xref: /freebsd/sys/kern/imgact_elf.c (revision 817420dc8eac7df799c78f5309b75092b7f7cd40)
1 /*-
2  * Copyright (c) 1995-1996 S�ren Schmidt
3  * Copyright (c) 1996 Peter Wemm
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer
11  *    in this position and unchanged.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. The name of the author may not be used to endorse or promote products
16  *    derived from this software withough specific prior written permission
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28  *
29  * $FreeBSD$
30  */
31 
32 #include "opt_rlimit.h"
33 
34 #include <sys/param.h>
35 #include <sys/exec.h>
36 #include <sys/fcntl.h>
37 #include <sys/imgact.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/malloc.h>
41 #include <sys/mutex.h>
42 #include <sys/mman.h>
43 #include <sys/namei.h>
44 #include <sys/pioctl.h>
45 #include <sys/proc.h>
46 #include <sys/procfs.h>
47 #include <sys/resourcevar.h>
48 #include <sys/systm.h>
49 #include <sys/signalvar.h>
50 #include <sys/stat.h>
51 #include <sys/syscall.h>
52 #include <sys/sysctl.h>
53 #include <sys/sysent.h>
54 #include <sys/vnode.h>
55 
56 #include <vm/vm.h>
57 #include <vm/vm_kern.h>
58 #include <vm/vm_param.h>
59 #include <vm/pmap.h>
60 #include <sys/lock.h>
61 #include <vm/vm_map.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_extern.h>
64 
65 #include <machine/elf.h>
66 #include <machine/md_var.h>
67 
68 #define OLD_EI_BRAND	8
69 
70 __ElfType(Brandinfo);
71 __ElfType(Auxargs);
72 
73 static int elf_check_header __P((const Elf_Ehdr *hdr));
74 static int elf_freebsd_fixup __P((register_t **stack_base,
75     struct image_params *imgp));
76 static int elf_load_file __P((struct proc *p, const char *file, u_long *addr,
77     u_long *entry));
78 static int elf_load_section __P((struct proc *p,
79     struct vmspace *vmspace, struct vnode *vp,
80     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
81     vm_prot_t prot));
82 static int exec_elf_imgact __P((struct image_params *imgp));
83 
84 static int elf_trace = 0;
85 SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
86 
87 struct sysentvec elf_freebsd_sysvec = {
88         SYS_MAXSYSCALL,
89         sysent,
90         0,
91         0,
92         0,
93         0,
94         0,
95         0,
96         elf_freebsd_fixup,
97         sendsig,
98         sigcode,
99         &szsigcode,
100         0,
101 	"FreeBSD ELF",
102 	elf_coredump,
103 	NULL,
104 	MINSIGSTKSZ
105 };
106 
107 static Elf_Brandinfo freebsd_brand_info = {
108 						ELFOSABI_FREEBSD,
109 						"",
110 						"/usr/libexec/ld-elf.so.1",
111 						&elf_freebsd_sysvec
112 					  };
113 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
114 							&freebsd_brand_info,
115 							NULL, NULL, NULL,
116 							NULL, NULL, NULL, NULL
117 						    };
118 
119 int
120 elf_insert_brand_entry(Elf_Brandinfo *entry)
121 {
122 	int i;
123 
124 	for (i=1; i<MAX_BRANDS; i++) {
125 		if (elf_brand_list[i] == NULL) {
126 			elf_brand_list[i] = entry;
127 			break;
128 		}
129 	}
130 	if (i == MAX_BRANDS)
131 		return -1;
132 	return 0;
133 }
134 
135 int
136 elf_remove_brand_entry(Elf_Brandinfo *entry)
137 {
138 	int i;
139 
140 	for (i=1; i<MAX_BRANDS; i++) {
141 		if (elf_brand_list[i] == entry) {
142 			elf_brand_list[i] = NULL;
143 			break;
144 		}
145 	}
146 	if (i == MAX_BRANDS)
147 		return -1;
148 	return 0;
149 }
150 
151 int
152 elf_brand_inuse(Elf_Brandinfo *entry)
153 {
154 	struct proc *p;
155 	int rval = FALSE;
156 
157 	lockmgr(&allproc_lock, LK_SHARED, NULL, CURPROC);
158 	LIST_FOREACH(p, &allproc, p_list) {
159 		if (p->p_sysent == entry->sysvec) {
160 			rval = TRUE;
161 			break;
162 		}
163 	}
164 	lockmgr(&allproc_lock, LK_RELEASE, NULL, CURPROC);
165 
166 	return (rval);
167 }
168 
169 static int
170 elf_check_header(const Elf_Ehdr *hdr)
171 {
172 	if (!IS_ELF(*hdr) ||
173 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
174 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
175 	    hdr->e_ident[EI_VERSION] != EV_CURRENT)
176 		return ENOEXEC;
177 
178 	if (!ELF_MACHINE_OK(hdr->e_machine))
179 		return ENOEXEC;
180 
181 	if (hdr->e_version != ELF_TARG_VER)
182 		return ENOEXEC;
183 
184 	return 0;
185 }
186 
187 static int
188 elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp, vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
189 {
190 	size_t map_len;
191 	vm_offset_t map_addr;
192 	int error, rv;
193 	size_t copy_len;
194 	vm_object_t object;
195 	vm_offset_t file_addr;
196 	vm_offset_t data_buf = 0;
197 
198 	VOP_GETVOBJECT(vp, &object);
199 	error = 0;
200 
201 	/*
202 	 * It's necessary to fail if the filsz + offset taken from the
203 	 * header is greater than the actual file pager object's size.
204 	 * If we were to allow this, then the vm_map_find() below would
205 	 * walk right off the end of the file object and into the ether.
206 	 *
207 	 * While I'm here, might as well check for something else that
208 	 * is invalid: filsz cannot be greater than memsz.
209 	 */
210 	if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
211 	    filsz > memsz) {
212 		uprintf("elf_load_section: truncated ELF file\n");
213 		return (ENOEXEC);
214 	}
215 
216 	map_addr = trunc_page((vm_offset_t)vmaddr);
217 	file_addr = trunc_page(offset);
218 
219 	/*
220 	 * We have two choices.  We can either clear the data in the last page
221 	 * of an oversized mapping, or we can start the anon mapping a page
222 	 * early and copy the initialized data into that first page.  We
223 	 * choose the second..
224 	 */
225 	if (memsz > filsz)
226 		map_len = trunc_page(offset+filsz) - file_addr;
227 	else
228 		map_len = round_page(offset+filsz) - file_addr;
229 
230 	if (map_len != 0) {
231 		vm_object_reference(object);
232 		vm_map_lock(&vmspace->vm_map);
233 		rv = vm_map_insert(&vmspace->vm_map,
234 				      object,
235 				      file_addr,	/* file offset */
236 				      map_addr,		/* virtual start */
237 				      map_addr + map_len,/* virtual end */
238 				      prot,
239 				      VM_PROT_ALL,
240 				      MAP_COPY_ON_WRITE | MAP_PREFAULT);
241 		vm_map_unlock(&vmspace->vm_map);
242 		if (rv != KERN_SUCCESS) {
243 			vm_object_deallocate(object);
244 			return EINVAL;
245 		}
246 
247 		/* we can stop now if we've covered it all */
248 		if (memsz == filsz)
249 			return 0;
250 	}
251 
252 
253 	/*
254 	 * We have to get the remaining bit of the file into the first part
255 	 * of the oversized map segment.  This is normally because the .data
256 	 * segment in the file is extended to provide bss.  It's a neat idea
257 	 * to try and save a page, but it's a pain in the behind to implement.
258 	 */
259 	copy_len = (offset + filsz) - trunc_page(offset + filsz);
260 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
261 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
262 
263 	/* This had damn well better be true! */
264         if (map_len != 0) {
265 		vm_map_lock(&vmspace->vm_map);
266 		rv = vm_map_insert(&vmspace->vm_map, NULL, 0,
267 					map_addr, map_addr + map_len,
268 					VM_PROT_ALL, VM_PROT_ALL, 0);
269 		vm_map_unlock(&vmspace->vm_map);
270 		if (rv != KERN_SUCCESS)
271 			return EINVAL;
272 	}
273 
274 	if (copy_len != 0) {
275 		vm_object_reference(object);
276 		rv = vm_map_find(exec_map,
277 				 object,
278 				 trunc_page(offset + filsz),
279 				 &data_buf,
280 				 PAGE_SIZE,
281 				 TRUE,
282 				 VM_PROT_READ,
283 				 VM_PROT_ALL,
284 				 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
285 		if (rv != KERN_SUCCESS) {
286 			vm_object_deallocate(object);
287 			return EINVAL;
288 		}
289 
290 		/* send the page fragment to user space */
291 		error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len);
292 		vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
293 		if (error)
294 			return (error);
295 	}
296 
297 	/*
298 	 * set it to the specified protection
299 	 */
300 	vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len,  prot,
301 		       FALSE);
302 
303 	return error;
304 }
305 
306 /*
307  * Load the file "file" into memory.  It may be either a shared object
308  * or an executable.
309  *
310  * The "addr" reference parameter is in/out.  On entry, it specifies
311  * the address where a shared object should be loaded.  If the file is
312  * an executable, this value is ignored.  On exit, "addr" specifies
313  * where the file was actually loaded.
314  *
315  * The "entry" reference parameter is out only.  On exit, it specifies
316  * the entry point for the loaded file.
317  */
318 static int
319 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
320 {
321 	const Elf_Ehdr *hdr = NULL;
322 	const Elf_Phdr *phdr = NULL;
323 	struct nameidata nd;
324 	struct vmspace *vmspace = p->p_vmspace;
325 	struct vattr attr;
326 	struct image_params image_params, *imgp;
327 	vm_prot_t prot;
328 	u_long rbase;
329 	u_long base_addr = 0;
330 	int error, i, numsegs;
331 
332 	imgp = &image_params;
333 	/*
334 	 * Initialize part of the common data
335 	 */
336 	imgp->proc = p;
337 	imgp->uap = NULL;
338 	imgp->attr = &attr;
339 	imgp->firstpage = NULL;
340 	imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
341 
342 	if (imgp->image_header == NULL) {
343 		nd.ni_vp = NULL;
344 		error = ENOMEM;
345 		goto fail;
346 	}
347 
348         NDINIT(&nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, p);
349 
350 	if ((error = namei(&nd)) != 0) {
351 		nd.ni_vp = NULL;
352 		goto fail;
353 	}
354 	NDFREE(&nd, NDF_ONLY_PNBUF);
355 	imgp->vp = nd.ni_vp;
356 
357 	/*
358 	 * Check permissions, modes, uid, etc on the file, and "open" it.
359 	 */
360 	error = exec_check_permissions(imgp);
361 	if (error) {
362 		VOP_UNLOCK(nd.ni_vp, 0, p);
363 		goto fail;
364 	}
365 
366 	error = exec_map_first_page(imgp);
367 	/*
368 	 * Also make certain that the interpreter stays the same, so set
369 	 * its VTEXT flag, too.
370 	 */
371 	if (error == 0)
372 		nd.ni_vp->v_flag |= VTEXT;
373 	VOP_UNLOCK(nd.ni_vp, 0, p);
374 	if (error)
375                 goto fail;
376 
377 	hdr = (const Elf_Ehdr *)imgp->image_header;
378 	if ((error = elf_check_header(hdr)) != 0)
379 		goto fail;
380 	if (hdr->e_type == ET_DYN)
381 		rbase = *addr;
382 	else if (hdr->e_type == ET_EXEC)
383 		rbase = 0;
384 	else {
385 		error = ENOEXEC;
386 		goto fail;
387 	}
388 
389 	/* Only support headers that fit within first page for now */
390 	if ((hdr->e_phoff > PAGE_SIZE) ||
391 	    (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
392 		error = ENOEXEC;
393 		goto fail;
394 	}
395 
396 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
397 
398 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
399 		if (phdr[i].p_type == PT_LOAD) {	/* Loadable segment */
400 			prot = 0;
401 			if (phdr[i].p_flags & PF_X)
402   				prot |= VM_PROT_EXECUTE;
403 			if (phdr[i].p_flags & PF_W)
404   				prot |= VM_PROT_WRITE;
405 			if (phdr[i].p_flags & PF_R)
406   				prot |= VM_PROT_READ;
407 
408 			if ((error = elf_load_section(p, vmspace, nd.ni_vp,
409   						     phdr[i].p_offset,
410   						     (caddr_t)phdr[i].p_vaddr +
411 							rbase,
412   						     phdr[i].p_memsz,
413   						     phdr[i].p_filesz, prot)) != 0)
414 				goto fail;
415 			/*
416 			 * Establish the base address if this is the
417 			 * first segment.
418 			 */
419 			if (numsegs == 0)
420   				base_addr = trunc_page(phdr[i].p_vaddr + rbase);
421 			numsegs++;
422 		}
423 	}
424 	*addr = base_addr;
425 	*entry=(unsigned long)hdr->e_entry + rbase;
426 
427 fail:
428 	if (imgp->firstpage)
429 		exec_unmap_first_page(imgp);
430 	if (imgp->image_header)
431 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
432 			PAGE_SIZE);
433 	if (nd.ni_vp)
434 		vrele(nd.ni_vp);
435 
436 	return error;
437 }
438 
439 static int fallback_elf_brand = ELFOSABI_FREEBSD;
440 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
441 		&fallback_elf_brand, ELFOSABI_FREEBSD,
442 		"ELF brand of last resort");
443 
444 static int
445 exec_elf_imgact(struct image_params *imgp)
446 {
447 	const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
448 	const Elf_Phdr *phdr;
449 	Elf_Auxargs *elf_auxargs = NULL;
450 	struct vmspace *vmspace;
451 	vm_prot_t prot;
452 	u_long text_size = 0, data_size = 0;
453 	u_long text_addr = 0, data_addr = 0;
454 	u_long addr, entry = 0, proghdr = 0;
455 	int error, i;
456 	const char *interp = NULL;
457 	Elf_Brandinfo *brand_info;
458 	char path[MAXPATHLEN];
459 
460 	/*
461 	 * Do we have a valid ELF header ?
462 	 */
463 	if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
464 		return -1;
465 
466 	/*
467 	 * From here on down, we return an errno, not -1, as we've
468 	 * detected an ELF file.
469 	 */
470 
471 	if ((hdr->e_phoff > PAGE_SIZE) ||
472 	    (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
473 		/* Only support headers in first page for now */
474 		return ENOEXEC;
475 	}
476 	phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
477 
478 	/*
479 	 * From this point on, we may have resources that need to be freed.
480 	 */
481 
482 	/*
483 	 * Yeah, I'm paranoid.  There is every reason in the world to get
484 	 * VTEXT now since from here on out, there are places we can have
485 	 * a context switch.  Better safe than sorry; I really don't want
486 	 * the file to change while it's being loaded.
487 	 */
488 	mtx_enter(&imgp->vp->v_interlock, MTX_DEF);
489 	imgp->vp->v_flag |= VTEXT;
490 	mtx_exit(&imgp->vp->v_interlock, MTX_DEF);
491 
492 	if ((error = exec_extract_strings(imgp)) != 0)
493 		goto fail;
494 
495 	exec_new_vmspace(imgp);
496 
497 	vmspace = imgp->proc->p_vmspace;
498 
499 	for (i = 0; i < hdr->e_phnum; i++) {
500 		switch(phdr[i].p_type) {
501 
502 		case PT_LOAD:	/* Loadable segment */
503 			prot = 0;
504 			if (phdr[i].p_flags & PF_X)
505   				prot |= VM_PROT_EXECUTE;
506 			if (phdr[i].p_flags & PF_W)
507   				prot |= VM_PROT_WRITE;
508 			if (phdr[i].p_flags & PF_R)
509   				prot |= VM_PROT_READ;
510 
511 			if ((error = elf_load_section(imgp->proc,
512 						     vmspace, imgp->vp,
513   						     phdr[i].p_offset,
514   						     (caddr_t)phdr[i].p_vaddr,
515   						     phdr[i].p_memsz,
516   						     phdr[i].p_filesz, prot)) != 0)
517   				goto fail;
518 
519 			/*
520 			 * Is this .text or .data ??
521 			 *
522 			 * We only handle one each of those yet XXX
523 			 */
524 			if (hdr->e_entry >= phdr[i].p_vaddr &&
525 			hdr->e_entry <(phdr[i].p_vaddr+phdr[i].p_memsz)) {
526   				text_addr = trunc_page(phdr[i].p_vaddr);
527   				text_size = round_page(phdr[i].p_memsz +
528 						       phdr[i].p_vaddr -
529 						       text_addr);
530 				entry = (u_long)hdr->e_entry;
531 			} else {
532   				data_addr = trunc_page(phdr[i].p_vaddr);
533   				data_size = round_page(phdr[i].p_memsz +
534 						       phdr[i].p_vaddr -
535 						       data_addr);
536 			}
537 			break;
538 	  	case PT_INTERP:	/* Path to interpreter */
539 			if (phdr[i].p_filesz > MAXPATHLEN ||
540 			    phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
541 				error = ENOEXEC;
542 				goto fail;
543 			}
544 			interp = imgp->image_header + phdr[i].p_offset;
545 			break;
546 		case PT_PHDR: 	/* Program header table info */
547 			proghdr = phdr[i].p_vaddr;
548 			break;
549 		default:
550 			break;
551 		}
552 	}
553 
554 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
555 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
556 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
557 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
558 
559 	addr = ELF_RTLD_ADDR(vmspace);
560 
561 	imgp->entry_addr = entry;
562 
563 	brand_info = NULL;
564 
565 	/* XXX  For now we look for the magic "FreeBSD" that we used to put
566 	 * into the ELF header at the EI_ABIVERSION location.  If found use
567 	 * that information rather than figuring out the ABI from proper
568 	 * branding.  This should be removed for 5.0-RELEASE. The Linux caes
569 	 * can be figured out from the `interp_path' field.
570 	 */
571 	if (strcmp("FreeBSD", (const char *)&hdr->e_ident[OLD_EI_BRAND]) == 0)
572 		brand_info = &freebsd_brand_info;
573 
574 	/* If the executable has a brand, search for it in the brand list. */
575 	if (brand_info == NULL) {
576 		for (i = 0;  i < MAX_BRANDS;  i++) {
577 			Elf_Brandinfo *bi = elf_brand_list[i];
578 
579 			if (bi != NULL && hdr->e_ident[EI_OSABI] == bi->brand) {
580 				brand_info = bi;
581 				break;
582 			}
583 		}
584 	}
585 
586 	/* Lacking a known brand, search for a recognized interpreter. */
587 	if (brand_info == NULL && interp != NULL) {
588 		for (i = 0;  i < MAX_BRANDS;  i++) {
589 			Elf_Brandinfo *bi = elf_brand_list[i];
590 
591 			if (bi != NULL &&
592 			    strcmp(interp, bi->interp_path) == 0) {
593 				brand_info = bi;
594 				break;
595 			}
596 		}
597 	}
598 
599 	/* Lacking a recognized interpreter, try the default brand */
600 	if (brand_info == NULL) {
601 		for (i = 0; i < MAX_BRANDS; i++) {
602 			Elf_Brandinfo *bi = elf_brand_list[i];
603 
604 			if (bi != NULL && fallback_elf_brand == bi->brand) {
605 				brand_info = bi;
606 				break;
607 			}
608 		}
609 	}
610 
611 	/* XXX - Assume FreeBSD after the branding method change. */
612 	if (brand_info == NULL)
613 		brand_info = &freebsd_brand_info;
614 
615 	if (brand_info == NULL) {
616 		uprintf("ELF binary type \"%u\" not known.\n",
617 		    hdr->e_ident[EI_OSABI]);
618 		error = ENOEXEC;
619 		goto fail;
620 	}
621 
622 	imgp->proc->p_sysent = brand_info->sysvec;
623 	if (interp != NULL) {
624 	        snprintf(path, sizeof(path), "%s%s",
625 			 brand_info->emul_path, interp);
626 		if ((error = elf_load_file(imgp->proc, path, &addr,
627 					   &imgp->entry_addr)) != 0) {
628 		        if ((error = elf_load_file(imgp->proc, interp, &addr,
629 						   &imgp->entry_addr)) != 0) {
630 			        uprintf("ELF interpreter %s not found\n", path);
631 				goto fail;
632 			}
633                 }
634 	}
635 
636 	/*
637 	 * Construct auxargs table (used by the fixup routine)
638 	 */
639 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
640 	elf_auxargs->execfd = -1;
641 	elf_auxargs->phdr = proghdr;
642 	elf_auxargs->phent = hdr->e_phentsize;
643 	elf_auxargs->phnum = hdr->e_phnum;
644 	elf_auxargs->pagesz = PAGE_SIZE;
645 	elf_auxargs->base = addr;
646 	elf_auxargs->flags = 0;
647 	elf_auxargs->entry = entry;
648 	elf_auxargs->trace = elf_trace;
649 
650 	imgp->auxargs = elf_auxargs;
651 	imgp->interpreted = 0;
652 
653 fail:
654 	return error;
655 }
656 
657 static int
658 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
659 {
660 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
661 	register_t *pos;
662 
663 	pos = *stack_base + (imgp->argc + imgp->envc + 2);
664 
665 	if (args->trace) {
666 		AUXARGS_ENTRY(pos, AT_DEBUG, 1);
667 	}
668 	if (args->execfd != -1) {
669 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
670 	}
671 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
672 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
673 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
674 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
675 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
676 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
677 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
678 	AUXARGS_ENTRY(pos, AT_NULL, 0);
679 
680 	free(imgp->auxargs, M_TEMP);
681 	imgp->auxargs = NULL;
682 
683 	(*stack_base)--;
684 	suword(*stack_base, (long) imgp->argc);
685 	return 0;
686 }
687 
688 /*
689  * Code for generating ELF core dumps.
690  */
691 
692 typedef void (*segment_callback) __P((vm_map_entry_t, void *));
693 
694 /* Closure for cb_put_phdr(). */
695 struct phdr_closure {
696 	Elf_Phdr *phdr;		/* Program header to fill in */
697 	Elf_Off offset;		/* Offset of segment in core file */
698 };
699 
700 /* Closure for cb_size_segment(). */
701 struct sseg_closure {
702 	int count;		/* Count of writable segments. */
703 	size_t size;		/* Total size of all writable segments. */
704 };
705 
706 static void cb_put_phdr __P((vm_map_entry_t, void *));
707 static void cb_size_segment __P((vm_map_entry_t, void *));
708 static void each_writable_segment __P((struct proc *, segment_callback,
709     void *));
710 static int elf_corehdr __P((struct proc *, struct vnode *, struct ucred *,
711     int, void *, size_t));
712 static void elf_puthdr __P((struct proc *, void *, size_t *,
713     const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int));
714 static void elf_putnote __P((void *, size_t *, const char *, int,
715     const void *, size_t));
716 
717 extern int osreldate;
718 
719 int
720 elf_coredump(p, vp, limit)
721 	register struct proc *p;
722 	register struct vnode *vp;
723 	off_t limit;
724 {
725 	register struct ucred *cred = p->p_ucred;
726 	int error = 0;
727 	struct sseg_closure seginfo;
728 	void *hdr;
729 	size_t hdrsize;
730 
731 	/* Size the program segments. */
732 	seginfo.count = 0;
733 	seginfo.size = 0;
734 	each_writable_segment(p, cb_size_segment, &seginfo);
735 
736 	/*
737 	 * Calculate the size of the core file header area by making
738 	 * a dry run of generating it.  Nothing is written, but the
739 	 * size is calculated.
740 	 */
741 	hdrsize = 0;
742 	elf_puthdr((struct proc *)NULL, (void *)NULL, &hdrsize,
743 	    (const prstatus_t *)NULL, (const prfpregset_t *)NULL,
744 	    (const prpsinfo_t *)NULL, seginfo.count);
745 
746 	if (hdrsize + seginfo.size >= limit)
747 		return (EFAULT);
748 
749 	/*
750 	 * Allocate memory for building the header, fill it up,
751 	 * and write it out.
752 	 */
753 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
754 	if (hdr == NULL) {
755 		return EINVAL;
756 	}
757 	error = elf_corehdr(p, vp, cred, seginfo.count, hdr, hdrsize);
758 
759 	/* Write the contents of all of the writable segments. */
760 	if (error == 0) {
761 		Elf_Phdr *php;
762 		off_t offset;
763 		int i;
764 
765 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
766 		offset = hdrsize;
767 		for (i = 0;  i < seginfo.count;  i++) {
768 			error = vn_rdwr(UIO_WRITE, vp, (caddr_t)php->p_vaddr,
769 			    php->p_filesz, offset, UIO_USERSPACE,
770 			    IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, p);
771 			if (error != 0)
772 				break;
773 			offset += php->p_filesz;
774 			php++;
775 		}
776 	}
777 	free(hdr, M_TEMP);
778 
779 	return error;
780 }
781 
782 /*
783  * A callback for each_writable_segment() to write out the segment's
784  * program header entry.
785  */
786 static void
787 cb_put_phdr(entry, closure)
788 	vm_map_entry_t entry;
789 	void *closure;
790 {
791 	struct phdr_closure *phc = (struct phdr_closure *)closure;
792 	Elf_Phdr *phdr = phc->phdr;
793 
794 	phc->offset = round_page(phc->offset);
795 
796 	phdr->p_type = PT_LOAD;
797 	phdr->p_offset = phc->offset;
798 	phdr->p_vaddr = entry->start;
799 	phdr->p_paddr = 0;
800 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
801 	phdr->p_align = PAGE_SIZE;
802 	phdr->p_flags = 0;
803 	if (entry->protection & VM_PROT_READ)
804 		phdr->p_flags |= PF_R;
805 	if (entry->protection & VM_PROT_WRITE)
806 		phdr->p_flags |= PF_W;
807 	if (entry->protection & VM_PROT_EXECUTE)
808 		phdr->p_flags |= PF_X;
809 
810 	phc->offset += phdr->p_filesz;
811 	phc->phdr++;
812 }
813 
814 /*
815  * A callback for each_writable_segment() to gather information about
816  * the number of segments and their total size.
817  */
818 static void
819 cb_size_segment(entry, closure)
820 	vm_map_entry_t entry;
821 	void *closure;
822 {
823 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
824 
825 	ssc->count++;
826 	ssc->size += entry->end - entry->start;
827 }
828 
829 /*
830  * For each writable segment in the process's memory map, call the given
831  * function with a pointer to the map entry and some arbitrary
832  * caller-supplied data.
833  */
834 static void
835 each_writable_segment(p, func, closure)
836 	struct proc *p;
837 	segment_callback func;
838 	void *closure;
839 {
840 	vm_map_t map = &p->p_vmspace->vm_map;
841 	vm_map_entry_t entry;
842 
843 	for (entry = map->header.next;  entry != &map->header;
844 	    entry = entry->next) {
845 		vm_object_t obj;
846 
847 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) ||
848 		    (entry->protection & (VM_PROT_READ|VM_PROT_WRITE)) !=
849 		    (VM_PROT_READ|VM_PROT_WRITE))
850 			continue;
851 
852 		/*
853 		** Dont include memory segment in the coredump if
854 		** MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
855 		** madvise(2).
856 		*/
857 		if (entry->eflags & MAP_ENTRY_NOCOREDUMP)
858 			continue;
859 
860 		if ((obj = entry->object.vm_object) == NULL)
861 			continue;
862 
863 		/* Find the deepest backing object. */
864 		while (obj->backing_object != NULL)
865 			obj = obj->backing_object;
866 
867 		/* Ignore memory-mapped devices and such things. */
868 		if (obj->type != OBJT_DEFAULT &&
869 		    obj->type != OBJT_SWAP &&
870 		    obj->type != OBJT_VNODE)
871 			continue;
872 
873 		(*func)(entry, closure);
874 	}
875 }
876 
877 /*
878  * Write the core file header to the file, including padding up to
879  * the page boundary.
880  */
881 static int
882 elf_corehdr(p, vp, cred, numsegs, hdr, hdrsize)
883 	struct proc *p;
884 	struct vnode *vp;
885 	struct ucred *cred;
886 	int numsegs;
887 	size_t hdrsize;
888 	void *hdr;
889 {
890 	size_t off;
891 	prstatus_t status;
892 	prfpregset_t fpregset;
893 	prpsinfo_t psinfo;
894 
895 	/* Gather the information for the header. */
896 	bzero(&status, sizeof status);
897 	status.pr_version = PRSTATUS_VERSION;
898 	status.pr_statussz = sizeof(prstatus_t);
899 	status.pr_gregsetsz = sizeof(gregset_t);
900 	status.pr_fpregsetsz = sizeof(fpregset_t);
901 	status.pr_osreldate = osreldate;
902 	status.pr_cursig = p->p_sig;
903 	status.pr_pid = p->p_pid;
904 	fill_regs(p, &status.pr_reg);
905 
906 	fill_fpregs(p, &fpregset);
907 
908 	bzero(&psinfo, sizeof psinfo);
909 	psinfo.pr_version = PRPSINFO_VERSION;
910 	psinfo.pr_psinfosz = sizeof(prpsinfo_t);
911 	strncpy(psinfo.pr_fname, p->p_comm, MAXCOMLEN);
912 	/* XXX - We don't fill in the command line arguments properly yet. */
913 	strncpy(psinfo.pr_psargs, p->p_comm, PRARGSZ);
914 
915 	/* Fill in the header. */
916 	bzero(hdr, hdrsize);
917 	off = 0;
918 	elf_puthdr(p, hdr, &off, &status, &fpregset, &psinfo, numsegs);
919 
920 	/* Write it to the core file. */
921 	return vn_rdwr(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
922 	    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, NULL, p);
923 }
924 
925 static void
926 elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status,
927     const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
928 {
929 	size_t ehoff;
930 	size_t phoff;
931 	size_t noteoff;
932 	size_t notesz;
933 
934 	ehoff = *off;
935 	*off += sizeof(Elf_Ehdr);
936 
937 	phoff = *off;
938 	*off += (numsegs + 1) * sizeof(Elf_Phdr);
939 
940 	noteoff = *off;
941 	elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status,
942 	    sizeof *status);
943 	elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
944 	    sizeof *fpregset);
945 	elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
946 	    sizeof *psinfo);
947 	notesz = *off - noteoff;
948 
949 	/* Align up to a page boundary for the program segments. */
950 	*off = round_page(*off);
951 
952 	if (dst != NULL) {
953 		Elf_Ehdr *ehdr;
954 		Elf_Phdr *phdr;
955 		struct phdr_closure phc;
956 
957 		/*
958 		 * Fill in the ELF header.
959 		 */
960 		ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
961 		ehdr->e_ident[EI_MAG0] = ELFMAG0;
962 		ehdr->e_ident[EI_MAG1] = ELFMAG1;
963 		ehdr->e_ident[EI_MAG2] = ELFMAG2;
964 		ehdr->e_ident[EI_MAG3] = ELFMAG3;
965 		ehdr->e_ident[EI_CLASS] = ELF_CLASS;
966 		ehdr->e_ident[EI_DATA] = ELF_DATA;
967 		ehdr->e_ident[EI_VERSION] = EV_CURRENT;
968 		ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
969 		ehdr->e_ident[EI_ABIVERSION] = 0;
970 		ehdr->e_ident[EI_PAD] = 0;
971 		ehdr->e_type = ET_CORE;
972 		ehdr->e_machine = ELF_ARCH;
973 		ehdr->e_version = EV_CURRENT;
974 		ehdr->e_entry = 0;
975 		ehdr->e_phoff = phoff;
976 		ehdr->e_flags = 0;
977 		ehdr->e_ehsize = sizeof(Elf_Ehdr);
978 		ehdr->e_phentsize = sizeof(Elf_Phdr);
979 		ehdr->e_phnum = numsegs + 1;
980 		ehdr->e_shentsize = sizeof(Elf_Shdr);
981 		ehdr->e_shnum = 0;
982 		ehdr->e_shstrndx = SHN_UNDEF;
983 
984 		/*
985 		 * Fill in the program header entries.
986 		 */
987 		phdr = (Elf_Phdr *)((char *)dst + phoff);
988 
989 		/* The note segement. */
990 		phdr->p_type = PT_NOTE;
991 		phdr->p_offset = noteoff;
992 		phdr->p_vaddr = 0;
993 		phdr->p_paddr = 0;
994 		phdr->p_filesz = notesz;
995 		phdr->p_memsz = 0;
996 		phdr->p_flags = 0;
997 		phdr->p_align = 0;
998 		phdr++;
999 
1000 		/* All the writable segments from the program. */
1001 		phc.phdr = phdr;
1002 		phc.offset = *off;
1003 		each_writable_segment(p, cb_put_phdr, &phc);
1004 	}
1005 }
1006 
1007 static void
1008 elf_putnote(void *dst, size_t *off, const char *name, int type,
1009     const void *desc, size_t descsz)
1010 {
1011 	Elf_Note note;
1012 
1013 	note.n_namesz = strlen(name) + 1;
1014 	note.n_descsz = descsz;
1015 	note.n_type = type;
1016 	if (dst != NULL)
1017 		bcopy(&note, (char *)dst + *off, sizeof note);
1018 	*off += sizeof note;
1019 	if (dst != NULL)
1020 		bcopy(name, (char *)dst + *off, note.n_namesz);
1021 	*off += roundup2(note.n_namesz, sizeof(Elf_Size));
1022 	if (dst != NULL)
1023 		bcopy(desc, (char *)dst + *off, note.n_descsz);
1024 	*off += roundup2(note.n_descsz, sizeof(Elf_Size));
1025 }
1026 
1027 /*
1028  * Tell kern_execve.c about it, with a little help from the linker.
1029  */
1030 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1031 EXEC_SET(elf, elf_execsw);
1032