xref: /netbsd-src/sys/kern/exec_elf.c (revision f89f6560d453f5e37386cc7938c072d2f528b9fa)
1 /*	$NetBSD: exec_elf.c,v 1.71 2015/03/20 20:36:28 maxv Exp $	*/
2 
3 /*-
4  * Copyright (c) 1994, 2000, 2005 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Christos Zoulas.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1996 Christopher G. Demetriou
34  * All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. The name of the author may not be used to endorse or promote products
45  *    derived from this software without specific prior written permission
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
48  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
49  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
50  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
51  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
52  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
53  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
54  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
55  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
56  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
57  */
58 
59 #include <sys/cdefs.h>
60 __KERNEL_RCSID(1, "$NetBSD: exec_elf.c,v 1.71 2015/03/20 20:36:28 maxv Exp $");
61 
62 #ifdef _KERNEL_OPT
63 #include "opt_pax.h"
64 #endif /* _KERNEL_OPT */
65 
66 #include <sys/param.h>
67 #include <sys/proc.h>
68 #include <sys/kmem.h>
69 #include <sys/namei.h>
70 #include <sys/vnode.h>
71 #include <sys/exec.h>
72 #include <sys/exec_elf.h>
73 #include <sys/syscall.h>
74 #include <sys/signalvar.h>
75 #include <sys/mount.h>
76 #include <sys/stat.h>
77 #include <sys/kauth.h>
78 #include <sys/bitops.h>
79 #include <sys/cprng.h>
80 
81 #include <sys/cpu.h>
82 #include <machine/reg.h>
83 
84 #include <compat/common/compat_util.h>
85 
86 #include <sys/pax.h>
87 #include <uvm/uvm_param.h>
88 
89 extern struct emul emul_netbsd;
90 
91 #define elf_check_header	ELFNAME(check_header)
92 #define elf_copyargs		ELFNAME(copyargs)
93 #define elf_load_interp		ELFNAME(load_interp)
94 #define elf_load_psection	ELFNAME(load_psection)
95 #define exec_elf_makecmds	ELFNAME2(exec,makecmds)
96 #define netbsd_elf_signature	ELFNAME2(netbsd,signature)
97 #define netbsd_elf_probe	ELFNAME2(netbsd,probe)
98 #define	coredump		ELFNAMEEND(coredump)
99 #define	elf_free_emul_arg	ELFNAME(free_emul_arg)
100 
101 static int
102 elf_load_interp(struct lwp *, struct exec_package *, char *,
103     struct exec_vmcmd_set *, u_long *, Elf_Addr *);
104 static void
105 elf_load_psection(struct exec_vmcmd_set *, struct vnode *, const Elf_Phdr *,
106     Elf_Addr *, u_long *, int);
107 
108 int	netbsd_elf_signature(struct lwp *, struct exec_package *, Elf_Ehdr *);
109 int	netbsd_elf_probe(struct lwp *, struct exec_package *, void *, char *,
110 	    vaddr_t *);
111 
112 static void	elf_free_emul_arg(void *);
113 
114 /* round up and down to page boundaries. */
115 #define	ELF_ROUND(a, b)		(((a) + (b) - 1) & ~((b) - 1))
116 #define	ELF_TRUNC(a, b)		((a) & ~((b) - 1))
117 
118 static void
119 elf_placedynexec(struct lwp *l, struct exec_package *epp, Elf_Ehdr *eh,
120     Elf_Phdr *ph)
121 {
122 	Elf_Addr align, offset;
123 	int i;
124 
125 	for (align = i = 0; i < eh->e_phnum; i++)
126 		if (ph[i].p_type == PT_LOAD && ph[i].p_align > align)
127 			align = ph[i].p_align;
128 
129 #ifdef PAX_ASLR
130 	if (pax_aslr_active(l)) {
131 		size_t pax_align, l2, delta;
132 		uint32_t r;
133 
134 		pax_align = align;
135 
136 		r = cprng_fast32();
137 
138 		if (pax_align == 0)
139 			pax_align = PGSHIFT;
140 		l2 = ilog2(pax_align);
141 		delta = PAX_ASLR_DELTA(r, l2, PAX_ASLR_DELTA_EXEC_LEN);
142 		offset = ELF_TRUNC(delta, pax_align) + PAGE_SIZE;
143 #ifdef PAX_ASLR_DEBUG
144 		uprintf("r=0x%x l2=0x%zx PGSHIFT=0x%x Delta=0x%zx\n", r, l2,
145 		    PGSHIFT, delta);
146 		uprintf("pax offset=0x%llx entry=0x%llx\n",
147 		    (unsigned long long)offset,
148 		    (unsigned long long)eh->e_entry);
149 #endif /* PAX_ASLR_DEBUG */
150 	} else
151 #endif /* PAX_ASLR */
152 		offset = MAX(align, PAGE_SIZE);
153 
154 	offset += epp->ep_vm_minaddr;
155 
156 	for (i = 0; i < eh->e_phnum; i++)
157 		ph[i].p_vaddr += offset;
158 	epp->ep_entryoffset = offset;
159 	eh->e_entry += offset;
160 }
161 
162 /*
163  * Copy arguments onto the stack in the normal way, but add some
164  * extra information in case of dynamic binding.
165  */
166 int
167 elf_copyargs(struct lwp *l, struct exec_package *pack,
168     struct ps_strings *arginfo, char **stackp, void *argp)
169 {
170 	size_t len, vlen;
171 	AuxInfo ai[ELF_AUX_ENTRIES], *a, *execname;
172 	struct elf_args *ap;
173 	int error;
174 
175 	if ((error = copyargs(l, pack, arginfo, stackp, argp)) != 0)
176 		return error;
177 
178 	a = ai;
179 	execname = NULL;
180 
181 	memset(ai, 0, sizeof(ai));
182 
183 	/*
184 	 * Push extra arguments on the stack needed by dynamically
185 	 * linked binaries
186 	 */
187 	if ((ap = (struct elf_args *)pack->ep_emul_arg)) {
188 		struct vattr *vap = pack->ep_vap;
189 
190 		a->a_type = AT_PHDR;
191 		a->a_v = ap->arg_phaddr;
192 		a++;
193 
194 		a->a_type = AT_PHENT;
195 		a->a_v = ap->arg_phentsize;
196 		a++;
197 
198 		a->a_type = AT_PHNUM;
199 		a->a_v = ap->arg_phnum;
200 		a++;
201 
202 		a->a_type = AT_PAGESZ;
203 		a->a_v = PAGE_SIZE;
204 		a++;
205 
206 		a->a_type = AT_BASE;
207 		a->a_v = ap->arg_interp;
208 		a++;
209 
210 		a->a_type = AT_FLAGS;
211 		a->a_v = 0;
212 		a++;
213 
214 		a->a_type = AT_ENTRY;
215 		a->a_v = ap->arg_entry;
216 		a++;
217 
218 		a->a_type = AT_EUID;
219 		if (vap->va_mode & S_ISUID)
220 			a->a_v = vap->va_uid;
221 		else
222 			a->a_v = kauth_cred_geteuid(l->l_cred);
223 		a++;
224 
225 		a->a_type = AT_RUID;
226 		a->a_v = kauth_cred_getuid(l->l_cred);
227 		a++;
228 
229 		a->a_type = AT_EGID;
230 		if (vap->va_mode & S_ISGID)
231 			a->a_v = vap->va_gid;
232 		else
233 			a->a_v = kauth_cred_getegid(l->l_cred);
234 		a++;
235 
236 		a->a_type = AT_RGID;
237 		a->a_v = kauth_cred_getgid(l->l_cred);
238 		a++;
239 
240 		a->a_type = AT_STACKBASE;
241 		a->a_v = l->l_proc->p_stackbase;
242 		a++;
243 
244 		if (pack->ep_path) {
245 			execname = a;
246 			a->a_type = AT_SUN_EXECNAME;
247 			a++;
248 		}
249 
250 		exec_free_emul_arg(pack);
251 	}
252 
253 	a->a_type = AT_NULL;
254 	a->a_v = 0;
255 	a++;
256 
257 	vlen = (a - ai) * sizeof(ai[0]);
258 
259 	KASSERT(vlen <= sizeof(ai));
260 
261 	if (execname) {
262 		char *path = pack->ep_path;
263 		execname->a_v = (uintptr_t)(*stackp + vlen);
264 		len = strlen(path) + 1;
265 		if ((error = copyout(path, (*stackp + vlen), len)) != 0)
266 			return error;
267 		len = ALIGN(len);
268 	} else
269 		len = 0;
270 
271 	if ((error = copyout(ai, *stackp, vlen)) != 0)
272 		return error;
273 	*stackp += vlen + len;
274 
275 	return 0;
276 }
277 
278 /*
279  * elf_check_header():
280  *
281  * Check header for validity; return 0 if ok, ENOEXEC if error
282  */
283 int
284 elf_check_header(Elf_Ehdr *eh)
285 {
286 
287 	if (memcmp(eh->e_ident, ELFMAG, SELFMAG) != 0 ||
288 	    eh->e_ident[EI_CLASS] != ELFCLASS)
289 		return ENOEXEC;
290 
291 	switch (eh->e_machine) {
292 
293 	ELFDEFNNAME(MACHDEP_ID_CASES)
294 
295 	default:
296 		return ENOEXEC;
297 	}
298 
299 	if (ELF_EHDR_FLAGS_OK(eh) == 0)
300 		return ENOEXEC;
301 
302 	if (eh->e_shnum > ELF_MAXSHNUM || eh->e_phnum > ELF_MAXPHNUM)
303 		return ENOEXEC;
304 
305 	return 0;
306 }
307 
308 /*
309  * elf_load_psection():
310  *
311  * Load a psection at the appropriate address
312  */
313 static void
314 elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp,
315     const Elf_Phdr *ph, Elf_Addr *addr, u_long *size, int flags)
316 {
317 	u_long msize, psize, rm, rf;
318 	long diff, offset;
319 	int vmprot = 0;
320 
321 	/*
322 	 * If the user specified an address, then we load there.
323 	 */
324 	if (*addr == ELFDEFNNAME(NO_ADDR))
325 		*addr = ph->p_vaddr;
326 
327 	if (ph->p_align > 1) {
328 		/*
329 		 * Make sure we are virtually aligned as we are supposed to be.
330 		 */
331 		diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align);
332 		KASSERT(*addr - diff == ELF_TRUNC(*addr, ph->p_align));
333 		/*
334 		 * But make sure to not map any pages before the start of the
335 		 * psection by limiting the difference to within a page.
336 		 */
337 		diff &= PAGE_MASK;
338 	} else
339 		diff = 0;
340 
341 	vmprot |= (ph->p_flags & PF_R) ? VM_PROT_READ : 0;
342 	vmprot |= (ph->p_flags & PF_W) ? VM_PROT_WRITE : 0;
343 	vmprot |= (ph->p_flags & PF_X) ? VM_PROT_EXECUTE : 0;
344 
345 	/*
346 	 * Adjust everything so it all starts on a page boundary.
347 	 */
348 	*addr -= diff;
349 	offset = ph->p_offset - diff;
350 	*size = ph->p_filesz + diff;
351 	msize = ph->p_memsz + diff;
352 
353 	if (ph->p_align >= PAGE_SIZE) {
354 		if ((ph->p_flags & PF_W) != 0) {
355 			/*
356 			 * Because the pagedvn pager can't handle zero fill
357 			 * of the last data page if it's not page aligned we
358 			 * map the last page readvn.
359 			 */
360 			psize = trunc_page(*size);
361 		} else {
362 			psize = round_page(*size);
363 		}
364 	} else {
365 		psize = *size;
366 	}
367 
368 	if (psize > 0) {
369 		NEW_VMCMD2(vcset, ph->p_align < PAGE_SIZE ?
370 		    vmcmd_map_readvn : vmcmd_map_pagedvn, psize, *addr, vp,
371 		    offset, vmprot, flags);
372 		flags &= VMCMD_RELATIVE;
373 	}
374 	if (psize < *size) {
375 		NEW_VMCMD2(vcset, vmcmd_map_readvn, *size - psize,
376 		    *addr + psize, vp, offset + psize, vmprot, flags);
377 	}
378 
379 	/*
380 	 * Check if we need to extend the size of the segment (does
381 	 * bss extend page the next page boundary)?
382 	 */
383 	rm = round_page(*addr + msize);
384 	rf = round_page(*addr + *size);
385 
386 	if (rm != rf) {
387 		NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP,
388 		    0, vmprot, flags & VMCMD_RELATIVE);
389 		*size = msize;
390 	}
391 }
392 
393 /*
394  * elf_load_interp():
395  *
396  * Load an interpreter pointed to by path.
397  */
398 static int
399 elf_load_interp(struct lwp *l, struct exec_package *epp, char *path,
400     struct exec_vmcmd_set *vcset, u_long *entryoff, Elf_Addr *last)
401 {
402 	int error, i;
403 	struct vnode *vp;
404 	struct vattr attr;
405 	Elf_Ehdr eh;
406 	Elf_Phdr *ph = NULL;
407 	const Elf_Phdr *base_ph;
408 	const Elf_Phdr *last_ph;
409 	u_long phsize;
410 	Elf_Addr addr = *last;
411 	struct proc *p;
412 	bool use_topdown;
413 
414 	p = l->l_proc;
415 
416 	KASSERT(p->p_vmspace);
417 	if (__predict_true(p->p_vmspace != proc0.p_vmspace)) {
418 		use_topdown = p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN;
419 	} else {
420 #ifdef __USE_TOPDOWN_VM
421 		use_topdown = epp->ep_flags & EXEC_TOPDOWN_VM;
422 #else
423 		use_topdown = false;
424 #endif
425 	}
426 
427 	/*
428 	 * 1. open file
429 	 * 2. read filehdr
430 	 * 3. map text, data, and bss out of it using VM_*
431 	 */
432 	vp = epp->ep_interp;
433 	if (vp == NULL) {
434 		error = emul_find_interp(l, epp, path);
435 		if (error != 0)
436 			return error;
437 		vp = epp->ep_interp;
438 	}
439 	/* We'll tidy this ourselves - otherwise we have locking issues */
440 	epp->ep_interp = NULL;
441 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
442 
443 	/*
444 	 * Similarly, if it's not marked as executable, or it's not a regular
445 	 * file, we don't allow it to be used.
446 	 */
447 	if (vp->v_type != VREG) {
448 		error = EACCES;
449 		goto badunlock;
450 	}
451 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
452 		goto badunlock;
453 
454 	/* get attributes */
455 	if ((error = VOP_GETATTR(vp, &attr, l->l_cred)) != 0)
456 		goto badunlock;
457 
458 	/*
459 	 * Check mount point.  Though we're not trying to exec this binary,
460 	 * we will be executing code from it, so if the mount point
461 	 * disallows execution or set-id-ness, we punt or kill the set-id.
462 	 */
463 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
464 		error = EACCES;
465 		goto badunlock;
466 	}
467 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
468 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
469 
470 #ifdef notyet /* XXX cgd 960926 */
471 	XXX cgd 960926: (maybe) VOP_OPEN it (and VOP_CLOSE in copyargs?)
472 
473 	XXXps: this problem will make it impossible to use an interpreter
474 	from a file system which actually does something in VOP_OPEN
475 #endif
476 
477 	error = vn_marktext(vp);
478 	if (error)
479 		goto badunlock;
480 
481 	VOP_UNLOCK(vp);
482 
483 	if ((error = exec_read_from(l, vp, 0, &eh, sizeof(eh))) != 0)
484 		goto bad;
485 
486 	if ((error = elf_check_header(&eh)) != 0)
487 		goto bad;
488 	if (eh.e_type != ET_DYN || eh.e_phnum == 0) {
489 		error = ENOEXEC;
490 		goto bad;
491 	}
492 
493 	phsize = eh.e_phnum * sizeof(Elf_Phdr);
494 	ph = kmem_alloc(phsize, KM_SLEEP);
495 
496 	if ((error = exec_read_from(l, vp, eh.e_phoff, ph, phsize)) != 0)
497 		goto bad;
498 
499 #ifdef ELF_INTERP_NON_RELOCATABLE
500 	/*
501 	 * Evil hack:  Only MIPS should be non-relocatable, and the
502 	 * psections should have a high address (typically 0x5ffe0000).
503 	 * If it's now relocatable, it should be linked at 0 and the
504 	 * psections should have zeros in the upper part of the address.
505 	 * Otherwise, force the load at the linked address.
506 	 */
507 	if (*last == ELF_LINK_ADDR && (ph->p_vaddr & 0xffff0000) == 0)
508 		*last = ELFDEFNNAME(NO_ADDR);
509 #endif
510 
511 	/*
512 	 * If no position to load the interpreter was set by a probe
513 	 * function, pick the same address that a non-fixed mmap(0, ..)
514 	 * would (i.e. something safely out of the way).
515 	 */
516 	if (*last == ELFDEFNNAME(NO_ADDR)) {
517 		u_long limit = 0;
518 		/*
519 		 * Find the start and ending addresses of the psections to
520 		 * be loaded.  This will give us the size.
521 		 */
522 		for (i = 0, base_ph = NULL; i < eh.e_phnum; i++) {
523 			if (ph[i].p_type == PT_LOAD) {
524 				u_long psize = ph[i].p_vaddr + ph[i].p_memsz;
525 				if (base_ph == NULL)
526 					base_ph = &ph[i];
527 				if (psize > limit)
528 					limit = psize;
529 			}
530 		}
531 
532 		if (base_ph == NULL) {
533 			error = ENOEXEC;
534 			goto bad;
535 		}
536 
537 		/*
538 		 * Now compute the size and load address.
539 		 */
540 		addr = (*epp->ep_esch->es_emul->e_vm_default_addr)(p,
541 		    epp->ep_daddr,
542 		    round_page(limit) - trunc_page(base_ph->p_vaddr));
543 	} else
544 		addr = *last; /* may be ELF_LINK_ADDR */
545 
546 	/*
547 	 * Load all the necessary sections
548 	 */
549 	for (i = 0, base_ph = NULL, last_ph = NULL; i < eh.e_phnum; i++) {
550 		switch (ph[i].p_type) {
551 		case PT_LOAD: {
552 			u_long size;
553 			int flags;
554 
555 			if (base_ph == NULL) {
556 				/*
557 				 * First encountered psection is always the
558 				 * base psection.  Make sure it's aligned
559 				 * properly (align down for topdown and align
560 				 * upwards for not topdown).
561 				 */
562 				base_ph = &ph[i];
563 				flags = VMCMD_BASE;
564 				if (addr == ELF_LINK_ADDR)
565 					addr = ph[i].p_vaddr;
566 				if (use_topdown)
567 					addr = ELF_TRUNC(addr, ph[i].p_align);
568 				else
569 					addr = ELF_ROUND(addr, ph[i].p_align);
570 			} else {
571 				u_long limit = round_page(last_ph->p_vaddr
572 				    + last_ph->p_memsz);
573 				u_long base = trunc_page(ph[i].p_vaddr);
574 
575 				/*
576 				 * If there is a gap in between the psections,
577 				 * map it as inaccessible so nothing else
578 				 * mmap'ed will be placed there.
579 				 */
580 				if (limit != base) {
581 					NEW_VMCMD2(vcset, vmcmd_map_zero,
582 					    base - limit,
583 					    limit - base_ph->p_vaddr, NULLVP,
584 					    0, VM_PROT_NONE, VMCMD_RELATIVE);
585 				}
586 
587 				addr = ph[i].p_vaddr - base_ph->p_vaddr;
588 				flags = VMCMD_RELATIVE;
589 			}
590 			last_ph = &ph[i];
591 			elf_load_psection(vcset, vp, &ph[i], &addr,
592 			    &size, flags);
593 			/*
594 			 * If entry is within this psection then this
595 			 * must contain the .text section.  *entryoff is
596 			 * relative to the base psection.
597 			 */
598 			if (eh.e_entry >= ph[i].p_vaddr &&
599 			    eh.e_entry < (ph[i].p_vaddr + size)) {
600 				*entryoff = eh.e_entry - base_ph->p_vaddr;
601 			}
602 			addr += size;
603 			break;
604 		}
605 
606 		default:
607 			break;
608 		}
609 	}
610 
611 	kmem_free(ph, phsize);
612 	/*
613 	 * This value is ignored if TOPDOWN.
614 	 */
615 	*last = addr;
616 	vrele(vp);
617 	return 0;
618 
619 badunlock:
620 	VOP_UNLOCK(vp);
621 
622 bad:
623 	if (ph != NULL)
624 		kmem_free(ph, phsize);
625 #ifdef notyet /* XXX cgd 960926 */
626 	(maybe) VOP_CLOSE it
627 #endif
628 	vrele(vp);
629 	return error;
630 }
631 
632 /*
633  * exec_elf_makecmds(): Prepare an Elf binary's exec package
634  *
635  * First, set of the various offsets/lengths in the exec package.
636  *
637  * Then, mark the text image busy (so it can be demand paged) or error
638  * out if this is not possible.  Finally, set up vmcmds for the
639  * text, data, bss, and stack segments.
640  */
641 int
642 exec_elf_makecmds(struct lwp *l, struct exec_package *epp)
643 {
644 	Elf_Ehdr *eh = epp->ep_hdr;
645 	Elf_Phdr *ph, *pp;
646 	Elf_Addr phdr = 0, computed_phdr = 0, pos = 0, end_text = 0;
647 	int error, i;
648 	char *interp = NULL;
649 	u_long phsize;
650 	struct elf_args *ap;
651 	bool is_dyn = false;
652 
653 	if (epp->ep_hdrvalid < sizeof(Elf_Ehdr))
654 		return ENOEXEC;
655 	if ((error = elf_check_header(eh)) != 0)
656 		return error;
657 
658 	if (eh->e_type == ET_DYN)
659 		/*
660 		 * XXX allow for executing shared objects. It seems silly
661 		 * but other ELF-based systems allow it as well.
662 		 */
663 		is_dyn = true;
664 	else if (eh->e_type != ET_EXEC)
665 		return ENOEXEC;
666 
667 	if (eh->e_phnum == 0)
668 		return ENOEXEC;
669 
670 	error = vn_marktext(epp->ep_vp);
671 	if (error)
672 		return error;
673 
674 	/*
675 	 * Allocate space to hold all the program headers, and read them
676 	 * from the file
677 	 */
678 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
679 	ph = kmem_alloc(phsize, KM_SLEEP);
680 
681 	if ((error = exec_read_from(l, epp->ep_vp, eh->e_phoff, ph, phsize)) !=
682 	    0)
683 		goto bad;
684 
685 	epp->ep_taddr = epp->ep_tsize = ELFDEFNNAME(NO_ADDR);
686 	epp->ep_daddr = epp->ep_dsize = ELFDEFNNAME(NO_ADDR);
687 
688 	for (i = 0; i < eh->e_phnum; i++) {
689 		pp = &ph[i];
690 		if (pp->p_type == PT_INTERP) {
691 			if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN) {
692 				error = ENOEXEC;
693 				goto bad;
694 			}
695 			interp = PNBUF_GET();
696 			if ((error = exec_read_from(l, epp->ep_vp,
697 			    pp->p_offset, interp, pp->p_filesz)) != 0)
698 				goto bad;
699 			/* Ensure interp is NUL-terminated and of the expected length */
700 			if (strnlen(interp, pp->p_filesz) != pp->p_filesz - 1) {
701 				error = ENOEXEC;
702 				goto bad;
703 			}
704 			break;
705 		}
706 	}
707 
708 	/*
709 	 * On the same architecture, we may be emulating different systems.
710 	 * See which one will accept this executable.
711 	 *
712 	 * Probe functions would normally see if the interpreter (if any)
713 	 * exists. Emulation packages may possibly replace the interpreter in
714 	 * interp[] with a changed path (/emul/xxx/<path>).
715 	 */
716 	pos = ELFDEFNNAME(NO_ADDR);
717 	if (epp->ep_esch->u.elf_probe_func) {
718 		vaddr_t startp = (vaddr_t)pos;
719 
720 		error = (*epp->ep_esch->u.elf_probe_func)(l, epp, eh, interp,
721 							  &startp);
722 		if (error)
723 			goto bad;
724 		pos = (Elf_Addr)startp;
725 	}
726 
727 #if defined(PAX_MPROTECT) || defined(PAX_SEGVGUARD) || defined(PAX_ASLR)
728 	l->l_proc->p_pax = epp->ep_pax_flags;
729 #endif /* PAX_MPROTECT || PAX_SEGVGUARD || PAX_ASLR */
730 
731 	if (is_dyn)
732 		elf_placedynexec(l, epp, eh, ph);
733 
734 	/*
735 	 * Load all the necessary sections
736 	 */
737 	for (i = 0; i < eh->e_phnum; i++) {
738 		Elf_Addr addr = ELFDEFNNAME(NO_ADDR);
739 		u_long size = 0;
740 
741 		switch (ph[i].p_type) {
742 		case PT_LOAD:
743 			elf_load_psection(&epp->ep_vmcmds, epp->ep_vp,
744 			    &ph[i], &addr, &size, VMCMD_FIXED);
745 
746 			/*
747 			 * Consider this as text segment, if it is executable.
748 			 * If there is more than one text segment, pick the
749 			 * largest.
750 			 */
751 			if (ph[i].p_flags & PF_X) {
752 				if (epp->ep_taddr == ELFDEFNNAME(NO_ADDR) ||
753 				    size > epp->ep_tsize) {
754 					epp->ep_taddr = addr;
755 					epp->ep_tsize = size;
756 				}
757 				end_text = addr + size;
758 			} else {
759 				epp->ep_daddr = addr;
760 				epp->ep_dsize = size;
761 			}
762 			if (ph[i].p_offset == 0) {
763 				computed_phdr = ph[i].p_vaddr + eh->e_phoff;
764 			}
765 			break;
766 
767 		case PT_SHLIB:
768 			/* SCO has these sections. */
769 		case PT_INTERP:
770 			/* Already did this one. */
771 		case PT_DYNAMIC:
772 		case PT_NOTE:
773 			break;
774 		case PT_PHDR:
775 			/* Note address of program headers (in text segment) */
776 			phdr = ph[i].p_vaddr;
777 			break;
778 
779 		default:
780 			/*
781 			 * Not fatal; we don't need to understand everything.
782 			 */
783 			break;
784 		}
785 	}
786 
787 	if (epp->ep_vmcmds.evs_used == 0) {
788 		/* No VMCMD; there was no PT_LOAD section, or those
789 		 * sections were empty */
790 		error = ENOEXEC;
791 		goto bad;
792 	}
793 
794 	if (epp->ep_daddr == ELFDEFNNAME(NO_ADDR)) {
795 		epp->ep_daddr = round_page(end_text);
796 		epp->ep_dsize = 0;
797 	}
798 
799 	/*
800 	 * Check if we found a dynamically linked binary and arrange to load
801 	 * its interpreter
802 	 */
803 	if (interp) {
804 		u_int nused = epp->ep_vmcmds.evs_used;
805 		u_long interp_offset = 0;
806 
807 		if ((error = elf_load_interp(l, epp, interp,
808 		    &epp->ep_vmcmds, &interp_offset, &pos)) != 0) {
809 			goto bad;
810 		}
811 		if (epp->ep_vmcmds.evs_used == nused) {
812 			/* elf_load_interp() has not set up any new VMCMD */
813 			error = ENOEXEC;
814 			goto bad;
815 		}
816 
817 		ap = kmem_alloc(sizeof(*ap), KM_SLEEP);
818 		ap->arg_interp = epp->ep_vmcmds.evs_cmds[nused].ev_addr;
819 		epp->ep_entryoffset = interp_offset;
820 		epp->ep_entry = ap->arg_interp + interp_offset;
821 		PNBUF_PUT(interp);
822 	} else {
823 		epp->ep_entry = eh->e_entry;
824 		if (epp->ep_flags & EXEC_FORCEAUX) {
825 			ap = kmem_alloc(sizeof(*ap), KM_SLEEP);
826 			ap->arg_interp = (vaddr_t)NULL;
827 		} else
828 			ap = NULL;
829 	}
830 
831 	if (ap) {
832 		ap->arg_phaddr = phdr ? phdr : computed_phdr;
833 		ap->arg_phentsize = eh->e_phentsize;
834 		ap->arg_phnum = eh->e_phnum;
835 		ap->arg_entry = eh->e_entry;
836 		epp->ep_emul_arg = ap;
837 		epp->ep_emul_arg_free = elf_free_emul_arg;
838 	}
839 
840 #ifdef ELF_MAP_PAGE_ZERO
841 	/* Dell SVR4 maps page zero, yeuch! */
842 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, PAGE_SIZE, 0,
843 	    epp->ep_vp, 0, VM_PROT_READ);
844 #endif
845 	kmem_free(ph, phsize);
846 	return (*epp->ep_esch->es_setup_stack)(l, epp);
847 
848 bad:
849 	if (interp)
850 		PNBUF_PUT(interp);
851 	exec_free_emul_arg(epp);
852 	kmem_free(ph, phsize);
853 	kill_vmcmds(&epp->ep_vmcmds);
854 	return error;
855 }
856 
857 int
858 netbsd_elf_signature(struct lwp *l, struct exec_package *epp,
859     Elf_Ehdr *eh)
860 {
861 	size_t i;
862 	Elf_Shdr *sh;
863 	Elf_Nhdr *np;
864 	size_t shsize, nsize;
865 	int error;
866 	int isnetbsd = 0;
867 	char *ndata, *ndesc;
868 
869 #ifdef DIAGNOSTIC
870 	const char *badnote;
871 #define BADNOTE(n) badnote = (n)
872 #else
873 #define BADNOTE(n)
874 #endif
875 
876 	epp->ep_pax_flags = 0;
877 	if (eh->e_shnum > ELF_MAXSHNUM || eh->e_shnum == 0)
878 		return ENOEXEC;
879 
880 	shsize = eh->e_shnum * sizeof(Elf_Shdr);
881 	sh = kmem_alloc(shsize, KM_SLEEP);
882 	error = exec_read_from(l, epp->ep_vp, eh->e_shoff, sh, shsize);
883 	if (error)
884 		goto out;
885 
886 	np = kmem_alloc(ELF_MAXNOTESIZE, KM_SLEEP);
887 	for (i = 0; i < eh->e_shnum; i++) {
888 		Elf_Shdr *shp = &sh[i];
889 
890 		if (shp->sh_type != SHT_NOTE ||
891 		    shp->sh_size > ELF_MAXNOTESIZE ||
892 		    shp->sh_size < sizeof(Elf_Nhdr) + ELF_NOTE_NETBSD_NAMESZ)
893 			continue;
894 
895 		error = exec_read_from(l, epp->ep_vp, shp->sh_offset, np,
896 		    shp->sh_size);
897 		if (error)
898 			continue;
899 
900 		/* Point to the note, skip the header */
901 		ndata = (char *)(np + 1);
902 
903 		/*
904 		 * Padding is present if necessary to ensure 4-byte alignment.
905 		 * The actual section size is therefore:
906 		 *    header size + 4-byte aligned name + 4-byte aligned desc
907 		 * Ensure this size is consistent with what is indicated
908 		 * in sh_size. The first check avoids integer overflows.
909 		 *
910 		 * Binaries from before NetBSD 1.6 have two notes in the same
911 		 * note section.  The second note was never used, so as long as
912 		 * the section is at least as big as it should be, it's ok.
913 		 * These binaries also have a second note section with a note of
914 		 * type ELF_NOTE_TYPE_NETBSD_TAG, which can be ignored as well.
915 		 */
916 		if (np->n_namesz > shp->sh_size || np->n_descsz > shp->sh_size) {
917 			BADNOTE("note size limit");
918 			goto bad;
919 		}
920 		nsize = sizeof(*np) + roundup(np->n_namesz, 4) +
921 		    roundup(np->n_descsz, 4);
922 		if (nsize > shp->sh_size) {
923 			BADNOTE("note size");
924 			goto bad;
925 		}
926 		ndesc = ndata + roundup(np->n_namesz, 4);
927 
928 		switch (np->n_type) {
929 		case ELF_NOTE_TYPE_NETBSD_TAG:
930 			/* It is us */
931 			if (np->n_namesz == ELF_NOTE_NETBSD_NAMESZ &&
932 			    np->n_descsz == ELF_NOTE_NETBSD_DESCSZ &&
933 			    memcmp(ndata, ELF_NOTE_NETBSD_NAME,
934 			    ELF_NOTE_NETBSD_NAMESZ) == 0) {
935 				memcpy(&epp->ep_osversion, ndesc,
936 				    ELF_NOTE_NETBSD_DESCSZ);
937 				isnetbsd = 1;
938 				break;
939 			}
940 
941 			/*
942 			 * Ignore SuSE tags; SuSE's n_type is the same as NetBSD's
943 			 * one.
944 			 */
945 			if (np->n_namesz == ELF_NOTE_SUSE_NAMESZ &&
946 			    memcmp(ndata, ELF_NOTE_SUSE_NAME,
947 			    ELF_NOTE_SUSE_NAMESZ) == 0)
948 				break;
949 			BADNOTE("NetBSD tag");
950 			goto bad;
951 
952 		case ELF_NOTE_TYPE_PAX_TAG:
953 			if (np->n_namesz == ELF_NOTE_PAX_NAMESZ &&
954 			    np->n_descsz == ELF_NOTE_PAX_DESCSZ &&
955 			    memcmp(ndata, ELF_NOTE_PAX_NAME,
956 			    ELF_NOTE_PAX_NAMESZ) == 0) {
957 				memcpy(&epp->ep_pax_flags, ndesc,
958 				    sizeof(epp->ep_pax_flags));
959 				break;
960 			}
961 			BADNOTE("PaX tag");
962 			goto bad;
963 
964 		case ELF_NOTE_TYPE_MARCH_TAG:
965 			/* Copy the machine arch into the package. */
966 			if (np->n_namesz == ELF_NOTE_MARCH_NAMESZ
967 			    && memcmp(ndata, ELF_NOTE_MARCH_NAME,
968 				    ELF_NOTE_MARCH_NAMESZ) == 0) {
969 				/* Do not truncate the buffer */
970 				if (np->n_descsz > sizeof(epp->ep_machine_arch)) {
971 					BADNOTE("description size limit");
972 					goto bad;
973 				}
974 				/*
975 				 * Ensure ndesc is NUL-terminated and of the
976 				 * expected length.
977 				 */
978 				if (strnlen(ndesc, np->n_descsz) + 1 !=
979 				    np->n_descsz) {
980 					BADNOTE("description size");
981 					goto bad;
982 				}
983 				strlcpy(epp->ep_machine_arch, ndesc,
984 				    sizeof(epp->ep_machine_arch));
985 				break;
986 			}
987 			BADNOTE("march tag");
988 			goto bad;
989 
990 		case ELF_NOTE_TYPE_MCMODEL_TAG:
991 			/* arch specific check for code model */
992 #ifdef ELF_MD_MCMODEL_CHECK
993 			if (np->n_namesz == ELF_NOTE_MCMODEL_NAMESZ
994 			    && memcmp(ndata, ELF_NOTE_MCMODEL_NAME,
995 				    ELF_NOTE_MCMODEL_NAMESZ) == 0) {
996 				ELF_MD_MCMODEL_CHECK(epp, ndesc, np->n_descsz);
997 				break;
998 			}
999 			BADNOTE("mcmodel tag");
1000 			goto bad;
1001 #endif
1002 			break;
1003 
1004 		case ELF_NOTE_TYPE_SUSE_VERSION_TAG:
1005 			break;
1006 
1007 		default:
1008 			BADNOTE("unknown tag");
1009 bad:
1010 #ifdef DIAGNOSTIC
1011 			/* Ignore GNU tags */
1012 			if (np->n_namesz == ELF_NOTE_GNU_NAMESZ &&
1013 			    memcmp(ndata, ELF_NOTE_GNU_NAME,
1014 			    ELF_NOTE_GNU_NAMESZ) == 0)
1015 			    break;
1016 
1017 			int ns = MIN(np->n_namesz, shp->sh_size - sizeof(*np));
1018 			printf("%s: Unknown elf note type %d (%s): "
1019 			    "[namesz=%d, descsz=%d name=%-*.*s]\n",
1020 			    epp->ep_kname, np->n_type, badnote, np->n_namesz,
1021 			    np->n_descsz, ns, ns, ndata);
1022 #endif
1023 			break;
1024 		}
1025 	}
1026 	kmem_free(np, ELF_MAXNOTESIZE);
1027 
1028 	error = isnetbsd ? 0 : ENOEXEC;
1029 out:
1030 	kmem_free(sh, shsize);
1031 	return error;
1032 }
1033 
1034 int
1035 netbsd_elf_probe(struct lwp *l, struct exec_package *epp, void *eh, char *itp,
1036     vaddr_t *pos)
1037 {
1038 	int error;
1039 
1040 	if ((error = netbsd_elf_signature(l, epp, eh)) != 0)
1041 		return error;
1042 #ifdef ELF_MD_PROBE_FUNC
1043 	if ((error = ELF_MD_PROBE_FUNC(l, epp, eh, itp, pos)) != 0)
1044 		return error;
1045 #elif defined(ELF_INTERP_NON_RELOCATABLE)
1046 	*pos = ELF_LINK_ADDR;
1047 #endif
1048 	epp->ep_flags |= EXEC_FORCEAUX;
1049 	return 0;
1050 }
1051 
1052 void
1053 elf_free_emul_arg(void *arg)
1054 {
1055 	struct elf_args *ap = arg;
1056 	KASSERT(ap != NULL);
1057 	kmem_free(ap, sizeof(*ap));
1058 }
1059