xref: /freebsd/sys/kern/imgact_elf.c (revision d06955f9bdb1416d9196043ed781f9b36dae9adc)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2017 Dell EMC
5  * Copyright (c) 2000 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 #include "opt_compat.h"
39 #include "opt_gzio.h"
40 
41 #include <sys/param.h>
42 #include <sys/capsicum.h>
43 #include <sys/exec.h>
44 #include <sys/fcntl.h>
45 #include <sys/gzio.h>
46 #include <sys/imgact.h>
47 #include <sys/imgact_elf.h>
48 #include <sys/jail.h>
49 #include <sys/kernel.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mman.h>
54 #include <sys/namei.h>
55 #include <sys/pioctl.h>
56 #include <sys/proc.h>
57 #include <sys/procfs.h>
58 #include <sys/ptrace.h>
59 #include <sys/racct.h>
60 #include <sys/resourcevar.h>
61 #include <sys/rwlock.h>
62 #include <sys/sbuf.h>
63 #include <sys/sf_buf.h>
64 #include <sys/smp.h>
65 #include <sys/systm.h>
66 #include <sys/signalvar.h>
67 #include <sys/stat.h>
68 #include <sys/sx.h>
69 #include <sys/syscall.h>
70 #include <sys/sysctl.h>
71 #include <sys/sysent.h>
72 #include <sys/vnode.h>
73 #include <sys/syslog.h>
74 #include <sys/eventhandler.h>
75 #include <sys/user.h>
76 
77 #include <vm/vm.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_param.h>
80 #include <vm/pmap.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_extern.h>
84 
85 #include <machine/elf.h>
86 #include <machine/md_var.h>
87 
88 #define ELF_NOTE_ROUNDSIZE	4
89 #define OLD_EI_BRAND	8
90 
91 static int __elfN(check_header)(const Elf_Ehdr *hdr);
92 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
93     const char *interp, int interp_name_len, int32_t *osrel);
94 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
95     u_long *entry, size_t pagesize);
96 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
97     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
98     size_t pagesize);
99 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
100 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
101     int32_t *osrel);
102 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
103 static boolean_t __elfN(check_note)(struct image_params *imgp,
104     Elf_Brandnote *checknote, int32_t *osrel);
105 static vm_prot_t __elfN(trans_prot)(Elf_Word);
106 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
107 
108 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
109     "");
110 
111 #define	CORE_BUF_SIZE	(16 * 1024)
112 
113 int __elfN(fallback_brand) = -1;
114 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
115     fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
116     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
117 
118 static int elf_legacy_coredump = 0;
119 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
120     &elf_legacy_coredump, 0,
121     "include all and only RW pages in core dumps");
122 
123 int __elfN(nxstack) =
124 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
125     (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
126 	1;
127 #else
128 	0;
129 #endif
130 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
131     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
132     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
133 
134 #if __ELF_WORD_SIZE == 32
135 #if defined(__amd64__)
136 int i386_read_exec = 0;
137 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
138     "enable execution from readable segments");
139 #endif
140 #endif
141 
142 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
143 
144 #define	trunc_page_ps(va, ps)	rounddown2(va, ps)
145 #define	round_page_ps(va, ps)	roundup2(va, ps)
146 #define	aligned(a, t)	(trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
147 
148 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
149 
150 Elf_Brandnote __elfN(freebsd_brandnote) = {
151 	.hdr.n_namesz	= sizeof(FREEBSD_ABI_VENDOR),
152 	.hdr.n_descsz	= sizeof(int32_t),
153 	.hdr.n_type	= NT_FREEBSD_ABI_TAG,
154 	.vendor		= FREEBSD_ABI_VENDOR,
155 	.flags		= BN_TRANSLATE_OSREL,
156 	.trans_osrel	= __elfN(freebsd_trans_osrel)
157 };
158 
159 static boolean_t
160 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
161 {
162 	uintptr_t p;
163 
164 	p = (uintptr_t)(note + 1);
165 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
166 	*osrel = *(const int32_t *)(p);
167 
168 	return (TRUE);
169 }
170 
171 static const char GNU_ABI_VENDOR[] = "GNU";
172 static int GNU_KFREEBSD_ABI_DESC = 3;
173 
174 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
175 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
176 	.hdr.n_descsz	= 16,	/* XXX at least 16 */
177 	.hdr.n_type	= 1,
178 	.vendor		= GNU_ABI_VENDOR,
179 	.flags		= BN_TRANSLATE_OSREL,
180 	.trans_osrel	= kfreebsd_trans_osrel
181 };
182 
183 static boolean_t
184 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
185 {
186 	const Elf32_Word *desc;
187 	uintptr_t p;
188 
189 	p = (uintptr_t)(note + 1);
190 	p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
191 
192 	desc = (const Elf32_Word *)p;
193 	if (desc[0] != GNU_KFREEBSD_ABI_DESC)
194 		return (FALSE);
195 
196 	/*
197 	 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
198 	 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
199 	 */
200 	*osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
201 
202 	return (TRUE);
203 }
204 
205 int
206 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
207 {
208 	int i;
209 
210 	for (i = 0; i < MAX_BRANDS; i++) {
211 		if (elf_brand_list[i] == NULL) {
212 			elf_brand_list[i] = entry;
213 			break;
214 		}
215 	}
216 	if (i == MAX_BRANDS) {
217 		printf("WARNING: %s: could not insert brandinfo entry: %p\n",
218 			__func__, entry);
219 		return (-1);
220 	}
221 	return (0);
222 }
223 
224 int
225 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
226 {
227 	int i;
228 
229 	for (i = 0; i < MAX_BRANDS; i++) {
230 		if (elf_brand_list[i] == entry) {
231 			elf_brand_list[i] = NULL;
232 			break;
233 		}
234 	}
235 	if (i == MAX_BRANDS)
236 		return (-1);
237 	return (0);
238 }
239 
240 int
241 __elfN(brand_inuse)(Elf_Brandinfo *entry)
242 {
243 	struct proc *p;
244 	int rval = FALSE;
245 
246 	sx_slock(&allproc_lock);
247 	FOREACH_PROC_IN_SYSTEM(p) {
248 		if (p->p_sysent == entry->sysvec) {
249 			rval = TRUE;
250 			break;
251 		}
252 	}
253 	sx_sunlock(&allproc_lock);
254 
255 	return (rval);
256 }
257 
258 static Elf_Brandinfo *
259 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
260     int interp_name_len, int32_t *osrel)
261 {
262 	const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
263 	Elf_Brandinfo *bi, *bi_m;
264 	boolean_t ret;
265 	int i;
266 
267 	/*
268 	 * We support four types of branding -- (1) the ELF EI_OSABI field
269 	 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
270 	 * branding w/in the ELF header, (3) path of the `interp_path'
271 	 * field, and (4) the ".note.ABI-tag" ELF section.
272 	 */
273 
274 	/* Look for an ".note.ABI-tag" ELF section */
275 	bi_m = NULL;
276 	for (i = 0; i < MAX_BRANDS; i++) {
277 		bi = elf_brand_list[i];
278 		if (bi == NULL)
279 			continue;
280 		if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
281 			continue;
282 		if (hdr->e_machine == bi->machine && (bi->flags &
283 		    (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
284 			ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
285 			/* Give brand a chance to veto check_note's guess */
286 			if (ret && bi->header_supported)
287 				ret = bi->header_supported(imgp);
288 			/*
289 			 * If note checker claimed the binary, but the
290 			 * interpreter path in the image does not
291 			 * match default one for the brand, try to
292 			 * search for other brands with the same
293 			 * interpreter.  Either there is better brand
294 			 * with the right interpreter, or, failing
295 			 * this, we return first brand which accepted
296 			 * our note and, optionally, header.
297 			 */
298 			if (ret && bi_m == NULL && interp != NULL &&
299 			    (bi->interp_path == NULL ||
300 			    (strlen(bi->interp_path) + 1 != interp_name_len ||
301 			    strncmp(interp, bi->interp_path, interp_name_len)
302 			    != 0))) {
303 				bi_m = bi;
304 				ret = 0;
305 			}
306 			if (ret)
307 				return (bi);
308 		}
309 	}
310 	if (bi_m != NULL)
311 		return (bi_m);
312 
313 	/* If the executable has a brand, search for it in the brand list. */
314 	for (i = 0; i < MAX_BRANDS; i++) {
315 		bi = elf_brand_list[i];
316 		if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
317 		    (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
318 			continue;
319 		if (hdr->e_machine == bi->machine &&
320 		    (hdr->e_ident[EI_OSABI] == bi->brand ||
321 		    (bi->compat_3_brand != NULL &&
322 		    strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
323 		    bi->compat_3_brand) == 0))) {
324 			/* Looks good, but give brand a chance to veto */
325 			if (!bi->header_supported ||
326 			    bi->header_supported(imgp)) {
327 				/*
328 				 * Again, prefer strictly matching
329 				 * interpreter path.
330 				 */
331 				if (interp_name_len == 0 &&
332 				    bi->interp_path == NULL)
333 					return (bi);
334 				if (bi->interp_path != NULL &&
335 				    strlen(bi->interp_path) + 1 ==
336 				    interp_name_len && strncmp(interp,
337 				    bi->interp_path, interp_name_len) == 0)
338 					return (bi);
339 				if (bi_m == NULL)
340 					bi_m = bi;
341 			}
342 		}
343 	}
344 	if (bi_m != NULL)
345 		return (bi_m);
346 
347 	/* No known brand, see if the header is recognized by any brand */
348 	for (i = 0; i < MAX_BRANDS; i++) {
349 		bi = elf_brand_list[i];
350 		if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
351 		    bi->header_supported == NULL)
352 			continue;
353 		if (hdr->e_machine == bi->machine) {
354 			ret = bi->header_supported(imgp);
355 			if (ret)
356 				return (bi);
357 		}
358 	}
359 
360 	/* Lacking a known brand, search for a recognized interpreter. */
361 	if (interp != NULL) {
362 		for (i = 0; i < MAX_BRANDS; i++) {
363 			bi = elf_brand_list[i];
364 			if (bi == NULL || (bi->flags &
365 			    (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
366 			    != 0)
367 				continue;
368 			if (hdr->e_machine == bi->machine &&
369 			    bi->interp_path != NULL &&
370 			    /* ELF image p_filesz includes terminating zero */
371 			    strlen(bi->interp_path) + 1 == interp_name_len &&
372 			    strncmp(interp, bi->interp_path, interp_name_len)
373 			    == 0)
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 			return (bi);
387 	}
388 	return (NULL);
389 }
390 
391 static int
392 __elfN(check_header)(const Elf_Ehdr *hdr)
393 {
394 	Elf_Brandinfo *bi;
395 	int i;
396 
397 	if (!IS_ELF(*hdr) ||
398 	    hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
399 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
400 	    hdr->e_ident[EI_VERSION] != EV_CURRENT ||
401 	    hdr->e_phentsize != sizeof(Elf_Phdr) ||
402 	    hdr->e_version != ELF_TARG_VER)
403 		return (ENOEXEC);
404 
405 	/*
406 	 * Make sure we have at least one brand for this machine.
407 	 */
408 
409 	for (i = 0; i < MAX_BRANDS; i++) {
410 		bi = elf_brand_list[i];
411 		if (bi != NULL && bi->machine == hdr->e_machine)
412 			break;
413 	}
414 	if (i == MAX_BRANDS)
415 		return (ENOEXEC);
416 
417 	return (0);
418 }
419 
420 static int
421 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
422     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
423 {
424 	struct sf_buf *sf;
425 	int error;
426 	vm_offset_t off;
427 
428 	/*
429 	 * Create the page if it doesn't exist yet. Ignore errors.
430 	 */
431 	vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
432 	    trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
433 
434 	/*
435 	 * Find the page from the underlying object.
436 	 */
437 	if (object != NULL) {
438 		sf = vm_imgact_map_page(object, offset);
439 		if (sf == NULL)
440 			return (KERN_FAILURE);
441 		off = offset - trunc_page(offset);
442 		error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
443 		    end - start);
444 		vm_imgact_unmap_page(sf);
445 		if (error != 0)
446 			return (KERN_FAILURE);
447 	}
448 
449 	return (KERN_SUCCESS);
450 }
451 
452 static int
453 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
454     vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
455     int cow)
456 {
457 	struct sf_buf *sf;
458 	vm_offset_t off;
459 	vm_size_t sz;
460 	int error, locked, rv;
461 
462 	if (start != trunc_page(start)) {
463 		rv = __elfN(map_partial)(map, object, offset, start,
464 		    round_page(start), prot);
465 		if (rv != KERN_SUCCESS)
466 			return (rv);
467 		offset += round_page(start) - start;
468 		start = round_page(start);
469 	}
470 	if (end != round_page(end)) {
471 		rv = __elfN(map_partial)(map, object, offset +
472 		    trunc_page(end) - start, trunc_page(end), end, prot);
473 		if (rv != KERN_SUCCESS)
474 			return (rv);
475 		end = trunc_page(end);
476 	}
477 	if (start >= end)
478 		return (KERN_SUCCESS);
479 	if ((offset & PAGE_MASK) != 0) {
480 		/*
481 		 * The mapping is not page aligned.  This means that we have
482 		 * to copy the data.
483 		 */
484 		rv = vm_map_fixed(map, NULL, 0, start, end - start,
485 		    prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
486 		if (rv != KERN_SUCCESS)
487 			return (rv);
488 		if (object == NULL)
489 			return (KERN_SUCCESS);
490 		for (; start < end; start += sz) {
491 			sf = vm_imgact_map_page(object, offset);
492 			if (sf == NULL)
493 				return (KERN_FAILURE);
494 			off = offset - trunc_page(offset);
495 			sz = end - start;
496 			if (sz > PAGE_SIZE - off)
497 				sz = PAGE_SIZE - off;
498 			error = copyout((caddr_t)sf_buf_kva(sf) + off,
499 			    (caddr_t)start, sz);
500 			vm_imgact_unmap_page(sf);
501 			if (error != 0)
502 				return (KERN_FAILURE);
503 			offset += sz;
504 		}
505 	} else {
506 		vm_object_reference(object);
507 		rv = vm_map_fixed(map, object, offset, start, end - start,
508 		    prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
509 		if (rv != KERN_SUCCESS) {
510 			locked = VOP_ISLOCKED(imgp->vp);
511 			VOP_UNLOCK(imgp->vp, 0);
512 			vm_object_deallocate(object);
513 			vn_lock(imgp->vp, locked | LK_RETRY);
514 			return (rv);
515 		}
516 	}
517 	return (KERN_SUCCESS);
518 }
519 
520 static int
521 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
522     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
523     size_t pagesize)
524 {
525 	struct sf_buf *sf;
526 	size_t map_len;
527 	vm_map_t map;
528 	vm_object_t object;
529 	vm_offset_t off, map_addr;
530 	int error, rv, cow;
531 	size_t copy_len;
532 	vm_ooffset_t file_addr;
533 
534 	/*
535 	 * It's necessary to fail if the filsz + offset taken from the
536 	 * header is greater than the actual file pager object's size.
537 	 * If we were to allow this, then the vm_map_find() below would
538 	 * walk right off the end of the file object and into the ether.
539 	 *
540 	 * While I'm here, might as well check for something else that
541 	 * is invalid: filsz cannot be greater than memsz.
542 	 */
543 	if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
544 	    filsz > memsz) {
545 		uprintf("elf_load_section: truncated ELF file\n");
546 		return (ENOEXEC);
547 	}
548 
549 	object = imgp->object;
550 	map = &imgp->proc->p_vmspace->vm_map;
551 	map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
552 	file_addr = trunc_page_ps(offset, pagesize);
553 
554 	/*
555 	 * We have two choices.  We can either clear the data in the last page
556 	 * of an oversized mapping, or we can start the anon mapping a page
557 	 * early and copy the initialized data into that first page.  We
558 	 * choose the second.
559 	 */
560 	if (filsz == 0)
561 		map_len = 0;
562 	else if (memsz > filsz)
563 		map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
564 	else
565 		map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
566 
567 	if (map_len != 0) {
568 		/* cow flags: don't dump readonly sections in core */
569 		cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
570 		    (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
571 
572 		rv = __elfN(map_insert)(imgp, map,
573 				      object,
574 				      file_addr,	/* file offset */
575 				      map_addr,		/* virtual start */
576 				      map_addr + map_len,/* virtual end */
577 				      prot,
578 				      cow);
579 		if (rv != KERN_SUCCESS)
580 			return (EINVAL);
581 
582 		/* we can stop now if we've covered it all */
583 		if (memsz == filsz)
584 			return (0);
585 	}
586 
587 
588 	/*
589 	 * We have to get the remaining bit of the file into the first part
590 	 * of the oversized map segment.  This is normally because the .data
591 	 * segment in the file is extended to provide bss.  It's a neat idea
592 	 * to try and save a page, but it's a pain in the behind to implement.
593 	 */
594 	copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
595 	    filsz, pagesize);
596 	map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
597 	map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
598 	    map_addr;
599 
600 	/* This had damn well better be true! */
601 	if (map_len != 0) {
602 		rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
603 		    map_addr + map_len, prot, 0);
604 		if (rv != KERN_SUCCESS)
605 			return (EINVAL);
606 	}
607 
608 	if (copy_len != 0) {
609 		sf = vm_imgact_map_page(object, offset + filsz);
610 		if (sf == NULL)
611 			return (EIO);
612 
613 		/* send the page fragment to user space */
614 		off = trunc_page_ps(offset + filsz, pagesize) -
615 		    trunc_page(offset + filsz);
616 		error = copyout((caddr_t)sf_buf_kva(sf) + off,
617 		    (caddr_t)map_addr, copy_len);
618 		vm_imgact_unmap_page(sf);
619 		if (error != 0)
620 			return (error);
621 	}
622 
623 	/*
624 	 * Remove write access to the page if it was only granted by map_insert
625 	 * to allow copyout.
626 	 */
627 	if ((prot & VM_PROT_WRITE) == 0)
628 		vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
629 		    map_len), prot, FALSE);
630 
631 	return (0);
632 }
633 
634 /*
635  * Load the file "file" into memory.  It may be either a shared object
636  * or an executable.
637  *
638  * The "addr" reference parameter is in/out.  On entry, it specifies
639  * the address where a shared object should be loaded.  If the file is
640  * an executable, this value is ignored.  On exit, "addr" specifies
641  * where the file was actually loaded.
642  *
643  * The "entry" reference parameter is out only.  On exit, it specifies
644  * the entry point for the loaded file.
645  */
646 static int
647 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
648 	u_long *entry, size_t pagesize)
649 {
650 	struct {
651 		struct nameidata nd;
652 		struct vattr attr;
653 		struct image_params image_params;
654 	} *tempdata;
655 	const Elf_Ehdr *hdr = NULL;
656 	const Elf_Phdr *phdr = NULL;
657 	struct nameidata *nd;
658 	struct vattr *attr;
659 	struct image_params *imgp;
660 	vm_prot_t prot;
661 	u_long rbase;
662 	u_long base_addr = 0;
663 	int error, i, numsegs;
664 
665 #ifdef CAPABILITY_MODE
666 	/*
667 	 * XXXJA: This check can go away once we are sufficiently confident
668 	 * that the checks in namei() are correct.
669 	 */
670 	if (IN_CAPABILITY_MODE(curthread))
671 		return (ECAPMODE);
672 #endif
673 
674 	tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
675 	nd = &tempdata->nd;
676 	attr = &tempdata->attr;
677 	imgp = &tempdata->image_params;
678 
679 	/*
680 	 * Initialize part of the common data
681 	 */
682 	imgp->proc = p;
683 	imgp->attr = attr;
684 	imgp->firstpage = NULL;
685 	imgp->image_header = NULL;
686 	imgp->object = NULL;
687 	imgp->execlabel = NULL;
688 
689 	NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
690 	if ((error = namei(nd)) != 0) {
691 		nd->ni_vp = NULL;
692 		goto fail;
693 	}
694 	NDFREE(nd, NDF_ONLY_PNBUF);
695 	imgp->vp = nd->ni_vp;
696 
697 	/*
698 	 * Check permissions, modes, uid, etc on the file, and "open" it.
699 	 */
700 	error = exec_check_permissions(imgp);
701 	if (error)
702 		goto fail;
703 
704 	error = exec_map_first_page(imgp);
705 	if (error)
706 		goto fail;
707 
708 	/*
709 	 * Also make certain that the interpreter stays the same, so set
710 	 * its VV_TEXT flag, too.
711 	 */
712 	VOP_SET_TEXT(nd->ni_vp);
713 
714 	imgp->object = nd->ni_vp->v_object;
715 
716 	hdr = (const Elf_Ehdr *)imgp->image_header;
717 	if ((error = __elfN(check_header)(hdr)) != 0)
718 		goto fail;
719 	if (hdr->e_type == ET_DYN)
720 		rbase = *addr;
721 	else if (hdr->e_type == ET_EXEC)
722 		rbase = 0;
723 	else {
724 		error = ENOEXEC;
725 		goto fail;
726 	}
727 
728 	/* Only support headers that fit within first page for now      */
729 	if ((hdr->e_phoff > PAGE_SIZE) ||
730 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
731 		error = ENOEXEC;
732 		goto fail;
733 	}
734 
735 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
736 	if (!aligned(phdr, Elf_Addr)) {
737 		error = ENOEXEC;
738 		goto fail;
739 	}
740 
741 	for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
742 		if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
743 			/* Loadable segment */
744 			prot = __elfN(trans_prot)(phdr[i].p_flags);
745 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
746 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
747 			    phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
748 			if (error != 0)
749 				goto fail;
750 			/*
751 			 * Establish the base address if this is the
752 			 * first segment.
753 			 */
754 			if (numsegs == 0)
755   				base_addr = trunc_page(phdr[i].p_vaddr +
756 				    rbase);
757 			numsegs++;
758 		}
759 	}
760 	*addr = base_addr;
761 	*entry = (unsigned long)hdr->e_entry + rbase;
762 
763 fail:
764 	if (imgp->firstpage)
765 		exec_unmap_first_page(imgp);
766 
767 	if (nd->ni_vp)
768 		vput(nd->ni_vp);
769 
770 	free(tempdata, M_TEMP);
771 
772 	return (error);
773 }
774 
775 static int
776 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
777 {
778 	struct thread *td;
779 	const Elf_Ehdr *hdr;
780 	const Elf_Phdr *phdr;
781 	Elf_Auxargs *elf_auxargs;
782 	struct vmspace *vmspace;
783 	const char *err_str, *newinterp;
784 	char *interp, *interp_buf, *path;
785 	Elf_Brandinfo *brand_info;
786 	struct sysentvec *sv;
787 	vm_prot_t prot;
788 	u_long text_size, data_size, total_size, text_addr, data_addr;
789 	u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
790 	int32_t osrel;
791 	int error, i, n, interp_name_len, have_interp;
792 
793 	hdr = (const Elf_Ehdr *)imgp->image_header;
794 
795 	/*
796 	 * Do we have a valid ELF header ?
797 	 *
798 	 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
799 	 * if particular brand doesn't support it.
800 	 */
801 	if (__elfN(check_header)(hdr) != 0 ||
802 	    (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
803 		return (-1);
804 
805 	/*
806 	 * From here on down, we return an errno, not -1, as we've
807 	 * detected an ELF file.
808 	 */
809 
810 	if ((hdr->e_phoff > PAGE_SIZE) ||
811 	    (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
812 		/* Only support headers in first page for now */
813 		uprintf("Program headers not in the first page\n");
814 		return (ENOEXEC);
815 	}
816 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
817 	if (!aligned(phdr, Elf_Addr)) {
818 		uprintf("Unaligned program headers\n");
819 		return (ENOEXEC);
820 	}
821 
822 	n = error = 0;
823 	baddr = 0;
824 	osrel = 0;
825 	text_size = data_size = total_size = text_addr = data_addr = 0;
826 	entry = proghdr = 0;
827 	interp_name_len = 0;
828 	err_str = newinterp = NULL;
829 	interp = interp_buf = NULL;
830 	td = curthread;
831 
832 	for (i = 0; i < hdr->e_phnum; i++) {
833 		switch (phdr[i].p_type) {
834 		case PT_LOAD:
835 			if (n == 0)
836 				baddr = phdr[i].p_vaddr;
837 			n++;
838 			break;
839 		case PT_INTERP:
840 			/* Path to interpreter */
841 			if (phdr[i].p_filesz > MAXPATHLEN) {
842 				uprintf("Invalid PT_INTERP\n");
843 				error = ENOEXEC;
844 				goto ret;
845 			}
846 			if (interp != NULL) {
847 				uprintf("Multiple PT_INTERP headers\n");
848 				error = ENOEXEC;
849 				goto ret;
850 			}
851 			interp_name_len = phdr[i].p_filesz;
852 			if (phdr[i].p_offset > PAGE_SIZE ||
853 			    interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
854 				VOP_UNLOCK(imgp->vp, 0);
855 				interp_buf = malloc(interp_name_len + 1, M_TEMP,
856 				    M_WAITOK);
857 				vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
858 				error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
859 				    interp_name_len, phdr[i].p_offset,
860 				    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
861 				    NOCRED, NULL, td);
862 				if (error != 0) {
863 					uprintf("i/o error PT_INTERP\n");
864 					goto ret;
865 				}
866 				interp_buf[interp_name_len] = '\0';
867 				interp = interp_buf;
868 			} else {
869 				interp = __DECONST(char *, imgp->image_header) +
870 				    phdr[i].p_offset;
871 			}
872 			break;
873 		case PT_GNU_STACK:
874 			if (__elfN(nxstack))
875 				imgp->stack_prot =
876 				    __elfN(trans_prot)(phdr[i].p_flags);
877 			imgp->stack_sz = phdr[i].p_memsz;
878 			break;
879 		}
880 	}
881 
882 	brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
883 	    &osrel);
884 	if (brand_info == NULL) {
885 		uprintf("ELF binary type \"%u\" not known.\n",
886 		    hdr->e_ident[EI_OSABI]);
887 		error = ENOEXEC;
888 		goto ret;
889 	}
890 	et_dyn_addr = 0;
891 	if (hdr->e_type == ET_DYN) {
892 		if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
893 			uprintf("Cannot execute shared object\n");
894 			error = ENOEXEC;
895 			goto ret;
896 		}
897 		/*
898 		 * Honour the base load address from the dso if it is
899 		 * non-zero for some reason.
900 		 */
901 		if (baddr == 0)
902 			et_dyn_addr = ET_DYN_LOAD_ADDR;
903 	}
904 	sv = brand_info->sysvec;
905 	if (interp != NULL && brand_info->interp_newpath != NULL)
906 		newinterp = brand_info->interp_newpath;
907 
908 	/*
909 	 * Avoid a possible deadlock if the current address space is destroyed
910 	 * and that address space maps the locked vnode.  In the common case,
911 	 * the locked vnode's v_usecount is decremented but remains greater
912 	 * than zero.  Consequently, the vnode lock is not needed by vrele().
913 	 * However, in cases where the vnode lock is external, such as nullfs,
914 	 * v_usecount may become zero.
915 	 *
916 	 * The VV_TEXT flag prevents modifications to the executable while
917 	 * the vnode is unlocked.
918 	 */
919 	VOP_UNLOCK(imgp->vp, 0);
920 
921 	error = exec_new_vmspace(imgp, sv);
922 	imgp->proc->p_sysent = sv;
923 
924 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
925 	if (error != 0)
926 		goto ret;
927 
928 	for (i = 0; i < hdr->e_phnum; i++) {
929 		switch (phdr[i].p_type) {
930 		case PT_LOAD:	/* Loadable segment */
931 			if (phdr[i].p_memsz == 0)
932 				break;
933 			prot = __elfN(trans_prot)(phdr[i].p_flags);
934 			error = __elfN(load_section)(imgp, phdr[i].p_offset,
935 			    (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
936 			    phdr[i].p_memsz, phdr[i].p_filesz, prot,
937 			    sv->sv_pagesize);
938 			if (error != 0)
939 				goto ret;
940 
941 			/*
942 			 * If this segment contains the program headers,
943 			 * remember their virtual address for the AT_PHDR
944 			 * aux entry. Static binaries don't usually include
945 			 * a PT_PHDR entry.
946 			 */
947 			if (phdr[i].p_offset == 0 &&
948 			    hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
949 				<= phdr[i].p_filesz)
950 				proghdr = phdr[i].p_vaddr + hdr->e_phoff +
951 				    et_dyn_addr;
952 
953 			seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
954 			seg_size = round_page(phdr[i].p_memsz +
955 			    phdr[i].p_vaddr + et_dyn_addr - seg_addr);
956 
957 			/*
958 			 * Make the largest executable segment the official
959 			 * text segment and all others data.
960 			 *
961 			 * Note that obreak() assumes that data_addr +
962 			 * data_size == end of data load area, and the ELF
963 			 * file format expects segments to be sorted by
964 			 * address.  If multiple data segments exist, the
965 			 * last one will be used.
966 			 */
967 
968 			if (phdr[i].p_flags & PF_X && text_size < seg_size) {
969 				text_size = seg_size;
970 				text_addr = seg_addr;
971 			} else {
972 				data_size = seg_size;
973 				data_addr = seg_addr;
974 			}
975 			total_size += seg_size;
976 			break;
977 		case PT_PHDR: 	/* Program header table info */
978 			proghdr = phdr[i].p_vaddr + et_dyn_addr;
979 			break;
980 		default:
981 			break;
982 		}
983 	}
984 
985 	if (data_addr == 0 && data_size == 0) {
986 		data_addr = text_addr;
987 		data_size = text_size;
988 	}
989 
990 	entry = (u_long)hdr->e_entry + et_dyn_addr;
991 
992 	/*
993 	 * Check limits.  It should be safe to check the
994 	 * limits after loading the segments since we do
995 	 * not actually fault in all the segments pages.
996 	 */
997 	PROC_LOCK(imgp->proc);
998 	if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
999 		err_str = "Data segment size exceeds process limit";
1000 	else if (text_size > maxtsiz)
1001 		err_str = "Text segment size exceeds system limit";
1002 	else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1003 		err_str = "Total segment size exceeds process limit";
1004 	else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1005 		err_str = "Data segment size exceeds resource limit";
1006 	else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1007 		err_str = "Total segment size exceeds resource limit";
1008 	if (err_str != NULL) {
1009 		PROC_UNLOCK(imgp->proc);
1010 		uprintf("%s\n", err_str);
1011 		error = ENOMEM;
1012 		goto ret;
1013 	}
1014 
1015 	vmspace = imgp->proc->p_vmspace;
1016 	vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1017 	vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1018 	vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1019 	vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1020 
1021 	/*
1022 	 * We load the dynamic linker where a userland call
1023 	 * to mmap(0, ...) would put it.  The rationale behind this
1024 	 * calculation is that it leaves room for the heap to grow to
1025 	 * its maximum allowed size.
1026 	 */
1027 	addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1028 	    RLIMIT_DATA));
1029 	PROC_UNLOCK(imgp->proc);
1030 
1031 	imgp->entry_addr = entry;
1032 
1033 	if (interp != NULL) {
1034 		have_interp = FALSE;
1035 		VOP_UNLOCK(imgp->vp, 0);
1036 		if (brand_info->emul_path != NULL &&
1037 		    brand_info->emul_path[0] != '\0') {
1038 			path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1039 			snprintf(path, MAXPATHLEN, "%s%s",
1040 			    brand_info->emul_path, interp);
1041 			error = __elfN(load_file)(imgp->proc, path, &addr,
1042 			    &imgp->entry_addr, sv->sv_pagesize);
1043 			free(path, M_TEMP);
1044 			if (error == 0)
1045 				have_interp = TRUE;
1046 		}
1047 		if (!have_interp && newinterp != NULL &&
1048 		    (brand_info->interp_path == NULL ||
1049 		    strcmp(interp, brand_info->interp_path) == 0)) {
1050 			error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1051 			    &imgp->entry_addr, sv->sv_pagesize);
1052 			if (error == 0)
1053 				have_interp = TRUE;
1054 		}
1055 		if (!have_interp) {
1056 			error = __elfN(load_file)(imgp->proc, interp, &addr,
1057 			    &imgp->entry_addr, sv->sv_pagesize);
1058 		}
1059 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1060 		if (error != 0) {
1061 			uprintf("ELF interpreter %s not found, error %d\n",
1062 			    interp, error);
1063 			goto ret;
1064 		}
1065 	} else
1066 		addr = et_dyn_addr;
1067 
1068 	/*
1069 	 * Construct auxargs table (used by the fixup routine)
1070 	 */
1071 	elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1072 	elf_auxargs->execfd = -1;
1073 	elf_auxargs->phdr = proghdr;
1074 	elf_auxargs->phent = hdr->e_phentsize;
1075 	elf_auxargs->phnum = hdr->e_phnum;
1076 	elf_auxargs->pagesz = PAGE_SIZE;
1077 	elf_auxargs->base = addr;
1078 	elf_auxargs->flags = 0;
1079 	elf_auxargs->entry = entry;
1080 	elf_auxargs->hdr_eflags = hdr->e_flags;
1081 
1082 	imgp->auxargs = elf_auxargs;
1083 	imgp->interpreted = 0;
1084 	imgp->reloc_base = addr;
1085 	imgp->proc->p_osrel = osrel;
1086 	imgp->proc->p_elf_machine = hdr->e_machine;
1087 	imgp->proc->p_elf_flags = hdr->e_flags;
1088 
1089 ret:
1090 	free(interp_buf, M_TEMP);
1091 	return (error);
1092 }
1093 
1094 #define	suword __CONCAT(suword, __ELF_WORD_SIZE)
1095 
1096 int
1097 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1098 {
1099 	Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1100 	Elf_Addr *base;
1101 	Elf_Addr *pos;
1102 
1103 	base = (Elf_Addr *)*stack_base;
1104 	pos = base + (imgp->args->argc + imgp->args->envc + 2);
1105 
1106 	if (args->execfd != -1)
1107 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1108 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1109 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1110 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1111 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1112 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1113 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1114 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
1115 	AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1116 	if (imgp->execpathp != 0)
1117 		AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1118 	AUXARGS_ENTRY(pos, AT_OSRELDATE,
1119 	    imgp->proc->p_ucred->cr_prison->pr_osreldate);
1120 	if (imgp->canary != 0) {
1121 		AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1122 		AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1123 	}
1124 	AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1125 	if (imgp->pagesizes != 0) {
1126 		AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1127 		AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1128 	}
1129 	if (imgp->sysent->sv_timekeep_base != 0) {
1130 		AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1131 		    imgp->sysent->sv_timekeep_base);
1132 	}
1133 	AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1134 	    != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1135 	    imgp->sysent->sv_stackprot);
1136 	if (imgp->sysent->sv_hwcap != NULL)
1137 		AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1138 	if (imgp->sysent->sv_hwcap2 != NULL)
1139 		AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1140 	AUXARGS_ENTRY(pos, AT_NULL, 0);
1141 
1142 	free(imgp->auxargs, M_TEMP);
1143 	imgp->auxargs = NULL;
1144 
1145 	base--;
1146 	suword(base, (long)imgp->args->argc);
1147 	*stack_base = (register_t *)base;
1148 	return (0);
1149 }
1150 
1151 /*
1152  * Code for generating ELF core dumps.
1153  */
1154 
1155 typedef void (*segment_callback)(vm_map_entry_t, void *);
1156 
1157 /* Closure for cb_put_phdr(). */
1158 struct phdr_closure {
1159 	Elf_Phdr *phdr;		/* Program header to fill in */
1160 	Elf_Off offset;		/* Offset of segment in core file */
1161 };
1162 
1163 /* Closure for cb_size_segment(). */
1164 struct sseg_closure {
1165 	int count;		/* Count of writable segments. */
1166 	size_t size;		/* Total size of all writable segments. */
1167 };
1168 
1169 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1170 
1171 struct note_info {
1172 	int		type;		/* Note type. */
1173 	outfunc_t 	outfunc; 	/* Output function. */
1174 	void		*outarg;	/* Argument for the output function. */
1175 	size_t		outsize;	/* Output size. */
1176 	TAILQ_ENTRY(note_info) link;	/* Link to the next note info. */
1177 };
1178 
1179 TAILQ_HEAD(note_info_list, note_info);
1180 
1181 /* Coredump output parameters. */
1182 struct coredump_params {
1183 	off_t		offset;
1184 	struct ucred	*active_cred;
1185 	struct ucred	*file_cred;
1186 	struct thread	*td;
1187 	struct vnode	*vp;
1188 	struct gzio_stream *gzs;
1189 };
1190 
1191 static void cb_put_phdr(vm_map_entry_t, void *);
1192 static void cb_size_segment(vm_map_entry_t, void *);
1193 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1194     enum uio_seg);
1195 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1196 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1197     struct note_info_list *, size_t);
1198 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1199     size_t *);
1200 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1201 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1202 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1203 static int sbuf_drain_core_output(void *, const char *, int);
1204 static int sbuf_drain_count(void *arg, const char *data, int len);
1205 
1206 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1207 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1208 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1209 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1210 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1211 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1212 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1213 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1214 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1215 static void note_procstat_files(void *, struct sbuf *, size_t *);
1216 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1217 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1218 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1219 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1220 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1221 
1222 #ifdef GZIO
1223 extern int compress_user_cores_gzlevel;
1224 
1225 /*
1226  * Write out a core segment to the compression stream.
1227  */
1228 static int
1229 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1230 {
1231 	u_int chunk_len;
1232 	int error;
1233 
1234 	while (len > 0) {
1235 		chunk_len = MIN(len, CORE_BUF_SIZE);
1236 
1237 		/*
1238 		 * We can get EFAULT error here.
1239 		 * In that case zero out the current chunk of the segment.
1240 		 */
1241 		error = copyin(base, buf, chunk_len);
1242 		if (error != 0)
1243 			bzero(buf, chunk_len);
1244 		error = gzio_write(p->gzs, buf, chunk_len);
1245 		if (error != 0)
1246 			break;
1247 		base += chunk_len;
1248 		len -= chunk_len;
1249 	}
1250 	return (error);
1251 }
1252 
1253 static int
1254 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1255 {
1256 
1257 	return (core_write((struct coredump_params *)arg, base, len, offset,
1258 	    UIO_SYSSPACE));
1259 }
1260 #endif /* GZIO */
1261 
1262 static int
1263 core_write(struct coredump_params *p, const void *base, size_t len,
1264     off_t offset, enum uio_seg seg)
1265 {
1266 
1267 	return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1268 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1269 	    p->active_cred, p->file_cred, NULL, p->td));
1270 }
1271 
1272 static int
1273 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1274     void *tmpbuf)
1275 {
1276 	int error;
1277 
1278 #ifdef GZIO
1279 	if (p->gzs != NULL)
1280 		return (compress_chunk(p, base, tmpbuf, len));
1281 #endif
1282 	/*
1283 	 * EFAULT is a non-fatal error that we can get, for example,
1284 	 * if the segment is backed by a file but extends beyond its
1285 	 * end.
1286 	 */
1287 	error = core_write(p, base, len, offset, UIO_USERSPACE);
1288 	if (error == EFAULT) {
1289 		log(LOG_WARNING, "Failed to fully fault in a core file segment "
1290 		    "at VA %p with size 0x%zx to be written at offset 0x%jx "
1291 		    "for process %s\n", base, len, offset, curproc->p_comm);
1292 
1293 		/*
1294 		 * Write a "real" zero byte at the end of the target region
1295 		 * in the case this is the last segment.
1296 		 * The intermediate space will be implicitly zero-filled.
1297 		 */
1298 		error = core_write(p, zero_region, 1, offset + len - 1,
1299 		    UIO_SYSSPACE);
1300 	}
1301 	return (error);
1302 }
1303 
1304 /*
1305  * Drain into a core file.
1306  */
1307 static int
1308 sbuf_drain_core_output(void *arg, const char *data, int len)
1309 {
1310 	struct coredump_params *p;
1311 	int error, locked;
1312 
1313 	p = (struct coredump_params *)arg;
1314 
1315 	/*
1316 	 * Some kern_proc out routines that print to this sbuf may
1317 	 * call us with the process lock held. Draining with the
1318 	 * non-sleepable lock held is unsafe. The lock is needed for
1319 	 * those routines when dumping a live process. In our case we
1320 	 * can safely release the lock before draining and acquire
1321 	 * again after.
1322 	 */
1323 	locked = PROC_LOCKED(p->td->td_proc);
1324 	if (locked)
1325 		PROC_UNLOCK(p->td->td_proc);
1326 #ifdef GZIO
1327 	if (p->gzs != NULL)
1328 		error = gzio_write(p->gzs, __DECONST(char *, data), len);
1329 	else
1330 #endif
1331 		error = core_write(p, __DECONST(void *, data), len, p->offset,
1332 		    UIO_SYSSPACE);
1333 	if (locked)
1334 		PROC_LOCK(p->td->td_proc);
1335 	if (error != 0)
1336 		return (-error);
1337 	p->offset += len;
1338 	return (len);
1339 }
1340 
1341 /*
1342  * Drain into a counter.
1343  */
1344 static int
1345 sbuf_drain_count(void *arg, const char *data __unused, int len)
1346 {
1347 	size_t *sizep;
1348 
1349 	sizep = (size_t *)arg;
1350 	*sizep += len;
1351 	return (len);
1352 }
1353 
1354 int
1355 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1356 {
1357 	struct ucred *cred = td->td_ucred;
1358 	int error = 0;
1359 	struct sseg_closure seginfo;
1360 	struct note_info_list notelst;
1361 	struct coredump_params params;
1362 	struct note_info *ninfo;
1363 	void *hdr, *tmpbuf;
1364 	size_t hdrsize, notesz, coresize;
1365 #ifdef GZIO
1366 	boolean_t compress;
1367 
1368 	compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1369 #endif
1370 	hdr = NULL;
1371 	tmpbuf = NULL;
1372 	TAILQ_INIT(&notelst);
1373 
1374 	/* Size the program segments. */
1375 	seginfo.count = 0;
1376 	seginfo.size = 0;
1377 	each_dumpable_segment(td, cb_size_segment, &seginfo);
1378 
1379 	/*
1380 	 * Collect info about the core file header area.
1381 	 */
1382 	hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1383 	if (seginfo.count + 1 >= PN_XNUM)
1384 		hdrsize += sizeof(Elf_Shdr);
1385 	__elfN(prepare_notes)(td, &notelst, &notesz);
1386 	coresize = round_page(hdrsize + notesz) + seginfo.size;
1387 
1388 	/* Set up core dump parameters. */
1389 	params.offset = 0;
1390 	params.active_cred = cred;
1391 	params.file_cred = NOCRED;
1392 	params.td = td;
1393 	params.vp = vp;
1394 	params.gzs = NULL;
1395 
1396 #ifdef RACCT
1397 	if (racct_enable) {
1398 		PROC_LOCK(td->td_proc);
1399 		error = racct_add(td->td_proc, RACCT_CORE, coresize);
1400 		PROC_UNLOCK(td->td_proc);
1401 		if (error != 0) {
1402 			error = EFAULT;
1403 			goto done;
1404 		}
1405 	}
1406 #endif
1407 	if (coresize >= limit) {
1408 		error = EFAULT;
1409 		goto done;
1410 	}
1411 
1412 #ifdef GZIO
1413 	/* Create a compression stream if necessary. */
1414 	if (compress) {
1415 		params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1416 		    CORE_BUF_SIZE, compress_user_cores_gzlevel, &params);
1417 		if (params.gzs == NULL) {
1418 			error = EFAULT;
1419 			goto done;
1420 		}
1421 		tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1422         }
1423 #endif
1424 
1425 	/*
1426 	 * Allocate memory for building the header, fill it up,
1427 	 * and write it out following the notes.
1428 	 */
1429 	hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1430 	error = __elfN(corehdr)(&params, seginfo.count, hdr, hdrsize, &notelst,
1431 	    notesz);
1432 
1433 	/* Write the contents of all of the writable segments. */
1434 	if (error == 0) {
1435 		Elf_Phdr *php;
1436 		off_t offset;
1437 		int i;
1438 
1439 		php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1440 		offset = round_page(hdrsize + notesz);
1441 		for (i = 0; i < seginfo.count; i++) {
1442 			error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1443 			    php->p_filesz, offset, &params, tmpbuf);
1444 			if (error != 0)
1445 				break;
1446 			offset += php->p_filesz;
1447 			php++;
1448 		}
1449 #ifdef GZIO
1450 		if (error == 0 && compress)
1451 			error = gzio_flush(params.gzs);
1452 #endif
1453 	}
1454 	if (error) {
1455 		log(LOG_WARNING,
1456 		    "Failed to write core file for process %s (error %d)\n",
1457 		    curproc->p_comm, error);
1458 	}
1459 
1460 done:
1461 #ifdef GZIO
1462 	if (compress) {
1463 		free(tmpbuf, M_TEMP);
1464 		if (params.gzs != NULL)
1465 			gzio_fini(params.gzs);
1466 	}
1467 #endif
1468 	while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1469 		TAILQ_REMOVE(&notelst, ninfo, link);
1470 		free(ninfo, M_TEMP);
1471 	}
1472 	if (hdr != NULL)
1473 		free(hdr, M_TEMP);
1474 
1475 	return (error);
1476 }
1477 
1478 /*
1479  * A callback for each_dumpable_segment() to write out the segment's
1480  * program header entry.
1481  */
1482 static void
1483 cb_put_phdr(entry, closure)
1484 	vm_map_entry_t entry;
1485 	void *closure;
1486 {
1487 	struct phdr_closure *phc = (struct phdr_closure *)closure;
1488 	Elf_Phdr *phdr = phc->phdr;
1489 
1490 	phc->offset = round_page(phc->offset);
1491 
1492 	phdr->p_type = PT_LOAD;
1493 	phdr->p_offset = phc->offset;
1494 	phdr->p_vaddr = entry->start;
1495 	phdr->p_paddr = 0;
1496 	phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1497 	phdr->p_align = PAGE_SIZE;
1498 	phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1499 
1500 	phc->offset += phdr->p_filesz;
1501 	phc->phdr++;
1502 }
1503 
1504 /*
1505  * A callback for each_dumpable_segment() to gather information about
1506  * the number of segments and their total size.
1507  */
1508 static void
1509 cb_size_segment(vm_map_entry_t entry, void *closure)
1510 {
1511 	struct sseg_closure *ssc = (struct sseg_closure *)closure;
1512 
1513 	ssc->count++;
1514 	ssc->size += entry->end - entry->start;
1515 }
1516 
1517 /*
1518  * For each writable segment in the process's memory map, call the given
1519  * function with a pointer to the map entry and some arbitrary
1520  * caller-supplied data.
1521  */
1522 static void
1523 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1524 {
1525 	struct proc *p = td->td_proc;
1526 	vm_map_t map = &p->p_vmspace->vm_map;
1527 	vm_map_entry_t entry;
1528 	vm_object_t backing_object, object;
1529 	boolean_t ignore_entry;
1530 
1531 	vm_map_lock_read(map);
1532 	for (entry = map->header.next; entry != &map->header;
1533 	    entry = entry->next) {
1534 		/*
1535 		 * Don't dump inaccessible mappings, deal with legacy
1536 		 * coredump mode.
1537 		 *
1538 		 * Note that read-only segments related to the elf binary
1539 		 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1540 		 * need to arbitrarily ignore such segments.
1541 		 */
1542 		if (elf_legacy_coredump) {
1543 			if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1544 				continue;
1545 		} else {
1546 			if ((entry->protection & VM_PROT_ALL) == 0)
1547 				continue;
1548 		}
1549 
1550 		/*
1551 		 * Dont include memory segment in the coredump if
1552 		 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1553 		 * madvise(2).  Do not dump submaps (i.e. parts of the
1554 		 * kernel map).
1555 		 */
1556 		if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1557 			continue;
1558 
1559 		if ((object = entry->object.vm_object) == NULL)
1560 			continue;
1561 
1562 		/* Ignore memory-mapped devices and such things. */
1563 		VM_OBJECT_RLOCK(object);
1564 		while ((backing_object = object->backing_object) != NULL) {
1565 			VM_OBJECT_RLOCK(backing_object);
1566 			VM_OBJECT_RUNLOCK(object);
1567 			object = backing_object;
1568 		}
1569 		ignore_entry = object->type != OBJT_DEFAULT &&
1570 		    object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1571 		    object->type != OBJT_PHYS;
1572 		VM_OBJECT_RUNLOCK(object);
1573 		if (ignore_entry)
1574 			continue;
1575 
1576 		(*func)(entry, closure);
1577 	}
1578 	vm_map_unlock_read(map);
1579 }
1580 
1581 /*
1582  * Write the core file header to the file, including padding up to
1583  * the page boundary.
1584  */
1585 static int
1586 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1587     size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1588 {
1589 	struct note_info *ninfo;
1590 	struct sbuf *sb;
1591 	int error;
1592 
1593 	/* Fill in the header. */
1594 	bzero(hdr, hdrsize);
1595 	__elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1596 
1597 	sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1598 	sbuf_set_drain(sb, sbuf_drain_core_output, p);
1599 	sbuf_start_section(sb, NULL);
1600 	sbuf_bcat(sb, hdr, hdrsize);
1601 	TAILQ_FOREACH(ninfo, notelst, link)
1602 	    __elfN(putnote)(ninfo, sb);
1603 	/* Align up to a page boundary for the program segments. */
1604 	sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1605 	error = sbuf_finish(sb);
1606 	sbuf_delete(sb);
1607 
1608 	return (error);
1609 }
1610 
1611 static void
1612 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1613     size_t *sizep)
1614 {
1615 	struct proc *p;
1616 	struct thread *thr;
1617 	size_t size;
1618 
1619 	p = td->td_proc;
1620 	size = 0;
1621 
1622 	size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1623 
1624 	/*
1625 	 * To have the debugger select the right thread (LWP) as the initial
1626 	 * thread, we dump the state of the thread passed to us in td first.
1627 	 * This is the thread that causes the core dump and thus likely to
1628 	 * be the right thread one wants to have selected in the debugger.
1629 	 */
1630 	thr = td;
1631 	while (thr != NULL) {
1632 		size += register_note(list, NT_PRSTATUS,
1633 		    __elfN(note_prstatus), thr);
1634 		size += register_note(list, NT_FPREGSET,
1635 		    __elfN(note_fpregset), thr);
1636 		size += register_note(list, NT_THRMISC,
1637 		    __elfN(note_thrmisc), thr);
1638 		size += register_note(list, NT_PTLWPINFO,
1639 		    __elfN(note_ptlwpinfo), thr);
1640 		size += register_note(list, -1,
1641 		    __elfN(note_threadmd), thr);
1642 
1643 		thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1644 		    TAILQ_NEXT(thr, td_plist);
1645 		if (thr == td)
1646 			thr = TAILQ_NEXT(thr, td_plist);
1647 	}
1648 
1649 	size += register_note(list, NT_PROCSTAT_PROC,
1650 	    __elfN(note_procstat_proc), p);
1651 	size += register_note(list, NT_PROCSTAT_FILES,
1652 	    note_procstat_files, p);
1653 	size += register_note(list, NT_PROCSTAT_VMMAP,
1654 	    note_procstat_vmmap, p);
1655 	size += register_note(list, NT_PROCSTAT_GROUPS,
1656 	    note_procstat_groups, p);
1657 	size += register_note(list, NT_PROCSTAT_UMASK,
1658 	    note_procstat_umask, p);
1659 	size += register_note(list, NT_PROCSTAT_RLIMIT,
1660 	    note_procstat_rlimit, p);
1661 	size += register_note(list, NT_PROCSTAT_OSREL,
1662 	    note_procstat_osrel, p);
1663 	size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1664 	    __elfN(note_procstat_psstrings), p);
1665 	size += register_note(list, NT_PROCSTAT_AUXV,
1666 	    __elfN(note_procstat_auxv), p);
1667 
1668 	*sizep = size;
1669 }
1670 
1671 static void
1672 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1673     size_t notesz)
1674 {
1675 	Elf_Ehdr *ehdr;
1676 	Elf_Phdr *phdr;
1677 	Elf_Shdr *shdr;
1678 	struct phdr_closure phc;
1679 
1680 	ehdr = (Elf_Ehdr *)hdr;
1681 
1682 	ehdr->e_ident[EI_MAG0] = ELFMAG0;
1683 	ehdr->e_ident[EI_MAG1] = ELFMAG1;
1684 	ehdr->e_ident[EI_MAG2] = ELFMAG2;
1685 	ehdr->e_ident[EI_MAG3] = ELFMAG3;
1686 	ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1687 	ehdr->e_ident[EI_DATA] = ELF_DATA;
1688 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1689 	ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1690 	ehdr->e_ident[EI_ABIVERSION] = 0;
1691 	ehdr->e_ident[EI_PAD] = 0;
1692 	ehdr->e_type = ET_CORE;
1693 	ehdr->e_machine = td->td_proc->p_elf_machine;
1694 	ehdr->e_version = EV_CURRENT;
1695 	ehdr->e_entry = 0;
1696 	ehdr->e_phoff = sizeof(Elf_Ehdr);
1697 	ehdr->e_flags = td->td_proc->p_elf_flags;
1698 	ehdr->e_ehsize = sizeof(Elf_Ehdr);
1699 	ehdr->e_phentsize = sizeof(Elf_Phdr);
1700 	ehdr->e_shentsize = sizeof(Elf_Shdr);
1701 	ehdr->e_shstrndx = SHN_UNDEF;
1702 	if (numsegs + 1 < PN_XNUM) {
1703 		ehdr->e_phnum = numsegs + 1;
1704 		ehdr->e_shnum = 0;
1705 	} else {
1706 		ehdr->e_phnum = PN_XNUM;
1707 		ehdr->e_shnum = 1;
1708 
1709 		ehdr->e_shoff = ehdr->e_phoff +
1710 		    (numsegs + 1) * ehdr->e_phentsize;
1711 		KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1712 		    ("e_shoff: %zu, hdrsize - shdr: %zu",
1713 		     (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1714 
1715 		shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1716 		memset(shdr, 0, sizeof(*shdr));
1717 		/*
1718 		 * A special first section is used to hold large segment and
1719 		 * section counts.  This was proposed by Sun Microsystems in
1720 		 * Solaris and has been adopted by Linux; the standard ELF
1721 		 * tools are already familiar with the technique.
1722 		 *
1723 		 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1724 		 * (or 12-7 depending on the version of the document) for more
1725 		 * details.
1726 		 */
1727 		shdr->sh_type = SHT_NULL;
1728 		shdr->sh_size = ehdr->e_shnum;
1729 		shdr->sh_link = ehdr->e_shstrndx;
1730 		shdr->sh_info = numsegs + 1;
1731 	}
1732 
1733 	/*
1734 	 * Fill in the program header entries.
1735 	 */
1736 	phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1737 
1738 	/* The note segement. */
1739 	phdr->p_type = PT_NOTE;
1740 	phdr->p_offset = hdrsize;
1741 	phdr->p_vaddr = 0;
1742 	phdr->p_paddr = 0;
1743 	phdr->p_filesz = notesz;
1744 	phdr->p_memsz = 0;
1745 	phdr->p_flags = PF_R;
1746 	phdr->p_align = ELF_NOTE_ROUNDSIZE;
1747 	phdr++;
1748 
1749 	/* All the writable segments from the program. */
1750 	phc.phdr = phdr;
1751 	phc.offset = round_page(hdrsize + notesz);
1752 	each_dumpable_segment(td, cb_put_phdr, &phc);
1753 }
1754 
1755 static size_t
1756 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1757 {
1758 	struct note_info *ninfo;
1759 	size_t size, notesize;
1760 
1761 	size = 0;
1762 	out(arg, NULL, &size);
1763 	ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1764 	ninfo->type = type;
1765 	ninfo->outfunc = out;
1766 	ninfo->outarg = arg;
1767 	ninfo->outsize = size;
1768 	TAILQ_INSERT_TAIL(list, ninfo, link);
1769 
1770 	if (type == -1)
1771 		return (size);
1772 
1773 	notesize = sizeof(Elf_Note) +		/* note header */
1774 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1775 						/* note name */
1776 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1777 
1778 	return (notesize);
1779 }
1780 
1781 static size_t
1782 append_note_data(const void *src, void *dst, size_t len)
1783 {
1784 	size_t padded_len;
1785 
1786 	padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1787 	if (dst != NULL) {
1788 		bcopy(src, dst, len);
1789 		bzero((char *)dst + len, padded_len - len);
1790 	}
1791 	return (padded_len);
1792 }
1793 
1794 size_t
1795 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1796 {
1797 	Elf_Note *note;
1798 	char *buf;
1799 	size_t notesize;
1800 
1801 	buf = dst;
1802 	if (buf != NULL) {
1803 		note = (Elf_Note *)buf;
1804 		note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1805 		note->n_descsz = size;
1806 		note->n_type = type;
1807 		buf += sizeof(*note);
1808 		buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1809 		    sizeof(FREEBSD_ABI_VENDOR));
1810 		append_note_data(src, buf, size);
1811 		if (descp != NULL)
1812 			*descp = buf;
1813 	}
1814 
1815 	notesize = sizeof(Elf_Note) +		/* note header */
1816 	    roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1817 						/* note name */
1818 	    roundup2(size, ELF_NOTE_ROUNDSIZE);	/* note description */
1819 
1820 	return (notesize);
1821 }
1822 
1823 static void
1824 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1825 {
1826 	Elf_Note note;
1827 	ssize_t old_len, sect_len;
1828 	size_t new_len, descsz, i;
1829 
1830 	if (ninfo->type == -1) {
1831 		ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1832 		return;
1833 	}
1834 
1835 	note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1836 	note.n_descsz = ninfo->outsize;
1837 	note.n_type = ninfo->type;
1838 
1839 	sbuf_bcat(sb, &note, sizeof(note));
1840 	sbuf_start_section(sb, &old_len);
1841 	sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1842 	sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1843 	if (note.n_descsz == 0)
1844 		return;
1845 	sbuf_start_section(sb, &old_len);
1846 	ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1847 	sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1848 	if (sect_len < 0)
1849 		return;
1850 
1851 	new_len = (size_t)sect_len;
1852 	descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1853 	if (new_len < descsz) {
1854 		/*
1855 		 * It is expected that individual note emitters will correctly
1856 		 * predict their expected output size and fill up to that size
1857 		 * themselves, padding in a format-specific way if needed.
1858 		 * However, in case they don't, just do it here with zeros.
1859 		 */
1860 		for (i = 0; i < descsz - new_len; i++)
1861 			sbuf_putc(sb, 0);
1862 	} else if (new_len > descsz) {
1863 		/*
1864 		 * We can't always truncate sb -- we may have drained some
1865 		 * of it already.
1866 		 */
1867 		KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1868 		    "read it (%zu > %zu).  Since it is longer than "
1869 		    "expected, this coredump's notes are corrupt.  THIS "
1870 		    "IS A BUG in the note_procstat routine for type %u.\n",
1871 		    __func__, (unsigned)note.n_type, new_len, descsz,
1872 		    (unsigned)note.n_type));
1873 	}
1874 }
1875 
1876 /*
1877  * Miscellaneous note out functions.
1878  */
1879 
1880 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1881 #include <compat/freebsd32/freebsd32.h>
1882 #include <compat/freebsd32/freebsd32_signal.h>
1883 
1884 typedef struct prstatus32 elf_prstatus_t;
1885 typedef struct prpsinfo32 elf_prpsinfo_t;
1886 typedef struct fpreg32 elf_prfpregset_t;
1887 typedef struct fpreg32 elf_fpregset_t;
1888 typedef struct reg32 elf_gregset_t;
1889 typedef struct thrmisc32 elf_thrmisc_t;
1890 #define ELF_KERN_PROC_MASK	KERN_PROC_MASK32
1891 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1892 typedef uint32_t elf_ps_strings_t;
1893 #else
1894 typedef prstatus_t elf_prstatus_t;
1895 typedef prpsinfo_t elf_prpsinfo_t;
1896 typedef prfpregset_t elf_prfpregset_t;
1897 typedef prfpregset_t elf_fpregset_t;
1898 typedef gregset_t elf_gregset_t;
1899 typedef thrmisc_t elf_thrmisc_t;
1900 #define ELF_KERN_PROC_MASK	0
1901 typedef struct kinfo_proc elf_kinfo_proc_t;
1902 typedef vm_offset_t elf_ps_strings_t;
1903 #endif
1904 
1905 static void
1906 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1907 {
1908 	struct sbuf sbarg;
1909 	size_t len;
1910 	char *cp, *end;
1911 	struct proc *p;
1912 	elf_prpsinfo_t *psinfo;
1913 	int error;
1914 
1915 	p = (struct proc *)arg;
1916 	if (sb != NULL) {
1917 		KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1918 		psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1919 		psinfo->pr_version = PRPSINFO_VERSION;
1920 		psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1921 		strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1922 		PROC_LOCK(p);
1923 		if (p->p_args != NULL) {
1924 			len = sizeof(psinfo->pr_psargs) - 1;
1925 			if (len > p->p_args->ar_length)
1926 				len = p->p_args->ar_length;
1927 			memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1928 			PROC_UNLOCK(p);
1929 			error = 0;
1930 		} else {
1931 			_PHOLD(p);
1932 			PROC_UNLOCK(p);
1933 			sbuf_new(&sbarg, psinfo->pr_psargs,
1934 			    sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1935 			error = proc_getargv(curthread, p, &sbarg);
1936 			PRELE(p);
1937 			if (sbuf_finish(&sbarg) == 0)
1938 				len = sbuf_len(&sbarg) - 1;
1939 			else
1940 				len = sizeof(psinfo->pr_psargs) - 1;
1941 			sbuf_delete(&sbarg);
1942 		}
1943 		if (error || len == 0)
1944 			strlcpy(psinfo->pr_psargs, p->p_comm,
1945 			    sizeof(psinfo->pr_psargs));
1946 		else {
1947 			KASSERT(len < sizeof(psinfo->pr_psargs),
1948 			    ("len is too long: %zu vs %zu", len,
1949 			    sizeof(psinfo->pr_psargs)));
1950 			cp = psinfo->pr_psargs;
1951 			end = cp + len - 1;
1952 			for (;;) {
1953 				cp = memchr(cp, '\0', end - cp);
1954 				if (cp == NULL)
1955 					break;
1956 				*cp = ' ';
1957 			}
1958 		}
1959 		psinfo->pr_pid = p->p_pid;
1960 		sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1961 		free(psinfo, M_TEMP);
1962 	}
1963 	*sizep = sizeof(*psinfo);
1964 }
1965 
1966 static void
1967 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1968 {
1969 	struct thread *td;
1970 	elf_prstatus_t *status;
1971 
1972 	td = (struct thread *)arg;
1973 	if (sb != NULL) {
1974 		KASSERT(*sizep == sizeof(*status), ("invalid size"));
1975 		status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1976 		status->pr_version = PRSTATUS_VERSION;
1977 		status->pr_statussz = sizeof(elf_prstatus_t);
1978 		status->pr_gregsetsz = sizeof(elf_gregset_t);
1979 		status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1980 		status->pr_osreldate = osreldate;
1981 		status->pr_cursig = td->td_proc->p_sig;
1982 		status->pr_pid = td->td_tid;
1983 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1984 		fill_regs32(td, &status->pr_reg);
1985 #else
1986 		fill_regs(td, &status->pr_reg);
1987 #endif
1988 		sbuf_bcat(sb, status, sizeof(*status));
1989 		free(status, M_TEMP);
1990 	}
1991 	*sizep = sizeof(*status);
1992 }
1993 
1994 static void
1995 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1996 {
1997 	struct thread *td;
1998 	elf_prfpregset_t *fpregset;
1999 
2000 	td = (struct thread *)arg;
2001 	if (sb != NULL) {
2002 		KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2003 		fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2004 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2005 		fill_fpregs32(td, fpregset);
2006 #else
2007 		fill_fpregs(td, fpregset);
2008 #endif
2009 		sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2010 		free(fpregset, M_TEMP);
2011 	}
2012 	*sizep = sizeof(*fpregset);
2013 }
2014 
2015 static void
2016 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2017 {
2018 	struct thread *td;
2019 	elf_thrmisc_t thrmisc;
2020 
2021 	td = (struct thread *)arg;
2022 	if (sb != NULL) {
2023 		KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2024 		bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2025 		strcpy(thrmisc.pr_tname, td->td_name);
2026 		sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2027 	}
2028 	*sizep = sizeof(thrmisc);
2029 }
2030 
2031 static void
2032 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2033 {
2034 	struct thread *td;
2035 	size_t size;
2036 	int structsize;
2037 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2038 	struct ptrace_lwpinfo32 pl;
2039 #else
2040 	struct ptrace_lwpinfo pl;
2041 #endif
2042 
2043 	td = (struct thread *)arg;
2044 	size = sizeof(structsize) + sizeof(pl);
2045 	if (sb != NULL) {
2046 		KASSERT(*sizep == size, ("invalid size"));
2047 		structsize = sizeof(pl);
2048 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2049 		bzero(&pl, sizeof(pl));
2050 		pl.pl_lwpid = td->td_tid;
2051 		pl.pl_event = PL_EVENT_NONE;
2052 		pl.pl_sigmask = td->td_sigmask;
2053 		pl.pl_siglist = td->td_siglist;
2054 		if (td->td_si.si_signo != 0) {
2055 			pl.pl_event = PL_EVENT_SIGNAL;
2056 			pl.pl_flags |= PL_FLAG_SI;
2057 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2058 			siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2059 #else
2060 			pl.pl_siginfo = td->td_si;
2061 #endif
2062 		}
2063 		strcpy(pl.pl_tdname, td->td_name);
2064 		/* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2065 		sbuf_bcat(sb, &pl, sizeof(pl));
2066 	}
2067 	*sizep = size;
2068 }
2069 
2070 /*
2071  * Allow for MD specific notes, as well as any MD
2072  * specific preparations for writing MI notes.
2073  */
2074 static void
2075 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2076 {
2077 	struct thread *td;
2078 	void *buf;
2079 	size_t size;
2080 
2081 	td = (struct thread *)arg;
2082 	size = *sizep;
2083 	if (size != 0 && sb != NULL)
2084 		buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2085 	else
2086 		buf = NULL;
2087 	size = 0;
2088 	__elfN(dump_thread)(td, buf, &size);
2089 	KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2090 	if (size != 0 && sb != NULL)
2091 		sbuf_bcat(sb, buf, size);
2092 	free(buf, M_TEMP);
2093 	*sizep = size;
2094 }
2095 
2096 #ifdef KINFO_PROC_SIZE
2097 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2098 #endif
2099 
2100 static void
2101 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2102 {
2103 	struct proc *p;
2104 	size_t size;
2105 	int structsize;
2106 
2107 	p = (struct proc *)arg;
2108 	size = sizeof(structsize) + p->p_numthreads *
2109 	    sizeof(elf_kinfo_proc_t);
2110 
2111 	if (sb != NULL) {
2112 		KASSERT(*sizep == size, ("invalid size"));
2113 		structsize = sizeof(elf_kinfo_proc_t);
2114 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2115 		sx_slock(&proctree_lock);
2116 		PROC_LOCK(p);
2117 		kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2118 		sx_sunlock(&proctree_lock);
2119 	}
2120 	*sizep = size;
2121 }
2122 
2123 #ifdef KINFO_FILE_SIZE
2124 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2125 #endif
2126 
2127 static void
2128 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2129 {
2130 	struct proc *p;
2131 	size_t size, sect_sz, i;
2132 	ssize_t start_len, sect_len;
2133 	int structsize, filedesc_flags;
2134 
2135 	if (coredump_pack_fileinfo)
2136 		filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2137 	else
2138 		filedesc_flags = 0;
2139 
2140 	p = (struct proc *)arg;
2141 	structsize = sizeof(struct kinfo_file);
2142 	if (sb == NULL) {
2143 		size = 0;
2144 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2145 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2146 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2147 		PROC_LOCK(p);
2148 		kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2149 		sbuf_finish(sb);
2150 		sbuf_delete(sb);
2151 		*sizep = size;
2152 	} else {
2153 		sbuf_start_section(sb, &start_len);
2154 
2155 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2156 		PROC_LOCK(p);
2157 		kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2158 		    filedesc_flags);
2159 
2160 		sect_len = sbuf_end_section(sb, start_len, 0, 0);
2161 		if (sect_len < 0)
2162 			return;
2163 		sect_sz = sect_len;
2164 
2165 		KASSERT(sect_sz <= *sizep,
2166 		    ("kern_proc_filedesc_out did not respect maxlen; "
2167 		     "requested %zu, got %zu", *sizep - sizeof(structsize),
2168 		     sect_sz - sizeof(structsize)));
2169 
2170 		for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2171 			sbuf_putc(sb, 0);
2172 	}
2173 }
2174 
2175 #ifdef KINFO_VMENTRY_SIZE
2176 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2177 #endif
2178 
2179 static void
2180 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2181 {
2182 	struct proc *p;
2183 	size_t size;
2184 	int structsize, vmmap_flags;
2185 
2186 	if (coredump_pack_vmmapinfo)
2187 		vmmap_flags = KERN_VMMAP_PACK_KINFO;
2188 	else
2189 		vmmap_flags = 0;
2190 
2191 	p = (struct proc *)arg;
2192 	structsize = sizeof(struct kinfo_vmentry);
2193 	if (sb == NULL) {
2194 		size = 0;
2195 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2196 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2197 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2198 		PROC_LOCK(p);
2199 		kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2200 		sbuf_finish(sb);
2201 		sbuf_delete(sb);
2202 		*sizep = size;
2203 	} else {
2204 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2205 		PROC_LOCK(p);
2206 		kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2207 		    vmmap_flags);
2208 	}
2209 }
2210 
2211 static void
2212 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2213 {
2214 	struct proc *p;
2215 	size_t size;
2216 	int structsize;
2217 
2218 	p = (struct proc *)arg;
2219 	size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2220 	if (sb != NULL) {
2221 		KASSERT(*sizep == size, ("invalid size"));
2222 		structsize = sizeof(gid_t);
2223 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2224 		sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2225 		    sizeof(gid_t));
2226 	}
2227 	*sizep = size;
2228 }
2229 
2230 static void
2231 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2232 {
2233 	struct proc *p;
2234 	size_t size;
2235 	int structsize;
2236 
2237 	p = (struct proc *)arg;
2238 	size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2239 	if (sb != NULL) {
2240 		KASSERT(*sizep == size, ("invalid size"));
2241 		structsize = sizeof(p->p_fd->fd_cmask);
2242 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2243 		sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2244 	}
2245 	*sizep = size;
2246 }
2247 
2248 static void
2249 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2250 {
2251 	struct proc *p;
2252 	struct rlimit rlim[RLIM_NLIMITS];
2253 	size_t size;
2254 	int structsize, i;
2255 
2256 	p = (struct proc *)arg;
2257 	size = sizeof(structsize) + sizeof(rlim);
2258 	if (sb != NULL) {
2259 		KASSERT(*sizep == size, ("invalid size"));
2260 		structsize = sizeof(rlim);
2261 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2262 		PROC_LOCK(p);
2263 		for (i = 0; i < RLIM_NLIMITS; i++)
2264 			lim_rlimit_proc(p, i, &rlim[i]);
2265 		PROC_UNLOCK(p);
2266 		sbuf_bcat(sb, rlim, sizeof(rlim));
2267 	}
2268 	*sizep = size;
2269 }
2270 
2271 static void
2272 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2273 {
2274 	struct proc *p;
2275 	size_t size;
2276 	int structsize;
2277 
2278 	p = (struct proc *)arg;
2279 	size = sizeof(structsize) + sizeof(p->p_osrel);
2280 	if (sb != NULL) {
2281 		KASSERT(*sizep == size, ("invalid size"));
2282 		structsize = sizeof(p->p_osrel);
2283 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2284 		sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2285 	}
2286 	*sizep = size;
2287 }
2288 
2289 static void
2290 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2291 {
2292 	struct proc *p;
2293 	elf_ps_strings_t ps_strings;
2294 	size_t size;
2295 	int structsize;
2296 
2297 	p = (struct proc *)arg;
2298 	size = sizeof(structsize) + sizeof(ps_strings);
2299 	if (sb != NULL) {
2300 		KASSERT(*sizep == size, ("invalid size"));
2301 		structsize = sizeof(ps_strings);
2302 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2303 		ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2304 #else
2305 		ps_strings = p->p_sysent->sv_psstrings;
2306 #endif
2307 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2308 		sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2309 	}
2310 	*sizep = size;
2311 }
2312 
2313 static void
2314 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2315 {
2316 	struct proc *p;
2317 	size_t size;
2318 	int structsize;
2319 
2320 	p = (struct proc *)arg;
2321 	if (sb == NULL) {
2322 		size = 0;
2323 		sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2324 		sbuf_set_drain(sb, sbuf_drain_count, &size);
2325 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2326 		PHOLD(p);
2327 		proc_getauxv(curthread, p, sb);
2328 		PRELE(p);
2329 		sbuf_finish(sb);
2330 		sbuf_delete(sb);
2331 		*sizep = size;
2332 	} else {
2333 		structsize = sizeof(Elf_Auxinfo);
2334 		sbuf_bcat(sb, &structsize, sizeof(structsize));
2335 		PHOLD(p);
2336 		proc_getauxv(curthread, p, sb);
2337 		PRELE(p);
2338 	}
2339 }
2340 
2341 static boolean_t
2342 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2343     int32_t *osrel, const Elf_Phdr *pnote)
2344 {
2345 	const Elf_Note *note, *note0, *note_end;
2346 	const char *note_name;
2347 	char *buf;
2348 	int i, error;
2349 	boolean_t res;
2350 
2351 	/* We need some limit, might as well use PAGE_SIZE. */
2352 	if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2353 		return (FALSE);
2354 	ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2355 	if (pnote->p_offset > PAGE_SIZE ||
2356 	    pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2357 		VOP_UNLOCK(imgp->vp, 0);
2358 		buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2359 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2360 		error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2361 		    pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2362 		    curthread->td_ucred, NOCRED, NULL, curthread);
2363 		if (error != 0) {
2364 			uprintf("i/o error PT_NOTE\n");
2365 			res = FALSE;
2366 			goto ret;
2367 		}
2368 		note = note0 = (const Elf_Note *)buf;
2369 		note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2370 	} else {
2371 		note = note0 = (const Elf_Note *)(imgp->image_header +
2372 		    pnote->p_offset);
2373 		note_end = (const Elf_Note *)(imgp->image_header +
2374 		    pnote->p_offset + pnote->p_filesz);
2375 		buf = NULL;
2376 	}
2377 	for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2378 		if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2379 		    (const char *)note < sizeof(Elf_Note)) {
2380 			res = FALSE;
2381 			goto ret;
2382 		}
2383 		if (note->n_namesz != checknote->hdr.n_namesz ||
2384 		    note->n_descsz != checknote->hdr.n_descsz ||
2385 		    note->n_type != checknote->hdr.n_type)
2386 			goto nextnote;
2387 		note_name = (const char *)(note + 1);
2388 		if (note_name + checknote->hdr.n_namesz >=
2389 		    (const char *)note_end || strncmp(checknote->vendor,
2390 		    note_name, checknote->hdr.n_namesz) != 0)
2391 			goto nextnote;
2392 
2393 		/*
2394 		 * Fetch the osreldate for binary
2395 		 * from the ELF OSABI-note if necessary.
2396 		 */
2397 		if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2398 		    checknote->trans_osrel != NULL) {
2399 			res = checknote->trans_osrel(note, osrel);
2400 			goto ret;
2401 		}
2402 		res = TRUE;
2403 		goto ret;
2404 nextnote:
2405 		note = (const Elf_Note *)((const char *)(note + 1) +
2406 		    roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2407 		    roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2408 	}
2409 	res = FALSE;
2410 ret:
2411 	free(buf, M_TEMP);
2412 	return (res);
2413 }
2414 
2415 /*
2416  * Try to find the appropriate ABI-note section for checknote,
2417  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2418  * first page of the image is searched, the same as for headers.
2419  */
2420 static boolean_t
2421 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2422     int32_t *osrel)
2423 {
2424 	const Elf_Phdr *phdr;
2425 	const Elf_Ehdr *hdr;
2426 	int i;
2427 
2428 	hdr = (const Elf_Ehdr *)imgp->image_header;
2429 	phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2430 
2431 	for (i = 0; i < hdr->e_phnum; i++) {
2432 		if (phdr[i].p_type == PT_NOTE &&
2433 		    __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2434 			return (TRUE);
2435 	}
2436 	return (FALSE);
2437 
2438 }
2439 
2440 /*
2441  * Tell kern_execve.c about it, with a little help from the linker.
2442  */
2443 static struct execsw __elfN(execsw) = {
2444 	__CONCAT(exec_, __elfN(imgact)),
2445 	__XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2446 };
2447 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2448 
2449 static vm_prot_t
2450 __elfN(trans_prot)(Elf_Word flags)
2451 {
2452 	vm_prot_t prot;
2453 
2454 	prot = 0;
2455 	if (flags & PF_X)
2456 		prot |= VM_PROT_EXECUTE;
2457 	if (flags & PF_W)
2458 		prot |= VM_PROT_WRITE;
2459 	if (flags & PF_R)
2460 		prot |= VM_PROT_READ;
2461 #if __ELF_WORD_SIZE == 32
2462 #if defined(__amd64__)
2463 	if (i386_read_exec && (flags & PF_R))
2464 		prot |= VM_PROT_EXECUTE;
2465 #endif
2466 #endif
2467 	return (prot);
2468 }
2469 
2470 static Elf_Word
2471 __elfN(untrans_prot)(vm_prot_t prot)
2472 {
2473 	Elf_Word flags;
2474 
2475 	flags = 0;
2476 	if (prot & VM_PROT_EXECUTE)
2477 		flags |= PF_X;
2478 	if (prot & VM_PROT_READ)
2479 		flags |= PF_R;
2480 	if (prot & VM_PROT_WRITE)
2481 		flags |= PF_W;
2482 	return (flags);
2483 }
2484