kernel - Implement static stackgap feature in addition to random stackgap
[dragonfly.git] / sys / kern / kern_exec.c
CommitLineData
984263bc
MD
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/sysproto.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>
2bd9d75c 40#include <sys/kern_syscall.h>
984263bc
MD
41#include <sys/wait.h>
42#include <sys/malloc.h>
43#include <sys/proc.h>
895c1f85 44#include <sys/priv.h>
29f58392 45#include <sys/ktrace.h>
984263bc
MD
46#include <sys/signalvar.h>
47#include <sys/pioctl.h>
fad57d0e 48#include <sys/nlookup.h>
984263bc
MD
49#include <sys/sysent.h>
50#include <sys/shm.h>
51#include <sys/sysctl.h>
52#include <sys/vnode.h>
a6f89c72 53#include <sys/vmmeter.h>
17a7ca1f 54#include <sys/libkern.h>
984263bc 55
5c5185ae
SG
56#include <cpu/lwbuf.h>
57
984263bc
MD
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>
a55afca2 67#include <vm/vnode_pager.h>
984263bc
MD
68#include <vm/vm_pager.h>
69
70#include <sys/user.h>
527fddf7 71#include <sys/reg.h>
984263bc 72
19bfc8ab 73#include <sys/refcount.h>
5fd012e0 74#include <sys/thread2.h>
684a93c4 75#include <sys/mplock2.h>
5fd012e0 76
984263bc 77MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
18b3f457 78MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
984263bc 79
402ed7e1 80static register_t *exec_copyout_strings (struct image_params *);
984263bc
MD
81
82/* XXX This should be vm_size_t. */
83static u_long ps_strings = PS_STRINGS;
84SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
85
86/* XXX This should be vm_size_t. */
87static u_long usrstack = USRSTACK;
88SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
89
90u_long ps_arg_cache_limit = PAGE_SIZE / 16;
91SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
92 &ps_arg_cache_limit, 0, "");
93
94int ps_argsopen = 1;
95SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
96
5c627295
MD
97static int ktrace_suid = 0;
98SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
99
2bd9d75c
DRJ
100void print_execve_args(struct image_args *args);
101int debug_execve_args = 0;
102SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
103 0, "");
104
17a7ca1f 105/*
18b3f457
MD
106 * Exec arguments object cache
107 */
108static struct objcache *exec_objcache;
109
110static
111void
112exec_objcache_init(void *arg __unused)
113{
0aa16b5d 114 int cluster_limit;
1f29f2c5
MD
115 size_t limsize;
116
117 /*
118 * Maximum number of concurrent execs. This can be limiting on
119 * systems with a lot of cpu cores but it also eats a significant
120 * amount of memory.
121 */
e93da8d4 122 cluster_limit = (ncpus < 16) ? 16 : ncpus;
1f29f2c5
MD
123 limsize = kmem_lim_size();
124 if (limsize > 7 * 1024)
125 cluster_limit *= 2;
126 if (limsize > 15 * 1024)
127 cluster_limit *= 2;
0aa16b5d 128
18b3f457
MD
129 exec_objcache = objcache_create_mbacked(
130 M_EXECARGS, PATH_MAX + ARG_MAX,
765b1ae0 131 &cluster_limit, 8,
18b3f457
MD
132 NULL, NULL, NULL);
133}
134SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
135
136/*
17a7ca1f
MD
137 * stackgap_random specifies if the stackgap should have a random size added
138 * to it. It must be a power of 2. If non-zero, the stack gap will be
0ced1954 139 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
17a7ca1f
MD
140 */
141static int stackgap_random = 1024;
142static int
143sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
144{
145 int error, new_val;
146 new_val = stackgap_random;
147 error = sysctl_handle_int(oidp, &new_val, 0, req);
148 if (error != 0 || req->newptr == NULL)
149 return (error);
d0e92318 150 if (new_val > 0 && ((new_val > 16 * PAGE_SIZE) || !powerof2(new_val)))
17a7ca1f
MD
151 return (EINVAL);
152 stackgap_random = new_val;
153
154 return(0);
155}
156
d0e92318
MD
157SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT,
158 0, 0, sysctl_kern_stackgap, "IU",
159 "Max random stack gap (power of 2), static gap if negative");
17a7ca1f 160
2bd9d75c
DRJ
161void
162print_execve_args(struct image_args *args)
163{
164 char *cp;
165 int ndx;
166
167 cp = args->begin_argv;
168 for (ndx = 0; ndx < args->argc; ndx++) {
6ea70f76 169 kprintf("\targv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
170 while (*cp++ != '\0');
171 }
172 for (ndx = 0; ndx < args->envc; ndx++) {
6ea70f76 173 kprintf("\tenvv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
174 while (*cp++ != '\0');
175 }
176}
177
984263bc
MD
178/*
179 * Each of the items is a pointer to a `const struct execsw', hence the
180 * double pointer here.
181 */
182static const struct execsw **execsw;
183
5417cd57
SS
184/*
185 * Replace current vmspace with a new binary.
186 * Returns 0 on success, > 0 on recoverable error (use as errno).
90947cb1 187 * Returns -1 on lethal error which demands killing of the current
5417cd57
SS
188 * process!
189 */
984263bc 190int
fad57d0e 191kern_execve(struct nlookupdata *nd, struct image_args *args)
984263bc 192{
dadab5e9 193 struct thread *td = curthread;
08f2f1bb 194 struct lwp *lp = td->td_lwp;
dadab5e9 195 struct proc *p = td->td_proc;
a2ee730d 196 struct vnode *ovp;
984263bc 197 register_t *stack_base;
19bfc8ab 198 struct pargs *pa;
6fa9e71a
MD
199 struct sigacts *ops;
200 struct sigacts *nps;
984263bc
MD
201 int error, len, i;
202 struct image_params image_params, *imgp;
203 struct vattr attr;
402ed7e1 204 int (*img_first) (struct image_params *);
984263bc 205
2bd9d75c 206 if (debug_execve_args) {
6ea70f76 207 kprintf("%s()\n", __func__);
2bd9d75c
DRJ
208 print_execve_args(args);
209 }
210
dadab5e9 211 KKASSERT(p);
19bfc8ab 212 lwkt_gettoken(&p->p_token);
984263bc
MD
213 imgp = &image_params;
214
215 /*
dc52e1cc
MD
216 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
217 * no modifications to the process at all until we get there.
218 *
219 * Note that multiple threads may be trying to exec at the same
220 * time. exec_new_vmspace() handles that too.
984263bc 221 */
984263bc
MD
222
223 /*
224 * Initialize part of the common data
225 */
226 imgp->proc = p;
2bd9d75c 227 imgp->args = args;
984263bc 228 imgp->attr = &attr;
984263bc 229 imgp->entry_addr = 0;
29802dbb 230 imgp->resident = 0;
984263bc
MD
231 imgp->vmspace_destroyed = 0;
232 imgp->interpreted = 0;
2abfe22f 233 imgp->interpreter_name[0] = 0;
984263bc
MD
234 imgp->auxargs = NULL;
235 imgp->vp = NULL;
236 imgp->firstpage = NULL;
237 imgp->ps_strings = 0;
adc42cf3
JM
238 imgp->execpath = imgp->freepath = NULL;
239 imgp->execpathp = 0;
18f40545 240 imgp->image_header = NULL;
2bd9d75c
DRJ
241
242interpret:
984263bc
MD
243
244 /*
fad57d0e
MD
245 * Translate the file name to a vnode. Unlock the cache entry to
246 * improve parallelism for programs exec'd in parallel.
984263bc 247 */
fad57d0e
MD
248 if ((error = nlookup(nd)) != 0)
249 goto exec_fail;
28623bf9 250 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
fad57d0e
MD
251 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
252 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 253 cache_unlock(&nd->nl_nch);
fad57d0e 254 if (error)
984263bc 255 goto exec_fail;
984263bc
MD
256
257 /*
246693ac
AHJ
258 * Check file permissions (also 'opens' file).
259 * Include also the top level mount in the check.
984263bc 260 */
246693ac 261 error = exec_check_permissions(imgp, nd->nl_nch.mount);
984263bc 262 if (error) {
a11aaa81 263 vn_unlock(imgp->vp);
984263bc
MD
264 goto exec_fail_dealloc;
265 }
266
267 error = exec_map_first_page(imgp);
a11aaa81 268 vn_unlock(imgp->vp);
984263bc
MD
269 if (error)
270 goto exec_fail_dealloc;
271
315b8b8b
JM
272 imgp->proc->p_osrel = 0;
273
2bd9d75c 274 if (debug_execve_args && imgp->interpreted) {
6ea70f76
SW
275 kprintf(" target is interpreted -- recursive pass\n");
276 kprintf(" interpreter: %s\n", imgp->interpreter_name);
2bd9d75c
DRJ
277 print_execve_args(args);
278 }
279
984263bc
MD
280 /*
281 * If the current process has a special image activator it
282 * wants to try first, call it. For example, emulating shell
283 * scripts differently.
284 */
285 error = -1;
286 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
287 error = img_first(imgp);
288
289 /*
29802dbb
MD
290 * If the vnode has a registered vmspace, exec the vmspace
291 */
292 if (error == -1 && imgp->vp->v_resident) {
293 error = exec_resident_imgact(imgp);
294 }
295
296 /*
984263bc
MD
297 * Loop through the list of image activators, calling each one.
298 * An activator returns -1 if there is no match, 0 on success,
299 * and an error otherwise.
300 */
301 for (i = 0; error == -1 && execsw[i]; ++i) {
302 if (execsw[i]->ex_imgact == NULL ||
303 execsw[i]->ex_imgact == img_first) {
304 continue;
305 }
306 error = (*execsw[i]->ex_imgact)(imgp);
307 }
308
309 if (error) {
310 if (error == -1)
311 error = ENOEXEC;
312 goto exec_fail_dealloc;
313 }
314
315 /*
316 * Special interpreter operation, cleanup and loop up to try to
317 * activate the interpreter.
318 */
319 if (imgp->interpreted) {
320 exec_unmap_first_page(imgp);
fad57d0e
MD
321 nlookup_done(nd);
322 vrele(imgp->vp);
323 imgp->vp = NULL;
324 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
325 NLC_FOLLOW);
326 if (error)
327 goto exec_fail;
984263bc
MD
328 goto interpret;
329 }
330
331 /*
adc42cf3
JM
332 * Do the best to calculate the full path to the image file
333 */
334 if (imgp->auxargs != NULL &&
335 ((args->fname != NULL && args->fname[0] == '/') ||
336 vn_fullpath(imgp->proc,
337 imgp->vp,
338 &imgp->execpath,
339 &imgp->freepath,
340 0) != 0))
341 imgp->execpath = args->fname;
342
343 /*
984263bc
MD
344 * Copy out strings (args and env) and initialize stack base
345 */
346 stack_base = exec_copyout_strings(imgp);
347 p->p_vmspace->vm_minsaddr = (char *)stack_base;
348
349 /*
350 * If custom stack fixup routine present for this process
29802dbb
MD
351 * let it do the stack setup. If we are running a resident
352 * image there is no auxinfo or other image activator context
353 * so don't try to add fixups to the stack.
354 *
984263bc
MD
355 * Else stuff argument count as first item on stack
356 */
29802dbb 357 if (p->p_sysent->sv_fixup && imgp->resident == 0)
984263bc
MD
358 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
359 else
2bd9d75c 360 suword(--stack_base, imgp->args->argc);
984263bc
MD
361
362 /*
363 * For security and other reasons, the file descriptor table cannot
364 * be shared after an exec.
365 */
366 if (p->p_fd->fd_refcnt > 1) {
367 struct filedesc *tmp;
368
2994659f
VS
369 error = fdcopy(p, &tmp);
370 if (error != 0)
371 goto exec_fail;
0a4a9c77 372 fdfree(p, tmp);
984263bc
MD
373 }
374
375 /*
376 * For security and other reasons, signal handlers cannot
377 * be shared after an exec. The new proces gets a copy of the old
378 * handlers. In execsigs(), the new process will have its signals
379 * reset.
380 */
6fa9e71a
MD
381 ops = p->p_sigacts;
382 if (ops->ps_refcnt > 1) {
383 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
384 bcopy(ops, nps, sizeof(*nps));
385 refcount_init(&nps->ps_refcnt, 1);
386 p->p_sigacts = nps;
387 if (refcount_release(&ops->ps_refcnt)) {
388 kfree(ops, M_SUBPROC);
389 ops = NULL;
390 }
984263bc
MD
391 }
392
0daa37a5
MD
393 /*
394 * For security and other reasons virtual kernels cannot be
395 * inherited by an exec. This also allows a virtual kernel
396 * to fork/exec unrelated applications.
397 */
4a22e893
MD
398 if (p->p_vkernel)
399 vkernel_exit(p);
0daa37a5 400
984263bc
MD
401 /* Stop profiling */
402 stopprofclock(p);
403
404 /* close files on exec */
405 fdcloseexec(p);
406
407 /* reset caught signals */
408 execsigs(p);
409
410 /* name this process - nameiexec(p, ndp) */
28623bf9
MD
411 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
412 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
984263bc 413 p->p_comm[len] = 0;
08f2f1bb 414 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
984263bc
MD
415
416 /*
417 * mark as execed, wakeup the process that vforked (if any) and tell
418 * it that it now has its own resources back
419 */
4643740a
MD
420 p->p_flags |= P_EXEC;
421 if (p->p_pptr && (p->p_flags & P_PPWAIT)) {
422 p->p_flags &= ~P_PPWAIT;
984263bc
MD
423 wakeup((caddr_t)p->p_pptr);
424 }
425
426 /*
427 * Implement image setuid/setgid.
428 *
429 * Don't honor setuid/setgid if the filesystem prohibits it or if
430 * the process is being traced.
431 */
432 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
433 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
434 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
4643740a 435 (p->p_flags & P_TRACED) == 0) {
984263bc
MD
436 /*
437 * Turn off syscall tracing for set-id programs, except for
438 * root. Record any set-id flags first to make sure that
439 * we do not regain any tracing during a possible block.
440 */
41c20dac 441 setsugid();
5c627295
MD
442 if (p->p_tracenode && ktrace_suid == 0 &&
443 priv_check(td, PRIV_ROOT) != 0) {
29f58392
MD
444 ktrdestroy(&p->p_tracenode);
445 p->p_traceflag = 0;
984263bc
MD
446 }
447 /* Close any file descriptors 0..2 that reference procfs */
448 setugidsafety(p);
449 /* Make sure file descriptors 0..2 are in use. */
f3a2d8c4 450 error = fdcheckstd(lp);
984263bc
MD
451 if (error != 0)
452 goto exec_fail_dealloc;
453 /*
454 * Set the new credentials.
455 */
e9a372eb 456 cratom(&p->p_ucred);
984263bc 457 if (attr.va_mode & VSUID)
41c20dac 458 change_euid(attr.va_uid);
984263bc
MD
459 if (attr.va_mode & VSGID)
460 p->p_ucred->cr_gid = attr.va_gid;
98a7f915
MD
461
462 /*
463 * Clear local varsym variables
464 */
465 varsymset_clean(&p->p_varsymset);
984263bc 466 } else {
41c20dac
MD
467 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
468 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
4643740a 469 p->p_flags &= ~P_SUGID;
984263bc
MD
470 }
471
472 /*
473 * Implement correct POSIX saved-id behavior.
474 */
115ccd83
MD
475 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
476 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
477 cratom(&p->p_ucred);
478 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
479 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
480 }
984263bc
MD
481
482 /*
a2ee730d
MD
483 * Store the vp for use in procfs. Be sure to keep p_textvp
484 * consistent if we block during the switch-over.
984263bc 485 */
a2ee730d
MD
486 ovp = p->p_textvp;
487 vref(imgp->vp); /* ref new vp */
fad57d0e 488 p->p_textvp = imgp->vp;
a2ee730d
MD
489 if (ovp) /* release old vp */
490 vrele(ovp);
984263bc 491
8ba5f7ef
AH
492 /* Release old namecache handle to text file */
493 if (p->p_textnch.ncp)
494 cache_drop(&p->p_textnch);
495
496 if (nd->nl_nch.mount)
497 cache_copy(&nd->nl_nch, &p->p_textnch);
498
984263bc
MD
499 /*
500 * Notify others that we exec'd, and clear the P_INEXEC flag
501 * as we're now a bona fide freshly-execed process.
502 */
503 KNOTE(&p->p_klist, NOTE_EXEC);
4643740a 504 p->p_flags &= ~P_INEXEC;
984263bc
MD
505
506 /*
507 * If tracing the process, trap to debugger so breakpoints
508 * can be set before the program executes.
509 */
510 STOPEVENT(p, S_EXEC, 0);
511
4643740a 512 if (p->p_flags & P_TRACED)
84204577 513 ksignal(p, SIGTRAP);
984263bc
MD
514
515 /* clear "fork but no exec" flag, as we _are_ execing */
516 p->p_acflag &= ~AFORK;
517
518 /* Set values passed into the program in registers. */
08f2f1bb 519 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
a2ee730d 520 imgp->ps_strings);
984263bc 521
349433c9
MD
522 /* Set the access time on the vnode */
523 vn_mark_atime(imgp->vp, td);
524
19bfc8ab
MD
525 /*
526 * Free any previous argument cache
527 */
528 pa = p->p_args;
984263bc 529 p->p_args = NULL;
19bfc8ab
MD
530 if (pa && refcount_release(&pa->ar_ref)) {
531 kfree(pa, M_PARGS);
532 pa = NULL;
533 }
984263bc 534
19bfc8ab
MD
535 /*
536 * Cache arguments if they fit inside our allowance
537 */
2bd9d75c 538 i = imgp->args->begin_envv - imgp->args->begin_argv;
19bfc8ab
MD
539 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
540 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
541 refcount_init(&pa->ar_ref, 1);
542 pa->ar_length = i;
543 bcopy(imgp->args->begin_argv, pa->ar_args, i);
544 KKASSERT(p->p_args == NULL);
545 p->p_args = pa;
984263bc
MD
546 }
547
548exec_fail_dealloc:
549
550 /*
551 * free various allocated resources
552 */
553 if (imgp->firstpage)
554 exec_unmap_first_page(imgp);
555
984263bc 556 if (imgp->vp) {
984263bc 557 vrele(imgp->vp);
fad57d0e 558 imgp->vp = NULL;
984263bc
MD
559 }
560
bf52a6ac
MD
561 if (imgp->freepath)
562 kfree(imgp->freepath, M_TEMP);
563
a6f89c72
DR
564 if (error == 0) {
565 ++mycpu->gd_cnt.v_exec;
19bfc8ab 566 lwkt_reltoken(&p->p_token);
984263bc 567 return (0);
a6f89c72 568 }
984263bc
MD
569
570exec_fail:
dc52e1cc
MD
571 /*
572 * we're done here, clear P_INEXEC if we were the ones that
573 * set it. Otherwise if vmspace_destroyed is still set we
574 * raced another thread and that thread is responsible for
575 * clearing it.
576 */
577 if (imgp->vmspace_destroyed & 2)
4643740a 578 p->p_flags &= ~P_INEXEC;
19bfc8ab 579 lwkt_reltoken(&p->p_token);
984263bc 580 if (imgp->vmspace_destroyed) {
5417cd57
SS
581 /*
582 * Sorry, no more process anymore. exit gracefully.
583 * However we can't die right here, because our
584 * caller might have to clean up, so indicate a
90947cb1 585 * lethal error by returning -1.
5417cd57
SS
586 */
587 return(-1);
984263bc
MD
588 } else {
589 return(error);
590 }
591}
592
2bd9d75c
DRJ
593/*
594 * execve() system call.
595 */
596int
753fd850 597sys_execve(struct execve_args *uap)
2bd9d75c 598{
fad57d0e 599 struct nlookupdata nd;
2bd9d75c
DRJ
600 struct image_args args;
601 int error;
602
118cec38 603 bzero(&args, sizeof(args));
3919ced0 604
3919ced0 605 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
fad57d0e
MD
606 if (error == 0) {
607 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
608 uap->argv, uap->envv);
609 }
2abfe22f
MD
610 if (error == 0)
611 error = kern_execve(&nd, &args);
fad57d0e 612 nlookup_done(&nd);
2bd9d75c
DRJ
613 exec_free_args(&args);
614
5417cd57 615 if (error < 0) {
90947cb1 616 /* We hit a lethal error condition. Let's die now. */
5417cd57
SS
617 exit1(W_EXITCODE(0, SIGABRT));
618 /* NOTREACHED */
619 }
620
2bd9d75c
DRJ
621 /*
622 * The syscall result is returned in registers to the new program.
623 * Linux will register %edx as an atexit function and we must be
624 * sure to set it to 0. XXX
625 */
626 if (error == 0)
627 uap->sysmsg_result64 = 0;
628
629 return (error);
630}
631
984263bc 632int
7c34d798 633exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
5c5185ae 634 struct lwbuf **plwb, const char **pdata)
984263bc 635{
1b9d3514
MD
636 int rv;
637 vm_page_t ma;
06ecca5a 638 vm_page_t m;
984263bc
MD
639 vm_object_t object;
640
fad57d0e 641 /*
7540ab49 642 * The file has to be mappable.
fad57d0e 643 */
7540ab49 644 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 645 return (EIO);
984263bc 646
7c34d798
SS
647 if (pageno >= object->size)
648 return (EIO);
649
b12defdc 650 vm_object_hold(object);
7c34d798 651 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1b9d3514
MD
652 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
653 ma = m;
984263bc 654
06ecca5a
MD
655 /*
656 * get_pages unbusies all the requested pages except the
a55afca2
MD
657 * primary page (at index 0 in this case). The primary
658 * page may have been wired during the pagein (e.g. by
659 * the buffer cache) so vnode_pager_freepage() must be
660 * used to properly release it.
06ecca5a 661 */
1b9d3514 662 rv = vm_pager_get_page(object, &ma, 1);
7c34d798 663 m = vm_page_lookup(object, pageno);
984263bc 664
06ecca5a
MD
665 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
666 if (m) {
667 vm_page_protect(m, VM_PROT_NONE);
a55afca2 668 vnode_pager_freepage(m);
984263bc 669 }
471bd290 670 vm_object_drop(object);
984263bc
MD
671 return EIO;
672 }
673 }
b12defdc 674 vm_page_hold(m);
06ecca5a 675 vm_page_wakeup(m); /* unbusy the page */
b12defdc 676 vm_object_drop(object);
984263bc 677
7a683a24 678 *plwb = lwbuf_alloc(m, *plwb);
5c5185ae 679 *pdata = (void *)lwbuf_kva(*plwb);
7c34d798
SS
680
681 return (0);
682}
683
7484bdd2
MD
684/*
685 * Map the first page of an executable image.
686 *
687 * NOTE: If the mapping fails we have to NULL-out firstpage which may
688 * still be pointing to our supplied lwp structure.
689 */
7c34d798
SS
690int
691exec_map_first_page(struct image_params *imgp)
692{
693 int err;
694
695 if (imgp->firstpage)
696 exec_unmap_first_page(imgp);
697
7a683a24 698 imgp->firstpage = &imgp->firstpage_cache;
7c34d798
SS
699 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
700
7484bdd2
MD
701 if (err) {
702 imgp->firstpage = NULL;
7c34d798 703 return err;
7484bdd2 704 }
984263bc
MD
705
706 return 0;
707}
708
709void
5c5185ae 710exec_unmap_page(struct lwbuf *lwb)
984263bc 711{
18f40545
MD
712 vm_page_t m;
713
5fd012e0 714 crit_enter();
5c5185ae
SG
715 if (lwb != NULL) {
716 m = lwbuf_page(lwb);
717 lwbuf_free(lwb);
e4ba7818 718 vm_page_unhold(m);
984263bc 719 }
5fd012e0 720 crit_exit();
984263bc
MD
721}
722
7c34d798
SS
723void
724exec_unmap_first_page(struct image_params *imgp)
725{
726 exec_unmap_page(imgp->firstpage);
727 imgp->firstpage = NULL;
728 imgp->image_header = NULL;
729}
730
984263bc
MD
731/*
732 * Destroy old address space, and allocate a new stack
733 * The new stack is only SGROWSIZ large because it is grown
734 * automatically in trap.c.
dc52e1cc
MD
735 *
736 * This is the point of no return.
984263bc
MD
737 */
738int
29802dbb 739exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 740{
984263bc
MD
741 struct vmspace *vmspace = imgp->proc->p_vmspace;
742 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 743 struct proc *p;
29802dbb 744 vm_map_t map;
dc52e1cc 745 int error;
984263bc 746
dc52e1cc
MD
747 /*
748 * Indicate that we cannot gracefully error out any more, kill
749 * any other threads present, and set P_INEXEC to indicate that
750 * we are now messing with the process structure proper.
751 *
752 * If killalllwps() races return an error which coupled with
753 * vmspace_destroyed will cause us to exit. This is what we
754 * want since another thread is patiently waiting for us to exit
755 * in that case.
756 */
757 p = curproc;
984263bc
MD
758 imgp->vmspace_destroyed = 1;
759
dc52e1cc
MD
760 if (curthread->td_proc->p_nthreads > 1) {
761 error = killalllwps(1);
762 if (error)
763 return (error);
764 }
765 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
4643740a 766 p->p_flags |= P_INEXEC;
08f2f1bb
SS
767
768 /*
82354ad8
MD
769 * After setting P_INEXEC wait for any remaining references to
770 * the process (p) to go away.
771 *
772 * In particular, a vfork/exec sequence will replace p->p_vmspace
773 * and we must interlock anyone trying to access the space (aka
774 * procfs or sys_process.c calling procfs_domem()).
775 *
776 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us.
777 */
778 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0));
779
780 /*
984263bc
MD
781 * Blow away entire process VM, if address space not shared,
782 * otherwise, create a new VM space so that other threads are
29802dbb
MD
783 * not disrupted. If we are execing a resident vmspace we
784 * create a duplicate of it and remap the stack.
984263bc 785 */
29802dbb
MD
786 map = &vmspace->vm_map;
787 if (vmcopy) {
788 vmspace_exec(imgp->proc, vmcopy);
789 vmspace = imgp->proc->p_vmspace;
790 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
791 map = &vmspace->vm_map;
a2ee730d 792 } else if (vmspace->vm_sysref.refcnt == 1) {
fef0fdf2 793 shmexit(vmspace);
239b4df9 794 if (vmspace->vm_upcalls)
08f2f1bb 795 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08 796 pmap_remove_pages(vmspace_pmap(vmspace),
b0c15cdf 797 0, VM_MAX_USER_ADDRESS);
88181b08 798 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 799 } else {
29802dbb 800 vmspace_exec(imgp->proc, NULL);
984263bc
MD
801 vmspace = imgp->proc->p_vmspace;
802 map = &vmspace->vm_map;
803 }
804
805 /* Allocate a new stack */
806 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
c809941b 807 0, VM_PROT_ALL, VM_PROT_ALL, 0);
984263bc
MD
808 if (error)
809 return (error);
810
811 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
812 * VM_STACK case, but they are still used to monitor the size of the
813 * process stack so we can check the stack rlimit.
814 */
815 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
816 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
817
818 return(0);
819}
820
821/*
822 * Copy out argument and environment strings from the old process
823 * address space into the temporary string buffer.
824 */
825int
2bd9d75c 826exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 827 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 828{
984263bc 829 char *argp, *envp;
2bd9d75c 830 int error = 0;
984263bc
MD
831 size_t length;
832
18b3f457 833 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
834 if (args->buf == NULL)
835 return (ENOMEM);
836 args->begin_argv = args->buf;
837 args->endp = args->begin_argv;
838 args->space = ARG_MAX;
2bd9d75c
DRJ
839
840 args->fname = args->buf + ARG_MAX;
841
984263bc 842 /*
2bd9d75c 843 * Copy the file name.
984263bc 844 */
2bd9d75c
DRJ
845 if (segflg == PATH_SYSSPACE) {
846 error = copystr(fname, args->fname, PATH_MAX, &length);
847 } else if (segflg == PATH_USERSPACE) {
848 error = copyinstr(fname, args->fname, PATH_MAX, &length);
849 }
984263bc 850
2bd9d75c 851 /*
66be6566
MD
852 * Extract argument strings. argv may not be NULL. The argv
853 * array is terminated by a NULL entry. We special-case the
854 * situation where argv[0] is NULL by passing { filename, NULL }
855 * to the new program to guarentee that the interpreter knows what
856 * file to open in case we exec an interpreted file. Note that
857 * a NULL argv[0] terminates the argv[] array.
858 *
859 * XXX the special-casing of argv[0] is historical and needs to be
860 * revisited.
2bd9d75c 861 */
66be6566
MD
862 if (argv == NULL)
863 error = EFAULT;
864 if (error == 0) {
865 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
866 if (argp == (caddr_t)-1) {
2bd9d75c 867 error = EFAULT;
66be6566 868 break;
2bd9d75c
DRJ
869 }
870 error = copyinstr(argp, args->endp,
d0e92318 871 args->space, &length);
66be6566
MD
872 if (error) {
873 if (error == ENAMETOOLONG)
874 error = E2BIG;
875 break;
876 }
2bd9d75c
DRJ
877 args->space -= length;
878 args->endp += length;
879 args->argc++;
984263bc 880 }
66be6566
MD
881 if (args->argc == 0 && error == 0) {
882 length = strlen(args->fname) + 1;
883 if (length > args->space) {
884 error = E2BIG;
885 } else {
886 bcopy(args->fname, args->endp, length);
887 args->space -= length;
888 args->endp += length;
889 args->argc++;
890 }
891 }
984263bc
MD
892 }
893
2bd9d75c 894 args->begin_envv = args->endp;
984263bc
MD
895
896 /*
66be6566 897 * extract environment strings. envv may be NULL.
984263bc 898 */
2abfe22f 899 if (envv && error == 0) {
984263bc 900 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
901 if (envp == (caddr_t) -1) {
902 error = EFAULT;
66be6566 903 break;
984263bc 904 }
d0e92318
MD
905 error = copyinstr(envp, args->endp,
906 args->space, &length);
66be6566
MD
907 if (error) {
908 if (error == ENAMETOOLONG)
909 error = E2BIG;
910 break;
911 }
2bd9d75c
DRJ
912 args->space -= length;
913 args->endp += length;
914 args->envc++;
984263bc
MD
915 }
916 }
2bd9d75c
DRJ
917 return (error);
918}
919
920void
921exec_free_args(struct image_args *args)
922{
2abfe22f 923 if (args->buf) {
18b3f457 924 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
925 args->buf = NULL;
926 }
984263bc
MD
927}
928
929/*
930 * Copy strings out to the new process address space, constructing
d0e92318
MD
931 * new arg and env vector tables. Return a pointer to the base
932 * so that it can be used as the initial stack pointer.
933 *
934 * The format is, roughly:
935 *
936 * [argv[]] <-- vectp
937 * [envp[]]
938 * [ELF_Auxargs]
939 *
940 * [args & env] <-- destp
941 * [sgap]
942 * [SPARE_USRSPACE]
943 * [execpath]
944 * [szsigcode]
945 * [ps_strings] top of user stack
946 *
984263bc
MD
947 */
948register_t *
2bd9d75c 949exec_copyout_strings(struct image_params *imgp)
984263bc 950{
17a7ca1f 951 int argc, envc, sgap;
d0e92318
MD
952 int gap;
953 int argsenvspace;
984263bc
MD
954 char **vectp;
955 char *stringp, *destp;
956 register_t *stack_base;
957 struct ps_strings *arginfo;
adc42cf3 958 size_t execpath_len;
984263bc
MD
959 int szsigcode;
960
961 /*
962 * Calculate string base and vector table pointers.
963 * Also deal with signal trampoline code for this exec type.
964 */
adc42cf3
JM
965 if (imgp->execpath != NULL && imgp->auxargs != NULL)
966 execpath_len = strlen(imgp->execpath) + 1;
967 else
968 execpath_len = 0;
984263bc
MD
969 arginfo = (struct ps_strings *)PS_STRINGS;
970 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
d0e92318
MD
971
972 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *));
973 gap = stackgap_random;
974 cpu_ccfence();
975 if (gap != 0) {
976 if (gap < 0)
977 sgap = ALIGN(-gap);
978 else
979 sgap = ALIGN(karc4random() & (stackgap_random - 1));
980 } else {
981 sgap = 0;
982 }
983
17a7ca1f 984 if (stackgap_random != 0)
0ced1954 985 sgap = ALIGN(karc4random() & (stackgap_random - 1));
17a7ca1f
MD
986 else
987 sgap = 0;
d0e92318
MD
988
989 /*
990 * Calculate destp, which points to [args & env] and above.
991 */
992 destp = (caddr_t)arginfo - szsigcode -
993 roundup(execpath_len, sizeof(char *)) -
994 SPARE_USRSPACE -
995 sgap -
996 argsenvspace;
984263bc
MD
997
998 /*
999 * install sigcode
1000 */
d0e92318 1001 if (szsigcode) {
984263bc 1002 copyout(imgp->proc->p_sysent->sv_sigcode,
d0e92318
MD
1003 ((caddr_t)arginfo - szsigcode), szsigcode);
1004 }
984263bc
MD
1005
1006 /*
adc42cf3
JM
1007 * Copy the image path for the rtld
1008 */
d0e92318 1009 if (execpath_len) {
adc42cf3
JM
1010 imgp->execpathp = (uintptr_t)arginfo
1011 - szsigcode
d0e92318 1012 - roundup(execpath_len, sizeof(char *));
adc42cf3
JM
1013 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
1014 }
1015
1016 /*
d0e92318 1017 * Calculate base for argv[], envp[], and ELF_Auxargs.
984263bc 1018 */
d0e92318
MD
1019 vectp = (char **)destp - (AT_COUNT * 2);
1020 vectp -= imgp->args->argc + imgp->args->envc + 2;
984263bc 1021
984263bc
MD
1022 stack_base = (register_t *)vectp;
1023
2bd9d75c
DRJ
1024 stringp = imgp->args->begin_argv;
1025 argc = imgp->args->argc;
1026 envc = imgp->args->envc;
984263bc
MD
1027
1028 /*
d0e92318 1029 * Copy out strings - arguments and environment (at destp)
984263bc 1030 */
2bd9d75c 1031 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
1032
1033 /*
1034 * Fill in "ps_strings" struct for ps, w, etc.
1035 */
1036 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
3f938eea 1037 suword32(&arginfo->ps_nargvstr, argc);
984263bc
MD
1038
1039 /*
1040 * Fill in argument portion of vector table.
1041 */
1042 for (; argc > 0; --argc) {
1043 suword(vectp++, (long)(intptr_t)destp);
1044 while (*stringp++ != 0)
1045 destp++;
1046 destp++;
1047 }
1048
7a44d1cb 1049 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
1050 suword(vectp++, 0);
1051
1052 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
3f938eea 1053 suword32(&arginfo->ps_nenvstr, envc);
984263bc
MD
1054
1055 /*
1056 * Fill in environment portion of vector table.
1057 */
1058 for (; envc > 0; --envc) {
1059 suword(vectp++, (long)(intptr_t)destp);
1060 while (*stringp++ != 0)
1061 destp++;
1062 destp++;
1063 }
1064
1065 /* end of vector table is a null pointer */
1066 suword(vectp, 0);
1067
1068 return (stack_base);
1069}
1070
1071/*
1072 * Check permissions of file to execute.
1073 * Return 0 for success or error code on failure.
1074 */
1075int
246693ac 1076exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
984263bc
MD
1077{
1078 struct proc *p = imgp->proc;
1079 struct vnode *vp = imgp->vp;
1080 struct vattr *attr = imgp->attr;
1081 int error;
1082
1083 /* Get file attributes */
87de5057 1084 error = VOP_GETATTR(vp, attr);
984263bc
MD
1085 if (error)
1086 return (error);
1087
1088 /*
1089 * 1) Check if file execution is disabled for the filesystem that this
1090 * file resides on.
1091 * 2) Insure that at least one execute bit is on - otherwise root
1092 * will always succeed, and we don't want to happen unless the
1093 * file really is executable.
1094 * 3) Insure that the file is a regular file.
1095 */
1096 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
246693ac 1097 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
984263bc
MD
1098 ((attr->va_mode & 0111) == 0) ||
1099 (attr->va_type != VREG)) {
1100 return (EACCES);
1101 }
1102
1103 /*
1104 * Zero length files can't be exec'd
1105 */
1106 if (attr->va_size == 0)
1107 return (ENOEXEC);
1108
1109 /*
1110 * Check for execute permission to file based on current credentials.
1111 */
90a83564 1112 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
1113 if (error)
1114 return (error);
1115
1116 /*
1117 * Check number of open-for-writes on the file and deny execution
1118 * if there are any.
1119 */
1120 if (vp->v_writecount)
1121 return (ETXTBSY);
1122
1123 /*
7540ab49
MD
1124 * Call filesystem specific open routine, which allows us to read,
1125 * write, and mmap the file. Without the VOP_OPEN we can only
1126 * stat the file.
984263bc 1127 */
87de5057 1128 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
1129 if (error)
1130 return (error);
1131
1132 return (0);
1133}
1134
1135/*
1136 * Exec handler registration
1137 */
1138int
77153250 1139exec_register(const struct execsw *execsw_arg)
984263bc
MD
1140{
1141 const struct execsw **es, **xs, **newexecsw;
1142 int count = 2; /* New slot and trailing NULL */
1143
1144 if (execsw)
1145 for (es = execsw; *es; es++)
1146 count++;
efda3bd0 1147 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1148 xs = newexecsw;
1149 if (execsw)
1150 for (es = execsw; *es; es++)
1151 *xs++ = *es;
1152 *xs++ = execsw_arg;
1153 *xs = NULL;
1154 if (execsw)
efda3bd0 1155 kfree(execsw, M_TEMP);
984263bc
MD
1156 execsw = newexecsw;
1157 return 0;
1158}
1159
1160int
77153250 1161exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1162{
1163 const struct execsw **es, **xs, **newexecsw;
1164 int count = 1;
1165
1166 if (execsw == NULL)
ae7e74d4 1167 panic("unregister with no handlers left?");
984263bc
MD
1168
1169 for (es = execsw; *es; es++) {
1170 if (*es == execsw_arg)
1171 break;
1172 }
1173 if (*es == NULL)
1174 return ENOENT;
1175 for (es = execsw; *es; es++)
1176 if (*es != execsw_arg)
1177 count++;
efda3bd0 1178 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1179 xs = newexecsw;
1180 for (es = execsw; *es; es++)
1181 if (*es != execsw_arg)
1182 *xs++ = *es;
1183 *xs = NULL;
1184 if (execsw)
efda3bd0 1185 kfree(execsw, M_TEMP);
984263bc
MD
1186 execsw = newexecsw;
1187 return 0;
1188}