xref: /freebsd/sys/kern/imgact_elf.c (revision 0957b409a90fd597c1e9124cbaf3edd2b488f4ac)
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, int interp_name_len, 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, size_t pagesize);
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     size_t pagesize);
97 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
98 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99     int32_t *osrel);
100 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
101 static boolean_t __elfN(check_note)(struct image_params *imgp,
102     Elf_Brandnote *checknote, int32_t *osrel, uint32_t *fctl0);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
107     "");
108 
109 #define	CORE_BUF_SIZE	(16 * 1024)
110 
111 int __elfN(fallback_brand) = -1;
112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
113     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
114     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 
116 static int elf_legacy_coredump = 0;
117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
118     &elf_legacy_coredump, 0,
119     "include all and only RW pages in core dumps");
120 
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
124     defined(__riscv)
125 	1;
126 #else
127 	0;
128 #endif
129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
130     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
131     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 
133 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136     "enable execution from readable segments");
137 #endif
138 
139 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
140     "");
141 #define	ASLR_NODE_OID	__CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
142 
143 static int __elfN(aslr_enabled) = 0;
144 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
145     &__elfN(aslr_enabled), 0,
146     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
147     ": enable address map randomization");
148 
149 static int __elfN(pie_aslr_enabled) = 0;
150 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
151     &__elfN(pie_aslr_enabled), 0,
152     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
153     ": enable address map randomization for PIE binaries");
154 
155 static int __elfN(aslr_honor_sbrk) = 1;
156 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
157     &__elfN(aslr_honor_sbrk), 0,
158     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
159 
160 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
161 
162 #define	trunc_page_ps(va, ps)	rounddown2(va, ps)
163 #define	round_page_ps(va, ps)	roundup2(va, ps)
164 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
165 
166 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
167 
168 Elf_Brandnote __elfN(freebsd_brandnote) = {
169 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
170 	.hdr.n_descsz	= sizeof(int32_t),
171 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
172 	.vendor		= FREEBSD_ABI_VENDOR,
173 	.flags		= BN_TRANSLATE_OSREL,
174 	.trans_osrel	= __elfN(freebsd_trans_osrel)
175 };
176 
177 static bool
178 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
179 {
180 	uintptr_t p;
181 
182 	p = (uintptr_t)(note + 1);
183 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
184 	*osrel = *(const int32_t *)(p);
185 
186 	return (true);
187 }
188 
189 static const char GNU_ABI_VENDOR[] = "GNU";
190 static int GNU_KFREEBSD_ABI_DESC = 3;
191 
192 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
193 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
194 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
195 	.hdr.n_type	= 1,
196 	.vendor		= GNU_ABI_VENDOR,
197 	.flags		= BN_TRANSLATE_OSREL,
198 	.trans_osrel	= kfreebsd_trans_osrel
199 };
200 
201 static bool
202 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
203 {
204 	const Elf32_Word *desc;
205 	uintptr_t p;
206 
207 	p = (uintptr_t)(note + 1);
208 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
209 
210 	desc = (const Elf32_Word *)p;
211 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
212 		return (false);
213 
214 	/*
215 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
216 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
217 	 */
218 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
219 
220 	return (true);
221 }
222 
223 int
224 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
225 {
226 	int i;
227 
228 	for (i = 0; i < MAX_BRANDS; i++) {
229 		if (elf_brand_list[i] == NULL) {
230 			elf_brand_list[i] = entry;
231 			break;
232 		}
233 	}
234 	if (i == MAX_BRANDS) {
235 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
236 			__func__, entry);
237 		return (-1);
238 	}
239 	return (0);
240 }
241 
242 int
243 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
244 {
245 	int i;
246 
247 	for (i = 0; i < MAX_BRANDS; i++) {
248 		if (elf_brand_list[i] == entry) {
249 			elf_brand_list[i] = NULL;
250 			break;
251 		}
252 	}
253 	if (i == MAX_BRANDS)
254 		return (-1);
255 	return (0);
256 }
257 
258 int
259 __elfN(brand_inuse)(Elf_Brandinfo *entry)
260 {
261 	struct proc *p;
262 	int rval = FALSE;
263 
264 	sx_slock(&allproc_lock);
265 	FOREACH_PROC_IN_SYSTEM(p) {
266 		if (p->p_sysent == entry->sysvec) {
267 			rval = TRUE;
268 			break;
269 		}
270 	}
271 	sx_sunlock(&allproc_lock);
272 
273 	return (rval);
274 }
275 
276 static Elf_Brandinfo *
277 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
278     int interp_name_len, int32_t *osrel, uint32_t *fctl0)
279 {
280 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
281 	Elf_Brandinfo *bi, *bi_m;
282 	boolean_t ret;
283 	int i;
284 
285 	/*
286 	 * We support four types of branding -- (1) the ELF EI_OSABI field
287 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
288 	 * branding w/in the ELF header, (3) path of the `interp_path'
289 	 * field, and (4) the ".note.ABI-tag" ELF section.
290 	 */
291 
292 	/* Look for an ".note.ABI-tag" ELF section */
293 	bi_m = NULL;
294 	for (i = 0; i < MAX_BRANDS; i++) {
295 		bi = elf_brand_list[i];
296 		if (bi == NULL)
297 			continue;
298 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
299 			continue;
300 		if (hdr->e_machine == bi->machine && (bi->flags &
301 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
302 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
303 			    fctl0);
304 			/* Give brand a chance to veto check_note's guess */
305 			if (ret && bi->header_supported)
306 				ret = bi->header_supported(imgp);
307 			/*
308 			 * If note checker claimed the binary, but the
309 			 * interpreter path in the image does not
310 			 * match default one for the brand, try to
311 			 * search for other brands with the same
312 			 * interpreter.  Either there is better brand
313 			 * with the right interpreter, or, failing
314 			 * this, we return first brand which accepted
315 			 * our note and, optionally, header.
316 			 */
317 			if (ret && bi_m == NULL && interp != NULL &&
318 			    (bi->interp_path == NULL ||
319 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
320 			    strncmp(interp, bi->interp_path, interp_name_len)
321 			    != 0))) {
322 				bi_m = bi;
323 				ret = 0;
324 			}
325 			if (ret)
326 				return (bi);
327 		}
328 	}
329 	if (bi_m != NULL)
330 		return (bi_m);
331 
332 	/* If the executable has a brand, search for it in the brand list. */
333 	for (i = 0; i < MAX_BRANDS; i++) {
334 		bi = elf_brand_list[i];
335 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
336 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
337 			continue;
338 		if (hdr->e_machine == bi->machine &&
339 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
340 		    (bi->compat_3_brand != NULL &&
341 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
342 		    bi->compat_3_brand) == 0))) {
343 			/* Looks good, but give brand a chance to veto */
344 			if (bi->header_supported == NULL ||
345 			    bi->header_supported(imgp)) {
346 				/*
347 				 * Again, prefer strictly matching
348 				 * interpreter path.
349 				 */
350 				if (interp_name_len == 0 &&
351 				    bi->interp_path == NULL)
352 					return (bi);
353 				if (bi->interp_path != NULL &&
354 				    strlen(bi->interp_path) + 1 ==
355 				    interp_name_len && strncmp(interp,
356 				    bi->interp_path, interp_name_len) == 0)
357 					return (bi);
358 				if (bi_m == NULL)
359 					bi_m = bi;
360 			}
361 		}
362 	}
363 	if (bi_m != NULL)
364 		return (bi_m);
365 
366 	/* No known brand, see if the header is recognized by any brand */
367 	for (i = 0; i < MAX_BRANDS; i++) {
368 		bi = elf_brand_list[i];
369 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
370 		    bi->header_supported == NULL)
371 			continue;
372 		if (hdr->e_machine == bi->machine) {
373 			ret = bi->header_supported(imgp);
374 			if (ret)
375 				return (bi);
376 		}
377 	}
378 
379 	/* Lacking a known brand, search for a recognized interpreter. */
380 	if (interp != NULL) {
381 		for (i = 0; i < MAX_BRANDS; i++) {
382 			bi = elf_brand_list[i];
383 			if (bi == NULL || (bi->flags &
384 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
385 			    != 0)
386 				continue;
387 			if (hdr->e_machine == bi->machine &&
388 			    bi->interp_path != NULL &&
389 			    /* ELF image p_filesz includes terminating zero */
390 			    strlen(bi->interp_path) + 1 == interp_name_len &&
391 			    strncmp(interp, bi->interp_path, interp_name_len)
392 			    == 0 && (bi->header_supported == NULL ||
393 			    bi->header_supported(imgp)))
394 				return (bi);
395 		}
396 	}
397 
398 	/* Lacking a recognized interpreter, try the default brand */
399 	for (i = 0; i < MAX_BRANDS; i++) {
400 		bi = elf_brand_list[i];
401 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
402 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
403 			continue;
404 		if (hdr->e_machine == bi->machine &&
405 		    __elfN(fallback_brand) == bi->brand &&
406 		    (bi->header_supported == NULL ||
407 		    bi->header_supported(imgp)))
408 			return (bi);
409 	}
410 	return (NULL);
411 }
412 
413 static int
414 __elfN(check_header)(const Elf_Ehdr *hdr)
415 {
416 	Elf_Brandinfo *bi;
417 	int i;
418 
419 	if (!IS_ELF(*hdr) ||
420 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
421 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
422 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
423 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
424 	    hdr->e_version != ELF_TARG_VER)
425 		return (ENOEXEC);
426 
427 	/*
428 	 * Make sure we have at least one brand for this machine.
429 	 */
430 
431 	for (i = 0; i < MAX_BRANDS; i++) {
432 		bi = elf_brand_list[i];
433 		if (bi != NULL && bi->machine == hdr->e_machine)
434 			break;
435 	}
436 	if (i == MAX_BRANDS)
437 		return (ENOEXEC);
438 
439 	return (0);
440 }
441 
442 static int
443 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
444     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
445 {
446 	struct sf_buf *sf;
447 	int error;
448 	vm_offset_t off;
449 
450 	/*
451 	 * Create the page if it doesn't exist yet. Ignore errors.
452 	 */
453 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
454 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
455 
456 	/*
457 	 * Find the page from the underlying object.
458 	 */
459 	if (object != NULL) {
460 		sf = vm_imgact_map_page(object, offset);
461 		if (sf == NULL)
462 			return (KERN_FAILURE);
463 		off = offset - trunc_page(offset);
464 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
465 		    end - start);
466 		vm_imgact_unmap_page(sf);
467 		if (error != 0)
468 			return (KERN_FAILURE);
469 	}
470 
471 	return (KERN_SUCCESS);
472 }
473 
474 static int
475 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
476     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
477     int cow)
478 {
479 	struct sf_buf *sf;
480 	vm_offset_t off;
481 	vm_size_t sz;
482 	int error, locked, rv;
483 
484 	if (start != trunc_page(start)) {
485 		rv = __elfN(map_partial)(map, object, offset, start,
486 		    round_page(start), prot);
487 		if (rv != KERN_SUCCESS)
488 			return (rv);
489 		offset += round_page(start) - start;
490 		start = round_page(start);
491 	}
492 	if (end != round_page(end)) {
493 		rv = __elfN(map_partial)(map, object, offset +
494 		    trunc_page(end) - start, trunc_page(end), end, prot);
495 		if (rv != KERN_SUCCESS)
496 			return (rv);
497 		end = trunc_page(end);
498 	}
499 	if (start >= end)
500 		return (KERN_SUCCESS);
501 	if ((offset & PAGE_MASK) != 0) {
502 		/*
503 		 * The mapping is not page aligned.  This means that we have
504 		 * to copy the data.
505 		 */
506 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
507 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
508 		if (rv != KERN_SUCCESS)
509 			return (rv);
510 		if (object == NULL)
511 			return (KERN_SUCCESS);
512 		for (; start < end; start += sz) {
513 			sf = vm_imgact_map_page(object, offset);
514 			if (sf == NULL)
515 				return (KERN_FAILURE);
516 			off = offset - trunc_page(offset);
517 			sz = end - start;
518 			if (sz > PAGE_SIZE - off)
519 				sz = PAGE_SIZE - off;
520 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
521 			    (caddr_t)start, sz);
522 			vm_imgact_unmap_page(sf);
523 			if (error != 0)
524 				return (KERN_FAILURE);
525 			offset += sz;
526 		}
527 	} else {
528 		vm_object_reference(object);
529 		rv = vm_map_fixed(map, object, offset, start, end - start,
530 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
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 		}
538 	}
539 	return (KERN_SUCCESS);
540 }
541 
542 static int
543 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
544     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
545     size_t pagesize)
546 {
547 	struct sf_buf *sf;
548 	size_t map_len;
549 	vm_map_t map;
550 	vm_object_t object;
551 	vm_offset_t off, map_addr;
552 	int error, rv, cow;
553 	size_t copy_len;
554 	vm_ooffset_t file_addr;
555 
556 	/*
557 	 * It's necessary to fail if the filsz + offset taken from the
558 	 * header is greater than the actual file pager object's size.
559 	 * If we were to allow this, then the vm_map_find() below would
560 	 * walk right off the end of the file object and into the ether.
561 	 *
562 	 * While I'm here, might as well check for something else that
563 	 * is invalid: filsz cannot be greater than memsz.
564 	 */
565 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
566 	    filsz > memsz) {
567 		uprintf("elf_load_section: truncated ELF file\n");
568 		return (ENOEXEC);
569 	}
570 
571 	object = imgp->object;
572 	map = &imgp->proc->p_vmspace->vm_map;
573 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
574 	file_addr = trunc_page_ps(offset, pagesize);
575 
576 	/*
577 	 * We have two choices.  We can either clear the data in the last page
578 	 * of an oversized mapping, or we can start the anon mapping a page
579 	 * early and copy the initialized data into that first page.  We
580 	 * choose the second.
581 	 */
582 	if (filsz == 0)
583 		map_len = 0;
584 	else if (memsz > filsz)
585 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
586 	else
587 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
588 
589 	if (map_len != 0) {
590 		/* cow flags: don't dump readonly sections in core */
591 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
592 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
593 
594 		rv = __elfN(map_insert)(imgp, map,
595 				      object,
596 				      file_addr,	/* file offset */
597 				      map_addr,		/* virtual start */
598 				      map_addr + map_len,/* virtual end */
599 				      prot,
600 				      cow);
601 		if (rv != KERN_SUCCESS)
602 			return (EINVAL);
603 
604 		/* we can stop now if we've covered it all */
605 		if (memsz == filsz)
606 			return (0);
607 	}
608 
609 
610 	/*
611 	 * We have to get the remaining bit of the file into the first part
612 	 * of the oversized map segment.  This is normally because the .data
613 	 * segment in the file is extended to provide bss.  It's a neat idea
614 	 * to try and save a page, but it's a pain in the behind to implement.
615 	 */
616 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
617 	    filsz, pagesize);
618 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
619 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
620 	    map_addr;
621 
622 	/* This had damn well better be true! */
623 	if (map_len != 0) {
624 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
625 		    map_addr + map_len, prot, 0);
626 		if (rv != KERN_SUCCESS)
627 			return (EINVAL);
628 	}
629 
630 	if (copy_len != 0) {
631 		sf = vm_imgact_map_page(object, offset + filsz);
632 		if (sf == NULL)
633 			return (EIO);
634 
635 		/* send the page fragment to user space */
636 		off = trunc_page_ps(offset + filsz, pagesize) -
637 		    trunc_page(offset + filsz);
638 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
639 		    (caddr_t)map_addr, copy_len);
640 		vm_imgact_unmap_page(sf);
641 		if (error != 0)
642 			return (error);
643 	}
644 
645 	/*
646 	 * Remove write access to the page if it was only granted by map_insert
647 	 * to allow copyout.
648 	 */
649 	if ((prot & VM_PROT_WRITE) == 0)
650 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
651 		    map_len), prot, FALSE);
652 
653 	return (0);
654 }
655 
656 /*
657  * Load the file "file" into memory.  It may be either a shared object
658  * or an executable.
659  *
660  * The "addr" reference parameter is in/out.  On entry, it specifies
661  * the address where a shared object should be loaded.  If the file is
662  * an executable, this value is ignored.  On exit, "addr" specifies
663  * where the file was actually loaded.
664  *
665  * The "entry" reference parameter is out only.  On exit, it specifies
666  * the entry point for the loaded file.
667  */
668 static int
669 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
670 	u_long *entry, size_t pagesize)
671 {
672 	struct {
673 		struct nameidata nd;
674 		struct vattr attr;
675 		struct image_params image_params;
676 	} *tempdata;
677 	const Elf_Ehdr *hdr = NULL;
678 	const Elf_Phdr *phdr = NULL;
679 	struct nameidata *nd;
680 	struct vattr *attr;
681 	struct image_params *imgp;
682 	vm_prot_t prot;
683 	u_long rbase;
684 	u_long base_addr = 0;
685 	int error, i, numsegs;
686 
687 #ifdef CAPABILITY_MODE
688 	/*
689 	 * XXXJA: This check can go away once we are sufficiently confident
690 	 * that the checks in namei() are correct.
691 	 */
692 	if (IN_CAPABILITY_MODE(curthread))
693 		return (ECAPMODE);
694 #endif
695 
696 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
697 	nd = &tempdata->nd;
698 	attr = &tempdata->attr;
699 	imgp = &tempdata->image_params;
700 
701 	/*
702 	 * Initialize part of the common data
703 	 */
704 	imgp->proc = p;
705 	imgp->attr = attr;
706 	imgp->firstpage = NULL;
707 	imgp->image_header = NULL;
708 	imgp->object = NULL;
709 	imgp->execlabel = NULL;
710 
711 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
712 	if ((error = namei(nd)) != 0) {
713 		nd->ni_vp = NULL;
714 		goto fail;
715 	}
716 	NDFREE(nd, NDF_ONLY_PNBUF);
717 	imgp->vp = nd->ni_vp;
718 
719 	/*
720 	 * Check permissions, modes, uid, etc on the file, and "open" it.
721 	 */
722 	error = exec_check_permissions(imgp);
723 	if (error)
724 		goto fail;
725 
726 	error = exec_map_first_page(imgp);
727 	if (error)
728 		goto fail;
729 
730 	/*
731 	 * Also make certain that the interpreter stays the same, so set
732 	 * its VV_TEXT flag, too.
733 	 */
734 	VOP_SET_TEXT(nd->ni_vp);
735 
736 	imgp->object = nd->ni_vp->v_object;
737 
738 	hdr = (const Elf_Ehdr *)imgp->image_header;
739 	if ((error = __elfN(check_header)(hdr)) != 0)
740 		goto fail;
741 	if (hdr->e_type == ET_DYN)
742 		rbase = *addr;
743 	else if (hdr->e_type == ET_EXEC)
744 		rbase = 0;
745 	else {
746 		error = ENOEXEC;
747 		goto fail;
748 	}
749 
750 	/* Only support headers that fit within first page for now      */
751 	if ((hdr->e_phoff > PAGE_SIZE) ||
752 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
753 		error = ENOEXEC;
754 		goto fail;
755 	}
756 
757 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
758 	if (!aligned(phdr, Elf_Addr)) {
759 		error = ENOEXEC;
760 		goto fail;
761 	}
762 
763 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
764 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
765 			/* Loadable segment */
766 			prot = __elfN(trans_prot)(phdr[i].p_flags);
767 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
768 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
769 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
770 			if (error != 0)
771 				goto fail;
772 			/*
773 			 * Establish the base address if this is the
774 			 * first segment.
775 			 */
776 			if (numsegs == 0)
777   				base_addr = trunc_page(phdr[i].p_vaddr +
778 				    rbase);
779 			numsegs++;
780 		}
781 	}
782 	*addr = base_addr;
783 	*entry = (unsigned long)hdr->e_entry + rbase;
784 
785 fail:
786 	if (imgp->firstpage)
787 		exec_unmap_first_page(imgp);
788 
789 	if (nd->ni_vp)
790 		vput(nd->ni_vp);
791 
792 	free(tempdata, M_TEMP);
793 
794 	return (error);
795 }
796 
797 static u_long
798 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
799     u_int align)
800 {
801 	u_long rbase, res;
802 
803 	MPASS(vm_map_min(map) <= minv);
804 	MPASS(maxv <= vm_map_max(map));
805 	MPASS(minv < maxv);
806 	MPASS(minv + align < maxv);
807 	arc4rand(&rbase, sizeof(rbase), 0);
808 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
809 	res &= ~((u_long)align - 1);
810 	if (res >= maxv)
811 		res -= align;
812 	KASSERT(res >= minv,
813 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
814 	    res, minv, maxv, rbase));
815 	KASSERT(res < maxv,
816 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
817 	    res, maxv, minv, rbase));
818 	return (res);
819 }
820 
821 /*
822  * Impossible et_dyn_addr initial value indicating that the real base
823  * must be calculated later with some randomization applied.
824  */
825 #define	ET_DYN_ADDR_RAND	1
826 
827 static int
828 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
829 {
830 	struct thread *td;
831 	const Elf_Ehdr *hdr;
832 	const Elf_Phdr *phdr;
833 	Elf_Auxargs *elf_auxargs;
834 	struct vmspace *vmspace;
835 	vm_map_t map;
836 	const char *err_str, *newinterp;
837 	char *interp, *interp_buf, *path;
838 	Elf_Brandinfo *brand_info;
839 	struct sysentvec *sv;
840 	vm_prot_t prot;
841 	u_long text_size, data_size, total_size, text_addr, data_addr;
842 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
843 	u_long maxalign, mapsz, maxv, maxv1;
844 	uint32_t fctl0;
845 	int32_t osrel;
846 	int error, i, n, interp_name_len, have_interp;
847 
848 	hdr = (const Elf_Ehdr *)imgp->image_header;
849 
850 	/*
851 	 * Do we have a valid ELF header ?
852 	 *
853 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
854 	 * if particular brand doesn't support it.
855 	 */
856 	if (__elfN(check_header)(hdr) != 0 ||
857 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
858 		return (-1);
859 
860 	/*
861 	 * From here on down, we return an errno, not -1, as we've
862 	 * detected an ELF file.
863 	 */
864 
865 	if ((hdr->e_phoff > PAGE_SIZE) ||
866 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
867 		/* Only support headers in first page for now */
868 		uprintf("Program headers not in the first page\n");
869 		return (ENOEXEC);
870 	}
871 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
872 	if (!aligned(phdr, Elf_Addr)) {
873 		uprintf("Unaligned program headers\n");
874 		return (ENOEXEC);
875 	}
876 
877 	n = error = 0;
878 	baddr = 0;
879 	osrel = 0;
880 	fctl0 = 0;
881 	text_size = data_size = total_size = text_addr = data_addr = 0;
882 	entry = proghdr = 0;
883 	interp_name_len = 0;
884 	err_str = newinterp = NULL;
885 	interp = interp_buf = NULL;
886 	td = curthread;
887 	maxalign = PAGE_SIZE;
888 	mapsz = 0;
889 
890 	for (i = 0; i < hdr->e_phnum; i++) {
891 		switch (phdr[i].p_type) {
892 		case PT_LOAD:
893 			if (n == 0)
894 				baddr = phdr[i].p_vaddr;
895 			if (phdr[i].p_align > maxalign)
896 				maxalign = phdr[i].p_align;
897 			mapsz += phdr[i].p_memsz;
898 			n++;
899 			break;
900 		case PT_INTERP:
901 			/* Path to interpreter */
902 			if (phdr[i].p_filesz < 2 ||
903 			    phdr[i].p_filesz > MAXPATHLEN) {
904 				uprintf("Invalid PT_INTERP\n");
905 				error = ENOEXEC;
906 				goto ret;
907 			}
908 			if (interp != NULL) {
909 				uprintf("Multiple PT_INTERP headers\n");
910 				error = ENOEXEC;
911 				goto ret;
912 			}
913 			interp_name_len = phdr[i].p_filesz;
914 			if (phdr[i].p_offset > PAGE_SIZE ||
915 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
916 				VOP_UNLOCK(imgp->vp, 0);
917 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
918 				    M_WAITOK);
919 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
920 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
921 				    interp_name_len, phdr[i].p_offset,
922 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
923 				    NOCRED, NULL, td);
924 				if (error != 0) {
925 					uprintf("i/o error PT_INTERP %d\n",
926 					    error);
927 					goto ret;
928 				}
929 				interp_buf[interp_name_len] = '\0';
930 				interp = interp_buf;
931 			} else {
932 				interp = __DECONST(char *, imgp->image_header) +
933 				    phdr[i].p_offset;
934 				if (interp[interp_name_len - 1] != '\0') {
935 					uprintf("Invalid PT_INTERP\n");
936 					error = ENOEXEC;
937 					goto ret;
938 				}
939 			}
940 			break;
941 		case PT_GNU_STACK:
942 			if (__elfN(nxstack))
943 				imgp->stack_prot =
944 				    __elfN(trans_prot)(phdr[i].p_flags);
945 			imgp->stack_sz = phdr[i].p_memsz;
946 			break;
947 		}
948 	}
949 
950 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
951 	    &osrel, &fctl0);
952 	if (brand_info == NULL) {
953 		uprintf("ELF binary type \"%u\" not known.\n",
954 		    hdr->e_ident[EI_OSABI]);
955 		error = ENOEXEC;
956 		goto ret;
957 	}
958 	sv = brand_info->sysvec;
959 	et_dyn_addr = 0;
960 	if (hdr->e_type == ET_DYN) {
961 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
962 			uprintf("Cannot execute shared object\n");
963 			error = ENOEXEC;
964 			goto ret;
965 		}
966 		/*
967 		 * Honour the base load address from the dso if it is
968 		 * non-zero for some reason.
969 		 */
970 		if (baddr == 0) {
971 			if ((sv->sv_flags & SV_ASLR) == 0 ||
972 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
973 				et_dyn_addr = ET_DYN_LOAD_ADDR;
974 			else if ((__elfN(pie_aslr_enabled) &&
975 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
976 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
977 				et_dyn_addr = ET_DYN_ADDR_RAND;
978 			else
979 				et_dyn_addr = ET_DYN_LOAD_ADDR;
980 		}
981 	}
982 	if (interp != NULL && brand_info->interp_newpath != NULL)
983 		newinterp = brand_info->interp_newpath;
984 
985 	/*
986 	 * Avoid a possible deadlock if the current address space is destroyed
987 	 * and that address space maps the locked vnode.  In the common case,
988 	 * the locked vnode's v_usecount is decremented but remains greater
989 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
990 	 * However, in cases where the vnode lock is external, such as nullfs,
991 	 * v_usecount may become zero.
992 	 *
993 	 * The VV_TEXT flag prevents modifications to the executable while
994 	 * the vnode is unlocked.
995 	 */
996 	VOP_UNLOCK(imgp->vp, 0);
997 
998 	/*
999 	 * Decide whether to enable randomization of user mappings.
1000 	 * First, reset user preferences for the setid binaries.
1001 	 * Then, account for the support of the randomization by the
1002 	 * ABI, by user preferences, and make special treatment for
1003 	 * PIE binaries.
1004 	 */
1005 	if (imgp->credential_setid) {
1006 		PROC_LOCK(imgp->proc);
1007 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1008 		PROC_UNLOCK(imgp->proc);
1009 	}
1010 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1011 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1012 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1013 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1014 		    ("et_dyn_addr == RAND and !ASLR"));
1015 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1016 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1017 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1018 		imgp->map_flags |= MAP_ASLR;
1019 		/*
1020 		 * If user does not care about sbrk, utilize the bss
1021 		 * grow region for mappings as well.  We can select
1022 		 * the base for the image anywere and still not suffer
1023 		 * from the fragmentation.
1024 		 */
1025 		if (!__elfN(aslr_honor_sbrk) ||
1026 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1027 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1028 	}
1029 
1030 	error = exec_new_vmspace(imgp, sv);
1031 	vmspace = imgp->proc->p_vmspace;
1032 	map = &vmspace->vm_map;
1033 
1034 	imgp->proc->p_sysent = sv;
1035 
1036 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1037 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1038 		KASSERT((map->flags & MAP_ASLR) != 0,
1039 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1040 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1041 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1042 		    /* reserve half of the address space to interpreter */
1043 		    maxv / 2, 1UL << flsl(maxalign));
1044 	}
1045 
1046 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1047 	if (error != 0)
1048 		goto ret;
1049 
1050 	for (i = 0; i < hdr->e_phnum; i++) {
1051 		switch (phdr[i].p_type) {
1052 		case PT_LOAD:	/* Loadable segment */
1053 			if (phdr[i].p_memsz == 0)
1054 				break;
1055 			prot = __elfN(trans_prot)(phdr[i].p_flags);
1056 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
1057 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
1058 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
1059 			    sv->sv_pagesize);
1060 			if (error != 0)
1061 				goto ret;
1062 
1063 			/*
1064 			 * If this segment contains the program headers,
1065 			 * remember their virtual address for the AT_PHDR
1066 			 * aux entry. Static binaries don't usually include
1067 			 * a PT_PHDR entry.
1068 			 */
1069 			if (phdr[i].p_offset == 0 &&
1070 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1071 				<= phdr[i].p_filesz)
1072 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
1073 				    et_dyn_addr;
1074 
1075 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
1076 			seg_size = round_page(phdr[i].p_memsz +
1077 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
1078 
1079 			/*
1080 			 * Make the largest executable segment the official
1081 			 * text segment and all others data.
1082 			 *
1083 			 * Note that obreak() assumes that data_addr +
1084 			 * data_size == end of data load area, and the ELF
1085 			 * file format expects segments to be sorted by
1086 			 * address.  If multiple data segments exist, the
1087 			 * last one will be used.
1088 			 */
1089 
1090 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
1091 				text_size = seg_size;
1092 				text_addr = seg_addr;
1093 			} else {
1094 				data_size = seg_size;
1095 				data_addr = seg_addr;
1096 			}
1097 			total_size += seg_size;
1098 			break;
1099 		case PT_PHDR: 	/* Program header table info */
1100 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
1101 			break;
1102 		default:
1103 			break;
1104 		}
1105 	}
1106 
1107 	if (data_addr == 0 && data_size == 0) {
1108 		data_addr = text_addr;
1109 		data_size = text_size;
1110 	}
1111 
1112 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1113 
1114 	/*
1115 	 * Check limits.  It should be safe to check the
1116 	 * limits after loading the segments since we do
1117 	 * not actually fault in all the segments pages.
1118 	 */
1119 	PROC_LOCK(imgp->proc);
1120 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1121 		err_str = "Data segment size exceeds process limit";
1122 	else if (text_size > maxtsiz)
1123 		err_str = "Text segment size exceeds system limit";
1124 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1125 		err_str = "Total segment size exceeds process limit";
1126 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1127 		err_str = "Data segment size exceeds resource limit";
1128 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1129 		err_str = "Total segment size exceeds resource limit";
1130 	if (err_str != NULL) {
1131 		PROC_UNLOCK(imgp->proc);
1132 		uprintf("%s\n", err_str);
1133 		error = ENOMEM;
1134 		goto ret;
1135 	}
1136 
1137 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1138 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1139 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1140 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1141 
1142 	/*
1143 	 * We load the dynamic linker where a userland call
1144 	 * to mmap(0, ...) would put it.  The rationale behind this
1145 	 * calculation is that it leaves room for the heap to grow to
1146 	 * its maximum allowed size.
1147 	 */
1148 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1149 	    RLIMIT_DATA));
1150 	if ((map->flags & MAP_ASLR) != 0) {
1151 		maxv1 = maxv / 2 + addr / 2;
1152 		MPASS(maxv1 >= addr);	/* No overflow */
1153 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1154 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1155 	} else {
1156 		map->anon_loc = addr;
1157 	}
1158 	PROC_UNLOCK(imgp->proc);
1159 
1160 	imgp->entry_addr = entry;
1161 
1162 	if (interp != NULL) {
1163 		have_interp = FALSE;
1164 		VOP_UNLOCK(imgp->vp, 0);
1165 		if ((map->flags & MAP_ASLR) != 0) {
1166 			/* Assume that interpeter fits into 1/4 of AS */
1167 			maxv1 = maxv / 2 + addr / 2;
1168 			MPASS(maxv1 >= addr);	/* No overflow */
1169 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1170 			    maxv1, PAGE_SIZE);
1171 		}
1172 		if (brand_info->emul_path != NULL &&
1173 		    brand_info->emul_path[0] != '\0') {
1174 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1175 			snprintf(path, MAXPATHLEN, "%s%s",
1176 			    brand_info->emul_path, interp);
1177 			error = __elfN(load_file)(imgp->proc, path, &addr,
1178 			    &imgp->entry_addr, sv->sv_pagesize);
1179 			free(path, M_TEMP);
1180 			if (error == 0)
1181 				have_interp = TRUE;
1182 		}
1183 		if (!have_interp && newinterp != NULL &&
1184 		    (brand_info->interp_path == NULL ||
1185 		    strcmp(interp, brand_info->interp_path) == 0)) {
1186 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1187 			    &imgp->entry_addr, sv->sv_pagesize);
1188 			if (error == 0)
1189 				have_interp = TRUE;
1190 		}
1191 		if (!have_interp) {
1192 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1193 			    &imgp->entry_addr, sv->sv_pagesize);
1194 		}
1195 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1196 		if (error != 0) {
1197 			uprintf("ELF interpreter %s not found, error %d\n",
1198 			    interp, error);
1199 			goto ret;
1200 		}
1201 	} else
1202 		addr = et_dyn_addr;
1203 
1204 	/*
1205 	 * Construct auxargs table (used by the fixup routine)
1206 	 */
1207 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1208 	elf_auxargs->execfd = -1;
1209 	elf_auxargs->phdr = proghdr;
1210 	elf_auxargs->phent = hdr->e_phentsize;
1211 	elf_auxargs->phnum = hdr->e_phnum;
1212 	elf_auxargs->pagesz = PAGE_SIZE;
1213 	elf_auxargs->base = addr;
1214 	elf_auxargs->flags = 0;
1215 	elf_auxargs->entry = entry;
1216 	elf_auxargs->hdr_eflags = hdr->e_flags;
1217 
1218 	imgp->auxargs = elf_auxargs;
1219 	imgp->interpreted = 0;
1220 	imgp->reloc_base = addr;
1221 	imgp->proc->p_osrel = osrel;
1222 	imgp->proc->p_fctl0 = fctl0;
1223 	imgp->proc->p_elf_machine = hdr->e_machine;
1224 	imgp->proc->p_elf_flags = hdr->e_flags;
1225 
1226 ret:
1227 	free(interp_buf, M_TEMP);
1228 	return (error);
1229 }
1230 
1231 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1232 
1233 int
1234 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1235 {
1236 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1237 	Elf_Auxinfo *argarray, *pos;
1238 	Elf_Addr *base, *auxbase;
1239 	int error;
1240 
1241 	base = (Elf_Addr *)*stack_base;
1242 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1243 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1244 	    M_WAITOK | M_ZERO);
1245 
1246 	if (args->execfd != -1)
1247 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1248 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1249 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1250 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1251 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1252 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1253 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1254 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1255 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1256 	if (imgp->execpathp != 0)
1257 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1258 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1259 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1260 	if (imgp->canary != 0) {
1261 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1262 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1263 	}
1264 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1265 	if (imgp->pagesizes != 0) {
1266 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1267 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1268 	}
1269 	if (imgp->sysent->sv_timekeep_base != 0) {
1270 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1271 		    imgp->sysent->sv_timekeep_base);
1272 	}
1273 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1274 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1275 	    imgp->sysent->sv_stackprot);
1276 	if (imgp->sysent->sv_hwcap != NULL)
1277 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1278 	if (imgp->sysent->sv_hwcap2 != NULL)
1279 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1280 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1281 
1282 	free(imgp->auxargs, M_TEMP);
1283 	imgp->auxargs = NULL;
1284 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1285 
1286 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1287 	free(argarray, M_TEMP);
1288 	if (error != 0)
1289 		return (error);
1290 
1291 	base--;
1292 	if (suword(base, imgp->args->argc) == -1)
1293 		return (EFAULT);
1294 	*stack_base = (register_t *)base;
1295 	return (0);
1296 }
1297 
1298 /*
1299  * Code for generating ELF core dumps.
1300  */
1301 
1302 typedef void (*segment_callback)(vm_map_entry_t, void *);
1303 
1304 /* Closure for cb_put_phdr(). */
1305 struct phdr_closure {
1306 	Elf_Phdr *phdr;		/* Program header to fill in */
1307 	Elf_Off offset;		/* Offset of segment in core file */
1308 };
1309 
1310 /* Closure for cb_size_segment(). */
1311 struct sseg_closure {
1312 	int count;		/* Count of writable segments. */
1313 	size_t size;		/* Total size of all writable segments. */
1314 };
1315 
1316 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1317 
1318 struct note_info {
1319 	int		type;		/* Note type. */
1320 	outfunc_t 	outfunc; 	/* Output function. */
1321 	void		*outarg;	/* Argument for the output function. */
1322 	size_t		outsize;	/* Output size. */
1323 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1324 };
1325 
1326 TAILQ_HEAD(note_info_list, note_info);
1327 
1328 /* Coredump output parameters. */
1329 struct coredump_params {
1330 	off_t		offset;
1331 	struct ucred	*active_cred;
1332 	struct ucred	*file_cred;
1333 	struct thread	*td;
1334 	struct vnode	*vp;
1335 	struct compressor *comp;
1336 };
1337 
1338 extern int compress_user_cores;
1339 extern int compress_user_cores_level;
1340 
1341 static void cb_put_phdr(vm_map_entry_t, void *);
1342 static void cb_size_segment(vm_map_entry_t, void *);
1343 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1344     enum uio_seg);
1345 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1346 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1347     struct note_info_list *, size_t);
1348 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1349     size_t *);
1350 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1351 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1352 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1353 static int sbuf_drain_core_output(void *, const char *, int);
1354 static int sbuf_drain_count(void *arg, const char *data, int len);
1355 
1356 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1357 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1358 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1359 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1360 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1361 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1362 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1363 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1364 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1365 static void note_procstat_files(void *, struct sbuf *, size_t *);
1366 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1367 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1368 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1369 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1370 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1371 
1372 /*
1373  * Write out a core segment to the compression stream.
1374  */
1375 static int
1376 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1377 {
1378 	u_int chunk_len;
1379 	int error;
1380 
1381 	while (len > 0) {
1382 		chunk_len = MIN(len, CORE_BUF_SIZE);
1383 
1384 		/*
1385 		 * We can get EFAULT error here.
1386 		 * In that case zero out the current chunk of the segment.
1387 		 */
1388 		error = copyin(base, buf, chunk_len);
1389 		if (error != 0)
1390 			bzero(buf, chunk_len);
1391 		error = compressor_write(p->comp, buf, chunk_len);
1392 		if (error != 0)
1393 			break;
1394 		base += chunk_len;
1395 		len -= chunk_len;
1396 	}
1397 	return (error);
1398 }
1399 
1400 static int
1401 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1402 {
1403 
1404 	return (core_write((struct coredump_params *)arg, base, len, offset,
1405 	    UIO_SYSSPACE));
1406 }
1407 
1408 static int
1409 core_write(struct coredump_params *p, const void *base, size_t len,
1410     off_t offset, enum uio_seg seg)
1411 {
1412 
1413 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1414 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1415 	    p->active_cred, p->file_cred, NULL, p->td));
1416 }
1417 
1418 static int
1419 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1420     void *tmpbuf)
1421 {
1422 	int error;
1423 
1424 	if (p->comp != NULL)
1425 		return (compress_chunk(p, base, tmpbuf, len));
1426 
1427 	/*
1428 	 * EFAULT is a non-fatal error that we can get, for example,
1429 	 * if the segment is backed by a file but extends beyond its
1430 	 * end.
1431 	 */
1432 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1433 	if (error == EFAULT) {
1434 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1435 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1436 		    "for process %s\n", base, len, offset, curproc->p_comm);
1437 
1438 		/*
1439 		 * Write a "real" zero byte at the end of the target region
1440 		 * in the case this is the last segment.
1441 		 * The intermediate space will be implicitly zero-filled.
1442 		 */
1443 		error = core_write(p, zero_region, 1, offset + len - 1,
1444 		    UIO_SYSSPACE);
1445 	}
1446 	return (error);
1447 }
1448 
1449 /*
1450  * Drain into a core file.
1451  */
1452 static int
1453 sbuf_drain_core_output(void *arg, const char *data, int len)
1454 {
1455 	struct coredump_params *p;
1456 	int error, locked;
1457 
1458 	p = (struct coredump_params *)arg;
1459 
1460 	/*
1461 	 * Some kern_proc out routines that print to this sbuf may
1462 	 * call us with the process lock held. Draining with the
1463 	 * non-sleepable lock held is unsafe. The lock is needed for
1464 	 * those routines when dumping a live process. In our case we
1465 	 * can safely release the lock before draining and acquire
1466 	 * again after.
1467 	 */
1468 	locked = PROC_LOCKED(p->td->td_proc);
1469 	if (locked)
1470 		PROC_UNLOCK(p->td->td_proc);
1471 	if (p->comp != NULL)
1472 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1473 	else
1474 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1475 		    UIO_SYSSPACE);
1476 	if (locked)
1477 		PROC_LOCK(p->td->td_proc);
1478 	if (error != 0)
1479 		return (-error);
1480 	p->offset += len;
1481 	return (len);
1482 }
1483 
1484 /*
1485  * Drain into a counter.
1486  */
1487 static int
1488 sbuf_drain_count(void *arg, const char *data __unused, int len)
1489 {
1490 	size_t *sizep;
1491 
1492 	sizep = (size_t *)arg;
1493 	*sizep += len;
1494 	return (len);
1495 }
1496 
1497 int
1498 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1499 {
1500 	struct ucred *cred = td->td_ucred;
1501 	int error = 0;
1502 	struct sseg_closure seginfo;
1503 	struct note_info_list notelst;
1504 	struct coredump_params params;
1505 	struct note_info *ninfo;
1506 	void *hdr, *tmpbuf;
1507 	size_t hdrsize, notesz, coresize;
1508 
1509 	hdr = NULL;
1510 	tmpbuf = NULL;
1511 	TAILQ_INIT(&notelst);
1512 
1513 	/* Size the program segments. */
1514 	seginfo.count = 0;
1515 	seginfo.size = 0;
1516 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1517 
1518 	/*
1519 	 * Collect info about the core file header area.
1520 	 */
1521 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1522 	if (seginfo.count + 1 >= PN_XNUM)
1523 		hdrsize += sizeof(Elf_Shdr);
1524 	__elfN(prepare_notes)(td, &notelst, &notesz);
1525 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1526 
1527 	/* Set up core dump parameters. */
1528 	params.offset = 0;
1529 	params.active_cred = cred;
1530 	params.file_cred = NOCRED;
1531 	params.td = td;
1532 	params.vp = vp;
1533 	params.comp = NULL;
1534 
1535 #ifdef RACCT
1536 	if (racct_enable) {
1537 		PROC_LOCK(td->td_proc);
1538 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1539 		PROC_UNLOCK(td->td_proc);
1540 		if (error != 0) {
1541 			error = EFAULT;
1542 			goto done;
1543 		}
1544 	}
1545 #endif
1546 	if (coresize >= limit) {
1547 		error = EFAULT;
1548 		goto done;
1549 	}
1550 
1551 	/* Create a compression stream if necessary. */
1552 	if (compress_user_cores != 0) {
1553 		params.comp = compressor_init(core_compressed_write,
1554 		    compress_user_cores, CORE_BUF_SIZE,
1555 		    compress_user_cores_level, &params);
1556 		if (params.comp == NULL) {
1557 			error = EFAULT;
1558 			goto done;
1559 		}
1560 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1561         }
1562 
1563 	/*
1564 	 * Allocate memory for building the header, fill it up,
1565 	 * and write it out following the notes.
1566 	 */
1567 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1568 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1569 	    notesz);
1570 
1571 	/* Write the contents of all of the writable segments. */
1572 	if (error == 0) {
1573 		Elf_Phdr *php;
1574 		off_t offset;
1575 		int i;
1576 
1577 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1578 		offset = round_page(hdrsize + notesz);
1579 		for (i = 0; i < seginfo.count; i++) {
1580 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1581 			    php->p_filesz, offset, &params, tmpbuf);
1582 			if (error != 0)
1583 				break;
1584 			offset += php->p_filesz;
1585 			php++;
1586 		}
1587 		if (error == 0 && params.comp != NULL)
1588 			error = compressor_flush(params.comp);
1589 	}
1590 	if (error) {
1591 		log(LOG_WARNING,
1592 		    "Failed to write core file for process %s (error %d)\n",
1593 		    curproc->p_comm, error);
1594 	}
1595 
1596 done:
1597 	free(tmpbuf, M_TEMP);
1598 	if (params.comp != NULL)
1599 		compressor_fini(params.comp);
1600 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1601 		TAILQ_REMOVE(&notelst, ninfo, link);
1602 		free(ninfo, M_TEMP);
1603 	}
1604 	if (hdr != NULL)
1605 		free(hdr, M_TEMP);
1606 
1607 	return (error);
1608 }
1609 
1610 /*
1611  * A callback for each_dumpable_segment() to write out the segment's
1612  * program header entry.
1613  */
1614 static void
1615 cb_put_phdr(vm_map_entry_t entry, void *closure)
1616 {
1617 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1618 	Elf_Phdr *phdr = phc->phdr;
1619 
1620 	phc->offset = round_page(phc->offset);
1621 
1622 	phdr->p_type = PT_LOAD;
1623 	phdr->p_offset = phc->offset;
1624 	phdr->p_vaddr = entry->start;
1625 	phdr->p_paddr = 0;
1626 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1627 	phdr->p_align = PAGE_SIZE;
1628 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1629 
1630 	phc->offset += phdr->p_filesz;
1631 	phc->phdr++;
1632 }
1633 
1634 /*
1635  * A callback for each_dumpable_segment() to gather information about
1636  * the number of segments and their total size.
1637  */
1638 static void
1639 cb_size_segment(vm_map_entry_t entry, void *closure)
1640 {
1641 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1642 
1643 	ssc->count++;
1644 	ssc->size += entry->end - entry->start;
1645 }
1646 
1647 /*
1648  * For each writable segment in the process's memory map, call the given
1649  * function with a pointer to the map entry and some arbitrary
1650  * caller-supplied data.
1651  */
1652 static void
1653 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1654 {
1655 	struct proc *p = td->td_proc;
1656 	vm_map_t map = &p->p_vmspace->vm_map;
1657 	vm_map_entry_t entry;
1658 	vm_object_t backing_object, object;
1659 	boolean_t ignore_entry;
1660 
1661 	vm_map_lock_read(map);
1662 	for (entry = map->header.next; entry != &map->header;
1663 	    entry = entry->next) {
1664 		/*
1665 		 * Don't dump inaccessible mappings, deal with legacy
1666 		 * coredump mode.
1667 		 *
1668 		 * Note that read-only segments related to the elf binary
1669 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1670 		 * need to arbitrarily ignore such segments.
1671 		 */
1672 		if (elf_legacy_coredump) {
1673 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1674 				continue;
1675 		} else {
1676 			if ((entry->protection & VM_PROT_ALL) == 0)
1677 				continue;
1678 		}
1679 
1680 		/*
1681 		 * Dont include memory segment in the coredump if
1682 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1683 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1684 		 * kernel map).
1685 		 */
1686 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1687 			continue;
1688 
1689 		if ((object = entry->object.vm_object) == NULL)
1690 			continue;
1691 
1692 		/* Ignore memory-mapped devices and such things. */
1693 		VM_OBJECT_RLOCK(object);
1694 		while ((backing_object = object->backing_object) != NULL) {
1695 			VM_OBJECT_RLOCK(backing_object);
1696 			VM_OBJECT_RUNLOCK(object);
1697 			object = backing_object;
1698 		}
1699 		ignore_entry = object->type != OBJT_DEFAULT &&
1700 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1701 		    object->type != OBJT_PHYS;
1702 		VM_OBJECT_RUNLOCK(object);
1703 		if (ignore_entry)
1704 			continue;
1705 
1706 		(*func)(entry, closure);
1707 	}
1708 	vm_map_unlock_read(map);
1709 }
1710 
1711 /*
1712  * Write the core file header to the file, including padding up to
1713  * the page boundary.
1714  */
1715 static int
1716 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1717     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1718 {
1719 	struct note_info *ninfo;
1720 	struct sbuf *sb;
1721 	int error;
1722 
1723 	/* Fill in the header. */
1724 	bzero(hdr, hdrsize);
1725 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1726 
1727 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1728 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1729 	sbuf_start_section(sb, NULL);
1730 	sbuf_bcat(sb, hdr, hdrsize);
1731 	TAILQ_FOREACH(ninfo, notelst, link)
1732 	    __elfN(putnote)(ninfo, sb);
1733 	/* Align up to a page boundary for the program segments. */
1734 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1735 	error = sbuf_finish(sb);
1736 	sbuf_delete(sb);
1737 
1738 	return (error);
1739 }
1740 
1741 static void
1742 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1743     size_t *sizep)
1744 {
1745 	struct proc *p;
1746 	struct thread *thr;
1747 	size_t size;
1748 
1749 	p = td->td_proc;
1750 	size = 0;
1751 
1752 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1753 
1754 	/*
1755 	 * To have the debugger select the right thread (LWP) as the initial
1756 	 * thread, we dump the state of the thread passed to us in td first.
1757 	 * This is the thread that causes the core dump and thus likely to
1758 	 * be the right thread one wants to have selected in the debugger.
1759 	 */
1760 	thr = td;
1761 	while (thr != NULL) {
1762 		size += register_note(list, NT_PRSTATUS,
1763 		    __elfN(note_prstatus), thr);
1764 		size += register_note(list, NT_FPREGSET,
1765 		    __elfN(note_fpregset), thr);
1766 		size += register_note(list, NT_THRMISC,
1767 		    __elfN(note_thrmisc), thr);
1768 		size += register_note(list, NT_PTLWPINFO,
1769 		    __elfN(note_ptlwpinfo), thr);
1770 		size += register_note(list, -1,
1771 		    __elfN(note_threadmd), thr);
1772 
1773 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1774 		    TAILQ_NEXT(thr, td_plist);
1775 		if (thr == td)
1776 			thr = TAILQ_NEXT(thr, td_plist);
1777 	}
1778 
1779 	size += register_note(list, NT_PROCSTAT_PROC,
1780 	    __elfN(note_procstat_proc), p);
1781 	size += register_note(list, NT_PROCSTAT_FILES,
1782 	    note_procstat_files, p);
1783 	size += register_note(list, NT_PROCSTAT_VMMAP,
1784 	    note_procstat_vmmap, p);
1785 	size += register_note(list, NT_PROCSTAT_GROUPS,
1786 	    note_procstat_groups, p);
1787 	size += register_note(list, NT_PROCSTAT_UMASK,
1788 	    note_procstat_umask, p);
1789 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1790 	    note_procstat_rlimit, p);
1791 	size += register_note(list, NT_PROCSTAT_OSREL,
1792 	    note_procstat_osrel, p);
1793 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1794 	    __elfN(note_procstat_psstrings), p);
1795 	size += register_note(list, NT_PROCSTAT_AUXV,
1796 	    __elfN(note_procstat_auxv), p);
1797 
1798 	*sizep = size;
1799 }
1800 
1801 static void
1802 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1803     size_t notesz)
1804 {
1805 	Elf_Ehdr *ehdr;
1806 	Elf_Phdr *phdr;
1807 	Elf_Shdr *shdr;
1808 	struct phdr_closure phc;
1809 
1810 	ehdr = (Elf_Ehdr *)hdr;
1811 
1812 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1813 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1814 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1815 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1816 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1817 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1818 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1819 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1820 	ehdr->e_ident[EI_ABIVERSION] = 0;
1821 	ehdr->e_ident[EI_PAD] = 0;
1822 	ehdr->e_type = ET_CORE;
1823 	ehdr->e_machine = td->td_proc->p_elf_machine;
1824 	ehdr->e_version = EV_CURRENT;
1825 	ehdr->e_entry = 0;
1826 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1827 	ehdr->e_flags = td->td_proc->p_elf_flags;
1828 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1829 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1830 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1831 	ehdr->e_shstrndx = SHN_UNDEF;
1832 	if (numsegs + 1 < PN_XNUM) {
1833 		ehdr->e_phnum = numsegs + 1;
1834 		ehdr->e_shnum = 0;
1835 	} else {
1836 		ehdr->e_phnum = PN_XNUM;
1837 		ehdr->e_shnum = 1;
1838 
1839 		ehdr->e_shoff = ehdr->e_phoff +
1840 		    (numsegs + 1) * ehdr->e_phentsize;
1841 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1842 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1843 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1844 
1845 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1846 		memset(shdr, 0, sizeof(*shdr));
1847 		/*
1848 		 * A special first section is used to hold large segment and
1849 		 * section counts.  This was proposed by Sun Microsystems in
1850 		 * Solaris and has been adopted by Linux; the standard ELF
1851 		 * tools are already familiar with the technique.
1852 		 *
1853 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1854 		 * (or 12-7 depending on the version of the document) for more
1855 		 * details.
1856 		 */
1857 		shdr->sh_type = SHT_NULL;
1858 		shdr->sh_size = ehdr->e_shnum;
1859 		shdr->sh_link = ehdr->e_shstrndx;
1860 		shdr->sh_info = numsegs + 1;
1861 	}
1862 
1863 	/*
1864 	 * Fill in the program header entries.
1865 	 */
1866 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1867 
1868 	/* The note segement. */
1869 	phdr->p_type = PT_NOTE;
1870 	phdr->p_offset = hdrsize;
1871 	phdr->p_vaddr = 0;
1872 	phdr->p_paddr = 0;
1873 	phdr->p_filesz = notesz;
1874 	phdr->p_memsz = 0;
1875 	phdr->p_flags = PF_R;
1876 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1877 	phdr++;
1878 
1879 	/* All the writable segments from the program. */
1880 	phc.phdr = phdr;
1881 	phc.offset = round_page(hdrsize + notesz);
1882 	each_dumpable_segment(td, cb_put_phdr, &phc);
1883 }
1884 
1885 static size_t
1886 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1887 {
1888 	struct note_info *ninfo;
1889 	size_t size, notesize;
1890 
1891 	size = 0;
1892 	out(arg, NULL, &size);
1893 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1894 	ninfo->type = type;
1895 	ninfo->outfunc = out;
1896 	ninfo->outarg = arg;
1897 	ninfo->outsize = size;
1898 	TAILQ_INSERT_TAIL(list, ninfo, link);
1899 
1900 	if (type == -1)
1901 		return (size);
1902 
1903 	notesize = sizeof(Elf_Note) +		/* note header */
1904 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1905 						/* note name */
1906 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1907 
1908 	return (notesize);
1909 }
1910 
1911 static size_t
1912 append_note_data(const void *src, void *dst, size_t len)
1913 {
1914 	size_t padded_len;
1915 
1916 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1917 	if (dst != NULL) {
1918 		bcopy(src, dst, len);
1919 		bzero((char *)dst + len, padded_len - len);
1920 	}
1921 	return (padded_len);
1922 }
1923 
1924 size_t
1925 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1926 {
1927 	Elf_Note *note;
1928 	char *buf;
1929 	size_t notesize;
1930 
1931 	buf = dst;
1932 	if (buf != NULL) {
1933 		note = (Elf_Note *)buf;
1934 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1935 		note->n_descsz = size;
1936 		note->n_type = type;
1937 		buf += sizeof(*note);
1938 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1939 		    sizeof(FREEBSD_ABI_VENDOR));
1940 		append_note_data(src, buf, size);
1941 		if (descp != NULL)
1942 			*descp = buf;
1943 	}
1944 
1945 	notesize = sizeof(Elf_Note) +		/* note header */
1946 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1947 						/* note name */
1948 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1949 
1950 	return (notesize);
1951 }
1952 
1953 static void
1954 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1955 {
1956 	Elf_Note note;
1957 	ssize_t old_len, sect_len;
1958 	size_t new_len, descsz, i;
1959 
1960 	if (ninfo->type == -1) {
1961 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1962 		return;
1963 	}
1964 
1965 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1966 	note.n_descsz = ninfo->outsize;
1967 	note.n_type = ninfo->type;
1968 
1969 	sbuf_bcat(sb, &note, sizeof(note));
1970 	sbuf_start_section(sb, &old_len);
1971 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1972 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1973 	if (note.n_descsz == 0)
1974 		return;
1975 	sbuf_start_section(sb, &old_len);
1976 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1977 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1978 	if (sect_len < 0)
1979 		return;
1980 
1981 	new_len = (size_t)sect_len;
1982 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1983 	if (new_len < descsz) {
1984 		/*
1985 		 * It is expected that individual note emitters will correctly
1986 		 * predict their expected output size and fill up to that size
1987 		 * themselves, padding in a format-specific way if needed.
1988 		 * However, in case they don't, just do it here with zeros.
1989 		 */
1990 		for (i = 0; i < descsz - new_len; i++)
1991 			sbuf_putc(sb, 0);
1992 	} else if (new_len > descsz) {
1993 		/*
1994 		 * We can't always truncate sb -- we may have drained some
1995 		 * of it already.
1996 		 */
1997 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1998 		    "read it (%zu > %zu).  Since it is longer than "
1999 		    "expected, this coredump's notes are corrupt.  THIS "
2000 		    "IS A BUG in the note_procstat routine for type %u.\n",
2001 		    __func__, (unsigned)note.n_type, new_len, descsz,
2002 		    (unsigned)note.n_type));
2003 	}
2004 }
2005 
2006 /*
2007  * Miscellaneous note out functions.
2008  */
2009 
2010 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2011 #include <compat/freebsd32/freebsd32.h>
2012 #include <compat/freebsd32/freebsd32_signal.h>
2013 
2014 typedef struct prstatus32 elf_prstatus_t;
2015 typedef struct prpsinfo32 elf_prpsinfo_t;
2016 typedef struct fpreg32 elf_prfpregset_t;
2017 typedef struct fpreg32 elf_fpregset_t;
2018 typedef struct reg32 elf_gregset_t;
2019 typedef struct thrmisc32 elf_thrmisc_t;
2020 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2021 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2022 typedef uint32_t elf_ps_strings_t;
2023 #else
2024 typedef prstatus_t elf_prstatus_t;
2025 typedef prpsinfo_t elf_prpsinfo_t;
2026 typedef prfpregset_t elf_prfpregset_t;
2027 typedef prfpregset_t elf_fpregset_t;
2028 typedef gregset_t elf_gregset_t;
2029 typedef thrmisc_t elf_thrmisc_t;
2030 #define ELF_KERN_PROC_MASK	0
2031 typedef struct kinfo_proc elf_kinfo_proc_t;
2032 typedef vm_offset_t elf_ps_strings_t;
2033 #endif
2034 
2035 static void
2036 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2037 {
2038 	struct sbuf sbarg;
2039 	size_t len;
2040 	char *cp, *end;
2041 	struct proc *p;
2042 	elf_prpsinfo_t *psinfo;
2043 	int error;
2044 
2045 	p = (struct proc *)arg;
2046 	if (sb != NULL) {
2047 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2048 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2049 		psinfo->pr_version = PRPSINFO_VERSION;
2050 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2051 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2052 		PROC_LOCK(p);
2053 		if (p->p_args != NULL) {
2054 			len = sizeof(psinfo->pr_psargs) - 1;
2055 			if (len > p->p_args->ar_length)
2056 				len = p->p_args->ar_length;
2057 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2058 			PROC_UNLOCK(p);
2059 			error = 0;
2060 		} else {
2061 			_PHOLD(p);
2062 			PROC_UNLOCK(p);
2063 			sbuf_new(&sbarg, psinfo->pr_psargs,
2064 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2065 			error = proc_getargv(curthread, p, &sbarg);
2066 			PRELE(p);
2067 			if (sbuf_finish(&sbarg) == 0)
2068 				len = sbuf_len(&sbarg) - 1;
2069 			else
2070 				len = sizeof(psinfo->pr_psargs) - 1;
2071 			sbuf_delete(&sbarg);
2072 		}
2073 		if (error || len == 0)
2074 			strlcpy(psinfo->pr_psargs, p->p_comm,
2075 			    sizeof(psinfo->pr_psargs));
2076 		else {
2077 			KASSERT(len < sizeof(psinfo->pr_psargs),
2078 			    ("len is too long: %zu vs %zu", len,
2079 			    sizeof(psinfo->pr_psargs)));
2080 			cp = psinfo->pr_psargs;
2081 			end = cp + len - 1;
2082 			for (;;) {
2083 				cp = memchr(cp, '\0', end - cp);
2084 				if (cp == NULL)
2085 					break;
2086 				*cp = ' ';
2087 			}
2088 		}
2089 		psinfo->pr_pid = p->p_pid;
2090 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2091 		free(psinfo, M_TEMP);
2092 	}
2093 	*sizep = sizeof(*psinfo);
2094 }
2095 
2096 static void
2097 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2098 {
2099 	struct thread *td;
2100 	elf_prstatus_t *status;
2101 
2102 	td = (struct thread *)arg;
2103 	if (sb != NULL) {
2104 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2105 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2106 		status->pr_version = PRSTATUS_VERSION;
2107 		status->pr_statussz = sizeof(elf_prstatus_t);
2108 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2109 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2110 		status->pr_osreldate = osreldate;
2111 		status->pr_cursig = td->td_proc->p_sig;
2112 		status->pr_pid = td->td_tid;
2113 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2114 		fill_regs32(td, &status->pr_reg);
2115 #else
2116 		fill_regs(td, &status->pr_reg);
2117 #endif
2118 		sbuf_bcat(sb, status, sizeof(*status));
2119 		free(status, M_TEMP);
2120 	}
2121 	*sizep = sizeof(*status);
2122 }
2123 
2124 static void
2125 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2126 {
2127 	struct thread *td;
2128 	elf_prfpregset_t *fpregset;
2129 
2130 	td = (struct thread *)arg;
2131 	if (sb != NULL) {
2132 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2133 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2134 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2135 		fill_fpregs32(td, fpregset);
2136 #else
2137 		fill_fpregs(td, fpregset);
2138 #endif
2139 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2140 		free(fpregset, M_TEMP);
2141 	}
2142 	*sizep = sizeof(*fpregset);
2143 }
2144 
2145 static void
2146 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2147 {
2148 	struct thread *td;
2149 	elf_thrmisc_t thrmisc;
2150 
2151 	td = (struct thread *)arg;
2152 	if (sb != NULL) {
2153 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2154 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2155 		strcpy(thrmisc.pr_tname, td->td_name);
2156 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2157 	}
2158 	*sizep = sizeof(thrmisc);
2159 }
2160 
2161 static void
2162 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2163 {
2164 	struct thread *td;
2165 	size_t size;
2166 	int structsize;
2167 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2168 	struct ptrace_lwpinfo32 pl;
2169 #else
2170 	struct ptrace_lwpinfo pl;
2171 #endif
2172 
2173 	td = (struct thread *)arg;
2174 	size = sizeof(structsize) + sizeof(pl);
2175 	if (sb != NULL) {
2176 		KASSERT(*sizep == size, ("invalid size"));
2177 		structsize = sizeof(pl);
2178 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2179 		bzero(&pl, sizeof(pl));
2180 		pl.pl_lwpid = td->td_tid;
2181 		pl.pl_event = PL_EVENT_NONE;
2182 		pl.pl_sigmask = td->td_sigmask;
2183 		pl.pl_siglist = td->td_siglist;
2184 		if (td->td_si.si_signo != 0) {
2185 			pl.pl_event = PL_EVENT_SIGNAL;
2186 			pl.pl_flags |= PL_FLAG_SI;
2187 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2188 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2189 #else
2190 			pl.pl_siginfo = td->td_si;
2191 #endif
2192 		}
2193 		strcpy(pl.pl_tdname, td->td_name);
2194 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2195 		sbuf_bcat(sb, &pl, sizeof(pl));
2196 	}
2197 	*sizep = size;
2198 }
2199 
2200 /*
2201  * Allow for MD specific notes, as well as any MD
2202  * specific preparations for writing MI notes.
2203  */
2204 static void
2205 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2206 {
2207 	struct thread *td;
2208 	void *buf;
2209 	size_t size;
2210 
2211 	td = (struct thread *)arg;
2212 	size = *sizep;
2213 	if (size != 0 && sb != NULL)
2214 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2215 	else
2216 		buf = NULL;
2217 	size = 0;
2218 	__elfN(dump_thread)(td, buf, &size);
2219 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2220 	if (size != 0 && sb != NULL)
2221 		sbuf_bcat(sb, buf, size);
2222 	free(buf, M_TEMP);
2223 	*sizep = size;
2224 }
2225 
2226 #ifdef KINFO_PROC_SIZE
2227 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2228 #endif
2229 
2230 static void
2231 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2232 {
2233 	struct proc *p;
2234 	size_t size;
2235 	int structsize;
2236 
2237 	p = (struct proc *)arg;
2238 	size = sizeof(structsize) + p->p_numthreads *
2239 	    sizeof(elf_kinfo_proc_t);
2240 
2241 	if (sb != NULL) {
2242 		KASSERT(*sizep == size, ("invalid size"));
2243 		structsize = sizeof(elf_kinfo_proc_t);
2244 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2245 		PROC_LOCK(p);
2246 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2247 	}
2248 	*sizep = size;
2249 }
2250 
2251 #ifdef KINFO_FILE_SIZE
2252 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2253 #endif
2254 
2255 static void
2256 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2257 {
2258 	struct proc *p;
2259 	size_t size, sect_sz, i;
2260 	ssize_t start_len, sect_len;
2261 	int structsize, filedesc_flags;
2262 
2263 	if (coredump_pack_fileinfo)
2264 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2265 	else
2266 		filedesc_flags = 0;
2267 
2268 	p = (struct proc *)arg;
2269 	structsize = sizeof(struct kinfo_file);
2270 	if (sb == NULL) {
2271 		size = 0;
2272 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2273 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2274 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2275 		PROC_LOCK(p);
2276 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2277 		sbuf_finish(sb);
2278 		sbuf_delete(sb);
2279 		*sizep = size;
2280 	} else {
2281 		sbuf_start_section(sb, &start_len);
2282 
2283 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2284 		PROC_LOCK(p);
2285 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2286 		    filedesc_flags);
2287 
2288 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2289 		if (sect_len < 0)
2290 			return;
2291 		sect_sz = sect_len;
2292 
2293 		KASSERT(sect_sz <= *sizep,
2294 		    ("kern_proc_filedesc_out did not respect maxlen; "
2295 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2296 		     sect_sz - sizeof(structsize)));
2297 
2298 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2299 			sbuf_putc(sb, 0);
2300 	}
2301 }
2302 
2303 #ifdef KINFO_VMENTRY_SIZE
2304 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2305 #endif
2306 
2307 static void
2308 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2309 {
2310 	struct proc *p;
2311 	size_t size;
2312 	int structsize, vmmap_flags;
2313 
2314 	if (coredump_pack_vmmapinfo)
2315 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2316 	else
2317 		vmmap_flags = 0;
2318 
2319 	p = (struct proc *)arg;
2320 	structsize = sizeof(struct kinfo_vmentry);
2321 	if (sb == NULL) {
2322 		size = 0;
2323 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2324 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2325 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2326 		PROC_LOCK(p);
2327 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2328 		sbuf_finish(sb);
2329 		sbuf_delete(sb);
2330 		*sizep = size;
2331 	} else {
2332 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2333 		PROC_LOCK(p);
2334 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2335 		    vmmap_flags);
2336 	}
2337 }
2338 
2339 static void
2340 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2341 {
2342 	struct proc *p;
2343 	size_t size;
2344 	int structsize;
2345 
2346 	p = (struct proc *)arg;
2347 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2348 	if (sb != NULL) {
2349 		KASSERT(*sizep == size, ("invalid size"));
2350 		structsize = sizeof(gid_t);
2351 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2352 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2353 		    sizeof(gid_t));
2354 	}
2355 	*sizep = size;
2356 }
2357 
2358 static void
2359 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2360 {
2361 	struct proc *p;
2362 	size_t size;
2363 	int structsize;
2364 
2365 	p = (struct proc *)arg;
2366 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2367 	if (sb != NULL) {
2368 		KASSERT(*sizep == size, ("invalid size"));
2369 		structsize = sizeof(p->p_fd->fd_cmask);
2370 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2371 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2372 	}
2373 	*sizep = size;
2374 }
2375 
2376 static void
2377 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2378 {
2379 	struct proc *p;
2380 	struct rlimit rlim[RLIM_NLIMITS];
2381 	size_t size;
2382 	int structsize, i;
2383 
2384 	p = (struct proc *)arg;
2385 	size = sizeof(structsize) + sizeof(rlim);
2386 	if (sb != NULL) {
2387 		KASSERT(*sizep == size, ("invalid size"));
2388 		structsize = sizeof(rlim);
2389 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2390 		PROC_LOCK(p);
2391 		for (i = 0; i < RLIM_NLIMITS; i++)
2392 			lim_rlimit_proc(p, i, &rlim[i]);
2393 		PROC_UNLOCK(p);
2394 		sbuf_bcat(sb, rlim, sizeof(rlim));
2395 	}
2396 	*sizep = size;
2397 }
2398 
2399 static void
2400 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2401 {
2402 	struct proc *p;
2403 	size_t size;
2404 	int structsize;
2405 
2406 	p = (struct proc *)arg;
2407 	size = sizeof(structsize) + sizeof(p->p_osrel);
2408 	if (sb != NULL) {
2409 		KASSERT(*sizep == size, ("invalid size"));
2410 		structsize = sizeof(p->p_osrel);
2411 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2412 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2413 	}
2414 	*sizep = size;
2415 }
2416 
2417 static void
2418 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2419 {
2420 	struct proc *p;
2421 	elf_ps_strings_t ps_strings;
2422 	size_t size;
2423 	int structsize;
2424 
2425 	p = (struct proc *)arg;
2426 	size = sizeof(structsize) + sizeof(ps_strings);
2427 	if (sb != NULL) {
2428 		KASSERT(*sizep == size, ("invalid size"));
2429 		structsize = sizeof(ps_strings);
2430 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2431 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2432 #else
2433 		ps_strings = p->p_sysent->sv_psstrings;
2434 #endif
2435 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2436 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2437 	}
2438 	*sizep = size;
2439 }
2440 
2441 static void
2442 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2443 {
2444 	struct proc *p;
2445 	size_t size;
2446 	int structsize;
2447 
2448 	p = (struct proc *)arg;
2449 	if (sb == NULL) {
2450 		size = 0;
2451 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2452 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2453 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2454 		PHOLD(p);
2455 		proc_getauxv(curthread, p, sb);
2456 		PRELE(p);
2457 		sbuf_finish(sb);
2458 		sbuf_delete(sb);
2459 		*sizep = size;
2460 	} else {
2461 		structsize = sizeof(Elf_Auxinfo);
2462 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2463 		PHOLD(p);
2464 		proc_getauxv(curthread, p, sb);
2465 		PRELE(p);
2466 	}
2467 }
2468 
2469 static boolean_t
2470 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2471     const char *note_vendor, const Elf_Phdr *pnote,
2472     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2473 {
2474 	const Elf_Note *note, *note0, *note_end;
2475 	const char *note_name;
2476 	char *buf;
2477 	int i, error;
2478 	boolean_t res;
2479 
2480 	/* We need some limit, might as well use PAGE_SIZE. */
2481 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2482 		return (FALSE);
2483 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2484 	if (pnote->p_offset > PAGE_SIZE ||
2485 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2486 		VOP_UNLOCK(imgp->vp, 0);
2487 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2488 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2489 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2490 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2491 		    curthread->td_ucred, NOCRED, NULL, curthread);
2492 		if (error != 0) {
2493 			uprintf("i/o error PT_NOTE\n");
2494 			goto retf;
2495 		}
2496 		note = note0 = (const Elf_Note *)buf;
2497 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2498 	} else {
2499 		note = note0 = (const Elf_Note *)(imgp->image_header +
2500 		    pnote->p_offset);
2501 		note_end = (const Elf_Note *)(imgp->image_header +
2502 		    pnote->p_offset + pnote->p_filesz);
2503 		buf = NULL;
2504 	}
2505 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2506 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2507 		    (const char *)note < sizeof(Elf_Note)) {
2508 			goto retf;
2509 		}
2510 		if (note->n_namesz != checknote->n_namesz ||
2511 		    note->n_descsz != checknote->n_descsz ||
2512 		    note->n_type != checknote->n_type)
2513 			goto nextnote;
2514 		note_name = (const char *)(note + 1);
2515 		if (note_name + checknote->n_namesz >=
2516 		    (const char *)note_end || strncmp(note_vendor,
2517 		    note_name, checknote->n_namesz) != 0)
2518 			goto nextnote;
2519 
2520 		if (cb(note, cb_arg, &res))
2521 			goto ret;
2522 nextnote:
2523 		note = (const Elf_Note *)((const char *)(note + 1) +
2524 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2525 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2526 	}
2527 retf:
2528 	res = FALSE;
2529 ret:
2530 	free(buf, M_TEMP);
2531 	return (res);
2532 }
2533 
2534 struct brandnote_cb_arg {
2535 	Elf_Brandnote *brandnote;
2536 	int32_t *osrel;
2537 };
2538 
2539 static boolean_t
2540 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2541 {
2542 	struct brandnote_cb_arg *arg;
2543 
2544 	arg = arg0;
2545 
2546 	/*
2547 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2548 	 * necessary.
2549 	 */
2550 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2551 	    arg->brandnote->trans_osrel != NULL ?
2552 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2553 
2554 	return (TRUE);
2555 }
2556 
2557 static Elf_Note fctl_note = {
2558 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2559 	.n_descsz = sizeof(uint32_t),
2560 	.n_type = NT_FREEBSD_FEATURE_CTL,
2561 };
2562 
2563 struct fctl_cb_arg {
2564 	uint32_t *fctl0;
2565 };
2566 
2567 static boolean_t
2568 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2569 {
2570 	struct fctl_cb_arg *arg;
2571 	const Elf32_Word *desc;
2572 	uintptr_t p;
2573 
2574 	arg = arg0;
2575 	p = (uintptr_t)(note + 1);
2576 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2577 	desc = (const Elf32_Word *)p;
2578 	*arg->fctl0 = desc[0];
2579 	return (TRUE);
2580 }
2581 
2582 /*
2583  * Try to find the appropriate ABI-note section for checknote, fetch
2584  * the osreldate and feature control flags for binary from the ELF
2585  * OSABI-note.  Only the first page of the image is searched, the same
2586  * as for headers.
2587  */
2588 static boolean_t
2589 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2590     int32_t *osrel, uint32_t *fctl0)
2591 {
2592 	const Elf_Phdr *phdr;
2593 	const Elf_Ehdr *hdr;
2594 	struct brandnote_cb_arg b_arg;
2595 	struct fctl_cb_arg f_arg;
2596 	int i, j;
2597 
2598 	hdr = (const Elf_Ehdr *)imgp->image_header;
2599 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2600 	b_arg.brandnote = brandnote;
2601 	b_arg.osrel = osrel;
2602 	f_arg.fctl0 = fctl0;
2603 
2604 	for (i = 0; i < hdr->e_phnum; i++) {
2605 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2606 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2607 		    &b_arg)) {
2608 			for (j = 0; j < hdr->e_phnum; j++) {
2609 				if (phdr[j].p_type == PT_NOTE &&
2610 				    __elfN(parse_notes)(imgp, &fctl_note,
2611 				    FREEBSD_ABI_VENDOR, &phdr[j],
2612 				    note_fctl_cb, &f_arg))
2613 					break;
2614 			}
2615 			return (TRUE);
2616 		}
2617 	}
2618 	return (FALSE);
2619 
2620 }
2621 
2622 /*
2623  * Tell kern_execve.c about it, with a little help from the linker.
2624  */
2625 static struct execsw __elfN(execsw) = {
2626 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2627 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2628 };
2629 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2630 
2631 static vm_prot_t
2632 __elfN(trans_prot)(Elf_Word flags)
2633 {
2634 	vm_prot_t prot;
2635 
2636 	prot = 0;
2637 	if (flags & PF_X)
2638 		prot |= VM_PROT_EXECUTE;
2639 	if (flags & PF_W)
2640 		prot |= VM_PROT_WRITE;
2641 	if (flags & PF_R)
2642 		prot |= VM_PROT_READ;
2643 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2644 	if (i386_read_exec && (flags & PF_R))
2645 		prot |= VM_PROT_EXECUTE;
2646 #endif
2647 	return (prot);
2648 }
2649 
2650 static Elf_Word
2651 __elfN(untrans_prot)(vm_prot_t prot)
2652 {
2653 	Elf_Word flags;
2654 
2655 	flags = 0;
2656 	if (prot & VM_PROT_EXECUTE)
2657 		flags |= PF_X;
2658 	if (prot & VM_PROT_READ)
2659 		flags |= PF_R;
2660 	if (prot & VM_PROT_WRITE)
2661 		flags |= PF_W;
2662 	return (flags);
2663 }
2664