xref: /freebsd/sys/kern/imgact_elf.c (revision 79483833d06343afe6b87b59b1332e186d1339f8)
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);
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     int interp_name_len, 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;
281 
282 	/*
283 	 * We support four types of branding -- (1) the ELF EI_OSABI field
284 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
285 	 * branding w/in the ELF header, (3) path of the `interp_path'
286 	 * field, and (4) the ".note.ABI-tag" ELF section.
287 	 */
288 
289 	/* Look for an ".note.ABI-tag" ELF section */
290 	bi_m = NULL;
291 	for (i = 0; i < MAX_BRANDS; i++) {
292 		bi = elf_brand_list[i];
293 		if (bi == NULL)
294 			continue;
295 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
296 			continue;
297 		if (hdr->e_machine == bi->machine && (bi->flags &
298 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
299 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
300 			    fctl0);
301 			/* Give brand a chance to veto check_note's guess */
302 			if (ret && bi->header_supported)
303 				ret = bi->header_supported(imgp);
304 			/*
305 			 * If note checker claimed the binary, but the
306 			 * interpreter path in the image does not
307 			 * match default one for the brand, try to
308 			 * search for other brands with the same
309 			 * interpreter.  Either there is better brand
310 			 * with the right interpreter, or, failing
311 			 * this, we return first brand which accepted
312 			 * our note and, optionally, header.
313 			 */
314 			if (ret && bi_m == NULL && interp != NULL &&
315 			    (bi->interp_path == NULL ||
316 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
317 			    strncmp(interp, bi->interp_path, interp_name_len)
318 			    != 0))) {
319 				bi_m = bi;
320 				ret = 0;
321 			}
322 			if (ret)
323 				return (bi);
324 		}
325 	}
326 	if (bi_m != NULL)
327 		return (bi_m);
328 
329 	/* If the executable has a brand, search for it in the brand list. */
330 	for (i = 0; i < MAX_BRANDS; i++) {
331 		bi = elf_brand_list[i];
332 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
333 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
334 			continue;
335 		if (hdr->e_machine == bi->machine &&
336 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
337 		    (bi->compat_3_brand != NULL &&
338 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
339 		    bi->compat_3_brand) == 0))) {
340 			/* Looks good, but give brand a chance to veto */
341 			if (bi->header_supported == NULL ||
342 			    bi->header_supported(imgp)) {
343 				/*
344 				 * Again, prefer strictly matching
345 				 * interpreter path.
346 				 */
347 				if (interp_name_len == 0 &&
348 				    bi->interp_path == NULL)
349 					return (bi);
350 				if (bi->interp_path != NULL &&
351 				    strlen(bi->interp_path) + 1 ==
352 				    interp_name_len && strncmp(interp,
353 				    bi->interp_path, interp_name_len) == 0)
354 					return (bi);
355 				if (bi_m == NULL)
356 					bi_m = bi;
357 			}
358 		}
359 	}
360 	if (bi_m != NULL)
361 		return (bi_m);
362 
363 	/* No known brand, see if the header is recognized by any brand */
364 	for (i = 0; i < MAX_BRANDS; i++) {
365 		bi = elf_brand_list[i];
366 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
367 		    bi->header_supported == NULL)
368 			continue;
369 		if (hdr->e_machine == bi->machine) {
370 			ret = bi->header_supported(imgp);
371 			if (ret)
372 				return (bi);
373 		}
374 	}
375 
376 	/* Lacking a known brand, search for a recognized interpreter. */
377 	if (interp != NULL) {
378 		for (i = 0; i < MAX_BRANDS; i++) {
379 			bi = elf_brand_list[i];
380 			if (bi == NULL || (bi->flags &
381 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
382 			    != 0)
383 				continue;
384 			if (hdr->e_machine == bi->machine &&
385 			    bi->interp_path != NULL &&
386 			    /* ELF image p_filesz includes terminating zero */
387 			    strlen(bi->interp_path) + 1 == interp_name_len &&
388 			    strncmp(interp, bi->interp_path, interp_name_len)
389 			    == 0 && (bi->header_supported == NULL ||
390 			    bi->header_supported(imgp)))
391 				return (bi);
392 		}
393 	}
394 
395 	/* Lacking a recognized interpreter, try the default brand */
396 	for (i = 0; i < MAX_BRANDS; i++) {
397 		bi = elf_brand_list[i];
398 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
399 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
400 			continue;
401 		if (hdr->e_machine == bi->machine &&
402 		    __elfN(fallback_brand) == bi->brand &&
403 		    (bi->header_supported == NULL ||
404 		    bi->header_supported(imgp)))
405 			return (bi);
406 	}
407 	return (NULL);
408 }
409 
410 static int
411 __elfN(check_header)(const Elf_Ehdr *hdr)
412 {
413 	Elf_Brandinfo *bi;
414 	int i;
415 
416 	if (!IS_ELF(*hdr) ||
417 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
418 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
419 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
420 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
421 	    hdr->e_version != ELF_TARG_VER)
422 		return (ENOEXEC);
423 
424 	/*
425 	 * Make sure we have at least one brand for this machine.
426 	 */
427 
428 	for (i = 0; i < MAX_BRANDS; i++) {
429 		bi = elf_brand_list[i];
430 		if (bi != NULL && bi->machine == hdr->e_machine)
431 			break;
432 	}
433 	if (i == MAX_BRANDS)
434 		return (ENOEXEC);
435 
436 	return (0);
437 }
438 
439 static int
440 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
441     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
442 {
443 	struct sf_buf *sf;
444 	int error;
445 	vm_offset_t off;
446 
447 	/*
448 	 * Create the page if it doesn't exist yet. Ignore errors.
449 	 */
450 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
451 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
452 
453 	/*
454 	 * Find the page from the underlying object.
455 	 */
456 	if (object != NULL) {
457 		sf = vm_imgact_map_page(object, offset);
458 		if (sf == NULL)
459 			return (KERN_FAILURE);
460 		off = offset - trunc_page(offset);
461 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
462 		    end - start);
463 		vm_imgact_unmap_page(sf);
464 		if (error != 0)
465 			return (KERN_FAILURE);
466 	}
467 
468 	return (KERN_SUCCESS);
469 }
470 
471 static int
472 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
473     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
474     int cow)
475 {
476 	struct sf_buf *sf;
477 	vm_offset_t off;
478 	vm_size_t sz;
479 	int error, locked, rv;
480 
481 	if (start != trunc_page(start)) {
482 		rv = __elfN(map_partial)(map, object, offset, start,
483 		    round_page(start), prot);
484 		if (rv != KERN_SUCCESS)
485 			return (rv);
486 		offset += round_page(start) - start;
487 		start = round_page(start);
488 	}
489 	if (end != round_page(end)) {
490 		rv = __elfN(map_partial)(map, object, offset +
491 		    trunc_page(end) - start, trunc_page(end), end, prot);
492 		if (rv != KERN_SUCCESS)
493 			return (rv);
494 		end = trunc_page(end);
495 	}
496 	if (start >= end)
497 		return (KERN_SUCCESS);
498 	if ((offset & PAGE_MASK) != 0) {
499 		/*
500 		 * The mapping is not page aligned.  This means that we have
501 		 * to copy the data.
502 		 */
503 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
504 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
505 		if (rv != KERN_SUCCESS)
506 			return (rv);
507 		if (object == NULL)
508 			return (KERN_SUCCESS);
509 		for (; start < end; start += sz) {
510 			sf = vm_imgact_map_page(object, offset);
511 			if (sf == NULL)
512 				return (KERN_FAILURE);
513 			off = offset - trunc_page(offset);
514 			sz = end - start;
515 			if (sz > PAGE_SIZE - off)
516 				sz = PAGE_SIZE - off;
517 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
518 			    (caddr_t)start, sz);
519 			vm_imgact_unmap_page(sf);
520 			if (error != 0)
521 				return (KERN_FAILURE);
522 			offset += sz;
523 		}
524 	} else {
525 		vm_object_reference(object);
526 		rv = vm_map_fixed(map, object, offset, start, end - start,
527 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
528 		if (rv != KERN_SUCCESS) {
529 			locked = VOP_ISLOCKED(imgp->vp);
530 			VOP_UNLOCK(imgp->vp, 0);
531 			vm_object_deallocate(object);
532 			vn_lock(imgp->vp, locked | LK_RETRY);
533 			return (rv);
534 		}
535 	}
536 	return (KERN_SUCCESS);
537 }
538 
539 static int
540 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
541     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
542 {
543 	struct sf_buf *sf;
544 	size_t map_len;
545 	vm_map_t map;
546 	vm_object_t object;
547 	vm_offset_t off, map_addr;
548 	int error, rv, cow;
549 	size_t copy_len;
550 	vm_ooffset_t file_addr;
551 
552 	/*
553 	 * It's necessary to fail if the filsz + offset taken from the
554 	 * header is greater than the actual file pager object's size.
555 	 * If we were to allow this, then the vm_map_find() below would
556 	 * walk right off the end of the file object and into the ether.
557 	 *
558 	 * While I'm here, might as well check for something else that
559 	 * is invalid: filsz cannot be greater than memsz.
560 	 */
561 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
562 	    filsz > memsz) {
563 		uprintf("elf_load_section: truncated ELF file\n");
564 		return (ENOEXEC);
565 	}
566 
567 	object = imgp->object;
568 	map = &imgp->proc->p_vmspace->vm_map;
569 	map_addr = trunc_page((vm_offset_t)vmaddr);
570 	file_addr = trunc_page(offset);
571 
572 	/*
573 	 * We have two choices.  We can either clear the data in the last page
574 	 * of an oversized mapping, or we can start the anon mapping a page
575 	 * early and copy the initialized data into that first page.  We
576 	 * choose the second.
577 	 */
578 	if (filsz == 0)
579 		map_len = 0;
580 	else if (memsz > filsz)
581 		map_len = trunc_page(offset + filsz) - file_addr;
582 	else
583 		map_len = round_page(offset + filsz) - file_addr;
584 
585 	if (map_len != 0) {
586 		/* cow flags: don't dump readonly sections in core */
587 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
588 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
589 
590 		rv = __elfN(map_insert)(imgp, map,
591 				      object,
592 				      file_addr,	/* file offset */
593 				      map_addr,		/* virtual start */
594 				      map_addr + map_len,/* virtual end */
595 				      prot,
596 				      cow);
597 		if (rv != KERN_SUCCESS)
598 			return (EINVAL);
599 
600 		/* we can stop now if we've covered it all */
601 		if (memsz == filsz)
602 			return (0);
603 	}
604 
605 
606 	/*
607 	 * We have to get the remaining bit of the file into the first part
608 	 * of the oversized map segment.  This is normally because the .data
609 	 * segment in the file is extended to provide bss.  It's a neat idea
610 	 * to try and save a page, but it's a pain in the behind to implement.
611 	 */
612 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
613 	    filsz);
614 	map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
615 	map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
616 
617 	/* This had damn well better be true! */
618 	if (map_len != 0) {
619 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
620 		    map_addr + map_len, prot, 0);
621 		if (rv != KERN_SUCCESS)
622 			return (EINVAL);
623 	}
624 
625 	if (copy_len != 0) {
626 		sf = vm_imgact_map_page(object, offset + filsz);
627 		if (sf == NULL)
628 			return (EIO);
629 
630 		/* send the page fragment to user space */
631 		off = trunc_page(offset + filsz) - trunc_page(offset + filsz);
632 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
633 		    (caddr_t)map_addr, 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 /*
651  * Load the file "file" into memory.  It may be either a shared object
652  * or an executable.
653  *
654  * The "addr" reference parameter is in/out.  On entry, it specifies
655  * the address where a shared object should be loaded.  If the file is
656  * an executable, this value is ignored.  On exit, "addr" specifies
657  * where the file was actually loaded.
658  *
659  * The "entry" reference parameter is out only.  On exit, it specifies
660  * the entry point for the loaded file.
661  */
662 static int
663 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
664 	u_long *entry)
665 {
666 	struct {
667 		struct nameidata nd;
668 		struct vattr attr;
669 		struct image_params image_params;
670 	} *tempdata;
671 	const Elf_Ehdr *hdr = NULL;
672 	const Elf_Phdr *phdr = NULL;
673 	struct nameidata *nd;
674 	struct vattr *attr;
675 	struct image_params *imgp;
676 	vm_prot_t prot;
677 	u_long rbase;
678 	u_long base_addr = 0;
679 	int error, i, numsegs;
680 
681 #ifdef CAPABILITY_MODE
682 	/*
683 	 * XXXJA: This check can go away once we are sufficiently confident
684 	 * that the checks in namei() are correct.
685 	 */
686 	if (IN_CAPABILITY_MODE(curthread))
687 		return (ECAPMODE);
688 #endif
689 
690 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
691 	nd = &tempdata->nd;
692 	attr = &tempdata->attr;
693 	imgp = &tempdata->image_params;
694 
695 	/*
696 	 * Initialize part of the common data
697 	 */
698 	imgp->proc = p;
699 	imgp->attr = attr;
700 	imgp->firstpage = NULL;
701 	imgp->image_header = NULL;
702 	imgp->object = NULL;
703 	imgp->execlabel = NULL;
704 
705 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
706 	if ((error = namei(nd)) != 0) {
707 		nd->ni_vp = NULL;
708 		goto fail;
709 	}
710 	NDFREE(nd, NDF_ONLY_PNBUF);
711 	imgp->vp = nd->ni_vp;
712 
713 	/*
714 	 * Check permissions, modes, uid, etc on the file, and "open" it.
715 	 */
716 	error = exec_check_permissions(imgp);
717 	if (error)
718 		goto fail;
719 
720 	error = exec_map_first_page(imgp);
721 	if (error)
722 		goto fail;
723 
724 	/*
725 	 * Also make certain that the interpreter stays the same, so set
726 	 * its VV_TEXT flag, too.
727 	 */
728 	VOP_SET_TEXT(nd->ni_vp);
729 
730 	imgp->object = nd->ni_vp->v_object;
731 
732 	hdr = (const Elf_Ehdr *)imgp->image_header;
733 	if ((error = __elfN(check_header)(hdr)) != 0)
734 		goto fail;
735 	if (hdr->e_type == ET_DYN)
736 		rbase = *addr;
737 	else if (hdr->e_type == ET_EXEC)
738 		rbase = 0;
739 	else {
740 		error = ENOEXEC;
741 		goto fail;
742 	}
743 
744 	/* Only support headers that fit within first page for now      */
745 	if ((hdr->e_phoff > PAGE_SIZE) ||
746 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
747 		error = ENOEXEC;
748 		goto fail;
749 	}
750 
751 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
752 	if (!aligned(phdr, Elf_Addr)) {
753 		error = ENOEXEC;
754 		goto fail;
755 	}
756 
757 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
758 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
759 			/* Loadable segment */
760 			prot = __elfN(trans_prot)(phdr[i].p_flags);
761 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
762 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
763 			    phdr[i].p_memsz, phdr[i].p_filesz, prot);
764 			if (error != 0)
765 				goto fail;
766 			/*
767 			 * Establish the base address if this is the
768 			 * first segment.
769 			 */
770 			if (numsegs == 0)
771   				base_addr = trunc_page(phdr[i].p_vaddr +
772 				    rbase);
773 			numsegs++;
774 		}
775 	}
776 	*addr = base_addr;
777 	*entry = (unsigned long)hdr->e_entry + rbase;
778 
779 fail:
780 	if (imgp->firstpage)
781 		exec_unmap_first_page(imgp);
782 
783 	if (nd->ni_vp)
784 		vput(nd->ni_vp);
785 
786 	free(tempdata, M_TEMP);
787 
788 	return (error);
789 }
790 
791 static u_long
792 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
793     u_int align)
794 {
795 	u_long rbase, res;
796 
797 	MPASS(vm_map_min(map) <= minv);
798 	MPASS(maxv <= vm_map_max(map));
799 	MPASS(minv < maxv);
800 	MPASS(minv + align < maxv);
801 	arc4rand(&rbase, sizeof(rbase), 0);
802 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
803 	res &= ~((u_long)align - 1);
804 	if (res >= maxv)
805 		res -= align;
806 	KASSERT(res >= minv,
807 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
808 	    res, minv, maxv, rbase));
809 	KASSERT(res < maxv,
810 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
811 	    res, maxv, minv, rbase));
812 	return (res);
813 }
814 
815 /*
816  * Impossible et_dyn_addr initial value indicating that the real base
817  * must be calculated later with some randomization applied.
818  */
819 #define	ET_DYN_ADDR_RAND	1
820 
821 static int
822 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
823 {
824 	struct thread *td;
825 	const Elf_Ehdr *hdr;
826 	const Elf_Phdr *phdr;
827 	Elf_Auxargs *elf_auxargs;
828 	struct vmspace *vmspace;
829 	vm_map_t map;
830 	const char *err_str, *newinterp;
831 	char *interp, *interp_buf, *path;
832 	Elf_Brandinfo *brand_info;
833 	struct sysentvec *sv;
834 	vm_prot_t prot;
835 	u_long text_size, data_size, total_size, text_addr, data_addr;
836 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
837 	u_long maxalign, mapsz, maxv, maxv1;
838 	uint32_t fctl0;
839 	int32_t osrel;
840 	int error, i, n, interp_name_len, have_interp;
841 
842 	hdr = (const Elf_Ehdr *)imgp->image_header;
843 
844 	/*
845 	 * Do we have a valid ELF header ?
846 	 *
847 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
848 	 * if particular brand doesn't support it.
849 	 */
850 	if (__elfN(check_header)(hdr) != 0 ||
851 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
852 		return (-1);
853 
854 	/*
855 	 * From here on down, we return an errno, not -1, as we've
856 	 * detected an ELF file.
857 	 */
858 
859 	if ((hdr->e_phoff > PAGE_SIZE) ||
860 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
861 		/* Only support headers in first page for now */
862 		uprintf("Program headers not in the first page\n");
863 		return (ENOEXEC);
864 	}
865 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
866 	if (!aligned(phdr, Elf_Addr)) {
867 		uprintf("Unaligned program headers\n");
868 		return (ENOEXEC);
869 	}
870 
871 	n = error = 0;
872 	baddr = 0;
873 	osrel = 0;
874 	fctl0 = 0;
875 	text_size = data_size = total_size = text_addr = data_addr = 0;
876 	entry = proghdr = 0;
877 	interp_name_len = 0;
878 	err_str = newinterp = NULL;
879 	interp = interp_buf = NULL;
880 	td = curthread;
881 	maxalign = PAGE_SIZE;
882 	mapsz = 0;
883 
884 	for (i = 0; i < hdr->e_phnum; i++) {
885 		switch (phdr[i].p_type) {
886 		case PT_LOAD:
887 			if (n == 0)
888 				baddr = phdr[i].p_vaddr;
889 			if (phdr[i].p_align > maxalign)
890 				maxalign = phdr[i].p_align;
891 			mapsz += phdr[i].p_memsz;
892 			n++;
893 			break;
894 		case PT_INTERP:
895 			/* Path to interpreter */
896 			if (phdr[i].p_filesz < 2 ||
897 			    phdr[i].p_filesz > MAXPATHLEN) {
898 				uprintf("Invalid PT_INTERP\n");
899 				error = ENOEXEC;
900 				goto ret;
901 			}
902 			if (interp != NULL) {
903 				uprintf("Multiple PT_INTERP headers\n");
904 				error = ENOEXEC;
905 				goto ret;
906 			}
907 			interp_name_len = phdr[i].p_filesz;
908 			if (phdr[i].p_offset > PAGE_SIZE ||
909 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
910 				VOP_UNLOCK(imgp->vp, 0);
911 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
912 				    M_WAITOK);
913 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
914 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
915 				    interp_name_len, phdr[i].p_offset,
916 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
917 				    NOCRED, NULL, td);
918 				if (error != 0) {
919 					uprintf("i/o error PT_INTERP %d\n",
920 					    error);
921 					goto ret;
922 				}
923 				interp_buf[interp_name_len] = '\0';
924 				interp = interp_buf;
925 			} else {
926 				interp = __DECONST(char *, imgp->image_header) +
927 				    phdr[i].p_offset;
928 				if (interp[interp_name_len - 1] != '\0') {
929 					uprintf("Invalid PT_INTERP\n");
930 					error = ENOEXEC;
931 					goto ret;
932 				}
933 			}
934 			break;
935 		case PT_GNU_STACK:
936 			if (__elfN(nxstack))
937 				imgp->stack_prot =
938 				    __elfN(trans_prot)(phdr[i].p_flags);
939 			imgp->stack_sz = phdr[i].p_memsz;
940 			break;
941 		}
942 	}
943 
944 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
945 	    &osrel, &fctl0);
946 	if (brand_info == NULL) {
947 		uprintf("ELF binary type \"%u\" not known.\n",
948 		    hdr->e_ident[EI_OSABI]);
949 		error = ENOEXEC;
950 		goto ret;
951 	}
952 	sv = brand_info->sysvec;
953 	et_dyn_addr = 0;
954 	if (hdr->e_type == ET_DYN) {
955 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
956 			uprintf("Cannot execute shared object\n");
957 			error = ENOEXEC;
958 			goto ret;
959 		}
960 		/*
961 		 * Honour the base load address from the dso if it is
962 		 * non-zero for some reason.
963 		 */
964 		if (baddr == 0) {
965 			if ((sv->sv_flags & SV_ASLR) == 0 ||
966 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
967 				et_dyn_addr = ET_DYN_LOAD_ADDR;
968 			else if ((__elfN(pie_aslr_enabled) &&
969 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
970 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
971 				et_dyn_addr = ET_DYN_ADDR_RAND;
972 			else
973 				et_dyn_addr = ET_DYN_LOAD_ADDR;
974 		}
975 	}
976 	if (interp != NULL && brand_info->interp_newpath != NULL)
977 		newinterp = brand_info->interp_newpath;
978 
979 	/*
980 	 * Avoid a possible deadlock if the current address space is destroyed
981 	 * and that address space maps the locked vnode.  In the common case,
982 	 * the locked vnode's v_usecount is decremented but remains greater
983 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
984 	 * However, in cases where the vnode lock is external, such as nullfs,
985 	 * v_usecount may become zero.
986 	 *
987 	 * The VV_TEXT flag prevents modifications to the executable while
988 	 * the vnode is unlocked.
989 	 */
990 	VOP_UNLOCK(imgp->vp, 0);
991 
992 	/*
993 	 * Decide whether to enable randomization of user mappings.
994 	 * First, reset user preferences for the setid binaries.
995 	 * Then, account for the support of the randomization by the
996 	 * ABI, by user preferences, and make special treatment for
997 	 * PIE binaries.
998 	 */
999 	if (imgp->credential_setid) {
1000 		PROC_LOCK(imgp->proc);
1001 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1002 		PROC_UNLOCK(imgp->proc);
1003 	}
1004 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1005 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1006 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1007 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1008 		    ("et_dyn_addr == RAND and !ASLR"));
1009 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1010 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1011 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1012 		imgp->map_flags |= MAP_ASLR;
1013 		/*
1014 		 * If user does not care about sbrk, utilize the bss
1015 		 * grow region for mappings as well.  We can select
1016 		 * the base for the image anywere and still not suffer
1017 		 * from the fragmentation.
1018 		 */
1019 		if (!__elfN(aslr_honor_sbrk) ||
1020 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1021 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1022 	}
1023 
1024 	error = exec_new_vmspace(imgp, sv);
1025 	vmspace = imgp->proc->p_vmspace;
1026 	map = &vmspace->vm_map;
1027 
1028 	imgp->proc->p_sysent = sv;
1029 
1030 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1031 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1032 		KASSERT((map->flags & MAP_ASLR) != 0,
1033 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1034 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1035 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1036 		    /* reserve half of the address space to interpreter */
1037 		    maxv / 2, 1UL << flsl(maxalign));
1038 	}
1039 
1040 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1041 	if (error != 0)
1042 		goto ret;
1043 
1044 	for (i = 0; i < hdr->e_phnum; i++) {
1045 		switch (phdr[i].p_type) {
1046 		case PT_LOAD:	/* Loadable segment */
1047 			if (phdr[i].p_memsz == 0)
1048 				break;
1049 			prot = __elfN(trans_prot)(phdr[i].p_flags);
1050 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
1051 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
1052 			    phdr[i].p_memsz, phdr[i].p_filesz, prot);
1053 			if (error != 0)
1054 				goto ret;
1055 
1056 			/*
1057 			 * If this segment contains the program headers,
1058 			 * remember their virtual address for the AT_PHDR
1059 			 * aux entry. Static binaries don't usually include
1060 			 * a PT_PHDR entry.
1061 			 */
1062 			if (phdr[i].p_offset == 0 &&
1063 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1064 				<= phdr[i].p_filesz)
1065 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
1066 				    et_dyn_addr;
1067 
1068 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
1069 			seg_size = round_page(phdr[i].p_memsz +
1070 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
1071 
1072 			/*
1073 			 * Make the largest executable segment the official
1074 			 * text segment and all others data.
1075 			 *
1076 			 * Note that obreak() assumes that data_addr +
1077 			 * data_size == end of data load area, and the ELF
1078 			 * file format expects segments to be sorted by
1079 			 * address.  If multiple data segments exist, the
1080 			 * last one will be used.
1081 			 */
1082 
1083 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
1084 				text_size = seg_size;
1085 				text_addr = seg_addr;
1086 			} else {
1087 				data_size = seg_size;
1088 				data_addr = seg_addr;
1089 			}
1090 			total_size += seg_size;
1091 			break;
1092 		case PT_PHDR: 	/* Program header table info */
1093 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
1094 			break;
1095 		default:
1096 			break;
1097 		}
1098 	}
1099 
1100 	if (data_addr == 0 && data_size == 0) {
1101 		data_addr = text_addr;
1102 		data_size = text_size;
1103 	}
1104 
1105 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1106 
1107 	/*
1108 	 * Check limits.  It should be safe to check the
1109 	 * limits after loading the segments since we do
1110 	 * not actually fault in all the segments pages.
1111 	 */
1112 	PROC_LOCK(imgp->proc);
1113 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1114 		err_str = "Data segment size exceeds process limit";
1115 	else if (text_size > maxtsiz)
1116 		err_str = "Text segment size exceeds system limit";
1117 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1118 		err_str = "Total segment size exceeds process limit";
1119 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1120 		err_str = "Data segment size exceeds resource limit";
1121 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1122 		err_str = "Total segment size exceeds resource limit";
1123 	if (err_str != NULL) {
1124 		PROC_UNLOCK(imgp->proc);
1125 		uprintf("%s\n", err_str);
1126 		error = ENOMEM;
1127 		goto ret;
1128 	}
1129 
1130 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1131 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1132 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1133 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1134 
1135 	/*
1136 	 * We load the dynamic linker where a userland call
1137 	 * to mmap(0, ...) would put it.  The rationale behind this
1138 	 * calculation is that it leaves room for the heap to grow to
1139 	 * its maximum allowed size.
1140 	 */
1141 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1142 	    RLIMIT_DATA));
1143 	if ((map->flags & MAP_ASLR) != 0) {
1144 		maxv1 = maxv / 2 + addr / 2;
1145 		MPASS(maxv1 >= addr);	/* No overflow */
1146 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1147 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1148 	} else {
1149 		map->anon_loc = addr;
1150 	}
1151 	PROC_UNLOCK(imgp->proc);
1152 
1153 	imgp->entry_addr = entry;
1154 
1155 	if (interp != NULL) {
1156 		have_interp = FALSE;
1157 		VOP_UNLOCK(imgp->vp, 0);
1158 		if ((map->flags & MAP_ASLR) != 0) {
1159 			/* Assume that interpeter fits into 1/4 of AS */
1160 			maxv1 = maxv / 2 + addr / 2;
1161 			MPASS(maxv1 >= addr);	/* No overflow */
1162 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1163 			    maxv1, PAGE_SIZE);
1164 		}
1165 		if (brand_info->emul_path != NULL &&
1166 		    brand_info->emul_path[0] != '\0') {
1167 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1168 			snprintf(path, MAXPATHLEN, "%s%s",
1169 			    brand_info->emul_path, interp);
1170 			error = __elfN(load_file)(imgp->proc, path, &addr,
1171 			    &imgp->entry_addr);
1172 			free(path, M_TEMP);
1173 			if (error == 0)
1174 				have_interp = TRUE;
1175 		}
1176 		if (!have_interp && newinterp != NULL &&
1177 		    (brand_info->interp_path == NULL ||
1178 		    strcmp(interp, brand_info->interp_path) == 0)) {
1179 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1180 			    &imgp->entry_addr);
1181 			if (error == 0)
1182 				have_interp = TRUE;
1183 		}
1184 		if (!have_interp) {
1185 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1186 			    &imgp->entry_addr);
1187 		}
1188 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1189 		if (error != 0) {
1190 			uprintf("ELF interpreter %s not found, error %d\n",
1191 			    interp, error);
1192 			goto ret;
1193 		}
1194 	} else
1195 		addr = et_dyn_addr;
1196 
1197 	/*
1198 	 * Construct auxargs table (used by the fixup routine)
1199 	 */
1200 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1201 	elf_auxargs->execfd = -1;
1202 	elf_auxargs->phdr = proghdr;
1203 	elf_auxargs->phent = hdr->e_phentsize;
1204 	elf_auxargs->phnum = hdr->e_phnum;
1205 	elf_auxargs->pagesz = PAGE_SIZE;
1206 	elf_auxargs->base = addr;
1207 	elf_auxargs->flags = 0;
1208 	elf_auxargs->entry = entry;
1209 	elf_auxargs->hdr_eflags = hdr->e_flags;
1210 
1211 	imgp->auxargs = elf_auxargs;
1212 	imgp->interpreted = 0;
1213 	imgp->reloc_base = addr;
1214 	imgp->proc->p_osrel = osrel;
1215 	imgp->proc->p_fctl0 = fctl0;
1216 	imgp->proc->p_elf_machine = hdr->e_machine;
1217 	imgp->proc->p_elf_flags = hdr->e_flags;
1218 
1219 ret:
1220 	free(interp_buf, M_TEMP);
1221 	return (error);
1222 }
1223 
1224 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1225 
1226 int
1227 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1228 {
1229 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1230 	Elf_Auxinfo *argarray, *pos;
1231 	Elf_Addr *base, *auxbase;
1232 	int error;
1233 
1234 	base = (Elf_Addr *)*stack_base;
1235 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1236 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1237 	    M_WAITOK | M_ZERO);
1238 
1239 	if (args->execfd != -1)
1240 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1241 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1242 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1243 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1244 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1245 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1246 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1247 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1248 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1249 	if (imgp->execpathp != 0)
1250 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1251 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1252 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1253 	if (imgp->canary != 0) {
1254 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1255 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1256 	}
1257 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1258 	if (imgp->pagesizes != 0) {
1259 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1260 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1261 	}
1262 	if (imgp->sysent->sv_timekeep_base != 0) {
1263 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1264 		    imgp->sysent->sv_timekeep_base);
1265 	}
1266 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1267 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1268 	    imgp->sysent->sv_stackprot);
1269 	if (imgp->sysent->sv_hwcap != NULL)
1270 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1271 	if (imgp->sysent->sv_hwcap2 != NULL)
1272 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1273 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1274 
1275 	free(imgp->auxargs, M_TEMP);
1276 	imgp->auxargs = NULL;
1277 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1278 
1279 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1280 	free(argarray, M_TEMP);
1281 	if (error != 0)
1282 		return (error);
1283 
1284 	base--;
1285 	if (suword(base, imgp->args->argc) == -1)
1286 		return (EFAULT);
1287 	*stack_base = (register_t *)base;
1288 	return (0);
1289 }
1290 
1291 /*
1292  * Code for generating ELF core dumps.
1293  */
1294 
1295 typedef void (*segment_callback)(vm_map_entry_t, void *);
1296 
1297 /* Closure for cb_put_phdr(). */
1298 struct phdr_closure {
1299 	Elf_Phdr *phdr;		/* Program header to fill in */
1300 	Elf_Off offset;		/* Offset of segment in core file */
1301 };
1302 
1303 /* Closure for cb_size_segment(). */
1304 struct sseg_closure {
1305 	int count;		/* Count of writable segments. */
1306 	size_t size;		/* Total size of all writable segments. */
1307 };
1308 
1309 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1310 
1311 struct note_info {
1312 	int		type;		/* Note type. */
1313 	outfunc_t 	outfunc; 	/* Output function. */
1314 	void		*outarg;	/* Argument for the output function. */
1315 	size_t		outsize;	/* Output size. */
1316 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1317 };
1318 
1319 TAILQ_HEAD(note_info_list, note_info);
1320 
1321 /* Coredump output parameters. */
1322 struct coredump_params {
1323 	off_t		offset;
1324 	struct ucred	*active_cred;
1325 	struct ucred	*file_cred;
1326 	struct thread	*td;
1327 	struct vnode	*vp;
1328 	struct compressor *comp;
1329 };
1330 
1331 extern int compress_user_cores;
1332 extern int compress_user_cores_level;
1333 
1334 static void cb_put_phdr(vm_map_entry_t, void *);
1335 static void cb_size_segment(vm_map_entry_t, void *);
1336 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1337     enum uio_seg);
1338 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1339 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1340     struct note_info_list *, size_t);
1341 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1342     size_t *);
1343 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1344 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1345 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1346 static int sbuf_drain_core_output(void *, const char *, int);
1347 static int sbuf_drain_count(void *arg, const char *data, int len);
1348 
1349 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1350 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1351 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1352 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1353 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1354 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1355 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1356 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1357 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1358 static void note_procstat_files(void *, struct sbuf *, size_t *);
1359 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1360 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1361 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1362 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1363 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1364 
1365 /*
1366  * Write out a core segment to the compression stream.
1367  */
1368 static int
1369 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1370 {
1371 	u_int chunk_len;
1372 	int error;
1373 
1374 	while (len > 0) {
1375 		chunk_len = MIN(len, CORE_BUF_SIZE);
1376 
1377 		/*
1378 		 * We can get EFAULT error here.
1379 		 * In that case zero out the current chunk of the segment.
1380 		 */
1381 		error = copyin(base, buf, chunk_len);
1382 		if (error != 0)
1383 			bzero(buf, chunk_len);
1384 		error = compressor_write(p->comp, buf, chunk_len);
1385 		if (error != 0)
1386 			break;
1387 		base += chunk_len;
1388 		len -= chunk_len;
1389 	}
1390 	return (error);
1391 }
1392 
1393 static int
1394 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1395 {
1396 
1397 	return (core_write((struct coredump_params *)arg, base, len, offset,
1398 	    UIO_SYSSPACE));
1399 }
1400 
1401 static int
1402 core_write(struct coredump_params *p, const void *base, size_t len,
1403     off_t offset, enum uio_seg seg)
1404 {
1405 
1406 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1407 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1408 	    p->active_cred, p->file_cred, NULL, p->td));
1409 }
1410 
1411 static int
1412 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1413     void *tmpbuf)
1414 {
1415 	int error;
1416 
1417 	if (p->comp != NULL)
1418 		return (compress_chunk(p, base, tmpbuf, len));
1419 
1420 	/*
1421 	 * EFAULT is a non-fatal error that we can get, for example,
1422 	 * if the segment is backed by a file but extends beyond its
1423 	 * end.
1424 	 */
1425 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1426 	if (error == EFAULT) {
1427 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1428 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1429 		    "for process %s\n", base, len, offset, curproc->p_comm);
1430 
1431 		/*
1432 		 * Write a "real" zero byte at the end of the target region
1433 		 * in the case this is the last segment.
1434 		 * The intermediate space will be implicitly zero-filled.
1435 		 */
1436 		error = core_write(p, zero_region, 1, offset + len - 1,
1437 		    UIO_SYSSPACE);
1438 	}
1439 	return (error);
1440 }
1441 
1442 /*
1443  * Drain into a core file.
1444  */
1445 static int
1446 sbuf_drain_core_output(void *arg, const char *data, int len)
1447 {
1448 	struct coredump_params *p;
1449 	int error, locked;
1450 
1451 	p = (struct coredump_params *)arg;
1452 
1453 	/*
1454 	 * Some kern_proc out routines that print to this sbuf may
1455 	 * call us with the process lock held. Draining with the
1456 	 * non-sleepable lock held is unsafe. The lock is needed for
1457 	 * those routines when dumping a live process. In our case we
1458 	 * can safely release the lock before draining and acquire
1459 	 * again after.
1460 	 */
1461 	locked = PROC_LOCKED(p->td->td_proc);
1462 	if (locked)
1463 		PROC_UNLOCK(p->td->td_proc);
1464 	if (p->comp != NULL)
1465 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1466 	else
1467 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1468 		    UIO_SYSSPACE);
1469 	if (locked)
1470 		PROC_LOCK(p->td->td_proc);
1471 	if (error != 0)
1472 		return (-error);
1473 	p->offset += len;
1474 	return (len);
1475 }
1476 
1477 /*
1478  * Drain into a counter.
1479  */
1480 static int
1481 sbuf_drain_count(void *arg, const char *data __unused, int len)
1482 {
1483 	size_t *sizep;
1484 
1485 	sizep = (size_t *)arg;
1486 	*sizep += len;
1487 	return (len);
1488 }
1489 
1490 int
1491 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1492 {
1493 	struct ucred *cred = td->td_ucred;
1494 	int error = 0;
1495 	struct sseg_closure seginfo;
1496 	struct note_info_list notelst;
1497 	struct coredump_params params;
1498 	struct note_info *ninfo;
1499 	void *hdr, *tmpbuf;
1500 	size_t hdrsize, notesz, coresize;
1501 
1502 	hdr = NULL;
1503 	tmpbuf = NULL;
1504 	TAILQ_INIT(&notelst);
1505 
1506 	/* Size the program segments. */
1507 	seginfo.count = 0;
1508 	seginfo.size = 0;
1509 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1510 
1511 	/*
1512 	 * Collect info about the core file header area.
1513 	 */
1514 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1515 	if (seginfo.count + 1 >= PN_XNUM)
1516 		hdrsize += sizeof(Elf_Shdr);
1517 	__elfN(prepare_notes)(td, &notelst, &notesz);
1518 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1519 
1520 	/* Set up core dump parameters. */
1521 	params.offset = 0;
1522 	params.active_cred = cred;
1523 	params.file_cred = NOCRED;
1524 	params.td = td;
1525 	params.vp = vp;
1526 	params.comp = NULL;
1527 
1528 #ifdef RACCT
1529 	if (racct_enable) {
1530 		PROC_LOCK(td->td_proc);
1531 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1532 		PROC_UNLOCK(td->td_proc);
1533 		if (error != 0) {
1534 			error = EFAULT;
1535 			goto done;
1536 		}
1537 	}
1538 #endif
1539 	if (coresize >= limit) {
1540 		error = EFAULT;
1541 		goto done;
1542 	}
1543 
1544 	/* Create a compression stream if necessary. */
1545 	if (compress_user_cores != 0) {
1546 		params.comp = compressor_init(core_compressed_write,
1547 		    compress_user_cores, CORE_BUF_SIZE,
1548 		    compress_user_cores_level, &params);
1549 		if (params.comp == NULL) {
1550 			error = EFAULT;
1551 			goto done;
1552 		}
1553 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1554         }
1555 
1556 	/*
1557 	 * Allocate memory for building the header, fill it up,
1558 	 * and write it out following the notes.
1559 	 */
1560 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1561 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1562 	    notesz);
1563 
1564 	/* Write the contents of all of the writable segments. */
1565 	if (error == 0) {
1566 		Elf_Phdr *php;
1567 		off_t offset;
1568 		int i;
1569 
1570 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1571 		offset = round_page(hdrsize + notesz);
1572 		for (i = 0; i < seginfo.count; i++) {
1573 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1574 			    php->p_filesz, offset, &params, tmpbuf);
1575 			if (error != 0)
1576 				break;
1577 			offset += php->p_filesz;
1578 			php++;
1579 		}
1580 		if (error == 0 && params.comp != NULL)
1581 			error = compressor_flush(params.comp);
1582 	}
1583 	if (error) {
1584 		log(LOG_WARNING,
1585 		    "Failed to write core file for process %s (error %d)\n",
1586 		    curproc->p_comm, error);
1587 	}
1588 
1589 done:
1590 	free(tmpbuf, M_TEMP);
1591 	if (params.comp != NULL)
1592 		compressor_fini(params.comp);
1593 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1594 		TAILQ_REMOVE(&notelst, ninfo, link);
1595 		free(ninfo, M_TEMP);
1596 	}
1597 	if (hdr != NULL)
1598 		free(hdr, M_TEMP);
1599 
1600 	return (error);
1601 }
1602 
1603 /*
1604  * A callback for each_dumpable_segment() to write out the segment's
1605  * program header entry.
1606  */
1607 static void
1608 cb_put_phdr(vm_map_entry_t entry, void *closure)
1609 {
1610 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1611 	Elf_Phdr *phdr = phc->phdr;
1612 
1613 	phc->offset = round_page(phc->offset);
1614 
1615 	phdr->p_type = PT_LOAD;
1616 	phdr->p_offset = phc->offset;
1617 	phdr->p_vaddr = entry->start;
1618 	phdr->p_paddr = 0;
1619 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1620 	phdr->p_align = PAGE_SIZE;
1621 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1622 
1623 	phc->offset += phdr->p_filesz;
1624 	phc->phdr++;
1625 }
1626 
1627 /*
1628  * A callback for each_dumpable_segment() to gather information about
1629  * the number of segments and their total size.
1630  */
1631 static void
1632 cb_size_segment(vm_map_entry_t entry, void *closure)
1633 {
1634 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1635 
1636 	ssc->count++;
1637 	ssc->size += entry->end - entry->start;
1638 }
1639 
1640 /*
1641  * For each writable segment in the process's memory map, call the given
1642  * function with a pointer to the map entry and some arbitrary
1643  * caller-supplied data.
1644  */
1645 static void
1646 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1647 {
1648 	struct proc *p = td->td_proc;
1649 	vm_map_t map = &p->p_vmspace->vm_map;
1650 	vm_map_entry_t entry;
1651 	vm_object_t backing_object, object;
1652 	boolean_t ignore_entry;
1653 
1654 	vm_map_lock_read(map);
1655 	for (entry = map->header.next; entry != &map->header;
1656 	    entry = entry->next) {
1657 		/*
1658 		 * Don't dump inaccessible mappings, deal with legacy
1659 		 * coredump mode.
1660 		 *
1661 		 * Note that read-only segments related to the elf binary
1662 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1663 		 * need to arbitrarily ignore such segments.
1664 		 */
1665 		if (elf_legacy_coredump) {
1666 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1667 				continue;
1668 		} else {
1669 			if ((entry->protection & VM_PROT_ALL) == 0)
1670 				continue;
1671 		}
1672 
1673 		/*
1674 		 * Dont include memory segment in the coredump if
1675 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1676 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1677 		 * kernel map).
1678 		 */
1679 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1680 			continue;
1681 
1682 		if ((object = entry->object.vm_object) == NULL)
1683 			continue;
1684 
1685 		/* Ignore memory-mapped devices and such things. */
1686 		VM_OBJECT_RLOCK(object);
1687 		while ((backing_object = object->backing_object) != NULL) {
1688 			VM_OBJECT_RLOCK(backing_object);
1689 			VM_OBJECT_RUNLOCK(object);
1690 			object = backing_object;
1691 		}
1692 		ignore_entry = object->type != OBJT_DEFAULT &&
1693 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1694 		    object->type != OBJT_PHYS;
1695 		VM_OBJECT_RUNLOCK(object);
1696 		if (ignore_entry)
1697 			continue;
1698 
1699 		(*func)(entry, closure);
1700 	}
1701 	vm_map_unlock_read(map);
1702 }
1703 
1704 /*
1705  * Write the core file header to the file, including padding up to
1706  * the page boundary.
1707  */
1708 static int
1709 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1710     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1711 {
1712 	struct note_info *ninfo;
1713 	struct sbuf *sb;
1714 	int error;
1715 
1716 	/* Fill in the header. */
1717 	bzero(hdr, hdrsize);
1718 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1719 
1720 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1721 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1722 	sbuf_start_section(sb, NULL);
1723 	sbuf_bcat(sb, hdr, hdrsize);
1724 	TAILQ_FOREACH(ninfo, notelst, link)
1725 	    __elfN(putnote)(ninfo, sb);
1726 	/* Align up to a page boundary for the program segments. */
1727 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1728 	error = sbuf_finish(sb);
1729 	sbuf_delete(sb);
1730 
1731 	return (error);
1732 }
1733 
1734 static void
1735 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1736     size_t *sizep)
1737 {
1738 	struct proc *p;
1739 	struct thread *thr;
1740 	size_t size;
1741 
1742 	p = td->td_proc;
1743 	size = 0;
1744 
1745 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1746 
1747 	/*
1748 	 * To have the debugger select the right thread (LWP) as the initial
1749 	 * thread, we dump the state of the thread passed to us in td first.
1750 	 * This is the thread that causes the core dump and thus likely to
1751 	 * be the right thread one wants to have selected in the debugger.
1752 	 */
1753 	thr = td;
1754 	while (thr != NULL) {
1755 		size += register_note(list, NT_PRSTATUS,
1756 		    __elfN(note_prstatus), thr);
1757 		size += register_note(list, NT_FPREGSET,
1758 		    __elfN(note_fpregset), thr);
1759 		size += register_note(list, NT_THRMISC,
1760 		    __elfN(note_thrmisc), thr);
1761 		size += register_note(list, NT_PTLWPINFO,
1762 		    __elfN(note_ptlwpinfo), thr);
1763 		size += register_note(list, -1,
1764 		    __elfN(note_threadmd), thr);
1765 
1766 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1767 		    TAILQ_NEXT(thr, td_plist);
1768 		if (thr == td)
1769 			thr = TAILQ_NEXT(thr, td_plist);
1770 	}
1771 
1772 	size += register_note(list, NT_PROCSTAT_PROC,
1773 	    __elfN(note_procstat_proc), p);
1774 	size += register_note(list, NT_PROCSTAT_FILES,
1775 	    note_procstat_files, p);
1776 	size += register_note(list, NT_PROCSTAT_VMMAP,
1777 	    note_procstat_vmmap, p);
1778 	size += register_note(list, NT_PROCSTAT_GROUPS,
1779 	    note_procstat_groups, p);
1780 	size += register_note(list, NT_PROCSTAT_UMASK,
1781 	    note_procstat_umask, p);
1782 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1783 	    note_procstat_rlimit, p);
1784 	size += register_note(list, NT_PROCSTAT_OSREL,
1785 	    note_procstat_osrel, p);
1786 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1787 	    __elfN(note_procstat_psstrings), p);
1788 	size += register_note(list, NT_PROCSTAT_AUXV,
1789 	    __elfN(note_procstat_auxv), p);
1790 
1791 	*sizep = size;
1792 }
1793 
1794 static void
1795 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1796     size_t notesz)
1797 {
1798 	Elf_Ehdr *ehdr;
1799 	Elf_Phdr *phdr;
1800 	Elf_Shdr *shdr;
1801 	struct phdr_closure phc;
1802 
1803 	ehdr = (Elf_Ehdr *)hdr;
1804 
1805 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1806 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1807 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1808 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1809 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1810 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1811 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1812 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1813 	ehdr->e_ident[EI_ABIVERSION] = 0;
1814 	ehdr->e_ident[EI_PAD] = 0;
1815 	ehdr->e_type = ET_CORE;
1816 	ehdr->e_machine = td->td_proc->p_elf_machine;
1817 	ehdr->e_version = EV_CURRENT;
1818 	ehdr->e_entry = 0;
1819 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1820 	ehdr->e_flags = td->td_proc->p_elf_flags;
1821 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1822 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1823 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1824 	ehdr->e_shstrndx = SHN_UNDEF;
1825 	if (numsegs + 1 < PN_XNUM) {
1826 		ehdr->e_phnum = numsegs + 1;
1827 		ehdr->e_shnum = 0;
1828 	} else {
1829 		ehdr->e_phnum = PN_XNUM;
1830 		ehdr->e_shnum = 1;
1831 
1832 		ehdr->e_shoff = ehdr->e_phoff +
1833 		    (numsegs + 1) * ehdr->e_phentsize;
1834 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1835 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1836 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1837 
1838 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1839 		memset(shdr, 0, sizeof(*shdr));
1840 		/*
1841 		 * A special first section is used to hold large segment and
1842 		 * section counts.  This was proposed by Sun Microsystems in
1843 		 * Solaris and has been adopted by Linux; the standard ELF
1844 		 * tools are already familiar with the technique.
1845 		 *
1846 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1847 		 * (or 12-7 depending on the version of the document) for more
1848 		 * details.
1849 		 */
1850 		shdr->sh_type = SHT_NULL;
1851 		shdr->sh_size = ehdr->e_shnum;
1852 		shdr->sh_link = ehdr->e_shstrndx;
1853 		shdr->sh_info = numsegs + 1;
1854 	}
1855 
1856 	/*
1857 	 * Fill in the program header entries.
1858 	 */
1859 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1860 
1861 	/* The note segement. */
1862 	phdr->p_type = PT_NOTE;
1863 	phdr->p_offset = hdrsize;
1864 	phdr->p_vaddr = 0;
1865 	phdr->p_paddr = 0;
1866 	phdr->p_filesz = notesz;
1867 	phdr->p_memsz = 0;
1868 	phdr->p_flags = PF_R;
1869 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1870 	phdr++;
1871 
1872 	/* All the writable segments from the program. */
1873 	phc.phdr = phdr;
1874 	phc.offset = round_page(hdrsize + notesz);
1875 	each_dumpable_segment(td, cb_put_phdr, &phc);
1876 }
1877 
1878 static size_t
1879 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1880 {
1881 	struct note_info *ninfo;
1882 	size_t size, notesize;
1883 
1884 	size = 0;
1885 	out(arg, NULL, &size);
1886 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1887 	ninfo->type = type;
1888 	ninfo->outfunc = out;
1889 	ninfo->outarg = arg;
1890 	ninfo->outsize = size;
1891 	TAILQ_INSERT_TAIL(list, ninfo, link);
1892 
1893 	if (type == -1)
1894 		return (size);
1895 
1896 	notesize = sizeof(Elf_Note) +		/* note header */
1897 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1898 						/* note name */
1899 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1900 
1901 	return (notesize);
1902 }
1903 
1904 static size_t
1905 append_note_data(const void *src, void *dst, size_t len)
1906 {
1907 	size_t padded_len;
1908 
1909 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1910 	if (dst != NULL) {
1911 		bcopy(src, dst, len);
1912 		bzero((char *)dst + len, padded_len - len);
1913 	}
1914 	return (padded_len);
1915 }
1916 
1917 size_t
1918 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1919 {
1920 	Elf_Note *note;
1921 	char *buf;
1922 	size_t notesize;
1923 
1924 	buf = dst;
1925 	if (buf != NULL) {
1926 		note = (Elf_Note *)buf;
1927 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1928 		note->n_descsz = size;
1929 		note->n_type = type;
1930 		buf += sizeof(*note);
1931 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1932 		    sizeof(FREEBSD_ABI_VENDOR));
1933 		append_note_data(src, buf, size);
1934 		if (descp != NULL)
1935 			*descp = buf;
1936 	}
1937 
1938 	notesize = sizeof(Elf_Note) +		/* note header */
1939 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1940 						/* note name */
1941 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1942 
1943 	return (notesize);
1944 }
1945 
1946 static void
1947 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1948 {
1949 	Elf_Note note;
1950 	ssize_t old_len, sect_len;
1951 	size_t new_len, descsz, i;
1952 
1953 	if (ninfo->type == -1) {
1954 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1955 		return;
1956 	}
1957 
1958 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1959 	note.n_descsz = ninfo->outsize;
1960 	note.n_type = ninfo->type;
1961 
1962 	sbuf_bcat(sb, &note, sizeof(note));
1963 	sbuf_start_section(sb, &old_len);
1964 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1965 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1966 	if (note.n_descsz == 0)
1967 		return;
1968 	sbuf_start_section(sb, &old_len);
1969 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1970 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1971 	if (sect_len < 0)
1972 		return;
1973 
1974 	new_len = (size_t)sect_len;
1975 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1976 	if (new_len < descsz) {
1977 		/*
1978 		 * It is expected that individual note emitters will correctly
1979 		 * predict their expected output size and fill up to that size
1980 		 * themselves, padding in a format-specific way if needed.
1981 		 * However, in case they don't, just do it here with zeros.
1982 		 */
1983 		for (i = 0; i < descsz - new_len; i++)
1984 			sbuf_putc(sb, 0);
1985 	} else if (new_len > descsz) {
1986 		/*
1987 		 * We can't always truncate sb -- we may have drained some
1988 		 * of it already.
1989 		 */
1990 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1991 		    "read it (%zu > %zu).  Since it is longer than "
1992 		    "expected, this coredump's notes are corrupt.  THIS "
1993 		    "IS A BUG in the note_procstat routine for type %u.\n",
1994 		    __func__, (unsigned)note.n_type, new_len, descsz,
1995 		    (unsigned)note.n_type));
1996 	}
1997 }
1998 
1999 /*
2000  * Miscellaneous note out functions.
2001  */
2002 
2003 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2004 #include <compat/freebsd32/freebsd32.h>
2005 #include <compat/freebsd32/freebsd32_signal.h>
2006 
2007 typedef struct prstatus32 elf_prstatus_t;
2008 typedef struct prpsinfo32 elf_prpsinfo_t;
2009 typedef struct fpreg32 elf_prfpregset_t;
2010 typedef struct fpreg32 elf_fpregset_t;
2011 typedef struct reg32 elf_gregset_t;
2012 typedef struct thrmisc32 elf_thrmisc_t;
2013 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2014 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2015 typedef uint32_t elf_ps_strings_t;
2016 #else
2017 typedef prstatus_t elf_prstatus_t;
2018 typedef prpsinfo_t elf_prpsinfo_t;
2019 typedef prfpregset_t elf_prfpregset_t;
2020 typedef prfpregset_t elf_fpregset_t;
2021 typedef gregset_t elf_gregset_t;
2022 typedef thrmisc_t elf_thrmisc_t;
2023 #define ELF_KERN_PROC_MASK	0
2024 typedef struct kinfo_proc elf_kinfo_proc_t;
2025 typedef vm_offset_t elf_ps_strings_t;
2026 #endif
2027 
2028 static void
2029 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2030 {
2031 	struct sbuf sbarg;
2032 	size_t len;
2033 	char *cp, *end;
2034 	struct proc *p;
2035 	elf_prpsinfo_t *psinfo;
2036 	int error;
2037 
2038 	p = (struct proc *)arg;
2039 	if (sb != NULL) {
2040 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2041 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2042 		psinfo->pr_version = PRPSINFO_VERSION;
2043 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2044 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2045 		PROC_LOCK(p);
2046 		if (p->p_args != NULL) {
2047 			len = sizeof(psinfo->pr_psargs) - 1;
2048 			if (len > p->p_args->ar_length)
2049 				len = p->p_args->ar_length;
2050 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2051 			PROC_UNLOCK(p);
2052 			error = 0;
2053 		} else {
2054 			_PHOLD(p);
2055 			PROC_UNLOCK(p);
2056 			sbuf_new(&sbarg, psinfo->pr_psargs,
2057 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2058 			error = proc_getargv(curthread, p, &sbarg);
2059 			PRELE(p);
2060 			if (sbuf_finish(&sbarg) == 0)
2061 				len = sbuf_len(&sbarg) - 1;
2062 			else
2063 				len = sizeof(psinfo->pr_psargs) - 1;
2064 			sbuf_delete(&sbarg);
2065 		}
2066 		if (error || len == 0)
2067 			strlcpy(psinfo->pr_psargs, p->p_comm,
2068 			    sizeof(psinfo->pr_psargs));
2069 		else {
2070 			KASSERT(len < sizeof(psinfo->pr_psargs),
2071 			    ("len is too long: %zu vs %zu", len,
2072 			    sizeof(psinfo->pr_psargs)));
2073 			cp = psinfo->pr_psargs;
2074 			end = cp + len - 1;
2075 			for (;;) {
2076 				cp = memchr(cp, '\0', end - cp);
2077 				if (cp == NULL)
2078 					break;
2079 				*cp = ' ';
2080 			}
2081 		}
2082 		psinfo->pr_pid = p->p_pid;
2083 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2084 		free(psinfo, M_TEMP);
2085 	}
2086 	*sizep = sizeof(*psinfo);
2087 }
2088 
2089 static void
2090 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2091 {
2092 	struct thread *td;
2093 	elf_prstatus_t *status;
2094 
2095 	td = (struct thread *)arg;
2096 	if (sb != NULL) {
2097 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2098 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2099 		status->pr_version = PRSTATUS_VERSION;
2100 		status->pr_statussz = sizeof(elf_prstatus_t);
2101 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2102 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2103 		status->pr_osreldate = osreldate;
2104 		status->pr_cursig = td->td_proc->p_sig;
2105 		status->pr_pid = td->td_tid;
2106 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2107 		fill_regs32(td, &status->pr_reg);
2108 #else
2109 		fill_regs(td, &status->pr_reg);
2110 #endif
2111 		sbuf_bcat(sb, status, sizeof(*status));
2112 		free(status, M_TEMP);
2113 	}
2114 	*sizep = sizeof(*status);
2115 }
2116 
2117 static void
2118 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2119 {
2120 	struct thread *td;
2121 	elf_prfpregset_t *fpregset;
2122 
2123 	td = (struct thread *)arg;
2124 	if (sb != NULL) {
2125 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2126 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2127 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2128 		fill_fpregs32(td, fpregset);
2129 #else
2130 		fill_fpregs(td, fpregset);
2131 #endif
2132 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2133 		free(fpregset, M_TEMP);
2134 	}
2135 	*sizep = sizeof(*fpregset);
2136 }
2137 
2138 static void
2139 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2140 {
2141 	struct thread *td;
2142 	elf_thrmisc_t thrmisc;
2143 
2144 	td = (struct thread *)arg;
2145 	if (sb != NULL) {
2146 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2147 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2148 		strcpy(thrmisc.pr_tname, td->td_name);
2149 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2150 	}
2151 	*sizep = sizeof(thrmisc);
2152 }
2153 
2154 static void
2155 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2156 {
2157 	struct thread *td;
2158 	size_t size;
2159 	int structsize;
2160 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2161 	struct ptrace_lwpinfo32 pl;
2162 #else
2163 	struct ptrace_lwpinfo pl;
2164 #endif
2165 
2166 	td = (struct thread *)arg;
2167 	size = sizeof(structsize) + sizeof(pl);
2168 	if (sb != NULL) {
2169 		KASSERT(*sizep == size, ("invalid size"));
2170 		structsize = sizeof(pl);
2171 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2172 		bzero(&pl, sizeof(pl));
2173 		pl.pl_lwpid = td->td_tid;
2174 		pl.pl_event = PL_EVENT_NONE;
2175 		pl.pl_sigmask = td->td_sigmask;
2176 		pl.pl_siglist = td->td_siglist;
2177 		if (td->td_si.si_signo != 0) {
2178 			pl.pl_event = PL_EVENT_SIGNAL;
2179 			pl.pl_flags |= PL_FLAG_SI;
2180 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2181 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2182 #else
2183 			pl.pl_siginfo = td->td_si;
2184 #endif
2185 		}
2186 		strcpy(pl.pl_tdname, td->td_name);
2187 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2188 		sbuf_bcat(sb, &pl, sizeof(pl));
2189 	}
2190 	*sizep = size;
2191 }
2192 
2193 /*
2194  * Allow for MD specific notes, as well as any MD
2195  * specific preparations for writing MI notes.
2196  */
2197 static void
2198 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2199 {
2200 	struct thread *td;
2201 	void *buf;
2202 	size_t size;
2203 
2204 	td = (struct thread *)arg;
2205 	size = *sizep;
2206 	if (size != 0 && sb != NULL)
2207 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2208 	else
2209 		buf = NULL;
2210 	size = 0;
2211 	__elfN(dump_thread)(td, buf, &size);
2212 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2213 	if (size != 0 && sb != NULL)
2214 		sbuf_bcat(sb, buf, size);
2215 	free(buf, M_TEMP);
2216 	*sizep = size;
2217 }
2218 
2219 #ifdef KINFO_PROC_SIZE
2220 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2221 #endif
2222 
2223 static void
2224 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2225 {
2226 	struct proc *p;
2227 	size_t size;
2228 	int structsize;
2229 
2230 	p = (struct proc *)arg;
2231 	size = sizeof(structsize) + p->p_numthreads *
2232 	    sizeof(elf_kinfo_proc_t);
2233 
2234 	if (sb != NULL) {
2235 		KASSERT(*sizep == size, ("invalid size"));
2236 		structsize = sizeof(elf_kinfo_proc_t);
2237 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2238 		PROC_LOCK(p);
2239 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2240 	}
2241 	*sizep = size;
2242 }
2243 
2244 #ifdef KINFO_FILE_SIZE
2245 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2246 #endif
2247 
2248 static void
2249 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2250 {
2251 	struct proc *p;
2252 	size_t size, sect_sz, i;
2253 	ssize_t start_len, sect_len;
2254 	int structsize, filedesc_flags;
2255 
2256 	if (coredump_pack_fileinfo)
2257 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2258 	else
2259 		filedesc_flags = 0;
2260 
2261 	p = (struct proc *)arg;
2262 	structsize = sizeof(struct kinfo_file);
2263 	if (sb == NULL) {
2264 		size = 0;
2265 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2266 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2267 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2268 		PROC_LOCK(p);
2269 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2270 		sbuf_finish(sb);
2271 		sbuf_delete(sb);
2272 		*sizep = size;
2273 	} else {
2274 		sbuf_start_section(sb, &start_len);
2275 
2276 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2277 		PROC_LOCK(p);
2278 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2279 		    filedesc_flags);
2280 
2281 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2282 		if (sect_len < 0)
2283 			return;
2284 		sect_sz = sect_len;
2285 
2286 		KASSERT(sect_sz <= *sizep,
2287 		    ("kern_proc_filedesc_out did not respect maxlen; "
2288 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2289 		     sect_sz - sizeof(structsize)));
2290 
2291 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2292 			sbuf_putc(sb, 0);
2293 	}
2294 }
2295 
2296 #ifdef KINFO_VMENTRY_SIZE
2297 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2298 #endif
2299 
2300 static void
2301 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2302 {
2303 	struct proc *p;
2304 	size_t size;
2305 	int structsize, vmmap_flags;
2306 
2307 	if (coredump_pack_vmmapinfo)
2308 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2309 	else
2310 		vmmap_flags = 0;
2311 
2312 	p = (struct proc *)arg;
2313 	structsize = sizeof(struct kinfo_vmentry);
2314 	if (sb == NULL) {
2315 		size = 0;
2316 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2317 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2318 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2319 		PROC_LOCK(p);
2320 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2321 		sbuf_finish(sb);
2322 		sbuf_delete(sb);
2323 		*sizep = size;
2324 	} else {
2325 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2326 		PROC_LOCK(p);
2327 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2328 		    vmmap_flags);
2329 	}
2330 }
2331 
2332 static void
2333 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2334 {
2335 	struct proc *p;
2336 	size_t size;
2337 	int structsize;
2338 
2339 	p = (struct proc *)arg;
2340 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2341 	if (sb != NULL) {
2342 		KASSERT(*sizep == size, ("invalid size"));
2343 		structsize = sizeof(gid_t);
2344 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2345 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2346 		    sizeof(gid_t));
2347 	}
2348 	*sizep = size;
2349 }
2350 
2351 static void
2352 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2353 {
2354 	struct proc *p;
2355 	size_t size;
2356 	int structsize;
2357 
2358 	p = (struct proc *)arg;
2359 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2360 	if (sb != NULL) {
2361 		KASSERT(*sizep == size, ("invalid size"));
2362 		structsize = sizeof(p->p_fd->fd_cmask);
2363 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2364 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2365 	}
2366 	*sizep = size;
2367 }
2368 
2369 static void
2370 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2371 {
2372 	struct proc *p;
2373 	struct rlimit rlim[RLIM_NLIMITS];
2374 	size_t size;
2375 	int structsize, i;
2376 
2377 	p = (struct proc *)arg;
2378 	size = sizeof(structsize) + sizeof(rlim);
2379 	if (sb != NULL) {
2380 		KASSERT(*sizep == size, ("invalid size"));
2381 		structsize = sizeof(rlim);
2382 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2383 		PROC_LOCK(p);
2384 		for (i = 0; i < RLIM_NLIMITS; i++)
2385 			lim_rlimit_proc(p, i, &rlim[i]);
2386 		PROC_UNLOCK(p);
2387 		sbuf_bcat(sb, rlim, sizeof(rlim));
2388 	}
2389 	*sizep = size;
2390 }
2391 
2392 static void
2393 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2394 {
2395 	struct proc *p;
2396 	size_t size;
2397 	int structsize;
2398 
2399 	p = (struct proc *)arg;
2400 	size = sizeof(structsize) + sizeof(p->p_osrel);
2401 	if (sb != NULL) {
2402 		KASSERT(*sizep == size, ("invalid size"));
2403 		structsize = sizeof(p->p_osrel);
2404 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2405 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2406 	}
2407 	*sizep = size;
2408 }
2409 
2410 static void
2411 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2412 {
2413 	struct proc *p;
2414 	elf_ps_strings_t ps_strings;
2415 	size_t size;
2416 	int structsize;
2417 
2418 	p = (struct proc *)arg;
2419 	size = sizeof(structsize) + sizeof(ps_strings);
2420 	if (sb != NULL) {
2421 		KASSERT(*sizep == size, ("invalid size"));
2422 		structsize = sizeof(ps_strings);
2423 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2424 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2425 #else
2426 		ps_strings = p->p_sysent->sv_psstrings;
2427 #endif
2428 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2429 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2430 	}
2431 	*sizep = size;
2432 }
2433 
2434 static void
2435 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2436 {
2437 	struct proc *p;
2438 	size_t size;
2439 	int structsize;
2440 
2441 	p = (struct proc *)arg;
2442 	if (sb == NULL) {
2443 		size = 0;
2444 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2445 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2446 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2447 		PHOLD(p);
2448 		proc_getauxv(curthread, p, sb);
2449 		PRELE(p);
2450 		sbuf_finish(sb);
2451 		sbuf_delete(sb);
2452 		*sizep = size;
2453 	} else {
2454 		structsize = sizeof(Elf_Auxinfo);
2455 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2456 		PHOLD(p);
2457 		proc_getauxv(curthread, p, sb);
2458 		PRELE(p);
2459 	}
2460 }
2461 
2462 static boolean_t
2463 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2464     const char *note_vendor, const Elf_Phdr *pnote,
2465     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2466 {
2467 	const Elf_Note *note, *note0, *note_end;
2468 	const char *note_name;
2469 	char *buf;
2470 	int i, error;
2471 	boolean_t res;
2472 
2473 	/* We need some limit, might as well use PAGE_SIZE. */
2474 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2475 		return (FALSE);
2476 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2477 	if (pnote->p_offset > PAGE_SIZE ||
2478 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2479 		VOP_UNLOCK(imgp->vp, 0);
2480 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2481 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2482 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2483 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2484 		    curthread->td_ucred, NOCRED, NULL, curthread);
2485 		if (error != 0) {
2486 			uprintf("i/o error PT_NOTE\n");
2487 			goto retf;
2488 		}
2489 		note = note0 = (const Elf_Note *)buf;
2490 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2491 	} else {
2492 		note = note0 = (const Elf_Note *)(imgp->image_header +
2493 		    pnote->p_offset);
2494 		note_end = (const Elf_Note *)(imgp->image_header +
2495 		    pnote->p_offset + pnote->p_filesz);
2496 		buf = NULL;
2497 	}
2498 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2499 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2500 		    (const char *)note < sizeof(Elf_Note)) {
2501 			goto retf;
2502 		}
2503 		if (note->n_namesz != checknote->n_namesz ||
2504 		    note->n_descsz != checknote->n_descsz ||
2505 		    note->n_type != checknote->n_type)
2506 			goto nextnote;
2507 		note_name = (const char *)(note + 1);
2508 		if (note_name + checknote->n_namesz >=
2509 		    (const char *)note_end || strncmp(note_vendor,
2510 		    note_name, checknote->n_namesz) != 0)
2511 			goto nextnote;
2512 
2513 		if (cb(note, cb_arg, &res))
2514 			goto ret;
2515 nextnote:
2516 		note = (const Elf_Note *)((const char *)(note + 1) +
2517 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2518 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2519 	}
2520 retf:
2521 	res = FALSE;
2522 ret:
2523 	free(buf, M_TEMP);
2524 	return (res);
2525 }
2526 
2527 struct brandnote_cb_arg {
2528 	Elf_Brandnote *brandnote;
2529 	int32_t *osrel;
2530 };
2531 
2532 static boolean_t
2533 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2534 {
2535 	struct brandnote_cb_arg *arg;
2536 
2537 	arg = arg0;
2538 
2539 	/*
2540 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2541 	 * necessary.
2542 	 */
2543 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2544 	    arg->brandnote->trans_osrel != NULL ?
2545 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2546 
2547 	return (TRUE);
2548 }
2549 
2550 static Elf_Note fctl_note = {
2551 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2552 	.n_descsz = sizeof(uint32_t),
2553 	.n_type = NT_FREEBSD_FEATURE_CTL,
2554 };
2555 
2556 struct fctl_cb_arg {
2557 	uint32_t *fctl0;
2558 };
2559 
2560 static boolean_t
2561 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2562 {
2563 	struct fctl_cb_arg *arg;
2564 	const Elf32_Word *desc;
2565 	uintptr_t p;
2566 
2567 	arg = arg0;
2568 	p = (uintptr_t)(note + 1);
2569 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2570 	desc = (const Elf32_Word *)p;
2571 	*arg->fctl0 = desc[0];
2572 	return (TRUE);
2573 }
2574 
2575 /*
2576  * Try to find the appropriate ABI-note section for checknote, fetch
2577  * the osreldate and feature control flags for binary from the ELF
2578  * OSABI-note.  Only the first page of the image is searched, the same
2579  * as for headers.
2580  */
2581 static boolean_t
2582 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2583     int32_t *osrel, uint32_t *fctl0)
2584 {
2585 	const Elf_Phdr *phdr;
2586 	const Elf_Ehdr *hdr;
2587 	struct brandnote_cb_arg b_arg;
2588 	struct fctl_cb_arg f_arg;
2589 	int i, j;
2590 
2591 	hdr = (const Elf_Ehdr *)imgp->image_header;
2592 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2593 	b_arg.brandnote = brandnote;
2594 	b_arg.osrel = osrel;
2595 	f_arg.fctl0 = fctl0;
2596 
2597 	for (i = 0; i < hdr->e_phnum; i++) {
2598 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2599 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2600 		    &b_arg)) {
2601 			for (j = 0; j < hdr->e_phnum; j++) {
2602 				if (phdr[j].p_type == PT_NOTE &&
2603 				    __elfN(parse_notes)(imgp, &fctl_note,
2604 				    FREEBSD_ABI_VENDOR, &phdr[j],
2605 				    note_fctl_cb, &f_arg))
2606 					break;
2607 			}
2608 			return (TRUE);
2609 		}
2610 	}
2611 	return (FALSE);
2612 
2613 }
2614 
2615 /*
2616  * Tell kern_execve.c about it, with a little help from the linker.
2617  */
2618 static struct execsw __elfN(execsw) = {
2619 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2620 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2621 };
2622 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2623 
2624 static vm_prot_t
2625 __elfN(trans_prot)(Elf_Word flags)
2626 {
2627 	vm_prot_t prot;
2628 
2629 	prot = 0;
2630 	if (flags & PF_X)
2631 		prot |= VM_PROT_EXECUTE;
2632 	if (flags & PF_W)
2633 		prot |= VM_PROT_WRITE;
2634 	if (flags & PF_R)
2635 		prot |= VM_PROT_READ;
2636 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2637 	if (i386_read_exec && (flags & PF_R))
2638 		prot |= VM_PROT_EXECUTE;
2639 #endif
2640 	return (prot);
2641 }
2642 
2643 static Elf_Word
2644 __elfN(untrans_prot)(vm_prot_t prot)
2645 {
2646 	Elf_Word flags;
2647 
2648 	flags = 0;
2649 	if (prot & VM_PROT_EXECUTE)
2650 		flags |= PF_X;
2651 	if (prot & VM_PROT_READ)
2652 		flags |= PF_R;
2653 	if (prot & VM_PROT_WRITE)
2654 		flags |= PF_W;
2655 	return (flags);
2656 }
2657