Merge branch 'gcc442'
[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 $
0aa16b5d 27 * $DragonFly: src/sys/kern/kern_exec.c,v 1.64 2008/10/26 04:29:19 sephe Exp $
984263bc
MD
28 */
29
30#include <sys/param.h>
31#include <sys/systm.h>
32#include <sys/sysproto.h>
33#include <sys/kernel.h>
34#include <sys/mount.h>
35#include <sys/filedesc.h>
36#include <sys/fcntl.h>
37#include <sys/acct.h>
38#include <sys/exec.h>
39#include <sys/imgact.h>
40#include <sys/imgact_elf.h>
2bd9d75c 41#include <sys/kern_syscall.h>
984263bc
MD
42#include <sys/wait.h>
43#include <sys/malloc.h>
44#include <sys/proc.h>
895c1f85 45#include <sys/priv.h>
29f58392 46#include <sys/ktrace.h>
984263bc
MD
47#include <sys/signalvar.h>
48#include <sys/pioctl.h>
fad57d0e 49#include <sys/nlookup.h>
18f40545 50#include <sys/sfbuf.h>
984263bc
MD
51#include <sys/sysent.h>
52#include <sys/shm.h>
53#include <sys/sysctl.h>
54#include <sys/vnode.h>
a6f89c72 55#include <sys/vmmeter.h>
984263bc 56#include <sys/aio.h>
17a7ca1f 57#include <sys/libkern.h>
984263bc
MD
58
59#include <vm/vm.h>
60#include <vm/vm_param.h>
61#include <sys/lock.h>
62#include <vm/pmap.h>
63#include <vm/vm_page.h>
64#include <vm/vm_map.h>
65#include <vm/vm_kern.h>
66#include <vm/vm_extern.h>
67#include <vm/vm_object.h>
a55afca2 68#include <vm/vnode_pager.h>
984263bc
MD
69#include <vm/vm_pager.h>
70
71#include <sys/user.h>
527fddf7 72#include <sys/reg.h>
984263bc 73
5fd012e0
MD
74#include <sys/thread2.h>
75
984263bc 76MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
18b3f457 77MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
984263bc 78
402ed7e1 79static register_t *exec_copyout_strings (struct image_params *);
984263bc
MD
80
81/* XXX This should be vm_size_t. */
82static u_long ps_strings = PS_STRINGS;
83SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
84
85/* XXX This should be vm_size_t. */
86static u_long usrstack = USRSTACK;
87SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
88
89u_long ps_arg_cache_limit = PAGE_SIZE / 16;
90SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
91 &ps_arg_cache_limit, 0, "");
92
93int ps_argsopen = 1;
94SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
95
5c627295
MD
96static int ktrace_suid = 0;
97SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
98
2bd9d75c
DRJ
99void print_execve_args(struct image_args *args);
100int debug_execve_args = 0;
101SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
102 0, "");
103
18b3f457
MD
104/*
105 * Exec arguments object cache
106 */
107static struct objcache *exec_objcache;
108
109static
110void
111exec_objcache_init(void *arg __unused)
112{
0aa16b5d
SZ
113 int cluster_limit;
114
115 cluster_limit = 16; /* up to this many objects */
18b3f457
MD
116 exec_objcache = objcache_create_mbacked(
117 M_EXECARGS, PATH_MAX + ARG_MAX,
0aa16b5d 118 &cluster_limit,
18b3f457
MD
119 2, /* minimal magazine capacity */
120 NULL, NULL, NULL);
121}
122SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
123
17a7ca1f
MD
124/*
125 * stackgap_random specifies if the stackgap should have a random size added
126 * to it. It must be a power of 2. If non-zero, the stack gap will be
0ced1954 127 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
17a7ca1f
MD
128 */
129static int stackgap_random = 1024;
130static int
131sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
132{
133 int error, new_val;
134 new_val = stackgap_random;
135 error = sysctl_handle_int(oidp, &new_val, 0, req);
136 if (error != 0 || req->newptr == NULL)
137 return (error);
138 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val))
139 return (EINVAL);
140 stackgap_random = new_val;
141
142 return(0);
143}
144
145SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT,
146 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
147
2bd9d75c
DRJ
148void
149print_execve_args(struct image_args *args)
150{
151 char *cp;
152 int ndx;
153
154 cp = args->begin_argv;
155 for (ndx = 0; ndx < args->argc; ndx++) {
6ea70f76 156 kprintf("\targv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
157 while (*cp++ != '\0');
158 }
159 for (ndx = 0; ndx < args->envc; ndx++) {
6ea70f76 160 kprintf("\tenvv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
161 while (*cp++ != '\0');
162 }
163}
164
984263bc
MD
165/*
166 * Each of the items is a pointer to a `const struct execsw', hence the
167 * double pointer here.
168 */
169static const struct execsw **execsw;
170
5417cd57
SS
171/*
172 * Replace current vmspace with a new binary.
173 * Returns 0 on success, > 0 on recoverable error (use as errno).
90947cb1 174 * Returns -1 on lethal error which demands killing of the current
5417cd57
SS
175 * process!
176 */
984263bc 177int
fad57d0e 178kern_execve(struct nlookupdata *nd, struct image_args *args)
984263bc 179{
dadab5e9 180 struct thread *td = curthread;
08f2f1bb 181 struct lwp *lp = td->td_lwp;
dadab5e9 182 struct proc *p = td->td_proc;
984263bc
MD
183 register_t *stack_base;
184 int error, len, i;
185 struct image_params image_params, *imgp;
186 struct vattr attr;
402ed7e1 187 int (*img_first) (struct image_params *);
984263bc 188
2bd9d75c 189 if (debug_execve_args) {
6ea70f76 190 kprintf("%s()\n", __func__);
2bd9d75c
DRJ
191 print_execve_args(args);
192 }
193
dadab5e9 194 KKASSERT(p);
984263bc
MD
195 imgp = &image_params;
196
197 /*
dc52e1cc
MD
198 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
199 * no modifications to the process at all until we get there.
200 *
201 * Note that multiple threads may be trying to exec at the same
202 * time. exec_new_vmspace() handles that too.
984263bc 203 */
984263bc
MD
204
205 /*
206 * Initialize part of the common data
207 */
208 imgp->proc = p;
2bd9d75c 209 imgp->args = args;
984263bc 210 imgp->attr = &attr;
984263bc 211 imgp->entry_addr = 0;
29802dbb 212 imgp->resident = 0;
984263bc
MD
213 imgp->vmspace_destroyed = 0;
214 imgp->interpreted = 0;
2abfe22f 215 imgp->interpreter_name[0] = 0;
984263bc
MD
216 imgp->auxargs = NULL;
217 imgp->vp = NULL;
218 imgp->firstpage = NULL;
219 imgp->ps_strings = 0;
18f40545 220 imgp->image_header = NULL;
2bd9d75c
DRJ
221
222interpret:
984263bc
MD
223
224 /*
fad57d0e
MD
225 * Translate the file name to a vnode. Unlock the cache entry to
226 * improve parallelism for programs exec'd in parallel.
984263bc 227 */
fad57d0e
MD
228 if ((error = nlookup(nd)) != 0)
229 goto exec_fail;
28623bf9 230 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
fad57d0e
MD
231 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
232 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 233 cache_unlock(&nd->nl_nch);
fad57d0e 234 if (error)
984263bc 235 goto exec_fail;
984263bc
MD
236
237 /*
238 * Check file permissions (also 'opens' file)
239 */
240 error = exec_check_permissions(imgp);
241 if (error) {
a11aaa81 242 vn_unlock(imgp->vp);
984263bc
MD
243 goto exec_fail_dealloc;
244 }
245
246 error = exec_map_first_page(imgp);
a11aaa81 247 vn_unlock(imgp->vp);
984263bc
MD
248 if (error)
249 goto exec_fail_dealloc;
250
2bd9d75c 251 if (debug_execve_args && imgp->interpreted) {
6ea70f76
SW
252 kprintf(" target is interpreted -- recursive pass\n");
253 kprintf(" interpreter: %s\n", imgp->interpreter_name);
2bd9d75c
DRJ
254 print_execve_args(args);
255 }
256
984263bc
MD
257 /*
258 * If the current process has a special image activator it
259 * wants to try first, call it. For example, emulating shell
260 * scripts differently.
261 */
262 error = -1;
263 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
264 error = img_first(imgp);
265
29802dbb
MD
266 /*
267 * If the vnode has a registered vmspace, exec the vmspace
268 */
269 if (error == -1 && imgp->vp->v_resident) {
270 error = exec_resident_imgact(imgp);
271 }
272
984263bc
MD
273 /*
274 * Loop through the list of image activators, calling each one.
275 * An activator returns -1 if there is no match, 0 on success,
276 * and an error otherwise.
277 */
278 for (i = 0; error == -1 && execsw[i]; ++i) {
279 if (execsw[i]->ex_imgact == NULL ||
280 execsw[i]->ex_imgact == img_first) {
281 continue;
282 }
283 error = (*execsw[i]->ex_imgact)(imgp);
284 }
285
286 if (error) {
287 if (error == -1)
288 error = ENOEXEC;
289 goto exec_fail_dealloc;
290 }
291
292 /*
293 * Special interpreter operation, cleanup and loop up to try to
294 * activate the interpreter.
295 */
296 if (imgp->interpreted) {
297 exec_unmap_first_page(imgp);
fad57d0e
MD
298 nlookup_done(nd);
299 vrele(imgp->vp);
300 imgp->vp = NULL;
301 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
302 NLC_FOLLOW);
303 if (error)
304 goto exec_fail;
984263bc
MD
305 goto interpret;
306 }
307
308 /*
309 * Copy out strings (args and env) and initialize stack base
310 */
311 stack_base = exec_copyout_strings(imgp);
312 p->p_vmspace->vm_minsaddr = (char *)stack_base;
313
314 /*
315 * If custom stack fixup routine present for this process
29802dbb
MD
316 * let it do the stack setup. If we are running a resident
317 * image there is no auxinfo or other image activator context
318 * so don't try to add fixups to the stack.
319 *
984263bc
MD
320 * Else stuff argument count as first item on stack
321 */
29802dbb 322 if (p->p_sysent->sv_fixup && imgp->resident == 0)
984263bc
MD
323 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
324 else
2bd9d75c 325 suword(--stack_base, imgp->args->argc);
984263bc
MD
326
327 /*
328 * For security and other reasons, the file descriptor table cannot
329 * be shared after an exec.
330 */
331 if (p->p_fd->fd_refcnt > 1) {
332 struct filedesc *tmp;
333
334 tmp = fdcopy(p);
0a4a9c77 335 fdfree(p, tmp);
984263bc
MD
336 }
337
338 /*
339 * For security and other reasons, signal handlers cannot
340 * be shared after an exec. The new proces gets a copy of the old
341 * handlers. In execsigs(), the new process will have its signals
342 * reset.
343 */
b1b4e5a6
SS
344 if (p->p_sigacts->ps_refcnt > 1) {
345 struct sigacts *newsigacts;
984263bc 346
b1b4e5a6 347 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts),
984263bc 348 M_SUBPROC, M_WAITOK);
b1b4e5a6
SS
349 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts));
350 p->p_sigacts->ps_refcnt--;
351 p->p_sigacts = newsigacts;
352 p->p_sigacts->ps_refcnt = 1;
984263bc
MD
353 }
354
0daa37a5
MD
355 /*
356 * For security and other reasons virtual kernels cannot be
357 * inherited by an exec. This also allows a virtual kernel
358 * to fork/exec unrelated applications.
359 */
4a22e893
MD
360 if (p->p_vkernel)
361 vkernel_exit(p);
0daa37a5 362
984263bc
MD
363 /* Stop profiling */
364 stopprofclock(p);
365
366 /* close files on exec */
367 fdcloseexec(p);
368
369 /* reset caught signals */
370 execsigs(p);
371
372 /* name this process - nameiexec(p, ndp) */
28623bf9
MD
373 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
374 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
984263bc 375 p->p_comm[len] = 0;
08f2f1bb 376 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
984263bc
MD
377
378 /*
379 * mark as execed, wakeup the process that vforked (if any) and tell
380 * it that it now has its own resources back
381 */
382 p->p_flag |= P_EXEC;
383 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
384 p->p_flag &= ~P_PPWAIT;
385 wakeup((caddr_t)p->p_pptr);
386 }
387
388 /*
389 * Implement image setuid/setgid.
390 *
391 * Don't honor setuid/setgid if the filesystem prohibits it or if
392 * the process is being traced.
393 */
394 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
395 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
396 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
397 (p->p_flag & P_TRACED) == 0) {
398 /*
399 * Turn off syscall tracing for set-id programs, except for
400 * root. Record any set-id flags first to make sure that
401 * we do not regain any tracing during a possible block.
402 */
41c20dac 403 setsugid();
5c627295
MD
404 if (p->p_tracenode && ktrace_suid == 0 &&
405 priv_check(td, PRIV_ROOT) != 0) {
29f58392
MD
406 ktrdestroy(&p->p_tracenode);
407 p->p_traceflag = 0;
984263bc
MD
408 }
409 /* Close any file descriptors 0..2 that reference procfs */
410 setugidsafety(p);
411 /* Make sure file descriptors 0..2 are in use. */
412 error = fdcheckstd(p);
413 if (error != 0)
414 goto exec_fail_dealloc;
415 /*
416 * Set the new credentials.
417 */
e9a372eb 418 cratom(&p->p_ucred);
984263bc 419 if (attr.va_mode & VSUID)
41c20dac 420 change_euid(attr.va_uid);
984263bc
MD
421 if (attr.va_mode & VSGID)
422 p->p_ucred->cr_gid = attr.va_gid;
98a7f915
MD
423
424 /*
425 * Clear local varsym variables
426 */
427 varsymset_clean(&p->p_varsymset);
984263bc 428 } else {
41c20dac
MD
429 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
430 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
984263bc
MD
431 p->p_flag &= ~P_SUGID;
432 }
433
434 /*
435 * Implement correct POSIX saved-id behavior.
436 */
115ccd83
MD
437 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
438 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
439 cratom(&p->p_ucred);
440 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
441 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
442 }
984263bc
MD
443
444 /*
445 * Store the vp for use in procfs
446 */
447 if (p->p_textvp) /* release old reference */
448 vrele(p->p_textvp);
fad57d0e
MD
449 p->p_textvp = imgp->vp;
450 vref(p->p_textvp);
984263bc
MD
451
452 /*
453 * Notify others that we exec'd, and clear the P_INEXEC flag
454 * as we're now a bona fide freshly-execed process.
455 */
456 KNOTE(&p->p_klist, NOTE_EXEC);
457 p->p_flag &= ~P_INEXEC;
458
459 /*
460 * If tracing the process, trap to debugger so breakpoints
461 * can be set before the program executes.
462 */
463 STOPEVENT(p, S_EXEC, 0);
464
465 if (p->p_flag & P_TRACED)
84204577 466 ksignal(p, SIGTRAP);
984263bc
MD
467
468 /* clear "fork but no exec" flag, as we _are_ execing */
469 p->p_acflag &= ~AFORK;
470
471 /* Set values passed into the program in registers. */
08f2f1bb 472 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
984263bc
MD
473 imgp->ps_strings);
474
349433c9
MD
475 /* Set the access time on the vnode */
476 vn_mark_atime(imgp->vp, td);
477
984263bc
MD
478 /* Free any previous argument cache */
479 if (p->p_args && --p->p_args->ar_ref == 0)
480 FREE(p->p_args, M_PARGS);
481 p->p_args = NULL;
482
483 /* Cache arguments if they fit inside our allowance */
2bd9d75c 484 i = imgp->args->begin_envv - imgp->args->begin_argv;
984263bc
MD
485 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
486 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
487 M_PARGS, M_WAITOK);
488 p->p_args->ar_ref = 1;
489 p->p_args->ar_length = i;
2bd9d75c 490 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
984263bc
MD
491 }
492
493exec_fail_dealloc:
494
495 /*
496 * free various allocated resources
497 */
498 if (imgp->firstpage)
499 exec_unmap_first_page(imgp);
500
984263bc 501 if (imgp->vp) {
984263bc 502 vrele(imgp->vp);
fad57d0e 503 imgp->vp = NULL;
984263bc
MD
504 }
505
a6f89c72
DR
506 if (error == 0) {
507 ++mycpu->gd_cnt.v_exec;
984263bc 508 return (0);
a6f89c72 509 }
984263bc
MD
510
511exec_fail:
dc52e1cc
MD
512 /*
513 * we're done here, clear P_INEXEC if we were the ones that
514 * set it. Otherwise if vmspace_destroyed is still set we
515 * raced another thread and that thread is responsible for
516 * clearing it.
517 */
518 if (imgp->vmspace_destroyed & 2)
519 p->p_flag &= ~P_INEXEC;
984263bc 520 if (imgp->vmspace_destroyed) {
5417cd57
SS
521 /*
522 * Sorry, no more process anymore. exit gracefully.
523 * However we can't die right here, because our
524 * caller might have to clean up, so indicate a
90947cb1 525 * lethal error by returning -1.
5417cd57
SS
526 */
527 return(-1);
984263bc
MD
528 } else {
529 return(error);
530 }
531}
532
2bd9d75c
DRJ
533/*
534 * execve() system call.
535 */
536int
753fd850 537sys_execve(struct execve_args *uap)
2bd9d75c 538{
fad57d0e 539 struct nlookupdata nd;
2bd9d75c
DRJ
540 struct image_args args;
541 int error;
542
fad57d0e 543 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
118cec38 544 bzero(&args, sizeof(args));
fad57d0e
MD
545 if (error == 0) {
546 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
547 uap->argv, uap->envv);
548 }
2abfe22f
MD
549 if (error == 0)
550 error = kern_execve(&nd, &args);
fad57d0e 551 nlookup_done(&nd);
2bd9d75c
DRJ
552 exec_free_args(&args);
553
5417cd57 554 if (error < 0) {
90947cb1 555 /* We hit a lethal error condition. Let's die now. */
5417cd57
SS
556 exit1(W_EXITCODE(0, SIGABRT));
557 /* NOTREACHED */
558 }
559
2bd9d75c
DRJ
560 /*
561 * The syscall result is returned in registers to the new program.
562 * Linux will register %edx as an atexit function and we must be
563 * sure to set it to 0. XXX
564 */
565 if (error == 0)
566 uap->sysmsg_result64 = 0;
567
568 return (error);
569}
570
984263bc 571int
41c20dac 572exec_map_first_page(struct image_params *imgp)
984263bc 573{
5fd012e0 574 int rv, i;
984263bc
MD
575 int initial_pagein;
576 vm_page_t ma[VM_INITIAL_PAGEIN];
06ecca5a 577 vm_page_t m;
984263bc
MD
578 vm_object_t object;
579
18f40545 580 if (imgp->firstpage)
984263bc 581 exec_unmap_first_page(imgp);
984263bc 582
fad57d0e 583 /*
7540ab49 584 * The file has to be mappable.
fad57d0e 585 */
7540ab49 586 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 587 return (EIO);
984263bc 588
06ecca5a
MD
589 /*
590 * We shouldn't need protection for vm_page_grab() but we certainly
591 * need it for the lookup loop below (lookup/busy race), since
592 * an interrupt can unbusy and free the page before our busy check.
593 */
5fd012e0 594 crit_enter();
06ecca5a 595 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
984263bc 596
06ecca5a
MD
597 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
598 ma[0] = m;
984263bc
MD
599 initial_pagein = VM_INITIAL_PAGEIN;
600 if (initial_pagein > object->size)
601 initial_pagein = object->size;
602 for (i = 1; i < initial_pagein; i++) {
06ecca5a
MD
603 if ((m = vm_page_lookup(object, i)) != NULL) {
604 if ((m->flags & PG_BUSY) || m->busy)
984263bc 605 break;
06ecca5a 606 if (m->valid)
984263bc 607 break;
06ecca5a 608 vm_page_busy(m);
984263bc 609 } else {
06ecca5a
MD
610 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
611 if (m == NULL)
984263bc
MD
612 break;
613 }
06ecca5a 614 ma[i] = m;
984263bc
MD
615 }
616 initial_pagein = i;
617
06ecca5a
MD
618 /*
619 * get_pages unbusies all the requested pages except the
a55afca2
MD
620 * primary page (at index 0 in this case). The primary
621 * page may have been wired during the pagein (e.g. by
622 * the buffer cache) so vnode_pager_freepage() must be
623 * used to properly release it.
06ecca5a 624 */
984263bc 625 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
06ecca5a 626 m = vm_page_lookup(object, 0);
984263bc 627
06ecca5a
MD
628 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
629 if (m) {
630 vm_page_protect(m, VM_PROT_NONE);
a55afca2 631 vnode_pager_freepage(m);
984263bc 632 }
5fd012e0 633 crit_exit();
984263bc
MD
634 return EIO;
635 }
636 }
06ecca5a
MD
637 vm_page_hold(m);
638 vm_page_wakeup(m); /* unbusy the page */
5fd012e0 639 crit_exit();
984263bc 640
4f1640d6 641 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE);
18f40545 642 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage);
984263bc
MD
643
644 return 0;
645}
646
647void
77153250 648exec_unmap_first_page(struct image_params *imgp)
984263bc 649{
18f40545
MD
650 vm_page_t m;
651
5fd012e0 652 crit_enter();
18f40545
MD
653 if (imgp->firstpage != NULL) {
654 m = sf_buf_page(imgp->firstpage);
655 sf_buf_free(imgp->firstpage);
984263bc 656 imgp->firstpage = NULL;
18f40545 657 imgp->image_header = NULL;
e4ba7818 658 vm_page_unhold(m);
984263bc 659 }
5fd012e0 660 crit_exit();
984263bc
MD
661}
662
663/*
664 * Destroy old address space, and allocate a new stack
665 * The new stack is only SGROWSIZ large because it is grown
666 * automatically in trap.c.
dc52e1cc
MD
667 *
668 * This is the point of no return.
984263bc
MD
669 */
670int
29802dbb 671exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 672{
984263bc
MD
673 struct vmspace *vmspace = imgp->proc->p_vmspace;
674 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 675 struct proc *p;
29802dbb 676 vm_map_t map;
dc52e1cc 677 int error;
984263bc 678
dc52e1cc
MD
679 /*
680 * Indicate that we cannot gracefully error out any more, kill
681 * any other threads present, and set P_INEXEC to indicate that
682 * we are now messing with the process structure proper.
683 *
684 * If killalllwps() races return an error which coupled with
685 * vmspace_destroyed will cause us to exit. This is what we
686 * want since another thread is patiently waiting for us to exit
687 * in that case.
688 */
689 p = curproc;
984263bc
MD
690 imgp->vmspace_destroyed = 1;
691
dc52e1cc
MD
692 if (curthread->td_proc->p_nthreads > 1) {
693 error = killalllwps(1);
694 if (error)
695 return (error);
696 }
697 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
698 p->p_flag |= P_INEXEC;
08f2f1bb 699
984263bc
MD
700 /*
701 * Prevent a pending AIO from modifying the new address space.
702 */
703 aio_proc_rundown(imgp->proc);
704
705 /*
706 * Blow away entire process VM, if address space not shared,
707 * otherwise, create a new VM space so that other threads are
29802dbb
MD
708 * not disrupted. If we are execing a resident vmspace we
709 * create a duplicate of it and remap the stack.
239b4df9
MD
710 *
711 * The exitingcnt test is not strictly necessary but has been
712 * included for code sanity (to make the code more deterministic).
984263bc 713 */
29802dbb
MD
714 map = &vmspace->vm_map;
715 if (vmcopy) {
716 vmspace_exec(imgp->proc, vmcopy);
717 vmspace = imgp->proc->p_vmspace;
718 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
719 map = &vmspace->vm_map;
e3161323
MD
720 } else if (vmspace->vm_sysref.refcnt == 1 &&
721 vmspace->vm_exitingcnt == 0) {
fef0fdf2 722 shmexit(vmspace);
239b4df9 723 if (vmspace->vm_upcalls)
08f2f1bb 724 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08
MD
725 pmap_remove_pages(vmspace_pmap(vmspace),
726 0, VM_MAX_USER_ADDRESS);
727 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 728 } else {
29802dbb 729 vmspace_exec(imgp->proc, NULL);
984263bc
MD
730 vmspace = imgp->proc->p_vmspace;
731 map = &vmspace->vm_map;
732 }
733
734 /* Allocate a new stack */
735 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
c809941b 736 0, VM_PROT_ALL, VM_PROT_ALL, 0);
984263bc
MD
737 if (error)
738 return (error);
739
740 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
741 * VM_STACK case, but they are still used to monitor the size of the
742 * process stack so we can check the stack rlimit.
743 */
744 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
745 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
746
747 return(0);
748}
749
750/*
751 * Copy out argument and environment strings from the old process
752 * address space into the temporary string buffer.
753 */
754int
2bd9d75c 755exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 756 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 757{
984263bc 758 char *argp, *envp;
2bd9d75c 759 int error = 0;
984263bc
MD
760 size_t length;
761
18b3f457 762 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
763 if (args->buf == NULL)
764 return (ENOMEM);
765 args->begin_argv = args->buf;
766 args->endp = args->begin_argv;
767 args->space = ARG_MAX;
2bd9d75c
DRJ
768
769 args->fname = args->buf + ARG_MAX;
770
984263bc 771 /*
2bd9d75c 772 * Copy the file name.
984263bc 773 */
2bd9d75c
DRJ
774 if (segflg == PATH_SYSSPACE) {
775 error = copystr(fname, args->fname, PATH_MAX, &length);
776 } else if (segflg == PATH_USERSPACE) {
777 error = copyinstr(fname, args->fname, PATH_MAX, &length);
778 }
984263bc 779
2bd9d75c 780 /*
66be6566
MD
781 * Extract argument strings. argv may not be NULL. The argv
782 * array is terminated by a NULL entry. We special-case the
783 * situation where argv[0] is NULL by passing { filename, NULL }
784 * to the new program to guarentee that the interpreter knows what
785 * file to open in case we exec an interpreted file. Note that
786 * a NULL argv[0] terminates the argv[] array.
787 *
788 * XXX the special-casing of argv[0] is historical and needs to be
789 * revisited.
2bd9d75c 790 */
66be6566
MD
791 if (argv == NULL)
792 error = EFAULT;
793 if (error == 0) {
794 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
795 if (argp == (caddr_t)-1) {
2bd9d75c 796 error = EFAULT;
66be6566 797 break;
2bd9d75c
DRJ
798 }
799 error = copyinstr(argp, args->endp,
2abfe22f 800 args->space, &length);
66be6566
MD
801 if (error) {
802 if (error == ENAMETOOLONG)
803 error = E2BIG;
804 break;
805 }
2bd9d75c
DRJ
806 args->space -= length;
807 args->endp += length;
808 args->argc++;
984263bc 809 }
66be6566
MD
810 if (args->argc == 0 && error == 0) {
811 length = strlen(args->fname) + 1;
812 if (length > args->space) {
813 error = E2BIG;
814 } else {
815 bcopy(args->fname, args->endp, length);
816 args->space -= length;
817 args->endp += length;
818 args->argc++;
819 }
820 }
984263bc
MD
821 }
822
2bd9d75c 823 args->begin_envv = args->endp;
984263bc
MD
824
825 /*
66be6566 826 * extract environment strings. envv may be NULL.
984263bc 827 */
2abfe22f 828 if (envv && error == 0) {
984263bc 829 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
830 if (envp == (caddr_t) -1) {
831 error = EFAULT;
66be6566 832 break;
984263bc 833 }
2bd9d75c
DRJ
834 error = copyinstr(envp, args->endp, args->space,
835 &length);
66be6566
MD
836 if (error) {
837 if (error == ENAMETOOLONG)
838 error = E2BIG;
839 break;
840 }
2bd9d75c
DRJ
841 args->space -= length;
842 args->endp += length;
843 args->envc++;
984263bc
MD
844 }
845 }
2bd9d75c
DRJ
846 return (error);
847}
848
849void
850exec_free_args(struct image_args *args)
851{
2abfe22f 852 if (args->buf) {
18b3f457 853 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
854 args->buf = NULL;
855 }
984263bc
MD
856}
857
858/*
859 * Copy strings out to the new process address space, constructing
860 * new arg and env vector tables. Return a pointer to the base
861 * so that it can be used as the initial stack pointer.
862 */
863register_t *
2bd9d75c 864exec_copyout_strings(struct image_params *imgp)
984263bc 865{
17a7ca1f 866 int argc, envc, sgap;
984263bc
MD
867 char **vectp;
868 char *stringp, *destp;
869 register_t *stack_base;
870 struct ps_strings *arginfo;
871 int szsigcode;
872
873 /*
874 * Calculate string base and vector table pointers.
875 * Also deal with signal trampoline code for this exec type.
876 */
877 arginfo = (struct ps_strings *)PS_STRINGS;
878 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
17a7ca1f 879 if (stackgap_random != 0)
0ced1954 880 sgap = ALIGN(karc4random() & (stackgap_random - 1));
17a7ca1f
MD
881 else
882 sgap = 0;
883 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
2bd9d75c 884 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
984263bc
MD
885
886 /*
887 * install sigcode
888 */
889 if (szsigcode)
890 copyout(imgp->proc->p_sysent->sv_sigcode,
2bd9d75c 891 ((caddr_t)arginfo - szsigcode), szsigcode);
984263bc
MD
892
893 /*
894 * If we have a valid auxargs ptr, prepare some room
895 * on the stack.
7ec1918b 896 *
984263bc
MD
897 * The '+ 2' is for the null pointers at the end of each of the
898 * arg and env vector sets, and 'AT_COUNT*2' is room for the
899 * ELF Auxargs data.
900 */
7ec1918b
MD
901 if (imgp->auxargs) {
902 vectp = (char **)(destp - (imgp->args->argc +
903 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
904 } else {
2bd9d75c 905 vectp = (char **)(destp - (imgp->args->argc +
7ec1918b
MD
906 imgp->args->envc + 2) * sizeof(char*));
907 }
908
984263bc 909 /*
31bc42c8
MD
910 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
911 * be done in crt1.o. Note that GCC 3.x aligns the stack in main.
984263bc 912 */
984263bc
MD
913
914 /*
915 * vectp also becomes our initial stack base
916 */
917 stack_base = (register_t *)vectp;
918
2bd9d75c
DRJ
919 stringp = imgp->args->begin_argv;
920 argc = imgp->args->argc;
921 envc = imgp->args->envc;
984263bc
MD
922
923 /*
924 * Copy out strings - arguments and environment.
925 */
2bd9d75c 926 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
927
928 /*
929 * Fill in "ps_strings" struct for ps, w, etc.
930 */
931 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
932 suword(&arginfo->ps_nargvstr, argc);
933
934 /*
935 * Fill in argument portion of vector table.
936 */
937 for (; argc > 0; --argc) {
938 suword(vectp++, (long)(intptr_t)destp);
939 while (*stringp++ != 0)
940 destp++;
941 destp++;
942 }
943
7a44d1cb 944 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
945 suword(vectp++, 0);
946
947 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
948 suword(&arginfo->ps_nenvstr, envc);
949
950 /*
951 * Fill in environment portion of vector table.
952 */
953 for (; envc > 0; --envc) {
954 suword(vectp++, (long)(intptr_t)destp);
955 while (*stringp++ != 0)
956 destp++;
957 destp++;
958 }
959
960 /* end of vector table is a null pointer */
961 suword(vectp, 0);
962
963 return (stack_base);
964}
965
966/*
967 * Check permissions of file to execute.
968 * Return 0 for success or error code on failure.
969 */
970int
77153250 971exec_check_permissions(struct image_params *imgp)
984263bc
MD
972{
973 struct proc *p = imgp->proc;
974 struct vnode *vp = imgp->vp;
975 struct vattr *attr = imgp->attr;
976 int error;
977
978 /* Get file attributes */
87de5057 979 error = VOP_GETATTR(vp, attr);
984263bc
MD
980 if (error)
981 return (error);
982
983 /*
984 * 1) Check if file execution is disabled for the filesystem that this
985 * file resides on.
986 * 2) Insure that at least one execute bit is on - otherwise root
987 * will always succeed, and we don't want to happen unless the
988 * file really is executable.
989 * 3) Insure that the file is a regular file.
990 */
991 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
992 ((attr->va_mode & 0111) == 0) ||
993 (attr->va_type != VREG)) {
994 return (EACCES);
995 }
996
997 /*
998 * Zero length files can't be exec'd
999 */
1000 if (attr->va_size == 0)
1001 return (ENOEXEC);
1002
1003 /*
1004 * Check for execute permission to file based on current credentials.
1005 */
90a83564 1006 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
1007 if (error)
1008 return (error);
1009
1010 /*
1011 * Check number of open-for-writes on the file and deny execution
1012 * if there are any.
1013 */
1014 if (vp->v_writecount)
1015 return (ETXTBSY);
1016
1017 /*
7540ab49
MD
1018 * Call filesystem specific open routine, which allows us to read,
1019 * write, and mmap the file. Without the VOP_OPEN we can only
1020 * stat the file.
984263bc 1021 */
87de5057 1022 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
1023 if (error)
1024 return (error);
1025
1026 return (0);
1027}
1028
1029/*
1030 * Exec handler registration
1031 */
1032int
77153250 1033exec_register(const struct execsw *execsw_arg)
984263bc
MD
1034{
1035 const struct execsw **es, **xs, **newexecsw;
1036 int count = 2; /* New slot and trailing NULL */
1037
1038 if (execsw)
1039 for (es = execsw; *es; es++)
1040 count++;
efda3bd0 1041 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1042 xs = newexecsw;
1043 if (execsw)
1044 for (es = execsw; *es; es++)
1045 *xs++ = *es;
1046 *xs++ = execsw_arg;
1047 *xs = NULL;
1048 if (execsw)
efda3bd0 1049 kfree(execsw, M_TEMP);
984263bc
MD
1050 execsw = newexecsw;
1051 return 0;
1052}
1053
1054int
77153250 1055exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1056{
1057 const struct execsw **es, **xs, **newexecsw;
1058 int count = 1;
1059
1060 if (execsw == NULL)
ae7e74d4 1061 panic("unregister with no handlers left?");
984263bc
MD
1062
1063 for (es = execsw; *es; es++) {
1064 if (*es == execsw_arg)
1065 break;
1066 }
1067 if (*es == NULL)
1068 return ENOENT;
1069 for (es = execsw; *es; es++)
1070 if (*es != execsw_arg)
1071 count++;
efda3bd0 1072 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1073 xs = newexecsw;
1074 for (es = execsw; *es; es++)
1075 if (*es != execsw_arg)
1076 *xs++ = *es;
1077 *xs = NULL;
1078 if (execsw)
efda3bd0 1079 kfree(execsw, M_TEMP);
984263bc
MD
1080 execsw = newexecsw;
1081 return 0;
1082}