kernel - MPSAFE work - Finish tokenizing vm_page.c
[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 /*
240 * Check file permissions (also 'opens' file)
241 */
242 error = exec_check_permissions(imgp);
243 if (error) {
a11aaa81 244 vn_unlock(imgp->vp);
984263bc
MD
245 goto exec_fail_dealloc;
246 }
247
248 error = exec_map_first_page(imgp);
a11aaa81 249 vn_unlock(imgp->vp);
984263bc
MD
250 if (error)
251 goto exec_fail_dealloc;
252
2bd9d75c 253 if (debug_execve_args && imgp->interpreted) {
6ea70f76
SW
254 kprintf(" target is interpreted -- recursive pass\n");
255 kprintf(" interpreter: %s\n", imgp->interpreter_name);
2bd9d75c
DRJ
256 print_execve_args(args);
257 }
258
984263bc
MD
259 /*
260 * If the current process has a special image activator it
261 * wants to try first, call it. For example, emulating shell
262 * scripts differently.
263 */
264 error = -1;
265 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
266 error = img_first(imgp);
267
268 /*
29802dbb
MD
269 * If the vnode has a registered vmspace, exec the vmspace
270 */
271 if (error == -1 && imgp->vp->v_resident) {
272 error = exec_resident_imgact(imgp);
273 }
274
275 /*
984263bc
MD
276 * Loop through the list of image activators, calling each one.
277 * An activator returns -1 if there is no match, 0 on success,
278 * and an error otherwise.
279 */
280 for (i = 0; error == -1 && execsw[i]; ++i) {
281 if (execsw[i]->ex_imgact == NULL ||
282 execsw[i]->ex_imgact == img_first) {
283 continue;
284 }
285 error = (*execsw[i]->ex_imgact)(imgp);
286 }
287
288 if (error) {
289 if (error == -1)
290 error = ENOEXEC;
291 goto exec_fail_dealloc;
292 }
293
294 /*
295 * Special interpreter operation, cleanup and loop up to try to
296 * activate the interpreter.
297 */
298 if (imgp->interpreted) {
299 exec_unmap_first_page(imgp);
fad57d0e
MD
300 nlookup_done(nd);
301 vrele(imgp->vp);
302 imgp->vp = NULL;
303 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
304 NLC_FOLLOW);
305 if (error)
306 goto exec_fail;
984263bc
MD
307 goto interpret;
308 }
309
310 /*
311 * Copy out strings (args and env) and initialize stack base
312 */
313 stack_base = exec_copyout_strings(imgp);
314 p->p_vmspace->vm_minsaddr = (char *)stack_base;
315
316 /*
317 * If custom stack fixup routine present for this process
29802dbb
MD
318 * let it do the stack setup. If we are running a resident
319 * image there is no auxinfo or other image activator context
320 * so don't try to add fixups to the stack.
321 *
984263bc
MD
322 * Else stuff argument count as first item on stack
323 */
29802dbb 324 if (p->p_sysent->sv_fixup && imgp->resident == 0)
984263bc
MD
325 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
326 else
2bd9d75c 327 suword(--stack_base, imgp->args->argc);
984263bc
MD
328
329 /*
330 * For security and other reasons, the file descriptor table cannot
331 * be shared after an exec.
332 */
333 if (p->p_fd->fd_refcnt > 1) {
334 struct filedesc *tmp;
335
336 tmp = fdcopy(p);
0a4a9c77 337 fdfree(p, tmp);
984263bc
MD
338 }
339
340 /*
341 * For security and other reasons, signal handlers cannot
342 * be shared after an exec. The new proces gets a copy of the old
343 * handlers. In execsigs(), the new process will have its signals
344 * reset.
345 */
b1b4e5a6
SS
346 if (p->p_sigacts->ps_refcnt > 1) {
347 struct sigacts *newsigacts;
984263bc 348
b1b4e5a6 349 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts),
984263bc 350 M_SUBPROC, M_WAITOK);
b1b4e5a6
SS
351 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts));
352 p->p_sigacts->ps_refcnt--;
353 p->p_sigacts = newsigacts;
354 p->p_sigacts->ps_refcnt = 1;
984263bc
MD
355 }
356
0daa37a5
MD
357 /*
358 * For security and other reasons virtual kernels cannot be
359 * inherited by an exec. This also allows a virtual kernel
360 * to fork/exec unrelated applications.
361 */
4a22e893
MD
362 if (p->p_vkernel)
363 vkernel_exit(p);
0daa37a5 364
984263bc
MD
365 /* Stop profiling */
366 stopprofclock(p);
367
368 /* close files on exec */
369 fdcloseexec(p);
370
371 /* reset caught signals */
372 execsigs(p);
373
374 /* name this process - nameiexec(p, ndp) */
28623bf9
MD
375 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
376 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
984263bc 377 p->p_comm[len] = 0;
08f2f1bb 378 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
984263bc
MD
379
380 /*
381 * mark as execed, wakeup the process that vforked (if any) and tell
382 * it that it now has its own resources back
383 */
384 p->p_flag |= P_EXEC;
385 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
386 p->p_flag &= ~P_PPWAIT;
387 wakeup((caddr_t)p->p_pptr);
388 }
389
390 /*
391 * Implement image setuid/setgid.
392 *
393 * Don't honor setuid/setgid if the filesystem prohibits it or if
394 * the process is being traced.
395 */
396 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
397 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
398 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
399 (p->p_flag & P_TRACED) == 0) {
400 /*
401 * Turn off syscall tracing for set-id programs, except for
402 * root. Record any set-id flags first to make sure that
403 * we do not regain any tracing during a possible block.
404 */
41c20dac 405 setsugid();
5c627295
MD
406 if (p->p_tracenode && ktrace_suid == 0 &&
407 priv_check(td, PRIV_ROOT) != 0) {
29f58392
MD
408 ktrdestroy(&p->p_tracenode);
409 p->p_traceflag = 0;
984263bc
MD
410 }
411 /* Close any file descriptors 0..2 that reference procfs */
412 setugidsafety(p);
413 /* Make sure file descriptors 0..2 are in use. */
f3a2d8c4 414 error = fdcheckstd(lp);
984263bc
MD
415 if (error != 0)
416 goto exec_fail_dealloc;
417 /*
418 * Set the new credentials.
419 */
e9a372eb 420 cratom(&p->p_ucred);
984263bc 421 if (attr.va_mode & VSUID)
41c20dac 422 change_euid(attr.va_uid);
984263bc
MD
423 if (attr.va_mode & VSGID)
424 p->p_ucred->cr_gid = attr.va_gid;
98a7f915
MD
425
426 /*
427 * Clear local varsym variables
428 */
429 varsymset_clean(&p->p_varsymset);
984263bc 430 } else {
41c20dac
MD
431 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
432 p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
984263bc
MD
433 p->p_flag &= ~P_SUGID;
434 }
435
436 /*
437 * Implement correct POSIX saved-id behavior.
438 */
115ccd83
MD
439 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
440 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
441 cratom(&p->p_ucred);
442 p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
443 p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
444 }
984263bc
MD
445
446 /*
447 * Store the vp for use in procfs
448 */
449 if (p->p_textvp) /* release old reference */
450 vrele(p->p_textvp);
fad57d0e
MD
451 p->p_textvp = imgp->vp;
452 vref(p->p_textvp);
984263bc 453
8ba5f7ef
AH
454 /* Release old namecache handle to text file */
455 if (p->p_textnch.ncp)
456 cache_drop(&p->p_textnch);
457
458 if (nd->nl_nch.mount)
459 cache_copy(&nd->nl_nch, &p->p_textnch);
460
984263bc
MD
461 /*
462 * Notify others that we exec'd, and clear the P_INEXEC flag
463 * as we're now a bona fide freshly-execed process.
464 */
465 KNOTE(&p->p_klist, NOTE_EXEC);
466 p->p_flag &= ~P_INEXEC;
467
468 /*
469 * If tracing the process, trap to debugger so breakpoints
470 * can be set before the program executes.
471 */
472 STOPEVENT(p, S_EXEC, 0);
473
474 if (p->p_flag & P_TRACED)
84204577 475 ksignal(p, SIGTRAP);
984263bc
MD
476
477 /* clear "fork but no exec" flag, as we _are_ execing */
478 p->p_acflag &= ~AFORK;
479
480 /* Set values passed into the program in registers. */
08f2f1bb 481 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
984263bc
MD
482 imgp->ps_strings);
483
349433c9
MD
484 /* Set the access time on the vnode */
485 vn_mark_atime(imgp->vp, td);
486
984263bc
MD
487 /* Free any previous argument cache */
488 if (p->p_args && --p->p_args->ar_ref == 0)
489 FREE(p->p_args, M_PARGS);
490 p->p_args = NULL;
491
492 /* Cache arguments if they fit inside our allowance */
2bd9d75c 493 i = imgp->args->begin_envv - imgp->args->begin_argv;
984263bc
MD
494 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
495 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
496 M_PARGS, M_WAITOK);
497 p->p_args->ar_ref = 1;
498 p->p_args->ar_length = i;
2bd9d75c 499 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
984263bc
MD
500 }
501
502exec_fail_dealloc:
503
504 /*
505 * free various allocated resources
506 */
507 if (imgp->firstpage)
508 exec_unmap_first_page(imgp);
509
984263bc 510 if (imgp->vp) {
984263bc 511 vrele(imgp->vp);
fad57d0e 512 imgp->vp = NULL;
984263bc
MD
513 }
514
a6f89c72
DR
515 if (error == 0) {
516 ++mycpu->gd_cnt.v_exec;
984263bc 517 return (0);
a6f89c72 518 }
984263bc
MD
519
520exec_fail:
dc52e1cc
MD
521 /*
522 * we're done here, clear P_INEXEC if we were the ones that
523 * set it. Otherwise if vmspace_destroyed is still set we
524 * raced another thread and that thread is responsible for
525 * clearing it.
526 */
527 if (imgp->vmspace_destroyed & 2)
528 p->p_flag &= ~P_INEXEC;
984263bc 529 if (imgp->vmspace_destroyed) {
5417cd57
SS
530 /*
531 * Sorry, no more process anymore. exit gracefully.
532 * However we can't die right here, because our
533 * caller might have to clean up, so indicate a
90947cb1 534 * lethal error by returning -1.
5417cd57
SS
535 */
536 return(-1);
984263bc
MD
537 } else {
538 return(error);
539 }
540}
541
2bd9d75c
DRJ
542/*
543 * execve() system call.
3919ced0
MD
544 *
545 * MPALMOSTSAFE
2bd9d75c
DRJ
546 */
547int
753fd850 548sys_execve(struct execve_args *uap)
2bd9d75c 549{
fad57d0e 550 struct nlookupdata nd;
2bd9d75c
DRJ
551 struct image_args args;
552 int error;
553
118cec38 554 bzero(&args, sizeof(args));
3919ced0
MD
555
556 get_mplock();
557 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
fad57d0e
MD
558 if (error == 0) {
559 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
560 uap->argv, uap->envv);
561 }
2abfe22f
MD
562 if (error == 0)
563 error = kern_execve(&nd, &args);
fad57d0e 564 nlookup_done(&nd);
2bd9d75c
DRJ
565 exec_free_args(&args);
566
5417cd57 567 if (error < 0) {
90947cb1 568 /* We hit a lethal error condition. Let's die now. */
5417cd57
SS
569 exit1(W_EXITCODE(0, SIGABRT));
570 /* NOTREACHED */
571 }
3919ced0 572 rel_mplock();
5417cd57 573
2bd9d75c
DRJ
574 /*
575 * The syscall result is returned in registers to the new program.
576 * Linux will register %edx as an atexit function and we must be
577 * sure to set it to 0. XXX
578 */
579 if (error == 0)
580 uap->sysmsg_result64 = 0;
581
582 return (error);
583}
584
984263bc 585int
7c34d798 586exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
5c5185ae 587 struct lwbuf **plwb, const char **pdata)
984263bc 588{
1b9d3514
MD
589 int rv;
590 vm_page_t ma;
06ecca5a 591 vm_page_t m;
984263bc
MD
592 vm_object_t object;
593
fad57d0e 594 /*
7540ab49 595 * The file has to be mappable.
fad57d0e 596 */
7540ab49 597 if ((object = imgp->vp->v_object) == NULL)
fad57d0e 598 return (EIO);
984263bc 599
7c34d798
SS
600 if (pageno >= object->size)
601 return (EIO);
602
06ecca5a
MD
603 /*
604 * We shouldn't need protection for vm_page_grab() but we certainly
605 * need it for the lookup loop below (lookup/busy race), since
606 * an interrupt can unbusy and free the page before our busy check.
607 */
7c34d798 608 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
1b9d3514 609 crit_enter();
573fb415 610 lwkt_gettoken(&vm_token);
1b9d3514
MD
611 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
612 ma = m;
984263bc 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 */
1b9d3514 621 rv = vm_pager_get_page(object, &ma, 1);
7c34d798 622 m = vm_page_lookup(object, pageno);
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 }
573fb415 629 lwkt_reltoken(&vm_token);
5fd012e0 630 crit_exit();
984263bc
MD
631 return EIO;
632 }
633 }
573fb415 634 vm_page_hold(m); /* requires vm_token to be held */
06ecca5a 635 vm_page_wakeup(m); /* unbusy the page */
573fb415 636 lwkt_reltoken(&vm_token);
5fd012e0 637 crit_exit();
984263bc 638
5c5185ae
SG
639 *plwb = lwbuf_alloc(m);
640 *pdata = (void *)lwbuf_kva(*plwb);
7c34d798
SS
641
642 return (0);
643}
644
645int
646exec_map_first_page(struct image_params *imgp)
647{
648 int err;
649
650 if (imgp->firstpage)
651 exec_unmap_first_page(imgp);
652
653 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
654
655 if (err)
656 return err;
984263bc
MD
657
658 return 0;
659}
660
661void
5c5185ae 662exec_unmap_page(struct lwbuf *lwb)
984263bc 663{
18f40545
MD
664 vm_page_t m;
665
5fd012e0 666 crit_enter();
5c5185ae
SG
667 if (lwb != NULL) {
668 m = lwbuf_page(lwb);
669 lwbuf_free(lwb);
e4ba7818 670 vm_page_unhold(m);
984263bc 671 }
5fd012e0 672 crit_exit();
984263bc
MD
673}
674
7c34d798
SS
675void
676exec_unmap_first_page(struct image_params *imgp)
677{
678 exec_unmap_page(imgp->firstpage);
679 imgp->firstpage = NULL;
680 imgp->image_header = NULL;
681}
682
984263bc
MD
683/*
684 * Destroy old address space, and allocate a new stack
685 * The new stack is only SGROWSIZ large because it is grown
686 * automatically in trap.c.
dc52e1cc
MD
687 *
688 * This is the point of no return.
984263bc
MD
689 */
690int
29802dbb 691exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
984263bc 692{
984263bc
MD
693 struct vmspace *vmspace = imgp->proc->p_vmspace;
694 vm_offset_t stack_addr = USRSTACK - maxssiz;
dc52e1cc 695 struct proc *p;
29802dbb 696 vm_map_t map;
dc52e1cc 697 int error;
984263bc 698
dc52e1cc
MD
699 /*
700 * Indicate that we cannot gracefully error out any more, kill
701 * any other threads present, and set P_INEXEC to indicate that
702 * we are now messing with the process structure proper.
703 *
704 * If killalllwps() races return an error which coupled with
705 * vmspace_destroyed will cause us to exit. This is what we
706 * want since another thread is patiently waiting for us to exit
707 * in that case.
708 */
709 p = curproc;
984263bc
MD
710 imgp->vmspace_destroyed = 1;
711
dc52e1cc
MD
712 if (curthread->td_proc->p_nthreads > 1) {
713 error = killalllwps(1);
714 if (error)
715 return (error);
716 }
717 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */
718 p->p_flag |= P_INEXEC;
08f2f1bb
SS
719
720 /*
984263bc
MD
721 * Prevent a pending AIO from modifying the new address space.
722 */
723 aio_proc_rundown(imgp->proc);
724
725 /*
726 * Blow away entire process VM, if address space not shared,
727 * otherwise, create a new VM space so that other threads are
29802dbb
MD
728 * not disrupted. If we are execing a resident vmspace we
729 * create a duplicate of it and remap the stack.
239b4df9
MD
730 *
731 * The exitingcnt test is not strictly necessary but has been
732 * included for code sanity (to make the code more deterministic).
984263bc 733 */
29802dbb
MD
734 map = &vmspace->vm_map;
735 if (vmcopy) {
736 vmspace_exec(imgp->proc, vmcopy);
737 vmspace = imgp->proc->p_vmspace;
738 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
739 map = &vmspace->vm_map;
e3161323
MD
740 } else if (vmspace->vm_sysref.refcnt == 1 &&
741 vmspace->vm_exitingcnt == 0) {
fef0fdf2 742 shmexit(vmspace);
239b4df9 743 if (vmspace->vm_upcalls)
08f2f1bb 744 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
88181b08
MD
745 pmap_remove_pages(vmspace_pmap(vmspace),
746 0, VM_MAX_USER_ADDRESS);
747 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
984263bc 748 } else {
29802dbb 749 vmspace_exec(imgp->proc, NULL);
984263bc
MD
750 vmspace = imgp->proc->p_vmspace;
751 map = &vmspace->vm_map;
752 }
753
754 /* Allocate a new stack */
755 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
c809941b 756 0, VM_PROT_ALL, VM_PROT_ALL, 0);
984263bc
MD
757 if (error)
758 return (error);
759
760 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
761 * VM_STACK case, but they are still used to monitor the size of the
762 * process stack so we can check the stack rlimit.
763 */
764 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
765 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
766
767 return(0);
768}
769
770/*
771 * Copy out argument and environment strings from the old process
772 * address space into the temporary string buffer.
773 */
774int
2bd9d75c 775exec_copyin_args(struct image_args *args, char *fname,
2abfe22f 776 enum exec_path_segflg segflg, char **argv, char **envv)
984263bc 777{
984263bc 778 char *argp, *envp;
2bd9d75c 779 int error = 0;
984263bc
MD
780 size_t length;
781
18b3f457 782 args->buf = objcache_get(exec_objcache, M_WAITOK);
2bd9d75c
DRJ
783 if (args->buf == NULL)
784 return (ENOMEM);
785 args->begin_argv = args->buf;
786 args->endp = args->begin_argv;
787 args->space = ARG_MAX;
2bd9d75c
DRJ
788
789 args->fname = args->buf + ARG_MAX;
790
984263bc 791 /*
2bd9d75c 792 * Copy the file name.
984263bc 793 */
2bd9d75c
DRJ
794 if (segflg == PATH_SYSSPACE) {
795 error = copystr(fname, args->fname, PATH_MAX, &length);
796 } else if (segflg == PATH_USERSPACE) {
797 error = copyinstr(fname, args->fname, PATH_MAX, &length);
798 }
984263bc 799
2bd9d75c 800 /*
66be6566
MD
801 * Extract argument strings. argv may not be NULL. The argv
802 * array is terminated by a NULL entry. We special-case the
803 * situation where argv[0] is NULL by passing { filename, NULL }
804 * to the new program to guarentee that the interpreter knows what
805 * file to open in case we exec an interpreted file. Note that
806 * a NULL argv[0] terminates the argv[] array.
807 *
808 * XXX the special-casing of argv[0] is historical and needs to be
809 * revisited.
2bd9d75c 810 */
66be6566
MD
811 if (argv == NULL)
812 error = EFAULT;
813 if (error == 0) {
814 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
815 if (argp == (caddr_t)-1) {
2bd9d75c 816 error = EFAULT;
66be6566 817 break;
2bd9d75c
DRJ
818 }
819 error = copyinstr(argp, args->endp,
2abfe22f 820 args->space, &length);
66be6566
MD
821 if (error) {
822 if (error == ENAMETOOLONG)
823 error = E2BIG;
824 break;
825 }
2bd9d75c
DRJ
826 args->space -= length;
827 args->endp += length;
828 args->argc++;
984263bc 829 }
66be6566
MD
830 if (args->argc == 0 && error == 0) {
831 length = strlen(args->fname) + 1;
832 if (length > args->space) {
833 error = E2BIG;
834 } else {
835 bcopy(args->fname, args->endp, length);
836 args->space -= length;
837 args->endp += length;
838 args->argc++;
839 }
840 }
984263bc
MD
841 }
842
2bd9d75c 843 args->begin_envv = args->endp;
984263bc
MD
844
845 /*
66be6566 846 * extract environment strings. envv may be NULL.
984263bc 847 */
2abfe22f 848 if (envv && error == 0) {
984263bc 849 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
2bd9d75c
DRJ
850 if (envp == (caddr_t) -1) {
851 error = EFAULT;
66be6566 852 break;
984263bc 853 }
2bd9d75c
DRJ
854 error = copyinstr(envp, args->endp, args->space,
855 &length);
66be6566
MD
856 if (error) {
857 if (error == ENAMETOOLONG)
858 error = E2BIG;
859 break;
860 }
2bd9d75c
DRJ
861 args->space -= length;
862 args->endp += length;
863 args->envc++;
984263bc
MD
864 }
865 }
2bd9d75c
DRJ
866 return (error);
867}
868
869void
870exec_free_args(struct image_args *args)
871{
2abfe22f 872 if (args->buf) {
18b3f457 873 objcache_put(exec_objcache, args->buf);
2abfe22f
MD
874 args->buf = NULL;
875 }
984263bc
MD
876}
877
878/*
879 * Copy strings out to the new process address space, constructing
880 * new arg and env vector tables. Return a pointer to the base
881 * so that it can be used as the initial stack pointer.
882 */
883register_t *
2bd9d75c 884exec_copyout_strings(struct image_params *imgp)
984263bc 885{
17a7ca1f 886 int argc, envc, sgap;
984263bc
MD
887 char **vectp;
888 char *stringp, *destp;
889 register_t *stack_base;
890 struct ps_strings *arginfo;
891 int szsigcode;
892
893 /*
894 * Calculate string base and vector table pointers.
895 * Also deal with signal trampoline code for this exec type.
896 */
897 arginfo = (struct ps_strings *)PS_STRINGS;
898 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
17a7ca1f 899 if (stackgap_random != 0)
0ced1954 900 sgap = ALIGN(karc4random() & (stackgap_random - 1));
17a7ca1f
MD
901 else
902 sgap = 0;
903 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
2bd9d75c 904 roundup((ARG_MAX - imgp->args->space), sizeof(char *));
984263bc
MD
905
906 /*
907 * install sigcode
908 */
909 if (szsigcode)
910 copyout(imgp->proc->p_sysent->sv_sigcode,
2bd9d75c 911 ((caddr_t)arginfo - szsigcode), szsigcode);
984263bc
MD
912
913 /*
914 * If we have a valid auxargs ptr, prepare some room
915 * on the stack.
7ec1918b 916 *
984263bc
MD
917 * The '+ 2' is for the null pointers at the end of each of the
918 * arg and env vector sets, and 'AT_COUNT*2' is room for the
919 * ELF Auxargs data.
920 */
7ec1918b
MD
921 if (imgp->auxargs) {
922 vectp = (char **)(destp - (imgp->args->argc +
923 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
924 } else {
2bd9d75c 925 vectp = (char **)(destp - (imgp->args->argc +
7ec1918b
MD
926 imgp->args->envc + 2) * sizeof(char*));
927 }
928
984263bc 929 /*
31bc42c8
MD
930 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
931 * be done in crt1.o. Note that GCC 3.x aligns the stack in main.
984263bc 932 */
984263bc
MD
933
934 /*
935 * vectp also becomes our initial stack base
936 */
937 stack_base = (register_t *)vectp;
938
2bd9d75c
DRJ
939 stringp = imgp->args->begin_argv;
940 argc = imgp->args->argc;
941 envc = imgp->args->envc;
984263bc
MD
942
943 /*
944 * Copy out strings - arguments and environment.
945 */
2bd9d75c 946 copyout(stringp, destp, ARG_MAX - imgp->args->space);
984263bc
MD
947
948 /*
949 * Fill in "ps_strings" struct for ps, w, etc.
950 */
951 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
952 suword(&arginfo->ps_nargvstr, argc);
953
954 /*
955 * Fill in argument portion of vector table.
956 */
957 for (; argc > 0; --argc) {
958 suword(vectp++, (long)(intptr_t)destp);
959 while (*stringp++ != 0)
960 destp++;
961 destp++;
962 }
963
7a44d1cb 964 /* a null vector table pointer separates the argp's from the envp's */
984263bc
MD
965 suword(vectp++, 0);
966
967 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
968 suword(&arginfo->ps_nenvstr, envc);
969
970 /*
971 * Fill in environment portion of vector table.
972 */
973 for (; envc > 0; --envc) {
974 suword(vectp++, (long)(intptr_t)destp);
975 while (*stringp++ != 0)
976 destp++;
977 destp++;
978 }
979
980 /* end of vector table is a null pointer */
981 suword(vectp, 0);
982
983 return (stack_base);
984}
985
986/*
987 * Check permissions of file to execute.
988 * Return 0 for success or error code on failure.
989 */
990int
77153250 991exec_check_permissions(struct image_params *imgp)
984263bc
MD
992{
993 struct proc *p = imgp->proc;
994 struct vnode *vp = imgp->vp;
995 struct vattr *attr = imgp->attr;
996 int error;
997
998 /* Get file attributes */
87de5057 999 error = VOP_GETATTR(vp, attr);
984263bc
MD
1000 if (error)
1001 return (error);
1002
1003 /*
1004 * 1) Check if file execution is disabled for the filesystem that this
1005 * file resides on.
1006 * 2) Insure that at least one execute bit is on - otherwise root
1007 * will always succeed, and we don't want to happen unless the
1008 * file really is executable.
1009 * 3) Insure that the file is a regular file.
1010 */
1011 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1012 ((attr->va_mode & 0111) == 0) ||
1013 (attr->va_type != VREG)) {
1014 return (EACCES);
1015 }
1016
1017 /*
1018 * Zero length files can't be exec'd
1019 */
1020 if (attr->va_size == 0)
1021 return (ENOEXEC);
1022
1023 /*
1024 * Check for execute permission to file based on current credentials.
1025 */
90a83564 1026 error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
984263bc
MD
1027 if (error)
1028 return (error);
1029
1030 /*
1031 * Check number of open-for-writes on the file and deny execution
1032 * if there are any.
1033 */
1034 if (vp->v_writecount)
1035 return (ETXTBSY);
1036
1037 /*
7540ab49
MD
1038 * Call filesystem specific open routine, which allows us to read,
1039 * write, and mmap the file. Without the VOP_OPEN we can only
1040 * stat the file.
984263bc 1041 */
87de5057 1042 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
984263bc
MD
1043 if (error)
1044 return (error);
1045
1046 return (0);
1047}
1048
1049/*
1050 * Exec handler registration
1051 */
1052int
77153250 1053exec_register(const struct execsw *execsw_arg)
984263bc
MD
1054{
1055 const struct execsw **es, **xs, **newexecsw;
1056 int count = 2; /* New slot and trailing NULL */
1057
1058 if (execsw)
1059 for (es = execsw; *es; es++)
1060 count++;
efda3bd0 1061 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1062 xs = newexecsw;
1063 if (execsw)
1064 for (es = execsw; *es; es++)
1065 *xs++ = *es;
1066 *xs++ = execsw_arg;
1067 *xs = NULL;
1068 if (execsw)
efda3bd0 1069 kfree(execsw, M_TEMP);
984263bc
MD
1070 execsw = newexecsw;
1071 return 0;
1072}
1073
1074int
77153250 1075exec_unregister(const struct execsw *execsw_arg)
984263bc
MD
1076{
1077 const struct execsw **es, **xs, **newexecsw;
1078 int count = 1;
1079
1080 if (execsw == NULL)
ae7e74d4 1081 panic("unregister with no handlers left?");
984263bc
MD
1082
1083 for (es = execsw; *es; es++) {
1084 if (*es == execsw_arg)
1085 break;
1086 }
1087 if (*es == NULL)
1088 return ENOENT;
1089 for (es = execsw; *es; es++)
1090 if (*es != execsw_arg)
1091 count++;
efda3bd0 1092 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
984263bc
MD
1093 xs = newexecsw;
1094 for (es = execsw; *es; es++)
1095 if (*es != execsw_arg)
1096 *xs++ = *es;
1097 *xs = NULL;
1098 if (execsw)
efda3bd0 1099 kfree(execsw, M_TEMP);
984263bc
MD
1100 execsw = newexecsw;
1101 return 0;
1102}