xref: /freebsd/sys/kern/imgact_elf.c (revision 36d7818975359fd2aacb19e4f9442a841dc954bb)
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 		    (object != NULL ? MAP_VN_EXEC : 0));
531 		if (rv != KERN_SUCCESS) {
532 			locked = VOP_ISLOCKED(imgp->vp);
533 			VOP_UNLOCK(imgp->vp, 0);
534 			vm_object_deallocate(object);
535 			vn_lock(imgp->vp, locked | LK_RETRY);
536 			return (rv);
537 		} else if (object != NULL) {
538 			MPASS(imgp->vp->v_object == object);
539 			VOP_SET_TEXT_CHECKED(imgp->vp);
540 		}
541 	}
542 	return (KERN_SUCCESS);
543 }
544 
545 static int
546 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
547     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
548 {
549 	struct sf_buf *sf;
550 	size_t map_len;
551 	vm_map_t map;
552 	vm_object_t object;
553 	vm_offset_t map_addr;
554 	int error, rv, cow;
555 	size_t copy_len;
556 	vm_ooffset_t file_addr;
557 
558 	/*
559 	 * It's necessary to fail if the filsz + offset taken from the
560 	 * header is greater than the actual file pager object's size.
561 	 * If we were to allow this, then the vm_map_find() below would
562 	 * walk right off the end of the file object and into the ether.
563 	 *
564 	 * While I'm here, might as well check for something else that
565 	 * is invalid: filsz cannot be greater than memsz.
566 	 */
567 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
568 	    filsz > memsz) {
569 		uprintf("elf_load_section: truncated ELF file\n");
570 		return (ENOEXEC);
571 	}
572 
573 	object = imgp->object;
574 	map = &imgp->proc->p_vmspace->vm_map;
575 	map_addr = trunc_page((vm_offset_t)vmaddr);
576 	file_addr = trunc_page(offset);
577 
578 	/*
579 	 * We have two choices.  We can either clear the data in the last page
580 	 * of an oversized mapping, or we can start the anon mapping a page
581 	 * early and copy the initialized data into that first page.  We
582 	 * choose the second.
583 	 */
584 	if (filsz == 0)
585 		map_len = 0;
586 	else if (memsz > filsz)
587 		map_len = trunc_page(offset + filsz) - file_addr;
588 	else
589 		map_len = round_page(offset + filsz) - file_addr;
590 
591 	if (map_len != 0) {
592 		/* cow flags: don't dump readonly sections in core */
593 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
594 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
595 
596 		rv = __elfN(map_insert)(imgp, map, object, file_addr,
597 		    map_addr, map_addr + map_len, prot, cow);
598 		if (rv != KERN_SUCCESS)
599 			return (EINVAL);
600 
601 		/* we can stop now if we've covered it all */
602 		if (memsz == filsz)
603 			return (0);
604 	}
605 
606 
607 	/*
608 	 * We have to get the remaining bit of the file into the first part
609 	 * of the oversized map segment.  This is normally because the .data
610 	 * segment in the file is extended to provide bss.  It's a neat idea
611 	 * to try and save a page, but it's a pain in the behind to implement.
612 	 */
613 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
614 	    filsz);
615 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
616 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
617 
618 	/* This had damn well better be true! */
619 	if (map_len != 0) {
620 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
621 		    map_addr + map_len, prot, 0);
622 		if (rv != KERN_SUCCESS)
623 			return (EINVAL);
624 	}
625 
626 	if (copy_len != 0) {
627 		sf = vm_imgact_map_page(object, offset + filsz);
628 		if (sf == NULL)
629 			return (EIO);
630 
631 		/* send the page fragment to user space */
632 		error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
633 		    copy_len);
634 		vm_imgact_unmap_page(sf);
635 		if (error != 0)
636 			return (error);
637 	}
638 
639 	/*
640 	 * Remove write access to the page if it was only granted by map_insert
641 	 * to allow copyout.
642 	 */
643 	if ((prot & VM_PROT_WRITE) == 0)
644 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
645 		    map_len), prot, FALSE);
646 
647 	return (0);
648 }
649 
650 static int
651 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
652     const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
653 {
654 	vm_prot_t prot;
655 	u_long base_addr;
656 	bool first;
657 	int error, i;
658 
659 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
660 
661 	base_addr = 0;
662 	first = true;
663 
664 	for (i = 0; i < hdr->e_phnum; i++) {
665 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
666 			continue;
667 
668 		/* Loadable segment */
669 		prot = __elfN(trans_prot)(phdr[i].p_flags);
670 		error = __elfN(load_section)(imgp, phdr[i].p_offset,
671 		    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
672 		    phdr[i].p_memsz, phdr[i].p_filesz, prot);
673 		if (error != 0)
674 			return (error);
675 
676 		/*
677 		 * Establish the base address if this is the first segment.
678 		 */
679 		if (first) {
680   			base_addr = trunc_page(phdr[i].p_vaddr + rbase);
681 			first = false;
682 		}
683 	}
684 
685 	if (base_addrp != NULL)
686 		*base_addrp = base_addr;
687 
688 	return (0);
689 }
690 
691 /*
692  * Load the file "file" into memory.  It may be either a shared object
693  * or an executable.
694  *
695  * The "addr" reference parameter is in/out.  On entry, it specifies
696  * the address where a shared object should be loaded.  If the file is
697  * an executable, this value is ignored.  On exit, "addr" specifies
698  * where the file was actually loaded.
699  *
700  * The "entry" reference parameter is out only.  On exit, it specifies
701  * the entry point for the loaded file.
702  */
703 static int
704 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
705 	u_long *entry)
706 {
707 	struct {
708 		struct nameidata nd;
709 		struct vattr attr;
710 		struct image_params image_params;
711 	} *tempdata;
712 	const Elf_Ehdr *hdr = NULL;
713 	const Elf_Phdr *phdr = NULL;
714 	struct nameidata *nd;
715 	struct vattr *attr;
716 	struct image_params *imgp;
717 	u_long rbase;
718 	u_long base_addr = 0;
719 	int error;
720 
721 #ifdef CAPABILITY_MODE
722 	/*
723 	 * XXXJA: This check can go away once we are sufficiently confident
724 	 * that the checks in namei() are correct.
725 	 */
726 	if (IN_CAPABILITY_MODE(curthread))
727 		return (ECAPMODE);
728 #endif
729 
730 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
731 	nd = &tempdata->nd;
732 	attr = &tempdata->attr;
733 	imgp = &tempdata->image_params;
734 
735 	/*
736 	 * Initialize part of the common data
737 	 */
738 	imgp->proc = p;
739 	imgp->attr = attr;
740 	imgp->firstpage = NULL;
741 	imgp->image_header = NULL;
742 	imgp->object = NULL;
743 	imgp->execlabel = NULL;
744 
745 	NDINIT(nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, file,
746 	    curthread);
747 	if ((error = namei(nd)) != 0) {
748 		nd->ni_vp = NULL;
749 		goto fail;
750 	}
751 	NDFREE(nd, NDF_ONLY_PNBUF);
752 	imgp->vp = nd->ni_vp;
753 
754 	/*
755 	 * Check permissions, modes, uid, etc on the file, and "open" it.
756 	 */
757 	error = exec_check_permissions(imgp);
758 	if (error)
759 		goto fail;
760 
761 	error = exec_map_first_page(imgp);
762 	if (error)
763 		goto fail;
764 
765 	imgp->object = nd->ni_vp->v_object;
766 
767 	hdr = (const Elf_Ehdr *)imgp->image_header;
768 	if ((error = __elfN(check_header)(hdr)) != 0)
769 		goto fail;
770 	if (hdr->e_type == ET_DYN)
771 		rbase = *addr;
772 	else if (hdr->e_type == ET_EXEC)
773 		rbase = 0;
774 	else {
775 		error = ENOEXEC;
776 		goto fail;
777 	}
778 
779 	/* Only support headers that fit within first page for now      */
780 	if ((hdr->e_phoff > PAGE_SIZE) ||
781 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
782 		error = ENOEXEC;
783 		goto fail;
784 	}
785 
786 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
787 	if (!aligned(phdr, Elf_Addr)) {
788 		error = ENOEXEC;
789 		goto fail;
790 	}
791 
792 	error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
793 	if (error != 0)
794 		goto fail;
795 
796 	*addr = base_addr;
797 	*entry = (unsigned long)hdr->e_entry + rbase;
798 
799 fail:
800 	if (imgp->firstpage)
801 		exec_unmap_first_page(imgp);
802 
803 	if (nd->ni_vp) {
804 		if (imgp->textset)
805 			VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
806 		vput(nd->ni_vp);
807 	}
808 	free(tempdata, M_TEMP);
809 
810 	return (error);
811 }
812 
813 static u_long
814 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
815     u_int align)
816 {
817 	u_long rbase, res;
818 
819 	MPASS(vm_map_min(map) <= minv);
820 	MPASS(maxv <= vm_map_max(map));
821 	MPASS(minv < maxv);
822 	MPASS(minv + align < maxv);
823 	arc4rand(&rbase, sizeof(rbase), 0);
824 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
825 	res &= ~((u_long)align - 1);
826 	if (res >= maxv)
827 		res -= align;
828 	KASSERT(res >= minv,
829 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
830 	    res, minv, maxv, rbase));
831 	KASSERT(res < maxv,
832 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
833 	    res, maxv, minv, rbase));
834 	return (res);
835 }
836 
837 static int
838 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
839     const Elf_Phdr *phdr, u_long et_dyn_addr)
840 {
841 	struct vmspace *vmspace;
842 	const char *err_str;
843 	u_long text_size, data_size, total_size, text_addr, data_addr;
844 	u_long seg_size, seg_addr;
845 	int i;
846 
847 	err_str = NULL;
848 	text_size = data_size = total_size = text_addr = data_addr = 0;
849 
850 	for (i = 0; i < hdr->e_phnum; i++) {
851 		if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
852 			continue;
853 
854 		seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
855 		seg_size = round_page(phdr[i].p_memsz +
856 		    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
857 
858 		/*
859 		 * Make the largest executable segment the official
860 		 * text segment and all others data.
861 		 *
862 		 * Note that obreak() assumes that data_addr + data_size == end
863 		 * of data load area, and the ELF file format expects segments
864 		 * to be sorted by address.  If multiple data segments exist,
865 		 * the last one will be used.
866 		 */
867 
868 		if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
869 			text_size = seg_size;
870 			text_addr = seg_addr;
871 		} else {
872 			data_size = seg_size;
873 			data_addr = seg_addr;
874 		}
875 		total_size += seg_size;
876 	}
877 
878 	if (data_addr == 0 && data_size == 0) {
879 		data_addr = text_addr;
880 		data_size = text_size;
881 	}
882 
883 	/*
884 	 * Check limits.  It should be safe to check the
885 	 * limits after loading the segments since we do
886 	 * not actually fault in all the segments pages.
887 	 */
888 	PROC_LOCK(imgp->proc);
889 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
890 		err_str = "Data segment size exceeds process limit";
891 	else if (text_size > maxtsiz)
892 		err_str = "Text segment size exceeds system limit";
893 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
894 		err_str = "Total segment size exceeds process limit";
895 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
896 		err_str = "Data segment size exceeds resource limit";
897 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
898 		err_str = "Total segment size exceeds resource limit";
899 	PROC_UNLOCK(imgp->proc);
900 	if (err_str != NULL) {
901 		uprintf("%s\n", err_str);
902 		return (ENOMEM);
903 	}
904 
905 	vmspace = imgp->proc->p_vmspace;
906 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
907 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
908 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
909 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
910 
911 	return (0);
912 }
913 
914 static int
915 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
916     char **interpp, bool *free_interpp)
917 {
918 	struct thread *td;
919 	char *interp;
920 	int error, interp_name_len;
921 
922 	KASSERT(phdr->p_type == PT_INTERP,
923 	    ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
924 	ASSERT_VOP_LOCKED(imgp->vp, __func__);
925 
926 	td = curthread;
927 
928 	/* Path to interpreter */
929 	if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
930 		uprintf("Invalid PT_INTERP\n");
931 		return (ENOEXEC);
932 	}
933 
934 	interp_name_len = phdr->p_filesz;
935 	if (phdr->p_offset > PAGE_SIZE ||
936 	    interp_name_len > PAGE_SIZE - phdr->p_offset) {
937 		/*
938 		 * The vnode lock might be needed by the pagedaemon to
939 		 * clean pages owned by the vnode.  Do not allow sleep
940 		 * waiting for memory with the vnode locked, instead
941 		 * try non-sleepable allocation first, and if it
942 		 * fails, go to the slow path were we drop the lock
943 		 * and do M_WAITOK.  A text reference prevents
944 		 * modifications to the vnode content.
945 		 */
946 		interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
947 		if (interp == NULL) {
948 			VOP_UNLOCK(imgp->vp, 0);
949 			interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
950 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
951 		}
952 
953 		error = vn_rdwr(UIO_READ, imgp->vp, interp,
954 		    interp_name_len, phdr->p_offset,
955 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
956 		    NOCRED, NULL, td);
957 		if (error != 0) {
958 			free(interp, M_TEMP);
959 			uprintf("i/o error PT_INTERP %d\n", error);
960 			return (error);
961 		}
962 		interp[interp_name_len] = '\0';
963 
964 		*interpp = interp;
965 		*free_interpp = true;
966 		return (0);
967 	}
968 
969 	interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
970 	if (interp[interp_name_len - 1] != '\0') {
971 		uprintf("Invalid PT_INTERP\n");
972 		return (ENOEXEC);
973 	}
974 
975 	*interpp = interp;
976 	*free_interpp = false;
977 	return (0);
978 }
979 
980 static int
981 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
982     const char *interp, u_long *addr, u_long *entry)
983 {
984 	char *path;
985 	int error;
986 
987 	if (brand_info->emul_path != NULL &&
988 	    brand_info->emul_path[0] != '\0') {
989 		path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
990 		snprintf(path, MAXPATHLEN, "%s%s",
991 		    brand_info->emul_path, interp);
992 		error = __elfN(load_file)(imgp->proc, path, addr, entry);
993 		free(path, M_TEMP);
994 		if (error == 0)
995 			return (0);
996 	}
997 
998 	if (brand_info->interp_newpath != NULL &&
999 	    (brand_info->interp_path == NULL ||
1000 	    strcmp(interp, brand_info->interp_path) == 0)) {
1001 		error = __elfN(load_file)(imgp->proc,
1002 		    brand_info->interp_newpath, addr, entry);
1003 		if (error == 0)
1004 			return (0);
1005 	}
1006 
1007 	error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1008 	if (error == 0)
1009 		return (0);
1010 
1011 	uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1012 	return (error);
1013 }
1014 
1015 /*
1016  * Impossible et_dyn_addr initial value indicating that the real base
1017  * must be calculated later with some randomization applied.
1018  */
1019 #define	ET_DYN_ADDR_RAND	1
1020 
1021 static int
1022 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1023 {
1024 	struct thread *td;
1025 	const Elf_Ehdr *hdr;
1026 	const Elf_Phdr *phdr;
1027 	Elf_Auxargs *elf_auxargs;
1028 	struct vmspace *vmspace;
1029 	vm_map_t map;
1030 	char *interp;
1031 	Elf_Brandinfo *brand_info;
1032 	struct sysentvec *sv;
1033 	u_long addr, baddr, et_dyn_addr, entry, proghdr;
1034 	u_long maxalign, mapsz, maxv, maxv1;
1035 	uint32_t fctl0;
1036 	int32_t osrel;
1037 	bool free_interp;
1038 	int error, i, n;
1039 
1040 	hdr = (const Elf_Ehdr *)imgp->image_header;
1041 
1042 	/*
1043 	 * Do we have a valid ELF header ?
1044 	 *
1045 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1046 	 * if particular brand doesn't support it.
1047 	 */
1048 	if (__elfN(check_header)(hdr) != 0 ||
1049 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1050 		return (-1);
1051 
1052 	/*
1053 	 * From here on down, we return an errno, not -1, as we've
1054 	 * detected an ELF file.
1055 	 */
1056 
1057 	if ((hdr->e_phoff > PAGE_SIZE) ||
1058 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1059 		/* Only support headers in first page for now */
1060 		uprintf("Program headers not in the first page\n");
1061 		return (ENOEXEC);
1062 	}
1063 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1064 	if (!aligned(phdr, Elf_Addr)) {
1065 		uprintf("Unaligned program headers\n");
1066 		return (ENOEXEC);
1067 	}
1068 
1069 	n = error = 0;
1070 	baddr = 0;
1071 	osrel = 0;
1072 	fctl0 = 0;
1073 	entry = proghdr = 0;
1074 	interp = NULL;
1075 	free_interp = false;
1076 	td = curthread;
1077 	maxalign = PAGE_SIZE;
1078 	mapsz = 0;
1079 
1080 	for (i = 0; i < hdr->e_phnum; i++) {
1081 		switch (phdr[i].p_type) {
1082 		case PT_LOAD:
1083 			if (n == 0)
1084 				baddr = phdr[i].p_vaddr;
1085 			if (phdr[i].p_align > maxalign)
1086 				maxalign = phdr[i].p_align;
1087 			mapsz += phdr[i].p_memsz;
1088 			n++;
1089 
1090 			/*
1091 			 * If this segment contains the program headers,
1092 			 * remember their virtual address for the AT_PHDR
1093 			 * aux entry. Static binaries don't usually include
1094 			 * a PT_PHDR entry.
1095 			 */
1096 			if (phdr[i].p_offset == 0 &&
1097 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1098 				<= phdr[i].p_filesz)
1099 				proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1100 			break;
1101 		case PT_INTERP:
1102 			/* Path to interpreter */
1103 			if (interp != NULL) {
1104 				uprintf("Multiple PT_INTERP headers\n");
1105 				error = ENOEXEC;
1106 				goto ret;
1107 			}
1108 			error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1109 			    &free_interp);
1110 			if (error != 0)
1111 				goto ret;
1112 			break;
1113 		case PT_GNU_STACK:
1114 			if (__elfN(nxstack))
1115 				imgp->stack_prot =
1116 				    __elfN(trans_prot)(phdr[i].p_flags);
1117 			imgp->stack_sz = phdr[i].p_memsz;
1118 			break;
1119 		case PT_PHDR: 	/* Program header table info */
1120 			proghdr = phdr[i].p_vaddr;
1121 			break;
1122 		}
1123 	}
1124 
1125 	brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1126 	if (brand_info == NULL) {
1127 		uprintf("ELF binary type \"%u\" not known.\n",
1128 		    hdr->e_ident[EI_OSABI]);
1129 		error = ENOEXEC;
1130 		goto ret;
1131 	}
1132 	sv = brand_info->sysvec;
1133 	et_dyn_addr = 0;
1134 	if (hdr->e_type == ET_DYN) {
1135 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1136 			uprintf("Cannot execute shared object\n");
1137 			error = ENOEXEC;
1138 			goto ret;
1139 		}
1140 		/*
1141 		 * Honour the base load address from the dso if it is
1142 		 * non-zero for some reason.
1143 		 */
1144 		if (baddr == 0) {
1145 			if ((sv->sv_flags & SV_ASLR) == 0 ||
1146 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1147 				et_dyn_addr = ET_DYN_LOAD_ADDR;
1148 			else if ((__elfN(pie_aslr_enabled) &&
1149 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1150 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1151 				et_dyn_addr = ET_DYN_ADDR_RAND;
1152 			else
1153 				et_dyn_addr = ET_DYN_LOAD_ADDR;
1154 		}
1155 	}
1156 
1157 	/*
1158 	 * Avoid a possible deadlock if the current address space is destroyed
1159 	 * and that address space maps the locked vnode.  In the common case,
1160 	 * the locked vnode's v_usecount is decremented but remains greater
1161 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
1162 	 * However, in cases where the vnode lock is external, such as nullfs,
1163 	 * v_usecount may become zero.
1164 	 *
1165 	 * The VV_TEXT flag prevents modifications to the executable while
1166 	 * the vnode is unlocked.
1167 	 */
1168 	VOP_UNLOCK(imgp->vp, 0);
1169 
1170 	/*
1171 	 * Decide whether to enable randomization of user mappings.
1172 	 * First, reset user preferences for the setid binaries.
1173 	 * Then, account for the support of the randomization by the
1174 	 * ABI, by user preferences, and make special treatment for
1175 	 * PIE binaries.
1176 	 */
1177 	if (imgp->credential_setid) {
1178 		PROC_LOCK(imgp->proc);
1179 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1180 		PROC_UNLOCK(imgp->proc);
1181 	}
1182 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1183 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1184 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1185 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1186 		    ("et_dyn_addr == RAND and !ASLR"));
1187 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1188 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1189 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1190 		imgp->map_flags |= MAP_ASLR;
1191 		/*
1192 		 * If user does not care about sbrk, utilize the bss
1193 		 * grow region for mappings as well.  We can select
1194 		 * the base for the image anywere and still not suffer
1195 		 * from the fragmentation.
1196 		 */
1197 		if (!__elfN(aslr_honor_sbrk) ||
1198 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1199 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1200 	}
1201 
1202 	error = exec_new_vmspace(imgp, sv);
1203 	vmspace = imgp->proc->p_vmspace;
1204 	map = &vmspace->vm_map;
1205 
1206 	imgp->proc->p_sysent = sv;
1207 
1208 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1209 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1210 		KASSERT((map->flags & MAP_ASLR) != 0,
1211 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1212 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1213 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1214 		    /* reserve half of the address space to interpreter */
1215 		    maxv / 2, 1UL << flsl(maxalign));
1216 	}
1217 
1218 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1219 	if (error != 0)
1220 		goto ret;
1221 
1222 	error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1223 	if (error != 0)
1224 		goto ret;
1225 
1226 	error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1227 	if (error != 0)
1228 		goto ret;
1229 
1230 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1231 
1232 	/*
1233 	 * We load the dynamic linker where a userland call
1234 	 * to mmap(0, ...) would put it.  The rationale behind this
1235 	 * calculation is that it leaves room for the heap to grow to
1236 	 * its maximum allowed size.
1237 	 */
1238 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1239 	    RLIMIT_DATA));
1240 	if ((map->flags & MAP_ASLR) != 0) {
1241 		maxv1 = maxv / 2 + addr / 2;
1242 		MPASS(maxv1 >= addr);	/* No overflow */
1243 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1244 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1245 	} else {
1246 		map->anon_loc = addr;
1247 	}
1248 
1249 	imgp->entry_addr = entry;
1250 
1251 	if (interp != NULL) {
1252 		VOP_UNLOCK(imgp->vp, 0);
1253 		if ((map->flags & MAP_ASLR) != 0) {
1254 			/* Assume that interpeter fits into 1/4 of AS */
1255 			maxv1 = maxv / 2 + addr / 2;
1256 			MPASS(maxv1 >= addr);	/* No overflow */
1257 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1258 			    maxv1, PAGE_SIZE);
1259 		}
1260 		error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1261 		    &imgp->entry_addr);
1262 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1263 		if (error != 0)
1264 			goto ret;
1265 	} else
1266 		addr = et_dyn_addr;
1267 
1268 	/*
1269 	 * Construct auxargs table (used by the fixup routine)
1270 	 */
1271 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1272 	if (elf_auxargs == NULL) {
1273 		VOP_UNLOCK(imgp->vp, 0);
1274 		elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1275 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1276 	}
1277 	elf_auxargs->execfd = -1;
1278 	elf_auxargs->phdr = proghdr + et_dyn_addr;
1279 	elf_auxargs->phent = hdr->e_phentsize;
1280 	elf_auxargs->phnum = hdr->e_phnum;
1281 	elf_auxargs->pagesz = PAGE_SIZE;
1282 	elf_auxargs->base = addr;
1283 	elf_auxargs->flags = 0;
1284 	elf_auxargs->entry = entry;
1285 	elf_auxargs->hdr_eflags = hdr->e_flags;
1286 
1287 	imgp->auxargs = elf_auxargs;
1288 	imgp->interpreted = 0;
1289 	imgp->reloc_base = addr;
1290 	imgp->proc->p_osrel = osrel;
1291 	imgp->proc->p_fctl0 = fctl0;
1292 	imgp->proc->p_elf_machine = hdr->e_machine;
1293 	imgp->proc->p_elf_flags = hdr->e_flags;
1294 
1295 ret:
1296 	if (free_interp)
1297 		free(interp, M_TEMP);
1298 	return (error);
1299 }
1300 
1301 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1302 
1303 int
1304 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1305 {
1306 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1307 	Elf_Auxinfo *argarray, *pos;
1308 	Elf_Addr *base, *auxbase;
1309 	int error;
1310 
1311 	base = (Elf_Addr *)*stack_base;
1312 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1313 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1314 	    M_WAITOK | M_ZERO);
1315 
1316 	if (args->execfd != -1)
1317 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1318 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1319 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1320 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1321 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1322 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1323 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1324 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1325 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1326 	if (imgp->execpathp != 0)
1327 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1328 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1329 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1330 	if (imgp->canary != 0) {
1331 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1332 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1333 	}
1334 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1335 	if (imgp->pagesizes != 0) {
1336 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1337 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1338 	}
1339 	if (imgp->sysent->sv_timekeep_base != 0) {
1340 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1341 		    imgp->sysent->sv_timekeep_base);
1342 	}
1343 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1344 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1345 	    imgp->sysent->sv_stackprot);
1346 	if (imgp->sysent->sv_hwcap != NULL)
1347 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1348 	if (imgp->sysent->sv_hwcap2 != NULL)
1349 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1350 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1351 
1352 	free(imgp->auxargs, M_TEMP);
1353 	imgp->auxargs = NULL;
1354 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1355 
1356 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1357 	free(argarray, M_TEMP);
1358 	if (error != 0)
1359 		return (error);
1360 
1361 	base--;
1362 	if (suword(base, imgp->args->argc) == -1)
1363 		return (EFAULT);
1364 	*stack_base = (register_t *)base;
1365 	return (0);
1366 }
1367 
1368 /*
1369  * Code for generating ELF core dumps.
1370  */
1371 
1372 typedef void (*segment_callback)(vm_map_entry_t, void *);
1373 
1374 /* Closure for cb_put_phdr(). */
1375 struct phdr_closure {
1376 	Elf_Phdr *phdr;		/* Program header to fill in */
1377 	Elf_Off offset;		/* Offset of segment in core file */
1378 };
1379 
1380 /* Closure for cb_size_segment(). */
1381 struct sseg_closure {
1382 	int count;		/* Count of writable segments. */
1383 	size_t size;		/* Total size of all writable segments. */
1384 };
1385 
1386 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1387 
1388 struct note_info {
1389 	int		type;		/* Note type. */
1390 	outfunc_t 	outfunc; 	/* Output function. */
1391 	void		*outarg;	/* Argument for the output function. */
1392 	size_t		outsize;	/* Output size. */
1393 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1394 };
1395 
1396 TAILQ_HEAD(note_info_list, note_info);
1397 
1398 /* Coredump output parameters. */
1399 struct coredump_params {
1400 	off_t		offset;
1401 	struct ucred	*active_cred;
1402 	struct ucred	*file_cred;
1403 	struct thread	*td;
1404 	struct vnode	*vp;
1405 	struct compressor *comp;
1406 };
1407 
1408 extern int compress_user_cores;
1409 extern int compress_user_cores_level;
1410 
1411 static void cb_put_phdr(vm_map_entry_t, void *);
1412 static void cb_size_segment(vm_map_entry_t, void *);
1413 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1414     enum uio_seg);
1415 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1416 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1417     struct note_info_list *, size_t);
1418 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1419     size_t *);
1420 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1421 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1422 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1423 static int sbuf_drain_core_output(void *, const char *, int);
1424 
1425 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1426 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1427 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1428 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1429 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1430 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1431 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1432 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1433 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1434 static void note_procstat_files(void *, struct sbuf *, size_t *);
1435 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1436 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1437 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1438 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1439 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1440 
1441 /*
1442  * Write out a core segment to the compression stream.
1443  */
1444 static int
1445 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1446 {
1447 	u_int chunk_len;
1448 	int error;
1449 
1450 	while (len > 0) {
1451 		chunk_len = MIN(len, CORE_BUF_SIZE);
1452 
1453 		/*
1454 		 * We can get EFAULT error here.
1455 		 * In that case zero out the current chunk of the segment.
1456 		 */
1457 		error = copyin(base, buf, chunk_len);
1458 		if (error != 0)
1459 			bzero(buf, chunk_len);
1460 		error = compressor_write(p->comp, buf, chunk_len);
1461 		if (error != 0)
1462 			break;
1463 		base += chunk_len;
1464 		len -= chunk_len;
1465 	}
1466 	return (error);
1467 }
1468 
1469 static int
1470 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1471 {
1472 
1473 	return (core_write((struct coredump_params *)arg, base, len, offset,
1474 	    UIO_SYSSPACE));
1475 }
1476 
1477 static int
1478 core_write(struct coredump_params *p, const void *base, size_t len,
1479     off_t offset, enum uio_seg seg)
1480 {
1481 
1482 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1483 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1484 	    p->active_cred, p->file_cred, NULL, p->td));
1485 }
1486 
1487 static int
1488 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1489     void *tmpbuf)
1490 {
1491 	int error;
1492 
1493 	if (p->comp != NULL)
1494 		return (compress_chunk(p, base, tmpbuf, len));
1495 
1496 	/*
1497 	 * EFAULT is a non-fatal error that we can get, for example,
1498 	 * if the segment is backed by a file but extends beyond its
1499 	 * end.
1500 	 */
1501 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1502 	if (error == EFAULT) {
1503 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1504 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1505 		    "for process %s\n", base, len, offset, curproc->p_comm);
1506 
1507 		/*
1508 		 * Write a "real" zero byte at the end of the target region
1509 		 * in the case this is the last segment.
1510 		 * The intermediate space will be implicitly zero-filled.
1511 		 */
1512 		error = core_write(p, zero_region, 1, offset + len - 1,
1513 		    UIO_SYSSPACE);
1514 	}
1515 	return (error);
1516 }
1517 
1518 /*
1519  * Drain into a core file.
1520  */
1521 static int
1522 sbuf_drain_core_output(void *arg, const char *data, int len)
1523 {
1524 	struct coredump_params *p;
1525 	int error, locked;
1526 
1527 	p = (struct coredump_params *)arg;
1528 
1529 	/*
1530 	 * Some kern_proc out routines that print to this sbuf may
1531 	 * call us with the process lock held. Draining with the
1532 	 * non-sleepable lock held is unsafe. The lock is needed for
1533 	 * those routines when dumping a live process. In our case we
1534 	 * can safely release the lock before draining and acquire
1535 	 * again after.
1536 	 */
1537 	locked = PROC_LOCKED(p->td->td_proc);
1538 	if (locked)
1539 		PROC_UNLOCK(p->td->td_proc);
1540 	if (p->comp != NULL)
1541 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1542 	else
1543 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1544 		    UIO_SYSSPACE);
1545 	if (locked)
1546 		PROC_LOCK(p->td->td_proc);
1547 	if (error != 0)
1548 		return (-error);
1549 	p->offset += len;
1550 	return (len);
1551 }
1552 
1553 int
1554 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1555 {
1556 	struct ucred *cred = td->td_ucred;
1557 	int error = 0;
1558 	struct sseg_closure seginfo;
1559 	struct note_info_list notelst;
1560 	struct coredump_params params;
1561 	struct note_info *ninfo;
1562 	void *hdr, *tmpbuf;
1563 	size_t hdrsize, notesz, coresize;
1564 
1565 	hdr = NULL;
1566 	tmpbuf = NULL;
1567 	TAILQ_INIT(&notelst);
1568 
1569 	/* Size the program segments. */
1570 	seginfo.count = 0;
1571 	seginfo.size = 0;
1572 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1573 
1574 	/*
1575 	 * Collect info about the core file header area.
1576 	 */
1577 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1578 	if (seginfo.count + 1 >= PN_XNUM)
1579 		hdrsize += sizeof(Elf_Shdr);
1580 	__elfN(prepare_notes)(td, &notelst, &notesz);
1581 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1582 
1583 	/* Set up core dump parameters. */
1584 	params.offset = 0;
1585 	params.active_cred = cred;
1586 	params.file_cred = NOCRED;
1587 	params.td = td;
1588 	params.vp = vp;
1589 	params.comp = NULL;
1590 
1591 #ifdef RACCT
1592 	if (racct_enable) {
1593 		PROC_LOCK(td->td_proc);
1594 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1595 		PROC_UNLOCK(td->td_proc);
1596 		if (error != 0) {
1597 			error = EFAULT;
1598 			goto done;
1599 		}
1600 	}
1601 #endif
1602 	if (coresize >= limit) {
1603 		error = EFAULT;
1604 		goto done;
1605 	}
1606 
1607 	/* Create a compression stream if necessary. */
1608 	if (compress_user_cores != 0) {
1609 		params.comp = compressor_init(core_compressed_write,
1610 		    compress_user_cores, CORE_BUF_SIZE,
1611 		    compress_user_cores_level, &params);
1612 		if (params.comp == NULL) {
1613 			error = EFAULT;
1614 			goto done;
1615 		}
1616 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1617         }
1618 
1619 	/*
1620 	 * Allocate memory for building the header, fill it up,
1621 	 * and write it out following the notes.
1622 	 */
1623 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1624 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1625 	    notesz);
1626 
1627 	/* Write the contents of all of the writable segments. */
1628 	if (error == 0) {
1629 		Elf_Phdr *php;
1630 		off_t offset;
1631 		int i;
1632 
1633 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1634 		offset = round_page(hdrsize + notesz);
1635 		for (i = 0; i < seginfo.count; i++) {
1636 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1637 			    php->p_filesz, offset, &params, tmpbuf);
1638 			if (error != 0)
1639 				break;
1640 			offset += php->p_filesz;
1641 			php++;
1642 		}
1643 		if (error == 0 && params.comp != NULL)
1644 			error = compressor_flush(params.comp);
1645 	}
1646 	if (error) {
1647 		log(LOG_WARNING,
1648 		    "Failed to write core file for process %s (error %d)\n",
1649 		    curproc->p_comm, error);
1650 	}
1651 
1652 done:
1653 	free(tmpbuf, M_TEMP);
1654 	if (params.comp != NULL)
1655 		compressor_fini(params.comp);
1656 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1657 		TAILQ_REMOVE(&notelst, ninfo, link);
1658 		free(ninfo, M_TEMP);
1659 	}
1660 	if (hdr != NULL)
1661 		free(hdr, M_TEMP);
1662 
1663 	return (error);
1664 }
1665 
1666 /*
1667  * A callback for each_dumpable_segment() to write out the segment's
1668  * program header entry.
1669  */
1670 static void
1671 cb_put_phdr(vm_map_entry_t entry, void *closure)
1672 {
1673 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1674 	Elf_Phdr *phdr = phc->phdr;
1675 
1676 	phc->offset = round_page(phc->offset);
1677 
1678 	phdr->p_type = PT_LOAD;
1679 	phdr->p_offset = phc->offset;
1680 	phdr->p_vaddr = entry->start;
1681 	phdr->p_paddr = 0;
1682 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1683 	phdr->p_align = PAGE_SIZE;
1684 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1685 
1686 	phc->offset += phdr->p_filesz;
1687 	phc->phdr++;
1688 }
1689 
1690 /*
1691  * A callback for each_dumpable_segment() to gather information about
1692  * the number of segments and their total size.
1693  */
1694 static void
1695 cb_size_segment(vm_map_entry_t entry, void *closure)
1696 {
1697 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1698 
1699 	ssc->count++;
1700 	ssc->size += entry->end - entry->start;
1701 }
1702 
1703 /*
1704  * For each writable segment in the process's memory map, call the given
1705  * function with a pointer to the map entry and some arbitrary
1706  * caller-supplied data.
1707  */
1708 static void
1709 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1710 {
1711 	struct proc *p = td->td_proc;
1712 	vm_map_t map = &p->p_vmspace->vm_map;
1713 	vm_map_entry_t entry;
1714 	vm_object_t backing_object, object;
1715 	boolean_t ignore_entry;
1716 
1717 	vm_map_lock_read(map);
1718 	for (entry = map->header.next; entry != &map->header;
1719 	    entry = entry->next) {
1720 		/*
1721 		 * Don't dump inaccessible mappings, deal with legacy
1722 		 * coredump mode.
1723 		 *
1724 		 * Note that read-only segments related to the elf binary
1725 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1726 		 * need to arbitrarily ignore such segments.
1727 		 */
1728 		if (elf_legacy_coredump) {
1729 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1730 				continue;
1731 		} else {
1732 			if ((entry->protection & VM_PROT_ALL) == 0)
1733 				continue;
1734 		}
1735 
1736 		/*
1737 		 * Dont include memory segment in the coredump if
1738 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1739 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1740 		 * kernel map).
1741 		 */
1742 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1743 			continue;
1744 
1745 		if ((object = entry->object.vm_object) == NULL)
1746 			continue;
1747 
1748 		/* Ignore memory-mapped devices and such things. */
1749 		VM_OBJECT_RLOCK(object);
1750 		while ((backing_object = object->backing_object) != NULL) {
1751 			VM_OBJECT_RLOCK(backing_object);
1752 			VM_OBJECT_RUNLOCK(object);
1753 			object = backing_object;
1754 		}
1755 		ignore_entry = object->type != OBJT_DEFAULT &&
1756 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1757 		    object->type != OBJT_PHYS;
1758 		VM_OBJECT_RUNLOCK(object);
1759 		if (ignore_entry)
1760 			continue;
1761 
1762 		(*func)(entry, closure);
1763 	}
1764 	vm_map_unlock_read(map);
1765 }
1766 
1767 /*
1768  * Write the core file header to the file, including padding up to
1769  * the page boundary.
1770  */
1771 static int
1772 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1773     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1774 {
1775 	struct note_info *ninfo;
1776 	struct sbuf *sb;
1777 	int error;
1778 
1779 	/* Fill in the header. */
1780 	bzero(hdr, hdrsize);
1781 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1782 
1783 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1784 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1785 	sbuf_start_section(sb, NULL);
1786 	sbuf_bcat(sb, hdr, hdrsize);
1787 	TAILQ_FOREACH(ninfo, notelst, link)
1788 	    __elfN(putnote)(ninfo, sb);
1789 	/* Align up to a page boundary for the program segments. */
1790 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1791 	error = sbuf_finish(sb);
1792 	sbuf_delete(sb);
1793 
1794 	return (error);
1795 }
1796 
1797 static void
1798 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1799     size_t *sizep)
1800 {
1801 	struct proc *p;
1802 	struct thread *thr;
1803 	size_t size;
1804 
1805 	p = td->td_proc;
1806 	size = 0;
1807 
1808 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1809 
1810 	/*
1811 	 * To have the debugger select the right thread (LWP) as the initial
1812 	 * thread, we dump the state of the thread passed to us in td first.
1813 	 * This is the thread that causes the core dump and thus likely to
1814 	 * be the right thread one wants to have selected in the debugger.
1815 	 */
1816 	thr = td;
1817 	while (thr != NULL) {
1818 		size += register_note(list, NT_PRSTATUS,
1819 		    __elfN(note_prstatus), thr);
1820 		size += register_note(list, NT_FPREGSET,
1821 		    __elfN(note_fpregset), thr);
1822 		size += register_note(list, NT_THRMISC,
1823 		    __elfN(note_thrmisc), thr);
1824 		size += register_note(list, NT_PTLWPINFO,
1825 		    __elfN(note_ptlwpinfo), thr);
1826 		size += register_note(list, -1,
1827 		    __elfN(note_threadmd), thr);
1828 
1829 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1830 		    TAILQ_NEXT(thr, td_plist);
1831 		if (thr == td)
1832 			thr = TAILQ_NEXT(thr, td_plist);
1833 	}
1834 
1835 	size += register_note(list, NT_PROCSTAT_PROC,
1836 	    __elfN(note_procstat_proc), p);
1837 	size += register_note(list, NT_PROCSTAT_FILES,
1838 	    note_procstat_files, p);
1839 	size += register_note(list, NT_PROCSTAT_VMMAP,
1840 	    note_procstat_vmmap, p);
1841 	size += register_note(list, NT_PROCSTAT_GROUPS,
1842 	    note_procstat_groups, p);
1843 	size += register_note(list, NT_PROCSTAT_UMASK,
1844 	    note_procstat_umask, p);
1845 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1846 	    note_procstat_rlimit, p);
1847 	size += register_note(list, NT_PROCSTAT_OSREL,
1848 	    note_procstat_osrel, p);
1849 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1850 	    __elfN(note_procstat_psstrings), p);
1851 	size += register_note(list, NT_PROCSTAT_AUXV,
1852 	    __elfN(note_procstat_auxv), p);
1853 
1854 	*sizep = size;
1855 }
1856 
1857 static void
1858 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1859     size_t notesz)
1860 {
1861 	Elf_Ehdr *ehdr;
1862 	Elf_Phdr *phdr;
1863 	Elf_Shdr *shdr;
1864 	struct phdr_closure phc;
1865 
1866 	ehdr = (Elf_Ehdr *)hdr;
1867 
1868 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1869 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1870 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1871 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1872 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1873 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1874 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1875 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1876 	ehdr->e_ident[EI_ABIVERSION] = 0;
1877 	ehdr->e_ident[EI_PAD] = 0;
1878 	ehdr->e_type = ET_CORE;
1879 	ehdr->e_machine = td->td_proc->p_elf_machine;
1880 	ehdr->e_version = EV_CURRENT;
1881 	ehdr->e_entry = 0;
1882 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1883 	ehdr->e_flags = td->td_proc->p_elf_flags;
1884 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1885 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1886 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1887 	ehdr->e_shstrndx = SHN_UNDEF;
1888 	if (numsegs + 1 < PN_XNUM) {
1889 		ehdr->e_phnum = numsegs + 1;
1890 		ehdr->e_shnum = 0;
1891 	} else {
1892 		ehdr->e_phnum = PN_XNUM;
1893 		ehdr->e_shnum = 1;
1894 
1895 		ehdr->e_shoff = ehdr->e_phoff +
1896 		    (numsegs + 1) * ehdr->e_phentsize;
1897 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1898 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1899 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1900 
1901 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1902 		memset(shdr, 0, sizeof(*shdr));
1903 		/*
1904 		 * A special first section is used to hold large segment and
1905 		 * section counts.  This was proposed by Sun Microsystems in
1906 		 * Solaris and has been adopted by Linux; the standard ELF
1907 		 * tools are already familiar with the technique.
1908 		 *
1909 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1910 		 * (or 12-7 depending on the version of the document) for more
1911 		 * details.
1912 		 */
1913 		shdr->sh_type = SHT_NULL;
1914 		shdr->sh_size = ehdr->e_shnum;
1915 		shdr->sh_link = ehdr->e_shstrndx;
1916 		shdr->sh_info = numsegs + 1;
1917 	}
1918 
1919 	/*
1920 	 * Fill in the program header entries.
1921 	 */
1922 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1923 
1924 	/* The note segement. */
1925 	phdr->p_type = PT_NOTE;
1926 	phdr->p_offset = hdrsize;
1927 	phdr->p_vaddr = 0;
1928 	phdr->p_paddr = 0;
1929 	phdr->p_filesz = notesz;
1930 	phdr->p_memsz = 0;
1931 	phdr->p_flags = PF_R;
1932 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1933 	phdr++;
1934 
1935 	/* All the writable segments from the program. */
1936 	phc.phdr = phdr;
1937 	phc.offset = round_page(hdrsize + notesz);
1938 	each_dumpable_segment(td, cb_put_phdr, &phc);
1939 }
1940 
1941 static size_t
1942 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1943 {
1944 	struct note_info *ninfo;
1945 	size_t size, notesize;
1946 
1947 	size = 0;
1948 	out(arg, NULL, &size);
1949 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1950 	ninfo->type = type;
1951 	ninfo->outfunc = out;
1952 	ninfo->outarg = arg;
1953 	ninfo->outsize = size;
1954 	TAILQ_INSERT_TAIL(list, ninfo, link);
1955 
1956 	if (type == -1)
1957 		return (size);
1958 
1959 	notesize = sizeof(Elf_Note) +		/* note header */
1960 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1961 						/* note name */
1962 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1963 
1964 	return (notesize);
1965 }
1966 
1967 static size_t
1968 append_note_data(const void *src, void *dst, size_t len)
1969 {
1970 	size_t padded_len;
1971 
1972 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1973 	if (dst != NULL) {
1974 		bcopy(src, dst, len);
1975 		bzero((char *)dst + len, padded_len - len);
1976 	}
1977 	return (padded_len);
1978 }
1979 
1980 size_t
1981 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1982 {
1983 	Elf_Note *note;
1984 	char *buf;
1985 	size_t notesize;
1986 
1987 	buf = dst;
1988 	if (buf != NULL) {
1989 		note = (Elf_Note *)buf;
1990 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1991 		note->n_descsz = size;
1992 		note->n_type = type;
1993 		buf += sizeof(*note);
1994 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1995 		    sizeof(FREEBSD_ABI_VENDOR));
1996 		append_note_data(src, buf, size);
1997 		if (descp != NULL)
1998 			*descp = buf;
1999 	}
2000 
2001 	notesize = sizeof(Elf_Note) +		/* note header */
2002 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2003 						/* note name */
2004 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
2005 
2006 	return (notesize);
2007 }
2008 
2009 static void
2010 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2011 {
2012 	Elf_Note note;
2013 	ssize_t old_len, sect_len;
2014 	size_t new_len, descsz, i;
2015 
2016 	if (ninfo->type == -1) {
2017 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2018 		return;
2019 	}
2020 
2021 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2022 	note.n_descsz = ninfo->outsize;
2023 	note.n_type = ninfo->type;
2024 
2025 	sbuf_bcat(sb, &note, sizeof(note));
2026 	sbuf_start_section(sb, &old_len);
2027 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2028 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2029 	if (note.n_descsz == 0)
2030 		return;
2031 	sbuf_start_section(sb, &old_len);
2032 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2033 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2034 	if (sect_len < 0)
2035 		return;
2036 
2037 	new_len = (size_t)sect_len;
2038 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2039 	if (new_len < descsz) {
2040 		/*
2041 		 * It is expected that individual note emitters will correctly
2042 		 * predict their expected output size and fill up to that size
2043 		 * themselves, padding in a format-specific way if needed.
2044 		 * However, in case they don't, just do it here with zeros.
2045 		 */
2046 		for (i = 0; i < descsz - new_len; i++)
2047 			sbuf_putc(sb, 0);
2048 	} else if (new_len > descsz) {
2049 		/*
2050 		 * We can't always truncate sb -- we may have drained some
2051 		 * of it already.
2052 		 */
2053 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2054 		    "read it (%zu > %zu).  Since it is longer than "
2055 		    "expected, this coredump's notes are corrupt.  THIS "
2056 		    "IS A BUG in the note_procstat routine for type %u.\n",
2057 		    __func__, (unsigned)note.n_type, new_len, descsz,
2058 		    (unsigned)note.n_type));
2059 	}
2060 }
2061 
2062 /*
2063  * Miscellaneous note out functions.
2064  */
2065 
2066 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2067 #include <compat/freebsd32/freebsd32.h>
2068 #include <compat/freebsd32/freebsd32_signal.h>
2069 
2070 typedef struct prstatus32 elf_prstatus_t;
2071 typedef struct prpsinfo32 elf_prpsinfo_t;
2072 typedef struct fpreg32 elf_prfpregset_t;
2073 typedef struct fpreg32 elf_fpregset_t;
2074 typedef struct reg32 elf_gregset_t;
2075 typedef struct thrmisc32 elf_thrmisc_t;
2076 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2077 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2078 typedef uint32_t elf_ps_strings_t;
2079 #else
2080 typedef prstatus_t elf_prstatus_t;
2081 typedef prpsinfo_t elf_prpsinfo_t;
2082 typedef prfpregset_t elf_prfpregset_t;
2083 typedef prfpregset_t elf_fpregset_t;
2084 typedef gregset_t elf_gregset_t;
2085 typedef thrmisc_t elf_thrmisc_t;
2086 #define ELF_KERN_PROC_MASK	0
2087 typedef struct kinfo_proc elf_kinfo_proc_t;
2088 typedef vm_offset_t elf_ps_strings_t;
2089 #endif
2090 
2091 static void
2092 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2093 {
2094 	struct sbuf sbarg;
2095 	size_t len;
2096 	char *cp, *end;
2097 	struct proc *p;
2098 	elf_prpsinfo_t *psinfo;
2099 	int error;
2100 
2101 	p = (struct proc *)arg;
2102 	if (sb != NULL) {
2103 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2104 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2105 		psinfo->pr_version = PRPSINFO_VERSION;
2106 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2107 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2108 		PROC_LOCK(p);
2109 		if (p->p_args != NULL) {
2110 			len = sizeof(psinfo->pr_psargs) - 1;
2111 			if (len > p->p_args->ar_length)
2112 				len = p->p_args->ar_length;
2113 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2114 			PROC_UNLOCK(p);
2115 			error = 0;
2116 		} else {
2117 			_PHOLD(p);
2118 			PROC_UNLOCK(p);
2119 			sbuf_new(&sbarg, psinfo->pr_psargs,
2120 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2121 			error = proc_getargv(curthread, p, &sbarg);
2122 			PRELE(p);
2123 			if (sbuf_finish(&sbarg) == 0)
2124 				len = sbuf_len(&sbarg) - 1;
2125 			else
2126 				len = sizeof(psinfo->pr_psargs) - 1;
2127 			sbuf_delete(&sbarg);
2128 		}
2129 		if (error || len == 0)
2130 			strlcpy(psinfo->pr_psargs, p->p_comm,
2131 			    sizeof(psinfo->pr_psargs));
2132 		else {
2133 			KASSERT(len < sizeof(psinfo->pr_psargs),
2134 			    ("len is too long: %zu vs %zu", len,
2135 			    sizeof(psinfo->pr_psargs)));
2136 			cp = psinfo->pr_psargs;
2137 			end = cp + len - 1;
2138 			for (;;) {
2139 				cp = memchr(cp, '\0', end - cp);
2140 				if (cp == NULL)
2141 					break;
2142 				*cp = ' ';
2143 			}
2144 		}
2145 		psinfo->pr_pid = p->p_pid;
2146 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2147 		free(psinfo, M_TEMP);
2148 	}
2149 	*sizep = sizeof(*psinfo);
2150 }
2151 
2152 static void
2153 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2154 {
2155 	struct thread *td;
2156 	elf_prstatus_t *status;
2157 
2158 	td = (struct thread *)arg;
2159 	if (sb != NULL) {
2160 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2161 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2162 		status->pr_version = PRSTATUS_VERSION;
2163 		status->pr_statussz = sizeof(elf_prstatus_t);
2164 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2165 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2166 		status->pr_osreldate = osreldate;
2167 		status->pr_cursig = td->td_proc->p_sig;
2168 		status->pr_pid = td->td_tid;
2169 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2170 		fill_regs32(td, &status->pr_reg);
2171 #else
2172 		fill_regs(td, &status->pr_reg);
2173 #endif
2174 		sbuf_bcat(sb, status, sizeof(*status));
2175 		free(status, M_TEMP);
2176 	}
2177 	*sizep = sizeof(*status);
2178 }
2179 
2180 static void
2181 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2182 {
2183 	struct thread *td;
2184 	elf_prfpregset_t *fpregset;
2185 
2186 	td = (struct thread *)arg;
2187 	if (sb != NULL) {
2188 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2189 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2190 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2191 		fill_fpregs32(td, fpregset);
2192 #else
2193 		fill_fpregs(td, fpregset);
2194 #endif
2195 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2196 		free(fpregset, M_TEMP);
2197 	}
2198 	*sizep = sizeof(*fpregset);
2199 }
2200 
2201 static void
2202 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2203 {
2204 	struct thread *td;
2205 	elf_thrmisc_t thrmisc;
2206 
2207 	td = (struct thread *)arg;
2208 	if (sb != NULL) {
2209 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2210 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2211 		strcpy(thrmisc.pr_tname, td->td_name);
2212 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2213 	}
2214 	*sizep = sizeof(thrmisc);
2215 }
2216 
2217 static void
2218 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2219 {
2220 	struct thread *td;
2221 	size_t size;
2222 	int structsize;
2223 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2224 	struct ptrace_lwpinfo32 pl;
2225 #else
2226 	struct ptrace_lwpinfo pl;
2227 #endif
2228 
2229 	td = (struct thread *)arg;
2230 	size = sizeof(structsize) + sizeof(pl);
2231 	if (sb != NULL) {
2232 		KASSERT(*sizep == size, ("invalid size"));
2233 		structsize = sizeof(pl);
2234 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2235 		bzero(&pl, sizeof(pl));
2236 		pl.pl_lwpid = td->td_tid;
2237 		pl.pl_event = PL_EVENT_NONE;
2238 		pl.pl_sigmask = td->td_sigmask;
2239 		pl.pl_siglist = td->td_siglist;
2240 		if (td->td_si.si_signo != 0) {
2241 			pl.pl_event = PL_EVENT_SIGNAL;
2242 			pl.pl_flags |= PL_FLAG_SI;
2243 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2244 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2245 #else
2246 			pl.pl_siginfo = td->td_si;
2247 #endif
2248 		}
2249 		strcpy(pl.pl_tdname, td->td_name);
2250 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2251 		sbuf_bcat(sb, &pl, sizeof(pl));
2252 	}
2253 	*sizep = size;
2254 }
2255 
2256 /*
2257  * Allow for MD specific notes, as well as any MD
2258  * specific preparations for writing MI notes.
2259  */
2260 static void
2261 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2262 {
2263 	struct thread *td;
2264 	void *buf;
2265 	size_t size;
2266 
2267 	td = (struct thread *)arg;
2268 	size = *sizep;
2269 	if (size != 0 && sb != NULL)
2270 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2271 	else
2272 		buf = NULL;
2273 	size = 0;
2274 	__elfN(dump_thread)(td, buf, &size);
2275 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2276 	if (size != 0 && sb != NULL)
2277 		sbuf_bcat(sb, buf, size);
2278 	free(buf, M_TEMP);
2279 	*sizep = size;
2280 }
2281 
2282 #ifdef KINFO_PROC_SIZE
2283 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2284 #endif
2285 
2286 static void
2287 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2288 {
2289 	struct proc *p;
2290 	size_t size;
2291 	int structsize;
2292 
2293 	p = (struct proc *)arg;
2294 	size = sizeof(structsize) + p->p_numthreads *
2295 	    sizeof(elf_kinfo_proc_t);
2296 
2297 	if (sb != NULL) {
2298 		KASSERT(*sizep == size, ("invalid size"));
2299 		structsize = sizeof(elf_kinfo_proc_t);
2300 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2301 		PROC_LOCK(p);
2302 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2303 	}
2304 	*sizep = size;
2305 }
2306 
2307 #ifdef KINFO_FILE_SIZE
2308 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2309 #endif
2310 
2311 static void
2312 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2313 {
2314 	struct proc *p;
2315 	size_t size, sect_sz, i;
2316 	ssize_t start_len, sect_len;
2317 	int structsize, filedesc_flags;
2318 
2319 	if (coredump_pack_fileinfo)
2320 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2321 	else
2322 		filedesc_flags = 0;
2323 
2324 	p = (struct proc *)arg;
2325 	structsize = sizeof(struct kinfo_file);
2326 	if (sb == NULL) {
2327 		size = 0;
2328 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2329 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2330 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2331 		PROC_LOCK(p);
2332 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2333 		sbuf_finish(sb);
2334 		sbuf_delete(sb);
2335 		*sizep = size;
2336 	} else {
2337 		sbuf_start_section(sb, &start_len);
2338 
2339 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2340 		PROC_LOCK(p);
2341 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2342 		    filedesc_flags);
2343 
2344 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2345 		if (sect_len < 0)
2346 			return;
2347 		sect_sz = sect_len;
2348 
2349 		KASSERT(sect_sz <= *sizep,
2350 		    ("kern_proc_filedesc_out did not respect maxlen; "
2351 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2352 		     sect_sz - sizeof(structsize)));
2353 
2354 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2355 			sbuf_putc(sb, 0);
2356 	}
2357 }
2358 
2359 #ifdef KINFO_VMENTRY_SIZE
2360 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2361 #endif
2362 
2363 static void
2364 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2365 {
2366 	struct proc *p;
2367 	size_t size;
2368 	int structsize, vmmap_flags;
2369 
2370 	if (coredump_pack_vmmapinfo)
2371 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2372 	else
2373 		vmmap_flags = 0;
2374 
2375 	p = (struct proc *)arg;
2376 	structsize = sizeof(struct kinfo_vmentry);
2377 	if (sb == NULL) {
2378 		size = 0;
2379 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2380 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2381 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2382 		PROC_LOCK(p);
2383 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2384 		sbuf_finish(sb);
2385 		sbuf_delete(sb);
2386 		*sizep = size;
2387 	} else {
2388 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2389 		PROC_LOCK(p);
2390 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2391 		    vmmap_flags);
2392 	}
2393 }
2394 
2395 static void
2396 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2397 {
2398 	struct proc *p;
2399 	size_t size;
2400 	int structsize;
2401 
2402 	p = (struct proc *)arg;
2403 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2404 	if (sb != NULL) {
2405 		KASSERT(*sizep == size, ("invalid size"));
2406 		structsize = sizeof(gid_t);
2407 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2408 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2409 		    sizeof(gid_t));
2410 	}
2411 	*sizep = size;
2412 }
2413 
2414 static void
2415 note_procstat_umask(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) + sizeof(p->p_fd->fd_cmask);
2423 	if (sb != NULL) {
2424 		KASSERT(*sizep == size, ("invalid size"));
2425 		structsize = sizeof(p->p_fd->fd_cmask);
2426 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2427 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2428 	}
2429 	*sizep = size;
2430 }
2431 
2432 static void
2433 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2434 {
2435 	struct proc *p;
2436 	struct rlimit rlim[RLIM_NLIMITS];
2437 	size_t size;
2438 	int structsize, i;
2439 
2440 	p = (struct proc *)arg;
2441 	size = sizeof(structsize) + sizeof(rlim);
2442 	if (sb != NULL) {
2443 		KASSERT(*sizep == size, ("invalid size"));
2444 		structsize = sizeof(rlim);
2445 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2446 		PROC_LOCK(p);
2447 		for (i = 0; i < RLIM_NLIMITS; i++)
2448 			lim_rlimit_proc(p, i, &rlim[i]);
2449 		PROC_UNLOCK(p);
2450 		sbuf_bcat(sb, rlim, sizeof(rlim));
2451 	}
2452 	*sizep = size;
2453 }
2454 
2455 static void
2456 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2457 {
2458 	struct proc *p;
2459 	size_t size;
2460 	int structsize;
2461 
2462 	p = (struct proc *)arg;
2463 	size = sizeof(structsize) + sizeof(p->p_osrel);
2464 	if (sb != NULL) {
2465 		KASSERT(*sizep == size, ("invalid size"));
2466 		structsize = sizeof(p->p_osrel);
2467 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2468 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2469 	}
2470 	*sizep = size;
2471 }
2472 
2473 static void
2474 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2475 {
2476 	struct proc *p;
2477 	elf_ps_strings_t ps_strings;
2478 	size_t size;
2479 	int structsize;
2480 
2481 	p = (struct proc *)arg;
2482 	size = sizeof(structsize) + sizeof(ps_strings);
2483 	if (sb != NULL) {
2484 		KASSERT(*sizep == size, ("invalid size"));
2485 		structsize = sizeof(ps_strings);
2486 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2487 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2488 #else
2489 		ps_strings = p->p_sysent->sv_psstrings;
2490 #endif
2491 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2492 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2493 	}
2494 	*sizep = size;
2495 }
2496 
2497 static void
2498 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2499 {
2500 	struct proc *p;
2501 	size_t size;
2502 	int structsize;
2503 
2504 	p = (struct proc *)arg;
2505 	if (sb == NULL) {
2506 		size = 0;
2507 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2508 		sbuf_set_drain(sb, sbuf_count_drain, &size);
2509 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2510 		PHOLD(p);
2511 		proc_getauxv(curthread, p, sb);
2512 		PRELE(p);
2513 		sbuf_finish(sb);
2514 		sbuf_delete(sb);
2515 		*sizep = size;
2516 	} else {
2517 		structsize = sizeof(Elf_Auxinfo);
2518 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2519 		PHOLD(p);
2520 		proc_getauxv(curthread, p, sb);
2521 		PRELE(p);
2522 	}
2523 }
2524 
2525 static boolean_t
2526 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2527     const char *note_vendor, const Elf_Phdr *pnote,
2528     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2529 {
2530 	const Elf_Note *note, *note0, *note_end;
2531 	const char *note_name;
2532 	char *buf;
2533 	int i, error;
2534 	boolean_t res;
2535 
2536 	/* We need some limit, might as well use PAGE_SIZE. */
2537 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2538 		return (FALSE);
2539 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2540 	if (pnote->p_offset > PAGE_SIZE ||
2541 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2542 		buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2543 		if (buf == NULL) {
2544 			VOP_UNLOCK(imgp->vp, 0);
2545 			buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2546 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2547 		}
2548 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2549 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2550 		    curthread->td_ucred, NOCRED, NULL, curthread);
2551 		if (error != 0) {
2552 			uprintf("i/o error PT_NOTE\n");
2553 			goto retf;
2554 		}
2555 		note = note0 = (const Elf_Note *)buf;
2556 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2557 	} else {
2558 		note = note0 = (const Elf_Note *)(imgp->image_header +
2559 		    pnote->p_offset);
2560 		note_end = (const Elf_Note *)(imgp->image_header +
2561 		    pnote->p_offset + pnote->p_filesz);
2562 		buf = NULL;
2563 	}
2564 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2565 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2566 		    (const char *)note < sizeof(Elf_Note)) {
2567 			goto retf;
2568 		}
2569 		if (note->n_namesz != checknote->n_namesz ||
2570 		    note->n_descsz != checknote->n_descsz ||
2571 		    note->n_type != checknote->n_type)
2572 			goto nextnote;
2573 		note_name = (const char *)(note + 1);
2574 		if (note_name + checknote->n_namesz >=
2575 		    (const char *)note_end || strncmp(note_vendor,
2576 		    note_name, checknote->n_namesz) != 0)
2577 			goto nextnote;
2578 
2579 		if (cb(note, cb_arg, &res))
2580 			goto ret;
2581 nextnote:
2582 		note = (const Elf_Note *)((const char *)(note + 1) +
2583 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2584 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2585 	}
2586 retf:
2587 	res = FALSE;
2588 ret:
2589 	free(buf, M_TEMP);
2590 	return (res);
2591 }
2592 
2593 struct brandnote_cb_arg {
2594 	Elf_Brandnote *brandnote;
2595 	int32_t *osrel;
2596 };
2597 
2598 static boolean_t
2599 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2600 {
2601 	struct brandnote_cb_arg *arg;
2602 
2603 	arg = arg0;
2604 
2605 	/*
2606 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2607 	 * necessary.
2608 	 */
2609 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2610 	    arg->brandnote->trans_osrel != NULL ?
2611 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2612 
2613 	return (TRUE);
2614 }
2615 
2616 static Elf_Note fctl_note = {
2617 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2618 	.n_descsz = sizeof(uint32_t),
2619 	.n_type = NT_FREEBSD_FEATURE_CTL,
2620 };
2621 
2622 struct fctl_cb_arg {
2623 	uint32_t *fctl0;
2624 };
2625 
2626 static boolean_t
2627 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2628 {
2629 	struct fctl_cb_arg *arg;
2630 	const Elf32_Word *desc;
2631 	uintptr_t p;
2632 
2633 	arg = arg0;
2634 	p = (uintptr_t)(note + 1);
2635 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2636 	desc = (const Elf32_Word *)p;
2637 	*arg->fctl0 = desc[0];
2638 	return (TRUE);
2639 }
2640 
2641 /*
2642  * Try to find the appropriate ABI-note section for checknote, fetch
2643  * the osreldate and feature control flags for binary from the ELF
2644  * OSABI-note.  Only the first page of the image is searched, the same
2645  * as for headers.
2646  */
2647 static boolean_t
2648 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2649     int32_t *osrel, uint32_t *fctl0)
2650 {
2651 	const Elf_Phdr *phdr;
2652 	const Elf_Ehdr *hdr;
2653 	struct brandnote_cb_arg b_arg;
2654 	struct fctl_cb_arg f_arg;
2655 	int i, j;
2656 
2657 	hdr = (const Elf_Ehdr *)imgp->image_header;
2658 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2659 	b_arg.brandnote = brandnote;
2660 	b_arg.osrel = osrel;
2661 	f_arg.fctl0 = fctl0;
2662 
2663 	for (i = 0; i < hdr->e_phnum; i++) {
2664 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2665 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2666 		    &b_arg)) {
2667 			for (j = 0; j < hdr->e_phnum; j++) {
2668 				if (phdr[j].p_type == PT_NOTE &&
2669 				    __elfN(parse_notes)(imgp, &fctl_note,
2670 				    FREEBSD_ABI_VENDOR, &phdr[j],
2671 				    note_fctl_cb, &f_arg))
2672 					break;
2673 			}
2674 			return (TRUE);
2675 		}
2676 	}
2677 	return (FALSE);
2678 
2679 }
2680 
2681 /*
2682  * Tell kern_execve.c about it, with a little help from the linker.
2683  */
2684 static struct execsw __elfN(execsw) = {
2685 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2686 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2687 };
2688 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2689 
2690 static vm_prot_t
2691 __elfN(trans_prot)(Elf_Word flags)
2692 {
2693 	vm_prot_t prot;
2694 
2695 	prot = 0;
2696 	if (flags & PF_X)
2697 		prot |= VM_PROT_EXECUTE;
2698 	if (flags & PF_W)
2699 		prot |= VM_PROT_WRITE;
2700 	if (flags & PF_R)
2701 		prot |= VM_PROT_READ;
2702 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2703 	if (i386_read_exec && (flags & PF_R))
2704 		prot |= VM_PROT_EXECUTE;
2705 #endif
2706 	return (prot);
2707 }
2708 
2709 static Elf_Word
2710 __elfN(untrans_prot)(vm_prot_t prot)
2711 {
2712 	Elf_Word flags;
2713 
2714 	flags = 0;
2715 	if (prot & VM_PROT_EXECUTE)
2716 		flags |= PF_X;
2717 	if (prot & VM_PROT_READ)
2718 		flags |= PF_R;
2719 	if (prot & VM_PROT_WRITE)
2720 		flags |= PF_W;
2721 	return (flags);
2722 }
2723