xref: /netbsd-src/sys/kern/kern_exec.c (revision 7cc2f76925f078d01ddc9e640a98f4ccfc9f8c3b)
1 /*	$NetBSD: kern_exec.c,v 1.133 2000/12/11 05:29:02 mycroft Exp $	*/
2 
3 /*-
4  * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
5  * Copyright (C) 1992 Wolfgang Solfrank.
6  * Copyright (C) 1992 TooLs GmbH.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by TooLs GmbH.
20  * 4. The name of TooLs GmbH may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
28  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
29  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
30  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
31  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
32  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include "opt_ktrace.h"
36 #include "opt_syscall_debug.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/filedesc.h>
41 #include <sys/kernel.h>
42 #include <sys/proc.h>
43 #include <sys/mount.h>
44 #include <sys/malloc.h>
45 #include <sys/namei.h>
46 #include <sys/vnode.h>
47 #include <sys/file.h>
48 #include <sys/acct.h>
49 #include <sys/exec.h>
50 #include <sys/ktrace.h>
51 #include <sys/resourcevar.h>
52 #include <sys/wait.h>
53 #include <sys/mman.h>
54 #include <sys/signalvar.h>
55 #include <sys/stat.h>
56 #include <sys/syscall.h>
57 
58 #include <sys/syscallargs.h>
59 
60 #include <uvm/uvm_extern.h>
61 
62 #include <machine/cpu.h>
63 #include <machine/reg.h>
64 
65 /*
66  * Exec function switch:
67  *
68  * Note that each makecmds function is responsible for loading the
69  * exec package with the necessary functions for any exec-type-specific
70  * handling.
71  *
72  * Functions for specific exec types should be defined in their own
73  * header file.
74  */
75 extern const struct execsw execsw_builtin[];
76 extern int nexecs_builtin;
77 static const struct execsw **execsw = NULL;
78 static int nexecs;
79 int exec_maxhdrsz;		/* must not be static - netbsd32 needs it */
80 
81 #ifdef LKM
82 /* list of supported emulations */
83 static
84 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head);
85 struct emul_entry {
86 	LIST_ENTRY(emul_entry) el_list;
87 	const struct emul *el_emul;
88 	int ro_entry;
89 };
90 
91 /* list of dynamically loaded execsw entries */
92 static
93 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head);
94 struct exec_entry {
95 	LIST_ENTRY(exec_entry) ex_list;
96 	const struct execsw *es;
97 };
98 
99 /* structure used for building execw[] */
100 struct execsw_entry {
101 	struct execsw_entry *next;
102 	const struct execsw *es;
103 };
104 #endif /* LKM */
105 
106 /* NetBSD emul struct */
107 extern char sigcode[], esigcode[];
108 #ifdef SYSCALL_DEBUG
109 extern const char * const syscallnames[];
110 #endif
111 #ifdef __HAVE_SYSCALL_INTERN
112 void syscall_intern __P((struct proc *));
113 #else
114 void syscall __P((void));
115 #endif
116 
117 const struct emul emul_netbsd = {
118 	"netbsd",
119 	NULL,		/* emulation path */
120 #ifndef __HAVE_MINIMAL_EMUL
121 	EMUL_HAS_SYS___syscall,
122 	NULL,
123 	SYS_syscall,
124 	SYS_MAXSYSCALL,
125 #endif
126 	sysent,
127 #ifdef SYSCALL_DEBUG
128 	syscallnames,
129 #else
130 	NULL,
131 #endif
132 	sendsig,
133 	sigcode,
134 	esigcode,
135 	NULL,
136 	NULL,
137 	NULL,
138 #ifdef __HAVE_SYSCALL_INTERN
139 	syscall_intern,
140 #else
141 	syscall,
142 #endif
143 };
144 
145 /*
146  * Exec lock. Used to control access to execsw[] structures.
147  * This must not be static so that netbsd32 can access it, too.
148  */
149 struct lock exec_lock;
150 
151 #ifdef LKM
152 static const struct emul *	emul_search __P((const char *));
153 static void link_es __P((struct execsw_entry **, const struct execsw *));
154 #endif /* LKM */
155 
156 /*
157  * check exec:
158  * given an "executable" described in the exec package's namei info,
159  * see what we can do with it.
160  *
161  * ON ENTRY:
162  *	exec package with appropriate namei info
163  *	proc pointer of exec'ing proc
164  *	NO SELF-LOCKED VNODES
165  *
166  * ON EXIT:
167  *	error:	nothing held, etc.  exec header still allocated.
168  *	ok:	filled exec package, executable's vnode (unlocked).
169  *
170  * EXEC SWITCH ENTRY:
171  * 	Locked vnode to check, exec package, proc.
172  *
173  * EXEC SWITCH EXIT:
174  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
175  *	error:	destructive:
176  *			everything deallocated execept exec header.
177  *		non-destructive:
178  *			error code, executable's vnode (unlocked),
179  *			exec header unmodified.
180  */
181 int
182 check_exec(struct proc *p, struct exec_package *epp)
183 {
184 	int error, i;
185 	struct vnode *vp;
186 	struct nameidata *ndp;
187 	size_t resid;
188 
189 	ndp = epp->ep_ndp;
190 	ndp->ni_cnd.cn_nameiop = LOOKUP;
191 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
192 	/* first get the vnode */
193 	if ((error = namei(ndp)) != 0)
194 		return error;
195 	epp->ep_vp = vp = ndp->ni_vp;
196 
197 	/* check access and type */
198 	if (vp->v_type != VREG) {
199 		error = EACCES;
200 		goto bad1;
201 	}
202 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
203 		goto bad1;
204 
205 	/* get attributes */
206 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
207 		goto bad1;
208 
209 	/* Check mount point */
210 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
211 		error = EACCES;
212 		goto bad1;
213 	}
214 	if ((vp->v_mount->mnt_flag & MNT_NOSUID) || (p->p_flag & P_TRACED))
215 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
216 
217 	/* try to open it */
218 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
219 		goto bad1;
220 
221 	/* unlock vp, since we need it unlocked from here on out. */
222 	VOP_UNLOCK(vp, 0);
223 
224 	/* now we have the file, get the exec header */
225 	uvn_attach(vp, VM_PROT_READ);
226 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
227 			UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
228 	if (error)
229 		goto bad2;
230 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
231 
232 	/*
233 	 * set up the vmcmds for creation of the process
234 	 * address space
235 	 */
236 	error = ENOEXEC;
237 	for (i = 0; i < nexecs && error != 0; i++) {
238 		int newerror;
239 
240 		epp->ep_esch = execsw[i];
241 		newerror = (*execsw[i]->es_check)(p, epp);
242 		/* make sure the first "interesting" error code is saved. */
243 		if (!newerror || error == ENOEXEC)
244 			error = newerror;
245 
246 		/* if es_check call was successful, update epp->ep_es */
247 		if (!newerror && (epp->ep_flags & EXEC_HASES) == 0)
248 			epp->ep_es = execsw[i];
249 
250 		if (epp->ep_flags & EXEC_DESTR && error != 0)
251 			return error;
252 	}
253 	if (!error) {
254 		/* check that entry point is sane */
255 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
256 			error = ENOEXEC;
257 
258 		/* check limits */
259 		if ((epp->ep_tsize > MAXTSIZ) ||
260 		    (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur))
261 			error = ENOMEM;
262 
263 		if (!error)
264 			return (0);
265 	}
266 
267 	/*
268 	 * free any vmspace-creation commands,
269 	 * and release their references
270 	 */
271 	kill_vmcmds(&epp->ep_vmcmds);
272 
273 bad2:
274 	/*
275 	 * close and release the vnode, restore the old one, free the
276 	 * pathname buf, and punt.
277 	 */
278 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
279 	VOP_CLOSE(vp, FREAD, p->p_ucred, p);
280 	vput(vp);
281 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
282 	return error;
283 
284 bad1:
285 	/*
286 	 * free the namei pathname buffer, and put the vnode
287 	 * (which we don't yet have open).
288 	 */
289 	vput(vp);				/* was still locked */
290 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
291 	return error;
292 }
293 
294 /*
295  * exec system call
296  */
297 /* ARGSUSED */
298 int
299 sys_execve(struct proc *p, void *v, register_t *retval)
300 {
301 	struct sys_execve_args /* {
302 		syscallarg(const char *) path;
303 		syscallarg(char * const *) argp;
304 		syscallarg(char * const *) envp;
305 	} */ *uap = v;
306 	int error, i;
307 	struct exec_package pack;
308 	struct nameidata nid;
309 	struct vattr attr;
310 	struct ucred *cred = p->p_ucred;
311 	char *argp;
312 	char * const *cpp;
313 	char *dp, *sp;
314 	long argc, envc;
315 	size_t len;
316 	char *stack;
317 	struct ps_strings arginfo;
318 	struct vmspace *vm;
319 	char **tmpfap;
320 	int szsigcode;
321 	struct exec_vmcmd *base_vcp = NULL;
322 
323 	/*
324 	 * Init the namei data to point the file user's program name.
325 	 * This is done here rather than in check_exec(), so that it's
326 	 * possible to override this settings if any of makecmd/probe
327 	 * functions call check_exec() recursively - for example,
328 	 * see exec_script_makecmds().
329 	 */
330 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
331 
332 	/*
333 	 * initialize the fields of the exec package.
334 	 */
335 	pack.ep_name = SCARG(uap, path);
336 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
337 	pack.ep_hdrlen = exec_maxhdrsz;
338 	pack.ep_hdrvalid = 0;
339 	pack.ep_ndp = &nid;
340 	pack.ep_emul_arg = NULL;
341 	pack.ep_vmcmds.evs_cnt = 0;
342 	pack.ep_vmcmds.evs_used = 0;
343 	pack.ep_vap = &attr;
344 	pack.ep_flags = 0;
345 
346 	lockmgr(&exec_lock, LK_SHARED, NULL);
347 
348 	/* see if we can run it. */
349 	if ((error = check_exec(p, &pack)) != 0)
350 		goto freehdr;
351 
352 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
353 
354 	/* allocate an argument buffer */
355 	argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS);
356 #ifdef DIAGNOSTIC
357 	if (argp == (vaddr_t) 0)
358 		panic("execve: argp == NULL");
359 #endif
360 	dp = argp;
361 	argc = 0;
362 
363 	/* copy the fake args list, if there's one, freeing it as we go */
364 	if (pack.ep_flags & EXEC_HASARGL) {
365 		tmpfap = pack.ep_fa;
366 		while (*tmpfap != NULL) {
367 			char *cp;
368 
369 			cp = *tmpfap;
370 			while (*cp)
371 				*dp++ = *cp++;
372 			dp++;
373 
374 			FREE(*tmpfap, M_EXEC);
375 			tmpfap++; argc++;
376 		}
377 		FREE(pack.ep_fa, M_EXEC);
378 		pack.ep_flags &= ~EXEC_HASARGL;
379 	}
380 
381 	/* Now get argv & environment */
382 	if (!(cpp = SCARG(uap, argp))) {
383 		error = EINVAL;
384 		goto bad;
385 	}
386 
387 	if (pack.ep_flags & EXEC_SKIPARG)
388 		cpp++;
389 
390 	while (1) {
391 		len = argp + ARG_MAX - dp;
392 		if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
393 			goto bad;
394 		if (!sp)
395 			break;
396 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
397 			if (error == ENAMETOOLONG)
398 				error = E2BIG;
399 			goto bad;
400 		}
401 		dp += len;
402 		cpp++;
403 		argc++;
404 	}
405 
406 	envc = 0;
407 	/* environment need not be there */
408 	if ((cpp = SCARG(uap, envp)) != NULL ) {
409 		while (1) {
410 			len = argp + ARG_MAX - dp;
411 			if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
412 				goto bad;
413 			if (!sp)
414 				break;
415 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
416 				if (error == ENAMETOOLONG)
417 					error = E2BIG;
418 				goto bad;
419 			}
420 			dp += len;
421 			cpp++;
422 			envc++;
423 		}
424 	}
425 
426 	dp = (char *) ALIGN(dp);
427 
428 	szsigcode = pack.ep_es->es_emul->e_esigcode -
429 	    pack.ep_es->es_emul->e_sigcode;
430 
431 	/* Now check if args & environ fit into new stack */
432 	if (pack.ep_flags & EXEC_32)
433 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
434 		    sizeof(int) + sizeof(int) + dp + STACKGAPLEN +
435 		    szsigcode + sizeof(struct ps_strings)) - argp;
436 	else
437 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
438 		    sizeof(char *) + sizeof(int) + dp + STACKGAPLEN +
439 		    szsigcode + sizeof(struct ps_strings)) - argp;
440 
441 	len = ALIGN(len);	/* make the stack "safely" aligned */
442 
443 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
444 		error = ENOMEM;
445 		goto bad;
446 	}
447 
448 	/* adjust "active stack depth" for process VSZ */
449 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
450 
451 	/*
452 	 * Do whatever is necessary to prepare the address space
453 	 * for remapping.  Note that this might replace the current
454 	 * vmspace with another!
455 	 */
456 	uvmspace_exec(p);
457 
458 	/* Now map address space */
459 	vm = p->p_vmspace;
460 	vm->vm_taddr = (char *) pack.ep_taddr;
461 	vm->vm_tsize = btoc(pack.ep_tsize);
462 	vm->vm_daddr = (char *) pack.ep_daddr;
463 	vm->vm_dsize = btoc(pack.ep_dsize);
464 	vm->vm_ssize = btoc(pack.ep_ssize);
465 	vm->vm_maxsaddr = (char *) pack.ep_maxsaddr;
466 	vm->vm_minsaddr = (char *) pack.ep_minsaddr;
467 
468 	/* create the new process's VM space by running the vmcmds */
469 #ifdef DIAGNOSTIC
470 	if (pack.ep_vmcmds.evs_used == 0)
471 		panic("execve: no vmcmds");
472 #endif
473 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
474 		struct exec_vmcmd *vcp;
475 
476 		vcp = &pack.ep_vmcmds.evs_cmds[i];
477 		if (vcp->ev_flags & VMCMD_RELATIVE) {
478 #ifdef DIAGNOSTIC
479 			if (base_vcp == NULL)
480 				panic("execve: relative vmcmd with no base");
481 			if (vcp->ev_flags & VMCMD_BASE)
482 				panic("execve: illegal base & relative vmcmd");
483 #endif
484 			vcp->ev_addr += base_vcp->ev_addr;
485 		}
486 		error = (*vcp->ev_proc)(p, vcp);
487 #ifdef DEBUG
488 		if (error) {
489 			if (i > 0)
490 				printf("vmcmd[%d] = %#lx/%#lx @ %#lx\n", i-1,
491 				       vcp[-1].ev_addr, vcp[-1].ev_len,
492 				       vcp[-1].ev_offset);
493 			printf("vmcmd[%d] = %#lx/%#lx @ %#lx\n", i,
494 			       vcp->ev_addr, vcp->ev_len, vcp->ev_offset);
495 		}
496 #endif
497 		if (vcp->ev_flags & VMCMD_BASE)
498 			base_vcp = vcp;
499 	}
500 
501 	/* free the vmspace-creation commands, and release their references */
502 	kill_vmcmds(&pack.ep_vmcmds);
503 
504 	/* if an error happened, deallocate and punt */
505 	if (error) {
506 #ifdef DEBUG
507 		printf("execve: vmcmd %i failed: %d\n", i-1, error);
508 #endif
509 		goto exec_abort;
510 	}
511 
512 	/* remember information about the process */
513 	arginfo.ps_nargvstr = argc;
514 	arginfo.ps_nenvstr = envc;
515 
516 	stack = (char *) (vm->vm_minsaddr - len);
517 	/* Now copy argc, args & environ to new stack */
518 	if (!(*pack.ep_es->es_copyargs)(&pack, &arginfo, stack, argp)) {
519 #ifdef DEBUG
520 		printf("execve: copyargs failed\n");
521 #endif
522 		goto exec_abort;
523 	}
524 
525 	/* fill process ps_strings info */
526 	p->p_psstr = (struct ps_strings *)(vm->vm_minsaddr
527 		- sizeof(struct ps_strings));
528 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
529 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
530 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
531 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
532 
533 	/* copy out the process's ps_strings structure */
534 	if (copyout(&arginfo, (char *)p->p_psstr, sizeof(arginfo))) {
535 #ifdef DEBUG
536 		printf("execve: ps_strings copyout failed\n");
537 #endif
538 		goto exec_abort;
539 	}
540 
541 	/* copy out the process's signal trapoline code */
542 	if (szsigcode) {
543 		if (copyout((char *)pack.ep_es->es_emul->e_sigcode,
544 		    p->p_sigacts->ps_sigcode = (char *)p->p_psstr - szsigcode,
545 		    szsigcode)) {
546 #ifdef DEBUG
547 			printf("execve: sig trampoline copyout failed\n");
548 #endif
549 			goto exec_abort;
550 		}
551 #ifdef PMAP_NEED_PROCWR
552 		/* This is code. Let the pmap do what is needed. */
553 		pmap_procwr(p, (vaddr_t)p->p_sigacts->ps_sigcode, szsigcode);
554 #endif
555 	}
556 
557 	stopprofclock(p);	/* stop profiling */
558 	fdcloseexec(p);		/* handle close on exec */
559 	execsigs(p);		/* reset catched signals */
560 	p->p_ctxlink = NULL;	/* reset ucontext link */
561 
562 	/* set command name & other accounting info */
563 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
564 	memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len);
565 	p->p_comm[len] = 0;
566 	p->p_acflag &= ~AFORK;
567 
568 	/* record proc's vnode, for use by procfs and others */
569         if (p->p_textvp)
570                 vrele(p->p_textvp);
571 	VREF(pack.ep_vp);
572 	p->p_textvp = pack.ep_vp;
573 
574 	p->p_flag |= P_EXEC;
575 	if (p->p_flag & P_PPWAIT) {
576 		p->p_flag &= ~P_PPWAIT;
577 		wakeup((caddr_t) p->p_pptr);
578 	}
579 
580 	/*
581 	 * deal with set[ug]id.
582 	 * MNT_NOSUID and P_TRACED have already been used to disable s[ug]id.
583 	 */
584 	if (((attr.va_mode & S_ISUID) != 0 && p->p_ucred->cr_uid != attr.va_uid)
585 	 || ((attr.va_mode & S_ISGID) != 0 && p->p_ucred->cr_gid != attr.va_gid)){
586 		p->p_ucred = crcopy(cred);
587 #ifdef KTRACE
588 		/*
589 		 * If process is being ktraced, turn off - unless
590 		 * root set it.
591 		 */
592 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT))
593 			ktrderef(p);
594 #endif
595 		if (attr.va_mode & S_ISUID)
596 			p->p_ucred->cr_uid = attr.va_uid;
597 		if (attr.va_mode & S_ISGID)
598 			p->p_ucred->cr_gid = attr.va_gid;
599 		p_sugid(p);
600 	} else
601 		p->p_flag &= ~P_SUGID;
602 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
603 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
604 
605 	doexechooks(p);
606 
607 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
608 
609 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
610 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
611 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
612 	vput(pack.ep_vp);
613 
614 	/* setup new registers and do misc. setup. */
615 	(*pack.ep_es->es_setregs)(p, &pack, (u_long) stack);
616 
617 	if (p->p_flag & P_TRACED)
618 		psignal(p, SIGTRAP);
619 
620 	free(pack.ep_hdr, M_EXEC);
621 
622 	/*
623 	 * Call emulation specific exec hook. This can setup setup per-process
624 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
625 	 *
626 	 * If we are executing process of different emulation than the
627 	 * original forked process, call e_proc_exit() of the old emulation
628 	 * first, then e_proc_exec() of new emulation. If the emulation is
629 	 * same, the exec hook code should deallocate any old emulation
630 	 * resources held previously by this process.
631 	 */
632 	if (p->p_emul && p->p_emul->e_proc_exit
633 	    && p->p_emul != pack.ep_es->es_emul)
634 		(*p->p_emul->e_proc_exit)(p);
635 
636 	/*
637 	 * Call exec hook. Emulation code may NOT store reference to anything
638 	 * from &pack.
639 	 */
640         if (pack.ep_es->es_emul->e_proc_exec)
641                 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack);
642 
643 	/* update p_emul, the old value is no longer needed */
644 	p->p_emul = pack.ep_es->es_emul;
645 #ifdef __HAVE_SYSCALL_INTERN
646 	(*p->p_emul->e_syscall_intern)(p);
647 #endif
648 #ifdef KTRACE
649 	if (KTRPOINT(p, KTR_EMUL))
650 		ktremul(p);
651 #endif
652 
653 	lockmgr(&exec_lock, LK_RELEASE, NULL);
654 
655 	return (EJUSTRETURN);
656 
657 bad:
658 	/* free the vmspace-creation commands, and release their references */
659 	kill_vmcmds(&pack.ep_vmcmds);
660 	/* kill any opened file descriptor, if necessary */
661 	if (pack.ep_flags & EXEC_HASFD) {
662 		pack.ep_flags &= ~EXEC_HASFD;
663 		(void) fdrelease(p, pack.ep_fd);
664 	}
665 	/* close and put the exec'd file */
666 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
667 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
668 	vput(pack.ep_vp);
669 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
670 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
671 
672 freehdr:
673 	lockmgr(&exec_lock, LK_RELEASE, NULL);
674 
675 	free(pack.ep_hdr, M_EXEC);
676 	return error;
677 
678 exec_abort:
679 	lockmgr(&exec_lock, LK_RELEASE, NULL);
680 
681 	/*
682 	 * the old process doesn't exist anymore.  exit gracefully.
683 	 * get rid of the (new) address space we have created, if any, get rid
684 	 * of our namei data and vnode, and exit noting failure
685 	 */
686 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
687 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
688 	if (pack.ep_emul_arg)
689 		FREE(pack.ep_emul_arg, M_TEMP);
690 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
691 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
692 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
693 	vput(pack.ep_vp);
694 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
695 	free(pack.ep_hdr, M_EXEC);
696 	exit1(p, W_EXITCODE(0, SIGABRT));
697 	exit1(p, -1);
698 
699 	/* NOTREACHED */
700 	return 0;
701 }
702 
703 
704 void *
705 copyargs(struct exec_package *pack, struct ps_strings *arginfo,
706     void *stack, void *argp)
707 {
708 	char **cpp = stack;
709 	char *dp, *sp;
710 	size_t len;
711 	void *nullp = NULL;
712 	long argc = arginfo->ps_nargvstr;
713 	long envc = arginfo->ps_nenvstr;
714 
715 #ifdef __sparc_v9__
716 	/* XXX Temporary hack for argc format conversion. */
717 	argc <<= 32;
718 #endif
719 	if (copyout(&argc, cpp++, sizeof(argc)))
720 		return NULL;
721 #ifdef __sparc_v9__
722 	/* XXX Temporary hack for argc format conversion. */
723 	argc >>= 32;
724 #endif
725 
726 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen);
727 	sp = argp;
728 
729 	/* XXX don't copy them out, remap them! */
730 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
731 
732 	for (; --argc >= 0; sp += len, dp += len)
733 		if (copyout(&dp, cpp++, sizeof(dp)) ||
734 		    copyoutstr(sp, dp, ARG_MAX, &len))
735 			return NULL;
736 
737 	if (copyout(&nullp, cpp++, sizeof(nullp)))
738 		return NULL;
739 
740 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
741 
742 	for (; --envc >= 0; sp += len, dp += len)
743 		if (copyout(&dp, cpp++, sizeof(dp)) ||
744 		    copyoutstr(sp, dp, ARG_MAX, &len))
745 			return NULL;
746 
747 	if (copyout(&nullp, cpp++, sizeof(nullp)))
748 		return NULL;
749 
750 	return cpp;
751 }
752 
753 #ifdef LKM
754 /*
755  * Find an emulation of given name in list of emulations.
756  */
757 static const struct emul *
758 emul_search(name)
759 	const char *name;
760 {
761 	struct emul_entry *it;
762 
763 	LIST_FOREACH(it, &el_head, el_list) {
764 		if (strcmp(name, it->el_emul->e_name) == 0)
765 			return it->el_emul;
766 	}
767 
768 	return NULL;
769 }
770 
771 /*
772  * Add an emulation to list, if it's not there already.
773  */
774 int
775 emul_register(emul, ro_entry)
776 	const struct emul *emul;
777 	int ro_entry;
778 {
779 	struct emul_entry *ee;
780 	int error = 0;
781 
782 	lockmgr(&exec_lock, LK_SHARED, NULL);
783 
784 	if (emul_search(emul->e_name)) {
785 		error = EEXIST;
786 		goto out;
787 	}
788 
789 	MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry),
790 		M_EXEC, M_WAITOK);
791 	ee->el_emul = emul;
792 	ee->ro_entry = ro_entry;
793 	LIST_INSERT_HEAD(&el_head, ee, el_list);
794 
795     out:
796 	lockmgr(&exec_lock, LK_RELEASE, NULL);
797 	return error;
798 }
799 
800 /*
801  * Remove emulation with name 'name' from list of supported emulations.
802  */
803 int
804 emul_unregister(name)
805 	const char *name;
806 {
807 	struct emul_entry *it;
808 	int i, error = 0;
809 	const struct proclist_desc *pd;
810 	struct proc *ptmp;
811 
812 	lockmgr(&exec_lock, LK_SHARED, NULL);
813 
814 	LIST_FOREACH(it, &el_head, el_list) {
815 		if (strcmp(it->el_emul->e_name, name) == 0)
816 			break;
817 	}
818 
819 	if (!it) {
820 		error = ENOENT;
821 		goto out;
822 	}
823 
824 	if (it->ro_entry) {
825 		error = EBUSY;
826 		goto out;
827 	}
828 
829 	/* test if any execw[] entry is still using this */
830 	for(i=0; i < nexecs; i++) {
831 		if (execsw[i]->es_emul == it->el_emul) {
832 			error = EBUSY;
833 			goto out;
834 		}
835 	}
836 
837 	/*
838 	 * Test if any process is running under this emulation - since
839 	 * emul_unregister() is running quite sendomly, it's better
840 	 * to do expensive check here than to use any locking.
841 	 */
842 	proclist_lock_read();
843 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
844 		LIST_FOREACH(ptmp, pd->pd_list, p_list) {
845 			if (ptmp->p_emul == it->el_emul) {
846 				error = EBUSY;
847 				break;
848 			}
849 		}
850 	}
851 	proclist_unlock_read();
852 
853 	if (error)
854 		goto out;
855 
856 
857 	/* entry is not used, remove it */
858 	LIST_REMOVE(it, el_list);
859 	FREE(it, M_EXEC);
860 
861     out:
862 	lockmgr(&exec_lock, LK_RELEASE, NULL);
863 	return error;
864 }
865 
866 /*
867  * Add execsw[] entry.
868  */
869 int
870 exec_add(esp, e_name)
871 	struct execsw *esp;
872 	const char *e_name;
873 {
874 	struct exec_entry *it;
875 	int error = 0;
876 
877 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
878 
879 	if (!esp->es_emul) {
880 		esp->es_emul = emul_search(e_name);
881 		if (!esp->es_emul) {
882 			error = ENOENT;
883 			goto out;
884 		}
885 	}
886 
887 	LIST_FOREACH(it, &ex_head, ex_list) {
888 		/* assume tuple (makecmds, probe_func, emulation) is unique */
889 		if (it->es->es_check == esp->es_check
890 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
891 		    && it->es->es_emul == esp->es_emul) {
892 			error = EEXIST;
893 			goto out;
894 		}
895 	}
896 
897 	/* if we got here, the entry doesn't exist yet */
898 	MALLOC(it, struct exec_entry *, sizeof(struct exec_entry),
899 		M_EXEC, M_WAITOK);
900 	it->es = esp;
901 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
902 
903 	/* update execsw[] */
904 	exec_init(0);
905 
906     out:
907 	lockmgr(&exec_lock, LK_RELEASE, NULL);
908 	return error;
909 }
910 
911 /*
912  * Remove execsw[] entry.
913  */
914 int
915 exec_remove(esp)
916 	const struct execsw *esp;
917 {
918 	struct exec_entry *it;
919 	int error = 0;
920 
921 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
922 
923 	LIST_FOREACH(it, &ex_head, ex_list) {
924 		/* assume tuple (makecmds, probe_func, emulation) is unique */
925 		if (it->es->es_check == esp->es_check
926 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
927 		    && it->es->es_emul == esp->es_emul)
928 			break;
929 	}
930 	if (!it) {
931 		error = ENOENT;
932 		goto out;
933 	}
934 
935 	/* remove item from list and free resources */
936 	LIST_REMOVE(it, ex_list);
937 	FREE(it, M_EXEC);
938 
939 	/* update execsw[] */
940 	exec_init(0);
941 
942     out:
943 	lockmgr(&exec_lock, LK_RELEASE, NULL);
944 	return error;
945 }
946 
947 static void
948 link_es(listp, esp)
949 	struct execsw_entry **listp;
950 	const struct execsw *esp;
951 {
952 	struct execsw_entry *et, *e1;
953 
954 	MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry),
955 			M_TEMP, M_WAITOK);
956 	et->next = NULL;
957 	et->es = esp;
958 	if (*listp == NULL) {
959 		*listp = et;
960 		return;
961 	}
962 
963 	switch(et->es->es_prio) {
964 	case EXECSW_PRIO_FIRST:
965 		/* put new entry as the first */
966 		et->next = *listp;
967 		*listp = et;
968 		break;
969 	case EXECSW_PRIO_ANY:
970 		/* put new entry after all *_FIRST and *_ANY entries */
971 		for(e1 = *listp; e1->next
972 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
973 			e1 = e1->next);
974 		et->next = e1->next;
975 		e1->next = et;
976 		break;
977 	case EXECSW_PRIO_LAST:
978 		/* put new entry as the last one */
979 		for(e1 = *listp; e1->next; e1 = e1->next);
980 		e1->next = et;
981 		break;
982 	default:
983 #ifdef DIAGNOSTIC
984 		panic("execw[] entry with unknown priority %d found\n",
985 			et->es->es_prio);
986 #endif
987 		break;
988 	}
989 }
990 
991 /*
992  * Initialize exec structures. If init_boot is true, also does necessary
993  * one-time initialization (it's called from main() that way).
994  * Once system is multiuser, this should be called with exec_lock hold,
995  * i.e. via exec_{add|remove}().
996  */
997 int
998 exec_init(init_boot)
999 	int init_boot;
1000 {
1001 	const struct execsw **new_es, * const *old_es;
1002 	struct execsw_entry *list, *e1;
1003 	struct exec_entry *e2;
1004 	int i, es_sz;
1005 
1006 	if (init_boot) {
1007 		/* do one-time initializations */
1008 		lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1009 
1010 		/* register compiled-in emulations */
1011 		for(i=0; i < nexecs_builtin; i++) {
1012 			if (execsw_builtin[i].es_emul)
1013 				emul_register(execsw_builtin[i].es_emul, 1);
1014 		}
1015 #ifdef DIAGNOSTIC
1016 		if (i == 0)
1017 			panic("no emulations found in execsw_builtin[]\n");
1018 #endif
1019 	}
1020 
1021 	/*
1022 	 * Build execsw[] array from builtin entries and entries added
1023 	 * at runtime.
1024 	 */
1025 	list = NULL;
1026 	for(i=0; i < nexecs_builtin; i++)
1027 		link_es(&list, &execsw_builtin[i]);
1028 
1029 	/* Add dynamically loaded entries */
1030 	es_sz = nexecs_builtin;
1031 	LIST_FOREACH(e2, &ex_head, ex_list) {
1032 		link_es(&list, e2->es);
1033 		es_sz++;
1034 	}
1035 
1036 	/*
1037 	 * Now that we have sorted all execw entries, create new execsw[]
1038 	 * and free no longer needed memory in the process.
1039 	 */
1040 	new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK);
1041 	for(i=0; list; i++) {
1042 		new_es[i] = list->es;
1043 		e1 = list->next;
1044 		FREE(list, M_TEMP);
1045 		list = e1;
1046 	}
1047 
1048 	/*
1049 	 * New execsw[] array built, now replace old execsw[] and free
1050 	 * used memory.
1051 	 */
1052 	old_es = execsw;
1053 	execsw = new_es;
1054 	nexecs = es_sz;
1055 	if (old_es)
1056 		free((void *)old_es, M_EXEC);
1057 
1058 	/*
1059 	 * Figure out the maximum size of an exec header.
1060 	 */
1061 	exec_maxhdrsz = 0;
1062 	for (i = 0; i < nexecs; i++) {
1063 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1064 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1065 	}
1066 
1067 	return 0;
1068 }
1069 #endif
1070 
1071 #ifndef LKM
1072 /*
1073  * Simplified exec_init() for kernels without LKMs. Only initialize
1074  * exec_maxhdrsz and execsw[].
1075  */
1076 int
1077 exec_init(init_boot)
1078 	int init_boot;
1079 {
1080 	int i;
1081 
1082 #ifdef DIAGNOSTIC
1083 	if (!init_boot)
1084 		panic("exec_init(): called with init_boot == 0");
1085 #endif
1086 
1087 	/* do one-time initializations */
1088 	lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1089 
1090 	nexecs = nexecs_builtin;
1091 	execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK);
1092 
1093 	/*
1094 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1095 	 */
1096 	exec_maxhdrsz = 0;
1097 	for(i=0; i < nexecs; i++) {
1098 		execsw[i] = &execsw_builtin[i];
1099 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1100 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1101 	}
1102 
1103 	return 0;
1104 
1105 }
1106 #endif /* !LKM */
1107