xref: /freebsd/sys/kern/imgact_elf.c (revision 7d4374f65f7b3df3d2567029c510f2e1576f0f69)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2017 Dell EMC
5  * Copyright (c) 2000-2001, 2003 David O'Brien
6  * Copyright (c) 1995-1996 Søren Schmidt
7  * Copyright (c) 1996 Peter Wemm
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_capsicum.h"
38 
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mman.h>
52 #include <sys/namei.h>
53 #include <sys/proc.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sbuf.h>
60 #include <sys/sf_buf.h>
61 #include <sys/smp.h>
62 #include <sys/systm.h>
63 #include <sys/signalvar.h>
64 #include <sys/stat.h>
65 #include <sys/sx.h>
66 #include <sys/syscall.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/vnode.h>
70 #include <sys/syslog.h>
71 #include <sys/eventhandler.h>
72 #include <sys/user.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_extern.h>
81 
82 #include <machine/elf.h>
83 #include <machine/md_var.h>
84 
85 #define ELF_NOTE_ROUNDSIZE	4
86 #define OLD_EI_BRAND	8
87 
88 static int __elfN(check_header)(const Elf_Ehdr *hdr);
89 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
90     const char *interp, int32_t *osrel, uint32_t *fctl0);
91 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
92     u_long *entry);
93 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
94     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
97     int32_t *osrel);
98 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100     Elf_Brandnote *checknote, int32_t *osrel, boolean_t *has_fctl0,
101     uint32_t *fctl0);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
106     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
107     "");
108 
109 #define	CORE_BUF_SIZE	(16 * 1024)
110 
111 int __elfN(fallback_brand) = -1;
112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
113     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
114     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 
116 static int elf_legacy_coredump = 0;
117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
118     &elf_legacy_coredump, 0,
119     "include all and only RW pages in core dumps");
120 
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
124     defined(__riscv)
125 	1;
126 #else
127 	0;
128 #endif
129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
130     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
131     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 
133 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136     "enable execution from readable segments");
137 #endif
138 
139 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
140 static int
141 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
142 {
143 	u_long val;
144 	int error;
145 
146 	val = __elfN(pie_base);
147 	error = sysctl_handle_long(oidp, &val, 0, req);
148 	if (error != 0 || req->newptr == NULL)
149 		return (error);
150 	if ((val & PAGE_MASK) != 0)
151 		return (EINVAL);
152 	__elfN(pie_base) = val;
153 	return (0);
154 }
155 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
156     CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
157     sysctl_pie_base, "LU",
158     "PIE load base without randomization");
159 
160 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
161     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
162     "");
163 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
164 
165 static int __elfN(aslr_enabled) = 0;
166 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
167     &__elfN(aslr_enabled), 0,
168     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
169     ": enable address map randomization");
170 
171 static int __elfN(pie_aslr_enabled) = 0;
172 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
173     &__elfN(pie_aslr_enabled), 0,
174     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
175     ": enable address map randomization for PIE binaries");
176 
177 static int __elfN(aslr_honor_sbrk) = 1;
178 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
179     &__elfN(aslr_honor_sbrk), 0,
180     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
181 
182 static int __elfN(aslr_stack_gap) = 3;
183 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
184     &__elfN(aslr_stack_gap), 0,
185     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
186     ": maximum percentage of main stack to waste on a random gap");
187 
188 static int __elfN(sigfastblock) = 1;
189 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
190     CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
191     "enable sigfastblock for new processes");
192 
193 static 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, has_fctl0;
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 			has_fctl0 = false;
336 			*fctl0 = 0;
337 			*osrel = 0;
338 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
339 			    &has_fctl0, fctl0);
340 			/* Give brand a chance to veto check_note's guess */
341 			if (ret && bi->header_supported) {
342 				ret = bi->header_supported(imgp, osrel,
343 				    has_fctl0 ? fctl0 : NULL);
344 			}
345 			/*
346 			 * If note checker claimed the binary, but the
347 			 * interpreter path in the image does not
348 			 * match default one for the brand, try to
349 			 * search for other brands with the same
350 			 * interpreter.  Either there is better brand
351 			 * with the right interpreter, or, failing
352 			 * this, we return first brand which accepted
353 			 * our note and, optionally, header.
354 			 */
355 			if (ret && bi_m == NULL && interp != NULL &&
356 			    (bi->interp_path == NULL ||
357 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
358 			    strncmp(interp, bi->interp_path, interp_name_len)
359 			    != 0))) {
360 				bi_m = bi;
361 				ret = 0;
362 			}
363 			if (ret)
364 				return (bi);
365 		}
366 	}
367 	if (bi_m != NULL)
368 		return (bi_m);
369 
370 	/* If the executable has a brand, search for it in the brand list. */
371 	for (i = 0; i < MAX_BRANDS; i++) {
372 		bi = elf_brand_list[i];
373 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
374 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
375 			continue;
376 		if (hdr->e_machine == bi->machine &&
377 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
378 		    (bi->compat_3_brand != NULL &&
379 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
380 		    bi->compat_3_brand) == 0))) {
381 			/* Looks good, but give brand a chance to veto */
382 			if (bi->header_supported == NULL ||
383 			    bi->header_supported(imgp, NULL, NULL)) {
384 				/*
385 				 * Again, prefer strictly matching
386 				 * interpreter path.
387 				 */
388 				if (interp_name_len == 0 &&
389 				    bi->interp_path == NULL)
390 					return (bi);
391 				if (bi->interp_path != NULL &&
392 				    strlen(bi->interp_path) + 1 ==
393 				    interp_name_len && strncmp(interp,
394 				    bi->interp_path, interp_name_len) == 0)
395 					return (bi);
396 				if (bi_m == NULL)
397 					bi_m = bi;
398 			}
399 		}
400 	}
401 	if (bi_m != NULL)
402 		return (bi_m);
403 
404 	/* No known brand, see if the header is recognized by any brand */
405 	for (i = 0; i < MAX_BRANDS; i++) {
406 		bi = elf_brand_list[i];
407 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
408 		    bi->header_supported == NULL)
409 			continue;
410 		if (hdr->e_machine == bi->machine) {
411 			ret = bi->header_supported(imgp, NULL, NULL);
412 			if (ret)
413 				return (bi);
414 		}
415 	}
416 
417 	/* Lacking a known brand, search for a recognized interpreter. */
418 	if (interp != NULL) {
419 		for (i = 0; i < MAX_BRANDS; i++) {
420 			bi = elf_brand_list[i];
421 			if (bi == NULL || (bi->flags &
422 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
423 			    != 0)
424 				continue;
425 			if (hdr->e_machine == bi->machine &&
426 			    bi->interp_path != NULL &&
427 			    /* ELF image p_filesz includes terminating zero */
428 			    strlen(bi->interp_path) + 1 == interp_name_len &&
429 			    strncmp(interp, bi->interp_path, interp_name_len)
430 			    == 0 && (bi->header_supported == NULL ||
431 			    bi->header_supported(imgp, NULL, NULL)))
432 				return (bi);
433 		}
434 	}
435 
436 	/* Lacking a recognized interpreter, try the default brand */
437 	for (i = 0; i < MAX_BRANDS; i++) {
438 		bi = elf_brand_list[i];
439 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
440 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
441 			continue;
442 		if (hdr->e_machine == bi->machine &&
443 		    __elfN(fallback_brand) == bi->brand &&
444 		    (bi->header_supported == NULL ||
445 		    bi->header_supported(imgp, NULL, NULL)))
446 			return (bi);
447 	}
448 	return (NULL);
449 }
450 
451 static int
452 __elfN(check_header)(const Elf_Ehdr *hdr)
453 {
454 	Elf_Brandinfo *bi;
455 	int i;
456 
457 	if (!IS_ELF(*hdr) ||
458 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
459 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
460 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
461 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
462 	    hdr->e_version != ELF_TARG_VER)
463 		return (ENOEXEC);
464 
465 	/*
466 	 * Make sure we have at least one brand for this machine.
467 	 */
468 
469 	for (i = 0; i < MAX_BRANDS; i++) {
470 		bi = elf_brand_list[i];
471 		if (bi != NULL && bi->machine == hdr->e_machine)
472 			break;
473 	}
474 	if (i == MAX_BRANDS)
475 		return (ENOEXEC);
476 
477 	return (0);
478 }
479 
480 static int
481 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
482     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
483 {
484 	struct sf_buf *sf;
485 	int error;
486 	vm_offset_t off;
487 
488 	/*
489 	 * Create the page if it doesn't exist yet. Ignore errors.
490 	 */
491 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
492 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
493 
494 	/*
495 	 * Find the page from the underlying object.
496 	 */
497 	if (object != NULL) {
498 		sf = vm_imgact_map_page(object, offset);
499 		if (sf == NULL)
500 			return (KERN_FAILURE);
501 		off = offset - trunc_page(offset);
502 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
503 		    end - start);
504 		vm_imgact_unmap_page(sf);
505 		if (error != 0)
506 			return (KERN_FAILURE);
507 	}
508 
509 	return (KERN_SUCCESS);
510 }
511 
512 static int
513 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
514     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
515     int cow)
516 {
517 	struct sf_buf *sf;
518 	vm_offset_t off;
519 	vm_size_t sz;
520 	int error, locked, rv;
521 
522 	if (start != trunc_page(start)) {
523 		rv = __elfN(map_partial)(map, object, offset, start,
524 		    round_page(start), prot);
525 		if (rv != KERN_SUCCESS)
526 			return (rv);
527 		offset += round_page(start) - start;
528 		start = round_page(start);
529 	}
530 	if (end != round_page(end)) {
531 		rv = __elfN(map_partial)(map, object, offset +
532 		    trunc_page(end) - start, trunc_page(end), end, prot);
533 		if (rv != KERN_SUCCESS)
534 			return (rv);
535 		end = trunc_page(end);
536 	}
537 	if (start >= end)
538 		return (KERN_SUCCESS);
539 	if ((offset & PAGE_MASK) != 0) {
540 		/*
541 		 * The mapping is not page aligned.  This means that we have
542 		 * to copy the data.
543 		 */
544 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
545 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
546 		if (rv != KERN_SUCCESS)
547 			return (rv);
548 		if (object == NULL)
549 			return (KERN_SUCCESS);
550 		for (; start < end; start += sz) {
551 			sf = vm_imgact_map_page(object, offset);
552 			if (sf == NULL)
553 				return (KERN_FAILURE);
554 			off = offset - trunc_page(offset);
555 			sz = end - start;
556 			if (sz > PAGE_SIZE - off)
557 				sz = PAGE_SIZE - off;
558 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
559 			    (caddr_t)start, sz);
560 			vm_imgact_unmap_page(sf);
561 			if (error != 0)
562 				return (KERN_FAILURE);
563 			offset += sz;
564 		}
565 	} else {
566 		vm_object_reference(object);
567 		rv = vm_map_fixed(map, object, offset, start, end - start,
568 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
569 		    (object != NULL ? MAP_VN_EXEC : 0));
570 		if (rv != KERN_SUCCESS) {
571 			locked = VOP_ISLOCKED(imgp->vp);
572 			VOP_UNLOCK(imgp->vp);
573 			vm_object_deallocate(object);
574 			vn_lock(imgp->vp, locked | LK_RETRY);
575 			return (rv);
576 		} else if (object != NULL) {
577 			MPASS(imgp->vp->v_object == object);
578 			VOP_SET_TEXT_CHECKED(imgp->vp);
579 		}
580 	}
581 	return (KERN_SUCCESS);
582 }
583 
584 static int
585 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
586     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
587 {
588 	struct sf_buf *sf;
589 	size_t map_len;
590 	vm_map_t map;
591 	vm_object_t object;
592 	vm_offset_t map_addr;
593 	int error, rv, cow;
594 	size_t copy_len;
595 	vm_ooffset_t file_addr;
596 
597 	/*
598 	 * It's necessary to fail if the filsz + offset taken from the
599 	 * header is greater than the actual file pager object's size.
600 	 * If we were to allow this, then the vm_map_find() below would
601 	 * walk right off the end of the file object and into the ether.
602 	 *
603 	 * While I'm here, might as well check for something else that
604 	 * is invalid: filsz cannot be greater than memsz.
605 	 */
606 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
607 	    filsz > memsz) {
608 		uprintf("elf_load_section: truncated ELF file\n");
609 		return (ENOEXEC);
610 	}
611 
612 	object = imgp->object;
613 	map = &imgp->proc->p_vmspace->vm_map;
614 	map_addr = trunc_page((vm_offset_t)vmaddr);
615 	file_addr = trunc_page(offset);
616 
617 	/*
618 	 * We have two choices.  We can either clear the data in the last page
619 	 * of an oversized mapping, or we can start the anon mapping a page
620 	 * early and copy the initialized data into that first page.  We
621 	 * choose the second.
622 	 */
623 	if (filsz == 0)
624 		map_len = 0;
625 	else if (memsz > filsz)
626 		map_len = trunc_page(offset + filsz) - file_addr;
627 	else
628 		map_len = round_page(offset + filsz) - file_addr;
629 
630 	if (map_len != 0) {
631 		/* cow flags: don't dump readonly sections in core */
632 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
633 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
634 
635 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
636 		    map_addr, map_addr + map_len, prot, cow);
637 		if (rv != KERN_SUCCESS)
638 			return (EINVAL);
639 
640 		/* we can stop now if we've covered it all */
641 		if (memsz == filsz)
642 			return (0);
643 	}
644 
645 	/*
646 	 * We have to get the remaining bit of the file into the first part
647 	 * of the oversized map segment.  This is normally because the .data
648 	 * segment in the file is extended to provide bss.  It's a neat idea
649 	 * to try and save a page, but it's a pain in the behind to implement.
650 	 */
651 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
652 	    filsz);
653 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
654 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
655 
656 	/* This had damn well better be true! */
657 	if (map_len != 0) {
658 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
659 		    map_addr + map_len, prot, 0);
660 		if (rv != KERN_SUCCESS)
661 			return (EINVAL);
662 	}
663 
664 	if (copy_len != 0) {
665 		sf = vm_imgact_map_page(object, offset + filsz);
666 		if (sf == NULL)
667 			return (EIO);
668 
669 		/* send the page fragment to user space */
670 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
671 		    copy_len);
672 		vm_imgact_unmap_page(sf);
673 		if (error != 0)
674 			return (error);
675 	}
676 
677 	/*
678 	 * Remove write access to the page if it was only granted by map_insert
679 	 * to allow copyout.
680 	 */
681 	if ((prot & VM_PROT_WRITE) == 0)
682 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
683 		    map_len), prot, FALSE);
684 
685 	return (0);
686 }
687 
688 static int
689 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
690     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
691 {
692 	vm_prot_t prot;
693 	u_long base_addr;
694 	bool first;
695 	int error, i;
696 
697 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
698 
699 	base_addr = 0;
700 	first = true;
701 
702 	for (i = 0; i < hdr->e_phnum; i++) {
703 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
704 			continue;
705 
706 		/* Loadable segment */
707 		prot = __elfN(trans_prot)(phdr[i].p_flags);
708 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
709 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
710 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
711 		if (error != 0)
712 			return (error);
713 
714 		/*
715 		 * Establish the base address if this is the first segment.
716 		 */
717 		if (first) {
718   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
719 			first = false;
720 		}
721 	}
722 
723 	if (base_addrp != NULL)
724 		*base_addrp = base_addr;
725 
726 	return (0);
727 }
728 
729 /*
730  * Load the file "file" into memory.  It may be either a shared object
731  * or an executable.
732  *
733  * The "addr" reference parameter is in/out.  On entry, it specifies
734  * the address where a shared object should be loaded.  If the file is
735  * an executable, this value is ignored.  On exit, "addr" specifies
736  * where the file was actually loaded.
737  *
738  * The "entry" reference parameter is out only.  On exit, it specifies
739  * the entry point for the loaded file.
740  */
741 static int
742 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
743 	u_long *entry)
744 {
745 	struct {
746 		struct nameidata nd;
747 		struct vattr attr;
748 		struct image_params image_params;
749 	} *tempdata;
750 	const Elf_Ehdr *hdr = NULL;
751 	const Elf_Phdr *phdr = NULL;
752 	struct nameidata *nd;
753 	struct vattr *attr;
754 	struct image_params *imgp;
755 	u_long rbase;
756 	u_long base_addr = 0;
757 	int error;
758 
759 #ifdef CAPABILITY_MODE
760 	/*
761 	 * XXXJA: This check can go away once we are sufficiently confident
762 	 * that the checks in namei() are correct.
763 	 */
764 	if (IN_CAPABILITY_MODE(curthread))
765 		return (ECAPMODE);
766 #endif
767 
768 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
769 	nd = &tempdata->nd;
770 	attr = &tempdata->attr;
771 	imgp = &tempdata->image_params;
772 
773 	/*
774 	 * Initialize part of the common data
775 	 */
776 	imgp->proc = p;
777 	imgp->attr = attr;
778 
779 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
780 	    UIO_SYSSPACE, file, curthread);
781 	if ((error = namei(nd)) != 0) {
782 		nd->ni_vp = NULL;
783 		goto fail;
784 	}
785 	NDFREE(nd, NDF_ONLY_PNBUF);
786 	imgp->vp = nd->ni_vp;
787 
788 	/*
789 	 * Check permissions, modes, uid, etc on the file, and "open" it.
790 	 */
791 	error = exec_check_permissions(imgp);
792 	if (error)
793 		goto fail;
794 
795 	error = exec_map_first_page(imgp);
796 	if (error)
797 		goto fail;
798 
799 	imgp->object = nd->ni_vp->v_object;
800 
801 	hdr = (const Elf_Ehdr *)imgp->image_header;
802 	if ((error = __elfN(check_header)(hdr)) != 0)
803 		goto fail;
804 	if (hdr->e_type == ET_DYN)
805 		rbase = *addr;
806 	else if (hdr->e_type == ET_EXEC)
807 		rbase = 0;
808 	else {
809 		error = ENOEXEC;
810 		goto fail;
811 	}
812 
813 	/* Only support headers that fit within first page for now      */
814 	if ((hdr->e_phoff > PAGE_SIZE) ||
815 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
816 		error = ENOEXEC;
817 		goto fail;
818 	}
819 
820 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
821 	if (!aligned(phdr, Elf_Addr)) {
822 		error = ENOEXEC;
823 		goto fail;
824 	}
825 
826 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
827 	if (error != 0)
828 		goto fail;
829 
830 	*addr = base_addr;
831 	*entry = (unsigned long)hdr->e_entry + rbase;
832 
833 fail:
834 	if (imgp->firstpage)
835 		exec_unmap_first_page(imgp);
836 
837 	if (nd->ni_vp) {
838 		if (imgp->textset)
839 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
840 		vput(nd->ni_vp);
841 	}
842 	free(tempdata, M_TEMP);
843 
844 	return (error);
845 }
846 
847 static u_long
848 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
849     u_int align)
850 {
851 	u_long rbase, res;
852 
853 	MPASS(vm_map_min(map) <= minv);
854 	MPASS(maxv <= vm_map_max(map));
855 	MPASS(minv < maxv);
856 	MPASS(minv + align < maxv);
857 	arc4rand(&rbase, sizeof(rbase), 0);
858 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
859 	res &= ~((u_long)align - 1);
860 	if (res >= maxv)
861 		res -= align;
862 	KASSERT(res >= minv,
863 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
864 	    res, minv, maxv, rbase));
865 	KASSERT(res < maxv,
866 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
867 	    res, maxv, minv, rbase));
868 	return (res);
869 }
870 
871 static int
872 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
873     const Elf_Phdr *phdr, u_long et_dyn_addr)
874 {
875 	struct vmspace *vmspace;
876 	const char *err_str;
877 	u_long text_size, data_size, total_size, text_addr, data_addr;
878 	u_long seg_size, seg_addr;
879 	int i;
880 
881 	err_str = NULL;
882 	text_size = data_size = total_size = text_addr = data_addr = 0;
883 
884 	for (i = 0; i < hdr->e_phnum; i++) {
885 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
886 			continue;
887 
888 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
889 		seg_size = round_page(phdr[i].p_memsz +
890 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
891 
892 		/*
893 		 * Make the largest executable segment the official
894 		 * text segment and all others data.
895 		 *
896 		 * Note that obreak() assumes that data_addr + data_size == end
897 		 * of data load area, and the ELF file format expects segments
898 		 * to be sorted by address.  If multiple data segments exist,
899 		 * the last one will be used.
900 		 */
901 
902 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
903 			text_size = seg_size;
904 			text_addr = seg_addr;
905 		} else {
906 			data_size = seg_size;
907 			data_addr = seg_addr;
908 		}
909 		total_size += seg_size;
910 	}
911 
912 	if (data_addr == 0 && data_size == 0) {
913 		data_addr = text_addr;
914 		data_size = text_size;
915 	}
916 
917 	/*
918 	 * Check limits.  It should be safe to check the
919 	 * limits after loading the segments since we do
920 	 * not actually fault in all the segments pages.
921 	 */
922 	PROC_LOCK(imgp->proc);
923 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
924 		err_str = "Data segment size exceeds process limit";
925 	else if (text_size > maxtsiz)
926 		err_str = "Text segment size exceeds system limit";
927 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
928 		err_str = "Total segment size exceeds process limit";
929 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
930 		err_str = "Data segment size exceeds resource limit";
931 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
932 		err_str = "Total segment size exceeds resource limit";
933 	PROC_UNLOCK(imgp->proc);
934 	if (err_str != NULL) {
935 		uprintf("%s\n", err_str);
936 		return (ENOMEM);
937 	}
938 
939 	vmspace = imgp->proc->p_vmspace;
940 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
941 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
942 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
943 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
944 
945 	return (0);
946 }
947 
948 static int
949 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
950     char **interpp, bool *free_interpp)
951 {
952 	struct thread *td;
953 	char *interp;
954 	int error, interp_name_len;
955 
956 	KASSERT(phdr->p_type == PT_INTERP,
957 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
958 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
959 
960 	td = curthread;
961 
962 	/* Path to interpreter */
963 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
964 		uprintf("Invalid PT_INTERP\n");
965 		return (ENOEXEC);
966 	}
967 
968 	interp_name_len = phdr->p_filesz;
969 	if (phdr->p_offset > PAGE_SIZE ||
970 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
971 		/*
972 		 * The vnode lock might be needed by the pagedaemon to
973 		 * clean pages owned by the vnode.  Do not allow sleep
974 		 * waiting for memory with the vnode locked, instead
975 		 * try non-sleepable allocation first, and if it
976 		 * fails, go to the slow path were we drop the lock
977 		 * and do M_WAITOK.  A text reference prevents
978 		 * modifications to the vnode content.
979 		 */
980 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
981 		if (interp == NULL) {
982 			VOP_UNLOCK(imgp->vp);
983 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
984 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
985 		}
986 
987 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
988 		    interp_name_len, phdr->p_offset,
989 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
990 		    NOCRED, NULL, td);
991 		if (error != 0) {
992 			free(interp, M_TEMP);
993 			uprintf("i/o error PT_INTERP %d\n", error);
994 			return (error);
995 		}
996 		interp[interp_name_len] = '\0';
997 
998 		*interpp = interp;
999 		*free_interpp = true;
1000 		return (0);
1001 	}
1002 
1003 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1004 	if (interp[interp_name_len - 1] != '\0') {
1005 		uprintf("Invalid PT_INTERP\n");
1006 		return (ENOEXEC);
1007 	}
1008 
1009 	*interpp = interp;
1010 	*free_interpp = false;
1011 	return (0);
1012 }
1013 
1014 static int
1015 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1016     const char *interp, u_long *addr, u_long *entry)
1017 {
1018 	char *path;
1019 	int error;
1020 
1021 	if (brand_info->emul_path != NULL &&
1022 	    brand_info->emul_path[0] != '\0') {
1023 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1024 		snprintf(path, MAXPATHLEN, "%s%s",
1025 		    brand_info->emul_path, interp);
1026 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1027 		free(path, M_TEMP);
1028 		if (error == 0)
1029 			return (0);
1030 	}
1031 
1032 	if (brand_info->interp_newpath != NULL &&
1033 	    (brand_info->interp_path == NULL ||
1034 	    strcmp(interp, brand_info->interp_path) == 0)) {
1035 		error = __elfN(load_file)(imgp->proc,
1036 		    brand_info->interp_newpath, addr, entry);
1037 		if (error == 0)
1038 			return (0);
1039 	}
1040 
1041 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1042 	if (error == 0)
1043 		return (0);
1044 
1045 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1046 	return (error);
1047 }
1048 
1049 /*
1050  * Impossible et_dyn_addr initial value indicating that the real base
1051  * must be calculated later with some randomization applied.
1052  */
1053 #define	ET_DYN_ADDR_RAND	1
1054 
1055 static int
1056 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1057 {
1058 	struct thread *td;
1059 	const Elf_Ehdr *hdr;
1060 	const Elf_Phdr *phdr;
1061 	Elf_Auxargs *elf_auxargs;
1062 	struct vmspace *vmspace;
1063 	vm_map_t map;
1064 	char *interp;
1065 	Elf_Brandinfo *brand_info;
1066 	struct sysentvec *sv;
1067 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1068 	u_long maxalign, mapsz, maxv, maxv1;
1069 	uint32_t fctl0;
1070 	int32_t osrel;
1071 	bool free_interp;
1072 	int error, i, n;
1073 
1074 	hdr = (const Elf_Ehdr *)imgp->image_header;
1075 
1076 	/*
1077 	 * Do we have a valid ELF header ?
1078 	 *
1079 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1080 	 * if particular brand doesn't support it.
1081 	 */
1082 	if (__elfN(check_header)(hdr) != 0 ||
1083 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1084 		return (-1);
1085 
1086 	/*
1087 	 * From here on down, we return an errno, not -1, as we've
1088 	 * detected an ELF file.
1089 	 */
1090 
1091 	if ((hdr->e_phoff > PAGE_SIZE) ||
1092 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1093 		/* Only support headers in first page for now */
1094 		uprintf("Program headers not in the first page\n");
1095 		return (ENOEXEC);
1096 	}
1097 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1098 	if (!aligned(phdr, Elf_Addr)) {
1099 		uprintf("Unaligned program headers\n");
1100 		return (ENOEXEC);
1101 	}
1102 
1103 	n = error = 0;
1104 	baddr = 0;
1105 	osrel = 0;
1106 	fctl0 = 0;
1107 	entry = proghdr = 0;
1108 	interp = NULL;
1109 	free_interp = false;
1110 	td = curthread;
1111 	maxalign = PAGE_SIZE;
1112 	mapsz = 0;
1113 
1114 	for (i = 0; i < hdr->e_phnum; i++) {
1115 		switch (phdr[i].p_type) {
1116 		case PT_LOAD:
1117 			if (n == 0)
1118 				baddr = phdr[i].p_vaddr;
1119 			if (phdr[i].p_align > maxalign)
1120 				maxalign = phdr[i].p_align;
1121 			mapsz += phdr[i].p_memsz;
1122 			n++;
1123 
1124 			/*
1125 			 * If this segment contains the program headers,
1126 			 * remember their virtual address for the AT_PHDR
1127 			 * aux entry. Static binaries don't usually include
1128 			 * a PT_PHDR entry.
1129 			 */
1130 			if (phdr[i].p_offset == 0 &&
1131 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1132 				<= phdr[i].p_filesz)
1133 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1134 			break;
1135 		case PT_INTERP:
1136 			/* Path to interpreter */
1137 			if (interp != NULL) {
1138 				uprintf("Multiple PT_INTERP headers\n");
1139 				error = ENOEXEC;
1140 				goto ret;
1141 			}
1142 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1143 			    &free_interp);
1144 			if (error != 0)
1145 				goto ret;
1146 			break;
1147 		case PT_GNU_STACK:
1148 			if (__elfN(nxstack))
1149 				imgp->stack_prot =
1150 				    __elfN(trans_prot)(phdr[i].p_flags);
1151 			imgp->stack_sz = phdr[i].p_memsz;
1152 			break;
1153 		case PT_PHDR: 	/* Program header table info */
1154 			proghdr = phdr[i].p_vaddr;
1155 			break;
1156 		}
1157 	}
1158 
1159 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1160 	if (brand_info == NULL) {
1161 		uprintf("ELF binary type \"%u\" not known.\n",
1162 		    hdr->e_ident[EI_OSABI]);
1163 		error = ENOEXEC;
1164 		goto ret;
1165 	}
1166 	sv = brand_info->sysvec;
1167 	et_dyn_addr = 0;
1168 	if (hdr->e_type == ET_DYN) {
1169 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1170 			uprintf("Cannot execute shared object\n");
1171 			error = ENOEXEC;
1172 			goto ret;
1173 		}
1174 		/*
1175 		 * Honour the base load address from the dso if it is
1176 		 * non-zero for some reason.
1177 		 */
1178 		if (baddr == 0) {
1179 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1180 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1181 				et_dyn_addr = __elfN(pie_base);
1182 			else if ((__elfN(pie_aslr_enabled) &&
1183 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1184 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1185 				et_dyn_addr = ET_DYN_ADDR_RAND;
1186 			else
1187 				et_dyn_addr = __elfN(pie_base);
1188 		}
1189 	}
1190 
1191 	/*
1192 	 * Avoid a possible deadlock if the current address space is destroyed
1193 	 * and that address space maps the locked vnode.  In the common case,
1194 	 * the locked vnode's v_usecount is decremented but remains greater
1195 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1196 	 * However, in cases where the vnode lock is external, such as nullfs,
1197 	 * v_usecount may become zero.
1198 	 *
1199 	 * The VV_TEXT flag prevents modifications to the executable while
1200 	 * the vnode is unlocked.
1201 	 */
1202 	VOP_UNLOCK(imgp->vp);
1203 
1204 	/*
1205 	 * Decide whether to enable randomization of user mappings.
1206 	 * First, reset user preferences for the setid binaries.
1207 	 * Then, account for the support of the randomization by the
1208 	 * ABI, by user preferences, and make special treatment for
1209 	 * PIE binaries.
1210 	 */
1211 	if (imgp->credential_setid) {
1212 		PROC_LOCK(imgp->proc);
1213 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1214 		PROC_UNLOCK(imgp->proc);
1215 	}
1216 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1217 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1218 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1219 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1220 		    ("et_dyn_addr == RAND and !ASLR"));
1221 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1222 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1223 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1224 		imgp->map_flags |= MAP_ASLR;
1225 		/*
1226 		 * If user does not care about sbrk, utilize the bss
1227 		 * grow region for mappings as well.  We can select
1228 		 * the base for the image anywere and still not suffer
1229 		 * from the fragmentation.
1230 		 */
1231 		if (!__elfN(aslr_honor_sbrk) ||
1232 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1233 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1234 	}
1235 
1236 	error = exec_new_vmspace(imgp, sv);
1237 	vmspace = imgp->proc->p_vmspace;
1238 	map = &vmspace->vm_map;
1239 
1240 	imgp->proc->p_sysent = sv;
1241 
1242 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1243 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1244 		KASSERT((map->flags & MAP_ASLR) != 0,
1245 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1246 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1247 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1248 		    /* reserve half of the address space to interpreter */
1249 		    maxv / 2, 1UL << flsl(maxalign));
1250 	}
1251 
1252 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1253 	if (error != 0)
1254 		goto ret;
1255 
1256 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1257 	if (error != 0)
1258 		goto ret;
1259 
1260 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1261 	if (error != 0)
1262 		goto ret;
1263 
1264 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1265 
1266 	/*
1267 	 * We load the dynamic linker where a userland call
1268 	 * to mmap(0, ...) would put it.  The rationale behind this
1269 	 * calculation is that it leaves room for the heap to grow to
1270 	 * its maximum allowed size.
1271 	 */
1272 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1273 	    RLIMIT_DATA));
1274 	if ((map->flags & MAP_ASLR) != 0) {
1275 		maxv1 = maxv / 2 + addr / 2;
1276 		MPASS(maxv1 >= addr);	/* No overflow */
1277 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1278 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1279 	} else {
1280 		map->anon_loc = addr;
1281 	}
1282 
1283 	imgp->entry_addr = entry;
1284 
1285 	if (interp != NULL) {
1286 		VOP_UNLOCK(imgp->vp);
1287 		if ((map->flags & MAP_ASLR) != 0) {
1288 			/* Assume that interpeter fits into 1/4 of AS */
1289 			maxv1 = maxv / 2 + addr / 2;
1290 			MPASS(maxv1 >= addr);	/* No overflow */
1291 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1292 			    maxv1, PAGE_SIZE);
1293 		}
1294 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1295 		    &imgp->entry_addr);
1296 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1297 		if (error != 0)
1298 			goto ret;
1299 	} else
1300 		addr = et_dyn_addr;
1301 
1302 	/*
1303 	 * Construct auxargs table (used by the copyout_auxargs routine)
1304 	 */
1305 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1306 	if (elf_auxargs == NULL) {
1307 		VOP_UNLOCK(imgp->vp);
1308 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1309 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1310 	}
1311 	elf_auxargs->execfd = -1;
1312 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1313 	elf_auxargs->phent = hdr->e_phentsize;
1314 	elf_auxargs->phnum = hdr->e_phnum;
1315 	elf_auxargs->pagesz = PAGE_SIZE;
1316 	elf_auxargs->base = addr;
1317 	elf_auxargs->flags = 0;
1318 	elf_auxargs->entry = entry;
1319 	elf_auxargs->hdr_eflags = hdr->e_flags;
1320 
1321 	imgp->auxargs = elf_auxargs;
1322 	imgp->interpreted = 0;
1323 	imgp->reloc_base = addr;
1324 	imgp->proc->p_osrel = osrel;
1325 	imgp->proc->p_fctl0 = fctl0;
1326 	imgp->proc->p_elf_machine = hdr->e_machine;
1327 	imgp->proc->p_elf_flags = hdr->e_flags;
1328 
1329 ret:
1330 	if (free_interp)
1331 		free(interp, M_TEMP);
1332 	return (error);
1333 }
1334 
1335 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1336 
1337 int
1338 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1339 {
1340 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1341 	Elf_Auxinfo *argarray, *pos;
1342 	int error;
1343 
1344 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1345 	    M_WAITOK | M_ZERO);
1346 
1347 	if (args->execfd != -1)
1348 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1349 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1350 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1351 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1352 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1353 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1354 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1355 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1356 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1357 	if (imgp->execpathp != 0)
1358 		AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1359 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1360 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1361 	if (imgp->canary != 0) {
1362 		AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1363 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1364 	}
1365 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1366 	if (imgp->pagesizes != 0) {
1367 		AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1368 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1369 	}
1370 	if (imgp->sysent->sv_timekeep_base != 0) {
1371 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1372 		    imgp->sysent->sv_timekeep_base);
1373 	}
1374 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1375 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1376 	    imgp->sysent->sv_stackprot);
1377 	if (imgp->sysent->sv_hwcap != NULL)
1378 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1379 	if (imgp->sysent->sv_hwcap2 != NULL)
1380 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1381 	AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1382 	    ELF_BSDF_SIGFASTBLK : 0);
1383 	AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1384 	AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1385 	AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1386 	AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1387 	AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1388 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1389 
1390 	free(imgp->auxargs, M_TEMP);
1391 	imgp->auxargs = NULL;
1392 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1393 
1394 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1395 	free(argarray, M_TEMP);
1396 	return (error);
1397 }
1398 
1399 int
1400 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1401 {
1402 	Elf_Addr *base;
1403 
1404 	base = (Elf_Addr *)*stack_base;
1405 	base--;
1406 	if (suword(base, imgp->args->argc) == -1)
1407 		return (EFAULT);
1408 	*stack_base = (uintptr_t)base;
1409 	return (0);
1410 }
1411 
1412 /*
1413  * Code for generating ELF core dumps.
1414  */
1415 
1416 typedef void (*segment_callback)(vm_map_entry_t, void *);
1417 
1418 /* Closure for cb_put_phdr(). */
1419 struct phdr_closure {
1420 	Elf_Phdr *phdr;		/* Program header to fill in */
1421 	Elf_Off offset;		/* Offset of segment in core file */
1422 };
1423 
1424 /* Closure for cb_size_segment(). */
1425 struct sseg_closure {
1426 	int count;		/* Count of writable segments. */
1427 	size_t size;		/* Total size of all writable segments. */
1428 };
1429 
1430 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1431 
1432 struct note_info {
1433 	int		type;		/* Note type. */
1434 	outfunc_t 	outfunc; 	/* Output function. */
1435 	void		*outarg;	/* Argument for the output function. */
1436 	size_t		outsize;	/* Output size. */
1437 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1438 };
1439 
1440 TAILQ_HEAD(note_info_list, note_info);
1441 
1442 /* Coredump output parameters. */
1443 struct coredump_params {
1444 	off_t		offset;
1445 	struct ucred	*active_cred;
1446 	struct ucred	*file_cred;
1447 	struct thread	*td;
1448 	struct vnode	*vp;
1449 	struct compressor *comp;
1450 };
1451 
1452 extern int compress_user_cores;
1453 extern int compress_user_cores_level;
1454 
1455 static void cb_put_phdr(vm_map_entry_t, void *);
1456 static void cb_size_segment(vm_map_entry_t, void *);
1457 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1458     enum uio_seg);
1459 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1460 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1461     struct note_info_list *, size_t);
1462 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1463     size_t *);
1464 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1465 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1466 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1467 static int sbuf_drain_core_output(void *, const char *, int);
1468 
1469 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1470 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1471 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1472 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1473 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1474 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1475 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1476 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1477 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1478 static void note_procstat_files(void *, struct sbuf *, size_t *);
1479 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1480 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1481 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1482 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1483 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1484 
1485 /*
1486  * Write out a core segment to the compression stream.
1487  */
1488 static int
1489 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1490 {
1491 	u_int chunk_len;
1492 	int error;
1493 
1494 	while (len > 0) {
1495 		chunk_len = MIN(len, CORE_BUF_SIZE);
1496 
1497 		/*
1498 		 * We can get EFAULT error here.
1499 		 * In that case zero out the current chunk of the segment.
1500 		 */
1501 		error = copyin(base, buf, chunk_len);
1502 		if (error != 0)
1503 			bzero(buf, chunk_len);
1504 		error = compressor_write(p->comp, buf, chunk_len);
1505 		if (error != 0)
1506 			break;
1507 		base += chunk_len;
1508 		len -= chunk_len;
1509 	}
1510 	return (error);
1511 }
1512 
1513 static int
1514 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1515 {
1516 
1517 	return (core_write((struct coredump_params *)arg, base, len, offset,
1518 	    UIO_SYSSPACE));
1519 }
1520 
1521 static int
1522 core_write(struct coredump_params *p, const void *base, size_t len,
1523     off_t offset, enum uio_seg seg)
1524 {
1525 
1526 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1527 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1528 	    p->active_cred, p->file_cred, NULL, p->td));
1529 }
1530 
1531 static int
1532 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1533     void *tmpbuf)
1534 {
1535 	int error;
1536 
1537 	if (p->comp != NULL)
1538 		return (compress_chunk(p, base, tmpbuf, len));
1539 
1540 	/*
1541 	 * EFAULT is a non-fatal error that we can get, for example,
1542 	 * if the segment is backed by a file but extends beyond its
1543 	 * end.
1544 	 */
1545 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1546 	if (error == EFAULT) {
1547 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1548 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1549 		    "for process %s\n", base, len, offset, curproc->p_comm);
1550 
1551 		/*
1552 		 * Write a "real" zero byte at the end of the target region
1553 		 * in the case this is the last segment.
1554 		 * The intermediate space will be implicitly zero-filled.
1555 		 */
1556 		error = core_write(p, zero_region, 1, offset + len - 1,
1557 		    UIO_SYSSPACE);
1558 	}
1559 	return (error);
1560 }
1561 
1562 /*
1563  * Drain into a core file.
1564  */
1565 static int
1566 sbuf_drain_core_output(void *arg, const char *data, int len)
1567 {
1568 	struct coredump_params *p;
1569 	int error, locked;
1570 
1571 	p = (struct coredump_params *)arg;
1572 
1573 	/*
1574 	 * Some kern_proc out routines that print to this sbuf may
1575 	 * call us with the process lock held. Draining with the
1576 	 * non-sleepable lock held is unsafe. The lock is needed for
1577 	 * those routines when dumping a live process. In our case we
1578 	 * can safely release the lock before draining and acquire
1579 	 * again after.
1580 	 */
1581 	locked = PROC_LOCKED(p->td->td_proc);
1582 	if (locked)
1583 		PROC_UNLOCK(p->td->td_proc);
1584 	if (p->comp != NULL)
1585 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1586 	else
1587 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1588 		    UIO_SYSSPACE);
1589 	if (locked)
1590 		PROC_LOCK(p->td->td_proc);
1591 	if (error != 0)
1592 		return (-error);
1593 	p->offset += len;
1594 	return (len);
1595 }
1596 
1597 int
1598 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1599 {
1600 	struct ucred *cred = td->td_ucred;
1601 	int error = 0;
1602 	struct sseg_closure seginfo;
1603 	struct note_info_list notelst;
1604 	struct coredump_params params;
1605 	struct note_info *ninfo;
1606 	void *hdr, *tmpbuf;
1607 	size_t hdrsize, notesz, coresize;
1608 
1609 	hdr = NULL;
1610 	tmpbuf = NULL;
1611 	TAILQ_INIT(&notelst);
1612 
1613 	/* Size the program segments. */
1614 	seginfo.count = 0;
1615 	seginfo.size = 0;
1616 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1617 
1618 	/*
1619 	 * Collect info about the core file header area.
1620 	 */
1621 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1622 	if (seginfo.count + 1 >= PN_XNUM)
1623 		hdrsize += sizeof(Elf_Shdr);
1624 	__elfN(prepare_notes)(td, &notelst, &notesz);
1625 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1626 
1627 	/* Set up core dump parameters. */
1628 	params.offset = 0;
1629 	params.active_cred = cred;
1630 	params.file_cred = NOCRED;
1631 	params.td = td;
1632 	params.vp = vp;
1633 	params.comp = NULL;
1634 
1635 #ifdef RACCT
1636 	if (racct_enable) {
1637 		PROC_LOCK(td->td_proc);
1638 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1639 		PROC_UNLOCK(td->td_proc);
1640 		if (error != 0) {
1641 			error = EFAULT;
1642 			goto done;
1643 		}
1644 	}
1645 #endif
1646 	if (coresize >= limit) {
1647 		error = EFAULT;
1648 		goto done;
1649 	}
1650 
1651 	/* Create a compression stream if necessary. */
1652 	if (compress_user_cores != 0) {
1653 		params.comp = compressor_init(core_compressed_write,
1654 		    compress_user_cores, CORE_BUF_SIZE,
1655 		    compress_user_cores_level, &params);
1656 		if (params.comp == NULL) {
1657 			error = EFAULT;
1658 			goto done;
1659 		}
1660 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1661         }
1662 
1663 	/*
1664 	 * Allocate memory for building the header, fill it up,
1665 	 * and write it out following the notes.
1666 	 */
1667 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1668 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1669 	    notesz);
1670 
1671 	/* Write the contents of all of the writable segments. */
1672 	if (error == 0) {
1673 		Elf_Phdr *php;
1674 		off_t offset;
1675 		int i;
1676 
1677 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1678 		offset = round_page(hdrsize + notesz);
1679 		for (i = 0; i < seginfo.count; i++) {
1680 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1681 			    php->p_filesz, offset, &params, tmpbuf);
1682 			if (error != 0)
1683 				break;
1684 			offset += php->p_filesz;
1685 			php++;
1686 		}
1687 		if (error == 0 && params.comp != NULL)
1688 			error = compressor_flush(params.comp);
1689 	}
1690 	if (error) {
1691 		log(LOG_WARNING,
1692 		    "Failed to write core file for process %s (error %d)\n",
1693 		    curproc->p_comm, error);
1694 	}
1695 
1696 done:
1697 	free(tmpbuf, M_TEMP);
1698 	if (params.comp != NULL)
1699 		compressor_fini(params.comp);
1700 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1701 		TAILQ_REMOVE(&notelst, ninfo, link);
1702 		free(ninfo, M_TEMP);
1703 	}
1704 	if (hdr != NULL)
1705 		free(hdr, M_TEMP);
1706 
1707 	return (error);
1708 }
1709 
1710 /*
1711  * A callback for each_dumpable_segment() to write out the segment's
1712  * program header entry.
1713  */
1714 static void
1715 cb_put_phdr(vm_map_entry_t entry, void *closure)
1716 {
1717 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1718 	Elf_Phdr *phdr = phc->phdr;
1719 
1720 	phc->offset = round_page(phc->offset);
1721 
1722 	phdr->p_type = PT_LOAD;
1723 	phdr->p_offset = phc->offset;
1724 	phdr->p_vaddr = entry->start;
1725 	phdr->p_paddr = 0;
1726 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1727 	phdr->p_align = PAGE_SIZE;
1728 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1729 
1730 	phc->offset += phdr->p_filesz;
1731 	phc->phdr++;
1732 }
1733 
1734 /*
1735  * A callback for each_dumpable_segment() to gather information about
1736  * the number of segments and their total size.
1737  */
1738 static void
1739 cb_size_segment(vm_map_entry_t entry, void *closure)
1740 {
1741 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1742 
1743 	ssc->count++;
1744 	ssc->size += entry->end - entry->start;
1745 }
1746 
1747 /*
1748  * For each writable segment in the process's memory map, call the given
1749  * function with a pointer to the map entry and some arbitrary
1750  * caller-supplied data.
1751  */
1752 static void
1753 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1754 {
1755 	struct proc *p = td->td_proc;
1756 	vm_map_t map = &p->p_vmspace->vm_map;
1757 	vm_map_entry_t entry;
1758 	vm_object_t backing_object, object;
1759 	boolean_t ignore_entry;
1760 
1761 	vm_map_lock_read(map);
1762 	VM_MAP_ENTRY_FOREACH(entry, map) {
1763 		/*
1764 		 * Don't dump inaccessible mappings, deal with legacy
1765 		 * coredump mode.
1766 		 *
1767 		 * Note that read-only segments related to the elf binary
1768 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1769 		 * need to arbitrarily ignore such segments.
1770 		 */
1771 		if (elf_legacy_coredump) {
1772 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1773 				continue;
1774 		} else {
1775 			if ((entry->protection & VM_PROT_ALL) == 0)
1776 				continue;
1777 		}
1778 
1779 		/*
1780 		 * Dont include memory segment in the coredump if
1781 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1782 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1783 		 * kernel map).
1784 		 */
1785 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1786 			continue;
1787 
1788 		if ((object = entry->object.vm_object) == NULL)
1789 			continue;
1790 
1791 		/* Ignore memory-mapped devices and such things. */
1792 		VM_OBJECT_RLOCK(object);
1793 		while ((backing_object = object->backing_object) != NULL) {
1794 			VM_OBJECT_RLOCK(backing_object);
1795 			VM_OBJECT_RUNLOCK(object);
1796 			object = backing_object;
1797 		}
1798 		ignore_entry = object->type != OBJT_DEFAULT &&
1799 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1800 		    object->type != OBJT_PHYS;
1801 		VM_OBJECT_RUNLOCK(object);
1802 		if (ignore_entry)
1803 			continue;
1804 
1805 		(*func)(entry, closure);
1806 	}
1807 	vm_map_unlock_read(map);
1808 }
1809 
1810 /*
1811  * Write the core file header to the file, including padding up to
1812  * the page boundary.
1813  */
1814 static int
1815 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1816     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1817 {
1818 	struct note_info *ninfo;
1819 	struct sbuf *sb;
1820 	int error;
1821 
1822 	/* Fill in the header. */
1823 	bzero(hdr, hdrsize);
1824 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1825 
1826 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1827 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1828 	sbuf_start_section(sb, NULL);
1829 	sbuf_bcat(sb, hdr, hdrsize);
1830 	TAILQ_FOREACH(ninfo, notelst, link)
1831 	    __elfN(putnote)(ninfo, sb);
1832 	/* Align up to a page boundary for the program segments. */
1833 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1834 	error = sbuf_finish(sb);
1835 	sbuf_delete(sb);
1836 
1837 	return (error);
1838 }
1839 
1840 static void
1841 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1842     size_t *sizep)
1843 {
1844 	struct proc *p;
1845 	struct thread *thr;
1846 	size_t size;
1847 
1848 	p = td->td_proc;
1849 	size = 0;
1850 
1851 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1852 
1853 	/*
1854 	 * To have the debugger select the right thread (LWP) as the initial
1855 	 * thread, we dump the state of the thread passed to us in td first.
1856 	 * This is the thread that causes the core dump and thus likely to
1857 	 * be the right thread one wants to have selected in the debugger.
1858 	 */
1859 	thr = td;
1860 	while (thr != NULL) {
1861 		size += register_note(list, NT_PRSTATUS,
1862 		    __elfN(note_prstatus), thr);
1863 		size += register_note(list, NT_FPREGSET,
1864 		    __elfN(note_fpregset), thr);
1865 		size += register_note(list, NT_THRMISC,
1866 		    __elfN(note_thrmisc), thr);
1867 		size += register_note(list, NT_PTLWPINFO,
1868 		    __elfN(note_ptlwpinfo), thr);
1869 		size += register_note(list, -1,
1870 		    __elfN(note_threadmd), thr);
1871 
1872 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1873 		    TAILQ_NEXT(thr, td_plist);
1874 		if (thr == td)
1875 			thr = TAILQ_NEXT(thr, td_plist);
1876 	}
1877 
1878 	size += register_note(list, NT_PROCSTAT_PROC,
1879 	    __elfN(note_procstat_proc), p);
1880 	size += register_note(list, NT_PROCSTAT_FILES,
1881 	    note_procstat_files, p);
1882 	size += register_note(list, NT_PROCSTAT_VMMAP,
1883 	    note_procstat_vmmap, p);
1884 	size += register_note(list, NT_PROCSTAT_GROUPS,
1885 	    note_procstat_groups, p);
1886 	size += register_note(list, NT_PROCSTAT_UMASK,
1887 	    note_procstat_umask, p);
1888 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1889 	    note_procstat_rlimit, p);
1890 	size += register_note(list, NT_PROCSTAT_OSREL,
1891 	    note_procstat_osrel, p);
1892 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1893 	    __elfN(note_procstat_psstrings), p);
1894 	size += register_note(list, NT_PROCSTAT_AUXV,
1895 	    __elfN(note_procstat_auxv), p);
1896 
1897 	*sizep = size;
1898 }
1899 
1900 static void
1901 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1902     size_t notesz)
1903 {
1904 	Elf_Ehdr *ehdr;
1905 	Elf_Phdr *phdr;
1906 	Elf_Shdr *shdr;
1907 	struct phdr_closure phc;
1908 
1909 	ehdr = (Elf_Ehdr *)hdr;
1910 
1911 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1912 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1913 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1914 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1915 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1916 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1917 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1918 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1919 	ehdr->e_ident[EI_ABIVERSION] = 0;
1920 	ehdr->e_ident[EI_PAD] = 0;
1921 	ehdr->e_type = ET_CORE;
1922 	ehdr->e_machine = td->td_proc->p_elf_machine;
1923 	ehdr->e_version = EV_CURRENT;
1924 	ehdr->e_entry = 0;
1925 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1926 	ehdr->e_flags = td->td_proc->p_elf_flags;
1927 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1928 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1929 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1930 	ehdr->e_shstrndx = SHN_UNDEF;
1931 	if (numsegs + 1 < PN_XNUM) {
1932 		ehdr->e_phnum = numsegs + 1;
1933 		ehdr->e_shnum = 0;
1934 	} else {
1935 		ehdr->e_phnum = PN_XNUM;
1936 		ehdr->e_shnum = 1;
1937 
1938 		ehdr->e_shoff = ehdr->e_phoff +
1939 		    (numsegs + 1) * ehdr->e_phentsize;
1940 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1941 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1942 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1943 
1944 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1945 		memset(shdr, 0, sizeof(*shdr));
1946 		/*
1947 		 * A special first section is used to hold large segment and
1948 		 * section counts.  This was proposed by Sun Microsystems in
1949 		 * Solaris and has been adopted by Linux; the standard ELF
1950 		 * tools are already familiar with the technique.
1951 		 *
1952 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1953 		 * (or 12-7 depending on the version of the document) for more
1954 		 * details.
1955 		 */
1956 		shdr->sh_type = SHT_NULL;
1957 		shdr->sh_size = ehdr->e_shnum;
1958 		shdr->sh_link = ehdr->e_shstrndx;
1959 		shdr->sh_info = numsegs + 1;
1960 	}
1961 
1962 	/*
1963 	 * Fill in the program header entries.
1964 	 */
1965 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1966 
1967 	/* The note segement. */
1968 	phdr->p_type = PT_NOTE;
1969 	phdr->p_offset = hdrsize;
1970 	phdr->p_vaddr = 0;
1971 	phdr->p_paddr = 0;
1972 	phdr->p_filesz = notesz;
1973 	phdr->p_memsz = 0;
1974 	phdr->p_flags = PF_R;
1975 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1976 	phdr++;
1977 
1978 	/* All the writable segments from the program. */
1979 	phc.phdr = phdr;
1980 	phc.offset = round_page(hdrsize + notesz);
1981 	each_dumpable_segment(td, cb_put_phdr, &phc);
1982 }
1983 
1984 static size_t
1985 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1986 {
1987 	struct note_info *ninfo;
1988 	size_t size, notesize;
1989 
1990 	size = 0;
1991 	out(arg, NULL, &size);
1992 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1993 	ninfo->type = type;
1994 	ninfo->outfunc = out;
1995 	ninfo->outarg = arg;
1996 	ninfo->outsize = size;
1997 	TAILQ_INSERT_TAIL(list, ninfo, link);
1998 
1999 	if (type == -1)
2000 		return (size);
2001 
2002 	notesize = sizeof(Elf_Note) +		/* note header */
2003 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2004 						/* note name */
2005 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2006 
2007 	return (notesize);
2008 }
2009 
2010 static size_t
2011 append_note_data(const void *src, void *dst, size_t len)
2012 {
2013 	size_t padded_len;
2014 
2015 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2016 	if (dst != NULL) {
2017 		bcopy(src, dst, len);
2018 		bzero((char *)dst + len, padded_len - len);
2019 	}
2020 	return (padded_len);
2021 }
2022 
2023 size_t
2024 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2025 {
2026 	Elf_Note *note;
2027 	char *buf;
2028 	size_t notesize;
2029 
2030 	buf = dst;
2031 	if (buf != NULL) {
2032 		note = (Elf_Note *)buf;
2033 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2034 		note->n_descsz = size;
2035 		note->n_type = type;
2036 		buf += sizeof(*note);
2037 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2038 		    sizeof(FREEBSD_ABI_VENDOR));
2039 		append_note_data(src, buf, size);
2040 		if (descp != NULL)
2041 			*descp = buf;
2042 	}
2043 
2044 	notesize = sizeof(Elf_Note) +		/* note header */
2045 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2046 						/* note name */
2047 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2048 
2049 	return (notesize);
2050 }
2051 
2052 static void
2053 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2054 {
2055 	Elf_Note note;
2056 	ssize_t old_len, sect_len;
2057 	size_t new_len, descsz, i;
2058 
2059 	if (ninfo->type == -1) {
2060 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2061 		return;
2062 	}
2063 
2064 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2065 	note.n_descsz = ninfo->outsize;
2066 	note.n_type = ninfo->type;
2067 
2068 	sbuf_bcat(sb, &note, sizeof(note));
2069 	sbuf_start_section(sb, &old_len);
2070 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2071 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2072 	if (note.n_descsz == 0)
2073 		return;
2074 	sbuf_start_section(sb, &old_len);
2075 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2076 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2077 	if (sect_len < 0)
2078 		return;
2079 
2080 	new_len = (size_t)sect_len;
2081 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2082 	if (new_len < descsz) {
2083 		/*
2084 		 * It is expected that individual note emitters will correctly
2085 		 * predict their expected output size and fill up to that size
2086 		 * themselves, padding in a format-specific way if needed.
2087 		 * However, in case they don't, just do it here with zeros.
2088 		 */
2089 		for (i = 0; i < descsz - new_len; i++)
2090 			sbuf_putc(sb, 0);
2091 	} else if (new_len > descsz) {
2092 		/*
2093 		 * We can't always truncate sb -- we may have drained some
2094 		 * of it already.
2095 		 */
2096 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2097 		    "read it (%zu > %zu).  Since it is longer than "
2098 		    "expected, this coredump's notes are corrupt.  THIS "
2099 		    "IS A BUG in the note_procstat routine for type %u.\n",
2100 		    __func__, (unsigned)note.n_type, new_len, descsz,
2101 		    (unsigned)note.n_type));
2102 	}
2103 }
2104 
2105 /*
2106  * Miscellaneous note out functions.
2107  */
2108 
2109 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2110 #include <compat/freebsd32/freebsd32.h>
2111 #include <compat/freebsd32/freebsd32_signal.h>
2112 
2113 typedef struct prstatus32 elf_prstatus_t;
2114 typedef struct prpsinfo32 elf_prpsinfo_t;
2115 typedef struct fpreg32 elf_prfpregset_t;
2116 typedef struct fpreg32 elf_fpregset_t;
2117 typedef struct reg32 elf_gregset_t;
2118 typedef struct thrmisc32 elf_thrmisc_t;
2119 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2120 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2121 typedef uint32_t elf_ps_strings_t;
2122 #else
2123 typedef prstatus_t elf_prstatus_t;
2124 typedef prpsinfo_t elf_prpsinfo_t;
2125 typedef prfpregset_t elf_prfpregset_t;
2126 typedef prfpregset_t elf_fpregset_t;
2127 typedef gregset_t elf_gregset_t;
2128 typedef thrmisc_t elf_thrmisc_t;
2129 #define ELF_KERN_PROC_MASK	0
2130 typedef struct kinfo_proc elf_kinfo_proc_t;
2131 typedef vm_offset_t elf_ps_strings_t;
2132 #endif
2133 
2134 static void
2135 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2136 {
2137 	struct sbuf sbarg;
2138 	size_t len;
2139 	char *cp, *end;
2140 	struct proc *p;
2141 	elf_prpsinfo_t *psinfo;
2142 	int error;
2143 
2144 	p = (struct proc *)arg;
2145 	if (sb != NULL) {
2146 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2147 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2148 		psinfo->pr_version = PRPSINFO_VERSION;
2149 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2150 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2151 		PROC_LOCK(p);
2152 		if (p->p_args != NULL) {
2153 			len = sizeof(psinfo->pr_psargs) - 1;
2154 			if (len > p->p_args->ar_length)
2155 				len = p->p_args->ar_length;
2156 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2157 			PROC_UNLOCK(p);
2158 			error = 0;
2159 		} else {
2160 			_PHOLD(p);
2161 			PROC_UNLOCK(p);
2162 			sbuf_new(&sbarg, psinfo->pr_psargs,
2163 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2164 			error = proc_getargv(curthread, p, &sbarg);
2165 			PRELE(p);
2166 			if (sbuf_finish(&sbarg) == 0)
2167 				len = sbuf_len(&sbarg) - 1;
2168 			else
2169 				len = sizeof(psinfo->pr_psargs) - 1;
2170 			sbuf_delete(&sbarg);
2171 		}
2172 		if (error || len == 0)
2173 			strlcpy(psinfo->pr_psargs, p->p_comm,
2174 			    sizeof(psinfo->pr_psargs));
2175 		else {
2176 			KASSERT(len < sizeof(psinfo->pr_psargs),
2177 			    ("len is too long: %zu vs %zu", len,
2178 			    sizeof(psinfo->pr_psargs)));
2179 			cp = psinfo->pr_psargs;
2180 			end = cp + len - 1;
2181 			for (;;) {
2182 				cp = memchr(cp, '\0', end - cp);
2183 				if (cp == NULL)
2184 					break;
2185 				*cp = ' ';
2186 			}
2187 		}
2188 		psinfo->pr_pid = p->p_pid;
2189 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2190 		free(psinfo, M_TEMP);
2191 	}
2192 	*sizep = sizeof(*psinfo);
2193 }
2194 
2195 static void
2196 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2197 {
2198 	struct thread *td;
2199 	elf_prstatus_t *status;
2200 
2201 	td = (struct thread *)arg;
2202 	if (sb != NULL) {
2203 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2204 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2205 		status->pr_version = PRSTATUS_VERSION;
2206 		status->pr_statussz = sizeof(elf_prstatus_t);
2207 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2208 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2209 		status->pr_osreldate = osreldate;
2210 		status->pr_cursig = td->td_proc->p_sig;
2211 		status->pr_pid = td->td_tid;
2212 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2213 		fill_regs32(td, &status->pr_reg);
2214 #else
2215 		fill_regs(td, &status->pr_reg);
2216 #endif
2217 		sbuf_bcat(sb, status, sizeof(*status));
2218 		free(status, M_TEMP);
2219 	}
2220 	*sizep = sizeof(*status);
2221 }
2222 
2223 static void
2224 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2225 {
2226 	struct thread *td;
2227 	elf_prfpregset_t *fpregset;
2228 
2229 	td = (struct thread *)arg;
2230 	if (sb != NULL) {
2231 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2232 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2233 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2234 		fill_fpregs32(td, fpregset);
2235 #else
2236 		fill_fpregs(td, fpregset);
2237 #endif
2238 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2239 		free(fpregset, M_TEMP);
2240 	}
2241 	*sizep = sizeof(*fpregset);
2242 }
2243 
2244 static void
2245 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2246 {
2247 	struct thread *td;
2248 	elf_thrmisc_t thrmisc;
2249 
2250 	td = (struct thread *)arg;
2251 	if (sb != NULL) {
2252 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2253 		bzero(&thrmisc, sizeof(thrmisc));
2254 		strcpy(thrmisc.pr_tname, td->td_name);
2255 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2256 	}
2257 	*sizep = sizeof(thrmisc);
2258 }
2259 
2260 static void
2261 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2262 {
2263 	struct thread *td;
2264 	size_t size;
2265 	int structsize;
2266 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2267 	struct ptrace_lwpinfo32 pl;
2268 #else
2269 	struct ptrace_lwpinfo pl;
2270 #endif
2271 
2272 	td = (struct thread *)arg;
2273 	size = sizeof(structsize) + sizeof(pl);
2274 	if (sb != NULL) {
2275 		KASSERT(*sizep == size, ("invalid size"));
2276 		structsize = sizeof(pl);
2277 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2278 		bzero(&pl, sizeof(pl));
2279 		pl.pl_lwpid = td->td_tid;
2280 		pl.pl_event = PL_EVENT_NONE;
2281 		pl.pl_sigmask = td->td_sigmask;
2282 		pl.pl_siglist = td->td_siglist;
2283 		if (td->td_si.si_signo != 0) {
2284 			pl.pl_event = PL_EVENT_SIGNAL;
2285 			pl.pl_flags |= PL_FLAG_SI;
2286 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2287 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2288 #else
2289 			pl.pl_siginfo = td->td_si;
2290 #endif
2291 		}
2292 		strcpy(pl.pl_tdname, td->td_name);
2293 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2294 		sbuf_bcat(sb, &pl, sizeof(pl));
2295 	}
2296 	*sizep = size;
2297 }
2298 
2299 /*
2300  * Allow for MD specific notes, as well as any MD
2301  * specific preparations for writing MI notes.
2302  */
2303 static void
2304 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2305 {
2306 	struct thread *td;
2307 	void *buf;
2308 	size_t size;
2309 
2310 	td = (struct thread *)arg;
2311 	size = *sizep;
2312 	if (size != 0 && sb != NULL)
2313 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2314 	else
2315 		buf = NULL;
2316 	size = 0;
2317 	__elfN(dump_thread)(td, buf, &size);
2318 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2319 	if (size != 0 && sb != NULL)
2320 		sbuf_bcat(sb, buf, size);
2321 	free(buf, M_TEMP);
2322 	*sizep = size;
2323 }
2324 
2325 #ifdef KINFO_PROC_SIZE
2326 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2327 #endif
2328 
2329 static void
2330 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2331 {
2332 	struct proc *p;
2333 	size_t size;
2334 	int structsize;
2335 
2336 	p = (struct proc *)arg;
2337 	size = sizeof(structsize) + p->p_numthreads *
2338 	    sizeof(elf_kinfo_proc_t);
2339 
2340 	if (sb != NULL) {
2341 		KASSERT(*sizep == size, ("invalid size"));
2342 		structsize = sizeof(elf_kinfo_proc_t);
2343 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2344 		PROC_LOCK(p);
2345 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2346 	}
2347 	*sizep = size;
2348 }
2349 
2350 #ifdef KINFO_FILE_SIZE
2351 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2352 #endif
2353 
2354 static void
2355 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2356 {
2357 	struct proc *p;
2358 	size_t size, sect_sz, i;
2359 	ssize_t start_len, sect_len;
2360 	int structsize, filedesc_flags;
2361 
2362 	if (coredump_pack_fileinfo)
2363 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2364 	else
2365 		filedesc_flags = 0;
2366 
2367 	p = (struct proc *)arg;
2368 	structsize = sizeof(struct kinfo_file);
2369 	if (sb == NULL) {
2370 		size = 0;
2371 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2372 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2373 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2374 		PROC_LOCK(p);
2375 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2376 		sbuf_finish(sb);
2377 		sbuf_delete(sb);
2378 		*sizep = size;
2379 	} else {
2380 		sbuf_start_section(sb, &start_len);
2381 
2382 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2383 		PROC_LOCK(p);
2384 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2385 		    filedesc_flags);
2386 
2387 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2388 		if (sect_len < 0)
2389 			return;
2390 		sect_sz = sect_len;
2391 
2392 		KASSERT(sect_sz <= *sizep,
2393 		    ("kern_proc_filedesc_out did not respect maxlen; "
2394 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2395 		     sect_sz - sizeof(structsize)));
2396 
2397 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2398 			sbuf_putc(sb, 0);
2399 	}
2400 }
2401 
2402 #ifdef KINFO_VMENTRY_SIZE
2403 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2404 #endif
2405 
2406 static void
2407 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2408 {
2409 	struct proc *p;
2410 	size_t size;
2411 	int structsize, vmmap_flags;
2412 
2413 	if (coredump_pack_vmmapinfo)
2414 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2415 	else
2416 		vmmap_flags = 0;
2417 
2418 	p = (struct proc *)arg;
2419 	structsize = sizeof(struct kinfo_vmentry);
2420 	if (sb == NULL) {
2421 		size = 0;
2422 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2423 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2424 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2425 		PROC_LOCK(p);
2426 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2427 		sbuf_finish(sb);
2428 		sbuf_delete(sb);
2429 		*sizep = size;
2430 	} else {
2431 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2432 		PROC_LOCK(p);
2433 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2434 		    vmmap_flags);
2435 	}
2436 }
2437 
2438 static void
2439 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2440 {
2441 	struct proc *p;
2442 	size_t size;
2443 	int structsize;
2444 
2445 	p = (struct proc *)arg;
2446 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2447 	if (sb != NULL) {
2448 		KASSERT(*sizep == size, ("invalid size"));
2449 		structsize = sizeof(gid_t);
2450 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2451 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2452 		    sizeof(gid_t));
2453 	}
2454 	*sizep = size;
2455 }
2456 
2457 static void
2458 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2459 {
2460 	struct proc *p;
2461 	size_t size;
2462 	int structsize;
2463 
2464 	p = (struct proc *)arg;
2465 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2466 	if (sb != NULL) {
2467 		KASSERT(*sizep == size, ("invalid size"));
2468 		structsize = sizeof(p->p_fd->fd_cmask);
2469 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2470 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2471 	}
2472 	*sizep = size;
2473 }
2474 
2475 static void
2476 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2477 {
2478 	struct proc *p;
2479 	struct rlimit rlim[RLIM_NLIMITS];
2480 	size_t size;
2481 	int structsize, i;
2482 
2483 	p = (struct proc *)arg;
2484 	size = sizeof(structsize) + sizeof(rlim);
2485 	if (sb != NULL) {
2486 		KASSERT(*sizep == size, ("invalid size"));
2487 		structsize = sizeof(rlim);
2488 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2489 		PROC_LOCK(p);
2490 		for (i = 0; i < RLIM_NLIMITS; i++)
2491 			lim_rlimit_proc(p, i, &rlim[i]);
2492 		PROC_UNLOCK(p);
2493 		sbuf_bcat(sb, rlim, sizeof(rlim));
2494 	}
2495 	*sizep = size;
2496 }
2497 
2498 static void
2499 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2500 {
2501 	struct proc *p;
2502 	size_t size;
2503 	int structsize;
2504 
2505 	p = (struct proc *)arg;
2506 	size = sizeof(structsize) + sizeof(p->p_osrel);
2507 	if (sb != NULL) {
2508 		KASSERT(*sizep == size, ("invalid size"));
2509 		structsize = sizeof(p->p_osrel);
2510 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2511 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2512 	}
2513 	*sizep = size;
2514 }
2515 
2516 static void
2517 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2518 {
2519 	struct proc *p;
2520 	elf_ps_strings_t ps_strings;
2521 	size_t size;
2522 	int structsize;
2523 
2524 	p = (struct proc *)arg;
2525 	size = sizeof(structsize) + sizeof(ps_strings);
2526 	if (sb != NULL) {
2527 		KASSERT(*sizep == size, ("invalid size"));
2528 		structsize = sizeof(ps_strings);
2529 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2530 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2531 #else
2532 		ps_strings = p->p_sysent->sv_psstrings;
2533 #endif
2534 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2535 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2536 	}
2537 	*sizep = size;
2538 }
2539 
2540 static void
2541 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2542 {
2543 	struct proc *p;
2544 	size_t size;
2545 	int structsize;
2546 
2547 	p = (struct proc *)arg;
2548 	if (sb == NULL) {
2549 		size = 0;
2550 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2551 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2552 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2553 		PHOLD(p);
2554 		proc_getauxv(curthread, p, sb);
2555 		PRELE(p);
2556 		sbuf_finish(sb);
2557 		sbuf_delete(sb);
2558 		*sizep = size;
2559 	} else {
2560 		structsize = sizeof(Elf_Auxinfo);
2561 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2562 		PHOLD(p);
2563 		proc_getauxv(curthread, p, sb);
2564 		PRELE(p);
2565 	}
2566 }
2567 
2568 static boolean_t
2569 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2570     const char *note_vendor, const Elf_Phdr *pnote,
2571     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2572 {
2573 	const Elf_Note *note, *note0, *note_end;
2574 	const char *note_name;
2575 	char *buf;
2576 	int i, error;
2577 	boolean_t res;
2578 
2579 	/* We need some limit, might as well use PAGE_SIZE. */
2580 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2581 		return (FALSE);
2582 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2583 	if (pnote->p_offset > PAGE_SIZE ||
2584 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2585 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2586 		if (buf == NULL) {
2587 			VOP_UNLOCK(imgp->vp);
2588 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2589 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2590 		}
2591 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2592 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2593 		    curthread->td_ucred, NOCRED, NULL, curthread);
2594 		if (error != 0) {
2595 			uprintf("i/o error PT_NOTE\n");
2596 			goto retf;
2597 		}
2598 		note = note0 = (const Elf_Note *)buf;
2599 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2600 	} else {
2601 		note = note0 = (const Elf_Note *)(imgp->image_header +
2602 		    pnote->p_offset);
2603 		note_end = (const Elf_Note *)(imgp->image_header +
2604 		    pnote->p_offset + pnote->p_filesz);
2605 		buf = NULL;
2606 	}
2607 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2608 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2609 		    (const char *)note < sizeof(Elf_Note)) {
2610 			goto retf;
2611 		}
2612 		if (note->n_namesz != checknote->n_namesz ||
2613 		    note->n_descsz != checknote->n_descsz ||
2614 		    note->n_type != checknote->n_type)
2615 			goto nextnote;
2616 		note_name = (const char *)(note + 1);
2617 		if (note_name + checknote->n_namesz >=
2618 		    (const char *)note_end || strncmp(note_vendor,
2619 		    note_name, checknote->n_namesz) != 0)
2620 			goto nextnote;
2621 
2622 		if (cb(note, cb_arg, &res))
2623 			goto ret;
2624 nextnote:
2625 		note = (const Elf_Note *)((const char *)(note + 1) +
2626 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2627 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2628 	}
2629 retf:
2630 	res = FALSE;
2631 ret:
2632 	free(buf, M_TEMP);
2633 	return (res);
2634 }
2635 
2636 struct brandnote_cb_arg {
2637 	Elf_Brandnote *brandnote;
2638 	int32_t *osrel;
2639 };
2640 
2641 static boolean_t
2642 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2643 {
2644 	struct brandnote_cb_arg *arg;
2645 
2646 	arg = arg0;
2647 
2648 	/*
2649 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2650 	 * necessary.
2651 	 */
2652 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2653 	    arg->brandnote->trans_osrel != NULL ?
2654 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2655 
2656 	return (TRUE);
2657 }
2658 
2659 static Elf_Note fctl_note = {
2660 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2661 	.n_descsz = sizeof(uint32_t),
2662 	.n_type = NT_FREEBSD_FEATURE_CTL,
2663 };
2664 
2665 struct fctl_cb_arg {
2666 	boolean_t *has_fctl0;
2667 	uint32_t *fctl0;
2668 };
2669 
2670 static boolean_t
2671 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2672 {
2673 	struct fctl_cb_arg *arg;
2674 	const Elf32_Word *desc;
2675 	uintptr_t p;
2676 
2677 	arg = arg0;
2678 	p = (uintptr_t)(note + 1);
2679 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2680 	desc = (const Elf32_Word *)p;
2681 	*arg->has_fctl0 = TRUE;
2682 	*arg->fctl0 = desc[0];
2683 	return (TRUE);
2684 }
2685 
2686 /*
2687  * Try to find the appropriate ABI-note section for checknote, fetch
2688  * the osreldate and feature control flags for binary from the ELF
2689  * OSABI-note.  Only the first page of the image is searched, the same
2690  * as for headers.
2691  */
2692 static boolean_t
2693 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2694     int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2695 {
2696 	const Elf_Phdr *phdr;
2697 	const Elf_Ehdr *hdr;
2698 	struct brandnote_cb_arg b_arg;
2699 	struct fctl_cb_arg f_arg;
2700 	int i, j;
2701 
2702 	hdr = (const Elf_Ehdr *)imgp->image_header;
2703 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2704 	b_arg.brandnote = brandnote;
2705 	b_arg.osrel = osrel;
2706 	f_arg.has_fctl0 = has_fctl0;
2707 	f_arg.fctl0 = fctl0;
2708 
2709 	for (i = 0; i < hdr->e_phnum; i++) {
2710 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2711 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2712 		    &b_arg)) {
2713 			for (j = 0; j < hdr->e_phnum; j++) {
2714 				if (phdr[j].p_type == PT_NOTE &&
2715 				    __elfN(parse_notes)(imgp, &fctl_note,
2716 				    FREEBSD_ABI_VENDOR, &phdr[j],
2717 				    note_fctl_cb, &f_arg))
2718 					break;
2719 			}
2720 			return (TRUE);
2721 		}
2722 	}
2723 	return (FALSE);
2724 
2725 }
2726 
2727 /*
2728  * Tell kern_execve.c about it, with a little help from the linker.
2729  */
2730 static struct execsw __elfN(execsw) = {
2731 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2732 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2733 };
2734 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2735 
2736 static vm_prot_t
2737 __elfN(trans_prot)(Elf_Word flags)
2738 {
2739 	vm_prot_t prot;
2740 
2741 	prot = 0;
2742 	if (flags & PF_X)
2743 		prot |= VM_PROT_EXECUTE;
2744 	if (flags & PF_W)
2745 		prot |= VM_PROT_WRITE;
2746 	if (flags & PF_R)
2747 		prot |= VM_PROT_READ;
2748 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2749 	if (i386_read_exec && (flags & PF_R))
2750 		prot |= VM_PROT_EXECUTE;
2751 #endif
2752 	return (prot);
2753 }
2754 
2755 static Elf_Word
2756 __elfN(untrans_prot)(vm_prot_t prot)
2757 {
2758 	Elf_Word flags;
2759 
2760 	flags = 0;
2761 	if (prot & VM_PROT_EXECUTE)
2762 		flags |= PF_X;
2763 	if (prot & VM_PROT_READ)
2764 		flags |= PF_R;
2765 	if (prot & VM_PROT_WRITE)
2766 		flags |= PF_W;
2767 	return (flags);
2768 }
2769 
2770 void
2771 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2772 {
2773 	uintptr_t range, rbase, gap;
2774 	int pct;
2775 
2776 	if ((imgp->map_flags & MAP_ASLR) == 0)
2777 		return;
2778 	pct = __elfN(aslr_stack_gap);
2779 	if (pct == 0)
2780 		return;
2781 	if (pct > 50)
2782 		pct = 50;
2783 	range = imgp->eff_stack_sz * pct / 100;
2784 	arc4rand(&rbase, sizeof(rbase), 0);
2785 	gap = rbase % range;
2786 	gap &= ~(sizeof(u_long) - 1);
2787 	*stack_base -= gap;
2788 }
2789