */
1587 if (envs == NULL)
1588 goto done;
1589 i = 0;
1590 error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1591 if (error != 0) {
1592 DPRINTF(("%s: copyin env %d\n", __func__, error));
1593 return error;
1594 }
1595 data->ed_envc += i;
1596
1597 done:
1598 *dpp = dp;
1599
1600 return 0;
1601 }
1602
1603 static int
copyinargstrs(struct execve_data * restrict data,char * const * strs,execve_fetch_element_t fetch_element,char ** dpp,size_t * ip,void (* ktr)(const void *,size_t))1604 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1605 execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1606 void (*ktr)(const void *, size_t))
1607 {
1608 char *dp, *sp;
1609 size_t i;
1610 int error;
1611
1612 dp = *dpp;
1613
1614 i = 0;
1615 while (1) {
1616 const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1617 size_t len;
1618
1619 if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1620 return error;
1621 }
1622 if (!sp)
1623 break;
1624 if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1625 if (error == ENAMETOOLONG)
1626 error = E2BIG;
1627 return error;
1628 }
1629 if (__predict_false(ktrace_on))
1630 (*ktr)(dp, len - 1);
1631 dp += len;
1632 i++;
1633 }
1634
1635 *dpp = dp;
1636 *ip = i;
1637
1638 return 0;
1639 }
1640
1641 /*
1642 * Copy argv and env strings from kernel buffer (argp) to the new stack.
1643 * Those strings are located just after auxinfo.
1644 */
1645 int
copyargs(struct lwp * l,struct exec_package * pack,struct ps_strings * arginfo,char ** stackp,void * argp)1646 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1647 char **stackp, void *argp)
1648 {
1649 char **cpp, *dp, *sp;
1650 size_t len;
1651 void *nullp;
1652 long argc, envc;
1653 int error;
1654
1655 cpp = (char **)*stackp;
1656 nullp = NULL;
1657 argc = arginfo->ps_nargvstr;
1658 envc = arginfo->ps_nenvstr;
1659
1660 /* argc on stack is long */
1661 CTASSERT(sizeof(*cpp) == sizeof(argc));
1662
1663 dp = (char *)(cpp +
1664 1 + /* long argc */
1665 argc + /* char *argv[] */
1666 1 + /* \0 */
1667 envc + /* char *env[] */
1668 1 + /* \0 */
1669 /* XXX auxinfo multiplied by ptr size? */
1670 pack->ep_esch->es_arglen); /* auxinfo */
1671 sp = argp;
1672
1673 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1674 COPYPRINTF("", cpp - 1, sizeof(argc));
1675 return error;
1676 }
1677
1678 /* XXX don't copy them out, remap them! */
1679 arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1680
1681 for (; --argc >= 0; sp += len, dp += len) {
1682 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1683 COPYPRINTF("", cpp - 1, sizeof(dp));
1684 return error;
1685 }
1686 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1687 COPYPRINTF("str", dp, (size_t)ARG_MAX);
1688 return error;
1689 }
1690 }
1691
1692 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1693 COPYPRINTF("", cpp - 1, sizeof(nullp));
1694 return error;
1695 }
1696
1697 arginfo->ps_envstr = cpp; /* remember location of envp for later */
1698
1699 for (; --envc >= 0; sp += len, dp += len) {
1700 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1701 COPYPRINTF("", cpp - 1, sizeof(dp));
1702 return error;
1703 }
1704 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1705 COPYPRINTF("str", dp, (size_t)ARG_MAX);
1706 return error;
1707 }
1708
1709 }
1710
1711 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1712 COPYPRINTF("", cpp - 1, sizeof(nullp));
1713 return error;
1714 }
1715
1716 *stackp = (char *)cpp;
1717 return 0;
1718 }
1719
1720
1721 /*
1722 * Add execsw[] entries.
1723 */
1724 int
exec_add(struct execsw * esp,int count)1725 exec_add(struct execsw *esp, int count)
1726 {
1727 struct exec_entry *it;
1728 int i;
1729
1730 if (count == 0) {
1731 return 0;
1732 }
1733
1734 /* Check for duplicates. */
1735 rw_enter(&exec_lock, RW_WRITER);
1736 for (i = 0; i < count; i++) {
1737 LIST_FOREACH(it, &ex_head, ex_list) {
1738 /* assume unique (makecmds, probe_func, emulation) */
1739 if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1740 it->ex_sw->u.elf_probe_func ==
1741 esp[i].u.elf_probe_func &&
1742 it->ex_sw->es_emul == esp[i].es_emul) {
1743 rw_exit(&exec_lock);
1744 return EEXIST;
1745 }
1746 }
1747 }
1748
1749 /* Allocate new entries. */
1750 for (i = 0; i < count; i++) {
1751 it = kmem_alloc(sizeof(*it), KM_SLEEP);
1752 it->ex_sw = &esp[i];
1753 LIST_INSERT_HEAD(&ex_head, it, ex_list);
1754 }
1755
1756 /* update execsw[] */
1757 exec_init(0);
1758 rw_exit(&exec_lock);
1759 return 0;
1760 }
1761
1762 /*
1763 * Remove execsw[] entry.
1764 */
1765 int
exec_remove(struct execsw * esp,int count)1766 exec_remove(struct execsw *esp, int count)
1767 {
1768 struct exec_entry *it, *next;
1769 int i;
1770 const struct proclist_desc *pd;
1771 proc_t *p;
1772
1773 if (count == 0) {
1774 return 0;
1775 }
1776
1777 /* Abort if any are busy. */
1778 rw_enter(&exec_lock, RW_WRITER);
1779 for (i = 0; i < count; i++) {
1780 mutex_enter(proc_lock);
1781 for (pd = proclists; pd->pd_list != NULL; pd++) {
1782 PROCLIST_FOREACH(p, pd->pd_list) {
1783 if (p->p_execsw == &esp[i]) {
1784 mutex_exit(proc_lock);
1785 rw_exit(&exec_lock);
1786 return EBUSY;
1787 }
1788 }
1789 }
1790 mutex_exit(proc_lock);
1791 }
1792
1793 /* None are busy, so remove them all. */
1794 for (i = 0; i < count; i++) {
1795 for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1796 next = LIST_NEXT(it, ex_list);
1797 if (it->ex_sw == &esp[i]) {
1798 LIST_REMOVE(it, ex_list);
1799 kmem_free(it, sizeof(*it));
1800 break;
1801 }
1802 }
1803 }
1804
1805 /* update execsw[] */
1806 exec_init(0);
1807 rw_exit(&exec_lock);
1808 return 0;
1809 }
1810
1811 /*
1812 * Initialize exec structures. If init_boot is true, also does necessary
1813 * one-time initialization (it's called from main() that way).
1814 * Once system is multiuser, this should be called with exec_lock held,
1815 * i.e. via exec_{add|remove}().
1816 */
1817 int
exec_init(int init_boot)1818 exec_init(int init_boot)
1819 {
1820 const struct execsw **sw;
1821 struct exec_entry *ex;
1822 SLIST_HEAD(,exec_entry) first;
1823 SLIST_HEAD(,exec_entry) any;
1824 SLIST_HEAD(,exec_entry) last;
1825 int i, sz;
1826
1827 if (init_boot) {
1828 /* do one-time initializations */
1829 rw_init(&exec_lock);
1830 mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1831 pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1832 "execargs", &exec_palloc, IPL_NONE);
1833 pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1834 } else {
1835 KASSERT(rw_write_held(&exec_lock));
1836 }
1837
1838 /* Sort each entry onto the appropriate queue. */
1839 SLIST_INIT(&first);
1840 SLIST_INIT(&any);
1841 SLIST_INIT(&last);
1842 sz = 0;
1843 LIST_FOREACH(ex, &ex_head, ex_list) {
1844 switch(ex->ex_sw->es_prio) {
1845 case EXECSW_PRIO_FIRST:
1846 SLIST_INSERT_HEAD(&first, ex, ex_slist);
1847 break;
1848 case EXECSW_PRIO_ANY:
1849 SLIST_INSERT_HEAD(&any, ex, ex_slist);
1850 break;
1851 case EXECSW_PRIO_LAST:
1852 SLIST_INSERT_HEAD(&last, ex, ex_slist);
1853 break;
1854 default:
1855 panic("%s", __func__);
1856 break;
1857 }
1858 sz++;
1859 }
1860
1861 /*
1862 * Create new execsw[]. Ensure we do not try a zero-sized
1863 * allocation.
1864 */
1865 sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1866 i = 0;
1867 SLIST_FOREACH(ex, &first, ex_slist) {
1868 sw[i++] = ex->ex_sw;
1869 }
1870 SLIST_FOREACH(ex, &any, ex_slist) {
1871 sw[i++] = ex->ex_sw;
1872 }
1873 SLIST_FOREACH(ex, &last, ex_slist) {
1874 sw[i++] = ex->ex_sw;
1875 }
1876
1877 /* Replace old execsw[] and free used memory. */
1878 if (execsw != NULL) {
1879 kmem_free(__UNCONST(execsw),
1880 nexecs * sizeof(struct execsw *) + 1);
1881 }
1882 execsw = sw;
1883 nexecs = sz;
1884
1885 /* Figure out the maximum size of an exec header. */
1886 exec_maxhdrsz = sizeof(int);
1887 for (i = 0; i < nexecs; i++) {
1888 if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1889 exec_maxhdrsz = execsw[i]->es_hdrsz;
1890 }
1891
1892 return 0;
1893 }
1894
1895 static int
exec_sigcode_map(struct proc * p,const struct emul * e)1896 exec_sigcode_map(struct proc *p, const struct emul *e)
1897 {
1898 vaddr_t va;
1899 vsize_t sz;
1900 int error;
1901 struct uvm_object *uobj;
1902
1903 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1904
1905 if (e->e_sigobject == NULL || sz == 0) {
1906 return 0;
1907 }
1908
1909 /*
1910 * If we don't have a sigobject for this emulation, create one.
1911 *
1912 * sigobject is an anonymous memory object (just like SYSV shared
1913 * memory) that we keep a permanent reference to and that we map
1914 * in all processes that need this sigcode. The creation is simple,
1915 * we create an object, add a permanent reference to it, map it in
1916 * kernel space, copy out the sigcode to it and unmap it.
1917 * We map it with PROT_READ|PROT_EXEC into the process just
1918 * the way sys_mmap() would map it.
1919 */
1920
1921 uobj = *e->e_sigobject;
1922 if (uobj == NULL) {
1923 mutex_enter(&sigobject_lock);
1924 if ((uobj = *e->e_sigobject) == NULL) {
1925 uobj = uao_create(sz, 0);
1926 (*uobj->pgops->pgo_reference)(uobj);
1927 va = vm_map_min(kernel_map);
1928 if ((error = uvm_map(kernel_map, &va, round_page(sz),
1929 uobj, 0, 0,
1930 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1931 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1932 printf("kernel mapping failed %d\n", error);
1933 (*uobj->pgops->pgo_detach)(uobj);
1934 mutex_exit(&sigobject_lock);
1935 return error;
1936 }
1937 memcpy((void *)va, e->e_sigcode, sz);
1938 #ifdef PMAP_NEED_PROCWR
1939 pmap_procwr(&proc0, va, sz);
1940 #endif
1941 uvm_unmap(kernel_map, va, va + round_page(sz));
1942 *e->e_sigobject = uobj;
1943 }
1944 mutex_exit(&sigobject_lock);
1945 }
1946
1947 /* Just a hint to uvm_map where to put it. */
1948 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1949 round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
1950
1951 #ifdef __alpha__
1952 /*
1953 * Tru64 puts /sbin/loader at the end of user virtual memory,
1954 * which causes the above calculation to put the sigcode at
1955 * an invalid address. Put it just below the text instead.
1956 */
1957 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1958 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1959 }
1960 #endif
1961
1962 (*uobj->pgops->pgo_reference)(uobj);
1963 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1964 uobj, 0, 0,
1965 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1966 UVM_ADV_RANDOM, 0));
1967 if (error) {
1968 DPRINTF(("%s, %d: map %p "
1969 "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1970 __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1971 va, error));
1972 (*uobj->pgops->pgo_detach)(uobj);
1973 return error;
1974 }
1975 p->p_sigctx.ps_sigcode = (void *)va;
1976 return 0;
1977 }
1978
1979 /*
1980 * Release a refcount on spawn_exec_data and destroy memory, if this
1981 * was the last one.
1982 */
1983 static void
spawn_exec_data_release(struct spawn_exec_data * data)1984 spawn_exec_data_release(struct spawn_exec_data *data)
1985 {
1986 if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1987 return;
1988
1989 cv_destroy(&data->sed_cv_child_ready);
1990 mutex_destroy(&data->sed_mtx_child);
1991
1992 if (data->sed_actions)
1993 posix_spawn_fa_free(data->sed_actions,
1994 data->sed_actions->len);
1995 if (data->sed_attrs)
1996 kmem_free(data->sed_attrs,
1997 sizeof(*data->sed_attrs));
1998 kmem_free(data, sizeof(*data));
1999 }
2000
2001 /*
2002 * A child lwp of a posix_spawn operation starts here and ends up in
2003 * cpu_spawn_return, dealing with all filedescriptor and scheduler
2004 * manipulations in between.
2005 * The parent waits for the child, as it is not clear whether the child
2006 * will be able to acquire its own exec_lock. If it can, the parent can
2007 * be released early and continue running in parallel. If not (or if the
2008 * magic debug flag is passed in the scheduler attribute struct), the
2009 * child rides on the parent's exec lock until it is ready to return to
2010 * to userland - and only then releases the parent. This method loses
2011 * concurrency, but improves error reporting.
2012 */
2013 static void
spawn_return(void * arg)2014 spawn_return(void *arg)
2015 {
2016 struct spawn_exec_data *spawn_data = arg;
2017 struct lwp *l = curlwp;
2018 int error, newfd;
2019 int ostat;
2020 size_t i;
2021 const struct posix_spawn_file_actions_entry *fae;
2022 pid_t ppid;
2023 register_t retval;
2024 bool have_reflock;
2025 bool parent_is_waiting = true;
2026
2027 /*
2028 * Check if we can release parent early.
2029 * We either need to have no sed_attrs, or sed_attrs does not
2030 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2031 * safe access to the parent proc (passed in sed_parent).
2032 * We then try to get the exec_lock, and only if that works, we can
2033 * release the parent here already.
2034 */
2035 ppid = spawn_data->sed_parent->p_pid;
2036 if ((!spawn_data->sed_attrs
2037 || (spawn_data->sed_attrs->sa_flags
2038 & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2039 && rw_tryenter(&exec_lock, RW_READER)) {
2040 parent_is_waiting = false;
2041 mutex_enter(&spawn_data->sed_mtx_child);
2042 cv_signal(&spawn_data->sed_cv_child_ready);
2043 mutex_exit(&spawn_data->sed_mtx_child);
2044 }
2045
2046 /* don't allow debugger access yet */
2047 rw_enter(&l->l_proc->p_reflock, RW_WRITER);
2048 have_reflock = true;
2049
2050 error = 0;
2051 /* handle posix_spawn_file_actions */
2052 if (spawn_data->sed_actions != NULL) {
2053 for (i = 0; i < spawn_data->sed_actions->len; i++) {
2054 fae = &spawn_data->sed_actions->fae[i];
2055 switch (fae->fae_action) {
2056 case FAE_OPEN:
2057 if (fd_getfile(fae->fae_fildes) != NULL) {
2058 error = fd_close(fae->fae_fildes);
2059 if (error)
2060 break;
2061 }
2062 error = fd_open(fae->fae_path, fae->fae_oflag,
2063 fae->fae_mode, &newfd);
2064 if (error)
2065 break;
2066 if (newfd != fae->fae_fildes) {
2067 error = dodup(l, newfd,
2068 fae->fae_fildes, 0, &retval);
2069 if (fd_getfile(newfd) != NULL)
2070 fd_close(newfd);
2071 }
2072 break;
2073 case FAE_DUP2:
2074 error = dodup(l, fae->fae_fildes,
2075 fae->fae_newfildes, 0, &retval);
2076 break;
2077 case FAE_CLOSE:
2078 if (fd_getfile(fae->fae_fildes) == NULL) {
2079 error = EBADF;
2080 break;
2081 }
2082 error = fd_close(fae->fae_fildes);
2083 break;
2084 }
2085 if (error)
2086 goto report_error;
2087 }
2088 }
2089
2090 /* handle posix_spawnattr */
2091 if (spawn_data->sed_attrs != NULL) {
2092 struct sigaction sigact;
2093 sigact._sa_u._sa_handler = SIG_DFL;
2094 sigact.sa_flags = 0;
2095
2096 /*
2097 * set state to SSTOP so that this proc can be found by pid.
2098 * see proc_enterprp, do_sched_setparam below
2099 */
2100 mutex_enter(proc_lock);
2101 /*
2102 * p_stat should be SACTIVE, so we need to adjust the
2103 * parent's p_nstopchild here. For safety, just make
2104 * we're on the good side of SDEAD before we adjust.
2105 */
2106 ostat = l->l_proc->p_stat;
2107 KASSERT(ostat < SSTOP);
2108 l->l_proc->p_stat = SSTOP;
2109 l->l_proc->p_waited = 0;
2110 l->l_proc->p_pptr->p_nstopchild++;
2111 mutex_exit(proc_lock);
2112
2113 /* Set process group */
2114 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2115 pid_t mypid = l->l_proc->p_pid,
2116 pgrp = spawn_data->sed_attrs->sa_pgroup;
2117
2118 if (pgrp == 0)
2119 pgrp = mypid;
2120
2121 error = proc_enterpgrp(spawn_data->sed_parent,
2122 mypid, pgrp, false);
2123 if (error)
2124 goto report_error_stopped;
2125 }
2126
2127 /* Set scheduler policy */
2128 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2129 error = do_sched_setparam(l->l_proc->p_pid, 0,
2130 spawn_data->sed_attrs->sa_schedpolicy,
2131 &spawn_data->sed_attrs->sa_schedparam);
2132 else if (spawn_data->sed_attrs->sa_flags
2133 & POSIX_SPAWN_SETSCHEDPARAM) {
2134 error = do_sched_setparam(ppid, 0,
2135 SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2136 }
2137 if (error)
2138 goto report_error_stopped;
2139
2140 /* Reset user ID's */
2141 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2142 error = do_setresuid(l, -1,
2143 kauth_cred_getgid(l->l_cred), -1,
2144 ID_E_EQ_R | ID_E_EQ_S);
2145 if (error)
2146 goto report_error_stopped;
2147 error = do_setresuid(l, -1,
2148 kauth_cred_getuid(l->l_cred), -1,
2149 ID_E_EQ_R | ID_E_EQ_S);
2150 if (error)
2151 goto report_error_stopped;
2152 }
2153
2154 /* Set signal masks/defaults */
2155 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2156 mutex_enter(l->l_proc->p_lock);
2157 error = sigprocmask1(l, SIG_SETMASK,
2158 &spawn_data->sed_attrs->sa_sigmask, NULL);
2159 mutex_exit(l->l_proc->p_lock);
2160 if (error)
2161 goto report_error_stopped;
2162 }
2163
2164 if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2165 /*
2166 * The following sigaction call is using a sigaction
2167 * version 0 trampoline which is in the compatibility
2168 * code only. This is not a problem because for SIG_DFL
2169 * and SIG_IGN, the trampolines are now ignored. If they
2170 * were not, this would be a problem because we are
2171 * holding the exec_lock, and the compat code needs
2172 * to do the same in order to replace the trampoline
2173 * code of the process.
2174 */
2175 for (i = 1; i <= NSIG; i++) {
2176 if (sigismember(
2177 &spawn_data->sed_attrs->sa_sigdefault, i))
2178 sigaction1(l, i, &sigact, NULL, NULL,
2179 0);
2180 }
2181 }
2182 mutex_enter(proc_lock);
2183 l->l_proc->p_stat = ostat;
2184 l->l_proc->p_pptr->p_nstopchild--;
2185 mutex_exit(proc_lock);
2186 }
2187
2188 /* now do the real exec */
2189 error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2190 true);
2191 have_reflock = false;
2192 if (error == EJUSTRETURN)
2193 error = 0;
2194 else if (error)
2195 goto report_error;
2196
2197 if (parent_is_waiting) {
2198 mutex_enter(&spawn_data->sed_mtx_child);
2199 cv_signal(&spawn_data->sed_cv_child_ready);
2200 mutex_exit(&spawn_data->sed_mtx_child);
2201 }
2202
2203 /* release our refcount on the data */
2204 spawn_exec_data_release(spawn_data);
2205
2206 /* and finally: leave to userland for the first time */
2207 cpu_spawn_return(l);
2208
2209 /* NOTREACHED */
2210 return;
2211
2212 report_error_stopped:
2213 mutex_enter(proc_lock);
2214 l->l_proc->p_stat = ostat;
2215 l->l_proc->p_pptr->p_nstopchild--;
2216 mutex_exit(proc_lock);
2217 report_error:
2218 if (have_reflock) {
2219 /*
2220 * We have not passed through execve_runproc(),
2221 * which would have released the p_reflock and also
2222 * taken ownership of the sed_exec part of spawn_data,
2223 * so release/free both here.
2224 */
2225 rw_exit(&l->l_proc->p_reflock);
2226 execve_free_data(&spawn_data->sed_exec);
2227 }
2228
2229 if (parent_is_waiting) {
2230 /* pass error to parent */
2231 mutex_enter(&spawn_data->sed_mtx_child);
2232 spawn_data->sed_error = error;
2233 cv_signal(&spawn_data->sed_cv_child_ready);
2234 mutex_exit(&spawn_data->sed_mtx_child);
2235 } else {
2236 rw_exit(&exec_lock);
2237 }
2238
2239 /* release our refcount on the data */
2240 spawn_exec_data_release(spawn_data);
2241
2242 /* done, exit */
2243 mutex_enter(l->l_proc->p_lock);
2244 /*
2245 * Posix explicitly asks for an exit code of 127 if we report
2246 * errors from the child process - so, unfortunately, there
2247 * is no way to report a more exact error code.
2248 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2249 * flag bit in the attrp argument to posix_spawn(2), see above.
2250 */
2251 exit1(l, 127, 0);
2252 }
2253
2254 void
posix_spawn_fa_free(struct posix_spawn_file_actions * fa,size_t len)2255 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2256 {
2257
2258 for (size_t i = 0; i < len; i++) {
2259 struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2260 if (fae->fae_action != FAE_OPEN)
2261 continue;
2262 kmem_free(fae->fae_path, strlen(fae->fae_path) + 1);
2263 }
2264 if (fa->len > 0)
2265 kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2266 kmem_free(fa, sizeof(*fa));
2267 }
2268
2269 static int
posix_spawn_fa_alloc(struct posix_spawn_file_actions ** fap,const struct posix_spawn_file_actions * ufa,rlim_t lim)2270 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2271 const struct posix_spawn_file_actions *ufa, rlim_t lim)
2272 {
2273 struct posix_spawn_file_actions *fa;
2274 struct posix_spawn_file_actions_entry *fae;
2275 char *pbuf = NULL;
2276 int error;
2277 size_t i = 0;
2278
2279 fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2280 error = copyin(ufa, fa, sizeof(*fa));
2281 if (error || fa->len == 0) {
2282 kmem_free(fa, sizeof(*fa));
2283 return error; /* 0 if not an error, and len == 0 */
2284 }
2285
2286 if (fa->len > lim) {
2287 kmem_free(fa, sizeof(*fa));
2288 return EINVAL;
2289 }
2290
2291 fa->size = fa->len;
2292 size_t fal = fa->len * sizeof(*fae);
2293 fae = fa->fae;
2294 fa->fae = kmem_alloc(fal, KM_SLEEP);
2295 error = copyin(fae, fa->fae, fal);
2296 if (error)
2297 goto out;
2298
2299 pbuf = PNBUF_GET();
2300 for (; i < fa->len; i++) {
2301 fae = &fa->fae[i];
2302 if (fae->fae_action != FAE_OPEN)
2303 continue;
2304 error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2305 if (error)
2306 goto out;
2307 fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2308 memcpy(fae->fae_path, pbuf, fal);
2309 }
2310 PNBUF_PUT(pbuf);
2311
2312 *fap = fa;
2313 return 0;
2314 out:
2315 if (pbuf)
2316 PNBUF_PUT(pbuf);
2317 posix_spawn_fa_free(fa, i);
2318 return error;
2319 }
2320
2321 int
check_posix_spawn(struct lwp * l1)2322 check_posix_spawn(struct lwp *l1)
2323 {
2324 int error, tnprocs, count;
2325 uid_t uid;
2326 struct proc *p1;
2327
2328 p1 = l1->l_proc;
2329 uid = kauth_cred_getuid(l1->l_cred);
2330 tnprocs = atomic_inc_uint_nv(&nprocs);
2331
2332 /*
2333 * Although process entries are dynamically created, we still keep
2334 * a global limit on the maximum number we will create.
2335 */
2336 if (__predict_false(tnprocs >= maxproc))
2337 error = -1;
2338 else
2339 error = kauth_authorize_process(l1->l_cred,
2340 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2341
2342 if (error) {
2343 atomic_dec_uint(&nprocs);
2344 return EAGAIN;
2345 }
2346
2347 /*
2348 * Enforce limits.
2349 */
2350 count = chgproccnt(uid, 1);
2351 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2352 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2353 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2354 __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2355 (void)chgproccnt(uid, -1);
2356 atomic_dec_uint(&nprocs);
2357 return EAGAIN;
2358 }
2359
2360 return 0;
2361 }
2362
2363 int
do_posix_spawn(struct lwp * l1,pid_t * pid_res,bool * child_ok,const char * path,struct posix_spawn_file_actions * fa,struct posix_spawnattr * sa,char * const * argv,char * const * envp,execve_fetch_element_t fetch)2364 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2365 struct posix_spawn_file_actions *fa,
2366 struct posix_spawnattr *sa,
2367 char *const *argv, char *const *envp,
2368 execve_fetch_element_t fetch)
2369 {
2370
2371 struct proc *p1, *p2;
2372 struct lwp *l2;
2373 int error;
2374 struct spawn_exec_data *spawn_data;
2375 vaddr_t uaddr;
2376 pid_t pid;
2377 bool have_exec_lock = false;
2378
2379 p1 = l1->l_proc;
2380
2381 /* Allocate and init spawn_data */
2382 spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2383 spawn_data->sed_refcnt = 1; /* only parent so far */
2384 cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2385 mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2386 mutex_enter(&spawn_data->sed_mtx_child);
2387
2388 /*
2389 * Do the first part of the exec now, collect state
2390 * in spawn_data.
2391 */
2392 error = execve_loadvm(l1, path, argv,
2393 envp, fetch, &spawn_data->sed_exec);
2394 if (error == EJUSTRETURN)
2395 error = 0;
2396 else if (error)
2397 goto error_exit;
2398
2399 have_exec_lock = true;
2400
2401 /*
2402 * Allocate virtual address space for the U-area now, while it
2403 * is still easy to abort the fork operation if we're out of
2404 * kernel virtual address space.
2405 */
2406 uaddr = uvm_uarea_alloc();
2407 if (__predict_false(uaddr == 0)) {
2408 error = ENOMEM;
2409 goto error_exit;
2410 }
2411
2412 /*
2413 * Allocate new proc. Borrow proc0 vmspace for it, we will
2414 * replace it with its own before returning to userland
2415 * in the child.
2416 * This is a point of no return, we will have to go through
2417 * the child proc to properly clean it up past this point.
2418 */
2419 p2 = proc_alloc();
2420 pid = p2->p_pid;
2421
2422 /*
2423 * Make a proc table entry for the new process.
2424 * Start by zeroing the section of proc that is zero-initialized,
2425 * then copy the section that is copied directly from the parent.
2426 */
2427 memset(&p2->p_startzero, 0,
2428 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2429 memcpy(&p2->p_startcopy, &p1->p_startcopy,
2430 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2431 p2->p_vmspace = proc0.p_vmspace;
2432
2433 TAILQ_INIT(&p2->p_sigpend.sp_info);
2434
2435 LIST_INIT(&p2->p_lwps);
2436 LIST_INIT(&p2->p_sigwaiters);
2437
2438 /*
2439 * Duplicate sub-structures as needed.
2440 * Increase reference counts on shared objects.
2441 * Inherit flags we want to keep. The flags related to SIGCHLD
2442 * handling are important in order to keep a consistent behaviour
2443 * for the child after the fork. If we are a 32-bit process, the
2444 * child will be too.
2445 */
2446 p2->p_flag =
2447 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2448 p2->p_emul = p1->p_emul;
2449 p2->p_execsw = p1->p_execsw;
2450
2451 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2452 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2453 rw_init(&p2->p_reflock);
2454 cv_init(&p2->p_waitcv, "wait");
2455 cv_init(&p2->p_lwpcv, "lwpwait");
2456
2457 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2458
2459 kauth_proc_fork(p1, p2);
2460
2461 p2->p_raslist = NULL;
2462 p2->p_fd = fd_copy();
2463
2464 /* XXX racy */
2465 p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2466
2467 p2->p_cwdi = cwdinit();
2468
2469 /*
2470 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2471 * we just need increase pl_refcnt.
2472 */
2473 if (!p1->p_limit->pl_writeable) {
2474 lim_addref(p1->p_limit);
2475 p2->p_limit = p1->p_limit;
2476 } else {
2477 p2->p_limit = lim_copy(p1->p_limit);
2478 }
2479
2480 p2->p_lflag = 0;
2481 p2->p_sflag = 0;
2482 p2->p_slflag = 0;
2483 p2->p_pptr = p1;
2484 p2->p_ppid = p1->p_pid;
2485 LIST_INIT(&p2->p_children);
2486
2487 p2->p_aio = NULL;
2488
2489 #ifdef KTRACE
2490 /*
2491 * Copy traceflag and tracefile if enabled.
2492 * If not inherited, these were zeroed above.
2493 */
2494 if (p1->p_traceflag & KTRFAC_INHERIT) {
2495 mutex_enter(&ktrace_lock);
2496 p2->p_traceflag = p1->p_traceflag;
2497 if ((p2->p_tracep = p1->p_tracep) != NULL)
2498 ktradref(p2);
2499 mutex_exit(&ktrace_lock);
2500 }
2501 #endif
2502
2503 /*
2504 * Create signal actions for the child process.
2505 */
2506 p2->p_sigacts = sigactsinit(p1, 0);
2507 mutex_enter(p1->p_lock);
2508 p2->p_sflag |=
2509 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2510 sched_proc_fork(p1, p2);
2511 mutex_exit(p1->p_lock);
2512
2513 p2->p_stflag = p1->p_stflag;
2514
2515 /*
2516 * p_stats.
2517 * Copy parts of p_stats, and zero out the rest.
2518 */
2519 p2->p_stats = pstatscopy(p1->p_stats);
2520
2521 /* copy over machdep flags to the new proc */
2522 cpu_proc_fork(p1, p2);
2523
2524 /*
2525 * Prepare remaining parts of spawn data
2526 */
2527 spawn_data->sed_actions = fa;
2528 spawn_data->sed_attrs = sa;
2529
2530 spawn_data->sed_parent = p1;
2531
2532 /* create LWP */
2533 lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2534 &l2, l1->l_class);
2535 l2->l_ctxlink = NULL; /* reset ucontext link */
2536
2537 /*
2538 * Copy the credential so other references don't see our changes.
2539 * Test to see if this is necessary first, since in the common case
2540 * we won't need a private reference.
2541 */
2542 if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2543 kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2544 l2->l_cred = kauth_cred_copy(l2->l_cred);
2545 kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2546 kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2547 }
2548
2549 /* Update the master credentials. */
2550 if (l2->l_cred != p2->p_cred) {
2551 kauth_cred_t ocred;
2552
2553 kauth_cred_hold(l2->l_cred);
2554 mutex_enter(p2->p_lock);
2555 ocred = p2->p_cred;
2556 p2->p_cred = l2->l_cred;
2557 mutex_exit(p2->p_lock);
2558 kauth_cred_free(ocred);
2559 }
2560
2561 *child_ok = true;
2562 spawn_data->sed_refcnt = 2; /* child gets it as well */
2563 #if 0
2564 l2->l_nopreempt = 1; /* start it non-preemptable */
2565 #endif
2566
2567 /*
2568 * It's now safe for the scheduler and other processes to see the
2569 * child process.
2570 */
2571 mutex_enter(proc_lock);
2572
2573 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2574 p2->p_lflag |= PL_CONTROLT;
2575
2576 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2577 p2->p_exitsig = SIGCHLD; /* signal for parent on exit */
2578
2579 LIST_INSERT_AFTER(p1, p2, p_pglist);
2580 LIST_INSERT_HEAD(&allproc, p2, p_list);
2581
2582 p2->p_trace_enabled = trace_is_enabled(p2);
2583 #ifdef __HAVE_SYSCALL_INTERN
2584 (*p2->p_emul->e_syscall_intern)(p2);
2585 #endif
2586
2587 /*
2588 * Make child runnable, set start time, and add to run queue except
2589 * if the parent requested the child to start in SSTOP state.
2590 */
2591 mutex_enter(p2->p_lock);
2592
2593 getmicrotime(&p2->p_stats->p_start);
2594
2595 lwp_lock(l2);
2596 KASSERT(p2->p_nrlwps == 1);
2597 p2->p_nrlwps = 1;
2598 p2->p_stat = SACTIVE;
2599 l2->l_stat = LSRUN;
2600 sched_enqueue(l2, false);
2601 lwp_unlock(l2);
2602
2603 mutex_exit(p2->p_lock);
2604 mutex_exit(proc_lock);
2605
2606 cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2607 error = spawn_data->sed_error;
2608 mutex_exit(&spawn_data->sed_mtx_child);
2609 spawn_exec_data_release(spawn_data);
2610
2611 rw_exit(&p1->p_reflock);
2612 rw_exit(&exec_lock);
2613 have_exec_lock = false;
2614
2615 *pid_res = pid;
2616 return error;
2617
2618 error_exit:
2619 if (have_exec_lock) {
2620 execve_free_data(&spawn_data->sed_exec);
2621 rw_exit(&p1->p_reflock);
2622 rw_exit(&exec_lock);
2623 }
2624 mutex_exit(&spawn_data->sed_mtx_child);
2625 spawn_exec_data_release(spawn_data);
2626
2627 return error;
2628 }
2629
2630 int
sys_posix_spawn(struct lwp * l1,const struct sys_posix_spawn_args * uap,register_t * retval)2631 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2632 register_t *retval)
2633 {
2634 /* {
2635 syscallarg(pid_t *) pid;
2636 syscallarg(const char *) path;
2637 syscallarg(const struct posix_spawn_file_actions *) file_actions;
2638 syscallarg(const struct posix_spawnattr *) attrp;
2639 syscallarg(char *const *) argv;
2640 syscallarg(char *const *) envp;
2641 } */
2642
2643 int error;
2644 struct posix_spawn_file_actions *fa = NULL;
2645 struct posix_spawnattr *sa = NULL;
2646 pid_t pid;
2647 bool child_ok = false;
2648 rlim_t max_fileactions;
2649 proc_t *p = l1->l_proc;
2650
2651 error = check_posix_spawn(l1);
2652 if (error) {
2653 *retval = error;
2654 return 0;
2655 }
2656
2657 /* copy in file_actions struct */
2658 if (SCARG(uap, file_actions) != NULL) {
2659 max_fileactions = 2 * min(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2660 maxfiles);
2661 error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2662 max_fileactions);
2663 if (error)
2664 goto error_exit;
2665 }
2666
2667 /* copyin posix_spawnattr struct */
2668 if (SCARG(uap, attrp) != NULL) {
2669 sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2670 error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2671 if (error)
2672 goto error_exit;
2673 }
2674
2675 /*
2676 * Do the spawn
2677 */
2678 error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2679 SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2680 if (error)
2681 goto error_exit;
2682
2683 if (error == 0 && SCARG(uap, pid) != NULL)
2684 error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2685
2686 *retval = error;
2687 return 0;
2688
2689 error_exit:
2690 if (!child_ok) {
2691 (void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2692 atomic_dec_uint(&nprocs);
2693
2694 if (sa)
2695 kmem_free(sa, sizeof(*sa));
2696 if (fa)
2697 posix_spawn_fa_free(fa, fa->len);
2698 }
2699
2700 *retval = error;
2701 return 0;
2702 }
2703
2704 void
exec_free_emul_arg(struct exec_package * epp)2705 exec_free_emul_arg(struct exec_package *epp)
2706 {
2707 if (epp->ep_emul_arg_free != NULL) {
2708 KASSERT(epp->ep_emul_arg != NULL);
2709 (*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2710 epp->ep_emul_arg_free = NULL;
2711 epp->ep_emul_arg = NULL;
2712 } else {
2713 KASSERT(epp->ep_emul_arg == NULL);
2714 }
2715 }
2716
2717 #ifdef DEBUG_EXEC
2718 static void
dump_vmcmds(const struct exec_package * const epp,size_t x,int error)2719 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2720 {
2721 struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2722 size_t j;
2723
2724 if (error == 0)
2725 DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2726 else
2727 DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2728 epp->ep_vmcmds.evs_used, error));
2729
2730 for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2731 DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2732 PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2733 PRIxVSIZE" prot=0%o flags=%d\n", j,
2734 vp[j].ev_proc == vmcmd_map_pagedvn ?
2735 "pagedvn" :
2736 vp[j].ev_proc == vmcmd_map_readvn ?
2737 "readvn" :
2738 vp[j].ev_proc == vmcmd_map_zero ?
2739 "zero" : "*unknown*",
2740 vp[j].ev_addr, vp[j].ev_len,
2741 vp[j].ev_offset, vp[j].ev_prot,
2742 vp[j].ev_flags));
2743 if (error != 0 && j == x)
2744 DPRINTF((" ^--- failed\n"));
2745 }
2746 }
2747 #endif
2748