xref: /freebsd/sys/kern/imgact_elf.c (revision cd8537910406e68d4719136a5b0cf6d23bb1b23b)
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 bool
452 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
453 {
454 	return (hdr->e_phoff <= PAGE_SIZE &&
455 	    (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
456 }
457 
458 static int
459 __elfN(check_header)(const Elf_Ehdr *hdr)
460 {
461 	Elf_Brandinfo *bi;
462 	int i;
463 
464 	if (!IS_ELF(*hdr) ||
465 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
466 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
467 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
468 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
469 	    hdr->e_version != ELF_TARG_VER)
470 		return (ENOEXEC);
471 
472 	/*
473 	 * Make sure we have at least one brand for this machine.
474 	 */
475 
476 	for (i = 0; i < MAX_BRANDS; i++) {
477 		bi = elf_brand_list[i];
478 		if (bi != NULL && bi->machine == hdr->e_machine)
479 			break;
480 	}
481 	if (i == MAX_BRANDS)
482 		return (ENOEXEC);
483 
484 	return (0);
485 }
486 
487 static int
488 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
489     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
490 {
491 	struct sf_buf *sf;
492 	int error;
493 	vm_offset_t off;
494 
495 	/*
496 	 * Create the page if it doesn't exist yet. Ignore errors.
497 	 */
498 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
499 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
500 
501 	/*
502 	 * Find the page from the underlying object.
503 	 */
504 	if (object != NULL) {
505 		sf = vm_imgact_map_page(object, offset);
506 		if (sf == NULL)
507 			return (KERN_FAILURE);
508 		off = offset - trunc_page(offset);
509 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
510 		    end - start);
511 		vm_imgact_unmap_page(sf);
512 		if (error != 0)
513 			return (KERN_FAILURE);
514 	}
515 
516 	return (KERN_SUCCESS);
517 }
518 
519 static int
520 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
521     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
522     int cow)
523 {
524 	struct sf_buf *sf;
525 	vm_offset_t off;
526 	vm_size_t sz;
527 	int error, locked, rv;
528 
529 	if (start != trunc_page(start)) {
530 		rv = __elfN(map_partial)(map, object, offset, start,
531 		    round_page(start), prot);
532 		if (rv != KERN_SUCCESS)
533 			return (rv);
534 		offset += round_page(start) - start;
535 		start = round_page(start);
536 	}
537 	if (end != round_page(end)) {
538 		rv = __elfN(map_partial)(map, object, offset +
539 		    trunc_page(end) - start, trunc_page(end), end, prot);
540 		if (rv != KERN_SUCCESS)
541 			return (rv);
542 		end = trunc_page(end);
543 	}
544 	if (start >= end)
545 		return (KERN_SUCCESS);
546 	if ((offset & PAGE_MASK) != 0) {
547 		/*
548 		 * The mapping is not page aligned.  This means that we have
549 		 * to copy the data.
550 		 */
551 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
552 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
553 		if (rv != KERN_SUCCESS)
554 			return (rv);
555 		if (object == NULL)
556 			return (KERN_SUCCESS);
557 		for (; start < end; start += sz) {
558 			sf = vm_imgact_map_page(object, offset);
559 			if (sf == NULL)
560 				return (KERN_FAILURE);
561 			off = offset - trunc_page(offset);
562 			sz = end - start;
563 			if (sz > PAGE_SIZE - off)
564 				sz = PAGE_SIZE - off;
565 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
566 			    (caddr_t)start, sz);
567 			vm_imgact_unmap_page(sf);
568 			if (error != 0)
569 				return (KERN_FAILURE);
570 			offset += sz;
571 		}
572 	} else {
573 		vm_object_reference(object);
574 		rv = vm_map_fixed(map, object, offset, start, end - start,
575 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
576 		    (object != NULL ? MAP_VN_EXEC : 0));
577 		if (rv != KERN_SUCCESS) {
578 			locked = VOP_ISLOCKED(imgp->vp);
579 			VOP_UNLOCK(imgp->vp);
580 			vm_object_deallocate(object);
581 			vn_lock(imgp->vp, locked | LK_RETRY);
582 			return (rv);
583 		} else if (object != NULL) {
584 			MPASS(imgp->vp->v_object == object);
585 			VOP_SET_TEXT_CHECKED(imgp->vp);
586 		}
587 	}
588 	return (KERN_SUCCESS);
589 }
590 
591 static int
592 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
593     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
594 {
595 	struct sf_buf *sf;
596 	size_t map_len;
597 	vm_map_t map;
598 	vm_object_t object;
599 	vm_offset_t map_addr;
600 	int error, rv, cow;
601 	size_t copy_len;
602 	vm_ooffset_t file_addr;
603 
604 	/*
605 	 * It's necessary to fail if the filsz + offset taken from the
606 	 * header is greater than the actual file pager object's size.
607 	 * If we were to allow this, then the vm_map_find() below would
608 	 * walk right off the end of the file object and into the ether.
609 	 *
610 	 * While I'm here, might as well check for something else that
611 	 * is invalid: filsz cannot be greater than memsz.
612 	 */
613 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
614 	    filsz > memsz) {
615 		uprintf("elf_load_section: truncated ELF file\n");
616 		return (ENOEXEC);
617 	}
618 
619 	object = imgp->object;
620 	map = &imgp->proc->p_vmspace->vm_map;
621 	map_addr = trunc_page((vm_offset_t)vmaddr);
622 	file_addr = trunc_page(offset);
623 
624 	/*
625 	 * We have two choices.  We can either clear the data in the last page
626 	 * of an oversized mapping, or we can start the anon mapping a page
627 	 * early and copy the initialized data into that first page.  We
628 	 * choose the second.
629 	 */
630 	if (filsz == 0)
631 		map_len = 0;
632 	else if (memsz > filsz)
633 		map_len = trunc_page(offset + filsz) - file_addr;
634 	else
635 		map_len = round_page(offset + filsz) - file_addr;
636 
637 	if (map_len != 0) {
638 		/* cow flags: don't dump readonly sections in core */
639 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
640 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
641 
642 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
643 		    map_addr, map_addr + map_len, prot, cow);
644 		if (rv != KERN_SUCCESS)
645 			return (EINVAL);
646 
647 		/* we can stop now if we've covered it all */
648 		if (memsz == filsz)
649 			return (0);
650 	}
651 
652 	/*
653 	 * We have to get the remaining bit of the file into the first part
654 	 * of the oversized map segment.  This is normally because the .data
655 	 * segment in the file is extended to provide bss.  It's a neat idea
656 	 * to try and save a page, but it's a pain in the behind to implement.
657 	 */
658 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
659 	    filsz);
660 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
661 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
662 
663 	/* This had damn well better be true! */
664 	if (map_len != 0) {
665 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
666 		    map_addr + map_len, prot, 0);
667 		if (rv != KERN_SUCCESS)
668 			return (EINVAL);
669 	}
670 
671 	if (copy_len != 0) {
672 		sf = vm_imgact_map_page(object, offset + filsz);
673 		if (sf == NULL)
674 			return (EIO);
675 
676 		/* send the page fragment to user space */
677 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
678 		    copy_len);
679 		vm_imgact_unmap_page(sf);
680 		if (error != 0)
681 			return (error);
682 	}
683 
684 	/*
685 	 * Remove write access to the page if it was only granted by map_insert
686 	 * to allow copyout.
687 	 */
688 	if ((prot & VM_PROT_WRITE) == 0)
689 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
690 		    map_len), prot, FALSE);
691 
692 	return (0);
693 }
694 
695 static int
696 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
697     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
698 {
699 	vm_prot_t prot;
700 	u_long base_addr;
701 	bool first;
702 	int error, i;
703 
704 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
705 
706 	base_addr = 0;
707 	first = true;
708 
709 	for (i = 0; i < hdr->e_phnum; i++) {
710 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
711 			continue;
712 
713 		/* Loadable segment */
714 		prot = __elfN(trans_prot)(phdr[i].p_flags);
715 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
716 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
717 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
718 		if (error != 0)
719 			return (error);
720 
721 		/*
722 		 * Establish the base address if this is the first segment.
723 		 */
724 		if (first) {
725   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
726 			first = false;
727 		}
728 	}
729 
730 	if (base_addrp != NULL)
731 		*base_addrp = base_addr;
732 
733 	return (0);
734 }
735 
736 /*
737  * Load the file "file" into memory.  It may be either a shared object
738  * or an executable.
739  *
740  * The "addr" reference parameter is in/out.  On entry, it specifies
741  * the address where a shared object should be loaded.  If the file is
742  * an executable, this value is ignored.  On exit, "addr" specifies
743  * where the file was actually loaded.
744  *
745  * The "entry" reference parameter is out only.  On exit, it specifies
746  * the entry point for the loaded file.
747  */
748 static int
749 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
750 	u_long *entry)
751 {
752 	struct {
753 		struct nameidata nd;
754 		struct vattr attr;
755 		struct image_params image_params;
756 	} *tempdata;
757 	const Elf_Ehdr *hdr = NULL;
758 	const Elf_Phdr *phdr = NULL;
759 	struct nameidata *nd;
760 	struct vattr *attr;
761 	struct image_params *imgp;
762 	u_long rbase;
763 	u_long base_addr = 0;
764 	int error;
765 
766 #ifdef CAPABILITY_MODE
767 	/*
768 	 * XXXJA: This check can go away once we are sufficiently confident
769 	 * that the checks in namei() are correct.
770 	 */
771 	if (IN_CAPABILITY_MODE(curthread))
772 		return (ECAPMODE);
773 #endif
774 
775 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
776 	nd = &tempdata->nd;
777 	attr = &tempdata->attr;
778 	imgp = &tempdata->image_params;
779 
780 	/*
781 	 * Initialize part of the common data
782 	 */
783 	imgp->proc = p;
784 	imgp->attr = attr;
785 
786 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
787 	    UIO_SYSSPACE, file, curthread);
788 	if ((error = namei(nd)) != 0) {
789 		nd->ni_vp = NULL;
790 		goto fail;
791 	}
792 	NDFREE(nd, NDF_ONLY_PNBUF);
793 	imgp->vp = nd->ni_vp;
794 
795 	/*
796 	 * Check permissions, modes, uid, etc on the file, and "open" it.
797 	 */
798 	error = exec_check_permissions(imgp);
799 	if (error)
800 		goto fail;
801 
802 	error = exec_map_first_page(imgp);
803 	if (error)
804 		goto fail;
805 
806 	imgp->object = nd->ni_vp->v_object;
807 
808 	hdr = (const Elf_Ehdr *)imgp->image_header;
809 	if ((error = __elfN(check_header)(hdr)) != 0)
810 		goto fail;
811 	if (hdr->e_type == ET_DYN)
812 		rbase = *addr;
813 	else if (hdr->e_type == ET_EXEC)
814 		rbase = 0;
815 	else {
816 		error = ENOEXEC;
817 		goto fail;
818 	}
819 
820 	/* Only support headers that fit within first page for now      */
821 	if (!__elfN(phdr_in_zero_page)(hdr)) {
822 		error = ENOEXEC;
823 		goto fail;
824 	}
825 
826 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
827 	if (!aligned(phdr, Elf_Addr)) {
828 		error = ENOEXEC;
829 		goto fail;
830 	}
831 
832 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
833 	if (error != 0)
834 		goto fail;
835 
836 	*addr = base_addr;
837 	*entry = (unsigned long)hdr->e_entry + rbase;
838 
839 fail:
840 	if (imgp->firstpage)
841 		exec_unmap_first_page(imgp);
842 
843 	if (nd->ni_vp) {
844 		if (imgp->textset)
845 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
846 		vput(nd->ni_vp);
847 	}
848 	free(tempdata, M_TEMP);
849 
850 	return (error);
851 }
852 
853 static u_long
854 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
855     u_int align)
856 {
857 	u_long rbase, res;
858 
859 	MPASS(vm_map_min(map) <= minv);
860 	MPASS(maxv <= vm_map_max(map));
861 	MPASS(minv < maxv);
862 	MPASS(minv + align < maxv);
863 	arc4rand(&rbase, sizeof(rbase), 0);
864 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
865 	res &= ~((u_long)align - 1);
866 	if (res >= maxv)
867 		res -= align;
868 	KASSERT(res >= minv,
869 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
870 	    res, minv, maxv, rbase));
871 	KASSERT(res < maxv,
872 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
873 	    res, maxv, minv, rbase));
874 	return (res);
875 }
876 
877 static int
878 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
879     const Elf_Phdr *phdr, u_long et_dyn_addr)
880 {
881 	struct vmspace *vmspace;
882 	const char *err_str;
883 	u_long text_size, data_size, total_size, text_addr, data_addr;
884 	u_long seg_size, seg_addr;
885 	int i;
886 
887 	err_str = NULL;
888 	text_size = data_size = total_size = text_addr = data_addr = 0;
889 
890 	for (i = 0; i < hdr->e_phnum; i++) {
891 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
892 			continue;
893 
894 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
895 		seg_size = round_page(phdr[i].p_memsz +
896 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
897 
898 		/*
899 		 * Make the largest executable segment the official
900 		 * text segment and all others data.
901 		 *
902 		 * Note that obreak() assumes that data_addr + data_size == end
903 		 * of data load area, and the ELF file format expects segments
904 		 * to be sorted by address.  If multiple data segments exist,
905 		 * the last one will be used.
906 		 */
907 
908 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
909 			text_size = seg_size;
910 			text_addr = seg_addr;
911 		} else {
912 			data_size = seg_size;
913 			data_addr = seg_addr;
914 		}
915 		total_size += seg_size;
916 	}
917 
918 	if (data_addr == 0 && data_size == 0) {
919 		data_addr = text_addr;
920 		data_size = text_size;
921 	}
922 
923 	/*
924 	 * Check limits.  It should be safe to check the
925 	 * limits after loading the segments since we do
926 	 * not actually fault in all the segments pages.
927 	 */
928 	PROC_LOCK(imgp->proc);
929 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
930 		err_str = "Data segment size exceeds process limit";
931 	else if (text_size > maxtsiz)
932 		err_str = "Text segment size exceeds system limit";
933 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
934 		err_str = "Total segment size exceeds process limit";
935 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
936 		err_str = "Data segment size exceeds resource limit";
937 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
938 		err_str = "Total segment size exceeds resource limit";
939 	PROC_UNLOCK(imgp->proc);
940 	if (err_str != NULL) {
941 		uprintf("%s\n", err_str);
942 		return (ENOMEM);
943 	}
944 
945 	vmspace = imgp->proc->p_vmspace;
946 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
947 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
948 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
949 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
950 
951 	return (0);
952 }
953 
954 static int
955 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
956     char **interpp, bool *free_interpp)
957 {
958 	struct thread *td;
959 	char *interp;
960 	int error, interp_name_len;
961 
962 	KASSERT(phdr->p_type == PT_INTERP,
963 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
964 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
965 
966 	td = curthread;
967 
968 	/* Path to interpreter */
969 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
970 		uprintf("Invalid PT_INTERP\n");
971 		return (ENOEXEC);
972 	}
973 
974 	interp_name_len = phdr->p_filesz;
975 	if (phdr->p_offset > PAGE_SIZE ||
976 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
977 		/*
978 		 * The vnode lock might be needed by the pagedaemon to
979 		 * clean pages owned by the vnode.  Do not allow sleep
980 		 * waiting for memory with the vnode locked, instead
981 		 * try non-sleepable allocation first, and if it
982 		 * fails, go to the slow path were we drop the lock
983 		 * and do M_WAITOK.  A text reference prevents
984 		 * modifications to the vnode content.
985 		 */
986 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
987 		if (interp == NULL) {
988 			VOP_UNLOCK(imgp->vp);
989 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
990 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
991 		}
992 
993 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
994 		    interp_name_len, phdr->p_offset,
995 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
996 		    NOCRED, NULL, td);
997 		if (error != 0) {
998 			free(interp, M_TEMP);
999 			uprintf("i/o error PT_INTERP %d\n", error);
1000 			return (error);
1001 		}
1002 		interp[interp_name_len] = '\0';
1003 
1004 		*interpp = interp;
1005 		*free_interpp = true;
1006 		return (0);
1007 	}
1008 
1009 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1010 	if (interp[interp_name_len - 1] != '\0') {
1011 		uprintf("Invalid PT_INTERP\n");
1012 		return (ENOEXEC);
1013 	}
1014 
1015 	*interpp = interp;
1016 	*free_interpp = false;
1017 	return (0);
1018 }
1019 
1020 static int
1021 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1022     const char *interp, u_long *addr, u_long *entry)
1023 {
1024 	char *path;
1025 	int error;
1026 
1027 	if (brand_info->emul_path != NULL &&
1028 	    brand_info->emul_path[0] != '\0') {
1029 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1030 		snprintf(path, MAXPATHLEN, "%s%s",
1031 		    brand_info->emul_path, interp);
1032 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1033 		free(path, M_TEMP);
1034 		if (error == 0)
1035 			return (0);
1036 	}
1037 
1038 	if (brand_info->interp_newpath != NULL &&
1039 	    (brand_info->interp_path == NULL ||
1040 	    strcmp(interp, brand_info->interp_path) == 0)) {
1041 		error = __elfN(load_file)(imgp->proc,
1042 		    brand_info->interp_newpath, addr, entry);
1043 		if (error == 0)
1044 			return (0);
1045 	}
1046 
1047 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1048 	if (error == 0)
1049 		return (0);
1050 
1051 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1052 	return (error);
1053 }
1054 
1055 /*
1056  * Impossible et_dyn_addr initial value indicating that the real base
1057  * must be calculated later with some randomization applied.
1058  */
1059 #define	ET_DYN_ADDR_RAND	1
1060 
1061 static int
1062 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1063 {
1064 	struct thread *td;
1065 	const Elf_Ehdr *hdr;
1066 	const Elf_Phdr *phdr;
1067 	Elf_Auxargs *elf_auxargs;
1068 	struct vmspace *vmspace;
1069 	vm_map_t map;
1070 	char *interp;
1071 	Elf_Brandinfo *brand_info;
1072 	struct sysentvec *sv;
1073 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1074 	u_long maxalign, mapsz, maxv, maxv1;
1075 	uint32_t fctl0;
1076 	int32_t osrel;
1077 	bool free_interp;
1078 	int error, i, n;
1079 
1080 	hdr = (const Elf_Ehdr *)imgp->image_header;
1081 
1082 	/*
1083 	 * Do we have a valid ELF header ?
1084 	 *
1085 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1086 	 * if particular brand doesn't support it.
1087 	 */
1088 	if (__elfN(check_header)(hdr) != 0 ||
1089 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1090 		return (-1);
1091 
1092 	/*
1093 	 * From here on down, we return an errno, not -1, as we've
1094 	 * detected an ELF file.
1095 	 */
1096 
1097 	if (!__elfN(phdr_in_zero_page)(hdr)) {
1098 		uprintf("Program headers not in the first page\n");
1099 		return (ENOEXEC);
1100 	}
1101 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1102 	if (!aligned(phdr, Elf_Addr)) {
1103 		uprintf("Unaligned program headers\n");
1104 		return (ENOEXEC);
1105 	}
1106 
1107 	n = error = 0;
1108 	baddr = 0;
1109 	osrel = 0;
1110 	fctl0 = 0;
1111 	entry = proghdr = 0;
1112 	interp = NULL;
1113 	free_interp = false;
1114 	td = curthread;
1115 	maxalign = PAGE_SIZE;
1116 	mapsz = 0;
1117 
1118 	for (i = 0; i < hdr->e_phnum; i++) {
1119 		switch (phdr[i].p_type) {
1120 		case PT_LOAD:
1121 			if (n == 0)
1122 				baddr = phdr[i].p_vaddr;
1123 			if (phdr[i].p_align > maxalign)
1124 				maxalign = phdr[i].p_align;
1125 			mapsz += phdr[i].p_memsz;
1126 			n++;
1127 
1128 			/*
1129 			 * If this segment contains the program headers,
1130 			 * remember their virtual address for the AT_PHDR
1131 			 * aux entry. Static binaries don't usually include
1132 			 * a PT_PHDR entry.
1133 			 */
1134 			if (phdr[i].p_offset == 0 &&
1135 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1136 				<= phdr[i].p_filesz)
1137 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1138 			break;
1139 		case PT_INTERP:
1140 			/* Path to interpreter */
1141 			if (interp != NULL) {
1142 				uprintf("Multiple PT_INTERP headers\n");
1143 				error = ENOEXEC;
1144 				goto ret;
1145 			}
1146 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1147 			    &free_interp);
1148 			if (error != 0)
1149 				goto ret;
1150 			break;
1151 		case PT_GNU_STACK:
1152 			if (__elfN(nxstack))
1153 				imgp->stack_prot =
1154 				    __elfN(trans_prot)(phdr[i].p_flags);
1155 			imgp->stack_sz = phdr[i].p_memsz;
1156 			break;
1157 		case PT_PHDR: 	/* Program header table info */
1158 			proghdr = phdr[i].p_vaddr;
1159 			break;
1160 		}
1161 	}
1162 
1163 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1164 	if (brand_info == NULL) {
1165 		uprintf("ELF binary type \"%u\" not known.\n",
1166 		    hdr->e_ident[EI_OSABI]);
1167 		error = ENOEXEC;
1168 		goto ret;
1169 	}
1170 	sv = brand_info->sysvec;
1171 	et_dyn_addr = 0;
1172 	if (hdr->e_type == ET_DYN) {
1173 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1174 			uprintf("Cannot execute shared object\n");
1175 			error = ENOEXEC;
1176 			goto ret;
1177 		}
1178 		/*
1179 		 * Honour the base load address from the dso if it is
1180 		 * non-zero for some reason.
1181 		 */
1182 		if (baddr == 0) {
1183 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1184 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1185 				et_dyn_addr = __elfN(pie_base);
1186 			else if ((__elfN(pie_aslr_enabled) &&
1187 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1188 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1189 				et_dyn_addr = ET_DYN_ADDR_RAND;
1190 			else
1191 				et_dyn_addr = __elfN(pie_base);
1192 		}
1193 	}
1194 
1195 	/*
1196 	 * Avoid a possible deadlock if the current address space is destroyed
1197 	 * and that address space maps the locked vnode.  In the common case,
1198 	 * the locked vnode's v_usecount is decremented but remains greater
1199 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1200 	 * However, in cases where the vnode lock is external, such as nullfs,
1201 	 * v_usecount may become zero.
1202 	 *
1203 	 * The VV_TEXT flag prevents modifications to the executable while
1204 	 * the vnode is unlocked.
1205 	 */
1206 	VOP_UNLOCK(imgp->vp);
1207 
1208 	/*
1209 	 * Decide whether to enable randomization of user mappings.
1210 	 * First, reset user preferences for the setid binaries.
1211 	 * Then, account for the support of the randomization by the
1212 	 * ABI, by user preferences, and make special treatment for
1213 	 * PIE binaries.
1214 	 */
1215 	if (imgp->credential_setid) {
1216 		PROC_LOCK(imgp->proc);
1217 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1218 		PROC_UNLOCK(imgp->proc);
1219 	}
1220 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1221 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1222 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1223 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1224 		    ("et_dyn_addr == RAND and !ASLR"));
1225 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1226 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1227 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1228 		imgp->map_flags |= MAP_ASLR;
1229 		/*
1230 		 * If user does not care about sbrk, utilize the bss
1231 		 * grow region for mappings as well.  We can select
1232 		 * the base for the image anywere and still not suffer
1233 		 * from the fragmentation.
1234 		 */
1235 		if (!__elfN(aslr_honor_sbrk) ||
1236 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1237 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1238 	}
1239 
1240 	error = exec_new_vmspace(imgp, sv);
1241 	vmspace = imgp->proc->p_vmspace;
1242 	map = &vmspace->vm_map;
1243 
1244 	imgp->proc->p_sysent = sv;
1245 
1246 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1247 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1248 		KASSERT((map->flags & MAP_ASLR) != 0,
1249 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1250 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1251 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1252 		    /* reserve half of the address space to interpreter */
1253 		    maxv / 2, 1UL << flsl(maxalign));
1254 	}
1255 
1256 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1257 	if (error != 0)
1258 		goto ret;
1259 
1260 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1261 	if (error != 0)
1262 		goto ret;
1263 
1264 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1265 	if (error != 0)
1266 		goto ret;
1267 
1268 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1269 
1270 	/*
1271 	 * We load the dynamic linker where a userland call
1272 	 * to mmap(0, ...) would put it.  The rationale behind this
1273 	 * calculation is that it leaves room for the heap to grow to
1274 	 * its maximum allowed size.
1275 	 */
1276 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1277 	    RLIMIT_DATA));
1278 	if ((map->flags & MAP_ASLR) != 0) {
1279 		maxv1 = maxv / 2 + addr / 2;
1280 		MPASS(maxv1 >= addr);	/* No overflow */
1281 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1282 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1283 	} else {
1284 		map->anon_loc = addr;
1285 	}
1286 
1287 	imgp->entry_addr = entry;
1288 
1289 	if (interp != NULL) {
1290 		VOP_UNLOCK(imgp->vp);
1291 		if ((map->flags & MAP_ASLR) != 0) {
1292 			/* Assume that interpeter fits into 1/4 of AS */
1293 			maxv1 = maxv / 2 + addr / 2;
1294 			MPASS(maxv1 >= addr);	/* No overflow */
1295 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1296 			    maxv1, PAGE_SIZE);
1297 		}
1298 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1299 		    &imgp->entry_addr);
1300 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1301 		if (error != 0)
1302 			goto ret;
1303 	} else
1304 		addr = et_dyn_addr;
1305 
1306 	/*
1307 	 * Construct auxargs table (used by the copyout_auxargs routine)
1308 	 */
1309 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1310 	if (elf_auxargs == NULL) {
1311 		VOP_UNLOCK(imgp->vp);
1312 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1313 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1314 	}
1315 	elf_auxargs->execfd = -1;
1316 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1317 	elf_auxargs->phent = hdr->e_phentsize;
1318 	elf_auxargs->phnum = hdr->e_phnum;
1319 	elf_auxargs->pagesz = PAGE_SIZE;
1320 	elf_auxargs->base = addr;
1321 	elf_auxargs->flags = 0;
1322 	elf_auxargs->entry = entry;
1323 	elf_auxargs->hdr_eflags = hdr->e_flags;
1324 
1325 	imgp->auxargs = elf_auxargs;
1326 	imgp->interpreted = 0;
1327 	imgp->reloc_base = addr;
1328 	imgp->proc->p_osrel = osrel;
1329 	imgp->proc->p_fctl0 = fctl0;
1330 	imgp->proc->p_elf_machine = hdr->e_machine;
1331 	imgp->proc->p_elf_flags = hdr->e_flags;
1332 
1333 ret:
1334 	if (free_interp)
1335 		free(interp, M_TEMP);
1336 	return (error);
1337 }
1338 
1339 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1340 
1341 int
1342 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1343 {
1344 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1345 	Elf_Auxinfo *argarray, *pos;
1346 	int error;
1347 
1348 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1349 	    M_WAITOK | M_ZERO);
1350 
1351 	if (args->execfd != -1)
1352 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1353 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1354 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1355 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1356 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1357 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1358 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1359 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1360 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1361 	if (imgp->execpathp != 0)
1362 		AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1363 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1364 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1365 	if (imgp->canary != 0) {
1366 		AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1367 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1368 	}
1369 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1370 	if (imgp->pagesizes != 0) {
1371 		AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1372 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1373 	}
1374 	if (imgp->sysent->sv_timekeep_base != 0) {
1375 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1376 		    imgp->sysent->sv_timekeep_base);
1377 	}
1378 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1379 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1380 	    imgp->sysent->sv_stackprot);
1381 	if (imgp->sysent->sv_hwcap != NULL)
1382 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1383 	if (imgp->sysent->sv_hwcap2 != NULL)
1384 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1385 	AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1386 	    ELF_BSDF_SIGFASTBLK : 0);
1387 	AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1388 	AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1389 	AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1390 	AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1391 	AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1392 	if (imgp->sysent->sv_fxrng_gen_base != 0)
1393 		AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1394 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1395 
1396 	free(imgp->auxargs, M_TEMP);
1397 	imgp->auxargs = NULL;
1398 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1399 
1400 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1401 	free(argarray, M_TEMP);
1402 	return (error);
1403 }
1404 
1405 int
1406 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1407 {
1408 	Elf_Addr *base;
1409 
1410 	base = (Elf_Addr *)*stack_base;
1411 	base--;
1412 	if (suword(base, imgp->args->argc) == -1)
1413 		return (EFAULT);
1414 	*stack_base = (uintptr_t)base;
1415 	return (0);
1416 }
1417 
1418 /*
1419  * Code for generating ELF core dumps.
1420  */
1421 
1422 typedef void (*segment_callback)(vm_map_entry_t, void *);
1423 
1424 /* Closure for cb_put_phdr(). */
1425 struct phdr_closure {
1426 	Elf_Phdr *phdr;		/* Program header to fill in */
1427 	Elf_Off offset;		/* Offset of segment in core file */
1428 };
1429 
1430 /* Closure for cb_size_segment(). */
1431 struct sseg_closure {
1432 	int count;		/* Count of writable segments. */
1433 	size_t size;		/* Total size of all writable segments. */
1434 };
1435 
1436 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1437 
1438 struct note_info {
1439 	int		type;		/* Note type. */
1440 	outfunc_t 	outfunc; 	/* Output function. */
1441 	void		*outarg;	/* Argument for the output function. */
1442 	size_t		outsize;	/* Output size. */
1443 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1444 };
1445 
1446 TAILQ_HEAD(note_info_list, note_info);
1447 
1448 /* Coredump output parameters. */
1449 struct coredump_params {
1450 	off_t		offset;
1451 	struct ucred	*active_cred;
1452 	struct ucred	*file_cred;
1453 	struct thread	*td;
1454 	struct vnode	*vp;
1455 	struct compressor *comp;
1456 };
1457 
1458 extern int compress_user_cores;
1459 extern int compress_user_cores_level;
1460 
1461 static void cb_put_phdr(vm_map_entry_t, void *);
1462 static void cb_size_segment(vm_map_entry_t, void *);
1463 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1464     enum uio_seg, size_t *);
1465 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1466 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1467     struct note_info_list *, size_t);
1468 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1469     size_t *);
1470 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1471 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1472 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1473 static int sbuf_drain_core_output(void *, const char *, int);
1474 
1475 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1476 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1477 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1478 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1479 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1480 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1481 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1482 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1483 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1484 static void note_procstat_files(void *, struct sbuf *, size_t *);
1485 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1486 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1487 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1488 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1489 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1490 
1491 /*
1492  * Write out a core segment to the compression stream.
1493  */
1494 static int
1495 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1496 {
1497 	u_int chunk_len;
1498 	int error;
1499 
1500 	while (len > 0) {
1501 		chunk_len = MIN(len, CORE_BUF_SIZE);
1502 
1503 		/*
1504 		 * We can get EFAULT error here.
1505 		 * In that case zero out the current chunk of the segment.
1506 		 */
1507 		error = copyin(base, buf, chunk_len);
1508 		if (error != 0)
1509 			bzero(buf, chunk_len);
1510 		error = compressor_write(p->comp, buf, chunk_len);
1511 		if (error != 0)
1512 			break;
1513 		base += chunk_len;
1514 		len -= chunk_len;
1515 	}
1516 	return (error);
1517 }
1518 
1519 static int
1520 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1521 {
1522 
1523 	return (core_write((struct coredump_params *)arg, base, len, offset,
1524 	    UIO_SYSSPACE, NULL));
1525 }
1526 
1527 static int
1528 core_write(struct coredump_params *p, const void *base, size_t len,
1529     off_t offset, enum uio_seg seg, size_t *resid)
1530 {
1531 
1532 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1533 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1534 	    p->active_cred, p->file_cred, resid, p->td));
1535 }
1536 
1537 static int
1538 core_output(char *base, size_t len, off_t offset, struct coredump_params *p,
1539     void *tmpbuf)
1540 {
1541 	vm_map_t map;
1542 	struct mount *mp;
1543 	size_t resid, runlen;
1544 	int error;
1545 	bool success;
1546 
1547 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1548 	    ("%s: user address %p is not page-aligned", __func__, base));
1549 
1550 	if (p->comp != NULL)
1551 		return (compress_chunk(p, base, tmpbuf, len));
1552 
1553 	map = &p->td->td_proc->p_vmspace->vm_map;
1554 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1555 		/*
1556 		 * Attempt to page in all virtual pages in the range.  If a
1557 		 * virtual page is not backed by the pager, it is represented as
1558 		 * a hole in the file.  This can occur with zero-filled
1559 		 * anonymous memory or truncated files, for example.
1560 		 */
1561 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
1562 			error = vm_fault(map, (uintptr_t)base + runlen,
1563 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
1564 			if (runlen == 0)
1565 				success = error == KERN_SUCCESS;
1566 			else if ((error == KERN_SUCCESS) != success)
1567 				break;
1568 		}
1569 
1570 		if (success) {
1571 			error = core_write(p, base, runlen, offset,
1572 			    UIO_USERSPACE, &resid);
1573 			if (error != 0) {
1574 				if (error != EFAULT)
1575 					break;
1576 
1577 				/*
1578 				 * EFAULT may be returned if the user mapping
1579 				 * could not be accessed, e.g., because a mapped
1580 				 * file has been truncated.  Skip the page if no
1581 				 * progress was made, to protect against a
1582 				 * hypothetical scenario where vm_fault() was
1583 				 * successful but core_write() returns EFAULT
1584 				 * anyway.
1585 				 */
1586 				runlen -= resid;
1587 				if (runlen == 0) {
1588 					success = false;
1589 					runlen = PAGE_SIZE;
1590 				}
1591 			}
1592 		}
1593 		if (!success) {
1594 			error = vn_start_write(p->vp, &mp, V_WAIT);
1595 			if (error != 0)
1596 				break;
1597 			vn_lock(p->vp, LK_EXCLUSIVE | LK_RETRY);
1598 			error = vn_truncate_locked(p->vp, offset + runlen,
1599 			    false, p->td->td_ucred);
1600 			VOP_UNLOCK(p->vp);
1601 			vn_finished_write(mp);
1602 			if (error != 0)
1603 				break;
1604 		}
1605 	}
1606 	return (error);
1607 }
1608 
1609 /*
1610  * Drain into a core file.
1611  */
1612 static int
1613 sbuf_drain_core_output(void *arg, const char *data, int len)
1614 {
1615 	struct coredump_params *p;
1616 	int error, locked;
1617 
1618 	p = (struct coredump_params *)arg;
1619 
1620 	/*
1621 	 * Some kern_proc out routines that print to this sbuf may
1622 	 * call us with the process lock held. Draining with the
1623 	 * non-sleepable lock held is unsafe. The lock is needed for
1624 	 * those routines when dumping a live process. In our case we
1625 	 * can safely release the lock before draining and acquire
1626 	 * again after.
1627 	 */
1628 	locked = PROC_LOCKED(p->td->td_proc);
1629 	if (locked)
1630 		PROC_UNLOCK(p->td->td_proc);
1631 	if (p->comp != NULL)
1632 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1633 	else
1634 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1635 		    UIO_SYSSPACE, NULL);
1636 	if (locked)
1637 		PROC_LOCK(p->td->td_proc);
1638 	if (error != 0)
1639 		return (-error);
1640 	p->offset += len;
1641 	return (len);
1642 }
1643 
1644 int
1645 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1646 {
1647 	struct ucred *cred = td->td_ucred;
1648 	int error = 0;
1649 	struct sseg_closure seginfo;
1650 	struct note_info_list notelst;
1651 	struct coredump_params params;
1652 	struct note_info *ninfo;
1653 	void *hdr, *tmpbuf;
1654 	size_t hdrsize, notesz, coresize;
1655 
1656 	hdr = NULL;
1657 	tmpbuf = NULL;
1658 	TAILQ_INIT(&notelst);
1659 
1660 	/* Size the program segments. */
1661 	seginfo.count = 0;
1662 	seginfo.size = 0;
1663 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1664 
1665 	/*
1666 	 * Collect info about the core file header area.
1667 	 */
1668 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1669 	if (seginfo.count + 1 >= PN_XNUM)
1670 		hdrsize += sizeof(Elf_Shdr);
1671 	__elfN(prepare_notes)(td, &notelst, &notesz);
1672 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1673 
1674 	/* Set up core dump parameters. */
1675 	params.offset = 0;
1676 	params.active_cred = cred;
1677 	params.file_cred = NOCRED;
1678 	params.td = td;
1679 	params.vp = vp;
1680 	params.comp = NULL;
1681 
1682 #ifdef RACCT
1683 	if (racct_enable) {
1684 		PROC_LOCK(td->td_proc);
1685 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1686 		PROC_UNLOCK(td->td_proc);
1687 		if (error != 0) {
1688 			error = EFAULT;
1689 			goto done;
1690 		}
1691 	}
1692 #endif
1693 	if (coresize >= limit) {
1694 		error = EFAULT;
1695 		goto done;
1696 	}
1697 
1698 	/* Create a compression stream if necessary. */
1699 	if (compress_user_cores != 0) {
1700 		params.comp = compressor_init(core_compressed_write,
1701 		    compress_user_cores, CORE_BUF_SIZE,
1702 		    compress_user_cores_level, &params);
1703 		if (params.comp == NULL) {
1704 			error = EFAULT;
1705 			goto done;
1706 		}
1707 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1708         }
1709 
1710 	/*
1711 	 * Allocate memory for building the header, fill it up,
1712 	 * and write it out following the notes.
1713 	 */
1714 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1715 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1716 	    notesz);
1717 
1718 	/* Write the contents of all of the writable segments. */
1719 	if (error == 0) {
1720 		Elf_Phdr *php;
1721 		off_t offset;
1722 		int i;
1723 
1724 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1725 		offset = round_page(hdrsize + notesz);
1726 		for (i = 0; i < seginfo.count; i++) {
1727 			error = core_output((char *)(uintptr_t)php->p_vaddr,
1728 			    php->p_filesz, offset, &params, tmpbuf);
1729 			if (error != 0)
1730 				break;
1731 			offset += php->p_filesz;
1732 			php++;
1733 		}
1734 		if (error == 0 && params.comp != NULL)
1735 			error = compressor_flush(params.comp);
1736 	}
1737 	if (error) {
1738 		log(LOG_WARNING,
1739 		    "Failed to write core file for process %s (error %d)\n",
1740 		    curproc->p_comm, error);
1741 	}
1742 
1743 done:
1744 	free(tmpbuf, M_TEMP);
1745 	if (params.comp != NULL)
1746 		compressor_fini(params.comp);
1747 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1748 		TAILQ_REMOVE(&notelst, ninfo, link);
1749 		free(ninfo, M_TEMP);
1750 	}
1751 	if (hdr != NULL)
1752 		free(hdr, M_TEMP);
1753 
1754 	return (error);
1755 }
1756 
1757 /*
1758  * A callback for each_dumpable_segment() to write out the segment's
1759  * program header entry.
1760  */
1761 static void
1762 cb_put_phdr(vm_map_entry_t entry, void *closure)
1763 {
1764 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1765 	Elf_Phdr *phdr = phc->phdr;
1766 
1767 	phc->offset = round_page(phc->offset);
1768 
1769 	phdr->p_type = PT_LOAD;
1770 	phdr->p_offset = phc->offset;
1771 	phdr->p_vaddr = entry->start;
1772 	phdr->p_paddr = 0;
1773 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1774 	phdr->p_align = PAGE_SIZE;
1775 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1776 
1777 	phc->offset += phdr->p_filesz;
1778 	phc->phdr++;
1779 }
1780 
1781 /*
1782  * A callback for each_dumpable_segment() to gather information about
1783  * the number of segments and their total size.
1784  */
1785 static void
1786 cb_size_segment(vm_map_entry_t entry, void *closure)
1787 {
1788 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1789 
1790 	ssc->count++;
1791 	ssc->size += entry->end - entry->start;
1792 }
1793 
1794 /*
1795  * For each writable segment in the process's memory map, call the given
1796  * function with a pointer to the map entry and some arbitrary
1797  * caller-supplied data.
1798  */
1799 static void
1800 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1801 {
1802 	struct proc *p = td->td_proc;
1803 	vm_map_t map = &p->p_vmspace->vm_map;
1804 	vm_map_entry_t entry;
1805 	vm_object_t backing_object, object;
1806 	bool ignore_entry;
1807 
1808 	vm_map_lock_read(map);
1809 	VM_MAP_ENTRY_FOREACH(entry, map) {
1810 		/*
1811 		 * Don't dump inaccessible mappings, deal with legacy
1812 		 * coredump mode.
1813 		 *
1814 		 * Note that read-only segments related to the elf binary
1815 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1816 		 * need to arbitrarily ignore such segments.
1817 		 */
1818 		if (elf_legacy_coredump) {
1819 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1820 				continue;
1821 		} else {
1822 			if ((entry->protection & VM_PROT_ALL) == 0)
1823 				continue;
1824 		}
1825 
1826 		/*
1827 		 * Dont include memory segment in the coredump if
1828 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1829 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1830 		 * kernel map).
1831 		 */
1832 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1833 			continue;
1834 
1835 		if ((object = entry->object.vm_object) == NULL)
1836 			continue;
1837 
1838 		/* Ignore memory-mapped devices and such things. */
1839 		VM_OBJECT_RLOCK(object);
1840 		while ((backing_object = object->backing_object) != NULL) {
1841 			VM_OBJECT_RLOCK(backing_object);
1842 			VM_OBJECT_RUNLOCK(object);
1843 			object = backing_object;
1844 		}
1845 		ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1846 		VM_OBJECT_RUNLOCK(object);
1847 		if (ignore_entry)
1848 			continue;
1849 
1850 		(*func)(entry, closure);
1851 	}
1852 	vm_map_unlock_read(map);
1853 }
1854 
1855 /*
1856  * Write the core file header to the file, including padding up to
1857  * the page boundary.
1858  */
1859 static int
1860 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1861     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1862 {
1863 	struct note_info *ninfo;
1864 	struct sbuf *sb;
1865 	int error;
1866 
1867 	/* Fill in the header. */
1868 	bzero(hdr, hdrsize);
1869 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1870 
1871 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1872 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1873 	sbuf_start_section(sb, NULL);
1874 	sbuf_bcat(sb, hdr, hdrsize);
1875 	TAILQ_FOREACH(ninfo, notelst, link)
1876 	    __elfN(putnote)(ninfo, sb);
1877 	/* Align up to a page boundary for the program segments. */
1878 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1879 	error = sbuf_finish(sb);
1880 	sbuf_delete(sb);
1881 
1882 	return (error);
1883 }
1884 
1885 static void
1886 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1887     size_t *sizep)
1888 {
1889 	struct proc *p;
1890 	struct thread *thr;
1891 	size_t size;
1892 
1893 	p = td->td_proc;
1894 	size = 0;
1895 
1896 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1897 
1898 	/*
1899 	 * To have the debugger select the right thread (LWP) as the initial
1900 	 * thread, we dump the state of the thread passed to us in td first.
1901 	 * This is the thread that causes the core dump and thus likely to
1902 	 * be the right thread one wants to have selected in the debugger.
1903 	 */
1904 	thr = td;
1905 	while (thr != NULL) {
1906 		size += register_note(list, NT_PRSTATUS,
1907 		    __elfN(note_prstatus), thr);
1908 		size += register_note(list, NT_FPREGSET,
1909 		    __elfN(note_fpregset), thr);
1910 		size += register_note(list, NT_THRMISC,
1911 		    __elfN(note_thrmisc), thr);
1912 		size += register_note(list, NT_PTLWPINFO,
1913 		    __elfN(note_ptlwpinfo), thr);
1914 		size += register_note(list, -1,
1915 		    __elfN(note_threadmd), thr);
1916 
1917 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1918 		    TAILQ_NEXT(thr, td_plist);
1919 		if (thr == td)
1920 			thr = TAILQ_NEXT(thr, td_plist);
1921 	}
1922 
1923 	size += register_note(list, NT_PROCSTAT_PROC,
1924 	    __elfN(note_procstat_proc), p);
1925 	size += register_note(list, NT_PROCSTAT_FILES,
1926 	    note_procstat_files, p);
1927 	size += register_note(list, NT_PROCSTAT_VMMAP,
1928 	    note_procstat_vmmap, p);
1929 	size += register_note(list, NT_PROCSTAT_GROUPS,
1930 	    note_procstat_groups, p);
1931 	size += register_note(list, NT_PROCSTAT_UMASK,
1932 	    note_procstat_umask, p);
1933 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1934 	    note_procstat_rlimit, p);
1935 	size += register_note(list, NT_PROCSTAT_OSREL,
1936 	    note_procstat_osrel, p);
1937 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1938 	    __elfN(note_procstat_psstrings), p);
1939 	size += register_note(list, NT_PROCSTAT_AUXV,
1940 	    __elfN(note_procstat_auxv), p);
1941 
1942 	*sizep = size;
1943 }
1944 
1945 static void
1946 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1947     size_t notesz)
1948 {
1949 	Elf_Ehdr *ehdr;
1950 	Elf_Phdr *phdr;
1951 	Elf_Shdr *shdr;
1952 	struct phdr_closure phc;
1953 
1954 	ehdr = (Elf_Ehdr *)hdr;
1955 
1956 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1957 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1958 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1959 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1960 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1961 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1962 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1963 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1964 	ehdr->e_ident[EI_ABIVERSION] = 0;
1965 	ehdr->e_ident[EI_PAD] = 0;
1966 	ehdr->e_type = ET_CORE;
1967 	ehdr->e_machine = td->td_proc->p_elf_machine;
1968 	ehdr->e_version = EV_CURRENT;
1969 	ehdr->e_entry = 0;
1970 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1971 	ehdr->e_flags = td->td_proc->p_elf_flags;
1972 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1973 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1974 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1975 	ehdr->e_shstrndx = SHN_UNDEF;
1976 	if (numsegs + 1 < PN_XNUM) {
1977 		ehdr->e_phnum = numsegs + 1;
1978 		ehdr->e_shnum = 0;
1979 	} else {
1980 		ehdr->e_phnum = PN_XNUM;
1981 		ehdr->e_shnum = 1;
1982 
1983 		ehdr->e_shoff = ehdr->e_phoff +
1984 		    (numsegs + 1) * ehdr->e_phentsize;
1985 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1986 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1987 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1988 
1989 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1990 		memset(shdr, 0, sizeof(*shdr));
1991 		/*
1992 		 * A special first section is used to hold large segment and
1993 		 * section counts.  This was proposed by Sun Microsystems in
1994 		 * Solaris and has been adopted by Linux; the standard ELF
1995 		 * tools are already familiar with the technique.
1996 		 *
1997 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1998 		 * (or 12-7 depending on the version of the document) for more
1999 		 * details.
2000 		 */
2001 		shdr->sh_type = SHT_NULL;
2002 		shdr->sh_size = ehdr->e_shnum;
2003 		shdr->sh_link = ehdr->e_shstrndx;
2004 		shdr->sh_info = numsegs + 1;
2005 	}
2006 
2007 	/*
2008 	 * Fill in the program header entries.
2009 	 */
2010 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
2011 
2012 	/* The note segement. */
2013 	phdr->p_type = PT_NOTE;
2014 	phdr->p_offset = hdrsize;
2015 	phdr->p_vaddr = 0;
2016 	phdr->p_paddr = 0;
2017 	phdr->p_filesz = notesz;
2018 	phdr->p_memsz = 0;
2019 	phdr->p_flags = PF_R;
2020 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
2021 	phdr++;
2022 
2023 	/* All the writable segments from the program. */
2024 	phc.phdr = phdr;
2025 	phc.offset = round_page(hdrsize + notesz);
2026 	each_dumpable_segment(td, cb_put_phdr, &phc);
2027 }
2028 
2029 static size_t
2030 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
2031 {
2032 	struct note_info *ninfo;
2033 	size_t size, notesize;
2034 
2035 	size = 0;
2036 	out(arg, NULL, &size);
2037 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2038 	ninfo->type = type;
2039 	ninfo->outfunc = out;
2040 	ninfo->outarg = arg;
2041 	ninfo->outsize = size;
2042 	TAILQ_INSERT_TAIL(list, ninfo, link);
2043 
2044 	if (type == -1)
2045 		return (size);
2046 
2047 	notesize = sizeof(Elf_Note) +		/* note header */
2048 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2049 						/* note name */
2050 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2051 
2052 	return (notesize);
2053 }
2054 
2055 static size_t
2056 append_note_data(const void *src, void *dst, size_t len)
2057 {
2058 	size_t padded_len;
2059 
2060 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2061 	if (dst != NULL) {
2062 		bcopy(src, dst, len);
2063 		bzero((char *)dst + len, padded_len - len);
2064 	}
2065 	return (padded_len);
2066 }
2067 
2068 size_t
2069 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2070 {
2071 	Elf_Note *note;
2072 	char *buf;
2073 	size_t notesize;
2074 
2075 	buf = dst;
2076 	if (buf != NULL) {
2077 		note = (Elf_Note *)buf;
2078 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2079 		note->n_descsz = size;
2080 		note->n_type = type;
2081 		buf += sizeof(*note);
2082 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2083 		    sizeof(FREEBSD_ABI_VENDOR));
2084 		append_note_data(src, buf, size);
2085 		if (descp != NULL)
2086 			*descp = buf;
2087 	}
2088 
2089 	notesize = sizeof(Elf_Note) +		/* note header */
2090 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2091 						/* note name */
2092 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2093 
2094 	return (notesize);
2095 }
2096 
2097 static void
2098 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2099 {
2100 	Elf_Note note;
2101 	ssize_t old_len, sect_len;
2102 	size_t new_len, descsz, i;
2103 
2104 	if (ninfo->type == -1) {
2105 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2106 		return;
2107 	}
2108 
2109 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2110 	note.n_descsz = ninfo->outsize;
2111 	note.n_type = ninfo->type;
2112 
2113 	sbuf_bcat(sb, &note, sizeof(note));
2114 	sbuf_start_section(sb, &old_len);
2115 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2116 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2117 	if (note.n_descsz == 0)
2118 		return;
2119 	sbuf_start_section(sb, &old_len);
2120 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2121 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2122 	if (sect_len < 0)
2123 		return;
2124 
2125 	new_len = (size_t)sect_len;
2126 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2127 	if (new_len < descsz) {
2128 		/*
2129 		 * It is expected that individual note emitters will correctly
2130 		 * predict their expected output size and fill up to that size
2131 		 * themselves, padding in a format-specific way if needed.
2132 		 * However, in case they don't, just do it here with zeros.
2133 		 */
2134 		for (i = 0; i < descsz - new_len; i++)
2135 			sbuf_putc(sb, 0);
2136 	} else if (new_len > descsz) {
2137 		/*
2138 		 * We can't always truncate sb -- we may have drained some
2139 		 * of it already.
2140 		 */
2141 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2142 		    "read it (%zu > %zu).  Since it is longer than "
2143 		    "expected, this coredump's notes are corrupt.  THIS "
2144 		    "IS A BUG in the note_procstat routine for type %u.\n",
2145 		    __func__, (unsigned)note.n_type, new_len, descsz,
2146 		    (unsigned)note.n_type));
2147 	}
2148 }
2149 
2150 /*
2151  * Miscellaneous note out functions.
2152  */
2153 
2154 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2155 #include <compat/freebsd32/freebsd32.h>
2156 #include <compat/freebsd32/freebsd32_signal.h>
2157 
2158 typedef struct prstatus32 elf_prstatus_t;
2159 typedef struct prpsinfo32 elf_prpsinfo_t;
2160 typedef struct fpreg32 elf_prfpregset_t;
2161 typedef struct fpreg32 elf_fpregset_t;
2162 typedef struct reg32 elf_gregset_t;
2163 typedef struct thrmisc32 elf_thrmisc_t;
2164 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2165 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2166 typedef uint32_t elf_ps_strings_t;
2167 #else
2168 typedef prstatus_t elf_prstatus_t;
2169 typedef prpsinfo_t elf_prpsinfo_t;
2170 typedef prfpregset_t elf_prfpregset_t;
2171 typedef prfpregset_t elf_fpregset_t;
2172 typedef gregset_t elf_gregset_t;
2173 typedef thrmisc_t elf_thrmisc_t;
2174 #define ELF_KERN_PROC_MASK	0
2175 typedef struct kinfo_proc elf_kinfo_proc_t;
2176 typedef vm_offset_t elf_ps_strings_t;
2177 #endif
2178 
2179 static void
2180 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2181 {
2182 	struct sbuf sbarg;
2183 	size_t len;
2184 	char *cp, *end;
2185 	struct proc *p;
2186 	elf_prpsinfo_t *psinfo;
2187 	int error;
2188 
2189 	p = (struct proc *)arg;
2190 	if (sb != NULL) {
2191 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2192 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2193 		psinfo->pr_version = PRPSINFO_VERSION;
2194 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2195 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2196 		PROC_LOCK(p);
2197 		if (p->p_args != NULL) {
2198 			len = sizeof(psinfo->pr_psargs) - 1;
2199 			if (len > p->p_args->ar_length)
2200 				len = p->p_args->ar_length;
2201 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2202 			PROC_UNLOCK(p);
2203 			error = 0;
2204 		} else {
2205 			_PHOLD(p);
2206 			PROC_UNLOCK(p);
2207 			sbuf_new(&sbarg, psinfo->pr_psargs,
2208 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2209 			error = proc_getargv(curthread, p, &sbarg);
2210 			PRELE(p);
2211 			if (sbuf_finish(&sbarg) == 0)
2212 				len = sbuf_len(&sbarg) - 1;
2213 			else
2214 				len = sizeof(psinfo->pr_psargs) - 1;
2215 			sbuf_delete(&sbarg);
2216 		}
2217 		if (error || len == 0)
2218 			strlcpy(psinfo->pr_psargs, p->p_comm,
2219 			    sizeof(psinfo->pr_psargs));
2220 		else {
2221 			KASSERT(len < sizeof(psinfo->pr_psargs),
2222 			    ("len is too long: %zu vs %zu", len,
2223 			    sizeof(psinfo->pr_psargs)));
2224 			cp = psinfo->pr_psargs;
2225 			end = cp + len - 1;
2226 			for (;;) {
2227 				cp = memchr(cp, '\0', end - cp);
2228 				if (cp == NULL)
2229 					break;
2230 				*cp = ' ';
2231 			}
2232 		}
2233 		psinfo->pr_pid = p->p_pid;
2234 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2235 		free(psinfo, M_TEMP);
2236 	}
2237 	*sizep = sizeof(*psinfo);
2238 }
2239 
2240 static void
2241 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2242 {
2243 	struct thread *td;
2244 	elf_prstatus_t *status;
2245 
2246 	td = (struct thread *)arg;
2247 	if (sb != NULL) {
2248 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2249 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2250 		status->pr_version = PRSTATUS_VERSION;
2251 		status->pr_statussz = sizeof(elf_prstatus_t);
2252 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2253 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2254 		status->pr_osreldate = osreldate;
2255 		status->pr_cursig = td->td_proc->p_sig;
2256 		status->pr_pid = td->td_tid;
2257 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2258 		fill_regs32(td, &status->pr_reg);
2259 #else
2260 		fill_regs(td, &status->pr_reg);
2261 #endif
2262 		sbuf_bcat(sb, status, sizeof(*status));
2263 		free(status, M_TEMP);
2264 	}
2265 	*sizep = sizeof(*status);
2266 }
2267 
2268 static void
2269 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2270 {
2271 	struct thread *td;
2272 	elf_prfpregset_t *fpregset;
2273 
2274 	td = (struct thread *)arg;
2275 	if (sb != NULL) {
2276 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2277 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2278 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2279 		fill_fpregs32(td, fpregset);
2280 #else
2281 		fill_fpregs(td, fpregset);
2282 #endif
2283 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2284 		free(fpregset, M_TEMP);
2285 	}
2286 	*sizep = sizeof(*fpregset);
2287 }
2288 
2289 static void
2290 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2291 {
2292 	struct thread *td;
2293 	elf_thrmisc_t thrmisc;
2294 
2295 	td = (struct thread *)arg;
2296 	if (sb != NULL) {
2297 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2298 		bzero(&thrmisc, sizeof(thrmisc));
2299 		strcpy(thrmisc.pr_tname, td->td_name);
2300 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2301 	}
2302 	*sizep = sizeof(thrmisc);
2303 }
2304 
2305 static void
2306 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2307 {
2308 	struct thread *td;
2309 	size_t size;
2310 	int structsize;
2311 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2312 	struct ptrace_lwpinfo32 pl;
2313 #else
2314 	struct ptrace_lwpinfo pl;
2315 #endif
2316 
2317 	td = (struct thread *)arg;
2318 	size = sizeof(structsize) + sizeof(pl);
2319 	if (sb != NULL) {
2320 		KASSERT(*sizep == size, ("invalid size"));
2321 		structsize = sizeof(pl);
2322 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2323 		bzero(&pl, sizeof(pl));
2324 		pl.pl_lwpid = td->td_tid;
2325 		pl.pl_event = PL_EVENT_NONE;
2326 		pl.pl_sigmask = td->td_sigmask;
2327 		pl.pl_siglist = td->td_siglist;
2328 		if (td->td_si.si_signo != 0) {
2329 			pl.pl_event = PL_EVENT_SIGNAL;
2330 			pl.pl_flags |= PL_FLAG_SI;
2331 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2332 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2333 #else
2334 			pl.pl_siginfo = td->td_si;
2335 #endif
2336 		}
2337 		strcpy(pl.pl_tdname, td->td_name);
2338 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2339 		sbuf_bcat(sb, &pl, sizeof(pl));
2340 	}
2341 	*sizep = size;
2342 }
2343 
2344 /*
2345  * Allow for MD specific notes, as well as any MD
2346  * specific preparations for writing MI notes.
2347  */
2348 static void
2349 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2350 {
2351 	struct thread *td;
2352 	void *buf;
2353 	size_t size;
2354 
2355 	td = (struct thread *)arg;
2356 	size = *sizep;
2357 	if (size != 0 && sb != NULL)
2358 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2359 	else
2360 		buf = NULL;
2361 	size = 0;
2362 	__elfN(dump_thread)(td, buf, &size);
2363 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2364 	if (size != 0 && sb != NULL)
2365 		sbuf_bcat(sb, buf, size);
2366 	free(buf, M_TEMP);
2367 	*sizep = size;
2368 }
2369 
2370 #ifdef KINFO_PROC_SIZE
2371 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2372 #endif
2373 
2374 static void
2375 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2376 {
2377 	struct proc *p;
2378 	size_t size;
2379 	int structsize;
2380 
2381 	p = (struct proc *)arg;
2382 	size = sizeof(structsize) + p->p_numthreads *
2383 	    sizeof(elf_kinfo_proc_t);
2384 
2385 	if (sb != NULL) {
2386 		KASSERT(*sizep == size, ("invalid size"));
2387 		structsize = sizeof(elf_kinfo_proc_t);
2388 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2389 		PROC_LOCK(p);
2390 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2391 	}
2392 	*sizep = size;
2393 }
2394 
2395 #ifdef KINFO_FILE_SIZE
2396 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2397 #endif
2398 
2399 static void
2400 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2401 {
2402 	struct proc *p;
2403 	size_t size, sect_sz, i;
2404 	ssize_t start_len, sect_len;
2405 	int structsize, filedesc_flags;
2406 
2407 	if (coredump_pack_fileinfo)
2408 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2409 	else
2410 		filedesc_flags = 0;
2411 
2412 	p = (struct proc *)arg;
2413 	structsize = sizeof(struct kinfo_file);
2414 	if (sb == NULL) {
2415 		size = 0;
2416 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2417 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2418 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2419 		PROC_LOCK(p);
2420 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2421 		sbuf_finish(sb);
2422 		sbuf_delete(sb);
2423 		*sizep = size;
2424 	} else {
2425 		sbuf_start_section(sb, &start_len);
2426 
2427 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2428 		PROC_LOCK(p);
2429 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2430 		    filedesc_flags);
2431 
2432 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2433 		if (sect_len < 0)
2434 			return;
2435 		sect_sz = sect_len;
2436 
2437 		KASSERT(sect_sz <= *sizep,
2438 		    ("kern_proc_filedesc_out did not respect maxlen; "
2439 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2440 		     sect_sz - sizeof(structsize)));
2441 
2442 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2443 			sbuf_putc(sb, 0);
2444 	}
2445 }
2446 
2447 #ifdef KINFO_VMENTRY_SIZE
2448 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2449 #endif
2450 
2451 static void
2452 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2453 {
2454 	struct proc *p;
2455 	size_t size;
2456 	int structsize, vmmap_flags;
2457 
2458 	if (coredump_pack_vmmapinfo)
2459 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2460 	else
2461 		vmmap_flags = 0;
2462 
2463 	p = (struct proc *)arg;
2464 	structsize = sizeof(struct kinfo_vmentry);
2465 	if (sb == NULL) {
2466 		size = 0;
2467 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2468 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2469 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2470 		PROC_LOCK(p);
2471 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2472 		sbuf_finish(sb);
2473 		sbuf_delete(sb);
2474 		*sizep = size;
2475 	} else {
2476 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2477 		PROC_LOCK(p);
2478 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2479 		    vmmap_flags);
2480 	}
2481 }
2482 
2483 static void
2484 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2485 {
2486 	struct proc *p;
2487 	size_t size;
2488 	int structsize;
2489 
2490 	p = (struct proc *)arg;
2491 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2492 	if (sb != NULL) {
2493 		KASSERT(*sizep == size, ("invalid size"));
2494 		structsize = sizeof(gid_t);
2495 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2496 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2497 		    sizeof(gid_t));
2498 	}
2499 	*sizep = size;
2500 }
2501 
2502 static void
2503 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2504 {
2505 	struct proc *p;
2506 	size_t size;
2507 	int structsize;
2508 
2509 	p = (struct proc *)arg;
2510 	size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2511 	if (sb != NULL) {
2512 		KASSERT(*sizep == size, ("invalid size"));
2513 		structsize = sizeof(p->p_pd->pd_cmask);
2514 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2515 		sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2516 	}
2517 	*sizep = size;
2518 }
2519 
2520 static void
2521 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2522 {
2523 	struct proc *p;
2524 	struct rlimit rlim[RLIM_NLIMITS];
2525 	size_t size;
2526 	int structsize, i;
2527 
2528 	p = (struct proc *)arg;
2529 	size = sizeof(structsize) + sizeof(rlim);
2530 	if (sb != NULL) {
2531 		KASSERT(*sizep == size, ("invalid size"));
2532 		structsize = sizeof(rlim);
2533 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2534 		PROC_LOCK(p);
2535 		for (i = 0; i < RLIM_NLIMITS; i++)
2536 			lim_rlimit_proc(p, i, &rlim[i]);
2537 		PROC_UNLOCK(p);
2538 		sbuf_bcat(sb, rlim, sizeof(rlim));
2539 	}
2540 	*sizep = size;
2541 }
2542 
2543 static void
2544 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2545 {
2546 	struct proc *p;
2547 	size_t size;
2548 	int structsize;
2549 
2550 	p = (struct proc *)arg;
2551 	size = sizeof(structsize) + sizeof(p->p_osrel);
2552 	if (sb != NULL) {
2553 		KASSERT(*sizep == size, ("invalid size"));
2554 		structsize = sizeof(p->p_osrel);
2555 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2556 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2557 	}
2558 	*sizep = size;
2559 }
2560 
2561 static void
2562 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2563 {
2564 	struct proc *p;
2565 	elf_ps_strings_t ps_strings;
2566 	size_t size;
2567 	int structsize;
2568 
2569 	p = (struct proc *)arg;
2570 	size = sizeof(structsize) + sizeof(ps_strings);
2571 	if (sb != NULL) {
2572 		KASSERT(*sizep == size, ("invalid size"));
2573 		structsize = sizeof(ps_strings);
2574 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2575 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2576 #else
2577 		ps_strings = p->p_sysent->sv_psstrings;
2578 #endif
2579 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2580 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2581 	}
2582 	*sizep = size;
2583 }
2584 
2585 static void
2586 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2587 {
2588 	struct proc *p;
2589 	size_t size;
2590 	int structsize;
2591 
2592 	p = (struct proc *)arg;
2593 	if (sb == NULL) {
2594 		size = 0;
2595 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2596 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2597 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2598 		PHOLD(p);
2599 		proc_getauxv(curthread, p, sb);
2600 		PRELE(p);
2601 		sbuf_finish(sb);
2602 		sbuf_delete(sb);
2603 		*sizep = size;
2604 	} else {
2605 		structsize = sizeof(Elf_Auxinfo);
2606 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2607 		PHOLD(p);
2608 		proc_getauxv(curthread, p, sb);
2609 		PRELE(p);
2610 	}
2611 }
2612 
2613 static boolean_t
2614 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2615     const char *note_vendor, const Elf_Phdr *pnote,
2616     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2617 {
2618 	const Elf_Note *note, *note0, *note_end;
2619 	const char *note_name;
2620 	char *buf;
2621 	int i, error;
2622 	boolean_t res;
2623 
2624 	/* We need some limit, might as well use PAGE_SIZE. */
2625 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2626 		return (FALSE);
2627 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2628 	if (pnote->p_offset > PAGE_SIZE ||
2629 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2630 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2631 		if (buf == NULL) {
2632 			VOP_UNLOCK(imgp->vp);
2633 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2634 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2635 		}
2636 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2637 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2638 		    curthread->td_ucred, NOCRED, NULL, curthread);
2639 		if (error != 0) {
2640 			uprintf("i/o error PT_NOTE\n");
2641 			goto retf;
2642 		}
2643 		note = note0 = (const Elf_Note *)buf;
2644 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2645 	} else {
2646 		note = note0 = (const Elf_Note *)(imgp->image_header +
2647 		    pnote->p_offset);
2648 		note_end = (const Elf_Note *)(imgp->image_header +
2649 		    pnote->p_offset + pnote->p_filesz);
2650 		buf = NULL;
2651 	}
2652 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2653 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2654 		    (const char *)note < sizeof(Elf_Note)) {
2655 			goto retf;
2656 		}
2657 		if (note->n_namesz != checknote->n_namesz ||
2658 		    note->n_descsz != checknote->n_descsz ||
2659 		    note->n_type != checknote->n_type)
2660 			goto nextnote;
2661 		note_name = (const char *)(note + 1);
2662 		if (note_name + checknote->n_namesz >=
2663 		    (const char *)note_end || strncmp(note_vendor,
2664 		    note_name, checknote->n_namesz) != 0)
2665 			goto nextnote;
2666 
2667 		if (cb(note, cb_arg, &res))
2668 			goto ret;
2669 nextnote:
2670 		note = (const Elf_Note *)((const char *)(note + 1) +
2671 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2672 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2673 	}
2674 retf:
2675 	res = FALSE;
2676 ret:
2677 	free(buf, M_TEMP);
2678 	return (res);
2679 }
2680 
2681 struct brandnote_cb_arg {
2682 	Elf_Brandnote *brandnote;
2683 	int32_t *osrel;
2684 };
2685 
2686 static boolean_t
2687 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2688 {
2689 	struct brandnote_cb_arg *arg;
2690 
2691 	arg = arg0;
2692 
2693 	/*
2694 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2695 	 * necessary.
2696 	 */
2697 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2698 	    arg->brandnote->trans_osrel != NULL ?
2699 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2700 
2701 	return (TRUE);
2702 }
2703 
2704 static Elf_Note fctl_note = {
2705 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2706 	.n_descsz = sizeof(uint32_t),
2707 	.n_type = NT_FREEBSD_FEATURE_CTL,
2708 };
2709 
2710 struct fctl_cb_arg {
2711 	boolean_t *has_fctl0;
2712 	uint32_t *fctl0;
2713 };
2714 
2715 static boolean_t
2716 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2717 {
2718 	struct fctl_cb_arg *arg;
2719 	const Elf32_Word *desc;
2720 	uintptr_t p;
2721 
2722 	arg = arg0;
2723 	p = (uintptr_t)(note + 1);
2724 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2725 	desc = (const Elf32_Word *)p;
2726 	*arg->has_fctl0 = TRUE;
2727 	*arg->fctl0 = desc[0];
2728 	return (TRUE);
2729 }
2730 
2731 /*
2732  * Try to find the appropriate ABI-note section for checknote, fetch
2733  * the osreldate and feature control flags for binary from the ELF
2734  * OSABI-note.  Only the first page of the image is searched, the same
2735  * as for headers.
2736  */
2737 static boolean_t
2738 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2739     int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2740 {
2741 	const Elf_Phdr *phdr;
2742 	const Elf_Ehdr *hdr;
2743 	struct brandnote_cb_arg b_arg;
2744 	struct fctl_cb_arg f_arg;
2745 	int i, j;
2746 
2747 	hdr = (const Elf_Ehdr *)imgp->image_header;
2748 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2749 	b_arg.brandnote = brandnote;
2750 	b_arg.osrel = osrel;
2751 	f_arg.has_fctl0 = has_fctl0;
2752 	f_arg.fctl0 = fctl0;
2753 
2754 	for (i = 0; i < hdr->e_phnum; i++) {
2755 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2756 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2757 		    &b_arg)) {
2758 			for (j = 0; j < hdr->e_phnum; j++) {
2759 				if (phdr[j].p_type == PT_NOTE &&
2760 				    __elfN(parse_notes)(imgp, &fctl_note,
2761 				    FREEBSD_ABI_VENDOR, &phdr[j],
2762 				    note_fctl_cb, &f_arg))
2763 					break;
2764 			}
2765 			return (TRUE);
2766 		}
2767 	}
2768 	return (FALSE);
2769 
2770 }
2771 
2772 /*
2773  * Tell kern_execve.c about it, with a little help from the linker.
2774  */
2775 static struct execsw __elfN(execsw) = {
2776 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2777 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2778 };
2779 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2780 
2781 static vm_prot_t
2782 __elfN(trans_prot)(Elf_Word flags)
2783 {
2784 	vm_prot_t prot;
2785 
2786 	prot = 0;
2787 	if (flags & PF_X)
2788 		prot |= VM_PROT_EXECUTE;
2789 	if (flags & PF_W)
2790 		prot |= VM_PROT_WRITE;
2791 	if (flags & PF_R)
2792 		prot |= VM_PROT_READ;
2793 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2794 	if (i386_read_exec && (flags & PF_R))
2795 		prot |= VM_PROT_EXECUTE;
2796 #endif
2797 	return (prot);
2798 }
2799 
2800 static Elf_Word
2801 __elfN(untrans_prot)(vm_prot_t prot)
2802 {
2803 	Elf_Word flags;
2804 
2805 	flags = 0;
2806 	if (prot & VM_PROT_EXECUTE)
2807 		flags |= PF_X;
2808 	if (prot & VM_PROT_READ)
2809 		flags |= PF_R;
2810 	if (prot & VM_PROT_WRITE)
2811 		flags |= PF_W;
2812 	return (flags);
2813 }
2814 
2815 void
2816 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2817 {
2818 	uintptr_t range, rbase, gap;
2819 	int pct;
2820 
2821 	if ((imgp->map_flags & MAP_ASLR) == 0)
2822 		return;
2823 	pct = __elfN(aslr_stack_gap);
2824 	if (pct == 0)
2825 		return;
2826 	if (pct > 50)
2827 		pct = 50;
2828 	range = imgp->eff_stack_sz * pct / 100;
2829 	arc4rand(&rbase, sizeof(rbase), 0);
2830 	gap = rbase % range;
2831 	gap &= ~(sizeof(u_long) - 1);
2832 	*stack_base -= gap;
2833 }
2834