xref: /freebsd/sys/kern/imgact_elf.c (revision da5069e1f7daaef1e7157876d6044de6f3a08ce2)
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 Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
160 
161 #define	aligned(a, t)	(rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
162 
163 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
164 
165 Elf_Brandnote __elfN(freebsd_brandnote) = {
166 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
167 	.hdr.n_descsz	= sizeof(int32_t),
168 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
169 	.vendor		= FREEBSD_ABI_VENDOR,
170 	.flags		= BN_TRANSLATE_OSREL,
171 	.trans_osrel	= __elfN(freebsd_trans_osrel)
172 };
173 
174 static bool
175 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
176 {
177 	uintptr_t p;
178 
179 	p = (uintptr_t)(note + 1);
180 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
181 	*osrel = *(const int32_t *)(p);
182 
183 	return (true);
184 }
185 
186 static const char GNU_ABI_VENDOR[] = "GNU";
187 static int GNU_KFREEBSD_ABI_DESC = 3;
188 
189 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
190 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
191 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
192 	.hdr.n_type	= 1,
193 	.vendor		= GNU_ABI_VENDOR,
194 	.flags		= BN_TRANSLATE_OSREL,
195 	.trans_osrel	= kfreebsd_trans_osrel
196 };
197 
198 static bool
199 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
200 {
201 	const Elf32_Word *desc;
202 	uintptr_t p;
203 
204 	p = (uintptr_t)(note + 1);
205 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
206 
207 	desc = (const Elf32_Word *)p;
208 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
209 		return (false);
210 
211 	/*
212 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
213 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
214 	 */
215 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
216 
217 	return (true);
218 }
219 
220 int
221 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
222 {
223 	int i;
224 
225 	for (i = 0; i < MAX_BRANDS; i++) {
226 		if (elf_brand_list[i] == NULL) {
227 			elf_brand_list[i] = entry;
228 			break;
229 		}
230 	}
231 	if (i == MAX_BRANDS) {
232 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
233 			__func__, entry);
234 		return (-1);
235 	}
236 	return (0);
237 }
238 
239 int
240 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
241 {
242 	int i;
243 
244 	for (i = 0; i < MAX_BRANDS; i++) {
245 		if (elf_brand_list[i] == entry) {
246 			elf_brand_list[i] = NULL;
247 			break;
248 		}
249 	}
250 	if (i == MAX_BRANDS)
251 		return (-1);
252 	return (0);
253 }
254 
255 int
256 __elfN(brand_inuse)(Elf_Brandinfo *entry)
257 {
258 	struct proc *p;
259 	int rval = FALSE;
260 
261 	sx_slock(&allproc_lock);
262 	FOREACH_PROC_IN_SYSTEM(p) {
263 		if (p->p_sysent == entry->sysvec) {
264 			rval = TRUE;
265 			break;
266 		}
267 	}
268 	sx_sunlock(&allproc_lock);
269 
270 	return (rval);
271 }
272 
273 static Elf_Brandinfo *
274 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
275     int32_t *osrel, uint32_t *fctl0)
276 {
277 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
278 	Elf_Brandinfo *bi, *bi_m;
279 	boolean_t ret;
280 	int i, interp_name_len;
281 
282 	interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
283 
284 	/*
285 	 * We support four types of branding -- (1) the ELF EI_OSABI field
286 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
287 	 * branding w/in the ELF header, (3) path of the `interp_path'
288 	 * field, and (4) the ".note.ABI-tag" ELF section.
289 	 */
290 
291 	/* Look for an ".note.ABI-tag" ELF section */
292 	bi_m = NULL;
293 	for (i = 0; i < MAX_BRANDS; i++) {
294 		bi = elf_brand_list[i];
295 		if (bi == NULL)
296 			continue;
297 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
298 			continue;
299 		if (hdr->e_machine == bi->machine && (bi->flags &
300 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
301 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
302 			    fctl0);
303 			/* Give brand a chance to veto check_note's guess */
304 			if (ret && bi->header_supported)
305 				ret = bi->header_supported(imgp);
306 			/*
307 			 * If note checker claimed the binary, but the
308 			 * interpreter path in the image does not
309 			 * match default one for the brand, try to
310 			 * search for other brands with the same
311 			 * interpreter.  Either there is better brand
312 			 * with the right interpreter, or, failing
313 			 * this, we return first brand which accepted
314 			 * our note and, optionally, header.
315 			 */
316 			if (ret && bi_m == NULL && interp != NULL &&
317 			    (bi->interp_path == NULL ||
318 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
319 			    strncmp(interp, bi->interp_path, interp_name_len)
320 			    != 0))) {
321 				bi_m = bi;
322 				ret = 0;
323 			}
324 			if (ret)
325 				return (bi);
326 		}
327 	}
328 	if (bi_m != NULL)
329 		return (bi_m);
330 
331 	/* If the executable has a brand, search for it in the brand list. */
332 	for (i = 0; i < MAX_BRANDS; i++) {
333 		bi = elf_brand_list[i];
334 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
335 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
336 			continue;
337 		if (hdr->e_machine == bi->machine &&
338 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
339 		    (bi->compat_3_brand != NULL &&
340 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
341 		    bi->compat_3_brand) == 0))) {
342 			/* Looks good, but give brand a chance to veto */
343 			if (bi->header_supported == NULL ||
344 			    bi->header_supported(imgp)) {
345 				/*
346 				 * Again, prefer strictly matching
347 				 * interpreter path.
348 				 */
349 				if (interp_name_len == 0 &&
350 				    bi->interp_path == NULL)
351 					return (bi);
352 				if (bi->interp_path != NULL &&
353 				    strlen(bi->interp_path) + 1 ==
354 				    interp_name_len && strncmp(interp,
355 				    bi->interp_path, interp_name_len) == 0)
356 					return (bi);
357 				if (bi_m == NULL)
358 					bi_m = bi;
359 			}
360 		}
361 	}
362 	if (bi_m != NULL)
363 		return (bi_m);
364 
365 	/* No known brand, see if the header is recognized by any brand */
366 	for (i = 0; i < MAX_BRANDS; i++) {
367 		bi = elf_brand_list[i];
368 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
369 		    bi->header_supported == NULL)
370 			continue;
371 		if (hdr->e_machine == bi->machine) {
372 			ret = bi->header_supported(imgp);
373 			if (ret)
374 				return (bi);
375 		}
376 	}
377 
378 	/* Lacking a known brand, search for a recognized interpreter. */
379 	if (interp != NULL) {
380 		for (i = 0; i < MAX_BRANDS; i++) {
381 			bi = elf_brand_list[i];
382 			if (bi == NULL || (bi->flags &
383 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
384 			    != 0)
385 				continue;
386 			if (hdr->e_machine == bi->machine &&
387 			    bi->interp_path != NULL &&
388 			    /* ELF image p_filesz includes terminating zero */
389 			    strlen(bi->interp_path) + 1 == interp_name_len &&
390 			    strncmp(interp, bi->interp_path, interp_name_len)
391 			    == 0 && (bi->header_supported == NULL ||
392 			    bi->header_supported(imgp)))
393 				return (bi);
394 		}
395 	}
396 
397 	/* Lacking a recognized interpreter, try the default brand */
398 	for (i = 0; i < MAX_BRANDS; i++) {
399 		bi = elf_brand_list[i];
400 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
401 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
402 			continue;
403 		if (hdr->e_machine == bi->machine &&
404 		    __elfN(fallback_brand) == bi->brand &&
405 		    (bi->header_supported == NULL ||
406 		    bi->header_supported(imgp)))
407 			return (bi);
408 	}
409 	return (NULL);
410 }
411 
412 static int
413 __elfN(check_header)(const Elf_Ehdr *hdr)
414 {
415 	Elf_Brandinfo *bi;
416 	int i;
417 
418 	if (!IS_ELF(*hdr) ||
419 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
420 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
421 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
422 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
423 	    hdr->e_version != ELF_TARG_VER)
424 		return (ENOEXEC);
425 
426 	/*
427 	 * Make sure we have at least one brand for this machine.
428 	 */
429 
430 	for (i = 0; i < MAX_BRANDS; i++) {
431 		bi = elf_brand_list[i];
432 		if (bi != NULL && bi->machine == hdr->e_machine)
433 			break;
434 	}
435 	if (i == MAX_BRANDS)
436 		return (ENOEXEC);
437 
438 	return (0);
439 }
440 
441 static int
442 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
443     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
444 {
445 	struct sf_buf *sf;
446 	int error;
447 	vm_offset_t off;
448 
449 	/*
450 	 * Create the page if it doesn't exist yet. Ignore errors.
451 	 */
452 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
453 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
454 
455 	/*
456 	 * Find the page from the underlying object.
457 	 */
458 	if (object != NULL) {
459 		sf = vm_imgact_map_page(object, offset);
460 		if (sf == NULL)
461 			return (KERN_FAILURE);
462 		off = offset - trunc_page(offset);
463 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
464 		    end - start);
465 		vm_imgact_unmap_page(sf);
466 		if (error != 0)
467 			return (KERN_FAILURE);
468 	}
469 
470 	return (KERN_SUCCESS);
471 }
472 
473 static int
474 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
475     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
476     int cow)
477 {
478 	struct sf_buf *sf;
479 	vm_offset_t off;
480 	vm_size_t sz;
481 	int error, locked, rv;
482 
483 	if (start != trunc_page(start)) {
484 		rv = __elfN(map_partial)(map, object, offset, start,
485 		    round_page(start), prot);
486 		if (rv != KERN_SUCCESS)
487 			return (rv);
488 		offset += round_page(start) - start;
489 		start = round_page(start);
490 	}
491 	if (end != round_page(end)) {
492 		rv = __elfN(map_partial)(map, object, offset +
493 		    trunc_page(end) - start, trunc_page(end), end, prot);
494 		if (rv != KERN_SUCCESS)
495 			return (rv);
496 		end = trunc_page(end);
497 	}
498 	if (start >= end)
499 		return (KERN_SUCCESS);
500 	if ((offset & PAGE_MASK) != 0) {
501 		/*
502 		 * The mapping is not page aligned.  This means that we have
503 		 * to copy the data.
504 		 */
505 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
506 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
507 		if (rv != KERN_SUCCESS)
508 			return (rv);
509 		if (object == NULL)
510 			return (KERN_SUCCESS);
511 		for (; start < end; start += sz) {
512 			sf = vm_imgact_map_page(object, offset);
513 			if (sf == NULL)
514 				return (KERN_FAILURE);
515 			off = offset - trunc_page(offset);
516 			sz = end - start;
517 			if (sz > PAGE_SIZE - off)
518 				sz = PAGE_SIZE - off;
519 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
520 			    (caddr_t)start, sz);
521 			vm_imgact_unmap_page(sf);
522 			if (error != 0)
523 				return (KERN_FAILURE);
524 			offset += sz;
525 		}
526 	} else {
527 		vm_object_reference(object);
528 		rv = vm_map_fixed(map, object, offset, start, end - start,
529 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
530 		if (rv != KERN_SUCCESS) {
531 			locked = VOP_ISLOCKED(imgp->vp);
532 			VOP_UNLOCK(imgp->vp, 0);
533 			vm_object_deallocate(object);
534 			vn_lock(imgp->vp, locked | LK_RETRY);
535 			return (rv);
536 		}
537 	}
538 	return (KERN_SUCCESS);
539 }
540 
541 static int
542 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
543     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
544 {
545 	struct sf_buf *sf;
546 	size_t map_len;
547 	vm_map_t map;
548 	vm_object_t object;
549 	vm_offset_t off, map_addr;
550 	int error, rv, cow;
551 	size_t copy_len;
552 	vm_ooffset_t file_addr;
553 
554 	/*
555 	 * It's necessary to fail if the filsz + offset taken from the
556 	 * header is greater than the actual file pager object's size.
557 	 * If we were to allow this, then the vm_map_find() below would
558 	 * walk right off the end of the file object and into the ether.
559 	 *
560 	 * While I'm here, might as well check for something else that
561 	 * is invalid: filsz cannot be greater than memsz.
562 	 */
563 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
564 	    filsz > memsz) {
565 		uprintf("elf_load_section: truncated ELF file\n");
566 		return (ENOEXEC);
567 	}
568 
569 	object = imgp->object;
570 	map = &imgp->proc->p_vmspace->vm_map;
571 	map_addr = trunc_page((vm_offset_t)vmaddr);
572 	file_addr = trunc_page(offset);
573 
574 	/*
575 	 * We have two choices.  We can either clear the data in the last page
576 	 * of an oversized mapping, or we can start the anon mapping a page
577 	 * early and copy the initialized data into that first page.  We
578 	 * choose the second.
579 	 */
580 	if (filsz == 0)
581 		map_len = 0;
582 	else if (memsz > filsz)
583 		map_len = trunc_page(offset + filsz) - file_addr;
584 	else
585 		map_len = round_page(offset + filsz) - file_addr;
586 
587 	if (map_len != 0) {
588 		/* cow flags: don't dump readonly sections in core */
589 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
590 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
591 
592 		rv = __elfN(map_insert)(imgp, map,
593 				      object,
594 				      file_addr,	/* file offset */
595 				      map_addr,		/* virtual start */
596 				      map_addr + map_len,/* virtual end */
597 				      prot,
598 				      cow);
599 		if (rv != KERN_SUCCESS)
600 			return (EINVAL);
601 
602 		/* we can stop now if we've covered it all */
603 		if (memsz == filsz)
604 			return (0);
605 	}
606 
607 
608 	/*
609 	 * We have to get the remaining bit of the file into the first part
610 	 * of the oversized map segment.  This is normally because the .data
611 	 * segment in the file is extended to provide bss.  It's a neat idea
612 	 * to try and save a page, but it's a pain in the behind to implement.
613 	 */
614 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
615 	    filsz);
616 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
617 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
618 
619 	/* This had damn well better be true! */
620 	if (map_len != 0) {
621 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
622 		    map_addr + map_len, prot, 0);
623 		if (rv != KERN_SUCCESS)
624 			return (EINVAL);
625 	}
626 
627 	if (copy_len != 0) {
628 		sf = vm_imgact_map_page(object, offset + filsz);
629 		if (sf == NULL)
630 			return (EIO);
631 
632 		/* send the page fragment to user space */
633 		off = trunc_page(offset + filsz) - trunc_page(offset + filsz);
634 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
635 		    (caddr_t)map_addr, copy_len);
636 		vm_imgact_unmap_page(sf);
637 		if (error != 0)
638 			return (error);
639 	}
640 
641 	/*
642 	 * Remove write access to the page if it was only granted by map_insert
643 	 * to allow copyout.
644 	 */
645 	if ((prot & VM_PROT_WRITE) == 0)
646 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
647 		    map_len), prot, FALSE);
648 
649 	return (0);
650 }
651 
652 static int
653 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
654     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
655 {
656 	vm_prot_t prot;
657 	u_long base_addr;
658 	bool first;
659 	int error, i;
660 
661 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
662 
663 	base_addr = 0;
664 	first = true;
665 
666 	for (i = 0; i < hdr->e_phnum; i++) {
667 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
668 			continue;
669 
670 		/* Loadable segment */
671 		prot = __elfN(trans_prot)(phdr[i].p_flags);
672 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
673 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
674 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
675 		if (error != 0)
676 			return (error);
677 
678 		/*
679 		 * Establish the base address if this is the first segment.
680 		 */
681 		if (first) {
682   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
683 			first = false;
684 		}
685 	}
686 
687 	if (base_addrp != NULL)
688 		*base_addrp = base_addr;
689 
690 	return (0);
691 }
692 
693 /*
694  * Load the file "file" into memory.  It may be either a shared object
695  * or an executable.
696  *
697  * The "addr" reference parameter is in/out.  On entry, it specifies
698  * the address where a shared object should be loaded.  If the file is
699  * an executable, this value is ignored.  On exit, "addr" specifies
700  * where the file was actually loaded.
701  *
702  * The "entry" reference parameter is out only.  On exit, it specifies
703  * the entry point for the loaded file.
704  */
705 static int
706 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
707 	u_long *entry)
708 {
709 	struct {
710 		struct nameidata nd;
711 		struct vattr attr;
712 		struct image_params image_params;
713 	} *tempdata;
714 	const Elf_Ehdr *hdr = NULL;
715 	const Elf_Phdr *phdr = NULL;
716 	struct nameidata *nd;
717 	struct vattr *attr;
718 	struct image_params *imgp;
719 	u_long flags, rbase;
720 	u_long base_addr = 0;
721 	int error;
722 
723 #ifdef CAPABILITY_MODE
724 	/*
725 	 * XXXJA: This check can go away once we are sufficiently confident
726 	 * that the checks in namei() are correct.
727 	 */
728 	if (IN_CAPABILITY_MODE(curthread))
729 		return (ECAPMODE);
730 #endif
731 
732 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
733 	nd = &tempdata->nd;
734 	attr = &tempdata->attr;
735 	imgp = &tempdata->image_params;
736 
737 	/*
738 	 * Initialize part of the common data
739 	 */
740 	imgp->proc = p;
741 	imgp->attr = attr;
742 	imgp->firstpage = NULL;
743 	imgp->image_header = NULL;
744 	imgp->object = NULL;
745 	imgp->execlabel = NULL;
746 
747 	flags = FOLLOW | LOCKSHARED | LOCKLEAF;
748 
749 again:
750 	NDINIT(nd, LOOKUP, flags, UIO_SYSSPACE, file, curthread);
751 	if ((error = namei(nd)) != 0) {
752 		nd->ni_vp = NULL;
753 		goto fail;
754 	}
755 	NDFREE(nd, NDF_ONLY_PNBUF);
756 	imgp->vp = nd->ni_vp;
757 
758 	/*
759 	 * Check permissions, modes, uid, etc on the file, and "open" it.
760 	 */
761 	error = exec_check_permissions(imgp);
762 	if (error)
763 		goto fail;
764 
765 	/*
766 	 * Also make certain that the interpreter stays the same,
767 	 * so set its VV_TEXT flag, too.  Since this function is only
768 	 * used to load the interpreter, the VV_TEXT is almost always
769 	 * already set.
770 	 */
771 	if (VOP_IS_TEXT(nd->ni_vp) == 0) {
772 		if (VOP_ISLOCKED(nd->ni_vp) != LK_EXCLUSIVE) {
773 			/*
774 			 * LK_UPGRADE could have resulted in dropping
775 			 * the lock.  Just try again from the start,
776 			 * this time with exclusive vnode lock.
777 			 */
778 			vput(nd->ni_vp);
779 			flags &= ~LOCKSHARED;
780 			goto again;
781 		}
782 
783 		VOP_SET_TEXT(nd->ni_vp);
784 	}
785 
786 	error = exec_map_first_page(imgp);
787 	if (error)
788 		goto fail;
789 
790 	imgp->object = nd->ni_vp->v_object;
791 
792 	hdr = (const Elf_Ehdr *)imgp->image_header;
793 	if ((error = __elfN(check_header)(hdr)) != 0)
794 		goto fail;
795 	if (hdr->e_type == ET_DYN)
796 		rbase = *addr;
797 	else if (hdr->e_type == ET_EXEC)
798 		rbase = 0;
799 	else {
800 		error = ENOEXEC;
801 		goto fail;
802 	}
803 
804 	/* Only support headers that fit within first page for now      */
805 	if ((hdr->e_phoff > PAGE_SIZE) ||
806 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
807 		error = ENOEXEC;
808 		goto fail;
809 	}
810 
811 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
812 	if (!aligned(phdr, Elf_Addr)) {
813 		error = ENOEXEC;
814 		goto fail;
815 	}
816 
817 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
818 	if (error != 0)
819 		goto fail;
820 
821 	*addr = base_addr;
822 	*entry = (unsigned long)hdr->e_entry + rbase;
823 
824 fail:
825 	if (imgp->firstpage)
826 		exec_unmap_first_page(imgp);
827 
828 	if (nd->ni_vp)
829 		vput(nd->ni_vp);
830 
831 	free(tempdata, M_TEMP);
832 
833 	return (error);
834 }
835 
836 static u_long
837 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
838     u_int align)
839 {
840 	u_long rbase, res;
841 
842 	MPASS(vm_map_min(map) <= minv);
843 	MPASS(maxv <= vm_map_max(map));
844 	MPASS(minv < maxv);
845 	MPASS(minv + align < maxv);
846 	arc4rand(&rbase, sizeof(rbase), 0);
847 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
848 	res &= ~((u_long)align - 1);
849 	if (res >= maxv)
850 		res -= align;
851 	KASSERT(res >= minv,
852 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
853 	    res, minv, maxv, rbase));
854 	KASSERT(res < maxv,
855 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
856 	    res, maxv, minv, rbase));
857 	return (res);
858 }
859 
860 static int
861 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
862     const Elf_Phdr *phdr, u_long et_dyn_addr)
863 {
864 	struct vmspace *vmspace;
865 	const char *err_str;
866 	u_long text_size, data_size, total_size, text_addr, data_addr;
867 	u_long seg_size, seg_addr;
868 	int i;
869 
870 	err_str = NULL;
871 	text_size = data_size = total_size = text_addr = data_addr = 0;
872 
873 	for (i = 0; i < hdr->e_phnum; i++) {
874 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
875 			continue;
876 
877 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
878 		seg_size = round_page(phdr[i].p_memsz +
879 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
880 
881 		/*
882 		 * Make the largest executable segment the official
883 		 * text segment and all others data.
884 		 *
885 		 * Note that obreak() assumes that data_addr + data_size == end
886 		 * of data load area, and the ELF file format expects segments
887 		 * to be sorted by address.  If multiple data segments exist,
888 		 * the last one will be used.
889 		 */
890 
891 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
892 			text_size = seg_size;
893 			text_addr = seg_addr;
894 		} else {
895 			data_size = seg_size;
896 			data_addr = seg_addr;
897 		}
898 		total_size += seg_size;
899 	}
900 
901 	if (data_addr == 0 && data_size == 0) {
902 		data_addr = text_addr;
903 		data_size = text_size;
904 	}
905 
906 	/*
907 	 * Check limits.  It should be safe to check the
908 	 * limits after loading the segments since we do
909 	 * not actually fault in all the segments pages.
910 	 */
911 	PROC_LOCK(imgp->proc);
912 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
913 		err_str = "Data segment size exceeds process limit";
914 	else if (text_size > maxtsiz)
915 		err_str = "Text segment size exceeds system limit";
916 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
917 		err_str = "Total segment size exceeds process limit";
918 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
919 		err_str = "Data segment size exceeds resource limit";
920 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
921 		err_str = "Total segment size exceeds resource limit";
922 	PROC_UNLOCK(imgp->proc);
923 	if (err_str != NULL) {
924 		uprintf("%s\n", err_str);
925 		return (ENOMEM);
926 	}
927 
928 	vmspace = imgp->proc->p_vmspace;
929 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
930 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
931 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
932 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
933 
934 	return (0);
935 }
936 
937 static int
938 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
939     char **interpp, bool *free_interpp)
940 {
941 	struct thread *td;
942 	char *interp;
943 	int error, interp_name_len;
944 
945 	KASSERT(phdr->p_type == PT_INTERP,
946 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
947 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
948 
949 	td = curthread;
950 
951 	/* Path to interpreter */
952 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
953 		uprintf("Invalid PT_INTERP\n");
954 		return (ENOEXEC);
955 	}
956 
957 	interp_name_len = phdr->p_filesz;
958 	if (phdr->p_offset > PAGE_SIZE ||
959 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
960 		VOP_UNLOCK(imgp->vp, 0);
961 		interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
962 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
963 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
964 		    interp_name_len, phdr->p_offset,
965 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
966 		    NOCRED, NULL, td);
967 		if (error != 0) {
968 			free(interp, M_TEMP);
969 			uprintf("i/o error PT_INTERP %d\n", error);
970 			return (error);
971 		}
972 		interp[interp_name_len] = '\0';
973 
974 		*interpp = interp;
975 		*free_interpp = true;
976 		return (0);
977 	}
978 
979 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
980 	if (interp[interp_name_len - 1] != '\0') {
981 		uprintf("Invalid PT_INTERP\n");
982 		return (ENOEXEC);
983 	}
984 
985 	*interpp = interp;
986 	*free_interpp = false;
987 	return (0);
988 }
989 
990 static int
991 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
992     const char *interp, u_long *addr, u_long *entry)
993 {
994 	char *path;
995 	int error;
996 
997 	if (brand_info->emul_path != NULL &&
998 	    brand_info->emul_path[0] != '\0') {
999 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1000 		snprintf(path, MAXPATHLEN, "%s%s",
1001 		    brand_info->emul_path, interp);
1002 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
1003 		free(path, M_TEMP);
1004 		if (error == 0)
1005 			return (0);
1006 	}
1007 
1008 	if (brand_info->interp_newpath != NULL &&
1009 	    (brand_info->interp_path == NULL ||
1010 	    strcmp(interp, brand_info->interp_path) == 0)) {
1011 		error = __elfN(load_file)(imgp->proc,
1012 		    brand_info->interp_newpath, addr, entry);
1013 		if (error == 0)
1014 			return (0);
1015 	}
1016 
1017 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1018 	if (error == 0)
1019 		return (0);
1020 
1021 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1022 	return (error);
1023 }
1024 
1025 /*
1026  * Impossible et_dyn_addr initial value indicating that the real base
1027  * must be calculated later with some randomization applied.
1028  */
1029 #define	ET_DYN_ADDR_RAND	1
1030 
1031 static int
1032 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1033 {
1034 	struct thread *td;
1035 	const Elf_Ehdr *hdr;
1036 	const Elf_Phdr *phdr;
1037 	Elf_Auxargs *elf_auxargs;
1038 	struct vmspace *vmspace;
1039 	vm_map_t map;
1040 	char *interp;
1041 	Elf_Brandinfo *brand_info;
1042 	struct sysentvec *sv;
1043 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1044 	u_long maxalign, mapsz, maxv, maxv1;
1045 	uint32_t fctl0;
1046 	int32_t osrel;
1047 	bool free_interp;
1048 	int error, i, n;
1049 
1050 	hdr = (const Elf_Ehdr *)imgp->image_header;
1051 
1052 	/*
1053 	 * Do we have a valid ELF header ?
1054 	 *
1055 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1056 	 * if particular brand doesn't support it.
1057 	 */
1058 	if (__elfN(check_header)(hdr) != 0 ||
1059 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1060 		return (-1);
1061 
1062 	/*
1063 	 * From here on down, we return an errno, not -1, as we've
1064 	 * detected an ELF file.
1065 	 */
1066 
1067 	if ((hdr->e_phoff > PAGE_SIZE) ||
1068 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1069 		/* Only support headers in first page for now */
1070 		uprintf("Program headers not in the first page\n");
1071 		return (ENOEXEC);
1072 	}
1073 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1074 	if (!aligned(phdr, Elf_Addr)) {
1075 		uprintf("Unaligned program headers\n");
1076 		return (ENOEXEC);
1077 	}
1078 
1079 	n = error = 0;
1080 	baddr = 0;
1081 	osrel = 0;
1082 	fctl0 = 0;
1083 	entry = proghdr = 0;
1084 	interp = NULL;
1085 	free_interp = false;
1086 	td = curthread;
1087 	maxalign = PAGE_SIZE;
1088 	mapsz = 0;
1089 
1090 	for (i = 0; i < hdr->e_phnum; i++) {
1091 		switch (phdr[i].p_type) {
1092 		case PT_LOAD:
1093 			if (n == 0)
1094 				baddr = phdr[i].p_vaddr;
1095 			if (phdr[i].p_align > maxalign)
1096 				maxalign = phdr[i].p_align;
1097 			mapsz += phdr[i].p_memsz;
1098 			n++;
1099 
1100 			/*
1101 			 * If this segment contains the program headers,
1102 			 * remember their virtual address for the AT_PHDR
1103 			 * aux entry. Static binaries don't usually include
1104 			 * a PT_PHDR entry.
1105 			 */
1106 			if (phdr[i].p_offset == 0 &&
1107 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1108 				<= phdr[i].p_filesz)
1109 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1110 			break;
1111 		case PT_INTERP:
1112 			/* Path to interpreter */
1113 			if (interp != NULL) {
1114 				uprintf("Multiple PT_INTERP headers\n");
1115 				error = ENOEXEC;
1116 				goto ret;
1117 			}
1118 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1119 			    &free_interp);
1120 			if (error != 0)
1121 				goto ret;
1122 			break;
1123 		case PT_GNU_STACK:
1124 			if (__elfN(nxstack))
1125 				imgp->stack_prot =
1126 				    __elfN(trans_prot)(phdr[i].p_flags);
1127 			imgp->stack_sz = phdr[i].p_memsz;
1128 			break;
1129 		case PT_PHDR: 	/* Program header table info */
1130 			proghdr = phdr[i].p_vaddr;
1131 			break;
1132 		}
1133 	}
1134 
1135 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1136 	if (brand_info == NULL) {
1137 		uprintf("ELF binary type \"%u\" not known.\n",
1138 		    hdr->e_ident[EI_OSABI]);
1139 		error = ENOEXEC;
1140 		goto ret;
1141 	}
1142 	sv = brand_info->sysvec;
1143 	et_dyn_addr = 0;
1144 	if (hdr->e_type == ET_DYN) {
1145 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1146 			uprintf("Cannot execute shared object\n");
1147 			error = ENOEXEC;
1148 			goto ret;
1149 		}
1150 		/*
1151 		 * Honour the base load address from the dso if it is
1152 		 * non-zero for some reason.
1153 		 */
1154 		if (baddr == 0) {
1155 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1156 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1157 				et_dyn_addr = ET_DYN_LOAD_ADDR;
1158 			else if ((__elfN(pie_aslr_enabled) &&
1159 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1160 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1161 				et_dyn_addr = ET_DYN_ADDR_RAND;
1162 			else
1163 				et_dyn_addr = ET_DYN_LOAD_ADDR;
1164 		}
1165 	}
1166 
1167 	/*
1168 	 * Avoid a possible deadlock if the current address space is destroyed
1169 	 * and that address space maps the locked vnode.  In the common case,
1170 	 * the locked vnode's v_usecount is decremented but remains greater
1171 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1172 	 * However, in cases where the vnode lock is external, such as nullfs,
1173 	 * v_usecount may become zero.
1174 	 *
1175 	 * The VV_TEXT flag prevents modifications to the executable while
1176 	 * the vnode is unlocked.
1177 	 */
1178 	VOP_UNLOCK(imgp->vp, 0);
1179 
1180 	/*
1181 	 * Decide whether to enable randomization of user mappings.
1182 	 * First, reset user preferences for the setid binaries.
1183 	 * Then, account for the support of the randomization by the
1184 	 * ABI, by user preferences, and make special treatment for
1185 	 * PIE binaries.
1186 	 */
1187 	if (imgp->credential_setid) {
1188 		PROC_LOCK(imgp->proc);
1189 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1190 		PROC_UNLOCK(imgp->proc);
1191 	}
1192 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1193 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1194 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1195 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1196 		    ("et_dyn_addr == RAND and !ASLR"));
1197 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1198 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1199 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1200 		imgp->map_flags |= MAP_ASLR;
1201 		/*
1202 		 * If user does not care about sbrk, utilize the bss
1203 		 * grow region for mappings as well.  We can select
1204 		 * the base for the image anywere and still not suffer
1205 		 * from the fragmentation.
1206 		 */
1207 		if (!__elfN(aslr_honor_sbrk) ||
1208 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1209 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1210 	}
1211 
1212 	error = exec_new_vmspace(imgp, sv);
1213 	vmspace = imgp->proc->p_vmspace;
1214 	map = &vmspace->vm_map;
1215 
1216 	imgp->proc->p_sysent = sv;
1217 
1218 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1219 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1220 		KASSERT((map->flags & MAP_ASLR) != 0,
1221 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1222 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1223 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1224 		    /* reserve half of the address space to interpreter */
1225 		    maxv / 2, 1UL << flsl(maxalign));
1226 	}
1227 
1228 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1229 	if (error != 0)
1230 		goto ret;
1231 
1232 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1233 	if (error != 0)
1234 		goto ret;
1235 
1236 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1237 	if (error != 0)
1238 		goto ret;
1239 
1240 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1241 
1242 	/*
1243 	 * We load the dynamic linker where a userland call
1244 	 * to mmap(0, ...) would put it.  The rationale behind this
1245 	 * calculation is that it leaves room for the heap to grow to
1246 	 * its maximum allowed size.
1247 	 */
1248 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1249 	    RLIMIT_DATA));
1250 	if ((map->flags & MAP_ASLR) != 0) {
1251 		maxv1 = maxv / 2 + addr / 2;
1252 		MPASS(maxv1 >= addr);	/* No overflow */
1253 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1254 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1255 	} else {
1256 		map->anon_loc = addr;
1257 	}
1258 
1259 	imgp->entry_addr = entry;
1260 
1261 	if (interp != NULL) {
1262 		VOP_UNLOCK(imgp->vp, 0);
1263 		if ((map->flags & MAP_ASLR) != 0) {
1264 			/* Assume that interpeter fits into 1/4 of AS */
1265 			maxv1 = maxv / 2 + addr / 2;
1266 			MPASS(maxv1 >= addr);	/* No overflow */
1267 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1268 			    maxv1, PAGE_SIZE);
1269 		}
1270 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1271 		    &imgp->entry_addr);
1272 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1273 		if (error != 0)
1274 			goto ret;
1275 	} else
1276 		addr = et_dyn_addr;
1277 
1278 	/*
1279 	 * Construct auxargs table (used by the fixup routine)
1280 	 */
1281 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1282 	elf_auxargs->execfd = -1;
1283 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1284 	elf_auxargs->phent = hdr->e_phentsize;
1285 	elf_auxargs->phnum = hdr->e_phnum;
1286 	elf_auxargs->pagesz = PAGE_SIZE;
1287 	elf_auxargs->base = addr;
1288 	elf_auxargs->flags = 0;
1289 	elf_auxargs->entry = entry;
1290 	elf_auxargs->hdr_eflags = hdr->e_flags;
1291 
1292 	imgp->auxargs = elf_auxargs;
1293 	imgp->interpreted = 0;
1294 	imgp->reloc_base = addr;
1295 	imgp->proc->p_osrel = osrel;
1296 	imgp->proc->p_fctl0 = fctl0;
1297 	imgp->proc->p_elf_machine = hdr->e_machine;
1298 	imgp->proc->p_elf_flags = hdr->e_flags;
1299 
1300 ret:
1301 	if (free_interp)
1302 		free(interp, M_TEMP);
1303 	return (error);
1304 }
1305 
1306 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1307 
1308 int
1309 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1310 {
1311 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1312 	Elf_Auxinfo *argarray, *pos;
1313 	Elf_Addr *base, *auxbase;
1314 	int error;
1315 
1316 	base = (Elf_Addr *)*stack_base;
1317 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1318 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1319 	    M_WAITOK | M_ZERO);
1320 
1321 	if (args->execfd != -1)
1322 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1323 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1324 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1325 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1326 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1327 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1328 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1329 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1330 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1331 	if (imgp->execpathp != 0)
1332 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1333 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1334 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1335 	if (imgp->canary != 0) {
1336 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1337 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1338 	}
1339 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1340 	if (imgp->pagesizes != 0) {
1341 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1342 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1343 	}
1344 	if (imgp->sysent->sv_timekeep_base != 0) {
1345 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1346 		    imgp->sysent->sv_timekeep_base);
1347 	}
1348 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1349 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1350 	    imgp->sysent->sv_stackprot);
1351 	if (imgp->sysent->sv_hwcap != NULL)
1352 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1353 	if (imgp->sysent->sv_hwcap2 != NULL)
1354 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1355 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1356 
1357 	free(imgp->auxargs, M_TEMP);
1358 	imgp->auxargs = NULL;
1359 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1360 
1361 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1362 	free(argarray, M_TEMP);
1363 	if (error != 0)
1364 		return (error);
1365 
1366 	base--;
1367 	if (suword(base, imgp->args->argc) == -1)
1368 		return (EFAULT);
1369 	*stack_base = (register_t *)base;
1370 	return (0);
1371 }
1372 
1373 /*
1374  * Code for generating ELF core dumps.
1375  */
1376 
1377 typedef void (*segment_callback)(vm_map_entry_t, void *);
1378 
1379 /* Closure for cb_put_phdr(). */
1380 struct phdr_closure {
1381 	Elf_Phdr *phdr;		/* Program header to fill in */
1382 	Elf_Off offset;		/* Offset of segment in core file */
1383 };
1384 
1385 /* Closure for cb_size_segment(). */
1386 struct sseg_closure {
1387 	int count;		/* Count of writable segments. */
1388 	size_t size;		/* Total size of all writable segments. */
1389 };
1390 
1391 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1392 
1393 struct note_info {
1394 	int		type;		/* Note type. */
1395 	outfunc_t 	outfunc; 	/* Output function. */
1396 	void		*outarg;	/* Argument for the output function. */
1397 	size_t		outsize;	/* Output size. */
1398 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1399 };
1400 
1401 TAILQ_HEAD(note_info_list, note_info);
1402 
1403 /* Coredump output parameters. */
1404 struct coredump_params {
1405 	off_t		offset;
1406 	struct ucred	*active_cred;
1407 	struct ucred	*file_cred;
1408 	struct thread	*td;
1409 	struct vnode	*vp;
1410 	struct compressor *comp;
1411 };
1412 
1413 extern int compress_user_cores;
1414 extern int compress_user_cores_level;
1415 
1416 static void cb_put_phdr(vm_map_entry_t, void *);
1417 static void cb_size_segment(vm_map_entry_t, void *);
1418 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1419     enum uio_seg);
1420 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1421 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1422     struct note_info_list *, size_t);
1423 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1424     size_t *);
1425 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1426 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1427 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1428 static int sbuf_drain_core_output(void *, const char *, int);
1429 static int sbuf_drain_count(void *arg, const char *data, int len);
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 /*
1560  * Drain into a counter.
1561  */
1562 static int
1563 sbuf_drain_count(void *arg, const char *data __unused, int len)
1564 {
1565 	size_t *sizep;
1566 
1567 	sizep = (size_t *)arg;
1568 	*sizep += len;
1569 	return (len);
1570 }
1571 
1572 int
1573 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1574 {
1575 	struct ucred *cred = td->td_ucred;
1576 	int error = 0;
1577 	struct sseg_closure seginfo;
1578 	struct note_info_list notelst;
1579 	struct coredump_params params;
1580 	struct note_info *ninfo;
1581 	void *hdr, *tmpbuf;
1582 	size_t hdrsize, notesz, coresize;
1583 
1584 	hdr = NULL;
1585 	tmpbuf = NULL;
1586 	TAILQ_INIT(&notelst);
1587 
1588 	/* Size the program segments. */
1589 	seginfo.count = 0;
1590 	seginfo.size = 0;
1591 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1592 
1593 	/*
1594 	 * Collect info about the core file header area.
1595 	 */
1596 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1597 	if (seginfo.count + 1 >= PN_XNUM)
1598 		hdrsize += sizeof(Elf_Shdr);
1599 	__elfN(prepare_notes)(td, &notelst, &notesz);
1600 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1601 
1602 	/* Set up core dump parameters. */
1603 	params.offset = 0;
1604 	params.active_cred = cred;
1605 	params.file_cred = NOCRED;
1606 	params.td = td;
1607 	params.vp = vp;
1608 	params.comp = NULL;
1609 
1610 #ifdef RACCT
1611 	if (racct_enable) {
1612 		PROC_LOCK(td->td_proc);
1613 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1614 		PROC_UNLOCK(td->td_proc);
1615 		if (error != 0) {
1616 			error = EFAULT;
1617 			goto done;
1618 		}
1619 	}
1620 #endif
1621 	if (coresize >= limit) {
1622 		error = EFAULT;
1623 		goto done;
1624 	}
1625 
1626 	/* Create a compression stream if necessary. */
1627 	if (compress_user_cores != 0) {
1628 		params.comp = compressor_init(core_compressed_write,
1629 		    compress_user_cores, CORE_BUF_SIZE,
1630 		    compress_user_cores_level, &params);
1631 		if (params.comp == NULL) {
1632 			error = EFAULT;
1633 			goto done;
1634 		}
1635 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1636         }
1637 
1638 	/*
1639 	 * Allocate memory for building the header, fill it up,
1640 	 * and write it out following the notes.
1641 	 */
1642 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1643 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1644 	    notesz);
1645 
1646 	/* Write the contents of all of the writable segments. */
1647 	if (error == 0) {
1648 		Elf_Phdr *php;
1649 		off_t offset;
1650 		int i;
1651 
1652 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1653 		offset = round_page(hdrsize + notesz);
1654 		for (i = 0; i < seginfo.count; i++) {
1655 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1656 			    php->p_filesz, offset, &params, tmpbuf);
1657 			if (error != 0)
1658 				break;
1659 			offset += php->p_filesz;
1660 			php++;
1661 		}
1662 		if (error == 0 && params.comp != NULL)
1663 			error = compressor_flush(params.comp);
1664 	}
1665 	if (error) {
1666 		log(LOG_WARNING,
1667 		    "Failed to write core file for process %s (error %d)\n",
1668 		    curproc->p_comm, error);
1669 	}
1670 
1671 done:
1672 	free(tmpbuf, M_TEMP);
1673 	if (params.comp != NULL)
1674 		compressor_fini(params.comp);
1675 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1676 		TAILQ_REMOVE(&notelst, ninfo, link);
1677 		free(ninfo, M_TEMP);
1678 	}
1679 	if (hdr != NULL)
1680 		free(hdr, M_TEMP);
1681 
1682 	return (error);
1683 }
1684 
1685 /*
1686  * A callback for each_dumpable_segment() to write out the segment's
1687  * program header entry.
1688  */
1689 static void
1690 cb_put_phdr(vm_map_entry_t entry, void *closure)
1691 {
1692 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1693 	Elf_Phdr *phdr = phc->phdr;
1694 
1695 	phc->offset = round_page(phc->offset);
1696 
1697 	phdr->p_type = PT_LOAD;
1698 	phdr->p_offset = phc->offset;
1699 	phdr->p_vaddr = entry->start;
1700 	phdr->p_paddr = 0;
1701 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1702 	phdr->p_align = PAGE_SIZE;
1703 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1704 
1705 	phc->offset += phdr->p_filesz;
1706 	phc->phdr++;
1707 }
1708 
1709 /*
1710  * A callback for each_dumpable_segment() to gather information about
1711  * the number of segments and their total size.
1712  */
1713 static void
1714 cb_size_segment(vm_map_entry_t entry, void *closure)
1715 {
1716 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1717 
1718 	ssc->count++;
1719 	ssc->size += entry->end - entry->start;
1720 }
1721 
1722 /*
1723  * For each writable segment in the process's memory map, call the given
1724  * function with a pointer to the map entry and some arbitrary
1725  * caller-supplied data.
1726  */
1727 static void
1728 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1729 {
1730 	struct proc *p = td->td_proc;
1731 	vm_map_t map = &p->p_vmspace->vm_map;
1732 	vm_map_entry_t entry;
1733 	vm_object_t backing_object, object;
1734 	boolean_t ignore_entry;
1735 
1736 	vm_map_lock_read(map);
1737 	for (entry = map->header.next; entry != &map->header;
1738 	    entry = entry->next) {
1739 		/*
1740 		 * Don't dump inaccessible mappings, deal with legacy
1741 		 * coredump mode.
1742 		 *
1743 		 * Note that read-only segments related to the elf binary
1744 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1745 		 * need to arbitrarily ignore such segments.
1746 		 */
1747 		if (elf_legacy_coredump) {
1748 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1749 				continue;
1750 		} else {
1751 			if ((entry->protection & VM_PROT_ALL) == 0)
1752 				continue;
1753 		}
1754 
1755 		/*
1756 		 * Dont include memory segment in the coredump if
1757 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1758 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1759 		 * kernel map).
1760 		 */
1761 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1762 			continue;
1763 
1764 		if ((object = entry->object.vm_object) == NULL)
1765 			continue;
1766 
1767 		/* Ignore memory-mapped devices and such things. */
1768 		VM_OBJECT_RLOCK(object);
1769 		while ((backing_object = object->backing_object) != NULL) {
1770 			VM_OBJECT_RLOCK(backing_object);
1771 			VM_OBJECT_RUNLOCK(object);
1772 			object = backing_object;
1773 		}
1774 		ignore_entry = object->type != OBJT_DEFAULT &&
1775 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1776 		    object->type != OBJT_PHYS;
1777 		VM_OBJECT_RUNLOCK(object);
1778 		if (ignore_entry)
1779 			continue;
1780 
1781 		(*func)(entry, closure);
1782 	}
1783 	vm_map_unlock_read(map);
1784 }
1785 
1786 /*
1787  * Write the core file header to the file, including padding up to
1788  * the page boundary.
1789  */
1790 static int
1791 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1792     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1793 {
1794 	struct note_info *ninfo;
1795 	struct sbuf *sb;
1796 	int error;
1797 
1798 	/* Fill in the header. */
1799 	bzero(hdr, hdrsize);
1800 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1801 
1802 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1803 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1804 	sbuf_start_section(sb, NULL);
1805 	sbuf_bcat(sb, hdr, hdrsize);
1806 	TAILQ_FOREACH(ninfo, notelst, link)
1807 	    __elfN(putnote)(ninfo, sb);
1808 	/* Align up to a page boundary for the program segments. */
1809 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1810 	error = sbuf_finish(sb);
1811 	sbuf_delete(sb);
1812 
1813 	return (error);
1814 }
1815 
1816 static void
1817 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1818     size_t *sizep)
1819 {
1820 	struct proc *p;
1821 	struct thread *thr;
1822 	size_t size;
1823 
1824 	p = td->td_proc;
1825 	size = 0;
1826 
1827 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1828 
1829 	/*
1830 	 * To have the debugger select the right thread (LWP) as the initial
1831 	 * thread, we dump the state of the thread passed to us in td first.
1832 	 * This is the thread that causes the core dump and thus likely to
1833 	 * be the right thread one wants to have selected in the debugger.
1834 	 */
1835 	thr = td;
1836 	while (thr != NULL) {
1837 		size += register_note(list, NT_PRSTATUS,
1838 		    __elfN(note_prstatus), thr);
1839 		size += register_note(list, NT_FPREGSET,
1840 		    __elfN(note_fpregset), thr);
1841 		size += register_note(list, NT_THRMISC,
1842 		    __elfN(note_thrmisc), thr);
1843 		size += register_note(list, NT_PTLWPINFO,
1844 		    __elfN(note_ptlwpinfo), thr);
1845 		size += register_note(list, -1,
1846 		    __elfN(note_threadmd), thr);
1847 
1848 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1849 		    TAILQ_NEXT(thr, td_plist);
1850 		if (thr == td)
1851 			thr = TAILQ_NEXT(thr, td_plist);
1852 	}
1853 
1854 	size += register_note(list, NT_PROCSTAT_PROC,
1855 	    __elfN(note_procstat_proc), p);
1856 	size += register_note(list, NT_PROCSTAT_FILES,
1857 	    note_procstat_files, p);
1858 	size += register_note(list, NT_PROCSTAT_VMMAP,
1859 	    note_procstat_vmmap, p);
1860 	size += register_note(list, NT_PROCSTAT_GROUPS,
1861 	    note_procstat_groups, p);
1862 	size += register_note(list, NT_PROCSTAT_UMASK,
1863 	    note_procstat_umask, p);
1864 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1865 	    note_procstat_rlimit, p);
1866 	size += register_note(list, NT_PROCSTAT_OSREL,
1867 	    note_procstat_osrel, p);
1868 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1869 	    __elfN(note_procstat_psstrings), p);
1870 	size += register_note(list, NT_PROCSTAT_AUXV,
1871 	    __elfN(note_procstat_auxv), p);
1872 
1873 	*sizep = size;
1874 }
1875 
1876 static void
1877 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1878     size_t notesz)
1879 {
1880 	Elf_Ehdr *ehdr;
1881 	Elf_Phdr *phdr;
1882 	Elf_Shdr *shdr;
1883 	struct phdr_closure phc;
1884 
1885 	ehdr = (Elf_Ehdr *)hdr;
1886 
1887 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1888 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1889 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1890 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1891 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1892 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1893 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1894 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1895 	ehdr->e_ident[EI_ABIVERSION] = 0;
1896 	ehdr->e_ident[EI_PAD] = 0;
1897 	ehdr->e_type = ET_CORE;
1898 	ehdr->e_machine = td->td_proc->p_elf_machine;
1899 	ehdr->e_version = EV_CURRENT;
1900 	ehdr->e_entry = 0;
1901 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1902 	ehdr->e_flags = td->td_proc->p_elf_flags;
1903 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1904 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1905 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1906 	ehdr->e_shstrndx = SHN_UNDEF;
1907 	if (numsegs + 1 < PN_XNUM) {
1908 		ehdr->e_phnum = numsegs + 1;
1909 		ehdr->e_shnum = 0;
1910 	} else {
1911 		ehdr->e_phnum = PN_XNUM;
1912 		ehdr->e_shnum = 1;
1913 
1914 		ehdr->e_shoff = ehdr->e_phoff +
1915 		    (numsegs + 1) * ehdr->e_phentsize;
1916 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1917 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1918 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1919 
1920 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1921 		memset(shdr, 0, sizeof(*shdr));
1922 		/*
1923 		 * A special first section is used to hold large segment and
1924 		 * section counts.  This was proposed by Sun Microsystems in
1925 		 * Solaris and has been adopted by Linux; the standard ELF
1926 		 * tools are already familiar with the technique.
1927 		 *
1928 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1929 		 * (or 12-7 depending on the version of the document) for more
1930 		 * details.
1931 		 */
1932 		shdr->sh_type = SHT_NULL;
1933 		shdr->sh_size = ehdr->e_shnum;
1934 		shdr->sh_link = ehdr->e_shstrndx;
1935 		shdr->sh_info = numsegs + 1;
1936 	}
1937 
1938 	/*
1939 	 * Fill in the program header entries.
1940 	 */
1941 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1942 
1943 	/* The note segement. */
1944 	phdr->p_type = PT_NOTE;
1945 	phdr->p_offset = hdrsize;
1946 	phdr->p_vaddr = 0;
1947 	phdr->p_paddr = 0;
1948 	phdr->p_filesz = notesz;
1949 	phdr->p_memsz = 0;
1950 	phdr->p_flags = PF_R;
1951 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1952 	phdr++;
1953 
1954 	/* All the writable segments from the program. */
1955 	phc.phdr = phdr;
1956 	phc.offset = round_page(hdrsize + notesz);
1957 	each_dumpable_segment(td, cb_put_phdr, &phc);
1958 }
1959 
1960 static size_t
1961 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1962 {
1963 	struct note_info *ninfo;
1964 	size_t size, notesize;
1965 
1966 	size = 0;
1967 	out(arg, NULL, &size);
1968 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1969 	ninfo->type = type;
1970 	ninfo->outfunc = out;
1971 	ninfo->outarg = arg;
1972 	ninfo->outsize = size;
1973 	TAILQ_INSERT_TAIL(list, ninfo, link);
1974 
1975 	if (type == -1)
1976 		return (size);
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 size_t
1987 append_note_data(const void *src, void *dst, size_t len)
1988 {
1989 	size_t padded_len;
1990 
1991 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1992 	if (dst != NULL) {
1993 		bcopy(src, dst, len);
1994 		bzero((char *)dst + len, padded_len - len);
1995 	}
1996 	return (padded_len);
1997 }
1998 
1999 size_t
2000 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2001 {
2002 	Elf_Note *note;
2003 	char *buf;
2004 	size_t notesize;
2005 
2006 	buf = dst;
2007 	if (buf != NULL) {
2008 		note = (Elf_Note *)buf;
2009 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2010 		note->n_descsz = size;
2011 		note->n_type = type;
2012 		buf += sizeof(*note);
2013 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2014 		    sizeof(FREEBSD_ABI_VENDOR));
2015 		append_note_data(src, buf, size);
2016 		if (descp != NULL)
2017 			*descp = buf;
2018 	}
2019 
2020 	notesize = sizeof(Elf_Note) +		/* note header */
2021 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2022 						/* note name */
2023 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2024 
2025 	return (notesize);
2026 }
2027 
2028 static void
2029 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2030 {
2031 	Elf_Note note;
2032 	ssize_t old_len, sect_len;
2033 	size_t new_len, descsz, i;
2034 
2035 	if (ninfo->type == -1) {
2036 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2037 		return;
2038 	}
2039 
2040 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2041 	note.n_descsz = ninfo->outsize;
2042 	note.n_type = ninfo->type;
2043 
2044 	sbuf_bcat(sb, &note, sizeof(note));
2045 	sbuf_start_section(sb, &old_len);
2046 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2047 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2048 	if (note.n_descsz == 0)
2049 		return;
2050 	sbuf_start_section(sb, &old_len);
2051 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2052 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2053 	if (sect_len < 0)
2054 		return;
2055 
2056 	new_len = (size_t)sect_len;
2057 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2058 	if (new_len < descsz) {
2059 		/*
2060 		 * It is expected that individual note emitters will correctly
2061 		 * predict their expected output size and fill up to that size
2062 		 * themselves, padding in a format-specific way if needed.
2063 		 * However, in case they don't, just do it here with zeros.
2064 		 */
2065 		for (i = 0; i < descsz - new_len; i++)
2066 			sbuf_putc(sb, 0);
2067 	} else if (new_len > descsz) {
2068 		/*
2069 		 * We can't always truncate sb -- we may have drained some
2070 		 * of it already.
2071 		 */
2072 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2073 		    "read it (%zu > %zu).  Since it is longer than "
2074 		    "expected, this coredump's notes are corrupt.  THIS "
2075 		    "IS A BUG in the note_procstat routine for type %u.\n",
2076 		    __func__, (unsigned)note.n_type, new_len, descsz,
2077 		    (unsigned)note.n_type));
2078 	}
2079 }
2080 
2081 /*
2082  * Miscellaneous note out functions.
2083  */
2084 
2085 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2086 #include <compat/freebsd32/freebsd32.h>
2087 #include <compat/freebsd32/freebsd32_signal.h>
2088 
2089 typedef struct prstatus32 elf_prstatus_t;
2090 typedef struct prpsinfo32 elf_prpsinfo_t;
2091 typedef struct fpreg32 elf_prfpregset_t;
2092 typedef struct fpreg32 elf_fpregset_t;
2093 typedef struct reg32 elf_gregset_t;
2094 typedef struct thrmisc32 elf_thrmisc_t;
2095 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2096 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2097 typedef uint32_t elf_ps_strings_t;
2098 #else
2099 typedef prstatus_t elf_prstatus_t;
2100 typedef prpsinfo_t elf_prpsinfo_t;
2101 typedef prfpregset_t elf_prfpregset_t;
2102 typedef prfpregset_t elf_fpregset_t;
2103 typedef gregset_t elf_gregset_t;
2104 typedef thrmisc_t elf_thrmisc_t;
2105 #define ELF_KERN_PROC_MASK	0
2106 typedef struct kinfo_proc elf_kinfo_proc_t;
2107 typedef vm_offset_t elf_ps_strings_t;
2108 #endif
2109 
2110 static void
2111 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2112 {
2113 	struct sbuf sbarg;
2114 	size_t len;
2115 	char *cp, *end;
2116 	struct proc *p;
2117 	elf_prpsinfo_t *psinfo;
2118 	int error;
2119 
2120 	p = (struct proc *)arg;
2121 	if (sb != NULL) {
2122 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2123 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2124 		psinfo->pr_version = PRPSINFO_VERSION;
2125 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2126 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2127 		PROC_LOCK(p);
2128 		if (p->p_args != NULL) {
2129 			len = sizeof(psinfo->pr_psargs) - 1;
2130 			if (len > p->p_args->ar_length)
2131 				len = p->p_args->ar_length;
2132 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2133 			PROC_UNLOCK(p);
2134 			error = 0;
2135 		} else {
2136 			_PHOLD(p);
2137 			PROC_UNLOCK(p);
2138 			sbuf_new(&sbarg, psinfo->pr_psargs,
2139 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2140 			error = proc_getargv(curthread, p, &sbarg);
2141 			PRELE(p);
2142 			if (sbuf_finish(&sbarg) == 0)
2143 				len = sbuf_len(&sbarg) - 1;
2144 			else
2145 				len = sizeof(psinfo->pr_psargs) - 1;
2146 			sbuf_delete(&sbarg);
2147 		}
2148 		if (error || len == 0)
2149 			strlcpy(psinfo->pr_psargs, p->p_comm,
2150 			    sizeof(psinfo->pr_psargs));
2151 		else {
2152 			KASSERT(len < sizeof(psinfo->pr_psargs),
2153 			    ("len is too long: %zu vs %zu", len,
2154 			    sizeof(psinfo->pr_psargs)));
2155 			cp = psinfo->pr_psargs;
2156 			end = cp + len - 1;
2157 			for (;;) {
2158 				cp = memchr(cp, '\0', end - cp);
2159 				if (cp == NULL)
2160 					break;
2161 				*cp = ' ';
2162 			}
2163 		}
2164 		psinfo->pr_pid = p->p_pid;
2165 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2166 		free(psinfo, M_TEMP);
2167 	}
2168 	*sizep = sizeof(*psinfo);
2169 }
2170 
2171 static void
2172 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2173 {
2174 	struct thread *td;
2175 	elf_prstatus_t *status;
2176 
2177 	td = (struct thread *)arg;
2178 	if (sb != NULL) {
2179 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2180 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2181 		status->pr_version = PRSTATUS_VERSION;
2182 		status->pr_statussz = sizeof(elf_prstatus_t);
2183 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2184 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2185 		status->pr_osreldate = osreldate;
2186 		status->pr_cursig = td->td_proc->p_sig;
2187 		status->pr_pid = td->td_tid;
2188 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2189 		fill_regs32(td, &status->pr_reg);
2190 #else
2191 		fill_regs(td, &status->pr_reg);
2192 #endif
2193 		sbuf_bcat(sb, status, sizeof(*status));
2194 		free(status, M_TEMP);
2195 	}
2196 	*sizep = sizeof(*status);
2197 }
2198 
2199 static void
2200 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2201 {
2202 	struct thread *td;
2203 	elf_prfpregset_t *fpregset;
2204 
2205 	td = (struct thread *)arg;
2206 	if (sb != NULL) {
2207 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2208 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2209 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2210 		fill_fpregs32(td, fpregset);
2211 #else
2212 		fill_fpregs(td, fpregset);
2213 #endif
2214 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2215 		free(fpregset, M_TEMP);
2216 	}
2217 	*sizep = sizeof(*fpregset);
2218 }
2219 
2220 static void
2221 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2222 {
2223 	struct thread *td;
2224 	elf_thrmisc_t thrmisc;
2225 
2226 	td = (struct thread *)arg;
2227 	if (sb != NULL) {
2228 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2229 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2230 		strcpy(thrmisc.pr_tname, td->td_name);
2231 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2232 	}
2233 	*sizep = sizeof(thrmisc);
2234 }
2235 
2236 static void
2237 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2238 {
2239 	struct thread *td;
2240 	size_t size;
2241 	int structsize;
2242 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2243 	struct ptrace_lwpinfo32 pl;
2244 #else
2245 	struct ptrace_lwpinfo pl;
2246 #endif
2247 
2248 	td = (struct thread *)arg;
2249 	size = sizeof(structsize) + sizeof(pl);
2250 	if (sb != NULL) {
2251 		KASSERT(*sizep == size, ("invalid size"));
2252 		structsize = sizeof(pl);
2253 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2254 		bzero(&pl, sizeof(pl));
2255 		pl.pl_lwpid = td->td_tid;
2256 		pl.pl_event = PL_EVENT_NONE;
2257 		pl.pl_sigmask = td->td_sigmask;
2258 		pl.pl_siglist = td->td_siglist;
2259 		if (td->td_si.si_signo != 0) {
2260 			pl.pl_event = PL_EVENT_SIGNAL;
2261 			pl.pl_flags |= PL_FLAG_SI;
2262 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2263 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2264 #else
2265 			pl.pl_siginfo = td->td_si;
2266 #endif
2267 		}
2268 		strcpy(pl.pl_tdname, td->td_name);
2269 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2270 		sbuf_bcat(sb, &pl, sizeof(pl));
2271 	}
2272 	*sizep = size;
2273 }
2274 
2275 /*
2276  * Allow for MD specific notes, as well as any MD
2277  * specific preparations for writing MI notes.
2278  */
2279 static void
2280 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2281 {
2282 	struct thread *td;
2283 	void *buf;
2284 	size_t size;
2285 
2286 	td = (struct thread *)arg;
2287 	size = *sizep;
2288 	if (size != 0 && sb != NULL)
2289 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2290 	else
2291 		buf = NULL;
2292 	size = 0;
2293 	__elfN(dump_thread)(td, buf, &size);
2294 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2295 	if (size != 0 && sb != NULL)
2296 		sbuf_bcat(sb, buf, size);
2297 	free(buf, M_TEMP);
2298 	*sizep = size;
2299 }
2300 
2301 #ifdef KINFO_PROC_SIZE
2302 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2303 #endif
2304 
2305 static void
2306 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2307 {
2308 	struct proc *p;
2309 	size_t size;
2310 	int structsize;
2311 
2312 	p = (struct proc *)arg;
2313 	size = sizeof(structsize) + p->p_numthreads *
2314 	    sizeof(elf_kinfo_proc_t);
2315 
2316 	if (sb != NULL) {
2317 		KASSERT(*sizep == size, ("invalid size"));
2318 		structsize = sizeof(elf_kinfo_proc_t);
2319 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2320 		PROC_LOCK(p);
2321 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2322 	}
2323 	*sizep = size;
2324 }
2325 
2326 #ifdef KINFO_FILE_SIZE
2327 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2328 #endif
2329 
2330 static void
2331 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2332 {
2333 	struct proc *p;
2334 	size_t size, sect_sz, i;
2335 	ssize_t start_len, sect_len;
2336 	int structsize, filedesc_flags;
2337 
2338 	if (coredump_pack_fileinfo)
2339 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2340 	else
2341 		filedesc_flags = 0;
2342 
2343 	p = (struct proc *)arg;
2344 	structsize = sizeof(struct kinfo_file);
2345 	if (sb == NULL) {
2346 		size = 0;
2347 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2348 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2349 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2350 		PROC_LOCK(p);
2351 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2352 		sbuf_finish(sb);
2353 		sbuf_delete(sb);
2354 		*sizep = size;
2355 	} else {
2356 		sbuf_start_section(sb, &start_len);
2357 
2358 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2359 		PROC_LOCK(p);
2360 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2361 		    filedesc_flags);
2362 
2363 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2364 		if (sect_len < 0)
2365 			return;
2366 		sect_sz = sect_len;
2367 
2368 		KASSERT(sect_sz <= *sizep,
2369 		    ("kern_proc_filedesc_out did not respect maxlen; "
2370 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2371 		     sect_sz - sizeof(structsize)));
2372 
2373 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2374 			sbuf_putc(sb, 0);
2375 	}
2376 }
2377 
2378 #ifdef KINFO_VMENTRY_SIZE
2379 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2380 #endif
2381 
2382 static void
2383 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2384 {
2385 	struct proc *p;
2386 	size_t size;
2387 	int structsize, vmmap_flags;
2388 
2389 	if (coredump_pack_vmmapinfo)
2390 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2391 	else
2392 		vmmap_flags = 0;
2393 
2394 	p = (struct proc *)arg;
2395 	structsize = sizeof(struct kinfo_vmentry);
2396 	if (sb == NULL) {
2397 		size = 0;
2398 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2399 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2400 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2401 		PROC_LOCK(p);
2402 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2403 		sbuf_finish(sb);
2404 		sbuf_delete(sb);
2405 		*sizep = size;
2406 	} else {
2407 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2408 		PROC_LOCK(p);
2409 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2410 		    vmmap_flags);
2411 	}
2412 }
2413 
2414 static void
2415 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2416 {
2417 	struct proc *p;
2418 	size_t size;
2419 	int structsize;
2420 
2421 	p = (struct proc *)arg;
2422 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2423 	if (sb != NULL) {
2424 		KASSERT(*sizep == size, ("invalid size"));
2425 		structsize = sizeof(gid_t);
2426 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2427 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2428 		    sizeof(gid_t));
2429 	}
2430 	*sizep = size;
2431 }
2432 
2433 static void
2434 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2435 {
2436 	struct proc *p;
2437 	size_t size;
2438 	int structsize;
2439 
2440 	p = (struct proc *)arg;
2441 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2442 	if (sb != NULL) {
2443 		KASSERT(*sizep == size, ("invalid size"));
2444 		structsize = sizeof(p->p_fd->fd_cmask);
2445 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2446 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2447 	}
2448 	*sizep = size;
2449 }
2450 
2451 static void
2452 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2453 {
2454 	struct proc *p;
2455 	struct rlimit rlim[RLIM_NLIMITS];
2456 	size_t size;
2457 	int structsize, i;
2458 
2459 	p = (struct proc *)arg;
2460 	size = sizeof(structsize) + sizeof(rlim);
2461 	if (sb != NULL) {
2462 		KASSERT(*sizep == size, ("invalid size"));
2463 		structsize = sizeof(rlim);
2464 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2465 		PROC_LOCK(p);
2466 		for (i = 0; i < RLIM_NLIMITS; i++)
2467 			lim_rlimit_proc(p, i, &rlim[i]);
2468 		PROC_UNLOCK(p);
2469 		sbuf_bcat(sb, rlim, sizeof(rlim));
2470 	}
2471 	*sizep = size;
2472 }
2473 
2474 static void
2475 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2476 {
2477 	struct proc *p;
2478 	size_t size;
2479 	int structsize;
2480 
2481 	p = (struct proc *)arg;
2482 	size = sizeof(structsize) + sizeof(p->p_osrel);
2483 	if (sb != NULL) {
2484 		KASSERT(*sizep == size, ("invalid size"));
2485 		structsize = sizeof(p->p_osrel);
2486 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2487 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2488 	}
2489 	*sizep = size;
2490 }
2491 
2492 static void
2493 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2494 {
2495 	struct proc *p;
2496 	elf_ps_strings_t ps_strings;
2497 	size_t size;
2498 	int structsize;
2499 
2500 	p = (struct proc *)arg;
2501 	size = sizeof(structsize) + sizeof(ps_strings);
2502 	if (sb != NULL) {
2503 		KASSERT(*sizep == size, ("invalid size"));
2504 		structsize = sizeof(ps_strings);
2505 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2506 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2507 #else
2508 		ps_strings = p->p_sysent->sv_psstrings;
2509 #endif
2510 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2511 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2512 	}
2513 	*sizep = size;
2514 }
2515 
2516 static void
2517 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2518 {
2519 	struct proc *p;
2520 	size_t size;
2521 	int structsize;
2522 
2523 	p = (struct proc *)arg;
2524 	if (sb == NULL) {
2525 		size = 0;
2526 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2527 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2528 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2529 		PHOLD(p);
2530 		proc_getauxv(curthread, p, sb);
2531 		PRELE(p);
2532 		sbuf_finish(sb);
2533 		sbuf_delete(sb);
2534 		*sizep = size;
2535 	} else {
2536 		structsize = sizeof(Elf_Auxinfo);
2537 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2538 		PHOLD(p);
2539 		proc_getauxv(curthread, p, sb);
2540 		PRELE(p);
2541 	}
2542 }
2543 
2544 static boolean_t
2545 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2546     const char *note_vendor, const Elf_Phdr *pnote,
2547     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2548 {
2549 	const Elf_Note *note, *note0, *note_end;
2550 	const char *note_name;
2551 	char *buf;
2552 	int i, error;
2553 	boolean_t res;
2554 
2555 	/* We need some limit, might as well use PAGE_SIZE. */
2556 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2557 		return (FALSE);
2558 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2559 	if (pnote->p_offset > PAGE_SIZE ||
2560 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2561 		VOP_UNLOCK(imgp->vp, 0);
2562 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2563 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2564 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2565 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2566 		    curthread->td_ucred, NOCRED, NULL, curthread);
2567 		if (error != 0) {
2568 			uprintf("i/o error PT_NOTE\n");
2569 			goto retf;
2570 		}
2571 		note = note0 = (const Elf_Note *)buf;
2572 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2573 	} else {
2574 		note = note0 = (const Elf_Note *)(imgp->image_header +
2575 		    pnote->p_offset);
2576 		note_end = (const Elf_Note *)(imgp->image_header +
2577 		    pnote->p_offset + pnote->p_filesz);
2578 		buf = NULL;
2579 	}
2580 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2581 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2582 		    (const char *)note < sizeof(Elf_Note)) {
2583 			goto retf;
2584 		}
2585 		if (note->n_namesz != checknote->n_namesz ||
2586 		    note->n_descsz != checknote->n_descsz ||
2587 		    note->n_type != checknote->n_type)
2588 			goto nextnote;
2589 		note_name = (const char *)(note + 1);
2590 		if (note_name + checknote->n_namesz >=
2591 		    (const char *)note_end || strncmp(note_vendor,
2592 		    note_name, checknote->n_namesz) != 0)
2593 			goto nextnote;
2594 
2595 		if (cb(note, cb_arg, &res))
2596 			goto ret;
2597 nextnote:
2598 		note = (const Elf_Note *)((const char *)(note + 1) +
2599 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2600 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2601 	}
2602 retf:
2603 	res = FALSE;
2604 ret:
2605 	free(buf, M_TEMP);
2606 	return (res);
2607 }
2608 
2609 struct brandnote_cb_arg {
2610 	Elf_Brandnote *brandnote;
2611 	int32_t *osrel;
2612 };
2613 
2614 static boolean_t
2615 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2616 {
2617 	struct brandnote_cb_arg *arg;
2618 
2619 	arg = arg0;
2620 
2621 	/*
2622 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2623 	 * necessary.
2624 	 */
2625 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2626 	    arg->brandnote->trans_osrel != NULL ?
2627 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2628 
2629 	return (TRUE);
2630 }
2631 
2632 static Elf_Note fctl_note = {
2633 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2634 	.n_descsz = sizeof(uint32_t),
2635 	.n_type = NT_FREEBSD_FEATURE_CTL,
2636 };
2637 
2638 struct fctl_cb_arg {
2639 	uint32_t *fctl0;
2640 };
2641 
2642 static boolean_t
2643 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2644 {
2645 	struct fctl_cb_arg *arg;
2646 	const Elf32_Word *desc;
2647 	uintptr_t p;
2648 
2649 	arg = arg0;
2650 	p = (uintptr_t)(note + 1);
2651 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2652 	desc = (const Elf32_Word *)p;
2653 	*arg->fctl0 = desc[0];
2654 	return (TRUE);
2655 }
2656 
2657 /*
2658  * Try to find the appropriate ABI-note section for checknote, fetch
2659  * the osreldate and feature control flags for binary from the ELF
2660  * OSABI-note.  Only the first page of the image is searched, the same
2661  * as for headers.
2662  */
2663 static boolean_t
2664 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2665     int32_t *osrel, uint32_t *fctl0)
2666 {
2667 	const Elf_Phdr *phdr;
2668 	const Elf_Ehdr *hdr;
2669 	struct brandnote_cb_arg b_arg;
2670 	struct fctl_cb_arg f_arg;
2671 	int i, j;
2672 
2673 	hdr = (const Elf_Ehdr *)imgp->image_header;
2674 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2675 	b_arg.brandnote = brandnote;
2676 	b_arg.osrel = osrel;
2677 	f_arg.fctl0 = fctl0;
2678 
2679 	for (i = 0; i < hdr->e_phnum; i++) {
2680 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2681 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2682 		    &b_arg)) {
2683 			for (j = 0; j < hdr->e_phnum; j++) {
2684 				if (phdr[j].p_type == PT_NOTE &&
2685 				    __elfN(parse_notes)(imgp, &fctl_note,
2686 				    FREEBSD_ABI_VENDOR, &phdr[j],
2687 				    note_fctl_cb, &f_arg))
2688 					break;
2689 			}
2690 			return (TRUE);
2691 		}
2692 	}
2693 	return (FALSE);
2694 
2695 }
2696 
2697 /*
2698  * Tell kern_execve.c about it, with a little help from the linker.
2699  */
2700 static struct execsw __elfN(execsw) = {
2701 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2702 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2703 };
2704 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2705 
2706 static vm_prot_t
2707 __elfN(trans_prot)(Elf_Word flags)
2708 {
2709 	vm_prot_t prot;
2710 
2711 	prot = 0;
2712 	if (flags & PF_X)
2713 		prot |= VM_PROT_EXECUTE;
2714 	if (flags & PF_W)
2715 		prot |= VM_PROT_WRITE;
2716 	if (flags & PF_R)
2717 		prot |= VM_PROT_READ;
2718 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2719 	if (i386_read_exec && (flags & PF_R))
2720 		prot |= VM_PROT_EXECUTE;
2721 #endif
2722 	return (prot);
2723 }
2724 
2725 static Elf_Word
2726 __elfN(untrans_prot)(vm_prot_t prot)
2727 {
2728 	Elf_Word flags;
2729 
2730 	flags = 0;
2731 	if (prot & VM_PROT_EXECUTE)
2732 		flags |= PF_X;
2733 	if (prot & VM_PROT_READ)
2734 		flags |= PF_R;
2735 	if (prot & VM_PROT_WRITE)
2736 		flags |= PF_W;
2737 	return (flags);
2738 }
2739