1 /*
2 * Copyright (c) 1993, David Greenman
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD: src/sys/kern/kern_exec.c,v 1.107.2.15 2002/07/30 15:40:46 nectar Exp $
27 */
28
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysmsg.h>
32 #include <sys/kernel.h>
33 #include <sys/mount.h>
34 #include <sys/filedesc.h>
35 #include <sys/fcntl.h>
36 #include <sys/acct.h>
37 #include <sys/exec.h>
38 #include <sys/imgact.h>
39 #include <sys/imgact_elf.h>
40 #include <sys/kern_syscall.h>
41 #include <sys/wait.h>
42 #include <sys/malloc.h>
43 #include <sys/proc.h>
44 #include <sys/caps.h>
45 #include <sys/ktrace.h>
46 #include <sys/signalvar.h>
47 #include <sys/pioctl.h>
48 #include <sys/nlookup.h>
49 #include <sys/sysent.h>
50 #include <sys/shm.h>
51 #include <sys/sysctl.h>
52 #include <sys/vnode.h>
53 #include <sys/vmmeter.h>
54 #include <sys/libkern.h>
55
56 #include <cpu/lwbuf.h>
57
58 #include <vm/vm.h>
59 #include <vm/vm_param.h>
60 #include <sys/lock.h>
61 #include <vm/pmap.h>
62 #include <vm/vm_page.h>
63 #include <vm/vm_map.h>
64 #include <vm/vm_kern.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_object.h>
67 #include <vm/vnode_pager.h>
68 #include <vm/vm_pager.h>
69
70 #include <sys/reg.h>
71
72 #include <sys/objcache.h>
73 #include <sys/refcount.h>
74 #include <sys/thread2.h>
75 #include <vm/vm_page2.h>
76
77 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
78 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
79
80 enum exec_path_segflg {
81 PATH_SYSSPACE,
82 PATH_USERSPACE,
83 };
84
85 static register_t *exec_copyout_strings(struct image_params *);
86 static int exec_copyin_args(struct image_args *, char *,
87 enum exec_path_segflg, char **, char **);
88 static void exec_free_args(struct image_args *);
89 static void print_execve_args(struct image_args *args);
90
91 /* XXX This should be vm_size_t. */
92 __read_mostly static u_long ps_strings = PS_STRINGS;
93 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
94
95 /* XXX This should be vm_size_t. */
96 __read_mostly static u_long usrstack = USRSTACK;
97 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
98
99 __read_mostly u_long ps_arg_cache_limit = PAGE_SIZE / 16;
100 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
101 &ps_arg_cache_limit, 0, "");
102
103 __read_mostly int ps_argsopen = 1;
104 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
105
106 __read_mostly static int ktrace_suid = 0;
107 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
108
109 __read_mostly static int debug_execve_args = 0;
110 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
111 0, "");
112
113 /*
114 * Exec arguments object cache
115 */
116 __read_mostly static struct objcache *exec_objcache;
117
118 static
119 void
exec_objcache_init(void * arg __unused)120 exec_objcache_init(void *arg __unused)
121 {
122 int cluster_limit;
123 size_t limsize;
124
125 /*
126 * Maximum number of concurrent execs. This can be limiting on
127 * systems with a lot of cpu cores but it also eats a significant
128 * amount of memory.
129 */
130 cluster_limit = (ncpus < 16) ? 16 : ncpus;
131 limsize = kmem_lim_size();
132 if (limsize > 7 * 1024)
133 cluster_limit *= 2;
134 if (limsize > 15 * 1024)
135 cluster_limit *= 2;
136
137 exec_objcache = objcache_create_mbacked(
138 M_EXECARGS, PATH_MAX + ARG_MAX,
139 cluster_limit, 8,
140 NULL, NULL, NULL);
141 }
142 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
143
144 /*
145 * stackgap_random specifies if the stackgap should have a random size added
146 * to it. It must be a power of 2. If non-zero, the stack gap will be
147 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
148 */
149 __read_mostly static int stackgap_random = 1024;
150
151 static int
sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)152 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
153 {
154 int error, new_val;
155 new_val = stackgap_random;
156 error = sysctl_handle_int(oidp, &new_val, 0, req);
157 if (error != 0 || req->newptr == NULL)
158 return (error);
159 if (new_val > 0 && ((new_val > 16 * PAGE_SIZE) || !powerof2(new_val)))
160 return (EINVAL);
161 stackgap_random = new_val;
162
163 return(0);
164 }
165
166 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT,
167 0, 0, sysctl_kern_stackgap, "I",
168 "Max random stack gap (power of 2), static gap if negative");
169
170 static void
print_execve_args(struct image_args * args)171 print_execve_args(struct image_args *args)
172 {
173 char *cp;
174 int ndx;
175
176 cp = args->begin_argv;
177 for (ndx = 0; ndx < args->argc; ndx++) {
178 kprintf("\targv[%d]: %s\n", ndx, cp);
179 while (*cp++ != '\0');
180 }
181 for (ndx = 0; ndx < args->envc; ndx++) {
182 kprintf("\tenvv[%d]: %s\n", ndx, cp);
183 while (*cp++ != '\0');
184 }
185 }
186
187 /*
188 * Each of the items is a pointer to a `const struct execsw', hence the
189 * double pointer here.
190 */
191 __read_mostly static const struct execsw **execsw;
192
193 /*
194 * Replace current vmspace with a new binary.
195 * Returns 0 on success, > 0 on recoverable error (use as errno).
196 * Returns -1 on lethal error which demands killing of the current
197 * process!
198 */
199 int
kern_execve(struct nlookupdata * nd,struct file * fp,char fileflags,struct image_args * args)200 kern_execve(struct nlookupdata *nd, struct file *fp, char fileflags,
201 struct image_args *args)
202 {
203 static const char *proctitle = "(execve)";
204 register_t *stack_base;
205 struct thread *td = curthread;
206 struct lwp *lp = td->td_lwp;
207 struct proc *p = td->td_proc;
208 struct vnode *ovp;
209 struct pargs *pa;
210 struct sigacts *ops;
211 struct sigacts *nps;
212 struct image_params image_params, *imgp;
213 struct filedesc *fds;
214 struct nchandle *nch;
215 struct nlookupdata nd_interpreter;
216 struct vattr_lite lva;
217 int error, len, i;
218 int (*img_first) (struct image_params *);
219
220 if (debug_execve_args) {
221 kprintf("%s()\n", __func__);
222 print_execve_args(args);
223 }
224
225 KKASSERT(p);
226 lwkt_gettoken(&p->p_token);
227 imgp = &image_params;
228
229 /*
230 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
231 * no modifications to the process at all until we get there.
232 *
233 * Note that multiple threads may be trying to exec at the same
234 * time. exec_new_vmspace() handles that too.
235 */
236
237 /*
238 * Initialize part of the common data
239 */
240 imgp->proc = p;
241 imgp->args = args;
242 imgp->lvap = &lva;
243 imgp->entry_addr = 0;
244 imgp->resident = 0;
245 imgp->vmspace_destroyed = 0;
246 imgp->interpreted = 0;
247 imgp->interpreter_name[0] = 0;
248 imgp->auxargs = NULL;
249 imgp->vp = NULL;
250 imgp->firstpage = NULL;
251 imgp->ps_strings = 0;
252 imgp->execpath = imgp->freepath = NULL;
253 imgp->execpathp = 0;
254 imgp->image_header = NULL;
255
256 interpret:
257
258 if (nd) {
259 /*
260 * Translate the file name to a vnode. Unlock the cache
261 * entry to improve parallelism for programs exec'd in
262 * parallel.
263 */
264 nch = &nd->nl_nch;
265 nd->nl_flags |= NLC_SHAREDLOCK;
266 if ((error = nlookup(nd)) != 0)
267 goto failed;
268
269 error = cache_vget(nch, nd->nl_cred, LK_SHARED, &imgp->vp);
270 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
271 nd->nl_flags &= ~NLC_NCPISLOCKED;
272 cache_unlock(nch);
273 } else {
274 nch = &fp->f_nchandle;
275 imgp->vp = fp->f_data;
276 error = vget(imgp->vp, LK_SHARED);
277 }
278 if (error) {
279 imgp->vp = NULL;
280 goto failed;
281 }
282
283 /*
284 * Check file permissions (also 'opens' file).
285 * Include also the top level mount in the check.
286 */
287 error = exec_check_permissions(imgp, nch->mount);
288 if (error) {
289 vn_unlock(imgp->vp);
290 goto failed;
291 }
292
293 error = exec_map_first_page(imgp);
294 vn_unlock(imgp->vp);
295 if (error)
296 goto failed;
297
298 imgp->proc->p_osrel = 0;
299
300 if (debug_execve_args && imgp->interpreted) {
301 kprintf(" target is interpreted -- recursive pass\n");
302 kprintf(" interpreter: %s\n", imgp->interpreter_name);
303 print_execve_args(args);
304 }
305
306 /*
307 * If the current process has a special image activator it
308 * wants to try first, call it. For example, emulating shell
309 * scripts differently.
310 */
311 error = -1;
312 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
313 error = img_first(imgp);
314
315 /*
316 * If the vnode has a registered vmspace, exec the vmspace
317 */
318 if (error == -1 && imgp->vp->v_resident)
319 error = exec_resident_imgact(imgp);
320
321 /*
322 * Loop through the list of image activators, calling each one.
323 * An activator returns -1 if there is no match, 0 on success,
324 * and an error otherwise.
325 */
326 for (i = 0; error == -1 && execsw[i]; ++i) {
327 if (execsw[i]->ex_imgact == NULL ||
328 execsw[i]->ex_imgact == img_first) {
329 continue;
330 }
331 error = (*execsw[i]->ex_imgact)(imgp);
332 }
333
334 if (error) {
335 if (error == -1)
336 error = ENOEXEC;
337 goto failed;
338 }
339
340 /*
341 * Special interpreter operation, cleanup and loop up to try to
342 * activate the interpreter.
343 */
344 if (imgp->interpreted) {
345 exec_unmap_first_page(imgp);
346 vrele(imgp->vp);
347 imgp->vp = NULL;
348
349 nd = &nd_interpreter;
350 error = nlookup_init(nd, imgp->interpreter_name,
351 UIO_SYSSPACE, NLC_FOLLOW);
352 if (error)
353 goto failed;
354
355 if (fp && (fileflags & UF_EXCLOSE)) {
356 /*
357 * Fexecve'ing an interpreted file opened with
358 * O_CLOEXEC flag, return ENOENT.
359 */
360 error = ENOENT;
361 goto failed;
362 }
363
364 goto interpret;
365 }
366
367 /*
368 * Do the best to calculate the full path to the image file
369 */
370 if (imgp->auxargs != NULL &&
371 ((args->fname != NULL && args->fname[0] == '/') ||
372 vn_fullpath(imgp->proc, imgp->vp, &imgp->execpath,
373 &imgp->freepath, 0) != 0))
374 {
375 imgp->execpath = args->fname;
376 }
377
378 /*
379 * Copy out strings (args and env) and initialize stack base
380 */
381 stack_base = exec_copyout_strings(imgp);
382 p->p_vmspace->vm_minsaddr = (char *)stack_base;
383
384 /*
385 * If custom stack fixup routine present for this process
386 * let it do the stack setup. If we are running a resident
387 * image there is no auxinfo or other image activator context
388 * so don't try to add fixups to the stack.
389 *
390 * Else stuff argument count as first item on stack
391 */
392 if (p->p_sysent->sv_fixup && imgp->resident == 0)
393 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
394 else
395 suword64(--stack_base, imgp->args->argc);
396
397 /*
398 * For security and other reasons, the file descriptor table cannot
399 * be shared after an exec.
400 */
401 if (p->p_fd->fd_refcnt > 1) {
402 if ((error = fdcopy(p, &fds)) != 0)
403 goto failed;
404
405 fdfree(p, fds);
406 }
407
408 /*
409 * For security and other reasons, signal handlers cannot
410 * be shared after an exec. The new proces gets a copy of the old
411 * handlers. In execsigs(), the new process will have its signals
412 * reset.
413 */
414 ops = p->p_sigacts;
415 if (ops->ps_refcnt > 1) {
416 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
417 bcopy(ops, nps, sizeof(*nps));
418 refcount_init(&nps->ps_refcnt, 1);
419 p->p_sigacts = nps;
420 if (refcount_release(&ops->ps_refcnt)) {
421 kfree(ops, M_SUBPROC);
422 ops = NULL;
423 }
424 }
425
426 /*
427 * Clean up shared pages, the new program will allocate fresh
428 * copies as needed. This is also for security purposes and
429 * to ensure (for example) that things like sys_lpmap->blockallsigs
430 * state is properly reset on exec.
431 */
432 lwp_userunmap(lp);
433 proc_userunmap(p);
434
435 /*
436 * For security and other reasons virtual kernels cannot be
437 * inherited by an exec. This also allows a virtual kernel
438 * to fork/exec unrelated applications.
439 */
440 if (p->p_vkernel)
441 vkernel_exit(p);
442
443 /* Stop profiling */
444 stopprofclock(p);
445
446 /* close files on exec */
447 fdcloseexec(p);
448
449 /* reset caught signals */
450 execsigs(p);
451
452 /* name this process */
453 if (nch->ncp) {
454 len = min(nch->ncp->nc_nlen, MAXCOMLEN);
455 bcopy(nch->ncp->nc_name, p->p_comm, len);
456 } else {
457 len = sizeof(proctitle) - 1;
458 bcopy(proctitle, p->p_comm, len);
459 }
460 p->p_comm[len] = 0;
461 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
462
463 /*
464 * mark as execed, wakeup the process that vforked (if any) and tell
465 * it that it now has its own resources back
466 *
467 * We are using the P_PPWAIT as an interlock so an atomic op is
468 * necessary to synchronize with the parent's cpu.
469 */
470 p->p_flags |= P_EXEC;
471 if (p->p_pptr && (p->p_flags & P_PPWAIT)) {
472 if (p->p_pptr->p_upmap)
473 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1);
474 atomic_clear_int(&p->p_flags, P_PPWAIT);
475 wakeup(p->p_pptr);
476 }
477
478 /*
479 * Implement image setuid/setgid.
480 *
481 * Don't honor setuid/setgid if the filesystem prohibits it or if
482 * the process is being traced.
483 */
484 if ((((lva.va_mode & VSUID) && p->p_ucred->cr_uid != lva.va_uid) ||
485 ((lva.va_mode & VSGID) && p->p_ucred->cr_gid != lva.va_gid)) &&
486 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
487 (p->p_flags & P_TRACED) == 0) {
488 /*
489 * Turn off syscall tracing for set-id programs, except for
490 * root. Record any set-id flags first to make sure that
491 * we do not regain any tracing during a possible block.
492 */
493 setsugid();
494 if (p->p_tracenode && ktrace_suid == 0 &&
495 caps_priv_check_td(td, SYSCAP_RESTRICTEDROOT) != 0)
496 {
497 ktrdestroy(&p->p_tracenode);
498 p->p_traceflag = 0;
499 }
500
501 /* Clear any PROC_PDEATHSIG_CTL setting */
502 p->p_deathsig = 0;
503
504 /* Close any file descriptors 0..2 that reference procfs */
505 setugidsafety(p);
506 /* Make sure file descriptors 0..2 are in use. */
507 error = fdcheckstd(lp);
508 if (error != 0)
509 goto failed;
510
511 /*
512 * Set the new credentials.
513 */
514 cratom_proc(p);
515 if (lva.va_mode & VSUID)
516 change_euid(lva.va_uid);
517 if (lva.va_mode & VSGID)
518 p->p_ucred->cr_gid = lva.va_gid;
519
520 /* Clear local varsym variables */
521 varsymset_clean(&p->p_varsymset);
522 } else {
523 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
524 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
525 p->p_flags &= ~P_SUGID;
526 }
527
528 /*
529 * Implement correct POSIX saved-id behavior.
530 */
531 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
532 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
533 cratom_proc(p);
534 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
535 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
536 }
537
538 /*
539 * Store the vp for use in procfs. Be sure to keep p_textvp
540 * consistent if we block during the switch-over.
541 */
542 ovp = p->p_textvp;
543 vref(imgp->vp); /* ref new vp */
544 p->p_textvp = imgp->vp;
545 if (ovp) /* release old vp */
546 vrele(ovp);
547
548 /* Release old namecache handle to text file */
549 if (p->p_textnch.ncp)
550 cache_drop(&p->p_textnch);
551 if (nch->mount)
552 cache_copy(nch, &p->p_textnch);
553
554 /*
555 * Adjust capabilities in ucred if necessasry
556 */
557 caps_exec(p);
558
559 /*
560 * Notify others that we exec'd, and clear the P_INEXEC flag
561 * as we're now a bona fide freshly-execed process.
562 */
563 KNOTE(&p->p_klist, NOTE_EXEC);
564 p->p_flags &= ~P_INEXEC;
565 if (p->p_stops)
566 wakeup(&p->p_stype);
567
568 /*
569 * If tracing the process, trap to debugger so breakpoints
570 * can be set before the program executes.
571 */
572 STOPEVENT(p, S_EXEC, 0);
573
574 if (p->p_flags & P_TRACED)
575 ksignal(p, SIGTRAP);
576
577 /* clear "fork but no exec" flag, as we _are_ execing */
578 p->p_acflag &= ~AFORK;
579
580 /* Set values passed into the program in registers. */
581 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
582 imgp->ps_strings);
583
584 /* Set the access time on the vnode */
585 vn_mark_atime(imgp->vp, td);
586
587 /*
588 * Free any previous argument cache
589 */
590 pa = p->p_args;
591 p->p_args = NULL;
592 if (pa && refcount_release(&pa->ar_ref)) {
593 kfree(pa, M_PARGS);
594 pa = NULL;
595 }
596
597 /*
598 * Cache arguments if they fit inside our allowance
599 */
600 i = imgp->args->begin_envv - imgp->args->begin_argv;
601 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
602 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
603 refcount_init(&pa->ar_ref, 1);
604 pa->ar_length = i;
605 bcopy(imgp->args->begin_argv, pa->ar_args, i);
606 KKASSERT(p->p_args == NULL);
607 p->p_args = pa;
608 }
609
610 failed:
611
612 /*
613 * free various allocated resources
614 */
615 if (imgp->firstpage)
616 exec_unmap_first_page(imgp);
617 if (imgp->vp)
618 vrele(imgp->vp);
619 if (imgp->freepath)
620 kfree(imgp->freepath, M_TEMP);
621 if (nd == &nd_interpreter)
622 nlookup_done(nd);
623
624 if (error == 0) {
625 ++mycpu->gd_cnt.v_exec;
626 lwkt_reltoken(&p->p_token);
627 return (0);
628 }
629
630 /*
631 * we're done here, clear P_INEXEC if we were the ones that
632 * set it. Otherwise if vmspace_destroyed is still set we
633 * raced another thread and that thread is responsible for
634 * clearing it.
635 */
636 if (imgp->vmspace_destroyed & 2) {
637 p->p_flags &= ~P_INEXEC;
638 if (p->p_stops)
639 wakeup(&p->p_stype);
640 }
641 lwkt_reltoken(&p->p_token);
642 if (imgp->vmspace_destroyed) {
643 /*
644 * Sorry, no more process anymore. exit gracefully.
645 * However we can't die right here, because our
646 * caller might have to clean up, so indicate a
647 * lethal error by returning -1.
648 */
649 return (-1);
650 } else {
651 return (error);
652 }
653 }
654
655 /*
656 * execve() system call.
657 */
658 int
sys_execve(struct sysmsg * sysmsg,const struct execve_args * uap)659 sys_execve(struct sysmsg *sysmsg, const struct execve_args *uap)
660 {
661 struct nlookupdata nd;
662 struct image_args args;
663 int error;
664
665 /*
666 * General exec ok?
667 */
668 if (caps_priv_check_self(SYSCAP_NOEXEC))
669 return EACCES;
670
671 /*
672 * Exec path
673 */
674 bzero(&args, sizeof(args));
675
676 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
677 if (error == 0) {
678 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
679 uap->argv, uap->envv);
680 }
681 if (error == 0)
682 error = kern_execve(&nd, NULL, 0, &args);
683 nlookup_done(&nd);
684 exec_free_args(&args);
685
686 if (error < 0) {
687 /* We hit a lethal error condition. Let's die now. */
688 exit1(W_EXITCODE(0, SIGABRT));
689 /* NOTREACHED */
690 }
691
692 /*
693 * The syscall result is returned in registers to the new program.
694 * Linux will register %edx as an atexit function and we must be
695 * sure to set it to 0. XXX
696 */
697 if (error == 0)
698 sysmsg->sysmsg_result64 = 0;
699
700 return (error);
701 }
702
703 /*
704 * fexecve() system call.
705 */
706 int
sys_fexecve(struct sysmsg * sysmsg,const struct fexecve_args * uap)707 sys_fexecve(struct sysmsg *sysmsg, const struct fexecve_args *uap)
708 {
709 struct image_args args;
710 struct thread *td = curthread;
711 struct file *fp;
712 char fileflags;
713 char fname[32]; /* "/dev/fd/xxx" */
714 int error;
715
716 /*
717 * General exec ok?
718 */
719 if (caps_priv_check_self(SYSCAP_NOEXEC))
720 return EACCES;
721
722 /*
723 * Exec descriptor
724 */
725 if ((error = holdvnode2(td, uap->fd, &fp, &fileflags)) != 0)
726 return (error);
727
728 /*
729 * Require a descriptor opened only with O_RDONLY or O_EXEC.
730 * XXX: missing O_EXEC support
731 */
732 if ((fp->f_flag & FWRITE) != 0 || (fp->f_flag & FREAD) == 0) {
733 error = EBADF;
734 goto done;
735 }
736
737 /*
738 * The 'fname' argument is required when executing an
739 * interpreted program because the interpreter must know
740 * the script path. Supply it with '/dev/fd/xxx'.
741 */
742 ksnprintf(fname, sizeof(fname), "/dev/fd/%d", uap->fd);
743 bzero(&args, sizeof(args));
744 error = exec_copyin_args(&args, fname, PATH_SYSSPACE,
745 uap->argv, uap->envv);
746 if (error == 0)
747 error = kern_execve(NULL, fp, fileflags, &args);
748 exec_free_args(&args);
749
750 if (error < 0) {
751 /* We hit a lethal error condition. Let's die now. */
752 exit1(W_EXITCODE(0, SIGABRT));
753 /* NOTREACHED */
754 }
755
756 /*
757 * The syscall result is returned in registers to the new program.
758 * Linux will register %edx as an atexit function and we must be
759 * sure to set it to 0. XXX
760 */
761 if (error == 0)
762 sysmsg->sysmsg_result64 = 0;
763
764 done:
765 fdrop(fp);
766 return (error);
767 }
768
769 int
exec_map_page(struct image_params * imgp,vm_pindex_t pageno,struct lwbuf ** plwb,const char ** pdata)770 exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
771 struct lwbuf **plwb, const char **pdata)
772 {
773 int rv;
774 vm_page_t ma;
775 vm_page_t m;
776 vm_object_t object;
777
778 /*
779 * The file has to be mappable.
780 */
781 if ((object = imgp->vp->v_object) == NULL)
782 return (EIO);
783
784 if (pageno >= object->size)
785 return (EIO);
786
787 /*
788 * Shortcut using shared locks, improve concurrent execs.
789 */
790 vm_object_hold_shared(object);
791 m = vm_page_lookup(object, pageno);
792 if (m) {
793 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) {
794 vm_page_hold(m);
795 vm_page_sleep_busy(m, FALSE, "execpg");
796 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL &&
797 m->object == object && m->pindex == pageno) {
798 vm_object_drop(object);
799 goto done;
800 }
801 vm_page_unhold(m);
802 }
803 }
804 vm_object_drop(object);
805
806 /*
807 * Do it the hard way
808 */
809 vm_object_hold(object);
810 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
811 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
812 ma = m;
813
814 /*
815 * get_pages unbusies all the requested pages except the
816 * primary page (at index 0 in this case). The primary
817 * page may have been wired during the pagein (e.g. by
818 * the buffer cache) so vnode_pager_freepage() must be
819 * used to properly release it.
820 */
821 rv = vm_pager_get_page(object, pageno, &ma, 1);
822 m = vm_page_lookup(object, pageno);
823
824 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
825 if (m) {
826 vm_page_protect(m, VM_PROT_NONE);
827 vnode_pager_freepage(m);
828 }
829 vm_object_drop(object);
830 return EIO;
831 }
832 }
833 vm_page_hold(m);
834 vm_page_wakeup(m); /* unbusy the page */
835 vm_object_drop(object);
836
837 done:
838 *plwb = lwbuf_alloc(m, *plwb);
839 *pdata = (void *)lwbuf_kva(*plwb);
840
841 return (0);
842 }
843
844 /*
845 * Map the first page of an executable image.
846 *
847 * NOTE: If the mapping fails we have to NULL-out firstpage which may
848 * still be pointing to our supplied lwp structure.
849 */
850 int
exec_map_first_page(struct image_params * imgp)851 exec_map_first_page(struct image_params *imgp)
852 {
853 int err;
854
855 if (imgp->firstpage)
856 exec_unmap_first_page(imgp);
857
858 imgp->firstpage = &imgp->firstpage_cache;
859 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
860
861 if (err) {
862 imgp->firstpage = NULL;
863 return err;
864 }
865
866 return 0;
867 }
868
869 void
exec_unmap_page(struct lwbuf * lwb)870 exec_unmap_page(struct lwbuf *lwb)
871 {
872 vm_page_t m;
873
874 crit_enter();
875 if (lwb != NULL) {
876 m = lwbuf_page(lwb);
877 lwbuf_free(lwb);
878 vm_page_unhold(m);
879 }
880 crit_exit();
881 }
882
883 void
exec_unmap_first_page(struct image_params * imgp)884 exec_unmap_first_page(struct image_params *imgp)
885 {
886 exec_unmap_page(imgp->firstpage);
887 imgp->firstpage = NULL;
888 imgp->image_header = NULL;
889 }
890
891 /*
892 * Destroy old address space, and allocate a new stack
893 * The new stack is only SGROWSIZ large because it is grown
894 * automatically in trap.c.
895 *
896 * This is the point of no return.
897 */
898 int
exec_new_vmspace(struct image_params * imgp,struct vmspace * vmcopy)899 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
900 {
901 struct vmspace *vmspace = imgp->proc->p_vmspace;
902 vm_offset_t stack_addr = USRSTACK - maxssiz;
903 struct lwp *lp;
904 struct proc *p;
905 vm_map_t map;
906 int error;
907
908 /*
909 * Indicate that we cannot gracefully error out any more, kill
910 * any other threads present, and set P_INEXEC to indicate that
911 * we are now messing with the process structure proper.
912 *
913 * If killalllwps() races return an error which coupled with
914 * vmspace_destroyed will cause us to exit. This is what we
915 * want since another thread is patiently waiting for us to exit
916 * in that case.
917 */
918 lp = curthread->td_lwp;
919 p = lp->lwp_proc;
920 imgp->vmspace_destroyed = 1;
921
922 if (curthread->td_proc->p_nthreads > 1) {
923 error = killalllwps(1);
924 if (error)
925 return (error);
926 }
927 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
928 p->p_flags |= P_INEXEC;
929
930 /*
931 * Tell procfs to release its hold on the process. It
932 * will return EAGAIN.
933 */
934 if (p->p_stops)
935 wakeup(&p->p_stype);
936
937 /*
938 * After setting P_INEXEC wait for any remaining references to
939 * the process (p) to go away.
940 *
941 * In particular, a vfork/exec sequence will replace p->p_vmspace
942 * and we must interlock anyone trying to access the space (aka
943 * procfs or sys_process.c calling procfs_domem()).
944 *
945 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us.
946 */
947 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0));
948
949 /*
950 * Blow away entire process VM, if address space not shared,
951 * otherwise, create a new VM space so that other threads are
952 * not disrupted. If we are execing a resident vmspace we
953 * create a duplicate of it and remap the stack.
954 */
955 map = &vmspace->vm_map;
956 if (vmcopy) {
957 vmspace_exec(imgp->proc, vmcopy);
958 vmspace = imgp->proc->p_vmspace;
959 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
960 map = &vmspace->vm_map;
961 } else if (vmspace_getrefs(vmspace) == 1) {
962 shmexit(vmspace);
963 pmap_remove_pages(vmspace_pmap(vmspace),
964 0, VM_MAX_USER_ADDRESS);
965 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
966 } else {
967 vmspace_exec(imgp->proc, NULL);
968 vmspace = imgp->proc->p_vmspace;
969 map = &vmspace->vm_map;
970 }
971
972 /*
973 * Really make sure lwp-specific and process-specific mappings
974 * are gone.
975 *
976 * Once we've done that, and because we are the only LWP left, with
977 * no TID-dependent mappings, we can reset the TID to 1 (the RB tree
978 * will remain consistent since it has only one entry). This way
979 * the exec'd program gets a nice deterministic tid of 1.
980 */
981 lwp_userunmap(lp);
982 proc_userunmap(p);
983 lp->lwp_tid = 1;
984 p->p_lasttid = 1;
985
986 /*
987 * Allocate a new stack, generally make the stack non-executable
988 * but allow the program to adjust that (the program may desire to
989 * use areas of the stack for executable code).
990 */
991 error = vm_map_stack(&vmspace->vm_map, &stack_addr, (vm_size_t)maxssiz,
992 0,
993 VM_PROT_READ|VM_PROT_WRITE,
994 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE,
995 0);
996 if (error)
997 return (error);
998
999 /*
1000 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1001 * VM_STACK case, but they are still used to monitor the size of the
1002 * process stack so we can check the stack rlimit.
1003 */
1004 vmspace->vm_ssize = sgrowsiz; /* in bytes */
1005 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
1006
1007 return(0);
1008 }
1009
1010 /*
1011 * Copy out argument and environment strings from the old process
1012 * address space into the temporary string buffer.
1013 */
1014 static int
exec_copyin_args(struct image_args * args,char * fname,enum exec_path_segflg segflg,char ** argv,char ** envv)1015 exec_copyin_args(struct image_args *args, char *fname,
1016 enum exec_path_segflg segflg, char **argv, char **envv)
1017 {
1018 char *argp, *envp;
1019 int error = 0;
1020 size_t length;
1021
1022 args->buf = objcache_get(exec_objcache, M_WAITOK);
1023 if (args->buf == NULL)
1024 return (ENOMEM);
1025
1026 args->begin_argv = args->buf;
1027 args->endp = args->begin_argv;
1028 args->space = ARG_MAX;
1029
1030 args->fname = args->buf + ARG_MAX;
1031
1032 /*
1033 * Copy the file name.
1034 */
1035 if (segflg == PATH_SYSSPACE)
1036 error = copystr(fname, args->fname, PATH_MAX, &length);
1037 else
1038 error = copyinstr(fname, args->fname, PATH_MAX, &length);
1039 if (error)
1040 return (error);
1041
1042 /*
1043 * Extract argument strings. argv may not be NULL. The argv
1044 * array is terminated by a NULL entry. We special-case the
1045 * situation where argv[0] is NULL by passing { filename, NULL }
1046 * to the new program to guarentee that the interpreter knows what
1047 * file to open in case we exec an interpreted file. Note that
1048 * a NULL argv[0] terminates the argv[] array.
1049 *
1050 * XXX the special-casing of argv[0] is historical and needs to be
1051 * revisited.
1052 */
1053 if (argv == NULL)
1054 error = EFAULT;
1055 if (error == 0) {
1056 while ((argp = (caddr_t)(intptr_t)
1057 fuword64((uintptr_t *)argv++)) != NULL) {
1058 if (argp == (caddr_t)-1) {
1059 error = EFAULT;
1060 break;
1061 }
1062 error = copyinstr(argp, args->endp,
1063 args->space, &length);
1064 if (error) {
1065 if (error == ENAMETOOLONG)
1066 error = E2BIG;
1067 break;
1068 }
1069 args->space -= length;
1070 args->endp += length;
1071 args->argc++;
1072 }
1073 if (args->argc == 0 && error == 0) {
1074 length = strlen(args->fname) + 1;
1075 if (length > args->space) {
1076 error = E2BIG;
1077 } else {
1078 bcopy(args->fname, args->endp, length);
1079 args->space -= length;
1080 args->endp += length;
1081 args->argc++;
1082 }
1083 }
1084 }
1085
1086 args->begin_envv = args->endp;
1087
1088 /*
1089 * extract environment strings. envv may be NULL.
1090 */
1091 if (envv && error == 0) {
1092 while ((envp = (caddr_t)(intptr_t)
1093 fuword64((uintptr_t *)envv++))) {
1094 if (envp == (caddr_t) -1) {
1095 error = EFAULT;
1096 break;
1097 }
1098 error = copyinstr(envp, args->endp,
1099 args->space, &length);
1100 if (error) {
1101 if (error == ENAMETOOLONG)
1102 error = E2BIG;
1103 break;
1104 }
1105 args->space -= length;
1106 args->endp += length;
1107 args->envc++;
1108 }
1109 }
1110
1111 return (error);
1112 }
1113
1114 static void
exec_free_args(struct image_args * args)1115 exec_free_args(struct image_args *args)
1116 {
1117 if (args->buf) {
1118 objcache_put(exec_objcache, args->buf);
1119 args->buf = NULL;
1120 }
1121 }
1122
1123 /*
1124 * Copy strings out to the new process address space, constructing
1125 * new arg and env vector tables. Return a pointer to the base
1126 * so that it can be used as the initial stack pointer.
1127 *
1128 * The format is, roughly:
1129 *
1130 * [argv[]] <-- vectp
1131 * [envp[]]
1132 * [ELF_Auxargs]
1133 *
1134 * [args & env] <-- destp
1135 * [sgap]
1136 * [SPARE_USRSPACE]
1137 * [execpath]
1138 * [szsigcode] RO|NX
1139 * [ps_strings] RO|NX Top of user stack
1140 *
1141 */
1142 static register_t *
exec_copyout_strings(struct image_params * imgp)1143 exec_copyout_strings(struct image_params *imgp)
1144 {
1145 int argc, envc, sgap;
1146 int gap;
1147 int argsenvspace;
1148 char **vectp;
1149 char *stringp, *destp, *szsigbase;
1150 register_t *stack_base;
1151 struct ps_strings *arginfo;
1152 size_t execpath_len;
1153 int szsigcode;
1154
1155 /*
1156 * Calculate string base and vector table pointers.
1157 * Also deal with signal trampoline code for this exec type.
1158 */
1159 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1160 execpath_len = strlen(imgp->execpath) + 1;
1161 else
1162 execpath_len = 0;
1163 arginfo = (struct ps_strings *)PS_STRINGS;
1164 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
1165
1166 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *));
1167 gap = stackgap_random;
1168 cpu_ccfence();
1169 if (gap != 0) {
1170 if (gap < 0)
1171 sgap = ALIGN(-gap);
1172 else
1173 sgap = ALIGN(karc4random() & (gap - 1));
1174 } else {
1175 sgap = 0;
1176 }
1177
1178 /*
1179 * Calculate destp, which points to [args & env] and above.
1180 */
1181 szsigbase = (char *)(intptr_t)
1182 trunc_page64((intptr_t)arginfo - szsigcode);
1183 szsigbase -= SZSIGCODE_EXTRA_BYTES;
1184 destp = szsigbase -
1185 roundup(execpath_len, sizeof(char *)) -
1186 SPARE_USRSPACE -
1187 sgap -
1188 argsenvspace;
1189
1190 /*
1191 * install sigcode
1192 */
1193 if (szsigcode)
1194 copyout(imgp->proc->p_sysent->sv_sigcode, szsigbase, szsigcode);
1195
1196 /*
1197 * Copy the image path for the rtld
1198 */
1199 if (execpath_len) {
1200 imgp->execpathp = (uintptr_t)szsigbase -
1201 roundup(execpath_len, sizeof(char *));
1202 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
1203 }
1204
1205 /*
1206 * Calculate base for argv[], envp[], and ELF_Auxargs.
1207 */
1208 vectp = (char **)destp - (AT_COUNT * 2);
1209 vectp -= imgp->args->argc + imgp->args->envc + 2;
1210
1211 stack_base = (register_t *)vectp;
1212
1213 stringp = imgp->args->begin_argv;
1214 argc = imgp->args->argc;
1215 envc = imgp->args->envc;
1216
1217 /*
1218 * Copy out strings - arguments and environment (at destp)
1219 */
1220 copyout(stringp, destp, ARG_MAX - imgp->args->space);
1221
1222 /*
1223 * Fill in "ps_strings" struct for ps, w, etc.
1224 */
1225 suword64((void *)&arginfo->ps_argvstr, (uint64_t)(intptr_t)vectp);
1226 suword32((void *)&arginfo->ps_nargvstr, argc);
1227
1228 /*
1229 * Fill in argument portion of vector table.
1230 */
1231 for (; argc > 0; --argc) {
1232 suword64((void *)vectp++, (uintptr_t)destp);
1233 while (*stringp++ != 0)
1234 destp++;
1235 destp++;
1236 }
1237
1238 /* a null vector table pointer separates the argp's from the envp's */
1239 suword64((void *)vectp++, 0);
1240
1241 suword64((void *)&arginfo->ps_envstr, (uintptr_t)vectp);
1242 suword32((void *)&arginfo->ps_nenvstr, envc);
1243
1244 /*
1245 * Fill in environment portion of vector table.
1246 */
1247 for (; envc > 0; --envc) {
1248 suword64((void *)vectp++, (uintptr_t)destp);
1249 while (*stringp++ != 0)
1250 destp++;
1251 destp++;
1252 }
1253
1254 /* end of vector table is a null pointer */
1255 suword64((void *)vectp, 0);
1256
1257 /*
1258 * Make the signal trampoline executable and read-only.
1259 */
1260 vm_map_protect(&imgp->proc->p_vmspace->vm_map,
1261 (vm_offset_t)szsigbase,
1262 (vm_offset_t)szsigbase + PAGE_SIZE,
1263 VM_PROT_READ|VM_PROT_EXECUTE, FALSE);
1264
1265 return (stack_base);
1266 }
1267
1268 /*
1269 * Check permissions of file to execute.
1270 * Return 0 for success or error code on failure.
1271 */
1272 int
exec_check_permissions(struct image_params * imgp,struct mount * topmnt)1273 exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
1274 {
1275 struct proc *p = imgp->proc;
1276 struct vnode *vp = imgp->vp;
1277 struct vattr_lite *lvap = imgp->lvap;
1278 int error;
1279
1280 /* Get file attributes */
1281 error = VOP_GETATTR_LITE(vp, lvap);
1282 if (error)
1283 return (error);
1284
1285 /*
1286 * 1) Check if file execution is disabled for the filesystem that this
1287 * file resides on.
1288 * 2) Insure that at least one execute bit is on - otherwise root
1289 * will always succeed, and we don't want to happen unless the
1290 * file really is executable.
1291 * 3) Insure that the file is a regular file.
1292 */
1293 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1294 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
1295 ((lvap->va_mode & 0111) == 0) ||
1296 (lvap->va_type != VREG)) {
1297 return (EACCES);
1298 }
1299
1300 /*
1301 * Capability restrictions on suid or sgid exec?
1302 */
1303 if ((lvap->va_mode & VSUID) && caps_priv_check_self(SYSCAP_NOEXEC_SUID))
1304 return EACCES;
1305 if ((lvap->va_mode & VSGID) && caps_priv_check_self(SYSCAP_NOEXEC_SGID))
1306 return EACCES;
1307
1308 /*
1309 * Zero length files can't be exec'd
1310 */
1311 if (lvap->va_size == 0)
1312 return (ENOEXEC);
1313
1314 /*
1315 * Check for execute permission to file based on current credentials.
1316 */
1317 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1318 if (error)
1319 return (error);
1320
1321 /*
1322 * Check number of open-for-writes on the file and deny execution
1323 * if there are any.
1324 */
1325 if (vp->v_writecount)
1326 return (ETXTBSY);
1327
1328 /*
1329 * Call filesystem specific open routine, which allows us to read,
1330 * write, and mmap the file. Without the VOP_OPEN we can only
1331 * stat the file.
1332 */
1333 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1334 if (error)
1335 return (error);
1336
1337 return (0);
1338 }
1339
1340 /*
1341 * Exec handler registration
1342 */
1343 int
exec_register(const struct execsw * execsw_arg)1344 exec_register(const struct execsw *execsw_arg)
1345 {
1346 const struct execsw **es, **xs, **newexecsw;
1347 int count = 2; /* New slot and trailing NULL */
1348
1349 if (execsw)
1350 for (es = execsw; *es; es++)
1351 count++;
1352 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1353 xs = newexecsw;
1354 if (execsw)
1355 for (es = execsw; *es; es++)
1356 *xs++ = *es;
1357 *xs++ = execsw_arg;
1358 *xs = NULL;
1359 if (execsw)
1360 kfree(execsw, M_TEMP);
1361 execsw = newexecsw;
1362 return 0;
1363 }
1364
1365 int
exec_unregister(const struct execsw * execsw_arg)1366 exec_unregister(const struct execsw *execsw_arg)
1367 {
1368 const struct execsw **es, **xs, **newexecsw;
1369 int count = 1;
1370
1371 if (execsw == NULL)
1372 panic("unregister with no handlers left?");
1373
1374 for (es = execsw; *es; es++) {
1375 if (*es == execsw_arg)
1376 break;
1377 }
1378 if (*es == NULL)
1379 return ENOENT;
1380 for (es = execsw; *es; es++)
1381 if (*es != execsw_arg)
1382 count++;
1383 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1384 xs = newexecsw;
1385 for (es = execsw; *es; es++)
1386 if (*es != execsw_arg)
1387 *xs++ = *es;
1388 *xs = NULL;
1389 if (execsw)
1390 kfree(execsw, M_TEMP);
1391 execsw = newexecsw;
1392 return 0;
1393 }
1394