xref: /freebsd/sys/kern/imgact_elf.c (revision 3332f1b444d4a73238e9f59cca27bfc95fe936bd)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2017 Dell EMC
5  * Copyright (c) 2000-2001, 2003 David O'Brien
6  * Copyright (c) 1995-1996 Søren Schmidt
7  * Copyright (c) 1996 Peter Wemm
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_capsicum.h"
38 
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mman.h>
52 #include <sys/namei.h>
53 #include <sys/proc.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
57 #include <sys/reg.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sbuf.h>
61 #include <sys/sf_buf.h>
62 #include <sys/smp.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
65 #include <sys/stat.h>
66 #include <sys/sx.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
73 #include <sys/user.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
85 
86 #define ELF_NOTE_ROUNDSIZE	4
87 #define OLD_EI_BRAND	8
88 
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91     const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93     u_long *entry);
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98     int32_t *osrel);
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101     Elf_Brandnote *checknote, int32_t *osrel, boolean_t *has_fctl0,
102     uint32_t *fctl0);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
107     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
108     "");
109 
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117     &elf_legacy_coredump, 0,
118     "include all and only RW pages in core dumps");
119 
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
123     defined(__riscv)
124 	1;
125 #else
126 	0;
127 #endif
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 
132 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
133 int i386_read_exec = 0;
134 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
135     "enable execution from readable segments");
136 #endif
137 
138 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
139 static int
140 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
141 {
142 	u_long val;
143 	int error;
144 
145 	val = __elfN(pie_base);
146 	error = sysctl_handle_long(oidp, &val, 0, req);
147 	if (error != 0 || req->newptr == NULL)
148 		return (error);
149 	if ((val & PAGE_MASK) != 0)
150 		return (EINVAL);
151 	__elfN(pie_base) = val;
152 	return (0);
153 }
154 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
155     CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
156     sysctl_pie_base, "LU",
157     "PIE load base without randomization");
158 
159 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
160     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
161     "");
162 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
163 
164 static int __elfN(aslr_enabled) = 0;
165 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
166     &__elfN(aslr_enabled), 0,
167     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
168     ": enable address map randomization");
169 
170 static int __elfN(pie_aslr_enabled) = 0;
171 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
172     &__elfN(pie_aslr_enabled), 0,
173     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
174     ": enable address map randomization for PIE binaries");
175 
176 static int __elfN(aslr_honor_sbrk) = 1;
177 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
178     &__elfN(aslr_honor_sbrk), 0,
179     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
180 
181 static int __elfN(aslr_stack_gap) = 3;
182 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
183     &__elfN(aslr_stack_gap), 0,
184     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
185     ": maximum percentage of main stack to waste on a random gap");
186 
187 static int __elfN(sigfastblock) = 1;
188 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
189     CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
190     "enable sigfastblock for new processes");
191 
192 static bool __elfN(allow_wx) = true;
193 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
194     CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
195     "Allow pages to be mapped simultaneously writable and executable");
196 
197 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
198 
199 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
200 
201 Elf_Brandnote __elfN(freebsd_brandnote) = {
202 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
203 	.hdr.n_descsz	= sizeof(int32_t),
204 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
205 	.vendor		= FREEBSD_ABI_VENDOR,
206 	.flags		= BN_TRANSLATE_OSREL,
207 	.trans_osrel	= __elfN(freebsd_trans_osrel)
208 };
209 
210 static bool
211 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
212 {
213 	uintptr_t p;
214 
215 	p = (uintptr_t)(note + 1);
216 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
217 	*osrel = *(const int32_t *)(p);
218 
219 	return (true);
220 }
221 
222 static const char GNU_ABI_VENDOR[] = "GNU";
223 static int GNU_KFREEBSD_ABI_DESC = 3;
224 
225 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
226 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
227 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
228 	.hdr.n_type	= 1,
229 	.vendor		= GNU_ABI_VENDOR,
230 	.flags		= BN_TRANSLATE_OSREL,
231 	.trans_osrel	= kfreebsd_trans_osrel
232 };
233 
234 static bool
235 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
236 {
237 	const Elf32_Word *desc;
238 	uintptr_t p;
239 
240 	p = (uintptr_t)(note + 1);
241 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
242 
243 	desc = (const Elf32_Word *)p;
244 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
245 		return (false);
246 
247 	/*
248 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
249 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
250 	 */
251 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
252 
253 	return (true);
254 }
255 
256 int
257 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
258 {
259 	int i;
260 
261 	for (i = 0; i < MAX_BRANDS; i++) {
262 		if (elf_brand_list[i] == NULL) {
263 			elf_brand_list[i] = entry;
264 			break;
265 		}
266 	}
267 	if (i == MAX_BRANDS) {
268 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
269 			__func__, entry);
270 		return (-1);
271 	}
272 	return (0);
273 }
274 
275 int
276 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
277 {
278 	int i;
279 
280 	for (i = 0; i < MAX_BRANDS; i++) {
281 		if (elf_brand_list[i] == entry) {
282 			elf_brand_list[i] = NULL;
283 			break;
284 		}
285 	}
286 	if (i == MAX_BRANDS)
287 		return (-1);
288 	return (0);
289 }
290 
291 int
292 __elfN(brand_inuse)(Elf_Brandinfo *entry)
293 {
294 	struct proc *p;
295 	int rval = FALSE;
296 
297 	sx_slock(&allproc_lock);
298 	FOREACH_PROC_IN_SYSTEM(p) {
299 		if (p->p_sysent == entry->sysvec) {
300 			rval = TRUE;
301 			break;
302 		}
303 	}
304 	sx_sunlock(&allproc_lock);
305 
306 	return (rval);
307 }
308 
309 static Elf_Brandinfo *
310 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
311     int32_t *osrel, uint32_t *fctl0)
312 {
313 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
314 	Elf_Brandinfo *bi, *bi_m;
315 	boolean_t ret, has_fctl0;
316 	int i, interp_name_len;
317 
318 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
319 
320 	/*
321 	 * We support four types of branding -- (1) the ELF EI_OSABI field
322 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
323 	 * branding w/in the ELF header, (3) path of the `interp_path'
324 	 * field, and (4) the ".note.ABI-tag" ELF section.
325 	 */
326 
327 	/* Look for an ".note.ABI-tag" ELF section */
328 	bi_m = NULL;
329 	for (i = 0; i < MAX_BRANDS; i++) {
330 		bi = elf_brand_list[i];
331 		if (bi == NULL)
332 			continue;
333 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
334 			continue;
335 		if (hdr->e_machine == bi->machine && (bi->flags &
336 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
337 			has_fctl0 = false;
338 			*fctl0 = 0;
339 			*osrel = 0;
340 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
341 			    &has_fctl0, fctl0);
342 			/* Give brand a chance to veto check_note's guess */
343 			if (ret && bi->header_supported) {
344 				ret = bi->header_supported(imgp, osrel,
345 				    has_fctl0 ? fctl0 : NULL);
346 			}
347 			/*
348 			 * If note checker claimed the binary, but the
349 			 * interpreter path in the image does not
350 			 * match default one for the brand, try to
351 			 * search for other brands with the same
352 			 * interpreter.  Either there is better brand
353 			 * with the right interpreter, or, failing
354 			 * this, we return first brand which accepted
355 			 * our note and, optionally, header.
356 			 */
357 			if (ret && bi_m == NULL && interp != NULL &&
358 			    (bi->interp_path == NULL ||
359 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
360 			    strncmp(interp, bi->interp_path, interp_name_len)
361 			    != 0))) {
362 				bi_m = bi;
363 				ret = 0;
364 			}
365 			if (ret)
366 				return (bi);
367 		}
368 	}
369 	if (bi_m != NULL)
370 		return (bi_m);
371 
372 	/* If the executable has a brand, search for it in the brand list. */
373 	for (i = 0; i < MAX_BRANDS; i++) {
374 		bi = elf_brand_list[i];
375 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
376 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
377 			continue;
378 		if (hdr->e_machine == bi->machine &&
379 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
380 		    (bi->compat_3_brand != NULL &&
381 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
382 		    bi->compat_3_brand) == 0))) {
383 			/* Looks good, but give brand a chance to veto */
384 			if (bi->header_supported == NULL ||
385 			    bi->header_supported(imgp, NULL, NULL)) {
386 				/*
387 				 * Again, prefer strictly matching
388 				 * interpreter path.
389 				 */
390 				if (interp_name_len == 0 &&
391 				    bi->interp_path == NULL)
392 					return (bi);
393 				if (bi->interp_path != NULL &&
394 				    strlen(bi->interp_path) + 1 ==
395 				    interp_name_len && strncmp(interp,
396 				    bi->interp_path, interp_name_len) == 0)
397 					return (bi);
398 				if (bi_m == NULL)
399 					bi_m = bi;
400 			}
401 		}
402 	}
403 	if (bi_m != NULL)
404 		return (bi_m);
405 
406 	/* No known brand, see if the header is recognized by any brand */
407 	for (i = 0; i < MAX_BRANDS; i++) {
408 		bi = elf_brand_list[i];
409 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
410 		    bi->header_supported == NULL)
411 			continue;
412 		if (hdr->e_machine == bi->machine) {
413 			ret = bi->header_supported(imgp, NULL, NULL);
414 			if (ret)
415 				return (bi);
416 		}
417 	}
418 
419 	/* Lacking a known brand, search for a recognized interpreter. */
420 	if (interp != NULL) {
421 		for (i = 0; i < MAX_BRANDS; i++) {
422 			bi = elf_brand_list[i];
423 			if (bi == NULL || (bi->flags &
424 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
425 			    != 0)
426 				continue;
427 			if (hdr->e_machine == bi->machine &&
428 			    bi->interp_path != NULL &&
429 			    /* ELF image p_filesz includes terminating zero */
430 			    strlen(bi->interp_path) + 1 == interp_name_len &&
431 			    strncmp(interp, bi->interp_path, interp_name_len)
432 			    == 0 && (bi->header_supported == NULL ||
433 			    bi->header_supported(imgp, NULL, NULL)))
434 				return (bi);
435 		}
436 	}
437 
438 	/* Lacking a recognized interpreter, try the default brand */
439 	for (i = 0; i < MAX_BRANDS; i++) {
440 		bi = elf_brand_list[i];
441 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
442 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
443 			continue;
444 		if (hdr->e_machine == bi->machine &&
445 		    __elfN(fallback_brand) == bi->brand &&
446 		    (bi->header_supported == NULL ||
447 		    bi->header_supported(imgp, NULL, NULL)))
448 			return (bi);
449 	}
450 	return (NULL);
451 }
452 
453 static bool
454 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
455 {
456 	return (hdr->e_phoff <= PAGE_SIZE &&
457 	    (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
458 }
459 
460 static int
461 __elfN(check_header)(const Elf_Ehdr *hdr)
462 {
463 	Elf_Brandinfo *bi;
464 	int i;
465 
466 	if (!IS_ELF(*hdr) ||
467 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
468 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
469 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
470 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
471 	    hdr->e_version != ELF_TARG_VER)
472 		return (ENOEXEC);
473 
474 	/*
475 	 * Make sure we have at least one brand for this machine.
476 	 */
477 
478 	for (i = 0; i < MAX_BRANDS; i++) {
479 		bi = elf_brand_list[i];
480 		if (bi != NULL && bi->machine == hdr->e_machine)
481 			break;
482 	}
483 	if (i == MAX_BRANDS)
484 		return (ENOEXEC);
485 
486 	return (0);
487 }
488 
489 static int
490 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
491     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
492 {
493 	struct sf_buf *sf;
494 	int error;
495 	vm_offset_t off;
496 
497 	/*
498 	 * Create the page if it doesn't exist yet. Ignore errors.
499 	 */
500 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
501 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
502 
503 	/*
504 	 * Find the page from the underlying object.
505 	 */
506 	if (object != NULL) {
507 		sf = vm_imgact_map_page(object, offset);
508 		if (sf == NULL)
509 			return (KERN_FAILURE);
510 		off = offset - trunc_page(offset);
511 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
512 		    end - start);
513 		vm_imgact_unmap_page(sf);
514 		if (error != 0)
515 			return (KERN_FAILURE);
516 	}
517 
518 	return (KERN_SUCCESS);
519 }
520 
521 static int
522 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
523     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
524     int cow)
525 {
526 	struct sf_buf *sf;
527 	vm_offset_t off;
528 	vm_size_t sz;
529 	int error, locked, rv;
530 
531 	if (start != trunc_page(start)) {
532 		rv = __elfN(map_partial)(map, object, offset, start,
533 		    round_page(start), prot);
534 		if (rv != KERN_SUCCESS)
535 			return (rv);
536 		offset += round_page(start) - start;
537 		start = round_page(start);
538 	}
539 	if (end != round_page(end)) {
540 		rv = __elfN(map_partial)(map, object, offset +
541 		    trunc_page(end) - start, trunc_page(end), end, prot);
542 		if (rv != KERN_SUCCESS)
543 			return (rv);
544 		end = trunc_page(end);
545 	}
546 	if (start >= end)
547 		return (KERN_SUCCESS);
548 	if ((offset & PAGE_MASK) != 0) {
549 		/*
550 		 * The mapping is not page aligned.  This means that we have
551 		 * to copy the data.
552 		 */
553 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
554 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
555 		if (rv != KERN_SUCCESS)
556 			return (rv);
557 		if (object == NULL)
558 			return (KERN_SUCCESS);
559 		for (; start < end; start += sz) {
560 			sf = vm_imgact_map_page(object, offset);
561 			if (sf == NULL)
562 				return (KERN_FAILURE);
563 			off = offset - trunc_page(offset);
564 			sz = end - start;
565 			if (sz > PAGE_SIZE - off)
566 				sz = PAGE_SIZE - off;
567 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
568 			    (caddr_t)start, sz);
569 			vm_imgact_unmap_page(sf);
570 			if (error != 0)
571 				return (KERN_FAILURE);
572 			offset += sz;
573 		}
574 	} else {
575 		vm_object_reference(object);
576 		rv = vm_map_fixed(map, object, offset, start, end - start,
577 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
578 		    (object != NULL ? MAP_VN_EXEC : 0));
579 		if (rv != KERN_SUCCESS) {
580 			locked = VOP_ISLOCKED(imgp->vp);
581 			VOP_UNLOCK(imgp->vp);
582 			vm_object_deallocate(object);
583 			vn_lock(imgp->vp, locked | LK_RETRY);
584 			return (rv);
585 		} else if (object != NULL) {
586 			MPASS(imgp->vp->v_object == object);
587 			VOP_SET_TEXT_CHECKED(imgp->vp);
588 		}
589 	}
590 	return (KERN_SUCCESS);
591 }
592 
593 static int
594 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
595     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
596 {
597 	struct sf_buf *sf;
598 	size_t map_len;
599 	vm_map_t map;
600 	vm_object_t object;
601 	vm_offset_t map_addr;
602 	int error, rv, cow;
603 	size_t copy_len;
604 	vm_ooffset_t file_addr;
605 
606 	/*
607 	 * It's necessary to fail if the filsz + offset taken from the
608 	 * header is greater than the actual file pager object's size.
609 	 * If we were to allow this, then the vm_map_find() below would
610 	 * walk right off the end of the file object and into the ether.
611 	 *
612 	 * While I'm here, might as well check for something else that
613 	 * is invalid: filsz cannot be greater than memsz.
614 	 */
615 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
616 	    filsz > memsz) {
617 		uprintf("elf_load_section: truncated ELF file\n");
618 		return (ENOEXEC);
619 	}
620 
621 	object = imgp->object;
622 	map = &imgp->proc->p_vmspace->vm_map;
623 	map_addr = trunc_page((vm_offset_t)vmaddr);
624 	file_addr = trunc_page(offset);
625 
626 	/*
627 	 * We have two choices.  We can either clear the data in the last page
628 	 * of an oversized mapping, or we can start the anon mapping a page
629 	 * early and copy the initialized data into that first page.  We
630 	 * choose the second.
631 	 */
632 	if (filsz == 0)
633 		map_len = 0;
634 	else if (memsz > filsz)
635 		map_len = trunc_page(offset + filsz) - file_addr;
636 	else
637 		map_len = round_page(offset + filsz) - file_addr;
638 
639 	if (map_len != 0) {
640 		/* cow flags: don't dump readonly sections in core */
641 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
642 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
643 
644 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
645 		    map_addr, map_addr + map_len, prot, cow);
646 		if (rv != KERN_SUCCESS)
647 			return (EINVAL);
648 
649 		/* we can stop now if we've covered it all */
650 		if (memsz == filsz)
651 			return (0);
652 	}
653 
654 	/*
655 	 * We have to get the remaining bit of the file into the first part
656 	 * of the oversized map segment.  This is normally because the .data
657 	 * segment in the file is extended to provide bss.  It's a neat idea
658 	 * to try and save a page, but it's a pain in the behind to implement.
659 	 */
660 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
661 	    filsz);
662 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
663 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
664 
665 	/* This had damn well better be true! */
666 	if (map_len != 0) {
667 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
668 		    map_addr + map_len, prot, 0);
669 		if (rv != KERN_SUCCESS)
670 			return (EINVAL);
671 	}
672 
673 	if (copy_len != 0) {
674 		sf = vm_imgact_map_page(object, offset + filsz);
675 		if (sf == NULL)
676 			return (EIO);
677 
678 		/* send the page fragment to user space */
679 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
680 		    copy_len);
681 		vm_imgact_unmap_page(sf);
682 		if (error != 0)
683 			return (error);
684 	}
685 
686 	/*
687 	 * Remove write access to the page if it was only granted by map_insert
688 	 * to allow copyout.
689 	 */
690 	if ((prot & VM_PROT_WRITE) == 0)
691 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
692 		    map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
693 
694 	return (0);
695 }
696 
697 static int
698 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
699     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
700 {
701 	vm_prot_t prot;
702 	u_long base_addr;
703 	bool first;
704 	int error, i;
705 
706 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
707 
708 	base_addr = 0;
709 	first = true;
710 
711 	for (i = 0; i < hdr->e_phnum; i++) {
712 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
713 			continue;
714 
715 		/* Loadable segment */
716 		prot = __elfN(trans_prot)(phdr[i].p_flags);
717 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
718 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
719 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
720 		if (error != 0)
721 			return (error);
722 
723 		/*
724 		 * Establish the base address if this is the first segment.
725 		 */
726 		if (first) {
727   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
728 			first = false;
729 		}
730 	}
731 
732 	if (base_addrp != NULL)
733 		*base_addrp = base_addr;
734 
735 	return (0);
736 }
737 
738 /*
739  * Load the file "file" into memory.  It may be either a shared object
740  * or an executable.
741  *
742  * The "addr" reference parameter is in/out.  On entry, it specifies
743  * the address where a shared object should be loaded.  If the file is
744  * an executable, this value is ignored.  On exit, "addr" specifies
745  * where the file was actually loaded.
746  *
747  * The "entry" reference parameter is out only.  On exit, it specifies
748  * the entry point for the loaded file.
749  */
750 static int
751 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
752 	u_long *entry)
753 {
754 	struct {
755 		struct nameidata nd;
756 		struct vattr attr;
757 		struct image_params image_params;
758 	} *tempdata;
759 	const Elf_Ehdr *hdr = NULL;
760 	const Elf_Phdr *phdr = NULL;
761 	struct nameidata *nd;
762 	struct vattr *attr;
763 	struct image_params *imgp;
764 	u_long rbase;
765 	u_long base_addr = 0;
766 	int error;
767 
768 #ifdef CAPABILITY_MODE
769 	/*
770 	 * XXXJA: This check can go away once we are sufficiently confident
771 	 * that the checks in namei() are correct.
772 	 */
773 	if (IN_CAPABILITY_MODE(curthread))
774 		return (ECAPMODE);
775 #endif
776 
777 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
778 	nd = &tempdata->nd;
779 	attr = &tempdata->attr;
780 	imgp = &tempdata->image_params;
781 
782 	/*
783 	 * Initialize part of the common data
784 	 */
785 	imgp->proc = p;
786 	imgp->attr = attr;
787 
788 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
789 	    UIO_SYSSPACE, file, curthread);
790 	if ((error = namei(nd)) != 0) {
791 		nd->ni_vp = NULL;
792 		goto fail;
793 	}
794 	NDFREE(nd, NDF_ONLY_PNBUF);
795 	imgp->vp = nd->ni_vp;
796 
797 	/*
798 	 * Check permissions, modes, uid, etc on the file, and "open" it.
799 	 */
800 	error = exec_check_permissions(imgp);
801 	if (error)
802 		goto fail;
803 
804 	error = exec_map_first_page(imgp);
805 	if (error)
806 		goto fail;
807 
808 	imgp->object = nd->ni_vp->v_object;
809 
810 	hdr = (const Elf_Ehdr *)imgp->image_header;
811 	if ((error = __elfN(check_header)(hdr)) != 0)
812 		goto fail;
813 	if (hdr->e_type == ET_DYN)
814 		rbase = *addr;
815 	else if (hdr->e_type == ET_EXEC)
816 		rbase = 0;
817 	else {
818 		error = ENOEXEC;
819 		goto fail;
820 	}
821 
822 	/* Only support headers that fit within first page for now      */
823 	if (!__elfN(phdr_in_zero_page)(hdr)) {
824 		error = ENOEXEC;
825 		goto fail;
826 	}
827 
828 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
829 	if (!aligned(phdr, Elf_Addr)) {
830 		error = ENOEXEC;
831 		goto fail;
832 	}
833 
834 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
835 	if (error != 0)
836 		goto fail;
837 
838 	*addr = base_addr;
839 	*entry = (unsigned long)hdr->e_entry + rbase;
840 
841 fail:
842 	if (imgp->firstpage)
843 		exec_unmap_first_page(imgp);
844 
845 	if (nd->ni_vp) {
846 		if (imgp->textset)
847 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
848 		vput(nd->ni_vp);
849 	}
850 	free(tempdata, M_TEMP);
851 
852 	return (error);
853 }
854 
855 static u_long
856 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
857     u_int align)
858 {
859 	u_long rbase, res;
860 
861 	MPASS(vm_map_min(map) <= minv);
862 	MPASS(maxv <= vm_map_max(map));
863 	MPASS(minv < maxv);
864 	MPASS(minv + align < maxv);
865 	arc4rand(&rbase, sizeof(rbase), 0);
866 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
867 	res &= ~((u_long)align - 1);
868 	if (res >= maxv)
869 		res -= align;
870 	KASSERT(res >= minv,
871 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
872 	    res, minv, maxv, rbase));
873 	KASSERT(res < maxv,
874 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
875 	    res, maxv, minv, rbase));
876 	return (res);
877 }
878 
879 static int
880 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
881     const Elf_Phdr *phdr, u_long et_dyn_addr)
882 {
883 	struct vmspace *vmspace;
884 	const char *err_str;
885 	u_long text_size, data_size, total_size, text_addr, data_addr;
886 	u_long seg_size, seg_addr;
887 	int i;
888 
889 	err_str = NULL;
890 	text_size = data_size = total_size = text_addr = data_addr = 0;
891 
892 	for (i = 0; i < hdr->e_phnum; i++) {
893 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
894 			continue;
895 
896 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
897 		seg_size = round_page(phdr[i].p_memsz +
898 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
899 
900 		/*
901 		 * Make the largest executable segment the official
902 		 * text segment and all others data.
903 		 *
904 		 * Note that obreak() assumes that data_addr + data_size == end
905 		 * of data load area, and the ELF file format expects segments
906 		 * to be sorted by address.  If multiple data segments exist,
907 		 * the last one will be used.
908 		 */
909 
910 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
911 			text_size = seg_size;
912 			text_addr = seg_addr;
913 		} else {
914 			data_size = seg_size;
915 			data_addr = seg_addr;
916 		}
917 		total_size += seg_size;
918 	}
919 
920 	if (data_addr == 0 && data_size == 0) {
921 		data_addr = text_addr;
922 		data_size = text_size;
923 	}
924 
925 	/*
926 	 * Check limits.  It should be safe to check the
927 	 * limits after loading the segments since we do
928 	 * not actually fault in all the segments pages.
929 	 */
930 	PROC_LOCK(imgp->proc);
931 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
932 		err_str = "Data segment size exceeds process limit";
933 	else if (text_size > maxtsiz)
934 		err_str = "Text segment size exceeds system limit";
935 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
936 		err_str = "Total segment size exceeds process limit";
937 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
938 		err_str = "Data segment size exceeds resource limit";
939 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
940 		err_str = "Total segment size exceeds resource limit";
941 	PROC_UNLOCK(imgp->proc);
942 	if (err_str != NULL) {
943 		uprintf("%s\n", err_str);
944 		return (ENOMEM);
945 	}
946 
947 	vmspace = imgp->proc->p_vmspace;
948 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
949 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
950 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
951 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
952 
953 	return (0);
954 }
955 
956 static int
957 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
958     char **interpp, bool *free_interpp)
959 {
960 	struct thread *td;
961 	char *interp;
962 	int error, interp_name_len;
963 
964 	KASSERT(phdr->p_type == PT_INTERP,
965 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
966 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
967 
968 	td = curthread;
969 
970 	/* Path to interpreter */
971 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
972 		uprintf("Invalid PT_INTERP\n");
973 		return (ENOEXEC);
974 	}
975 
976 	interp_name_len = phdr->p_filesz;
977 	if (phdr->p_offset > PAGE_SIZE ||
978 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
979 		/*
980 		 * The vnode lock might be needed by the pagedaemon to
981 		 * clean pages owned by the vnode.  Do not allow sleep
982 		 * waiting for memory with the vnode locked, instead
983 		 * try non-sleepable allocation first, and if it
984 		 * fails, go to the slow path were we drop the lock
985 		 * and do M_WAITOK.  A text reference prevents
986 		 * modifications to the vnode content.
987 		 */
988 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
989 		if (interp == NULL) {
990 			VOP_UNLOCK(imgp->vp);
991 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
992 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
993 		}
994 
995 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
996 		    interp_name_len, phdr->p_offset,
997 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
998 		    NOCRED, NULL, td);
999 		if (error != 0) {
1000 			free(interp, M_TEMP);
1001 			uprintf("i/o error PT_INTERP %d\n", error);
1002 			return (error);
1003 		}
1004 		interp[interp_name_len] = '\0';
1005 
1006 		*interpp = interp;
1007 		*free_interpp = true;
1008 		return (0);
1009 	}
1010 
1011 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1012 	if (interp[interp_name_len - 1] != '\0') {
1013 		uprintf("Invalid PT_INTERP\n");
1014 		return (ENOEXEC);
1015 	}
1016 
1017 	*interpp = interp;
1018 	*free_interpp = false;
1019 	return (0);
1020 }
1021 
1022 static int
1023 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1024     const char *interp, u_long *addr, u_long *entry)
1025 {
1026 	char *path;
1027 	int error;
1028 
1029 	if (brand_info->emul_path != NULL &&
1030 	    brand_info->emul_path[0] != '\0') {
1031 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1032 		snprintf(path, MAXPATHLEN, "%s%s",
1033 		    brand_info->emul_path, interp);
1034 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1035 		free(path, M_TEMP);
1036 		if (error == 0)
1037 			return (0);
1038 	}
1039 
1040 	if (brand_info->interp_newpath != NULL &&
1041 	    (brand_info->interp_path == NULL ||
1042 	    strcmp(interp, brand_info->interp_path) == 0)) {
1043 		error = __elfN(load_file)(imgp->proc,
1044 		    brand_info->interp_newpath, addr, entry);
1045 		if (error == 0)
1046 			return (0);
1047 	}
1048 
1049 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1050 	if (error == 0)
1051 		return (0);
1052 
1053 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1054 	return (error);
1055 }
1056 
1057 /*
1058  * Impossible et_dyn_addr initial value indicating that the real base
1059  * must be calculated later with some randomization applied.
1060  */
1061 #define	ET_DYN_ADDR_RAND	1
1062 
1063 static int
1064 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1065 {
1066 	struct thread *td;
1067 	const Elf_Ehdr *hdr;
1068 	const Elf_Phdr *phdr;
1069 	Elf_Auxargs *elf_auxargs;
1070 	struct vmspace *vmspace;
1071 	vm_map_t map;
1072 	char *interp;
1073 	Elf_Brandinfo *brand_info;
1074 	struct sysentvec *sv;
1075 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1076 	u_long maxalign, mapsz, maxv, maxv1;
1077 	uint32_t fctl0;
1078 	int32_t osrel;
1079 	bool free_interp;
1080 	int error, i, n;
1081 
1082 	hdr = (const Elf_Ehdr *)imgp->image_header;
1083 
1084 	/*
1085 	 * Do we have a valid ELF header ?
1086 	 *
1087 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1088 	 * if particular brand doesn't support it.
1089 	 */
1090 	if (__elfN(check_header)(hdr) != 0 ||
1091 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1092 		return (-1);
1093 
1094 	/*
1095 	 * From here on down, we return an errno, not -1, as we've
1096 	 * detected an ELF file.
1097 	 */
1098 
1099 	if (!__elfN(phdr_in_zero_page)(hdr)) {
1100 		uprintf("Program headers not in the first page\n");
1101 		return (ENOEXEC);
1102 	}
1103 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1104 	if (!aligned(phdr, Elf_Addr)) {
1105 		uprintf("Unaligned program headers\n");
1106 		return (ENOEXEC);
1107 	}
1108 
1109 	n = error = 0;
1110 	baddr = 0;
1111 	osrel = 0;
1112 	fctl0 = 0;
1113 	entry = proghdr = 0;
1114 	interp = NULL;
1115 	free_interp = false;
1116 	td = curthread;
1117 	maxalign = PAGE_SIZE;
1118 	mapsz = 0;
1119 
1120 	for (i = 0; i < hdr->e_phnum; i++) {
1121 		switch (phdr[i].p_type) {
1122 		case PT_LOAD:
1123 			if (n == 0)
1124 				baddr = phdr[i].p_vaddr;
1125 			if (phdr[i].p_align > maxalign)
1126 				maxalign = phdr[i].p_align;
1127 			mapsz += phdr[i].p_memsz;
1128 			n++;
1129 
1130 			/*
1131 			 * If this segment contains the program headers,
1132 			 * remember their virtual address for the AT_PHDR
1133 			 * aux entry. Static binaries don't usually include
1134 			 * a PT_PHDR entry.
1135 			 */
1136 			if (phdr[i].p_offset == 0 &&
1137 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1138 				<= phdr[i].p_filesz)
1139 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1140 			break;
1141 		case PT_INTERP:
1142 			/* Path to interpreter */
1143 			if (interp != NULL) {
1144 				uprintf("Multiple PT_INTERP headers\n");
1145 				error = ENOEXEC;
1146 				goto ret;
1147 			}
1148 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1149 			    &free_interp);
1150 			if (error != 0)
1151 				goto ret;
1152 			break;
1153 		case PT_GNU_STACK:
1154 			if (__elfN(nxstack))
1155 				imgp->stack_prot =
1156 				    __elfN(trans_prot)(phdr[i].p_flags);
1157 			imgp->stack_sz = phdr[i].p_memsz;
1158 			break;
1159 		case PT_PHDR: 	/* Program header table info */
1160 			proghdr = phdr[i].p_vaddr;
1161 			break;
1162 		}
1163 	}
1164 
1165 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1166 	if (brand_info == NULL) {
1167 		uprintf("ELF binary type \"%u\" not known.\n",
1168 		    hdr->e_ident[EI_OSABI]);
1169 		error = ENOEXEC;
1170 		goto ret;
1171 	}
1172 	sv = brand_info->sysvec;
1173 	et_dyn_addr = 0;
1174 	if (hdr->e_type == ET_DYN) {
1175 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1176 			uprintf("Cannot execute shared object\n");
1177 			error = ENOEXEC;
1178 			goto ret;
1179 		}
1180 		/*
1181 		 * Honour the base load address from the dso if it is
1182 		 * non-zero for some reason.
1183 		 */
1184 		if (baddr == 0) {
1185 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1186 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1187 				et_dyn_addr = __elfN(pie_base);
1188 			else if ((__elfN(pie_aslr_enabled) &&
1189 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1190 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1191 				et_dyn_addr = ET_DYN_ADDR_RAND;
1192 			else
1193 				et_dyn_addr = __elfN(pie_base);
1194 		}
1195 	}
1196 
1197 	/*
1198 	 * Avoid a possible deadlock if the current address space is destroyed
1199 	 * and that address space maps the locked vnode.  In the common case,
1200 	 * the locked vnode's v_usecount is decremented but remains greater
1201 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1202 	 * However, in cases where the vnode lock is external, such as nullfs,
1203 	 * v_usecount may become zero.
1204 	 *
1205 	 * The VV_TEXT flag prevents modifications to the executable while
1206 	 * the vnode is unlocked.
1207 	 */
1208 	VOP_UNLOCK(imgp->vp);
1209 
1210 	/*
1211 	 * Decide whether to enable randomization of user mappings.
1212 	 * First, reset user preferences for the setid binaries.
1213 	 * Then, account for the support of the randomization by the
1214 	 * ABI, by user preferences, and make special treatment for
1215 	 * PIE binaries.
1216 	 */
1217 	if (imgp->credential_setid) {
1218 		PROC_LOCK(imgp->proc);
1219 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1220 		    P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1221 		PROC_UNLOCK(imgp->proc);
1222 	}
1223 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1224 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1225 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1226 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1227 		    ("et_dyn_addr == RAND and !ASLR"));
1228 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1229 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1230 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1231 		imgp->map_flags |= MAP_ASLR;
1232 		/*
1233 		 * If user does not care about sbrk, utilize the bss
1234 		 * grow region for mappings as well.  We can select
1235 		 * the base for the image anywere and still not suffer
1236 		 * from the fragmentation.
1237 		 */
1238 		if (!__elfN(aslr_honor_sbrk) ||
1239 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1240 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1241 	}
1242 
1243 	if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1244 	    (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1245 	    (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1246 		imgp->map_flags |= MAP_WXORX;
1247 
1248 	error = exec_new_vmspace(imgp, sv);
1249 	vmspace = imgp->proc->p_vmspace;
1250 	map = &vmspace->vm_map;
1251 
1252 	imgp->proc->p_sysent = sv;
1253 	imgp->proc->p_elf_brandinfo = brand_info;
1254 
1255 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1256 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1257 		KASSERT((map->flags & MAP_ASLR) != 0,
1258 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1259 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1260 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1261 		    /* reserve half of the address space to interpreter */
1262 		    maxv / 2, 1UL << flsl(maxalign));
1263 	}
1264 
1265 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1266 	if (error != 0)
1267 		goto ret;
1268 
1269 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1270 	if (error != 0)
1271 		goto ret;
1272 
1273 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1274 	if (error != 0)
1275 		goto ret;
1276 
1277 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1278 
1279 	/*
1280 	 * We load the dynamic linker where a userland call
1281 	 * to mmap(0, ...) would put it.  The rationale behind this
1282 	 * calculation is that it leaves room for the heap to grow to
1283 	 * its maximum allowed size.
1284 	 */
1285 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1286 	    RLIMIT_DATA));
1287 	if ((map->flags & MAP_ASLR) != 0) {
1288 		maxv1 = maxv / 2 + addr / 2;
1289 		MPASS(maxv1 >= addr);	/* No overflow */
1290 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1291 		    (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1292 		    pagesizes[1] : pagesizes[0]);
1293 	} else {
1294 		map->anon_loc = addr;
1295 	}
1296 
1297 	imgp->entry_addr = entry;
1298 
1299 	if (interp != NULL) {
1300 		VOP_UNLOCK(imgp->vp);
1301 		if ((map->flags & MAP_ASLR) != 0) {
1302 			/* Assume that interpreter fits into 1/4 of AS */
1303 			maxv1 = maxv / 2 + addr / 2;
1304 			MPASS(maxv1 >= addr);	/* No overflow */
1305 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1306 			    maxv1, PAGE_SIZE);
1307 		}
1308 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1309 		    &imgp->entry_addr);
1310 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1311 		if (error != 0)
1312 			goto ret;
1313 	} else
1314 		addr = et_dyn_addr;
1315 
1316 	/*
1317 	 * Construct auxargs table (used by the copyout_auxargs routine)
1318 	 */
1319 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1320 	if (elf_auxargs == NULL) {
1321 		VOP_UNLOCK(imgp->vp);
1322 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1323 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1324 	}
1325 	elf_auxargs->execfd = -1;
1326 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1327 	elf_auxargs->phent = hdr->e_phentsize;
1328 	elf_auxargs->phnum = hdr->e_phnum;
1329 	elf_auxargs->pagesz = PAGE_SIZE;
1330 	elf_auxargs->base = addr;
1331 	elf_auxargs->flags = 0;
1332 	elf_auxargs->entry = entry;
1333 	elf_auxargs->hdr_eflags = hdr->e_flags;
1334 
1335 	imgp->auxargs = elf_auxargs;
1336 	imgp->interpreted = 0;
1337 	imgp->reloc_base = addr;
1338 	imgp->proc->p_osrel = osrel;
1339 	imgp->proc->p_fctl0 = fctl0;
1340 	imgp->proc->p_elf_flags = hdr->e_flags;
1341 
1342 ret:
1343 	if (free_interp)
1344 		free(interp, M_TEMP);
1345 	return (error);
1346 }
1347 
1348 #define	elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1349 
1350 int
1351 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1352 {
1353 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1354 	Elf_Auxinfo *argarray, *pos;
1355 	int error;
1356 
1357 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1358 	    M_WAITOK | M_ZERO);
1359 
1360 	if (args->execfd != -1)
1361 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1362 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1363 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1364 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1365 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1366 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1367 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1368 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1369 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1370 	if (imgp->execpathp != 0)
1371 		AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1372 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1373 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1374 	if (imgp->canary != 0) {
1375 		AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1376 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1377 	}
1378 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1379 	if (imgp->pagesizes != 0) {
1380 		AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1381 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1382 	}
1383 	if (imgp->sysent->sv_timekeep_base != 0) {
1384 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1385 		    imgp->sysent->sv_timekeep_base);
1386 	}
1387 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1388 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1389 	    imgp->sysent->sv_stackprot);
1390 	if (imgp->sysent->sv_hwcap != NULL)
1391 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1392 	if (imgp->sysent->sv_hwcap2 != NULL)
1393 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1394 	AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1395 	    ELF_BSDF_SIGFASTBLK : 0);
1396 	AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1397 	AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1398 	AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1399 	AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1400 	AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1401 	if (imgp->sysent->sv_fxrng_gen_base != 0)
1402 		AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1403 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1404 
1405 	free(imgp->auxargs, M_TEMP);
1406 	imgp->auxargs = NULL;
1407 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1408 
1409 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1410 	free(argarray, M_TEMP);
1411 	return (error);
1412 }
1413 
1414 int
1415 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1416 {
1417 	Elf_Addr *base;
1418 
1419 	base = (Elf_Addr *)*stack_base;
1420 	base--;
1421 	if (elf_suword(base, imgp->args->argc) == -1)
1422 		return (EFAULT);
1423 	*stack_base = (uintptr_t)base;
1424 	return (0);
1425 }
1426 
1427 /*
1428  * Code for generating ELF core dumps.
1429  */
1430 
1431 typedef void (*segment_callback)(vm_map_entry_t, void *);
1432 
1433 /* Closure for cb_put_phdr(). */
1434 struct phdr_closure {
1435 	Elf_Phdr *phdr;		/* Program header to fill in */
1436 	Elf_Off offset;		/* Offset of segment in core file */
1437 };
1438 
1439 struct note_info {
1440 	int		type;		/* Note type. */
1441 	outfunc_t 	outfunc; 	/* Output function. */
1442 	void		*outarg;	/* Argument for the output function. */
1443 	size_t		outsize;	/* Output size. */
1444 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1445 };
1446 
1447 TAILQ_HEAD(note_info_list, note_info);
1448 
1449 extern int compress_user_cores;
1450 extern int compress_user_cores_level;
1451 
1452 static void cb_put_phdr(vm_map_entry_t, void *);
1453 static void cb_size_segment(vm_map_entry_t, void *);
1454 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1455     int);
1456 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1457     struct note_info_list *, size_t, int);
1458 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1459 
1460 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1461 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1462 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1463 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1464 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1465 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1466 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1467 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1468 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1469 static void note_procstat_files(void *, struct sbuf *, size_t *);
1470 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1471 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1472 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1473 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1474 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1475 
1476 static int
1477 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1478 {
1479 
1480 	return (core_write((struct coredump_params *)arg, base, len, offset,
1481 	    UIO_SYSSPACE, NULL));
1482 }
1483 
1484 int
1485 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1486 {
1487 	struct ucred *cred = td->td_ucred;
1488 	int compm, error = 0;
1489 	struct sseg_closure seginfo;
1490 	struct note_info_list notelst;
1491 	struct coredump_params params;
1492 	struct note_info *ninfo;
1493 	void *hdr, *tmpbuf;
1494 	size_t hdrsize, notesz, coresize;
1495 
1496 	hdr = NULL;
1497 	tmpbuf = NULL;
1498 	TAILQ_INIT(&notelst);
1499 
1500 	/* Size the program segments. */
1501 	__elfN(size_segments)(td, &seginfo, flags);
1502 
1503 	/*
1504 	 * Collect info about the core file header area.
1505 	 */
1506 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1507 	if (seginfo.count + 1 >= PN_XNUM)
1508 		hdrsize += sizeof(Elf_Shdr);
1509 	td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, &notelst, &notesz);
1510 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1511 
1512 	/* Set up core dump parameters. */
1513 	params.offset = 0;
1514 	params.active_cred = cred;
1515 	params.file_cred = NOCRED;
1516 	params.td = td;
1517 	params.vp = vp;
1518 	params.comp = NULL;
1519 
1520 #ifdef RACCT
1521 	if (racct_enable) {
1522 		PROC_LOCK(td->td_proc);
1523 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1524 		PROC_UNLOCK(td->td_proc);
1525 		if (error != 0) {
1526 			error = EFAULT;
1527 			goto done;
1528 		}
1529 	}
1530 #endif
1531 	if (coresize >= limit) {
1532 		error = EFAULT;
1533 		goto done;
1534 	}
1535 
1536 	/* Create a compression stream if necessary. */
1537 	compm = compress_user_cores;
1538 	if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1539 	    compm == 0)
1540 		compm = COMPRESS_GZIP;
1541 	if (compm != 0) {
1542 		params.comp = compressor_init(core_compressed_write,
1543 		    compm, CORE_BUF_SIZE,
1544 		    compress_user_cores_level, &params);
1545 		if (params.comp == NULL) {
1546 			error = EFAULT;
1547 			goto done;
1548 		}
1549 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1550         }
1551 
1552 	/*
1553 	 * Allocate memory for building the header, fill it up,
1554 	 * and write it out following the notes.
1555 	 */
1556 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1557 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1558 	    notesz, flags);
1559 
1560 	/* Write the contents of all of the writable segments. */
1561 	if (error == 0) {
1562 		Elf_Phdr *php;
1563 		off_t offset;
1564 		int i;
1565 
1566 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1567 		offset = round_page(hdrsize + notesz);
1568 		for (i = 0; i < seginfo.count; i++) {
1569 			error = core_output((char *)(uintptr_t)php->p_vaddr,
1570 			    php->p_filesz, offset, &params, tmpbuf);
1571 			if (error != 0)
1572 				break;
1573 			offset += php->p_filesz;
1574 			php++;
1575 		}
1576 		if (error == 0 && params.comp != NULL)
1577 			error = compressor_flush(params.comp);
1578 	}
1579 	if (error) {
1580 		log(LOG_WARNING,
1581 		    "Failed to write core file for process %s (error %d)\n",
1582 		    curproc->p_comm, error);
1583 	}
1584 
1585 done:
1586 	free(tmpbuf, M_TEMP);
1587 	if (params.comp != NULL)
1588 		compressor_fini(params.comp);
1589 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1590 		TAILQ_REMOVE(&notelst, ninfo, link);
1591 		free(ninfo, M_TEMP);
1592 	}
1593 	if (hdr != NULL)
1594 		free(hdr, M_TEMP);
1595 
1596 	return (error);
1597 }
1598 
1599 /*
1600  * A callback for each_dumpable_segment() to write out the segment's
1601  * program header entry.
1602  */
1603 static void
1604 cb_put_phdr(vm_map_entry_t entry, void *closure)
1605 {
1606 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1607 	Elf_Phdr *phdr = phc->phdr;
1608 
1609 	phc->offset = round_page(phc->offset);
1610 
1611 	phdr->p_type = PT_LOAD;
1612 	phdr->p_offset = phc->offset;
1613 	phdr->p_vaddr = entry->start;
1614 	phdr->p_paddr = 0;
1615 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1616 	phdr->p_align = PAGE_SIZE;
1617 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1618 
1619 	phc->offset += phdr->p_filesz;
1620 	phc->phdr++;
1621 }
1622 
1623 /*
1624  * A callback for each_dumpable_segment() to gather information about
1625  * the number of segments and their total size.
1626  */
1627 static void
1628 cb_size_segment(vm_map_entry_t entry, void *closure)
1629 {
1630 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1631 
1632 	ssc->count++;
1633 	ssc->size += entry->end - entry->start;
1634 }
1635 
1636 void
1637 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1638     int flags)
1639 {
1640 	seginfo->count = 0;
1641 	seginfo->size = 0;
1642 
1643 	each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1644 }
1645 
1646 /*
1647  * For each writable segment in the process's memory map, call the given
1648  * function with a pointer to the map entry and some arbitrary
1649  * caller-supplied data.
1650  */
1651 static void
1652 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1653     int flags)
1654 {
1655 	struct proc *p = td->td_proc;
1656 	vm_map_t map = &p->p_vmspace->vm_map;
1657 	vm_map_entry_t entry;
1658 	vm_object_t backing_object, object;
1659 	bool ignore_entry;
1660 
1661 	vm_map_lock_read(map);
1662 	VM_MAP_ENTRY_FOREACH(entry, map) {
1663 		/*
1664 		 * Don't dump inaccessible mappings, deal with legacy
1665 		 * coredump mode.
1666 		 *
1667 		 * Note that read-only segments related to the elf binary
1668 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1669 		 * need to arbitrarily ignore such segments.
1670 		 */
1671 		if ((flags & SVC_ALL) == 0) {
1672 			if (elf_legacy_coredump) {
1673 				if ((entry->protection & VM_PROT_RW) !=
1674 				    VM_PROT_RW)
1675 					continue;
1676 			} else {
1677 				if ((entry->protection & VM_PROT_ALL) == 0)
1678 					continue;
1679 			}
1680 		}
1681 
1682 		/*
1683 		 * Dont include memory segment in the coredump if
1684 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1685 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1686 		 * kernel map).
1687 		 */
1688 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1689 			continue;
1690 		if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1691 		    (flags & SVC_ALL) == 0)
1692 			continue;
1693 		if ((object = entry->object.vm_object) == NULL)
1694 			continue;
1695 
1696 		/* Ignore memory-mapped devices and such things. */
1697 		VM_OBJECT_RLOCK(object);
1698 		while ((backing_object = object->backing_object) != NULL) {
1699 			VM_OBJECT_RLOCK(backing_object);
1700 			VM_OBJECT_RUNLOCK(object);
1701 			object = backing_object;
1702 		}
1703 		ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1704 		VM_OBJECT_RUNLOCK(object);
1705 		if (ignore_entry)
1706 			continue;
1707 
1708 		(*func)(entry, closure);
1709 	}
1710 	vm_map_unlock_read(map);
1711 }
1712 
1713 /*
1714  * Write the core file header to the file, including padding up to
1715  * the page boundary.
1716  */
1717 static int
1718 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1719     size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1720     int flags)
1721 {
1722 	struct note_info *ninfo;
1723 	struct sbuf *sb;
1724 	int error;
1725 
1726 	/* Fill in the header. */
1727 	bzero(hdr, hdrsize);
1728 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1729 
1730 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1731 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1732 	sbuf_start_section(sb, NULL);
1733 	sbuf_bcat(sb, hdr, hdrsize);
1734 	TAILQ_FOREACH(ninfo, notelst, link)
1735 	    __elfN(putnote)(p->td, ninfo, sb);
1736 	/* Align up to a page boundary for the program segments. */
1737 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1738 	error = sbuf_finish(sb);
1739 	sbuf_delete(sb);
1740 
1741 	return (error);
1742 }
1743 
1744 void
1745 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1746     size_t *sizep)
1747 {
1748 	struct proc *p;
1749 	struct thread *thr;
1750 	size_t size;
1751 
1752 	p = td->td_proc;
1753 	size = 0;
1754 
1755 	size += __elfN(register_note)(td, list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1756 
1757 	/*
1758 	 * To have the debugger select the right thread (LWP) as the initial
1759 	 * thread, we dump the state of the thread passed to us in td first.
1760 	 * This is the thread that causes the core dump and thus likely to
1761 	 * be the right thread one wants to have selected in the debugger.
1762 	 */
1763 	thr = td;
1764 	while (thr != NULL) {
1765 		size += __elfN(register_note)(td, list, NT_PRSTATUS,
1766 		    __elfN(note_prstatus), thr);
1767 		size += __elfN(register_note)(td, list, NT_FPREGSET,
1768 		    __elfN(note_fpregset), thr);
1769 		size += __elfN(register_note)(td, list, NT_THRMISC,
1770 		    __elfN(note_thrmisc), thr);
1771 		size += __elfN(register_note)(td, list, NT_PTLWPINFO,
1772 		    __elfN(note_ptlwpinfo), thr);
1773 		size += __elfN(register_note)(td, list, -1,
1774 		    __elfN(note_threadmd), thr);
1775 
1776 		thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1777 		    TAILQ_NEXT(thr, td_plist);
1778 		if (thr == td)
1779 			thr = TAILQ_NEXT(thr, td_plist);
1780 	}
1781 
1782 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1783 	    __elfN(note_procstat_proc), p);
1784 	size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1785 	    note_procstat_files, p);
1786 	size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1787 	    note_procstat_vmmap, p);
1788 	size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1789 	    note_procstat_groups, p);
1790 	size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1791 	    note_procstat_umask, p);
1792 	size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1793 	    note_procstat_rlimit, p);
1794 	size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1795 	    note_procstat_osrel, p);
1796 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1797 	    __elfN(note_procstat_psstrings), p);
1798 	size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1799 	    __elfN(note_procstat_auxv), p);
1800 
1801 	*sizep = size;
1802 }
1803 
1804 void
1805 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1806     size_t notesz, int flags)
1807 {
1808 	Elf_Ehdr *ehdr;
1809 	Elf_Phdr *phdr;
1810 	Elf_Shdr *shdr;
1811 	struct phdr_closure phc;
1812 	Elf_Brandinfo *bi;
1813 
1814 	ehdr = (Elf_Ehdr *)hdr;
1815 	bi = td->td_proc->p_elf_brandinfo;
1816 
1817 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1818 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1819 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1820 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1821 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1822 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1823 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1824 	ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1825 	ehdr->e_ident[EI_ABIVERSION] = 0;
1826 	ehdr->e_ident[EI_PAD] = 0;
1827 	ehdr->e_type = ET_CORE;
1828 	ehdr->e_machine = bi->machine;
1829 	ehdr->e_version = EV_CURRENT;
1830 	ehdr->e_entry = 0;
1831 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1832 	ehdr->e_flags = td->td_proc->p_elf_flags;
1833 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1834 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1835 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1836 	ehdr->e_shstrndx = SHN_UNDEF;
1837 	if (numsegs + 1 < PN_XNUM) {
1838 		ehdr->e_phnum = numsegs + 1;
1839 		ehdr->e_shnum = 0;
1840 	} else {
1841 		ehdr->e_phnum = PN_XNUM;
1842 		ehdr->e_shnum = 1;
1843 
1844 		ehdr->e_shoff = ehdr->e_phoff +
1845 		    (numsegs + 1) * ehdr->e_phentsize;
1846 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1847 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1848 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1849 
1850 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1851 		memset(shdr, 0, sizeof(*shdr));
1852 		/*
1853 		 * A special first section is used to hold large segment and
1854 		 * section counts.  This was proposed by Sun Microsystems in
1855 		 * Solaris and has been adopted by Linux; the standard ELF
1856 		 * tools are already familiar with the technique.
1857 		 *
1858 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1859 		 * (or 12-7 depending on the version of the document) for more
1860 		 * details.
1861 		 */
1862 		shdr->sh_type = SHT_NULL;
1863 		shdr->sh_size = ehdr->e_shnum;
1864 		shdr->sh_link = ehdr->e_shstrndx;
1865 		shdr->sh_info = numsegs + 1;
1866 	}
1867 
1868 	/*
1869 	 * Fill in the program header entries.
1870 	 */
1871 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1872 
1873 	/* The note segement. */
1874 	phdr->p_type = PT_NOTE;
1875 	phdr->p_offset = hdrsize;
1876 	phdr->p_vaddr = 0;
1877 	phdr->p_paddr = 0;
1878 	phdr->p_filesz = notesz;
1879 	phdr->p_memsz = 0;
1880 	phdr->p_flags = PF_R;
1881 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1882 	phdr++;
1883 
1884 	/* All the writable segments from the program. */
1885 	phc.phdr = phdr;
1886 	phc.offset = round_page(hdrsize + notesz);
1887 	each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1888 }
1889 
1890 size_t
1891 __elfN(register_note)(struct thread *td, struct note_info_list *list,
1892     int type, outfunc_t out, void *arg)
1893 {
1894 	const struct sysentvec *sv;
1895 	struct note_info *ninfo;
1896 	size_t size, notesize;
1897 
1898 	sv = td->td_proc->p_sysent;
1899 	size = 0;
1900 	out(arg, NULL, &size);
1901 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1902 	ninfo->type = type;
1903 	ninfo->outfunc = out;
1904 	ninfo->outarg = arg;
1905 	ninfo->outsize = size;
1906 	TAILQ_INSERT_TAIL(list, ninfo, link);
1907 
1908 	if (type == -1)
1909 		return (size);
1910 
1911 	notesize = sizeof(Elf_Note) +		/* note header */
1912 	    roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1913 						/* note name */
1914 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1915 
1916 	return (notesize);
1917 }
1918 
1919 static size_t
1920 append_note_data(const void *src, void *dst, size_t len)
1921 {
1922 	size_t padded_len;
1923 
1924 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1925 	if (dst != NULL) {
1926 		bcopy(src, dst, len);
1927 		bzero((char *)dst + len, padded_len - len);
1928 	}
1929 	return (padded_len);
1930 }
1931 
1932 size_t
1933 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1934 {
1935 	Elf_Note *note;
1936 	char *buf;
1937 	size_t notesize;
1938 
1939 	buf = dst;
1940 	if (buf != NULL) {
1941 		note = (Elf_Note *)buf;
1942 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1943 		note->n_descsz = size;
1944 		note->n_type = type;
1945 		buf += sizeof(*note);
1946 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1947 		    sizeof(FREEBSD_ABI_VENDOR));
1948 		append_note_data(src, buf, size);
1949 		if (descp != NULL)
1950 			*descp = buf;
1951 	}
1952 
1953 	notesize = sizeof(Elf_Note) +		/* note header */
1954 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1955 						/* note name */
1956 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1957 
1958 	return (notesize);
1959 }
1960 
1961 static void
1962 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
1963 {
1964 	Elf_Note note;
1965 	const struct sysentvec *sv;
1966 	ssize_t old_len, sect_len;
1967 	size_t new_len, descsz, i;
1968 
1969 	if (ninfo->type == -1) {
1970 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1971 		return;
1972 	}
1973 
1974 	sv = td->td_proc->p_sysent;
1975 
1976 	note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
1977 	note.n_descsz = ninfo->outsize;
1978 	note.n_type = ninfo->type;
1979 
1980 	sbuf_bcat(sb, &note, sizeof(note));
1981 	sbuf_start_section(sb, &old_len);
1982 	sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
1983 	    strlen(sv->sv_elf_core_abi_vendor) + 1);
1984 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1985 	if (note.n_descsz == 0)
1986 		return;
1987 	sbuf_start_section(sb, &old_len);
1988 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1989 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1990 	if (sect_len < 0)
1991 		return;
1992 
1993 	new_len = (size_t)sect_len;
1994 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1995 	if (new_len < descsz) {
1996 		/*
1997 		 * It is expected that individual note emitters will correctly
1998 		 * predict their expected output size and fill up to that size
1999 		 * themselves, padding in a format-specific way if needed.
2000 		 * However, in case they don't, just do it here with zeros.
2001 		 */
2002 		for (i = 0; i < descsz - new_len; i++)
2003 			sbuf_putc(sb, 0);
2004 	} else if (new_len > descsz) {
2005 		/*
2006 		 * We can't always truncate sb -- we may have drained some
2007 		 * of it already.
2008 		 */
2009 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2010 		    "read it (%zu > %zu).  Since it is longer than "
2011 		    "expected, this coredump's notes are corrupt.  THIS "
2012 		    "IS A BUG in the note_procstat routine for type %u.\n",
2013 		    __func__, (unsigned)note.n_type, new_len, descsz,
2014 		    (unsigned)note.n_type));
2015 	}
2016 }
2017 
2018 /*
2019  * Miscellaneous note out functions.
2020  */
2021 
2022 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2023 #include <compat/freebsd32/freebsd32.h>
2024 #include <compat/freebsd32/freebsd32_signal.h>
2025 
2026 typedef struct prstatus32 elf_prstatus_t;
2027 typedef struct prpsinfo32 elf_prpsinfo_t;
2028 typedef struct fpreg32 elf_prfpregset_t;
2029 typedef struct fpreg32 elf_fpregset_t;
2030 typedef struct reg32 elf_gregset_t;
2031 typedef struct thrmisc32 elf_thrmisc_t;
2032 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2033 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2034 typedef uint32_t elf_ps_strings_t;
2035 #else
2036 typedef prstatus_t elf_prstatus_t;
2037 typedef prpsinfo_t elf_prpsinfo_t;
2038 typedef prfpregset_t elf_prfpregset_t;
2039 typedef prfpregset_t elf_fpregset_t;
2040 typedef gregset_t elf_gregset_t;
2041 typedef thrmisc_t elf_thrmisc_t;
2042 #define ELF_KERN_PROC_MASK	0
2043 typedef struct kinfo_proc elf_kinfo_proc_t;
2044 typedef vm_offset_t elf_ps_strings_t;
2045 #endif
2046 
2047 static void
2048 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2049 {
2050 	struct sbuf sbarg;
2051 	size_t len;
2052 	char *cp, *end;
2053 	struct proc *p;
2054 	elf_prpsinfo_t *psinfo;
2055 	int error;
2056 
2057 	p = arg;
2058 	if (sb != NULL) {
2059 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2060 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2061 		psinfo->pr_version = PRPSINFO_VERSION;
2062 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2063 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2064 		PROC_LOCK(p);
2065 		if (p->p_args != NULL) {
2066 			len = sizeof(psinfo->pr_psargs) - 1;
2067 			if (len > p->p_args->ar_length)
2068 				len = p->p_args->ar_length;
2069 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2070 			PROC_UNLOCK(p);
2071 			error = 0;
2072 		} else {
2073 			_PHOLD(p);
2074 			PROC_UNLOCK(p);
2075 			sbuf_new(&sbarg, psinfo->pr_psargs,
2076 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2077 			error = proc_getargv(curthread, p, &sbarg);
2078 			PRELE(p);
2079 			if (sbuf_finish(&sbarg) == 0)
2080 				len = sbuf_len(&sbarg) - 1;
2081 			else
2082 				len = sizeof(psinfo->pr_psargs) - 1;
2083 			sbuf_delete(&sbarg);
2084 		}
2085 		if (error || len == 0)
2086 			strlcpy(psinfo->pr_psargs, p->p_comm,
2087 			    sizeof(psinfo->pr_psargs));
2088 		else {
2089 			KASSERT(len < sizeof(psinfo->pr_psargs),
2090 			    ("len is too long: %zu vs %zu", len,
2091 			    sizeof(psinfo->pr_psargs)));
2092 			cp = psinfo->pr_psargs;
2093 			end = cp + len - 1;
2094 			for (;;) {
2095 				cp = memchr(cp, '\0', end - cp);
2096 				if (cp == NULL)
2097 					break;
2098 				*cp = ' ';
2099 			}
2100 		}
2101 		psinfo->pr_pid = p->p_pid;
2102 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2103 		free(psinfo, M_TEMP);
2104 	}
2105 	*sizep = sizeof(*psinfo);
2106 }
2107 
2108 static void
2109 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2110 {
2111 	struct thread *td;
2112 	elf_prstatus_t *status;
2113 
2114 	td = arg;
2115 	if (sb != NULL) {
2116 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2117 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2118 		status->pr_version = PRSTATUS_VERSION;
2119 		status->pr_statussz = sizeof(elf_prstatus_t);
2120 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2121 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2122 		status->pr_osreldate = osreldate;
2123 		status->pr_cursig = td->td_proc->p_sig;
2124 		status->pr_pid = td->td_tid;
2125 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2126 		fill_regs32(td, &status->pr_reg);
2127 #else
2128 		fill_regs(td, &status->pr_reg);
2129 #endif
2130 		sbuf_bcat(sb, status, sizeof(*status));
2131 		free(status, M_TEMP);
2132 	}
2133 	*sizep = sizeof(*status);
2134 }
2135 
2136 static void
2137 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2138 {
2139 	struct thread *td;
2140 	elf_prfpregset_t *fpregset;
2141 
2142 	td = arg;
2143 	if (sb != NULL) {
2144 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2145 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2146 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2147 		fill_fpregs32(td, fpregset);
2148 #else
2149 		fill_fpregs(td, fpregset);
2150 #endif
2151 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2152 		free(fpregset, M_TEMP);
2153 	}
2154 	*sizep = sizeof(*fpregset);
2155 }
2156 
2157 static void
2158 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2159 {
2160 	struct thread *td;
2161 	elf_thrmisc_t thrmisc;
2162 
2163 	td = arg;
2164 	if (sb != NULL) {
2165 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2166 		bzero(&thrmisc, sizeof(thrmisc));
2167 		strcpy(thrmisc.pr_tname, td->td_name);
2168 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2169 	}
2170 	*sizep = sizeof(thrmisc);
2171 }
2172 
2173 static void
2174 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2175 {
2176 	struct thread *td;
2177 	size_t size;
2178 	int structsize;
2179 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2180 	struct ptrace_lwpinfo32 pl;
2181 #else
2182 	struct ptrace_lwpinfo pl;
2183 #endif
2184 
2185 	td = arg;
2186 	size = sizeof(structsize) + sizeof(pl);
2187 	if (sb != NULL) {
2188 		KASSERT(*sizep == size, ("invalid size"));
2189 		structsize = sizeof(pl);
2190 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2191 		bzero(&pl, sizeof(pl));
2192 		pl.pl_lwpid = td->td_tid;
2193 		pl.pl_event = PL_EVENT_NONE;
2194 		pl.pl_sigmask = td->td_sigmask;
2195 		pl.pl_siglist = td->td_siglist;
2196 		if (td->td_si.si_signo != 0) {
2197 			pl.pl_event = PL_EVENT_SIGNAL;
2198 			pl.pl_flags |= PL_FLAG_SI;
2199 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2200 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2201 #else
2202 			pl.pl_siginfo = td->td_si;
2203 #endif
2204 		}
2205 		strcpy(pl.pl_tdname, td->td_name);
2206 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2207 		sbuf_bcat(sb, &pl, sizeof(pl));
2208 	}
2209 	*sizep = size;
2210 }
2211 
2212 /*
2213  * Allow for MD specific notes, as well as any MD
2214  * specific preparations for writing MI notes.
2215  */
2216 static void
2217 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2218 {
2219 	struct thread *td;
2220 	void *buf;
2221 	size_t size;
2222 
2223 	td = (struct thread *)arg;
2224 	size = *sizep;
2225 	if (size != 0 && sb != NULL)
2226 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2227 	else
2228 		buf = NULL;
2229 	size = 0;
2230 	__elfN(dump_thread)(td, buf, &size);
2231 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2232 	if (size != 0 && sb != NULL)
2233 		sbuf_bcat(sb, buf, size);
2234 	free(buf, M_TEMP);
2235 	*sizep = size;
2236 }
2237 
2238 #ifdef KINFO_PROC_SIZE
2239 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2240 #endif
2241 
2242 static void
2243 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2244 {
2245 	struct proc *p;
2246 	size_t size;
2247 	int structsize;
2248 
2249 	p = arg;
2250 	size = sizeof(structsize) + p->p_numthreads *
2251 	    sizeof(elf_kinfo_proc_t);
2252 
2253 	if (sb != NULL) {
2254 		KASSERT(*sizep == size, ("invalid size"));
2255 		structsize = sizeof(elf_kinfo_proc_t);
2256 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2257 		sx_slock(&proctree_lock);
2258 		PROC_LOCK(p);
2259 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2260 		sx_sunlock(&proctree_lock);
2261 	}
2262 	*sizep = size;
2263 }
2264 
2265 #ifdef KINFO_FILE_SIZE
2266 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2267 #endif
2268 
2269 static void
2270 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2271 {
2272 	struct proc *p;
2273 	size_t size, sect_sz, i;
2274 	ssize_t start_len, sect_len;
2275 	int structsize, filedesc_flags;
2276 
2277 	if (coredump_pack_fileinfo)
2278 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2279 	else
2280 		filedesc_flags = 0;
2281 
2282 	p = arg;
2283 	structsize = sizeof(struct kinfo_file);
2284 	if (sb == NULL) {
2285 		size = 0;
2286 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2287 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2288 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2289 		PROC_LOCK(p);
2290 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2291 		sbuf_finish(sb);
2292 		sbuf_delete(sb);
2293 		*sizep = size;
2294 	} else {
2295 		sbuf_start_section(sb, &start_len);
2296 
2297 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2298 		PROC_LOCK(p);
2299 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2300 		    filedesc_flags);
2301 
2302 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2303 		if (sect_len < 0)
2304 			return;
2305 		sect_sz = sect_len;
2306 
2307 		KASSERT(sect_sz <= *sizep,
2308 		    ("kern_proc_filedesc_out did not respect maxlen; "
2309 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2310 		     sect_sz - sizeof(structsize)));
2311 
2312 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2313 			sbuf_putc(sb, 0);
2314 	}
2315 }
2316 
2317 #ifdef KINFO_VMENTRY_SIZE
2318 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2319 #endif
2320 
2321 static void
2322 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2323 {
2324 	struct proc *p;
2325 	size_t size;
2326 	int structsize, vmmap_flags;
2327 
2328 	if (coredump_pack_vmmapinfo)
2329 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2330 	else
2331 		vmmap_flags = 0;
2332 
2333 	p = arg;
2334 	structsize = sizeof(struct kinfo_vmentry);
2335 	if (sb == NULL) {
2336 		size = 0;
2337 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2338 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2339 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2340 		PROC_LOCK(p);
2341 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2342 		sbuf_finish(sb);
2343 		sbuf_delete(sb);
2344 		*sizep = size;
2345 	} else {
2346 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2347 		PROC_LOCK(p);
2348 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2349 		    vmmap_flags);
2350 	}
2351 }
2352 
2353 static void
2354 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2355 {
2356 	struct proc *p;
2357 	size_t size;
2358 	int structsize;
2359 
2360 	p = arg;
2361 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2362 	if (sb != NULL) {
2363 		KASSERT(*sizep == size, ("invalid size"));
2364 		structsize = sizeof(gid_t);
2365 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2366 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2367 		    sizeof(gid_t));
2368 	}
2369 	*sizep = size;
2370 }
2371 
2372 static void
2373 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2374 {
2375 	struct proc *p;
2376 	size_t size;
2377 	int structsize;
2378 
2379 	p = arg;
2380 	size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2381 	if (sb != NULL) {
2382 		KASSERT(*sizep == size, ("invalid size"));
2383 		structsize = sizeof(p->p_pd->pd_cmask);
2384 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2385 		sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2386 	}
2387 	*sizep = size;
2388 }
2389 
2390 static void
2391 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2392 {
2393 	struct proc *p;
2394 	struct rlimit rlim[RLIM_NLIMITS];
2395 	size_t size;
2396 	int structsize, i;
2397 
2398 	p = arg;
2399 	size = sizeof(structsize) + sizeof(rlim);
2400 	if (sb != NULL) {
2401 		KASSERT(*sizep == size, ("invalid size"));
2402 		structsize = sizeof(rlim);
2403 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2404 		PROC_LOCK(p);
2405 		for (i = 0; i < RLIM_NLIMITS; i++)
2406 			lim_rlimit_proc(p, i, &rlim[i]);
2407 		PROC_UNLOCK(p);
2408 		sbuf_bcat(sb, rlim, sizeof(rlim));
2409 	}
2410 	*sizep = size;
2411 }
2412 
2413 static void
2414 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2415 {
2416 	struct proc *p;
2417 	size_t size;
2418 	int structsize;
2419 
2420 	p = arg;
2421 	size = sizeof(structsize) + sizeof(p->p_osrel);
2422 	if (sb != NULL) {
2423 		KASSERT(*sizep == size, ("invalid size"));
2424 		structsize = sizeof(p->p_osrel);
2425 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2426 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2427 	}
2428 	*sizep = size;
2429 }
2430 
2431 static void
2432 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2433 {
2434 	struct proc *p;
2435 	elf_ps_strings_t ps_strings;
2436 	size_t size;
2437 	int structsize;
2438 
2439 	p = arg;
2440 	size = sizeof(structsize) + sizeof(ps_strings);
2441 	if (sb != NULL) {
2442 		KASSERT(*sizep == size, ("invalid size"));
2443 		structsize = sizeof(ps_strings);
2444 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2445 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2446 #else
2447 		ps_strings = p->p_sysent->sv_psstrings;
2448 #endif
2449 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2450 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2451 	}
2452 	*sizep = size;
2453 }
2454 
2455 static void
2456 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2457 {
2458 	struct proc *p;
2459 	size_t size;
2460 	int structsize;
2461 
2462 	p = arg;
2463 	if (sb == NULL) {
2464 		size = 0;
2465 		sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2466 		    SBUF_FIXEDLEN);
2467 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2468 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2469 		PHOLD(p);
2470 		proc_getauxv(curthread, p, sb);
2471 		PRELE(p);
2472 		sbuf_finish(sb);
2473 		sbuf_delete(sb);
2474 		*sizep = size;
2475 	} else {
2476 		structsize = sizeof(Elf_Auxinfo);
2477 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2478 		PHOLD(p);
2479 		proc_getauxv(curthread, p, sb);
2480 		PRELE(p);
2481 	}
2482 }
2483 
2484 static boolean_t
2485 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2486     const char *note_vendor, const Elf_Phdr *pnote,
2487     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2488 {
2489 	const Elf_Note *note, *note0, *note_end;
2490 	const char *note_name;
2491 	char *buf;
2492 	int i, error;
2493 	boolean_t res;
2494 
2495 	/* We need some limit, might as well use PAGE_SIZE. */
2496 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2497 		return (FALSE);
2498 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2499 	if (pnote->p_offset > PAGE_SIZE ||
2500 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2501 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2502 		if (buf == NULL) {
2503 			VOP_UNLOCK(imgp->vp);
2504 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2505 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2506 		}
2507 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2508 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2509 		    curthread->td_ucred, NOCRED, NULL, curthread);
2510 		if (error != 0) {
2511 			uprintf("i/o error PT_NOTE\n");
2512 			goto retf;
2513 		}
2514 		note = note0 = (const Elf_Note *)buf;
2515 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2516 	} else {
2517 		note = note0 = (const Elf_Note *)(imgp->image_header +
2518 		    pnote->p_offset);
2519 		note_end = (const Elf_Note *)(imgp->image_header +
2520 		    pnote->p_offset + pnote->p_filesz);
2521 		buf = NULL;
2522 	}
2523 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2524 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2525 		    (const char *)note < sizeof(Elf_Note)) {
2526 			goto retf;
2527 		}
2528 		if (note->n_namesz != checknote->n_namesz ||
2529 		    note->n_descsz != checknote->n_descsz ||
2530 		    note->n_type != checknote->n_type)
2531 			goto nextnote;
2532 		note_name = (const char *)(note + 1);
2533 		if (note_name + checknote->n_namesz >=
2534 		    (const char *)note_end || strncmp(note_vendor,
2535 		    note_name, checknote->n_namesz) != 0)
2536 			goto nextnote;
2537 
2538 		if (cb(note, cb_arg, &res))
2539 			goto ret;
2540 nextnote:
2541 		note = (const Elf_Note *)((const char *)(note + 1) +
2542 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2543 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2544 	}
2545 retf:
2546 	res = FALSE;
2547 ret:
2548 	free(buf, M_TEMP);
2549 	return (res);
2550 }
2551 
2552 struct brandnote_cb_arg {
2553 	Elf_Brandnote *brandnote;
2554 	int32_t *osrel;
2555 };
2556 
2557 static boolean_t
2558 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2559 {
2560 	struct brandnote_cb_arg *arg;
2561 
2562 	arg = arg0;
2563 
2564 	/*
2565 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2566 	 * necessary.
2567 	 */
2568 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2569 	    arg->brandnote->trans_osrel != NULL ?
2570 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2571 
2572 	return (TRUE);
2573 }
2574 
2575 static Elf_Note fctl_note = {
2576 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2577 	.n_descsz = sizeof(uint32_t),
2578 	.n_type = NT_FREEBSD_FEATURE_CTL,
2579 };
2580 
2581 struct fctl_cb_arg {
2582 	boolean_t *has_fctl0;
2583 	uint32_t *fctl0;
2584 };
2585 
2586 static boolean_t
2587 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2588 {
2589 	struct fctl_cb_arg *arg;
2590 	const Elf32_Word *desc;
2591 	uintptr_t p;
2592 
2593 	arg = arg0;
2594 	p = (uintptr_t)(note + 1);
2595 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2596 	desc = (const Elf32_Word *)p;
2597 	*arg->has_fctl0 = TRUE;
2598 	*arg->fctl0 = desc[0];
2599 	*res = TRUE;
2600 	return (TRUE);
2601 }
2602 
2603 /*
2604  * Try to find the appropriate ABI-note section for checknote, fetch
2605  * the osreldate and feature control flags for binary from the ELF
2606  * OSABI-note.  Only the first page of the image is searched, the same
2607  * as for headers.
2608  */
2609 static boolean_t
2610 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2611     int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2612 {
2613 	const Elf_Phdr *phdr;
2614 	const Elf_Ehdr *hdr;
2615 	struct brandnote_cb_arg b_arg;
2616 	struct fctl_cb_arg f_arg;
2617 	int i, j;
2618 
2619 	hdr = (const Elf_Ehdr *)imgp->image_header;
2620 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2621 	b_arg.brandnote = brandnote;
2622 	b_arg.osrel = osrel;
2623 	f_arg.has_fctl0 = has_fctl0;
2624 	f_arg.fctl0 = fctl0;
2625 
2626 	for (i = 0; i < hdr->e_phnum; i++) {
2627 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2628 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2629 		    &b_arg)) {
2630 			for (j = 0; j < hdr->e_phnum; j++) {
2631 				if (phdr[j].p_type == PT_NOTE &&
2632 				    __elfN(parse_notes)(imgp, &fctl_note,
2633 				    FREEBSD_ABI_VENDOR, &phdr[j],
2634 				    note_fctl_cb, &f_arg))
2635 					break;
2636 			}
2637 			return (TRUE);
2638 		}
2639 	}
2640 	return (FALSE);
2641 
2642 }
2643 
2644 /*
2645  * Tell kern_execve.c about it, with a little help from the linker.
2646  */
2647 static struct execsw __elfN(execsw) = {
2648 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2649 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2650 };
2651 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2652 
2653 static vm_prot_t
2654 __elfN(trans_prot)(Elf_Word flags)
2655 {
2656 	vm_prot_t prot;
2657 
2658 	prot = 0;
2659 	if (flags & PF_X)
2660 		prot |= VM_PROT_EXECUTE;
2661 	if (flags & PF_W)
2662 		prot |= VM_PROT_WRITE;
2663 	if (flags & PF_R)
2664 		prot |= VM_PROT_READ;
2665 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2666 	if (i386_read_exec && (flags & PF_R))
2667 		prot |= VM_PROT_EXECUTE;
2668 #endif
2669 	return (prot);
2670 }
2671 
2672 static Elf_Word
2673 __elfN(untrans_prot)(vm_prot_t prot)
2674 {
2675 	Elf_Word flags;
2676 
2677 	flags = 0;
2678 	if (prot & VM_PROT_EXECUTE)
2679 		flags |= PF_X;
2680 	if (prot & VM_PROT_READ)
2681 		flags |= PF_R;
2682 	if (prot & VM_PROT_WRITE)
2683 		flags |= PF_W;
2684 	return (flags);
2685 }
2686 
2687 vm_size_t
2688 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2689 {
2690 	uintptr_t range, rbase, gap;
2691 	int pct;
2692 
2693 	pct = __elfN(aslr_stack_gap);
2694 	if (pct == 0)
2695 		return (0);
2696 	if (pct > 50)
2697 		pct = 50;
2698 	range = imgp->eff_stack_sz * pct / 100;
2699 	arc4rand(&rbase, sizeof(rbase), 0);
2700 	gap = rbase % range;
2701 	gap &= ~(sizeof(u_long) - 1);
2702 	*stack_base -= gap;
2703 	return (gap);
2704 }
2705