xref: /openbsd-src/sys/kern/exec_elf.c (revision 1a0bd10ffe9f658eda7b09dfc58adeac4e2daf87)
1 /*	$OpenBSD: exec_elf.c,v 1.111 2015/02/06 23:58:12 deraadt Exp $	*/
2 
3 /*
4  * Copyright (c) 1996 Per Fogelstrom
5  * All rights reserved.
6  *
7  * Copyright (c) 1994 Christos Zoulas
8  * All rights reserved.
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  * 3. The name of the author may not be used to endorse or promote products
19  *    derived from this software without specific prior written permission
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
24  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31  *
32  */
33 
34 /*
35  * Copyright (c) 2001 Wasabi Systems, Inc.
36  * All rights reserved.
37  *
38  * Written by Jason R. Thorpe for Wasabi Systems, Inc.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. All advertising materials mentioning features or use of this software
49  *    must display the following acknowledgement:
50  *	This product includes software developed for the NetBSD Project by
51  *	Wasabi Systems, Inc.
52  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
53  *    or promote products derived from this software without specific prior
54  *    written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
57  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
58  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
59  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
60  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
61  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
62  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
63  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
64  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
65  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
66  * POSSIBILITY OF SUCH DAMAGE.
67  */
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/kernel.h>
72 #include <sys/proc.h>
73 #include <sys/malloc.h>
74 #include <sys/pool.h>
75 #include <sys/mount.h>
76 #include <sys/namei.h>
77 #include <sys/vnode.h>
78 #include <sys/core.h>
79 #include <sys/exec.h>
80 #include <sys/exec_elf.h>
81 #include <sys/file.h>
82 #include <sys/ptrace.h>
83 #include <sys/syscall.h>
84 #include <sys/signalvar.h>
85 #include <sys/stat.h>
86 
87 #include <sys/mman.h>
88 
89 #include <uvm/uvm_extern.h>
90 
91 #include <machine/reg.h>
92 #include <machine/exec.h>
93 
94 #ifdef COMPAT_LINUX
95 #include <compat/linux/linux_exec.h>
96 #endif
97 
98 struct ELFNAME(probe_entry) {
99 	int (*func)(struct proc *, struct exec_package *, char *,
100 	    u_long *);
101 } ELFNAME(probes)[] = {
102 	/* XXX - bogus, shouldn't be size independent.. */
103 #ifdef COMPAT_LINUX
104 	{ linux_elf_probe },
105 #endif
106 	{ NULL }
107 };
108 
109 int ELFNAME(load_file)(struct proc *, char *, struct exec_package *,
110 	struct elf_args *, Elf_Addr *);
111 int ELFNAME(check_header)(Elf_Ehdr *);
112 int ELFNAME(read_from)(struct proc *, struct vnode *, u_long, caddr_t, int);
113 void ELFNAME(load_psection)(struct exec_vmcmd_set *, struct vnode *,
114 	Elf_Phdr *, Elf_Addr *, Elf_Addr *, int *, int);
115 int ELFNAMEEND(coredump)(struct proc *, void *);
116 
117 extern char sigcode[], esigcode[];
118 #ifdef SYSCALL_DEBUG
119 extern char *syscallnames[];
120 #endif
121 
122 /* round up and down to page boundaries. */
123 #define ELF_ROUND(a, b)		(((a) + (b) - 1) & ~((b) - 1))
124 #define ELF_TRUNC(a, b)		((a) & ~((b) - 1))
125 
126 /*
127  * We limit the number of program headers to 32, this should
128  * be a reasonable limit for ELF, the most we have seen so far is 12
129  */
130 #define ELF_MAX_VALID_PHDR 32
131 
132 /*
133  * This is the basic elf emul. elf_probe_funcs may change to other emuls.
134  */
135 struct emul ELFNAMEEND(emul) = {
136 	"native",
137 	NULL,
138 	sendsig,
139 	SYS_syscall,
140 	SYS_MAXSYSCALL,
141 	sysent,
142 #ifdef SYSCALL_DEBUG
143 	syscallnames,
144 #else
145 	NULL,
146 #endif
147 	(sizeof(AuxInfo) * ELF_AUX_ENTRIES / sizeof(char *)),
148 	ELFNAME(copyargs),
149 	setregs,
150 	ELFNAME2(exec,fixup),
151 	ELFNAMEEND(coredump),
152 	sigcode,
153 	esigcode,
154 	EMUL_ENABLED | EMUL_NATIVE,
155 };
156 
157 /*
158  * Copy arguments onto the stack in the normal way, but add some
159  * space for extra information in case of dynamic binding.
160  */
161 void *
162 ELFNAME(copyargs)(struct exec_package *pack, struct ps_strings *arginfo,
163 		void *stack, void *argp)
164 {
165 	stack = copyargs(pack, arginfo, stack, argp);
166 	if (!stack)
167 		return (NULL);
168 
169 	/*
170 	 * Push space for extra arguments on the stack needed by
171 	 * dynamically linked binaries.
172 	 */
173 	if (pack->ep_emul_arg != NULL) {
174 		pack->ep_emul_argp = stack;
175 		stack = (char *)stack + ELF_AUX_ENTRIES * sizeof (AuxInfo);
176 	}
177 	return (stack);
178 }
179 
180 /*
181  * Check header for validity; return 0 for ok, ENOEXEC if error
182  */
183 int
184 ELFNAME(check_header)(Elf_Ehdr *ehdr)
185 {
186 	/*
187 	 * We need to check magic, class size, endianess, and version before
188 	 * we look at the rest of the Elf_Ehdr structure. These few elements
189 	 * are represented in a machine independent fashion.
190 	 */
191 	if (!IS_ELF(*ehdr) ||
192 	    ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
193 	    ehdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
194 	    ehdr->e_ident[EI_VERSION] != ELF_TARG_VER)
195 		return (ENOEXEC);
196 
197 	/* Now check the machine dependent header */
198 	if (ehdr->e_machine != ELF_TARG_MACH ||
199 	    ehdr->e_version != ELF_TARG_VER)
200 		return (ENOEXEC);
201 
202 	/* Don't allow an insane amount of sections. */
203 	if (ehdr->e_phnum > ELF_MAX_VALID_PHDR)
204 		return (ENOEXEC);
205 
206 	return (0);
207 }
208 
209 /*
210  * Load a psection at the appropriate address
211  */
212 void
213 ELFNAME(load_psection)(struct exec_vmcmd_set *vcset, struct vnode *vp,
214 	Elf_Phdr *ph, Elf_Addr *addr, Elf_Addr *size, int *prot, int flags)
215 {
216 	u_long uaddr, msize, lsize, psize, rm, rf;
217 	long diff, offset, bdiff;
218 	Elf_Addr base;
219 
220 	/*
221 	 * If the user specified an address, then we load there.
222 	 */
223 	if (*addr != ELFDEFNNAME(NO_ADDR)) {
224 		if (ph->p_align > 1) {
225 			*addr = ELF_TRUNC(*addr, ph->p_align);
226 			diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align);
227 			/* page align vaddr */
228 			base = *addr + trunc_page(ph->p_vaddr)
229 			    - ELF_TRUNC(ph->p_vaddr, ph->p_align);
230 
231 			bdiff = ph->p_vaddr - trunc_page(ph->p_vaddr);
232 
233 		} else
234 			diff = 0;
235 	} else {
236 		*addr = uaddr = ph->p_vaddr;
237 		if (ph->p_align > 1)
238 			*addr = ELF_TRUNC(uaddr, ph->p_align);
239 		base = trunc_page(uaddr);
240 		bdiff = uaddr - base;
241 		diff = uaddr - *addr;
242 	}
243 
244 	*prot |= (ph->p_flags & PF_R) ? PROT_READ : 0;
245 	*prot |= (ph->p_flags & PF_W) ? PROT_WRITE : 0;
246 	*prot |= (ph->p_flags & PF_X) ? PROT_EXEC : 0;
247 
248 	msize = ph->p_memsz + diff;
249 	offset = ph->p_offset - bdiff;
250 	lsize = ph->p_filesz + bdiff;
251 	psize = round_page(lsize);
252 
253 	/*
254 	 * Because the pagedvn pager can't handle zero fill of the last
255 	 * data page if it's not page aligned we map the last page readvn.
256 	 */
257 	if (ph->p_flags & PF_W) {
258 		psize = trunc_page(lsize);
259 		if (psize > 0)
260 			NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp,
261 			    offset, *prot, flags);
262 		if (psize != lsize) {
263 			NEW_VMCMD2(vcset, vmcmd_map_readvn, lsize - psize,
264 			    base + psize, vp, offset + psize, *prot, flags);
265 		}
266 	} else {
267 		NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp, offset,
268 		    *prot, flags);
269 	}
270 
271 	/*
272 	 * Check if we need to extend the size of the segment
273 	 */
274 	rm = round_page(*addr + ph->p_memsz + diff);
275 	rf = round_page(*addr + ph->p_filesz + diff);
276 
277 	if (rm != rf) {
278 		NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP, 0,
279 		    *prot, flags);
280 	}
281 	*size = msize;
282 }
283 
284 /*
285  * Read from vnode into buffer at offset.
286  */
287 int
288 ELFNAME(read_from)(struct proc *p, struct vnode *vp, u_long off, caddr_t buf,
289 	int size)
290 {
291 	int error;
292 	size_t resid;
293 
294 	if ((error = vn_rdwr(UIO_READ, vp, buf, size, off, UIO_SYSSPACE,
295 	    0, p->p_ucred, &resid, p)) != 0)
296 		return error;
297 	/*
298 	 * See if we got all of it
299 	 */
300 	if (resid != 0)
301 		return (ENOEXEC);
302 	return (0);
303 }
304 
305 /*
306  * Load a file (interpreter/library) pointed to by path [stolen from
307  * coff_load_shlib()]. Made slightly generic so it might be used externally.
308  */
309 int
310 ELFNAME(load_file)(struct proc *p, char *path, struct exec_package *epp,
311 	struct elf_args *ap, Elf_Addr *last)
312 {
313 	int error, i;
314 	struct nameidata nd;
315 	Elf_Ehdr eh;
316 	Elf_Phdr *ph = NULL;
317 	u_long phsize;
318 	Elf_Addr addr;
319 	struct vnode *vp;
320 	Elf_Phdr *base_ph = NULL;
321 	struct interp_ld_sec {
322 		Elf_Addr vaddr;
323 		u_long memsz;
324 	} loadmap[ELF_MAX_VALID_PHDR];
325 	int nload, idx = 0;
326 	Elf_Addr pos = *last;
327 	int file_align;
328 	int loop;
329 	size_t randomizequota = ELF_RANDOMIZE_LIMIT;
330 
331 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path, p);
332 	if ((error = namei(&nd)) != 0) {
333 		return (error);
334 	}
335 	vp = nd.ni_vp;
336 	if (vp->v_type != VREG) {
337 		error = EACCES;
338 		goto bad;
339 	}
340 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
341 		goto bad;
342 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
343 		error = EACCES;
344 		goto bad;
345 	}
346 	if ((error = VOP_ACCESS(vp, VREAD, p->p_ucred, p)) != 0)
347 		goto bad1;
348 	if ((error = ELFNAME(read_from)(p, nd.ni_vp, 0,
349 				    (caddr_t)&eh, sizeof(eh))) != 0)
350 		goto bad1;
351 
352 	if (ELFNAME(check_header)(&eh) || eh.e_type != ET_DYN) {
353 		error = ENOEXEC;
354 		goto bad1;
355 	}
356 
357 	ph = mallocarray(eh.e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK);
358 	phsize = eh.e_phnum * sizeof(Elf_Phdr);
359 
360 	if ((error = ELFNAME(read_from)(p, nd.ni_vp, eh.e_phoff, (caddr_t)ph,
361 	    phsize)) != 0)
362 		goto bad1;
363 
364 	for (i = 0; i < eh.e_phnum; i++) {
365 		if (ph[i].p_type == PT_LOAD) {
366 			loadmap[idx].vaddr = trunc_page(ph[i].p_vaddr);
367 			loadmap[idx].memsz = round_page (ph[i].p_vaddr +
368 			    ph[i].p_memsz - loadmap[idx].vaddr);
369 			file_align = ph[i].p_align;
370 			idx++;
371 		}
372 	}
373 	nload = idx;
374 
375 	/*
376 	 * If no position to load the interpreter was set by a probe
377 	 * function, pick the same address that a non-fixed mmap(0, ..)
378 	 * would (i.e. something safely out of the way).
379 	 */
380 	if (pos == ELFDEFNNAME(NO_ADDR)) {
381 		pos = uvm_map_hint(p->p_vmspace, PROT_EXEC);
382 	}
383 
384 	pos = ELF_ROUND(pos, file_align);
385 	*last = epp->ep_interp_pos = pos;
386 	loop = 0;
387 	for (i = 0; i < nload;/**/) {
388 		vaddr_t	addr;
389 		struct	uvm_object *uobj;
390 		off_t	uoff;
391 		size_t	size;
392 
393 #ifdef this_needs_fixing
394 		if (i == 0) {
395 			uobj = &vp->v_uvm.u_obj;
396 			/* need to fix uoff */
397 		} else {
398 #endif
399 			uobj = NULL;
400 			uoff = 0;
401 #ifdef this_needs_fixing
402 		}
403 #endif
404 
405 		addr = trunc_page(pos + loadmap[i].vaddr);
406 		size =  round_page(addr + loadmap[i].memsz) - addr;
407 
408 		/* CRAP - map_findspace does not avoid daddr+BRKSIZ */
409 		if ((addr + size > (vaddr_t)p->p_vmspace->vm_daddr) &&
410 		    (addr < (vaddr_t)p->p_vmspace->vm_daddr + BRKSIZ))
411 			addr = round_page((vaddr_t)p->p_vmspace->vm_daddr +
412 			    BRKSIZ);
413 
414 		if (uvm_map_mquery(&p->p_vmspace->vm_map, &addr, size,
415 		    (i == 0 ? uoff : UVM_UNKNOWN_OFFSET), 0) != 0) {
416 			if (loop == 0) {
417 				loop = 1;
418 				i = 0;
419 				*last = epp->ep_interp_pos = pos = 0;
420 				continue;
421 			}
422 			error = ENOMEM;
423 			goto bad1;
424 		}
425 		if (addr != pos + loadmap[i].vaddr) {
426 			/* base changed. */
427 			pos = addr - trunc_page(loadmap[i].vaddr);
428 			pos = ELF_ROUND(pos,file_align);
429 			epp->ep_interp_pos = *last = pos;
430 			i = 0;
431 			continue;
432 		}
433 
434 		i++;
435 	}
436 
437 	/*
438 	 * Load all the necessary sections
439 	 */
440 	for (i = 0; i < eh.e_phnum; i++) {
441 		Elf_Addr size = 0;
442 		int prot = 0;
443 		int flags;
444 
445 		switch (ph[i].p_type) {
446 		case PT_LOAD:
447 			if (base_ph == NULL) {
448 				flags = VMCMD_BASE;
449 				addr = *last;
450 				base_ph = &ph[i];
451 			} else {
452 				flags = VMCMD_RELATIVE;
453 				addr = ph[i].p_vaddr - base_ph->p_vaddr;
454 			}
455 			ELFNAME(load_psection)(&epp->ep_vmcmds, nd.ni_vp,
456 			    &ph[i], &addr, &size, &prot, flags);
457 			/* If entry is within this section it must be text */
458 			if (eh.e_entry >= ph[i].p_vaddr &&
459 			    eh.e_entry < (ph[i].p_vaddr + size)) {
460  				epp->ep_entry = addr + eh.e_entry -
461 				    ELF_TRUNC(ph[i].p_vaddr,ph[i].p_align);
462 				ap->arg_interp = addr;
463 			}
464 			addr += size;
465 			break;
466 
467 		case PT_DYNAMIC:
468 		case PT_PHDR:
469 		case PT_NOTE:
470 			break;
471 
472 		case PT_OPENBSD_RANDOMIZE:
473 			if (ph[i].p_memsz > randomizequota) {
474 				error = ENOMEM;
475 				goto bad1;
476 			}
477 			randomizequota -= ph[i].p_memsz;
478 			NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize,
479 			    ph[i].p_memsz, ph[i].p_vaddr + pos, NULLVP, 0, 0);
480 			break;
481 
482 		default:
483 			break;
484 		}
485 	}
486 
487 	vn_marktext(nd.ni_vp);
488 
489 bad1:
490 	VOP_CLOSE(nd.ni_vp, FREAD, p->p_ucred, p);
491 bad:
492 	free(ph, M_TEMP, phsize);
493 
494 	*last = addr;
495 	vput(nd.ni_vp);
496 	return (error);
497 }
498 
499 /*
500  * Prepare an Elf binary's exec package
501  *
502  * First, set of the various offsets/lengths in the exec package.
503  *
504  * Then, mark the text image busy (so it can be demand paged) or error out if
505  * this is not possible.  Finally, set up vmcmds for the text, data, bss, and
506  * stack segments.
507  */
508 int
509 ELFNAME2(exec,makecmds)(struct proc *p, struct exec_package *epp)
510 {
511 	Elf_Ehdr *eh = epp->ep_hdr;
512 	Elf_Phdr *ph, *pp, *base_ph = NULL;
513 	Elf_Addr phdr = 0, exe_base = 0;
514 	int error, i, has_phdr = 0;
515 	char *interp = NULL;
516 	u_long pos = 0, phsize;
517 	size_t randomizequota = ELF_RANDOMIZE_LIMIT;
518 
519 	if (epp->ep_hdrvalid < sizeof(Elf_Ehdr))
520 		return (ENOEXEC);
521 
522 	if (ELFNAME(check_header)(eh) ||
523 	   (eh->e_type != ET_EXEC && eh->e_type != ET_DYN))
524 		return (ENOEXEC);
525 
526 	/*
527 	 * check if vnode is in open for writing, because we want to demand-
528 	 * page out of it.  if it is, don't do it, for various reasons.
529 	 */
530 	if (epp->ep_vp->v_writecount != 0) {
531 #ifdef DIAGNOSTIC
532 		if (epp->ep_vp->v_flag & VTEXT)
533 			panic("exec: a VTEXT vnode has writecount != 0");
534 #endif
535 		return (ETXTBSY);
536 	}
537 	/*
538 	 * Allocate space to hold all the program headers, and read them
539 	 * from the file
540 	 */
541 	ph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK);
542 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
543 
544 	if ((error = ELFNAME(read_from)(p, epp->ep_vp, eh->e_phoff, (caddr_t)ph,
545 	    phsize)) != 0)
546 		goto bad;
547 
548 	epp->ep_tsize = ELFDEFNNAME(NO_ADDR);
549 	epp->ep_dsize = ELFDEFNNAME(NO_ADDR);
550 
551 	for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) {
552 		if (pp->p_type == PT_INTERP && !interp) {
553 			if (pp->p_filesz >= MAXPATHLEN)
554 				goto bad;
555 			interp = pool_get(&namei_pool, PR_WAITOK);
556 			if ((error = ELFNAME(read_from)(p, epp->ep_vp,
557 			    pp->p_offset, interp, pp->p_filesz)) != 0) {
558 				goto bad;
559 			}
560 		} else if (pp->p_type == PT_LOAD) {
561 			if (base_ph == NULL)
562 				base_ph = pp;
563 		} else if (pp->p_type == PT_PHDR) {
564 			has_phdr = 1;
565 		}
566 	}
567 
568 	if (eh->e_type == ET_DYN) {
569 		/* need phdr and load sections for PIE */
570 		if (!has_phdr || base_ph == NULL) {
571 			error = EINVAL;
572 			goto bad;
573 		}
574 		/* randomize exe_base for PIE */
575 		exe_base = uvm_map_pie(base_ph->p_align);
576 	}
577 
578 	/*
579 	 * OK, we want a slightly different twist of the
580 	 * standard emulation package for "real" elf.
581 	 */
582 	epp->ep_emul = &ELFNAMEEND(emul);
583 	pos = ELFDEFNNAME(NO_ADDR);
584 
585 	/*
586 	 * On the same architecture, we may be emulating different systems.
587 	 * See which one will accept this executable.
588 	 *
589 	 * Probe functions would normally see if the interpreter (if any)
590 	 * exists. Emulation packages may possibly replace the interpreter in
591 	 * *interp with a changed path (/emul/xxx/<path>), and also
592 	 * set the ep_emul field in the exec package structure.
593 	 */
594 	error = ENOEXEC;
595 	if (eh->e_ident[EI_OSABI] != ELFOSABI_OPENBSD &&
596 	    ELFNAME(os_pt_note)(p, epp, epp->ep_hdr, "OpenBSD", 8, 4) != 0) {
597 		for (i = 0; ELFNAME(probes)[i].func != NULL && error; i++)
598 			error = (*ELFNAME(probes)[i].func)(p, epp, interp, &pos);
599 		if (error)
600 			goto bad;
601 	}
602 
603 	/*
604 	 * Load all the necessary sections
605 	 */
606 	for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) {
607 		Elf_Addr addr, size = 0;
608 		int prot = 0;
609 		int flags = 0;
610 
611 		switch (pp->p_type) {
612 		case PT_LOAD:
613 			if (exe_base != 0) {
614 				if (pp == base_ph) {
615 					flags = VMCMD_BASE;
616 					addr = exe_base;
617 				} else {
618 					flags = VMCMD_RELATIVE;
619 					addr = pp->p_vaddr - base_ph->p_vaddr;
620 				}
621 			} else
622 				addr = ELFDEFNNAME(NO_ADDR);
623 
624 			/*
625 			 * Calculates size of text and data segments
626 			 * by starting at first and going to end of last.
627 			 * 'rwx' sections are treated as data.
628 			 * this is correct for BSS_PLT, but may not be
629 			 * for DATA_PLT, is fine for TEXT_PLT.
630 			 */
631 			ELFNAME(load_psection)(&epp->ep_vmcmds, epp->ep_vp,
632 			    pp, &addr, &size, &prot, flags);
633 
634 			/*
635 			 * Update exe_base in case alignment was off.
636 			 * For PIE, addr is relative to exe_base so
637 			 * adjust it (non PIE exe_base is 0 so no change).
638 			 */
639 			if (flags == VMCMD_BASE)
640 				exe_base = addr;
641 			else
642 				addr += exe_base;
643 
644 			/*
645 			 * Decide whether it's text or data by looking
646 			 * at the protection of the section
647 			 */
648 			if (prot & PROT_WRITE) {
649 				/* data section */
650 				if (epp->ep_dsize == ELFDEFNNAME(NO_ADDR)) {
651 					epp->ep_daddr = addr;
652 					epp->ep_dsize = size;
653 				} else {
654 					if (addr < epp->ep_daddr) {
655 						epp->ep_dsize =
656 						    epp->ep_dsize +
657 						    epp->ep_daddr -
658 						    addr;
659 						epp->ep_daddr = addr;
660 					} else
661 						epp->ep_dsize = addr+size -
662 						    epp->ep_daddr;
663 				}
664 			} else if (prot & PROT_EXEC) {
665 				/* text section */
666 				if (epp->ep_tsize == ELFDEFNNAME(NO_ADDR)) {
667 					epp->ep_taddr = addr;
668 					epp->ep_tsize = size;
669 				} else {
670 					if (addr < epp->ep_taddr) {
671 						epp->ep_tsize =
672 						    epp->ep_tsize +
673 						    epp->ep_taddr -
674 						    addr;
675 						epp->ep_taddr = addr;
676 					} else
677 						epp->ep_tsize = addr+size -
678 						    epp->ep_taddr;
679 				}
680 			}
681 			break;
682 
683 		case PT_SHLIB:
684 			error = ENOEXEC;
685 			goto bad;
686 
687 		case PT_INTERP:
688 			/* Already did this one */
689 		case PT_DYNAMIC:
690 		case PT_NOTE:
691 			break;
692 
693 		case PT_PHDR:
694 			/* Note address of program headers (in text segment) */
695 			phdr = pp->p_vaddr;
696 			break;
697 
698 		case PT_OPENBSD_RANDOMIZE:
699 			if (ph[i].p_memsz > randomizequota) {
700 				error = ENOMEM;
701 				goto bad;
702 			}
703 			randomizequota -= ph[i].p_memsz;
704 			NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize,
705 			    ph[i].p_memsz, ph[i].p_vaddr + exe_base, NULLVP, 0, 0);
706 			break;
707 
708 		default:
709 			/*
710 			 * Not fatal, we don't need to understand everything
711 			 * :-)
712 			 */
713 			break;
714 		}
715 	}
716 
717 	phdr += exe_base;
718 
719 	/*
720 	 * Strangely some linux programs may have all load sections marked
721 	 * writeable, in this case, textsize is not -1, but rather 0;
722 	 */
723 	if (epp->ep_tsize == ELFDEFNNAME(NO_ADDR))
724 		epp->ep_tsize = 0;
725 	/*
726 	 * Another possibility is that it has all load sections marked
727 	 * read-only.  Fake a zero-sized data segment right after the
728 	 * text segment.
729 	 */
730 	if (epp->ep_dsize == ELFDEFNNAME(NO_ADDR)) {
731 		epp->ep_daddr = round_page(epp->ep_taddr + epp->ep_tsize);
732 		epp->ep_dsize = 0;
733 	}
734 
735 	epp->ep_interp = interp;
736 	epp->ep_entry = eh->e_entry + exe_base;
737 
738 	/*
739 	 * Check if we found a dynamically linked binary and arrange to load
740 	 * its interpreter when the exec file is released.
741 	 */
742 	if (interp || eh->e_type == ET_DYN) {
743 		struct elf_args *ap;
744 
745 		ap = malloc(sizeof(*ap), M_TEMP, M_WAITOK);
746 
747 		ap->arg_phaddr = phdr;
748 		ap->arg_phentsize = eh->e_phentsize;
749 		ap->arg_phnum = eh->e_phnum;
750 		ap->arg_entry = eh->e_entry + exe_base;
751 		ap->arg_interp = exe_base;
752 
753 		epp->ep_emul_arg = ap;
754 		epp->ep_interp_pos = pos;
755 	}
756 
757 	free(ph, M_TEMP, phsize);
758 	vn_marktext(epp->ep_vp);
759 	return (exec_setup_stack(p, epp));
760 
761 bad:
762 	if (interp)
763 		pool_put(&namei_pool, interp);
764 	free(ph, M_TEMP, phsize);
765 	kill_vmcmds(&epp->ep_vmcmds);
766 	if (error == 0)
767 		return (ENOEXEC);
768 	return (error);
769 }
770 
771 /*
772  * Phase II of load. It is now safe to load the interpreter. Info collected
773  * when loading the program is available for setup of the interpreter.
774  */
775 int
776 ELFNAME2(exec,fixup)(struct proc *p, struct exec_package *epp)
777 {
778 	char	*interp;
779 	int	error = 0;
780 	struct	elf_args *ap;
781 	AuxInfo ai[ELF_AUX_ENTRIES], *a;
782 	Elf_Addr	pos = epp->ep_interp_pos;
783 
784 	if (epp->ep_emul_arg == NULL) {
785 		return (0);
786 	}
787 
788 	interp = epp->ep_interp;
789 	ap = epp->ep_emul_arg;
790 
791 	if (interp &&
792 	    (error = ELFNAME(load_file)(p, interp, epp, ap, &pos)) != 0) {
793 		free(ap, M_TEMP, 0);
794 		pool_put(&namei_pool, interp);
795 		kill_vmcmds(&epp->ep_vmcmds);
796 		return (error);
797 	}
798 	/*
799 	 * We have to do this ourselves...
800 	 */
801 	error = exec_process_vmcmds(p, epp);
802 
803 	/*
804 	 * Push extra arguments on the stack needed by dynamically
805 	 * linked binaries
806 	 */
807 	if (error == 0) {
808 		a = ai;
809 
810 		a->au_id = AUX_phdr;
811 		a->au_v = ap->arg_phaddr;
812 		a++;
813 
814 		a->au_id = AUX_phent;
815 		a->au_v = ap->arg_phentsize;
816 		a++;
817 
818 		a->au_id = AUX_phnum;
819 		a->au_v = ap->arg_phnum;
820 		a++;
821 
822 		a->au_id = AUX_pagesz;
823 		a->au_v = PAGE_SIZE;
824 		a++;
825 
826 		a->au_id = AUX_base;
827 		a->au_v = ap->arg_interp;
828 		a++;
829 
830 		a->au_id = AUX_flags;
831 		a->au_v = 0;
832 		a++;
833 
834 		a->au_id = AUX_entry;
835 		a->au_v = ap->arg_entry;
836 		a++;
837 
838 		a->au_id = AUX_null;
839 		a->au_v = 0;
840 		a++;
841 
842 		error = copyout(ai, epp->ep_emul_argp, sizeof ai);
843 	}
844 	free(ap, M_TEMP, 0);
845 	if (interp)
846 		pool_put(&namei_pool, interp);
847 	return (error);
848 }
849 
850 /*
851  * Older ELF binaries use EI_ABIVERSION (formerly EI_BRAND) to brand
852  * executables.  Newer ELF binaries use EI_OSABI instead.
853  */
854 char *
855 ELFNAME(check_brand)(Elf_Ehdr *eh)
856 {
857 	if (eh->e_ident[EI_ABIVERSION] == '\0')
858 		return (NULL);
859 	return (&eh->e_ident[EI_ABIVERSION]);
860 }
861 
862 int
863 ELFNAME(os_pt_note)(struct proc *p, struct exec_package *epp, Elf_Ehdr *eh,
864 	char *os_name, size_t name_size, size_t desc_size)
865 {
866 	Elf_Phdr *hph, *ph;
867 	Elf_Note *np = NULL;
868 	size_t phsize;
869 	int error;
870 
871 	hph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK);
872 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
873 	if ((error = ELFNAME(read_from)(p, epp->ep_vp, eh->e_phoff,
874 	    (caddr_t)hph, phsize)) != 0)
875 		goto out1;
876 
877 	for (ph = hph;  ph < &hph[eh->e_phnum]; ph++) {
878 		if (ph->p_type != PT_NOTE ||
879 		    ph->p_filesz > 1024 ||
880 		    ph->p_filesz < sizeof(Elf_Note) + name_size)
881 			continue;
882 
883 		np = malloc(ph->p_filesz, M_TEMP, M_WAITOK);
884 		if ((error = ELFNAME(read_from)(p, epp->ep_vp, ph->p_offset,
885 		    (caddr_t)np, ph->p_filesz)) != 0)
886 			goto out2;
887 
888 #if 0
889 		if (np->type != ELF_NOTE_TYPE_OSVERSION) {
890 			free(np, M_TEMP, ph->p_filesz);
891 			np = NULL;
892 			continue;
893 		}
894 #endif
895 
896 		/* Check the name and description sizes. */
897 		if (np->namesz != name_size ||
898 		    np->descsz != desc_size)
899 			goto out3;
900 
901 		if (memcmp((np + 1), os_name, name_size))
902 			goto out3;
903 
904 		/* XXX: We could check for the specific emulation here */
905 		/* All checks succeeded. */
906 		error = 0;
907 		goto out2;
908 	}
909 
910 out3:
911 	error = ENOEXEC;
912 out2:
913 	free(np, M_TEMP, ph->p_filesz);
914 out1:
915 	free(hph, M_TEMP, phsize);
916 	return error;
917 }
918 
919 struct countsegs_state {
920 	int	npsections;
921 };
922 
923 int	ELFNAMEEND(coredump_countsegs)(struct proc *, void *,
924 	    struct uvm_coredump_state *);
925 
926 struct writesegs_state {
927 	Elf_Phdr *psections;
928 	off_t	secoff;
929 };
930 
931 int	ELFNAMEEND(coredump_writeseghdrs)(struct proc *, void *,
932 	    struct uvm_coredump_state *);
933 
934 int	ELFNAMEEND(coredump_notes)(struct proc *, void *, size_t *);
935 int	ELFNAMEEND(coredump_note)(struct proc *, void *, size_t *);
936 int	ELFNAMEEND(coredump_writenote)(struct proc *, void *, Elf_Note *,
937 	    const char *, void *);
938 
939 #define	ELFROUNDSIZE	4	/* XXX Should it be sizeof(Elf_Word)? */
940 #define	elfround(x)	roundup((x), ELFROUNDSIZE)
941 
942 int
943 ELFNAMEEND(coredump)(struct proc *p, void *cookie)
944 {
945 #ifdef SMALL_KERNEL
946 	return EPERM;
947 #else
948 	Elf_Ehdr ehdr;
949 	Elf_Phdr *psections = NULL;
950 	struct countsegs_state cs;
951 	struct writesegs_state ws;
952 	off_t notestart, secstart, offset;
953 	size_t notesize, psectionslen;
954 	int error, i;
955 
956 	/*
957 	 * We have to make a total of 3 passes across the map:
958 	 *
959 	 *	1. Count the number of map entries (the number of
960 	 *	   PT_LOAD sections).
961 	 *
962 	 *	2. Write the P-section headers.
963 	 *
964 	 *	3. Write the P-sections.
965 	 */
966 
967 	/* Pass 1: count the entries. */
968 	cs.npsections = 0;
969 	error = uvm_coredump_walkmap(p, NULL,
970 	    ELFNAMEEND(coredump_countsegs), &cs);
971 	if (error)
972 		goto out;
973 
974 	/* Count the PT_NOTE section. */
975 	cs.npsections++;
976 
977 	/* Get the size of the notes. */
978 	error = ELFNAMEEND(coredump_notes)(p, NULL, &notesize);
979 	if (error)
980 		goto out;
981 
982 	memset(&ehdr, 0, sizeof(ehdr));
983 	memcpy(ehdr.e_ident, ELFMAG, SELFMAG);
984 	ehdr.e_ident[EI_CLASS] = ELF_TARG_CLASS;
985 	ehdr.e_ident[EI_DATA] = ELF_TARG_DATA;
986 	ehdr.e_ident[EI_VERSION] = EV_CURRENT;
987 	/* XXX Should be the OSABI/ABI version of the executable. */
988 	ehdr.e_ident[EI_OSABI] = ELFOSABI_SYSV;
989 	ehdr.e_ident[EI_ABIVERSION] = 0;
990 	ehdr.e_type = ET_CORE;
991 	/* XXX This should be the e_machine of the executable. */
992 	ehdr.e_machine = ELF_TARG_MACH;
993 	ehdr.e_version = EV_CURRENT;
994 	ehdr.e_entry = 0;
995 	ehdr.e_phoff = sizeof(ehdr);
996 	ehdr.e_shoff = 0;
997 	ehdr.e_flags = 0;
998 	ehdr.e_ehsize = sizeof(ehdr);
999 	ehdr.e_phentsize = sizeof(Elf_Phdr);
1000 	ehdr.e_phnum = cs.npsections;
1001 	ehdr.e_shentsize = 0;
1002 	ehdr.e_shnum = 0;
1003 	ehdr.e_shstrndx = 0;
1004 
1005 	/* Write out the ELF header. */
1006 	error = coredump_write(cookie, UIO_SYSSPACE, &ehdr, sizeof(ehdr));
1007 	if (error)
1008 		goto out;
1009 
1010 	psections = mallocarray(cs.npsections, sizeof(Elf_Phdr),
1011 	    M_TEMP, M_WAITOK|M_ZERO);
1012 	psectionslen = cs.npsections * sizeof(Elf_Phdr);
1013 
1014 	offset = sizeof(ehdr);
1015 	notestart = offset + psectionslen;
1016 	secstart = notestart + notesize;
1017 
1018 	/* Pass 2: now write the P-section headers. */
1019 	ws.secoff = secstart;
1020 	ws.psections = psections;
1021 	error = uvm_coredump_walkmap(p, cookie,
1022 	    ELFNAMEEND(coredump_writeseghdrs), &ws);
1023 	if (error)
1024 		goto out;
1025 
1026 	/* Write out the PT_NOTE header. */
1027 	ws.psections->p_type = PT_NOTE;
1028 	ws.psections->p_offset = notestart;
1029 	ws.psections->p_vaddr = 0;
1030 	ws.psections->p_paddr = 0;
1031 	ws.psections->p_filesz = notesize;
1032 	ws.psections->p_memsz = 0;
1033 	ws.psections->p_flags = PF_R;
1034 	ws.psections->p_align = ELFROUNDSIZE;
1035 
1036 	error = coredump_write(cookie, UIO_SYSSPACE, psections, psectionslen);
1037 	if (error)
1038 		goto out;
1039 
1040 #ifdef DIAGNOSTIC
1041 	offset += psectionslen;
1042 	if (offset != notestart)
1043 		panic("coredump: offset %lld != notestart %lld",
1044 		    (long long) offset, (long long) notestart);
1045 #endif
1046 
1047 	/* Write out the notes. */
1048 	error = ELFNAMEEND(coredump_notes)(p, cookie, &notesize);
1049 	if (error)
1050 		goto out;
1051 
1052 #ifdef DIAGNOSTIC
1053 	offset += notesize;
1054 	if (offset != secstart)
1055 		panic("coredump: offset %lld != secstart %lld",
1056 		    (long long) offset, (long long) secstart);
1057 #endif
1058 
1059 	/* Pass 3: finally, write the sections themselves. */
1060 	for (i = 0; i < cs.npsections - 1; i++) {
1061 		if (psections[i].p_filesz == 0)
1062 			continue;
1063 
1064 #ifdef DIAGNOSTIC
1065 		if (offset != psections[i].p_offset)
1066 			panic("coredump: offset %lld != p_offset[%d] %lld",
1067 			    (long long) offset, i,
1068 			    (long long) psections[i].p_filesz);
1069 #endif
1070 
1071 		error = coredump_write(cookie, UIO_USERSPACE,
1072 		    (void *)(vaddr_t)psections[i].p_vaddr,
1073 		    psections[i].p_filesz);
1074 		if (error)
1075 			goto out;
1076 
1077 		coredump_unmap(cookie, (vaddr_t)psections[i].p_vaddr,
1078 		    (vaddr_t)psections[i].p_vaddr + psections[i].p_filesz);
1079 
1080 #ifdef DIAGNOSTIC
1081 		offset += psections[i].p_filesz;
1082 #endif
1083 	}
1084 
1085 out:
1086 	free(psections, M_TEMP, psectionslen);
1087 	return (error);
1088 #endif
1089 }
1090 
1091 int
1092 ELFNAMEEND(coredump_countsegs)(struct proc *p, void *iocookie,
1093     struct uvm_coredump_state *us)
1094 {
1095 #ifndef SMALL_KERNEL
1096 	struct countsegs_state *cs = us->cookie;
1097 
1098 	cs->npsections++;
1099 #endif
1100 	return (0);
1101 }
1102 
1103 int
1104 ELFNAMEEND(coredump_writeseghdrs)(struct proc *p, void *iocookie,
1105     struct uvm_coredump_state *us)
1106 {
1107 #ifndef SMALL_KERNEL
1108 	struct writesegs_state *ws = us->cookie;
1109 	Elf_Phdr phdr;
1110 	vsize_t size, realsize;
1111 
1112 	size = us->end - us->start;
1113 	realsize = us->realend - us->start;
1114 
1115 	phdr.p_type = PT_LOAD;
1116 	phdr.p_offset = ws->secoff;
1117 	phdr.p_vaddr = us->start;
1118 	phdr.p_paddr = 0;
1119 	phdr.p_filesz = realsize;
1120 	phdr.p_memsz = size;
1121 	phdr.p_flags = 0;
1122 	if (us->prot & PROT_READ)
1123 		phdr.p_flags |= PF_R;
1124 	if (us->prot & PROT_WRITE)
1125 		phdr.p_flags |= PF_W;
1126 	if (us->prot & PROT_EXEC)
1127 		phdr.p_flags |= PF_X;
1128 	phdr.p_align = PAGE_SIZE;
1129 
1130 	ws->secoff += phdr.p_filesz;
1131 	*ws->psections++ = phdr;
1132 #endif
1133 
1134 	return (0);
1135 }
1136 
1137 int
1138 ELFNAMEEND(coredump_notes)(struct proc *p, void *iocookie, size_t *sizep)
1139 {
1140 #ifndef SMALL_KERNEL
1141 	struct ps_strings pss;
1142 	struct iovec iov;
1143 	struct uio uio;
1144 	struct elfcore_procinfo cpi;
1145 	Elf_Note nhdr;
1146 	struct process *pr = p->p_p;
1147 	struct proc *q;
1148 	size_t size, notesize;
1149 	int error;
1150 
1151 	size = 0;
1152 
1153 	/* First, write an elfcore_procinfo. */
1154 	notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) +
1155 	    elfround(sizeof(cpi));
1156 	if (iocookie) {
1157 		memset(&cpi, 0, sizeof(cpi));
1158 
1159 		cpi.cpi_version = ELFCORE_PROCINFO_VERSION;
1160 		cpi.cpi_cpisize = sizeof(cpi);
1161 		cpi.cpi_signo = p->p_sisig;
1162 		cpi.cpi_sigcode = p->p_sicode;
1163 
1164 		cpi.cpi_sigpend = p->p_siglist;
1165 		cpi.cpi_sigmask = p->p_sigmask;
1166 		cpi.cpi_sigignore = pr->ps_sigacts->ps_sigignore;
1167 		cpi.cpi_sigcatch = pr->ps_sigacts->ps_sigcatch;
1168 
1169 		cpi.cpi_pid = pr->ps_pid;
1170 		cpi.cpi_ppid = pr->ps_pptr->ps_pid;
1171 		cpi.cpi_pgrp = pr->ps_pgid;
1172 		if (pr->ps_session->s_leader)
1173 			cpi.cpi_sid = pr->ps_session->s_leader->ps_pid;
1174 		else
1175 			cpi.cpi_sid = 0;
1176 
1177 		cpi.cpi_ruid = p->p_ucred->cr_ruid;
1178 		cpi.cpi_euid = p->p_ucred->cr_uid;
1179 		cpi.cpi_svuid = p->p_ucred->cr_svuid;
1180 
1181 		cpi.cpi_rgid = p->p_ucred->cr_rgid;
1182 		cpi.cpi_egid = p->p_ucred->cr_gid;
1183 		cpi.cpi_svgid = p->p_ucred->cr_svgid;
1184 
1185 		(void)strlcpy(cpi.cpi_name, p->p_comm, sizeof(cpi.cpi_name));
1186 
1187 		nhdr.namesz = sizeof("OpenBSD");
1188 		nhdr.descsz = sizeof(cpi);
1189 		nhdr.type = NT_OPENBSD_PROCINFO;
1190 
1191 		error = ELFNAMEEND(coredump_writenote)(p, iocookie, &nhdr,
1192 		    "OpenBSD", &cpi);
1193 		if (error)
1194 			return (error);
1195 	}
1196 	size += notesize;
1197 
1198 	/* Second, write an NT_OPENBSD_AUXV note. */
1199 	notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) +
1200 	    elfround(pr->ps_emul->e_arglen * sizeof(char *));
1201 	if (iocookie) {
1202 		iov.iov_base = &pss;
1203 		iov.iov_len = sizeof(pss);
1204 		uio.uio_iov = &iov;
1205 		uio.uio_iovcnt = 1;
1206 		uio.uio_offset = (off_t)pr->ps_strings;
1207 		uio.uio_resid = sizeof(pss);
1208 		uio.uio_segflg = UIO_SYSSPACE;
1209 		uio.uio_rw = UIO_READ;
1210 		uio.uio_procp = NULL;
1211 
1212 		error = uvm_io(&p->p_vmspace->vm_map, &uio, 0);
1213 		if (error)
1214 			return (error);
1215 
1216 		if (pss.ps_envstr == NULL)
1217 			return (EIO);
1218 
1219 		nhdr.namesz = sizeof("OpenBSD");
1220 		nhdr.descsz = pr->ps_emul->e_arglen * sizeof(char *);
1221 		nhdr.type = NT_OPENBSD_AUXV;
1222 
1223 		error = coredump_write(iocookie, UIO_SYSSPACE,
1224 		    &nhdr, sizeof(nhdr));
1225 		if (error)
1226 			return (error);
1227 
1228 		error = coredump_write(iocookie, UIO_SYSSPACE,
1229 		    "OpenBSD", elfround(nhdr.namesz));
1230 		if (error)
1231 			return (error);
1232 
1233 		error = coredump_write(iocookie, UIO_USERSPACE,
1234 		    pss.ps_envstr + pss.ps_nenvstr + 1, nhdr.descsz);
1235 		if (error)
1236 			return (error);
1237 	}
1238 	size += notesize;
1239 
1240 #ifdef PT_WCOOKIE
1241 	notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) +
1242 	    elfround(sizeof(register_t));
1243 	if (iocookie) {
1244 		register_t wcookie;
1245 
1246 		nhdr.namesz = sizeof("OpenBSD");
1247 		nhdr.descsz = sizeof(register_t);
1248 		nhdr.type = NT_OPENBSD_WCOOKIE;
1249 
1250 		wcookie = process_get_wcookie(p);
1251 		error = ELFNAMEEND(coredump_writenote)(p, iocookie, &nhdr,
1252 		    "OpenBSD", &wcookie);
1253 		if (error)
1254 			return (error);
1255 	}
1256 	size += notesize;
1257 #endif
1258 
1259 	/*
1260 	 * Now write the register info for the thread that caused the
1261 	 * coredump.
1262 	 */
1263 	error = ELFNAMEEND(coredump_note)(p, iocookie, &notesize);
1264 	if (error)
1265 		return (error);
1266 	size += notesize;
1267 
1268 	/*
1269 	 * Now, for each thread, write the register info and any other
1270 	 * per-thread notes.  Since we're dumping core, all the other
1271 	 * threads in the process have been stopped and the list can't
1272 	 * change.
1273 	 */
1274 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
1275 		if (q == p)		/* we've taken care of this thread */
1276 			continue;
1277 		error = ELFNAMEEND(coredump_note)(q, iocookie, &notesize);
1278 		if (error)
1279 			return (error);
1280 		size += notesize;
1281 	}
1282 
1283 	*sizep = size;
1284 #endif
1285 	return (0);
1286 }
1287 
1288 int
1289 ELFNAMEEND(coredump_note)(struct proc *p, void *iocookie, size_t *sizep)
1290 {
1291 #ifndef SMALL_KERNEL
1292 	Elf_Note nhdr;
1293 	int size, notesize, error;
1294 	int namesize;
1295 	char name[64+ELFROUNDSIZE];
1296 	struct reg intreg;
1297 #ifdef PT_GETFPREGS
1298 	struct fpreg freg;
1299 #endif
1300 
1301 	size = 0;
1302 
1303 	snprintf(name, sizeof(name)-ELFROUNDSIZE, "%s@%d",
1304 	    "OpenBSD", p->p_pid);
1305 	namesize = strlen(name) + 1;
1306 	memset(name + namesize, 0, elfround(namesize) - namesize);
1307 
1308 	notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(intreg));
1309 	if (iocookie) {
1310 		error = process_read_regs(p, &intreg);
1311 		if (error)
1312 			return (error);
1313 
1314 		nhdr.namesz = namesize;
1315 		nhdr.descsz = sizeof(intreg);
1316 		nhdr.type = NT_OPENBSD_REGS;
1317 
1318 		error = ELFNAMEEND(coredump_writenote)(p, iocookie, &nhdr,
1319 		    name, &intreg);
1320 		if (error)
1321 			return (error);
1322 
1323 	}
1324 	size += notesize;
1325 
1326 #ifdef PT_GETFPREGS
1327 	notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(freg));
1328 	if (iocookie) {
1329 		error = process_read_fpregs(p, &freg);
1330 		if (error)
1331 			return (error);
1332 
1333 		nhdr.namesz = namesize;
1334 		nhdr.descsz = sizeof(freg);
1335 		nhdr.type = NT_OPENBSD_FPREGS;
1336 
1337 		error = ELFNAMEEND(coredump_writenote)(p, iocookie, &nhdr,
1338 		    name, &freg);
1339 		if (error)
1340 			return (error);
1341 	}
1342 	size += notesize;
1343 #endif
1344 
1345 	*sizep = size;
1346 	/* XXX Add hook for machdep per-LWP notes. */
1347 #endif
1348 	return (0);
1349 }
1350 
1351 int
1352 ELFNAMEEND(coredump_writenote)(struct proc *p, void *cookie, Elf_Note *nhdr,
1353     const char *name, void *data)
1354 {
1355 #ifdef SMALL_KERNEL
1356 	return EPERM;
1357 #else
1358 	int error;
1359 
1360 	error = coredump_write(cookie, UIO_SYSSPACE, nhdr, sizeof(*nhdr));
1361 	if (error)
1362 		return error;
1363 
1364 	error = coredump_write(cookie, UIO_SYSSPACE, name,
1365 	    elfround(nhdr->namesz));
1366 	if (error)
1367 		return error;
1368 
1369 	return coredump_write(cookie, UIO_SYSSPACE, data, nhdr->descsz);
1370 #endif
1371 }
1372