xref: /freebsd/sys/kern/imgact_elf.c (revision 752d135e0dacd9a463d24ffb89779b67ce0a7ea0)
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);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93     u_long *entry, size_t pagesize);
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
96     size_t pagesize);
97 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
98 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99     int32_t *osrel);
100 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
101 static boolean_t __elfN(check_note)(struct image_params *imgp,
102     Elf_Brandnote *checknote, int32_t *osrel);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
107     "");
108 
109 #define	CORE_BUF_SIZE	(16 * 1024)
110 
111 int __elfN(fallback_brand) = -1;
112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
113     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
114     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 
116 static int elf_legacy_coredump = 0;
117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
118     &elf_legacy_coredump, 0,
119     "include all and only RW pages in core dumps");
120 
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
124     defined(__riscv)
125 	1;
126 #else
127 	0;
128 #endif
129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
130     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
131     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 
133 #if __ELF_WORD_SIZE == 32
134 #if defined(__amd64__)
135 int i386_read_exec = 0;
136 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
137     "enable execution from readable segments");
138 #endif
139 #endif
140 
141 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
142 
143 #define	trunc_page_ps(va, ps)	rounddown2(va, ps)
144 #define	round_page_ps(va, ps)	roundup2(va, ps)
145 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
146 
147 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
148 
149 Elf_Brandnote __elfN(freebsd_brandnote) = {
150 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
151 	.hdr.n_descsz	= sizeof(int32_t),
152 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
153 	.vendor		= FREEBSD_ABI_VENDOR,
154 	.flags		= BN_TRANSLATE_OSREL,
155 	.trans_osrel	= __elfN(freebsd_trans_osrel)
156 };
157 
158 static bool
159 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
160 {
161 	uintptr_t p;
162 
163 	p = (uintptr_t)(note + 1);
164 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
165 	*osrel = *(const int32_t *)(p);
166 
167 	return (true);
168 }
169 
170 static const char GNU_ABI_VENDOR[] = "GNU";
171 static int GNU_KFREEBSD_ABI_DESC = 3;
172 
173 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
174 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
175 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
176 	.hdr.n_type	= 1,
177 	.vendor		= GNU_ABI_VENDOR,
178 	.flags		= BN_TRANSLATE_OSREL,
179 	.trans_osrel	= kfreebsd_trans_osrel
180 };
181 
182 static bool
183 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
184 {
185 	const Elf32_Word *desc;
186 	uintptr_t p;
187 
188 	p = (uintptr_t)(note + 1);
189 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
190 
191 	desc = (const Elf32_Word *)p;
192 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
193 		return (false);
194 
195 	/*
196 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
197 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
198 	 */
199 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
200 
201 	return (true);
202 }
203 
204 int
205 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
206 {
207 	int i;
208 
209 	for (i = 0; i < MAX_BRANDS; i++) {
210 		if (elf_brand_list[i] == NULL) {
211 			elf_brand_list[i] = entry;
212 			break;
213 		}
214 	}
215 	if (i == MAX_BRANDS) {
216 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
217 			__func__, entry);
218 		return (-1);
219 	}
220 	return (0);
221 }
222 
223 int
224 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
225 {
226 	int i;
227 
228 	for (i = 0; i < MAX_BRANDS; i++) {
229 		if (elf_brand_list[i] == entry) {
230 			elf_brand_list[i] = NULL;
231 			break;
232 		}
233 	}
234 	if (i == MAX_BRANDS)
235 		return (-1);
236 	return (0);
237 }
238 
239 int
240 __elfN(brand_inuse)(Elf_Brandinfo *entry)
241 {
242 	struct proc *p;
243 	int rval = FALSE;
244 
245 	sx_slock(&allproc_lock);
246 	FOREACH_PROC_IN_SYSTEM(p) {
247 		if (p->p_sysent == entry->sysvec) {
248 			rval = TRUE;
249 			break;
250 		}
251 	}
252 	sx_sunlock(&allproc_lock);
253 
254 	return (rval);
255 }
256 
257 static Elf_Brandinfo *
258 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
259     int interp_name_len, int32_t *osrel)
260 {
261 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
262 	Elf_Brandinfo *bi, *bi_m;
263 	boolean_t ret;
264 	int i;
265 
266 	/*
267 	 * We support four types of branding -- (1) the ELF EI_OSABI field
268 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
269 	 * branding w/in the ELF header, (3) path of the `interp_path'
270 	 * field, and (4) the ".note.ABI-tag" ELF section.
271 	 */
272 
273 	/* Look for an ".note.ABI-tag" ELF section */
274 	bi_m = NULL;
275 	for (i = 0; i < MAX_BRANDS; i++) {
276 		bi = elf_brand_list[i];
277 		if (bi == NULL)
278 			continue;
279 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
280 			continue;
281 		if (hdr->e_machine == bi->machine && (bi->flags &
282 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
283 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
284 			/* Give brand a chance to veto check_note's guess */
285 			if (ret && bi->header_supported)
286 				ret = bi->header_supported(imgp);
287 			/*
288 			 * If note checker claimed the binary, but the
289 			 * interpreter path in the image does not
290 			 * match default one for the brand, try to
291 			 * search for other brands with the same
292 			 * interpreter.  Either there is better brand
293 			 * with the right interpreter, or, failing
294 			 * this, we return first brand which accepted
295 			 * our note and, optionally, header.
296 			 */
297 			if (ret && bi_m == NULL && interp != NULL &&
298 			    (bi->interp_path == NULL ||
299 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
300 			    strncmp(interp, bi->interp_path, interp_name_len)
301 			    != 0))) {
302 				bi_m = bi;
303 				ret = 0;
304 			}
305 			if (ret)
306 				return (bi);
307 		}
308 	}
309 	if (bi_m != NULL)
310 		return (bi_m);
311 
312 	/* If the executable has a brand, search for it in the brand list. */
313 	for (i = 0; i < MAX_BRANDS; i++) {
314 		bi = elf_brand_list[i];
315 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
316 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
317 			continue;
318 		if (hdr->e_machine == bi->machine &&
319 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
320 		    (bi->compat_3_brand != NULL &&
321 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
322 		    bi->compat_3_brand) == 0))) {
323 			/* Looks good, but give brand a chance to veto */
324 			if (bi->header_supported == NULL ||
325 			    bi->header_supported(imgp)) {
326 				/*
327 				 * Again, prefer strictly matching
328 				 * interpreter path.
329 				 */
330 				if (interp_name_len == 0 &&
331 				    bi->interp_path == NULL)
332 					return (bi);
333 				if (bi->interp_path != NULL &&
334 				    strlen(bi->interp_path) + 1 ==
335 				    interp_name_len && strncmp(interp,
336 				    bi->interp_path, interp_name_len) == 0)
337 					return (bi);
338 				if (bi_m == NULL)
339 					bi_m = bi;
340 			}
341 		}
342 	}
343 	if (bi_m != NULL)
344 		return (bi_m);
345 
346 	/* No known brand, see if the header is recognized by any brand */
347 	for (i = 0; i < MAX_BRANDS; i++) {
348 		bi = elf_brand_list[i];
349 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
350 		    bi->header_supported == NULL)
351 			continue;
352 		if (hdr->e_machine == bi->machine) {
353 			ret = bi->header_supported(imgp);
354 			if (ret)
355 				return (bi);
356 		}
357 	}
358 
359 	/* Lacking a known brand, search for a recognized interpreter. */
360 	if (interp != NULL) {
361 		for (i = 0; i < MAX_BRANDS; i++) {
362 			bi = elf_brand_list[i];
363 			if (bi == NULL || (bi->flags &
364 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
365 			    != 0)
366 				continue;
367 			if (hdr->e_machine == bi->machine &&
368 			    bi->interp_path != NULL &&
369 			    /* ELF image p_filesz includes terminating zero */
370 			    strlen(bi->interp_path) + 1 == interp_name_len &&
371 			    strncmp(interp, bi->interp_path, interp_name_len)
372 			    == 0 && (bi->header_supported == NULL ||
373 			    bi->header_supported(imgp)))
374 				return (bi);
375 		}
376 	}
377 
378 	/* Lacking a recognized interpreter, try the default brand */
379 	for (i = 0; i < MAX_BRANDS; i++) {
380 		bi = elf_brand_list[i];
381 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
382 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
383 			continue;
384 		if (hdr->e_machine == bi->machine &&
385 		    __elfN(fallback_brand) == bi->brand &&
386 		    (bi->header_supported == NULL ||
387 		    bi->header_supported(imgp)))
388 			return (bi);
389 	}
390 	return (NULL);
391 }
392 
393 static int
394 __elfN(check_header)(const Elf_Ehdr *hdr)
395 {
396 	Elf_Brandinfo *bi;
397 	int i;
398 
399 	if (!IS_ELF(*hdr) ||
400 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
401 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
402 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
403 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
404 	    hdr->e_version != ELF_TARG_VER)
405 		return (ENOEXEC);
406 
407 	/*
408 	 * Make sure we have at least one brand for this machine.
409 	 */
410 
411 	for (i = 0; i < MAX_BRANDS; i++) {
412 		bi = elf_brand_list[i];
413 		if (bi != NULL && bi->machine == hdr->e_machine)
414 			break;
415 	}
416 	if (i == MAX_BRANDS)
417 		return (ENOEXEC);
418 
419 	return (0);
420 }
421 
422 static int
423 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
424     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
425 {
426 	struct sf_buf *sf;
427 	int error;
428 	vm_offset_t off;
429 
430 	/*
431 	 * Create the page if it doesn't exist yet. Ignore errors.
432 	 */
433 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
434 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
435 
436 	/*
437 	 * Find the page from the underlying object.
438 	 */
439 	if (object != NULL) {
440 		sf = vm_imgact_map_page(object, offset);
441 		if (sf == NULL)
442 			return (KERN_FAILURE);
443 		off = offset - trunc_page(offset);
444 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
445 		    end - start);
446 		vm_imgact_unmap_page(sf);
447 		if (error != 0)
448 			return (KERN_FAILURE);
449 	}
450 
451 	return (KERN_SUCCESS);
452 }
453 
454 static int
455 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
456     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
457     int cow)
458 {
459 	struct sf_buf *sf;
460 	vm_offset_t off;
461 	vm_size_t sz;
462 	int error, locked, rv;
463 
464 	if (start != trunc_page(start)) {
465 		rv = __elfN(map_partial)(map, object, offset, start,
466 		    round_page(start), prot);
467 		if (rv != KERN_SUCCESS)
468 			return (rv);
469 		offset += round_page(start) - start;
470 		start = round_page(start);
471 	}
472 	if (end != round_page(end)) {
473 		rv = __elfN(map_partial)(map, object, offset +
474 		    trunc_page(end) - start, trunc_page(end), end, prot);
475 		if (rv != KERN_SUCCESS)
476 			return (rv);
477 		end = trunc_page(end);
478 	}
479 	if (start >= end)
480 		return (KERN_SUCCESS);
481 	if ((offset & PAGE_MASK) != 0) {
482 		/*
483 		 * The mapping is not page aligned.  This means that we have
484 		 * to copy the data.
485 		 */
486 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
487 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
488 		if (rv != KERN_SUCCESS)
489 			return (rv);
490 		if (object == NULL)
491 			return (KERN_SUCCESS);
492 		for (; start < end; start += sz) {
493 			sf = vm_imgact_map_page(object, offset);
494 			if (sf == NULL)
495 				return (KERN_FAILURE);
496 			off = offset - trunc_page(offset);
497 			sz = end - start;
498 			if (sz > PAGE_SIZE - off)
499 				sz = PAGE_SIZE - off;
500 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
501 			    (caddr_t)start, sz);
502 			vm_imgact_unmap_page(sf);
503 			if (error != 0)
504 				return (KERN_FAILURE);
505 			offset += sz;
506 		}
507 	} else {
508 		vm_object_reference(object);
509 		rv = vm_map_fixed(map, object, offset, start, end - start,
510 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
511 		if (rv != KERN_SUCCESS) {
512 			locked = VOP_ISLOCKED(imgp->vp);
513 			VOP_UNLOCK(imgp->vp, 0);
514 			vm_object_deallocate(object);
515 			vn_lock(imgp->vp, locked | LK_RETRY);
516 			return (rv);
517 		}
518 	}
519 	return (KERN_SUCCESS);
520 }
521 
522 static int
523 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
524     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
525     size_t pagesize)
526 {
527 	struct sf_buf *sf;
528 	size_t map_len;
529 	vm_map_t map;
530 	vm_object_t object;
531 	vm_offset_t off, map_addr;
532 	int error, rv, cow;
533 	size_t copy_len;
534 	vm_ooffset_t file_addr;
535 
536 	/*
537 	 * It's necessary to fail if the filsz + offset taken from the
538 	 * header is greater than the actual file pager object's size.
539 	 * If we were to allow this, then the vm_map_find() below would
540 	 * walk right off the end of the file object and into the ether.
541 	 *
542 	 * While I'm here, might as well check for something else that
543 	 * is invalid: filsz cannot be greater than memsz.
544 	 */
545 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
546 	    filsz > memsz) {
547 		uprintf("elf_load_section: truncated ELF file\n");
548 		return (ENOEXEC);
549 	}
550 
551 	object = imgp->object;
552 	map = &imgp->proc->p_vmspace->vm_map;
553 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
554 	file_addr = trunc_page_ps(offset, pagesize);
555 
556 	/*
557 	 * We have two choices.  We can either clear the data in the last page
558 	 * of an oversized mapping, or we can start the anon mapping a page
559 	 * early and copy the initialized data into that first page.  We
560 	 * choose the second.
561 	 */
562 	if (filsz == 0)
563 		map_len = 0;
564 	else if (memsz > filsz)
565 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
566 	else
567 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
568 
569 	if (map_len != 0) {
570 		/* cow flags: don't dump readonly sections in core */
571 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
572 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
573 
574 		rv = __elfN(map_insert)(imgp, map,
575 				      object,
576 				      file_addr,	/* file offset */
577 				      map_addr,		/* virtual start */
578 				      map_addr + map_len,/* virtual end */
579 				      prot,
580 				      cow);
581 		if (rv != KERN_SUCCESS)
582 			return (EINVAL);
583 
584 		/* we can stop now if we've covered it all */
585 		if (memsz == filsz)
586 			return (0);
587 	}
588 
589 
590 	/*
591 	 * We have to get the remaining bit of the file into the first part
592 	 * of the oversized map segment.  This is normally because the .data
593 	 * segment in the file is extended to provide bss.  It's a neat idea
594 	 * to try and save a page, but it's a pain in the behind to implement.
595 	 */
596 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
597 	    filsz, pagesize);
598 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
599 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
600 	    map_addr;
601 
602 	/* This had damn well better be true! */
603 	if (map_len != 0) {
604 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
605 		    map_addr + map_len, prot, 0);
606 		if (rv != KERN_SUCCESS)
607 			return (EINVAL);
608 	}
609 
610 	if (copy_len != 0) {
611 		sf = vm_imgact_map_page(object, offset + filsz);
612 		if (sf == NULL)
613 			return (EIO);
614 
615 		/* send the page fragment to user space */
616 		off = trunc_page_ps(offset + filsz, pagesize) -
617 		    trunc_page(offset + filsz);
618 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
619 		    (caddr_t)map_addr, copy_len);
620 		vm_imgact_unmap_page(sf);
621 		if (error != 0)
622 			return (error);
623 	}
624 
625 	/*
626 	 * Remove write access to the page if it was only granted by map_insert
627 	 * to allow copyout.
628 	 */
629 	if ((prot & VM_PROT_WRITE) == 0)
630 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
631 		    map_len), prot, FALSE);
632 
633 	return (0);
634 }
635 
636 /*
637  * Load the file "file" into memory.  It may be either a shared object
638  * or an executable.
639  *
640  * The "addr" reference parameter is in/out.  On entry, it specifies
641  * the address where a shared object should be loaded.  If the file is
642  * an executable, this value is ignored.  On exit, "addr" specifies
643  * where the file was actually loaded.
644  *
645  * The "entry" reference parameter is out only.  On exit, it specifies
646  * the entry point for the loaded file.
647  */
648 static int
649 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
650 	u_long *entry, size_t pagesize)
651 {
652 	struct {
653 		struct nameidata nd;
654 		struct vattr attr;
655 		struct image_params image_params;
656 	} *tempdata;
657 	const Elf_Ehdr *hdr = NULL;
658 	const Elf_Phdr *phdr = NULL;
659 	struct nameidata *nd;
660 	struct vattr *attr;
661 	struct image_params *imgp;
662 	vm_prot_t prot;
663 	u_long rbase;
664 	u_long base_addr = 0;
665 	int error, i, numsegs;
666 
667 #ifdef CAPABILITY_MODE
668 	/*
669 	 * XXXJA: This check can go away once we are sufficiently confident
670 	 * that the checks in namei() are correct.
671 	 */
672 	if (IN_CAPABILITY_MODE(curthread))
673 		return (ECAPMODE);
674 #endif
675 
676 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
677 	nd = &tempdata->nd;
678 	attr = &tempdata->attr;
679 	imgp = &tempdata->image_params;
680 
681 	/*
682 	 * Initialize part of the common data
683 	 */
684 	imgp->proc = p;
685 	imgp->attr = attr;
686 	imgp->firstpage = NULL;
687 	imgp->image_header = NULL;
688 	imgp->object = NULL;
689 	imgp->execlabel = NULL;
690 
691 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
692 	if ((error = namei(nd)) != 0) {
693 		nd->ni_vp = NULL;
694 		goto fail;
695 	}
696 	NDFREE(nd, NDF_ONLY_PNBUF);
697 	imgp->vp = nd->ni_vp;
698 
699 	/*
700 	 * Check permissions, modes, uid, etc on the file, and "open" it.
701 	 */
702 	error = exec_check_permissions(imgp);
703 	if (error)
704 		goto fail;
705 
706 	error = exec_map_first_page(imgp);
707 	if (error)
708 		goto fail;
709 
710 	/*
711 	 * Also make certain that the interpreter stays the same, so set
712 	 * its VV_TEXT flag, too.
713 	 */
714 	VOP_SET_TEXT(nd->ni_vp);
715 
716 	imgp->object = nd->ni_vp->v_object;
717 
718 	hdr = (const Elf_Ehdr *)imgp->image_header;
719 	if ((error = __elfN(check_header)(hdr)) != 0)
720 		goto fail;
721 	if (hdr->e_type == ET_DYN)
722 		rbase = *addr;
723 	else if (hdr->e_type == ET_EXEC)
724 		rbase = 0;
725 	else {
726 		error = ENOEXEC;
727 		goto fail;
728 	}
729 
730 	/* Only support headers that fit within first page for now      */
731 	if ((hdr->e_phoff > PAGE_SIZE) ||
732 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
733 		error = ENOEXEC;
734 		goto fail;
735 	}
736 
737 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
738 	if (!aligned(phdr, Elf_Addr)) {
739 		error = ENOEXEC;
740 		goto fail;
741 	}
742 
743 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
744 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
745 			/* Loadable segment */
746 			prot = __elfN(trans_prot)(phdr[i].p_flags);
747 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
748 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
749 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
750 			if (error != 0)
751 				goto fail;
752 			/*
753 			 * Establish the base address if this is the
754 			 * first segment.
755 			 */
756 			if (numsegs == 0)
757   				base_addr = trunc_page(phdr[i].p_vaddr +
758 				    rbase);
759 			numsegs++;
760 		}
761 	}
762 	*addr = base_addr;
763 	*entry = (unsigned long)hdr->e_entry + rbase;
764 
765 fail:
766 	if (imgp->firstpage)
767 		exec_unmap_first_page(imgp);
768 
769 	if (nd->ni_vp)
770 		vput(nd->ni_vp);
771 
772 	free(tempdata, M_TEMP);
773 
774 	return (error);
775 }
776 
777 static int
778 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
779 {
780 	struct thread *td;
781 	const Elf_Ehdr *hdr;
782 	const Elf_Phdr *phdr;
783 	Elf_Auxargs *elf_auxargs;
784 	struct vmspace *vmspace;
785 	const char *err_str, *newinterp;
786 	char *interp, *interp_buf, *path;
787 	Elf_Brandinfo *brand_info;
788 	struct sysentvec *sv;
789 	vm_prot_t prot;
790 	u_long text_size, data_size, total_size, text_addr, data_addr;
791 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
792 	int32_t osrel;
793 	int error, i, n, interp_name_len, have_interp;
794 
795 	hdr = (const Elf_Ehdr *)imgp->image_header;
796 
797 	/*
798 	 * Do we have a valid ELF header ?
799 	 *
800 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
801 	 * if particular brand doesn't support it.
802 	 */
803 	if (__elfN(check_header)(hdr) != 0 ||
804 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
805 		return (-1);
806 
807 	/*
808 	 * From here on down, we return an errno, not -1, as we've
809 	 * detected an ELF file.
810 	 */
811 
812 	if ((hdr->e_phoff > PAGE_SIZE) ||
813 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
814 		/* Only support headers in first page for now */
815 		uprintf("Program headers not in the first page\n");
816 		return (ENOEXEC);
817 	}
818 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
819 	if (!aligned(phdr, Elf_Addr)) {
820 		uprintf("Unaligned program headers\n");
821 		return (ENOEXEC);
822 	}
823 
824 	n = error = 0;
825 	baddr = 0;
826 	osrel = 0;
827 	text_size = data_size = total_size = text_addr = data_addr = 0;
828 	entry = proghdr = 0;
829 	interp_name_len = 0;
830 	err_str = newinterp = NULL;
831 	interp = interp_buf = NULL;
832 	td = curthread;
833 
834 	for (i = 0; i < hdr->e_phnum; i++) {
835 		switch (phdr[i].p_type) {
836 		case PT_LOAD:
837 			if (n == 0)
838 				baddr = phdr[i].p_vaddr;
839 			n++;
840 			break;
841 		case PT_INTERP:
842 			/* Path to interpreter */
843 			if (phdr[i].p_filesz < 2 ||
844 			    phdr[i].p_filesz > MAXPATHLEN) {
845 				uprintf("Invalid PT_INTERP\n");
846 				error = ENOEXEC;
847 				goto ret;
848 			}
849 			if (interp != NULL) {
850 				uprintf("Multiple PT_INTERP headers\n");
851 				error = ENOEXEC;
852 				goto ret;
853 			}
854 			interp_name_len = phdr[i].p_filesz;
855 			if (phdr[i].p_offset > PAGE_SIZE ||
856 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
857 				VOP_UNLOCK(imgp->vp, 0);
858 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
859 				    M_WAITOK);
860 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
861 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
862 				    interp_name_len, phdr[i].p_offset,
863 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
864 				    NOCRED, NULL, td);
865 				if (error != 0) {
866 					uprintf("i/o error PT_INTERP %d\n",
867 					    error);
868 					goto ret;
869 				}
870 				interp_buf[interp_name_len] = '\0';
871 				interp = interp_buf;
872 			} else {
873 				interp = __DECONST(char *, imgp->image_header) +
874 				    phdr[i].p_offset;
875 				if (interp[interp_name_len - 1] != '\0') {
876 					uprintf("Invalid PT_INTERP\n");
877 					error = ENOEXEC;
878 					goto ret;
879 				}
880 			}
881 			break;
882 		case PT_GNU_STACK:
883 			if (__elfN(nxstack))
884 				imgp->stack_prot =
885 				    __elfN(trans_prot)(phdr[i].p_flags);
886 			imgp->stack_sz = phdr[i].p_memsz;
887 			break;
888 		}
889 	}
890 
891 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
892 	    &osrel);
893 	if (brand_info == NULL) {
894 		uprintf("ELF binary type \"%u\" not known.\n",
895 		    hdr->e_ident[EI_OSABI]);
896 		error = ENOEXEC;
897 		goto ret;
898 	}
899 	et_dyn_addr = 0;
900 	if (hdr->e_type == ET_DYN) {
901 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
902 			uprintf("Cannot execute shared object\n");
903 			error = ENOEXEC;
904 			goto ret;
905 		}
906 		/*
907 		 * Honour the base load address from the dso if it is
908 		 * non-zero for some reason.
909 		 */
910 		if (baddr == 0)
911 			et_dyn_addr = ET_DYN_LOAD_ADDR;
912 	}
913 	sv = brand_info->sysvec;
914 	if (interp != NULL && brand_info->interp_newpath != NULL)
915 		newinterp = brand_info->interp_newpath;
916 
917 	/*
918 	 * Avoid a possible deadlock if the current address space is destroyed
919 	 * and that address space maps the locked vnode.  In the common case,
920 	 * the locked vnode's v_usecount is decremented but remains greater
921 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
922 	 * However, in cases where the vnode lock is external, such as nullfs,
923 	 * v_usecount may become zero.
924 	 *
925 	 * The VV_TEXT flag prevents modifications to the executable while
926 	 * the vnode is unlocked.
927 	 */
928 	VOP_UNLOCK(imgp->vp, 0);
929 
930 	error = exec_new_vmspace(imgp, sv);
931 	imgp->proc->p_sysent = sv;
932 
933 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
934 	if (error != 0)
935 		goto ret;
936 
937 	for (i = 0; i < hdr->e_phnum; i++) {
938 		switch (phdr[i].p_type) {
939 		case PT_LOAD:	/* Loadable segment */
940 			if (phdr[i].p_memsz == 0)
941 				break;
942 			prot = __elfN(trans_prot)(phdr[i].p_flags);
943 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
944 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
945 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
946 			    sv->sv_pagesize);
947 			if (error != 0)
948 				goto ret;
949 
950 			/*
951 			 * If this segment contains the program headers,
952 			 * remember their virtual address for the AT_PHDR
953 			 * aux entry. Static binaries don't usually include
954 			 * a PT_PHDR entry.
955 			 */
956 			if (phdr[i].p_offset == 0 &&
957 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
958 				<= phdr[i].p_filesz)
959 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
960 				    et_dyn_addr;
961 
962 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
963 			seg_size = round_page(phdr[i].p_memsz +
964 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
965 
966 			/*
967 			 * Make the largest executable segment the official
968 			 * text segment and all others data.
969 			 *
970 			 * Note that obreak() assumes that data_addr +
971 			 * data_size == end of data load area, and the ELF
972 			 * file format expects segments to be sorted by
973 			 * address.  If multiple data segments exist, the
974 			 * last one will be used.
975 			 */
976 
977 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
978 				text_size = seg_size;
979 				text_addr = seg_addr;
980 			} else {
981 				data_size = seg_size;
982 				data_addr = seg_addr;
983 			}
984 			total_size += seg_size;
985 			break;
986 		case PT_PHDR: 	/* Program header table info */
987 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
988 			break;
989 		default:
990 			break;
991 		}
992 	}
993 
994 	if (data_addr == 0 && data_size == 0) {
995 		data_addr = text_addr;
996 		data_size = text_size;
997 	}
998 
999 	entry = (u_long)hdr->e_entry + et_dyn_addr;
1000 
1001 	/*
1002 	 * Check limits.  It should be safe to check the
1003 	 * limits after loading the segments since we do
1004 	 * not actually fault in all the segments pages.
1005 	 */
1006 	PROC_LOCK(imgp->proc);
1007 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1008 		err_str = "Data segment size exceeds process limit";
1009 	else if (text_size > maxtsiz)
1010 		err_str = "Text segment size exceeds system limit";
1011 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1012 		err_str = "Total segment size exceeds process limit";
1013 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1014 		err_str = "Data segment size exceeds resource limit";
1015 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1016 		err_str = "Total segment size exceeds resource limit";
1017 	if (err_str != NULL) {
1018 		PROC_UNLOCK(imgp->proc);
1019 		uprintf("%s\n", err_str);
1020 		error = ENOMEM;
1021 		goto ret;
1022 	}
1023 
1024 	vmspace = imgp->proc->p_vmspace;
1025 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1026 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1027 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1028 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1029 
1030 	/*
1031 	 * We load the dynamic linker where a userland call
1032 	 * to mmap(0, ...) would put it.  The rationale behind this
1033 	 * calculation is that it leaves room for the heap to grow to
1034 	 * its maximum allowed size.
1035 	 */
1036 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1037 	    RLIMIT_DATA));
1038 	PROC_UNLOCK(imgp->proc);
1039 
1040 	imgp->entry_addr = entry;
1041 
1042 	if (interp != NULL) {
1043 		have_interp = FALSE;
1044 		VOP_UNLOCK(imgp->vp, 0);
1045 		if (brand_info->emul_path != NULL &&
1046 		    brand_info->emul_path[0] != '\0') {
1047 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1048 			snprintf(path, MAXPATHLEN, "%s%s",
1049 			    brand_info->emul_path, interp);
1050 			error = __elfN(load_file)(imgp->proc, path, &addr,
1051 			    &imgp->entry_addr, sv->sv_pagesize);
1052 			free(path, M_TEMP);
1053 			if (error == 0)
1054 				have_interp = TRUE;
1055 		}
1056 		if (!have_interp && newinterp != NULL &&
1057 		    (brand_info->interp_path == NULL ||
1058 		    strcmp(interp, brand_info->interp_path) == 0)) {
1059 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1060 			    &imgp->entry_addr, sv->sv_pagesize);
1061 			if (error == 0)
1062 				have_interp = TRUE;
1063 		}
1064 		if (!have_interp) {
1065 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1066 			    &imgp->entry_addr, sv->sv_pagesize);
1067 		}
1068 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1069 		if (error != 0) {
1070 			uprintf("ELF interpreter %s not found, error %d\n",
1071 			    interp, error);
1072 			goto ret;
1073 		}
1074 	} else
1075 		addr = et_dyn_addr;
1076 
1077 	/*
1078 	 * Construct auxargs table (used by the fixup routine)
1079 	 */
1080 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1081 	elf_auxargs->execfd = -1;
1082 	elf_auxargs->phdr = proghdr;
1083 	elf_auxargs->phent = hdr->e_phentsize;
1084 	elf_auxargs->phnum = hdr->e_phnum;
1085 	elf_auxargs->pagesz = PAGE_SIZE;
1086 	elf_auxargs->base = addr;
1087 	elf_auxargs->flags = 0;
1088 	elf_auxargs->entry = entry;
1089 	elf_auxargs->hdr_eflags = hdr->e_flags;
1090 
1091 	imgp->auxargs = elf_auxargs;
1092 	imgp->interpreted = 0;
1093 	imgp->reloc_base = addr;
1094 	imgp->proc->p_osrel = osrel;
1095 	imgp->proc->p_elf_machine = hdr->e_machine;
1096 	imgp->proc->p_elf_flags = hdr->e_flags;
1097 
1098 ret:
1099 	free(interp_buf, M_TEMP);
1100 	return (error);
1101 }
1102 
1103 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1104 
1105 int
1106 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1107 {
1108 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1109 	Elf_Auxinfo *argarray, *pos;
1110 	Elf_Addr *base, *auxbase;
1111 	int error;
1112 
1113 	base = (Elf_Addr *)*stack_base;
1114 	auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1115 	argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1116 	    M_WAITOK | M_ZERO);
1117 
1118 	if (args->execfd != -1)
1119 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1120 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1121 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1122 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1123 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1124 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1125 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1126 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1127 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1128 	if (imgp->execpathp != 0)
1129 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1130 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1131 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1132 	if (imgp->canary != 0) {
1133 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1134 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1135 	}
1136 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1137 	if (imgp->pagesizes != 0) {
1138 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1139 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1140 	}
1141 	if (imgp->sysent->sv_timekeep_base != 0) {
1142 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1143 		    imgp->sysent->sv_timekeep_base);
1144 	}
1145 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1146 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1147 	    imgp->sysent->sv_stackprot);
1148 	if (imgp->sysent->sv_hwcap != NULL)
1149 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1150 	if (imgp->sysent->sv_hwcap2 != NULL)
1151 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1152 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1153 
1154 	free(imgp->auxargs, M_TEMP);
1155 	imgp->auxargs = NULL;
1156 	KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1157 
1158 	error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1159 	free(argarray, M_TEMP);
1160 	if (error != 0)
1161 		return (error);
1162 
1163 	base--;
1164 	if (suword(base, imgp->args->argc) == -1)
1165 		return (EFAULT);
1166 	*stack_base = (register_t *)base;
1167 	return (0);
1168 }
1169 
1170 /*
1171  * Code for generating ELF core dumps.
1172  */
1173 
1174 typedef void (*segment_callback)(vm_map_entry_t, void *);
1175 
1176 /* Closure for cb_put_phdr(). */
1177 struct phdr_closure {
1178 	Elf_Phdr *phdr;		/* Program header to fill in */
1179 	Elf_Off offset;		/* Offset of segment in core file */
1180 };
1181 
1182 /* Closure for cb_size_segment(). */
1183 struct sseg_closure {
1184 	int count;		/* Count of writable segments. */
1185 	size_t size;		/* Total size of all writable segments. */
1186 };
1187 
1188 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1189 
1190 struct note_info {
1191 	int		type;		/* Note type. */
1192 	outfunc_t 	outfunc; 	/* Output function. */
1193 	void		*outarg;	/* Argument for the output function. */
1194 	size_t		outsize;	/* Output size. */
1195 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1196 };
1197 
1198 TAILQ_HEAD(note_info_list, note_info);
1199 
1200 /* Coredump output parameters. */
1201 struct coredump_params {
1202 	off_t		offset;
1203 	struct ucred	*active_cred;
1204 	struct ucred	*file_cred;
1205 	struct thread	*td;
1206 	struct vnode	*vp;
1207 	struct compressor *comp;
1208 };
1209 
1210 extern int compress_user_cores;
1211 extern int compress_user_cores_level;
1212 
1213 static void cb_put_phdr(vm_map_entry_t, void *);
1214 static void cb_size_segment(vm_map_entry_t, void *);
1215 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1216     enum uio_seg);
1217 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1218 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1219     struct note_info_list *, size_t);
1220 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1221     size_t *);
1222 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1223 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1224 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1225 static int sbuf_drain_core_output(void *, const char *, int);
1226 static int sbuf_drain_count(void *arg, const char *data, int len);
1227 
1228 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1229 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1230 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1231 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1232 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1233 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1234 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1235 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1236 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1237 static void note_procstat_files(void *, struct sbuf *, size_t *);
1238 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1239 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1240 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1241 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1242 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1243 
1244 /*
1245  * Write out a core segment to the compression stream.
1246  */
1247 static int
1248 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1249 {
1250 	u_int chunk_len;
1251 	int error;
1252 
1253 	while (len > 0) {
1254 		chunk_len = MIN(len, CORE_BUF_SIZE);
1255 
1256 		/*
1257 		 * We can get EFAULT error here.
1258 		 * In that case zero out the current chunk of the segment.
1259 		 */
1260 		error = copyin(base, buf, chunk_len);
1261 		if (error != 0)
1262 			bzero(buf, chunk_len);
1263 		error = compressor_write(p->comp, buf, chunk_len);
1264 		if (error != 0)
1265 			break;
1266 		base += chunk_len;
1267 		len -= chunk_len;
1268 	}
1269 	return (error);
1270 }
1271 
1272 static int
1273 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1274 {
1275 
1276 	return (core_write((struct coredump_params *)arg, base, len, offset,
1277 	    UIO_SYSSPACE));
1278 }
1279 
1280 static int
1281 core_write(struct coredump_params *p, const void *base, size_t len,
1282     off_t offset, enum uio_seg seg)
1283 {
1284 
1285 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1286 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1287 	    p->active_cred, p->file_cred, NULL, p->td));
1288 }
1289 
1290 static int
1291 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1292     void *tmpbuf)
1293 {
1294 	int error;
1295 
1296 	if (p->comp != NULL)
1297 		return (compress_chunk(p, base, tmpbuf, len));
1298 
1299 	/*
1300 	 * EFAULT is a non-fatal error that we can get, for example,
1301 	 * if the segment is backed by a file but extends beyond its
1302 	 * end.
1303 	 */
1304 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1305 	if (error == EFAULT) {
1306 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1307 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1308 		    "for process %s\n", base, len, offset, curproc->p_comm);
1309 
1310 		/*
1311 		 * Write a "real" zero byte at the end of the target region
1312 		 * in the case this is the last segment.
1313 		 * The intermediate space will be implicitly zero-filled.
1314 		 */
1315 		error = core_write(p, zero_region, 1, offset + len - 1,
1316 		    UIO_SYSSPACE);
1317 	}
1318 	return (error);
1319 }
1320 
1321 /*
1322  * Drain into a core file.
1323  */
1324 static int
1325 sbuf_drain_core_output(void *arg, const char *data, int len)
1326 {
1327 	struct coredump_params *p;
1328 	int error, locked;
1329 
1330 	p = (struct coredump_params *)arg;
1331 
1332 	/*
1333 	 * Some kern_proc out routines that print to this sbuf may
1334 	 * call us with the process lock held. Draining with the
1335 	 * non-sleepable lock held is unsafe. The lock is needed for
1336 	 * those routines when dumping a live process. In our case we
1337 	 * can safely release the lock before draining and acquire
1338 	 * again after.
1339 	 */
1340 	locked = PROC_LOCKED(p->td->td_proc);
1341 	if (locked)
1342 		PROC_UNLOCK(p->td->td_proc);
1343 	if (p->comp != NULL)
1344 		error = compressor_write(p->comp, __DECONST(char *, data), len);
1345 	else
1346 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1347 		    UIO_SYSSPACE);
1348 	if (locked)
1349 		PROC_LOCK(p->td->td_proc);
1350 	if (error != 0)
1351 		return (-error);
1352 	p->offset += len;
1353 	return (len);
1354 }
1355 
1356 /*
1357  * Drain into a counter.
1358  */
1359 static int
1360 sbuf_drain_count(void *arg, const char *data __unused, int len)
1361 {
1362 	size_t *sizep;
1363 
1364 	sizep = (size_t *)arg;
1365 	*sizep += len;
1366 	return (len);
1367 }
1368 
1369 int
1370 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1371 {
1372 	struct ucred *cred = td->td_ucred;
1373 	int error = 0;
1374 	struct sseg_closure seginfo;
1375 	struct note_info_list notelst;
1376 	struct coredump_params params;
1377 	struct note_info *ninfo;
1378 	void *hdr, *tmpbuf;
1379 	size_t hdrsize, notesz, coresize;
1380 
1381 	hdr = NULL;
1382 	tmpbuf = NULL;
1383 	TAILQ_INIT(&notelst);
1384 
1385 	/* Size the program segments. */
1386 	seginfo.count = 0;
1387 	seginfo.size = 0;
1388 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1389 
1390 	/*
1391 	 * Collect info about the core file header area.
1392 	 */
1393 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1394 	if (seginfo.count + 1 >= PN_XNUM)
1395 		hdrsize += sizeof(Elf_Shdr);
1396 	__elfN(prepare_notes)(td, &notelst, &notesz);
1397 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1398 
1399 	/* Set up core dump parameters. */
1400 	params.offset = 0;
1401 	params.active_cred = cred;
1402 	params.file_cred = NOCRED;
1403 	params.td = td;
1404 	params.vp = vp;
1405 	params.comp = NULL;
1406 
1407 #ifdef RACCT
1408 	if (racct_enable) {
1409 		PROC_LOCK(td->td_proc);
1410 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1411 		PROC_UNLOCK(td->td_proc);
1412 		if (error != 0) {
1413 			error = EFAULT;
1414 			goto done;
1415 		}
1416 	}
1417 #endif
1418 	if (coresize >= limit) {
1419 		error = EFAULT;
1420 		goto done;
1421 	}
1422 
1423 	/* Create a compression stream if necessary. */
1424 	if (compress_user_cores != 0) {
1425 		params.comp = compressor_init(core_compressed_write,
1426 		    compress_user_cores, CORE_BUF_SIZE,
1427 		    compress_user_cores_level, &params);
1428 		if (params.comp == NULL) {
1429 			error = EFAULT;
1430 			goto done;
1431 		}
1432 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1433         }
1434 
1435 	/*
1436 	 * Allocate memory for building the header, fill it up,
1437 	 * and write it out following the notes.
1438 	 */
1439 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1440 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1441 	    notesz);
1442 
1443 	/* Write the contents of all of the writable segments. */
1444 	if (error == 0) {
1445 		Elf_Phdr *php;
1446 		off_t offset;
1447 		int i;
1448 
1449 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1450 		offset = round_page(hdrsize + notesz);
1451 		for (i = 0; i < seginfo.count; i++) {
1452 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1453 			    php->p_filesz, offset, &params, tmpbuf);
1454 			if (error != 0)
1455 				break;
1456 			offset += php->p_filesz;
1457 			php++;
1458 		}
1459 		if (error == 0 && params.comp != NULL)
1460 			error = compressor_flush(params.comp);
1461 	}
1462 	if (error) {
1463 		log(LOG_WARNING,
1464 		    "Failed to write core file for process %s (error %d)\n",
1465 		    curproc->p_comm, error);
1466 	}
1467 
1468 done:
1469 	free(tmpbuf, M_TEMP);
1470 	if (params.comp != NULL)
1471 		compressor_fini(params.comp);
1472 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1473 		TAILQ_REMOVE(&notelst, ninfo, link);
1474 		free(ninfo, M_TEMP);
1475 	}
1476 	if (hdr != NULL)
1477 		free(hdr, M_TEMP);
1478 
1479 	return (error);
1480 }
1481 
1482 /*
1483  * A callback for each_dumpable_segment() to write out the segment's
1484  * program header entry.
1485  */
1486 static void
1487 cb_put_phdr(vm_map_entry_t entry, void *closure)
1488 {
1489 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1490 	Elf_Phdr *phdr = phc->phdr;
1491 
1492 	phc->offset = round_page(phc->offset);
1493 
1494 	phdr->p_type = PT_LOAD;
1495 	phdr->p_offset = phc->offset;
1496 	phdr->p_vaddr = entry->start;
1497 	phdr->p_paddr = 0;
1498 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1499 	phdr->p_align = PAGE_SIZE;
1500 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1501 
1502 	phc->offset += phdr->p_filesz;
1503 	phc->phdr++;
1504 }
1505 
1506 /*
1507  * A callback for each_dumpable_segment() to gather information about
1508  * the number of segments and their total size.
1509  */
1510 static void
1511 cb_size_segment(vm_map_entry_t entry, void *closure)
1512 {
1513 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1514 
1515 	ssc->count++;
1516 	ssc->size += entry->end - entry->start;
1517 }
1518 
1519 /*
1520  * For each writable segment in the process's memory map, call the given
1521  * function with a pointer to the map entry and some arbitrary
1522  * caller-supplied data.
1523  */
1524 static void
1525 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1526 {
1527 	struct proc *p = td->td_proc;
1528 	vm_map_t map = &p->p_vmspace->vm_map;
1529 	vm_map_entry_t entry;
1530 	vm_object_t backing_object, object;
1531 	boolean_t ignore_entry;
1532 
1533 	vm_map_lock_read(map);
1534 	for (entry = map->header.next; entry != &map->header;
1535 	    entry = entry->next) {
1536 		/*
1537 		 * Don't dump inaccessible mappings, deal with legacy
1538 		 * coredump mode.
1539 		 *
1540 		 * Note that read-only segments related to the elf binary
1541 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1542 		 * need to arbitrarily ignore such segments.
1543 		 */
1544 		if (elf_legacy_coredump) {
1545 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1546 				continue;
1547 		} else {
1548 			if ((entry->protection & VM_PROT_ALL) == 0)
1549 				continue;
1550 		}
1551 
1552 		/*
1553 		 * Dont include memory segment in the coredump if
1554 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1555 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1556 		 * kernel map).
1557 		 */
1558 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1559 			continue;
1560 
1561 		if ((object = entry->object.vm_object) == NULL)
1562 			continue;
1563 
1564 		/* Ignore memory-mapped devices and such things. */
1565 		VM_OBJECT_RLOCK(object);
1566 		while ((backing_object = object->backing_object) != NULL) {
1567 			VM_OBJECT_RLOCK(backing_object);
1568 			VM_OBJECT_RUNLOCK(object);
1569 			object = backing_object;
1570 		}
1571 		ignore_entry = object->type != OBJT_DEFAULT &&
1572 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1573 		    object->type != OBJT_PHYS;
1574 		VM_OBJECT_RUNLOCK(object);
1575 		if (ignore_entry)
1576 			continue;
1577 
1578 		(*func)(entry, closure);
1579 	}
1580 	vm_map_unlock_read(map);
1581 }
1582 
1583 /*
1584  * Write the core file header to the file, including padding up to
1585  * the page boundary.
1586  */
1587 static int
1588 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1589     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1590 {
1591 	struct note_info *ninfo;
1592 	struct sbuf *sb;
1593 	int error;
1594 
1595 	/* Fill in the header. */
1596 	bzero(hdr, hdrsize);
1597 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1598 
1599 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1600 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1601 	sbuf_start_section(sb, NULL);
1602 	sbuf_bcat(sb, hdr, hdrsize);
1603 	TAILQ_FOREACH(ninfo, notelst, link)
1604 	    __elfN(putnote)(ninfo, sb);
1605 	/* Align up to a page boundary for the program segments. */
1606 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1607 	error = sbuf_finish(sb);
1608 	sbuf_delete(sb);
1609 
1610 	return (error);
1611 }
1612 
1613 static void
1614 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1615     size_t *sizep)
1616 {
1617 	struct proc *p;
1618 	struct thread *thr;
1619 	size_t size;
1620 
1621 	p = td->td_proc;
1622 	size = 0;
1623 
1624 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1625 
1626 	/*
1627 	 * To have the debugger select the right thread (LWP) as the initial
1628 	 * thread, we dump the state of the thread passed to us in td first.
1629 	 * This is the thread that causes the core dump and thus likely to
1630 	 * be the right thread one wants to have selected in the debugger.
1631 	 */
1632 	thr = td;
1633 	while (thr != NULL) {
1634 		size += register_note(list, NT_PRSTATUS,
1635 		    __elfN(note_prstatus), thr);
1636 		size += register_note(list, NT_FPREGSET,
1637 		    __elfN(note_fpregset), thr);
1638 		size += register_note(list, NT_THRMISC,
1639 		    __elfN(note_thrmisc), thr);
1640 		size += register_note(list, NT_PTLWPINFO,
1641 		    __elfN(note_ptlwpinfo), thr);
1642 		size += register_note(list, -1,
1643 		    __elfN(note_threadmd), thr);
1644 
1645 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1646 		    TAILQ_NEXT(thr, td_plist);
1647 		if (thr == td)
1648 			thr = TAILQ_NEXT(thr, td_plist);
1649 	}
1650 
1651 	size += register_note(list, NT_PROCSTAT_PROC,
1652 	    __elfN(note_procstat_proc), p);
1653 	size += register_note(list, NT_PROCSTAT_FILES,
1654 	    note_procstat_files, p);
1655 	size += register_note(list, NT_PROCSTAT_VMMAP,
1656 	    note_procstat_vmmap, p);
1657 	size += register_note(list, NT_PROCSTAT_GROUPS,
1658 	    note_procstat_groups, p);
1659 	size += register_note(list, NT_PROCSTAT_UMASK,
1660 	    note_procstat_umask, p);
1661 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1662 	    note_procstat_rlimit, p);
1663 	size += register_note(list, NT_PROCSTAT_OSREL,
1664 	    note_procstat_osrel, p);
1665 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1666 	    __elfN(note_procstat_psstrings), p);
1667 	size += register_note(list, NT_PROCSTAT_AUXV,
1668 	    __elfN(note_procstat_auxv), p);
1669 
1670 	*sizep = size;
1671 }
1672 
1673 static void
1674 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1675     size_t notesz)
1676 {
1677 	Elf_Ehdr *ehdr;
1678 	Elf_Phdr *phdr;
1679 	Elf_Shdr *shdr;
1680 	struct phdr_closure phc;
1681 
1682 	ehdr = (Elf_Ehdr *)hdr;
1683 
1684 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1685 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1686 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1687 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1688 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1689 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1690 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1691 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1692 	ehdr->e_ident[EI_ABIVERSION] = 0;
1693 	ehdr->e_ident[EI_PAD] = 0;
1694 	ehdr->e_type = ET_CORE;
1695 	ehdr->e_machine = td->td_proc->p_elf_machine;
1696 	ehdr->e_version = EV_CURRENT;
1697 	ehdr->e_entry = 0;
1698 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1699 	ehdr->e_flags = td->td_proc->p_elf_flags;
1700 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1701 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1702 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1703 	ehdr->e_shstrndx = SHN_UNDEF;
1704 	if (numsegs + 1 < PN_XNUM) {
1705 		ehdr->e_phnum = numsegs + 1;
1706 		ehdr->e_shnum = 0;
1707 	} else {
1708 		ehdr->e_phnum = PN_XNUM;
1709 		ehdr->e_shnum = 1;
1710 
1711 		ehdr->e_shoff = ehdr->e_phoff +
1712 		    (numsegs + 1) * ehdr->e_phentsize;
1713 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1714 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1715 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1716 
1717 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1718 		memset(shdr, 0, sizeof(*shdr));
1719 		/*
1720 		 * A special first section is used to hold large segment and
1721 		 * section counts.  This was proposed by Sun Microsystems in
1722 		 * Solaris and has been adopted by Linux; the standard ELF
1723 		 * tools are already familiar with the technique.
1724 		 *
1725 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1726 		 * (or 12-7 depending on the version of the document) for more
1727 		 * details.
1728 		 */
1729 		shdr->sh_type = SHT_NULL;
1730 		shdr->sh_size = ehdr->e_shnum;
1731 		shdr->sh_link = ehdr->e_shstrndx;
1732 		shdr->sh_info = numsegs + 1;
1733 	}
1734 
1735 	/*
1736 	 * Fill in the program header entries.
1737 	 */
1738 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1739 
1740 	/* The note segement. */
1741 	phdr->p_type = PT_NOTE;
1742 	phdr->p_offset = hdrsize;
1743 	phdr->p_vaddr = 0;
1744 	phdr->p_paddr = 0;
1745 	phdr->p_filesz = notesz;
1746 	phdr->p_memsz = 0;
1747 	phdr->p_flags = PF_R;
1748 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1749 	phdr++;
1750 
1751 	/* All the writable segments from the program. */
1752 	phc.phdr = phdr;
1753 	phc.offset = round_page(hdrsize + notesz);
1754 	each_dumpable_segment(td, cb_put_phdr, &phc);
1755 }
1756 
1757 static size_t
1758 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1759 {
1760 	struct note_info *ninfo;
1761 	size_t size, notesize;
1762 
1763 	size = 0;
1764 	out(arg, NULL, &size);
1765 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1766 	ninfo->type = type;
1767 	ninfo->outfunc = out;
1768 	ninfo->outarg = arg;
1769 	ninfo->outsize = size;
1770 	TAILQ_INSERT_TAIL(list, ninfo, link);
1771 
1772 	if (type == -1)
1773 		return (size);
1774 
1775 	notesize = sizeof(Elf_Note) +		/* note header */
1776 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1777 						/* note name */
1778 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1779 
1780 	return (notesize);
1781 }
1782 
1783 static size_t
1784 append_note_data(const void *src, void *dst, size_t len)
1785 {
1786 	size_t padded_len;
1787 
1788 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1789 	if (dst != NULL) {
1790 		bcopy(src, dst, len);
1791 		bzero((char *)dst + len, padded_len - len);
1792 	}
1793 	return (padded_len);
1794 }
1795 
1796 size_t
1797 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1798 {
1799 	Elf_Note *note;
1800 	char *buf;
1801 	size_t notesize;
1802 
1803 	buf = dst;
1804 	if (buf != NULL) {
1805 		note = (Elf_Note *)buf;
1806 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1807 		note->n_descsz = size;
1808 		note->n_type = type;
1809 		buf += sizeof(*note);
1810 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1811 		    sizeof(FREEBSD_ABI_VENDOR));
1812 		append_note_data(src, buf, size);
1813 		if (descp != NULL)
1814 			*descp = buf;
1815 	}
1816 
1817 	notesize = sizeof(Elf_Note) +		/* note header */
1818 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1819 						/* note name */
1820 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1821 
1822 	return (notesize);
1823 }
1824 
1825 static void
1826 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1827 {
1828 	Elf_Note note;
1829 	ssize_t old_len, sect_len;
1830 	size_t new_len, descsz, i;
1831 
1832 	if (ninfo->type == -1) {
1833 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1834 		return;
1835 	}
1836 
1837 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1838 	note.n_descsz = ninfo->outsize;
1839 	note.n_type = ninfo->type;
1840 
1841 	sbuf_bcat(sb, &note, sizeof(note));
1842 	sbuf_start_section(sb, &old_len);
1843 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1844 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1845 	if (note.n_descsz == 0)
1846 		return;
1847 	sbuf_start_section(sb, &old_len);
1848 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1849 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1850 	if (sect_len < 0)
1851 		return;
1852 
1853 	new_len = (size_t)sect_len;
1854 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1855 	if (new_len < descsz) {
1856 		/*
1857 		 * It is expected that individual note emitters will correctly
1858 		 * predict their expected output size and fill up to that size
1859 		 * themselves, padding in a format-specific way if needed.
1860 		 * However, in case they don't, just do it here with zeros.
1861 		 */
1862 		for (i = 0; i < descsz - new_len; i++)
1863 			sbuf_putc(sb, 0);
1864 	} else if (new_len > descsz) {
1865 		/*
1866 		 * We can't always truncate sb -- we may have drained some
1867 		 * of it already.
1868 		 */
1869 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1870 		    "read it (%zu > %zu).  Since it is longer than "
1871 		    "expected, this coredump's notes are corrupt.  THIS "
1872 		    "IS A BUG in the note_procstat routine for type %u.\n",
1873 		    __func__, (unsigned)note.n_type, new_len, descsz,
1874 		    (unsigned)note.n_type));
1875 	}
1876 }
1877 
1878 /*
1879  * Miscellaneous note out functions.
1880  */
1881 
1882 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1883 #include <compat/freebsd32/freebsd32.h>
1884 #include <compat/freebsd32/freebsd32_signal.h>
1885 
1886 typedef struct prstatus32 elf_prstatus_t;
1887 typedef struct prpsinfo32 elf_prpsinfo_t;
1888 typedef struct fpreg32 elf_prfpregset_t;
1889 typedef struct fpreg32 elf_fpregset_t;
1890 typedef struct reg32 elf_gregset_t;
1891 typedef struct thrmisc32 elf_thrmisc_t;
1892 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1893 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1894 typedef uint32_t elf_ps_strings_t;
1895 #else
1896 typedef prstatus_t elf_prstatus_t;
1897 typedef prpsinfo_t elf_prpsinfo_t;
1898 typedef prfpregset_t elf_prfpregset_t;
1899 typedef prfpregset_t elf_fpregset_t;
1900 typedef gregset_t elf_gregset_t;
1901 typedef thrmisc_t elf_thrmisc_t;
1902 #define ELF_KERN_PROC_MASK	0
1903 typedef struct kinfo_proc elf_kinfo_proc_t;
1904 typedef vm_offset_t elf_ps_strings_t;
1905 #endif
1906 
1907 static void
1908 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1909 {
1910 	struct sbuf sbarg;
1911 	size_t len;
1912 	char *cp, *end;
1913 	struct proc *p;
1914 	elf_prpsinfo_t *psinfo;
1915 	int error;
1916 
1917 	p = (struct proc *)arg;
1918 	if (sb != NULL) {
1919 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1920 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1921 		psinfo->pr_version = PRPSINFO_VERSION;
1922 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1923 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1924 		PROC_LOCK(p);
1925 		if (p->p_args != NULL) {
1926 			len = sizeof(psinfo->pr_psargs) - 1;
1927 			if (len > p->p_args->ar_length)
1928 				len = p->p_args->ar_length;
1929 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1930 			PROC_UNLOCK(p);
1931 			error = 0;
1932 		} else {
1933 			_PHOLD(p);
1934 			PROC_UNLOCK(p);
1935 			sbuf_new(&sbarg, psinfo->pr_psargs,
1936 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1937 			error = proc_getargv(curthread, p, &sbarg);
1938 			PRELE(p);
1939 			if (sbuf_finish(&sbarg) == 0)
1940 				len = sbuf_len(&sbarg) - 1;
1941 			else
1942 				len = sizeof(psinfo->pr_psargs) - 1;
1943 			sbuf_delete(&sbarg);
1944 		}
1945 		if (error || len == 0)
1946 			strlcpy(psinfo->pr_psargs, p->p_comm,
1947 			    sizeof(psinfo->pr_psargs));
1948 		else {
1949 			KASSERT(len < sizeof(psinfo->pr_psargs),
1950 			    ("len is too long: %zu vs %zu", len,
1951 			    sizeof(psinfo->pr_psargs)));
1952 			cp = psinfo->pr_psargs;
1953 			end = cp + len - 1;
1954 			for (;;) {
1955 				cp = memchr(cp, '\0', end - cp);
1956 				if (cp == NULL)
1957 					break;
1958 				*cp = ' ';
1959 			}
1960 		}
1961 		psinfo->pr_pid = p->p_pid;
1962 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1963 		free(psinfo, M_TEMP);
1964 	}
1965 	*sizep = sizeof(*psinfo);
1966 }
1967 
1968 static void
1969 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1970 {
1971 	struct thread *td;
1972 	elf_prstatus_t *status;
1973 
1974 	td = (struct thread *)arg;
1975 	if (sb != NULL) {
1976 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1977 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1978 		status->pr_version = PRSTATUS_VERSION;
1979 		status->pr_statussz = sizeof(elf_prstatus_t);
1980 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1981 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1982 		status->pr_osreldate = osreldate;
1983 		status->pr_cursig = td->td_proc->p_sig;
1984 		status->pr_pid = td->td_tid;
1985 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1986 		fill_regs32(td, &status->pr_reg);
1987 #else
1988 		fill_regs(td, &status->pr_reg);
1989 #endif
1990 		sbuf_bcat(sb, status, sizeof(*status));
1991 		free(status, M_TEMP);
1992 	}
1993 	*sizep = sizeof(*status);
1994 }
1995 
1996 static void
1997 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1998 {
1999 	struct thread *td;
2000 	elf_prfpregset_t *fpregset;
2001 
2002 	td = (struct thread *)arg;
2003 	if (sb != NULL) {
2004 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2005 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2006 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2007 		fill_fpregs32(td, fpregset);
2008 #else
2009 		fill_fpregs(td, fpregset);
2010 #endif
2011 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2012 		free(fpregset, M_TEMP);
2013 	}
2014 	*sizep = sizeof(*fpregset);
2015 }
2016 
2017 static void
2018 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2019 {
2020 	struct thread *td;
2021 	elf_thrmisc_t thrmisc;
2022 
2023 	td = (struct thread *)arg;
2024 	if (sb != NULL) {
2025 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2026 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2027 		strcpy(thrmisc.pr_tname, td->td_name);
2028 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2029 	}
2030 	*sizep = sizeof(thrmisc);
2031 }
2032 
2033 static void
2034 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2035 {
2036 	struct thread *td;
2037 	size_t size;
2038 	int structsize;
2039 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2040 	struct ptrace_lwpinfo32 pl;
2041 #else
2042 	struct ptrace_lwpinfo pl;
2043 #endif
2044 
2045 	td = (struct thread *)arg;
2046 	size = sizeof(structsize) + sizeof(pl);
2047 	if (sb != NULL) {
2048 		KASSERT(*sizep == size, ("invalid size"));
2049 		structsize = sizeof(pl);
2050 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2051 		bzero(&pl, sizeof(pl));
2052 		pl.pl_lwpid = td->td_tid;
2053 		pl.pl_event = PL_EVENT_NONE;
2054 		pl.pl_sigmask = td->td_sigmask;
2055 		pl.pl_siglist = td->td_siglist;
2056 		if (td->td_si.si_signo != 0) {
2057 			pl.pl_event = PL_EVENT_SIGNAL;
2058 			pl.pl_flags |= PL_FLAG_SI;
2059 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2060 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2061 #else
2062 			pl.pl_siginfo = td->td_si;
2063 #endif
2064 		}
2065 		strcpy(pl.pl_tdname, td->td_name);
2066 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2067 		sbuf_bcat(sb, &pl, sizeof(pl));
2068 	}
2069 	*sizep = size;
2070 }
2071 
2072 /*
2073  * Allow for MD specific notes, as well as any MD
2074  * specific preparations for writing MI notes.
2075  */
2076 static void
2077 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2078 {
2079 	struct thread *td;
2080 	void *buf;
2081 	size_t size;
2082 
2083 	td = (struct thread *)arg;
2084 	size = *sizep;
2085 	if (size != 0 && sb != NULL)
2086 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2087 	else
2088 		buf = NULL;
2089 	size = 0;
2090 	__elfN(dump_thread)(td, buf, &size);
2091 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2092 	if (size != 0 && sb != NULL)
2093 		sbuf_bcat(sb, buf, size);
2094 	free(buf, M_TEMP);
2095 	*sizep = size;
2096 }
2097 
2098 #ifdef KINFO_PROC_SIZE
2099 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2100 #endif
2101 
2102 static void
2103 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2104 {
2105 	struct proc *p;
2106 	size_t size;
2107 	int structsize;
2108 
2109 	p = (struct proc *)arg;
2110 	size = sizeof(structsize) + p->p_numthreads *
2111 	    sizeof(elf_kinfo_proc_t);
2112 
2113 	if (sb != NULL) {
2114 		KASSERT(*sizep == size, ("invalid size"));
2115 		structsize = sizeof(elf_kinfo_proc_t);
2116 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2117 		sx_slock(&proctree_lock);
2118 		PROC_LOCK(p);
2119 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2120 		sx_sunlock(&proctree_lock);
2121 	}
2122 	*sizep = size;
2123 }
2124 
2125 #ifdef KINFO_FILE_SIZE
2126 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2127 #endif
2128 
2129 static void
2130 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2131 {
2132 	struct proc *p;
2133 	size_t size, sect_sz, i;
2134 	ssize_t start_len, sect_len;
2135 	int structsize, filedesc_flags;
2136 
2137 	if (coredump_pack_fileinfo)
2138 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2139 	else
2140 		filedesc_flags = 0;
2141 
2142 	p = (struct proc *)arg;
2143 	structsize = sizeof(struct kinfo_file);
2144 	if (sb == NULL) {
2145 		size = 0;
2146 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2147 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2148 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2149 		PROC_LOCK(p);
2150 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2151 		sbuf_finish(sb);
2152 		sbuf_delete(sb);
2153 		*sizep = size;
2154 	} else {
2155 		sbuf_start_section(sb, &start_len);
2156 
2157 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2158 		PROC_LOCK(p);
2159 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2160 		    filedesc_flags);
2161 
2162 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2163 		if (sect_len < 0)
2164 			return;
2165 		sect_sz = sect_len;
2166 
2167 		KASSERT(sect_sz <= *sizep,
2168 		    ("kern_proc_filedesc_out did not respect maxlen; "
2169 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2170 		     sect_sz - sizeof(structsize)));
2171 
2172 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2173 			sbuf_putc(sb, 0);
2174 	}
2175 }
2176 
2177 #ifdef KINFO_VMENTRY_SIZE
2178 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2179 #endif
2180 
2181 static void
2182 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2183 {
2184 	struct proc *p;
2185 	size_t size;
2186 	int structsize, vmmap_flags;
2187 
2188 	if (coredump_pack_vmmapinfo)
2189 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2190 	else
2191 		vmmap_flags = 0;
2192 
2193 	p = (struct proc *)arg;
2194 	structsize = sizeof(struct kinfo_vmentry);
2195 	if (sb == NULL) {
2196 		size = 0;
2197 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2198 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2199 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2200 		PROC_LOCK(p);
2201 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2202 		sbuf_finish(sb);
2203 		sbuf_delete(sb);
2204 		*sizep = size;
2205 	} else {
2206 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2207 		PROC_LOCK(p);
2208 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2209 		    vmmap_flags);
2210 	}
2211 }
2212 
2213 static void
2214 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2215 {
2216 	struct proc *p;
2217 	size_t size;
2218 	int structsize;
2219 
2220 	p = (struct proc *)arg;
2221 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2222 	if (sb != NULL) {
2223 		KASSERT(*sizep == size, ("invalid size"));
2224 		structsize = sizeof(gid_t);
2225 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2226 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2227 		    sizeof(gid_t));
2228 	}
2229 	*sizep = size;
2230 }
2231 
2232 static void
2233 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2234 {
2235 	struct proc *p;
2236 	size_t size;
2237 	int structsize;
2238 
2239 	p = (struct proc *)arg;
2240 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2241 	if (sb != NULL) {
2242 		KASSERT(*sizep == size, ("invalid size"));
2243 		structsize = sizeof(p->p_fd->fd_cmask);
2244 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2245 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2246 	}
2247 	*sizep = size;
2248 }
2249 
2250 static void
2251 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2252 {
2253 	struct proc *p;
2254 	struct rlimit rlim[RLIM_NLIMITS];
2255 	size_t size;
2256 	int structsize, i;
2257 
2258 	p = (struct proc *)arg;
2259 	size = sizeof(structsize) + sizeof(rlim);
2260 	if (sb != NULL) {
2261 		KASSERT(*sizep == size, ("invalid size"));
2262 		structsize = sizeof(rlim);
2263 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2264 		PROC_LOCK(p);
2265 		for (i = 0; i < RLIM_NLIMITS; i++)
2266 			lim_rlimit_proc(p, i, &rlim[i]);
2267 		PROC_UNLOCK(p);
2268 		sbuf_bcat(sb, rlim, sizeof(rlim));
2269 	}
2270 	*sizep = size;
2271 }
2272 
2273 static void
2274 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2275 {
2276 	struct proc *p;
2277 	size_t size;
2278 	int structsize;
2279 
2280 	p = (struct proc *)arg;
2281 	size = sizeof(structsize) + sizeof(p->p_osrel);
2282 	if (sb != NULL) {
2283 		KASSERT(*sizep == size, ("invalid size"));
2284 		structsize = sizeof(p->p_osrel);
2285 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2286 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2287 	}
2288 	*sizep = size;
2289 }
2290 
2291 static void
2292 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2293 {
2294 	struct proc *p;
2295 	elf_ps_strings_t ps_strings;
2296 	size_t size;
2297 	int structsize;
2298 
2299 	p = (struct proc *)arg;
2300 	size = sizeof(structsize) + sizeof(ps_strings);
2301 	if (sb != NULL) {
2302 		KASSERT(*sizep == size, ("invalid size"));
2303 		structsize = sizeof(ps_strings);
2304 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2305 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2306 #else
2307 		ps_strings = p->p_sysent->sv_psstrings;
2308 #endif
2309 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2310 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2311 	}
2312 	*sizep = size;
2313 }
2314 
2315 static void
2316 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2317 {
2318 	struct proc *p;
2319 	size_t size;
2320 	int structsize;
2321 
2322 	p = (struct proc *)arg;
2323 	if (sb == NULL) {
2324 		size = 0;
2325 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2326 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2327 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2328 		PHOLD(p);
2329 		proc_getauxv(curthread, p, sb);
2330 		PRELE(p);
2331 		sbuf_finish(sb);
2332 		sbuf_delete(sb);
2333 		*sizep = size;
2334 	} else {
2335 		structsize = sizeof(Elf_Auxinfo);
2336 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2337 		PHOLD(p);
2338 		proc_getauxv(curthread, p, sb);
2339 		PRELE(p);
2340 	}
2341 }
2342 
2343 static boolean_t
2344 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2345     int32_t *osrel, const Elf_Phdr *pnote)
2346 {
2347 	const Elf_Note *note, *note0, *note_end;
2348 	const char *note_name;
2349 	char *buf;
2350 	int i, error;
2351 	boolean_t res;
2352 
2353 	/* We need some limit, might as well use PAGE_SIZE. */
2354 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2355 		return (FALSE);
2356 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2357 	if (pnote->p_offset > PAGE_SIZE ||
2358 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2359 		VOP_UNLOCK(imgp->vp, 0);
2360 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2361 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2362 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2363 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2364 		    curthread->td_ucred, NOCRED, NULL, curthread);
2365 		if (error != 0) {
2366 			uprintf("i/o error PT_NOTE\n");
2367 			res = FALSE;
2368 			goto ret;
2369 		}
2370 		note = note0 = (const Elf_Note *)buf;
2371 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2372 	} else {
2373 		note = note0 = (const Elf_Note *)(imgp->image_header +
2374 		    pnote->p_offset);
2375 		note_end = (const Elf_Note *)(imgp->image_header +
2376 		    pnote->p_offset + pnote->p_filesz);
2377 		buf = NULL;
2378 	}
2379 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2380 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2381 		    (const char *)note < sizeof(Elf_Note)) {
2382 			res = FALSE;
2383 			goto ret;
2384 		}
2385 		if (note->n_namesz != checknote->hdr.n_namesz ||
2386 		    note->n_descsz != checknote->hdr.n_descsz ||
2387 		    note->n_type != checknote->hdr.n_type)
2388 			goto nextnote;
2389 		note_name = (const char *)(note + 1);
2390 		if (note_name + checknote->hdr.n_namesz >=
2391 		    (const char *)note_end || strncmp(checknote->vendor,
2392 		    note_name, checknote->hdr.n_namesz) != 0)
2393 			goto nextnote;
2394 
2395 		/*
2396 		 * Fetch the osreldate for binary
2397 		 * from the ELF OSABI-note if necessary.
2398 		 */
2399 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2400 		    checknote->trans_osrel != NULL) {
2401 			res = checknote->trans_osrel(note, osrel);
2402 			goto ret;
2403 		}
2404 		res = TRUE;
2405 		goto ret;
2406 nextnote:
2407 		note = (const Elf_Note *)((const char *)(note + 1) +
2408 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2409 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2410 	}
2411 	res = FALSE;
2412 ret:
2413 	free(buf, M_TEMP);
2414 	return (res);
2415 }
2416 
2417 /*
2418  * Try to find the appropriate ABI-note section for checknote,
2419  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2420  * first page of the image is searched, the same as for headers.
2421  */
2422 static boolean_t
2423 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2424     int32_t *osrel)
2425 {
2426 	const Elf_Phdr *phdr;
2427 	const Elf_Ehdr *hdr;
2428 	int i;
2429 
2430 	hdr = (const Elf_Ehdr *)imgp->image_header;
2431 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2432 
2433 	for (i = 0; i < hdr->e_phnum; i++) {
2434 		if (phdr[i].p_type == PT_NOTE &&
2435 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2436 			return (TRUE);
2437 	}
2438 	return (FALSE);
2439 
2440 }
2441 
2442 /*
2443  * Tell kern_execve.c about it, with a little help from the linker.
2444  */
2445 static struct execsw __elfN(execsw) = {
2446 	.ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2447 	.ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2448 };
2449 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2450 
2451 static vm_prot_t
2452 __elfN(trans_prot)(Elf_Word flags)
2453 {
2454 	vm_prot_t prot;
2455 
2456 	prot = 0;
2457 	if (flags & PF_X)
2458 		prot |= VM_PROT_EXECUTE;
2459 	if (flags & PF_W)
2460 		prot |= VM_PROT_WRITE;
2461 	if (flags & PF_R)
2462 		prot |= VM_PROT_READ;
2463 #if __ELF_WORD_SIZE == 32
2464 #if defined(__amd64__)
2465 	if (i386_read_exec && (flags & PF_R))
2466 		prot |= VM_PROT_EXECUTE;
2467 #endif
2468 #endif
2469 	return (prot);
2470 }
2471 
2472 static Elf_Word
2473 __elfN(untrans_prot)(vm_prot_t prot)
2474 {
2475 	Elf_Word flags;
2476 
2477 	flags = 0;
2478 	if (prot & VM_PROT_EXECUTE)
2479 		flags |= PF_X;
2480 	if (prot & VM_PROT_READ)
2481 		flags |= PF_R;
2482 	if (prot & VM_PROT_WRITE)
2483 		flags |= PF_W;
2484 	return (flags);
2485 }
2486