kernel - Increase cluster_limit for exec args on machines w/ > 16 cpus
[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);
150 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val))
151 return (EINVAL);
152 stackgap_random = new_val;
153
154 return(0);
155}
156
157SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT,
158 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
159
2bd9d75c
DRJ
160void
161print_execve_args(struct image_args *args)
162{
163 char *cp;
164 int ndx;
165
166 cp = args->begin_argv;
167 for (ndx = 0; ndx < args->argc; ndx++) {
6ea70f76 168 kprintf("\targv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
169 while (*cp++ != '\0');
170 }
171 for (ndx = 0; ndx < args->envc; ndx++) {
6ea70f76 172 kprintf("\tenvv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
173 while (*cp++ != '\0');
174 }
175}
176
984263bc
MD
177/*
178 * Each of the items is a pointer to a `const struct execsw', hence the
179 * double pointer here.
180 */
181static const struct execsw **execsw;
182
5417cd57
SS
183/*
184 * Replace current vmspace with a new binary.
185 * Returns 0 on success, > 0 on recoverable error (use as errno).
90947cb1 186 * Returns -1 on lethal error which demands killing of the current
5417cd57
SS
187 * process!
188 */
984263bc 189int
fad57d0e 190kern_execve(struct nlookupdata *nd, struct image_args *args)
984263bc 191{
dadab5e9 192 struct thread *td = curthread;
08f2f1bb 193 struct lwp *lp = td->td_lwp;
dadab5e9 194 struct proc *p = td->td_proc;
a2ee730d 195 struct vnode *ovp;
984263bc 196 register_t *stack_base;
19bfc8ab 197 struct pargs *pa;
6fa9e71a
MD
198 struct sigacts *ops;
199 struct sigacts *nps;
984263bc
MD
200 int error, len, i;
201 struct image_params image_params, *imgp;
202 struct vattr attr;
402ed7e1 203 int (*img_first) (struct image_params *);
984263bc 204
2bd9d75c 205 if (debug_execve_args) {
6ea70f76 206 kprintf("%s()\n", __func__);
2bd9d75c
DRJ
207 print_execve_args(args);
208 }
209
dadab5e9 210 KKASSERT(p);
19bfc8ab 211 lwkt_gettoken(&p->p_token);
984263bc
MD
212 imgp = &image_params;
213
214 /*
dc52e1cc
MD
215 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
216 * no modifications to the process at all until we get there.
217 *
218 * Note that multiple threads may be trying to exec at the same
219 * time. exec_new_vmspace() handles that too.
984263bc 220 */
984263bc
MD
221
222 /*
223 * Initialize part of the common data
224 */
225 imgp->proc = p;
2bd9d75c 226 imgp->args = args;
984263bc 227 imgp->attr = &attr;
984263bc 228 imgp->entry_addr = 0;
29802dbb 229 imgp->resident = 0;
984263bc
MD
230 imgp->vmspace_destroyed = 0;
231 imgp->interpreted = 0;
2abfe22f 232 imgp->interpreter_name[0] = 0;
984263bc
MD
233 imgp->auxargs = NULL;
234 imgp->vp = NULL;
235 imgp->firstpage = NULL;
236 imgp->ps_strings = 0;
adc42cf3
JM
237 imgp->execpath = imgp->freepath = NULL;
238 imgp->execpathp = 0;
18f40545 239 imgp->image_header = NULL;
2bd9d75c
DRJ
240
241interpret:
984263bc
MD
242
243 /*
fad57d0e
MD
244 * Translate the file name to a vnode. Unlock the cache entry to
245 * improve parallelism for programs exec'd in parallel.
984263bc 246 */
fad57d0e
MD
247 if ((error = nlookup(nd)) != 0)
248 goto exec_fail;
28623bf9 249 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
fad57d0e
MD
250 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
251 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 252 cache_unlock(&nd->nl_nch);
fad57d0e 253 if (error)
984263bc 254 goto exec_fail;
984263bc
MD
255
256 /*
246693ac
AHJ
257 * Check file permissions (also 'opens' file).
258 * Include also the top level mount in the check.
984263bc 259 */
246693ac 260 error = exec_check_permissions(imgp, nd->nl_nch.mount);
984263bc 261 if (error) {
a11aaa81 262 vn_unlock(imgp->vp);
984263bc
MD
263 goto exec_fail_dealloc;
264 }
265
266 error = exec_map_first_page(imgp);
a11aaa81 267 vn_unlock(imgp->vp);
984263bc
MD
268 if (error)
269 goto exec_fail_dealloc;
270
315b8b8b
JM
271 imgp->proc->p_osrel = 0;
272
2bd9d75c 273 if (debug_execve_args && imgp->interpreted) {
6ea70f76
SW
274 kprintf(" target is interpreted -- recursive pass\n");
275 kprintf(" interpreter: %s\n", imgp->interpreter_name);
2bd9d75c
DRJ
276 print_execve_args(args);
277 }
278
984263bc
MD
279 /*
280 * If the current process has a special image activator it
281 * wants to try first, call it. For example, emulating shell
282 * scripts differently.
283 */
284 error = -1;
285 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
286 error = img_first(imgp);
287
288 /*
29802dbb
MD
289 * If the vnode has a registered vmspace, exec the vmspace
290 */
291 if (error == -1 && imgp->vp->v_resident) {
292 error = exec_resident_imgact(imgp);
293 }
294
295 /*
984263bc
MD
296 * Loop through the list of image activators, calling each one.
297 * An activator returns -1 if there is no match, 0 on success,
298 * and an error otherwise.
299 */
300 for (i = 0; error == -1 && execsw[i]; ++i) {
301 if (execsw[i]->ex_imgact == NULL ||
302 execsw[i]->ex_imgact == img_first) {
303 continue;
304 }
305 error = (*execsw[i]->ex_imgact)(imgp);
306 }
307
308 if (error) {
309 if (error == -1)
310 error = ENOEXEC;
311 goto exec_fail_dealloc;
312 }
313
314 /*
315 * Special interpreter operation, cleanup and loop up to try to
316 * activate the interpreter.
317 */
318 if (imgp->interpreted) {
319 exec_unmap_first_page(imgp);
fad57d0e
MD
320 nlookup_done(nd);
321 vrele(imgp->vp);
322 imgp->vp = NULL;
323 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
324 NLC_FOLLOW);
325 if (error)
326 goto exec_fail;
984263bc
MD
327 goto interpret;
328 }
329
330 /*
adc42cf3
JM
331 * Do the best to calculate the full path to the image file
332 */
333 if (imgp->auxargs != NULL &&
334 ((args->fname != NULL && args->fname[0] == '/') ||
335 vn_fullpath(imgp->proc,
336 imgp->vp,
337 &imgp->execpath,
338 &imgp->freepath,
339 0) != 0))
340 imgp->execpath = args->fname;
341
342 /*
984263bc
MD
343 * Copy out strings (args and env) and initialize stack base
344 */
345 stack_base = exec_copyout_strings(imgp);
346 p->p_vmspace->vm_minsaddr = (char *)stack_base;
347
348 /*
349 * If custom stack fixup routine present for this process
29802dbb
MD
350 * let it do the stack setup. If we are running a resident
351 * image there is no auxinfo or other image activator context
352 * so don't try to add fixups to the stack.
353 *
984263bc
MD
354 * Else stuff argument count as first item on stack
355 */
29802dbb 356 if (p->p_sysent->sv_fixup && imgp->resident == 0)
984263bc
MD
357 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
358 else
2bd9d75c 359 suword(--stack_base, imgp->args->argc);
984263bc
MD
360
361 /*
362 * For security and other reasons, the file descriptor table cannot
363 * be shared after an exec.
364 */
365 if (p->p_fd->fd_refcnt > 1) {
366 struct filedesc *tmp;
367
2994659f
VS
368 error = fdcopy(p, &tmp);
369 if (error != 0)
370 goto exec_fail;
0a4a9c77 371 fdfree(p, tmp);
984263bc
MD
372 }
373
374 /*
375 * For security and other reasons, signal handlers cannot
376 * be shared after an exec. The new proces gets a copy of the old
377 * handlers. In execsigs(), the new process will have its signals
378 * reset.
379 */
6fa9e71a
MD
380 ops = p->p_sigacts;
381 if (ops->ps_refcnt > 1) {
382 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
383 bcopy(ops, nps, sizeof(*nps));
384 refcount_init(&nps->ps_refcnt, 1);
385 p->p_sigacts = nps;
386 if (refcount_release(&ops->ps_refcnt)) {
387 kfree(ops, M_SUBPROC);
388 ops = NULL;
389 }
984263bc
MD
390 }
391
0daa37a5
MD
392 /*
393 * For security and other reasons virtual kernels cannot be
394 * inherited by an exec. This also allows a virtual kernel
395 * to fork/exec unrelated applications.
396 */
4a22e893
MD
397 if (p->p_vkernel)
398 vkernel_exit(p);
0daa37a5 399
984263bc
MD
400 /* Stop profiling */
401 stopprofclock(p);
402
403 /* close files on exec */
404 fdcloseexec(p);
405
406 /* reset caught signals */
407 execsigs(p);
408
409 /* name this process - nameiexec(p, ndp) */
28623bf9
MD
410 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
411 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
984263bc 412 p->p_comm[len] = 0;
08f2f1bb 413 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
984263bc
MD
414
415 /*
416 * mark as execed, wakeup the process that vforked (if any) and tell
417 * it that it now has its own resources back
418 */
4643740a
MD
419 p->p_flags |= P_EXEC;
420 if (p->p_pptr && (p->p_flags & P_PPWAIT)) {
421 p->p_flags &= ~P_PPWAIT;
984263bc
MD
422 wakeup((caddr_t)p->p_pptr);
423 }
424
425 /*
426 * Implement image setuid/setgid.
427 *
428 * Don't honor setuid/setgid if the filesystem prohibits it or if
429 * the process is being traced.
430 */
431 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
432 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
433 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
4643740a 434 (p->p_flags & P_TRACED) == 0) {
984263bc
MD
435 /*
436 * Turn off syscall tracing for set-id programs, except for
437 * root. Record any set-id flags first to make sure that
438 * we do not regain any tracing during a possible block.
439 */
41c20dac 440 setsugid();
5c627295
MD
441 if (p->p_tracenode && ktrace_suid == 0 &&
442 priv_check(td, PRIV_ROOT) != 0) {
29f58392
MD
443 ktrdestroy(&p->p_tracenode);
444 p->p_traceflag = 0;
984263bc
MD
445 }
446 /* Close any file descriptors 0..2 that reference procfs */
447 setugidsafety(p);
448 /* Make sure file descriptors 0..2 are in use. */
f3a2d8c4 449 error = fdcheckstd(lp);
984263bc
MD
450 if (error != 0)
451 goto exec_fail_dealloc;
452 /*
453 * Set the new credentials.
454 */
e9a372eb 455 cratom(&p->p_ucred);
984263bc 456 if (attr.va_mode & VSUID)
41c20dac 457 change_euid(attr.va_uid);
984263bc
MD
458 if (attr.va_mode & VSGID)
459 p->p_ucred->cr_gid = attr.va_gid;
98a7f915
MD
460
461 /*
462 * Clear local varsym variables
463 */
464 varsymset_clean(&p->p_varsymset);
984263bc 465 } else {
41c20dac
MD
466 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
467 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
4643740a 468 p->p_flags &= ~P_SUGID;
984263bc
MD
469 }
470
471 /*
472 * Implement correct POSIX saved-id behavior.
473 */
115ccd83
MD
474 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
475 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
476 cratom(&p->p_ucred);
477 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
478 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
479 }
984263bc
MD
480
481 /*
a2ee730d
MD
482 * Store the vp for use in procfs. Be sure to keep p_textvp
483 * consistent if we block during the switch-over.
984263bc 484 */
a2ee730d
MD
485 ovp = p->p_textvp;
486 vref(imgp->vp); /* ref new vp */
fad57d0e 487 p->p_textvp = imgp->vp;
a2ee730d
MD
488 if (ovp) /* release old vp */
489 vrele(ovp);
984263bc 490
8ba5f7ef
AH
491 /* Release old namecache handle to text file */
492 if (p->p_textnch.ncp)
493 cache_drop(&p->p_textnch);
494
495 if (nd->nl_nch.mount)
496 cache_copy(&nd->nl_nch, &p->p_textnch);
497
984263bc
MD
498 /*
499 * Notify others that we exec'd, and clear the P_INEXEC flag
500 * as we're now a bona fide freshly-execed process.
501 */
502 KNOTE(&p->p_klist, NOTE_EXEC);
4643740a 503 p->p_flags &= ~P_INEXEC;
984263bc
MD
504
505 /*
506 * If tracing the process, trap to debugger so breakpoints
507 * can be set before the program executes.
508 */
509 STOPEVENT(p, S_EXEC, 0);
510
4643740a 511 if (p->p_flags & P_TRACED)
84204577 512 ksignal(p, SIGTRAP);
984263bc
MD
513
514 /* clear "fork but no exec" flag, as we _are_ execing */
515 p->p_acflag &= ~AFORK;
516
517 /* Set values passed into the program in registers. */
08f2f1bb 518 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
a2ee730d 519 imgp->ps_strings);
984263bc 520
349433c9
MD
521 /* Set the access time on the vnode */
522 vn_mark_atime(imgp->vp, td);
523
19bfc8ab
MD
524 /*
525 * Free any previous argument cache
526 */
527 pa = p->p_args;
984263bc 528 p->p_args = NULL;
19bfc8ab
MD
529 if (pa && refcount_release(&pa->ar_ref)) {
530 kfree(pa, M_PARGS);
531 pa = NULL;
532 }
984263bc 533
19bfc8ab
MD
534 /*
535 * Cache arguments if they fit inside our allowance
536 */
2bd9d75c 537 i = imgp->args->begin_envv - imgp->args->begin_argv;
19bfc8ab
MD
538 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
539 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
540 refcount_init(&pa->ar_ref, 1);
541 pa->ar_length = i;
542 bcopy(imgp->args->begin_argv, pa->ar_args, i);
543 KKASSERT(p->p_args == NULL);
544 p->p_args = pa;
984263bc
MD
545 }
546
547exec_fail_dealloc:
548
549 /*
550 * free various allocated resources
551 */
552 if (imgp->firstpage)
553 exec_unmap_first_page(imgp);
554
984263bc 555 if (imgp->vp) {
984263bc 556 vrele(imgp->vp);
fad57d0e 557 imgp->vp = NULL;
984263bc
MD
558 }
559
bf52a6ac
MD
560 if (imgp->freepath)
561 kfree(imgp->freepath, M_TEMP);
562
a6f89c72
DR
563 if (error == 0) {
564 ++mycpu->gd_cnt.v_exec;
19bfc8ab 565 lwkt_reltoken(&p->p_token);
984263bc 566 return (0);
a6f89c72 567 }
984263bc
MD
568
569exec_fail:
dc52e1cc
MD
570 /*
571 * we're done here, clear P_INEXEC if we were the ones that
572 * set it. Otherwise if vmspace_destroyed is still set we
573 * raced another thread and that thread is responsible for
574 * clearing it.
575 */
576 if (imgp->vmspace_destroyed & 2)
4643740a 577 p->p_flags &= ~P_INEXEC;
19bfc8ab 578 lwkt_reltoken(&p->p_token);
984263bc 579 if (imgp->vmspace_destroyed) {
5417cd57
SS
580 /*
581 * Sorry, no more process anymore. exit gracefully.
582 * However we can't die right here, because our
583 * caller might have to clean up, so indicate a
90947cb1 584 * lethal error by returning -1.
5417cd57
SS
585 */
586 return(-1);
984263bc
MD
587 } else {
588 return(error);
589 }
590}
591
2bd9d75c
DRJ
592/*
593 * execve() system call.
594 */
595int
753fd850 596sys_execve(struct execve_args *uap)
2bd9d75c 597{
fad57d0e 598 struct nlookupdata nd;
2bd9d75c
DRJ
599 struct image_args args;
600 int error;
601
118cec38 602 bzero(&args, sizeof(args));
3919ced0 603
3919ced0 604 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
fad57d0e
MD
605 if (error == 0) {
606 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
607 uap->argv, uap->envv);
608 }
2abfe22f
MD
609 if (error == 0)
610 error = kern_execve(&nd, &args);
fad57d0e 611 nlookup_done(&nd);
2bd9d75c
DRJ
612 exec_free_args(&args);
613
5417cd57 614 if (error < 0) {
90947cb1 615 /* We hit a lethal error condition. Let's die now. */
5417cd57
SS
616 exit1(W_EXITCODE(0, SIGABRT));
617 /* NOTREACHED */
618 }
619
2bd9d75c
DRJ
620 /*
621 * The syscall result is returned in registers to the new program.
622 * Linux will register %edx as an atexit function and we must be
623 * sure to set it to 0. XXX
624 */
625 if (error == 0)
626 uap->sysmsg_result64 = 0;
627
628 return (error);
629}
630
984263bc 631int
7c34d798 632exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
5c5185ae 633 struct lwbuf **plwb, const char **pdata)
984263bc 634{
1b9d3514
MD
635 int rv;
636 vm_page_t ma;
06ecca5a 637 vm_page_t m;
984263bc
MD
638 vm_object_t object;
639
fad57d0e 640 /*
7540ab49 641 * The file has to be mappable.
fad57d0e 642 */
7540ab49 643 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 644 return (EIO);
984263bc 645
7c34d798
SS
646 if (pageno >= object->size)
647 return (EIO);
648
b12defdc 649 vm_object_hold(object);
7c34d798 650 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1b9d3514
MD
651 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
652 ma = m;
984263bc 653
06ecca5a
MD
654 /*
655 * get_pages unbusies all the requested pages except the
a55afca2
MD
656 * primary page (at index 0 in this case). The primary
657 * page may have been wired during the pagein (e.g. by
658 * the buffer cache) so vnode_pager_freepage() must be
659 * used to properly release it.
06ecca5a 660 */
1b9d3514 661 rv = vm_pager_get_page(object, &ma, 1);
7c34d798 662 m = vm_page_lookup(object, pageno);
984263bc 663
06ecca5a
MD
664 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
665 if (m) {
666 vm_page_protect(m, VM_PROT_NONE);
a55afca2 667 vnode_pager_freepage(m);
984263bc 668 }
471bd290 669 vm_object_drop(object);
984263bc
MD
670 return EIO;
671 }
672 }
b12defdc 673 vm_page_hold(m);
06ecca5a 674 vm_page_wakeup(m); /* unbusy the page */
b12defdc 675 vm_object_drop(object);
984263bc 676
7a683a24 677 *plwb = lwbuf_alloc(m, *plwb);
5c5185ae 678 *pdata = (void *)lwbuf_kva(*plwb);
7c34d798
SS
679
680 return (0);
681}
682
7484bdd2
MD
683/*
684 * Map the first page of an executable image.
685 *
686 * NOTE: If the mapping fails we have to NULL-out firstpage which may
687 * still be pointing to our supplied lwp structure.
688 */
7c34d798
SS
689int
690exec_map_first_page(struct image_params *imgp)
691{
692 int err;
693
694 if (imgp->firstpage)
695 exec_unmap_first_page(imgp);
696
7a683a24 697 imgp->firstpage = &imgp->firstpage_cache;
7c34d798
SS
698 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
699
7484bdd2
MD
700 if (err) {
701 imgp->firstpage = NULL;
7c34d798 702 return err;
7484bdd2 703 }
984263bc
MD
704
705 return 0;
706}
707
708void
5c5185ae 709exec_unmap_page(struct lwbuf *lwb)
984263bc 710{
18f40545
MD
711 vm_page_t m;
712
5fd012e0 713 crit_enter();
5c5185ae
SG
714 if (lwb != NULL) {
715 m = lwbuf_page(lwb);
716 lwbuf_free(lwb);
e4ba7818 717 vm_page_unhold(m);
984263bc 718 }
5fd012e0 719 crit_exit();
984263bc
MD
720}
721
7c34d798
SS
722void
723exec_unmap_first_page(struct image_params *imgp)
724{
725 exec_unmap_page(imgp->firstpage);
726 imgp->firstpage = NULL;
727 imgp->image_header = NULL;
728}
729
984263bc
MD
730/*
731 * Destroy old address space, and allocate a new stack
732 * The new stack is only SGROWSIZ large because it is grown
733 * automatically in trap.c.
dc52e1cc
MD
734 *
735 * This is the point of no return.
984263bc
MD
736 */
737int
29802dbb 738exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 739{
984263bc
MD
740 struct vmspace *vmspace = imgp->proc->p_vmspace;
741 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 742 struct proc *p;
29802dbb 743 vm_map_t map;
dc52e1cc 744 int error;
984263bc 745
dc52e1cc
MD
746 /*
747 * Indicate that we cannot gracefully error out any more, kill
748 * any other threads present, and set P_INEXEC to indicate that
749 * we are now messing with the process structure proper.
750 *
751 * If killalllwps() races return an error which coupled with
752 * vmspace_destroyed will cause us to exit. This is what we
753 * want since another thread is patiently waiting for us to exit
754 * in that case.
755 */
756 p = curproc;
984263bc
MD
757 imgp->vmspace_destroyed = 1;
758
dc52e1cc
MD
759 if (curthread->td_proc->p_nthreads > 1) {
760 error = killalllwps(1);
761 if (error)
762 return (error);
763 }
764 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
4643740a 765 p->p_flags |= P_INEXEC;
08f2f1bb
SS
766
767 /*
82354ad8
MD
768 * After setting P_INEXEC wait for any remaining references to
769 * the process (p) to go away.
770 *
771 * In particular, a vfork/exec sequence will replace p->p_vmspace
772 * and we must interlock anyone trying to access the space (aka
773 * procfs or sys_process.c calling procfs_domem()).
774 *
775 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us.
776 */
777 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0));
778
779 /*
984263bc
MD
780 * Blow away entire process VM, if address space not shared,
781 * otherwise, create a new VM space so that other threads are
29802dbb
MD
782 * not disrupted. If we are execing a resident vmspace we
783 * create a duplicate of it and remap the stack.
984263bc 784 */
29802dbb
MD
785 map = &vmspace->vm_map;
786 if (vmcopy) {
787 vmspace_exec(imgp->proc, vmcopy);
788 vmspace = imgp->proc->p_vmspace;
789 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
790 map = &vmspace->vm_map;
a2ee730d 791 } else if (vmspace->vm_sysref.refcnt == 1) {
fef0fdf2 792 shmexit(vmspace);
239b4df9 793 if (vmspace->vm_upcalls)
08f2f1bb 794 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08 795 pmap_remove_pages(vmspace_pmap(vmspace),
b0c15cdf 796 0, VM_MAX_USER_ADDRESS);
88181b08 797 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 798 } else {
29802dbb 799 vmspace_exec(imgp->proc, NULL);
984263bc
MD
800 vmspace = imgp->proc->p_vmspace;
801 map = &vmspace->vm_map;
802 }
803
804 /* Allocate a new stack */
805 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
c809941b 806 0, VM_PROT_ALL, VM_PROT_ALL, 0);
984263bc
MD
807 if (error)
808 return (error);
809
810 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
811 * VM_STACK case, but they are still used to monitor the size of the
812 * process stack so we can check the stack rlimit.
813 */
814 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
815 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
816
817 return(0);
818}
819
820/*
821 * Copy out argument and environment strings from the old process
822 * address space into the temporary string buffer.
823 */
824int
2bd9d75c 825exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 826 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 827{
984263bc 828 char *argp, *envp;
2bd9d75c 829 int error = 0;
984263bc
MD
830 size_t length;
831
18b3f457 832 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
833 if (args->buf == NULL)
834 return (ENOMEM);
835 args->begin_argv = args->buf;
836 args->endp = args->begin_argv;
837 args->space = ARG_MAX;
2bd9d75c
DRJ
838
839 args->fname = args->buf + ARG_MAX;
840
984263bc 841 /*
2bd9d75c 842 * Copy the file name.
984263bc 843 */
2bd9d75c
DRJ
844 if (segflg == PATH_SYSSPACE) {
845 error = copystr(fname, args->fname, PATH_MAX, &length);
846 } else if (segflg == PATH_USERSPACE) {
847 error = copyinstr(fname, args->fname, PATH_MAX, &length);
848 }
984263bc 849
2bd9d75c 850 /*
66be6566
MD
851 * Extract argument strings. argv may not be NULL. The argv
852 * array is terminated by a NULL entry. We special-case the
853 * situation where argv[0] is NULL by passing { filename, NULL }
854 * to the new program to guarentee that the interpreter knows what
855 * file to open in case we exec an interpreted file. Note that
856 * a NULL argv[0] terminates the argv[] array.
857 *
858 * XXX the special-casing of argv[0] is historical and needs to be
859 * revisited.
2bd9d75c 860 */
66be6566
MD
861 if (argv == NULL)
862 error = EFAULT;
863 if (error == 0) {
864 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
865 if (argp == (caddr_t)-1) {
2bd9d75c 866 error = EFAULT;
66be6566 867 break;
2bd9d75c
DRJ
868 }
869 error = copyinstr(argp, args->endp,
2abfe22f 870 args->space, &length);
66be6566
MD
871 if (error) {
872 if (error == ENAMETOOLONG)
873 error = E2BIG;
874 break;
875 }
2bd9d75c
DRJ
876 args->space -= length;
877 args->endp += length;
878 args->argc++;
984263bc 879 }
66be6566
MD
880 if (args->argc == 0 && error == 0) {
881 length = strlen(args->fname) + 1;
882 if (length > args->space) {
883 error = E2BIG;
884 } else {
885 bcopy(args->fname, args->endp, length);
886 args->space -= length;
887 args->endp += length;
888 args->argc++;
889 }
890 }
984263bc
MD
891 }
892
2bd9d75c 893 args->begin_envv = args->endp;
984263bc
MD
894
895 /*
66be6566 896 * extract environment strings. envv may be NULL.
984263bc 897 */
2abfe22f 898 if (envv && error == 0) {
984263bc 899 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
900 if (envp == (caddr_t) -1) {
901 error = EFAULT;
66be6566 902 break;
984263bc 903 }
2bd9d75c
DRJ
904 error = copyinstr(envp, args->endp, args->space,
905 &length);
66be6566
MD
906 if (error) {
907 if (error == ENAMETOOLONG)
908 error = E2BIG;
909 break;
910 }
2bd9d75c
DRJ
911 args->space -= length;
912 args->endp += length;
913 args->envc++;
984263bc
MD
914 }
915 }
2bd9d75c
DRJ
916 return (error);
917}
918
919void
920exec_free_args(struct image_args *args)
921{
2abfe22f 922 if (args->buf) {
18b3f457 923 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
924 args->buf = NULL;
925 }
984263bc
MD
926}
927
928/*
929 * Copy strings out to the new process address space, constructing
930 * new arg and env vector tables. Return a pointer to the base
931 * so that it can be used as the initial stack pointer.
932 */
933register_t *
2bd9d75c 934exec_copyout_strings(struct image_params *imgp)
984263bc 935{
17a7ca1f 936 int argc, envc, sgap;
984263bc
MD
937 char **vectp;
938 char *stringp, *destp;
939 register_t *stack_base;
940 struct ps_strings *arginfo;
adc42cf3 941 size_t execpath_len;
984263bc
MD
942 int szsigcode;
943
944 /*
945 * Calculate string base and vector table pointers.
946 * Also deal with signal trampoline code for this exec type.
947 */
adc42cf3
JM
948 if (imgp->execpath != NULL && imgp->auxargs != NULL)
949 execpath_len = strlen(imgp->execpath) + 1;
950 else
951 execpath_len = 0;
984263bc
MD
952 arginfo = (struct ps_strings *)PS_STRINGS;
953 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
17a7ca1f 954 if (stackgap_random != 0)
0ced1954 955 sgap = ALIGN(karc4random() & (stackgap_random - 1));
17a7ca1f
MD
956 else
957 sgap = 0;
958 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
adc42cf3 959 roundup(execpath_len, sizeof(char *)) -
2bd9d75c 960 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
984263bc
MD
961
962 /*
963 * install sigcode
964 */
965 if (szsigcode)
966 copyout(imgp->proc->p_sysent->sv_sigcode,
2bd9d75c 967 ((caddr_t)arginfo - szsigcode), szsigcode);
984263bc
MD
968
969 /*
adc42cf3
JM
970 * Copy the image path for the rtld
971 */
972 if (execpath_len != 0) {
973 imgp->execpathp = (uintptr_t)arginfo
974 - szsigcode
975 - execpath_len;
976 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
977 }
978
979 /*
984263bc
MD
980 * If we have a valid auxargs ptr, prepare some room
981 * on the stack.
7ec1918b 982 *
984263bc
MD
983 * The '+ 2' is for the null pointers at the end of each of the
984 * arg and env vector sets, and 'AT_COUNT*2' is room for the
985 * ELF Auxargs data.
986 */
7ec1918b
MD
987 if (imgp->auxargs) {
988 vectp = (char **)(destp - (imgp->args->argc +
adc42cf3
JM
989 imgp->args->envc + 2 + (AT_COUNT * 2) + execpath_len) *
990 sizeof(char*));
7ec1918b 991 } else {
2bd9d75c 992 vectp = (char **)(destp - (imgp->args->argc +
7ec1918b
MD
993 imgp->args->envc + 2) * sizeof(char*));
994 }
995
984263bc 996 /*
31bc42c8
MD
997 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
998 * be done in crt1.o. Note that GCC 3.x aligns the stack in main.
984263bc 999 */
984263bc
MD
1000
1001 /*
1002 * vectp also becomes our initial stack base
1003 */
1004 stack_base = (register_t *)vectp;
1005
2bd9d75c
DRJ
1006 stringp = imgp->args->begin_argv;
1007 argc = imgp->args->argc;
1008 envc = imgp->args->envc;
984263bc
MD
1009
1010 /*
1011 * Copy out strings - arguments and environment.
1012 */
2bd9d75c 1013 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
1014
1015 /*
1016 * Fill in "ps_strings" struct for ps, w, etc.
1017 */
1018 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
3f938eea 1019 suword32(&arginfo->ps_nargvstr, argc);
984263bc
MD
1020
1021 /*
1022 * Fill in argument portion of vector table.
1023 */
1024 for (; argc > 0; --argc) {
1025 suword(vectp++, (long)(intptr_t)destp);
1026 while (*stringp++ != 0)
1027 destp++;
1028 destp++;
1029 }
1030
7a44d1cb 1031 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
1032 suword(vectp++, 0);
1033
1034 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
3f938eea 1035 suword32(&arginfo->ps_nenvstr, envc);
984263bc
MD
1036
1037 /*
1038 * Fill in environment portion of vector table.
1039 */
1040 for (; envc > 0; --envc) {
1041 suword(vectp++, (long)(intptr_t)destp);
1042 while (*stringp++ != 0)
1043 destp++;
1044 destp++;
1045 }
1046
1047 /* end of vector table is a null pointer */
1048 suword(vectp, 0);
1049
1050 return (stack_base);
1051}
1052
1053/*
1054 * Check permissions of file to execute.
1055 * Return 0 for success or error code on failure.
1056 */
1057int
246693ac 1058exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
984263bc
MD
1059{
1060 struct proc *p = imgp->proc;
1061 struct vnode *vp = imgp->vp;
1062 struct vattr *attr = imgp->attr;
1063 int error;
1064
1065 /* Get file attributes */
87de5057 1066 error = VOP_GETATTR(vp, attr);
984263bc
MD
1067 if (error)
1068 return (error);
1069
1070 /*
1071 * 1) Check if file execution is disabled for the filesystem that this
1072 * file resides on.
1073 * 2) Insure that at least one execute bit is on - otherwise root
1074 * will always succeed, and we don't want to happen unless the
1075 * file really is executable.
1076 * 3) Insure that the file is a regular file.
1077 */
1078 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
246693ac 1079 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
984263bc
MD
1080 ((attr->va_mode & 0111) == 0) ||
1081 (attr->va_type != VREG)) {
1082 return (EACCES);
1083 }
1084
1085 /*
1086 * Zero length files can't be exec'd
1087 */
1088 if (attr->va_size == 0)
1089 return (ENOEXEC);
1090
1091 /*
1092 * Check for execute permission to file based on current credentials.
1093 */
90a83564 1094 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
1095 if (error)
1096 return (error);
1097
1098 /*
1099 * Check number of open-for-writes on the file and deny execution
1100 * if there are any.
1101 */
1102 if (vp->v_writecount)
1103 return (ETXTBSY);
1104
1105 /*
7540ab49
MD
1106 * Call filesystem specific open routine, which allows us to read,
1107 * write, and mmap the file. Without the VOP_OPEN we can only
1108 * stat the file.
984263bc 1109 */
87de5057 1110 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
1111 if (error)
1112 return (error);
1113
1114 return (0);
1115}
1116
1117/*
1118 * Exec handler registration
1119 */
1120int
77153250 1121exec_register(const struct execsw *execsw_arg)
984263bc
MD
1122{
1123 const struct execsw **es, **xs, **newexecsw;
1124 int count = 2; /* New slot and trailing NULL */
1125
1126 if (execsw)
1127 for (es = execsw; *es; es++)
1128 count++;
efda3bd0 1129 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1130 xs = newexecsw;
1131 if (execsw)
1132 for (es = execsw; *es; es++)
1133 *xs++ = *es;
1134 *xs++ = execsw_arg;
1135 *xs = NULL;
1136 if (execsw)
efda3bd0 1137 kfree(execsw, M_TEMP);
984263bc
MD
1138 execsw = newexecsw;
1139 return 0;
1140}
1141
1142int
77153250 1143exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1144{
1145 const struct execsw **es, **xs, **newexecsw;
1146 int count = 1;
1147
1148 if (execsw == NULL)
ae7e74d4 1149 panic("unregister with no handlers left?");
984263bc
MD
1150
1151 for (es = execsw; *es; es++) {
1152 if (*es == execsw_arg)
1153 break;
1154 }
1155 if (*es == NULL)
1156 return ENOENT;
1157 for (es = execsw; *es; es++)
1158 if (*es != execsw_arg)
1159 count++;
efda3bd0 1160 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1161 xs = newexecsw;
1162 for (es = execsw; *es; es++)
1163 if (*es != execsw_arg)
1164 *xs++ = *es;
1165 *xs = NULL;
1166 if (execsw)
efda3bd0 1167 kfree(execsw, M_TEMP);
984263bc
MD
1168 execsw = newexecsw;
1169 return 0;
1170}