xref: /freebsd/sys/kern/imgact_elf.c (revision f2530c80db7b29b95368fce956b3a778f096b368)
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 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
139 static int
140 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
141 {
142 	u_long val;
143 	int error;
144 
145 	val = __elfN(pie_base);
146 	error = sysctl_handle_long(oidp, &val, 0, req);
147 	if (error != 0 || req->newptr == NULL)
148 		return (error);
149 	if ((val & PAGE_MASK) != 0)
150 		return (EINVAL);
151 	__elfN(pie_base) = val;
152 	return (0);
153 }
154 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
155     CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
156     sysctl_pie_base, "LU",
157     "PIE load base without randomization");
158 
159 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
160     "");
161 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
162 
163 static int __elfN(aslr_enabled) = 0;
164 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
165     &__elfN(aslr_enabled), 0,
166     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
167     ": enable address map randomization");
168 
169 static int __elfN(pie_aslr_enabled) = 0;
170 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
171     &__elfN(pie_aslr_enabled), 0,
172     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
173     ": enable address map randomization for PIE binaries");
174 
175 static int __elfN(aslr_honor_sbrk) = 1;
176 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
177     &__elfN(aslr_honor_sbrk), 0,
178     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
179 
180 static int __elfN(aslr_stack_gap) = 3;
181 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
182     &__elfN(aslr_stack_gap), 0,
183     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
184     ": maximum percentage of main stack to waste on a random gap");
185 
186 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
187 
188 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
189 
190 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
191 
192 Elf_Brandnote __elfN(freebsd_brandnote) = {
193 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
194 	.hdr.n_descsz	= sizeof(int32_t),
195 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
196 	.vendor		= FREEBSD_ABI_VENDOR,
197 	.flags		= BN_TRANSLATE_OSREL,
198 	.trans_osrel	= __elfN(freebsd_trans_osrel)
199 };
200 
201 static bool
202 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
203 {
204 	uintptr_t p;
205 
206 	p = (uintptr_t)(note + 1);
207 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
208 	*osrel = *(const int32_t *)(p);
209 
210 	return (true);
211 }
212 
213 static const char GNU_ABI_VENDOR[] = "GNU";
214 static int GNU_KFREEBSD_ABI_DESC = 3;
215 
216 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
217 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
218 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
219 	.hdr.n_type	= 1,
220 	.vendor		= GNU_ABI_VENDOR,
221 	.flags		= BN_TRANSLATE_OSREL,
222 	.trans_osrel	= kfreebsd_trans_osrel
223 };
224 
225 static bool
226 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
227 {
228 	const Elf32_Word *desc;
229 	uintptr_t p;
230 
231 	p = (uintptr_t)(note + 1);
232 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
233 
234 	desc = (const Elf32_Word *)p;
235 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
236 		return (false);
237 
238 	/*
239 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
240 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
241 	 */
242 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
243 
244 	return (true);
245 }
246 
247 int
248 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
249 {
250 	int i;
251 
252 	for (i = 0; i < MAX_BRANDS; i++) {
253 		if (elf_brand_list[i] == NULL) {
254 			elf_brand_list[i] = entry;
255 			break;
256 		}
257 	}
258 	if (i == MAX_BRANDS) {
259 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
260 			__func__, entry);
261 		return (-1);
262 	}
263 	return (0);
264 }
265 
266 int
267 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
268 {
269 	int i;
270 
271 	for (i = 0; i < MAX_BRANDS; i++) {
272 		if (elf_brand_list[i] == entry) {
273 			elf_brand_list[i] = NULL;
274 			break;
275 		}
276 	}
277 	if (i == MAX_BRANDS)
278 		return (-1);
279 	return (0);
280 }
281 
282 int
283 __elfN(brand_inuse)(Elf_Brandinfo *entry)
284 {
285 	struct proc *p;
286 	int rval = FALSE;
287 
288 	sx_slock(&allproc_lock);
289 	FOREACH_PROC_IN_SYSTEM(p) {
290 		if (p->p_sysent == entry->sysvec) {
291 			rval = TRUE;
292 			break;
293 		}
294 	}
295 	sx_sunlock(&allproc_lock);
296 
297 	return (rval);
298 }
299 
300 static Elf_Brandinfo *
301 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
302     int32_t *osrel, uint32_t *fctl0)
303 {
304 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
305 	Elf_Brandinfo *bi, *bi_m;
306 	boolean_t ret;
307 	int i, interp_name_len;
308 
309 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
310 
311 	/*
312 	 * We support four types of branding -- (1) the ELF EI_OSABI field
313 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
314 	 * branding w/in the ELF header, (3) path of the `interp_path'
315 	 * field, and (4) the ".note.ABI-tag" ELF section.
316 	 */
317 
318 	/* Look for an ".note.ABI-tag" ELF section */
319 	bi_m = NULL;
320 	for (i = 0; i < MAX_BRANDS; i++) {
321 		bi = elf_brand_list[i];
322 		if (bi == NULL)
323 			continue;
324 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
325 			continue;
326 		if (hdr->e_machine == bi->machine && (bi->flags &
327 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
328 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
329 			    fctl0);
330 			/* Give brand a chance to veto check_note's guess */
331 			if (ret && bi->header_supported)
332 				ret = bi->header_supported(imgp);
333 			/*
334 			 * If note checker claimed the binary, but the
335 			 * interpreter path in the image does not
336 			 * match default one for the brand, try to
337 			 * search for other brands with the same
338 			 * interpreter.  Either there is better brand
339 			 * with the right interpreter, or, failing
340 			 * this, we return first brand which accepted
341 			 * our note and, optionally, header.
342 			 */
343 			if (ret && bi_m == NULL && interp != NULL &&
344 			    (bi->interp_path == NULL ||
345 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
346 			    strncmp(interp, bi->interp_path, interp_name_len)
347 			    != 0))) {
348 				bi_m = bi;
349 				ret = 0;
350 			}
351 			if (ret)
352 				return (bi);
353 		}
354 	}
355 	if (bi_m != NULL)
356 		return (bi_m);
357 
358 	/* If the executable has a brand, search for it in the brand list. */
359 	for (i = 0; i < MAX_BRANDS; i++) {
360 		bi = elf_brand_list[i];
361 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
362 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
363 			continue;
364 		if (hdr->e_machine == bi->machine &&
365 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
366 		    (bi->compat_3_brand != NULL &&
367 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
368 		    bi->compat_3_brand) == 0))) {
369 			/* Looks good, but give brand a chance to veto */
370 			if (bi->header_supported == NULL ||
371 			    bi->header_supported(imgp)) {
372 				/*
373 				 * Again, prefer strictly matching
374 				 * interpreter path.
375 				 */
376 				if (interp_name_len == 0 &&
377 				    bi->interp_path == NULL)
378 					return (bi);
379 				if (bi->interp_path != NULL &&
380 				    strlen(bi->interp_path) + 1 ==
381 				    interp_name_len && strncmp(interp,
382 				    bi->interp_path, interp_name_len) == 0)
383 					return (bi);
384 				if (bi_m == NULL)
385 					bi_m = bi;
386 			}
387 		}
388 	}
389 	if (bi_m != NULL)
390 		return (bi_m);
391 
392 	/* No known brand, see if the header is recognized by any brand */
393 	for (i = 0; i < MAX_BRANDS; i++) {
394 		bi = elf_brand_list[i];
395 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
396 		    bi->header_supported == NULL)
397 			continue;
398 		if (hdr->e_machine == bi->machine) {
399 			ret = bi->header_supported(imgp);
400 			if (ret)
401 				return (bi);
402 		}
403 	}
404 
405 	/* Lacking a known brand, search for a recognized interpreter. */
406 	if (interp != NULL) {
407 		for (i = 0; i < MAX_BRANDS; i++) {
408 			bi = elf_brand_list[i];
409 			if (bi == NULL || (bi->flags &
410 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
411 			    != 0)
412 				continue;
413 			if (hdr->e_machine == bi->machine &&
414 			    bi->interp_path != NULL &&
415 			    /* ELF image p_filesz includes terminating zero */
416 			    strlen(bi->interp_path) + 1 == interp_name_len &&
417 			    strncmp(interp, bi->interp_path, interp_name_len)
418 			    == 0 && (bi->header_supported == NULL ||
419 			    bi->header_supported(imgp)))
420 				return (bi);
421 		}
422 	}
423 
424 	/* Lacking a recognized interpreter, try the default brand */
425 	for (i = 0; i < MAX_BRANDS; i++) {
426 		bi = elf_brand_list[i];
427 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
428 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
429 			continue;
430 		if (hdr->e_machine == bi->machine &&
431 		    __elfN(fallback_brand) == bi->brand &&
432 		    (bi->header_supported == NULL ||
433 		    bi->header_supported(imgp)))
434 			return (bi);
435 	}
436 	return (NULL);
437 }
438 
439 static int
440 __elfN(check_header)(const Elf_Ehdr *hdr)
441 {
442 	Elf_Brandinfo *bi;
443 	int i;
444 
445 	if (!IS_ELF(*hdr) ||
446 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
447 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
448 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
449 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
450 	    hdr->e_version != ELF_TARG_VER)
451 		return (ENOEXEC);
452 
453 	/*
454 	 * Make sure we have at least one brand for this machine.
455 	 */
456 
457 	for (i = 0; i < MAX_BRANDS; i++) {
458 		bi = elf_brand_list[i];
459 		if (bi != NULL && bi->machine == hdr->e_machine)
460 			break;
461 	}
462 	if (i == MAX_BRANDS)
463 		return (ENOEXEC);
464 
465 	return (0);
466 }
467 
468 static int
469 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
470     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
471 {
472 	struct sf_buf *sf;
473 	int error;
474 	vm_offset_t off;
475 
476 	/*
477 	 * Create the page if it doesn't exist yet. Ignore errors.
478 	 */
479 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
480 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
481 
482 	/*
483 	 * Find the page from the underlying object.
484 	 */
485 	if (object != NULL) {
486 		sf = vm_imgact_map_page(object, offset);
487 		if (sf == NULL)
488 			return (KERN_FAILURE);
489 		off = offset - trunc_page(offset);
490 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
491 		    end - start);
492 		vm_imgact_unmap_page(sf);
493 		if (error != 0)
494 			return (KERN_FAILURE);
495 	}
496 
497 	return (KERN_SUCCESS);
498 }
499 
500 static int
501 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
502     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
503     int cow)
504 {
505 	struct sf_buf *sf;
506 	vm_offset_t off;
507 	vm_size_t sz;
508 	int error, locked, rv;
509 
510 	if (start != trunc_page(start)) {
511 		rv = __elfN(map_partial)(map, object, offset, start,
512 		    round_page(start), prot);
513 		if (rv != KERN_SUCCESS)
514 			return (rv);
515 		offset += round_page(start) - start;
516 		start = round_page(start);
517 	}
518 	if (end != round_page(end)) {
519 		rv = __elfN(map_partial)(map, object, offset +
520 		    trunc_page(end) - start, trunc_page(end), end, prot);
521 		if (rv != KERN_SUCCESS)
522 			return (rv);
523 		end = trunc_page(end);
524 	}
525 	if (start >= end)
526 		return (KERN_SUCCESS);
527 	if ((offset & PAGE_MASK) != 0) {
528 		/*
529 		 * The mapping is not page aligned.  This means that we have
530 		 * to copy the data.
531 		 */
532 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
533 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
534 		if (rv != KERN_SUCCESS)
535 			return (rv);
536 		if (object == NULL)
537 			return (KERN_SUCCESS);
538 		for (; start < end; start += sz) {
539 			sf = vm_imgact_map_page(object, offset);
540 			if (sf == NULL)
541 				return (KERN_FAILURE);
542 			off = offset - trunc_page(offset);
543 			sz = end - start;
544 			if (sz > PAGE_SIZE - off)
545 				sz = PAGE_SIZE - off;
546 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
547 			    (caddr_t)start, sz);
548 			vm_imgact_unmap_page(sf);
549 			if (error != 0)
550 				return (KERN_FAILURE);
551 			offset += sz;
552 		}
553 	} else {
554 		vm_object_reference(object);
555 		rv = vm_map_fixed(map, object, offset, start, end - start,
556 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
557 		    (object != NULL ? MAP_VN_EXEC : 0));
558 		if (rv != KERN_SUCCESS) {
559 			locked = VOP_ISLOCKED(imgp->vp);
560 			VOP_UNLOCK(imgp->vp);
561 			vm_object_deallocate(object);
562 			vn_lock(imgp->vp, locked | LK_RETRY);
563 			return (rv);
564 		} else if (object != NULL) {
565 			MPASS(imgp->vp->v_object == object);
566 			VOP_SET_TEXT_CHECKED(imgp->vp);
567 		}
568 	}
569 	return (KERN_SUCCESS);
570 }
571 
572 static int
573 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
574     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
575 {
576 	struct sf_buf *sf;
577 	size_t map_len;
578 	vm_map_t map;
579 	vm_object_t object;
580 	vm_offset_t map_addr;
581 	int error, rv, cow;
582 	size_t copy_len;
583 	vm_ooffset_t file_addr;
584 
585 	/*
586 	 * It's necessary to fail if the filsz + offset taken from the
587 	 * header is greater than the actual file pager object's size.
588 	 * If we were to allow this, then the vm_map_find() below would
589 	 * walk right off the end of the file object and into the ether.
590 	 *
591 	 * While I'm here, might as well check for something else that
592 	 * is invalid: filsz cannot be greater than memsz.
593 	 */
594 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
595 	    filsz > memsz) {
596 		uprintf("elf_load_section: truncated ELF file\n");
597 		return (ENOEXEC);
598 	}
599 
600 	object = imgp->object;
601 	map = &imgp->proc->p_vmspace->vm_map;
602 	map_addr = trunc_page((vm_offset_t)vmaddr);
603 	file_addr = trunc_page(offset);
604 
605 	/*
606 	 * We have two choices.  We can either clear the data in the last page
607 	 * of an oversized mapping, or we can start the anon mapping a page
608 	 * early and copy the initialized data into that first page.  We
609 	 * choose the second.
610 	 */
611 	if (filsz == 0)
612 		map_len = 0;
613 	else if (memsz > filsz)
614 		map_len = trunc_page(offset + filsz) - file_addr;
615 	else
616 		map_len = round_page(offset + filsz) - file_addr;
617 
618 	if (map_len != 0) {
619 		/* cow flags: don't dump readonly sections in core */
620 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
621 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
622 
623 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
624 		    map_addr, map_addr + map_len, prot, cow);
625 		if (rv != KERN_SUCCESS)
626 			return (EINVAL);
627 
628 		/* we can stop now if we've covered it all */
629 		if (memsz == filsz)
630 			return (0);
631 	}
632 
633 	/*
634 	 * We have to get the remaining bit of the file into the first part
635 	 * of the oversized map segment.  This is normally because the .data
636 	 * segment in the file is extended to provide bss.  It's a neat idea
637 	 * to try and save a page, but it's a pain in the behind to implement.
638 	 */
639 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
640 	    filsz);
641 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
642 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
643 
644 	/* This had damn well better be true! */
645 	if (map_len != 0) {
646 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
647 		    map_addr + map_len, prot, 0);
648 		if (rv != KERN_SUCCESS)
649 			return (EINVAL);
650 	}
651 
652 	if (copy_len != 0) {
653 		sf = vm_imgact_map_page(object, offset + filsz);
654 		if (sf == NULL)
655 			return (EIO);
656 
657 		/* send the page fragment to user space */
658 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
659 		    copy_len);
660 		vm_imgact_unmap_page(sf);
661 		if (error != 0)
662 			return (error);
663 	}
664 
665 	/*
666 	 * Remove write access to the page if it was only granted by map_insert
667 	 * to allow copyout.
668 	 */
669 	if ((prot & VM_PROT_WRITE) == 0)
670 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
671 		    map_len), prot, FALSE);
672 
673 	return (0);
674 }
675 
676 static int
677 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
678     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
679 {
680 	vm_prot_t prot;
681 	u_long base_addr;
682 	bool first;
683 	int error, i;
684 
685 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
686 
687 	base_addr = 0;
688 	first = true;
689 
690 	for (i = 0; i < hdr->e_phnum; i++) {
691 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
692 			continue;
693 
694 		/* Loadable segment */
695 		prot = __elfN(trans_prot)(phdr[i].p_flags);
696 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
697 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
698 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
699 		if (error != 0)
700 			return (error);
701 
702 		/*
703 		 * Establish the base address if this is the first segment.
704 		 */
705 		if (first) {
706   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
707 			first = false;
708 		}
709 	}
710 
711 	if (base_addrp != NULL)
712 		*base_addrp = base_addr;
713 
714 	return (0);
715 }
716 
717 /*
718  * Load the file "file" into memory.  It may be either a shared object
719  * or an executable.
720  *
721  * The "addr" reference parameter is in/out.  On entry, it specifies
722  * the address where a shared object should be loaded.  If the file is
723  * an executable, this value is ignored.  On exit, "addr" specifies
724  * where the file was actually loaded.
725  *
726  * The "entry" reference parameter is out only.  On exit, it specifies
727  * the entry point for the loaded file.
728  */
729 static int
730 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
731 	u_long *entry)
732 {
733 	struct {
734 		struct nameidata nd;
735 		struct vattr attr;
736 		struct image_params image_params;
737 	} *tempdata;
738 	const Elf_Ehdr *hdr = NULL;
739 	const Elf_Phdr *phdr = NULL;
740 	struct nameidata *nd;
741 	struct vattr *attr;
742 	struct image_params *imgp;
743 	u_long rbase;
744 	u_long base_addr = 0;
745 	int error;
746 
747 #ifdef CAPABILITY_MODE
748 	/*
749 	 * XXXJA: This check can go away once we are sufficiently confident
750 	 * that the checks in namei() are correct.
751 	 */
752 	if (IN_CAPABILITY_MODE(curthread))
753 		return (ECAPMODE);
754 #endif
755 
756 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
757 	nd = &tempdata->nd;
758 	attr = &tempdata->attr;
759 	imgp = &tempdata->image_params;
760 
761 	/*
762 	 * Initialize part of the common data
763 	 */
764 	imgp->proc = p;
765 	imgp->attr = attr;
766 
767 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
768 	    UIO_SYSSPACE, file, curthread);
769 	if ((error = namei(nd)) != 0) {
770 		nd->ni_vp = NULL;
771 		goto fail;
772 	}
773 	NDFREE(nd, NDF_ONLY_PNBUF);
774 	imgp->vp = nd->ni_vp;
775 
776 	/*
777 	 * Check permissions, modes, uid, etc on the file, and "open" it.
778 	 */
779 	error = exec_check_permissions(imgp);
780 	if (error)
781 		goto fail;
782 
783 	error = exec_map_first_page(imgp);
784 	if (error)
785 		goto fail;
786 
787 	imgp->object = nd->ni_vp->v_object;
788 
789 	hdr = (const Elf_Ehdr *)imgp->image_header;
790 	if ((error = __elfN(check_header)(hdr)) != 0)
791 		goto fail;
792 	if (hdr->e_type == ET_DYN)
793 		rbase = *addr;
794 	else if (hdr->e_type == ET_EXEC)
795 		rbase = 0;
796 	else {
797 		error = ENOEXEC;
798 		goto fail;
799 	}
800 
801 	/* Only support headers that fit within first page for now      */
802 	if ((hdr->e_phoff > PAGE_SIZE) ||
803 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
804 		error = ENOEXEC;
805 		goto fail;
806 	}
807 
808 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
809 	if (!aligned(phdr, Elf_Addr)) {
810 		error = ENOEXEC;
811 		goto fail;
812 	}
813 
814 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
815 	if (error != 0)
816 		goto fail;
817 
818 	*addr = base_addr;
819 	*entry = (unsigned long)hdr->e_entry + rbase;
820 
821 fail:
822 	if (imgp->firstpage)
823 		exec_unmap_first_page(imgp);
824 
825 	if (nd->ni_vp) {
826 		if (imgp->textset)
827 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
828 		vput(nd->ni_vp);
829 	}
830 	free(tempdata, M_TEMP);
831 
832 	return (error);
833 }
834 
835 static u_long
836 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
837     u_int align)
838 {
839 	u_long rbase, res;
840 
841 	MPASS(vm_map_min(map) <= minv);
842 	MPASS(maxv <= vm_map_max(map));
843 	MPASS(minv < maxv);
844 	MPASS(minv + align < maxv);
845 	arc4rand(&rbase, sizeof(rbase), 0);
846 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
847 	res &= ~((u_long)align - 1);
848 	if (res >= maxv)
849 		res -= align;
850 	KASSERT(res >= minv,
851 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
852 	    res, minv, maxv, rbase));
853 	KASSERT(res < maxv,
854 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
855 	    res, maxv, minv, rbase));
856 	return (res);
857 }
858 
859 static int
860 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
861     const Elf_Phdr *phdr, u_long et_dyn_addr)
862 {
863 	struct vmspace *vmspace;
864 	const char *err_str;
865 	u_long text_size, data_size, total_size, text_addr, data_addr;
866 	u_long seg_size, seg_addr;
867 	int i;
868 
869 	err_str = NULL;
870 	text_size = data_size = total_size = text_addr = data_addr = 0;
871 
872 	for (i = 0; i < hdr->e_phnum; i++) {
873 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
874 			continue;
875 
876 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
877 		seg_size = round_page(phdr[i].p_memsz +
878 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
879 
880 		/*
881 		 * Make the largest executable segment the official
882 		 * text segment and all others data.
883 		 *
884 		 * Note that obreak() assumes that data_addr + data_size == end
885 		 * of data load area, and the ELF file format expects segments
886 		 * to be sorted by address.  If multiple data segments exist,
887 		 * the last one will be used.
888 		 */
889 
890 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
891 			text_size = seg_size;
892 			text_addr = seg_addr;
893 		} else {
894 			data_size = seg_size;
895 			data_addr = seg_addr;
896 		}
897 		total_size += seg_size;
898 	}
899 
900 	if (data_addr == 0 && data_size == 0) {
901 		data_addr = text_addr;
902 		data_size = text_size;
903 	}
904 
905 	/*
906 	 * Check limits.  It should be safe to check the
907 	 * limits after loading the segments since we do
908 	 * not actually fault in all the segments pages.
909 	 */
910 	PROC_LOCK(imgp->proc);
911 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
912 		err_str = "Data segment size exceeds process limit";
913 	else if (text_size > maxtsiz)
914 		err_str = "Text segment size exceeds system limit";
915 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
916 		err_str = "Total segment size exceeds process limit";
917 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
918 		err_str = "Data segment size exceeds resource limit";
919 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
920 		err_str = "Total segment size exceeds resource limit";
921 	PROC_UNLOCK(imgp->proc);
922 	if (err_str != NULL) {
923 		uprintf("%s\n", err_str);
924 		return (ENOMEM);
925 	}
926 
927 	vmspace = imgp->proc->p_vmspace;
928 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
929 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
930 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
931 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
932 
933 	return (0);
934 }
935 
936 static int
937 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
938     char **interpp, bool *free_interpp)
939 {
940 	struct thread *td;
941 	char *interp;
942 	int error, interp_name_len;
943 
944 	KASSERT(phdr->p_type == PT_INTERP,
945 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
946 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
947 
948 	td = curthread;
949 
950 	/* Path to interpreter */
951 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
952 		uprintf("Invalid PT_INTERP\n");
953 		return (ENOEXEC);
954 	}
955 
956 	interp_name_len = phdr->p_filesz;
957 	if (phdr->p_offset > PAGE_SIZE ||
958 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
959 		/*
960 		 * The vnode lock might be needed by the pagedaemon to
961 		 * clean pages owned by the vnode.  Do not allow sleep
962 		 * waiting for memory with the vnode locked, instead
963 		 * try non-sleepable allocation first, and if it
964 		 * fails, go to the slow path were we drop the lock
965 		 * and do M_WAITOK.  A text reference prevents
966 		 * modifications to the vnode content.
967 		 */
968 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
969 		if (interp == NULL) {
970 			VOP_UNLOCK(imgp->vp);
971 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
972 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
973 		}
974 
975 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
976 		    interp_name_len, phdr->p_offset,
977 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
978 		    NOCRED, NULL, td);
979 		if (error != 0) {
980 			free(interp, M_TEMP);
981 			uprintf("i/o error PT_INTERP %d\n", error);
982 			return (error);
983 		}
984 		interp[interp_name_len] = '\0';
985 
986 		*interpp = interp;
987 		*free_interpp = true;
988 		return (0);
989 	}
990 
991 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
992 	if (interp[interp_name_len - 1] != '\0') {
993 		uprintf("Invalid PT_INTERP\n");
994 		return (ENOEXEC);
995 	}
996 
997 	*interpp = interp;
998 	*free_interpp = false;
999 	return (0);
1000 }
1001 
1002 static int
1003 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1004     const char *interp, u_long *addr, u_long *entry)
1005 {
1006 	char *path;
1007 	int error;
1008 
1009 	if (brand_info->emul_path != NULL &&
1010 	    brand_info->emul_path[0] != '\0') {
1011 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1012 		snprintf(path, MAXPATHLEN, "%s%s",
1013 		    brand_info->emul_path, interp);
1014 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1015 		free(path, M_TEMP);
1016 		if (error == 0)
1017 			return (0);
1018 	}
1019 
1020 	if (brand_info->interp_newpath != NULL &&
1021 	    (brand_info->interp_path == NULL ||
1022 	    strcmp(interp, brand_info->interp_path) == 0)) {
1023 		error = __elfN(load_file)(imgp->proc,
1024 		    brand_info->interp_newpath, addr, entry);
1025 		if (error == 0)
1026 			return (0);
1027 	}
1028 
1029 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1030 	if (error == 0)
1031 		return (0);
1032 
1033 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1034 	return (error);
1035 }
1036 
1037 /*
1038  * Impossible et_dyn_addr initial value indicating that the real base
1039  * must be calculated later with some randomization applied.
1040  */
1041 #define	ET_DYN_ADDR_RAND	1
1042 
1043 static int
1044 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1045 {
1046 	struct thread *td;
1047 	const Elf_Ehdr *hdr;
1048 	const Elf_Phdr *phdr;
1049 	Elf_Auxargs *elf_auxargs;
1050 	struct vmspace *vmspace;
1051 	vm_map_t map;
1052 	char *interp;
1053 	Elf_Brandinfo *brand_info;
1054 	struct sysentvec *sv;
1055 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1056 	u_long maxalign, mapsz, maxv, maxv1;
1057 	uint32_t fctl0;
1058 	int32_t osrel;
1059 	bool free_interp;
1060 	int error, i, n;
1061 
1062 	hdr = (const Elf_Ehdr *)imgp->image_header;
1063 
1064 	/*
1065 	 * Do we have a valid ELF header ?
1066 	 *
1067 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1068 	 * if particular brand doesn't support it.
1069 	 */
1070 	if (__elfN(check_header)(hdr) != 0 ||
1071 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1072 		return (-1);
1073 
1074 	/*
1075 	 * From here on down, we return an errno, not -1, as we've
1076 	 * detected an ELF file.
1077 	 */
1078 
1079 	if ((hdr->e_phoff > PAGE_SIZE) ||
1080 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1081 		/* Only support headers in first page for now */
1082 		uprintf("Program headers not in the first page\n");
1083 		return (ENOEXEC);
1084 	}
1085 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1086 	if (!aligned(phdr, Elf_Addr)) {
1087 		uprintf("Unaligned program headers\n");
1088 		return (ENOEXEC);
1089 	}
1090 
1091 	n = error = 0;
1092 	baddr = 0;
1093 	osrel = 0;
1094 	fctl0 = 0;
1095 	entry = proghdr = 0;
1096 	interp = NULL;
1097 	free_interp = false;
1098 	td = curthread;
1099 	maxalign = PAGE_SIZE;
1100 	mapsz = 0;
1101 
1102 	for (i = 0; i < hdr->e_phnum; i++) {
1103 		switch (phdr[i].p_type) {
1104 		case PT_LOAD:
1105 			if (n == 0)
1106 				baddr = phdr[i].p_vaddr;
1107 			if (phdr[i].p_align > maxalign)
1108 				maxalign = phdr[i].p_align;
1109 			mapsz += phdr[i].p_memsz;
1110 			n++;
1111 
1112 			/*
1113 			 * If this segment contains the program headers,
1114 			 * remember their virtual address for the AT_PHDR
1115 			 * aux entry. Static binaries don't usually include
1116 			 * a PT_PHDR entry.
1117 			 */
1118 			if (phdr[i].p_offset == 0 &&
1119 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1120 				<= phdr[i].p_filesz)
1121 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1122 			break;
1123 		case PT_INTERP:
1124 			/* Path to interpreter */
1125 			if (interp != NULL) {
1126 				uprintf("Multiple PT_INTERP headers\n");
1127 				error = ENOEXEC;
1128 				goto ret;
1129 			}
1130 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1131 			    &free_interp);
1132 			if (error != 0)
1133 				goto ret;
1134 			break;
1135 		case PT_GNU_STACK:
1136 			if (__elfN(nxstack))
1137 				imgp->stack_prot =
1138 				    __elfN(trans_prot)(phdr[i].p_flags);
1139 			imgp->stack_sz = phdr[i].p_memsz;
1140 			break;
1141 		case PT_PHDR: 	/* Program header table info */
1142 			proghdr = phdr[i].p_vaddr;
1143 			break;
1144 		}
1145 	}
1146 
1147 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1148 	if (brand_info == NULL) {
1149 		uprintf("ELF binary type \"%u\" not known.\n",
1150 		    hdr->e_ident[EI_OSABI]);
1151 		error = ENOEXEC;
1152 		goto ret;
1153 	}
1154 	sv = brand_info->sysvec;
1155 	et_dyn_addr = 0;
1156 	if (hdr->e_type == ET_DYN) {
1157 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1158 			uprintf("Cannot execute shared object\n");
1159 			error = ENOEXEC;
1160 			goto ret;
1161 		}
1162 		/*
1163 		 * Honour the base load address from the dso if it is
1164 		 * non-zero for some reason.
1165 		 */
1166 		if (baddr == 0) {
1167 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1168 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1169 				et_dyn_addr = __elfN(pie_base);
1170 			else if ((__elfN(pie_aslr_enabled) &&
1171 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1172 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1173 				et_dyn_addr = ET_DYN_ADDR_RAND;
1174 			else
1175 				et_dyn_addr = __elfN(pie_base);
1176 		}
1177 	}
1178 
1179 	/*
1180 	 * Avoid a possible deadlock if the current address space is destroyed
1181 	 * and that address space maps the locked vnode.  In the common case,
1182 	 * the locked vnode's v_usecount is decremented but remains greater
1183 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1184 	 * However, in cases where the vnode lock is external, such as nullfs,
1185 	 * v_usecount may become zero.
1186 	 *
1187 	 * The VV_TEXT flag prevents modifications to the executable while
1188 	 * the vnode is unlocked.
1189 	 */
1190 	VOP_UNLOCK(imgp->vp);
1191 
1192 	/*
1193 	 * Decide whether to enable randomization of user mappings.
1194 	 * First, reset user preferences for the setid binaries.
1195 	 * Then, account for the support of the randomization by the
1196 	 * ABI, by user preferences, and make special treatment for
1197 	 * PIE binaries.
1198 	 */
1199 	if (imgp->credential_setid) {
1200 		PROC_LOCK(imgp->proc);
1201 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1202 		PROC_UNLOCK(imgp->proc);
1203 	}
1204 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1205 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1206 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1207 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1208 		    ("et_dyn_addr == RAND and !ASLR"));
1209 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1210 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1211 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1212 		imgp->map_flags |= MAP_ASLR;
1213 		/*
1214 		 * If user does not care about sbrk, utilize the bss
1215 		 * grow region for mappings as well.  We can select
1216 		 * the base for the image anywere and still not suffer
1217 		 * from the fragmentation.
1218 		 */
1219 		if (!__elfN(aslr_honor_sbrk) ||
1220 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1221 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1222 	}
1223 
1224 	error = exec_new_vmspace(imgp, sv);
1225 	vmspace = imgp->proc->p_vmspace;
1226 	map = &vmspace->vm_map;
1227 
1228 	imgp->proc->p_sysent = sv;
1229 
1230 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1231 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1232 		KASSERT((map->flags & MAP_ASLR) != 0,
1233 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1234 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1235 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1236 		    /* reserve half of the address space to interpreter */
1237 		    maxv / 2, 1UL << flsl(maxalign));
1238 	}
1239 
1240 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1241 	if (error != 0)
1242 		goto ret;
1243 
1244 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1245 	if (error != 0)
1246 		goto ret;
1247 
1248 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1249 	if (error != 0)
1250 		goto ret;
1251 
1252 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1253 
1254 	/*
1255 	 * We load the dynamic linker where a userland call
1256 	 * to mmap(0, ...) would put it.  The rationale behind this
1257 	 * calculation is that it leaves room for the heap to grow to
1258 	 * its maximum allowed size.
1259 	 */
1260 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1261 	    RLIMIT_DATA));
1262 	if ((map->flags & MAP_ASLR) != 0) {
1263 		maxv1 = maxv / 2 + addr / 2;
1264 		MPASS(maxv1 >= addr);	/* No overflow */
1265 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1266 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1267 	} else {
1268 		map->anon_loc = addr;
1269 	}
1270 
1271 	imgp->entry_addr = entry;
1272 
1273 	if (interp != NULL) {
1274 		VOP_UNLOCK(imgp->vp);
1275 		if ((map->flags & MAP_ASLR) != 0) {
1276 			/* Assume that interpeter fits into 1/4 of AS */
1277 			maxv1 = maxv / 2 + addr / 2;
1278 			MPASS(maxv1 >= addr);	/* No overflow */
1279 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1280 			    maxv1, PAGE_SIZE);
1281 		}
1282 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1283 		    &imgp->entry_addr);
1284 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1285 		if (error != 0)
1286 			goto ret;
1287 	} else
1288 		addr = et_dyn_addr;
1289 
1290 	/*
1291 	 * Construct auxargs table (used by the copyout_auxargs routine)
1292 	 */
1293 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1294 	if (elf_auxargs == NULL) {
1295 		VOP_UNLOCK(imgp->vp);
1296 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1297 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1298 	}
1299 	elf_auxargs->execfd = -1;
1300 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1301 	elf_auxargs->phent = hdr->e_phentsize;
1302 	elf_auxargs->phnum = hdr->e_phnum;
1303 	elf_auxargs->pagesz = PAGE_SIZE;
1304 	elf_auxargs->base = addr;
1305 	elf_auxargs->flags = 0;
1306 	elf_auxargs->entry = entry;
1307 	elf_auxargs->hdr_eflags = hdr->e_flags;
1308 
1309 	imgp->auxargs = elf_auxargs;
1310 	imgp->interpreted = 0;
1311 	imgp->reloc_base = addr;
1312 	imgp->proc->p_osrel = osrel;
1313 	imgp->proc->p_fctl0 = fctl0;
1314 	imgp->proc->p_elf_machine = hdr->e_machine;
1315 	imgp->proc->p_elf_flags = hdr->e_flags;
1316 
1317 ret:
1318 	if (free_interp)
1319 		free(interp, M_TEMP);
1320 	return (error);
1321 }
1322 
1323 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1324 
1325 int
1326 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1327 {
1328 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1329 	Elf_Auxinfo *argarray, *pos;
1330 	int error;
1331 
1332 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1333 	    M_WAITOK | M_ZERO);
1334 
1335 	if (args->execfd != -1)
1336 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1337 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1338 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1339 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1340 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1341 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1342 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1343 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1344 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1345 	if (imgp->execpathp != 0)
1346 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1347 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1348 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1349 	if (imgp->canary != 0) {
1350 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1351 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1352 	}
1353 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1354 	if (imgp->pagesizes != 0) {
1355 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1356 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1357 	}
1358 	if (imgp->sysent->sv_timekeep_base != 0) {
1359 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1360 		    imgp->sysent->sv_timekeep_base);
1361 	}
1362 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1363 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1364 	    imgp->sysent->sv_stackprot);
1365 	if (imgp->sysent->sv_hwcap != NULL)
1366 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1367 	if (imgp->sysent->sv_hwcap2 != NULL)
1368 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1369 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1370 
1371 	free(imgp->auxargs, M_TEMP);
1372 	imgp->auxargs = NULL;
1373 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1374 
1375 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1376 	free(argarray, M_TEMP);
1377 	return (error);
1378 }
1379 
1380 int
1381 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1382 {
1383 	Elf_Addr *base;
1384 
1385 	base = (Elf_Addr *)*stack_base;
1386 	base--;
1387 	if (suword(base, imgp->args->argc) == -1)
1388 		return (EFAULT);
1389 	*stack_base = (uintptr_t)base;
1390 	return (0);
1391 }
1392 
1393 /*
1394  * Code for generating ELF core dumps.
1395  */
1396 
1397 typedef void (*segment_callback)(vm_map_entry_t, void *);
1398 
1399 /* Closure for cb_put_phdr(). */
1400 struct phdr_closure {
1401 	Elf_Phdr *phdr;		/* Program header to fill in */
1402 	Elf_Off offset;		/* Offset of segment in core file */
1403 };
1404 
1405 /* Closure for cb_size_segment(). */
1406 struct sseg_closure {
1407 	int count;		/* Count of writable segments. */
1408 	size_t size;		/* Total size of all writable segments. */
1409 };
1410 
1411 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1412 
1413 struct note_info {
1414 	int		type;		/* Note type. */
1415 	outfunc_t 	outfunc; 	/* Output function. */
1416 	void		*outarg;	/* Argument for the output function. */
1417 	size_t		outsize;	/* Output size. */
1418 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1419 };
1420 
1421 TAILQ_HEAD(note_info_list, note_info);
1422 
1423 /* Coredump output parameters. */
1424 struct coredump_params {
1425 	off_t		offset;
1426 	struct ucred	*active_cred;
1427 	struct ucred	*file_cred;
1428 	struct thread	*td;
1429 	struct vnode	*vp;
1430 	struct compressor *comp;
1431 };
1432 
1433 extern int compress_user_cores;
1434 extern int compress_user_cores_level;
1435 
1436 static void cb_put_phdr(vm_map_entry_t, void *);
1437 static void cb_size_segment(vm_map_entry_t, void *);
1438 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1439     enum uio_seg);
1440 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1441 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1442     struct note_info_list *, size_t);
1443 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1444     size_t *);
1445 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1446 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1447 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1448 static int sbuf_drain_core_output(void *, const char *, int);
1449 
1450 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1451 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1452 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1453 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1454 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1455 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1456 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1457 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1458 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1459 static void note_procstat_files(void *, struct sbuf *, size_t *);
1460 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1461 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1462 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1463 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1464 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1465 
1466 /*
1467  * Write out a core segment to the compression stream.
1468  */
1469 static int
1470 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1471 {
1472 	u_int chunk_len;
1473 	int error;
1474 
1475 	while (len > 0) {
1476 		chunk_len = MIN(len, CORE_BUF_SIZE);
1477 
1478 		/*
1479 		 * We can get EFAULT error here.
1480 		 * In that case zero out the current chunk of the segment.
1481 		 */
1482 		error = copyin(base, buf, chunk_len);
1483 		if (error != 0)
1484 			bzero(buf, chunk_len);
1485 		error = compressor_write(p->comp, buf, chunk_len);
1486 		if (error != 0)
1487 			break;
1488 		base += chunk_len;
1489 		len -= chunk_len;
1490 	}
1491 	return (error);
1492 }
1493 
1494 static int
1495 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1496 {
1497 
1498 	return (core_write((struct coredump_params *)arg, base, len, offset,
1499 	    UIO_SYSSPACE));
1500 }
1501 
1502 static int
1503 core_write(struct coredump_params *p, const void *base, size_t len,
1504     off_t offset, enum uio_seg seg)
1505 {
1506 
1507 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1508 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1509 	    p->active_cred, p->file_cred, NULL, p->td));
1510 }
1511 
1512 static int
1513 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1514     void *tmpbuf)
1515 {
1516 	int error;
1517 
1518 	if (p->comp != NULL)
1519 		return (compress_chunk(p, base, tmpbuf, len));
1520 
1521 	/*
1522 	 * EFAULT is a non-fatal error that we can get, for example,
1523 	 * if the segment is backed by a file but extends beyond its
1524 	 * end.
1525 	 */
1526 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1527 	if (error == EFAULT) {
1528 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1529 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1530 		    "for process %s\n", base, len, offset, curproc->p_comm);
1531 
1532 		/*
1533 		 * Write a "real" zero byte at the end of the target region
1534 		 * in the case this is the last segment.
1535 		 * The intermediate space will be implicitly zero-filled.
1536 		 */
1537 		error = core_write(p, zero_region, 1, offset + len - 1,
1538 		    UIO_SYSSPACE);
1539 	}
1540 	return (error);
1541 }
1542 
1543 /*
1544  * Drain into a core file.
1545  */
1546 static int
1547 sbuf_drain_core_output(void *arg, const char *data, int len)
1548 {
1549 	struct coredump_params *p;
1550 	int error, locked;
1551 
1552 	p = (struct coredump_params *)arg;
1553 
1554 	/*
1555 	 * Some kern_proc out routines that print to this sbuf may
1556 	 * call us with the process lock held. Draining with the
1557 	 * non-sleepable lock held is unsafe. The lock is needed for
1558 	 * those routines when dumping a live process. In our case we
1559 	 * can safely release the lock before draining and acquire
1560 	 * again after.
1561 	 */
1562 	locked = PROC_LOCKED(p->td->td_proc);
1563 	if (locked)
1564 		PROC_UNLOCK(p->td->td_proc);
1565 	if (p->comp != NULL)
1566 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1567 	else
1568 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1569 		    UIO_SYSSPACE);
1570 	if (locked)
1571 		PROC_LOCK(p->td->td_proc);
1572 	if (error != 0)
1573 		return (-error);
1574 	p->offset += len;
1575 	return (len);
1576 }
1577 
1578 int
1579 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1580 {
1581 	struct ucred *cred = td->td_ucred;
1582 	int error = 0;
1583 	struct sseg_closure seginfo;
1584 	struct note_info_list notelst;
1585 	struct coredump_params params;
1586 	struct note_info *ninfo;
1587 	void *hdr, *tmpbuf;
1588 	size_t hdrsize, notesz, coresize;
1589 
1590 	hdr = NULL;
1591 	tmpbuf = NULL;
1592 	TAILQ_INIT(&notelst);
1593 
1594 	/* Size the program segments. */
1595 	seginfo.count = 0;
1596 	seginfo.size = 0;
1597 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1598 
1599 	/*
1600 	 * Collect info about the core file header area.
1601 	 */
1602 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1603 	if (seginfo.count + 1 >= PN_XNUM)
1604 		hdrsize += sizeof(Elf_Shdr);
1605 	__elfN(prepare_notes)(td, &notelst, &notesz);
1606 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1607 
1608 	/* Set up core dump parameters. */
1609 	params.offset = 0;
1610 	params.active_cred = cred;
1611 	params.file_cred = NOCRED;
1612 	params.td = td;
1613 	params.vp = vp;
1614 	params.comp = NULL;
1615 
1616 #ifdef RACCT
1617 	if (racct_enable) {
1618 		PROC_LOCK(td->td_proc);
1619 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1620 		PROC_UNLOCK(td->td_proc);
1621 		if (error != 0) {
1622 			error = EFAULT;
1623 			goto done;
1624 		}
1625 	}
1626 #endif
1627 	if (coresize >= limit) {
1628 		error = EFAULT;
1629 		goto done;
1630 	}
1631 
1632 	/* Create a compression stream if necessary. */
1633 	if (compress_user_cores != 0) {
1634 		params.comp = compressor_init(core_compressed_write,
1635 		    compress_user_cores, CORE_BUF_SIZE,
1636 		    compress_user_cores_level, &params);
1637 		if (params.comp == NULL) {
1638 			error = EFAULT;
1639 			goto done;
1640 		}
1641 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1642         }
1643 
1644 	/*
1645 	 * Allocate memory for building the header, fill it up,
1646 	 * and write it out following the notes.
1647 	 */
1648 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1649 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1650 	    notesz);
1651 
1652 	/* Write the contents of all of the writable segments. */
1653 	if (error == 0) {
1654 		Elf_Phdr *php;
1655 		off_t offset;
1656 		int i;
1657 
1658 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1659 		offset = round_page(hdrsize + notesz);
1660 		for (i = 0; i < seginfo.count; i++) {
1661 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1662 			    php->p_filesz, offset, &params, tmpbuf);
1663 			if (error != 0)
1664 				break;
1665 			offset += php->p_filesz;
1666 			php++;
1667 		}
1668 		if (error == 0 && params.comp != NULL)
1669 			error = compressor_flush(params.comp);
1670 	}
1671 	if (error) {
1672 		log(LOG_WARNING,
1673 		    "Failed to write core file for process %s (error %d)\n",
1674 		    curproc->p_comm, error);
1675 	}
1676 
1677 done:
1678 	free(tmpbuf, M_TEMP);
1679 	if (params.comp != NULL)
1680 		compressor_fini(params.comp);
1681 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1682 		TAILQ_REMOVE(&notelst, ninfo, link);
1683 		free(ninfo, M_TEMP);
1684 	}
1685 	if (hdr != NULL)
1686 		free(hdr, M_TEMP);
1687 
1688 	return (error);
1689 }
1690 
1691 /*
1692  * A callback for each_dumpable_segment() to write out the segment's
1693  * program header entry.
1694  */
1695 static void
1696 cb_put_phdr(vm_map_entry_t entry, void *closure)
1697 {
1698 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1699 	Elf_Phdr *phdr = phc->phdr;
1700 
1701 	phc->offset = round_page(phc->offset);
1702 
1703 	phdr->p_type = PT_LOAD;
1704 	phdr->p_offset = phc->offset;
1705 	phdr->p_vaddr = entry->start;
1706 	phdr->p_paddr = 0;
1707 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1708 	phdr->p_align = PAGE_SIZE;
1709 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1710 
1711 	phc->offset += phdr->p_filesz;
1712 	phc->phdr++;
1713 }
1714 
1715 /*
1716  * A callback for each_dumpable_segment() to gather information about
1717  * the number of segments and their total size.
1718  */
1719 static void
1720 cb_size_segment(vm_map_entry_t entry, void *closure)
1721 {
1722 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1723 
1724 	ssc->count++;
1725 	ssc->size += entry->end - entry->start;
1726 }
1727 
1728 /*
1729  * For each writable segment in the process's memory map, call the given
1730  * function with a pointer to the map entry and some arbitrary
1731  * caller-supplied data.
1732  */
1733 static void
1734 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1735 {
1736 	struct proc *p = td->td_proc;
1737 	vm_map_t map = &p->p_vmspace->vm_map;
1738 	vm_map_entry_t entry;
1739 	vm_object_t backing_object, object;
1740 	boolean_t ignore_entry;
1741 
1742 	vm_map_lock_read(map);
1743 	VM_MAP_ENTRY_FOREACH(entry, map) {
1744 		/*
1745 		 * Don't dump inaccessible mappings, deal with legacy
1746 		 * coredump mode.
1747 		 *
1748 		 * Note that read-only segments related to the elf binary
1749 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1750 		 * need to arbitrarily ignore such segments.
1751 		 */
1752 		if (elf_legacy_coredump) {
1753 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1754 				continue;
1755 		} else {
1756 			if ((entry->protection & VM_PROT_ALL) == 0)
1757 				continue;
1758 		}
1759 
1760 		/*
1761 		 * Dont include memory segment in the coredump if
1762 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1763 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1764 		 * kernel map).
1765 		 */
1766 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1767 			continue;
1768 
1769 		if ((object = entry->object.vm_object) == NULL)
1770 			continue;
1771 
1772 		/* Ignore memory-mapped devices and such things. */
1773 		VM_OBJECT_RLOCK(object);
1774 		while ((backing_object = object->backing_object) != NULL) {
1775 			VM_OBJECT_RLOCK(backing_object);
1776 			VM_OBJECT_RUNLOCK(object);
1777 			object = backing_object;
1778 		}
1779 		ignore_entry = object->type != OBJT_DEFAULT &&
1780 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1781 		    object->type != OBJT_PHYS;
1782 		VM_OBJECT_RUNLOCK(object);
1783 		if (ignore_entry)
1784 			continue;
1785 
1786 		(*func)(entry, closure);
1787 	}
1788 	vm_map_unlock_read(map);
1789 }
1790 
1791 /*
1792  * Write the core file header to the file, including padding up to
1793  * the page boundary.
1794  */
1795 static int
1796 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1797     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1798 {
1799 	struct note_info *ninfo;
1800 	struct sbuf *sb;
1801 	int error;
1802 
1803 	/* Fill in the header. */
1804 	bzero(hdr, hdrsize);
1805 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1806 
1807 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1808 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1809 	sbuf_start_section(sb, NULL);
1810 	sbuf_bcat(sb, hdr, hdrsize);
1811 	TAILQ_FOREACH(ninfo, notelst, link)
1812 	    __elfN(putnote)(ninfo, sb);
1813 	/* Align up to a page boundary for the program segments. */
1814 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1815 	error = sbuf_finish(sb);
1816 	sbuf_delete(sb);
1817 
1818 	return (error);
1819 }
1820 
1821 static void
1822 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1823     size_t *sizep)
1824 {
1825 	struct proc *p;
1826 	struct thread *thr;
1827 	size_t size;
1828 
1829 	p = td->td_proc;
1830 	size = 0;
1831 
1832 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1833 
1834 	/*
1835 	 * To have the debugger select the right thread (LWP) as the initial
1836 	 * thread, we dump the state of the thread passed to us in td first.
1837 	 * This is the thread that causes the core dump and thus likely to
1838 	 * be the right thread one wants to have selected in the debugger.
1839 	 */
1840 	thr = td;
1841 	while (thr != NULL) {
1842 		size += register_note(list, NT_PRSTATUS,
1843 		    __elfN(note_prstatus), thr);
1844 		size += register_note(list, NT_FPREGSET,
1845 		    __elfN(note_fpregset), thr);
1846 		size += register_note(list, NT_THRMISC,
1847 		    __elfN(note_thrmisc), thr);
1848 		size += register_note(list, NT_PTLWPINFO,
1849 		    __elfN(note_ptlwpinfo), thr);
1850 		size += register_note(list, -1,
1851 		    __elfN(note_threadmd), thr);
1852 
1853 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1854 		    TAILQ_NEXT(thr, td_plist);
1855 		if (thr == td)
1856 			thr = TAILQ_NEXT(thr, td_plist);
1857 	}
1858 
1859 	size += register_note(list, NT_PROCSTAT_PROC,
1860 	    __elfN(note_procstat_proc), p);
1861 	size += register_note(list, NT_PROCSTAT_FILES,
1862 	    note_procstat_files, p);
1863 	size += register_note(list, NT_PROCSTAT_VMMAP,
1864 	    note_procstat_vmmap, p);
1865 	size += register_note(list, NT_PROCSTAT_GROUPS,
1866 	    note_procstat_groups, p);
1867 	size += register_note(list, NT_PROCSTAT_UMASK,
1868 	    note_procstat_umask, p);
1869 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1870 	    note_procstat_rlimit, p);
1871 	size += register_note(list, NT_PROCSTAT_OSREL,
1872 	    note_procstat_osrel, p);
1873 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1874 	    __elfN(note_procstat_psstrings), p);
1875 	size += register_note(list, NT_PROCSTAT_AUXV,
1876 	    __elfN(note_procstat_auxv), p);
1877 
1878 	*sizep = size;
1879 }
1880 
1881 static void
1882 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1883     size_t notesz)
1884 {
1885 	Elf_Ehdr *ehdr;
1886 	Elf_Phdr *phdr;
1887 	Elf_Shdr *shdr;
1888 	struct phdr_closure phc;
1889 
1890 	ehdr = (Elf_Ehdr *)hdr;
1891 
1892 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1893 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1894 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1895 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1896 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1897 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1898 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1899 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1900 	ehdr->e_ident[EI_ABIVERSION] = 0;
1901 	ehdr->e_ident[EI_PAD] = 0;
1902 	ehdr->e_type = ET_CORE;
1903 	ehdr->e_machine = td->td_proc->p_elf_machine;
1904 	ehdr->e_version = EV_CURRENT;
1905 	ehdr->e_entry = 0;
1906 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1907 	ehdr->e_flags = td->td_proc->p_elf_flags;
1908 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1909 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1910 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1911 	ehdr->e_shstrndx = SHN_UNDEF;
1912 	if (numsegs + 1 < PN_XNUM) {
1913 		ehdr->e_phnum = numsegs + 1;
1914 		ehdr->e_shnum = 0;
1915 	} else {
1916 		ehdr->e_phnum = PN_XNUM;
1917 		ehdr->e_shnum = 1;
1918 
1919 		ehdr->e_shoff = ehdr->e_phoff +
1920 		    (numsegs + 1) * ehdr->e_phentsize;
1921 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1922 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1923 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1924 
1925 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1926 		memset(shdr, 0, sizeof(*shdr));
1927 		/*
1928 		 * A special first section is used to hold large segment and
1929 		 * section counts.  This was proposed by Sun Microsystems in
1930 		 * Solaris and has been adopted by Linux; the standard ELF
1931 		 * tools are already familiar with the technique.
1932 		 *
1933 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1934 		 * (or 12-7 depending on the version of the document) for more
1935 		 * details.
1936 		 */
1937 		shdr->sh_type = SHT_NULL;
1938 		shdr->sh_size = ehdr->e_shnum;
1939 		shdr->sh_link = ehdr->e_shstrndx;
1940 		shdr->sh_info = numsegs + 1;
1941 	}
1942 
1943 	/*
1944 	 * Fill in the program header entries.
1945 	 */
1946 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1947 
1948 	/* The note segement. */
1949 	phdr->p_type = PT_NOTE;
1950 	phdr->p_offset = hdrsize;
1951 	phdr->p_vaddr = 0;
1952 	phdr->p_paddr = 0;
1953 	phdr->p_filesz = notesz;
1954 	phdr->p_memsz = 0;
1955 	phdr->p_flags = PF_R;
1956 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1957 	phdr++;
1958 
1959 	/* All the writable segments from the program. */
1960 	phc.phdr = phdr;
1961 	phc.offset = round_page(hdrsize + notesz);
1962 	each_dumpable_segment(td, cb_put_phdr, &phc);
1963 }
1964 
1965 static size_t
1966 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1967 {
1968 	struct note_info *ninfo;
1969 	size_t size, notesize;
1970 
1971 	size = 0;
1972 	out(arg, NULL, &size);
1973 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1974 	ninfo->type = type;
1975 	ninfo->outfunc = out;
1976 	ninfo->outarg = arg;
1977 	ninfo->outsize = size;
1978 	TAILQ_INSERT_TAIL(list, ninfo, link);
1979 
1980 	if (type == -1)
1981 		return (size);
1982 
1983 	notesize = sizeof(Elf_Note) +		/* note header */
1984 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1985 						/* note name */
1986 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1987 
1988 	return (notesize);
1989 }
1990 
1991 static size_t
1992 append_note_data(const void *src, void *dst, size_t len)
1993 {
1994 	size_t padded_len;
1995 
1996 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1997 	if (dst != NULL) {
1998 		bcopy(src, dst, len);
1999 		bzero((char *)dst + len, padded_len - len);
2000 	}
2001 	return (padded_len);
2002 }
2003 
2004 size_t
2005 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2006 {
2007 	Elf_Note *note;
2008 	char *buf;
2009 	size_t notesize;
2010 
2011 	buf = dst;
2012 	if (buf != NULL) {
2013 		note = (Elf_Note *)buf;
2014 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2015 		note->n_descsz = size;
2016 		note->n_type = type;
2017 		buf += sizeof(*note);
2018 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2019 		    sizeof(FREEBSD_ABI_VENDOR));
2020 		append_note_data(src, buf, size);
2021 		if (descp != NULL)
2022 			*descp = buf;
2023 	}
2024 
2025 	notesize = sizeof(Elf_Note) +		/* note header */
2026 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2027 						/* note name */
2028 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2029 
2030 	return (notesize);
2031 }
2032 
2033 static void
2034 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2035 {
2036 	Elf_Note note;
2037 	ssize_t old_len, sect_len;
2038 	size_t new_len, descsz, i;
2039 
2040 	if (ninfo->type == -1) {
2041 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2042 		return;
2043 	}
2044 
2045 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2046 	note.n_descsz = ninfo->outsize;
2047 	note.n_type = ninfo->type;
2048 
2049 	sbuf_bcat(sb, &note, sizeof(note));
2050 	sbuf_start_section(sb, &old_len);
2051 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2052 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2053 	if (note.n_descsz == 0)
2054 		return;
2055 	sbuf_start_section(sb, &old_len);
2056 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2057 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2058 	if (sect_len < 0)
2059 		return;
2060 
2061 	new_len = (size_t)sect_len;
2062 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2063 	if (new_len < descsz) {
2064 		/*
2065 		 * It is expected that individual note emitters will correctly
2066 		 * predict their expected output size and fill up to that size
2067 		 * themselves, padding in a format-specific way if needed.
2068 		 * However, in case they don't, just do it here with zeros.
2069 		 */
2070 		for (i = 0; i < descsz - new_len; i++)
2071 			sbuf_putc(sb, 0);
2072 	} else if (new_len > descsz) {
2073 		/*
2074 		 * We can't always truncate sb -- we may have drained some
2075 		 * of it already.
2076 		 */
2077 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2078 		    "read it (%zu > %zu).  Since it is longer than "
2079 		    "expected, this coredump's notes are corrupt.  THIS "
2080 		    "IS A BUG in the note_procstat routine for type %u.\n",
2081 		    __func__, (unsigned)note.n_type, new_len, descsz,
2082 		    (unsigned)note.n_type));
2083 	}
2084 }
2085 
2086 /*
2087  * Miscellaneous note out functions.
2088  */
2089 
2090 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2091 #include <compat/freebsd32/freebsd32.h>
2092 #include <compat/freebsd32/freebsd32_signal.h>
2093 
2094 typedef struct prstatus32 elf_prstatus_t;
2095 typedef struct prpsinfo32 elf_prpsinfo_t;
2096 typedef struct fpreg32 elf_prfpregset_t;
2097 typedef struct fpreg32 elf_fpregset_t;
2098 typedef struct reg32 elf_gregset_t;
2099 typedef struct thrmisc32 elf_thrmisc_t;
2100 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2101 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2102 typedef uint32_t elf_ps_strings_t;
2103 #else
2104 typedef prstatus_t elf_prstatus_t;
2105 typedef prpsinfo_t elf_prpsinfo_t;
2106 typedef prfpregset_t elf_prfpregset_t;
2107 typedef prfpregset_t elf_fpregset_t;
2108 typedef gregset_t elf_gregset_t;
2109 typedef thrmisc_t elf_thrmisc_t;
2110 #define ELF_KERN_PROC_MASK	0
2111 typedef struct kinfo_proc elf_kinfo_proc_t;
2112 typedef vm_offset_t elf_ps_strings_t;
2113 #endif
2114 
2115 static void
2116 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2117 {
2118 	struct sbuf sbarg;
2119 	size_t len;
2120 	char *cp, *end;
2121 	struct proc *p;
2122 	elf_prpsinfo_t *psinfo;
2123 	int error;
2124 
2125 	p = (struct proc *)arg;
2126 	if (sb != NULL) {
2127 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2128 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2129 		psinfo->pr_version = PRPSINFO_VERSION;
2130 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2131 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2132 		PROC_LOCK(p);
2133 		if (p->p_args != NULL) {
2134 			len = sizeof(psinfo->pr_psargs) - 1;
2135 			if (len > p->p_args->ar_length)
2136 				len = p->p_args->ar_length;
2137 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2138 			PROC_UNLOCK(p);
2139 			error = 0;
2140 		} else {
2141 			_PHOLD(p);
2142 			PROC_UNLOCK(p);
2143 			sbuf_new(&sbarg, psinfo->pr_psargs,
2144 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2145 			error = proc_getargv(curthread, p, &sbarg);
2146 			PRELE(p);
2147 			if (sbuf_finish(&sbarg) == 0)
2148 				len = sbuf_len(&sbarg) - 1;
2149 			else
2150 				len = sizeof(psinfo->pr_psargs) - 1;
2151 			sbuf_delete(&sbarg);
2152 		}
2153 		if (error || len == 0)
2154 			strlcpy(psinfo->pr_psargs, p->p_comm,
2155 			    sizeof(psinfo->pr_psargs));
2156 		else {
2157 			KASSERT(len < sizeof(psinfo->pr_psargs),
2158 			    ("len is too long: %zu vs %zu", len,
2159 			    sizeof(psinfo->pr_psargs)));
2160 			cp = psinfo->pr_psargs;
2161 			end = cp + len - 1;
2162 			for (;;) {
2163 				cp = memchr(cp, '\0', end - cp);
2164 				if (cp == NULL)
2165 					break;
2166 				*cp = ' ';
2167 			}
2168 		}
2169 		psinfo->pr_pid = p->p_pid;
2170 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2171 		free(psinfo, M_TEMP);
2172 	}
2173 	*sizep = sizeof(*psinfo);
2174 }
2175 
2176 static void
2177 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2178 {
2179 	struct thread *td;
2180 	elf_prstatus_t *status;
2181 
2182 	td = (struct thread *)arg;
2183 	if (sb != NULL) {
2184 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2185 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2186 		status->pr_version = PRSTATUS_VERSION;
2187 		status->pr_statussz = sizeof(elf_prstatus_t);
2188 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2189 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2190 		status->pr_osreldate = osreldate;
2191 		status->pr_cursig = td->td_proc->p_sig;
2192 		status->pr_pid = td->td_tid;
2193 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2194 		fill_regs32(td, &status->pr_reg);
2195 #else
2196 		fill_regs(td, &status->pr_reg);
2197 #endif
2198 		sbuf_bcat(sb, status, sizeof(*status));
2199 		free(status, M_TEMP);
2200 	}
2201 	*sizep = sizeof(*status);
2202 }
2203 
2204 static void
2205 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2206 {
2207 	struct thread *td;
2208 	elf_prfpregset_t *fpregset;
2209 
2210 	td = (struct thread *)arg;
2211 	if (sb != NULL) {
2212 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2213 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2214 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2215 		fill_fpregs32(td, fpregset);
2216 #else
2217 		fill_fpregs(td, fpregset);
2218 #endif
2219 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2220 		free(fpregset, M_TEMP);
2221 	}
2222 	*sizep = sizeof(*fpregset);
2223 }
2224 
2225 static void
2226 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2227 {
2228 	struct thread *td;
2229 	elf_thrmisc_t thrmisc;
2230 
2231 	td = (struct thread *)arg;
2232 	if (sb != NULL) {
2233 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2234 		bzero(&thrmisc, sizeof(thrmisc));
2235 		strcpy(thrmisc.pr_tname, td->td_name);
2236 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2237 	}
2238 	*sizep = sizeof(thrmisc);
2239 }
2240 
2241 static void
2242 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2243 {
2244 	struct thread *td;
2245 	size_t size;
2246 	int structsize;
2247 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2248 	struct ptrace_lwpinfo32 pl;
2249 #else
2250 	struct ptrace_lwpinfo pl;
2251 #endif
2252 
2253 	td = (struct thread *)arg;
2254 	size = sizeof(structsize) + sizeof(pl);
2255 	if (sb != NULL) {
2256 		KASSERT(*sizep == size, ("invalid size"));
2257 		structsize = sizeof(pl);
2258 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2259 		bzero(&pl, sizeof(pl));
2260 		pl.pl_lwpid = td->td_tid;
2261 		pl.pl_event = PL_EVENT_NONE;
2262 		pl.pl_sigmask = td->td_sigmask;
2263 		pl.pl_siglist = td->td_siglist;
2264 		if (td->td_si.si_signo != 0) {
2265 			pl.pl_event = PL_EVENT_SIGNAL;
2266 			pl.pl_flags |= PL_FLAG_SI;
2267 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2268 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2269 #else
2270 			pl.pl_siginfo = td->td_si;
2271 #endif
2272 		}
2273 		strcpy(pl.pl_tdname, td->td_name);
2274 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2275 		sbuf_bcat(sb, &pl, sizeof(pl));
2276 	}
2277 	*sizep = size;
2278 }
2279 
2280 /*
2281  * Allow for MD specific notes, as well as any MD
2282  * specific preparations for writing MI notes.
2283  */
2284 static void
2285 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2286 {
2287 	struct thread *td;
2288 	void *buf;
2289 	size_t size;
2290 
2291 	td = (struct thread *)arg;
2292 	size = *sizep;
2293 	if (size != 0 && sb != NULL)
2294 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2295 	else
2296 		buf = NULL;
2297 	size = 0;
2298 	__elfN(dump_thread)(td, buf, &size);
2299 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2300 	if (size != 0 && sb != NULL)
2301 		sbuf_bcat(sb, buf, size);
2302 	free(buf, M_TEMP);
2303 	*sizep = size;
2304 }
2305 
2306 #ifdef KINFO_PROC_SIZE
2307 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2308 #endif
2309 
2310 static void
2311 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2312 {
2313 	struct proc *p;
2314 	size_t size;
2315 	int structsize;
2316 
2317 	p = (struct proc *)arg;
2318 	size = sizeof(structsize) + p->p_numthreads *
2319 	    sizeof(elf_kinfo_proc_t);
2320 
2321 	if (sb != NULL) {
2322 		KASSERT(*sizep == size, ("invalid size"));
2323 		structsize = sizeof(elf_kinfo_proc_t);
2324 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2325 		PROC_LOCK(p);
2326 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2327 	}
2328 	*sizep = size;
2329 }
2330 
2331 #ifdef KINFO_FILE_SIZE
2332 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2333 #endif
2334 
2335 static void
2336 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2337 {
2338 	struct proc *p;
2339 	size_t size, sect_sz, i;
2340 	ssize_t start_len, sect_len;
2341 	int structsize, filedesc_flags;
2342 
2343 	if (coredump_pack_fileinfo)
2344 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2345 	else
2346 		filedesc_flags = 0;
2347 
2348 	p = (struct proc *)arg;
2349 	structsize = sizeof(struct kinfo_file);
2350 	if (sb == NULL) {
2351 		size = 0;
2352 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2353 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2354 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2355 		PROC_LOCK(p);
2356 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2357 		sbuf_finish(sb);
2358 		sbuf_delete(sb);
2359 		*sizep = size;
2360 	} else {
2361 		sbuf_start_section(sb, &start_len);
2362 
2363 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2364 		PROC_LOCK(p);
2365 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2366 		    filedesc_flags);
2367 
2368 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2369 		if (sect_len < 0)
2370 			return;
2371 		sect_sz = sect_len;
2372 
2373 		KASSERT(sect_sz <= *sizep,
2374 		    ("kern_proc_filedesc_out did not respect maxlen; "
2375 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2376 		     sect_sz - sizeof(structsize)));
2377 
2378 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2379 			sbuf_putc(sb, 0);
2380 	}
2381 }
2382 
2383 #ifdef KINFO_VMENTRY_SIZE
2384 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2385 #endif
2386 
2387 static void
2388 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2389 {
2390 	struct proc *p;
2391 	size_t size;
2392 	int structsize, vmmap_flags;
2393 
2394 	if (coredump_pack_vmmapinfo)
2395 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2396 	else
2397 		vmmap_flags = 0;
2398 
2399 	p = (struct proc *)arg;
2400 	structsize = sizeof(struct kinfo_vmentry);
2401 	if (sb == NULL) {
2402 		size = 0;
2403 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2404 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2405 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2406 		PROC_LOCK(p);
2407 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2408 		sbuf_finish(sb);
2409 		sbuf_delete(sb);
2410 		*sizep = size;
2411 	} else {
2412 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2413 		PROC_LOCK(p);
2414 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2415 		    vmmap_flags);
2416 	}
2417 }
2418 
2419 static void
2420 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2421 {
2422 	struct proc *p;
2423 	size_t size;
2424 	int structsize;
2425 
2426 	p = (struct proc *)arg;
2427 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2428 	if (sb != NULL) {
2429 		KASSERT(*sizep == size, ("invalid size"));
2430 		structsize = sizeof(gid_t);
2431 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2432 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2433 		    sizeof(gid_t));
2434 	}
2435 	*sizep = size;
2436 }
2437 
2438 static void
2439 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2440 {
2441 	struct proc *p;
2442 	size_t size;
2443 	int structsize;
2444 
2445 	p = (struct proc *)arg;
2446 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2447 	if (sb != NULL) {
2448 		KASSERT(*sizep == size, ("invalid size"));
2449 		structsize = sizeof(p->p_fd->fd_cmask);
2450 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2451 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2452 	}
2453 	*sizep = size;
2454 }
2455 
2456 static void
2457 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2458 {
2459 	struct proc *p;
2460 	struct rlimit rlim[RLIM_NLIMITS];
2461 	size_t size;
2462 	int structsize, i;
2463 
2464 	p = (struct proc *)arg;
2465 	size = sizeof(structsize) + sizeof(rlim);
2466 	if (sb != NULL) {
2467 		KASSERT(*sizep == size, ("invalid size"));
2468 		structsize = sizeof(rlim);
2469 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2470 		PROC_LOCK(p);
2471 		for (i = 0; i < RLIM_NLIMITS; i++)
2472 			lim_rlimit_proc(p, i, &rlim[i]);
2473 		PROC_UNLOCK(p);
2474 		sbuf_bcat(sb, rlim, sizeof(rlim));
2475 	}
2476 	*sizep = size;
2477 }
2478 
2479 static void
2480 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2481 {
2482 	struct proc *p;
2483 	size_t size;
2484 	int structsize;
2485 
2486 	p = (struct proc *)arg;
2487 	size = sizeof(structsize) + sizeof(p->p_osrel);
2488 	if (sb != NULL) {
2489 		KASSERT(*sizep == size, ("invalid size"));
2490 		structsize = sizeof(p->p_osrel);
2491 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2492 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2493 	}
2494 	*sizep = size;
2495 }
2496 
2497 static void
2498 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2499 {
2500 	struct proc *p;
2501 	elf_ps_strings_t ps_strings;
2502 	size_t size;
2503 	int structsize;
2504 
2505 	p = (struct proc *)arg;
2506 	size = sizeof(structsize) + sizeof(ps_strings);
2507 	if (sb != NULL) {
2508 		KASSERT(*sizep == size, ("invalid size"));
2509 		structsize = sizeof(ps_strings);
2510 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2511 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2512 #else
2513 		ps_strings = p->p_sysent->sv_psstrings;
2514 #endif
2515 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2516 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2517 	}
2518 	*sizep = size;
2519 }
2520 
2521 static void
2522 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2523 {
2524 	struct proc *p;
2525 	size_t size;
2526 	int structsize;
2527 
2528 	p = (struct proc *)arg;
2529 	if (sb == NULL) {
2530 		size = 0;
2531 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2532 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2533 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2534 		PHOLD(p);
2535 		proc_getauxv(curthread, p, sb);
2536 		PRELE(p);
2537 		sbuf_finish(sb);
2538 		sbuf_delete(sb);
2539 		*sizep = size;
2540 	} else {
2541 		structsize = sizeof(Elf_Auxinfo);
2542 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2543 		PHOLD(p);
2544 		proc_getauxv(curthread, p, sb);
2545 		PRELE(p);
2546 	}
2547 }
2548 
2549 static boolean_t
2550 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2551     const char *note_vendor, const Elf_Phdr *pnote,
2552     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2553 {
2554 	const Elf_Note *note, *note0, *note_end;
2555 	const char *note_name;
2556 	char *buf;
2557 	int i, error;
2558 	boolean_t res;
2559 
2560 	/* We need some limit, might as well use PAGE_SIZE. */
2561 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2562 		return (FALSE);
2563 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2564 	if (pnote->p_offset > PAGE_SIZE ||
2565 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2566 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2567 		if (buf == NULL) {
2568 			VOP_UNLOCK(imgp->vp);
2569 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2570 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2571 		}
2572 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2573 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2574 		    curthread->td_ucred, NOCRED, NULL, curthread);
2575 		if (error != 0) {
2576 			uprintf("i/o error PT_NOTE\n");
2577 			goto retf;
2578 		}
2579 		note = note0 = (const Elf_Note *)buf;
2580 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2581 	} else {
2582 		note = note0 = (const Elf_Note *)(imgp->image_header +
2583 		    pnote->p_offset);
2584 		note_end = (const Elf_Note *)(imgp->image_header +
2585 		    pnote->p_offset + pnote->p_filesz);
2586 		buf = NULL;
2587 	}
2588 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2589 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2590 		    (const char *)note < sizeof(Elf_Note)) {
2591 			goto retf;
2592 		}
2593 		if (note->n_namesz != checknote->n_namesz ||
2594 		    note->n_descsz != checknote->n_descsz ||
2595 		    note->n_type != checknote->n_type)
2596 			goto nextnote;
2597 		note_name = (const char *)(note + 1);
2598 		if (note_name + checknote->n_namesz >=
2599 		    (const char *)note_end || strncmp(note_vendor,
2600 		    note_name, checknote->n_namesz) != 0)
2601 			goto nextnote;
2602 
2603 		if (cb(note, cb_arg, &res))
2604 			goto ret;
2605 nextnote:
2606 		note = (const Elf_Note *)((const char *)(note + 1) +
2607 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2608 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2609 	}
2610 retf:
2611 	res = FALSE;
2612 ret:
2613 	free(buf, M_TEMP);
2614 	return (res);
2615 }
2616 
2617 struct brandnote_cb_arg {
2618 	Elf_Brandnote *brandnote;
2619 	int32_t *osrel;
2620 };
2621 
2622 static boolean_t
2623 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2624 {
2625 	struct brandnote_cb_arg *arg;
2626 
2627 	arg = arg0;
2628 
2629 	/*
2630 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2631 	 * necessary.
2632 	 */
2633 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2634 	    arg->brandnote->trans_osrel != NULL ?
2635 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2636 
2637 	return (TRUE);
2638 }
2639 
2640 static Elf_Note fctl_note = {
2641 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2642 	.n_descsz = sizeof(uint32_t),
2643 	.n_type = NT_FREEBSD_FEATURE_CTL,
2644 };
2645 
2646 struct fctl_cb_arg {
2647 	uint32_t *fctl0;
2648 };
2649 
2650 static boolean_t
2651 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2652 {
2653 	struct fctl_cb_arg *arg;
2654 	const Elf32_Word *desc;
2655 	uintptr_t p;
2656 
2657 	arg = arg0;
2658 	p = (uintptr_t)(note + 1);
2659 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2660 	desc = (const Elf32_Word *)p;
2661 	*arg->fctl0 = desc[0];
2662 	return (TRUE);
2663 }
2664 
2665 /*
2666  * Try to find the appropriate ABI-note section for checknote, fetch
2667  * the osreldate and feature control flags for binary from the ELF
2668  * OSABI-note.  Only the first page of the image is searched, the same
2669  * as for headers.
2670  */
2671 static boolean_t
2672 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2673     int32_t *osrel, uint32_t *fctl0)
2674 {
2675 	const Elf_Phdr *phdr;
2676 	const Elf_Ehdr *hdr;
2677 	struct brandnote_cb_arg b_arg;
2678 	struct fctl_cb_arg f_arg;
2679 	int i, j;
2680 
2681 	hdr = (const Elf_Ehdr *)imgp->image_header;
2682 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2683 	b_arg.brandnote = brandnote;
2684 	b_arg.osrel = osrel;
2685 	f_arg.fctl0 = fctl0;
2686 
2687 	for (i = 0; i < hdr->e_phnum; i++) {
2688 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2689 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2690 		    &b_arg)) {
2691 			for (j = 0; j < hdr->e_phnum; j++) {
2692 				if (phdr[j].p_type == PT_NOTE &&
2693 				    __elfN(parse_notes)(imgp, &fctl_note,
2694 				    FREEBSD_ABI_VENDOR, &phdr[j],
2695 				    note_fctl_cb, &f_arg))
2696 					break;
2697 			}
2698 			return (TRUE);
2699 		}
2700 	}
2701 	return (FALSE);
2702 
2703 }
2704 
2705 /*
2706  * Tell kern_execve.c about it, with a little help from the linker.
2707  */
2708 static struct execsw __elfN(execsw) = {
2709 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2710 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2711 };
2712 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2713 
2714 static vm_prot_t
2715 __elfN(trans_prot)(Elf_Word flags)
2716 {
2717 	vm_prot_t prot;
2718 
2719 	prot = 0;
2720 	if (flags & PF_X)
2721 		prot |= VM_PROT_EXECUTE;
2722 	if (flags & PF_W)
2723 		prot |= VM_PROT_WRITE;
2724 	if (flags & PF_R)
2725 		prot |= VM_PROT_READ;
2726 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2727 	if (i386_read_exec && (flags & PF_R))
2728 		prot |= VM_PROT_EXECUTE;
2729 #endif
2730 	return (prot);
2731 }
2732 
2733 static Elf_Word
2734 __elfN(untrans_prot)(vm_prot_t prot)
2735 {
2736 	Elf_Word flags;
2737 
2738 	flags = 0;
2739 	if (prot & VM_PROT_EXECUTE)
2740 		flags |= PF_X;
2741 	if (prot & VM_PROT_READ)
2742 		flags |= PF_R;
2743 	if (prot & VM_PROT_WRITE)
2744 		flags |= PF_W;
2745 	return (flags);
2746 }
2747 
2748 void
2749 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2750 {
2751 	uintptr_t range, rbase, gap;
2752 	int pct;
2753 
2754 	if ((imgp->map_flags & MAP_ASLR) == 0)
2755 		return;
2756 	pct = __elfN(aslr_stack_gap);
2757 	if (pct == 0)
2758 		return;
2759 	if (pct > 50)
2760 		pct = 50;
2761 	range = imgp->eff_stack_sz * pct / 100;
2762 	arc4rand(&rbase, sizeof(rbase), 0);
2763 	gap = rbase % range;
2764 	gap &= ~(sizeof(u_long) - 1);
2765 	*stack_base -= gap;
2766 }
2767