xref: /freebsd/sys/kern/imgact_elf.c (revision f6a3b357e9be4c6423c85eff9a847163a0d307c8)
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/pioctl.h>
54 #include <sys/proc.h>
55 #include <sys/procfs.h>
56 #include <sys/ptrace.h>
57 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sbuf.h>
61 #include <sys/sf_buf.h>
62 #include <sys/smp.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
65 #include <sys/stat.h>
66 #include <sys/sx.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
73 #include <sys/user.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
85 
86 #define ELF_NOTE_ROUNDSIZE	4
87 #define OLD_EI_BRAND	8
88 
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91     const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93     u_long *entry);
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98     int32_t *osrel);
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101     Elf_Brandnote *checknote, int32_t *osrel, uint32_t *fctl0);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
106     "");
107 
108 #define	CORE_BUF_SIZE	(16 * 1024)
109 
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117     &elf_legacy_coredump, 0,
118     "include all and only RW pages in core dumps");
119 
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
123     defined(__riscv)
124 	1;
125 #else
126 	0;
127 #endif
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 
132 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
133 int i386_read_exec = 0;
134 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
135     "enable execution from readable segments");
136 #endif
137 
138 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
139     "");
140 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
141 
142 static int __elfN(aslr_enabled) = 0;
143 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
144     &__elfN(aslr_enabled), 0,
145     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
146     ": enable address map randomization");
147 
148 static int __elfN(pie_aslr_enabled) = 0;
149 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
150     &__elfN(pie_aslr_enabled), 0,
151     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
152     ": enable address map randomization for PIE binaries");
153 
154 static int __elfN(aslr_honor_sbrk) = 1;
155 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
156     &__elfN(aslr_honor_sbrk), 0,
157     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
158 
159 static int __elfN(aslr_stack_gap) = 3;
160 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
161     &__elfN(aslr_stack_gap), 0,
162     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
163     ": maximum percentage of main stack to waste on a random gap");
164 
165 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
166 
167 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
168 
169 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
170 
171 Elf_Brandnote __elfN(freebsd_brandnote) = {
172 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
173 	.hdr.n_descsz	= sizeof(int32_t),
174 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
175 	.vendor		= FREEBSD_ABI_VENDOR,
176 	.flags		= BN_TRANSLATE_OSREL,
177 	.trans_osrel	= __elfN(freebsd_trans_osrel)
178 };
179 
180 static bool
181 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
182 {
183 	uintptr_t p;
184 
185 	p = (uintptr_t)(note + 1);
186 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
187 	*osrel = *(const int32_t *)(p);
188 
189 	return (true);
190 }
191 
192 static const char GNU_ABI_VENDOR[] = "GNU";
193 static int GNU_KFREEBSD_ABI_DESC = 3;
194 
195 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
196 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
197 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
198 	.hdr.n_type	= 1,
199 	.vendor		= GNU_ABI_VENDOR,
200 	.flags		= BN_TRANSLATE_OSREL,
201 	.trans_osrel	= kfreebsd_trans_osrel
202 };
203 
204 static bool
205 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
206 {
207 	const Elf32_Word *desc;
208 	uintptr_t p;
209 
210 	p = (uintptr_t)(note + 1);
211 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
212 
213 	desc = (const Elf32_Word *)p;
214 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
215 		return (false);
216 
217 	/*
218 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
219 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
220 	 */
221 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
222 
223 	return (true);
224 }
225 
226 int
227 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
228 {
229 	int i;
230 
231 	for (i = 0; i < MAX_BRANDS; i++) {
232 		if (elf_brand_list[i] == NULL) {
233 			elf_brand_list[i] = entry;
234 			break;
235 		}
236 	}
237 	if (i == MAX_BRANDS) {
238 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
239 			__func__, entry);
240 		return (-1);
241 	}
242 	return (0);
243 }
244 
245 int
246 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
247 {
248 	int i;
249 
250 	for (i = 0; i < MAX_BRANDS; i++) {
251 		if (elf_brand_list[i] == entry) {
252 			elf_brand_list[i] = NULL;
253 			break;
254 		}
255 	}
256 	if (i == MAX_BRANDS)
257 		return (-1);
258 	return (0);
259 }
260 
261 int
262 __elfN(brand_inuse)(Elf_Brandinfo *entry)
263 {
264 	struct proc *p;
265 	int rval = FALSE;
266 
267 	sx_slock(&allproc_lock);
268 	FOREACH_PROC_IN_SYSTEM(p) {
269 		if (p->p_sysent == entry->sysvec) {
270 			rval = TRUE;
271 			break;
272 		}
273 	}
274 	sx_sunlock(&allproc_lock);
275 
276 	return (rval);
277 }
278 
279 static Elf_Brandinfo *
280 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
281     int32_t *osrel, uint32_t *fctl0)
282 {
283 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
284 	Elf_Brandinfo *bi, *bi_m;
285 	boolean_t ret;
286 	int i, interp_name_len;
287 
288 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
289 
290 	/*
291 	 * We support four types of branding -- (1) the ELF EI_OSABI field
292 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
293 	 * branding w/in the ELF header, (3) path of the `interp_path'
294 	 * field, and (4) the ".note.ABI-tag" ELF section.
295 	 */
296 
297 	/* Look for an ".note.ABI-tag" ELF section */
298 	bi_m = NULL;
299 	for (i = 0; i < MAX_BRANDS; i++) {
300 		bi = elf_brand_list[i];
301 		if (bi == NULL)
302 			continue;
303 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
304 			continue;
305 		if (hdr->e_machine == bi->machine && (bi->flags &
306 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
307 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
308 			    fctl0);
309 			/* Give brand a chance to veto check_note's guess */
310 			if (ret && bi->header_supported)
311 				ret = bi->header_supported(imgp);
312 			/*
313 			 * If note checker claimed the binary, but the
314 			 * interpreter path in the image does not
315 			 * match default one for the brand, try to
316 			 * search for other brands with the same
317 			 * interpreter.  Either there is better brand
318 			 * with the right interpreter, or, failing
319 			 * this, we return first brand which accepted
320 			 * our note and, optionally, header.
321 			 */
322 			if (ret && bi_m == NULL && interp != NULL &&
323 			    (bi->interp_path == NULL ||
324 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
325 			    strncmp(interp, bi->interp_path, interp_name_len)
326 			    != 0))) {
327 				bi_m = bi;
328 				ret = 0;
329 			}
330 			if (ret)
331 				return (bi);
332 		}
333 	}
334 	if (bi_m != NULL)
335 		return (bi_m);
336 
337 	/* If the executable has a brand, search for it in the brand list. */
338 	for (i = 0; i < MAX_BRANDS; i++) {
339 		bi = elf_brand_list[i];
340 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
341 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
342 			continue;
343 		if (hdr->e_machine == bi->machine &&
344 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
345 		    (bi->compat_3_brand != NULL &&
346 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
347 		    bi->compat_3_brand) == 0))) {
348 			/* Looks good, but give brand a chance to veto */
349 			if (bi->header_supported == NULL ||
350 			    bi->header_supported(imgp)) {
351 				/*
352 				 * Again, prefer strictly matching
353 				 * interpreter path.
354 				 */
355 				if (interp_name_len == 0 &&
356 				    bi->interp_path == NULL)
357 					return (bi);
358 				if (bi->interp_path != NULL &&
359 				    strlen(bi->interp_path) + 1 ==
360 				    interp_name_len && strncmp(interp,
361 				    bi->interp_path, interp_name_len) == 0)
362 					return (bi);
363 				if (bi_m == NULL)
364 					bi_m = bi;
365 			}
366 		}
367 	}
368 	if (bi_m != NULL)
369 		return (bi_m);
370 
371 	/* No known brand, see if the header is recognized by any brand */
372 	for (i = 0; i < MAX_BRANDS; i++) {
373 		bi = elf_brand_list[i];
374 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
375 		    bi->header_supported == NULL)
376 			continue;
377 		if (hdr->e_machine == bi->machine) {
378 			ret = bi->header_supported(imgp);
379 			if (ret)
380 				return (bi);
381 		}
382 	}
383 
384 	/* Lacking a known brand, search for a recognized interpreter. */
385 	if (interp != NULL) {
386 		for (i = 0; i < MAX_BRANDS; i++) {
387 			bi = elf_brand_list[i];
388 			if (bi == NULL || (bi->flags &
389 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
390 			    != 0)
391 				continue;
392 			if (hdr->e_machine == bi->machine &&
393 			    bi->interp_path != NULL &&
394 			    /* ELF image p_filesz includes terminating zero */
395 			    strlen(bi->interp_path) + 1 == interp_name_len &&
396 			    strncmp(interp, bi->interp_path, interp_name_len)
397 			    == 0 && (bi->header_supported == NULL ||
398 			    bi->header_supported(imgp)))
399 				return (bi);
400 		}
401 	}
402 
403 	/* Lacking a recognized interpreter, try the default brand */
404 	for (i = 0; i < MAX_BRANDS; i++) {
405 		bi = elf_brand_list[i];
406 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
407 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
408 			continue;
409 		if (hdr->e_machine == bi->machine &&
410 		    __elfN(fallback_brand) == bi->brand &&
411 		    (bi->header_supported == NULL ||
412 		    bi->header_supported(imgp)))
413 			return (bi);
414 	}
415 	return (NULL);
416 }
417 
418 static int
419 __elfN(check_header)(const Elf_Ehdr *hdr)
420 {
421 	Elf_Brandinfo *bi;
422 	int i;
423 
424 	if (!IS_ELF(*hdr) ||
425 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
426 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
427 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
428 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
429 	    hdr->e_version != ELF_TARG_VER)
430 		return (ENOEXEC);
431 
432 	/*
433 	 * Make sure we have at least one brand for this machine.
434 	 */
435 
436 	for (i = 0; i < MAX_BRANDS; i++) {
437 		bi = elf_brand_list[i];
438 		if (bi != NULL && bi->machine == hdr->e_machine)
439 			break;
440 	}
441 	if (i == MAX_BRANDS)
442 		return (ENOEXEC);
443 
444 	return (0);
445 }
446 
447 static int
448 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
449     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
450 {
451 	struct sf_buf *sf;
452 	int error;
453 	vm_offset_t off;
454 
455 	/*
456 	 * Create the page if it doesn't exist yet. Ignore errors.
457 	 */
458 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
459 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
460 
461 	/*
462 	 * Find the page from the underlying object.
463 	 */
464 	if (object != NULL) {
465 		sf = vm_imgact_map_page(object, offset);
466 		if (sf == NULL)
467 			return (KERN_FAILURE);
468 		off = offset - trunc_page(offset);
469 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
470 		    end - start);
471 		vm_imgact_unmap_page(sf);
472 		if (error != 0)
473 			return (KERN_FAILURE);
474 	}
475 
476 	return (KERN_SUCCESS);
477 }
478 
479 static int
480 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
481     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
482     int cow)
483 {
484 	struct sf_buf *sf;
485 	vm_offset_t off;
486 	vm_size_t sz;
487 	int error, locked, rv;
488 
489 	if (start != trunc_page(start)) {
490 		rv = __elfN(map_partial)(map, object, offset, start,
491 		    round_page(start), prot);
492 		if (rv != KERN_SUCCESS)
493 			return (rv);
494 		offset += round_page(start) - start;
495 		start = round_page(start);
496 	}
497 	if (end != round_page(end)) {
498 		rv = __elfN(map_partial)(map, object, offset +
499 		    trunc_page(end) - start, trunc_page(end), end, prot);
500 		if (rv != KERN_SUCCESS)
501 			return (rv);
502 		end = trunc_page(end);
503 	}
504 	if (start >= end)
505 		return (KERN_SUCCESS);
506 	if ((offset & PAGE_MASK) != 0) {
507 		/*
508 		 * The mapping is not page aligned.  This means that we have
509 		 * to copy the data.
510 		 */
511 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
512 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
513 		if (rv != KERN_SUCCESS)
514 			return (rv);
515 		if (object == NULL)
516 			return (KERN_SUCCESS);
517 		for (; start < end; start += sz) {
518 			sf = vm_imgact_map_page(object, offset);
519 			if (sf == NULL)
520 				return (KERN_FAILURE);
521 			off = offset - trunc_page(offset);
522 			sz = end - start;
523 			if (sz > PAGE_SIZE - off)
524 				sz = PAGE_SIZE - off;
525 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
526 			    (caddr_t)start, sz);
527 			vm_imgact_unmap_page(sf);
528 			if (error != 0)
529 				return (KERN_FAILURE);
530 			offset += sz;
531 		}
532 	} else {
533 		vm_object_reference(object);
534 		rv = vm_map_fixed(map, object, offset, start, end - start,
535 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
536 		    (object != NULL ? MAP_VN_EXEC : 0));
537 		if (rv != KERN_SUCCESS) {
538 			locked = VOP_ISLOCKED(imgp->vp);
539 			VOP_UNLOCK(imgp->vp, 0);
540 			vm_object_deallocate(object);
541 			vn_lock(imgp->vp, locked | LK_RETRY);
542 			return (rv);
543 		} else if (object != NULL) {
544 			MPASS(imgp->vp->v_object == object);
545 			VOP_SET_TEXT_CHECKED(imgp->vp);
546 		}
547 	}
548 	return (KERN_SUCCESS);
549 }
550 
551 static int
552 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
553     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
554 {
555 	struct sf_buf *sf;
556 	size_t map_len;
557 	vm_map_t map;
558 	vm_object_t object;
559 	vm_offset_t map_addr;
560 	int error, rv, cow;
561 	size_t copy_len;
562 	vm_ooffset_t file_addr;
563 
564 	/*
565 	 * It's necessary to fail if the filsz + offset taken from the
566 	 * header is greater than the actual file pager object's size.
567 	 * If we were to allow this, then the vm_map_find() below would
568 	 * walk right off the end of the file object and into the ether.
569 	 *
570 	 * While I'm here, might as well check for something else that
571 	 * is invalid: filsz cannot be greater than memsz.
572 	 */
573 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
574 	    filsz > memsz) {
575 		uprintf("elf_load_section: truncated ELF file\n");
576 		return (ENOEXEC);
577 	}
578 
579 	object = imgp->object;
580 	map = &imgp->proc->p_vmspace->vm_map;
581 	map_addr = trunc_page((vm_offset_t)vmaddr);
582 	file_addr = trunc_page(offset);
583 
584 	/*
585 	 * We have two choices.  We can either clear the data in the last page
586 	 * of an oversized mapping, or we can start the anon mapping a page
587 	 * early and copy the initialized data into that first page.  We
588 	 * choose the second.
589 	 */
590 	if (filsz == 0)
591 		map_len = 0;
592 	else if (memsz > filsz)
593 		map_len = trunc_page(offset + filsz) - file_addr;
594 	else
595 		map_len = round_page(offset + filsz) - file_addr;
596 
597 	if (map_len != 0) {
598 		/* cow flags: don't dump readonly sections in core */
599 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
600 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
601 
602 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
603 		    map_addr, map_addr + map_len, prot, cow);
604 		if (rv != KERN_SUCCESS)
605 			return (EINVAL);
606 
607 		/* we can stop now if we've covered it all */
608 		if (memsz == filsz)
609 			return (0);
610 	}
611 
612 
613 	/*
614 	 * We have to get the remaining bit of the file into the first part
615 	 * of the oversized map segment.  This is normally because the .data
616 	 * segment in the file is extended to provide bss.  It's a neat idea
617 	 * to try and save a page, but it's a pain in the behind to implement.
618 	 */
619 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
620 	    filsz);
621 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
622 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
623 
624 	/* This had damn well better be true! */
625 	if (map_len != 0) {
626 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
627 		    map_addr + map_len, prot, 0);
628 		if (rv != KERN_SUCCESS)
629 			return (EINVAL);
630 	}
631 
632 	if (copy_len != 0) {
633 		sf = vm_imgact_map_page(object, offset + filsz);
634 		if (sf == NULL)
635 			return (EIO);
636 
637 		/* send the page fragment to user space */
638 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
639 		    copy_len);
640 		vm_imgact_unmap_page(sf);
641 		if (error != 0)
642 			return (error);
643 	}
644 
645 	/*
646 	 * Remove write access to the page if it was only granted by map_insert
647 	 * to allow copyout.
648 	 */
649 	if ((prot & VM_PROT_WRITE) == 0)
650 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
651 		    map_len), prot, FALSE);
652 
653 	return (0);
654 }
655 
656 static int
657 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
658     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
659 {
660 	vm_prot_t prot;
661 	u_long base_addr;
662 	bool first;
663 	int error, i;
664 
665 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
666 
667 	base_addr = 0;
668 	first = true;
669 
670 	for (i = 0; i < hdr->e_phnum; i++) {
671 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
672 			continue;
673 
674 		/* Loadable segment */
675 		prot = __elfN(trans_prot)(phdr[i].p_flags);
676 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
677 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
678 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
679 		if (error != 0)
680 			return (error);
681 
682 		/*
683 		 * Establish the base address if this is the first segment.
684 		 */
685 		if (first) {
686   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
687 			first = false;
688 		}
689 	}
690 
691 	if (base_addrp != NULL)
692 		*base_addrp = base_addr;
693 
694 	return (0);
695 }
696 
697 /*
698  * Load the file "file" into memory.  It may be either a shared object
699  * or an executable.
700  *
701  * The "addr" reference parameter is in/out.  On entry, it specifies
702  * the address where a shared object should be loaded.  If the file is
703  * an executable, this value is ignored.  On exit, "addr" specifies
704  * where the file was actually loaded.
705  *
706  * The "entry" reference parameter is out only.  On exit, it specifies
707  * the entry point for the loaded file.
708  */
709 static int
710 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
711 	u_long *entry)
712 {
713 	struct {
714 		struct nameidata nd;
715 		struct vattr attr;
716 		struct image_params image_params;
717 	} *tempdata;
718 	const Elf_Ehdr *hdr = NULL;
719 	const Elf_Phdr *phdr = NULL;
720 	struct nameidata *nd;
721 	struct vattr *attr;
722 	struct image_params *imgp;
723 	u_long rbase;
724 	u_long base_addr = 0;
725 	int error;
726 
727 #ifdef CAPABILITY_MODE
728 	/*
729 	 * XXXJA: This check can go away once we are sufficiently confident
730 	 * that the checks in namei() are correct.
731 	 */
732 	if (IN_CAPABILITY_MODE(curthread))
733 		return (ECAPMODE);
734 #endif
735 
736 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
737 	nd = &tempdata->nd;
738 	attr = &tempdata->attr;
739 	imgp = &tempdata->image_params;
740 
741 	/*
742 	 * Initialize part of the common data
743 	 */
744 	imgp->proc = p;
745 	imgp->attr = attr;
746 	imgp->firstpage = NULL;
747 	imgp->image_header = NULL;
748 	imgp->object = NULL;
749 	imgp->execlabel = NULL;
750 
751 	NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
752 	    UIO_SYSSPACE, file, curthread);
753 	if ((error = namei(nd)) != 0) {
754 		nd->ni_vp = NULL;
755 		goto fail;
756 	}
757 	NDFREE(nd, NDF_ONLY_PNBUF);
758 	imgp->vp = nd->ni_vp;
759 
760 	/*
761 	 * Check permissions, modes, uid, etc on the file, and "open" it.
762 	 */
763 	error = exec_check_permissions(imgp);
764 	if (error)
765 		goto fail;
766 
767 	error = exec_map_first_page(imgp);
768 	if (error)
769 		goto fail;
770 
771 	imgp->object = nd->ni_vp->v_object;
772 
773 	hdr = (const Elf_Ehdr *)imgp->image_header;
774 	if ((error = __elfN(check_header)(hdr)) != 0)
775 		goto fail;
776 	if (hdr->e_type == ET_DYN)
777 		rbase = *addr;
778 	else if (hdr->e_type == ET_EXEC)
779 		rbase = 0;
780 	else {
781 		error = ENOEXEC;
782 		goto fail;
783 	}
784 
785 	/* Only support headers that fit within first page for now      */
786 	if ((hdr->e_phoff > PAGE_SIZE) ||
787 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
788 		error = ENOEXEC;
789 		goto fail;
790 	}
791 
792 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
793 	if (!aligned(phdr, Elf_Addr)) {
794 		error = ENOEXEC;
795 		goto fail;
796 	}
797 
798 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
799 	if (error != 0)
800 		goto fail;
801 
802 	*addr = base_addr;
803 	*entry = (unsigned long)hdr->e_entry + rbase;
804 
805 fail:
806 	if (imgp->firstpage)
807 		exec_unmap_first_page(imgp);
808 
809 	if (nd->ni_vp) {
810 		if (imgp->textset)
811 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
812 		vput(nd->ni_vp);
813 	}
814 	free(tempdata, M_TEMP);
815 
816 	return (error);
817 }
818 
819 static u_long
820 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
821     u_int align)
822 {
823 	u_long rbase, res;
824 
825 	MPASS(vm_map_min(map) <= minv);
826 	MPASS(maxv <= vm_map_max(map));
827 	MPASS(minv < maxv);
828 	MPASS(minv + align < maxv);
829 	arc4rand(&rbase, sizeof(rbase), 0);
830 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
831 	res &= ~((u_long)align - 1);
832 	if (res >= maxv)
833 		res -= align;
834 	KASSERT(res >= minv,
835 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
836 	    res, minv, maxv, rbase));
837 	KASSERT(res < maxv,
838 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
839 	    res, maxv, minv, rbase));
840 	return (res);
841 }
842 
843 static int
844 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
845     const Elf_Phdr *phdr, u_long et_dyn_addr)
846 {
847 	struct vmspace *vmspace;
848 	const char *err_str;
849 	u_long text_size, data_size, total_size, text_addr, data_addr;
850 	u_long seg_size, seg_addr;
851 	int i;
852 
853 	err_str = NULL;
854 	text_size = data_size = total_size = text_addr = data_addr = 0;
855 
856 	for (i = 0; i < hdr->e_phnum; i++) {
857 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
858 			continue;
859 
860 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
861 		seg_size = round_page(phdr[i].p_memsz +
862 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
863 
864 		/*
865 		 * Make the largest executable segment the official
866 		 * text segment and all others data.
867 		 *
868 		 * Note that obreak() assumes that data_addr + data_size == end
869 		 * of data load area, and the ELF file format expects segments
870 		 * to be sorted by address.  If multiple data segments exist,
871 		 * the last one will be used.
872 		 */
873 
874 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
875 			text_size = seg_size;
876 			text_addr = seg_addr;
877 		} else {
878 			data_size = seg_size;
879 			data_addr = seg_addr;
880 		}
881 		total_size += seg_size;
882 	}
883 
884 	if (data_addr == 0 && data_size == 0) {
885 		data_addr = text_addr;
886 		data_size = text_size;
887 	}
888 
889 	/*
890 	 * Check limits.  It should be safe to check the
891 	 * limits after loading the segments since we do
892 	 * not actually fault in all the segments pages.
893 	 */
894 	PROC_LOCK(imgp->proc);
895 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
896 		err_str = "Data segment size exceeds process limit";
897 	else if (text_size > maxtsiz)
898 		err_str = "Text segment size exceeds system limit";
899 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
900 		err_str = "Total segment size exceeds process limit";
901 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
902 		err_str = "Data segment size exceeds resource limit";
903 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
904 		err_str = "Total segment size exceeds resource limit";
905 	PROC_UNLOCK(imgp->proc);
906 	if (err_str != NULL) {
907 		uprintf("%s\n", err_str);
908 		return (ENOMEM);
909 	}
910 
911 	vmspace = imgp->proc->p_vmspace;
912 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
913 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
914 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
915 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
916 
917 	return (0);
918 }
919 
920 static int
921 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
922     char **interpp, bool *free_interpp)
923 {
924 	struct thread *td;
925 	char *interp;
926 	int error, interp_name_len;
927 
928 	KASSERT(phdr->p_type == PT_INTERP,
929 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
930 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
931 
932 	td = curthread;
933 
934 	/* Path to interpreter */
935 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
936 		uprintf("Invalid PT_INTERP\n");
937 		return (ENOEXEC);
938 	}
939 
940 	interp_name_len = phdr->p_filesz;
941 	if (phdr->p_offset > PAGE_SIZE ||
942 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
943 		/*
944 		 * The vnode lock might be needed by the pagedaemon to
945 		 * clean pages owned by the vnode.  Do not allow sleep
946 		 * waiting for memory with the vnode locked, instead
947 		 * try non-sleepable allocation first, and if it
948 		 * fails, go to the slow path were we drop the lock
949 		 * and do M_WAITOK.  A text reference prevents
950 		 * modifications to the vnode content.
951 		 */
952 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
953 		if (interp == NULL) {
954 			VOP_UNLOCK(imgp->vp, 0);
955 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
956 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
957 		}
958 
959 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
960 		    interp_name_len, phdr->p_offset,
961 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
962 		    NOCRED, NULL, td);
963 		if (error != 0) {
964 			free(interp, M_TEMP);
965 			uprintf("i/o error PT_INTERP %d\n", error);
966 			return (error);
967 		}
968 		interp[interp_name_len] = '\0';
969 
970 		*interpp = interp;
971 		*free_interpp = true;
972 		return (0);
973 	}
974 
975 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
976 	if (interp[interp_name_len - 1] != '\0') {
977 		uprintf("Invalid PT_INTERP\n");
978 		return (ENOEXEC);
979 	}
980 
981 	*interpp = interp;
982 	*free_interpp = false;
983 	return (0);
984 }
985 
986 static int
987 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
988     const char *interp, u_long *addr, u_long *entry)
989 {
990 	char *path;
991 	int error;
992 
993 	if (brand_info->emul_path != NULL &&
994 	    brand_info->emul_path[0] != '\0') {
995 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
996 		snprintf(path, MAXPATHLEN, "%s%s",
997 		    brand_info->emul_path, interp);
998 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
999 		free(path, M_TEMP);
1000 		if (error == 0)
1001 			return (0);
1002 	}
1003 
1004 	if (brand_info->interp_newpath != NULL &&
1005 	    (brand_info->interp_path == NULL ||
1006 	    strcmp(interp, brand_info->interp_path) == 0)) {
1007 		error = __elfN(load_file)(imgp->proc,
1008 		    brand_info->interp_newpath, addr, entry);
1009 		if (error == 0)
1010 			return (0);
1011 	}
1012 
1013 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1014 	if (error == 0)
1015 		return (0);
1016 
1017 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1018 	return (error);
1019 }
1020 
1021 /*
1022  * Impossible et_dyn_addr initial value indicating that the real base
1023  * must be calculated later with some randomization applied.
1024  */
1025 #define	ET_DYN_ADDR_RAND	1
1026 
1027 static int
1028 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1029 {
1030 	struct thread *td;
1031 	const Elf_Ehdr *hdr;
1032 	const Elf_Phdr *phdr;
1033 	Elf_Auxargs *elf_auxargs;
1034 	struct vmspace *vmspace;
1035 	vm_map_t map;
1036 	char *interp;
1037 	Elf_Brandinfo *brand_info;
1038 	struct sysentvec *sv;
1039 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1040 	u_long maxalign, mapsz, maxv, maxv1;
1041 	uint32_t fctl0;
1042 	int32_t osrel;
1043 	bool free_interp;
1044 	int error, i, n;
1045 
1046 	hdr = (const Elf_Ehdr *)imgp->image_header;
1047 
1048 	/*
1049 	 * Do we have a valid ELF header ?
1050 	 *
1051 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1052 	 * if particular brand doesn't support it.
1053 	 */
1054 	if (__elfN(check_header)(hdr) != 0 ||
1055 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1056 		return (-1);
1057 
1058 	/*
1059 	 * From here on down, we return an errno, not -1, as we've
1060 	 * detected an ELF file.
1061 	 */
1062 
1063 	if ((hdr->e_phoff > PAGE_SIZE) ||
1064 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1065 		/* Only support headers in first page for now */
1066 		uprintf("Program headers not in the first page\n");
1067 		return (ENOEXEC);
1068 	}
1069 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1070 	if (!aligned(phdr, Elf_Addr)) {
1071 		uprintf("Unaligned program headers\n");
1072 		return (ENOEXEC);
1073 	}
1074 
1075 	n = error = 0;
1076 	baddr = 0;
1077 	osrel = 0;
1078 	fctl0 = 0;
1079 	entry = proghdr = 0;
1080 	interp = NULL;
1081 	free_interp = false;
1082 	td = curthread;
1083 	maxalign = PAGE_SIZE;
1084 	mapsz = 0;
1085 
1086 	for (i = 0; i < hdr->e_phnum; i++) {
1087 		switch (phdr[i].p_type) {
1088 		case PT_LOAD:
1089 			if (n == 0)
1090 				baddr = phdr[i].p_vaddr;
1091 			if (phdr[i].p_align > maxalign)
1092 				maxalign = phdr[i].p_align;
1093 			mapsz += phdr[i].p_memsz;
1094 			n++;
1095 
1096 			/*
1097 			 * If this segment contains the program headers,
1098 			 * remember their virtual address for the AT_PHDR
1099 			 * aux entry. Static binaries don't usually include
1100 			 * a PT_PHDR entry.
1101 			 */
1102 			if (phdr[i].p_offset == 0 &&
1103 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1104 				<= phdr[i].p_filesz)
1105 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1106 			break;
1107 		case PT_INTERP:
1108 			/* Path to interpreter */
1109 			if (interp != NULL) {
1110 				uprintf("Multiple PT_INTERP headers\n");
1111 				error = ENOEXEC;
1112 				goto ret;
1113 			}
1114 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1115 			    &free_interp);
1116 			if (error != 0)
1117 				goto ret;
1118 			break;
1119 		case PT_GNU_STACK:
1120 			if (__elfN(nxstack))
1121 				imgp->stack_prot =
1122 				    __elfN(trans_prot)(phdr[i].p_flags);
1123 			imgp->stack_sz = phdr[i].p_memsz;
1124 			break;
1125 		case PT_PHDR: 	/* Program header table info */
1126 			proghdr = phdr[i].p_vaddr;
1127 			break;
1128 		}
1129 	}
1130 
1131 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1132 	if (brand_info == NULL) {
1133 		uprintf("ELF binary type \"%u\" not known.\n",
1134 		    hdr->e_ident[EI_OSABI]);
1135 		error = ENOEXEC;
1136 		goto ret;
1137 	}
1138 	sv = brand_info->sysvec;
1139 	et_dyn_addr = 0;
1140 	if (hdr->e_type == ET_DYN) {
1141 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1142 			uprintf("Cannot execute shared object\n");
1143 			error = ENOEXEC;
1144 			goto ret;
1145 		}
1146 		/*
1147 		 * Honour the base load address from the dso if it is
1148 		 * non-zero for some reason.
1149 		 */
1150 		if (baddr == 0) {
1151 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1152 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1153 				et_dyn_addr = ET_DYN_LOAD_ADDR;
1154 			else if ((__elfN(pie_aslr_enabled) &&
1155 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1156 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1157 				et_dyn_addr = ET_DYN_ADDR_RAND;
1158 			else
1159 				et_dyn_addr = ET_DYN_LOAD_ADDR;
1160 		}
1161 	}
1162 
1163 	/*
1164 	 * Avoid a possible deadlock if the current address space is destroyed
1165 	 * and that address space maps the locked vnode.  In the common case,
1166 	 * the locked vnode's v_usecount is decremented but remains greater
1167 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1168 	 * However, in cases where the vnode lock is external, such as nullfs,
1169 	 * v_usecount may become zero.
1170 	 *
1171 	 * The VV_TEXT flag prevents modifications to the executable while
1172 	 * the vnode is unlocked.
1173 	 */
1174 	VOP_UNLOCK(imgp->vp, 0);
1175 
1176 	/*
1177 	 * Decide whether to enable randomization of user mappings.
1178 	 * First, reset user preferences for the setid binaries.
1179 	 * Then, account for the support of the randomization by the
1180 	 * ABI, by user preferences, and make special treatment for
1181 	 * PIE binaries.
1182 	 */
1183 	if (imgp->credential_setid) {
1184 		PROC_LOCK(imgp->proc);
1185 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1186 		PROC_UNLOCK(imgp->proc);
1187 	}
1188 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1189 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1190 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1191 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1192 		    ("et_dyn_addr == RAND and !ASLR"));
1193 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1194 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1195 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1196 		imgp->map_flags |= MAP_ASLR;
1197 		/*
1198 		 * If user does not care about sbrk, utilize the bss
1199 		 * grow region for mappings as well.  We can select
1200 		 * the base for the image anywere and still not suffer
1201 		 * from the fragmentation.
1202 		 */
1203 		if (!__elfN(aslr_honor_sbrk) ||
1204 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1205 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1206 	}
1207 
1208 	error = exec_new_vmspace(imgp, sv);
1209 	vmspace = imgp->proc->p_vmspace;
1210 	map = &vmspace->vm_map;
1211 
1212 	imgp->proc->p_sysent = sv;
1213 
1214 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1215 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1216 		KASSERT((map->flags & MAP_ASLR) != 0,
1217 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1218 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1219 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1220 		    /* reserve half of the address space to interpreter */
1221 		    maxv / 2, 1UL << flsl(maxalign));
1222 	}
1223 
1224 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1225 	if (error != 0)
1226 		goto ret;
1227 
1228 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1229 	if (error != 0)
1230 		goto ret;
1231 
1232 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1233 	if (error != 0)
1234 		goto ret;
1235 
1236 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1237 
1238 	/*
1239 	 * We load the dynamic linker where a userland call
1240 	 * to mmap(0, ...) would put it.  The rationale behind this
1241 	 * calculation is that it leaves room for the heap to grow to
1242 	 * its maximum allowed size.
1243 	 */
1244 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1245 	    RLIMIT_DATA));
1246 	if ((map->flags & MAP_ASLR) != 0) {
1247 		maxv1 = maxv / 2 + addr / 2;
1248 		MPASS(maxv1 >= addr);	/* No overflow */
1249 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1250 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1251 	} else {
1252 		map->anon_loc = addr;
1253 	}
1254 
1255 	imgp->entry_addr = entry;
1256 
1257 	if (interp != NULL) {
1258 		VOP_UNLOCK(imgp->vp, 0);
1259 		if ((map->flags & MAP_ASLR) != 0) {
1260 			/* Assume that interpeter fits into 1/4 of AS */
1261 			maxv1 = maxv / 2 + addr / 2;
1262 			MPASS(maxv1 >= addr);	/* No overflow */
1263 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1264 			    maxv1, PAGE_SIZE);
1265 		}
1266 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1267 		    &imgp->entry_addr);
1268 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1269 		if (error != 0)
1270 			goto ret;
1271 	} else
1272 		addr = et_dyn_addr;
1273 
1274 	/*
1275 	 * Construct auxargs table (used by the fixup routine)
1276 	 */
1277 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1278 	if (elf_auxargs == NULL) {
1279 		VOP_UNLOCK(imgp->vp, 0);
1280 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1281 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1282 	}
1283 	elf_auxargs->execfd = -1;
1284 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1285 	elf_auxargs->phent = hdr->e_phentsize;
1286 	elf_auxargs->phnum = hdr->e_phnum;
1287 	elf_auxargs->pagesz = PAGE_SIZE;
1288 	elf_auxargs->base = addr;
1289 	elf_auxargs->flags = 0;
1290 	elf_auxargs->entry = entry;
1291 	elf_auxargs->hdr_eflags = hdr->e_flags;
1292 
1293 	imgp->auxargs = elf_auxargs;
1294 	imgp->interpreted = 0;
1295 	imgp->reloc_base = addr;
1296 	imgp->proc->p_osrel = osrel;
1297 	imgp->proc->p_fctl0 = fctl0;
1298 	imgp->proc->p_elf_machine = hdr->e_machine;
1299 	imgp->proc->p_elf_flags = hdr->e_flags;
1300 
1301 ret:
1302 	if (free_interp)
1303 		free(interp, M_TEMP);
1304 	return (error);
1305 }
1306 
1307 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1308 
1309 int
1310 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1311 {
1312 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1313 	Elf_Auxinfo *argarray, *pos;
1314 	Elf_Addr *base, *auxbase;
1315 	int error;
1316 
1317 	base = (Elf_Addr *)*stack_base;
1318 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1319 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1320 	    M_WAITOK | M_ZERO);
1321 
1322 	if (args->execfd != -1)
1323 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1324 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1325 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1326 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1327 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1328 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1329 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1330 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1331 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1332 	if (imgp->execpathp != 0)
1333 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1334 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1335 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1336 	if (imgp->canary != 0) {
1337 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1338 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1339 	}
1340 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1341 	if (imgp->pagesizes != 0) {
1342 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1343 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1344 	}
1345 	if (imgp->sysent->sv_timekeep_base != 0) {
1346 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1347 		    imgp->sysent->sv_timekeep_base);
1348 	}
1349 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1350 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1351 	    imgp->sysent->sv_stackprot);
1352 	if (imgp->sysent->sv_hwcap != NULL)
1353 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1354 	if (imgp->sysent->sv_hwcap2 != NULL)
1355 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1356 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1357 
1358 	free(imgp->auxargs, M_TEMP);
1359 	imgp->auxargs = NULL;
1360 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1361 
1362 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1363 	free(argarray, M_TEMP);
1364 	if (error != 0)
1365 		return (error);
1366 
1367 	base--;
1368 	if (suword(base, imgp->args->argc) == -1)
1369 		return (EFAULT);
1370 	*stack_base = (register_t *)base;
1371 	return (0);
1372 }
1373 
1374 /*
1375  * Code for generating ELF core dumps.
1376  */
1377 
1378 typedef void (*segment_callback)(vm_map_entry_t, void *);
1379 
1380 /* Closure for cb_put_phdr(). */
1381 struct phdr_closure {
1382 	Elf_Phdr *phdr;		/* Program header to fill in */
1383 	Elf_Off offset;		/* Offset of segment in core file */
1384 };
1385 
1386 /* Closure for cb_size_segment(). */
1387 struct sseg_closure {
1388 	int count;		/* Count of writable segments. */
1389 	size_t size;		/* Total size of all writable segments. */
1390 };
1391 
1392 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1393 
1394 struct note_info {
1395 	int		type;		/* Note type. */
1396 	outfunc_t 	outfunc; 	/* Output function. */
1397 	void		*outarg;	/* Argument for the output function. */
1398 	size_t		outsize;	/* Output size. */
1399 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1400 };
1401 
1402 TAILQ_HEAD(note_info_list, note_info);
1403 
1404 /* Coredump output parameters. */
1405 struct coredump_params {
1406 	off_t		offset;
1407 	struct ucred	*active_cred;
1408 	struct ucred	*file_cred;
1409 	struct thread	*td;
1410 	struct vnode	*vp;
1411 	struct compressor *comp;
1412 };
1413 
1414 extern int compress_user_cores;
1415 extern int compress_user_cores_level;
1416 
1417 static void cb_put_phdr(vm_map_entry_t, void *);
1418 static void cb_size_segment(vm_map_entry_t, void *);
1419 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1420     enum uio_seg);
1421 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1422 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1423     struct note_info_list *, size_t);
1424 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1425     size_t *);
1426 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1427 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1428 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1429 static int sbuf_drain_core_output(void *, const char *, int);
1430 
1431 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1432 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1433 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1434 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1435 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1436 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1437 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1438 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1439 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1440 static void note_procstat_files(void *, struct sbuf *, size_t *);
1441 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1442 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1443 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1444 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1445 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1446 
1447 /*
1448  * Write out a core segment to the compression stream.
1449  */
1450 static int
1451 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1452 {
1453 	u_int chunk_len;
1454 	int error;
1455 
1456 	while (len > 0) {
1457 		chunk_len = MIN(len, CORE_BUF_SIZE);
1458 
1459 		/*
1460 		 * We can get EFAULT error here.
1461 		 * In that case zero out the current chunk of the segment.
1462 		 */
1463 		error = copyin(base, buf, chunk_len);
1464 		if (error != 0)
1465 			bzero(buf, chunk_len);
1466 		error = compressor_write(p->comp, buf, chunk_len);
1467 		if (error != 0)
1468 			break;
1469 		base += chunk_len;
1470 		len -= chunk_len;
1471 	}
1472 	return (error);
1473 }
1474 
1475 static int
1476 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1477 {
1478 
1479 	return (core_write((struct coredump_params *)arg, base, len, offset,
1480 	    UIO_SYSSPACE));
1481 }
1482 
1483 static int
1484 core_write(struct coredump_params *p, const void *base, size_t len,
1485     off_t offset, enum uio_seg seg)
1486 {
1487 
1488 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1489 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1490 	    p->active_cred, p->file_cred, NULL, p->td));
1491 }
1492 
1493 static int
1494 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1495     void *tmpbuf)
1496 {
1497 	int error;
1498 
1499 	if (p->comp != NULL)
1500 		return (compress_chunk(p, base, tmpbuf, len));
1501 
1502 	/*
1503 	 * EFAULT is a non-fatal error that we can get, for example,
1504 	 * if the segment is backed by a file but extends beyond its
1505 	 * end.
1506 	 */
1507 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1508 	if (error == EFAULT) {
1509 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1510 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1511 		    "for process %s\n", base, len, offset, curproc->p_comm);
1512 
1513 		/*
1514 		 * Write a "real" zero byte at the end of the target region
1515 		 * in the case this is the last segment.
1516 		 * The intermediate space will be implicitly zero-filled.
1517 		 */
1518 		error = core_write(p, zero_region, 1, offset + len - 1,
1519 		    UIO_SYSSPACE);
1520 	}
1521 	return (error);
1522 }
1523 
1524 /*
1525  * Drain into a core file.
1526  */
1527 static int
1528 sbuf_drain_core_output(void *arg, const char *data, int len)
1529 {
1530 	struct coredump_params *p;
1531 	int error, locked;
1532 
1533 	p = (struct coredump_params *)arg;
1534 
1535 	/*
1536 	 * Some kern_proc out routines that print to this sbuf may
1537 	 * call us with the process lock held. Draining with the
1538 	 * non-sleepable lock held is unsafe. The lock is needed for
1539 	 * those routines when dumping a live process. In our case we
1540 	 * can safely release the lock before draining and acquire
1541 	 * again after.
1542 	 */
1543 	locked = PROC_LOCKED(p->td->td_proc);
1544 	if (locked)
1545 		PROC_UNLOCK(p->td->td_proc);
1546 	if (p->comp != NULL)
1547 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1548 	else
1549 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1550 		    UIO_SYSSPACE);
1551 	if (locked)
1552 		PROC_LOCK(p->td->td_proc);
1553 	if (error != 0)
1554 		return (-error);
1555 	p->offset += len;
1556 	return (len);
1557 }
1558 
1559 int
1560 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1561 {
1562 	struct ucred *cred = td->td_ucred;
1563 	int error = 0;
1564 	struct sseg_closure seginfo;
1565 	struct note_info_list notelst;
1566 	struct coredump_params params;
1567 	struct note_info *ninfo;
1568 	void *hdr, *tmpbuf;
1569 	size_t hdrsize, notesz, coresize;
1570 
1571 	hdr = NULL;
1572 	tmpbuf = NULL;
1573 	TAILQ_INIT(&notelst);
1574 
1575 	/* Size the program segments. */
1576 	seginfo.count = 0;
1577 	seginfo.size = 0;
1578 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1579 
1580 	/*
1581 	 * Collect info about the core file header area.
1582 	 */
1583 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1584 	if (seginfo.count + 1 >= PN_XNUM)
1585 		hdrsize += sizeof(Elf_Shdr);
1586 	__elfN(prepare_notes)(td, &notelst, &notesz);
1587 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1588 
1589 	/* Set up core dump parameters. */
1590 	params.offset = 0;
1591 	params.active_cred = cred;
1592 	params.file_cred = NOCRED;
1593 	params.td = td;
1594 	params.vp = vp;
1595 	params.comp = NULL;
1596 
1597 #ifdef RACCT
1598 	if (racct_enable) {
1599 		PROC_LOCK(td->td_proc);
1600 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1601 		PROC_UNLOCK(td->td_proc);
1602 		if (error != 0) {
1603 			error = EFAULT;
1604 			goto done;
1605 		}
1606 	}
1607 #endif
1608 	if (coresize >= limit) {
1609 		error = EFAULT;
1610 		goto done;
1611 	}
1612 
1613 	/* Create a compression stream if necessary. */
1614 	if (compress_user_cores != 0) {
1615 		params.comp = compressor_init(core_compressed_write,
1616 		    compress_user_cores, CORE_BUF_SIZE,
1617 		    compress_user_cores_level, &params);
1618 		if (params.comp == NULL) {
1619 			error = EFAULT;
1620 			goto done;
1621 		}
1622 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1623         }
1624 
1625 	/*
1626 	 * Allocate memory for building the header, fill it up,
1627 	 * and write it out following the notes.
1628 	 */
1629 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1630 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1631 	    notesz);
1632 
1633 	/* Write the contents of all of the writable segments. */
1634 	if (error == 0) {
1635 		Elf_Phdr *php;
1636 		off_t offset;
1637 		int i;
1638 
1639 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1640 		offset = round_page(hdrsize + notesz);
1641 		for (i = 0; i < seginfo.count; i++) {
1642 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1643 			    php->p_filesz, offset, &params, tmpbuf);
1644 			if (error != 0)
1645 				break;
1646 			offset += php->p_filesz;
1647 			php++;
1648 		}
1649 		if (error == 0 && params.comp != NULL)
1650 			error = compressor_flush(params.comp);
1651 	}
1652 	if (error) {
1653 		log(LOG_WARNING,
1654 		    "Failed to write core file for process %s (error %d)\n",
1655 		    curproc->p_comm, error);
1656 	}
1657 
1658 done:
1659 	free(tmpbuf, M_TEMP);
1660 	if (params.comp != NULL)
1661 		compressor_fini(params.comp);
1662 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1663 		TAILQ_REMOVE(&notelst, ninfo, link);
1664 		free(ninfo, M_TEMP);
1665 	}
1666 	if (hdr != NULL)
1667 		free(hdr, M_TEMP);
1668 
1669 	return (error);
1670 }
1671 
1672 /*
1673  * A callback for each_dumpable_segment() to write out the segment's
1674  * program header entry.
1675  */
1676 static void
1677 cb_put_phdr(vm_map_entry_t entry, void *closure)
1678 {
1679 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1680 	Elf_Phdr *phdr = phc->phdr;
1681 
1682 	phc->offset = round_page(phc->offset);
1683 
1684 	phdr->p_type = PT_LOAD;
1685 	phdr->p_offset = phc->offset;
1686 	phdr->p_vaddr = entry->start;
1687 	phdr->p_paddr = 0;
1688 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1689 	phdr->p_align = PAGE_SIZE;
1690 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1691 
1692 	phc->offset += phdr->p_filesz;
1693 	phc->phdr++;
1694 }
1695 
1696 /*
1697  * A callback for each_dumpable_segment() to gather information about
1698  * the number of segments and their total size.
1699  */
1700 static void
1701 cb_size_segment(vm_map_entry_t entry, void *closure)
1702 {
1703 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1704 
1705 	ssc->count++;
1706 	ssc->size += entry->end - entry->start;
1707 }
1708 
1709 /*
1710  * For each writable segment in the process's memory map, call the given
1711  * function with a pointer to the map entry and some arbitrary
1712  * caller-supplied data.
1713  */
1714 static void
1715 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1716 {
1717 	struct proc *p = td->td_proc;
1718 	vm_map_t map = &p->p_vmspace->vm_map;
1719 	vm_map_entry_t entry;
1720 	vm_object_t backing_object, object;
1721 	boolean_t ignore_entry;
1722 
1723 	vm_map_lock_read(map);
1724 	for (entry = map->header.next; entry != &map->header;
1725 	    entry = entry->next) {
1726 		/*
1727 		 * Don't dump inaccessible mappings, deal with legacy
1728 		 * coredump mode.
1729 		 *
1730 		 * Note that read-only segments related to the elf binary
1731 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1732 		 * need to arbitrarily ignore such segments.
1733 		 */
1734 		if (elf_legacy_coredump) {
1735 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1736 				continue;
1737 		} else {
1738 			if ((entry->protection & VM_PROT_ALL) == 0)
1739 				continue;
1740 		}
1741 
1742 		/*
1743 		 * Dont include memory segment in the coredump if
1744 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1745 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1746 		 * kernel map).
1747 		 */
1748 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1749 			continue;
1750 
1751 		if ((object = entry->object.vm_object) == NULL)
1752 			continue;
1753 
1754 		/* Ignore memory-mapped devices and such things. */
1755 		VM_OBJECT_RLOCK(object);
1756 		while ((backing_object = object->backing_object) != NULL) {
1757 			VM_OBJECT_RLOCK(backing_object);
1758 			VM_OBJECT_RUNLOCK(object);
1759 			object = backing_object;
1760 		}
1761 		ignore_entry = object->type != OBJT_DEFAULT &&
1762 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1763 		    object->type != OBJT_PHYS;
1764 		VM_OBJECT_RUNLOCK(object);
1765 		if (ignore_entry)
1766 			continue;
1767 
1768 		(*func)(entry, closure);
1769 	}
1770 	vm_map_unlock_read(map);
1771 }
1772 
1773 /*
1774  * Write the core file header to the file, including padding up to
1775  * the page boundary.
1776  */
1777 static int
1778 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1779     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1780 {
1781 	struct note_info *ninfo;
1782 	struct sbuf *sb;
1783 	int error;
1784 
1785 	/* Fill in the header. */
1786 	bzero(hdr, hdrsize);
1787 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1788 
1789 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1790 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1791 	sbuf_start_section(sb, NULL);
1792 	sbuf_bcat(sb, hdr, hdrsize);
1793 	TAILQ_FOREACH(ninfo, notelst, link)
1794 	    __elfN(putnote)(ninfo, sb);
1795 	/* Align up to a page boundary for the program segments. */
1796 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1797 	error = sbuf_finish(sb);
1798 	sbuf_delete(sb);
1799 
1800 	return (error);
1801 }
1802 
1803 static void
1804 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1805     size_t *sizep)
1806 {
1807 	struct proc *p;
1808 	struct thread *thr;
1809 	size_t size;
1810 
1811 	p = td->td_proc;
1812 	size = 0;
1813 
1814 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1815 
1816 	/*
1817 	 * To have the debugger select the right thread (LWP) as the initial
1818 	 * thread, we dump the state of the thread passed to us in td first.
1819 	 * This is the thread that causes the core dump and thus likely to
1820 	 * be the right thread one wants to have selected in the debugger.
1821 	 */
1822 	thr = td;
1823 	while (thr != NULL) {
1824 		size += register_note(list, NT_PRSTATUS,
1825 		    __elfN(note_prstatus), thr);
1826 		size += register_note(list, NT_FPREGSET,
1827 		    __elfN(note_fpregset), thr);
1828 		size += register_note(list, NT_THRMISC,
1829 		    __elfN(note_thrmisc), thr);
1830 		size += register_note(list, NT_PTLWPINFO,
1831 		    __elfN(note_ptlwpinfo), thr);
1832 		size += register_note(list, -1,
1833 		    __elfN(note_threadmd), thr);
1834 
1835 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1836 		    TAILQ_NEXT(thr, td_plist);
1837 		if (thr == td)
1838 			thr = TAILQ_NEXT(thr, td_plist);
1839 	}
1840 
1841 	size += register_note(list, NT_PROCSTAT_PROC,
1842 	    __elfN(note_procstat_proc), p);
1843 	size += register_note(list, NT_PROCSTAT_FILES,
1844 	    note_procstat_files, p);
1845 	size += register_note(list, NT_PROCSTAT_VMMAP,
1846 	    note_procstat_vmmap, p);
1847 	size += register_note(list, NT_PROCSTAT_GROUPS,
1848 	    note_procstat_groups, p);
1849 	size += register_note(list, NT_PROCSTAT_UMASK,
1850 	    note_procstat_umask, p);
1851 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1852 	    note_procstat_rlimit, p);
1853 	size += register_note(list, NT_PROCSTAT_OSREL,
1854 	    note_procstat_osrel, p);
1855 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1856 	    __elfN(note_procstat_psstrings), p);
1857 	size += register_note(list, NT_PROCSTAT_AUXV,
1858 	    __elfN(note_procstat_auxv), p);
1859 
1860 	*sizep = size;
1861 }
1862 
1863 static void
1864 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1865     size_t notesz)
1866 {
1867 	Elf_Ehdr *ehdr;
1868 	Elf_Phdr *phdr;
1869 	Elf_Shdr *shdr;
1870 	struct phdr_closure phc;
1871 
1872 	ehdr = (Elf_Ehdr *)hdr;
1873 
1874 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1875 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1876 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1877 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1878 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1879 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1880 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1881 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1882 	ehdr->e_ident[EI_ABIVERSION] = 0;
1883 	ehdr->e_ident[EI_PAD] = 0;
1884 	ehdr->e_type = ET_CORE;
1885 	ehdr->e_machine = td->td_proc->p_elf_machine;
1886 	ehdr->e_version = EV_CURRENT;
1887 	ehdr->e_entry = 0;
1888 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1889 	ehdr->e_flags = td->td_proc->p_elf_flags;
1890 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1891 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1892 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1893 	ehdr->e_shstrndx = SHN_UNDEF;
1894 	if (numsegs + 1 < PN_XNUM) {
1895 		ehdr->e_phnum = numsegs + 1;
1896 		ehdr->e_shnum = 0;
1897 	} else {
1898 		ehdr->e_phnum = PN_XNUM;
1899 		ehdr->e_shnum = 1;
1900 
1901 		ehdr->e_shoff = ehdr->e_phoff +
1902 		    (numsegs + 1) * ehdr->e_phentsize;
1903 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1904 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1905 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1906 
1907 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1908 		memset(shdr, 0, sizeof(*shdr));
1909 		/*
1910 		 * A special first section is used to hold large segment and
1911 		 * section counts.  This was proposed by Sun Microsystems in
1912 		 * Solaris and has been adopted by Linux; the standard ELF
1913 		 * tools are already familiar with the technique.
1914 		 *
1915 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1916 		 * (or 12-7 depending on the version of the document) for more
1917 		 * details.
1918 		 */
1919 		shdr->sh_type = SHT_NULL;
1920 		shdr->sh_size = ehdr->e_shnum;
1921 		shdr->sh_link = ehdr->e_shstrndx;
1922 		shdr->sh_info = numsegs + 1;
1923 	}
1924 
1925 	/*
1926 	 * Fill in the program header entries.
1927 	 */
1928 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1929 
1930 	/* The note segement. */
1931 	phdr->p_type = PT_NOTE;
1932 	phdr->p_offset = hdrsize;
1933 	phdr->p_vaddr = 0;
1934 	phdr->p_paddr = 0;
1935 	phdr->p_filesz = notesz;
1936 	phdr->p_memsz = 0;
1937 	phdr->p_flags = PF_R;
1938 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1939 	phdr++;
1940 
1941 	/* All the writable segments from the program. */
1942 	phc.phdr = phdr;
1943 	phc.offset = round_page(hdrsize + notesz);
1944 	each_dumpable_segment(td, cb_put_phdr, &phc);
1945 }
1946 
1947 static size_t
1948 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1949 {
1950 	struct note_info *ninfo;
1951 	size_t size, notesize;
1952 
1953 	size = 0;
1954 	out(arg, NULL, &size);
1955 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1956 	ninfo->type = type;
1957 	ninfo->outfunc = out;
1958 	ninfo->outarg = arg;
1959 	ninfo->outsize = size;
1960 	TAILQ_INSERT_TAIL(list, ninfo, link);
1961 
1962 	if (type == -1)
1963 		return (size);
1964 
1965 	notesize = sizeof(Elf_Note) +		/* note header */
1966 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1967 						/* note name */
1968 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1969 
1970 	return (notesize);
1971 }
1972 
1973 static size_t
1974 append_note_data(const void *src, void *dst, size_t len)
1975 {
1976 	size_t padded_len;
1977 
1978 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1979 	if (dst != NULL) {
1980 		bcopy(src, dst, len);
1981 		bzero((char *)dst + len, padded_len - len);
1982 	}
1983 	return (padded_len);
1984 }
1985 
1986 size_t
1987 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1988 {
1989 	Elf_Note *note;
1990 	char *buf;
1991 	size_t notesize;
1992 
1993 	buf = dst;
1994 	if (buf != NULL) {
1995 		note = (Elf_Note *)buf;
1996 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1997 		note->n_descsz = size;
1998 		note->n_type = type;
1999 		buf += sizeof(*note);
2000 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2001 		    sizeof(FREEBSD_ABI_VENDOR));
2002 		append_note_data(src, buf, size);
2003 		if (descp != NULL)
2004 			*descp = buf;
2005 	}
2006 
2007 	notesize = sizeof(Elf_Note) +		/* note header */
2008 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2009 						/* note name */
2010 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2011 
2012 	return (notesize);
2013 }
2014 
2015 static void
2016 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2017 {
2018 	Elf_Note note;
2019 	ssize_t old_len, sect_len;
2020 	size_t new_len, descsz, i;
2021 
2022 	if (ninfo->type == -1) {
2023 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2024 		return;
2025 	}
2026 
2027 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2028 	note.n_descsz = ninfo->outsize;
2029 	note.n_type = ninfo->type;
2030 
2031 	sbuf_bcat(sb, &note, sizeof(note));
2032 	sbuf_start_section(sb, &old_len);
2033 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2034 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2035 	if (note.n_descsz == 0)
2036 		return;
2037 	sbuf_start_section(sb, &old_len);
2038 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2039 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2040 	if (sect_len < 0)
2041 		return;
2042 
2043 	new_len = (size_t)sect_len;
2044 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2045 	if (new_len < descsz) {
2046 		/*
2047 		 * It is expected that individual note emitters will correctly
2048 		 * predict their expected output size and fill up to that size
2049 		 * themselves, padding in a format-specific way if needed.
2050 		 * However, in case they don't, just do it here with zeros.
2051 		 */
2052 		for (i = 0; i < descsz - new_len; i++)
2053 			sbuf_putc(sb, 0);
2054 	} else if (new_len > descsz) {
2055 		/*
2056 		 * We can't always truncate sb -- we may have drained some
2057 		 * of it already.
2058 		 */
2059 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2060 		    "read it (%zu > %zu).  Since it is longer than "
2061 		    "expected, this coredump's notes are corrupt.  THIS "
2062 		    "IS A BUG in the note_procstat routine for type %u.\n",
2063 		    __func__, (unsigned)note.n_type, new_len, descsz,
2064 		    (unsigned)note.n_type));
2065 	}
2066 }
2067 
2068 /*
2069  * Miscellaneous note out functions.
2070  */
2071 
2072 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2073 #include <compat/freebsd32/freebsd32.h>
2074 #include <compat/freebsd32/freebsd32_signal.h>
2075 
2076 typedef struct prstatus32 elf_prstatus_t;
2077 typedef struct prpsinfo32 elf_prpsinfo_t;
2078 typedef struct fpreg32 elf_prfpregset_t;
2079 typedef struct fpreg32 elf_fpregset_t;
2080 typedef struct reg32 elf_gregset_t;
2081 typedef struct thrmisc32 elf_thrmisc_t;
2082 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2083 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2084 typedef uint32_t elf_ps_strings_t;
2085 #else
2086 typedef prstatus_t elf_prstatus_t;
2087 typedef prpsinfo_t elf_prpsinfo_t;
2088 typedef prfpregset_t elf_prfpregset_t;
2089 typedef prfpregset_t elf_fpregset_t;
2090 typedef gregset_t elf_gregset_t;
2091 typedef thrmisc_t elf_thrmisc_t;
2092 #define ELF_KERN_PROC_MASK	0
2093 typedef struct kinfo_proc elf_kinfo_proc_t;
2094 typedef vm_offset_t elf_ps_strings_t;
2095 #endif
2096 
2097 static void
2098 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2099 {
2100 	struct sbuf sbarg;
2101 	size_t len;
2102 	char *cp, *end;
2103 	struct proc *p;
2104 	elf_prpsinfo_t *psinfo;
2105 	int error;
2106 
2107 	p = (struct proc *)arg;
2108 	if (sb != NULL) {
2109 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2110 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2111 		psinfo->pr_version = PRPSINFO_VERSION;
2112 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2113 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2114 		PROC_LOCK(p);
2115 		if (p->p_args != NULL) {
2116 			len = sizeof(psinfo->pr_psargs) - 1;
2117 			if (len > p->p_args->ar_length)
2118 				len = p->p_args->ar_length;
2119 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2120 			PROC_UNLOCK(p);
2121 			error = 0;
2122 		} else {
2123 			_PHOLD(p);
2124 			PROC_UNLOCK(p);
2125 			sbuf_new(&sbarg, psinfo->pr_psargs,
2126 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2127 			error = proc_getargv(curthread, p, &sbarg);
2128 			PRELE(p);
2129 			if (sbuf_finish(&sbarg) == 0)
2130 				len = sbuf_len(&sbarg) - 1;
2131 			else
2132 				len = sizeof(psinfo->pr_psargs) - 1;
2133 			sbuf_delete(&sbarg);
2134 		}
2135 		if (error || len == 0)
2136 			strlcpy(psinfo->pr_psargs, p->p_comm,
2137 			    sizeof(psinfo->pr_psargs));
2138 		else {
2139 			KASSERT(len < sizeof(psinfo->pr_psargs),
2140 			    ("len is too long: %zu vs %zu", len,
2141 			    sizeof(psinfo->pr_psargs)));
2142 			cp = psinfo->pr_psargs;
2143 			end = cp + len - 1;
2144 			for (;;) {
2145 				cp = memchr(cp, '\0', end - cp);
2146 				if (cp == NULL)
2147 					break;
2148 				*cp = ' ';
2149 			}
2150 		}
2151 		psinfo->pr_pid = p->p_pid;
2152 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2153 		free(psinfo, M_TEMP);
2154 	}
2155 	*sizep = sizeof(*psinfo);
2156 }
2157 
2158 static void
2159 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2160 {
2161 	struct thread *td;
2162 	elf_prstatus_t *status;
2163 
2164 	td = (struct thread *)arg;
2165 	if (sb != NULL) {
2166 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2167 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2168 		status->pr_version = PRSTATUS_VERSION;
2169 		status->pr_statussz = sizeof(elf_prstatus_t);
2170 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2171 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2172 		status->pr_osreldate = osreldate;
2173 		status->pr_cursig = td->td_proc->p_sig;
2174 		status->pr_pid = td->td_tid;
2175 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2176 		fill_regs32(td, &status->pr_reg);
2177 #else
2178 		fill_regs(td, &status->pr_reg);
2179 #endif
2180 		sbuf_bcat(sb, status, sizeof(*status));
2181 		free(status, M_TEMP);
2182 	}
2183 	*sizep = sizeof(*status);
2184 }
2185 
2186 static void
2187 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2188 {
2189 	struct thread *td;
2190 	elf_prfpregset_t *fpregset;
2191 
2192 	td = (struct thread *)arg;
2193 	if (sb != NULL) {
2194 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2195 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2196 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2197 		fill_fpregs32(td, fpregset);
2198 #else
2199 		fill_fpregs(td, fpregset);
2200 #endif
2201 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2202 		free(fpregset, M_TEMP);
2203 	}
2204 	*sizep = sizeof(*fpregset);
2205 }
2206 
2207 static void
2208 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2209 {
2210 	struct thread *td;
2211 	elf_thrmisc_t thrmisc;
2212 
2213 	td = (struct thread *)arg;
2214 	if (sb != NULL) {
2215 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2216 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2217 		strcpy(thrmisc.pr_tname, td->td_name);
2218 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2219 	}
2220 	*sizep = sizeof(thrmisc);
2221 }
2222 
2223 static void
2224 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2225 {
2226 	struct thread *td;
2227 	size_t size;
2228 	int structsize;
2229 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2230 	struct ptrace_lwpinfo32 pl;
2231 #else
2232 	struct ptrace_lwpinfo pl;
2233 #endif
2234 
2235 	td = (struct thread *)arg;
2236 	size = sizeof(structsize) + sizeof(pl);
2237 	if (sb != NULL) {
2238 		KASSERT(*sizep == size, ("invalid size"));
2239 		structsize = sizeof(pl);
2240 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2241 		bzero(&pl, sizeof(pl));
2242 		pl.pl_lwpid = td->td_tid;
2243 		pl.pl_event = PL_EVENT_NONE;
2244 		pl.pl_sigmask = td->td_sigmask;
2245 		pl.pl_siglist = td->td_siglist;
2246 		if (td->td_si.si_signo != 0) {
2247 			pl.pl_event = PL_EVENT_SIGNAL;
2248 			pl.pl_flags |= PL_FLAG_SI;
2249 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2250 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2251 #else
2252 			pl.pl_siginfo = td->td_si;
2253 #endif
2254 		}
2255 		strcpy(pl.pl_tdname, td->td_name);
2256 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2257 		sbuf_bcat(sb, &pl, sizeof(pl));
2258 	}
2259 	*sizep = size;
2260 }
2261 
2262 /*
2263  * Allow for MD specific notes, as well as any MD
2264  * specific preparations for writing MI notes.
2265  */
2266 static void
2267 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2268 {
2269 	struct thread *td;
2270 	void *buf;
2271 	size_t size;
2272 
2273 	td = (struct thread *)arg;
2274 	size = *sizep;
2275 	if (size != 0 && sb != NULL)
2276 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2277 	else
2278 		buf = NULL;
2279 	size = 0;
2280 	__elfN(dump_thread)(td, buf, &size);
2281 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2282 	if (size != 0 && sb != NULL)
2283 		sbuf_bcat(sb, buf, size);
2284 	free(buf, M_TEMP);
2285 	*sizep = size;
2286 }
2287 
2288 #ifdef KINFO_PROC_SIZE
2289 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2290 #endif
2291 
2292 static void
2293 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2294 {
2295 	struct proc *p;
2296 	size_t size;
2297 	int structsize;
2298 
2299 	p = (struct proc *)arg;
2300 	size = sizeof(structsize) + p->p_numthreads *
2301 	    sizeof(elf_kinfo_proc_t);
2302 
2303 	if (sb != NULL) {
2304 		KASSERT(*sizep == size, ("invalid size"));
2305 		structsize = sizeof(elf_kinfo_proc_t);
2306 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2307 		PROC_LOCK(p);
2308 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2309 	}
2310 	*sizep = size;
2311 }
2312 
2313 #ifdef KINFO_FILE_SIZE
2314 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2315 #endif
2316 
2317 static void
2318 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2319 {
2320 	struct proc *p;
2321 	size_t size, sect_sz, i;
2322 	ssize_t start_len, sect_len;
2323 	int structsize, filedesc_flags;
2324 
2325 	if (coredump_pack_fileinfo)
2326 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2327 	else
2328 		filedesc_flags = 0;
2329 
2330 	p = (struct proc *)arg;
2331 	structsize = sizeof(struct kinfo_file);
2332 	if (sb == NULL) {
2333 		size = 0;
2334 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2335 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2336 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2337 		PROC_LOCK(p);
2338 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2339 		sbuf_finish(sb);
2340 		sbuf_delete(sb);
2341 		*sizep = size;
2342 	} else {
2343 		sbuf_start_section(sb, &start_len);
2344 
2345 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2346 		PROC_LOCK(p);
2347 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2348 		    filedesc_flags);
2349 
2350 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2351 		if (sect_len < 0)
2352 			return;
2353 		sect_sz = sect_len;
2354 
2355 		KASSERT(sect_sz <= *sizep,
2356 		    ("kern_proc_filedesc_out did not respect maxlen; "
2357 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2358 		     sect_sz - sizeof(structsize)));
2359 
2360 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2361 			sbuf_putc(sb, 0);
2362 	}
2363 }
2364 
2365 #ifdef KINFO_VMENTRY_SIZE
2366 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2367 #endif
2368 
2369 static void
2370 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2371 {
2372 	struct proc *p;
2373 	size_t size;
2374 	int structsize, vmmap_flags;
2375 
2376 	if (coredump_pack_vmmapinfo)
2377 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2378 	else
2379 		vmmap_flags = 0;
2380 
2381 	p = (struct proc *)arg;
2382 	structsize = sizeof(struct kinfo_vmentry);
2383 	if (sb == NULL) {
2384 		size = 0;
2385 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2386 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2387 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2388 		PROC_LOCK(p);
2389 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2390 		sbuf_finish(sb);
2391 		sbuf_delete(sb);
2392 		*sizep = size;
2393 	} else {
2394 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2395 		PROC_LOCK(p);
2396 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2397 		    vmmap_flags);
2398 	}
2399 }
2400 
2401 static void
2402 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2403 {
2404 	struct proc *p;
2405 	size_t size;
2406 	int structsize;
2407 
2408 	p = (struct proc *)arg;
2409 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2410 	if (sb != NULL) {
2411 		KASSERT(*sizep == size, ("invalid size"));
2412 		structsize = sizeof(gid_t);
2413 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2414 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2415 		    sizeof(gid_t));
2416 	}
2417 	*sizep = size;
2418 }
2419 
2420 static void
2421 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2422 {
2423 	struct proc *p;
2424 	size_t size;
2425 	int structsize;
2426 
2427 	p = (struct proc *)arg;
2428 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2429 	if (sb != NULL) {
2430 		KASSERT(*sizep == size, ("invalid size"));
2431 		structsize = sizeof(p->p_fd->fd_cmask);
2432 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2433 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2434 	}
2435 	*sizep = size;
2436 }
2437 
2438 static void
2439 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2440 {
2441 	struct proc *p;
2442 	struct rlimit rlim[RLIM_NLIMITS];
2443 	size_t size;
2444 	int structsize, i;
2445 
2446 	p = (struct proc *)arg;
2447 	size = sizeof(structsize) + sizeof(rlim);
2448 	if (sb != NULL) {
2449 		KASSERT(*sizep == size, ("invalid size"));
2450 		structsize = sizeof(rlim);
2451 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2452 		PROC_LOCK(p);
2453 		for (i = 0; i < RLIM_NLIMITS; i++)
2454 			lim_rlimit_proc(p, i, &rlim[i]);
2455 		PROC_UNLOCK(p);
2456 		sbuf_bcat(sb, rlim, sizeof(rlim));
2457 	}
2458 	*sizep = size;
2459 }
2460 
2461 static void
2462 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2463 {
2464 	struct proc *p;
2465 	size_t size;
2466 	int structsize;
2467 
2468 	p = (struct proc *)arg;
2469 	size = sizeof(structsize) + sizeof(p->p_osrel);
2470 	if (sb != NULL) {
2471 		KASSERT(*sizep == size, ("invalid size"));
2472 		structsize = sizeof(p->p_osrel);
2473 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2474 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2475 	}
2476 	*sizep = size;
2477 }
2478 
2479 static void
2480 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2481 {
2482 	struct proc *p;
2483 	elf_ps_strings_t ps_strings;
2484 	size_t size;
2485 	int structsize;
2486 
2487 	p = (struct proc *)arg;
2488 	size = sizeof(structsize) + sizeof(ps_strings);
2489 	if (sb != NULL) {
2490 		KASSERT(*sizep == size, ("invalid size"));
2491 		structsize = sizeof(ps_strings);
2492 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2493 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2494 #else
2495 		ps_strings = p->p_sysent->sv_psstrings;
2496 #endif
2497 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2498 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2499 	}
2500 	*sizep = size;
2501 }
2502 
2503 static void
2504 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2505 {
2506 	struct proc *p;
2507 	size_t size;
2508 	int structsize;
2509 
2510 	p = (struct proc *)arg;
2511 	if (sb == NULL) {
2512 		size = 0;
2513 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2514 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2515 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2516 		PHOLD(p);
2517 		proc_getauxv(curthread, p, sb);
2518 		PRELE(p);
2519 		sbuf_finish(sb);
2520 		sbuf_delete(sb);
2521 		*sizep = size;
2522 	} else {
2523 		structsize = sizeof(Elf_Auxinfo);
2524 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2525 		PHOLD(p);
2526 		proc_getauxv(curthread, p, sb);
2527 		PRELE(p);
2528 	}
2529 }
2530 
2531 static boolean_t
2532 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2533     const char *note_vendor, const Elf_Phdr *pnote,
2534     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2535 {
2536 	const Elf_Note *note, *note0, *note_end;
2537 	const char *note_name;
2538 	char *buf;
2539 	int i, error;
2540 	boolean_t res;
2541 
2542 	/* We need some limit, might as well use PAGE_SIZE. */
2543 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2544 		return (FALSE);
2545 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2546 	if (pnote->p_offset > PAGE_SIZE ||
2547 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2548 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2549 		if (buf == NULL) {
2550 			VOP_UNLOCK(imgp->vp, 0);
2551 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2552 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2553 		}
2554 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2555 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2556 		    curthread->td_ucred, NOCRED, NULL, curthread);
2557 		if (error != 0) {
2558 			uprintf("i/o error PT_NOTE\n");
2559 			goto retf;
2560 		}
2561 		note = note0 = (const Elf_Note *)buf;
2562 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2563 	} else {
2564 		note = note0 = (const Elf_Note *)(imgp->image_header +
2565 		    pnote->p_offset);
2566 		note_end = (const Elf_Note *)(imgp->image_header +
2567 		    pnote->p_offset + pnote->p_filesz);
2568 		buf = NULL;
2569 	}
2570 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2571 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2572 		    (const char *)note < sizeof(Elf_Note)) {
2573 			goto retf;
2574 		}
2575 		if (note->n_namesz != checknote->n_namesz ||
2576 		    note->n_descsz != checknote->n_descsz ||
2577 		    note->n_type != checknote->n_type)
2578 			goto nextnote;
2579 		note_name = (const char *)(note + 1);
2580 		if (note_name + checknote->n_namesz >=
2581 		    (const char *)note_end || strncmp(note_vendor,
2582 		    note_name, checknote->n_namesz) != 0)
2583 			goto nextnote;
2584 
2585 		if (cb(note, cb_arg, &res))
2586 			goto ret;
2587 nextnote:
2588 		note = (const Elf_Note *)((const char *)(note + 1) +
2589 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2590 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2591 	}
2592 retf:
2593 	res = FALSE;
2594 ret:
2595 	free(buf, M_TEMP);
2596 	return (res);
2597 }
2598 
2599 struct brandnote_cb_arg {
2600 	Elf_Brandnote *brandnote;
2601 	int32_t *osrel;
2602 };
2603 
2604 static boolean_t
2605 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2606 {
2607 	struct brandnote_cb_arg *arg;
2608 
2609 	arg = arg0;
2610 
2611 	/*
2612 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2613 	 * necessary.
2614 	 */
2615 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2616 	    arg->brandnote->trans_osrel != NULL ?
2617 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2618 
2619 	return (TRUE);
2620 }
2621 
2622 static Elf_Note fctl_note = {
2623 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2624 	.n_descsz = sizeof(uint32_t),
2625 	.n_type = NT_FREEBSD_FEATURE_CTL,
2626 };
2627 
2628 struct fctl_cb_arg {
2629 	uint32_t *fctl0;
2630 };
2631 
2632 static boolean_t
2633 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2634 {
2635 	struct fctl_cb_arg *arg;
2636 	const Elf32_Word *desc;
2637 	uintptr_t p;
2638 
2639 	arg = arg0;
2640 	p = (uintptr_t)(note + 1);
2641 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2642 	desc = (const Elf32_Word *)p;
2643 	*arg->fctl0 = desc[0];
2644 	return (TRUE);
2645 }
2646 
2647 /*
2648  * Try to find the appropriate ABI-note section for checknote, fetch
2649  * the osreldate and feature control flags for binary from the ELF
2650  * OSABI-note.  Only the first page of the image is searched, the same
2651  * as for headers.
2652  */
2653 static boolean_t
2654 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2655     int32_t *osrel, uint32_t *fctl0)
2656 {
2657 	const Elf_Phdr *phdr;
2658 	const Elf_Ehdr *hdr;
2659 	struct brandnote_cb_arg b_arg;
2660 	struct fctl_cb_arg f_arg;
2661 	int i, j;
2662 
2663 	hdr = (const Elf_Ehdr *)imgp->image_header;
2664 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2665 	b_arg.brandnote = brandnote;
2666 	b_arg.osrel = osrel;
2667 	f_arg.fctl0 = fctl0;
2668 
2669 	for (i = 0; i < hdr->e_phnum; i++) {
2670 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2671 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2672 		    &b_arg)) {
2673 			for (j = 0; j < hdr->e_phnum; j++) {
2674 				if (phdr[j].p_type == PT_NOTE &&
2675 				    __elfN(parse_notes)(imgp, &fctl_note,
2676 				    FREEBSD_ABI_VENDOR, &phdr[j],
2677 				    note_fctl_cb, &f_arg))
2678 					break;
2679 			}
2680 			return (TRUE);
2681 		}
2682 	}
2683 	return (FALSE);
2684 
2685 }
2686 
2687 /*
2688  * Tell kern_execve.c about it, with a little help from the linker.
2689  */
2690 static struct execsw __elfN(execsw) = {
2691 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2692 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2693 };
2694 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2695 
2696 static vm_prot_t
2697 __elfN(trans_prot)(Elf_Word flags)
2698 {
2699 	vm_prot_t prot;
2700 
2701 	prot = 0;
2702 	if (flags & PF_X)
2703 		prot |= VM_PROT_EXECUTE;
2704 	if (flags & PF_W)
2705 		prot |= VM_PROT_WRITE;
2706 	if (flags & PF_R)
2707 		prot |= VM_PROT_READ;
2708 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2709 	if (i386_read_exec && (flags & PF_R))
2710 		prot |= VM_PROT_EXECUTE;
2711 #endif
2712 	return (prot);
2713 }
2714 
2715 static Elf_Word
2716 __elfN(untrans_prot)(vm_prot_t prot)
2717 {
2718 	Elf_Word flags;
2719 
2720 	flags = 0;
2721 	if (prot & VM_PROT_EXECUTE)
2722 		flags |= PF_X;
2723 	if (prot & VM_PROT_READ)
2724 		flags |= PF_R;
2725 	if (prot & VM_PROT_WRITE)
2726 		flags |= PF_W;
2727 	return (flags);
2728 }
2729 
2730 void
2731 __elfN(stackgap)(struct image_params *imgp, u_long *stack_base)
2732 {
2733 	u_long range, rbase, gap;
2734 	int pct;
2735 
2736 	if ((imgp->map_flags & MAP_ASLR) == 0)
2737 		return;
2738 	pct = __elfN(aslr_stack_gap);
2739 	if (pct == 0)
2740 		return;
2741 	if (pct > 50)
2742 		pct = 50;
2743 	range = imgp->eff_stack_sz * pct / 100;
2744 	arc4rand(&rbase, sizeof(rbase), 0);
2745 	gap = rbase % range;
2746 	gap &= ~(sizeof(u_long) - 1);
2747 	*stack_base -= gap;
2748 }
2749