xref: /openbsd-src/sys/kern/exec_elf.c (revision 5a38ef86d0b61900239c7913d24a05e7b88a58f0)
1 /*	$OpenBSD: exec_elf.c,v 1.165 2021/12/09 00:26:10 guenther 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	elf_os_pt_note_name(Elf_Note *);
103 int	elf_os_pt_note(struct proc *, struct exec_package *, Elf_Ehdr *, int *);
104 
105 /* round up and down to page boundaries. */
106 #define ELF_ROUND(a, b)		(((a) + (b) - 1) & ~((b) - 1))
107 #define ELF_TRUNC(a, b)		((a) & ~((b) - 1))
108 
109 /*
110  * We limit the number of program headers to 32, this should
111  * be a reasonable limit for ELF, the most we have seen so far is 12
112  */
113 #define ELF_MAX_VALID_PHDR 32
114 
115 #define ELF_NOTE_NAME_OPENBSD	0x01
116 
117 struct elf_note_name {
118 	char *name;
119 	int id;
120 } elf_note_names[] = {
121 	{ "OpenBSD",	ELF_NOTE_NAME_OPENBSD },
122 };
123 
124 #define	ELFROUNDSIZE	sizeof(Elf_Word)
125 #define	elfround(x)	roundup((x), ELFROUNDSIZE)
126 
127 
128 /*
129  * Check header for validity; return 0 for ok, ENOEXEC if error
130  */
131 int
132 elf_check_header(Elf_Ehdr *ehdr)
133 {
134 	/*
135 	 * We need to check magic, class size, endianness, and version before
136 	 * we look at the rest of the Elf_Ehdr structure. These few elements
137 	 * are represented in a machine independent fashion.
138 	 */
139 	if (!IS_ELF(*ehdr) ||
140 	    ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
141 	    ehdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
142 	    ehdr->e_ident[EI_VERSION] != ELF_TARG_VER)
143 		return (ENOEXEC);
144 
145 	/* Now check the machine dependent header */
146 	if (ehdr->e_machine != ELF_TARG_MACH ||
147 	    ehdr->e_version != ELF_TARG_VER)
148 		return (ENOEXEC);
149 
150 	/* Don't allow an insane amount of sections. */
151 	if (ehdr->e_phnum > ELF_MAX_VALID_PHDR)
152 		return (ENOEXEC);
153 
154 	return (0);
155 }
156 
157 /*
158  * Load a psection at the appropriate address
159  */
160 void
161 elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp,
162     Elf_Phdr *ph, Elf_Addr *addr, Elf_Addr *size, int *prot, int flags)
163 {
164 	u_long msize, lsize, psize, rm, rf;
165 	long diff, offset, bdiff;
166 	Elf_Addr base;
167 
168 	/*
169 	 * If the user specified an address, then we load there.
170 	 */
171 	if (*addr != ELF_NO_ADDR) {
172 		if (ph->p_align > 1) {
173 			*addr = ELF_TRUNC(*addr, ph->p_align);
174 			diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align);
175 			/* page align vaddr */
176 			base = *addr + trunc_page(ph->p_vaddr)
177 			    - ELF_TRUNC(ph->p_vaddr, ph->p_align);
178 		} else {
179 			diff = 0;
180 			base = *addr + trunc_page(ph->p_vaddr) - ph->p_vaddr;
181 		}
182 	} else {
183 		*addr = ph->p_vaddr;
184 		if (ph->p_align > 1)
185 			*addr = ELF_TRUNC(*addr, ph->p_align);
186 		base = trunc_page(ph->p_vaddr);
187 		diff = ph->p_vaddr - *addr;
188 	}
189 	bdiff = ph->p_vaddr - trunc_page(ph->p_vaddr);
190 
191 	/*
192 	 * Enforce W^X and map W|X segments without X permission
193 	 * initially.  The dynamic linker will make these read-only
194 	 * and add back X permission after relocation processing.
195 	 * Static executables with W|X segments will probably crash.
196 	 */
197 	*prot |= (ph->p_flags & PF_R) ? PROT_READ : 0;
198 	*prot |= (ph->p_flags & PF_W) ? PROT_WRITE : 0;
199 	if ((ph->p_flags & PF_W) == 0)
200 		*prot |= (ph->p_flags & PF_X) ? PROT_EXEC : 0;
201 
202 	msize = ph->p_memsz + diff;
203 	offset = ph->p_offset - bdiff;
204 	lsize = ph->p_filesz + bdiff;
205 	psize = round_page(lsize);
206 
207 	/*
208 	 * Because the pagedvn pager can't handle zero fill of the last
209 	 * data page if it's not page aligned we map the last page readvn.
210 	 */
211 	if (ph->p_flags & PF_W) {
212 		psize = trunc_page(lsize);
213 		if (psize > 0)
214 			NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp,
215 			    offset, *prot, flags);
216 		if (psize != lsize) {
217 			NEW_VMCMD2(vcset, vmcmd_map_readvn, lsize - psize,
218 			    base + psize, vp, offset + psize, *prot, flags);
219 		}
220 	} else {
221 		NEW_VMCMD2(vcset, vmcmd_map_pagedvn, psize, base, vp, offset,
222 		    *prot, flags);
223 	}
224 
225 	/*
226 	 * Check if we need to extend the size of the segment
227 	 */
228 	rm = round_page(*addr + ph->p_memsz + diff);
229 	rf = round_page(*addr + ph->p_filesz + diff);
230 
231 	if (rm != rf) {
232 		NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP, 0,
233 		    *prot, flags);
234 	}
235 	*size = msize;
236 }
237 
238 /*
239  * Read from vnode into buffer at offset.
240  */
241 int
242 elf_read_from(struct proc *p, struct vnode *vp, u_long off, void *buf,
243     int size)
244 {
245 	int error;
246 	size_t resid;
247 
248 	if ((error = vn_rdwr(UIO_READ, vp, buf, size, off, UIO_SYSSPACE,
249 	    0, p->p_ucred, &resid, p)) != 0)
250 		return error;
251 	/*
252 	 * See if we got all of it
253 	 */
254 	if (resid != 0)
255 		return (ENOEXEC);
256 	return (0);
257 }
258 
259 /*
260  * Load a file (interpreter/library) pointed to by path [stolen from
261  * coff_load_shlib()]. Made slightly generic so it might be used externally.
262  */
263 int
264 elf_load_file(struct proc *p, char *path, struct exec_package *epp,
265     struct elf_args *ap)
266 {
267 	int error, i;
268 	struct nameidata nd;
269 	Elf_Ehdr eh;
270 	Elf_Phdr *ph = NULL;
271 	u_long phsize = 0;
272 	Elf_Addr addr;
273 	struct vnode *vp;
274 	Elf_Phdr *base_ph = NULL;
275 	struct interp_ld_sec {
276 		Elf_Addr vaddr;
277 		u_long memsz;
278 	} loadmap[ELF_MAX_VALID_PHDR];
279 	int nload, idx = 0;
280 	Elf_Addr pos;
281 	int file_align;
282 	int loop;
283 	size_t randomizequota = ELF_RANDOMIZE_LIMIT;
284 
285 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path, p);
286 	nd.ni_pledge = PLEDGE_RPATH;
287 	nd.ni_unveil = UNVEIL_READ;
288 	if ((error = namei(&nd)) != 0) {
289 		return (error);
290 	}
291 	vp = nd.ni_vp;
292 	if (vp->v_type != VREG) {
293 		error = EACCES;
294 		goto bad;
295 	}
296 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
297 		goto bad;
298 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
299 		error = EACCES;
300 		goto bad;
301 	}
302 	if ((error = VOP_ACCESS(vp, VREAD, p->p_ucred, p)) != 0)
303 		goto bad1;
304 	if ((error = elf_read_from(p, nd.ni_vp, 0, &eh, sizeof(eh))) != 0)
305 		goto bad1;
306 
307 	if (elf_check_header(&eh) || eh.e_type != ET_DYN) {
308 		error = ENOEXEC;
309 		goto bad1;
310 	}
311 
312 	ph = mallocarray(eh.e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK);
313 	phsize = eh.e_phnum * sizeof(Elf_Phdr);
314 
315 	if ((error = elf_read_from(p, nd.ni_vp, eh.e_phoff, ph, phsize)) != 0)
316 		goto bad1;
317 
318 	for (i = 0; i < eh.e_phnum; i++) {
319 		if (ph[i].p_type == PT_LOAD) {
320 			if (ph[i].p_filesz > ph[i].p_memsz ||
321 			    ph[i].p_memsz == 0) {
322 				error = EINVAL;
323 				goto bad1;
324 			}
325 			loadmap[idx].vaddr = trunc_page(ph[i].p_vaddr);
326 			loadmap[idx].memsz = round_page (ph[i].p_vaddr +
327 			    ph[i].p_memsz - loadmap[idx].vaddr);
328 			file_align = ph[i].p_align;
329 			idx++;
330 		}
331 	}
332 	nload = idx;
333 
334 	/*
335 	 * Load the interpreter where a non-fixed mmap(NULL, ...)
336 	 * would (i.e. something safely out of the way).
337 	 */
338 	pos = uvm_map_hint(p->p_vmspace, PROT_EXEC, VM_MIN_ADDRESS,
339 	    VM_MAXUSER_ADDRESS);
340 	pos = ELF_ROUND(pos, file_align);
341 
342 	loop = 0;
343 	for (i = 0; i < nload;/**/) {
344 		vaddr_t	addr;
345 		struct	uvm_object *uobj;
346 		off_t	uoff;
347 		size_t	size;
348 
349 #ifdef this_needs_fixing
350 		if (i == 0) {
351 			uobj = &vp->v_uvm.u_obj;
352 			/* need to fix uoff */
353 		} else {
354 #endif
355 			uobj = NULL;
356 			uoff = 0;
357 #ifdef this_needs_fixing
358 		}
359 #endif
360 
361 		addr = trunc_page(pos + loadmap[i].vaddr);
362 		size =  round_page(addr + loadmap[i].memsz) - addr;
363 
364 		/* CRAP - map_findspace does not avoid daddr+BRKSIZ */
365 		if ((addr + size > (vaddr_t)p->p_vmspace->vm_daddr) &&
366 		    (addr < (vaddr_t)p->p_vmspace->vm_daddr + BRKSIZ))
367 			addr = round_page((vaddr_t)p->p_vmspace->vm_daddr +
368 			    BRKSIZ);
369 
370 		if (uvm_map_mquery(&p->p_vmspace->vm_map, &addr, size,
371 		    (i == 0 ? uoff : UVM_UNKNOWN_OFFSET), 0) != 0) {
372 			if (loop == 0) {
373 				loop = 1;
374 				i = 0;
375 				pos = 0;
376 				continue;
377 			}
378 			error = ENOMEM;
379 			goto bad1;
380 		}
381 		if (addr != pos + loadmap[i].vaddr) {
382 			/* base changed. */
383 			pos = addr - trunc_page(loadmap[i].vaddr);
384 			pos = ELF_ROUND(pos,file_align);
385 			i = 0;
386 			continue;
387 		}
388 
389 		i++;
390 	}
391 
392 	/*
393 	 * Load all the necessary sections
394 	 */
395 	for (i = 0; i < eh.e_phnum; i++) {
396 		Elf_Addr size = 0;
397 		int prot = 0;
398 		int flags;
399 
400 		switch (ph[i].p_type) {
401 		case PT_LOAD:
402 			if (base_ph == NULL) {
403 				flags = VMCMD_BASE;
404 				addr = pos;
405 				base_ph = &ph[i];
406 			} else {
407 				flags = VMCMD_RELATIVE;
408 				addr = ph[i].p_vaddr - base_ph->p_vaddr;
409 			}
410 			elf_load_psection(&epp->ep_vmcmds, nd.ni_vp,
411 			    &ph[i], &addr, &size, &prot, flags | VMCMD_SYSCALL);
412 			/* If entry is within this section it must be text */
413 			if (eh.e_entry >= ph[i].p_vaddr &&
414 			    eh.e_entry < (ph[i].p_vaddr + size)) {
415  				epp->ep_entry = addr + eh.e_entry -
416 				    ELF_TRUNC(ph[i].p_vaddr,ph[i].p_align);
417 				if (flags == VMCMD_RELATIVE)
418 					epp->ep_entry += pos;
419 				ap->arg_interp = pos;
420 			}
421 			addr += size;
422 			break;
423 
424 		case PT_DYNAMIC:
425 		case PT_PHDR:
426 		case PT_NOTE:
427 			break;
428 
429 		case PT_OPENBSD_RANDOMIZE:
430 			if (ph[i].p_memsz > randomizequota) {
431 				error = ENOMEM;
432 				goto bad1;
433 			}
434 			randomizequota -= ph[i].p_memsz;
435 			NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize,
436 			    ph[i].p_memsz, ph[i].p_vaddr + pos, NULLVP, 0, 0);
437 			break;
438 
439 		default:
440 			break;
441 		}
442 	}
443 
444 	vn_marktext(nd.ni_vp);
445 
446 bad1:
447 	VOP_CLOSE(nd.ni_vp, FREAD, p->p_ucred, p);
448 bad:
449 	free(ph, M_TEMP, phsize);
450 
451 	vput(nd.ni_vp);
452 	return (error);
453 }
454 
455 /*
456  * Prepare an Elf binary's exec package
457  *
458  * First, set of the various offsets/lengths in the exec package.
459  *
460  * Then, mark the text image busy (so it can be demand paged) or error out if
461  * this is not possible.  Finally, set up vmcmds for the text, data, bss, and
462  * stack segments.
463  */
464 int
465 exec_elf_makecmds(struct proc *p, struct exec_package *epp)
466 {
467 	Elf_Ehdr *eh = epp->ep_hdr;
468 	Elf_Phdr *ph, *pp, *base_ph = NULL;
469 	Elf_Addr phdr = 0, exe_base = 0;
470 	int error, i, has_phdr = 0, names = 0;
471 	char *interp = NULL;
472 	u_long phsize;
473 	size_t randomizequota = ELF_RANDOMIZE_LIMIT;
474 
475 	if (epp->ep_hdrvalid < sizeof(Elf_Ehdr))
476 		return (ENOEXEC);
477 
478 	if (elf_check_header(eh) ||
479 	   (eh->e_type != ET_EXEC && eh->e_type != ET_DYN))
480 		return (ENOEXEC);
481 
482 	/*
483 	 * check if vnode is in open for writing, because we want to demand-
484 	 * page out of it.  if it is, don't do it, for various reasons.
485 	 */
486 	if (epp->ep_vp->v_writecount != 0) {
487 #ifdef DIAGNOSTIC
488 		if (epp->ep_vp->v_flag & VTEXT)
489 			panic("exec: a VTEXT vnode has writecount != 0");
490 #endif
491 		return (ETXTBSY);
492 	}
493 	/*
494 	 * Allocate space to hold all the program headers, and read them
495 	 * from the file
496 	 */
497 	ph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK);
498 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
499 
500 	if ((error = elf_read_from(p, epp->ep_vp, eh->e_phoff, ph,
501 	    phsize)) != 0)
502 		goto bad;
503 
504 	epp->ep_tsize = ELF_NO_ADDR;
505 	epp->ep_dsize = ELF_NO_ADDR;
506 
507 	for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) {
508 		if (pp->p_type == PT_INTERP && !interp) {
509 			if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN)
510 				goto bad;
511 			interp = pool_get(&namei_pool, PR_WAITOK);
512 			if ((error = elf_read_from(p, epp->ep_vp,
513 			    pp->p_offset, interp, pp->p_filesz)) != 0) {
514 				goto bad;
515 			}
516 			if (interp[pp->p_filesz - 1] != '\0')
517 				goto bad;
518 		} else if (pp->p_type == PT_LOAD) {
519 			if (pp->p_filesz > pp->p_memsz ||
520 			    pp->p_memsz == 0) {
521 				error = EINVAL;
522 				goto bad;
523 			}
524 			if (base_ph == NULL)
525 				base_ph = pp;
526 		} else if (pp->p_type == PT_PHDR) {
527 			has_phdr = 1;
528 		}
529 	}
530 
531 	if (eh->e_type == ET_DYN) {
532 		/* need phdr and load sections for PIE */
533 		if (!has_phdr || base_ph == NULL) {
534 			error = EINVAL;
535 			goto bad;
536 		}
537 		/* randomize exe_base for PIE */
538 		exe_base = uvm_map_pie(base_ph->p_align);
539 	}
540 
541 	/*
542 	 * Verify this is an OpenBSD executable.  If it's marked that way
543 	 * via a PT_NOTE then also check for a PT_OPENBSD_WXNEEDED segment.
544 	 */
545 	if ((error = elf_os_pt_note(p, epp, epp->ep_hdr, &names)) != 0)
546 		goto bad;
547 	if (eh->e_ident[EI_OSABI] == ELFOSABI_OPENBSD)
548 		names |= ELF_NOTE_NAME_OPENBSD;
549 
550 	/*
551 	 * Load all the necessary sections
552 	 */
553 	for (i = 0, pp = ph; i < eh->e_phnum; i++, pp++) {
554 		Elf_Addr addr, size = 0;
555 		int prot = 0;
556 		int flags = 0;
557 
558 		switch (pp->p_type) {
559 		case PT_LOAD:
560 			if (exe_base != 0) {
561 				if (pp == base_ph) {
562 					flags = VMCMD_BASE;
563 					addr = exe_base;
564 				} else {
565 					flags = VMCMD_RELATIVE;
566 					addr = pp->p_vaddr - base_ph->p_vaddr;
567 				}
568 			} else
569 				addr = ELF_NO_ADDR;
570 
571 			/* Permit system calls in specific main-programs */
572 			if (interp == NULL) {
573 				/* statics. Also block the ld.so syscall-grant */
574 				flags |= VMCMD_SYSCALL;
575 				p->p_vmspace->vm_map.flags |= VM_MAP_SYSCALL_ONCE;
576 			}
577 
578 			/*
579 			 * Calculates size of text and data segments
580 			 * by starting at first and going to end of last.
581 			 * 'rwx' sections are treated as data.
582 			 * this is correct for BSS_PLT, but may not be
583 			 * for DATA_PLT, is fine for TEXT_PLT.
584 			 */
585 			elf_load_psection(&epp->ep_vmcmds, epp->ep_vp,
586 			    pp, &addr, &size, &prot, flags);
587 
588 			/*
589 			 * Update exe_base in case alignment was off.
590 			 * For PIE, addr is relative to exe_base so
591 			 * adjust it (non PIE exe_base is 0 so no change).
592 			 */
593 			if (flags == VMCMD_BASE)
594 				exe_base = addr;
595 			else
596 				addr += exe_base;
597 
598 			/*
599 			 * Decide whether it's text or data by looking
600 			 * at the protection of the section
601 			 */
602 			if (prot & PROT_WRITE) {
603 				/* data section */
604 				if (epp->ep_dsize == ELF_NO_ADDR) {
605 					epp->ep_daddr = addr;
606 					epp->ep_dsize = size;
607 				} else {
608 					if (addr < epp->ep_daddr) {
609 						epp->ep_dsize =
610 						    epp->ep_dsize +
611 						    epp->ep_daddr -
612 						    addr;
613 						epp->ep_daddr = addr;
614 					} else
615 						epp->ep_dsize = addr+size -
616 						    epp->ep_daddr;
617 				}
618 			} else if (prot & PROT_EXEC) {
619 				/* text section */
620 				if (epp->ep_tsize == ELF_NO_ADDR) {
621 					epp->ep_taddr = addr;
622 					epp->ep_tsize = size;
623 				} else {
624 					if (addr < epp->ep_taddr) {
625 						epp->ep_tsize =
626 						    epp->ep_tsize +
627 						    epp->ep_taddr -
628 						    addr;
629 						epp->ep_taddr = addr;
630 					} else
631 						epp->ep_tsize = addr+size -
632 						    epp->ep_taddr;
633 				}
634 			}
635 			break;
636 
637 		case PT_SHLIB:
638 			error = ENOEXEC;
639 			goto bad;
640 
641 		case PT_INTERP:
642 			/* Already did this one */
643 		case PT_DYNAMIC:
644 		case PT_NOTE:
645 			break;
646 
647 		case PT_PHDR:
648 			/* Note address of program headers (in text segment) */
649 			phdr = pp->p_vaddr;
650 			break;
651 
652 		case PT_OPENBSD_RANDOMIZE:
653 			if (ph[i].p_memsz > randomizequota) {
654 				error = ENOMEM;
655 				goto bad;
656 			}
657 			randomizequota -= ph[i].p_memsz;
658 			NEW_VMCMD(&epp->ep_vmcmds, vmcmd_randomize,
659 			    ph[i].p_memsz, ph[i].p_vaddr + exe_base, NULLVP, 0, 0);
660 			break;
661 
662 		default:
663 			/*
664 			 * Not fatal, we don't need to understand everything
665 			 * :-)
666 			 */
667 			break;
668 		}
669 	}
670 
671 	phdr += exe_base;
672 
673 	/*
674 	 * Strangely some linux programs may have all load sections marked
675 	 * writeable, in this case, textsize is not -1, but rather 0;
676 	 */
677 	if (epp->ep_tsize == ELF_NO_ADDR)
678 		epp->ep_tsize = 0;
679 	/*
680 	 * Another possibility is that it has all load sections marked
681 	 * read-only.  Fake a zero-sized data segment right after the
682 	 * text segment.
683 	 */
684 	if (epp->ep_dsize == ELF_NO_ADDR) {
685 		epp->ep_daddr = round_page(epp->ep_taddr + epp->ep_tsize);
686 		epp->ep_dsize = 0;
687 	}
688 
689 	epp->ep_interp = interp;
690 	epp->ep_entry = eh->e_entry + exe_base;
691 
692 	/*
693 	 * Check if we found a dynamically linked binary and arrange to load
694 	 * its interpreter when the exec file is released.
695 	 */
696 	if (interp || eh->e_type == ET_DYN) {
697 		struct elf_args *ap;
698 
699 		ap = malloc(sizeof(*ap), M_TEMP, M_WAITOK);
700 
701 		ap->arg_phaddr = phdr;
702 		ap->arg_phentsize = eh->e_phentsize;
703 		ap->arg_phnum = eh->e_phnum;
704 		ap->arg_entry = eh->e_entry + exe_base;
705 		ap->arg_interp = exe_base;
706 
707 		epp->ep_args = ap;
708 	}
709 
710 	free(ph, M_TEMP, phsize);
711 	vn_marktext(epp->ep_vp);
712 	return (exec_setup_stack(p, epp));
713 
714 bad:
715 	if (interp)
716 		pool_put(&namei_pool, interp);
717 	free(ph, M_TEMP, phsize);
718 	kill_vmcmds(&epp->ep_vmcmds);
719 	if (error == 0)
720 		return (ENOEXEC);
721 	return (error);
722 }
723 
724 /*
725  * Phase II of load. It is now safe to load the interpreter. Info collected
726  * when loading the program is available for setup of the interpreter.
727  */
728 int
729 exec_elf_fixup(struct proc *p, struct exec_package *epp)
730 {
731 	char	*interp;
732 	int	error = 0;
733 	struct	elf_args *ap;
734 	AuxInfo ai[ELF_AUX_ENTRIES], *a;
735 
736 	ap = epp->ep_args;
737 	if (ap == NULL) {
738 		return (0);
739 	}
740 
741 	interp = epp->ep_interp;
742 
743 	if (interp &&
744 	    (error = elf_load_file(p, interp, epp, ap)) != 0) {
745 		free(ap, M_TEMP, sizeof *ap);
746 		pool_put(&namei_pool, interp);
747 		kill_vmcmds(&epp->ep_vmcmds);
748 		return (error);
749 	}
750 	/*
751 	 * We have to do this ourselves...
752 	 */
753 	error = exec_process_vmcmds(p, epp);
754 
755 	/*
756 	 * Push extra arguments on the stack needed by dynamically
757 	 * linked binaries
758 	 */
759 	if (error == 0) {
760 		memset(&ai, 0, sizeof ai);
761 		a = ai;
762 
763 		a->au_id = AUX_phdr;
764 		a->au_v = ap->arg_phaddr;
765 		a++;
766 
767 		a->au_id = AUX_phent;
768 		a->au_v = ap->arg_phentsize;
769 		a++;
770 
771 		a->au_id = AUX_phnum;
772 		a->au_v = ap->arg_phnum;
773 		a++;
774 
775 		a->au_id = AUX_pagesz;
776 		a->au_v = PAGE_SIZE;
777 		a++;
778 
779 		a->au_id = AUX_base;
780 		a->au_v = ap->arg_interp;
781 		a++;
782 
783 		a->au_id = AUX_flags;
784 		a->au_v = 0;
785 		a++;
786 
787 		a->au_id = AUX_entry;
788 		a->au_v = ap->arg_entry;
789 		a++;
790 
791 		a->au_id = AUX_openbsd_timekeep;
792 		a->au_v = p->p_p->ps_timekeep;
793 		a++;
794 
795 		a->au_id = AUX_null;
796 		a->au_v = 0;
797 		a++;
798 
799 		error = copyout(ai, epp->ep_auxinfo, sizeof ai);
800 	}
801 	free(ap, M_TEMP, sizeof *ap);
802 	if (interp)
803 		pool_put(&namei_pool, interp);
804 	return (error);
805 }
806 
807 int
808 elf_os_pt_note_name(Elf_Note *np)
809 {
810 	int i, j;
811 
812 	for (i = 0; i < nitems(elf_note_names); i++) {
813 		size_t namlen = strlen(elf_note_names[i].name);
814 		if (np->namesz < namlen)
815 			continue;
816 		/* verify name padding (after the NUL) is NUL */
817 		for (j = namlen + 1; j < elfround(np->namesz); j++)
818 			if (((char *)(np + 1))[j] != '\0')
819 				continue;
820 		/* verify desc padding is NUL */
821 		for (j = np->descsz; j < elfround(np->descsz); j++)
822 			if (((char *)(np + 1))[j] != '\0')
823 				continue;
824 		if (strcmp((char *)(np + 1), elf_note_names[i].name) == 0)
825 			return elf_note_names[i].id;
826 	}
827 	return (0);
828 }
829 
830 int
831 elf_os_pt_note(struct proc *p, struct exec_package *epp, Elf_Ehdr *eh, int *namesp)
832 {
833 	Elf_Phdr *hph, *ph;
834 	Elf_Note *np = NULL;
835 	size_t phsize, offset, pfilesz = 0, total;
836 	int error, names = 0;
837 
838 	hph = mallocarray(eh->e_phnum, sizeof(Elf_Phdr), M_TEMP, M_WAITOK);
839 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
840 	if ((error = elf_read_from(p, epp->ep_vp, eh->e_phoff,
841 	    hph, phsize)) != 0)
842 		goto out1;
843 
844 	for (ph = hph;  ph < &hph[eh->e_phnum]; ph++) {
845 		if (ph->p_type == PT_OPENBSD_WXNEEDED) {
846 			epp->ep_flags |= EXEC_WXNEEDED;
847 			continue;
848 		}
849 
850 		if (ph->p_type != PT_NOTE || ph->p_filesz > 1024)
851 			continue;
852 
853 		if (np && ph->p_filesz != pfilesz) {
854 			free(np, M_TEMP, pfilesz);
855 			np = NULL;
856 		}
857 		if (!np)
858 			np = malloc(ph->p_filesz, M_TEMP, M_WAITOK);
859 		pfilesz = ph->p_filesz;
860 		if ((error = elf_read_from(p, epp->ep_vp, ph->p_offset,
861 		    np, ph->p_filesz)) != 0)
862 			goto out2;
863 
864 		for (offset = 0; offset < ph->p_filesz; offset += total) {
865 			Elf_Note *np2 = (Elf_Note *)((char *)np + offset);
866 
867 			if (offset + sizeof(Elf_Note) > ph->p_filesz)
868 				break;
869 			total = sizeof(Elf_Note) + elfround(np2->namesz) +
870 			    elfround(np2->descsz);
871 			if (offset + total > ph->p_filesz)
872 				break;
873 			names |= elf_os_pt_note_name(np2);
874 		}
875 	}
876 
877 out2:
878 	free(np, M_TEMP, pfilesz);
879 out1:
880 	free(hph, M_TEMP, phsize);
881 	*namesp = names;
882 	return ((names & ELF_NOTE_NAME_OPENBSD) ? 0 : ENOEXEC);
883 }
884 
885 /*
886  * Start of routines related to dumping core
887  */
888 
889 #ifdef SMALL_KERNEL
890 int
891 coredump_elf(struct proc *p, void *cookie)
892 {
893 	return EPERM;
894 }
895 #else /* !SMALL_KERNEL */
896 
897 struct writesegs_state {
898 	off_t	notestart;
899 	off_t	secstart;
900 	off_t	secoff;
901 	struct	proc *p;
902 	void	*iocookie;
903 	Elf_Phdr *psections;
904 	size_t	psectionslen;
905 	size_t	notesize;
906 	int	npsections;
907 };
908 
909 uvm_coredump_setup_cb	coredump_setup_elf;
910 uvm_coredump_walk_cb	coredump_walk_elf;
911 
912 int	coredump_notes_elf(struct proc *, void *, size_t *);
913 int	coredump_note_elf(struct proc *, void *, size_t *);
914 int	coredump_writenote_elf(struct proc *, void *, Elf_Note *,
915 	    const char *, void *);
916 
917 int
918 coredump_elf(struct proc *p, void *cookie)
919 {
920 #ifdef DIAGNOSTIC
921 	off_t offset;
922 #endif
923 	struct writesegs_state ws;
924 	size_t notesize;
925 	int error, i;
926 
927 	ws.p = p;
928 	ws.iocookie = cookie;
929 	ws.psections = NULL;
930 
931 	/*
932 	 * Walk the map to get all the segment offsets and lengths,
933 	 * write out the ELF header.
934 	 */
935 	error = uvm_coredump_walkmap(p, coredump_setup_elf,
936 	    coredump_walk_elf, &ws);
937 	if (error)
938 		goto out;
939 
940 	error = coredump_write(cookie, UIO_SYSSPACE, ws.psections,
941 	    ws.psectionslen);
942 	if (error)
943 		goto out;
944 
945 	/* Write out the notes. */
946 	error = coredump_notes_elf(p, cookie, &notesize);
947 	if (error)
948 		goto out;
949 
950 #ifdef DIAGNOSTIC
951 	if (notesize != ws.notesize)
952 		panic("coredump: notesize changed: %zu != %zu",
953 		    ws.notesize, notesize);
954 	offset = ws.notestart + notesize;
955 	if (offset != ws.secstart)
956 		panic("coredump: offset %lld != secstart %lld",
957 		    (long long) offset, (long long) ws.secstart);
958 #endif
959 
960 	/* Pass 3: finally, write the sections themselves. */
961 	for (i = 0; i < ws.npsections - 1; i++) {
962 		Elf_Phdr *pent = &ws.psections[i];
963 		if (pent->p_filesz == 0)
964 			continue;
965 
966 #ifdef DIAGNOSTIC
967 		if (offset != pent->p_offset)
968 			panic("coredump: offset %lld != p_offset[%d] %lld",
969 			    (long long) offset, i,
970 			    (long long) pent->p_filesz);
971 #endif
972 
973 		error = coredump_write(cookie, UIO_USERSPACE,
974 		    (void *)(vaddr_t)pent->p_vaddr, pent->p_filesz);
975 		if (error)
976 			goto out;
977 
978 		coredump_unmap(cookie, (vaddr_t)pent->p_vaddr,
979 		    (vaddr_t)pent->p_vaddr + pent->p_filesz);
980 
981 #ifdef DIAGNOSTIC
982 		offset += ws.psections[i].p_filesz;
983 #endif
984 	}
985 
986 out:
987 	free(ws.psections, M_TEMP, ws.psectionslen);
988 	return (error);
989 }
990 
991 
992 /*
993  * Normally we lay out core files like this:
994  *	[ELF Header] [Program headers] [Notes] [data for PT_LOAD segments]
995  *
996  * However, if there's >= 65535 segments then it overflows the field
997  * in the ELF header, so the standard specifies putting a magic
998  * number there and saving the real count in the .sh_info field of
999  * the first *section* header...which requires generating a section
1000  * header.  To avoid confusing tools, we include an .shstrtab section
1001  * as well so all the indexes look valid.  So in this case we lay
1002  * out the core file like this:
1003  *	[ELF Header] [Section Headers] [.shstrtab] [Program headers] \
1004  *	[Notes] [data for PT_LOAD segments]
1005  *
1006  * The 'shstrtab' structure below is data for the second of the two
1007  * section headers, plus the .shstrtab itself, in one const buffer.
1008  */
1009 static const struct {
1010     Elf_Shdr	shdr;
1011     char	shstrtab[sizeof(ELF_SHSTRTAB) + 1];
1012 } shstrtab = {
1013     .shdr = {
1014 	.sh_name = 1,			/* offset in .shstrtab below */
1015 	.sh_type = SHT_STRTAB,
1016 	.sh_offset = sizeof(Elf_Ehdr) + 2*sizeof(Elf_Shdr),
1017 	.sh_size = sizeof(ELF_SHSTRTAB) + 1,
1018 	.sh_addralign = 1,
1019     },
1020     .shstrtab = "\0" ELF_SHSTRTAB,
1021 };
1022 
1023 int
1024 coredump_setup_elf(int segment_count, void *cookie)
1025 {
1026 	Elf_Ehdr ehdr;
1027 	struct writesegs_state *ws = cookie;
1028 	Elf_Phdr *note;
1029 	int error;
1030 
1031 	/* Get the count of segments, plus one for the PT_NOTE */
1032 	ws->npsections = segment_count + 1;
1033 
1034 	/* Get the size of the notes. */
1035 	error = coredump_notes_elf(ws->p, NULL, &ws->notesize);
1036 	if (error)
1037 		return error;
1038 
1039 	/* Setup the ELF header */
1040 	memset(&ehdr, 0, sizeof(ehdr));
1041 	memcpy(ehdr.e_ident, ELFMAG, SELFMAG);
1042 	ehdr.e_ident[EI_CLASS] = ELF_TARG_CLASS;
1043 	ehdr.e_ident[EI_DATA] = ELF_TARG_DATA;
1044 	ehdr.e_ident[EI_VERSION] = EV_CURRENT;
1045 	/* XXX Should be the OSABI/ABI version of the executable. */
1046 	ehdr.e_ident[EI_OSABI] = ELFOSABI_SYSV;
1047 	ehdr.e_ident[EI_ABIVERSION] = 0;
1048 	ehdr.e_type = ET_CORE;
1049 	/* XXX This should be the e_machine of the executable. */
1050 	ehdr.e_machine = ELF_TARG_MACH;
1051 	ehdr.e_version = EV_CURRENT;
1052 	ehdr.e_entry = 0;
1053 	ehdr.e_flags = 0;
1054 	ehdr.e_ehsize = sizeof(ehdr);
1055 	ehdr.e_phentsize = sizeof(Elf_Phdr);
1056 
1057 	if (ws->npsections < PN_XNUM) {
1058 		ehdr.e_phoff = sizeof(ehdr);
1059 		ehdr.e_shoff = 0;
1060 		ehdr.e_phnum = ws->npsections;
1061 		ehdr.e_shentsize = 0;
1062 		ehdr.e_shnum = 0;
1063 		ehdr.e_shstrndx = 0;
1064 	} else {
1065 		/* too many segments, use extension setup */
1066 		ehdr.e_shoff = sizeof(ehdr);
1067 		ehdr.e_phnum = PN_XNUM;
1068 		ehdr.e_shentsize = sizeof(Elf_Shdr);
1069 		ehdr.e_shnum = 2;
1070 		ehdr.e_shstrndx = 1;
1071 		ehdr.e_phoff = shstrtab.shdr.sh_offset + shstrtab.shdr.sh_size;
1072 	}
1073 
1074 	/* Write out the ELF header. */
1075 	error = coredump_write(ws->iocookie, UIO_SYSSPACE, &ehdr, sizeof(ehdr));
1076 	if (error)
1077 		return error;
1078 
1079 	/*
1080 	 * If an section header is needed to store extension info, write
1081 	 * it out after the ELF header and before the program header.
1082 	 */
1083 	if (ehdr.e_shnum != 0) {
1084 		Elf_Shdr shdr = { .sh_info = ws->npsections };
1085 		error = coredump_write(ws->iocookie, UIO_SYSSPACE, &shdr,
1086 		    sizeof shdr);
1087 		if (error)
1088 			return error;
1089 		error = coredump_write(ws->iocookie, UIO_SYSSPACE, &shstrtab,
1090 		    sizeof(shstrtab.shdr) + sizeof(shstrtab.shstrtab));
1091 		if (error)
1092 			return error;
1093 	}
1094 
1095 	/*
1096 	 * Allocate the segment header array and setup to collect
1097 	 * the section sizes and offsets
1098 	 */
1099 	ws->psections = mallocarray(ws->npsections, sizeof(Elf_Phdr),
1100 	    M_TEMP, M_WAITOK|M_CANFAIL|M_ZERO);
1101 	if (ws->psections == NULL)
1102 		return ENOMEM;
1103 	ws->psectionslen = ws->npsections * sizeof(Elf_Phdr);
1104 
1105 	ws->notestart = ehdr.e_phoff + ws->psectionslen;
1106 	ws->secstart = ws->notestart + ws->notesize;
1107 	ws->secoff = ws->secstart;
1108 
1109 	/* Fill in the PT_NOTE segment header in the last slot */
1110 	note = &ws->psections[ws->npsections - 1];
1111 	note->p_type = PT_NOTE;
1112 	note->p_offset = ws->notestart;
1113 	note->p_vaddr = 0;
1114 	note->p_paddr = 0;
1115 	note->p_filesz = ws->notesize;
1116 	note->p_memsz = 0;
1117 	note->p_flags = PF_R;
1118 	note->p_align = ELFROUNDSIZE;
1119 
1120 	return (0);
1121 }
1122 
1123 int
1124 coredump_walk_elf(vaddr_t start, vaddr_t realend, vaddr_t end, vm_prot_t prot,
1125     int nsegment, void *cookie)
1126 {
1127 	struct writesegs_state *ws = cookie;
1128 	Elf_Phdr phdr;
1129 	vsize_t size, realsize;
1130 
1131 	size = end - start;
1132 	realsize = realend - start;
1133 
1134 	phdr.p_type = PT_LOAD;
1135 	phdr.p_offset = ws->secoff;
1136 	phdr.p_vaddr = start;
1137 	phdr.p_paddr = 0;
1138 	phdr.p_filesz = realsize;
1139 	phdr.p_memsz = size;
1140 	phdr.p_flags = 0;
1141 	if (prot & PROT_READ)
1142 		phdr.p_flags |= PF_R;
1143 	if (prot & PROT_WRITE)
1144 		phdr.p_flags |= PF_W;
1145 	if (prot & PROT_EXEC)
1146 		phdr.p_flags |= PF_X;
1147 	phdr.p_align = PAGE_SIZE;
1148 
1149 	ws->secoff += phdr.p_filesz;
1150 	ws->psections[nsegment] = phdr;
1151 
1152 	return (0);
1153 }
1154 
1155 int
1156 coredump_notes_elf(struct proc *p, void *iocookie, size_t *sizep)
1157 {
1158 	struct ps_strings pss;
1159 	struct iovec iov;
1160 	struct uio uio;
1161 	struct elfcore_procinfo cpi;
1162 	Elf_Note nhdr;
1163 	struct process *pr = p->p_p;
1164 	struct proc *q;
1165 	size_t size, notesize;
1166 	int error;
1167 
1168 	size = 0;
1169 
1170 	/* First, write an elfcore_procinfo. */
1171 	notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) +
1172 	    elfround(sizeof(cpi));
1173 	if (iocookie) {
1174 		memset(&cpi, 0, sizeof(cpi));
1175 
1176 		cpi.cpi_version = ELFCORE_PROCINFO_VERSION;
1177 		cpi.cpi_cpisize = sizeof(cpi);
1178 		cpi.cpi_signo = p->p_sisig;
1179 		cpi.cpi_sigcode = p->p_sicode;
1180 
1181 		cpi.cpi_sigpend = p->p_siglist | pr->ps_siglist;
1182 		cpi.cpi_sigmask = p->p_sigmask;
1183 		cpi.cpi_sigignore = pr->ps_sigacts->ps_sigignore;
1184 		cpi.cpi_sigcatch = pr->ps_sigacts->ps_sigcatch;
1185 
1186 		cpi.cpi_pid = pr->ps_pid;
1187 		cpi.cpi_ppid = pr->ps_ppid;
1188 		cpi.cpi_pgrp = pr->ps_pgid;
1189 		if (pr->ps_session->s_leader)
1190 			cpi.cpi_sid = pr->ps_session->s_leader->ps_pid;
1191 		else
1192 			cpi.cpi_sid = 0;
1193 
1194 		cpi.cpi_ruid = p->p_ucred->cr_ruid;
1195 		cpi.cpi_euid = p->p_ucred->cr_uid;
1196 		cpi.cpi_svuid = p->p_ucred->cr_svuid;
1197 
1198 		cpi.cpi_rgid = p->p_ucred->cr_rgid;
1199 		cpi.cpi_egid = p->p_ucred->cr_gid;
1200 		cpi.cpi_svgid = p->p_ucred->cr_svgid;
1201 
1202 		(void)strlcpy(cpi.cpi_name, pr->ps_comm, sizeof(cpi.cpi_name));
1203 
1204 		nhdr.namesz = sizeof("OpenBSD");
1205 		nhdr.descsz = sizeof(cpi);
1206 		nhdr.type = NT_OPENBSD_PROCINFO;
1207 
1208 		error = coredump_writenote_elf(p, iocookie, &nhdr,
1209 		    "OpenBSD", &cpi);
1210 		if (error)
1211 			return (error);
1212 	}
1213 	size += notesize;
1214 
1215 	/* Second, write an NT_OPENBSD_AUXV note. */
1216 	notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) +
1217 	    elfround(ELF_AUX_WORDS * sizeof(char *));
1218 	if (iocookie) {
1219 		iov.iov_base = &pss;
1220 		iov.iov_len = sizeof(pss);
1221 		uio.uio_iov = &iov;
1222 		uio.uio_iovcnt = 1;
1223 		uio.uio_offset = (off_t)pr->ps_strings;
1224 		uio.uio_resid = sizeof(pss);
1225 		uio.uio_segflg = UIO_SYSSPACE;
1226 		uio.uio_rw = UIO_READ;
1227 		uio.uio_procp = NULL;
1228 
1229 		error = uvm_io(&p->p_vmspace->vm_map, &uio, 0);
1230 		if (error)
1231 			return (error);
1232 
1233 		if (pss.ps_envstr == NULL)
1234 			return (EIO);
1235 
1236 		nhdr.namesz = sizeof("OpenBSD");
1237 		nhdr.descsz = ELF_AUX_WORDS * sizeof(char *);
1238 		nhdr.type = NT_OPENBSD_AUXV;
1239 
1240 		error = coredump_write(iocookie, UIO_SYSSPACE,
1241 		    &nhdr, sizeof(nhdr));
1242 		if (error)
1243 			return (error);
1244 
1245 		error = coredump_write(iocookie, UIO_SYSSPACE,
1246 		    "OpenBSD", elfround(nhdr.namesz));
1247 		if (error)
1248 			return (error);
1249 
1250 		error = coredump_write(iocookie, UIO_USERSPACE,
1251 		    pss.ps_envstr + pss.ps_nenvstr + 1, nhdr.descsz);
1252 		if (error)
1253 			return (error);
1254 	}
1255 	size += notesize;
1256 
1257 #ifdef PT_WCOOKIE
1258 	notesize = sizeof(nhdr) + elfround(sizeof("OpenBSD")) +
1259 	    elfround(sizeof(register_t));
1260 	if (iocookie) {
1261 		register_t wcookie;
1262 
1263 		nhdr.namesz = sizeof("OpenBSD");
1264 		nhdr.descsz = sizeof(register_t);
1265 		nhdr.type = NT_OPENBSD_WCOOKIE;
1266 
1267 		wcookie = process_get_wcookie(p);
1268 		error = coredump_writenote_elf(p, iocookie, &nhdr,
1269 		    "OpenBSD", &wcookie);
1270 		if (error)
1271 			return (error);
1272 	}
1273 	size += notesize;
1274 #endif
1275 
1276 	/*
1277 	 * Now write the register info for the thread that caused the
1278 	 * coredump.
1279 	 */
1280 	error = coredump_note_elf(p, iocookie, &notesize);
1281 	if (error)
1282 		return (error);
1283 	size += notesize;
1284 
1285 	/*
1286 	 * Now, for each thread, write the register info and any other
1287 	 * per-thread notes.  Since we're dumping core, all the other
1288 	 * threads in the process have been stopped and the list can't
1289 	 * change.
1290 	 */
1291 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
1292 		if (q == p)		/* we've taken care of this thread */
1293 			continue;
1294 		error = coredump_note_elf(q, iocookie, &notesize);
1295 		if (error)
1296 			return (error);
1297 		size += notesize;
1298 	}
1299 
1300 	*sizep = size;
1301 	return (0);
1302 }
1303 
1304 int
1305 coredump_note_elf(struct proc *p, void *iocookie, size_t *sizep)
1306 {
1307 	Elf_Note nhdr;
1308 	int size, notesize, error;
1309 	int namesize;
1310 	char name[64+ELFROUNDSIZE];
1311 	struct reg intreg;
1312 #ifdef PT_GETFPREGS
1313 	struct fpreg freg;
1314 #endif
1315 
1316 	size = 0;
1317 
1318 	snprintf(name, sizeof(name)-ELFROUNDSIZE, "%s@%d",
1319 	    "OpenBSD", p->p_tid + THREAD_PID_OFFSET);
1320 	namesize = strlen(name) + 1;
1321 	memset(name + namesize, 0, elfround(namesize) - namesize);
1322 
1323 	notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(intreg));
1324 	if (iocookie) {
1325 		error = process_read_regs(p, &intreg);
1326 		if (error)
1327 			return (error);
1328 
1329 		nhdr.namesz = namesize;
1330 		nhdr.descsz = sizeof(intreg);
1331 		nhdr.type = NT_OPENBSD_REGS;
1332 
1333 		error = coredump_writenote_elf(p, iocookie, &nhdr,
1334 		    name, &intreg);
1335 		if (error)
1336 			return (error);
1337 
1338 	}
1339 	size += notesize;
1340 
1341 #ifdef PT_GETFPREGS
1342 	notesize = sizeof(nhdr) + elfround(namesize) + elfround(sizeof(freg));
1343 	if (iocookie) {
1344 		error = process_read_fpregs(p, &freg);
1345 		if (error)
1346 			return (error);
1347 
1348 		nhdr.namesz = namesize;
1349 		nhdr.descsz = sizeof(freg);
1350 		nhdr.type = NT_OPENBSD_FPREGS;
1351 
1352 		error = coredump_writenote_elf(p, iocookie, &nhdr, name, &freg);
1353 		if (error)
1354 			return (error);
1355 	}
1356 	size += notesize;
1357 #endif
1358 
1359 	*sizep = size;
1360 	/* XXX Add hook for machdep per-LWP notes. */
1361 	return (0);
1362 }
1363 
1364 int
1365 coredump_writenote_elf(struct proc *p, void *cookie, Elf_Note *nhdr,
1366     const char *name, void *data)
1367 {
1368 	int error;
1369 
1370 	error = coredump_write(cookie, UIO_SYSSPACE, nhdr, sizeof(*nhdr));
1371 	if (error)
1372 		return error;
1373 
1374 	error = coredump_write(cookie, UIO_SYSSPACE, name,
1375 	    elfround(nhdr->namesz));
1376 	if (error)
1377 		return error;
1378 
1379 	return coredump_write(cookie, UIO_SYSSPACE, data, nhdr->descsz);
1380 }
1381 #endif /* !SMALL_KERNEL */
1382