xref: /freebsd/sys/kern/imgact_elf.c (revision cfd6422a5217410fbd66f7a7a8a64d9d85e61229)
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/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sbuf.h>
60 #include <sys/sf_buf.h>
61 #include <sys/smp.h>
62 #include <sys/systm.h>
63 #include <sys/signalvar.h>
64 #include <sys/stat.h>
65 #include <sys/sx.h>
66 #include <sys/syscall.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/vnode.h>
70 #include <sys/syslog.h>
71 #include <sys/eventhandler.h>
72 #include <sys/user.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_extern.h>
81 
82 #include <machine/elf.h>
83 #include <machine/md_var.h>
84 
85 #define ELF_NOTE_ROUNDSIZE	4
86 #define OLD_EI_BRAND	8
87 
88 static int __elfN(check_header)(const Elf_Ehdr *hdr);
89 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
90     const char *interp, int32_t *osrel, uint32_t *fctl0);
91 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
92     u_long *entry);
93 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
94     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
97     int32_t *osrel);
98 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100     Elf_Brandnote *checknote, int32_t *osrel, boolean_t *has_fctl0,
101     uint32_t *fctl0);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
106     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
107     "");
108 
109 #define	CORE_BUF_SIZE	(16 * 1024)
110 
111 int __elfN(fallback_brand) = -1;
112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
113     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
114     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 
116 static int elf_legacy_coredump = 0;
117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
118     &elf_legacy_coredump, 0,
119     "include all and only RW pages in core dumps");
120 
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
124     defined(__riscv)
125 	1;
126 #else
127 	0;
128 #endif
129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
130     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
131     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 
133 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136     "enable execution from readable segments");
137 #endif
138 
139 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
140 static int
141 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
142 {
143 	u_long val;
144 	int error;
145 
146 	val = __elfN(pie_base);
147 	error = sysctl_handle_long(oidp, &val, 0, req);
148 	if (error != 0 || req->newptr == NULL)
149 		return (error);
150 	if ((val & PAGE_MASK) != 0)
151 		return (EINVAL);
152 	__elfN(pie_base) = val;
153 	return (0);
154 }
155 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
156     CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
157     sysctl_pie_base, "LU",
158     "PIE load base without randomization");
159 
160 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
161     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
162     "");
163 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
164 
165 static int __elfN(aslr_enabled) = 0;
166 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
167     &__elfN(aslr_enabled), 0,
168     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
169     ": enable address map randomization");
170 
171 static int __elfN(pie_aslr_enabled) = 0;
172 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
173     &__elfN(pie_aslr_enabled), 0,
174     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
175     ": enable address map randomization for PIE binaries");
176 
177 static int __elfN(aslr_honor_sbrk) = 1;
178 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
179     &__elfN(aslr_honor_sbrk), 0,
180     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
181 
182 static int __elfN(aslr_stack_gap) = 3;
183 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
184     &__elfN(aslr_stack_gap), 0,
185     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
186     ": maximum percentage of main stack to waste on a random gap");
187 
188 static int __elfN(sigfastblock) = 1;
189 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
190     CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
191     "enable sigfastblock for new processes");
192 
193 static bool __elfN(allow_wx) = true;
194 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
195     CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
196     "Allow pages to be mapped simultaneously writable and executable");
197 
198 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
199 
200 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
201 
202 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
203 
204 Elf_Brandnote __elfN(freebsd_brandnote) = {
205 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
206 	.hdr.n_descsz	= sizeof(int32_t),
207 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
208 	.vendor		= FREEBSD_ABI_VENDOR,
209 	.flags		= BN_TRANSLATE_OSREL,
210 	.trans_osrel	= __elfN(freebsd_trans_osrel)
211 };
212 
213 static bool
214 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
215 {
216 	uintptr_t p;
217 
218 	p = (uintptr_t)(note + 1);
219 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
220 	*osrel = *(const int32_t *)(p);
221 
222 	return (true);
223 }
224 
225 static const char GNU_ABI_VENDOR[] = "GNU";
226 static int GNU_KFREEBSD_ABI_DESC = 3;
227 
228 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
229 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
230 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
231 	.hdr.n_type	= 1,
232 	.vendor		= GNU_ABI_VENDOR,
233 	.flags		= BN_TRANSLATE_OSREL,
234 	.trans_osrel	= kfreebsd_trans_osrel
235 };
236 
237 static bool
238 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
239 {
240 	const Elf32_Word *desc;
241 	uintptr_t p;
242 
243 	p = (uintptr_t)(note + 1);
244 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
245 
246 	desc = (const Elf32_Word *)p;
247 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
248 		return (false);
249 
250 	/*
251 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
252 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
253 	 */
254 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
255 
256 	return (true);
257 }
258 
259 int
260 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
261 {
262 	int i;
263 
264 	for (i = 0; i < MAX_BRANDS; i++) {
265 		if (elf_brand_list[i] == NULL) {
266 			elf_brand_list[i] = entry;
267 			break;
268 		}
269 	}
270 	if (i == MAX_BRANDS) {
271 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
272 			__func__, entry);
273 		return (-1);
274 	}
275 	return (0);
276 }
277 
278 int
279 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
280 {
281 	int i;
282 
283 	for (i = 0; i < MAX_BRANDS; i++) {
284 		if (elf_brand_list[i] == entry) {
285 			elf_brand_list[i] = NULL;
286 			break;
287 		}
288 	}
289 	if (i == MAX_BRANDS)
290 		return (-1);
291 	return (0);
292 }
293 
294 int
295 __elfN(brand_inuse)(Elf_Brandinfo *entry)
296 {
297 	struct proc *p;
298 	int rval = FALSE;
299 
300 	sx_slock(&allproc_lock);
301 	FOREACH_PROC_IN_SYSTEM(p) {
302 		if (p->p_sysent == entry->sysvec) {
303 			rval = TRUE;
304 			break;
305 		}
306 	}
307 	sx_sunlock(&allproc_lock);
308 
309 	return (rval);
310 }
311 
312 static Elf_Brandinfo *
313 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
314     int32_t *osrel, uint32_t *fctl0)
315 {
316 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
317 	Elf_Brandinfo *bi, *bi_m;
318 	boolean_t ret, has_fctl0;
319 	int i, interp_name_len;
320 
321 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
322 
323 	/*
324 	 * We support four types of branding -- (1) the ELF EI_OSABI field
325 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
326 	 * branding w/in the ELF header, (3) path of the `interp_path'
327 	 * field, and (4) the ".note.ABI-tag" ELF section.
328 	 */
329 
330 	/* Look for an ".note.ABI-tag" ELF section */
331 	bi_m = NULL;
332 	for (i = 0; i < MAX_BRANDS; i++) {
333 		bi = elf_brand_list[i];
334 		if (bi == NULL)
335 			continue;
336 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
337 			continue;
338 		if (hdr->e_machine == bi->machine && (bi->flags &
339 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
340 			has_fctl0 = false;
341 			*fctl0 = 0;
342 			*osrel = 0;
343 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
344 			    &has_fctl0, fctl0);
345 			/* Give brand a chance to veto check_note's guess */
346 			if (ret && bi->header_supported) {
347 				ret = bi->header_supported(imgp, osrel,
348 				    has_fctl0 ? fctl0 : NULL);
349 			}
350 			/*
351 			 * If note checker claimed the binary, but the
352 			 * interpreter path in the image does not
353 			 * match default one for the brand, try to
354 			 * search for other brands with the same
355 			 * interpreter.  Either there is better brand
356 			 * with the right interpreter, or, failing
357 			 * this, we return first brand which accepted
358 			 * our note and, optionally, header.
359 			 */
360 			if (ret && bi_m == NULL && interp != NULL &&
361 			    (bi->interp_path == NULL ||
362 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
363 			    strncmp(interp, bi->interp_path, interp_name_len)
364 			    != 0))) {
365 				bi_m = bi;
366 				ret = 0;
367 			}
368 			if (ret)
369 				return (bi);
370 		}
371 	}
372 	if (bi_m != NULL)
373 		return (bi_m);
374 
375 	/* If the executable has a brand, search for it in the brand list. */
376 	for (i = 0; i < MAX_BRANDS; i++) {
377 		bi = elf_brand_list[i];
378 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
379 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
380 			continue;
381 		if (hdr->e_machine == bi->machine &&
382 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
383 		    (bi->compat_3_brand != NULL &&
384 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
385 		    bi->compat_3_brand) == 0))) {
386 			/* Looks good, but give brand a chance to veto */
387 			if (bi->header_supported == NULL ||
388 			    bi->header_supported(imgp, NULL, NULL)) {
389 				/*
390 				 * Again, prefer strictly matching
391 				 * interpreter path.
392 				 */
393 				if (interp_name_len == 0 &&
394 				    bi->interp_path == NULL)
395 					return (bi);
396 				if (bi->interp_path != NULL &&
397 				    strlen(bi->interp_path) + 1 ==
398 				    interp_name_len && strncmp(interp,
399 				    bi->interp_path, interp_name_len) == 0)
400 					return (bi);
401 				if (bi_m == NULL)
402 					bi_m = bi;
403 			}
404 		}
405 	}
406 	if (bi_m != NULL)
407 		return (bi_m);
408 
409 	/* No known brand, see if the header is recognized by any brand */
410 	for (i = 0; i < MAX_BRANDS; i++) {
411 		bi = elf_brand_list[i];
412 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
413 		    bi->header_supported == NULL)
414 			continue;
415 		if (hdr->e_machine == bi->machine) {
416 			ret = bi->header_supported(imgp, NULL, NULL);
417 			if (ret)
418 				return (bi);
419 		}
420 	}
421 
422 	/* Lacking a known brand, search for a recognized interpreter. */
423 	if (interp != NULL) {
424 		for (i = 0; i < MAX_BRANDS; i++) {
425 			bi = elf_brand_list[i];
426 			if (bi == NULL || (bi->flags &
427 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
428 			    != 0)
429 				continue;
430 			if (hdr->e_machine == bi->machine &&
431 			    bi->interp_path != NULL &&
432 			    /* ELF image p_filesz includes terminating zero */
433 			    strlen(bi->interp_path) + 1 == interp_name_len &&
434 			    strncmp(interp, bi->interp_path, interp_name_len)
435 			    == 0 && (bi->header_supported == NULL ||
436 			    bi->header_supported(imgp, NULL, NULL)))
437 				return (bi);
438 		}
439 	}
440 
441 	/* Lacking a recognized interpreter, try the default brand */
442 	for (i = 0; i < MAX_BRANDS; i++) {
443 		bi = elf_brand_list[i];
444 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
445 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
446 			continue;
447 		if (hdr->e_machine == bi->machine &&
448 		    __elfN(fallback_brand) == bi->brand &&
449 		    (bi->header_supported == NULL ||
450 		    bi->header_supported(imgp, NULL, NULL)))
451 			return (bi);
452 	}
453 	return (NULL);
454 }
455 
456 static bool
457 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
458 {
459 	return (hdr->e_phoff <= PAGE_SIZE &&
460 	    (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
461 }
462 
463 static int
464 __elfN(check_header)(const Elf_Ehdr *hdr)
465 {
466 	Elf_Brandinfo *bi;
467 	int i;
468 
469 	if (!IS_ELF(*hdr) ||
470 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
471 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
472 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
473 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
474 	    hdr->e_version != ELF_TARG_VER)
475 		return (ENOEXEC);
476 
477 	/*
478 	 * Make sure we have at least one brand for this machine.
479 	 */
480 
481 	for (i = 0; i < MAX_BRANDS; i++) {
482 		bi = elf_brand_list[i];
483 		if (bi != NULL && bi->machine == hdr->e_machine)
484 			break;
485 	}
486 	if (i == MAX_BRANDS)
487 		return (ENOEXEC);
488 
489 	return (0);
490 }
491 
492 static int
493 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
494     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
495 {
496 	struct sf_buf *sf;
497 	int error;
498 	vm_offset_t off;
499 
500 	/*
501 	 * Create the page if it doesn't exist yet. Ignore errors.
502 	 */
503 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
504 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
505 
506 	/*
507 	 * Find the page from the underlying object.
508 	 */
509 	if (object != NULL) {
510 		sf = vm_imgact_map_page(object, offset);
511 		if (sf == NULL)
512 			return (KERN_FAILURE);
513 		off = offset - trunc_page(offset);
514 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
515 		    end - start);
516 		vm_imgact_unmap_page(sf);
517 		if (error != 0)
518 			return (KERN_FAILURE);
519 	}
520 
521 	return (KERN_SUCCESS);
522 }
523 
524 static int
525 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
526     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
527     int cow)
528 {
529 	struct sf_buf *sf;
530 	vm_offset_t off;
531 	vm_size_t sz;
532 	int error, locked, rv;
533 
534 	if (start != trunc_page(start)) {
535 		rv = __elfN(map_partial)(map, object, offset, start,
536 		    round_page(start), prot);
537 		if (rv != KERN_SUCCESS)
538 			return (rv);
539 		offset += round_page(start) - start;
540 		start = round_page(start);
541 	}
542 	if (end != round_page(end)) {
543 		rv = __elfN(map_partial)(map, object, offset +
544 		    trunc_page(end) - start, trunc_page(end), end, prot);
545 		if (rv != KERN_SUCCESS)
546 			return (rv);
547 		end = trunc_page(end);
548 	}
549 	if (start >= end)
550 		return (KERN_SUCCESS);
551 	if ((offset & PAGE_MASK) != 0) {
552 		/*
553 		 * The mapping is not page aligned.  This means that we have
554 		 * to copy the data.
555 		 */
556 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
557 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
558 		if (rv != KERN_SUCCESS)
559 			return (rv);
560 		if (object == NULL)
561 			return (KERN_SUCCESS);
562 		for (; start < end; start += sz) {
563 			sf = vm_imgact_map_page(object, offset);
564 			if (sf == NULL)
565 				return (KERN_FAILURE);
566 			off = offset - trunc_page(offset);
567 			sz = end - start;
568 			if (sz > PAGE_SIZE - off)
569 				sz = PAGE_SIZE - off;
570 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
571 			    (caddr_t)start, sz);
572 			vm_imgact_unmap_page(sf);
573 			if (error != 0)
574 				return (KERN_FAILURE);
575 			offset += sz;
576 		}
577 	} else {
578 		vm_object_reference(object);
579 		rv = vm_map_fixed(map, object, offset, start, end - start,
580 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
581 		    (object != NULL ? MAP_VN_EXEC : 0));
582 		if (rv != KERN_SUCCESS) {
583 			locked = VOP_ISLOCKED(imgp->vp);
584 			VOP_UNLOCK(imgp->vp);
585 			vm_object_deallocate(object);
586 			vn_lock(imgp->vp, locked | LK_RETRY);
587 			return (rv);
588 		} else if (object != NULL) {
589 			MPASS(imgp->vp->v_object == object);
590 			VOP_SET_TEXT_CHECKED(imgp->vp);
591 		}
592 	}
593 	return (KERN_SUCCESS);
594 }
595 
596 static int
597 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
598     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
599 {
600 	struct sf_buf *sf;
601 	size_t map_len;
602 	vm_map_t map;
603 	vm_object_t object;
604 	vm_offset_t map_addr;
605 	int error, rv, cow;
606 	size_t copy_len;
607 	vm_ooffset_t file_addr;
608 
609 	/*
610 	 * It's necessary to fail if the filsz + offset taken from the
611 	 * header is greater than the actual file pager object's size.
612 	 * If we were to allow this, then the vm_map_find() below would
613 	 * walk right off the end of the file object and into the ether.
614 	 *
615 	 * While I'm here, might as well check for something else that
616 	 * is invalid: filsz cannot be greater than memsz.
617 	 */
618 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
619 	    filsz > memsz) {
620 		uprintf("elf_load_section: truncated ELF file\n");
621 		return (ENOEXEC);
622 	}
623 
624 	object = imgp->object;
625 	map = &imgp->proc->p_vmspace->vm_map;
626 	map_addr = trunc_page((vm_offset_t)vmaddr);
627 	file_addr = trunc_page(offset);
628 
629 	/*
630 	 * We have two choices.  We can either clear the data in the last page
631 	 * of an oversized mapping, or we can start the anon mapping a page
632 	 * early and copy the initialized data into that first page.  We
633 	 * choose the second.
634 	 */
635 	if (filsz == 0)
636 		map_len = 0;
637 	else if (memsz > filsz)
638 		map_len = trunc_page(offset + filsz) - file_addr;
639 	else
640 		map_len = round_page(offset + filsz) - file_addr;
641 
642 	if (map_len != 0) {
643 		/* cow flags: don't dump readonly sections in core */
644 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
645 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
646 
647 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
648 		    map_addr, map_addr + map_len, prot, cow);
649 		if (rv != KERN_SUCCESS)
650 			return (EINVAL);
651 
652 		/* we can stop now if we've covered it all */
653 		if (memsz == filsz)
654 			return (0);
655 	}
656 
657 	/*
658 	 * We have to get the remaining bit of the file into the first part
659 	 * of the oversized map segment.  This is normally because the .data
660 	 * segment in the file is extended to provide bss.  It's a neat idea
661 	 * to try and save a page, but it's a pain in the behind to implement.
662 	 */
663 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
664 	    filsz);
665 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
666 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
667 
668 	/* This had damn well better be true! */
669 	if (map_len != 0) {
670 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
671 		    map_addr + map_len, prot, 0);
672 		if (rv != KERN_SUCCESS)
673 			return (EINVAL);
674 	}
675 
676 	if (copy_len != 0) {
677 		sf = vm_imgact_map_page(object, offset + filsz);
678 		if (sf == NULL)
679 			return (EIO);
680 
681 		/* send the page fragment to user space */
682 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
683 		    copy_len);
684 		vm_imgact_unmap_page(sf);
685 		if (error != 0)
686 			return (error);
687 	}
688 
689 	/*
690 	 * Remove write access to the page if it was only granted by map_insert
691 	 * to allow copyout.
692 	 */
693 	if ((prot & VM_PROT_WRITE) == 0)
694 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
695 		    map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
696 
697 	return (0);
698 }
699 
700 static int
701 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
702     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
703 {
704 	vm_prot_t prot;
705 	u_long base_addr;
706 	bool first;
707 	int error, i;
708 
709 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
710 
711 	base_addr = 0;
712 	first = true;
713 
714 	for (i = 0; i < hdr->e_phnum; i++) {
715 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
716 			continue;
717 
718 		/* Loadable segment */
719 		prot = __elfN(trans_prot)(phdr[i].p_flags);
720 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
721 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
722 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
723 		if (error != 0)
724 			return (error);
725 
726 		/*
727 		 * Establish the base address if this is the first segment.
728 		 */
729 		if (first) {
730   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
731 			first = false;
732 		}
733 	}
734 
735 	if (base_addrp != NULL)
736 		*base_addrp = base_addr;
737 
738 	return (0);
739 }
740 
741 /*
742  * Load the file "file" into memory.  It may be either a shared object
743  * or an executable.
744  *
745  * The "addr" reference parameter is in/out.  On entry, it specifies
746  * the address where a shared object should be loaded.  If the file is
747  * an executable, this value is ignored.  On exit, "addr" specifies
748  * where the file was actually loaded.
749  *
750  * The "entry" reference parameter is out only.  On exit, it specifies
751  * the entry point for the loaded file.
752  */
753 static int
754 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
755 	u_long *entry)
756 {
757 	struct {
758 		struct nameidata nd;
759 		struct vattr attr;
760 		struct image_params image_params;
761 	} *tempdata;
762 	const Elf_Ehdr *hdr = NULL;
763 	const Elf_Phdr *phdr = NULL;
764 	struct nameidata *nd;
765 	struct vattr *attr;
766 	struct image_params *imgp;
767 	u_long rbase;
768 	u_long base_addr = 0;
769 	int error;
770 
771 #ifdef CAPABILITY_MODE
772 	/*
773 	 * XXXJA: This check can go away once we are sufficiently confident
774 	 * that the checks in namei() are correct.
775 	 */
776 	if (IN_CAPABILITY_MODE(curthread))
777 		return (ECAPMODE);
778 #endif
779 
780 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
781 	nd = &tempdata->nd;
782 	attr = &tempdata->attr;
783 	imgp = &tempdata->image_params;
784 
785 	/*
786 	 * Initialize part of the common data
787 	 */
788 	imgp->proc = p;
789 	imgp->attr = attr;
790 
791 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
792 	    UIO_SYSSPACE, file, curthread);
793 	if ((error = namei(nd)) != 0) {
794 		nd->ni_vp = NULL;
795 		goto fail;
796 	}
797 	NDFREE(nd, NDF_ONLY_PNBUF);
798 	imgp->vp = nd->ni_vp;
799 
800 	/*
801 	 * Check permissions, modes, uid, etc on the file, and "open" it.
802 	 */
803 	error = exec_check_permissions(imgp);
804 	if (error)
805 		goto fail;
806 
807 	error = exec_map_first_page(imgp);
808 	if (error)
809 		goto fail;
810 
811 	imgp->object = nd->ni_vp->v_object;
812 
813 	hdr = (const Elf_Ehdr *)imgp->image_header;
814 	if ((error = __elfN(check_header)(hdr)) != 0)
815 		goto fail;
816 	if (hdr->e_type == ET_DYN)
817 		rbase = *addr;
818 	else if (hdr->e_type == ET_EXEC)
819 		rbase = 0;
820 	else {
821 		error = ENOEXEC;
822 		goto fail;
823 	}
824 
825 	/* Only support headers that fit within first page for now      */
826 	if (!__elfN(phdr_in_zero_page)(hdr)) {
827 		error = ENOEXEC;
828 		goto fail;
829 	}
830 
831 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
832 	if (!aligned(phdr, Elf_Addr)) {
833 		error = ENOEXEC;
834 		goto fail;
835 	}
836 
837 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
838 	if (error != 0)
839 		goto fail;
840 
841 	*addr = base_addr;
842 	*entry = (unsigned long)hdr->e_entry + rbase;
843 
844 fail:
845 	if (imgp->firstpage)
846 		exec_unmap_first_page(imgp);
847 
848 	if (nd->ni_vp) {
849 		if (imgp->textset)
850 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
851 		vput(nd->ni_vp);
852 	}
853 	free(tempdata, M_TEMP);
854 
855 	return (error);
856 }
857 
858 static u_long
859 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
860     u_int align)
861 {
862 	u_long rbase, res;
863 
864 	MPASS(vm_map_min(map) <= minv);
865 	MPASS(maxv <= vm_map_max(map));
866 	MPASS(minv < maxv);
867 	MPASS(minv + align < maxv);
868 	arc4rand(&rbase, sizeof(rbase), 0);
869 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
870 	res &= ~((u_long)align - 1);
871 	if (res >= maxv)
872 		res -= align;
873 	KASSERT(res >= minv,
874 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
875 	    res, minv, maxv, rbase));
876 	KASSERT(res < maxv,
877 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
878 	    res, maxv, minv, rbase));
879 	return (res);
880 }
881 
882 static int
883 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
884     const Elf_Phdr *phdr, u_long et_dyn_addr)
885 {
886 	struct vmspace *vmspace;
887 	const char *err_str;
888 	u_long text_size, data_size, total_size, text_addr, data_addr;
889 	u_long seg_size, seg_addr;
890 	int i;
891 
892 	err_str = NULL;
893 	text_size = data_size = total_size = text_addr = data_addr = 0;
894 
895 	for (i = 0; i < hdr->e_phnum; i++) {
896 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
897 			continue;
898 
899 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
900 		seg_size = round_page(phdr[i].p_memsz +
901 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
902 
903 		/*
904 		 * Make the largest executable segment the official
905 		 * text segment and all others data.
906 		 *
907 		 * Note that obreak() assumes that data_addr + data_size == end
908 		 * of data load area, and the ELF file format expects segments
909 		 * to be sorted by address.  If multiple data segments exist,
910 		 * the last one will be used.
911 		 */
912 
913 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
914 			text_size = seg_size;
915 			text_addr = seg_addr;
916 		} else {
917 			data_size = seg_size;
918 			data_addr = seg_addr;
919 		}
920 		total_size += seg_size;
921 	}
922 
923 	if (data_addr == 0 && data_size == 0) {
924 		data_addr = text_addr;
925 		data_size = text_size;
926 	}
927 
928 	/*
929 	 * Check limits.  It should be safe to check the
930 	 * limits after loading the segments since we do
931 	 * not actually fault in all the segments pages.
932 	 */
933 	PROC_LOCK(imgp->proc);
934 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
935 		err_str = "Data segment size exceeds process limit";
936 	else if (text_size > maxtsiz)
937 		err_str = "Text segment size exceeds system limit";
938 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
939 		err_str = "Total segment size exceeds process limit";
940 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
941 		err_str = "Data segment size exceeds resource limit";
942 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
943 		err_str = "Total segment size exceeds resource limit";
944 	PROC_UNLOCK(imgp->proc);
945 	if (err_str != NULL) {
946 		uprintf("%s\n", err_str);
947 		return (ENOMEM);
948 	}
949 
950 	vmspace = imgp->proc->p_vmspace;
951 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
952 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
953 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
954 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
955 
956 	return (0);
957 }
958 
959 static int
960 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
961     char **interpp, bool *free_interpp)
962 {
963 	struct thread *td;
964 	char *interp;
965 	int error, interp_name_len;
966 
967 	KASSERT(phdr->p_type == PT_INTERP,
968 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
969 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
970 
971 	td = curthread;
972 
973 	/* Path to interpreter */
974 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
975 		uprintf("Invalid PT_INTERP\n");
976 		return (ENOEXEC);
977 	}
978 
979 	interp_name_len = phdr->p_filesz;
980 	if (phdr->p_offset > PAGE_SIZE ||
981 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
982 		/*
983 		 * The vnode lock might be needed by the pagedaemon to
984 		 * clean pages owned by the vnode.  Do not allow sleep
985 		 * waiting for memory with the vnode locked, instead
986 		 * try non-sleepable allocation first, and if it
987 		 * fails, go to the slow path were we drop the lock
988 		 * and do M_WAITOK.  A text reference prevents
989 		 * modifications to the vnode content.
990 		 */
991 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
992 		if (interp == NULL) {
993 			VOP_UNLOCK(imgp->vp);
994 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
995 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
996 		}
997 
998 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
999 		    interp_name_len, phdr->p_offset,
1000 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1001 		    NOCRED, NULL, td);
1002 		if (error != 0) {
1003 			free(interp, M_TEMP);
1004 			uprintf("i/o error PT_INTERP %d\n", error);
1005 			return (error);
1006 		}
1007 		interp[interp_name_len] = '\0';
1008 
1009 		*interpp = interp;
1010 		*free_interpp = true;
1011 		return (0);
1012 	}
1013 
1014 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1015 	if (interp[interp_name_len - 1] != '\0') {
1016 		uprintf("Invalid PT_INTERP\n");
1017 		return (ENOEXEC);
1018 	}
1019 
1020 	*interpp = interp;
1021 	*free_interpp = false;
1022 	return (0);
1023 }
1024 
1025 static int
1026 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1027     const char *interp, u_long *addr, u_long *entry)
1028 {
1029 	char *path;
1030 	int error;
1031 
1032 	if (brand_info->emul_path != NULL &&
1033 	    brand_info->emul_path[0] != '\0') {
1034 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1035 		snprintf(path, MAXPATHLEN, "%s%s",
1036 		    brand_info->emul_path, interp);
1037 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1038 		free(path, M_TEMP);
1039 		if (error == 0)
1040 			return (0);
1041 	}
1042 
1043 	if (brand_info->interp_newpath != NULL &&
1044 	    (brand_info->interp_path == NULL ||
1045 	    strcmp(interp, brand_info->interp_path) == 0)) {
1046 		error = __elfN(load_file)(imgp->proc,
1047 		    brand_info->interp_newpath, addr, entry);
1048 		if (error == 0)
1049 			return (0);
1050 	}
1051 
1052 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1053 	if (error == 0)
1054 		return (0);
1055 
1056 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1057 	return (error);
1058 }
1059 
1060 /*
1061  * Impossible et_dyn_addr initial value indicating that the real base
1062  * must be calculated later with some randomization applied.
1063  */
1064 #define	ET_DYN_ADDR_RAND	1
1065 
1066 static int
1067 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1068 {
1069 	struct thread *td;
1070 	const Elf_Ehdr *hdr;
1071 	const Elf_Phdr *phdr;
1072 	Elf_Auxargs *elf_auxargs;
1073 	struct vmspace *vmspace;
1074 	vm_map_t map;
1075 	char *interp;
1076 	Elf_Brandinfo *brand_info;
1077 	struct sysentvec *sv;
1078 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1079 	u_long maxalign, mapsz, maxv, maxv1;
1080 	uint32_t fctl0;
1081 	int32_t osrel;
1082 	bool free_interp;
1083 	int error, i, n;
1084 
1085 	hdr = (const Elf_Ehdr *)imgp->image_header;
1086 
1087 	/*
1088 	 * Do we have a valid ELF header ?
1089 	 *
1090 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1091 	 * if particular brand doesn't support it.
1092 	 */
1093 	if (__elfN(check_header)(hdr) != 0 ||
1094 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1095 		return (-1);
1096 
1097 	/*
1098 	 * From here on down, we return an errno, not -1, as we've
1099 	 * detected an ELF file.
1100 	 */
1101 
1102 	if (!__elfN(phdr_in_zero_page)(hdr)) {
1103 		uprintf("Program headers not in the first page\n");
1104 		return (ENOEXEC);
1105 	}
1106 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1107 	if (!aligned(phdr, Elf_Addr)) {
1108 		uprintf("Unaligned program headers\n");
1109 		return (ENOEXEC);
1110 	}
1111 
1112 	n = error = 0;
1113 	baddr = 0;
1114 	osrel = 0;
1115 	fctl0 = 0;
1116 	entry = proghdr = 0;
1117 	interp = NULL;
1118 	free_interp = false;
1119 	td = curthread;
1120 	maxalign = PAGE_SIZE;
1121 	mapsz = 0;
1122 
1123 	for (i = 0; i < hdr->e_phnum; i++) {
1124 		switch (phdr[i].p_type) {
1125 		case PT_LOAD:
1126 			if (n == 0)
1127 				baddr = phdr[i].p_vaddr;
1128 			if (phdr[i].p_align > maxalign)
1129 				maxalign = phdr[i].p_align;
1130 			mapsz += phdr[i].p_memsz;
1131 			n++;
1132 
1133 			/*
1134 			 * If this segment contains the program headers,
1135 			 * remember their virtual address for the AT_PHDR
1136 			 * aux entry. Static binaries don't usually include
1137 			 * a PT_PHDR entry.
1138 			 */
1139 			if (phdr[i].p_offset == 0 &&
1140 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1141 				<= phdr[i].p_filesz)
1142 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1143 			break;
1144 		case PT_INTERP:
1145 			/* Path to interpreter */
1146 			if (interp != NULL) {
1147 				uprintf("Multiple PT_INTERP headers\n");
1148 				error = ENOEXEC;
1149 				goto ret;
1150 			}
1151 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1152 			    &free_interp);
1153 			if (error != 0)
1154 				goto ret;
1155 			break;
1156 		case PT_GNU_STACK:
1157 			if (__elfN(nxstack))
1158 				imgp->stack_prot =
1159 				    __elfN(trans_prot)(phdr[i].p_flags);
1160 			imgp->stack_sz = phdr[i].p_memsz;
1161 			break;
1162 		case PT_PHDR: 	/* Program header table info */
1163 			proghdr = phdr[i].p_vaddr;
1164 			break;
1165 		}
1166 	}
1167 
1168 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1169 	if (brand_info == NULL) {
1170 		uprintf("ELF binary type \"%u\" not known.\n",
1171 		    hdr->e_ident[EI_OSABI]);
1172 		error = ENOEXEC;
1173 		goto ret;
1174 	}
1175 	sv = brand_info->sysvec;
1176 	et_dyn_addr = 0;
1177 	if (hdr->e_type == ET_DYN) {
1178 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1179 			uprintf("Cannot execute shared object\n");
1180 			error = ENOEXEC;
1181 			goto ret;
1182 		}
1183 		/*
1184 		 * Honour the base load address from the dso if it is
1185 		 * non-zero for some reason.
1186 		 */
1187 		if (baddr == 0) {
1188 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1189 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1190 				et_dyn_addr = __elfN(pie_base);
1191 			else if ((__elfN(pie_aslr_enabled) &&
1192 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1193 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1194 				et_dyn_addr = ET_DYN_ADDR_RAND;
1195 			else
1196 				et_dyn_addr = __elfN(pie_base);
1197 		}
1198 	}
1199 
1200 	/*
1201 	 * Avoid a possible deadlock if the current address space is destroyed
1202 	 * and that address space maps the locked vnode.  In the common case,
1203 	 * the locked vnode's v_usecount is decremented but remains greater
1204 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1205 	 * However, in cases where the vnode lock is external, such as nullfs,
1206 	 * v_usecount may become zero.
1207 	 *
1208 	 * The VV_TEXT flag prevents modifications to the executable while
1209 	 * the vnode is unlocked.
1210 	 */
1211 	VOP_UNLOCK(imgp->vp);
1212 
1213 	/*
1214 	 * Decide whether to enable randomization of user mappings.
1215 	 * First, reset user preferences for the setid binaries.
1216 	 * Then, account for the support of the randomization by the
1217 	 * ABI, by user preferences, and make special treatment for
1218 	 * PIE binaries.
1219 	 */
1220 	if (imgp->credential_setid) {
1221 		PROC_LOCK(imgp->proc);
1222 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1223 		PROC_UNLOCK(imgp->proc);
1224 	}
1225 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1226 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1227 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1228 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1229 		    ("et_dyn_addr == RAND and !ASLR"));
1230 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1231 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1232 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1233 		imgp->map_flags |= MAP_ASLR;
1234 		/*
1235 		 * If user does not care about sbrk, utilize the bss
1236 		 * grow region for mappings as well.  We can select
1237 		 * the base for the image anywere and still not suffer
1238 		 * from the fragmentation.
1239 		 */
1240 		if (!__elfN(aslr_honor_sbrk) ||
1241 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1242 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1243 	}
1244 
1245 	if (!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0)
1246 		imgp->map_flags |= MAP_WXORX;
1247 
1248 	error = exec_new_vmspace(imgp, sv);
1249 	vmspace = imgp->proc->p_vmspace;
1250 	map = &vmspace->vm_map;
1251 
1252 	imgp->proc->p_sysent = sv;
1253 
1254 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1255 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1256 		KASSERT((map->flags & MAP_ASLR) != 0,
1257 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1258 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1259 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1260 		    /* reserve half of the address space to interpreter */
1261 		    maxv / 2, 1UL << flsl(maxalign));
1262 	}
1263 
1264 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1265 	if (error != 0)
1266 		goto ret;
1267 
1268 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1269 	if (error != 0)
1270 		goto ret;
1271 
1272 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1273 	if (error != 0)
1274 		goto ret;
1275 
1276 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1277 
1278 	/*
1279 	 * We load the dynamic linker where a userland call
1280 	 * to mmap(0, ...) would put it.  The rationale behind this
1281 	 * calculation is that it leaves room for the heap to grow to
1282 	 * its maximum allowed size.
1283 	 */
1284 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1285 	    RLIMIT_DATA));
1286 	if ((map->flags & MAP_ASLR) != 0) {
1287 		maxv1 = maxv / 2 + addr / 2;
1288 		MPASS(maxv1 >= addr);	/* No overflow */
1289 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1290 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1291 	} else {
1292 		map->anon_loc = addr;
1293 	}
1294 
1295 	imgp->entry_addr = entry;
1296 
1297 	if (interp != NULL) {
1298 		VOP_UNLOCK(imgp->vp);
1299 		if ((map->flags & MAP_ASLR) != 0) {
1300 			/* Assume that interpeter fits into 1/4 of AS */
1301 			maxv1 = maxv / 2 + addr / 2;
1302 			MPASS(maxv1 >= addr);	/* No overflow */
1303 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1304 			    maxv1, PAGE_SIZE);
1305 		}
1306 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1307 		    &imgp->entry_addr);
1308 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1309 		if (error != 0)
1310 			goto ret;
1311 	} else
1312 		addr = et_dyn_addr;
1313 
1314 	/*
1315 	 * Construct auxargs table (used by the copyout_auxargs routine)
1316 	 */
1317 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1318 	if (elf_auxargs == NULL) {
1319 		VOP_UNLOCK(imgp->vp);
1320 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1321 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1322 	}
1323 	elf_auxargs->execfd = -1;
1324 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1325 	elf_auxargs->phent = hdr->e_phentsize;
1326 	elf_auxargs->phnum = hdr->e_phnum;
1327 	elf_auxargs->pagesz = PAGE_SIZE;
1328 	elf_auxargs->base = addr;
1329 	elf_auxargs->flags = 0;
1330 	elf_auxargs->entry = entry;
1331 	elf_auxargs->hdr_eflags = hdr->e_flags;
1332 
1333 	imgp->auxargs = elf_auxargs;
1334 	imgp->interpreted = 0;
1335 	imgp->reloc_base = addr;
1336 	imgp->proc->p_osrel = osrel;
1337 	imgp->proc->p_fctl0 = fctl0;
1338 	imgp->proc->p_elf_machine = hdr->e_machine;
1339 	imgp->proc->p_elf_flags = hdr->e_flags;
1340 
1341 ret:
1342 	if (free_interp)
1343 		free(interp, M_TEMP);
1344 	return (error);
1345 }
1346 
1347 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1348 
1349 int
1350 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1351 {
1352 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1353 	Elf_Auxinfo *argarray, *pos;
1354 	int error;
1355 
1356 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1357 	    M_WAITOK | M_ZERO);
1358 
1359 	if (args->execfd != -1)
1360 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1361 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1362 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1363 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1364 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1365 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1366 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1367 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1368 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1369 	if (imgp->execpathp != 0)
1370 		AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1371 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1372 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1373 	if (imgp->canary != 0) {
1374 		AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1375 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1376 	}
1377 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1378 	if (imgp->pagesizes != 0) {
1379 		AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1380 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1381 	}
1382 	if (imgp->sysent->sv_timekeep_base != 0) {
1383 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1384 		    imgp->sysent->sv_timekeep_base);
1385 	}
1386 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1387 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1388 	    imgp->sysent->sv_stackprot);
1389 	if (imgp->sysent->sv_hwcap != NULL)
1390 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1391 	if (imgp->sysent->sv_hwcap2 != NULL)
1392 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1393 	AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1394 	    ELF_BSDF_SIGFASTBLK : 0);
1395 	AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1396 	AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1397 	AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1398 	AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1399 	AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1400 	if (imgp->sysent->sv_fxrng_gen_base != 0)
1401 		AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1402 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1403 
1404 	free(imgp->auxargs, M_TEMP);
1405 	imgp->auxargs = NULL;
1406 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1407 
1408 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1409 	free(argarray, M_TEMP);
1410 	return (error);
1411 }
1412 
1413 int
1414 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1415 {
1416 	Elf_Addr *base;
1417 
1418 	base = (Elf_Addr *)*stack_base;
1419 	base--;
1420 	if (suword(base, imgp->args->argc) == -1)
1421 		return (EFAULT);
1422 	*stack_base = (uintptr_t)base;
1423 	return (0);
1424 }
1425 
1426 /*
1427  * Code for generating ELF core dumps.
1428  */
1429 
1430 typedef void (*segment_callback)(vm_map_entry_t, void *);
1431 
1432 /* Closure for cb_put_phdr(). */
1433 struct phdr_closure {
1434 	Elf_Phdr *phdr;		/* Program header to fill in */
1435 	Elf_Off offset;		/* Offset of segment in core file */
1436 };
1437 
1438 /* Closure for cb_size_segment(). */
1439 struct sseg_closure {
1440 	int count;		/* Count of writable segments. */
1441 	size_t size;		/* Total size of all writable segments. */
1442 };
1443 
1444 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1445 
1446 struct note_info {
1447 	int		type;		/* Note type. */
1448 	outfunc_t 	outfunc; 	/* Output function. */
1449 	void		*outarg;	/* Argument for the output function. */
1450 	size_t		outsize;	/* Output size. */
1451 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1452 };
1453 
1454 TAILQ_HEAD(note_info_list, note_info);
1455 
1456 /* Coredump output parameters. */
1457 struct coredump_params {
1458 	off_t		offset;
1459 	struct ucred	*active_cred;
1460 	struct ucred	*file_cred;
1461 	struct thread	*td;
1462 	struct vnode	*vp;
1463 	struct compressor *comp;
1464 };
1465 
1466 extern int compress_user_cores;
1467 extern int compress_user_cores_level;
1468 
1469 static void cb_put_phdr(vm_map_entry_t, void *);
1470 static void cb_size_segment(vm_map_entry_t, void *);
1471 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1472     enum uio_seg, size_t *);
1473 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1474 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1475     struct note_info_list *, size_t);
1476 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1477     size_t *);
1478 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1479 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1480 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1481 static int sbuf_drain_core_output(void *, const char *, int);
1482 
1483 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1484 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1485 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1486 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1487 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1488 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1489 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1490 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1491 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1492 static void note_procstat_files(void *, struct sbuf *, size_t *);
1493 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1494 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1495 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1496 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1497 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1498 
1499 /*
1500  * Write out a core segment to the compression stream.
1501  */
1502 static int
1503 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1504 {
1505 	u_int chunk_len;
1506 	int error;
1507 
1508 	while (len > 0) {
1509 		chunk_len = MIN(len, CORE_BUF_SIZE);
1510 
1511 		/*
1512 		 * We can get EFAULT error here.
1513 		 * In that case zero out the current chunk of the segment.
1514 		 */
1515 		error = copyin(base, buf, chunk_len);
1516 		if (error != 0)
1517 			bzero(buf, chunk_len);
1518 		error = compressor_write(p->comp, buf, chunk_len);
1519 		if (error != 0)
1520 			break;
1521 		base += chunk_len;
1522 		len -= chunk_len;
1523 	}
1524 	return (error);
1525 }
1526 
1527 static int
1528 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1529 {
1530 
1531 	return (core_write((struct coredump_params *)arg, base, len, offset,
1532 	    UIO_SYSSPACE, NULL));
1533 }
1534 
1535 static int
1536 core_write(struct coredump_params *p, const void *base, size_t len,
1537     off_t offset, enum uio_seg seg, size_t *resid)
1538 {
1539 
1540 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1541 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1542 	    p->active_cred, p->file_cred, resid, p->td));
1543 }
1544 
1545 static int
1546 core_output(char *base, size_t len, off_t offset, struct coredump_params *p,
1547     void *tmpbuf)
1548 {
1549 	vm_map_t map;
1550 	struct mount *mp;
1551 	size_t resid, runlen;
1552 	int error;
1553 	bool success;
1554 
1555 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1556 	    ("%s: user address %p is not page-aligned", __func__, base));
1557 
1558 	if (p->comp != NULL)
1559 		return (compress_chunk(p, base, tmpbuf, len));
1560 
1561 	map = &p->td->td_proc->p_vmspace->vm_map;
1562 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1563 		/*
1564 		 * Attempt to page in all virtual pages in the range.  If a
1565 		 * virtual page is not backed by the pager, it is represented as
1566 		 * a hole in the file.  This can occur with zero-filled
1567 		 * anonymous memory or truncated files, for example.
1568 		 */
1569 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
1570 			error = vm_fault(map, (uintptr_t)base + runlen,
1571 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
1572 			if (runlen == 0)
1573 				success = error == KERN_SUCCESS;
1574 			else if ((error == KERN_SUCCESS) != success)
1575 				break;
1576 		}
1577 
1578 		if (success) {
1579 			error = core_write(p, base, runlen, offset,
1580 			    UIO_USERSPACE, &resid);
1581 			if (error != 0) {
1582 				if (error != EFAULT)
1583 					break;
1584 
1585 				/*
1586 				 * EFAULT may be returned if the user mapping
1587 				 * could not be accessed, e.g., because a mapped
1588 				 * file has been truncated.  Skip the page if no
1589 				 * progress was made, to protect against a
1590 				 * hypothetical scenario where vm_fault() was
1591 				 * successful but core_write() returns EFAULT
1592 				 * anyway.
1593 				 */
1594 				runlen -= resid;
1595 				if (runlen == 0) {
1596 					success = false;
1597 					runlen = PAGE_SIZE;
1598 				}
1599 			}
1600 		}
1601 		if (!success) {
1602 			error = vn_start_write(p->vp, &mp, V_WAIT);
1603 			if (error != 0)
1604 				break;
1605 			vn_lock(p->vp, LK_EXCLUSIVE | LK_RETRY);
1606 			error = vn_truncate_locked(p->vp, offset + runlen,
1607 			    false, p->td->td_ucred);
1608 			VOP_UNLOCK(p->vp);
1609 			vn_finished_write(mp);
1610 			if (error != 0)
1611 				break;
1612 		}
1613 	}
1614 	return (error);
1615 }
1616 
1617 /*
1618  * Drain into a core file.
1619  */
1620 static int
1621 sbuf_drain_core_output(void *arg, const char *data, int len)
1622 {
1623 	struct coredump_params *p;
1624 	int error, locked;
1625 
1626 	p = (struct coredump_params *)arg;
1627 
1628 	/*
1629 	 * Some kern_proc out routines that print to this sbuf may
1630 	 * call us with the process lock held. Draining with the
1631 	 * non-sleepable lock held is unsafe. The lock is needed for
1632 	 * those routines when dumping a live process. In our case we
1633 	 * can safely release the lock before draining and acquire
1634 	 * again after.
1635 	 */
1636 	locked = PROC_LOCKED(p->td->td_proc);
1637 	if (locked)
1638 		PROC_UNLOCK(p->td->td_proc);
1639 	if (p->comp != NULL)
1640 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1641 	else
1642 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1643 		    UIO_SYSSPACE, NULL);
1644 	if (locked)
1645 		PROC_LOCK(p->td->td_proc);
1646 	if (error != 0)
1647 		return (-error);
1648 	p->offset += len;
1649 	return (len);
1650 }
1651 
1652 int
1653 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1654 {
1655 	struct ucred *cred = td->td_ucred;
1656 	int error = 0;
1657 	struct sseg_closure seginfo;
1658 	struct note_info_list notelst;
1659 	struct coredump_params params;
1660 	struct note_info *ninfo;
1661 	void *hdr, *tmpbuf;
1662 	size_t hdrsize, notesz, coresize;
1663 
1664 	hdr = NULL;
1665 	tmpbuf = NULL;
1666 	TAILQ_INIT(&notelst);
1667 
1668 	/* Size the program segments. */
1669 	seginfo.count = 0;
1670 	seginfo.size = 0;
1671 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1672 
1673 	/*
1674 	 * Collect info about the core file header area.
1675 	 */
1676 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1677 	if (seginfo.count + 1 >= PN_XNUM)
1678 		hdrsize += sizeof(Elf_Shdr);
1679 	__elfN(prepare_notes)(td, &notelst, &notesz);
1680 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1681 
1682 	/* Set up core dump parameters. */
1683 	params.offset = 0;
1684 	params.active_cred = cred;
1685 	params.file_cred = NOCRED;
1686 	params.td = td;
1687 	params.vp = vp;
1688 	params.comp = NULL;
1689 
1690 #ifdef RACCT
1691 	if (racct_enable) {
1692 		PROC_LOCK(td->td_proc);
1693 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1694 		PROC_UNLOCK(td->td_proc);
1695 		if (error != 0) {
1696 			error = EFAULT;
1697 			goto done;
1698 		}
1699 	}
1700 #endif
1701 	if (coresize >= limit) {
1702 		error = EFAULT;
1703 		goto done;
1704 	}
1705 
1706 	/* Create a compression stream if necessary. */
1707 	if (compress_user_cores != 0) {
1708 		params.comp = compressor_init(core_compressed_write,
1709 		    compress_user_cores, CORE_BUF_SIZE,
1710 		    compress_user_cores_level, &params);
1711 		if (params.comp == NULL) {
1712 			error = EFAULT;
1713 			goto done;
1714 		}
1715 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1716         }
1717 
1718 	/*
1719 	 * Allocate memory for building the header, fill it up,
1720 	 * and write it out following the notes.
1721 	 */
1722 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1723 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1724 	    notesz);
1725 
1726 	/* Write the contents of all of the writable segments. */
1727 	if (error == 0) {
1728 		Elf_Phdr *php;
1729 		off_t offset;
1730 		int i;
1731 
1732 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1733 		offset = round_page(hdrsize + notesz);
1734 		for (i = 0; i < seginfo.count; i++) {
1735 			error = core_output((char *)(uintptr_t)php->p_vaddr,
1736 			    php->p_filesz, offset, &params, tmpbuf);
1737 			if (error != 0)
1738 				break;
1739 			offset += php->p_filesz;
1740 			php++;
1741 		}
1742 		if (error == 0 && params.comp != NULL)
1743 			error = compressor_flush(params.comp);
1744 	}
1745 	if (error) {
1746 		log(LOG_WARNING,
1747 		    "Failed to write core file for process %s (error %d)\n",
1748 		    curproc->p_comm, error);
1749 	}
1750 
1751 done:
1752 	free(tmpbuf, M_TEMP);
1753 	if (params.comp != NULL)
1754 		compressor_fini(params.comp);
1755 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1756 		TAILQ_REMOVE(&notelst, ninfo, link);
1757 		free(ninfo, M_TEMP);
1758 	}
1759 	if (hdr != NULL)
1760 		free(hdr, M_TEMP);
1761 
1762 	return (error);
1763 }
1764 
1765 /*
1766  * A callback for each_dumpable_segment() to write out the segment's
1767  * program header entry.
1768  */
1769 static void
1770 cb_put_phdr(vm_map_entry_t entry, void *closure)
1771 {
1772 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1773 	Elf_Phdr *phdr = phc->phdr;
1774 
1775 	phc->offset = round_page(phc->offset);
1776 
1777 	phdr->p_type = PT_LOAD;
1778 	phdr->p_offset = phc->offset;
1779 	phdr->p_vaddr = entry->start;
1780 	phdr->p_paddr = 0;
1781 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1782 	phdr->p_align = PAGE_SIZE;
1783 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1784 
1785 	phc->offset += phdr->p_filesz;
1786 	phc->phdr++;
1787 }
1788 
1789 /*
1790  * A callback for each_dumpable_segment() to gather information about
1791  * the number of segments and their total size.
1792  */
1793 static void
1794 cb_size_segment(vm_map_entry_t entry, void *closure)
1795 {
1796 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1797 
1798 	ssc->count++;
1799 	ssc->size += entry->end - entry->start;
1800 }
1801 
1802 /*
1803  * For each writable segment in the process's memory map, call the given
1804  * function with a pointer to the map entry and some arbitrary
1805  * caller-supplied data.
1806  */
1807 static void
1808 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1809 {
1810 	struct proc *p = td->td_proc;
1811 	vm_map_t map = &p->p_vmspace->vm_map;
1812 	vm_map_entry_t entry;
1813 	vm_object_t backing_object, object;
1814 	bool ignore_entry;
1815 
1816 	vm_map_lock_read(map);
1817 	VM_MAP_ENTRY_FOREACH(entry, map) {
1818 		/*
1819 		 * Don't dump inaccessible mappings, deal with legacy
1820 		 * coredump mode.
1821 		 *
1822 		 * Note that read-only segments related to the elf binary
1823 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1824 		 * need to arbitrarily ignore such segments.
1825 		 */
1826 		if (elf_legacy_coredump) {
1827 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1828 				continue;
1829 		} else {
1830 			if ((entry->protection & VM_PROT_ALL) == 0)
1831 				continue;
1832 		}
1833 
1834 		/*
1835 		 * Dont include memory segment in the coredump if
1836 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1837 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1838 		 * kernel map).
1839 		 */
1840 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1841 			continue;
1842 
1843 		if ((object = entry->object.vm_object) == NULL)
1844 			continue;
1845 
1846 		/* Ignore memory-mapped devices and such things. */
1847 		VM_OBJECT_RLOCK(object);
1848 		while ((backing_object = object->backing_object) != NULL) {
1849 			VM_OBJECT_RLOCK(backing_object);
1850 			VM_OBJECT_RUNLOCK(object);
1851 			object = backing_object;
1852 		}
1853 		ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1854 		VM_OBJECT_RUNLOCK(object);
1855 		if (ignore_entry)
1856 			continue;
1857 
1858 		(*func)(entry, closure);
1859 	}
1860 	vm_map_unlock_read(map);
1861 }
1862 
1863 /*
1864  * Write the core file header to the file, including padding up to
1865  * the page boundary.
1866  */
1867 static int
1868 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1869     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1870 {
1871 	struct note_info *ninfo;
1872 	struct sbuf *sb;
1873 	int error;
1874 
1875 	/* Fill in the header. */
1876 	bzero(hdr, hdrsize);
1877 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1878 
1879 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1880 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1881 	sbuf_start_section(sb, NULL);
1882 	sbuf_bcat(sb, hdr, hdrsize);
1883 	TAILQ_FOREACH(ninfo, notelst, link)
1884 	    __elfN(putnote)(ninfo, sb);
1885 	/* Align up to a page boundary for the program segments. */
1886 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1887 	error = sbuf_finish(sb);
1888 	sbuf_delete(sb);
1889 
1890 	return (error);
1891 }
1892 
1893 static void
1894 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1895     size_t *sizep)
1896 {
1897 	struct proc *p;
1898 	struct thread *thr;
1899 	size_t size;
1900 
1901 	p = td->td_proc;
1902 	size = 0;
1903 
1904 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1905 
1906 	/*
1907 	 * To have the debugger select the right thread (LWP) as the initial
1908 	 * thread, we dump the state of the thread passed to us in td first.
1909 	 * This is the thread that causes the core dump and thus likely to
1910 	 * be the right thread one wants to have selected in the debugger.
1911 	 */
1912 	thr = td;
1913 	while (thr != NULL) {
1914 		size += register_note(list, NT_PRSTATUS,
1915 		    __elfN(note_prstatus), thr);
1916 		size += register_note(list, NT_FPREGSET,
1917 		    __elfN(note_fpregset), thr);
1918 		size += register_note(list, NT_THRMISC,
1919 		    __elfN(note_thrmisc), thr);
1920 		size += register_note(list, NT_PTLWPINFO,
1921 		    __elfN(note_ptlwpinfo), thr);
1922 		size += register_note(list, -1,
1923 		    __elfN(note_threadmd), thr);
1924 
1925 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1926 		    TAILQ_NEXT(thr, td_plist);
1927 		if (thr == td)
1928 			thr = TAILQ_NEXT(thr, td_plist);
1929 	}
1930 
1931 	size += register_note(list, NT_PROCSTAT_PROC,
1932 	    __elfN(note_procstat_proc), p);
1933 	size += register_note(list, NT_PROCSTAT_FILES,
1934 	    note_procstat_files, p);
1935 	size += register_note(list, NT_PROCSTAT_VMMAP,
1936 	    note_procstat_vmmap, p);
1937 	size += register_note(list, NT_PROCSTAT_GROUPS,
1938 	    note_procstat_groups, p);
1939 	size += register_note(list, NT_PROCSTAT_UMASK,
1940 	    note_procstat_umask, p);
1941 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1942 	    note_procstat_rlimit, p);
1943 	size += register_note(list, NT_PROCSTAT_OSREL,
1944 	    note_procstat_osrel, p);
1945 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1946 	    __elfN(note_procstat_psstrings), p);
1947 	size += register_note(list, NT_PROCSTAT_AUXV,
1948 	    __elfN(note_procstat_auxv), p);
1949 
1950 	*sizep = size;
1951 }
1952 
1953 static void
1954 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1955     size_t notesz)
1956 {
1957 	Elf_Ehdr *ehdr;
1958 	Elf_Phdr *phdr;
1959 	Elf_Shdr *shdr;
1960 	struct phdr_closure phc;
1961 
1962 	ehdr = (Elf_Ehdr *)hdr;
1963 
1964 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1965 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1966 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1967 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1968 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1969 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1970 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1971 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1972 	ehdr->e_ident[EI_ABIVERSION] = 0;
1973 	ehdr->e_ident[EI_PAD] = 0;
1974 	ehdr->e_type = ET_CORE;
1975 	ehdr->e_machine = td->td_proc->p_elf_machine;
1976 	ehdr->e_version = EV_CURRENT;
1977 	ehdr->e_entry = 0;
1978 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1979 	ehdr->e_flags = td->td_proc->p_elf_flags;
1980 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1981 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1982 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1983 	ehdr->e_shstrndx = SHN_UNDEF;
1984 	if (numsegs + 1 < PN_XNUM) {
1985 		ehdr->e_phnum = numsegs + 1;
1986 		ehdr->e_shnum = 0;
1987 	} else {
1988 		ehdr->e_phnum = PN_XNUM;
1989 		ehdr->e_shnum = 1;
1990 
1991 		ehdr->e_shoff = ehdr->e_phoff +
1992 		    (numsegs + 1) * ehdr->e_phentsize;
1993 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1994 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1995 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1996 
1997 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1998 		memset(shdr, 0, sizeof(*shdr));
1999 		/*
2000 		 * A special first section is used to hold large segment and
2001 		 * section counts.  This was proposed by Sun Microsystems in
2002 		 * Solaris and has been adopted by Linux; the standard ELF
2003 		 * tools are already familiar with the technique.
2004 		 *
2005 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
2006 		 * (or 12-7 depending on the version of the document) for more
2007 		 * details.
2008 		 */
2009 		shdr->sh_type = SHT_NULL;
2010 		shdr->sh_size = ehdr->e_shnum;
2011 		shdr->sh_link = ehdr->e_shstrndx;
2012 		shdr->sh_info = numsegs + 1;
2013 	}
2014 
2015 	/*
2016 	 * Fill in the program header entries.
2017 	 */
2018 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
2019 
2020 	/* The note segement. */
2021 	phdr->p_type = PT_NOTE;
2022 	phdr->p_offset = hdrsize;
2023 	phdr->p_vaddr = 0;
2024 	phdr->p_paddr = 0;
2025 	phdr->p_filesz = notesz;
2026 	phdr->p_memsz = 0;
2027 	phdr->p_flags = PF_R;
2028 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
2029 	phdr++;
2030 
2031 	/* All the writable segments from the program. */
2032 	phc.phdr = phdr;
2033 	phc.offset = round_page(hdrsize + notesz);
2034 	each_dumpable_segment(td, cb_put_phdr, &phc);
2035 }
2036 
2037 static size_t
2038 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
2039 {
2040 	struct note_info *ninfo;
2041 	size_t size, notesize;
2042 
2043 	size = 0;
2044 	out(arg, NULL, &size);
2045 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2046 	ninfo->type = type;
2047 	ninfo->outfunc = out;
2048 	ninfo->outarg = arg;
2049 	ninfo->outsize = size;
2050 	TAILQ_INSERT_TAIL(list, ninfo, link);
2051 
2052 	if (type == -1)
2053 		return (size);
2054 
2055 	notesize = sizeof(Elf_Note) +		/* note header */
2056 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2057 						/* note name */
2058 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2059 
2060 	return (notesize);
2061 }
2062 
2063 static size_t
2064 append_note_data(const void *src, void *dst, size_t len)
2065 {
2066 	size_t padded_len;
2067 
2068 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2069 	if (dst != NULL) {
2070 		bcopy(src, dst, len);
2071 		bzero((char *)dst + len, padded_len - len);
2072 	}
2073 	return (padded_len);
2074 }
2075 
2076 size_t
2077 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2078 {
2079 	Elf_Note *note;
2080 	char *buf;
2081 	size_t notesize;
2082 
2083 	buf = dst;
2084 	if (buf != NULL) {
2085 		note = (Elf_Note *)buf;
2086 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2087 		note->n_descsz = size;
2088 		note->n_type = type;
2089 		buf += sizeof(*note);
2090 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2091 		    sizeof(FREEBSD_ABI_VENDOR));
2092 		append_note_data(src, buf, size);
2093 		if (descp != NULL)
2094 			*descp = buf;
2095 	}
2096 
2097 	notesize = sizeof(Elf_Note) +		/* note header */
2098 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2099 						/* note name */
2100 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2101 
2102 	return (notesize);
2103 }
2104 
2105 static void
2106 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2107 {
2108 	Elf_Note note;
2109 	ssize_t old_len, sect_len;
2110 	size_t new_len, descsz, i;
2111 
2112 	if (ninfo->type == -1) {
2113 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2114 		return;
2115 	}
2116 
2117 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2118 	note.n_descsz = ninfo->outsize;
2119 	note.n_type = ninfo->type;
2120 
2121 	sbuf_bcat(sb, &note, sizeof(note));
2122 	sbuf_start_section(sb, &old_len);
2123 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2124 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2125 	if (note.n_descsz == 0)
2126 		return;
2127 	sbuf_start_section(sb, &old_len);
2128 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2129 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2130 	if (sect_len < 0)
2131 		return;
2132 
2133 	new_len = (size_t)sect_len;
2134 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2135 	if (new_len < descsz) {
2136 		/*
2137 		 * It is expected that individual note emitters will correctly
2138 		 * predict their expected output size and fill up to that size
2139 		 * themselves, padding in a format-specific way if needed.
2140 		 * However, in case they don't, just do it here with zeros.
2141 		 */
2142 		for (i = 0; i < descsz - new_len; i++)
2143 			sbuf_putc(sb, 0);
2144 	} else if (new_len > descsz) {
2145 		/*
2146 		 * We can't always truncate sb -- we may have drained some
2147 		 * of it already.
2148 		 */
2149 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2150 		    "read it (%zu > %zu).  Since it is longer than "
2151 		    "expected, this coredump's notes are corrupt.  THIS "
2152 		    "IS A BUG in the note_procstat routine for type %u.\n",
2153 		    __func__, (unsigned)note.n_type, new_len, descsz,
2154 		    (unsigned)note.n_type));
2155 	}
2156 }
2157 
2158 /*
2159  * Miscellaneous note out functions.
2160  */
2161 
2162 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2163 #include <compat/freebsd32/freebsd32.h>
2164 #include <compat/freebsd32/freebsd32_signal.h>
2165 
2166 typedef struct prstatus32 elf_prstatus_t;
2167 typedef struct prpsinfo32 elf_prpsinfo_t;
2168 typedef struct fpreg32 elf_prfpregset_t;
2169 typedef struct fpreg32 elf_fpregset_t;
2170 typedef struct reg32 elf_gregset_t;
2171 typedef struct thrmisc32 elf_thrmisc_t;
2172 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2173 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2174 typedef uint32_t elf_ps_strings_t;
2175 #else
2176 typedef prstatus_t elf_prstatus_t;
2177 typedef prpsinfo_t elf_prpsinfo_t;
2178 typedef prfpregset_t elf_prfpregset_t;
2179 typedef prfpregset_t elf_fpregset_t;
2180 typedef gregset_t elf_gregset_t;
2181 typedef thrmisc_t elf_thrmisc_t;
2182 #define ELF_KERN_PROC_MASK	0
2183 typedef struct kinfo_proc elf_kinfo_proc_t;
2184 typedef vm_offset_t elf_ps_strings_t;
2185 #endif
2186 
2187 static void
2188 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2189 {
2190 	struct sbuf sbarg;
2191 	size_t len;
2192 	char *cp, *end;
2193 	struct proc *p;
2194 	elf_prpsinfo_t *psinfo;
2195 	int error;
2196 
2197 	p = (struct proc *)arg;
2198 	if (sb != NULL) {
2199 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2200 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2201 		psinfo->pr_version = PRPSINFO_VERSION;
2202 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2203 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2204 		PROC_LOCK(p);
2205 		if (p->p_args != NULL) {
2206 			len = sizeof(psinfo->pr_psargs) - 1;
2207 			if (len > p->p_args->ar_length)
2208 				len = p->p_args->ar_length;
2209 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2210 			PROC_UNLOCK(p);
2211 			error = 0;
2212 		} else {
2213 			_PHOLD(p);
2214 			PROC_UNLOCK(p);
2215 			sbuf_new(&sbarg, psinfo->pr_psargs,
2216 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2217 			error = proc_getargv(curthread, p, &sbarg);
2218 			PRELE(p);
2219 			if (sbuf_finish(&sbarg) == 0)
2220 				len = sbuf_len(&sbarg) - 1;
2221 			else
2222 				len = sizeof(psinfo->pr_psargs) - 1;
2223 			sbuf_delete(&sbarg);
2224 		}
2225 		if (error || len == 0)
2226 			strlcpy(psinfo->pr_psargs, p->p_comm,
2227 			    sizeof(psinfo->pr_psargs));
2228 		else {
2229 			KASSERT(len < sizeof(psinfo->pr_psargs),
2230 			    ("len is too long: %zu vs %zu", len,
2231 			    sizeof(psinfo->pr_psargs)));
2232 			cp = psinfo->pr_psargs;
2233 			end = cp + len - 1;
2234 			for (;;) {
2235 				cp = memchr(cp, '\0', end - cp);
2236 				if (cp == NULL)
2237 					break;
2238 				*cp = ' ';
2239 			}
2240 		}
2241 		psinfo->pr_pid = p->p_pid;
2242 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2243 		free(psinfo, M_TEMP);
2244 	}
2245 	*sizep = sizeof(*psinfo);
2246 }
2247 
2248 static void
2249 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2250 {
2251 	struct thread *td;
2252 	elf_prstatus_t *status;
2253 
2254 	td = (struct thread *)arg;
2255 	if (sb != NULL) {
2256 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2257 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2258 		status->pr_version = PRSTATUS_VERSION;
2259 		status->pr_statussz = sizeof(elf_prstatus_t);
2260 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2261 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2262 		status->pr_osreldate = osreldate;
2263 		status->pr_cursig = td->td_proc->p_sig;
2264 		status->pr_pid = td->td_tid;
2265 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2266 		fill_regs32(td, &status->pr_reg);
2267 #else
2268 		fill_regs(td, &status->pr_reg);
2269 #endif
2270 		sbuf_bcat(sb, status, sizeof(*status));
2271 		free(status, M_TEMP);
2272 	}
2273 	*sizep = sizeof(*status);
2274 }
2275 
2276 static void
2277 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2278 {
2279 	struct thread *td;
2280 	elf_prfpregset_t *fpregset;
2281 
2282 	td = (struct thread *)arg;
2283 	if (sb != NULL) {
2284 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2285 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2286 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2287 		fill_fpregs32(td, fpregset);
2288 #else
2289 		fill_fpregs(td, fpregset);
2290 #endif
2291 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2292 		free(fpregset, M_TEMP);
2293 	}
2294 	*sizep = sizeof(*fpregset);
2295 }
2296 
2297 static void
2298 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2299 {
2300 	struct thread *td;
2301 	elf_thrmisc_t thrmisc;
2302 
2303 	td = (struct thread *)arg;
2304 	if (sb != NULL) {
2305 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2306 		bzero(&thrmisc, sizeof(thrmisc));
2307 		strcpy(thrmisc.pr_tname, td->td_name);
2308 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2309 	}
2310 	*sizep = sizeof(thrmisc);
2311 }
2312 
2313 static void
2314 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2315 {
2316 	struct thread *td;
2317 	size_t size;
2318 	int structsize;
2319 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2320 	struct ptrace_lwpinfo32 pl;
2321 #else
2322 	struct ptrace_lwpinfo pl;
2323 #endif
2324 
2325 	td = (struct thread *)arg;
2326 	size = sizeof(structsize) + sizeof(pl);
2327 	if (sb != NULL) {
2328 		KASSERT(*sizep == size, ("invalid size"));
2329 		structsize = sizeof(pl);
2330 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2331 		bzero(&pl, sizeof(pl));
2332 		pl.pl_lwpid = td->td_tid;
2333 		pl.pl_event = PL_EVENT_NONE;
2334 		pl.pl_sigmask = td->td_sigmask;
2335 		pl.pl_siglist = td->td_siglist;
2336 		if (td->td_si.si_signo != 0) {
2337 			pl.pl_event = PL_EVENT_SIGNAL;
2338 			pl.pl_flags |= PL_FLAG_SI;
2339 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2340 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2341 #else
2342 			pl.pl_siginfo = td->td_si;
2343 #endif
2344 		}
2345 		strcpy(pl.pl_tdname, td->td_name);
2346 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2347 		sbuf_bcat(sb, &pl, sizeof(pl));
2348 	}
2349 	*sizep = size;
2350 }
2351 
2352 /*
2353  * Allow for MD specific notes, as well as any MD
2354  * specific preparations for writing MI notes.
2355  */
2356 static void
2357 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2358 {
2359 	struct thread *td;
2360 	void *buf;
2361 	size_t size;
2362 
2363 	td = (struct thread *)arg;
2364 	size = *sizep;
2365 	if (size != 0 && sb != NULL)
2366 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2367 	else
2368 		buf = NULL;
2369 	size = 0;
2370 	__elfN(dump_thread)(td, buf, &size);
2371 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2372 	if (size != 0 && sb != NULL)
2373 		sbuf_bcat(sb, buf, size);
2374 	free(buf, M_TEMP);
2375 	*sizep = size;
2376 }
2377 
2378 #ifdef KINFO_PROC_SIZE
2379 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2380 #endif
2381 
2382 static void
2383 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2384 {
2385 	struct proc *p;
2386 	size_t size;
2387 	int structsize;
2388 
2389 	p = (struct proc *)arg;
2390 	size = sizeof(structsize) + p->p_numthreads *
2391 	    sizeof(elf_kinfo_proc_t);
2392 
2393 	if (sb != NULL) {
2394 		KASSERT(*sizep == size, ("invalid size"));
2395 		structsize = sizeof(elf_kinfo_proc_t);
2396 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2397 		sx_slock(&proctree_lock);
2398 		PROC_LOCK(p);
2399 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2400 		sx_sunlock(&proctree_lock);
2401 	}
2402 	*sizep = size;
2403 }
2404 
2405 #ifdef KINFO_FILE_SIZE
2406 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2407 #endif
2408 
2409 static void
2410 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2411 {
2412 	struct proc *p;
2413 	size_t size, sect_sz, i;
2414 	ssize_t start_len, sect_len;
2415 	int structsize, filedesc_flags;
2416 
2417 	if (coredump_pack_fileinfo)
2418 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2419 	else
2420 		filedesc_flags = 0;
2421 
2422 	p = (struct proc *)arg;
2423 	structsize = sizeof(struct kinfo_file);
2424 	if (sb == NULL) {
2425 		size = 0;
2426 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2427 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2428 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2429 		PROC_LOCK(p);
2430 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2431 		sbuf_finish(sb);
2432 		sbuf_delete(sb);
2433 		*sizep = size;
2434 	} else {
2435 		sbuf_start_section(sb, &start_len);
2436 
2437 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2438 		PROC_LOCK(p);
2439 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2440 		    filedesc_flags);
2441 
2442 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2443 		if (sect_len < 0)
2444 			return;
2445 		sect_sz = sect_len;
2446 
2447 		KASSERT(sect_sz <= *sizep,
2448 		    ("kern_proc_filedesc_out did not respect maxlen; "
2449 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2450 		     sect_sz - sizeof(structsize)));
2451 
2452 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2453 			sbuf_putc(sb, 0);
2454 	}
2455 }
2456 
2457 #ifdef KINFO_VMENTRY_SIZE
2458 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2459 #endif
2460 
2461 static void
2462 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2463 {
2464 	struct proc *p;
2465 	size_t size;
2466 	int structsize, vmmap_flags;
2467 
2468 	if (coredump_pack_vmmapinfo)
2469 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2470 	else
2471 		vmmap_flags = 0;
2472 
2473 	p = (struct proc *)arg;
2474 	structsize = sizeof(struct kinfo_vmentry);
2475 	if (sb == NULL) {
2476 		size = 0;
2477 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2478 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2479 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2480 		PROC_LOCK(p);
2481 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2482 		sbuf_finish(sb);
2483 		sbuf_delete(sb);
2484 		*sizep = size;
2485 	} else {
2486 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2487 		PROC_LOCK(p);
2488 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2489 		    vmmap_flags);
2490 	}
2491 }
2492 
2493 static void
2494 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2495 {
2496 	struct proc *p;
2497 	size_t size;
2498 	int structsize;
2499 
2500 	p = (struct proc *)arg;
2501 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2502 	if (sb != NULL) {
2503 		KASSERT(*sizep == size, ("invalid size"));
2504 		structsize = sizeof(gid_t);
2505 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2506 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2507 		    sizeof(gid_t));
2508 	}
2509 	*sizep = size;
2510 }
2511 
2512 static void
2513 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2514 {
2515 	struct proc *p;
2516 	size_t size;
2517 	int structsize;
2518 
2519 	p = (struct proc *)arg;
2520 	size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2521 	if (sb != NULL) {
2522 		KASSERT(*sizep == size, ("invalid size"));
2523 		structsize = sizeof(p->p_pd->pd_cmask);
2524 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2525 		sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2526 	}
2527 	*sizep = size;
2528 }
2529 
2530 static void
2531 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2532 {
2533 	struct proc *p;
2534 	struct rlimit rlim[RLIM_NLIMITS];
2535 	size_t size;
2536 	int structsize, i;
2537 
2538 	p = (struct proc *)arg;
2539 	size = sizeof(structsize) + sizeof(rlim);
2540 	if (sb != NULL) {
2541 		KASSERT(*sizep == size, ("invalid size"));
2542 		structsize = sizeof(rlim);
2543 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2544 		PROC_LOCK(p);
2545 		for (i = 0; i < RLIM_NLIMITS; i++)
2546 			lim_rlimit_proc(p, i, &rlim[i]);
2547 		PROC_UNLOCK(p);
2548 		sbuf_bcat(sb, rlim, sizeof(rlim));
2549 	}
2550 	*sizep = size;
2551 }
2552 
2553 static void
2554 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2555 {
2556 	struct proc *p;
2557 	size_t size;
2558 	int structsize;
2559 
2560 	p = (struct proc *)arg;
2561 	size = sizeof(structsize) + sizeof(p->p_osrel);
2562 	if (sb != NULL) {
2563 		KASSERT(*sizep == size, ("invalid size"));
2564 		structsize = sizeof(p->p_osrel);
2565 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2566 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2567 	}
2568 	*sizep = size;
2569 }
2570 
2571 static void
2572 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2573 {
2574 	struct proc *p;
2575 	elf_ps_strings_t ps_strings;
2576 	size_t size;
2577 	int structsize;
2578 
2579 	p = (struct proc *)arg;
2580 	size = sizeof(structsize) + sizeof(ps_strings);
2581 	if (sb != NULL) {
2582 		KASSERT(*sizep == size, ("invalid size"));
2583 		structsize = sizeof(ps_strings);
2584 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2585 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2586 #else
2587 		ps_strings = p->p_sysent->sv_psstrings;
2588 #endif
2589 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2590 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2591 	}
2592 	*sizep = size;
2593 }
2594 
2595 static void
2596 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2597 {
2598 	struct proc *p;
2599 	size_t size;
2600 	int structsize;
2601 
2602 	p = (struct proc *)arg;
2603 	if (sb == NULL) {
2604 		size = 0;
2605 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2606 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2607 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2608 		PHOLD(p);
2609 		proc_getauxv(curthread, p, sb);
2610 		PRELE(p);
2611 		sbuf_finish(sb);
2612 		sbuf_delete(sb);
2613 		*sizep = size;
2614 	} else {
2615 		structsize = sizeof(Elf_Auxinfo);
2616 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2617 		PHOLD(p);
2618 		proc_getauxv(curthread, p, sb);
2619 		PRELE(p);
2620 	}
2621 }
2622 
2623 static boolean_t
2624 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2625     const char *note_vendor, const Elf_Phdr *pnote,
2626     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2627 {
2628 	const Elf_Note *note, *note0, *note_end;
2629 	const char *note_name;
2630 	char *buf;
2631 	int i, error;
2632 	boolean_t res;
2633 
2634 	/* We need some limit, might as well use PAGE_SIZE. */
2635 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2636 		return (FALSE);
2637 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2638 	if (pnote->p_offset > PAGE_SIZE ||
2639 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2640 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2641 		if (buf == NULL) {
2642 			VOP_UNLOCK(imgp->vp);
2643 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2644 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2645 		}
2646 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2647 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2648 		    curthread->td_ucred, NOCRED, NULL, curthread);
2649 		if (error != 0) {
2650 			uprintf("i/o error PT_NOTE\n");
2651 			goto retf;
2652 		}
2653 		note = note0 = (const Elf_Note *)buf;
2654 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2655 	} else {
2656 		note = note0 = (const Elf_Note *)(imgp->image_header +
2657 		    pnote->p_offset);
2658 		note_end = (const Elf_Note *)(imgp->image_header +
2659 		    pnote->p_offset + pnote->p_filesz);
2660 		buf = NULL;
2661 	}
2662 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2663 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2664 		    (const char *)note < sizeof(Elf_Note)) {
2665 			goto retf;
2666 		}
2667 		if (note->n_namesz != checknote->n_namesz ||
2668 		    note->n_descsz != checknote->n_descsz ||
2669 		    note->n_type != checknote->n_type)
2670 			goto nextnote;
2671 		note_name = (const char *)(note + 1);
2672 		if (note_name + checknote->n_namesz >=
2673 		    (const char *)note_end || strncmp(note_vendor,
2674 		    note_name, checknote->n_namesz) != 0)
2675 			goto nextnote;
2676 
2677 		if (cb(note, cb_arg, &res))
2678 			goto ret;
2679 nextnote:
2680 		note = (const Elf_Note *)((const char *)(note + 1) +
2681 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2682 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2683 	}
2684 retf:
2685 	res = FALSE;
2686 ret:
2687 	free(buf, M_TEMP);
2688 	return (res);
2689 }
2690 
2691 struct brandnote_cb_arg {
2692 	Elf_Brandnote *brandnote;
2693 	int32_t *osrel;
2694 };
2695 
2696 static boolean_t
2697 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2698 {
2699 	struct brandnote_cb_arg *arg;
2700 
2701 	arg = arg0;
2702 
2703 	/*
2704 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2705 	 * necessary.
2706 	 */
2707 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2708 	    arg->brandnote->trans_osrel != NULL ?
2709 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2710 
2711 	return (TRUE);
2712 }
2713 
2714 static Elf_Note fctl_note = {
2715 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2716 	.n_descsz = sizeof(uint32_t),
2717 	.n_type = NT_FREEBSD_FEATURE_CTL,
2718 };
2719 
2720 struct fctl_cb_arg {
2721 	boolean_t *has_fctl0;
2722 	uint32_t *fctl0;
2723 };
2724 
2725 static boolean_t
2726 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2727 {
2728 	struct fctl_cb_arg *arg;
2729 	const Elf32_Word *desc;
2730 	uintptr_t p;
2731 
2732 	arg = arg0;
2733 	p = (uintptr_t)(note + 1);
2734 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2735 	desc = (const Elf32_Word *)p;
2736 	*arg->has_fctl0 = TRUE;
2737 	*arg->fctl0 = desc[0];
2738 	return (TRUE);
2739 }
2740 
2741 /*
2742  * Try to find the appropriate ABI-note section for checknote, fetch
2743  * the osreldate and feature control flags for binary from the ELF
2744  * OSABI-note.  Only the first page of the image is searched, the same
2745  * as for headers.
2746  */
2747 static boolean_t
2748 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2749     int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2750 {
2751 	const Elf_Phdr *phdr;
2752 	const Elf_Ehdr *hdr;
2753 	struct brandnote_cb_arg b_arg;
2754 	struct fctl_cb_arg f_arg;
2755 	int i, j;
2756 
2757 	hdr = (const Elf_Ehdr *)imgp->image_header;
2758 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2759 	b_arg.brandnote = brandnote;
2760 	b_arg.osrel = osrel;
2761 	f_arg.has_fctl0 = has_fctl0;
2762 	f_arg.fctl0 = fctl0;
2763 
2764 	for (i = 0; i < hdr->e_phnum; i++) {
2765 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2766 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2767 		    &b_arg)) {
2768 			for (j = 0; j < hdr->e_phnum; j++) {
2769 				if (phdr[j].p_type == PT_NOTE &&
2770 				    __elfN(parse_notes)(imgp, &fctl_note,
2771 				    FREEBSD_ABI_VENDOR, &phdr[j],
2772 				    note_fctl_cb, &f_arg))
2773 					break;
2774 			}
2775 			return (TRUE);
2776 		}
2777 	}
2778 	return (FALSE);
2779 
2780 }
2781 
2782 /*
2783  * Tell kern_execve.c about it, with a little help from the linker.
2784  */
2785 static struct execsw __elfN(execsw) = {
2786 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2787 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2788 };
2789 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2790 
2791 static vm_prot_t
2792 __elfN(trans_prot)(Elf_Word flags)
2793 {
2794 	vm_prot_t prot;
2795 
2796 	prot = 0;
2797 	if (flags & PF_X)
2798 		prot |= VM_PROT_EXECUTE;
2799 	if (flags & PF_W)
2800 		prot |= VM_PROT_WRITE;
2801 	if (flags & PF_R)
2802 		prot |= VM_PROT_READ;
2803 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2804 	if (i386_read_exec && (flags & PF_R))
2805 		prot |= VM_PROT_EXECUTE;
2806 #endif
2807 	return (prot);
2808 }
2809 
2810 static Elf_Word
2811 __elfN(untrans_prot)(vm_prot_t prot)
2812 {
2813 	Elf_Word flags;
2814 
2815 	flags = 0;
2816 	if (prot & VM_PROT_EXECUTE)
2817 		flags |= PF_X;
2818 	if (prot & VM_PROT_READ)
2819 		flags |= PF_R;
2820 	if (prot & VM_PROT_WRITE)
2821 		flags |= PF_W;
2822 	return (flags);
2823 }
2824 
2825 void
2826 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2827 {
2828 	uintptr_t range, rbase, gap;
2829 	int pct;
2830 
2831 	pct = __elfN(aslr_stack_gap);
2832 	if (pct == 0)
2833 		return;
2834 	if (pct > 50)
2835 		pct = 50;
2836 	range = imgp->eff_stack_sz * pct / 100;
2837 	arc4rand(&rbase, sizeof(rbase), 0);
2838 	gap = rbase % range;
2839 	gap &= ~(sizeof(u_long) - 1);
2840 	*stack_base -= gap;
2841 }
2842