xref: /freebsd/sys/kern/imgact_elf.c (revision c6ec7d31830ab1c80edae95ad5e4b9dba10c47ac)
1 /*-
2  * Copyright (c) 2000 David O'Brien
3  * Copyright (c) 1995-1996 Søren 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_capsicum.h"
35 #include "opt_compat.h"
36 #include "opt_core.h"
37 
38 #include <sys/param.h>
39 #include <sys/capability.h>
40 #include <sys/exec.h>
41 #include <sys/fcntl.h>
42 #include <sys/imgact.h>
43 #include <sys/imgact_elf.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mutex.h>
49 #include <sys/mman.h>
50 #include <sys/namei.h>
51 #include <sys/pioctl.h>
52 #include <sys/proc.h>
53 #include <sys/procfs.h>
54 #include <sys/racct.h>
55 #include <sys/resourcevar.h>
56 #include <sys/sf_buf.h>
57 #include <sys/smp.h>
58 #include <sys/systm.h>
59 #include <sys/signalvar.h>
60 #include <sys/stat.h>
61 #include <sys/sx.h>
62 #include <sys/syscall.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/vnode.h>
66 #include <sys/syslog.h>
67 #include <sys/eventhandler.h>
68 
69 #include <net/zlib.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_kern.h>
73 #include <vm/vm_param.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_map.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_extern.h>
78 
79 #include <machine/elf.h>
80 #include <machine/md_var.h>
81 
82 #define OLD_EI_BRAND	8
83 
84 static int __elfN(check_header)(const Elf_Ehdr *hdr);
85 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
86     const char *interp, int interp_name_len, int32_t *osrel);
87 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
88     u_long *entry, size_t pagesize);
89 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
90     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
91     size_t pagesize);
92 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
93 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
94     int32_t *osrel);
95 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
96 static boolean_t __elfN(check_note)(struct image_params *imgp,
97     Elf_Brandnote *checknote, int32_t *osrel);
98 static vm_prot_t __elfN(trans_prot)(Elf_Word);
99 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
100 
101 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
102     "");
103 
104 #ifdef COMPRESS_USER_CORES
105 static int compress_core(gzFile, char *, char *, unsigned int,
106     struct thread * td);
107 #define CORE_BUF_SIZE	(16 * 1024)
108 #endif
109 
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112     fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
113     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
115     &__elfN(fallback_brand));
116 
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119     &elf_legacy_coredump, 0, "");
120 
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
123 	1;
124 #else
125 	0;
126 #endif
127 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
128     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
129     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
130 
131 #if __ELF_WORD_SIZE == 32
132 #if defined(__amd64__) || defined(__ia64__)
133 int i386_read_exec = 0;
134 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
135     "enable execution from readable segments");
136 #endif
137 #endif
138 
139 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
140 
141 #define	trunc_page_ps(va, ps)	((va) & ~(ps - 1))
142 #define	round_page_ps(va, ps)	(((va) + (ps - 1)) & ~(ps - 1))
143 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
144 
145 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
146 
147 Elf_Brandnote __elfN(freebsd_brandnote) = {
148 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
149 	.hdr.n_descsz	= sizeof(int32_t),
150 	.hdr.n_type	= 1,
151 	.vendor		= FREEBSD_ABI_VENDOR,
152 	.flags		= BN_TRANSLATE_OSREL,
153 	.trans_osrel	= __elfN(freebsd_trans_osrel)
154 };
155 
156 static boolean_t
157 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
158 {
159 	uintptr_t p;
160 
161 	p = (uintptr_t)(note + 1);
162 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
163 	*osrel = *(const int32_t *)(p);
164 
165 	return (TRUE);
166 }
167 
168 static const char GNU_ABI_VENDOR[] = "GNU";
169 static int GNU_KFREEBSD_ABI_DESC = 3;
170 
171 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
172 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
173 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
174 	.hdr.n_type	= 1,
175 	.vendor		= GNU_ABI_VENDOR,
176 	.flags		= BN_TRANSLATE_OSREL,
177 	.trans_osrel	= kfreebsd_trans_osrel
178 };
179 
180 static boolean_t
181 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
182 {
183 	const Elf32_Word *desc;
184 	uintptr_t p;
185 
186 	p = (uintptr_t)(note + 1);
187 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
188 
189 	desc = (const Elf32_Word *)p;
190 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
191 		return (FALSE);
192 
193 	/*
194 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
195 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
196 	 */
197 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
198 
199 	return (TRUE);
200 }
201 
202 int
203 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
204 {
205 	int i;
206 
207 	for (i = 0; i < MAX_BRANDS; i++) {
208 		if (elf_brand_list[i] == NULL) {
209 			elf_brand_list[i] = entry;
210 			break;
211 		}
212 	}
213 	if (i == MAX_BRANDS) {
214 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
215 			__func__, entry);
216 		return (-1);
217 	}
218 	return (0);
219 }
220 
221 int
222 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
223 {
224 	int i;
225 
226 	for (i = 0; i < MAX_BRANDS; i++) {
227 		if (elf_brand_list[i] == entry) {
228 			elf_brand_list[i] = NULL;
229 			break;
230 		}
231 	}
232 	if (i == MAX_BRANDS)
233 		return (-1);
234 	return (0);
235 }
236 
237 int
238 __elfN(brand_inuse)(Elf_Brandinfo *entry)
239 {
240 	struct proc *p;
241 	int rval = FALSE;
242 
243 	sx_slock(&allproc_lock);
244 	FOREACH_PROC_IN_SYSTEM(p) {
245 		if (p->p_sysent == entry->sysvec) {
246 			rval = TRUE;
247 			break;
248 		}
249 	}
250 	sx_sunlock(&allproc_lock);
251 
252 	return (rval);
253 }
254 
255 static Elf_Brandinfo *
256 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
257     int interp_name_len, int32_t *osrel)
258 {
259 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
260 	Elf_Brandinfo *bi;
261 	boolean_t ret;
262 	int i;
263 
264 	/*
265 	 * We support four types of branding -- (1) the ELF EI_OSABI field
266 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
267 	 * branding w/in the ELF header, (3) path of the `interp_path'
268 	 * field, and (4) the ".note.ABI-tag" ELF section.
269 	 */
270 
271 	/* Look for an ".note.ABI-tag" ELF section */
272 	for (i = 0; i < MAX_BRANDS; i++) {
273 		bi = elf_brand_list[i];
274 		if (bi == NULL)
275 			continue;
276 		if (hdr->e_machine == bi->machine && (bi->flags &
277 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
278 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
279 			if (ret)
280 				return (bi);
281 		}
282 	}
283 
284 	/* If the executable has a brand, search for it in the brand list. */
285 	for (i = 0; i < MAX_BRANDS; i++) {
286 		bi = elf_brand_list[i];
287 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
288 			continue;
289 		if (hdr->e_machine == bi->machine &&
290 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
291 		    strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
292 		    bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
293 			return (bi);
294 	}
295 
296 	/* Lacking a known brand, search for a recognized interpreter. */
297 	if (interp != NULL) {
298 		for (i = 0; i < MAX_BRANDS; i++) {
299 			bi = elf_brand_list[i];
300 			if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
301 				continue;
302 			if (hdr->e_machine == bi->machine &&
303 			    /* ELF image p_filesz includes terminating zero */
304 			    strlen(bi->interp_path) + 1 == interp_name_len &&
305 			    strncmp(interp, bi->interp_path, interp_name_len)
306 			    == 0)
307 				return (bi);
308 		}
309 	}
310 
311 	/* Lacking a recognized interpreter, try the default brand */
312 	for (i = 0; i < MAX_BRANDS; i++) {
313 		bi = elf_brand_list[i];
314 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
315 			continue;
316 		if (hdr->e_machine == bi->machine &&
317 		    __elfN(fallback_brand) == bi->brand)
318 			return (bi);
319 	}
320 	return (NULL);
321 }
322 
323 static int
324 __elfN(check_header)(const Elf_Ehdr *hdr)
325 {
326 	Elf_Brandinfo *bi;
327 	int i;
328 
329 	if (!IS_ELF(*hdr) ||
330 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
331 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
332 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
333 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
334 	    hdr->e_version != ELF_TARG_VER)
335 		return (ENOEXEC);
336 
337 	/*
338 	 * Make sure we have at least one brand for this machine.
339 	 */
340 
341 	for (i = 0; i < MAX_BRANDS; i++) {
342 		bi = elf_brand_list[i];
343 		if (bi != NULL && bi->machine == hdr->e_machine)
344 			break;
345 	}
346 	if (i == MAX_BRANDS)
347 		return (ENOEXEC);
348 
349 	return (0);
350 }
351 
352 static int
353 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
354     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
355 {
356 	struct sf_buf *sf;
357 	int error;
358 	vm_offset_t off;
359 
360 	/*
361 	 * Create the page if it doesn't exist yet. Ignore errors.
362 	 */
363 	vm_map_lock(map);
364 	vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
365 	    VM_PROT_ALL, VM_PROT_ALL, 0);
366 	vm_map_unlock(map);
367 
368 	/*
369 	 * Find the page from the underlying object.
370 	 */
371 	if (object) {
372 		sf = vm_imgact_map_page(object, offset);
373 		if (sf == NULL)
374 			return (KERN_FAILURE);
375 		off = offset - trunc_page(offset);
376 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
377 		    end - start);
378 		vm_imgact_unmap_page(sf);
379 		if (error) {
380 			return (KERN_FAILURE);
381 		}
382 	}
383 
384 	return (KERN_SUCCESS);
385 }
386 
387 static int
388 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
389     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
390 {
391 	struct sf_buf *sf;
392 	vm_offset_t off;
393 	vm_size_t sz;
394 	int error, rv;
395 
396 	if (start != trunc_page(start)) {
397 		rv = __elfN(map_partial)(map, object, offset, start,
398 		    round_page(start), prot);
399 		if (rv)
400 			return (rv);
401 		offset += round_page(start) - start;
402 		start = round_page(start);
403 	}
404 	if (end != round_page(end)) {
405 		rv = __elfN(map_partial)(map, object, offset +
406 		    trunc_page(end) - start, trunc_page(end), end, prot);
407 		if (rv)
408 			return (rv);
409 		end = trunc_page(end);
410 	}
411 	if (end > start) {
412 		if (offset & PAGE_MASK) {
413 			/*
414 			 * The mapping is not page aligned. This means we have
415 			 * to copy the data. Sigh.
416 			 */
417 			rv = vm_map_find(map, NULL, 0, &start, end - start,
418 			    FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
419 			if (rv)
420 				return (rv);
421 			if (object == NULL)
422 				return (KERN_SUCCESS);
423 			for (; start < end; start += sz) {
424 				sf = vm_imgact_map_page(object, offset);
425 				if (sf == NULL)
426 					return (KERN_FAILURE);
427 				off = offset - trunc_page(offset);
428 				sz = end - start;
429 				if (sz > PAGE_SIZE - off)
430 					sz = PAGE_SIZE - off;
431 				error = copyout((caddr_t)sf_buf_kva(sf) + off,
432 				    (caddr_t)start, sz);
433 				vm_imgact_unmap_page(sf);
434 				if (error) {
435 					return (KERN_FAILURE);
436 				}
437 				offset += sz;
438 			}
439 			rv = KERN_SUCCESS;
440 		} else {
441 			vm_object_reference(object);
442 			vm_map_lock(map);
443 			rv = vm_map_insert(map, object, offset, start, end,
444 			    prot, VM_PROT_ALL, cow);
445 			vm_map_unlock(map);
446 			if (rv != KERN_SUCCESS)
447 				vm_object_deallocate(object);
448 		}
449 		return (rv);
450 	} else {
451 		return (KERN_SUCCESS);
452 	}
453 }
454 
455 static int
456 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
457     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
458     size_t pagesize)
459 {
460 	struct sf_buf *sf;
461 	size_t map_len;
462 	vm_map_t map;
463 	vm_object_t object;
464 	vm_offset_t map_addr;
465 	int error, rv, cow;
466 	size_t copy_len;
467 	vm_offset_t file_addr;
468 
469 	/*
470 	 * It's necessary to fail if the filsz + offset taken from the
471 	 * header is greater than the actual file pager object's size.
472 	 * If we were to allow this, then the vm_map_find() below would
473 	 * walk right off the end of the file object and into the ether.
474 	 *
475 	 * While I'm here, might as well check for something else that
476 	 * is invalid: filsz cannot be greater than memsz.
477 	 */
478 	if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
479 		uprintf("elf_load_section: truncated ELF file\n");
480 		return (ENOEXEC);
481 	}
482 
483 	object = imgp->object;
484 	map = &imgp->proc->p_vmspace->vm_map;
485 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
486 	file_addr = trunc_page_ps(offset, pagesize);
487 
488 	/*
489 	 * We have two choices.  We can either clear the data in the last page
490 	 * of an oversized mapping, or we can start the anon mapping a page
491 	 * early and copy the initialized data into that first page.  We
492 	 * choose the second..
493 	 */
494 	if (memsz > filsz)
495 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
496 	else
497 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
498 
499 	if (map_len != 0) {
500 		/* cow flags: don't dump readonly sections in core */
501 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
502 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
503 
504 		rv = __elfN(map_insert)(map,
505 				      object,
506 				      file_addr,	/* file offset */
507 				      map_addr,		/* virtual start */
508 				      map_addr + map_len,/* virtual end */
509 				      prot,
510 				      cow);
511 		if (rv != KERN_SUCCESS)
512 			return (EINVAL);
513 
514 		/* we can stop now if we've covered it all */
515 		if (memsz == filsz) {
516 			return (0);
517 		}
518 	}
519 
520 
521 	/*
522 	 * We have to get the remaining bit of the file into the first part
523 	 * of the oversized map segment.  This is normally because the .data
524 	 * segment in the file is extended to provide bss.  It's a neat idea
525 	 * to try and save a page, but it's a pain in the behind to implement.
526 	 */
527 	copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
528 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
529 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
530 	    map_addr;
531 
532 	/* This had damn well better be true! */
533 	if (map_len != 0) {
534 		rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
535 		    map_len, VM_PROT_ALL, 0);
536 		if (rv != KERN_SUCCESS) {
537 			return (EINVAL);
538 		}
539 	}
540 
541 	if (copy_len != 0) {
542 		vm_offset_t off;
543 
544 		sf = vm_imgact_map_page(object, offset + filsz);
545 		if (sf == NULL)
546 			return (EIO);
547 
548 		/* send the page fragment to user space */
549 		off = trunc_page_ps(offset + filsz, pagesize) -
550 		    trunc_page(offset + filsz);
551 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
552 		    (caddr_t)map_addr, copy_len);
553 		vm_imgact_unmap_page(sf);
554 		if (error) {
555 			return (error);
556 		}
557 	}
558 
559 	/*
560 	 * set it to the specified protection.
561 	 * XXX had better undo the damage from pasting over the cracks here!
562 	 */
563 	vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
564 	    map_len), prot, FALSE);
565 
566 	return (0);
567 }
568 
569 /*
570  * Load the file "file" into memory.  It may be either a shared object
571  * or an executable.
572  *
573  * The "addr" reference parameter is in/out.  On entry, it specifies
574  * the address where a shared object should be loaded.  If the file is
575  * an executable, this value is ignored.  On exit, "addr" specifies
576  * where the file was actually loaded.
577  *
578  * The "entry" reference parameter is out only.  On exit, it specifies
579  * the entry point for the loaded file.
580  */
581 static int
582 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
583 	u_long *entry, size_t pagesize)
584 {
585 	struct {
586 		struct nameidata nd;
587 		struct vattr attr;
588 		struct image_params image_params;
589 	} *tempdata;
590 	const Elf_Ehdr *hdr = NULL;
591 	const Elf_Phdr *phdr = NULL;
592 	struct nameidata *nd;
593 	struct vattr *attr;
594 	struct image_params *imgp;
595 	vm_prot_t prot;
596 	u_long rbase;
597 	u_long base_addr = 0;
598 	int error, i, numsegs;
599 
600 #ifdef CAPABILITY_MODE
601 	/*
602 	 * XXXJA: This check can go away once we are sufficiently confident
603 	 * that the checks in namei() are correct.
604 	 */
605 	if (IN_CAPABILITY_MODE(curthread))
606 		return (ECAPMODE);
607 #endif
608 
609 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
610 	nd = &tempdata->nd;
611 	attr = &tempdata->attr;
612 	imgp = &tempdata->image_params;
613 
614 	/*
615 	 * Initialize part of the common data
616 	 */
617 	imgp->proc = p;
618 	imgp->attr = attr;
619 	imgp->firstpage = NULL;
620 	imgp->image_header = NULL;
621 	imgp->object = NULL;
622 	imgp->execlabel = NULL;
623 
624 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
625 	if ((error = namei(nd)) != 0) {
626 		nd->ni_vp = NULL;
627 		goto fail;
628 	}
629 	NDFREE(nd, NDF_ONLY_PNBUF);
630 	imgp->vp = nd->ni_vp;
631 
632 	/*
633 	 * Check permissions, modes, uid, etc on the file, and "open" it.
634 	 */
635 	error = exec_check_permissions(imgp);
636 	if (error)
637 		goto fail;
638 
639 	error = exec_map_first_page(imgp);
640 	if (error)
641 		goto fail;
642 
643 	/*
644 	 * Also make certain that the interpreter stays the same, so set
645 	 * its VV_TEXT flag, too.
646 	 */
647 	VOP_SET_TEXT(nd->ni_vp);
648 
649 	imgp->object = nd->ni_vp->v_object;
650 
651 	hdr = (const Elf_Ehdr *)imgp->image_header;
652 	if ((error = __elfN(check_header)(hdr)) != 0)
653 		goto fail;
654 	if (hdr->e_type == ET_DYN)
655 		rbase = *addr;
656 	else if (hdr->e_type == ET_EXEC)
657 		rbase = 0;
658 	else {
659 		error = ENOEXEC;
660 		goto fail;
661 	}
662 
663 	/* Only support headers that fit within first page for now      */
664 	/*    (multiplication of two Elf_Half fields will not overflow) */
665 	if ((hdr->e_phoff > PAGE_SIZE) ||
666 	    (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
667 		error = ENOEXEC;
668 		goto fail;
669 	}
670 
671 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
672 	if (!aligned(phdr, Elf_Addr)) {
673 		error = ENOEXEC;
674 		goto fail;
675 	}
676 
677 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
678 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
679 			/* Loadable segment */
680 			prot = __elfN(trans_prot)(phdr[i].p_flags);
681 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
682 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
683 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
684 			if (error != 0)
685 				goto fail;
686 			/*
687 			 * Establish the base address if this is the
688 			 * first segment.
689 			 */
690 			if (numsegs == 0)
691   				base_addr = trunc_page(phdr[i].p_vaddr +
692 				    rbase);
693 			numsegs++;
694 		}
695 	}
696 	*addr = base_addr;
697 	*entry = (unsigned long)hdr->e_entry + rbase;
698 
699 fail:
700 	if (imgp->firstpage)
701 		exec_unmap_first_page(imgp);
702 
703 	if (nd->ni_vp)
704 		vput(nd->ni_vp);
705 
706 	free(tempdata, M_TEMP);
707 
708 	return (error);
709 }
710 
711 static int
712 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
713 {
714 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
715 	const Elf_Phdr *phdr;
716 	Elf_Auxargs *elf_auxargs;
717 	struct vmspace *vmspace;
718 	vm_prot_t prot;
719 	u_long text_size = 0, data_size = 0, total_size = 0;
720 	u_long text_addr = 0, data_addr = 0;
721 	u_long seg_size, seg_addr;
722 	u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
723 	int32_t osrel = 0;
724 	int error = 0, i, n, interp_name_len = 0;
725 	const char *interp = NULL, *newinterp = NULL;
726 	Elf_Brandinfo *brand_info;
727 	char *path;
728 	struct sysentvec *sv;
729 
730 	/*
731 	 * Do we have a valid ELF header ?
732 	 *
733 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
734 	 * if particular brand doesn't support it.
735 	 */
736 	if (__elfN(check_header)(hdr) != 0 ||
737 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
738 		return (-1);
739 
740 	/*
741 	 * From here on down, we return an errno, not -1, as we've
742 	 * detected an ELF file.
743 	 */
744 
745 	if ((hdr->e_phoff > PAGE_SIZE) ||
746 	    (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
747 		/* Only support headers in first page for now */
748 		return (ENOEXEC);
749 	}
750 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
751 	if (!aligned(phdr, Elf_Addr))
752 		return (ENOEXEC);
753 	n = 0;
754 	baddr = 0;
755 	for (i = 0; i < hdr->e_phnum; i++) {
756 		switch (phdr[i].p_type) {
757 		case PT_LOAD:
758 			if (n == 0)
759 				baddr = phdr[i].p_vaddr;
760 			n++;
761 			break;
762 		case PT_INTERP:
763 			/* Path to interpreter */
764 			if (phdr[i].p_filesz > MAXPATHLEN ||
765 			    phdr[i].p_offset >= PAGE_SIZE ||
766 			    phdr[i].p_offset + phdr[i].p_filesz >= PAGE_SIZE)
767 				return (ENOEXEC);
768 			interp = imgp->image_header + phdr[i].p_offset;
769 			interp_name_len = phdr[i].p_filesz;
770 			break;
771 		case PT_GNU_STACK:
772 			if (__elfN(nxstack))
773 				imgp->stack_prot =
774 				    __elfN(trans_prot)(phdr[i].p_flags);
775 			break;
776 		}
777 	}
778 
779 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
780 	    &osrel);
781 	if (brand_info == NULL) {
782 		uprintf("ELF binary type \"%u\" not known.\n",
783 		    hdr->e_ident[EI_OSABI]);
784 		return (ENOEXEC);
785 	}
786 	if (hdr->e_type == ET_DYN) {
787 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
788 			return (ENOEXEC);
789 		/*
790 		 * Honour the base load address from the dso if it is
791 		 * non-zero for some reason.
792 		 */
793 		if (baddr == 0)
794 			et_dyn_addr = ET_DYN_LOAD_ADDR;
795 		else
796 			et_dyn_addr = 0;
797 	} else
798 		et_dyn_addr = 0;
799 	sv = brand_info->sysvec;
800 	if (interp != NULL && brand_info->interp_newpath != NULL)
801 		newinterp = brand_info->interp_newpath;
802 
803 	/*
804 	 * Avoid a possible deadlock if the current address space is destroyed
805 	 * and that address space maps the locked vnode.  In the common case,
806 	 * the locked vnode's v_usecount is decremented but remains greater
807 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
808 	 * However, in cases where the vnode lock is external, such as nullfs,
809 	 * v_usecount may become zero.
810 	 *
811 	 * The VV_TEXT flag prevents modifications to the executable while
812 	 * the vnode is unlocked.
813 	 */
814 	VOP_UNLOCK(imgp->vp, 0);
815 
816 	error = exec_new_vmspace(imgp, sv);
817 	imgp->proc->p_sysent = sv;
818 
819 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
820 	if (error)
821 		return (error);
822 
823 	for (i = 0; i < hdr->e_phnum; i++) {
824 		switch (phdr[i].p_type) {
825 		case PT_LOAD:	/* Loadable segment */
826 			if (phdr[i].p_memsz == 0)
827 				break;
828 			prot = __elfN(trans_prot)(phdr[i].p_flags);
829 
830 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
831 			/*
832 			 * Some x86 binaries assume read == executable,
833 			 * notably the M3 runtime and therefore cvsup
834 			 */
835 			if (prot & VM_PROT_READ)
836 				prot |= VM_PROT_EXECUTE;
837 #endif
838 
839 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
840 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
841 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
842 			    sv->sv_pagesize);
843 			if (error != 0)
844 				return (error);
845 
846 			/*
847 			 * If this segment contains the program headers,
848 			 * remember their virtual address for the AT_PHDR
849 			 * aux entry. Static binaries don't usually include
850 			 * a PT_PHDR entry.
851 			 */
852 			if (phdr[i].p_offset == 0 &&
853 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
854 				<= phdr[i].p_filesz)
855 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
856 				    et_dyn_addr;
857 
858 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
859 			seg_size = round_page(phdr[i].p_memsz +
860 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
861 
862 			/*
863 			 * Make the largest executable segment the official
864 			 * text segment and all others data.
865 			 *
866 			 * Note that obreak() assumes that data_addr +
867 			 * data_size == end of data load area, and the ELF
868 			 * file format expects segments to be sorted by
869 			 * address.  If multiple data segments exist, the
870 			 * last one will be used.
871 			 */
872 
873 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
874 				text_size = seg_size;
875 				text_addr = seg_addr;
876 			} else {
877 				data_size = seg_size;
878 				data_addr = seg_addr;
879 			}
880 			total_size += seg_size;
881 			break;
882 		case PT_PHDR: 	/* Program header table info */
883 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
884 			break;
885 		default:
886 			break;
887 		}
888 	}
889 
890 	if (data_addr == 0 && data_size == 0) {
891 		data_addr = text_addr;
892 		data_size = text_size;
893 	}
894 
895 	entry = (u_long)hdr->e_entry + et_dyn_addr;
896 
897 	/*
898 	 * Check limits.  It should be safe to check the
899 	 * limits after loading the segments since we do
900 	 * not actually fault in all the segments pages.
901 	 */
902 	PROC_LOCK(imgp->proc);
903 	if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
904 	    text_size > maxtsiz ||
905 	    total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
906 	    racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
907 	    racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
908 		PROC_UNLOCK(imgp->proc);
909 		return (ENOMEM);
910 	}
911 
912 	vmspace = imgp->proc->p_vmspace;
913 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
914 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
915 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
916 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
917 
918 	/*
919 	 * We load the dynamic linker where a userland call
920 	 * to mmap(0, ...) would put it.  The rationale behind this
921 	 * calculation is that it leaves room for the heap to grow to
922 	 * its maximum allowed size.
923 	 */
924 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
925 	    RLIMIT_DATA));
926 	PROC_UNLOCK(imgp->proc);
927 
928 	imgp->entry_addr = entry;
929 
930 	if (interp != NULL) {
931 		int have_interp = FALSE;
932 		VOP_UNLOCK(imgp->vp, 0);
933 		if (brand_info->emul_path != NULL &&
934 		    brand_info->emul_path[0] != '\0') {
935 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
936 			snprintf(path, MAXPATHLEN, "%s%s",
937 			    brand_info->emul_path, interp);
938 			error = __elfN(load_file)(imgp->proc, path, &addr,
939 			    &imgp->entry_addr, sv->sv_pagesize);
940 			free(path, M_TEMP);
941 			if (error == 0)
942 				have_interp = TRUE;
943 		}
944 		if (!have_interp && newinterp != NULL) {
945 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
946 			    &imgp->entry_addr, sv->sv_pagesize);
947 			if (error == 0)
948 				have_interp = TRUE;
949 		}
950 		if (!have_interp) {
951 			error = __elfN(load_file)(imgp->proc, interp, &addr,
952 			    &imgp->entry_addr, sv->sv_pagesize);
953 		}
954 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
955 		if (error != 0) {
956 			uprintf("ELF interpreter %s not found\n", interp);
957 			return (error);
958 		}
959 	} else
960 		addr = et_dyn_addr;
961 
962 	/*
963 	 * Construct auxargs table (used by the fixup routine)
964 	 */
965 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
966 	elf_auxargs->execfd = -1;
967 	elf_auxargs->phdr = proghdr;
968 	elf_auxargs->phent = hdr->e_phentsize;
969 	elf_auxargs->phnum = hdr->e_phnum;
970 	elf_auxargs->pagesz = PAGE_SIZE;
971 	elf_auxargs->base = addr;
972 	elf_auxargs->flags = 0;
973 	elf_auxargs->entry = entry;
974 
975 	imgp->auxargs = elf_auxargs;
976 	imgp->interpreted = 0;
977 	imgp->reloc_base = addr;
978 	imgp->proc->p_osrel = osrel;
979 
980 	return (error);
981 }
982 
983 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
984 
985 int
986 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
987 {
988 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
989 	Elf_Addr *base;
990 	Elf_Addr *pos;
991 
992 	base = (Elf_Addr *)*stack_base;
993 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
994 
995 	if (args->execfd != -1)
996 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
997 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
998 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
999 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1000 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1001 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1002 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1003 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1004 	if (imgp->execpathp != 0)
1005 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1006 	AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate);
1007 	if (imgp->canary != 0) {
1008 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1009 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1010 	}
1011 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1012 	if (imgp->pagesizes != 0) {
1013 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1014 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1015 	}
1016 	if (imgp->sysent->sv_timekeep_base != 0) {
1017 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1018 		    imgp->sysent->sv_timekeep_base);
1019 	}
1020 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1021 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1022 	    imgp->sysent->sv_stackprot);
1023 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1024 
1025 	free(imgp->auxargs, M_TEMP);
1026 	imgp->auxargs = NULL;
1027 
1028 	base--;
1029 	suword(base, (long)imgp->args->argc);
1030 	*stack_base = (register_t *)base;
1031 	return (0);
1032 }
1033 
1034 /*
1035  * Code for generating ELF core dumps.
1036  */
1037 
1038 typedef void (*segment_callback)(vm_map_entry_t, void *);
1039 
1040 /* Closure for cb_put_phdr(). */
1041 struct phdr_closure {
1042 	Elf_Phdr *phdr;		/* Program header to fill in */
1043 	Elf_Off offset;		/* Offset of segment in core file */
1044 };
1045 
1046 /* Closure for cb_size_segment(). */
1047 struct sseg_closure {
1048 	int count;		/* Count of writable segments. */
1049 	size_t size;		/* Total size of all writable segments. */
1050 };
1051 
1052 static void cb_put_phdr(vm_map_entry_t, void *);
1053 static void cb_size_segment(vm_map_entry_t, void *);
1054 static void each_writable_segment(struct thread *, segment_callback, void *);
1055 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1056     int, void *, size_t, gzFile);
1057 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
1058 static void __elfN(putnote)(void *, size_t *, const char *, int,
1059     const void *, size_t);
1060 
1061 #ifdef COMPRESS_USER_CORES
1062 extern int compress_user_cores;
1063 extern int compress_user_cores_gzlevel;
1064 #endif
1065 
1066 static int
1067 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1068     struct ucred *active_cred, struct ucred *file_cred,
1069     struct thread *td, char *core_buf, gzFile gzfile) {
1070 
1071 	int error;
1072 	if (gzfile) {
1073 #ifdef COMPRESS_USER_CORES
1074 		error = compress_core(gzfile, base, core_buf, len, td);
1075 #else
1076 		panic("shouldn't be here");
1077 #endif
1078 	} else {
1079 		error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1080 		    UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred,
1081 		    NULL, td);
1082 	}
1083 	return (error);
1084 }
1085 
1086 int
1087 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1088 {
1089 	struct ucred *cred = td->td_ucred;
1090 	int error = 0;
1091 	struct sseg_closure seginfo;
1092 	void *hdr;
1093 	size_t hdrsize;
1094 
1095 	gzFile gzfile = Z_NULL;
1096 	char *core_buf = NULL;
1097 #ifdef COMPRESS_USER_CORES
1098 	char gzopen_flags[8];
1099 	char *p;
1100 	int doing_compress = flags & IMGACT_CORE_COMPRESS;
1101 #endif
1102 
1103 	hdr = NULL;
1104 
1105 #ifdef COMPRESS_USER_CORES
1106         if (doing_compress) {
1107                 p = gzopen_flags;
1108                 *p++ = 'w';
1109                 if (compress_user_cores_gzlevel >= 0 &&
1110                     compress_user_cores_gzlevel <= 9)
1111                         *p++ = '0' + compress_user_cores_gzlevel;
1112                 *p = 0;
1113                 gzfile = gz_open("", gzopen_flags, vp);
1114                 if (gzfile == Z_NULL) {
1115                         error = EFAULT;
1116                         goto done;
1117                 }
1118                 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1119                 if (!core_buf) {
1120                         error = ENOMEM;
1121                         goto done;
1122                 }
1123         }
1124 #endif
1125 
1126 	/* Size the program segments. */
1127 	seginfo.count = 0;
1128 	seginfo.size = 0;
1129 	each_writable_segment(td, cb_size_segment, &seginfo);
1130 
1131 	/*
1132 	 * Calculate the size of the core file header area by making
1133 	 * a dry run of generating it.  Nothing is written, but the
1134 	 * size is calculated.
1135 	 */
1136 	hdrsize = 0;
1137 	__elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
1138 
1139 #ifdef RACCT
1140 	PROC_LOCK(td->td_proc);
1141 	error = racct_add(td->td_proc, RACCT_CORE, hdrsize + seginfo.size);
1142 	PROC_UNLOCK(td->td_proc);
1143 	if (error != 0) {
1144 		error = EFAULT;
1145 		goto done;
1146 	}
1147 #endif
1148 	if (hdrsize + seginfo.size >= limit) {
1149 		error = EFAULT;
1150 		goto done;
1151 	}
1152 
1153 	/*
1154 	 * Allocate memory for building the header, fill it up,
1155 	 * and write it out.
1156 	 */
1157 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1158 	if (hdr == NULL) {
1159 		error = EINVAL;
1160 		goto done;
1161 	}
1162 	error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1163 	    gzfile);
1164 
1165 	/* Write the contents of all of the writable segments. */
1166 	if (error == 0) {
1167 		Elf_Phdr *php;
1168 		off_t offset;
1169 		int i;
1170 
1171 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1172 		offset = hdrsize;
1173 		for (i = 0; i < seginfo.count; i++) {
1174 			error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1175 			    php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1176 			if (error != 0)
1177 				break;
1178 			offset += php->p_filesz;
1179 			php++;
1180 		}
1181 	}
1182 	if (error) {
1183 		log(LOG_WARNING,
1184 		    "Failed to write core file for process %s (error %d)\n",
1185 		    curproc->p_comm, error);
1186 	}
1187 
1188 done:
1189 #ifdef COMPRESS_USER_CORES
1190 	if (core_buf)
1191 		free(core_buf, M_TEMP);
1192 	if (gzfile)
1193 		gzclose(gzfile);
1194 #endif
1195 
1196 	free(hdr, M_TEMP);
1197 
1198 	return (error);
1199 }
1200 
1201 /*
1202  * A callback for each_writable_segment() to write out the segment's
1203  * program header entry.
1204  */
1205 static void
1206 cb_put_phdr(entry, closure)
1207 	vm_map_entry_t entry;
1208 	void *closure;
1209 {
1210 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1211 	Elf_Phdr *phdr = phc->phdr;
1212 
1213 	phc->offset = round_page(phc->offset);
1214 
1215 	phdr->p_type = PT_LOAD;
1216 	phdr->p_offset = phc->offset;
1217 	phdr->p_vaddr = entry->start;
1218 	phdr->p_paddr = 0;
1219 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1220 	phdr->p_align = PAGE_SIZE;
1221 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1222 
1223 	phc->offset += phdr->p_filesz;
1224 	phc->phdr++;
1225 }
1226 
1227 /*
1228  * A callback for each_writable_segment() to gather information about
1229  * the number of segments and their total size.
1230  */
1231 static void
1232 cb_size_segment(entry, closure)
1233 	vm_map_entry_t entry;
1234 	void *closure;
1235 {
1236 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1237 
1238 	ssc->count++;
1239 	ssc->size += entry->end - entry->start;
1240 }
1241 
1242 /*
1243  * For each writable segment in the process's memory map, call the given
1244  * function with a pointer to the map entry and some arbitrary
1245  * caller-supplied data.
1246  */
1247 static void
1248 each_writable_segment(td, func, closure)
1249 	struct thread *td;
1250 	segment_callback func;
1251 	void *closure;
1252 {
1253 	struct proc *p = td->td_proc;
1254 	vm_map_t map = &p->p_vmspace->vm_map;
1255 	vm_map_entry_t entry;
1256 	vm_object_t backing_object, object;
1257 	boolean_t ignore_entry;
1258 
1259 	vm_map_lock_read(map);
1260 	for (entry = map->header.next; entry != &map->header;
1261 	    entry = entry->next) {
1262 		/*
1263 		 * Don't dump inaccessible mappings, deal with legacy
1264 		 * coredump mode.
1265 		 *
1266 		 * Note that read-only segments related to the elf binary
1267 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1268 		 * need to arbitrarily ignore such segments.
1269 		 */
1270 		if (elf_legacy_coredump) {
1271 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1272 				continue;
1273 		} else {
1274 			if ((entry->protection & VM_PROT_ALL) == 0)
1275 				continue;
1276 		}
1277 
1278 		/*
1279 		 * Dont include memory segment in the coredump if
1280 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1281 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1282 		 * kernel map).
1283 		 */
1284 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1285 			continue;
1286 
1287 		if ((object = entry->object.vm_object) == NULL)
1288 			continue;
1289 
1290 		/* Ignore memory-mapped devices and such things. */
1291 		VM_OBJECT_LOCK(object);
1292 		while ((backing_object = object->backing_object) != NULL) {
1293 			VM_OBJECT_LOCK(backing_object);
1294 			VM_OBJECT_UNLOCK(object);
1295 			object = backing_object;
1296 		}
1297 		ignore_entry = object->type != OBJT_DEFAULT &&
1298 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1299 		VM_OBJECT_UNLOCK(object);
1300 		if (ignore_entry)
1301 			continue;
1302 
1303 		(*func)(entry, closure);
1304 	}
1305 	vm_map_unlock_read(map);
1306 }
1307 
1308 /*
1309  * Write the core file header to the file, including padding up to
1310  * the page boundary.
1311  */
1312 static int
1313 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize, gzfile)
1314 	struct thread *td;
1315 	struct vnode *vp;
1316 	struct ucred *cred;
1317 	int numsegs;
1318 	size_t hdrsize;
1319 	void *hdr;
1320 	gzFile gzfile;
1321 {
1322 	size_t off;
1323 
1324 	/* Fill in the header. */
1325 	bzero(hdr, hdrsize);
1326 	off = 0;
1327 	__elfN(puthdr)(td, hdr, &off, numsegs);
1328 
1329 	if (!gzfile) {
1330 		/* Write it to the core file. */
1331 		return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1332 			UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1333 			td));
1334 	} else {
1335 #ifdef COMPRESS_USER_CORES
1336 		if (gzwrite(gzfile, hdr, hdrsize) != hdrsize) {
1337 			log(LOG_WARNING,
1338 			    "Failed to compress core file header for process"
1339 			    " %s.\n", curproc->p_comm);
1340 			return (EFAULT);
1341 		}
1342 		else {
1343 			return (0);
1344 		}
1345 #else
1346 		panic("shouldn't be here");
1347 #endif
1348 	}
1349 }
1350 
1351 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1352 #include <compat/freebsd32/freebsd32.h>
1353 
1354 typedef struct prstatus32 elf_prstatus_t;
1355 typedef struct prpsinfo32 elf_prpsinfo_t;
1356 typedef struct fpreg32 elf_prfpregset_t;
1357 typedef struct fpreg32 elf_fpregset_t;
1358 typedef struct reg32 elf_gregset_t;
1359 typedef struct thrmisc32 elf_thrmisc_t;
1360 #else
1361 typedef prstatus_t elf_prstatus_t;
1362 typedef prpsinfo_t elf_prpsinfo_t;
1363 typedef prfpregset_t elf_prfpregset_t;
1364 typedef prfpregset_t elf_fpregset_t;
1365 typedef gregset_t elf_gregset_t;
1366 typedef thrmisc_t elf_thrmisc_t;
1367 #endif
1368 
1369 static void
1370 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1371 {
1372 	struct {
1373 		elf_prstatus_t status;
1374 		elf_prfpregset_t fpregset;
1375 		elf_prpsinfo_t psinfo;
1376 		elf_thrmisc_t thrmisc;
1377 	} *tempdata;
1378 	elf_prstatus_t *status;
1379 	elf_prfpregset_t *fpregset;
1380 	elf_prpsinfo_t *psinfo;
1381 	elf_thrmisc_t *thrmisc;
1382 	struct proc *p;
1383 	struct thread *thr;
1384 	size_t ehoff, noteoff, notesz, phoff;
1385 
1386 	p = td->td_proc;
1387 
1388 	ehoff = *off;
1389 	*off += sizeof(Elf_Ehdr);
1390 
1391 	phoff = *off;
1392 	*off += (numsegs + 1) * sizeof(Elf_Phdr);
1393 
1394 	noteoff = *off;
1395 	/*
1396 	 * Don't allocate space for the notes if we're just calculating
1397 	 * the size of the header. We also don't collect the data.
1398 	 */
1399 	if (dst != NULL) {
1400 		tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1401 		status = &tempdata->status;
1402 		fpregset = &tempdata->fpregset;
1403 		psinfo = &tempdata->psinfo;
1404 		thrmisc = &tempdata->thrmisc;
1405 	} else {
1406 		tempdata = NULL;
1407 		status = NULL;
1408 		fpregset = NULL;
1409 		psinfo = NULL;
1410 		thrmisc = NULL;
1411 	}
1412 
1413 	if (dst != NULL) {
1414 		psinfo->pr_version = PRPSINFO_VERSION;
1415 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1416 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1417 		/*
1418 		 * XXX - We don't fill in the command line arguments properly
1419 		 * yet.
1420 		 */
1421 		strlcpy(psinfo->pr_psargs, p->p_comm,
1422 		    sizeof(psinfo->pr_psargs));
1423 	}
1424 	__elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1425 	    sizeof *psinfo);
1426 
1427 	/*
1428 	 * To have the debugger select the right thread (LWP) as the initial
1429 	 * thread, we dump the state of the thread passed to us in td first.
1430 	 * This is the thread that causes the core dump and thus likely to
1431 	 * be the right thread one wants to have selected in the debugger.
1432 	 */
1433 	thr = td;
1434 	while (thr != NULL) {
1435 		if (dst != NULL) {
1436 			status->pr_version = PRSTATUS_VERSION;
1437 			status->pr_statussz = sizeof(elf_prstatus_t);
1438 			status->pr_gregsetsz = sizeof(elf_gregset_t);
1439 			status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1440 			status->pr_osreldate = osreldate;
1441 			status->pr_cursig = p->p_sig;
1442 			status->pr_pid = thr->td_tid;
1443 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1444 			fill_regs32(thr, &status->pr_reg);
1445 			fill_fpregs32(thr, fpregset);
1446 #else
1447 			fill_regs(thr, &status->pr_reg);
1448 			fill_fpregs(thr, fpregset);
1449 #endif
1450 			memset(&thrmisc->_pad, 0, sizeof (thrmisc->_pad));
1451 			strcpy(thrmisc->pr_tname, thr->td_name);
1452 		}
1453 		__elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1454 		    sizeof *status);
1455 		__elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1456 		    sizeof *fpregset);
1457 		__elfN(putnote)(dst, off, "FreeBSD", NT_THRMISC, thrmisc,
1458 		    sizeof *thrmisc);
1459 		/*
1460 		 * Allow for MD specific notes, as well as any MD
1461 		 * specific preparations for writing MI notes.
1462 		 */
1463 		__elfN(dump_thread)(thr, dst, off);
1464 
1465 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1466 		    TAILQ_NEXT(thr, td_plist);
1467 		if (thr == td)
1468 			thr = TAILQ_NEXT(thr, td_plist);
1469 	}
1470 
1471 	notesz = *off - noteoff;
1472 
1473 	if (dst != NULL)
1474 		free(tempdata, M_TEMP);
1475 
1476 	/* Align up to a page boundary for the program segments. */
1477 	*off = round_page(*off);
1478 
1479 	if (dst != NULL) {
1480 		Elf_Ehdr *ehdr;
1481 		Elf_Phdr *phdr;
1482 		struct phdr_closure phc;
1483 
1484 		/*
1485 		 * Fill in the ELF header.
1486 		 */
1487 		ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1488 		ehdr->e_ident[EI_MAG0] = ELFMAG0;
1489 		ehdr->e_ident[EI_MAG1] = ELFMAG1;
1490 		ehdr->e_ident[EI_MAG2] = ELFMAG2;
1491 		ehdr->e_ident[EI_MAG3] = ELFMAG3;
1492 		ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1493 		ehdr->e_ident[EI_DATA] = ELF_DATA;
1494 		ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1495 		ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1496 		ehdr->e_ident[EI_ABIVERSION] = 0;
1497 		ehdr->e_ident[EI_PAD] = 0;
1498 		ehdr->e_type = ET_CORE;
1499 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1500 		ehdr->e_machine = ELF_ARCH32;
1501 #else
1502 		ehdr->e_machine = ELF_ARCH;
1503 #endif
1504 		ehdr->e_version = EV_CURRENT;
1505 		ehdr->e_entry = 0;
1506 		ehdr->e_phoff = phoff;
1507 		ehdr->e_flags = 0;
1508 		ehdr->e_ehsize = sizeof(Elf_Ehdr);
1509 		ehdr->e_phentsize = sizeof(Elf_Phdr);
1510 		ehdr->e_phnum = numsegs + 1;
1511 		ehdr->e_shentsize = sizeof(Elf_Shdr);
1512 		ehdr->e_shnum = 0;
1513 		ehdr->e_shstrndx = SHN_UNDEF;
1514 
1515 		/*
1516 		 * Fill in the program header entries.
1517 		 */
1518 		phdr = (Elf_Phdr *)((char *)dst + phoff);
1519 
1520 		/* The note segement. */
1521 		phdr->p_type = PT_NOTE;
1522 		phdr->p_offset = noteoff;
1523 		phdr->p_vaddr = 0;
1524 		phdr->p_paddr = 0;
1525 		phdr->p_filesz = notesz;
1526 		phdr->p_memsz = 0;
1527 		phdr->p_flags = 0;
1528 		phdr->p_align = 0;
1529 		phdr++;
1530 
1531 		/* All the writable segments from the program. */
1532 		phc.phdr = phdr;
1533 		phc.offset = *off;
1534 		each_writable_segment(td, cb_put_phdr, &phc);
1535 	}
1536 }
1537 
1538 static void
1539 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1540     const void *desc, size_t descsz)
1541 {
1542 	Elf_Note note;
1543 
1544 	note.n_namesz = strlen(name) + 1;
1545 	note.n_descsz = descsz;
1546 	note.n_type = type;
1547 	if (dst != NULL)
1548 		bcopy(&note, (char *)dst + *off, sizeof note);
1549 	*off += sizeof note;
1550 	if (dst != NULL)
1551 		bcopy(name, (char *)dst + *off, note.n_namesz);
1552 	*off += roundup2(note.n_namesz, sizeof(Elf_Size));
1553 	if (dst != NULL)
1554 		bcopy(desc, (char *)dst + *off, note.n_descsz);
1555 	*off += roundup2(note.n_descsz, sizeof(Elf_Size));
1556 }
1557 
1558 static boolean_t
1559 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
1560     int32_t *osrel, const Elf_Phdr *pnote)
1561 {
1562 	const Elf_Note *note, *note0, *note_end;
1563 	const char *note_name;
1564 	int i;
1565 
1566 	if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
1567 	    pnote->p_filesz > PAGE_SIZE ||
1568 	    pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
1569 		return (FALSE);
1570 
1571 	note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1572 	note_end = (const Elf_Note *)(imgp->image_header +
1573 	    pnote->p_offset + pnote->p_filesz);
1574 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
1575 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
1576 		    (const char *)note < sizeof(Elf_Note))
1577 			return (FALSE);
1578 		if (note->n_namesz != checknote->hdr.n_namesz ||
1579 		    note->n_descsz != checknote->hdr.n_descsz ||
1580 		    note->n_type != checknote->hdr.n_type)
1581 			goto nextnote;
1582 		note_name = (const char *)(note + 1);
1583 		if (note_name + checknote->hdr.n_namesz >=
1584 		    (const char *)note_end || strncmp(checknote->vendor,
1585 		    note_name, checknote->hdr.n_namesz) != 0)
1586 			goto nextnote;
1587 
1588 		/*
1589 		 * Fetch the osreldate for binary
1590 		 * from the ELF OSABI-note if necessary.
1591 		 */
1592 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
1593 		    checknote->trans_osrel != NULL)
1594 			return (checknote->trans_osrel(note, osrel));
1595 		return (TRUE);
1596 
1597 nextnote:
1598 		note = (const Elf_Note *)((const char *)(note + 1) +
1599 		    roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
1600 		    roundup2(note->n_descsz, sizeof(Elf32_Addr)));
1601 	}
1602 
1603 	return (FALSE);
1604 }
1605 
1606 /*
1607  * Try to find the appropriate ABI-note section for checknote,
1608  * fetch the osreldate for binary from the ELF OSABI-note. Only the
1609  * first page of the image is searched, the same as for headers.
1610  */
1611 static boolean_t
1612 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
1613     int32_t *osrel)
1614 {
1615 	const Elf_Phdr *phdr;
1616 	const Elf_Ehdr *hdr;
1617 	int i;
1618 
1619 	hdr = (const Elf_Ehdr *)imgp->image_header;
1620 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1621 
1622 	for (i = 0; i < hdr->e_phnum; i++) {
1623 		if (phdr[i].p_type == PT_NOTE &&
1624 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
1625 			return (TRUE);
1626 	}
1627 	return (FALSE);
1628 
1629 }
1630 
1631 /*
1632  * Tell kern_execve.c about it, with a little help from the linker.
1633  */
1634 static struct execsw __elfN(execsw) = {
1635 	__CONCAT(exec_, __elfN(imgact)),
1636 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1637 };
1638 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
1639 
1640 #ifdef COMPRESS_USER_CORES
1641 /*
1642  * Compress and write out a core segment for a user process.
1643  *
1644  * 'inbuf' is the starting address of a VM segment in the process' address
1645  * space that is to be compressed and written out to the core file.  'dest_buf'
1646  * is a buffer in the kernel's address space.  The segment is copied from
1647  * 'inbuf' to 'dest_buf' first before being processed by the compression
1648  * routine gzwrite().  This copying is necessary because the content of the VM
1649  * segment may change between the compression pass and the crc-computation pass
1650  * in gzwrite().  This is because realtime threads may preempt the UNIX kernel.
1651  */
1652 static int
1653 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
1654     struct thread *td)
1655 {
1656 	int len_compressed;
1657 	int error = 0;
1658 	unsigned int chunk_len;
1659 
1660 	while (len) {
1661 		chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
1662 		copyin(inbuf, dest_buf, chunk_len);
1663 		len_compressed = gzwrite(file, dest_buf, chunk_len);
1664 
1665 		EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
1666 
1667 		if ((unsigned int)len_compressed != chunk_len) {
1668 			log(LOG_WARNING,
1669 			    "compress_core: length mismatch (0x%x returned, "
1670 			    "0x%x expected)\n", len_compressed, chunk_len);
1671 			EVENTHANDLER_INVOKE(app_coredump_error, td,
1672 			    "compress_core: length mismatch %x -> %x",
1673 			    chunk_len, len_compressed);
1674 			error = EFAULT;
1675 			break;
1676 		}
1677 		inbuf += chunk_len;
1678 		len -= chunk_len;
1679 		maybe_yield();
1680 	}
1681 
1682 	return (error);
1683 }
1684 #endif /* COMPRESS_USER_CORES */
1685 
1686 static vm_prot_t
1687 __elfN(trans_prot)(Elf_Word flags)
1688 {
1689 	vm_prot_t prot;
1690 
1691 	prot = 0;
1692 	if (flags & PF_X)
1693 		prot |= VM_PROT_EXECUTE;
1694 	if (flags & PF_W)
1695 		prot |= VM_PROT_WRITE;
1696 	if (flags & PF_R)
1697 		prot |= VM_PROT_READ;
1698 #if __ELF_WORD_SIZE == 32
1699 #if defined(__amd64__) || defined(__ia64__)
1700 	if (i386_read_exec && (flags & PF_R))
1701 		prot |= VM_PROT_EXECUTE;
1702 #endif
1703 #endif
1704 	return (prot);
1705 }
1706 
1707 static Elf_Word
1708 __elfN(untrans_prot)(vm_prot_t prot)
1709 {
1710 	Elf_Word flags;
1711 
1712 	flags = 0;
1713 	if (prot & VM_PROT_EXECUTE)
1714 		flags |= PF_X;
1715 	if (prot & VM_PROT_READ)
1716 		flags |= PF_R;
1717 	if (prot & VM_PROT_WRITE)
1718 		flags |= PF_W;
1719 	return (flags);
1720 }
1721