xref: /freebsd/sys/kern/imgact_elf.c (revision aaa3852435aee8c1b5abfe38ed7d43f9ba57e6dd)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2017 Dell EMC
5  * Copyright (c) 2000-2001, 2003 David O'Brien
6  * Copyright (c) 1995-1996 Søren Schmidt
7  * Copyright (c) 1996 Peter Wemm
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_capsicum.h"
38 
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mman.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
54 #include <sys/proc.h>
55 #include <sys/procfs.h>
56 #include <sys/ptrace.h>
57 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sbuf.h>
61 #include <sys/sf_buf.h>
62 #include <sys/smp.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
65 #include <sys/stat.h>
66 #include <sys/sx.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
73 #include <sys/user.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
85 
86 #define ELF_NOTE_ROUNDSIZE	4
87 #define OLD_EI_BRAND	8
88 
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91     const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93     u_long *entry);
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98     int32_t *osrel);
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101     Elf_Brandnote *checknote, int32_t *osrel, uint32_t *fctl0);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
106     "");
107 
108 #define	CORE_BUF_SIZE	(16 * 1024)
109 
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117     &elf_legacy_coredump, 0,
118     "include all and only RW pages in core dumps");
119 
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
123     defined(__riscv)
124 	1;
125 #else
126 	0;
127 #endif
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 
132 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
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 
138 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
139     "");
140 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
141 
142 static int __elfN(aslr_enabled) = 0;
143 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
144     &__elfN(aslr_enabled), 0,
145     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
146     ": enable address map randomization");
147 
148 static int __elfN(pie_aslr_enabled) = 0;
149 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
150     &__elfN(pie_aslr_enabled), 0,
151     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
152     ": enable address map randomization for PIE binaries");
153 
154 static int __elfN(aslr_honor_sbrk) = 1;
155 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
156     &__elfN(aslr_honor_sbrk), 0,
157     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
158 
159 static int __elfN(aslr_stack_gap) = 3;
160 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
161     &__elfN(aslr_stack_gap), 0,
162     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
163     ": maximum percentage of main stack to waste on a random gap");
164 
165 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
166 
167 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
168 
169 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
170 
171 Elf_Brandnote __elfN(freebsd_brandnote) = {
172 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
173 	.hdr.n_descsz	= sizeof(int32_t),
174 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
175 	.vendor		= FREEBSD_ABI_VENDOR,
176 	.flags		= BN_TRANSLATE_OSREL,
177 	.trans_osrel	= __elfN(freebsd_trans_osrel)
178 };
179 
180 static bool
181 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
182 {
183 	uintptr_t p;
184 
185 	p = (uintptr_t)(note + 1);
186 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
187 	*osrel = *(const int32_t *)(p);
188 
189 	return (true);
190 }
191 
192 static const char GNU_ABI_VENDOR[] = "GNU";
193 static int GNU_KFREEBSD_ABI_DESC = 3;
194 
195 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
196 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
197 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
198 	.hdr.n_type	= 1,
199 	.vendor		= GNU_ABI_VENDOR,
200 	.flags		= BN_TRANSLATE_OSREL,
201 	.trans_osrel	= kfreebsd_trans_osrel
202 };
203 
204 static bool
205 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
206 {
207 	const Elf32_Word *desc;
208 	uintptr_t p;
209 
210 	p = (uintptr_t)(note + 1);
211 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
212 
213 	desc = (const Elf32_Word *)p;
214 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
215 		return (false);
216 
217 	/*
218 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
219 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
220 	 */
221 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
222 
223 	return (true);
224 }
225 
226 int
227 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
228 {
229 	int i;
230 
231 	for (i = 0; i < MAX_BRANDS; i++) {
232 		if (elf_brand_list[i] == NULL) {
233 			elf_brand_list[i] = entry;
234 			break;
235 		}
236 	}
237 	if (i == MAX_BRANDS) {
238 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
239 			__func__, entry);
240 		return (-1);
241 	}
242 	return (0);
243 }
244 
245 int
246 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
247 {
248 	int i;
249 
250 	for (i = 0; i < MAX_BRANDS; i++) {
251 		if (elf_brand_list[i] == entry) {
252 			elf_brand_list[i] = NULL;
253 			break;
254 		}
255 	}
256 	if (i == MAX_BRANDS)
257 		return (-1);
258 	return (0);
259 }
260 
261 int
262 __elfN(brand_inuse)(Elf_Brandinfo *entry)
263 {
264 	struct proc *p;
265 	int rval = FALSE;
266 
267 	sx_slock(&allproc_lock);
268 	FOREACH_PROC_IN_SYSTEM(p) {
269 		if (p->p_sysent == entry->sysvec) {
270 			rval = TRUE;
271 			break;
272 		}
273 	}
274 	sx_sunlock(&allproc_lock);
275 
276 	return (rval);
277 }
278 
279 static Elf_Brandinfo *
280 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
281     int32_t *osrel, uint32_t *fctl0)
282 {
283 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
284 	Elf_Brandinfo *bi, *bi_m;
285 	boolean_t ret;
286 	int i, interp_name_len;
287 
288 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
289 
290 	/*
291 	 * We support four types of branding -- (1) the ELF EI_OSABI field
292 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
293 	 * branding w/in the ELF header, (3) path of the `interp_path'
294 	 * field, and (4) the ".note.ABI-tag" ELF section.
295 	 */
296 
297 	/* Look for an ".note.ABI-tag" ELF section */
298 	bi_m = NULL;
299 	for (i = 0; i < MAX_BRANDS; i++) {
300 		bi = elf_brand_list[i];
301 		if (bi == NULL)
302 			continue;
303 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
304 			continue;
305 		if (hdr->e_machine == bi->machine && (bi->flags &
306 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
307 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
308 			    fctl0);
309 			/* Give brand a chance to veto check_note's guess */
310 			if (ret && bi->header_supported)
311 				ret = bi->header_supported(imgp);
312 			/*
313 			 * If note checker claimed the binary, but the
314 			 * interpreter path in the image does not
315 			 * match default one for the brand, try to
316 			 * search for other brands with the same
317 			 * interpreter.  Either there is better brand
318 			 * with the right interpreter, or, failing
319 			 * this, we return first brand which accepted
320 			 * our note and, optionally, header.
321 			 */
322 			if (ret && bi_m == NULL && interp != NULL &&
323 			    (bi->interp_path == NULL ||
324 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
325 			    strncmp(interp, bi->interp_path, interp_name_len)
326 			    != 0))) {
327 				bi_m = bi;
328 				ret = 0;
329 			}
330 			if (ret)
331 				return (bi);
332 		}
333 	}
334 	if (bi_m != NULL)
335 		return (bi_m);
336 
337 	/* If the executable has a brand, search for it in the brand list. */
338 	for (i = 0; i < MAX_BRANDS; i++) {
339 		bi = elf_brand_list[i];
340 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
341 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
342 			continue;
343 		if (hdr->e_machine == bi->machine &&
344 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
345 		    (bi->compat_3_brand != NULL &&
346 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
347 		    bi->compat_3_brand) == 0))) {
348 			/* Looks good, but give brand a chance to veto */
349 			if (bi->header_supported == NULL ||
350 			    bi->header_supported(imgp)) {
351 				/*
352 				 * Again, prefer strictly matching
353 				 * interpreter path.
354 				 */
355 				if (interp_name_len == 0 &&
356 				    bi->interp_path == NULL)
357 					return (bi);
358 				if (bi->interp_path != NULL &&
359 				    strlen(bi->interp_path) + 1 ==
360 				    interp_name_len && strncmp(interp,
361 				    bi->interp_path, interp_name_len) == 0)
362 					return (bi);
363 				if (bi_m == NULL)
364 					bi_m = bi;
365 			}
366 		}
367 	}
368 	if (bi_m != NULL)
369 		return (bi_m);
370 
371 	/* No known brand, see if the header is recognized by any brand */
372 	for (i = 0; i < MAX_BRANDS; i++) {
373 		bi = elf_brand_list[i];
374 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
375 		    bi->header_supported == NULL)
376 			continue;
377 		if (hdr->e_machine == bi->machine) {
378 			ret = bi->header_supported(imgp);
379 			if (ret)
380 				return (bi);
381 		}
382 	}
383 
384 	/* Lacking a known brand, search for a recognized interpreter. */
385 	if (interp != NULL) {
386 		for (i = 0; i < MAX_BRANDS; i++) {
387 			bi = elf_brand_list[i];
388 			if (bi == NULL || (bi->flags &
389 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
390 			    != 0)
391 				continue;
392 			if (hdr->e_machine == bi->machine &&
393 			    bi->interp_path != NULL &&
394 			    /* ELF image p_filesz includes terminating zero */
395 			    strlen(bi->interp_path) + 1 == interp_name_len &&
396 			    strncmp(interp, bi->interp_path, interp_name_len)
397 			    == 0 && (bi->header_supported == NULL ||
398 			    bi->header_supported(imgp)))
399 				return (bi);
400 		}
401 	}
402 
403 	/* Lacking a recognized interpreter, try the default brand */
404 	for (i = 0; i < MAX_BRANDS; i++) {
405 		bi = elf_brand_list[i];
406 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
407 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
408 			continue;
409 		if (hdr->e_machine == bi->machine &&
410 		    __elfN(fallback_brand) == bi->brand &&
411 		    (bi->header_supported == NULL ||
412 		    bi->header_supported(imgp)))
413 			return (bi);
414 	}
415 	return (NULL);
416 }
417 
418 static int
419 __elfN(check_header)(const Elf_Ehdr *hdr)
420 {
421 	Elf_Brandinfo *bi;
422 	int i;
423 
424 	if (!IS_ELF(*hdr) ||
425 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
426 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
427 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
428 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
429 	    hdr->e_version != ELF_TARG_VER)
430 		return (ENOEXEC);
431 
432 	/*
433 	 * Make sure we have at least one brand for this machine.
434 	 */
435 
436 	for (i = 0; i < MAX_BRANDS; i++) {
437 		bi = elf_brand_list[i];
438 		if (bi != NULL && bi->machine == hdr->e_machine)
439 			break;
440 	}
441 	if (i == MAX_BRANDS)
442 		return (ENOEXEC);
443 
444 	return (0);
445 }
446 
447 static int
448 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
449     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
450 {
451 	struct sf_buf *sf;
452 	int error;
453 	vm_offset_t off;
454 
455 	/*
456 	 * Create the page if it doesn't exist yet. Ignore errors.
457 	 */
458 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
459 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
460 
461 	/*
462 	 * Find the page from the underlying object.
463 	 */
464 	if (object != NULL) {
465 		sf = vm_imgact_map_page(object, offset);
466 		if (sf == NULL)
467 			return (KERN_FAILURE);
468 		off = offset - trunc_page(offset);
469 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
470 		    end - start);
471 		vm_imgact_unmap_page(sf);
472 		if (error != 0)
473 			return (KERN_FAILURE);
474 	}
475 
476 	return (KERN_SUCCESS);
477 }
478 
479 static int
480 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
481     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
482     int cow)
483 {
484 	struct sf_buf *sf;
485 	vm_offset_t off;
486 	vm_size_t sz;
487 	int error, locked, rv;
488 
489 	if (start != trunc_page(start)) {
490 		rv = __elfN(map_partial)(map, object, offset, start,
491 		    round_page(start), prot);
492 		if (rv != KERN_SUCCESS)
493 			return (rv);
494 		offset += round_page(start) - start;
495 		start = round_page(start);
496 	}
497 	if (end != round_page(end)) {
498 		rv = __elfN(map_partial)(map, object, offset +
499 		    trunc_page(end) - start, trunc_page(end), end, prot);
500 		if (rv != KERN_SUCCESS)
501 			return (rv);
502 		end = trunc_page(end);
503 	}
504 	if (start >= end)
505 		return (KERN_SUCCESS);
506 	if ((offset & PAGE_MASK) != 0) {
507 		/*
508 		 * The mapping is not page aligned.  This means that we have
509 		 * to copy the data.
510 		 */
511 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
512 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
513 		if (rv != KERN_SUCCESS)
514 			return (rv);
515 		if (object == NULL)
516 			return (KERN_SUCCESS);
517 		for (; start < end; start += sz) {
518 			sf = vm_imgact_map_page(object, offset);
519 			if (sf == NULL)
520 				return (KERN_FAILURE);
521 			off = offset - trunc_page(offset);
522 			sz = end - start;
523 			if (sz > PAGE_SIZE - off)
524 				sz = PAGE_SIZE - off;
525 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
526 			    (caddr_t)start, sz);
527 			vm_imgact_unmap_page(sf);
528 			if (error != 0)
529 				return (KERN_FAILURE);
530 			offset += sz;
531 		}
532 	} else {
533 		vm_object_reference(object);
534 		rv = vm_map_fixed(map, object, offset, start, end - start,
535 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
536 		    (object != NULL ? MAP_VN_EXEC : 0));
537 		if (rv != KERN_SUCCESS) {
538 			locked = VOP_ISLOCKED(imgp->vp);
539 			VOP_UNLOCK(imgp->vp, 0);
540 			vm_object_deallocate(object);
541 			vn_lock(imgp->vp, locked | LK_RETRY);
542 			return (rv);
543 		} else if (object != NULL) {
544 			MPASS(imgp->vp->v_object == object);
545 			VOP_SET_TEXT_CHECKED(imgp->vp);
546 		}
547 	}
548 	return (KERN_SUCCESS);
549 }
550 
551 static int
552 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
553     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
554 {
555 	struct sf_buf *sf;
556 	size_t map_len;
557 	vm_map_t map;
558 	vm_object_t object;
559 	vm_offset_t map_addr;
560 	int error, rv, cow;
561 	size_t copy_len;
562 	vm_ooffset_t file_addr;
563 
564 	/*
565 	 * It's necessary to fail if the filsz + offset taken from the
566 	 * header is greater than the actual file pager object's size.
567 	 * If we were to allow this, then the vm_map_find() below would
568 	 * walk right off the end of the file object and into the ether.
569 	 *
570 	 * While I'm here, might as well check for something else that
571 	 * is invalid: filsz cannot be greater than memsz.
572 	 */
573 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
574 	    filsz > memsz) {
575 		uprintf("elf_load_section: truncated ELF file\n");
576 		return (ENOEXEC);
577 	}
578 
579 	object = imgp->object;
580 	map = &imgp->proc->p_vmspace->vm_map;
581 	map_addr = trunc_page((vm_offset_t)vmaddr);
582 	file_addr = trunc_page(offset);
583 
584 	/*
585 	 * We have two choices.  We can either clear the data in the last page
586 	 * of an oversized mapping, or we can start the anon mapping a page
587 	 * early and copy the initialized data into that first page.  We
588 	 * choose the second.
589 	 */
590 	if (filsz == 0)
591 		map_len = 0;
592 	else if (memsz > filsz)
593 		map_len = trunc_page(offset + filsz) - file_addr;
594 	else
595 		map_len = round_page(offset + filsz) - file_addr;
596 
597 	if (map_len != 0) {
598 		/* cow flags: don't dump readonly sections in core */
599 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
600 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
601 
602 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
603 		    map_addr, map_addr + map_len, prot, cow);
604 		if (rv != KERN_SUCCESS)
605 			return (EINVAL);
606 
607 		/* we can stop now if we've covered it all */
608 		if (memsz == filsz)
609 			return (0);
610 	}
611 
612 
613 	/*
614 	 * We have to get the remaining bit of the file into the first part
615 	 * of the oversized map segment.  This is normally because the .data
616 	 * segment in the file is extended to provide bss.  It's a neat idea
617 	 * to try and save a page, but it's a pain in the behind to implement.
618 	 */
619 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
620 	    filsz);
621 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
622 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
623 
624 	/* This had damn well better be true! */
625 	if (map_len != 0) {
626 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
627 		    map_addr + map_len, prot, 0);
628 		if (rv != KERN_SUCCESS)
629 			return (EINVAL);
630 	}
631 
632 	if (copy_len != 0) {
633 		sf = vm_imgact_map_page(object, offset + filsz);
634 		if (sf == NULL)
635 			return (EIO);
636 
637 		/* send the page fragment to user space */
638 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
639 		    copy_len);
640 		vm_imgact_unmap_page(sf);
641 		if (error != 0)
642 			return (error);
643 	}
644 
645 	/*
646 	 * Remove write access to the page if it was only granted by map_insert
647 	 * to allow copyout.
648 	 */
649 	if ((prot & VM_PROT_WRITE) == 0)
650 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
651 		    map_len), prot, FALSE);
652 
653 	return (0);
654 }
655 
656 static int
657 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
658     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
659 {
660 	vm_prot_t prot;
661 	u_long base_addr;
662 	bool first;
663 	int error, i;
664 
665 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
666 
667 	base_addr = 0;
668 	first = true;
669 
670 	for (i = 0; i < hdr->e_phnum; i++) {
671 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
672 			continue;
673 
674 		/* Loadable segment */
675 		prot = __elfN(trans_prot)(phdr[i].p_flags);
676 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
677 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
678 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
679 		if (error != 0)
680 			return (error);
681 
682 		/*
683 		 * Establish the base address if this is the first segment.
684 		 */
685 		if (first) {
686   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
687 			first = false;
688 		}
689 	}
690 
691 	if (base_addrp != NULL)
692 		*base_addrp = base_addr;
693 
694 	return (0);
695 }
696 
697 /*
698  * Load the file "file" into memory.  It may be either a shared object
699  * or an executable.
700  *
701  * The "addr" reference parameter is in/out.  On entry, it specifies
702  * the address where a shared object should be loaded.  If the file is
703  * an executable, this value is ignored.  On exit, "addr" specifies
704  * where the file was actually loaded.
705  *
706  * The "entry" reference parameter is out only.  On exit, it specifies
707  * the entry point for the loaded file.
708  */
709 static int
710 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
711 	u_long *entry)
712 {
713 	struct {
714 		struct nameidata nd;
715 		struct vattr attr;
716 		struct image_params image_params;
717 	} *tempdata;
718 	const Elf_Ehdr *hdr = NULL;
719 	const Elf_Phdr *phdr = NULL;
720 	struct nameidata *nd;
721 	struct vattr *attr;
722 	struct image_params *imgp;
723 	u_long rbase;
724 	u_long base_addr = 0;
725 	int error;
726 
727 #ifdef CAPABILITY_MODE
728 	/*
729 	 * XXXJA: This check can go away once we are sufficiently confident
730 	 * that the checks in namei() are correct.
731 	 */
732 	if (IN_CAPABILITY_MODE(curthread))
733 		return (ECAPMODE);
734 #endif
735 
736 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
737 	nd = &tempdata->nd;
738 	attr = &tempdata->attr;
739 	imgp = &tempdata->image_params;
740 
741 	/*
742 	 * Initialize part of the common data
743 	 */
744 	imgp->proc = p;
745 	imgp->attr = attr;
746 
747 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
748 	    UIO_SYSSPACE, file, curthread);
749 	if ((error = namei(nd)) != 0) {
750 		nd->ni_vp = NULL;
751 		goto fail;
752 	}
753 	NDFREE(nd, NDF_ONLY_PNBUF);
754 	imgp->vp = nd->ni_vp;
755 
756 	/*
757 	 * Check permissions, modes, uid, etc on the file, and "open" it.
758 	 */
759 	error = exec_check_permissions(imgp);
760 	if (error)
761 		goto fail;
762 
763 	error = exec_map_first_page(imgp);
764 	if (error)
765 		goto fail;
766 
767 	imgp->object = nd->ni_vp->v_object;
768 
769 	hdr = (const Elf_Ehdr *)imgp->image_header;
770 	if ((error = __elfN(check_header)(hdr)) != 0)
771 		goto fail;
772 	if (hdr->e_type == ET_DYN)
773 		rbase = *addr;
774 	else if (hdr->e_type == ET_EXEC)
775 		rbase = 0;
776 	else {
777 		error = ENOEXEC;
778 		goto fail;
779 	}
780 
781 	/* Only support headers that fit within first page for now      */
782 	if ((hdr->e_phoff > PAGE_SIZE) ||
783 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
784 		error = ENOEXEC;
785 		goto fail;
786 	}
787 
788 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
789 	if (!aligned(phdr, Elf_Addr)) {
790 		error = ENOEXEC;
791 		goto fail;
792 	}
793 
794 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
795 	if (error != 0)
796 		goto fail;
797 
798 	*addr = base_addr;
799 	*entry = (unsigned long)hdr->e_entry + rbase;
800 
801 fail:
802 	if (imgp->firstpage)
803 		exec_unmap_first_page(imgp);
804 
805 	if (nd->ni_vp) {
806 		if (imgp->textset)
807 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
808 		vput(nd->ni_vp);
809 	}
810 	free(tempdata, M_TEMP);
811 
812 	return (error);
813 }
814 
815 static u_long
816 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
817     u_int align)
818 {
819 	u_long rbase, res;
820 
821 	MPASS(vm_map_min(map) <= minv);
822 	MPASS(maxv <= vm_map_max(map));
823 	MPASS(minv < maxv);
824 	MPASS(minv + align < maxv);
825 	arc4rand(&rbase, sizeof(rbase), 0);
826 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
827 	res &= ~((u_long)align - 1);
828 	if (res >= maxv)
829 		res -= align;
830 	KASSERT(res >= minv,
831 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
832 	    res, minv, maxv, rbase));
833 	KASSERT(res < maxv,
834 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
835 	    res, maxv, minv, rbase));
836 	return (res);
837 }
838 
839 static int
840 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
841     const Elf_Phdr *phdr, u_long et_dyn_addr)
842 {
843 	struct vmspace *vmspace;
844 	const char *err_str;
845 	u_long text_size, data_size, total_size, text_addr, data_addr;
846 	u_long seg_size, seg_addr;
847 	int i;
848 
849 	err_str = NULL;
850 	text_size = data_size = total_size = text_addr = data_addr = 0;
851 
852 	for (i = 0; i < hdr->e_phnum; i++) {
853 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
854 			continue;
855 
856 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
857 		seg_size = round_page(phdr[i].p_memsz +
858 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
859 
860 		/*
861 		 * Make the largest executable segment the official
862 		 * text segment and all others data.
863 		 *
864 		 * Note that obreak() assumes that data_addr + data_size == end
865 		 * of data load area, and the ELF file format expects segments
866 		 * to be sorted by address.  If multiple data segments exist,
867 		 * the last one will be used.
868 		 */
869 
870 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
871 			text_size = seg_size;
872 			text_addr = seg_addr;
873 		} else {
874 			data_size = seg_size;
875 			data_addr = seg_addr;
876 		}
877 		total_size += seg_size;
878 	}
879 
880 	if (data_addr == 0 && data_size == 0) {
881 		data_addr = text_addr;
882 		data_size = text_size;
883 	}
884 
885 	/*
886 	 * Check limits.  It should be safe to check the
887 	 * limits after loading the segments since we do
888 	 * not actually fault in all the segments pages.
889 	 */
890 	PROC_LOCK(imgp->proc);
891 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
892 		err_str = "Data segment size exceeds process limit";
893 	else if (text_size > maxtsiz)
894 		err_str = "Text segment size exceeds system limit";
895 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
896 		err_str = "Total segment size exceeds process limit";
897 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
898 		err_str = "Data segment size exceeds resource limit";
899 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
900 		err_str = "Total segment size exceeds resource limit";
901 	PROC_UNLOCK(imgp->proc);
902 	if (err_str != NULL) {
903 		uprintf("%s\n", err_str);
904 		return (ENOMEM);
905 	}
906 
907 	vmspace = imgp->proc->p_vmspace;
908 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
909 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
910 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
911 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
912 
913 	return (0);
914 }
915 
916 static int
917 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
918     char **interpp, bool *free_interpp)
919 {
920 	struct thread *td;
921 	char *interp;
922 	int error, interp_name_len;
923 
924 	KASSERT(phdr->p_type == PT_INTERP,
925 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
926 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
927 
928 	td = curthread;
929 
930 	/* Path to interpreter */
931 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
932 		uprintf("Invalid PT_INTERP\n");
933 		return (ENOEXEC);
934 	}
935 
936 	interp_name_len = phdr->p_filesz;
937 	if (phdr->p_offset > PAGE_SIZE ||
938 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
939 		/*
940 		 * The vnode lock might be needed by the pagedaemon to
941 		 * clean pages owned by the vnode.  Do not allow sleep
942 		 * waiting for memory with the vnode locked, instead
943 		 * try non-sleepable allocation first, and if it
944 		 * fails, go to the slow path were we drop the lock
945 		 * and do M_WAITOK.  A text reference prevents
946 		 * modifications to the vnode content.
947 		 */
948 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
949 		if (interp == NULL) {
950 			VOP_UNLOCK(imgp->vp, 0);
951 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
952 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
953 		}
954 
955 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
956 		    interp_name_len, phdr->p_offset,
957 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
958 		    NOCRED, NULL, td);
959 		if (error != 0) {
960 			free(interp, M_TEMP);
961 			uprintf("i/o error PT_INTERP %d\n", error);
962 			return (error);
963 		}
964 		interp[interp_name_len] = '\0';
965 
966 		*interpp = interp;
967 		*free_interpp = true;
968 		return (0);
969 	}
970 
971 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
972 	if (interp[interp_name_len - 1] != '\0') {
973 		uprintf("Invalid PT_INTERP\n");
974 		return (ENOEXEC);
975 	}
976 
977 	*interpp = interp;
978 	*free_interpp = false;
979 	return (0);
980 }
981 
982 static int
983 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
984     const char *interp, u_long *addr, u_long *entry)
985 {
986 	char *path;
987 	int error;
988 
989 	if (brand_info->emul_path != NULL &&
990 	    brand_info->emul_path[0] != '\0') {
991 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
992 		snprintf(path, MAXPATHLEN, "%s%s",
993 		    brand_info->emul_path, interp);
994 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
995 		free(path, M_TEMP);
996 		if (error == 0)
997 			return (0);
998 	}
999 
1000 	if (brand_info->interp_newpath != NULL &&
1001 	    (brand_info->interp_path == NULL ||
1002 	    strcmp(interp, brand_info->interp_path) == 0)) {
1003 		error = __elfN(load_file)(imgp->proc,
1004 		    brand_info->interp_newpath, addr, entry);
1005 		if (error == 0)
1006 			return (0);
1007 	}
1008 
1009 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1010 	if (error == 0)
1011 		return (0);
1012 
1013 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1014 	return (error);
1015 }
1016 
1017 /*
1018  * Impossible et_dyn_addr initial value indicating that the real base
1019  * must be calculated later with some randomization applied.
1020  */
1021 #define	ET_DYN_ADDR_RAND	1
1022 
1023 static int
1024 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1025 {
1026 	struct thread *td;
1027 	const Elf_Ehdr *hdr;
1028 	const Elf_Phdr *phdr;
1029 	Elf_Auxargs *elf_auxargs;
1030 	struct vmspace *vmspace;
1031 	vm_map_t map;
1032 	char *interp;
1033 	Elf_Brandinfo *brand_info;
1034 	struct sysentvec *sv;
1035 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1036 	u_long maxalign, mapsz, maxv, maxv1;
1037 	uint32_t fctl0;
1038 	int32_t osrel;
1039 	bool free_interp;
1040 	int error, i, n;
1041 
1042 	hdr = (const Elf_Ehdr *)imgp->image_header;
1043 
1044 	/*
1045 	 * Do we have a valid ELF header ?
1046 	 *
1047 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1048 	 * if particular brand doesn't support it.
1049 	 */
1050 	if (__elfN(check_header)(hdr) != 0 ||
1051 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1052 		return (-1);
1053 
1054 	/*
1055 	 * From here on down, we return an errno, not -1, as we've
1056 	 * detected an ELF file.
1057 	 */
1058 
1059 	if ((hdr->e_phoff > PAGE_SIZE) ||
1060 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1061 		/* Only support headers in first page for now */
1062 		uprintf("Program headers not in the first page\n");
1063 		return (ENOEXEC);
1064 	}
1065 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1066 	if (!aligned(phdr, Elf_Addr)) {
1067 		uprintf("Unaligned program headers\n");
1068 		return (ENOEXEC);
1069 	}
1070 
1071 	n = error = 0;
1072 	baddr = 0;
1073 	osrel = 0;
1074 	fctl0 = 0;
1075 	entry = proghdr = 0;
1076 	interp = NULL;
1077 	free_interp = false;
1078 	td = curthread;
1079 	maxalign = PAGE_SIZE;
1080 	mapsz = 0;
1081 
1082 	for (i = 0; i < hdr->e_phnum; i++) {
1083 		switch (phdr[i].p_type) {
1084 		case PT_LOAD:
1085 			if (n == 0)
1086 				baddr = phdr[i].p_vaddr;
1087 			if (phdr[i].p_align > maxalign)
1088 				maxalign = phdr[i].p_align;
1089 			mapsz += phdr[i].p_memsz;
1090 			n++;
1091 
1092 			/*
1093 			 * If this segment contains the program headers,
1094 			 * remember their virtual address for the AT_PHDR
1095 			 * aux entry. Static binaries don't usually include
1096 			 * a PT_PHDR entry.
1097 			 */
1098 			if (phdr[i].p_offset == 0 &&
1099 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1100 				<= phdr[i].p_filesz)
1101 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1102 			break;
1103 		case PT_INTERP:
1104 			/* Path to interpreter */
1105 			if (interp != NULL) {
1106 				uprintf("Multiple PT_INTERP headers\n");
1107 				error = ENOEXEC;
1108 				goto ret;
1109 			}
1110 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1111 			    &free_interp);
1112 			if (error != 0)
1113 				goto ret;
1114 			break;
1115 		case PT_GNU_STACK:
1116 			if (__elfN(nxstack))
1117 				imgp->stack_prot =
1118 				    __elfN(trans_prot)(phdr[i].p_flags);
1119 			imgp->stack_sz = phdr[i].p_memsz;
1120 			break;
1121 		case PT_PHDR: 	/* Program header table info */
1122 			proghdr = phdr[i].p_vaddr;
1123 			break;
1124 		}
1125 	}
1126 
1127 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1128 	if (brand_info == NULL) {
1129 		uprintf("ELF binary type \"%u\" not known.\n",
1130 		    hdr->e_ident[EI_OSABI]);
1131 		error = ENOEXEC;
1132 		goto ret;
1133 	}
1134 	sv = brand_info->sysvec;
1135 	et_dyn_addr = 0;
1136 	if (hdr->e_type == ET_DYN) {
1137 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1138 			uprintf("Cannot execute shared object\n");
1139 			error = ENOEXEC;
1140 			goto ret;
1141 		}
1142 		/*
1143 		 * Honour the base load address from the dso if it is
1144 		 * non-zero for some reason.
1145 		 */
1146 		if (baddr == 0) {
1147 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1148 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1149 				et_dyn_addr = ET_DYN_LOAD_ADDR;
1150 			else if ((__elfN(pie_aslr_enabled) &&
1151 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1152 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1153 				et_dyn_addr = ET_DYN_ADDR_RAND;
1154 			else
1155 				et_dyn_addr = ET_DYN_LOAD_ADDR;
1156 		}
1157 	}
1158 
1159 	/*
1160 	 * Avoid a possible deadlock if the current address space is destroyed
1161 	 * and that address space maps the locked vnode.  In the common case,
1162 	 * the locked vnode's v_usecount is decremented but remains greater
1163 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1164 	 * However, in cases where the vnode lock is external, such as nullfs,
1165 	 * v_usecount may become zero.
1166 	 *
1167 	 * The VV_TEXT flag prevents modifications to the executable while
1168 	 * the vnode is unlocked.
1169 	 */
1170 	VOP_UNLOCK(imgp->vp, 0);
1171 
1172 	/*
1173 	 * Decide whether to enable randomization of user mappings.
1174 	 * First, reset user preferences for the setid binaries.
1175 	 * Then, account for the support of the randomization by the
1176 	 * ABI, by user preferences, and make special treatment for
1177 	 * PIE binaries.
1178 	 */
1179 	if (imgp->credential_setid) {
1180 		PROC_LOCK(imgp->proc);
1181 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1182 		PROC_UNLOCK(imgp->proc);
1183 	}
1184 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1185 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1186 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1187 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1188 		    ("et_dyn_addr == RAND and !ASLR"));
1189 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1190 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1191 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1192 		imgp->map_flags |= MAP_ASLR;
1193 		/*
1194 		 * If user does not care about sbrk, utilize the bss
1195 		 * grow region for mappings as well.  We can select
1196 		 * the base for the image anywere and still not suffer
1197 		 * from the fragmentation.
1198 		 */
1199 		if (!__elfN(aslr_honor_sbrk) ||
1200 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1201 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1202 	}
1203 
1204 	error = exec_new_vmspace(imgp, sv);
1205 	vmspace = imgp->proc->p_vmspace;
1206 	map = &vmspace->vm_map;
1207 
1208 	imgp->proc->p_sysent = sv;
1209 
1210 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1211 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1212 		KASSERT((map->flags & MAP_ASLR) != 0,
1213 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1214 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1215 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1216 		    /* reserve half of the address space to interpreter */
1217 		    maxv / 2, 1UL << flsl(maxalign));
1218 	}
1219 
1220 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1221 	if (error != 0)
1222 		goto ret;
1223 
1224 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1225 	if (error != 0)
1226 		goto ret;
1227 
1228 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1229 	if (error != 0)
1230 		goto ret;
1231 
1232 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1233 
1234 	/*
1235 	 * We load the dynamic linker where a userland call
1236 	 * to mmap(0, ...) would put it.  The rationale behind this
1237 	 * calculation is that it leaves room for the heap to grow to
1238 	 * its maximum allowed size.
1239 	 */
1240 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1241 	    RLIMIT_DATA));
1242 	if ((map->flags & MAP_ASLR) != 0) {
1243 		maxv1 = maxv / 2 + addr / 2;
1244 		MPASS(maxv1 >= addr);	/* No overflow */
1245 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1246 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1247 	} else {
1248 		map->anon_loc = addr;
1249 	}
1250 
1251 	imgp->entry_addr = entry;
1252 
1253 	if (interp != NULL) {
1254 		VOP_UNLOCK(imgp->vp, 0);
1255 		if ((map->flags & MAP_ASLR) != 0) {
1256 			/* Assume that interpeter fits into 1/4 of AS */
1257 			maxv1 = maxv / 2 + addr / 2;
1258 			MPASS(maxv1 >= addr);	/* No overflow */
1259 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1260 			    maxv1, PAGE_SIZE);
1261 		}
1262 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1263 		    &imgp->entry_addr);
1264 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1265 		if (error != 0)
1266 			goto ret;
1267 	} else
1268 		addr = et_dyn_addr;
1269 
1270 	/*
1271 	 * Construct auxargs table (used by the fixup routine)
1272 	 */
1273 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1274 	if (elf_auxargs == NULL) {
1275 		VOP_UNLOCK(imgp->vp, 0);
1276 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1277 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1278 	}
1279 	elf_auxargs->execfd = -1;
1280 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1281 	elf_auxargs->phent = hdr->e_phentsize;
1282 	elf_auxargs->phnum = hdr->e_phnum;
1283 	elf_auxargs->pagesz = PAGE_SIZE;
1284 	elf_auxargs->base = addr;
1285 	elf_auxargs->flags = 0;
1286 	elf_auxargs->entry = entry;
1287 	elf_auxargs->hdr_eflags = hdr->e_flags;
1288 
1289 	imgp->auxargs = elf_auxargs;
1290 	imgp->interpreted = 0;
1291 	imgp->reloc_base = addr;
1292 	imgp->proc->p_osrel = osrel;
1293 	imgp->proc->p_fctl0 = fctl0;
1294 	imgp->proc->p_elf_machine = hdr->e_machine;
1295 	imgp->proc->p_elf_flags = hdr->e_flags;
1296 
1297 ret:
1298 	if (free_interp)
1299 		free(interp, M_TEMP);
1300 	return (error);
1301 }
1302 
1303 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1304 
1305 int
1306 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1307 {
1308 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1309 	Elf_Auxinfo *argarray, *pos;
1310 	Elf_Addr *base, *auxbase;
1311 	int error;
1312 
1313 	base = (Elf_Addr *)*stack_base;
1314 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1315 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1316 	    M_WAITOK | M_ZERO);
1317 
1318 	if (args->execfd != -1)
1319 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1320 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1321 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1322 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1323 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1324 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1325 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1326 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1327 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1328 	if (imgp->execpathp != 0)
1329 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1330 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1331 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1332 	if (imgp->canary != 0) {
1333 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1334 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1335 	}
1336 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1337 	if (imgp->pagesizes != 0) {
1338 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1339 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1340 	}
1341 	if (imgp->sysent->sv_timekeep_base != 0) {
1342 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1343 		    imgp->sysent->sv_timekeep_base);
1344 	}
1345 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1346 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1347 	    imgp->sysent->sv_stackprot);
1348 	if (imgp->sysent->sv_hwcap != NULL)
1349 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1350 	if (imgp->sysent->sv_hwcap2 != NULL)
1351 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1352 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1353 
1354 	free(imgp->auxargs, M_TEMP);
1355 	imgp->auxargs = NULL;
1356 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1357 
1358 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1359 	free(argarray, M_TEMP);
1360 	if (error != 0)
1361 		return (error);
1362 
1363 	base--;
1364 	if (suword(base, imgp->args->argc) == -1)
1365 		return (EFAULT);
1366 	*stack_base = (register_t *)base;
1367 	return (0);
1368 }
1369 
1370 /*
1371  * Code for generating ELF core dumps.
1372  */
1373 
1374 typedef void (*segment_callback)(vm_map_entry_t, void *);
1375 
1376 /* Closure for cb_put_phdr(). */
1377 struct phdr_closure {
1378 	Elf_Phdr *phdr;		/* Program header to fill in */
1379 	Elf_Off offset;		/* Offset of segment in core file */
1380 };
1381 
1382 /* Closure for cb_size_segment(). */
1383 struct sseg_closure {
1384 	int count;		/* Count of writable segments. */
1385 	size_t size;		/* Total size of all writable segments. */
1386 };
1387 
1388 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1389 
1390 struct note_info {
1391 	int		type;		/* Note type. */
1392 	outfunc_t 	outfunc; 	/* Output function. */
1393 	void		*outarg;	/* Argument for the output function. */
1394 	size_t		outsize;	/* Output size. */
1395 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1396 };
1397 
1398 TAILQ_HEAD(note_info_list, note_info);
1399 
1400 /* Coredump output parameters. */
1401 struct coredump_params {
1402 	off_t		offset;
1403 	struct ucred	*active_cred;
1404 	struct ucred	*file_cred;
1405 	struct thread	*td;
1406 	struct vnode	*vp;
1407 	struct compressor *comp;
1408 };
1409 
1410 extern int compress_user_cores;
1411 extern int compress_user_cores_level;
1412 
1413 static void cb_put_phdr(vm_map_entry_t, void *);
1414 static void cb_size_segment(vm_map_entry_t, void *);
1415 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1416     enum uio_seg);
1417 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1418 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1419     struct note_info_list *, size_t);
1420 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1421     size_t *);
1422 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1423 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1424 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1425 static int sbuf_drain_core_output(void *, const char *, int);
1426 
1427 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1428 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1429 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1430 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1431 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1432 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1433 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1434 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1435 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1436 static void note_procstat_files(void *, struct sbuf *, size_t *);
1437 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1438 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1439 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1440 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1441 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1442 
1443 /*
1444  * Write out a core segment to the compression stream.
1445  */
1446 static int
1447 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1448 {
1449 	u_int chunk_len;
1450 	int error;
1451 
1452 	while (len > 0) {
1453 		chunk_len = MIN(len, CORE_BUF_SIZE);
1454 
1455 		/*
1456 		 * We can get EFAULT error here.
1457 		 * In that case zero out the current chunk of the segment.
1458 		 */
1459 		error = copyin(base, buf, chunk_len);
1460 		if (error != 0)
1461 			bzero(buf, chunk_len);
1462 		error = compressor_write(p->comp, buf, chunk_len);
1463 		if (error != 0)
1464 			break;
1465 		base += chunk_len;
1466 		len -= chunk_len;
1467 	}
1468 	return (error);
1469 }
1470 
1471 static int
1472 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1473 {
1474 
1475 	return (core_write((struct coredump_params *)arg, base, len, offset,
1476 	    UIO_SYSSPACE));
1477 }
1478 
1479 static int
1480 core_write(struct coredump_params *p, const void *base, size_t len,
1481     off_t offset, enum uio_seg seg)
1482 {
1483 
1484 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1485 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1486 	    p->active_cred, p->file_cred, NULL, p->td));
1487 }
1488 
1489 static int
1490 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1491     void *tmpbuf)
1492 {
1493 	int error;
1494 
1495 	if (p->comp != NULL)
1496 		return (compress_chunk(p, base, tmpbuf, len));
1497 
1498 	/*
1499 	 * EFAULT is a non-fatal error that we can get, for example,
1500 	 * if the segment is backed by a file but extends beyond its
1501 	 * end.
1502 	 */
1503 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1504 	if (error == EFAULT) {
1505 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1506 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1507 		    "for process %s\n", base, len, offset, curproc->p_comm);
1508 
1509 		/*
1510 		 * Write a "real" zero byte at the end of the target region
1511 		 * in the case this is the last segment.
1512 		 * The intermediate space will be implicitly zero-filled.
1513 		 */
1514 		error = core_write(p, zero_region, 1, offset + len - 1,
1515 		    UIO_SYSSPACE);
1516 	}
1517 	return (error);
1518 }
1519 
1520 /*
1521  * Drain into a core file.
1522  */
1523 static int
1524 sbuf_drain_core_output(void *arg, const char *data, int len)
1525 {
1526 	struct coredump_params *p;
1527 	int error, locked;
1528 
1529 	p = (struct coredump_params *)arg;
1530 
1531 	/*
1532 	 * Some kern_proc out routines that print to this sbuf may
1533 	 * call us with the process lock held. Draining with the
1534 	 * non-sleepable lock held is unsafe. The lock is needed for
1535 	 * those routines when dumping a live process. In our case we
1536 	 * can safely release the lock before draining and acquire
1537 	 * again after.
1538 	 */
1539 	locked = PROC_LOCKED(p->td->td_proc);
1540 	if (locked)
1541 		PROC_UNLOCK(p->td->td_proc);
1542 	if (p->comp != NULL)
1543 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1544 	else
1545 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1546 		    UIO_SYSSPACE);
1547 	if (locked)
1548 		PROC_LOCK(p->td->td_proc);
1549 	if (error != 0)
1550 		return (-error);
1551 	p->offset += len;
1552 	return (len);
1553 }
1554 
1555 int
1556 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1557 {
1558 	struct ucred *cred = td->td_ucred;
1559 	int error = 0;
1560 	struct sseg_closure seginfo;
1561 	struct note_info_list notelst;
1562 	struct coredump_params params;
1563 	struct note_info *ninfo;
1564 	void *hdr, *tmpbuf;
1565 	size_t hdrsize, notesz, coresize;
1566 
1567 	hdr = NULL;
1568 	tmpbuf = NULL;
1569 	TAILQ_INIT(&notelst);
1570 
1571 	/* Size the program segments. */
1572 	seginfo.count = 0;
1573 	seginfo.size = 0;
1574 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1575 
1576 	/*
1577 	 * Collect info about the core file header area.
1578 	 */
1579 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1580 	if (seginfo.count + 1 >= PN_XNUM)
1581 		hdrsize += sizeof(Elf_Shdr);
1582 	__elfN(prepare_notes)(td, &notelst, &notesz);
1583 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1584 
1585 	/* Set up core dump parameters. */
1586 	params.offset = 0;
1587 	params.active_cred = cred;
1588 	params.file_cred = NOCRED;
1589 	params.td = td;
1590 	params.vp = vp;
1591 	params.comp = NULL;
1592 
1593 #ifdef RACCT
1594 	if (racct_enable) {
1595 		PROC_LOCK(td->td_proc);
1596 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1597 		PROC_UNLOCK(td->td_proc);
1598 		if (error != 0) {
1599 			error = EFAULT;
1600 			goto done;
1601 		}
1602 	}
1603 #endif
1604 	if (coresize >= limit) {
1605 		error = EFAULT;
1606 		goto done;
1607 	}
1608 
1609 	/* Create a compression stream if necessary. */
1610 	if (compress_user_cores != 0) {
1611 		params.comp = compressor_init(core_compressed_write,
1612 		    compress_user_cores, CORE_BUF_SIZE,
1613 		    compress_user_cores_level, &params);
1614 		if (params.comp == NULL) {
1615 			error = EFAULT;
1616 			goto done;
1617 		}
1618 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1619         }
1620 
1621 	/*
1622 	 * Allocate memory for building the header, fill it up,
1623 	 * and write it out following the notes.
1624 	 */
1625 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1626 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1627 	    notesz);
1628 
1629 	/* Write the contents of all of the writable segments. */
1630 	if (error == 0) {
1631 		Elf_Phdr *php;
1632 		off_t offset;
1633 		int i;
1634 
1635 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1636 		offset = round_page(hdrsize + notesz);
1637 		for (i = 0; i < seginfo.count; i++) {
1638 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1639 			    php->p_filesz, offset, &params, tmpbuf);
1640 			if (error != 0)
1641 				break;
1642 			offset += php->p_filesz;
1643 			php++;
1644 		}
1645 		if (error == 0 && params.comp != NULL)
1646 			error = compressor_flush(params.comp);
1647 	}
1648 	if (error) {
1649 		log(LOG_WARNING,
1650 		    "Failed to write core file for process %s (error %d)\n",
1651 		    curproc->p_comm, error);
1652 	}
1653 
1654 done:
1655 	free(tmpbuf, M_TEMP);
1656 	if (params.comp != NULL)
1657 		compressor_fini(params.comp);
1658 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1659 		TAILQ_REMOVE(&notelst, ninfo, link);
1660 		free(ninfo, M_TEMP);
1661 	}
1662 	if (hdr != NULL)
1663 		free(hdr, M_TEMP);
1664 
1665 	return (error);
1666 }
1667 
1668 /*
1669  * A callback for each_dumpable_segment() to write out the segment's
1670  * program header entry.
1671  */
1672 static void
1673 cb_put_phdr(vm_map_entry_t entry, void *closure)
1674 {
1675 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1676 	Elf_Phdr *phdr = phc->phdr;
1677 
1678 	phc->offset = round_page(phc->offset);
1679 
1680 	phdr->p_type = PT_LOAD;
1681 	phdr->p_offset = phc->offset;
1682 	phdr->p_vaddr = entry->start;
1683 	phdr->p_paddr = 0;
1684 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1685 	phdr->p_align = PAGE_SIZE;
1686 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1687 
1688 	phc->offset += phdr->p_filesz;
1689 	phc->phdr++;
1690 }
1691 
1692 /*
1693  * A callback for each_dumpable_segment() to gather information about
1694  * the number of segments and their total size.
1695  */
1696 static void
1697 cb_size_segment(vm_map_entry_t entry, void *closure)
1698 {
1699 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1700 
1701 	ssc->count++;
1702 	ssc->size += entry->end - entry->start;
1703 }
1704 
1705 /*
1706  * For each writable segment in the process's memory map, call the given
1707  * function with a pointer to the map entry and some arbitrary
1708  * caller-supplied data.
1709  */
1710 static void
1711 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1712 {
1713 	struct proc *p = td->td_proc;
1714 	vm_map_t map = &p->p_vmspace->vm_map;
1715 	vm_map_entry_t entry;
1716 	vm_object_t backing_object, object;
1717 	boolean_t ignore_entry;
1718 
1719 	vm_map_lock_read(map);
1720 	for (entry = map->header.next; entry != &map->header;
1721 	    entry = entry->next) {
1722 		/*
1723 		 * Don't dump inaccessible mappings, deal with legacy
1724 		 * coredump mode.
1725 		 *
1726 		 * Note that read-only segments related to the elf binary
1727 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1728 		 * need to arbitrarily ignore such segments.
1729 		 */
1730 		if (elf_legacy_coredump) {
1731 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1732 				continue;
1733 		} else {
1734 			if ((entry->protection & VM_PROT_ALL) == 0)
1735 				continue;
1736 		}
1737 
1738 		/*
1739 		 * Dont include memory segment in the coredump if
1740 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1741 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1742 		 * kernel map).
1743 		 */
1744 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1745 			continue;
1746 
1747 		if ((object = entry->object.vm_object) == NULL)
1748 			continue;
1749 
1750 		/* Ignore memory-mapped devices and such things. */
1751 		VM_OBJECT_RLOCK(object);
1752 		while ((backing_object = object->backing_object) != NULL) {
1753 			VM_OBJECT_RLOCK(backing_object);
1754 			VM_OBJECT_RUNLOCK(object);
1755 			object = backing_object;
1756 		}
1757 		ignore_entry = object->type != OBJT_DEFAULT &&
1758 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1759 		    object->type != OBJT_PHYS;
1760 		VM_OBJECT_RUNLOCK(object);
1761 		if (ignore_entry)
1762 			continue;
1763 
1764 		(*func)(entry, closure);
1765 	}
1766 	vm_map_unlock_read(map);
1767 }
1768 
1769 /*
1770  * Write the core file header to the file, including padding up to
1771  * the page boundary.
1772  */
1773 static int
1774 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1775     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1776 {
1777 	struct note_info *ninfo;
1778 	struct sbuf *sb;
1779 	int error;
1780 
1781 	/* Fill in the header. */
1782 	bzero(hdr, hdrsize);
1783 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1784 
1785 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1786 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1787 	sbuf_start_section(sb, NULL);
1788 	sbuf_bcat(sb, hdr, hdrsize);
1789 	TAILQ_FOREACH(ninfo, notelst, link)
1790 	    __elfN(putnote)(ninfo, sb);
1791 	/* Align up to a page boundary for the program segments. */
1792 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1793 	error = sbuf_finish(sb);
1794 	sbuf_delete(sb);
1795 
1796 	return (error);
1797 }
1798 
1799 static void
1800 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1801     size_t *sizep)
1802 {
1803 	struct proc *p;
1804 	struct thread *thr;
1805 	size_t size;
1806 
1807 	p = td->td_proc;
1808 	size = 0;
1809 
1810 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1811 
1812 	/*
1813 	 * To have the debugger select the right thread (LWP) as the initial
1814 	 * thread, we dump the state of the thread passed to us in td first.
1815 	 * This is the thread that causes the core dump and thus likely to
1816 	 * be the right thread one wants to have selected in the debugger.
1817 	 */
1818 	thr = td;
1819 	while (thr != NULL) {
1820 		size += register_note(list, NT_PRSTATUS,
1821 		    __elfN(note_prstatus), thr);
1822 		size += register_note(list, NT_FPREGSET,
1823 		    __elfN(note_fpregset), thr);
1824 		size += register_note(list, NT_THRMISC,
1825 		    __elfN(note_thrmisc), thr);
1826 		size += register_note(list, NT_PTLWPINFO,
1827 		    __elfN(note_ptlwpinfo), thr);
1828 		size += register_note(list, -1,
1829 		    __elfN(note_threadmd), thr);
1830 
1831 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1832 		    TAILQ_NEXT(thr, td_plist);
1833 		if (thr == td)
1834 			thr = TAILQ_NEXT(thr, td_plist);
1835 	}
1836 
1837 	size += register_note(list, NT_PROCSTAT_PROC,
1838 	    __elfN(note_procstat_proc), p);
1839 	size += register_note(list, NT_PROCSTAT_FILES,
1840 	    note_procstat_files, p);
1841 	size += register_note(list, NT_PROCSTAT_VMMAP,
1842 	    note_procstat_vmmap, p);
1843 	size += register_note(list, NT_PROCSTAT_GROUPS,
1844 	    note_procstat_groups, p);
1845 	size += register_note(list, NT_PROCSTAT_UMASK,
1846 	    note_procstat_umask, p);
1847 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1848 	    note_procstat_rlimit, p);
1849 	size += register_note(list, NT_PROCSTAT_OSREL,
1850 	    note_procstat_osrel, p);
1851 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1852 	    __elfN(note_procstat_psstrings), p);
1853 	size += register_note(list, NT_PROCSTAT_AUXV,
1854 	    __elfN(note_procstat_auxv), p);
1855 
1856 	*sizep = size;
1857 }
1858 
1859 static void
1860 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1861     size_t notesz)
1862 {
1863 	Elf_Ehdr *ehdr;
1864 	Elf_Phdr *phdr;
1865 	Elf_Shdr *shdr;
1866 	struct phdr_closure phc;
1867 
1868 	ehdr = (Elf_Ehdr *)hdr;
1869 
1870 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1871 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1872 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1873 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1874 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1875 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1876 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1877 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1878 	ehdr->e_ident[EI_ABIVERSION] = 0;
1879 	ehdr->e_ident[EI_PAD] = 0;
1880 	ehdr->e_type = ET_CORE;
1881 	ehdr->e_machine = td->td_proc->p_elf_machine;
1882 	ehdr->e_version = EV_CURRENT;
1883 	ehdr->e_entry = 0;
1884 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1885 	ehdr->e_flags = td->td_proc->p_elf_flags;
1886 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1887 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1888 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1889 	ehdr->e_shstrndx = SHN_UNDEF;
1890 	if (numsegs + 1 < PN_XNUM) {
1891 		ehdr->e_phnum = numsegs + 1;
1892 		ehdr->e_shnum = 0;
1893 	} else {
1894 		ehdr->e_phnum = PN_XNUM;
1895 		ehdr->e_shnum = 1;
1896 
1897 		ehdr->e_shoff = ehdr->e_phoff +
1898 		    (numsegs + 1) * ehdr->e_phentsize;
1899 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1900 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1901 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1902 
1903 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1904 		memset(shdr, 0, sizeof(*shdr));
1905 		/*
1906 		 * A special first section is used to hold large segment and
1907 		 * section counts.  This was proposed by Sun Microsystems in
1908 		 * Solaris and has been adopted by Linux; the standard ELF
1909 		 * tools are already familiar with the technique.
1910 		 *
1911 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1912 		 * (or 12-7 depending on the version of the document) for more
1913 		 * details.
1914 		 */
1915 		shdr->sh_type = SHT_NULL;
1916 		shdr->sh_size = ehdr->e_shnum;
1917 		shdr->sh_link = ehdr->e_shstrndx;
1918 		shdr->sh_info = numsegs + 1;
1919 	}
1920 
1921 	/*
1922 	 * Fill in the program header entries.
1923 	 */
1924 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1925 
1926 	/* The note segement. */
1927 	phdr->p_type = PT_NOTE;
1928 	phdr->p_offset = hdrsize;
1929 	phdr->p_vaddr = 0;
1930 	phdr->p_paddr = 0;
1931 	phdr->p_filesz = notesz;
1932 	phdr->p_memsz = 0;
1933 	phdr->p_flags = PF_R;
1934 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1935 	phdr++;
1936 
1937 	/* All the writable segments from the program. */
1938 	phc.phdr = phdr;
1939 	phc.offset = round_page(hdrsize + notesz);
1940 	each_dumpable_segment(td, cb_put_phdr, &phc);
1941 }
1942 
1943 static size_t
1944 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1945 {
1946 	struct note_info *ninfo;
1947 	size_t size, notesize;
1948 
1949 	size = 0;
1950 	out(arg, NULL, &size);
1951 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1952 	ninfo->type = type;
1953 	ninfo->outfunc = out;
1954 	ninfo->outarg = arg;
1955 	ninfo->outsize = size;
1956 	TAILQ_INSERT_TAIL(list, ninfo, link);
1957 
1958 	if (type == -1)
1959 		return (size);
1960 
1961 	notesize = sizeof(Elf_Note) +		/* note header */
1962 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1963 						/* note name */
1964 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1965 
1966 	return (notesize);
1967 }
1968 
1969 static size_t
1970 append_note_data(const void *src, void *dst, size_t len)
1971 {
1972 	size_t padded_len;
1973 
1974 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1975 	if (dst != NULL) {
1976 		bcopy(src, dst, len);
1977 		bzero((char *)dst + len, padded_len - len);
1978 	}
1979 	return (padded_len);
1980 }
1981 
1982 size_t
1983 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1984 {
1985 	Elf_Note *note;
1986 	char *buf;
1987 	size_t notesize;
1988 
1989 	buf = dst;
1990 	if (buf != NULL) {
1991 		note = (Elf_Note *)buf;
1992 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1993 		note->n_descsz = size;
1994 		note->n_type = type;
1995 		buf += sizeof(*note);
1996 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1997 		    sizeof(FREEBSD_ABI_VENDOR));
1998 		append_note_data(src, buf, size);
1999 		if (descp != NULL)
2000 			*descp = buf;
2001 	}
2002 
2003 	notesize = sizeof(Elf_Note) +		/* note header */
2004 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2005 						/* note name */
2006 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2007 
2008 	return (notesize);
2009 }
2010 
2011 static void
2012 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2013 {
2014 	Elf_Note note;
2015 	ssize_t old_len, sect_len;
2016 	size_t new_len, descsz, i;
2017 
2018 	if (ninfo->type == -1) {
2019 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2020 		return;
2021 	}
2022 
2023 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2024 	note.n_descsz = ninfo->outsize;
2025 	note.n_type = ninfo->type;
2026 
2027 	sbuf_bcat(sb, &note, sizeof(note));
2028 	sbuf_start_section(sb, &old_len);
2029 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2030 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2031 	if (note.n_descsz == 0)
2032 		return;
2033 	sbuf_start_section(sb, &old_len);
2034 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2035 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2036 	if (sect_len < 0)
2037 		return;
2038 
2039 	new_len = (size_t)sect_len;
2040 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2041 	if (new_len < descsz) {
2042 		/*
2043 		 * It is expected that individual note emitters will correctly
2044 		 * predict their expected output size and fill up to that size
2045 		 * themselves, padding in a format-specific way if needed.
2046 		 * However, in case they don't, just do it here with zeros.
2047 		 */
2048 		for (i = 0; i < descsz - new_len; i++)
2049 			sbuf_putc(sb, 0);
2050 	} else if (new_len > descsz) {
2051 		/*
2052 		 * We can't always truncate sb -- we may have drained some
2053 		 * of it already.
2054 		 */
2055 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2056 		    "read it (%zu > %zu).  Since it is longer than "
2057 		    "expected, this coredump's notes are corrupt.  THIS "
2058 		    "IS A BUG in the note_procstat routine for type %u.\n",
2059 		    __func__, (unsigned)note.n_type, new_len, descsz,
2060 		    (unsigned)note.n_type));
2061 	}
2062 }
2063 
2064 /*
2065  * Miscellaneous note out functions.
2066  */
2067 
2068 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2069 #include <compat/freebsd32/freebsd32.h>
2070 #include <compat/freebsd32/freebsd32_signal.h>
2071 
2072 typedef struct prstatus32 elf_prstatus_t;
2073 typedef struct prpsinfo32 elf_prpsinfo_t;
2074 typedef struct fpreg32 elf_prfpregset_t;
2075 typedef struct fpreg32 elf_fpregset_t;
2076 typedef struct reg32 elf_gregset_t;
2077 typedef struct thrmisc32 elf_thrmisc_t;
2078 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2079 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2080 typedef uint32_t elf_ps_strings_t;
2081 #else
2082 typedef prstatus_t elf_prstatus_t;
2083 typedef prpsinfo_t elf_prpsinfo_t;
2084 typedef prfpregset_t elf_prfpregset_t;
2085 typedef prfpregset_t elf_fpregset_t;
2086 typedef gregset_t elf_gregset_t;
2087 typedef thrmisc_t elf_thrmisc_t;
2088 #define ELF_KERN_PROC_MASK	0
2089 typedef struct kinfo_proc elf_kinfo_proc_t;
2090 typedef vm_offset_t elf_ps_strings_t;
2091 #endif
2092 
2093 static void
2094 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2095 {
2096 	struct sbuf sbarg;
2097 	size_t len;
2098 	char *cp, *end;
2099 	struct proc *p;
2100 	elf_prpsinfo_t *psinfo;
2101 	int error;
2102 
2103 	p = (struct proc *)arg;
2104 	if (sb != NULL) {
2105 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2106 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2107 		psinfo->pr_version = PRPSINFO_VERSION;
2108 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2109 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2110 		PROC_LOCK(p);
2111 		if (p->p_args != NULL) {
2112 			len = sizeof(psinfo->pr_psargs) - 1;
2113 			if (len > p->p_args->ar_length)
2114 				len = p->p_args->ar_length;
2115 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2116 			PROC_UNLOCK(p);
2117 			error = 0;
2118 		} else {
2119 			_PHOLD(p);
2120 			PROC_UNLOCK(p);
2121 			sbuf_new(&sbarg, psinfo->pr_psargs,
2122 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2123 			error = proc_getargv(curthread, p, &sbarg);
2124 			PRELE(p);
2125 			if (sbuf_finish(&sbarg) == 0)
2126 				len = sbuf_len(&sbarg) - 1;
2127 			else
2128 				len = sizeof(psinfo->pr_psargs) - 1;
2129 			sbuf_delete(&sbarg);
2130 		}
2131 		if (error || len == 0)
2132 			strlcpy(psinfo->pr_psargs, p->p_comm,
2133 			    sizeof(psinfo->pr_psargs));
2134 		else {
2135 			KASSERT(len < sizeof(psinfo->pr_psargs),
2136 			    ("len is too long: %zu vs %zu", len,
2137 			    sizeof(psinfo->pr_psargs)));
2138 			cp = psinfo->pr_psargs;
2139 			end = cp + len - 1;
2140 			for (;;) {
2141 				cp = memchr(cp, '\0', end - cp);
2142 				if (cp == NULL)
2143 					break;
2144 				*cp = ' ';
2145 			}
2146 		}
2147 		psinfo->pr_pid = p->p_pid;
2148 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2149 		free(psinfo, M_TEMP);
2150 	}
2151 	*sizep = sizeof(*psinfo);
2152 }
2153 
2154 static void
2155 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2156 {
2157 	struct thread *td;
2158 	elf_prstatus_t *status;
2159 
2160 	td = (struct thread *)arg;
2161 	if (sb != NULL) {
2162 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2163 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2164 		status->pr_version = PRSTATUS_VERSION;
2165 		status->pr_statussz = sizeof(elf_prstatus_t);
2166 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2167 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2168 		status->pr_osreldate = osreldate;
2169 		status->pr_cursig = td->td_proc->p_sig;
2170 		status->pr_pid = td->td_tid;
2171 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2172 		fill_regs32(td, &status->pr_reg);
2173 #else
2174 		fill_regs(td, &status->pr_reg);
2175 #endif
2176 		sbuf_bcat(sb, status, sizeof(*status));
2177 		free(status, M_TEMP);
2178 	}
2179 	*sizep = sizeof(*status);
2180 }
2181 
2182 static void
2183 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2184 {
2185 	struct thread *td;
2186 	elf_prfpregset_t *fpregset;
2187 
2188 	td = (struct thread *)arg;
2189 	if (sb != NULL) {
2190 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2191 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2192 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2193 		fill_fpregs32(td, fpregset);
2194 #else
2195 		fill_fpregs(td, fpregset);
2196 #endif
2197 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2198 		free(fpregset, M_TEMP);
2199 	}
2200 	*sizep = sizeof(*fpregset);
2201 }
2202 
2203 static void
2204 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2205 {
2206 	struct thread *td;
2207 	elf_thrmisc_t thrmisc;
2208 
2209 	td = (struct thread *)arg;
2210 	if (sb != NULL) {
2211 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2212 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2213 		strcpy(thrmisc.pr_tname, td->td_name);
2214 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2215 	}
2216 	*sizep = sizeof(thrmisc);
2217 }
2218 
2219 static void
2220 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2221 {
2222 	struct thread *td;
2223 	size_t size;
2224 	int structsize;
2225 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2226 	struct ptrace_lwpinfo32 pl;
2227 #else
2228 	struct ptrace_lwpinfo pl;
2229 #endif
2230 
2231 	td = (struct thread *)arg;
2232 	size = sizeof(structsize) + sizeof(pl);
2233 	if (sb != NULL) {
2234 		KASSERT(*sizep == size, ("invalid size"));
2235 		structsize = sizeof(pl);
2236 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2237 		bzero(&pl, sizeof(pl));
2238 		pl.pl_lwpid = td->td_tid;
2239 		pl.pl_event = PL_EVENT_NONE;
2240 		pl.pl_sigmask = td->td_sigmask;
2241 		pl.pl_siglist = td->td_siglist;
2242 		if (td->td_si.si_signo != 0) {
2243 			pl.pl_event = PL_EVENT_SIGNAL;
2244 			pl.pl_flags |= PL_FLAG_SI;
2245 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2246 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2247 #else
2248 			pl.pl_siginfo = td->td_si;
2249 #endif
2250 		}
2251 		strcpy(pl.pl_tdname, td->td_name);
2252 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2253 		sbuf_bcat(sb, &pl, sizeof(pl));
2254 	}
2255 	*sizep = size;
2256 }
2257 
2258 /*
2259  * Allow for MD specific notes, as well as any MD
2260  * specific preparations for writing MI notes.
2261  */
2262 static void
2263 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2264 {
2265 	struct thread *td;
2266 	void *buf;
2267 	size_t size;
2268 
2269 	td = (struct thread *)arg;
2270 	size = *sizep;
2271 	if (size != 0 && sb != NULL)
2272 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2273 	else
2274 		buf = NULL;
2275 	size = 0;
2276 	__elfN(dump_thread)(td, buf, &size);
2277 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2278 	if (size != 0 && sb != NULL)
2279 		sbuf_bcat(sb, buf, size);
2280 	free(buf, M_TEMP);
2281 	*sizep = size;
2282 }
2283 
2284 #ifdef KINFO_PROC_SIZE
2285 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2286 #endif
2287 
2288 static void
2289 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2290 {
2291 	struct proc *p;
2292 	size_t size;
2293 	int structsize;
2294 
2295 	p = (struct proc *)arg;
2296 	size = sizeof(structsize) + p->p_numthreads *
2297 	    sizeof(elf_kinfo_proc_t);
2298 
2299 	if (sb != NULL) {
2300 		KASSERT(*sizep == size, ("invalid size"));
2301 		structsize = sizeof(elf_kinfo_proc_t);
2302 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2303 		PROC_LOCK(p);
2304 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2305 	}
2306 	*sizep = size;
2307 }
2308 
2309 #ifdef KINFO_FILE_SIZE
2310 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2311 #endif
2312 
2313 static void
2314 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2315 {
2316 	struct proc *p;
2317 	size_t size, sect_sz, i;
2318 	ssize_t start_len, sect_len;
2319 	int structsize, filedesc_flags;
2320 
2321 	if (coredump_pack_fileinfo)
2322 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2323 	else
2324 		filedesc_flags = 0;
2325 
2326 	p = (struct proc *)arg;
2327 	structsize = sizeof(struct kinfo_file);
2328 	if (sb == NULL) {
2329 		size = 0;
2330 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2331 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2332 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2333 		PROC_LOCK(p);
2334 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2335 		sbuf_finish(sb);
2336 		sbuf_delete(sb);
2337 		*sizep = size;
2338 	} else {
2339 		sbuf_start_section(sb, &start_len);
2340 
2341 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2342 		PROC_LOCK(p);
2343 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2344 		    filedesc_flags);
2345 
2346 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2347 		if (sect_len < 0)
2348 			return;
2349 		sect_sz = sect_len;
2350 
2351 		KASSERT(sect_sz <= *sizep,
2352 		    ("kern_proc_filedesc_out did not respect maxlen; "
2353 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2354 		     sect_sz - sizeof(structsize)));
2355 
2356 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2357 			sbuf_putc(sb, 0);
2358 	}
2359 }
2360 
2361 #ifdef KINFO_VMENTRY_SIZE
2362 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2363 #endif
2364 
2365 static void
2366 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2367 {
2368 	struct proc *p;
2369 	size_t size;
2370 	int structsize, vmmap_flags;
2371 
2372 	if (coredump_pack_vmmapinfo)
2373 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2374 	else
2375 		vmmap_flags = 0;
2376 
2377 	p = (struct proc *)arg;
2378 	structsize = sizeof(struct kinfo_vmentry);
2379 	if (sb == NULL) {
2380 		size = 0;
2381 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2382 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2383 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2384 		PROC_LOCK(p);
2385 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2386 		sbuf_finish(sb);
2387 		sbuf_delete(sb);
2388 		*sizep = size;
2389 	} else {
2390 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2391 		PROC_LOCK(p);
2392 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2393 		    vmmap_flags);
2394 	}
2395 }
2396 
2397 static void
2398 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2399 {
2400 	struct proc *p;
2401 	size_t size;
2402 	int structsize;
2403 
2404 	p = (struct proc *)arg;
2405 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2406 	if (sb != NULL) {
2407 		KASSERT(*sizep == size, ("invalid size"));
2408 		structsize = sizeof(gid_t);
2409 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2410 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2411 		    sizeof(gid_t));
2412 	}
2413 	*sizep = size;
2414 }
2415 
2416 static void
2417 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2418 {
2419 	struct proc *p;
2420 	size_t size;
2421 	int structsize;
2422 
2423 	p = (struct proc *)arg;
2424 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2425 	if (sb != NULL) {
2426 		KASSERT(*sizep == size, ("invalid size"));
2427 		structsize = sizeof(p->p_fd->fd_cmask);
2428 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2429 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2430 	}
2431 	*sizep = size;
2432 }
2433 
2434 static void
2435 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2436 {
2437 	struct proc *p;
2438 	struct rlimit rlim[RLIM_NLIMITS];
2439 	size_t size;
2440 	int structsize, i;
2441 
2442 	p = (struct proc *)arg;
2443 	size = sizeof(structsize) + sizeof(rlim);
2444 	if (sb != NULL) {
2445 		KASSERT(*sizep == size, ("invalid size"));
2446 		structsize = sizeof(rlim);
2447 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2448 		PROC_LOCK(p);
2449 		for (i = 0; i < RLIM_NLIMITS; i++)
2450 			lim_rlimit_proc(p, i, &rlim[i]);
2451 		PROC_UNLOCK(p);
2452 		sbuf_bcat(sb, rlim, sizeof(rlim));
2453 	}
2454 	*sizep = size;
2455 }
2456 
2457 static void
2458 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2459 {
2460 	struct proc *p;
2461 	size_t size;
2462 	int structsize;
2463 
2464 	p = (struct proc *)arg;
2465 	size = sizeof(structsize) + sizeof(p->p_osrel);
2466 	if (sb != NULL) {
2467 		KASSERT(*sizep == size, ("invalid size"));
2468 		structsize = sizeof(p->p_osrel);
2469 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2470 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2471 	}
2472 	*sizep = size;
2473 }
2474 
2475 static void
2476 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2477 {
2478 	struct proc *p;
2479 	elf_ps_strings_t ps_strings;
2480 	size_t size;
2481 	int structsize;
2482 
2483 	p = (struct proc *)arg;
2484 	size = sizeof(structsize) + sizeof(ps_strings);
2485 	if (sb != NULL) {
2486 		KASSERT(*sizep == size, ("invalid size"));
2487 		structsize = sizeof(ps_strings);
2488 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2489 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2490 #else
2491 		ps_strings = p->p_sysent->sv_psstrings;
2492 #endif
2493 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2494 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2495 	}
2496 	*sizep = size;
2497 }
2498 
2499 static void
2500 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2501 {
2502 	struct proc *p;
2503 	size_t size;
2504 	int structsize;
2505 
2506 	p = (struct proc *)arg;
2507 	if (sb == NULL) {
2508 		size = 0;
2509 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2510 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2511 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2512 		PHOLD(p);
2513 		proc_getauxv(curthread, p, sb);
2514 		PRELE(p);
2515 		sbuf_finish(sb);
2516 		sbuf_delete(sb);
2517 		*sizep = size;
2518 	} else {
2519 		structsize = sizeof(Elf_Auxinfo);
2520 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2521 		PHOLD(p);
2522 		proc_getauxv(curthread, p, sb);
2523 		PRELE(p);
2524 	}
2525 }
2526 
2527 static boolean_t
2528 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2529     const char *note_vendor, const Elf_Phdr *pnote,
2530     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2531 {
2532 	const Elf_Note *note, *note0, *note_end;
2533 	const char *note_name;
2534 	char *buf;
2535 	int i, error;
2536 	boolean_t res;
2537 
2538 	/* We need some limit, might as well use PAGE_SIZE. */
2539 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2540 		return (FALSE);
2541 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2542 	if (pnote->p_offset > PAGE_SIZE ||
2543 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2544 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2545 		if (buf == NULL) {
2546 			VOP_UNLOCK(imgp->vp, 0);
2547 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2548 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2549 		}
2550 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2551 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2552 		    curthread->td_ucred, NOCRED, NULL, curthread);
2553 		if (error != 0) {
2554 			uprintf("i/o error PT_NOTE\n");
2555 			goto retf;
2556 		}
2557 		note = note0 = (const Elf_Note *)buf;
2558 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2559 	} else {
2560 		note = note0 = (const Elf_Note *)(imgp->image_header +
2561 		    pnote->p_offset);
2562 		note_end = (const Elf_Note *)(imgp->image_header +
2563 		    pnote->p_offset + pnote->p_filesz);
2564 		buf = NULL;
2565 	}
2566 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2567 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2568 		    (const char *)note < sizeof(Elf_Note)) {
2569 			goto retf;
2570 		}
2571 		if (note->n_namesz != checknote->n_namesz ||
2572 		    note->n_descsz != checknote->n_descsz ||
2573 		    note->n_type != checknote->n_type)
2574 			goto nextnote;
2575 		note_name = (const char *)(note + 1);
2576 		if (note_name + checknote->n_namesz >=
2577 		    (const char *)note_end || strncmp(note_vendor,
2578 		    note_name, checknote->n_namesz) != 0)
2579 			goto nextnote;
2580 
2581 		if (cb(note, cb_arg, &res))
2582 			goto ret;
2583 nextnote:
2584 		note = (const Elf_Note *)((const char *)(note + 1) +
2585 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2586 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2587 	}
2588 retf:
2589 	res = FALSE;
2590 ret:
2591 	free(buf, M_TEMP);
2592 	return (res);
2593 }
2594 
2595 struct brandnote_cb_arg {
2596 	Elf_Brandnote *brandnote;
2597 	int32_t *osrel;
2598 };
2599 
2600 static boolean_t
2601 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2602 {
2603 	struct brandnote_cb_arg *arg;
2604 
2605 	arg = arg0;
2606 
2607 	/*
2608 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2609 	 * necessary.
2610 	 */
2611 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2612 	    arg->brandnote->trans_osrel != NULL ?
2613 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2614 
2615 	return (TRUE);
2616 }
2617 
2618 static Elf_Note fctl_note = {
2619 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2620 	.n_descsz = sizeof(uint32_t),
2621 	.n_type = NT_FREEBSD_FEATURE_CTL,
2622 };
2623 
2624 struct fctl_cb_arg {
2625 	uint32_t *fctl0;
2626 };
2627 
2628 static boolean_t
2629 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2630 {
2631 	struct fctl_cb_arg *arg;
2632 	const Elf32_Word *desc;
2633 	uintptr_t p;
2634 
2635 	arg = arg0;
2636 	p = (uintptr_t)(note + 1);
2637 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2638 	desc = (const Elf32_Word *)p;
2639 	*arg->fctl0 = desc[0];
2640 	return (TRUE);
2641 }
2642 
2643 /*
2644  * Try to find the appropriate ABI-note section for checknote, fetch
2645  * the osreldate and feature control flags for binary from the ELF
2646  * OSABI-note.  Only the first page of the image is searched, the same
2647  * as for headers.
2648  */
2649 static boolean_t
2650 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2651     int32_t *osrel, uint32_t *fctl0)
2652 {
2653 	const Elf_Phdr *phdr;
2654 	const Elf_Ehdr *hdr;
2655 	struct brandnote_cb_arg b_arg;
2656 	struct fctl_cb_arg f_arg;
2657 	int i, j;
2658 
2659 	hdr = (const Elf_Ehdr *)imgp->image_header;
2660 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2661 	b_arg.brandnote = brandnote;
2662 	b_arg.osrel = osrel;
2663 	f_arg.fctl0 = fctl0;
2664 
2665 	for (i = 0; i < hdr->e_phnum; i++) {
2666 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2667 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2668 		    &b_arg)) {
2669 			for (j = 0; j < hdr->e_phnum; j++) {
2670 				if (phdr[j].p_type == PT_NOTE &&
2671 				    __elfN(parse_notes)(imgp, &fctl_note,
2672 				    FREEBSD_ABI_VENDOR, &phdr[j],
2673 				    note_fctl_cb, &f_arg))
2674 					break;
2675 			}
2676 			return (TRUE);
2677 		}
2678 	}
2679 	return (FALSE);
2680 
2681 }
2682 
2683 /*
2684  * Tell kern_execve.c about it, with a little help from the linker.
2685  */
2686 static struct execsw __elfN(execsw) = {
2687 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2688 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2689 };
2690 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2691 
2692 static vm_prot_t
2693 __elfN(trans_prot)(Elf_Word flags)
2694 {
2695 	vm_prot_t prot;
2696 
2697 	prot = 0;
2698 	if (flags & PF_X)
2699 		prot |= VM_PROT_EXECUTE;
2700 	if (flags & PF_W)
2701 		prot |= VM_PROT_WRITE;
2702 	if (flags & PF_R)
2703 		prot |= VM_PROT_READ;
2704 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2705 	if (i386_read_exec && (flags & PF_R))
2706 		prot |= VM_PROT_EXECUTE;
2707 #endif
2708 	return (prot);
2709 }
2710 
2711 static Elf_Word
2712 __elfN(untrans_prot)(vm_prot_t prot)
2713 {
2714 	Elf_Word flags;
2715 
2716 	flags = 0;
2717 	if (prot & VM_PROT_EXECUTE)
2718 		flags |= PF_X;
2719 	if (prot & VM_PROT_READ)
2720 		flags |= PF_R;
2721 	if (prot & VM_PROT_WRITE)
2722 		flags |= PF_W;
2723 	return (flags);
2724 }
2725 
2726 void
2727 __elfN(stackgap)(struct image_params *imgp, u_long *stack_base)
2728 {
2729 	u_long range, rbase, gap;
2730 	int pct;
2731 
2732 	if ((imgp->map_flags & MAP_ASLR) == 0)
2733 		return;
2734 	pct = __elfN(aslr_stack_gap);
2735 	if (pct == 0)
2736 		return;
2737 	if (pct > 50)
2738 		pct = 50;
2739 	range = imgp->eff_stack_sz * pct / 100;
2740 	arc4rand(&rbase, sizeof(rbase), 0);
2741 	gap = rbase % range;
2742 	gap &= ~(sizeof(u_long) - 1);
2743 	*stack_base -= gap;
2744 }
2745