xref: /openbsd-src/sys/kern/exec_elf.c (revision 4b70baf6e17fc8b27fc1f7fa7929335753fa94c3)
1 /*	$OpenBSD: exec_elf.c,v 1.148 2019/04/20 23:11:20 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/syslog.h>
80 #include <sys/exec.h>
81 #include <sys/exec_elf.h>
82 #include <sys/fcntl.h>
83 #include <sys/ptrace.h>
84 #include <sys/syscall.h>
85 #include <sys/signalvar.h>
86 #include <sys/stat.h>
87 #include <sys/pledge.h>
88 
89 #include <sys/mman.h>
90 
91 #include <uvm/uvm_extern.h>
92 
93 #include <machine/reg.h>
94 #include <machine/exec.h>
95 
96 int	elf_load_file(struct proc *, char *, struct exec_package *,
97 	    struct elf_args *);
98 int	elf_check_header(Elf_Ehdr *);
99 int	elf_read_from(struct proc *, struct vnode *, u_long, void *, int);
100 void	elf_load_psection(struct exec_vmcmd_set *, struct vnode *,
101 	    Elf_Phdr *, Elf_Addr *, Elf_Addr *, int *, int);
102 int	coredump_elf(struct proc *, void *);
103 void	*elf_copyargs(struct exec_package *, struct ps_strings *, void *,
104 	    void *);
105 int	exec_elf_fixup(struct proc *, struct exec_package *);
106 int	elf_os_pt_note(struct proc *, struct exec_package *, Elf_Ehdr *,
107 	    char *, size_t, size_t);
108 
109 extern char sigcode[], esigcode[], sigcoderet[];
110 #ifdef SYSCALL_DEBUG
111 extern char *syscallnames[];
112 #endif
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 /*
119  * We limit the number of program headers to 32, this should
120  * be a reasonable limit for ELF, the most we have seen so far is 12
121  */
122 #define ELF_MAX_VALID_PHDR 32
123 
124 /*
125  * How many entries are in the AuxInfo array we pass to the process?
126  */
127 #define ELF_AUX_ENTRIES	8
128 
129 /*
130  * This is the OpenBSD ELF emul
131  */
132 struct emul emul_elf = {
133 	"native",
134 	NULL,
135 	SYS_syscall,
136 	SYS_MAXSYSCALL,
137 	sysent,
138 #ifdef SYSCALL_DEBUG
139 	syscallnames,
140 #else
141 	NULL,
142 #endif
143 	(sizeof(AuxInfo) * ELF_AUX_ENTRIES / sizeof(char *)),
144 	elf_copyargs,
145 	setregs,
146 	exec_elf_fixup,
147 	coredump_elf,
148 	sigcode,
149 	esigcode,
150 	sigcoderet
151 };
152 
153 /*
154  * Copy arguments onto the stack in the normal way, but add some
155  * space for extra information in case of dynamic binding.
156  */
157 void *
158 elf_copyargs(struct exec_package *pack, struct ps_strings *arginfo,
159 		void *stack, void *argp)
160 {
161 	stack = copyargs(pack, arginfo, stack, argp);
162 	if (!stack)
163 		return (NULL);
164 
165 	/*
166 	 * Push space for extra arguments on the stack needed by
167 	 * dynamically linked binaries.
168 	 */
169 	if (pack->ep_emul_arg != NULL) {
170 		pack->ep_emul_argp = stack;
171 		stack = (char *)stack + ELF_AUX_ENTRIES * sizeof (AuxInfo);
172 	}
173 	return (stack);
174 }
175 
176 /*
177  * Check header for validity; return 0 for ok, ENOEXEC if error
178  */
179 int
180 elf_check_header(Elf_Ehdr *ehdr)
181 {
182 	/*
183 	 * We need to check magic, class size, endianess, and version before
184 	 * we look at the rest of the Elf_Ehdr structure. These few elements
185 	 * are represented in a machine independent fashion.
186 	 */
187 	if (!IS_ELF(*ehdr) ||
188 	    ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
189 	    ehdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
190 	    ehdr->e_ident[EI_VERSION] != ELF_TARG_VER)
191 		return (ENOEXEC);
192 
193 	/* Now check the machine dependent header */
194 	if (ehdr->e_machine != ELF_TARG_MACH ||
195 	    ehdr->e_version != ELF_TARG_VER)
196 		return (ENOEXEC);
197 
198 	/* Don't allow an insane amount of sections. */
199 	if (ehdr->e_phnum > ELF_MAX_VALID_PHDR)
200 		return (ENOEXEC);
201 
202 	return (0);
203 }
204 
205 /*
206  * Load a psection at the appropriate address
207  */
208 void
209 elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp,
210     Elf_Phdr *ph, Elf_Addr *addr, Elf_Addr *size, int *prot, int flags)
211 {
212 	u_long msize, lsize, psize, rm, rf;
213 	long diff, offset, bdiff;
214 	Elf_Addr base;
215 
216 	/*
217 	 * If the user specified an address, then we load there.
218 	 */
219 	if (*addr != ELF_NO_ADDR) {
220 		if (ph->p_align > 1) {
221 			*addr = ELF_TRUNC(*addr, ph->p_align);
222 			diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align);
223 			/* page align vaddr */
224 			base = *addr + trunc_page(ph->p_vaddr)
225 			    - ELF_TRUNC(ph->p_vaddr, ph->p_align);
226 		} else {
227 			diff = 0;
228 			base = *addr + trunc_page(ph->p_vaddr) - ph->p_vaddr;
229 		}
230 	} else {
231 		*addr = ph->p_vaddr;
232 		if (ph->p_align > 1)
233 			*addr = ELF_TRUNC(*addr, ph->p_align);
234 		base = trunc_page(ph->p_vaddr);
235 		diff = ph->p_vaddr - *addr;
236 	}
237 	bdiff = ph->p_vaddr - trunc_page(ph->p_vaddr);
238 
239 	/*
240 	 * Enforce W^X and map W|X segments without X permission
241 	 * initially.  The dynamic linker will make these read-only
242 	 * and add back X permission after relocation processing.
243 	 * Static executables with W|X segments will probably crash.
244 	 */
245 	*prot |= (ph->p_flags & PF_R) ? PROT_READ : 0;
246 	*prot |= (ph->p_flags & PF_W) ? PROT_WRITE : 0;
247 	if ((ph->p_flags & PF_W) == 0)
248 		*prot |= (ph->p_flags & PF_X) ? PROT_EXEC : 0;
249 
250 	msize = ph->p_memsz + diff;
251 	offset = ph->p_offset - bdiff;
252 	lsize = ph->p_filesz + bdiff;
253 	psize = round_page(lsize);
254 
255 	/*
256 	 * Because the pagedvn pager can't handle zero fill of the last
257 	 * data page if it's not page aligned we map the last page readvn.
258 	 */
259 	if (ph->p_flags & PF_W) {
260 		psize = trunc_page(lsize);
261 		if (psize > 0)
262 			NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp,
263 			    offset, *prot, flags);
264 		if (psize != lsize) {
265 			NEW_VMCMD2(vcset, vmcmd_map_readvn, lsize - psize,
266 			    base + psize, vp, offset + psize, *prot, flags);
267 		}
268 	} else {
269 		NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp, offset,
270 		    *prot, flags);
271 	}
272 
273 	/*
274 	 * Check if we need to extend the size of the segment
275 	 */
276 	rm = round_page(*addr + ph->p_memsz + diff);
277 	rf = round_page(*addr + ph->p_filesz + diff);
278 
279 	if (rm != rf) {
280 		NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP, 0,
281 		    *prot, flags);
282 	}
283 	*size = msize;
284 }
285 
286 /*
287  * Read from vnode into buffer at offset.
288  */
289 int
290 elf_read_from(struct proc *p, struct vnode *vp, u_long off, void *buf,
291     int size)
292 {
293 	int error;
294 	size_t resid;
295 
296 	if ((error = vn_rdwr(UIO_READ, vp, buf, size, off, UIO_SYSSPACE,
297 	    0, p->p_ucred, &resid, p)) != 0)
298 		return error;
299 	/*
300 	 * See if we got all of it
301 	 */
302 	if (resid != 0)
303 		return (ENOEXEC);
304 	return (0);
305 }
306 
307 /*
308  * Load a file (interpreter/library) pointed to by path [stolen from
309  * coff_load_shlib()]. Made slightly generic so it might be used externally.
310  */
311 int
312 elf_load_file(struct proc *p, char *path, struct exec_package *epp,
313     struct elf_args *ap)
314 {
315 	int error, i;
316 	struct nameidata nd;
317 	Elf_Ehdr eh;
318 	Elf_Phdr *ph = NULL;
319 	u_long phsize = 0;
320 	Elf_Addr addr;
321 	struct vnode *vp;
322 	Elf_Phdr *base_ph = NULL;
323 	struct interp_ld_sec {
324 		Elf_Addr vaddr;
325 		u_long memsz;
326 	} loadmap[ELF_MAX_VALID_PHDR];
327 	int nload, idx = 0;
328 	Elf_Addr pos;
329 	int file_align;
330 	int loop;
331 	size_t randomizequota = ELF_RANDOMIZE_LIMIT;
332 
333 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path, p);
334 	nd.ni_pledge = PLEDGE_RPATH;
335 	nd.ni_unveil = UNVEIL_READ;
336 	if ((error = namei(&nd)) != 0) {
337 		return (error);
338 	}
339 	vp = nd.ni_vp;
340 	if (vp->v_type != VREG) {
341 		error = EACCES;
342 		goto bad;
343 	}
344 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
345 		goto bad;
346 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
347 		error = EACCES;
348 		goto bad;
349 	}
350 	if ((error = VOP_ACCESS(vp, VREAD, p->p_ucred, p)) != 0)
351 		goto bad1;
352 	if ((error = elf_read_from(p, nd.ni_vp, 0, &eh, sizeof(eh))) != 0)
353 		goto bad1;
354 
355 	if (elf_check_header(&eh) || eh.e_type != ET_DYN) {
356 		error = ENOEXEC;
357 		goto bad1;
358 	}
359 
360 	ph = mallocarray(eh.e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK);
361 	phsize = eh.e_phnum * sizeof(Elf_Phdr);
362 
363 	if ((error = elf_read_from(p, nd.ni_vp, eh.e_phoff, ph, phsize)) != 0)
364 		goto bad1;
365 
366 	for (i = 0; i < eh.e_phnum; i++) {
367 		if (ph[i].p_type == PT_LOAD) {
368 			if (ph[i].p_filesz > ph[i].p_memsz ||
369 			    ph[i].p_memsz == 0) {
370 				error = EINVAL;
371 				goto bad1;
372 			}
373 			loadmap[idx].vaddr = trunc_page(ph[i].p_vaddr);
374 			loadmap[idx].memsz = round_page (ph[i].p_vaddr +
375 			    ph[i].p_memsz - loadmap[idx].vaddr);
376 			file_align = ph[i].p_align;
377 			idx++;
378 		}
379 	}
380 	nload = idx;
381 
382 	/*
383 	 * Load the interpreter where a non-fixed mmap(NULL, ...)
384 	 * would (i.e. something safely out of the way).
385 	 */
386 	pos = uvm_map_hint(p->p_vmspace, PROT_EXEC, VM_MIN_ADDRESS,
387 	    VM_MAXUSER_ADDRESS);
388 	pos = ELF_ROUND(pos, file_align);
389 
390 	loop = 0;
391 	for (i = 0; i < nload;/**/) {
392 		vaddr_t	addr;
393 		struct	uvm_object *uobj;
394 		off_t	uoff;
395 		size_t	size;
396 
397 #ifdef this_needs_fixing
398 		if (i == 0) {
399 			uobj = &vp->v_uvm.u_obj;
400 			/* need to fix uoff */
401 		} else {
402 #endif
403 			uobj = NULL;
404 			uoff = 0;
405 #ifdef this_needs_fixing
406 		}
407 #endif
408 
409 		addr = trunc_page(pos + loadmap[i].vaddr);
410 		size =  round_page(addr + loadmap[i].memsz) - addr;
411 
412 		/* CRAP - map_findspace does not avoid daddr+BRKSIZ */
413 		if ((addr + size > (vaddr_t)p->p_vmspace->vm_daddr) &&
414 		    (addr < (vaddr_t)p->p_vmspace->vm_daddr + BRKSIZ))
415 			addr = round_page((vaddr_t)p->p_vmspace->vm_daddr +
416 			    BRKSIZ);
417 
418 		if (uvm_map_mquery(&p->p_vmspace->vm_map, &addr, size,
419 		    (i == 0 ? uoff : UVM_UNKNOWN_OFFSET), 0) != 0) {
420 			if (loop == 0) {
421 				loop = 1;
422 				i = 0;
423 				pos = 0;
424 				continue;
425 			}
426 			error = ENOMEM;
427 			goto bad1;
428 		}
429 		if (addr != pos + loadmap[i].vaddr) {
430 			/* base changed. */
431 			pos = addr - trunc_page(loadmap[i].vaddr);
432 			pos = ELF_ROUND(pos,file_align);
433 			i = 0;
434 			continue;
435 		}
436 
437 		i++;
438 	}
439 
440 	/*
441 	 * Load all the necessary sections
442 	 */
443 	for (i = 0; i < eh.e_phnum; i++) {
444 		Elf_Addr size = 0;
445 		int prot = 0;
446 		int flags;
447 
448 		switch (ph[i].p_type) {
449 		case PT_LOAD:
450 			if (base_ph == NULL) {
451 				flags = VMCMD_BASE;
452 				addr = pos;
453 				base_ph = &ph[i];
454 			} else {
455 				flags = VMCMD_RELATIVE;
456 				addr = ph[i].p_vaddr - base_ph->p_vaddr;
457 			}
458 			elf_load_psection(&epp->ep_vmcmds, nd.ni_vp,
459 			    &ph[i], &addr, &size, &prot, flags);
460 			/* If entry is within this section it must be text */
461 			if (eh.e_entry >= ph[i].p_vaddr &&
462 			    eh.e_entry < (ph[i].p_vaddr + size)) {
463  				epp->ep_entry = addr + eh.e_entry -
464 				    ELF_TRUNC(ph[i].p_vaddr,ph[i].p_align);
465 				if (flags == VMCMD_RELATIVE)
466 					epp->ep_entry += pos;
467 				ap->arg_interp = pos;
468 			}
469 			addr += size;
470 			break;
471 
472 		case PT_DYNAMIC:
473 		case PT_PHDR:
474 		case PT_NOTE:
475 			break;
476 
477 		case PT_OPENBSD_RANDOMIZE:
478 			if (ph[i].p_memsz > randomizequota) {
479 				error = ENOMEM;
480 				goto bad1;
481 			}
482 			randomizequota -= ph[i].p_memsz;
483 			NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize,
484 			    ph[i].p_memsz, ph[i].p_vaddr + pos, NULLVP, 0, 0);
485 			break;
486 
487 		default:
488 			break;
489 		}
490 	}
491 
492 	vn_marktext(nd.ni_vp);
493 
494 bad1:
495 	VOP_CLOSE(nd.ni_vp, FREAD, p->p_ucred, p);
496 bad:
497 	free(ph, M_TEMP, phsize);
498 
499 	vput(nd.ni_vp);
500 	return (error);
501 }
502 
503 /*
504  * Prepare an Elf binary's exec package
505  *
506  * First, set of the various offsets/lengths in the exec package.
507  *
508  * Then, mark the text image busy (so it can be demand paged) or error out if
509  * this is not possible.  Finally, set up vmcmds for the text, data, bss, and
510  * stack segments.
511  */
512 int
513 exec_elf_makecmds(struct proc *p, struct exec_package *epp)
514 {
515 	Elf_Ehdr *eh = epp->ep_hdr;
516 	Elf_Phdr *ph, *pp, *base_ph = NULL;
517 	Elf_Addr phdr = 0, exe_base = 0;
518 	int error, i, has_phdr = 0;
519 	char *interp = NULL;
520 	u_long phsize;
521 	size_t randomizequota = ELF_RANDOMIZE_LIMIT;
522 
523 	if (epp->ep_hdrvalid < sizeof(Elf_Ehdr))
524 		return (ENOEXEC);
525 
526 	if (elf_check_header(eh) ||
527 	   (eh->e_type != ET_EXEC && eh->e_type != ET_DYN))
528 		return (ENOEXEC);
529 
530 	/*
531 	 * check if vnode is in open for writing, because we want to demand-
532 	 * page out of it.  if it is, don't do it, for various reasons.
533 	 */
534 	if (epp->ep_vp->v_writecount != 0) {
535 #ifdef DIAGNOSTIC
536 		if (epp->ep_vp->v_flag & VTEXT)
537 			panic("exec: a VTEXT vnode has writecount != 0");
538 #endif
539 		return (ETXTBSY);
540 	}
541 	/*
542 	 * Allocate space to hold all the program headers, and read them
543 	 * from the file
544 	 */
545 	ph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK);
546 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
547 
548 	if ((error = elf_read_from(p, epp->ep_vp, eh->e_phoff, ph,
549 	    phsize)) != 0)
550 		goto bad;
551 
552 	epp->ep_tsize = ELF_NO_ADDR;
553 	epp->ep_dsize = ELF_NO_ADDR;
554 
555 	for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) {
556 		if (pp->p_type == PT_INTERP && !interp) {
557 			if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN)
558 				goto bad;
559 			interp = pool_get(&namei_pool, PR_WAITOK);
560 			if ((error = elf_read_from(p, epp->ep_vp,
561 			    pp->p_offset, interp, pp->p_filesz)) != 0) {
562 				goto bad;
563 			}
564 			if (interp[pp->p_filesz - 1] != '\0')
565 				goto bad;
566 		} else if (pp->p_type == PT_LOAD) {
567 			if (pp->p_filesz > pp->p_memsz ||
568 			    pp->p_memsz == 0) {
569 				error = EINVAL;
570 				goto bad;
571 			}
572 			if (base_ph == NULL)
573 				base_ph = pp;
574 		} else if (pp->p_type == PT_PHDR) {
575 			has_phdr = 1;
576 		}
577 	}
578 
579 	if (eh->e_type == ET_DYN) {
580 		/* need phdr and load sections for PIE */
581 		if (!has_phdr || base_ph == NULL) {
582 			error = EINVAL;
583 			goto bad;
584 		}
585 		/* randomize exe_base for PIE */
586 		exe_base = uvm_map_pie(base_ph->p_align);
587 	}
588 
589 	/*
590 	 * OK, we want a slightly different twist of the
591 	 * standard emulation package for "real" elf.
592 	 */
593 	epp->ep_emul = &emul_elf;
594 
595 	/*
596 	 * Verify this is an OpenBSD executable.  If it's marked that way
597 	 * via a PT_NOTE then also check for a PT_OPENBSD_WXNEEDED segment.
598 	 */
599 	if (eh->e_ident[EI_OSABI] != ELFOSABI_OPENBSD && (error =
600 	    elf_os_pt_note(p, epp, epp->ep_hdr, "OpenBSD", 8, 4)) != 0) {
601 		goto bad;
602 	}
603 
604 	/*
605 	 * Load all the necessary sections
606 	 */
607 	for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) {
608 		Elf_Addr addr, size = 0;
609 		int prot = 0;
610 		int flags = 0;
611 
612 		switch (pp->p_type) {
613 		case PT_LOAD:
614 			if (exe_base != 0) {
615 				if (pp == base_ph) {
616 					flags = VMCMD_BASE;
617 					addr = exe_base;
618 				} else {
619 					flags = VMCMD_RELATIVE;
620 					addr = pp->p_vaddr - base_ph->p_vaddr;
621 				}
622 			} else
623 				addr = ELF_NO_ADDR;
624 
625 			/*
626 			 * Calculates size of text and data segments
627 			 * by starting at first and going to end of last.
628 			 * 'rwx' sections are treated as data.
629 			 * this is correct for BSS_PLT, but may not be
630 			 * for DATA_PLT, is fine for TEXT_PLT.
631 			 */
632 			elf_load_psection(&epp->ep_vmcmds, epp->ep_vp,
633 			    pp, &addr, &size, &prot, flags);
634 
635 			/*
636 			 * Update exe_base in case alignment was off.
637 			 * For PIE, addr is relative to exe_base so
638 			 * adjust it (non PIE exe_base is 0 so no change).
639 			 */
640 			if (flags == VMCMD_BASE)
641 				exe_base = addr;
642 			else
643 				addr += exe_base;
644 
645 			/*
646 			 * Decide whether it's text or data by looking
647 			 * at the protection of the section
648 			 */
649 			if (prot & PROT_WRITE) {
650 				/* data section */
651 				if (epp->ep_dsize == ELF_NO_ADDR) {
652 					epp->ep_daddr = addr;
653 					epp->ep_dsize = size;
654 				} else {
655 					if (addr < epp->ep_daddr) {
656 						epp->ep_dsize =
657 						    epp->ep_dsize +
658 						    epp->ep_daddr -
659 						    addr;
660 						epp->ep_daddr = addr;
661 					} else
662 						epp->ep_dsize = addr+size -
663 						    epp->ep_daddr;
664 				}
665 			} else if (prot & PROT_EXEC) {
666 				/* text section */
667 				if (epp->ep_tsize == ELF_NO_ADDR) {
668 					epp->ep_taddr = addr;
669 					epp->ep_tsize = size;
670 				} else {
671 					if (addr < epp->ep_taddr) {
672 						epp->ep_tsize =
673 						    epp->ep_tsize +
674 						    epp->ep_taddr -
675 						    addr;
676 						epp->ep_taddr = addr;
677 					} else
678 						epp->ep_tsize = addr+size -
679 						    epp->ep_taddr;
680 				}
681 			}
682 			break;
683 
684 		case PT_SHLIB:
685 			error = ENOEXEC;
686 			goto bad;
687 
688 		case PT_INTERP:
689 			/* Already did this one */
690 		case PT_DYNAMIC:
691 		case PT_NOTE:
692 			break;
693 
694 		case PT_PHDR:
695 			/* Note address of program headers (in text segment) */
696 			phdr = pp->p_vaddr;
697 			break;
698 
699 		case PT_OPENBSD_RANDOMIZE:
700 			if (ph[i].p_memsz > randomizequota) {
701 				error = ENOMEM;
702 				goto bad;
703 			}
704 			randomizequota -= ph[i].p_memsz;
705 			NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize,
706 			    ph[i].p_memsz, ph[i].p_vaddr + exe_base, NULLVP, 0, 0);
707 			break;
708 
709 		default:
710 			/*
711 			 * Not fatal, we don't need to understand everything
712 			 * :-)
713 			 */
714 			break;
715 		}
716 	}
717 
718 	phdr += exe_base;
719 
720 	/*
721 	 * Strangely some linux programs may have all load sections marked
722 	 * writeable, in this case, textsize is not -1, but rather 0;
723 	 */
724 	if (epp->ep_tsize == ELF_NO_ADDR)
725 		epp->ep_tsize = 0;
726 	/*
727 	 * Another possibility is that it has all load sections marked
728 	 * read-only.  Fake a zero-sized data segment right after the
729 	 * text segment.
730 	 */
731 	if (epp->ep_dsize == ELF_NO_ADDR) {
732 		epp->ep_daddr = round_page(epp->ep_taddr + epp->ep_tsize);
733 		epp->ep_dsize = 0;
734 	}
735 
736 	epp->ep_interp = interp;
737 	epp->ep_entry = eh->e_entry + exe_base;
738 
739 	/*
740 	 * Check if we found a dynamically linked binary and arrange to load
741 	 * its interpreter when the exec file is released.
742 	 */
743 	if (interp || eh->e_type == ET_DYN) {
744 		struct elf_args *ap;
745 
746 		ap = malloc(sizeof(*ap), M_TEMP, M_WAITOK);
747 
748 		ap->arg_phaddr = phdr;
749 		ap->arg_phentsize = eh->e_phentsize;
750 		ap->arg_phnum = eh->e_phnum;
751 		ap->arg_entry = eh->e_entry + exe_base;
752 		ap->arg_interp = exe_base;
753 
754 		epp->ep_emul_arg = ap;
755 		epp->ep_emul_argsize = sizeof *ap;
756 	}
757 
758 	free(ph, M_TEMP, phsize);
759 	vn_marktext(epp->ep_vp);
760 	return (exec_setup_stack(p, epp));
761 
762 bad:
763 	if (interp)
764 		pool_put(&namei_pool, interp);
765 	free(ph, M_TEMP, phsize);
766 	kill_vmcmds(&epp->ep_vmcmds);
767 	if (error == 0)
768 		return (ENOEXEC);
769 	return (error);
770 }
771 
772 /*
773  * Phase II of load. It is now safe to load the interpreter. Info collected
774  * when loading the program is available for setup of the interpreter.
775  */
776 int
777 exec_elf_fixup(struct proc *p, struct exec_package *epp)
778 {
779 	char	*interp;
780 	int	error = 0;
781 	struct	elf_args *ap;
782 	AuxInfo ai[ELF_AUX_ENTRIES], *a;
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 = elf_load_file(p, interp, epp, ap)) != 0) {
793 		free(ap, M_TEMP, epp->ep_emul_argsize);
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 		memset(&ai, 0, sizeof ai);
809 		a = ai;
810 
811 		a->au_id = AUX_phdr;
812 		a->au_v = ap->arg_phaddr;
813 		a++;
814 
815 		a->au_id = AUX_phent;
816 		a->au_v = ap->arg_phentsize;
817 		a++;
818 
819 		a->au_id = AUX_phnum;
820 		a->au_v = ap->arg_phnum;
821 		a++;
822 
823 		a->au_id = AUX_pagesz;
824 		a->au_v = PAGE_SIZE;
825 		a++;
826 
827 		a->au_id = AUX_base;
828 		a->au_v = ap->arg_interp;
829 		a++;
830 
831 		a->au_id = AUX_flags;
832 		a->au_v = 0;
833 		a++;
834 
835 		a->au_id = AUX_entry;
836 		a->au_v = ap->arg_entry;
837 		a++;
838 
839 		a->au_id = AUX_null;
840 		a->au_v = 0;
841 		a++;
842 
843 		error = copyout(ai, epp->ep_emul_argp, sizeof ai);
844 	}
845 	free(ap, M_TEMP, epp->ep_emul_argsize);
846 	if (interp)
847 		pool_put(&namei_pool, interp);
848 	return (error);
849 }
850 
851 int
852 elf_os_pt_note(struct proc *p, struct exec_package *epp, Elf_Ehdr *eh,
853     char *os_name, size_t name_size, size_t desc_size)
854 {
855 	char pathbuf[MAXPATHLEN];
856 	Elf_Phdr *hph, *ph;
857 	Elf_Note *np = NULL;
858 	size_t phsize;
859 	int error;
860 
861 	hph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK);
862 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
863 	if ((error = elf_read_from(p, epp->ep_vp, eh->e_phoff,
864 	    hph, phsize)) != 0)
865 		goto out1;
866 
867 	for (ph = hph;  ph < &hph[eh->e_phnum]; ph++) {
868 		if (ph->p_type == PT_OPENBSD_WXNEEDED) {
869 			int wxallowed = (epp->ep_vp->v_mount &&
870 			    (epp->ep_vp->v_mount->mnt_flag & MNT_WXALLOWED));
871 
872 			if (!wxallowed) {
873 				error = copyinstr(epp->ep_name, &pathbuf,
874 				    sizeof(pathbuf), NULL);
875 				log(LOG_NOTICE,
876 				    "%s(%d): W^X binary outside wxallowed mountpoint\n",
877 				    error ? "" : pathbuf, p->p_p->ps_pid);
878 				error = EACCES;
879 				goto out1;
880 			}
881 			epp->ep_flags |= EXEC_WXNEEDED;
882 			break;
883 		}
884 	}
885 
886 	for (ph = hph;  ph < &hph[eh->e_phnum]; ph++) {
887 		if (ph->p_type != PT_NOTE ||
888 		    ph->p_filesz > 1024 ||
889 		    ph->p_filesz < sizeof(Elf_Note) + name_size)
890 			continue;
891 
892 		np = malloc(ph->p_filesz, M_TEMP, M_WAITOK);
893 		if ((error = elf_read_from(p, epp->ep_vp, ph->p_offset,
894 		    np, ph->p_filesz)) != 0)
895 			goto out2;
896 
897 #if 0
898 		if (np->type != ELF_NOTE_TYPE_OSVERSION) {
899 			free(np, M_TEMP, ph->p_filesz);
900 			np = NULL;
901 			continue;
902 		}
903 #endif
904 
905 		/* Check the name and description sizes. */
906 		if (np->namesz != name_size ||
907 		    np->descsz != desc_size)
908 			goto out3;
909 
910 		if (memcmp((np + 1), os_name, name_size))
911 			goto out3;
912 
913 		/* XXX: We could check for the specific emulation here */
914 		/* All checks succeeded. */
915 		error = 0;
916 		goto out2;
917 	}
918 
919 out3:
920 	error = ENOEXEC;
921 out2:
922 	free(np, M_TEMP, ph->p_filesz);
923 out1:
924 	free(hph, M_TEMP, phsize);
925 	return error;
926 }
927 
928 /*
929  * Start of routines related to dumping core
930  */
931 
932 #ifdef SMALL_KERNEL
933 int
934 coredump_elf(struct proc *p, void *cookie)
935 {
936 	return EPERM;
937 }
938 #else /* !SMALL_KERNEL */
939 
940 struct writesegs_state {
941 	off_t	notestart;
942 	off_t	secstart;
943 	off_t	secoff;
944 	struct	proc *p;
945 	void	*iocookie;
946 	Elf_Phdr *psections;
947 	size_t	psectionslen;
948 	size_t	notesize;
949 	int	npsections;
950 };
951 
952 uvm_coredump_setup_cb	coredump_setup_elf;
953 uvm_coredump_walk_cb	coredump_walk_elf;
954 
955 int	coredump_notes_elf(struct proc *, void *, size_t *);
956 int	coredump_note_elf(struct proc *, void *, size_t *);
957 int	coredump_writenote_elf(struct proc *, void *, Elf_Note *,
958 	    const char *, void *);
959 
960 #define	ELFROUNDSIZE	sizeof(Elf_Word)
961 #define	elfround(x)	roundup((x), ELFROUNDSIZE)
962 
963 int
964 coredump_elf(struct proc *p, void *cookie)
965 {
966 #ifdef DIAGNOSTIC
967 	off_t offset;
968 #endif
969 	struct writesegs_state ws;
970 	size_t notesize;
971 	int error, i;
972 
973 	ws.p = p;
974 	ws.iocookie = cookie;
975 	ws.psections = NULL;
976 
977 	/*
978 	 * Walk the map to get all the segment offsets and lengths,
979 	 * write out the ELF header.
980 	 */
981 	error = uvm_coredump_walkmap(p, coredump_setup_elf,
982 	    coredump_walk_elf, &ws);
983 	if (error)
984 		goto out;
985 
986 	error = coredump_write(cookie, UIO_SYSSPACE, ws.psections,
987 	    ws.psectionslen);
988 	if (error)
989 		goto out;
990 
991 	/* Write out the notes. */
992 	error = coredump_notes_elf(p, cookie, &notesize);
993 	if (error)
994 		goto out;
995 
996 #ifdef DIAGNOSTIC
997 	if (notesize != ws.notesize)
998 		panic("coredump: notesize changed: %zu != %zu",
999 		    ws.notesize, notesize);
1000 	offset = ws.notestart + notesize;
1001 	if (offset != ws.secstart)
1002 		panic("coredump: offset %lld != secstart %lld",
1003 		    (long long) offset, (long long) ws.secstart);
1004 #endif
1005 
1006 	/* Pass 3: finally, write the sections themselves. */
1007 	for (i = 0; i < ws.npsections - 1; i++) {
1008 		Elf_Phdr *pent = &ws.psections[i];
1009 		if (pent->p_filesz == 0)
1010 			continue;
1011 
1012 #ifdef DIAGNOSTIC
1013 		if (offset != pent->p_offset)
1014 			panic("coredump: offset %lld != p_offset[%d] %lld",
1015 			    (long long) offset, i,
1016 			    (long long) pent->p_filesz);
1017 #endif
1018 
1019 		error = coredump_write(cookie, UIO_USERSPACE,
1020 		    (void *)(vaddr_t)pent->p_vaddr, pent->p_filesz);
1021 		if (error)
1022 			goto out;
1023 
1024 		coredump_unmap(cookie, (vaddr_t)pent->p_vaddr,
1025 		    (vaddr_t)pent->p_vaddr + pent->p_filesz);
1026 
1027 #ifdef DIAGNOSTIC
1028 		offset += ws.psections[i].p_filesz;
1029 #endif
1030 	}
1031 
1032 out:
1033 	free(ws.psections, M_TEMP, ws.psectionslen);
1034 	return (error);
1035 }
1036 
1037 
1038 /*
1039  * Normally we lay out core files like this:
1040  *	[ELF Header] [Program headers] [Notes] [data for PT_LOAD segments]
1041  *
1042  * However, if there's >= 65535 segments then it overflows the field
1043  * in the ELF header, so the standard specifies putting a magic
1044  * number there and saving the real count in the .sh_info field of
1045  * the first *section* header...which requires generating a section
1046  * header.  To avoid confusing tools, we include an .shstrtab section
1047  * as well so all the indexes look valid.  So in this case we lay
1048  * out the core file like this:
1049  *	[ELF Header] [Section Headers] [.shstrtab] [Program headers] \
1050  *	[Notes] [data for PT_LOAD segments]
1051  *
1052  * The 'shstrtab' structure below is data for the second of the two
1053  * section headers, plus the .shstrtab itself, in one const buffer.
1054  */
1055 static const struct {
1056     Elf_Shdr	shdr;
1057     char	shstrtab[sizeof(ELF_SHSTRTAB) + 1];
1058 } shstrtab = {
1059     .shdr = {
1060 	.sh_name = 1,			/* offset in .shstrtab below */
1061 	.sh_type = SHT_STRTAB,
1062 	.sh_offset = sizeof(Elf_Ehdr) + 2*sizeof(Elf_Shdr),
1063 	.sh_size = sizeof(ELF_SHSTRTAB) + 1,
1064 	.sh_addralign = 1,
1065     },
1066     .shstrtab = "\0" ELF_SHSTRTAB,
1067 };
1068 
1069 int
1070 coredump_setup_elf(int segment_count, void *cookie)
1071 {
1072 	Elf_Ehdr ehdr;
1073 	struct writesegs_state *ws = cookie;
1074 	Elf_Phdr *note;
1075 	int error;
1076 
1077 	/* Get the count of segments, plus one for the PT_NOTE */
1078 	ws->npsections = segment_count + 1;
1079 
1080 	/* Get the size of the notes. */
1081 	error = coredump_notes_elf(ws->p, NULL, &ws->notesize);
1082 	if (error)
1083 		return error;
1084 
1085 	/* Setup the ELF header */
1086 	memset(&ehdr, 0, sizeof(ehdr));
1087 	memcpy(ehdr.e_ident, ELFMAG, SELFMAG);
1088 	ehdr.e_ident[EI_CLASS] = ELF_TARG_CLASS;
1089 	ehdr.e_ident[EI_DATA] = ELF_TARG_DATA;
1090 	ehdr.e_ident[EI_VERSION] = EV_CURRENT;
1091 	/* XXX Should be the OSABI/ABI version of the executable. */
1092 	ehdr.e_ident[EI_OSABI] = ELFOSABI_SYSV;
1093 	ehdr.e_ident[EI_ABIVERSION] = 0;
1094 	ehdr.e_type = ET_CORE;
1095 	/* XXX This should be the e_machine of the executable. */
1096 	ehdr.e_machine = ELF_TARG_MACH;
1097 	ehdr.e_version = EV_CURRENT;
1098 	ehdr.e_entry = 0;
1099 	ehdr.e_flags = 0;
1100 	ehdr.e_ehsize = sizeof(ehdr);
1101 	ehdr.e_phentsize = sizeof(Elf_Phdr);
1102 
1103 	if (ws->npsections < PN_XNUM) {
1104 		ehdr.e_phoff = sizeof(ehdr);
1105 		ehdr.e_shoff = 0;
1106 		ehdr.e_phnum = ws->npsections;
1107 		ehdr.e_shentsize = 0;
1108 		ehdr.e_shnum = 0;
1109 		ehdr.e_shstrndx = 0;
1110 	} else {
1111 		/* too many segments, use extension setup */
1112 		ehdr.e_shoff = sizeof(ehdr);
1113 		ehdr.e_phnum = PN_XNUM;
1114 		ehdr.e_shentsize = sizeof(Elf_Shdr);
1115 		ehdr.e_shnum = 2;
1116 		ehdr.e_shstrndx = 1;
1117 		ehdr.e_phoff = shstrtab.shdr.sh_offset + shstrtab.shdr.sh_size;
1118 	}
1119 
1120 	/* Write out the ELF header. */
1121 	error = coredump_write(ws->iocookie, UIO_SYSSPACE, &ehdr, sizeof(ehdr));
1122 	if (error)
1123 		return error;
1124 
1125 	/*
1126 	 * If an section header is needed to store extension info, write
1127 	 * it out after the ELF header and before the program header.
1128 	 */
1129 	if (ehdr.e_shnum != 0) {
1130 		Elf_Shdr shdr = { .sh_info = ws->npsections };
1131 		error = coredump_write(ws->iocookie, UIO_SYSSPACE, &shdr,
1132 		    sizeof shdr);
1133 		if (error)
1134 			return error;
1135 		error = coredump_write(ws->iocookie, UIO_SYSSPACE, &shstrtab,
1136 		    sizeof(shstrtab.shdr) + sizeof(shstrtab.shstrtab));
1137 		if (error)
1138 			return error;
1139 	}
1140 
1141 	/*
1142 	 * Allocate the segment header array and setup to collect
1143 	 * the section sizes and offsets
1144 	 */
1145 	ws->psections = mallocarray(ws->npsections, sizeof(Elf_Phdr),
1146 	    M_TEMP, M_WAITOK|M_ZERO);
1147 	ws->psectionslen = ws->npsections * sizeof(Elf_Phdr);
1148 
1149 	ws->notestart = ehdr.e_phoff + ws->psectionslen;
1150 	ws->secstart = ws->notestart + ws->notesize;
1151 	ws->secoff = ws->secstart;
1152 
1153 	/* Fill in the PT_NOTE segment header in the last slot */
1154 	note = &ws->psections[ws->npsections - 1];
1155 	note->p_type = PT_NOTE;
1156 	note->p_offset = ws->notestart;
1157 	note->p_vaddr = 0;
1158 	note->p_paddr = 0;
1159 	note->p_filesz = ws->notesize;
1160 	note->p_memsz = 0;
1161 	note->p_flags = PF_R;
1162 	note->p_align = ELFROUNDSIZE;
1163 
1164 	return (0);
1165 }
1166 
1167 int
1168 coredump_walk_elf(vaddr_t start, vaddr_t realend, vaddr_t end, vm_prot_t prot,
1169     int nsegment, void *cookie)
1170 {
1171 	struct writesegs_state *ws = cookie;
1172 	Elf_Phdr phdr;
1173 	vsize_t size, realsize;
1174 
1175 	size = end - start;
1176 	realsize = realend - start;
1177 
1178 	phdr.p_type = PT_LOAD;
1179 	phdr.p_offset = ws->secoff;
1180 	phdr.p_vaddr = start;
1181 	phdr.p_paddr = 0;
1182 	phdr.p_filesz = realsize;
1183 	phdr.p_memsz = size;
1184 	phdr.p_flags = 0;
1185 	if (prot & PROT_READ)
1186 		phdr.p_flags |= PF_R;
1187 	if (prot & PROT_WRITE)
1188 		phdr.p_flags |= PF_W;
1189 	if (prot & PROT_EXEC)
1190 		phdr.p_flags |= PF_X;
1191 	phdr.p_align = PAGE_SIZE;
1192 
1193 	ws->secoff += phdr.p_filesz;
1194 	ws->psections[nsegment] = phdr;
1195 
1196 	return (0);
1197 }
1198 
1199 int
1200 coredump_notes_elf(struct proc *p, void *iocookie, size_t *sizep)
1201 {
1202 	struct ps_strings pss;
1203 	struct iovec iov;
1204 	struct uio uio;
1205 	struct elfcore_procinfo cpi;
1206 	Elf_Note nhdr;
1207 	struct process *pr = p->p_p;
1208 	struct proc *q;
1209 	size_t size, notesize;
1210 	int error;
1211 
1212 	size = 0;
1213 
1214 	/* First, write an elfcore_procinfo. */
1215 	notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) +
1216 	    elfround(sizeof(cpi));
1217 	if (iocookie) {
1218 		memset(&cpi, 0, sizeof(cpi));
1219 
1220 		cpi.cpi_version = ELFCORE_PROCINFO_VERSION;
1221 		cpi.cpi_cpisize = sizeof(cpi);
1222 		cpi.cpi_signo = p->p_sisig;
1223 		cpi.cpi_sigcode = p->p_sicode;
1224 
1225 		cpi.cpi_sigpend = p->p_siglist;
1226 		cpi.cpi_sigmask = p->p_sigmask;
1227 		cpi.cpi_sigignore = pr->ps_sigacts->ps_sigignore;
1228 		cpi.cpi_sigcatch = pr->ps_sigacts->ps_sigcatch;
1229 
1230 		cpi.cpi_pid = pr->ps_pid;
1231 		cpi.cpi_ppid = pr->ps_pptr->ps_pid;
1232 		cpi.cpi_pgrp = pr->ps_pgid;
1233 		if (pr->ps_session->s_leader)
1234 			cpi.cpi_sid = pr->ps_session->s_leader->ps_pid;
1235 		else
1236 			cpi.cpi_sid = 0;
1237 
1238 		cpi.cpi_ruid = p->p_ucred->cr_ruid;
1239 		cpi.cpi_euid = p->p_ucred->cr_uid;
1240 		cpi.cpi_svuid = p->p_ucred->cr_svuid;
1241 
1242 		cpi.cpi_rgid = p->p_ucred->cr_rgid;
1243 		cpi.cpi_egid = p->p_ucred->cr_gid;
1244 		cpi.cpi_svgid = p->p_ucred->cr_svgid;
1245 
1246 		(void)strlcpy(cpi.cpi_name, pr->ps_comm, sizeof(cpi.cpi_name));
1247 
1248 		nhdr.namesz = sizeof("OpenBSD");
1249 		nhdr.descsz = sizeof(cpi);
1250 		nhdr.type = NT_OPENBSD_PROCINFO;
1251 
1252 		error = coredump_writenote_elf(p, iocookie, &nhdr,
1253 		    "OpenBSD", &cpi);
1254 		if (error)
1255 			return (error);
1256 	}
1257 	size += notesize;
1258 
1259 	/* Second, write an NT_OPENBSD_AUXV note. */
1260 	notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) +
1261 	    elfround(pr->ps_emul->e_arglen * sizeof(char *));
1262 	if (iocookie) {
1263 		iov.iov_base = &pss;
1264 		iov.iov_len = sizeof(pss);
1265 		uio.uio_iov = &iov;
1266 		uio.uio_iovcnt = 1;
1267 		uio.uio_offset = (off_t)pr->ps_strings;
1268 		uio.uio_resid = sizeof(pss);
1269 		uio.uio_segflg = UIO_SYSSPACE;
1270 		uio.uio_rw = UIO_READ;
1271 		uio.uio_procp = NULL;
1272 
1273 		error = uvm_io(&p->p_vmspace->vm_map, &uio, 0);
1274 		if (error)
1275 			return (error);
1276 
1277 		if (pss.ps_envstr == NULL)
1278 			return (EIO);
1279 
1280 		nhdr.namesz = sizeof("OpenBSD");
1281 		nhdr.descsz = pr->ps_emul->e_arglen * sizeof(char *);
1282 		nhdr.type = NT_OPENBSD_AUXV;
1283 
1284 		error = coredump_write(iocookie, UIO_SYSSPACE,
1285 		    &nhdr, sizeof(nhdr));
1286 		if (error)
1287 			return (error);
1288 
1289 		error = coredump_write(iocookie, UIO_SYSSPACE,
1290 		    "OpenBSD", elfround(nhdr.namesz));
1291 		if (error)
1292 			return (error);
1293 
1294 		error = coredump_write(iocookie, UIO_USERSPACE,
1295 		    pss.ps_envstr + pss.ps_nenvstr + 1, nhdr.descsz);
1296 		if (error)
1297 			return (error);
1298 	}
1299 	size += notesize;
1300 
1301 #ifdef PT_WCOOKIE
1302 	notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) +
1303 	    elfround(sizeof(register_t));
1304 	if (iocookie) {
1305 		register_t wcookie;
1306 
1307 		nhdr.namesz = sizeof("OpenBSD");
1308 		nhdr.descsz = sizeof(register_t);
1309 		nhdr.type = NT_OPENBSD_WCOOKIE;
1310 
1311 		wcookie = process_get_wcookie(p);
1312 		error = coredump_writenote_elf(p, iocookie, &nhdr,
1313 		    "OpenBSD", &wcookie);
1314 		if (error)
1315 			return (error);
1316 	}
1317 	size += notesize;
1318 #endif
1319 
1320 	/*
1321 	 * Now write the register info for the thread that caused the
1322 	 * coredump.
1323 	 */
1324 	error = coredump_note_elf(p, iocookie, &notesize);
1325 	if (error)
1326 		return (error);
1327 	size += notesize;
1328 
1329 	/*
1330 	 * Now, for each thread, write the register info and any other
1331 	 * per-thread notes.  Since we're dumping core, all the other
1332 	 * threads in the process have been stopped and the list can't
1333 	 * change.
1334 	 */
1335 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
1336 		if (q == p)		/* we've taken care of this thread */
1337 			continue;
1338 		error = coredump_note_elf(q, iocookie, &notesize);
1339 		if (error)
1340 			return (error);
1341 		size += notesize;
1342 	}
1343 
1344 	*sizep = size;
1345 	return (0);
1346 }
1347 
1348 int
1349 coredump_note_elf(struct proc *p, void *iocookie, size_t *sizep)
1350 {
1351 	Elf_Note nhdr;
1352 	int size, notesize, error;
1353 	int namesize;
1354 	char name[64+ELFROUNDSIZE];
1355 	struct reg intreg;
1356 #ifdef PT_GETFPREGS
1357 	struct fpreg freg;
1358 #endif
1359 
1360 	size = 0;
1361 
1362 	snprintf(name, sizeof(name)-ELFROUNDSIZE, "%s@%d",
1363 	    "OpenBSD", p->p_tid + THREAD_PID_OFFSET);
1364 	namesize = strlen(name) + 1;
1365 	memset(name + namesize, 0, elfround(namesize) - namesize);
1366 
1367 	notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(intreg));
1368 	if (iocookie) {
1369 		error = process_read_regs(p, &intreg);
1370 		if (error)
1371 			return (error);
1372 
1373 		nhdr.namesz = namesize;
1374 		nhdr.descsz = sizeof(intreg);
1375 		nhdr.type = NT_OPENBSD_REGS;
1376 
1377 		error = coredump_writenote_elf(p, iocookie, &nhdr,
1378 		    name, &intreg);
1379 		if (error)
1380 			return (error);
1381 
1382 	}
1383 	size += notesize;
1384 
1385 #ifdef PT_GETFPREGS
1386 	notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(freg));
1387 	if (iocookie) {
1388 		error = process_read_fpregs(p, &freg);
1389 		if (error)
1390 			return (error);
1391 
1392 		nhdr.namesz = namesize;
1393 		nhdr.descsz = sizeof(freg);
1394 		nhdr.type = NT_OPENBSD_FPREGS;
1395 
1396 		error = coredump_writenote_elf(p, iocookie, &nhdr, name, &freg);
1397 		if (error)
1398 			return (error);
1399 	}
1400 	size += notesize;
1401 #endif
1402 
1403 	*sizep = size;
1404 	/* XXX Add hook for machdep per-LWP notes. */
1405 	return (0);
1406 }
1407 
1408 int
1409 coredump_writenote_elf(struct proc *p, void *cookie, Elf_Note *nhdr,
1410     const char *name, void *data)
1411 {
1412 	int error;
1413 
1414 	error = coredump_write(cookie, UIO_SYSSPACE, nhdr, sizeof(*nhdr));
1415 	if (error)
1416 		return error;
1417 
1418 	error = coredump_write(cookie, UIO_SYSSPACE, name,
1419 	    elfround(nhdr->namesz));
1420 	if (error)
1421 		return error;
1422 
1423 	return coredump_write(cookie, UIO_SYSSPACE, data, nhdr->descsz);
1424 }
1425 #endif /* !SMALL_KERNEL */
1426