xref: /freebsd/sys/kern/imgact_elf.c (revision 9336e0699bda8a301cd2bfa37106b6ec5e32012e)
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 #define	trunc_page_ps(va, ps)	((va) & ~(ps - 1))
110 #define	round_page_ps(va, ps)	(((va) + (ps - 1)) & ~(ps - 1))
111 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
112 
113 int
114 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
115 {
116 	int i;
117 
118 	for (i = 0; i < MAX_BRANDS; i++) {
119 		if (elf_brand_list[i] == NULL) {
120 			elf_brand_list[i] = entry;
121 			break;
122 		}
123 	}
124 	if (i == MAX_BRANDS)
125 		return (-1);
126 	return (0);
127 }
128 
129 int
130 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
131 {
132 	int i;
133 
134 	for (i = 0; i < MAX_BRANDS; i++) {
135 		if (elf_brand_list[i] == entry) {
136 			elf_brand_list[i] = NULL;
137 			break;
138 		}
139 	}
140 	if (i == MAX_BRANDS)
141 		return (-1);
142 	return (0);
143 }
144 
145 int
146 __elfN(brand_inuse)(Elf_Brandinfo *entry)
147 {
148 	struct proc *p;
149 	int rval = FALSE;
150 
151 	sx_slock(&allproc_lock);
152 	FOREACH_PROC_IN_SYSTEM(p) {
153 		if (p->p_sysent == entry->sysvec) {
154 			rval = TRUE;
155 			break;
156 		}
157 	}
158 	sx_sunlock(&allproc_lock);
159 
160 	return (rval);
161 }
162 
163 static Elf_Brandinfo *
164 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
165 {
166 	Elf_Brandinfo *bi;
167 	int i;
168 
169 	/*
170 	 * We support three types of branding -- (1) the ELF EI_OSABI field
171 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
172 	 * branding w/in the ELF header, and (3) path of the `interp_path'
173 	 * field.  We should also look for an ".note.ABI-tag" ELF section now
174 	 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
175 	 */
176 
177 	/* If the executable has a brand, search for it in the brand list. */
178 	for (i = 0; i < MAX_BRANDS; i++) {
179 		bi = elf_brand_list[i];
180 		if (bi != NULL && hdr->e_machine == bi->machine &&
181 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
182 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
183 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
184 			return (bi);
185 	}
186 
187 	/* Lacking a known brand, search for a recognized interpreter. */
188 	if (interp != NULL) {
189 		for (i = 0; i < MAX_BRANDS; i++) {
190 			bi = elf_brand_list[i];
191 			if (bi != NULL && hdr->e_machine == bi->machine &&
192 			    strcmp(interp, bi->interp_path) == 0)
193 				return (bi);
194 		}
195 	}
196 
197 	/* Lacking a recognized interpreter, try the default brand */
198 	for (i = 0; i < MAX_BRANDS; i++) {
199 		bi = elf_brand_list[i];
200 		if (bi != NULL && hdr->e_machine == bi->machine &&
201 		    __elfN(fallback_brand) == bi->brand)
202 			return (bi);
203 	}
204 	return (NULL);
205 }
206 
207 static int
208 __elfN(check_header)(const Elf_Ehdr *hdr)
209 {
210 	Elf_Brandinfo *bi;
211 	int i;
212 
213 	if (!IS_ELF(*hdr) ||
214 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
215 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
216 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
217 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
218 	    hdr->e_version != ELF_TARG_VER)
219 		return (ENOEXEC);
220 
221 	/*
222 	 * Make sure we have at least one brand for this machine.
223 	 */
224 
225 	for (i = 0; i < MAX_BRANDS; i++) {
226 		bi = elf_brand_list[i];
227 		if (bi != NULL && bi->machine == hdr->e_machine)
228 			break;
229 	}
230 	if (i == MAX_BRANDS)
231 		return (ENOEXEC);
232 
233 	return (0);
234 }
235 
236 static int
237 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
238     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
239 {
240 	struct sf_buf *sf;
241 	int error;
242 	vm_offset_t off;
243 
244 	/*
245 	 * Create the page if it doesn't exist yet. Ignore errors.
246 	 */
247 	vm_map_lock(map);
248 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
249 	    VM_PROT_ALL, VM_PROT_ALL, 0);
250 	vm_map_unlock(map);
251 
252 	/*
253 	 * Find the page from the underlying object.
254 	 */
255 	if (object) {
256 		sf = vm_imgact_map_page(object, offset);
257 		if (sf == NULL)
258 			return (KERN_FAILURE);
259 		off = offset - trunc_page(offset);
260 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
261 		    end - start);
262 		vm_imgact_unmap_page(sf);
263 		if (error) {
264 			return (KERN_FAILURE);
265 		}
266 	}
267 
268 	return (KERN_SUCCESS);
269 }
270 
271 static int
272 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
273     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
274 {
275 	struct sf_buf *sf;
276 	vm_offset_t off;
277 	vm_size_t sz;
278 	int error, rv;
279 
280 	if (start != trunc_page(start)) {
281 		rv = __elfN(map_partial)(map, object, offset, start,
282 		    round_page(start), prot);
283 		if (rv)
284 			return (rv);
285 		offset += round_page(start) - start;
286 		start = round_page(start);
287 	}
288 	if (end != round_page(end)) {
289 		rv = __elfN(map_partial)(map, object, offset +
290 		    trunc_page(end) - start, trunc_page(end), end, prot);
291 		if (rv)
292 			return (rv);
293 		end = trunc_page(end);
294 	}
295 	if (end > start) {
296 		if (offset & PAGE_MASK) {
297 			/*
298 			 * The mapping is not page aligned. This means we have
299 			 * to copy the data. Sigh.
300 			 */
301 			rv = vm_map_find(map, NULL, 0, &start, end - start,
302 			    FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
303 			if (rv)
304 				return (rv);
305 			if (object == NULL)
306 				return (KERN_SUCCESS);
307 			for (; start < end; start += sz) {
308 				sf = vm_imgact_map_page(object, offset);
309 				if (sf == NULL)
310 					return (KERN_FAILURE);
311 				off = offset - trunc_page(offset);
312 				sz = end - start;
313 				if (sz > PAGE_SIZE - off)
314 					sz = PAGE_SIZE - off;
315 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
316 				    (caddr_t)start, sz);
317 				vm_imgact_unmap_page(sf);
318 				if (error) {
319 					return (KERN_FAILURE);
320 				}
321 				offset += sz;
322 			}
323 			rv = KERN_SUCCESS;
324 		} else {
325 			vm_object_reference(object);
326 			vm_map_lock(map);
327 			rv = vm_map_insert(map, object, offset, start, end,
328 			    prot, VM_PROT_ALL, cow);
329 			vm_map_unlock(map);
330 			if (rv != KERN_SUCCESS)
331 				vm_object_deallocate(object);
332 		}
333 		return (rv);
334 	} else {
335 		return (KERN_SUCCESS);
336 	}
337 }
338 
339 static int
340 __elfN(load_section)(struct vmspace *vmspace,
341 	vm_object_t object, vm_offset_t offset,
342 	caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
343 	size_t pagesize)
344 {
345 	struct sf_buf *sf;
346 	size_t map_len;
347 	vm_offset_t map_addr;
348 	int error, rv, cow;
349 	size_t copy_len;
350 	vm_offset_t file_addr;
351 
352 	/*
353 	 * It's necessary to fail if the filsz + offset taken from the
354 	 * header is greater than the actual file pager object's size.
355 	 * If we were to allow this, then the vm_map_find() below would
356 	 * walk right off the end of the file object and into the ether.
357 	 *
358 	 * While I'm here, might as well check for something else that
359 	 * is invalid: filsz cannot be greater than memsz.
360 	 */
361 	if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
362 	    filsz > memsz) {
363 		uprintf("elf_load_section: truncated ELF file\n");
364 		return (ENOEXEC);
365 	}
366 
367 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
368 	file_addr = trunc_page_ps(offset, pagesize);
369 
370 	/*
371 	 * We have two choices.  We can either clear the data in the last page
372 	 * of an oversized mapping, or we can start the anon mapping a page
373 	 * early and copy the initialized data into that first page.  We
374 	 * choose the second..
375 	 */
376 	if (memsz > filsz)
377 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
378 	else
379 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
380 
381 	if (map_len != 0) {
382 		/* cow flags: don't dump readonly sections in core */
383 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
384 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
385 
386 		rv = __elfN(map_insert)(&vmspace->vm_map,
387 				      object,
388 				      file_addr,	/* file offset */
389 				      map_addr,		/* virtual start */
390 				      map_addr + map_len,/* virtual end */
391 				      prot,
392 				      cow);
393 		if (rv != KERN_SUCCESS)
394 			return (EINVAL);
395 
396 		/* we can stop now if we've covered it all */
397 		if (memsz == filsz) {
398 			return (0);
399 		}
400 	}
401 
402 
403 	/*
404 	 * We have to get the remaining bit of the file into the first part
405 	 * of the oversized map segment.  This is normally because the .data
406 	 * segment in the file is extended to provide bss.  It's a neat idea
407 	 * to try and save a page, but it's a pain in the behind to implement.
408 	 */
409 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
410 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
411 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
412 	    map_addr;
413 
414 	/* This had damn well better be true! */
415 	if (map_len != 0) {
416 		rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
417 		    map_addr + map_len, VM_PROT_ALL, 0);
418 		if (rv != KERN_SUCCESS) {
419 			return (EINVAL);
420 		}
421 	}
422 
423 	if (copy_len != 0) {
424 		vm_offset_t off;
425 
426 		sf = vm_imgact_map_page(object, offset + filsz);
427 		if (sf == NULL)
428 			return (EIO);
429 
430 		/* send the page fragment to user space */
431 		off = trunc_page_ps(offset + filsz, pagesize) -
432 		    trunc_page(offset + filsz);
433 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
434 		    (caddr_t)map_addr, copy_len);
435 		vm_imgact_unmap_page(sf);
436 		if (error) {
437 			return (error);
438 		}
439 	}
440 
441 	/*
442 	 * set it to the specified protection.
443 	 * XXX had better undo the damage from pasting over the cracks here!
444 	 */
445 	vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
446 	    round_page(map_addr + map_len),  prot, FALSE);
447 
448 	return (0);
449 }
450 
451 /*
452  * Load the file "file" into memory.  It may be either a shared object
453  * or an executable.
454  *
455  * The "addr" reference parameter is in/out.  On entry, it specifies
456  * the address where a shared object should be loaded.  If the file is
457  * an executable, this value is ignored.  On exit, "addr" specifies
458  * where the file was actually loaded.
459  *
460  * The "entry" reference parameter is out only.  On exit, it specifies
461  * the entry point for the loaded file.
462  */
463 static int
464 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
465 	u_long *entry, size_t pagesize)
466 {
467 	struct {
468 		struct nameidata nd;
469 		struct vattr attr;
470 		struct image_params image_params;
471 	} *tempdata;
472 	const Elf_Ehdr *hdr = NULL;
473 	const Elf_Phdr *phdr = NULL;
474 	struct nameidata *nd;
475 	struct vmspace *vmspace = p->p_vmspace;
476 	struct vattr *attr;
477 	struct image_params *imgp;
478 	vm_prot_t prot;
479 	u_long rbase;
480 	u_long base_addr = 0;
481 	int vfslocked, error, i, numsegs;
482 
483 	if (curthread->td_proc != p)
484 		panic("elf_load_file - thread");	/* XXXKSE DIAGNOSTIC */
485 
486 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
487 	nd = &tempdata->nd;
488 	attr = &tempdata->attr;
489 	imgp = &tempdata->image_params;
490 
491 	/*
492 	 * Initialize part of the common data
493 	 */
494 	imgp->proc = p;
495 	imgp->attr = attr;
496 	imgp->firstpage = NULL;
497 	imgp->image_header = NULL;
498 	imgp->object = NULL;
499 	imgp->execlabel = NULL;
500 
501 	/* XXXKSE */
502 	NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
503 	    curthread);
504 	vfslocked = 0;
505 	if ((error = namei(nd)) != 0) {
506 		nd->ni_vp = NULL;
507 		goto fail;
508 	}
509 	vfslocked = NDHASGIANT(nd);
510 	NDFREE(nd, NDF_ONLY_PNBUF);
511 	imgp->vp = nd->ni_vp;
512 
513 	/*
514 	 * Check permissions, modes, uid, etc on the file, and "open" it.
515 	 */
516 	error = exec_check_permissions(imgp);
517 	if (error)
518 		goto fail;
519 
520 	error = exec_map_first_page(imgp);
521 	if (error)
522 		goto fail;
523 
524 	/*
525 	 * Also make certain that the interpreter stays the same, so set
526 	 * its VV_TEXT flag, too.
527 	 */
528 	nd->ni_vp->v_vflag |= VV_TEXT;
529 
530 	imgp->object = nd->ni_vp->v_object;
531 
532 	hdr = (const Elf_Ehdr *)imgp->image_header;
533 	if ((error = __elfN(check_header)(hdr)) != 0)
534 		goto fail;
535 	if (hdr->e_type == ET_DYN)
536 		rbase = *addr;
537 	else if (hdr->e_type == ET_EXEC)
538 		rbase = 0;
539 	else {
540 		error = ENOEXEC;
541 		goto fail;
542 	}
543 
544 	/* Only support headers that fit within first page for now      */
545 	/*    (multiplication of two Elf_Half fields will not overflow) */
546 	if ((hdr->e_phoff > PAGE_SIZE) ||
547 	    (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
548 		error = ENOEXEC;
549 		goto fail;
550 	}
551 
552 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
553 	if (!aligned(phdr, Elf_Addr)) {
554 		error = ENOEXEC;
555 		goto fail;
556 	}
557 
558 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
559 		if (phdr[i].p_type == PT_LOAD) {	/* Loadable segment */
560 			prot = 0;
561 			if (phdr[i].p_flags & PF_X)
562   				prot |= VM_PROT_EXECUTE;
563 			if (phdr[i].p_flags & PF_W)
564   				prot |= VM_PROT_WRITE;
565 			if (phdr[i].p_flags & PF_R)
566   				prot |= VM_PROT_READ;
567 
568 			if ((error = __elfN(load_section)(vmspace,
569 			    imgp->object, phdr[i].p_offset,
570 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
571 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
572 			    pagesize)) != 0)
573 				goto fail;
574 			/*
575 			 * Establish the base address if this is the
576 			 * first segment.
577 			 */
578 			if (numsegs == 0)
579   				base_addr = trunc_page(phdr[i].p_vaddr +
580 				    rbase);
581 			numsegs++;
582 		}
583 	}
584 	*addr = base_addr;
585 	*entry = (unsigned long)hdr->e_entry + rbase;
586 
587 fail:
588 	if (imgp->firstpage)
589 		exec_unmap_first_page(imgp);
590 
591 	if (nd->ni_vp)
592 		vput(nd->ni_vp);
593 
594 	VFS_UNLOCK_GIANT(vfslocked);
595 	free(tempdata, M_TEMP);
596 
597 	return (error);
598 }
599 
600 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
601 
602 static int
603 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
604 {
605 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
606 	const Elf_Phdr *phdr, *pnote = NULL;
607 	Elf_Auxargs *elf_auxargs;
608 	struct vmspace *vmspace;
609 	vm_prot_t prot;
610 	u_long text_size = 0, data_size = 0, total_size = 0;
611 	u_long text_addr = 0, data_addr = 0;
612 	u_long seg_size, seg_addr;
613 	u_long addr, entry = 0, proghdr = 0;
614 	int error = 0, i;
615 	const char *interp = NULL;
616 	Elf_Brandinfo *brand_info;
617 	const Elf_Note *note, *note_end;
618 	char *path;
619 	const char *note_name;
620 	struct thread *td = curthread;
621 	struct sysentvec *sv;
622 
623 	/*
624 	 * Do we have a valid ELF header ?
625 	 *
626 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
627 	 * if particular brand doesn't support it.
628 	 */
629 	if (__elfN(check_header)(hdr) != 0 ||
630 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
631 		return (-1);
632 
633 	/*
634 	 * From here on down, we return an errno, not -1, as we've
635 	 * detected an ELF file.
636 	 */
637 
638 	if ((hdr->e_phoff > PAGE_SIZE) ||
639 	    (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
640 		/* Only support headers in first page for now */
641 		return (ENOEXEC);
642 	}
643 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
644 	if (!aligned(phdr, Elf_Addr))
645 		return (ENOEXEC);
646 	for (i = 0; i < hdr->e_phnum; i++) {
647 		if (phdr[i].p_type == PT_INTERP) {
648 			/* Path to interpreter */
649 			if (phdr[i].p_filesz > MAXPATHLEN ||
650 			    phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
651 				return (ENOEXEC);
652 			interp = imgp->image_header + phdr[i].p_offset;
653 			break;
654 		}
655 	}
656 
657 	brand_info = __elfN(get_brandinfo)(hdr, interp);
658 	if (brand_info == NULL) {
659 		uprintf("ELF binary type \"%u\" not known.\n",
660 		    hdr->e_ident[EI_OSABI]);
661 		return (ENOEXEC);
662 	}
663 	if (hdr->e_type == ET_DYN &&
664 	    (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
665 		return (ENOEXEC);
666 	sv = brand_info->sysvec;
667 	if (interp != NULL && brand_info->interp_newpath != NULL)
668 		interp = brand_info->interp_newpath;
669 
670 	/*
671 	 * Avoid a possible deadlock if the current address space is destroyed
672 	 * and that address space maps the locked vnode.  In the common case,
673 	 * the locked vnode's v_usecount is decremented but remains greater
674 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
675 	 * However, in cases where the vnode lock is external, such as nullfs,
676 	 * v_usecount may become zero.
677 	 */
678 	VOP_UNLOCK(imgp->vp, 0, td);
679 
680 	error = exec_new_vmspace(imgp, sv);
681 	imgp->proc->p_sysent = sv;
682 
683 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
684 	if (error)
685 		return (error);
686 
687 	vmspace = imgp->proc->p_vmspace;
688 
689 	for (i = 0; i < hdr->e_phnum; i++) {
690 		switch (phdr[i].p_type) {
691 		case PT_LOAD:	/* Loadable segment */
692 			prot = 0;
693 			if (phdr[i].p_flags & PF_X)
694   				prot |= VM_PROT_EXECUTE;
695 			if (phdr[i].p_flags & PF_W)
696   				prot |= VM_PROT_WRITE;
697 			if (phdr[i].p_flags & PF_R)
698   				prot |= VM_PROT_READ;
699 
700 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
701 			/*
702 			 * Some x86 binaries assume read == executable,
703 			 * notably the M3 runtime and therefore cvsup
704 			 */
705 			if (prot & VM_PROT_READ)
706 				prot |= VM_PROT_EXECUTE;
707 #endif
708 
709 			if ((error = __elfN(load_section)(vmspace,
710 			    imgp->object, phdr[i].p_offset,
711 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr,
712 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
713 			    sv->sv_pagesize)) != 0)
714 				return (error);
715 
716 			/*
717 			 * If this segment contains the program headers,
718 			 * remember their virtual address for the AT_PHDR
719 			 * aux entry. Static binaries don't usually include
720 			 * a PT_PHDR entry.
721 			 */
722 			if (phdr[i].p_offset == 0 &&
723 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
724 				<= phdr[i].p_filesz)
725 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
726 
727 			seg_addr = trunc_page(phdr[i].p_vaddr);
728 			seg_size = round_page(phdr[i].p_memsz +
729 			    phdr[i].p_vaddr - seg_addr);
730 
731 			/*
732 			 * Is this .text or .data?  We can't use
733 			 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
734 			 * alpha terribly and possibly does other bad
735 			 * things so we stick to the old way of figuring
736 			 * it out:  If the segment contains the program
737 			 * entry point, it's a text segment, otherwise it
738 			 * is a data segment.
739 			 *
740 			 * Note that obreak() assumes that data_addr +
741 			 * data_size == end of data load area, and the ELF
742 			 * file format expects segments to be sorted by
743 			 * address.  If multiple data segments exist, the
744 			 * last one will be used.
745 			 */
746 			if (hdr->e_entry >= phdr[i].p_vaddr &&
747 			    hdr->e_entry < (phdr[i].p_vaddr +
748 			    phdr[i].p_memsz)) {
749 				text_size = seg_size;
750 				text_addr = seg_addr;
751 				entry = (u_long)hdr->e_entry;
752 			} else {
753 				data_size = seg_size;
754 				data_addr = seg_addr;
755 			}
756 			total_size += seg_size;
757 			break;
758 		case PT_PHDR: 	/* Program header table info */
759 			proghdr = phdr[i].p_vaddr;
760 			break;
761 		case PT_NOTE:
762 			pnote = &phdr[i];
763 			break;
764 		default:
765 			break;
766 		}
767 	}
768 
769 	if (data_addr == 0 && data_size == 0) {
770 		data_addr = text_addr;
771 		data_size = text_size;
772 	}
773 
774 	/*
775 	 * Check limits.  It should be safe to check the
776 	 * limits after loading the segments since we do
777 	 * not actually fault in all the segments pages.
778 	 */
779 	PROC_LOCK(imgp->proc);
780 	if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
781 	    text_size > maxtsiz ||
782 	    total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
783 		PROC_UNLOCK(imgp->proc);
784 		return (ENOMEM);
785 	}
786 
787 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
788 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
789 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
790 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
791 
792 	/*
793 	 * We load the dynamic linker where a userland call
794 	 * to mmap(0, ...) would put it.  The rationale behind this
795 	 * calculation is that it leaves room for the heap to grow to
796 	 * its maximum allowed size.
797 	 */
798 	addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
799 	    lim_max(imgp->proc, RLIMIT_DATA));
800 	PROC_UNLOCK(imgp->proc);
801 
802 	imgp->entry_addr = entry;
803 
804 	if (interp != NULL) {
805 		VOP_UNLOCK(imgp->vp, 0, td);
806 		if (brand_info->emul_path != NULL &&
807 		    brand_info->emul_path[0] != '\0') {
808 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
809 			snprintf(path, MAXPATHLEN, "%s%s",
810 			    brand_info->emul_path, interp);
811 			error = __elfN(load_file)(imgp->proc, path, &addr,
812 			    &imgp->entry_addr, sv->sv_pagesize);
813 			free(path, M_TEMP);
814 			if (error == 0)
815 				interp = NULL;
816 		}
817 		if (interp != NULL) {
818 			error = __elfN(load_file)(imgp->proc, interp, &addr,
819 			    &imgp->entry_addr, sv->sv_pagesize);
820 		}
821 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
822 		if (error != 0) {
823 			uprintf("ELF interpreter %s not found\n", interp);
824 			return (error);
825 		}
826 	}
827 
828 	/*
829 	 * Construct auxargs table (used by the fixup routine)
830 	 */
831 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
832 	elf_auxargs->execfd = -1;
833 	elf_auxargs->phdr = proghdr;
834 	elf_auxargs->phent = hdr->e_phentsize;
835 	elf_auxargs->phnum = hdr->e_phnum;
836 	elf_auxargs->pagesz = PAGE_SIZE;
837 	elf_auxargs->base = addr;
838 	elf_auxargs->flags = 0;
839 	elf_auxargs->entry = entry;
840 	elf_auxargs->trace = elf_trace;
841 
842 	imgp->auxargs = elf_auxargs;
843 	imgp->interpreted = 0;
844 
845 	/*
846 	 * Try to fetch the osreldate for FreeBSD binary from the ELF
847 	 * OSABI-note. Only the first page of the image is searched,
848 	 * the same as for headers.
849 	 */
850 	if (pnote != NULL && pnote->p_offset < PAGE_SIZE &&
851 	    pnote->p_offset + pnote->p_filesz < PAGE_SIZE ) {
852 		note = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
853 		if (!aligned(note, Elf32_Addr)) {
854 			free(imgp->auxargs, M_TEMP);
855 			imgp->auxargs = NULL;
856 			return (ENOEXEC);
857 		}
858 		note_end = (const Elf_Note *)(imgp->image_header + pnote->p_offset +
859 		    pnote->p_filesz);
860 		while (note < note_end) {
861 			if (note->n_namesz == sizeof(FREEBSD_ABI_VENDOR) &&
862 			    note->n_descsz == sizeof(int32_t) &&
863 			    note->n_type == 1 /* ABI_NOTETYPE */) {
864 				note_name = (const char *)(note + 1);
865 				if (strncmp(FREEBSD_ABI_VENDOR, note_name,
866 				    sizeof(FREEBSD_ABI_VENDOR)) == 0) {
867 					imgp->proc->p_osrel = *(const int32_t *)
868 					    (note_name +
869 					    round_page_ps(sizeof(FREEBSD_ABI_VENDOR),
870 						sizeof(Elf32_Addr)));
871 					break;
872 				}
873 			}
874 			note = (const Elf_Note *)((const char *)(note + 1) +
875 			    round_page_ps(note->n_namesz, sizeof(Elf32_Addr)) +
876 			    round_page_ps(note->n_descsz, sizeof(Elf32_Addr)));
877 		}
878 	}
879 
880 	return (error);
881 }
882 
883 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
884 
885 int
886 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
887 {
888 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
889 	Elf_Addr *base;
890 	Elf_Addr *pos;
891 
892 	base = (Elf_Addr *)*stack_base;
893 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
894 
895 	if (args->trace) {
896 		AUXARGS_ENTRY(pos, AT_DEBUG, 1);
897 	}
898 	if (args->execfd != -1) {
899 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
900 	}
901 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
902 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
903 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
904 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
905 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
906 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
907 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
908 	AUXARGS_ENTRY(pos, AT_NULL, 0);
909 
910 	free(imgp->auxargs, M_TEMP);
911 	imgp->auxargs = NULL;
912 
913 	base--;
914 	suword(base, (long)imgp->args->argc);
915 	*stack_base = (register_t *)base;
916 	return (0);
917 }
918 
919 /*
920  * Code for generating ELF core dumps.
921  */
922 
923 typedef void (*segment_callback)(vm_map_entry_t, void *);
924 
925 /* Closure for cb_put_phdr(). */
926 struct phdr_closure {
927 	Elf_Phdr *phdr;		/* Program header to fill in */
928 	Elf_Off offset;		/* Offset of segment in core file */
929 };
930 
931 /* Closure for cb_size_segment(). */
932 struct sseg_closure {
933 	int count;		/* Count of writable segments. */
934 	size_t size;		/* Total size of all writable segments. */
935 };
936 
937 static void cb_put_phdr(vm_map_entry_t, void *);
938 static void cb_size_segment(vm_map_entry_t, void *);
939 static void each_writable_segment(struct thread *, segment_callback, void *);
940 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
941     int, void *, size_t);
942 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
943 static void __elfN(putnote)(void *, size_t *, const char *, int,
944     const void *, size_t);
945 
946 int
947 __elfN(coredump)(td, vp, limit)
948 	struct thread *td;
949 	struct vnode *vp;
950 	off_t limit;
951 {
952 	struct ucred *cred = td->td_ucred;
953 	int error = 0;
954 	struct sseg_closure seginfo;
955 	void *hdr;
956 	size_t hdrsize;
957 
958 	/* Size the program segments. */
959 	seginfo.count = 0;
960 	seginfo.size = 0;
961 	each_writable_segment(td, cb_size_segment, &seginfo);
962 
963 	/*
964 	 * Calculate the size of the core file header area by making
965 	 * a dry run of generating it.  Nothing is written, but the
966 	 * size is calculated.
967 	 */
968 	hdrsize = 0;
969 	__elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
970 
971 	if (hdrsize + seginfo.size >= limit)
972 		return (EFAULT);
973 
974 	/*
975 	 * Allocate memory for building the header, fill it up,
976 	 * and write it out.
977 	 */
978 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
979 	if (hdr == NULL) {
980 		return (EINVAL);
981 	}
982 	error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
983 
984 	/* Write the contents of all of the writable segments. */
985 	if (error == 0) {
986 		Elf_Phdr *php;
987 		off_t offset;
988 		int i;
989 
990 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
991 		offset = hdrsize;
992 		for (i = 0; i < seginfo.count; i++) {
993 			error = vn_rdwr_inchunks(UIO_WRITE, vp,
994 			    (caddr_t)(uintptr_t)php->p_vaddr,
995 			    php->p_filesz, offset, UIO_USERSPACE,
996 			    IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
997 			    curthread); /* XXXKSE */
998 			if (error != 0)
999 				break;
1000 			offset += php->p_filesz;
1001 			php++;
1002 		}
1003 	}
1004 	free(hdr, M_TEMP);
1005 
1006 	return (error);
1007 }
1008 
1009 /*
1010  * A callback for each_writable_segment() to write out the segment's
1011  * program header entry.
1012  */
1013 static void
1014 cb_put_phdr(entry, closure)
1015 	vm_map_entry_t entry;
1016 	void *closure;
1017 {
1018 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1019 	Elf_Phdr *phdr = phc->phdr;
1020 
1021 	phc->offset = round_page(phc->offset);
1022 
1023 	phdr->p_type = PT_LOAD;
1024 	phdr->p_offset = phc->offset;
1025 	phdr->p_vaddr = entry->start;
1026 	phdr->p_paddr = 0;
1027 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1028 	phdr->p_align = PAGE_SIZE;
1029 	phdr->p_flags = 0;
1030 	if (entry->protection & VM_PROT_READ)
1031 		phdr->p_flags |= PF_R;
1032 	if (entry->protection & VM_PROT_WRITE)
1033 		phdr->p_flags |= PF_W;
1034 	if (entry->protection & VM_PROT_EXECUTE)
1035 		phdr->p_flags |= PF_X;
1036 
1037 	phc->offset += phdr->p_filesz;
1038 	phc->phdr++;
1039 }
1040 
1041 /*
1042  * A callback for each_writable_segment() to gather information about
1043  * the number of segments and their total size.
1044  */
1045 static void
1046 cb_size_segment(entry, closure)
1047 	vm_map_entry_t entry;
1048 	void *closure;
1049 {
1050 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1051 
1052 	ssc->count++;
1053 	ssc->size += entry->end - entry->start;
1054 }
1055 
1056 /*
1057  * For each writable segment in the process's memory map, call the given
1058  * function with a pointer to the map entry and some arbitrary
1059  * caller-supplied data.
1060  */
1061 static void
1062 each_writable_segment(td, func, closure)
1063 	struct thread *td;
1064 	segment_callback func;
1065 	void *closure;
1066 {
1067 	struct proc *p = td->td_proc;
1068 	vm_map_t map = &p->p_vmspace->vm_map;
1069 	vm_map_entry_t entry;
1070 	vm_object_t backing_object, object;
1071 	boolean_t ignore_entry;
1072 
1073 	vm_map_lock_read(map);
1074 	for (entry = map->header.next; entry != &map->header;
1075 	    entry = entry->next) {
1076 		/*
1077 		 * Don't dump inaccessible mappings, deal with legacy
1078 		 * coredump mode.
1079 		 *
1080 		 * Note that read-only segments related to the elf binary
1081 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1082 		 * need to arbitrarily ignore such segments.
1083 		 */
1084 		if (elf_legacy_coredump) {
1085 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1086 				continue;
1087 		} else {
1088 			if ((entry->protection & VM_PROT_ALL) == 0)
1089 				continue;
1090 		}
1091 
1092 		/*
1093 		 * Dont include memory segment in the coredump if
1094 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1095 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1096 		 * kernel map).
1097 		 */
1098 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1099 			continue;
1100 
1101 		if ((object = entry->object.vm_object) == NULL)
1102 			continue;
1103 
1104 		/* Ignore memory-mapped devices and such things. */
1105 		VM_OBJECT_LOCK(object);
1106 		while ((backing_object = object->backing_object) != NULL) {
1107 			VM_OBJECT_LOCK(backing_object);
1108 			VM_OBJECT_UNLOCK(object);
1109 			object = backing_object;
1110 		}
1111 		ignore_entry = object->type != OBJT_DEFAULT &&
1112 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1113 		VM_OBJECT_UNLOCK(object);
1114 		if (ignore_entry)
1115 			continue;
1116 
1117 		(*func)(entry, closure);
1118 	}
1119 	vm_map_unlock_read(map);
1120 }
1121 
1122 /*
1123  * Write the core file header to the file, including padding up to
1124  * the page boundary.
1125  */
1126 static int
1127 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1128 	struct thread *td;
1129 	struct vnode *vp;
1130 	struct ucred *cred;
1131 	int numsegs;
1132 	size_t hdrsize;
1133 	void *hdr;
1134 {
1135 	size_t off;
1136 
1137 	/* Fill in the header. */
1138 	bzero(hdr, hdrsize);
1139 	off = 0;
1140 	__elfN(puthdr)(td, hdr, &off, numsegs);
1141 
1142 	/* Write it to the core file. */
1143 	return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1144 	    UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1145 	    td)); /* XXXKSE */
1146 }
1147 
1148 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1149 typedef struct prstatus32 elf_prstatus_t;
1150 typedef struct prpsinfo32 elf_prpsinfo_t;
1151 typedef struct fpreg32 elf_prfpregset_t;
1152 typedef struct fpreg32 elf_fpregset_t;
1153 typedef struct reg32 elf_gregset_t;
1154 #else
1155 typedef prstatus_t elf_prstatus_t;
1156 typedef prpsinfo_t elf_prpsinfo_t;
1157 typedef prfpregset_t elf_prfpregset_t;
1158 typedef prfpregset_t elf_fpregset_t;
1159 typedef gregset_t elf_gregset_t;
1160 #endif
1161 
1162 static void
1163 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1164 {
1165 	struct {
1166 		elf_prstatus_t status;
1167 		elf_prfpregset_t fpregset;
1168 		elf_prpsinfo_t psinfo;
1169 	} *tempdata;
1170 	elf_prstatus_t *status;
1171 	elf_prfpregset_t *fpregset;
1172 	elf_prpsinfo_t *psinfo;
1173 	struct proc *p;
1174 	struct thread *thr;
1175 	size_t ehoff, noteoff, notesz, phoff;
1176 
1177 	p = td->td_proc;
1178 
1179 	ehoff = *off;
1180 	*off += sizeof(Elf_Ehdr);
1181 
1182 	phoff = *off;
1183 	*off += (numsegs + 1) * sizeof(Elf_Phdr);
1184 
1185 	noteoff = *off;
1186 	/*
1187 	 * Don't allocate space for the notes if we're just calculating
1188 	 * the size of the header. We also don't collect the data.
1189 	 */
1190 	if (dst != NULL) {
1191 		tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1192 		status = &tempdata->status;
1193 		fpregset = &tempdata->fpregset;
1194 		psinfo = &tempdata->psinfo;
1195 	} else {
1196 		tempdata = NULL;
1197 		status = NULL;
1198 		fpregset = NULL;
1199 		psinfo = NULL;
1200 	}
1201 
1202 	if (dst != NULL) {
1203 		psinfo->pr_version = PRPSINFO_VERSION;
1204 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1205 		strlcpy(psinfo->pr_fname, td->td_name, sizeof(psinfo->pr_fname));
1206 		/*
1207 		 * XXX - We don't fill in the command line arguments properly
1208 		 * yet.
1209 		 */
1210 		strlcpy(psinfo->pr_psargs, td->td_name,
1211 		    sizeof(psinfo->pr_psargs));
1212 	}
1213 	__elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1214 	    sizeof *psinfo);
1215 
1216 	/*
1217 	 * To have the debugger select the right thread (LWP) as the initial
1218 	 * thread, we dump the state of the thread passed to us in td first.
1219 	 * This is the thread that causes the core dump and thus likely to
1220 	 * be the right thread one wants to have selected in the debugger.
1221 	 */
1222 	thr = td;
1223 	while (thr != NULL) {
1224 		if (dst != NULL) {
1225 			status->pr_version = PRSTATUS_VERSION;
1226 			status->pr_statussz = sizeof(elf_prstatus_t);
1227 			status->pr_gregsetsz = sizeof(elf_gregset_t);
1228 			status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1229 			status->pr_osreldate = osreldate;
1230 			status->pr_cursig = p->p_sig;
1231 			status->pr_pid = thr->td_tid;
1232 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1233 			fill_regs32(thr, &status->pr_reg);
1234 			fill_fpregs32(thr, fpregset);
1235 #else
1236 			fill_regs(thr, &status->pr_reg);
1237 			fill_fpregs(thr, fpregset);
1238 #endif
1239 		}
1240 		__elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1241 		    sizeof *status);
1242 		__elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1243 		    sizeof *fpregset);
1244 		/*
1245 		 * Allow for MD specific notes, as well as any MD
1246 		 * specific preparations for writing MI notes.
1247 		 */
1248 		__elfN(dump_thread)(thr, dst, off);
1249 
1250 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1251 		    TAILQ_NEXT(thr, td_plist);
1252 		if (thr == td)
1253 			thr = TAILQ_NEXT(thr, td_plist);
1254 	}
1255 
1256 	notesz = *off - noteoff;
1257 
1258 	if (dst != NULL)
1259 		free(tempdata, M_TEMP);
1260 
1261 	/* Align up to a page boundary for the program segments. */
1262 	*off = round_page(*off);
1263 
1264 	if (dst != NULL) {
1265 		Elf_Ehdr *ehdr;
1266 		Elf_Phdr *phdr;
1267 		struct phdr_closure phc;
1268 
1269 		/*
1270 		 * Fill in the ELF header.
1271 		 */
1272 		ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1273 		ehdr->e_ident[EI_MAG0] = ELFMAG0;
1274 		ehdr->e_ident[EI_MAG1] = ELFMAG1;
1275 		ehdr->e_ident[EI_MAG2] = ELFMAG2;
1276 		ehdr->e_ident[EI_MAG3] = ELFMAG3;
1277 		ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1278 		ehdr->e_ident[EI_DATA] = ELF_DATA;
1279 		ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1280 		ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1281 		ehdr->e_ident[EI_ABIVERSION] = 0;
1282 		ehdr->e_ident[EI_PAD] = 0;
1283 		ehdr->e_type = ET_CORE;
1284 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1285 		ehdr->e_machine = EM_386;
1286 #else
1287 		ehdr->e_machine = ELF_ARCH;
1288 #endif
1289 		ehdr->e_version = EV_CURRENT;
1290 		ehdr->e_entry = 0;
1291 		ehdr->e_phoff = phoff;
1292 		ehdr->e_flags = 0;
1293 		ehdr->e_ehsize = sizeof(Elf_Ehdr);
1294 		ehdr->e_phentsize = sizeof(Elf_Phdr);
1295 		ehdr->e_phnum = numsegs + 1;
1296 		ehdr->e_shentsize = sizeof(Elf_Shdr);
1297 		ehdr->e_shnum = 0;
1298 		ehdr->e_shstrndx = SHN_UNDEF;
1299 
1300 		/*
1301 		 * Fill in the program header entries.
1302 		 */
1303 		phdr = (Elf_Phdr *)((char *)dst + phoff);
1304 
1305 		/* The note segement. */
1306 		phdr->p_type = PT_NOTE;
1307 		phdr->p_offset = noteoff;
1308 		phdr->p_vaddr = 0;
1309 		phdr->p_paddr = 0;
1310 		phdr->p_filesz = notesz;
1311 		phdr->p_memsz = 0;
1312 		phdr->p_flags = 0;
1313 		phdr->p_align = 0;
1314 		phdr++;
1315 
1316 		/* All the writable segments from the program. */
1317 		phc.phdr = phdr;
1318 		phc.offset = *off;
1319 		each_writable_segment(td, cb_put_phdr, &phc);
1320 	}
1321 }
1322 
1323 static void
1324 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1325     const void *desc, size_t descsz)
1326 {
1327 	Elf_Note note;
1328 
1329 	note.n_namesz = strlen(name) + 1;
1330 	note.n_descsz = descsz;
1331 	note.n_type = type;
1332 	if (dst != NULL)
1333 		bcopy(&note, (char *)dst + *off, sizeof note);
1334 	*off += sizeof note;
1335 	if (dst != NULL)
1336 		bcopy(name, (char *)dst + *off, note.n_namesz);
1337 	*off += roundup2(note.n_namesz, sizeof(Elf_Size));
1338 	if (dst != NULL)
1339 		bcopy(desc, (char *)dst + *off, note.n_descsz);
1340 	*off += roundup2(note.n_descsz, sizeof(Elf_Size));
1341 }
1342 
1343 /*
1344  * Tell kern_execve.c about it, with a little help from the linker.
1345  */
1346 static struct execsw __elfN(execsw) = {
1347 	__CONCAT(exec_, __elfN(imgact)),
1348 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1349 };
1350 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
1351