xref: /freebsd/sys/kern/imgact_elf.c (revision 123af6ec70016f5556da5972d4d63c7d175c06d3)
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	trunc_page_ps(va, ps)	rounddown2(va, ps)
162 #define	round_page_ps(va, ps)	roundup2(va, ps)
163 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
164 
165 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
166 
167 Elf_Brandnote __elfN(freebsd_brandnote) = {
168 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
169 	.hdr.n_descsz	= sizeof(int32_t),
170 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
171 	.vendor		= FREEBSD_ABI_VENDOR,
172 	.flags		= BN_TRANSLATE_OSREL,
173 	.trans_osrel	= __elfN(freebsd_trans_osrel)
174 };
175 
176 static bool
177 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
178 {
179 	uintptr_t p;
180 
181 	p = (uintptr_t)(note + 1);
182 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
183 	*osrel = *(const int32_t *)(p);
184 
185 	return (true);
186 }
187 
188 static const char GNU_ABI_VENDOR[] = "GNU";
189 static int GNU_KFREEBSD_ABI_DESC = 3;
190 
191 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
192 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
193 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
194 	.hdr.n_type	= 1,
195 	.vendor		= GNU_ABI_VENDOR,
196 	.flags		= BN_TRANSLATE_OSREL,
197 	.trans_osrel	= kfreebsd_trans_osrel
198 };
199 
200 static bool
201 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
202 {
203 	const Elf32_Word *desc;
204 	uintptr_t p;
205 
206 	p = (uintptr_t)(note + 1);
207 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
208 
209 	desc = (const Elf32_Word *)p;
210 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
211 		return (false);
212 
213 	/*
214 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
215 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
216 	 */
217 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
218 
219 	return (true);
220 }
221 
222 int
223 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
224 {
225 	int i;
226 
227 	for (i = 0; i < MAX_BRANDS; i++) {
228 		if (elf_brand_list[i] == NULL) {
229 			elf_brand_list[i] = entry;
230 			break;
231 		}
232 	}
233 	if (i == MAX_BRANDS) {
234 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
235 			__func__, entry);
236 		return (-1);
237 	}
238 	return (0);
239 }
240 
241 int
242 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
243 {
244 	int i;
245 
246 	for (i = 0; i < MAX_BRANDS; i++) {
247 		if (elf_brand_list[i] == entry) {
248 			elf_brand_list[i] = NULL;
249 			break;
250 		}
251 	}
252 	if (i == MAX_BRANDS)
253 		return (-1);
254 	return (0);
255 }
256 
257 int
258 __elfN(brand_inuse)(Elf_Brandinfo *entry)
259 {
260 	struct proc *p;
261 	int rval = FALSE;
262 
263 	sx_slock(&allproc_lock);
264 	FOREACH_PROC_IN_SYSTEM(p) {
265 		if (p->p_sysent == entry->sysvec) {
266 			rval = TRUE;
267 			break;
268 		}
269 	}
270 	sx_sunlock(&allproc_lock);
271 
272 	return (rval);
273 }
274 
275 static Elf_Brandinfo *
276 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
277     int interp_name_len, int32_t *osrel, uint32_t *fctl0)
278 {
279 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
280 	Elf_Brandinfo *bi, *bi_m;
281 	boolean_t ret;
282 	int i;
283 
284 	/*
285 	 * We support four types of branding -- (1) the ELF EI_OSABI field
286 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
287 	 * branding w/in the ELF header, (3) path of the `interp_path'
288 	 * field, and (4) the ".note.ABI-tag" ELF section.
289 	 */
290 
291 	/* Look for an ".note.ABI-tag" ELF section */
292 	bi_m = NULL;
293 	for (i = 0; i < MAX_BRANDS; i++) {
294 		bi = elf_brand_list[i];
295 		if (bi == NULL)
296 			continue;
297 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
298 			continue;
299 		if (hdr->e_machine == bi->machine && (bi->flags &
300 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
301 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
302 			    fctl0);
303 			/* Give brand a chance to veto check_note's guess */
304 			if (ret && bi->header_supported)
305 				ret = bi->header_supported(imgp);
306 			/*
307 			 * If note checker claimed the binary, but the
308 			 * interpreter path in the image does not
309 			 * match default one for the brand, try to
310 			 * search for other brands with the same
311 			 * interpreter.  Either there is better brand
312 			 * with the right interpreter, or, failing
313 			 * this, we return first brand which accepted
314 			 * our note and, optionally, header.
315 			 */
316 			if (ret && bi_m == NULL && interp != NULL &&
317 			    (bi->interp_path == NULL ||
318 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
319 			    strncmp(interp, bi->interp_path, interp_name_len)
320 			    != 0))) {
321 				bi_m = bi;
322 				ret = 0;
323 			}
324 			if (ret)
325 				return (bi);
326 		}
327 	}
328 	if (bi_m != NULL)
329 		return (bi_m);
330 
331 	/* If the executable has a brand, search for it in the brand list. */
332 	for (i = 0; i < MAX_BRANDS; i++) {
333 		bi = elf_brand_list[i];
334 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
335 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
336 			continue;
337 		if (hdr->e_machine == bi->machine &&
338 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
339 		    (bi->compat_3_brand != NULL &&
340 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
341 		    bi->compat_3_brand) == 0))) {
342 			/* Looks good, but give brand a chance to veto */
343 			if (bi->header_supported == NULL ||
344 			    bi->header_supported(imgp)) {
345 				/*
346 				 * Again, prefer strictly matching
347 				 * interpreter path.
348 				 */
349 				if (interp_name_len == 0 &&
350 				    bi->interp_path == NULL)
351 					return (bi);
352 				if (bi->interp_path != NULL &&
353 				    strlen(bi->interp_path) + 1 ==
354 				    interp_name_len && strncmp(interp,
355 				    bi->interp_path, interp_name_len) == 0)
356 					return (bi);
357 				if (bi_m == NULL)
358 					bi_m = bi;
359 			}
360 		}
361 	}
362 	if (bi_m != NULL)
363 		return (bi_m);
364 
365 	/* No known brand, see if the header is recognized by any brand */
366 	for (i = 0; i < MAX_BRANDS; i++) {
367 		bi = elf_brand_list[i];
368 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
369 		    bi->header_supported == NULL)
370 			continue;
371 		if (hdr->e_machine == bi->machine) {
372 			ret = bi->header_supported(imgp);
373 			if (ret)
374 				return (bi);
375 		}
376 	}
377 
378 	/* Lacking a known brand, search for a recognized interpreter. */
379 	if (interp != NULL) {
380 		for (i = 0; i < MAX_BRANDS; i++) {
381 			bi = elf_brand_list[i];
382 			if (bi == NULL || (bi->flags &
383 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
384 			    != 0)
385 				continue;
386 			if (hdr->e_machine == bi->machine &&
387 			    bi->interp_path != NULL &&
388 			    /* ELF image p_filesz includes terminating zero */
389 			    strlen(bi->interp_path) + 1 == interp_name_len &&
390 			    strncmp(interp, bi->interp_path, interp_name_len)
391 			    == 0 && (bi->header_supported == NULL ||
392 			    bi->header_supported(imgp)))
393 				return (bi);
394 		}
395 	}
396 
397 	/* Lacking a recognized interpreter, try the default brand */
398 	for (i = 0; i < MAX_BRANDS; i++) {
399 		bi = elf_brand_list[i];
400 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
401 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
402 			continue;
403 		if (hdr->e_machine == bi->machine &&
404 		    __elfN(fallback_brand) == bi->brand &&
405 		    (bi->header_supported == NULL ||
406 		    bi->header_supported(imgp)))
407 			return (bi);
408 	}
409 	return (NULL);
410 }
411 
412 static int
413 __elfN(check_header)(const Elf_Ehdr *hdr)
414 {
415 	Elf_Brandinfo *bi;
416 	int i;
417 
418 	if (!IS_ELF(*hdr) ||
419 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
420 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
421 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
422 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
423 	    hdr->e_version != ELF_TARG_VER)
424 		return (ENOEXEC);
425 
426 	/*
427 	 * Make sure we have at least one brand for this machine.
428 	 */
429 
430 	for (i = 0; i < MAX_BRANDS; i++) {
431 		bi = elf_brand_list[i];
432 		if (bi != NULL && bi->machine == hdr->e_machine)
433 			break;
434 	}
435 	if (i == MAX_BRANDS)
436 		return (ENOEXEC);
437 
438 	return (0);
439 }
440 
441 static int
442 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
443     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
444 {
445 	struct sf_buf *sf;
446 	int error;
447 	vm_offset_t off;
448 
449 	/*
450 	 * Create the page if it doesn't exist yet. Ignore errors.
451 	 */
452 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
453 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
454 
455 	/*
456 	 * Find the page from the underlying object.
457 	 */
458 	if (object != NULL) {
459 		sf = vm_imgact_map_page(object, offset);
460 		if (sf == NULL)
461 			return (KERN_FAILURE);
462 		off = offset - trunc_page(offset);
463 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
464 		    end - start);
465 		vm_imgact_unmap_page(sf);
466 		if (error != 0)
467 			return (KERN_FAILURE);
468 	}
469 
470 	return (KERN_SUCCESS);
471 }
472 
473 static int
474 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
475     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
476     int cow)
477 {
478 	struct sf_buf *sf;
479 	vm_offset_t off;
480 	vm_size_t sz;
481 	int error, locked, rv;
482 
483 	if (start != trunc_page(start)) {
484 		rv = __elfN(map_partial)(map, object, offset, start,
485 		    round_page(start), prot);
486 		if (rv != KERN_SUCCESS)
487 			return (rv);
488 		offset += round_page(start) - start;
489 		start = round_page(start);
490 	}
491 	if (end != round_page(end)) {
492 		rv = __elfN(map_partial)(map, object, offset +
493 		    trunc_page(end) - start, trunc_page(end), end, prot);
494 		if (rv != KERN_SUCCESS)
495 			return (rv);
496 		end = trunc_page(end);
497 	}
498 	if (start >= end)
499 		return (KERN_SUCCESS);
500 	if ((offset & PAGE_MASK) != 0) {
501 		/*
502 		 * The mapping is not page aligned.  This means that we have
503 		 * to copy the data.
504 		 */
505 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
506 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
507 		if (rv != KERN_SUCCESS)
508 			return (rv);
509 		if (object == NULL)
510 			return (KERN_SUCCESS);
511 		for (; start < end; start += sz) {
512 			sf = vm_imgact_map_page(object, offset);
513 			if (sf == NULL)
514 				return (KERN_FAILURE);
515 			off = offset - trunc_page(offset);
516 			sz = end - start;
517 			if (sz > PAGE_SIZE - off)
518 				sz = PAGE_SIZE - off;
519 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
520 			    (caddr_t)start, sz);
521 			vm_imgact_unmap_page(sf);
522 			if (error != 0)
523 				return (KERN_FAILURE);
524 			offset += sz;
525 		}
526 	} else {
527 		vm_object_reference(object);
528 		rv = vm_map_fixed(map, object, offset, start, end - start,
529 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
530 		if (rv != KERN_SUCCESS) {
531 			locked = VOP_ISLOCKED(imgp->vp);
532 			VOP_UNLOCK(imgp->vp, 0);
533 			vm_object_deallocate(object);
534 			vn_lock(imgp->vp, locked | LK_RETRY);
535 			return (rv);
536 		}
537 	}
538 	return (KERN_SUCCESS);
539 }
540 
541 static int
542 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
543     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
544 {
545 	struct sf_buf *sf;
546 	size_t map_len;
547 	vm_map_t map;
548 	vm_object_t object;
549 	vm_offset_t off, map_addr;
550 	int error, rv, cow;
551 	size_t copy_len;
552 	vm_ooffset_t file_addr;
553 
554 	/*
555 	 * It's necessary to fail if the filsz + offset taken from the
556 	 * header is greater than the actual file pager object's size.
557 	 * If we were to allow this, then the vm_map_find() below would
558 	 * walk right off the end of the file object and into the ether.
559 	 *
560 	 * While I'm here, might as well check for something else that
561 	 * is invalid: filsz cannot be greater than memsz.
562 	 */
563 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
564 	    filsz > memsz) {
565 		uprintf("elf_load_section: truncated ELF file\n");
566 		return (ENOEXEC);
567 	}
568 
569 	object = imgp->object;
570 	map = &imgp->proc->p_vmspace->vm_map;
571 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, PAGE_SIZE);
572 	file_addr = trunc_page_ps(offset, PAGE_SIZE);
573 
574 	/*
575 	 * We have two choices.  We can either clear the data in the last page
576 	 * of an oversized mapping, or we can start the anon mapping a page
577 	 * early and copy the initialized data into that first page.  We
578 	 * choose the second.
579 	 */
580 	if (filsz == 0)
581 		map_len = 0;
582 	else if (memsz > filsz)
583 		map_len = trunc_page_ps(offset + filsz, PAGE_SIZE) - file_addr;
584 	else
585 		map_len = round_page_ps(offset + filsz, PAGE_SIZE) - file_addr;
586 
587 	if (map_len != 0) {
588 		/* cow flags: don't dump readonly sections in core */
589 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
590 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
591 
592 		rv = __elfN(map_insert)(imgp, map,
593 				      object,
594 				      file_addr,	/* file offset */
595 				      map_addr,		/* virtual start */
596 				      map_addr + map_len,/* virtual end */
597 				      prot,
598 				      cow);
599 		if (rv != KERN_SUCCESS)
600 			return (EINVAL);
601 
602 		/* we can stop now if we've covered it all */
603 		if (memsz == filsz)
604 			return (0);
605 	}
606 
607 
608 	/*
609 	 * We have to get the remaining bit of the file into the first part
610 	 * of the oversized map segment.  This is normally because the .data
611 	 * segment in the file is extended to provide bss.  It's a neat idea
612 	 * to try and save a page, but it's a pain in the behind to implement.
613 	 */
614 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
615 	    filsz, PAGE_SIZE);
616 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, PAGE_SIZE);
617 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, PAGE_SIZE) -
618 	    map_addr;
619 
620 	/* This had damn well better be true! */
621 	if (map_len != 0) {
622 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
623 		    map_addr + map_len, prot, 0);
624 		if (rv != KERN_SUCCESS)
625 			return (EINVAL);
626 	}
627 
628 	if (copy_len != 0) {
629 		sf = vm_imgact_map_page(object, offset + filsz);
630 		if (sf == NULL)
631 			return (EIO);
632 
633 		/* send the page fragment to user space */
634 		off = trunc_page_ps(offset + filsz, PAGE_SIZE) -
635 		    trunc_page(offset + filsz);
636 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
637 		    (caddr_t)map_addr, copy_len);
638 		vm_imgact_unmap_page(sf);
639 		if (error != 0)
640 			return (error);
641 	}
642 
643 	/*
644 	 * Remove write access to the page if it was only granted by map_insert
645 	 * to allow copyout.
646 	 */
647 	if ((prot & VM_PROT_WRITE) == 0)
648 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
649 		    map_len), prot, FALSE);
650 
651 	return (0);
652 }
653 
654 /*
655  * Load the file "file" into memory.  It may be either a shared object
656  * or an executable.
657  *
658  * The "addr" reference parameter is in/out.  On entry, it specifies
659  * the address where a shared object should be loaded.  If the file is
660  * an executable, this value is ignored.  On exit, "addr" specifies
661  * where the file was actually loaded.
662  *
663  * The "entry" reference parameter is out only.  On exit, it specifies
664  * the entry point for the loaded file.
665  */
666 static int
667 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
668 	u_long *entry)
669 {
670 	struct {
671 		struct nameidata nd;
672 		struct vattr attr;
673 		struct image_params image_params;
674 	} *tempdata;
675 	const Elf_Ehdr *hdr = NULL;
676 	const Elf_Phdr *phdr = NULL;
677 	struct nameidata *nd;
678 	struct vattr *attr;
679 	struct image_params *imgp;
680 	vm_prot_t prot;
681 	u_long rbase;
682 	u_long base_addr = 0;
683 	int error, i, numsegs;
684 
685 #ifdef CAPABILITY_MODE
686 	/*
687 	 * XXXJA: This check can go away once we are sufficiently confident
688 	 * that the checks in namei() are correct.
689 	 */
690 	if (IN_CAPABILITY_MODE(curthread))
691 		return (ECAPMODE);
692 #endif
693 
694 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
695 	nd = &tempdata->nd;
696 	attr = &tempdata->attr;
697 	imgp = &tempdata->image_params;
698 
699 	/*
700 	 * Initialize part of the common data
701 	 */
702 	imgp->proc = p;
703 	imgp->attr = attr;
704 	imgp->firstpage = NULL;
705 	imgp->image_header = NULL;
706 	imgp->object = NULL;
707 	imgp->execlabel = NULL;
708 
709 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
710 	if ((error = namei(nd)) != 0) {
711 		nd->ni_vp = NULL;
712 		goto fail;
713 	}
714 	NDFREE(nd, NDF_ONLY_PNBUF);
715 	imgp->vp = nd->ni_vp;
716 
717 	/*
718 	 * Check permissions, modes, uid, etc on the file, and "open" it.
719 	 */
720 	error = exec_check_permissions(imgp);
721 	if (error)
722 		goto fail;
723 
724 	error = exec_map_first_page(imgp);
725 	if (error)
726 		goto fail;
727 
728 	/*
729 	 * Also make certain that the interpreter stays the same, so set
730 	 * its VV_TEXT flag, too.
731 	 */
732 	VOP_SET_TEXT(nd->ni_vp);
733 
734 	imgp->object = nd->ni_vp->v_object;
735 
736 	hdr = (const Elf_Ehdr *)imgp->image_header;
737 	if ((error = __elfN(check_header)(hdr)) != 0)
738 		goto fail;
739 	if (hdr->e_type == ET_DYN)
740 		rbase = *addr;
741 	else if (hdr->e_type == ET_EXEC)
742 		rbase = 0;
743 	else {
744 		error = ENOEXEC;
745 		goto fail;
746 	}
747 
748 	/* Only support headers that fit within first page for now      */
749 	if ((hdr->e_phoff > PAGE_SIZE) ||
750 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
751 		error = ENOEXEC;
752 		goto fail;
753 	}
754 
755 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
756 	if (!aligned(phdr, Elf_Addr)) {
757 		error = ENOEXEC;
758 		goto fail;
759 	}
760 
761 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
762 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
763 			/* Loadable segment */
764 			prot = __elfN(trans_prot)(phdr[i].p_flags);
765 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
766 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
767 			    phdr[i].p_memsz, phdr[i].p_filesz, prot);
768 			if (error != 0)
769 				goto fail;
770 			/*
771 			 * Establish the base address if this is the
772 			 * first segment.
773 			 */
774 			if (numsegs == 0)
775   				base_addr = trunc_page(phdr[i].p_vaddr +
776 				    rbase);
777 			numsegs++;
778 		}
779 	}
780 	*addr = base_addr;
781 	*entry = (unsigned long)hdr->e_entry + rbase;
782 
783 fail:
784 	if (imgp->firstpage)
785 		exec_unmap_first_page(imgp);
786 
787 	if (nd->ni_vp)
788 		vput(nd->ni_vp);
789 
790 	free(tempdata, M_TEMP);
791 
792 	return (error);
793 }
794 
795 static u_long
796 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
797     u_int align)
798 {
799 	u_long rbase, res;
800 
801 	MPASS(vm_map_min(map) <= minv);
802 	MPASS(maxv <= vm_map_max(map));
803 	MPASS(minv < maxv);
804 	MPASS(minv + align < maxv);
805 	arc4rand(&rbase, sizeof(rbase), 0);
806 	res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
807 	res &= ~((u_long)align - 1);
808 	if (res >= maxv)
809 		res -= align;
810 	KASSERT(res >= minv,
811 	    ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
812 	    res, minv, maxv, rbase));
813 	KASSERT(res < maxv,
814 	    ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
815 	    res, maxv, minv, rbase));
816 	return (res);
817 }
818 
819 /*
820  * Impossible et_dyn_addr initial value indicating that the real base
821  * must be calculated later with some randomization applied.
822  */
823 #define	ET_DYN_ADDR_RAND	1
824 
825 static int
826 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
827 {
828 	struct thread *td;
829 	const Elf_Ehdr *hdr;
830 	const Elf_Phdr *phdr;
831 	Elf_Auxargs *elf_auxargs;
832 	struct vmspace *vmspace;
833 	vm_map_t map;
834 	const char *err_str, *newinterp;
835 	char *interp, *interp_buf, *path;
836 	Elf_Brandinfo *brand_info;
837 	struct sysentvec *sv;
838 	vm_prot_t prot;
839 	u_long text_size, data_size, total_size, text_addr, data_addr;
840 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
841 	u_long maxalign, mapsz, maxv, maxv1;
842 	uint32_t fctl0;
843 	int32_t osrel;
844 	int error, i, n, interp_name_len, have_interp;
845 
846 	hdr = (const Elf_Ehdr *)imgp->image_header;
847 
848 	/*
849 	 * Do we have a valid ELF header ?
850 	 *
851 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
852 	 * if particular brand doesn't support it.
853 	 */
854 	if (__elfN(check_header)(hdr) != 0 ||
855 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
856 		return (-1);
857 
858 	/*
859 	 * From here on down, we return an errno, not -1, as we've
860 	 * detected an ELF file.
861 	 */
862 
863 	if ((hdr->e_phoff > PAGE_SIZE) ||
864 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
865 		/* Only support headers in first page for now */
866 		uprintf("Program headers not in the first page\n");
867 		return (ENOEXEC);
868 	}
869 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
870 	if (!aligned(phdr, Elf_Addr)) {
871 		uprintf("Unaligned program headers\n");
872 		return (ENOEXEC);
873 	}
874 
875 	n = error = 0;
876 	baddr = 0;
877 	osrel = 0;
878 	fctl0 = 0;
879 	text_size = data_size = total_size = text_addr = data_addr = 0;
880 	entry = proghdr = 0;
881 	interp_name_len = 0;
882 	err_str = newinterp = NULL;
883 	interp = interp_buf = NULL;
884 	td = curthread;
885 	maxalign = PAGE_SIZE;
886 	mapsz = 0;
887 
888 	for (i = 0; i < hdr->e_phnum; i++) {
889 		switch (phdr[i].p_type) {
890 		case PT_LOAD:
891 			if (n == 0)
892 				baddr = phdr[i].p_vaddr;
893 			if (phdr[i].p_align > maxalign)
894 				maxalign = phdr[i].p_align;
895 			mapsz += phdr[i].p_memsz;
896 			n++;
897 			break;
898 		case PT_INTERP:
899 			/* Path to interpreter */
900 			if (phdr[i].p_filesz < 2 ||
901 			    phdr[i].p_filesz > MAXPATHLEN) {
902 				uprintf("Invalid PT_INTERP\n");
903 				error = ENOEXEC;
904 				goto ret;
905 			}
906 			if (interp != NULL) {
907 				uprintf("Multiple PT_INTERP headers\n");
908 				error = ENOEXEC;
909 				goto ret;
910 			}
911 			interp_name_len = phdr[i].p_filesz;
912 			if (phdr[i].p_offset > PAGE_SIZE ||
913 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
914 				VOP_UNLOCK(imgp->vp, 0);
915 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
916 				    M_WAITOK);
917 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
918 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
919 				    interp_name_len, phdr[i].p_offset,
920 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
921 				    NOCRED, NULL, td);
922 				if (error != 0) {
923 					uprintf("i/o error PT_INTERP %d\n",
924 					    error);
925 					goto ret;
926 				}
927 				interp_buf[interp_name_len] = '\0';
928 				interp = interp_buf;
929 			} else {
930 				interp = __DECONST(char *, imgp->image_header) +
931 				    phdr[i].p_offset;
932 				if (interp[interp_name_len - 1] != '\0') {
933 					uprintf("Invalid PT_INTERP\n");
934 					error = ENOEXEC;
935 					goto ret;
936 				}
937 			}
938 			break;
939 		case PT_GNU_STACK:
940 			if (__elfN(nxstack))
941 				imgp->stack_prot =
942 				    __elfN(trans_prot)(phdr[i].p_flags);
943 			imgp->stack_sz = phdr[i].p_memsz;
944 			break;
945 		}
946 	}
947 
948 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
949 	    &osrel, &fctl0);
950 	if (brand_info == NULL) {
951 		uprintf("ELF binary type \"%u\" not known.\n",
952 		    hdr->e_ident[EI_OSABI]);
953 		error = ENOEXEC;
954 		goto ret;
955 	}
956 	sv = brand_info->sysvec;
957 	et_dyn_addr = 0;
958 	if (hdr->e_type == ET_DYN) {
959 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
960 			uprintf("Cannot execute shared object\n");
961 			error = ENOEXEC;
962 			goto ret;
963 		}
964 		/*
965 		 * Honour the base load address from the dso if it is
966 		 * non-zero for some reason.
967 		 */
968 		if (baddr == 0) {
969 			if ((sv->sv_flags & SV_ASLR) == 0 ||
970 			    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
971 				et_dyn_addr = ET_DYN_LOAD_ADDR;
972 			else if ((__elfN(pie_aslr_enabled) &&
973 			    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
974 			    (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
975 				et_dyn_addr = ET_DYN_ADDR_RAND;
976 			else
977 				et_dyn_addr = ET_DYN_LOAD_ADDR;
978 		}
979 	}
980 	if (interp != NULL && brand_info->interp_newpath != NULL)
981 		newinterp = brand_info->interp_newpath;
982 
983 	/*
984 	 * Avoid a possible deadlock if the current address space is destroyed
985 	 * and that address space maps the locked vnode.  In the common case,
986 	 * the locked vnode's v_usecount is decremented but remains greater
987 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
988 	 * However, in cases where the vnode lock is external, such as nullfs,
989 	 * v_usecount may become zero.
990 	 *
991 	 * The VV_TEXT flag prevents modifications to the executable while
992 	 * the vnode is unlocked.
993 	 */
994 	VOP_UNLOCK(imgp->vp, 0);
995 
996 	/*
997 	 * Decide whether to enable randomization of user mappings.
998 	 * First, reset user preferences for the setid binaries.
999 	 * Then, account for the support of the randomization by the
1000 	 * ABI, by user preferences, and make special treatment for
1001 	 * PIE binaries.
1002 	 */
1003 	if (imgp->credential_setid) {
1004 		PROC_LOCK(imgp->proc);
1005 		imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1006 		PROC_UNLOCK(imgp->proc);
1007 	}
1008 	if ((sv->sv_flags & SV_ASLR) == 0 ||
1009 	    (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1010 	    (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1011 		KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1012 		    ("et_dyn_addr == RAND and !ASLR"));
1013 	} else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1014 	    (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1015 	    et_dyn_addr == ET_DYN_ADDR_RAND) {
1016 		imgp->map_flags |= MAP_ASLR;
1017 		/*
1018 		 * If user does not care about sbrk, utilize the bss
1019 		 * grow region for mappings as well.  We can select
1020 		 * the base for the image anywere and still not suffer
1021 		 * from the fragmentation.
1022 		 */
1023 		if (!__elfN(aslr_honor_sbrk) ||
1024 		    (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1025 			imgp->map_flags |= MAP_ASLR_IGNSTART;
1026 	}
1027 
1028 	error = exec_new_vmspace(imgp, sv);
1029 	vmspace = imgp->proc->p_vmspace;
1030 	map = &vmspace->vm_map;
1031 
1032 	imgp->proc->p_sysent = sv;
1033 
1034 	maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1035 	if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1036 		KASSERT((map->flags & MAP_ASLR) != 0,
1037 		    ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1038 		et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1039 		    vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1040 		    /* reserve half of the address space to interpreter */
1041 		    maxv / 2, 1UL << flsl(maxalign));
1042 	}
1043 
1044 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1045 	if (error != 0)
1046 		goto ret;
1047 
1048 	for (i = 0; i < hdr->e_phnum; i++) {
1049 		switch (phdr[i].p_type) {
1050 		case PT_LOAD:	/* Loadable segment */
1051 			if (phdr[i].p_memsz == 0)
1052 				break;
1053 			prot = __elfN(trans_prot)(phdr[i].p_flags);
1054 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
1055 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
1056 			    phdr[i].p_memsz, phdr[i].p_filesz, prot);
1057 			if (error != 0)
1058 				goto ret;
1059 
1060 			/*
1061 			 * If this segment contains the program headers,
1062 			 * remember their virtual address for the AT_PHDR
1063 			 * aux entry. Static binaries don't usually include
1064 			 * a PT_PHDR entry.
1065 			 */
1066 			if (phdr[i].p_offset == 0 &&
1067 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1068 				<= phdr[i].p_filesz)
1069 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
1070 				    et_dyn_addr;
1071 
1072 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
1073 			seg_size = round_page(phdr[i].p_memsz +
1074 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
1075 
1076 			/*
1077 			 * Make the largest executable segment the official
1078 			 * text segment and all others data.
1079 			 *
1080 			 * Note that obreak() assumes that data_addr +
1081 			 * data_size == end of data load area, and the ELF
1082 			 * file format expects segments to be sorted by
1083 			 * address.  If multiple data segments exist, the
1084 			 * last one will be used.
1085 			 */
1086 
1087 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
1088 				text_size = seg_size;
1089 				text_addr = seg_addr;
1090 			} else {
1091 				data_size = seg_size;
1092 				data_addr = seg_addr;
1093 			}
1094 			total_size += seg_size;
1095 			break;
1096 		case PT_PHDR: 	/* Program header table info */
1097 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
1098 			break;
1099 		default:
1100 			break;
1101 		}
1102 	}
1103 
1104 	if (data_addr == 0 && data_size == 0) {
1105 		data_addr = text_addr;
1106 		data_size = text_size;
1107 	}
1108 
1109 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1110 
1111 	/*
1112 	 * Check limits.  It should be safe to check the
1113 	 * limits after loading the segments since we do
1114 	 * not actually fault in all the segments pages.
1115 	 */
1116 	PROC_LOCK(imgp->proc);
1117 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1118 		err_str = "Data segment size exceeds process limit";
1119 	else if (text_size > maxtsiz)
1120 		err_str = "Text segment size exceeds system limit";
1121 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1122 		err_str = "Total segment size exceeds process limit";
1123 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1124 		err_str = "Data segment size exceeds resource limit";
1125 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1126 		err_str = "Total segment size exceeds resource limit";
1127 	if (err_str != NULL) {
1128 		PROC_UNLOCK(imgp->proc);
1129 		uprintf("%s\n", err_str);
1130 		error = ENOMEM;
1131 		goto ret;
1132 	}
1133 
1134 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1135 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1136 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1137 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1138 
1139 	/*
1140 	 * We load the dynamic linker where a userland call
1141 	 * to mmap(0, ...) would put it.  The rationale behind this
1142 	 * calculation is that it leaves room for the heap to grow to
1143 	 * its maximum allowed size.
1144 	 */
1145 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1146 	    RLIMIT_DATA));
1147 	if ((map->flags & MAP_ASLR) != 0) {
1148 		maxv1 = maxv / 2 + addr / 2;
1149 		MPASS(maxv1 >= addr);	/* No overflow */
1150 		map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1151 		    MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1152 	} else {
1153 		map->anon_loc = addr;
1154 	}
1155 	PROC_UNLOCK(imgp->proc);
1156 
1157 	imgp->entry_addr = entry;
1158 
1159 	if (interp != NULL) {
1160 		have_interp = FALSE;
1161 		VOP_UNLOCK(imgp->vp, 0);
1162 		if ((map->flags & MAP_ASLR) != 0) {
1163 			/* Assume that interpeter fits into 1/4 of AS */
1164 			maxv1 = maxv / 2 + addr / 2;
1165 			MPASS(maxv1 >= addr);	/* No overflow */
1166 			addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1167 			    maxv1, PAGE_SIZE);
1168 		}
1169 		if (brand_info->emul_path != NULL &&
1170 		    brand_info->emul_path[0] != '\0') {
1171 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1172 			snprintf(path, MAXPATHLEN, "%s%s",
1173 			    brand_info->emul_path, interp);
1174 			error = __elfN(load_file)(imgp->proc, path, &addr,
1175 			    &imgp->entry_addr);
1176 			free(path, M_TEMP);
1177 			if (error == 0)
1178 				have_interp = TRUE;
1179 		}
1180 		if (!have_interp && newinterp != NULL &&
1181 		    (brand_info->interp_path == NULL ||
1182 		    strcmp(interp, brand_info->interp_path) == 0)) {
1183 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1184 			    &imgp->entry_addr);
1185 			if (error == 0)
1186 				have_interp = TRUE;
1187 		}
1188 		if (!have_interp) {
1189 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1190 			    &imgp->entry_addr);
1191 		}
1192 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1193 		if (error != 0) {
1194 			uprintf("ELF interpreter %s not found, error %d\n",
1195 			    interp, error);
1196 			goto ret;
1197 		}
1198 	} else
1199 		addr = et_dyn_addr;
1200 
1201 	/*
1202 	 * Construct auxargs table (used by the fixup routine)
1203 	 */
1204 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1205 	elf_auxargs->execfd = -1;
1206 	elf_auxargs->phdr = proghdr;
1207 	elf_auxargs->phent = hdr->e_phentsize;
1208 	elf_auxargs->phnum = hdr->e_phnum;
1209 	elf_auxargs->pagesz = PAGE_SIZE;
1210 	elf_auxargs->base = addr;
1211 	elf_auxargs->flags = 0;
1212 	elf_auxargs->entry = entry;
1213 	elf_auxargs->hdr_eflags = hdr->e_flags;
1214 
1215 	imgp->auxargs = elf_auxargs;
1216 	imgp->interpreted = 0;
1217 	imgp->reloc_base = addr;
1218 	imgp->proc->p_osrel = osrel;
1219 	imgp->proc->p_fctl0 = fctl0;
1220 	imgp->proc->p_elf_machine = hdr->e_machine;
1221 	imgp->proc->p_elf_flags = hdr->e_flags;
1222 
1223 ret:
1224 	free(interp_buf, M_TEMP);
1225 	return (error);
1226 }
1227 
1228 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1229 
1230 int
1231 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1232 {
1233 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1234 	Elf_Auxinfo *argarray, *pos;
1235 	Elf_Addr *base, *auxbase;
1236 	int error;
1237 
1238 	base = (Elf_Addr *)*stack_base;
1239 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1240 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1241 	    M_WAITOK | M_ZERO);
1242 
1243 	if (args->execfd != -1)
1244 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1245 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1246 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1247 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1248 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1249 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1250 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1251 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1252 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1253 	if (imgp->execpathp != 0)
1254 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1255 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1256 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1257 	if (imgp->canary != 0) {
1258 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1259 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1260 	}
1261 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1262 	if (imgp->pagesizes != 0) {
1263 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1264 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1265 	}
1266 	if (imgp->sysent->sv_timekeep_base != 0) {
1267 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1268 		    imgp->sysent->sv_timekeep_base);
1269 	}
1270 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1271 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1272 	    imgp->sysent->sv_stackprot);
1273 	if (imgp->sysent->sv_hwcap != NULL)
1274 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1275 	if (imgp->sysent->sv_hwcap2 != NULL)
1276 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1277 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1278 
1279 	free(imgp->auxargs, M_TEMP);
1280 	imgp->auxargs = NULL;
1281 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1282 
1283 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1284 	free(argarray, M_TEMP);
1285 	if (error != 0)
1286 		return (error);
1287 
1288 	base--;
1289 	if (suword(base, imgp->args->argc) == -1)
1290 		return (EFAULT);
1291 	*stack_base = (register_t *)base;
1292 	return (0);
1293 }
1294 
1295 /*
1296  * Code for generating ELF core dumps.
1297  */
1298 
1299 typedef void (*segment_callback)(vm_map_entry_t, void *);
1300 
1301 /* Closure for cb_put_phdr(). */
1302 struct phdr_closure {
1303 	Elf_Phdr *phdr;		/* Program header to fill in */
1304 	Elf_Off offset;		/* Offset of segment in core file */
1305 };
1306 
1307 /* Closure for cb_size_segment(). */
1308 struct sseg_closure {
1309 	int count;		/* Count of writable segments. */
1310 	size_t size;		/* Total size of all writable segments. */
1311 };
1312 
1313 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1314 
1315 struct note_info {
1316 	int		type;		/* Note type. */
1317 	outfunc_t 	outfunc; 	/* Output function. */
1318 	void		*outarg;	/* Argument for the output function. */
1319 	size_t		outsize;	/* Output size. */
1320 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1321 };
1322 
1323 TAILQ_HEAD(note_info_list, note_info);
1324 
1325 /* Coredump output parameters. */
1326 struct coredump_params {
1327 	off_t		offset;
1328 	struct ucred	*active_cred;
1329 	struct ucred	*file_cred;
1330 	struct thread	*td;
1331 	struct vnode	*vp;
1332 	struct compressor *comp;
1333 };
1334 
1335 extern int compress_user_cores;
1336 extern int compress_user_cores_level;
1337 
1338 static void cb_put_phdr(vm_map_entry_t, void *);
1339 static void cb_size_segment(vm_map_entry_t, void *);
1340 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1341     enum uio_seg);
1342 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1343 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1344     struct note_info_list *, size_t);
1345 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1346     size_t *);
1347 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1348 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1349 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1350 static int sbuf_drain_core_output(void *, const char *, int);
1351 static int sbuf_drain_count(void *arg, const char *data, int len);
1352 
1353 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1354 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1355 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1356 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1357 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1358 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1359 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1360 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1361 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1362 static void note_procstat_files(void *, struct sbuf *, size_t *);
1363 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1364 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1365 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1366 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1367 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1368 
1369 /*
1370  * Write out a core segment to the compression stream.
1371  */
1372 static int
1373 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1374 {
1375 	u_int chunk_len;
1376 	int error;
1377 
1378 	while (len > 0) {
1379 		chunk_len = MIN(len, CORE_BUF_SIZE);
1380 
1381 		/*
1382 		 * We can get EFAULT error here.
1383 		 * In that case zero out the current chunk of the segment.
1384 		 */
1385 		error = copyin(base, buf, chunk_len);
1386 		if (error != 0)
1387 			bzero(buf, chunk_len);
1388 		error = compressor_write(p->comp, buf, chunk_len);
1389 		if (error != 0)
1390 			break;
1391 		base += chunk_len;
1392 		len -= chunk_len;
1393 	}
1394 	return (error);
1395 }
1396 
1397 static int
1398 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1399 {
1400 
1401 	return (core_write((struct coredump_params *)arg, base, len, offset,
1402 	    UIO_SYSSPACE));
1403 }
1404 
1405 static int
1406 core_write(struct coredump_params *p, const void *base, size_t len,
1407     off_t offset, enum uio_seg seg)
1408 {
1409 
1410 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1411 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1412 	    p->active_cred, p->file_cred, NULL, p->td));
1413 }
1414 
1415 static int
1416 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1417     void *tmpbuf)
1418 {
1419 	int error;
1420 
1421 	if (p->comp != NULL)
1422 		return (compress_chunk(p, base, tmpbuf, len));
1423 
1424 	/*
1425 	 * EFAULT is a non-fatal error that we can get, for example,
1426 	 * if the segment is backed by a file but extends beyond its
1427 	 * end.
1428 	 */
1429 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1430 	if (error == EFAULT) {
1431 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1432 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1433 		    "for process %s\n", base, len, offset, curproc->p_comm);
1434 
1435 		/*
1436 		 * Write a "real" zero byte at the end of the target region
1437 		 * in the case this is the last segment.
1438 		 * The intermediate space will be implicitly zero-filled.
1439 		 */
1440 		error = core_write(p, zero_region, 1, offset + len - 1,
1441 		    UIO_SYSSPACE);
1442 	}
1443 	return (error);
1444 }
1445 
1446 /*
1447  * Drain into a core file.
1448  */
1449 static int
1450 sbuf_drain_core_output(void *arg, const char *data, int len)
1451 {
1452 	struct coredump_params *p;
1453 	int error, locked;
1454 
1455 	p = (struct coredump_params *)arg;
1456 
1457 	/*
1458 	 * Some kern_proc out routines that print to this sbuf may
1459 	 * call us with the process lock held. Draining with the
1460 	 * non-sleepable lock held is unsafe. The lock is needed for
1461 	 * those routines when dumping a live process. In our case we
1462 	 * can safely release the lock before draining and acquire
1463 	 * again after.
1464 	 */
1465 	locked = PROC_LOCKED(p->td->td_proc);
1466 	if (locked)
1467 		PROC_UNLOCK(p->td->td_proc);
1468 	if (p->comp != NULL)
1469 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1470 	else
1471 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1472 		    UIO_SYSSPACE);
1473 	if (locked)
1474 		PROC_LOCK(p->td->td_proc);
1475 	if (error != 0)
1476 		return (-error);
1477 	p->offset += len;
1478 	return (len);
1479 }
1480 
1481 /*
1482  * Drain into a counter.
1483  */
1484 static int
1485 sbuf_drain_count(void *arg, const char *data __unused, int len)
1486 {
1487 	size_t *sizep;
1488 
1489 	sizep = (size_t *)arg;
1490 	*sizep += len;
1491 	return (len);
1492 }
1493 
1494 int
1495 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1496 {
1497 	struct ucred *cred = td->td_ucred;
1498 	int error = 0;
1499 	struct sseg_closure seginfo;
1500 	struct note_info_list notelst;
1501 	struct coredump_params params;
1502 	struct note_info *ninfo;
1503 	void *hdr, *tmpbuf;
1504 	size_t hdrsize, notesz, coresize;
1505 
1506 	hdr = NULL;
1507 	tmpbuf = NULL;
1508 	TAILQ_INIT(&notelst);
1509 
1510 	/* Size the program segments. */
1511 	seginfo.count = 0;
1512 	seginfo.size = 0;
1513 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1514 
1515 	/*
1516 	 * Collect info about the core file header area.
1517 	 */
1518 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1519 	if (seginfo.count + 1 >= PN_XNUM)
1520 		hdrsize += sizeof(Elf_Shdr);
1521 	__elfN(prepare_notes)(td, &notelst, &notesz);
1522 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1523 
1524 	/* Set up core dump parameters. */
1525 	params.offset = 0;
1526 	params.active_cred = cred;
1527 	params.file_cred = NOCRED;
1528 	params.td = td;
1529 	params.vp = vp;
1530 	params.comp = NULL;
1531 
1532 #ifdef RACCT
1533 	if (racct_enable) {
1534 		PROC_LOCK(td->td_proc);
1535 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1536 		PROC_UNLOCK(td->td_proc);
1537 		if (error != 0) {
1538 			error = EFAULT;
1539 			goto done;
1540 		}
1541 	}
1542 #endif
1543 	if (coresize >= limit) {
1544 		error = EFAULT;
1545 		goto done;
1546 	}
1547 
1548 	/* Create a compression stream if necessary. */
1549 	if (compress_user_cores != 0) {
1550 		params.comp = compressor_init(core_compressed_write,
1551 		    compress_user_cores, CORE_BUF_SIZE,
1552 		    compress_user_cores_level, &params);
1553 		if (params.comp == NULL) {
1554 			error = EFAULT;
1555 			goto done;
1556 		}
1557 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1558         }
1559 
1560 	/*
1561 	 * Allocate memory for building the header, fill it up,
1562 	 * and write it out following the notes.
1563 	 */
1564 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1565 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1566 	    notesz);
1567 
1568 	/* Write the contents of all of the writable segments. */
1569 	if (error == 0) {
1570 		Elf_Phdr *php;
1571 		off_t offset;
1572 		int i;
1573 
1574 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1575 		offset = round_page(hdrsize + notesz);
1576 		for (i = 0; i < seginfo.count; i++) {
1577 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1578 			    php->p_filesz, offset, &params, tmpbuf);
1579 			if (error != 0)
1580 				break;
1581 			offset += php->p_filesz;
1582 			php++;
1583 		}
1584 		if (error == 0 && params.comp != NULL)
1585 			error = compressor_flush(params.comp);
1586 	}
1587 	if (error) {
1588 		log(LOG_WARNING,
1589 		    "Failed to write core file for process %s (error %d)\n",
1590 		    curproc->p_comm, error);
1591 	}
1592 
1593 done:
1594 	free(tmpbuf, M_TEMP);
1595 	if (params.comp != NULL)
1596 		compressor_fini(params.comp);
1597 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1598 		TAILQ_REMOVE(&notelst, ninfo, link);
1599 		free(ninfo, M_TEMP);
1600 	}
1601 	if (hdr != NULL)
1602 		free(hdr, M_TEMP);
1603 
1604 	return (error);
1605 }
1606 
1607 /*
1608  * A callback for each_dumpable_segment() to write out the segment's
1609  * program header entry.
1610  */
1611 static void
1612 cb_put_phdr(vm_map_entry_t entry, void *closure)
1613 {
1614 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1615 	Elf_Phdr *phdr = phc->phdr;
1616 
1617 	phc->offset = round_page(phc->offset);
1618 
1619 	phdr->p_type = PT_LOAD;
1620 	phdr->p_offset = phc->offset;
1621 	phdr->p_vaddr = entry->start;
1622 	phdr->p_paddr = 0;
1623 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1624 	phdr->p_align = PAGE_SIZE;
1625 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1626 
1627 	phc->offset += phdr->p_filesz;
1628 	phc->phdr++;
1629 }
1630 
1631 /*
1632  * A callback for each_dumpable_segment() to gather information about
1633  * the number of segments and their total size.
1634  */
1635 static void
1636 cb_size_segment(vm_map_entry_t entry, void *closure)
1637 {
1638 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1639 
1640 	ssc->count++;
1641 	ssc->size += entry->end - entry->start;
1642 }
1643 
1644 /*
1645  * For each writable segment in the process's memory map, call the given
1646  * function with a pointer to the map entry and some arbitrary
1647  * caller-supplied data.
1648  */
1649 static void
1650 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1651 {
1652 	struct proc *p = td->td_proc;
1653 	vm_map_t map = &p->p_vmspace->vm_map;
1654 	vm_map_entry_t entry;
1655 	vm_object_t backing_object, object;
1656 	boolean_t ignore_entry;
1657 
1658 	vm_map_lock_read(map);
1659 	for (entry = map->header.next; entry != &map->header;
1660 	    entry = entry->next) {
1661 		/*
1662 		 * Don't dump inaccessible mappings, deal with legacy
1663 		 * coredump mode.
1664 		 *
1665 		 * Note that read-only segments related to the elf binary
1666 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1667 		 * need to arbitrarily ignore such segments.
1668 		 */
1669 		if (elf_legacy_coredump) {
1670 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1671 				continue;
1672 		} else {
1673 			if ((entry->protection & VM_PROT_ALL) == 0)
1674 				continue;
1675 		}
1676 
1677 		/*
1678 		 * Dont include memory segment in the coredump if
1679 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1680 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1681 		 * kernel map).
1682 		 */
1683 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1684 			continue;
1685 
1686 		if ((object = entry->object.vm_object) == NULL)
1687 			continue;
1688 
1689 		/* Ignore memory-mapped devices and such things. */
1690 		VM_OBJECT_RLOCK(object);
1691 		while ((backing_object = object->backing_object) != NULL) {
1692 			VM_OBJECT_RLOCK(backing_object);
1693 			VM_OBJECT_RUNLOCK(object);
1694 			object = backing_object;
1695 		}
1696 		ignore_entry = object->type != OBJT_DEFAULT &&
1697 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1698 		    object->type != OBJT_PHYS;
1699 		VM_OBJECT_RUNLOCK(object);
1700 		if (ignore_entry)
1701 			continue;
1702 
1703 		(*func)(entry, closure);
1704 	}
1705 	vm_map_unlock_read(map);
1706 }
1707 
1708 /*
1709  * Write the core file header to the file, including padding up to
1710  * the page boundary.
1711  */
1712 static int
1713 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1714     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1715 {
1716 	struct note_info *ninfo;
1717 	struct sbuf *sb;
1718 	int error;
1719 
1720 	/* Fill in the header. */
1721 	bzero(hdr, hdrsize);
1722 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1723 
1724 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1725 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1726 	sbuf_start_section(sb, NULL);
1727 	sbuf_bcat(sb, hdr, hdrsize);
1728 	TAILQ_FOREACH(ninfo, notelst, link)
1729 	    __elfN(putnote)(ninfo, sb);
1730 	/* Align up to a page boundary for the program segments. */
1731 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1732 	error = sbuf_finish(sb);
1733 	sbuf_delete(sb);
1734 
1735 	return (error);
1736 }
1737 
1738 static void
1739 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1740     size_t *sizep)
1741 {
1742 	struct proc *p;
1743 	struct thread *thr;
1744 	size_t size;
1745 
1746 	p = td->td_proc;
1747 	size = 0;
1748 
1749 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1750 
1751 	/*
1752 	 * To have the debugger select the right thread (LWP) as the initial
1753 	 * thread, we dump the state of the thread passed to us in td first.
1754 	 * This is the thread that causes the core dump and thus likely to
1755 	 * be the right thread one wants to have selected in the debugger.
1756 	 */
1757 	thr = td;
1758 	while (thr != NULL) {
1759 		size += register_note(list, NT_PRSTATUS,
1760 		    __elfN(note_prstatus), thr);
1761 		size += register_note(list, NT_FPREGSET,
1762 		    __elfN(note_fpregset), thr);
1763 		size += register_note(list, NT_THRMISC,
1764 		    __elfN(note_thrmisc), thr);
1765 		size += register_note(list, NT_PTLWPINFO,
1766 		    __elfN(note_ptlwpinfo), thr);
1767 		size += register_note(list, -1,
1768 		    __elfN(note_threadmd), thr);
1769 
1770 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1771 		    TAILQ_NEXT(thr, td_plist);
1772 		if (thr == td)
1773 			thr = TAILQ_NEXT(thr, td_plist);
1774 	}
1775 
1776 	size += register_note(list, NT_PROCSTAT_PROC,
1777 	    __elfN(note_procstat_proc), p);
1778 	size += register_note(list, NT_PROCSTAT_FILES,
1779 	    note_procstat_files, p);
1780 	size += register_note(list, NT_PROCSTAT_VMMAP,
1781 	    note_procstat_vmmap, p);
1782 	size += register_note(list, NT_PROCSTAT_GROUPS,
1783 	    note_procstat_groups, p);
1784 	size += register_note(list, NT_PROCSTAT_UMASK,
1785 	    note_procstat_umask, p);
1786 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1787 	    note_procstat_rlimit, p);
1788 	size += register_note(list, NT_PROCSTAT_OSREL,
1789 	    note_procstat_osrel, p);
1790 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1791 	    __elfN(note_procstat_psstrings), p);
1792 	size += register_note(list, NT_PROCSTAT_AUXV,
1793 	    __elfN(note_procstat_auxv), p);
1794 
1795 	*sizep = size;
1796 }
1797 
1798 static void
1799 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1800     size_t notesz)
1801 {
1802 	Elf_Ehdr *ehdr;
1803 	Elf_Phdr *phdr;
1804 	Elf_Shdr *shdr;
1805 	struct phdr_closure phc;
1806 
1807 	ehdr = (Elf_Ehdr *)hdr;
1808 
1809 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1810 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1811 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1812 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1813 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1814 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1815 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1816 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1817 	ehdr->e_ident[EI_ABIVERSION] = 0;
1818 	ehdr->e_ident[EI_PAD] = 0;
1819 	ehdr->e_type = ET_CORE;
1820 	ehdr->e_machine = td->td_proc->p_elf_machine;
1821 	ehdr->e_version = EV_CURRENT;
1822 	ehdr->e_entry = 0;
1823 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1824 	ehdr->e_flags = td->td_proc->p_elf_flags;
1825 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1826 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1827 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1828 	ehdr->e_shstrndx = SHN_UNDEF;
1829 	if (numsegs + 1 < PN_XNUM) {
1830 		ehdr->e_phnum = numsegs + 1;
1831 		ehdr->e_shnum = 0;
1832 	} else {
1833 		ehdr->e_phnum = PN_XNUM;
1834 		ehdr->e_shnum = 1;
1835 
1836 		ehdr->e_shoff = ehdr->e_phoff +
1837 		    (numsegs + 1) * ehdr->e_phentsize;
1838 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1839 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1840 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1841 
1842 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1843 		memset(shdr, 0, sizeof(*shdr));
1844 		/*
1845 		 * A special first section is used to hold large segment and
1846 		 * section counts.  This was proposed by Sun Microsystems in
1847 		 * Solaris and has been adopted by Linux; the standard ELF
1848 		 * tools are already familiar with the technique.
1849 		 *
1850 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1851 		 * (or 12-7 depending on the version of the document) for more
1852 		 * details.
1853 		 */
1854 		shdr->sh_type = SHT_NULL;
1855 		shdr->sh_size = ehdr->e_shnum;
1856 		shdr->sh_link = ehdr->e_shstrndx;
1857 		shdr->sh_info = numsegs + 1;
1858 	}
1859 
1860 	/*
1861 	 * Fill in the program header entries.
1862 	 */
1863 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1864 
1865 	/* The note segement. */
1866 	phdr->p_type = PT_NOTE;
1867 	phdr->p_offset = hdrsize;
1868 	phdr->p_vaddr = 0;
1869 	phdr->p_paddr = 0;
1870 	phdr->p_filesz = notesz;
1871 	phdr->p_memsz = 0;
1872 	phdr->p_flags = PF_R;
1873 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1874 	phdr++;
1875 
1876 	/* All the writable segments from the program. */
1877 	phc.phdr = phdr;
1878 	phc.offset = round_page(hdrsize + notesz);
1879 	each_dumpable_segment(td, cb_put_phdr, &phc);
1880 }
1881 
1882 static size_t
1883 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1884 {
1885 	struct note_info *ninfo;
1886 	size_t size, notesize;
1887 
1888 	size = 0;
1889 	out(arg, NULL, &size);
1890 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1891 	ninfo->type = type;
1892 	ninfo->outfunc = out;
1893 	ninfo->outarg = arg;
1894 	ninfo->outsize = size;
1895 	TAILQ_INSERT_TAIL(list, ninfo, link);
1896 
1897 	if (type == -1)
1898 		return (size);
1899 
1900 	notesize = sizeof(Elf_Note) +		/* note header */
1901 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1902 						/* note name */
1903 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1904 
1905 	return (notesize);
1906 }
1907 
1908 static size_t
1909 append_note_data(const void *src, void *dst, size_t len)
1910 {
1911 	size_t padded_len;
1912 
1913 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1914 	if (dst != NULL) {
1915 		bcopy(src, dst, len);
1916 		bzero((char *)dst + len, padded_len - len);
1917 	}
1918 	return (padded_len);
1919 }
1920 
1921 size_t
1922 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1923 {
1924 	Elf_Note *note;
1925 	char *buf;
1926 	size_t notesize;
1927 
1928 	buf = dst;
1929 	if (buf != NULL) {
1930 		note = (Elf_Note *)buf;
1931 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1932 		note->n_descsz = size;
1933 		note->n_type = type;
1934 		buf += sizeof(*note);
1935 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1936 		    sizeof(FREEBSD_ABI_VENDOR));
1937 		append_note_data(src, buf, size);
1938 		if (descp != NULL)
1939 			*descp = buf;
1940 	}
1941 
1942 	notesize = sizeof(Elf_Note) +		/* note header */
1943 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1944 						/* note name */
1945 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1946 
1947 	return (notesize);
1948 }
1949 
1950 static void
1951 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1952 {
1953 	Elf_Note note;
1954 	ssize_t old_len, sect_len;
1955 	size_t new_len, descsz, i;
1956 
1957 	if (ninfo->type == -1) {
1958 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1959 		return;
1960 	}
1961 
1962 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1963 	note.n_descsz = ninfo->outsize;
1964 	note.n_type = ninfo->type;
1965 
1966 	sbuf_bcat(sb, &note, sizeof(note));
1967 	sbuf_start_section(sb, &old_len);
1968 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1969 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1970 	if (note.n_descsz == 0)
1971 		return;
1972 	sbuf_start_section(sb, &old_len);
1973 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1974 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1975 	if (sect_len < 0)
1976 		return;
1977 
1978 	new_len = (size_t)sect_len;
1979 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1980 	if (new_len < descsz) {
1981 		/*
1982 		 * It is expected that individual note emitters will correctly
1983 		 * predict their expected output size and fill up to that size
1984 		 * themselves, padding in a format-specific way if needed.
1985 		 * However, in case they don't, just do it here with zeros.
1986 		 */
1987 		for (i = 0; i < descsz - new_len; i++)
1988 			sbuf_putc(sb, 0);
1989 	} else if (new_len > descsz) {
1990 		/*
1991 		 * We can't always truncate sb -- we may have drained some
1992 		 * of it already.
1993 		 */
1994 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1995 		    "read it (%zu > %zu).  Since it is longer than "
1996 		    "expected, this coredump's notes are corrupt.  THIS "
1997 		    "IS A BUG in the note_procstat routine for type %u.\n",
1998 		    __func__, (unsigned)note.n_type, new_len, descsz,
1999 		    (unsigned)note.n_type));
2000 	}
2001 }
2002 
2003 /*
2004  * Miscellaneous note out functions.
2005  */
2006 
2007 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2008 #include <compat/freebsd32/freebsd32.h>
2009 #include <compat/freebsd32/freebsd32_signal.h>
2010 
2011 typedef struct prstatus32 elf_prstatus_t;
2012 typedef struct prpsinfo32 elf_prpsinfo_t;
2013 typedef struct fpreg32 elf_prfpregset_t;
2014 typedef struct fpreg32 elf_fpregset_t;
2015 typedef struct reg32 elf_gregset_t;
2016 typedef struct thrmisc32 elf_thrmisc_t;
2017 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
2018 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2019 typedef uint32_t elf_ps_strings_t;
2020 #else
2021 typedef prstatus_t elf_prstatus_t;
2022 typedef prpsinfo_t elf_prpsinfo_t;
2023 typedef prfpregset_t elf_prfpregset_t;
2024 typedef prfpregset_t elf_fpregset_t;
2025 typedef gregset_t elf_gregset_t;
2026 typedef thrmisc_t elf_thrmisc_t;
2027 #define ELF_KERN_PROC_MASK	0
2028 typedef struct kinfo_proc elf_kinfo_proc_t;
2029 typedef vm_offset_t elf_ps_strings_t;
2030 #endif
2031 
2032 static void
2033 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2034 {
2035 	struct sbuf sbarg;
2036 	size_t len;
2037 	char *cp, *end;
2038 	struct proc *p;
2039 	elf_prpsinfo_t *psinfo;
2040 	int error;
2041 
2042 	p = (struct proc *)arg;
2043 	if (sb != NULL) {
2044 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2045 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2046 		psinfo->pr_version = PRPSINFO_VERSION;
2047 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2048 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2049 		PROC_LOCK(p);
2050 		if (p->p_args != NULL) {
2051 			len = sizeof(psinfo->pr_psargs) - 1;
2052 			if (len > p->p_args->ar_length)
2053 				len = p->p_args->ar_length;
2054 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2055 			PROC_UNLOCK(p);
2056 			error = 0;
2057 		} else {
2058 			_PHOLD(p);
2059 			PROC_UNLOCK(p);
2060 			sbuf_new(&sbarg, psinfo->pr_psargs,
2061 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2062 			error = proc_getargv(curthread, p, &sbarg);
2063 			PRELE(p);
2064 			if (sbuf_finish(&sbarg) == 0)
2065 				len = sbuf_len(&sbarg) - 1;
2066 			else
2067 				len = sizeof(psinfo->pr_psargs) - 1;
2068 			sbuf_delete(&sbarg);
2069 		}
2070 		if (error || len == 0)
2071 			strlcpy(psinfo->pr_psargs, p->p_comm,
2072 			    sizeof(psinfo->pr_psargs));
2073 		else {
2074 			KASSERT(len < sizeof(psinfo->pr_psargs),
2075 			    ("len is too long: %zu vs %zu", len,
2076 			    sizeof(psinfo->pr_psargs)));
2077 			cp = psinfo->pr_psargs;
2078 			end = cp + len - 1;
2079 			for (;;) {
2080 				cp = memchr(cp, '\0', end - cp);
2081 				if (cp == NULL)
2082 					break;
2083 				*cp = ' ';
2084 			}
2085 		}
2086 		psinfo->pr_pid = p->p_pid;
2087 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2088 		free(psinfo, M_TEMP);
2089 	}
2090 	*sizep = sizeof(*psinfo);
2091 }
2092 
2093 static void
2094 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2095 {
2096 	struct thread *td;
2097 	elf_prstatus_t *status;
2098 
2099 	td = (struct thread *)arg;
2100 	if (sb != NULL) {
2101 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
2102 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2103 		status->pr_version = PRSTATUS_VERSION;
2104 		status->pr_statussz = sizeof(elf_prstatus_t);
2105 		status->pr_gregsetsz = sizeof(elf_gregset_t);
2106 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2107 		status->pr_osreldate = osreldate;
2108 		status->pr_cursig = td->td_proc->p_sig;
2109 		status->pr_pid = td->td_tid;
2110 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2111 		fill_regs32(td, &status->pr_reg);
2112 #else
2113 		fill_regs(td, &status->pr_reg);
2114 #endif
2115 		sbuf_bcat(sb, status, sizeof(*status));
2116 		free(status, M_TEMP);
2117 	}
2118 	*sizep = sizeof(*status);
2119 }
2120 
2121 static void
2122 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2123 {
2124 	struct thread *td;
2125 	elf_prfpregset_t *fpregset;
2126 
2127 	td = (struct thread *)arg;
2128 	if (sb != NULL) {
2129 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2130 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2131 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2132 		fill_fpregs32(td, fpregset);
2133 #else
2134 		fill_fpregs(td, fpregset);
2135 #endif
2136 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2137 		free(fpregset, M_TEMP);
2138 	}
2139 	*sizep = sizeof(*fpregset);
2140 }
2141 
2142 static void
2143 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2144 {
2145 	struct thread *td;
2146 	elf_thrmisc_t thrmisc;
2147 
2148 	td = (struct thread *)arg;
2149 	if (sb != NULL) {
2150 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2151 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2152 		strcpy(thrmisc.pr_tname, td->td_name);
2153 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2154 	}
2155 	*sizep = sizeof(thrmisc);
2156 }
2157 
2158 static void
2159 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2160 {
2161 	struct thread *td;
2162 	size_t size;
2163 	int structsize;
2164 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2165 	struct ptrace_lwpinfo32 pl;
2166 #else
2167 	struct ptrace_lwpinfo pl;
2168 #endif
2169 
2170 	td = (struct thread *)arg;
2171 	size = sizeof(structsize) + sizeof(pl);
2172 	if (sb != NULL) {
2173 		KASSERT(*sizep == size, ("invalid size"));
2174 		structsize = sizeof(pl);
2175 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2176 		bzero(&pl, sizeof(pl));
2177 		pl.pl_lwpid = td->td_tid;
2178 		pl.pl_event = PL_EVENT_NONE;
2179 		pl.pl_sigmask = td->td_sigmask;
2180 		pl.pl_siglist = td->td_siglist;
2181 		if (td->td_si.si_signo != 0) {
2182 			pl.pl_event = PL_EVENT_SIGNAL;
2183 			pl.pl_flags |= PL_FLAG_SI;
2184 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2185 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2186 #else
2187 			pl.pl_siginfo = td->td_si;
2188 #endif
2189 		}
2190 		strcpy(pl.pl_tdname, td->td_name);
2191 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2192 		sbuf_bcat(sb, &pl, sizeof(pl));
2193 	}
2194 	*sizep = size;
2195 }
2196 
2197 /*
2198  * Allow for MD specific notes, as well as any MD
2199  * specific preparations for writing MI notes.
2200  */
2201 static void
2202 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2203 {
2204 	struct thread *td;
2205 	void *buf;
2206 	size_t size;
2207 
2208 	td = (struct thread *)arg;
2209 	size = *sizep;
2210 	if (size != 0 && sb != NULL)
2211 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2212 	else
2213 		buf = NULL;
2214 	size = 0;
2215 	__elfN(dump_thread)(td, buf, &size);
2216 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2217 	if (size != 0 && sb != NULL)
2218 		sbuf_bcat(sb, buf, size);
2219 	free(buf, M_TEMP);
2220 	*sizep = size;
2221 }
2222 
2223 #ifdef KINFO_PROC_SIZE
2224 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2225 #endif
2226 
2227 static void
2228 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2229 {
2230 	struct proc *p;
2231 	size_t size;
2232 	int structsize;
2233 
2234 	p = (struct proc *)arg;
2235 	size = sizeof(structsize) + p->p_numthreads *
2236 	    sizeof(elf_kinfo_proc_t);
2237 
2238 	if (sb != NULL) {
2239 		KASSERT(*sizep == size, ("invalid size"));
2240 		structsize = sizeof(elf_kinfo_proc_t);
2241 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2242 		PROC_LOCK(p);
2243 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2244 	}
2245 	*sizep = size;
2246 }
2247 
2248 #ifdef KINFO_FILE_SIZE
2249 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2250 #endif
2251 
2252 static void
2253 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2254 {
2255 	struct proc *p;
2256 	size_t size, sect_sz, i;
2257 	ssize_t start_len, sect_len;
2258 	int structsize, filedesc_flags;
2259 
2260 	if (coredump_pack_fileinfo)
2261 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2262 	else
2263 		filedesc_flags = 0;
2264 
2265 	p = (struct proc *)arg;
2266 	structsize = sizeof(struct kinfo_file);
2267 	if (sb == NULL) {
2268 		size = 0;
2269 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2270 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2271 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2272 		PROC_LOCK(p);
2273 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2274 		sbuf_finish(sb);
2275 		sbuf_delete(sb);
2276 		*sizep = size;
2277 	} else {
2278 		sbuf_start_section(sb, &start_len);
2279 
2280 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2281 		PROC_LOCK(p);
2282 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2283 		    filedesc_flags);
2284 
2285 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2286 		if (sect_len < 0)
2287 			return;
2288 		sect_sz = sect_len;
2289 
2290 		KASSERT(sect_sz <= *sizep,
2291 		    ("kern_proc_filedesc_out did not respect maxlen; "
2292 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2293 		     sect_sz - sizeof(structsize)));
2294 
2295 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2296 			sbuf_putc(sb, 0);
2297 	}
2298 }
2299 
2300 #ifdef KINFO_VMENTRY_SIZE
2301 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2302 #endif
2303 
2304 static void
2305 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2306 {
2307 	struct proc *p;
2308 	size_t size;
2309 	int structsize, vmmap_flags;
2310 
2311 	if (coredump_pack_vmmapinfo)
2312 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2313 	else
2314 		vmmap_flags = 0;
2315 
2316 	p = (struct proc *)arg;
2317 	structsize = sizeof(struct kinfo_vmentry);
2318 	if (sb == NULL) {
2319 		size = 0;
2320 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2321 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2322 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2323 		PROC_LOCK(p);
2324 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2325 		sbuf_finish(sb);
2326 		sbuf_delete(sb);
2327 		*sizep = size;
2328 	} else {
2329 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2330 		PROC_LOCK(p);
2331 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2332 		    vmmap_flags);
2333 	}
2334 }
2335 
2336 static void
2337 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2338 {
2339 	struct proc *p;
2340 	size_t size;
2341 	int structsize;
2342 
2343 	p = (struct proc *)arg;
2344 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2345 	if (sb != NULL) {
2346 		KASSERT(*sizep == size, ("invalid size"));
2347 		structsize = sizeof(gid_t);
2348 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2349 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2350 		    sizeof(gid_t));
2351 	}
2352 	*sizep = size;
2353 }
2354 
2355 static void
2356 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2357 {
2358 	struct proc *p;
2359 	size_t size;
2360 	int structsize;
2361 
2362 	p = (struct proc *)arg;
2363 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2364 	if (sb != NULL) {
2365 		KASSERT(*sizep == size, ("invalid size"));
2366 		structsize = sizeof(p->p_fd->fd_cmask);
2367 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2368 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2369 	}
2370 	*sizep = size;
2371 }
2372 
2373 static void
2374 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2375 {
2376 	struct proc *p;
2377 	struct rlimit rlim[RLIM_NLIMITS];
2378 	size_t size;
2379 	int structsize, i;
2380 
2381 	p = (struct proc *)arg;
2382 	size = sizeof(structsize) + sizeof(rlim);
2383 	if (sb != NULL) {
2384 		KASSERT(*sizep == size, ("invalid size"));
2385 		structsize = sizeof(rlim);
2386 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2387 		PROC_LOCK(p);
2388 		for (i = 0; i < RLIM_NLIMITS; i++)
2389 			lim_rlimit_proc(p, i, &rlim[i]);
2390 		PROC_UNLOCK(p);
2391 		sbuf_bcat(sb, rlim, sizeof(rlim));
2392 	}
2393 	*sizep = size;
2394 }
2395 
2396 static void
2397 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2398 {
2399 	struct proc *p;
2400 	size_t size;
2401 	int structsize;
2402 
2403 	p = (struct proc *)arg;
2404 	size = sizeof(structsize) + sizeof(p->p_osrel);
2405 	if (sb != NULL) {
2406 		KASSERT(*sizep == size, ("invalid size"));
2407 		structsize = sizeof(p->p_osrel);
2408 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2409 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2410 	}
2411 	*sizep = size;
2412 }
2413 
2414 static void
2415 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2416 {
2417 	struct proc *p;
2418 	elf_ps_strings_t ps_strings;
2419 	size_t size;
2420 	int structsize;
2421 
2422 	p = (struct proc *)arg;
2423 	size = sizeof(structsize) + sizeof(ps_strings);
2424 	if (sb != NULL) {
2425 		KASSERT(*sizep == size, ("invalid size"));
2426 		structsize = sizeof(ps_strings);
2427 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2428 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2429 #else
2430 		ps_strings = p->p_sysent->sv_psstrings;
2431 #endif
2432 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2433 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2434 	}
2435 	*sizep = size;
2436 }
2437 
2438 static void
2439 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2440 {
2441 	struct proc *p;
2442 	size_t size;
2443 	int structsize;
2444 
2445 	p = (struct proc *)arg;
2446 	if (sb == NULL) {
2447 		size = 0;
2448 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2449 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2450 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2451 		PHOLD(p);
2452 		proc_getauxv(curthread, p, sb);
2453 		PRELE(p);
2454 		sbuf_finish(sb);
2455 		sbuf_delete(sb);
2456 		*sizep = size;
2457 	} else {
2458 		structsize = sizeof(Elf_Auxinfo);
2459 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2460 		PHOLD(p);
2461 		proc_getauxv(curthread, p, sb);
2462 		PRELE(p);
2463 	}
2464 }
2465 
2466 static boolean_t
2467 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2468     const char *note_vendor, const Elf_Phdr *pnote,
2469     boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2470 {
2471 	const Elf_Note *note, *note0, *note_end;
2472 	const char *note_name;
2473 	char *buf;
2474 	int i, error;
2475 	boolean_t res;
2476 
2477 	/* We need some limit, might as well use PAGE_SIZE. */
2478 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2479 		return (FALSE);
2480 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2481 	if (pnote->p_offset > PAGE_SIZE ||
2482 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2483 		VOP_UNLOCK(imgp->vp, 0);
2484 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2485 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2486 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2487 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2488 		    curthread->td_ucred, NOCRED, NULL, curthread);
2489 		if (error != 0) {
2490 			uprintf("i/o error PT_NOTE\n");
2491 			goto retf;
2492 		}
2493 		note = note0 = (const Elf_Note *)buf;
2494 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2495 	} else {
2496 		note = note0 = (const Elf_Note *)(imgp->image_header +
2497 		    pnote->p_offset);
2498 		note_end = (const Elf_Note *)(imgp->image_header +
2499 		    pnote->p_offset + pnote->p_filesz);
2500 		buf = NULL;
2501 	}
2502 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2503 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2504 		    (const char *)note < sizeof(Elf_Note)) {
2505 			goto retf;
2506 		}
2507 		if (note->n_namesz != checknote->n_namesz ||
2508 		    note->n_descsz != checknote->n_descsz ||
2509 		    note->n_type != checknote->n_type)
2510 			goto nextnote;
2511 		note_name = (const char *)(note + 1);
2512 		if (note_name + checknote->n_namesz >=
2513 		    (const char *)note_end || strncmp(note_vendor,
2514 		    note_name, checknote->n_namesz) != 0)
2515 			goto nextnote;
2516 
2517 		if (cb(note, cb_arg, &res))
2518 			goto ret;
2519 nextnote:
2520 		note = (const Elf_Note *)((const char *)(note + 1) +
2521 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2522 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2523 	}
2524 retf:
2525 	res = FALSE;
2526 ret:
2527 	free(buf, M_TEMP);
2528 	return (res);
2529 }
2530 
2531 struct brandnote_cb_arg {
2532 	Elf_Brandnote *brandnote;
2533 	int32_t *osrel;
2534 };
2535 
2536 static boolean_t
2537 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2538 {
2539 	struct brandnote_cb_arg *arg;
2540 
2541 	arg = arg0;
2542 
2543 	/*
2544 	 * Fetch the osreldate for binary from the ELF OSABI-note if
2545 	 * necessary.
2546 	 */
2547 	*res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2548 	    arg->brandnote->trans_osrel != NULL ?
2549 	    arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2550 
2551 	return (TRUE);
2552 }
2553 
2554 static Elf_Note fctl_note = {
2555 	.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2556 	.n_descsz = sizeof(uint32_t),
2557 	.n_type = NT_FREEBSD_FEATURE_CTL,
2558 };
2559 
2560 struct fctl_cb_arg {
2561 	uint32_t *fctl0;
2562 };
2563 
2564 static boolean_t
2565 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2566 {
2567 	struct fctl_cb_arg *arg;
2568 	const Elf32_Word *desc;
2569 	uintptr_t p;
2570 
2571 	arg = arg0;
2572 	p = (uintptr_t)(note + 1);
2573 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2574 	desc = (const Elf32_Word *)p;
2575 	*arg->fctl0 = desc[0];
2576 	return (TRUE);
2577 }
2578 
2579 /*
2580  * Try to find the appropriate ABI-note section for checknote, fetch
2581  * the osreldate and feature control flags for binary from the ELF
2582  * OSABI-note.  Only the first page of the image is searched, the same
2583  * as for headers.
2584  */
2585 static boolean_t
2586 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2587     int32_t *osrel, uint32_t *fctl0)
2588 {
2589 	const Elf_Phdr *phdr;
2590 	const Elf_Ehdr *hdr;
2591 	struct brandnote_cb_arg b_arg;
2592 	struct fctl_cb_arg f_arg;
2593 	int i, j;
2594 
2595 	hdr = (const Elf_Ehdr *)imgp->image_header;
2596 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2597 	b_arg.brandnote = brandnote;
2598 	b_arg.osrel = osrel;
2599 	f_arg.fctl0 = fctl0;
2600 
2601 	for (i = 0; i < hdr->e_phnum; i++) {
2602 		if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2603 		    &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2604 		    &b_arg)) {
2605 			for (j = 0; j < hdr->e_phnum; j++) {
2606 				if (phdr[j].p_type == PT_NOTE &&
2607 				    __elfN(parse_notes)(imgp, &fctl_note,
2608 				    FREEBSD_ABI_VENDOR, &phdr[j],
2609 				    note_fctl_cb, &f_arg))
2610 					break;
2611 			}
2612 			return (TRUE);
2613 		}
2614 	}
2615 	return (FALSE);
2616 
2617 }
2618 
2619 /*
2620  * Tell kern_execve.c about it, with a little help from the linker.
2621  */
2622 static struct execsw __elfN(execsw) = {
2623 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2624 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2625 };
2626 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2627 
2628 static vm_prot_t
2629 __elfN(trans_prot)(Elf_Word flags)
2630 {
2631 	vm_prot_t prot;
2632 
2633 	prot = 0;
2634 	if (flags & PF_X)
2635 		prot |= VM_PROT_EXECUTE;
2636 	if (flags & PF_W)
2637 		prot |= VM_PROT_WRITE;
2638 	if (flags & PF_R)
2639 		prot |= VM_PROT_READ;
2640 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2641 	if (i386_read_exec && (flags & PF_R))
2642 		prot |= VM_PROT_EXECUTE;
2643 #endif
2644 	return (prot);
2645 }
2646 
2647 static Elf_Word
2648 __elfN(untrans_prot)(vm_prot_t prot)
2649 {
2650 	Elf_Word flags;
2651 
2652 	flags = 0;
2653 	if (prot & VM_PROT_EXECUTE)
2654 		flags |= PF_X;
2655 	if (prot & VM_PROT_READ)
2656 		flags |= PF_R;
2657 	if (prot & VM_PROT_WRITE)
2658 		flags |= PF_W;
2659 	return (flags);
2660 }
2661