xref: /freebsd/sys/kern/imgact_elf.c (revision 8c27c5541e95164a37ac1beb7bd4ef43429aed55)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2017 Dell EMC
5  * Copyright (c) 2000-2001, 2003 David O'Brien
6  * Copyright (c) 1995-1996 Søren Schmidt
7  * Copyright (c) 1996 Peter Wemm
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_capsicum.h"
38 
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mman.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
54 #include <sys/proc.h>
55 #include <sys/procfs.h>
56 #include <sys/ptrace.h>
57 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sbuf.h>
61 #include <sys/sf_buf.h>
62 #include <sys/smp.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
65 #include <sys/stat.h>
66 #include <sys/sx.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
73 #include <sys/user.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
85 
86 #define ELF_NOTE_ROUNDSIZE	4
87 #define OLD_EI_BRAND	8
88 
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91     const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93     u_long *entry);
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98     int32_t *osrel);
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101     Elf_Brandnote *checknote, int32_t *osrel, uint32_t *fctl0);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
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 Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
194 
195 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
196 
197 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
198 
199 Elf_Brandnote __elfN(freebsd_brandnote) = {
200 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
201 	.hdr.n_descsz	= sizeof(int32_t),
202 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
203 	.vendor		= FREEBSD_ABI_VENDOR,
204 	.flags		= BN_TRANSLATE_OSREL,
205 	.trans_osrel	= __elfN(freebsd_trans_osrel)
206 };
207 
208 static bool
209 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
210 {
211 	uintptr_t p;
212 
213 	p = (uintptr_t)(note + 1);
214 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
215 	*osrel = *(const int32_t *)(p);
216 
217 	return (true);
218 }
219 
220 static const char GNU_ABI_VENDOR[] = "GNU";
221 static int GNU_KFREEBSD_ABI_DESC = 3;
222 
223 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
224 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
225 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
226 	.hdr.n_type	= 1,
227 	.vendor		= GNU_ABI_VENDOR,
228 	.flags		= BN_TRANSLATE_OSREL,
229 	.trans_osrel	= kfreebsd_trans_osrel
230 };
231 
232 static bool
233 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
234 {
235 	const Elf32_Word *desc;
236 	uintptr_t p;
237 
238 	p = (uintptr_t)(note + 1);
239 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
240 
241 	desc = (const Elf32_Word *)p;
242 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
243 		return (false);
244 
245 	/*
246 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
247 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
248 	 */
249 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
250 
251 	return (true);
252 }
253 
254 int
255 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
256 {
257 	int i;
258 
259 	for (i = 0; i < MAX_BRANDS; i++) {
260 		if (elf_brand_list[i] == NULL) {
261 			elf_brand_list[i] = entry;
262 			break;
263 		}
264 	}
265 	if (i == MAX_BRANDS) {
266 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
267 			__func__, entry);
268 		return (-1);
269 	}
270 	return (0);
271 }
272 
273 int
274 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
275 {
276 	int i;
277 
278 	for (i = 0; i < MAX_BRANDS; i++) {
279 		if (elf_brand_list[i] == entry) {
280 			elf_brand_list[i] = NULL;
281 			break;
282 		}
283 	}
284 	if (i == MAX_BRANDS)
285 		return (-1);
286 	return (0);
287 }
288 
289 int
290 __elfN(brand_inuse)(Elf_Brandinfo *entry)
291 {
292 	struct proc *p;
293 	int rval = FALSE;
294 
295 	sx_slock(&allproc_lock);
296 	FOREACH_PROC_IN_SYSTEM(p) {
297 		if (p->p_sysent == entry->sysvec) {
298 			rval = TRUE;
299 			break;
300 		}
301 	}
302 	sx_sunlock(&allproc_lock);
303 
304 	return (rval);
305 }
306 
307 static Elf_Brandinfo *
308 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
309     int32_t *osrel, uint32_t *fctl0)
310 {
311 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
312 	Elf_Brandinfo *bi, *bi_m;
313 	boolean_t ret;
314 	int i, interp_name_len;
315 
316 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
317 
318 	/*
319 	 * We support four types of branding -- (1) the ELF EI_OSABI field
320 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
321 	 * branding w/in the ELF header, (3) path of the `interp_path'
322 	 * field, and (4) the ".note.ABI-tag" ELF section.
323 	 */
324 
325 	/* Look for an ".note.ABI-tag" ELF section */
326 	bi_m = NULL;
327 	for (i = 0; i < MAX_BRANDS; i++) {
328 		bi = elf_brand_list[i];
329 		if (bi == NULL)
330 			continue;
331 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
332 			continue;
333 		if (hdr->e_machine == bi->machine && (bi->flags &
334 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
335 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
336 			    fctl0);
337 			/* Give brand a chance to veto check_note's guess */
338 			if (ret && bi->header_supported)
339 				ret = bi->header_supported(imgp);
340 			/*
341 			 * If note checker claimed the binary, but the
342 			 * interpreter path in the image does not
343 			 * match default one for the brand, try to
344 			 * search for other brands with the same
345 			 * interpreter.  Either there is better brand
346 			 * with the right interpreter, or, failing
347 			 * this, we return first brand which accepted
348 			 * our note and, optionally, header.
349 			 */
350 			if (ret && bi_m == NULL && interp != NULL &&
351 			    (bi->interp_path == NULL ||
352 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
353 			    strncmp(interp, bi->interp_path, interp_name_len)
354 			    != 0))) {
355 				bi_m = bi;
356 				ret = 0;
357 			}
358 			if (ret)
359 				return (bi);
360 		}
361 	}
362 	if (bi_m != NULL)
363 		return (bi_m);
364 
365 	/* If the executable has a brand, search for it in the brand list. */
366 	for (i = 0; i < MAX_BRANDS; i++) {
367 		bi = elf_brand_list[i];
368 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
369 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
370 			continue;
371 		if (hdr->e_machine == bi->machine &&
372 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
373 		    (bi->compat_3_brand != NULL &&
374 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
375 		    bi->compat_3_brand) == 0))) {
376 			/* Looks good, but give brand a chance to veto */
377 			if (bi->header_supported == NULL ||
378 			    bi->header_supported(imgp)) {
379 				/*
380 				 * Again, prefer strictly matching
381 				 * interpreter path.
382 				 */
383 				if (interp_name_len == 0 &&
384 				    bi->interp_path == NULL)
385 					return (bi);
386 				if (bi->interp_path != NULL &&
387 				    strlen(bi->interp_path) + 1 ==
388 				    interp_name_len && strncmp(interp,
389 				    bi->interp_path, interp_name_len) == 0)
390 					return (bi);
391 				if (bi_m == NULL)
392 					bi_m = bi;
393 			}
394 		}
395 	}
396 	if (bi_m != NULL)
397 		return (bi_m);
398 
399 	/* No known brand, see if the header is recognized by any brand */
400 	for (i = 0; i < MAX_BRANDS; i++) {
401 		bi = elf_brand_list[i];
402 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
403 		    bi->header_supported == NULL)
404 			continue;
405 		if (hdr->e_machine == bi->machine) {
406 			ret = bi->header_supported(imgp);
407 			if (ret)
408 				return (bi);
409 		}
410 	}
411 
412 	/* Lacking a known brand, search for a recognized interpreter. */
413 	if (interp != NULL) {
414 		for (i = 0; i < MAX_BRANDS; i++) {
415 			bi = elf_brand_list[i];
416 			if (bi == NULL || (bi->flags &
417 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
418 			    != 0)
419 				continue;
420 			if (hdr->e_machine == bi->machine &&
421 			    bi->interp_path != NULL &&
422 			    /* ELF image p_filesz includes terminating zero */
423 			    strlen(bi->interp_path) + 1 == interp_name_len &&
424 			    strncmp(interp, bi->interp_path, interp_name_len)
425 			    == 0 && (bi->header_supported == NULL ||
426 			    bi->header_supported(imgp)))
427 				return (bi);
428 		}
429 	}
430 
431 	/* Lacking a recognized interpreter, try the default 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) != 0 ||
435 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
436 			continue;
437 		if (hdr->e_machine == bi->machine &&
438 		    __elfN(fallback_brand) == bi->brand &&
439 		    (bi->header_supported == NULL ||
440 		    bi->header_supported(imgp)))
441 			return (bi);
442 	}
443 	return (NULL);
444 }
445 
446 static int
447 __elfN(check_header)(const Elf_Ehdr *hdr)
448 {
449 	Elf_Brandinfo *bi;
450 	int i;
451 
452 	if (!IS_ELF(*hdr) ||
453 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
454 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
455 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
456 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
457 	    hdr->e_version != ELF_TARG_VER)
458 		return (ENOEXEC);
459 
460 	/*
461 	 * Make sure we have at least one brand for this machine.
462 	 */
463 
464 	for (i = 0; i < MAX_BRANDS; i++) {
465 		bi = elf_brand_list[i];
466 		if (bi != NULL && bi->machine == hdr->e_machine)
467 			break;
468 	}
469 	if (i == MAX_BRANDS)
470 		return (ENOEXEC);
471 
472 	return (0);
473 }
474 
475 static int
476 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
477     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
478 {
479 	struct sf_buf *sf;
480 	int error;
481 	vm_offset_t off;
482 
483 	/*
484 	 * Create the page if it doesn't exist yet. Ignore errors.
485 	 */
486 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
487 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
488 
489 	/*
490 	 * Find the page from the underlying object.
491 	 */
492 	if (object != NULL) {
493 		sf = vm_imgact_map_page(object, offset);
494 		if (sf == NULL)
495 			return (KERN_FAILURE);
496 		off = offset - trunc_page(offset);
497 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
498 		    end - start);
499 		vm_imgact_unmap_page(sf);
500 		if (error != 0)
501 			return (KERN_FAILURE);
502 	}
503 
504 	return (KERN_SUCCESS);
505 }
506 
507 static int
508 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
509     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
510     int cow)
511 {
512 	struct sf_buf *sf;
513 	vm_offset_t off;
514 	vm_size_t sz;
515 	int error, locked, rv;
516 
517 	if (start != trunc_page(start)) {
518 		rv = __elfN(map_partial)(map, object, offset, start,
519 		    round_page(start), prot);
520 		if (rv != KERN_SUCCESS)
521 			return (rv);
522 		offset += round_page(start) - start;
523 		start = round_page(start);
524 	}
525 	if (end != round_page(end)) {
526 		rv = __elfN(map_partial)(map, object, offset +
527 		    trunc_page(end) - start, trunc_page(end), end, prot);
528 		if (rv != KERN_SUCCESS)
529 			return (rv);
530 		end = trunc_page(end);
531 	}
532 	if (start >= end)
533 		return (KERN_SUCCESS);
534 	if ((offset & PAGE_MASK) != 0) {
535 		/*
536 		 * The mapping is not page aligned.  This means that we have
537 		 * to copy the data.
538 		 */
539 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
540 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
541 		if (rv != KERN_SUCCESS)
542 			return (rv);
543 		if (object == NULL)
544 			return (KERN_SUCCESS);
545 		for (; start < end; start += sz) {
546 			sf = vm_imgact_map_page(object, offset);
547 			if (sf == NULL)
548 				return (KERN_FAILURE);
549 			off = offset - trunc_page(offset);
550 			sz = end - start;
551 			if (sz > PAGE_SIZE - off)
552 				sz = PAGE_SIZE - off;
553 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
554 			    (caddr_t)start, sz);
555 			vm_imgact_unmap_page(sf);
556 			if (error != 0)
557 				return (KERN_FAILURE);
558 			offset += sz;
559 		}
560 	} else {
561 		vm_object_reference(object);
562 		rv = vm_map_fixed(map, object, offset, start, end - start,
563 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
564 		    (object != NULL ? MAP_VN_EXEC : 0));
565 		if (rv != KERN_SUCCESS) {
566 			locked = VOP_ISLOCKED(imgp->vp);
567 			VOP_UNLOCK(imgp->vp);
568 			vm_object_deallocate(object);
569 			vn_lock(imgp->vp, locked | LK_RETRY);
570 			return (rv);
571 		} else if (object != NULL) {
572 			MPASS(imgp->vp->v_object == object);
573 			VOP_SET_TEXT_CHECKED(imgp->vp);
574 		}
575 	}
576 	return (KERN_SUCCESS);
577 }
578 
579 static int
580 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
581     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
582 {
583 	struct sf_buf *sf;
584 	size_t map_len;
585 	vm_map_t map;
586 	vm_object_t object;
587 	vm_offset_t map_addr;
588 	int error, rv, cow;
589 	size_t copy_len;
590 	vm_ooffset_t file_addr;
591 
592 	/*
593 	 * It's necessary to fail if the filsz + offset taken from the
594 	 * header is greater than the actual file pager object's size.
595 	 * If we were to allow this, then the vm_map_find() below would
596 	 * walk right off the end of the file object and into the ether.
597 	 *
598 	 * While I'm here, might as well check for something else that
599 	 * is invalid: filsz cannot be greater than memsz.
600 	 */
601 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
602 	    filsz > memsz) {
603 		uprintf("elf_load_section: truncated ELF file\n");
604 		return (ENOEXEC);
605 	}
606 
607 	object = imgp->object;
608 	map = &imgp->proc->p_vmspace->vm_map;
609 	map_addr = trunc_page((vm_offset_t)vmaddr);
610 	file_addr = trunc_page(offset);
611 
612 	/*
613 	 * We have two choices.  We can either clear the data in the last page
614 	 * of an oversized mapping, or we can start the anon mapping a page
615 	 * early and copy the initialized data into that first page.  We
616 	 * choose the second.
617 	 */
618 	if (filsz == 0)
619 		map_len = 0;
620 	else if (memsz > filsz)
621 		map_len = trunc_page(offset + filsz) - file_addr;
622 	else
623 		map_len = round_page(offset + filsz) - file_addr;
624 
625 	if (map_len != 0) {
626 		/* cow flags: don't dump readonly sections in core */
627 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
628 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
629 
630 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
631 		    map_addr, map_addr + map_len, prot, cow);
632 		if (rv != KERN_SUCCESS)
633 			return (EINVAL);
634 
635 		/* we can stop now if we've covered it all */
636 		if (memsz == filsz)
637 			return (0);
638 	}
639 
640 	/*
641 	 * We have to get the remaining bit of the file into the first part
642 	 * of the oversized map segment.  This is normally because the .data
643 	 * segment in the file is extended to provide bss.  It's a neat idea
644 	 * to try and save a page, but it's a pain in the behind to implement.
645 	 */
646 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
647 	    filsz);
648 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
649 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
650 
651 	/* This had damn well better be true! */
652 	if (map_len != 0) {
653 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
654 		    map_addr + map_len, prot, 0);
655 		if (rv != KERN_SUCCESS)
656 			return (EINVAL);
657 	}
658 
659 	if (copy_len != 0) {
660 		sf = vm_imgact_map_page(object, offset + filsz);
661 		if (sf == NULL)
662 			return (EIO);
663 
664 		/* send the page fragment to user space */
665 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
666 		    copy_len);
667 		vm_imgact_unmap_page(sf);
668 		if (error != 0)
669 			return (error);
670 	}
671 
672 	/*
673 	 * Remove write access to the page if it was only granted by map_insert
674 	 * to allow copyout.
675 	 */
676 	if ((prot & VM_PROT_WRITE) == 0)
677 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
678 		    map_len), prot, FALSE);
679 
680 	return (0);
681 }
682 
683 static int
684 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
685     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
686 {
687 	vm_prot_t prot;
688 	u_long base_addr;
689 	bool first;
690 	int error, i;
691 
692 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
693 
694 	base_addr = 0;
695 	first = true;
696 
697 	for (i = 0; i < hdr->e_phnum; i++) {
698 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
699 			continue;
700 
701 		/* Loadable segment */
702 		prot = __elfN(trans_prot)(phdr[i].p_flags);
703 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
704 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
705 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
706 		if (error != 0)
707 			return (error);
708 
709 		/*
710 		 * Establish the base address if this is the first segment.
711 		 */
712 		if (first) {
713   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
714 			first = false;
715 		}
716 	}
717 
718 	if (base_addrp != NULL)
719 		*base_addrp = base_addr;
720 
721 	return (0);
722 }
723 
724 /*
725  * Load the file "file" into memory.  It may be either a shared object
726  * or an executable.
727  *
728  * The "addr" reference parameter is in/out.  On entry, it specifies
729  * the address where a shared object should be loaded.  If the file is
730  * an executable, this value is ignored.  On exit, "addr" specifies
731  * where the file was actually loaded.
732  *
733  * The "entry" reference parameter is out only.  On exit, it specifies
734  * the entry point for the loaded file.
735  */
736 static int
737 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
738 	u_long *entry)
739 {
740 	struct {
741 		struct nameidata nd;
742 		struct vattr attr;
743 		struct image_params image_params;
744 	} *tempdata;
745 	const Elf_Ehdr *hdr = NULL;
746 	const Elf_Phdr *phdr = NULL;
747 	struct nameidata *nd;
748 	struct vattr *attr;
749 	struct image_params *imgp;
750 	u_long rbase;
751 	u_long base_addr = 0;
752 	int error;
753 
754 #ifdef CAPABILITY_MODE
755 	/*
756 	 * XXXJA: This check can go away once we are sufficiently confident
757 	 * that the checks in namei() are correct.
758 	 */
759 	if (IN_CAPABILITY_MODE(curthread))
760 		return (ECAPMODE);
761 #endif
762 
763 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
764 	nd = &tempdata->nd;
765 	attr = &tempdata->attr;
766 	imgp = &tempdata->image_params;
767 
768 	/*
769 	 * Initialize part of the common data
770 	 */
771 	imgp->proc = p;
772 	imgp->attr = attr;
773 
774 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
775 	    UIO_SYSSPACE, file, curthread);
776 	if ((error = namei(nd)) != 0) {
777 		nd->ni_vp = NULL;
778 		goto fail;
779 	}
780 	NDFREE(nd, NDF_ONLY_PNBUF);
781 	imgp->vp = nd->ni_vp;
782 
783 	/*
784 	 * Check permissions, modes, uid, etc on the file, and "open" it.
785 	 */
786 	error = exec_check_permissions(imgp);
787 	if (error)
788 		goto fail;
789 
790 	error = exec_map_first_page(imgp);
791 	if (error)
792 		goto fail;
793 
794 	imgp->object = nd->ni_vp->v_object;
795 
796 	hdr = (const Elf_Ehdr *)imgp->image_header;
797 	if ((error = __elfN(check_header)(hdr)) != 0)
798 		goto fail;
799 	if (hdr->e_type == ET_DYN)
800 		rbase = *addr;
801 	else if (hdr->e_type == ET_EXEC)
802 		rbase = 0;
803 	else {
804 		error = ENOEXEC;
805 		goto fail;
806 	}
807 
808 	/* Only support headers that fit within first page for now      */
809 	if ((hdr->e_phoff > PAGE_SIZE) ||
810 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
811 		error = ENOEXEC;
812 		goto fail;
813 	}
814 
815 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
816 	if (!aligned(phdr, Elf_Addr)) {
817 		error = ENOEXEC;
818 		goto fail;
819 	}
820 
821 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
822 	if (error != 0)
823 		goto fail;
824 
825 	*addr = base_addr;
826 	*entry = (unsigned long)hdr->e_entry + rbase;
827 
828 fail:
829 	if (imgp->firstpage)
830 		exec_unmap_first_page(imgp);
831 
832 	if (nd->ni_vp) {
833 		if (imgp->textset)
834 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
835 		vput(nd->ni_vp);
836 	}
837 	free(tempdata, M_TEMP);
838 
839 	return (error);
840 }
841 
842 static u_long
843 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
844     u_int align)
845 {
846 	u_long rbase, res;
847 
848 	MPASS(vm_map_min(map) <= minv);
849 	MPASS(maxv <= vm_map_max(map));
850 	MPASS(minv < maxv);
851 	MPASS(minv + align < maxv);
852 	arc4rand(&rbase, sizeof(rbase), 0);
853 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
854 	res &= ~((u_long)align - 1);
855 	if (res >= maxv)
856 		res -= align;
857 	KASSERT(res >= minv,
858 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
859 	    res, minv, maxv, rbase));
860 	KASSERT(res < maxv,
861 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
862 	    res, maxv, minv, rbase));
863 	return (res);
864 }
865 
866 static int
867 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
868     const Elf_Phdr *phdr, u_long et_dyn_addr)
869 {
870 	struct vmspace *vmspace;
871 	const char *err_str;
872 	u_long text_size, data_size, total_size, text_addr, data_addr;
873 	u_long seg_size, seg_addr;
874 	int i;
875 
876 	err_str = NULL;
877 	text_size = data_size = total_size = text_addr = data_addr = 0;
878 
879 	for (i = 0; i < hdr->e_phnum; i++) {
880 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
881 			continue;
882 
883 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
884 		seg_size = round_page(phdr[i].p_memsz +
885 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
886 
887 		/*
888 		 * Make the largest executable segment the official
889 		 * text segment and all others data.
890 		 *
891 		 * Note that obreak() assumes that data_addr + data_size == end
892 		 * of data load area, and the ELF file format expects segments
893 		 * to be sorted by address.  If multiple data segments exist,
894 		 * the last one will be used.
895 		 */
896 
897 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
898 			text_size = seg_size;
899 			text_addr = seg_addr;
900 		} else {
901 			data_size = seg_size;
902 			data_addr = seg_addr;
903 		}
904 		total_size += seg_size;
905 	}
906 
907 	if (data_addr == 0 && data_size == 0) {
908 		data_addr = text_addr;
909 		data_size = text_size;
910 	}
911 
912 	/*
913 	 * Check limits.  It should be safe to check the
914 	 * limits after loading the segments since we do
915 	 * not actually fault in all the segments pages.
916 	 */
917 	PROC_LOCK(imgp->proc);
918 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
919 		err_str = "Data segment size exceeds process limit";
920 	else if (text_size > maxtsiz)
921 		err_str = "Text segment size exceeds system limit";
922 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
923 		err_str = "Total segment size exceeds process limit";
924 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
925 		err_str = "Data segment size exceeds resource limit";
926 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
927 		err_str = "Total segment size exceeds resource limit";
928 	PROC_UNLOCK(imgp->proc);
929 	if (err_str != NULL) {
930 		uprintf("%s\n", err_str);
931 		return (ENOMEM);
932 	}
933 
934 	vmspace = imgp->proc->p_vmspace;
935 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
936 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
937 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
938 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
939 
940 	return (0);
941 }
942 
943 static int
944 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
945     char **interpp, bool *free_interpp)
946 {
947 	struct thread *td;
948 	char *interp;
949 	int error, interp_name_len;
950 
951 	KASSERT(phdr->p_type == PT_INTERP,
952 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
953 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
954 
955 	td = curthread;
956 
957 	/* Path to interpreter */
958 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
959 		uprintf("Invalid PT_INTERP\n");
960 		return (ENOEXEC);
961 	}
962 
963 	interp_name_len = phdr->p_filesz;
964 	if (phdr->p_offset > PAGE_SIZE ||
965 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
966 		/*
967 		 * The vnode lock might be needed by the pagedaemon to
968 		 * clean pages owned by the vnode.  Do not allow sleep
969 		 * waiting for memory with the vnode locked, instead
970 		 * try non-sleepable allocation first, and if it
971 		 * fails, go to the slow path were we drop the lock
972 		 * and do M_WAITOK.  A text reference prevents
973 		 * modifications to the vnode content.
974 		 */
975 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
976 		if (interp == NULL) {
977 			VOP_UNLOCK(imgp->vp);
978 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
979 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
980 		}
981 
982 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
983 		    interp_name_len, phdr->p_offset,
984 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
985 		    NOCRED, NULL, td);
986 		if (error != 0) {
987 			free(interp, M_TEMP);
988 			uprintf("i/o error PT_INTERP %d\n", error);
989 			return (error);
990 		}
991 		interp[interp_name_len] = '\0';
992 
993 		*interpp = interp;
994 		*free_interpp = true;
995 		return (0);
996 	}
997 
998 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
999 	if (interp[interp_name_len - 1] != '\0') {
1000 		uprintf("Invalid PT_INTERP\n");
1001 		return (ENOEXEC);
1002 	}
1003 
1004 	*interpp = interp;
1005 	*free_interpp = false;
1006 	return (0);
1007 }
1008 
1009 static int
1010 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1011     const char *interp, u_long *addr, u_long *entry)
1012 {
1013 	char *path;
1014 	int error;
1015 
1016 	if (brand_info->emul_path != NULL &&
1017 	    brand_info->emul_path[0] != '\0') {
1018 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1019 		snprintf(path, MAXPATHLEN, "%s%s",
1020 		    brand_info->emul_path, interp);
1021 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1022 		free(path, M_TEMP);
1023 		if (error == 0)
1024 			return (0);
1025 	}
1026 
1027 	if (brand_info->interp_newpath != NULL &&
1028 	    (brand_info->interp_path == NULL ||
1029 	    strcmp(interp, brand_info->interp_path) == 0)) {
1030 		error = __elfN(load_file)(imgp->proc,
1031 		    brand_info->interp_newpath, addr, entry);
1032 		if (error == 0)
1033 			return (0);
1034 	}
1035 
1036 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1037 	if (error == 0)
1038 		return (0);
1039 
1040 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1041 	return (error);
1042 }
1043 
1044 /*
1045  * Impossible et_dyn_addr initial value indicating that the real base
1046  * must be calculated later with some randomization applied.
1047  */
1048 #define	ET_DYN_ADDR_RAND	1
1049 
1050 static int
1051 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1052 {
1053 	struct thread *td;
1054 	const Elf_Ehdr *hdr;
1055 	const Elf_Phdr *phdr;
1056 	Elf_Auxargs *elf_auxargs;
1057 	struct vmspace *vmspace;
1058 	vm_map_t map;
1059 	char *interp;
1060 	Elf_Brandinfo *brand_info;
1061 	struct sysentvec *sv;
1062 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1063 	u_long maxalign, mapsz, maxv, maxv1;
1064 	uint32_t fctl0;
1065 	int32_t osrel;
1066 	bool free_interp;
1067 	int error, i, n;
1068 
1069 	hdr = (const Elf_Ehdr *)imgp->image_header;
1070 
1071 	/*
1072 	 * Do we have a valid ELF header ?
1073 	 *
1074 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1075 	 * if particular brand doesn't support it.
1076 	 */
1077 	if (__elfN(check_header)(hdr) != 0 ||
1078 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1079 		return (-1);
1080 
1081 	/*
1082 	 * From here on down, we return an errno, not -1, as we've
1083 	 * detected an ELF file.
1084 	 */
1085 
1086 	if ((hdr->e_phoff > PAGE_SIZE) ||
1087 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1088 		/* Only support headers in first page for now */
1089 		uprintf("Program headers not in the first page\n");
1090 		return (ENOEXEC);
1091 	}
1092 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1093 	if (!aligned(phdr, Elf_Addr)) {
1094 		uprintf("Unaligned program headers\n");
1095 		return (ENOEXEC);
1096 	}
1097 
1098 	n = error = 0;
1099 	baddr = 0;
1100 	osrel = 0;
1101 	fctl0 = 0;
1102 	entry = proghdr = 0;
1103 	interp = NULL;
1104 	free_interp = false;
1105 	td = curthread;
1106 	maxalign = PAGE_SIZE;
1107 	mapsz = 0;
1108 
1109 	for (i = 0; i < hdr->e_phnum; i++) {
1110 		switch (phdr[i].p_type) {
1111 		case PT_LOAD:
1112 			if (n == 0)
1113 				baddr = phdr[i].p_vaddr;
1114 			if (phdr[i].p_align > maxalign)
1115 				maxalign = phdr[i].p_align;
1116 			mapsz += phdr[i].p_memsz;
1117 			n++;
1118 
1119 			/*
1120 			 * If this segment contains the program headers,
1121 			 * remember their virtual address for the AT_PHDR
1122 			 * aux entry. Static binaries don't usually include
1123 			 * a PT_PHDR entry.
1124 			 */
1125 			if (phdr[i].p_offset == 0 &&
1126 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1127 				<= phdr[i].p_filesz)
1128 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1129 			break;
1130 		case PT_INTERP:
1131 			/* Path to interpreter */
1132 			if (interp != NULL) {
1133 				uprintf("Multiple PT_INTERP headers\n");
1134 				error = ENOEXEC;
1135 				goto ret;
1136 			}
1137 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1138 			    &free_interp);
1139 			if (error != 0)
1140 				goto ret;
1141 			break;
1142 		case PT_GNU_STACK:
1143 			if (__elfN(nxstack))
1144 				imgp->stack_prot =
1145 				    __elfN(trans_prot)(phdr[i].p_flags);
1146 			imgp->stack_sz = phdr[i].p_memsz;
1147 			break;
1148 		case PT_PHDR: 	/* Program header table info */
1149 			proghdr = phdr[i].p_vaddr;
1150 			break;
1151 		}
1152 	}
1153 
1154 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1155 	if (brand_info == NULL) {
1156 		uprintf("ELF binary type \"%u\" not known.\n",
1157 		    hdr->e_ident[EI_OSABI]);
1158 		error = ENOEXEC;
1159 		goto ret;
1160 	}
1161 	sv = brand_info->sysvec;
1162 	et_dyn_addr = 0;
1163 	if (hdr->e_type == ET_DYN) {
1164 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1165 			uprintf("Cannot execute shared object\n");
1166 			error = ENOEXEC;
1167 			goto ret;
1168 		}
1169 		/*
1170 		 * Honour the base load address from the dso if it is
1171 		 * non-zero for some reason.
1172 		 */
1173 		if (baddr == 0) {
1174 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1175 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1176 				et_dyn_addr = __elfN(pie_base);
1177 			else if ((__elfN(pie_aslr_enabled) &&
1178 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1179 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1180 				et_dyn_addr = ET_DYN_ADDR_RAND;
1181 			else
1182 				et_dyn_addr = __elfN(pie_base);
1183 		}
1184 	}
1185 
1186 	/*
1187 	 * Avoid a possible deadlock if the current address space is destroyed
1188 	 * and that address space maps the locked vnode.  In the common case,
1189 	 * the locked vnode's v_usecount is decremented but remains greater
1190 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1191 	 * However, in cases where the vnode lock is external, such as nullfs,
1192 	 * v_usecount may become zero.
1193 	 *
1194 	 * The VV_TEXT flag prevents modifications to the executable while
1195 	 * the vnode is unlocked.
1196 	 */
1197 	VOP_UNLOCK(imgp->vp);
1198 
1199 	/*
1200 	 * Decide whether to enable randomization of user mappings.
1201 	 * First, reset user preferences for the setid binaries.
1202 	 * Then, account for the support of the randomization by the
1203 	 * ABI, by user preferences, and make special treatment for
1204 	 * PIE binaries.
1205 	 */
1206 	if (imgp->credential_setid) {
1207 		PROC_LOCK(imgp->proc);
1208 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1209 		PROC_UNLOCK(imgp->proc);
1210 	}
1211 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1212 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1213 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1214 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1215 		    ("et_dyn_addr == RAND and !ASLR"));
1216 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1217 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1218 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1219 		imgp->map_flags |= MAP_ASLR;
1220 		/*
1221 		 * If user does not care about sbrk, utilize the bss
1222 		 * grow region for mappings as well.  We can select
1223 		 * the base for the image anywere and still not suffer
1224 		 * from the fragmentation.
1225 		 */
1226 		if (!__elfN(aslr_honor_sbrk) ||
1227 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1228 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1229 	}
1230 
1231 	error = exec_new_vmspace(imgp, sv);
1232 	vmspace = imgp->proc->p_vmspace;
1233 	map = &vmspace->vm_map;
1234 
1235 	imgp->proc->p_sysent = sv;
1236 
1237 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1238 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1239 		KASSERT((map->flags & MAP_ASLR) != 0,
1240 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1241 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1242 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1243 		    /* reserve half of the address space to interpreter */
1244 		    maxv / 2, 1UL << flsl(maxalign));
1245 	}
1246 
1247 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1248 	if (error != 0)
1249 		goto ret;
1250 
1251 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1252 	if (error != 0)
1253 		goto ret;
1254 
1255 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1256 	if (error != 0)
1257 		goto ret;
1258 
1259 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1260 
1261 	/*
1262 	 * We load the dynamic linker where a userland call
1263 	 * to mmap(0, ...) would put it.  The rationale behind this
1264 	 * calculation is that it leaves room for the heap to grow to
1265 	 * its maximum allowed size.
1266 	 */
1267 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1268 	    RLIMIT_DATA));
1269 	if ((map->flags & MAP_ASLR) != 0) {
1270 		maxv1 = maxv / 2 + addr / 2;
1271 		MPASS(maxv1 >= addr);	/* No overflow */
1272 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1273 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1274 	} else {
1275 		map->anon_loc = addr;
1276 	}
1277 
1278 	imgp->entry_addr = entry;
1279 
1280 	if (interp != NULL) {
1281 		VOP_UNLOCK(imgp->vp);
1282 		if ((map->flags & MAP_ASLR) != 0) {
1283 			/* Assume that interpeter fits into 1/4 of AS */
1284 			maxv1 = maxv / 2 + addr / 2;
1285 			MPASS(maxv1 >= addr);	/* No overflow */
1286 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1287 			    maxv1, PAGE_SIZE);
1288 		}
1289 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1290 		    &imgp->entry_addr);
1291 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1292 		if (error != 0)
1293 			goto ret;
1294 	} else
1295 		addr = et_dyn_addr;
1296 
1297 	/*
1298 	 * Construct auxargs table (used by the copyout_auxargs routine)
1299 	 */
1300 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1301 	if (elf_auxargs == NULL) {
1302 		VOP_UNLOCK(imgp->vp);
1303 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1304 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1305 	}
1306 	elf_auxargs->execfd = -1;
1307 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1308 	elf_auxargs->phent = hdr->e_phentsize;
1309 	elf_auxargs->phnum = hdr->e_phnum;
1310 	elf_auxargs->pagesz = PAGE_SIZE;
1311 	elf_auxargs->base = addr;
1312 	elf_auxargs->flags = 0;
1313 	elf_auxargs->entry = entry;
1314 	elf_auxargs->hdr_eflags = hdr->e_flags;
1315 
1316 	imgp->auxargs = elf_auxargs;
1317 	imgp->interpreted = 0;
1318 	imgp->reloc_base = addr;
1319 	imgp->proc->p_osrel = osrel;
1320 	imgp->proc->p_fctl0 = fctl0;
1321 	imgp->proc->p_elf_machine = hdr->e_machine;
1322 	imgp->proc->p_elf_flags = hdr->e_flags;
1323 
1324 ret:
1325 	if (free_interp)
1326 		free(interp, M_TEMP);
1327 	return (error);
1328 }
1329 
1330 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1331 
1332 int
1333 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1334 {
1335 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1336 	Elf_Auxinfo *argarray, *pos;
1337 	int error;
1338 
1339 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1340 	    M_WAITOK | M_ZERO);
1341 
1342 	if (args->execfd != -1)
1343 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1344 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1345 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1346 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1347 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1348 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1349 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1350 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1351 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1352 	if (imgp->execpathp != 0)
1353 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1354 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1355 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1356 	if (imgp->canary != 0) {
1357 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1358 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1359 	}
1360 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1361 	if (imgp->pagesizes != 0) {
1362 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1363 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1364 	}
1365 	if (imgp->sysent->sv_timekeep_base != 0) {
1366 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1367 		    imgp->sysent->sv_timekeep_base);
1368 	}
1369 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1370 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1371 	    imgp->sysent->sv_stackprot);
1372 	if (imgp->sysent->sv_hwcap != NULL)
1373 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1374 	if (imgp->sysent->sv_hwcap2 != NULL)
1375 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1376 	AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1377 	    ELF_BSDF_SIGFASTBLK : 0);
1378 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1379 
1380 	free(imgp->auxargs, M_TEMP);
1381 	imgp->auxargs = NULL;
1382 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1383 
1384 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1385 	free(argarray, M_TEMP);
1386 	return (error);
1387 }
1388 
1389 int
1390 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1391 {
1392 	Elf_Addr *base;
1393 
1394 	base = (Elf_Addr *)*stack_base;
1395 	base--;
1396 	if (suword(base, imgp->args->argc) == -1)
1397 		return (EFAULT);
1398 	*stack_base = (uintptr_t)base;
1399 	return (0);
1400 }
1401 
1402 /*
1403  * Code for generating ELF core dumps.
1404  */
1405 
1406 typedef void (*segment_callback)(vm_map_entry_t, void *);
1407 
1408 /* Closure for cb_put_phdr(). */
1409 struct phdr_closure {
1410 	Elf_Phdr *phdr;		/* Program header to fill in */
1411 	Elf_Off offset;		/* Offset of segment in core file */
1412 };
1413 
1414 /* Closure for cb_size_segment(). */
1415 struct sseg_closure {
1416 	int count;		/* Count of writable segments. */
1417 	size_t size;		/* Total size of all writable segments. */
1418 };
1419 
1420 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1421 
1422 struct note_info {
1423 	int		type;		/* Note type. */
1424 	outfunc_t 	outfunc; 	/* Output function. */
1425 	void		*outarg;	/* Argument for the output function. */
1426 	size_t		outsize;	/* Output size. */
1427 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1428 };
1429 
1430 TAILQ_HEAD(note_info_list, note_info);
1431 
1432 /* Coredump output parameters. */
1433 struct coredump_params {
1434 	off_t		offset;
1435 	struct ucred	*active_cred;
1436 	struct ucred	*file_cred;
1437 	struct thread	*td;
1438 	struct vnode	*vp;
1439 	struct compressor *comp;
1440 };
1441 
1442 extern int compress_user_cores;
1443 extern int compress_user_cores_level;
1444 
1445 static void cb_put_phdr(vm_map_entry_t, void *);
1446 static void cb_size_segment(vm_map_entry_t, void *);
1447 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1448     enum uio_seg);
1449 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1450 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1451     struct note_info_list *, size_t);
1452 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1453     size_t *);
1454 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1455 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1456 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1457 static int sbuf_drain_core_output(void *, const char *, int);
1458 
1459 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1460 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1461 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1462 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1463 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1464 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1465 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1466 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1467 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1468 static void note_procstat_files(void *, struct sbuf *, size_t *);
1469 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1470 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1471 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1472 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1473 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1474 
1475 /*
1476  * Write out a core segment to the compression stream.
1477  */
1478 static int
1479 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1480 {
1481 	u_int chunk_len;
1482 	int error;
1483 
1484 	while (len > 0) {
1485 		chunk_len = MIN(len, CORE_BUF_SIZE);
1486 
1487 		/*
1488 		 * We can get EFAULT error here.
1489 		 * In that case zero out the current chunk of the segment.
1490 		 */
1491 		error = copyin(base, buf, chunk_len);
1492 		if (error != 0)
1493 			bzero(buf, chunk_len);
1494 		error = compressor_write(p->comp, buf, chunk_len);
1495 		if (error != 0)
1496 			break;
1497 		base += chunk_len;
1498 		len -= chunk_len;
1499 	}
1500 	return (error);
1501 }
1502 
1503 static int
1504 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1505 {
1506 
1507 	return (core_write((struct coredump_params *)arg, base, len, offset,
1508 	    UIO_SYSSPACE));
1509 }
1510 
1511 static int
1512 core_write(struct coredump_params *p, const void *base, size_t len,
1513     off_t offset, enum uio_seg seg)
1514 {
1515 
1516 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1517 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1518 	    p->active_cred, p->file_cred, NULL, p->td));
1519 }
1520 
1521 static int
1522 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1523     void *tmpbuf)
1524 {
1525 	int error;
1526 
1527 	if (p->comp != NULL)
1528 		return (compress_chunk(p, base, tmpbuf, len));
1529 
1530 	/*
1531 	 * EFAULT is a non-fatal error that we can get, for example,
1532 	 * if the segment is backed by a file but extends beyond its
1533 	 * end.
1534 	 */
1535 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1536 	if (error == EFAULT) {
1537 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1538 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1539 		    "for process %s\n", base, len, offset, curproc->p_comm);
1540 
1541 		/*
1542 		 * Write a "real" zero byte at the end of the target region
1543 		 * in the case this is the last segment.
1544 		 * The intermediate space will be implicitly zero-filled.
1545 		 */
1546 		error = core_write(p, zero_region, 1, offset + len - 1,
1547 		    UIO_SYSSPACE);
1548 	}
1549 	return (error);
1550 }
1551 
1552 /*
1553  * Drain into a core file.
1554  */
1555 static int
1556 sbuf_drain_core_output(void *arg, const char *data, int len)
1557 {
1558 	struct coredump_params *p;
1559 	int error, locked;
1560 
1561 	p = (struct coredump_params *)arg;
1562 
1563 	/*
1564 	 * Some kern_proc out routines that print to this sbuf may
1565 	 * call us with the process lock held. Draining with the
1566 	 * non-sleepable lock held is unsafe. The lock is needed for
1567 	 * those routines when dumping a live process. In our case we
1568 	 * can safely release the lock before draining and acquire
1569 	 * again after.
1570 	 */
1571 	locked = PROC_LOCKED(p->td->td_proc);
1572 	if (locked)
1573 		PROC_UNLOCK(p->td->td_proc);
1574 	if (p->comp != NULL)
1575 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1576 	else
1577 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1578 		    UIO_SYSSPACE);
1579 	if (locked)
1580 		PROC_LOCK(p->td->td_proc);
1581 	if (error != 0)
1582 		return (-error);
1583 	p->offset += len;
1584 	return (len);
1585 }
1586 
1587 int
1588 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1589 {
1590 	struct ucred *cred = td->td_ucred;
1591 	int error = 0;
1592 	struct sseg_closure seginfo;
1593 	struct note_info_list notelst;
1594 	struct coredump_params params;
1595 	struct note_info *ninfo;
1596 	void *hdr, *tmpbuf;
1597 	size_t hdrsize, notesz, coresize;
1598 
1599 	hdr = NULL;
1600 	tmpbuf = NULL;
1601 	TAILQ_INIT(&notelst);
1602 
1603 	/* Size the program segments. */
1604 	seginfo.count = 0;
1605 	seginfo.size = 0;
1606 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1607 
1608 	/*
1609 	 * Collect info about the core file header area.
1610 	 */
1611 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1612 	if (seginfo.count + 1 >= PN_XNUM)
1613 		hdrsize += sizeof(Elf_Shdr);
1614 	__elfN(prepare_notes)(td, &notelst, &notesz);
1615 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1616 
1617 	/* Set up core dump parameters. */
1618 	params.offset = 0;
1619 	params.active_cred = cred;
1620 	params.file_cred = NOCRED;
1621 	params.td = td;
1622 	params.vp = vp;
1623 	params.comp = NULL;
1624 
1625 #ifdef RACCT
1626 	if (racct_enable) {
1627 		PROC_LOCK(td->td_proc);
1628 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1629 		PROC_UNLOCK(td->td_proc);
1630 		if (error != 0) {
1631 			error = EFAULT;
1632 			goto done;
1633 		}
1634 	}
1635 #endif
1636 	if (coresize >= limit) {
1637 		error = EFAULT;
1638 		goto done;
1639 	}
1640 
1641 	/* Create a compression stream if necessary. */
1642 	if (compress_user_cores != 0) {
1643 		params.comp = compressor_init(core_compressed_write,
1644 		    compress_user_cores, CORE_BUF_SIZE,
1645 		    compress_user_cores_level, &params);
1646 		if (params.comp == NULL) {
1647 			error = EFAULT;
1648 			goto done;
1649 		}
1650 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1651         }
1652 
1653 	/*
1654 	 * Allocate memory for building the header, fill it up,
1655 	 * and write it out following the notes.
1656 	 */
1657 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1658 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1659 	    notesz);
1660 
1661 	/* Write the contents of all of the writable segments. */
1662 	if (error == 0) {
1663 		Elf_Phdr *php;
1664 		off_t offset;
1665 		int i;
1666 
1667 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1668 		offset = round_page(hdrsize + notesz);
1669 		for (i = 0; i < seginfo.count; i++) {
1670 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1671 			    php->p_filesz, offset, &params, tmpbuf);
1672 			if (error != 0)
1673 				break;
1674 			offset += php->p_filesz;
1675 			php++;
1676 		}
1677 		if (error == 0 && params.comp != NULL)
1678 			error = compressor_flush(params.comp);
1679 	}
1680 	if (error) {
1681 		log(LOG_WARNING,
1682 		    "Failed to write core file for process %s (error %d)\n",
1683 		    curproc->p_comm, error);
1684 	}
1685 
1686 done:
1687 	free(tmpbuf, M_TEMP);
1688 	if (params.comp != NULL)
1689 		compressor_fini(params.comp);
1690 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1691 		TAILQ_REMOVE(&notelst, ninfo, link);
1692 		free(ninfo, M_TEMP);
1693 	}
1694 	if (hdr != NULL)
1695 		free(hdr, M_TEMP);
1696 
1697 	return (error);
1698 }
1699 
1700 /*
1701  * A callback for each_dumpable_segment() to write out the segment's
1702  * program header entry.
1703  */
1704 static void
1705 cb_put_phdr(vm_map_entry_t entry, void *closure)
1706 {
1707 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1708 	Elf_Phdr *phdr = phc->phdr;
1709 
1710 	phc->offset = round_page(phc->offset);
1711 
1712 	phdr->p_type = PT_LOAD;
1713 	phdr->p_offset = phc->offset;
1714 	phdr->p_vaddr = entry->start;
1715 	phdr->p_paddr = 0;
1716 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1717 	phdr->p_align = PAGE_SIZE;
1718 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1719 
1720 	phc->offset += phdr->p_filesz;
1721 	phc->phdr++;
1722 }
1723 
1724 /*
1725  * A callback for each_dumpable_segment() to gather information about
1726  * the number of segments and their total size.
1727  */
1728 static void
1729 cb_size_segment(vm_map_entry_t entry, void *closure)
1730 {
1731 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1732 
1733 	ssc->count++;
1734 	ssc->size += entry->end - entry->start;
1735 }
1736 
1737 /*
1738  * For each writable segment in the process's memory map, call the given
1739  * function with a pointer to the map entry and some arbitrary
1740  * caller-supplied data.
1741  */
1742 static void
1743 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1744 {
1745 	struct proc *p = td->td_proc;
1746 	vm_map_t map = &p->p_vmspace->vm_map;
1747 	vm_map_entry_t entry;
1748 	vm_object_t backing_object, object;
1749 	boolean_t ignore_entry;
1750 
1751 	vm_map_lock_read(map);
1752 	VM_MAP_ENTRY_FOREACH(entry, map) {
1753 		/*
1754 		 * Don't dump inaccessible mappings, deal with legacy
1755 		 * coredump mode.
1756 		 *
1757 		 * Note that read-only segments related to the elf binary
1758 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1759 		 * need to arbitrarily ignore such segments.
1760 		 */
1761 		if (elf_legacy_coredump) {
1762 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1763 				continue;
1764 		} else {
1765 			if ((entry->protection & VM_PROT_ALL) == 0)
1766 				continue;
1767 		}
1768 
1769 		/*
1770 		 * Dont include memory segment in the coredump if
1771 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1772 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1773 		 * kernel map).
1774 		 */
1775 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1776 			continue;
1777 
1778 		if ((object = entry->object.vm_object) == NULL)
1779 			continue;
1780 
1781 		/* Ignore memory-mapped devices and such things. */
1782 		VM_OBJECT_RLOCK(object);
1783 		while ((backing_object = object->backing_object) != NULL) {
1784 			VM_OBJECT_RLOCK(backing_object);
1785 			VM_OBJECT_RUNLOCK(object);
1786 			object = backing_object;
1787 		}
1788 		ignore_entry = object->type != OBJT_DEFAULT &&
1789 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1790 		    object->type != OBJT_PHYS;
1791 		VM_OBJECT_RUNLOCK(object);
1792 		if (ignore_entry)
1793 			continue;
1794 
1795 		(*func)(entry, closure);
1796 	}
1797 	vm_map_unlock_read(map);
1798 }
1799 
1800 /*
1801  * Write the core file header to the file, including padding up to
1802  * the page boundary.
1803  */
1804 static int
1805 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1806     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1807 {
1808 	struct note_info *ninfo;
1809 	struct sbuf *sb;
1810 	int error;
1811 
1812 	/* Fill in the header. */
1813 	bzero(hdr, hdrsize);
1814 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1815 
1816 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1817 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1818 	sbuf_start_section(sb, NULL);
1819 	sbuf_bcat(sb, hdr, hdrsize);
1820 	TAILQ_FOREACH(ninfo, notelst, link)
1821 	    __elfN(putnote)(ninfo, sb);
1822 	/* Align up to a page boundary for the program segments. */
1823 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1824 	error = sbuf_finish(sb);
1825 	sbuf_delete(sb);
1826 
1827 	return (error);
1828 }
1829 
1830 static void
1831 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1832     size_t *sizep)
1833 {
1834 	struct proc *p;
1835 	struct thread *thr;
1836 	size_t size;
1837 
1838 	p = td->td_proc;
1839 	size = 0;
1840 
1841 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1842 
1843 	/*
1844 	 * To have the debugger select the right thread (LWP) as the initial
1845 	 * thread, we dump the state of the thread passed to us in td first.
1846 	 * This is the thread that causes the core dump and thus likely to
1847 	 * be the right thread one wants to have selected in the debugger.
1848 	 */
1849 	thr = td;
1850 	while (thr != NULL) {
1851 		size += register_note(list, NT_PRSTATUS,
1852 		    __elfN(note_prstatus), thr);
1853 		size += register_note(list, NT_FPREGSET,
1854 		    __elfN(note_fpregset), thr);
1855 		size += register_note(list, NT_THRMISC,
1856 		    __elfN(note_thrmisc), thr);
1857 		size += register_note(list, NT_PTLWPINFO,
1858 		    __elfN(note_ptlwpinfo), thr);
1859 		size += register_note(list, -1,
1860 		    __elfN(note_threadmd), thr);
1861 
1862 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1863 		    TAILQ_NEXT(thr, td_plist);
1864 		if (thr == td)
1865 			thr = TAILQ_NEXT(thr, td_plist);
1866 	}
1867 
1868 	size += register_note(list, NT_PROCSTAT_PROC,
1869 	    __elfN(note_procstat_proc), p);
1870 	size += register_note(list, NT_PROCSTAT_FILES,
1871 	    note_procstat_files, p);
1872 	size += register_note(list, NT_PROCSTAT_VMMAP,
1873 	    note_procstat_vmmap, p);
1874 	size += register_note(list, NT_PROCSTAT_GROUPS,
1875 	    note_procstat_groups, p);
1876 	size += register_note(list, NT_PROCSTAT_UMASK,
1877 	    note_procstat_umask, p);
1878 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1879 	    note_procstat_rlimit, p);
1880 	size += register_note(list, NT_PROCSTAT_OSREL,
1881 	    note_procstat_osrel, p);
1882 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1883 	    __elfN(note_procstat_psstrings), p);
1884 	size += register_note(list, NT_PROCSTAT_AUXV,
1885 	    __elfN(note_procstat_auxv), p);
1886 
1887 	*sizep = size;
1888 }
1889 
1890 static void
1891 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1892     size_t notesz)
1893 {
1894 	Elf_Ehdr *ehdr;
1895 	Elf_Phdr *phdr;
1896 	Elf_Shdr *shdr;
1897 	struct phdr_closure phc;
1898 
1899 	ehdr = (Elf_Ehdr *)hdr;
1900 
1901 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1902 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1903 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1904 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1905 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1906 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1907 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1908 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1909 	ehdr->e_ident[EI_ABIVERSION] = 0;
1910 	ehdr->e_ident[EI_PAD] = 0;
1911 	ehdr->e_type = ET_CORE;
1912 	ehdr->e_machine = td->td_proc->p_elf_machine;
1913 	ehdr->e_version = EV_CURRENT;
1914 	ehdr->e_entry = 0;
1915 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1916 	ehdr->e_flags = td->td_proc->p_elf_flags;
1917 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1918 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1919 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1920 	ehdr->e_shstrndx = SHN_UNDEF;
1921 	if (numsegs + 1 < PN_XNUM) {
1922 		ehdr->e_phnum = numsegs + 1;
1923 		ehdr->e_shnum = 0;
1924 	} else {
1925 		ehdr->e_phnum = PN_XNUM;
1926 		ehdr->e_shnum = 1;
1927 
1928 		ehdr->e_shoff = ehdr->e_phoff +
1929 		    (numsegs + 1) * ehdr->e_phentsize;
1930 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1931 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1932 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1933 
1934 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1935 		memset(shdr, 0, sizeof(*shdr));
1936 		/*
1937 		 * A special first section is used to hold large segment and
1938 		 * section counts.  This was proposed by Sun Microsystems in
1939 		 * Solaris and has been adopted by Linux; the standard ELF
1940 		 * tools are already familiar with the technique.
1941 		 *
1942 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1943 		 * (or 12-7 depending on the version of the document) for more
1944 		 * details.
1945 		 */
1946 		shdr->sh_type = SHT_NULL;
1947 		shdr->sh_size = ehdr->e_shnum;
1948 		shdr->sh_link = ehdr->e_shstrndx;
1949 		shdr->sh_info = numsegs + 1;
1950 	}
1951 
1952 	/*
1953 	 * Fill in the program header entries.
1954 	 */
1955 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1956 
1957 	/* The note segement. */
1958 	phdr->p_type = PT_NOTE;
1959 	phdr->p_offset = hdrsize;
1960 	phdr->p_vaddr = 0;
1961 	phdr->p_paddr = 0;
1962 	phdr->p_filesz = notesz;
1963 	phdr->p_memsz = 0;
1964 	phdr->p_flags = PF_R;
1965 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1966 	phdr++;
1967 
1968 	/* All the writable segments from the program. */
1969 	phc.phdr = phdr;
1970 	phc.offset = round_page(hdrsize + notesz);
1971 	each_dumpable_segment(td, cb_put_phdr, &phc);
1972 }
1973 
1974 static size_t
1975 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1976 {
1977 	struct note_info *ninfo;
1978 	size_t size, notesize;
1979 
1980 	size = 0;
1981 	out(arg, NULL, &size);
1982 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1983 	ninfo->type = type;
1984 	ninfo->outfunc = out;
1985 	ninfo->outarg = arg;
1986 	ninfo->outsize = size;
1987 	TAILQ_INSERT_TAIL(list, ninfo, link);
1988 
1989 	if (type == -1)
1990 		return (size);
1991 
1992 	notesize = sizeof(Elf_Note) +		/* note header */
1993 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1994 						/* note name */
1995 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1996 
1997 	return (notesize);
1998 }
1999 
2000 static size_t
2001 append_note_data(const void *src, void *dst, size_t len)
2002 {
2003 	size_t padded_len;
2004 
2005 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2006 	if (dst != NULL) {
2007 		bcopy(src, dst, len);
2008 		bzero((char *)dst + len, padded_len - len);
2009 	}
2010 	return (padded_len);
2011 }
2012 
2013 size_t
2014 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2015 {
2016 	Elf_Note *note;
2017 	char *buf;
2018 	size_t notesize;
2019 
2020 	buf = dst;
2021 	if (buf != NULL) {
2022 		note = (Elf_Note *)buf;
2023 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2024 		note->n_descsz = size;
2025 		note->n_type = type;
2026 		buf += sizeof(*note);
2027 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2028 		    sizeof(FREEBSD_ABI_VENDOR));
2029 		append_note_data(src, buf, size);
2030 		if (descp != NULL)
2031 			*descp = buf;
2032 	}
2033 
2034 	notesize = sizeof(Elf_Note) +		/* note header */
2035 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2036 						/* note name */
2037 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2038 
2039 	return (notesize);
2040 }
2041 
2042 static void
2043 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2044 {
2045 	Elf_Note note;
2046 	ssize_t old_len, sect_len;
2047 	size_t new_len, descsz, i;
2048 
2049 	if (ninfo->type == -1) {
2050 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2051 		return;
2052 	}
2053 
2054 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2055 	note.n_descsz = ninfo->outsize;
2056 	note.n_type = ninfo->type;
2057 
2058 	sbuf_bcat(sb, &note, sizeof(note));
2059 	sbuf_start_section(sb, &old_len);
2060 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2061 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2062 	if (note.n_descsz == 0)
2063 		return;
2064 	sbuf_start_section(sb, &old_len);
2065 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2066 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2067 	if (sect_len < 0)
2068 		return;
2069 
2070 	new_len = (size_t)sect_len;
2071 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2072 	if (new_len < descsz) {
2073 		/*
2074 		 * It is expected that individual note emitters will correctly
2075 		 * predict their expected output size and fill up to that size
2076 		 * themselves, padding in a format-specific way if needed.
2077 		 * However, in case they don't, just do it here with zeros.
2078 		 */
2079 		for (i = 0; i < descsz - new_len; i++)
2080 			sbuf_putc(sb, 0);
2081 	} else if (new_len > descsz) {
2082 		/*
2083 		 * We can't always truncate sb -- we may have drained some
2084 		 * of it already.
2085 		 */
2086 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2087 		    "read it (%zu > %zu).  Since it is longer than "
2088 		    "expected, this coredump's notes are corrupt.  THIS "
2089 		    "IS A BUG in the note_procstat routine for type %u.\n",
2090 		    __func__, (unsigned)note.n_type, new_len, descsz,
2091 		    (unsigned)note.n_type));
2092 	}
2093 }
2094 
2095 /*
2096  * Miscellaneous note out functions.
2097  */
2098 
2099 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2100 #include <compat/freebsd32/freebsd32.h>
2101 #include <compat/freebsd32/freebsd32_signal.h>
2102 
2103 typedef struct prstatus32 elf_prstatus_t;
2104 typedef struct prpsinfo32 elf_prpsinfo_t;
2105 typedef struct fpreg32 elf_prfpregset_t;
2106 typedef struct fpreg32 elf_fpregset_t;
2107 typedef struct reg32 elf_gregset_t;
2108 typedef struct thrmisc32 elf_thrmisc_t;
2109 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2110 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2111 typedef uint32_t elf_ps_strings_t;
2112 #else
2113 typedef prstatus_t elf_prstatus_t;
2114 typedef prpsinfo_t elf_prpsinfo_t;
2115 typedef prfpregset_t elf_prfpregset_t;
2116 typedef prfpregset_t elf_fpregset_t;
2117 typedef gregset_t elf_gregset_t;
2118 typedef thrmisc_t elf_thrmisc_t;
2119 #define ELF_KERN_PROC_MASK	0
2120 typedef struct kinfo_proc elf_kinfo_proc_t;
2121 typedef vm_offset_t elf_ps_strings_t;
2122 #endif
2123 
2124 static void
2125 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2126 {
2127 	struct sbuf sbarg;
2128 	size_t len;
2129 	char *cp, *end;
2130 	struct proc *p;
2131 	elf_prpsinfo_t *psinfo;
2132 	int error;
2133 
2134 	p = (struct proc *)arg;
2135 	if (sb != NULL) {
2136 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2137 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2138 		psinfo->pr_version = PRPSINFO_VERSION;
2139 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2140 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2141 		PROC_LOCK(p);
2142 		if (p->p_args != NULL) {
2143 			len = sizeof(psinfo->pr_psargs) - 1;
2144 			if (len > p->p_args->ar_length)
2145 				len = p->p_args->ar_length;
2146 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2147 			PROC_UNLOCK(p);
2148 			error = 0;
2149 		} else {
2150 			_PHOLD(p);
2151 			PROC_UNLOCK(p);
2152 			sbuf_new(&sbarg, psinfo->pr_psargs,
2153 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2154 			error = proc_getargv(curthread, p, &sbarg);
2155 			PRELE(p);
2156 			if (sbuf_finish(&sbarg) == 0)
2157 				len = sbuf_len(&sbarg) - 1;
2158 			else
2159 				len = sizeof(psinfo->pr_psargs) - 1;
2160 			sbuf_delete(&sbarg);
2161 		}
2162 		if (error || len == 0)
2163 			strlcpy(psinfo->pr_psargs, p->p_comm,
2164 			    sizeof(psinfo->pr_psargs));
2165 		else {
2166 			KASSERT(len < sizeof(psinfo->pr_psargs),
2167 			    ("len is too long: %zu vs %zu", len,
2168 			    sizeof(psinfo->pr_psargs)));
2169 			cp = psinfo->pr_psargs;
2170 			end = cp + len - 1;
2171 			for (;;) {
2172 				cp = memchr(cp, '\0', end - cp);
2173 				if (cp == NULL)
2174 					break;
2175 				*cp = ' ';
2176 			}
2177 		}
2178 		psinfo->pr_pid = p->p_pid;
2179 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2180 		free(psinfo, M_TEMP);
2181 	}
2182 	*sizep = sizeof(*psinfo);
2183 }
2184 
2185 static void
2186 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2187 {
2188 	struct thread *td;
2189 	elf_prstatus_t *status;
2190 
2191 	td = (struct thread *)arg;
2192 	if (sb != NULL) {
2193 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2194 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2195 		status->pr_version = PRSTATUS_VERSION;
2196 		status->pr_statussz = sizeof(elf_prstatus_t);
2197 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2198 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2199 		status->pr_osreldate = osreldate;
2200 		status->pr_cursig = td->td_proc->p_sig;
2201 		status->pr_pid = td->td_tid;
2202 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2203 		fill_regs32(td, &status->pr_reg);
2204 #else
2205 		fill_regs(td, &status->pr_reg);
2206 #endif
2207 		sbuf_bcat(sb, status, sizeof(*status));
2208 		free(status, M_TEMP);
2209 	}
2210 	*sizep = sizeof(*status);
2211 }
2212 
2213 static void
2214 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2215 {
2216 	struct thread *td;
2217 	elf_prfpregset_t *fpregset;
2218 
2219 	td = (struct thread *)arg;
2220 	if (sb != NULL) {
2221 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2222 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2223 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2224 		fill_fpregs32(td, fpregset);
2225 #else
2226 		fill_fpregs(td, fpregset);
2227 #endif
2228 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2229 		free(fpregset, M_TEMP);
2230 	}
2231 	*sizep = sizeof(*fpregset);
2232 }
2233 
2234 static void
2235 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2236 {
2237 	struct thread *td;
2238 	elf_thrmisc_t thrmisc;
2239 
2240 	td = (struct thread *)arg;
2241 	if (sb != NULL) {
2242 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2243 		bzero(&thrmisc, sizeof(thrmisc));
2244 		strcpy(thrmisc.pr_tname, td->td_name);
2245 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2246 	}
2247 	*sizep = sizeof(thrmisc);
2248 }
2249 
2250 static void
2251 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2252 {
2253 	struct thread *td;
2254 	size_t size;
2255 	int structsize;
2256 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2257 	struct ptrace_lwpinfo32 pl;
2258 #else
2259 	struct ptrace_lwpinfo pl;
2260 #endif
2261 
2262 	td = (struct thread *)arg;
2263 	size = sizeof(structsize) + sizeof(pl);
2264 	if (sb != NULL) {
2265 		KASSERT(*sizep == size, ("invalid size"));
2266 		structsize = sizeof(pl);
2267 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2268 		bzero(&pl, sizeof(pl));
2269 		pl.pl_lwpid = td->td_tid;
2270 		pl.pl_event = PL_EVENT_NONE;
2271 		pl.pl_sigmask = td->td_sigmask;
2272 		pl.pl_siglist = td->td_siglist;
2273 		if (td->td_si.si_signo != 0) {
2274 			pl.pl_event = PL_EVENT_SIGNAL;
2275 			pl.pl_flags |= PL_FLAG_SI;
2276 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2277 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2278 #else
2279 			pl.pl_siginfo = td->td_si;
2280 #endif
2281 		}
2282 		strcpy(pl.pl_tdname, td->td_name);
2283 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2284 		sbuf_bcat(sb, &pl, sizeof(pl));
2285 	}
2286 	*sizep = size;
2287 }
2288 
2289 /*
2290  * Allow for MD specific notes, as well as any MD
2291  * specific preparations for writing MI notes.
2292  */
2293 static void
2294 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2295 {
2296 	struct thread *td;
2297 	void *buf;
2298 	size_t size;
2299 
2300 	td = (struct thread *)arg;
2301 	size = *sizep;
2302 	if (size != 0 && sb != NULL)
2303 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2304 	else
2305 		buf = NULL;
2306 	size = 0;
2307 	__elfN(dump_thread)(td, buf, &size);
2308 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2309 	if (size != 0 && sb != NULL)
2310 		sbuf_bcat(sb, buf, size);
2311 	free(buf, M_TEMP);
2312 	*sizep = size;
2313 }
2314 
2315 #ifdef KINFO_PROC_SIZE
2316 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2317 #endif
2318 
2319 static void
2320 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2321 {
2322 	struct proc *p;
2323 	size_t size;
2324 	int structsize;
2325 
2326 	p = (struct proc *)arg;
2327 	size = sizeof(structsize) + p->p_numthreads *
2328 	    sizeof(elf_kinfo_proc_t);
2329 
2330 	if (sb != NULL) {
2331 		KASSERT(*sizep == size, ("invalid size"));
2332 		structsize = sizeof(elf_kinfo_proc_t);
2333 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2334 		PROC_LOCK(p);
2335 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2336 	}
2337 	*sizep = size;
2338 }
2339 
2340 #ifdef KINFO_FILE_SIZE
2341 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2342 #endif
2343 
2344 static void
2345 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2346 {
2347 	struct proc *p;
2348 	size_t size, sect_sz, i;
2349 	ssize_t start_len, sect_len;
2350 	int structsize, filedesc_flags;
2351 
2352 	if (coredump_pack_fileinfo)
2353 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2354 	else
2355 		filedesc_flags = 0;
2356 
2357 	p = (struct proc *)arg;
2358 	structsize = sizeof(struct kinfo_file);
2359 	if (sb == NULL) {
2360 		size = 0;
2361 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2362 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2363 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2364 		PROC_LOCK(p);
2365 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2366 		sbuf_finish(sb);
2367 		sbuf_delete(sb);
2368 		*sizep = size;
2369 	} else {
2370 		sbuf_start_section(sb, &start_len);
2371 
2372 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2373 		PROC_LOCK(p);
2374 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2375 		    filedesc_flags);
2376 
2377 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2378 		if (sect_len < 0)
2379 			return;
2380 		sect_sz = sect_len;
2381 
2382 		KASSERT(sect_sz <= *sizep,
2383 		    ("kern_proc_filedesc_out did not respect maxlen; "
2384 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2385 		     sect_sz - sizeof(structsize)));
2386 
2387 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2388 			sbuf_putc(sb, 0);
2389 	}
2390 }
2391 
2392 #ifdef KINFO_VMENTRY_SIZE
2393 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2394 #endif
2395 
2396 static void
2397 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2398 {
2399 	struct proc *p;
2400 	size_t size;
2401 	int structsize, vmmap_flags;
2402 
2403 	if (coredump_pack_vmmapinfo)
2404 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2405 	else
2406 		vmmap_flags = 0;
2407 
2408 	p = (struct proc *)arg;
2409 	structsize = sizeof(struct kinfo_vmentry);
2410 	if (sb == NULL) {
2411 		size = 0;
2412 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2413 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2414 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2415 		PROC_LOCK(p);
2416 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2417 		sbuf_finish(sb);
2418 		sbuf_delete(sb);
2419 		*sizep = size;
2420 	} else {
2421 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2422 		PROC_LOCK(p);
2423 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2424 		    vmmap_flags);
2425 	}
2426 }
2427 
2428 static void
2429 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2430 {
2431 	struct proc *p;
2432 	size_t size;
2433 	int structsize;
2434 
2435 	p = (struct proc *)arg;
2436 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2437 	if (sb != NULL) {
2438 		KASSERT(*sizep == size, ("invalid size"));
2439 		structsize = sizeof(gid_t);
2440 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2441 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2442 		    sizeof(gid_t));
2443 	}
2444 	*sizep = size;
2445 }
2446 
2447 static void
2448 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2449 {
2450 	struct proc *p;
2451 	size_t size;
2452 	int structsize;
2453 
2454 	p = (struct proc *)arg;
2455 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2456 	if (sb != NULL) {
2457 		KASSERT(*sizep == size, ("invalid size"));
2458 		structsize = sizeof(p->p_fd->fd_cmask);
2459 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2460 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2461 	}
2462 	*sizep = size;
2463 }
2464 
2465 static void
2466 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2467 {
2468 	struct proc *p;
2469 	struct rlimit rlim[RLIM_NLIMITS];
2470 	size_t size;
2471 	int structsize, i;
2472 
2473 	p = (struct proc *)arg;
2474 	size = sizeof(structsize) + sizeof(rlim);
2475 	if (sb != NULL) {
2476 		KASSERT(*sizep == size, ("invalid size"));
2477 		structsize = sizeof(rlim);
2478 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2479 		PROC_LOCK(p);
2480 		for (i = 0; i < RLIM_NLIMITS; i++)
2481 			lim_rlimit_proc(p, i, &rlim[i]);
2482 		PROC_UNLOCK(p);
2483 		sbuf_bcat(sb, rlim, sizeof(rlim));
2484 	}
2485 	*sizep = size;
2486 }
2487 
2488 static void
2489 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2490 {
2491 	struct proc *p;
2492 	size_t size;
2493 	int structsize;
2494 
2495 	p = (struct proc *)arg;
2496 	size = sizeof(structsize) + sizeof(p->p_osrel);
2497 	if (sb != NULL) {
2498 		KASSERT(*sizep == size, ("invalid size"));
2499 		structsize = sizeof(p->p_osrel);
2500 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2501 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2502 	}
2503 	*sizep = size;
2504 }
2505 
2506 static void
2507 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2508 {
2509 	struct proc *p;
2510 	elf_ps_strings_t ps_strings;
2511 	size_t size;
2512 	int structsize;
2513 
2514 	p = (struct proc *)arg;
2515 	size = sizeof(structsize) + sizeof(ps_strings);
2516 	if (sb != NULL) {
2517 		KASSERT(*sizep == size, ("invalid size"));
2518 		structsize = sizeof(ps_strings);
2519 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2520 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2521 #else
2522 		ps_strings = p->p_sysent->sv_psstrings;
2523 #endif
2524 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2525 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2526 	}
2527 	*sizep = size;
2528 }
2529 
2530 static void
2531 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2532 {
2533 	struct proc *p;
2534 	size_t size;
2535 	int structsize;
2536 
2537 	p = (struct proc *)arg;
2538 	if (sb == NULL) {
2539 		size = 0;
2540 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2541 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2542 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2543 		PHOLD(p);
2544 		proc_getauxv(curthread, p, sb);
2545 		PRELE(p);
2546 		sbuf_finish(sb);
2547 		sbuf_delete(sb);
2548 		*sizep = size;
2549 	} else {
2550 		structsize = sizeof(Elf_Auxinfo);
2551 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2552 		PHOLD(p);
2553 		proc_getauxv(curthread, p, sb);
2554 		PRELE(p);
2555 	}
2556 }
2557 
2558 static boolean_t
2559 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2560     const char *note_vendor, const Elf_Phdr *pnote,
2561     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2562 {
2563 	const Elf_Note *note, *note0, *note_end;
2564 	const char *note_name;
2565 	char *buf;
2566 	int i, error;
2567 	boolean_t res;
2568 
2569 	/* We need some limit, might as well use PAGE_SIZE. */
2570 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2571 		return (FALSE);
2572 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2573 	if (pnote->p_offset > PAGE_SIZE ||
2574 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2575 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2576 		if (buf == NULL) {
2577 			VOP_UNLOCK(imgp->vp);
2578 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2579 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2580 		}
2581 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2582 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2583 		    curthread->td_ucred, NOCRED, NULL, curthread);
2584 		if (error != 0) {
2585 			uprintf("i/o error PT_NOTE\n");
2586 			goto retf;
2587 		}
2588 		note = note0 = (const Elf_Note *)buf;
2589 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2590 	} else {
2591 		note = note0 = (const Elf_Note *)(imgp->image_header +
2592 		    pnote->p_offset);
2593 		note_end = (const Elf_Note *)(imgp->image_header +
2594 		    pnote->p_offset + pnote->p_filesz);
2595 		buf = NULL;
2596 	}
2597 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2598 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2599 		    (const char *)note < sizeof(Elf_Note)) {
2600 			goto retf;
2601 		}
2602 		if (note->n_namesz != checknote->n_namesz ||
2603 		    note->n_descsz != checknote->n_descsz ||
2604 		    note->n_type != checknote->n_type)
2605 			goto nextnote;
2606 		note_name = (const char *)(note + 1);
2607 		if (note_name + checknote->n_namesz >=
2608 		    (const char *)note_end || strncmp(note_vendor,
2609 		    note_name, checknote->n_namesz) != 0)
2610 			goto nextnote;
2611 
2612 		if (cb(note, cb_arg, &res))
2613 			goto ret;
2614 nextnote:
2615 		note = (const Elf_Note *)((const char *)(note + 1) +
2616 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2617 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2618 	}
2619 retf:
2620 	res = FALSE;
2621 ret:
2622 	free(buf, M_TEMP);
2623 	return (res);
2624 }
2625 
2626 struct brandnote_cb_arg {
2627 	Elf_Brandnote *brandnote;
2628 	int32_t *osrel;
2629 };
2630 
2631 static boolean_t
2632 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2633 {
2634 	struct brandnote_cb_arg *arg;
2635 
2636 	arg = arg0;
2637 
2638 	/*
2639 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2640 	 * necessary.
2641 	 */
2642 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2643 	    arg->brandnote->trans_osrel != NULL ?
2644 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2645 
2646 	return (TRUE);
2647 }
2648 
2649 static Elf_Note fctl_note = {
2650 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2651 	.n_descsz = sizeof(uint32_t),
2652 	.n_type = NT_FREEBSD_FEATURE_CTL,
2653 };
2654 
2655 struct fctl_cb_arg {
2656 	uint32_t *fctl0;
2657 };
2658 
2659 static boolean_t
2660 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2661 {
2662 	struct fctl_cb_arg *arg;
2663 	const Elf32_Word *desc;
2664 	uintptr_t p;
2665 
2666 	arg = arg0;
2667 	p = (uintptr_t)(note + 1);
2668 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2669 	desc = (const Elf32_Word *)p;
2670 	*arg->fctl0 = desc[0];
2671 	return (TRUE);
2672 }
2673 
2674 /*
2675  * Try to find the appropriate ABI-note section for checknote, fetch
2676  * the osreldate and feature control flags for binary from the ELF
2677  * OSABI-note.  Only the first page of the image is searched, the same
2678  * as for headers.
2679  */
2680 static boolean_t
2681 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2682     int32_t *osrel, uint32_t *fctl0)
2683 {
2684 	const Elf_Phdr *phdr;
2685 	const Elf_Ehdr *hdr;
2686 	struct brandnote_cb_arg b_arg;
2687 	struct fctl_cb_arg f_arg;
2688 	int i, j;
2689 
2690 	hdr = (const Elf_Ehdr *)imgp->image_header;
2691 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2692 	b_arg.brandnote = brandnote;
2693 	b_arg.osrel = osrel;
2694 	f_arg.fctl0 = fctl0;
2695 
2696 	for (i = 0; i < hdr->e_phnum; i++) {
2697 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2698 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2699 		    &b_arg)) {
2700 			for (j = 0; j < hdr->e_phnum; j++) {
2701 				if (phdr[j].p_type == PT_NOTE &&
2702 				    __elfN(parse_notes)(imgp, &fctl_note,
2703 				    FREEBSD_ABI_VENDOR, &phdr[j],
2704 				    note_fctl_cb, &f_arg))
2705 					break;
2706 			}
2707 			return (TRUE);
2708 		}
2709 	}
2710 	return (FALSE);
2711 
2712 }
2713 
2714 /*
2715  * Tell kern_execve.c about it, with a little help from the linker.
2716  */
2717 static struct execsw __elfN(execsw) = {
2718 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2719 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2720 };
2721 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2722 
2723 static vm_prot_t
2724 __elfN(trans_prot)(Elf_Word flags)
2725 {
2726 	vm_prot_t prot;
2727 
2728 	prot = 0;
2729 	if (flags & PF_X)
2730 		prot |= VM_PROT_EXECUTE;
2731 	if (flags & PF_W)
2732 		prot |= VM_PROT_WRITE;
2733 	if (flags & PF_R)
2734 		prot |= VM_PROT_READ;
2735 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2736 	if (i386_read_exec && (flags & PF_R))
2737 		prot |= VM_PROT_EXECUTE;
2738 #endif
2739 	return (prot);
2740 }
2741 
2742 static Elf_Word
2743 __elfN(untrans_prot)(vm_prot_t prot)
2744 {
2745 	Elf_Word flags;
2746 
2747 	flags = 0;
2748 	if (prot & VM_PROT_EXECUTE)
2749 		flags |= PF_X;
2750 	if (prot & VM_PROT_READ)
2751 		flags |= PF_R;
2752 	if (prot & VM_PROT_WRITE)
2753 		flags |= PF_W;
2754 	return (flags);
2755 }
2756 
2757 void
2758 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2759 {
2760 	uintptr_t range, rbase, gap;
2761 	int pct;
2762 
2763 	if ((imgp->map_flags & MAP_ASLR) == 0)
2764 		return;
2765 	pct = __elfN(aslr_stack_gap);
2766 	if (pct == 0)
2767 		return;
2768 	if (pct > 50)
2769 		pct = 50;
2770 	range = imgp->eff_stack_sz * pct / 100;
2771 	arc4rand(&rbase, sizeof(rbase), 0);
2772 	gap = rbase % range;
2773 	gap &= ~(sizeof(u_long) - 1);
2774 	*stack_base -= gap;
2775 }
2776