kernel - Fix a 'vm_page_unhold: hold count < 0' panic in kern_execve()
[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
SZ
114 int cluster_limit;
115
116 cluster_limit = 16; /* up to this many objects */
18b3f457
MD
117 exec_objcache = objcache_create_mbacked(
118 M_EXECARGS, PATH_MAX + ARG_MAX,
765b1ae0 119 &cluster_limit, 8,
18b3f457
MD
120 NULL, NULL, NULL);
121}
122SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
123
124/*
17a7ca1f
MD
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 183 register_t *stack_base;
19bfc8ab 184 struct pargs *pa;
6fa9e71a
MD
185 struct sigacts *ops;
186 struct sigacts *nps;
984263bc
MD
187 int error, len, i;
188 struct image_params image_params, *imgp;
189 struct vattr attr;
402ed7e1 190 int (*img_first) (struct image_params *);
984263bc 191
2bd9d75c 192 if (debug_execve_args) {
6ea70f76 193 kprintf("%s()\n", __func__);
2bd9d75c
DRJ
194 print_execve_args(args);
195 }
196
dadab5e9 197 KKASSERT(p);
19bfc8ab 198 lwkt_gettoken(&p->p_token);
984263bc
MD
199 imgp = &image_params;
200
201 /*
dc52e1cc
MD
202 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
203 * no modifications to the process at all until we get there.
204 *
205 * Note that multiple threads may be trying to exec at the same
206 * time. exec_new_vmspace() handles that too.
984263bc 207 */
984263bc
MD
208
209 /*
210 * Initialize part of the common data
211 */
212 imgp->proc = p;
2bd9d75c 213 imgp->args = args;
984263bc 214 imgp->attr = &attr;
984263bc 215 imgp->entry_addr = 0;
29802dbb 216 imgp->resident = 0;
984263bc
MD
217 imgp->vmspace_destroyed = 0;
218 imgp->interpreted = 0;
2abfe22f 219 imgp->interpreter_name[0] = 0;
984263bc
MD
220 imgp->auxargs = NULL;
221 imgp->vp = NULL;
222 imgp->firstpage = NULL;
223 imgp->ps_strings = 0;
adc42cf3
JM
224 imgp->execpath = imgp->freepath = NULL;
225 imgp->execpathp = 0;
18f40545 226 imgp->image_header = NULL;
2bd9d75c
DRJ
227
228interpret:
984263bc
MD
229
230 /*
fad57d0e
MD
231 * Translate the file name to a vnode. Unlock the cache entry to
232 * improve parallelism for programs exec'd in parallel.
984263bc 233 */
fad57d0e
MD
234 if ((error = nlookup(nd)) != 0)
235 goto exec_fail;
28623bf9 236 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
fad57d0e
MD
237 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
238 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 239 cache_unlock(&nd->nl_nch);
fad57d0e 240 if (error)
984263bc 241 goto exec_fail;
984263bc
MD
242
243 /*
246693ac
AHJ
244 * Check file permissions (also 'opens' file).
245 * Include also the top level mount in the check.
984263bc 246 */
246693ac 247 error = exec_check_permissions(imgp, nd->nl_nch.mount);
984263bc 248 if (error) {
a11aaa81 249 vn_unlock(imgp->vp);
984263bc
MD
250 goto exec_fail_dealloc;
251 }
252
253 error = exec_map_first_page(imgp);
a11aaa81 254 vn_unlock(imgp->vp);
984263bc
MD
255 if (error)
256 goto exec_fail_dealloc;
257
315b8b8b
JM
258 imgp->proc->p_osrel = 0;
259
2bd9d75c 260 if (debug_execve_args && imgp->interpreted) {
6ea70f76
SW
261 kprintf(" target is interpreted -- recursive pass\n");
262 kprintf(" interpreter: %s\n", imgp->interpreter_name);
2bd9d75c
DRJ
263 print_execve_args(args);
264 }
265
984263bc
MD
266 /*
267 * If the current process has a special image activator it
268 * wants to try first, call it. For example, emulating shell
269 * scripts differently.
270 */
271 error = -1;
272 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
273 error = img_first(imgp);
274
275 /*
29802dbb
MD
276 * If the vnode has a registered vmspace, exec the vmspace
277 */
278 if (error == -1 && imgp->vp->v_resident) {
279 error = exec_resident_imgact(imgp);
280 }
281
282 /*
984263bc
MD
283 * Loop through the list of image activators, calling each one.
284 * An activator returns -1 if there is no match, 0 on success,
285 * and an error otherwise.
286 */
287 for (i = 0; error == -1 && execsw[i]; ++i) {
288 if (execsw[i]->ex_imgact == NULL ||
289 execsw[i]->ex_imgact == img_first) {
290 continue;
291 }
292 error = (*execsw[i]->ex_imgact)(imgp);
293 }
294
295 if (error) {
296 if (error == -1)
297 error = ENOEXEC;
298 goto exec_fail_dealloc;
299 }
300
301 /*
302 * Special interpreter operation, cleanup and loop up to try to
303 * activate the interpreter.
304 */
305 if (imgp->interpreted) {
306 exec_unmap_first_page(imgp);
fad57d0e
MD
307 nlookup_done(nd);
308 vrele(imgp->vp);
309 imgp->vp = NULL;
310 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
311 NLC_FOLLOW);
312 if (error)
313 goto exec_fail;
984263bc
MD
314 goto interpret;
315 }
316
317 /*
adc42cf3
JM
318 * Do the best to calculate the full path to the image file
319 */
320 if (imgp->auxargs != NULL &&
321 ((args->fname != NULL && args->fname[0] == '/') ||
322 vn_fullpath(imgp->proc,
323 imgp->vp,
324 &imgp->execpath,
325 &imgp->freepath,
326 0) != 0))
327 imgp->execpath = args->fname;
328
329 /*
984263bc
MD
330 * Copy out strings (args and env) and initialize stack base
331 */
332 stack_base = exec_copyout_strings(imgp);
333 p->p_vmspace->vm_minsaddr = (char *)stack_base;
334
335 /*
336 * If custom stack fixup routine present for this process
29802dbb
MD
337 * let it do the stack setup. If we are running a resident
338 * image there is no auxinfo or other image activator context
339 * so don't try to add fixups to the stack.
340 *
984263bc
MD
341 * Else stuff argument count as first item on stack
342 */
29802dbb 343 if (p->p_sysent->sv_fixup && imgp->resident == 0)
984263bc
MD
344 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
345 else
2bd9d75c 346 suword(--stack_base, imgp->args->argc);
984263bc
MD
347
348 /*
349 * For security and other reasons, the file descriptor table cannot
350 * be shared after an exec.
351 */
352 if (p->p_fd->fd_refcnt > 1) {
353 struct filedesc *tmp;
354
2994659f
VS
355 error = fdcopy(p, &tmp);
356 if (error != 0)
357 goto exec_fail;
0a4a9c77 358 fdfree(p, tmp);
984263bc
MD
359 }
360
361 /*
362 * For security and other reasons, signal handlers cannot
363 * be shared after an exec. The new proces gets a copy of the old
364 * handlers. In execsigs(), the new process will have its signals
365 * reset.
366 */
6fa9e71a
MD
367 ops = p->p_sigacts;
368 if (ops->ps_refcnt > 1) {
369 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
370 bcopy(ops, nps, sizeof(*nps));
371 refcount_init(&nps->ps_refcnt, 1);
372 p->p_sigacts = nps;
373 if (refcount_release(&ops->ps_refcnt)) {
374 kfree(ops, M_SUBPROC);
375 ops = NULL;
376 }
984263bc
MD
377 }
378
0daa37a5
MD
379 /*
380 * For security and other reasons virtual kernels cannot be
381 * inherited by an exec. This also allows a virtual kernel
382 * to fork/exec unrelated applications.
383 */
4a22e893
MD
384 if (p->p_vkernel)
385 vkernel_exit(p);
0daa37a5 386
984263bc
MD
387 /* Stop profiling */
388 stopprofclock(p);
389
390 /* close files on exec */
391 fdcloseexec(p);
392
393 /* reset caught signals */
394 execsigs(p);
395
396 /* name this process - nameiexec(p, ndp) */
28623bf9
MD
397 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
398 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
984263bc 399 p->p_comm[len] = 0;
08f2f1bb 400 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
984263bc
MD
401
402 /*
403 * mark as execed, wakeup the process that vforked (if any) and tell
404 * it that it now has its own resources back
405 */
406 p->p_flag |= P_EXEC;
407 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
408 p->p_flag &= ~P_PPWAIT;
409 wakeup((caddr_t)p->p_pptr);
410 }
411
412 /*
413 * Implement image setuid/setgid.
414 *
415 * Don't honor setuid/setgid if the filesystem prohibits it or if
416 * the process is being traced.
417 */
418 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
419 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
420 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
421 (p->p_flag & P_TRACED) == 0) {
422 /*
423 * Turn off syscall tracing for set-id programs, except for
424 * root. Record any set-id flags first to make sure that
425 * we do not regain any tracing during a possible block.
426 */
41c20dac 427 setsugid();
5c627295
MD
428 if (p->p_tracenode && ktrace_suid == 0 &&
429 priv_check(td, PRIV_ROOT) != 0) {
29f58392
MD
430 ktrdestroy(&p->p_tracenode);
431 p->p_traceflag = 0;
984263bc
MD
432 }
433 /* Close any file descriptors 0..2 that reference procfs */
434 setugidsafety(p);
435 /* Make sure file descriptors 0..2 are in use. */
f3a2d8c4 436 error = fdcheckstd(lp);
984263bc
MD
437 if (error != 0)
438 goto exec_fail_dealloc;
439 /*
440 * Set the new credentials.
441 */
e9a372eb 442 cratom(&p->p_ucred);
984263bc 443 if (attr.va_mode & VSUID)
41c20dac 444 change_euid(attr.va_uid);
984263bc
MD
445 if (attr.va_mode & VSGID)
446 p->p_ucred->cr_gid = attr.va_gid;
98a7f915
MD
447
448 /*
449 * Clear local varsym variables
450 */
451 varsymset_clean(&p->p_varsymset);
984263bc 452 } else {
41c20dac
MD
453 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
454 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
984263bc
MD
455 p->p_flag &= ~P_SUGID;
456 }
457
458 /*
459 * Implement correct POSIX saved-id behavior.
460 */
115ccd83
MD
461 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
462 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
463 cratom(&p->p_ucred);
464 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
465 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
466 }
984263bc
MD
467
468 /*
469 * Store the vp for use in procfs
470 */
471 if (p->p_textvp) /* release old reference */
472 vrele(p->p_textvp);
fad57d0e
MD
473 p->p_textvp = imgp->vp;
474 vref(p->p_textvp);
984263bc 475
8ba5f7ef
AH
476 /* Release old namecache handle to text file */
477 if (p->p_textnch.ncp)
478 cache_drop(&p->p_textnch);
479
480 if (nd->nl_nch.mount)
481 cache_copy(&nd->nl_nch, &p->p_textnch);
482
984263bc
MD
483 /*
484 * Notify others that we exec'd, and clear the P_INEXEC flag
485 * as we're now a bona fide freshly-execed process.
486 */
487 KNOTE(&p->p_klist, NOTE_EXEC);
488 p->p_flag &= ~P_INEXEC;
489
490 /*
491 * If tracing the process, trap to debugger so breakpoints
492 * can be set before the program executes.
493 */
494 STOPEVENT(p, S_EXEC, 0);
495
496 if (p->p_flag & P_TRACED)
84204577 497 ksignal(p, SIGTRAP);
984263bc
MD
498
499 /* clear "fork but no exec" flag, as we _are_ execing */
500 p->p_acflag &= ~AFORK;
501
502 /* Set values passed into the program in registers. */
08f2f1bb 503 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
984263bc
MD
504 imgp->ps_strings);
505
349433c9
MD
506 /* Set the access time on the vnode */
507 vn_mark_atime(imgp->vp, td);
508
19bfc8ab
MD
509 /*
510 * Free any previous argument cache
511 */
512 pa = p->p_args;
984263bc 513 p->p_args = NULL;
19bfc8ab
MD
514 if (pa && refcount_release(&pa->ar_ref)) {
515 kfree(pa, M_PARGS);
516 pa = NULL;
517 }
984263bc 518
19bfc8ab
MD
519 /*
520 * Cache arguments if they fit inside our allowance
521 */
2bd9d75c 522 i = imgp->args->begin_envv - imgp->args->begin_argv;
19bfc8ab
MD
523 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
524 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
525 refcount_init(&pa->ar_ref, 1);
526 pa->ar_length = i;
527 bcopy(imgp->args->begin_argv, pa->ar_args, i);
528 KKASSERT(p->p_args == NULL);
529 p->p_args = pa;
984263bc
MD
530 }
531
532exec_fail_dealloc:
533
534 /*
535 * free various allocated resources
536 */
537 if (imgp->firstpage)
538 exec_unmap_first_page(imgp);
539
984263bc 540 if (imgp->vp) {
984263bc 541 vrele(imgp->vp);
fad57d0e 542 imgp->vp = NULL;
984263bc
MD
543 }
544
a6f89c72
DR
545 if (error == 0) {
546 ++mycpu->gd_cnt.v_exec;
19bfc8ab 547 lwkt_reltoken(&p->p_token);
984263bc 548 return (0);
a6f89c72 549 }
984263bc 550
adc42cf3
JM
551 if (imgp->freepath)
552 kfree(imgp->freepath, M_TEMP);
553
984263bc 554exec_fail:
dc52e1cc
MD
555 /*
556 * we're done here, clear P_INEXEC if we were the ones that
557 * set it. Otherwise if vmspace_destroyed is still set we
558 * raced another thread and that thread is responsible for
559 * clearing it.
560 */
561 if (imgp->vmspace_destroyed & 2)
562 p->p_flag &= ~P_INEXEC;
19bfc8ab 563 lwkt_reltoken(&p->p_token);
984263bc 564 if (imgp->vmspace_destroyed) {
5417cd57
SS
565 /*
566 * Sorry, no more process anymore. exit gracefully.
567 * However we can't die right here, because our
568 * caller might have to clean up, so indicate a
90947cb1 569 * lethal error by returning -1.
5417cd57
SS
570 */
571 return(-1);
984263bc
MD
572 } else {
573 return(error);
574 }
575}
576
2bd9d75c
DRJ
577/*
578 * execve() system call.
3919ced0
MD
579 *
580 * MPALMOSTSAFE
2bd9d75c
DRJ
581 */
582int
753fd850 583sys_execve(struct execve_args *uap)
2bd9d75c 584{
fad57d0e 585 struct nlookupdata nd;
2bd9d75c
DRJ
586 struct image_args args;
587 int error;
588
118cec38 589 bzero(&args, sizeof(args));
3919ced0
MD
590
591 get_mplock();
592 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
fad57d0e
MD
593 if (error == 0) {
594 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
595 uap->argv, uap->envv);
596 }
2abfe22f
MD
597 if (error == 0)
598 error = kern_execve(&nd, &args);
fad57d0e 599 nlookup_done(&nd);
2bd9d75c
DRJ
600 exec_free_args(&args);
601
5417cd57 602 if (error < 0) {
90947cb1 603 /* We hit a lethal error condition. Let's die now. */
5417cd57
SS
604 exit1(W_EXITCODE(0, SIGABRT));
605 /* NOTREACHED */
606 }
3919ced0 607 rel_mplock();
5417cd57 608
2bd9d75c
DRJ
609 /*
610 * The syscall result is returned in registers to the new program.
611 * Linux will register %edx as an atexit function and we must be
612 * sure to set it to 0. XXX
613 */
614 if (error == 0)
615 uap->sysmsg_result64 = 0;
616
617 return (error);
618}
619
984263bc 620int
7c34d798 621exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
5c5185ae 622 struct lwbuf **plwb, const char **pdata)
984263bc 623{
1b9d3514
MD
624 int rv;
625 vm_page_t ma;
06ecca5a 626 vm_page_t m;
984263bc
MD
627 vm_object_t object;
628
fad57d0e 629 /*
7540ab49 630 * The file has to be mappable.
fad57d0e 631 */
7540ab49 632 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 633 return (EIO);
984263bc 634
7c34d798
SS
635 if (pageno >= object->size)
636 return (EIO);
637
7c34d798 638 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
77912481 639
573fb415 640 lwkt_gettoken(&vm_token);
1b9d3514
MD
641 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
642 ma = m;
984263bc 643
06ecca5a
MD
644 /*
645 * get_pages unbusies all the requested pages except the
a55afca2
MD
646 * primary page (at index 0 in this case). The primary
647 * page may have been wired during the pagein (e.g. by
648 * the buffer cache) so vnode_pager_freepage() must be
649 * used to properly release it.
06ecca5a 650 */
1b9d3514 651 rv = vm_pager_get_page(object, &ma, 1);
7c34d798 652 m = vm_page_lookup(object, pageno);
984263bc 653
06ecca5a
MD
654 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
655 if (m) {
656 vm_page_protect(m, VM_PROT_NONE);
a55afca2 657 vnode_pager_freepage(m);
984263bc 658 }
573fb415 659 lwkt_reltoken(&vm_token);
984263bc
MD
660 return EIO;
661 }
662 }
573fb415 663 vm_page_hold(m); /* requires vm_token to be held */
06ecca5a 664 vm_page_wakeup(m); /* unbusy the page */
573fb415 665 lwkt_reltoken(&vm_token);
984263bc 666
7a683a24 667 *plwb = lwbuf_alloc(m, *plwb);
5c5185ae 668 *pdata = (void *)lwbuf_kva(*plwb);
7c34d798
SS
669
670 return (0);
671}
672
3caf6b88
MD
673/*
674 * Map the first page of an executable image.
675 *
676 * NOTE: If the mapping fails we have to NULL-out firstpage which may
677 * still be pointing to our supplied lwp structure.
678 */
7c34d798
SS
679int
680exec_map_first_page(struct image_params *imgp)
681{
682 int err;
683
684 if (imgp->firstpage)
685 exec_unmap_first_page(imgp);
686
7a683a24 687 imgp->firstpage = &imgp->firstpage_cache;
7c34d798
SS
688 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
689
3caf6b88
MD
690 if (err) {
691 imgp->firstpage = NULL;
7c34d798 692 return err;
3caf6b88 693 }
984263bc
MD
694
695 return 0;
696}
697
698void
5c5185ae 699exec_unmap_page(struct lwbuf *lwb)
984263bc 700{
18f40545
MD
701 vm_page_t m;
702
5fd012e0 703 crit_enter();
5c5185ae
SG
704 if (lwb != NULL) {
705 m = lwbuf_page(lwb);
706 lwbuf_free(lwb);
e4ba7818 707 vm_page_unhold(m);
984263bc 708 }
5fd012e0 709 crit_exit();
984263bc
MD
710}
711
7c34d798
SS
712void
713exec_unmap_first_page(struct image_params *imgp)
714{
715 exec_unmap_page(imgp->firstpage);
716 imgp->firstpage = NULL;
717 imgp->image_header = NULL;
718}
719
984263bc
MD
720/*
721 * Destroy old address space, and allocate a new stack
722 * The new stack is only SGROWSIZ large because it is grown
723 * automatically in trap.c.
dc52e1cc
MD
724 *
725 * This is the point of no return.
984263bc
MD
726 */
727int
29802dbb 728exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 729{
984263bc
MD
730 struct vmspace *vmspace = imgp->proc->p_vmspace;
731 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 732 struct proc *p;
29802dbb 733 vm_map_t map;
dc52e1cc 734 int error;
984263bc 735
dc52e1cc
MD
736 /*
737 * Indicate that we cannot gracefully error out any more, kill
738 * any other threads present, and set P_INEXEC to indicate that
739 * we are now messing with the process structure proper.
740 *
741 * If killalllwps() races return an error which coupled with
742 * vmspace_destroyed will cause us to exit. This is what we
743 * want since another thread is patiently waiting for us to exit
744 * in that case.
745 */
746 p = curproc;
984263bc
MD
747 imgp->vmspace_destroyed = 1;
748
dc52e1cc
MD
749 if (curthread->td_proc->p_nthreads > 1) {
750 error = killalllwps(1);
751 if (error)
752 return (error);
753 }
754 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
755 p->p_flag |= P_INEXEC;
08f2f1bb
SS
756
757 /*
984263bc
MD
758 * Blow away entire process VM, if address space not shared,
759 * otherwise, create a new VM space so that other threads are
29802dbb
MD
760 * not disrupted. If we are execing a resident vmspace we
761 * create a duplicate of it and remap the stack.
239b4df9
MD
762 *
763 * The exitingcnt test is not strictly necessary but has been
764 * included for code sanity (to make the code more deterministic).
984263bc 765 */
29802dbb
MD
766 map = &vmspace->vm_map;
767 if (vmcopy) {
768 vmspace_exec(imgp->proc, vmcopy);
769 vmspace = imgp->proc->p_vmspace;
770 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
771 map = &vmspace->vm_map;
e3161323
MD
772 } else if (vmspace->vm_sysref.refcnt == 1 &&
773 vmspace->vm_exitingcnt == 0) {
fef0fdf2 774 shmexit(vmspace);
239b4df9 775 if (vmspace->vm_upcalls)
08f2f1bb 776 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08
MD
777 pmap_remove_pages(vmspace_pmap(vmspace),
778 0, VM_MAX_USER_ADDRESS);
779 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 780 } else {
29802dbb 781 vmspace_exec(imgp->proc, NULL);
984263bc
MD
782 vmspace = imgp->proc->p_vmspace;
783 map = &vmspace->vm_map;
784 }
785
786 /* Allocate a new stack */
787 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
c809941b 788 0, VM_PROT_ALL, VM_PROT_ALL, 0);
984263bc
MD
789 if (error)
790 return (error);
791
792 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
793 * VM_STACK case, but they are still used to monitor the size of the
794 * process stack so we can check the stack rlimit.
795 */
796 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
797 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
798
799 return(0);
800}
801
802/*
803 * Copy out argument and environment strings from the old process
804 * address space into the temporary string buffer.
805 */
806int
2bd9d75c 807exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 808 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 809{
984263bc 810 char *argp, *envp;
2bd9d75c 811 int error = 0;
984263bc
MD
812 size_t length;
813
18b3f457 814 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
815 if (args->buf == NULL)
816 return (ENOMEM);
817 args->begin_argv = args->buf;
818 args->endp = args->begin_argv;
819 args->space = ARG_MAX;
2bd9d75c
DRJ
820
821 args->fname = args->buf + ARG_MAX;
822
984263bc 823 /*
2bd9d75c 824 * Copy the file name.
984263bc 825 */
2bd9d75c
DRJ
826 if (segflg == PATH_SYSSPACE) {
827 error = copystr(fname, args->fname, PATH_MAX, &length);
828 } else if (segflg == PATH_USERSPACE) {
829 error = copyinstr(fname, args->fname, PATH_MAX, &length);
830 }
984263bc 831
2bd9d75c 832 /*
66be6566
MD
833 * Extract argument strings. argv may not be NULL. The argv
834 * array is terminated by a NULL entry. We special-case the
835 * situation where argv[0] is NULL by passing { filename, NULL }
836 * to the new program to guarentee that the interpreter knows what
837 * file to open in case we exec an interpreted file. Note that
838 * a NULL argv[0] terminates the argv[] array.
839 *
840 * XXX the special-casing of argv[0] is historical and needs to be
841 * revisited.
2bd9d75c 842 */
66be6566
MD
843 if (argv == NULL)
844 error = EFAULT;
845 if (error == 0) {
846 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
847 if (argp == (caddr_t)-1) {
2bd9d75c 848 error = EFAULT;
66be6566 849 break;
2bd9d75c
DRJ
850 }
851 error = copyinstr(argp, args->endp,
2abfe22f 852 args->space, &length);
66be6566
MD
853 if (error) {
854 if (error == ENAMETOOLONG)
855 error = E2BIG;
856 break;
857 }
2bd9d75c
DRJ
858 args->space -= length;
859 args->endp += length;
860 args->argc++;
984263bc 861 }
66be6566
MD
862 if (args->argc == 0 && error == 0) {
863 length = strlen(args->fname) + 1;
864 if (length > args->space) {
865 error = E2BIG;
866 } else {
867 bcopy(args->fname, args->endp, length);
868 args->space -= length;
869 args->endp += length;
870 args->argc++;
871 }
872 }
984263bc
MD
873 }
874
2bd9d75c 875 args->begin_envv = args->endp;
984263bc
MD
876
877 /*
66be6566 878 * extract environment strings. envv may be NULL.
984263bc 879 */
2abfe22f 880 if (envv && error == 0) {
984263bc 881 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
882 if (envp == (caddr_t) -1) {
883 error = EFAULT;
66be6566 884 break;
984263bc 885 }
2bd9d75c
DRJ
886 error = copyinstr(envp, args->endp, args->space,
887 &length);
66be6566
MD
888 if (error) {
889 if (error == ENAMETOOLONG)
890 error = E2BIG;
891 break;
892 }
2bd9d75c
DRJ
893 args->space -= length;
894 args->endp += length;
895 args->envc++;
984263bc
MD
896 }
897 }
2bd9d75c
DRJ
898 return (error);
899}
900
901void
902exec_free_args(struct image_args *args)
903{
2abfe22f 904 if (args->buf) {
18b3f457 905 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
906 args->buf = NULL;
907 }
984263bc
MD
908}
909
910/*
911 * Copy strings out to the new process address space, constructing
912 * new arg and env vector tables. Return a pointer to the base
913 * so that it can be used as the initial stack pointer.
914 */
915register_t *
2bd9d75c 916exec_copyout_strings(struct image_params *imgp)
984263bc 917{
17a7ca1f 918 int argc, envc, sgap;
984263bc
MD
919 char **vectp;
920 char *stringp, *destp;
921 register_t *stack_base;
922 struct ps_strings *arginfo;
adc42cf3 923 size_t execpath_len;
984263bc
MD
924 int szsigcode;
925
926 /*
927 * Calculate string base and vector table pointers.
928 * Also deal with signal trampoline code for this exec type.
929 */
adc42cf3
JM
930 if (imgp->execpath != NULL && imgp->auxargs != NULL)
931 execpath_len = strlen(imgp->execpath) + 1;
932 else
933 execpath_len = 0;
984263bc
MD
934 arginfo = (struct ps_strings *)PS_STRINGS;
935 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
17a7ca1f 936 if (stackgap_random != 0)
0ced1954 937 sgap = ALIGN(karc4random() & (stackgap_random - 1));
17a7ca1f
MD
938 else
939 sgap = 0;
940 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
adc42cf3 941 roundup(execpath_len, sizeof(char *)) -
2bd9d75c 942 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
984263bc
MD
943
944 /*
945 * install sigcode
946 */
947 if (szsigcode)
948 copyout(imgp->proc->p_sysent->sv_sigcode,
2bd9d75c 949 ((caddr_t)arginfo - szsigcode), szsigcode);
984263bc
MD
950
951 /*
adc42cf3
JM
952 * Copy the image path for the rtld
953 */
954 if (execpath_len != 0) {
955 imgp->execpathp = (uintptr_t)arginfo
956 - szsigcode
957 - execpath_len;
958 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
959 }
960
961 /*
984263bc
MD
962 * If we have a valid auxargs ptr, prepare some room
963 * on the stack.
7ec1918b 964 *
984263bc
MD
965 * The '+ 2' is for the null pointers at the end of each of the
966 * arg and env vector sets, and 'AT_COUNT*2' is room for the
967 * ELF Auxargs data.
968 */
7ec1918b
MD
969 if (imgp->auxargs) {
970 vectp = (char **)(destp - (imgp->args->argc +
adc42cf3
JM
971 imgp->args->envc + 2 + (AT_COUNT * 2) + execpath_len) *
972 sizeof(char*));
7ec1918b 973 } else {
2bd9d75c 974 vectp = (char **)(destp - (imgp->args->argc +
7ec1918b
MD
975 imgp->args->envc + 2) * sizeof(char*));
976 }
977
984263bc 978 /*
31bc42c8
MD
979 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
980 * be done in crt1.o. Note that GCC 3.x aligns the stack in main.
984263bc 981 */
984263bc
MD
982
983 /*
984 * vectp also becomes our initial stack base
985 */
986 stack_base = (register_t *)vectp;
987
2bd9d75c
DRJ
988 stringp = imgp->args->begin_argv;
989 argc = imgp->args->argc;
990 envc = imgp->args->envc;
984263bc
MD
991
992 /*
993 * Copy out strings - arguments and environment.
994 */
2bd9d75c 995 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
996
997 /*
998 * Fill in "ps_strings" struct for ps, w, etc.
999 */
1000 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1001 suword(&arginfo->ps_nargvstr, argc);
1002
1003 /*
1004 * Fill in argument portion of vector table.
1005 */
1006 for (; argc > 0; --argc) {
1007 suword(vectp++, (long)(intptr_t)destp);
1008 while (*stringp++ != 0)
1009 destp++;
1010 destp++;
1011 }
1012
7a44d1cb 1013 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
1014 suword(vectp++, 0);
1015
1016 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1017 suword(&arginfo->ps_nenvstr, envc);
1018
1019 /*
1020 * Fill in environment portion of vector table.
1021 */
1022 for (; envc > 0; --envc) {
1023 suword(vectp++, (long)(intptr_t)destp);
1024 while (*stringp++ != 0)
1025 destp++;
1026 destp++;
1027 }
1028
1029 /* end of vector table is a null pointer */
1030 suword(vectp, 0);
1031
1032 return (stack_base);
1033}
1034
1035/*
1036 * Check permissions of file to execute.
1037 * Return 0 for success or error code on failure.
1038 */
1039int
246693ac 1040exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
984263bc
MD
1041{
1042 struct proc *p = imgp->proc;
1043 struct vnode *vp = imgp->vp;
1044 struct vattr *attr = imgp->attr;
1045 int error;
1046
1047 /* Get file attributes */
87de5057 1048 error = VOP_GETATTR(vp, attr);
984263bc
MD
1049 if (error)
1050 return (error);
1051
1052 /*
1053 * 1) Check if file execution is disabled for the filesystem that this
1054 * file resides on.
1055 * 2) Insure that at least one execute bit is on - otherwise root
1056 * will always succeed, and we don't want to happen unless the
1057 * file really is executable.
1058 * 3) Insure that the file is a regular file.
1059 */
1060 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
246693ac 1061 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
984263bc
MD
1062 ((attr->va_mode & 0111) == 0) ||
1063 (attr->va_type != VREG)) {
1064 return (EACCES);
1065 }
1066
1067 /*
1068 * Zero length files can't be exec'd
1069 */
1070 if (attr->va_size == 0)
1071 return (ENOEXEC);
1072
1073 /*
1074 * Check for execute permission to file based on current credentials.
1075 */
90a83564 1076 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
1077 if (error)
1078 return (error);
1079
1080 /*
1081 * Check number of open-for-writes on the file and deny execution
1082 * if there are any.
1083 */
1084 if (vp->v_writecount)
1085 return (ETXTBSY);
1086
1087 /*
7540ab49
MD
1088 * Call filesystem specific open routine, which allows us to read,
1089 * write, and mmap the file. Without the VOP_OPEN we can only
1090 * stat the file.
984263bc 1091 */
87de5057 1092 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
1093 if (error)
1094 return (error);
1095
1096 return (0);
1097}
1098
1099/*
1100 * Exec handler registration
1101 */
1102int
77153250 1103exec_register(const struct execsw *execsw_arg)
984263bc
MD
1104{
1105 const struct execsw **es, **xs, **newexecsw;
1106 int count = 2; /* New slot and trailing NULL */
1107
1108 if (execsw)
1109 for (es = execsw; *es; es++)
1110 count++;
efda3bd0 1111 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1112 xs = newexecsw;
1113 if (execsw)
1114 for (es = execsw; *es; es++)
1115 *xs++ = *es;
1116 *xs++ = execsw_arg;
1117 *xs = NULL;
1118 if (execsw)
efda3bd0 1119 kfree(execsw, M_TEMP);
984263bc
MD
1120 execsw = newexecsw;
1121 return 0;
1122}
1123
1124int
77153250 1125exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1126{
1127 const struct execsw **es, **xs, **newexecsw;
1128 int count = 1;
1129
1130 if (execsw == NULL)
ae7e74d4 1131 panic("unregister with no handlers left?");
984263bc
MD
1132
1133 for (es = execsw; *es; es++) {
1134 if (*es == execsw_arg)
1135 break;
1136 }
1137 if (*es == NULL)
1138 return ENOENT;
1139 for (es = execsw; *es; es++)
1140 if (*es != execsw_arg)
1141 count++;
efda3bd0 1142 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1143 xs = newexecsw;
1144 for (es = execsw; *es; es++)
1145 if (*es != execsw_arg)
1146 *xs++ = *es;
1147 *xs = NULL;
1148 if (execsw)
efda3bd0 1149 kfree(execsw, M_TEMP);
984263bc
MD
1150 execsw = newexecsw;
1151 return 0;
1152}