SILI - Fix embarassing bug for non-NCQ disk accesses > 128G
[dragonfly.git] / sys / kern / kern_exec.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1993, David Greenman
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD: src/sys/kern/kern_exec.c,v 1.107.2.15 2002/07/30 15:40:46 nectar Exp $
0aa16b5d 27 * $DragonFly: src/sys/kern/kern_exec.c,v 1.64 2008/10/26 04:29:19 sephe Exp $
984263bc
MD
28 */
29
30#include <sys/param.h>
31#include <sys/systm.h>
32#include <sys/sysproto.h>
33#include <sys/kernel.h>
34#include <sys/mount.h>
35#include <sys/filedesc.h>
36#include <sys/fcntl.h>
37#include <sys/acct.h>
38#include <sys/exec.h>
39#include <sys/imgact.h>
40#include <sys/imgact_elf.h>
2bd9d75c 41#include <sys/kern_syscall.h>
984263bc
MD
42#include <sys/wait.h>
43#include <sys/malloc.h>
44#include <sys/proc.h>
895c1f85 45#include <sys/priv.h>
29f58392 46#include <sys/ktrace.h>
984263bc
MD
47#include <sys/signalvar.h>
48#include <sys/pioctl.h>
fad57d0e 49#include <sys/nlookup.h>
18f40545 50#include <sys/sfbuf.h>
984263bc
MD
51#include <sys/sysent.h>
52#include <sys/shm.h>
53#include <sys/sysctl.h>
54#include <sys/vnode.h>
a6f89c72 55#include <sys/vmmeter.h>
984263bc 56#include <sys/aio.h>
17a7ca1f 57#include <sys/libkern.h>
984263bc
MD
58
59#include <vm/vm.h>
60#include <vm/vm_param.h>
61#include <sys/lock.h>
62#include <vm/pmap.h>
63#include <vm/vm_page.h>
64#include <vm/vm_map.h>
65#include <vm/vm_kern.h>
66#include <vm/vm_extern.h>
67#include <vm/vm_object.h>
a55afca2 68#include <vm/vnode_pager.h>
984263bc
MD
69#include <vm/vm_pager.h>
70
71#include <sys/user.h>
527fddf7 72#include <sys/reg.h>
984263bc 73
5fd012e0
MD
74#include <sys/thread2.h>
75
984263bc 76MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
18b3f457 77MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
984263bc 78
402ed7e1 79static register_t *exec_copyout_strings (struct image_params *);
984263bc
MD
80
81/* XXX This should be vm_size_t. */
82static u_long ps_strings = PS_STRINGS;
83SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
84
85/* XXX This should be vm_size_t. */
86static u_long usrstack = USRSTACK;
87SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
88
89u_long ps_arg_cache_limit = PAGE_SIZE / 16;
90SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
91 &ps_arg_cache_limit, 0, "");
92
93int ps_argsopen = 1;
94SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
95
2bd9d75c
DRJ
96void print_execve_args(struct image_args *args);
97int debug_execve_args = 0;
98SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
99 0, "");
100
18b3f457
MD
101/*
102 * Exec arguments object cache
103 */
104static struct objcache *exec_objcache;
105
106static
107void
108exec_objcache_init(void *arg __unused)
109{
0aa16b5d
SZ
110 int cluster_limit;
111
112 cluster_limit = 16; /* up to this many objects */
18b3f457
MD
113 exec_objcache = objcache_create_mbacked(
114 M_EXECARGS, PATH_MAX + ARG_MAX,
0aa16b5d 115 &cluster_limit,
18b3f457
MD
116 2, /* minimal magazine capacity */
117 NULL, NULL, NULL);
118}
119SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
120
17a7ca1f
MD
121/*
122 * stackgap_random specifies if the stackgap should have a random size added
123 * to it. It must be a power of 2. If non-zero, the stack gap will be
0ced1954 124 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
17a7ca1f
MD
125 */
126static int stackgap_random = 1024;
127static int
128sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
129{
130 int error, new_val;
131 new_val = stackgap_random;
132 error = sysctl_handle_int(oidp, &new_val, 0, req);
133 if (error != 0 || req->newptr == NULL)
134 return (error);
135 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val))
136 return (EINVAL);
137 stackgap_random = new_val;
138
139 return(0);
140}
141
142SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT,
143 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
144
2bd9d75c
DRJ
145void
146print_execve_args(struct image_args *args)
147{
148 char *cp;
149 int ndx;
150
151 cp = args->begin_argv;
152 for (ndx = 0; ndx < args->argc; ndx++) {
6ea70f76 153 kprintf("\targv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
154 while (*cp++ != '\0');
155 }
156 for (ndx = 0; ndx < args->envc; ndx++) {
6ea70f76 157 kprintf("\tenvv[%d]: %s\n", ndx, cp);
2bd9d75c
DRJ
158 while (*cp++ != '\0');
159 }
160}
161
984263bc
MD
162/*
163 * Each of the items is a pointer to a `const struct execsw', hence the
164 * double pointer here.
165 */
166static const struct execsw **execsw;
167
5417cd57
SS
168/*
169 * Replace current vmspace with a new binary.
170 * Returns 0 on success, > 0 on recoverable error (use as errno).
90947cb1 171 * Returns -1 on lethal error which demands killing of the current
5417cd57
SS
172 * process!
173 */
984263bc 174int
fad57d0e 175kern_execve(struct nlookupdata *nd, struct image_args *args)
984263bc 176{
dadab5e9 177 struct thread *td = curthread;
08f2f1bb 178 struct lwp *lp = td->td_lwp;
dadab5e9 179 struct proc *p = td->td_proc;
984263bc
MD
180 register_t *stack_base;
181 int error, len, i;
182 struct image_params image_params, *imgp;
183 struct vattr attr;
402ed7e1 184 int (*img_first) (struct image_params *);
984263bc 185
2bd9d75c 186 if (debug_execve_args) {
6ea70f76 187 kprintf("%s()\n", __func__);
2bd9d75c
DRJ
188 print_execve_args(args);
189 }
190
dadab5e9 191 KKASSERT(p);
984263bc
MD
192 imgp = &image_params;
193
194 /*
dc52e1cc
MD
195 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make
196 * no modifications to the process at all until we get there.
197 *
198 * Note that multiple threads may be trying to exec at the same
199 * time. exec_new_vmspace() handles that too.
984263bc 200 */
984263bc
MD
201
202 /*
203 * Initialize part of the common data
204 */
205 imgp->proc = p;
2bd9d75c 206 imgp->args = args;
984263bc 207 imgp->attr = &attr;
984263bc 208 imgp->entry_addr = 0;
29802dbb 209 imgp->resident = 0;
984263bc
MD
210 imgp->vmspace_destroyed = 0;
211 imgp->interpreted = 0;
2abfe22f 212 imgp->interpreter_name[0] = 0;
984263bc
MD
213 imgp->auxargs = NULL;
214 imgp->vp = NULL;
215 imgp->firstpage = NULL;
216 imgp->ps_strings = 0;
18f40545 217 imgp->image_header = NULL;
2bd9d75c
DRJ
218
219interpret:
984263bc
MD
220
221 /*
fad57d0e
MD
222 * Translate the file name to a vnode. Unlock the cache entry to
223 * improve parallelism for programs exec'd in parallel.
984263bc 224 */
fad57d0e
MD
225 if ((error = nlookup(nd)) != 0)
226 goto exec_fail;
28623bf9 227 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
fad57d0e
MD
228 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
229 nd->nl_flags &= ~NLC_NCPISLOCKED;
28623bf9 230 cache_unlock(&nd->nl_nch);
fad57d0e 231 if (error)
984263bc 232 goto exec_fail;
984263bc
MD
233
234 /*
235 * Check file permissions (also 'opens' file)
236 */
237 error = exec_check_permissions(imgp);
238 if (error) {
a11aaa81 239 vn_unlock(imgp->vp);
984263bc
MD
240 goto exec_fail_dealloc;
241 }
242
243 error = exec_map_first_page(imgp);
a11aaa81 244 vn_unlock(imgp->vp);
984263bc
MD
245 if (error)
246 goto exec_fail_dealloc;
247
2bd9d75c 248 if (debug_execve_args && imgp->interpreted) {
6ea70f76
SW
249 kprintf(" target is interpreted -- recursive pass\n");
250 kprintf(" interpreter: %s\n", imgp->interpreter_name);
2bd9d75c
DRJ
251 print_execve_args(args);
252 }
253
984263bc
MD
254 /*
255 * If the current process has a special image activator it
256 * wants to try first, call it. For example, emulating shell
257 * scripts differently.
258 */
259 error = -1;
260 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
261 error = img_first(imgp);
262
29802dbb
MD
263 /*
264 * If the vnode has a registered vmspace, exec the vmspace
265 */
266 if (error == -1 && imgp->vp->v_resident) {
267 error = exec_resident_imgact(imgp);
268 }
269
984263bc
MD
270 /*
271 * Loop through the list of image activators, calling each one.
272 * An activator returns -1 if there is no match, 0 on success,
273 * and an error otherwise.
274 */
275 for (i = 0; error == -1 && execsw[i]; ++i) {
276 if (execsw[i]->ex_imgact == NULL ||
277 execsw[i]->ex_imgact == img_first) {
278 continue;
279 }
280 error = (*execsw[i]->ex_imgact)(imgp);
281 }
282
283 if (error) {
284 if (error == -1)
285 error = ENOEXEC;
286 goto exec_fail_dealloc;
287 }
288
289 /*
290 * Special interpreter operation, cleanup and loop up to try to
291 * activate the interpreter.
292 */
293 if (imgp->interpreted) {
294 exec_unmap_first_page(imgp);
fad57d0e
MD
295 nlookup_done(nd);
296 vrele(imgp->vp);
297 imgp->vp = NULL;
298 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
299 NLC_FOLLOW);
300 if (error)
301 goto exec_fail;
984263bc
MD
302 goto interpret;
303 }
304
305 /*
306 * Copy out strings (args and env) and initialize stack base
307 */
308 stack_base = exec_copyout_strings(imgp);
309 p->p_vmspace->vm_minsaddr = (char *)stack_base;
310
311 /*
312 * If custom stack fixup routine present for this process
29802dbb
MD
313 * let it do the stack setup. If we are running a resident
314 * image there is no auxinfo or other image activator context
315 * so don't try to add fixups to the stack.
316 *
984263bc
MD
317 * Else stuff argument count as first item on stack
318 */
29802dbb 319 if (p->p_sysent->sv_fixup && imgp->resident == 0)
984263bc
MD
320 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
321 else
2bd9d75c 322 suword(--stack_base, imgp->args->argc);
984263bc
MD
323
324 /*
325 * For security and other reasons, the file descriptor table cannot
326 * be shared after an exec.
327 */
328 if (p->p_fd->fd_refcnt > 1) {
329 struct filedesc *tmp;
330
331 tmp = fdcopy(p);
332 fdfree(p);
333 p->p_fd = tmp;
334 }
335
336 /*
337 * For security and other reasons, signal handlers cannot
338 * be shared after an exec. The new proces gets a copy of the old
339 * handlers. In execsigs(), the new process will have its signals
340 * reset.
341 */
b1b4e5a6
SS
342 if (p->p_sigacts->ps_refcnt > 1) {
343 struct sigacts *newsigacts;
984263bc 344
b1b4e5a6 345 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts),
984263bc 346 M_SUBPROC, M_WAITOK);
b1b4e5a6
SS
347 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts));
348 p->p_sigacts->ps_refcnt--;
349 p->p_sigacts = newsigacts;
350 p->p_sigacts->ps_refcnt = 1;
984263bc
MD
351 }
352
0daa37a5
MD
353 /*
354 * For security and other reasons virtual kernels cannot be
355 * inherited by an exec. This also allows a virtual kernel
356 * to fork/exec unrelated applications.
357 */
4a22e893
MD
358 if (p->p_vkernel)
359 vkernel_exit(p);
0daa37a5 360
984263bc
MD
361 /* Stop profiling */
362 stopprofclock(p);
363
364 /* close files on exec */
365 fdcloseexec(p);
366
367 /* reset caught signals */
368 execsigs(p);
369
370 /* name this process - nameiexec(p, ndp) */
28623bf9
MD
371 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
372 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
984263bc 373 p->p_comm[len] = 0;
08f2f1bb 374 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
984263bc
MD
375
376 /*
377 * mark as execed, wakeup the process that vforked (if any) and tell
378 * it that it now has its own resources back
379 */
380 p->p_flag |= P_EXEC;
381 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
382 p->p_flag &= ~P_PPWAIT;
383 wakeup((caddr_t)p->p_pptr);
384 }
385
386 /*
387 * Implement image setuid/setgid.
388 *
389 * Don't honor setuid/setgid if the filesystem prohibits it or if
390 * the process is being traced.
391 */
392 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
393 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
394 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
395 (p->p_flag & P_TRACED) == 0) {
396 /*
397 * Turn off syscall tracing for set-id programs, except for
398 * root. Record any set-id flags first to make sure that
399 * we do not regain any tracing during a possible block.
400 */
41c20dac 401 setsugid();
895c1f85 402 if (p->p_tracenode && priv_check(td, PRIV_ROOT) != 0) {
29f58392
MD
403 ktrdestroy(&p->p_tracenode);
404 p->p_traceflag = 0;
984263bc
MD
405 }
406 /* Close any file descriptors 0..2 that reference procfs */
407 setugidsafety(p);
408 /* Make sure file descriptors 0..2 are in use. */
409 error = fdcheckstd(p);
410 if (error != 0)
411 goto exec_fail_dealloc;
412 /*
413 * Set the new credentials.
414 */
e9a372eb 415 cratom(&p->p_ucred);
984263bc 416 if (attr.va_mode & VSUID)
41c20dac 417 change_euid(attr.va_uid);
984263bc
MD
418 if (attr.va_mode & VSGID)
419 p->p_ucred->cr_gid = attr.va_gid;
98a7f915
MD
420
421 /*
422 * Clear local varsym variables
423 */
424 varsymset_clean(&p->p_varsymset);
984263bc 425 } else {
41c20dac
MD
426 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
427 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
984263bc
MD
428 p->p_flag &= ~P_SUGID;
429 }
430
431 /*
432 * Implement correct POSIX saved-id behavior.
433 */
115ccd83
MD
434 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
435 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
436 cratom(&p->p_ucred);
437 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
438 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
439 }
984263bc
MD
440
441 /*
442 * Store the vp for use in procfs
443 */
444 if (p->p_textvp) /* release old reference */
445 vrele(p->p_textvp);
fad57d0e
MD
446 p->p_textvp = imgp->vp;
447 vref(p->p_textvp);
984263bc
MD
448
449 /*
450 * Notify others that we exec'd, and clear the P_INEXEC flag
451 * as we're now a bona fide freshly-execed process.
452 */
453 KNOTE(&p->p_klist, NOTE_EXEC);
454 p->p_flag &= ~P_INEXEC;
455
456 /*
457 * If tracing the process, trap to debugger so breakpoints
458 * can be set before the program executes.
459 */
460 STOPEVENT(p, S_EXEC, 0);
461
462 if (p->p_flag & P_TRACED)
84204577 463 ksignal(p, SIGTRAP);
984263bc
MD
464
465 /* clear "fork but no exec" flag, as we _are_ execing */
466 p->p_acflag &= ~AFORK;
467
468 /* Set values passed into the program in registers. */
08f2f1bb 469 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
984263bc
MD
470 imgp->ps_strings);
471
349433c9
MD
472 /* Set the access time on the vnode */
473 vn_mark_atime(imgp->vp, td);
474
984263bc
MD
475 /* Free any previous argument cache */
476 if (p->p_args && --p->p_args->ar_ref == 0)
477 FREE(p->p_args, M_PARGS);
478 p->p_args = NULL;
479
480 /* Cache arguments if they fit inside our allowance */
2bd9d75c 481 i = imgp->args->begin_envv - imgp->args->begin_argv;
984263bc
MD
482 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
483 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
484 M_PARGS, M_WAITOK);
485 p->p_args->ar_ref = 1;
486 p->p_args->ar_length = i;
2bd9d75c 487 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
984263bc
MD
488 }
489
490exec_fail_dealloc:
491
492 /*
493 * free various allocated resources
494 */
495 if (imgp->firstpage)
496 exec_unmap_first_page(imgp);
497
984263bc 498 if (imgp->vp) {
984263bc 499 vrele(imgp->vp);
fad57d0e 500 imgp->vp = NULL;
984263bc
MD
501 }
502
a6f89c72
DR
503 if (error == 0) {
504 ++mycpu->gd_cnt.v_exec;
984263bc 505 return (0);
a6f89c72 506 }
984263bc
MD
507
508exec_fail:
dc52e1cc
MD
509 /*
510 * we're done here, clear P_INEXEC if we were the ones that
511 * set it. Otherwise if vmspace_destroyed is still set we
512 * raced another thread and that thread is responsible for
513 * clearing it.
514 */
515 if (imgp->vmspace_destroyed & 2)
516 p->p_flag &= ~P_INEXEC;
984263bc 517 if (imgp->vmspace_destroyed) {
5417cd57
SS
518 /*
519 * Sorry, no more process anymore. exit gracefully.
520 * However we can't die right here, because our
521 * caller might have to clean up, so indicate a
90947cb1 522 * lethal error by returning -1.
5417cd57
SS
523 */
524 return(-1);
984263bc
MD
525 } else {
526 return(error);
527 }
528}
529
2bd9d75c
DRJ
530/*
531 * execve() system call.
532 */
533int
753fd850 534sys_execve(struct execve_args *uap)
2bd9d75c 535{
fad57d0e 536 struct nlookupdata nd;
2bd9d75c
DRJ
537 struct image_args args;
538 int error;
539
fad57d0e
MD
540 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
541 if (error == 0) {
542 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
543 uap->argv, uap->envv);
544 }
2abfe22f
MD
545 if (error == 0)
546 error = kern_execve(&nd, &args);
fad57d0e 547 nlookup_done(&nd);
2bd9d75c
DRJ
548 exec_free_args(&args);
549
5417cd57 550 if (error < 0) {
90947cb1 551 /* We hit a lethal error condition. Let's die now. */
5417cd57
SS
552 exit1(W_EXITCODE(0, SIGABRT));
553 /* NOTREACHED */
554 }
555
2bd9d75c
DRJ
556 /*
557 * The syscall result is returned in registers to the new program.
558 * Linux will register %edx as an atexit function and we must be
559 * sure to set it to 0. XXX
560 */
561 if (error == 0)
562 uap->sysmsg_result64 = 0;
563
564 return (error);
565}
566
984263bc 567int
41c20dac 568exec_map_first_page(struct image_params *imgp)
984263bc 569{
5fd012e0 570 int rv, i;
984263bc
MD
571 int initial_pagein;
572 vm_page_t ma[VM_INITIAL_PAGEIN];
06ecca5a 573 vm_page_t m;
984263bc
MD
574 vm_object_t object;
575
18f40545 576 if (imgp->firstpage)
984263bc 577 exec_unmap_first_page(imgp);
984263bc 578
fad57d0e 579 /*
7540ab49 580 * The file has to be mappable.
fad57d0e 581 */
7540ab49 582 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 583 return (EIO);
984263bc 584
06ecca5a
MD
585 /*
586 * We shouldn't need protection for vm_page_grab() but we certainly
587 * need it for the lookup loop below (lookup/busy race), since
588 * an interrupt can unbusy and free the page before our busy check.
589 */
5fd012e0 590 crit_enter();
06ecca5a 591 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
984263bc 592
06ecca5a
MD
593 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
594 ma[0] = m;
984263bc
MD
595 initial_pagein = VM_INITIAL_PAGEIN;
596 if (initial_pagein > object->size)
597 initial_pagein = object->size;
598 for (i = 1; i < initial_pagein; i++) {
06ecca5a
MD
599 if ((m = vm_page_lookup(object, i)) != NULL) {
600 if ((m->flags & PG_BUSY) || m->busy)
984263bc 601 break;
06ecca5a 602 if (m->valid)
984263bc 603 break;
06ecca5a 604 vm_page_busy(m);
984263bc 605 } else {
06ecca5a
MD
606 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
607 if (m == NULL)
984263bc
MD
608 break;
609 }
06ecca5a 610 ma[i] = m;
984263bc
MD
611 }
612 initial_pagein = i;
613
06ecca5a
MD
614 /*
615 * get_pages unbusies all the requested pages except the
a55afca2
MD
616 * primary page (at index 0 in this case). The primary
617 * page may have been wired during the pagein (e.g. by
618 * the buffer cache) so vnode_pager_freepage() must be
619 * used to properly release it.
06ecca5a 620 */
984263bc 621 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
06ecca5a 622 m = vm_page_lookup(object, 0);
984263bc 623
06ecca5a
MD
624 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
625 if (m) {
626 vm_page_protect(m, VM_PROT_NONE);
a55afca2 627 vnode_pager_freepage(m);
984263bc 628 }
5fd012e0 629 crit_exit();
984263bc
MD
630 return EIO;
631 }
632 }
06ecca5a
MD
633 vm_page_hold(m);
634 vm_page_wakeup(m); /* unbusy the page */
5fd012e0 635 crit_exit();
984263bc 636
4f1640d6 637 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE);
18f40545 638 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage);
984263bc
MD
639
640 return 0;
641}
642
643void
77153250 644exec_unmap_first_page(struct image_params *imgp)
984263bc 645{
18f40545
MD
646 vm_page_t m;
647
5fd012e0 648 crit_enter();
18f40545
MD
649 if (imgp->firstpage != NULL) {
650 m = sf_buf_page(imgp->firstpage);
651 sf_buf_free(imgp->firstpage);
984263bc 652 imgp->firstpage = NULL;
18f40545 653 imgp->image_header = NULL;
e4ba7818 654 vm_page_unhold(m);
984263bc 655 }
5fd012e0 656 crit_exit();
984263bc
MD
657}
658
659/*
660 * Destroy old address space, and allocate a new stack
661 * The new stack is only SGROWSIZ large because it is grown
662 * automatically in trap.c.
dc52e1cc
MD
663 *
664 * This is the point of no return.
984263bc
MD
665 */
666int
29802dbb 667exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 668{
984263bc
MD
669 struct vmspace *vmspace = imgp->proc->p_vmspace;
670 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 671 struct proc *p;
29802dbb 672 vm_map_t map;
dc52e1cc 673 int error;
984263bc 674
dc52e1cc
MD
675 /*
676 * Indicate that we cannot gracefully error out any more, kill
677 * any other threads present, and set P_INEXEC to indicate that
678 * we are now messing with the process structure proper.
679 *
680 * If killalllwps() races return an error which coupled with
681 * vmspace_destroyed will cause us to exit. This is what we
682 * want since another thread is patiently waiting for us to exit
683 * in that case.
684 */
685 p = curproc;
984263bc
MD
686 imgp->vmspace_destroyed = 1;
687
dc52e1cc
MD
688 if (curthread->td_proc->p_nthreads > 1) {
689 error = killalllwps(1);
690 if (error)
691 return (error);
692 }
693 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
694 p->p_flag |= P_INEXEC;
08f2f1bb 695
984263bc
MD
696 /*
697 * Prevent a pending AIO from modifying the new address space.
698 */
699 aio_proc_rundown(imgp->proc);
700
701 /*
702 * Blow away entire process VM, if address space not shared,
703 * otherwise, create a new VM space so that other threads are
29802dbb
MD
704 * not disrupted. If we are execing a resident vmspace we
705 * create a duplicate of it and remap the stack.
239b4df9
MD
706 *
707 * The exitingcnt test is not strictly necessary but has been
708 * included for code sanity (to make the code more deterministic).
984263bc 709 */
29802dbb
MD
710 map = &vmspace->vm_map;
711 if (vmcopy) {
712 vmspace_exec(imgp->proc, vmcopy);
713 vmspace = imgp->proc->p_vmspace;
714 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
715 map = &vmspace->vm_map;
e3161323
MD
716 } else if (vmspace->vm_sysref.refcnt == 1 &&
717 vmspace->vm_exitingcnt == 0) {
fef0fdf2 718 shmexit(vmspace);
239b4df9 719 if (vmspace->vm_upcalls)
08f2f1bb 720 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08
MD
721 pmap_remove_pages(vmspace_pmap(vmspace),
722 0, VM_MAX_USER_ADDRESS);
723 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 724 } else {
29802dbb 725 vmspace_exec(imgp->proc, NULL);
984263bc
MD
726 vmspace = imgp->proc->p_vmspace;
727 map = &vmspace->vm_map;
728 }
729
730 /* Allocate a new stack */
731 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
c809941b 732 0, VM_PROT_ALL, VM_PROT_ALL, 0);
984263bc
MD
733 if (error)
734 return (error);
735
736 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
737 * VM_STACK case, but they are still used to monitor the size of the
738 * process stack so we can check the stack rlimit.
739 */
740 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
741 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
742
743 return(0);
744}
745
746/*
747 * Copy out argument and environment strings from the old process
748 * address space into the temporary string buffer.
749 */
750int
2bd9d75c 751exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 752 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 753{
984263bc 754 char *argp, *envp;
2bd9d75c 755 int error = 0;
984263bc
MD
756 size_t length;
757
2abfe22f 758 bzero(args, sizeof(*args));
18b3f457
MD
759
760 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
761 if (args->buf == NULL)
762 return (ENOMEM);
763 args->begin_argv = args->buf;
764 args->endp = args->begin_argv;
765 args->space = ARG_MAX;
2bd9d75c
DRJ
766
767 args->fname = args->buf + ARG_MAX;
768
984263bc 769 /*
2bd9d75c 770 * Copy the file name.
984263bc 771 */
2bd9d75c
DRJ
772 if (segflg == PATH_SYSSPACE) {
773 error = copystr(fname, args->fname, PATH_MAX, &length);
774 } else if (segflg == PATH_USERSPACE) {
775 error = copyinstr(fname, args->fname, PATH_MAX, &length);
776 }
984263bc 777
2bd9d75c 778 /*
66be6566
MD
779 * Extract argument strings. argv may not be NULL. The argv
780 * array is terminated by a NULL entry. We special-case the
781 * situation where argv[0] is NULL by passing { filename, NULL }
782 * to the new program to guarentee that the interpreter knows what
783 * file to open in case we exec an interpreted file. Note that
784 * a NULL argv[0] terminates the argv[] array.
785 *
786 * XXX the special-casing of argv[0] is historical and needs to be
787 * revisited.
2bd9d75c 788 */
66be6566
MD
789 if (argv == NULL)
790 error = EFAULT;
791 if (error == 0) {
792 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
793 if (argp == (caddr_t)-1) {
2bd9d75c 794 error = EFAULT;
66be6566 795 break;
2bd9d75c
DRJ
796 }
797 error = copyinstr(argp, args->endp,
2abfe22f 798 args->space, &length);
66be6566
MD
799 if (error) {
800 if (error == ENAMETOOLONG)
801 error = E2BIG;
802 break;
803 }
2bd9d75c
DRJ
804 args->space -= length;
805 args->endp += length;
806 args->argc++;
984263bc 807 }
66be6566
MD
808 if (args->argc == 0 && error == 0) {
809 length = strlen(args->fname) + 1;
810 if (length > args->space) {
811 error = E2BIG;
812 } else {
813 bcopy(args->fname, args->endp, length);
814 args->space -= length;
815 args->endp += length;
816 args->argc++;
817 }
818 }
984263bc
MD
819 }
820
2bd9d75c 821 args->begin_envv = args->endp;
984263bc
MD
822
823 /*
66be6566 824 * extract environment strings. envv may be NULL.
984263bc 825 */
2abfe22f 826 if (envv && error == 0) {
984263bc 827 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
828 if (envp == (caddr_t) -1) {
829 error = EFAULT;
66be6566 830 break;
984263bc 831 }
2bd9d75c
DRJ
832 error = copyinstr(envp, args->endp, args->space,
833 &length);
66be6566
MD
834 if (error) {
835 if (error == ENAMETOOLONG)
836 error = E2BIG;
837 break;
838 }
2bd9d75c
DRJ
839 args->space -= length;
840 args->endp += length;
841 args->envc++;
984263bc
MD
842 }
843 }
2bd9d75c
DRJ
844 return (error);
845}
846
847void
848exec_free_args(struct image_args *args)
849{
2abfe22f 850 if (args->buf) {
18b3f457 851 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
852 args->buf = NULL;
853 }
984263bc
MD
854}
855
856/*
857 * Copy strings out to the new process address space, constructing
858 * new arg and env vector tables. Return a pointer to the base
859 * so that it can be used as the initial stack pointer.
860 */
861register_t *
2bd9d75c 862exec_copyout_strings(struct image_params *imgp)
984263bc 863{
17a7ca1f 864 int argc, envc, sgap;
984263bc
MD
865 char **vectp;
866 char *stringp, *destp;
867 register_t *stack_base;
868 struct ps_strings *arginfo;
869 int szsigcode;
870
871 /*
872 * Calculate string base and vector table pointers.
873 * Also deal with signal trampoline code for this exec type.
874 */
875 arginfo = (struct ps_strings *)PS_STRINGS;
876 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
17a7ca1f 877 if (stackgap_random != 0)
0ced1954 878 sgap = ALIGN(karc4random() & (stackgap_random - 1));
17a7ca1f
MD
879 else
880 sgap = 0;
881 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
2bd9d75c 882 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
984263bc
MD
883
884 /*
885 * install sigcode
886 */
887 if (szsigcode)
888 copyout(imgp->proc->p_sysent->sv_sigcode,
2bd9d75c 889 ((caddr_t)arginfo - szsigcode), szsigcode);
984263bc
MD
890
891 /*
892 * If we have a valid auxargs ptr, prepare some room
893 * on the stack.
7ec1918b 894 *
984263bc
MD
895 * The '+ 2' is for the null pointers at the end of each of the
896 * arg and env vector sets, and 'AT_COUNT*2' is room for the
897 * ELF Auxargs data.
898 */
7ec1918b
MD
899 if (imgp->auxargs) {
900 vectp = (char **)(destp - (imgp->args->argc +
901 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
902 } else {
2bd9d75c 903 vectp = (char **)(destp - (imgp->args->argc +
7ec1918b
MD
904 imgp->args->envc + 2) * sizeof(char*));
905 }
906
984263bc 907 /*
31bc42c8
MD
908 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
909 * be done in crt1.o. Note that GCC 3.x aligns the stack in main.
984263bc 910 */
984263bc
MD
911
912 /*
913 * vectp also becomes our initial stack base
914 */
915 stack_base = (register_t *)vectp;
916
2bd9d75c
DRJ
917 stringp = imgp->args->begin_argv;
918 argc = imgp->args->argc;
919 envc = imgp->args->envc;
984263bc
MD
920
921 /*
922 * Copy out strings - arguments and environment.
923 */
2bd9d75c 924 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
925
926 /*
927 * Fill in "ps_strings" struct for ps, w, etc.
928 */
929 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
930 suword(&arginfo->ps_nargvstr, argc);
931
932 /*
933 * Fill in argument portion of vector table.
934 */
935 for (; argc > 0; --argc) {
936 suword(vectp++, (long)(intptr_t)destp);
937 while (*stringp++ != 0)
938 destp++;
939 destp++;
940 }
941
7a44d1cb 942 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
943 suword(vectp++, 0);
944
945 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
946 suword(&arginfo->ps_nenvstr, envc);
947
948 /*
949 * Fill in environment portion of vector table.
950 */
951 for (; envc > 0; --envc) {
952 suword(vectp++, (long)(intptr_t)destp);
953 while (*stringp++ != 0)
954 destp++;
955 destp++;
956 }
957
958 /* end of vector table is a null pointer */
959 suword(vectp, 0);
960
961 return (stack_base);
962}
963
964/*
965 * Check permissions of file to execute.
966 * Return 0 for success or error code on failure.
967 */
968int
77153250 969exec_check_permissions(struct image_params *imgp)
984263bc
MD
970{
971 struct proc *p = imgp->proc;
972 struct vnode *vp = imgp->vp;
973 struct vattr *attr = imgp->attr;
974 int error;
975
976 /* Get file attributes */
87de5057 977 error = VOP_GETATTR(vp, attr);
984263bc
MD
978 if (error)
979 return (error);
980
981 /*
982 * 1) Check if file execution is disabled for the filesystem that this
983 * file resides on.
984 * 2) Insure that at least one execute bit is on - otherwise root
985 * will always succeed, and we don't want to happen unless the
986 * file really is executable.
987 * 3) Insure that the file is a regular file.
988 */
989 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
990 ((attr->va_mode & 0111) == 0) ||
991 (attr->va_type != VREG)) {
992 return (EACCES);
993 }
994
995 /*
996 * Zero length files can't be exec'd
997 */
998 if (attr->va_size == 0)
999 return (ENOEXEC);
1000
1001 /*
1002 * Check for execute permission to file based on current credentials.
1003 */
87de5057 1004 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
1005 if (error)
1006 return (error);
1007
1008 /*
1009 * Check number of open-for-writes on the file and deny execution
1010 * if there are any.
1011 */
1012 if (vp->v_writecount)
1013 return (ETXTBSY);
1014
1015 /*
7540ab49
MD
1016 * Call filesystem specific open routine, which allows us to read,
1017 * write, and mmap the file. Without the VOP_OPEN we can only
1018 * stat the file.
984263bc 1019 */
87de5057 1020 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
1021 if (error)
1022 return (error);
1023
1024 return (0);
1025}
1026
1027/*
1028 * Exec handler registration
1029 */
1030int
77153250 1031exec_register(const struct execsw *execsw_arg)
984263bc
MD
1032{
1033 const struct execsw **es, **xs, **newexecsw;
1034 int count = 2; /* New slot and trailing NULL */
1035
1036 if (execsw)
1037 for (es = execsw; *es; es++)
1038 count++;
efda3bd0 1039 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1040 xs = newexecsw;
1041 if (execsw)
1042 for (es = execsw; *es; es++)
1043 *xs++ = *es;
1044 *xs++ = execsw_arg;
1045 *xs = NULL;
1046 if (execsw)
efda3bd0 1047 kfree(execsw, M_TEMP);
984263bc
MD
1048 execsw = newexecsw;
1049 return 0;
1050}
1051
1052int
77153250 1053exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1054{
1055 const struct execsw **es, **xs, **newexecsw;
1056 int count = 1;
1057
1058 if (execsw == NULL)
ae7e74d4 1059 panic("unregister with no handlers left?");
984263bc
MD
1060
1061 for (es = execsw; *es; es++) {
1062 if (*es == execsw_arg)
1063 break;
1064 }
1065 if (*es == NULL)
1066 return ENOENT;
1067 for (es = execsw; *es; es++)
1068 if (*es != execsw_arg)
1069 count++;
efda3bd0 1070 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1071 xs = newexecsw;
1072 for (es = execsw; *es; es++)
1073 if (*es != execsw_arg)
1074 *xs++ = *es;
1075 *xs = NULL;
1076 if (execsw)
efda3bd0 1077 kfree(execsw, M_TEMP);
984263bc
MD
1078 execsw = newexecsw;
1079 return 0;
1080}