xref: /freebsd/sys/kern/imgact_elf.c (revision bdcfd222ce6369e7aeaceb9a92ffdde84bdbf6cd)
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_gap) = 3;
205 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
206     &__elfN(aslr_stack_gap), 0,
207     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
208     ": maximum percentage of main stack to waste on a random gap");
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 static u_long
879 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
880     u_int align)
881 {
882 	u_long rbase, res;
883 
884 	MPASS(vm_map_min(map) <= minv);
885 	MPASS(maxv <= vm_map_max(map));
886 	MPASS(minv < maxv);
887 	MPASS(minv + align < maxv);
888 	arc4rand(&rbase, sizeof(rbase), 0);
889 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
890 	res &= ~((u_long)align - 1);
891 	if (res >= maxv)
892 		res -= align;
893 	KASSERT(res >= minv,
894 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
895 	    res, minv, maxv, rbase));
896 	KASSERT(res < maxv,
897 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
898 	    res, maxv, minv, rbase));
899 	return (res);
900 }
901 
902 static int
903 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
904     const Elf_Phdr *phdr, u_long et_dyn_addr)
905 {
906 	struct vmspace *vmspace;
907 	const char *err_str;
908 	u_long text_size, data_size, total_size, text_addr, data_addr;
909 	u_long seg_size, seg_addr;
910 	int i;
911 
912 	err_str = NULL;
913 	text_size = data_size = total_size = text_addr = data_addr = 0;
914 
915 	for (i = 0; i < hdr->e_phnum; i++) {
916 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
917 			continue;
918 
919 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
920 		seg_size = round_page(phdr[i].p_memsz +
921 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
922 
923 		/*
924 		 * Make the largest executable segment the official
925 		 * text segment and all others data.
926 		 *
927 		 * Note that obreak() assumes that data_addr + data_size == end
928 		 * of data load area, and the ELF file format expects segments
929 		 * to be sorted by address.  If multiple data segments exist,
930 		 * the last one will be used.
931 		 */
932 
933 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
934 			text_size = seg_size;
935 			text_addr = seg_addr;
936 		} else {
937 			data_size = seg_size;
938 			data_addr = seg_addr;
939 		}
940 		total_size += seg_size;
941 	}
942 
943 	if (data_addr == 0 && data_size == 0) {
944 		data_addr = text_addr;
945 		data_size = text_size;
946 	}
947 
948 	/*
949 	 * Check limits.  It should be safe to check the
950 	 * limits after loading the segments since we do
951 	 * not actually fault in all the segments pages.
952 	 */
953 	PROC_LOCK(imgp->proc);
954 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
955 		err_str = "Data segment size exceeds process limit";
956 	else if (text_size > maxtsiz)
957 		err_str = "Text segment size exceeds system limit";
958 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
959 		err_str = "Total segment size exceeds process limit";
960 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
961 		err_str = "Data segment size exceeds resource limit";
962 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
963 		err_str = "Total segment size exceeds resource limit";
964 	PROC_UNLOCK(imgp->proc);
965 	if (err_str != NULL) {
966 		uprintf("%s\n", err_str);
967 		return (ENOMEM);
968 	}
969 
970 	vmspace = imgp->proc->p_vmspace;
971 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
972 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
973 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
974 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
975 
976 	return (0);
977 }
978 
979 static int
980 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
981     char **interpp, bool *free_interpp)
982 {
983 	struct thread *td;
984 	char *interp;
985 	int error, interp_name_len;
986 
987 	KASSERT(phdr->p_type == PT_INTERP,
988 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
989 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
990 
991 	td = curthread;
992 
993 	/* Path to interpreter */
994 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
995 		uprintf("Invalid PT_INTERP\n");
996 		return (ENOEXEC);
997 	}
998 
999 	interp_name_len = phdr->p_filesz;
1000 	if (phdr->p_offset > PAGE_SIZE ||
1001 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
1002 		/*
1003 		 * The vnode lock might be needed by the pagedaemon to
1004 		 * clean pages owned by the vnode.  Do not allow sleep
1005 		 * waiting for memory with the vnode locked, instead
1006 		 * try non-sleepable allocation first, and if it
1007 		 * fails, go to the slow path were we drop the lock
1008 		 * and do M_WAITOK.  A text reference prevents
1009 		 * modifications to the vnode content.
1010 		 */
1011 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1012 		if (interp == NULL) {
1013 			VOP_UNLOCK(imgp->vp);
1014 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1015 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1016 		}
1017 
1018 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
1019 		    interp_name_len, phdr->p_offset,
1020 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1021 		    NOCRED, NULL, td);
1022 		if (error != 0) {
1023 			free(interp, M_TEMP);
1024 			uprintf("i/o error PT_INTERP %d\n", error);
1025 			return (error);
1026 		}
1027 		interp[interp_name_len] = '\0';
1028 
1029 		*interpp = interp;
1030 		*free_interpp = true;
1031 		return (0);
1032 	}
1033 
1034 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1035 	if (interp[interp_name_len - 1] != '\0') {
1036 		uprintf("Invalid PT_INTERP\n");
1037 		return (ENOEXEC);
1038 	}
1039 
1040 	*interpp = interp;
1041 	*free_interpp = false;
1042 	return (0);
1043 }
1044 
1045 static int
1046 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1047     const char *interp, u_long *addr, u_long *entry)
1048 {
1049 	char *path;
1050 	int error;
1051 
1052 	if (brand_info->emul_path != NULL &&
1053 	    brand_info->emul_path[0] != '\0') {
1054 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1055 		snprintf(path, MAXPATHLEN, "%s%s",
1056 		    brand_info->emul_path, interp);
1057 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1058 		free(path, M_TEMP);
1059 		if (error == 0)
1060 			return (0);
1061 	}
1062 
1063 	if (brand_info->interp_newpath != NULL &&
1064 	    (brand_info->interp_path == NULL ||
1065 	    strcmp(interp, brand_info->interp_path) == 0)) {
1066 		error = __elfN(load_file)(imgp->proc,
1067 		    brand_info->interp_newpath, addr, entry);
1068 		if (error == 0)
1069 			return (0);
1070 	}
1071 
1072 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1073 	if (error == 0)
1074 		return (0);
1075 
1076 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1077 	return (error);
1078 }
1079 
1080 /*
1081  * Impossible et_dyn_addr initial value indicating that the real base
1082  * must be calculated later with some randomization applied.
1083  */
1084 #define	ET_DYN_ADDR_RAND	1
1085 
1086 static int
1087 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1088 {
1089 	struct thread *td;
1090 	const Elf_Ehdr *hdr;
1091 	const Elf_Phdr *phdr;
1092 	Elf_Auxargs *elf_auxargs;
1093 	struct vmspace *vmspace;
1094 	vm_map_t map;
1095 	char *interp;
1096 	Elf_Brandinfo *brand_info;
1097 	struct sysentvec *sv;
1098 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1099 	u_long maxalign, mapsz, maxv, maxv1;
1100 	uint32_t fctl0;
1101 	int32_t osrel;
1102 	bool free_interp;
1103 	int error, i, n;
1104 
1105 	hdr = (const Elf_Ehdr *)imgp->image_header;
1106 
1107 	/*
1108 	 * Do we have a valid ELF header ?
1109 	 *
1110 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1111 	 * if particular brand doesn't support it.
1112 	 */
1113 	if (__elfN(check_header)(hdr) != 0 ||
1114 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1115 		return (-1);
1116 
1117 	/*
1118 	 * From here on down, we return an errno, not -1, as we've
1119 	 * detected an ELF file.
1120 	 */
1121 
1122 	if (!__elfN(phdr_in_zero_page)(hdr)) {
1123 		uprintf("Program headers not in the first page\n");
1124 		return (ENOEXEC);
1125 	}
1126 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1127 	if (!aligned(phdr, Elf_Addr)) {
1128 		uprintf("Unaligned program headers\n");
1129 		return (ENOEXEC);
1130 	}
1131 
1132 	n = error = 0;
1133 	baddr = 0;
1134 	osrel = 0;
1135 	fctl0 = 0;
1136 	entry = proghdr = 0;
1137 	interp = NULL;
1138 	free_interp = false;
1139 	td = curthread;
1140 	maxalign = PAGE_SIZE;
1141 	mapsz = 0;
1142 
1143 	for (i = 0; i < hdr->e_phnum; i++) {
1144 		switch (phdr[i].p_type) {
1145 		case PT_LOAD:
1146 			if (n == 0)
1147 				baddr = phdr[i].p_vaddr;
1148 			if (phdr[i].p_align > maxalign)
1149 				maxalign = phdr[i].p_align;
1150 			mapsz += phdr[i].p_memsz;
1151 			n++;
1152 
1153 			/*
1154 			 * If this segment contains the program headers,
1155 			 * remember their virtual address for the AT_PHDR
1156 			 * aux entry. Static binaries don't usually include
1157 			 * a PT_PHDR entry.
1158 			 */
1159 			if (phdr[i].p_offset == 0 &&
1160 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1161 				<= phdr[i].p_filesz)
1162 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1163 			break;
1164 		case PT_INTERP:
1165 			/* Path to interpreter */
1166 			if (interp != NULL) {
1167 				uprintf("Multiple PT_INTERP headers\n");
1168 				error = ENOEXEC;
1169 				goto ret;
1170 			}
1171 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1172 			    &free_interp);
1173 			if (error != 0)
1174 				goto ret;
1175 			break;
1176 		case PT_GNU_STACK:
1177 			if (__elfN(nxstack))
1178 				imgp->stack_prot =
1179 				    __elfN(trans_prot)(phdr[i].p_flags);
1180 			imgp->stack_sz = phdr[i].p_memsz;
1181 			break;
1182 		case PT_PHDR: 	/* Program header table info */
1183 			proghdr = phdr[i].p_vaddr;
1184 			break;
1185 		}
1186 	}
1187 
1188 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1189 	if (brand_info == NULL) {
1190 		uprintf("ELF binary type \"%u\" not known.\n",
1191 		    hdr->e_ident[EI_OSABI]);
1192 		error = ENOEXEC;
1193 		goto ret;
1194 	}
1195 	sv = brand_info->sysvec;
1196 	et_dyn_addr = 0;
1197 	if (hdr->e_type == ET_DYN) {
1198 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1199 			uprintf("Cannot execute shared object\n");
1200 			error = ENOEXEC;
1201 			goto ret;
1202 		}
1203 		/*
1204 		 * Honour the base load address from the dso if it is
1205 		 * non-zero for some reason.
1206 		 */
1207 		if (baddr == 0) {
1208 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1209 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1210 				et_dyn_addr = __elfN(pie_base);
1211 			else if ((__elfN(pie_aslr_enabled) &&
1212 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1213 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1214 				et_dyn_addr = ET_DYN_ADDR_RAND;
1215 			else
1216 				et_dyn_addr = __elfN(pie_base);
1217 		}
1218 	}
1219 
1220 	/*
1221 	 * Avoid a possible deadlock if the current address space is destroyed
1222 	 * and that address space maps the locked vnode.  In the common case,
1223 	 * the locked vnode's v_usecount is decremented but remains greater
1224 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1225 	 * However, in cases where the vnode lock is external, such as nullfs,
1226 	 * v_usecount may become zero.
1227 	 *
1228 	 * The VV_TEXT flag prevents modifications to the executable while
1229 	 * the vnode is unlocked.
1230 	 */
1231 	VOP_UNLOCK(imgp->vp);
1232 
1233 	/*
1234 	 * Decide whether to enable randomization of user mappings.
1235 	 * First, reset user preferences for the setid binaries.
1236 	 * Then, account for the support of the randomization by the
1237 	 * ABI, by user preferences, and make special treatment for
1238 	 * PIE binaries.
1239 	 */
1240 	if (imgp->credential_setid) {
1241 		PROC_LOCK(imgp->proc);
1242 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1243 		    P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1244 		PROC_UNLOCK(imgp->proc);
1245 	}
1246 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1247 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1248 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1249 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1250 		    ("et_dyn_addr == RAND and !ASLR"));
1251 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1252 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1253 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1254 		imgp->map_flags |= MAP_ASLR;
1255 		/*
1256 		 * If user does not care about sbrk, utilize the bss
1257 		 * grow region for mappings as well.  We can select
1258 		 * the base for the image anywere and still not suffer
1259 		 * from the fragmentation.
1260 		 */
1261 		if (!__elfN(aslr_honor_sbrk) ||
1262 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1263 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1264 	}
1265 
1266 	if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1267 	    (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1268 	    (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1269 		imgp->map_flags |= MAP_WXORX;
1270 
1271 	error = exec_new_vmspace(imgp, sv);
1272 	vmspace = imgp->proc->p_vmspace;
1273 	map = &vmspace->vm_map;
1274 
1275 	imgp->proc->p_sysent = sv;
1276 	imgp->proc->p_elf_brandinfo = brand_info;
1277 
1278 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1279 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1280 		KASSERT((map->flags & MAP_ASLR) != 0,
1281 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1282 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1283 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1284 		    /* reserve half of the address space to interpreter */
1285 		    maxv / 2, 1UL << flsl(maxalign));
1286 	}
1287 
1288 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1289 	if (error != 0)
1290 		goto ret;
1291 
1292 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1293 	if (error != 0)
1294 		goto ret;
1295 
1296 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1297 	if (error != 0)
1298 		goto ret;
1299 
1300 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1301 
1302 	/*
1303 	 * We load the dynamic linker where a userland call
1304 	 * to mmap(0, ...) would put it.  The rationale behind this
1305 	 * calculation is that it leaves room for the heap to grow to
1306 	 * its maximum allowed size.
1307 	 */
1308 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1309 	    RLIMIT_DATA));
1310 	if ((map->flags & MAP_ASLR) != 0) {
1311 		maxv1 = maxv / 2 + addr / 2;
1312 		MPASS(maxv1 >= addr);	/* No overflow */
1313 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1314 		    (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1315 		    pagesizes[1] : pagesizes[0]);
1316 	} else {
1317 		map->anon_loc = addr;
1318 	}
1319 
1320 	imgp->entry_addr = entry;
1321 
1322 	if (interp != NULL) {
1323 		VOP_UNLOCK(imgp->vp);
1324 		if ((map->flags & MAP_ASLR) != 0) {
1325 			/* Assume that interpreter fits into 1/4 of AS */
1326 			maxv1 = maxv / 2 + addr / 2;
1327 			MPASS(maxv1 >= addr);	/* No overflow */
1328 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1329 			    maxv1, PAGE_SIZE);
1330 		}
1331 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1332 		    &imgp->entry_addr);
1333 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1334 		if (error != 0)
1335 			goto ret;
1336 	} else
1337 		addr = et_dyn_addr;
1338 
1339 	/*
1340 	 * Construct auxargs table (used by the copyout_auxargs routine)
1341 	 */
1342 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1343 	if (elf_auxargs == NULL) {
1344 		VOP_UNLOCK(imgp->vp);
1345 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1346 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1347 	}
1348 	elf_auxargs->execfd = -1;
1349 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1350 	elf_auxargs->phent = hdr->e_phentsize;
1351 	elf_auxargs->phnum = hdr->e_phnum;
1352 	elf_auxargs->pagesz = PAGE_SIZE;
1353 	elf_auxargs->base = addr;
1354 	elf_auxargs->flags = 0;
1355 	elf_auxargs->entry = entry;
1356 	elf_auxargs->hdr_eflags = hdr->e_flags;
1357 
1358 	imgp->auxargs = elf_auxargs;
1359 	imgp->interpreted = 0;
1360 	imgp->reloc_base = addr;
1361 	imgp->proc->p_osrel = osrel;
1362 	imgp->proc->p_fctl0 = fctl0;
1363 	imgp->proc->p_elf_flags = hdr->e_flags;
1364 
1365 ret:
1366 	if (free_interp)
1367 		free(interp, M_TEMP);
1368 	return (error);
1369 }
1370 
1371 #define	elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1372 
1373 int
1374 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1375 {
1376 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1377 	Elf_Auxinfo *argarray, *pos;
1378 	int error;
1379 
1380 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1381 	    M_WAITOK | M_ZERO);
1382 
1383 	if (args->execfd != -1)
1384 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1385 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1386 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1387 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1388 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1389 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1390 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1391 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1392 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1393 	if (imgp->execpathp != 0)
1394 		AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1395 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1396 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1397 	if (imgp->canary != 0) {
1398 		AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1399 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1400 	}
1401 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1402 	if (imgp->pagesizes != 0) {
1403 		AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1404 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1405 	}
1406 	if (imgp->sysent->sv_timekeep_base != 0) {
1407 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1408 		    imgp->sysent->sv_timekeep_base);
1409 	}
1410 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1411 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1412 	    imgp->sysent->sv_stackprot);
1413 	if (imgp->sysent->sv_hwcap != NULL)
1414 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1415 	if (imgp->sysent->sv_hwcap2 != NULL)
1416 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1417 	AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1418 	    ELF_BSDF_SIGFASTBLK : 0);
1419 	AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1420 	AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1421 	AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1422 	AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1423 	AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1424 	if (imgp->sysent->sv_fxrng_gen_base != 0)
1425 		AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1426 	if (imgp->sysent->sv_vdso_base != 0 && __elfN(vdso) != 0)
1427 		AUXARGS_ENTRY(pos, AT_KPRELOAD, imgp->sysent->sv_vdso_base);
1428 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1429 
1430 	free(imgp->auxargs, M_TEMP);
1431 	imgp->auxargs = NULL;
1432 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1433 
1434 	error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1435 	free(argarray, M_TEMP);
1436 	return (error);
1437 }
1438 
1439 int
1440 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1441 {
1442 	Elf_Addr *base;
1443 
1444 	base = (Elf_Addr *)*stack_base;
1445 	base--;
1446 	if (elf_suword(base, imgp->args->argc) == -1)
1447 		return (EFAULT);
1448 	*stack_base = (uintptr_t)base;
1449 	return (0);
1450 }
1451 
1452 /*
1453  * Code for generating ELF core dumps.
1454  */
1455 
1456 typedef void (*segment_callback)(vm_map_entry_t, void *);
1457 
1458 /* Closure for cb_put_phdr(). */
1459 struct phdr_closure {
1460 	Elf_Phdr *phdr;		/* Program header to fill in */
1461 	Elf_Off offset;		/* Offset of segment in core file */
1462 };
1463 
1464 struct note_info {
1465 	int		type;		/* Note type. */
1466 	outfunc_t 	outfunc; 	/* Output function. */
1467 	void		*outarg;	/* Argument for the output function. */
1468 	size_t		outsize;	/* Output size. */
1469 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1470 };
1471 
1472 TAILQ_HEAD(note_info_list, note_info);
1473 
1474 extern int compress_user_cores;
1475 extern int compress_user_cores_level;
1476 
1477 static void cb_put_phdr(vm_map_entry_t, void *);
1478 static void cb_size_segment(vm_map_entry_t, void *);
1479 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1480     int);
1481 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1482     struct note_info_list *, size_t, int);
1483 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1484 
1485 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1486 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1487 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1488 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1489 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1490 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1491 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1492 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1493 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1494 static void note_procstat_files(void *, struct sbuf *, size_t *);
1495 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1496 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1497 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1498 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1499 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1500 
1501 static int
1502 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1503 {
1504 
1505 	return (core_write((struct coredump_params *)arg, base, len, offset,
1506 	    UIO_SYSSPACE, NULL));
1507 }
1508 
1509 int
1510 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1511 {
1512 	struct ucred *cred = td->td_ucred;
1513 	int compm, error = 0;
1514 	struct sseg_closure seginfo;
1515 	struct note_info_list notelst;
1516 	struct coredump_params params;
1517 	struct note_info *ninfo;
1518 	void *hdr, *tmpbuf;
1519 	size_t hdrsize, notesz, coresize;
1520 
1521 	hdr = NULL;
1522 	tmpbuf = NULL;
1523 	TAILQ_INIT(&notelst);
1524 
1525 	/* Size the program segments. */
1526 	__elfN(size_segments)(td, &seginfo, flags);
1527 
1528 	/*
1529 	 * Collect info about the core file header area.
1530 	 */
1531 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1532 	if (seginfo.count + 1 >= PN_XNUM)
1533 		hdrsize += sizeof(Elf_Shdr);
1534 	td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, &notelst, &notesz);
1535 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1536 
1537 	/* Set up core dump parameters. */
1538 	params.offset = 0;
1539 	params.active_cred = cred;
1540 	params.file_cred = NOCRED;
1541 	params.td = td;
1542 	params.vp = vp;
1543 	params.comp = NULL;
1544 
1545 #ifdef RACCT
1546 	if (racct_enable) {
1547 		PROC_LOCK(td->td_proc);
1548 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1549 		PROC_UNLOCK(td->td_proc);
1550 		if (error != 0) {
1551 			error = EFAULT;
1552 			goto done;
1553 		}
1554 	}
1555 #endif
1556 	if (coresize >= limit) {
1557 		error = EFAULT;
1558 		goto done;
1559 	}
1560 
1561 	/* Create a compression stream if necessary. */
1562 	compm = compress_user_cores;
1563 	if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1564 	    compm == 0)
1565 		compm = COMPRESS_GZIP;
1566 	if (compm != 0) {
1567 		params.comp = compressor_init(core_compressed_write,
1568 		    compm, CORE_BUF_SIZE,
1569 		    compress_user_cores_level, &params);
1570 		if (params.comp == NULL) {
1571 			error = EFAULT;
1572 			goto done;
1573 		}
1574 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1575         }
1576 
1577 	/*
1578 	 * Allocate memory for building the header, fill it up,
1579 	 * and write it out following the notes.
1580 	 */
1581 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1582 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1583 	    notesz, flags);
1584 
1585 	/* Write the contents of all of the writable segments. */
1586 	if (error == 0) {
1587 		Elf_Phdr *php;
1588 		off_t offset;
1589 		int i;
1590 
1591 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1592 		offset = round_page(hdrsize + notesz);
1593 		for (i = 0; i < seginfo.count; i++) {
1594 			error = core_output((char *)(uintptr_t)php->p_vaddr,
1595 			    php->p_filesz, offset, &params, tmpbuf);
1596 			if (error != 0)
1597 				break;
1598 			offset += php->p_filesz;
1599 			php++;
1600 		}
1601 		if (error == 0 && params.comp != NULL)
1602 			error = compressor_flush(params.comp);
1603 	}
1604 	if (error) {
1605 		log(LOG_WARNING,
1606 		    "Failed to write core file for process %s (error %d)\n",
1607 		    curproc->p_comm, error);
1608 	}
1609 
1610 done:
1611 	free(tmpbuf, M_TEMP);
1612 	if (params.comp != NULL)
1613 		compressor_fini(params.comp);
1614 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1615 		TAILQ_REMOVE(&notelst, ninfo, link);
1616 		free(ninfo, M_TEMP);
1617 	}
1618 	if (hdr != NULL)
1619 		free(hdr, M_TEMP);
1620 
1621 	return (error);
1622 }
1623 
1624 /*
1625  * A callback for each_dumpable_segment() to write out the segment's
1626  * program header entry.
1627  */
1628 static void
1629 cb_put_phdr(vm_map_entry_t entry, void *closure)
1630 {
1631 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1632 	Elf_Phdr *phdr = phc->phdr;
1633 
1634 	phc->offset = round_page(phc->offset);
1635 
1636 	phdr->p_type = PT_LOAD;
1637 	phdr->p_offset = phc->offset;
1638 	phdr->p_vaddr = entry->start;
1639 	phdr->p_paddr = 0;
1640 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1641 	phdr->p_align = PAGE_SIZE;
1642 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1643 
1644 	phc->offset += phdr->p_filesz;
1645 	phc->phdr++;
1646 }
1647 
1648 /*
1649  * A callback for each_dumpable_segment() to gather information about
1650  * the number of segments and their total size.
1651  */
1652 static void
1653 cb_size_segment(vm_map_entry_t entry, void *closure)
1654 {
1655 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1656 
1657 	ssc->count++;
1658 	ssc->size += entry->end - entry->start;
1659 }
1660 
1661 void
1662 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1663     int flags)
1664 {
1665 	seginfo->count = 0;
1666 	seginfo->size = 0;
1667 
1668 	each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1669 }
1670 
1671 /*
1672  * For each writable segment in the process's memory map, call the given
1673  * function with a pointer to the map entry and some arbitrary
1674  * caller-supplied data.
1675  */
1676 static void
1677 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1678     int flags)
1679 {
1680 	struct proc *p = td->td_proc;
1681 	vm_map_t map = &p->p_vmspace->vm_map;
1682 	vm_map_entry_t entry;
1683 	vm_object_t backing_object, object;
1684 	bool ignore_entry;
1685 
1686 	vm_map_lock_read(map);
1687 	VM_MAP_ENTRY_FOREACH(entry, map) {
1688 		/*
1689 		 * Don't dump inaccessible mappings, deal with legacy
1690 		 * coredump mode.
1691 		 *
1692 		 * Note that read-only segments related to the elf binary
1693 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1694 		 * need to arbitrarily ignore such segments.
1695 		 */
1696 		if ((flags & SVC_ALL) == 0) {
1697 			if (elf_legacy_coredump) {
1698 				if ((entry->protection & VM_PROT_RW) !=
1699 				    VM_PROT_RW)
1700 					continue;
1701 			} else {
1702 				if ((entry->protection & VM_PROT_ALL) == 0)
1703 					continue;
1704 			}
1705 		}
1706 
1707 		/*
1708 		 * Dont include memory segment in the coredump if
1709 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1710 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1711 		 * kernel map).
1712 		 */
1713 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1714 			continue;
1715 		if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1716 		    (flags & SVC_ALL) == 0)
1717 			continue;
1718 		if ((object = entry->object.vm_object) == NULL)
1719 			continue;
1720 
1721 		/* Ignore memory-mapped devices and such things. */
1722 		VM_OBJECT_RLOCK(object);
1723 		while ((backing_object = object->backing_object) != NULL) {
1724 			VM_OBJECT_RLOCK(backing_object);
1725 			VM_OBJECT_RUNLOCK(object);
1726 			object = backing_object;
1727 		}
1728 		ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1729 		VM_OBJECT_RUNLOCK(object);
1730 		if (ignore_entry)
1731 			continue;
1732 
1733 		(*func)(entry, closure);
1734 	}
1735 	vm_map_unlock_read(map);
1736 }
1737 
1738 /*
1739  * Write the core file header to the file, including padding up to
1740  * the page boundary.
1741  */
1742 static int
1743 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1744     size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1745     int flags)
1746 {
1747 	struct note_info *ninfo;
1748 	struct sbuf *sb;
1749 	int error;
1750 
1751 	/* Fill in the header. */
1752 	bzero(hdr, hdrsize);
1753 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1754 
1755 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1756 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1757 	sbuf_start_section(sb, NULL);
1758 	sbuf_bcat(sb, hdr, hdrsize);
1759 	TAILQ_FOREACH(ninfo, notelst, link)
1760 	    __elfN(putnote)(p->td, ninfo, sb);
1761 	/* Align up to a page boundary for the program segments. */
1762 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1763 	error = sbuf_finish(sb);
1764 	sbuf_delete(sb);
1765 
1766 	return (error);
1767 }
1768 
1769 void
1770 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1771     size_t *sizep)
1772 {
1773 	struct proc *p;
1774 	struct thread *thr;
1775 	size_t size;
1776 
1777 	p = td->td_proc;
1778 	size = 0;
1779 
1780 	size += __elfN(register_note)(td, list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1781 
1782 	/*
1783 	 * To have the debugger select the right thread (LWP) as the initial
1784 	 * thread, we dump the state of the thread passed to us in td first.
1785 	 * This is the thread that causes the core dump and thus likely to
1786 	 * be the right thread one wants to have selected in the debugger.
1787 	 */
1788 	thr = td;
1789 	while (thr != NULL) {
1790 		size += __elfN(register_note)(td, list, NT_PRSTATUS,
1791 		    __elfN(note_prstatus), thr);
1792 		size += __elfN(register_note)(td, list, NT_FPREGSET,
1793 		    __elfN(note_fpregset), thr);
1794 		size += __elfN(register_note)(td, list, NT_THRMISC,
1795 		    __elfN(note_thrmisc), thr);
1796 		size += __elfN(register_note)(td, list, NT_PTLWPINFO,
1797 		    __elfN(note_ptlwpinfo), thr);
1798 		size += __elfN(register_note)(td, list, -1,
1799 		    __elfN(note_threadmd), thr);
1800 
1801 		thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1802 		    TAILQ_NEXT(thr, td_plist);
1803 		if (thr == td)
1804 			thr = TAILQ_NEXT(thr, td_plist);
1805 	}
1806 
1807 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1808 	    __elfN(note_procstat_proc), p);
1809 	size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1810 	    note_procstat_files, p);
1811 	size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1812 	    note_procstat_vmmap, p);
1813 	size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1814 	    note_procstat_groups, p);
1815 	size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1816 	    note_procstat_umask, p);
1817 	size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1818 	    note_procstat_rlimit, p);
1819 	size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1820 	    note_procstat_osrel, p);
1821 	size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1822 	    __elfN(note_procstat_psstrings), p);
1823 	size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1824 	    __elfN(note_procstat_auxv), p);
1825 
1826 	*sizep = size;
1827 }
1828 
1829 void
1830 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1831     size_t notesz, int flags)
1832 {
1833 	Elf_Ehdr *ehdr;
1834 	Elf_Phdr *phdr;
1835 	Elf_Shdr *shdr;
1836 	struct phdr_closure phc;
1837 	Elf_Brandinfo *bi;
1838 
1839 	ehdr = (Elf_Ehdr *)hdr;
1840 	bi = td->td_proc->p_elf_brandinfo;
1841 
1842 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1843 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1844 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1845 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1846 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1847 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1848 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1849 	ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1850 	ehdr->e_ident[EI_ABIVERSION] = 0;
1851 	ehdr->e_ident[EI_PAD] = 0;
1852 	ehdr->e_type = ET_CORE;
1853 	ehdr->e_machine = bi->machine;
1854 	ehdr->e_version = EV_CURRENT;
1855 	ehdr->e_entry = 0;
1856 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1857 	ehdr->e_flags = td->td_proc->p_elf_flags;
1858 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1859 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1860 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1861 	ehdr->e_shstrndx = SHN_UNDEF;
1862 	if (numsegs + 1 < PN_XNUM) {
1863 		ehdr->e_phnum = numsegs + 1;
1864 		ehdr->e_shnum = 0;
1865 	} else {
1866 		ehdr->e_phnum = PN_XNUM;
1867 		ehdr->e_shnum = 1;
1868 
1869 		ehdr->e_shoff = ehdr->e_phoff +
1870 		    (numsegs + 1) * ehdr->e_phentsize;
1871 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1872 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1873 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1874 
1875 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1876 		memset(shdr, 0, sizeof(*shdr));
1877 		/*
1878 		 * A special first section is used to hold large segment and
1879 		 * section counts.  This was proposed by Sun Microsystems in
1880 		 * Solaris and has been adopted by Linux; the standard ELF
1881 		 * tools are already familiar with the technique.
1882 		 *
1883 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1884 		 * (or 12-7 depending on the version of the document) for more
1885 		 * details.
1886 		 */
1887 		shdr->sh_type = SHT_NULL;
1888 		shdr->sh_size = ehdr->e_shnum;
1889 		shdr->sh_link = ehdr->e_shstrndx;
1890 		shdr->sh_info = numsegs + 1;
1891 	}
1892 
1893 	/*
1894 	 * Fill in the program header entries.
1895 	 */
1896 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1897 
1898 	/* The note segement. */
1899 	phdr->p_type = PT_NOTE;
1900 	phdr->p_offset = hdrsize;
1901 	phdr->p_vaddr = 0;
1902 	phdr->p_paddr = 0;
1903 	phdr->p_filesz = notesz;
1904 	phdr->p_memsz = 0;
1905 	phdr->p_flags = PF_R;
1906 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1907 	phdr++;
1908 
1909 	/* All the writable segments from the program. */
1910 	phc.phdr = phdr;
1911 	phc.offset = round_page(hdrsize + notesz);
1912 	each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1913 }
1914 
1915 size_t
1916 __elfN(register_note)(struct thread *td, struct note_info_list *list,
1917     int type, outfunc_t out, void *arg)
1918 {
1919 	const struct sysentvec *sv;
1920 	struct note_info *ninfo;
1921 	size_t size, notesize;
1922 
1923 	sv = td->td_proc->p_sysent;
1924 	size = 0;
1925 	out(arg, NULL, &size);
1926 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1927 	ninfo->type = type;
1928 	ninfo->outfunc = out;
1929 	ninfo->outarg = arg;
1930 	ninfo->outsize = size;
1931 	TAILQ_INSERT_TAIL(list, ninfo, link);
1932 
1933 	if (type == -1)
1934 		return (size);
1935 
1936 	notesize = sizeof(Elf_Note) +		/* note header */
1937 	    roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1938 						/* note name */
1939 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1940 
1941 	return (notesize);
1942 }
1943 
1944 static size_t
1945 append_note_data(const void *src, void *dst, size_t len)
1946 {
1947 	size_t padded_len;
1948 
1949 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1950 	if (dst != NULL) {
1951 		bcopy(src, dst, len);
1952 		bzero((char *)dst + len, padded_len - len);
1953 	}
1954 	return (padded_len);
1955 }
1956 
1957 size_t
1958 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1959 {
1960 	Elf_Note *note;
1961 	char *buf;
1962 	size_t notesize;
1963 
1964 	buf = dst;
1965 	if (buf != NULL) {
1966 		note = (Elf_Note *)buf;
1967 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1968 		note->n_descsz = size;
1969 		note->n_type = type;
1970 		buf += sizeof(*note);
1971 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1972 		    sizeof(FREEBSD_ABI_VENDOR));
1973 		append_note_data(src, buf, size);
1974 		if (descp != NULL)
1975 			*descp = buf;
1976 	}
1977 
1978 	notesize = sizeof(Elf_Note) +		/* note header */
1979 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1980 						/* note name */
1981 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1982 
1983 	return (notesize);
1984 }
1985 
1986 static void
1987 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
1988 {
1989 	Elf_Note note;
1990 	const struct sysentvec *sv;
1991 	ssize_t old_len, sect_len;
1992 	size_t new_len, descsz, i;
1993 
1994 	if (ninfo->type == -1) {
1995 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1996 		return;
1997 	}
1998 
1999 	sv = td->td_proc->p_sysent;
2000 
2001 	note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2002 	note.n_descsz = ninfo->outsize;
2003 	note.n_type = ninfo->type;
2004 
2005 	sbuf_bcat(sb, &note, sizeof(note));
2006 	sbuf_start_section(sb, &old_len);
2007 	sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2008 	    strlen(sv->sv_elf_core_abi_vendor) + 1);
2009 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2010 	if (note.n_descsz == 0)
2011 		return;
2012 	sbuf_start_section(sb, &old_len);
2013 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2014 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2015 	if (sect_len < 0)
2016 		return;
2017 
2018 	new_len = (size_t)sect_len;
2019 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2020 	if (new_len < descsz) {
2021 		/*
2022 		 * It is expected that individual note emitters will correctly
2023 		 * predict their expected output size and fill up to that size
2024 		 * themselves, padding in a format-specific way if needed.
2025 		 * However, in case they don't, just do it here with zeros.
2026 		 */
2027 		for (i = 0; i < descsz - new_len; i++)
2028 			sbuf_putc(sb, 0);
2029 	} else if (new_len > descsz) {
2030 		/*
2031 		 * We can't always truncate sb -- we may have drained some
2032 		 * of it already.
2033 		 */
2034 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2035 		    "read it (%zu > %zu).  Since it is longer than "
2036 		    "expected, this coredump's notes are corrupt.  THIS "
2037 		    "IS A BUG in the note_procstat routine for type %u.\n",
2038 		    __func__, (unsigned)note.n_type, new_len, descsz,
2039 		    (unsigned)note.n_type));
2040 	}
2041 }
2042 
2043 /*
2044  * Miscellaneous note out functions.
2045  */
2046 
2047 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2048 #include <compat/freebsd32/freebsd32.h>
2049 #include <compat/freebsd32/freebsd32_signal.h>
2050 
2051 typedef struct prstatus32 elf_prstatus_t;
2052 typedef struct prpsinfo32 elf_prpsinfo_t;
2053 typedef struct fpreg32 elf_prfpregset_t;
2054 typedef struct fpreg32 elf_fpregset_t;
2055 typedef struct reg32 elf_gregset_t;
2056 typedef struct thrmisc32 elf_thrmisc_t;
2057 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2058 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2059 typedef uint32_t elf_ps_strings_t;
2060 #else
2061 typedef prstatus_t elf_prstatus_t;
2062 typedef prpsinfo_t elf_prpsinfo_t;
2063 typedef prfpregset_t elf_prfpregset_t;
2064 typedef prfpregset_t elf_fpregset_t;
2065 typedef gregset_t elf_gregset_t;
2066 typedef thrmisc_t elf_thrmisc_t;
2067 #define ELF_KERN_PROC_MASK	0
2068 typedef struct kinfo_proc elf_kinfo_proc_t;
2069 typedef vm_offset_t elf_ps_strings_t;
2070 #endif
2071 
2072 static void
2073 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2074 {
2075 	struct sbuf sbarg;
2076 	size_t len;
2077 	char *cp, *end;
2078 	struct proc *p;
2079 	elf_prpsinfo_t *psinfo;
2080 	int error;
2081 
2082 	p = arg;
2083 	if (sb != NULL) {
2084 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2085 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2086 		psinfo->pr_version = PRPSINFO_VERSION;
2087 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2088 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2089 		PROC_LOCK(p);
2090 		if (p->p_args != NULL) {
2091 			len = sizeof(psinfo->pr_psargs) - 1;
2092 			if (len > p->p_args->ar_length)
2093 				len = p->p_args->ar_length;
2094 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2095 			PROC_UNLOCK(p);
2096 			error = 0;
2097 		} else {
2098 			_PHOLD(p);
2099 			PROC_UNLOCK(p);
2100 			sbuf_new(&sbarg, psinfo->pr_psargs,
2101 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2102 			error = proc_getargv(curthread, p, &sbarg);
2103 			PRELE(p);
2104 			if (sbuf_finish(&sbarg) == 0)
2105 				len = sbuf_len(&sbarg) - 1;
2106 			else
2107 				len = sizeof(psinfo->pr_psargs) - 1;
2108 			sbuf_delete(&sbarg);
2109 		}
2110 		if (error || len == 0)
2111 			strlcpy(psinfo->pr_psargs, p->p_comm,
2112 			    sizeof(psinfo->pr_psargs));
2113 		else {
2114 			KASSERT(len < sizeof(psinfo->pr_psargs),
2115 			    ("len is too long: %zu vs %zu", len,
2116 			    sizeof(psinfo->pr_psargs)));
2117 			cp = psinfo->pr_psargs;
2118 			end = cp + len - 1;
2119 			for (;;) {
2120 				cp = memchr(cp, '\0', end - cp);
2121 				if (cp == NULL)
2122 					break;
2123 				*cp = ' ';
2124 			}
2125 		}
2126 		psinfo->pr_pid = p->p_pid;
2127 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2128 		free(psinfo, M_TEMP);
2129 	}
2130 	*sizep = sizeof(*psinfo);
2131 }
2132 
2133 static void
2134 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2135 {
2136 	struct thread *td;
2137 	elf_prstatus_t *status;
2138 
2139 	td = arg;
2140 	if (sb != NULL) {
2141 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2142 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2143 		status->pr_version = PRSTATUS_VERSION;
2144 		status->pr_statussz = sizeof(elf_prstatus_t);
2145 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2146 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2147 		status->pr_osreldate = osreldate;
2148 		status->pr_cursig = td->td_proc->p_sig;
2149 		status->pr_pid = td->td_tid;
2150 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2151 		fill_regs32(td, &status->pr_reg);
2152 #else
2153 		fill_regs(td, &status->pr_reg);
2154 #endif
2155 		sbuf_bcat(sb, status, sizeof(*status));
2156 		free(status, M_TEMP);
2157 	}
2158 	*sizep = sizeof(*status);
2159 }
2160 
2161 static void
2162 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2163 {
2164 	struct thread *td;
2165 	elf_prfpregset_t *fpregset;
2166 
2167 	td = arg;
2168 	if (sb != NULL) {
2169 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2170 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2171 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2172 		fill_fpregs32(td, fpregset);
2173 #else
2174 		fill_fpregs(td, fpregset);
2175 #endif
2176 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2177 		free(fpregset, M_TEMP);
2178 	}
2179 	*sizep = sizeof(*fpregset);
2180 }
2181 
2182 static void
2183 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2184 {
2185 	struct thread *td;
2186 	elf_thrmisc_t thrmisc;
2187 
2188 	td = arg;
2189 	if (sb != NULL) {
2190 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2191 		bzero(&thrmisc, sizeof(thrmisc));
2192 		strcpy(thrmisc.pr_tname, td->td_name);
2193 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2194 	}
2195 	*sizep = sizeof(thrmisc);
2196 }
2197 
2198 static void
2199 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2200 {
2201 	struct thread *td;
2202 	size_t size;
2203 	int structsize;
2204 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2205 	struct ptrace_lwpinfo32 pl;
2206 #else
2207 	struct ptrace_lwpinfo pl;
2208 #endif
2209 
2210 	td = arg;
2211 	size = sizeof(structsize) + sizeof(pl);
2212 	if (sb != NULL) {
2213 		KASSERT(*sizep == size, ("invalid size"));
2214 		structsize = sizeof(pl);
2215 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2216 		bzero(&pl, sizeof(pl));
2217 		pl.pl_lwpid = td->td_tid;
2218 		pl.pl_event = PL_EVENT_NONE;
2219 		pl.pl_sigmask = td->td_sigmask;
2220 		pl.pl_siglist = td->td_siglist;
2221 		if (td->td_si.si_signo != 0) {
2222 			pl.pl_event = PL_EVENT_SIGNAL;
2223 			pl.pl_flags |= PL_FLAG_SI;
2224 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2225 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2226 #else
2227 			pl.pl_siginfo = td->td_si;
2228 #endif
2229 		}
2230 		strcpy(pl.pl_tdname, td->td_name);
2231 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2232 		sbuf_bcat(sb, &pl, sizeof(pl));
2233 	}
2234 	*sizep = size;
2235 }
2236 
2237 /*
2238  * Allow for MD specific notes, as well as any MD
2239  * specific preparations for writing MI notes.
2240  */
2241 static void
2242 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2243 {
2244 	struct thread *td;
2245 	void *buf;
2246 	size_t size;
2247 
2248 	td = (struct thread *)arg;
2249 	size = *sizep;
2250 	if (size != 0 && sb != NULL)
2251 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2252 	else
2253 		buf = NULL;
2254 	size = 0;
2255 	__elfN(dump_thread)(td, buf, &size);
2256 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2257 	if (size != 0 && sb != NULL)
2258 		sbuf_bcat(sb, buf, size);
2259 	free(buf, M_TEMP);
2260 	*sizep = size;
2261 }
2262 
2263 #ifdef KINFO_PROC_SIZE
2264 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2265 #endif
2266 
2267 static void
2268 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2269 {
2270 	struct proc *p;
2271 	size_t size;
2272 	int structsize;
2273 
2274 	p = arg;
2275 	size = sizeof(structsize) + p->p_numthreads *
2276 	    sizeof(elf_kinfo_proc_t);
2277 
2278 	if (sb != NULL) {
2279 		KASSERT(*sizep == size, ("invalid size"));
2280 		structsize = sizeof(elf_kinfo_proc_t);
2281 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2282 		sx_slock(&proctree_lock);
2283 		PROC_LOCK(p);
2284 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2285 		sx_sunlock(&proctree_lock);
2286 	}
2287 	*sizep = size;
2288 }
2289 
2290 #ifdef KINFO_FILE_SIZE
2291 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2292 #endif
2293 
2294 static void
2295 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2296 {
2297 	struct proc *p;
2298 	size_t size, sect_sz, i;
2299 	ssize_t start_len, sect_len;
2300 	int structsize, filedesc_flags;
2301 
2302 	if (coredump_pack_fileinfo)
2303 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2304 	else
2305 		filedesc_flags = 0;
2306 
2307 	p = arg;
2308 	structsize = sizeof(struct kinfo_file);
2309 	if (sb == NULL) {
2310 		size = 0;
2311 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2312 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2313 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2314 		PROC_LOCK(p);
2315 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2316 		sbuf_finish(sb);
2317 		sbuf_delete(sb);
2318 		*sizep = size;
2319 	} else {
2320 		sbuf_start_section(sb, &start_len);
2321 
2322 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2323 		PROC_LOCK(p);
2324 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2325 		    filedesc_flags);
2326 
2327 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2328 		if (sect_len < 0)
2329 			return;
2330 		sect_sz = sect_len;
2331 
2332 		KASSERT(sect_sz <= *sizep,
2333 		    ("kern_proc_filedesc_out did not respect maxlen; "
2334 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2335 		     sect_sz - sizeof(structsize)));
2336 
2337 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2338 			sbuf_putc(sb, 0);
2339 	}
2340 }
2341 
2342 #ifdef KINFO_VMENTRY_SIZE
2343 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2344 #endif
2345 
2346 static void
2347 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2348 {
2349 	struct proc *p;
2350 	size_t size;
2351 	int structsize, vmmap_flags;
2352 
2353 	if (coredump_pack_vmmapinfo)
2354 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2355 	else
2356 		vmmap_flags = 0;
2357 
2358 	p = arg;
2359 	structsize = sizeof(struct kinfo_vmentry);
2360 	if (sb == NULL) {
2361 		size = 0;
2362 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2363 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2364 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2365 		PROC_LOCK(p);
2366 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2367 		sbuf_finish(sb);
2368 		sbuf_delete(sb);
2369 		*sizep = size;
2370 	} else {
2371 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2372 		PROC_LOCK(p);
2373 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2374 		    vmmap_flags);
2375 	}
2376 }
2377 
2378 static void
2379 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2380 {
2381 	struct proc *p;
2382 	size_t size;
2383 	int structsize;
2384 
2385 	p = arg;
2386 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2387 	if (sb != NULL) {
2388 		KASSERT(*sizep == size, ("invalid size"));
2389 		structsize = sizeof(gid_t);
2390 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2391 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2392 		    sizeof(gid_t));
2393 	}
2394 	*sizep = size;
2395 }
2396 
2397 static void
2398 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2399 {
2400 	struct proc *p;
2401 	size_t size;
2402 	int structsize;
2403 
2404 	p = arg;
2405 	size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2406 	if (sb != NULL) {
2407 		KASSERT(*sizep == size, ("invalid size"));
2408 		structsize = sizeof(p->p_pd->pd_cmask);
2409 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2410 		sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2411 	}
2412 	*sizep = size;
2413 }
2414 
2415 static void
2416 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2417 {
2418 	struct proc *p;
2419 	struct rlimit rlim[RLIM_NLIMITS];
2420 	size_t size;
2421 	int structsize, i;
2422 
2423 	p = arg;
2424 	size = sizeof(structsize) + sizeof(rlim);
2425 	if (sb != NULL) {
2426 		KASSERT(*sizep == size, ("invalid size"));
2427 		structsize = sizeof(rlim);
2428 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2429 		PROC_LOCK(p);
2430 		for (i = 0; i < RLIM_NLIMITS; i++)
2431 			lim_rlimit_proc(p, i, &rlim[i]);
2432 		PROC_UNLOCK(p);
2433 		sbuf_bcat(sb, rlim, sizeof(rlim));
2434 	}
2435 	*sizep = size;
2436 }
2437 
2438 static void
2439 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2440 {
2441 	struct proc *p;
2442 	size_t size;
2443 	int structsize;
2444 
2445 	p = arg;
2446 	size = sizeof(structsize) + sizeof(p->p_osrel);
2447 	if (sb != NULL) {
2448 		KASSERT(*sizep == size, ("invalid size"));
2449 		structsize = sizeof(p->p_osrel);
2450 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2451 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2452 	}
2453 	*sizep = size;
2454 }
2455 
2456 static void
2457 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2458 {
2459 	struct proc *p;
2460 	elf_ps_strings_t ps_strings;
2461 	size_t size;
2462 	int structsize;
2463 
2464 	p = arg;
2465 	size = sizeof(structsize) + sizeof(ps_strings);
2466 	if (sb != NULL) {
2467 		KASSERT(*sizep == size, ("invalid size"));
2468 		structsize = sizeof(ps_strings);
2469 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2470 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2471 #else
2472 		ps_strings = p->p_sysent->sv_psstrings;
2473 #endif
2474 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2475 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2476 	}
2477 	*sizep = size;
2478 }
2479 
2480 static void
2481 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2482 {
2483 	struct proc *p;
2484 	size_t size;
2485 	int structsize;
2486 
2487 	p = arg;
2488 	if (sb == NULL) {
2489 		size = 0;
2490 		sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2491 		    SBUF_FIXEDLEN);
2492 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2493 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2494 		PHOLD(p);
2495 		proc_getauxv(curthread, p, sb);
2496 		PRELE(p);
2497 		sbuf_finish(sb);
2498 		sbuf_delete(sb);
2499 		*sizep = size;
2500 	} else {
2501 		structsize = sizeof(Elf_Auxinfo);
2502 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2503 		PHOLD(p);
2504 		proc_getauxv(curthread, p, sb);
2505 		PRELE(p);
2506 	}
2507 }
2508 
2509 static bool
2510 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2511     const char *note_vendor, const Elf_Phdr *pnote,
2512     bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2513 {
2514 	const Elf_Note *note, *note0, *note_end;
2515 	const char *note_name;
2516 	char *buf;
2517 	int i, error;
2518 	bool res;
2519 
2520 	/* We need some limit, might as well use PAGE_SIZE. */
2521 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2522 		return (false);
2523 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2524 	if (pnote->p_offset > PAGE_SIZE ||
2525 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2526 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2527 		if (buf == NULL) {
2528 			VOP_UNLOCK(imgp->vp);
2529 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2530 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2531 		}
2532 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2533 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2534 		    curthread->td_ucred, NOCRED, NULL, curthread);
2535 		if (error != 0) {
2536 			uprintf("i/o error PT_NOTE\n");
2537 			goto retf;
2538 		}
2539 		note = note0 = (const Elf_Note *)buf;
2540 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2541 	} else {
2542 		note = note0 = (const Elf_Note *)(imgp->image_header +
2543 		    pnote->p_offset);
2544 		note_end = (const Elf_Note *)(imgp->image_header +
2545 		    pnote->p_offset + pnote->p_filesz);
2546 		buf = NULL;
2547 	}
2548 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2549 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2550 		    (const char *)note < sizeof(Elf_Note)) {
2551 			goto retf;
2552 		}
2553 		if (note->n_namesz != checknote->n_namesz ||
2554 		    note->n_descsz != checknote->n_descsz ||
2555 		    note->n_type != checknote->n_type)
2556 			goto nextnote;
2557 		note_name = (const char *)(note + 1);
2558 		if (note_name + checknote->n_namesz >=
2559 		    (const char *)note_end || strncmp(note_vendor,
2560 		    note_name, checknote->n_namesz) != 0)
2561 			goto nextnote;
2562 
2563 		if (cb(note, cb_arg, &res))
2564 			goto ret;
2565 nextnote:
2566 		note = (const Elf_Note *)((const char *)(note + 1) +
2567 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2568 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2569 	}
2570 retf:
2571 	res = false;
2572 ret:
2573 	free(buf, M_TEMP);
2574 	return (res);
2575 }
2576 
2577 struct brandnote_cb_arg {
2578 	Elf_Brandnote *brandnote;
2579 	int32_t *osrel;
2580 };
2581 
2582 static bool
2583 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2584 {
2585 	struct brandnote_cb_arg *arg;
2586 
2587 	arg = arg0;
2588 
2589 	/*
2590 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2591 	 * necessary.
2592 	 */
2593 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2594 	    arg->brandnote->trans_osrel != NULL ?
2595 	    arg->brandnote->trans_osrel(note, arg->osrel) : true;
2596 
2597 	return (true);
2598 }
2599 
2600 static Elf_Note fctl_note = {
2601 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2602 	.n_descsz = sizeof(uint32_t),
2603 	.n_type = NT_FREEBSD_FEATURE_CTL,
2604 };
2605 
2606 struct fctl_cb_arg {
2607 	bool *has_fctl0;
2608 	uint32_t *fctl0;
2609 };
2610 
2611 static bool
2612 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2613 {
2614 	struct fctl_cb_arg *arg;
2615 	const Elf32_Word *desc;
2616 	uintptr_t p;
2617 
2618 	arg = arg0;
2619 	p = (uintptr_t)(note + 1);
2620 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2621 	desc = (const Elf32_Word *)p;
2622 	*arg->has_fctl0 = true;
2623 	*arg->fctl0 = desc[0];
2624 	*res = true;
2625 	return (true);
2626 }
2627 
2628 /*
2629  * Try to find the appropriate ABI-note section for checknote, fetch
2630  * the osreldate and feature control flags for binary from the ELF
2631  * OSABI-note.  Only the first page of the image is searched, the same
2632  * as for headers.
2633  */
2634 static bool
2635 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2636     int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2637 {
2638 	const Elf_Phdr *phdr;
2639 	const Elf_Ehdr *hdr;
2640 	struct brandnote_cb_arg b_arg;
2641 	struct fctl_cb_arg f_arg;
2642 	int i, j;
2643 
2644 	hdr = (const Elf_Ehdr *)imgp->image_header;
2645 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2646 	b_arg.brandnote = brandnote;
2647 	b_arg.osrel = osrel;
2648 	f_arg.has_fctl0 = has_fctl0;
2649 	f_arg.fctl0 = fctl0;
2650 
2651 	for (i = 0; i < hdr->e_phnum; i++) {
2652 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2653 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2654 		    &b_arg)) {
2655 			for (j = 0; j < hdr->e_phnum; j++) {
2656 				if (phdr[j].p_type == PT_NOTE &&
2657 				    __elfN(parse_notes)(imgp, &fctl_note,
2658 				    FREEBSD_ABI_VENDOR, &phdr[j],
2659 				    note_fctl_cb, &f_arg))
2660 					break;
2661 			}
2662 			return (true);
2663 		}
2664 	}
2665 	return (false);
2666 
2667 }
2668 
2669 /*
2670  * Tell kern_execve.c about it, with a little help from the linker.
2671  */
2672 static struct execsw __elfN(execsw) = {
2673 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2674 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2675 };
2676 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2677 
2678 static vm_prot_t
2679 __elfN(trans_prot)(Elf_Word flags)
2680 {
2681 	vm_prot_t prot;
2682 
2683 	prot = 0;
2684 	if (flags & PF_X)
2685 		prot |= VM_PROT_EXECUTE;
2686 	if (flags & PF_W)
2687 		prot |= VM_PROT_WRITE;
2688 	if (flags & PF_R)
2689 		prot |= VM_PROT_READ;
2690 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2691 	if (i386_read_exec && (flags & PF_R))
2692 		prot |= VM_PROT_EXECUTE;
2693 #endif
2694 	return (prot);
2695 }
2696 
2697 static Elf_Word
2698 __elfN(untrans_prot)(vm_prot_t prot)
2699 {
2700 	Elf_Word flags;
2701 
2702 	flags = 0;
2703 	if (prot & VM_PROT_EXECUTE)
2704 		flags |= PF_X;
2705 	if (prot & VM_PROT_READ)
2706 		flags |= PF_R;
2707 	if (prot & VM_PROT_WRITE)
2708 		flags |= PF_W;
2709 	return (flags);
2710 }
2711 
2712 vm_size_t
2713 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2714 {
2715 	uintptr_t range, rbase, gap;
2716 	int pct;
2717 
2718 	pct = __elfN(aslr_stack_gap);
2719 	if (pct == 0)
2720 		return (0);
2721 	if (pct > 50)
2722 		pct = 50;
2723 	range = imgp->eff_stack_sz * pct / 100;
2724 	arc4rand(&rbase, sizeof(rbase), 0);
2725 	gap = rbase % range;
2726 	gap &= ~(sizeof(u_long) - 1);
2727 	*stack_base -= gap;
2728 	return (gap);
2729 }
2730