xref: /freebsd/sys/kern/imgact_elf.c (revision 7bd6fde3)
1 /*-
2  * Copyright (c) 2000 David O'Brien
3  * Copyright (c) 1995-1996 S�ren Schmidt
4  * Copyright (c) 1996 Peter Wemm
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_compat.h"
35 
36 #include <sys/param.h>
37 #include <sys/exec.h>
38 #include <sys/fcntl.h>
39 #include <sys/imgact.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mount.h>
45 #include <sys/mutex.h>
46 #include <sys/mman.h>
47 #include <sys/namei.h>
48 #include <sys/pioctl.h>
49 #include <sys/proc.h>
50 #include <sys/procfs.h>
51 #include <sys/resourcevar.h>
52 #include <sys/sf_buf.h>
53 #include <sys/systm.h>
54 #include <sys/signalvar.h>
55 #include <sys/stat.h>
56 #include <sys/sx.h>
57 #include <sys/syscall.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysent.h>
60 #include <sys/vnode.h>
61 
62 #include <vm/vm.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_param.h>
65 #include <vm/pmap.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_extern.h>
69 
70 #include <machine/elf.h>
71 #include <machine/md_var.h>
72 
73 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
74 #include <machine/fpu.h>
75 #include <compat/ia32/ia32_reg.h>
76 #endif
77 
78 #define OLD_EI_BRAND	8
79 
80 static int __elfN(check_header)(const Elf_Ehdr *hdr);
81 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr,
82     const char *interp);
83 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
84     u_long *entry, size_t pagesize);
85 static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object,
86     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
87     vm_prot_t prot, size_t pagesize);
88 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
89 
90 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
91     "");
92 
93 int __elfN(fallback_brand) = -1;
94 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
95     fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
96     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
97 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
98     &__elfN(fallback_brand));
99 
100 static int elf_trace = 0;
101 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
102 
103 static int elf_legacy_coredump = 0;
104 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
105     &elf_legacy_coredump, 0, "");
106 
107 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
108 
109 int
110 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
111 {
112 	int i;
113 
114 	for (i = 0; i < MAX_BRANDS; i++) {
115 		if (elf_brand_list[i] == NULL) {
116 			elf_brand_list[i] = entry;
117 			break;
118 		}
119 	}
120 	if (i == MAX_BRANDS)
121 		return (-1);
122 	return (0);
123 }
124 
125 int
126 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
127 {
128 	int i;
129 
130 	for (i = 0; i < MAX_BRANDS; i++) {
131 		if (elf_brand_list[i] == entry) {
132 			elf_brand_list[i] = NULL;
133 			break;
134 		}
135 	}
136 	if (i == MAX_BRANDS)
137 		return (-1);
138 	return (0);
139 }
140 
141 int
142 __elfN(brand_inuse)(Elf_Brandinfo *entry)
143 {
144 	struct proc *p;
145 	int rval = FALSE;
146 
147 	sx_slock(&allproc_lock);
148 	FOREACH_PROC_IN_SYSTEM(p) {
149 		if (p->p_sysent == entry->sysvec) {
150 			rval = TRUE;
151 			break;
152 		}
153 	}
154 	sx_sunlock(&allproc_lock);
155 
156 	return (rval);
157 }
158 
159 static Elf_Brandinfo *
160 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
161 {
162 	Elf_Brandinfo *bi;
163 	int i;
164 
165 	/*
166 	 * We support three types of branding -- (1) the ELF EI_OSABI field
167 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
168 	 * branding w/in the ELF header, and (3) path of the `interp_path'
169 	 * field.  We should also look for an ".note.ABI-tag" ELF section now
170 	 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
171 	 */
172 
173 	/* If the executable has a brand, search for it in the brand list. */
174 	for (i = 0; i < MAX_BRANDS; i++) {
175 		bi = elf_brand_list[i];
176 		if (bi != NULL && hdr->e_machine == bi->machine &&
177 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
178 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
179 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
180 			return (bi);
181 	}
182 
183 	/* Lacking a known brand, search for a recognized interpreter. */
184 	if (interp != NULL) {
185 		for (i = 0; i < MAX_BRANDS; i++) {
186 			bi = elf_brand_list[i];
187 			if (bi != NULL && hdr->e_machine == bi->machine &&
188 			    strcmp(interp, bi->interp_path) == 0)
189 				return (bi);
190 		}
191 	}
192 
193 	/* Lacking a recognized interpreter, try the default brand */
194 	for (i = 0; i < MAX_BRANDS; i++) {
195 		bi = elf_brand_list[i];
196 		if (bi != NULL && hdr->e_machine == bi->machine &&
197 		    __elfN(fallback_brand) == bi->brand)
198 			return (bi);
199 	}
200 	return (NULL);
201 }
202 
203 static int
204 __elfN(check_header)(const Elf_Ehdr *hdr)
205 {
206 	Elf_Brandinfo *bi;
207 	int i;
208 
209 	if (!IS_ELF(*hdr) ||
210 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
211 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
212 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
213 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
214 	    hdr->e_version != ELF_TARG_VER)
215 		return (ENOEXEC);
216 
217 	/*
218 	 * Make sure we have at least one brand for this machine.
219 	 */
220 
221 	for (i = 0; i < MAX_BRANDS; i++) {
222 		bi = elf_brand_list[i];
223 		if (bi != NULL && bi->machine == hdr->e_machine)
224 			break;
225 	}
226 	if (i == MAX_BRANDS)
227 		return (ENOEXEC);
228 
229 	return (0);
230 }
231 
232 static int
233 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
234     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
235 {
236 	struct sf_buf *sf;
237 	int error;
238 	vm_offset_t off;
239 
240 	/*
241 	 * Create the page if it doesn't exist yet. Ignore errors.
242 	 */
243 	vm_map_lock(map);
244 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
245 	    VM_PROT_ALL, VM_PROT_ALL, 0);
246 	vm_map_unlock(map);
247 
248 	/*
249 	 * Find the page from the underlying object.
250 	 */
251 	if (object) {
252 		sf = vm_imgact_map_page(object, offset);
253 		if (sf == NULL)
254 			return (KERN_FAILURE);
255 		off = offset - trunc_page(offset);
256 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
257 		    end - start);
258 		vm_imgact_unmap_page(sf);
259 		if (error) {
260 			return (KERN_FAILURE);
261 		}
262 	}
263 
264 	return (KERN_SUCCESS);
265 }
266 
267 static int
268 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
269     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
270 {
271 	struct sf_buf *sf;
272 	vm_offset_t off;
273 	vm_size_t sz;
274 	int error, rv;
275 
276 	if (start != trunc_page(start)) {
277 		rv = __elfN(map_partial)(map, object, offset, start,
278 		    round_page(start), prot);
279 		if (rv)
280 			return (rv);
281 		offset += round_page(start) - start;
282 		start = round_page(start);
283 	}
284 	if (end != round_page(end)) {
285 		rv = __elfN(map_partial)(map, object, offset +
286 		    trunc_page(end) - start, trunc_page(end), end, prot);
287 		if (rv)
288 			return (rv);
289 		end = trunc_page(end);
290 	}
291 	if (end > start) {
292 		if (offset & PAGE_MASK) {
293 			/*
294 			 * The mapping is not page aligned. This means we have
295 			 * to copy the data. Sigh.
296 			 */
297 			rv = vm_map_find(map, NULL, 0, &start, end - start,
298 			    FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
299 			if (rv)
300 				return (rv);
301 			if (object == NULL)
302 				return (KERN_SUCCESS);
303 			for (; start < end; start += sz) {
304 				sf = vm_imgact_map_page(object, offset);
305 				if (sf == NULL)
306 					return (KERN_FAILURE);
307 				off = offset - trunc_page(offset);
308 				sz = end - start;
309 				if (sz > PAGE_SIZE - off)
310 					sz = PAGE_SIZE - off;
311 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
312 				    (caddr_t)start, sz);
313 				vm_imgact_unmap_page(sf);
314 				if (error) {
315 					return (KERN_FAILURE);
316 				}
317 				offset += sz;
318 			}
319 			rv = KERN_SUCCESS;
320 		} else {
321 			vm_object_reference(object);
322 			vm_map_lock(map);
323 			rv = vm_map_insert(map, object, offset, start, end,
324 			    prot, VM_PROT_ALL, cow);
325 			vm_map_unlock(map);
326 			if (rv != KERN_SUCCESS)
327 				vm_object_deallocate(object);
328 		}
329 		return (rv);
330 	} else {
331 		return (KERN_SUCCESS);
332 	}
333 }
334 
335 static int
336 __elfN(load_section)(struct vmspace *vmspace,
337 	vm_object_t object, vm_offset_t offset,
338 	caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
339 	size_t pagesize)
340 {
341 	struct sf_buf *sf;
342 	size_t map_len;
343 	vm_offset_t map_addr;
344 	int error, rv, cow;
345 	size_t copy_len;
346 	vm_offset_t file_addr;
347 
348 	/*
349 	 * It's necessary to fail if the filsz + offset taken from the
350 	 * header is greater than the actual file pager object's size.
351 	 * If we were to allow this, then the vm_map_find() below would
352 	 * walk right off the end of the file object and into the ether.
353 	 *
354 	 * While I'm here, might as well check for something else that
355 	 * is invalid: filsz cannot be greater than memsz.
356 	 */
357 	if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
358 	    filsz > memsz) {
359 		uprintf("elf_load_section: truncated ELF file\n");
360 		return (ENOEXEC);
361 	}
362 
363 #define trunc_page_ps(va, ps)	((va) & ~(ps - 1))
364 #define round_page_ps(va, ps)	(((va) + (ps - 1)) & ~(ps - 1))
365 
366 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
367 	file_addr = trunc_page_ps(offset, pagesize);
368 
369 	/*
370 	 * We have two choices.  We can either clear the data in the last page
371 	 * of an oversized mapping, or we can start the anon mapping a page
372 	 * early and copy the initialized data into that first page.  We
373 	 * choose the second..
374 	 */
375 	if (memsz > filsz)
376 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
377 	else
378 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
379 
380 	if (map_len != 0) {
381 		/* cow flags: don't dump readonly sections in core */
382 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
383 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
384 
385 		rv = __elfN(map_insert)(&vmspace->vm_map,
386 				      object,
387 				      file_addr,	/* file offset */
388 				      map_addr,		/* virtual start */
389 				      map_addr + map_len,/* virtual end */
390 				      prot,
391 				      cow);
392 		if (rv != KERN_SUCCESS)
393 			return (EINVAL);
394 
395 		/* we can stop now if we've covered it all */
396 		if (memsz == filsz) {
397 			return (0);
398 		}
399 	}
400 
401 
402 	/*
403 	 * We have to get the remaining bit of the file into the first part
404 	 * of the oversized map segment.  This is normally because the .data
405 	 * segment in the file is extended to provide bss.  It's a neat idea
406 	 * to try and save a page, but it's a pain in the behind to implement.
407 	 */
408 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
409 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
410 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
411 	    map_addr;
412 
413 	/* This had damn well better be true! */
414 	if (map_len != 0) {
415 		rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
416 		    map_addr + map_len, VM_PROT_ALL, 0);
417 		if (rv != KERN_SUCCESS) {
418 			return (EINVAL);
419 		}
420 	}
421 
422 	if (copy_len != 0) {
423 		vm_offset_t off;
424 
425 		sf = vm_imgact_map_page(object, offset + filsz);
426 		if (sf == NULL)
427 			return (EIO);
428 
429 		/* send the page fragment to user space */
430 		off = trunc_page_ps(offset + filsz, pagesize) -
431 		    trunc_page(offset + filsz);
432 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
433 		    (caddr_t)map_addr, copy_len);
434 		vm_imgact_unmap_page(sf);
435 		if (error) {
436 			return (error);
437 		}
438 	}
439 
440 	/*
441 	 * set it to the specified protection.
442 	 * XXX had better undo the damage from pasting over the cracks here!
443 	 */
444 	vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
445 	    round_page(map_addr + map_len),  prot, FALSE);
446 
447 	return (0);
448 }
449 
450 /*
451  * Load the file "file" into memory.  It may be either a shared object
452  * or an executable.
453  *
454  * The "addr" reference parameter is in/out.  On entry, it specifies
455  * the address where a shared object should be loaded.  If the file is
456  * an executable, this value is ignored.  On exit, "addr" specifies
457  * where the file was actually loaded.
458  *
459  * The "entry" reference parameter is out only.  On exit, it specifies
460  * the entry point for the loaded file.
461  */
462 static int
463 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
464 	u_long *entry, size_t pagesize)
465 {
466 	struct {
467 		struct nameidata nd;
468 		struct vattr attr;
469 		struct image_params image_params;
470 	} *tempdata;
471 	const Elf_Ehdr *hdr = NULL;
472 	const Elf_Phdr *phdr = NULL;
473 	struct nameidata *nd;
474 	struct vmspace *vmspace = p->p_vmspace;
475 	struct vattr *attr;
476 	struct image_params *imgp;
477 	vm_prot_t prot;
478 	u_long rbase;
479 	u_long base_addr = 0;
480 	int vfslocked, error, i, numsegs;
481 
482 	if (curthread->td_proc != p)
483 		panic("elf_load_file - thread");	/* XXXKSE DIAGNOSTIC */
484 
485 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
486 	nd = &tempdata->nd;
487 	attr = &tempdata->attr;
488 	imgp = &tempdata->image_params;
489 
490 	/*
491 	 * Initialize part of the common data
492 	 */
493 	imgp->proc = p;
494 	imgp->attr = attr;
495 	imgp->firstpage = NULL;
496 	imgp->image_header = NULL;
497 	imgp->object = NULL;
498 	imgp->execlabel = NULL;
499 
500 	/* XXXKSE */
501 	NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
502 	    curthread);
503 	vfslocked = 0;
504 	if ((error = namei(nd)) != 0) {
505 		nd->ni_vp = NULL;
506 		goto fail;
507 	}
508 	vfslocked = NDHASGIANT(nd);
509 	NDFREE(nd, NDF_ONLY_PNBUF);
510 	imgp->vp = nd->ni_vp;
511 
512 	/*
513 	 * Check permissions, modes, uid, etc on the file, and "open" it.
514 	 */
515 	error = exec_check_permissions(imgp);
516 	if (error)
517 		goto fail;
518 
519 	error = exec_map_first_page(imgp);
520 	if (error)
521 		goto fail;
522 
523 	/*
524 	 * Also make certain that the interpreter stays the same, so set
525 	 * its VV_TEXT flag, too.
526 	 */
527 	nd->ni_vp->v_vflag |= VV_TEXT;
528 
529 	imgp->object = nd->ni_vp->v_object;
530 
531 	hdr = (const Elf_Ehdr *)imgp->image_header;
532 	if ((error = __elfN(check_header)(hdr)) != 0)
533 		goto fail;
534 	if (hdr->e_type == ET_DYN)
535 		rbase = *addr;
536 	else if (hdr->e_type == ET_EXEC)
537 		rbase = 0;
538 	else {
539 		error = ENOEXEC;
540 		goto fail;
541 	}
542 
543 	/* Only support headers that fit within first page for now      */
544 	/*    (multiplication of two Elf_Half fields will not overflow) */
545 	if ((hdr->e_phoff > PAGE_SIZE) ||
546 	    (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
547 		error = ENOEXEC;
548 		goto fail;
549 	}
550 
551 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
552 
553 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
554 		if (phdr[i].p_type == PT_LOAD) {	/* Loadable segment */
555 			prot = 0;
556 			if (phdr[i].p_flags & PF_X)
557   				prot |= VM_PROT_EXECUTE;
558 			if (phdr[i].p_flags & PF_W)
559   				prot |= VM_PROT_WRITE;
560 			if (phdr[i].p_flags & PF_R)
561   				prot |= VM_PROT_READ;
562 
563 			if ((error = __elfN(load_section)(vmspace,
564 			    imgp->object, phdr[i].p_offset,
565 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
566 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
567 			    pagesize)) != 0)
568 				goto fail;
569 			/*
570 			 * Establish the base address if this is the
571 			 * first segment.
572 			 */
573 			if (numsegs == 0)
574   				base_addr = trunc_page(phdr[i].p_vaddr +
575 				    rbase);
576 			numsegs++;
577 		}
578 	}
579 	*addr = base_addr;
580 	*entry = (unsigned long)hdr->e_entry + rbase;
581 
582 fail:
583 	if (imgp->firstpage)
584 		exec_unmap_first_page(imgp);
585 
586 	if (nd->ni_vp)
587 		vput(nd->ni_vp);
588 
589 	VFS_UNLOCK_GIANT(vfslocked);
590 	free(tempdata, M_TEMP);
591 
592 	return (error);
593 }
594 
595 static int
596 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
597 {
598 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
599 	const Elf_Phdr *phdr;
600 	Elf_Auxargs *elf_auxargs;
601 	struct vmspace *vmspace;
602 	vm_prot_t prot;
603 	u_long text_size = 0, data_size = 0, total_size = 0;
604 	u_long text_addr = 0, data_addr = 0;
605 	u_long seg_size, seg_addr;
606 	u_long addr, entry = 0, proghdr = 0;
607 	int error = 0, i;
608 	const char *interp = NULL;
609 	Elf_Brandinfo *brand_info;
610 	char *path;
611 	struct thread *td = curthread;
612 	struct sysentvec *sv;
613 
614 	/*
615 	 * Do we have a valid ELF header ?
616 	 *
617 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
618 	 * if particular brand doesn't support it.
619 	 */
620 	if (__elfN(check_header)(hdr) != 0 ||
621 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
622 		return (-1);
623 
624 	/*
625 	 * From here on down, we return an errno, not -1, as we've
626 	 * detected an ELF file.
627 	 */
628 
629 	if ((hdr->e_phoff > PAGE_SIZE) ||
630 	    (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
631 		/* Only support headers in first page for now */
632 		return (ENOEXEC);
633 	}
634 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
635 	for (i = 0; i < hdr->e_phnum; i++) {
636 		if (phdr[i].p_type == PT_INTERP) {
637 			/* Path to interpreter */
638 			if (phdr[i].p_filesz > MAXPATHLEN ||
639 			    phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
640 				return (ENOEXEC);
641 			interp = imgp->image_header + phdr[i].p_offset;
642 			break;
643 		}
644 	}
645 
646 	brand_info = __elfN(get_brandinfo)(hdr, interp);
647 	if (brand_info == NULL) {
648 		uprintf("ELF binary type \"%u\" not known.\n",
649 		    hdr->e_ident[EI_OSABI]);
650 		return (ENOEXEC);
651 	}
652 	if (hdr->e_type == ET_DYN &&
653 	    (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
654 		return (ENOEXEC);
655 	sv = brand_info->sysvec;
656 	if (interp != NULL && brand_info->interp_newpath != NULL)
657 		interp = brand_info->interp_newpath;
658 
659 	/*
660 	 * Avoid a possible deadlock if the current address space is destroyed
661 	 * and that address space maps the locked vnode.  In the common case,
662 	 * the locked vnode's v_usecount is decremented but remains greater
663 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
664 	 * However, in cases where the vnode lock is external, such as nullfs,
665 	 * v_usecount may become zero.
666 	 */
667 	VOP_UNLOCK(imgp->vp, 0, td);
668 
669 	exec_new_vmspace(imgp, sv);
670 
671 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
672 
673 	vmspace = imgp->proc->p_vmspace;
674 
675 	for (i = 0; i < hdr->e_phnum; i++) {
676 		switch (phdr[i].p_type) {
677 		case PT_LOAD:	/* Loadable segment */
678 			prot = 0;
679 			if (phdr[i].p_flags & PF_X)
680   				prot |= VM_PROT_EXECUTE;
681 			if (phdr[i].p_flags & PF_W)
682   				prot |= VM_PROT_WRITE;
683 			if (phdr[i].p_flags & PF_R)
684   				prot |= VM_PROT_READ;
685 
686 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
687 			/*
688 			 * Some x86 binaries assume read == executable,
689 			 * notably the M3 runtime and therefore cvsup
690 			 */
691 			if (prot & VM_PROT_READ)
692 				prot |= VM_PROT_EXECUTE;
693 #endif
694 
695 			if ((error = __elfN(load_section)(vmspace,
696 			    imgp->object, phdr[i].p_offset,
697 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr,
698 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
699 			    sv->sv_pagesize)) != 0)
700 				return (error);
701 
702 			/*
703 			 * If this segment contains the program headers,
704 			 * remember their virtual address for the AT_PHDR
705 			 * aux entry. Static binaries don't usually include
706 			 * a PT_PHDR entry.
707 			 */
708 			if (phdr[i].p_offset == 0 &&
709 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
710 				<= phdr[i].p_filesz)
711 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
712 
713 			seg_addr = trunc_page(phdr[i].p_vaddr);
714 			seg_size = round_page(phdr[i].p_memsz +
715 			    phdr[i].p_vaddr - seg_addr);
716 
717 			/*
718 			 * Is this .text or .data?  We can't use
719 			 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
720 			 * alpha terribly and possibly does other bad
721 			 * things so we stick to the old way of figuring
722 			 * it out:  If the segment contains the program
723 			 * entry point, it's a text segment, otherwise it
724 			 * is a data segment.
725 			 *
726 			 * Note that obreak() assumes that data_addr +
727 			 * data_size == end of data load area, and the ELF
728 			 * file format expects segments to be sorted by
729 			 * address.  If multiple data segments exist, the
730 			 * last one will be used.
731 			 */
732 			if (hdr->e_entry >= phdr[i].p_vaddr &&
733 			    hdr->e_entry < (phdr[i].p_vaddr +
734 			    phdr[i].p_memsz)) {
735 				text_size = seg_size;
736 				text_addr = seg_addr;
737 				entry = (u_long)hdr->e_entry;
738 			} else {
739 				data_size = seg_size;
740 				data_addr = seg_addr;
741 			}
742 			total_size += seg_size;
743 			break;
744 		case PT_PHDR: 	/* Program header table info */
745 			proghdr = phdr[i].p_vaddr;
746 			break;
747 		default:
748 			break;
749 		}
750 	}
751 
752 	if (data_addr == 0 && data_size == 0) {
753 		data_addr = text_addr;
754 		data_size = text_size;
755 	}
756 
757 	/*
758 	 * Check limits.  It should be safe to check the
759 	 * limits after loading the segments since we do
760 	 * not actually fault in all the segments pages.
761 	 */
762 	PROC_LOCK(imgp->proc);
763 	if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
764 	    text_size > maxtsiz ||
765 	    total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
766 		PROC_UNLOCK(imgp->proc);
767 		return (ENOMEM);
768 	}
769 
770 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
771 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
772 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
773 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
774 
775 	/*
776 	 * We load the dynamic linker where a userland call
777 	 * to mmap(0, ...) would put it.  The rationale behind this
778 	 * calculation is that it leaves room for the heap to grow to
779 	 * its maximum allowed size.
780 	 */
781 	addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
782 	    lim_max(imgp->proc, RLIMIT_DATA));
783 	PROC_UNLOCK(imgp->proc);
784 
785 	imgp->entry_addr = entry;
786 
787 	imgp->proc->p_sysent = sv;
788 	if (interp != NULL) {
789 		VOP_UNLOCK(imgp->vp, 0, td);
790 		if (brand_info->emul_path != NULL &&
791 		    brand_info->emul_path[0] != '\0') {
792 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
793 			snprintf(path, MAXPATHLEN, "%s%s",
794 			    brand_info->emul_path, interp);
795 			error = __elfN(load_file)(imgp->proc, path, &addr,
796 			    &imgp->entry_addr, sv->sv_pagesize);
797 			free(path, M_TEMP);
798 			if (error == 0)
799 				interp = NULL;
800 		}
801 		if (interp != NULL) {
802 			error = __elfN(load_file)(imgp->proc, interp, &addr,
803 			    &imgp->entry_addr, sv->sv_pagesize);
804 		}
805 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
806 		if (error != 0) {
807 			uprintf("ELF interpreter %s not found\n", interp);
808 			return (error);
809 		}
810 	}
811 
812 	/*
813 	 * Construct auxargs table (used by the fixup routine)
814 	 */
815 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
816 	elf_auxargs->execfd = -1;
817 	elf_auxargs->phdr = proghdr;
818 	elf_auxargs->phent = hdr->e_phentsize;
819 	elf_auxargs->phnum = hdr->e_phnum;
820 	elf_auxargs->pagesz = PAGE_SIZE;
821 	elf_auxargs->base = addr;
822 	elf_auxargs->flags = 0;
823 	elf_auxargs->entry = entry;
824 	elf_auxargs->trace = elf_trace;
825 
826 	imgp->auxargs = elf_auxargs;
827 	imgp->interpreted = 0;
828 
829 	return (error);
830 }
831 
832 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
833 
834 int
835 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
836 {
837 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
838 	Elf_Addr *base;
839 	Elf_Addr *pos;
840 
841 	base = (Elf_Addr *)*stack_base;
842 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
843 
844 	if (args->trace) {
845 		AUXARGS_ENTRY(pos, AT_DEBUG, 1);
846 	}
847 	if (args->execfd != -1) {
848 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
849 	}
850 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
851 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
852 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
853 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
854 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
855 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
856 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
857 	AUXARGS_ENTRY(pos, AT_NULL, 0);
858 
859 	free(imgp->auxargs, M_TEMP);
860 	imgp->auxargs = NULL;
861 
862 	base--;
863 	suword(base, (long)imgp->args->argc);
864 	*stack_base = (register_t *)base;
865 	return (0);
866 }
867 
868 /*
869  * Code for generating ELF core dumps.
870  */
871 
872 typedef void (*segment_callback)(vm_map_entry_t, void *);
873 
874 /* Closure for cb_put_phdr(). */
875 struct phdr_closure {
876 	Elf_Phdr *phdr;		/* Program header to fill in */
877 	Elf_Off offset;		/* Offset of segment in core file */
878 };
879 
880 /* Closure for cb_size_segment(). */
881 struct sseg_closure {
882 	int count;		/* Count of writable segments. */
883 	size_t size;		/* Total size of all writable segments. */
884 };
885 
886 static void cb_put_phdr(vm_map_entry_t, void *);
887 static void cb_size_segment(vm_map_entry_t, void *);
888 static void each_writable_segment(struct thread *, segment_callback, void *);
889 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
890     int, void *, size_t);
891 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
892 static void __elfN(putnote)(void *, size_t *, const char *, int,
893     const void *, size_t);
894 
895 extern int osreldate;
896 
897 int
898 __elfN(coredump)(td, vp, limit)
899 	struct thread *td;
900 	struct vnode *vp;
901 	off_t limit;
902 {
903 	struct ucred *cred = td->td_ucred;
904 	int error = 0;
905 	struct sseg_closure seginfo;
906 	void *hdr;
907 	size_t hdrsize;
908 
909 	/* Size the program segments. */
910 	seginfo.count = 0;
911 	seginfo.size = 0;
912 	each_writable_segment(td, cb_size_segment, &seginfo);
913 
914 	/*
915 	 * Calculate the size of the core file header area by making
916 	 * a dry run of generating it.  Nothing is written, but the
917 	 * size is calculated.
918 	 */
919 	hdrsize = 0;
920 	__elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
921 
922 	if (hdrsize + seginfo.size >= limit)
923 		return (EFAULT);
924 
925 	/*
926 	 * Allocate memory for building the header, fill it up,
927 	 * and write it out.
928 	 */
929 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
930 	if (hdr == NULL) {
931 		return (EINVAL);
932 	}
933 	error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
934 
935 	/* Write the contents of all of the writable segments. */
936 	if (error == 0) {
937 		Elf_Phdr *php;
938 		off_t offset;
939 		int i;
940 
941 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
942 		offset = hdrsize;
943 		for (i = 0; i < seginfo.count; i++) {
944 			error = vn_rdwr_inchunks(UIO_WRITE, vp,
945 			    (caddr_t)(uintptr_t)php->p_vaddr,
946 			    php->p_filesz, offset, UIO_USERSPACE,
947 			    IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
948 			    curthread); /* XXXKSE */
949 			if (error != 0)
950 				break;
951 			offset += php->p_filesz;
952 			php++;
953 		}
954 	}
955 	free(hdr, M_TEMP);
956 
957 	return (error);
958 }
959 
960 /*
961  * A callback for each_writable_segment() to write out the segment's
962  * program header entry.
963  */
964 static void
965 cb_put_phdr(entry, closure)
966 	vm_map_entry_t entry;
967 	void *closure;
968 {
969 	struct phdr_closure *phc = (struct phdr_closure *)closure;
970 	Elf_Phdr *phdr = phc->phdr;
971 
972 	phc->offset = round_page(phc->offset);
973 
974 	phdr->p_type = PT_LOAD;
975 	phdr->p_offset = phc->offset;
976 	phdr->p_vaddr = entry->start;
977 	phdr->p_paddr = 0;
978 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
979 	phdr->p_align = PAGE_SIZE;
980 	phdr->p_flags = 0;
981 	if (entry->protection & VM_PROT_READ)
982 		phdr->p_flags |= PF_R;
983 	if (entry->protection & VM_PROT_WRITE)
984 		phdr->p_flags |= PF_W;
985 	if (entry->protection & VM_PROT_EXECUTE)
986 		phdr->p_flags |= PF_X;
987 
988 	phc->offset += phdr->p_filesz;
989 	phc->phdr++;
990 }
991 
992 /*
993  * A callback for each_writable_segment() to gather information about
994  * the number of segments and their total size.
995  */
996 static void
997 cb_size_segment(entry, closure)
998 	vm_map_entry_t entry;
999 	void *closure;
1000 {
1001 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1002 
1003 	ssc->count++;
1004 	ssc->size += entry->end - entry->start;
1005 }
1006 
1007 /*
1008  * For each writable segment in the process's memory map, call the given
1009  * function with a pointer to the map entry and some arbitrary
1010  * caller-supplied data.
1011  */
1012 static void
1013 each_writable_segment(td, func, closure)
1014 	struct thread *td;
1015 	segment_callback func;
1016 	void *closure;
1017 {
1018 	struct proc *p = td->td_proc;
1019 	vm_map_t map = &p->p_vmspace->vm_map;
1020 	vm_map_entry_t entry;
1021 	vm_object_t backing_object, object;
1022 	boolean_t ignore_entry;
1023 
1024 	vm_map_lock_read(map);
1025 	for (entry = map->header.next; entry != &map->header;
1026 	    entry = entry->next) {
1027 		/*
1028 		 * Don't dump inaccessible mappings, deal with legacy
1029 		 * coredump mode.
1030 		 *
1031 		 * Note that read-only segments related to the elf binary
1032 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1033 		 * need to arbitrarily ignore such segments.
1034 		 */
1035 		if (elf_legacy_coredump) {
1036 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1037 				continue;
1038 		} else {
1039 			if ((entry->protection & VM_PROT_ALL) == 0)
1040 				continue;
1041 		}
1042 
1043 		/*
1044 		 * Dont include memory segment in the coredump if
1045 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1046 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1047 		 * kernel map).
1048 		 */
1049 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1050 			continue;
1051 
1052 		if ((object = entry->object.vm_object) == NULL)
1053 			continue;
1054 
1055 		/* Ignore memory-mapped devices and such things. */
1056 		VM_OBJECT_LOCK(object);
1057 		while ((backing_object = object->backing_object) != NULL) {
1058 			VM_OBJECT_LOCK(backing_object);
1059 			VM_OBJECT_UNLOCK(object);
1060 			object = backing_object;
1061 		}
1062 		ignore_entry = object->type != OBJT_DEFAULT &&
1063 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1064 		VM_OBJECT_UNLOCK(object);
1065 		if (ignore_entry)
1066 			continue;
1067 
1068 		(*func)(entry, closure);
1069 	}
1070 	vm_map_unlock_read(map);
1071 }
1072 
1073 /*
1074  * Write the core file header to the file, including padding up to
1075  * the page boundary.
1076  */
1077 static int
1078 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1079 	struct thread *td;
1080 	struct vnode *vp;
1081 	struct ucred *cred;
1082 	int numsegs;
1083 	size_t hdrsize;
1084 	void *hdr;
1085 {
1086 	size_t off;
1087 
1088 	/* Fill in the header. */
1089 	bzero(hdr, hdrsize);
1090 	off = 0;
1091 	__elfN(puthdr)(td, hdr, &off, numsegs);
1092 
1093 	/* Write it to the core file. */
1094 	return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1095 	    UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1096 	    td)); /* XXXKSE */
1097 }
1098 
1099 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1100 typedef struct prstatus32 elf_prstatus_t;
1101 typedef struct prpsinfo32 elf_prpsinfo_t;
1102 typedef struct fpreg32 elf_prfpregset_t;
1103 typedef struct fpreg32 elf_fpregset_t;
1104 typedef struct reg32 elf_gregset_t;
1105 #else
1106 typedef prstatus_t elf_prstatus_t;
1107 typedef prpsinfo_t elf_prpsinfo_t;
1108 typedef prfpregset_t elf_prfpregset_t;
1109 typedef prfpregset_t elf_fpregset_t;
1110 typedef gregset_t elf_gregset_t;
1111 #endif
1112 
1113 static void
1114 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1115 {
1116 	struct {
1117 		elf_prstatus_t status;
1118 		elf_prfpregset_t fpregset;
1119 		elf_prpsinfo_t psinfo;
1120 	} *tempdata;
1121 	elf_prstatus_t *status;
1122 	elf_prfpregset_t *fpregset;
1123 	elf_prpsinfo_t *psinfo;
1124 	struct proc *p;
1125 	struct thread *thr;
1126 	size_t ehoff, noteoff, notesz, phoff;
1127 
1128 	p = td->td_proc;
1129 
1130 	ehoff = *off;
1131 	*off += sizeof(Elf_Ehdr);
1132 
1133 	phoff = *off;
1134 	*off += (numsegs + 1) * sizeof(Elf_Phdr);
1135 
1136 	noteoff = *off;
1137 	/*
1138 	 * Don't allocate space for the notes if we're just calculating
1139 	 * the size of the header. We also don't collect the data.
1140 	 */
1141 	if (dst != NULL) {
1142 		tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1143 		status = &tempdata->status;
1144 		fpregset = &tempdata->fpregset;
1145 		psinfo = &tempdata->psinfo;
1146 	} else {
1147 		tempdata = NULL;
1148 		status = NULL;
1149 		fpregset = NULL;
1150 		psinfo = NULL;
1151 	}
1152 
1153 	if (dst != NULL) {
1154 		psinfo->pr_version = PRPSINFO_VERSION;
1155 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1156 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1157 		/*
1158 		 * XXX - We don't fill in the command line arguments properly
1159 		 * yet.
1160 		 */
1161 		strlcpy(psinfo->pr_psargs, p->p_comm,
1162 		    sizeof(psinfo->pr_psargs));
1163 	}
1164 	__elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1165 	    sizeof *psinfo);
1166 
1167 	/*
1168 	 * To have the debugger select the right thread (LWP) as the initial
1169 	 * thread, we dump the state of the thread passed to us in td first.
1170 	 * This is the thread that causes the core dump and thus likely to
1171 	 * be the right thread one wants to have selected in the debugger.
1172 	 */
1173 	thr = td;
1174 	while (thr != NULL) {
1175 		if (dst != NULL) {
1176 			status->pr_version = PRSTATUS_VERSION;
1177 			status->pr_statussz = sizeof(elf_prstatus_t);
1178 			status->pr_gregsetsz = sizeof(elf_gregset_t);
1179 			status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1180 			status->pr_osreldate = osreldate;
1181 			status->pr_cursig = p->p_sig;
1182 			status->pr_pid = thr->td_tid;
1183 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1184 			fill_regs32(thr, &status->pr_reg);
1185 			fill_fpregs32(thr, fpregset);
1186 #else
1187 			fill_regs(thr, &status->pr_reg);
1188 			fill_fpregs(thr, fpregset);
1189 #endif
1190 		}
1191 		__elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1192 		    sizeof *status);
1193 		__elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1194 		    sizeof *fpregset);
1195 		/*
1196 		 * Allow for MD specific notes, as well as any MD
1197 		 * specific preparations for writing MI notes.
1198 		 */
1199 		__elfN(dump_thread)(thr, dst, off);
1200 
1201 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1202 		    TAILQ_NEXT(thr, td_plist);
1203 		if (thr == td)
1204 			thr = TAILQ_NEXT(thr, td_plist);
1205 	}
1206 
1207 	notesz = *off - noteoff;
1208 
1209 	if (dst != NULL)
1210 		free(tempdata, M_TEMP);
1211 
1212 	/* Align up to a page boundary for the program segments. */
1213 	*off = round_page(*off);
1214 
1215 	if (dst != NULL) {
1216 		Elf_Ehdr *ehdr;
1217 		Elf_Phdr *phdr;
1218 		struct phdr_closure phc;
1219 
1220 		/*
1221 		 * Fill in the ELF header.
1222 		 */
1223 		ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1224 		ehdr->e_ident[EI_MAG0] = ELFMAG0;
1225 		ehdr->e_ident[EI_MAG1] = ELFMAG1;
1226 		ehdr->e_ident[EI_MAG2] = ELFMAG2;
1227 		ehdr->e_ident[EI_MAG3] = ELFMAG3;
1228 		ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1229 		ehdr->e_ident[EI_DATA] = ELF_DATA;
1230 		ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1231 		ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1232 		ehdr->e_ident[EI_ABIVERSION] = 0;
1233 		ehdr->e_ident[EI_PAD] = 0;
1234 		ehdr->e_type = ET_CORE;
1235 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1236 		ehdr->e_machine = EM_386;
1237 #else
1238 		ehdr->e_machine = ELF_ARCH;
1239 #endif
1240 		ehdr->e_version = EV_CURRENT;
1241 		ehdr->e_entry = 0;
1242 		ehdr->e_phoff = phoff;
1243 		ehdr->e_flags = 0;
1244 		ehdr->e_ehsize = sizeof(Elf_Ehdr);
1245 		ehdr->e_phentsize = sizeof(Elf_Phdr);
1246 		ehdr->e_phnum = numsegs + 1;
1247 		ehdr->e_shentsize = sizeof(Elf_Shdr);
1248 		ehdr->e_shnum = 0;
1249 		ehdr->e_shstrndx = SHN_UNDEF;
1250 
1251 		/*
1252 		 * Fill in the program header entries.
1253 		 */
1254 		phdr = (Elf_Phdr *)((char *)dst + phoff);
1255 
1256 		/* The note segement. */
1257 		phdr->p_type = PT_NOTE;
1258 		phdr->p_offset = noteoff;
1259 		phdr->p_vaddr = 0;
1260 		phdr->p_paddr = 0;
1261 		phdr->p_filesz = notesz;
1262 		phdr->p_memsz = 0;
1263 		phdr->p_flags = 0;
1264 		phdr->p_align = 0;
1265 		phdr++;
1266 
1267 		/* All the writable segments from the program. */
1268 		phc.phdr = phdr;
1269 		phc.offset = *off;
1270 		each_writable_segment(td, cb_put_phdr, &phc);
1271 	}
1272 }
1273 
1274 static void
1275 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1276     const void *desc, size_t descsz)
1277 {
1278 	Elf_Note note;
1279 
1280 	note.n_namesz = strlen(name) + 1;
1281 	note.n_descsz = descsz;
1282 	note.n_type = type;
1283 	if (dst != NULL)
1284 		bcopy(&note, (char *)dst + *off, sizeof note);
1285 	*off += sizeof note;
1286 	if (dst != NULL)
1287 		bcopy(name, (char *)dst + *off, note.n_namesz);
1288 	*off += roundup2(note.n_namesz, sizeof(Elf_Size));
1289 	if (dst != NULL)
1290 		bcopy(desc, (char *)dst + *off, note.n_descsz);
1291 	*off += roundup2(note.n_descsz, sizeof(Elf_Size));
1292 }
1293 
1294 /*
1295  * Tell kern_execve.c about it, with a little help from the linker.
1296  */
1297 static struct execsw __elfN(execsw) = {
1298 	__CONCAT(exec_, __elfN(imgact)),
1299 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1300 };
1301 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
1302