Kernel - Fix numerous procfs/ptrace issues
[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 157SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT,
1237359e 158 0, 0, sysctl_kern_stackgap, "I",
d0e92318 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;
7c37ea07
MD
505 if (p->p_stops)
506 wakeup(&p->p_stype);
984263bc
MD
507
508 /*
509 * If tracing the process, trap to debugger so breakpoints
510 * can be set before the program executes.
511 */
512 STOPEVENT(p, S_EXEC, 0);
513
4643740a 514 if (p->p_flags & P_TRACED)
84204577 515 ksignal(p, SIGTRAP);
984263bc
MD
516
517 /* clear "fork but no exec" flag, as we _are_ execing */
518 p->p_acflag &= ~AFORK;
519
520 /* Set values passed into the program in registers. */
08f2f1bb 521 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
a2ee730d 522 imgp->ps_strings);
984263bc 523
349433c9
MD
524 /* Set the access time on the vnode */
525 vn_mark_atime(imgp->vp, td);
526
19bfc8ab
MD
527 /*
528 * Free any previous argument cache
529 */
530 pa = p->p_args;
984263bc 531 p->p_args = NULL;
19bfc8ab
MD
532 if (pa && refcount_release(&pa->ar_ref)) {
533 kfree(pa, M_PARGS);
534 pa = NULL;
535 }
984263bc 536
19bfc8ab
MD
537 /*
538 * Cache arguments if they fit inside our allowance
539 */
2bd9d75c 540 i = imgp->args->begin_envv - imgp->args->begin_argv;
19bfc8ab
MD
541 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
542 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
543 refcount_init(&pa->ar_ref, 1);
544 pa->ar_length = i;
545 bcopy(imgp->args->begin_argv, pa->ar_args, i);
546 KKASSERT(p->p_args == NULL);
547 p->p_args = pa;
984263bc
MD
548 }
549
550exec_fail_dealloc:
551
552 /*
553 * free various allocated resources
554 */
555 if (imgp->firstpage)
556 exec_unmap_first_page(imgp);
557
984263bc 558 if (imgp->vp) {
984263bc 559 vrele(imgp->vp);
fad57d0e 560 imgp->vp = NULL;
984263bc
MD
561 }
562
bf52a6ac
MD
563 if (imgp->freepath)
564 kfree(imgp->freepath, M_TEMP);
565
a6f89c72
DR
566 if (error == 0) {
567 ++mycpu->gd_cnt.v_exec;
19bfc8ab 568 lwkt_reltoken(&p->p_token);
984263bc 569 return (0);
a6f89c72 570 }
984263bc
MD
571
572exec_fail:
dc52e1cc
MD
573 /*
574 * we're done here, clear P_INEXEC if we were the ones that
575 * set it. Otherwise if vmspace_destroyed is still set we
576 * raced another thread and that thread is responsible for
577 * clearing it.
578 */
7c37ea07 579 if (imgp->vmspace_destroyed & 2) {
4643740a 580 p->p_flags &= ~P_INEXEC;
7c37ea07
MD
581 if (p->p_stops)
582 wakeup(&p->p_stype);
583 }
19bfc8ab 584 lwkt_reltoken(&p->p_token);
984263bc 585 if (imgp->vmspace_destroyed) {
5417cd57
SS
586 /*
587 * Sorry, no more process anymore. exit gracefully.
588 * However we can't die right here, because our
589 * caller might have to clean up, so indicate a
90947cb1 590 * lethal error by returning -1.
5417cd57
SS
591 */
592 return(-1);
984263bc
MD
593 } else {
594 return(error);
595 }
596}
597
2bd9d75c
DRJ
598/*
599 * execve() system call.
600 */
601int
753fd850 602sys_execve(struct execve_args *uap)
2bd9d75c 603{
fad57d0e 604 struct nlookupdata nd;
2bd9d75c
DRJ
605 struct image_args args;
606 int error;
607
118cec38 608 bzero(&args, sizeof(args));
3919ced0 609
3919ced0 610 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
fad57d0e
MD
611 if (error == 0) {
612 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
613 uap->argv, uap->envv);
614 }
2abfe22f
MD
615 if (error == 0)
616 error = kern_execve(&nd, &args);
fad57d0e 617 nlookup_done(&nd);
2bd9d75c
DRJ
618 exec_free_args(&args);
619
5417cd57 620 if (error < 0) {
90947cb1 621 /* We hit a lethal error condition. Let's die now. */
5417cd57
SS
622 exit1(W_EXITCODE(0, SIGABRT));
623 /* NOTREACHED */
624 }
625
2bd9d75c
DRJ
626 /*
627 * The syscall result is returned in registers to the new program.
628 * Linux will register %edx as an atexit function and we must be
629 * sure to set it to 0. XXX
630 */
631 if (error == 0)
632 uap->sysmsg_result64 = 0;
633
634 return (error);
635}
636
984263bc 637int
7c34d798 638exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
5c5185ae 639 struct lwbuf **plwb, const char **pdata)
984263bc 640{
1b9d3514
MD
641 int rv;
642 vm_page_t ma;
06ecca5a 643 vm_page_t m;
984263bc
MD
644 vm_object_t object;
645
fad57d0e 646 /*
7540ab49 647 * The file has to be mappable.
fad57d0e 648 */
7540ab49 649 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 650 return (EIO);
984263bc 651
7c34d798
SS
652 if (pageno >= object->size)
653 return (EIO);
654
b12defdc 655 vm_object_hold(object);
7c34d798 656 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1b9d3514
MD
657 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
658 ma = m;
984263bc 659
06ecca5a
MD
660 /*
661 * get_pages unbusies all the requested pages except the
a55afca2
MD
662 * primary page (at index 0 in this case). The primary
663 * page may have been wired during the pagein (e.g. by
664 * the buffer cache) so vnode_pager_freepage() must be
665 * used to properly release it.
06ecca5a 666 */
1b9d3514 667 rv = vm_pager_get_page(object, &ma, 1);
7c34d798 668 m = vm_page_lookup(object, pageno);
984263bc 669
06ecca5a
MD
670 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
671 if (m) {
672 vm_page_protect(m, VM_PROT_NONE);
a55afca2 673 vnode_pager_freepage(m);
984263bc 674 }
471bd290 675 vm_object_drop(object);
984263bc
MD
676 return EIO;
677 }
678 }
b12defdc 679 vm_page_hold(m);
06ecca5a 680 vm_page_wakeup(m); /* unbusy the page */
b12defdc 681 vm_object_drop(object);
984263bc 682
7a683a24 683 *plwb = lwbuf_alloc(m, *plwb);
5c5185ae 684 *pdata = (void *)lwbuf_kva(*plwb);
7c34d798
SS
685
686 return (0);
687}
688
7484bdd2
MD
689/*
690 * Map the first page of an executable image.
691 *
692 * NOTE: If the mapping fails we have to NULL-out firstpage which may
693 * still be pointing to our supplied lwp structure.
694 */
7c34d798
SS
695int
696exec_map_first_page(struct image_params *imgp)
697{
698 int err;
699
700 if (imgp->firstpage)
701 exec_unmap_first_page(imgp);
702
7a683a24 703 imgp->firstpage = &imgp->firstpage_cache;
7c34d798
SS
704 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
705
7484bdd2
MD
706 if (err) {
707 imgp->firstpage = NULL;
7c34d798 708 return err;
7484bdd2 709 }
984263bc
MD
710
711 return 0;
712}
713
714void
5c5185ae 715exec_unmap_page(struct lwbuf *lwb)
984263bc 716{
18f40545
MD
717 vm_page_t m;
718
5fd012e0 719 crit_enter();
5c5185ae
SG
720 if (lwb != NULL) {
721 m = lwbuf_page(lwb);
722 lwbuf_free(lwb);
e4ba7818 723 vm_page_unhold(m);
984263bc 724 }
5fd012e0 725 crit_exit();
984263bc
MD
726}
727
7c34d798
SS
728void
729exec_unmap_first_page(struct image_params *imgp)
730{
731 exec_unmap_page(imgp->firstpage);
732 imgp->firstpage = NULL;
733 imgp->image_header = NULL;
734}
735
984263bc
MD
736/*
737 * Destroy old address space, and allocate a new stack
738 * The new stack is only SGROWSIZ large because it is grown
739 * automatically in trap.c.
dc52e1cc
MD
740 *
741 * This is the point of no return.
984263bc
MD
742 */
743int
29802dbb 744exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 745{
984263bc
MD
746 struct vmspace *vmspace = imgp->proc->p_vmspace;
747 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 748 struct proc *p;
29802dbb 749 vm_map_t map;
dc52e1cc 750 int error;
984263bc 751
dc52e1cc
MD
752 /*
753 * Indicate that we cannot gracefully error out any more, kill
754 * any other threads present, and set P_INEXEC to indicate that
755 * we are now messing with the process structure proper.
756 *
757 * If killalllwps() races return an error which coupled with
758 * vmspace_destroyed will cause us to exit. This is what we
759 * want since another thread is patiently waiting for us to exit
760 * in that case.
761 */
762 p = curproc;
984263bc
MD
763 imgp->vmspace_destroyed = 1;
764
dc52e1cc
MD
765 if (curthread->td_proc->p_nthreads > 1) {
766 error = killalllwps(1);
767 if (error)
768 return (error);
769 }
770 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
4643740a 771 p->p_flags |= P_INEXEC;
08f2f1bb
SS
772
773 /*
7c37ea07
MD
774 * Tell procfs to release its hold on the process. It
775 * will return EAGAIN.
776 */
777 if (p->p_stops)
778 wakeup(&p->p_stype);
779
780 /*
82354ad8
MD
781 * After setting P_INEXEC wait for any remaining references to
782 * the process (p) to go away.
783 *
784 * In particular, a vfork/exec sequence will replace p->p_vmspace
785 * and we must interlock anyone trying to access the space (aka
786 * procfs or sys_process.c calling procfs_domem()).
787 *
788 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us.
789 */
790 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0));
791
792 /*
984263bc
MD
793 * Blow away entire process VM, if address space not shared,
794 * otherwise, create a new VM space so that other threads are
29802dbb
MD
795 * not disrupted. If we are execing a resident vmspace we
796 * create a duplicate of it and remap the stack.
984263bc 797 */
29802dbb
MD
798 map = &vmspace->vm_map;
799 if (vmcopy) {
800 vmspace_exec(imgp->proc, vmcopy);
801 vmspace = imgp->proc->p_vmspace;
802 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
803 map = &vmspace->vm_map;
a2ee730d 804 } else if (vmspace->vm_sysref.refcnt == 1) {
fef0fdf2 805 shmexit(vmspace);
239b4df9 806 if (vmspace->vm_upcalls)
08f2f1bb 807 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08 808 pmap_remove_pages(vmspace_pmap(vmspace),
b0c15cdf 809 0, VM_MAX_USER_ADDRESS);
88181b08 810 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 811 } else {
29802dbb 812 vmspace_exec(imgp->proc, NULL);
984263bc
MD
813 vmspace = imgp->proc->p_vmspace;
814 map = &vmspace->vm_map;
815 }
816
817 /* Allocate a new stack */
818 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
c809941b 819 0, VM_PROT_ALL, VM_PROT_ALL, 0);
984263bc
MD
820 if (error)
821 return (error);
822
823 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
824 * VM_STACK case, but they are still used to monitor the size of the
825 * process stack so we can check the stack rlimit.
826 */
827 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
828 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
829
830 return(0);
831}
832
833/*
834 * Copy out argument and environment strings from the old process
835 * address space into the temporary string buffer.
836 */
837int
2bd9d75c 838exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 839 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 840{
984263bc 841 char *argp, *envp;
2bd9d75c 842 int error = 0;
984263bc
MD
843 size_t length;
844
18b3f457 845 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
846 if (args->buf == NULL)
847 return (ENOMEM);
848 args->begin_argv = args->buf;
849 args->endp = args->begin_argv;
850 args->space = ARG_MAX;
2bd9d75c
DRJ
851
852 args->fname = args->buf + ARG_MAX;
853
984263bc 854 /*
2bd9d75c 855 * Copy the file name.
984263bc 856 */
2bd9d75c
DRJ
857 if (segflg == PATH_SYSSPACE) {
858 error = copystr(fname, args->fname, PATH_MAX, &length);
859 } else if (segflg == PATH_USERSPACE) {
860 error = copyinstr(fname, args->fname, PATH_MAX, &length);
861 }
984263bc 862
2bd9d75c 863 /*
66be6566
MD
864 * Extract argument strings. argv may not be NULL. The argv
865 * array is terminated by a NULL entry. We special-case the
866 * situation where argv[0] is NULL by passing { filename, NULL }
867 * to the new program to guarentee that the interpreter knows what
868 * file to open in case we exec an interpreted file. Note that
869 * a NULL argv[0] terminates the argv[] array.
870 *
871 * XXX the special-casing of argv[0] is historical and needs to be
872 * revisited.
2bd9d75c 873 */
66be6566
MD
874 if (argv == NULL)
875 error = EFAULT;
876 if (error == 0) {
877 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
878 if (argp == (caddr_t)-1) {
2bd9d75c 879 error = EFAULT;
66be6566 880 break;
2bd9d75c
DRJ
881 }
882 error = copyinstr(argp, args->endp,
d0e92318 883 args->space, &length);
66be6566
MD
884 if (error) {
885 if (error == ENAMETOOLONG)
886 error = E2BIG;
887 break;
888 }
2bd9d75c
DRJ
889 args->space -= length;
890 args->endp += length;
891 args->argc++;
984263bc 892 }
66be6566
MD
893 if (args->argc == 0 && error == 0) {
894 length = strlen(args->fname) + 1;
895 if (length > args->space) {
896 error = E2BIG;
897 } else {
898 bcopy(args->fname, args->endp, length);
899 args->space -= length;
900 args->endp += length;
901 args->argc++;
902 }
903 }
984263bc
MD
904 }
905
2bd9d75c 906 args->begin_envv = args->endp;
984263bc
MD
907
908 /*
66be6566 909 * extract environment strings. envv may be NULL.
984263bc 910 */
2abfe22f 911 if (envv && error == 0) {
984263bc 912 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
913 if (envp == (caddr_t) -1) {
914 error = EFAULT;
66be6566 915 break;
984263bc 916 }
d0e92318
MD
917 error = copyinstr(envp, args->endp,
918 args->space, &length);
66be6566
MD
919 if (error) {
920 if (error == ENAMETOOLONG)
921 error = E2BIG;
922 break;
923 }
2bd9d75c
DRJ
924 args->space -= length;
925 args->endp += length;
926 args->envc++;
984263bc
MD
927 }
928 }
2bd9d75c
DRJ
929 return (error);
930}
931
932void
933exec_free_args(struct image_args *args)
934{
2abfe22f 935 if (args->buf) {
18b3f457 936 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
937 args->buf = NULL;
938 }
984263bc
MD
939}
940
941/*
942 * Copy strings out to the new process address space, constructing
d0e92318
MD
943 * new arg and env vector tables. Return a pointer to the base
944 * so that it can be used as the initial stack pointer.
945 *
946 * The format is, roughly:
947 *
948 * [argv[]] <-- vectp
949 * [envp[]]
950 * [ELF_Auxargs]
951 *
952 * [args & env] <-- destp
953 * [sgap]
954 * [SPARE_USRSPACE]
955 * [execpath]
956 * [szsigcode]
957 * [ps_strings] top of user stack
958 *
984263bc
MD
959 */
960register_t *
2bd9d75c 961exec_copyout_strings(struct image_params *imgp)
984263bc 962{
17a7ca1f 963 int argc, envc, sgap;
d0e92318
MD
964 int gap;
965 int argsenvspace;
984263bc
MD
966 char **vectp;
967 char *stringp, *destp;
968 register_t *stack_base;
969 struct ps_strings *arginfo;
adc42cf3 970 size_t execpath_len;
984263bc
MD
971 int szsigcode;
972
973 /*
974 * Calculate string base and vector table pointers.
975 * Also deal with signal trampoline code for this exec type.
976 */
adc42cf3
JM
977 if (imgp->execpath != NULL && imgp->auxargs != NULL)
978 execpath_len = strlen(imgp->execpath) + 1;
979 else
980 execpath_len = 0;
984263bc
MD
981 arginfo = (struct ps_strings *)PS_STRINGS;
982 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
d0e92318
MD
983
984 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *));
985 gap = stackgap_random;
986 cpu_ccfence();
987 if (gap != 0) {
988 if (gap < 0)
989 sgap = ALIGN(-gap);
990 else
1bd7d69f 991 sgap = ALIGN(karc4random() & (gap - 1));
d0e92318
MD
992 } else {
993 sgap = 0;
994 }
995
d0e92318
MD
996 /*
997 * Calculate destp, which points to [args & env] and above.
998 */
1bd7d69f
MD
999 destp = (caddr_t)arginfo -
1000 szsigcode -
d0e92318
MD
1001 roundup(execpath_len, sizeof(char *)) -
1002 SPARE_USRSPACE -
1003 sgap -
1004 argsenvspace;
984263bc
MD
1005
1006 /*
1007 * install sigcode
1008 */
d0e92318 1009 if (szsigcode) {
984263bc 1010 copyout(imgp->proc->p_sysent->sv_sigcode,
d0e92318
MD
1011 ((caddr_t)arginfo - szsigcode), szsigcode);
1012 }
984263bc
MD
1013
1014 /*
adc42cf3
JM
1015 * Copy the image path for the rtld
1016 */
d0e92318 1017 if (execpath_len) {
adc42cf3
JM
1018 imgp->execpathp = (uintptr_t)arginfo
1019 - szsigcode
d0e92318 1020 - roundup(execpath_len, sizeof(char *));
adc42cf3
JM
1021 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
1022 }
1023
1024 /*
d0e92318 1025 * Calculate base for argv[], envp[], and ELF_Auxargs.
984263bc 1026 */
d0e92318
MD
1027 vectp = (char **)destp - (AT_COUNT * 2);
1028 vectp -= imgp->args->argc + imgp->args->envc + 2;
984263bc 1029
984263bc
MD
1030 stack_base = (register_t *)vectp;
1031
2bd9d75c
DRJ
1032 stringp = imgp->args->begin_argv;
1033 argc = imgp->args->argc;
1034 envc = imgp->args->envc;
984263bc
MD
1035
1036 /*
d0e92318 1037 * Copy out strings - arguments and environment (at destp)
984263bc 1038 */
2bd9d75c 1039 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
1040
1041 /*
1042 * Fill in "ps_strings" struct for ps, w, etc.
1043 */
1044 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
3f938eea 1045 suword32(&arginfo->ps_nargvstr, argc);
984263bc
MD
1046
1047 /*
1048 * Fill in argument portion of vector table.
1049 */
1050 for (; argc > 0; --argc) {
1051 suword(vectp++, (long)(intptr_t)destp);
1052 while (*stringp++ != 0)
1053 destp++;
1054 destp++;
1055 }
1056
7a44d1cb 1057 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
1058 suword(vectp++, 0);
1059
1060 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
3f938eea 1061 suword32(&arginfo->ps_nenvstr, envc);
984263bc
MD
1062
1063 /*
1064 * Fill in environment portion of vector table.
1065 */
1066 for (; envc > 0; --envc) {
1067 suword(vectp++, (long)(intptr_t)destp);
1068 while (*stringp++ != 0)
1069 destp++;
1070 destp++;
1071 }
1072
1073 /* end of vector table is a null pointer */
1074 suword(vectp, 0);
1075
1076 return (stack_base);
1077}
1078
1079/*
1080 * Check permissions of file to execute.
1081 * Return 0 for success or error code on failure.
1082 */
1083int
246693ac 1084exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
984263bc
MD
1085{
1086 struct proc *p = imgp->proc;
1087 struct vnode *vp = imgp->vp;
1088 struct vattr *attr = imgp->attr;
1089 int error;
1090
1091 /* Get file attributes */
87de5057 1092 error = VOP_GETATTR(vp, attr);
984263bc
MD
1093 if (error)
1094 return (error);
1095
1096 /*
1097 * 1) Check if file execution is disabled for the filesystem that this
1098 * file resides on.
1099 * 2) Insure that at least one execute bit is on - otherwise root
1100 * will always succeed, and we don't want to happen unless the
1101 * file really is executable.
1102 * 3) Insure that the file is a regular file.
1103 */
1104 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
246693ac 1105 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
984263bc
MD
1106 ((attr->va_mode & 0111) == 0) ||
1107 (attr->va_type != VREG)) {
1108 return (EACCES);
1109 }
1110
1111 /*
1112 * Zero length files can't be exec'd
1113 */
1114 if (attr->va_size == 0)
1115 return (ENOEXEC);
1116
1117 /*
1118 * Check for execute permission to file based on current credentials.
1119 */
90a83564 1120 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
1121 if (error)
1122 return (error);
1123
1124 /*
1125 * Check number of open-for-writes on the file and deny execution
1126 * if there are any.
1127 */
1128 if (vp->v_writecount)
1129 return (ETXTBSY);
1130
1131 /*
7540ab49
MD
1132 * Call filesystem specific open routine, which allows us to read,
1133 * write, and mmap the file. Without the VOP_OPEN we can only
1134 * stat the file.
984263bc 1135 */
87de5057 1136 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
1137 if (error)
1138 return (error);
1139
1140 return (0);
1141}
1142
1143/*
1144 * Exec handler registration
1145 */
1146int
77153250 1147exec_register(const struct execsw *execsw_arg)
984263bc
MD
1148{
1149 const struct execsw **es, **xs, **newexecsw;
1150 int count = 2; /* New slot and trailing NULL */
1151
1152 if (execsw)
1153 for (es = execsw; *es; es++)
1154 count++;
efda3bd0 1155 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1156 xs = newexecsw;
1157 if (execsw)
1158 for (es = execsw; *es; es++)
1159 *xs++ = *es;
1160 *xs++ = execsw_arg;
1161 *xs = NULL;
1162 if (execsw)
efda3bd0 1163 kfree(execsw, M_TEMP);
984263bc
MD
1164 execsw = newexecsw;
1165 return 0;
1166}
1167
1168int
77153250 1169exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1170{
1171 const struct execsw **es, **xs, **newexecsw;
1172 int count = 1;
1173
1174 if (execsw == NULL)
ae7e74d4 1175 panic("unregister with no handlers left?");
984263bc
MD
1176
1177 for (es = execsw; *es; es++) {
1178 if (*es == execsw_arg)
1179 break;
1180 }
1181 if (*es == NULL)
1182 return ENOENT;
1183 for (es = execsw; *es; es++)
1184 if (*es != execsw_arg)
1185 count++;
efda3bd0 1186 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1187 xs = newexecsw;
1188 for (es = execsw; *es; es++)
1189 if (*es != execsw_arg)
1190 *xs++ = *es;
1191 *xs = NULL;
1192 if (execsw)
efda3bd0 1193 kfree(execsw, M_TEMP);
984263bc
MD
1194 execsw = newexecsw;
1195 return 0;
1196}