xref: /freebsd/sys/kern/imgact_elf.c (revision 1b6c76a2fe091c74f08427e6c870851025a9cf67)
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 withough 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  * $FreeBSD$
31  */
32 
33 #include "opt_rlimit.h"
34 
35 #include <sys/param.h>
36 #include <sys/exec.h>
37 #include <sys/fcntl.h>
38 #include <sys/imgact.h>
39 #include <sys/imgact_elf.h>
40 #include <sys/kernel.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mutex.h>
44 #include <sys/mman.h>
45 #include <sys/namei.h>
46 #include <sys/pioctl.h>
47 #include <sys/proc.h>
48 #include <sys/procfs.h>
49 #include <sys/resourcevar.h>
50 #include <sys/systm.h>
51 #include <sys/signalvar.h>
52 #include <sys/stat.h>
53 #include <sys/sx.h>
54 #include <sys/syscall.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysent.h>
57 #include <sys/vnode.h>
58 
59 #include <vm/vm.h>
60 #include <vm/vm_kern.h>
61 #include <vm/vm_param.h>
62 #include <vm/pmap.h>
63 #include <vm/vm_map.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_extern.h>
66 
67 #include <machine/elf.h>
68 #include <machine/md_var.h>
69 
70 #define OLD_EI_BRAND	8
71 
72 __ElfType(Brandinfo);
73 __ElfType(Auxargs);
74 
75 static int elf_check_header __P((const Elf_Ehdr *hdr));
76 static int elf_freebsd_fixup __P((register_t **stack_base,
77     struct image_params *imgp));
78 static int elf_load_file __P((struct proc *p, const char *file, u_long *addr,
79     u_long *entry));
80 static int elf_load_section __P((struct proc *p,
81     struct vmspace *vmspace, struct vnode *vp,
82     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
83     vm_prot_t prot));
84 static int exec_elf_imgact __P((struct image_params *imgp));
85 
86 static int elf_trace = 0;
87 SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
88 
89 struct sysentvec elf_freebsd_sysvec = {
90         SYS_MAXSYSCALL,
91         sysent,
92         0,
93         0,
94         0,
95         0,
96         0,
97         0,
98         elf_freebsd_fixup,
99         sendsig,
100         sigcode,
101         &szsigcode,
102         0,
103 	"FreeBSD ELF",
104 	elf_coredump,
105 	NULL,
106 	MINSIGSTKSZ
107 };
108 
109 static Elf_Brandinfo freebsd_brand_info = {
110 						ELFOSABI_FREEBSD,
111 						"FreeBSD",
112 						"",
113 						"/usr/libexec/ld-elf.so.1",
114 						&elf_freebsd_sysvec
115 					  };
116 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
117 							&freebsd_brand_info,
118 							NULL, NULL, NULL,
119 							NULL, NULL, NULL, NULL
120 						    };
121 
122 int
123 elf_insert_brand_entry(Elf_Brandinfo *entry)
124 {
125 	int i;
126 
127 	for (i=1; i<MAX_BRANDS; i++) {
128 		if (elf_brand_list[i] == NULL) {
129 			elf_brand_list[i] = entry;
130 			break;
131 		}
132 	}
133 	if (i == MAX_BRANDS)
134 		return -1;
135 	return 0;
136 }
137 
138 int
139 elf_remove_brand_entry(Elf_Brandinfo *entry)
140 {
141 	int i;
142 
143 	for (i=1; i<MAX_BRANDS; i++) {
144 		if (elf_brand_list[i] == entry) {
145 			elf_brand_list[i] = NULL;
146 			break;
147 		}
148 	}
149 	if (i == MAX_BRANDS)
150 		return -1;
151 	return 0;
152 }
153 
154 int
155 elf_brand_inuse(Elf_Brandinfo *entry)
156 {
157 	struct proc *p;
158 	int rval = FALSE;
159 
160 	sx_slock(&allproc_lock);
161 	LIST_FOREACH(p, &allproc, p_list) {
162 		if (p->p_sysent == entry->sysvec) {
163 			rval = TRUE;
164 			break;
165 		}
166 	}
167 	sx_sunlock(&allproc_lock);
168 
169 	return (rval);
170 }
171 
172 static int
173 elf_check_header(const Elf_Ehdr *hdr)
174 {
175 	if (!IS_ELF(*hdr) ||
176 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
177 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
178 	    hdr->e_ident[EI_VERSION] != EV_CURRENT)
179 		return ENOEXEC;
180 
181 	if (!ELF_MACHINE_OK(hdr->e_machine))
182 		return ENOEXEC;
183 
184 	if (hdr->e_version != ELF_TARG_VER)
185 		return ENOEXEC;
186 
187 	return 0;
188 }
189 
190 static int
191 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)
192 {
193 	size_t map_len;
194 	vm_offset_t map_addr;
195 	int error, rv;
196 	size_t copy_len;
197 	vm_object_t object;
198 	vm_offset_t file_addr;
199 	vm_offset_t data_buf = 0;
200 
201 	VOP_GETVOBJECT(vp, &object);
202 	error = 0;
203 
204 	/*
205 	 * It's necessary to fail if the filsz + offset taken from the
206 	 * header is greater than the actual file pager object's size.
207 	 * If we were to allow this, then the vm_map_find() below would
208 	 * walk right off the end of the file object and into the ether.
209 	 *
210 	 * While I'm here, might as well check for something else that
211 	 * is invalid: filsz cannot be greater than memsz.
212 	 */
213 	if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
214 	    filsz > memsz) {
215 		uprintf("elf_load_section: truncated ELF file\n");
216 		return (ENOEXEC);
217 	}
218 
219 	map_addr = trunc_page((vm_offset_t)vmaddr);
220 	file_addr = trunc_page(offset);
221 
222 	/*
223 	 * We have two choices.  We can either clear the data in the last page
224 	 * of an oversized mapping, or we can start the anon mapping a page
225 	 * early and copy the initialized data into that first page.  We
226 	 * choose the second..
227 	 */
228 	if (memsz > filsz)
229 		map_len = trunc_page(offset+filsz) - file_addr;
230 	else
231 		map_len = round_page(offset+filsz) - file_addr;
232 
233 	mtx_lock(&vm_mtx);
234 	if (map_len != 0) {
235 		vm_object_reference(object);
236 		vm_map_lock(&vmspace->vm_map);
237 		rv = vm_map_insert(&vmspace->vm_map,
238 				      object,
239 				      file_addr,	/* file offset */
240 				      map_addr,		/* virtual start */
241 				      map_addr + map_len,/* virtual end */
242 				      prot,
243 				      VM_PROT_ALL,
244 				      MAP_COPY_ON_WRITE | MAP_PREFAULT);
245 		vm_map_unlock(&vmspace->vm_map);
246 		if (rv != KERN_SUCCESS) {
247 			vm_object_deallocate(object);
248 			mtx_unlock(&vm_mtx);
249 			return EINVAL;
250 		}
251 
252 		/* we can stop now if we've covered it all */
253 		if (memsz == filsz) {
254 			mtx_unlock(&vm_mtx);
255 			return 0;
256 		}
257 	}
258 
259 
260 	/*
261 	 * We have to get the remaining bit of the file into the first part
262 	 * of the oversized map segment.  This is normally because the .data
263 	 * segment in the file is extended to provide bss.  It's a neat idea
264 	 * to try and save a page, but it's a pain in the behind to implement.
265 	 */
266 	copy_len = (offset + filsz) - trunc_page(offset + filsz);
267 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
268 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
269 
270 	/* This had damn well better be true! */
271         if (map_len != 0) {
272 		vm_map_lock(&vmspace->vm_map);
273 		rv = vm_map_insert(&vmspace->vm_map, NULL, 0,
274 					map_addr, map_addr + map_len,
275 					VM_PROT_ALL, VM_PROT_ALL, 0);
276 		vm_map_unlock(&vmspace->vm_map);
277 		if (rv != KERN_SUCCESS) {
278 			mtx_unlock(&vm_mtx);
279 			return EINVAL;
280 		}
281 	}
282 
283 	if (copy_len != 0) {
284 		vm_object_reference(object);
285 		rv = vm_map_find(exec_map,
286 				 object,
287 				 trunc_page(offset + filsz),
288 				 &data_buf,
289 				 PAGE_SIZE,
290 				 TRUE,
291 				 VM_PROT_READ,
292 				 VM_PROT_ALL,
293 				 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
294 		if (rv != KERN_SUCCESS) {
295 			vm_object_deallocate(object);
296 			mtx_unlock(&vm_mtx);
297 			return EINVAL;
298 		}
299 
300 		/* send the page fragment to user space */
301 		mtx_unlock(&vm_mtx);
302 		error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len);
303 		mtx_lock(&vm_mtx);
304 		vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
305 		if (error) {
306 			mtx_unlock(&vm_mtx);
307 			return (error);
308 		}
309 	}
310 
311 	/*
312 	 * set it to the specified protection
313 	 */
314 	vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len,  prot,
315 		       FALSE);
316 
317 	mtx_unlock(&vm_mtx);
318 	return error;
319 }
320 
321 /*
322  * Load the file "file" into memory.  It may be either a shared object
323  * or an executable.
324  *
325  * The "addr" reference parameter is in/out.  On entry, it specifies
326  * the address where a shared object should be loaded.  If the file is
327  * an executable, this value is ignored.  On exit, "addr" specifies
328  * where the file was actually loaded.
329  *
330  * The "entry" reference parameter is out only.  On exit, it specifies
331  * the entry point for the loaded file.
332  */
333 static int
334 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
335 {
336 	const Elf_Ehdr *hdr = NULL;
337 	const Elf_Phdr *phdr = NULL;
338 	struct nameidata nd;
339 	struct vmspace *vmspace = p->p_vmspace;
340 	struct vattr attr;
341 	struct image_params image_params, *imgp;
342 	vm_prot_t prot;
343 	u_long rbase;
344 	u_long base_addr = 0;
345 	int error, i, numsegs;
346 
347 	imgp = &image_params;
348 	/*
349 	 * Initialize part of the common data
350 	 */
351 	imgp->proc = p;
352 	imgp->uap = NULL;
353 	imgp->attr = &attr;
354 	imgp->firstpage = NULL;
355 	imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
356 
357 	if (imgp->image_header == NULL) {
358 		nd.ni_vp = NULL;
359 		error = ENOMEM;
360 		goto fail;
361 	}
362 
363         NDINIT(&nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, p);
364 
365 	if ((error = namei(&nd)) != 0) {
366 		nd.ni_vp = NULL;
367 		goto fail;
368 	}
369 	NDFREE(&nd, NDF_ONLY_PNBUF);
370 	imgp->vp = nd.ni_vp;
371 
372 	/*
373 	 * Check permissions, modes, uid, etc on the file, and "open" it.
374 	 */
375 	error = exec_check_permissions(imgp);
376 	if (error) {
377 		VOP_UNLOCK(nd.ni_vp, 0, p);
378 		goto fail;
379 	}
380 
381 	error = exec_map_first_page(imgp);
382 	/*
383 	 * Also make certain that the interpreter stays the same, so set
384 	 * its VTEXT flag, too.
385 	 */
386 	if (error == 0)
387 		nd.ni_vp->v_flag |= VTEXT;
388 	VOP_UNLOCK(nd.ni_vp, 0, p);
389 	if (error)
390                 goto fail;
391 
392 	hdr = (const Elf_Ehdr *)imgp->image_header;
393 	if ((error = elf_check_header(hdr)) != 0)
394 		goto fail;
395 	if (hdr->e_type == ET_DYN)
396 		rbase = *addr;
397 	else if (hdr->e_type == ET_EXEC)
398 		rbase = 0;
399 	else {
400 		error = ENOEXEC;
401 		goto fail;
402 	}
403 
404 	/* Only support headers that fit within first page for now */
405 	if ((hdr->e_phoff > PAGE_SIZE) ||
406 	    (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
407 		error = ENOEXEC;
408 		goto fail;
409 	}
410 
411 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
412 
413 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
414 		if (phdr[i].p_type == PT_LOAD) {	/* Loadable segment */
415 			prot = 0;
416 			if (phdr[i].p_flags & PF_X)
417   				prot |= VM_PROT_EXECUTE;
418 			if (phdr[i].p_flags & PF_W)
419   				prot |= VM_PROT_WRITE;
420 			if (phdr[i].p_flags & PF_R)
421   				prot |= VM_PROT_READ;
422 
423 			if ((error = elf_load_section(p, vmspace, nd.ni_vp,
424   						     phdr[i].p_offset,
425   						     (caddr_t)phdr[i].p_vaddr +
426 							rbase,
427   						     phdr[i].p_memsz,
428   						     phdr[i].p_filesz, prot)) != 0)
429 				goto fail;
430 			/*
431 			 * Establish the base address if this is the
432 			 * first segment.
433 			 */
434 			if (numsegs == 0)
435   				base_addr = trunc_page(phdr[i].p_vaddr + rbase);
436 			numsegs++;
437 		}
438 	}
439 	*addr = base_addr;
440 	*entry=(unsigned long)hdr->e_entry + rbase;
441 
442 fail:
443 	if (imgp->firstpage)
444 		exec_unmap_first_page(imgp);
445 	if (imgp->image_header)
446 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
447 			PAGE_SIZE);
448 	if (nd.ni_vp)
449 		vrele(nd.ni_vp);
450 
451 	return error;
452 }
453 
454 /*
455  * non static, as it can be overridden by start_init()
456  */
457 int fallback_elf_brand = -1;
458 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
459 		&fallback_elf_brand, -1,
460 		"ELF brand of last resort");
461 
462 static int
463 exec_elf_imgact(struct image_params *imgp)
464 {
465 	const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
466 	const Elf_Phdr *phdr;
467 	Elf_Auxargs *elf_auxargs = NULL;
468 	struct vmspace *vmspace;
469 	vm_prot_t prot;
470 	u_long text_size = 0, data_size = 0;
471 	u_long text_addr = 0, data_addr = 0;
472 	u_long addr, entry = 0, proghdr = 0;
473 	int error, i;
474 	const char *interp = NULL;
475 	Elf_Brandinfo *brand_info;
476 	char path[MAXPATHLEN];
477 
478 	/*
479 	 * Do we have a valid ELF header ?
480 	 */
481 	if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
482 		return -1;
483 
484 	/*
485 	 * From here on down, we return an errno, not -1, as we've
486 	 * detected an ELF file.
487 	 */
488 
489 	if ((hdr->e_phoff > PAGE_SIZE) ||
490 	    (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
491 		/* Only support headers in first page for now */
492 		return ENOEXEC;
493 	}
494 	phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
495 
496 	/*
497 	 * From this point on, we may have resources that need to be freed.
498 	 */
499 
500 	/*
501 	 * Yeah, I'm paranoid.  There is every reason in the world to get
502 	 * VTEXT now since from here on out, there are places we can have
503 	 * a context switch.  Better safe than sorry; I really don't want
504 	 * the file to change while it's being loaded.
505 	 */
506 	mtx_lock(&imgp->vp->v_interlock);
507 	imgp->vp->v_flag |= VTEXT;
508 	mtx_unlock(&imgp->vp->v_interlock);
509 
510 	if ((error = exec_extract_strings(imgp)) != 0)
511 		goto fail;
512 
513 	mtx_lock(&vm_mtx);
514 	exec_new_vmspace(imgp);
515 
516 	vmspace = imgp->proc->p_vmspace;
517 	mtx_unlock(&vm_mtx);
518 
519 	for (i = 0; i < hdr->e_phnum; i++) {
520 		switch(phdr[i].p_type) {
521 
522 		case PT_LOAD:	/* Loadable segment */
523 			prot = 0;
524 			if (phdr[i].p_flags & PF_X)
525   				prot |= VM_PROT_EXECUTE;
526 			if (phdr[i].p_flags & PF_W)
527   				prot |= VM_PROT_WRITE;
528 			if (phdr[i].p_flags & PF_R)
529   				prot |= VM_PROT_READ;
530 
531 			if ((error = elf_load_section(imgp->proc,
532 						     vmspace, imgp->vp,
533   						     phdr[i].p_offset,
534   						     (caddr_t)phdr[i].p_vaddr,
535   						     phdr[i].p_memsz,
536   						     phdr[i].p_filesz, prot)) != 0)
537   				goto fail;
538 
539 			/*
540 			 * Is this .text or .data ??
541 			 *
542 			 * We only handle one each of those yet XXX
543 			 */
544 			if (hdr->e_entry >= phdr[i].p_vaddr &&
545 			hdr->e_entry <(phdr[i].p_vaddr+phdr[i].p_memsz)) {
546   				text_addr = trunc_page(phdr[i].p_vaddr);
547   				text_size = round_page(phdr[i].p_memsz +
548 						       phdr[i].p_vaddr -
549 						       text_addr);
550 				entry = (u_long)hdr->e_entry;
551 			} else {
552   				data_addr = trunc_page(phdr[i].p_vaddr);
553   				data_size = round_page(phdr[i].p_memsz +
554 						       phdr[i].p_vaddr -
555 						       data_addr);
556 			}
557 			break;
558 	  	case PT_INTERP:	/* Path to interpreter */
559 			if (phdr[i].p_filesz > MAXPATHLEN ||
560 			    phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
561 				error = ENOEXEC;
562 				goto fail;
563 			}
564 			interp = imgp->image_header + phdr[i].p_offset;
565 			break;
566 		case PT_PHDR: 	/* Program header table info */
567 			proghdr = phdr[i].p_vaddr;
568 			break;
569 		default:
570 			break;
571 		}
572 	}
573 
574 	mtx_lock(&vm_mtx);
575 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
576 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
577 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
578 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
579 	mtx_unlock(&vm_mtx);
580 
581 	addr = ELF_RTLD_ADDR(vmspace);
582 
583 	imgp->entry_addr = entry;
584 
585 	brand_info = NULL;
586 
587 	/* We support three types of branding -- (1) the ELF EI_OSABI field
588 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
589 	 * branding w/in the ELF header, and (3) path of the `interp_path'
590 	 * field.  We should also look for an ".note.ABI-tag" ELF section now
591 	 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
592 	 */
593 
594 	/* If the executable has a brand, search for it in the brand list. */
595 	if (brand_info == NULL) {
596 		for (i = 0;  i < MAX_BRANDS;  i++) {
597 			Elf_Brandinfo *bi = elf_brand_list[i];
598 
599 			if (bi != NULL &&
600 			    (hdr->e_ident[EI_OSABI] == bi->brand
601 			    || 0 ==
602 			    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
603 			    bi->compat_3_brand, strlen(bi->compat_3_brand)))) {
604 				brand_info = bi;
605 				break;
606 			}
607 		}
608 	}
609 
610 	/* Lacking a known brand, search for a recognized interpreter. */
611 	if (brand_info == NULL && interp != NULL) {
612 		for (i = 0;  i < MAX_BRANDS;  i++) {
613 			Elf_Brandinfo *bi = elf_brand_list[i];
614 
615 			if (bi != NULL &&
616 			    strcmp(interp, bi->interp_path) == 0) {
617 				brand_info = bi;
618 				break;
619 			}
620 		}
621 	}
622 
623 	/* Lacking a recognized interpreter, try the default brand */
624 	if (brand_info == NULL) {
625 		for (i = 0; i < MAX_BRANDS; i++) {
626 			Elf_Brandinfo *bi = elf_brand_list[i];
627 
628 			if (bi != NULL && fallback_elf_brand == bi->brand) {
629 				brand_info = bi;
630 				break;
631 			}
632 		}
633 	}
634 
635 	if (brand_info == NULL) {
636 		uprintf("ELF binary type \"%u\" not known.\n",
637 		    hdr->e_ident[EI_OSABI]);
638 		error = ENOEXEC;
639 		goto fail;
640 	}
641 
642 	imgp->proc->p_sysent = brand_info->sysvec;
643 	if (interp != NULL) {
644 	        snprintf(path, sizeof(path), "%s%s",
645 			 brand_info->emul_path, interp);
646 		if ((error = elf_load_file(imgp->proc, path, &addr,
647 					   &imgp->entry_addr)) != 0) {
648 		        if ((error = elf_load_file(imgp->proc, interp, &addr,
649 						   &imgp->entry_addr)) != 0) {
650 			        uprintf("ELF interpreter %s not found\n", path);
651 				goto fail;
652 			}
653                 }
654 	}
655 
656 	/*
657 	 * Construct auxargs table (used by the fixup routine)
658 	 */
659 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
660 	elf_auxargs->execfd = -1;
661 	elf_auxargs->phdr = proghdr;
662 	elf_auxargs->phent = hdr->e_phentsize;
663 	elf_auxargs->phnum = hdr->e_phnum;
664 	elf_auxargs->pagesz = PAGE_SIZE;
665 	elf_auxargs->base = addr;
666 	elf_auxargs->flags = 0;
667 	elf_auxargs->entry = entry;
668 	elf_auxargs->trace = elf_trace;
669 
670 	imgp->auxargs = elf_auxargs;
671 	imgp->interpreted = 0;
672 
673 fail:
674 	return error;
675 }
676 
677 static int
678 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
679 {
680 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
681 	register_t *pos;
682 
683 	pos = *stack_base + (imgp->argc + imgp->envc + 2);
684 
685 	if (args->trace) {
686 		AUXARGS_ENTRY(pos, AT_DEBUG, 1);
687 	}
688 	if (args->execfd != -1) {
689 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
690 	}
691 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
692 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
693 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
694 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
695 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
696 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
697 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
698 	AUXARGS_ENTRY(pos, AT_NULL, 0);
699 
700 	free(imgp->auxargs, M_TEMP);
701 	imgp->auxargs = NULL;
702 
703 	(*stack_base)--;
704 	suword(*stack_base, (long) imgp->argc);
705 	return 0;
706 }
707 
708 /*
709  * Code for generating ELF core dumps.
710  */
711 
712 typedef void (*segment_callback) __P((vm_map_entry_t, void *));
713 
714 /* Closure for cb_put_phdr(). */
715 struct phdr_closure {
716 	Elf_Phdr *phdr;		/* Program header to fill in */
717 	Elf_Off offset;		/* Offset of segment in core file */
718 };
719 
720 /* Closure for cb_size_segment(). */
721 struct sseg_closure {
722 	int count;		/* Count of writable segments. */
723 	size_t size;		/* Total size of all writable segments. */
724 };
725 
726 static void cb_put_phdr __P((vm_map_entry_t, void *));
727 static void cb_size_segment __P((vm_map_entry_t, void *));
728 static void each_writable_segment __P((struct proc *, segment_callback,
729     void *));
730 static int elf_corehdr __P((struct proc *, struct vnode *, struct ucred *,
731     int, void *, size_t));
732 static void elf_puthdr __P((struct proc *, void *, size_t *,
733     const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int));
734 static void elf_putnote __P((void *, size_t *, const char *, int,
735     const void *, size_t));
736 
737 extern int osreldate;
738 
739 int
740 elf_coredump(p, vp, limit)
741 	register struct proc *p;
742 	register struct vnode *vp;
743 	off_t limit;
744 {
745 	register struct ucred *cred = p->p_ucred;
746 	int error = 0;
747 	struct sseg_closure seginfo;
748 	void *hdr;
749 	size_t hdrsize;
750 
751 	/* Size the program segments. */
752 	seginfo.count = 0;
753 	seginfo.size = 0;
754 	each_writable_segment(p, cb_size_segment, &seginfo);
755 
756 	/*
757 	 * Calculate the size of the core file header area by making
758 	 * a dry run of generating it.  Nothing is written, but the
759 	 * size is calculated.
760 	 */
761 	hdrsize = 0;
762 	elf_puthdr((struct proc *)NULL, (void *)NULL, &hdrsize,
763 	    (const prstatus_t *)NULL, (const prfpregset_t *)NULL,
764 	    (const prpsinfo_t *)NULL, seginfo.count);
765 
766 	if (hdrsize + seginfo.size >= limit)
767 		return (EFAULT);
768 
769 	/*
770 	 * Allocate memory for building the header, fill it up,
771 	 * and write it out.
772 	 */
773 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
774 	if (hdr == NULL) {
775 		return EINVAL;
776 	}
777 	error = elf_corehdr(p, vp, cred, seginfo.count, hdr, hdrsize);
778 
779 	/* Write the contents of all of the writable segments. */
780 	if (error == 0) {
781 		Elf_Phdr *php;
782 		off_t offset;
783 		int i;
784 
785 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
786 		offset = hdrsize;
787 		for (i = 0;  i < seginfo.count;  i++) {
788 			error = vn_rdwr(UIO_WRITE, vp, (caddr_t)php->p_vaddr,
789 			    php->p_filesz, offset, UIO_USERSPACE,
790 			    IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, p);
791 			if (error != 0)
792 				break;
793 			offset += php->p_filesz;
794 			php++;
795 		}
796 	}
797 	free(hdr, M_TEMP);
798 
799 	return error;
800 }
801 
802 /*
803  * A callback for each_writable_segment() to write out the segment's
804  * program header entry.
805  */
806 static void
807 cb_put_phdr(entry, closure)
808 	vm_map_entry_t entry;
809 	void *closure;
810 {
811 	struct phdr_closure *phc = (struct phdr_closure *)closure;
812 	Elf_Phdr *phdr = phc->phdr;
813 
814 	phc->offset = round_page(phc->offset);
815 
816 	phdr->p_type = PT_LOAD;
817 	phdr->p_offset = phc->offset;
818 	phdr->p_vaddr = entry->start;
819 	phdr->p_paddr = 0;
820 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
821 	phdr->p_align = PAGE_SIZE;
822 	phdr->p_flags = 0;
823 	if (entry->protection & VM_PROT_READ)
824 		phdr->p_flags |= PF_R;
825 	if (entry->protection & VM_PROT_WRITE)
826 		phdr->p_flags |= PF_W;
827 	if (entry->protection & VM_PROT_EXECUTE)
828 		phdr->p_flags |= PF_X;
829 
830 	phc->offset += phdr->p_filesz;
831 	phc->phdr++;
832 }
833 
834 /*
835  * A callback for each_writable_segment() to gather information about
836  * the number of segments and their total size.
837  */
838 static void
839 cb_size_segment(entry, closure)
840 	vm_map_entry_t entry;
841 	void *closure;
842 {
843 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
844 
845 	ssc->count++;
846 	ssc->size += entry->end - entry->start;
847 }
848 
849 /*
850  * For each writable segment in the process's memory map, call the given
851  * function with a pointer to the map entry and some arbitrary
852  * caller-supplied data.
853  */
854 static void
855 each_writable_segment(p, func, closure)
856 	struct proc *p;
857 	segment_callback func;
858 	void *closure;
859 {
860 	vm_map_t map = &p->p_vmspace->vm_map;
861 	vm_map_entry_t entry;
862 
863 	for (entry = map->header.next;  entry != &map->header;
864 	    entry = entry->next) {
865 		vm_object_t obj;
866 
867 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) ||
868 		    (entry->protection & (VM_PROT_READ|VM_PROT_WRITE)) !=
869 		    (VM_PROT_READ|VM_PROT_WRITE))
870 			continue;
871 
872 		/*
873 		** Dont include memory segment in the coredump if
874 		** MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
875 		** madvise(2).
876 		*/
877 		if (entry->eflags & MAP_ENTRY_NOCOREDUMP)
878 			continue;
879 
880 		if ((obj = entry->object.vm_object) == NULL)
881 			continue;
882 
883 		/* Find the deepest backing object. */
884 		while (obj->backing_object != NULL)
885 			obj = obj->backing_object;
886 
887 		/* Ignore memory-mapped devices and such things. */
888 		if (obj->type != OBJT_DEFAULT &&
889 		    obj->type != OBJT_SWAP &&
890 		    obj->type != OBJT_VNODE)
891 			continue;
892 
893 		(*func)(entry, closure);
894 	}
895 }
896 
897 /*
898  * Write the core file header to the file, including padding up to
899  * the page boundary.
900  */
901 static int
902 elf_corehdr(p, vp, cred, numsegs, hdr, hdrsize)
903 	struct proc *p;
904 	struct vnode *vp;
905 	struct ucred *cred;
906 	int numsegs;
907 	size_t hdrsize;
908 	void *hdr;
909 {
910 	size_t off;
911 	prstatus_t status;
912 	prfpregset_t fpregset;
913 	prpsinfo_t psinfo;
914 
915 	/* Gather the information for the header. */
916 	bzero(&status, sizeof status);
917 	status.pr_version = PRSTATUS_VERSION;
918 	status.pr_statussz = sizeof(prstatus_t);
919 	status.pr_gregsetsz = sizeof(gregset_t);
920 	status.pr_fpregsetsz = sizeof(fpregset_t);
921 	status.pr_osreldate = osreldate;
922 	status.pr_cursig = p->p_sig;
923 	status.pr_pid = p->p_pid;
924 	fill_regs(p, &status.pr_reg);
925 
926 	fill_fpregs(p, &fpregset);
927 
928 	bzero(&psinfo, sizeof psinfo);
929 	psinfo.pr_version = PRPSINFO_VERSION;
930 	psinfo.pr_psinfosz = sizeof(prpsinfo_t);
931 	strncpy(psinfo.pr_fname, p->p_comm, MAXCOMLEN);
932 	/* XXX - We don't fill in the command line arguments properly yet. */
933 	strncpy(psinfo.pr_psargs, p->p_comm, PRARGSZ);
934 
935 	/* Fill in the header. */
936 	bzero(hdr, hdrsize);
937 	off = 0;
938 	elf_puthdr(p, hdr, &off, &status, &fpregset, &psinfo, numsegs);
939 
940 	/* Write it to the core file. */
941 	return vn_rdwr(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
942 	    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, NULL, p);
943 }
944 
945 static void
946 elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status,
947     const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
948 {
949 	size_t ehoff;
950 	size_t phoff;
951 	size_t noteoff;
952 	size_t notesz;
953 
954 	ehoff = *off;
955 	*off += sizeof(Elf_Ehdr);
956 
957 	phoff = *off;
958 	*off += (numsegs + 1) * sizeof(Elf_Phdr);
959 
960 	noteoff = *off;
961 	elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status,
962 	    sizeof *status);
963 	elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
964 	    sizeof *fpregset);
965 	elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
966 	    sizeof *psinfo);
967 	notesz = *off - noteoff;
968 
969 	/* Align up to a page boundary for the program segments. */
970 	*off = round_page(*off);
971 
972 	if (dst != NULL) {
973 		Elf_Ehdr *ehdr;
974 		Elf_Phdr *phdr;
975 		struct phdr_closure phc;
976 
977 		/*
978 		 * Fill in the ELF header.
979 		 */
980 		ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
981 		ehdr->e_ident[EI_MAG0] = ELFMAG0;
982 		ehdr->e_ident[EI_MAG1] = ELFMAG1;
983 		ehdr->e_ident[EI_MAG2] = ELFMAG2;
984 		ehdr->e_ident[EI_MAG3] = ELFMAG3;
985 		ehdr->e_ident[EI_CLASS] = ELF_CLASS;
986 		ehdr->e_ident[EI_DATA] = ELF_DATA;
987 		ehdr->e_ident[EI_VERSION] = EV_CURRENT;
988 		ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
989 		ehdr->e_ident[EI_ABIVERSION] = 0;
990 		ehdr->e_ident[EI_PAD] = 0;
991 		ehdr->e_type = ET_CORE;
992 		ehdr->e_machine = ELF_ARCH;
993 		ehdr->e_version = EV_CURRENT;
994 		ehdr->e_entry = 0;
995 		ehdr->e_phoff = phoff;
996 		ehdr->e_flags = 0;
997 		ehdr->e_ehsize = sizeof(Elf_Ehdr);
998 		ehdr->e_phentsize = sizeof(Elf_Phdr);
999 		ehdr->e_phnum = numsegs + 1;
1000 		ehdr->e_shentsize = sizeof(Elf_Shdr);
1001 		ehdr->e_shnum = 0;
1002 		ehdr->e_shstrndx = SHN_UNDEF;
1003 
1004 		/*
1005 		 * Fill in the program header entries.
1006 		 */
1007 		phdr = (Elf_Phdr *)((char *)dst + phoff);
1008 
1009 		/* The note segement. */
1010 		phdr->p_type = PT_NOTE;
1011 		phdr->p_offset = noteoff;
1012 		phdr->p_vaddr = 0;
1013 		phdr->p_paddr = 0;
1014 		phdr->p_filesz = notesz;
1015 		phdr->p_memsz = 0;
1016 		phdr->p_flags = 0;
1017 		phdr->p_align = 0;
1018 		phdr++;
1019 
1020 		/* All the writable segments from the program. */
1021 		phc.phdr = phdr;
1022 		phc.offset = *off;
1023 		each_writable_segment(p, cb_put_phdr, &phc);
1024 	}
1025 }
1026 
1027 static void
1028 elf_putnote(void *dst, size_t *off, const char *name, int type,
1029     const void *desc, size_t descsz)
1030 {
1031 	Elf_Note note;
1032 
1033 	note.n_namesz = strlen(name) + 1;
1034 	note.n_descsz = descsz;
1035 	note.n_type = type;
1036 	if (dst != NULL)
1037 		bcopy(&note, (char *)dst + *off, sizeof note);
1038 	*off += sizeof note;
1039 	if (dst != NULL)
1040 		bcopy(name, (char *)dst + *off, note.n_namesz);
1041 	*off += roundup2(note.n_namesz, sizeof(Elf_Size));
1042 	if (dst != NULL)
1043 		bcopy(desc, (char *)dst + *off, note.n_descsz);
1044 	*off += roundup2(note.n_descsz, sizeof(Elf_Size));
1045 }
1046 
1047 /*
1048  * Tell kern_execve.c about it, with a little help from the linker.
1049  */
1050 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1051 EXEC_SET(elf, elf_execsw);
1052