When an inode collision occurs a partially initialized vnode will be
[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 $
dc52e1cc 27 * $DragonFly: src/sys/kern/kern_exec.c,v 1.59 2007/07/12 21:56:22 dillon 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>
29f58392 45#include <sys/ktrace.h>
984263bc
MD
46#include <sys/signalvar.h>
47#include <sys/pioctl.h>
fad57d0e 48#include <sys/nlookup.h>
18f40545 49#include <sys/sfbuf.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
MD
57
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>
67#include <vm/vm_pager.h>
68
69#include <sys/user.h>
527fddf7 70#include <sys/reg.h>
984263bc 71
5fd012e0
MD
72#include <sys/thread2.h>
73
984263bc 74MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
18b3f457 75MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
984263bc 76
402ed7e1 77static register_t *exec_copyout_strings (struct image_params *);
984263bc
MD
78
79/* XXX This should be vm_size_t. */
80static u_long ps_strings = PS_STRINGS;
81SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
82
83/* XXX This should be vm_size_t. */
84static u_long usrstack = USRSTACK;
85SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
86
87u_long ps_arg_cache_limit = PAGE_SIZE / 16;
88SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
89 &ps_arg_cache_limit, 0, "");
90
91int ps_argsopen = 1;
92SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
93
2bd9d75c
DRJ
94void print_execve_args(struct image_args *args);
95int debug_execve_args = 0;
96SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
97 0, "");
98
18b3f457
MD
99/*
100 * Exec arguments object cache
101 */
102static struct objcache *exec_objcache;
103
104static
105void
106exec_objcache_init(void *arg __unused)
107{
108 exec_objcache = objcache_create_mbacked(
109 M_EXECARGS, PATH_MAX + ARG_MAX,
110 16, /* up to this many objects */
111 2, /* minimal magazine capacity */
112 NULL, NULL, NULL);
113}
114SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
115
17a7ca1f
MD
116/*
117 * stackgap_random specifies if the stackgap should have a random size added
118 * to it. It must be a power of 2. If non-zero, the stack gap will be
0ced1954 119 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
17a7ca1f
MD
120 */
121static int stackgap_random = 1024;
122static int
123sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
124{
125 int error, new_val;
126 new_val = stackgap_random;
127 error = sysctl_handle_int(oidp, &new_val, 0, req);
128 if (error != 0 || req->newptr == NULL)
129 return (error);
130 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val))
131 return (EINVAL);
132 stackgap_random = new_val;
133
134 return(0);
135}
136
137SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT,
138 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
139
2bd9d75c
DRJ
140void
141print_execve_args(struct image_args *args)
142{
143 char *cp;
144 int ndx;
145
146 cp = args->begin_argv;
147 for (ndx = 0; ndx < args->argc; ndx++) {
6ea70f76 148 kprintf("\targv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
149 while (*cp++ != '\0');
150 }
151 for (ndx = 0; ndx < args->envc; ndx++) {
6ea70f76 152 kprintf("\tenvv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
153 while (*cp++ != '\0');
154 }
155}
156
984263bc
MD
157/*
158 * Each of the items is a pointer to a `const struct execsw', hence the
159 * double pointer here.
160 */
161static const struct execsw **execsw;
162
984263bc 163int
fad57d0e 164kern_execve(struct nlookupdata *nd, struct image_args *args)
984263bc 165{
dadab5e9 166 struct thread *td = curthread;
08f2f1bb 167 struct lwp *lp = td->td_lwp;
dadab5e9 168 struct proc *p = td->td_proc;
984263bc
MD
169 register_t *stack_base;
170 int error, len, i;
171 struct image_params image_params, *imgp;
172 struct vattr attr;
402ed7e1 173 int (*img_first) (struct image_params *);
984263bc 174
2bd9d75c 175 if (debug_execve_args) {
6ea70f76 176 kprintf("%s()\n", __func__);
2bd9d75c
DRJ
177 print_execve_args(args);
178 }
179
dadab5e9 180 KKASSERT(p);
984263bc
MD
181 imgp = &image_params;
182
183 /*
dc52e1cc
MD
184 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
185 * no modifications to the process at all until we get there.
186 *
187 * Note that multiple threads may be trying to exec at the same
188 * time. exec_new_vmspace() handles that too.
984263bc 189 */
984263bc
MD
190
191 /*
192 * Initialize part of the common data
193 */
194 imgp->proc = p;
2bd9d75c 195 imgp->args = args;
984263bc 196 imgp->attr = &attr;
984263bc 197 imgp->entry_addr = 0;
29802dbb 198 imgp->resident = 0;
984263bc
MD
199 imgp->vmspace_destroyed = 0;
200 imgp->interpreted = 0;
2abfe22f 201 imgp->interpreter_name[0] = 0;
984263bc
MD
202 imgp->auxargs = NULL;
203 imgp->vp = NULL;
204 imgp->firstpage = NULL;
205 imgp->ps_strings = 0;
18f40545 206 imgp->image_header = NULL;
2bd9d75c
DRJ
207
208interpret:
984263bc
MD
209
210 /*
fad57d0e
MD
211 * Translate the file name to a vnode. Unlock the cache entry to
212 * improve parallelism for programs exec'd in parallel.
984263bc 213 */
fad57d0e
MD
214 if ((error = nlookup(nd)) != 0)
215 goto exec_fail;
28623bf9 216 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
fad57d0e
MD
217 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
218 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 219 cache_unlock(&nd->nl_nch);
fad57d0e 220 if (error)
984263bc 221 goto exec_fail;
984263bc
MD
222
223 /*
224 * Check file permissions (also 'opens' file)
225 */
226 error = exec_check_permissions(imgp);
227 if (error) {
a11aaa81 228 vn_unlock(imgp->vp);
984263bc
MD
229 goto exec_fail_dealloc;
230 }
231
232 error = exec_map_first_page(imgp);
a11aaa81 233 vn_unlock(imgp->vp);
984263bc
MD
234 if (error)
235 goto exec_fail_dealloc;
236
2bd9d75c 237 if (debug_execve_args && imgp->interpreted) {
6ea70f76
SW
238 kprintf(" target is interpreted -- recursive pass\n");
239 kprintf(" interpreter: %s\n", imgp->interpreter_name);
2bd9d75c
DRJ
240 print_execve_args(args);
241 }
242
984263bc
MD
243 /*
244 * If the current process has a special image activator it
245 * wants to try first, call it. For example, emulating shell
246 * scripts differently.
247 */
248 error = -1;
249 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
250 error = img_first(imgp);
251
29802dbb
MD
252 /*
253 * If the vnode has a registered vmspace, exec the vmspace
254 */
255 if (error == -1 && imgp->vp->v_resident) {
256 error = exec_resident_imgact(imgp);
257 }
258
984263bc
MD
259 /*
260 * Loop through the list of image activators, calling each one.
261 * An activator returns -1 if there is no match, 0 on success,
262 * and an error otherwise.
263 */
264 for (i = 0; error == -1 && execsw[i]; ++i) {
265 if (execsw[i]->ex_imgact == NULL ||
266 execsw[i]->ex_imgact == img_first) {
267 continue;
268 }
269 error = (*execsw[i]->ex_imgact)(imgp);
270 }
271
272 if (error) {
273 if (error == -1)
274 error = ENOEXEC;
275 goto exec_fail_dealloc;
276 }
277
278 /*
279 * Special interpreter operation, cleanup and loop up to try to
280 * activate the interpreter.
281 */
282 if (imgp->interpreted) {
283 exec_unmap_first_page(imgp);
fad57d0e
MD
284 nlookup_done(nd);
285 vrele(imgp->vp);
286 imgp->vp = NULL;
287 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
288 NLC_FOLLOW);
289 if (error)
290 goto exec_fail;
984263bc
MD
291 goto interpret;
292 }
293
294 /*
295 * Copy out strings (args and env) and initialize stack base
296 */
297 stack_base = exec_copyout_strings(imgp);
298 p->p_vmspace->vm_minsaddr = (char *)stack_base;
299
300 /*
301 * If custom stack fixup routine present for this process
29802dbb
MD
302 * let it do the stack setup. If we are running a resident
303 * image there is no auxinfo or other image activator context
304 * so don't try to add fixups to the stack.
305 *
984263bc
MD
306 * Else stuff argument count as first item on stack
307 */
29802dbb 308 if (p->p_sysent->sv_fixup && imgp->resident == 0)
984263bc
MD
309 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
310 else
2bd9d75c 311 suword(--stack_base, imgp->args->argc);
984263bc
MD
312
313 /*
314 * For security and other reasons, the file descriptor table cannot
315 * be shared after an exec.
316 */
317 if (p->p_fd->fd_refcnt > 1) {
318 struct filedesc *tmp;
319
320 tmp = fdcopy(p);
321 fdfree(p);
322 p->p_fd = tmp;
323 }
324
325 /*
326 * For security and other reasons, signal handlers cannot
327 * be shared after an exec. The new proces gets a copy of the old
328 * handlers. In execsigs(), the new process will have its signals
329 * reset.
330 */
b1b4e5a6
SS
331 if (p->p_sigacts->ps_refcnt > 1) {
332 struct sigacts *newsigacts;
984263bc 333
b1b4e5a6 334 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts),
984263bc 335 M_SUBPROC, M_WAITOK);
b1b4e5a6
SS
336 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts));
337 p->p_sigacts->ps_refcnt--;
338 p->p_sigacts = newsigacts;
339 p->p_sigacts->ps_refcnt = 1;
984263bc
MD
340 }
341
0daa37a5
MD
342 /*
343 * For security and other reasons virtual kernels cannot be
344 * inherited by an exec. This also allows a virtual kernel
345 * to fork/exec unrelated applications.
346 */
4a22e893
MD
347 if (p->p_vkernel)
348 vkernel_exit(p);
0daa37a5 349
984263bc
MD
350 /* Stop profiling */
351 stopprofclock(p);
352
353 /* close files on exec */
354 fdcloseexec(p);
355
356 /* reset caught signals */
357 execsigs(p);
358
359 /* name this process - nameiexec(p, ndp) */
28623bf9
MD
360 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
361 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
984263bc 362 p->p_comm[len] = 0;
08f2f1bb 363 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
984263bc
MD
364
365 /*
366 * mark as execed, wakeup the process that vforked (if any) and tell
367 * it that it now has its own resources back
368 */
369 p->p_flag |= P_EXEC;
370 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
371 p->p_flag &= ~P_PPWAIT;
372 wakeup((caddr_t)p->p_pptr);
373 }
374
375 /*
376 * Implement image setuid/setgid.
377 *
378 * Don't honor setuid/setgid if the filesystem prohibits it or if
379 * the process is being traced.
380 */
381 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
382 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
383 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
384 (p->p_flag & P_TRACED) == 0) {
385 /*
386 * Turn off syscall tracing for set-id programs, except for
387 * root. Record any set-id flags first to make sure that
388 * we do not regain any tracing during a possible block.
389 */
41c20dac 390 setsugid();
29f58392
MD
391 if (p->p_tracenode && suser(td) != 0) {
392 ktrdestroy(&p->p_tracenode);
393 p->p_traceflag = 0;
984263bc
MD
394 }
395 /* Close any file descriptors 0..2 that reference procfs */
396 setugidsafety(p);
397 /* Make sure file descriptors 0..2 are in use. */
398 error = fdcheckstd(p);
399 if (error != 0)
400 goto exec_fail_dealloc;
401 /*
402 * Set the new credentials.
403 */
e9a372eb 404 cratom(&p->p_ucred);
984263bc 405 if (attr.va_mode & VSUID)
41c20dac 406 change_euid(attr.va_uid);
984263bc
MD
407 if (attr.va_mode & VSGID)
408 p->p_ucred->cr_gid = attr.va_gid;
98a7f915
MD
409
410 /*
411 * Clear local varsym variables
412 */
413 varsymset_clean(&p->p_varsymset);
984263bc 414 } else {
41c20dac
MD
415 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
416 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
984263bc
MD
417 p->p_flag &= ~P_SUGID;
418 }
419
420 /*
421 * Implement correct POSIX saved-id behavior.
422 */
115ccd83
MD
423 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
424 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
425 cratom(&p->p_ucred);
426 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
427 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
428 }
984263bc
MD
429
430 /*
431 * Store the vp for use in procfs
432 */
433 if (p->p_textvp) /* release old reference */
434 vrele(p->p_textvp);
fad57d0e
MD
435 p->p_textvp = imgp->vp;
436 vref(p->p_textvp);
984263bc
MD
437
438 /*
439 * Notify others that we exec'd, and clear the P_INEXEC flag
440 * as we're now a bona fide freshly-execed process.
441 */
442 KNOTE(&p->p_klist, NOTE_EXEC);
443 p->p_flag &= ~P_INEXEC;
444
445 /*
446 * If tracing the process, trap to debugger so breakpoints
447 * can be set before the program executes.
448 */
449 STOPEVENT(p, S_EXEC, 0);
450
451 if (p->p_flag & P_TRACED)
84204577 452 ksignal(p, SIGTRAP);
984263bc
MD
453
454 /* clear "fork but no exec" flag, as we _are_ execing */
455 p->p_acflag &= ~AFORK;
456
457 /* Set values passed into the program in registers. */
08f2f1bb 458 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
984263bc
MD
459 imgp->ps_strings);
460
461 /* Free any previous argument cache */
462 if (p->p_args && --p->p_args->ar_ref == 0)
463 FREE(p->p_args, M_PARGS);
464 p->p_args = NULL;
465
466 /* Cache arguments if they fit inside our allowance */
2bd9d75c 467 i = imgp->args->begin_envv - imgp->args->begin_argv;
984263bc
MD
468 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
469 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
470 M_PARGS, M_WAITOK);
471 p->p_args->ar_ref = 1;
472 p->p_args->ar_length = i;
2bd9d75c 473 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
984263bc
MD
474 }
475
476exec_fail_dealloc:
477
478 /*
479 * free various allocated resources
480 */
481 if (imgp->firstpage)
482 exec_unmap_first_page(imgp);
483
984263bc 484 if (imgp->vp) {
984263bc 485 vrele(imgp->vp);
fad57d0e 486 imgp->vp = NULL;
984263bc
MD
487 }
488
a6f89c72
DR
489 if (error == 0) {
490 ++mycpu->gd_cnt.v_exec;
984263bc 491 return (0);
a6f89c72 492 }
984263bc
MD
493
494exec_fail:
dc52e1cc
MD
495 /*
496 * we're done here, clear P_INEXEC if we were the ones that
497 * set it. Otherwise if vmspace_destroyed is still set we
498 * raced another thread and that thread is responsible for
499 * clearing it.
500 */
501 if (imgp->vmspace_destroyed & 2)
502 p->p_flag &= ~P_INEXEC;
984263bc
MD
503 if (imgp->vmspace_destroyed) {
504 /* sorry, no more process anymore. exit gracefully */
41c20dac 505 exit1(W_EXITCODE(0, SIGABRT));
984263bc
MD
506 /* NOT REACHED */
507 return(0);
508 } else {
509 return(error);
510 }
511}
512
2bd9d75c
DRJ
513/*
514 * execve() system call.
515 */
516int
753fd850 517sys_execve(struct execve_args *uap)
2bd9d75c 518{
fad57d0e 519 struct nlookupdata nd;
2bd9d75c
DRJ
520 struct image_args args;
521 int error;
522
fad57d0e
MD
523 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
524 if (error == 0) {
525 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
526 uap->argv, uap->envv);
527 }
2abfe22f
MD
528 if (error == 0)
529 error = kern_execve(&nd, &args);
fad57d0e 530 nlookup_done(&nd);
2bd9d75c
DRJ
531 exec_free_args(&args);
532
533 /*
534 * The syscall result is returned in registers to the new program.
535 * Linux will register %edx as an atexit function and we must be
536 * sure to set it to 0. XXX
537 */
538 if (error == 0)
539 uap->sysmsg_result64 = 0;
540
541 return (error);
542}
543
984263bc 544int
41c20dac 545exec_map_first_page(struct image_params *imgp)
984263bc 546{
5fd012e0 547 int rv, i;
984263bc
MD
548 int initial_pagein;
549 vm_page_t ma[VM_INITIAL_PAGEIN];
06ecca5a 550 vm_page_t m;
984263bc
MD
551 vm_object_t object;
552
18f40545 553 if (imgp->firstpage)
984263bc 554 exec_unmap_first_page(imgp);
984263bc 555
fad57d0e 556 /*
7540ab49 557 * The file has to be mappable.
fad57d0e 558 */
7540ab49 559 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 560 return (EIO);
984263bc 561
06ecca5a
MD
562 /*
563 * We shouldn't need protection for vm_page_grab() but we certainly
564 * need it for the lookup loop below (lookup/busy race), since
565 * an interrupt can unbusy and free the page before our busy check.
566 */
5fd012e0 567 crit_enter();
06ecca5a 568 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
984263bc 569
06ecca5a
MD
570 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
571 ma[0] = m;
984263bc
MD
572 initial_pagein = VM_INITIAL_PAGEIN;
573 if (initial_pagein > object->size)
574 initial_pagein = object->size;
575 for (i = 1; i < initial_pagein; i++) {
06ecca5a
MD
576 if ((m = vm_page_lookup(object, i)) != NULL) {
577 if ((m->flags & PG_BUSY) || m->busy)
984263bc 578 break;
06ecca5a 579 if (m->valid)
984263bc 580 break;
06ecca5a 581 vm_page_busy(m);
984263bc 582 } else {
06ecca5a
MD
583 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
584 if (m == NULL)
984263bc
MD
585 break;
586 }
06ecca5a 587 ma[i] = m;
984263bc
MD
588 }
589 initial_pagein = i;
590
06ecca5a
MD
591 /*
592 * get_pages unbusies all the requested pages except the
593 * primary page (at index 0 in this case).
594 */
984263bc 595 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
06ecca5a 596 m = vm_page_lookup(object, 0);
984263bc 597
06ecca5a
MD
598 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
599 if (m) {
600 vm_page_protect(m, VM_PROT_NONE);
601 vm_page_free(m);
984263bc 602 }
5fd012e0 603 crit_exit();
984263bc
MD
604 return EIO;
605 }
606 }
06ecca5a
MD
607 vm_page_hold(m);
608 vm_page_wakeup(m); /* unbusy the page */
5fd012e0 609 crit_exit();
984263bc 610
4f1640d6 611 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE);
18f40545 612 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage);
984263bc
MD
613
614 return 0;
615}
616
617void
77153250 618exec_unmap_first_page(struct image_params *imgp)
984263bc 619{
18f40545
MD
620 vm_page_t m;
621
5fd012e0 622 crit_enter();
18f40545
MD
623 if (imgp->firstpage != NULL) {
624 m = sf_buf_page(imgp->firstpage);
625 sf_buf_free(imgp->firstpage);
984263bc 626 imgp->firstpage = NULL;
18f40545 627 imgp->image_header = NULL;
e4ba7818 628 vm_page_unhold(m);
984263bc 629 }
5fd012e0 630 crit_exit();
984263bc
MD
631}
632
633/*
634 * Destroy old address space, and allocate a new stack
635 * The new stack is only SGROWSIZ large because it is grown
636 * automatically in trap.c.
dc52e1cc
MD
637 *
638 * This is the point of no return.
984263bc
MD
639 */
640int
29802dbb 641exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 642{
984263bc
MD
643 struct vmspace *vmspace = imgp->proc->p_vmspace;
644 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 645 struct proc *p;
29802dbb 646 vm_map_t map;
dc52e1cc 647 int error;
984263bc 648
dc52e1cc
MD
649 /*
650 * Indicate that we cannot gracefully error out any more, kill
651 * any other threads present, and set P_INEXEC to indicate that
652 * we are now messing with the process structure proper.
653 *
654 * If killalllwps() races return an error which coupled with
655 * vmspace_destroyed will cause us to exit. This is what we
656 * want since another thread is patiently waiting for us to exit
657 * in that case.
658 */
659 p = curproc;
984263bc
MD
660 imgp->vmspace_destroyed = 1;
661
dc52e1cc
MD
662 if (curthread->td_proc->p_nthreads > 1) {
663 error = killalllwps(1);
664 if (error)
665 return (error);
666 }
667 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
668 p->p_flag |= P_INEXEC;
08f2f1bb 669
984263bc
MD
670 /*
671 * Prevent a pending AIO from modifying the new address space.
672 */
673 aio_proc_rundown(imgp->proc);
674
675 /*
676 * Blow away entire process VM, if address space not shared,
677 * otherwise, create a new VM space so that other threads are
29802dbb
MD
678 * not disrupted. If we are execing a resident vmspace we
679 * create a duplicate of it and remap the stack.
239b4df9
MD
680 *
681 * The exitingcnt test is not strictly necessary but has been
682 * included for code sanity (to make the code more deterministic).
984263bc 683 */
29802dbb
MD
684 map = &vmspace->vm_map;
685 if (vmcopy) {
686 vmspace_exec(imgp->proc, vmcopy);
687 vmspace = imgp->proc->p_vmspace;
688 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
689 map = &vmspace->vm_map;
e3161323
MD
690 } else if (vmspace->vm_sysref.refcnt == 1 &&
691 vmspace->vm_exitingcnt == 0) {
fef0fdf2 692 shmexit(vmspace);
239b4df9 693 if (vmspace->vm_upcalls)
08f2f1bb 694 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08
MD
695 pmap_remove_pages(vmspace_pmap(vmspace),
696 0, VM_MAX_USER_ADDRESS);
697 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 698 } else {
29802dbb 699 vmspace_exec(imgp->proc, NULL);
984263bc
MD
700 vmspace = imgp->proc->p_vmspace;
701 map = &vmspace->vm_map;
702 }
703
704 /* Allocate a new stack */
705 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
706 VM_PROT_ALL, VM_PROT_ALL, 0);
707 if (error)
708 return (error);
709
710 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
711 * VM_STACK case, but they are still used to monitor the size of the
712 * process stack so we can check the stack rlimit.
713 */
714 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
715 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
716
717 return(0);
718}
719
720/*
721 * Copy out argument and environment strings from the old process
722 * address space into the temporary string buffer.
723 */
724int
2bd9d75c 725exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 726 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 727{
984263bc 728 char *argp, *envp;
2bd9d75c 729 int error = 0;
984263bc
MD
730 size_t length;
731
2abfe22f 732 bzero(args, sizeof(*args));
18b3f457
MD
733
734 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
735 if (args->buf == NULL)
736 return (ENOMEM);
737 args->begin_argv = args->buf;
738 args->endp = args->begin_argv;
739 args->space = ARG_MAX;
2bd9d75c
DRJ
740
741 args->fname = args->buf + ARG_MAX;
742
984263bc 743 /*
2bd9d75c 744 * Copy the file name.
984263bc 745 */
2bd9d75c
DRJ
746 if (segflg == PATH_SYSSPACE) {
747 error = copystr(fname, args->fname, PATH_MAX, &length);
748 } else if (segflg == PATH_USERSPACE) {
749 error = copyinstr(fname, args->fname, PATH_MAX, &length);
750 }
984263bc 751
2bd9d75c 752 /*
66be6566
MD
753 * Extract argument strings. argv may not be NULL. The argv
754 * array is terminated by a NULL entry. We special-case the
755 * situation where argv[0] is NULL by passing { filename, NULL }
756 * to the new program to guarentee that the interpreter knows what
757 * file to open in case we exec an interpreted file. Note that
758 * a NULL argv[0] terminates the argv[] array.
759 *
760 * XXX the special-casing of argv[0] is historical and needs to be
761 * revisited.
2bd9d75c 762 */
66be6566
MD
763 if (argv == NULL)
764 error = EFAULT;
765 if (error == 0) {
766 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
767 if (argp == (caddr_t)-1) {
2bd9d75c 768 error = EFAULT;
66be6566 769 break;
2bd9d75c
DRJ
770 }
771 error = copyinstr(argp, args->endp,
2abfe22f 772 args->space, &length);
66be6566
MD
773 if (error) {
774 if (error == ENAMETOOLONG)
775 error = E2BIG;
776 break;
777 }
2bd9d75c
DRJ
778 args->space -= length;
779 args->endp += length;
780 args->argc++;
984263bc 781 }
66be6566
MD
782 if (args->argc == 0 && error == 0) {
783 length = strlen(args->fname) + 1;
784 if (length > args->space) {
785 error = E2BIG;
786 } else {
787 bcopy(args->fname, args->endp, length);
788 args->space -= length;
789 args->endp += length;
790 args->argc++;
791 }
792 }
984263bc
MD
793 }
794
2bd9d75c 795 args->begin_envv = args->endp;
984263bc
MD
796
797 /*
66be6566 798 * extract environment strings. envv may be NULL.
984263bc 799 */
2abfe22f 800 if (envv && error == 0) {
984263bc 801 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
802 if (envp == (caddr_t) -1) {
803 error = EFAULT;
66be6566 804 break;
984263bc 805 }
2bd9d75c
DRJ
806 error = copyinstr(envp, args->endp, args->space,
807 &length);
66be6566
MD
808 if (error) {
809 if (error == ENAMETOOLONG)
810 error = E2BIG;
811 break;
812 }
2bd9d75c
DRJ
813 args->space -= length;
814 args->endp += length;
815 args->envc++;
984263bc
MD
816 }
817 }
2bd9d75c
DRJ
818 return (error);
819}
820
821void
822exec_free_args(struct image_args *args)
823{
2abfe22f 824 if (args->buf) {
18b3f457 825 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
826 args->buf = NULL;
827 }
984263bc
MD
828}
829
830/*
831 * Copy strings out to the new process address space, constructing
832 * new arg and env vector tables. Return a pointer to the base
833 * so that it can be used as the initial stack pointer.
834 */
835register_t *
2bd9d75c 836exec_copyout_strings(struct image_params *imgp)
984263bc 837{
17a7ca1f 838 int argc, envc, sgap;
984263bc
MD
839 char **vectp;
840 char *stringp, *destp;
841 register_t *stack_base;
842 struct ps_strings *arginfo;
843 int szsigcode;
844
845 /*
846 * Calculate string base and vector table pointers.
847 * Also deal with signal trampoline code for this exec type.
848 */
849 arginfo = (struct ps_strings *)PS_STRINGS;
850 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
17a7ca1f 851 if (stackgap_random != 0)
0ced1954 852 sgap = ALIGN(karc4random() & (stackgap_random - 1));
17a7ca1f
MD
853 else
854 sgap = 0;
855 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
2bd9d75c 856 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
984263bc
MD
857
858 /*
859 * install sigcode
860 */
861 if (szsigcode)
862 copyout(imgp->proc->p_sysent->sv_sigcode,
2bd9d75c 863 ((caddr_t)arginfo - szsigcode), szsigcode);
984263bc
MD
864
865 /*
866 * If we have a valid auxargs ptr, prepare some room
867 * on the stack.
7ec1918b 868 *
984263bc
MD
869 * The '+ 2' is for the null pointers at the end of each of the
870 * arg and env vector sets, and 'AT_COUNT*2' is room for the
871 * ELF Auxargs data.
872 */
7ec1918b
MD
873 if (imgp->auxargs) {
874 vectp = (char **)(destp - (imgp->args->argc +
875 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
876 } else {
2bd9d75c 877 vectp = (char **)(destp - (imgp->args->argc +
7ec1918b
MD
878 imgp->args->envc + 2) * sizeof(char*));
879 }
880
984263bc 881 /*
31bc42c8
MD
882 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
883 * be done in crt1.o. Note that GCC 3.x aligns the stack in main.
984263bc 884 */
984263bc
MD
885
886 /*
887 * vectp also becomes our initial stack base
888 */
889 stack_base = (register_t *)vectp;
890
2bd9d75c
DRJ
891 stringp = imgp->args->begin_argv;
892 argc = imgp->args->argc;
893 envc = imgp->args->envc;
984263bc
MD
894
895 /*
896 * Copy out strings - arguments and environment.
897 */
2bd9d75c 898 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
899
900 /*
901 * Fill in "ps_strings" struct for ps, w, etc.
902 */
903 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
904 suword(&arginfo->ps_nargvstr, argc);
905
906 /*
907 * Fill in argument portion of vector table.
908 */
909 for (; argc > 0; --argc) {
910 suword(vectp++, (long)(intptr_t)destp);
911 while (*stringp++ != 0)
912 destp++;
913 destp++;
914 }
915
7a44d1cb 916 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
917 suword(vectp++, 0);
918
919 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
920 suword(&arginfo->ps_nenvstr, envc);
921
922 /*
923 * Fill in environment portion of vector table.
924 */
925 for (; envc > 0; --envc) {
926 suword(vectp++, (long)(intptr_t)destp);
927 while (*stringp++ != 0)
928 destp++;
929 destp++;
930 }
931
932 /* end of vector table is a null pointer */
933 suword(vectp, 0);
934
935 return (stack_base);
936}
937
938/*
939 * Check permissions of file to execute.
940 * Return 0 for success or error code on failure.
941 */
942int
77153250 943exec_check_permissions(struct image_params *imgp)
984263bc
MD
944{
945 struct proc *p = imgp->proc;
946 struct vnode *vp = imgp->vp;
947 struct vattr *attr = imgp->attr;
948 int error;
949
950 /* Get file attributes */
87de5057 951 error = VOP_GETATTR(vp, attr);
984263bc
MD
952 if (error)
953 return (error);
954
955 /*
956 * 1) Check if file execution is disabled for the filesystem that this
957 * file resides on.
958 * 2) Insure that at least one execute bit is on - otherwise root
959 * will always succeed, and we don't want to happen unless the
960 * file really is executable.
961 * 3) Insure that the file is a regular file.
962 */
963 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
964 ((attr->va_mode & 0111) == 0) ||
965 (attr->va_type != VREG)) {
966 return (EACCES);
967 }
968
969 /*
970 * Zero length files can't be exec'd
971 */
972 if (attr->va_size == 0)
973 return (ENOEXEC);
974
975 /*
976 * Check for execute permission to file based on current credentials.
977 */
87de5057 978 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
979 if (error)
980 return (error);
981
982 /*
983 * Check number of open-for-writes on the file and deny execution
984 * if there are any.
985 */
986 if (vp->v_writecount)
987 return (ETXTBSY);
988
989 /*
7540ab49
MD
990 * Call filesystem specific open routine, which allows us to read,
991 * write, and mmap the file. Without the VOP_OPEN we can only
992 * stat the file.
984263bc 993 */
87de5057 994 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
995 if (error)
996 return (error);
997
998 return (0);
999}
1000
1001/*
1002 * Exec handler registration
1003 */
1004int
77153250 1005exec_register(const struct execsw *execsw_arg)
984263bc
MD
1006{
1007 const struct execsw **es, **xs, **newexecsw;
1008 int count = 2; /* New slot and trailing NULL */
1009
1010 if (execsw)
1011 for (es = execsw; *es; es++)
1012 count++;
efda3bd0 1013 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1014 if (newexecsw == NULL)
1015 return ENOMEM;
1016 xs = newexecsw;
1017 if (execsw)
1018 for (es = execsw; *es; es++)
1019 *xs++ = *es;
1020 *xs++ = execsw_arg;
1021 *xs = NULL;
1022 if (execsw)
efda3bd0 1023 kfree(execsw, M_TEMP);
984263bc
MD
1024 execsw = newexecsw;
1025 return 0;
1026}
1027
1028int
77153250 1029exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1030{
1031 const struct execsw **es, **xs, **newexecsw;
1032 int count = 1;
1033
1034 if (execsw == NULL)
ae7e74d4 1035 panic("unregister with no handlers left?");
984263bc
MD
1036
1037 for (es = execsw; *es; es++) {
1038 if (*es == execsw_arg)
1039 break;
1040 }
1041 if (*es == NULL)
1042 return ENOENT;
1043 for (es = execsw; *es; es++)
1044 if (*es != execsw_arg)
1045 count++;
efda3bd0 1046 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1047 if (newexecsw == NULL)
1048 return ENOMEM;
1049 xs = newexecsw;
1050 for (es = execsw; *es; es++)
1051 if (*es != execsw_arg)
1052 *xs++ = *es;
1053 *xs = NULL;
1054 if (execsw)
efda3bd0 1055 kfree(execsw, M_TEMP);
984263bc
MD
1056 execsw = newexecsw;
1057 return 0;
1058}