xref: /freebsd/sys/kern/imgact_elf.c (revision f7c32ed617858bcd22f8d1b03199099d50125721)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2017 Dell EMC
5  * Copyright (c) 2000-2001, 2003 David O'Brien
6  * Copyright (c) 1995-1996 Søren Schmidt
7  * Copyright (c) 1996 Peter Wemm
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_capsicum.h"
38 
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mman.h>
52 #include <sys/namei.h>
53 #include <sys/proc.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
57 #include <sys/reg.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sbuf.h>
61 #include <sys/sf_buf.h>
62 #include <sys/smp.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
65 #include <sys/stat.h>
66 #include <sys/sx.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
73 #include <sys/user.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
85 
86 #define ELF_NOTE_ROUNDSIZE	4
87 #define OLD_EI_BRAND	8
88 
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91     const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93     u_long *entry);
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98     int32_t *osrel);
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static bool __elfN(check_note)(struct image_params *imgp,
101     Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0,
102     uint32_t *fctl0);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
107     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
108     "");
109 
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117     &elf_legacy_coredump, 0,
118     "include all and only RW pages in core dumps");
119 
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
123     defined(__riscv)
124 	1;
125 #else
126 	0;
127 #endif
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 
132 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
133 int i386_read_exec = 0;
134 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
135     "enable execution from readable segments");
136 #endif
137 
138 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
139 static int
140 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
141 {
142 	u_long val;
143 	int error;
144 
145 	val = __elfN(pie_base);
146 	error = sysctl_handle_long(oidp, &val, 0, req);
147 	if (error != 0 || req->newptr == NULL)
148 		return (error);
149 	if ((val & PAGE_MASK) != 0)
150 		return (EINVAL);
151 	__elfN(pie_base) = val;
152 	return (0);
153 }
154 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
155     CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
156     sysctl_pie_base, "LU",
157     "PIE load base without randomization");
158 
159 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
160     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
161     "");
162 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
163 
164 /*
165  * While for 64-bit machines ASLR works properly, there are
166  * still some problems when using 32-bit architectures. For this
167  * reason ASLR is only enabled by default when running native
168  * 64-bit non-PIE executables.
169  */
170 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
171 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
172     &__elfN(aslr_enabled), 0,
173     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
174     ": enable address map randomization");
175 
176 /*
177  * Enable ASLR only for 64-bit PIE binaries by default.
178  */
179 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
180 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
181     &__elfN(pie_aslr_enabled), 0,
182     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
183     ": enable address map randomization for PIE binaries");
184 
185 /*
186  * Sbrk is now deprecated and it can be assumed, that in most
187  * cases it will not be used anyway. This setting is valid only
188  * for the ASLR enabled and allows for utilizing the bss grow region.
189  */
190 static int __elfN(aslr_honor_sbrk) = 0;
191 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
192     &__elfN(aslr_honor_sbrk), 0,
193     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
194 
195 static int __elfN(aslr_stack_gap) = 3;
196 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
197     &__elfN(aslr_stack_gap), 0,
198     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
199     ": maximum percentage of main stack to waste on a random gap");
200 
201 static int __elfN(sigfastblock) = 1;
202 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
203     CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
204     "enable sigfastblock for new processes");
205 
206 static bool __elfN(allow_wx) = true;
207 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
208     CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
209     "Allow pages to be mapped simultaneously writable and executable");
210 
211 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
212 
213 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
214 
215 Elf_Brandnote __elfN(freebsd_brandnote) = {
216 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
217 	.hdr.n_descsz	= sizeof(int32_t),
218 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
219 	.vendor		= FREEBSD_ABI_VENDOR,
220 	.flags		= BN_TRANSLATE_OSREL,
221 	.trans_osrel	= __elfN(freebsd_trans_osrel)
222 };
223 
224 static bool
225 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
226 {
227 	uintptr_t p;
228 
229 	p = (uintptr_t)(note + 1);
230 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
231 	*osrel = *(const int32_t *)(p);
232 
233 	return (true);
234 }
235 
236 static const char GNU_ABI_VENDOR[] = "GNU";
237 static int GNU_KFREEBSD_ABI_DESC = 3;
238 
239 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
240 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
241 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
242 	.hdr.n_type	= 1,
243 	.vendor		= GNU_ABI_VENDOR,
244 	.flags		= BN_TRANSLATE_OSREL,
245 	.trans_osrel	= kfreebsd_trans_osrel
246 };
247 
248 static bool
249 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
250 {
251 	const Elf32_Word *desc;
252 	uintptr_t p;
253 
254 	p = (uintptr_t)(note + 1);
255 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
256 
257 	desc = (const Elf32_Word *)p;
258 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
259 		return (false);
260 
261 	/*
262 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
263 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
264 	 */
265 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
266 
267 	return (true);
268 }
269 
270 int
271 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
272 {
273 	int i;
274 
275 	for (i = 0; i < MAX_BRANDS; i++) {
276 		if (elf_brand_list[i] == NULL) {
277 			elf_brand_list[i] = entry;
278 			break;
279 		}
280 	}
281 	if (i == MAX_BRANDS) {
282 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
283 			__func__, entry);
284 		return (-1);
285 	}
286 	return (0);
287 }
288 
289 int
290 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
291 {
292 	int i;
293 
294 	for (i = 0; i < MAX_BRANDS; i++) {
295 		if (elf_brand_list[i] == entry) {
296 			elf_brand_list[i] = NULL;
297 			break;
298 		}
299 	}
300 	if (i == MAX_BRANDS)
301 		return (-1);
302 	return (0);
303 }
304 
305 bool
306 __elfN(brand_inuse)(Elf_Brandinfo *entry)
307 {
308 	struct proc *p;
309 	bool rval = false;
310 
311 	sx_slock(&allproc_lock);
312 	FOREACH_PROC_IN_SYSTEM(p) {
313 		if (p->p_sysent == entry->sysvec) {
314 			rval = true;
315 			break;
316 		}
317 	}
318 	sx_sunlock(&allproc_lock);
319 
320 	return (rval);
321 }
322 
323 static Elf_Brandinfo *
324 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
325     int32_t *osrel, uint32_t *fctl0)
326 {
327 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
328 	Elf_Brandinfo *bi, *bi_m;
329 	bool ret, has_fctl0;
330 	int i, interp_name_len;
331 
332 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
333 
334 	/*
335 	 * We support four types of branding -- (1) the ELF EI_OSABI field
336 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
337 	 * branding w/in the ELF header, (3) path of the `interp_path'
338 	 * field, and (4) the ".note.ABI-tag" ELF section.
339 	 */
340 
341 	/* Look for an ".note.ABI-tag" ELF section */
342 	bi_m = NULL;
343 	for (i = 0; i < MAX_BRANDS; i++) {
344 		bi = elf_brand_list[i];
345 		if (bi == NULL)
346 			continue;
347 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
348 			continue;
349 		if (hdr->e_machine == bi->machine && (bi->flags &
350 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
351 			has_fctl0 = false;
352 			*fctl0 = 0;
353 			*osrel = 0;
354 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
355 			    &has_fctl0, fctl0);
356 			/* Give brand a chance to veto check_note's guess */
357 			if (ret && bi->header_supported) {
358 				ret = bi->header_supported(imgp, osrel,
359 				    has_fctl0 ? fctl0 : NULL);
360 			}
361 			/*
362 			 * If note checker claimed the binary, but the
363 			 * interpreter path in the image does not
364 			 * match default one for the brand, try to
365 			 * search for other brands with the same
366 			 * interpreter.  Either there is better brand
367 			 * with the right interpreter, or, failing
368 			 * this, we return first brand which accepted
369 			 * our note and, optionally, header.
370 			 */
371 			if (ret && bi_m == NULL && interp != NULL &&
372 			    (bi->interp_path == NULL ||
373 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
374 			    strncmp(interp, bi->interp_path, interp_name_len)
375 			    != 0))) {
376 				bi_m = bi;
377 				ret = 0;
378 			}
379 			if (ret)
380 				return (bi);
381 		}
382 	}
383 	if (bi_m != NULL)
384 		return (bi_m);
385 
386 	/* If the executable has a brand, search for it in the brand list. */
387 	for (i = 0; i < MAX_BRANDS; i++) {
388 		bi = elf_brand_list[i];
389 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
390 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
391 			continue;
392 		if (hdr->e_machine == bi->machine &&
393 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
394 		    (bi->compat_3_brand != NULL &&
395 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
396 		    bi->compat_3_brand) == 0))) {
397 			/* Looks good, but give brand a chance to veto */
398 			if (bi->header_supported == NULL ||
399 			    bi->header_supported(imgp, NULL, NULL)) {
400 				/*
401 				 * Again, prefer strictly matching
402 				 * interpreter path.
403 				 */
404 				if (interp_name_len == 0 &&
405 				    bi->interp_path == NULL)
406 					return (bi);
407 				if (bi->interp_path != NULL &&
408 				    strlen(bi->interp_path) + 1 ==
409 				    interp_name_len && strncmp(interp,
410 				    bi->interp_path, interp_name_len) == 0)
411 					return (bi);
412 				if (bi_m == NULL)
413 					bi_m = bi;
414 			}
415 		}
416 	}
417 	if (bi_m != NULL)
418 		return (bi_m);
419 
420 	/* No known brand, see if the header is recognized by any brand */
421 	for (i = 0; i < MAX_BRANDS; i++) {
422 		bi = elf_brand_list[i];
423 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
424 		    bi->header_supported == NULL)
425 			continue;
426 		if (hdr->e_machine == bi->machine) {
427 			ret = bi->header_supported(imgp, NULL, NULL);
428 			if (ret)
429 				return (bi);
430 		}
431 	}
432 
433 	/* Lacking a known brand, search for a recognized interpreter. */
434 	if (interp != NULL) {
435 		for (i = 0; i < MAX_BRANDS; i++) {
436 			bi = elf_brand_list[i];
437 			if (bi == NULL || (bi->flags &
438 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
439 			    != 0)
440 				continue;
441 			if (hdr->e_machine == bi->machine &&
442 			    bi->interp_path != NULL &&
443 			    /* ELF image p_filesz includes terminating zero */
444 			    strlen(bi->interp_path) + 1 == interp_name_len &&
445 			    strncmp(interp, bi->interp_path, interp_name_len)
446 			    == 0 && (bi->header_supported == NULL ||
447 			    bi->header_supported(imgp, NULL, NULL)))
448 				return (bi);
449 		}
450 	}
451 
452 	/* Lacking a recognized interpreter, try the default brand */
453 	for (i = 0; i < MAX_BRANDS; i++) {
454 		bi = elf_brand_list[i];
455 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
456 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
457 			continue;
458 		if (hdr->e_machine == bi->machine &&
459 		    __elfN(fallback_brand) == bi->brand &&
460 		    (bi->header_supported == NULL ||
461 		    bi->header_supported(imgp, NULL, NULL)))
462 			return (bi);
463 	}
464 	return (NULL);
465 }
466 
467 static bool
468 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
469 {
470 	return (hdr->e_phoff <= PAGE_SIZE &&
471 	    (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
472 }
473 
474 static int
475 __elfN(check_header)(const Elf_Ehdr *hdr)
476 {
477 	Elf_Brandinfo *bi;
478 	int i;
479 
480 	if (!IS_ELF(*hdr) ||
481 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
482 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
483 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
484 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
485 	    hdr->e_version != ELF_TARG_VER)
486 		return (ENOEXEC);
487 
488 	/*
489 	 * Make sure we have at least one brand for this machine.
490 	 */
491 
492 	for (i = 0; i < MAX_BRANDS; i++) {
493 		bi = elf_brand_list[i];
494 		if (bi != NULL && bi->machine == hdr->e_machine)
495 			break;
496 	}
497 	if (i == MAX_BRANDS)
498 		return (ENOEXEC);
499 
500 	return (0);
501 }
502 
503 static int
504 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
505     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
506 {
507 	struct sf_buf *sf;
508 	int error;
509 	vm_offset_t off;
510 
511 	/*
512 	 * Create the page if it doesn't exist yet. Ignore errors.
513 	 */
514 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
515 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
516 
517 	/*
518 	 * Find the page from the underlying object.
519 	 */
520 	if (object != NULL) {
521 		sf = vm_imgact_map_page(object, offset);
522 		if (sf == NULL)
523 			return (KERN_FAILURE);
524 		off = offset - trunc_page(offset);
525 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
526 		    end - start);
527 		vm_imgact_unmap_page(sf);
528 		if (error != 0)
529 			return (KERN_FAILURE);
530 	}
531 
532 	return (KERN_SUCCESS);
533 }
534 
535 static int
536 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
537     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
538     int cow)
539 {
540 	struct sf_buf *sf;
541 	vm_offset_t off;
542 	vm_size_t sz;
543 	int error, locked, rv;
544 
545 	if (start != trunc_page(start)) {
546 		rv = __elfN(map_partial)(map, object, offset, start,
547 		    round_page(start), prot);
548 		if (rv != KERN_SUCCESS)
549 			return (rv);
550 		offset += round_page(start) - start;
551 		start = round_page(start);
552 	}
553 	if (end != round_page(end)) {
554 		rv = __elfN(map_partial)(map, object, offset +
555 		    trunc_page(end) - start, trunc_page(end), end, prot);
556 		if (rv != KERN_SUCCESS)
557 			return (rv);
558 		end = trunc_page(end);
559 	}
560 	if (start >= end)
561 		return (KERN_SUCCESS);
562 	if ((offset & PAGE_MASK) != 0) {
563 		/*
564 		 * The mapping is not page aligned.  This means that we have
565 		 * to copy the data.
566 		 */
567 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
568 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
569 		if (rv != KERN_SUCCESS)
570 			return (rv);
571 		if (object == NULL)
572 			return (KERN_SUCCESS);
573 		for (; start < end; start += sz) {
574 			sf = vm_imgact_map_page(object, offset);
575 			if (sf == NULL)
576 				return (KERN_FAILURE);
577 			off = offset - trunc_page(offset);
578 			sz = end - start;
579 			if (sz > PAGE_SIZE - off)
580 				sz = PAGE_SIZE - off;
581 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
582 			    (caddr_t)start, sz);
583 			vm_imgact_unmap_page(sf);
584 			if (error != 0)
585 				return (KERN_FAILURE);
586 			offset += sz;
587 		}
588 	} else {
589 		vm_object_reference(object);
590 		rv = vm_map_fixed(map, object, offset, start, end - start,
591 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
592 		    (object != NULL ? MAP_VN_EXEC : 0));
593 		if (rv != KERN_SUCCESS) {
594 			locked = VOP_ISLOCKED(imgp->vp);
595 			VOP_UNLOCK(imgp->vp);
596 			vm_object_deallocate(object);
597 			vn_lock(imgp->vp, locked | LK_RETRY);
598 			return (rv);
599 		} else if (object != NULL) {
600 			MPASS(imgp->vp->v_object == object);
601 			VOP_SET_TEXT_CHECKED(imgp->vp);
602 		}
603 	}
604 	return (KERN_SUCCESS);
605 }
606 
607 static int
608 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
609     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
610 {
611 	struct sf_buf *sf;
612 	size_t map_len;
613 	vm_map_t map;
614 	vm_object_t object;
615 	vm_offset_t map_addr;
616 	int error, rv, cow;
617 	size_t copy_len;
618 	vm_ooffset_t file_addr;
619 
620 	/*
621 	 * It's necessary to fail if the filsz + offset taken from the
622 	 * header is greater than the actual file pager object's size.
623 	 * If we were to allow this, then the vm_map_find() below would
624 	 * walk right off the end of the file object and into the ether.
625 	 *
626 	 * While I'm here, might as well check for something else that
627 	 * is invalid: filsz cannot be greater than memsz.
628 	 */
629 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
630 	    filsz > memsz) {
631 		uprintf("elf_load_section: truncated ELF file\n");
632 		return (ENOEXEC);
633 	}
634 
635 	object = imgp->object;
636 	map = &imgp->proc->p_vmspace->vm_map;
637 	map_addr = trunc_page((vm_offset_t)vmaddr);
638 	file_addr = trunc_page(offset);
639 
640 	/*
641 	 * We have two choices.  We can either clear the data in the last page
642 	 * of an oversized mapping, or we can start the anon mapping a page
643 	 * early and copy the initialized data into that first page.  We
644 	 * choose the second.
645 	 */
646 	if (filsz == 0)
647 		map_len = 0;
648 	else if (memsz > filsz)
649 		map_len = trunc_page(offset + filsz) - file_addr;
650 	else
651 		map_len = round_page(offset + filsz) - file_addr;
652 
653 	if (map_len != 0) {
654 		/* cow flags: don't dump readonly sections in core */
655 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
656 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
657 
658 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
659 		    map_addr, map_addr + map_len, prot, cow);
660 		if (rv != KERN_SUCCESS)
661 			return (EINVAL);
662 
663 		/* we can stop now if we've covered it all */
664 		if (memsz == filsz)
665 			return (0);
666 	}
667 
668 	/*
669 	 * We have to get the remaining bit of the file into the first part
670 	 * of the oversized map segment.  This is normally because the .data
671 	 * segment in the file is extended to provide bss.  It's a neat idea
672 	 * to try and save a page, but it's a pain in the behind to implement.
673 	 */
674 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
675 	    filsz);
676 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
677 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
678 
679 	/* This had damn well better be true! */
680 	if (map_len != 0) {
681 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
682 		    map_addr + map_len, prot, 0);
683 		if (rv != KERN_SUCCESS)
684 			return (EINVAL);
685 	}
686 
687 	if (copy_len != 0) {
688 		sf = vm_imgact_map_page(object, offset + filsz);
689 		if (sf == NULL)
690 			return (EIO);
691 
692 		/* send the page fragment to user space */
693 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
694 		    copy_len);
695 		vm_imgact_unmap_page(sf);
696 		if (error != 0)
697 			return (error);
698 	}
699 
700 	/*
701 	 * Remove write access to the page if it was only granted by map_insert
702 	 * to allow copyout.
703 	 */
704 	if ((prot & VM_PROT_WRITE) == 0)
705 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
706 		    map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
707 
708 	return (0);
709 }
710 
711 static int
712 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
713     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
714 {
715 	vm_prot_t prot;
716 	u_long base_addr;
717 	bool first;
718 	int error, i;
719 
720 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
721 
722 	base_addr = 0;
723 	first = true;
724 
725 	for (i = 0; i < hdr->e_phnum; i++) {
726 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
727 			continue;
728 
729 		/* Loadable segment */
730 		prot = __elfN(trans_prot)(phdr[i].p_flags);
731 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
732 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
733 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
734 		if (error != 0)
735 			return (error);
736 
737 		/*
738 		 * Establish the base address if this is the first segment.
739 		 */
740 		if (first) {
741   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
742 			first = false;
743 		}
744 	}
745 
746 	if (base_addrp != NULL)
747 		*base_addrp = base_addr;
748 
749 	return (0);
750 }
751 
752 /*
753  * Load the file "file" into memory.  It may be either a shared object
754  * or an executable.
755  *
756  * The "addr" reference parameter is in/out.  On entry, it specifies
757  * the address where a shared object should be loaded.  If the file is
758  * an executable, this value is ignored.  On exit, "addr" specifies
759  * where the file was actually loaded.
760  *
761  * The "entry" reference parameter is out only.  On exit, it specifies
762  * the entry point for the loaded file.
763  */
764 static int
765 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
766 	u_long *entry)
767 {
768 	struct {
769 		struct nameidata nd;
770 		struct vattr attr;
771 		struct image_params image_params;
772 	} *tempdata;
773 	const Elf_Ehdr *hdr = NULL;
774 	const Elf_Phdr *phdr = NULL;
775 	struct nameidata *nd;
776 	struct vattr *attr;
777 	struct image_params *imgp;
778 	u_long rbase;
779 	u_long base_addr = 0;
780 	int error;
781 
782 #ifdef CAPABILITY_MODE
783 	/*
784 	 * XXXJA: This check can go away once we are sufficiently confident
785 	 * that the checks in namei() are correct.
786 	 */
787 	if (IN_CAPABILITY_MODE(curthread))
788 		return (ECAPMODE);
789 #endif
790 
791 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
792 	nd = &tempdata->nd;
793 	attr = &tempdata->attr;
794 	imgp = &tempdata->image_params;
795 
796 	/*
797 	 * Initialize part of the common data
798 	 */
799 	imgp->proc = p;
800 	imgp->attr = attr;
801 
802 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
803 	    UIO_SYSSPACE, file, curthread);
804 	if ((error = namei(nd)) != 0) {
805 		nd->ni_vp = NULL;
806 		goto fail;
807 	}
808 	NDFREE(nd, NDF_ONLY_PNBUF);
809 	imgp->vp = nd->ni_vp;
810 
811 	/*
812 	 * Check permissions, modes, uid, etc on the file, and "open" it.
813 	 */
814 	error = exec_check_permissions(imgp);
815 	if (error)
816 		goto fail;
817 
818 	error = exec_map_first_page(imgp);
819 	if (error)
820 		goto fail;
821 
822 	imgp->object = nd->ni_vp->v_object;
823 
824 	hdr = (const Elf_Ehdr *)imgp->image_header;
825 	if ((error = __elfN(check_header)(hdr)) != 0)
826 		goto fail;
827 	if (hdr->e_type == ET_DYN)
828 		rbase = *addr;
829 	else if (hdr->e_type == ET_EXEC)
830 		rbase = 0;
831 	else {
832 		error = ENOEXEC;
833 		goto fail;
834 	}
835 
836 	/* Only support headers that fit within first page for now      */
837 	if (!__elfN(phdr_in_zero_page)(hdr)) {
838 		error = ENOEXEC;
839 		goto fail;
840 	}
841 
842 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
843 	if (!aligned(phdr, Elf_Addr)) {
844 		error = ENOEXEC;
845 		goto fail;
846 	}
847 
848 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
849 	if (error != 0)
850 		goto fail;
851 
852 	*addr = base_addr;
853 	*entry = (unsigned long)hdr->e_entry + rbase;
854 
855 fail:
856 	if (imgp->firstpage)
857 		exec_unmap_first_page(imgp);
858 
859 	if (nd->ni_vp) {
860 		if (imgp->textset)
861 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
862 		vput(nd->ni_vp);
863 	}
864 	free(tempdata, M_TEMP);
865 
866 	return (error);
867 }
868 
869 static u_long
870 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
871     u_int align)
872 {
873 	u_long rbase, res;
874 
875 	MPASS(vm_map_min(map) <= minv);
876 	MPASS(maxv <= vm_map_max(map));
877 	MPASS(minv < maxv);
878 	MPASS(minv + align < maxv);
879 	arc4rand(&rbase, sizeof(rbase), 0);
880 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
881 	res &= ~((u_long)align - 1);
882 	if (res >= maxv)
883 		res -= align;
884 	KASSERT(res >= minv,
885 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
886 	    res, minv, maxv, rbase));
887 	KASSERT(res < maxv,
888 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
889 	    res, maxv, minv, rbase));
890 	return (res);
891 }
892 
893 static int
894 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
895     const Elf_Phdr *phdr, u_long et_dyn_addr)
896 {
897 	struct vmspace *vmspace;
898 	const char *err_str;
899 	u_long text_size, data_size, total_size, text_addr, data_addr;
900 	u_long seg_size, seg_addr;
901 	int i;
902 
903 	err_str = NULL;
904 	text_size = data_size = total_size = text_addr = data_addr = 0;
905 
906 	for (i = 0; i < hdr->e_phnum; i++) {
907 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
908 			continue;
909 
910 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
911 		seg_size = round_page(phdr[i].p_memsz +
912 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
913 
914 		/*
915 		 * Make the largest executable segment the official
916 		 * text segment and all others data.
917 		 *
918 		 * Note that obreak() assumes that data_addr + data_size == end
919 		 * of data load area, and the ELF file format expects segments
920 		 * to be sorted by address.  If multiple data segments exist,
921 		 * the last one will be used.
922 		 */
923 
924 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
925 			text_size = seg_size;
926 			text_addr = seg_addr;
927 		} else {
928 			data_size = seg_size;
929 			data_addr = seg_addr;
930 		}
931 		total_size += seg_size;
932 	}
933 
934 	if (data_addr == 0 && data_size == 0) {
935 		data_addr = text_addr;
936 		data_size = text_size;
937 	}
938 
939 	/*
940 	 * Check limits.  It should be safe to check the
941 	 * limits after loading the segments since we do
942 	 * not actually fault in all the segments pages.
943 	 */
944 	PROC_LOCK(imgp->proc);
945 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
946 		err_str = "Data segment size exceeds process limit";
947 	else if (text_size > maxtsiz)
948 		err_str = "Text segment size exceeds system limit";
949 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
950 		err_str = "Total segment size exceeds process limit";
951 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
952 		err_str = "Data segment size exceeds resource limit";
953 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
954 		err_str = "Total segment size exceeds resource limit";
955 	PROC_UNLOCK(imgp->proc);
956 	if (err_str != NULL) {
957 		uprintf("%s\n", err_str);
958 		return (ENOMEM);
959 	}
960 
961 	vmspace = imgp->proc->p_vmspace;
962 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
963 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
964 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
965 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
966 
967 	return (0);
968 }
969 
970 static int
971 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
972     char **interpp, bool *free_interpp)
973 {
974 	struct thread *td;
975 	char *interp;
976 	int error, interp_name_len;
977 
978 	KASSERT(phdr->p_type == PT_INTERP,
979 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
980 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
981 
982 	td = curthread;
983 
984 	/* Path to interpreter */
985 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
986 		uprintf("Invalid PT_INTERP\n");
987 		return (ENOEXEC);
988 	}
989 
990 	interp_name_len = phdr->p_filesz;
991 	if (phdr->p_offset > PAGE_SIZE ||
992 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
993 		/*
994 		 * The vnode lock might be needed by the pagedaemon to
995 		 * clean pages owned by the vnode.  Do not allow sleep
996 		 * waiting for memory with the vnode locked, instead
997 		 * try non-sleepable allocation first, and if it
998 		 * fails, go to the slow path were we drop the lock
999 		 * and do M_WAITOK.  A text reference prevents
1000 		 * modifications to the vnode content.
1001 		 */
1002 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1003 		if (interp == NULL) {
1004 			VOP_UNLOCK(imgp->vp);
1005 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1006 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1007 		}
1008 
1009 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
1010 		    interp_name_len, phdr->p_offset,
1011 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1012 		    NOCRED, NULL, td);
1013 		if (error != 0) {
1014 			free(interp, M_TEMP);
1015 			uprintf("i/o error PT_INTERP %d\n", error);
1016 			return (error);
1017 		}
1018 		interp[interp_name_len] = '\0';
1019 
1020 		*interpp = interp;
1021 		*free_interpp = true;
1022 		return (0);
1023 	}
1024 
1025 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1026 	if (interp[interp_name_len - 1] != '\0') {
1027 		uprintf("Invalid PT_INTERP\n");
1028 		return (ENOEXEC);
1029 	}
1030 
1031 	*interpp = interp;
1032 	*free_interpp = false;
1033 	return (0);
1034 }
1035 
1036 static int
1037 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1038     const char *interp, u_long *addr, u_long *entry)
1039 {
1040 	char *path;
1041 	int error;
1042 
1043 	if (brand_info->emul_path != NULL &&
1044 	    brand_info->emul_path[0] != '\0') {
1045 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1046 		snprintf(path, MAXPATHLEN, "%s%s",
1047 		    brand_info->emul_path, interp);
1048 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1049 		free(path, M_TEMP);
1050 		if (error == 0)
1051 			return (0);
1052 	}
1053 
1054 	if (brand_info->interp_newpath != NULL &&
1055 	    (brand_info->interp_path == NULL ||
1056 	    strcmp(interp, brand_info->interp_path) == 0)) {
1057 		error = __elfN(load_file)(imgp->proc,
1058 		    brand_info->interp_newpath, addr, entry);
1059 		if (error == 0)
1060 			return (0);
1061 	}
1062 
1063 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1064 	if (error == 0)
1065 		return (0);
1066 
1067 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1068 	return (error);
1069 }
1070 
1071 /*
1072  * Impossible et_dyn_addr initial value indicating that the real base
1073  * must be calculated later with some randomization applied.
1074  */
1075 #define	ET_DYN_ADDR_RAND	1
1076 
1077 static int
1078 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1079 {
1080 	struct thread *td;
1081 	const Elf_Ehdr *hdr;
1082 	const Elf_Phdr *phdr;
1083 	Elf_Auxargs *elf_auxargs;
1084 	struct vmspace *vmspace;
1085 	vm_map_t map;
1086 	char *interp;
1087 	Elf_Brandinfo *brand_info;
1088 	struct sysentvec *sv;
1089 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1090 	u_long maxalign, mapsz, maxv, maxv1;
1091 	uint32_t fctl0;
1092 	int32_t osrel;
1093 	bool free_interp;
1094 	int error, i, n;
1095 
1096 	hdr = (const Elf_Ehdr *)imgp->image_header;
1097 
1098 	/*
1099 	 * Do we have a valid ELF header ?
1100 	 *
1101 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1102 	 * if particular brand doesn't support it.
1103 	 */
1104 	if (__elfN(check_header)(hdr) != 0 ||
1105 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1106 		return (-1);
1107 
1108 	/*
1109 	 * From here on down, we return an errno, not -1, as we've
1110 	 * detected an ELF file.
1111 	 */
1112 
1113 	if (!__elfN(phdr_in_zero_page)(hdr)) {
1114 		uprintf("Program headers not in the first page\n");
1115 		return (ENOEXEC);
1116 	}
1117 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1118 	if (!aligned(phdr, Elf_Addr)) {
1119 		uprintf("Unaligned program headers\n");
1120 		return (ENOEXEC);
1121 	}
1122 
1123 	n = error = 0;
1124 	baddr = 0;
1125 	osrel = 0;
1126 	fctl0 = 0;
1127 	entry = proghdr = 0;
1128 	interp = NULL;
1129 	free_interp = false;
1130 	td = curthread;
1131 	maxalign = PAGE_SIZE;
1132 	mapsz = 0;
1133 
1134 	for (i = 0; i < hdr->e_phnum; i++) {
1135 		switch (phdr[i].p_type) {
1136 		case PT_LOAD:
1137 			if (n == 0)
1138 				baddr = phdr[i].p_vaddr;
1139 			if (phdr[i].p_align > maxalign)
1140 				maxalign = phdr[i].p_align;
1141 			mapsz += phdr[i].p_memsz;
1142 			n++;
1143 
1144 			/*
1145 			 * If this segment contains the program headers,
1146 			 * remember their virtual address for the AT_PHDR
1147 			 * aux entry. Static binaries don't usually include
1148 			 * a PT_PHDR entry.
1149 			 */
1150 			if (phdr[i].p_offset == 0 &&
1151 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1152 				<= phdr[i].p_filesz)
1153 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1154 			break;
1155 		case PT_INTERP:
1156 			/* Path to interpreter */
1157 			if (interp != NULL) {
1158 				uprintf("Multiple PT_INTERP headers\n");
1159 				error = ENOEXEC;
1160 				goto ret;
1161 			}
1162 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1163 			    &free_interp);
1164 			if (error != 0)
1165 				goto ret;
1166 			break;
1167 		case PT_GNU_STACK:
1168 			if (__elfN(nxstack))
1169 				imgp->stack_prot =
1170 				    __elfN(trans_prot)(phdr[i].p_flags);
1171 			imgp->stack_sz = phdr[i].p_memsz;
1172 			break;
1173 		case PT_PHDR: 	/* Program header table info */
1174 			proghdr = phdr[i].p_vaddr;
1175 			break;
1176 		}
1177 	}
1178 
1179 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1180 	if (brand_info == NULL) {
1181 		uprintf("ELF binary type \"%u\" not known.\n",
1182 		    hdr->e_ident[EI_OSABI]);
1183 		error = ENOEXEC;
1184 		goto ret;
1185 	}
1186 	sv = brand_info->sysvec;
1187 	et_dyn_addr = 0;
1188 	if (hdr->e_type == ET_DYN) {
1189 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1190 			uprintf("Cannot execute shared object\n");
1191 			error = ENOEXEC;
1192 			goto ret;
1193 		}
1194 		/*
1195 		 * Honour the base load address from the dso if it is
1196 		 * non-zero for some reason.
1197 		 */
1198 		if (baddr == 0) {
1199 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1200 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1201 				et_dyn_addr = __elfN(pie_base);
1202 			else if ((__elfN(pie_aslr_enabled) &&
1203 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1204 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1205 				et_dyn_addr = ET_DYN_ADDR_RAND;
1206 			else
1207 				et_dyn_addr = __elfN(pie_base);
1208 		}
1209 	}
1210 
1211 	/*
1212 	 * Avoid a possible deadlock if the current address space is destroyed
1213 	 * and that address space maps the locked vnode.  In the common case,
1214 	 * the locked vnode's v_usecount is decremented but remains greater
1215 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1216 	 * However, in cases where the vnode lock is external, such as nullfs,
1217 	 * v_usecount may become zero.
1218 	 *
1219 	 * The VV_TEXT flag prevents modifications to the executable while
1220 	 * the vnode is unlocked.
1221 	 */
1222 	VOP_UNLOCK(imgp->vp);
1223 
1224 	/*
1225 	 * Decide whether to enable randomization of user mappings.
1226 	 * First, reset user preferences for the setid binaries.
1227 	 * Then, account for the support of the randomization by the
1228 	 * ABI, by user preferences, and make special treatment for
1229 	 * PIE binaries.
1230 	 */
1231 	if (imgp->credential_setid) {
1232 		PROC_LOCK(imgp->proc);
1233 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1234 		    P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1235 		PROC_UNLOCK(imgp->proc);
1236 	}
1237 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1238 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1239 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1240 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1241 		    ("et_dyn_addr == RAND and !ASLR"));
1242 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1243 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1244 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1245 		imgp->map_flags |= MAP_ASLR;
1246 		/*
1247 		 * If user does not care about sbrk, utilize the bss
1248 		 * grow region for mappings as well.  We can select
1249 		 * the base for the image anywere and still not suffer
1250 		 * from the fragmentation.
1251 		 */
1252 		if (!__elfN(aslr_honor_sbrk) ||
1253 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1254 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1255 	}
1256 
1257 	if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1258 	    (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1259 	    (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1260 		imgp->map_flags |= MAP_WXORX;
1261 
1262 	error = exec_new_vmspace(imgp, sv);
1263 	vmspace = imgp->proc->p_vmspace;
1264 	map = &vmspace->vm_map;
1265 
1266 	imgp->proc->p_sysent = sv;
1267 	imgp->proc->p_elf_brandinfo = brand_info;
1268 
1269 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1270 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1271 		KASSERT((map->flags & MAP_ASLR) != 0,
1272 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1273 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1274 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1275 		    /* reserve half of the address space to interpreter */
1276 		    maxv / 2, 1UL << flsl(maxalign));
1277 	}
1278 
1279 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1280 	if (error != 0)
1281 		goto ret;
1282 
1283 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1284 	if (error != 0)
1285 		goto ret;
1286 
1287 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1288 	if (error != 0)
1289 		goto ret;
1290 
1291 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1292 
1293 	/*
1294 	 * We load the dynamic linker where a userland call
1295 	 * to mmap(0, ...) would put it.  The rationale behind this
1296 	 * calculation is that it leaves room for the heap to grow to
1297 	 * its maximum allowed size.
1298 	 */
1299 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1300 	    RLIMIT_DATA));
1301 	if ((map->flags & MAP_ASLR) != 0) {
1302 		maxv1 = maxv / 2 + addr / 2;
1303 		MPASS(maxv1 >= addr);	/* No overflow */
1304 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1305 		    (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1306 		    pagesizes[1] : pagesizes[0]);
1307 	} else {
1308 		map->anon_loc = addr;
1309 	}
1310 
1311 	imgp->entry_addr = entry;
1312 
1313 	if (interp != NULL) {
1314 		VOP_UNLOCK(imgp->vp);
1315 		if ((map->flags & MAP_ASLR) != 0) {
1316 			/* Assume that interpreter fits into 1/4 of AS */
1317 			maxv1 = maxv / 2 + addr / 2;
1318 			MPASS(maxv1 >= addr);	/* No overflow */
1319 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1320 			    maxv1, PAGE_SIZE);
1321 		}
1322 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1323 		    &imgp->entry_addr);
1324 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1325 		if (error != 0)
1326 			goto ret;
1327 	} else
1328 		addr = et_dyn_addr;
1329 
1330 	/*
1331 	 * Construct auxargs table (used by the copyout_auxargs routine)
1332 	 */
1333 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1334 	if (elf_auxargs == NULL) {
1335 		VOP_UNLOCK(imgp->vp);
1336 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1337 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1338 	}
1339 	elf_auxargs->execfd = -1;
1340 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1341 	elf_auxargs->phent = hdr->e_phentsize;
1342 	elf_auxargs->phnum = hdr->e_phnum;
1343 	elf_auxargs->pagesz = PAGE_SIZE;
1344 	elf_auxargs->base = addr;
1345 	elf_auxargs->flags = 0;
1346 	elf_auxargs->entry = entry;
1347 	elf_auxargs->hdr_eflags = hdr->e_flags;
1348 
1349 	imgp->auxargs = elf_auxargs;
1350 	imgp->interpreted = 0;
1351 	imgp->reloc_base = addr;
1352 	imgp->proc->p_osrel = osrel;
1353 	imgp->proc->p_fctl0 = fctl0;
1354 	imgp->proc->p_elf_flags = hdr->e_flags;
1355 
1356 ret:
1357 	if (free_interp)
1358 		free(interp, M_TEMP);
1359 	return (error);
1360 }
1361 
1362 #define	elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1363 
1364 int
1365 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1366 {
1367 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1368 	Elf_Auxinfo *argarray, *pos;
1369 	int error;
1370 
1371 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1372 	    M_WAITOK | M_ZERO);
1373 
1374 	if (args->execfd != -1)
1375 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1376 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1377 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1378 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1379 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1380 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1381 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1382 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1383 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1384 	if (imgp->execpathp != 0)
1385 		AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1386 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1387 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1388 	if (imgp->canary != 0) {
1389 		AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1390 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1391 	}
1392 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1393 	if (imgp->pagesizes != 0) {
1394 		AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1395 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1396 	}
1397 	if (imgp->sysent->sv_timekeep_base != 0) {
1398 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1399 		    imgp->sysent->sv_timekeep_base);
1400 	}
1401 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1402 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1403 	    imgp->sysent->sv_stackprot);
1404 	if (imgp->sysent->sv_hwcap != NULL)
1405 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1406 	if (imgp->sysent->sv_hwcap2 != NULL)
1407 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1408 	AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1409 	    ELF_BSDF_SIGFASTBLK : 0);
1410 	AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1411 	AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1412 	AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1413 	AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1414 	AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1415 	if (imgp->sysent->sv_fxrng_gen_base != 0)
1416 		AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1417 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1418 
1419 	free(imgp->auxargs, M_TEMP);
1420 	imgp->auxargs = NULL;
1421 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1422 
1423 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1424 	free(argarray, M_TEMP);
1425 	return (error);
1426 }
1427 
1428 int
1429 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1430 {
1431 	Elf_Addr *base;
1432 
1433 	base = (Elf_Addr *)*stack_base;
1434 	base--;
1435 	if (elf_suword(base, imgp->args->argc) == -1)
1436 		return (EFAULT);
1437 	*stack_base = (uintptr_t)base;
1438 	return (0);
1439 }
1440 
1441 /*
1442  * Code for generating ELF core dumps.
1443  */
1444 
1445 typedef void (*segment_callback)(vm_map_entry_t, void *);
1446 
1447 /* Closure for cb_put_phdr(). */
1448 struct phdr_closure {
1449 	Elf_Phdr *phdr;		/* Program header to fill in */
1450 	Elf_Off offset;		/* Offset of segment in core file */
1451 };
1452 
1453 struct note_info {
1454 	int		type;		/* Note type. */
1455 	outfunc_t 	outfunc; 	/* Output function. */
1456 	void		*outarg;	/* Argument for the output function. */
1457 	size_t		outsize;	/* Output size. */
1458 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1459 };
1460 
1461 TAILQ_HEAD(note_info_list, note_info);
1462 
1463 extern int compress_user_cores;
1464 extern int compress_user_cores_level;
1465 
1466 static void cb_put_phdr(vm_map_entry_t, void *);
1467 static void cb_size_segment(vm_map_entry_t, void *);
1468 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1469     int);
1470 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1471     struct note_info_list *, size_t, int);
1472 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1473 
1474 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1475 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1476 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1477 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1478 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1479 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1480 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1481 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1482 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1483 static void note_procstat_files(void *, struct sbuf *, size_t *);
1484 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1485 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1486 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1487 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1488 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1489 
1490 static int
1491 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1492 {
1493 
1494 	return (core_write((struct coredump_params *)arg, base, len, offset,
1495 	    UIO_SYSSPACE, NULL));
1496 }
1497 
1498 int
1499 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1500 {
1501 	struct ucred *cred = td->td_ucred;
1502 	int compm, error = 0;
1503 	struct sseg_closure seginfo;
1504 	struct note_info_list notelst;
1505 	struct coredump_params params;
1506 	struct note_info *ninfo;
1507 	void *hdr, *tmpbuf;
1508 	size_t hdrsize, notesz, coresize;
1509 
1510 	hdr = NULL;
1511 	tmpbuf = NULL;
1512 	TAILQ_INIT(&notelst);
1513 
1514 	/* Size the program segments. */
1515 	__elfN(size_segments)(td, &seginfo, flags);
1516 
1517 	/*
1518 	 * Collect info about the core file header area.
1519 	 */
1520 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1521 	if (seginfo.count + 1 >= PN_XNUM)
1522 		hdrsize += sizeof(Elf_Shdr);
1523 	td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, &notelst, &notesz);
1524 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1525 
1526 	/* Set up core dump parameters. */
1527 	params.offset = 0;
1528 	params.active_cred = cred;
1529 	params.file_cred = NOCRED;
1530 	params.td = td;
1531 	params.vp = vp;
1532 	params.comp = NULL;
1533 
1534 #ifdef RACCT
1535 	if (racct_enable) {
1536 		PROC_LOCK(td->td_proc);
1537 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1538 		PROC_UNLOCK(td->td_proc);
1539 		if (error != 0) {
1540 			error = EFAULT;
1541 			goto done;
1542 		}
1543 	}
1544 #endif
1545 	if (coresize >= limit) {
1546 		error = EFAULT;
1547 		goto done;
1548 	}
1549 
1550 	/* Create a compression stream if necessary. */
1551 	compm = compress_user_cores;
1552 	if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1553 	    compm == 0)
1554 		compm = COMPRESS_GZIP;
1555 	if (compm != 0) {
1556 		params.comp = compressor_init(core_compressed_write,
1557 		    compm, CORE_BUF_SIZE,
1558 		    compress_user_cores_level, &params);
1559 		if (params.comp == NULL) {
1560 			error = EFAULT;
1561 			goto done;
1562 		}
1563 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1564         }
1565 
1566 	/*
1567 	 * Allocate memory for building the header, fill it up,
1568 	 * and write it out following the notes.
1569 	 */
1570 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1571 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1572 	    notesz, flags);
1573 
1574 	/* Write the contents of all of the writable segments. */
1575 	if (error == 0) {
1576 		Elf_Phdr *php;
1577 		off_t offset;
1578 		int i;
1579 
1580 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1581 		offset = round_page(hdrsize + notesz);
1582 		for (i = 0; i < seginfo.count; i++) {
1583 			error = core_output((char *)(uintptr_t)php->p_vaddr,
1584 			    php->p_filesz, offset, &params, tmpbuf);
1585 			if (error != 0)
1586 				break;
1587 			offset += php->p_filesz;
1588 			php++;
1589 		}
1590 		if (error == 0 && params.comp != NULL)
1591 			error = compressor_flush(params.comp);
1592 	}
1593 	if (error) {
1594 		log(LOG_WARNING,
1595 		    "Failed to write core file for process %s (error %d)\n",
1596 		    curproc->p_comm, error);
1597 	}
1598 
1599 done:
1600 	free(tmpbuf, M_TEMP);
1601 	if (params.comp != NULL)
1602 		compressor_fini(params.comp);
1603 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1604 		TAILQ_REMOVE(&notelst, ninfo, link);
1605 		free(ninfo, M_TEMP);
1606 	}
1607 	if (hdr != NULL)
1608 		free(hdr, M_TEMP);
1609 
1610 	return (error);
1611 }
1612 
1613 /*
1614  * A callback for each_dumpable_segment() to write out the segment's
1615  * program header entry.
1616  */
1617 static void
1618 cb_put_phdr(vm_map_entry_t entry, void *closure)
1619 {
1620 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1621 	Elf_Phdr *phdr = phc->phdr;
1622 
1623 	phc->offset = round_page(phc->offset);
1624 
1625 	phdr->p_type = PT_LOAD;
1626 	phdr->p_offset = phc->offset;
1627 	phdr->p_vaddr = entry->start;
1628 	phdr->p_paddr = 0;
1629 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1630 	phdr->p_align = PAGE_SIZE;
1631 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1632 
1633 	phc->offset += phdr->p_filesz;
1634 	phc->phdr++;
1635 }
1636 
1637 /*
1638  * A callback for each_dumpable_segment() to gather information about
1639  * the number of segments and their total size.
1640  */
1641 static void
1642 cb_size_segment(vm_map_entry_t entry, void *closure)
1643 {
1644 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1645 
1646 	ssc->count++;
1647 	ssc->size += entry->end - entry->start;
1648 }
1649 
1650 void
1651 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1652     int flags)
1653 {
1654 	seginfo->count = 0;
1655 	seginfo->size = 0;
1656 
1657 	each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1658 }
1659 
1660 /*
1661  * For each writable segment in the process's memory map, call the given
1662  * function with a pointer to the map entry and some arbitrary
1663  * caller-supplied data.
1664  */
1665 static void
1666 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1667     int flags)
1668 {
1669 	struct proc *p = td->td_proc;
1670 	vm_map_t map = &p->p_vmspace->vm_map;
1671 	vm_map_entry_t entry;
1672 	vm_object_t backing_object, object;
1673 	bool ignore_entry;
1674 
1675 	vm_map_lock_read(map);
1676 	VM_MAP_ENTRY_FOREACH(entry, map) {
1677 		/*
1678 		 * Don't dump inaccessible mappings, deal with legacy
1679 		 * coredump mode.
1680 		 *
1681 		 * Note that read-only segments related to the elf binary
1682 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1683 		 * need to arbitrarily ignore such segments.
1684 		 */
1685 		if ((flags & SVC_ALL) == 0) {
1686 			if (elf_legacy_coredump) {
1687 				if ((entry->protection & VM_PROT_RW) !=
1688 				    VM_PROT_RW)
1689 					continue;
1690 			} else {
1691 				if ((entry->protection & VM_PROT_ALL) == 0)
1692 					continue;
1693 			}
1694 		}
1695 
1696 		/*
1697 		 * Dont include memory segment in the coredump if
1698 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1699 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1700 		 * kernel map).
1701 		 */
1702 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1703 			continue;
1704 		if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1705 		    (flags & SVC_ALL) == 0)
1706 			continue;
1707 		if ((object = entry->object.vm_object) == NULL)
1708 			continue;
1709 
1710 		/* Ignore memory-mapped devices and such things. */
1711 		VM_OBJECT_RLOCK(object);
1712 		while ((backing_object = object->backing_object) != NULL) {
1713 			VM_OBJECT_RLOCK(backing_object);
1714 			VM_OBJECT_RUNLOCK(object);
1715 			object = backing_object;
1716 		}
1717 		ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1718 		VM_OBJECT_RUNLOCK(object);
1719 		if (ignore_entry)
1720 			continue;
1721 
1722 		(*func)(entry, closure);
1723 	}
1724 	vm_map_unlock_read(map);
1725 }
1726 
1727 /*
1728  * Write the core file header to the file, including padding up to
1729  * the page boundary.
1730  */
1731 static int
1732 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1733     size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1734     int flags)
1735 {
1736 	struct note_info *ninfo;
1737 	struct sbuf *sb;
1738 	int error;
1739 
1740 	/* Fill in the header. */
1741 	bzero(hdr, hdrsize);
1742 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1743 
1744 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1745 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1746 	sbuf_start_section(sb, NULL);
1747 	sbuf_bcat(sb, hdr, hdrsize);
1748 	TAILQ_FOREACH(ninfo, notelst, link)
1749 	    __elfN(putnote)(p->td, ninfo, sb);
1750 	/* Align up to a page boundary for the program segments. */
1751 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1752 	error = sbuf_finish(sb);
1753 	sbuf_delete(sb);
1754 
1755 	return (error);
1756 }
1757 
1758 void
1759 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1760     size_t *sizep)
1761 {
1762 	struct proc *p;
1763 	struct thread *thr;
1764 	size_t size;
1765 
1766 	p = td->td_proc;
1767 	size = 0;
1768 
1769 	size += __elfN(register_note)(td, list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1770 
1771 	/*
1772 	 * To have the debugger select the right thread (LWP) as the initial
1773 	 * thread, we dump the state of the thread passed to us in td first.
1774 	 * This is the thread that causes the core dump and thus likely to
1775 	 * be the right thread one wants to have selected in the debugger.
1776 	 */
1777 	thr = td;
1778 	while (thr != NULL) {
1779 		size += __elfN(register_note)(td, list, NT_PRSTATUS,
1780 		    __elfN(note_prstatus), thr);
1781 		size += __elfN(register_note)(td, list, NT_FPREGSET,
1782 		    __elfN(note_fpregset), thr);
1783 		size += __elfN(register_note)(td, list, NT_THRMISC,
1784 		    __elfN(note_thrmisc), thr);
1785 		size += __elfN(register_note)(td, list, NT_PTLWPINFO,
1786 		    __elfN(note_ptlwpinfo), thr);
1787 		size += __elfN(register_note)(td, list, -1,
1788 		    __elfN(note_threadmd), thr);
1789 
1790 		thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1791 		    TAILQ_NEXT(thr, td_plist);
1792 		if (thr == td)
1793 			thr = TAILQ_NEXT(thr, td_plist);
1794 	}
1795 
1796 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1797 	    __elfN(note_procstat_proc), p);
1798 	size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1799 	    note_procstat_files, p);
1800 	size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1801 	    note_procstat_vmmap, p);
1802 	size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1803 	    note_procstat_groups, p);
1804 	size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1805 	    note_procstat_umask, p);
1806 	size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1807 	    note_procstat_rlimit, p);
1808 	size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1809 	    note_procstat_osrel, p);
1810 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1811 	    __elfN(note_procstat_psstrings), p);
1812 	size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1813 	    __elfN(note_procstat_auxv), p);
1814 
1815 	*sizep = size;
1816 }
1817 
1818 void
1819 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1820     size_t notesz, int flags)
1821 {
1822 	Elf_Ehdr *ehdr;
1823 	Elf_Phdr *phdr;
1824 	Elf_Shdr *shdr;
1825 	struct phdr_closure phc;
1826 	Elf_Brandinfo *bi;
1827 
1828 	ehdr = (Elf_Ehdr *)hdr;
1829 	bi = td->td_proc->p_elf_brandinfo;
1830 
1831 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1832 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1833 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1834 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1835 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1836 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1837 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1838 	ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1839 	ehdr->e_ident[EI_ABIVERSION] = 0;
1840 	ehdr->e_ident[EI_PAD] = 0;
1841 	ehdr->e_type = ET_CORE;
1842 	ehdr->e_machine = bi->machine;
1843 	ehdr->e_version = EV_CURRENT;
1844 	ehdr->e_entry = 0;
1845 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1846 	ehdr->e_flags = td->td_proc->p_elf_flags;
1847 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1848 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1849 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1850 	ehdr->e_shstrndx = SHN_UNDEF;
1851 	if (numsegs + 1 < PN_XNUM) {
1852 		ehdr->e_phnum = numsegs + 1;
1853 		ehdr->e_shnum = 0;
1854 	} else {
1855 		ehdr->e_phnum = PN_XNUM;
1856 		ehdr->e_shnum = 1;
1857 
1858 		ehdr->e_shoff = ehdr->e_phoff +
1859 		    (numsegs + 1) * ehdr->e_phentsize;
1860 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1861 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1862 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1863 
1864 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1865 		memset(shdr, 0, sizeof(*shdr));
1866 		/*
1867 		 * A special first section is used to hold large segment and
1868 		 * section counts.  This was proposed by Sun Microsystems in
1869 		 * Solaris and has been adopted by Linux; the standard ELF
1870 		 * tools are already familiar with the technique.
1871 		 *
1872 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1873 		 * (or 12-7 depending on the version of the document) for more
1874 		 * details.
1875 		 */
1876 		shdr->sh_type = SHT_NULL;
1877 		shdr->sh_size = ehdr->e_shnum;
1878 		shdr->sh_link = ehdr->e_shstrndx;
1879 		shdr->sh_info = numsegs + 1;
1880 	}
1881 
1882 	/*
1883 	 * Fill in the program header entries.
1884 	 */
1885 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1886 
1887 	/* The note segement. */
1888 	phdr->p_type = PT_NOTE;
1889 	phdr->p_offset = hdrsize;
1890 	phdr->p_vaddr = 0;
1891 	phdr->p_paddr = 0;
1892 	phdr->p_filesz = notesz;
1893 	phdr->p_memsz = 0;
1894 	phdr->p_flags = PF_R;
1895 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1896 	phdr++;
1897 
1898 	/* All the writable segments from the program. */
1899 	phc.phdr = phdr;
1900 	phc.offset = round_page(hdrsize + notesz);
1901 	each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1902 }
1903 
1904 size_t
1905 __elfN(register_note)(struct thread *td, struct note_info_list *list,
1906     int type, outfunc_t out, void *arg)
1907 {
1908 	const struct sysentvec *sv;
1909 	struct note_info *ninfo;
1910 	size_t size, notesize;
1911 
1912 	sv = td->td_proc->p_sysent;
1913 	size = 0;
1914 	out(arg, NULL, &size);
1915 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1916 	ninfo->type = type;
1917 	ninfo->outfunc = out;
1918 	ninfo->outarg = arg;
1919 	ninfo->outsize = size;
1920 	TAILQ_INSERT_TAIL(list, ninfo, link);
1921 
1922 	if (type == -1)
1923 		return (size);
1924 
1925 	notesize = sizeof(Elf_Note) +		/* note header */
1926 	    roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1927 						/* note name */
1928 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1929 
1930 	return (notesize);
1931 }
1932 
1933 static size_t
1934 append_note_data(const void *src, void *dst, size_t len)
1935 {
1936 	size_t padded_len;
1937 
1938 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1939 	if (dst != NULL) {
1940 		bcopy(src, dst, len);
1941 		bzero((char *)dst + len, padded_len - len);
1942 	}
1943 	return (padded_len);
1944 }
1945 
1946 size_t
1947 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1948 {
1949 	Elf_Note *note;
1950 	char *buf;
1951 	size_t notesize;
1952 
1953 	buf = dst;
1954 	if (buf != NULL) {
1955 		note = (Elf_Note *)buf;
1956 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1957 		note->n_descsz = size;
1958 		note->n_type = type;
1959 		buf += sizeof(*note);
1960 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1961 		    sizeof(FREEBSD_ABI_VENDOR));
1962 		append_note_data(src, buf, size);
1963 		if (descp != NULL)
1964 			*descp = buf;
1965 	}
1966 
1967 	notesize = sizeof(Elf_Note) +		/* note header */
1968 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1969 						/* note name */
1970 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1971 
1972 	return (notesize);
1973 }
1974 
1975 static void
1976 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
1977 {
1978 	Elf_Note note;
1979 	const struct sysentvec *sv;
1980 	ssize_t old_len, sect_len;
1981 	size_t new_len, descsz, i;
1982 
1983 	if (ninfo->type == -1) {
1984 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1985 		return;
1986 	}
1987 
1988 	sv = td->td_proc->p_sysent;
1989 
1990 	note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
1991 	note.n_descsz = ninfo->outsize;
1992 	note.n_type = ninfo->type;
1993 
1994 	sbuf_bcat(sb, &note, sizeof(note));
1995 	sbuf_start_section(sb, &old_len);
1996 	sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
1997 	    strlen(sv->sv_elf_core_abi_vendor) + 1);
1998 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1999 	if (note.n_descsz == 0)
2000 		return;
2001 	sbuf_start_section(sb, &old_len);
2002 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2003 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2004 	if (sect_len < 0)
2005 		return;
2006 
2007 	new_len = (size_t)sect_len;
2008 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2009 	if (new_len < descsz) {
2010 		/*
2011 		 * It is expected that individual note emitters will correctly
2012 		 * predict their expected output size and fill up to that size
2013 		 * themselves, padding in a format-specific way if needed.
2014 		 * However, in case they don't, just do it here with zeros.
2015 		 */
2016 		for (i = 0; i < descsz - new_len; i++)
2017 			sbuf_putc(sb, 0);
2018 	} else if (new_len > descsz) {
2019 		/*
2020 		 * We can't always truncate sb -- we may have drained some
2021 		 * of it already.
2022 		 */
2023 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2024 		    "read it (%zu > %zu).  Since it is longer than "
2025 		    "expected, this coredump's notes are corrupt.  THIS "
2026 		    "IS A BUG in the note_procstat routine for type %u.\n",
2027 		    __func__, (unsigned)note.n_type, new_len, descsz,
2028 		    (unsigned)note.n_type));
2029 	}
2030 }
2031 
2032 /*
2033  * Miscellaneous note out functions.
2034  */
2035 
2036 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2037 #include <compat/freebsd32/freebsd32.h>
2038 #include <compat/freebsd32/freebsd32_signal.h>
2039 
2040 typedef struct prstatus32 elf_prstatus_t;
2041 typedef struct prpsinfo32 elf_prpsinfo_t;
2042 typedef struct fpreg32 elf_prfpregset_t;
2043 typedef struct fpreg32 elf_fpregset_t;
2044 typedef struct reg32 elf_gregset_t;
2045 typedef struct thrmisc32 elf_thrmisc_t;
2046 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2047 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2048 typedef uint32_t elf_ps_strings_t;
2049 #else
2050 typedef prstatus_t elf_prstatus_t;
2051 typedef prpsinfo_t elf_prpsinfo_t;
2052 typedef prfpregset_t elf_prfpregset_t;
2053 typedef prfpregset_t elf_fpregset_t;
2054 typedef gregset_t elf_gregset_t;
2055 typedef thrmisc_t elf_thrmisc_t;
2056 #define ELF_KERN_PROC_MASK	0
2057 typedef struct kinfo_proc elf_kinfo_proc_t;
2058 typedef vm_offset_t elf_ps_strings_t;
2059 #endif
2060 
2061 static void
2062 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2063 {
2064 	struct sbuf sbarg;
2065 	size_t len;
2066 	char *cp, *end;
2067 	struct proc *p;
2068 	elf_prpsinfo_t *psinfo;
2069 	int error;
2070 
2071 	p = arg;
2072 	if (sb != NULL) {
2073 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2074 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2075 		psinfo->pr_version = PRPSINFO_VERSION;
2076 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2077 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2078 		PROC_LOCK(p);
2079 		if (p->p_args != NULL) {
2080 			len = sizeof(psinfo->pr_psargs) - 1;
2081 			if (len > p->p_args->ar_length)
2082 				len = p->p_args->ar_length;
2083 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2084 			PROC_UNLOCK(p);
2085 			error = 0;
2086 		} else {
2087 			_PHOLD(p);
2088 			PROC_UNLOCK(p);
2089 			sbuf_new(&sbarg, psinfo->pr_psargs,
2090 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2091 			error = proc_getargv(curthread, p, &sbarg);
2092 			PRELE(p);
2093 			if (sbuf_finish(&sbarg) == 0)
2094 				len = sbuf_len(&sbarg) - 1;
2095 			else
2096 				len = sizeof(psinfo->pr_psargs) - 1;
2097 			sbuf_delete(&sbarg);
2098 		}
2099 		if (error || len == 0)
2100 			strlcpy(psinfo->pr_psargs, p->p_comm,
2101 			    sizeof(psinfo->pr_psargs));
2102 		else {
2103 			KASSERT(len < sizeof(psinfo->pr_psargs),
2104 			    ("len is too long: %zu vs %zu", len,
2105 			    sizeof(psinfo->pr_psargs)));
2106 			cp = psinfo->pr_psargs;
2107 			end = cp + len - 1;
2108 			for (;;) {
2109 				cp = memchr(cp, '\0', end - cp);
2110 				if (cp == NULL)
2111 					break;
2112 				*cp = ' ';
2113 			}
2114 		}
2115 		psinfo->pr_pid = p->p_pid;
2116 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2117 		free(psinfo, M_TEMP);
2118 	}
2119 	*sizep = sizeof(*psinfo);
2120 }
2121 
2122 static void
2123 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2124 {
2125 	struct thread *td;
2126 	elf_prstatus_t *status;
2127 
2128 	td = arg;
2129 	if (sb != NULL) {
2130 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2131 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2132 		status->pr_version = PRSTATUS_VERSION;
2133 		status->pr_statussz = sizeof(elf_prstatus_t);
2134 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2135 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2136 		status->pr_osreldate = osreldate;
2137 		status->pr_cursig = td->td_proc->p_sig;
2138 		status->pr_pid = td->td_tid;
2139 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2140 		fill_regs32(td, &status->pr_reg);
2141 #else
2142 		fill_regs(td, &status->pr_reg);
2143 #endif
2144 		sbuf_bcat(sb, status, sizeof(*status));
2145 		free(status, M_TEMP);
2146 	}
2147 	*sizep = sizeof(*status);
2148 }
2149 
2150 static void
2151 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2152 {
2153 	struct thread *td;
2154 	elf_prfpregset_t *fpregset;
2155 
2156 	td = arg;
2157 	if (sb != NULL) {
2158 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2159 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2160 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2161 		fill_fpregs32(td, fpregset);
2162 #else
2163 		fill_fpregs(td, fpregset);
2164 #endif
2165 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2166 		free(fpregset, M_TEMP);
2167 	}
2168 	*sizep = sizeof(*fpregset);
2169 }
2170 
2171 static void
2172 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2173 {
2174 	struct thread *td;
2175 	elf_thrmisc_t thrmisc;
2176 
2177 	td = arg;
2178 	if (sb != NULL) {
2179 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2180 		bzero(&thrmisc, sizeof(thrmisc));
2181 		strcpy(thrmisc.pr_tname, td->td_name);
2182 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2183 	}
2184 	*sizep = sizeof(thrmisc);
2185 }
2186 
2187 static void
2188 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2189 {
2190 	struct thread *td;
2191 	size_t size;
2192 	int structsize;
2193 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2194 	struct ptrace_lwpinfo32 pl;
2195 #else
2196 	struct ptrace_lwpinfo pl;
2197 #endif
2198 
2199 	td = arg;
2200 	size = sizeof(structsize) + sizeof(pl);
2201 	if (sb != NULL) {
2202 		KASSERT(*sizep == size, ("invalid size"));
2203 		structsize = sizeof(pl);
2204 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2205 		bzero(&pl, sizeof(pl));
2206 		pl.pl_lwpid = td->td_tid;
2207 		pl.pl_event = PL_EVENT_NONE;
2208 		pl.pl_sigmask = td->td_sigmask;
2209 		pl.pl_siglist = td->td_siglist;
2210 		if (td->td_si.si_signo != 0) {
2211 			pl.pl_event = PL_EVENT_SIGNAL;
2212 			pl.pl_flags |= PL_FLAG_SI;
2213 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2214 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2215 #else
2216 			pl.pl_siginfo = td->td_si;
2217 #endif
2218 		}
2219 		strcpy(pl.pl_tdname, td->td_name);
2220 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2221 		sbuf_bcat(sb, &pl, sizeof(pl));
2222 	}
2223 	*sizep = size;
2224 }
2225 
2226 /*
2227  * Allow for MD specific notes, as well as any MD
2228  * specific preparations for writing MI notes.
2229  */
2230 static void
2231 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2232 {
2233 	struct thread *td;
2234 	void *buf;
2235 	size_t size;
2236 
2237 	td = (struct thread *)arg;
2238 	size = *sizep;
2239 	if (size != 0 && sb != NULL)
2240 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2241 	else
2242 		buf = NULL;
2243 	size = 0;
2244 	__elfN(dump_thread)(td, buf, &size);
2245 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2246 	if (size != 0 && sb != NULL)
2247 		sbuf_bcat(sb, buf, size);
2248 	free(buf, M_TEMP);
2249 	*sizep = size;
2250 }
2251 
2252 #ifdef KINFO_PROC_SIZE
2253 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2254 #endif
2255 
2256 static void
2257 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2258 {
2259 	struct proc *p;
2260 	size_t size;
2261 	int structsize;
2262 
2263 	p = arg;
2264 	size = sizeof(structsize) + p->p_numthreads *
2265 	    sizeof(elf_kinfo_proc_t);
2266 
2267 	if (sb != NULL) {
2268 		KASSERT(*sizep == size, ("invalid size"));
2269 		structsize = sizeof(elf_kinfo_proc_t);
2270 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2271 		sx_slock(&proctree_lock);
2272 		PROC_LOCK(p);
2273 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2274 		sx_sunlock(&proctree_lock);
2275 	}
2276 	*sizep = size;
2277 }
2278 
2279 #ifdef KINFO_FILE_SIZE
2280 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2281 #endif
2282 
2283 static void
2284 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2285 {
2286 	struct proc *p;
2287 	size_t size, sect_sz, i;
2288 	ssize_t start_len, sect_len;
2289 	int structsize, filedesc_flags;
2290 
2291 	if (coredump_pack_fileinfo)
2292 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2293 	else
2294 		filedesc_flags = 0;
2295 
2296 	p = arg;
2297 	structsize = sizeof(struct kinfo_file);
2298 	if (sb == NULL) {
2299 		size = 0;
2300 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2301 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2302 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2303 		PROC_LOCK(p);
2304 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2305 		sbuf_finish(sb);
2306 		sbuf_delete(sb);
2307 		*sizep = size;
2308 	} else {
2309 		sbuf_start_section(sb, &start_len);
2310 
2311 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2312 		PROC_LOCK(p);
2313 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2314 		    filedesc_flags);
2315 
2316 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2317 		if (sect_len < 0)
2318 			return;
2319 		sect_sz = sect_len;
2320 
2321 		KASSERT(sect_sz <= *sizep,
2322 		    ("kern_proc_filedesc_out did not respect maxlen; "
2323 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2324 		     sect_sz - sizeof(structsize)));
2325 
2326 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2327 			sbuf_putc(sb, 0);
2328 	}
2329 }
2330 
2331 #ifdef KINFO_VMENTRY_SIZE
2332 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2333 #endif
2334 
2335 static void
2336 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2337 {
2338 	struct proc *p;
2339 	size_t size;
2340 	int structsize, vmmap_flags;
2341 
2342 	if (coredump_pack_vmmapinfo)
2343 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2344 	else
2345 		vmmap_flags = 0;
2346 
2347 	p = arg;
2348 	structsize = sizeof(struct kinfo_vmentry);
2349 	if (sb == NULL) {
2350 		size = 0;
2351 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2352 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2353 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2354 		PROC_LOCK(p);
2355 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2356 		sbuf_finish(sb);
2357 		sbuf_delete(sb);
2358 		*sizep = size;
2359 	} else {
2360 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2361 		PROC_LOCK(p);
2362 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2363 		    vmmap_flags);
2364 	}
2365 }
2366 
2367 static void
2368 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2369 {
2370 	struct proc *p;
2371 	size_t size;
2372 	int structsize;
2373 
2374 	p = arg;
2375 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2376 	if (sb != NULL) {
2377 		KASSERT(*sizep == size, ("invalid size"));
2378 		structsize = sizeof(gid_t);
2379 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2380 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2381 		    sizeof(gid_t));
2382 	}
2383 	*sizep = size;
2384 }
2385 
2386 static void
2387 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2388 {
2389 	struct proc *p;
2390 	size_t size;
2391 	int structsize;
2392 
2393 	p = arg;
2394 	size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2395 	if (sb != NULL) {
2396 		KASSERT(*sizep == size, ("invalid size"));
2397 		structsize = sizeof(p->p_pd->pd_cmask);
2398 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2399 		sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2400 	}
2401 	*sizep = size;
2402 }
2403 
2404 static void
2405 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2406 {
2407 	struct proc *p;
2408 	struct rlimit rlim[RLIM_NLIMITS];
2409 	size_t size;
2410 	int structsize, i;
2411 
2412 	p = arg;
2413 	size = sizeof(structsize) + sizeof(rlim);
2414 	if (sb != NULL) {
2415 		KASSERT(*sizep == size, ("invalid size"));
2416 		structsize = sizeof(rlim);
2417 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2418 		PROC_LOCK(p);
2419 		for (i = 0; i < RLIM_NLIMITS; i++)
2420 			lim_rlimit_proc(p, i, &rlim[i]);
2421 		PROC_UNLOCK(p);
2422 		sbuf_bcat(sb, rlim, sizeof(rlim));
2423 	}
2424 	*sizep = size;
2425 }
2426 
2427 static void
2428 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2429 {
2430 	struct proc *p;
2431 	size_t size;
2432 	int structsize;
2433 
2434 	p = arg;
2435 	size = sizeof(structsize) + sizeof(p->p_osrel);
2436 	if (sb != NULL) {
2437 		KASSERT(*sizep == size, ("invalid size"));
2438 		structsize = sizeof(p->p_osrel);
2439 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2440 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2441 	}
2442 	*sizep = size;
2443 }
2444 
2445 static void
2446 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2447 {
2448 	struct proc *p;
2449 	elf_ps_strings_t ps_strings;
2450 	size_t size;
2451 	int structsize;
2452 
2453 	p = arg;
2454 	size = sizeof(structsize) + sizeof(ps_strings);
2455 	if (sb != NULL) {
2456 		KASSERT(*sizep == size, ("invalid size"));
2457 		structsize = sizeof(ps_strings);
2458 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2459 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2460 #else
2461 		ps_strings = p->p_sysent->sv_psstrings;
2462 #endif
2463 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2464 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2465 	}
2466 	*sizep = size;
2467 }
2468 
2469 static void
2470 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2471 {
2472 	struct proc *p;
2473 	size_t size;
2474 	int structsize;
2475 
2476 	p = arg;
2477 	if (sb == NULL) {
2478 		size = 0;
2479 		sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2480 		    SBUF_FIXEDLEN);
2481 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2482 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2483 		PHOLD(p);
2484 		proc_getauxv(curthread, p, sb);
2485 		PRELE(p);
2486 		sbuf_finish(sb);
2487 		sbuf_delete(sb);
2488 		*sizep = size;
2489 	} else {
2490 		structsize = sizeof(Elf_Auxinfo);
2491 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2492 		PHOLD(p);
2493 		proc_getauxv(curthread, p, sb);
2494 		PRELE(p);
2495 	}
2496 }
2497 
2498 static bool
2499 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2500     const char *note_vendor, const Elf_Phdr *pnote,
2501     bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2502 {
2503 	const Elf_Note *note, *note0, *note_end;
2504 	const char *note_name;
2505 	char *buf;
2506 	int i, error;
2507 	bool res;
2508 
2509 	/* We need some limit, might as well use PAGE_SIZE. */
2510 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2511 		return (false);
2512 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2513 	if (pnote->p_offset > PAGE_SIZE ||
2514 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2515 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2516 		if (buf == NULL) {
2517 			VOP_UNLOCK(imgp->vp);
2518 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2519 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2520 		}
2521 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2522 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2523 		    curthread->td_ucred, NOCRED, NULL, curthread);
2524 		if (error != 0) {
2525 			uprintf("i/o error PT_NOTE\n");
2526 			goto retf;
2527 		}
2528 		note = note0 = (const Elf_Note *)buf;
2529 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2530 	} else {
2531 		note = note0 = (const Elf_Note *)(imgp->image_header +
2532 		    pnote->p_offset);
2533 		note_end = (const Elf_Note *)(imgp->image_header +
2534 		    pnote->p_offset + pnote->p_filesz);
2535 		buf = NULL;
2536 	}
2537 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2538 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2539 		    (const char *)note < sizeof(Elf_Note)) {
2540 			goto retf;
2541 		}
2542 		if (note->n_namesz != checknote->n_namesz ||
2543 		    note->n_descsz != checknote->n_descsz ||
2544 		    note->n_type != checknote->n_type)
2545 			goto nextnote;
2546 		note_name = (const char *)(note + 1);
2547 		if (note_name + checknote->n_namesz >=
2548 		    (const char *)note_end || strncmp(note_vendor,
2549 		    note_name, checknote->n_namesz) != 0)
2550 			goto nextnote;
2551 
2552 		if (cb(note, cb_arg, &res))
2553 			goto ret;
2554 nextnote:
2555 		note = (const Elf_Note *)((const char *)(note + 1) +
2556 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2557 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2558 	}
2559 retf:
2560 	res = false;
2561 ret:
2562 	free(buf, M_TEMP);
2563 	return (res);
2564 }
2565 
2566 struct brandnote_cb_arg {
2567 	Elf_Brandnote *brandnote;
2568 	int32_t *osrel;
2569 };
2570 
2571 static bool
2572 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2573 {
2574 	struct brandnote_cb_arg *arg;
2575 
2576 	arg = arg0;
2577 
2578 	/*
2579 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2580 	 * necessary.
2581 	 */
2582 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2583 	    arg->brandnote->trans_osrel != NULL ?
2584 	    arg->brandnote->trans_osrel(note, arg->osrel) : true;
2585 
2586 	return (true);
2587 }
2588 
2589 static Elf_Note fctl_note = {
2590 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2591 	.n_descsz = sizeof(uint32_t),
2592 	.n_type = NT_FREEBSD_FEATURE_CTL,
2593 };
2594 
2595 struct fctl_cb_arg {
2596 	bool *has_fctl0;
2597 	uint32_t *fctl0;
2598 };
2599 
2600 static bool
2601 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2602 {
2603 	struct fctl_cb_arg *arg;
2604 	const Elf32_Word *desc;
2605 	uintptr_t p;
2606 
2607 	arg = arg0;
2608 	p = (uintptr_t)(note + 1);
2609 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2610 	desc = (const Elf32_Word *)p;
2611 	*arg->has_fctl0 = true;
2612 	*arg->fctl0 = desc[0];
2613 	*res = true;
2614 	return (true);
2615 }
2616 
2617 /*
2618  * Try to find the appropriate ABI-note section for checknote, fetch
2619  * the osreldate and feature control flags for binary from the ELF
2620  * OSABI-note.  Only the first page of the image is searched, the same
2621  * as for headers.
2622  */
2623 static bool
2624 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2625     int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2626 {
2627 	const Elf_Phdr *phdr;
2628 	const Elf_Ehdr *hdr;
2629 	struct brandnote_cb_arg b_arg;
2630 	struct fctl_cb_arg f_arg;
2631 	int i, j;
2632 
2633 	hdr = (const Elf_Ehdr *)imgp->image_header;
2634 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2635 	b_arg.brandnote = brandnote;
2636 	b_arg.osrel = osrel;
2637 	f_arg.has_fctl0 = has_fctl0;
2638 	f_arg.fctl0 = fctl0;
2639 
2640 	for (i = 0; i < hdr->e_phnum; i++) {
2641 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2642 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2643 		    &b_arg)) {
2644 			for (j = 0; j < hdr->e_phnum; j++) {
2645 				if (phdr[j].p_type == PT_NOTE &&
2646 				    __elfN(parse_notes)(imgp, &fctl_note,
2647 				    FREEBSD_ABI_VENDOR, &phdr[j],
2648 				    note_fctl_cb, &f_arg))
2649 					break;
2650 			}
2651 			return (true);
2652 		}
2653 	}
2654 	return (false);
2655 
2656 }
2657 
2658 /*
2659  * Tell kern_execve.c about it, with a little help from the linker.
2660  */
2661 static struct execsw __elfN(execsw) = {
2662 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2663 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2664 };
2665 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2666 
2667 static vm_prot_t
2668 __elfN(trans_prot)(Elf_Word flags)
2669 {
2670 	vm_prot_t prot;
2671 
2672 	prot = 0;
2673 	if (flags & PF_X)
2674 		prot |= VM_PROT_EXECUTE;
2675 	if (flags & PF_W)
2676 		prot |= VM_PROT_WRITE;
2677 	if (flags & PF_R)
2678 		prot |= VM_PROT_READ;
2679 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2680 	if (i386_read_exec && (flags & PF_R))
2681 		prot |= VM_PROT_EXECUTE;
2682 #endif
2683 	return (prot);
2684 }
2685 
2686 static Elf_Word
2687 __elfN(untrans_prot)(vm_prot_t prot)
2688 {
2689 	Elf_Word flags;
2690 
2691 	flags = 0;
2692 	if (prot & VM_PROT_EXECUTE)
2693 		flags |= PF_X;
2694 	if (prot & VM_PROT_READ)
2695 		flags |= PF_R;
2696 	if (prot & VM_PROT_WRITE)
2697 		flags |= PF_W;
2698 	return (flags);
2699 }
2700 
2701 vm_size_t
2702 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2703 {
2704 	uintptr_t range, rbase, gap;
2705 	int pct;
2706 
2707 	pct = __elfN(aslr_stack_gap);
2708 	if (pct == 0)
2709 		return (0);
2710 	if (pct > 50)
2711 		pct = 50;
2712 	range = imgp->eff_stack_sz * pct / 100;
2713 	arc4rand(&rbase, sizeof(rbase), 0);
2714 	gap = rbase % range;
2715 	gap &= ~(sizeof(u_long) - 1);
2716 	*stack_base -= gap;
2717 	return (gap);
2718 }
2719