1 /* $NetBSD: kern_exec.c,v 1.528 2025/04/27 17:40:55 riastradh Exp $ */
2
3 /*-
4 * Copyright (c) 2008, 2019, 2020 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Andrew Doran.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 /*-
33 * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
34 * Copyright (C) 1992 Wolfgang Solfrank.
35 * Copyright (C) 1992 TooLs GmbH.
36 * All rights reserved.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by TooLs GmbH.
49 * 4. The name of TooLs GmbH may not be used to endorse or promote products
50 * derived from this software without specific prior written permission.
51 *
52 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
53 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
54 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
55 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
56 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
57 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
58 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
59 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
60 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
61 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
62 */
63
64 #include <sys/cdefs.h>
65 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.528 2025/04/27 17:40:55 riastradh Exp $");
66
67 #include "opt_exec.h"
68 #include "opt_execfmt.h"
69 #include "opt_ktrace.h"
70 #include "opt_modular.h"
71 #include "opt_pax.h"
72 #include "opt_syscall_debug.h"
73 #include "veriexec.h"
74
75 #include <sys/param.h>
76 #include <sys/types.h>
77
78 #include <sys/acct.h>
79 #include <sys/atomic.h>
80 #include <sys/cprng.h>
81 #include <sys/cpu.h>
82 #include <sys/exec.h>
83 #include <sys/file.h>
84 #include <sys/filedesc.h>
85 #include <sys/futex.h>
86 #include <sys/kauth.h>
87 #include <sys/kernel.h>
88 #include <sys/kmem.h>
89 #include <sys/ktrace.h>
90 #include <sys/lwpctl.h>
91 #include <sys/mman.h>
92 #include <sys/module.h>
93 #include <sys/mount.h>
94 #include <sys/namei.h>
95 #include <sys/pax.h>
96 #include <sys/proc.h>
97 #include <sys/prot.h>
98 #include <sys/ptrace.h>
99 #include <sys/ras.h>
100 #include <sys/sdt.h>
101 #include <sys/signalvar.h>
102 #include <sys/spawn.h>
103 #include <sys/stat.h>
104 #include <sys/syscall.h>
105 #include <sys/syscallargs.h>
106 #include <sys/syscallvar.h>
107 #include <sys/systm.h>
108 #include <sys/uidinfo.h>
109 #if NVERIEXEC > 0
110 #include <sys/verified_exec.h>
111 #endif /* NVERIEXEC > 0 */
112 #include <sys/vfs_syscalls.h>
113 #include <sys/vnode.h>
114 #include <sys/wait.h>
115
116 #include <uvm/uvm_extern.h>
117
118 #include <machine/reg.h>
119
120 #include <compat/common/compat_util.h>
121
122 #ifndef MD_TOPDOWN_INIT
123 #ifdef __USE_TOPDOWN_VM
124 #define MD_TOPDOWN_INIT(epp) (epp)->ep_flags |= EXEC_TOPDOWN_VM
125 #else
126 #define MD_TOPDOWN_INIT(epp)
127 #endif
128 #endif
129
130 struct execve_data;
131
132 extern int user_va0_disable;
133
134 static size_t calcargs(struct execve_data * restrict, const size_t);
135 static size_t calcstack(struct execve_data * restrict, const size_t);
136 static int copyoutargs(struct execve_data * restrict, struct lwp *,
137 char * const);
138 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
139 static int copyinargs(struct execve_data * restrict, char * const *,
140 char * const *, execve_fetch_element_t, char **);
141 static int copyinargstrs(struct execve_data * restrict, char * const *,
142 execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
143 static int exec_sigcode_map(struct proc *, const struct emul *);
144
145 #if defined(DEBUG) && !defined(DEBUG_EXEC)
146 #define DEBUG_EXEC
147 #endif
148 #ifdef DEBUG_EXEC
149 #define DPRINTF(a) printf a
150 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
151 __LINE__, (s), (a), (b))
152 static void dump_vmcmds(const struct exec_package * const, size_t, int);
153 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
154 #else
155 #define DPRINTF(a)
156 #define COPYPRINTF(s, a, b)
157 #define DUMPVMCMDS(p, x, e) do {} while (0)
158 #endif /* DEBUG_EXEC */
159
160 /*
161 * DTrace SDT provider definitions
162 */
163 SDT_PROVIDER_DECLARE(proc);
164 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
165 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
166 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
167
168 /*
169 * Exec function switch:
170 *
171 * Note that each makecmds function is responsible for loading the
172 * exec package with the necessary functions for any exec-type-specific
173 * handling.
174 *
175 * Functions for specific exec types should be defined in their own
176 * header file.
177 */
178 static const struct execsw **execsw = NULL;
179 static int nexecs;
180
181 u_int exec_maxhdrsz; /* must not be static - used by netbsd32 */
182
183 /* list of dynamically loaded execsw entries */
184 static LIST_HEAD(execlist_head, exec_entry) ex_head =
185 LIST_HEAD_INITIALIZER(ex_head);
186 struct exec_entry {
187 LIST_ENTRY(exec_entry) ex_list;
188 SLIST_ENTRY(exec_entry) ex_slist;
189 const struct execsw *ex_sw;
190 };
191
192 #ifndef __HAVE_SYSCALL_INTERN
193 void syscall(void);
194 #endif
195
196 /* NetBSD autoloadable syscalls */
197 #ifdef MODULAR
198 #include <kern/syscalls_autoload.c>
199 #endif
200
201 /* NetBSD emul struct */
202 struct emul emul_netbsd = {
203 .e_name = "netbsd",
204 #ifdef EMUL_NATIVEROOT
205 .e_path = EMUL_NATIVEROOT,
206 #else
207 .e_path = NULL,
208 #endif
209 #ifndef __HAVE_MINIMAL_EMUL
210 .e_flags = EMUL_HAS_SYS___syscall,
211 .e_errno = NULL,
212 .e_nosys = SYS_syscall,
213 .e_nsysent = SYS_NSYSENT,
214 #endif
215 #ifdef MODULAR
216 .e_sc_autoload = netbsd_syscalls_autoload,
217 #endif
218 .e_sysent = sysent,
219 .e_nomodbits = sysent_nomodbits,
220 #ifdef SYSCALL_DEBUG
221 .e_syscallnames = syscallnames,
222 #else
223 .e_syscallnames = NULL,
224 #endif
225 .e_sendsig = sendsig,
226 .e_trapsignal = trapsignal,
227 .e_sigcode = NULL,
228 .e_esigcode = NULL,
229 .e_sigobject = NULL,
230 .e_setregs = setregs,
231 .e_proc_exec = NULL,
232 .e_proc_fork = NULL,
233 .e_proc_exit = NULL,
234 .e_lwp_fork = NULL,
235 .e_lwp_exit = NULL,
236 #ifdef __HAVE_SYSCALL_INTERN
237 .e_syscall_intern = syscall_intern,
238 #else
239 .e_syscall = syscall,
240 #endif
241 .e_sysctlovly = NULL,
242 .e_vm_default_addr = uvm_default_mapaddr,
243 .e_usertrap = NULL,
244 .e_ucsize = sizeof(ucontext_t),
245 .e_startlwp = startlwp
246 };
247
248 /*
249 * Exec lock. Used to control access to execsw[] structures.
250 * This must not be static so that netbsd32 can access it, too.
251 */
252 krwlock_t exec_lock __cacheline_aligned;
253
254 /*
255 * Data used between a loadvm and execve part of an "exec" operation
256 */
257 struct execve_data {
258 struct exec_package ed_pack;
259 struct pathbuf *ed_pathbuf;
260 struct vattr ed_attr;
261 struct ps_strings ed_arginfo;
262 char *ed_argp;
263 const char *ed_pathstring;
264 char *ed_resolvedname;
265 size_t ed_ps_strings_sz;
266 int ed_szsigcode;
267 size_t ed_argslen;
268 long ed_argc;
269 long ed_envc;
270 };
271
272 /*
273 * data passed from parent lwp to child during a posix_spawn()
274 */
275 struct spawn_exec_data {
276 struct execve_data sed_exec;
277 struct posix_spawn_file_actions
278 *sed_actions;
279 struct posix_spawnattr *sed_attrs;
280 struct proc *sed_parent;
281 kcondvar_t sed_cv_child_ready;
282 kmutex_t sed_mtx_child;
283 int sed_error;
284 bool sed_child_ready;
285 volatile uint32_t sed_refcnt;
286 };
287
288 static struct vm_map *exec_map;
289 static struct pool exec_pool;
290
291 static void *
exec_pool_alloc(struct pool * pp,int flags)292 exec_pool_alloc(struct pool *pp, int flags)
293 {
294
295 return (void *)uvm_km_alloc(exec_map, NCARGS, 0,
296 UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
297 }
298
299 static void
exec_pool_free(struct pool * pp,void * addr)300 exec_pool_free(struct pool *pp, void *addr)
301 {
302
303 uvm_km_free(exec_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
304 }
305
306 static struct pool_allocator exec_palloc = {
307 .pa_alloc = exec_pool_alloc,
308 .pa_free = exec_pool_free,
309 .pa_pagesz = NCARGS
310 };
311
312 static void
exec_path_free(struct execve_data * data)313 exec_path_free(struct execve_data *data)
314 {
315 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
316 pathbuf_destroy(data->ed_pathbuf);
317 if (data->ed_resolvedname)
318 PNBUF_PUT(data->ed_resolvedname);
319 }
320
321 static int
exec_resolvename(struct lwp * l,struct exec_package * epp,struct vnode * vp,char ** rpath)322 exec_resolvename(struct lwp *l, struct exec_package *epp, struct vnode *vp,
323 char **rpath)
324 {
325 int error;
326 char *p;
327
328 KASSERT(rpath != NULL);
329
330 *rpath = PNBUF_GET();
331 error = vnode_to_path(*rpath, MAXPATHLEN, vp, l, l->l_proc);
332 if (error) {
333 DPRINTF(("%s: can't resolve name for %s, error %d\n",
334 __func__, epp->ep_kname, error));
335 PNBUF_PUT(*rpath);
336 *rpath = NULL;
337 return error;
338 }
339 epp->ep_resolvedname = *rpath;
340 if ((p = strrchr(*rpath, '/')) != NULL)
341 epp->ep_kname = p + 1;
342 return 0;
343 }
344
345
346 /*
347 * check exec:
348 * given an "executable" described in the exec package's namei info,
349 * see what we can do with it.
350 *
351 * ON ENTRY:
352 * exec package with appropriate namei info
353 * lwp pointer of exec'ing lwp
354 * NO SELF-LOCKED VNODES
355 *
356 * ON EXIT:
357 * error: nothing held, etc. exec header still allocated.
358 * ok: filled exec package, executable's vnode (unlocked).
359 *
360 * EXEC SWITCH ENTRY:
361 * Locked vnode to check, exec package, proc.
362 *
363 * EXEC SWITCH EXIT:
364 * ok: return 0, filled exec package, executable's vnode (unlocked).
365 * error: destructive:
366 * everything deallocated execept exec header.
367 * non-destructive:
368 * error code, executable's vnode (unlocked),
369 * exec header unmodified.
370 */
371 int
372 /*ARGSUSED*/
check_exec(struct lwp * l,struct exec_package * epp,struct pathbuf * pb,char ** rpath)373 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb,
374 char **rpath)
375 {
376 int error, i;
377 struct vnode *vp;
378 size_t resid;
379
380 if (epp->ep_resolvedname) {
381 struct nameidata nd;
382
383 // grab the absolute pathbuf here before namei() trashes it.
384 pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX);
385 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
386
387 /* first get the vnode */
388 if ((error = namei(&nd)) != 0)
389 return error;
390
391 epp->ep_vp = vp = nd.ni_vp;
392 #ifdef DIAGNOSTIC
393 /* paranoia (take this out once namei stuff stabilizes) */
394 memset(nd.ni_pnbuf, '~', PATH_MAX);
395 #endif
396 } else {
397 struct file *fp;
398
399 if ((error = fd_getvnode(epp->ep_xfd, &fp)) != 0)
400 return error;
401 epp->ep_vp = vp = fp->f_vnode;
402 vref(vp);
403 fd_putfile(epp->ep_xfd);
404 if ((error = exec_resolvename(l, epp, vp, rpath)) != 0)
405 return error;
406 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
407 }
408
409 /* check access and type */
410 if (vp->v_type != VREG) {
411 error = SET_ERROR(EACCES);
412 goto bad1;
413 }
414 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
415 goto bad1;
416
417 /* get attributes */
418 /* XXX VOP_GETATTR is the only thing that needs LK_EXCLUSIVE here */
419 if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
420 goto bad1;
421
422 /* Check mount point */
423 if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
424 error = SET_ERROR(EACCES);
425 goto bad1;
426 }
427 if (vp->v_mount->mnt_flag & MNT_NOSUID)
428 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
429
430 /* try to open it */
431 if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
432 goto bad1;
433
434 /* now we have the file, get the exec header */
435 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
436 UIO_SYSSPACE, IO_NODELOCKED, l->l_cred, &resid, NULL);
437 if (error)
438 goto bad1;
439
440 /* unlock vp, since we need it unlocked from here on out. */
441 VOP_UNLOCK(vp);
442
443 #if NVERIEXEC > 0
444 error = veriexec_verify(l, vp,
445 epp->ep_resolvedname ? epp->ep_resolvedname : epp->ep_kname,
446 epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
447 NULL);
448 if (error)
449 goto bad2;
450 #endif /* NVERIEXEC > 0 */
451
452 #ifdef PAX_SEGVGUARD
453 error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
454 if (error)
455 goto bad2;
456 #endif /* PAX_SEGVGUARD */
457
458 epp->ep_hdrvalid = epp->ep_hdrlen - resid;
459
460 /*
461 * Set up default address space limits. Can be overridden
462 * by individual exec packages.
463 */
464 epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS);
465 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
466
467 /*
468 * set up the vmcmds for creation of the process
469 * address space
470 */
471 error = nexecs == 0 ? SET_ERROR(ENOEXEC) : ENOEXEC;
472 for (i = 0; i < nexecs; i++) {
473 int newerror;
474
475 epp->ep_esch = execsw[i];
476 newerror = (*execsw[i]->es_makecmds)(l, epp);
477
478 if (!newerror) {
479 /* Seems ok: check that entry point is not too high */
480 if (epp->ep_entry >= epp->ep_vm_maxaddr) {
481 #ifdef DIAGNOSTIC
482 printf("%s: rejecting %p due to "
483 "too high entry address (>= %p)\n",
484 __func__, (void *)epp->ep_entry,
485 (void *)epp->ep_vm_maxaddr);
486 #endif
487 error = SET_ERROR(ENOEXEC);
488 break;
489 }
490 /* Seems ok: check that entry point is not too low */
491 if (epp->ep_entry < epp->ep_vm_minaddr) {
492 #ifdef DIAGNOSTIC
493 printf("%s: rejecting %p due to "
494 "too low entry address (< %p)\n",
495 __func__, (void *)epp->ep_entry,
496 (void *)epp->ep_vm_minaddr);
497 #endif
498 error = SET_ERROR(ENOEXEC);
499 break;
500 }
501
502 /* check limits */
503 #ifdef DIAGNOSTIC
504 #define LMSG "%s: rejecting due to %s limit (%ju > %ju)\n"
505 #endif
506 #ifdef MAXTSIZ
507 if (epp->ep_tsize > MAXTSIZ) {
508 #ifdef DIAGNOSTIC
509 printf(LMSG, __func__, "text",
510 (uintmax_t)epp->ep_tsize,
511 (uintmax_t)MAXTSIZ);
512 #endif
513 error = SET_ERROR(ENOMEM);
514 break;
515 }
516 #endif
517 vsize_t dlimit =
518 (vsize_t)l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur;
519 if (epp->ep_dsize > dlimit) {
520 #ifdef DIAGNOSTIC
521 printf(LMSG, __func__, "data",
522 (uintmax_t)epp->ep_dsize,
523 (uintmax_t)dlimit);
524 #endif
525 error = SET_ERROR(ENOMEM);
526 break;
527 }
528 return 0;
529 }
530
531 /*
532 * Reset all the fields that may have been modified by the
533 * loader.
534 */
535 KASSERT(epp->ep_emul_arg == NULL);
536 if (epp->ep_emul_root != NULL) {
537 vrele(epp->ep_emul_root);
538 epp->ep_emul_root = NULL;
539 }
540 if (epp->ep_interp != NULL) {
541 vrele(epp->ep_interp);
542 epp->ep_interp = NULL;
543 }
544 epp->ep_pax_flags = 0;
545
546 /* make sure the first "interesting" error code is saved. */
547 if (error == ENOEXEC)
548 error = newerror;
549
550 if (epp->ep_flags & EXEC_DESTR)
551 /* Error from "#!" code, tidied up by recursive call */
552 return error;
553 }
554
555 /* not found, error */
556
557 /*
558 * free any vmspace-creation commands,
559 * and release their references
560 */
561 kill_vmcmds(&epp->ep_vmcmds);
562
563 #if NVERIEXEC > 0 || defined(PAX_SEGVGUARD)
564 bad2:
565 #endif
566 /*
567 * close and release the vnode, restore the old one, free the
568 * pathname buf, and punt.
569 */
570 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
571 VOP_CLOSE(vp, FREAD, l->l_cred);
572 vput(vp);
573 return error;
574
575 bad1:
576 /*
577 * free the namei pathname buffer, and put the vnode
578 * (which we don't yet have open).
579 */
580 vput(vp); /* was still locked */
581 return error;
582 }
583
584 #ifdef __MACHINE_STACK_GROWS_UP
585 #define STACK_PTHREADSPACE NBPG
586 #else
587 #define STACK_PTHREADSPACE 0
588 #endif
589
590 static int
execve_fetch_element(char * const * array,size_t index,char ** value)591 execve_fetch_element(char * const *array, size_t index, char **value)
592 {
593 return copyin(array + index, value, sizeof(*value));
594 }
595
596 /*
597 * exec system call
598 */
599 int
sys_execve(struct lwp * l,const struct sys_execve_args * uap,register_t * retval)600 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
601 {
602 /* {
603 syscallarg(const char *) path;
604 syscallarg(char * const *) argp;
605 syscallarg(char * const *) envp;
606 } */
607
608 return execve1(l, true, SCARG(uap, path), -1, SCARG(uap, argp),
609 SCARG(uap, envp), execve_fetch_element);
610 }
611
612 int
sys_fexecve(struct lwp * l,const struct sys_fexecve_args * uap,register_t * retval)613 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
614 register_t *retval)
615 {
616 /* {
617 syscallarg(int) fd;
618 syscallarg(char * const *) argp;
619 syscallarg(char * const *) envp;
620 } */
621
622 return execve1(l, false, NULL, SCARG(uap, fd), SCARG(uap, argp),
623 SCARG(uap, envp), execve_fetch_element);
624 }
625
626 /*
627 * Load modules to try and execute an image that we do not understand.
628 * If no execsw entries are present, we load those likely to be needed
629 * in order to run native images only. Otherwise, we autoload all
630 * possible modules that could let us run the binary. XXX lame
631 */
632 static void
exec_autoload(void)633 exec_autoload(void)
634 {
635 #ifdef MODULAR
636 static const char * const native[] = {
637 "exec_elf32",
638 "exec_elf64",
639 "exec_script",
640 NULL
641 };
642 static const char * const compat[] = {
643 "exec_elf32",
644 "exec_elf64",
645 "exec_script",
646 "exec_aout",
647 "exec_coff",
648 "exec_ecoff",
649 "compat_aoutm68k",
650 "compat_netbsd32",
651 #if 0
652 "compat_linux",
653 "compat_linux32",
654 #endif
655 "compat_sunos",
656 "compat_sunos32",
657 "compat_ultrix",
658 NULL
659 };
660 char const * const *list;
661 int i;
662
663 list = nexecs == 0 ? native : compat;
664 for (i = 0; list[i] != NULL; i++) {
665 if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
666 continue;
667 }
668 yield();
669 }
670 #endif
671 }
672
673 /*
674 * Copy the user or kernel supplied upath to the allocated pathbuffer pbp
675 * making it absolute in the process, by prepending the current working
676 * directory if it is not. If offs is supplied it will contain the offset
677 * where the original supplied copy of upath starts.
678 */
679 int
exec_makepathbuf(struct lwp * l,const char * upath,enum uio_seg seg,struct pathbuf ** pbp,size_t * offs)680 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg,
681 struct pathbuf **pbp, size_t *offs)
682 {
683 char *path, *bp;
684 size_t len, tlen;
685 int error;
686 struct cwdinfo *cwdi;
687
688 path = PNBUF_GET();
689 if (seg == UIO_SYSSPACE) {
690 error = copystr(upath, path, MAXPATHLEN, &len);
691 } else {
692 error = copyinstr(upath, path, MAXPATHLEN, &len);
693 }
694 if (error)
695 goto err;
696
697 if (path[0] == '/') {
698 if (offs)
699 *offs = 0;
700 goto out;
701 }
702
703 len++;
704 if (len + 1 >= MAXPATHLEN) {
705 error = SET_ERROR(ENAMETOOLONG);
706 goto err;
707 }
708 bp = path + MAXPATHLEN - len;
709 memmove(bp, path, len);
710 *(--bp) = '/';
711
712 cwdi = l->l_proc->p_cwdi;
713 rw_enter(&cwdi->cwdi_lock, RW_READER);
714 error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
715 GETCWD_CHECK_ACCESS, l);
716 rw_exit(&cwdi->cwdi_lock);
717
718 if (error)
719 goto err;
720 tlen = path + MAXPATHLEN - bp;
721
722 memmove(path, bp, tlen);
723 path[tlen - 1] = '\0';
724 if (offs)
725 *offs = tlen - len;
726 out:
727 *pbp = pathbuf_assimilate(path);
728 return 0;
729 err:
730 PNBUF_PUT(path);
731 return error;
732 }
733
734 vaddr_t
exec_vm_minaddr(vaddr_t va_min)735 exec_vm_minaddr(vaddr_t va_min)
736 {
737 /*
738 * Increase va_min if we don't want NULL to be mappable by the
739 * process.
740 */
741 #define VM_MIN_GUARD PAGE_SIZE
742 if (user_va0_disable && (va_min < VM_MIN_GUARD))
743 return VM_MIN_GUARD;
744 return va_min;
745 }
746
747 static int
execve_loadvm(struct lwp * l,bool has_path,const char * path,int fd,char * const * args,char * const * envs,execve_fetch_element_t fetch_element,struct execve_data * restrict data)748 execve_loadvm(struct lwp *l, bool has_path, const char *path, int fd,
749 char * const *args, char * const *envs,
750 execve_fetch_element_t fetch_element,
751 struct execve_data * restrict data)
752 {
753 struct exec_package * const epp = &data->ed_pack;
754 int error;
755 struct proc *p;
756 char *dp;
757 u_int modgen;
758
759 KASSERT(data != NULL);
760
761 p = l->l_proc;
762 modgen = 0;
763
764 SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
765
766 /*
767 * Check if we have exceeded our number of processes limit.
768 * This is so that we handle the case where a root daemon
769 * forked, ran setuid to become the desired user and is trying
770 * to exec. The obvious place to do the reference counting check
771 * is setuid(), but we don't do the reference counting check there
772 * like other OS's do because then all the programs that use setuid()
773 * must be modified to check the return code of setuid() and exit().
774 * It is dangerous to make setuid() fail, because it fails open and
775 * the program will continue to run as root. If we make it succeed
776 * and return an error code, again we are not enforcing the limit.
777 * The best place to enforce the limit is here, when the process tries
778 * to execute a new image, because eventually the process will need
779 * to call exec in order to do something useful.
780 */
781 retry:
782 if (p->p_flag & PK_SUGID) {
783 if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
784 p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
785 &p->p_rlimit[RLIMIT_NPROC],
786 KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
787 chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
788 p->p_rlimit[RLIMIT_NPROC].rlim_cur)
789 return SET_ERROR(EAGAIN);
790 }
791
792 /*
793 * Drain existing references and forbid new ones. The process
794 * should be left alone until we're done here. This is necessary
795 * to avoid race conditions - e.g. in ptrace() - that might allow
796 * a local user to illicitly obtain elevated privileges.
797 */
798 rw_enter(&p->p_reflock, RW_WRITER);
799
800 if (has_path) {
801 size_t offs;
802 /*
803 * Init the namei data to point the file user's program name.
804 * This is done here rather than in check_exec(), so that it's
805 * possible to override this settings if any of makecmd/probe
806 * functions call check_exec() recursively - for example,
807 * see exec_script_makecmds().
808 */
809 if ((error = exec_makepathbuf(l, path, UIO_USERSPACE,
810 &data->ed_pathbuf, &offs)) != 0)
811 goto clrflg;
812 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
813 epp->ep_kname = data->ed_pathstring + offs;
814 data->ed_resolvedname = PNBUF_GET();
815 epp->ep_resolvedname = data->ed_resolvedname;
816 epp->ep_xfd = -1;
817 } else {
818 data->ed_pathbuf = pathbuf_assimilate(strcpy(PNBUF_GET(), "/"));
819 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
820 epp->ep_kname = "*fexecve*";
821 data->ed_resolvedname = NULL;
822 epp->ep_resolvedname = NULL;
823 epp->ep_xfd = fd;
824 }
825
826
827 /*
828 * initialize the fields of the exec package.
829 */
830 epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
831 epp->ep_hdrlen = exec_maxhdrsz;
832 epp->ep_hdrvalid = 0;
833 epp->ep_emul_arg = NULL;
834 epp->ep_emul_arg_free = NULL;
835 memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
836 epp->ep_vap = &data->ed_attr;
837 epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
838 MD_TOPDOWN_INIT(epp);
839 epp->ep_emul_root = NULL;
840 epp->ep_interp = NULL;
841 epp->ep_esch = NULL;
842 epp->ep_pax_flags = 0;
843 memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
844
845 rw_enter(&exec_lock, RW_READER);
846
847 /* see if we can run it. */
848 if ((error = check_exec(l, epp, data->ed_pathbuf,
849 &data->ed_resolvedname)) != 0) {
850 if (error != ENOENT && error != EACCES && error != ENOEXEC) {
851 DPRINTF(("%s: check exec failed for %s, error %d\n",
852 __func__, epp->ep_kname, error));
853 }
854 goto freehdr;
855 }
856
857 /* allocate an argument buffer */
858 data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
859 KASSERT(data->ed_argp != NULL);
860 dp = data->ed_argp;
861
862 if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
863 goto bad;
864 }
865
866 /*
867 * Calculate the new stack size.
868 */
869
870 #ifdef __MACHINE_STACK_GROWS_UP
871 /*
872 * copyargs() fills argc/argv/envp from the lower address even on
873 * __MACHINE_STACK_GROWS_UP machines. Reserve a few words just below the SP
874 * so that _rtld() use it.
875 */
876 #define RTLD_GAP 32
877 #else
878 #define RTLD_GAP 0
879 #endif
880
881 const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
882
883 data->ed_argslen = calcargs(data, argenvstrlen);
884
885 const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
886
887 if (len > epp->ep_ssize) {
888 /* in effect, compare to initial limit */
889 DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
890 error = SET_ERROR(ENOMEM);
891 goto bad;
892 }
893 /* adjust "active stack depth" for process VSZ */
894 epp->ep_ssize = len;
895
896 return 0;
897
898 bad:
899 /* free the vmspace-creation commands, and release their references */
900 kill_vmcmds(&epp->ep_vmcmds);
901 /* kill any opened file descriptor, if necessary */
902 if (epp->ep_flags & EXEC_HASFD) {
903 epp->ep_flags &= ~EXEC_HASFD;
904 fd_close(epp->ep_fd);
905 }
906 /* close and put the exec'd file */
907 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
908 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
909 vput(epp->ep_vp);
910 pool_put(&exec_pool, data->ed_argp);
911
912 freehdr:
913 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
914 if (epp->ep_emul_root != NULL)
915 vrele(epp->ep_emul_root);
916 if (epp->ep_interp != NULL)
917 vrele(epp->ep_interp);
918
919 rw_exit(&exec_lock);
920
921 exec_path_free(data);
922
923 clrflg:
924 rw_exit(&p->p_reflock);
925
926 if (modgen != module_gen && error == ENOEXEC) {
927 modgen = module_gen;
928 exec_autoload();
929 goto retry;
930 }
931
932 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
933 return error;
934 }
935
936 static int
execve_dovmcmds(struct lwp * l,struct execve_data * restrict data)937 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
938 {
939 struct exec_package * const epp = &data->ed_pack;
940 struct proc *p = l->l_proc;
941 struct exec_vmcmd *base_vcp;
942 int error = 0;
943 size_t i;
944
945 /* record proc's vnode, for use by procfs and others */
946 if (p->p_textvp)
947 vrele(p->p_textvp);
948 vref(epp->ep_vp);
949 p->p_textvp = epp->ep_vp;
950
951 /* create the new process's VM space by running the vmcmds */
952 KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
953
954 #ifdef TRACE_EXEC
955 DUMPVMCMDS(epp, 0, 0);
956 #endif
957
958 base_vcp = NULL;
959
960 for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
961 struct exec_vmcmd *vcp;
962
963 vcp = &epp->ep_vmcmds.evs_cmds[i];
964 if (vcp->ev_flags & VMCMD_RELATIVE) {
965 KASSERTMSG(base_vcp != NULL,
966 "%s: relative vmcmd with no base", __func__);
967 KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
968 "%s: illegal base & relative vmcmd", __func__);
969 vcp->ev_addr += base_vcp->ev_addr;
970 }
971 error = (*vcp->ev_proc)(l, vcp);
972 if (error)
973 DUMPVMCMDS(epp, i, error);
974 if (vcp->ev_flags & VMCMD_BASE)
975 base_vcp = vcp;
976 }
977
978 /* free the vmspace-creation commands, and release their references */
979 kill_vmcmds(&epp->ep_vmcmds);
980
981 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
982 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
983 vput(epp->ep_vp);
984
985 /* if an error happened, deallocate and punt */
986 if (error != 0) {
987 DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
988 }
989 return error;
990 }
991
992 static void
execve_free_data(struct execve_data * data)993 execve_free_data(struct execve_data *data)
994 {
995 struct exec_package * const epp = &data->ed_pack;
996
997 /* free the vmspace-creation commands, and release their references */
998 kill_vmcmds(&epp->ep_vmcmds);
999 /* kill any opened file descriptor, if necessary */
1000 if (epp->ep_flags & EXEC_HASFD) {
1001 epp->ep_flags &= ~EXEC_HASFD;
1002 fd_close(epp->ep_fd);
1003 }
1004
1005 /* close and put the exec'd file */
1006 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
1007 VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
1008 vput(epp->ep_vp);
1009 pool_put(&exec_pool, data->ed_argp);
1010
1011 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1012 if (epp->ep_emul_root != NULL)
1013 vrele(epp->ep_emul_root);
1014 if (epp->ep_interp != NULL)
1015 vrele(epp->ep_interp);
1016
1017 exec_path_free(data);
1018 }
1019
1020 static void
pathexec(struct proc * p,const char * resolvedname)1021 pathexec(struct proc *p, const char *resolvedname)
1022 {
1023 /* set command name & other accounting info */
1024 const char *cmdname;
1025
1026 if (resolvedname == NULL) {
1027 cmdname = "*fexecve*";
1028 resolvedname = "/";
1029 } else {
1030 cmdname = strrchr(resolvedname, '/') + 1;
1031 }
1032 KASSERTMSG(resolvedname[0] == '/', "bad resolvedname `%s'",
1033 resolvedname);
1034
1035 strlcpy(p->p_comm, cmdname, sizeof(p->p_comm));
1036
1037 kmem_strfree(p->p_path);
1038 p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP);
1039 }
1040
1041 /* XXX elsewhere */
1042 static int
credexec(struct lwp * l,struct execve_data * data)1043 credexec(struct lwp *l, struct execve_data *data)
1044 {
1045 struct proc *p = l->l_proc;
1046 struct vattr *attr = &data->ed_attr;
1047 int error;
1048
1049 /*
1050 * Deal with set[ug]id. MNT_NOSUID has already been used to disable
1051 * s[ug]id. It's OK to check for PSL_TRACED here as we have blocked
1052 * out additional references on the process for the moment.
1053 */
1054 if ((p->p_slflag & PSL_TRACED) == 0 &&
1055
1056 (((attr->va_mode & S_ISUID) != 0 &&
1057 kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
1058
1059 ((attr->va_mode & S_ISGID) != 0 &&
1060 kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
1061 /*
1062 * Mark the process as SUGID before we do
1063 * anything that might block.
1064 */
1065 proc_crmod_enter();
1066 proc_crmod_leave(NULL, NULL, true);
1067 if (data->ed_argc == 0) {
1068 DPRINTF((
1069 "%s: not executing set[ug]id binary with no args\n",
1070 __func__));
1071 return SET_ERROR(EINVAL);
1072 }
1073
1074 /* Make sure file descriptors 0..2 are in use. */
1075 if ((error = fd_checkstd()) != 0) {
1076 DPRINTF(("%s: fdcheckstd failed %d\n",
1077 __func__, error));
1078 return error;
1079 }
1080
1081 /*
1082 * Copy the credential so other references don't see our
1083 * changes.
1084 */
1085 l->l_cred = kauth_cred_copy(l->l_cred);
1086 #ifdef KTRACE
1087 /*
1088 * If the persistent trace flag isn't set, turn off.
1089 */
1090 if (p->p_tracep) {
1091 mutex_enter(&ktrace_lock);
1092 if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1093 ktrderef(p);
1094 mutex_exit(&ktrace_lock);
1095 }
1096 #endif
1097 if (attr->va_mode & S_ISUID)
1098 kauth_cred_seteuid(l->l_cred, attr->va_uid);
1099 if (attr->va_mode & S_ISGID)
1100 kauth_cred_setegid(l->l_cred, attr->va_gid);
1101 } else {
1102 if (kauth_cred_geteuid(l->l_cred) ==
1103 kauth_cred_getuid(l->l_cred) &&
1104 kauth_cred_getegid(l->l_cred) ==
1105 kauth_cred_getgid(l->l_cred))
1106 p->p_flag &= ~PK_SUGID;
1107 }
1108
1109 /*
1110 * Copy the credential so other references don't see our changes.
1111 * Test to see if this is necessary first, since in the common case
1112 * we won't need a private reference.
1113 */
1114 if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1115 kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1116 l->l_cred = kauth_cred_copy(l->l_cred);
1117 kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1118 kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1119 }
1120
1121 /* Update the master credentials. */
1122 if (l->l_cred != p->p_cred) {
1123 kauth_cred_t ocred;
1124 mutex_enter(p->p_lock);
1125 ocred = p->p_cred;
1126 p->p_cred = kauth_cred_hold(l->l_cred);
1127 mutex_exit(p->p_lock);
1128 kauth_cred_free(ocred);
1129 }
1130
1131 return 0;
1132 }
1133
1134 static void
emulexec(struct lwp * l,struct exec_package * epp)1135 emulexec(struct lwp *l, struct exec_package *epp)
1136 {
1137 struct proc *p = l->l_proc;
1138
1139 /* The emulation root will usually have been found when we looked
1140 * for the elf interpreter (or similar), if not look now. */
1141 if (epp->ep_esch->es_emul->e_path != NULL &&
1142 epp->ep_emul_root == NULL)
1143 emul_find_root(l, epp);
1144
1145 /* Any old emulation root got removed by fdcloseexec */
1146 rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1147 p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1148 rw_exit(&p->p_cwdi->cwdi_lock);
1149 epp->ep_emul_root = NULL;
1150 if (epp->ep_interp != NULL)
1151 vrele(epp->ep_interp);
1152
1153 /*
1154 * Call emulation specific exec hook. This can setup per-process
1155 * p->p_emuldata or do any other per-process stuff an emulation needs.
1156 *
1157 * If we are executing process of different emulation than the
1158 * original forked process, call e_proc_exit() of the old emulation
1159 * first, then e_proc_exec() of new emulation. If the emulation is
1160 * same, the exec hook code should deallocate any old emulation
1161 * resources held previously by this process.
1162 */
1163 if (p->p_emul && p->p_emul->e_proc_exit
1164 && p->p_emul != epp->ep_esch->es_emul)
1165 (*p->p_emul->e_proc_exit)(p);
1166
1167 /*
1168 * Call exec hook. Emulation code may NOT store reference to anything
1169 * from &pack.
1170 */
1171 if (epp->ep_esch->es_emul->e_proc_exec)
1172 (*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1173
1174 /* update p_emul, the old value is no longer needed */
1175 p->p_emul = epp->ep_esch->es_emul;
1176
1177 /* ...and the same for p_execsw */
1178 p->p_execsw = epp->ep_esch;
1179
1180 #ifdef __HAVE_SYSCALL_INTERN
1181 (*p->p_emul->e_syscall_intern)(p);
1182 #endif
1183 ktremul();
1184 }
1185
1186 static int
execve_runproc(struct lwp * l,struct execve_data * restrict data,bool no_local_exec_lock,bool is_spawn)1187 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1188 bool no_local_exec_lock, bool is_spawn)
1189 {
1190 struct exec_package * const epp = &data->ed_pack;
1191 int error = 0;
1192 struct proc *p;
1193 struct vmspace *vm;
1194
1195 /*
1196 * In case of a posix_spawn operation, the child doing the exec
1197 * might not hold the reader lock on exec_lock, but the parent
1198 * will do this instead.
1199 */
1200 KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1201 KASSERT(!no_local_exec_lock || is_spawn);
1202 KASSERT(data != NULL);
1203
1204 p = l->l_proc;
1205
1206 /* Get rid of other LWPs. */
1207 if (p->p_nlwps > 1) {
1208 mutex_enter(p->p_lock);
1209 exit_lwps(l);
1210 mutex_exit(p->p_lock);
1211 }
1212 KDASSERT(p->p_nlwps == 1);
1213
1214 /*
1215 * All of the other LWPs got rid of their robust futexes
1216 * when they exited above, but we might still have some
1217 * to dispose of. Do that now.
1218 */
1219 if (__predict_false(l->l_robust_head != 0)) {
1220 futex_release_all_lwp(l);
1221 /*
1222 * Since this LWP will live on with a different
1223 * program image, we need to clear the robust
1224 * futex list pointer here.
1225 */
1226 l->l_robust_head = 0;
1227 }
1228
1229 /* Destroy any lwpctl info. */
1230 if (p->p_lwpctl != NULL)
1231 lwp_ctl_exit();
1232
1233 /* Remove POSIX timers */
1234 ptimers_free(p, TIMERS_POSIX);
1235
1236 /* Set the PaX flags. */
1237 pax_set_flags(epp, p);
1238
1239 /*
1240 * Do whatever is necessary to prepare the address space
1241 * for remapping. Note that this might replace the current
1242 * vmspace with another!
1243 *
1244 * vfork(): do not touch any user space data in the new child
1245 * until we have awoken the parent below, or it will defeat
1246 * lazy pmap switching (on x86).
1247 */
1248 uvmspace_exec(l, epp->ep_vm_minaddr, epp->ep_vm_maxaddr,
1249 epp->ep_flags & EXEC_TOPDOWN_VM);
1250 vm = p->p_vmspace;
1251
1252 vm->vm_taddr = (void *)epp->ep_taddr;
1253 vm->vm_tsize = btoc(epp->ep_tsize);
1254 vm->vm_daddr = (void*)epp->ep_daddr;
1255 vm->vm_dsize = btoc(epp->ep_dsize);
1256 vm->vm_ssize = btoc(epp->ep_ssize);
1257 vm->vm_issize = 0;
1258 vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1259 vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1260
1261 pax_aslr_init_vm(l, vm, epp);
1262
1263 cwdexec(p);
1264 fd_closeexec(); /* handle close on exec */
1265
1266 if (__predict_false(ktrace_on))
1267 fd_ktrexecfd();
1268
1269 execsigs(p); /* reset caught signals */
1270
1271 mutex_enter(p->p_lock);
1272 l->l_ctxlink = NULL; /* reset ucontext link */
1273 p->p_acflag &= ~AFORK;
1274 p->p_flag |= PK_EXEC;
1275 mutex_exit(p->p_lock);
1276
1277 error = credexec(l, data);
1278 if (error)
1279 goto exec_abort;
1280
1281 #if defined(__HAVE_RAS)
1282 /*
1283 * Remove all RASs from the address space.
1284 */
1285 ras_purgeall();
1286 #endif
1287
1288 /*
1289 * Stop profiling.
1290 */
1291 if ((p->p_stflag & PST_PROFIL) != 0) {
1292 mutex_spin_enter(&p->p_stmutex);
1293 stopprofclock(p);
1294 mutex_spin_exit(&p->p_stmutex);
1295 }
1296
1297 /*
1298 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1299 * exited and exec()/exit() are the only places it will be cleared.
1300 *
1301 * Once the parent has been awoken, curlwp may teleport to a new CPU
1302 * in sched_vforkexec(), and it's then OK to start messing with user
1303 * data. See comment above.
1304 */
1305 if ((p->p_lflag & PL_PPWAIT) != 0) {
1306 bool samecpu;
1307 lwp_t *lp;
1308
1309 mutex_enter(&proc_lock);
1310 lp = p->p_vforklwp;
1311 p->p_vforklwp = NULL;
1312 l->l_lwpctl = NULL; /* was on loan from blocked parent */
1313
1314 /* Clear flags after cv_broadcast() (scheduler needs them). */
1315 p->p_lflag &= ~PL_PPWAIT;
1316 lp->l_vforkwaiting = false;
1317
1318 /* If parent is still on same CPU, teleport curlwp elsewhere. */
1319 samecpu = (lp->l_cpu == curlwp->l_cpu);
1320 cv_broadcast(&lp->l_waitcv);
1321 mutex_exit(&proc_lock);
1322
1323 /* Give the parent its CPU back - find a new home. */
1324 KASSERT(!is_spawn);
1325 sched_vforkexec(l, samecpu);
1326 }
1327
1328 /* Now map address space. */
1329 error = execve_dovmcmds(l, data);
1330 if (error != 0)
1331 goto exec_abort;
1332
1333 pathexec(p, epp->ep_resolvedname);
1334
1335 char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1336
1337 error = copyoutargs(data, l, newstack);
1338 if (error != 0)
1339 goto exec_abort;
1340
1341 doexechooks(p);
1342
1343 /*
1344 * Set initial SP at the top of the stack.
1345 *
1346 * Note that on machines where stack grows up (e.g. hppa), SP points to
1347 * the end of arg/env strings. Userland guesses the address of argc
1348 * via ps_strings::ps_argvstr.
1349 */
1350
1351 /* Setup new registers and do misc. setup. */
1352 (*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1353 if (epp->ep_esch->es_setregs)
1354 (*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1355
1356 /* Provide a consistent LWP private setting */
1357 (void)lwp_setprivate(l, NULL);
1358
1359 /* Discard all PCU state; need to start fresh */
1360 pcu_discard_all(l);
1361
1362 /* map the process's signal trampoline code */
1363 if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1364 DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1365 goto exec_abort;
1366 }
1367
1368 pool_put(&exec_pool, data->ed_argp);
1369
1370 /*
1371 * Notify anyone who might care that we've exec'd.
1372 *
1373 * This is slightly racy; someone could sneak in and
1374 * attach a knote after we've decided not to notify,
1375 * or vice-versa, but that's not particularly bothersome.
1376 * knote_proc_exec() will acquire p->p_lock as needed.
1377 */
1378 if (!SLIST_EMPTY(&p->p_klist)) {
1379 knote_proc_exec(p);
1380 }
1381
1382 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1383
1384 SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1385
1386 emulexec(l, epp);
1387
1388 /* Allow new references from the debugger/procfs. */
1389 rw_exit(&p->p_reflock);
1390 if (!no_local_exec_lock)
1391 rw_exit(&exec_lock);
1392
1393 mutex_enter(&proc_lock);
1394
1395 /* posix_spawn(3) reports a single event with implied exec(3) */
1396 if ((p->p_slflag & PSL_TRACED) && !is_spawn) {
1397 mutex_enter(p->p_lock);
1398 eventswitch(TRAP_EXEC, 0, 0);
1399 mutex_enter(&proc_lock);
1400 }
1401
1402 if (p->p_sflag & PS_STOPEXEC) {
1403 ksiginfoq_t kq;
1404
1405 KASSERT(l->l_blcnt == 0);
1406 p->p_pptr->p_nstopchild++;
1407 p->p_waited = 0;
1408 mutex_enter(p->p_lock);
1409 ksiginfo_queue_init(&kq);
1410 sigclearall(p, &contsigmask, &kq);
1411 lwp_lock(l);
1412 l->l_stat = LSSTOP;
1413 p->p_stat = SSTOP;
1414 p->p_nrlwps--;
1415 lwp_unlock(l);
1416 mutex_exit(p->p_lock);
1417 mutex_exit(&proc_lock);
1418 lwp_lock(l);
1419 spc_lock(l->l_cpu);
1420 mi_switch(l);
1421 ksiginfo_queue_drain(&kq);
1422 } else {
1423 mutex_exit(&proc_lock);
1424 }
1425
1426 exec_path_free(data);
1427 #ifdef TRACE_EXEC
1428 DPRINTF(("%s finished\n", __func__));
1429 #endif
1430 return EJUSTRETURN;
1431
1432 exec_abort:
1433 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1434 rw_exit(&p->p_reflock);
1435 if (!no_local_exec_lock)
1436 rw_exit(&exec_lock);
1437
1438 exec_path_free(data);
1439
1440 /*
1441 * the old process doesn't exist anymore. exit gracefully.
1442 * get rid of the (new) address space we have created, if any, get rid
1443 * of our namei data and vnode, and exit noting failure
1444 */
1445 if (vm != NULL) {
1446 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1447 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1448 }
1449
1450 exec_free_emul_arg(epp);
1451 pool_put(&exec_pool, data->ed_argp);
1452 kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1453 if (epp->ep_emul_root != NULL)
1454 vrele(epp->ep_emul_root);
1455 if (epp->ep_interp != NULL)
1456 vrele(epp->ep_interp);
1457
1458 /* Acquire the sched-state mutex (exit1() will release it). */
1459 if (!is_spawn) {
1460 mutex_enter(p->p_lock);
1461 exit1(l, error, SIGABRT);
1462 }
1463
1464 return error;
1465 }
1466
1467 int
execve1(struct lwp * l,bool has_path,const char * path,int fd,char * const * args,char * const * envs,execve_fetch_element_t fetch_element)1468 execve1(struct lwp *l, bool has_path, const char *path, int fd,
1469 char * const *args, char * const *envs,
1470 execve_fetch_element_t fetch_element)
1471 {
1472 struct execve_data data;
1473 int error;
1474
1475 error = execve_loadvm(l, has_path, path, fd, args, envs, fetch_element,
1476 &data);
1477 if (error)
1478 return error;
1479 error = execve_runproc(l, &data, false, false);
1480 return error;
1481 }
1482
1483 static size_t
fromptrsz(const struct exec_package * epp)1484 fromptrsz(const struct exec_package *epp)
1485 {
1486 return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1487 }
1488
1489 static size_t
ptrsz(const struct exec_package * epp)1490 ptrsz(const struct exec_package *epp)
1491 {
1492 return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1493 }
1494
1495 static size_t
calcargs(struct execve_data * restrict data,const size_t argenvstrlen)1496 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1497 {
1498 struct exec_package * const epp = &data->ed_pack;
1499
1500 const size_t nargenvptrs =
1501 1 + /* long argc */
1502 data->ed_argc + /* char *argv[] */
1503 1 + /* \0 */
1504 data->ed_envc + /* char *env[] */
1505 1; /* \0 */
1506
1507 return (nargenvptrs * ptrsz(epp)) /* pointers */
1508 + argenvstrlen /* strings */
1509 + epp->ep_esch->es_arglen; /* auxinfo */
1510 }
1511
1512 static size_t
calcstack(struct execve_data * restrict data,const size_t gaplen)1513 calcstack(struct execve_data * restrict data, const size_t gaplen)
1514 {
1515 struct exec_package * const epp = &data->ed_pack;
1516
1517 data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1518 epp->ep_esch->es_emul->e_sigcode;
1519
1520 data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1521 sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1522
1523 const size_t sigcode_psstr_sz =
1524 data->ed_szsigcode + /* sigcode */
1525 data->ed_ps_strings_sz + /* ps_strings */
1526 STACK_PTHREADSPACE; /* pthread space */
1527
1528 const size_t stacklen =
1529 data->ed_argslen +
1530 gaplen +
1531 sigcode_psstr_sz;
1532
1533 /* make the stack "safely" aligned */
1534 return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1535 }
1536
1537 static int
copyoutargs(struct execve_data * restrict data,struct lwp * l,char * const newstack)1538 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1539 char * const newstack)
1540 {
1541 struct exec_package * const epp = &data->ed_pack;
1542 struct proc *p = l->l_proc;
1543 int error;
1544
1545 memset(&data->ed_arginfo, 0, sizeof(data->ed_arginfo));
1546
1547 /* remember information about the process */
1548 data->ed_arginfo.ps_nargvstr = data->ed_argc;
1549 data->ed_arginfo.ps_nenvstr = data->ed_envc;
1550
1551 /*
1552 * Allocate the stack address passed to the newly execve()'ed process.
1553 *
1554 * The new stack address will be set to the SP (stack pointer) register
1555 * in setregs().
1556 */
1557
1558 char *newargs = STACK_ALLOC(
1559 STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1560
1561 error = (*epp->ep_esch->es_copyargs)(l, epp,
1562 &data->ed_arginfo, &newargs, data->ed_argp);
1563
1564 if (error) {
1565 DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1566 return error;
1567 }
1568
1569 error = copyoutpsstrs(data, p);
1570 if (error != 0)
1571 return error;
1572
1573 return 0;
1574 }
1575
1576 static int
copyoutpsstrs(struct execve_data * restrict data,struct proc * p)1577 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1578 {
1579 struct exec_package * const epp = &data->ed_pack;
1580 struct ps_strings32 arginfo32;
1581 void *aip;
1582 int error;
1583
1584 /* fill process ps_strings info */
1585 p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1586 STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1587
1588 if (epp->ep_flags & EXEC_32) {
1589 aip = &arginfo32;
1590 arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1591 arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1592 arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1593 arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1594 } else
1595 aip = &data->ed_arginfo;
1596
1597 /* copy out the process's ps_strings structure */
1598 if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1599 != 0) {
1600 DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1601 __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1602 return error;
1603 }
1604
1605 return 0;
1606 }
1607
1608 static int
copyinargs(struct execve_data * restrict data,char * const * args,char * const * envs,execve_fetch_element_t fetch_element,char ** dpp)1609 copyinargs(struct execve_data * restrict data, char * const *args,
1610 char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1611 {
1612 struct exec_package * const epp = &data->ed_pack;
1613 char *dp;
1614 size_t i;
1615 int error;
1616
1617 dp = *dpp;
1618
1619 data->ed_argc = 0;
1620
1621 /* copy the fake args list, if there's one, freeing it as we go */
1622 if (epp->ep_flags & EXEC_HASARGL) {
1623 struct exec_fakearg *fa = epp->ep_fa;
1624
1625 while (fa->fa_arg != NULL) {
1626 const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1627 size_t len;
1628
1629 len = strlcpy(dp, fa->fa_arg, maxlen);
1630 /* Count NUL into len. */
1631 if (len < maxlen)
1632 len++;
1633 else {
1634 while (fa->fa_arg != NULL) {
1635 kmem_free(fa->fa_arg, fa->fa_len);
1636 fa++;
1637 }
1638 kmem_free(epp->ep_fa, epp->ep_fa_len);
1639 epp->ep_flags &= ~EXEC_HASARGL;
1640 return SET_ERROR(E2BIG);
1641 }
1642 ktrexecarg(fa->fa_arg, len - 1);
1643 dp += len;
1644
1645 kmem_free(fa->fa_arg, fa->fa_len);
1646 fa++;
1647 data->ed_argc++;
1648 }
1649 kmem_free(epp->ep_fa, epp->ep_fa_len);
1650 epp->ep_flags &= ~EXEC_HASARGL;
1651 }
1652
1653 /*
1654 * Read and count argument strings from user.
1655 */
1656
1657 if (args == NULL) {
1658 DPRINTF(("%s: null args\n", __func__));
1659 return SET_ERROR(EINVAL);
1660 }
1661 if (epp->ep_flags & EXEC_SKIPARG)
1662 args = (const void *)((const char *)args + fromptrsz(epp));
1663 i = 0;
1664 error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1665 if (error != 0) {
1666 DPRINTF(("%s: copyin arg %d\n", __func__, error));
1667 return error;
1668 }
1669 data->ed_argc += i;
1670
1671 /*
1672 * Read and count environment strings from user.
1673 */
1674
1675 data->ed_envc = 0;
1676 /* environment need not be there */
1677 if (envs == NULL)
1678 goto done;
1679 i = 0;
1680 error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1681 if (error != 0) {
1682 DPRINTF(("%s: copyin env %d\n", __func__, error));
1683 return error;
1684 }
1685 data->ed_envc += i;
1686
1687 done:
1688 *dpp = dp;
1689
1690 return 0;
1691 }
1692
1693 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))1694 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1695 execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1696 void (*ktr)(const void *, size_t))
1697 {
1698 char *dp, *sp;
1699 size_t i;
1700 int error;
1701
1702 dp = *dpp;
1703
1704 i = 0;
1705 while (1) {
1706 const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1707 size_t len;
1708
1709 if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1710 return error;
1711 }
1712 if (!sp)
1713 break;
1714 if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1715 if (error == ENAMETOOLONG)
1716 error = SET_ERROR(E2BIG);
1717 return error;
1718 }
1719 if (__predict_false(ktrace_on))
1720 (*ktr)(dp, len - 1);
1721 dp += len;
1722 i++;
1723 }
1724
1725 *dpp = dp;
1726 *ip = i;
1727
1728 return 0;
1729 }
1730
1731 /*
1732 * Copy argv and env strings from kernel buffer (argp) to the new stack.
1733 * Those strings are located just after auxinfo.
1734 */
1735 int
copyargs(struct lwp * l,struct exec_package * pack,struct ps_strings * arginfo,char ** stackp,void * argp)1736 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1737 char **stackp, void *argp)
1738 {
1739 char **cpp, *dp, *sp;
1740 size_t len;
1741 void *nullp;
1742 long argc, envc;
1743 int error;
1744
1745 cpp = (char **)*stackp;
1746 nullp = NULL;
1747 argc = arginfo->ps_nargvstr;
1748 envc = arginfo->ps_nenvstr;
1749
1750 /* argc on stack is long */
1751 CTASSERT(sizeof(*cpp) == sizeof(argc));
1752
1753 dp = (char *)(cpp +
1754 1 + /* long argc */
1755 argc + /* char *argv[] */
1756 1 + /* \0 */
1757 envc + /* char *env[] */
1758 1) + /* \0 */
1759 pack->ep_esch->es_arglen; /* auxinfo */
1760 sp = argp;
1761
1762 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1763 COPYPRINTF("", cpp - 1, sizeof(argc));
1764 return error;
1765 }
1766
1767 /* XXX don't copy them out, remap them! */
1768 arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1769
1770 for (; --argc >= 0; sp += len, dp += len) {
1771 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1772 COPYPRINTF("", cpp - 1, sizeof(dp));
1773 return error;
1774 }
1775 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1776 COPYPRINTF("str", dp, (size_t)ARG_MAX);
1777 return error;
1778 }
1779 }
1780
1781 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1782 COPYPRINTF("", cpp - 1, sizeof(nullp));
1783 return error;
1784 }
1785
1786 arginfo->ps_envstr = cpp; /* remember location of envp for later */
1787
1788 for (; --envc >= 0; sp += len, dp += len) {
1789 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1790 COPYPRINTF("", cpp - 1, sizeof(dp));
1791 return error;
1792 }
1793 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1794 COPYPRINTF("str", dp, (size_t)ARG_MAX);
1795 return error;
1796 }
1797
1798 }
1799
1800 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1801 COPYPRINTF("", cpp - 1, sizeof(nullp));
1802 return error;
1803 }
1804
1805 *stackp = (char *)cpp;
1806 return 0;
1807 }
1808
1809
1810 /*
1811 * Add execsw[] entries.
1812 */
1813 int
exec_add(struct execsw * esp,int count)1814 exec_add(struct execsw *esp, int count)
1815 {
1816 struct exec_entry *it;
1817 int i, error = 0;
1818
1819 if (count == 0) {
1820 return 0;
1821 }
1822
1823 /* Check for duplicates. */
1824 rw_enter(&exec_lock, RW_WRITER);
1825 for (i = 0; i < count; i++) {
1826 LIST_FOREACH(it, &ex_head, ex_list) {
1827 /* assume unique (makecmds, probe_func, emulation) */
1828 if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1829 it->ex_sw->u.elf_probe_func ==
1830 esp[i].u.elf_probe_func &&
1831 it->ex_sw->es_emul == esp[i].es_emul) {
1832 rw_exit(&exec_lock);
1833 return SET_ERROR(EEXIST);
1834 }
1835 }
1836 }
1837
1838 /* Allocate new entries. */
1839 for (i = 0; i < count; i++) {
1840 it = kmem_alloc(sizeof(*it), KM_SLEEP);
1841 it->ex_sw = &esp[i];
1842 error = exec_sigcode_alloc(it->ex_sw->es_emul);
1843 if (error != 0) {
1844 kmem_free(it, sizeof(*it));
1845 break;
1846 }
1847 LIST_INSERT_HEAD(&ex_head, it, ex_list);
1848 }
1849 /* If even one fails, remove them all back. */
1850 if (error != 0) {
1851 for (i--; i >= 0; i--) {
1852 it = LIST_FIRST(&ex_head);
1853 LIST_REMOVE(it, ex_list);
1854 exec_sigcode_free(it->ex_sw->es_emul);
1855 kmem_free(it, sizeof(*it));
1856 }
1857 rw_exit(&exec_lock);
1858 return error;
1859 }
1860
1861 /* update execsw[] */
1862 exec_init(0);
1863 rw_exit(&exec_lock);
1864 return 0;
1865 }
1866
1867 /*
1868 * Remove execsw[] entry.
1869 */
1870 int
exec_remove(struct execsw * esp,int count)1871 exec_remove(struct execsw *esp, int count)
1872 {
1873 struct exec_entry *it, *next;
1874 int i;
1875 const struct proclist_desc *pd;
1876 proc_t *p;
1877
1878 if (count == 0) {
1879 return 0;
1880 }
1881
1882 /* Abort if any are busy. */
1883 rw_enter(&exec_lock, RW_WRITER);
1884 for (i = 0; i < count; i++) {
1885 mutex_enter(&proc_lock);
1886 for (pd = proclists; pd->pd_list != NULL; pd++) {
1887 PROCLIST_FOREACH(p, pd->pd_list) {
1888 if (p->p_execsw == &esp[i]) {
1889 mutex_exit(&proc_lock);
1890 rw_exit(&exec_lock);
1891 return SET_ERROR(EBUSY);
1892 }
1893 }
1894 }
1895 mutex_exit(&proc_lock);
1896 }
1897
1898 /* None are busy, so remove them all. */
1899 for (i = 0; i < count; i++) {
1900 for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1901 next = LIST_NEXT(it, ex_list);
1902 if (it->ex_sw == &esp[i]) {
1903 LIST_REMOVE(it, ex_list);
1904 exec_sigcode_free(it->ex_sw->es_emul);
1905 kmem_free(it, sizeof(*it));
1906 break;
1907 }
1908 }
1909 }
1910
1911 /* update execsw[] */
1912 exec_init(0);
1913 rw_exit(&exec_lock);
1914 return 0;
1915 }
1916
1917 /*
1918 * Initialize exec structures. If init_boot is true, also does necessary
1919 * one-time initialization (it's called from main() that way).
1920 * Once system is multiuser, this should be called with exec_lock held,
1921 * i.e. via exec_{add|remove}().
1922 */
1923 int
exec_init(int init_boot)1924 exec_init(int init_boot)
1925 {
1926 const struct execsw **sw;
1927 struct exec_entry *ex;
1928 SLIST_HEAD(,exec_entry) first;
1929 SLIST_HEAD(,exec_entry) any;
1930 SLIST_HEAD(,exec_entry) last;
1931 int i, sz;
1932
1933 if (init_boot) {
1934 /* do one-time initializations */
1935 vaddr_t vmin = 0, vmax;
1936
1937 rw_init(&exec_lock);
1938 exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax,
1939 maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL);
1940 pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1941 "execargs", &exec_palloc, IPL_NONE);
1942 pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1943 } else {
1944 KASSERT(rw_write_held(&exec_lock));
1945 }
1946
1947 /* Sort each entry onto the appropriate queue. */
1948 SLIST_INIT(&first);
1949 SLIST_INIT(&any);
1950 SLIST_INIT(&last);
1951 sz = 0;
1952 LIST_FOREACH(ex, &ex_head, ex_list) {
1953 switch(ex->ex_sw->es_prio) {
1954 case EXECSW_PRIO_FIRST:
1955 SLIST_INSERT_HEAD(&first, ex, ex_slist);
1956 break;
1957 case EXECSW_PRIO_ANY:
1958 SLIST_INSERT_HEAD(&any, ex, ex_slist);
1959 break;
1960 case EXECSW_PRIO_LAST:
1961 SLIST_INSERT_HEAD(&last, ex, ex_slist);
1962 break;
1963 default:
1964 panic("%s", __func__);
1965 break;
1966 }
1967 sz++;
1968 }
1969
1970 /*
1971 * Create new execsw[]. Ensure we do not try a zero-sized
1972 * allocation.
1973 */
1974 sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1975 i = 0;
1976 SLIST_FOREACH(ex, &first, ex_slist) {
1977 sw[i++] = ex->ex_sw;
1978 }
1979 SLIST_FOREACH(ex, &any, ex_slist) {
1980 sw[i++] = ex->ex_sw;
1981 }
1982 SLIST_FOREACH(ex, &last, ex_slist) {
1983 sw[i++] = ex->ex_sw;
1984 }
1985
1986 /* Replace old execsw[] and free used memory. */
1987 if (execsw != NULL) {
1988 kmem_free(__UNCONST(execsw),
1989 nexecs * sizeof(struct execsw *) + 1);
1990 }
1991 execsw = sw;
1992 nexecs = sz;
1993
1994 /* Figure out the maximum size of an exec header. */
1995 exec_maxhdrsz = sizeof(int);
1996 for (i = 0; i < nexecs; i++) {
1997 if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1998 exec_maxhdrsz = execsw[i]->es_hdrsz;
1999 }
2000
2001 return 0;
2002 }
2003
2004 int
exec_sigcode_alloc(const struct emul * e)2005 exec_sigcode_alloc(const struct emul *e)
2006 {
2007 vaddr_t va;
2008 vsize_t sz;
2009 int error;
2010 struct uvm_object *uobj;
2011
2012 KASSERT(rw_lock_held(&exec_lock));
2013
2014 if (e == NULL || e->e_sigobject == NULL)
2015 return 0;
2016
2017 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
2018 if (sz == 0)
2019 return 0;
2020
2021 /*
2022 * Create a sigobject for this emulation.
2023 *
2024 * sigobject is an anonymous memory object (just like SYSV shared
2025 * memory) that we keep a permanent reference to and that we map
2026 * in all processes that need this sigcode. The creation is simple,
2027 * we create an object, add a permanent reference to it, map it in
2028 * kernel space, copy out the sigcode to it and unmap it.
2029 * We map it with PROT_READ|PROT_EXEC into the process just
2030 * the way sys_mmap() would map it.
2031 */
2032 if (*e->e_sigobject == NULL) {
2033 uobj = uao_create(sz, 0);
2034 (*uobj->pgops->pgo_reference)(uobj);
2035 va = vm_map_min(kernel_map);
2036 if ((error = uvm_map(kernel_map, &va, round_page(sz),
2037 uobj, 0, 0,
2038 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
2039 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
2040 printf("sigcode kernel mapping failed %d\n", error);
2041 (*uobj->pgops->pgo_detach)(uobj);
2042 return error;
2043 }
2044 memcpy((void *)va, e->e_sigcode, sz);
2045 #ifdef PMAP_NEED_PROCWR
2046 pmap_procwr(&proc0, va, sz);
2047 #endif
2048 uvm_unmap(kernel_map, va, va + round_page(sz));
2049 *e->e_sigobject = uobj;
2050 KASSERT(uobj->uo_refs == 1);
2051 } else {
2052 /* if already created, reference++ */
2053 uobj = *e->e_sigobject;
2054 (*uobj->pgops->pgo_reference)(uobj);
2055 }
2056
2057 return 0;
2058 }
2059
2060 void
exec_sigcode_free(const struct emul * e)2061 exec_sigcode_free(const struct emul *e)
2062 {
2063 struct uvm_object *uobj;
2064
2065 KASSERT(rw_lock_held(&exec_lock));
2066
2067 if (e == NULL || e->e_sigobject == NULL)
2068 return;
2069
2070 uobj = *e->e_sigobject;
2071 if (uobj == NULL)
2072 return;
2073
2074 if (uobj->uo_refs == 1)
2075 *e->e_sigobject = NULL; /* I'm the last person to reference. */
2076 (*uobj->pgops->pgo_detach)(uobj);
2077 }
2078
2079 static int
exec_sigcode_map(struct proc * p,const struct emul * e)2080 exec_sigcode_map(struct proc *p, const struct emul *e)
2081 {
2082 vaddr_t va;
2083 vsize_t sz;
2084 int error;
2085 struct uvm_object *uobj;
2086
2087 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
2088 if (e->e_sigobject == NULL || sz == 0)
2089 return 0;
2090
2091 uobj = *e->e_sigobject;
2092 if (uobj == NULL)
2093 return 0;
2094
2095 /* Just a hint to uvm_map where to put it. */
2096 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
2097 round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
2098
2099 #ifdef __alpha__
2100 /*
2101 * Tru64 puts /sbin/loader at the end of user virtual memory,
2102 * which causes the above calculation to put the sigcode at
2103 * an invalid address. Put it just below the text instead.
2104 */
2105 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
2106 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
2107 }
2108 #endif
2109
2110 (*uobj->pgops->pgo_reference)(uobj);
2111 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
2112 uobj, 0, 0,
2113 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
2114 UVM_ADV_RANDOM, 0));
2115 if (error) {
2116 DPRINTF(("%s, %d: map %p "
2117 "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
2118 __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
2119 va, error));
2120 (*uobj->pgops->pgo_detach)(uobj);
2121 return error;
2122 }
2123 p->p_sigctx.ps_sigcode = (void *)va;
2124 return 0;
2125 }
2126
2127 /*
2128 * Release a refcount on spawn_exec_data and destroy memory, if this
2129 * was the last one.
2130 */
2131 static void
spawn_exec_data_release(struct spawn_exec_data * data)2132 spawn_exec_data_release(struct spawn_exec_data *data)
2133 {
2134
2135 membar_release();
2136 if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
2137 return;
2138 membar_acquire();
2139
2140 cv_destroy(&data->sed_cv_child_ready);
2141 mutex_destroy(&data->sed_mtx_child);
2142
2143 if (data->sed_actions)
2144 posix_spawn_fa_free(data->sed_actions,
2145 data->sed_actions->len);
2146 if (data->sed_attrs)
2147 kmem_free(data->sed_attrs,
2148 sizeof(*data->sed_attrs));
2149 kmem_free(data, sizeof(*data));
2150 }
2151
2152 static int
handle_posix_spawn_file_actions(struct posix_spawn_file_actions * actions)2153 handle_posix_spawn_file_actions(struct posix_spawn_file_actions *actions)
2154 {
2155 struct lwp *l = curlwp;
2156 register_t retval;
2157 int error, newfd;
2158
2159 if (actions == NULL)
2160 return 0;
2161
2162 for (size_t i = 0; i < actions->len; i++) {
2163 const struct posix_spawn_file_actions_entry *fae =
2164 &actions->fae[i];
2165 switch (fae->fae_action) {
2166 case FAE_OPEN:
2167 if (fd_getfile(fae->fae_fildes) != NULL) {
2168 error = fd_close(fae->fae_fildes);
2169 if (error)
2170 return error;
2171 }
2172 error = fd_open(fae->fae_path, fae->fae_oflag,
2173 fae->fae_mode, &newfd);
2174 if (error)
2175 return error;
2176 if (newfd != fae->fae_fildes) {
2177 error = dodup(l, newfd,
2178 fae->fae_fildes, 0, &retval);
2179 if (fd_getfile(newfd) != NULL)
2180 fd_close(newfd);
2181 }
2182 break;
2183 case FAE_DUP2:
2184 error = dodup(l, fae->fae_fildes,
2185 fae->fae_newfildes, 0, &retval);
2186 break;
2187 case FAE_CLOSE:
2188 if (fd_getfile(fae->fae_fildes) == NULL) {
2189 return SET_ERROR(EBADF);
2190 }
2191 error = fd_close(fae->fae_fildes);
2192 break;
2193 case FAE_CHDIR:
2194 error = do_sys_chdir(l, fae->fae_chdir_path,
2195 UIO_SYSSPACE, &retval);
2196 break;
2197 case FAE_FCHDIR:
2198 error = do_sys_fchdir(l, fae->fae_fildes, &retval);
2199 break;
2200 }
2201 if (error)
2202 return error;
2203 }
2204 return 0;
2205 }
2206
2207 static int
handle_posix_spawn_attrs(struct posix_spawnattr * attrs,struct proc * parent)2208 handle_posix_spawn_attrs(struct posix_spawnattr *attrs, struct proc *parent)
2209 {
2210 struct sigaction sigact;
2211 int error = 0;
2212 struct proc *p = curproc;
2213 struct lwp *l = curlwp;
2214
2215 if (attrs == NULL)
2216 return 0;
2217
2218 memset(&sigact, 0, sizeof(sigact));
2219 sigact._sa_u._sa_handler = SIG_DFL;
2220 sigact.sa_flags = 0;
2221
2222 /*
2223 * set state to SSTOP so that this proc can be found by pid.
2224 * see proc_enterprp, do_sched_setparam below
2225 */
2226 mutex_enter(&proc_lock);
2227 /*
2228 * p_stat should be SACTIVE, so we need to adjust the
2229 * parent's p_nstopchild here. For safety, just make
2230 * we're on the good side of SDEAD before we adjust.
2231 */
2232 int ostat = p->p_stat;
2233 KASSERT(ostat < SSTOP);
2234 p->p_stat = SSTOP;
2235 p->p_waited = 0;
2236 p->p_pptr->p_nstopchild++;
2237 mutex_exit(&proc_lock);
2238
2239 /* Set process group */
2240 if (attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2241 pid_t mypid = p->p_pid;
2242 pid_t pgrp = attrs->sa_pgroup;
2243
2244 if (pgrp == 0)
2245 pgrp = mypid;
2246
2247 error = proc_enterpgrp(parent, mypid, pgrp, false);
2248 if (error)
2249 goto out;
2250 }
2251
2252 /* Set scheduler policy */
2253 if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2254 error = do_sched_setparam(p->p_pid, 0, attrs->sa_schedpolicy,
2255 &attrs->sa_schedparam);
2256 else if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDPARAM) {
2257 error = do_sched_setparam(parent->p_pid, 0,
2258 SCHED_NONE, &attrs->sa_schedparam);
2259 }
2260 if (error)
2261 goto out;
2262
2263 /* Reset user ID's */
2264 if (attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2265 error = do_setresgid(l, -1, kauth_cred_getgid(l->l_cred), -1,
2266 ID_E_EQ_R | ID_E_EQ_S);
2267 if (error)
2268 return error;
2269 error = do_setresuid(l, -1, kauth_cred_getuid(l->l_cred), -1,
2270 ID_E_EQ_R | ID_E_EQ_S);
2271 if (error)
2272 goto out;
2273 }
2274
2275 /* Set signal masks/defaults */
2276 if (attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2277 mutex_enter(p->p_lock);
2278 error = sigprocmask1(l, SIG_SETMASK, &attrs->sa_sigmask, NULL);
2279 mutex_exit(p->p_lock);
2280 if (error)
2281 goto out;
2282 }
2283
2284 if (attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2285 /*
2286 * The following sigaction call is using a sigaction
2287 * version 0 trampoline which is in the compatibility
2288 * code only. This is not a problem because for SIG_DFL
2289 * and SIG_IGN, the trampolines are now ignored. If they
2290 * were not, this would be a problem because we are
2291 * holding the exec_lock, and the compat code needs
2292 * to do the same in order to replace the trampoline
2293 * code of the process.
2294 */
2295 for (int i = 1; i <= NSIG; i++) {
2296 if (sigismember(&attrs->sa_sigdefault, i))
2297 sigaction1(l, i, &sigact, NULL, NULL, 0);
2298 }
2299 }
2300 out:
2301 mutex_enter(&proc_lock);
2302 p->p_stat = ostat;
2303 p->p_pptr->p_nstopchild--;
2304 mutex_exit(&proc_lock);
2305 return error;
2306 }
2307
2308 /*
2309 * A child lwp of a posix_spawn operation starts here and ends up in
2310 * cpu_spawn_return, dealing with all filedescriptor and scheduler
2311 * manipulations in between.
2312 * The parent waits for the child, as it is not clear whether the child
2313 * will be able to acquire its own exec_lock. If it can, the parent can
2314 * be released early and continue running in parallel. If not (or if the
2315 * magic debug flag is passed in the scheduler attribute struct), the
2316 * child rides on the parent's exec lock until it is ready to return to
2317 * to userland - and only then releases the parent. This method loses
2318 * concurrency, but improves error reporting.
2319 */
2320 static void
spawn_return(void * arg)2321 spawn_return(void *arg)
2322 {
2323 struct spawn_exec_data *spawn_data = arg;
2324 struct lwp *l = curlwp;
2325 struct proc *p = l->l_proc;
2326 int error;
2327 bool have_reflock;
2328 bool parent_is_waiting = true;
2329
2330 /*
2331 * Check if we can release parent early.
2332 * We either need to have no sed_attrs, or sed_attrs does not
2333 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2334 * safe access to the parent proc (passed in sed_parent).
2335 * We then try to get the exec_lock, and only if that works, we can
2336 * release the parent here already.
2337 */
2338 struct posix_spawnattr *attrs = spawn_data->sed_attrs;
2339 if ((!attrs || (attrs->sa_flags
2340 & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2341 && rw_tryenter(&exec_lock, RW_READER)) {
2342 parent_is_waiting = false;
2343 mutex_enter(&spawn_data->sed_mtx_child);
2344 KASSERT(!spawn_data->sed_child_ready);
2345 spawn_data->sed_error = 0;
2346 spawn_data->sed_child_ready = true;
2347 cv_signal(&spawn_data->sed_cv_child_ready);
2348 mutex_exit(&spawn_data->sed_mtx_child);
2349 }
2350
2351 /* don't allow debugger access yet */
2352 rw_enter(&p->p_reflock, RW_WRITER);
2353 have_reflock = true;
2354
2355 /* handle posix_spawnattr */
2356 error = handle_posix_spawn_attrs(attrs, spawn_data->sed_parent);
2357 if (error)
2358 goto report_error;
2359
2360 /* handle posix_spawn_file_actions */
2361 error = handle_posix_spawn_file_actions(spawn_data->sed_actions);
2362 if (error)
2363 goto report_error;
2364
2365 /* now do the real exec */
2366 error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2367 true);
2368 have_reflock = false;
2369 if (error == EJUSTRETURN)
2370 error = 0;
2371 else if (error)
2372 goto report_error;
2373
2374 if (parent_is_waiting) {
2375 mutex_enter(&spawn_data->sed_mtx_child);
2376 KASSERT(!spawn_data->sed_child_ready);
2377 spawn_data->sed_error = 0;
2378 spawn_data->sed_child_ready = true;
2379 cv_signal(&spawn_data->sed_cv_child_ready);
2380 mutex_exit(&spawn_data->sed_mtx_child);
2381 }
2382
2383 /* release our refcount on the data */
2384 spawn_exec_data_release(spawn_data);
2385
2386 if ((p->p_slflag & (PSL_TRACED|PSL_TRACEDCHILD)) ==
2387 (PSL_TRACED|PSL_TRACEDCHILD)) {
2388 eventswitchchild(p, TRAP_CHLD, PTRACE_POSIX_SPAWN);
2389 }
2390
2391 /* and finally: leave to userland for the first time */
2392 cpu_spawn_return(l);
2393
2394 /* NOTREACHED */
2395 return;
2396
2397 report_error:
2398 if (have_reflock) {
2399 /*
2400 * We have not passed through execve_runproc(),
2401 * which would have released the p_reflock and also
2402 * taken ownership of the sed_exec part of spawn_data,
2403 * so release/free both here.
2404 */
2405 rw_exit(&p->p_reflock);
2406 execve_free_data(&spawn_data->sed_exec);
2407 }
2408
2409 if (parent_is_waiting) {
2410 /* pass error to parent */
2411 mutex_enter(&spawn_data->sed_mtx_child);
2412 KASSERT(!spawn_data->sed_child_ready);
2413 spawn_data->sed_error = error;
2414 spawn_data->sed_child_ready = true;
2415 cv_signal(&spawn_data->sed_cv_child_ready);
2416 mutex_exit(&spawn_data->sed_mtx_child);
2417 } else {
2418 rw_exit(&exec_lock);
2419 }
2420
2421 /* release our refcount on the data */
2422 spawn_exec_data_release(spawn_data);
2423
2424 /* done, exit */
2425 mutex_enter(p->p_lock);
2426 /*
2427 * Posix explicitly asks for an exit code of 127 if we report
2428 * errors from the child process - so, unfortunately, there
2429 * is no way to report a more exact error code.
2430 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2431 * flag bit in the attrp argument to posix_spawn(2), see above.
2432 */
2433 exit1(l, 127, 0);
2434 }
2435
2436 static __inline char **
posix_spawn_fae_path(struct posix_spawn_file_actions_entry * fae)2437 posix_spawn_fae_path(struct posix_spawn_file_actions_entry *fae)
2438 {
2439 switch (fae->fae_action) {
2440 case FAE_OPEN:
2441 return &fae->fae_path;
2442 case FAE_CHDIR:
2443 return &fae->fae_chdir_path;
2444 default:
2445 return NULL;
2446 }
2447 }
2448
2449 void
posix_spawn_fa_free(struct posix_spawn_file_actions * fa,size_t len)2450 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2451 {
2452
2453 for (size_t i = 0; i < len; i++) {
2454 char **pathp = posix_spawn_fae_path(&fa->fae[i]);
2455 if (pathp)
2456 kmem_strfree(*pathp);
2457 }
2458 if (fa->len > 0)
2459 kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2460 kmem_free(fa, sizeof(*fa));
2461 }
2462
2463 static int
posix_spawn_fa_alloc(struct posix_spawn_file_actions ** fap,const struct posix_spawn_file_actions * ufa,rlim_t lim)2464 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2465 const struct posix_spawn_file_actions *ufa, rlim_t lim)
2466 {
2467 struct posix_spawn_file_actions *fa;
2468 struct posix_spawn_file_actions_entry *fae;
2469 char *pbuf = NULL;
2470 int error;
2471 size_t i = 0;
2472
2473 fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2474 error = copyin(ufa, fa, sizeof(*fa));
2475 if (error || fa->len == 0) {
2476 kmem_free(fa, sizeof(*fa));
2477 return error; /* 0 if not an error, and len == 0 */
2478 }
2479
2480 if (fa->len > lim) {
2481 kmem_free(fa, sizeof(*fa));
2482 return SET_ERROR(EINVAL);
2483 }
2484
2485 fa->size = fa->len;
2486 size_t fal = fa->len * sizeof(*fae);
2487 fae = fa->fae;
2488 fa->fae = kmem_alloc(fal, KM_SLEEP);
2489 error = copyin(fae, fa->fae, fal);
2490 if (error)
2491 goto out;
2492
2493 pbuf = PNBUF_GET();
2494 for (; i < fa->len; i++) {
2495 char **pathp = posix_spawn_fae_path(&fa->fae[i]);
2496 if (pathp == NULL)
2497 continue;
2498 error = copyinstr(*pathp, pbuf, MAXPATHLEN, &fal);
2499 if (error)
2500 goto out;
2501 *pathp = kmem_alloc(fal, KM_SLEEP);
2502 memcpy(*pathp, pbuf, fal);
2503 }
2504 PNBUF_PUT(pbuf);
2505
2506 *fap = fa;
2507 return 0;
2508 out:
2509 if (pbuf)
2510 PNBUF_PUT(pbuf);
2511 posix_spawn_fa_free(fa, i);
2512 return error;
2513 }
2514
2515 /*
2516 * N.B. increments nprocs upon success. Callers need to drop nprocs if
2517 * they fail for some other reason.
2518 */
2519 int
check_posix_spawn(struct lwp * l1)2520 check_posix_spawn(struct lwp *l1)
2521 {
2522 int error, tnprocs, count;
2523 uid_t uid;
2524 struct proc *p1;
2525
2526 p1 = l1->l_proc;
2527 uid = kauth_cred_getuid(l1->l_cred);
2528 tnprocs = atomic_inc_uint_nv(&nprocs);
2529
2530 /*
2531 * Although process entries are dynamically created, we still keep
2532 * a global limit on the maximum number we will create.
2533 */
2534 if (__predict_false(tnprocs >= maxproc))
2535 error = -1;
2536 else
2537 error = kauth_authorize_process(l1->l_cred,
2538 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2539
2540 if (error) {
2541 atomic_dec_uint(&nprocs);
2542 return SET_ERROR(EAGAIN);
2543 }
2544
2545 /*
2546 * Enforce limits.
2547 */
2548 count = chgproccnt(uid, 1);
2549 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2550 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2551 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2552 __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2553 (void)chgproccnt(uid, -1);
2554 atomic_dec_uint(&nprocs);
2555 return SET_ERROR(EAGAIN);
2556 }
2557
2558 return 0;
2559 }
2560
2561 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)2562 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2563 struct posix_spawn_file_actions *fa,
2564 struct posix_spawnattr *sa,
2565 char *const *argv, char *const *envp,
2566 execve_fetch_element_t fetch)
2567 {
2568
2569 struct proc *p1, *p2;
2570 struct lwp *l2;
2571 int error;
2572 struct spawn_exec_data *spawn_data;
2573 vaddr_t uaddr = 0;
2574 pid_t pid;
2575 bool have_exec_lock = false;
2576
2577 p1 = l1->l_proc;
2578
2579 /* Allocate and init spawn_data */
2580 spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2581 spawn_data->sed_refcnt = 1; /* only parent so far */
2582 cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2583 mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2584 mutex_enter(&spawn_data->sed_mtx_child);
2585
2586 /*
2587 * Do the first part of the exec now, collect state
2588 * in spawn_data.
2589 */
2590 error = execve_loadvm(l1, true, path, -1, argv,
2591 envp, fetch, &spawn_data->sed_exec);
2592 if (error == EJUSTRETURN)
2593 error = 0;
2594 else if (error)
2595 goto error_exit;
2596
2597 have_exec_lock = true;
2598
2599 /*
2600 * Allocate virtual address space for the U-area now, while it
2601 * is still easy to abort the fork operation if we're out of
2602 * kernel virtual address space.
2603 */
2604 uaddr = uvm_uarea_alloc();
2605 if (__predict_false(uaddr == 0)) {
2606 error = SET_ERROR(ENOMEM);
2607 goto error_exit;
2608 }
2609
2610 /*
2611 * Allocate new proc. Borrow proc0 vmspace for it, we will
2612 * replace it with its own before returning to userland
2613 * in the child.
2614 */
2615 p2 = proc_alloc();
2616 if (p2 == NULL) {
2617 /* We were unable to allocate a process ID. */
2618 error = SET_ERROR(EAGAIN);
2619 goto error_exit;
2620 }
2621
2622 /*
2623 * This is a point of no return, we will have to go through
2624 * the child proc to properly clean it up past this point.
2625 */
2626 pid = p2->p_pid;
2627
2628 /*
2629 * Make a proc table entry for the new process.
2630 * Start by zeroing the section of proc that is zero-initialized,
2631 * then copy the section that is copied directly from the parent.
2632 */
2633 memset(&p2->p_startzero, 0,
2634 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2635 memcpy(&p2->p_startcopy, &p1->p_startcopy,
2636 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2637
2638 /*
2639 * Allocate an empty user vmspace for the new process now.
2640 * The min/max and topdown parameters given here are just placeholders,
2641 * the right values will be assigned in uvmspace_exec().
2642 */
2643 p2->p_vmspace = uvmspace_alloc(exec_vm_minaddr(VM_MIN_ADDRESS),
2644 VM_MAXUSER_ADDRESS, true);
2645
2646 TAILQ_INIT(&p2->p_sigpend.sp_info);
2647
2648 LIST_INIT(&p2->p_lwps);
2649 LIST_INIT(&p2->p_sigwaiters);
2650
2651 /*
2652 * Duplicate sub-structures as needed.
2653 * Increase reference counts on shared objects.
2654 * Inherit flags we want to keep. The flags related to SIGCHLD
2655 * handling are important in order to keep a consistent behaviour
2656 * for the child after the fork. If we are a 32-bit process, the
2657 * child will be too.
2658 */
2659 p2->p_flag =
2660 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2661 p2->p_emul = p1->p_emul;
2662 p2->p_execsw = p1->p_execsw;
2663
2664 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2665 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2666 rw_init(&p2->p_reflock);
2667 cv_init(&p2->p_waitcv, "wait");
2668 cv_init(&p2->p_lwpcv, "lwpwait");
2669
2670 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2671
2672 kauth_proc_fork(p1, p2);
2673
2674 p2->p_raslist = NULL;
2675 p2->p_fd = fd_copy();
2676
2677 /* XXX racy */
2678 p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2679
2680 p2->p_cwdi = cwdinit();
2681
2682 /*
2683 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2684 * we just need increase pl_refcnt.
2685 */
2686 if (!p1->p_limit->pl_writeable) {
2687 lim_addref(p1->p_limit);
2688 p2->p_limit = p1->p_limit;
2689 } else {
2690 p2->p_limit = lim_copy(p1->p_limit);
2691 }
2692
2693 p2->p_lflag = 0;
2694 l1->l_vforkwaiting = false;
2695 p2->p_sflag = 0;
2696 p2->p_slflag = 0;
2697 p2->p_pptr = p1;
2698 p2->p_ppid = p1->p_pid;
2699 LIST_INIT(&p2->p_children);
2700
2701 p2->p_aio = NULL;
2702
2703 #ifdef KTRACE
2704 /*
2705 * Copy traceflag and tracefile if enabled.
2706 * If not inherited, these were zeroed above.
2707 */
2708 if (p1->p_traceflag & KTRFAC_INHERIT) {
2709 mutex_enter(&ktrace_lock);
2710 p2->p_traceflag = p1->p_traceflag;
2711 if ((p2->p_tracep = p1->p_tracep) != NULL)
2712 ktradref(p2);
2713 mutex_exit(&ktrace_lock);
2714 }
2715 #endif
2716
2717 /*
2718 * Create signal actions for the child process.
2719 */
2720 p2->p_sigacts = sigactsinit(p1, 0);
2721 mutex_enter(p1->p_lock);
2722 p2->p_sflag |=
2723 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2724 sched_proc_fork(p1, p2);
2725 mutex_exit(p1->p_lock);
2726
2727 p2->p_stflag = p1->p_stflag;
2728
2729 /*
2730 * p_stats.
2731 * Copy parts of p_stats, and zero out the rest.
2732 */
2733 p2->p_stats = pstatscopy(p1->p_stats);
2734
2735 /* copy over machdep flags to the new proc */
2736 cpu_proc_fork(p1, p2);
2737
2738 /*
2739 * Prepare remaining parts of spawn data
2740 */
2741 spawn_data->sed_actions = fa;
2742 spawn_data->sed_attrs = sa;
2743
2744 spawn_data->sed_parent = p1;
2745
2746 /* create LWP */
2747 lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2748 &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
2749 l2->l_ctxlink = NULL; /* reset ucontext link */
2750
2751 /*
2752 * Copy the credential so other references don't see our changes.
2753 * Test to see if this is necessary first, since in the common case
2754 * we won't need a private reference.
2755 */
2756 if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2757 kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2758 l2->l_cred = kauth_cred_copy(l2->l_cred);
2759 kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2760 kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2761 }
2762
2763 /* Update the master credentials. */
2764 if (l2->l_cred != p2->p_cred) {
2765 kauth_cred_t ocred;
2766 mutex_enter(p2->p_lock);
2767 ocred = p2->p_cred;
2768 p2->p_cred = kauth_cred_hold(l2->l_cred);
2769 mutex_exit(p2->p_lock);
2770 kauth_cred_free(ocred);
2771 }
2772
2773 *child_ok = true;
2774 spawn_data->sed_refcnt = 2; /* child gets it as well */
2775 #if 0
2776 l2->l_nopreempt = 1; /* start it non-preemptable */
2777 #endif
2778
2779 /*
2780 * It's now safe for the scheduler and other processes to see the
2781 * child process.
2782 */
2783 mutex_enter(&proc_lock);
2784
2785 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2786 p2->p_lflag |= PL_CONTROLT;
2787
2788 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2789 p2->p_exitsig = SIGCHLD; /* signal for parent on exit */
2790
2791 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) ==
2792 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2793 proc_changeparent(p2, p1->p_pptr);
2794 SET(p2->p_slflag, PSL_TRACEDCHILD);
2795 }
2796
2797 p2->p_oppid = p1->p_pid; /* Remember the original parent id. */
2798
2799 LIST_INSERT_AFTER(p1, p2, p_pglist);
2800 LIST_INSERT_HEAD(&allproc, p2, p_list);
2801
2802 p2->p_trace_enabled = trace_is_enabled(p2);
2803 #ifdef __HAVE_SYSCALL_INTERN
2804 (*p2->p_emul->e_syscall_intern)(p2);
2805 #endif
2806
2807 /*
2808 * Make child runnable, set start time, and add to run queue except
2809 * if the parent requested the child to start in SSTOP state.
2810 */
2811 mutex_enter(p2->p_lock);
2812
2813 getmicrotime(&p2->p_stats->p_start);
2814
2815 lwp_lock(l2);
2816 KASSERT(p2->p_nrlwps == 1);
2817 KASSERT(l2->l_stat == LSIDL);
2818 p2->p_nrlwps = 1;
2819 p2->p_stat = SACTIVE;
2820 setrunnable(l2);
2821 /* LWP now unlocked */
2822
2823 mutex_exit(p2->p_lock);
2824 mutex_exit(&proc_lock);
2825
2826 while (!spawn_data->sed_child_ready) {
2827 cv_wait(&spawn_data->sed_cv_child_ready,
2828 &spawn_data->sed_mtx_child);
2829 }
2830 error = spawn_data->sed_error;
2831 mutex_exit(&spawn_data->sed_mtx_child);
2832 spawn_exec_data_release(spawn_data);
2833
2834 rw_exit(&p1->p_reflock);
2835 rw_exit(&exec_lock);
2836 have_exec_lock = false;
2837
2838 *pid_res = pid;
2839
2840 if (error)
2841 return error;
2842
2843 if (p1->p_slflag & PSL_TRACED) {
2844 /* Paranoid check */
2845 mutex_enter(&proc_lock);
2846 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) !=
2847 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2848 mutex_exit(&proc_lock);
2849 return 0;
2850 }
2851
2852 mutex_enter(p1->p_lock);
2853 eventswitch(TRAP_CHLD, PTRACE_POSIX_SPAWN, pid);
2854 }
2855 return 0;
2856
2857 error_exit:
2858 if (have_exec_lock) {
2859 execve_free_data(&spawn_data->sed_exec);
2860 rw_exit(&p1->p_reflock);
2861 rw_exit(&exec_lock);
2862 }
2863 mutex_exit(&spawn_data->sed_mtx_child);
2864 spawn_exec_data_release(spawn_data);
2865 if (uaddr != 0)
2866 uvm_uarea_free(uaddr);
2867
2868 return error;
2869 }
2870
2871 int
sys_posix_spawn(struct lwp * l1,const struct sys_posix_spawn_args * uap,register_t * retval)2872 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2873 register_t *retval)
2874 {
2875 /* {
2876 syscallarg(pid_t *) pid;
2877 syscallarg(const char *) path;
2878 syscallarg(const struct posix_spawn_file_actions *) file_actions;
2879 syscallarg(const struct posix_spawnattr *) attrp;
2880 syscallarg(char *const *) argv;
2881 syscallarg(char *const *) envp;
2882 } */
2883
2884 int error;
2885 struct posix_spawn_file_actions *fa = NULL;
2886 struct posix_spawnattr *sa = NULL;
2887 pid_t pid;
2888 bool child_ok = false;
2889 rlim_t max_fileactions;
2890 proc_t *p = l1->l_proc;
2891
2892 /* check_posix_spawn() increments nprocs for us. */
2893 error = check_posix_spawn(l1);
2894 if (error) {
2895 *retval = error;
2896 return 0;
2897 }
2898
2899 /* copy in file_actions struct */
2900 if (SCARG(uap, file_actions) != NULL) {
2901 max_fileactions = 2 * uimin(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2902 maxfiles);
2903 error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2904 max_fileactions);
2905 if (error)
2906 goto error_exit;
2907 }
2908
2909 /* copyin posix_spawnattr struct */
2910 if (SCARG(uap, attrp) != NULL) {
2911 sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2912 error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2913 if (error)
2914 goto error_exit;
2915 }
2916
2917 /*
2918 * Do the spawn
2919 */
2920 error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2921 SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2922 if (error)
2923 goto error_exit;
2924
2925 if (error == 0 && SCARG(uap, pid) != NULL)
2926 error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2927
2928 *retval = error;
2929 return 0;
2930
2931 error_exit:
2932 if (!child_ok) {
2933 (void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2934 atomic_dec_uint(&nprocs);
2935
2936 if (sa)
2937 kmem_free(sa, sizeof(*sa));
2938 if (fa)
2939 posix_spawn_fa_free(fa, fa->len);
2940 }
2941
2942 *retval = error;
2943 return 0;
2944 }
2945
2946 void
exec_free_emul_arg(struct exec_package * epp)2947 exec_free_emul_arg(struct exec_package *epp)
2948 {
2949 if (epp->ep_emul_arg_free != NULL) {
2950 KASSERT(epp->ep_emul_arg != NULL);
2951 (*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2952 epp->ep_emul_arg_free = NULL;
2953 epp->ep_emul_arg = NULL;
2954 } else {
2955 KASSERT(epp->ep_emul_arg == NULL);
2956 }
2957 }
2958
2959 #ifdef DEBUG_EXEC
2960 static void
dump_vmcmds(const struct exec_package * const epp,size_t x,int error)2961 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2962 {
2963 struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2964 size_t j;
2965
2966 if (error == 0)
2967 DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2968 else
2969 DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2970 epp->ep_vmcmds.evs_used, error));
2971
2972 for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2973 DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2974 PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2975 PRIxVSIZE" prot=0%o flags=%d\n", j,
2976 vp[j].ev_proc == vmcmd_map_pagedvn ?
2977 "pagedvn" :
2978 vp[j].ev_proc == vmcmd_map_readvn ?
2979 "readvn" :
2980 vp[j].ev_proc == vmcmd_map_zero ?
2981 "zero" : "*unknown*",
2982 vp[j].ev_addr, vp[j].ev_len,
2983 vp[j].ev_offset, vp[j].ev_prot,
2984 vp[j].ev_flags));
2985 if (error != 0 && j == x)
2986 DPRINTF((" ^--- failed\n"));
2987 }
2988 }
2989 #endif
2990