kernel - Optimize the x86-64 lwbuf API
[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>
984263bc
MD
50#include <sys/sysent.h>
51#include <sys/shm.h>
52#include <sys/sysctl.h>
53#include <sys/vnode.h>
a6f89c72 54#include <sys/vmmeter.h>
984263bc 55#include <sys/aio.h>
17a7ca1f 56#include <sys/libkern.h>
984263bc 57
5c5185ae
SG
58#include <cpu/lwbuf.h>
59
984263bc
MD
60#include <vm/vm.h>
61#include <vm/vm_param.h>
62#include <sys/lock.h>
63#include <vm/pmap.h>
64#include <vm/vm_page.h>
65#include <vm/vm_map.h>
66#include <vm/vm_kern.h>
67#include <vm/vm_extern.h>
68#include <vm/vm_object.h>
a55afca2 69#include <vm/vnode_pager.h>
984263bc
MD
70#include <vm/vm_pager.h>
71
72#include <sys/user.h>
527fddf7 73#include <sys/reg.h>
984263bc 74
5fd012e0 75#include <sys/thread2.h>
684a93c4 76#include <sys/mplock2.h>
5fd012e0 77
984263bc 78MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
18b3f457 79MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
984263bc 80
402ed7e1 81static register_t *exec_copyout_strings (struct image_params *);
984263bc
MD
82
83/* XXX This should be vm_size_t. */
84static u_long ps_strings = PS_STRINGS;
85SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
86
87/* XXX This should be vm_size_t. */
88static u_long usrstack = USRSTACK;
89SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
90
91u_long ps_arg_cache_limit = PAGE_SIZE / 16;
92SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
93 &ps_arg_cache_limit, 0, "");
94
95int ps_argsopen = 1;
96SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
97
5c627295
MD
98static int ktrace_suid = 0;
99SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
100
2bd9d75c
DRJ
101void print_execve_args(struct image_args *args);
102int debug_execve_args = 0;
103SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
104 0, "");
105
17a7ca1f 106/*
18b3f457
MD
107 * Exec arguments object cache
108 */
109static struct objcache *exec_objcache;
110
111static
112void
113exec_objcache_init(void *arg __unused)
114{
0aa16b5d
SZ
115 int cluster_limit;
116
117 cluster_limit = 16; /* up to this many objects */
18b3f457
MD
118 exec_objcache = objcache_create_mbacked(
119 M_EXECARGS, PATH_MAX + ARG_MAX,
0aa16b5d 120 &cluster_limit,
18b3f457
MD
121 2, /* minimal magazine capacity */
122 NULL, NULL, NULL);
123}
124SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
125
126/*
17a7ca1f
MD
127 * stackgap_random specifies if the stackgap should have a random size added
128 * to it. It must be a power of 2. If non-zero, the stack gap will be
0ced1954 129 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
17a7ca1f
MD
130 */
131static int stackgap_random = 1024;
132static int
133sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
134{
135 int error, new_val;
136 new_val = stackgap_random;
137 error = sysctl_handle_int(oidp, &new_val, 0, req);
138 if (error != 0 || req->newptr == NULL)
139 return (error);
140 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val))
141 return (EINVAL);
142 stackgap_random = new_val;
143
144 return(0);
145}
146
147SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT,
148 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
149
2bd9d75c
DRJ
150void
151print_execve_args(struct image_args *args)
152{
153 char *cp;
154 int ndx;
155
156 cp = args->begin_argv;
157 for (ndx = 0; ndx < args->argc; ndx++) {
6ea70f76 158 kprintf("\targv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
159 while (*cp++ != '\0');
160 }
161 for (ndx = 0; ndx < args->envc; ndx++) {
6ea70f76 162 kprintf("\tenvv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
163 while (*cp++ != '\0');
164 }
165}
166
984263bc
MD
167/*
168 * Each of the items is a pointer to a `const struct execsw', hence the
169 * double pointer here.
170 */
171static const struct execsw **execsw;
172
5417cd57
SS
173/*
174 * Replace current vmspace with a new binary.
175 * Returns 0 on success, > 0 on recoverable error (use as errno).
90947cb1 176 * Returns -1 on lethal error which demands killing of the current
5417cd57
SS
177 * process!
178 */
984263bc 179int
fad57d0e 180kern_execve(struct nlookupdata *nd, struct image_args *args)
984263bc 181{
dadab5e9 182 struct thread *td = curthread;
08f2f1bb 183 struct lwp *lp = td->td_lwp;
dadab5e9 184 struct proc *p = td->td_proc;
984263bc
MD
185 register_t *stack_base;
186 int error, len, i;
187 struct image_params image_params, *imgp;
188 struct vattr attr;
402ed7e1 189 int (*img_first) (struct image_params *);
984263bc 190
2bd9d75c 191 if (debug_execve_args) {
6ea70f76 192 kprintf("%s()\n", __func__);
2bd9d75c
DRJ
193 print_execve_args(args);
194 }
195
dadab5e9 196 KKASSERT(p);
984263bc
MD
197 imgp = &image_params;
198
199 /*
dc52e1cc
MD
200 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
201 * no modifications to the process at all until we get there.
202 *
203 * Note that multiple threads may be trying to exec at the same
204 * time. exec_new_vmspace() handles that too.
984263bc 205 */
984263bc
MD
206
207 /*
208 * Initialize part of the common data
209 */
210 imgp->proc = p;
2bd9d75c 211 imgp->args = args;
984263bc 212 imgp->attr = &attr;
984263bc 213 imgp->entry_addr = 0;
29802dbb 214 imgp->resident = 0;
984263bc
MD
215 imgp->vmspace_destroyed = 0;
216 imgp->interpreted = 0;
2abfe22f 217 imgp->interpreter_name[0] = 0;
984263bc
MD
218 imgp->auxargs = NULL;
219 imgp->vp = NULL;
220 imgp->firstpage = NULL;
221 imgp->ps_strings = 0;
18f40545 222 imgp->image_header = NULL;
2bd9d75c
DRJ
223
224interpret:
984263bc
MD
225
226 /*
fad57d0e
MD
227 * Translate the file name to a vnode. Unlock the cache entry to
228 * improve parallelism for programs exec'd in parallel.
984263bc 229 */
fad57d0e
MD
230 if ((error = nlookup(nd)) != 0)
231 goto exec_fail;
28623bf9 232 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
fad57d0e
MD
233 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
234 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 235 cache_unlock(&nd->nl_nch);
fad57d0e 236 if (error)
984263bc 237 goto exec_fail;
984263bc
MD
238
239 /*
246693ac
AHJ
240 * Check file permissions (also 'opens' file).
241 * Include also the top level mount in the check.
984263bc 242 */
246693ac 243 error = exec_check_permissions(imgp, nd->nl_nch.mount);
984263bc 244 if (error) {
a11aaa81 245 vn_unlock(imgp->vp);
984263bc
MD
246 goto exec_fail_dealloc;
247 }
248
249 error = exec_map_first_page(imgp);
a11aaa81 250 vn_unlock(imgp->vp);
984263bc
MD
251 if (error)
252 goto exec_fail_dealloc;
253
2bd9d75c 254 if (debug_execve_args && imgp->interpreted) {
6ea70f76
SW
255 kprintf(" target is interpreted -- recursive pass\n");
256 kprintf(" interpreter: %s\n", imgp->interpreter_name);
2bd9d75c
DRJ
257 print_execve_args(args);
258 }
259
984263bc
MD
260 /*
261 * If the current process has a special image activator it
262 * wants to try first, call it. For example, emulating shell
263 * scripts differently.
264 */
265 error = -1;
266 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
267 error = img_first(imgp);
268
269 /*
29802dbb
MD
270 * If the vnode has a registered vmspace, exec the vmspace
271 */
272 if (error == -1 && imgp->vp->v_resident) {
273 error = exec_resident_imgact(imgp);
274 }
275
276 /*
984263bc
MD
277 * Loop through the list of image activators, calling each one.
278 * An activator returns -1 if there is no match, 0 on success,
279 * and an error otherwise.
280 */
281 for (i = 0; error == -1 && execsw[i]; ++i) {
282 if (execsw[i]->ex_imgact == NULL ||
283 execsw[i]->ex_imgact == img_first) {
284 continue;
285 }
286 error = (*execsw[i]->ex_imgact)(imgp);
287 }
288
289 if (error) {
290 if (error == -1)
291 error = ENOEXEC;
292 goto exec_fail_dealloc;
293 }
294
295 /*
296 * Special interpreter operation, cleanup and loop up to try to
297 * activate the interpreter.
298 */
299 if (imgp->interpreted) {
300 exec_unmap_first_page(imgp);
fad57d0e
MD
301 nlookup_done(nd);
302 vrele(imgp->vp);
303 imgp->vp = NULL;
304 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
305 NLC_FOLLOW);
306 if (error)
307 goto exec_fail;
984263bc
MD
308 goto interpret;
309 }
310
311 /*
312 * Copy out strings (args and env) and initialize stack base
313 */
314 stack_base = exec_copyout_strings(imgp);
315 p->p_vmspace->vm_minsaddr = (char *)stack_base;
316
317 /*
318 * If custom stack fixup routine present for this process
29802dbb
MD
319 * let it do the stack setup. If we are running a resident
320 * image there is no auxinfo or other image activator context
321 * so don't try to add fixups to the stack.
322 *
984263bc
MD
323 * Else stuff argument count as first item on stack
324 */
29802dbb 325 if (p->p_sysent->sv_fixup && imgp->resident == 0)
984263bc
MD
326 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
327 else
2bd9d75c 328 suword(--stack_base, imgp->args->argc);
984263bc
MD
329
330 /*
331 * For security and other reasons, the file descriptor table cannot
332 * be shared after an exec.
333 */
334 if (p->p_fd->fd_refcnt > 1) {
335 struct filedesc *tmp;
336
337 tmp = fdcopy(p);
0a4a9c77 338 fdfree(p, tmp);
984263bc
MD
339 }
340
341 /*
342 * For security and other reasons, signal handlers cannot
343 * be shared after an exec. The new proces gets a copy of the old
344 * handlers. In execsigs(), the new process will have its signals
345 * reset.
346 */
b1b4e5a6
SS
347 if (p->p_sigacts->ps_refcnt > 1) {
348 struct sigacts *newsigacts;
984263bc 349
b1b4e5a6 350 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts),
984263bc 351 M_SUBPROC, M_WAITOK);
b1b4e5a6
SS
352 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts));
353 p->p_sigacts->ps_refcnt--;
354 p->p_sigacts = newsigacts;
355 p->p_sigacts->ps_refcnt = 1;
984263bc
MD
356 }
357
0daa37a5
MD
358 /*
359 * For security and other reasons virtual kernels cannot be
360 * inherited by an exec. This also allows a virtual kernel
361 * to fork/exec unrelated applications.
362 */
4a22e893
MD
363 if (p->p_vkernel)
364 vkernel_exit(p);
0daa37a5 365
984263bc
MD
366 /* Stop profiling */
367 stopprofclock(p);
368
369 /* close files on exec */
370 fdcloseexec(p);
371
372 /* reset caught signals */
373 execsigs(p);
374
375 /* name this process - nameiexec(p, ndp) */
28623bf9
MD
376 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
377 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
984263bc 378 p->p_comm[len] = 0;
08f2f1bb 379 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
984263bc
MD
380
381 /*
382 * mark as execed, wakeup the process that vforked (if any) and tell
383 * it that it now has its own resources back
384 */
385 p->p_flag |= P_EXEC;
386 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
387 p->p_flag &= ~P_PPWAIT;
388 wakeup((caddr_t)p->p_pptr);
389 }
390
391 /*
392 * Implement image setuid/setgid.
393 *
394 * Don't honor setuid/setgid if the filesystem prohibits it or if
395 * the process is being traced.
396 */
397 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
398 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
399 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
400 (p->p_flag & P_TRACED) == 0) {
401 /*
402 * Turn off syscall tracing for set-id programs, except for
403 * root. Record any set-id flags first to make sure that
404 * we do not regain any tracing during a possible block.
405 */
41c20dac 406 setsugid();
5c627295
MD
407 if (p->p_tracenode && ktrace_suid == 0 &&
408 priv_check(td, PRIV_ROOT) != 0) {
29f58392
MD
409 ktrdestroy(&p->p_tracenode);
410 p->p_traceflag = 0;
984263bc
MD
411 }
412 /* Close any file descriptors 0..2 that reference procfs */
413 setugidsafety(p);
414 /* Make sure file descriptors 0..2 are in use. */
f3a2d8c4 415 error = fdcheckstd(lp);
984263bc
MD
416 if (error != 0)
417 goto exec_fail_dealloc;
418 /*
419 * Set the new credentials.
420 */
e9a372eb 421 cratom(&p->p_ucred);
984263bc 422 if (attr.va_mode & VSUID)
41c20dac 423 change_euid(attr.va_uid);
984263bc
MD
424 if (attr.va_mode & VSGID)
425 p->p_ucred->cr_gid = attr.va_gid;
98a7f915
MD
426
427 /*
428 * Clear local varsym variables
429 */
430 varsymset_clean(&p->p_varsymset);
984263bc 431 } else {
41c20dac
MD
432 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
433 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
984263bc
MD
434 p->p_flag &= ~P_SUGID;
435 }
436
437 /*
438 * Implement correct POSIX saved-id behavior.
439 */
115ccd83
MD
440 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
441 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
442 cratom(&p->p_ucred);
443 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
444 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
445 }
984263bc
MD
446
447 /*
448 * Store the vp for use in procfs
449 */
450 if (p->p_textvp) /* release old reference */
451 vrele(p->p_textvp);
fad57d0e
MD
452 p->p_textvp = imgp->vp;
453 vref(p->p_textvp);
984263bc 454
8ba5f7ef
AH
455 /* Release old namecache handle to text file */
456 if (p->p_textnch.ncp)
457 cache_drop(&p->p_textnch);
458
459 if (nd->nl_nch.mount)
460 cache_copy(&nd->nl_nch, &p->p_textnch);
461
984263bc
MD
462 /*
463 * Notify others that we exec'd, and clear the P_INEXEC flag
464 * as we're now a bona fide freshly-execed process.
465 */
466 KNOTE(&p->p_klist, NOTE_EXEC);
467 p->p_flag &= ~P_INEXEC;
468
469 /*
470 * If tracing the process, trap to debugger so breakpoints
471 * can be set before the program executes.
472 */
473 STOPEVENT(p, S_EXEC, 0);
474
475 if (p->p_flag & P_TRACED)
84204577 476 ksignal(p, SIGTRAP);
984263bc
MD
477
478 /* clear "fork but no exec" flag, as we _are_ execing */
479 p->p_acflag &= ~AFORK;
480
481 /* Set values passed into the program in registers. */
08f2f1bb 482 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
984263bc
MD
483 imgp->ps_strings);
484
349433c9
MD
485 /* Set the access time on the vnode */
486 vn_mark_atime(imgp->vp, td);
487
984263bc
MD
488 /* Free any previous argument cache */
489 if (p->p_args && --p->p_args->ar_ref == 0)
490 FREE(p->p_args, M_PARGS);
491 p->p_args = NULL;
492
493 /* Cache arguments if they fit inside our allowance */
2bd9d75c 494 i = imgp->args->begin_envv - imgp->args->begin_argv;
984263bc
MD
495 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
496 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
497 M_PARGS, M_WAITOK);
498 p->p_args->ar_ref = 1;
499 p->p_args->ar_length = i;
2bd9d75c 500 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
984263bc
MD
501 }
502
503exec_fail_dealloc:
504
505 /*
506 * free various allocated resources
507 */
508 if (imgp->firstpage)
509 exec_unmap_first_page(imgp);
510
984263bc 511 if (imgp->vp) {
984263bc 512 vrele(imgp->vp);
fad57d0e 513 imgp->vp = NULL;
984263bc
MD
514 }
515
a6f89c72
DR
516 if (error == 0) {
517 ++mycpu->gd_cnt.v_exec;
984263bc 518 return (0);
a6f89c72 519 }
984263bc
MD
520
521exec_fail:
dc52e1cc
MD
522 /*
523 * we're done here, clear P_INEXEC if we were the ones that
524 * set it. Otherwise if vmspace_destroyed is still set we
525 * raced another thread and that thread is responsible for
526 * clearing it.
527 */
528 if (imgp->vmspace_destroyed & 2)
529 p->p_flag &= ~P_INEXEC;
984263bc 530 if (imgp->vmspace_destroyed) {
5417cd57
SS
531 /*
532 * Sorry, no more process anymore. exit gracefully.
533 * However we can't die right here, because our
534 * caller might have to clean up, so indicate a
90947cb1 535 * lethal error by returning -1.
5417cd57
SS
536 */
537 return(-1);
984263bc
MD
538 } else {
539 return(error);
540 }
541}
542
2bd9d75c
DRJ
543/*
544 * execve() system call.
3919ced0
MD
545 *
546 * MPALMOSTSAFE
2bd9d75c
DRJ
547 */
548int
753fd850 549sys_execve(struct execve_args *uap)
2bd9d75c 550{
fad57d0e 551 struct nlookupdata nd;
2bd9d75c
DRJ
552 struct image_args args;
553 int error;
554
118cec38 555 bzero(&args, sizeof(args));
3919ced0
MD
556
557 get_mplock();
558 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
fad57d0e
MD
559 if (error == 0) {
560 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
561 uap->argv, uap->envv);
562 }
2abfe22f
MD
563 if (error == 0)
564 error = kern_execve(&nd, &args);
fad57d0e 565 nlookup_done(&nd);
2bd9d75c
DRJ
566 exec_free_args(&args);
567
5417cd57 568 if (error < 0) {
90947cb1 569 /* We hit a lethal error condition. Let's die now. */
5417cd57
SS
570 exit1(W_EXITCODE(0, SIGABRT));
571 /* NOTREACHED */
572 }
3919ced0 573 rel_mplock();
5417cd57 574
2bd9d75c
DRJ
575 /*
576 * The syscall result is returned in registers to the new program.
577 * Linux will register %edx as an atexit function and we must be
578 * sure to set it to 0. XXX
579 */
580 if (error == 0)
581 uap->sysmsg_result64 = 0;
582
583 return (error);
584}
585
984263bc 586int
7c34d798 587exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
5c5185ae 588 struct lwbuf **plwb, const char **pdata)
984263bc 589{
1b9d3514
MD
590 int rv;
591 vm_page_t ma;
06ecca5a 592 vm_page_t m;
984263bc
MD
593 vm_object_t object;
594
fad57d0e 595 /*
7540ab49 596 * The file has to be mappable.
fad57d0e 597 */
7540ab49 598 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 599 return (EIO);
984263bc 600
7c34d798
SS
601 if (pageno >= object->size)
602 return (EIO);
603
7c34d798 604 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
77912481 605
573fb415 606 lwkt_gettoken(&vm_token);
1b9d3514
MD
607 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
608 ma = m;
984263bc 609
06ecca5a
MD
610 /*
611 * get_pages unbusies all the requested pages except the
a55afca2
MD
612 * primary page (at index 0 in this case). The primary
613 * page may have been wired during the pagein (e.g. by
614 * the buffer cache) so vnode_pager_freepage() must be
615 * used to properly release it.
06ecca5a 616 */
1b9d3514 617 rv = vm_pager_get_page(object, &ma, 1);
7c34d798 618 m = vm_page_lookup(object, pageno);
984263bc 619
06ecca5a
MD
620 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
621 if (m) {
622 vm_page_protect(m, VM_PROT_NONE);
a55afca2 623 vnode_pager_freepage(m);
984263bc 624 }
573fb415 625 lwkt_reltoken(&vm_token);
984263bc
MD
626 return EIO;
627 }
628 }
573fb415 629 vm_page_hold(m); /* requires vm_token to be held */
06ecca5a 630 vm_page_wakeup(m); /* unbusy the page */
573fb415 631 lwkt_reltoken(&vm_token);
984263bc 632
7a683a24 633 *plwb = lwbuf_alloc(m, *plwb);
5c5185ae 634 *pdata = (void *)lwbuf_kva(*plwb);
7c34d798
SS
635
636 return (0);
637}
638
639int
640exec_map_first_page(struct image_params *imgp)
641{
642 int err;
643
644 if (imgp->firstpage)
645 exec_unmap_first_page(imgp);
646
7a683a24 647 imgp->firstpage = &imgp->firstpage_cache;
7c34d798
SS
648 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
649
650 if (err)
651 return err;
984263bc
MD
652
653 return 0;
654}
655
656void
5c5185ae 657exec_unmap_page(struct lwbuf *lwb)
984263bc 658{
18f40545
MD
659 vm_page_t m;
660
5fd012e0 661 crit_enter();
5c5185ae
SG
662 if (lwb != NULL) {
663 m = lwbuf_page(lwb);
664 lwbuf_free(lwb);
e4ba7818 665 vm_page_unhold(m);
984263bc 666 }
5fd012e0 667 crit_exit();
984263bc
MD
668}
669
7c34d798
SS
670void
671exec_unmap_first_page(struct image_params *imgp)
672{
673 exec_unmap_page(imgp->firstpage);
674 imgp->firstpage = NULL;
675 imgp->image_header = NULL;
676}
677
984263bc
MD
678/*
679 * Destroy old address space, and allocate a new stack
680 * The new stack is only SGROWSIZ large because it is grown
681 * automatically in trap.c.
dc52e1cc
MD
682 *
683 * This is the point of no return.
984263bc
MD
684 */
685int
29802dbb 686exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 687{
984263bc
MD
688 struct vmspace *vmspace = imgp->proc->p_vmspace;
689 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 690 struct proc *p;
29802dbb 691 vm_map_t map;
dc52e1cc 692 int error;
984263bc 693
dc52e1cc
MD
694 /*
695 * Indicate that we cannot gracefully error out any more, kill
696 * any other threads present, and set P_INEXEC to indicate that
697 * we are now messing with the process structure proper.
698 *
699 * If killalllwps() races return an error which coupled with
700 * vmspace_destroyed will cause us to exit. This is what we
701 * want since another thread is patiently waiting for us to exit
702 * in that case.
703 */
704 p = curproc;
984263bc
MD
705 imgp->vmspace_destroyed = 1;
706
dc52e1cc
MD
707 if (curthread->td_proc->p_nthreads > 1) {
708 error = killalllwps(1);
709 if (error)
710 return (error);
711 }
712 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
713 p->p_flag |= P_INEXEC;
08f2f1bb
SS
714
715 /*
984263bc
MD
716 * Prevent a pending AIO from modifying the new address space.
717 */
718 aio_proc_rundown(imgp->proc);
719
720 /*
721 * Blow away entire process VM, if address space not shared,
722 * otherwise, create a new VM space so that other threads are
29802dbb
MD
723 * not disrupted. If we are execing a resident vmspace we
724 * create a duplicate of it and remap the stack.
239b4df9
MD
725 *
726 * The exitingcnt test is not strictly necessary but has been
727 * included for code sanity (to make the code more deterministic).
984263bc 728 */
29802dbb
MD
729 map = &vmspace->vm_map;
730 if (vmcopy) {
731 vmspace_exec(imgp->proc, vmcopy);
732 vmspace = imgp->proc->p_vmspace;
733 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
734 map = &vmspace->vm_map;
e3161323
MD
735 } else if (vmspace->vm_sysref.refcnt == 1 &&
736 vmspace->vm_exitingcnt == 0) {
fef0fdf2 737 shmexit(vmspace);
239b4df9 738 if (vmspace->vm_upcalls)
08f2f1bb 739 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08
MD
740 pmap_remove_pages(vmspace_pmap(vmspace),
741 0, VM_MAX_USER_ADDRESS);
742 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 743 } else {
29802dbb 744 vmspace_exec(imgp->proc, NULL);
984263bc
MD
745 vmspace = imgp->proc->p_vmspace;
746 map = &vmspace->vm_map;
747 }
748
749 /* Allocate a new stack */
750 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
c809941b 751 0, VM_PROT_ALL, VM_PROT_ALL, 0);
984263bc
MD
752 if (error)
753 return (error);
754
755 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
756 * VM_STACK case, but they are still used to monitor the size of the
757 * process stack so we can check the stack rlimit.
758 */
759 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
760 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
761
762 return(0);
763}
764
765/*
766 * Copy out argument and environment strings from the old process
767 * address space into the temporary string buffer.
768 */
769int
2bd9d75c 770exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 771 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 772{
984263bc 773 char *argp, *envp;
2bd9d75c 774 int error = 0;
984263bc
MD
775 size_t length;
776
18b3f457 777 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
778 if (args->buf == NULL)
779 return (ENOMEM);
780 args->begin_argv = args->buf;
781 args->endp = args->begin_argv;
782 args->space = ARG_MAX;
2bd9d75c
DRJ
783
784 args->fname = args->buf + ARG_MAX;
785
984263bc 786 /*
2bd9d75c 787 * Copy the file name.
984263bc 788 */
2bd9d75c
DRJ
789 if (segflg == PATH_SYSSPACE) {
790 error = copystr(fname, args->fname, PATH_MAX, &length);
791 } else if (segflg == PATH_USERSPACE) {
792 error = copyinstr(fname, args->fname, PATH_MAX, &length);
793 }
984263bc 794
2bd9d75c 795 /*
66be6566
MD
796 * Extract argument strings. argv may not be NULL. The argv
797 * array is terminated by a NULL entry. We special-case the
798 * situation where argv[0] is NULL by passing { filename, NULL }
799 * to the new program to guarentee that the interpreter knows what
800 * file to open in case we exec an interpreted file. Note that
801 * a NULL argv[0] terminates the argv[] array.
802 *
803 * XXX the special-casing of argv[0] is historical and needs to be
804 * revisited.
2bd9d75c 805 */
66be6566
MD
806 if (argv == NULL)
807 error = EFAULT;
808 if (error == 0) {
809 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
810 if (argp == (caddr_t)-1) {
2bd9d75c 811 error = EFAULT;
66be6566 812 break;
2bd9d75c
DRJ
813 }
814 error = copyinstr(argp, args->endp,
2abfe22f 815 args->space, &length);
66be6566
MD
816 if (error) {
817 if (error == ENAMETOOLONG)
818 error = E2BIG;
819 break;
820 }
2bd9d75c
DRJ
821 args->space -= length;
822 args->endp += length;
823 args->argc++;
984263bc 824 }
66be6566
MD
825 if (args->argc == 0 && error == 0) {
826 length = strlen(args->fname) + 1;
827 if (length > args->space) {
828 error = E2BIG;
829 } else {
830 bcopy(args->fname, args->endp, length);
831 args->space -= length;
832 args->endp += length;
833 args->argc++;
834 }
835 }
984263bc
MD
836 }
837
2bd9d75c 838 args->begin_envv = args->endp;
984263bc
MD
839
840 /*
66be6566 841 * extract environment strings. envv may be NULL.
984263bc 842 */
2abfe22f 843 if (envv && error == 0) {
984263bc 844 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
845 if (envp == (caddr_t) -1) {
846 error = EFAULT;
66be6566 847 break;
984263bc 848 }
2bd9d75c
DRJ
849 error = copyinstr(envp, args->endp, args->space,
850 &length);
66be6566
MD
851 if (error) {
852 if (error == ENAMETOOLONG)
853 error = E2BIG;
854 break;
855 }
2bd9d75c
DRJ
856 args->space -= length;
857 args->endp += length;
858 args->envc++;
984263bc
MD
859 }
860 }
2bd9d75c
DRJ
861 return (error);
862}
863
864void
865exec_free_args(struct image_args *args)
866{
2abfe22f 867 if (args->buf) {
18b3f457 868 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
869 args->buf = NULL;
870 }
984263bc
MD
871}
872
873/*
874 * Copy strings out to the new process address space, constructing
875 * new arg and env vector tables. Return a pointer to the base
876 * so that it can be used as the initial stack pointer.
877 */
878register_t *
2bd9d75c 879exec_copyout_strings(struct image_params *imgp)
984263bc 880{
17a7ca1f 881 int argc, envc, sgap;
984263bc
MD
882 char **vectp;
883 char *stringp, *destp;
884 register_t *stack_base;
885 struct ps_strings *arginfo;
886 int szsigcode;
887
888 /*
889 * Calculate string base and vector table pointers.
890 * Also deal with signal trampoline code for this exec type.
891 */
892 arginfo = (struct ps_strings *)PS_STRINGS;
893 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
17a7ca1f 894 if (stackgap_random != 0)
0ced1954 895 sgap = ALIGN(karc4random() & (stackgap_random - 1));
17a7ca1f
MD
896 else
897 sgap = 0;
898 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
2bd9d75c 899 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
984263bc
MD
900
901 /*
902 * install sigcode
903 */
904 if (szsigcode)
905 copyout(imgp->proc->p_sysent->sv_sigcode,
2bd9d75c 906 ((caddr_t)arginfo - szsigcode), szsigcode);
984263bc
MD
907
908 /*
909 * If we have a valid auxargs ptr, prepare some room
910 * on the stack.
7ec1918b 911 *
984263bc
MD
912 * The '+ 2' is for the null pointers at the end of each of the
913 * arg and env vector sets, and 'AT_COUNT*2' is room for the
914 * ELF Auxargs data.
915 */
7ec1918b
MD
916 if (imgp->auxargs) {
917 vectp = (char **)(destp - (imgp->args->argc +
918 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
919 } else {
2bd9d75c 920 vectp = (char **)(destp - (imgp->args->argc +
7ec1918b
MD
921 imgp->args->envc + 2) * sizeof(char*));
922 }
923
984263bc 924 /*
31bc42c8
MD
925 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
926 * be done in crt1.o. Note that GCC 3.x aligns the stack in main.
984263bc 927 */
984263bc
MD
928
929 /*
930 * vectp also becomes our initial stack base
931 */
932 stack_base = (register_t *)vectp;
933
2bd9d75c
DRJ
934 stringp = imgp->args->begin_argv;
935 argc = imgp->args->argc;
936 envc = imgp->args->envc;
984263bc
MD
937
938 /*
939 * Copy out strings - arguments and environment.
940 */
2bd9d75c 941 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
942
943 /*
944 * Fill in "ps_strings" struct for ps, w, etc.
945 */
946 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
947 suword(&arginfo->ps_nargvstr, argc);
948
949 /*
950 * Fill in argument portion of vector table.
951 */
952 for (; argc > 0; --argc) {
953 suword(vectp++, (long)(intptr_t)destp);
954 while (*stringp++ != 0)
955 destp++;
956 destp++;
957 }
958
7a44d1cb 959 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
960 suword(vectp++, 0);
961
962 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
963 suword(&arginfo->ps_nenvstr, envc);
964
965 /*
966 * Fill in environment portion of vector table.
967 */
968 for (; envc > 0; --envc) {
969 suword(vectp++, (long)(intptr_t)destp);
970 while (*stringp++ != 0)
971 destp++;
972 destp++;
973 }
974
975 /* end of vector table is a null pointer */
976 suword(vectp, 0);
977
978 return (stack_base);
979}
980
981/*
982 * Check permissions of file to execute.
983 * Return 0 for success or error code on failure.
984 */
985int
246693ac 986exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
984263bc
MD
987{
988 struct proc *p = imgp->proc;
989 struct vnode *vp = imgp->vp;
990 struct vattr *attr = imgp->attr;
991 int error;
992
993 /* Get file attributes */
87de5057 994 error = VOP_GETATTR(vp, attr);
984263bc
MD
995 if (error)
996 return (error);
997
998 /*
999 * 1) Check if file execution is disabled for the filesystem that this
1000 * file resides on.
1001 * 2) Insure that at least one execute bit is on - otherwise root
1002 * will always succeed, and we don't want to happen unless the
1003 * file really is executable.
1004 * 3) Insure that the file is a regular file.
1005 */
1006 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
246693ac 1007 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
984263bc
MD
1008 ((attr->va_mode & 0111) == 0) ||
1009 (attr->va_type != VREG)) {
1010 return (EACCES);
1011 }
1012
1013 /*
1014 * Zero length files can't be exec'd
1015 */
1016 if (attr->va_size == 0)
1017 return (ENOEXEC);
1018
1019 /*
1020 * Check for execute permission to file based on current credentials.
1021 */
90a83564 1022 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
1023 if (error)
1024 return (error);
1025
1026 /*
1027 * Check number of open-for-writes on the file and deny execution
1028 * if there are any.
1029 */
1030 if (vp->v_writecount)
1031 return (ETXTBSY);
1032
1033 /*
7540ab49
MD
1034 * Call filesystem specific open routine, which allows us to read,
1035 * write, and mmap the file. Without the VOP_OPEN we can only
1036 * stat the file.
984263bc 1037 */
87de5057 1038 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
1039 if (error)
1040 return (error);
1041
1042 return (0);
1043}
1044
1045/*
1046 * Exec handler registration
1047 */
1048int
77153250 1049exec_register(const struct execsw *execsw_arg)
984263bc
MD
1050{
1051 const struct execsw **es, **xs, **newexecsw;
1052 int count = 2; /* New slot and trailing NULL */
1053
1054 if (execsw)
1055 for (es = execsw; *es; es++)
1056 count++;
efda3bd0 1057 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1058 xs = newexecsw;
1059 if (execsw)
1060 for (es = execsw; *es; es++)
1061 *xs++ = *es;
1062 *xs++ = execsw_arg;
1063 *xs = NULL;
1064 if (execsw)
efda3bd0 1065 kfree(execsw, M_TEMP);
984263bc
MD
1066 execsw = newexecsw;
1067 return 0;
1068}
1069
1070int
77153250 1071exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1072{
1073 const struct execsw **es, **xs, **newexecsw;
1074 int count = 1;
1075
1076 if (execsw == NULL)
ae7e74d4 1077 panic("unregister with no handlers left?");
984263bc
MD
1078
1079 for (es = execsw; *es; es++) {
1080 if (*es == execsw_arg)
1081 break;
1082 }
1083 if (*es == NULL)
1084 return ENOENT;
1085 for (es = execsw; *es; es++)
1086 if (*es != execsw_arg)
1087 count++;
efda3bd0 1088 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1089 xs = newexecsw;
1090 for (es = execsw; *es; es++)
1091 if (*es != execsw_arg)
1092 *xs++ = *es;
1093 *xs = NULL;
1094 if (execsw)
efda3bd0 1095 kfree(execsw, M_TEMP);
984263bc
MD
1096 execsw = newexecsw;
1097 return 0;
1098}