xref: /freebsd/sys/kern/imgact_elf.c (revision 6e563a1b608438504d963c2d7c70e50d2e75af46)
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/proc.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
57 #include <sys/reg.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 bool __elfN(check_note)(struct image_params *imgp,
101     Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0,
102     uint32_t *fctl0);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 static size_t __elfN(prepare_register_notes)(struct thread *td,
106     struct note_info_list *list, struct thread *target_td);
107 
108 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
109     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
110     "");
111 
112 int __elfN(fallback_brand) = -1;
113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
114     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
115     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
116 
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119     &elf_legacy_coredump, 0,
120     "include all and only RW pages in core dumps");
121 
122 int __elfN(nxstack) =
123 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
124     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
125     defined(__riscv)
126 	1;
127 #else
128 	0;
129 #endif
130 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
131     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
132     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
133 
134 #if defined(__amd64__)
135 static int __elfN(vdso) = 1;
136 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
137     vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0,
138     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading");
139 #else
140 static int __elfN(vdso) = 0;
141 #endif
142 
143 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
144 int i386_read_exec = 0;
145 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
146     "enable execution from readable segments");
147 #endif
148 
149 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
150 static int
151 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
152 {
153 	u_long val;
154 	int error;
155 
156 	val = __elfN(pie_base);
157 	error = sysctl_handle_long(oidp, &val, 0, req);
158 	if (error != 0 || req->newptr == NULL)
159 		return (error);
160 	if ((val & PAGE_MASK) != 0)
161 		return (EINVAL);
162 	__elfN(pie_base) = val;
163 	return (0);
164 }
165 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
166     CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
167     sysctl_pie_base, "LU",
168     "PIE load base without randomization");
169 
170 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
171     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
172     "");
173 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
174 
175 /*
176  * While for 64-bit machines ASLR works properly, there are
177  * still some problems when using 32-bit architectures. For this
178  * reason ASLR is only enabled by default when running native
179  * 64-bit non-PIE executables.
180  */
181 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
182 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
183     &__elfN(aslr_enabled), 0,
184     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
185     ": enable address map randomization");
186 
187 /*
188  * Enable ASLR only for 64-bit PIE binaries by default.
189  */
190 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
191 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
192     &__elfN(pie_aslr_enabled), 0,
193     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
194     ": enable address map randomization for PIE binaries");
195 
196 /*
197  * Sbrk is now deprecated and it can be assumed, that in most
198  * cases it will not be used anyway. This setting is valid only
199  * for the ASLR enabled and allows for utilizing the bss grow region.
200  */
201 static int __elfN(aslr_honor_sbrk) = 0;
202 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
203     &__elfN(aslr_honor_sbrk), 0,
204     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
205 
206 static int __elfN(aslr_stack) = 1;
207 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
208     &__elfN(aslr_stack), 0,
209     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
210     ": enable stack address randomization");
211 
212 static int __elfN(sigfastblock) = 1;
213 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
214     CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
215     "enable sigfastblock for new processes");
216 
217 static bool __elfN(allow_wx) = true;
218 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
219     CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
220     "Allow pages to be mapped simultaneously writable and executable");
221 
222 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
223 
224 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
225 
226 Elf_Brandnote __elfN(freebsd_brandnote) = {
227 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
228 	.hdr.n_descsz	= sizeof(int32_t),
229 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
230 	.vendor		= FREEBSD_ABI_VENDOR,
231 	.flags		= BN_TRANSLATE_OSREL,
232 	.trans_osrel	= __elfN(freebsd_trans_osrel)
233 };
234 
235 static bool
236 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
237 {
238 	uintptr_t p;
239 
240 	p = (uintptr_t)(note + 1);
241 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
242 	*osrel = *(const int32_t *)(p);
243 
244 	return (true);
245 }
246 
247 static const char GNU_ABI_VENDOR[] = "GNU";
248 static int GNU_KFREEBSD_ABI_DESC = 3;
249 
250 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
251 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
252 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
253 	.hdr.n_type	= 1,
254 	.vendor		= GNU_ABI_VENDOR,
255 	.flags		= BN_TRANSLATE_OSREL,
256 	.trans_osrel	= kfreebsd_trans_osrel
257 };
258 
259 static bool
260 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
261 {
262 	const Elf32_Word *desc;
263 	uintptr_t p;
264 
265 	p = (uintptr_t)(note + 1);
266 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
267 
268 	desc = (const Elf32_Word *)p;
269 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
270 		return (false);
271 
272 	/*
273 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
274 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
275 	 */
276 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
277 
278 	return (true);
279 }
280 
281 int
282 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
283 {
284 	int i;
285 
286 	for (i = 0; i < MAX_BRANDS; i++) {
287 		if (elf_brand_list[i] == NULL) {
288 			elf_brand_list[i] = entry;
289 			break;
290 		}
291 	}
292 	if (i == MAX_BRANDS) {
293 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
294 			__func__, entry);
295 		return (-1);
296 	}
297 	return (0);
298 }
299 
300 int
301 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
302 {
303 	int i;
304 
305 	for (i = 0; i < MAX_BRANDS; i++) {
306 		if (elf_brand_list[i] == entry) {
307 			elf_brand_list[i] = NULL;
308 			break;
309 		}
310 	}
311 	if (i == MAX_BRANDS)
312 		return (-1);
313 	return (0);
314 }
315 
316 bool
317 __elfN(brand_inuse)(Elf_Brandinfo *entry)
318 {
319 	struct proc *p;
320 	bool rval = false;
321 
322 	sx_slock(&allproc_lock);
323 	FOREACH_PROC_IN_SYSTEM(p) {
324 		if (p->p_sysent == entry->sysvec) {
325 			rval = true;
326 			break;
327 		}
328 	}
329 	sx_sunlock(&allproc_lock);
330 
331 	return (rval);
332 }
333 
334 static Elf_Brandinfo *
335 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
336     int32_t *osrel, uint32_t *fctl0)
337 {
338 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
339 	Elf_Brandinfo *bi, *bi_m;
340 	bool ret, has_fctl0;
341 	int i, interp_name_len;
342 
343 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
344 
345 	/*
346 	 * We support four types of branding -- (1) the ELF EI_OSABI field
347 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
348 	 * branding w/in the ELF header, (3) path of the `interp_path'
349 	 * field, and (4) the ".note.ABI-tag" ELF section.
350 	 */
351 
352 	/* Look for an ".note.ABI-tag" ELF section */
353 	bi_m = NULL;
354 	for (i = 0; i < MAX_BRANDS; i++) {
355 		bi = elf_brand_list[i];
356 		if (bi == NULL)
357 			continue;
358 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
359 			continue;
360 		if (hdr->e_machine == bi->machine && (bi->flags &
361 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
362 			has_fctl0 = false;
363 			*fctl0 = 0;
364 			*osrel = 0;
365 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
366 			    &has_fctl0, fctl0);
367 			/* Give brand a chance to veto check_note's guess */
368 			if (ret && bi->header_supported) {
369 				ret = bi->header_supported(imgp, osrel,
370 				    has_fctl0 ? fctl0 : NULL);
371 			}
372 			/*
373 			 * If note checker claimed the binary, but the
374 			 * interpreter path in the image does not
375 			 * match default one for the brand, try to
376 			 * search for other brands with the same
377 			 * interpreter.  Either there is better brand
378 			 * with the right interpreter, or, failing
379 			 * this, we return first brand which accepted
380 			 * our note and, optionally, header.
381 			 */
382 			if (ret && bi_m == NULL && interp != NULL &&
383 			    (bi->interp_path == NULL ||
384 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
385 			    strncmp(interp, bi->interp_path, interp_name_len)
386 			    != 0))) {
387 				bi_m = bi;
388 				ret = 0;
389 			}
390 			if (ret)
391 				return (bi);
392 		}
393 	}
394 	if (bi_m != NULL)
395 		return (bi_m);
396 
397 	/* If the executable has a brand, search for it in the brand list. */
398 	for (i = 0; i < MAX_BRANDS; i++) {
399 		bi = elf_brand_list[i];
400 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
401 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
402 			continue;
403 		if (hdr->e_machine == bi->machine &&
404 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
405 		    (bi->compat_3_brand != NULL &&
406 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
407 		    bi->compat_3_brand) == 0))) {
408 			/* Looks good, but give brand a chance to veto */
409 			if (bi->header_supported == NULL ||
410 			    bi->header_supported(imgp, NULL, NULL)) {
411 				/*
412 				 * Again, prefer strictly matching
413 				 * interpreter path.
414 				 */
415 				if (interp_name_len == 0 &&
416 				    bi->interp_path == NULL)
417 					return (bi);
418 				if (bi->interp_path != NULL &&
419 				    strlen(bi->interp_path) + 1 ==
420 				    interp_name_len && strncmp(interp,
421 				    bi->interp_path, interp_name_len) == 0)
422 					return (bi);
423 				if (bi_m == NULL)
424 					bi_m = bi;
425 			}
426 		}
427 	}
428 	if (bi_m != NULL)
429 		return (bi_m);
430 
431 	/* No known brand, see if the header is recognized by any brand */
432 	for (i = 0; i < MAX_BRANDS; i++) {
433 		bi = elf_brand_list[i];
434 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
435 		    bi->header_supported == NULL)
436 			continue;
437 		if (hdr->e_machine == bi->machine) {
438 			ret = bi->header_supported(imgp, NULL, NULL);
439 			if (ret)
440 				return (bi);
441 		}
442 	}
443 
444 	/* Lacking a known brand, search for a recognized interpreter. */
445 	if (interp != NULL) {
446 		for (i = 0; i < MAX_BRANDS; i++) {
447 			bi = elf_brand_list[i];
448 			if (bi == NULL || (bi->flags &
449 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
450 			    != 0)
451 				continue;
452 			if (hdr->e_machine == bi->machine &&
453 			    bi->interp_path != NULL &&
454 			    /* ELF image p_filesz includes terminating zero */
455 			    strlen(bi->interp_path) + 1 == interp_name_len &&
456 			    strncmp(interp, bi->interp_path, interp_name_len)
457 			    == 0 && (bi->header_supported == NULL ||
458 			    bi->header_supported(imgp, NULL, NULL)))
459 				return (bi);
460 		}
461 	}
462 
463 	/* Lacking a recognized interpreter, try the default brand */
464 	for (i = 0; i < MAX_BRANDS; i++) {
465 		bi = elf_brand_list[i];
466 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
467 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
468 			continue;
469 		if (hdr->e_machine == bi->machine &&
470 		    __elfN(fallback_brand) == bi->brand &&
471 		    (bi->header_supported == NULL ||
472 		    bi->header_supported(imgp, NULL, NULL)))
473 			return (bi);
474 	}
475 	return (NULL);
476 }
477 
478 static bool
479 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
480 {
481 	return (hdr->e_phoff <= PAGE_SIZE &&
482 	    (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
483 }
484 
485 static int
486 __elfN(check_header)(const Elf_Ehdr *hdr)
487 {
488 	Elf_Brandinfo *bi;
489 	int i;
490 
491 	if (!IS_ELF(*hdr) ||
492 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
493 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
494 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
495 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
496 	    hdr->e_version != ELF_TARG_VER)
497 		return (ENOEXEC);
498 
499 	/*
500 	 * Make sure we have at least one brand for this machine.
501 	 */
502 
503 	for (i = 0; i < MAX_BRANDS; i++) {
504 		bi = elf_brand_list[i];
505 		if (bi != NULL && bi->machine == hdr->e_machine)
506 			break;
507 	}
508 	if (i == MAX_BRANDS)
509 		return (ENOEXEC);
510 
511 	return (0);
512 }
513 
514 static int
515 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
516     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
517 {
518 	struct sf_buf *sf;
519 	int error;
520 	vm_offset_t off;
521 
522 	/*
523 	 * Create the page if it doesn't exist yet. Ignore errors.
524 	 */
525 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
526 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
527 
528 	/*
529 	 * Find the page from the underlying object.
530 	 */
531 	if (object != NULL) {
532 		sf = vm_imgact_map_page(object, offset);
533 		if (sf == NULL)
534 			return (KERN_FAILURE);
535 		off = offset - trunc_page(offset);
536 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
537 		    end - start);
538 		vm_imgact_unmap_page(sf);
539 		if (error != 0)
540 			return (KERN_FAILURE);
541 	}
542 
543 	return (KERN_SUCCESS);
544 }
545 
546 static int
547 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
548     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
549     int cow)
550 {
551 	struct sf_buf *sf;
552 	vm_offset_t off;
553 	vm_size_t sz;
554 	int error, locked, rv;
555 
556 	if (start != trunc_page(start)) {
557 		rv = __elfN(map_partial)(map, object, offset, start,
558 		    round_page(start), prot);
559 		if (rv != KERN_SUCCESS)
560 			return (rv);
561 		offset += round_page(start) - start;
562 		start = round_page(start);
563 	}
564 	if (end != round_page(end)) {
565 		rv = __elfN(map_partial)(map, object, offset +
566 		    trunc_page(end) - start, trunc_page(end), end, prot);
567 		if (rv != KERN_SUCCESS)
568 			return (rv);
569 		end = trunc_page(end);
570 	}
571 	if (start >= end)
572 		return (KERN_SUCCESS);
573 	if ((offset & PAGE_MASK) != 0) {
574 		/*
575 		 * The mapping is not page aligned.  This means that we have
576 		 * to copy the data.
577 		 */
578 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
579 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
580 		if (rv != KERN_SUCCESS)
581 			return (rv);
582 		if (object == NULL)
583 			return (KERN_SUCCESS);
584 		for (; start < end; start += sz) {
585 			sf = vm_imgact_map_page(object, offset);
586 			if (sf == NULL)
587 				return (KERN_FAILURE);
588 			off = offset - trunc_page(offset);
589 			sz = end - start;
590 			if (sz > PAGE_SIZE - off)
591 				sz = PAGE_SIZE - off;
592 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
593 			    (caddr_t)start, sz);
594 			vm_imgact_unmap_page(sf);
595 			if (error != 0)
596 				return (KERN_FAILURE);
597 			offset += sz;
598 		}
599 	} else {
600 		vm_object_reference(object);
601 		rv = vm_map_fixed(map, object, offset, start, end - start,
602 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
603 		    (object != NULL ? MAP_VN_EXEC : 0));
604 		if (rv != KERN_SUCCESS) {
605 			locked = VOP_ISLOCKED(imgp->vp);
606 			VOP_UNLOCK(imgp->vp);
607 			vm_object_deallocate(object);
608 			vn_lock(imgp->vp, locked | LK_RETRY);
609 			return (rv);
610 		} else if (object != NULL) {
611 			MPASS(imgp->vp->v_object == object);
612 			VOP_SET_TEXT_CHECKED(imgp->vp);
613 		}
614 	}
615 	return (KERN_SUCCESS);
616 }
617 
618 static int
619 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
620     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
621 {
622 	struct sf_buf *sf;
623 	size_t map_len;
624 	vm_map_t map;
625 	vm_object_t object;
626 	vm_offset_t map_addr;
627 	int error, rv, cow;
628 	size_t copy_len;
629 	vm_ooffset_t file_addr;
630 
631 	/*
632 	 * It's necessary to fail if the filsz + offset taken from the
633 	 * header is greater than the actual file pager object's size.
634 	 * If we were to allow this, then the vm_map_find() below would
635 	 * walk right off the end of the file object and into the ether.
636 	 *
637 	 * While I'm here, might as well check for something else that
638 	 * is invalid: filsz cannot be greater than memsz.
639 	 */
640 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
641 	    filsz > memsz) {
642 		uprintf("elf_load_section: truncated ELF file\n");
643 		return (ENOEXEC);
644 	}
645 
646 	object = imgp->object;
647 	map = &imgp->proc->p_vmspace->vm_map;
648 	map_addr = trunc_page((vm_offset_t)vmaddr);
649 	file_addr = trunc_page(offset);
650 
651 	/*
652 	 * We have two choices.  We can either clear the data in the last page
653 	 * of an oversized mapping, or we can start the anon mapping a page
654 	 * early and copy the initialized data into that first page.  We
655 	 * choose the second.
656 	 */
657 	if (filsz == 0)
658 		map_len = 0;
659 	else if (memsz > filsz)
660 		map_len = trunc_page(offset + filsz) - file_addr;
661 	else
662 		map_len = round_page(offset + filsz) - file_addr;
663 
664 	if (map_len != 0) {
665 		/* cow flags: don't dump readonly sections in core */
666 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
667 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
668 
669 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
670 		    map_addr, map_addr + map_len, prot, cow);
671 		if (rv != KERN_SUCCESS)
672 			return (EINVAL);
673 
674 		/* we can stop now if we've covered it all */
675 		if (memsz == filsz)
676 			return (0);
677 	}
678 
679 	/*
680 	 * We have to get the remaining bit of the file into the first part
681 	 * of the oversized map segment.  This is normally because the .data
682 	 * segment in the file is extended to provide bss.  It's a neat idea
683 	 * to try and save a page, but it's a pain in the behind to implement.
684 	 */
685 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
686 	    filsz);
687 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
688 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
689 
690 	/* This had damn well better be true! */
691 	if (map_len != 0) {
692 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
693 		    map_addr + map_len, prot, 0);
694 		if (rv != KERN_SUCCESS)
695 			return (EINVAL);
696 	}
697 
698 	if (copy_len != 0) {
699 		sf = vm_imgact_map_page(object, offset + filsz);
700 		if (sf == NULL)
701 			return (EIO);
702 
703 		/* send the page fragment to user space */
704 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
705 		    copy_len);
706 		vm_imgact_unmap_page(sf);
707 		if (error != 0)
708 			return (error);
709 	}
710 
711 	/*
712 	 * Remove write access to the page if it was only granted by map_insert
713 	 * to allow copyout.
714 	 */
715 	if ((prot & VM_PROT_WRITE) == 0)
716 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
717 		    map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
718 
719 	return (0);
720 }
721 
722 static int
723 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
724     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
725 {
726 	vm_prot_t prot;
727 	u_long base_addr;
728 	bool first;
729 	int error, i;
730 
731 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
732 
733 	base_addr = 0;
734 	first = true;
735 
736 	for (i = 0; i < hdr->e_phnum; i++) {
737 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
738 			continue;
739 
740 		/* Loadable segment */
741 		prot = __elfN(trans_prot)(phdr[i].p_flags);
742 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
743 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
744 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
745 		if (error != 0)
746 			return (error);
747 
748 		/*
749 		 * Establish the base address if this is the first segment.
750 		 */
751 		if (first) {
752   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
753 			first = false;
754 		}
755 	}
756 
757 	if (base_addrp != NULL)
758 		*base_addrp = base_addr;
759 
760 	return (0);
761 }
762 
763 /*
764  * Load the file "file" into memory.  It may be either a shared object
765  * or an executable.
766  *
767  * The "addr" reference parameter is in/out.  On entry, it specifies
768  * the address where a shared object should be loaded.  If the file is
769  * an executable, this value is ignored.  On exit, "addr" specifies
770  * where the file was actually loaded.
771  *
772  * The "entry" reference parameter is out only.  On exit, it specifies
773  * the entry point for the loaded file.
774  */
775 static int
776 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
777 	u_long *entry)
778 {
779 	struct {
780 		struct nameidata nd;
781 		struct vattr attr;
782 		struct image_params image_params;
783 	} *tempdata;
784 	const Elf_Ehdr *hdr = NULL;
785 	const Elf_Phdr *phdr = NULL;
786 	struct nameidata *nd;
787 	struct vattr *attr;
788 	struct image_params *imgp;
789 	u_long rbase;
790 	u_long base_addr = 0;
791 	int error;
792 
793 #ifdef CAPABILITY_MODE
794 	/*
795 	 * XXXJA: This check can go away once we are sufficiently confident
796 	 * that the checks in namei() are correct.
797 	 */
798 	if (IN_CAPABILITY_MODE(curthread))
799 		return (ECAPMODE);
800 #endif
801 
802 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
803 	nd = &tempdata->nd;
804 	attr = &tempdata->attr;
805 	imgp = &tempdata->image_params;
806 
807 	/*
808 	 * Initialize part of the common data
809 	 */
810 	imgp->proc = p;
811 	imgp->attr = attr;
812 
813 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
814 	    UIO_SYSSPACE, file);
815 	if ((error = namei(nd)) != 0) {
816 		nd->ni_vp = NULL;
817 		goto fail;
818 	}
819 	NDFREE_PNBUF(nd);
820 	imgp->vp = nd->ni_vp;
821 
822 	/*
823 	 * Check permissions, modes, uid, etc on the file, and "open" it.
824 	 */
825 	error = exec_check_permissions(imgp);
826 	if (error)
827 		goto fail;
828 
829 	error = exec_map_first_page(imgp);
830 	if (error)
831 		goto fail;
832 
833 	imgp->object = nd->ni_vp->v_object;
834 
835 	hdr = (const Elf_Ehdr *)imgp->image_header;
836 	if ((error = __elfN(check_header)(hdr)) != 0)
837 		goto fail;
838 	if (hdr->e_type == ET_DYN)
839 		rbase = *addr;
840 	else if (hdr->e_type == ET_EXEC)
841 		rbase = 0;
842 	else {
843 		error = ENOEXEC;
844 		goto fail;
845 	}
846 
847 	/* Only support headers that fit within first page for now      */
848 	if (!__elfN(phdr_in_zero_page)(hdr)) {
849 		error = ENOEXEC;
850 		goto fail;
851 	}
852 
853 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
854 	if (!aligned(phdr, Elf_Addr)) {
855 		error = ENOEXEC;
856 		goto fail;
857 	}
858 
859 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
860 	if (error != 0)
861 		goto fail;
862 
863 	*addr = base_addr;
864 	*entry = (unsigned long)hdr->e_entry + rbase;
865 
866 fail:
867 	if (imgp->firstpage)
868 		exec_unmap_first_page(imgp);
869 
870 	if (nd->ni_vp) {
871 		if (imgp->textset)
872 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
873 		vput(nd->ni_vp);
874 	}
875 	free(tempdata, M_TEMP);
876 
877 	return (error);
878 }
879 
880 /*
881  * Select randomized valid address in the map map, between minv and
882  * maxv, with specified alignment.  The [minv, maxv) range must belong
883  * to the map.  Note that function only allocates the address, it is
884  * up to caller to clamp maxv in a way that the final allocation
885  * length fit into the map.
886  *
887  * Result is returned in *resp, error code indicates that arguments
888  * did not pass sanity checks for overflow and range correctness.
889  */
890 static int
891 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
892     u_int align, u_long *resp)
893 {
894 	u_long rbase, res;
895 
896 	MPASS(vm_map_min(map) <= minv);
897 
898 	if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
899 		uprintf("Invalid ELF segments layout\n");
900 		return (ENOEXEC);
901 	}
902 
903 	arc4rand(&rbase, sizeof(rbase), 0);
904 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
905 	res &= ~((u_long)align - 1);
906 	if (res >= maxv)
907 		res -= align;
908 
909 	KASSERT(res >= minv,
910 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
911 	    res, minv, maxv, rbase));
912 	KASSERT(res < maxv,
913 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
914 	    res, maxv, minv, rbase));
915 
916 	*resp = res;
917 	return (0);
918 }
919 
920 static int
921 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
922     const Elf_Phdr *phdr, u_long et_dyn_addr)
923 {
924 	struct vmspace *vmspace;
925 	const char *err_str;
926 	u_long text_size, data_size, total_size, text_addr, data_addr;
927 	u_long seg_size, seg_addr;
928 	int i;
929 
930 	err_str = NULL;
931 	text_size = data_size = total_size = text_addr = data_addr = 0;
932 
933 	for (i = 0; i < hdr->e_phnum; i++) {
934 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
935 			continue;
936 
937 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
938 		seg_size = round_page(phdr[i].p_memsz +
939 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
940 
941 		/*
942 		 * Make the largest executable segment the official
943 		 * text segment and all others data.
944 		 *
945 		 * Note that obreak() assumes that data_addr + data_size == end
946 		 * of data load area, and the ELF file format expects segments
947 		 * to be sorted by address.  If multiple data segments exist,
948 		 * the last one will be used.
949 		 */
950 
951 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
952 			text_size = seg_size;
953 			text_addr = seg_addr;
954 		} else {
955 			data_size = seg_size;
956 			data_addr = seg_addr;
957 		}
958 		total_size += seg_size;
959 	}
960 
961 	if (data_addr == 0 && data_size == 0) {
962 		data_addr = text_addr;
963 		data_size = text_size;
964 	}
965 
966 	/*
967 	 * Check limits.  It should be safe to check the
968 	 * limits after loading the segments since we do
969 	 * not actually fault in all the segments pages.
970 	 */
971 	PROC_LOCK(imgp->proc);
972 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
973 		err_str = "Data segment size exceeds process limit";
974 	else if (text_size > maxtsiz)
975 		err_str = "Text segment size exceeds system limit";
976 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
977 		err_str = "Total segment size exceeds process limit";
978 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
979 		err_str = "Data segment size exceeds resource limit";
980 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
981 		err_str = "Total segment size exceeds resource limit";
982 	PROC_UNLOCK(imgp->proc);
983 	if (err_str != NULL) {
984 		uprintf("%s\n", err_str);
985 		return (ENOMEM);
986 	}
987 
988 	vmspace = imgp->proc->p_vmspace;
989 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
990 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
991 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
992 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
993 
994 	return (0);
995 }
996 
997 static int
998 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
999     char **interpp, bool *free_interpp)
1000 {
1001 	struct thread *td;
1002 	char *interp;
1003 	int error, interp_name_len;
1004 
1005 	KASSERT(phdr->p_type == PT_INTERP,
1006 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
1007 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
1008 
1009 	td = curthread;
1010 
1011 	/* Path to interpreter */
1012 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
1013 		uprintf("Invalid PT_INTERP\n");
1014 		return (ENOEXEC);
1015 	}
1016 
1017 	interp_name_len = phdr->p_filesz;
1018 	if (phdr->p_offset > PAGE_SIZE ||
1019 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
1020 		/*
1021 		 * The vnode lock might be needed by the pagedaemon to
1022 		 * clean pages owned by the vnode.  Do not allow sleep
1023 		 * waiting for memory with the vnode locked, instead
1024 		 * try non-sleepable allocation first, and if it
1025 		 * fails, go to the slow path were we drop the lock
1026 		 * and do M_WAITOK.  A text reference prevents
1027 		 * modifications to the vnode content.
1028 		 */
1029 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1030 		if (interp == NULL) {
1031 			VOP_UNLOCK(imgp->vp);
1032 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1033 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1034 		}
1035 
1036 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
1037 		    interp_name_len, phdr->p_offset,
1038 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1039 		    NOCRED, NULL, td);
1040 		if (error != 0) {
1041 			free(interp, M_TEMP);
1042 			uprintf("i/o error PT_INTERP %d\n", error);
1043 			return (error);
1044 		}
1045 		interp[interp_name_len] = '\0';
1046 
1047 		*interpp = interp;
1048 		*free_interpp = true;
1049 		return (0);
1050 	}
1051 
1052 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1053 	if (interp[interp_name_len - 1] != '\0') {
1054 		uprintf("Invalid PT_INTERP\n");
1055 		return (ENOEXEC);
1056 	}
1057 
1058 	*interpp = interp;
1059 	*free_interpp = false;
1060 	return (0);
1061 }
1062 
1063 static int
1064 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1065     const char *interp, u_long *addr, u_long *entry)
1066 {
1067 	char *path;
1068 	int error;
1069 
1070 	if (brand_info->emul_path != NULL &&
1071 	    brand_info->emul_path[0] != '\0') {
1072 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1073 		snprintf(path, MAXPATHLEN, "%s%s",
1074 		    brand_info->emul_path, interp);
1075 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1076 		free(path, M_TEMP);
1077 		if (error == 0)
1078 			return (0);
1079 	}
1080 
1081 	if (brand_info->interp_newpath != NULL &&
1082 	    (brand_info->interp_path == NULL ||
1083 	    strcmp(interp, brand_info->interp_path) == 0)) {
1084 		error = __elfN(load_file)(imgp->proc,
1085 		    brand_info->interp_newpath, addr, entry);
1086 		if (error == 0)
1087 			return (0);
1088 	}
1089 
1090 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1091 	if (error == 0)
1092 		return (0);
1093 
1094 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1095 	return (error);
1096 }
1097 
1098 /*
1099  * Impossible et_dyn_addr initial value indicating that the real base
1100  * must be calculated later with some randomization applied.
1101  */
1102 #define	ET_DYN_ADDR_RAND	1
1103 
1104 static int
1105 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1106 {
1107 	struct thread *td;
1108 	const Elf_Ehdr *hdr;
1109 	const Elf_Phdr *phdr;
1110 	Elf_Auxargs *elf_auxargs;
1111 	struct vmspace *vmspace;
1112 	vm_map_t map;
1113 	char *interp;
1114 	Elf_Brandinfo *brand_info;
1115 	struct sysentvec *sv;
1116 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1117 	u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1118 	uint32_t fctl0;
1119 	int32_t osrel;
1120 	bool free_interp;
1121 	int error, i, n;
1122 
1123 	hdr = (const Elf_Ehdr *)imgp->image_header;
1124 
1125 	/*
1126 	 * Do we have a valid ELF header ?
1127 	 *
1128 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1129 	 * if particular brand doesn't support it.
1130 	 */
1131 	if (__elfN(check_header)(hdr) != 0 ||
1132 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1133 		return (-1);
1134 
1135 	/*
1136 	 * From here on down, we return an errno, not -1, as we've
1137 	 * detected an ELF file.
1138 	 */
1139 
1140 	if (!__elfN(phdr_in_zero_page)(hdr)) {
1141 		uprintf("Program headers not in the first page\n");
1142 		return (ENOEXEC);
1143 	}
1144 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1145 	if (!aligned(phdr, Elf_Addr)) {
1146 		uprintf("Unaligned program headers\n");
1147 		return (ENOEXEC);
1148 	}
1149 
1150 	n = error = 0;
1151 	baddr = 0;
1152 	osrel = 0;
1153 	fctl0 = 0;
1154 	entry = proghdr = 0;
1155 	interp = NULL;
1156 	free_interp = false;
1157 	td = curthread;
1158 
1159 	/*
1160 	 * Somewhat arbitrary, limit accepted max alignment for the
1161 	 * loadable segment to the max supported superpage size. Too
1162 	 * large alignment requests are not useful and are indicators
1163 	 * of corrupted or outright malicious binary.
1164 	 */
1165 	maxalign = PAGE_SIZE;
1166 	maxsalign = PAGE_SIZE * 1024;
1167 	for (i = MAXPAGESIZES - 1; i > 0; i--) {
1168 		if (pagesizes[i] > maxsalign)
1169 			maxsalign = pagesizes[i];
1170 	}
1171 
1172 	mapsz = 0;
1173 
1174 	for (i = 0; i < hdr->e_phnum; i++) {
1175 		switch (phdr[i].p_type) {
1176 		case PT_LOAD:
1177 			if (n == 0)
1178 				baddr = phdr[i].p_vaddr;
1179 			if (!powerof2(phdr[i].p_align) ||
1180 			    phdr[i].p_align > maxsalign) {
1181 				uprintf("Invalid segment alignment\n");
1182 				error = ENOEXEC;
1183 				goto ret;
1184 			}
1185 			if (phdr[i].p_align > maxalign)
1186 				maxalign = phdr[i].p_align;
1187 			if (mapsz + phdr[i].p_memsz < mapsz) {
1188 				uprintf("Mapsize overflow\n");
1189 				error = ENOEXEC;
1190 				goto ret;
1191 			}
1192 			mapsz += phdr[i].p_memsz;
1193 			n++;
1194 
1195 			/*
1196 			 * If this segment contains the program headers,
1197 			 * remember their virtual address for the AT_PHDR
1198 			 * aux entry. Static binaries don't usually include
1199 			 * a PT_PHDR entry.
1200 			 */
1201 			if (phdr[i].p_offset == 0 &&
1202 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1203 			    phdr[i].p_filesz)
1204 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1205 			break;
1206 		case PT_INTERP:
1207 			/* Path to interpreter */
1208 			if (interp != NULL) {
1209 				uprintf("Multiple PT_INTERP headers\n");
1210 				error = ENOEXEC;
1211 				goto ret;
1212 			}
1213 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1214 			    &free_interp);
1215 			if (error != 0)
1216 				goto ret;
1217 			break;
1218 		case PT_GNU_STACK:
1219 			if (__elfN(nxstack))
1220 				imgp->stack_prot =
1221 				    __elfN(trans_prot)(phdr[i].p_flags);
1222 			imgp->stack_sz = phdr[i].p_memsz;
1223 			break;
1224 		case PT_PHDR: 	/* Program header table info */
1225 			proghdr = phdr[i].p_vaddr;
1226 			break;
1227 		}
1228 	}
1229 
1230 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1231 	if (brand_info == NULL) {
1232 		uprintf("ELF binary type \"%u\" not known.\n",
1233 		    hdr->e_ident[EI_OSABI]);
1234 		error = ENOEXEC;
1235 		goto ret;
1236 	}
1237 	sv = brand_info->sysvec;
1238 	et_dyn_addr = 0;
1239 	if (hdr->e_type == ET_DYN) {
1240 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1241 			uprintf("Cannot execute shared object\n");
1242 			error = ENOEXEC;
1243 			goto ret;
1244 		}
1245 		/*
1246 		 * Honour the base load address from the dso if it is
1247 		 * non-zero for some reason.
1248 		 */
1249 		if (baddr == 0) {
1250 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1251 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1252 				et_dyn_addr = __elfN(pie_base);
1253 			else if ((__elfN(pie_aslr_enabled) &&
1254 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1255 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1256 				et_dyn_addr = ET_DYN_ADDR_RAND;
1257 			else
1258 				et_dyn_addr = __elfN(pie_base);
1259 		}
1260 	}
1261 
1262 	/*
1263 	 * Avoid a possible deadlock if the current address space is destroyed
1264 	 * and that address space maps the locked vnode.  In the common case,
1265 	 * the locked vnode's v_usecount is decremented but remains greater
1266 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1267 	 * However, in cases where the vnode lock is external, such as nullfs,
1268 	 * v_usecount may become zero.
1269 	 *
1270 	 * The VV_TEXT flag prevents modifications to the executable while
1271 	 * the vnode is unlocked.
1272 	 */
1273 	VOP_UNLOCK(imgp->vp);
1274 
1275 	/*
1276 	 * Decide whether to enable randomization of user mappings.
1277 	 * First, reset user preferences for the setid binaries.
1278 	 * Then, account for the support of the randomization by the
1279 	 * ABI, by user preferences, and make special treatment for
1280 	 * PIE binaries.
1281 	 */
1282 	if (imgp->credential_setid) {
1283 		PROC_LOCK(imgp->proc);
1284 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1285 		    P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1286 		PROC_UNLOCK(imgp->proc);
1287 	}
1288 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1289 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1290 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1291 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1292 		    ("et_dyn_addr == RAND and !ASLR"));
1293 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1294 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1295 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1296 		imgp->map_flags |= MAP_ASLR;
1297 		/*
1298 		 * If user does not care about sbrk, utilize the bss
1299 		 * grow region for mappings as well.  We can select
1300 		 * the base for the image anywere and still not suffer
1301 		 * from the fragmentation.
1302 		 */
1303 		if (!__elfN(aslr_honor_sbrk) ||
1304 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1305 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1306 		if (__elfN(aslr_stack))
1307 			imgp->map_flags |= MAP_ASLR_STACK;
1308 	}
1309 
1310 	if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1311 	    (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1312 	    (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1313 		imgp->map_flags |= MAP_WXORX;
1314 
1315 	error = exec_new_vmspace(imgp, sv);
1316 
1317 	imgp->proc->p_sysent = sv;
1318 	imgp->proc->p_elf_brandinfo = brand_info;
1319 
1320 	vmspace = imgp->proc->p_vmspace;
1321 	map = &vmspace->vm_map;
1322 	maxv = sv->sv_usrstack;
1323 	if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1324 		maxv -= lim_max(td, RLIMIT_STACK);
1325 	if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1326 		uprintf("Excessive mapping size\n");
1327 		error = ENOEXEC;
1328 	}
1329 
1330 	if (error == 0 && et_dyn_addr == ET_DYN_ADDR_RAND) {
1331 		KASSERT((map->flags & MAP_ASLR) != 0,
1332 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1333 		error = __CONCAT(rnd_, __elfN(base))(map,
1334 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1335 		    /* reserve half of the address space to interpreter */
1336 		    maxv / 2, maxalign, &et_dyn_addr);
1337 	}
1338 
1339 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1340 	if (error != 0)
1341 		goto ret;
1342 
1343 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1344 	if (error != 0)
1345 		goto ret;
1346 
1347 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1348 	if (error != 0)
1349 		goto ret;
1350 
1351 	/*
1352 	 * We load the dynamic linker where a userland call
1353 	 * to mmap(0, ...) would put it.  The rationale behind this
1354 	 * calculation is that it leaves room for the heap to grow to
1355 	 * its maximum allowed size.
1356 	 */
1357 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1358 	    RLIMIT_DATA));
1359 	if ((map->flags & MAP_ASLR) != 0) {
1360 		maxv1 = maxv / 2 + addr / 2;
1361 		error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1362 		    (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1363 		    pagesizes[1] : pagesizes[0], &anon_loc);
1364 		if (error != 0)
1365 			goto ret;
1366 		map->anon_loc = anon_loc;
1367 	} else {
1368 		map->anon_loc = addr;
1369 	}
1370 
1371 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1372 	imgp->entry_addr = entry;
1373 
1374 	if (interp != NULL) {
1375 		VOP_UNLOCK(imgp->vp);
1376 		if ((map->flags & MAP_ASLR) != 0) {
1377 			/* Assume that interpreter fits into 1/4 of AS */
1378 			maxv1 = maxv / 2 + addr / 2;
1379 			error = __CONCAT(rnd_, __elfN(base))(map, addr,
1380 			    maxv1, PAGE_SIZE, &addr);
1381 		}
1382 		if (error == 0) {
1383 			error = __elfN(load_interp)(imgp, brand_info, interp,
1384 			    &addr, &imgp->entry_addr);
1385 		}
1386 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1387 		if (error != 0)
1388 			goto ret;
1389 	} else
1390 		addr = et_dyn_addr;
1391 
1392 	error = exec_map_stack(imgp);
1393 	if (error != 0)
1394 		goto ret;
1395 
1396 	/*
1397 	 * Construct auxargs table (used by the copyout_auxargs routine)
1398 	 */
1399 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1400 	if (elf_auxargs == NULL) {
1401 		VOP_UNLOCK(imgp->vp);
1402 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1403 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1404 	}
1405 	elf_auxargs->execfd = -1;
1406 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1407 	elf_auxargs->phent = hdr->e_phentsize;
1408 	elf_auxargs->phnum = hdr->e_phnum;
1409 	elf_auxargs->pagesz = PAGE_SIZE;
1410 	elf_auxargs->base = addr;
1411 	elf_auxargs->flags = 0;
1412 	elf_auxargs->entry = entry;
1413 	elf_auxargs->hdr_eflags = hdr->e_flags;
1414 
1415 	imgp->auxargs = elf_auxargs;
1416 	imgp->interpreted = 0;
1417 	imgp->reloc_base = addr;
1418 	imgp->proc->p_osrel = osrel;
1419 	imgp->proc->p_fctl0 = fctl0;
1420 	imgp->proc->p_elf_flags = hdr->e_flags;
1421 
1422 ret:
1423 	ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1424 	if (free_interp)
1425 		free(interp, M_TEMP);
1426 	return (error);
1427 }
1428 
1429 #define	elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1430 
1431 int
1432 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1433 {
1434 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1435 	Elf_Auxinfo *argarray, *pos;
1436 	int error;
1437 
1438 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1439 	    M_WAITOK | M_ZERO);
1440 
1441 	if (args->execfd != -1)
1442 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1443 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1444 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1445 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1446 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1447 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1448 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1449 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1450 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1451 	if (imgp->execpathp != 0)
1452 		AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1453 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1454 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1455 	if (imgp->canary != 0) {
1456 		AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1457 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1458 	}
1459 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1460 	if (imgp->pagesizes != 0) {
1461 		AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1462 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1463 	}
1464 	if (imgp->sysent->sv_timekeep_base != 0) {
1465 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1466 		    imgp->sysent->sv_timekeep_base);
1467 	}
1468 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1469 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1470 	    imgp->sysent->sv_stackprot);
1471 	if (imgp->sysent->sv_hwcap != NULL)
1472 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1473 	if (imgp->sysent->sv_hwcap2 != NULL)
1474 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1475 	AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1476 	    ELF_BSDF_SIGFASTBLK : 0);
1477 	AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1478 	AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1479 	AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1480 	AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1481 	AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1482 	if (imgp->sysent->sv_fxrng_gen_base != 0)
1483 		AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1484 	if (imgp->sysent->sv_vdso_base != 0 && __elfN(vdso) != 0)
1485 		AUXARGS_ENTRY(pos, AT_KPRELOAD, imgp->sysent->sv_vdso_base);
1486 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1487 
1488 	free(imgp->auxargs, M_TEMP);
1489 	imgp->auxargs = NULL;
1490 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1491 
1492 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1493 	free(argarray, M_TEMP);
1494 	return (error);
1495 }
1496 
1497 int
1498 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1499 {
1500 	Elf_Addr *base;
1501 
1502 	base = (Elf_Addr *)*stack_base;
1503 	base--;
1504 	if (elf_suword(base, imgp->args->argc) == -1)
1505 		return (EFAULT);
1506 	*stack_base = (uintptr_t)base;
1507 	return (0);
1508 }
1509 
1510 /*
1511  * Code for generating ELF core dumps.
1512  */
1513 
1514 typedef void (*segment_callback)(vm_map_entry_t, void *);
1515 
1516 /* Closure for cb_put_phdr(). */
1517 struct phdr_closure {
1518 	Elf_Phdr *phdr;		/* Program header to fill in */
1519 	Elf_Off offset;		/* Offset of segment in core file */
1520 };
1521 
1522 struct note_info {
1523 	int		type;		/* Note type. */
1524 	struct regset	*regset;	/* Register set. */
1525 	outfunc_t 	outfunc; 	/* Output function. */
1526 	void		*outarg;	/* Argument for the output function. */
1527 	size_t		outsize;	/* Output size. */
1528 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1529 };
1530 
1531 TAILQ_HEAD(note_info_list, note_info);
1532 
1533 extern int compress_user_cores;
1534 extern int compress_user_cores_level;
1535 
1536 static void cb_put_phdr(vm_map_entry_t, void *);
1537 static void cb_size_segment(vm_map_entry_t, void *);
1538 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1539     int);
1540 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1541     struct note_info_list *, size_t, int);
1542 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1543 
1544 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1545 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1546 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1547 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1548 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1549 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1550 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1551 static void note_procstat_files(void *, struct sbuf *, size_t *);
1552 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1553 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1554 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1555 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1556 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1557 
1558 static int
1559 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1560 {
1561 
1562 	return (core_write((struct coredump_params *)arg, base, len, offset,
1563 	    UIO_SYSSPACE, NULL));
1564 }
1565 
1566 int
1567 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1568 {
1569 	struct ucred *cred = td->td_ucred;
1570 	int compm, error = 0;
1571 	struct sseg_closure seginfo;
1572 	struct note_info_list notelst;
1573 	struct coredump_params params;
1574 	struct note_info *ninfo;
1575 	void *hdr, *tmpbuf;
1576 	size_t hdrsize, notesz, coresize;
1577 
1578 	hdr = NULL;
1579 	tmpbuf = NULL;
1580 	TAILQ_INIT(&notelst);
1581 
1582 	/* Size the program segments. */
1583 	__elfN(size_segments)(td, &seginfo, flags);
1584 
1585 	/*
1586 	 * Collect info about the core file header area.
1587 	 */
1588 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1589 	if (seginfo.count + 1 >= PN_XNUM)
1590 		hdrsize += sizeof(Elf_Shdr);
1591 	td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, &notelst, &notesz);
1592 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1593 
1594 	/* Set up core dump parameters. */
1595 	params.offset = 0;
1596 	params.active_cred = cred;
1597 	params.file_cred = NOCRED;
1598 	params.td = td;
1599 	params.vp = vp;
1600 	params.comp = NULL;
1601 
1602 #ifdef RACCT
1603 	if (racct_enable) {
1604 		PROC_LOCK(td->td_proc);
1605 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1606 		PROC_UNLOCK(td->td_proc);
1607 		if (error != 0) {
1608 			error = EFAULT;
1609 			goto done;
1610 		}
1611 	}
1612 #endif
1613 	if (coresize >= limit) {
1614 		error = EFAULT;
1615 		goto done;
1616 	}
1617 
1618 	/* Create a compression stream if necessary. */
1619 	compm = compress_user_cores;
1620 	if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1621 	    compm == 0)
1622 		compm = COMPRESS_GZIP;
1623 	if (compm != 0) {
1624 		params.comp = compressor_init(core_compressed_write,
1625 		    compm, CORE_BUF_SIZE,
1626 		    compress_user_cores_level, &params);
1627 		if (params.comp == NULL) {
1628 			error = EFAULT;
1629 			goto done;
1630 		}
1631 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1632         }
1633 
1634 	/*
1635 	 * Allocate memory for building the header, fill it up,
1636 	 * and write it out following the notes.
1637 	 */
1638 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1639 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1640 	    notesz, flags);
1641 
1642 	/* Write the contents of all of the writable segments. */
1643 	if (error == 0) {
1644 		Elf_Phdr *php;
1645 		off_t offset;
1646 		int i;
1647 
1648 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1649 		offset = round_page(hdrsize + notesz);
1650 		for (i = 0; i < seginfo.count; i++) {
1651 			error = core_output((char *)(uintptr_t)php->p_vaddr,
1652 			    php->p_filesz, offset, &params, tmpbuf);
1653 			if (error != 0)
1654 				break;
1655 			offset += php->p_filesz;
1656 			php++;
1657 		}
1658 		if (error == 0 && params.comp != NULL)
1659 			error = compressor_flush(params.comp);
1660 	}
1661 	if (error) {
1662 		log(LOG_WARNING,
1663 		    "Failed to write core file for process %s (error %d)\n",
1664 		    curproc->p_comm, error);
1665 	}
1666 
1667 done:
1668 	free(tmpbuf, M_TEMP);
1669 	if (params.comp != NULL)
1670 		compressor_fini(params.comp);
1671 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1672 		TAILQ_REMOVE(&notelst, ninfo, link);
1673 		free(ninfo, M_TEMP);
1674 	}
1675 	if (hdr != NULL)
1676 		free(hdr, M_TEMP);
1677 
1678 	return (error);
1679 }
1680 
1681 /*
1682  * A callback for each_dumpable_segment() to write out the segment's
1683  * program header entry.
1684  */
1685 static void
1686 cb_put_phdr(vm_map_entry_t entry, void *closure)
1687 {
1688 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1689 	Elf_Phdr *phdr = phc->phdr;
1690 
1691 	phc->offset = round_page(phc->offset);
1692 
1693 	phdr->p_type = PT_LOAD;
1694 	phdr->p_offset = phc->offset;
1695 	phdr->p_vaddr = entry->start;
1696 	phdr->p_paddr = 0;
1697 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1698 	phdr->p_align = PAGE_SIZE;
1699 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1700 
1701 	phc->offset += phdr->p_filesz;
1702 	phc->phdr++;
1703 }
1704 
1705 /*
1706  * A callback for each_dumpable_segment() to gather information about
1707  * the number of segments and their total size.
1708  */
1709 static void
1710 cb_size_segment(vm_map_entry_t entry, void *closure)
1711 {
1712 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1713 
1714 	ssc->count++;
1715 	ssc->size += entry->end - entry->start;
1716 }
1717 
1718 void
1719 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1720     int flags)
1721 {
1722 	seginfo->count = 0;
1723 	seginfo->size = 0;
1724 
1725 	each_dumpable_segment(td, cb_size_segment, seginfo, flags);
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     int flags)
1736 {
1737 	struct proc *p = td->td_proc;
1738 	vm_map_t map = &p->p_vmspace->vm_map;
1739 	vm_map_entry_t entry;
1740 	vm_object_t backing_object, object;
1741 	bool ignore_entry;
1742 
1743 	vm_map_lock_read(map);
1744 	VM_MAP_ENTRY_FOREACH(entry, map) {
1745 		/*
1746 		 * Don't dump inaccessible mappings, deal with legacy
1747 		 * coredump mode.
1748 		 *
1749 		 * Note that read-only segments related to the elf binary
1750 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1751 		 * need to arbitrarily ignore such segments.
1752 		 */
1753 		if ((flags & SVC_ALL) == 0) {
1754 			if (elf_legacy_coredump) {
1755 				if ((entry->protection & VM_PROT_RW) !=
1756 				    VM_PROT_RW)
1757 					continue;
1758 			} else {
1759 				if ((entry->protection & VM_PROT_ALL) == 0)
1760 					continue;
1761 			}
1762 		}
1763 
1764 		/*
1765 		 * Dont include memory segment in the coredump if
1766 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1767 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1768 		 * kernel map).
1769 		 */
1770 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1771 			continue;
1772 		if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1773 		    (flags & SVC_ALL) == 0)
1774 			continue;
1775 		if ((object = entry->object.vm_object) == NULL)
1776 			continue;
1777 
1778 		/* Ignore memory-mapped devices and such things. */
1779 		VM_OBJECT_RLOCK(object);
1780 		while ((backing_object = object->backing_object) != NULL) {
1781 			VM_OBJECT_RLOCK(backing_object);
1782 			VM_OBJECT_RUNLOCK(object);
1783 			object = backing_object;
1784 		}
1785 		ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1786 		VM_OBJECT_RUNLOCK(object);
1787 		if (ignore_entry)
1788 			continue;
1789 
1790 		(*func)(entry, closure);
1791 	}
1792 	vm_map_unlock_read(map);
1793 }
1794 
1795 /*
1796  * Write the core file header to the file, including padding up to
1797  * the page boundary.
1798  */
1799 static int
1800 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1801     size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1802     int flags)
1803 {
1804 	struct note_info *ninfo;
1805 	struct sbuf *sb;
1806 	int error;
1807 
1808 	/* Fill in the header. */
1809 	bzero(hdr, hdrsize);
1810 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1811 
1812 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1813 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1814 	sbuf_start_section(sb, NULL);
1815 	sbuf_bcat(sb, hdr, hdrsize);
1816 	TAILQ_FOREACH(ninfo, notelst, link)
1817 	    __elfN(putnote)(p->td, ninfo, sb);
1818 	/* Align up to a page boundary for the program segments. */
1819 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1820 	error = sbuf_finish(sb);
1821 	sbuf_delete(sb);
1822 
1823 	return (error);
1824 }
1825 
1826 void
1827 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1828     size_t *sizep)
1829 {
1830 	struct proc *p;
1831 	struct thread *thr;
1832 	size_t size;
1833 
1834 	p = td->td_proc;
1835 	size = 0;
1836 
1837 	size += __elfN(register_note)(td, list, NT_PRPSINFO,
1838 	    __elfN(note_prpsinfo), p);
1839 
1840 	/*
1841 	 * To have the debugger select the right thread (LWP) as the initial
1842 	 * thread, we dump the state of the thread passed to us in td first.
1843 	 * This is the thread that causes the core dump and thus likely to
1844 	 * be the right thread one wants to have selected in the debugger.
1845 	 */
1846 	thr = td;
1847 	while (thr != NULL) {
1848 		size += __elfN(prepare_register_notes)(td, list, thr);
1849 		size += __elfN(register_note)(td, list, NT_THRMISC,
1850 		    __elfN(note_thrmisc), thr);
1851 		size += __elfN(register_note)(td, list, NT_PTLWPINFO,
1852 		    __elfN(note_ptlwpinfo), thr);
1853 		size += __elfN(register_note)(td, list, -1,
1854 		    __elfN(note_threadmd), thr);
1855 
1856 		thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1857 		    TAILQ_NEXT(thr, td_plist);
1858 		if (thr == td)
1859 			thr = TAILQ_NEXT(thr, td_plist);
1860 	}
1861 
1862 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1863 	    __elfN(note_procstat_proc), p);
1864 	size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1865 	    note_procstat_files, p);
1866 	size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1867 	    note_procstat_vmmap, p);
1868 	size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1869 	    note_procstat_groups, p);
1870 	size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1871 	    note_procstat_umask, p);
1872 	size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1873 	    note_procstat_rlimit, p);
1874 	size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1875 	    note_procstat_osrel, p);
1876 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1877 	    __elfN(note_procstat_psstrings), p);
1878 	size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1879 	    __elfN(note_procstat_auxv), p);
1880 
1881 	*sizep = size;
1882 }
1883 
1884 void
1885 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1886     size_t notesz, int flags)
1887 {
1888 	Elf_Ehdr *ehdr;
1889 	Elf_Phdr *phdr;
1890 	Elf_Shdr *shdr;
1891 	struct phdr_closure phc;
1892 	Elf_Brandinfo *bi;
1893 
1894 	ehdr = (Elf_Ehdr *)hdr;
1895 	bi = td->td_proc->p_elf_brandinfo;
1896 
1897 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1898 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1899 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1900 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1901 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1902 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1903 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1904 	ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1905 	ehdr->e_ident[EI_ABIVERSION] = 0;
1906 	ehdr->e_ident[EI_PAD] = 0;
1907 	ehdr->e_type = ET_CORE;
1908 	ehdr->e_machine = bi->machine;
1909 	ehdr->e_version = EV_CURRENT;
1910 	ehdr->e_entry = 0;
1911 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1912 	ehdr->e_flags = td->td_proc->p_elf_flags;
1913 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1914 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1915 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1916 	ehdr->e_shstrndx = SHN_UNDEF;
1917 	if (numsegs + 1 < PN_XNUM) {
1918 		ehdr->e_phnum = numsegs + 1;
1919 		ehdr->e_shnum = 0;
1920 	} else {
1921 		ehdr->e_phnum = PN_XNUM;
1922 		ehdr->e_shnum = 1;
1923 
1924 		ehdr->e_shoff = ehdr->e_phoff +
1925 		    (numsegs + 1) * ehdr->e_phentsize;
1926 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1927 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1928 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1929 
1930 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1931 		memset(shdr, 0, sizeof(*shdr));
1932 		/*
1933 		 * A special first section is used to hold large segment and
1934 		 * section counts.  This was proposed by Sun Microsystems in
1935 		 * Solaris and has been adopted by Linux; the standard ELF
1936 		 * tools are already familiar with the technique.
1937 		 *
1938 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1939 		 * (or 12-7 depending on the version of the document) for more
1940 		 * details.
1941 		 */
1942 		shdr->sh_type = SHT_NULL;
1943 		shdr->sh_size = ehdr->e_shnum;
1944 		shdr->sh_link = ehdr->e_shstrndx;
1945 		shdr->sh_info = numsegs + 1;
1946 	}
1947 
1948 	/*
1949 	 * Fill in the program header entries.
1950 	 */
1951 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1952 
1953 	/* The note segement. */
1954 	phdr->p_type = PT_NOTE;
1955 	phdr->p_offset = hdrsize;
1956 	phdr->p_vaddr = 0;
1957 	phdr->p_paddr = 0;
1958 	phdr->p_filesz = notesz;
1959 	phdr->p_memsz = 0;
1960 	phdr->p_flags = PF_R;
1961 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1962 	phdr++;
1963 
1964 	/* All the writable segments from the program. */
1965 	phc.phdr = phdr;
1966 	phc.offset = round_page(hdrsize + notesz);
1967 	each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1968 }
1969 
1970 static size_t
1971 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list,
1972     struct regset *regset, struct thread *target_td)
1973 {
1974 	const struct sysentvec *sv;
1975 	struct note_info *ninfo;
1976 	size_t size, notesize;
1977 
1978 	size = 0;
1979 	if (!regset->get(regset, target_td, NULL, &size) || size == 0)
1980 		return (0);
1981 
1982 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1983 	ninfo->type = regset->note;
1984 	ninfo->regset = regset;
1985 	ninfo->outarg = target_td;
1986 	ninfo->outsize = size;
1987 	TAILQ_INSERT_TAIL(list, ninfo, link);
1988 
1989 	sv = td->td_proc->p_sysent;
1990 	notesize = sizeof(Elf_Note) +		/* note header */
1991 	    roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1992 						/* note name */
1993 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1994 
1995 	return (notesize);
1996 }
1997 
1998 size_t
1999 __elfN(register_note)(struct thread *td, struct note_info_list *list,
2000     int type, outfunc_t out, void *arg)
2001 {
2002 	const struct sysentvec *sv;
2003 	struct note_info *ninfo;
2004 	size_t size, notesize;
2005 
2006 	sv = td->td_proc->p_sysent;
2007 	size = 0;
2008 	out(arg, NULL, &size);
2009 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2010 	ninfo->type = type;
2011 	ninfo->outfunc = out;
2012 	ninfo->outarg = arg;
2013 	ninfo->outsize = size;
2014 	TAILQ_INSERT_TAIL(list, ninfo, link);
2015 
2016 	if (type == -1)
2017 		return (size);
2018 
2019 	notesize = sizeof(Elf_Note) +		/* note header */
2020 	    roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2021 						/* note name */
2022 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2023 
2024 	return (notesize);
2025 }
2026 
2027 static size_t
2028 append_note_data(const void *src, void *dst, size_t len)
2029 {
2030 	size_t padded_len;
2031 
2032 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2033 	if (dst != NULL) {
2034 		bcopy(src, dst, len);
2035 		bzero((char *)dst + len, padded_len - len);
2036 	}
2037 	return (padded_len);
2038 }
2039 
2040 size_t
2041 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2042 {
2043 	Elf_Note *note;
2044 	char *buf;
2045 	size_t notesize;
2046 
2047 	buf = dst;
2048 	if (buf != NULL) {
2049 		note = (Elf_Note *)buf;
2050 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2051 		note->n_descsz = size;
2052 		note->n_type = type;
2053 		buf += sizeof(*note);
2054 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2055 		    sizeof(FREEBSD_ABI_VENDOR));
2056 		append_note_data(src, buf, size);
2057 		if (descp != NULL)
2058 			*descp = buf;
2059 	}
2060 
2061 	notesize = sizeof(Elf_Note) +		/* note header */
2062 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2063 						/* note name */
2064 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2065 
2066 	return (notesize);
2067 }
2068 
2069 static void
2070 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
2071 {
2072 	Elf_Note note;
2073 	const struct sysentvec *sv;
2074 	ssize_t old_len, sect_len;
2075 	size_t new_len, descsz, i;
2076 
2077 	if (ninfo->type == -1) {
2078 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2079 		return;
2080 	}
2081 
2082 	sv = td->td_proc->p_sysent;
2083 
2084 	note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2085 	note.n_descsz = ninfo->outsize;
2086 	note.n_type = ninfo->type;
2087 
2088 	sbuf_bcat(sb, &note, sizeof(note));
2089 	sbuf_start_section(sb, &old_len);
2090 	sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2091 	    strlen(sv->sv_elf_core_abi_vendor) + 1);
2092 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2093 	if (note.n_descsz == 0)
2094 		return;
2095 	sbuf_start_section(sb, &old_len);
2096 	if (ninfo->regset != NULL) {
2097 		struct regset *regset = ninfo->regset;
2098 		void *buf;
2099 
2100 		buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK);
2101 		(void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize);
2102 		sbuf_bcat(sb, buf, ninfo->outsize);
2103 		free(buf, M_TEMP);
2104 	} else
2105 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2106 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2107 	if (sect_len < 0)
2108 		return;
2109 
2110 	new_len = (size_t)sect_len;
2111 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2112 	if (new_len < descsz) {
2113 		/*
2114 		 * It is expected that individual note emitters will correctly
2115 		 * predict their expected output size and fill up to that size
2116 		 * themselves, padding in a format-specific way if needed.
2117 		 * However, in case they don't, just do it here with zeros.
2118 		 */
2119 		for (i = 0; i < descsz - new_len; i++)
2120 			sbuf_putc(sb, 0);
2121 	} else if (new_len > descsz) {
2122 		/*
2123 		 * We can't always truncate sb -- we may have drained some
2124 		 * of it already.
2125 		 */
2126 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2127 		    "read it (%zu > %zu).  Since it is longer than "
2128 		    "expected, this coredump's notes are corrupt.  THIS "
2129 		    "IS A BUG in the note_procstat routine for type %u.\n",
2130 		    __func__, (unsigned)note.n_type, new_len, descsz,
2131 		    (unsigned)note.n_type));
2132 	}
2133 }
2134 
2135 /*
2136  * Miscellaneous note out functions.
2137  */
2138 
2139 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2140 #include <compat/freebsd32/freebsd32.h>
2141 #include <compat/freebsd32/freebsd32_signal.h>
2142 
2143 typedef struct prstatus32 elf_prstatus_t;
2144 typedef struct prpsinfo32 elf_prpsinfo_t;
2145 typedef struct fpreg32 elf_prfpregset_t;
2146 typedef struct fpreg32 elf_fpregset_t;
2147 typedef struct reg32 elf_gregset_t;
2148 typedef struct thrmisc32 elf_thrmisc_t;
2149 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2150 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2151 typedef uint32_t elf_ps_strings_t;
2152 #else
2153 typedef prstatus_t elf_prstatus_t;
2154 typedef prpsinfo_t elf_prpsinfo_t;
2155 typedef prfpregset_t elf_prfpregset_t;
2156 typedef prfpregset_t elf_fpregset_t;
2157 typedef gregset_t elf_gregset_t;
2158 typedef thrmisc_t elf_thrmisc_t;
2159 #define ELF_KERN_PROC_MASK	0
2160 typedef struct kinfo_proc elf_kinfo_proc_t;
2161 typedef vm_offset_t elf_ps_strings_t;
2162 #endif
2163 
2164 static void
2165 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2166 {
2167 	struct sbuf sbarg;
2168 	size_t len;
2169 	char *cp, *end;
2170 	struct proc *p;
2171 	elf_prpsinfo_t *psinfo;
2172 	int error;
2173 
2174 	p = arg;
2175 	if (sb != NULL) {
2176 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2177 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2178 		psinfo->pr_version = PRPSINFO_VERSION;
2179 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2180 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2181 		PROC_LOCK(p);
2182 		if (p->p_args != NULL) {
2183 			len = sizeof(psinfo->pr_psargs) - 1;
2184 			if (len > p->p_args->ar_length)
2185 				len = p->p_args->ar_length;
2186 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2187 			PROC_UNLOCK(p);
2188 			error = 0;
2189 		} else {
2190 			_PHOLD(p);
2191 			PROC_UNLOCK(p);
2192 			sbuf_new(&sbarg, psinfo->pr_psargs,
2193 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2194 			error = proc_getargv(curthread, p, &sbarg);
2195 			PRELE(p);
2196 			if (sbuf_finish(&sbarg) == 0)
2197 				len = sbuf_len(&sbarg) - 1;
2198 			else
2199 				len = sizeof(psinfo->pr_psargs) - 1;
2200 			sbuf_delete(&sbarg);
2201 		}
2202 		if (error || len == 0)
2203 			strlcpy(psinfo->pr_psargs, p->p_comm,
2204 			    sizeof(psinfo->pr_psargs));
2205 		else {
2206 			KASSERT(len < sizeof(psinfo->pr_psargs),
2207 			    ("len is too long: %zu vs %zu", len,
2208 			    sizeof(psinfo->pr_psargs)));
2209 			cp = psinfo->pr_psargs;
2210 			end = cp + len - 1;
2211 			for (;;) {
2212 				cp = memchr(cp, '\0', end - cp);
2213 				if (cp == NULL)
2214 					break;
2215 				*cp = ' ';
2216 			}
2217 		}
2218 		psinfo->pr_pid = p->p_pid;
2219 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2220 		free(psinfo, M_TEMP);
2221 	}
2222 	*sizep = sizeof(*psinfo);
2223 }
2224 
2225 static bool
2226 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf,
2227     size_t *sizep)
2228 {
2229 	elf_prstatus_t *status;
2230 
2231 	if (buf != NULL) {
2232 		KASSERT(*sizep == sizeof(*status), ("%s: invalid size",
2233 		    __func__));
2234 		status = buf;
2235 		memset(status, 0, *sizep);
2236 		status->pr_version = PRSTATUS_VERSION;
2237 		status->pr_statussz = sizeof(elf_prstatus_t);
2238 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2239 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2240 		status->pr_osreldate = osreldate;
2241 		status->pr_cursig = td->td_proc->p_sig;
2242 		status->pr_pid = td->td_tid;
2243 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2244 		fill_regs32(td, &status->pr_reg);
2245 #else
2246 		fill_regs(td, &status->pr_reg);
2247 #endif
2248 	}
2249 	*sizep = sizeof(*status);
2250 	return (true);
2251 }
2252 
2253 static bool
2254 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf,
2255     size_t size)
2256 {
2257 	elf_prstatus_t *status;
2258 
2259 	KASSERT(size == sizeof(*status), ("%s: invalid size", __func__));
2260 	status = buf;
2261 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2262 	set_regs32(td, &status->pr_reg);
2263 #else
2264 	set_regs(td, &status->pr_reg);
2265 #endif
2266 	return (true);
2267 }
2268 
2269 static struct regset __elfN(regset_prstatus) = {
2270 	.note = NT_PRSTATUS,
2271 	.size = sizeof(elf_prstatus_t),
2272 	.get = __elfN(get_prstatus),
2273 	.set = __elfN(set_prstatus),
2274 };
2275 ELF_REGSET(__elfN(regset_prstatus));
2276 
2277 static bool
2278 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf,
2279     size_t *sizep)
2280 {
2281 	elf_prfpregset_t *fpregset;
2282 
2283 	if (buf != NULL) {
2284 		KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size",
2285 		    __func__));
2286 		fpregset = buf;
2287 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2288 		fill_fpregs32(td, fpregset);
2289 #else
2290 		fill_fpregs(td, fpregset);
2291 #endif
2292 	}
2293 	*sizep = sizeof(*fpregset);
2294 	return (true);
2295 }
2296 
2297 static bool
2298 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf,
2299     size_t size)
2300 {
2301 	elf_prfpregset_t *fpregset;
2302 
2303 	fpregset = buf;
2304 	KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__));
2305 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2306 	set_fpregs32(td, fpregset);
2307 #else
2308 	set_fpregs(td, fpregset);
2309 #endif
2310 	return (true);
2311 }
2312 
2313 static struct regset __elfN(regset_fpregset) = {
2314 	.note = NT_FPREGSET,
2315 	.size = sizeof(elf_prfpregset_t),
2316 	.get = __elfN(get_fpregset),
2317 	.set = __elfN(set_fpregset),
2318 };
2319 ELF_REGSET(__elfN(regset_fpregset));
2320 
2321 static size_t
2322 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list,
2323     struct thread *target_td)
2324 {
2325 	struct sysentvec *sv = td->td_proc->p_sysent;
2326 	struct regset **regsetp, **regset_end, *regset;
2327 	size_t size;
2328 
2329 	size = 0;
2330 
2331 	/* NT_PRSTATUS must be the first register set note. */
2332 	size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus),
2333 	    target_td);
2334 
2335 	regsetp = sv->sv_regset_begin;
2336 	if (regsetp == NULL) {
2337 		/* XXX: This shouldn't be true for any FreeBSD ABIs. */
2338 		size += __elfN(register_regset_note)(td, list,
2339 		    &__elfN(regset_fpregset), target_td);
2340 		return (size);
2341 	}
2342 	regset_end = sv->sv_regset_end;
2343 	MPASS(regset_end != NULL);
2344 	for (; regsetp < regset_end; regsetp++) {
2345 		regset = *regsetp;
2346 		if (regset->note == NT_PRSTATUS)
2347 			continue;
2348 		size += __elfN(register_regset_note)(td, list, regset,
2349 		    target_td);
2350 	}
2351 	return (size);
2352 }
2353 
2354 static void
2355 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2356 {
2357 	struct thread *td;
2358 	elf_thrmisc_t thrmisc;
2359 
2360 	td = arg;
2361 	if (sb != NULL) {
2362 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2363 		bzero(&thrmisc, sizeof(thrmisc));
2364 		strcpy(thrmisc.pr_tname, td->td_name);
2365 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2366 	}
2367 	*sizep = sizeof(thrmisc);
2368 }
2369 
2370 static void
2371 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2372 {
2373 	struct thread *td;
2374 	size_t size;
2375 	int structsize;
2376 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2377 	struct ptrace_lwpinfo32 pl;
2378 #else
2379 	struct ptrace_lwpinfo pl;
2380 #endif
2381 
2382 	td = arg;
2383 	size = sizeof(structsize) + sizeof(pl);
2384 	if (sb != NULL) {
2385 		KASSERT(*sizep == size, ("invalid size"));
2386 		structsize = sizeof(pl);
2387 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2388 		bzero(&pl, sizeof(pl));
2389 		pl.pl_lwpid = td->td_tid;
2390 		pl.pl_event = PL_EVENT_NONE;
2391 		pl.pl_sigmask = td->td_sigmask;
2392 		pl.pl_siglist = td->td_siglist;
2393 		if (td->td_si.si_signo != 0) {
2394 			pl.pl_event = PL_EVENT_SIGNAL;
2395 			pl.pl_flags |= PL_FLAG_SI;
2396 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2397 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2398 #else
2399 			pl.pl_siginfo = td->td_si;
2400 #endif
2401 		}
2402 		strcpy(pl.pl_tdname, td->td_name);
2403 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2404 		sbuf_bcat(sb, &pl, sizeof(pl));
2405 	}
2406 	*sizep = size;
2407 }
2408 
2409 /*
2410  * Allow for MD specific notes, as well as any MD
2411  * specific preparations for writing MI notes.
2412  */
2413 static void
2414 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2415 {
2416 	struct thread *td;
2417 	void *buf;
2418 	size_t size;
2419 
2420 	td = (struct thread *)arg;
2421 	size = *sizep;
2422 	if (size != 0 && sb != NULL)
2423 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2424 	else
2425 		buf = NULL;
2426 	size = 0;
2427 	__elfN(dump_thread)(td, buf, &size);
2428 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2429 	if (size != 0 && sb != NULL)
2430 		sbuf_bcat(sb, buf, size);
2431 	free(buf, M_TEMP);
2432 	*sizep = size;
2433 }
2434 
2435 #ifdef KINFO_PROC_SIZE
2436 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2437 #endif
2438 
2439 static void
2440 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2441 {
2442 	struct proc *p;
2443 	size_t size;
2444 	int structsize;
2445 
2446 	p = arg;
2447 	size = sizeof(structsize) + p->p_numthreads *
2448 	    sizeof(elf_kinfo_proc_t);
2449 
2450 	if (sb != NULL) {
2451 		KASSERT(*sizep == size, ("invalid size"));
2452 		structsize = sizeof(elf_kinfo_proc_t);
2453 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2454 		sx_slock(&proctree_lock);
2455 		PROC_LOCK(p);
2456 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2457 		sx_sunlock(&proctree_lock);
2458 	}
2459 	*sizep = size;
2460 }
2461 
2462 #ifdef KINFO_FILE_SIZE
2463 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2464 #endif
2465 
2466 static void
2467 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2468 {
2469 	struct proc *p;
2470 	size_t size, sect_sz, i;
2471 	ssize_t start_len, sect_len;
2472 	int structsize, filedesc_flags;
2473 
2474 	if (coredump_pack_fileinfo)
2475 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2476 	else
2477 		filedesc_flags = 0;
2478 
2479 	p = arg;
2480 	structsize = sizeof(struct kinfo_file);
2481 	if (sb == NULL) {
2482 		size = 0;
2483 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2484 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2485 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2486 		PROC_LOCK(p);
2487 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2488 		sbuf_finish(sb);
2489 		sbuf_delete(sb);
2490 		*sizep = size;
2491 	} else {
2492 		sbuf_start_section(sb, &start_len);
2493 
2494 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2495 		PROC_LOCK(p);
2496 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2497 		    filedesc_flags);
2498 
2499 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2500 		if (sect_len < 0)
2501 			return;
2502 		sect_sz = sect_len;
2503 
2504 		KASSERT(sect_sz <= *sizep,
2505 		    ("kern_proc_filedesc_out did not respect maxlen; "
2506 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2507 		     sect_sz - sizeof(structsize)));
2508 
2509 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2510 			sbuf_putc(sb, 0);
2511 	}
2512 }
2513 
2514 #ifdef KINFO_VMENTRY_SIZE
2515 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2516 #endif
2517 
2518 static void
2519 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2520 {
2521 	struct proc *p;
2522 	size_t size;
2523 	int structsize, vmmap_flags;
2524 
2525 	if (coredump_pack_vmmapinfo)
2526 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2527 	else
2528 		vmmap_flags = 0;
2529 
2530 	p = arg;
2531 	structsize = sizeof(struct kinfo_vmentry);
2532 	if (sb == NULL) {
2533 		size = 0;
2534 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2535 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2536 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2537 		PROC_LOCK(p);
2538 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2539 		sbuf_finish(sb);
2540 		sbuf_delete(sb);
2541 		*sizep = size;
2542 	} else {
2543 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2544 		PROC_LOCK(p);
2545 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2546 		    vmmap_flags);
2547 	}
2548 }
2549 
2550 static void
2551 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2552 {
2553 	struct proc *p;
2554 	size_t size;
2555 	int structsize;
2556 
2557 	p = arg;
2558 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2559 	if (sb != NULL) {
2560 		KASSERT(*sizep == size, ("invalid size"));
2561 		structsize = sizeof(gid_t);
2562 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2563 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2564 		    sizeof(gid_t));
2565 	}
2566 	*sizep = size;
2567 }
2568 
2569 static void
2570 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2571 {
2572 	struct proc *p;
2573 	size_t size;
2574 	int structsize;
2575 
2576 	p = arg;
2577 	size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2578 	if (sb != NULL) {
2579 		KASSERT(*sizep == size, ("invalid size"));
2580 		structsize = sizeof(p->p_pd->pd_cmask);
2581 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2582 		sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2583 	}
2584 	*sizep = size;
2585 }
2586 
2587 static void
2588 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2589 {
2590 	struct proc *p;
2591 	struct rlimit rlim[RLIM_NLIMITS];
2592 	size_t size;
2593 	int structsize, i;
2594 
2595 	p = arg;
2596 	size = sizeof(structsize) + sizeof(rlim);
2597 	if (sb != NULL) {
2598 		KASSERT(*sizep == size, ("invalid size"));
2599 		structsize = sizeof(rlim);
2600 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2601 		PROC_LOCK(p);
2602 		for (i = 0; i < RLIM_NLIMITS; i++)
2603 			lim_rlimit_proc(p, i, &rlim[i]);
2604 		PROC_UNLOCK(p);
2605 		sbuf_bcat(sb, rlim, sizeof(rlim));
2606 	}
2607 	*sizep = size;
2608 }
2609 
2610 static void
2611 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2612 {
2613 	struct proc *p;
2614 	size_t size;
2615 	int structsize;
2616 
2617 	p = arg;
2618 	size = sizeof(structsize) + sizeof(p->p_osrel);
2619 	if (sb != NULL) {
2620 		KASSERT(*sizep == size, ("invalid size"));
2621 		structsize = sizeof(p->p_osrel);
2622 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2623 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2624 	}
2625 	*sizep = size;
2626 }
2627 
2628 static void
2629 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2630 {
2631 	struct proc *p;
2632 	elf_ps_strings_t ps_strings;
2633 	size_t size;
2634 	int structsize;
2635 
2636 	p = arg;
2637 	size = sizeof(structsize) + sizeof(ps_strings);
2638 	if (sb != NULL) {
2639 		KASSERT(*sizep == size, ("invalid size"));
2640 		structsize = sizeof(ps_strings);
2641 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2642 		ps_strings = PTROUT(PROC_PS_STRINGS(p));
2643 #else
2644 		ps_strings = PROC_PS_STRINGS(p);
2645 #endif
2646 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2647 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2648 	}
2649 	*sizep = size;
2650 }
2651 
2652 static void
2653 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2654 {
2655 	struct proc *p;
2656 	size_t size;
2657 	int structsize;
2658 
2659 	p = arg;
2660 	if (sb == NULL) {
2661 		size = 0;
2662 		sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2663 		    SBUF_FIXEDLEN);
2664 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2665 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2666 		PHOLD(p);
2667 		proc_getauxv(curthread, p, sb);
2668 		PRELE(p);
2669 		sbuf_finish(sb);
2670 		sbuf_delete(sb);
2671 		*sizep = size;
2672 	} else {
2673 		structsize = sizeof(Elf_Auxinfo);
2674 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2675 		PHOLD(p);
2676 		proc_getauxv(curthread, p, sb);
2677 		PRELE(p);
2678 	}
2679 }
2680 
2681 static bool
2682 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2683     const char *note_vendor, const Elf_Phdr *pnote,
2684     bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2685 {
2686 	const Elf_Note *note, *note0, *note_end;
2687 	const char *note_name;
2688 	char *buf;
2689 	int i, error;
2690 	bool res;
2691 
2692 	/* We need some limit, might as well use PAGE_SIZE. */
2693 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2694 		return (false);
2695 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2696 	if (pnote->p_offset > PAGE_SIZE ||
2697 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2698 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2699 		if (buf == NULL) {
2700 			VOP_UNLOCK(imgp->vp);
2701 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2702 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2703 		}
2704 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2705 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2706 		    curthread->td_ucred, NOCRED, NULL, curthread);
2707 		if (error != 0) {
2708 			uprintf("i/o error PT_NOTE\n");
2709 			goto retf;
2710 		}
2711 		note = note0 = (const Elf_Note *)buf;
2712 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2713 	} else {
2714 		note = note0 = (const Elf_Note *)(imgp->image_header +
2715 		    pnote->p_offset);
2716 		note_end = (const Elf_Note *)(imgp->image_header +
2717 		    pnote->p_offset + pnote->p_filesz);
2718 		buf = NULL;
2719 	}
2720 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2721 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2722 		    (const char *)note < sizeof(Elf_Note)) {
2723 			goto retf;
2724 		}
2725 		if (note->n_namesz != checknote->n_namesz ||
2726 		    note->n_descsz != checknote->n_descsz ||
2727 		    note->n_type != checknote->n_type)
2728 			goto nextnote;
2729 		note_name = (const char *)(note + 1);
2730 		if (note_name + checknote->n_namesz >=
2731 		    (const char *)note_end || strncmp(note_vendor,
2732 		    note_name, checknote->n_namesz) != 0)
2733 			goto nextnote;
2734 
2735 		if (cb(note, cb_arg, &res))
2736 			goto ret;
2737 nextnote:
2738 		note = (const Elf_Note *)((const char *)(note + 1) +
2739 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2740 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2741 	}
2742 retf:
2743 	res = false;
2744 ret:
2745 	free(buf, M_TEMP);
2746 	return (res);
2747 }
2748 
2749 struct brandnote_cb_arg {
2750 	Elf_Brandnote *brandnote;
2751 	int32_t *osrel;
2752 };
2753 
2754 static bool
2755 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2756 {
2757 	struct brandnote_cb_arg *arg;
2758 
2759 	arg = arg0;
2760 
2761 	/*
2762 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2763 	 * necessary.
2764 	 */
2765 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2766 	    arg->brandnote->trans_osrel != NULL ?
2767 	    arg->brandnote->trans_osrel(note, arg->osrel) : true;
2768 
2769 	return (true);
2770 }
2771 
2772 static Elf_Note fctl_note = {
2773 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2774 	.n_descsz = sizeof(uint32_t),
2775 	.n_type = NT_FREEBSD_FEATURE_CTL,
2776 };
2777 
2778 struct fctl_cb_arg {
2779 	bool *has_fctl0;
2780 	uint32_t *fctl0;
2781 };
2782 
2783 static bool
2784 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2785 {
2786 	struct fctl_cb_arg *arg;
2787 	const Elf32_Word *desc;
2788 	uintptr_t p;
2789 
2790 	arg = arg0;
2791 	p = (uintptr_t)(note + 1);
2792 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2793 	desc = (const Elf32_Word *)p;
2794 	*arg->has_fctl0 = true;
2795 	*arg->fctl0 = desc[0];
2796 	*res = true;
2797 	return (true);
2798 }
2799 
2800 /*
2801  * Try to find the appropriate ABI-note section for checknote, fetch
2802  * the osreldate and feature control flags for binary from the ELF
2803  * OSABI-note.  Only the first page of the image is searched, the same
2804  * as for headers.
2805  */
2806 static bool
2807 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2808     int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2809 {
2810 	const Elf_Phdr *phdr;
2811 	const Elf_Ehdr *hdr;
2812 	struct brandnote_cb_arg b_arg;
2813 	struct fctl_cb_arg f_arg;
2814 	int i, j;
2815 
2816 	hdr = (const Elf_Ehdr *)imgp->image_header;
2817 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2818 	b_arg.brandnote = brandnote;
2819 	b_arg.osrel = osrel;
2820 	f_arg.has_fctl0 = has_fctl0;
2821 	f_arg.fctl0 = fctl0;
2822 
2823 	for (i = 0; i < hdr->e_phnum; i++) {
2824 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2825 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2826 		    &b_arg)) {
2827 			for (j = 0; j < hdr->e_phnum; j++) {
2828 				if (phdr[j].p_type == PT_NOTE &&
2829 				    __elfN(parse_notes)(imgp, &fctl_note,
2830 				    FREEBSD_ABI_VENDOR, &phdr[j],
2831 				    note_fctl_cb, &f_arg))
2832 					break;
2833 			}
2834 			return (true);
2835 		}
2836 	}
2837 	return (false);
2838 
2839 }
2840 
2841 /*
2842  * Tell kern_execve.c about it, with a little help from the linker.
2843  */
2844 static struct execsw __elfN(execsw) = {
2845 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2846 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2847 };
2848 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2849 
2850 static vm_prot_t
2851 __elfN(trans_prot)(Elf_Word flags)
2852 {
2853 	vm_prot_t prot;
2854 
2855 	prot = 0;
2856 	if (flags & PF_X)
2857 		prot |= VM_PROT_EXECUTE;
2858 	if (flags & PF_W)
2859 		prot |= VM_PROT_WRITE;
2860 	if (flags & PF_R)
2861 		prot |= VM_PROT_READ;
2862 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2863 	if (i386_read_exec && (flags & PF_R))
2864 		prot |= VM_PROT_EXECUTE;
2865 #endif
2866 	return (prot);
2867 }
2868 
2869 static Elf_Word
2870 __elfN(untrans_prot)(vm_prot_t prot)
2871 {
2872 	Elf_Word flags;
2873 
2874 	flags = 0;
2875 	if (prot & VM_PROT_EXECUTE)
2876 		flags |= PF_X;
2877 	if (prot & VM_PROT_READ)
2878 		flags |= PF_R;
2879 	if (prot & VM_PROT_WRITE)
2880 		flags |= PF_W;
2881 	return (flags);
2882 }
2883