xref: /freebsd/sys/kern/imgact_elf.c (revision 13ec1e3155c7e9bf037b12af186351b7fa9b9450)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2017 Dell EMC
5  * Copyright (c) 2000-2001, 2003 David O'Brien
6  * Copyright (c) 1995-1996 Søren Schmidt
7  * Copyright (c) 1996 Peter Wemm
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_capsicum.h"
38 
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mman.h>
52 #include <sys/namei.h>
53 #include <sys/proc.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
57 #include <sys/reg.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sbuf.h>
61 #include <sys/sf_buf.h>
62 #include <sys/smp.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
65 #include <sys/stat.h>
66 #include <sys/sx.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
73 #include <sys/user.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
85 
86 #define ELF_NOTE_ROUNDSIZE	4
87 #define OLD_EI_BRAND	8
88 
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91     const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93     u_long *entry);
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98     int32_t *osrel);
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static bool __elfN(check_note)(struct image_params *imgp,
101     Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0,
102     uint32_t *fctl0);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
107     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
108     "");
109 
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117     &elf_legacy_coredump, 0,
118     "include all and only RW pages in core dumps");
119 
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
123     defined(__riscv)
124 	1;
125 #else
126 	0;
127 #endif
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 
132 #if defined(__amd64__)
133 static int __elfN(vdso) = 1;
134 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
135     vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0,
136     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading");
137 #else
138 static int __elfN(vdso) = 0;
139 #endif
140 
141 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
142 int i386_read_exec = 0;
143 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
144     "enable execution from readable segments");
145 #endif
146 
147 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
148 static int
149 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
150 {
151 	u_long val;
152 	int error;
153 
154 	val = __elfN(pie_base);
155 	error = sysctl_handle_long(oidp, &val, 0, req);
156 	if (error != 0 || req->newptr == NULL)
157 		return (error);
158 	if ((val & PAGE_MASK) != 0)
159 		return (EINVAL);
160 	__elfN(pie_base) = val;
161 	return (0);
162 }
163 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
164     CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
165     sysctl_pie_base, "LU",
166     "PIE load base without randomization");
167 
168 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
169     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
170     "");
171 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
172 
173 /*
174  * While for 64-bit machines ASLR works properly, there are
175  * still some problems when using 32-bit architectures. For this
176  * reason ASLR is only enabled by default when running native
177  * 64-bit non-PIE executables.
178  */
179 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
180 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
181     &__elfN(aslr_enabled), 0,
182     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
183     ": enable address map randomization");
184 
185 /*
186  * Enable ASLR only for 64-bit PIE binaries by default.
187  */
188 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
189 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
190     &__elfN(pie_aslr_enabled), 0,
191     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
192     ": enable address map randomization for PIE binaries");
193 
194 /*
195  * Sbrk is now deprecated and it can be assumed, that in most
196  * cases it will not be used anyway. This setting is valid only
197  * for the ASLR enabled and allows for utilizing the bss grow region.
198  */
199 static int __elfN(aslr_honor_sbrk) = 0;
200 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
201     &__elfN(aslr_honor_sbrk), 0,
202     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
203 
204 static int __elfN(aslr_stack) = 1;
205 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
206     &__elfN(aslr_stack), 0,
207     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
208     ": enable stack address randomization");
209 
210 static int __elfN(sigfastblock) = 1;
211 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
212     CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
213     "enable sigfastblock for new processes");
214 
215 static bool __elfN(allow_wx) = true;
216 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
217     CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
218     "Allow pages to be mapped simultaneously writable and executable");
219 
220 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
221 
222 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
223 
224 Elf_Brandnote __elfN(freebsd_brandnote) = {
225 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
226 	.hdr.n_descsz	= sizeof(int32_t),
227 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
228 	.vendor		= FREEBSD_ABI_VENDOR,
229 	.flags		= BN_TRANSLATE_OSREL,
230 	.trans_osrel	= __elfN(freebsd_trans_osrel)
231 };
232 
233 static bool
234 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
235 {
236 	uintptr_t p;
237 
238 	p = (uintptr_t)(note + 1);
239 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
240 	*osrel = *(const int32_t *)(p);
241 
242 	return (true);
243 }
244 
245 static const char GNU_ABI_VENDOR[] = "GNU";
246 static int GNU_KFREEBSD_ABI_DESC = 3;
247 
248 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
249 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
250 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
251 	.hdr.n_type	= 1,
252 	.vendor		= GNU_ABI_VENDOR,
253 	.flags		= BN_TRANSLATE_OSREL,
254 	.trans_osrel	= kfreebsd_trans_osrel
255 };
256 
257 static bool
258 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
259 {
260 	const Elf32_Word *desc;
261 	uintptr_t p;
262 
263 	p = (uintptr_t)(note + 1);
264 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
265 
266 	desc = (const Elf32_Word *)p;
267 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
268 		return (false);
269 
270 	/*
271 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
272 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
273 	 */
274 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
275 
276 	return (true);
277 }
278 
279 int
280 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
281 {
282 	int i;
283 
284 	for (i = 0; i < MAX_BRANDS; i++) {
285 		if (elf_brand_list[i] == NULL) {
286 			elf_brand_list[i] = entry;
287 			break;
288 		}
289 	}
290 	if (i == MAX_BRANDS) {
291 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
292 			__func__, entry);
293 		return (-1);
294 	}
295 	return (0);
296 }
297 
298 int
299 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
300 {
301 	int i;
302 
303 	for (i = 0; i < MAX_BRANDS; i++) {
304 		if (elf_brand_list[i] == entry) {
305 			elf_brand_list[i] = NULL;
306 			break;
307 		}
308 	}
309 	if (i == MAX_BRANDS)
310 		return (-1);
311 	return (0);
312 }
313 
314 bool
315 __elfN(brand_inuse)(Elf_Brandinfo *entry)
316 {
317 	struct proc *p;
318 	bool rval = false;
319 
320 	sx_slock(&allproc_lock);
321 	FOREACH_PROC_IN_SYSTEM(p) {
322 		if (p->p_sysent == entry->sysvec) {
323 			rval = true;
324 			break;
325 		}
326 	}
327 	sx_sunlock(&allproc_lock);
328 
329 	return (rval);
330 }
331 
332 static Elf_Brandinfo *
333 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
334     int32_t *osrel, uint32_t *fctl0)
335 {
336 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
337 	Elf_Brandinfo *bi, *bi_m;
338 	bool ret, has_fctl0;
339 	int i, interp_name_len;
340 
341 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
342 
343 	/*
344 	 * We support four types of branding -- (1) the ELF EI_OSABI field
345 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
346 	 * branding w/in the ELF header, (3) path of the `interp_path'
347 	 * field, and (4) the ".note.ABI-tag" ELF section.
348 	 */
349 
350 	/* Look for an ".note.ABI-tag" ELF section */
351 	bi_m = NULL;
352 	for (i = 0; i < MAX_BRANDS; i++) {
353 		bi = elf_brand_list[i];
354 		if (bi == NULL)
355 			continue;
356 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
357 			continue;
358 		if (hdr->e_machine == bi->machine && (bi->flags &
359 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
360 			has_fctl0 = false;
361 			*fctl0 = 0;
362 			*osrel = 0;
363 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
364 			    &has_fctl0, fctl0);
365 			/* Give brand a chance to veto check_note's guess */
366 			if (ret && bi->header_supported) {
367 				ret = bi->header_supported(imgp, osrel,
368 				    has_fctl0 ? fctl0 : NULL);
369 			}
370 			/*
371 			 * If note checker claimed the binary, but the
372 			 * interpreter path in the image does not
373 			 * match default one for the brand, try to
374 			 * search for other brands with the same
375 			 * interpreter.  Either there is better brand
376 			 * with the right interpreter, or, failing
377 			 * this, we return first brand which accepted
378 			 * our note and, optionally, header.
379 			 */
380 			if (ret && bi_m == NULL && interp != NULL &&
381 			    (bi->interp_path == NULL ||
382 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
383 			    strncmp(interp, bi->interp_path, interp_name_len)
384 			    != 0))) {
385 				bi_m = bi;
386 				ret = 0;
387 			}
388 			if (ret)
389 				return (bi);
390 		}
391 	}
392 	if (bi_m != NULL)
393 		return (bi_m);
394 
395 	/* If the executable has a brand, search for it in the brand list. */
396 	for (i = 0; i < MAX_BRANDS; i++) {
397 		bi = elf_brand_list[i];
398 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
399 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
400 			continue;
401 		if (hdr->e_machine == bi->machine &&
402 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
403 		    (bi->compat_3_brand != NULL &&
404 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
405 		    bi->compat_3_brand) == 0))) {
406 			/* Looks good, but give brand a chance to veto */
407 			if (bi->header_supported == NULL ||
408 			    bi->header_supported(imgp, NULL, NULL)) {
409 				/*
410 				 * Again, prefer strictly matching
411 				 * interpreter path.
412 				 */
413 				if (interp_name_len == 0 &&
414 				    bi->interp_path == NULL)
415 					return (bi);
416 				if (bi->interp_path != NULL &&
417 				    strlen(bi->interp_path) + 1 ==
418 				    interp_name_len && strncmp(interp,
419 				    bi->interp_path, interp_name_len) == 0)
420 					return (bi);
421 				if (bi_m == NULL)
422 					bi_m = bi;
423 			}
424 		}
425 	}
426 	if (bi_m != NULL)
427 		return (bi_m);
428 
429 	/* No known brand, see if the header is recognized by any brand */
430 	for (i = 0; i < MAX_BRANDS; i++) {
431 		bi = elf_brand_list[i];
432 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
433 		    bi->header_supported == NULL)
434 			continue;
435 		if (hdr->e_machine == bi->machine) {
436 			ret = bi->header_supported(imgp, NULL, NULL);
437 			if (ret)
438 				return (bi);
439 		}
440 	}
441 
442 	/* Lacking a known brand, search for a recognized interpreter. */
443 	if (interp != NULL) {
444 		for (i = 0; i < MAX_BRANDS; i++) {
445 			bi = elf_brand_list[i];
446 			if (bi == NULL || (bi->flags &
447 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
448 			    != 0)
449 				continue;
450 			if (hdr->e_machine == bi->machine &&
451 			    bi->interp_path != NULL &&
452 			    /* ELF image p_filesz includes terminating zero */
453 			    strlen(bi->interp_path) + 1 == interp_name_len &&
454 			    strncmp(interp, bi->interp_path, interp_name_len)
455 			    == 0 && (bi->header_supported == NULL ||
456 			    bi->header_supported(imgp, NULL, NULL)))
457 				return (bi);
458 		}
459 	}
460 
461 	/* Lacking a recognized interpreter, try the default brand */
462 	for (i = 0; i < MAX_BRANDS; i++) {
463 		bi = elf_brand_list[i];
464 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
465 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
466 			continue;
467 		if (hdr->e_machine == bi->machine &&
468 		    __elfN(fallback_brand) == bi->brand &&
469 		    (bi->header_supported == NULL ||
470 		    bi->header_supported(imgp, NULL, NULL)))
471 			return (bi);
472 	}
473 	return (NULL);
474 }
475 
476 static bool
477 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
478 {
479 	return (hdr->e_phoff <= PAGE_SIZE &&
480 	    (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
481 }
482 
483 static int
484 __elfN(check_header)(const Elf_Ehdr *hdr)
485 {
486 	Elf_Brandinfo *bi;
487 	int i;
488 
489 	if (!IS_ELF(*hdr) ||
490 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
491 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
492 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
493 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
494 	    hdr->e_version != ELF_TARG_VER)
495 		return (ENOEXEC);
496 
497 	/*
498 	 * Make sure we have at least one brand for this machine.
499 	 */
500 
501 	for (i = 0; i < MAX_BRANDS; i++) {
502 		bi = elf_brand_list[i];
503 		if (bi != NULL && bi->machine == hdr->e_machine)
504 			break;
505 	}
506 	if (i == MAX_BRANDS)
507 		return (ENOEXEC);
508 
509 	return (0);
510 }
511 
512 static int
513 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
514     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
515 {
516 	struct sf_buf *sf;
517 	int error;
518 	vm_offset_t off;
519 
520 	/*
521 	 * Create the page if it doesn't exist yet. Ignore errors.
522 	 */
523 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
524 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
525 
526 	/*
527 	 * Find the page from the underlying object.
528 	 */
529 	if (object != NULL) {
530 		sf = vm_imgact_map_page(object, offset);
531 		if (sf == NULL)
532 			return (KERN_FAILURE);
533 		off = offset - trunc_page(offset);
534 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
535 		    end - start);
536 		vm_imgact_unmap_page(sf);
537 		if (error != 0)
538 			return (KERN_FAILURE);
539 	}
540 
541 	return (KERN_SUCCESS);
542 }
543 
544 static int
545 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
546     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
547     int cow)
548 {
549 	struct sf_buf *sf;
550 	vm_offset_t off;
551 	vm_size_t sz;
552 	int error, locked, rv;
553 
554 	if (start != trunc_page(start)) {
555 		rv = __elfN(map_partial)(map, object, offset, start,
556 		    round_page(start), prot);
557 		if (rv != KERN_SUCCESS)
558 			return (rv);
559 		offset += round_page(start) - start;
560 		start = round_page(start);
561 	}
562 	if (end != round_page(end)) {
563 		rv = __elfN(map_partial)(map, object, offset +
564 		    trunc_page(end) - start, trunc_page(end), end, prot);
565 		if (rv != KERN_SUCCESS)
566 			return (rv);
567 		end = trunc_page(end);
568 	}
569 	if (start >= end)
570 		return (KERN_SUCCESS);
571 	if ((offset & PAGE_MASK) != 0) {
572 		/*
573 		 * The mapping is not page aligned.  This means that we have
574 		 * to copy the data.
575 		 */
576 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
577 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
578 		if (rv != KERN_SUCCESS)
579 			return (rv);
580 		if (object == NULL)
581 			return (KERN_SUCCESS);
582 		for (; start < end; start += sz) {
583 			sf = vm_imgact_map_page(object, offset);
584 			if (sf == NULL)
585 				return (KERN_FAILURE);
586 			off = offset - trunc_page(offset);
587 			sz = end - start;
588 			if (sz > PAGE_SIZE - off)
589 				sz = PAGE_SIZE - off;
590 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
591 			    (caddr_t)start, sz);
592 			vm_imgact_unmap_page(sf);
593 			if (error != 0)
594 				return (KERN_FAILURE);
595 			offset += sz;
596 		}
597 	} else {
598 		vm_object_reference(object);
599 		rv = vm_map_fixed(map, object, offset, start, end - start,
600 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
601 		    (object != NULL ? MAP_VN_EXEC : 0));
602 		if (rv != KERN_SUCCESS) {
603 			locked = VOP_ISLOCKED(imgp->vp);
604 			VOP_UNLOCK(imgp->vp);
605 			vm_object_deallocate(object);
606 			vn_lock(imgp->vp, locked | LK_RETRY);
607 			return (rv);
608 		} else if (object != NULL) {
609 			MPASS(imgp->vp->v_object == object);
610 			VOP_SET_TEXT_CHECKED(imgp->vp);
611 		}
612 	}
613 	return (KERN_SUCCESS);
614 }
615 
616 static int
617 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
618     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
619 {
620 	struct sf_buf *sf;
621 	size_t map_len;
622 	vm_map_t map;
623 	vm_object_t object;
624 	vm_offset_t map_addr;
625 	int error, rv, cow;
626 	size_t copy_len;
627 	vm_ooffset_t file_addr;
628 
629 	/*
630 	 * It's necessary to fail if the filsz + offset taken from the
631 	 * header is greater than the actual file pager object's size.
632 	 * If we were to allow this, then the vm_map_find() below would
633 	 * walk right off the end of the file object and into the ether.
634 	 *
635 	 * While I'm here, might as well check for something else that
636 	 * is invalid: filsz cannot be greater than memsz.
637 	 */
638 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
639 	    filsz > memsz) {
640 		uprintf("elf_load_section: truncated ELF file\n");
641 		return (ENOEXEC);
642 	}
643 
644 	object = imgp->object;
645 	map = &imgp->proc->p_vmspace->vm_map;
646 	map_addr = trunc_page((vm_offset_t)vmaddr);
647 	file_addr = trunc_page(offset);
648 
649 	/*
650 	 * We have two choices.  We can either clear the data in the last page
651 	 * of an oversized mapping, or we can start the anon mapping a page
652 	 * early and copy the initialized data into that first page.  We
653 	 * choose the second.
654 	 */
655 	if (filsz == 0)
656 		map_len = 0;
657 	else if (memsz > filsz)
658 		map_len = trunc_page(offset + filsz) - file_addr;
659 	else
660 		map_len = round_page(offset + filsz) - file_addr;
661 
662 	if (map_len != 0) {
663 		/* cow flags: don't dump readonly sections in core */
664 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
665 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
666 
667 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
668 		    map_addr, map_addr + map_len, prot, cow);
669 		if (rv != KERN_SUCCESS)
670 			return (EINVAL);
671 
672 		/* we can stop now if we've covered it all */
673 		if (memsz == filsz)
674 			return (0);
675 	}
676 
677 	/*
678 	 * We have to get the remaining bit of the file into the first part
679 	 * of the oversized map segment.  This is normally because the .data
680 	 * segment in the file is extended to provide bss.  It's a neat idea
681 	 * to try and save a page, but it's a pain in the behind to implement.
682 	 */
683 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
684 	    filsz);
685 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
686 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
687 
688 	/* This had damn well better be true! */
689 	if (map_len != 0) {
690 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
691 		    map_addr + map_len, prot, 0);
692 		if (rv != KERN_SUCCESS)
693 			return (EINVAL);
694 	}
695 
696 	if (copy_len != 0) {
697 		sf = vm_imgact_map_page(object, offset + filsz);
698 		if (sf == NULL)
699 			return (EIO);
700 
701 		/* send the page fragment to user space */
702 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
703 		    copy_len);
704 		vm_imgact_unmap_page(sf);
705 		if (error != 0)
706 			return (error);
707 	}
708 
709 	/*
710 	 * Remove write access to the page if it was only granted by map_insert
711 	 * to allow copyout.
712 	 */
713 	if ((prot & VM_PROT_WRITE) == 0)
714 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
715 		    map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
716 
717 	return (0);
718 }
719 
720 static int
721 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
722     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
723 {
724 	vm_prot_t prot;
725 	u_long base_addr;
726 	bool first;
727 	int error, i;
728 
729 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
730 
731 	base_addr = 0;
732 	first = true;
733 
734 	for (i = 0; i < hdr->e_phnum; i++) {
735 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
736 			continue;
737 
738 		/* Loadable segment */
739 		prot = __elfN(trans_prot)(phdr[i].p_flags);
740 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
741 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
742 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
743 		if (error != 0)
744 			return (error);
745 
746 		/*
747 		 * Establish the base address if this is the first segment.
748 		 */
749 		if (first) {
750   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
751 			first = false;
752 		}
753 	}
754 
755 	if (base_addrp != NULL)
756 		*base_addrp = base_addr;
757 
758 	return (0);
759 }
760 
761 /*
762  * Load the file "file" into memory.  It may be either a shared object
763  * or an executable.
764  *
765  * The "addr" reference parameter is in/out.  On entry, it specifies
766  * the address where a shared object should be loaded.  If the file is
767  * an executable, this value is ignored.  On exit, "addr" specifies
768  * where the file was actually loaded.
769  *
770  * The "entry" reference parameter is out only.  On exit, it specifies
771  * the entry point for the loaded file.
772  */
773 static int
774 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
775 	u_long *entry)
776 {
777 	struct {
778 		struct nameidata nd;
779 		struct vattr attr;
780 		struct image_params image_params;
781 	} *tempdata;
782 	const Elf_Ehdr *hdr = NULL;
783 	const Elf_Phdr *phdr = NULL;
784 	struct nameidata *nd;
785 	struct vattr *attr;
786 	struct image_params *imgp;
787 	u_long rbase;
788 	u_long base_addr = 0;
789 	int error;
790 
791 #ifdef CAPABILITY_MODE
792 	/*
793 	 * XXXJA: This check can go away once we are sufficiently confident
794 	 * that the checks in namei() are correct.
795 	 */
796 	if (IN_CAPABILITY_MODE(curthread))
797 		return (ECAPMODE);
798 #endif
799 
800 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
801 	nd = &tempdata->nd;
802 	attr = &tempdata->attr;
803 	imgp = &tempdata->image_params;
804 
805 	/*
806 	 * Initialize part of the common data
807 	 */
808 	imgp->proc = p;
809 	imgp->attr = attr;
810 
811 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
812 	    UIO_SYSSPACE, file);
813 	if ((error = namei(nd)) != 0) {
814 		nd->ni_vp = NULL;
815 		goto fail;
816 	}
817 	NDFREE(nd, NDF_ONLY_PNBUF);
818 	imgp->vp = nd->ni_vp;
819 
820 	/*
821 	 * Check permissions, modes, uid, etc on the file, and "open" it.
822 	 */
823 	error = exec_check_permissions(imgp);
824 	if (error)
825 		goto fail;
826 
827 	error = exec_map_first_page(imgp);
828 	if (error)
829 		goto fail;
830 
831 	imgp->object = nd->ni_vp->v_object;
832 
833 	hdr = (const Elf_Ehdr *)imgp->image_header;
834 	if ((error = __elfN(check_header)(hdr)) != 0)
835 		goto fail;
836 	if (hdr->e_type == ET_DYN)
837 		rbase = *addr;
838 	else if (hdr->e_type == ET_EXEC)
839 		rbase = 0;
840 	else {
841 		error = ENOEXEC;
842 		goto fail;
843 	}
844 
845 	/* Only support headers that fit within first page for now      */
846 	if (!__elfN(phdr_in_zero_page)(hdr)) {
847 		error = ENOEXEC;
848 		goto fail;
849 	}
850 
851 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
852 	if (!aligned(phdr, Elf_Addr)) {
853 		error = ENOEXEC;
854 		goto fail;
855 	}
856 
857 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
858 	if (error != 0)
859 		goto fail;
860 
861 	*addr = base_addr;
862 	*entry = (unsigned long)hdr->e_entry + rbase;
863 
864 fail:
865 	if (imgp->firstpage)
866 		exec_unmap_first_page(imgp);
867 
868 	if (nd->ni_vp) {
869 		if (imgp->textset)
870 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
871 		vput(nd->ni_vp);
872 	}
873 	free(tempdata, M_TEMP);
874 
875 	return (error);
876 }
877 
878 /*
879  * Select randomized valid address in the map map, between minv and
880  * maxv, with specified alignment.  The [minv, maxv) range must belong
881  * to the map.  Note that function only allocates the address, it is
882  * up to caller to clamp maxv in a way that the final allocation
883  * length fit into the map.
884  *
885  * Result is returned in *resp, error code indicates that arguments
886  * did not pass sanity checks for overflow and range correctness.
887  */
888 static int
889 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
890     u_int align, u_long *resp)
891 {
892 	u_long rbase, res;
893 
894 	MPASS(vm_map_min(map) <= minv);
895 
896 	if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
897 		uprintf("Invalid ELF segments layout\n");
898 		return (ENOEXEC);
899 	}
900 
901 	arc4rand(&rbase, sizeof(rbase), 0);
902 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
903 	res &= ~((u_long)align - 1);
904 	if (res >= maxv)
905 		res -= align;
906 
907 	KASSERT(res >= minv,
908 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
909 	    res, minv, maxv, rbase));
910 	KASSERT(res < maxv,
911 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
912 	    res, maxv, minv, rbase));
913 
914 	*resp = res;
915 	return (0);
916 }
917 
918 static int
919 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
920     const Elf_Phdr *phdr, u_long et_dyn_addr)
921 {
922 	struct vmspace *vmspace;
923 	const char *err_str;
924 	u_long text_size, data_size, total_size, text_addr, data_addr;
925 	u_long seg_size, seg_addr;
926 	int i;
927 
928 	err_str = NULL;
929 	text_size = data_size = total_size = text_addr = data_addr = 0;
930 
931 	for (i = 0; i < hdr->e_phnum; i++) {
932 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
933 			continue;
934 
935 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
936 		seg_size = round_page(phdr[i].p_memsz +
937 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
938 
939 		/*
940 		 * Make the largest executable segment the official
941 		 * text segment and all others data.
942 		 *
943 		 * Note that obreak() assumes that data_addr + data_size == end
944 		 * of data load area, and the ELF file format expects segments
945 		 * to be sorted by address.  If multiple data segments exist,
946 		 * the last one will be used.
947 		 */
948 
949 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
950 			text_size = seg_size;
951 			text_addr = seg_addr;
952 		} else {
953 			data_size = seg_size;
954 			data_addr = seg_addr;
955 		}
956 		total_size += seg_size;
957 	}
958 
959 	if (data_addr == 0 && data_size == 0) {
960 		data_addr = text_addr;
961 		data_size = text_size;
962 	}
963 
964 	/*
965 	 * Check limits.  It should be safe to check the
966 	 * limits after loading the segments since we do
967 	 * not actually fault in all the segments pages.
968 	 */
969 	PROC_LOCK(imgp->proc);
970 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
971 		err_str = "Data segment size exceeds process limit";
972 	else if (text_size > maxtsiz)
973 		err_str = "Text segment size exceeds system limit";
974 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
975 		err_str = "Total segment size exceeds process limit";
976 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
977 		err_str = "Data segment size exceeds resource limit";
978 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
979 		err_str = "Total segment size exceeds resource limit";
980 	PROC_UNLOCK(imgp->proc);
981 	if (err_str != NULL) {
982 		uprintf("%s\n", err_str);
983 		return (ENOMEM);
984 	}
985 
986 	vmspace = imgp->proc->p_vmspace;
987 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
988 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
989 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
990 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
991 
992 	return (0);
993 }
994 
995 static int
996 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
997     char **interpp, bool *free_interpp)
998 {
999 	struct thread *td;
1000 	char *interp;
1001 	int error, interp_name_len;
1002 
1003 	KASSERT(phdr->p_type == PT_INTERP,
1004 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
1005 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
1006 
1007 	td = curthread;
1008 
1009 	/* Path to interpreter */
1010 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
1011 		uprintf("Invalid PT_INTERP\n");
1012 		return (ENOEXEC);
1013 	}
1014 
1015 	interp_name_len = phdr->p_filesz;
1016 	if (phdr->p_offset > PAGE_SIZE ||
1017 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
1018 		/*
1019 		 * The vnode lock might be needed by the pagedaemon to
1020 		 * clean pages owned by the vnode.  Do not allow sleep
1021 		 * waiting for memory with the vnode locked, instead
1022 		 * try non-sleepable allocation first, and if it
1023 		 * fails, go to the slow path were we drop the lock
1024 		 * and do M_WAITOK.  A text reference prevents
1025 		 * modifications to the vnode content.
1026 		 */
1027 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1028 		if (interp == NULL) {
1029 			VOP_UNLOCK(imgp->vp);
1030 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1031 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1032 		}
1033 
1034 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
1035 		    interp_name_len, phdr->p_offset,
1036 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1037 		    NOCRED, NULL, td);
1038 		if (error != 0) {
1039 			free(interp, M_TEMP);
1040 			uprintf("i/o error PT_INTERP %d\n", error);
1041 			return (error);
1042 		}
1043 		interp[interp_name_len] = '\0';
1044 
1045 		*interpp = interp;
1046 		*free_interpp = true;
1047 		return (0);
1048 	}
1049 
1050 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1051 	if (interp[interp_name_len - 1] != '\0') {
1052 		uprintf("Invalid PT_INTERP\n");
1053 		return (ENOEXEC);
1054 	}
1055 
1056 	*interpp = interp;
1057 	*free_interpp = false;
1058 	return (0);
1059 }
1060 
1061 static int
1062 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1063     const char *interp, u_long *addr, u_long *entry)
1064 {
1065 	char *path;
1066 	int error;
1067 
1068 	if (brand_info->emul_path != NULL &&
1069 	    brand_info->emul_path[0] != '\0') {
1070 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1071 		snprintf(path, MAXPATHLEN, "%s%s",
1072 		    brand_info->emul_path, interp);
1073 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1074 		free(path, M_TEMP);
1075 		if (error == 0)
1076 			return (0);
1077 	}
1078 
1079 	if (brand_info->interp_newpath != NULL &&
1080 	    (brand_info->interp_path == NULL ||
1081 	    strcmp(interp, brand_info->interp_path) == 0)) {
1082 		error = __elfN(load_file)(imgp->proc,
1083 		    brand_info->interp_newpath, addr, entry);
1084 		if (error == 0)
1085 			return (0);
1086 	}
1087 
1088 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1089 	if (error == 0)
1090 		return (0);
1091 
1092 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1093 	return (error);
1094 }
1095 
1096 /*
1097  * Impossible et_dyn_addr initial value indicating that the real base
1098  * must be calculated later with some randomization applied.
1099  */
1100 #define	ET_DYN_ADDR_RAND	1
1101 
1102 static int
1103 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1104 {
1105 	struct thread *td;
1106 	const Elf_Ehdr *hdr;
1107 	const Elf_Phdr *phdr;
1108 	Elf_Auxargs *elf_auxargs;
1109 	struct vmspace *vmspace;
1110 	vm_map_t map;
1111 	char *interp;
1112 	Elf_Brandinfo *brand_info;
1113 	struct sysentvec *sv;
1114 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1115 	u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1116 	uint32_t fctl0;
1117 	int32_t osrel;
1118 	bool free_interp;
1119 	int error, i, n;
1120 
1121 	hdr = (const Elf_Ehdr *)imgp->image_header;
1122 
1123 	/*
1124 	 * Do we have a valid ELF header ?
1125 	 *
1126 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1127 	 * if particular brand doesn't support it.
1128 	 */
1129 	if (__elfN(check_header)(hdr) != 0 ||
1130 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1131 		return (-1);
1132 
1133 	/*
1134 	 * From here on down, we return an errno, not -1, as we've
1135 	 * detected an ELF file.
1136 	 */
1137 
1138 	if (!__elfN(phdr_in_zero_page)(hdr)) {
1139 		uprintf("Program headers not in the first page\n");
1140 		return (ENOEXEC);
1141 	}
1142 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1143 	if (!aligned(phdr, Elf_Addr)) {
1144 		uprintf("Unaligned program headers\n");
1145 		return (ENOEXEC);
1146 	}
1147 
1148 	n = error = 0;
1149 	baddr = 0;
1150 	osrel = 0;
1151 	fctl0 = 0;
1152 	entry = proghdr = 0;
1153 	interp = NULL;
1154 	free_interp = false;
1155 	td = curthread;
1156 
1157 	/*
1158 	 * Somewhat arbitrary, limit accepted max alignment for the
1159 	 * loadable segment to the max supported superpage size. Too
1160 	 * large alignment requests are not useful and are indicators
1161 	 * of corrupted or outright malicious binary.
1162 	 */
1163 	maxalign = PAGE_SIZE;
1164 	maxsalign = PAGE_SIZE * 1024;
1165 	for (i = MAXPAGESIZES - 1; i > 0; i--) {
1166 		if (pagesizes[i] > maxsalign)
1167 			maxsalign = pagesizes[i];
1168 	}
1169 
1170 	mapsz = 0;
1171 
1172 	for (i = 0; i < hdr->e_phnum; i++) {
1173 		switch (phdr[i].p_type) {
1174 		case PT_LOAD:
1175 			if (n == 0)
1176 				baddr = phdr[i].p_vaddr;
1177 			if (!powerof2(phdr[i].p_align) ||
1178 			    phdr[i].p_align > maxsalign) {
1179 				uprintf("Invalid segment alignment\n");
1180 				error = ENOEXEC;
1181 				goto ret;
1182 			}
1183 			if (phdr[i].p_align > maxalign)
1184 				maxalign = phdr[i].p_align;
1185 			if (mapsz + phdr[i].p_memsz < mapsz) {
1186 				uprintf("Mapsize overflow\n");
1187 				error = ENOEXEC;
1188 				goto ret;
1189 			}
1190 			mapsz += phdr[i].p_memsz;
1191 			n++;
1192 
1193 			/*
1194 			 * If this segment contains the program headers,
1195 			 * remember their virtual address for the AT_PHDR
1196 			 * aux entry. Static binaries don't usually include
1197 			 * a PT_PHDR entry.
1198 			 */
1199 			if (phdr[i].p_offset == 0 &&
1200 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1201 			    phdr[i].p_filesz)
1202 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1203 			break;
1204 		case PT_INTERP:
1205 			/* Path to interpreter */
1206 			if (interp != NULL) {
1207 				uprintf("Multiple PT_INTERP headers\n");
1208 				error = ENOEXEC;
1209 				goto ret;
1210 			}
1211 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1212 			    &free_interp);
1213 			if (error != 0)
1214 				goto ret;
1215 			break;
1216 		case PT_GNU_STACK:
1217 			if (__elfN(nxstack))
1218 				imgp->stack_prot =
1219 				    __elfN(trans_prot)(phdr[i].p_flags);
1220 			imgp->stack_sz = phdr[i].p_memsz;
1221 			break;
1222 		case PT_PHDR: 	/* Program header table info */
1223 			proghdr = phdr[i].p_vaddr;
1224 			break;
1225 		}
1226 	}
1227 
1228 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1229 	if (brand_info == NULL) {
1230 		uprintf("ELF binary type \"%u\" not known.\n",
1231 		    hdr->e_ident[EI_OSABI]);
1232 		error = ENOEXEC;
1233 		goto ret;
1234 	}
1235 	sv = brand_info->sysvec;
1236 	et_dyn_addr = 0;
1237 	if (hdr->e_type == ET_DYN) {
1238 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1239 			uprintf("Cannot execute shared object\n");
1240 			error = ENOEXEC;
1241 			goto ret;
1242 		}
1243 		/*
1244 		 * Honour the base load address from the dso if it is
1245 		 * non-zero for some reason.
1246 		 */
1247 		if (baddr == 0) {
1248 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1249 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1250 				et_dyn_addr = __elfN(pie_base);
1251 			else if ((__elfN(pie_aslr_enabled) &&
1252 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1253 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1254 				et_dyn_addr = ET_DYN_ADDR_RAND;
1255 			else
1256 				et_dyn_addr = __elfN(pie_base);
1257 		}
1258 	}
1259 
1260 	/*
1261 	 * Avoid a possible deadlock if the current address space is destroyed
1262 	 * and that address space maps the locked vnode.  In the common case,
1263 	 * the locked vnode's v_usecount is decremented but remains greater
1264 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1265 	 * However, in cases where the vnode lock is external, such as nullfs,
1266 	 * v_usecount may become zero.
1267 	 *
1268 	 * The VV_TEXT flag prevents modifications to the executable while
1269 	 * the vnode is unlocked.
1270 	 */
1271 	VOP_UNLOCK(imgp->vp);
1272 
1273 	/*
1274 	 * Decide whether to enable randomization of user mappings.
1275 	 * First, reset user preferences for the setid binaries.
1276 	 * Then, account for the support of the randomization by the
1277 	 * ABI, by user preferences, and make special treatment for
1278 	 * PIE binaries.
1279 	 */
1280 	if (imgp->credential_setid) {
1281 		PROC_LOCK(imgp->proc);
1282 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1283 		    P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1284 		PROC_UNLOCK(imgp->proc);
1285 	}
1286 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1287 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1288 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1289 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1290 		    ("et_dyn_addr == RAND and !ASLR"));
1291 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1292 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1293 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1294 		imgp->map_flags |= MAP_ASLR;
1295 		/*
1296 		 * If user does not care about sbrk, utilize the bss
1297 		 * grow region for mappings as well.  We can select
1298 		 * the base for the image anywere and still not suffer
1299 		 * from the fragmentation.
1300 		 */
1301 		if (!__elfN(aslr_honor_sbrk) ||
1302 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1303 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1304 		if (__elfN(aslr_stack))
1305 			imgp->map_flags |= MAP_ASLR_STACK;
1306 	}
1307 
1308 	if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1309 	    (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1310 	    (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1311 		imgp->map_flags |= MAP_WXORX;
1312 
1313 	error = exec_new_vmspace(imgp, sv);
1314 
1315 	imgp->proc->p_sysent = sv;
1316 	imgp->proc->p_elf_brandinfo = brand_info;
1317 
1318 	vmspace = imgp->proc->p_vmspace;
1319 	map = &vmspace->vm_map;
1320 	maxv = sv->sv_usrstack;
1321 	if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1322 		maxv -= lim_max(td, RLIMIT_STACK);
1323 	if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1324 		uprintf("Excessive mapping size\n");
1325 		error = ENOEXEC;
1326 	}
1327 
1328 	if (error == 0 && et_dyn_addr == ET_DYN_ADDR_RAND) {
1329 		KASSERT((map->flags & MAP_ASLR) != 0,
1330 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1331 		error = __CONCAT(rnd_, __elfN(base))(map,
1332 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1333 		    /* reserve half of the address space to interpreter */
1334 		    maxv / 2, maxalign, &et_dyn_addr);
1335 	}
1336 
1337 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1338 	if (error != 0)
1339 		goto ret;
1340 
1341 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1342 	if (error != 0)
1343 		goto ret;
1344 
1345 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1346 	if (error != 0)
1347 		goto ret;
1348 
1349 	/*
1350 	 * We load the dynamic linker where a userland call
1351 	 * to mmap(0, ...) would put it.  The rationale behind this
1352 	 * calculation is that it leaves room for the heap to grow to
1353 	 * its maximum allowed size.
1354 	 */
1355 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1356 	    RLIMIT_DATA));
1357 	if ((map->flags & MAP_ASLR) != 0) {
1358 		maxv1 = maxv / 2 + addr / 2;
1359 		error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1360 		    (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1361 		    pagesizes[1] : pagesizes[0], &anon_loc);
1362 		if (error != 0)
1363 			goto ret;
1364 		map->anon_loc = anon_loc;
1365 	} else {
1366 		map->anon_loc = addr;
1367 	}
1368 
1369 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1370 	imgp->entry_addr = entry;
1371 
1372 	if (interp != NULL) {
1373 		VOP_UNLOCK(imgp->vp);
1374 		if ((map->flags & MAP_ASLR) != 0) {
1375 			/* Assume that interpreter fits into 1/4 of AS */
1376 			maxv1 = maxv / 2 + addr / 2;
1377 			error = __CONCAT(rnd_, __elfN(base))(map, addr,
1378 			    maxv1, PAGE_SIZE, &addr);
1379 		}
1380 		if (error == 0) {
1381 			error = __elfN(load_interp)(imgp, brand_info, interp,
1382 			    &addr, &imgp->entry_addr);
1383 		}
1384 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1385 		if (error != 0)
1386 			goto ret;
1387 	} else
1388 		addr = et_dyn_addr;
1389 
1390 	error = exec_map_stack(imgp);
1391 	if (error != 0)
1392 		goto ret;
1393 
1394 	/*
1395 	 * Construct auxargs table (used by the copyout_auxargs routine)
1396 	 */
1397 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1398 	if (elf_auxargs == NULL) {
1399 		VOP_UNLOCK(imgp->vp);
1400 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1401 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1402 	}
1403 	elf_auxargs->execfd = -1;
1404 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1405 	elf_auxargs->phent = hdr->e_phentsize;
1406 	elf_auxargs->phnum = hdr->e_phnum;
1407 	elf_auxargs->pagesz = PAGE_SIZE;
1408 	elf_auxargs->base = addr;
1409 	elf_auxargs->flags = 0;
1410 	elf_auxargs->entry = entry;
1411 	elf_auxargs->hdr_eflags = hdr->e_flags;
1412 
1413 	imgp->auxargs = elf_auxargs;
1414 	imgp->interpreted = 0;
1415 	imgp->reloc_base = addr;
1416 	imgp->proc->p_osrel = osrel;
1417 	imgp->proc->p_fctl0 = fctl0;
1418 	imgp->proc->p_elf_flags = hdr->e_flags;
1419 
1420 ret:
1421 	ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1422 	if (free_interp)
1423 		free(interp, M_TEMP);
1424 	return (error);
1425 }
1426 
1427 #define	elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1428 
1429 int
1430 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1431 {
1432 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1433 	Elf_Auxinfo *argarray, *pos;
1434 	int error;
1435 
1436 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1437 	    M_WAITOK | M_ZERO);
1438 
1439 	if (args->execfd != -1)
1440 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1441 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1442 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1443 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1444 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1445 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1446 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1447 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1448 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1449 	if (imgp->execpathp != 0)
1450 		AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1451 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1452 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1453 	if (imgp->canary != 0) {
1454 		AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1455 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1456 	}
1457 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1458 	if (imgp->pagesizes != 0) {
1459 		AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1460 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1461 	}
1462 	if (imgp->sysent->sv_timekeep_base != 0) {
1463 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1464 		    imgp->sysent->sv_timekeep_base);
1465 	}
1466 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1467 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1468 	    imgp->sysent->sv_stackprot);
1469 	if (imgp->sysent->sv_hwcap != NULL)
1470 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1471 	if (imgp->sysent->sv_hwcap2 != NULL)
1472 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1473 	AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1474 	    ELF_BSDF_SIGFASTBLK : 0);
1475 	AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1476 	AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1477 	AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1478 	AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1479 	AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1480 	if (imgp->sysent->sv_fxrng_gen_base != 0)
1481 		AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1482 	if (imgp->sysent->sv_vdso_base != 0 && __elfN(vdso) != 0)
1483 		AUXARGS_ENTRY(pos, AT_KPRELOAD, imgp->sysent->sv_vdso_base);
1484 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1485 
1486 	free(imgp->auxargs, M_TEMP);
1487 	imgp->auxargs = NULL;
1488 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1489 
1490 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1491 	free(argarray, M_TEMP);
1492 	return (error);
1493 }
1494 
1495 int
1496 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1497 {
1498 	Elf_Addr *base;
1499 
1500 	base = (Elf_Addr *)*stack_base;
1501 	base--;
1502 	if (elf_suword(base, imgp->args->argc) == -1)
1503 		return (EFAULT);
1504 	*stack_base = (uintptr_t)base;
1505 	return (0);
1506 }
1507 
1508 /*
1509  * Code for generating ELF core dumps.
1510  */
1511 
1512 typedef void (*segment_callback)(vm_map_entry_t, void *);
1513 
1514 /* Closure for cb_put_phdr(). */
1515 struct phdr_closure {
1516 	Elf_Phdr *phdr;		/* Program header to fill in */
1517 	Elf_Off offset;		/* Offset of segment in core file */
1518 };
1519 
1520 struct note_info {
1521 	int		type;		/* Note type. */
1522 	outfunc_t 	outfunc; 	/* Output function. */
1523 	void		*outarg;	/* Argument for the output function. */
1524 	size_t		outsize;	/* Output size. */
1525 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1526 };
1527 
1528 TAILQ_HEAD(note_info_list, note_info);
1529 
1530 extern int compress_user_cores;
1531 extern int compress_user_cores_level;
1532 
1533 static void cb_put_phdr(vm_map_entry_t, void *);
1534 static void cb_size_segment(vm_map_entry_t, void *);
1535 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1536     int);
1537 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1538     struct note_info_list *, size_t, int);
1539 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1540 
1541 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1542 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1543 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1544 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1545 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1546 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1547 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1548 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1549 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1550 static void note_procstat_files(void *, struct sbuf *, size_t *);
1551 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1552 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1553 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1554 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1555 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1556 
1557 static int
1558 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1559 {
1560 
1561 	return (core_write((struct coredump_params *)arg, base, len, offset,
1562 	    UIO_SYSSPACE, NULL));
1563 }
1564 
1565 int
1566 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1567 {
1568 	struct ucred *cred = td->td_ucred;
1569 	int compm, error = 0;
1570 	struct sseg_closure seginfo;
1571 	struct note_info_list notelst;
1572 	struct coredump_params params;
1573 	struct note_info *ninfo;
1574 	void *hdr, *tmpbuf;
1575 	size_t hdrsize, notesz, coresize;
1576 
1577 	hdr = NULL;
1578 	tmpbuf = NULL;
1579 	TAILQ_INIT(&notelst);
1580 
1581 	/* Size the program segments. */
1582 	__elfN(size_segments)(td, &seginfo, flags);
1583 
1584 	/*
1585 	 * Collect info about the core file header area.
1586 	 */
1587 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1588 	if (seginfo.count + 1 >= PN_XNUM)
1589 		hdrsize += sizeof(Elf_Shdr);
1590 	td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, &notelst, &notesz);
1591 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1592 
1593 	/* Set up core dump parameters. */
1594 	params.offset = 0;
1595 	params.active_cred = cred;
1596 	params.file_cred = NOCRED;
1597 	params.td = td;
1598 	params.vp = vp;
1599 	params.comp = NULL;
1600 
1601 #ifdef RACCT
1602 	if (racct_enable) {
1603 		PROC_LOCK(td->td_proc);
1604 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1605 		PROC_UNLOCK(td->td_proc);
1606 		if (error != 0) {
1607 			error = EFAULT;
1608 			goto done;
1609 		}
1610 	}
1611 #endif
1612 	if (coresize >= limit) {
1613 		error = EFAULT;
1614 		goto done;
1615 	}
1616 
1617 	/* Create a compression stream if necessary. */
1618 	compm = compress_user_cores;
1619 	if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1620 	    compm == 0)
1621 		compm = COMPRESS_GZIP;
1622 	if (compm != 0) {
1623 		params.comp = compressor_init(core_compressed_write,
1624 		    compm, CORE_BUF_SIZE,
1625 		    compress_user_cores_level, &params);
1626 		if (params.comp == NULL) {
1627 			error = EFAULT;
1628 			goto done;
1629 		}
1630 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1631         }
1632 
1633 	/*
1634 	 * Allocate memory for building the header, fill it up,
1635 	 * and write it out following the notes.
1636 	 */
1637 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1638 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1639 	    notesz, flags);
1640 
1641 	/* Write the contents of all of the writable segments. */
1642 	if (error == 0) {
1643 		Elf_Phdr *php;
1644 		off_t offset;
1645 		int i;
1646 
1647 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1648 		offset = round_page(hdrsize + notesz);
1649 		for (i = 0; i < seginfo.count; i++) {
1650 			error = core_output((char *)(uintptr_t)php->p_vaddr,
1651 			    php->p_filesz, offset, &params, tmpbuf);
1652 			if (error != 0)
1653 				break;
1654 			offset += php->p_filesz;
1655 			php++;
1656 		}
1657 		if (error == 0 && params.comp != NULL)
1658 			error = compressor_flush(params.comp);
1659 	}
1660 	if (error) {
1661 		log(LOG_WARNING,
1662 		    "Failed to write core file for process %s (error %d)\n",
1663 		    curproc->p_comm, error);
1664 	}
1665 
1666 done:
1667 	free(tmpbuf, M_TEMP);
1668 	if (params.comp != NULL)
1669 		compressor_fini(params.comp);
1670 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1671 		TAILQ_REMOVE(&notelst, ninfo, link);
1672 		free(ninfo, M_TEMP);
1673 	}
1674 	if (hdr != NULL)
1675 		free(hdr, M_TEMP);
1676 
1677 	return (error);
1678 }
1679 
1680 /*
1681  * A callback for each_dumpable_segment() to write out the segment's
1682  * program header entry.
1683  */
1684 static void
1685 cb_put_phdr(vm_map_entry_t entry, void *closure)
1686 {
1687 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1688 	Elf_Phdr *phdr = phc->phdr;
1689 
1690 	phc->offset = round_page(phc->offset);
1691 
1692 	phdr->p_type = PT_LOAD;
1693 	phdr->p_offset = phc->offset;
1694 	phdr->p_vaddr = entry->start;
1695 	phdr->p_paddr = 0;
1696 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1697 	phdr->p_align = PAGE_SIZE;
1698 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1699 
1700 	phc->offset += phdr->p_filesz;
1701 	phc->phdr++;
1702 }
1703 
1704 /*
1705  * A callback for each_dumpable_segment() to gather information about
1706  * the number of segments and their total size.
1707  */
1708 static void
1709 cb_size_segment(vm_map_entry_t entry, void *closure)
1710 {
1711 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1712 
1713 	ssc->count++;
1714 	ssc->size += entry->end - entry->start;
1715 }
1716 
1717 void
1718 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1719     int flags)
1720 {
1721 	seginfo->count = 0;
1722 	seginfo->size = 0;
1723 
1724 	each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1725 }
1726 
1727 /*
1728  * For each writable segment in the process's memory map, call the given
1729  * function with a pointer to the map entry and some arbitrary
1730  * caller-supplied data.
1731  */
1732 static void
1733 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1734     int flags)
1735 {
1736 	struct proc *p = td->td_proc;
1737 	vm_map_t map = &p->p_vmspace->vm_map;
1738 	vm_map_entry_t entry;
1739 	vm_object_t backing_object, object;
1740 	bool ignore_entry;
1741 
1742 	vm_map_lock_read(map);
1743 	VM_MAP_ENTRY_FOREACH(entry, map) {
1744 		/*
1745 		 * Don't dump inaccessible mappings, deal with legacy
1746 		 * coredump mode.
1747 		 *
1748 		 * Note that read-only segments related to the elf binary
1749 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1750 		 * need to arbitrarily ignore such segments.
1751 		 */
1752 		if ((flags & SVC_ALL) == 0) {
1753 			if (elf_legacy_coredump) {
1754 				if ((entry->protection & VM_PROT_RW) !=
1755 				    VM_PROT_RW)
1756 					continue;
1757 			} else {
1758 				if ((entry->protection & VM_PROT_ALL) == 0)
1759 					continue;
1760 			}
1761 		}
1762 
1763 		/*
1764 		 * Dont include memory segment in the coredump if
1765 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1766 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1767 		 * kernel map).
1768 		 */
1769 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1770 			continue;
1771 		if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1772 		    (flags & SVC_ALL) == 0)
1773 			continue;
1774 		if ((object = entry->object.vm_object) == NULL)
1775 			continue;
1776 
1777 		/* Ignore memory-mapped devices and such things. */
1778 		VM_OBJECT_RLOCK(object);
1779 		while ((backing_object = object->backing_object) != NULL) {
1780 			VM_OBJECT_RLOCK(backing_object);
1781 			VM_OBJECT_RUNLOCK(object);
1782 			object = backing_object;
1783 		}
1784 		ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1785 		VM_OBJECT_RUNLOCK(object);
1786 		if (ignore_entry)
1787 			continue;
1788 
1789 		(*func)(entry, closure);
1790 	}
1791 	vm_map_unlock_read(map);
1792 }
1793 
1794 /*
1795  * Write the core file header to the file, including padding up to
1796  * the page boundary.
1797  */
1798 static int
1799 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1800     size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1801     int flags)
1802 {
1803 	struct note_info *ninfo;
1804 	struct sbuf *sb;
1805 	int error;
1806 
1807 	/* Fill in the header. */
1808 	bzero(hdr, hdrsize);
1809 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1810 
1811 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1812 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1813 	sbuf_start_section(sb, NULL);
1814 	sbuf_bcat(sb, hdr, hdrsize);
1815 	TAILQ_FOREACH(ninfo, notelst, link)
1816 	    __elfN(putnote)(p->td, ninfo, sb);
1817 	/* Align up to a page boundary for the program segments. */
1818 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1819 	error = sbuf_finish(sb);
1820 	sbuf_delete(sb);
1821 
1822 	return (error);
1823 }
1824 
1825 void
1826 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1827     size_t *sizep)
1828 {
1829 	struct proc *p;
1830 	struct thread *thr;
1831 	size_t size;
1832 
1833 	p = td->td_proc;
1834 	size = 0;
1835 
1836 	size += __elfN(register_note)(td, list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1837 
1838 	/*
1839 	 * To have the debugger select the right thread (LWP) as the initial
1840 	 * thread, we dump the state of the thread passed to us in td first.
1841 	 * This is the thread that causes the core dump and thus likely to
1842 	 * be the right thread one wants to have selected in the debugger.
1843 	 */
1844 	thr = td;
1845 	while (thr != NULL) {
1846 		size += __elfN(register_note)(td, list, NT_PRSTATUS,
1847 		    __elfN(note_prstatus), thr);
1848 		size += __elfN(register_note)(td, list, NT_FPREGSET,
1849 		    __elfN(note_fpregset), thr);
1850 		size += __elfN(register_note)(td, list, NT_THRMISC,
1851 		    __elfN(note_thrmisc), thr);
1852 		size += __elfN(register_note)(td, list, NT_PTLWPINFO,
1853 		    __elfN(note_ptlwpinfo), thr);
1854 		size += __elfN(register_note)(td, list, -1,
1855 		    __elfN(note_threadmd), thr);
1856 
1857 		thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1858 		    TAILQ_NEXT(thr, td_plist);
1859 		if (thr == td)
1860 			thr = TAILQ_NEXT(thr, td_plist);
1861 	}
1862 
1863 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1864 	    __elfN(note_procstat_proc), p);
1865 	size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1866 	    note_procstat_files, p);
1867 	size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1868 	    note_procstat_vmmap, p);
1869 	size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1870 	    note_procstat_groups, p);
1871 	size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1872 	    note_procstat_umask, p);
1873 	size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1874 	    note_procstat_rlimit, p);
1875 	size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1876 	    note_procstat_osrel, p);
1877 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1878 	    __elfN(note_procstat_psstrings), p);
1879 	size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1880 	    __elfN(note_procstat_auxv), p);
1881 
1882 	*sizep = size;
1883 }
1884 
1885 void
1886 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1887     size_t notesz, int flags)
1888 {
1889 	Elf_Ehdr *ehdr;
1890 	Elf_Phdr *phdr;
1891 	Elf_Shdr *shdr;
1892 	struct phdr_closure phc;
1893 	Elf_Brandinfo *bi;
1894 
1895 	ehdr = (Elf_Ehdr *)hdr;
1896 	bi = td->td_proc->p_elf_brandinfo;
1897 
1898 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1899 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1900 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1901 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1902 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1903 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1904 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1905 	ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1906 	ehdr->e_ident[EI_ABIVERSION] = 0;
1907 	ehdr->e_ident[EI_PAD] = 0;
1908 	ehdr->e_type = ET_CORE;
1909 	ehdr->e_machine = bi->machine;
1910 	ehdr->e_version = EV_CURRENT;
1911 	ehdr->e_entry = 0;
1912 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1913 	ehdr->e_flags = td->td_proc->p_elf_flags;
1914 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1915 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1916 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1917 	ehdr->e_shstrndx = SHN_UNDEF;
1918 	if (numsegs + 1 < PN_XNUM) {
1919 		ehdr->e_phnum = numsegs + 1;
1920 		ehdr->e_shnum = 0;
1921 	} else {
1922 		ehdr->e_phnum = PN_XNUM;
1923 		ehdr->e_shnum = 1;
1924 
1925 		ehdr->e_shoff = ehdr->e_phoff +
1926 		    (numsegs + 1) * ehdr->e_phentsize;
1927 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1928 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1929 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1930 
1931 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1932 		memset(shdr, 0, sizeof(*shdr));
1933 		/*
1934 		 * A special first section is used to hold large segment and
1935 		 * section counts.  This was proposed by Sun Microsystems in
1936 		 * Solaris and has been adopted by Linux; the standard ELF
1937 		 * tools are already familiar with the technique.
1938 		 *
1939 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1940 		 * (or 12-7 depending on the version of the document) for more
1941 		 * details.
1942 		 */
1943 		shdr->sh_type = SHT_NULL;
1944 		shdr->sh_size = ehdr->e_shnum;
1945 		shdr->sh_link = ehdr->e_shstrndx;
1946 		shdr->sh_info = numsegs + 1;
1947 	}
1948 
1949 	/*
1950 	 * Fill in the program header entries.
1951 	 */
1952 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1953 
1954 	/* The note segement. */
1955 	phdr->p_type = PT_NOTE;
1956 	phdr->p_offset = hdrsize;
1957 	phdr->p_vaddr = 0;
1958 	phdr->p_paddr = 0;
1959 	phdr->p_filesz = notesz;
1960 	phdr->p_memsz = 0;
1961 	phdr->p_flags = PF_R;
1962 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1963 	phdr++;
1964 
1965 	/* All the writable segments from the program. */
1966 	phc.phdr = phdr;
1967 	phc.offset = round_page(hdrsize + notesz);
1968 	each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1969 }
1970 
1971 size_t
1972 __elfN(register_note)(struct thread *td, struct note_info_list *list,
1973     int type, outfunc_t out, void *arg)
1974 {
1975 	const struct sysentvec *sv;
1976 	struct note_info *ninfo;
1977 	size_t size, notesize;
1978 
1979 	sv = td->td_proc->p_sysent;
1980 	size = 0;
1981 	out(arg, NULL, &size);
1982 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1983 	ninfo->type = type;
1984 	ninfo->outfunc = out;
1985 	ninfo->outarg = arg;
1986 	ninfo->outsize = size;
1987 	TAILQ_INSERT_TAIL(list, ninfo, link);
1988 
1989 	if (type == -1)
1990 		return (size);
1991 
1992 	notesize = sizeof(Elf_Note) +		/* note header */
1993 	    roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1994 						/* note name */
1995 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1996 
1997 	return (notesize);
1998 }
1999 
2000 static size_t
2001 append_note_data(const void *src, void *dst, size_t len)
2002 {
2003 	size_t padded_len;
2004 
2005 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2006 	if (dst != NULL) {
2007 		bcopy(src, dst, len);
2008 		bzero((char *)dst + len, padded_len - len);
2009 	}
2010 	return (padded_len);
2011 }
2012 
2013 size_t
2014 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2015 {
2016 	Elf_Note *note;
2017 	char *buf;
2018 	size_t notesize;
2019 
2020 	buf = dst;
2021 	if (buf != NULL) {
2022 		note = (Elf_Note *)buf;
2023 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2024 		note->n_descsz = size;
2025 		note->n_type = type;
2026 		buf += sizeof(*note);
2027 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2028 		    sizeof(FREEBSD_ABI_VENDOR));
2029 		append_note_data(src, buf, size);
2030 		if (descp != NULL)
2031 			*descp = buf;
2032 	}
2033 
2034 	notesize = sizeof(Elf_Note) +		/* note header */
2035 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2036 						/* note name */
2037 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2038 
2039 	return (notesize);
2040 }
2041 
2042 static void
2043 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
2044 {
2045 	Elf_Note note;
2046 	const struct sysentvec *sv;
2047 	ssize_t old_len, sect_len;
2048 	size_t new_len, descsz, i;
2049 
2050 	if (ninfo->type == -1) {
2051 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2052 		return;
2053 	}
2054 
2055 	sv = td->td_proc->p_sysent;
2056 
2057 	note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2058 	note.n_descsz = ninfo->outsize;
2059 	note.n_type = ninfo->type;
2060 
2061 	sbuf_bcat(sb, &note, sizeof(note));
2062 	sbuf_start_section(sb, &old_len);
2063 	sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2064 	    strlen(sv->sv_elf_core_abi_vendor) + 1);
2065 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2066 	if (note.n_descsz == 0)
2067 		return;
2068 	sbuf_start_section(sb, &old_len);
2069 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2070 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2071 	if (sect_len < 0)
2072 		return;
2073 
2074 	new_len = (size_t)sect_len;
2075 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2076 	if (new_len < descsz) {
2077 		/*
2078 		 * It is expected that individual note emitters will correctly
2079 		 * predict their expected output size and fill up to that size
2080 		 * themselves, padding in a format-specific way if needed.
2081 		 * However, in case they don't, just do it here with zeros.
2082 		 */
2083 		for (i = 0; i < descsz - new_len; i++)
2084 			sbuf_putc(sb, 0);
2085 	} else if (new_len > descsz) {
2086 		/*
2087 		 * We can't always truncate sb -- we may have drained some
2088 		 * of it already.
2089 		 */
2090 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2091 		    "read it (%zu > %zu).  Since it is longer than "
2092 		    "expected, this coredump's notes are corrupt.  THIS "
2093 		    "IS A BUG in the note_procstat routine for type %u.\n",
2094 		    __func__, (unsigned)note.n_type, new_len, descsz,
2095 		    (unsigned)note.n_type));
2096 	}
2097 }
2098 
2099 /*
2100  * Miscellaneous note out functions.
2101  */
2102 
2103 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2104 #include <compat/freebsd32/freebsd32.h>
2105 #include <compat/freebsd32/freebsd32_signal.h>
2106 
2107 typedef struct prstatus32 elf_prstatus_t;
2108 typedef struct prpsinfo32 elf_prpsinfo_t;
2109 typedef struct fpreg32 elf_prfpregset_t;
2110 typedef struct fpreg32 elf_fpregset_t;
2111 typedef struct reg32 elf_gregset_t;
2112 typedef struct thrmisc32 elf_thrmisc_t;
2113 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2114 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2115 typedef uint32_t elf_ps_strings_t;
2116 #else
2117 typedef prstatus_t elf_prstatus_t;
2118 typedef prpsinfo_t elf_prpsinfo_t;
2119 typedef prfpregset_t elf_prfpregset_t;
2120 typedef prfpregset_t elf_fpregset_t;
2121 typedef gregset_t elf_gregset_t;
2122 typedef thrmisc_t elf_thrmisc_t;
2123 #define ELF_KERN_PROC_MASK	0
2124 typedef struct kinfo_proc elf_kinfo_proc_t;
2125 typedef vm_offset_t elf_ps_strings_t;
2126 #endif
2127 
2128 static void
2129 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2130 {
2131 	struct sbuf sbarg;
2132 	size_t len;
2133 	char *cp, *end;
2134 	struct proc *p;
2135 	elf_prpsinfo_t *psinfo;
2136 	int error;
2137 
2138 	p = arg;
2139 	if (sb != NULL) {
2140 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2141 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2142 		psinfo->pr_version = PRPSINFO_VERSION;
2143 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2144 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2145 		PROC_LOCK(p);
2146 		if (p->p_args != NULL) {
2147 			len = sizeof(psinfo->pr_psargs) - 1;
2148 			if (len > p->p_args->ar_length)
2149 				len = p->p_args->ar_length;
2150 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2151 			PROC_UNLOCK(p);
2152 			error = 0;
2153 		} else {
2154 			_PHOLD(p);
2155 			PROC_UNLOCK(p);
2156 			sbuf_new(&sbarg, psinfo->pr_psargs,
2157 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2158 			error = proc_getargv(curthread, p, &sbarg);
2159 			PRELE(p);
2160 			if (sbuf_finish(&sbarg) == 0)
2161 				len = sbuf_len(&sbarg) - 1;
2162 			else
2163 				len = sizeof(psinfo->pr_psargs) - 1;
2164 			sbuf_delete(&sbarg);
2165 		}
2166 		if (error || len == 0)
2167 			strlcpy(psinfo->pr_psargs, p->p_comm,
2168 			    sizeof(psinfo->pr_psargs));
2169 		else {
2170 			KASSERT(len < sizeof(psinfo->pr_psargs),
2171 			    ("len is too long: %zu vs %zu", len,
2172 			    sizeof(psinfo->pr_psargs)));
2173 			cp = psinfo->pr_psargs;
2174 			end = cp + len - 1;
2175 			for (;;) {
2176 				cp = memchr(cp, '\0', end - cp);
2177 				if (cp == NULL)
2178 					break;
2179 				*cp = ' ';
2180 			}
2181 		}
2182 		psinfo->pr_pid = p->p_pid;
2183 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2184 		free(psinfo, M_TEMP);
2185 	}
2186 	*sizep = sizeof(*psinfo);
2187 }
2188 
2189 static bool
2190 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf,
2191     size_t *sizep)
2192 {
2193 	elf_prstatus_t *status;
2194 
2195 	if (buf != NULL) {
2196 		KASSERT(*sizep == sizeof(*status), ("%s: invalid size",
2197 		    __func__));
2198 		status = buf;
2199 		status->pr_version = PRSTATUS_VERSION;
2200 		status->pr_statussz = sizeof(elf_prstatus_t);
2201 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2202 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2203 		status->pr_osreldate = osreldate;
2204 		status->pr_cursig = td->td_proc->p_sig;
2205 		status->pr_pid = td->td_tid;
2206 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2207 		fill_regs32(td, &status->pr_reg);
2208 #else
2209 		fill_regs(td, &status->pr_reg);
2210 #endif
2211 	}
2212 	*sizep = sizeof(*status);
2213 	return (true);
2214 }
2215 
2216 static bool
2217 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf,
2218     size_t size)
2219 {
2220 	elf_prstatus_t *status;
2221 
2222 	KASSERT(size == sizeof(*status), ("%s: invalid size", __func__));
2223 	status = buf;
2224 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2225 	set_regs32(td, &status->pr_reg);
2226 #else
2227 	set_regs(td, &status->pr_reg);
2228 #endif
2229 	return (true);
2230 }
2231 
2232 static struct regset __elfN(regset_prstatus) = {
2233 	.note = NT_PRSTATUS,
2234 	.size = sizeof(elf_prstatus_t),
2235 	.get = __elfN(get_prstatus),
2236 	.set = __elfN(set_prstatus),
2237 };
2238 ELF_REGSET(__elfN(regset_prstatus));
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 = arg;
2247 	if (sb != NULL) {
2248 		KASSERT(*sizep == sizeof(*status), ("%s: invalid size",
2249 		    __func__));
2250 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2251 		__elfN(get_prstatus)(NULL, td, status, sizep);
2252 		sbuf_bcat(sb, status, sizeof(*status));
2253 		free(status, M_TEMP);
2254 	}
2255 	*sizep = sizeof(*status);
2256 }
2257 
2258 static bool
2259 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf,
2260     size_t *sizep)
2261 {
2262 	elf_prfpregset_t *fpregset;
2263 
2264 	if (buf != NULL) {
2265 		KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size",
2266 		    __func__));
2267 		fpregset = buf;
2268 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2269 		fill_fpregs32(td, fpregset);
2270 #else
2271 		fill_fpregs(td, fpregset);
2272 #endif
2273 	}
2274 	*sizep = sizeof(fpregset);
2275 	return (true);
2276 }
2277 
2278 static bool
2279 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf,
2280     size_t size)
2281 {
2282 	elf_prfpregset_t *fpregset;
2283 
2284 	fpregset = buf;
2285 	KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__));
2286 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2287 	set_fpregs32(td, fpregset);
2288 #else
2289 	set_fpregs(td, fpregset);
2290 #endif
2291 	return (true);
2292 }
2293 
2294 static struct regset __elfN(regset_fpregset) = {
2295 	.note = NT_FPREGSET,
2296 	.size = sizeof(elf_prfpregset_t),
2297 	.get = __elfN(get_fpregset),
2298 	.set = __elfN(set_fpregset),
2299 };
2300 ELF_REGSET(__elfN(regset_fpregset));
2301 
2302 static void
2303 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2304 {
2305 	struct thread *td;
2306 	elf_prfpregset_t *fpregset;
2307 
2308 	td = arg;
2309 	if (sb != NULL) {
2310 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2311 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2312 		__elfN(get_fpregset)(NULL, td, fpregset, sizep);
2313 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2314 		free(fpregset, M_TEMP);
2315 	}
2316 	*sizep = sizeof(*fpregset);
2317 }
2318 
2319 static void
2320 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2321 {
2322 	struct thread *td;
2323 	elf_thrmisc_t thrmisc;
2324 
2325 	td = arg;
2326 	if (sb != NULL) {
2327 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2328 		bzero(&thrmisc, sizeof(thrmisc));
2329 		strcpy(thrmisc.pr_tname, td->td_name);
2330 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2331 	}
2332 	*sizep = sizeof(thrmisc);
2333 }
2334 
2335 static void
2336 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2337 {
2338 	struct thread *td;
2339 	size_t size;
2340 	int structsize;
2341 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2342 	struct ptrace_lwpinfo32 pl;
2343 #else
2344 	struct ptrace_lwpinfo pl;
2345 #endif
2346 
2347 	td = arg;
2348 	size = sizeof(structsize) + sizeof(pl);
2349 	if (sb != NULL) {
2350 		KASSERT(*sizep == size, ("invalid size"));
2351 		structsize = sizeof(pl);
2352 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2353 		bzero(&pl, sizeof(pl));
2354 		pl.pl_lwpid = td->td_tid;
2355 		pl.pl_event = PL_EVENT_NONE;
2356 		pl.pl_sigmask = td->td_sigmask;
2357 		pl.pl_siglist = td->td_siglist;
2358 		if (td->td_si.si_signo != 0) {
2359 			pl.pl_event = PL_EVENT_SIGNAL;
2360 			pl.pl_flags |= PL_FLAG_SI;
2361 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2362 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2363 #else
2364 			pl.pl_siginfo = td->td_si;
2365 #endif
2366 		}
2367 		strcpy(pl.pl_tdname, td->td_name);
2368 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2369 		sbuf_bcat(sb, &pl, sizeof(pl));
2370 	}
2371 	*sizep = size;
2372 }
2373 
2374 /*
2375  * Allow for MD specific notes, as well as any MD
2376  * specific preparations for writing MI notes.
2377  */
2378 static void
2379 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2380 {
2381 	struct thread *td;
2382 	void *buf;
2383 	size_t size;
2384 
2385 	td = (struct thread *)arg;
2386 	size = *sizep;
2387 	if (size != 0 && sb != NULL)
2388 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2389 	else
2390 		buf = NULL;
2391 	size = 0;
2392 	__elfN(dump_thread)(td, buf, &size);
2393 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2394 	if (size != 0 && sb != NULL)
2395 		sbuf_bcat(sb, buf, size);
2396 	free(buf, M_TEMP);
2397 	*sizep = size;
2398 }
2399 
2400 #ifdef KINFO_PROC_SIZE
2401 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2402 #endif
2403 
2404 static void
2405 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2406 {
2407 	struct proc *p;
2408 	size_t size;
2409 	int structsize;
2410 
2411 	p = arg;
2412 	size = sizeof(structsize) + p->p_numthreads *
2413 	    sizeof(elf_kinfo_proc_t);
2414 
2415 	if (sb != NULL) {
2416 		KASSERT(*sizep == size, ("invalid size"));
2417 		structsize = sizeof(elf_kinfo_proc_t);
2418 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2419 		sx_slock(&proctree_lock);
2420 		PROC_LOCK(p);
2421 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2422 		sx_sunlock(&proctree_lock);
2423 	}
2424 	*sizep = size;
2425 }
2426 
2427 #ifdef KINFO_FILE_SIZE
2428 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2429 #endif
2430 
2431 static void
2432 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2433 {
2434 	struct proc *p;
2435 	size_t size, sect_sz, i;
2436 	ssize_t start_len, sect_len;
2437 	int structsize, filedesc_flags;
2438 
2439 	if (coredump_pack_fileinfo)
2440 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2441 	else
2442 		filedesc_flags = 0;
2443 
2444 	p = arg;
2445 	structsize = sizeof(struct kinfo_file);
2446 	if (sb == NULL) {
2447 		size = 0;
2448 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2449 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2450 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2451 		PROC_LOCK(p);
2452 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2453 		sbuf_finish(sb);
2454 		sbuf_delete(sb);
2455 		*sizep = size;
2456 	} else {
2457 		sbuf_start_section(sb, &start_len);
2458 
2459 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2460 		PROC_LOCK(p);
2461 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2462 		    filedesc_flags);
2463 
2464 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2465 		if (sect_len < 0)
2466 			return;
2467 		sect_sz = sect_len;
2468 
2469 		KASSERT(sect_sz <= *sizep,
2470 		    ("kern_proc_filedesc_out did not respect maxlen; "
2471 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2472 		     sect_sz - sizeof(structsize)));
2473 
2474 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2475 			sbuf_putc(sb, 0);
2476 	}
2477 }
2478 
2479 #ifdef KINFO_VMENTRY_SIZE
2480 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2481 #endif
2482 
2483 static void
2484 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2485 {
2486 	struct proc *p;
2487 	size_t size;
2488 	int structsize, vmmap_flags;
2489 
2490 	if (coredump_pack_vmmapinfo)
2491 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2492 	else
2493 		vmmap_flags = 0;
2494 
2495 	p = arg;
2496 	structsize = sizeof(struct kinfo_vmentry);
2497 	if (sb == NULL) {
2498 		size = 0;
2499 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2500 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2501 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2502 		PROC_LOCK(p);
2503 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2504 		sbuf_finish(sb);
2505 		sbuf_delete(sb);
2506 		*sizep = size;
2507 	} else {
2508 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2509 		PROC_LOCK(p);
2510 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2511 		    vmmap_flags);
2512 	}
2513 }
2514 
2515 static void
2516 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2517 {
2518 	struct proc *p;
2519 	size_t size;
2520 	int structsize;
2521 
2522 	p = arg;
2523 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2524 	if (sb != NULL) {
2525 		KASSERT(*sizep == size, ("invalid size"));
2526 		structsize = sizeof(gid_t);
2527 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2528 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2529 		    sizeof(gid_t));
2530 	}
2531 	*sizep = size;
2532 }
2533 
2534 static void
2535 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2536 {
2537 	struct proc *p;
2538 	size_t size;
2539 	int structsize;
2540 
2541 	p = arg;
2542 	size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2543 	if (sb != NULL) {
2544 		KASSERT(*sizep == size, ("invalid size"));
2545 		structsize = sizeof(p->p_pd->pd_cmask);
2546 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2547 		sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2548 	}
2549 	*sizep = size;
2550 }
2551 
2552 static void
2553 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2554 {
2555 	struct proc *p;
2556 	struct rlimit rlim[RLIM_NLIMITS];
2557 	size_t size;
2558 	int structsize, i;
2559 
2560 	p = arg;
2561 	size = sizeof(structsize) + sizeof(rlim);
2562 	if (sb != NULL) {
2563 		KASSERT(*sizep == size, ("invalid size"));
2564 		structsize = sizeof(rlim);
2565 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2566 		PROC_LOCK(p);
2567 		for (i = 0; i < RLIM_NLIMITS; i++)
2568 			lim_rlimit_proc(p, i, &rlim[i]);
2569 		PROC_UNLOCK(p);
2570 		sbuf_bcat(sb, rlim, sizeof(rlim));
2571 	}
2572 	*sizep = size;
2573 }
2574 
2575 static void
2576 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2577 {
2578 	struct proc *p;
2579 	size_t size;
2580 	int structsize;
2581 
2582 	p = arg;
2583 	size = sizeof(structsize) + sizeof(p->p_osrel);
2584 	if (sb != NULL) {
2585 		KASSERT(*sizep == size, ("invalid size"));
2586 		structsize = sizeof(p->p_osrel);
2587 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2588 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2589 	}
2590 	*sizep = size;
2591 }
2592 
2593 static void
2594 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2595 {
2596 	struct proc *p;
2597 	elf_ps_strings_t ps_strings;
2598 	size_t size;
2599 	int structsize;
2600 
2601 	p = arg;
2602 	size = sizeof(structsize) + sizeof(ps_strings);
2603 	if (sb != NULL) {
2604 		KASSERT(*sizep == size, ("invalid size"));
2605 		structsize = sizeof(ps_strings);
2606 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2607 		ps_strings = PTROUT(PROC_PS_STRINGS(p));
2608 #else
2609 		ps_strings = PROC_PS_STRINGS(p);
2610 #endif
2611 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2612 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2613 	}
2614 	*sizep = size;
2615 }
2616 
2617 static void
2618 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2619 {
2620 	struct proc *p;
2621 	size_t size;
2622 	int structsize;
2623 
2624 	p = arg;
2625 	if (sb == NULL) {
2626 		size = 0;
2627 		sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2628 		    SBUF_FIXEDLEN);
2629 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2630 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2631 		PHOLD(p);
2632 		proc_getauxv(curthread, p, sb);
2633 		PRELE(p);
2634 		sbuf_finish(sb);
2635 		sbuf_delete(sb);
2636 		*sizep = size;
2637 	} else {
2638 		structsize = sizeof(Elf_Auxinfo);
2639 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2640 		PHOLD(p);
2641 		proc_getauxv(curthread, p, sb);
2642 		PRELE(p);
2643 	}
2644 }
2645 
2646 static bool
2647 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2648     const char *note_vendor, const Elf_Phdr *pnote,
2649     bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2650 {
2651 	const Elf_Note *note, *note0, *note_end;
2652 	const char *note_name;
2653 	char *buf;
2654 	int i, error;
2655 	bool res;
2656 
2657 	/* We need some limit, might as well use PAGE_SIZE. */
2658 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2659 		return (false);
2660 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2661 	if (pnote->p_offset > PAGE_SIZE ||
2662 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2663 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2664 		if (buf == NULL) {
2665 			VOP_UNLOCK(imgp->vp);
2666 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2667 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2668 		}
2669 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2670 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2671 		    curthread->td_ucred, NOCRED, NULL, curthread);
2672 		if (error != 0) {
2673 			uprintf("i/o error PT_NOTE\n");
2674 			goto retf;
2675 		}
2676 		note = note0 = (const Elf_Note *)buf;
2677 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2678 	} else {
2679 		note = note0 = (const Elf_Note *)(imgp->image_header +
2680 		    pnote->p_offset);
2681 		note_end = (const Elf_Note *)(imgp->image_header +
2682 		    pnote->p_offset + pnote->p_filesz);
2683 		buf = NULL;
2684 	}
2685 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2686 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2687 		    (const char *)note < sizeof(Elf_Note)) {
2688 			goto retf;
2689 		}
2690 		if (note->n_namesz != checknote->n_namesz ||
2691 		    note->n_descsz != checknote->n_descsz ||
2692 		    note->n_type != checknote->n_type)
2693 			goto nextnote;
2694 		note_name = (const char *)(note + 1);
2695 		if (note_name + checknote->n_namesz >=
2696 		    (const char *)note_end || strncmp(note_vendor,
2697 		    note_name, checknote->n_namesz) != 0)
2698 			goto nextnote;
2699 
2700 		if (cb(note, cb_arg, &res))
2701 			goto ret;
2702 nextnote:
2703 		note = (const Elf_Note *)((const char *)(note + 1) +
2704 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2705 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2706 	}
2707 retf:
2708 	res = false;
2709 ret:
2710 	free(buf, M_TEMP);
2711 	return (res);
2712 }
2713 
2714 struct brandnote_cb_arg {
2715 	Elf_Brandnote *brandnote;
2716 	int32_t *osrel;
2717 };
2718 
2719 static bool
2720 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2721 {
2722 	struct brandnote_cb_arg *arg;
2723 
2724 	arg = arg0;
2725 
2726 	/*
2727 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2728 	 * necessary.
2729 	 */
2730 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2731 	    arg->brandnote->trans_osrel != NULL ?
2732 	    arg->brandnote->trans_osrel(note, arg->osrel) : true;
2733 
2734 	return (true);
2735 }
2736 
2737 static Elf_Note fctl_note = {
2738 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2739 	.n_descsz = sizeof(uint32_t),
2740 	.n_type = NT_FREEBSD_FEATURE_CTL,
2741 };
2742 
2743 struct fctl_cb_arg {
2744 	bool *has_fctl0;
2745 	uint32_t *fctl0;
2746 };
2747 
2748 static bool
2749 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2750 {
2751 	struct fctl_cb_arg *arg;
2752 	const Elf32_Word *desc;
2753 	uintptr_t p;
2754 
2755 	arg = arg0;
2756 	p = (uintptr_t)(note + 1);
2757 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2758 	desc = (const Elf32_Word *)p;
2759 	*arg->has_fctl0 = true;
2760 	*arg->fctl0 = desc[0];
2761 	*res = true;
2762 	return (true);
2763 }
2764 
2765 /*
2766  * Try to find the appropriate ABI-note section for checknote, fetch
2767  * the osreldate and feature control flags for binary from the ELF
2768  * OSABI-note.  Only the first page of the image is searched, the same
2769  * as for headers.
2770  */
2771 static bool
2772 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2773     int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2774 {
2775 	const Elf_Phdr *phdr;
2776 	const Elf_Ehdr *hdr;
2777 	struct brandnote_cb_arg b_arg;
2778 	struct fctl_cb_arg f_arg;
2779 	int i, j;
2780 
2781 	hdr = (const Elf_Ehdr *)imgp->image_header;
2782 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2783 	b_arg.brandnote = brandnote;
2784 	b_arg.osrel = osrel;
2785 	f_arg.has_fctl0 = has_fctl0;
2786 	f_arg.fctl0 = fctl0;
2787 
2788 	for (i = 0; i < hdr->e_phnum; i++) {
2789 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2790 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2791 		    &b_arg)) {
2792 			for (j = 0; j < hdr->e_phnum; j++) {
2793 				if (phdr[j].p_type == PT_NOTE &&
2794 				    __elfN(parse_notes)(imgp, &fctl_note,
2795 				    FREEBSD_ABI_VENDOR, &phdr[j],
2796 				    note_fctl_cb, &f_arg))
2797 					break;
2798 			}
2799 			return (true);
2800 		}
2801 	}
2802 	return (false);
2803 
2804 }
2805 
2806 /*
2807  * Tell kern_execve.c about it, with a little help from the linker.
2808  */
2809 static struct execsw __elfN(execsw) = {
2810 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2811 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2812 };
2813 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2814 
2815 static vm_prot_t
2816 __elfN(trans_prot)(Elf_Word flags)
2817 {
2818 	vm_prot_t prot;
2819 
2820 	prot = 0;
2821 	if (flags & PF_X)
2822 		prot |= VM_PROT_EXECUTE;
2823 	if (flags & PF_W)
2824 		prot |= VM_PROT_WRITE;
2825 	if (flags & PF_R)
2826 		prot |= VM_PROT_READ;
2827 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2828 	if (i386_read_exec && (flags & PF_R))
2829 		prot |= VM_PROT_EXECUTE;
2830 #endif
2831 	return (prot);
2832 }
2833 
2834 static Elf_Word
2835 __elfN(untrans_prot)(vm_prot_t prot)
2836 {
2837 	Elf_Word flags;
2838 
2839 	flags = 0;
2840 	if (prot & VM_PROT_EXECUTE)
2841 		flags |= PF_X;
2842 	if (prot & VM_PROT_READ)
2843 		flags |= PF_R;
2844 	if (prot & VM_PROT_WRITE)
2845 		flags |= PF_W;
2846 	return (flags);
2847 }
2848